Update bcachefs sources to ff3a76e1af bcachefs: Change need_whiteout_for_snapshot() to clone iterator

.bcachefs_revision

@@ -1 +1 @@
-635ca475f4f40ddcb2976f8f20a89df4c574aa22
+ff3a76e1af04f51506f45e0f71d53f7e6dd51a75

.gitignore

@@ -18,4 +18,4 @@ tests/__pycache__/
 
 mount/target
 mount.bcachefs
-doc/bcachefs.5.rst
+bcachefs-principles-of-operation.*

Makefile

@@ -28,15 +28,6 @@ PYTEST_CMD?=$(shell \
 )
 PYTEST:=$(PYTEST_CMD) $(PYTEST_ARGS)
 
-RST2MAN_ARGS?=
-RST2MAN_CMD?=$(shell \
-	command -v rst2man \
-	|| which rst2man \
-	|| command -v rst2man.py \
-	|| which rst2man.py \
-)
-RST2MAN:=$(RST2MAN_CMD) $(RST2MAN_ARGS)
-
 CARGO_ARGS=
 CARGO=cargo $(CARGO_ARGS)
 CARGO_PROFILE=release
@@ -108,18 +99,6 @@ TAGS:
 tags:
 	ctags -R .
 
-DOCSRC := opts_macro.h bcachefs.5.rst.tmpl
-DOCGENERATED := bcachefs.5 doc/bcachefs.5.rst
-DOCDEPS := $(addprefix ./doc/,$(DOCSRC))
-bcachefs.5: $(DOCDEPS)  libbcachefs/opts.h
-ifneq (,$(RST2MAN_CMD))
-	$(CC) doc/opts_macro.h -I libbcachefs -I include -E 2>/dev/null	\
-		| doc/macro2rst.py
-	$(RST2MAN) doc/bcachefs.5.rst bcachefs.5
-else
-	@echo "WARNING: no rst2man found! Man page not generated."
-endif
-
 SRCS=$(shell find . -type f -iname '*.c')
 DEPS=$(SRCS:.c=.d)
 -include $(DEPS)
@@ -184,6 +163,11 @@ clean:
 deb: all
 	debuild -us -uc -nc -b -i -I
 
+bcachefs-principles-of-operation.pdf: bcachefs-principles-of-operation.tex
+	pdflatex bcachefs-principles-of-operation.tex && pdflatex bcachefs-principles-of-operation.tex
+
+doc: bcachefs-principles-of-operation.pdf
+
 .PHONY: update-bcachefs-sources
 update-bcachefs-sources:
 	git rm -rf --ignore-unmatch libbcachefs

libbcachefs/bkey_sort.c

@@ -117,23 +117,6 @@ bch2_key_sort_fix_overlapping(struct bch_fs *c, struct bset *dst,
 	return nr;
 }
 
-static void extent_sort_append(struct bch_fs *c,
-			       struct bkey_format *f,
-			       struct btree_nr_keys *nr,
-			       struct bkey_packed **out,
-			       struct bkey_s k)
-{
-	if (!bkey_deleted(k.k)) {
-		if (!bch2_bkey_pack_key(*out, k.k, f))
-			memcpy_u64s_small(*out, k.k, BKEY_U64s);
-
-		memcpy_u64s_small(bkeyp_val(f, *out), k.v, bkey_val_u64s(k.k));
-
-		btree_keys_account_key_add(nr, 0, *out);
-		*out = bkey_next(*out);
-	}
-}
-
 /* Sort + repack in a new format: */
 struct btree_nr_keys
 bch2_sort_repack(struct bset *dst, struct btree *src,
@@ -144,6 +127,7 @@ bch2_sort_repack(struct bset *dst, struct btree *src,
 	struct bkey_format *in_f = &src->format;
 	struct bkey_packed *in, *out = vstruct_last(dst);
 	struct btree_nr_keys nr;
+	bool transform = memcmp(out_f, &src->format, sizeof(*out_f));
 
 	memset(&nr, 0, sizeof(nr));
 
@@ -151,8 +135,10 @@ bch2_sort_repack(struct bset *dst, struct btree *src,
 		if (filter_whiteouts && bkey_deleted(in))
 			continue;
 
-		if (bch2_bkey_transform(out_f, out, bkey_packed(in)
+		if (!transform)
-				       ? in_f : &bch2_bkey_format_current, in))
+			bkey_copy(out, in);
+		else if (bch2_bkey_transform(out_f, out, bkey_packed(in)
+					     ? in_f : &bch2_bkey_format_current, in))
 			out->format = KEY_FORMAT_LOCAL_BTREE;
 		else
 			bch2_bkey_unpack(src, (void *) out, in);
@@ -165,47 +151,6 @@ bch2_sort_repack(struct bset *dst, struct btree *src,
 	return nr;
 }
 
-/* Sort, repack, and call bch2_bkey_normalize() to drop stale pointers: */
-struct btree_nr_keys
-bch2_sort_repack_merge(struct bch_fs *c,
-		       struct bset *dst, struct btree *src,
-		       struct btree_node_iter *iter,
-		       struct bkey_format *out_f,
-		       bool filter_whiteouts)
-{
-	struct bkey_packed *out = vstruct_last(dst), *k_packed;
-	struct bkey_buf k;
-	struct btree_nr_keys nr;
-
-	memset(&nr, 0, sizeof(nr));
-	bch2_bkey_buf_init(&k);
-
-	while ((k_packed = bch2_btree_node_iter_next_all(iter, src))) {
-		if (filter_whiteouts && bkey_deleted(k_packed))
-			continue;
-
-		/*
-		 * NOTE:
-		 * bch2_bkey_normalize may modify the key we pass it (dropping
-		 * stale pointers) and we don't have a write lock on the src
-		 * node; we have to make a copy of the entire key before calling
-		 * normalize
-		 */
-		bch2_bkey_buf_realloc(&k, c, k_packed->u64s + BKEY_U64s);
-		bch2_bkey_unpack(src, k.k, k_packed);
-
-		if (filter_whiteouts &&
-		    bch2_bkey_normalize(c, bkey_i_to_s(k.k)))
-			continue;
-
-		extent_sort_append(c, out_f, &nr, &out, bkey_i_to_s(k.k));
-	}
-
-	dst->u64s = cpu_to_le16((u64 *) out - dst->_data);
-	bch2_bkey_buf_exit(&k, c);
-	return nr;
-}
-
 static inline int sort_keys_cmp(struct btree *b,
 				struct bkey_packed *l,
 				struct bkey_packed *r)

libbcachefs/bkey_sort.h

@@ -37,11 +37,6 @@ struct btree_nr_keys
 bch2_sort_repack(struct bset *, struct btree *,
 		 struct btree_node_iter *,
 		 struct bkey_format *, bool);
-struct btree_nr_keys
-bch2_sort_repack_merge(struct bch_fs *,
-		       struct bset *, struct btree *,
-		       struct btree_node_iter *,
-		       struct bkey_format *, bool);
 
 unsigned bch2_sort_keys(struct bkey_packed *,
 			struct sort_iter *, bool);

libbcachefs/btree_io.c

@@ -391,16 +391,10 @@ void bch2_btree_sort_into(struct bch_fs *c,
 
 	bch2_btree_node_iter_init_from_start(&src_iter, src);
 
-	if (btree_node_is_extents(src))
+	nr = bch2_sort_repack(btree_bset_first(dst),
-		nr = bch2_sort_repack_merge(c, btree_bset_first(dst),
+			src, &src_iter,
-				src, &src_iter,
+			&dst->format,
-				&dst->format,
+			true);
-				true);
-	else
-		nr = bch2_sort_repack(btree_bset_first(dst),
-				src, &src_iter,
-				&dst->format,
-				true);
 
 	bch2_time_stats_update(&c->times[BCH_TIME_btree_node_sort],
 			       start_time);

libbcachefs/btree_iter.c

@@ -155,11 +155,19 @@ void __bch2_btree_node_lock_write(struct btree_trans *trans, struct btree *b)
 	 * goes to 0, and it's safe because we have the node intent
 	 * locked:
 	 */
-	atomic64_sub(__SIX_VAL(read_lock, readers),
+	if (!b->c.lock.readers)
-		     &b->c.lock.state.counter);
+		atomic64_sub(__SIX_VAL(read_lock, readers),
+			     &b->c.lock.state.counter);
+	else
+		this_cpu_sub(*b->c.lock.readers, readers);
+
 	btree_node_lock_type(trans->c, b, SIX_LOCK_write);
-	atomic64_add(__SIX_VAL(read_lock, readers),
+
-		     &b->c.lock.state.counter);
+	if (!b->c.lock.readers)
+		atomic64_add(__SIX_VAL(read_lock, readers),
+			     &b->c.lock.state.counter);
+	else
+		this_cpu_add(*b->c.lock.readers, readers);
 }
 
 bool __bch2_btree_node_relock(struct btree_trans *trans,
@@ -369,19 +377,16 @@ bool __bch2_btree_node_lock(struct btree_trans *trans,
 	if (six_trylock_type(&b->c.lock, type))
 		return true;
 
-#ifdef CONFIG_BCACHEFS_DEBUG
 	trans->locking_path_idx = path->idx;
 	trans->locking_pos	= pos;
 	trans->locking_btree_id	= path->btree_id;
 	trans->locking_level	= level;
 	trans->locking		= b;
-#endif
 
 	ret = six_lock_type(&b->c.lock, type, should_sleep_fn, p) == 0;
 
-#ifdef CONFIG_BCACHEFS_DEBUG
 	trans->locking = NULL;
-#endif
+
 	if (ret)
 		bch2_time_stats_update(&trans->c->times[lock_to_time_stat(type)],
 				       start_time);
@@ -2796,12 +2801,10 @@ void bch2_trans_init(struct btree_trans *trans, struct bch_fs *c,
 
 	trans->srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
 
-#ifdef CONFIG_BCACHEFS_DEBUG
 	trans->pid = current->pid;
 	mutex_lock(&c->btree_trans_lock);
 	list_add(&trans->list, &c->btree_trans_list);
 	mutex_unlock(&c->btree_trans_lock);
-#endif
 }
 
 static void check_btree_paths_leaked(struct btree_trans *trans)
@@ -2840,11 +2843,9 @@ void bch2_trans_exit(struct btree_trans *trans)
 
 	check_btree_paths_leaked(trans);
 
-#ifdef CONFIG_BCACHEFS_DEBUG
 	mutex_lock(&c->btree_trans_lock);
 	list_del(&trans->list);
 	mutex_unlock(&c->btree_trans_lock);
-#endif
 
 	srcu_read_unlock(&c->btree_trans_barrier, trans->srcu_idx);
 
@@ -2888,7 +2889,6 @@ bch2_btree_path_node_to_text(struct printbuf *out,
 	bch2_bpos_to_text(out, btree_node_pos(_b, cached));
 }
 
-#ifdef CONFIG_BCACHEFS_DEBUG
 static bool trans_has_locks(struct btree_trans *trans)
 {
 	struct btree_path *path;
@@ -2898,11 +2898,9 @@ static bool trans_has_locks(struct btree_trans *trans)
 			return true;
 	return false;
 }
-#endif
 
 void bch2_btree_trans_to_text(struct printbuf *out, struct bch_fs *c)
 {
-#ifdef CONFIG_BCACHEFS_DEBUG
 	struct btree_trans *trans;
 	struct btree_path *path;
 	struct btree *b;
@@ -2956,7 +2954,6 @@ void bch2_btree_trans_to_text(struct printbuf *out, struct bch_fs *c)
 		}
 	}
 	mutex_unlock(&c->btree_trans_lock);
-#endif
 }
 
 void bch2_fs_btree_iter_exit(struct bch_fs *c)

libbcachefs/btree_types.h

@@ -366,7 +366,6 @@ struct btree_trans_commit_hook {
 
 struct btree_trans {
 	struct bch_fs		*c;
-#ifdef CONFIG_BCACHEFS_DEBUG
 	struct list_head	list;
 	struct btree		*locking;
 	unsigned		locking_path_idx;
@@ -374,7 +373,6 @@ struct btree_trans {
 	u8			locking_btree_id;
 	u8			locking_level;
 	pid_t			pid;
-#endif
 	unsigned long		ip;
 	int			srcu_idx;
 

libbcachefs/btree_update_leaf.c

@@ -1271,22 +1271,24 @@ err:
  * When deleting, check if we need to emit a whiteout (because we're overwriting
  * something in an ancestor snapshot)
  */
-static int need_whiteout_for_snapshot(struct btree_trans *trans,
+static int need_whiteout_for_snapshot(struct btree_trans *trans, struct btree_iter *orig)
-				      enum btree_id btree_id, struct bpos pos)
 {
 	struct btree_iter iter;
 	struct bkey_s_c k;
-	u32 snapshot = pos.snapshot;
+	u32 snapshot = orig->pos.snapshot;
 	int ret;
 
-	if (!bch2_snapshot_parent(trans->c, pos.snapshot))
+	if (!bch2_snapshot_parent(trans->c, snapshot))
 		return 0;
 
-	pos.snapshot++;
+	bch2_trans_copy_iter(&iter, orig);
+	iter.flags &= BTREE_ITER_FILTER_SNAPSHOTS;
+	iter.flags |= BTREE_ITER_ALL_SNAPSHOTS;
 
-	for_each_btree_key_norestart(trans, iter, btree_id, pos,
+	bch2_btree_iter_advance(&iter);
-			   BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
+
-		if (bkey_cmp(k.k->p, pos))
+	for_each_btree_key_continue_norestart(iter, 0, k, ret) {
+		if (bkey_cmp(k.k->p, orig->pos))
 			break;
 
 		if (bch2_snapshot_is_ancestor(trans->c, snapshot,
@@ -1312,6 +1314,7 @@ int __must_check bch2_trans_update(struct btree_trans *trans, struct btree_iter
 
 	BUG_ON(trans->nr_updates >= BTREE_ITER_MAX);
 	BUG_ON(bpos_cmp(k->k.p, iter->path->pos));
+	BUG_ON(bpos_cmp(k->k.p, iter->pos));
 
 	n = (struct btree_insert_entry) {
 		.flags		= flags,
@@ -1332,7 +1335,7 @@ int __must_check bch2_trans_update(struct btree_trans *trans, struct btree_iter
 
 	if (bkey_deleted(&n.k->k) &&
 	    (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)) {
-		int ret = need_whiteout_for_snapshot(trans, n.btree_id, n.k->k.p);
+		int ret = need_whiteout_for_snapshot(trans, iter);
 		if (unlikely(ret < 0))
 			return ret;
 

libbcachefs/fs-io.c

@@ -2705,7 +2705,8 @@ int bch2_truncate(struct user_namespace *mnt_userns,
 			U64_MAX, &i_sectors_delta);
 	i_sectors_acct(c, inode, NULL, i_sectors_delta);
 
-	BUG_ON(!inode->v.i_size && inode->v.i_blocks);
+	WARN_ON(!inode->v.i_size && inode->v.i_blocks &&
+		!bch2_journal_error(&c->journal));
 
 	if (unlikely(ret))
 		goto err;

libbcachefs/sysfs.c

@@ -262,21 +262,6 @@ static long data_progress_to_text(struct printbuf *out, struct bch_fs *c)
 	return ret;
 }
 
-static int fs_alloc_debug_to_text(struct printbuf *out, struct bch_fs *c)
-{
-	struct bch_fs_usage_online *fs_usage = bch2_fs_usage_read(c);
-
-	if (!fs_usage)
-		return -ENOMEM;
-
-	bch2_fs_usage_to_text(out, c, fs_usage);
-
-	percpu_up_read(&c->mark_lock);
-
-	kfree(fs_usage);
-	return 0;
-}
-
 static int bch2_compression_stats_to_text(struct printbuf *out, struct bch_fs *c)
 {
 	struct btree_trans trans;
@@ -386,9 +371,6 @@ SHOW(bch2_fs)
 
 	/* Debugging: */
 
-	if (attr == &sysfs_alloc_debug)
-		return fs_alloc_debug_to_text(&out, c) ?: out.pos - buf;
-
 	if (attr == &sysfs_journal_debug) {
 		bch2_journal_debug_to_text(&out, &c->journal);
 		return out.pos - buf;
@@ -580,7 +562,6 @@ STORE(bch2_fs_internal)
 SYSFS_OPS(bch2_fs_internal);
 
 struct attribute *bch2_fs_internal_files[] = {
-	&sysfs_alloc_debug,
 	&sysfs_journal_debug,
 	&sysfs_journal_pins,
 	&sysfs_btree_updates,
@@ -588,17 +569,21 @@ struct attribute *bch2_fs_internal_files[] = {
 	&sysfs_btree_cache,
 	&sysfs_btree_key_cache,
 	&sysfs_btree_transactions,
+	&sysfs_new_stripes,
 	&sysfs_stripes_heap,
 	&sysfs_open_buckets,
+	&sysfs_io_timers_read,
+	&sysfs_io_timers_write,
+
+	&sysfs_trigger_journal_flush,
+	&sysfs_trigger_gc,
+	&sysfs_prune_cache,
 
 	&sysfs_read_realloc_races,
 	&sysfs_extent_migrate_done,
 	&sysfs_extent_migrate_raced,
 
-	&sysfs_trigger_journal_flush,
-	&sysfs_trigger_gc,
 	&sysfs_gc_gens_pos,
-	&sysfs_prune_cache,
 
 	&sysfs_copy_gc_enabled,
 	&sysfs_copy_gc_wait,
@@ -607,11 +592,6 @@ struct attribute *bch2_fs_internal_files[] = {
 	&sysfs_rebalance_work,
 	sysfs_pd_controller_files(rebalance),
 
-	&sysfs_new_stripes,
-
-	&sysfs_io_timers_read,
-	&sysfs_io_timers_write,
-
 	&sysfs_data_op_data_progress,
 
 	&sysfs_internal_uuid,

libbcachefs/tests.c

@@ -14,12 +14,14 @@ static void delete_test_keys(struct bch_fs *c)
 	int ret;
 
 	ret = bch2_btree_delete_range(c, BTREE_ID_extents,
-				      POS(0, 0), POS(0, U64_MAX),
+				      SPOS(0, 0, U32_MAX),
+				      SPOS(0, U64_MAX, U32_MAX),
 				      NULL);
 	BUG_ON(ret);
 
 	ret = bch2_btree_delete_range(c, BTREE_ID_xattrs,
-				      POS(0, 0), POS(0, U64_MAX),
+				      SPOS(0, 0, U32_MAX),
+				      SPOS(0, U64_MAX, U32_MAX),
 				      NULL);
 	BUG_ON(ret);
 }
@@ -541,10 +543,11 @@ static int rand_lookup(struct bch_fs *c, u64 nr)
 	u64 i;
 
 	bch2_trans_init(&trans, c, 0, 0);
-	bch2_trans_iter_init(&trans, &iter, BTREE_ID_xattrs, POS_MIN, 0);
+	bch2_trans_iter_init(&trans, &iter, BTREE_ID_xattrs,
+			     SPOS(0, 0, U32_MAX), 0);
 
 	for (i = 0; i < nr; i++) {
-		bch2_btree_iter_set_pos(&iter, POS(0, test_rand()));
+		bch2_btree_iter_set_pos(&iter, SPOS(0, test_rand(), U32_MAX));
 
 		k = bch2_btree_iter_peek(&iter);
 		ret = bkey_err(k);
@@ -567,7 +570,7 @@ static int rand_mixed_trans(struct btree_trans *trans,
 	struct bkey_s_c k;
 	int ret;
 
-	bch2_btree_iter_set_pos(iter, POS(0, pos));
+	bch2_btree_iter_set_pos(iter, SPOS(0, pos, U32_MAX));
 
 	k = bch2_btree_iter_peek(iter);
 	ret = bkey_err(k);
@@ -594,7 +597,8 @@ static int rand_mixed(struct bch_fs *c, u64 nr)
 	u64 i, rand;
 
 	bch2_trans_init(&trans, c, 0, 0);
-	bch2_trans_iter_init(&trans, &iter, BTREE_ID_xattrs, POS_MIN, 0);
+	bch2_trans_iter_init(&trans, &iter, BTREE_ID_xattrs,
+			     SPOS(0, 0, U32_MAX), 0);
 
 	for (i = 0; i < nr; i++) {
 		rand = test_rand();
@@ -673,7 +677,7 @@ static int seq_insert(struct bch_fs *c, u64 nr)
 
 	bch2_trans_init(&trans, c, 0, 0);
 
-	for_each_btree_key(&trans, iter, BTREE_ID_xattrs, POS_MIN,
+	for_each_btree_key(&trans, iter, BTREE_ID_xattrs, SPOS(0, 0, U32_MAX),
 			   BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
 		insert.k.p = iter.pos;
 
@@ -703,7 +707,8 @@ static int seq_lookup(struct bch_fs *c, u64 nr)
 
 	bch2_trans_init(&trans, c, 0, 0);
 
-	for_each_btree_key(&trans, iter, BTREE_ID_xattrs, POS_MIN, 0, k, ret)
+	for_each_btree_key(&trans, iter, BTREE_ID_xattrs,
+			   SPOS(0, 0, U32_MAX), 0, k, ret)
 		;
 	bch2_trans_iter_exit(&trans, &iter);
 
@@ -720,7 +725,8 @@ static int seq_overwrite(struct bch_fs *c, u64 nr)
 
 	bch2_trans_init(&trans, c, 0, 0);
 
-	for_each_btree_key(&trans, iter, BTREE_ID_xattrs, POS_MIN,
+	for_each_btree_key(&trans, iter, BTREE_ID_xattrs,
+			   SPOS(0, 0, U32_MAX),
 			   BTREE_ITER_INTENT, k, ret) {
 		struct bkey_i_cookie u;
 
@@ -745,8 +751,7 @@ static int seq_delete(struct bch_fs *c, u64 nr)
 	int ret;
 
 	ret = bch2_btree_delete_range(c, BTREE_ID_xattrs,
-				      POS(0, 0), POS(0, U64_MAX),
+				      SPOS(0, 0, U32_MAX), POS_MAX, NULL);
-				      NULL);
 	if (ret)
 		bch_err(c, "error in seq_delete: %i", ret);
 	return ret;