fix: use cargo workspace

This way bch_bindgen actually gets used by various cargo subcommands,
like clippy.

Signed-off-by: Thomas Mühlbacher <tmuehlbacher@posteo.net>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

.bcachefs_revision

@@ -1 +1 @@
-63bbe0ca416791095c994aba7bea388e947dd60a
+7fdc3fa3cb5fb561f5945b4de418d48d1a726a8d

.github/workflows/nix-flake.yml

@@ -1,22 +1,31 @@
-name: "Nix-Tests"
+name: Nix Flake actions
+
 on:
   pull_request:
   push:
+
 jobs:
-  nix-flake-check:
+  nix-matrix:
     runs-on: ubuntu-latest
+    outputs:
+      matrix: ${{ steps.set-matrix.outputs.matrix }}
     steps:
-    - uses: actions/checkout@v4
-    - uses: cachix/install-nix-action@v27
-      with:
-        extra_nix_config: |
-          experimental-features = nix-command flakes
-          access-tokens = github.com=${{ secrets.GITHUB_TOKEN }}
-    - uses: cachix/cachix-action@v15
-      with:
-        name: bcachefs-tools
-        # If you chose API tokens for write access OR if you have a private cache
-        authToken: '${{ secrets.CACHIX_AUTH_TOKEN }}'
-    - run: nix flake show
-    - run: nix flake check --print-build-logs
-    - run: nix build --print-build-logs
+      - uses: actions/checkout@v4
+      - uses: cachix/install-nix-action@v30
+      - id: set-matrix
+        name: Generate Nix Matrix
+        run: |
+          set -Eeu
+          matrix="$(nix eval --json '.#githubActions.matrix')"
+          echo "matrix=$matrix" >> "$GITHUB_OUTPUT"
+
+  nix-build:
+    name: ${{ matrix.name }} (${{ matrix.system }})
+    needs: nix-matrix
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix: ${{fromJSON(needs.nix-matrix.outputs.matrix)}}
+    steps:
+      - uses: actions/checkout@v4
+      - uses: cachix/install-nix-action@v30
+      - run: nix build -L '.#${{ matrix.attr }}'

Cargo.lock

@@ -68,7 +68,7 @@ checksum = "86fdf8605db99b54d3cd748a44c6d04df638eb5dafb219b135d0149bd0db01f6"
 
 [[package]]
 name = "bcachefs-tools"
-version = "1.12.0"
+version = "1.25.0"
 dependencies = [
  "anyhow",
  "bch_bindgen",

Cargo.toml

@@ -1,9 +1,13 @@
+[workspace]
+resolver = "2"
+default-members = [".", "bch_bindgen"]
+
 [package]
 name = "bcachefs-tools"
-version = "1.20.0"
+version = "1.25.0"
 authors = ["Yuxuan Shui <yshuiv7@gmail.com>", "Kayla Firestack <dev@kaylafire.me>", "Kent Overstreet <kent.overstreet@linux.dev>" ]
 edition = "2021"
-rust-version = "1.77"
+rust-version = "1.77.0"
 
 [[bin]]
 name = "bcachefs"

INSTALL.md

@@ -39,20 +39,19 @@ Starting from Debian Trixie and Ubuntu 23.10, you will additionally need:
 apt install -y systemd-dev
 ```
 
-Fedora: install the "Development tools" group along with:
+Fedora: install build dependencies either with `dnf builddep bcachefs-tools` or with:
 ```shell
-dnf install -y libaio-devel libsodium-devel \
+dnf install -y @c-development libaio-devel libsodium-devel \
     libblkid-devel libzstd-devel zlib-devel userspace-rcu-devel \
     lz4-devel libuuid-devel valgrind-devel keyutils-libs-devel \
-    findutils udev systemd-devel llvm-devel
+    findutils systemd-devel clang-devel llvm-devel rust cargo
 ```
 
 openSUSE: install build dependencies with:
 ```shell
-zypper install -y libaio-devel libsodium-devel \
-    libblkid-devel liburcu-devel libzstd-devel zlib-devel \
-    liblz4-devel libuuid-devel valgrind-devel keyutils-devel \
-    findutils udev systemd-devel llvm-devel
+zypper in -y libaio-devel libsodium-devel libblkid-devel liburcu-devel \
+    libzstd-devel zlib-devel liblz4-devel libuuid-devel valgrind-devel \
+    keyutils-devel findutils udev systemd-devel llvm-devel
 ```
 
 Arch: install bcachefs-tools-git from the AUR.
@@ -88,6 +87,11 @@ Arch:
 pacman -S fuse3
 ```
 
+openSUSE:
+```shell
+zypper in -y fuse3-devel
+```
+
 Then, make using the `BCACHEFS_FUSE` environment variable (make clean first if
 previously built without fuse support):
 

Makefile

@@ -1,4 +1,4 @@
-VERSION=1.20.0
+VERSION=1.25.0
 
 PREFIX?=/usr/local
 LIBEXECDIR?=$(PREFIX)/libexec
@@ -22,6 +22,13 @@ else
   CARGO_CLEAN_ARGS = --quiet
 endif
 
+# when cross compiling, cargo places the built binary in a different location
+ifdef CARGO_BUILD_TARGET
+	BUILT_BIN = target/$(CARGO_BUILD_TARGET)/release/bcachefs
+else
+	BUILT_BIN = target/release/bcachefs
+endif
+
 # Prevent recursive expansions of $(CFLAGS) to avoid repeatedly performing
 # compile tests
 CFLAGS:=$(CFLAGS)
@@ -195,7 +202,7 @@ cmd_version.o : .version
 install: INITRAMFS_HOOK=$(INITRAMFS_DIR)/hooks/bcachefs
 install: INITRAMFS_SCRIPT=$(INITRAMFS_DIR)/scripts/local-premount/bcachefs
 install: bcachefs $(optional_install)
-	$(INSTALL) -m0755 -D target/release/bcachefs -t $(DESTDIR)$(ROOT_SBINDIR)
+	$(INSTALL) -m0755 -D $(BUILT_BIN)  -t $(DESTDIR)$(ROOT_SBINDIR)
 	$(INSTALL) -m0644 -D bcachefs.8    -t $(DESTDIR)$(PREFIX)/share/man/man8/
 	$(INSTALL) -m0755 -D initramfs/script $(DESTDIR)$(INITRAMFS_SCRIPT)
 	$(INSTALL) -m0755 -D initramfs/hook   $(DESTDIR)$(INITRAMFS_HOOK)

bch_bindgen/src/opts.rs

@@ -1,3 +1,7 @@
+use crate::c;
+use crate::fs::Fs;
+use std::ffi::{CString, c_char};
+
 #[macro_export]
 macro_rules! opt_set {
     ($opts:ident, $n:ident, $v:expr) => {
@@ -33,3 +37,29 @@ macro_rules! opt_get {
         }
     };
 }
+
+pub fn parse_mount_opts(fs: Option<&mut Fs>, optstr: Option<&str>, ignore_unknown: bool)
+        -> Result<c::bch_opts, c::bch_errcode> {
+    let mut opts: c::bch_opts = Default::default();
+
+    if let Some(optstr) = optstr {
+        let optstr = CString::new(optstr).unwrap();
+        let optstr_ptr = optstr.as_ptr();
+
+        let ret = unsafe {
+            c::bch2_parse_mount_opts(fs.map_or(std::ptr::null_mut(), |f| f.raw),
+                                     &mut opts as *mut c::bch_opts,
+                                     std::ptr::null_mut(),
+                                     optstr_ptr as *mut c_char,
+                                     ignore_unknown)
+        };
+
+        drop(optstr);
+
+        if ret != 0 {
+            let err: c::bch_errcode = unsafe { std::mem::transmute(-ret) };
+            return Err(err);
+        }
+    }
+    Ok(opts)
+}

c_src/bcachefs.c

@@ -33,6 +33,7 @@ void bcachefs_usage(void)
 	     "Superblock commands:\n"
 	     "  format                   Format a new filesystem\n"
 	     "  show-super               Dump superblock information to stdout\n"
+	     "  recover-super            Attempt to recover overwritten superblock from backups\n"
 	     "  set-fs-option            Set a filesystem option\n"
 	     "  reset-counters           Reset all counters on an unmounted device\n"
 	     "\n"
@@ -92,9 +93,11 @@ void bcachefs_usage(void)
 	     "  list                     List filesystem metadata in textual form\n"
 	     "  list_journal             List contents of journal\n"
 	     "\n"
+#ifdef BCACHEFS_FUSE
 	     "FUSE:\n"
 	     "  fusemount                Mount a filesystem via FUSE\n"
 	     "\n"
+#endif
 	     "Miscellaneous:\n"
          "  completions              Generate shell completions\n"
 	     "  version                  Display the version of the invoked bcachefs tool\n");
@@ -115,16 +118,15 @@ int fs_cmds(int argc, char *argv[])
 {
 	char *cmd = pop_cmd(&argc, argv);
 
-	if (argc < 1) {
-		bcachefs_usage();
-		exit(EXIT_FAILURE);
-	}
+	if (argc < 1)
+		return fs_usage();
 	if (!strcmp(cmd, "usage"))
 		return cmd_fs_usage(argc, argv);
 	if (!strcmp(cmd, "top"))
 		return cmd_fs_top(argc, argv);
 
-	return 0;
+	fs_usage();
+	return -EINVAL;
 }
 
 int device_cmds(int argc, char *argv[])
@@ -150,7 +152,8 @@ int device_cmds(int argc, char *argv[])
 	if (!strcmp(cmd, "resize-journal"))
 		return cmd_device_resize_journal(argc, argv);
 
-	return 0;
+	device_usage();
+	return -EINVAL;
 }
 
 int data_cmds(int argc, char *argv[])
@@ -166,5 +169,6 @@ int data_cmds(int argc, char *argv[])
 	if (!strcmp(cmd, "job"))
 		return cmd_data_job(argc, argv);
 
-	return 0;
+	data_usage();
+	return -EINVAL;
 }

c_src/cmd_device.c

@@ -47,11 +47,11 @@ static void device_add_usage(void)
 	puts("bcachefs device add - add a device to an existing filesystem\n"
 	     "Usage: bcachefs device add [OPTION]... filesystem device\n"
 	     "\n"
-	     "Options:\n"
-	     "  -S, --fs_size=size          Size of filesystem on device\n"
-	     "  -B, --bucket=size           Bucket size\n"
-	     "  -D, --discard               Enable discards\n"
-	     "  -l, --label=label           Disk label\n"
+	     "Options:\n");
+
+	bch2_opts_usage(OPT_FORMAT|OPT_DEVICE);
+
+	puts("  -l, --label=label           Disk label\n"
 	     "  -f, --force                 Use device even if it appears to already be formatted\n"
 	     "  -h, --help                  Display this help and exit\n"
 	     "\n"
@@ -61,9 +61,6 @@ static void device_add_usage(void)
 int cmd_device_add(int argc, char *argv[])
 {
 	static const struct option longopts[] = {
-		{ "fs_size",		required_argument,	NULL, 'S' },
-		{ "bucket",		required_argument,	NULL, 'B' },
-		{ "discard",		no_argument,		NULL, 'D' },
 		{ "label",		required_argument,	NULL, 'l' },
 		{ "force",		no_argument,		NULL, 'f' },
 		{ "help",		no_argument,		NULL, 'h' },
@@ -72,22 +69,31 @@ int cmd_device_add(int argc, char *argv[])
 	struct format_opts format_opts	= format_opts_default();
 	struct dev_opts dev_opts	= dev_opts_default();
 	bool force = false;
-	int opt;
 
-	while ((opt = getopt_long(argc, argv, "S:B:Dl:fh",
-				  longopts, NULL)) != -1)
-		switch (opt) {
-		case 'S':
-			if (bch2_strtoull_h(optarg, &dev_opts.size))
-				die("invalid filesystem size");
-			break;
-		case 'B':
-			if (bch2_strtoull_h(optarg, &dev_opts.bucket_size))
-				die("bad bucket_size %s", optarg);
-			break;
-		case 'D':
-			dev_opts.discard = true;
+	while (true) {
+		const struct bch_option *opt =
+			bch2_cmdline_opt_parse(argc, argv, OPT_FORMAT|OPT_DEVICE);
+		if (opt) {
+			unsigned id = opt - bch2_opt_table;
+			u64 v;
+			struct printbuf err = PRINTBUF;
+			int ret = bch2_opt_parse(NULL, opt, optarg, &v, &err);
+			if (ret)
+				die("invalid %s: %s", opt->attr.name, err.buf);
+
+			if (opt->flags & OPT_DEVICE)
+				bch2_opt_set_by_id(&dev_opts.opts, id, v);
+			else
+				die("got bch_opt of wrong type %s", opt->attr.name);
+
+			continue;
+		}
+
+		int optid = getopt_long(argc, argv, "S:B:Dl:fh", longopts, NULL);
+		if (optid == -1)
 			break;
+
+		switch (optid) {
 		case 'l':
 			dev_opts.label = strdup(optarg);
 			break;
@@ -97,7 +103,11 @@ int cmd_device_add(int argc, char *argv[])
 		case 'h':
 			device_add_usage();
 			exit(EXIT_SUCCESS);
+		case '?':
+			exit(EXIT_FAILURE);
+			break;
 		}
+}
 	args_shift(optind);
 
 	char *fs_path = arg_pop();
@@ -127,10 +137,11 @@ int cmd_device_add(int argc, char *argv[])
 	opt_set(fs_opts, btree_node_size,
 		read_file_u64(fs.sysfs_fd, "options/btree_node_size"));
 
-	struct bch_sb *sb = bch2_format(fs_opt_strs,
-					fs_opts,
-					format_opts,
-					&dev_opts, 1);
+	dev_opts_list devs = {};
+	darray_push(&devs, dev_opts);
+
+	struct bch_sb *sb = bch2_format(fs_opt_strs, fs_opts, format_opts, devs);
+	darray_exit(&devs);
 	free(sb);
 	bchu_disk_add(fs, dev_opts.path);
 	return 0;

c_src/cmd_format.c

@@ -39,11 +39,7 @@ x('L',	fs_label,		required_argument)	\
 x('U',	uuid,			required_argument)	\
 x(0,	fs_size,		required_argument)	\
 x(0,	superblock_size,	required_argument)	\
-x(0,	bucket_size,		required_argument)	\
 x('l',	label,			required_argument)	\
-x(0,	discard,		no_argument)		\
-x(0,	data_allowed,		required_argument)	\
-x(0,	durability,		required_argument)	\
 x(0,	version,		required_argument)	\
 x(0,	no_initialize,		no_argument)		\
 x(0,	source,			required_argument)	\
@@ -52,17 +48,16 @@ x('q',	quiet,			no_argument)		\
 x('v',	verbose,		no_argument)		\
 x('h',	help,			no_argument)
 
-static void usage(void)
+static void format_usage(void)
 {
 	puts("bcachefs format - create a new bcachefs filesystem on one or more devices\n"
 	     "Usage: bcachefs format [OPTION]... <devices>\n"
 	     "\n"
 	     "Options:");
 
-	bch2_opts_usage(OPT_FORMAT);
+	bch2_opts_usage(OPT_FORMAT|OPT_FS);
 
-	puts(
-	     "      --replicas=#            Sets both data and metadata replicas\n"
+	puts("      --replicas=#            Sets both data and metadata replicas\n"
 	     "      --encrypted             Enable whole filesystem encryption (chacha20/poly1305)\n"
 	     "      --no_passphrase         Don't encrypt master encryption key\n"
 	     "  -L, --fs_label=label\n"
@@ -72,9 +67,10 @@ static void usage(void)
 	     "\n"
 	     "Device specific options:");
 
-	bch2_opts_usage(OPT_DEVICE);
+	bch2_opts_usage(OPT_FORMAT|OPT_DEVICE);
 
-	puts("  -l, --label=label           Disk label\n"
+	puts("      --fs_size=size          Size of filesystem on device\n"
+	     "  -l, --label=label           Disk label\n"
 	     "\n"
 	     "  -f, --force\n"
 	     "  -q, --quiet                 Only print errors\n"
@@ -125,32 +121,60 @@ void build_fs(struct bch_fs *c, const char *src_path)
 	if (!S_ISDIR(stat.st_mode))
 		die("%s is not a directory", src_path);
 
-	copy_fs(c, src_fd, src_path, &s);
+	copy_fs(c, src_fd, src_path, &s, 0);
 }
 
 int cmd_format(int argc, char *argv[])
 {
-	DARRAY(struct dev_opts) devices = { 0 };
-	DARRAY(char *) device_paths = { 0 };
+	dev_opts_list devices = {};
+	darray_str device_paths = {};
 	struct format_opts opts	= format_opts_default();
 	struct dev_opts dev_opts = dev_opts_default();
 	bool force = false, no_passphrase = false, quiet = false, initialize = true, verbose = false;
 	bool unconsumed_dev_option = false;
 	unsigned v;
-	int opt;
 
-	struct bch_opt_strs fs_opt_strs =
-		bch2_cmdline_opts_get(&argc, argv, OPT_FORMAT);
-	struct bch_opts fs_opts = bch2_parse_opts(fs_opt_strs);
+	struct bch_opt_strs fs_opt_strs = {};
+	struct bch_opts fs_opts = bch2_opts_empty();
 
 	if (getenv("BCACHEFS_KERNEL_ONLY"))
 		initialize = false;
 
-	while ((opt = getopt_long(argc, argv,
-				  "-L:l:U:g:fqhv",
-				  format_opts,
-				  NULL)) != -1)
-		switch (opt) {
+	while (true) {
+		const struct bch_option *opt =
+			bch2_cmdline_opt_parse(argc, argv, OPT_FORMAT|OPT_FS|OPT_DEVICE);
+		if (opt) {
+			unsigned id = opt - bch2_opt_table;
+			u64 v;
+			struct printbuf err = PRINTBUF;
+			int ret = bch2_opt_parse(NULL, opt, optarg, &v, &err);
+			if (ret == -BCH_ERR_option_needs_open_fs) {
+				fs_opt_strs.by_id[id] = strdup(optarg);
+				continue;
+			}
+			if (ret)
+				die("invalid option: %s", err.buf);
+
+			if (opt->flags & OPT_DEVICE) {
+				bch2_opt_set_by_id(&dev_opts.opts, id, v);
+				unconsumed_dev_option = true;
+			} else if (opt->flags & OPT_FS) {
+				bch2_opt_set_by_id(&fs_opts, id, v);
+			} else {
+				die("got bch_opt of wrong type %s", opt->attr.name);
+			}
+
+			continue;
+		}
+
+		int optid = getopt_long(argc, argv,
+					"-L:l:U:g:fqhv",
+					format_opts,
+					NULL);
+		if (optid == -1)
+			break;
+
+		switch (optid) {
 		case O_replicas:
 			if (kstrtouint(optarg, 10, &v) ||
 			    !v ||
@@ -183,7 +207,7 @@ int cmd_format(int argc, char *argv[])
 			force = true;
 			break;
 		case O_fs_size:
-			if (bch2_strtoull_h(optarg, &dev_opts.size))
+			if (bch2_strtoull_h(optarg, &dev_opts.fs_size))
 				die("invalid filesystem size");
 			unconsumed_dev_option = true;
 			break;
@@ -193,32 +217,11 @@ int cmd_format(int argc, char *argv[])
 
 			opts.superblock_size >>= 9;
 			break;
-		case O_bucket_size:
-			if (bch2_strtoull_h(optarg, &dev_opts.bucket_size))
-				die("bad bucket_size %s", optarg);
-			unconsumed_dev_option = true;
-			break;
 		case O_label:
 		case 'l':
 			dev_opts.label = optarg;
 			unconsumed_dev_option = true;
 			break;
-		case O_discard:
-			dev_opts.discard = true;
-			unconsumed_dev_option = true;
-			break;
-		case O_data_allowed:
-			dev_opts.data_allowed =
-				read_flag_list_or_die(optarg,
-					__bch2_data_types, "data type");
-			unconsumed_dev_option = true;
-			break;
-		case O_durability:
-			if (kstrtouint(optarg, 10, &dev_opts.durability) ||
-			    dev_opts.durability > BCH_REPLICAS_MAX)
-				die("invalid durability");
-			unconsumed_dev_option = true;
-			break;
 		case O_version:
 			opts.version = version_parse(optarg);
 			break;
@@ -229,7 +232,7 @@ int cmd_format(int argc, char *argv[])
 			darray_push(&device_paths, optarg);
 			dev_opts.path = optarg;
 			darray_push(&devices, dev_opts);
-			dev_opts.size = 0;
+			dev_opts.fs_size = 0;
 			unconsumed_dev_option = false;
 			break;
 		case O_quiet:
@@ -241,13 +244,16 @@ int cmd_format(int argc, char *argv[])
 			break;
 		case O_help:
 		case 'h':
-			usage();
+			format_usage();
 			exit(EXIT_SUCCESS);
 			break;
 		case '?':
 			exit(EXIT_FAILURE);
 			break;
+		default:
+			die("getopt ret %i %c", optid, optid);
 		}
+	}
 
 	if (unconsumed_dev_option)
 		die("Options for devices apply to subsequent devices; got a device option with no device");
@@ -269,11 +275,7 @@ int cmd_format(int argc, char *argv[])
 			die("Error opening %s: %s", dev_opts.path, strerror(-ret));
 	}
 
-	struct bch_sb *sb =
-		bch2_format(fs_opt_strs,
-			    fs_opts,
-			    opts,
-			    devices.data, devices.nr);
+	struct bch_sb *sb = bch2_format(fs_opt_strs, fs_opts, opts, devices);
 	bch2_opt_strs_free(&fs_opt_strs);
 
 	if (!quiet) {
@@ -433,3 +435,204 @@ int cmd_show_super(int argc, char *argv[])
 	printbuf_exit(&buf);
 	return 0;
 }
+
+#include "libbcachefs/super-io.h"
+#include "libbcachefs/sb-members.h"
+
+typedef DARRAY(struct bch_sb *) probed_sb_list;
+
+static void probe_one_super(int dev_fd, unsigned sb_size, u64 offset,
+			    probed_sb_list *sbs, bool verbose)
+{
+	darray_char sb_buf = {};
+	darray_resize(&sb_buf, sb_size);
+
+	xpread(dev_fd, sb_buf.data, sb_buf.size, offset);
+
+	struct printbuf err = PRINTBUF;
+	int ret = bch2_sb_validate((void *) sb_buf.data, offset >> 9, 0, &err);
+	printbuf_exit(&err);
+
+	if (!ret) {
+		if (verbose) {
+			struct printbuf buf = PRINTBUF;
+			prt_human_readable_u64(&buf, offset);
+			printf("found superblock at %s\n", buf.buf);
+			printbuf_exit(&buf);
+		}
+
+		darray_push(sbs, (void *) sb_buf.data);
+		sb_buf.data = NULL;
+	}
+
+	darray_exit(&sb_buf);
+}
+
+static void probe_sb_range(int dev_fd, u64 start_offset, u64 end_offset,
+			   probed_sb_list *sbs, bool verbose)
+{
+	start_offset	&= ~((u64) 511);
+	end_offset	&= ~((u64) 511);
+
+	size_t buflen = end_offset - start_offset;
+	void *buf = malloc(buflen);
+	xpread(dev_fd, buf, buflen, start_offset);
+
+	for (u64 offset = 0; offset < buflen; offset += 512) {
+		struct bch_sb *sb = buf + offset;
+
+		if (!uuid_equal(&sb->magic, &BCACHE_MAGIC) &&
+		    !uuid_equal(&sb->magic, &BCHFS_MAGIC))
+			continue;
+
+		size_t bytes = vstruct_bytes(sb);
+		if (offset + bytes > buflen) {
+			fprintf(stderr, "found sb %llu size %zu that overran buffer\n",
+				start_offset + offset, bytes);
+			continue;
+		}
+		struct printbuf err = PRINTBUF;
+		int ret = bch2_sb_validate(sb, (start_offset + offset) >> 9, 0, &err);
+		if (ret)
+			fprintf(stderr, "found sb %llu that failed to validate: %s\n",
+				start_offset + offset, err.buf);
+		printbuf_exit(&err);
+
+		if (ret)
+			continue;
+
+		if (verbose) {
+			struct printbuf buf = PRINTBUF;
+			prt_human_readable_u64(&buf, start_offset + offset);
+			printf("found superblock at %s\n", buf.buf);
+			printbuf_exit(&buf);
+		}
+
+		void *sb_copy = malloc(bytes);
+		memcpy(sb_copy, sb, bytes);
+		darray_push(sbs, sb_copy);
+	}
+
+	free(buf);
+}
+
+static u64 bch2_sb_last_mount_time(struct bch_sb *sb)
+{
+	u64 ret = 0;
+	for (unsigned i = 0; i < sb->nr_devices; i++)
+		ret = max(ret, le64_to_cpu(bch2_sb_member_get(sb, i).last_mount));
+	return ret;
+}
+
+static int bch2_sb_time_cmp(struct bch_sb *l, struct bch_sb *r)
+{
+	return cmp_int(bch2_sb_last_mount_time(l),
+		       bch2_sb_last_mount_time(r));
+}
+
+static void recover_super_usage(void)
+{
+	puts("bcachefs recover-super \n"
+	     "Usage: bcachefs recover-super [OPTION].. device\n"
+	     "\n"
+	     "Attempt to recover a filesystem on a device that has had the main superblock\n"
+	     "and superblock layout overwritten.\n"
+	     "All options will be guessed if not provided\n"
+	     "\n"
+	     "Options:\n"
+	     "  -d, --dev_size              size of filessytem on device, in bytes \n"
+	     "  -o, --offset                offset to probe, in bytes\n"
+	     "  -y, --yes                   Recover without prompting\n"
+	     "  -v, --verbose               Increase logging level\n"
+	     "  -h, --help                  display this help and exit\n"
+	     "Report bugs to <linux-bcachefs@vger.kernel.org>");
+	exit(EXIT_SUCCESS);
+}
+
+int cmd_recover_super(int argc, char *argv[])
+{
+	static const struct option longopts[] = {
+		{ "dev_size",			1, NULL, 'd' },
+		{ "offset",			1, NULL, 'o' },
+		{ "yes",			0, NULL, 'y' },
+		{ "verbose",			0, NULL, 'v' },
+		{ "help",			0, NULL, 'h' },
+		{ NULL }
+	};
+	u64 dev_size = 0, offset = 0;
+	bool yes = false, verbose = false;
+	int opt;
+
+	while ((opt = getopt_long(argc, argv, "d:o:yvh", longopts, NULL)) != -1)
+		switch (opt) {
+		case 'd':
+			if (bch2_strtoull_h(optarg, &dev_size))
+				die("invalid offset");
+			break;
+		case 'o':
+			if (bch2_strtoull_h(optarg, &offset))
+				die("invalid offset");
+
+			if (offset & 511)
+				die("offset must be a multiple of 512");
+			break;
+		case 'y':
+			yes = true;
+			break;
+		case 'v':
+			verbose = true;
+			break;
+		case 'h':
+			recover_super_usage();
+			break;
+		}
+	args_shift(optind);
+
+	char *dev_path = arg_pop();
+	if (!dev_path)
+		die("please supply a device");
+	if (argc)
+		die("too many arguments");
+
+	int dev_fd = xopen(dev_path, O_RDWR);
+
+	if (!dev_size)
+		dev_size = get_size(dev_fd);
+
+	probed_sb_list sbs = {};
+
+	if (offset) {
+		probe_one_super(dev_fd, SUPERBLOCK_SIZE_DEFAULT, offset, &sbs, verbose);
+	} else {
+		unsigned scan_len = 16 << 20; /* 16MB, start and end of device */
+
+		probe_sb_range(dev_fd, 4096, scan_len, &sbs, verbose);
+		probe_sb_range(dev_fd, dev_size - scan_len, dev_size, &sbs, verbose);
+	}
+
+	if (!sbs.nr) {
+		printf("Found no bcachefs superblocks\n");
+		exit(EXIT_FAILURE);
+	}
+
+	struct bch_sb *best = NULL;
+	darray_for_each(sbs, sb)
+		if (!best || bch2_sb_time_cmp(best, *sb) < 0)
+			best = *sb;
+
+	struct printbuf buf = PRINTBUF;
+	bch2_sb_to_text(&buf, best, true, BIT_ULL(BCH_SB_FIELD_members_v2));
+
+	printf("Found superblock:\n%s", buf.buf);
+	printf("Recover?");
+
+	if (yes || ask_yn())
+		bch2_super_write(dev_fd, best);
+
+	printbuf_exit(&buf);
+	darray_for_each(sbs, sb)
+		kfree(*sb);
+	darray_exit(&sbs);
+
+	return 0;
+}

c_src/cmd_fs.c

@@ -237,6 +237,18 @@ static void accounting_sort(darray_accounting_p *sorted,
 	sort(sorted->data, sorted->nr, sizeof(sorted->data[0]), accounting_p_cmp, NULL);
 }
 
+static void accounting_swab_if_old(struct bch_ioctl_query_accounting *in)
+{
+	unsigned kernel_version = bcachefs_kernel_version();
+
+	if (kernel_version &&
+	    kernel_version < bcachefs_metadata_version_disk_accounting_big_endian)
+		for (struct bkey_i_accounting *a = in->accounting;
+		     a < (struct bkey_i_accounting *) ((u64 *) in->accounting + in->accounting_u64s);
+		     a = bkey_i_to_accounting(bkey_next(&a->k_i)))
+			bch2_bpos_swab(&a->k.p);
+}
+
 static int fs_usage_v1_to_text(struct printbuf *out,
 			       struct bchfs_handle fs,
 			       dev_names dev_names)
@@ -251,6 +263,8 @@ static int fs_usage_v1_to_text(struct printbuf *out,
 	if (!a)
 		return -1;
 
+	accounting_swab_if_old(a);
+
 	darray_accounting_p a_sorted = {};
 
 	accounting_sort(&a_sorted, a);
@@ -477,6 +491,19 @@ devs:
 	bcache_fs_close(fs);
 }
 
+int fs_usage(void)
+{
+	puts("bcachefs fs - manage a running filesystem\n"
+	     "Usage: bcachefs fs <CMD> [OPTIONS]\n"
+	     "\n"
+	     "Commands:\n"
+	     "  usage                   Display detailed filesystem usage\n"
+	     "  top                     Show runtime performance information\n"
+	     "\n"
+	     "Report bugs to <linux-bcachefs@vger.kernel.org>");
+	return 0;
+}
+
 static void fs_usage_usage(void)
 {
 	puts("bcachefs fs usage - display detailed filesystem usage\n"

c_src/cmd_fsck.c

@@ -21,7 +21,6 @@ static void fsck_usage(void)
 	     "  -y                      Assume \"yes\" to all questions\n"
 	     "  -f                      Force checking even if filesystem is marked clean\n"
 	     "  -r, --ratelimit_errors  Don't display more than 10 errors of a given type\n"
-	     "  -R, --reconstruct_alloc Reconstruct the alloc btree\n"
 	     "  -k, --kernel            Use the in-kernel fsck implementation\n"
 	     "  -v                      Be verbose\n"
 	     "  -h, --help              Display this help and exit\n"
@@ -117,9 +116,7 @@ static bool should_use_kernel_fsck(darray_str devs)
 {
 	system("modprobe bcachefs");
 
-	unsigned kernel_version = !access("/sys/module/bcachefs/parameters/version", R_OK)
-	    ? read_file_u64(AT_FDCWD, "/sys/module/bcachefs/parameters/version")
-	    : 0;
+	unsigned kernel_version = bcachefs_kernel_version();
 
 	if (!kernel_version)
 		return false;
@@ -205,7 +202,6 @@ int cmd_fsck(int argc, char *argv[])
 {
 	static const struct option longopts[] = {
 		{ "ratelimit_errors",	no_argument,		NULL, 'r' },
-		{ "reconstruct_alloc",	no_argument,		NULL, 'R' },
 		{ "kernel",		no_argument,		NULL, 'k' },
 		{ "no-kernel",		no_argument,		NULL, 'K' },
 		{ "help",		no_argument,		NULL, 'h' },
@@ -224,10 +220,13 @@ int cmd_fsck(int argc, char *argv[])
 	append_opt(&opts_str, "read_only");
 
 	while ((opt = getopt_long(argc, argv,
-				  "apynfo:rRkKvh",
+				  "apynfo:rkKvh",
 				  longopts, NULL)) != -1)
 		switch (opt) {
-		case 'a': /* outdated alias for -p */
+		case 'a':
+			/* "automatic" run, called by the system, for us to do checks as needed.
+			 *  we don't need checks here: */
+			exit(EXIT_SUCCESS);
 		case 'p':
 		case 'y':
 			append_opt(&opts_str, "fix_errors=yes");
@@ -245,9 +244,6 @@ int cmd_fsck(int argc, char *argv[])
 		case 'r':
 			append_opt(&opts_str, "ratelimit_errors");
 			break;
-		case 'R':
-			append_opt(&opts_str, "reconstruct_alloc");
-			break;
 		case 'k':
 			kernel = true;
 			break;
@@ -323,7 +319,7 @@ kernel_fsck_err:
 	} else {
 userland_fsck:
 		printf("Running userspace offline fsck\n");
-		ret = bch2_parse_mount_opts(NULL, &opts, &parse_later, opts_str.buf);
+		ret = bch2_parse_mount_opts(NULL, &opts, &parse_later, opts_str.buf, false);
 		if (ret)
 			return ret;
 

c_src/cmd_fusemount.c

@@ -19,7 +19,7 @@
 #include "libbcachefs/dirent.h"
 #include "libbcachefs/errcode.h"
 #include "libbcachefs/error.h"
-#include "libbcachefs/fs-common.h"
+#include "libbcachefs/namei.h"
 #include "libbcachefs/inode.h"
 #include "libbcachefs/io_read.h"
 #include "libbcachefs/io_write.h"
@@ -31,9 +31,6 @@
 
 #include <linux/dcache.h>
 
-/* XXX cut and pasted from fsck.c */
-#define QSTR(n) { { { .len = strlen(n) } }, .name = n }
-
 /* used by write_aligned function for waiting on bch2_write closure */
 struct write_aligned_op_t {
         struct closure cl;
@@ -478,10 +475,9 @@ static int read_aligned(struct bch_fs *c, subvol_inum inum, size_t aligned_size,
 	closure_init_stack(&cl);
 
 	closure_get(&cl);
-	rbio.bio.bi_end_io		= bcachefs_fuse_read_endio;
-	rbio.bio.bi_private		= &cl;
+	rbio.bio.bi_private = &cl;
 
-	bch2_read(c, rbio_init(&rbio.bio, io_opts), inum);
+	bch2_read(c, rbio_init(&rbio.bio, c, io_opts, bcachefs_fuse_read_endio), inum);
 
 	closure_sync(&cl);
 

c_src/cmd_list_journal.c

@@ -65,9 +65,21 @@ static bool bkey_matches_filter(d_bbpos_range filter, struct jset_entry *entry,
 		struct bbpos k_start	= BBPOS(entry->btree_id, bkey_start_pos(&k->k));
 		struct bbpos k_end	= BBPOS(entry->btree_id, k->k.p);
 
-		if (bbpos_cmp(k_start, i->start) >= 0 &&
-		    bbpos_cmp(k_end, i->end) <= 0)
-			return true;
+		if (!i->start.pos.snapshot &&
+		    !i->end.pos.snapshot) {
+			k_start.pos.snapshot = 0;
+			k_end.pos.snapshot = 0;
+		}
+
+		if (!k->k.size) {
+			if (bbpos_cmp(k_start, i->start) >= 0 &&
+			    bbpos_cmp(k_end, i->end) <= 0)
+				return true;
+		} else {
+			if (bbpos_cmp(i->start, k_end) <= 0 &&
+			    bbpos_cmp(i->end, k_start) >= 0)
+				return true;
+		}
 	}
 	return false;
 }
@@ -75,9 +87,9 @@ static bool bkey_matches_filter(d_bbpos_range filter, struct jset_entry *entry,
 static bool entry_matches_transaction_filter(struct jset_entry *entry,
 					     d_bbpos_range filter)
 {
-	if (entry->type == BCH_JSET_ENTRY_btree_root ||
-	    entry->type == BCH_JSET_ENTRY_btree_keys ||
-	    entry->type == BCH_JSET_ENTRY_overwrite)
+	if (!entry->level &&
+	    (entry->type == BCH_JSET_ENTRY_btree_keys ||
+	     entry->type == BCH_JSET_ENTRY_overwrite))
 		jset_entry_for_each_key(entry, k)
 			if (bkey_matches_filter(filter, entry, k))
 				return true;
@@ -90,6 +102,8 @@ static bool should_print_transaction(struct jset_entry *entry, struct jset_entry
 {
 	struct jset_entry_log *l = container_of(entry, struct jset_entry_log, entry);
 	unsigned b = jset_entry_log_msg_bytes(l);
+	bool have_log_messages = false;
+	bool have_non_log_messages = false;
 
 	darray_for_each(msg_filter, i)
 		if (!strncmp(*i, l->d, b))
@@ -100,11 +114,19 @@ static bool should_print_transaction(struct jset_entry *entry, struct jset_entry
 
 	for (entry = vstruct_next(entry);
 	     entry != end && !entry_is_transaction_start(entry);
-	     entry = vstruct_next(entry))
-		if (entry_is_log_msg(entry) ||
-		    entry_matches_transaction_filter(entry, key_filter))
+	     entry = vstruct_next(entry)) {
+		if (entry_matches_transaction_filter(entry, key_filter))
 			return true;
 
+		if (entry_is_log_msg(entry))
+			have_log_messages = true;
+		else
+			have_non_log_messages = true;
+	}
+
+	if (have_log_messages && !have_non_log_messages)
+		return true;
+
 	return false;
 }
 
@@ -134,6 +156,7 @@ static void journal_entry_header_to_text(struct printbuf *out,
 		prt_str(out, "blacklisted ");
 
 	prt_printf(out,
+		   "\n"
 		   "journal entry     %llu\n"
 		   "  version         %u\n"
 		   "  last seq        %llu\n"

c_src/cmd_migrate.c

@@ -159,9 +159,9 @@ static void find_superblock_space(ranges extents,
 {
 	darray_for_each(extents, i) {
 		u64 start = round_up(max(256ULL << 10, i->start),
-				     dev->bucket_size << 9);
+				     dev->opts.bucket_size << 9);
 		u64 end = round_down(i->end,
-				     dev->bucket_size << 9);
+				     dev->opts.bucket_size << 9);
 
 		/* Need space for two superblocks: */
 		if (start + (opts.superblock_size << 9) * 2 <= end) {
@@ -209,38 +209,43 @@ static int migrate_fs(const char		*fs_path,
 	if (!S_ISDIR(stat.st_mode))
 		die("%s is not a directory", fs_path);
 
-	struct dev_opts dev = dev_opts_default();
+	dev_opts_list devs = {};
+	darray_push(&devs, dev_opts_default());
 
-	dev.path = dev_t_to_path(stat.st_dev);
-	dev.file = bdev_file_open_by_path(dev.path, BLK_OPEN_READ|BLK_OPEN_WRITE, &dev, NULL);
+	struct dev_opts *dev = &devs.data[0];
 
-	int ret = PTR_ERR_OR_ZERO(dev.file);
+	dev->path = dev_t_to_path(stat.st_dev);
+	dev->file = bdev_file_open_by_path(dev->path, BLK_OPEN_READ|BLK_OPEN_WRITE, dev, NULL);
+
+	int ret = PTR_ERR_OR_ZERO(dev->file);
 	if (ret < 0)
-		die("Error opening device to format %s: %s", dev.path, strerror(-ret));
-	dev.bdev = file_bdev(dev.file);
+		die("Error opening device to format %s: %s", dev->path, strerror(-ret));
+	dev->bdev = file_bdev(dev->file);
 
-	opt_set(fs_opts, block_size, get_blocksize(dev.bdev->bd_fd));
+	opt_set(fs_opts, block_size, get_blocksize(dev->bdev->bd_fd));
 
 	char *file_path = mprintf("%s/bcachefs", fs_path);
 	printf("Creating new filesystem on %s in space reserved at %s\n",
-	       dev.path, file_path);
+	       dev->path, file_path);
 
-	dev.size	= get_size(dev.bdev->bd_fd);
-	dev.bucket_size = bch2_pick_bucket_size(fs_opts, &dev);
-	dev.nbuckets	= dev.size / dev.bucket_size;
+	dev->fs_size		= get_size(dev->bdev->bd_fd);
+	opt_set(dev->opts, bucket_size, bch2_pick_bucket_size(fs_opts, devs));
 
-	bch2_check_bucket_size(fs_opts, &dev);
+	dev->nbuckets		= dev->fs_size / dev->opts.bucket_size;
+
+	bch2_check_bucket_size(fs_opts, dev);
 
 	u64 bcachefs_inum;
 	ranges extents = reserve_new_fs_space(file_path,
 				fs_opts.block_size >> 9,
-				get_size(dev.bdev->bd_fd) / 5,
+				get_size(dev->bdev->bd_fd) / 5,
 				&bcachefs_inum, stat.st_dev, force);
 
-	find_superblock_space(extents, format_opts, &dev);
+	find_superblock_space(extents, format_opts, dev);
+
+	struct bch_sb *sb = bch2_format(fs_opt_strs, fs_opts, format_opts, devs);
+	darray_exit(&devs);
 
-	struct bch_sb *sb = bch2_format(fs_opt_strs,
-					fs_opts, format_opts, &dev, 1);
 	u64 sb_offset = le64_to_cpu(sb->layout.sb_offset[0]);
 
 	if (format_opts.passphrase)
@@ -248,16 +253,14 @@ static int migrate_fs(const char		*fs_path,
 
 	free(sb);
 
-	struct bch_opts opts = bch2_opts_empty();
-	struct bch_fs *c = NULL;
-	char *path[1] = { dev.path };
+	char *path[1] = { dev->path };
 
+	struct bch_opts opts = bch2_opts_empty();
 	opt_set(opts, sb,	sb_offset);
 	opt_set(opts, nostart,	true);
 	opt_set(opts, noexcl,	true);
-	opt_set(opts, nostart, true);
 
-	c = bch2_fs_open(path, 1, opts);
+	struct bch_fs *c = bch2_fs_open(path, 1, opts);
 	if (IS_ERR(c))
 		die("Error opening new filesystem: %s", bch2_err_str(PTR_ERR(c)));
 
@@ -265,10 +268,6 @@ static int migrate_fs(const char		*fs_path,
 	if (ret)
 		die("Error allocating buckets_nouse: %s", bch2_err_str(ret));
 
-	ret = bch2_fs_start(c);
-	if (IS_ERR(c))
-		die("Error starting new filesystem: %s", bch2_err_str(ret));
-
 	mark_unreserved_space(c, extents);
 
 	ret = bch2_fs_start(c);
@@ -282,7 +281,10 @@ static int migrate_fs(const char		*fs_path,
 		.type		= BCH_MIGRATE_migrate,
 	};
 
-	copy_fs(c, fs_fd, fs_path, &s);
+	u64 reserve_start = round_up((format_opts.superblock_size * 2 + 8) << 9,
+				     dev->opts.bucket_size);
+
+	copy_fs(c, fs_fd, fs_path, &s, reserve_start);
 
 	bch2_fs_stop(c);
 
@@ -310,7 +312,7 @@ static int migrate_fs(const char		*fs_path,
 	       "filesystem. That file can be deleted once the old filesystem is\n"
 	       "no longer needed (and should be deleted prior to running\n"
 	       "bcachefs migrate-superblock)\n",
-	       sb_offset, dev.path, dev.path, sb_offset);
+	       sb_offset, dev->path, dev->path, sb_offset);
 	return 0;
 }
 
@@ -374,7 +376,7 @@ static void migrate_superblock_usage(void)
 int cmd_migrate_superblock(int argc, char *argv[])
 {
 	char *dev = NULL;
-	u64 offset = 0;
+	u64 sb_offset = 0;
 	int opt, ret;
 
 	while ((opt = getopt(argc, argv, "d:o:h")) != -1)
@@ -383,7 +385,7 @@ int cmd_migrate_superblock(int argc, char *argv[])
 				dev = optarg;
 				break;
 			case 'o':
-				ret = kstrtou64(optarg, 10, &offset);
+				ret = kstrtou64(optarg, 10, &sb_offset);
 				if (ret)
 					die("Invalid offset");
 				break;
@@ -395,29 +397,72 @@ int cmd_migrate_superblock(int argc, char *argv[])
 	if (!dev)
 		die("Please specify a device");
 
-	if (!offset)
+	if (!sb_offset)
 		die("Please specify offset of existing superblock");
 
 	int fd = xopen(dev, O_RDWR);
-	struct bch_sb *sb = __bch2_super_read(fd, offset);
+	struct bch_sb *sb = __bch2_super_read(fd, sb_offset);
+	unsigned sb_size = 1U << sb->layout.sb_max_size_bits;
 
 	if (sb->layout.nr_superblocks >= ARRAY_SIZE(sb->layout.sb_offset))
 		die("Can't add superblock: no space left in superblock layout");
 
-	unsigned i;
-	for (i = 0; i < sb->layout.nr_superblocks; i++)
-		if (le64_to_cpu(sb->layout.sb_offset[i]) == BCH_SB_SECTOR)
-			die("Superblock layout already has default superblock");
+	for (unsigned i = 0; i < sb->layout.nr_superblocks; i++)
+		if (le64_to_cpu(sb->layout.sb_offset[i]) == BCH_SB_SECTOR ||
+		    le64_to_cpu(sb->layout.sb_offset[i]) == BCH_SB_SECTOR + sb_size)
+			die("Superblock layout already has default superblocks");
 
-	memmove(&sb->layout.sb_offset[1],
+	memmove(&sb->layout.sb_offset[2],
 		&sb->layout.sb_offset[0],
 		sb->layout.nr_superblocks * sizeof(u64));
-	sb->layout.nr_superblocks++;
-
+	sb->layout.nr_superblocks += 2;
 	sb->layout.sb_offset[0] = cpu_to_le64(BCH_SB_SECTOR);
+	sb->layout.sb_offset[1] = cpu_to_le64(BCH_SB_SECTOR + sb_size);
+
+	/* also write first 0-3.5k bytes with zeroes, ensure we blow away old
+	 * superblock */
+	static const char zeroes[BCH_SB_SECTOR << 9];
+	xpwrite(fd, zeroes, BCH_SB_SECTOR << 9, 0, "zeroing start of disk");
 
 	bch2_super_write(fd, sb);
 	close(fd);
 
+	/* mark new superblocks */
+
+	struct bch_opts opts = bch2_opts_empty();
+	opt_set(opts, nostart,	true);
+	opt_set(opts, sb,	sb_offset);
+
+	struct bch_fs *c = bch2_fs_open(&dev, 1, opts);
+	ret =   PTR_ERR_OR_ZERO(c) ?:
+		bch2_buckets_nouse_alloc(c);
+	if (ret)
+		die("error opening filesystem: %s", bch2_err_str(ret));
+
+	struct bch_dev *ca = c->devs[0];
+	for (u64 b = 0; bucket_to_sector(ca, b) < BCH_SB_SECTOR + sb_size * 2; b++)
+		set_bit(b, ca->buckets_nouse);
+
+	ret = bch2_fs_start(c);
+	if (ret)
+		die("Error starting filesystem: %s", bch2_err_str(ret));
+
+	bch2_fs_stop(c);
+
+	opts = bch2_opts_empty();
+	opt_set(opts, fsck, true);
+	opt_set(opts, fix_errors, true);
+
+	/*
+	 * Hack: the free space counters are coming out wrong after marking the
+	 * new superblock, but it's just the device counters so it's
+	 * inconsequential:
+	 */
+
+	c = bch2_fs_open(&dev, 1, opts);
+	ret =   PTR_ERR_OR_ZERO(c);
+	if (ret)
+		die("error opening filesystem: %s", bch2_err_str(ret));
+	bch2_fs_stop(c);
 	return 0;
 }

c_src/cmd_option.c

@@ -30,7 +30,7 @@ static void set_option_usage(void)
 	     "Usage: bcachefs set-fs-option [OPTION].. device\n"
 	     "\n"
 	     "Options:\n");
-	bch2_opts_usage(OPT_MOUNT);
+	bch2_opts_usage(OPT_MOUNT|OPT_RUNTIME);
 	puts("  -d, --dev-idx               index for device specific options\n"
 	     "  -h, --help                  display this help and exit\n"
 	     "Report bugs to <linux-bcachefs@vger.kernel.org>");
@@ -111,16 +111,16 @@ int cmd_set_option(int argc, char *argv[])
 			if (!bch2_opt_defined_by_id(&new_opts, i))
 				continue;
 
-			ret = bch2_opt_check_may_set(c, i, v);
-			if (ret < 0) {
-				fprintf(stderr, "error setting %s: %i\n", opt->attr.name, ret);
-				continue;
-			}
-
 			if (!(opt->flags & (OPT_FS|OPT_DEVICE)))
 				fprintf(stderr, "Can't set option %s\n", opt->attr.name);
 
 			if (opt->flags & OPT_FS) {
+				ret = bch2_opt_check_may_set(c, NULL, i, v);
+				if (ret < 0) {
+					fprintf(stderr, "error setting %s: %i\n", opt->attr.name, ret);
+					continue;
+				}
+
 				bch2_opt_set_sb(c, NULL, opt, v);
 			}
 
@@ -133,6 +133,12 @@ int cmd_set_option(int argc, char *argv[])
 							continue;
 						}
 
+						ret = bch2_opt_check_may_set(c, ca, i, v);
+						if (ret < 0) {
+							fprintf(stderr, "error setting %s: %i\n", opt->attr.name, ret);
+							continue;
+						}
+
 						bch2_opt_set_sb(c, ca, opt, v);
 						bch2_dev_put(ca);
 					}

c_src/cmds.h

@@ -11,9 +11,11 @@
 
 int cmd_format(int argc, char *argv[]);
 int cmd_show_super(int argc, char *argv[]);
+int cmd_recover_super(int argc, char *argv[]);
 int cmd_reset_counters(int argc, char *argv[]);
 int cmd_set_option(int argc, char *argv[]);
 
+int fs_usage(void);
 int cmd_fs_usage(int argc, char *argv[]);
 int cmd_fs_top(int argc, char *argv[]);
 

c_src/libbcachefs.c

@@ -16,6 +16,8 @@
 
 #include <uuid/uuid.h>
 
+#include <linux/mm.h>
+
 #include "libbcachefs.h"
 #include "crypto.h"
 #include "libbcachefs/bcachefs_format.h"
@@ -31,10 +33,10 @@
 
 #define NSEC_PER_SEC	1000000000L
 
-static void init_layout(struct bch_sb_layout *l,
-			unsigned block_size,
-			unsigned sb_size,
-			u64 sb_start, u64 sb_end)
+void bch2_sb_layout_init(struct bch_sb_layout *l,
+			 unsigned block_size,
+			 unsigned sb_size,
+			 u64 sb_start, u64 sb_end)
 {
 	u64 sb_pos = sb_start;
 	unsigned i;
@@ -60,83 +62,86 @@ static void init_layout(struct bch_sb_layout *l,
 		    sb_start, sb_pos, sb_end, sb_size);
 }
 
-/* minimum size filesystem we can create, given a bucket size: */
-static u64 min_size(unsigned bucket_size)
+static u64 dev_max_bucket_size(u64 dev_size)
 {
-	return BCH_MIN_NR_NBUCKETS * bucket_size;
+	return rounddown_pow_of_two(dev_size / (BCH_MIN_NR_NBUCKETS * 4));
 }
 
-u64 bch2_pick_bucket_size(struct bch_opts opts, struct dev_opts *dev)
+u64 bch2_pick_bucket_size(struct bch_opts opts, dev_opts_list devs)
 {
-	u64 bucket_size;
-
-	if (dev->size < min_size(opts.block_size))
-		die("cannot format %s, too small (%llu bytes, min %llu)",
-		    dev->path, dev->size, min_size(opts.block_size));
-
 	/* Bucket size must be >= block size: */
-	bucket_size = opts.block_size;
+	u64 bucket_size = opts.block_size;
 
 	/* Bucket size must be >= btree node size: */
 	if (opt_defined(opts, btree_node_size))
-		bucket_size = max_t(unsigned, bucket_size,
-					 opts.btree_node_size);
+		bucket_size = max_t(u64, bucket_size, opts.btree_node_size);
 
-	/* Want a bucket size of at least 128k, if possible: */
-	bucket_size = max(bucket_size, 128ULL << 10);
+	u64 min_dev_size = BCH_MIN_NR_NBUCKETS * bucket_size;
+	darray_for_each(devs, i)
+		if (i->fs_size < min_dev_size)
+			die("cannot format %s, too small (%llu bytes, min %llu)",
+			    i->path, i->fs_size, min_dev_size);
 
-	if (dev->size >= min_size(bucket_size)) {
-		unsigned scale = max(1,
-			ilog2(dev->size / min_size(bucket_size)) / 4);
+	u64 total_fs_size = 0;
+	darray_for_each(devs, i)
+		total_fs_size += i->fs_size;
 
-		scale = rounddown_pow_of_two(scale);
+	struct sysinfo info;
+	si_meminfo(&info);
 
-		/* max bucket size 1 mb */
-		bucket_size = min(bucket_size * scale, 1ULL << 20);
-	} else {
-		do {
-			bucket_size /= 2;
-		} while (dev->size < min_size(bucket_size));
-	}
+	/*
+	 * Large fudge factor to allow for other fsck processes and devices
+	 * being added after creation
+	 */
+	u64 mem_available_for_fsck = info.totalram / 8;
+	u64 buckets_can_fsck = mem_available_for_fsck / (sizeof(struct bucket) * 1.5);
+	u64 mem_lower_bound = roundup_pow_of_two(total_fs_size / buckets_can_fsck);
+
+	/*
+	 * Lower bound to avoid fragmenting encoded (checksummed, compressed)
+	 * extents too much as they're moved:
+	 */
+	bucket_size = max(bucket_size, opt_get(opts, encoded_extent_max) * 4);
+
+	/* Lower bound to ensure we can fsck: */
+	bucket_size = max(bucket_size, mem_lower_bound);
+
+	u64 perf_lower_bound = min(2ULL << 20, total_fs_size / (1ULL << 20));
+
+	/* We also prefer larger buckets for performance, up to 2MB at 2T */
+	bucket_size = max(bucket_size, perf_lower_bound);
 
 	return bucket_size;
 }
 
 void bch2_check_bucket_size(struct bch_opts opts, struct dev_opts *dev)
 {
-	if (dev->bucket_size < opts.block_size)
-		die("Bucket size (%llu) cannot be smaller than block size (%u)",
-		    dev->bucket_size, opts.block_size);
+	if (dev->opts.bucket_size < opts.block_size)
+		die("Bucket size (%u) cannot be smaller than block size (%u)",
+		    dev->opts.bucket_size, opts.block_size);
 
 	if (opt_defined(opts, btree_node_size) &&
-	    dev->bucket_size < opts.btree_node_size)
-		die("Bucket size (%llu) cannot be smaller than btree node size (%u)",
-		    dev->bucket_size, opts.btree_node_size);
+	    dev->opts.bucket_size < opts.btree_node_size)
+		die("Bucket size (%u) cannot be smaller than btree node size (%u)",
+		    dev->opts.bucket_size, opts.btree_node_size);
 
 	if (dev->nbuckets < BCH_MIN_NR_NBUCKETS)
-		die("Not enough buckets: %llu, need %u (bucket size %llu)",
-		    dev->nbuckets, BCH_MIN_NR_NBUCKETS, dev->bucket_size);
-
-	if (dev->bucket_size > (u32) U16_MAX << 9)
-		die("Bucket size (%llu) too big (max %u)",
-		    dev->bucket_size, (u32) U16_MAX << 9);
+		die("Not enough buckets: %llu, need %u (bucket size %u)",
+		    dev->nbuckets, BCH_MIN_NR_NBUCKETS, dev->opts.bucket_size);
 }
 
 static unsigned parse_target(struct bch_sb_handle *sb,
-			     struct dev_opts *devs, size_t nr_devs,
+			     dev_opts_list devs,
 			     const char *s)
 {
-	struct dev_opts *i;
-	int idx;
-
 	if (!s)
 		return 0;
 
-	for (i = devs; i < devs + nr_devs; i++)
+	darray_for_each(devs, i)
 		if (!strcmp(s, i->path))
-			return dev_to_target(i - devs);
+			return dev_to_target(i - devs.data);
 
-	idx = bch2_disk_path_find(sb, s);
+	int idx = bch2_disk_path_find(sb, s);
 	if (idx >= 0)
 		return group_to_target(idx);
 
@@ -144,56 +149,61 @@ static unsigned parse_target(struct bch_sb_handle *sb,
 	return 0;
 }
 
+static void bch2_opt_set_sb_all(struct bch_sb *sb, int dev_idx, struct bch_opts *opts)
+{
+	for (unsigned id = 0; id < bch2_opts_nr; id++) {
+		u64 v = bch2_opt_defined_by_id(opts, id)
+			? bch2_opt_get_by_id(opts, id)
+			: bch2_opt_get_by_id(&bch2_opts_default, id);
+
+		__bch2_opt_set_sb(sb, dev_idx, &bch2_opt_table[id], v);
+	}
+}
+
 struct bch_sb *bch2_format(struct bch_opt_strs	fs_opt_strs,
 			   struct bch_opts	fs_opts,
 			   struct format_opts	opts,
-			   struct dev_opts	*devs,
-			   size_t		nr_devs)
+			   dev_opts_list devs)
 {
 	struct bch_sb_handle sb = { NULL };
-	struct dev_opts *i;
 	unsigned max_dev_block_size = 0;
-	unsigned opt_id;
-	u64 min_bucket_size = U64_MAX;
 
-	for (i = devs; i < devs + nr_devs; i++)
+	darray_for_each(devs, i)
 		max_dev_block_size = max(max_dev_block_size, get_blocksize(i->bdev->bd_fd));
 
 	/* calculate block size: */
-	if (!opt_defined(fs_opts, block_size)) {
+	if (!opt_defined(fs_opts, block_size))
 		opt_set(fs_opts, block_size, max_dev_block_size);
-	} else if (fs_opts.block_size < max_dev_block_size)
+
+	if (fs_opts.block_size < max_dev_block_size)
 		die("blocksize too small: %u, must be greater than device blocksize %u",
 		    fs_opts.block_size, max_dev_block_size);
 
 	/* get device size, if it wasn't specified: */
-	for (i = devs; i < devs + nr_devs; i++)
-		if (!i->size)
-			i->size = get_size(i->bdev->bd_fd);
+	darray_for_each(devs, i)
+		if (!i->fs_size)
+			i->fs_size = get_size(i->bdev->bd_fd);
 
 	/* calculate bucket sizes: */
-	for (i = devs; i < devs + nr_devs; i++)
-		min_bucket_size = min(min_bucket_size,
-			i->bucket_size ?: bch2_pick_bucket_size(fs_opts, i));
+	u64 fs_bucket_size = bch2_pick_bucket_size(fs_opts, devs);
 
-	for (i = devs; i < devs + nr_devs; i++)
-		if (!i->bucket_size)
-			i->bucket_size = min_bucket_size;
+	darray_for_each(devs, i)
+		if (!opt_defined(i->opts, bucket_size))
+			opt_set(i->opts, bucket_size,
+				min(fs_bucket_size, dev_max_bucket_size(i->fs_size)));
 
-	for (i = devs; i < devs + nr_devs; i++) {
-		i->nbuckets = i->size / i->bucket_size;
+	darray_for_each(devs, i) {
+		i->nbuckets = i->fs_size / i->opts.bucket_size;
 		bch2_check_bucket_size(fs_opts, i);
 	}
 
 	/* calculate btree node size: */
 	if (!opt_defined(fs_opts, btree_node_size)) {
-		/* 256k default btree node size */
-		opt_set(fs_opts, btree_node_size, 256 << 10);
+		unsigned s = bch2_opts_default.btree_node_size;
 
-		for (i = devs; i < devs + nr_devs; i++)
-			fs_opts.btree_node_size =
-				min_t(unsigned, fs_opts.btree_node_size,
-				      i->bucket_size);
+		darray_for_each(devs, i)
+			s = min(s, i->opts.bucket_size);
+		opt_set(fs_opts, btree_node_size, s);
 	}
 
 	if (uuid_is_null(opts.uuid.b))
@@ -206,7 +216,8 @@ struct bch_sb *bch2_format(struct bch_opt_strs	fs_opt_strs,
 	sb.sb->version_min	= le16_to_cpu(opts.version);
 	sb.sb->magic		= BCHFS_MAGIC;
 	sb.sb->user_uuid	= opts.uuid;
-	sb.sb->nr_devices	= nr_devs;
+	sb.sb->nr_devices	= devs.nr;
+	SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(sb.sb, opts.version);
 
 	if (opts.version == bcachefs_metadata_version_current)
 		sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
@@ -218,17 +229,7 @@ struct bch_sb *bch2_format(struct bch_opt_strs	fs_opt_strs,
 		       opts.label,
 		       min(strlen(opts.label), sizeof(sb.sb->label)));
 
-	for (opt_id = 0;
-	     opt_id < bch2_opts_nr;
-	     opt_id++) {
-		u64 v;
-
-		v = bch2_opt_defined_by_id(&fs_opts, opt_id)
-			? bch2_opt_get_by_id(&fs_opts, opt_id)
-			: bch2_opt_get_by_id(&bch2_opts_default, opt_id);
-
-		__bch2_opt_set_sb(sb.sb, -1, &bch2_opt_table[opt_id], v);
-	}
+	bch2_opt_set_sb_all(sb.sb, -1, &fs_opts);
 
 	struct timespec now;
 	if (clock_gettime(CLOCK_REALTIME, &now))
@@ -240,31 +241,25 @@ struct bch_sb *bch2_format(struct bch_opt_strs	fs_opt_strs,
 	/* Member info: */
 	struct bch_sb_field_members_v2 *mi =
 		bch2_sb_field_resize(&sb, members_v2,
-			(sizeof(*mi) + sizeof(struct bch_member) *
-			nr_devs) / sizeof(u64));
+			(sizeof(*mi) + sizeof(struct bch_member) * devs.nr) / sizeof(u64));
+
 	mi->member_bytes = cpu_to_le16(sizeof(struct bch_member));
-	for (i = devs; i < devs + nr_devs; i++) {
-		struct bch_member *m = bch2_members_v2_get_mut(sb.sb, (i - devs));
+	darray_for_each(devs, i) {
+		unsigned idx = i - devs.data;
+		struct bch_member *m = bch2_members_v2_get_mut(sb.sb, idx);
 
 		uuid_generate(m->uuid.b);
 		m->nbuckets	= cpu_to_le64(i->nbuckets);
 		m->first_bucket	= 0;
-		m->bucket_size	= cpu_to_le16(i->bucket_size >> 9);
-
-		SET_BCH_MEMBER_DISCARD(m,	i->discard);
-		SET_BCH_MEMBER_DATA_ALLOWED(m,	i->data_allowed);
-		SET_BCH_MEMBER_DURABILITY(m,	i->durability + 1);
+		bch2_opt_set_sb_all(sb.sb, idx, &i->opts);
 	}
 
 	/* Disk labels*/
-	for (i = devs; i < devs + nr_devs; i++) {
-		struct bch_member *m;
-		int idx;
-
+	darray_for_each(devs, i) {
 		if (!i->label)
 			continue;
 
-		idx = bch2_disk_path_find_or_create(&sb, i->label);
+		int idx = bch2_disk_path_find_or_create(&sb, i->label);
 		if (idx < 0)
 			die("error creating disk path: %s", strerror(-idx));
 
@@ -272,18 +267,18 @@ struct bch_sb *bch2_format(struct bch_opt_strs	fs_opt_strs,
 		 * Recompute mi and m after each sb modification: its location
 		 * in memory may have changed due to reallocation.
 		 */
-		m = bch2_members_v2_get_mut(sb.sb, (i - devs));
+		struct bch_member *m = bch2_members_v2_get_mut(sb.sb, (i - devs.data));
 		SET_BCH_MEMBER_GROUP(m,	idx + 1);
 	}
 
 	SET_BCH_SB_FOREGROUND_TARGET(sb.sb,
-		parse_target(&sb, devs, nr_devs, fs_opt_strs.foreground_target));
+		parse_target(&sb, devs, fs_opt_strs.foreground_target));
 	SET_BCH_SB_BACKGROUND_TARGET(sb.sb,
-		parse_target(&sb, devs, nr_devs, fs_opt_strs.background_target));
+		parse_target(&sb, devs, fs_opt_strs.background_target));
 	SET_BCH_SB_PROMOTE_TARGET(sb.sb,
-		parse_target(&sb, devs, nr_devs, fs_opt_strs.promote_target));
+		parse_target(&sb, devs, fs_opt_strs.promote_target));
 	SET_BCH_SB_METADATA_TARGET(sb.sb,
-		parse_target(&sb, devs, nr_devs, fs_opt_strs.metadata_target));
+		parse_target(&sb, devs, fs_opt_strs.metadata_target));
 
 	/* Crypt: */
 	if (opts.encrypted) {
@@ -296,19 +291,19 @@ struct bch_sb *bch2_format(struct bch_opt_strs	fs_opt_strs,
 
 	bch2_sb_members_cpy_v2_v1(&sb);
 
-	for (i = devs; i < devs + nr_devs; i++) {
-		u64 size_sectors = i->size >> 9;
+	darray_for_each(devs, i) {
+		u64 size_sectors = i->fs_size >> 9;
 
-		sb.sb->dev_idx = i - devs;
+		sb.sb->dev_idx = i - devs.data;
 
 		if (!i->sb_offset) {
 			i->sb_offset	= BCH_SB_SECTOR;
 			i->sb_end	= size_sectors;
 		}
 
-		init_layout(&sb.sb->layout, fs_opts.block_size,
-			    opts.superblock_size,
-			    i->sb_offset, i->sb_end);
+		bch2_sb_layout_init(&sb.sb->layout, fs_opts.block_size,
+				    opts.superblock_size,
+				    i->sb_offset, i->sb_end);
 
 		/*
 		 * Also create a backup superblock at the end of the disk:
@@ -321,7 +316,7 @@ struct bch_sb *bch2_format(struct bch_opt_strs	fs_opt_strs,
 			struct bch_sb_layout *l = &sb.sb->layout;
 			u64 backup_sb = size_sectors - (1 << l->sb_max_size_bits);
 
-			backup_sb = rounddown(backup_sb, i->bucket_size >> 9);
+			backup_sb = rounddown(backup_sb, i->opts.bucket_size >> 9);
 			l->sb_offset[l->nr_superblocks++] = cpu_to_le64(backup_sb);
 		}
 
@@ -352,9 +347,9 @@ void bch2_super_write(int fd, struct bch_sb *sb)
 		if (sb->offset == BCH_SB_SECTOR) {
 			/* Write backup layout */
 
-			BUG_ON(bs > 4096);
+			unsigned buflen = max(bs, 4096);
 
-			char *buf = aligned_alloc(bs, bs);
+			char *buf = aligned_alloc(buflen, buflen);
 			xpread(fd, buf, bs, 4096 - bs);
 			memcpy(buf + bs - sizeof(sb->layout),
 			       &sb->layout,
@@ -618,6 +613,8 @@ int bchu_data(struct bchfs_handle fs, struct bch_ioctl_data cmd)
 
 /* option parsing */
 
+#include <getopt.h>
+
 void bch2_opt_strs_free(struct bch_opt_strs *opts)
 {
 	unsigned i;
@@ -628,6 +625,64 @@ void bch2_opt_strs_free(struct bch_opt_strs *opts)
 	}
 }
 
+static bool opt_type_filter(const struct bch_option *opt, unsigned opt_types)
+{
+	if (!(opt->flags & opt_types))
+		return false;
+
+	if ((opt_types & OPT_FORMAT) &&
+	    !opt->set_sb && !opt->set_member)
+		return false;
+
+	return true;
+}
+
+const struct bch_option *bch2_cmdline_opt_parse(int argc, char *argv[],
+						unsigned opt_types)
+{
+	if (optind >= argc)
+		return NULL;
+
+	if (argv[optind][0] != '-' ||
+	    argv[optind][1] != '-')
+		return NULL;
+
+	char *optstr = strdup(argv[optind] + 2);
+	optarg = argv[optind + 1];
+
+	char *eq = strchr(optstr, '=');
+	if (eq) {
+		*eq = '\0';
+		optarg = eq + 1;
+	}
+
+	if (!optarg)
+		optarg = "1";
+
+
+	int optid = bch2_opt_lookup(optstr);
+	if (optid < 0)
+		goto noopt;
+
+	const struct bch_option *opt = bch2_opt_table + optid;
+	if (!opt_type_filter(opt, opt_types))
+		goto noopt;
+
+	optind++;
+
+	if (opt->type != BCH_OPT_BOOL) {
+		if (optarg == argv[optind])
+			optind++;
+	} else {
+		optarg = NULL;
+	}
+
+	return opt;
+noopt:
+	free(optstr);
+	return NULL;
+}
+
 struct bch_opt_strs bch2_cmdline_opts_get(int *argc, char *argv[],
 					  unsigned opt_types)
 {
@@ -715,19 +770,17 @@ struct bch_opts bch2_parse_opts(struct bch_opt_strs strs)
 #define newline(c)		\
 	do {			\
 		printf("\n");	\
-		c = 0;	 	\
+		c = 0;		\
 	} while(0)
 void bch2_opts_usage(unsigned opt_types)
 {
 	const struct bch_option *opt;
 	unsigned i, c = 0, helpcol = 30;
 
-
-
 	for (opt = bch2_opt_table;
 	     opt < bch2_opt_table + bch2_opts_nr;
 	     opt++) {
-		if (!(opt->flags & opt_types))
+		if (!opt_type_filter(opt, opt_types))
 			continue;
 
 		c += printf("      --%s", opt->attr.name);

c_src/libbcachefs.h

@@ -28,6 +28,9 @@ struct {
 };
 
 void bch2_opt_strs_free(struct bch_opt_strs *);
+
+const struct bch_option *bch2_cmdline_opt_parse(int argc, char *argv[],
+						unsigned opt_types);
 struct bch_opt_strs bch2_cmdline_opts_get(int *, char *[], unsigned);
 struct bch_opts bch2_parse_opts(struct bch_opt_strs);
 void bch2_opts_usage(unsigned);
@@ -42,14 +45,18 @@ struct format_opts {
 	char		*source;
 };
 
+static inline unsigned bcachefs_kernel_version(void)
+{
+	return !access("/sys/module/bcachefs/parameters/version", R_OK)
+	    ? read_file_u64(AT_FDCWD, "/sys/module/bcachefs/parameters/version")
+	    : 0;
+}
+
 static inline struct format_opts format_opts_default()
 {
-	unsigned version = !access(   "/sys/module/bcachefs/parameters/version", R_OK)
-	    ? read_file_u64(AT_FDCWD, "/sys/module/bcachefs/parameters/version")
-	    : bcachefs_metadata_version_current;
-
 	return (struct format_opts) {
-		.version		= version,
+		.version		= bcachefs_kernel_version() ?:
+			bcachefs_metadata_version_current,
 		.superblock_size	= SUPERBLOCK_SIZE_DEFAULT,
 	};
 }
@@ -58,33 +65,35 @@ struct dev_opts {
 	struct file	*file;
 	struct block_device *bdev;
 	char		*path;
-	u64		size;		/* bytes*/
-	u64		bucket_size;	/* bytes */
-	const char	*label;
-	unsigned	data_allowed;
-	unsigned	durability;
-	bool		discard;
-
-	u64		nbuckets;
 
 	u64		sb_offset;
 	u64		sb_end;
+
+	u64		nbuckets;
+	u64		fs_size;
+
+	const char	*label; /* make this a bch_opt */
+
+	struct bch_opts	opts;
 };
 
+typedef DARRAY(struct dev_opts) dev_opts_list;
+
 static inline struct dev_opts dev_opts_default()
 {
-	return (struct dev_opts) {
-		.data_allowed		= ~0U << 2,
-		.durability		= 1,
-	};
+	return (struct dev_opts) { .opts = bch2_opts_empty() };
 }
 
-u64 bch2_pick_bucket_size(struct bch_opts, struct dev_opts *);
+void bch2_sb_layout_init(struct bch_sb_layout *,
+			 unsigned, unsigned, u64, u64);
+
+u64 bch2_pick_bucket_size(struct bch_opts, dev_opts_list);
 void bch2_check_bucket_size(struct bch_opts, struct dev_opts *);
 
 struct bch_sb *bch2_format(struct bch_opt_strs,
 			   struct bch_opts,
-			   struct format_opts, struct dev_opts *, size_t);
+			   struct format_opts,
+			   dev_opts_list devs);
 
 void bch2_super_write(int, struct bch_sb *);
 struct bch_sb *__bch2_super_read(int, u64);

c_src/posix_to_bcachefs.c

@@ -6,8 +6,8 @@
 #include "posix_to_bcachefs.h"
 #include "libbcachefs/alloc_foreground.h"
 #include "libbcachefs/buckets.h"
-#include "libbcachefs/fs-common.h"
 #include "libbcachefs/io_write.h"
+#include "libbcachefs/namei.h"
 #include "libbcachefs/str_hash.h"
 #include "libbcachefs/xattr.h"
 
@@ -264,7 +264,8 @@ void copy_link(struct bch_fs *c, struct bch_inode_unpacked *dst,
 
 static void copy_file(struct bch_fs *c, struct bch_inode_unpacked *dst,
 		      int src_fd, u64 src_size,
-		      char *src_path, struct copy_fs_state *s)
+		      char *src_path, struct copy_fs_state *s,
+		      u64 reserve_start)
 {
 	struct fiemap_iter iter;
 	struct fiemap_extent e;
@@ -295,11 +296,8 @@ static void copy_file(struct bch_fs *c, struct bch_inode_unpacked *dst,
 			continue;
 		}
 
-		/*
-		 * if the data is below 1 MB, copy it so it doesn't conflict
-		 * with bcachefs's potentially larger superblock:
-		 */
-		if (e.fe_physical < 1 << 20) {
+		/* If the data is in bcachefs's superblock region, copy it: */
+		if (e.fe_physical < reserve_start) {
 			copy_data(c, dst, src_fd, e.fe_logical,
 				  e.fe_logical + min(src_size - e.fe_logical,
 				      e.fe_length));
@@ -318,7 +316,8 @@ static void copy_file(struct bch_fs *c, struct bch_inode_unpacked *dst,
 static void copy_dir(struct copy_fs_state *s,
 		     struct bch_fs *c,
 		     struct bch_inode_unpacked *dst,
-		     int src_fd, const char *src_path)
+		     int src_fd, const char *src_path,
+		     u64 reserve_start)
 {
 	DIR *dir = fdopendir(src_fd);
 	struct dirent *d;
@@ -369,7 +368,7 @@ static void copy_dir(struct copy_fs_state *s,
 		switch (mode_to_type(stat.st_mode)) {
 		case DT_DIR:
 			fd = xopen(d->d_name, O_RDONLY|O_NOATIME);
-			copy_dir(s, c, &inode, fd, child_path);
+			copy_dir(s, c, &inode, fd, child_path, reserve_start);
 			close(fd);
 			break;
 		case DT_REG:
@@ -377,7 +376,7 @@ static void copy_dir(struct copy_fs_state *s,
 
 			fd = xopen(d->d_name, O_RDONLY|O_NOATIME);
 			copy_file(c, &inode, fd, stat.st_size,
-				  child_path, s);
+				  child_path, s, reserve_start);
 			close(fd);
 			break;
 		case DT_LNK:
@@ -409,7 +408,8 @@ next:
 
 static void reserve_old_fs_space(struct bch_fs *c,
 				 struct bch_inode_unpacked *root_inode,
-				 ranges *extents)
+				 ranges *extents,
+				 u64 reserve_start)
 {
 	struct bch_dev *ca = c->devs[0];
 	struct bch_inode_unpacked dst;
@@ -422,14 +422,20 @@ static void reserve_old_fs_space(struct bch_fs *c,
 
 	ranges_sort_merge(extents);
 
-	for_each_hole(iter, *extents, bucket_to_sector(ca, ca->mi.nbuckets) << 9, i)
-		link_data(c, &dst, i.start, i.start, i.end - i.start);
+	for_each_hole(iter, *extents, bucket_to_sector(ca, ca->mi.nbuckets) << 9, i) {
+		if (i.end <= reserve_start)
+			continue;
+
+		u64 start = max(i.start, reserve_start);
+
+		link_data(c, &dst, start, start, i.end - start);
+	}
 
 	update_inode(c, &dst);
 }
 
 void copy_fs(struct bch_fs *c, int src_fd, const char *src_path,
-		    struct copy_fs_state *s)
+	     struct copy_fs_state *s, u64 reserve_start)
 {
 	syncfs(src_fd);
 
@@ -448,10 +454,10 @@ void copy_fs(struct bch_fs *c, int src_fd, const char *src_path,
 
 
 	/* now, copy: */
-	copy_dir(s, c, &root_inode, src_fd, src_path);
+	copy_dir(s, c, &root_inode, src_fd, src_path, reserve_start);
 
 	if (BCH_MIGRATE_migrate == s->type)
-		reserve_old_fs_space(c, &root_inode, &s->extents);
+		reserve_old_fs_space(c, &root_inode, &s->extents, reserve_start);
 
 	update_inode(c, &root_inode);
 

c_src/posix_to_bcachefs.h

@@ -50,5 +50,5 @@ struct copy_fs_state {
  * initialized (`hardlinks` is initialized with zeroes).
  */
 void copy_fs(struct bch_fs *c, int src_fd, const char *src_path,
-		    struct copy_fs_state *s);
+		    struct copy_fs_state *s, u64);
 #endif /* _LIBBCACHE_H */

flake.lock

@@ -1,17 +1,12 @@
 {
   "nodes": {
     "crane": {
-      "inputs": {
-        "nixpkgs": [
-          "nixpkgs"
-        ]
-      },
       "locked": {
-        "lastModified": 1721842668,
-        "narHash": "sha256-k3oiD2z2AAwBFLa4+xfU+7G5fisRXfkvrMTCJrjZzXo=",
+        "lastModified": 1742394900,
+        "narHash": "sha256-vVOAp9ahvnU+fQoKd4SEXB2JG2wbENkpqcwlkIXgUC0=",
         "owner": "ipetkov",
         "repo": "crane",
-        "rev": "529c1a0b1f29f0d78fa3086b8f6a134c71ef3aaf",
+        "rev": "70947c1908108c0c551ddfd73d4f750ff2ea67cd",
         "type": "github"
       },
       "original": {
@@ -20,35 +15,14 @@
         "type": "github"
       }
     },
-    "fenix": {
-      "inputs": {
-        "nixpkgs": [
-          "nixpkgs"
-        ],
-        "rust-analyzer-src": "rust-analyzer-src"
-      },
-      "locked": {
-        "lastModified": 1722320953,
-        "narHash": "sha256-DfGaJtgrzcwPQYLTvjL1KaVIjpvi85b2MpM6yEGvJzM=",
-        "owner": "nix-community",
-        "repo": "fenix",
-        "rev": "483df76def3e5010d709aa3a0418ba2088503994",
-        "type": "github"
-      },
-      "original": {
-        "owner": "nix-community",
-        "repo": "fenix",
-        "type": "github"
-      }
-    },
     "flake-compat": {
       "flake": false,
       "locked": {
-        "lastModified": 1696426674,
-        "narHash": "sha256-kvjfFW7WAETZlt09AgDn1MrtKzP7t90Vf7vypd3OL1U=",
+        "lastModified": 1733328505,
+        "narHash": "sha256-NeCCThCEP3eCl2l/+27kNNK7QrwZB1IJCrXfrbv5oqU=",
         "owner": "edolstra",
         "repo": "flake-compat",
-        "rev": "0f9255e01c2351cc7d116c072cb317785dd33b33",
+        "rev": "ff81ac966bb2cae68946d5ed5fc4994f96d0ffec",
         "type": "github"
       },
       "original": {
@@ -62,11 +36,11 @@
         "nixpkgs-lib": "nixpkgs-lib"
       },
       "locked": {
-        "lastModified": 1719994518,
-        "narHash": "sha256-pQMhCCHyQGRzdfAkdJ4cIWiw+JNuWsTX7f0ZYSyz0VY=",
+        "lastModified": 1741352980,
+        "narHash": "sha256-+u2UunDA4Cl5Fci3m7S643HzKmIDAe+fiXrLqYsR2fs=",
         "owner": "hercules-ci",
         "repo": "flake-parts",
-        "rev": "9227223f6d922fee3c7b190b2cc238a99527bbb7",
+        "rev": "f4330d22f1c5d2ba72d3d22df5597d123fdb60a9",
         "type": "github"
       },
       "original": {
@@ -75,13 +49,33 @@
         "type": "github"
       }
     },
+    "nix-github-actions": {
+      "inputs": {
+        "nixpkgs": [
+          "nixpkgs"
+        ]
+      },
+      "locked": {
+        "lastModified": 1737420293,
+        "narHash": "sha256-F1G5ifvqTpJq7fdkT34e/Jy9VCyzd5XfJ9TO8fHhJWE=",
+        "owner": "nix-community",
+        "repo": "nix-github-actions",
+        "rev": "f4158fa080ef4503c8f4c820967d946c2af31ec9",
+        "type": "github"
+      },
+      "original": {
+        "owner": "nix-community",
+        "repo": "nix-github-actions",
+        "type": "github"
+      }
+    },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1722185531,
-        "narHash": "sha256-veKR07psFoJjINLC8RK4DiLniGGMgF3QMlS4tb74S6k=",
+        "lastModified": 1742422364,
+        "narHash": "sha256-mNqIplmEohk5jRkqYqG19GA8MbQ/D4gQSK0Mu4LvfRQ=",
         "owner": "nixos",
         "repo": "nixpkgs",
-        "rev": "52ec9ac3b12395ad677e8b62106f0b98c1f8569d",
+        "rev": "a84ebe20c6bc2ecbcfb000a50776219f48d134cc",
         "type": "github"
       },
       "original": {
@@ -93,40 +87,47 @@
     },
     "nixpkgs-lib": {
       "locked": {
-        "lastModified": 1719876945,
-        "narHash": "sha256-Fm2rDDs86sHy0/1jxTOKB1118Q0O3Uc7EC0iXvXKpbI=",
-        "type": "tarball",
-        "url": "https://github.com/NixOS/nixpkgs/archive/5daf0514482af3f97abaefc78a6606365c9108e2.tar.gz"
+        "lastModified": 1740877520,
+        "narHash": "sha256-oiwv/ZK/2FhGxrCkQkB83i7GnWXPPLzoqFHpDD3uYpk=",
+        "owner": "nix-community",
+        "repo": "nixpkgs.lib",
+        "rev": "147dee35aab2193b174e4c0868bd80ead5ce755c",
+        "type": "github"
       },
       "original": {
-        "type": "tarball",
-        "url": "https://github.com/NixOS/nixpkgs/archive/5daf0514482af3f97abaefc78a6606365c9108e2.tar.gz"
+        "owner": "nix-community",
+        "repo": "nixpkgs.lib",
+        "type": "github"
       }
     },
     "root": {
       "inputs": {
         "crane": "crane",
-        "fenix": "fenix",
         "flake-compat": "flake-compat",
         "flake-parts": "flake-parts",
+        "nix-github-actions": "nix-github-actions",
         "nixpkgs": "nixpkgs",
+        "rust-overlay": "rust-overlay",
         "treefmt-nix": "treefmt-nix"
       }
     },
-    "rust-analyzer-src": {
-      "flake": false,
+    "rust-overlay": {
+      "inputs": {
+        "nixpkgs": [
+          "nixpkgs"
+        ]
+      },
       "locked": {
-        "lastModified": 1722262053,
-        "narHash": "sha256-KxjkPVn9rQqYam6DhiN/V2NcMXtYW25maxkJoiVMpmE=",
-        "owner": "rust-lang",
-        "repo": "rust-analyzer",
-        "rev": "a021b85be57d34b1eed687fcafd5d5ec64b2d853",
+        "lastModified": 1742524367,
+        "narHash": "sha256-KzTwk/5ETJavJZYV1DEWdCx05M4duFCxCpRbQSKWpng=",
+        "owner": "oxalica",
+        "repo": "rust-overlay",
+        "rev": "70bf752d176b2ce07417e346d85486acea9040ef",
         "type": "github"
       },
       "original": {
-        "owner": "rust-lang",
-        "ref": "nightly",
-        "repo": "rust-analyzer",
+        "owner": "oxalica",
+        "repo": "rust-overlay",
         "type": "github"
       }
     },
@@ -137,11 +138,11 @@
         ]
       },
       "locked": {
-        "lastModified": 1722330636,
-        "narHash": "sha256-uru7JzOa33YlSRwf9sfXpJG+UAV+bnBEYMjrzKrQZFw=",
+        "lastModified": 1742370146,
+        "narHash": "sha256-XRE8hL4vKIQyVMDXykFh4ceo3KSpuJF3ts8GKwh5bIU=",
         "owner": "numtide",
         "repo": "treefmt-nix",
-        "rev": "768acdb06968e53aa1ee8de207fd955335c754b7",
+        "rev": "adc195eef5da3606891cedf80c0d9ce2d3190808",
         "type": "github"
       },
       "original": {

flake.nix

@@ -11,13 +11,10 @@
       inputs.nixpkgs.follows = "nixpkgs";
     };
 
-    crane = {
-      url = "github:ipetkov/crane";
-      inputs.nixpkgs.follows = "nixpkgs";
-    };
+    crane.url = "github:ipetkov/crane";
 
-    fenix = {
-      url = "github:nix-community/fenix";
+    rust-overlay = {
+      url = "github:oxalica/rust-overlay";
       inputs.nixpkgs.follows = "nixpkgs";
     };
 
@@ -25,6 +22,11 @@
       url = "github:edolstra/flake-compat";
       flake = false;
     };
+
+    nix-github-actions = {
+      url = "github:nix-community/nix-github-actions";
+      inputs.nixpkgs.follows = "nixpkgs";
+    };
   };
 
   outputs =
@@ -33,27 +35,31 @@
       nixpkgs,
       flake-parts,
       treefmt-nix,
-      fenix,
       crane,
-      ...
+      rust-overlay,
+      flake-compat,
+      nix-github-actions,
     }:
+    let
+      systems = nixpkgs.lib.filter (s: nixpkgs.lib.hasSuffix "-linux" s) nixpkgs.lib.systems.flakeExposed;
+    in
     flake-parts.lib.mkFlake { inherit inputs; } {
       imports = [ inputs.treefmt-nix.flakeModule ];
 
-      # can be extended, but these have proper binary cache support in nixpkgs
-      # as of writing.
-      systems = [
-        "aarch64-linux"
-        "x86_64-linux"
-        "i686-linux"
-      ];
+      flake = {
+        githubActions = nix-github-actions.lib.mkGithubMatrix {
+          # github actions supports fewer architectures
+          checks = nixpkgs.lib.getAttrs [ "aarch64-linux" "x86_64-linux" ] self.checks;
+        };
+      };
+
+      inherit systems;
 
       perSystem =
         {
           self',
           config,
           lib,
-          pkgs,
           system,
           ...
         }:
@@ -62,119 +68,225 @@
           inherit (lib.lists) findFirst;
           inherit (lib.strings) hasPrefix removePrefix substring;
 
+          pkgs = import nixpkgs {
+            inherit system;
+            overlays = [ (import rust-overlay) ];
+          };
+
           cargoToml = builtins.fromTOML (builtins.readFile ./Cargo.toml);
           rustfmtToml = builtins.fromTOML (builtins.readFile ./rustfmt.toml);
 
-          craneLib = crane.mkLib pkgs;
-
           rev = self.shortRev or self.dirtyShortRev or (substring 0 8 self.lastModifiedDate);
           makefileVersion = removePrefix "VERSION=" (
             findFirst (line: hasPrefix "VERSION=" line) "VERSION=0.0.0" (split "\n" (readFile ./Makefile))
           );
           version = "${makefileVersion}+${rev}";
 
-          commonArgs = {
-            inherit version;
-            src = self;
+          mkCommon =
+            {
+              crane,
+              pkgs,
+              rustVersion ? "latest",
 
-            env = {
-              PKG_CONFIG_SYSTEMD_SYSTEMDSYSTEMUNITDIR = "${placeholder "out"}/lib/systemd/system";
-              PKG_CONFIG_UDEV_UDEVDIR = "${placeholder "out"}/lib/udev";
+              # build time
+              buildPackages,
+              pkg-config,
+              rustPlatform,
+              stdenv,
+
+              # run time
+              keyutils,
+              libaio,
+              libsodium,
+              liburcu,
+              libuuid,
+              lz4,
+              udev,
+              zlib,
+              zstd,
+            }:
+            let
+              inherit (stdenv) cc hostPlatform;
+
+              craneLib = (crane.mkLib pkgs).overrideToolchain (
+                p: p.rust-bin.stable."${rustVersion}".minimal.override { extensions = [ "clippy" ]; }
+              );
+
+              args = {
+                inherit version;
+                src = self;
+                strictDeps = true;
+
+                env = {
+                  PKG_CONFIG_SYSTEMD_SYSTEMDSYSTEMUNITDIR = "${placeholder "out"}/lib/systemd/system";
+                  PKG_CONFIG_UDEV_UDEVDIR = "${placeholder "out"}/lib/udev";
+
+                  CARGO_BUILD_TARGET = hostPlatform.rust.rustcTargetSpec;
+                  "CARGO_TARGET_${hostPlatform.rust.cargoEnvVarTarget}_LINKER" = "${cc.targetPrefix}cc";
+                  HOST_CC = "${cc.nativePrefix}cc";
+                  TARGET_CC = "${cc.targetPrefix}cc";
+                };
+
+                makeFlags = [
+                  "INITRAMFS_DIR=${placeholder "out"}/etc/initramfs-tools"
+                  "PREFIX=${placeholder "out"}"
+                  "VERSION=${version}"
+                ];
+
+                dontStrip = true;
+
+                depsBuildBuild = [
+                  buildPackages.stdenv.cc
+                ];
+
+                nativeBuildInputs = [
+                  pkg-config
+                  rustPlatform.bindgenHook
+                ];
+
+                buildInputs = [
+                  keyutils
+                  libaio
+                  libsodium
+                  liburcu
+                  libuuid
+                  lz4
+                  udev
+                  zlib
+                  zstd
+                ];
+
+                meta = {
+                  description = "Userspace tools for bcachefs";
+                  license = lib.licenses.gpl2Only;
+                  mainProgram = "bcachefs";
+                };
+              };
+
+              cargoArtifacts = craneLib.buildDepsOnly args;
+            in
+            {
+              inherit args cargoArtifacts craneLib;
             };
+          common = pkgs.callPackage mkCommon { inherit crane; };
 
-            makeFlags = [
-              "INITRAMFS_DIR=${placeholder "out"}/etc/initramfs-tools"
-              "PREFIX=${placeholder "out"}"
-              "VERSION=${version}"
-            ];
+          mkPackage =
+            { common, name }:
+            common.craneLib.buildPackage (
+              common.args
+              // {
+                inherit (common) cargoArtifacts;
+                pname = name;
 
-            dontStrip = true;
+                enableParallelBuilding = true;
+                buildPhaseCargoCommand = ''
+                  make ''${enableParallelBuilding:+-j''${NIX_BUILD_CORES}} $makeFlags
+                '';
+                doNotPostBuildInstallCargoBinaries = true;
+                installPhaseCommand = ''
+                  make ''${enableParallelBuilding:+-j''${NIX_BUILD_CORES}} $makeFlags install
+                '';
 
-            nativeBuildInputs = with pkgs; [
-              pkg-config
-              rustPlatform.bindgenHook
-            ];
+                doInstallCheck = true;
+                installCheckPhase = ''
+                  runHook preInstallCheck
 
-            buildInputs = with pkgs; [
-              attr
-              keyutils
-              libaio
-              libsodium
-              liburcu
-              libuuid
-              lz4
-              udev
-              zlib
-              zstd
-            ];
+                  test "$($out/bin/bcachefs version)" = "${version}"
 
-            meta = {
-              description = "Userspace tools for bcachefs";
-              license = lib.licenses.gpl2Only;
-              mainProgram = "bcachefs";
-            };
-          };
+                  runHook postInstallCheck
+                '';
+              }
+            );
 
-          cargoArtifacts = craneLib.buildDepsOnly (commonArgs // { pname = cargoToml.package.name; });
+          mkPackages =
+            name: systems:
+            let
+              packagesForSystem =
+                crossSystem:
+                let
+                  localSystem = system;
+                  pkgs' = import nixpkgs {
+                    inherit crossSystem localSystem;
+                    overlays = [ (import rust-overlay) ];
+                  };
+
+                  common = pkgs'.callPackage mkCommon { inherit crane; };
+                  package = pkgs'.callPackage mkPackage { inherit common name; };
+                  packageFuse = package.overrideAttrs (
+                    final: prev: {
+                      makeFlags = prev.makeFlags ++ [ "BCACHEFS_FUSE=1" ];
+                      buildInputs = prev.buildInputs ++ [ pkgs'.fuse3 ];
+                    }
+                  );
+                in
+                [
+                  (lib.nameValuePair "${name}-${crossSystem}" package)
+                  (lib.nameValuePair "${name}-fuse-${crossSystem}" packageFuse)
+                ];
+            in
+            lib.listToAttrs (lib.flatten (map packagesForSystem systems));
         in
         {
-          packages.default = config.packages.bcachefs-tools;
-          packages.bcachefs-tools = craneLib.buildPackage (
-            commonArgs
+          packages =
+            let
+              inherit (cargoToml.package) name;
+            in
+            (mkPackages name systems)
             // {
-              inherit cargoArtifacts;
+              ${name} = config.packages."${name}-${system}";
+              "${name}-fuse" = config.packages."${name}-fuse-${system}";
+              default = config.packages.${name};
+            };
 
-              enableParallelBuilding = true;
-              buildPhaseCargoCommand = ''
-                make ''${enableParallelBuilding:+-j''${NIX_BUILD_CORES}} $makeFlags
-              '';
-              installPhaseCommand = ''
-                make ''${enableParallelBuilding:+-j''${NIX_BUILD_CORES}} $makeFlags install
-              '';
+          checks = {
+            inherit (config.packages)
+              bcachefs-tools
+              bcachefs-tools-fuse
+              bcachefs-tools-fuse-i686-linux
+              ;
 
-              doInstallCheck = true;
-              installCheckPhase = ''
-                runHook preInstallCheck
+            cargo-clippy = common.craneLib.cargoClippy (
+              common.args
+              // {
+                inherit (common) cargoArtifacts;
+                cargoClippyExtraArgs = "--all-targets --all-features -- --deny warnings";
+              }
+            );
 
-                test "$($out/bin/bcachefs version)" = "${version}"
+            # we have to build our own `craneLib.cargoTest`
+            cargo-test = common.craneLib.mkCargoDerivation (
+              common.args
+              // {
+                inherit (common) cargoArtifacts;
+                doCheck = true;
 
-                runHook postInstallCheck
-              '';
-            }
-          );
+                enableParallelChecking = true;
 
-          packages.bcachefs-tools-fuse = config.packages.bcachefs-tools.overrideAttrs (
-            final: prev: {
-              makeFlags = prev.makeFlags ++ [ "BCACHEFS_FUSE=1" ];
-              buildInputs = prev.buildInputs ++ [ pkgs.fuse3 ];
-            }
-          );
+                pnameSuffix = "-test";
+                buildPhaseCargoCommand = "";
+                checkPhaseCargoCommand = ''
+                  make ''${enableParallelChecking:+-j''${NIX_BUILD_CORES}} $makeFlags libbcachefs.a
+                  cargo test --profile release -- --nocapture
+                '';
+              }
+            );
 
-          checks.cargo-clippy = craneLib.cargoClippy (
-            commonArgs
-            // {
-              inherit cargoArtifacts;
-              cargoClippyExtraArgs = "--all-targets -- --deny warnings";
-            }
-          );
-
-          # we have to build our own `craneLib.cargoTest`
-          checks.cargo-test = craneLib.mkCargoDerivation (
-            commonArgs
-            // {
-              inherit cargoArtifacts;
-              doCheck = true;
-
-              enableParallelChecking = true;
-
-              pnameSuffix = "-test";
-              buildPhaseCargoCommand = "";
-              checkPhaseCargoCommand = ''
-                make ''${enableParallelChecking:+-j''${NIX_BUILD_CORES}} $makeFlags libbcachefs.a
-                cargo test --profile release -- --nocapture
-              '';
-            }
-          );
+            # cargo clippy with the current minimum supported rust version
+            # according to Cargo.toml
+            msrv =
+              let
+                rustVersion = cargoToml.package.rust-version;
+                common = pkgs.callPackage mkCommon { inherit crane rustVersion; };
+              in
+              common.craneLib.cargoClippy (
+                common.args
+                // {
+                  pname = "msrv";
+                  inherit (common) cargoArtifacts;
+                  cargoClippyExtraArgs = "--all-targets --all-features -- --deny warnings";
+                }
+              );
+          };
 
           devShells.default = pkgs.mkShell {
             inputsFrom = [
@@ -190,9 +302,12 @@
               cargo-audit
               cargo-outdated
               clang-tools
-              clippy
-              rust-analyzer
-              rustc
+              (rust-bin.stable.latest.minimal.override {
+                extensions = [
+                  "rust-analyzer"
+                  "rust-src"
+                ];
+              })
             ];
           };
 
@@ -204,7 +319,7 @@
               nixfmt.enable = true;
               rustfmt.edition = rustfmtToml.edition;
               rustfmt.enable = true;
-              rustfmt.package = fenix.packages.${system}.default.rustfmt;
+              rustfmt.package = pkgs.rust-bin.selectLatestNightlyWith (toolchain: toolchain.rustfmt);
             };
           };
         };

include/crypto/sha2.h

@@ -7,6 +7,7 @@
 #define _CRYPTO_SHA_H
 
 #include <linux/types.h>
+#include <sodium/crypto_hash_sha256.h>
 
 #define SHA1_DIGEST_SIZE        20
 #define SHA1_BLOCK_SIZE         64
@@ -112,4 +113,9 @@ extern int crypto_sha512_update(struct shash_desc *desc, const u8 *data,
 
 extern int crypto_sha512_finup(struct shash_desc *desc, const u8 *data,
 			       unsigned int len, u8 *hash);
+
+static inline void sha256(const u8 *data, unsigned int len, u8 *out)
+{
+	crypto_hash_sha256(out, data, len);
+}
 #endif

include/linux/atomic.h

@@ -16,53 +16,6 @@ typedef struct {
 	u64		counter;
 } atomic64_t;
 
-#ifndef C11_ATOMICS
-
-#include <urcu/uatomic.h>
-
-#if (CAA_BITS_PER_LONG != 64)
-#define ATOMIC64_SPINLOCK
-#endif
-
-#define __ATOMIC_READ(p)		uatomic_read(p)
-#define __ATOMIC_SET(p, v)		uatomic_set(p, v)
-#define __ATOMIC_SET_RELEASE(p, v)	uatomic_set(p, v)
-#define __ATOMIC_ADD_RETURN(v, p)	uatomic_add_return(p, v)
-#define __ATOMIC_SUB_RETURN(v, p)	uatomic_sub_return(p, v)
-#define __ATOMIC_ADD(v, p)		uatomic_add(p, v)
-#define __ATOMIC_SUB(v, p)		uatomic_sub(p, v)
-#define __ATOMIC_INC(p)			uatomic_inc(p)
-#define __ATOMIC_DEC(p)			uatomic_dec(p)
-#define __ATOMIC_AND(v, p)		uatomic_and(p, v)
-#define __ATOMIC_OR(v, p)		uatomic_or(p, v)
-
-#define xchg(p, v)			uatomic_xchg(p, v)
-#define xchg_acquire(p, v)		uatomic_xchg(p, v)
-#define cmpxchg(p, old, new)		uatomic_cmpxchg(p, old, new)
-#define cmpxchg_acquire(p, old, new)	uatomic_cmpxchg(p, old, new)
-#define cmpxchg_release(p, old, new)	uatomic_cmpxchg(p, old, new)
-
-#define try_cmpxchg(p, _old, _new)				\
-({								\
-	typeof(*(_old)) _v = cmpxchg(p, *(_old), _new);		\
-	bool _ret = _v == *(_old);				\
-	*(_old) = _v;						\
-	_ret;							\
-})
-
-#define try_cmpxchg_acquire(p, _old, _new)			\
-	try_cmpxchg(p, _old, _new)
-
-#define smp_mb__before_atomic()		cmm_smp_mb__before_uatomic_add()
-#define smp_mb__after_atomic()		cmm_smp_mb__after_uatomic_add()
-#define smp_wmb()			cmm_smp_wmb()
-#define smp_rmb()			cmm_smp_rmb()
-#define smp_mb()			cmm_smp_mb()
-#define smp_read_barrier_depends()	cmm_smp_read_barrier_depends()
-#define smp_acquire__after_ctrl_dep()	cmm_smp_mb()
-
-#else /* C11_ATOMICS */
-
 #define __ATOMIC_READ(p)		__atomic_load_n(p,	__ATOMIC_RELAXED)
 #define __ATOMIC_SET(p, v)		__atomic_store_n(p, v,	__ATOMIC_RELAXED)
 #define __ATOMIC_SET_RELEASE(p, v)	__atomic_store_n(p, v,	__ATOMIC_RELEASE)
@@ -83,6 +36,11 @@ typedef struct {
 				    __ATOMIC_SEQ_CST,		\
 				    __ATOMIC_SEQ_CST)
 
+#define try_cmpxchg_acquire(p, old, new)			\
+	__atomic_compare_exchange_n((p), old, new, false,	\
+				    __ATOMIC_ACQUIRE,		\
+				    __ATOMIC_RELAXED)
+
 #define cmpxchg(p, old, new)					\
 ({								\
 	typeof(*(p)) __old = (old);				\
@@ -109,7 +67,7 @@ typedef struct {
 								\
 	__atomic_compare_exchange_n((p), &__old, new, false,	\
 				    __ATOMIC_RELEASE,		\
-				    __ATOMIC_RELEASE);		\
+				    __ATOMIC_RELAXED);		\
 	__old;							\
 })
 
@@ -119,9 +77,7 @@ typedef struct {
 #define smp_rmb()		__atomic_thread_fence(__ATOMIC_SEQ_CST)
 #define smp_mb()		__atomic_thread_fence(__ATOMIC_SEQ_CST)
 #define smp_read_barrier_depends()
-
-#endif
-
+#define smp_acquire__after_ctrl_dep()	__atomic_thread_fence(__ATOMIC_SEQ_CST)
 #define smp_store_mb(var, value)  do { WRITE_ONCE(var, value); smp_mb(); } while (0)
 
 #define smp_load_acquire(p)						\

include/linux/blk_types.h

@@ -10,6 +10,8 @@
 #include <linux/types.h>
 #include <linux/bvec.h>
 #include <linux/kobject.h>
+#include <linux/mutex.h>
+#include <linux/rwsem.h>
 
 struct bio_set;
 struct bio;
@@ -63,6 +65,8 @@ struct block_device {
 	struct gendisk *	bd_disk;
 	struct gendisk		__bd_disk;
 	int			bd_fd;
+
+	struct mutex		bd_holder_lock;
 };
 
 #define bdev_kobj(_bdev) (&((_bdev)->kobj))

include/linux/blkdev.h

@@ -65,7 +65,10 @@ unsigned bdev_logical_block_size(struct block_device *bdev);
 sector_t get_capacity(struct gendisk *disk);
 
 struct blk_holder_ops {
-        void (*mark_dead)(struct block_device *bdev);
+        void (*mark_dead)(struct block_device *bdev, bool surprise);
+	void (*sync)(struct block_device *bdev);
+	int (*freeze)(struct block_device *bdev);
+	int (*thaw)(struct block_device *bdev);
 };
 
 static inline struct block_device *file_bdev(struct file *file)
@@ -80,8 +83,12 @@ int lookup_bdev(const char *path, dev_t *);
 
 struct super_block {
 	void			*s_fs_info;
+	struct rw_semaphore	s_umount;
 };
 
+static inline void evict_inodes(struct super_block *sb) {}
+static inline int sync_filesystem(struct super_block *) { return 0; }
+
 /*
  * File types
  *

include/linux/compiler.h

@@ -67,6 +67,7 @@
 #define __same_type(a, b)	__builtin_types_compatible_p(typeof(a), typeof(b))
 #define fallthrough		__attribute__((__fallthrough__))
 #define __noreturn		__attribute__((__noreturn__))
+#define __no_kmsan_checks
 
 #ifndef __counted_by
 #define __counted_by(nr)

include/linux/dcache.h

@@ -9,6 +9,8 @@ struct dentry {
 	struct inode *d_inode;
 };
 
+static inline void shrink_dcache_sb(struct super_block *) {}
+
 #define QSTR_INIT(n,l) { { { .len = l } }, .name = n }
 #define QSTR(n) (struct qstr)QSTR_INIT(n, strlen(n))
 

include/linux/kernel.h

@@ -12,6 +12,7 @@
 #include <linux/byteorder.h>
 #include <linux/compiler.h>
 #include <linux/dcache.h>
+#include <linux/kmsan-checks.h>
 #include <linux/math.h>
 #include <linux/minmax.h>
 

include/linux/kmsan-checks.h

@@ -0,0 +1,98 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * KMSAN checks to be used for one-off annotations in subsystems.
+ *
+ * Copyright (C) 2017-2022 Google LLC
+ * Author: Alexander Potapenko <glider@google.com>
+ *
+ */
+
+#ifndef _LINUX_KMSAN_CHECKS_H
+#define _LINUX_KMSAN_CHECKS_H
+
+#include <linux/types.h>
+
+#ifdef CONFIG_KMSAN
+
+/**
+ * kmsan_poison_memory() - Mark the memory range as uninitialized.
+ * @address: address to start with.
+ * @size:    size of buffer to poison.
+ * @flags:   GFP flags for allocations done by this function.
+ *
+ * Until other data is written to this range, KMSAN will treat it as
+ * uninitialized. Error reports for this memory will reference the call site of
+ * kmsan_poison_memory() as origin.
+ */
+void kmsan_poison_memory(const void *address, size_t size, gfp_t flags);
+
+/**
+ * kmsan_unpoison_memory() -  Mark the memory range as initialized.
+ * @address: address to start with.
+ * @size:    size of buffer to unpoison.
+ *
+ * Until other data is written to this range, KMSAN will treat it as
+ * initialized.
+ */
+void kmsan_unpoison_memory(const void *address, size_t size);
+
+/**
+ * kmsan_check_memory() - Check the memory range for being initialized.
+ * @address: address to start with.
+ * @size:    size of buffer to check.
+ *
+ * If any piece of the given range is marked as uninitialized, KMSAN will report
+ * an error.
+ */
+void kmsan_check_memory(const void *address, size_t size);
+
+/**
+ * kmsan_copy_to_user() - Notify KMSAN about a data transfer to userspace.
+ * @to:      destination address in the userspace.
+ * @from:    source address in the kernel.
+ * @to_copy: number of bytes to copy.
+ * @left:    number of bytes not copied.
+ *
+ * If this is a real userspace data transfer, KMSAN checks the bytes that were
+ * actually copied to ensure there was no information leak. If @to belongs to
+ * the kernel space (which is possible for compat syscalls), KMSAN just copies
+ * the metadata.
+ */
+void kmsan_copy_to_user(void __user *to, const void *from, size_t to_copy,
+			size_t left);
+
+/**
+ * kmsan_memmove() - Notify KMSAN about a data copy within kernel.
+ * @to:   destination address in the kernel.
+ * @from: source address in the kernel.
+ * @size: number of bytes to copy.
+ *
+ * Invoked after non-instrumented version (e.g. implemented using assembly
+ * code) of memmove()/memcpy() is called, in order to copy KMSAN's metadata.
+ */
+void kmsan_memmove(void *to, const void *from, size_t to_copy);
+
+#else
+
+static inline void kmsan_poison_memory(const void *address, size_t size,
+				       gfp_t flags)
+{
+}
+static inline void kmsan_unpoison_memory(const void *address, size_t size)
+{
+}
+static inline void kmsan_check_memory(const void *address, size_t size)
+{
+}
+static inline void kmsan_copy_to_user(void __user *to, const void *from,
+				      size_t to_copy, size_t left)
+{
+}
+
+static inline void kmsan_memmove(void *to, const void *from, size_t to_copy)
+{
+}
+
+#endif
+
+#endif /* _LINUX_KMSAN_CHECKS_H */

include/linux/math64.h

@@ -82,4 +82,71 @@ static inline s64 div_s64(s64 dividend, s32 divisor)
 	return div_s64_rem(dividend, divisor, &remainder);
 }
 
+#ifndef mul_u32_u32
+/*
+ * Many a GCC version messes this up and generates a 64x64 mult :-(
+ */
+static inline u64 mul_u32_u32(u32 a, u32 b)
+{
+	return (u64)a * b;
+}
+#endif
+
+#if defined(CONFIG_ARCH_SUPPORTS_INT128) && defined(__SIZEOF_INT128__)
+
+#ifndef mul_u64_u64_shr
+static __always_inline u64 mul_u64_u64_shr(u64 a, u64 mul, unsigned int shift)
+{
+	return (u64)(((unsigned __int128)a * mul) >> shift);
+}
+#endif /* mul_u64_u64_shr */
+
+#else
+
+#ifndef mul_u64_u64_shr
+static inline u64 mul_u64_u64_shr(u64 a, u64 b, unsigned int shift)
+{
+	union {
+		u64 ll;
+		struct {
+#ifdef __BIG_ENDIAN
+			u32 high, low;
+#else
+			u32 low, high;
+#endif
+		} l;
+	} rl, rm, rn, rh, a0, b0;
+	u64 c;
+
+	a0.ll = a;
+	b0.ll = b;
+
+	rl.ll = mul_u32_u32(a0.l.low, b0.l.low);
+	rm.ll = mul_u32_u32(a0.l.low, b0.l.high);
+	rn.ll = mul_u32_u32(a0.l.high, b0.l.low);
+	rh.ll = mul_u32_u32(a0.l.high, b0.l.high);
+
+	/*
+	 * Each of these lines computes a 64-bit intermediate result into "c",
+	 * starting at bits 32-95.  The low 32-bits go into the result of the
+	 * multiplication, the high 32-bits are carried into the next step.
+	 */
+	rl.l.high = c = (u64)rl.l.high + rm.l.low + rn.l.low;
+	rh.l.low = c = (c >> 32) + rm.l.high + rn.l.high + rh.l.low;
+	rh.l.high = (c >> 32) + rh.l.high;
+
+	/*
+	 * The 128-bit result of the multiplication is in rl.ll and rh.ll,
+	 * shift it right and throw away the high part of the result.
+	 */
+	if (shift == 0)
+		return rl.ll;
+	if (shift < 64)
+		return (rl.ll >> shift) | (rh.ll << (64 - shift));
+	return rh.ll >> (shift & 63);
+}
+#endif /* mul_u64_u64_shr */
+
+#endif
+
 #endif /* _LINUX_MATH64_H */

include/linux/mm.h

@@ -4,6 +4,7 @@
 
 #include <sys/syscall.h>
 #include <unistd.h>
+#include <linux/bug.h>
 #include <linux/types.h>
 
 struct sysinfo {

include/linux/random.h

@@ -9,7 +9,9 @@
 #include <unistd.h>
 #include <sys/syscall.h>
 #include <linux/bug.h>
+#include <linux/kernel.h>
 #include <linux/log2.h>
+#include <linux/math64.h>
 
 #ifdef SYS_getrandom
 static inline int getrandom(void *buf, size_t buflen, unsigned int flags)
@@ -67,4 +69,24 @@ static inline u32 get_random_u32_below(u32 ceil)
 	}
 }
 
+static inline u32 __get_random_u32_below(u32 ceil)
+{
+	return get_random_u32_below(ceil);
+}
+
+static inline u64 get_random_u64_below(u64 ceil)
+{
+	if (ceil <= 1)
+		return 0;
+	if (ceil <= U32_MAX)
+		return get_random_u32_below(ceil);
+
+	for (;;) {
+		u64 rand = get_random_u64();
+		u64 mult = ceil * rand;
+		if (likely(mult >= -ceil % ceil))
+			return mul_u64_u64_shr(ceil, rand, 64);
+	}
+}
+
 #endif /* _LINUX_RANDOM_H */

include/linux/sched/sysctl.h

@@ -0,0 +1,6 @@
+#ifndef __TOOLS_LINUX_SCHED_SYSCTL_H
+#define __TOOLS_LINUX_SCHED_SYSCTL_H
+
+#define sysctl_hung_task_timeout_secs		(HZ * 10)
+
+#endif /* __TOOLS_LINUX_SCHED_SYSCTL_H */

libbcachefs/alloc_background.c

@@ -232,7 +232,7 @@ int bch2_alloc_v3_validate(struct bch_fs *c, struct bkey_s_c k,
 	int ret = 0;
 
 	bkey_fsck_err_on(bch2_alloc_unpack_v3(&u, k),
-			 c, alloc_v2_unpack_error,
+			 c, alloc_v3_unpack_error,
 			 "unpack error");
 fsck_err:
 	return ret;
@@ -589,6 +589,8 @@ iter_err:
 
 int bch2_alloc_read(struct bch_fs *c)
 {
+	down_read(&c->state_lock);
+
 	struct btree_trans *trans = bch2_trans_get(c);
 	struct bch_dev *ca = NULL;
 	int ret;
@@ -652,6 +654,7 @@ int bch2_alloc_read(struct bch_fs *c)
 	bch2_dev_put(ca);
 	bch2_trans_put(trans);
 
+	up_read(&c->state_lock);
 	bch_err_fn(c, ret);
 	return ret;
 }
@@ -673,8 +676,7 @@ static int __need_discard_or_freespace_err(struct btree_trans *trans,
 	bch2_bkey_val_to_text(&buf, c, alloc_k);
 
 	int ret = __bch2_fsck_err(NULL, trans, flags, err_id,
-				  "bucket incorrectly %sset in %s btree\n"
-				  "  %s",
+				  "bucket incorrectly %sset in %s btree\n%s",
 				  set ? "" : "un",
 				  bch2_btree_id_str(btree),
 				  buf.buf);
@@ -777,14 +779,12 @@ static inline int bch2_dev_data_type_accounting_mod(struct btree_trans *trans, s
 						    s64 delta_sectors,
 						    s64 delta_fragmented, unsigned flags)
 {
-	struct disk_accounting_pos acc = {
-		.type = BCH_DISK_ACCOUNTING_dev_data_type,
-		.dev_data_type.dev		= ca->dev_idx,
-		.dev_data_type.data_type	= data_type,
-	};
 	s64 d[3] = { delta_buckets, delta_sectors, delta_fragmented };
 
-	return bch2_disk_accounting_mod(trans, &acc, d, 3, flags & BTREE_TRIGGER_gc);
+	return bch2_disk_accounting_mod2(trans, flags & BTREE_TRIGGER_gc,
+					 d, dev_data_type,
+					 .dev		= ca->dev_idx,
+					 .data_type	= data_type);
 }
 
 int bch2_alloc_key_to_dev_counters(struct btree_trans *trans, struct bch_dev *ca,
@@ -837,7 +837,7 @@ int bch2_trigger_alloc(struct btree_trans *trans,
 
 	struct bch_dev *ca = bch2_dev_bucket_tryget(c, new.k->p);
 	if (!ca)
-		return -EIO;
+		return -BCH_ERR_trigger_alloc;
 
 	struct bch_alloc_v4 old_a_convert;
 	const struct bch_alloc_v4 *old_a = bch2_alloc_to_v4(old, &old_a_convert);
@@ -871,6 +871,9 @@ int bch2_trigger_alloc(struct btree_trans *trans,
 		if (data_type_is_empty(new_a->data_type) &&
 		    BCH_ALLOC_V4_NEED_INC_GEN(new_a) &&
 		    !bch2_bucket_is_open_safe(c, new.k->p.inode, new.k->p.offset)) {
+			if (new_a->oldest_gen == new_a->gen &&
+			    !bch2_bucket_sectors_total(*new_a))
+				new_a->oldest_gen++;
 			new_a->gen++;
 			SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false);
 			alloc_data_type_set(new_a, new_a->data_type);
@@ -889,26 +892,20 @@ int bch2_trigger_alloc(struct btree_trans *trans,
 		    !new_a->io_time[READ])
 			new_a->io_time[READ] = bch2_current_io_time(c, READ);
 
-		u64 old_lru = alloc_lru_idx_read(*old_a);
-		u64 new_lru = alloc_lru_idx_read(*new_a);
-		if (old_lru != new_lru) {
-			ret = bch2_lru_change(trans, new.k->p.inode,
-					      bucket_to_u64(new.k->p),
-					      old_lru, new_lru);
-			if (ret)
-				goto err;
-		}
+		ret = bch2_lru_change(trans, new.k->p.inode,
+				      bucket_to_u64(new.k->p),
+				      alloc_lru_idx_read(*old_a),
+				      alloc_lru_idx_read(*new_a));
+		if (ret)
+			goto err;
 
-		old_lru = alloc_lru_idx_fragmentation(*old_a, ca);
-		new_lru = alloc_lru_idx_fragmentation(*new_a, ca);
-		if (old_lru != new_lru) {
-			ret = bch2_lru_change(trans,
-					BCH_LRU_FRAGMENTATION_START,
-					bucket_to_u64(new.k->p),
-					old_lru, new_lru);
-			if (ret)
-				goto err;
-		}
+		ret = bch2_lru_change(trans,
+				      BCH_LRU_BUCKET_FRAGMENTATION,
+				      bucket_to_u64(new.k->p),
+				      alloc_lru_idx_fragmentation(*old_a, ca),
+				      alloc_lru_idx_fragmentation(*new_a, ca));
+		if (ret)
+			goto err;
 
 		if (old_a->gen != new_a->gen) {
 			ret = bch2_bucket_gen_update(trans, new.k->p, new_a->gen);
@@ -1032,9 +1029,9 @@ fsck_err:
 	bch2_dev_put(ca);
 	return ret;
 invalid_bucket:
-	bch2_fs_inconsistent(c, "reference to invalid bucket\n  %s",
+	bch2_fs_inconsistent(c, "reference to invalid bucket\n%s",
 			     (bch2_bkey_val_to_text(&buf, c, new.s_c), buf.buf));
-	ret = -EIO;
+	ret = -BCH_ERR_trigger_alloc;
 	goto err;
 }
 
@@ -1206,8 +1203,7 @@ int bch2_check_alloc_key(struct btree_trans *trans,
 
 	if (fsck_err_on(a->gen != alloc_gen(k, gens_offset),
 			trans, bucket_gens_key_wrong,
-			"incorrect gen in bucket_gens btree (got %u should be %u)\n"
-			"  %s",
+			"incorrect gen in bucket_gens btree (got %u should be %u)\n%s",
 			alloc_gen(k, gens_offset), a->gen,
 			(printbuf_reset(&buf),
 			 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
@@ -1265,7 +1261,7 @@ int bch2_check_alloc_hole_freespace(struct btree_trans *trans,
 	if (fsck_err_on(k.k->type != KEY_TYPE_set,
 			trans, freespace_hole_missing,
 			"hole in alloc btree missing in freespace btree\n"
-			"  device %llu buckets %llu-%llu",
+			"device %llu buckets %llu-%llu",
 			freespace_iter->pos.inode,
 			freespace_iter->pos.offset,
 			end->offset)) {
@@ -1424,7 +1420,7 @@ int bch2_check_discard_freespace_key(struct btree_trans *trans, struct btree_ite
 	    (state == BCH_DATA_free &&
 	     genbits != alloc_freespace_genbits(*a))) {
 		if (fsck_err(trans, need_discard_freespace_key_bad,
-			     "%s\n  incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
+			     "%s\nincorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
 			     (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
 			     bch2_btree_id_str(iter->btree_id),
 			     iter->pos.inode,
@@ -1505,7 +1501,7 @@ int bch2_check_bucket_gens_key(struct btree_trans *trans,
 	struct bch_dev *ca = bch2_dev_tryget_noerror(c, k.k->p.inode);
 	if (!ca) {
 		if (fsck_err(trans, bucket_gens_to_invalid_dev,
-			     "bucket_gens key for invalid device:\n  %s",
+			     "bucket_gens key for invalid device:\n%s",
 			     (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
 			ret = bch2_btree_delete_at(trans, iter, 0);
 		goto out;
@@ -1514,7 +1510,7 @@ int bch2_check_bucket_gens_key(struct btree_trans *trans,
 	if (fsck_err_on(end <= ca->mi.first_bucket ||
 			start >= ca->mi.nbuckets,
 			trans, bucket_gens_to_invalid_buckets,
-			"bucket_gens key for invalid buckets:\n  %s",
+			"bucket_gens key for invalid buckets:\n%s",
 			(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
 		ret = bch2_btree_delete_at(trans, iter, 0);
 		goto out;
@@ -1705,7 +1701,8 @@ static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
 
 	u64 lru_idx = alloc_lru_idx_fragmentation(*a, ca);
 	if (lru_idx) {
-		ret = bch2_lru_check_set(trans, BCH_LRU_FRAGMENTATION_START,
+		ret = bch2_lru_check_set(trans, BCH_LRU_BUCKET_FRAGMENTATION,
+					 bucket_to_u64(alloc_k.k->p),
 					 lru_idx, alloc_k, last_flushed);
 		if (ret)
 			goto err;
@@ -1716,8 +1713,7 @@ static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
 
 	if (fsck_err_on(!a->io_time[READ],
 			trans, alloc_key_cached_but_read_time_zero,
-			"cached bucket with read_time 0\n"
-			"  %s",
+			"cached bucket with read_time 0\n%s",
 		(printbuf_reset(&buf),
 		 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
 		struct bkey_i_alloc_v4 *a_mut =
@@ -1735,7 +1731,9 @@ static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
 		a = &a_mut->v;
 	}
 
-	ret = bch2_lru_check_set(trans, alloc_k.k->p.inode, a->io_time[READ],
+	ret = bch2_lru_check_set(trans, alloc_k.k->p.inode,
+				 bucket_to_u64(alloc_k.k->p),
+				 a->io_time[READ],
 				 alloc_k, last_flushed);
 	if (ret)
 		goto err;
@@ -1757,7 +1755,8 @@ int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
 		for_each_btree_key_commit(trans, iter, BTREE_ID_alloc,
 				POS_MIN, BTREE_ITER_prefetch, k,
 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
-			bch2_check_alloc_to_lru_ref(trans, &iter, &last_flushed)));
+			bch2_check_alloc_to_lru_ref(trans, &iter, &last_flushed))) ?:
+		bch2_check_stripe_to_lru_refs(c);
 
 	bch2_bkey_buf_exit(&last_flushed, c);
 	bch_err_fn(c, ret);
@@ -1805,6 +1804,19 @@ struct discard_buckets_state {
 	u64		discarded;
 };
 
+/*
+ * This is needed because discard is both a filesystem option and a device
+ * option, and mount options are supposed to apply to that mount and not be
+ * persisted, i.e. if it's set as a mount option we can't propagate it to the
+ * device.
+ */
+static inline bool discard_opt_enabled(struct bch_fs *c, struct bch_dev *ca)
+{
+	return test_bit(BCH_FS_discard_mount_opt_set, &c->flags)
+		? c->opts.discard
+		: ca->mi.discard;
+}
+
 static int bch2_discard_one_bucket(struct btree_trans *trans,
 				   struct bch_dev *ca,
 				   struct btree_iter *need_discard_iter,
@@ -1868,7 +1880,7 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 		s->discarded++;
 		*discard_pos_done = iter.pos;
 
-		if (ca->mi.discard && !c->opts.nochanges) {
+		if (discard_opt_enabled(c, ca) && !c->opts.nochanges) {
 			/*
 			 * This works without any other locks because this is the only
 			 * thread that removes items from the need_discard tree
@@ -2058,16 +2070,71 @@ put_ref:
 	bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast);
 }
 
+static int invalidate_one_bp(struct btree_trans *trans,
+			     struct bch_dev *ca,
+			     struct bkey_s_c_backpointer bp,
+			     struct bkey_buf *last_flushed)
+{
+	struct btree_iter extent_iter;
+	struct bkey_s_c extent_k =
+		bch2_backpointer_get_key(trans, bp, &extent_iter, 0, last_flushed);
+	int ret = bkey_err(extent_k);
+	if (ret)
+		return ret;
+
+	struct bkey_i *n =
+		bch2_bkey_make_mut(trans, &extent_iter, &extent_k,
+				   BTREE_UPDATE_internal_snapshot_node);
+	ret = PTR_ERR_OR_ZERO(n);
+	if (ret)
+		goto err;
+
+	bch2_bkey_drop_device(bkey_i_to_s(n), ca->dev_idx);
+err:
+	bch2_trans_iter_exit(trans, &extent_iter);
+	return ret;
+}
+
+static int invalidate_one_bucket_by_bps(struct btree_trans *trans,
+					struct bch_dev *ca,
+					struct bpos bucket,
+					u8 gen,
+					struct bkey_buf *last_flushed)
+{
+	struct bpos bp_start	= bucket_pos_to_bp_start(ca,	bucket);
+	struct bpos bp_end	= bucket_pos_to_bp_end(ca,	bucket);
+
+	return for_each_btree_key_max_commit(trans, iter, BTREE_ID_backpointers,
+				      bp_start, bp_end, 0, k,
+				      NULL, NULL,
+				      BCH_WATERMARK_btree|
+				      BCH_TRANS_COMMIT_no_enospc, ({
+		if (k.k->type != KEY_TYPE_backpointer)
+			continue;
+
+		struct bkey_s_c_backpointer bp = bkey_s_c_to_backpointer(k);
+
+		if (bp.v->bucket_gen != gen)
+			continue;
+
+		/* filter out bps with gens that don't match */
+
+		invalidate_one_bp(trans, ca, bp, last_flushed);
+	}));
+}
+
+noinline_for_stack
 static int invalidate_one_bucket(struct btree_trans *trans,
+				 struct bch_dev *ca,
 				 struct btree_iter *lru_iter,
 				 struct bkey_s_c lru_k,
+				 struct bkey_buf *last_flushed,
 				 s64 *nr_to_invalidate)
 {
 	struct bch_fs *c = trans->c;
-	struct bkey_i_alloc_v4 *a = NULL;
 	struct printbuf buf = PRINTBUF;
 	struct bpos bucket = u64_to_bucket(lru_k.k->p.offset);
-	unsigned cached_sectors;
+	struct btree_iter alloc_iter = {};
 	int ret = 0;
 
 	if (*nr_to_invalidate <= 0)
@@ -2084,35 +2151,37 @@ static int invalidate_one_bucket(struct btree_trans *trans,
 	if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset))
 		return 0;
 
-	a = bch2_trans_start_alloc_update(trans, bucket, BTREE_TRIGGER_bucket_invalidate);
-	ret = PTR_ERR_OR_ZERO(a);
+	struct bkey_s_c alloc_k = bch2_bkey_get_iter(trans, &alloc_iter,
+						     BTREE_ID_alloc, bucket,
+						     BTREE_ITER_cached);
+	ret = bkey_err(alloc_k);
 	if (ret)
-		goto out;
+		return ret;
+
+	struct bch_alloc_v4 a_convert;
+	const struct bch_alloc_v4 *a = bch2_alloc_to_v4(alloc_k, &a_convert);
 
 	/* We expect harmless races here due to the btree write buffer: */
-	if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(a->v))
+	if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(*a))
 		goto out;
 
-	BUG_ON(a->v.data_type != BCH_DATA_cached);
-	BUG_ON(a->v.dirty_sectors);
+	/*
+	 * Impossible since alloc_lru_idx_read() only returns nonzero if the
+	 * bucket is supposed to be on the cached bucket LRU (i.e.
+	 * BCH_DATA_cached)
+	 *
+	 * bch2_lru_validate() also disallows lru keys with lru_pos_time() == 0
+	 */
+	BUG_ON(a->data_type != BCH_DATA_cached);
+	BUG_ON(a->dirty_sectors);
 
-	if (!a->v.cached_sectors)
+	if (!a->cached_sectors)
 		bch_err(c, "invalidating empty bucket, confused");
 
-	cached_sectors = a->v.cached_sectors;
+	unsigned cached_sectors = a->cached_sectors;
+	u8 gen = a->gen;
 
-	SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
-	a->v.gen++;
-	a->v.data_type		= 0;
-	a->v.dirty_sectors	= 0;
-	a->v.stripe_sectors	= 0;
-	a->v.cached_sectors	= 0;
-	a->v.io_time[READ]	= bch2_current_io_time(c, READ);
-	a->v.io_time[WRITE]	= bch2_current_io_time(c, WRITE);
-
-	ret = bch2_trans_commit(trans, NULL, NULL,
-				BCH_WATERMARK_btree|
-				BCH_TRANS_COMMIT_no_enospc);
+	ret = invalidate_one_bucket_by_bps(trans, ca, bucket, gen, last_flushed);
 	if (ret)
 		goto out;
 
@@ -2120,6 +2189,7 @@ static int invalidate_one_bucket(struct btree_trans *trans,
 	--*nr_to_invalidate;
 out:
 fsck_err:
+	bch2_trans_iter_exit(trans, &alloc_iter);
 	printbuf_exit(&buf);
 	return ret;
 }
@@ -2146,6 +2216,10 @@ static void bch2_do_invalidates_work(struct work_struct *work)
 	struct btree_trans *trans = bch2_trans_get(c);
 	int ret = 0;
 
+	struct bkey_buf last_flushed;
+	bch2_bkey_buf_init(&last_flushed);
+	bkey_init(&last_flushed.k->k);
+
 	ret = bch2_btree_write_buffer_tryflush(trans);
 	if (ret)
 		goto err;
@@ -2170,7 +2244,7 @@ static void bch2_do_invalidates_work(struct work_struct *work)
 		if (!k.k)
 			break;
 
-		ret = invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate);
+		ret = invalidate_one_bucket(trans, ca, &iter, k, &last_flushed, &nr_to_invalidate);
 restart_err:
 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 			continue;
@@ -2183,6 +2257,7 @@ restart_err:
 err:
 	bch2_trans_put(trans);
 	percpu_ref_put(&ca->io_ref);
+	bch2_bkey_buf_exit(&last_flushed, c);
 	bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
 }
 

libbcachefs/alloc_background.h

@@ -131,7 +131,7 @@ static inline enum bch_data_type alloc_data_type(struct bch_alloc_v4 a,
 	if (a.stripe)
 		return data_type == BCH_DATA_parity ? data_type : BCH_DATA_stripe;
 	if (bch2_bucket_sectors_dirty(a))
-		return data_type;
+		return bucket_data_type(data_type);
 	if (a.cached_sectors)
 		return BCH_DATA_cached;
 	if (BCH_ALLOC_V4_NEED_DISCARD(&a))

libbcachefs/alloc_foreground.c

@@ -127,14 +127,14 @@ void __bch2_open_bucket_put(struct bch_fs *c, struct open_bucket *ob)
 
 void bch2_open_bucket_write_error(struct bch_fs *c,
 				  struct open_buckets *obs,
-				  unsigned dev)
+				  unsigned dev, int err)
 {
 	struct open_bucket *ob;
 	unsigned i;
 
 	open_bucket_for_each(c, obs, ob, i)
 		if (ob->dev == dev && ob->ec)
-			bch2_ec_bucket_cancel(c, ob);
+			bch2_ec_bucket_cancel(c, ob, err);
 }
 
 static struct open_bucket *bch2_open_bucket_alloc(struct bch_fs *c)
@@ -631,7 +631,7 @@ static inline void bch2_dev_stripe_increment_inlined(struct bch_dev *ca,
 			       struct bch_dev_usage *usage)
 {
 	u64 *v = stripe->next_alloc + ca->dev_idx;
-	u64 free_space = dev_buckets_available(ca, BCH_WATERMARK_normal);
+	u64 free_space = __dev_buckets_available(ca, *usage, BCH_WATERMARK_normal);
 	u64 free_space_inv = free_space
 		? div64_u64(1ULL << 48, free_space)
 		: 1ULL << 48;

libbcachefs/alloc_foreground.h

@@ -82,7 +82,7 @@ static inline struct open_bucket *ec_open_bucket(struct bch_fs *c,
 }
 
 void bch2_open_bucket_write_error(struct bch_fs *,
-			struct open_buckets *, unsigned);
+			struct open_buckets *, unsigned, int);
 
 void __bch2_open_bucket_put(struct bch_fs *, struct open_bucket *);
 

libbcachefs/backpointers.c

@@ -11,6 +11,7 @@
 #include "checksum.h"
 #include "disk_accounting.h"
 #include "error.h"
+#include "progress.h"
 
 #include <linux/mm.h>
 
@@ -49,6 +50,8 @@ void bch2_backpointer_to_text(struct printbuf *out, struct bch_fs *c, struct bke
 	}
 
 	bch2_btree_id_level_to_text(out, bp.v->btree_id, bp.v->level);
+	prt_str(out, " data_type=");
+	bch2_prt_data_type(out, bp.v->data_type);
 	prt_printf(out, " suboffset=%u len=%u gen=%u pos=",
 		   (u32) bp.k->p.offset & ~(~0U << MAX_EXTENT_COMPRESS_RATIO_SHIFT),
 		   bp.v->bucket_len,
@@ -93,6 +96,7 @@ static noinline int backpointer_mod_err(struct btree_trans *trans,
 {
 	struct bch_fs *c = trans->c;
 	struct printbuf buf = PRINTBUF;
+	int ret = 0;
 
 	if (insert) {
 		prt_printf(&buf, "existing backpointer found when inserting ");
@@ -122,17 +126,15 @@ static noinline int backpointer_mod_err(struct btree_trans *trans,
 
 		prt_printf(&buf, "for ");
 		bch2_bkey_val_to_text(&buf, c, orig_k);
-
-		bch_err(c, "%s", buf.buf);
 	}
 
+	if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_extents_to_backpointers &&
+	    __bch2_inconsistent_error(c, &buf))
+		ret = -BCH_ERR_erofs_unfixed_errors;
+
+	bch_err(c, "%s", buf.buf);
 	printbuf_exit(&buf);
-
-	if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_extents_to_backpointers) {
-		return bch2_inconsistent_error(c) ? BCH_ERR_erofs_unfixed_errors : 0;
-	} else {
-		return 0;
-	}
+	return ret;
 }
 
 int bch2_bucket_backpointer_mod_nowritebuffer(struct btree_trans *trans,
@@ -207,11 +209,11 @@ static int backpointer_target_not_found(struct btree_trans *trans,
 	if (ret)
 		return ret;
 
-	prt_printf(&buf, "backpointer doesn't match %s it points to:\n  ",
+	prt_printf(&buf, "backpointer doesn't match %s it points to:\n",
 		   bp.v->level ? "btree node" : "extent");
 	bch2_bkey_val_to_text(&buf, c, bp.s_c);
 
-	prt_printf(&buf, "\n  ");
+	prt_newline(&buf);
 	bch2_bkey_val_to_text(&buf, c, target_k);
 
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(target_k);
@@ -219,7 +221,7 @@ static int backpointer_target_not_found(struct btree_trans *trans,
 	struct extent_ptr_decoded p;
 	bkey_for_each_ptr_decode(target_k.k, ptrs, p, entry)
 		if (p.ptr.dev == bp.k->p.inode) {
-			prt_printf(&buf, "\n  ");
+			prt_newline(&buf);
 			struct bkey_i_backpointer bp2;
 			bch2_extent_ptr_to_bp(c, bp.v->btree_id, bp.v->level, target_k, p, entry, &bp2);
 			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&bp2.k_i));
@@ -440,12 +442,11 @@ found:
 	if (ret)
 		goto err;
 
-	prt_str(&buf, "extents pointing to same space, but first extent checksum bad:");
-	prt_printf(&buf, "\n  ");
+	prt_printf(&buf, "extents pointing to same space, but first extent checksum bad:\n");
 	bch2_btree_id_to_text(&buf, btree);
 	prt_str(&buf, " ");
 	bch2_bkey_val_to_text(&buf, c, extent);
-	prt_printf(&buf, "\n  ");
+	prt_newline(&buf);
 	bch2_btree_id_to_text(&buf, o_btree);
 	prt_str(&buf, " ");
 	bch2_bkey_val_to_text(&buf, c, extent2);
@@ -518,11 +519,27 @@ check_existing_bp:
 	if (!other_extent.k)
 		goto missing;
 
+	rcu_read_lock();
+	struct bch_dev *ca = bch2_dev_rcu_noerror(c, bp->k.p.inode);
+	if (ca) {
+		struct bkey_ptrs_c other_extent_ptrs = bch2_bkey_ptrs_c(other_extent);
+		bkey_for_each_ptr(other_extent_ptrs, ptr)
+			if (ptr->dev == bp->k.p.inode &&
+			    dev_ptr_stale_rcu(ca, ptr)) {
+				ret = drop_dev_and_update(trans, other_bp.v->btree_id,
+							  other_extent, bp->k.p.inode);
+				if (ret)
+					goto err;
+				goto out;
+			}
+	}
+	rcu_read_unlock();
+
 	if (bch2_extents_match(orig_k, other_extent)) {
 		printbuf_reset(&buf);
-		prt_printf(&buf, "duplicate versions of same extent, deleting smaller\n  ");
+		prt_printf(&buf, "duplicate versions of same extent, deleting smaller\n");
 		bch2_bkey_val_to_text(&buf, c, orig_k);
-		prt_str(&buf, "\n  ");
+		prt_newline(&buf);
 		bch2_bkey_val_to_text(&buf, c, other_extent);
 		bch_err(c, "%s", buf.buf);
 
@@ -561,20 +578,20 @@ check_existing_bp:
 	}
 
 	printbuf_reset(&buf);
-	prt_printf(&buf, "duplicate extents pointing to same space on dev %llu\n  ", bp->k.p.inode);
+	prt_printf(&buf, "duplicate extents pointing to same space on dev %llu\n", bp->k.p.inode);
 	bch2_bkey_val_to_text(&buf, c, orig_k);
-	prt_str(&buf, "\n  ");
+	prt_newline(&buf);
 	bch2_bkey_val_to_text(&buf, c, other_extent);
 	bch_err(c, "%s", buf.buf);
 	ret = -BCH_ERR_fsck_repair_unimplemented;
 	goto err;
 missing:
 	printbuf_reset(&buf);
-	prt_str(&buf, "missing backpointer\n  for:  ");
+	prt_str(&buf, "missing backpointer\nfor:  ");
 	bch2_bkey_val_to_text(&buf, c, orig_k);
-	prt_printf(&buf, "\n  want: ");
+	prt_printf(&buf, "\nwant: ");
 	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&bp->k_i));
-	prt_printf(&buf, "\n  got:  ");
+	prt_printf(&buf, "\ngot:  ");
 	bch2_bkey_val_to_text(&buf, c, bp_k);
 
 	if (fsck_err(trans, ptr_to_missing_backpointer, "%s", buf.buf))
@@ -594,9 +611,6 @@ static int check_extent_to_backpointers(struct btree_trans *trans,
 	struct extent_ptr_decoded p;
 
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
-		if (p.ptr.cached)
-			continue;
-
 		if (p.ptr.dev == BCH_SB_MEMBER_INVALID)
 			continue;
 
@@ -604,9 +618,11 @@ static int check_extent_to_backpointers(struct btree_trans *trans,
 		struct bch_dev *ca = bch2_dev_rcu_noerror(c, p.ptr.dev);
 		bool check = ca && test_bit(PTR_BUCKET_NR(ca, &p.ptr), ca->bucket_backpointer_mismatches);
 		bool empty = ca && test_bit(PTR_BUCKET_NR(ca, &p.ptr), ca->bucket_backpointer_empty);
+
+		bool stale = p.ptr.cached && (!ca || dev_ptr_stale_rcu(ca, &p.ptr));
 		rcu_read_unlock();
 
-		if (check || empty) {
+		if ((check || empty) && !stale) {
 			struct bkey_i_backpointer bp;
 			bch2_extent_ptr_to_bp(c, btree, level, k, p, entry, &bp);
 
@@ -719,71 +735,6 @@ static int bch2_get_btree_in_memory_pos(struct btree_trans *trans,
 	return ret;
 }
 
-struct progress_indicator_state {
-	unsigned long		next_print;
-	u64			nodes_seen;
-	u64			nodes_total;
-	struct btree		*last_node;
-};
-
-static inline void progress_init(struct progress_indicator_state *s,
-				 struct bch_fs *c,
-				 u64 btree_id_mask)
-{
-	memset(s, 0, sizeof(*s));
-
-	s->next_print = jiffies + HZ * 10;
-
-	for (unsigned i = 0; i < BTREE_ID_NR; i++) {
-		if (!(btree_id_mask & BIT_ULL(i)))
-			continue;
-
-		struct disk_accounting_pos acc = {
-			.type		= BCH_DISK_ACCOUNTING_btree,
-			.btree.id	= i,
-		};
-
-		u64 v;
-		bch2_accounting_mem_read(c, disk_accounting_pos_to_bpos(&acc), &v, 1);
-		s->nodes_total += div64_ul(v, btree_sectors(c));
-	}
-}
-
-static inline bool progress_update_p(struct progress_indicator_state *s)
-{
-	bool ret = time_after_eq(jiffies, s->next_print);
-
-	if (ret)
-		s->next_print = jiffies + HZ * 10;
-	return ret;
-}
-
-static void progress_update_iter(struct btree_trans *trans,
-				 struct progress_indicator_state *s,
-				 struct btree_iter *iter,
-				 const char *msg)
-{
-	struct bch_fs *c = trans->c;
-	struct btree *b = path_l(btree_iter_path(trans, iter))->b;
-
-	s->nodes_seen += b != s->last_node;
-	s->last_node = b;
-
-	if (progress_update_p(s)) {
-		struct printbuf buf = PRINTBUF;
-		unsigned percent = s->nodes_total
-			? div64_u64(s->nodes_seen * 100, s->nodes_total)
-			: 0;
-
-		prt_printf(&buf, "%s: %d%%, done %llu/%llu nodes, at ",
-			   msg, percent, s->nodes_seen, s->nodes_total);
-		bch2_bbpos_to_text(&buf, BBPOS(iter->btree_id, iter->pos));
-
-		bch_info(c, "%s", buf.buf);
-		printbuf_exit(&buf);
-	}
-}
-
 static int bch2_check_extents_to_backpointers_pass(struct btree_trans *trans,
 						   struct extents_to_bp_state *s)
 {
@@ -791,7 +742,7 @@ static int bch2_check_extents_to_backpointers_pass(struct btree_trans *trans,
 	struct progress_indicator_state progress;
 	int ret = 0;
 
-	progress_init(&progress, trans->c, BIT_ULL(BTREE_ID_extents)|BIT_ULL(BTREE_ID_reflink));
+	bch2_progress_init(&progress, trans->c, BIT_ULL(BTREE_ID_extents)|BIT_ULL(BTREE_ID_reflink));
 
 	for (enum btree_id btree_id = 0;
 	     btree_id < btree_id_nr_alive(c);
@@ -810,7 +761,7 @@ static int bch2_check_extents_to_backpointers_pass(struct btree_trans *trans,
 						  BTREE_ITER_prefetch);
 
 			ret = for_each_btree_key_continue(trans, iter, 0, k, ({
-				progress_update_iter(trans, &progress, &iter, "extents_to_backpointers");
+				bch2_progress_update_iter(trans, &progress, &iter, "extents_to_backpointers");
 				check_extent_to_backpointers(trans, s, btree_id, level, k) ?:
 				bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
 			}));
@@ -831,7 +782,7 @@ enum alloc_sector_counter {
 	ALLOC_SECTORS_NR
 };
 
-static enum alloc_sector_counter data_type_to_alloc_counter(enum bch_data_type t)
+static int data_type_to_alloc_counter(enum bch_data_type t)
 {
 	switch (t) {
 	case BCH_DATA_btree:
@@ -840,9 +791,10 @@ static enum alloc_sector_counter data_type_to_alloc_counter(enum bch_data_type t
 	case BCH_DATA_cached:
 		return ALLOC_cached;
 	case BCH_DATA_stripe:
+	case BCH_DATA_parity:
 		return ALLOC_stripe;
 	default:
-		BUG();
+		return -1;
 	}
 }
 
@@ -893,7 +845,11 @@ static int check_bucket_backpointer_mismatch(struct btree_trans *trans, struct b
 		if (bp.v->bucket_gen != a->gen)
 			continue;
 
-		sectors[data_type_to_alloc_counter(bp.v->data_type)] += bp.v->bucket_len;
+		int alloc_counter = data_type_to_alloc_counter(bp.v->data_type);
+		if (alloc_counter < 0)
+			continue;
+
+		sectors[alloc_counter] += bp.v->bucket_len;
 	};
 	bch2_trans_iter_exit(trans, &iter);
 	if (ret)
@@ -905,9 +861,8 @@ static int check_bucket_backpointer_mismatch(struct btree_trans *trans, struct b
 			goto err;
 	}
 
-	/* Cached pointers don't have backpointers: */
-
 	if (sectors[ALLOC_dirty]  != a->dirty_sectors ||
+	    sectors[ALLOC_cached] != a->cached_sectors ||
 	    sectors[ALLOC_stripe] != a->stripe_sectors) {
 		if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_backpointer_bucket_gen) {
 			ret = bch2_backpointers_maybe_flush(trans, alloc_k, last_flushed);
@@ -916,6 +871,7 @@ static int check_bucket_backpointer_mismatch(struct btree_trans *trans, struct b
 		}
 
 		if (sectors[ALLOC_dirty]  > a->dirty_sectors ||
+		    sectors[ALLOC_cached] > a->cached_sectors ||
 		    sectors[ALLOC_stripe] > a->stripe_sectors) {
 			ret = check_bucket_backpointers_to_extents(trans, ca, alloc_k.k->p) ?:
 				-BCH_ERR_transaction_restart_nested;
@@ -923,7 +879,8 @@ static int check_bucket_backpointer_mismatch(struct btree_trans *trans, struct b
 		}
 
 		if (!sectors[ALLOC_dirty] &&
-		    !sectors[ALLOC_stripe])
+		    !sectors[ALLOC_stripe] &&
+		    !sectors[ALLOC_cached])
 			__set_bit(alloc_k.k->p.offset, ca->bucket_backpointer_empty);
 		else
 			__set_bit(alloc_k.k->p.offset, ca->bucket_backpointer_mismatches);
@@ -1064,7 +1021,7 @@ int bch2_check_extents_to_backpointers(struct bch_fs *c)
 	 * Can't allow devices to come/go/resize while we have bucket bitmaps
 	 * allocated
 	 */
-	lockdep_assert_held(&c->state_lock);
+	down_read(&c->state_lock);
 
 	for_each_member_device(c, ca) {
 		BUG_ON(ca->bucket_backpointer_mismatches);
@@ -1149,6 +1106,7 @@ err_free_bitmaps:
 		ca->bucket_backpointer_mismatches = NULL;
 	}
 
+	up_read(&c->state_lock);
 	bch_err_fn(c, ret);
 	return ret;
 }
@@ -1210,11 +1168,11 @@ static int bch2_check_backpointers_to_extents_pass(struct btree_trans *trans,
 
 	bch2_bkey_buf_init(&last_flushed);
 	bkey_init(&last_flushed.k->k);
-	progress_init(&progress, trans->c, BIT_ULL(BTREE_ID_backpointers));
+	bch2_progress_init(&progress, trans->c, BIT_ULL(BTREE_ID_backpointers));
 
 	int ret = for_each_btree_key(trans, iter, BTREE_ID_backpointers,
 				     POS_MIN, BTREE_ITER_prefetch, k, ({
-			progress_update_iter(trans, &progress, &iter, "backpointers_to_extents");
+			bch2_progress_update_iter(trans, &progress, &iter, "backpointers_to_extents");
 			check_one_backpointer(trans, start, end, k, &last_flushed);
 	}));
 

libbcachefs/backpointers.h

@@ -1,6 +1,6 @@
 /* SPDX-License-Identifier: GPL-2.0 */
-#ifndef _BCACHEFS_BACKPOINTERS_BACKGROUND_H
-#define _BCACHEFS_BACKPOINTERS_BACKGROUND_H
+#ifndef _BCACHEFS_BACKPOINTERS_H
+#define _BCACHEFS_BACKPOINTERS_H
 
 #include "btree_cache.h"
 #include "btree_iter.h"
@@ -123,7 +123,12 @@ static inline enum bch_data_type bch2_bkey_ptr_data_type(struct bkey_s_c k,
 		return BCH_DATA_btree;
 	case KEY_TYPE_extent:
 	case KEY_TYPE_reflink_v:
-		return p.has_ec ? BCH_DATA_stripe : BCH_DATA_user;
+		if (p.has_ec)
+			return BCH_DATA_stripe;
+		if (p.ptr.cached)
+			return BCH_DATA_cached;
+		else
+			return BCH_DATA_user;
 	case KEY_TYPE_stripe: {
 		const struct bch_extent_ptr *ptr = &entry->ptr;
 		struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
@@ -147,7 +152,20 @@ static inline void bch2_extent_ptr_to_bp(struct bch_fs *c,
 			   struct bkey_i_backpointer *bp)
 {
 	bkey_backpointer_init(&bp->k_i);
-	bp->k.p = POS(p.ptr.dev, ((u64) p.ptr.offset << MAX_EXTENT_COMPRESS_RATIO_SHIFT) + p.crc.offset);
+	bp->k.p.inode = p.ptr.dev;
+
+	if (k.k->type != KEY_TYPE_stripe)
+		bp->k.p.offset = ((u64) p.ptr.offset << MAX_EXTENT_COMPRESS_RATIO_SHIFT) + p.crc.offset;
+	else {
+		/*
+		 * Put stripe backpointers where they won't collide with the
+		 * extent backpointers within the stripe:
+		 */
+		struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
+		bp->k.p.offset = ((u64) (p.ptr.offset + le16_to_cpu(s.v->sectors)) <<
+				  MAX_EXTENT_COMPRESS_RATIO_SHIFT) - 1;
+	}
+
 	bp->v	= (struct bch_backpointer) {
 		.btree_id	= btree_id,
 		.level		= level,

libbcachefs/bcachefs.h

@@ -203,6 +203,7 @@
 #include <linux/types.h>
 #include <linux/workqueue.h>
 #include <linux/zstd.h>
+#include <linux/unicode.h>
 
 #include "bcachefs_format.h"
 #include "btree_journal_iter_types.h"
@@ -444,6 +445,7 @@ BCH_DEBUG_PARAMS_DEBUG()
 	x(btree_node_sort)			\
 	x(btree_node_read)			\
 	x(btree_node_read_done)			\
+	x(btree_node_write)			\
 	x(btree_interior_update_foreground)	\
 	x(btree_interior_update_total)		\
 	x(btree_gc)				\
@@ -456,6 +458,7 @@ BCH_DEBUG_PARAMS_DEBUG()
 	x(blocked_journal_low_on_space)		\
 	x(blocked_journal_low_on_pin)		\
 	x(blocked_journal_max_in_flight)	\
+	x(blocked_journal_max_open)		\
 	x(blocked_key_cache_flush)		\
 	x(blocked_allocate)			\
 	x(blocked_allocate_open_bucket)		\
@@ -533,6 +536,7 @@ struct bch_dev {
 	 */
 	struct bch_member_cpu	mi;
 	atomic64_t		errors[BCH_MEMBER_ERROR_NR];
+	unsigned long		write_errors_start;
 
 	__uuid_t		uuid;
 	char			name[BDEVNAME_SIZE];
@@ -623,7 +627,8 @@ struct bch_dev {
 	x(topology_error)		\
 	x(errors_fixed)			\
 	x(errors_not_fixed)		\
-	x(no_invalid_checks)
+	x(no_invalid_checks)		\
+	x(discard_mount_opt_set)	\
 
 enum bch_fs_flags {
 #define x(n)		BCH_FS_##n,
@@ -697,6 +702,8 @@ enum bch_write_ref {
 	BCH_WRITE_REF_NR,
 };
 
+#define BCH_FS_DEFAULT_UTF8_ENCODING UNICODE_AGE(12, 1, 0)
+
 struct bch_fs {
 	struct closure		cl;
 
@@ -781,6 +788,9 @@ struct bch_fs {
 		u64		btrees_lost_data;
 	}			sb;
 
+#ifdef CONFIG_UNICODE
+	struct unicode_map	*cf_encoding;
+#endif
 
 	struct bch_sb_handle	disk_sb;
 
@@ -970,7 +980,6 @@ struct bch_fs {
 	mempool_t		compress_workspace[BCH_COMPRESSION_OPT_NR];
 	size_t			zstd_workspace_size;
 
-	struct crypto_shash	*sha256;
 	struct crypto_sync_skcipher *chacha20;
 	struct crypto_shash	*poly1305;
 
@@ -994,15 +1003,11 @@ struct bch_fs {
 	wait_queue_head_t	copygc_running_wq;
 
 	/* STRIPES: */
-	GENRADIX(struct stripe) stripes;
 	GENRADIX(struct gc_stripe) gc_stripes;
 
 	struct hlist_head	ec_stripes_new[32];
 	spinlock_t		ec_stripes_new_lock;
 
-	ec_stripes_heap		ec_stripes_heap;
-	struct mutex		ec_stripes_heap_lock;
-
 	/* ERASURE CODING */
 	struct list_head	ec_stripe_head_list;
 	struct mutex		ec_stripe_head_lock;

libbcachefs/bcachefs_format.h

@@ -686,7 +686,12 @@ struct bch_sb_field_ext {
 	x(inode_depth,			BCH_VERSION(1, 17))		\
 	x(persistent_inode_cursors,	BCH_VERSION(1, 18))		\
 	x(autofix_errors,		BCH_VERSION(1, 19))		\
-	x(directory_size,		BCH_VERSION(1, 20))
+	x(directory_size,		BCH_VERSION(1, 20))		\
+	x(cached_backpointers,		BCH_VERSION(1, 21))		\
+	x(stripe_backpointers,		BCH_VERSION(1, 22))		\
+	x(stripe_lru,			BCH_VERSION(1, 23))		\
+	x(casefolding,			BCH_VERSION(1, 24))		\
+	x(extent_flags,			BCH_VERSION(1, 25))
 
 enum bcachefs_metadata_version {
 	bcachefs_metadata_version_min = 9,
@@ -837,6 +842,7 @@ LE64_BITMASK(BCH_SB_SHARD_INUMS,	struct bch_sb, flags[3], 28, 29);
 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
+/* one free bit */
 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
 LE64_BITMASK(BCH_SB_NOCOW,		struct bch_sb, flags[4], 33, 34);
@@ -855,6 +861,8 @@ LE64_BITMASK(BCH_SB_VERSION_INCOMPAT,	struct bch_sb, flags[5], 32, 48);
 LE64_BITMASK(BCH_SB_VERSION_INCOMPAT_ALLOWED,
 					struct bch_sb, flags[5], 48, 64);
 LE64_BITMASK(BCH_SB_SHARD_INUMS_NBITS,	struct bch_sb, flags[6],  0,  4);
+LE64_BITMASK(BCH_SB_WRITE_ERROR_TIMEOUT,struct bch_sb, flags[6],  4, 14);
+LE64_BITMASK(BCH_SB_CSUM_ERR_RETRY_NR,	struct bch_sb, flags[6], 14, 20);
 
 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
 {
@@ -908,7 +916,8 @@ static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u
 	x(journal_no_flush,		16)	\
 	x(alloc_v2,			17)	\
 	x(extents_across_btree_nodes,	18)	\
-	x(incompat_version_field,	19)
+	x(incompat_version_field,	19)	\
+	x(casefolding,			20)
 
 #define BCH_SB_FEATURES_ALWAYS				\
 	(BIT_ULL(BCH_FEATURE_new_extent_overwrite)|	\
@@ -922,7 +931,8 @@ static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u
 	 BIT_ULL(BCH_FEATURE_new_siphash)|		\
 	 BIT_ULL(BCH_FEATURE_btree_ptr_v2)|		\
 	 BIT_ULL(BCH_FEATURE_new_varint)|		\
-	 BIT_ULL(BCH_FEATURE_journal_no_flush))
+	 BIT_ULL(BCH_FEATURE_journal_no_flush)|		\
+	 BIT_ULL(BCH_FEATURE_incompat_version_field))
 
 enum bch_sb_feature {
 #define x(f, n) BCH_FEATURE_##f,
@@ -1133,7 +1143,8 @@ static inline __u64 __bset_magic(struct bch_sb *sb)
 	x(log,			9)		\
 	x(overwrite,		10)		\
 	x(write_buffer_keys,	11)		\
-	x(datetime,		12)
+	x(datetime,		12)		\
+	x(log_bkey,		13)
 
 enum bch_jset_entry_type {
 #define x(f, nr)	BCH_JSET_ENTRY_##f	= nr,

libbcachefs/bcachefs_ioctl.h

@@ -234,6 +234,11 @@ enum bch_data_event {
 	BCH_DATA_EVENT_NR	= 1,
 };
 
+enum data_progress_data_type_special {
+	DATA_PROGRESS_DATA_TYPE_phys	= 254,
+	DATA_PROGRESS_DATA_TYPE_done	= 255,
+};
+
 struct bch_ioctl_data_progress {
 	__u8			data_type;
 	__u8			btree_id;

libbcachefs/btree_cache.c

@@ -203,7 +203,7 @@ struct btree *__bch2_btree_node_mem_alloc(struct bch_fs *c)
 		return NULL;
 	}
 
-	bch2_btree_lock_init(&b->c, 0);
+	bch2_btree_lock_init(&b->c, 0, GFP_KERNEL);
 
 	__bch2_btree_node_to_freelist(bc, b);
 	return b;
@@ -610,6 +610,7 @@ void bch2_fs_btree_cache_exit(struct bch_fs *c)
 		       btree_node_write_in_flight(b));
 
 		btree_node_data_free(bc, b);
+		cond_resched();
 	}
 
 	BUG_ON(!bch2_journal_error(&c->journal) &&
@@ -795,17 +796,18 @@ struct btree *bch2_btree_node_mem_alloc(struct btree_trans *trans, bool pcpu_rea
 		}
 
 	b = __btree_node_mem_alloc(c, GFP_NOWAIT|__GFP_NOWARN);
-	if (!b) {
+	if (b) {
+		bch2_btree_lock_init(&b->c, pcpu_read_locks ? SIX_LOCK_INIT_PCPU : 0, GFP_NOWAIT);
+	} else {
 		mutex_unlock(&bc->lock);
 		bch2_trans_unlock(trans);
 		b = __btree_node_mem_alloc(c, GFP_KERNEL);
 		if (!b)
 			goto err;
+		bch2_btree_lock_init(&b->c, pcpu_read_locks ? SIX_LOCK_INIT_PCPU : 0, GFP_KERNEL);
 		mutex_lock(&bc->lock);
 	}
 
-	bch2_btree_lock_init(&b->c, pcpu_read_locks ? SIX_LOCK_INIT_PCPU : 0);
-
 	BUG_ON(!six_trylock_intent(&b->c.lock));
 	BUG_ON(!six_trylock_write(&b->c.lock));
 
@@ -1415,7 +1417,7 @@ void __bch2_btree_pos_to_text(struct printbuf *out, struct bch_fs *c,
 		prt_printf(out, "%u", r->level);
 	else
 		prt_printf(out, "(unknown)");
-	prt_printf(out, "\n  ");
+	prt_newline(out);
 
 	bch2_bkey_val_to_text(out, c, k);
 }

libbcachefs/btree_gc.c

@@ -27,6 +27,7 @@
 #include "journal.h"
 #include "keylist.h"
 #include "move.h"
+#include "progress.h"
 #include "recovery_passes.h"
 #include "reflink.h"
 #include "recovery.h"
@@ -212,15 +213,15 @@ static int btree_check_node_boundaries(struct btree_trans *trans, struct btree *
 
 	prt_printf(&buf, "  at ");
 	bch2_btree_id_level_to_text(&buf, b->c.btree_id, b->c.level);
-	prt_printf(&buf, ":\n  parent: ");
+	prt_printf(&buf, ":\nparent: ");
 	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
 
 	if (prev) {
-		prt_printf(&buf, "\n  prev: ");
+		prt_printf(&buf, "\nprev: ");
 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&prev->key));
 	}
 
-	prt_str(&buf, "\n  next: ");
+	prt_str(&buf, "\nnext: ");
 	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&cur->key));
 
 	if (bpos_lt(expected_start, cur->data->min_key)) {				/* gap */
@@ -279,12 +280,12 @@ static int btree_repair_node_end(struct btree_trans *trans, struct btree *b,
 	if (bpos_eq(child->key.k.p, b->key.k.p))
 		return 0;
 
-	prt_printf(&buf, "  at ");
+	prt_printf(&buf, "\nat: ");
 	bch2_btree_id_level_to_text(&buf, b->c.btree_id, b->c.level);
-	prt_printf(&buf, ":\n  parent: ");
+	prt_printf(&buf, "\nparent: ");
 	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
 
-	prt_str(&buf, "\n  child: ");
+	prt_str(&buf, "\nchild: ");
 	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&child->key));
 
 	if (mustfix_fsck_err(trans, btree_node_topology_bad_max_key,
@@ -350,8 +351,7 @@ again:
 
 		if (mustfix_fsck_err_on(bch2_err_matches(ret, EIO),
 				trans, btree_node_read_error,
-				"Topology repair: unreadable btree node at\n"
-				"  %s",
+				"Topology repair: unreadable btree node at\n%s",
 				buf.buf)) {
 			bch2_btree_node_evict(trans, cur_k.k);
 			cur = NULL;
@@ -611,7 +611,7 @@ static int bch2_gc_mark_key(struct btree_trans *trans, enum btree_id btree_id,
 		if (fsck_err_on(btree_id != BTREE_ID_accounting &&
 				k.k->bversion.lo > atomic64_read(&c->key_version),
 				trans, bkey_version_in_future,
-				"key version number higher than recorded %llu\n  %s",
+				"key version number higher than recorded %llu\n%s",
 				atomic64_read(&c->key_version),
 				(bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
 			atomic64_set(&c->key_version, k.k->bversion.lo);
@@ -619,7 +619,7 @@ static int bch2_gc_mark_key(struct btree_trans *trans, enum btree_id btree_id,
 
 	if (mustfix_fsck_err_on(level && !bch2_dev_btree_bitmap_marked(c, k),
 				trans, btree_bitmap_not_marked,
-				"btree ptr not marked in member info btree allocated bitmap\n  %s",
+				"btree ptr not marked in member info btree allocated bitmap\n%s",
 				(printbuf_reset(&buf),
 				 bch2_bkey_val_to_text(&buf, c, k),
 				 buf.buf))) {
@@ -656,7 +656,9 @@ fsck_err:
 	return ret;
 }
 
-static int bch2_gc_btree(struct btree_trans *trans, enum btree_id btree, bool initial)
+static int bch2_gc_btree(struct btree_trans *trans,
+			 struct progress_indicator_state *progress,
+			 enum btree_id btree, bool initial)
 {
 	struct bch_fs *c = trans->c;
 	unsigned target_depth = btree_node_type_has_triggers(__btree_node_type(0, btree)) ? 0 : 1;
@@ -673,6 +675,7 @@ static int bch2_gc_btree(struct btree_trans *trans, enum btree_id btree, bool in
 					  BTREE_ITER_prefetch);
 
 		ret = for_each_btree_key_continue(trans, iter, 0, k, ({
+			bch2_progress_update_iter(trans, progress, &iter, "check_allocations");
 			gc_pos_set(c, gc_pos_btree(btree, level, k.k->p));
 			bch2_gc_mark_key(trans, btree, level, &prev, &iter, k, initial);
 		}));
@@ -717,22 +720,24 @@ static inline int btree_id_gc_phase_cmp(enum btree_id l, enum btree_id r)
 static int bch2_gc_btrees(struct bch_fs *c)
 {
 	struct btree_trans *trans = bch2_trans_get(c);
-	enum btree_id ids[BTREE_ID_NR];
 	struct printbuf buf = PRINTBUF;
-	unsigned i;
 	int ret = 0;
 
-	for (i = 0; i < BTREE_ID_NR; i++)
+	struct progress_indicator_state progress;
+	bch2_progress_init(&progress, c, ~0ULL);
+
+	enum btree_id ids[BTREE_ID_NR];
+	for (unsigned i = 0; i < BTREE_ID_NR; i++)
 		ids[i] = i;
 	bubble_sort(ids, BTREE_ID_NR, btree_id_gc_phase_cmp);
 
-	for (i = 0; i < btree_id_nr_alive(c) && !ret; i++) {
+	for (unsigned i = 0; i < btree_id_nr_alive(c) && !ret; i++) {
 		unsigned btree = i < BTREE_ID_NR ? ids[i] : i;
 
 		if (IS_ERR_OR_NULL(bch2_btree_id_root(c, btree)->b))
 			continue;
 
-		ret = bch2_gc_btree(trans, btree, true);
+		ret = bch2_gc_btree(trans, &progress, btree, true);
 	}
 
 	printbuf_exit(&buf);
@@ -1015,8 +1020,7 @@ int bch2_check_allocations(struct bch_fs *c)
 {
 	int ret;
 
-	lockdep_assert_held(&c->state_lock);
-
+	down_read(&c->state_lock);
 	down_write(&c->gc_lock);
 
 	bch2_btree_interior_updates_flush(c);
@@ -1054,6 +1058,7 @@ out:
 	percpu_up_write(&c->mark_lock);
 
 	up_write(&c->gc_lock);
+	up_read(&c->state_lock);
 
 	/*
 	 * At startup, allocations can happen directly instead of via the

libbcachefs/btree_io.c

@@ -525,8 +525,6 @@ static void btree_err_msg(struct printbuf *out, struct bch_fs *c,
 	prt_printf(out, "at btree ");
 	bch2_btree_pos_to_text(out, c, b);
 
-	printbuf_indent_add(out, 2);
-
 	prt_printf(out, "\nnode offset %u/%u",
 		   b->written, btree_ptr_sectors_written(bkey_i_to_s_c(&b->key)));
 	if (i)
@@ -550,23 +548,7 @@ static int __btree_err(int ret,
 		       enum bch_sb_error_id err_type,
 		       const char *fmt, ...)
 {
-	struct printbuf out = PRINTBUF;
 	bool silent = c->curr_recovery_pass == BCH_RECOVERY_PASS_scan_for_btree_nodes;
-	va_list args;
-
-	btree_err_msg(&out, c, ca, b, i, k, b->written, write);
-
-	va_start(args, fmt);
-	prt_vprintf(&out, fmt, args);
-	va_end(args);
-
-	if (write == WRITE) {
-		bch2_print_string_as_lines(KERN_ERR, out.buf);
-		ret = c->opts.errors == BCH_ON_ERROR_continue
-			? 0
-			: -BCH_ERR_fsck_errors_not_fixed;
-		goto out;
-	}
 
 	if (!have_retry && ret == -BCH_ERR_btree_node_read_err_want_retry)
 		ret = -BCH_ERR_btree_node_read_err_fixable;
@@ -576,6 +558,29 @@ static int __btree_err(int ret,
 	if (!silent && ret != -BCH_ERR_btree_node_read_err_fixable)
 		bch2_sb_error_count(c, err_type);
 
+	struct printbuf out = PRINTBUF;
+	if (write != WRITE && ret != -BCH_ERR_btree_node_read_err_fixable) {
+		printbuf_indent_add_nextline(&out, 2);
+#ifdef BCACHEFS_LOG_PREFIX
+		prt_printf(&out, bch2_log_msg(c, ""));
+#endif
+	}
+
+	btree_err_msg(&out, c, ca, b, i, k, b->written, write);
+
+	va_list args;
+	va_start(args, fmt);
+	prt_vprintf(&out, fmt, args);
+	va_end(args);
+
+	if (write == WRITE) {
+		prt_str(&out, ", ");
+		ret = __bch2_inconsistent_error(c, &out)
+			? -BCH_ERR_fsck_errors_not_fixed
+			: 0;
+		silent = false;
+	}
+
 	switch (ret) {
 	case -BCH_ERR_btree_node_read_err_fixable:
 		ret = !silent
@@ -585,25 +590,21 @@ static int __btree_err(int ret,
 		    ret != -BCH_ERR_fsck_ignore)
 			goto fsck_err;
 		ret = -BCH_ERR_fsck_fix;
-		break;
-	case -BCH_ERR_btree_node_read_err_want_retry:
-	case -BCH_ERR_btree_node_read_err_must_retry:
-		if (!silent)
-			bch2_print_string_as_lines(KERN_ERR, out.buf);
-		break;
+		goto out;
 	case -BCH_ERR_btree_node_read_err_bad_node:
-		if (!silent)
-			bch2_print_string_as_lines(KERN_ERR, out.buf);
-		ret = bch2_topology_error(c);
+		prt_str(&out, ", ");
+		ret = __bch2_topology_error(c, &out);
+		if (ret)
+			silent = false;
 		break;
 	case -BCH_ERR_btree_node_read_err_incompatible:
-		if (!silent)
-			bch2_print_string_as_lines(KERN_ERR, out.buf);
 		ret = -BCH_ERR_fsck_errors_not_fixed;
+		silent = false;
 		break;
-	default:
-		BUG();
 	}
+
+	if (!silent)
+		bch2_print_string_as_lines(KERN_ERR, out.buf);
 out:
 fsck_err:
 	printbuf_exit(&out);
@@ -817,7 +818,7 @@ static int validate_bset(struct bch_fs *c, struct bch_dev *ca,
 			     -BCH_ERR_btree_node_read_err_bad_node,
 			     c, ca, b, i, NULL,
 			     btree_node_bad_format,
-			     "invalid bkey format: %s\n  %s", buf1.buf,
+			     "invalid bkey format: %s\n%s", buf1.buf,
 			     (printbuf_reset(&buf2),
 			      bch2_bkey_format_to_text(&buf2, &bn->format), buf2.buf));
 		printbuf_reset(&buf1);
@@ -997,7 +998,7 @@ drop_this_key:
 		}
 got_good_key:
 		le16_add_cpu(&i->u64s, -next_good_key);
-		memmove_u64s_down(k, bkey_p_next(k), (u64 *) vstruct_end(i) - (u64 *) k);
+		memmove_u64s_down(k, (u64 *) k + next_good_key, (u64 *) vstruct_end(i) - (u64 *) k);
 		set_btree_node_need_rewrite(b);
 	}
 fsck_err:
@@ -1187,7 +1188,7 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
 			     le64_to_cpu(i->journal_seq),
 			     b->written, b->written + sectors, ptr_written);
 
-		b->written += sectors;
+		b->written = min(b->written + sectors, btree_sectors(c));
 
 		if (blacklisted && !first)
 			continue;
@@ -1329,6 +1330,7 @@ static void btree_node_read_work(struct work_struct *work)
 		bch_info(c, "retrying read");
 		ca = bch2_dev_get_ioref(c, rb->pick.ptr.dev, READ);
 		rb->have_ioref		= ca != NULL;
+		rb->start_time		= local_clock();
 		bio_reset(bio, NULL, REQ_OP_READ|REQ_SYNC|REQ_META);
 		bio->bi_iter.bi_sector	= rb->pick.ptr.offset;
 		bio->bi_iter.bi_size	= btree_buf_bytes(b);
@@ -1339,17 +1341,22 @@ static void btree_node_read_work(struct work_struct *work)
 		} else {
 			bio->bi_status = BLK_STS_REMOVED;
 		}
+
+		bch2_account_io_completion(ca, BCH_MEMBER_ERROR_read,
+					   rb->start_time, !bio->bi_status);
 start:
 		printbuf_reset(&buf);
 		bch2_btree_pos_to_text(&buf, c, b);
-		bch2_dev_io_err_on(ca && bio->bi_status, ca, BCH_MEMBER_ERROR_read,
-				   "btree read error %s for %s",
-				   bch2_blk_status_to_str(bio->bi_status), buf.buf);
+
+		if (ca && bio->bi_status)
+			bch_err_dev_ratelimited(ca,
+					"btree read error %s for %s",
+					bch2_blk_status_to_str(bio->bi_status), buf.buf);
 		if (rb->have_ioref)
 			percpu_ref_put(&ca->io_ref);
 		rb->have_ioref = false;
 
-		bch2_mark_io_failure(&failed, &rb->pick);
+		bch2_mark_io_failure(&failed, &rb->pick, false);
 
 		can_retry = bch2_bkey_pick_read_device(c,
 				bkey_i_to_s_c(&b->key),
@@ -1401,12 +1408,11 @@ static void btree_node_read_endio(struct bio *bio)
 	struct btree_read_bio *rb =
 		container_of(bio, struct btree_read_bio, bio);
 	struct bch_fs *c	= rb->c;
+	struct bch_dev *ca	= rb->have_ioref
+		? bch2_dev_have_ref(c, rb->pick.ptr.dev) : NULL;
 
-	if (rb->have_ioref) {
-		struct bch_dev *ca = bch2_dev_have_ref(c, rb->pick.ptr.dev);
-
-		bch2_latency_acct(ca, rb->start_time, READ);
-	}
+	bch2_account_io_completion(ca, BCH_MEMBER_ERROR_read,
+				   rb->start_time, !bio->bi_status);
 
 	queue_work(c->btree_read_complete_wq, &rb->work);
 }
@@ -2016,7 +2022,7 @@ static void bch2_btree_complete_write(struct bch_fs *c, struct btree *b,
 	bch2_journal_pin_drop(&c->journal, &w->journal);
 }
 
-static void __btree_node_write_done(struct bch_fs *c, struct btree *b)
+static void __btree_node_write_done(struct bch_fs *c, struct btree *b, u64 start_time)
 {
 	struct btree_write *w = btree_prev_write(b);
 	unsigned long old, new;
@@ -2024,6 +2030,9 @@ static void __btree_node_write_done(struct bch_fs *c, struct btree *b)
 
 	bch2_btree_complete_write(c, b, w);
 
+	if (start_time)
+		bch2_time_stats_update(&c->times[BCH_TIME_btree_node_write], start_time);
+
 	old = READ_ONCE(b->flags);
 	do {
 		new = old;
@@ -2054,7 +2063,7 @@ static void __btree_node_write_done(struct bch_fs *c, struct btree *b)
 		wake_up_bit(&b->flags, BTREE_NODE_write_in_flight);
 }
 
-static void btree_node_write_done(struct bch_fs *c, struct btree *b)
+static void btree_node_write_done(struct bch_fs *c, struct btree *b, u64 start_time)
 {
 	struct btree_trans *trans = bch2_trans_get(c);
 
@@ -2062,7 +2071,7 @@ static void btree_node_write_done(struct bch_fs *c, struct btree *b)
 
 	/* we don't need transaction context anymore after we got the lock. */
 	bch2_trans_put(trans);
-	__btree_node_write_done(c, b);
+	__btree_node_write_done(c, b, start_time);
 	six_unlock_read(&b->c.lock);
 }
 
@@ -2072,6 +2081,7 @@ static void btree_node_write_work(struct work_struct *work)
 		container_of(work, struct btree_write_bio, work);
 	struct bch_fs *c	= wbio->wbio.c;
 	struct btree *b		= wbio->wbio.bio.bi_private;
+	u64 start_time		= wbio->start_time;
 	int ret = 0;
 
 	btree_bounce_free(c,
@@ -2104,12 +2114,18 @@ static void btree_node_write_work(struct work_struct *work)
 	}
 out:
 	bio_put(&wbio->wbio.bio);
-	btree_node_write_done(c, b);
+	btree_node_write_done(c, b, start_time);
 	return;
 err:
 	set_btree_node_noevict(b);
-	bch2_fs_fatal_err_on(!bch2_err_matches(ret, EROFS), c,
-			     "writing btree node: %s", bch2_err_str(ret));
+
+	if (!bch2_err_matches(ret, EROFS)) {
+		struct printbuf buf = PRINTBUF;
+		prt_printf(&buf, "writing btree node: %s\n  ", bch2_err_str(ret));
+		bch2_btree_pos_to_text(&buf, c, b);
+		bch2_fs_fatal_error(c, "%s", buf.buf);
+		printbuf_exit(&buf);
+	}
 	goto out;
 }
 
@@ -2122,16 +2138,21 @@ static void btree_node_write_endio(struct bio *bio)
 	struct bch_fs *c		= wbio->c;
 	struct btree *b			= wbio->bio.bi_private;
 	struct bch_dev *ca		= wbio->have_ioref ? bch2_dev_have_ref(c, wbio->dev) : NULL;
-	unsigned long flags;
 
-	if (wbio->have_ioref)
-		bch2_latency_acct(ca, wbio->submit_time, WRITE);
+	bch2_account_io_completion(ca, BCH_MEMBER_ERROR_write,
+				   wbio->submit_time, !bio->bi_status);
 
-	if (!ca ||
-	    bch2_dev_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_write,
-			       "btree write error: %s",
-			       bch2_blk_status_to_str(bio->bi_status)) ||
-	    bch2_meta_write_fault("btree")) {
+	if (ca && bio->bi_status) {
+		struct printbuf buf = PRINTBUF;
+		prt_printf(&buf, "btree write error: %s\n  ",
+			   bch2_blk_status_to_str(bio->bi_status));
+		bch2_btree_pos_to_text(&buf, c, b);
+		bch_err_dev_ratelimited(ca, "%s", buf.buf);
+		printbuf_exit(&buf);
+	}
+
+	if (bio->bi_status) {
+		unsigned long flags;
 		spin_lock_irqsave(&c->btree_write_error_lock, flags);
 		bch2_dev_list_add_dev(&orig->failed, wbio->dev);
 		spin_unlock_irqrestore(&c->btree_write_error_lock, flags);
@@ -2208,6 +2229,7 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b, unsigned flags)
 	bool validate_before_checksum = false;
 	enum btree_write_type type = flags & BTREE_WRITE_TYPE_MASK;
 	void *data;
+	u64 start_time = local_clock();
 	int ret;
 
 	if (flags & BTREE_WRITE_ALREADY_STARTED)
@@ -2416,6 +2438,7 @@ do_write:
 	wbio->data			= data;
 	wbio->data_bytes		= bytes;
 	wbio->sector_offset		= b->written;
+	wbio->start_time		= start_time;
 	wbio->wbio.c			= c;
 	wbio->wbio.used_mempool		= used_mempool;
 	wbio->wbio.first_btree_write	= !b->written;
@@ -2443,7 +2466,7 @@ err:
 	b->written += sectors_to_write;
 nowrite:
 	btree_bounce_free(c, bytes, used_mempool, data);
-	__btree_node_write_done(c, b);
+	__btree_node_write_done(c, b, 0);
 }
 
 /*

libbcachefs/btree_io.h

@@ -52,6 +52,7 @@ struct btree_write_bio {
 	void			*data;
 	unsigned		data_bytes;
 	unsigned		sector_offset;
+	u64			start_time;
 	struct bch_write_bio	wbio;
 };
 

libbcachefs/btree_iter.c

@@ -562,20 +562,6 @@ static inline struct bkey_s_c btree_path_level_peek_all(struct bch_fs *c,
 			bch2_btree_node_iter_peek_all(&l->iter, l->b));
 }
 
-static inline struct bkey_s_c btree_path_level_peek(struct btree_trans *trans,
-						    struct btree_path *path,
-						    struct btree_path_level *l,
-						    struct bkey *u)
-{
-	struct bkey_s_c k = __btree_iter_unpack(trans->c, l, u,
-			bch2_btree_node_iter_peek(&l->iter, l->b));
-
-	path->pos = k.k ? k.k->p : l->b->key.k.p;
-	trans->paths_sorted = false;
-	bch2_btree_path_verify_level(trans, path, l - path->l);
-	return k;
-}
-
 static inline struct bkey_s_c btree_path_level_prev(struct btree_trans *trans,
 						    struct btree_path *path,
 						    struct btree_path_level *l,
@@ -1501,22 +1487,14 @@ void bch2_trans_updates_to_text(struct printbuf *buf, struct btree_trans *trans)
 
 	for (struct jset_entry *e = trans->journal_entries;
 	     e != btree_trans_journal_entries_top(trans);
-	     e = vstruct_next(e))
+	     e = vstruct_next(e)) {
 		bch2_journal_entry_to_text(buf, trans->c, e);
+		prt_newline(buf);
+	}
 
 	printbuf_indent_sub(buf, 2);
 }
 
-noinline __cold
-void bch2_dump_trans_updates(struct btree_trans *trans)
-{
-	struct printbuf buf = PRINTBUF;
-
-	bch2_trans_updates_to_text(&buf, trans);
-	bch2_print_str(trans->c, buf.buf);
-	printbuf_exit(&buf);
-}
-
 static void bch2_btree_path_to_text_short(struct printbuf *out, struct btree_trans *trans, btree_path_idx_t path_idx)
 {
 	struct btree_path *path = trans->paths + path_idx;
@@ -2357,6 +2335,12 @@ struct bkey_s_c bch2_btree_iter_peek_max(struct btree_iter *iter, struct bpos en
 	bch2_btree_iter_verify_entry_exit(iter);
 	EBUG_ON((iter->flags & BTREE_ITER_filter_snapshots) && bkey_eq(end, POS_MAX));
 
+	ret = trans_maybe_inject_restart(trans, _RET_IP_);
+	if (unlikely(ret)) {
+		k = bkey_s_c_err(ret);
+		goto out_no_locked;
+	}
+
 	if (iter->update_path) {
 		bch2_path_put_nokeep(trans, iter->update_path,
 				     iter->flags & BTREE_ITER_intent);
@@ -2622,6 +2606,12 @@ struct bkey_s_c bch2_btree_iter_peek_prev_min(struct btree_iter *iter, struct bp
 	bch2_btree_iter_verify_entry_exit(iter);
 	EBUG_ON((iter->flags & BTREE_ITER_filter_snapshots) && bpos_eq(end, POS_MIN));
 
+	int ret = trans_maybe_inject_restart(trans, _RET_IP_);
+	if (unlikely(ret)) {
+		k = bkey_s_c_err(ret);
+		goto out_no_locked;
+	}
+
 	while (1) {
 		k = __bch2_btree_iter_peek_prev(iter, search_key);
 		if (unlikely(!k.k))
@@ -2749,6 +2739,12 @@ struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *iter)
 	bch2_btree_iter_verify_entry_exit(iter);
 	EBUG_ON(btree_iter_path(trans, iter)->level && (iter->flags & BTREE_ITER_with_key_cache));
 
+	ret = trans_maybe_inject_restart(trans, _RET_IP_);
+	if (unlikely(ret)) {
+		k = bkey_s_c_err(ret);
+		goto out_no_locked;
+	}
+
 	/* extents can't span inode numbers: */
 	if ((iter->flags & BTREE_ITER_is_extents) &&
 	    unlikely(iter->pos.offset == KEY_OFFSET_MAX)) {
@@ -3106,6 +3102,10 @@ void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
 
 	WARN_ON_ONCE(new_bytes > BTREE_TRANS_MEM_MAX);
 
+	ret = trans_maybe_inject_restart(trans, _RET_IP_);
+	if (ret)
+		return ERR_PTR(ret);
+
 	struct btree_transaction_stats *s = btree_trans_stats(trans);
 	s->max_mem = max(s->max_mem, new_bytes);
 
@@ -3163,7 +3163,8 @@ out_new_mem:
 
 	if (old_bytes) {
 		trace_and_count(c, trans_restart_mem_realloced, trans, _RET_IP_, new_bytes);
-		return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_mem_realloced));
+		return ERR_PTR(btree_trans_restart_ip(trans,
+					BCH_ERR_transaction_restart_mem_realloced, _RET_IP_));
 	}
 out_change_top:
 	p = trans->mem + trans->mem_top;
@@ -3271,6 +3272,14 @@ u32 bch2_trans_begin(struct btree_trans *trans)
 
 	trans->last_begin_ip = _RET_IP_;
 
+#ifdef CONFIG_BCACHEFS_INJECT_TRANSACTION_RESTARTS
+	if (trans->restarted) {
+		trans->restart_count_this_trans++;
+	} else {
+		trans->restart_count_this_trans = 0;
+	}
+#endif
+
 	trans_set_locked(trans, false);
 
 	if (trans->restarted) {

libbcachefs/btree_iter.h

@@ -9,7 +9,6 @@
 void bch2_trans_updates_to_text(struct printbuf *, struct btree_trans *);
 void bch2_btree_path_to_text(struct printbuf *, struct btree_trans *, btree_path_idx_t);
 void bch2_trans_paths_to_text(struct printbuf *, struct btree_trans *);
-void bch2_dump_trans_updates(struct btree_trans *);
 void bch2_dump_trans_paths_updates(struct btree_trans *);
 
 static inline int __bkey_err(const struct bkey *k)
@@ -335,13 +334,20 @@ static inline void bch2_trans_verify_not_unlocked_or_in_restart(struct btree_tra
 }
 
 __always_inline
-static int btree_trans_restart_ip(struct btree_trans *trans, int err, unsigned long ip)
+static int btree_trans_restart_foreign_task(struct btree_trans *trans, int err, unsigned long ip)
 {
 	BUG_ON(err <= 0);
 	BUG_ON(!bch2_err_matches(-err, BCH_ERR_transaction_restart));
 
 	trans->restarted = err;
 	trans->last_restarted_ip = ip;
+	return -err;
+}
+
+__always_inline
+static int btree_trans_restart_ip(struct btree_trans *trans, int err, unsigned long ip)
+{
+	btree_trans_restart_foreign_task(trans, err, ip);
 #ifdef CONFIG_BCACHEFS_DEBUG
 	darray_exit(&trans->last_restarted_trace);
 	bch2_save_backtrace(&trans->last_restarted_trace, current, 0, GFP_NOWAIT);
@@ -355,6 +361,18 @@ static int btree_trans_restart(struct btree_trans *trans, int err)
 	return btree_trans_restart_ip(trans, err, _THIS_IP_);
 }
 
+static inline int trans_maybe_inject_restart(struct btree_trans *trans, unsigned long ip)
+{
+#ifdef CONFIG_BCACHEFS_INJECT_TRANSACTION_RESTARTS
+	if (!(ktime_get_ns() & ~(~0ULL << min(63, (10 + trans->restart_count_this_trans))))) {
+		trace_and_count(trans->c, trans_restart_injected, trans, ip);
+		return btree_trans_restart_ip(trans,
+					BCH_ERR_transaction_restart_fault_inject, ip);
+	}
+#endif
+	return 0;
+}
+
 bool bch2_btree_node_upgrade(struct btree_trans *,
 			     struct btree_path *, unsigned);
 
@@ -739,7 +757,7 @@ transaction_restart:							\
 	if (!_ret2)							\
 		bch2_trans_verify_not_restarted(_trans, _restart_count);\
 									\
-	_ret2 ?: trans_was_restarted(_trans, _restart_count);		\
+	_ret2 ?: trans_was_restarted(_trans, _orig_restart_count);		\
 })
 
 #define for_each_btree_key_max_continue(_trans, _iter,			\

libbcachefs/btree_journal_iter.c

@@ -644,6 +644,8 @@ void bch2_btree_and_journal_iter_init_node_iter(struct btree_trans *trans,
  */
 static int journal_sort_key_cmp(const void *_l, const void *_r)
 {
+	cond_resched();
+
 	const struct journal_key *l = _l;
 	const struct journal_key *r = _r;
 

libbcachefs/btree_key_cache.c

@@ -156,7 +156,7 @@ bkey_cached_alloc(struct btree_trans *trans, struct btree_path *path, unsigned k
 	}
 
 	if (ck) {
-		bch2_btree_lock_init(&ck->c, pcpu_readers ? SIX_LOCK_INIT_PCPU : 0);
+		bch2_btree_lock_init(&ck->c, pcpu_readers ? SIX_LOCK_INIT_PCPU : 0, GFP_KERNEL);
 		ck->c.cached = true;
 		goto lock;
 	}

libbcachefs/btree_locking.c

@@ -7,9 +7,10 @@
 static struct lock_class_key bch2_btree_node_lock_key;
 
 void bch2_btree_lock_init(struct btree_bkey_cached_common *b,
-			  enum six_lock_init_flags flags)
+			  enum six_lock_init_flags flags,
+			  gfp_t gfp)
 {
-	__six_lock_init(&b->lock, "b->c.lock", &bch2_btree_node_lock_key, flags);
+	__six_lock_init(&b->lock, "b->c.lock", &bch2_btree_node_lock_key, flags, gfp);
 	lockdep_set_notrack_class(&b->lock);
 }
 
@@ -90,10 +91,10 @@ static noinline void print_chain(struct printbuf *out, struct lock_graph *g)
 	struct trans_waiting_for_lock *i;
 
 	for (i = g->g; i != g->g + g->nr; i++) {
-		struct task_struct *task = i->trans->locking_wait.task;
+		struct task_struct *task = READ_ONCE(i->trans->locking_wait.task);
 		if (i != g->g)
 			prt_str(out, "<- ");
-		prt_printf(out, "%u ", task ?task->pid : 0);
+		prt_printf(out, "%u ", task ? task->pid : 0);
 	}
 	prt_newline(out);
 }
@@ -171,7 +172,9 @@ static int abort_lock(struct lock_graph *g, struct trans_waiting_for_lock *i)
 {
 	if (i == g->g) {
 		trace_would_deadlock(g, i->trans);
-		return btree_trans_restart(i->trans, BCH_ERR_transaction_restart_would_deadlock);
+		return btree_trans_restart_foreign_task(i->trans,
+					BCH_ERR_transaction_restart_would_deadlock,
+					_THIS_IP_);
 	} else {
 		i->trans->lock_must_abort = true;
 		wake_up_process(i->trans->locking_wait.task);

libbcachefs/btree_locking.h

@@ -13,7 +13,7 @@
 #include "btree_iter.h"
 #include "six.h"
 
-void bch2_btree_lock_init(struct btree_bkey_cached_common *, enum six_lock_init_flags);
+void bch2_btree_lock_init(struct btree_bkey_cached_common *, enum six_lock_init_flags, gfp_t gfp);
 
 void bch2_trans_unlock_noassert(struct btree_trans *);
 void bch2_trans_unlock_write(struct btree_trans *);

libbcachefs/btree_node_scan.c

@@ -13,6 +13,7 @@
 
 #include <linux/kthread.h>
 #include <linux/min_heap.h>
+#include <linux/sched/sysctl.h>
 #include <linux/sort.h>
 
 struct find_btree_nodes_worker {
@@ -166,11 +167,17 @@ static void try_read_btree_node(struct find_btree_nodes *f, struct bch_dev *ca,
 	bio->bi_iter.bi_sector	= offset;
 	bch2_bio_map(bio, bn, PAGE_SIZE);
 
+	u64 submit_time = local_clock();
 	submit_bio_wait(bio);
-	if (bch2_dev_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_read,
-			       "IO error in try_read_btree_node() at %llu: %s",
-			       offset, bch2_blk_status_to_str(bio->bi_status)))
+
+	bch2_account_io_completion(ca, BCH_MEMBER_ERROR_read, submit_time, !bio->bi_status);
+
+	if (bio->bi_status) {
+		bch_err_dev_ratelimited(ca,
+				"IO error in try_read_btree_node() at %llu: %s",
+				offset, bch2_blk_status_to_str(bio->bi_status));
 		return;
+	}
 
 	if (le64_to_cpu(bn->magic) != bset_magic(c))
 		return;
@@ -264,7 +271,7 @@ static int read_btree_nodes_worker(void *p)
 err:
 	bio_put(bio);
 	free_page((unsigned long) buf);
-	percpu_ref_get(&ca->io_ref);
+	percpu_ref_put(&ca->io_ref);
 	closure_put(w->cl);
 	kfree(w);
 	return 0;
@@ -283,32 +290,32 @@ static int read_btree_nodes(struct find_btree_nodes *f)
 			continue;
 
 		struct find_btree_nodes_worker *w = kmalloc(sizeof(*w), GFP_KERNEL);
-		struct task_struct *t;
-
 		if (!w) {
 			percpu_ref_put(&ca->io_ref);
 			ret = -ENOMEM;
 			goto err;
 		}
 
-		percpu_ref_get(&ca->io_ref);
-		closure_get(&cl);
 		w->cl		= &cl;
 		w->f		= f;
 		w->ca		= ca;
 
-		t = kthread_run(read_btree_nodes_worker, w, "read_btree_nodes/%s", ca->name);
+		struct task_struct *t = kthread_create(read_btree_nodes_worker, w, "read_btree_nodes/%s", ca->name);
 		ret = PTR_ERR_OR_ZERO(t);
 		if (ret) {
 			percpu_ref_put(&ca->io_ref);
-			closure_put(&cl);
-			f->ret = ret;
-			bch_err(c, "error starting kthread: %i", ret);
+			kfree(w);
+			bch_err_msg(c, ret, "starting kthread");
 			break;
 		}
+
+		closure_get(&cl);
+		percpu_ref_get(&ca->io_ref);
+		wake_up_process(t);
 	}
 err:
-	closure_sync(&cl);
+	while (closure_sync_timeout(&cl, sysctl_hung_task_timeout_secs * HZ / 2))
+		;
 	return f->ret ?: ret;
 }
 
@@ -572,10 +579,12 @@ int bch2_get_scanned_nodes(struct bch_fs *c, enum btree_id btree,
 
 		found_btree_node_to_key(&tmp.k, &n);
 
-		struct printbuf buf = PRINTBUF;
-		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&tmp.k));
-		bch_verbose(c, "%s(): recovering %s", __func__, buf.buf);
-		printbuf_exit(&buf);
+		if (c->opts.verbose) {
+			struct printbuf buf = PRINTBUF;
+			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&tmp.k));
+			bch_verbose(c, "%s(): recovering %s", __func__, buf.buf);
+			printbuf_exit(&buf);
+		}
 
 		BUG_ON(bch2_bkey_validate(c, bkey_i_to_s_c(&tmp.k),
 					  (struct bkey_validate_context) {

libbcachefs/btree_trans_commit.c

@@ -164,6 +164,7 @@ bool bch2_btree_bset_insert_key(struct btree_trans *trans,
 	EBUG_ON(bpos_gt(insert->k.p, b->data->max_key));
 	EBUG_ON(insert->k.u64s > bch2_btree_keys_u64s_remaining(b));
 	EBUG_ON(!b->c.level && !bpos_eq(insert->k.p, path->pos));
+	kmsan_check_memory(insert, bkey_bytes(&insert->k));
 
 	k = bch2_btree_node_iter_peek_all(node_iter, b);
 	if (k && bkey_cmp_left_packed(b, k, &insert->k.p))
@@ -336,6 +337,7 @@ static inline void btree_insert_entry_checks(struct btree_trans *trans,
 	BUG_ON(i->cached	!= path->cached);
 	BUG_ON(i->level		!= path->level);
 	BUG_ON(i->btree_id	!= path->btree_id);
+	BUG_ON(i->bkey_type	!= __btree_node_type(path->level, path->btree_id));
 	EBUG_ON(!i->level &&
 		btree_type_has_snapshots(i->btree_id) &&
 		!(i->flags & BTREE_UPDATE_internal_snapshot_node) &&
@@ -517,69 +519,45 @@ static int run_one_trans_trigger(struct btree_trans *trans, struct btree_insert_
 	}
 }
 
-static int run_btree_triggers(struct btree_trans *trans, enum btree_id btree_id,
-			      unsigned *btree_id_updates_start)
-{
-	bool trans_trigger_run;
-
-	/*
-	 * Running triggers will append more updates to the list of updates as
-	 * we're walking it:
-	 */
-	do {
-		trans_trigger_run = false;
-
-		for (unsigned i = *btree_id_updates_start;
-		     i < trans->nr_updates && trans->updates[i].btree_id <= btree_id;
-		     i++) {
-			if (trans->updates[i].btree_id < btree_id) {
-				*btree_id_updates_start = i;
-				continue;
-			}
-
-			int ret = run_one_trans_trigger(trans, trans->updates + i);
-			if (ret < 0)
-				return ret;
-			if (ret)
-				trans_trigger_run = true;
-		}
-	} while (trans_trigger_run);
-
-	trans_for_each_update(trans, i)
-		BUG_ON(!(i->flags & BTREE_TRIGGER_norun) &&
-		       i->btree_id == btree_id &&
-		       btree_node_type_has_trans_triggers(i->bkey_type) &&
-		       (!i->insert_trigger_run || !i->overwrite_trigger_run));
-
-	return 0;
-}
-
 static int bch2_trans_commit_run_triggers(struct btree_trans *trans)
 {
-	unsigned btree_id = 0, btree_id_updates_start = 0;
-	int ret = 0;
+	unsigned sort_id_start = 0;
 
-	/*
-	 *
-	 * For a given btree, this algorithm runs insert triggers before
-	 * overwrite triggers: this is so that when extents are being moved
-	 * (e.g. by FALLOCATE_FL_INSERT_RANGE), we don't drop references before
-	 * they are re-added.
-	 */
-	for (btree_id = 0; btree_id < BTREE_ID_NR; btree_id++) {
-		if (btree_id == BTREE_ID_alloc)
-			continue;
+	while (sort_id_start < trans->nr_updates) {
+		unsigned i, sort_id = trans->updates[sort_id_start].sort_order;
+		bool trans_trigger_run;
 
-		ret = run_btree_triggers(trans, btree_id, &btree_id_updates_start);
-		if (ret)
-			return ret;
+		/*
+		 * For a given btree, this algorithm runs insert triggers before
+		 * overwrite triggers: this is so that when extents are being
+		 * moved (e.g. by FALLOCATE_FL_INSERT_RANGE), we don't drop
+		 * references before they are re-added.
+		 *
+		 * Running triggers will append more updates to the list of
+		 * updates as we're walking it:
+		 */
+		do {
+			trans_trigger_run = false;
+
+			for (i = sort_id_start;
+			     i < trans->nr_updates && trans->updates[i].sort_order <= sort_id;
+			     i++) {
+				if (trans->updates[i].sort_order < sort_id) {
+					sort_id_start = i;
+					continue;
+				}
+
+				int ret = run_one_trans_trigger(trans, trans->updates + i);
+				if (ret < 0)
+					return ret;
+				if (ret)
+					trans_trigger_run = true;
+			}
+		} while (trans_trigger_run);
+
+		sort_id_start = i;
 	}
 
-	btree_id_updates_start = 0;
-	ret = run_btree_triggers(trans, BTREE_ID_alloc, &btree_id_updates_start);
-	if (ret)
-		return ret;
-
 #ifdef CONFIG_BCACHEFS_DEBUG
 	trans_for_each_update(trans, i)
 		BUG_ON(!(i->flags & BTREE_TRIGGER_norun) &&
@@ -903,6 +881,24 @@ int bch2_trans_commit_error(struct btree_trans *trans, unsigned flags,
 	struct bch_fs *c = trans->c;
 	enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
 
+	if (bch2_err_matches(ret, BCH_ERR_journal_res_blocked)) {
+		/*
+		 * XXX: this should probably be a separate BTREE_INSERT_NONBLOCK
+		 * flag
+		 */
+		if ((flags & BCH_TRANS_COMMIT_journal_reclaim) &&
+		    watermark < BCH_WATERMARK_reclaim) {
+			ret = -BCH_ERR_journal_reclaim_would_deadlock;
+			goto out;
+		}
+
+		ret = drop_locks_do(trans,
+			bch2_trans_journal_res_get(trans,
+					(flags & BCH_WATERMARK_MASK)|
+					JOURNAL_RES_GET_CHECK));
+		goto out;
+	}
+
 	switch (ret) {
 	case -BCH_ERR_btree_insert_btree_node_full:
 		ret = bch2_btree_split_leaf(trans, i->path, flags);
@@ -914,22 +910,6 @@ int bch2_trans_commit_error(struct btree_trans *trans, unsigned flags,
 		ret = drop_locks_do(trans,
 			bch2_accounting_update_sb(trans));
 		break;
-	case -BCH_ERR_journal_res_get_blocked:
-		/*
-		 * XXX: this should probably be a separate BTREE_INSERT_NONBLOCK
-		 * flag
-		 */
-		if ((flags & BCH_TRANS_COMMIT_journal_reclaim) &&
-		    watermark < BCH_WATERMARK_reclaim) {
-			ret = -BCH_ERR_journal_reclaim_would_deadlock;
-			break;
-		}
-
-		ret = drop_locks_do(trans,
-			bch2_trans_journal_res_get(trans,
-					(flags & BCH_WATERMARK_MASK)|
-					JOURNAL_RES_GET_CHECK));
-		break;
 	case -BCH_ERR_btree_insert_need_journal_reclaim:
 		bch2_trans_unlock(trans);
 
@@ -950,7 +930,7 @@ int bch2_trans_commit_error(struct btree_trans *trans, unsigned flags,
 		BUG_ON(ret >= 0);
 		break;
 	}
-
+out:
 	BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart) != !!trans->restarted);
 
 	bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOSPC) &&
@@ -999,6 +979,10 @@ int __bch2_trans_commit(struct btree_trans *trans, unsigned flags)
 
 	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 
+	ret = trans_maybe_inject_restart(trans, _RET_IP_);
+	if (unlikely(ret))
+		goto out_reset;
+
 	if (!trans->nr_updates &&
 	    !trans->journal_entries_u64s)
 		goto out_reset;

libbcachefs/btree_types.h

@@ -423,6 +423,7 @@ static inline struct bpos btree_node_pos(struct btree_bkey_cached_common *b)
 
 struct btree_insert_entry {
 	unsigned		flags;
+	u8			sort_order;
 	u8			bkey_type;
 	enum btree_id		btree_id:8;
 	u8			level:4;
@@ -509,6 +510,9 @@ struct btree_trans {
 	bool			notrace_relock_fail:1;
 	enum bch_errcode	restarted:16;
 	u32			restart_count;
+#ifdef CONFIG_BCACHEFS_INJECT_TRANSACTION_RESTARTS
+	u32			restart_count_this_trans;
+#endif
 
 	u64			last_begin_time;
 	unsigned long		last_begin_ip;
@@ -850,6 +854,18 @@ static inline bool btree_type_uses_write_buffer(enum btree_id btree)
 	return BIT_ULL(btree) & mask;
 }
 
+static inline u8 btree_trigger_order(enum btree_id btree)
+{
+	switch (btree) {
+	case BTREE_ID_alloc:
+		return U8_MAX;
+	case BTREE_ID_stripes:
+		return U8_MAX - 1;
+	default:
+		return btree;
+	}
+}
+
 struct btree_root {
 	struct btree		*b;
 

libbcachefs/btree_update.c

@@ -17,7 +17,7 @@
 static inline int btree_insert_entry_cmp(const struct btree_insert_entry *l,
 					 const struct btree_insert_entry *r)
 {
-	return   cmp_int(l->btree_id,	r->btree_id) ?:
+	return   cmp_int(l->sort_order,	r->sort_order) ?:
 		 cmp_int(l->cached,	r->cached) ?:
 		 -cmp_int(l->level,	r->level) ?:
 		 bpos_cmp(l->k->k.p,	r->k->k.p);
@@ -397,6 +397,7 @@ bch2_trans_update_by_path(struct btree_trans *trans, btree_path_idx_t path_idx,
 
 	n = (struct btree_insert_entry) {
 		.flags		= flags,
+		.sort_order	= btree_trigger_order(path->btree_id),
 		.bkey_type	= __btree_node_type(path->level, path->btree_id),
 		.btree_id	= path->btree_id,
 		.level		= path->level,
@@ -511,6 +512,8 @@ static noinline int bch2_trans_update_get_key_cache(struct btree_trans *trans,
 int __must_check bch2_trans_update(struct btree_trans *trans, struct btree_iter *iter,
 				   struct bkey_i *k, enum btree_iter_update_trigger_flags flags)
 {
+	kmsan_check_memory(k, bkey_bytes(&k->k));
+
 	btree_path_idx_t path_idx = iter->update_path ?: iter->path;
 	int ret;
 
@@ -843,6 +846,19 @@ int bch2_trans_log_msg(struct btree_trans *trans, struct printbuf *buf)
 	return 0;
 }
 
+int bch2_trans_log_bkey(struct btree_trans *trans, enum btree_id btree,
+			unsigned level, struct bkey_i *k)
+{
+	struct jset_entry *e = bch2_trans_jset_entry_alloc(trans, jset_u64s(k->k.u64s));
+	int ret = PTR_ERR_OR_ZERO(e);
+	if (ret)
+		return ret;
+
+	journal_entry_init(e, BCH_JSET_ENTRY_log_bkey, btree, level, k->k.u64s);
+	bkey_copy(e->start, k);
+	return 0;
+}
+
 __printf(3, 0)
 static int
 __bch2_fs_log_msg(struct bch_fs *c, unsigned commit_flags, const char *fmt,

libbcachefs/btree_update.h

@@ -126,10 +126,20 @@ bch2_trans_jset_entry_alloc(struct btree_trans *trans, unsigned u64s)
 
 int bch2_btree_insert_clone_trans(struct btree_trans *, enum btree_id, struct bkey_i *);
 
+int bch2_btree_write_buffer_insert_err(struct btree_trans *,
+				       enum btree_id, struct bkey_i *);
+
 static inline int __must_check bch2_trans_update_buffered(struct btree_trans *trans,
 					    enum btree_id btree,
 					    struct bkey_i *k)
 {
+	kmsan_check_memory(k, bkey_bytes(&k->k));
+
+	if (unlikely(!btree_type_uses_write_buffer(btree))) {
+		int ret = bch2_btree_write_buffer_insert_err(trans, btree, k);
+		dump_stack();
+		return ret;
+	}
 	/*
 	 * Most updates skip the btree write buffer until journal replay is
 	 * finished because synchronization with journal replay relies on having
@@ -160,6 +170,8 @@ void bch2_trans_commit_hook(struct btree_trans *,
 int __bch2_trans_commit(struct btree_trans *, unsigned);
 
 int bch2_trans_log_msg(struct btree_trans *, struct printbuf *);
+int bch2_trans_log_bkey(struct btree_trans *, enum btree_id, unsigned, struct bkey_i *);
+
 __printf(2, 3) int bch2_fs_log_msg(struct bch_fs *, const char *, ...);
 __printf(2, 3) int bch2_journal_log_msg(struct bch_fs *, const char *, ...);
 

libbcachefs/btree_update_interior.c

@@ -35,6 +35,8 @@ static const char * const bch2_btree_update_modes[] = {
 	NULL
 };
 
+static void bch2_btree_update_to_text(struct printbuf *, struct btree_update *);
+
 static int bch2_btree_insert_node(struct btree_update *, struct btree_trans *,
 				  btree_path_idx_t, struct btree *, struct keylist *);
 static void bch2_btree_update_add_new_node(struct btree_update *, struct btree *);
@@ -54,6 +56,8 @@ int bch2_btree_node_check_topology(struct btree_trans *trans, struct btree *b)
 	struct bkey_buf prev;
 	int ret = 0;
 
+	printbuf_indent_add_nextline(&buf, 2);
+
 	BUG_ON(b->key.k.type == KEY_TYPE_btree_ptr_v2 &&
 	       !bpos_eq(bkey_i_to_btree_ptr_v2(&b->key)->v.min_key,
 			b->data->min_key));
@@ -64,19 +68,20 @@ int bch2_btree_node_check_topology(struct btree_trans *trans, struct btree *b)
 
 	if (b == btree_node_root(c, b)) {
 		if (!bpos_eq(b->data->min_key, POS_MIN)) {
-			printbuf_reset(&buf);
+			ret = __bch2_topology_error(c, &buf);
+
 			bch2_bpos_to_text(&buf, b->data->min_key);
 			log_fsck_err(trans, btree_root_bad_min_key,
 				      "btree root with incorrect min_key: %s", buf.buf);
-			goto topology_repair;
+			goto out;
 		}
 
 		if (!bpos_eq(b->data->max_key, SPOS_MAX)) {
-			printbuf_reset(&buf);
+			ret = __bch2_topology_error(c, &buf);
 			bch2_bpos_to_text(&buf, b->data->max_key);
 			log_fsck_err(trans, btree_root_bad_max_key,
 				      "btree root with incorrect max_key: %s", buf.buf);
-			goto topology_repair;
+			goto out;
 		}
 	}
 
@@ -94,20 +99,19 @@ int bch2_btree_node_check_topology(struct btree_trans *trans, struct btree *b)
 			: bpos_successor(prev.k->k.p);
 
 		if (!bpos_eq(expected_min, bp.v->min_key)) {
-			bch2_topology_error(c);
+			ret = __bch2_topology_error(c, &buf);
 
-			printbuf_reset(&buf);
-			prt_str(&buf, "end of prev node doesn't match start of next node\n  in ");
+			prt_str(&buf, "end of prev node doesn't match start of next node\nin ");
 			bch2_btree_id_level_to_text(&buf, b->c.btree_id, b->c.level);
 			prt_str(&buf, " node ");
 			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
-			prt_str(&buf, "\n  prev ");
+			prt_str(&buf, "\nprev ");
 			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(prev.k));
-			prt_str(&buf, "\n  next ");
+			prt_str(&buf, "\nnext ");
 			bch2_bkey_val_to_text(&buf, c, k);
 
 			log_fsck_err(trans, btree_node_topology_bad_min_key, "%s", buf.buf);
-			goto topology_repair;
+			goto out;
 		}
 
 		bch2_bkey_buf_reassemble(&prev, c, k);
@@ -115,29 +119,25 @@ int bch2_btree_node_check_topology(struct btree_trans *trans, struct btree *b)
 	}
 
 	if (bkey_deleted(&prev.k->k)) {
-		bch2_topology_error(c);
+		ret = __bch2_topology_error(c, &buf);
 
-		printbuf_reset(&buf);
-		prt_str(&buf, "empty interior node\n  in ");
+		prt_str(&buf, "empty interior node\nin ");
 		bch2_btree_id_level_to_text(&buf, b->c.btree_id, b->c.level);
 		prt_str(&buf, " node ");
 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
 
 		log_fsck_err(trans, btree_node_topology_empty_interior_node, "%s", buf.buf);
-		goto topology_repair;
 	} else if (!bpos_eq(prev.k->k.p, b->key.k.p)) {
-		bch2_topology_error(c);
+		ret = __bch2_topology_error(c, &buf);
 
-		printbuf_reset(&buf);
-		prt_str(&buf, "last child node doesn't end at end of parent node\n  in ");
+		prt_str(&buf, "last child node doesn't end at end of parent node\nin ");
 		bch2_btree_id_level_to_text(&buf, b->c.btree_id, b->c.level);
 		prt_str(&buf, " node ");
 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
-		prt_str(&buf, "\n  last key ");
+		prt_str(&buf, "\nlast key ");
 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(prev.k));
 
 		log_fsck_err(trans, btree_node_topology_bad_max_key, "%s", buf.buf);
-		goto topology_repair;
 	}
 out:
 fsck_err:
@@ -145,9 +145,6 @@ fsck_err:
 	bch2_bkey_buf_exit(&prev, c);
 	printbuf_exit(&buf);
 	return ret;
-topology_repair:
-	ret = bch2_topology_error(c);
-	goto out;
 }
 
 /* Calculate ideal packed bkey format for new btree nodes: */
@@ -649,6 +646,14 @@ static int btree_update_nodes_written_trans(struct btree_trans *trans,
 	return 0;
 }
 
+/* If the node has been reused, we might be reading uninitialized memory - that's fine: */
+static noinline __no_kmsan_checks bool btree_node_seq_matches(struct btree *b, __le64 seq)
+{
+	struct btree_node *b_data = READ_ONCE(b->data);
+
+	return (b_data ? b_data->keys.seq : 0) == seq;
+}
+
 static void btree_update_nodes_written(struct btree_update *as)
 {
 	struct bch_fs *c = as->c;
@@ -677,15 +682,9 @@ static void btree_update_nodes_written(struct btree_update *as)
 	 * on disk:
 	 */
 	for (i = 0; i < as->nr_old_nodes; i++) {
-		__le64 seq;
-
 		b = as->old_nodes[i];
 
-		btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
-		seq = b->data ? b->data->keys.seq : 0;
-		six_unlock_read(&b->c.lock);
-
-		if (seq == as->old_nodes_seq[i])
+		if (btree_node_seq_matches(b, as->old_nodes_seq[i]))
 			wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight_inner,
 				       TASK_UNINTERRUPTIBLE);
 	}
@@ -1269,7 +1268,8 @@ err:
 	bch2_btree_update_free(as, trans);
 	if (!bch2_err_matches(ret, ENOSPC) &&
 	    !bch2_err_matches(ret, EROFS) &&
-	    ret != -BCH_ERR_journal_reclaim_would_deadlock)
+	    ret != -BCH_ERR_journal_reclaim_would_deadlock &&
+	    ret != -BCH_ERR_journal_shutdown)
 		bch_err_fn_ratelimited(c, ret);
 	return ERR_PTR(ret);
 }
@@ -1780,11 +1780,24 @@ static int bch2_btree_insert_node(struct btree_update *as, struct btree_trans *t
 	int ret;
 
 	lockdep_assert_held(&c->gc_lock);
-	BUG_ON(!btree_node_intent_locked(path, b->c.level));
 	BUG_ON(!b->c.level);
 	BUG_ON(!as || as->b);
 	bch2_verify_keylist_sorted(keys);
 
+	if (!btree_node_intent_locked(path, b->c.level)) {
+		struct printbuf buf = PRINTBUF;
+		bch2_log_msg_start(c, &buf);
+		prt_printf(&buf, "%s(): node not locked at level %u\n",
+			   __func__, b->c.level);
+		bch2_btree_update_to_text(&buf, as);
+		bch2_btree_path_to_text(&buf, trans, path_idx);
+
+		bch2_print_string_as_lines(KERN_ERR, buf.buf);
+		printbuf_exit(&buf);
+		bch2_fs_emergency_read_only(c);
+		return -EIO;
+	}
+
 	ret = bch2_btree_node_lock_write(trans, path, &b->c);
 	if (ret)
 		return ret;
@@ -2005,18 +2018,22 @@ int __bch2_foreground_maybe_merge(struct btree_trans *trans,
 	}
 
 	if (!bpos_eq(bpos_successor(prev->data->max_key), next->data->min_key)) {
-		struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
+		struct printbuf buf = PRINTBUF;
 
-		bch2_bpos_to_text(&buf1, prev->data->max_key);
-		bch2_bpos_to_text(&buf2, next->data->min_key);
-		bch_err(c,
-			"%s(): btree topology error:\n"
-			"  prev ends at   %s\n"
-			"  next starts at %s",
-			__func__, buf1.buf, buf2.buf);
-		printbuf_exit(&buf1);
-		printbuf_exit(&buf2);
-		ret = bch2_topology_error(c);
+		printbuf_indent_add_nextline(&buf, 2);
+		prt_printf(&buf, "%s(): ", __func__);
+		ret = __bch2_topology_error(c, &buf);
+		prt_newline(&buf);
+
+		prt_printf(&buf, "prev ends at   ");
+		bch2_bpos_to_text(&buf, prev->data->max_key);
+		prt_newline(&buf);
+
+		prt_printf(&buf, "next starts at ");
+		bch2_bpos_to_text(&buf, next->data->min_key);
+
+		bch_err(c, "%s", buf.buf);
+		printbuf_exit(&buf);
 		goto err;
 	}
 
@@ -2124,6 +2141,31 @@ err_free_update:
 	goto out;
 }
 
+static int get_iter_to_node(struct btree_trans *trans, struct btree_iter *iter,
+			    struct btree *b)
+{
+	bch2_trans_node_iter_init(trans, iter, b->c.btree_id, b->key.k.p,
+				  BTREE_MAX_DEPTH, b->c.level,
+				  BTREE_ITER_intent);
+	int ret = bch2_btree_iter_traverse(iter);
+	if (ret)
+		goto err;
+
+	/* has node been freed? */
+	if (btree_iter_path(trans, iter)->l[b->c.level].b != b) {
+		/* node has been freed: */
+		BUG_ON(!btree_node_dying(b));
+		ret = -BCH_ERR_btree_node_dying;
+		goto err;
+	}
+
+	BUG_ON(!btree_node_hashed(b));
+	return 0;
+err:
+	bch2_trans_iter_exit(trans, iter);
+	return ret;
+}
+
 int bch2_btree_node_rewrite(struct btree_trans *trans,
 			    struct btree_iter *iter,
 			    struct btree *b,
@@ -2189,7 +2231,29 @@ err:
 	goto out;
 }
 
-int bch2_btree_node_rewrite_key(struct btree_trans *trans,
+static int bch2_btree_node_rewrite_key(struct btree_trans *trans,
+				       enum btree_id btree, unsigned level,
+				       struct bkey_i *k, unsigned flags)
+{
+	struct btree_iter iter;
+	bch2_trans_node_iter_init(trans, &iter,
+				  btree, k->k.p,
+				  BTREE_MAX_DEPTH, level, 0);
+	struct btree *b = bch2_btree_iter_peek_node(&iter);
+	int ret = PTR_ERR_OR_ZERO(b);
+	if (ret)
+		goto out;
+
+	bool found = b && btree_ptr_hash_val(&b->key) == btree_ptr_hash_val(k);
+	ret = found
+		? bch2_btree_node_rewrite(trans, &iter, b, flags)
+		: -ENOENT;
+out:
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
+}
+
+int bch2_btree_node_rewrite_pos(struct btree_trans *trans,
 				enum btree_id btree, unsigned level,
 				struct bpos pos, unsigned flags)
 {
@@ -2209,6 +2273,19 @@ err:
 	return ret;
 }
 
+int bch2_btree_node_rewrite_key_get_iter(struct btree_trans *trans,
+					 struct btree *b, unsigned flags)
+{
+	struct btree_iter iter;
+	int ret = get_iter_to_node(trans, &iter, b);
+	if (ret)
+		return ret == -BCH_ERR_btree_node_dying ? 0 : ret;
+
+	ret = bch2_btree_node_rewrite(trans, &iter, b, flags);
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
+}
+
 struct async_btree_rewrite {
 	struct bch_fs		*c;
 	struct work_struct	work;
@@ -2218,58 +2295,17 @@ struct async_btree_rewrite {
 	struct bkey_buf		key;
 };
 
-static int async_btree_node_rewrite_trans(struct btree_trans *trans,
-					  struct async_btree_rewrite *a)
-{
-	struct btree_iter iter;
-	bch2_trans_node_iter_init(trans, &iter,
-				  a->btree_id, a->key.k->k.p,
-				  BTREE_MAX_DEPTH, a->level, 0);
-	struct btree *b = bch2_btree_iter_peek_node(&iter);
-	int ret = PTR_ERR_OR_ZERO(b);
-	if (ret)
-		goto out;
-
-	bool found = b && btree_ptr_hash_val(&b->key) == btree_ptr_hash_val(a->key.k);
-	ret = found
-		? bch2_btree_node_rewrite(trans, &iter, b, 0)
-		: -ENOENT;
-
-#if 0
-	/* Tracepoint... */
-	if (!ret || ret == -ENOENT) {
-		struct bch_fs *c = trans->c;
-		struct printbuf buf = PRINTBUF;
-
-		if (!ret) {
-			prt_printf(&buf, "rewrite node:\n  ");
-			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(a->key.k));
-		} else {
-			prt_printf(&buf, "node to rewrite not found:\n  want: ");
-			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(a->key.k));
-			prt_printf(&buf, "\n  got:  ");
-			if (b)
-				bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
-			else
-				prt_str(&buf, "(null)");
-		}
-		bch_info(c, "%s", buf.buf);
-		printbuf_exit(&buf);
-	}
-#endif
-out:
-	bch2_trans_iter_exit(trans, &iter);
-	return ret;
-}
-
 static void async_btree_node_rewrite_work(struct work_struct *work)
 {
 	struct async_btree_rewrite *a =
 		container_of(work, struct async_btree_rewrite, work);
 	struct bch_fs *c = a->c;
 
-	int ret = bch2_trans_do(c, async_btree_node_rewrite_trans(trans, a));
-	if (ret != -ENOENT)
+	int ret = bch2_trans_do(c, bch2_btree_node_rewrite_key(trans,
+						a->btree_id, a->level, a->key.k, 0));
+	if (ret != -ENOENT &&
+	    !bch2_err_matches(ret, EROFS) &&
+	    ret != -BCH_ERR_journal_shutdown)
 		bch_err_fn_ratelimited(c, ret);
 
 	spin_lock(&c->btree_node_rewrites_lock);
@@ -2512,30 +2548,15 @@ int bch2_btree_node_update_key_get_iter(struct btree_trans *trans,
 					unsigned commit_flags, bool skip_triggers)
 {
 	struct btree_iter iter;
-	int ret;
-
-	bch2_trans_node_iter_init(trans, &iter, b->c.btree_id, b->key.k.p,
-				  BTREE_MAX_DEPTH, b->c.level,
-				  BTREE_ITER_intent);
-	ret = bch2_btree_iter_traverse(&iter);
+	int ret = get_iter_to_node(trans, &iter, b);
 	if (ret)
-		goto out;
-
-	/* has node been freed? */
-	if (btree_iter_path(trans, &iter)->l[b->c.level].b != b) {
-		/* node has been freed: */
-		BUG_ON(!btree_node_dying(b));
-		goto out;
-	}
-
-	BUG_ON(!btree_node_hashed(b));
+		return ret == -BCH_ERR_btree_node_dying ? 0 : ret;
 
 	bch2_bkey_drop_ptrs(bkey_i_to_s(new_key), ptr,
 			    !bch2_bkey_has_device(bkey_i_to_s(&b->key), ptr->dev));
 
 	ret = bch2_btree_node_update_key(trans, &iter, b, new_key,
 					 commit_flags, skip_triggers);
-out:
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
 }

libbcachefs/btree_update_interior.h

@@ -169,9 +169,12 @@ static inline int bch2_foreground_maybe_merge(struct btree_trans *trans,
 
 int bch2_btree_node_rewrite(struct btree_trans *, struct btree_iter *,
 			    struct btree *, unsigned);
-int bch2_btree_node_rewrite_key(struct btree_trans *,
+int bch2_btree_node_rewrite_pos(struct btree_trans *,
 				enum btree_id, unsigned,
 				struct bpos, unsigned);
+int bch2_btree_node_rewrite_key_get_iter(struct btree_trans *,
+					 struct btree *, unsigned);
+
 void bch2_btree_node_rewrite_async(struct bch_fs *, struct btree *);
 
 int bch2_btree_node_update_key(struct btree_trans *, struct btree_iter *,
@@ -282,12 +285,12 @@ static inline struct btree_node_entry *want_new_bset(struct bch_fs *c, struct bt
 {
 	struct bset_tree *t = bset_tree_last(b);
 	struct btree_node_entry *bne = max(write_block(b),
-			(void *) btree_bkey_last(b, bset_tree_last(b)));
+			(void *) btree_bkey_last(b, t));
 	ssize_t remaining_space =
 		__bch2_btree_u64s_remaining(b, bne->keys.start);
 
 	if (unlikely(bset_written(b, bset(b, t)))) {
-		if (remaining_space > (ssize_t) (block_bytes(c) >> 3))
+		if (b->written + block_sectors(c) <= btree_sectors(c))
 			return bne;
 	} else {
 		if (unlikely(bset_u64s(t) * sizeof(u64) > btree_write_set_buffer(b)) &&

libbcachefs/btree_write_buffer.c

@@ -264,6 +264,22 @@ out:
 	BUG_ON(wb->sorted.size < wb->flushing.keys.nr);
 }
 
+int bch2_btree_write_buffer_insert_err(struct btree_trans *trans,
+				       enum btree_id btree, struct bkey_i *k)
+{
+	struct bch_fs *c = trans->c;
+	struct printbuf buf = PRINTBUF;
+
+	prt_printf(&buf, "attempting to do write buffer update on non wb btree=");
+	bch2_btree_id_to_text(&buf, btree);
+	prt_str(&buf, "\n");
+	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k));
+
+	bch2_fs_inconsistent(c, "%s", buf.buf);
+	printbuf_exit(&buf);
+	return -EROFS;
+}
+
 static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 {
 	struct bch_fs *c = trans->c;
@@ -312,7 +328,10 @@ static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 	darray_for_each(wb->sorted, i) {
 		struct btree_write_buffered_key *k = &wb->flushing.keys.data[i->idx];
 
-		BUG_ON(!btree_type_uses_write_buffer(k->btree));
+		if (unlikely(!btree_type_uses_write_buffer(k->btree))) {
+			ret = bch2_btree_write_buffer_insert_err(trans, k->btree, &k->k);
+			goto err;
+		}
 
 		for (struct wb_key_ref *n = i + 1; n < min(i + 4, &darray_top(wb->sorted)); n++)
 			prefetch(&wb->flushing.keys.data[n->idx]);

libbcachefs/buckets.c

@@ -381,6 +381,36 @@ err:
 	return ret;
 }
 
+static int bucket_ref_update_err(struct btree_trans *trans, struct printbuf *buf,
+				 struct bkey_s_c k, bool insert, enum bch_sb_error_id id)
+{
+	struct bch_fs *c = trans->c;
+	bool repeat = false, print = true, suppress = false;
+
+	prt_printf(buf, "\nwhile marking ");
+	bch2_bkey_val_to_text(buf, c, k);
+	prt_newline(buf);
+
+	__bch2_count_fsck_err(c, id, buf->buf, &repeat, &print, &suppress);
+
+	int ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_allocations);
+
+	if (insert) {
+		print = true;
+		suppress = false;
+
+		bch2_trans_updates_to_text(buf, trans);
+		__bch2_inconsistent_error(c, buf);
+		ret = -BCH_ERR_bucket_ref_update;
+	}
+
+	if (suppress)
+		prt_printf(buf, "Ratelimiting new instances of previous error\n");
+	if (print)
+		bch2_print_string_as_lines(KERN_ERR, buf->buf);
+	return ret;
+}
+
 int bch2_bucket_ref_update(struct btree_trans *trans, struct bch_dev *ca,
 			   struct bkey_s_c k,
 			   const struct bch_extent_ptr *ptr,
@@ -396,32 +426,29 @@ int bch2_bucket_ref_update(struct btree_trans *trans, struct bch_dev *ca,
 
 	BUG_ON(!sectors);
 
-	if (gen_after(ptr->gen, b_gen)) {
-		bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_allocations);
-		log_fsck_err(trans, ptr_gen_newer_than_bucket_gen,
-			"bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen\n"
-			"while marking %s",
+	if (unlikely(gen_after(ptr->gen, b_gen))) {
+		bch2_log_msg_start(c, &buf);
+		prt_printf(&buf,
+			"bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen",
 			ptr->dev, bucket_nr, b_gen,
 			bch2_data_type_str(bucket_data_type ?: ptr_data_type),
-			ptr->gen,
-			(bch2_bkey_val_to_text(&buf, c, k), buf.buf));
-		if (inserting)
-			goto err;
+			ptr->gen);
+
+		ret = bucket_ref_update_err(trans, &buf, k, inserting,
+					    BCH_FSCK_ERR_ptr_gen_newer_than_bucket_gen);
 		goto out;
 	}
 
-	if (gen_cmp(b_gen, ptr->gen) > BUCKET_GC_GEN_MAX) {
-		bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_allocations);
-		log_fsck_err(trans, ptr_too_stale,
-			"bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n"
-			"while marking %s",
+	if (unlikely(gen_cmp(b_gen, ptr->gen) > BUCKET_GC_GEN_MAX)) {
+		bch2_log_msg_start(c, &buf);
+		prt_printf(&buf,
+			"bucket %u:%zu gen %u data type %s: ptr gen %u too stale",
 			ptr->dev, bucket_nr, b_gen,
 			bch2_data_type_str(bucket_data_type ?: ptr_data_type),
-			ptr->gen,
-			(printbuf_reset(&buf),
-			 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
-		if (inserting)
-			goto err;
+			ptr->gen);
+
+		ret = bucket_ref_update_err(trans, &buf, k, inserting,
+					    BCH_FSCK_ERR_ptr_too_stale);
 		goto out;
 	}
 
@@ -430,62 +457,50 @@ int bch2_bucket_ref_update(struct btree_trans *trans, struct bch_dev *ca,
 		goto out;
 	}
 
-	if (b_gen != ptr->gen) {
-		bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_allocations);
-		log_fsck_err(trans, stale_dirty_ptr,
-			"bucket %u:%zu gen %u (mem gen %u) data type %s: stale dirty ptr (gen %u)\n"
-			"while marking %s",
+	if (unlikely(b_gen != ptr->gen)) {
+		bch2_log_msg_start(c, &buf);
+		prt_printf(&buf,
+			"bucket %u:%zu gen %u (mem gen %u) data type %s: stale dirty ptr (gen %u)",
 			ptr->dev, bucket_nr, b_gen,
 			bucket_gen_get(ca, bucket_nr),
 			bch2_data_type_str(bucket_data_type ?: ptr_data_type),
-			ptr->gen,
-			(printbuf_reset(&buf),
-			 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
-		if (inserting)
-			goto err;
+			ptr->gen);
+
+		ret = bucket_ref_update_err(trans, &buf, k, inserting,
+					    BCH_FSCK_ERR_stale_dirty_ptr);
 		goto out;
 	}
 
-	if (bucket_data_type_mismatch(bucket_data_type, ptr_data_type)) {
-		bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_allocations);
-		log_fsck_err(trans, ptr_bucket_data_type_mismatch,
-			"bucket %u:%zu gen %u different types of data in same bucket: %s, %s\n"
-			"while marking %s",
-			ptr->dev, bucket_nr, b_gen,
-			bch2_data_type_str(bucket_data_type),
-			bch2_data_type_str(ptr_data_type),
-			(printbuf_reset(&buf),
-			 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
-		if (inserting)
-			goto err;
+	if (unlikely(bucket_data_type_mismatch(bucket_data_type, ptr_data_type))) {
+		bch2_log_msg_start(c, &buf);
+		prt_printf(&buf, "bucket %u:%zu gen %u different types of data in same bucket: %s, %s",
+			   ptr->dev, bucket_nr, b_gen,
+			   bch2_data_type_str(bucket_data_type),
+			   bch2_data_type_str(ptr_data_type));
+
+		ret = bucket_ref_update_err(trans, &buf, k, inserting,
+					    BCH_FSCK_ERR_ptr_bucket_data_type_mismatch);
 		goto out;
 	}
 
-	if ((u64) *bucket_sectors + sectors > U32_MAX) {
-		bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_allocations);
-		log_fsck_err(trans, bucket_sector_count_overflow,
-			"bucket %u:%zu gen %u data type %s sector count overflow: %u + %lli > U32_MAX\n"
-			"while marking %s",
+	if (unlikely((u64) *bucket_sectors + sectors > U32_MAX)) {
+		bch2_log_msg_start(c, &buf);
+		prt_printf(&buf,
+			"bucket %u:%zu gen %u data type %s sector count overflow: %u + %lli > U32_MAX",
 			ptr->dev, bucket_nr, b_gen,
 			bch2_data_type_str(bucket_data_type ?: ptr_data_type),
-			*bucket_sectors, sectors,
-			(printbuf_reset(&buf),
-			 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
-		if (inserting)
-			goto err;
+			*bucket_sectors, sectors);
+
+		ret = bucket_ref_update_err(trans, &buf, k, inserting,
+					    BCH_FSCK_ERR_bucket_sector_count_overflow);
 		sectors = -*bucket_sectors;
+		goto out;
 	}
 
 	*bucket_sectors += sectors;
 out:
 	printbuf_exit(&buf);
 	return ret;
-err:
-fsck_err:
-	bch2_dump_trans_updates(trans);
-	bch2_inconsistent_error(c);
-	ret = -BCH_ERR_bucket_ref_update;
-	goto out;
 }
 
 void bch2_trans_account_disk_usage_change(struct btree_trans *trans)
@@ -590,11 +605,9 @@ static int bch2_trigger_pointer(struct btree_trans *trans,
 		if (ret)
 			goto err;
 
-		if (!p.ptr.cached) {
-			ret = bch2_bucket_backpointer_mod(trans, k, &bp, insert);
-			if (ret)
-				goto err;
-		}
+		ret = bch2_bucket_backpointer_mod(trans, k, &bp, insert);
+		if (ret)
+			goto err;
 	}
 
 	if (flags & BTREE_TRIGGER_gc) {
@@ -653,9 +666,9 @@ static int bch2_trigger_stripe_ptr(struct btree_trans *trans,
 			stripe_blockcount_get(&s->v, p.ec.block) +
 			sectors);
 
-		struct disk_accounting_pos acc = {
-			.type = BCH_DISK_ACCOUNTING_replicas,
-		};
+		struct disk_accounting_pos acc;
+		memset(&acc, 0, sizeof(acc));
+		acc.type = BCH_DISK_ACCOUNTING_replicas;
 		bch2_bkey_to_replicas(&acc.replicas, bkey_i_to_s_c(&s->k_i));
 		acc.replicas.data_type = data_type;
 		ret = bch2_disk_accounting_mod(trans, &acc, &sectors, 1, false);
@@ -674,26 +687,28 @@ err:
 			return -BCH_ERR_ENOMEM_mark_stripe_ptr;
 		}
 
-		mutex_lock(&c->ec_stripes_heap_lock);
+		gc_stripe_lock(m);
 
 		if (!m || !m->alive) {
-			mutex_unlock(&c->ec_stripes_heap_lock);
+			gc_stripe_unlock(m);
 			struct printbuf buf = PRINTBUF;
+			bch2_log_msg_start(c, &buf);
+			prt_printf(&buf, "pointer to nonexistent stripe %llu\n  while marking ",
+				   (u64) p.ec.idx);
 			bch2_bkey_val_to_text(&buf, c, k);
-			bch_err_ratelimited(c, "pointer to nonexistent stripe %llu\n  while marking %s",
-					    (u64) p.ec.idx, buf.buf);
+			__bch2_inconsistent_error(c, &buf);
+			bch2_print_string_as_lines(KERN_ERR, buf.buf);
 			printbuf_exit(&buf);
-			bch2_inconsistent_error(c);
 			return -BCH_ERR_trigger_stripe_pointer;
 		}
 
 		m->block_sectors[p.ec.block] += sectors;
 
-		struct disk_accounting_pos acc = {
-			.type = BCH_DISK_ACCOUNTING_replicas,
-		};
+		struct disk_accounting_pos acc;
+		memset(&acc, 0, sizeof(acc));
+		acc.type = BCH_DISK_ACCOUNTING_replicas;
 		memcpy(&acc.replicas, &m->r.e, replicas_entry_bytes(&m->r.e));
-		mutex_unlock(&c->ec_stripes_heap_lock);
+		gc_stripe_unlock(m);
 
 		acc.replicas.data_type = data_type;
 		int ret = bch2_disk_accounting_mod(trans, &acc, &sectors, 1, true);
@@ -719,16 +734,14 @@ static int __trigger_extent(struct btree_trans *trans,
 		: BCH_DATA_user;
 	int ret = 0;
 
-	struct disk_accounting_pos acc_replicas_key = {
-		.type			= BCH_DISK_ACCOUNTING_replicas,
-		.replicas.data_type	= data_type,
-		.replicas.nr_devs	= 0,
-		.replicas.nr_required	= 1,
-	};
+	struct disk_accounting_pos acc_replicas_key;
+	memset(&acc_replicas_key, 0, sizeof(acc_replicas_key));
+	acc_replicas_key.type = BCH_DISK_ACCOUNTING_replicas;
+	acc_replicas_key.replicas.data_type	= data_type;
+	acc_replicas_key.replicas.nr_devs	= 0;
+	acc_replicas_key.replicas.nr_required	= 1;
 
-	struct disk_accounting_pos acct_compression_key = {
-		.type			= BCH_DISK_ACCOUNTING_compression,
-	};
+	unsigned cur_compression_type = 0;
 	u64 compression_acct[3] = { 1, 0, 0 };
 
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
@@ -762,13 +775,13 @@ static int __trigger_extent(struct btree_trans *trans,
 			acc_replicas_key.replicas.nr_required = 0;
 		}
 
-		if (acct_compression_key.compression.type &&
-		    acct_compression_key.compression.type != p.crc.compression_type) {
+		if (cur_compression_type &&
+		    cur_compression_type != p.crc.compression_type) {
 			if (flags & BTREE_TRIGGER_overwrite)
 				bch2_u64s_neg(compression_acct, ARRAY_SIZE(compression_acct));
 
-			ret = bch2_disk_accounting_mod(trans, &acct_compression_key, compression_acct,
-						       ARRAY_SIZE(compression_acct), gc);
+			ret = bch2_disk_accounting_mod2(trans, gc, compression_acct,
+							compression, cur_compression_type);
 			if (ret)
 				return ret;
 
@@ -777,7 +790,7 @@ static int __trigger_extent(struct btree_trans *trans,
 			compression_acct[2] = 0;
 		}
 
-		acct_compression_key.compression.type = p.crc.compression_type;
+		cur_compression_type = p.crc.compression_type;
 		if (p.crc.compression_type) {
 			compression_acct[1] += p.crc.uncompressed_size;
 			compression_acct[2] += p.crc.compressed_size;
@@ -791,45 +804,34 @@ static int __trigger_extent(struct btree_trans *trans,
 	}
 
 	if (acc_replicas_key.replicas.nr_devs && !level && k.k->p.snapshot) {
-		struct disk_accounting_pos acc_snapshot_key = {
-			.type			= BCH_DISK_ACCOUNTING_snapshot,
-			.snapshot.id		= k.k->p.snapshot,
-		};
-		ret = bch2_disk_accounting_mod(trans, &acc_snapshot_key, replicas_sectors, 1, gc);
+		ret = bch2_disk_accounting_mod2_nr(trans, gc, replicas_sectors, 1, snapshot, k.k->p.snapshot);
 		if (ret)
 			return ret;
 	}
 
-	if (acct_compression_key.compression.type) {
+	if (cur_compression_type) {
 		if (flags & BTREE_TRIGGER_overwrite)
 			bch2_u64s_neg(compression_acct, ARRAY_SIZE(compression_acct));
 
-		ret = bch2_disk_accounting_mod(trans, &acct_compression_key, compression_acct,
-					       ARRAY_SIZE(compression_acct), gc);
+		ret = bch2_disk_accounting_mod2(trans, gc, compression_acct,
+						compression, cur_compression_type);
 		if (ret)
 			return ret;
 	}
 
 	if (level) {
-		struct disk_accounting_pos acc_btree_key = {
-			.type		= BCH_DISK_ACCOUNTING_btree,
-			.btree.id	= btree_id,
-		};
-		ret = bch2_disk_accounting_mod(trans, &acc_btree_key, replicas_sectors, 1, gc);
+		ret = bch2_disk_accounting_mod2_nr(trans, gc, replicas_sectors, 1, btree, btree_id);
 		if (ret)
 			return ret;
 	} else {
 		bool insert = !(flags & BTREE_TRIGGER_overwrite);
-		struct disk_accounting_pos acc_inum_key = {
-			.type		= BCH_DISK_ACCOUNTING_inum,
-			.inum.inum	= k.k->p.inode,
-		};
+
 		s64 v[3] = {
 			insert ? 1 : -1,
 			insert ? k.k->size : -((s64) k.k->size),
 			*replicas_sectors,
 		};
-		ret = bch2_disk_accounting_mod(trans, &acc_inum_key, v, ARRAY_SIZE(v), gc);
+		ret = bch2_disk_accounting_mod2(trans, gc, v, inum, k.k->p.inode);
 		if (ret)
 			return ret;
 	}
@@ -878,15 +880,15 @@ int bch2_trigger_extent(struct btree_trans *trans,
 		}
 
 		int need_rebalance_delta = 0;
-		s64 need_rebalance_sectors_delta = 0;
+		s64 need_rebalance_sectors_delta[1] = { 0 };
 
 		s64 s = bch2_bkey_sectors_need_rebalance(c, old);
 		need_rebalance_delta -= s != 0;
-		need_rebalance_sectors_delta -= s;
+		need_rebalance_sectors_delta[0] -= s;
 
 		s = bch2_bkey_sectors_need_rebalance(c, new.s_c);
 		need_rebalance_delta += s != 0;
-		need_rebalance_sectors_delta += s;
+		need_rebalance_sectors_delta[0] += s;
 
 		if ((flags & BTREE_TRIGGER_transactional) && need_rebalance_delta) {
 			int ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_rebalance_work,
@@ -895,12 +897,9 @@ int bch2_trigger_extent(struct btree_trans *trans,
 				return ret;
 		}
 
-		if (need_rebalance_sectors_delta) {
-			struct disk_accounting_pos acc = {
-				.type		= BCH_DISK_ACCOUNTING_rebalance_work,
-			};
-			int ret = bch2_disk_accounting_mod(trans, &acc, &need_rebalance_sectors_delta, 1,
-							   flags & BTREE_TRIGGER_gc);
+		if (need_rebalance_sectors_delta[0]) {
+			int ret = bch2_disk_accounting_mod2(trans, flags & BTREE_TRIGGER_gc,
+							    need_rebalance_sectors_delta, rebalance_work);
 			if (ret)
 				return ret;
 		}
@@ -916,17 +915,13 @@ static int __trigger_reservation(struct btree_trans *trans,
 			enum btree_iter_update_trigger_flags flags)
 {
 	if (flags & (BTREE_TRIGGER_transactional|BTREE_TRIGGER_gc)) {
-		s64 sectors = k.k->size;
+		s64 sectors[1] = { k.k->size };
 
 		if (flags & BTREE_TRIGGER_overwrite)
-			sectors = -sectors;
+			sectors[0] = -sectors[0];
 
-		struct disk_accounting_pos acc = {
-			.type = BCH_DISK_ACCOUNTING_persistent_reserved,
-			.persistent_reserved.nr_replicas = bkey_s_c_to_reservation(k).v->nr_replicas,
-		};
-
-		return bch2_disk_accounting_mod(trans, &acc, &sectors, 1, flags & BTREE_TRIGGER_gc);
+		return bch2_disk_accounting_mod2(trans, flags & BTREE_TRIGGER_gc, sectors,
+				persistent_reserved, bkey_s_c_to_reservation(k).v->nr_replicas);
 	}
 
 	return 0;

libbcachefs/buckets.h

@@ -39,33 +39,6 @@ static inline u64 sector_to_bucket_and_offset(const struct bch_dev *ca, sector_t
 	for (_b = (_buckets)->b + (_buckets)->first_bucket;	\
 	     _b < (_buckets)->b + (_buckets)->nbuckets; _b++)
 
-/*
- * Ugly hack alert:
- *
- * We need to cram a spinlock in a single byte, because that's what we have left
- * in struct bucket, and we care about the size of these - during fsck, we need
- * in memory state for every single bucket on every device.
- *
- * We used to do
- *   while (xchg(&b->lock, 1) cpu_relax();
- * but, it turns out not all architectures support xchg on a single byte.
- *
- * So now we use bit_spin_lock(), with fun games since we can't burn a whole
- * ulong for this - we just need to make sure the lock bit always ends up in the
- * first byte.
- */
-
-#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
-#define BUCKET_LOCK_BITNR	0
-#else
-#define BUCKET_LOCK_BITNR	(BITS_PER_LONG - 1)
-#endif
-
-union ulong_byte_assert {
-	ulong	ulong;
-	u8	byte;
-};
-
 static inline void bucket_unlock(struct bucket *b)
 {
 	BUILD_BUG_ON(!((union ulong_byte_assert) { .ulong = 1UL << BUCKET_LOCK_BITNR }).byte);
@@ -167,9 +140,7 @@ static inline int gen_cmp(u8 a, u8 b)
 
 static inline int gen_after(u8 a, u8 b)
 {
-	int r = gen_cmp(a, b);
-
-	return r > 0 ? r : 0;
+	return max(0, gen_cmp(a, b));
 }
 
 static inline int dev_ptr_stale_rcu(struct bch_dev *ca, const struct bch_extent_ptr *ptr)

libbcachefs/buckets_types.h

@@ -7,6 +7,33 @@
 
 #define BUCKET_JOURNAL_SEQ_BITS		16
 
+/*
+ * Ugly hack alert:
+ *
+ * We need to cram a spinlock in a single byte, because that's what we have left
+ * in struct bucket, and we care about the size of these - during fsck, we need
+ * in memory state for every single bucket on every device.
+ *
+ * We used to do
+ *   while (xchg(&b->lock, 1) cpu_relax();
+ * but, it turns out not all architectures support xchg on a single byte.
+ *
+ * So now we use bit_spin_lock(), with fun games since we can't burn a whole
+ * ulong for this - we just need to make sure the lock bit always ends up in the
+ * first byte.
+ */
+
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+#define BUCKET_LOCK_BITNR	0
+#else
+#define BUCKET_LOCK_BITNR	(BITS_PER_LONG - 1)
+#endif
+
+union ulong_byte_assert {
+	ulong	ulong;
+	u8	byte;
+};
+
 struct bucket {
 	u8			lock;
 	u8			gen_valid:1;

libbcachefs/chardev.c

@@ -315,8 +315,10 @@ static int bch2_data_thread(void *arg)
 	ctx->thr.ret = bch2_data_job(ctx->c, &ctx->stats, ctx->arg);
 	if (ctx->thr.ret == -BCH_ERR_device_offline)
 		ctx->stats.ret = BCH_IOCTL_DATA_EVENT_RET_device_offline;
-	else
+	else {
 		ctx->stats.ret = BCH_IOCTL_DATA_EVENT_RET_done;
+		ctx->stats.data_type = (int) DATA_PROGRESS_DATA_TYPE_done;
+	}
 	return 0;
 }
 
@@ -424,10 +426,8 @@ static long bch2_ioctl_fs_usage(struct bch_fs *c,
 	arg.replica_entries_bytes = replicas.nr;
 
 	for (unsigned i = 0; i < BCH_REPLICAS_MAX; i++) {
-		struct disk_accounting_pos k = {
-			.type = BCH_DISK_ACCOUNTING_persistent_reserved,
-			.persistent_reserved.nr_replicas = i,
-		};
+		struct disk_accounting_pos k;
+		disk_accounting_key_init(k, persistent_reserved, .nr_replicas = i);
 
 		bch2_accounting_mem_read(c,
 					 disk_accounting_pos_to_bpos(&k),

libbcachefs/checksum.c

@@ -466,7 +466,7 @@ int bch2_rechecksum_bio(struct bch_fs *c, struct bio *bio,
 		prt_str(&buf, ")");
 		WARN_RATELIMIT(1, "%s", buf.buf);
 		printbuf_exit(&buf);
-		return -EIO;
+		return -BCH_ERR_recompute_checksum;
 	}
 
 	for (i = splits; i < splits + ARRAY_SIZE(splits); i++) {
@@ -693,6 +693,14 @@ static int bch2_alloc_ciphers(struct bch_fs *c)
 	return 0;
 }
 
+#if 0
+
+/*
+ * This seems to be duplicating code in cmd_remove_passphrase() in
+ * bcachefs-tools, but we might want to switch userspace to use this - and
+ * perhaps add an ioctl for calling this at runtime, so we can take the
+ * passphrase off of a mounted filesystem (which has come up).
+ */
 int bch2_disable_encryption(struct bch_fs *c)
 {
 	struct bch_sb_field_crypt *crypt;
@@ -725,6 +733,10 @@ out:
 	return ret;
 }
 
+/*
+ * For enabling encryption on an existing filesystem: not hooked up yet, but it
+ * should be
+ */
 int bch2_enable_encryption(struct bch_fs *c, bool keyed)
 {
 	struct bch_encrypted_key key;
@@ -781,6 +793,7 @@ err:
 	memzero_explicit(&key, sizeof(key));
 	return ret;
 }
+#endif
 
 void bch2_fs_encryption_exit(struct bch_fs *c)
 {
@@ -788,8 +801,6 @@ void bch2_fs_encryption_exit(struct bch_fs *c)
 		crypto_free_shash(c->poly1305);
 	if (c->chacha20)
 		crypto_free_sync_skcipher(c->chacha20);
-	if (c->sha256)
-		crypto_free_shash(c->sha256);
 }
 
 int bch2_fs_encryption_init(struct bch_fs *c)
@@ -798,14 +809,6 @@ int bch2_fs_encryption_init(struct bch_fs *c)
 	struct bch_key key;
 	int ret = 0;
 
-	c->sha256 = crypto_alloc_shash("sha256", 0, 0);
-	ret = PTR_ERR_OR_ZERO(c->sha256);
-	if (ret) {
-		c->sha256 = NULL;
-		bch_err(c, "error requesting sha256 module: %s", bch2_err_str(ret));
-		goto out;
-	}
-
 	crypt = bch2_sb_field_get(c->disk_sb.sb, crypt);
 	if (!crypt)
 		goto out;

libbcachefs/checksum.h

@@ -103,8 +103,10 @@ extern const struct bch_sb_field_ops bch_sb_field_ops_crypt;
 int bch2_decrypt_sb_key(struct bch_fs *, struct bch_sb_field_crypt *,
 			struct bch_key *);
 
+#if 0
 int bch2_disable_encryption(struct bch_fs *);
 int bch2_enable_encryption(struct bch_fs *, bool);
+#endif
 
 void bch2_fs_encryption_exit(struct bch_fs *);
 int bch2_fs_encryption_init(struct bch_fs *);

libbcachefs/compress.c

@@ -177,7 +177,7 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 	size_t src_len = src->bi_iter.bi_size;
 	size_t dst_len = crc.uncompressed_size << 9;
 	void *workspace;
-	int ret;
+	int ret = 0, ret2;
 
 	enum bch_compression_opts opt = bch2_compression_type_to_opt(crc.compression_type);
 	mempool_t *workspace_pool = &c->compress_workspace[opt];
@@ -189,7 +189,7 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 		else
 			ret = -BCH_ERR_compression_workspace_not_initialized;
 		if (ret)
-			goto out;
+			goto err;
 	}
 
 	src_data = bio_map_or_bounce(c, src, READ);
@@ -197,10 +197,10 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 	switch (crc.compression_type) {
 	case BCH_COMPRESSION_TYPE_lz4_old:
 	case BCH_COMPRESSION_TYPE_lz4:
-		ret = LZ4_decompress_safe_partial(src_data.b, dst_data,
-						  src_len, dst_len, dst_len);
-		if (ret != dst_len)
-			goto err;
+		ret2 = LZ4_decompress_safe_partial(src_data.b, dst_data,
+						   src_len, dst_len, dst_len);
+		if (ret2 != dst_len)
+			ret = -BCH_ERR_decompress_lz4;
 		break;
 	case BCH_COMPRESSION_TYPE_gzip: {
 		z_stream strm = {
@@ -214,45 +214,43 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 
 		zlib_set_workspace(&strm, workspace);
 		zlib_inflateInit2(&strm, -MAX_WBITS);
-		ret = zlib_inflate(&strm, Z_FINISH);
+		ret2 = zlib_inflate(&strm, Z_FINISH);
 
 		mempool_free(workspace, workspace_pool);
 
-		if (ret != Z_STREAM_END)
-			goto err;
+		if (ret2 != Z_STREAM_END)
+			ret = -BCH_ERR_decompress_gzip;
 		break;
 	}
 	case BCH_COMPRESSION_TYPE_zstd: {
 		ZSTD_DCtx *ctx;
 		size_t real_src_len = le32_to_cpup(src_data.b);
 
-		if (real_src_len > src_len - 4)
+		if (real_src_len > src_len - 4) {
+			ret = -BCH_ERR_decompress_zstd_src_len_bad;
 			goto err;
+		}
 
 		workspace = mempool_alloc(workspace_pool, GFP_NOFS);
 		ctx = zstd_init_dctx(workspace, zstd_dctx_workspace_bound());
 
-		ret = zstd_decompress_dctx(ctx,
+		ret2 = zstd_decompress_dctx(ctx,
 				dst_data,	dst_len,
 				src_data.b + 4, real_src_len);
 
 		mempool_free(workspace, workspace_pool);
 
-		if (ret != dst_len)
-			goto err;
+		if (ret2 != dst_len)
+			ret = -BCH_ERR_decompress_zstd;
 		break;
 	}
 	default:
 		BUG();
 	}
-	ret = 0;
+err:
 fsck_err:
-out:
 	bio_unmap_or_unbounce(c, src_data);
 	return ret;
-err:
-	ret = -EIO;
-	goto out;
 }
 
 int bch2_bio_uncompress_inplace(struct bch_write_op *op,
@@ -268,27 +266,22 @@ int bch2_bio_uncompress_inplace(struct bch_write_op *op,
 	BUG_ON(!bio->bi_vcnt);
 	BUG_ON(DIV_ROUND_UP(crc->live_size, PAGE_SECTORS) > bio->bi_max_vecs);
 
-	if (crc->uncompressed_size << 9	> c->opts.encoded_extent_max ||
-	    crc->compressed_size << 9	> c->opts.encoded_extent_max) {
-		struct printbuf buf = PRINTBUF;
-		bch2_write_op_error(&buf, op);
-		prt_printf(&buf, "error rewriting existing data: extent too big");
-		bch_err_ratelimited(c, "%s", buf.buf);
-		printbuf_exit(&buf);
-		return -EIO;
+	if (crc->uncompressed_size << 9	> c->opts.encoded_extent_max) {
+		bch2_write_op_error(op, op->pos.offset,
+				    "extent too big to decompress (%u > %u)",
+				    crc->uncompressed_size << 9, c->opts.encoded_extent_max);
+		return -BCH_ERR_decompress_exceeded_max_encoded_extent;
 	}
 
 	data = __bounce_alloc(c, dst_len, WRITE);
 
-	if (__bio_uncompress(c, bio, data.b, *crc)) {
-		if (!c->opts.no_data_io) {
-			struct printbuf buf = PRINTBUF;
-			bch2_write_op_error(&buf, op);
-			prt_printf(&buf, "error rewriting existing data: decompression error");
-			bch_err_ratelimited(c, "%s", buf.buf);
-			printbuf_exit(&buf);
-		}
-		ret = -EIO;
+	ret = __bio_uncompress(c, bio, data.b, *crc);
+
+	if (c->opts.no_data_io)
+		ret = 0;
+
+	if (ret) {
+		bch2_write_op_error(op, op->pos.offset, "%s", bch2_err_str(ret));
 		goto err;
 	}
 
@@ -321,7 +314,7 @@ int bch2_bio_uncompress(struct bch_fs *c, struct bio *src,
 
 	if (crc.uncompressed_size << 9	> c->opts.encoded_extent_max ||
 	    crc.compressed_size << 9	> c->opts.encoded_extent_max)
-		return -EIO;
+		return -BCH_ERR_decompress_exceeded_max_encoded_extent;
 
 	dst_data = dst_len == dst_iter.bi_size
 		? __bio_map_or_bounce(c, dst, dst_iter, WRITE)

libbcachefs/data_update.c

@@ -22,6 +22,13 @@
 
 #include <linux/ioprio.h>
 
+static const char * const bch2_data_update_type_strs[] = {
+#define x(t, n, ...) [n] = #t,
+	BCH_DATA_UPDATE_TYPES()
+#undef x
+	NULL
+};
+
 static void bkey_put_dev_refs(struct bch_fs *c, struct bkey_s_c k)
 {
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
@@ -93,7 +100,7 @@ static bool bkey_nocow_lock(struct bch_fs *c, struct moving_context *ctxt, struc
 	return true;
 }
 
-static noinline void trace_move_extent_finish2(struct data_update *u,
+static noinline void trace_io_move_finish2(struct data_update *u,
 					       struct bkey_i *new,
 					       struct bkey_i *insert)
 {
@@ -113,11 +120,11 @@ static noinline void trace_move_extent_finish2(struct data_update *u,
 	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(insert));
 	prt_newline(&buf);
 
-	trace_move_extent_finish(c, buf.buf);
+	trace_io_move_finish(c, buf.buf);
 	printbuf_exit(&buf);
 }
 
-static void trace_move_extent_fail2(struct data_update *m,
+static void trace_io_move_fail2(struct data_update *m,
 			 struct bkey_s_c new,
 			 struct bkey_s_c wrote,
 			 struct bkey_i *insert,
@@ -128,7 +135,7 @@ static void trace_move_extent_fail2(struct data_update *m,
 	struct printbuf buf = PRINTBUF;
 	unsigned rewrites_found = 0;
 
-	if (!trace_move_extent_fail_enabled())
+	if (!trace_io_move_fail_enabled())
 		return;
 
 	prt_str(&buf, msg);
@@ -168,7 +175,7 @@ static void trace_move_extent_fail2(struct data_update *m,
 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(insert));
 	}
 
-	trace_move_extent_fail(c, buf.buf);
+	trace_io_move_fail(c, buf.buf);
 	printbuf_exit(&buf);
 }
 
@@ -181,6 +188,7 @@ static int __bch2_data_update_index_update(struct btree_trans *trans,
 		container_of(op, struct data_update, op);
 	struct keylist *keys = &op->insert_keys;
 	struct bkey_buf _new, _insert;
+	struct printbuf journal_msg = PRINTBUF;
 	int ret = 0;
 
 	bch2_bkey_buf_init(&_new);
@@ -216,7 +224,7 @@ static int __bch2_data_update_index_update(struct btree_trans *trans,
 		new = bkey_i_to_extent(bch2_keylist_front(keys));
 
 		if (!bch2_extents_match(k, old)) {
-			trace_move_extent_fail2(m, k, bkey_i_to_s_c(&new->k_i),
+			trace_io_move_fail2(m, k, bkey_i_to_s_c(&new->k_i),
 						NULL, "no match:");
 			goto nowork;
 		}
@@ -256,7 +264,7 @@ static int __bch2_data_update_index_update(struct btree_trans *trans,
 		if (m->data_opts.rewrite_ptrs &&
 		    !rewrites_found &&
 		    bch2_bkey_durability(c, k) >= m->op.opts.data_replicas) {
-			trace_move_extent_fail2(m, k, bkey_i_to_s_c(&new->k_i), insert, "no rewrites found:");
+			trace_io_move_fail2(m, k, bkey_i_to_s_c(&new->k_i), insert, "no rewrites found:");
 			goto nowork;
 		}
 
@@ -273,7 +281,7 @@ restart_drop_conflicting_replicas:
 			}
 
 		if (!bkey_val_u64s(&new->k)) {
-			trace_move_extent_fail2(m, k, bkey_i_to_s_c(&new->k_i), insert, "new replicas conflicted:");
+			trace_io_move_fail2(m, k, bkey_i_to_s_c(&new->k_i), insert, "new replicas conflicted:");
 			goto nowork;
 		}
 
@@ -342,6 +350,7 @@ restart_drop_extra_replicas:
 			struct printbuf buf = PRINTBUF;
 
 			prt_str(&buf, "about to insert invalid key in data update path");
+			prt_printf(&buf, "\nop.nonce: %u", m->op.nonce);
 			prt_str(&buf, "\nold: ");
 			bch2_bkey_val_to_text(&buf, c, old);
 			prt_str(&buf, "\nk:   ");
@@ -353,7 +362,7 @@ restart_drop_extra_replicas:
 			printbuf_exit(&buf);
 
 			bch2_fatal_error(c);
-			ret = -EIO;
+			ret = -BCH_ERR_invalid_bkey;
 			goto out;
 		}
 
@@ -371,7 +380,12 @@ restart_drop_extra_replicas:
 			printbuf_exit(&buf);
 		}
 
-		ret =   bch2_insert_snapshot_whiteouts(trans, m->btree_id,
+		printbuf_reset(&journal_msg);
+		prt_str(&journal_msg, bch2_data_update_type_strs[m->type]);
+
+		ret =   bch2_trans_log_msg(trans, &journal_msg) ?:
+			bch2_trans_log_bkey(trans, m->btree_id, 0, m->k.k) ?:
+			bch2_insert_snapshot_whiteouts(trans, m->btree_id,
 						k.k->p, bkey_start_pos(&insert->k)) ?:
 			bch2_insert_snapshot_whiteouts(trans, m->btree_id,
 						k.k->p, insert->k.p) ?:
@@ -386,9 +400,9 @@ restart_drop_extra_replicas:
 		if (!ret) {
 			bch2_btree_iter_set_pos(&iter, next_pos);
 
-			this_cpu_add(c->counters[BCH_COUNTER_move_extent_finish], new->k.size);
-			if (trace_move_extent_finish_enabled())
-				trace_move_extent_finish2(m, &new->k_i, insert);
+			this_cpu_add(c->counters[BCH_COUNTER_io_move_finish], new->k.size);
+			if (trace_io_move_finish_enabled())
+				trace_io_move_finish2(m, &new->k_i, insert);
 		}
 err:
 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
@@ -410,12 +424,13 @@ nowork:
 				     &m->stats->sectors_raced);
 		}
 
-		count_event(c, move_extent_fail);
+		count_event(c, io_move_fail);
 
 		bch2_btree_iter_advance(&iter);
 		goto next;
 	}
 out:
+	printbuf_exit(&journal_msg);
 	bch2_trans_iter_exit(trans, &iter);
 	bch2_bkey_buf_exit(&_insert, c);
 	bch2_bkey_buf_exit(&_new, c);
@@ -438,7 +453,7 @@ void bch2_data_update_read_done(struct data_update *m)
 	m->op.crc = m->rbio.pick.crc;
 	m->op.wbio.bio.bi_iter.bi_size = m->op.crc.compressed_size << 9;
 
-	this_cpu_add(m->op.c->counters[BCH_COUNTER_move_extent_write], m->k.k->k.size);
+	this_cpu_add(m->op.c->counters[BCH_COUNTER_io_move_write], m->k.k->k.size);
 
 	closure_call(&m->op.cl, bch2_write, NULL, NULL);
 }
@@ -572,11 +587,13 @@ void bch2_data_update_opts_to_text(struct printbuf *out, struct bch_fs *c,
 
 	prt_str_indented(out, "extra replicas:\t");
 	prt_u64(out, data_opts->extra_replicas);
-	prt_newline(out);
 }
 
 void bch2_data_update_to_text(struct printbuf *out, struct data_update *m)
 {
+	prt_str(out, bch2_data_update_type_strs[m->type]);
+	prt_newline(out);
+
 	bch2_data_update_opts_to_text(out, m->op.c, &m->op.opts, &m->data_opts);
 	prt_newline(out);
 
@@ -638,11 +655,46 @@ int bch2_extent_drop_ptrs(struct btree_trans *trans,
 		bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
 }
 
-static bool can_allocate_without_blocking(struct bch_fs *c,
-					  struct data_update *m)
+int bch2_data_update_bios_init(struct data_update *m, struct bch_fs *c,
+			       struct bch_io_opts *io_opts)
 {
-	if (unlikely(c->open_buckets_nr_free <= bch2_open_buckets_reserved(m->op.watermark)))
-		return false;
+	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(bkey_i_to_s_c(m->k.k));
+	const union bch_extent_entry *entry;
+	struct extent_ptr_decoded p;
+
+	/* write path might have to decompress data: */
+	unsigned buf_bytes = 0;
+	bkey_for_each_ptr_decode(&m->k.k->k, ptrs, p, entry)
+		buf_bytes = max_t(unsigned, buf_bytes, p.crc.uncompressed_size << 9);
+
+	unsigned nr_vecs = DIV_ROUND_UP(buf_bytes, PAGE_SIZE);
+
+	m->bvecs = kmalloc_array(nr_vecs, sizeof*(m->bvecs), GFP_KERNEL);
+	if (!m->bvecs)
+		return -ENOMEM;
+
+	bio_init(&m->rbio.bio,		NULL, m->bvecs, nr_vecs, REQ_OP_READ);
+	bio_init(&m->op.wbio.bio,	NULL, m->bvecs, nr_vecs, 0);
+
+	if (bch2_bio_alloc_pages(&m->op.wbio.bio, buf_bytes, GFP_KERNEL)) {
+		kfree(m->bvecs);
+		m->bvecs = NULL;
+		return -ENOMEM;
+	}
+
+	rbio_init(&m->rbio.bio, c, *io_opts, NULL);
+	m->rbio.data_update		= true;
+	m->rbio.bio.bi_iter.bi_size	= buf_bytes;
+	m->rbio.bio.bi_iter.bi_sector	= bkey_start_offset(&m->k.k->k);
+	m->op.wbio.bio.bi_ioprio	= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0);
+	return 0;
+}
+
+static int can_write_extent(struct bch_fs *c, struct data_update *m)
+{
+	if ((m->op.flags & BCH_WRITE_alloc_nowait) &&
+	    unlikely(c->open_buckets_nr_free <= bch2_open_buckets_reserved(m->op.watermark)))
+		return -BCH_ERR_data_update_done_would_block;
 
 	unsigned target = m->op.flags & BCH_WRITE_only_specified_devs
 		? m->op.target
@@ -669,7 +721,11 @@ static bool can_allocate_without_blocking(struct bch_fs *c,
 	}
 	rcu_read_unlock();
 
-	return nr_replicas >= m->op.nr_replicas;
+	if (!nr_replicas)
+		return -BCH_ERR_data_update_done_no_rw_devs;
+	if (nr_replicas < m->op.nr_replicas)
+		return -BCH_ERR_insufficient_devices;
+	return 0;
 }
 
 int bch2_data_update_init(struct btree_trans *trans,
@@ -677,7 +733,7 @@ int bch2_data_update_init(struct btree_trans *trans,
 			  struct moving_context *ctxt,
 			  struct data_update *m,
 			  struct write_point_specifier wp,
-			  struct bch_io_opts io_opts,
+			  struct bch_io_opts *io_opts,
 			  struct data_update_opts data_opts,
 			  enum btree_id btree_id,
 			  struct bkey_s_c k)
@@ -699,12 +755,15 @@ int bch2_data_update_init(struct btree_trans *trans,
 
 	bch2_bkey_buf_init(&m->k);
 	bch2_bkey_buf_reassemble(&m->k, c, k);
+	m->type		= data_opts.btree_insert_flags & BCH_WATERMARK_copygc
+		? BCH_DATA_UPDATE_copygc
+		: BCH_DATA_UPDATE_rebalance;
 	m->btree_id	= btree_id;
 	m->data_opts	= data_opts;
 	m->ctxt		= ctxt;
 	m->stats	= ctxt ? ctxt->stats : NULL;
 
-	bch2_write_op_init(&m->op, c, io_opts);
+	bch2_write_op_init(&m->op, c, *io_opts);
 	m->op.pos	= bkey_start_pos(k.k);
 	m->op.version	= k.k->bversion;
 	m->op.target	= data_opts.target;
@@ -715,7 +774,7 @@ int bch2_data_update_init(struct btree_trans *trans,
 		BCH_WRITE_data_encoded|
 		BCH_WRITE_move|
 		m->data_opts.write_flags;
-	m->op.compression_opt	= io_opts.background_compression;
+	m->op.compression_opt	= io_opts->background_compression;
 	m->op.watermark		= m->data_opts.btree_insert_flags & BCH_WATERMARK_MASK;
 
 	unsigned durability_have = 0, durability_removing = 0;
@@ -753,7 +812,7 @@ int bch2_data_update_init(struct btree_trans *trans,
 		ptr_bit <<= 1;
 	}
 
-	unsigned durability_required = max(0, (int) (io_opts.data_replicas - durability_have));
+	unsigned durability_required = max(0, (int) (io_opts->data_replicas - durability_have));
 
 	/*
 	 * If current extent durability is less than io_opts.data_replicas,
@@ -786,17 +845,28 @@ int bch2_data_update_init(struct btree_trans *trans,
 		m->data_opts.rewrite_ptrs = 0;
 		/* if iter == NULL, it's just a promote */
 		if (iter)
-			ret = bch2_extent_drop_ptrs(trans, iter, k, &io_opts, &m->data_opts);
+			ret = bch2_extent_drop_ptrs(trans, iter, k, io_opts, &m->data_opts);
 		if (!ret)
 			ret = -BCH_ERR_data_update_done_no_writes_needed;
 		goto out_bkey_buf_exit;
 	}
 
-	if ((m->op.flags & BCH_WRITE_alloc_nowait) &&
-	    !can_allocate_without_blocking(c, m)) {
-		ret = -BCH_ERR_data_update_done_would_block;
+	/*
+	 * Check if the allocation will succeed, to avoid getting an error later
+	 * in bch2_write() -> bch2_alloc_sectors_start() and doing a useless
+	 * read:
+	 *
+	 * This guards against
+	 * - BCH_WRITE_alloc_nowait allocations failing (promotes)
+	 * - Destination target full
+	 * - Device(s) in destination target offline
+	 * - Insufficient durability available in destination target
+	 *   (i.e. trying to move a durability=2 replica to a target with a
+	 *   single durability=2 device)
+	 */
+	ret = can_write_extent(c, m);
+	if (ret)
 		goto out_bkey_buf_exit;
-	}
 
 	if (reserve_sectors) {
 		ret = bch2_disk_reservation_add(c, &m->op.res, reserve_sectors,
@@ -824,33 +894,11 @@ int bch2_data_update_init(struct btree_trans *trans,
 		goto out_nocow_unlock;
 	}
 
-	/* write path might have to decompress data: */
-	unsigned buf_bytes = 0;
-	bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
-		buf_bytes = max_t(unsigned, buf_bytes, p.crc.uncompressed_size << 9);
-
-	unsigned nr_vecs = DIV_ROUND_UP(buf_bytes, PAGE_SIZE);
-
-	m->bvecs = kmalloc_array(nr_vecs, sizeof*(m->bvecs), GFP_KERNEL);
-	if (!m->bvecs)
-		goto enomem;
-
-	bio_init(&m->rbio.bio,		NULL, m->bvecs, nr_vecs, REQ_OP_READ);
-	bio_init(&m->op.wbio.bio,	NULL, m->bvecs, nr_vecs, 0);
-
-	if (bch2_bio_alloc_pages(&m->op.wbio.bio, buf_bytes, GFP_KERNEL))
-		goto enomem;
-
-	rbio_init(&m->rbio.bio, c, io_opts, NULL);
-	m->rbio.bio.bi_iter.bi_size	= buf_bytes;
-	m->rbio.bio.bi_iter.bi_sector	= bkey_start_offset(k.k);
-	m->op.wbio.bio.bi_ioprio	= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0);
+	ret = bch2_data_update_bios_init(m, c, io_opts);
+	if (ret)
+		goto out_nocow_unlock;
 
 	return 0;
-enomem:
-	ret = -ENOMEM;
-	kfree(m->bvecs);
-	m->bvecs = NULL;
 out_nocow_unlock:
 	if (c->opts.nocow_enabled)
 		bkey_nocow_unlock(c, k);

libbcachefs/data_update.h

@@ -24,7 +24,19 @@ struct data_update_opts {
 void bch2_data_update_opts_to_text(struct printbuf *, struct bch_fs *,
 				   struct bch_io_opts *, struct data_update_opts *);
 
+#define BCH_DATA_UPDATE_TYPES()		\
+	x(copygc,	0)		\
+	x(rebalance,	1)		\
+	x(promote,	2)
+
+enum bch_data_update_types {
+#define x(n, id)	BCH_DATA_UPDATE_##n = id,
+	BCH_DATA_UPDATE_TYPES()
+#undef x
+};
+
 struct data_update {
+	enum bch_data_update_types type;
 	/* extent being updated: */
 	bool			read_done;
 	enum btree_id		btree_id;
@@ -51,12 +63,15 @@ int bch2_extent_drop_ptrs(struct btree_trans *,
 			  struct bch_io_opts *,
 			  struct data_update_opts *);
 
+int bch2_data_update_bios_init(struct data_update *, struct bch_fs *,
+			       struct bch_io_opts *);
+
 void bch2_data_update_exit(struct data_update *);
 int bch2_data_update_init(struct btree_trans *, struct btree_iter *,
 			  struct moving_context *,
 			  struct data_update *,
 			  struct write_point_specifier,
-			  struct bch_io_opts, struct data_update_opts,
+			  struct bch_io_opts *, struct data_update_opts,
 			  enum btree_id, struct bkey_s_c);
 void bch2_data_update_opts_normalize(struct bkey_s_c, struct data_update_opts *);
 

libbcachefs/dirent.c

@@ -13,6 +13,40 @@
 
 #include <linux/dcache.h>
 
+static int bch2_casefold(struct btree_trans *trans, const struct bch_hash_info *info,
+				const struct qstr *str, struct qstr *out_cf)
+{
+	*out_cf = (struct qstr) QSTR_INIT(NULL, 0);
+
+#ifdef CONFIG_UNICODE
+	unsigned char *buf = bch2_trans_kmalloc(trans, BCH_NAME_MAX + 1);
+	int ret = PTR_ERR_OR_ZERO(buf);
+	if (ret)
+		return ret;
+
+	ret = utf8_casefold(info->cf_encoding, str, buf, BCH_NAME_MAX + 1);
+	if (ret <= 0)
+		return ret;
+
+	*out_cf = (struct qstr) QSTR_INIT(buf, ret);
+	return 0;
+#else
+	return -EOPNOTSUPP;
+#endif
+}
+
+static inline int bch2_maybe_casefold(struct btree_trans *trans,
+				      const struct bch_hash_info *info,
+				      const struct qstr *str, struct qstr *out_cf)
+{
+	if (likely(!info->cf_encoding)) {
+		*out_cf = *str;
+		return 0;
+	} else {
+		return bch2_casefold(trans, info, str, out_cf);
+	}
+}
+
 static unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
 {
 	if (bkey_val_bytes(d.k) < offsetof(struct bch_dirent, d_name))
@@ -28,13 +62,38 @@ static unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
 #endif
 
 	return bkey_bytes -
-		offsetof(struct bch_dirent, d_name) -
+		(d.v->d_casefold
+		? offsetof(struct bch_dirent, d_cf_name_block.d_names)
+		: offsetof(struct bch_dirent, d_name)) -
 		trailing_nuls;
 }
 
 struct qstr bch2_dirent_get_name(struct bkey_s_c_dirent d)
 {
-	return (struct qstr) QSTR_INIT(d.v->d_name, bch2_dirent_name_bytes(d));
+	if (d.v->d_casefold) {
+		unsigned name_len = le16_to_cpu(d.v->d_cf_name_block.d_name_len);
+		return (struct qstr) QSTR_INIT(&d.v->d_cf_name_block.d_names[0], name_len);
+	} else {
+		return (struct qstr) QSTR_INIT(d.v->d_name, bch2_dirent_name_bytes(d));
+	}
+}
+
+static struct qstr bch2_dirent_get_casefold_name(struct bkey_s_c_dirent d)
+{
+	if (d.v->d_casefold) {
+		unsigned name_len = le16_to_cpu(d.v->d_cf_name_block.d_name_len);
+		unsigned cf_name_len = le16_to_cpu(d.v->d_cf_name_block.d_cf_name_len);
+		return (struct qstr) QSTR_INIT(&d.v->d_cf_name_block.d_names[name_len], cf_name_len);
+	} else {
+		return (struct qstr) QSTR_INIT(NULL, 0);
+	}
+}
+
+static inline struct qstr bch2_dirent_get_lookup_name(struct bkey_s_c_dirent d)
+{
+	return d.v->d_casefold
+		? bch2_dirent_get_casefold_name(d)
+		: bch2_dirent_get_name(d);
 }
 
 static u64 bch2_dirent_hash(const struct bch_hash_info *info,
@@ -57,7 +116,7 @@ static u64 dirent_hash_key(const struct bch_hash_info *info, const void *key)
 static u64 dirent_hash_bkey(const struct bch_hash_info *info, struct bkey_s_c k)
 {
 	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
-	struct qstr name = bch2_dirent_get_name(d);
+	struct qstr name = bch2_dirent_get_lookup_name(d);
 
 	return bch2_dirent_hash(info, &name);
 }
@@ -65,7 +124,7 @@ static u64 dirent_hash_bkey(const struct bch_hash_info *info, struct bkey_s_c k)
 static bool dirent_cmp_key(struct bkey_s_c _l, const void *_r)
 {
 	struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
-	const struct qstr l_name = bch2_dirent_get_name(l);
+	const struct qstr l_name = bch2_dirent_get_lookup_name(l);
 	const struct qstr *r_name = _r;
 
 	return !qstr_eq(l_name, *r_name);
@@ -75,8 +134,8 @@ static bool dirent_cmp_bkey(struct bkey_s_c _l, struct bkey_s_c _r)
 {
 	struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
 	struct bkey_s_c_dirent r = bkey_s_c_to_dirent(_r);
-	const struct qstr l_name = bch2_dirent_get_name(l);
-	const struct qstr r_name = bch2_dirent_get_name(r);
+	const struct qstr l_name = bch2_dirent_get_lookup_name(l);
+	const struct qstr r_name = bch2_dirent_get_lookup_name(r);
 
 	return !qstr_eq(l_name, r_name);
 }
@@ -104,17 +163,19 @@ int bch2_dirent_validate(struct bch_fs *c, struct bkey_s_c k,
 			 struct bkey_validate_context from)
 {
 	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
+	unsigned name_block_len = bch2_dirent_name_bytes(d);
 	struct qstr d_name = bch2_dirent_get_name(d);
+	struct qstr d_cf_name = bch2_dirent_get_casefold_name(d);
 	int ret = 0;
 
 	bkey_fsck_err_on(!d_name.len,
 			 c, dirent_empty_name,
 			 "empty name");
 
-	bkey_fsck_err_on(bkey_val_u64s(k.k) > dirent_val_u64s(d_name.len),
+	bkey_fsck_err_on(d_name.len + d_cf_name.len > name_block_len,
 			 c, dirent_val_too_big,
-			 "value too big (%zu > %u)",
-			 bkey_val_u64s(k.k), dirent_val_u64s(d_name.len));
+			 "dirent names exceed bkey size (%d + %d > %d)",
+			 d_name.len, d_cf_name.len, name_block_len);
 
 	/*
 	 * Check new keys don't exceed the max length
@@ -142,6 +203,18 @@ int bch2_dirent_validate(struct bch_fs *c, struct bkey_s_c k,
 			 le64_to_cpu(d.v->d_inum) == d.k->p.inode,
 			 c, dirent_to_itself,
 			 "dirent points to own directory");
+
+	if (d.v->d_casefold) {
+		bkey_fsck_err_on(from.from == BKEY_VALIDATE_commit &&
+				 d_cf_name.len > BCH_NAME_MAX,
+				 c, dirent_cf_name_too_big,
+				 "dirent w/ cf name too big (%u > %u)",
+				 d_cf_name.len, BCH_NAME_MAX);
+
+		bkey_fsck_err_on(d_cf_name.len != strnlen(d_cf_name.name, d_cf_name.len),
+				 c, dirent_stray_data_after_cf_name,
+				 "dirent has stray data after cf name's NUL");
+	}
 fsck_err:
 	return ret;
 }
@@ -163,15 +236,14 @@ void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c
 	prt_printf(out, " type %s", bch2_d_type_str(d.v->d_type));
 }
 
-static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
-				subvol_inum dir, u8 type,
-				const struct qstr *name, u64 dst)
+static struct bkey_i_dirent *dirent_alloc_key(struct btree_trans *trans,
+				subvol_inum dir,
+				u8 type,
+				int name_len, int cf_name_len,
+				u64 dst)
 {
 	struct bkey_i_dirent *dirent;
-	unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
-
-	if (name->len > BCH_NAME_MAX)
-		return ERR_PTR(-ENAMETOOLONG);
+	unsigned u64s = BKEY_U64s + dirent_val_u64s(name_len, cf_name_len);
 
 	BUG_ON(u64s > U8_MAX);
 
@@ -190,14 +262,65 @@ static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
 	}
 
 	dirent->v.d_type = type;
+	dirent->v.d_unused = 0;
+	dirent->v.d_casefold = cf_name_len ? 1 : 0;
 
-	memcpy(dirent->v.d_name, name->name, name->len);
-	memset(dirent->v.d_name + name->len, 0,
-	       bkey_val_bytes(&dirent->k) -
-	       offsetof(struct bch_dirent, d_name) -
-	       name->len);
+	return dirent;
+}
 
-	EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
+static void dirent_init_regular_name(struct bkey_i_dirent *dirent,
+				     const struct qstr *name)
+{
+	EBUG_ON(dirent->v.d_casefold);
+
+	memcpy(&dirent->v.d_name[0], name->name, name->len);
+	memset(&dirent->v.d_name[name->len], 0,
+		bkey_val_bytes(&dirent->k) -
+		offsetof(struct bch_dirent, d_name) -
+		name->len);
+}
+
+static void dirent_init_casefolded_name(struct bkey_i_dirent *dirent,
+					const struct qstr *name,
+					const struct qstr *cf_name)
+{
+	EBUG_ON(!dirent->v.d_casefold);
+	EBUG_ON(!cf_name->len);
+
+	dirent->v.d_cf_name_block.d_name_len = name->len;
+	dirent->v.d_cf_name_block.d_cf_name_len = cf_name->len;
+	memcpy(&dirent->v.d_cf_name_block.d_names[0], name->name, name->len);
+	memcpy(&dirent->v.d_cf_name_block.d_names[name->len], cf_name->name, cf_name->len);
+	memset(&dirent->v.d_cf_name_block.d_names[name->len + cf_name->len], 0,
+		bkey_val_bytes(&dirent->k) -
+		offsetof(struct bch_dirent, d_cf_name_block.d_names) -
+		name->len + cf_name->len);
+
+	EBUG_ON(bch2_dirent_get_casefold_name(dirent_i_to_s_c(dirent)).len != cf_name->len);
+}
+
+static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
+				subvol_inum dir,
+				u8 type,
+				const struct qstr *name,
+				const struct qstr *cf_name,
+				u64 dst)
+{
+	struct bkey_i_dirent *dirent;
+
+	if (name->len > BCH_NAME_MAX)
+		return ERR_PTR(-ENAMETOOLONG);
+
+	dirent = dirent_alloc_key(trans, dir, type, name->len, cf_name ? cf_name->len : 0, dst);
+	if (IS_ERR(dirent))
+		return dirent;
+
+	if (cf_name)
+		dirent_init_casefolded_name(dirent, name, cf_name);
+	else
+		dirent_init_regular_name(dirent, name);
+
+	EBUG_ON(bch2_dirent_get_name(dirent_i_to_s_c(dirent)).len != name->len);
 
 	return dirent;
 }
@@ -213,7 +336,7 @@ int bch2_dirent_create_snapshot(struct btree_trans *trans,
 	struct bkey_i_dirent *dirent;
 	int ret;
 
-	dirent = dirent_create_key(trans, dir_inum, type, name, dst_inum);
+	dirent = dirent_create_key(trans, dir_inum, type, name, NULL, dst_inum);
 	ret = PTR_ERR_OR_ZERO(dirent);
 	if (ret)
 		return ret;
@@ -233,16 +356,28 @@ int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
 		       const struct bch_hash_info *hash_info,
 		       u8 type, const struct qstr *name, u64 dst_inum,
 		       u64 *dir_offset,
+		       u64 *i_size,
 		       enum btree_iter_update_trigger_flags flags)
 {
 	struct bkey_i_dirent *dirent;
 	int ret;
 
-	dirent = dirent_create_key(trans, dir, type, name, dst_inum);
+	if (hash_info->cf_encoding) {
+		struct qstr cf_name;
+		ret = bch2_casefold(trans, hash_info, name, &cf_name);
+		if (ret)
+			return ret;
+		dirent = dirent_create_key(trans, dir, type, name, &cf_name, dst_inum);
+	} else {
+		dirent = dirent_create_key(trans, dir, type, name, NULL, dst_inum);
+	}
+
 	ret = PTR_ERR_OR_ZERO(dirent);
 	if (ret)
 		return ret;
 
+	*i_size += bkey_bytes(&dirent->k);
+
 	ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
 			    dir, &dirent->k_i, flags);
 	*dir_offset = dirent->k.p.offset;
@@ -275,12 +410,13 @@ int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
 }
 
 int bch2_dirent_rename(struct btree_trans *trans,
-		subvol_inum src_dir, struct bch_hash_info *src_hash,
-		subvol_inum dst_dir, struct bch_hash_info *dst_hash,
+		subvol_inum src_dir, struct bch_hash_info *src_hash, u64 *src_dir_i_size,
+		subvol_inum dst_dir, struct bch_hash_info *dst_hash, u64 *dst_dir_i_size,
 		const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
 		const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
 		enum bch_rename_mode mode)
 {
+	struct qstr src_name_lookup, dst_name_lookup;
 	struct btree_iter src_iter = { NULL };
 	struct btree_iter dst_iter = { NULL };
 	struct bkey_s_c old_src, old_dst = bkey_s_c_null;
@@ -295,8 +431,11 @@ int bch2_dirent_rename(struct btree_trans *trans,
 	memset(dst_inum, 0, sizeof(*dst_inum));
 
 	/* Lookup src: */
+	ret = bch2_maybe_casefold(trans, src_hash, src_name, &src_name_lookup);
+	if (ret)
+		goto out;
 	old_src = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
-				   src_hash, src_dir, src_name,
+				   src_hash, src_dir, &src_name_lookup,
 				   BTREE_ITER_intent);
 	ret = bkey_err(old_src);
 	if (ret)
@@ -308,6 +447,9 @@ int bch2_dirent_rename(struct btree_trans *trans,
 		goto out;
 
 	/* Lookup dst: */
+	ret = bch2_maybe_casefold(trans, dst_hash, dst_name, &dst_name_lookup);
+	if (ret)
+		goto out;
 	if (mode == BCH_RENAME) {
 		/*
 		 * Note that we're _not_ checking if the target already exists -
@@ -315,12 +457,12 @@ int bch2_dirent_rename(struct btree_trans *trans,
 		 * correctness:
 		 */
 		ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
-				     dst_hash, dst_dir, dst_name);
+				     dst_hash, dst_dir, &dst_name_lookup);
 		if (ret)
 			goto out;
 	} else {
 		old_dst = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
-					    dst_hash, dst_dir, dst_name,
+					    dst_hash, dst_dir, &dst_name_lookup,
 					    BTREE_ITER_intent);
 		ret = bkey_err(old_dst);
 		if (ret)
@@ -336,7 +478,8 @@ int bch2_dirent_rename(struct btree_trans *trans,
 		*src_offset = dst_iter.pos.offset;
 
 	/* Create new dst key: */
-	new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
+	new_dst = dirent_create_key(trans, dst_dir, 0, dst_name,
+				    dst_hash->cf_encoding ? &dst_name_lookup : NULL, 0);
 	ret = PTR_ERR_OR_ZERO(new_dst);
 	if (ret)
 		goto out;
@@ -346,7 +489,8 @@ int bch2_dirent_rename(struct btree_trans *trans,
 
 	/* Create new src key: */
 	if (mode == BCH_RENAME_EXCHANGE) {
-		new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
+		new_src = dirent_create_key(trans, src_dir, 0, src_name,
+					    src_hash->cf_encoding ? &src_name_lookup : NULL, 0);
 		ret = PTR_ERR_OR_ZERO(new_src);
 		if (ret)
 			goto out;
@@ -406,6 +550,14 @@ int bch2_dirent_rename(struct btree_trans *trans,
 	    new_src->v.d_type == DT_SUBVOL)
 		new_src->v.d_parent_subvol = cpu_to_le32(src_dir.subvol);
 
+	if (old_dst.k)
+		*dst_dir_i_size -= bkey_bytes(old_dst.k);
+	*src_dir_i_size -= bkey_bytes(old_src.k);
+
+	if (mode == BCH_RENAME_EXCHANGE)
+		*src_dir_i_size += bkey_bytes(&new_src->k);
+	*dst_dir_i_size += bkey_bytes(&new_dst->k);
+
 	ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
 	if (ret)
 		goto out;
@@ -465,9 +617,14 @@ int bch2_dirent_lookup_trans(struct btree_trans *trans,
 			     const struct qstr *name, subvol_inum *inum,
 			     unsigned flags)
 {
+	struct qstr lookup_name;
+	int ret = bch2_maybe_casefold(trans, hash_info, name, &lookup_name);
+	if (ret)
+		return ret;
+
 	struct bkey_s_c k = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
-					     hash_info, dir, name, flags);
-	int ret = bkey_err(k);
+					     hash_info, dir, &lookup_name, flags);
+	ret = bkey_err(k);
 	if (ret)
 		goto err;
 
@@ -572,3 +729,54 @@ int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
 
 	return ret < 0 ? ret : 0;
 }
+
+/* fsck */
+
+static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
+			      struct bch_inode_unpacked *inode)
+{
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	int ret;
+
+	for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inode_nr),
+				     BTREE_ITER_all_snapshots, k, ret) {
+		if (k.k->p.offset != inode_nr)
+			break;
+		if (!bkey_is_inode(k.k))
+			continue;
+		ret = bch2_inode_unpack(k, inode);
+		goto found;
+	}
+	ret = -BCH_ERR_ENOENT_inode;
+found:
+	bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
+}
+
+int bch2_fsck_remove_dirent(struct btree_trans *trans, struct bpos pos)
+{
+	struct bch_fs *c = trans->c;
+	struct btree_iter iter;
+	struct bch_inode_unpacked dir_inode;
+	struct bch_hash_info dir_hash_info;
+	int ret;
+
+	ret = lookup_first_inode(trans, pos.inode, &dir_inode);
+	if (ret)
+		goto err;
+
+	dir_hash_info = bch2_hash_info_init(c, &dir_inode);
+
+	bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_intent);
+
+	ret =   bch2_btree_iter_traverse(&iter) ?:
+		bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
+				    &dir_hash_info, &iter,
+				    BTREE_UPDATE_internal_snapshot_node);
+	bch2_trans_iter_exit(trans, &iter);
+err:
+	bch_err_fn(c, ret);
+	return ret;
+}

libbcachefs/dirent.h

@@ -25,15 +25,13 @@ struct bch_inode_info;
 
 struct qstr bch2_dirent_get_name(struct bkey_s_c_dirent d);
 
-static inline unsigned dirent_val_u64s(unsigned len)
+static inline unsigned dirent_val_u64s(unsigned len, unsigned cf_len)
 {
-	return DIV_ROUND_UP(offsetof(struct bch_dirent, d_name) + len,
-			    sizeof(u64));
-}
+	unsigned bytes = cf_len
+		? offsetof(struct bch_dirent, d_cf_name_block.d_names) + len + cf_len
+		: offsetof(struct bch_dirent, d_name) + len;
 
-static inline unsigned int dirent_occupied_size(const struct qstr *name)
-{
-	return (BKEY_U64s + dirent_val_u64s(name->len)) * sizeof(u64);
+	return DIV_ROUND_UP(bytes, sizeof(u64));
 }
 
 int bch2_dirent_read_target(struct btree_trans *, subvol_inum,
@@ -52,7 +50,7 @@ int bch2_dirent_create_snapshot(struct btree_trans *, u32, u64, u32,
 			enum btree_iter_update_trigger_flags);
 int bch2_dirent_create(struct btree_trans *, subvol_inum,
 		       const struct bch_hash_info *, u8,
-		       const struct qstr *, u64, u64 *,
+		       const struct qstr *, u64, u64 *, u64 *,
 		       enum btree_iter_update_trigger_flags);
 
 static inline unsigned vfs_d_type(unsigned type)
@@ -67,8 +65,8 @@ enum bch_rename_mode {
 };
 
 int bch2_dirent_rename(struct btree_trans *,
-		       subvol_inum, struct bch_hash_info *,
-		       subvol_inum, struct bch_hash_info *,
+		       subvol_inum, struct bch_hash_info *, u64 *,
+		       subvol_inum, struct bch_hash_info *, u64 *,
 		       const struct qstr *, subvol_inum *, u64 *,
 		       const struct qstr *, subvol_inum *, u64 *,
 		       enum bch_rename_mode);
@@ -84,4 +82,6 @@ int bch2_empty_dir_snapshot(struct btree_trans *, u64, u32, u32);
 int bch2_empty_dir_trans(struct btree_trans *, subvol_inum);
 int bch2_readdir(struct bch_fs *, subvol_inum, struct dir_context *);
 
+int bch2_fsck_remove_dirent(struct btree_trans *, struct bpos);
+
 #endif /* _BCACHEFS_DIRENT_H */

libbcachefs/dirent_format.h

@@ -29,9 +29,25 @@ struct bch_dirent {
 	 * Copy of mode bits 12-15 from the target inode - so userspace can get
 	 * the filetype without having to do a stat()
 	 */
-	__u8			d_type;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u8			d_type:5,
+				d_unused:2,
+				d_casefold:1;
+#elif defined(__BIG_ENDIAN_BITFIELD)
+	__u8			d_casefold:1,
+				d_unused:2,
+				d_type:5;
+#endif
 
-	__u8			d_name[];
+	union {
+	struct {
+		__u8		d_pad;
+		__le16		d_name_len;
+		__le16		d_cf_name_len;
+		__u8		d_names[];
+	} d_cf_name_block __packed;
+	__DECLARE_FLEX_ARRAY(__u8, d_name);
+	} __packed;
 } __packed __aligned(8);
 
 #define DT_SUBVOL	16

libbcachefs/disk_accounting.c

@@ -114,10 +114,9 @@ int bch2_mod_dev_cached_sectors(struct btree_trans *trans,
 				unsigned dev, s64 sectors,
 				bool gc)
 {
-	struct disk_accounting_pos acc = {
-		.type = BCH_DISK_ACCOUNTING_replicas,
-	};
-
+	struct disk_accounting_pos acc;
+	memset(&acc, 0, sizeof(acc));
+	acc.type = BCH_DISK_ACCOUNTING_replicas;
 	bch2_replicas_entry_cached(&acc.replicas, dev);
 
 	return bch2_disk_accounting_mod(trans, &acc, &sectors, 1, gc);
@@ -135,6 +134,12 @@ static inline bool is_zero(char *start, char *end)
 
 #define field_end(p, member)	(((void *) (&p.member)) + sizeof(p.member))
 
+static const unsigned bch2_accounting_type_nr_counters[] = {
+#define x(f, id, nr)	[BCH_DISK_ACCOUNTING_##f]	= nr,
+	BCH_DISK_ACCOUNTING_TYPES()
+#undef x
+};
+
 int bch2_accounting_validate(struct bch_fs *c, struct bkey_s_c k,
 			     struct bkey_validate_context from)
 {
@@ -193,6 +198,11 @@ int bch2_accounting_validate(struct bch_fs *c, struct bkey_s_c k,
 	bkey_fsck_err_on(!is_zero(end, (void *) (&acc_k + 1)),
 			 c, accounting_key_junk_at_end,
 			 "junk at end of accounting key");
+
+	bkey_fsck_err_on(bch2_accounting_counters(k.k) != bch2_accounting_type_nr_counters[acc_k.type],
+			 c, accounting_key_nr_counters_wrong,
+			 "accounting key with %u counters, should be %u",
+			 bch2_accounting_counters(k.k), bch2_accounting_type_nr_counters[acc_k.type]);
 fsck_err:
 	return ret;
 }
@@ -635,7 +645,7 @@ static int bch2_disk_accounting_validate_late(struct btree_trans *trans,
 
 		if (fsck_err_on(!bch2_replicas_marked_locked(c, &r.e),
 				trans, accounting_replicas_not_marked,
-				"accounting not marked in superblock replicas\n  %s",
+				"accounting not marked in superblock replicas\n%s",
 				(printbuf_reset(&buf),
 				 bch2_accounting_key_to_text(&buf, &acc),
 				 buf.buf))) {
@@ -665,7 +675,7 @@ fsck_err:
 	return ret;
 invalid_device:
 	if (fsck_err(trans, accounting_to_invalid_device,
-		     "accounting entry points to invalid device %i\n  %s",
+		     "accounting entry points to invalid device %i\n%s",
 		     invalid_dev,
 		     (printbuf_reset(&buf),
 		      bch2_accounting_key_to_text(&buf, &acc),
@@ -726,7 +736,9 @@ int bch2_accounting_read(struct bch_fs *c)
 				break;
 
 			if (!bch2_accounting_is_mem(acc_k)) {
-				struct disk_accounting_pos next = { .type = acc_k.type + 1 };
+				struct disk_accounting_pos next;
+				memset(&next, 0, sizeof(next));
+				next.type = acc_k.type + 1;
 				bch2_btree_iter_set_pos(&iter, disk_accounting_pos_to_bpos(&next));
 				continue;
 			}
@@ -882,15 +894,13 @@ int bch2_dev_usage_remove(struct bch_fs *c, unsigned dev)
 int bch2_dev_usage_init(struct bch_dev *ca, bool gc)
 {
 	struct bch_fs *c = ca->fs;
-	struct disk_accounting_pos acc = {
-		.type = BCH_DISK_ACCOUNTING_dev_data_type,
-		.dev_data_type.dev = ca->dev_idx,
-		.dev_data_type.data_type = BCH_DATA_free,
-	};
 	u64 v[3] = { ca->mi.nbuckets - ca->mi.first_bucket, 0, 0 };
 
 	int ret = bch2_trans_do(c, ({
-		bch2_disk_accounting_mod(trans, &acc, v, ARRAY_SIZE(v), gc) ?:
+		bch2_disk_accounting_mod2(trans, gc,
+					  v, dev_data_type,
+					  .dev = ca->dev_idx,
+					  .data_type = BCH_DATA_free) ?:
 		(!gc ? bch2_trans_commit(trans, NULL, NULL, 0) : 0);
 	}));
 	bch_err_fn(c, ret);
@@ -917,7 +927,9 @@ void bch2_verify_accounting_clean(struct bch_fs *c)
 				break;
 
 			if (!bch2_accounting_is_mem(acc_k)) {
-				struct disk_accounting_pos next = { .type = acc_k.type + 1 };
+				struct disk_accounting_pos next;
+				memset(&next, 0, sizeof(next));
+				next.type = acc_k.type + 1;
 				bch2_btree_iter_set_pos(&iter, disk_accounting_pos_to_bpos(&next));
 				continue;
 			}

libbcachefs/disk_accounting.h

@@ -33,10 +33,12 @@ static inline bool bch2_accounting_key_is_zero(struct bkey_s_c_accounting a)
 static inline void bch2_accounting_accumulate(struct bkey_i_accounting *dst,
 					      struct bkey_s_c_accounting src)
 {
-	EBUG_ON(dst->k.u64s != src.k->u64s);
-
-	for (unsigned i = 0; i < bch2_accounting_counters(&dst->k); i++)
+	for (unsigned i = 0;
+	     i < min(bch2_accounting_counters(&dst->k),
+		     bch2_accounting_counters(src.k));
+	     i++)
 		dst->v.d[i] += src.v->d[i];
+
 	if (bversion_cmp(dst->k.bversion, src.k->bversion) < 0)
 		dst->k.bversion = src.k->bversion;
 }
@@ -85,6 +87,24 @@ static inline struct bpos disk_accounting_pos_to_bpos(struct disk_accounting_pos
 
 int bch2_disk_accounting_mod(struct btree_trans *, struct disk_accounting_pos *,
 			     s64 *, unsigned, bool);
+
+#define disk_accounting_key_init(_k, _type, ...)			\
+do {									\
+	memset(&(_k), 0, sizeof(_k));					\
+	(_k).type	= BCH_DISK_ACCOUNTING_##_type;			\
+	(_k)._type	= (struct bch_acct_##_type) { __VA_ARGS__ };	\
+} while (0)
+
+#define bch2_disk_accounting_mod2_nr(_trans, _gc, _v, _nr, ...)		\
+({									\
+	struct disk_accounting_pos pos;					\
+	disk_accounting_key_init(pos, __VA_ARGS__);			\
+	bch2_disk_accounting_mod(trans, &pos, _v, _nr, _gc);		\
+})
+
+#define bch2_disk_accounting_mod2(_trans, _gc, _v, ...)			\
+	bch2_disk_accounting_mod2_nr(_trans, _gc, _v, ARRAY_SIZE(_v), __VA_ARGS__)
+
 int bch2_mod_dev_cached_sectors(struct btree_trans *, unsigned, s64, bool);
 
 int bch2_accounting_validate(struct bch_fs *, struct bkey_s_c,
@@ -210,11 +230,13 @@ static inline void bch2_accounting_mem_read_counters(struct bch_accounting_mem *
 static inline void bch2_accounting_mem_read(struct bch_fs *c, struct bpos p,
 					    u64 *v, unsigned nr)
 {
+	percpu_down_read(&c->mark_lock);
 	struct bch_accounting_mem *acc = &c->accounting;
 	unsigned idx = eytzinger0_find(acc->k.data, acc->k.nr, sizeof(acc->k.data[0]),
 				       accounting_pos_cmp, &p);
 
 	bch2_accounting_mem_read_counters(acc, idx, v, nr, false);
+	percpu_up_read(&c->mark_lock);
 }
 
 static inline struct bversion journal_pos_to_bversion(struct journal_res *res, unsigned offset)

libbcachefs/disk_accounting_format.h

@@ -95,40 +95,81 @@ static inline bool data_type_is_hidden(enum bch_data_type type)
 	}
 }
 
+/*
+ * field 1: name
+ * field 2: id
+ * field 3: number of counters (max 3)
+ */
+
 #define BCH_DISK_ACCOUNTING_TYPES()		\
-	x(nr_inodes,		0)		\
-	x(persistent_reserved,	1)		\
-	x(replicas,		2)		\
-	x(dev_data_type,	3)		\
-	x(compression,		4)		\
-	x(snapshot,		5)		\
-	x(btree,		6)		\
-	x(rebalance_work,	7)		\
-	x(inum,			8)
+	x(nr_inodes,		0,	1)	\
+	x(persistent_reserved,	1,	1)	\
+	x(replicas,		2,	1)	\
+	x(dev_data_type,	3,	3)	\
+	x(compression,		4,	3)	\
+	x(snapshot,		5,	1)	\
+	x(btree,		6,	1)	\
+	x(rebalance_work,	7,	1)	\
+	x(inum,			8,	3)
 
 enum disk_accounting_type {
-#define x(f, nr)	BCH_DISK_ACCOUNTING_##f	= nr,
+#define x(f, nr, ...)	BCH_DISK_ACCOUNTING_##f	= nr,
 	BCH_DISK_ACCOUNTING_TYPES()
 #undef x
 	BCH_DISK_ACCOUNTING_TYPE_NR,
 };
 
-struct bch_nr_inodes {
+/*
+ * No subtypes - number of inodes in the entire filesystem
+ *
+ * XXX: perhaps we could add a per-subvolume counter?
+ */
+struct bch_acct_nr_inodes {
 };
 
-struct bch_persistent_reserved {
+/*
+ * Tracks KEY_TYPE_reservation sectors, broken out by number of replicas for the
+ * reservation:
+ */
+struct bch_acct_persistent_reserved {
 	__u8			nr_replicas;
 };
 
-struct bch_dev_data_type {
+/*
+ * device, data type counter fields:
+ * [
+ *   nr_buckets
+ *   live sectors (in buckets of that data type)
+ *   sectors of internal fragmentation
+ * ]
+ *
+ * XXX: live sectors should've been done differently, you can have multiple data
+ * types in the same bucket (user, stripe, cached) and this collapses them to
+ * the bucket data type, and makes the internal fragmentation counter redundant
+ */
+struct bch_acct_dev_data_type {
 	__u8			dev;
 	__u8			data_type;
 };
 
+/*
+ * Compression type fields:
+ * [
+ *   number of extents
+ *   uncompressed size
+ *   compressed size
+ * ]
+ *
+ * Compression ratio, average extent size (fragmentation).
+ */
 struct bch_acct_compression {
 	__u8			type;
 };
 
+/*
+ * On disk usage by snapshot id; counts same values as replicas counter, but
+ * aggregated differently
+ */
 struct bch_acct_snapshot {
 	__u32			id;
 } __packed;
@@ -137,10 +178,27 @@ struct bch_acct_btree {
 	__u32			id;
 } __packed;
 
+/*
+ * inum counter fields:
+ * [
+ *   number of extents
+ *   sum of extent sizes - bkey size
+ *     this field is similar to inode.bi_sectors, except here extents in
+ *     different snapshots but the same inode number are all collapsed to the
+ *     same counter
+ *   sum of on disk size - same values tracked by replicas counters
+ * ]
+ *
+ * This tracks on disk fragmentation.
+ */
 struct bch_acct_inum {
 	__u64			inum;
 } __packed;
 
+/*
+ * Simple counter of the amount of data (on disk sectors) rebalance needs to
+ * move, extents counted here are also in the rebalance_work btree.
+ */
 struct bch_acct_rebalance_work {
 };
 
@@ -149,10 +207,10 @@ struct disk_accounting_pos {
 	struct {
 		__u8				type;
 		union {
-		struct bch_nr_inodes		nr_inodes;
-		struct bch_persistent_reserved	persistent_reserved;
+		struct bch_acct_nr_inodes	nr_inodes;
+		struct bch_acct_persistent_reserved	persistent_reserved;
 		struct bch_replicas_entry_v1	replicas;
-		struct bch_dev_data_type	dev_data_type;
+		struct bch_acct_dev_data_type	dev_data_type;
 		struct bch_acct_compression	compression;
 		struct bch_acct_snapshot	snapshot;
 		struct bch_acct_btree		btree;

libbcachefs/ec.c

@@ -20,6 +20,7 @@
 #include "io_read.h"
 #include "io_write.h"
 #include "keylist.h"
+#include "lru.h"
 #include "recovery.h"
 #include "replicas.h"
 #include "super-io.h"
@@ -104,6 +105,7 @@ struct ec_bio {
 	struct bch_dev		*ca;
 	struct ec_stripe_buf	*buf;
 	size_t			idx;
+	u64			submit_time;
 	struct bio		bio;
 };
 
@@ -298,15 +300,27 @@ static int mark_stripe_bucket(struct btree_trans *trans,
 	struct bpos bucket = PTR_BUCKET_POS(ca, ptr);
 
 	if (flags & BTREE_TRIGGER_transactional) {
+		struct extent_ptr_decoded p = {
+			.ptr = *ptr,
+			.crc = bch2_extent_crc_unpack(s.k, NULL),
+		};
+		struct bkey_i_backpointer bp;
+		bch2_extent_ptr_to_bp(c, BTREE_ID_stripes, 0, s.s_c, p,
+				      (const union bch_extent_entry *) ptr, &bp);
+
 		struct bkey_i_alloc_v4 *a =
 			bch2_trans_start_alloc_update(trans, bucket, 0);
-		ret = PTR_ERR_OR_ZERO(a) ?:
-			__mark_stripe_bucket(trans, ca, s, ptr_idx, deleting, bucket, &a->v, flags);
+		ret   = PTR_ERR_OR_ZERO(a) ?:
+			__mark_stripe_bucket(trans, ca, s, ptr_idx, deleting, bucket, &a->v, flags) ?:
+			bch2_bucket_backpointer_mod(trans, s.s_c, &bp,
+						    !(flags & BTREE_TRIGGER_overwrite));
+		if (ret)
+			goto err;
 	}
 
 	if (flags & BTREE_TRIGGER_gc) {
 		struct bucket *g = gc_bucket(ca, bucket.offset);
-		if (bch2_fs_inconsistent_on(!g, c, "reference to invalid bucket on device %u\n  %s",
+		if (bch2_fs_inconsistent_on(!g, c, "reference to invalid bucket on device %u\n%s",
 					    ptr->dev,
 					    (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
 			ret = -BCH_ERR_mark_stripe;
@@ -366,19 +380,6 @@ static int mark_stripe_buckets(struct btree_trans *trans,
 	return 0;
 }
 
-static inline void stripe_to_mem(struct stripe *m, const struct bch_stripe *s)
-{
-	m->sectors	= le16_to_cpu(s->sectors);
-	m->algorithm	= s->algorithm;
-	m->nr_blocks	= s->nr_blocks;
-	m->nr_redundant	= s->nr_redundant;
-	m->disk_label	= s->disk_label;
-	m->blocks_nonempty = 0;
-
-	for (unsigned i = 0; i < s->nr_blocks; i++)
-		m->blocks_nonempty += !!stripe_blockcount_get(s, i);
-}
-
 int bch2_trigger_stripe(struct btree_trans *trans,
 			enum btree_id btree, unsigned level,
 			struct bkey_s_c old, struct bkey_s _new,
@@ -399,6 +400,15 @@ int bch2_trigger_stripe(struct btree_trans *trans,
 	       (new_s->nr_blocks	!= old_s->nr_blocks ||
 		new_s->nr_redundant	!= old_s->nr_redundant));
 
+	if (flags & BTREE_TRIGGER_transactional) {
+		int ret = bch2_lru_change(trans,
+					  BCH_LRU_STRIPE_FRAGMENTATION,
+					  idx,
+					  stripe_lru_pos(old_s),
+					  stripe_lru_pos(new_s));
+		if (ret)
+			return ret;
+	}
 
 	if (flags & (BTREE_TRIGGER_transactional|BTREE_TRIGGER_gc)) {
 		/*
@@ -443,9 +453,9 @@ int bch2_trigger_stripe(struct btree_trans *trans,
 		if (new_s) {
 			s64 sectors = (u64) le16_to_cpu(new_s->sectors) * new_s->nr_redundant;
 
-			struct disk_accounting_pos acc = {
-				.type = BCH_DISK_ACCOUNTING_replicas,
-			};
+			struct disk_accounting_pos acc;
+			memset(&acc, 0, sizeof(acc));
+			acc.type = BCH_DISK_ACCOUNTING_replicas;
 			bch2_bkey_to_replicas(&acc.replicas, new);
 			int ret = bch2_disk_accounting_mod(trans, &acc, &sectors, 1, gc);
 			if (ret)
@@ -458,9 +468,9 @@ int bch2_trigger_stripe(struct btree_trans *trans,
 		if (old_s) {
 			s64 sectors = -((s64) le16_to_cpu(old_s->sectors)) * old_s->nr_redundant;
 
-			struct disk_accounting_pos acc = {
-				.type = BCH_DISK_ACCOUNTING_replicas,
-			};
+			struct disk_accounting_pos acc;
+			memset(&acc, 0, sizeof(acc));
+			acc.type = BCH_DISK_ACCOUNTING_replicas;
 			bch2_bkey_to_replicas(&acc.replicas, old);
 			int ret = bch2_disk_accounting_mod(trans, &acc, &sectors, 1, gc);
 			if (ret)
@@ -472,38 +482,6 @@ int bch2_trigger_stripe(struct btree_trans *trans,
 			return ret;
 	}
 
-	if (flags & BTREE_TRIGGER_atomic) {
-		struct stripe *m = genradix_ptr(&c->stripes, idx);
-
-		if (!m) {
-			struct printbuf buf1 = PRINTBUF;
-			struct printbuf buf2 = PRINTBUF;
-
-			bch2_bkey_val_to_text(&buf1, c, old);
-			bch2_bkey_val_to_text(&buf2, c, new);
-			bch_err_ratelimited(c, "error marking nonexistent stripe %llu while marking\n"
-					    "old %s\n"
-					    "new %s", idx, buf1.buf, buf2.buf);
-			printbuf_exit(&buf2);
-			printbuf_exit(&buf1);
-			bch2_inconsistent_error(c);
-			return -1;
-		}
-
-		if (!new_s) {
-			bch2_stripes_heap_del(c, m, idx);
-
-			memset(m, 0, sizeof(*m));
-		} else {
-			stripe_to_mem(m, new_s);
-
-			if (!old_s)
-				bch2_stripes_heap_insert(c, m, idx);
-			else
-				bch2_stripes_heap_update(c, m, idx);
-		}
-	}
-
 	return 0;
 }
 
@@ -726,14 +704,15 @@ static void ec_block_endio(struct bio *bio)
 	struct bch_dev *ca = ec_bio->ca;
 	struct closure *cl = bio->bi_private;
 
-	if (bch2_dev_io_err_on(bio->bi_status, ca,
-			       bio_data_dir(bio)
-			       ? BCH_MEMBER_ERROR_write
-			       : BCH_MEMBER_ERROR_read,
-			       "erasure coding %s error: %s",
+	bch2_account_io_completion(ca, bio_data_dir(bio),
+				   ec_bio->submit_time, !bio->bi_status);
+
+	if (bio->bi_status) {
+		bch_err_dev_ratelimited(ca, "erasure coding %s error: %s",
 			       str_write_read(bio_data_dir(bio)),
-			       bch2_blk_status_to_str(bio->bi_status)))
+			       bch2_blk_status_to_str(bio->bi_status));
 		clear_bit(ec_bio->idx, ec_bio->buf->valid);
+	}
 
 	int stale = dev_ptr_stale(ca, ptr);
 	if (stale) {
@@ -796,6 +775,7 @@ static void ec_block_io(struct bch_fs *c, struct ec_stripe_buf *buf,
 		ec_bio->ca			= ca;
 		ec_bio->buf			= buf;
 		ec_bio->idx			= idx;
+		ec_bio->submit_time		= local_clock();
 
 		ec_bio->bio.bi_iter.bi_sector	= ptr->offset + buf->offset + (offset >> 9);
 		ec_bio->bio.bi_end_io		= ec_block_endio;
@@ -917,26 +897,6 @@ err:
 
 static int __ec_stripe_mem_alloc(struct bch_fs *c, size_t idx, gfp_t gfp)
 {
-	ec_stripes_heap n, *h = &c->ec_stripes_heap;
-
-	if (idx >= h->size) {
-		if (!init_heap(&n, max(1024UL, roundup_pow_of_two(idx + 1)), gfp))
-			return -BCH_ERR_ENOMEM_ec_stripe_mem_alloc;
-
-		mutex_lock(&c->ec_stripes_heap_lock);
-		if (n.size > h->size) {
-			memcpy(n.data, h->data, h->nr * sizeof(h->data[0]));
-			n.nr = h->nr;
-			swap(*h, n);
-		}
-		mutex_unlock(&c->ec_stripes_heap_lock);
-
-		free_heap(&n);
-	}
-
-	if (!genradix_ptr_alloc(&c->stripes, idx, gfp))
-		return -BCH_ERR_ENOMEM_ec_stripe_mem_alloc;
-
 	if (c->gc_pos.phase != GC_PHASE_not_running &&
 	    !genradix_ptr_alloc(&c->gc_stripes, idx, gfp))
 		return -BCH_ERR_ENOMEM_ec_stripe_mem_alloc;
@@ -1009,180 +969,50 @@ static void bch2_stripe_close(struct bch_fs *c, struct ec_stripe_new *s)
 	s->idx = 0;
 }
 
-/* Heap of all existing stripes, ordered by blocks_nonempty */
-
-static u64 stripe_idx_to_delete(struct bch_fs *c)
-{
-	ec_stripes_heap *h = &c->ec_stripes_heap;
-
-	lockdep_assert_held(&c->ec_stripes_heap_lock);
-
-	if (h->nr &&
-	    h->data[0].blocks_nonempty == 0 &&
-	    !bch2_stripe_is_open(c, h->data[0].idx))
-		return h->data[0].idx;
-
-	return 0;
-}
-
-static inline void ec_stripes_heap_set_backpointer(ec_stripes_heap *h,
-						   size_t i)
-{
-	struct bch_fs *c = container_of(h, struct bch_fs, ec_stripes_heap);
-
-	genradix_ptr(&c->stripes, h->data[i].idx)->heap_idx = i;
-}
-
-static inline bool ec_stripes_heap_cmp(const void *l, const void *r, void __always_unused *args)
-{
-	struct ec_stripe_heap_entry *_l = (struct ec_stripe_heap_entry *)l;
-	struct ec_stripe_heap_entry *_r = (struct ec_stripe_heap_entry *)r;
-
-	return ((_l->blocks_nonempty > _r->blocks_nonempty) <
-		(_l->blocks_nonempty < _r->blocks_nonempty));
-}
-
-static inline void ec_stripes_heap_swap(void *l, void *r, void *h)
-{
-	struct ec_stripe_heap_entry *_l = (struct ec_stripe_heap_entry *)l;
-	struct ec_stripe_heap_entry *_r = (struct ec_stripe_heap_entry *)r;
-	ec_stripes_heap *_h = (ec_stripes_heap *)h;
-	size_t i = _l - _h->data;
-	size_t j = _r - _h->data;
-
-	swap(*_l, *_r);
-
-	ec_stripes_heap_set_backpointer(_h, i);
-	ec_stripes_heap_set_backpointer(_h, j);
-}
-
-static const struct min_heap_callbacks callbacks = {
-	.less = ec_stripes_heap_cmp,
-	.swp = ec_stripes_heap_swap,
-};
-
-static void heap_verify_backpointer(struct bch_fs *c, size_t idx)
-{
-	ec_stripes_heap *h = &c->ec_stripes_heap;
-	struct stripe *m = genradix_ptr(&c->stripes, idx);
-
-	BUG_ON(m->heap_idx >= h->nr);
-	BUG_ON(h->data[m->heap_idx].idx != idx);
-}
-
-void bch2_stripes_heap_del(struct bch_fs *c,
-			   struct stripe *m, size_t idx)
-{
-	mutex_lock(&c->ec_stripes_heap_lock);
-	heap_verify_backpointer(c, idx);
-
-	min_heap_del(&c->ec_stripes_heap, m->heap_idx, &callbacks, &c->ec_stripes_heap);
-	mutex_unlock(&c->ec_stripes_heap_lock);
-}
-
-void bch2_stripes_heap_insert(struct bch_fs *c,
-			      struct stripe *m, size_t idx)
-{
-	mutex_lock(&c->ec_stripes_heap_lock);
-	BUG_ON(min_heap_full(&c->ec_stripes_heap));
-
-	genradix_ptr(&c->stripes, idx)->heap_idx = c->ec_stripes_heap.nr;
-	min_heap_push(&c->ec_stripes_heap, &((struct ec_stripe_heap_entry) {
-			.idx = idx,
-			.blocks_nonempty = m->blocks_nonempty,
-		}),
-		&callbacks,
-		&c->ec_stripes_heap);
-
-	heap_verify_backpointer(c, idx);
-	mutex_unlock(&c->ec_stripes_heap_lock);
-}
-
-void bch2_stripes_heap_update(struct bch_fs *c,
-			      struct stripe *m, size_t idx)
-{
-	ec_stripes_heap *h = &c->ec_stripes_heap;
-	bool do_deletes;
-	size_t i;
-
-	mutex_lock(&c->ec_stripes_heap_lock);
-	heap_verify_backpointer(c, idx);
-
-	h->data[m->heap_idx].blocks_nonempty = m->blocks_nonempty;
-
-	i = m->heap_idx;
-	min_heap_sift_up(h,	i, &callbacks, &c->ec_stripes_heap);
-	min_heap_sift_down(h, i, &callbacks, &c->ec_stripes_heap);
-
-	heap_verify_backpointer(c, idx);
-
-	do_deletes = stripe_idx_to_delete(c) != 0;
-	mutex_unlock(&c->ec_stripes_heap_lock);
-
-	if (do_deletes)
-		bch2_do_stripe_deletes(c);
-}
-
 /* stripe deletion */
 
 static int ec_stripe_delete(struct btree_trans *trans, u64 idx)
 {
-	struct bch_fs *c = trans->c;
 	struct btree_iter iter;
-	struct bkey_s_c k;
-	struct bkey_s_c_stripe s;
-	int ret;
-
-	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_stripes, POS(0, idx),
-			       BTREE_ITER_intent);
-	ret = bkey_err(k);
+	struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter,
+					       BTREE_ID_stripes, POS(0, idx),
+					       BTREE_ITER_intent);
+	int ret = bkey_err(k);
 	if (ret)
 		goto err;
 
-	if (k.k->type != KEY_TYPE_stripe) {
-		bch2_fs_inconsistent(c, "attempting to delete nonexistent stripe %llu", idx);
-		ret = -EINVAL;
-		goto err;
-	}
-
-	s = bkey_s_c_to_stripe(k);
-	for (unsigned i = 0; i < s.v->nr_blocks; i++)
-		if (stripe_blockcount_get(s.v, i)) {
-			struct printbuf buf = PRINTBUF;
-
-			bch2_bkey_val_to_text(&buf, c, k);
-			bch2_fs_inconsistent(c, "attempting to delete nonempty stripe %s", buf.buf);
-			printbuf_exit(&buf);
-			ret = -EINVAL;
-			goto err;
-		}
-
-	ret = bch2_btree_delete_at(trans, &iter, 0);
+	/*
+	 * We expect write buffer races here
+	 * Important: check stripe_is_open with stripe key locked:
+	 */
+	if (k.k->type == KEY_TYPE_stripe &&
+	    !bch2_stripe_is_open(trans->c, idx) &&
+	    stripe_lru_pos(bkey_s_c_to_stripe(k).v) == 1)
+		ret = bch2_btree_delete_at(trans, &iter, 0);
 err:
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
 }
 
+/*
+ * XXX
+ * can we kill this and delete stripes from the trigger?
+ */
 static void ec_stripe_delete_work(struct work_struct *work)
 {
 	struct bch_fs *c =
 		container_of(work, struct bch_fs, ec_stripe_delete_work);
 
-	while (1) {
-		mutex_lock(&c->ec_stripes_heap_lock);
-		u64 idx = stripe_idx_to_delete(c);
-		mutex_unlock(&c->ec_stripes_heap_lock);
-
-		if (!idx)
-			break;
-
-		int ret = bch2_trans_commit_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
-					ec_stripe_delete(trans, idx));
-		bch_err_fn(c, ret);
-		if (ret)
-			break;
-	}
-
+	bch2_trans_run(c,
+		bch2_btree_write_buffer_tryflush(trans) ?:
+		for_each_btree_key_max_commit(trans, lru_iter, BTREE_ID_lru,
+				lru_pos(BCH_LRU_STRIPE_FRAGMENTATION, 1, 0),
+				lru_pos(BCH_LRU_STRIPE_FRAGMENTATION, 1, LRU_TIME_MAX),
+				0, lru_k,
+				NULL, NULL,
+				BCH_TRANS_COMMIT_no_enospc, ({
+			ec_stripe_delete(trans, lru_k.k->p.offset);
+		})));
 	bch2_write_ref_put(c, BCH_WRITE_REF_stripe_delete);
 }
 
@@ -1294,7 +1124,7 @@ static int ec_stripe_update_extent(struct btree_trans *trans,
 
 		bch2_fs_inconsistent(c, "%s", buf.buf);
 		printbuf_exit(&buf);
-		return -EIO;
+		return -BCH_ERR_erasure_coding_found_btree_node;
 	}
 
 	k = bch2_backpointer_get_key(trans, bp, &iter, BTREE_ITER_intent, last_flushed);
@@ -1360,7 +1190,7 @@ static int ec_stripe_update_bucket(struct btree_trans *trans, struct ec_stripe_b
 
 	struct bch_dev *ca = bch2_dev_tryget(c, ptr.dev);
 	if (!ca)
-		return -EIO;
+		return -BCH_ERR_ENOENT_dev_not_found;
 
 	struct bpos bucket_pos = PTR_BUCKET_POS(ca, &ptr);
 
@@ -1380,8 +1210,12 @@ static int ec_stripe_update_bucket(struct btree_trans *trans, struct ec_stripe_b
 		if (bp_k.k->type != KEY_TYPE_backpointer)
 			continue;
 
+		struct bkey_s_c_backpointer bp = bkey_s_c_to_backpointer(bp_k);
+		if (bp.v->btree_id == BTREE_ID_stripes)
+			continue;
+
 		ec_stripe_update_extent(trans, ca, bucket_pos, ptr.gen, s,
-					bkey_s_c_to_backpointer(bp_k), &last_flushed);
+					bp, &last_flushed);
 	}));
 
 	bch2_bkey_buf_exit(&last_flushed, c);
@@ -1393,21 +1227,19 @@ static int ec_stripe_update_extents(struct bch_fs *c, struct ec_stripe_buf *s)
 {
 	struct btree_trans *trans = bch2_trans_get(c);
 	struct bch_stripe *v = &bkey_i_to_stripe(&s->key)->v;
-	unsigned i, nr_data = v->nr_blocks - v->nr_redundant;
-	int ret = 0;
+	unsigned nr_data = v->nr_blocks - v->nr_redundant;
 
-	ret = bch2_btree_write_buffer_flush_sync(trans);
+	int ret = bch2_btree_write_buffer_flush_sync(trans);
 	if (ret)
 		goto err;
 
-	for (i = 0; i < nr_data; i++) {
+	for (unsigned i = 0; i < nr_data; i++) {
 		ret = ec_stripe_update_bucket(trans, s, i);
 		if (ret)
 			break;
 	}
 err:
 	bch2_trans_put(trans);
-
 	return ret;
 }
 
@@ -1473,6 +1305,7 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 	if (s->err) {
 		if (!bch2_err_matches(s->err, EROFS))
 			bch_err(c, "error creating stripe: error writing data buckets");
+		ret = s->err;
 		goto err;
 	}
 
@@ -1481,6 +1314,7 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 
 		if (ec_do_recov(c, &s->existing_stripe)) {
 			bch_err(c, "error creating stripe: error reading existing stripe");
+			ret = -BCH_ERR_ec_block_read;
 			goto err;
 		}
 
@@ -1506,6 +1340,7 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 
 	if (ec_nr_failed(&s->new_stripe)) {
 		bch_err(c, "error creating stripe: error writing redundancy buckets");
+		ret = -BCH_ERR_ec_block_write;
 		goto err;
 	}
 
@@ -1527,6 +1362,8 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 	if (ret)
 		goto err;
 err:
+	trace_stripe_create(c, s->idx, ret);
+
 	bch2_disk_reservation_put(c, &s->res);
 
 	for (i = 0; i < v->nr_blocks; i++)
@@ -1612,11 +1449,11 @@ static void ec_stripe_new_cancel(struct bch_fs *c, struct ec_stripe_head *h, int
 	ec_stripe_new_set_pending(c, h);
 }
 
-void bch2_ec_bucket_cancel(struct bch_fs *c, struct open_bucket *ob)
+void bch2_ec_bucket_cancel(struct bch_fs *c, struct open_bucket *ob, int err)
 {
 	struct ec_stripe_new *s = ob->ec;
 
-	s->err = -EIO;
+	s->err = err;
 }
 
 void *bch2_writepoint_ec_buf(struct bch_fs *c, struct write_point *wp)
@@ -1968,39 +1805,40 @@ static int new_stripe_alloc_buckets(struct btree_trans *trans,
 	return 0;
 }
 
-static s64 get_existing_stripe(struct bch_fs *c,
-			       struct ec_stripe_head *head)
+static int __get_existing_stripe(struct btree_trans *trans,
+				 struct ec_stripe_head *head,
+				 struct ec_stripe_buf *stripe,
+				 u64 idx)
 {
-	ec_stripes_heap *h = &c->ec_stripes_heap;
-	struct stripe *m;
-	size_t heap_idx;
-	u64 stripe_idx;
-	s64 ret = -1;
+	struct bch_fs *c = trans->c;
 
-	if (may_create_new_stripe(c))
-		return -1;
+	struct btree_iter iter;
+	struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter,
+					  BTREE_ID_stripes, POS(0, idx), 0);
+	int ret = bkey_err(k);
+	if (ret)
+		goto err;
 
-	mutex_lock(&c->ec_stripes_heap_lock);
-	for (heap_idx = 0; heap_idx < h->nr; heap_idx++) {
-		/* No blocks worth reusing, stripe will just be deleted: */
-		if (!h->data[heap_idx].blocks_nonempty)
-			continue;
+	/* We expect write buffer races here */
+	if (k.k->type != KEY_TYPE_stripe)
+		goto out;
 
-		stripe_idx = h->data[heap_idx].idx;
+	struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
+	if (stripe_lru_pos(s.v) <= 1)
+		goto out;
 
-		m = genradix_ptr(&c->stripes, stripe_idx);
-
-		if (m->disk_label	== head->disk_label &&
-		    m->algorithm	== head->algo &&
-		    m->nr_redundant	== head->redundancy &&
-		    m->sectors		== head->blocksize &&
-		    m->blocks_nonempty	< m->nr_blocks - m->nr_redundant &&
-		    bch2_try_open_stripe(c, head->s, stripe_idx)) {
-			ret = stripe_idx;
-			break;
-		}
+	if (s.v->disk_label		== head->disk_label &&
+	    s.v->algorithm		== head->algo &&
+	    s.v->nr_redundant		== head->redundancy &&
+	    le16_to_cpu(s.v->sectors)	== head->blocksize &&
+	    bch2_try_open_stripe(c, head->s, idx)) {
+		bkey_reassemble(&stripe->key, k);
+		ret = 1;
 	}
-	mutex_unlock(&c->ec_stripes_heap_lock);
+out:
+	bch2_set_btree_iter_dontneed(&iter);
+err:
+	bch2_trans_iter_exit(trans, &iter);
 	return ret;
 }
 
@@ -2052,24 +1890,33 @@ static int __bch2_ec_stripe_head_reuse(struct btree_trans *trans, struct ec_stri
 				       struct ec_stripe_new *s)
 {
 	struct bch_fs *c = trans->c;
-	s64 idx;
-	int ret;
 
 	/*
 	 * If we can't allocate a new stripe, and there's no stripes with empty
 	 * blocks for us to reuse, that means we have to wait on copygc:
 	 */
-	idx = get_existing_stripe(c, h);
-	if (idx < 0)
-		return -BCH_ERR_stripe_alloc_blocked;
+	if (may_create_new_stripe(c))
+		return -1;
 
-	ret = get_stripe_key_trans(trans, idx, &s->existing_stripe);
-	bch2_fs_fatal_err_on(ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart), c,
-			     "reading stripe key: %s", bch2_err_str(ret));
-	if (ret) {
-		bch2_stripe_close(c, s);
-		return ret;
+	struct btree_iter lru_iter;
+	struct bkey_s_c lru_k;
+	int ret = 0;
+
+	for_each_btree_key_max_norestart(trans, lru_iter, BTREE_ID_lru,
+			lru_pos(BCH_LRU_STRIPE_FRAGMENTATION, 2, 0),
+			lru_pos(BCH_LRU_STRIPE_FRAGMENTATION, 2, LRU_TIME_MAX),
+			0, lru_k, ret) {
+		ret = __get_existing_stripe(trans, h, &s->existing_stripe, lru_k.k->p.offset);
+		if (ret)
+			break;
 	}
+	bch2_trans_iter_exit(trans, &lru_iter);
+	if (!ret)
+		ret = -BCH_ERR_stripe_alloc_blocked;
+	if (ret == 1)
+		ret = 0;
+	if (ret)
+		return ret;
 
 	return init_new_stripe_from_existing(c, s);
 }
@@ -2263,14 +2110,14 @@ static int bch2_invalidate_stripe_to_dev(struct btree_trans *trans, struct bkey_
 	if (ret)
 		return ret;
 
-	struct disk_accounting_pos acc = {
-		.type = BCH_DISK_ACCOUNTING_replicas,
-	};
+	struct disk_accounting_pos acc;
 
 	s64 sectors = 0;
 	for (unsigned i = 0; i < s->v.nr_blocks; i++)
 		sectors -= stripe_blockcount_get(&s->v, i);
 
+	memset(&acc, 0, sizeof(acc));
+	acc.type = BCH_DISK_ACCOUNTING_replicas;
 	bch2_bkey_to_replicas(&acc.replicas, bkey_i_to_s_c(&s->k_i));
 	acc.replicas.data_type = BCH_DATA_user;
 	ret = bch2_disk_accounting_mod(trans, &acc, &sectors, 1, false);
@@ -2284,6 +2131,8 @@ static int bch2_invalidate_stripe_to_dev(struct btree_trans *trans, struct bkey_
 
 	sectors = -sectors;
 
+	memset(&acc, 0, sizeof(acc));
+	acc.type = BCH_DISK_ACCOUNTING_replicas;
 	bch2_bkey_to_replicas(&acc.replicas, bkey_i_to_s_c(&s->k_i));
 	acc.replicas.data_type = BCH_DATA_user;
 	ret = bch2_disk_accounting_mod(trans, &acc, &sectors, 1, false);
@@ -2367,46 +2216,7 @@ void bch2_fs_ec_flush(struct bch_fs *c)
 
 int bch2_stripes_read(struct bch_fs *c)
 {
-	int ret = bch2_trans_run(c,
-		for_each_btree_key(trans, iter, BTREE_ID_stripes, POS_MIN,
-				   BTREE_ITER_prefetch, k, ({
-			if (k.k->type != KEY_TYPE_stripe)
-				continue;
-
-			ret = __ec_stripe_mem_alloc(c, k.k->p.offset, GFP_KERNEL);
-			if (ret)
-				break;
-
-			struct stripe *m = genradix_ptr(&c->stripes, k.k->p.offset);
-
-			stripe_to_mem(m, bkey_s_c_to_stripe(k).v);
-
-			bch2_stripes_heap_insert(c, m, k.k->p.offset);
-			0;
-		})));
-	bch_err_fn(c, ret);
-	return ret;
-}
-
-void bch2_stripes_heap_to_text(struct printbuf *out, struct bch_fs *c)
-{
-	ec_stripes_heap *h = &c->ec_stripes_heap;
-	struct stripe *m;
-	size_t i;
-
-	mutex_lock(&c->ec_stripes_heap_lock);
-	for (i = 0; i < min_t(size_t, h->nr, 50); i++) {
-		m = genradix_ptr(&c->stripes, h->data[i].idx);
-
-		prt_printf(out, "%zu %u/%u+%u", h->data[i].idx,
-		       h->data[i].blocks_nonempty,
-		       m->nr_blocks - m->nr_redundant,
-		       m->nr_redundant);
-		if (bch2_stripe_is_open(c, h->data[i].idx))
-			prt_str(out, " open");
-		prt_newline(out);
-	}
-	mutex_unlock(&c->ec_stripes_heap_lock);
+	return 0;
 }
 
 static void bch2_new_stripe_to_text(struct printbuf *out, struct bch_fs *c,
@@ -2477,15 +2287,12 @@ void bch2_fs_ec_exit(struct bch_fs *c)
 
 	BUG_ON(!list_empty(&c->ec_stripe_new_list));
 
-	free_heap(&c->ec_stripes_heap);
-	genradix_free(&c->stripes);
 	bioset_exit(&c->ec_bioset);
 }
 
 void bch2_fs_ec_init_early(struct bch_fs *c)
 {
 	spin_lock_init(&c->ec_stripes_new_lock);
-	mutex_init(&c->ec_stripes_heap_lock);
 
 	INIT_LIST_HEAD(&c->ec_stripe_head_list);
 	mutex_init(&c->ec_stripe_head_lock);
@@ -2503,3 +2310,40 @@ int bch2_fs_ec_init(struct bch_fs *c)
 	return bioset_init(&c->ec_bioset, 1, offsetof(struct ec_bio, bio),
 			   BIOSET_NEED_BVECS);
 }
+
+static int bch2_check_stripe_to_lru_ref(struct btree_trans *trans,
+					struct bkey_s_c k,
+					struct bkey_buf *last_flushed)
+{
+	if (k.k->type != KEY_TYPE_stripe)
+		return 0;
+
+	struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
+
+	u64 lru_idx = stripe_lru_pos(s.v);
+	if (lru_idx) {
+		int ret = bch2_lru_check_set(trans, BCH_LRU_STRIPE_FRAGMENTATION,
+					     k.k->p.offset, lru_idx, k, last_flushed);
+		if (ret)
+			return ret;
+	}
+	return 0;
+}
+
+int bch2_check_stripe_to_lru_refs(struct bch_fs *c)
+{
+	struct bkey_buf last_flushed;
+
+	bch2_bkey_buf_init(&last_flushed);
+	bkey_init(&last_flushed.k->k);
+
+	int ret = bch2_trans_run(c,
+		for_each_btree_key_commit(trans, iter, BTREE_ID_stripes,
+				POS_MIN, BTREE_ITER_prefetch, k,
+				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+			bch2_check_stripe_to_lru_ref(trans, k, &last_flushed)));
+
+	bch2_bkey_buf_exit(&last_flushed, c);
+	bch_err_fn(c, ret);
+	return ret;
+}

libbcachefs/ec.h

@@ -92,6 +92,29 @@ static inline void stripe_csum_set(struct bch_stripe *s,
 	memcpy(stripe_csum(s, block, csum_idx), &csum, bch_crc_bytes[s->csum_type]);
 }
 
+#define STRIPE_LRU_POS_EMPTY	1
+
+static inline u64 stripe_lru_pos(const struct bch_stripe *s)
+{
+	if (!s)
+		return 0;
+
+	unsigned nr_data = s->nr_blocks - s->nr_redundant, blocks_empty = 0;
+
+	for (unsigned i = 0; i < nr_data; i++)
+		blocks_empty += !stripe_blockcount_get(s, i);
+
+	/* Will be picked up by the stripe_delete worker */
+	if (blocks_empty == nr_data)
+		return STRIPE_LRU_POS_EMPTY;
+
+	if (!blocks_empty)
+		return 0;
+
+	/* invert: more blocks empty = reuse first */
+	return LRU_TIME_MAX - blocks_empty;
+}
+
 static inline bool __bch2_ptr_matches_stripe(const struct bch_extent_ptr *stripe_ptr,
 					     const struct bch_extent_ptr *data_ptr,
 					     unsigned sectors)
@@ -132,6 +155,20 @@ static inline bool bch2_ptr_matches_stripe_m(const struct gc_stripe *m,
 					 m->sectors);
 }
 
+static inline void gc_stripe_unlock(struct gc_stripe *s)
+{
+	BUILD_BUG_ON(!((union ulong_byte_assert) { .ulong = 1UL << BUCKET_LOCK_BITNR }).byte);
+
+	clear_bit_unlock(BUCKET_LOCK_BITNR, (void *) &s->lock);
+	wake_up_bit((void *) &s->lock, BUCKET_LOCK_BITNR);
+}
+
+static inline void gc_stripe_lock(struct gc_stripe *s)
+{
+	wait_on_bit_lock((void *) &s->lock, BUCKET_LOCK_BITNR,
+			 TASK_UNINTERRUPTIBLE);
+}
+
 struct bch_read_bio;
 
 struct ec_stripe_buf {
@@ -212,7 +249,7 @@ int bch2_ec_read_extent(struct btree_trans *, struct bch_read_bio *, struct bkey
 
 void *bch2_writepoint_ec_buf(struct bch_fs *, struct write_point *);
 
-void bch2_ec_bucket_cancel(struct bch_fs *, struct open_bucket *);
+void bch2_ec_bucket_cancel(struct bch_fs *, struct open_bucket *, int);
 
 int bch2_ec_stripe_new_alloc(struct bch_fs *, struct ec_stripe_head *);
 
@@ -221,10 +258,6 @@ struct ec_stripe_head *bch2_ec_stripe_head_get(struct btree_trans *,
 			unsigned, unsigned, unsigned,
 			enum bch_watermark, struct closure *);
 
-void bch2_stripes_heap_update(struct bch_fs *, struct stripe *, size_t);
-void bch2_stripes_heap_del(struct bch_fs *, struct stripe *, size_t);
-void bch2_stripes_heap_insert(struct bch_fs *, struct stripe *, size_t);
-
 void bch2_do_stripe_deletes(struct bch_fs *);
 void bch2_ec_do_stripe_creates(struct bch_fs *);
 void bch2_ec_stripe_new_free(struct bch_fs *, struct ec_stripe_new *);
@@ -261,11 +294,12 @@ void bch2_fs_ec_flush(struct bch_fs *);
 
 int bch2_stripes_read(struct bch_fs *);
 
-void bch2_stripes_heap_to_text(struct printbuf *, struct bch_fs *);
 void bch2_new_stripes_to_text(struct printbuf *, struct bch_fs *);
 
 void bch2_fs_ec_exit(struct bch_fs *);
 void bch2_fs_ec_init_early(struct bch_fs *);
 int bch2_fs_ec_init(struct bch_fs *);
 
+int bch2_check_stripe_to_lru_refs(struct bch_fs *);
+
 #endif /* _BCACHEFS_EC_H */

libbcachefs/ec_types.h

@@ -20,23 +20,15 @@ struct stripe {
 };
 
 struct gc_stripe {
+	u8			lock;
+	unsigned		alive:1; /* does a corresponding key exist in stripes btree? */
 	u16			sectors;
-
 	u8			nr_blocks;
 	u8			nr_redundant;
-
-	unsigned		alive:1; /* does a corresponding key exist in stripes btree? */
 	u16			block_sectors[BCH_BKEY_PTRS_MAX];
 	struct bch_extent_ptr	ptrs[BCH_BKEY_PTRS_MAX];
 
 	struct bch_replicas_padded r;
 };
 
-struct ec_stripe_heap_entry {
-	size_t			idx;
-	unsigned		blocks_nonempty;
-};
-
-typedef DEFINE_MIN_HEAP(struct ec_stripe_heap_entry, ec_stripes_heap) ec_stripes_heap;
-
 #endif /* _BCACHEFS_EC_TYPES_H */

libbcachefs/errcode.h

@@ -5,6 +5,8 @@
 #define BCH_ERRCODES()								\
 	x(ERANGE,			ERANGE_option_too_small)		\
 	x(ERANGE,			ERANGE_option_too_big)			\
+	x(EINVAL,			injected)				\
+	x(BCH_ERR_injected,		injected_fs_start)			\
 	x(EINVAL,			mount_option)				\
 	x(BCH_ERR_mount_option,		option_name)				\
 	x(BCH_ERR_mount_option,		option_value)				\
@@ -116,9 +118,11 @@
 	x(ENOENT,			ENOENT_snapshot_tree)			\
 	x(ENOENT,			ENOENT_dirent_doesnt_match_inode)	\
 	x(ENOENT,			ENOENT_dev_not_found)			\
+	x(ENOENT,			ENOENT_dev_bucket_not_found)		\
 	x(ENOENT,			ENOENT_dev_idx_not_found)		\
 	x(ENOENT,			ENOENT_inode_no_backpointer)		\
 	x(ENOENT,			ENOENT_no_snapshot_tree_subvol)		\
+	x(ENOENT,			btree_node_dying)			\
 	x(ENOTEMPTY,			ENOTEMPTY_dir_not_empty)		\
 	x(ENOTEMPTY,			ENOTEMPTY_subvol_not_empty)		\
 	x(EEXIST,			EEXIST_str_hash_set)			\
@@ -185,6 +189,7 @@
 	x(BCH_ERR_data_update_done,	data_update_done_no_writes_needed)	\
 	x(BCH_ERR_data_update_done,	data_update_done_no_snapshot)		\
 	x(BCH_ERR_data_update_done,	data_update_done_no_dev_refs)		\
+	x(BCH_ERR_data_update_done,	data_update_done_no_rw_devs)		\
 	x(EINVAL,			device_state_not_allowed)		\
 	x(EINVAL,			member_info_missing)			\
 	x(EINVAL,			mismatched_block_size)			\
@@ -205,6 +210,8 @@
 	x(EINVAL,			no_resize_with_buckets_nouse)		\
 	x(EINVAL,			inode_unpack_error)			\
 	x(EINVAL,			varint_decode_error)			\
+	x(EINVAL,			erasure_coding_found_btree_node)	\
+	x(EOPNOTSUPP,			may_not_use_incompat_feature)		\
 	x(EROFS,			erofs_trans_commit)			\
 	x(EROFS,			erofs_no_writes)			\
 	x(EROFS,			erofs_journal_err)			\
@@ -215,10 +222,18 @@
 	x(EROFS,			insufficient_devices)			\
 	x(0,				operation_blocked)			\
 	x(BCH_ERR_operation_blocked,	btree_cache_cannibalize_lock_blocked)	\
-	x(BCH_ERR_operation_blocked,	journal_res_get_blocked)		\
-	x(BCH_ERR_operation_blocked,	journal_preres_get_blocked)		\
-	x(BCH_ERR_operation_blocked,	bucket_alloc_blocked)			\
-	x(BCH_ERR_operation_blocked,	stripe_alloc_blocked)			\
+	x(BCH_ERR_operation_blocked,	journal_res_blocked)			\
+	x(BCH_ERR_journal_res_blocked,	journal_blocked)			\
+	x(BCH_ERR_journal_res_blocked,	journal_max_in_flight)			\
+	x(BCH_ERR_journal_res_blocked,	journal_max_open)			\
+	x(BCH_ERR_journal_res_blocked,	journal_full)				\
+	x(BCH_ERR_journal_res_blocked,	journal_pin_full)			\
+	x(BCH_ERR_journal_res_blocked,	journal_buf_enomem)			\
+	x(BCH_ERR_journal_res_blocked,	journal_stuck)				\
+	x(BCH_ERR_journal_res_blocked,	journal_retry_open)			\
+	x(BCH_ERR_journal_res_blocked,	journal_preres_get_blocked)		\
+	x(BCH_ERR_journal_res_blocked,	bucket_alloc_blocked)			\
+	x(BCH_ERR_journal_res_blocked,	stripe_alloc_blocked)			\
 	x(BCH_ERR_invalid,		invalid_sb)				\
 	x(BCH_ERR_invalid_sb,		invalid_sb_magic)			\
 	x(BCH_ERR_invalid_sb,		invalid_sb_version)			\
@@ -228,6 +243,7 @@
 	x(BCH_ERR_invalid_sb,		invalid_sb_csum)			\
 	x(BCH_ERR_invalid_sb,		invalid_sb_block_size)			\
 	x(BCH_ERR_invalid_sb,		invalid_sb_uuid)			\
+	x(BCH_ERR_invalid_sb,		invalid_sb_offset)			\
 	x(BCH_ERR_invalid_sb,		invalid_sb_too_many_members)		\
 	x(BCH_ERR_invalid_sb,		invalid_sb_dev_idx)			\
 	x(BCH_ERR_invalid_sb,		invalid_sb_time_precision)		\
@@ -255,6 +271,7 @@
 	x(BCH_ERR_operation_blocked,    nocow_lock_blocked)			\
 	x(EIO,				journal_shutdown)			\
 	x(EIO,				journal_flush_err)			\
+	x(EIO,				journal_write_err)			\
 	x(EIO,				btree_node_read_err)			\
 	x(BCH_ERR_btree_node_read_err,	btree_node_read_err_cached)		\
 	x(EIO,				sb_not_downgraded)			\
@@ -263,18 +280,53 @@
 	x(EIO,				btree_node_read_validate_error)		\
 	x(EIO,				btree_need_topology_repair)		\
 	x(EIO,				bucket_ref_update)			\
+	x(EIO,				trigger_alloc)				\
 	x(EIO,				trigger_pointer)			\
 	x(EIO,				trigger_stripe_pointer)			\
 	x(EIO,				metadata_bucket_inconsistency)		\
 	x(EIO,				mark_stripe)				\
 	x(EIO,				stripe_reconstruct)			\
 	x(EIO,				key_type_error)				\
-	x(EIO,				no_device_to_read_from)			\
+	x(EIO,				extent_poisened)			\
 	x(EIO,				missing_indirect_extent)		\
 	x(EIO,				invalidate_stripe_to_dev)		\
 	x(EIO,				no_encryption_key)			\
 	x(EIO,				insufficient_journal_devices)		\
 	x(EIO,				device_offline)				\
+	x(EIO,				EIO_fault_injected)			\
+	x(EIO,				ec_block_read)				\
+	x(EIO,				ec_block_write)				\
+	x(EIO,				recompute_checksum)			\
+	x(EIO,				decompress)				\
+	x(BCH_ERR_decompress,		decompress_exceeded_max_encoded_extent)	\
+	x(BCH_ERR_decompress,		decompress_lz4)				\
+	x(BCH_ERR_decompress,		decompress_gzip)			\
+	x(BCH_ERR_decompress,		decompress_zstd_src_len_bad)		\
+	x(BCH_ERR_decompress,		decompress_zstd)			\
+	x(EIO,				data_write)				\
+	x(BCH_ERR_data_write,		data_write_io)				\
+	x(BCH_ERR_data_write,		data_write_csum)			\
+	x(BCH_ERR_data_write,		data_write_invalid_ptr)			\
+	x(BCH_ERR_data_write,		data_write_misaligned)			\
+	x(BCH_ERR_decompress,		data_read)				\
+	x(BCH_ERR_data_read,		no_device_to_read_from)			\
+	x(BCH_ERR_data_read,		no_devices_valid)			\
+	x(BCH_ERR_data_read,		data_read_io_err)			\
+	x(BCH_ERR_data_read,		data_read_csum_err)			\
+	x(BCH_ERR_data_read,		data_read_retry)			\
+	x(BCH_ERR_data_read_retry,	data_read_retry_avoid)			\
+	x(BCH_ERR_data_read_retry_avoid,data_read_retry_device_offline)		\
+	x(BCH_ERR_data_read_retry_avoid,data_read_retry_io_err)			\
+	x(BCH_ERR_data_read_retry_avoid,data_read_retry_ec_reconstruct_err)	\
+	x(BCH_ERR_data_read_retry_avoid,data_read_retry_csum_err)		\
+	x(BCH_ERR_data_read_retry,	data_read_retry_csum_err_maybe_userspace)\
+	x(BCH_ERR_data_read,		data_read_decompress_err)		\
+	x(BCH_ERR_data_read,		data_read_decrypt_err)			\
+	x(BCH_ERR_data_read,		data_read_ptr_stale_race)		\
+	x(BCH_ERR_data_read_retry,	data_read_ptr_stale_retry)		\
+	x(BCH_ERR_data_read,		data_read_no_encryption_key)		\
+	x(BCH_ERR_data_read,		data_read_buffer_too_small)		\
+	x(BCH_ERR_data_read,		data_read_key_overwritten)		\
 	x(BCH_ERR_btree_node_read_err,	btree_node_read_err_fixable)		\
 	x(BCH_ERR_btree_node_read_err,	btree_node_read_err_want_retry)		\
 	x(BCH_ERR_btree_node_read_err,	btree_node_read_err_must_retry)		\

libbcachefs/error.c

@@ -3,15 +3,24 @@
 #include "btree_cache.h"
 #include "btree_iter.h"
 #include "error.h"
-#include "fs-common.h"
 #include "journal.h"
+#include "namei.h"
 #include "recovery_passes.h"
 #include "super.h"
 #include "thread_with_file.h"
 
 #define FSCK_ERR_RATELIMIT_NR	10
 
-bool bch2_inconsistent_error(struct bch_fs *c)
+void bch2_log_msg_start(struct bch_fs *c, struct printbuf *out)
+{
+	printbuf_indent_add_nextline(out, 2);
+
+#ifdef BCACHEFS_LOG_PREFIX
+	prt_printf(out, bch2_log_msg(c, ""));
+#endif
+}
+
+bool __bch2_inconsistent_error(struct bch_fs *c, struct printbuf *out)
 {
 	set_bit(BCH_FS_error, &c->flags);
 
@@ -21,10 +30,11 @@ bool bch2_inconsistent_error(struct bch_fs *c)
 	case BCH_ON_ERROR_fix_safe:
 	case BCH_ON_ERROR_ro:
 		if (bch2_fs_emergency_read_only(c))
-			bch_err(c, "inconsistency detected - emergency read only at journal seq %llu",
-				journal_cur_seq(&c->journal));
+			prt_printf(out, "inconsistency detected - emergency read only at journal seq %llu\n",
+				   journal_cur_seq(&c->journal));
 		return true;
 	case BCH_ON_ERROR_panic:
+		bch2_print_string_as_lines(KERN_ERR, out->buf);
 		panic(bch2_fmt(c, "panic after error"));
 		return true;
 	default:
@@ -32,11 +42,63 @@ bool bch2_inconsistent_error(struct bch_fs *c)
 	}
 }
 
-int bch2_topology_error(struct bch_fs *c)
+bool bch2_inconsistent_error(struct bch_fs *c)
 {
+	struct printbuf buf = PRINTBUF;
+	printbuf_indent_add_nextline(&buf, 2);
+
+	bool ret = __bch2_inconsistent_error(c, &buf);
+	if (ret)
+		bch_err(c, "%s", buf.buf);
+	printbuf_exit(&buf);
+	return ret;
+}
+
+__printf(3, 0)
+static bool bch2_fs_trans_inconsistent(struct bch_fs *c, struct btree_trans *trans,
+				       const char *fmt, va_list args)
+{
+	struct printbuf buf = PRINTBUF;
+
+	bch2_log_msg_start(c, &buf);
+
+	prt_vprintf(&buf, fmt, args);
+	prt_newline(&buf);
+
+	if (trans)
+		bch2_trans_updates_to_text(&buf, trans);
+	bool ret = __bch2_inconsistent_error(c, &buf);
+	bch2_print_string_as_lines(KERN_ERR, buf.buf);
+
+	printbuf_exit(&buf);
+	return ret;
+}
+
+bool bch2_fs_inconsistent(struct bch_fs *c, const char *fmt, ...)
+{
+	va_list args;
+	va_start(args, fmt);
+	bool ret = bch2_fs_trans_inconsistent(c, NULL, fmt, args);
+	va_end(args);
+	return ret;
+}
+
+bool bch2_trans_inconsistent(struct btree_trans *trans, const char *fmt, ...)
+{
+	va_list args;
+	va_start(args, fmt);
+	bool ret = bch2_fs_trans_inconsistent(trans->c, trans, fmt, args);
+	va_end(args);
+	return ret;
+}
+
+int __bch2_topology_error(struct bch_fs *c, struct printbuf *out)
+{
+	prt_printf(out, "btree topology error: ");
+
 	set_bit(BCH_FS_topology_error, &c->flags);
 	if (!test_bit(BCH_FS_recovery_running, &c->flags)) {
-		bch2_inconsistent_error(c);
+		__bch2_inconsistent_error(c, out);
 		return -BCH_ERR_btree_need_topology_repair;
 	} else {
 		return bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology) ?:
@@ -44,6 +106,24 @@ int bch2_topology_error(struct bch_fs *c)
 	}
 }
 
+int bch2_fs_topology_error(struct bch_fs *c, const char *fmt, ...)
+{
+	struct printbuf buf = PRINTBUF;
+
+	bch2_log_msg_start(c, &buf);
+
+	va_list args;
+	va_start(args, fmt);
+	prt_vprintf(&buf, fmt, args);
+	va_end(args);
+
+	int ret = __bch2_topology_error(c, &buf);
+	bch2_print_string_as_lines(KERN_ERR, buf.buf);
+
+	printbuf_exit(&buf);
+	return ret;
+}
+
 void bch2_fatal_error(struct bch_fs *c)
 {
 	if (bch2_fs_emergency_read_only(c))
@@ -54,25 +134,41 @@ void bch2_io_error_work(struct work_struct *work)
 {
 	struct bch_dev *ca = container_of(work, struct bch_dev, io_error_work);
 	struct bch_fs *c = ca->fs;
-	bool dev;
+
+	/* XXX: if it's reads or checksums that are failing, set it to failed */
 
 	down_write(&c->state_lock);
-	dev = bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_ro,
-				    BCH_FORCE_IF_DEGRADED);
-	if (dev
-	    ? __bch2_dev_set_state(c, ca, BCH_MEMBER_STATE_ro,
-				  BCH_FORCE_IF_DEGRADED)
-	    : bch2_fs_emergency_read_only(c))
+	unsigned long write_errors_start = READ_ONCE(ca->write_errors_start);
+
+	if (write_errors_start &&
+	    time_after(jiffies,
+		       write_errors_start + c->opts.write_error_timeout * HZ)) {
+		if (ca->mi.state >= BCH_MEMBER_STATE_ro)
+			goto out;
+
+		bool dev = !__bch2_dev_set_state(c, ca, BCH_MEMBER_STATE_ro,
+						 BCH_FORCE_IF_DEGRADED);
+
 		bch_err(ca,
-			"too many IO errors, setting %s RO",
+			"writes erroring for %u seconds, setting %s ro",
+			c->opts.write_error_timeout,
 			dev ? "device" : "filesystem");
+		if (!dev)
+			bch2_fs_emergency_read_only(c);
+
+	}
+out:
 	up_write(&c->state_lock);
 }
 
 void bch2_io_error(struct bch_dev *ca, enum bch_member_error_type type)
 {
 	atomic64_inc(&ca->errors[type]);
-	//queue_work(system_long_wq, &ca->io_error_work);
+
+	if (type == BCH_MEMBER_ERROR_write && !ca->write_errors_start)
+		ca->write_errors_start = jiffies;
+
+	queue_work(system_long_wq, &ca->io_error_work);
 }
 
 enum ask_yn {
@@ -168,7 +264,8 @@ static enum ask_yn bch2_fsck_ask_yn(struct bch_fs *c, struct btree_trans *trans)
 
 #endif
 
-static struct fsck_err_state *fsck_err_get(struct bch_fs *c, const char *fmt)
+static struct fsck_err_state *fsck_err_get(struct bch_fs *c,
+					   enum bch_sb_error_id id)
 {
 	struct fsck_err_state *s;
 
@@ -176,7 +273,7 @@ static struct fsck_err_state *fsck_err_get(struct bch_fs *c, const char *fmt)
 		return NULL;
 
 	list_for_each_entry(s, &c->fsck_error_msgs, list)
-		if (s->fmt == fmt) {
+		if (s->id == id) {
 			/*
 			 * move it to the head of the list: repeated fsck errors
 			 * are common
@@ -194,7 +291,7 @@ static struct fsck_err_state *fsck_err_get(struct bch_fs *c, const char *fmt)
 	}
 
 	INIT_LIST_HEAD(&s->list);
-	s->fmt = fmt;
+	s->id = id;
 	list_add(&s->list, &c->fsck_error_msgs);
 	return s;
 }
@@ -244,15 +341,59 @@ static int do_fsck_ask_yn(struct bch_fs *c,
 	return ask;
 }
 
+static struct fsck_err_state *count_fsck_err_locked(struct bch_fs *c,
+			  enum bch_sb_error_id id, const char *msg,
+			  bool *repeat, bool *print, bool *suppress)
+{
+	bch2_sb_error_count(c, id);
+
+	struct fsck_err_state *s = fsck_err_get(c, id);
+	if (s) {
+		/*
+		 * We may be called multiple times for the same error on
+		 * transaction restart - this memoizes instead of asking the user
+		 * multiple times for the same error:
+		 */
+		if (s->last_msg && !strcmp(msg, s->last_msg)) {
+			*repeat = true;
+			*print = false;
+			return s;
+		}
+
+		kfree(s->last_msg);
+		s->last_msg = kstrdup(msg, GFP_KERNEL);
+
+		if (c->opts.ratelimit_errors &&
+		    s->nr >= FSCK_ERR_RATELIMIT_NR) {
+			if (s->nr == FSCK_ERR_RATELIMIT_NR)
+				*suppress = true;
+			else
+				*print = false;
+		}
+
+		s->nr++;
+	}
+	return s;
+}
+
+void __bch2_count_fsck_err(struct bch_fs *c,
+			   enum bch_sb_error_id id, const char *msg,
+			   bool *repeat, bool *print, bool *suppress)
+{
+	bch2_sb_error_count(c, id);
+
+	mutex_lock(&c->fsck_error_msgs_lock);
+	count_fsck_err_locked(c, id, msg, repeat, print, suppress);
+	mutex_unlock(&c->fsck_error_msgs_lock);
+}
+
 int __bch2_fsck_err(struct bch_fs *c,
 		  struct btree_trans *trans,
 		  enum bch_fsck_flags flags,
 		  enum bch_sb_error_id err,
 		  const char *fmt, ...)
 {
-	struct fsck_err_state *s = NULL;
 	va_list args;
-	bool print = true, suppressing = false, inconsistent = false, exiting = false;
 	struct printbuf buf = PRINTBUF, *out = &buf;
 	int ret = -BCH_ERR_fsck_ignore;
 	const char *action_orig = "fix?", *action = action_orig;
@@ -287,7 +428,12 @@ int __bch2_fsck_err(struct bch_fs *c,
 			? -BCH_ERR_fsck_fix
 			: -BCH_ERR_fsck_ignore;
 
-	bch2_sb_error_count(c, err);
+	printbuf_indent_add_nextline(out, 2);
+
+#ifdef BCACHEFS_LOG_PREFIX
+	if (strncmp(fmt, "bcachefs", 8))
+		prt_printf(out, bch2_log_msg(c, ""));
+#endif
 
 	va_start(args, fmt);
 	prt_vprintf(out, fmt, args);
@@ -307,42 +453,15 @@ int __bch2_fsck_err(struct bch_fs *c,
 	}
 
 	mutex_lock(&c->fsck_error_msgs_lock);
-	s = fsck_err_get(c, fmt);
-	if (s) {
-		/*
-		 * We may be called multiple times for the same error on
-		 * transaction restart - this memoizes instead of asking the user
-		 * multiple times for the same error:
-		 */
-		if (s->last_msg && !strcmp(buf.buf, s->last_msg)) {
-			ret = s->ret;
-			goto err_unlock;
-		}
-
-		kfree(s->last_msg);
-		s->last_msg = kstrdup(buf.buf, GFP_KERNEL);
-		if (!s->last_msg) {
-			ret = -ENOMEM;
-			goto err_unlock;
-		}
-
-		if (c->opts.ratelimit_errors &&
-		    !(flags & FSCK_NO_RATELIMIT) &&
-		    s->nr >= FSCK_ERR_RATELIMIT_NR) {
-			if (s->nr == FSCK_ERR_RATELIMIT_NR)
-				suppressing = true;
-			else
-				print = false;
-		}
-
-		s->nr++;
+	bool repeat = false, print = true, suppress = false;
+	bool inconsistent = false, exiting = false;
+	struct fsck_err_state *s =
+		count_fsck_err_locked(c, err, buf.buf, &repeat, &print, &suppress);
+	if (repeat) {
+		ret = s->ret;
+		goto err_unlock;
 	}
 
-#ifdef BCACHEFS_LOG_PREFIX
-	if (!strncmp(fmt, "bcachefs:", 9))
-		prt_printf(out, bch2_log_msg(c, ""));
-#endif
-
 	if ((flags & FSCK_AUTOFIX) &&
 	    (c->opts.errors == BCH_ON_ERROR_continue ||
 	     c->opts.errors == BCH_ON_ERROR_fix_safe)) {
@@ -361,6 +480,7 @@ int __bch2_fsck_err(struct bch_fs *c,
 		    !(flags & (FSCK_CAN_FIX|FSCK_CAN_IGNORE))) {
 			prt_str(out, ", shutting down");
 			inconsistent = true;
+			print = true;
 			ret = -BCH_ERR_fsck_errors_not_fixed;
 		} else if (flags & FSCK_CAN_FIX) {
 			prt_str(out, ", ");
@@ -419,24 +539,30 @@ int __bch2_fsck_err(struct bch_fs *c,
 		print = true;
 	}
 print:
+	prt_newline(out);
+
+	if (inconsistent)
+		__bch2_inconsistent_error(c, out);
+	else if (exiting)
+		prt_printf(out, "Unable to continue, halting\n");
+	else if (suppress)
+		prt_printf(out, "Ratelimiting new instances of previous error\n");
+
 	if (print) {
+		/* possibly strip an empty line, from printbuf_indent_add */
+		while (out->pos && out->buf[out->pos - 1] == ' ')
+			--out->pos;
+		printbuf_nul_terminate(out);
+
 		if (bch2_fs_stdio_redirect(c))
-			bch2_print(c, "%s\n", out->buf);
+			bch2_print(c, "%s", out->buf);
 		else
 			bch2_print_string_as_lines(KERN_ERR, out->buf);
 	}
 
-	if (exiting)
-		bch_err(c, "Unable to continue, halting");
-	else if (suppressing)
-		bch_err(c, "Ratelimiting new instances of previous error");
-
 	if (s)
 		s->ret = ret;
 
-	if (inconsistent)
-		bch2_inconsistent_error(c);
-
 	/*
 	 * We don't yet track whether the filesystem currently has errors, for
 	 * log_fsck_err()s: that would require us to track for every error type
@@ -498,16 +624,14 @@ int __bch2_bkey_fsck_err(struct bch_fs *c,
 	prt_printf(&buf, " level=%u: ", from.level);
 
 	bch2_bkey_val_to_text(&buf, c, k);
-	prt_str(&buf, "\n  ");
+	prt_newline(&buf);
 
 	va_list args;
 	va_start(args, fmt);
 	prt_vprintf(&buf, fmt, args);
 	va_end(args);
 
-	prt_str(&buf, ": delete?");
-
-	int ret = __bch2_fsck_err(c, NULL, fsck_flags, err, "%s", buf.buf);
+	int ret = __bch2_fsck_err(c, NULL, fsck_flags, err, "%s, delete?", buf.buf);
 	printbuf_exit(&buf);
 	return ret;
 }
@@ -520,7 +644,7 @@ void bch2_flush_fsck_errs(struct bch_fs *c)
 
 	list_for_each_entry_safe(s, n, &c->fsck_error_msgs, list) {
 		if (s->ratelimited && s->last_msg)
-			bch_err(c, "Saw %llu errors like:\n    %s", s->nr, s->last_msg);
+			bch_err(c, "Saw %llu errors like:\n  %s", s->nr, s->last_msg);
 
 		list_del(&s->list);
 		kfree(s->last_msg);
@@ -530,35 +654,59 @@ void bch2_flush_fsck_errs(struct bch_fs *c)
 	mutex_unlock(&c->fsck_error_msgs_lock);
 }
 
-int bch2_inum_err_msg_trans(struct btree_trans *trans, struct printbuf *out, subvol_inum inum)
+int bch2_inum_offset_err_msg_trans(struct btree_trans *trans, struct printbuf *out,
+				    subvol_inum inum, u64 offset)
 {
 	u32 restart_count = trans->restart_count;
 	int ret = 0;
 
-	/* XXX: we don't yet attempt to print paths when we don't know the subvol */
-	if (inum.subvol)
-		ret = lockrestart_do(trans, bch2_inum_to_path(trans, inum, out));
+	if (inum.subvol) {
+		ret = bch2_inum_to_path(trans, inum, out);
+		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+			return ret;
+	}
 	if (!inum.subvol || ret)
 		prt_printf(out, "inum %llu:%llu", inum.subvol, inum.inum);
+	prt_printf(out, " offset %llu: ", offset);
 
 	return trans_was_restarted(trans, restart_count);
 }
 
-int bch2_inum_offset_err_msg_trans(struct btree_trans *trans, struct printbuf *out,
-				    subvol_inum inum, u64 offset)
-{
-	int ret = bch2_inum_err_msg_trans(trans, out, inum);
-	prt_printf(out, " offset %llu: ", offset);
-	return ret;
-}
-
-void bch2_inum_err_msg(struct bch_fs *c, struct printbuf *out, subvol_inum inum)
-{
-	bch2_trans_run(c, bch2_inum_err_msg_trans(trans, out, inum));
-}
-
 void bch2_inum_offset_err_msg(struct bch_fs *c, struct printbuf *out,
 			      subvol_inum inum, u64 offset)
 {
-	bch2_trans_run(c, bch2_inum_offset_err_msg_trans(trans, out, inum, offset));
+	bch2_trans_do(c, bch2_inum_offset_err_msg_trans(trans, out, inum, offset));
+}
+
+int bch2_inum_snap_offset_err_msg_trans(struct btree_trans *trans, struct printbuf *out,
+					struct bpos pos)
+{
+	struct bch_fs *c = trans->c;
+	int ret = 0;
+
+	if (!bch2_snapshot_is_leaf(c, pos.snapshot))
+		prt_str(out, "(multiple snapshots) ");
+
+	subvol_inum inum = {
+		.subvol	= bch2_snapshot_tree_oldest_subvol(c, pos.snapshot),
+		.inum	= pos.inode,
+	};
+
+	if (inum.subvol) {
+		ret = bch2_inum_to_path(trans, inum, out);
+		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+			return ret;
+	}
+
+	if (!inum.subvol || ret)
+		prt_printf(out, "inum %llu:%u", pos.inode, pos.snapshot);
+
+	prt_printf(out, " offset %llu: ", pos.offset << 8);
+	return 0;
+}
+
+void bch2_inum_snap_offset_err_msg(struct bch_fs *c, struct printbuf *out,
+				  struct bpos pos)
+{
+	bch2_trans_do(c, bch2_inum_snap_offset_err_msg_trans(trans, out, pos));
 }

libbcachefs/error.h

@@ -18,6 +18,8 @@ struct work_struct;
 
 /* Error messages: */
 
+void bch2_log_msg_start(struct bch_fs *, struct printbuf *);
+
 /*
  * Inconsistency errors: The on disk data is inconsistent. If these occur during
  * initial recovery, they don't indicate a bug in the running code - we walk all
@@ -29,21 +31,10 @@ struct work_struct;
  * BCH_ON_ERROR_CONTINUE mode
  */
 
+bool __bch2_inconsistent_error(struct bch_fs *, struct printbuf *);
 bool bch2_inconsistent_error(struct bch_fs *);
-
-int bch2_topology_error(struct bch_fs *);
-
-#define bch2_fs_topology_error(c, ...)					\
-({									\
-	bch_err(c, "btree topology error: " __VA_ARGS__);		\
-	bch2_topology_error(c);						\
-})
-
-#define bch2_fs_inconsistent(c, ...)					\
-({									\
-	bch_err(c, __VA_ARGS__);					\
-	bch2_inconsistent_error(c);					\
-})
+__printf(2, 3)
+bool bch2_fs_inconsistent(struct bch_fs *, const char *, ...);
 
 #define bch2_fs_inconsistent_on(cond, ...)				\
 ({									\
@@ -53,26 +44,21 @@ int bch2_topology_error(struct bch_fs *);
 	_ret;								\
 })
 
-/*
- * When a transaction update discovers or is causing a fs inconsistency, it's
- * helpful to also dump the pending updates:
- */
-#define bch2_trans_inconsistent(trans, ...)				\
-({									\
-	bch_err(trans->c, __VA_ARGS__);					\
-	bch2_dump_trans_updates(trans);					\
-	bch2_inconsistent_error(trans->c);				\
-})
+__printf(2, 3)
+bool bch2_trans_inconsistent(struct btree_trans *, const char *, ...);
 
-#define bch2_trans_inconsistent_on(cond, trans, ...)			\
+#define bch2_trans_inconsistent_on(cond, ...)				\
 ({									\
 	bool _ret = unlikely(!!(cond));					\
-									\
 	if (_ret)							\
-		bch2_trans_inconsistent(trans, __VA_ARGS__);		\
+		bch2_trans_inconsistent(__VA_ARGS__);			\
 	_ret;								\
 })
 
+int __bch2_topology_error(struct bch_fs *, struct printbuf *);
+__printf(2, 3)
+int bch2_fs_topology_error(struct bch_fs *, const char *, ...);
+
 /*
  * Fsck errors: inconsistency errors we detect at mount time, and should ideally
  * be able to repair:
@@ -80,7 +66,7 @@ int bch2_topology_error(struct bch_fs *);
 
 struct fsck_err_state {
 	struct list_head	list;
-	const char		*fmt;
+	enum bch_sb_error_id	id;
 	u64			nr;
 	bool			ratelimited;
 	int			ret;
@@ -90,6 +76,12 @@ struct fsck_err_state {
 
 #define fsck_err_count(_c, _err)	bch2_sb_err_count(_c, BCH_FSCK_ERR_##_err)
 
+void __bch2_count_fsck_err(struct bch_fs *,
+			   enum bch_sb_error_id, const char *,
+			   bool *, bool *, bool *);
+#define bch2_count_fsck_err(_c, _err, ...)				\
+	__bch2_count_fsck_err(_c, BCH_FSCK_ERR_##_err, __VA_ARGS__)
+
 __printf(5, 6) __cold
 int __bch2_fsck_err(struct bch_fs *, struct btree_trans *,
 		  enum bch_fsck_flags,
@@ -216,32 +208,43 @@ void bch2_io_error_work(struct work_struct *);
 /* Does the error handling without logging a message */
 void bch2_io_error(struct bch_dev *, enum bch_member_error_type);
 
-#define bch2_dev_io_err_on(cond, ca, _type, ...)			\
-({									\
-	bool _ret = (cond);						\
-									\
-	if (_ret) {							\
-		bch_err_dev_ratelimited(ca, __VA_ARGS__);		\
-		bch2_io_error(ca, _type);				\
-	}								\
-	_ret;								\
-})
+#ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
+void bch2_latency_acct(struct bch_dev *, u64, int);
+#else
+static inline void bch2_latency_acct(struct bch_dev *ca, u64 submit_time, int rw) {}
+#endif
 
-#define bch2_dev_inum_io_err_on(cond, ca, _type, ...)			\
-({									\
-	bool _ret = (cond);						\
-									\
-	if (_ret) {							\
-		bch_err_inum_offset_ratelimited(ca, __VA_ARGS__);	\
-		bch2_io_error(ca, _type);				\
-	}								\
-	_ret;								\
-})
+static inline void bch2_account_io_success_fail(struct bch_dev *ca,
+						enum bch_member_error_type type,
+						bool success)
+{
+	if (likely(success)) {
+		if (type == BCH_MEMBER_ERROR_write &&
+		    ca->write_errors_start)
+			ca->write_errors_start = 0;
+	} else {
+		bch2_io_error(ca, type);
+	}
+}
+
+static inline void bch2_account_io_completion(struct bch_dev *ca,
+					      enum bch_member_error_type type,
+					      u64 submit_time, bool success)
+{
+	if (unlikely(!ca))
+		return;
+
+	if (type != BCH_MEMBER_ERROR_checksum)
+		bch2_latency_acct(ca, submit_time, type);
+
+	bch2_account_io_success_fail(ca, type, success);
+}
 
-int bch2_inum_err_msg_trans(struct btree_trans *, struct printbuf *, subvol_inum);
 int bch2_inum_offset_err_msg_trans(struct btree_trans *, struct printbuf *, subvol_inum, u64);
 
-void bch2_inum_err_msg(struct bch_fs *, struct printbuf *, subvol_inum);
 void bch2_inum_offset_err_msg(struct bch_fs *, struct printbuf *, subvol_inum, u64);
 
+int bch2_inum_snap_offset_err_msg_trans(struct btree_trans *, struct printbuf *, struct bpos);
+void bch2_inum_snap_offset_err_msg(struct bch_fs *, struct printbuf *, struct bpos);
+
 #endif /* _BCACHEFS_ERROR_H */

libbcachefs/extents.c

@@ -28,6 +28,13 @@
 #include "trace.h"
 #include "util.h"
 
+static const char * const bch2_extent_flags_strs[] = {
+#define x(n, v)	[BCH_EXTENT_FLAG_##n] = #n,
+	BCH_EXTENT_FLAGS()
+#undef x
+	NULL,
+};
+
 static unsigned bch2_crc_field_size_max[] = {
 	[BCH_EXTENT_ENTRY_crc32] = CRC32_SIZE_MAX,
 	[BCH_EXTENT_ENTRY_crc64] = CRC64_SIZE_MAX,
@@ -51,7 +58,8 @@ struct bch_dev_io_failures *bch2_dev_io_failures(struct bch_io_failures *f,
 }
 
 void bch2_mark_io_failure(struct bch_io_failures *failed,
-			  struct extent_ptr_decoded *p)
+			  struct extent_ptr_decoded *p,
+			  bool csum_error)
 {
 	struct bch_dev_io_failures *f = bch2_dev_io_failures(failed, p->ptr.dev);
 
@@ -59,53 +67,57 @@ void bch2_mark_io_failure(struct bch_io_failures *failed,
 		BUG_ON(failed->nr >= ARRAY_SIZE(failed->devs));
 
 		f = &failed->devs[failed->nr++];
-		f->dev		= p->ptr.dev;
-		f->idx		= p->idx;
-		f->nr_failed	= 1;
-		f->nr_retries	= 0;
-	} else if (p->idx != f->idx) {
-		f->idx		= p->idx;
-		f->nr_failed	= 1;
-		f->nr_retries	= 0;
-	} else {
-		f->nr_failed++;
+		memset(f, 0, sizeof(*f));
+		f->dev = p->ptr.dev;
 	}
+
+	if (p->do_ec_reconstruct)
+		f->failed_ec = true;
+	else if (!csum_error)
+		f->failed_io = true;
+	else
+		f->failed_csum_nr++;
 }
 
-static inline u64 dev_latency(struct bch_fs *c, unsigned dev)
+static inline u64 dev_latency(struct bch_dev *ca)
 {
-	struct bch_dev *ca = bch2_dev_rcu(c, dev);
 	return ca ? atomic64_read(&ca->cur_latency[READ]) : S64_MAX;
 }
 
+static inline int dev_failed(struct bch_dev *ca)
+{
+	return !ca || ca->mi.state == BCH_MEMBER_STATE_failed;
+}
+
 /*
  * returns true if p1 is better than p2:
  */
 static inline bool ptr_better(struct bch_fs *c,
 			      const struct extent_ptr_decoded p1,
-			      const struct extent_ptr_decoded p2)
+			      u64 p1_latency,
+			      struct bch_dev *ca1,
+			      const struct extent_ptr_decoded p2,
+			      u64 p2_latency)
 {
-	if (likely(!p1.idx && !p2.idx)) {
-		u64 l1 = dev_latency(c, p1.ptr.dev);
-		u64 l2 = dev_latency(c, p2.ptr.dev);
+	struct bch_dev *ca2 = bch2_dev_rcu(c, p2.ptr.dev);
 
-		/*
-		 * Square the latencies, to bias more in favor of the faster
-		 * device - we never want to stop issuing reads to the slower
-		 * device altogether, so that we can update our latency numbers:
-		 */
-		l1 *= l1;
-		l2 *= l2;
+	int failed_delta = dev_failed(ca1) - dev_failed(ca2);
+	if (unlikely(failed_delta))
+		return failed_delta < 0;
 
-		/* Pick at random, biased in favor of the faster device: */
+	if (unlikely(bch2_force_reconstruct_read))
+		return p1.do_ec_reconstruct > p2.do_ec_reconstruct;
 
-		return bch2_rand_range(l1 + l2) > l1;
-	}
+	if (unlikely(p1.do_ec_reconstruct || p2.do_ec_reconstruct))
+		return p1.do_ec_reconstruct < p2.do_ec_reconstruct;
 
-	if (bch2_force_reconstruct_read)
-		return p1.idx > p2.idx;
+	int crc_retry_delta = (int) p1.crc_retry_nr - (int) p2.crc_retry_nr;
+	if (unlikely(crc_retry_delta))
+		return crc_retry_delta < 0;
 
-	return p1.idx < p2.idx;
+	/* Pick at random, biased in favor of the faster device: */
+
+	return bch2_get_random_u64_below(p1_latency + p2_latency) > p1_latency;
 }
 
 /*
@@ -114,70 +126,112 @@ static inline bool ptr_better(struct bch_fs *c,
  * other devices, it will still pick a pointer from avoid.
  */
 int bch2_bkey_pick_read_device(struct bch_fs *c, struct bkey_s_c k,
-			      struct bch_io_failures *failed,
-			      struct extent_ptr_decoded *pick,
-			      int dev)
+			       struct bch_io_failures *failed,
+			       struct extent_ptr_decoded *pick,
+			       int dev)
 {
-	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
-	const union bch_extent_entry *entry;
-	struct extent_ptr_decoded p;
-	struct bch_dev_io_failures *f;
-	int ret = 0;
+	bool have_csum_errors = false, have_io_errors = false, have_missing_devs = false;
+	bool have_dirty_ptrs = false, have_pick = false;
 
 	if (k.k->type == KEY_TYPE_error)
 		return -BCH_ERR_key_type_error;
 
+	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+
+	if (bch2_bkey_extent_ptrs_flags(ptrs) & BIT_ULL(BCH_EXTENT_FLAG_poisoned))
+		return -BCH_ERR_extent_poisened;
+
 	rcu_read_lock();
+	const union bch_extent_entry *entry;
+	struct extent_ptr_decoded p;
+	u64 pick_latency;
+
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
+		have_dirty_ptrs |= !p.ptr.cached;
+
 		/*
 		 * Unwritten extent: no need to actually read, treat it as a
 		 * hole and return 0s:
 		 */
 		if (p.ptr.unwritten) {
-			ret = 0;
-			break;
+			rcu_read_unlock();
+			return 0;
 		}
 
 		/* Are we being asked to read from a specific device? */
 		if (dev >= 0 && p.ptr.dev != dev)
 			continue;
 
-		/*
-		 * If there are any dirty pointers it's an error if we can't
-		 * read:
-		 */
-		if (!ret && !p.ptr.cached)
-			ret = -BCH_ERR_no_device_to_read_from;
-
 		struct bch_dev *ca = bch2_dev_rcu(c, p.ptr.dev);
 
 		if (p.ptr.cached && (!ca || dev_ptr_stale_rcu(ca, &p.ptr)))
 			continue;
 
-		f = failed ? bch2_dev_io_failures(failed, p.ptr.dev) : NULL;
-		if (f)
-			p.idx = f->nr_failed < f->nr_retries
-				? f->idx
-				: f->idx + 1;
+		struct bch_dev_io_failures *f =
+			unlikely(failed) ? bch2_dev_io_failures(failed, p.ptr.dev) : NULL;
+		if (unlikely(f)) {
+			p.crc_retry_nr	   = f->failed_csum_nr;
+			p.has_ec	  &= ~f->failed_ec;
 
-		if (!p.idx && (!ca || !bch2_dev_is_readable(ca)))
-			p.idx++;
+			if (ca && ca->mi.state != BCH_MEMBER_STATE_failed) {
+				have_io_errors	|= f->failed_io;
+				have_io_errors	|= f->failed_ec;
+			}
+			have_csum_errors	|= !!f->failed_csum_nr;
 
-		if (!p.idx && p.has_ec && bch2_force_reconstruct_read)
-			p.idx++;
+			if (p.has_ec && (f->failed_io || f->failed_csum_nr))
+				p.do_ec_reconstruct = true;
+			else if (f->failed_io ||
+				 f->failed_csum_nr > c->opts.checksum_err_retry_nr)
+				continue;
+		}
 
-		if (p.idx > (unsigned) p.has_ec)
-			continue;
+		have_missing_devs |= ca && !bch2_dev_is_online(ca);
 
-		if (ret > 0 && !ptr_better(c, p, *pick))
-			continue;
+		if (!ca || !bch2_dev_is_online(ca)) {
+			if (!p.has_ec)
+				continue;
+			p.do_ec_reconstruct = true;
+		}
 
-		*pick = p;
-		ret = 1;
+		if (bch2_force_reconstruct_read && p.has_ec)
+			p.do_ec_reconstruct = true;
+
+		u64 p_latency = dev_latency(ca);
+		/*
+		 * Square the latencies, to bias more in favor of the faster
+		 * device - we never want to stop issuing reads to the slower
+		 * device altogether, so that we can update our latency numbers:
+		 */
+		p_latency *= p_latency;
+
+		if (!have_pick ||
+		    ptr_better(c,
+			       p, p_latency, ca,
+			       *pick, pick_latency)) {
+			*pick = p;
+			pick_latency = p_latency;
+			have_pick = true;
+		}
 	}
 	rcu_read_unlock();
 
-	return ret;
+	if (have_pick)
+		return 1;
+	if (!have_dirty_ptrs)
+		return 0;
+	if (have_missing_devs)
+		return -BCH_ERR_no_device_to_read_from;
+	if (have_csum_errors)
+		return -BCH_ERR_data_read_csum_err;
+	if (have_io_errors)
+		return -BCH_ERR_data_read_io_err;
+
+	/*
+	 * If we get here, we have pointers (bkey_ptrs_validate() ensures that),
+	 * but they don't point to valid devices:
+	 */
+	return -BCH_ERR_no_devices_valid;
 }
 
 /* KEY_TYPE_btree_ptr: */
@@ -541,29 +595,35 @@ static void bch2_extent_crc_pack(union bch_extent_crc *dst,
 				 struct bch_extent_crc_unpacked src,
 				 enum bch_extent_entry_type type)
 {
-#define set_common_fields(_dst, _src)					\
-		_dst.type		= 1 << type;			\
-		_dst.csum_type		= _src.csum_type,		\
-		_dst.compression_type	= _src.compression_type,	\
-		_dst._compressed_size	= _src.compressed_size - 1,	\
-		_dst._uncompressed_size	= _src.uncompressed_size - 1,	\
-		_dst.offset		= _src.offset
+#define common_fields(_src)						\
+		.type			= BIT(type),			\
+		.csum_type		= _src.csum_type,		\
+		.compression_type	= _src.compression_type,	\
+		._compressed_size	= _src.compressed_size - 1,	\
+		._uncompressed_size	= _src.uncompressed_size - 1,	\
+		.offset			= _src.offset
 
 	switch (type) {
 	case BCH_EXTENT_ENTRY_crc32:
-		set_common_fields(dst->crc32, src);
-		dst->crc32.csum		= (u32 __force) *((__le32 *) &src.csum.lo);
+		dst->crc32		= (struct bch_extent_crc32) {
+			common_fields(src),
+			.csum		= (u32 __force) *((__le32 *) &src.csum.lo),
+		};
 		break;
 	case BCH_EXTENT_ENTRY_crc64:
-		set_common_fields(dst->crc64, src);
-		dst->crc64.nonce	= src.nonce;
-		dst->crc64.csum_lo	= (u64 __force) src.csum.lo;
-		dst->crc64.csum_hi	= (u64 __force) *((__le16 *) &src.csum.hi);
+		dst->crc64		= (struct bch_extent_crc64) {
+			common_fields(src),
+			.nonce		= src.nonce,
+			.csum_lo	= (u64 __force) src.csum.lo,
+			.csum_hi	= (u64 __force) *((__le16 *) &src.csum.hi),
+		};
 		break;
 	case BCH_EXTENT_ENTRY_crc128:
-		set_common_fields(dst->crc128, src);
-		dst->crc128.nonce	= src.nonce;
-		dst->crc128.csum	= src.csum;
+		dst->crc128		= (struct bch_extent_crc128) {
+			common_fields(src),
+			.nonce		= src.nonce,
+			.csum		= src.csum,
+		};
 		break;
 	default:
 		BUG();
@@ -1002,7 +1062,7 @@ static bool want_cached_ptr(struct bch_fs *c, struct bch_io_opts *opts,
 
 	struct bch_dev *ca = bch2_dev_rcu_noerror(c, ptr->dev);
 
-	return ca && bch2_dev_is_readable(ca) && !dev_ptr_stale_rcu(ca, ptr);
+	return ca && bch2_dev_is_healthy(ca) && !dev_ptr_stale_rcu(ca, ptr);
 }
 
 void bch2_extent_ptr_set_cached(struct bch_fs *c,
@@ -1225,6 +1285,10 @@ void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 			bch2_extent_rebalance_to_text(out, c, &entry->rebalance);
 			break;
 
+		case BCH_EXTENT_ENTRY_flags:
+			prt_bitflags(out, bch2_extent_flags_strs, entry->flags.flags);
+			break;
+
 		default:
 			prt_printf(out, "(invalid extent entry %.16llx)", *((u64 *) entry));
 			return;
@@ -1386,6 +1450,11 @@ int bch2_bkey_ptrs_validate(struct bch_fs *c, struct bkey_s_c k,
 #endif
 			break;
 		}
+		case BCH_EXTENT_ENTRY_flags:
+			bkey_fsck_err_on(entry != ptrs.start,
+					 c, extent_flags_not_at_start,
+					 "extent flags entry not at start");
+			break;
 		}
 	}
 
@@ -1452,6 +1521,28 @@ void bch2_ptr_swab(struct bkey_s k)
 	}
 }
 
+int bch2_bkey_extent_flags_set(struct bch_fs *c, struct bkey_i *k, u64 flags)
+{
+	int ret = bch2_request_incompat_feature(c, bcachefs_metadata_version_extent_flags);
+	if (ret)
+		return ret;
+
+	struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
+
+	if (ptrs.start != ptrs.end &&
+	    extent_entry_type(ptrs.start) == BCH_EXTENT_ENTRY_flags) {
+		ptrs.start->flags.flags = flags;
+	} else {
+		struct bch_extent_flags f = {
+			.type	= BIT(BCH_EXTENT_ENTRY_flags),
+			.flags	= flags,
+		};
+		__extent_entry_insert(k, ptrs.start, (union bch_extent_entry *) &f);
+	}
+
+	return 0;
+}
+
 /* Generic extent code: */
 
 int bch2_cut_front_s(struct bpos where, struct bkey_s k)
@@ -1497,8 +1588,8 @@ int bch2_cut_front_s(struct bpos where, struct bkey_s k)
 				entry->crc128.offset += sub;
 				break;
 			case BCH_EXTENT_ENTRY_stripe_ptr:
-				break;
 			case BCH_EXTENT_ENTRY_rebalance:
+			case BCH_EXTENT_ENTRY_flags:
 				break;
 			}
 

libbcachefs/extents.h

@@ -320,8 +320,9 @@ static inline struct bkey_ptrs bch2_bkey_ptrs(struct bkey_s k)
 ({									\
 	__label__ out;							\
 									\
-	(_ptr).idx	= 0;						\
-	(_ptr).has_ec	= false;					\
+	(_ptr).has_ec			= false;			\
+	(_ptr).do_ec_reconstruct	= false;			\
+	(_ptr).crc_retry_nr		= 0;				\
 									\
 	__bkey_extent_entry_for_each_from(_entry, _end, _entry)		\
 		switch (__extent_entry_type(_entry)) {			\
@@ -401,7 +402,7 @@ out:									\
 struct bch_dev_io_failures *bch2_dev_io_failures(struct bch_io_failures *,
 						 unsigned);
 void bch2_mark_io_failure(struct bch_io_failures *,
-			  struct extent_ptr_decoded *);
+			  struct extent_ptr_decoded *, bool);
 int bch2_bkey_pick_read_device(struct bch_fs *, struct bkey_s_c,
 			       struct bch_io_failures *,
 			       struct extent_ptr_decoded *, int);
@@ -704,7 +705,7 @@ static inline bool bch2_extent_ptr_eq(struct bch_extent_ptr ptr1,
 		ptr1.unwritten	== ptr2.unwritten &&
 		ptr1.offset	== ptr2.offset &&
 		ptr1.dev	== ptr2.dev &&
-		ptr1.dev	== ptr2.dev);
+		ptr1.gen	== ptr2.gen);
 }
 
 void bch2_ptr_swab(struct bkey_s);
@@ -753,4 +754,19 @@ static inline void bch2_key_resize(struct bkey *k, unsigned new_size)
 	k->size = new_size;
 }
 
+static inline u64 bch2_bkey_extent_ptrs_flags(struct bkey_ptrs_c ptrs)
+{
+	if (ptrs.start != ptrs.end &&
+	    extent_entry_type(ptrs.start) == BCH_EXTENT_ENTRY_flags)
+		return ptrs.start->flags.flags;
+	return 0;
+}
+
+static inline u64 bch2_bkey_extent_flags(struct bkey_s_c k)
+{
+	return bch2_bkey_extent_ptrs_flags(bch2_bkey_ptrs_c(k));
+}
+
+int bch2_bkey_extent_flags_set(struct bch_fs *, struct bkey_i *, u64);
+
 #endif /* _BCACHEFS_EXTENTS_H */

libbcachefs/extents_format.h

@@ -79,8 +79,9 @@
 	x(crc64,		2)		\
 	x(crc128,		3)		\
 	x(stripe_ptr,		4)		\
-	x(rebalance,		5)
-#define BCH_EXTENT_ENTRY_MAX	6
+	x(rebalance,		5)		\
+	x(flags,		6)
+#define BCH_EXTENT_ENTRY_MAX	7
 
 enum bch_extent_entry_type {
 #define x(f, n) BCH_EXTENT_ENTRY_##f = n,
@@ -201,6 +202,25 @@ struct bch_extent_stripe_ptr {
 #endif
 };
 
+#define BCH_EXTENT_FLAGS()		\
+	x(poisoned,		0)
+
+enum bch_extent_flags_e {
+#define x(n, v)	BCH_EXTENT_FLAG_##n = v,
+	BCH_EXTENT_FLAGS()
+#undef x
+};
+
+struct bch_extent_flags {
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u64			type:7,
+				flags:57;
+#elif defined (__BIG_ENDIAN_BITFIELD)
+	__u64			flags:57,
+				type:7;
+#endif
+};
+
 /* bch_extent_rebalance: */
 #include "rebalance_format.h"
 

libbcachefs/extents_types.h

@@ -20,8 +20,9 @@ struct bch_extent_crc_unpacked {
 };
 
 struct extent_ptr_decoded {
-	unsigned			idx;
 	bool				has_ec;
+	bool				do_ec_reconstruct;
+	u8				crc_retry_nr;
 	struct bch_extent_crc_unpacked	crc;
 	struct bch_extent_ptr		ptr;
 	struct bch_extent_stripe_ptr	ec;
@@ -31,10 +32,10 @@ struct bch_io_failures {
 	u8			nr;
 	struct bch_dev_io_failures {
 		u8		dev;
-		u8		idx;
-		u8		nr_failed;
-		u8		nr_retries;
-	}			devs[BCH_REPLICAS_MAX];
+		unsigned	failed_csum_nr:6,
+				failed_io:1,
+				failed_ec:1;
+	}			devs[BCH_REPLICAS_MAX + 1];
 };
 
 #endif /* _BCACHEFS_EXTENTS_TYPES_H */

libbcachefs/eytzinger.c

@@ -148,87 +148,97 @@ static int do_cmp(const void *a, const void *b, cmp_r_func_t cmp, const void *pr
 	return cmp(a, b, priv);
 }
 
-static inline int eytzinger0_do_cmp(void *base, size_t n, size_t size,
+static inline int eytzinger1_do_cmp(void *base1, size_t n, size_t size,
 			 cmp_r_func_t cmp_func, const void *priv,
 			 size_t l, size_t r)
 {
-	return do_cmp(base + inorder_to_eytzinger0(l, n) * size,
-		      base + inorder_to_eytzinger0(r, n) * size,
+	return do_cmp(base1 + inorder_to_eytzinger1(l, n) * size,
+		      base1 + inorder_to_eytzinger1(r, n) * size,
 		      cmp_func, priv);
 }
 
-static inline void eytzinger0_do_swap(void *base, size_t n, size_t size,
+static inline void eytzinger1_do_swap(void *base1, size_t n, size_t size,
 			   swap_r_func_t swap_func, const void *priv,
 			   size_t l, size_t r)
 {
-	do_swap(base + inorder_to_eytzinger0(l, n) * size,
-		base + inorder_to_eytzinger0(r, n) * size,
+	do_swap(base1 + inorder_to_eytzinger1(l, n) * size,
+		base1 + inorder_to_eytzinger1(r, n) * size,
 		size, swap_func, priv);
 }
 
+static void eytzinger1_sort_r(void *base1, size_t n, size_t size,
+			      cmp_r_func_t cmp_func,
+			      swap_r_func_t swap_func,
+			      const void *priv)
+{
+	unsigned i, j, k;
+
+	/* called from 'sort' without swap function, let's pick the default */
+	if (swap_func == SWAP_WRAPPER && !((struct wrapper *)priv)->swap_func)
+		swap_func = NULL;
+
+	if (!swap_func) {
+		if (is_aligned(base1, size, 8))
+			swap_func = SWAP_WORDS_64;
+		else if (is_aligned(base1, size, 4))
+			swap_func = SWAP_WORDS_32;
+		else
+			swap_func = SWAP_BYTES;
+	}
+
+	/* heapify */
+	for (i = n / 2; i >= 1; --i) {
+		/* Find the sift-down path all the way to the leaves. */
+		for (j = i; k = j * 2, k < n;)
+			j = eytzinger1_do_cmp(base1, n, size, cmp_func, priv, k, k + 1) > 0 ? k : k + 1;
+
+		/* Special case for the last leaf with no sibling. */
+		if (j * 2 == n)
+			j *= 2;
+
+		/* Backtrack to the correct location. */
+		while (j != i && eytzinger1_do_cmp(base1, n, size, cmp_func, priv, i, j) >= 0)
+			j /= 2;
+
+		/* Shift the element into its correct place. */
+		for (k = j; j != i;) {
+			j /= 2;
+			eytzinger1_do_swap(base1, n, size, swap_func, priv, j, k);
+		}
+	}
+
+	/* sort */
+	for (i = n; i > 1; --i) {
+		eytzinger1_do_swap(base1, n, size, swap_func, priv, 1, i);
+
+		/* Find the sift-down path all the way to the leaves. */
+		for (j = 1; k = j * 2, k + 1 < i;)
+			j = eytzinger1_do_cmp(base1, n, size, cmp_func, priv, k, k + 1) > 0 ? k : k + 1;
+
+		/* Special case for the last leaf with no sibling. */
+		if (j * 2 + 1 == i)
+			j *= 2;
+
+		/* Backtrack to the correct location. */
+		while (j >= 1 && eytzinger1_do_cmp(base1, n, size, cmp_func, priv, 1, j) >= 0)
+			j /= 2;
+
+		/* Shift the element into its correct place. */
+		for (k = j; j > 1;) {
+			j /= 2;
+			eytzinger1_do_swap(base1, n, size, swap_func, priv, j, k);
+		}
+	}
+}
+
 void eytzinger0_sort_r(void *base, size_t n, size_t size,
 		       cmp_r_func_t cmp_func,
 		       swap_r_func_t swap_func,
 		       const void *priv)
 {
-	int i, j, k;
+	void *base1 = base - size;
 
-	/* called from 'sort' without swap function, let's pick the default */
-	if (swap_func == SWAP_WRAPPER && !((struct wrapper *)priv)->swap_func)
-		swap_func = NULL;
-
-	if (!swap_func) {
-		if (is_aligned(base, size, 8))
-			swap_func = SWAP_WORDS_64;
-		else if (is_aligned(base, size, 4))
-			swap_func = SWAP_WORDS_32;
-		else
-			swap_func = SWAP_BYTES;
-	}
-
-	/* heapify */
-	for (i = n / 2 - 1; i >= 0; --i) {
-		/* Find the sift-down path all the way to the leaves. */
-		for (j = i; k = j * 2 + 1, k + 1 < n;)
-			j = eytzinger0_do_cmp(base, n, size, cmp_func, priv, k, k + 1) > 0 ? k : k + 1;
-
-		/* Special case for the last leaf with no sibling. */
-		if (j * 2 + 2 == n)
-			j = j * 2 + 1;
-
-		/* Backtrack to the correct location. */
-		while (j != i && eytzinger0_do_cmp(base, n, size, cmp_func, priv, i, j) >= 0)
-			j = (j - 1) / 2;
-
-		/* Shift the element into its correct place. */
-		for (k = j; j != i;) {
-			j = (j - 1) / 2;
-			eytzinger0_do_swap(base, n, size, swap_func, priv, j, k);
-		}
-	}
-
-	/* sort */
-	for (i = n - 1; i > 0; --i) {
-		eytzinger0_do_swap(base, n, size, swap_func, priv, 0, i);
-
-		/* Find the sift-down path all the way to the leaves. */
-		for (j = 0; k = j * 2 + 1, k + 1 < i;)
-			j = eytzinger0_do_cmp(base, n, size, cmp_func, priv, k, k + 1) > 0 ? k : k + 1;
-
-		/* Special case for the last leaf with no sibling. */
-		if (j * 2 + 2 == i)
-			j = j * 2 + 1;
-
-		/* Backtrack to the correct location. */
-		while (j && eytzinger0_do_cmp(base, n, size, cmp_func, priv, 0, j) >= 0)
-			j = (j - 1) / 2;
-
-		/* Shift the element into its correct place. */
-		for (k = j; j;) {
-			j = (j - 1) / 2;
-			eytzinger0_do_swap(base, n, size, swap_func, priv, j, k);
-		}
-	}
+	return eytzinger1_sort_r(base1, n, size, cmp_func, swap_func, priv);
 }
 
 void eytzinger0_sort(void *base, size_t n, size_t size,

libbcachefs/eytzinger.h

@@ -6,6 +6,7 @@
 #include <linux/log2.h>
 
 #ifdef EYTZINGER_DEBUG
+#include <linux/bug.h>
 #define EYTZINGER_BUG_ON(cond)		BUG_ON(cond)
 #else
 #define EYTZINGER_BUG_ON(cond)
@@ -56,24 +57,14 @@ static inline unsigned eytzinger1_last(unsigned size)
 	return rounddown_pow_of_two(size + 1) - 1;
 }
 
-/*
- * eytzinger1_next() and eytzinger1_prev() have the nice properties that
- *
- * eytzinger1_next(0) == eytzinger1_first())
- * eytzinger1_prev(0) == eytzinger1_last())
- *
- * eytzinger1_prev(eytzinger1_first()) == 0
- * eytzinger1_next(eytzinger1_last()) == 0
- */
-
 static inline unsigned eytzinger1_next(unsigned i, unsigned size)
 {
-	EYTZINGER_BUG_ON(i > size);
+	EYTZINGER_BUG_ON(i == 0 || i > size);
 
 	if (eytzinger1_right_child(i) <= size) {
 		i = eytzinger1_right_child(i);
 
-		i <<= __fls(size + 1) - __fls(i);
+		i <<= __fls(size) - __fls(i);
 		i >>= i > size;
 	} else {
 		i >>= ffz(i) + 1;
@@ -84,12 +75,12 @@ static inline unsigned eytzinger1_next(unsigned i, unsigned size)
 
 static inline unsigned eytzinger1_prev(unsigned i, unsigned size)
 {
-	EYTZINGER_BUG_ON(i > size);
+	EYTZINGER_BUG_ON(i == 0 || i > size);
 
 	if (eytzinger1_left_child(i) <= size) {
 		i = eytzinger1_left_child(i) + 1;
 
-		i <<= __fls(size + 1) - __fls(i);
+		i <<= __fls(size) - __fls(i);
 		i -= 1;
 		i >>= i > size;
 	} else {
@@ -243,73 +234,63 @@ static inline unsigned inorder_to_eytzinger0(unsigned i, unsigned size)
 	     (_i) != -1;				\
 	     (_i) = eytzinger0_next((_i), (_size)))
 
+#define eytzinger0_for_each_prev(_i, _size)		\
+	for (unsigned (_i) = eytzinger0_last((_size));	\
+	     (_i) != -1;				\
+	     (_i) = eytzinger0_prev((_i), (_size)))
+
 /* return greatest node <= @search, or -1 if not found */
 static inline int eytzinger0_find_le(void *base, size_t nr, size_t size,
 				     cmp_func_t cmp, const void *search)
 {
-	unsigned i, n = 0;
+	void *base1 = base - size;
+	unsigned n = 1;
 
-	if (!nr)
-		return -1;
-
-	do {
-		i = n;
-		n = eytzinger0_child(i, cmp(base + i * size, search) <= 0);
-	} while (n < nr);
-
-	if (n & 1) {
-		/*
-		 * @i was greater than @search, return previous node:
-		 *
-		 * if @i was leftmost/smallest element,
-		 * eytzinger0_prev(eytzinger0_first())) returns -1, as expected
-		 */
-		return eytzinger0_prev(i, nr);
-	} else {
-		return i;
-	}
+	while (n <= nr)
+		n = eytzinger1_child(n, cmp(base1 + n * size, search) <= 0);
+	n >>= __ffs(n) + 1;
+	return n - 1;
 }
 
+/* return smallest node > @search, or -1 if not found */
 static inline int eytzinger0_find_gt(void *base, size_t nr, size_t size,
 				     cmp_func_t cmp, const void *search)
 {
-	ssize_t idx = eytzinger0_find_le(base, nr, size, cmp, search);
+	void *base1 = base - size;
+	unsigned n = 1;
 
-	/*
-	 * if eytitzinger0_find_le() returned -1 - no element was <= search - we
-	 * want to return the first element; next/prev identities mean this work
-	 * as expected
-	 *
-	 * similarly if find_le() returns last element, we should return -1;
-	 * identities mean this all works out:
-	 */
-	return eytzinger0_next(idx, nr);
+	while (n <= nr)
+		n = eytzinger1_child(n, cmp(base1 + n * size, search) <= 0);
+	n >>= __ffs(n + 1) + 1;
+	return n - 1;
 }
 
+/* return smallest node >= @search, or -1 if not found */
 static inline int eytzinger0_find_ge(void *base, size_t nr, size_t size,
 				     cmp_func_t cmp, const void *search)
 {
-	ssize_t idx = eytzinger0_find_le(base, nr, size, cmp, search);
+	void *base1 = base - size;
+	unsigned n = 1;
 
-	if (idx < nr && !cmp(base + idx * size, search))
-		return idx;
-
-	return eytzinger0_next(idx, nr);
+	while (n <= nr)
+		n = eytzinger1_child(n, cmp(base1 + n * size, search) < 0);
+	n >>= __ffs(n + 1) + 1;
+	return n - 1;
 }
 
 #define eytzinger0_find(base, nr, size, _cmp, search)			\
 ({									\
-	void *_base		= (base);				\
+	size_t _size		= (size);				\
+	void *_base1		= (void *)(base) - _size;		\
 	const void *_search	= (search);				\
 	size_t _nr		= (nr);					\
-	size_t _size		= (size);				\
-	size_t _i		= 0;					\
+	size_t _i		= 1;					\
 	int _res;							\
 									\
-	while (_i < _nr &&						\
-	       (_res = _cmp(_search, _base + _i * _size)))		\
-		_i = eytzinger0_child(_i, _res > 0);			\
-	_i;								\
+	while (_i <= _nr &&						\
+	       (_res = _cmp(_search, _base1 + _i * _size)))		\
+		_i = eytzinger1_child(_i, _res > 0);			\
+	_i - 1;								\
 })
 
 void eytzinger0_sort_r(void *, size_t, size_t,

libbcachefs/fs-io-buffered.c

@@ -110,11 +110,21 @@ static int readpage_bio_extend(struct btree_trans *trans,
 			if (!get_more)
 				break;
 
+			unsigned sectors_remaining = sectors_this_extent - bio_sectors(bio);
+
+			if (sectors_remaining < PAGE_SECTORS << mapping_min_folio_order(iter->mapping))
+				break;
+
+			unsigned order = ilog2(rounddown_pow_of_two(sectors_remaining) / PAGE_SECTORS);
+
+			/* ensure proper alignment */
+			order = min(order, __ffs(folio_offset|BIT(31)));
+
 			folio = xa_load(&iter->mapping->i_pages, folio_offset);
 			if (folio && !xa_is_value(folio))
 				break;
 
-			folio = filemap_alloc_folio(readahead_gfp_mask(iter->mapping), 0);
+			folio = filemap_alloc_folio(readahead_gfp_mask(iter->mapping), order);
 			if (!folio)
 				break;
 
@@ -215,11 +225,11 @@ static void bchfs_read(struct btree_trans *trans,
 
 		bch2_read_extent(trans, rbio, iter.pos,
 				 data_btree, k, offset_into_extent, flags);
+		swap(rbio->bio.bi_iter.bi_size, bytes);
 
 		if (flags & BCH_READ_last_fragment)
 			break;
 
-		swap(rbio->bio.bi_iter.bi_size, bytes);
 		bio_advance(&rbio->bio, bytes);
 err:
 		if (ret &&
@@ -230,7 +240,8 @@ err:
 
 	if (ret) {
 		struct printbuf buf = PRINTBUF;
-		bch2_inum_offset_err_msg_trans(trans, &buf, inum, iter.pos.offset << 9);
+		lockrestart_do(trans,
+			bch2_inum_offset_err_msg_trans(trans, &buf, inum, iter.pos.offset << 9));
 		prt_printf(&buf, "read error %i from btree lookup", ret);
 		bch_err_ratelimited(c, "%s", buf.buf);
 		printbuf_exit(&buf);

libbcachefs/fs-io.c

@@ -466,6 +466,7 @@ int bchfs_truncate(struct mnt_idmap *idmap,
 	ret = bch2_truncate_folio(inode, iattr->ia_size);
 	if (unlikely(ret < 0))
 		goto err;
+	ret = 0;
 
 	truncate_setsize(&inode->v, iattr->ia_size);
 
@@ -998,17 +999,28 @@ static loff_t bch2_seek_hole(struct file *file, u64 offset)
 				   POS(inode->v.i_ino, offset >> 9),
 				   POS(inode->v.i_ino, U64_MAX),
 				   inum.subvol, BTREE_ITER_slots, k, ({
-			if (k.k->p.inode != inode->v.i_ino) {
-				next_hole = bch2_seek_pagecache_hole(&inode->v,
-						offset, MAX_LFS_FILESIZE, 0, false);
-				break;
-			} else if (!bkey_extent_is_data(k.k)) {
-				next_hole = bch2_seek_pagecache_hole(&inode->v,
-						max(offset, bkey_start_offset(k.k) << 9),
-						k.k->p.offset << 9, 0, false);
+			if (k.k->p.inode != inode->v.i_ino ||
+			    !bkey_extent_is_data(k.k)) {
+				loff_t start_offset = k.k->p.inode == inode->v.i_ino
+					? max(offset, bkey_start_offset(k.k) << 9)
+					: offset;
+				loff_t end_offset = k.k->p.inode == inode->v.i_ino
+					? MAX_LFS_FILESIZE
+					: k.k->p.offset << 9;
 
-				if (next_hole < k.k->p.offset << 9)
+				/*
+				 * Found a hole in the btree, now make sure it's
+				 * a hole in the pagecache. We might have to
+				 * keep searching if this hole is entirely dirty
+				 * in the page cache:
+				 */
+				bch2_trans_unlock(trans);
+				loff_t pagecache_hole = bch2_seek_pagecache_hole(&inode->v,
+								start_offset, end_offset, 0, false);
+				if (pagecache_hole < end_offset) {
+					next_hole = pagecache_hole;
 					break;
+				}
 			} else {
 				offset = max(offset, bkey_start_offset(k.k) << 9);
 			}

libbcachefs/fs-ioctl.c

@@ -5,8 +5,8 @@
 #include "chardev.h"
 #include "dirent.h"
 #include "fs.h"
-#include "fs-common.h"
 #include "fs-ioctl.h"
+#include "namei.h"
 #include "quota.h"
 
 #include <linux/compat.h>
@@ -54,6 +54,32 @@ static int bch2_inode_flags_set(struct btree_trans *trans,
 	    (newflags & (BCH_INODE_nodump|BCH_INODE_noatime)) != newflags)
 		return -EINVAL;
 
+	if ((newflags ^ oldflags) & BCH_INODE_casefolded) {
+#ifdef CONFIG_UNICODE
+		int ret = 0;
+		/* Not supported on individual files. */
+		if (!S_ISDIR(bi->bi_mode))
+			return -EOPNOTSUPP;
+
+		/*
+		 * Make sure the dir is empty, as otherwise we'd need to
+		 * rehash everything and update the dirent keys.
+		 */
+		ret = bch2_empty_dir_trans(trans, inode_inum(inode));
+		if (ret < 0)
+			return ret;
+
+		ret = bch2_request_incompat_feature(c,bcachefs_metadata_version_casefolding);
+		if (ret)
+			return ret;
+
+		bch2_check_set_feature(c, BCH_FEATURE_casefolding);
+#else
+		printk(KERN_ERR "Cannot use casefolding on a kernel without CONFIG_UNICODE\n");
+		return -EOPNOTSUPP;
+#endif
+	}
+
 	if (s->set_projinherit) {
 		bi->bi_fields_set &= ~(1 << Inode_opt_project);
 		bi->bi_fields_set |= ((int) s->projinherit << Inode_opt_project);
@@ -218,7 +244,7 @@ static int bch2_ioc_reinherit_attrs(struct bch_fs *c,
 	int ret = 0;
 	subvol_inum inum;
 
-	kname = kmalloc(BCH_NAME_MAX + 1, GFP_KERNEL);
+	kname = kmalloc(BCH_NAME_MAX, GFP_KERNEL);
 	if (!kname)
 		return -ENOMEM;
 
@@ -515,10 +541,12 @@ static long bch2_ioctl_subvolume_destroy(struct bch_fs *c, struct file *filp,
 		ret = -ENOENT;
 		goto err;
 	}
-	ret = __bch2_unlink(dir, victim, true);
+
+	ret =   inode_permission(file_mnt_idmap(filp), d_inode(victim), MAY_WRITE) ?:
+		__bch2_unlink(dir, victim, true);
 	if (!ret) {
 		fsnotify_rmdir(dir, victim);
-		d_delete(victim);
+		d_invalidate(victim);
 	}
 err:
 	inode_unlock(dir);

libbcachefs/fs-ioctl.h

@@ -6,19 +6,21 @@
 
 /* bcachefs inode flags -> vfs inode flags: */
 static const __maybe_unused unsigned bch_flags_to_vfs[] = {
-	[__BCH_INODE_sync]	= S_SYNC,
-	[__BCH_INODE_immutable]	= S_IMMUTABLE,
-	[__BCH_INODE_append]	= S_APPEND,
-	[__BCH_INODE_noatime]	= S_NOATIME,
+	[__BCH_INODE_sync]		= S_SYNC,
+	[__BCH_INODE_immutable]		= S_IMMUTABLE,
+	[__BCH_INODE_append]		= S_APPEND,
+	[__BCH_INODE_noatime]		= S_NOATIME,
+	[__BCH_INODE_casefolded]	= S_CASEFOLD,
 };
 
 /* bcachefs inode flags -> FS_IOC_GETFLAGS: */
 static const __maybe_unused unsigned bch_flags_to_uflags[] = {
-	[__BCH_INODE_sync]	= FS_SYNC_FL,
-	[__BCH_INODE_immutable]	= FS_IMMUTABLE_FL,
-	[__BCH_INODE_append]	= FS_APPEND_FL,
-	[__BCH_INODE_nodump]	= FS_NODUMP_FL,
-	[__BCH_INODE_noatime]	= FS_NOATIME_FL,
+	[__BCH_INODE_sync]		= FS_SYNC_FL,
+	[__BCH_INODE_immutable]		= FS_IMMUTABLE_FL,
+	[__BCH_INODE_append]		= FS_APPEND_FL,
+	[__BCH_INODE_nodump]		= FS_NODUMP_FL,
+	[__BCH_INODE_noatime]		= FS_NOATIME_FL,
+	[__BCH_INODE_casefolded]	= FS_CASEFOLD_FL,
 };
 
 /* bcachefs inode flags -> FS_IOC_FSGETXATTR: */

libbcachefs/fs.c

@@ -11,7 +11,6 @@
 #include "errcode.h"
 #include "extents.h"
 #include "fs.h"
-#include "fs-common.h"
 #include "fs-io.h"
 #include "fs-ioctl.h"
 #include "fs-io-buffered.h"
@@ -22,6 +21,7 @@
 #include "io_read.h"
 #include "journal.h"
 #include "keylist.h"
+#include "namei.h"
 #include "quota.h"
 #include "rebalance.h"
 #include "snapshot.h"
@@ -641,7 +641,9 @@ static struct bch_inode_info *bch2_lookup_trans(struct btree_trans *trans,
 	if (ret)
 		return ERR_PTR(ret);
 
-	ret = bch2_dirent_read_target(trans, dir, bkey_s_c_to_dirent(k), &inum);
+	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
+
+	ret = bch2_dirent_read_target(trans, dir, d, &inum);
 	if (ret > 0)
 		ret = -ENOENT;
 	if (ret)
@@ -651,30 +653,30 @@ static struct bch_inode_info *bch2_lookup_trans(struct btree_trans *trans,
 	if (inode)
 		goto out;
 
+	/*
+	 * Note: if check/repair needs it, we commit before
+	 * bch2_inode_hash_init_insert(), as after that point we can't take a
+	 * restart - not in the top level loop with a commit_do(), like we
+	 * usually do:
+	 */
+
 	struct bch_subvolume subvol;
 	struct bch_inode_unpacked inode_u;
 	ret =   bch2_subvolume_get(trans, inum.subvol, true, &subvol) ?:
 		bch2_inode_find_by_inum_nowarn_trans(trans, inum, &inode_u) ?:
+		bch2_check_dirent_target(trans, &dirent_iter, d, &inode_u, false) ?:
+		bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc) ?:
 		PTR_ERR_OR_ZERO(inode = bch2_inode_hash_init_insert(trans, inum, &inode_u, &subvol));
 
+	/*
+	 * don't remove it: check_inodes might find another inode that points
+	 * back to this dirent
+	 */
 	bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT),
-				c, "dirent to missing inode:\n  %s",
-				(bch2_bkey_val_to_text(&buf, c, k), buf.buf));
+				c, "dirent to missing inode:\n%s",
+				(bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf));
 	if (ret)
 		goto err;
-
-	/* regular files may have hardlinks: */
-	if (bch2_fs_inconsistent_on(bch2_inode_should_have_single_bp(&inode_u) &&
-				    !bkey_eq(k.k->p, POS(inode_u.bi_dir, inode_u.bi_dir_offset)),
-				    c,
-				    "dirent points to inode that does not point back:\n  %s",
-				    (bch2_bkey_val_to_text(&buf, c, k),
-				     prt_printf(&buf, "\n  "),
-				     bch2_inode_unpacked_to_text(&buf, &inode_u),
-				     buf.buf))) {
-		ret = -ENOENT;
-		goto err;
-	}
 out:
 	bch2_trans_iter_exit(trans, &dirent_iter);
 	printbuf_exit(&buf);
@@ -698,6 +700,23 @@ static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 	if (IS_ERR(inode))
 		inode = NULL;
 
+#ifdef CONFIG_UNICODE
+	if (!inode && IS_CASEFOLDED(vdir)) {
+		/*
+		 * Do not cache a negative dentry in casefolded directories
+		 * as it would need to be invalidated in the following situation:
+		 * - Lookup file "blAH" in a casefolded directory
+		 * - Creation of file "BLAH" in a casefolded directory
+		 * - Lookup file "blAH" in a casefolded directory
+		 * which would fail if we had a negative dentry.
+		 *
+		 * We should come back to this when VFS has a method to handle
+		 * this edgecase.
+		 */
+		return NULL;
+	}
+#endif
+
 	return d_splice_alias(&inode->v, dentry);
 }
 
@@ -1802,7 +1821,8 @@ static void bch2_vfs_inode_init(struct btree_trans *trans,
 		break;
 	}
 
-	mapping_set_large_folios(inode->v.i_mapping);
+	mapping_set_folio_min_order(inode->v.i_mapping,
+				    get_order(trans->c->opts.block_size));
 }
 
 static void bch2_free_inode(struct inode *vinode)
@@ -2008,44 +2028,6 @@ static struct bch_fs *bch2_path_to_fs(const char *path)
 	return c ?: ERR_PTR(-ENOENT);
 }
 
-static int bch2_remount(struct super_block *sb, int *flags,
-			struct bch_opts opts)
-{
-	struct bch_fs *c = sb->s_fs_info;
-	int ret = 0;
-
-	opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
-
-	if (opts.read_only != c->opts.read_only) {
-		down_write(&c->state_lock);
-
-		if (opts.read_only) {
-			bch2_fs_read_only(c);
-
-			sb->s_flags |= SB_RDONLY;
-		} else {
-			ret = bch2_fs_read_write(c);
-			if (ret) {
-				bch_err(c, "error going rw: %i", ret);
-				up_write(&c->state_lock);
-				ret = -EINVAL;
-				goto err;
-			}
-
-			sb->s_flags &= ~SB_RDONLY;
-		}
-
-		c->opts.read_only = opts.read_only;
-
-		up_write(&c->state_lock);
-	}
-
-	if (opt_defined(opts, errors))
-		c->opts.errors = opts.errors;
-err:
-	return bch2_err_class(ret);
-}
-
 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
 {
 	struct bch_fs *c = root->d_sb->s_fs_info;
@@ -2192,17 +2174,21 @@ static int bch2_fs_get_tree(struct fs_context *fc)
 	if (ret)
 		goto err;
 
+	if (opt_defined(opts, discard))
+		set_bit(BCH_FS_discard_mount_opt_set, &c->flags);
+
 	/* Some options can't be parsed until after the fs is started: */
 	opts = bch2_opts_empty();
-	ret = bch2_parse_mount_opts(c, &opts, NULL, opts_parse->parse_later.buf);
+	ret = bch2_parse_mount_opts(c, &opts, NULL, opts_parse->parse_later.buf, false);
 	if (ret)
 		goto err_stop_fs;
 
 	bch2_opts_apply(&c->opts, opts);
 
-	ret = bch2_fs_start(c);
-	if (ret)
-		goto err_stop_fs;
+	/*
+	 * need to initialise sb and set c->vfs_sb _before_ starting fs,
+	 * for blk_holder_ops
+	 */
 
 	sb = sget(fc->fs_type, NULL, bch2_set_super, fc->sb_flags|SB_NOSEC, c);
 	ret = PTR_ERR_OR_ZERO(sb);
@@ -2264,6 +2250,10 @@ got_sb:
 
 	sb->s_shrink->seeks = 0;
 
+	ret = bch2_fs_start(c);
+	if (ret)
+		goto err_put_super;
+
 	vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_SUBVOL_INUM);
 	ret = PTR_ERR_OR_ZERO(vinode);
 	bch_err_msg(c, ret, "mounting: error getting root inode");
@@ -2300,7 +2290,8 @@ err_stop_fs:
 	goto err;
 
 err_put_super:
-	__bch2_fs_stop(c);
+	if (!sb->s_root)
+		__bch2_fs_stop(c);
 	deactivate_locked_super(sb);
 	goto err;
 }
@@ -2343,6 +2334,8 @@ static int bch2_fs_parse_param(struct fs_context *fc,
 	int ret = bch2_parse_one_mount_opt(c, &opts->opts,
 					   &opts->parse_later, param->key,
 					   param->string);
+	if (ret)
+		pr_err("Error parsing option %s: %s", param->key, bch2_err_str(ret));
 
 	return bch2_err_class(ret);
 }
@@ -2351,8 +2344,39 @@ static int bch2_fs_reconfigure(struct fs_context *fc)
 {
 	struct super_block *sb = fc->root->d_sb;
 	struct bch2_opts_parse *opts = fc->fs_private;
+	struct bch_fs *c = sb->s_fs_info;
+	int ret = 0;
 
-	return bch2_remount(sb, &fc->sb_flags, opts->opts);
+	opt_set(opts->opts, read_only, (fc->sb_flags & SB_RDONLY) != 0);
+
+	if (opts->opts.read_only != c->opts.read_only) {
+		down_write(&c->state_lock);
+
+		if (opts->opts.read_only) {
+			bch2_fs_read_only(c);
+
+			sb->s_flags |= SB_RDONLY;
+		} else {
+			ret = bch2_fs_read_write(c);
+			if (ret) {
+				bch_err(c, "error going rw: %i", ret);
+				up_write(&c->state_lock);
+				ret = -EINVAL;
+				goto err;
+			}
+
+			sb->s_flags &= ~SB_RDONLY;
+		}
+
+		c->opts.read_only = opts->opts.read_only;
+
+		up_write(&c->state_lock);
+	}
+
+	if (opt_defined(opts->opts, errors))
+		c->opts.errors = opts->opts.errors;
+err:
+	return bch2_err_class(ret);
 }
 
 static const struct fs_context_operations bch2_context_ops = {

libbcachefs/fsck.c

@@ -10,10 +10,10 @@
 #include "dirent.h"
 #include "error.h"
 #include "fs.h"
-#include "fs-common.h"
 #include "fsck.h"
 #include "inode.h"
 #include "keylist.h"
+#include "namei.h"
 #include "recovery_passes.h"
 #include "snapshot.h"
 #include "super.h"
@@ -23,13 +23,6 @@
 #include <linux/bsearch.h>
 #include <linux/dcache.h> /* struct qstr */
 
-static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
-				   struct bkey_s_c_dirent d)
-{
-	return  inode->bi_dir		== d.k->p.inode &&
-		inode->bi_dir_offset	== d.k->p.offset;
-}
-
 static int dirent_points_to_inode_nowarn(struct bkey_s_c_dirent d,
 					 struct bch_inode_unpacked *inode)
 {
@@ -116,29 +109,6 @@ static int subvol_lookup(struct btree_trans *trans, u32 subvol,
 	return ret;
 }
 
-static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
-			      struct bch_inode_unpacked *inode)
-{
-	struct btree_iter iter;
-	struct bkey_s_c k;
-	int ret;
-
-	for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inode_nr),
-				     BTREE_ITER_all_snapshots, k, ret) {
-		if (k.k->p.offset != inode_nr)
-			break;
-		if (!bkey_is_inode(k.k))
-			continue;
-		ret = bch2_inode_unpack(k, inode);
-		goto found;
-	}
-	ret = -BCH_ERR_ENOENT_inode;
-found:
-	bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
-	bch2_trans_iter_exit(trans, &iter);
-	return ret;
-}
-
 static int lookup_inode(struct btree_trans *trans, u64 inode_nr, u32 snapshot,
 			struct bch_inode_unpacked *inode)
 {
@@ -179,32 +149,6 @@ static int lookup_dirent_in_snapshot(struct btree_trans *trans,
 	return 0;
 }
 
-static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
-{
-	struct bch_fs *c = trans->c;
-	struct btree_iter iter;
-	struct bch_inode_unpacked dir_inode;
-	struct bch_hash_info dir_hash_info;
-	int ret;
-
-	ret = lookup_first_inode(trans, pos.inode, &dir_inode);
-	if (ret)
-		goto err;
-
-	dir_hash_info = bch2_hash_info_init(c, &dir_inode);
-
-	bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_intent);
-
-	ret =   bch2_btree_iter_traverse(&iter) ?:
-		bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
-				    &dir_hash_info, &iter,
-				    BTREE_UPDATE_internal_snapshot_node);
-	bch2_trans_iter_exit(trans, &iter);
-err:
-	bch_err_fn(c, ret);
-	return ret;
-}
-
 /*
  * Find any subvolume associated with a tree of snapshots
  * We can't rely on master_subvol - it might have been deleted.
@@ -548,7 +492,7 @@ static int remove_backpointer(struct btree_trans *trans,
 				     SPOS(inode->bi_dir, inode->bi_dir_offset, inode->bi_snapshot));
 	int ret = bkey_err(d) ?:
 		  dirent_points_to_inode(c, d, inode) ?:
-		  __remove_dirent(trans, d.k->p);
+		  bch2_fsck_remove_dirent(trans, d.k->p);
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
 }
@@ -823,6 +767,7 @@ struct inode_walker_entry {
 	struct bch_inode_unpacked inode;
 	u32			snapshot;
 	u64			count;
+	u64			i_size;
 };
 
 struct inode_walker {
@@ -910,8 +855,9 @@ found:
 	if (k.k->p.snapshot != i->snapshot && !is_whiteout) {
 		struct inode_walker_entry new = *i;
 
-		new.snapshot = k.k->p.snapshot;
-		new.count = 0;
+		new.snapshot	= k.k->p.snapshot;
+		new.count	= 0;
+		new.i_size	= 0;
 
 		struct printbuf buf = PRINTBUF;
 		bch2_bkey_val_to_text(&buf, c, k);
@@ -1116,37 +1062,6 @@ err:
 	return ret;
 }
 
-static int check_directory_size(struct btree_trans *trans,
-				struct bch_inode_unpacked *inode_u,
-				struct bkey_s_c inode_k, bool *write_inode)
-{
-	struct btree_iter iter;
-	struct bkey_s_c k;
-	u64 new_size = 0;
-	int ret;
-
-	for_each_btree_key_max_norestart(trans, iter, BTREE_ID_dirents,
-			SPOS(inode_k.k->p.offset, 0, inode_k.k->p.snapshot),
-			POS(inode_k.k->p.offset, U64_MAX),
-			0, k, ret) {
-		if (k.k->type != KEY_TYPE_dirent)
-			continue;
-
-		struct bkey_s_c_dirent dirent = bkey_s_c_to_dirent(k);
-		struct qstr name = bch2_dirent_get_name(dirent);
-
-		new_size += dirent_occupied_size(&name);
-	}
-	bch2_trans_iter_exit(trans, &iter);
-
-	if (!ret && inode_u->bi_size != new_size) {
-		inode_u->bi_size = new_size;
-		*write_inode = true;
-	}
-
-	return ret;
-}
-
 static int check_inode(struct btree_trans *trans,
 		       struct btree_iter *iter,
 		       struct bkey_s_c k,
@@ -1335,16 +1250,6 @@ static int check_inode(struct btree_trans *trans,
 		u.bi_journal_seq = journal_cur_seq(&c->journal);
 		do_update = true;
 	}
-
-	if (S_ISDIR(u.bi_mode)) {
-		ret = check_directory_size(trans, &u, k, &do_update);
-
-		fsck_err_on(ret,
-			    trans, directory_size_mismatch,
-			    "directory inode %llu:%u with the mismatch directory size",
-			    u.bi_inum, k.k->p.snapshot);
-		ret = 0;
-	}
 do_update:
 	if (do_update) {
 		ret = __bch2_fsck_write_inode(trans, &u);
@@ -1516,14 +1421,14 @@ static int check_key_has_inode(struct btree_trans *trans,
 
 	if (fsck_err_on(!i,
 			trans, key_in_missing_inode,
-			"key in missing inode:\n  %s",
+			"key in missing inode:\n%s",
 			(printbuf_reset(&buf),
 			 bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
 		goto delete;
 
 	if (fsck_err_on(i && !btree_matches_i_mode(iter->btree_id, i->inode.bi_mode),
 			trans, key_in_wrong_inode_type,
-			"key for wrong inode mode %o:\n  %s",
+			"key for wrong inode mode %o:\n%s",
 			i->inode.bi_mode,
 			(printbuf_reset(&buf),
 			 bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
@@ -1666,13 +1571,13 @@ static int overlapping_extents_found(struct btree_trans *trans,
 	if (ret)
 		goto err;
 
-	prt_str(&buf, "\n  ");
+	prt_newline(&buf);
 	bch2_bkey_val_to_text(&buf, c, k1);
 
 	if (!bpos_eq(pos1, k1.k->p)) {
-		prt_str(&buf, "\n  wanted\n  ");
+		prt_str(&buf, "\nwanted\n  ");
 		bch2_bpos_to_text(&buf, pos1);
-		prt_str(&buf, "\n  ");
+		prt_str(&buf, "\n");
 		bch2_bkey_to_text(&buf, &pos2);
 
 		bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
@@ -1695,7 +1600,7 @@ static int overlapping_extents_found(struct btree_trans *trans,
 			break;
 	}
 
-	prt_str(&buf, "\n  ");
+	prt_newline(&buf);
 	bch2_bkey_val_to_text(&buf, c, k2);
 
 	if (bpos_gt(k2.k->p, pos2.p) ||
@@ -1706,7 +1611,7 @@ static int overlapping_extents_found(struct btree_trans *trans,
 		goto err;
 	}
 
-	prt_printf(&buf, "\n  overwriting %s extent",
+	prt_printf(&buf, "\noverwriting %s extent",
 		   pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
 
 	if (fsck_err(trans, extent_overlapping,
@@ -1727,6 +1632,8 @@ static int overlapping_extents_found(struct btree_trans *trans,
 			bch2_trans_commit(trans, &res, NULL, BCH_TRANS_COMMIT_no_enospc);
 		bch2_disk_reservation_put(c, &res);
 
+		bch_info(c, "repair ret %s", bch2_err_str(ret));
+
 		if (ret)
 			goto err;
 
@@ -1879,7 +1786,7 @@ static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
 			if (fsck_err_on(k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
 					!bkey_extent_is_reservation(k),
 					trans, extent_past_end_of_inode,
-					"extent type past end of inode %llu:%u, i_size %llu\n  %s",
+					"extent type past end of inode %llu:%u, i_size %llu\n%s",
 					i->inode.bi_inum, i->snapshot, i->inode.bi_size,
 					(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
 				struct btree_iter iter2;
@@ -2017,176 +1924,13 @@ fsck_err:
 	return ret;
 }
 
-static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
+static int check_subdir_dirents_count(struct btree_trans *trans, struct inode_walker *w)
 {
 	u32 restart_count = trans->restart_count;
 	return check_subdir_count_notnested(trans, w) ?:
 		trans_was_restarted(trans, restart_count);
 }
 
-noinline_for_stack
-static int check_dirent_inode_dirent(struct btree_trans *trans,
-				   struct btree_iter *iter,
-				   struct bkey_s_c_dirent d,
-				   struct bch_inode_unpacked *target)
-{
-	struct bch_fs *c = trans->c;
-	struct printbuf buf = PRINTBUF;
-	struct btree_iter bp_iter = { NULL };
-	int ret = 0;
-
-	if (inode_points_to_dirent(target, d))
-		return 0;
-
-	if (!target->bi_dir &&
-	    !target->bi_dir_offset) {
-		fsck_err_on(S_ISDIR(target->bi_mode),
-			    trans, inode_dir_missing_backpointer,
-			    "directory with missing backpointer\n%s",
-			    (printbuf_reset(&buf),
-			     bch2_bkey_val_to_text(&buf, c, d.s_c),
-			     prt_printf(&buf, "\n"),
-			     bch2_inode_unpacked_to_text(&buf, target),
-			     buf.buf));
-
-		fsck_err_on(target->bi_flags & BCH_INODE_unlinked,
-			    trans, inode_unlinked_but_has_dirent,
-			    "inode unlinked but has dirent\n%s",
-			    (printbuf_reset(&buf),
-			     bch2_bkey_val_to_text(&buf, c, d.s_c),
-			     prt_printf(&buf, "\n"),
-			     bch2_inode_unpacked_to_text(&buf, target),
-			     buf.buf));
-
-		target->bi_flags &= ~BCH_INODE_unlinked;
-		target->bi_dir		= d.k->p.inode;
-		target->bi_dir_offset	= d.k->p.offset;
-		return __bch2_fsck_write_inode(trans, target);
-	}
-
-	if (bch2_inode_should_have_single_bp(target) &&
-	    !fsck_err(trans, inode_wrong_backpointer,
-		      "dirent points to inode that does not point back:\n  %s",
-		      (bch2_bkey_val_to_text(&buf, c, d.s_c),
-		       prt_printf(&buf, "\n  "),
-		       bch2_inode_unpacked_to_text(&buf, target),
-		       buf.buf)))
-		goto err;
-
-	struct bkey_s_c_dirent bp_dirent = dirent_get_by_pos(trans, &bp_iter,
-			      SPOS(target->bi_dir, target->bi_dir_offset, target->bi_snapshot));
-	ret = bkey_err(bp_dirent);
-	if (ret && !bch2_err_matches(ret, ENOENT))
-		goto err;
-
-	bool backpointer_exists = !ret;
-	ret = 0;
-
-	if (fsck_err_on(!backpointer_exists,
-			trans, inode_wrong_backpointer,
-			"inode %llu:%u has wrong backpointer:\n"
-			"got       %llu:%llu\n"
-			"should be %llu:%llu",
-			target->bi_inum, target->bi_snapshot,
-			target->bi_dir,
-			target->bi_dir_offset,
-			d.k->p.inode,
-			d.k->p.offset)) {
-		target->bi_dir		= d.k->p.inode;
-		target->bi_dir_offset	= d.k->p.offset;
-		ret = __bch2_fsck_write_inode(trans, target);
-		goto out;
-	}
-
-	bch2_bkey_val_to_text(&buf, c, d.s_c);
-	prt_newline(&buf);
-	if (backpointer_exists)
-		bch2_bkey_val_to_text(&buf, c, bp_dirent.s_c);
-
-	if (fsck_err_on(backpointer_exists &&
-			(S_ISDIR(target->bi_mode) ||
-			 target->bi_subvol),
-			trans, inode_dir_multiple_links,
-			"%s %llu:%u with multiple links\n%s",
-			S_ISDIR(target->bi_mode) ? "directory" : "subvolume",
-			target->bi_inum, target->bi_snapshot, buf.buf)) {
-		ret = __remove_dirent(trans, d.k->p);
-		goto out;
-	}
-
-	/*
-	 * hardlinked file with nlink 0:
-	 * We're just adjusting nlink here so check_nlinks() will pick
-	 * it up, it ignores inodes with nlink 0
-	 */
-	if (fsck_err_on(backpointer_exists && !target->bi_nlink,
-			trans, inode_multiple_links_but_nlink_0,
-			"inode %llu:%u type %s has multiple links but i_nlink 0\n%s",
-			target->bi_inum, target->bi_snapshot, bch2_d_types[d.v->d_type], buf.buf)) {
-		target->bi_nlink++;
-		target->bi_flags &= ~BCH_INODE_unlinked;
-		ret = __bch2_fsck_write_inode(trans, target);
-		if (ret)
-			goto err;
-	}
-out:
-err:
-fsck_err:
-	bch2_trans_iter_exit(trans, &bp_iter);
-	printbuf_exit(&buf);
-	bch_err_fn(c, ret);
-	return ret;
-}
-
-noinline_for_stack
-static int check_dirent_target(struct btree_trans *trans,
-			       struct btree_iter *iter,
-			       struct bkey_s_c_dirent d,
-			       struct bch_inode_unpacked *target)
-{
-	struct bch_fs *c = trans->c;
-	struct bkey_i_dirent *n;
-	struct printbuf buf = PRINTBUF;
-	int ret = 0;
-
-	ret = check_dirent_inode_dirent(trans, iter, d, target);
-	if (ret)
-		goto err;
-
-	if (fsck_err_on(d.v->d_type != inode_d_type(target),
-			trans, dirent_d_type_wrong,
-			"incorrect d_type: got %s, should be %s:\n%s",
-			bch2_d_type_str(d.v->d_type),
-			bch2_d_type_str(inode_d_type(target)),
-			(printbuf_reset(&buf),
-			 bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
-		n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
-		ret = PTR_ERR_OR_ZERO(n);
-		if (ret)
-			goto err;
-
-		bkey_reassemble(&n->k_i, d.s_c);
-		n->v.d_type = inode_d_type(target);
-		if (n->v.d_type == DT_SUBVOL) {
-			n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
-			n->v.d_child_subvol = cpu_to_le32(target->bi_subvol);
-		} else {
-			n->v.d_inum = cpu_to_le64(target->bi_inum);
-		}
-
-		ret = bch2_trans_update(trans, iter, &n->k_i, 0);
-		if (ret)
-			goto err;
-
-		d = dirent_i_to_s_c(n);
-	}
-err:
-fsck_err:
-	printbuf_exit(&buf);
-	bch_err_fn(c, ret);
-	return ret;
-}
-
 /* find a subvolume that's a descendent of @snapshot: */
 static int find_snapshot_subvol(struct btree_trans *trans, u32 snapshot, u32 *subvolid)
 {
@@ -2286,7 +2030,7 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *
 		if (fsck_err(trans, dirent_to_missing_subvol,
 			     "dirent points to missing subvolume\n%s",
 			     (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)))
-			return __remove_dirent(trans, d.k->p);
+			return bch2_fsck_remove_dirent(trans, d.k->p);
 		ret = 0;
 		goto out;
 	}
@@ -2330,7 +2074,7 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *
 			goto err;
 	}
 
-	ret = check_dirent_target(trans, iter, d, &subvol_root);
+	ret = bch2_check_dirent_target(trans, iter, d, &subvol_root, true);
 	if (ret)
 		goto err;
 out:
@@ -2367,7 +2111,7 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 		goto out;
 
 	if (dir->last_pos.inode != k.k->p.inode && dir->have_inodes) {
-		ret = check_subdir_count(trans, dir);
+		ret = check_subdir_dirents_count(trans, dir);
 		if (ret)
 			goto err;
 	}
@@ -2417,13 +2161,13 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 				(printbuf_reset(&buf),
 				 bch2_bkey_val_to_text(&buf, c, k),
 				 buf.buf))) {
-			ret = __remove_dirent(trans, d.k->p);
+			ret = bch2_fsck_remove_dirent(trans, d.k->p);
 			if (ret)
 				goto err;
 		}
 
 		darray_for_each(target->inodes, i) {
-			ret = check_dirent_target(trans, iter, d, &i->inode);
+			ret = bch2_check_dirent_target(trans, iter, d, &i->inode, true);
 			if (ret)
 				goto err;
 		}
@@ -2457,9 +2201,11 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 	if (ret)
 		goto err;
 
-	if (d.v->d_type == DT_DIR)
-		for_each_visible_inode(c, s, dir, d.k->p.snapshot, i)
+	for_each_visible_inode(c, s, dir, d.k->p.snapshot, i) {
+		if (d.v->d_type == DT_DIR)
 			i->count++;
+		i->i_size += bkey_bytes(d.k);
+	}
 out:
 err:
 fsck_err:
@@ -3277,7 +3023,7 @@ long bch2_ioctl_fsck_offline(struct bch_ioctl_fsck_offline __user *user_arg)
 	if (arg.opts) {
 		char *optstr = strndup_user((char __user *)(unsigned long) arg.opts, 1 << 16);
 		ret =   PTR_ERR_OR_ZERO(optstr) ?:
-			bch2_parse_mount_opts(NULL, &thr->opts, NULL, optstr);
+			bch2_parse_mount_opts(NULL, &thr->opts, NULL, optstr, false);
 		if (!IS_ERR(optstr))
 			kfree(optstr);
 
@@ -3385,7 +3131,7 @@ long bch2_ioctl_fsck_online(struct bch_fs *c, struct bch_ioctl_fsck_online arg)
 		char *optstr = strndup_user((char __user *)(unsigned long) arg.opts, 1 << 16);
 
 		ret =   PTR_ERR_OR_ZERO(optstr) ?:
-			bch2_parse_mount_opts(c, &thr->opts, NULL, optstr);
+			bch2_parse_mount_opts(c, &thr->opts, NULL, optstr, false);
 		if (!IS_ERR(optstr))
 			kfree(optstr);
 

libbcachefs/inode.c

@@ -731,10 +731,9 @@ int bch2_trigger_inode(struct btree_trans *trans,
 		bkey_s_to_inode_v3(new).v->bi_journal_seq = cpu_to_le64(trans->journal_res.seq);
 	}
 
-	s64 nr = bkey_is_inode(new.k) - bkey_is_inode(old.k);
-	if ((flags & (BTREE_TRIGGER_transactional|BTREE_TRIGGER_gc)) && nr) {
-		struct disk_accounting_pos acc = { .type = BCH_DISK_ACCOUNTING_nr_inodes };
-		int ret = bch2_disk_accounting_mod(trans, &acc, &nr, 1, flags & BTREE_TRIGGER_gc);
+	s64 nr[1] = { bkey_is_inode(new.k) - bkey_is_inode(old.k) };
+	if ((flags & (BTREE_TRIGGER_transactional|BTREE_TRIGGER_gc)) && nr[0]) {
+		int ret = bch2_disk_accounting_mod2(trans, flags & BTREE_TRIGGER_gc, nr, nr_inodes);
 		if (ret)
 			return ret;
 	}
@@ -868,19 +867,6 @@ void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
 			     uid, gid, mode, rdev, parent);
 }
 
-static inline u32 bkey_generation(struct bkey_s_c k)
-{
-	switch (k.k->type) {
-	case KEY_TYPE_inode:
-	case KEY_TYPE_inode_v2:
-		BUG();
-	case KEY_TYPE_inode_generation:
-		return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
-	default:
-		return 0;
-	}
-}
-
 static struct bkey_i_inode_alloc_cursor *
 bch2_inode_alloc_cursor_get(struct btree_trans *trans, u64 cpu, u64 *min, u64 *max)
 {
@@ -1092,7 +1078,7 @@ retry:
 		bch2_fs_inconsistent(c,
 				     "inode %llu:%u not found when deleting",
 				     inum.inum, snapshot);
-		ret = -EIO;
+		ret = -BCH_ERR_ENOENT_inode;
 		goto err;
 	}
 
@@ -1198,6 +1184,7 @@ void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
 		opts->_name##_from_inode = true;			\
 	} else {							\
 		opts->_name = c->opts._name;				\
+		opts->_name##_from_inode = false;			\
 	}
 	BCH_INODE_OPTS()
 #undef x
@@ -1255,7 +1242,7 @@ retry:
 		bch2_fs_inconsistent(c,
 				     "inode %llu:%u not found when deleting",
 				     inum, snapshot);
-		ret = -EIO;
+		ret = -BCH_ERR_ENOENT_inode;
 		goto err;
 	}
 

libbcachefs/inode.h

@@ -277,6 +277,7 @@ static inline bool bch2_inode_should_have_single_bp(struct bch_inode_unpacked *i
 	bool inode_has_bp = inode->bi_dir || inode->bi_dir_offset;
 
 	return S_ISDIR(inode->bi_mode) ||
+		inode->bi_subvol ||
 		(!inode->bi_nlink && inode_has_bp);
 }
 

libbcachefs/inode_format.h

@@ -137,7 +137,8 @@ enum inode_opt_id {
 	x(i_sectors_dirty,		6)	\
 	x(unlinked,			7)	\
 	x(backptr_untrusted,		8)	\
-	x(has_child_snapshot,		9)
+	x(has_child_snapshot,		9)	\
+	x(casefolded,			10)
 
 /* bits 20+ reserved for packed fields below: */
 

libbcachefs/io_misc.c

@@ -115,7 +115,8 @@ err:
 		bch2_increment_clock(c, sectors_allocated, WRITE);
 	if (should_print_err(ret)) {
 		struct printbuf buf = PRINTBUF;
-		bch2_inum_offset_err_msg_trans(trans, &buf, inum, iter->pos.offset << 9);
+		lockrestart_do(trans,
+			bch2_inum_offset_err_msg_trans(trans, &buf, inum, iter->pos.offset << 9));
 		prt_printf(&buf, "fallocate error: %s", bch2_err_str(ret));
 		bch_err_ratelimited(c, "%s", buf.buf);
 		printbuf_exit(&buf);