Update bcachefs sources to 1dbc147a6eb0 bcachefs: Add version check for bch_btree_ptr_v2.sectors_written validate

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

.bcachefs_revision

@@ -1 +1 @@
-9ab6c94002b9def79b0d79d0efd4e5255100feb4
+1dbc147a6eb08a0137a81a1fe4eb528186594bfe

c_src/posix_to_bcachefs.c

@@ -32,7 +32,7 @@ void create_link(struct bch_fs *c,
 	struct bch_inode_unpacked parent_u;
 	struct bch_inode_unpacked inode;
 
-	int ret = bch2_trans_do(c, NULL, NULL, 0,
+	int ret = bch2_trans_commit_do(c, NULL, NULL, 0,
 		bch2_link_trans(trans,
 				(subvol_inum) { 1, parent->bi_inum }, &parent_u,
 				(subvol_inum) { 1, inum }, &inode, &qstr));
@@ -51,7 +51,7 @@ struct bch_inode_unpacked create_file(struct bch_fs *c,
 
 	bch2_inode_init_early(c, &new_inode);
 
-	int ret = bch2_trans_do(c, NULL, NULL, 0,
+	int ret = bch2_trans_commit_do(c, NULL, NULL, 0,
 		bch2_create_trans(trans,
 				  (subvol_inum) { 1, parent->bi_inum }, parent,
 				  &new_inode, &qstr,
@@ -125,7 +125,7 @@ void copy_xattrs(struct bch_fs *c, struct bch_inode_unpacked *dst,
 		if (IS_ERR(h))
 			continue;
 
-		int ret = bch2_trans_do(c, NULL, NULL, 0,
+		int ret = bch2_trans_commit_do(c, NULL, NULL, 0,
 				bch2_xattr_set(trans,
 					       (subvol_inum) { 1, dst->bi_inum },
 					       dst, &hash_info, attr,

include/linux/slab.h

@@ -111,6 +111,7 @@ static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
 #define kzfree(p)			free((void *) p)
 
 #define kvmalloc(size, flags)		kmalloc(size, flags)
+#define kvmalloc_noprof(size, flags)	kmalloc(size, flags)
 #define kvzalloc(size, flags)		kzalloc(size, flags)
 #define kvfree(p)			kfree(p)
 
@@ -274,6 +275,8 @@ static inline void *vmalloc(unsigned long size)
 	return __vmalloc(size, GFP_KERNEL);
 }
 
+#define vmalloc_noprof(...)	vmalloc(__VA_ARGS__)
+
 static inline void *vzalloc(unsigned long size)
 {
 	return __vmalloc(size, GFP_KERNEL|__GFP_ZERO);

include/linux/wait.h

@@ -96,6 +96,7 @@ do {									\
 #define wait_event_freezable(wq, condition)	({wait_event(wq, condition); 0; })
 #define wait_event_killable(wq, condition)	({wait_event(wq, condition); 0; })
 #define wait_event_interruptible(wq, condition)	({wait_event(wq, condition); 0; })
+#define wait_event_state(wq, condition, state)	({wait_event(wq, condition); 0; })
 
 #define __wait_event_timeout(wq, condition, timeout)			\
 	___wait_event(wq, ___wait_cond_timeout(condition),		\

libbcachefs/alloc_background.c

@@ -639,6 +639,16 @@ int bch2_alloc_read(struct bch_fs *c)
 				continue;
 			}
 
+			if (k.k->p.offset < ca->mi.first_bucket) {
+				bch2_btree_iter_set_pos(&iter, POS(k.k->p.inode, ca->mi.first_bucket));
+				continue;
+			}
+
+			if (k.k->p.offset >= ca->mi.nbuckets) {
+				bch2_btree_iter_set_pos(&iter, POS(k.k->p.inode + 1, 0));
+				continue;
+			}
+
 			struct bch_alloc_v4 a;
 			*bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen;
 			0;
@@ -1967,7 +1977,7 @@ static void bch2_do_discards_fast_work(struct work_struct *work)
 					     ca->mi.bucket_size,
 					     GFP_KERNEL);
 
-		int ret = bch2_trans_do(c, NULL, NULL,
+		int ret = bch2_trans_commit_do(c, NULL, NULL,
 			BCH_WATERMARK_btree|
 			BCH_TRANS_COMMIT_no_enospc,
 			bch2_clear_bucket_needs_discard(trans, POS(ca->dev_idx, bucket)));
@@ -2127,14 +2137,15 @@ static void bch2_do_invalidates_work(struct work_struct *work)
 
 		struct bkey_s_c k = next_lru_key(trans, &iter, ca, &wrapped);
 		ret = bkey_err(k);
-		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
-			continue;
 		if (ret)
-			break;
+			goto restart_err;
 		if (!k.k)
 			break;
 
 		ret = invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate);
+restart_err:
+		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+			continue;
 		if (ret)
 			break;
 
@@ -2340,24 +2351,19 @@ int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
 
 /* Bucket IO clocks: */
 
-int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
+static int __bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
-			      size_t bucket_nr, int rw)
+				size_t bucket_nr, int rw)
 {
 	struct bch_fs *c = trans->c;
+
 	struct btree_iter iter;
-	struct bkey_i_alloc_v4 *a;
+	struct bkey_i_alloc_v4 *a =
-	u64 now;
+		bch2_trans_start_alloc_update_noupdate(trans, &iter, POS(dev, bucket_nr));
-	int ret = 0;
+	int ret = PTR_ERR_OR_ZERO(a);
-
-	if (bch2_trans_relock(trans))
-		bch2_trans_begin(trans);
-
-	a = bch2_trans_start_alloc_update_noupdate(trans, &iter, POS(dev, bucket_nr));
-	ret = PTR_ERR_OR_ZERO(a);
 	if (ret)
 		return ret;
 
-	now = bch2_current_io_time(c, rw);
+	u64 now = bch2_current_io_time(c, rw);
 	if (a->v.io_time[rw] == now)
 		goto out;
 
@@ -2370,6 +2376,15 @@ out:
 	return ret;
 }
 
+int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
+			      size_t bucket_nr, int rw)
+{
+	if (bch2_trans_relock(trans))
+		bch2_trans_begin(trans);
+
+	return nested_lockrestart_do(trans, __bch2_bucket_io_time_reset(trans, dev, bucket_nr, rw));
+}
+
 /* Startup/shutdown (ro/rw): */
 
 void bch2_recalc_capacity(struct bch_fs *c)

libbcachefs/alloc_foreground.c

@@ -684,7 +684,7 @@ struct open_bucket *bch2_bucket_alloc(struct bch_fs *c, struct bch_dev *ca,
 	struct bch_dev_usage usage;
 	struct open_bucket *ob;
 
-	bch2_trans_do(c, NULL, NULL, 0,
+	bch2_trans_do(c,
 		      PTR_ERR_OR_ZERO(ob = bch2_bucket_alloc_trans(trans, ca, watermark,
 							data_type, cl, false, &usage)));
 	return ob;
@@ -1610,8 +1610,7 @@ void bch2_open_buckets_to_text(struct printbuf *out, struct bch_fs *c,
 	     ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
 	     ob++) {
 		spin_lock(&ob->lock);
-		if (ob->valid && !ob->on_partial_list &&
+		if (ob->valid && (!ca || ob->dev == ca->dev_idx))
-		    (!ca || ob->dev == ca->dev_idx))
 			bch2_open_bucket_to_text(out, c, ob);
 		spin_unlock(&ob->lock);
 	}

libbcachefs/btree_gc.c

@@ -829,12 +829,22 @@ static int bch2_alloc_write_key(struct btree_trans *trans,
 	 * fix that here:
 	 */
 	alloc_data_type_set(&gc, gc.data_type);
-
 	if (gc.data_type != old_gc.data_type ||
 	    gc.dirty_sectors != old_gc.dirty_sectors) {
 		ret = bch2_alloc_key_to_dev_counters(trans, ca, &old_gc, &gc, BTREE_TRIGGER_gc);
 		if (ret)
 			return ret;
+
+		/*
+		 * Ugly: alloc_key_to_dev_counters(..., BTREE_TRIGGER_gc) is not
+		 * safe w.r.t. transaction restarts, so fixup the gc_bucket so
+		 * we don't run it twice:
+		 */
+		percpu_down_read(&c->mark_lock);
+		struct bucket *gc_m = gc_bucket(ca, iter->pos.offset);
+		gc_m->data_type = gc.data_type;
+		gc_m->dirty_sectors = gc.dirty_sectors;
+		percpu_up_read(&c->mark_lock);
 	}
 
 	if (fsck_err_on(new.data_type != gc.data_type,

libbcachefs/btree_io.c

@@ -1872,7 +1872,7 @@ static void btree_node_write_work(struct work_struct *work)
 
 		}
 	} else {
-		ret = bch2_trans_do(c, NULL, NULL, 0,
+		ret = bch2_trans_do(c,
 			bch2_btree_node_update_key_get_iter(trans, b, &wbio->key,
 					BCH_WATERMARK_interior_updates|
 					BCH_TRANS_COMMIT_journal_reclaim|

libbcachefs/btree_iter.c

@@ -1414,9 +1414,17 @@ void __noreturn bch2_trans_restart_error(struct btree_trans *trans, u32 restart_
 
 void __noreturn bch2_trans_in_restart_error(struct btree_trans *trans)
 {
+#ifdef CONFIG_BCACHEFS_DEBUG
+	struct printbuf buf = PRINTBUF;
+	bch2_prt_backtrace(&buf, &trans->last_restarted_trace);
+	panic("in transaction restart: %s, last restarted by\n%s",
+	      bch2_err_str(trans->restarted),
+	      buf.buf);
+#else
 	panic("in transaction restart: %s, last restarted by %pS\n",
 	      bch2_err_str(trans->restarted),
 	      (void *) trans->last_restarted_ip);
+#endif
 }
 
 void __noreturn bch2_trans_unlocked_error(struct btree_trans *trans)
@@ -2292,12 +2300,6 @@ struct bkey_s_c bch2_btree_iter_peek_upto(struct btree_iter *iter, struct bpos e
 
 	bch2_btree_iter_verify_entry_exit(iter);
 
-	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(iter, search_key);
 		if (unlikely(!k.k))
@@ -2467,12 +2469,6 @@ struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *iter)
 	if (iter->flags & BTREE_ITER_with_journal)
 		return bkey_s_c_err(-BCH_ERR_btree_iter_with_journal_not_supported);
 
-	ret = trans_maybe_inject_restart(trans, _RET_IP_);
-	if (unlikely(ret)) {
-		k = bkey_s_c_err(ret);
-		goto out_no_locked;
-	}
-
 	bch2_btree_iter_verify(iter);
 	bch2_btree_iter_verify_entry_exit(iter);
 
@@ -2610,12 +2606,6 @@ 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)) {
@@ -2961,10 +2951,6 @@ 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);
 
@@ -3022,8 +3008,7 @@ 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_ip(trans,
+		return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_mem_realloced));
-					BCH_ERR_transaction_restart_mem_realloced, _RET_IP_));
 	}
 out_change_top:
 	p = trans->mem + trans->mem_top;
@@ -3131,14 +3116,6 @@ 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);
 
 	if (trans->restarted) {
@@ -3279,6 +3256,9 @@ void bch2_trans_put(struct btree_trans *trans)
 {
 	struct bch_fs *c = trans->c;
 
+	if (trans->restarted)
+		bch2_trans_in_restart_error(trans);
+
 	bch2_trans_unlock(trans);
 
 	trans_for_each_update(trans, i)
@@ -3302,6 +3282,10 @@ void bch2_trans_put(struct btree_trans *trans)
 	closure_return_sync(&trans->ref);
 	trans->locking_wait.task = NULL;
 
+#ifdef CONFIG_BCACHEFS_DEBUG
+	darray_exit(&trans->last_restarted_trace);
+#endif
+
 	unsigned long *paths_allocated = trans->paths_allocated;
 	trans->paths_allocated	= NULL;
 	trans->paths		= NULL;

libbcachefs/btree_iter.h

@@ -350,6 +350,10 @@ static int btree_trans_restart_ip(struct btree_trans *trans, int err, unsigned l
 
 	trans->restarted = err;
 	trans->last_restarted_ip = ip;
+#ifdef CONFIG_BCACHEFS_DEBUG
+	darray_exit(&trans->last_restarted_trace);
+	bch2_save_backtrace(&trans->last_restarted_trace, current, 0, GFP_KERNEL);
+#endif
 	return -err;
 }
 
@@ -359,18 +363,6 @@ 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);
 
@@ -923,6 +915,8 @@ struct bkey_s_c bch2_btree_iter_peek_and_restart_outlined(struct btree_iter *);
 	_ret;								\
 })
 
+#define bch2_trans_do(_c, _do)	bch2_trans_run(_c, lockrestart_do(trans, _do))
+
 struct btree_trans *__bch2_trans_get(struct bch_fs *, unsigned);
 void bch2_trans_put(struct btree_trans *);
 

libbcachefs/btree_trans_commit.c

@@ -1027,10 +1027,6 @@ int __bch2_trans_commit(struct btree_trans *trans, unsigned flags)
 	bch2_trans_verify_not_unlocked(trans);
 	bch2_trans_verify_not_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

@@ -509,13 +509,13 @@ 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;
 	unsigned long		last_restarted_ip;
+#ifdef CONFIG_BCACHEFS_DEBUG
+	bch_stacktrace		last_restarted_trace;
+#endif
 	unsigned long		last_unlock_ip;
 	unsigned long		srcu_lock_time;
 

libbcachefs/btree_update.c

@@ -668,7 +668,7 @@ int bch2_btree_insert(struct bch_fs *c, enum btree_id id, struct bkey_i *k,
 		      struct disk_reservation *disk_res, int flags,
 		      enum btree_iter_update_trigger_flags iter_flags)
 {
-	return bch2_trans_do(c, disk_res, NULL, flags,
+	return bch2_trans_commit_do(c, disk_res, NULL, flags,
 			     bch2_btree_insert_trans(trans, id, k, iter_flags));
 }
 
@@ -865,7 +865,7 @@ __bch2_fs_log_msg(struct bch_fs *c, unsigned commit_flags, const char *fmt,
 		memcpy(l->d, buf.buf, buf.pos);
 		c->journal.early_journal_entries.nr += jset_u64s(u64s);
 	} else {
-		ret = bch2_trans_do(c, NULL, NULL,
+		ret = bch2_trans_commit_do(c, NULL, NULL,
 			BCH_TRANS_COMMIT_lazy_rw|commit_flags,
 			__bch2_trans_log_msg(trans, &buf, u64s));
 	}

libbcachefs/btree_update.h

@@ -192,7 +192,7 @@ static inline int bch2_trans_commit(struct btree_trans *trans,
 	nested_lockrestart_do(_trans, _do ?: bch2_trans_commit(_trans, (_disk_res),\
 					(_journal_seq), (_flags)))
 
-#define bch2_trans_do(_c, _disk_res, _journal_seq, _flags, _do)		\
+#define bch2_trans_commit_do(_c, _disk_res, _journal_seq, _flags, _do)		\
 	bch2_trans_run(_c, commit_do(trans, _disk_res, _journal_seq, _flags, _do))
 
 #define trans_for_each_update(_trans, _i)				\

libbcachefs/btree_update_interior.c

@@ -2239,10 +2239,8 @@ 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;
 
-	ret = bch2_trans_do(c, NULL, NULL, 0,
+	int ret = bch2_trans_do(c, async_btree_node_rewrite_trans(trans, a));
-		      async_btree_node_rewrite_trans(trans, a));
 	bch_err_fn_ratelimited(c, ret);
 	bch2_write_ref_put(c, BCH_WRITE_REF_node_rewrite);
 	kfree(a);

libbcachefs/buckets.c

@@ -1160,11 +1160,11 @@ int bch2_trans_mark_dev_sbs(struct bch_fs *c)
 #define SECTORS_CACHE	1024
 
 int __bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
-			      u64 sectors, int flags)
+				u64 sectors, enum bch_reservation_flags flags)
 {
 	struct bch_fs_pcpu *pcpu;
 	u64 old, get;
-	s64 sectors_available;
+	u64 sectors_available;
 	int ret;
 
 	percpu_down_read(&c->mark_lock);
@@ -1202,6 +1202,9 @@ recalculate:
 	percpu_u64_set(&c->pcpu->sectors_available, 0);
 	sectors_available = avail_factor(__bch2_fs_usage_read_short(c).free);
 
+	if (sectors_available && (flags & BCH_DISK_RESERVATION_PARTIAL))
+		sectors = min(sectors, sectors_available);
+
 	if (sectors <= sectors_available ||
 	    (flags & BCH_DISK_RESERVATION_NOFAIL)) {
 		atomic64_set(&c->sectors_available,

libbcachefs/buckets.h

@@ -344,14 +344,16 @@ static inline void bch2_disk_reservation_put(struct bch_fs *c,
 	}
 }
 
-#define BCH_DISK_RESERVATION_NOFAIL		(1 << 0)
+enum bch_reservation_flags {
+	BCH_DISK_RESERVATION_NOFAIL	= 1 << 0,
+	BCH_DISK_RESERVATION_PARTIAL	= 1 << 1,
+};
 
-int __bch2_disk_reservation_add(struct bch_fs *,
+int __bch2_disk_reservation_add(struct bch_fs *, struct disk_reservation *,
-				struct disk_reservation *,
+				u64, enum bch_reservation_flags);
-				u64, int);
 
 static inline int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
-					    u64 sectors, int flags)
+					    u64 sectors, enum bch_reservation_flags flags)
 {
 #ifdef __KERNEL__
 	u64 old, new;

libbcachefs/chardev.c

@@ -225,6 +225,7 @@ static long bch2_ioctl_fsck_offline(struct bch_ioctl_fsck_offline __user *user_a
 
 	opt_set(thr->opts, stdio, (u64)(unsigned long)&thr->thr.stdio);
 	opt_set(thr->opts, read_only, 1);
+	opt_set(thr->opts, ratelimit_errors, 0);
 
 	/* We need request_key() to be called before we punt to kthread: */
 	opt_set(thr->opts, nostart, true);

libbcachefs/darray.c

@@ -2,6 +2,7 @@
 
 #include <linux/log2.h>
 #include <linux/slab.h>
+#include <linux/vmalloc.h>
 #include "darray.h"
 
 int __bch2_darray_resize_noprof(darray_char *d, size_t element_size, size_t new_size, gfp_t gfp)
@@ -9,7 +10,19 @@ int __bch2_darray_resize_noprof(darray_char *d, size_t element_size, size_t new_
 	if (new_size > d->size) {
 		new_size = roundup_pow_of_two(new_size);
 
-		void *data = kvmalloc_array_noprof(new_size, element_size, gfp);
+		/*
+		 * This is a workaround: kvmalloc() doesn't support > INT_MAX
+		 * allocations, but vmalloc() does.
+		 * The limit needs to be lifted from kvmalloc, and when it does
+		 * we'll go back to just using that.
+		 */
+		size_t bytes;
+		if (unlikely(check_mul_overflow(new_size, element_size, &bytes)))
+			return -ENOMEM;
+
+		void *data = likely(bytes < INT_MAX)
+			? kvmalloc_noprof(bytes, gfp)
+			: vmalloc_noprof(bytes);
 		if (!data)
 			return -ENOMEM;
 

libbcachefs/data_update.c

@@ -80,6 +80,7 @@ static bool bkey_nocow_lock(struct bch_fs *c, struct moving_context *ctxt, struc
 					if (ptr2 == ptr)
 						break;
 
+					ca = bch2_dev_have_ref(c, ptr2->dev);
 					bucket = PTR_BUCKET_POS(ca, ptr2);
 					bch2_bucket_nocow_unlock(&c->nocow_locks, bucket, 0);
 				}

libbcachefs/dirent.c

@@ -250,13 +250,6 @@ int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
 	return ret;
 }
 
-static void dirent_copy_target(struct bkey_i_dirent *dst,
-			       struct bkey_s_c_dirent src)
-{
-	dst->v.d_inum = src.v->d_inum;
-	dst->v.d_type = src.v->d_type;
-}
-
 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
 			    struct bkey_s_c_dirent d, subvol_inum *target)
 {

libbcachefs/dirent.h

@@ -34,6 +34,13 @@ static inline unsigned dirent_val_u64s(unsigned len)
 int bch2_dirent_read_target(struct btree_trans *, subvol_inum,
 			    struct bkey_s_c_dirent, subvol_inum *);
 
+static inline void dirent_copy_target(struct bkey_i_dirent *dst,
+				      struct bkey_s_c_dirent src)
+{
+	dst->v.d_inum = src.v->d_inum;
+	dst->v.d_type = src.v->d_type;
+}
+
 int bch2_dirent_create_snapshot(struct btree_trans *, u32, u64, u32,
 			const struct bch_hash_info *, u8,
 			const struct qstr *, u64, u64 *,

libbcachefs/disk_accounting.c

@@ -244,10 +244,10 @@ void bch2_accounting_swab(struct bkey_s k)
 }
 
 static inline void __accounting_to_replicas(struct bch_replicas_entry_v1 *r,
-					    struct disk_accounting_pos acc)
+					    struct disk_accounting_pos *acc)
 {
-	unsafe_memcpy(r, &acc.replicas,
+	unsafe_memcpy(r, &acc->replicas,
-		      replicas_entry_bytes(&acc.replicas),
+		      replicas_entry_bytes(&acc->replicas),
 		      "variable length struct");
 }
 
@@ -258,7 +258,7 @@ static inline bool accounting_to_replicas(struct bch_replicas_entry_v1 *r, struc
 
 	switch (acc_k.type) {
 	case BCH_DISK_ACCOUNTING_replicas:
-		__accounting_to_replicas(r, acc_k);
+		__accounting_to_replicas(r, &acc_k);
 		return true;
 	default:
 		return false;
@@ -626,7 +626,7 @@ static int bch2_disk_accounting_validate_late(struct btree_trans *trans,
 	switch (acc.type) {
 	case BCH_DISK_ACCOUNTING_replicas: {
 		struct bch_replicas_padded r;
-		__accounting_to_replicas(&r.e, acc);
+		__accounting_to_replicas(&r.e, &acc);
 
 		for (unsigned i = 0; i < r.e.nr_devs; i++)
 			if (r.e.devs[i] != BCH_SB_MEMBER_INVALID &&
@@ -857,8 +857,10 @@ int bch2_dev_usage_init(struct bch_dev *ca, bool gc)
 	};
 	u64 v[3] = { ca->mi.nbuckets - ca->mi.first_bucket, 0, 0 };
 
-	int ret = bch2_trans_do(c, NULL, NULL, 0,
+	int ret = bch2_trans_do(c, ({
-			bch2_disk_accounting_mod(trans, &acc, v, ARRAY_SIZE(v), gc));
+		bch2_disk_accounting_mod(trans, &acc, v, ARRAY_SIZE(v), gc) ?:
+		(!gc ? bch2_trans_commit(trans, NULL, NULL, 0) : 0);
+	}));
 	bch_err_fn(c, ret);
 	return ret;
 }

libbcachefs/ec.c

@@ -124,6 +124,11 @@ int bch2_stripe_validate(struct bch_fs *c, struct bkey_s_c k,
 			 "incorrect value size (%zu < %u)",
 			 bkey_val_u64s(k.k), stripe_val_u64s(s));
 
+	bkey_fsck_err_on(s->csum_granularity_bits >= 64,
+			 c, stripe_csum_granularity_bad,
+			 "invalid csum granularity (%u >= 64)",
+			 s->csum_granularity_bits);
+
 	ret = bch2_bkey_ptrs_validate(c, k, flags);
 fsck_err:
 	return ret;
@@ -145,7 +150,11 @@ void bch2_stripe_to_text(struct printbuf *out, struct bch_fs *c,
 		   nr_data,
 		   s.nr_redundant);
 	bch2_prt_csum_type(out, s.csum_type);
-	prt_printf(out, " gran %u", 1U << s.csum_granularity_bits);
+	prt_str(out, " gran ");
+	if (s.csum_granularity_bits < 64)
+		prt_printf(out, "%llu", 1ULL << s.csum_granularity_bits);
+	else
+		prt_printf(out, "(invalid shift %u)", s.csum_granularity_bits);
 
 	if (s.disk_label) {
 		prt_str(out, " label");
@@ -257,12 +266,12 @@ static int __mark_stripe_bucket(struct btree_trans *trans,
 	if (!deleting) {
 		a->stripe		= s.k->p.offset;
 		a->stripe_redundancy	= s.v->nr_redundant;
+		alloc_data_type_set(a, data_type);
 	} else {
 		a->stripe		= 0;
 		a->stripe_redundancy	= 0;
+		alloc_data_type_set(a, BCH_DATA_user);
 	}
-
-	alloc_data_type_set(a, data_type);
 err:
 	printbuf_exit(&buf);
 	return ret;
@@ -1177,7 +1186,7 @@ static void ec_stripe_delete_work(struct work_struct *work)
 		if (!idx)
 			break;
 
-		int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+		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)
@@ -1197,47 +1206,62 @@ void bch2_do_stripe_deletes(struct bch_fs *c)
 /* stripe creation: */
 
 static int ec_stripe_key_update(struct btree_trans *trans,
-				struct bkey_i_stripe *new,
+				struct bkey_i_stripe *old,
-				bool create)
+				struct bkey_i_stripe *new)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter iter;
+	bool create = !old;
-	struct bkey_s_c k;
-	int ret;
 
-	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_stripes,
+	struct btree_iter iter;
-			       new->k.p, BTREE_ITER_intent);
+	struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_stripes,
-	ret = bkey_err(k);
+					       new->k.p, BTREE_ITER_intent);
+	int ret = bkey_err(k);
 	if (ret)
 		goto err;
 
-	if (k.k->type != (create ? KEY_TYPE_deleted : KEY_TYPE_stripe)) {
+	if (bch2_fs_inconsistent_on(k.k->type != (create ? KEY_TYPE_deleted : KEY_TYPE_stripe),
-		bch2_fs_inconsistent(c, "error %s stripe: got existing key type %s",
+				    c, "error %s stripe: got existing key type %s",
-				     create ? "creating" : "updating",
+				    create ? "creating" : "updating",
-				     bch2_bkey_types[k.k->type]);
+				    bch2_bkey_types[k.k->type])) {
 		ret = -EINVAL;
 		goto err;
 	}
 
 	if (k.k->type == KEY_TYPE_stripe) {
-		const struct bch_stripe *old = bkey_s_c_to_stripe(k).v;
+		const struct bch_stripe *v = bkey_s_c_to_stripe(k).v;
-		unsigned i;
 
-		if (old->nr_blocks != new->v.nr_blocks) {
+		BUG_ON(old->v.nr_blocks != new->v.nr_blocks);
-			bch_err(c, "error updating stripe: nr_blocks does not match");
+		BUG_ON(old->v.nr_blocks != v->nr_blocks);
-			ret = -EINVAL;
-			goto err;
-		}
 
-		for (i = 0; i < new->v.nr_blocks; i++) {
+		for (unsigned i = 0; i < new->v.nr_blocks; i++) {
-			unsigned v = stripe_blockcount_get(old, i);
+			unsigned sectors = stripe_blockcount_get(v, i);
 
-			BUG_ON(v &&
+			if (!bch2_extent_ptr_eq(old->v.ptrs[i], new->v.ptrs[i]) && sectors) {
-			       (old->ptrs[i].dev != new->v.ptrs[i].dev ||
+				struct printbuf buf = PRINTBUF;
-				old->ptrs[i].gen != new->v.ptrs[i].gen ||
-				old->ptrs[i].offset != new->v.ptrs[i].offset));
 
-			stripe_blockcount_set(&new->v, i, v);
+				prt_printf(&buf, "stripe changed nonempty block %u", i);
+				prt_str(&buf, "\nold: ");
+				bch2_bkey_val_to_text(&buf, c, k);
+				prt_str(&buf, "\nnew: ");
+				bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&new->k_i));
+				bch2_fs_inconsistent(c, "%s", buf.buf);
+				printbuf_exit(&buf);
+				ret = -EINVAL;
+				goto err;
+			}
+
+			/*
+			 * If the stripe ptr changed underneath us, it must have
+			 * been dev_remove_stripes() -> * invalidate_stripe_to_dev()
+			 */
+			if (!bch2_extent_ptr_eq(old->v.ptrs[i], v->ptrs[i])) {
+				BUG_ON(v->ptrs[i].dev != BCH_SB_MEMBER_INVALID);
+
+				if (bch2_extent_ptr_eq(old->v.ptrs[i], new->v.ptrs[i]))
+					new->v.ptrs[i].dev = BCH_SB_MEMBER_INVALID;
+			}
+
+			stripe_blockcount_set(&new->v, i, sectors);
 		}
 	}
 
@@ -1495,12 +1519,14 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 		goto err;
 	}
 
-	ret = bch2_trans_do(c, &s->res, NULL,
+	ret = bch2_trans_commit_do(c, &s->res, NULL,
-			    BCH_TRANS_COMMIT_no_check_rw|
+		BCH_TRANS_COMMIT_no_check_rw|
-			    BCH_TRANS_COMMIT_no_enospc,
+		BCH_TRANS_COMMIT_no_enospc,
-			    ec_stripe_key_update(trans,
+		ec_stripe_key_update(trans,
-					bkey_i_to_stripe(&s->new_stripe.key),
+				     s->have_existing_stripe
-					!s->have_existing_stripe));
+				     ? bkey_i_to_stripe(&s->existing_stripe.key)
+				     : NULL,
+				     bkey_i_to_stripe(&s->new_stripe.key)));
 	bch_err_msg(c, ret, "creating stripe key");
 	if (ret) {
 		goto err;
@@ -1874,7 +1900,15 @@ static int new_stripe_alloc_buckets(struct btree_trans *trans,
 	bitmap_and(devs.d, devs.d, c->rw_devs[BCH_DATA_user].d, BCH_SB_MEMBERS_MAX);
 
 	for_each_set_bit(i, s->blocks_gotten, v->nr_blocks) {
-		__clear_bit(v->ptrs[i].dev, devs.d);
+		/*
+		 * Note: we don't yet repair invalid blocks (failed/removed
+		 * devices) when reusing stripes - we still need a codepath to
+		 * walk backpointers and update all extents that point to that
+		 * block when updating the stripe
+		 */
+		if (v->ptrs[i].dev != BCH_SB_MEMBER_INVALID)
+			__clear_bit(v->ptrs[i].dev, devs.d);
+
 		if (i < s->nr_data)
 			nr_have_data++;
 		else

libbcachefs/error.c

@@ -251,7 +251,10 @@ int __bch2_fsck_err(struct bch_fs *c,
 	 *   delete the key)
 	 * - and we don't need to warn if we're not prompting
 	 */
-	WARN_ON(!(flags & FSCK_AUTOFIX) && !trans && bch2_current_has_btree_trans(c));
+	WARN_ON((flags & FSCK_CAN_FIX) &&
+		!(flags & FSCK_AUTOFIX) &&
+		!trans &&
+		bch2_current_has_btree_trans(c));
 
 	if ((flags & FSCK_CAN_FIX) &&
 	    test_bit(err, c->sb.errors_silent))

libbcachefs/extents.c

@@ -203,7 +203,8 @@ int bch2_btree_ptr_v2_validate(struct bch_fs *c, struct bkey_s_c k,
 			 c, btree_ptr_v2_min_key_bad,
 			 "min_key > key");
 
-	if (flags & BCH_VALIDATE_write)
+	if ((flags & BCH_VALIDATE_write) &&
+	    c->sb.version_min >= bcachefs_metadata_version_btree_ptr_sectors_written)
 		bkey_fsck_err_on(!bp.v->sectors_written,
 				 c, btree_ptr_v2_written_0,
 				 "sectors_written == 0");

libbcachefs/extents.h

@@ -695,6 +695,16 @@ void bch2_bkey_ptrs_to_text(struct printbuf *, struct bch_fs *,
 int bch2_bkey_ptrs_validate(struct bch_fs *, struct bkey_s_c,
 			    enum bch_validate_flags);
 
+static inline bool bch2_extent_ptr_eq(struct bch_extent_ptr ptr1,
+				      struct bch_extent_ptr ptr2)
+{
+	return (ptr1.cached	== ptr2.cached &&
+		ptr1.unwritten	== ptr2.unwritten &&
+		ptr1.offset	== ptr2.offset &&
+		ptr1.dev	== ptr2.dev &&
+		ptr1.dev	== ptr2.dev);
+}
+
 void bch2_ptr_swab(struct bkey_s);
 
 const struct bch_extent_rebalance *bch2_bkey_rebalance_opts(struct bkey_s_c);

libbcachefs/fs-io-buffered.c

@@ -248,6 +248,7 @@ void bch2_readahead(struct readahead_control *ractl)
 	struct bch_io_opts opts;
 	struct folio *folio;
 	struct readpages_iter readpages_iter;
+	struct blk_plug plug;
 
 	bch2_inode_opts_get(&opts, c, &inode->ei_inode);
 
@@ -255,6 +256,16 @@ void bch2_readahead(struct readahead_control *ractl)
 	if (ret)
 		return;
 
+	/*
+	 * Besides being a general performance optimization, plugging helps with
+	 * avoiding btree transaction srcu warnings - submitting a bio can
+	 * block, and we don't want todo that with the transaction locked.
+	 *
+	 * However, plugged bios are submitted when we schedule; we ideally
+	 * would have our own scheduler hook to call unlock_long() before
+	 * scheduling.
+	 */
+	blk_start_plug(&plug);
 	bch2_pagecache_add_get(inode);
 
 	struct btree_trans *trans = bch2_trans_get(c);
@@ -281,7 +292,7 @@ void bch2_readahead(struct readahead_control *ractl)
 	bch2_trans_put(trans);
 
 	bch2_pagecache_add_put(inode);
-
+	blk_finish_plug(&plug);
 	darray_exit(&readpages_iter.folios);
 }
 
@@ -296,9 +307,13 @@ int bch2_read_single_folio(struct folio *folio, struct address_space *mapping)
 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
 	struct bch_read_bio *rbio;
 	struct bch_io_opts opts;
+	struct blk_plug plug;
 	int ret;
 	DECLARE_COMPLETION_ONSTACK(done);
 
+	BUG_ON(folio_test_uptodate(folio));
+	BUG_ON(folio_test_dirty(folio));
+
 	if (!bch2_folio_create(folio, GFP_KERNEL))
 		return -ENOMEM;
 
@@ -313,7 +328,9 @@ int bch2_read_single_folio(struct folio *folio, struct address_space *mapping)
 	rbio->bio.bi_iter.bi_sector = folio_sector(folio);
 	BUG_ON(!bio_add_folio(&rbio->bio, folio, folio_size(folio), 0));
 
+	blk_start_plug(&plug);
 	bch2_trans_run(c, (bchfs_read(trans, rbio, inode_inum(inode), NULL), 0));
+	blk_finish_plug(&plug);
 	wait_for_completion(&done);
 
 	ret = blk_status_to_errno(rbio->bio.bi_status);
@@ -856,6 +873,12 @@ static int __bch2_buffered_write(struct bch_inode_info *inode,
 				folios_trunc(&fs, fi);
 				end = min(end, folio_end_pos(darray_last(fs)));
 			} else {
+				if (!folio_test_uptodate(f)) {
+					ret = bch2_read_single_folio(f, mapping);
+					if (ret)
+						goto out;
+				}
+
 				folios_trunc(&fs, fi + 1);
 				end = f_pos + f_reserved;
 			}

libbcachefs/fs-io-direct.c

@@ -70,6 +70,7 @@ static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter)
 	struct bch_io_opts opts;
 	struct dio_read *dio;
 	struct bio *bio;
+	struct blk_plug plug;
 	loff_t offset = req->ki_pos;
 	bool sync = is_sync_kiocb(req);
 	size_t shorten;
@@ -128,6 +129,8 @@ static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter)
 	 */
 	dio->should_dirty = iter_is_iovec(iter);
 
+	blk_start_plug(&plug);
+
 	goto start;
 	while (iter->count) {
 		bio = bio_alloc_bioset(NULL,
@@ -160,6 +163,8 @@ start:
 		bch2_read(c, rbio_init(bio, opts), inode_inum(inode));
 	}
 
+	blk_finish_plug(&plug);
+
 	iter->count += shorten;
 
 	if (sync) {
@@ -369,6 +374,7 @@ static noinline void bch2_dio_write_flush(struct dio_write *dio)
 
 static __always_inline long bch2_dio_write_done(struct dio_write *dio)
 {
+	struct bch_fs *c = dio->op.c;
 	struct kiocb *req = dio->req;
 	struct bch_inode_info *inode = dio->inode;
 	bool sync = dio->sync;
@@ -387,7 +393,7 @@ static __always_inline long bch2_dio_write_done(struct dio_write *dio)
 	ret = dio->op.error ?: ((long) dio->written << 9);
 	bio_put(&dio->op.wbio.bio);
 
-	bch2_write_ref_put(dio->op.c, BCH_WRITE_REF_dio_write);
+	bch2_write_ref_put(c, BCH_WRITE_REF_dio_write);
 
 	/* inode->i_dio_count is our ref on inode and thus bch_fs */
 	inode_dio_end(&inode->v);

libbcachefs/fs-io-pagecache.c

@@ -399,14 +399,17 @@ void bch2_folio_reservation_put(struct bch_fs *c,
 	bch2_quota_reservation_put(c, inode, &res->quota);
 }
 
-int bch2_folio_reservation_get(struct bch_fs *c,
+static int __bch2_folio_reservation_get(struct bch_fs *c,
 			struct bch_inode_info *inode,
 			struct folio *folio,
 			struct bch2_folio_reservation *res,
-			size_t offset, size_t len)
+			size_t offset, size_t len,
+			bool partial)
 {
 	struct bch_folio *s = bch2_folio_create(folio, 0);
 	unsigned i, disk_sectors = 0, quota_sectors = 0;
+	struct disk_reservation disk_res = {};
+	size_t reserved = len;
 	int ret;
 
 	if (!s)
@@ -422,23 +425,56 @@ int bch2_folio_reservation_get(struct bch_fs *c,
 	}
 
 	if (disk_sectors) {
-		ret = bch2_disk_reservation_add(c, &res->disk, disk_sectors, 0);
+		ret = bch2_disk_reservation_add(c, &disk_res, disk_sectors,
+				partial ? BCH_DISK_RESERVATION_PARTIAL : 0);
 		if (unlikely(ret))
 			return ret;
+
+		if (unlikely(disk_res.sectors != disk_sectors)) {
+			disk_sectors = quota_sectors = 0;
+
+			for (i = round_down(offset, block_bytes(c)) >> 9;
+			     i < round_up(offset + len, block_bytes(c)) >> 9;
+			     i++) {
+				disk_sectors += sectors_to_reserve(&s->s[i], res->disk.nr_replicas);
+				if (disk_sectors > disk_res.sectors) {
+					/*
+					 * Make sure to get a reservation that's
+					 * aligned to the filesystem blocksize:
+					 */
+					unsigned reserved_offset = round_down(i << 9, block_bytes(c));
+					reserved = clamp(reserved_offset, offset, offset + len) - offset;
+
+					if (!reserved) {
+						bch2_disk_reservation_put(c, &disk_res);
+						return -BCH_ERR_ENOSPC_disk_reservation;
+					}
+					break;
+				}
+				quota_sectors += s->s[i].state == SECTOR_unallocated;
+			}
+		}
 	}
 
 	if (quota_sectors) {
 		ret = bch2_quota_reservation_add(c, inode, &res->quota, quota_sectors, true);
 		if (unlikely(ret)) {
-			struct disk_reservation tmp = { .sectors = disk_sectors };
+			bch2_disk_reservation_put(c, &disk_res);
-
-			bch2_disk_reservation_put(c, &tmp);
-			res->disk.sectors -= disk_sectors;
 			return ret;
 		}
 	}
 
-	return 0;
+	res->disk.sectors += disk_res.sectors;
+	return partial ? reserved : 0;
+}
+
+int bch2_folio_reservation_get(struct bch_fs *c,
+			struct bch_inode_info *inode,
+			struct folio *folio,
+			struct bch2_folio_reservation *res,
+			size_t offset, size_t len)
+{
+	return __bch2_folio_reservation_get(c, inode, folio, res, offset, len, false);
 }
 
 ssize_t bch2_folio_reservation_get_partial(struct bch_fs *c,
@@ -447,23 +483,7 @@ ssize_t bch2_folio_reservation_get_partial(struct bch_fs *c,
 			struct bch2_folio_reservation *res,
 			size_t offset, size_t len)
 {
-	size_t l, reserved = 0;
+	return __bch2_folio_reservation_get(c, inode, folio, res, offset, len, true);
-	int ret;
-
-	while ((l = len - reserved)) {
-		while ((ret = bch2_folio_reservation_get(c, inode, folio, res, offset, l))) {
-			if ((offset & (block_bytes(c) - 1)) + l <= block_bytes(c))
-				return reserved ?: ret;
-
-			len = reserved + l;
-			l /= 2;
-		}
-
-		offset += l;
-		reserved += l;
-	}
-
-	return reserved;
 }
 
 static void bch2_clear_folio_bits(struct folio *folio)

libbcachefs/fs-io.c

@@ -182,7 +182,7 @@ static int bch2_flush_inode(struct bch_fs *c,
 
 	struct bch_inode_unpacked u;
 	int ret = bch2_inode_find_by_inum(c, inode_inum(inode), &u) ?:
-		  bch2_journal_flush_seq(&c->journal, u.bi_journal_seq) ?:
+		  bch2_journal_flush_seq(&c->journal, u.bi_journal_seq, TASK_INTERRUPTIBLE) ?:
 		  bch2_inode_flush_nocow_writes(c, inode);
 	bch2_write_ref_put(c, BCH_WRITE_REF_fsync);
 	return ret;

libbcachefs/fs.c

@@ -658,7 +658,7 @@ static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 	struct bch_hash_info hash = bch2_hash_info_init(c, &dir->ei_inode);
 
 	struct bch_inode_info *inode;
-	bch2_trans_do(c, NULL, NULL, 0,
+	bch2_trans_do(c,
 		PTR_ERR_OR_ZERO(inode = bch2_lookup_trans(trans, inode_inum(dir),
 							  &hash, &dentry->d_name)));
 	if (IS_ERR(inode))
@@ -871,7 +871,7 @@ static int bch2_rename2(struct mnt_idmap *idmap,
 	ret   = bch2_subvol_is_ro_trans(trans, src_dir->ei_inum.subvol) ?:
 		bch2_subvol_is_ro_trans(trans, dst_dir->ei_inum.subvol);
 	if (ret)
-		goto err;
+		goto err_tx_restart;
 
 	if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
 		ret = bch2_fs_quota_transfer(c, src_inode,
@@ -1268,7 +1268,7 @@ static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 	bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
 			     POS(ei->v.i_ino, start), 0);
 
-	while (true) {
+	while (!ret || bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
 		enum btree_id data_btree = BTREE_ID_extents;
 
 		bch2_trans_begin(trans);
@@ -1276,14 +1276,14 @@ static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 		u32 snapshot;
 		ret = bch2_subvolume_get_snapshot(trans, ei->ei_inum.subvol, &snapshot);
 		if (ret)
-			goto err;
+			continue;
 
 		bch2_btree_iter_set_snapshot(&iter, snapshot);
 
 		k = bch2_btree_iter_peek_upto(&iter, end);
 		ret = bkey_err(k);
 		if (ret)
-			goto err;
+			continue;
 
 		if (!k.k)
 			break;
@@ -1303,7 +1303,7 @@ static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 		ret = bch2_read_indirect_extent(trans, &data_btree,
 					&offset_into_extent, &cur);
 		if (ret)
-			break;
+			continue;
 
 		k = bkey_i_to_s_c(cur.k);
 		bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
@@ -1331,10 +1331,6 @@ static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 
 		bch2_btree_iter_set_pos(&iter,
 			POS(iter.pos.inode, iter.pos.offset + sectors));
-err:
-		if (ret &&
-		    !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-			break;
 	}
 	bch2_trans_iter_exit(trans, &iter);
 

libbcachefs/fsck.c

@@ -934,35 +934,138 @@ static int get_visible_inodes(struct btree_trans *trans,
 	return ret;
 }
 
-static int hash_redo_key(struct btree_trans *trans,
+static int dirent_has_target(struct btree_trans *trans, struct bkey_s_c_dirent d)
-			 const struct bch_hash_desc desc,
-			 struct bch_hash_info *hash_info,
-			 struct btree_iter *k_iter, struct bkey_s_c k)
 {
-	struct bkey_i *delete;
+	if (d.v->d_type == DT_SUBVOL) {
-	struct bkey_i *tmp;
+		u32 snap;
+		u64 inum;
+		int ret = subvol_lookup(trans, le32_to_cpu(d.v->d_child_subvol), &snap, &inum);
+		if (ret && !bch2_err_matches(ret, ENOENT))
+			return ret;
+		return !ret;
+	} else {
+		struct btree_iter iter;
+		struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
+				SPOS(0, le64_to_cpu(d.v->d_inum), d.k->p.snapshot), 0);
+		int ret = bkey_err(k);
+		if (ret)
+			return ret;
 
-	delete = bch2_trans_kmalloc(trans, sizeof(*delete));
+		ret = bkey_is_inode(k.k);
-	if (IS_ERR(delete))
+		bch2_trans_iter_exit(trans, &iter);
-		return PTR_ERR(delete);
+		return ret;
+	}
+}
 
-	tmp = bch2_bkey_make_mut_noupdate(trans, k);
+/*
-	if (IS_ERR(tmp))
+ * Prefer to delete the first one, since that will be the one at the wrong
-		return PTR_ERR(tmp);
+ * offset:
+ * return value: 0 -> delete k1, 1 -> delete k2
+ */
+static int hash_pick_winner(struct btree_trans *trans,
+			    const struct bch_hash_desc desc,
+			    struct bch_hash_info *hash_info,
+			    struct bkey_s_c k1,
+			    struct bkey_s_c k2)
+{
+	if (bkey_val_bytes(k1.k) == bkey_val_bytes(k2.k) &&
+	    !memcmp(k1.v, k2.v, bkey_val_bytes(k1.k)))
+		return 0;
 
-	bkey_init(&delete->k);
+	switch (desc.btree_id) {
-	delete->k.p = k_iter->pos;
+	case BTREE_ID_dirents: {
-	return  bch2_btree_iter_traverse(k_iter) ?:
+		int ret = dirent_has_target(trans, bkey_s_c_to_dirent(k1));
-		bch2_trans_update(trans, k_iter, delete, 0) ?:
+		if (ret < 0)
-		bch2_hash_set_in_snapshot(trans, desc, hash_info,
+			return ret;
-				       (subvol_inum) { 0, k.k->p.inode },
+		if (!ret)
-				       k.k->p.snapshot, tmp,
+			return 0;
-				       STR_HASH_must_create|
+
-				       BTREE_UPDATE_internal_snapshot_node) ?:
+		ret = dirent_has_target(trans, bkey_s_c_to_dirent(k2));
-		bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
+		if (ret < 0)
+			return ret;
+		if (!ret)
+			return 1;
+		return 2;
+	}
+	default:
+		return 0;
+	}
+}
+
+static int fsck_update_backpointers(struct btree_trans *trans,
+				    struct snapshots_seen *s,
+				    const struct bch_hash_desc desc,
+				    struct bch_hash_info *hash_info,
+				    struct bkey_i *new)
+{
+	if (new->k.type != KEY_TYPE_dirent)
+		return 0;
+
+	struct bkey_i_dirent *d = bkey_i_to_dirent(new);
+	struct inode_walker target = inode_walker_init();
+	int ret = 0;
+
+	if (d->v.d_type == DT_SUBVOL) {
+		BUG();
+	} else {
+		ret = get_visible_inodes(trans, &target, s, le64_to_cpu(d->v.d_inum));
+		if (ret)
+			goto err;
+
+		darray_for_each(target.inodes, i) {
+			i->inode.bi_dir_offset = d->k.p.offset;
+			ret = __bch2_fsck_write_inode(trans, &i->inode);
+			if (ret)
+				goto err;
+		}
+	}
+err:
+	inode_walker_exit(&target);
+	return ret;
+}
+
+static int fsck_rename_dirent(struct btree_trans *trans,
+			      struct snapshots_seen *s,
+			      const struct bch_hash_desc desc,
+			      struct bch_hash_info *hash_info,
+			      struct bkey_s_c_dirent old)
+{
+	struct qstr old_name = bch2_dirent_get_name(old);
+	struct bkey_i_dirent *new = bch2_trans_kmalloc(trans, bkey_bytes(old.k) + 32);
+	int ret = PTR_ERR_OR_ZERO(new);
+	if (ret)
+		return ret;
+
+	bkey_dirent_init(&new->k_i);
+	dirent_copy_target(new, old);
+	new->k.p = old.k->p;
+
+	for (unsigned i = 0; i < 1000; i++) {
+		unsigned len = sprintf(new->v.d_name, "%.*s.fsck_renamed-%u",
+				       old_name.len, old_name.name, i);
+		unsigned u64s = BKEY_U64s + dirent_val_u64s(len);
+
+		if (u64s > U8_MAX)
+			return -EINVAL;
+
+		new->k.u64s = u64s;
+
+		ret = bch2_hash_set_in_snapshot(trans, bch2_dirent_hash_desc, hash_info,
+						(subvol_inum) { 0, old.k->p.inode },
+						old.k->p.snapshot, &new->k_i,
+						BTREE_UPDATE_internal_snapshot_node);
+		if (!bch2_err_matches(ret, EEXIST))
+			break;
+	}
+
+	if (ret)
+		return ret;
+
+	return fsck_update_backpointers(trans, s, desc, hash_info, &new->k_i);
 }
 
 static int hash_check_key(struct btree_trans *trans,
+			  struct snapshots_seen *s,
 			  const struct bch_hash_desc desc,
 			  struct bch_hash_info *hash_info,
 			  struct btree_iter *k_iter, struct bkey_s_c hash_k)
@@ -991,16 +1094,9 @@ static int hash_check_key(struct btree_trans *trans,
 		if (bkey_eq(k.k->p, hash_k.k->p))
 			break;
 
-		if (fsck_err_on(k.k->type == desc.key_type &&
+		if (k.k->type == desc.key_type &&
-				!desc.cmp_bkey(k, hash_k),
+		    !desc.cmp_bkey(k, hash_k))
-				trans, hash_table_key_duplicate,
+			goto duplicate_entries;
-				"duplicate hash table keys:\n%s",
-				(printbuf_reset(&buf),
-				 bch2_bkey_val_to_text(&buf, c, hash_k),
-				 buf.buf))) {
-			ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
-			break;
-		}
 
 		if (bkey_deleted(k.k)) {
 			bch2_trans_iter_exit(trans, &iter);
@@ -1013,18 +1109,66 @@ out:
 	return ret;
 bad_hash:
 	if (fsck_err(trans, hash_table_key_wrong_offset,
-		     "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
+		     "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n  %s",
 		     bch2_btree_id_str(desc.btree_id), hash_k.k->p.inode, hash_k.k->p.offset, hash,
 		     (printbuf_reset(&buf),
 		      bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
-		ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
+		struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, hash_k);
-		bch_err_fn(c, ret);
+		if (IS_ERR(new))
+			return PTR_ERR(new);
+
+		k = bch2_hash_set_or_get_in_snapshot(trans, &iter, desc, hash_info,
+				       (subvol_inum) { 0, hash_k.k->p.inode },
+				       hash_k.k->p.snapshot, new,
+				       STR_HASH_must_create|
+				       BTREE_ITER_with_updates|
+				       BTREE_UPDATE_internal_snapshot_node);
+		ret = bkey_err(k);
 		if (ret)
-			return ret;
+			goto out;
-		ret = -BCH_ERR_transaction_restart_nested;
+		if (k.k)
+			goto duplicate_entries;
+
+		ret =   bch2_hash_delete_at(trans, desc, hash_info, k_iter,
+					    BTREE_UPDATE_internal_snapshot_node) ?:
+			fsck_update_backpointers(trans, s, desc, hash_info, new) ?:
+			bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc) ?:
+			-BCH_ERR_transaction_restart_nested;
+		goto out;
 	}
 fsck_err:
 	goto out;
+duplicate_entries:
+	ret = hash_pick_winner(trans, desc, hash_info, hash_k, k);
+	if (ret < 0)
+		goto out;
+
+	if (!fsck_err(trans, hash_table_key_duplicate,
+		      "duplicate hash table keys%s:\n%s",
+		      ret != 2 ? "" : ", both point to valid inodes",
+		      (printbuf_reset(&buf),
+		       bch2_bkey_val_to_text(&buf, c, hash_k),
+		       prt_newline(&buf),
+		       bch2_bkey_val_to_text(&buf, c, k),
+		       buf.buf)))
+		goto out;
+
+	switch (ret) {
+	case 0:
+		ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0);
+		break;
+	case 1:
+		ret = bch2_hash_delete_at(trans, desc, hash_info, &iter, 0);
+		break;
+	case 2:
+		ret = fsck_rename_dirent(trans, s, desc, hash_info, bkey_s_c_to_dirent(hash_k)) ?:
+			bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0);
+		goto out;
+	}
+
+	ret = bch2_trans_commit(trans, NULL, NULL, 0) ?:
+		-BCH_ERR_transaction_restart_nested;
+	goto out;
 }
 
 static struct bkey_s_c_dirent inode_get_dirent(struct btree_trans *trans,
@@ -1094,10 +1238,36 @@ fsck_err:
 	return ret;
 }
 
+static int get_snapshot_root_inode(struct btree_trans *trans,
+				   struct bch_inode_unpacked *root,
+				   u64 inum)
+{
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	int ret = 0;
+
+	for_each_btree_key_reverse_norestart(trans, iter, BTREE_ID_inodes,
+					     SPOS(0, inum, U32_MAX),
+					     BTREE_ITER_all_snapshots, k, ret) {
+		if (k.k->p.offset != inum)
+			break;
+		if (bkey_is_inode(k.k))
+			goto found_root;
+	}
+	if (ret)
+		goto err;
+	BUG();
+found_root:
+	BUG_ON(bch2_inode_unpack(k, root));
+err:
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
+}
+
 static int check_inode(struct btree_trans *trans,
 		       struct btree_iter *iter,
 		       struct bkey_s_c k,
-		       struct bch_inode_unpacked *prev,
+		       struct bch_inode_unpacked *snapshot_root,
 		       struct snapshots_seen *s)
 {
 	struct bch_fs *c = trans->c;
@@ -1121,16 +1291,19 @@ static int check_inode(struct btree_trans *trans,
 
 	BUG_ON(bch2_inode_unpack(k, &u));
 
-	if (prev->bi_inum != u.bi_inum)
+	if (snapshot_root->bi_inum != u.bi_inum) {
-		*prev = u;
+		ret = get_snapshot_root_inode(trans, snapshot_root, u.bi_inum);
+		if (ret)
+			goto err;
+	}
 
-	if (fsck_err_on(prev->bi_hash_seed	!= u.bi_hash_seed ||
+	if (fsck_err_on(u.bi_hash_seed		!= snapshot_root->bi_hash_seed ||
-			inode_d_type(prev)	!= inode_d_type(&u),
+			INODE_STR_HASH(&u)	!= INODE_STR_HASH(snapshot_root),
 			trans, inode_snapshot_mismatch,
 			"inodes in different snapshots don't match")) {
-		bch_err(c, "repair not implemented yet");
+		u.bi_hash_seed = snapshot_root->bi_hash_seed;
-		ret = -BCH_ERR_fsck_repair_unimplemented;
+		SET_INODE_STR_HASH(&u, INODE_STR_HASH(snapshot_root));
-		goto err_noprint;
+		do_update = true;
 	}
 
 	if (u.bi_dir || u.bi_dir_offset) {
@@ -1283,7 +1456,7 @@ err_noprint:
 
 int bch2_check_inodes(struct bch_fs *c)
 {
-	struct bch_inode_unpacked prev = { 0 };
+	struct bch_inode_unpacked snapshot_root = {};
 	struct snapshots_seen s;
 
 	snapshots_seen_init(&s);
@@ -1293,7 +1466,7 @@ int bch2_check_inodes(struct bch_fs *c)
 				POS_MIN,
 				BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
-			check_inode(trans, &iter, k, &prev, &s)));
+			check_inode(trans, &iter, k, &snapshot_root, &s)));
 
 	snapshots_seen_exit(&s);
 	bch_err_fn(c, ret);
@@ -2305,7 +2478,7 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 		*hash_info = bch2_hash_info_init(c, &i->inode);
 	dir->first_this_inode = false;
 
-	ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
+	ret = hash_check_key(trans, s, bch2_dirent_hash_desc, hash_info, iter, k);
 	if (ret < 0)
 		goto err;
 	if (ret) {
@@ -2419,7 +2592,7 @@ static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
 		*hash_info = bch2_hash_info_init(c, &i->inode);
 	inode->first_this_inode = false;
 
-	ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
+	ret = hash_check_key(trans, NULL, bch2_xattr_hash_desc, hash_info, iter, k);
 	bch_err_fn(c, ret);
 	return ret;
 }
@@ -2507,7 +2680,7 @@ fsck_err:
 /* Get root directory, create if it doesn't exist: */
 int bch2_check_root(struct bch_fs *c)
 {
-	int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+	int ret = bch2_trans_commit_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 		check_root_trans(trans));
 	bch_err_fn(c, ret);
 	return ret;

libbcachefs/inode.c

@@ -163,8 +163,8 @@ static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
 	unsigned fieldnr = 0, field_bits;
 	int ret;
 
-#define x(_name, _bits)					\
+#define x(_name, _bits)							\
-	if (fieldnr++ == INODE_NR_FIELDS(inode.v)) {			\
+	if (fieldnr++ == INODEv1_NR_FIELDS(inode.v)) {			\
 		unsigned offset = offsetof(struct bch_inode_unpacked, _name);\
 		memset((void *) unpacked + offset, 0,			\
 		       sizeof(*unpacked) - offset);			\
@@ -283,6 +283,8 @@ static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
 {
 	memset(unpacked, 0, sizeof(*unpacked));
 
+	unpacked->bi_snapshot = k.k->p.snapshot;
+
 	switch (k.k->type) {
 	case KEY_TYPE_inode: {
 		struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
@@ -293,10 +295,10 @@ static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
 		unpacked->bi_flags	= le32_to_cpu(inode.v->bi_flags);
 		unpacked->bi_mode	= le16_to_cpu(inode.v->bi_mode);
 
-		if (INODE_NEW_VARINT(inode.v)) {
+		if (INODEv1_NEW_VARINT(inode.v)) {
 			return bch2_inode_unpack_v2(unpacked, inode.v->fields,
 						    bkey_val_end(inode),
-						    INODE_NR_FIELDS(inode.v));
+						    INODEv1_NR_FIELDS(inode.v));
 		} else {
 			return bch2_inode_unpack_v1(inode, unpacked);
 		}
@@ -471,10 +473,10 @@ int bch2_inode_validate(struct bch_fs *c, struct bkey_s_c k,
 	struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 	int ret = 0;
 
-	bkey_fsck_err_on(INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR,
+	bkey_fsck_err_on(INODEv1_STR_HASH(inode.v) >= BCH_STR_HASH_NR,
 			 c, inode_str_hash_invalid,
 			 "invalid str hash type (%llu >= %u)",
-			 INODE_STR_HASH(inode.v), BCH_STR_HASH_NR);
+			 INODEv1_STR_HASH(inode.v), BCH_STR_HASH_NR);
 
 	ret = __bch2_inode_validate(c, k, flags);
 fsck_err:
@@ -534,6 +536,9 @@ static void __bch2_inode_unpacked_to_text(struct printbuf *out,
 
 	prt_printf(out, "journal_seq=%llu\n",	inode->bi_journal_seq);
 	prt_printf(out, "hash_seed=%llx\n",	inode->bi_hash_seed);
+	prt_printf(out, "hash_type=");
+	bch2_prt_str_hash_type(out, INODE_STR_HASH(inode));
+	prt_newline(out);
 	prt_printf(out, "bi_size=%llu\n",	inode->bi_size);
 	prt_printf(out, "bi_sectors=%llu\n",	inode->bi_sectors);
 	prt_printf(out, "bi_version=%llu\n",	inode->bi_version);
@@ -801,10 +806,8 @@ void bch2_inode_init_early(struct bch_fs *c,
 
 	memset(inode_u, 0, sizeof(*inode_u));
 
-	/* ick */
+	SET_INODE_STR_HASH(inode_u, str_hash);
-	inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
+	get_random_bytes(&inode_u->bi_hash_seed, sizeof(inode_u->bi_hash_seed));
-	get_random_bytes(&inode_u->bi_hash_seed,
-			 sizeof(inode_u->bi_hash_seed));
 }
 
 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
@@ -1088,8 +1091,7 @@ int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
 			    struct bch_inode_unpacked *inode)
 {
-	return bch2_trans_do(c, NULL, NULL, 0,
+	return bch2_trans_do(c, bch2_inode_find_by_inum_trans(trans, inum, inode));
-		bch2_inode_find_by_inum_trans(trans, inum, inode));
 }
 
 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)

libbcachefs/inode.h

@@ -92,6 +92,7 @@ struct bch_inode_unpacked {
 	BCH_INODE_FIELDS_v3()
 #undef  x
 };
+BITMASK(INODE_STR_HASH,	struct bch_inode_unpacked, bi_flags, 20, 24);
 
 struct bkey_inode_buf {
 	struct bkey_i_inode_v3	inode;

libbcachefs/inode_format.h

@@ -150,9 +150,9 @@ enum __bch_inode_flags {
 #undef x
 };
 
-LE32_BITMASK(INODE_STR_HASH,	struct bch_inode, bi_flags, 20, 24);
+LE32_BITMASK(INODEv1_STR_HASH,	struct bch_inode, bi_flags, 20, 24);
-LE32_BITMASK(INODE_NR_FIELDS,	struct bch_inode, bi_flags, 24, 31);
+LE32_BITMASK(INODEv1_NR_FIELDS,	struct bch_inode, bi_flags, 24, 31);
-LE32_BITMASK(INODE_NEW_VARINT,	struct bch_inode, bi_flags, 31, 32);
+LE32_BITMASK(INODEv1_NEW_VARINT,struct bch_inode, bi_flags, 31, 32);
 
 LE64_BITMASK(INODEv2_STR_HASH,	struct bch_inode_v2, bi_flags, 20, 24);
 LE64_BITMASK(INODEv2_NR_FIELDS,	struct bch_inode_v2, bi_flags, 24, 31);

libbcachefs/io_misc.c

@@ -377,7 +377,7 @@ static int __bch2_resume_logged_op_finsert(struct btree_trans *trans,
 	 * check for missing subvolume before fpunch, as in resume we don't want
 	 * it to be a fatal error
 	 */
-	ret = __bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot, warn_errors);
+	ret = lockrestart_do(trans, __bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot, warn_errors));
 	if (ret)
 		return ret;
 

libbcachefs/io_read.c

@@ -409,8 +409,8 @@ retry:
 	bch2_trans_begin(trans);
 	rbio->bio.bi_status = 0;
 
-	k = bch2_btree_iter_peek_slot(&iter);
+	ret = lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
-	if (bkey_err(k))
+	if (ret)
 		goto err;
 
 	bch2_bkey_buf_reassemble(&sk, c, k);
@@ -557,8 +557,8 @@ out:
 
 static noinline void bch2_rbio_narrow_crcs(struct bch_read_bio *rbio)
 {
-	bch2_trans_do(rbio->c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+	bch2_trans_commit_do(rbio->c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
-		      __bch2_rbio_narrow_crcs(trans, rbio));
+			     __bch2_rbio_narrow_crcs(trans, rbio));
 }
 
 /* Inner part that may run in process context */

libbcachefs/io_write.c

@@ -1437,7 +1437,7 @@ again:
 		 * freeing up space on specific disks, which means that
 		 * allocations for specific disks may hang arbitrarily long:
 		 */
-		ret = bch2_trans_do(c, NULL, NULL, 0,
+		ret = bch2_trans_run(c, lockrestart_do(trans,
 			bch2_alloc_sectors_start_trans(trans,
 				op->target,
 				op->opts.erasure_code && !(op->flags & BCH_WRITE_CACHED),
@@ -1447,7 +1447,7 @@ again:
 				op->nr_replicas_required,
 				op->watermark,
 				op->flags,
-				&op->cl, &wp));
+				&op->cl, &wp)));
 		if (unlikely(ret)) {
 			if (bch2_err_matches(ret, BCH_ERR_operation_blocked))
 				break;

libbcachefs/journal.c

@@ -758,7 +758,7 @@ out:
 	return ret;
 }
 
-int bch2_journal_flush_seq(struct journal *j, u64 seq)
+int bch2_journal_flush_seq(struct journal *j, u64 seq, unsigned task_state)
 {
 	u64 start_time = local_clock();
 	int ret, ret2;
@@ -769,7 +769,9 @@ int bch2_journal_flush_seq(struct journal *j, u64 seq)
 	if (seq <= j->flushed_seq_ondisk)
 		return 0;
 
-	ret = wait_event_interruptible(j->wait, (ret2 = bch2_journal_flush_seq_async(j, seq, NULL)));
+	ret = wait_event_state(j->wait,
+			       (ret2 = bch2_journal_flush_seq_async(j, seq, NULL)),
+			       task_state);
 
 	if (!ret)
 		bch2_time_stats_update(j->flush_seq_time, start_time);
@@ -788,7 +790,7 @@ void bch2_journal_flush_async(struct journal *j, struct closure *parent)
 
 int bch2_journal_flush(struct journal *j)
 {
-	return bch2_journal_flush_seq(j, atomic64_read(&j->seq));
+	return bch2_journal_flush_seq(j, atomic64_read(&j->seq), TASK_UNINTERRUPTIBLE);
 }
 
 /*
@@ -846,7 +848,7 @@ static int __bch2_journal_meta(struct journal *j)
 
 	bch2_journal_res_put(j, &res);
 
-	return bch2_journal_flush_seq(j, res.seq);
+	return bch2_journal_flush_seq(j, res.seq, TASK_UNINTERRUPTIBLE);
 }
 
 int bch2_journal_meta(struct journal *j)

libbcachefs/journal.h

@@ -401,7 +401,7 @@ void bch2_journal_entry_res_resize(struct journal *,
 int bch2_journal_flush_seq_async(struct journal *, u64, struct closure *);
 void bch2_journal_flush_async(struct journal *, struct closure *);
 
-int bch2_journal_flush_seq(struct journal *, u64);
+int bch2_journal_flush_seq(struct journal *, u64, unsigned);
 int bch2_journal_flush(struct journal *);
 bool bch2_journal_noflush_seq(struct journal *, u64);
 int bch2_journal_meta(struct journal *);

libbcachefs/opts.c

@@ -63,7 +63,7 @@ const char * const bch2_compression_opts[] = {
 	NULL
 };
 
-const char * const bch2_str_hash_types[] = {
+const char * const __bch2_str_hash_types[] = {
 	BCH_STR_HASH_TYPES()
 	NULL
 };
@@ -115,6 +115,7 @@ PRT_STR_OPT_BOUNDSCHECKED(fs_usage_type,	enum bch_fs_usage_type);
 PRT_STR_OPT_BOUNDSCHECKED(data_type,		enum bch_data_type);
 PRT_STR_OPT_BOUNDSCHECKED(csum_type,		enum bch_csum_type);
 PRT_STR_OPT_BOUNDSCHECKED(compression_type,	enum bch_compression_type);
+PRT_STR_OPT_BOUNDSCHECKED(str_hash_type,	enum bch_str_hash_type);
 
 static int bch2_opt_fix_errors_parse(struct bch_fs *c, const char *val, u64 *res,
 				     struct printbuf *err)

libbcachefs/opts.h

@@ -18,7 +18,7 @@ extern const char * const bch2_sb_compat[];
 extern const char * const __bch2_btree_ids[];
 extern const char * const bch2_csum_opts[];
 extern const char * const bch2_compression_opts[];
-extern const char * const bch2_str_hash_types[];
+extern const char * const __bch2_str_hash_types[];
 extern const char * const bch2_str_hash_opts[];
 extern const char * const __bch2_data_types[];
 extern const char * const bch2_member_states[];
@@ -29,6 +29,7 @@ void bch2_prt_fs_usage_type(struct printbuf *,		enum bch_fs_usage_type);
 void bch2_prt_data_type(struct printbuf *,		enum bch_data_type);
 void bch2_prt_csum_type(struct printbuf *,		enum bch_csum_type);
 void bch2_prt_compression_type(struct printbuf *,	enum bch_compression_type);
+void bch2_prt_str_hash_type(struct printbuf *,		enum bch_str_hash_type);
 
 static inline const char *bch2_d_type_str(unsigned d_type)
 {

libbcachefs/quota.c

@@ -869,7 +869,7 @@ static int bch2_set_quota(struct super_block *sb, struct kqid qid,
 	bkey_quota_init(&new_quota.k_i);
 	new_quota.k.p = POS(qid.type, from_kqid(&init_user_ns, qid));
 
-	ret = bch2_trans_do(c, NULL, NULL, 0,
+	ret = bch2_trans_commit_do(c, NULL, NULL, 0,
 			    bch2_set_quota_trans(trans, &new_quota, qdq)) ?:
 		__bch2_quota_set(c, bkey_i_to_s_c(&new_quota.k_i), qdq);
 

libbcachefs/rebalance.c

@@ -70,7 +70,9 @@ err:
 
 int bch2_set_rebalance_needs_scan(struct bch_fs *c, u64 inum)
 {
-	int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc|BCH_TRANS_COMMIT_lazy_rw,
+	int ret = bch2_trans_commit_do(c, NULL, NULL,
+				       BCH_TRANS_COMMIT_no_enospc|
+				       BCH_TRANS_COMMIT_lazy_rw,
 			    __bch2_set_rebalance_needs_scan(trans, inum));
 	rebalance_wakeup(c);
 	return ret;

libbcachefs/recovery.c

@@ -321,7 +321,8 @@ int bch2_journal_replay(struct bch_fs *c)
 				BCH_TRANS_COMMIT_no_enospc|
 				BCH_TRANS_COMMIT_journal_reclaim|
 				BCH_TRANS_COMMIT_skip_accounting_apply|
-				BCH_TRANS_COMMIT_no_journal_res,
+				BCH_TRANS_COMMIT_no_journal_res|
+				BCH_WATERMARK_reclaim,
 			     bch2_journal_replay_accounting_key(trans, k));
 		if (bch2_fs_fatal_err_on(ret, c, "error replaying accounting; %s", bch2_err_str(ret)))
 			goto err;
@@ -1118,7 +1119,7 @@ int bch2_fs_initialize(struct bch_fs *c)
 
 	bch2_inode_init_early(c, &lostfound_inode);
 
-	ret = bch2_trans_do(c, NULL, NULL, 0,
+	ret = bch2_trans_commit_do(c, NULL, NULL, 0,
 		bch2_create_trans(trans,
 				  BCACHEFS_ROOT_SUBVOL_INUM,
 				  &root_inode, &lostfound_inode,

libbcachefs/sb-errors_format.h

@@ -180,6 +180,7 @@ enum bch_fsck_flags {
 	x(reflink_p_to_missing_reflink_v,			166,	0)		\
 	x(stripe_pos_bad,					167,	0)		\
 	x(stripe_val_size_bad,					168,	0)		\
+	x(stripe_csum_granularity_bad,				290,	0)		\
 	x(stripe_sector_count_wrong,				169,	0)		\
 	x(snapshot_tree_pos_bad,				170,	0)		\
 	x(snapshot_tree_to_missing_snapshot,			171,	0)		\
@@ -266,8 +267,8 @@ enum bch_fsck_flags {
 	x(journal_entry_dup_same_device,			246,	0)		\
 	x(inode_bi_subvol_missing,				247,	0)		\
 	x(inode_bi_subvol_wrong,				248,	0)		\
-	x(inode_points_to_missing_dirent,			249,	0)		\
+	x(inode_points_to_missing_dirent,			249,	FSCK_AUTOFIX)	\
-	x(inode_points_to_wrong_dirent,				250,	0)		\
+	x(inode_points_to_wrong_dirent,				250,	FSCK_AUTOFIX)	\
 	x(inode_bi_parent_nonzero,				251,	0)		\
 	x(dirent_to_missing_parent_subvol,			252,	0)		\
 	x(dirent_not_visible_in_parent_subvol,			253,	0)		\
@@ -301,7 +302,7 @@ enum bch_fsck_flags {
 	x(accounting_key_replicas_devs_unsorted,		280,	FSCK_AUTOFIX)	\
 	x(accounting_key_version_0,				282,	FSCK_AUTOFIX)	\
 	x(logged_op_but_clean,					283,	FSCK_AUTOFIX)	\
-	x(MAX,							290,	0)
+	x(MAX,							291,	0)
 
 enum bch_sb_error_id {
 #define x(t, n, ...) BCH_FSCK_ERR_##t = n,

libbcachefs/sb-members.c

@@ -163,6 +163,11 @@ static int validate_member(struct printbuf *err,
 		return -BCH_ERR_invalid_sb_members;
 	}
 
+	if (m.btree_bitmap_shift >= 64) {
+		prt_printf(err, "device %u: invalid btree_bitmap_shift %u", i, m.btree_bitmap_shift);
+		return -BCH_ERR_invalid_sb_members;
+	}
+
 	return 0;
 }
 

libbcachefs/str_hash.h

@@ -46,8 +46,7 @@ bch2_hash_info_init(struct bch_fs *c, const struct bch_inode_unpacked *bi)
 {
 	/* XXX ick */
 	struct bch_hash_info info = {
-		.type = (bi->bi_flags >> INODE_STR_HASH_OFFSET) &
+		.type = INODE_STR_HASH(bi),
-			~(~0U << INODE_STR_HASH_BITS),
 		.siphash_key = { .k0 = bi->bi_hash_seed }
 	};
 
@@ -253,19 +252,20 @@ int bch2_hash_needs_whiteout(struct btree_trans *trans,
 }
 
 static __always_inline
-int bch2_hash_set_in_snapshot(struct btree_trans *trans,
+struct bkey_s_c bch2_hash_set_or_get_in_snapshot(struct btree_trans *trans,
+			   struct btree_iter *iter,
 			   const struct bch_hash_desc desc,
 			   const struct bch_hash_info *info,
 			   subvol_inum inum, u32 snapshot,
 			   struct bkey_i *insert,
 			   enum btree_iter_update_trigger_flags flags)
 {
-	struct btree_iter iter, slot = { NULL };
+	struct btree_iter slot = {};
 	struct bkey_s_c k;
 	bool found = false;
 	int ret;
 
-	for_each_btree_key_upto_norestart(trans, iter, desc.btree_id,
+	for_each_btree_key_upto_norestart(trans, *iter, desc.btree_id,
 			   SPOS(insert->k.p.inode,
 				desc.hash_bkey(info, bkey_i_to_s_c(insert)),
 				snapshot),
@@ -280,7 +280,7 @@ int bch2_hash_set_in_snapshot(struct btree_trans *trans,
 		}
 
 		if (!slot.path && !(flags & STR_HASH_must_replace))
-			bch2_trans_copy_iter(&slot, &iter);
+			bch2_trans_copy_iter(&slot, iter);
 
 		if (k.k->type != KEY_TYPE_hash_whiteout)
 			goto not_found;
@@ -290,28 +290,49 @@ int bch2_hash_set_in_snapshot(struct btree_trans *trans,
 		ret = -BCH_ERR_ENOSPC_str_hash_create;
 out:
 	bch2_trans_iter_exit(trans, &slot);
-	bch2_trans_iter_exit(trans, &iter);
+	bch2_trans_iter_exit(trans, iter);
-
+	return ret ? bkey_s_c_err(ret) : bkey_s_c_null;
-	return ret;
 found:
 	found = true;
 not_found:
-
+	if (found && (flags & STR_HASH_must_create)) {
-	if (!found && (flags & STR_HASH_must_replace)) {
+		bch2_trans_iter_exit(trans, &slot);
+		return k;
+	} else if (!found && (flags & STR_HASH_must_replace)) {
 		ret = -BCH_ERR_ENOENT_str_hash_set_must_replace;
-	} else if (found && (flags & STR_HASH_must_create)) {
-		ret = -BCH_ERR_EEXIST_str_hash_set;
 	} else {
 		if (!found && slot.path)
-			swap(iter, slot);
+			swap(*iter, slot);
 
-		insert->k.p = iter.pos;
+		insert->k.p = iter->pos;
-		ret = bch2_trans_update(trans, &iter, insert, flags);
+		ret = bch2_trans_update(trans, iter, insert, flags);
 	}
 
 	goto out;
 }
 
+static __always_inline
+int bch2_hash_set_in_snapshot(struct btree_trans *trans,
+			   const struct bch_hash_desc desc,
+			   const struct bch_hash_info *info,
+			   subvol_inum inum, u32 snapshot,
+			   struct bkey_i *insert,
+			   enum btree_iter_update_trigger_flags flags)
+{
+	struct btree_iter iter;
+	struct bkey_s_c k = bch2_hash_set_or_get_in_snapshot(trans, &iter, desc, info, inum,
+							     snapshot, insert, flags);
+	int ret = bkey_err(k);
+	if (ret)
+		return ret;
+	if (k.k) {
+		bch2_trans_iter_exit(trans, &iter);
+		return -BCH_ERR_EEXIST_str_hash_set;
+	}
+
+	return 0;
+}
+
 static __always_inline
 int bch2_hash_set(struct btree_trans *trans,
 		  const struct bch_hash_desc desc,
@@ -363,8 +384,11 @@ int bch2_hash_delete(struct btree_trans *trans,
 	struct btree_iter iter;
 	struct bkey_s_c k = bch2_hash_lookup(trans, &iter, desc, info, inum, key,
 					     BTREE_ITER_intent);
-	int ret = bkey_err(k) ?:
+	int ret = bkey_err(k);
-		  bch2_hash_delete_at(trans, desc, info, &iter, 0);
+	if (ret)
+		return ret;
+
+	ret = bch2_hash_delete_at(trans, desc, info, &iter, 0);
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
 }

libbcachefs/subvolume.c

@@ -319,8 +319,7 @@ int bch2_subvol_is_ro_trans(struct btree_trans *trans, u32 subvol)
 
 int bch2_subvol_is_ro(struct bch_fs *c, u32 subvol)
 {
-	return bch2_trans_do(c, NULL, NULL, 0,
+	return bch2_trans_do(c, bch2_subvol_is_ro_trans(trans, subvol));
-		bch2_subvol_is_ro_trans(trans, subvol));
 }
 
 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
@@ -676,8 +675,8 @@ err:
 /* set bi_subvol on root inode */
 int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
 {
-	int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
+	int ret = bch2_trans_commit_do(c, NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
-				__bch2_fs_upgrade_for_subvolumes(trans));
+				       __bch2_fs_upgrade_for_subvolumes(trans));
 	bch_err_fn(c, ret);
 	return ret;
 }

libbcachefs/super.c

@@ -184,6 +184,7 @@ static DEFINE_MUTEX(bch_fs_list_lock);
 
 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
 
+static void bch2_dev_unlink(struct bch_dev *);
 static void bch2_dev_free(struct bch_dev *);
 static int bch2_dev_alloc(struct bch_fs *, unsigned);
 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
@@ -620,9 +621,7 @@ void __bch2_fs_stop(struct bch_fs *c)
 	up_write(&c->state_lock);
 
 	for_each_member_device(c, ca)
-		if (ca->kobj.state_in_sysfs &&
+		bch2_dev_unlink(ca);
-		    ca->disk_sb.bdev)
-			sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 
 	if (c->kobj.state_in_sysfs)
 		kobject_del(&c->kobj);
@@ -1188,9 +1187,7 @@ static void bch2_dev_free(struct bch_dev *ca)
 {
 	cancel_work_sync(&ca->io_error_work);
 
-	if (ca->kobj.state_in_sysfs &&
+	bch2_dev_unlink(ca);
-	    ca->disk_sb.bdev)
-		sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 
 	if (ca->kobj.state_in_sysfs)
 		kobject_del(&ca->kobj);
@@ -1227,10 +1224,7 @@ static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
 	percpu_ref_kill(&ca->io_ref);
 	wait_for_completion(&ca->io_ref_completion);
 
-	if (ca->kobj.state_in_sysfs) {
+	bch2_dev_unlink(ca);
-		sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
-		sysfs_remove_link(&ca->kobj, "block");
-	}
 
 	bch2_free_super(&ca->disk_sb);
 	bch2_dev_journal_exit(ca);
@@ -1252,6 +1246,26 @@ static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
 	complete(&ca->io_ref_completion);
 }
 
+static void bch2_dev_unlink(struct bch_dev *ca)
+{
+	struct kobject *b;
+
+	/*
+	 * This is racy w.r.t. the underlying block device being hot-removed,
+	 * which removes it from sysfs.
+	 *
+	 * It'd be lovely if we had a way to handle this race, but the sysfs
+	 * code doesn't appear to provide a good method and block/holder.c is
+	 * susceptible as well:
+	 */
+	if (ca->kobj.state_in_sysfs &&
+	    ca->disk_sb.bdev &&
+	    (b = bdev_kobj(ca->disk_sb.bdev))->state_in_sysfs) {
+		sysfs_remove_link(b, "bcachefs");
+		sysfs_remove_link(&ca->kobj, "block");
+	}
+}
+
 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
 {
 	int ret;
@@ -1959,7 +1973,7 @@ int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 		};
 		u64 v[3] = { nbuckets - old_nbuckets, 0, 0 };
 
-		ret   = bch2_trans_do(ca->fs, NULL, NULL, 0,
+		ret   = bch2_trans_commit_do(ca->fs, NULL, NULL, 0,
 				bch2_disk_accounting_mod(trans, &acc, v, ARRAY_SIZE(v), false)) ?:
 			bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
 		if (ret)

libbcachefs/tests.c

@@ -450,7 +450,7 @@ static int insert_test_overlapping_extent(struct bch_fs *c, u64 inum, u64 start,
 	k.k_i.k.p.snapshot = snapid;
 	k.k_i.k.size = len;
 
-	ret = bch2_trans_do(c, NULL, NULL, 0,
+	ret = bch2_trans_commit_do(c, NULL, NULL, 0,
 		bch2_btree_insert_nonextent(trans, BTREE_ID_extents, &k.k_i,
 					    BTREE_UPDATE_internal_snapshot_node));
 	bch_err_fn(c, ret);
@@ -510,7 +510,7 @@ static int test_snapshots(struct bch_fs *c, u64 nr)
 	if (ret)
 		return ret;
 
-	ret = bch2_trans_do(c, NULL, NULL, 0,
+	ret = bch2_trans_commit_do(c, NULL, NULL, 0,
 		      bch2_snapshot_node_create(trans, U32_MAX,
 						snapids,
 						snapid_subvols,

libbcachefs/xattr.c

@@ -330,7 +330,7 @@ static int bch2_xattr_get_handler(const struct xattr_handler *handler,
 {
 	struct bch_inode_info *inode = to_bch_ei(vinode);
 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
-	int ret = bch2_trans_do(c, NULL, NULL, 0,
+	int ret = bch2_trans_do(c,
 		bch2_xattr_get_trans(trans, inode, name, buffer, size, handler->flags));
 
 	if (ret < 0 && bch2_err_matches(ret, ENOENT))