Merge tag 'bcachefs-2025-06-04' of git://evilpiepirate.org/bcachefs

Pull more bcachefs updates from Kent Overstreet:
"More bcachefs updates:

- More stack usage improvements (~600 bytes)

- Define CLASS()es for some commonly used types, and convert most
rcu_read_lock() uses to the new lock guards

- New introspection:
- Superblock error counters are now available in sysfs:
previously, they were only visible with 'show-super', which
doesn't provide a live view
- New tracepoint, error_throw(), which is called any time we
return an error and start to unwind

- Repair
- check_fix_ptrs() can now repair btree node roots
- We can now repair when we've somehow ended up with the journal
using a superblock bucket

- Revert some leftovers from the aborted directory i_size feature,
and add repair code: some userspace programs (e.g. sshfs) were
getting confused

It seems in 6.15 there's a bug where i_nlink on the vfs inode has been
getting incorrectly set to 0, with some unfortunate results;
list_journal analysis showed bch2_inode_rm() being called (by
bch2_evict_inode()) when it clearly should not have been.

- bch2_inode_rm() now runs "should we be deleting this inode?" checks
that were previously only run when deleting unlinked inodes in
recovery

- check_subvol() was treating a dangling subvol (pointing to a
missing root inode) like a dangling dirent, and deleting it. This
was the really unfortunate one: check_subvol() will now recreate
the root inode if necessary

This took longer to debug than it should have, and we lost several
filesystems unnecessarily, because users have been ignoring the
release notes and blindly running 'fsck -y'. Debugging required
reconstructing what happened through analyzing the journal, when
ideally someone would have noticed 'hey, fsck is asking me if I want
to repair this: it usually doesn't, maybe I should run this in dry run
mode and check what's going on?'

As a reminder, fsck errors are being marked as autofix once we've
verified, in real world usage, that they're working correctly; blindly
running 'fsck -y' on an experimental filesystem is playing with fire

Up to this incident we've had an excellent track record of not losing
data, so let's try to learn from this one

This is a community effort, I wouldn't be able to get this done
without the help of all the people QAing and providing excellent bug
reports and feedback based on real world usage. But please don't
ignore advice and expect me to pick up the pieces

If an error isn't marked as autofix, and it /is/ happening in the
wild, that's also something I need to know about so we can check it
out and add it to the autofix list if repair looks good. I haven't
been getting those reports, and I should be; since we don't have any
sort of telemetry yet I am absolutely dependent on user reports

Now I'll be spending the weekend working on new repair code to see if
I can get a filesystem back for a user who didn't have backups"

* tag 'bcachefs-2025-06-04' of git://evilpiepirate.org/bcachefs: (69 commits)
bcachefs: add cond_resched() to handle_overwrites()
bcachefs: Make journal read log message a bit quieter
bcachefs: Fix subvol to missing root repair
bcachefs: Run may_delete_deleted_inode() checks in bch2_inode_rm()
bcachefs: delete dead code from may_delete_deleted_inode()
bcachefs: Add flags to subvolume_to_text()
bcachefs: Fix oops in btree_node_seq_matches()
bcachefs: Fix dirent_casefold_mismatch repair
bcachefs: Fix bch2_fsck_rename_dirent() for casefold
bcachefs: Redo bch2_dirent_init_name()
bcachefs: Fix -Wc23-extensions in bch2_check_dirents()
bcachefs: Run check_dirents second time if required
bcachefs: Run snapshot deletion out of system_long_wq
bcachefs: Make check_key_has_snapshot safer
bcachefs: BCH_RECOVERY_PASS_NO_RATELIMIT
bcachefs: bch2_require_recovery_pass()
bcachefs: bch_err_throw()
bcachefs: Repair code for directory i_size
bcachefs: Kill un-reverted directory i_size code
bcachefs: Delete redundant fsck_err()
...

fs/bcachefs/alloc_background.c

···
 #include "error.h"
 #include "lru.h"
 #include "recovery.h"
-#include "trace.h"
 #include "varint.h"
 
 #include <linux/kthread.h>
···
 	a->stripe_sectors	= swab32(a->stripe_sectors);
 }
 
-void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
+static inline void __bch2_alloc_v4_to_text(struct printbuf *out, struct bch_fs *c,
+					   unsigned dev, const struct bch_alloc_v4 *a)
 {
-	struct bch_alloc_v4 _a;
-	const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a);
-	struct bch_dev *ca = c ? bch2_dev_bucket_tryget_noerror(c, k.k->p) : NULL;
+	struct bch_dev *ca = c ? bch2_dev_tryget_noerror(c, dev) : NULL;
 
 	prt_newline(out);
 	printbuf_indent_add(out, 2);
···
 	printbuf_indent_sub(out, 2);
 
 	bch2_dev_put(ca);
+}
+
+void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
+{
+	struct bch_alloc_v4 _a;
+	const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a);
+
+	__bch2_alloc_v4_to_text(out, c, k.k->p.inode, a);
+}
+
+void bch2_alloc_v4_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
+{
+	__bch2_alloc_v4_to_text(out, c, k.k->p.inode, bkey_s_c_to_alloc_v4(k).v);
 }
 
 void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out)
···
 				  set ? "" : "un",
 				  bch2_btree_id_str(btree),
 				  buf.buf);
-	if (ret == -BCH_ERR_fsck_ignore ||
-	    ret == -BCH_ERR_fsck_errors_not_fixed)
+	if (bch2_err_matches(ret, BCH_ERR_fsck_ignore) ||
+	    bch2_err_matches(ret, BCH_ERR_fsck_errors_not_fixed))
 		ret = 0;
 
 	printbuf_exit(&buf);
···
 
 	struct bch_dev *ca = bch2_dev_bucket_tryget(c, new.k->p);
 	if (!ca)
-		return -BCH_ERR_trigger_alloc;
+		return bch_err_throw(c, trigger_alloc);
 
 	struct bch_alloc_v4 old_a_convert;
 	const struct bch_alloc_v4 *old_a = bch2_alloc_to_v4(old, &old_a_convert);
···
 		}
 
 		if (new_a->gen != old_a->gen) {
-			rcu_read_lock();
+			guard(rcu)();
 			u8 *gen = bucket_gen(ca, new.k->p.offset);
-			if (unlikely(!gen)) {
-				rcu_read_unlock();
+			if (unlikely(!gen))
 				goto invalid_bucket;
-			}
 			*gen = new_a->gen;
-			rcu_read_unlock();
 		}
 
 #define eval_state(_a, expr)		({ const struct bch_alloc_v4 *a = _a; expr; })
···
 	}
 
 	if ((flags & BTREE_TRIGGER_gc) && (flags & BTREE_TRIGGER_insert)) {
-		rcu_read_lock();
+		guard(rcu)();
 		struct bucket *g = gc_bucket(ca, new.k->p.offset);
-		if (unlikely(!g)) {
-			rcu_read_unlock();
+		if (unlikely(!g))
 			goto invalid_bucket;
-		}
 		g->gen_valid	= 1;
 		g->gen		= new_a->gen;
-		rcu_read_unlock();
 	}
 err:
 fsck_err:
···
 invalid_bucket:
 	bch2_fs_inconsistent(c, "reference to invalid bucket\n%s",
 			     (bch2_bkey_val_to_text(&buf, c, new.s_c), buf.buf));
-	ret = -BCH_ERR_trigger_alloc;
+	ret = bch_err_throw(c, trigger_alloc);
 	goto err;
 }
 
···
 		bucket->offset = 0;
 	}
 
-	rcu_read_lock();
+	guard(rcu)();
 	*ca = __bch2_next_dev_idx(c, bucket->inode, NULL);
 	if (*ca) {
 		*bucket = POS((*ca)->dev_idx, (*ca)->mi.first_bucket);
 		bch2_dev_get(*ca);
 	}
-	rcu_read_unlock();
 
 	return *ca != NULL;
 }
···
 		ret =   bch2_btree_bit_mod_iter(trans, iter, false) ?:
 			bch2_trans_commit(trans, NULL, NULL,
 				BCH_TRANS_COMMIT_no_enospc) ?:
-			-BCH_ERR_transaction_restart_commit;
+			bch_err_throw(c, transaction_restart_commit);
 		goto out;
 	} else {
 		/*
···
 
 static int discard_in_flight_add(struct bch_dev *ca, u64 bucket, bool in_progress)
 {
+	struct bch_fs *c = ca->fs;
 	int ret;
 
 	mutex_lock(&ca->discard_buckets_in_flight_lock);
-	darray_for_each(ca->discard_buckets_in_flight, i)
-		if (i->bucket == bucket) {
-			ret = -BCH_ERR_EEXIST_discard_in_flight_add;
-			goto out;
-		}
+	struct discard_in_flight *i =
+		darray_find_p(ca->discard_buckets_in_flight, i, i->bucket == bucket);
+	if (i) {
+		ret = bch_err_throw(c, EEXIST_discard_in_flight_add);
+		goto out;
+	}
 
 	ret = darray_push(&ca->discard_buckets_in_flight, ((struct discard_in_flight) {
 			   .in_progress = in_progress,
···
 static void discard_in_flight_remove(struct bch_dev *ca, u64 bucket)
 {
 	mutex_lock(&ca->discard_buckets_in_flight_lock);
-	darray_for_each(ca->discard_buckets_in_flight, i)
-		if (i->bucket == bucket) {
-			BUG_ON(!i->in_progress);
-			darray_remove_item(&ca->discard_buckets_in_flight, i);
-			goto found;
-		}
-	BUG();
-found:
+	struct discard_in_flight *i =
+		darray_find_p(ca->discard_buckets_in_flight, i, i->bucket == bucket);
+	BUG_ON(!i || !i->in_progress);
+
+	darray_remove_item(&ca->discard_buckets_in_flight, i);
 	mutex_unlock(&ca->discard_buckets_in_flight_lock);
 }
 
···
 
 	lockdep_assert_held(&c->state_lock);
 
-	rcu_read_lock();
+	guard(rcu)();
 	for_each_member_device_rcu(c, ca, NULL) {
 		struct block_device *bdev = READ_ONCE(ca->disk_sb.bdev);
 		if (bdev)
···
 		bucket_size_max = max_t(unsigned, bucket_size_max,
 					ca->mi.bucket_size);
 	}
-	rcu_read_unlock();
 
 	bch2_set_ra_pages(c, ra_pages);
 
···
 {
 	u64 ret = U64_MAX;
 
-	rcu_read_lock();
+	guard(rcu)();
 	for_each_rw_member_rcu(c, ca)
 		ret = min(ret, ca->mi.nbuckets * ca->mi.bucket_size);
-	rcu_read_unlock();
 	return ret;
 }
 

fs/bcachefs/alloc_background.h

···
 
 static inline bool bch2_dev_bucket_exists(struct bch_fs *c, struct bpos pos)
 {
-	rcu_read_lock();
+	guard(rcu)();
 	struct bch_dev *ca = bch2_dev_rcu_noerror(c, pos.inode);
-	bool ret = ca && bucket_valid(ca, pos.offset);
-	rcu_read_unlock();
-	return ret;
+	return ca && bucket_valid(ca, pos.offset);
 }
 
 static inline u64 bucket_to_u64(struct bpos bucket)
···
 			   struct bkey_validate_context);
 void bch2_alloc_v4_swab(struct bkey_s);
 void bch2_alloc_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
+void bch2_alloc_v4_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 
 #define bch2_bkey_ops_alloc ((struct bkey_ops) {	\
 	.key_validate	= bch2_alloc_v1_validate,	\
···
 
 #define bch2_bkey_ops_alloc_v4 ((struct bkey_ops) {	\
 	.key_validate	= bch2_alloc_v4_validate,	\
-	.val_to_text	= bch2_alloc_to_text,		\
+	.val_to_text	= bch2_alloc_v4_to_text,	\
 	.swab		= bch2_alloc_v4_swab,		\
 	.trigger	= bch2_trigger_alloc,		\
 	.min_val_size	= 48,				\

fs/bcachefs/alloc_foreground.c

···
 
 void bch2_reset_alloc_cursors(struct bch_fs *c)
 {
-	rcu_read_lock();
+	guard(rcu)();
 	for_each_member_device_rcu(c, ca, NULL)
 		memset(ca->alloc_cursor, 0, sizeof(ca->alloc_cursor));
-	rcu_read_unlock();
 }
 
 static void bch2_open_bucket_hash_add(struct bch_fs *c, struct open_bucket *ob)
···
 	       ARRAY_SIZE(c->open_buckets_partial));
 
 	spin_lock(&c->freelist_lock);
-	rcu_read_lock();
-	bch2_dev_rcu(c, ob->dev)->nr_partial_buckets++;
-	rcu_read_unlock();
+	scoped_guard(rcu)
+		bch2_dev_rcu(c, ob->dev)->nr_partial_buckets++;
 
 	ob->on_partial_list = true;
 	c->open_buckets_partial[c->open_buckets_partial_nr++] =
···
 
 		track_event_change(&c->times[BCH_TIME_blocked_allocate_open_bucket], true);
 		spin_unlock(&c->freelist_lock);
-		return ERR_PTR(-BCH_ERR_open_buckets_empty);
+		return ERR_PTR(bch_err_throw(c, open_buckets_empty));
 	}
 
 	/* Recheck under lock: */
···
 
 		track_event_change(&c->times[BCH_TIME_blocked_allocate], true);
 
-		ob = ERR_PTR(-BCH_ERR_freelist_empty);
+		ob = ERR_PTR(bch_err_throw(c, freelist_empty));
 		goto err;
 	}
 
···
 	}
 err:
 	if (!ob)
-		ob = ERR_PTR(-BCH_ERR_no_buckets_found);
+		ob = ERR_PTR(bch_err_throw(c, no_buckets_found));
 
 	if (!IS_ERR(ob))
 		ob->data_type = req->data_type;
···
 
 #define dev_stripe_cmp(l, r) __dev_stripe_cmp(stripe, l, r)
 
-struct dev_alloc_list bch2_dev_alloc_list(struct bch_fs *c,
-					  struct dev_stripe_state *stripe,
-					  struct bch_devs_mask *devs)
+void bch2_dev_alloc_list(struct bch_fs *c,
+			 struct dev_stripe_state *stripe,
+			 struct bch_devs_mask *devs,
+			 struct dev_alloc_list *ret)
 {
-	struct dev_alloc_list ret = { .nr = 0 };
+	ret->nr = 0;
+
 	unsigned i;
-
 	for_each_set_bit(i, devs->d, BCH_SB_MEMBERS_MAX)
-		ret.data[ret.nr++] = i;
+		ret->data[ret->nr++] = i;
 
-	bubble_sort(ret.data, ret.nr, dev_stripe_cmp);
-	return ret;
+	bubble_sort(ret->data, ret->nr, dev_stripe_cmp);
 }
 
 static const u64 stripe_clock_hand_rescale	= 1ULL << 62; /* trigger rescale at */
···
 	return 0;
 }
 
-int bch2_bucket_alloc_set_trans(struct btree_trans *trans,
-				struct alloc_request *req,
-				struct dev_stripe_state *stripe,
-				struct closure *cl)
+inline int bch2_bucket_alloc_set_trans(struct btree_trans *trans,
+				       struct alloc_request *req,
+				       struct dev_stripe_state *stripe,
+				       struct closure *cl)
 {
 	struct bch_fs *c = trans->c;
-	int ret = -BCH_ERR_insufficient_devices;
+	int ret = 0;
 
 	BUG_ON(req->nr_effective >= req->nr_replicas);
 
-	struct dev_alloc_list devs_sorted = bch2_dev_alloc_list(c, stripe, &req->devs_may_alloc);
-	darray_for_each(devs_sorted, i) {
+	bch2_dev_alloc_list(c, stripe, &req->devs_may_alloc, &req->devs_sorted);
+
+	darray_for_each(req->devs_sorted, i) {
 		req->ca = bch2_dev_tryget_noerror(c, *i);
 		if (!req->ca)
 			continue;
···
 			continue;
 		}
 
-		if (add_new_bucket(c, req, ob)) {
-			ret = 0;
+		ret = add_new_bucket(c, req, ob);
+		if (ret)
 			break;
-		}
 	}
 
-	return ret;
+	if (ret == 1)
+		return 0;
+	if (ret)
+		return ret;
+	return bch_err_throw(c, insufficient_devices);
 }
 
 /* Allocate from stripes: */
···
 	if (!h)
 		return 0;
 
-	struct dev_alloc_list devs_sorted =
-		bch2_dev_alloc_list(c, &req->wp->stripe, &req->devs_may_alloc);
-	darray_for_each(devs_sorted, i)
+	bch2_dev_alloc_list(c, &req->wp->stripe, &req->devs_may_alloc, &req->devs_sorted);
+
+	darray_for_each(req->devs_sorted, i)
 		for (unsigned ec_idx = 0; ec_idx < h->s->nr_data; ec_idx++) {
 			if (!h->s->blocks[ec_idx])
 				continue;
···
 					  i);
 			ob->on_partial_list = false;
 
-			rcu_read_lock();
-			bch2_dev_rcu(c, ob->dev)->nr_partial_buckets--;
-			rcu_read_unlock();
+			scoped_guard(rcu)
+				bch2_dev_rcu(c, ob->dev)->nr_partial_buckets--;
 
 			ret = add_new_bucket(c, req, ob);
 			if (ret)
···
 
 			ob->on_partial_list = false;
 
-			rcu_read_lock();
-			bch2_dev_rcu(c, ob->dev)->nr_partial_buckets--;
-			rcu_read_unlock();
+			scoped_guard(rcu)
+				bch2_dev_rcu(c, ob->dev)->nr_partial_buckets--;
 
 			spin_unlock(&c->freelist_lock);
 			bch2_open_bucket_put(c, ob);
···
 {
 	struct write_point *wp;
 
-	rcu_read_lock();
+	guard(rcu)();
 	hlist_for_each_entry_rcu(wp, head, node)
 		if (wp->write_point == write_point)
-			goto out;
-	wp = NULL;
-out:
-	rcu_read_unlock();
-	return wp;
+			return wp;
+	return NULL;
 }
 
 static inline bool too_many_writepoints(struct bch_fs *c, unsigned factor)
···
 	return stranded * factor > free;
 }
 
-static bool try_increase_writepoints(struct bch_fs *c)
+static noinline bool try_increase_writepoints(struct bch_fs *c)
 {
 	struct write_point *wp;
 
···
 	return true;
 }
 
-static bool try_decrease_writepoints(struct btree_trans *trans, unsigned old_nr)
+static noinline bool try_decrease_writepoints(struct btree_trans *trans, unsigned old_nr)
 {
 	struct bch_fs *c = trans->c;
 	struct write_point *wp;
···
 		goto retry;
 
 	if (cl && bch2_err_matches(ret, BCH_ERR_open_buckets_empty))
-		ret = -BCH_ERR_bucket_alloc_blocked;
+		ret = bch_err_throw(c, bucket_alloc_blocked);
 
 	if (cl && !(flags & BCH_WRITE_alloc_nowait) &&
 	    bch2_err_matches(ret, BCH_ERR_freelist_empty))
-		ret = -BCH_ERR_bucket_alloc_blocked;
+		ret = bch_err_throw(c, bucket_alloc_blocked);
 
 	return ret;
 }
···
 
 	bch2_printbuf_make_room(&buf, 4096);
 
-	rcu_read_lock();
 	buf.atomic++;
-
-	for_each_online_member_rcu(c, ca) {
-		prt_printf(&buf, "Dev %u:\n", ca->dev_idx);
-		printbuf_indent_add(&buf, 2);
-		bch2_dev_alloc_debug_to_text(&buf, ca);
-		printbuf_indent_sub(&buf, 2);
-		prt_newline(&buf);
-	}
-
+	scoped_guard(rcu)
+		for_each_online_member_rcu(c, ca) {
+			prt_printf(&buf, "Dev %u:\n", ca->dev_idx);
+			printbuf_indent_add(&buf, 2);
+			bch2_dev_alloc_debug_to_text(&buf, ca);
+			printbuf_indent_sub(&buf, 2);
+			prt_newline(&buf);
+		}
 	--buf.atomic;
-	rcu_read_unlock();
 
 	prt_printf(&buf, "Copygc debug:\n");
 	printbuf_indent_add(&buf, 2);

fs/bcachefs/alloc_foreground.h

···
 	struct bch_devs_mask	devs_may_alloc;
 
 	/* bch2_bucket_alloc_set_trans(): */
+	struct dev_alloc_list	devs_sorted;
 	struct bch_dev_usage	usage;
 
 	/* bch2_bucket_alloc_trans(): */
···
 	struct bch_devs_mask	scratch_devs_may_alloc;
 };
 
-struct dev_alloc_list bch2_dev_alloc_list(struct bch_fs *,
-					  struct dev_stripe_state *,
-					  struct bch_devs_mask *);
+void bch2_dev_alloc_list(struct bch_fs *,
+			 struct dev_stripe_state *,
+			 struct bch_devs_mask *,
+			 struct dev_alloc_list *);
 void bch2_dev_stripe_increment(struct bch_dev *, struct dev_stripe_state *);
 
 static inline struct bch_dev *ob_dev(struct bch_fs *c, struct open_bucket *ob)

fs/bcachefs/backpointers.c

···
 {
 	struct bkey_s_c_backpointer bp = bkey_s_c_to_backpointer(k);
 
-	rcu_read_lock();
-	struct bch_dev *ca = bch2_dev_rcu_noerror(c, bp.k->p.inode);
-	if (ca) {
-		u32 bucket_offset;
-		struct bpos bucket = bp_pos_to_bucket_and_offset(ca, bp.k->p, &bucket_offset);
-		rcu_read_unlock();
-		prt_printf(out, "bucket=%llu:%llu:%u ", bucket.inode, bucket.offset, bucket_offset);
-	} else {
-		rcu_read_unlock();
-		prt_printf(out, "sector=%llu:%llu ", bp.k->p.inode, bp.k->p.offset >> MAX_EXTENT_COMPRESS_RATIO_SHIFT);
+	struct bch_dev *ca;
+	u32 bucket_offset;
+	struct bpos bucket;
+	scoped_guard(rcu) {
+		ca = bch2_dev_rcu_noerror(c, bp.k->p.inode);
+		if (ca)
+			bucket = bp_pos_to_bucket_and_offset(ca, bp.k->p, &bucket_offset);
 	}
+
+	if (ca)
+		prt_printf(out, "bucket=%llu:%llu:%u ", bucket.inode, bucket.offset, bucket_offset);
+	else
+		prt_printf(out, "sector=%llu:%llu ", bp.k->p.inode, bp.k->p.offset >> MAX_EXTENT_COMPRESS_RATIO_SHIFT);
 
 	bch2_btree_id_level_to_text(out, bp.v->btree_id, bp.v->level);
 	prt_str(out, " data_type=");
···
 	}
 
 	if (!will_check && __bch2_inconsistent_error(c, &buf))
-		ret = -BCH_ERR_erofs_unfixed_errors;
+		ret = bch_err_throw(c, erofs_unfixed_errors);
 
 	bch_err(c, "%s", buf.buf);
 	printbuf_exit(&buf);
···
 		return b;
 
 	if (btree_node_will_make_reachable(b)) {
-		b = ERR_PTR(-BCH_ERR_backpointer_to_overwritten_btree_node);
+		b = ERR_PTR(bch_err_throw(c, backpointer_to_overwritten_btree_node));
 	} else {
 		int ret = backpointer_target_not_found(trans, bp, bkey_i_to_s_c(&b->key),
 						       last_flushed, commit);
···
 		return ret ? bkey_s_c_err(ret) : bkey_s_c_null;
 	} else {
 		struct btree *b = __bch2_backpointer_get_node(trans, bp, iter, last_flushed, commit);
-		if (b == ERR_PTR(-BCH_ERR_backpointer_to_overwritten_btree_node))
+		if (b == ERR_PTR(bch_err_throw(c, backpointer_to_overwritten_btree_node)))
 			return bkey_s_c_null;
 		if (IS_ERR_OR_NULL(b))
 			return ((struct bkey_s_c) { .k = ERR_CAST(b) });
···
 		bkey_for_each_ptr(other_extent_ptrs, ptr)
 			if (ptr->dev == bp->k.p.inode &&
 			    dev_ptr_stale_rcu(ca, ptr)) {
+				rcu_read_unlock();
 				ret = drop_dev_and_update(trans, other_bp.v->btree_id,
 							  other_extent, bp->k.p.inode);
 				if (ret)
···
 	prt_newline(&buf);
 	bch2_bkey_val_to_text(&buf, c, other_extent);
 	bch_err(c, "%s", buf.buf);
-	ret = -BCH_ERR_fsck_repair_unimplemented;
+	ret = bch_err_throw(c, fsck_repair_unimplemented);
 	goto err;
 missing:
 	printbuf_reset(&buf);
···
 		if (p.ptr.dev == BCH_SB_MEMBER_INVALID)
 			continue;
 
-		rcu_read_lock();
-		struct bch_dev *ca = bch2_dev_rcu_noerror(c, p.ptr.dev);
-		if (!ca) {
-			rcu_read_unlock();
-			continue;
-		}
+		bool empty;
+		{
+			/* scoped_guard() is a loop, so it breaks continue */
+			guard(rcu)();
+			struct bch_dev *ca = bch2_dev_rcu_noerror(c, p.ptr.dev);
+			if (!ca)
+				continue;
 
-		if (p.ptr.cached && dev_ptr_stale_rcu(ca, &p.ptr)) {
-			rcu_read_unlock();
-			continue;
-		}
+			if (p.ptr.cached && dev_ptr_stale_rcu(ca, &p.ptr))
+				continue;
 
-		u64 b = PTR_BUCKET_NR(ca, &p.ptr);
-		if (!bch2_bucket_bitmap_test(&ca->bucket_backpointer_mismatch, b)) {
-			rcu_read_unlock();
-			continue;
-		}
+			u64 b = PTR_BUCKET_NR(ca, &p.ptr);
+			if (!bch2_bucket_bitmap_test(&ca->bucket_backpointer_mismatch, b))
+				continue;
 
-		bool empty = bch2_bucket_bitmap_test(&ca->bucket_backpointer_empty, b);
-		rcu_read_unlock();
+			empty = bch2_bucket_bitmap_test(&ca->bucket_backpointer_empty, b);
+		}
 
 		struct bkey_i_backpointer bp;
 		bch2_extent_ptr_to_bp(c, btree, level, k, p, entry, &bp);
···
 		    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;
+				bch_err_throw(c, transaction_restart_nested);
 			goto err;
 		}
 
···
 	case KEY_TYPE_btree_ptr_v2: {
 		bool ret = false;
 
-		rcu_read_lock();
+		guard(rcu)();
 		struct bpos pos = bkey_s_c_to_btree_ptr_v2(k).v->min_key;
 		while (pos.inode <= k.k->p.inode) {
 			if (pos.inode >= c->sb.nr_devices)
···
 next:
 			pos = SPOS(pos.inode + 1, 0, 0);
 		}
-		rcu_read_unlock();
 
 		return ret;
 	}
···
 			b->buckets = kvcalloc(BITS_TO_LONGS(ca->mi.nbuckets),
 					      sizeof(unsigned long), GFP_KERNEL);
 			if (!b->buckets)
-				return -BCH_ERR_ENOMEM_backpointer_mismatches_bitmap;
+				return bch_err_throw(ca->fs, ENOMEM_backpointer_mismatches_bitmap);
 		}
 
 		b->nr += !__test_and_set_bit(bit, b->buckets);
···
 	return 0;
 }
 
-int bch2_bucket_bitmap_resize(struct bucket_bitmap *b, u64 old_size, u64 new_size)
+int bch2_bucket_bitmap_resize(struct bch_dev *ca, struct bucket_bitmap *b,
+			      u64 old_size, u64 new_size)
 {
 	scoped_guard(mutex, &b->lock) {
 		if (!b->buckets)
···
 		unsigned long *n = kvcalloc(BITS_TO_LONGS(new_size),
 					    sizeof(unsigned long), GFP_KERNEL);
 		if (!n)
-			return -BCH_ERR_ENOMEM_backpointer_mismatches_bitmap;
+			return bch_err_throw(ca->fs, ENOMEM_backpointer_mismatches_bitmap);
 
 		memcpy(n, b->buckets,
 		       BITS_TO_LONGS(min(old_size, new_size)) * sizeof(unsigned long));

fs/bcachefs/backpointers.h

···
 
 static inline bool bp_pos_to_bucket_nodev_noerror(struct bch_fs *c, struct bpos bp_pos, struct bpos *bucket)
 {
-	rcu_read_lock();
+	guard(rcu)();
 	struct bch_dev *ca = bch2_dev_rcu_noerror(c, bp_pos.inode);
 	if (ca)
 		*bucket = bp_pos_to_bucket(ca, bp_pos);
-	rcu_read_unlock();
 	return ca != NULL;
 }
 
···
 	return bitmap && test_bit(i, bitmap);
 }
 
-int bch2_bucket_bitmap_resize(struct bucket_bitmap *, u64, u64);
+int bch2_bucket_bitmap_resize(struct bch_dev *, struct bucket_bitmap *, u64, u64);
 void bch2_bucket_bitmap_free(struct bucket_bitmap *);
 
 #endif /* _BCACHEFS_BACKPOINTERS_BACKGROUND_H */

fs/bcachefs/bcachefs.h

···
 #define pr_fmt(fmt) "%s() " fmt "\n", __func__
 #endif
 
+#ifdef CONFIG_BCACHEFS_DEBUG
+#define ENUMERATED_REF_DEBUG
+#endif
+
+#ifndef dynamic_fault
+#define dynamic_fault(...)		0
+#endif
+
+#define race_fault(...)			dynamic_fault("bcachefs:race")
+
 #include <linux/backing-dev-defs.h>
 #include <linux/bug.h>
 #include <linux/bio.h>
···
 #include "time_stats.h"
 #include "util.h"
 
-#ifdef CONFIG_BCACHEFS_DEBUG
-#define ENUMERATED_REF_DEBUG
-#endif
+#include "alloc_types.h"
+#include "async_objs_types.h"
+#include "btree_gc_types.h"
+#include "btree_types.h"
+#include "btree_node_scan_types.h"
+#include "btree_write_buffer_types.h"
+#include "buckets_types.h"
+#include "buckets_waiting_for_journal_types.h"
+#include "clock_types.h"
+#include "disk_groups_types.h"
+#include "ec_types.h"
+#include "enumerated_ref_types.h"
+#include "journal_types.h"
+#include "keylist_types.h"
+#include "quota_types.h"
+#include "rebalance_types.h"
+#include "recovery_passes_types.h"
+#include "replicas_types.h"
+#include "sb-members_types.h"
+#include "subvolume_types.h"
+#include "super_types.h"
+#include "thread_with_file_types.h"
 
-#ifndef dynamic_fault
-#define dynamic_fault(...)		0
-#endif
-
-#define race_fault(...)			dynamic_fault("bcachefs:race")
+#include "trace.h"
 
 #define count_event(_c, _name)	this_cpu_inc((_c)->counters[BCH_COUNTER_##_name])
 
···
 		pr_info(fmt, ##__VA_ARGS__);				\
 } while (0)
 
+static inline int __bch2_err_trace(struct bch_fs *c, int err)
+{
+	trace_error_throw(c, err, _THIS_IP_);
+	return err;
+}
+
+#define bch_err_throw(_c, _err) __bch2_err_trace(_c, -BCH_ERR_##_err)
+
 /* Parameters that are useful for debugging, but should always be compiled in: */
 #define BCH_DEBUG_PARAMS_ALWAYS()					\
 	BCH_DEBUG_PARAM(key_merging_disabled,				\
···
 #undef x
 	BCH_TIME_STAT_NR
 };
-
-#include "alloc_types.h"
-#include "async_objs_types.h"
-#include "btree_gc_types.h"
-#include "btree_types.h"
-#include "btree_node_scan_types.h"
-#include "btree_write_buffer_types.h"
-#include "buckets_types.h"
-#include "buckets_waiting_for_journal_types.h"
-#include "clock_types.h"
-#include "disk_groups_types.h"
-#include "ec_types.h"
-#include "enumerated_ref_types.h"
-#include "journal_types.h"
-#include "keylist_types.h"
-#include "quota_types.h"
-#include "rebalance_types.h"
-#include "recovery_passes_types.h"
-#include "replicas_types.h"
-#include "sb-members_types.h"
-#include "subvolume_types.h"
-#include "super_types.h"
-#include "thread_with_file_types.h"
 
 /* Number of nodes btree coalesce will try to coalesce at once */
 #define GC_MERGE_NODES		4U

fs/bcachefs/btree_cache.c

···
 
 	b->data = kvmalloc(btree_buf_bytes(b), gfp);
 	if (!b->data)
-		return -BCH_ERR_ENOMEM_btree_node_mem_alloc;
+		return bch_err_throw(c, ENOMEM_btree_node_mem_alloc);
 #ifdef __KERNEL__
 	b->aux_data = kvmalloc(btree_aux_data_bytes(b), gfp);
 #else
···
 	if (!b->aux_data) {
 		kvfree(b->data);
 		b->data = NULL;
-		return -BCH_ERR_ENOMEM_btree_node_mem_alloc;
+		return bch_err_throw(c, ENOMEM_btree_node_mem_alloc);
 	}
 
 	return 0;
···
 
 	if (btree_node_noevict(b)) {
 		bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_noevict]++;
-		return -BCH_ERR_ENOMEM_btree_node_reclaim;
+		return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 	}
 	if (btree_node_write_blocked(b)) {
 		bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_write_blocked]++;
-		return -BCH_ERR_ENOMEM_btree_node_reclaim;
+		return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 	}
 	if (btree_node_will_make_reachable(b)) {
 		bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_will_make_reachable]++;
-		return -BCH_ERR_ENOMEM_btree_node_reclaim;
+		return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 	}
 
 	if (btree_node_dirty(b)) {
 		if (!flush) {
 			bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_dirty]++;
-			return -BCH_ERR_ENOMEM_btree_node_reclaim;
+			return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 		}
 
 		if (locked) {
···
 				bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_read_in_flight]++;
 			else if (btree_node_write_in_flight(b))
 				bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_write_in_flight]++;
-			return -BCH_ERR_ENOMEM_btree_node_reclaim;
+			return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 		}
 
 		if (locked)
···
 
 	if (!six_trylock_intent(&b->c.lock)) {
 		bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_lock_intent]++;
-		return -BCH_ERR_ENOMEM_btree_node_reclaim;
+		return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 	}
 
 	if (!six_trylock_write(&b->c.lock)) {
 		bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_lock_write]++;
 		six_unlock_intent(&b->c.lock);
-		return -BCH_ERR_ENOMEM_btree_node_reclaim;
+		return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 	}
 
 	/* recheck under lock */
···
 
 	return 0;
 err:
-	return -BCH_ERR_ENOMEM_fs_btree_cache_init;
+	return bch_err_throw(c, ENOMEM_fs_btree_cache_init);
 }
 
 void bch2_fs_btree_cache_init_early(struct btree_cache *bc)
···
 
 	if (!cl) {
 		trace_and_count(c, btree_cache_cannibalize_lock_fail, trans);
-		return -BCH_ERR_ENOMEM_btree_cache_cannibalize_lock;
+		return bch_err_throw(c, ENOMEM_btree_cache_cannibalize_lock);
 	}
 
 	closure_wait(&bc->alloc_wait, cl);
···
 	}
 
 	trace_and_count(c, btree_cache_cannibalize_lock_fail, trans);
-	return -BCH_ERR_btree_cache_cannibalize_lock_blocked;
+	return bch_err_throw(c, btree_cache_cannibalize_lock_blocked);
 
 success:
 	trace_and_count(c, btree_cache_cannibalize_lock, trans);

fs/bcachefs/btree_gc.c

···
 
 	new = kmalloc_array(BKEY_BTREE_PTR_U64s_MAX, sizeof(u64), GFP_KERNEL);
 	if (!new)
-		return -BCH_ERR_ENOMEM_gc_repair_key;
+		return bch_err_throw(c, ENOMEM_gc_repair_key);
 
 	btree_ptr_to_v2(b, new);
 	b->data->min_key	= new_min;
···
 
 	new = kmalloc_array(BKEY_BTREE_PTR_U64s_MAX, sizeof(u64), GFP_KERNEL);
 	if (!new)
-		return -BCH_ERR_ENOMEM_gc_repair_key;
+		return bch_err_throw(c, ENOMEM_gc_repair_key);
 
 	btree_ptr_to_v2(b, new);
 	b->data->max_key	= new_max;
···
 		ret = genradix_prealloc(&ca->buckets_gc, ca->mi.nbuckets, GFP_KERNEL);
 		if (ret) {
 			bch2_dev_put(ca);
-			ret = -BCH_ERR_ENOMEM_gc_alloc_start;
+			ret = bch_err_throw(c, ENOMEM_gc_alloc_start);
 			break;
 		}
 	}
···
 {
 	struct bch_fs *c = trans->c;
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
-	struct bkey_i *u;
-	int ret;
 
 	if (unlikely(test_bit(BCH_FS_going_ro, &c->flags)))
 		return -EROFS;
 
-	rcu_read_lock();
-	bkey_for_each_ptr(ptrs, ptr) {
-		struct bch_dev *ca = bch2_dev_rcu(c, ptr->dev);
-		if (!ca)
-			continue;
+	bool too_stale = false;
+	scoped_guard(rcu) {
+		bkey_for_each_ptr(ptrs, ptr) {
+			struct bch_dev *ca = bch2_dev_rcu(c, ptr->dev);
+			if (!ca)
+				continue;
 
-		if (dev_ptr_stale(ca, ptr) > 16) {
-			rcu_read_unlock();
-			goto update;
+			too_stale |= dev_ptr_stale(ca, ptr) > 16;
 		}
+
+		if (!too_stale)
+			bkey_for_each_ptr(ptrs, ptr) {
+				struct bch_dev *ca = bch2_dev_rcu(c, ptr->dev);
+				if (!ca)
+					continue;
+
+				u8 *gen = &ca->oldest_gen[PTR_BUCKET_NR(ca, ptr)];
+				if (gen_after(*gen, ptr->gen))
+					*gen = ptr->gen;
+			}
 	}
 
-	bkey_for_each_ptr(ptrs, ptr) {
-		struct bch_dev *ca = bch2_dev_rcu(c, ptr->dev);
-		if (!ca)
-			continue;
+	if (too_stale) {
+		struct bkey_i *u = bch2_bkey_make_mut(trans, iter, &k, 0);
+		int ret = PTR_ERR_OR_ZERO(u);
+		if (ret)
+			return ret;
 
-		u8 *gen = &ca->oldest_gen[PTR_BUCKET_NR(ca, ptr)];
-		if (gen_after(*gen, ptr->gen))
-			*gen = ptr->gen;
+		bch2_extent_normalize(c, bkey_i_to_s(u));
 	}
-	rcu_read_unlock();
-	return 0;
-update:
-	u = bch2_bkey_make_mut(trans, iter, &k, 0);
-	ret = PTR_ERR_OR_ZERO(u);
-	if (ret)
-		return ret;
 
-	bch2_extent_normalize(c, bkey_i_to_s(u));
 	return 0;
 }
 
···
 		ca->oldest_gen = kvmalloc(gens->nbuckets, GFP_KERNEL);
 		if (!ca->oldest_gen) {
 			bch2_dev_put(ca);
-			ret = -BCH_ERR_ENOMEM_gc_gens;
+			ret = bch_err_throw(c, ENOMEM_gc_gens);
 			goto err;
 		}
 

fs/bcachefs/btree_io.c

···
 		       const char *fmt, ...)
 {
 	if (c->recovery.curr_pass == BCH_RECOVERY_PASS_scan_for_btree_nodes)
-		return -BCH_ERR_fsck_fix;
+		return bch_err_throw(c, fsck_fix);
 
 	bool have_retry = false;
 	int ret2;
···
 	}
 
 	if (!have_retry && ret == -BCH_ERR_btree_node_read_err_want_retry)
-		ret = -BCH_ERR_btree_node_read_err_fixable;
+		ret = bch_err_throw(c, btree_node_read_err_fixable);
 	if (!have_retry && ret == -BCH_ERR_btree_node_read_err_must_retry)
-		ret = -BCH_ERR_btree_node_read_err_bad_node;
+		ret = bch_err_throw(c, btree_node_read_err_bad_node);
 
 	bch2_sb_error_count(c, err_type);
 
···
 		switch (ret) {
 		case -BCH_ERR_btree_node_read_err_fixable:
 			ret2 = bch2_fsck_err_opt(c, FSCK_CAN_FIX, err_type);
-			if (ret2 != -BCH_ERR_fsck_fix &&
-			    ret2 != -BCH_ERR_fsck_ignore) {
+			if (!bch2_err_matches(ret2, BCH_ERR_fsck_fix) &&
+			    !bch2_err_matches(ret2, BCH_ERR_fsck_ignore)) {
 				ret = ret2;
 				goto fsck_err;
 			}
 
 			if (!have_retry)
-				ret = -BCH_ERR_fsck_fix;
+				ret = bch_err_throw(c, fsck_fix);
 			goto out;
 		case -BCH_ERR_btree_node_read_err_bad_node:
 			prt_str(&out, ", ");
···
 	switch (ret) {
 	case -BCH_ERR_btree_node_read_err_fixable:
 		ret2 = __bch2_fsck_err(c, NULL, FSCK_CAN_FIX, err_type, "%s", out.buf);
-		if (ret2 != -BCH_ERR_fsck_fix &&
-		    ret2 != -BCH_ERR_fsck_ignore) {
+		if (!bch2_err_matches(ret2, BCH_ERR_fsck_fix) &&
+		    !bch2_err_matches(ret2, BCH_ERR_fsck_ignore)) {
 			ret = ret2;
 			goto fsck_err;
 		}
 
 		if (!have_retry)
-			ret = -BCH_ERR_fsck_fix;
+			ret = bch_err_throw(c, fsck_fix);
 		goto out;
 	case -BCH_ERR_btree_node_read_err_bad_node:
 		prt_str(&out, ", ");
···
 			       failed, err_msg,				\
 			       msg, ##__VA_ARGS__);			\
 									\
-	if (_ret != -BCH_ERR_fsck_fix) {				\
+	if (!bch2_err_matches(_ret, BCH_ERR_fsck_fix)) {		\
 		ret = _ret;						\
 		goto fsck_err;						\
 	}								\
···
 
 	btree_node_reset_sib_u64s(b);
 
-	rcu_read_lock();
-	bkey_for_each_ptr(bch2_bkey_ptrs(bkey_i_to_s(&b->key)), ptr) {
-		struct bch_dev *ca2 = bch2_dev_rcu(c, ptr->dev);
+	scoped_guard(rcu)
+		bkey_for_each_ptr(bch2_bkey_ptrs(bkey_i_to_s(&b->key)), ptr) {
+			struct bch_dev *ca2 = bch2_dev_rcu(c, ptr->dev);
 
-		if (!ca2 || ca2->mi.state != BCH_MEMBER_STATE_rw)
-			set_btree_node_need_rewrite(b);
-	}
-	rcu_read_unlock();
+			if (!ca2 || ca2->mi.state != BCH_MEMBER_STATE_rw)
+				set_btree_node_need_rewrite(b);
+		}
 
 	if (!ptr_written)
 		set_btree_node_need_rewrite(b);
···
 
 	ra = kzalloc(sizeof(*ra), GFP_NOFS);
 	if (!ra)
-		return -BCH_ERR_ENOMEM_btree_node_read_all_replicas;
+		return bch_err_throw(c, ENOMEM_btree_node_read_all_replicas);
 
 	closure_init(&ra->cl, NULL);
 	ra->c	= c;
···
 		bch2_btree_node_hash_remove(&c->btree_cache, b);
 		mutex_unlock(&c->btree_cache.lock);
 
-		ret = -BCH_ERR_btree_node_read_error;
+		ret = bch_err_throw(c, btree_node_read_error);
 		goto err;
 	}
 
···
 	struct bch_fs *c = trans->c;
 
 	if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_btree_node_scrub))
-		return -BCH_ERR_erofs_no_writes;
+		return bch_err_throw(c, erofs_no_writes);
 
 	struct extent_ptr_decoded pick;
 	int ret = bch2_bkey_pick_read_device(c, k, NULL, &pick, dev);
···
 	struct bch_dev *ca = bch2_dev_get_ioref(c, pick.ptr.dev, READ,
 						BCH_DEV_READ_REF_btree_node_scrub);
 	if (!ca) {
-		ret = -BCH_ERR_device_offline;
+		ret = bch_err_throw(c, device_offline);
 		goto err;
 	}
 
···
 		bch2_dev_list_has_dev(wbio->wbio.failed, ptr->dev));
 
 	if (!bch2_bkey_nr_ptrs(bkey_i_to_s_c(&wbio->key))) {
-		ret = -BCH_ERR_btree_node_write_all_failed;
+		ret = bch_err_throw(c, btree_node_write_all_failed);
 		goto err;
 	}
 

fs/bcachefs/btree_iter.c

···
 
 static noinline int btree_node_iter_and_journal_peek(struct btree_trans *trans,
 						     struct btree_path *path,
-						     unsigned flags,
-						     struct bkey_buf *out)
+						     unsigned flags)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_path_level *l = path_l(path);
···
 		goto err;
 	}
 
-	bch2_bkey_buf_reassemble(out, c, k);
+	bkey_reassemble(&trans->btree_path_down, k);
 
 	if ((flags & BTREE_ITER_prefetch) &&
 	    c->opts.btree_node_prefetch)
···
 err:
 	bch2_btree_and_journal_iter_exit(&jiter);
 	return ret;
+}
+
+static noinline_for_stack int btree_node_missing_err(struct btree_trans *trans,
+						     struct btree_path *path)
+{
+	struct bch_fs *c = trans->c;
+	struct printbuf buf = PRINTBUF;
+
+	prt_str(&buf, "node not found at pos ");
+	bch2_bpos_to_text(&buf, path->pos);
+	prt_str(&buf, " within parent node ");
+	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&path_l(path)->b->key));
+
+	bch2_fs_fatal_error(c, "%s", buf.buf);
+	printbuf_exit(&buf);
+	return bch_err_throw(c, btree_need_topology_repair);
 }
 
 static __always_inline int btree_path_down(struct btree_trans *trans,
···
 	struct btree *b;
 	unsigned level = path->level - 1;
 	enum six_lock_type lock_type = __btree_lock_want(path, level);
-	struct bkey_buf tmp;
 	int ret;
 
 	EBUG_ON(!btree_node_locked(path, path->level));
 
-	bch2_bkey_buf_init(&tmp);
-
 	if (unlikely(trans->journal_replay_not_finished)) {
-		ret = btree_node_iter_and_journal_peek(trans, path, flags, &tmp);
+		ret = btree_node_iter_and_journal_peek(trans, path, flags);
 		if (ret)
-			goto err;
+			return ret;
 	} else {
 		struct bkey_packed *k = bch2_btree_node_iter_peek(&l->iter, l->b);
-		if (!k) {
-			struct printbuf buf = PRINTBUF;
+		if (unlikely(!k))
+			return btree_node_missing_err(trans, path);
 
-			prt_str(&buf, "node not found at pos ");
-			bch2_bpos_to_text(&buf, path->pos);
-			prt_str(&buf, " within parent node ");
-			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&l->b->key));
+		bch2_bkey_unpack(l->b, &trans->btree_path_down, k);
 
-			bch2_fs_fatal_error(c, "%s", buf.buf);
-			printbuf_exit(&buf);
-			ret = -BCH_ERR_btree_need_topology_repair;
-			goto err;
-		}
-
-		bch2_bkey_buf_unpack(&tmp, c, l->b, k);
-
-		if ((flags & BTREE_ITER_prefetch) &&
+		if (unlikely((flags & BTREE_ITER_prefetch)) &&
 		    c->opts.btree_node_prefetch) {
 			ret = btree_path_prefetch(trans, path);
 			if (ret)
-				goto err;
+				return ret;
 		}
 	}
 
-	b = bch2_btree_node_get(trans, path, tmp.k, level, lock_type, trace_ip);
+	b = bch2_btree_node_get(trans, path, &trans->btree_path_down,
+				level, lock_type, trace_ip);
 	ret = PTR_ERR_OR_ZERO(b);
 	if (unlikely(ret))
-		goto err;
+		return ret;
 
-	if (likely(!trans->journal_replay_not_finished &&
-		   tmp.k->k.type == KEY_TYPE_btree_ptr_v2) &&
-	    unlikely(b != btree_node_mem_ptr(tmp.k)))
+	if (unlikely(b != btree_node_mem_ptr(&trans->btree_path_down)) &&
+	    likely(!trans->journal_replay_not_finished &&
+		   trans->btree_path_down.k.type == KEY_TYPE_btree_ptr_v2))
 		btree_node_mem_ptr_set(trans, path, level + 1, b);
 
 	if (btree_node_read_locked(path, level + 1))
···
 	bch2_btree_path_level_init(trans, path, b);
 
 	bch2_btree_path_verify_locks(trans, path);
-err:
-	bch2_bkey_buf_exit(&tmp, c);
-	return ret;
+	return 0;
 }
 
 static int bch2_btree_path_traverse_all(struct btree_trans *trans)
···
 	int ret = 0;
 
 	if (trans->in_traverse_all)
-		return -BCH_ERR_transaction_restart_in_traverse_all;
+		return bch_err_throw(c, transaction_restart_in_traverse_all);
 
 	trans->in_traverse_all = true;
 retry_all:
···
 				      struct btree_bkey_cached_common *b)
 {
 	struct six_lock_count c = six_lock_counts(&b->lock);
-	struct task_struct *owner;
 	pid_t pid;
 
-	rcu_read_lock();
-	owner = READ_ONCE(b->lock.owner);
-	pid = owner ? owner->pid : 0;
-	rcu_read_unlock();
+	scoped_guard(rcu) {
+		struct task_struct *owner = READ_ONCE(b->lock.owner);
+		pid = owner ? owner->pid : 0;
+	}
 
 	prt_printf(out, "\t%px %c ", b, b->cached ? 'c' : 'b');
 	bch2_btree_id_to_text(out, b->btree_id);
···
 	prt_printf(out, "%i %s\n", task ? task->pid : 0, trans->fn);
 
 	/* trans->paths is rcu protected vs. freeing */
-	rcu_read_lock();
+	guard(rcu)();
 	out->atomic++;
 
 	struct btree_path *paths = rcu_dereference(trans->paths);
···
 	}
 out:
 	--out->atomic;
-	rcu_read_unlock();
 }
 
 void bch2_fs_btree_iter_exit(struct bch_fs *c)

fs/bcachefs/btree_iter.h

···
 	_p;								\
 })
 
-#define bch2_trans_run(_c, _do)						\
-({									\
-	struct btree_trans *trans = bch2_trans_get(_c);			\
-	int _ret = (_do);						\
-	bch2_trans_put(trans);						\
-	_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 *);
 
···
 		trans_fn_idx = bch2_trans_get_fn_idx(__func__);		\
 	__bch2_trans_get(_c, trans_fn_idx);				\
 })
+
+/*
+ * We don't use DEFINE_CLASS() because using a function for the constructor
+ * breaks bch2_trans_get()'s use of __func__
+ */
+typedef struct btree_trans * class_btree_trans_t;
+static inline void class_btree_trans_destructor(struct btree_trans **p)
+{
+	struct btree_trans *trans = *p;
+	bch2_trans_put(trans);
+}
+
+#define class_btree_trans_constructor(_c)	bch2_trans_get(_c)
+
+#define bch2_trans_run(_c, _do)						\
+({									\
+	CLASS(btree_trans, trans)(_c);					\
+	(_do);								\
+})
+
+#define bch2_trans_do(_c, _do)	bch2_trans_run(_c, lockrestart_do(trans, _do))
 
 void bch2_btree_trans_to_text(struct printbuf *, struct btree_trans *);
 

fs/bcachefs/btree_journal_iter.c

···
 		if (!new_keys.data) {
 			bch_err(c, "%s: error allocating new key array (size %zu)",
 				__func__, new_keys.size);
-			return -BCH_ERR_ENOMEM_journal_key_insert;
+			return bch_err_throw(c, ENOMEM_journal_key_insert);
 		}
 
 		/* Since @keys was full, there was no gap: */
···
 
 	n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 	if (!n)
-		return -BCH_ERR_ENOMEM_journal_key_insert;
+		return bch_err_throw(c, ENOMEM_journal_key_insert);
 
 	bkey_copy(n, k);
 	ret = bch2_journal_key_insert_take(c, id, level, n);
···
 
 static struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
 {
-	struct bkey_s_c ret = bkey_s_c_null;
-
 	journal_iter_verify(iter);
 
-	rcu_read_lock();
+	guard(rcu)();
 	while (iter->idx < iter->keys->size) {
 		struct journal_key *k = iter->keys->data + iter->idx;
 
···
 			break;
 		BUG_ON(cmp);
 
-		if (!k->overwritten) {
-			ret = bkey_i_to_s_c(k->k);
-			break;
-		}
+		if (!k->overwritten)
+			return bkey_i_to_s_c(k->k);
 
 		if (k->overwritten_range)
 			iter->idx = idx_to_pos(iter->keys, rcu_dereference(k->overwritten_range)->end);
 		else
 			bch2_journal_iter_advance(iter);
 	}
-	rcu_read_unlock();
 
-	return ret;
+	return bkey_s_c_null;
 }
 
 static void bch2_journal_iter_exit(struct journal_iter *iter)
···
 				if (keys->nr * 8 > keys->size * 7) {
 					bch_err(c, "Too many journal keys for slowpath; have %zu compacted, buf size %zu, processed %zu keys at seq %llu",
 						keys->nr, keys->size, nr_read, le64_to_cpu(i->j.seq));
-					return -BCH_ERR_ENOMEM_journal_keys_sort;
+					return bch_err_throw(c, ENOMEM_journal_keys_sort);
 				}
 
 				BUG_ON(darray_push(keys, n));

fs/bcachefs/btree_key_cache.c

···
 static struct bkey_cached *
 bkey_cached_reuse(struct btree_key_cache *c)
 {
-	struct bucket_table *tbl;
+
+	guard(rcu)();
+	struct bucket_table *tbl = rht_dereference_rcu(c->table.tbl, &c->table);
 	struct rhash_head *pos;
 	struct bkey_cached *ck;
-	unsigned i;
 
-	rcu_read_lock();
-	tbl = rht_dereference_rcu(c->table.tbl, &c->table);
-	for (i = 0; i < tbl->size; i++)
+	for (unsigned i = 0; i < tbl->size; i++)
 		rht_for_each_entry_rcu(ck, pos, tbl, i, hash) {
 			if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags) &&
 			    bkey_cached_lock_for_evict(ck)) {
 				if (bkey_cached_evict(c, ck))
-					goto out;
+					return ck;
 				six_unlock_write(&ck->c.lock);
 				six_unlock_intent(&ck->c.lock);
 			}
 		}
-	ck = NULL;
-out:
-	rcu_read_unlock();
-	return ck;
+	return NULL;
 }
 
 static int btree_key_cache_create(struct btree_trans *trans,
···
 		if (unlikely(!ck)) {
 			bch_err(c, "error allocating memory for key cache item, btree %s",
 				bch2_btree_id_str(ck_path->btree_id));
-			return -BCH_ERR_ENOMEM_btree_key_cache_create;
+			return bch_err_throw(c, ENOMEM_btree_key_cache_create);
 		}
 	}
 
···
 		if (unlikely(!new_k)) {
 			bch_err(trans->c, "error allocating memory for key cache key, btree %s u64s %u",
 				bch2_btree_id_str(ck->key.btree_id), key_u64s);
-			ret = -BCH_ERR_ENOMEM_btree_key_cache_fill;
+			ret = bch_err_throw(c, ENOMEM_btree_key_cache_fill);
 		} else if (ret) {
 			kfree(new_k);
 			goto err;
···
 
 	bc->nr_pending = alloc_percpu(size_t);
 	if (!bc->nr_pending)
-		return -BCH_ERR_ENOMEM_fs_btree_cache_init;
+		return bch_err_throw(c, ENOMEM_fs_btree_cache_init);
 
 	if (rcu_pending_init(&bc->pending[0], &c->btree_trans_barrier, __bkey_cached_free) ||
 	    rcu_pending_init(&bc->pending[1], &c->btree_trans_barrier, __bkey_cached_free))
-		return -BCH_ERR_ENOMEM_fs_btree_cache_init;
+		return bch_err_throw(c, ENOMEM_fs_btree_cache_init);
 
 	if (rhashtable_init(&bc->table, &bch2_btree_key_cache_params))
-		return -BCH_ERR_ENOMEM_fs_btree_cache_init;
+		return bch_err_throw(c, ENOMEM_fs_btree_cache_init);
 
 	bc->table_init_done = true;
 
 	shrink = shrinker_alloc(0, "%s-btree_key_cache", c->name);
 	if (!shrink)
-		return -BCH_ERR_ENOMEM_fs_btree_cache_init;
+		return bch_err_throw(c, ENOMEM_fs_btree_cache_init);
 	bc->shrink = shrink;
 	shrink->count_objects	= bch2_btree_key_cache_count;
 	shrink->scan_objects	= bch2_btree_key_cache_scan;

fs/bcachefs/btree_locking.c

···
 	return 3;
 }
 
+static noinline __noreturn void break_cycle_fail(struct lock_graph *g)
+{
+	struct printbuf buf = PRINTBUF;
+	buf.atomic++;
+
+	prt_printf(&buf, bch2_fmt(g->g->trans->c, "cycle of nofail locks"));
+
+	for (struct trans_waiting_for_lock *i = g->g; i < g->g + g->nr; i++) {
+		struct btree_trans *trans = i->trans;
+
+		bch2_btree_trans_to_text(&buf, trans);
+
+		prt_printf(&buf, "backtrace:\n");
+		printbuf_indent_add(&buf, 2);
+		bch2_prt_task_backtrace(&buf, trans->locking_wait.task, 2, GFP_NOWAIT);
+		printbuf_indent_sub(&buf, 2);
+		prt_newline(&buf);
+	}
+
+	bch2_print_str_nonblocking(g->g->trans->c, KERN_ERR, buf.buf);
+	printbuf_exit(&buf);
+	BUG();
+}
+
 static noinline int break_cycle(struct lock_graph *g, struct printbuf *cycle,
 				struct trans_waiting_for_lock *from)
 {
···
 		}
 	}
 
-	if (unlikely(!best)) {
-		struct printbuf buf = PRINTBUF;
-		buf.atomic++;
-
-		prt_printf(&buf, bch2_fmt(g->g->trans->c, "cycle of nofail locks"));
-
-		for (i = g->g; i < g->g + g->nr; i++) {
-			struct btree_trans *trans = i->trans;
-
-			bch2_btree_trans_to_text(&buf, trans);
-
-			prt_printf(&buf, "backtrace:\n");
-			printbuf_indent_add(&buf, 2);
-			bch2_prt_task_backtrace(&buf, trans->locking_wait.task, 2, GFP_NOWAIT);
-			printbuf_indent_sub(&buf, 2);
-			prt_newline(&buf);
-		}
-
-		bch2_print_str_nonblocking(g->g->trans->c, KERN_ERR, buf.buf);
-		printbuf_exit(&buf);
-		BUG();
-	}
+	if (unlikely(!best))
+		break_cycle_fail(g);
 
 	ret = abort_lock(g, abort);
 out:
···
 			      struct printbuf *cycle)
 {
 	struct btree_trans *orig_trans = g->g->trans;
-	struct trans_waiting_for_lock *i;
 
-	for (i = g->g; i < g->g + g->nr; i++)
+	for (struct trans_waiting_for_lock *i = g->g; i < g->g + g->nr; i++)
 		if (i->trans == trans) {
 			closure_put(&trans->ref);
 			return break_cycle(g, cycle, i);
 		}
 
-	if (g->nr == ARRAY_SIZE(g->g)) {
+	if (unlikely(g->nr == ARRAY_SIZE(g->g))) {
 		closure_put(&trans->ref);
 
 		if (orig_trans->lock_may_not_fail)
···
 	lock_graph_down(&g, trans);
 
 	/* trans->paths is rcu protected vs. freeing */
-	rcu_read_lock();
+	guard(rcu)();
 	if (cycle)
 		cycle->atomic++;
 next:
···
 out:
 	if (cycle)
 		--cycle->atomic;
-	rcu_read_unlock();
 	return ret;
 }
 

fs/bcachefs/btree_node_scan.c

···
 				min_heap_sift_down(nodes_heap, 0, &found_btree_node_heap_cbs, NULL);
 			}
 		}
+
+		cond_resched();
 	}
 
 	return 0;

fs/bcachefs/btree_trans_commit.c

···
 				       struct btree *b, unsigned u64s)
 {
 	if (!bch2_btree_node_insert_fits(b, u64s))
-		return -BCH_ERR_btree_insert_btree_node_full;
+		return bch_err_throw(trans->c, btree_insert_btree_node_full);
 
 	return 0;
 }
···
 
 	new_k = kmalloc(new_u64s * sizeof(u64), GFP_KERNEL);
 	if (!new_k) {
-		bch_err(trans->c, "error allocating memory for key cache key, btree %s u64s %u",
+		struct bch_fs *c = trans->c;
+		bch_err(c, "error allocating memory for key cache key, btree %s u64s %u",
 			bch2_btree_id_str(path->btree_id), new_u64s);
-		return -BCH_ERR_ENOMEM_btree_key_cache_insert;
+		return bch_err_throw(c, ENOMEM_btree_key_cache_insert);
 	}
 
 	ret =   bch2_trans_relock(trans) ?:
···
 	if (watermark < BCH_WATERMARK_reclaim &&
 	    !test_bit(BKEY_CACHED_DIRTY, &ck->flags) &&
 	    bch2_btree_key_cache_must_wait(c))
-		return -BCH_ERR_btree_insert_need_journal_reclaim;
+		return bch_err_throw(c, btree_insert_need_journal_reclaim);
 
 	/*
 	 * bch2_varint_decode can read past the end of the buffer by at most 7
···
 		 */
 		if ((flags & BCH_TRANS_COMMIT_journal_reclaim) &&
 		    watermark < BCH_WATERMARK_reclaim) {
-			ret = -BCH_ERR_journal_reclaim_would_deadlock;
+			ret = bch_err_throw(c, journal_reclaim_would_deadlock);
 			goto out;
 		}
 
···
 
 	for (struct jset_entry *i = btree_trans_journal_entries_start(trans);
 	     i != btree_trans_journal_entries_top(trans);
-	     i = vstruct_next(i))
+	     i = vstruct_next(i)) {
 		if (i->type == BCH_JSET_ENTRY_btree_keys ||
 		    i->type == BCH_JSET_ENTRY_write_buffer_keys) {
-			int ret = bch2_journal_key_insert(c, i->btree_id, i->level, i->start);
-			if (ret)
-				return ret;
+			jset_entry_for_each_key(i, k) {
+				int ret = bch2_journal_key_insert(c, i->btree_id, i->level, k);
+				if (ret)
+					return ret;
+			}
 		}
+
+		if (i->type == BCH_JSET_ENTRY_btree_root) {
+			guard(mutex)(&c->btree_root_lock);
+
+			struct btree_root *r = bch2_btree_id_root(c, i->btree_id);
+
+			bkey_copy(&r->key, i->start);
+			r->level = i->level;
+			r->alive = true;
+		}
+	}
 
 	for (struct bkey_i *i = btree_trans_subbuf_base(trans, &trans->accounting);
 	     i != btree_trans_subbuf_top(trans, &trans->accounting);
···
 		if (unlikely(!test_bit(BCH_FS_may_go_rw, &c->flags)))
 			ret = do_bch2_trans_commit_to_journal_replay(trans);
 		else
-			ret = -BCH_ERR_erofs_trans_commit;
+			ret = bch_err_throw(c, erofs_trans_commit);
 		goto out_reset;
 	}
 
···
 	 * restart:
 	 */
 	if (flags & BCH_TRANS_COMMIT_no_journal_res) {
-		ret = -BCH_ERR_transaction_restart_nested;
+		ret = bch_err_throw(c, transaction_restart_nested);
 		goto out;
 	}
 

fs/bcachefs/btree_types.h

···
 	unsigned		journal_u64s;
 	unsigned		extra_disk_res; /* XXX kill */
 
+	__BKEY_PADDED(btree_path_down, BKEY_BTREE_PTR_VAL_U64s_MAX);
+
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 	struct lockdep_map	dep_map;
 #endif

fs/bcachefs/btree_update.c

···
 }
 
 int __bch2_insert_snapshot_whiteouts(struct btree_trans *trans,
-				   enum btree_id id,
-				   struct bpos old_pos,
-				   struct bpos new_pos)
+				     enum btree_id btree, struct bpos pos,
+				     snapshot_id_list *s)
 {
-	struct bch_fs *c = trans->c;
-	struct btree_iter old_iter, new_iter = {};
-	struct bkey_s_c old_k, new_k;
-	snapshot_id_list s;
-	struct bkey_i *update;
 	int ret = 0;
 
-	if (!bch2_snapshot_has_children(c, old_pos.snapshot))
-		return 0;
+	darray_for_each(*s, id) {
+		pos.snapshot = *id;
 
-	darray_init(&s);
-
-	bch2_trans_iter_init(trans, &old_iter, id, old_pos,
-			     BTREE_ITER_not_extents|
-			     BTREE_ITER_all_snapshots);
-	while ((old_k = bch2_btree_iter_prev(trans, &old_iter)).k &&
-	       !(ret = bkey_err(old_k)) &&
-	       bkey_eq(old_pos, old_k.k->p)) {
-		struct bpos whiteout_pos =
-			SPOS(new_pos.inode, new_pos.offset, old_k.k->p.snapshot);
-
-		if (!bch2_snapshot_is_ancestor(c, old_k.k->p.snapshot, old_pos.snapshot) ||
-		    snapshot_list_has_ancestor(c, &s, old_k.k->p.snapshot))
-			continue;
-
-		new_k = bch2_bkey_get_iter(trans, &new_iter, id, whiteout_pos,
-					   BTREE_ITER_not_extents|
-					   BTREE_ITER_intent);
-		ret = bkey_err(new_k);
+		struct btree_iter iter;
+		struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, btree, pos,
+						       BTREE_ITER_not_extents|
+						       BTREE_ITER_intent);
+		ret = bkey_err(k);
 		if (ret)
 			break;
 
-		if (new_k.k->type == KEY_TYPE_deleted) {
-			update = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
+		if (k.k->type == KEY_TYPE_deleted) {
+			struct bkey_i *update = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
 			ret = PTR_ERR_OR_ZERO(update);
-			if (ret)
+			if (ret) {
+				bch2_trans_iter_exit(trans, &iter);
 				break;
+			}
 
 			bkey_init(&update->k);
-			update->k.p		= whiteout_pos;
+			update->k.p		= pos;
 			update->k.type		= KEY_TYPE_whiteout;
 
-			ret = bch2_trans_update(trans, &new_iter, update,
+			ret = bch2_trans_update(trans, &iter, update,
 						BTREE_UPDATE_internal_snapshot_node);
 		}
-		bch2_trans_iter_exit(trans, &new_iter);
+		bch2_trans_iter_exit(trans, &iter);
 
-		ret = snapshot_list_add(c, &s, old_k.k->p.snapshot);
 		if (ret)
 			break;
 	}
-	bch2_trans_iter_exit(trans, &new_iter);
-	bch2_trans_iter_exit(trans, &old_iter);
-	darray_exit(&s);
 
+	darray_exit(s);
 	return ret;
 }
 
···
 	BUG_ON(k.k->type != KEY_TYPE_deleted);
 
 	if (bkey_gt(k.k->p, end)) {
-		ret = -BCH_ERR_ENOSPC_btree_slot;
+		ret = bch_err_throw(trans->c, ENOSPC_btree_slot);
 		goto err;
 	}
 

fs/bcachefs/btree_update.h

···
 
 #include "btree_iter.h"
 #include "journal.h"
+#include "snapshot.h"
 
 struct bch_fs;
 struct btree;
···
 }
 
 int __bch2_insert_snapshot_whiteouts(struct btree_trans *, enum btree_id,
-				     struct bpos, struct bpos);
+				     struct bpos, snapshot_id_list *);
 
 /*
  * For use when splitting extents in existing snapshots:
···
 						 struct bpos old_pos,
 						 struct bpos new_pos)
 {
+	BUG_ON(old_pos.snapshot != new_pos.snapshot);
+
 	if (!btree_type_has_snapshots(btree) ||
 	    bkey_eq(old_pos, new_pos))
 		return 0;
 
-	return __bch2_insert_snapshot_whiteouts(trans, btree, old_pos, new_pos);
+	snapshot_id_list s;
+	int ret = bch2_get_snapshot_overwrites(trans, btree, old_pos, &s);
+	if (ret)
+		return ret;
+
+	return s.nr
+		? __bch2_insert_snapshot_whiteouts(trans, btree, new_pos, &s)
+		: 0;
 }
 
 int bch2_trans_update_extent_overwrite(struct btree_trans *, struct btree_iter *,

fs/bcachefs/btree_update_interior.c

···
 	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));
···
 
 	if (b == btree_node_root(c, b)) {
 		if (!bpos_eq(b->data->min_key, POS_MIN)) {
-			ret = __bch2_topology_error(c, &buf);
-
+			bch2_log_msg_start(c, &buf);
+			prt_printf(&buf, "btree root with incorrect min_key: ");
 			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 out;
+			prt_newline(&buf);
+
+			bch2_count_fsck_err(c, btree_root_bad_min_key, &buf);
+			goto err;
 		}
 
 		if (!bpos_eq(b->data->max_key, SPOS_MAX)) {
-			ret = __bch2_topology_error(c, &buf);
+			bch2_log_msg_start(c, &buf);
+			prt_printf(&buf, "btree root with incorrect max_key: ");
 			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 out;
+			prt_newline(&buf);
+
+			bch2_count_fsck_err(c, btree_root_bad_max_key, &buf);
+			goto err;
 		}
 	}
 
···
 			: bpos_successor(prev.k->k.p);
 
 		if (!bpos_eq(expected_min, bp.v->min_key)) {
-			ret = __bch2_topology_error(c, &buf);
-
-			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, "end of prev node doesn't match start of next node");
 			prt_str(&buf, "\nprev ");
 			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(prev.k));
 			prt_str(&buf, "\nnext ");
 			bch2_bkey_val_to_text(&buf, c, k);
+			prt_newline(&buf);
 
-			log_fsck_err(trans, btree_node_topology_bad_min_key, "%s", buf.buf);
-			goto out;
+			bch2_count_fsck_err(c, btree_node_topology_bad_min_key, &buf);
+			goto err;
 		}
 
 		bch2_bkey_buf_reassemble(&prev, c, k);
···
 	}
 
 	if (bkey_deleted(&prev.k->k)) {
-		ret = __bch2_topology_error(c, &buf);
+		prt_printf(&buf, "empty interior node\n");
+		bch2_count_fsck_err(c, btree_node_topology_empty_interior_node, &buf);
+		goto err;
+	}
 
-		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);
-	} else if (!bpos_eq(prev.k->k.p, b->key.k.p)) {
-		ret = __bch2_topology_error(c, &buf);
-
-		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, "\nlast key ");
+	if (!bpos_eq(prev.k->k.p, b->key.k.p)) {
+		prt_str(&buf, "last child node doesn't end at end of parent node\nchild: ");
 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(prev.k));
+		prt_newline(&buf);
 
-		log_fsck_err(trans, btree_node_topology_bad_max_key, "%s", buf.buf);
+		bch2_count_fsck_err(c, btree_node_topology_bad_max_key, &buf);
+		goto err;
 	}
 out:
-fsck_err:
 	bch2_btree_and_journal_iter_exit(&iter);
 	bch2_bkey_buf_exit(&prev, c);
 	printbuf_exit(&buf);
 	return ret;
+err:
+	bch2_btree_id_level_to_text(&buf, b->c.btree_id, b->c.level);
+	prt_char(&buf, ' ');
+	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+	prt_newline(&buf);
+
+	ret = __bch2_topology_error(c, &buf);
+	bch2_print_str(c, KERN_ERR, buf.buf);
+	BUG_ON(!ret);
+	goto out;
 }
 
 /* Calculate ideal packed bkey format for new btree nodes: */
···
 
 	/*
 	 * Wait for any in flight writes to finish before we free the old nodes
-	 * on disk:
+	 * on disk. But we haven't pinned those old nodes in the btree cache,
+	 * they might have already been evicted.
+	 *
+	 * The update we're completing deleted references to those nodes from the
+	 * btree, so we know if they've been evicted they can't be pulled back in.
+	 * We just have to check if the nodes we have pointers to are still those
+	 * old nodes, and haven't been reused.
+	 *
+	 * This can't be done locklessly because the data buffer might have been
+	 * vmalloc allocated, and they're not RCU freed. We also need the
+	 * __no_kmsan_checks annotation because even with the btree node read
+	 * lock, nothing tells us that the data buffer has been initialized (if
+	 * the btree node has been reused for a different node, and the data
+	 * buffer swapped for a new data buffer).
 	 */
 	for (i = 0; i < as->nr_old_nodes; i++) {
 		b = as->old_nodes[i];
 
-		if (btree_node_seq_matches(b, as->old_nodes_seq[i]))
+		bch2_trans_begin(trans);
+		btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
+		bool seq_matches = btree_node_seq_matches(b, as->old_nodes_seq[i]);
+		six_unlock_read(&b->c.lock);
+		bch2_trans_unlock_long(trans);
+
+		if (seq_matches)
 			wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight_inner,
 				       TASK_UNINTERRUPTIBLE);
 	}
···
 		if (bch2_err_matches(ret, ENOSPC) &&
 		    (flags & BCH_TRANS_COMMIT_journal_reclaim) &&
 		    watermark < BCH_WATERMARK_reclaim) {
-			ret = -BCH_ERR_journal_reclaim_would_deadlock;
+			ret = bch_err_throw(c, journal_reclaim_would_deadlock);
 			goto err;
 		}
 
···
 	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;
+		ret = bch_err_throw(trans->c, btree_node_dying);
 		goto err;
 	}
 
···
 	c->btree_interior_update_worker =
 		alloc_workqueue("btree_update", WQ_UNBOUND|WQ_MEM_RECLAIM, 8);
 	if (!c->btree_interior_update_worker)
-		return -BCH_ERR_ENOMEM_btree_interior_update_worker_init;
+		return bch_err_throw(c, ENOMEM_btree_interior_update_worker_init);
 
 	c->btree_node_rewrite_worker =
 		alloc_ordered_workqueue("btree_node_rewrite", WQ_UNBOUND);
 	if (!c->btree_node_rewrite_worker)
-		return -BCH_ERR_ENOMEM_btree_interior_update_worker_init;
+		return bch_err_throw(c, ENOMEM_btree_interior_update_worker_init);
 
 	if (mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
 				      sizeof(struct btree_update)))
-		return -BCH_ERR_ENOMEM_btree_interior_update_pool_init;
+		return bch_err_throw(c, ENOMEM_btree_interior_update_pool_init);
 
 	return 0;
 }

fs/bcachefs/btree_write_buffer.c

···
 		bool accounting_accumulated = false;
 		do {
 			if (race_fault()) {
-				ret = -BCH_ERR_journal_reclaim_would_deadlock;
+				ret = bch_err_throw(c, journal_reclaim_would_deadlock);
 				break;
 			}
 
···
 	struct bch_fs *c = trans->c;
 
 	if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_btree_write_buffer))
-		return -BCH_ERR_erofs_no_writes;
+		return bch_err_throw(c, erofs_no_writes);
 
 	int ret = bch2_btree_write_buffer_flush_nocheck_rw(trans);
 	enumerated_ref_put(&c->writes, BCH_WRITE_REF_btree_write_buffer);
···
 			goto err;
 
 		bch2_bkey_buf_copy(last_flushed, c, tmp.k);
-		ret = -BCH_ERR_transaction_restart_write_buffer_flush;
+		ret = bch_err_throw(c, transaction_restart_write_buffer_flush);
 	}
 err:
 	bch2_bkey_buf_exit(&tmp, c);

fs/bcachefs/buckets.c

···
 			 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
 		if (!p.ptr.cached &&
 		    data_type == BCH_DATA_btree) {
+			switch (g->data_type) {
+			case BCH_DATA_sb:
+				bch_err(c, "btree and superblock in the same bucket - cannot repair");
+				ret = bch_err_throw(c, fsck_repair_unimplemented);
+				goto out;
+			case BCH_DATA_journal:
+				ret = bch2_dev_journal_bucket_delete(ca, PTR_BUCKET_NR(ca, &p.ptr));
+				bch_err_msg(c, ret, "error deleting journal bucket %zu",
+					    PTR_BUCKET_NR(ca, &p.ptr));
+				if (ret)
+					goto out;
+				break;
+			}
+
 			g->data_type		= data_type;
 			g->stripe_sectors	= 0;
 			g->dirty_sectors	= 0;
···
 	struct printbuf buf = PRINTBUF;
 	int ret = 0;
 
+	/* We don't yet do btree key updates correctly for when we're RW */
+	BUG_ON(test_bit(BCH_FS_rw, &c->flags));
+
 	bkey_for_each_ptr_decode(k.k, ptrs_c, p, entry_c) {
 		ret = bch2_check_fix_ptr(trans, k, p, entry_c, &do_update);
 		if (ret)
···
 	}
 
 	if (do_update) {
-		if (flags & BTREE_TRIGGER_is_root) {
-			bch_err(c, "cannot update btree roots yet");
-			ret = -EINVAL;
-			goto err;
-		}
-
 		struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k);
 		ret = PTR_ERR_OR_ZERO(new);
 		if (ret)
 			goto err;
 
-		rcu_read_lock();
-		bch2_bkey_drop_ptrs(bkey_i_to_s(new), ptr, !bch2_dev_exists(c, ptr->dev));
-		rcu_read_unlock();
+		scoped_guard(rcu)
+			bch2_bkey_drop_ptrs(bkey_i_to_s(new), ptr, !bch2_dev_exists(c, ptr->dev));
 
 		if (level) {
 			/*
···
 			 * sort it out:
 			 */
 			struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(new));
-			rcu_read_lock();
-			bkey_for_each_ptr(ptrs, ptr) {
-				struct bch_dev *ca = bch2_dev_rcu(c, ptr->dev);
-				struct bucket *g = PTR_GC_BUCKET(ca, ptr);
-
-				ptr->gen = g->gen;
-			}
-			rcu_read_unlock();
+			scoped_guard(rcu)
+				bkey_for_each_ptr(ptrs, ptr) {
+					struct bch_dev *ca = bch2_dev_rcu(c, ptr->dev);
+					ptr->gen = PTR_GC_BUCKET(ca, ptr)->gen;
+				}
 		} else {
 			struct bkey_ptrs ptrs;
 			union bch_extent_entry *entry;
···
 			bch_info(c, "new key %s", buf.buf);
 		}
 
-		struct btree_iter iter;
-		bch2_trans_node_iter_init(trans, &iter, btree, new->k.p, 0, level,
-					  BTREE_ITER_intent|BTREE_ITER_all_snapshots);
-		ret =   bch2_btree_iter_traverse(trans, &iter) ?:
-			bch2_trans_update(trans, &iter, new,
-					  BTREE_UPDATE_internal_snapshot_node|
-					  BTREE_TRIGGER_norun);
-		bch2_trans_iter_exit(trans, &iter);
-		if (ret)
-			goto err;
+		if (!(flags & BTREE_TRIGGER_is_root)) {
+			struct btree_iter iter;
+			bch2_trans_node_iter_init(trans, &iter, btree, new->k.p, 0, level,
+						  BTREE_ITER_intent|BTREE_ITER_all_snapshots);
+			ret =   bch2_btree_iter_traverse(trans, &iter) ?:
+				bch2_trans_update(trans, &iter, new,
+						  BTREE_UPDATE_internal_snapshot_node|
+						  BTREE_TRIGGER_norun);
+			bch2_trans_iter_exit(trans, &iter);
+			if (ret)
+				goto err;
 
-		if (level)
-			bch2_btree_node_update_key_early(trans, btree, level - 1, k, new);
+			if (level)
+				bch2_btree_node_update_key_early(trans, btree, level - 1, k, new);
+		} else {
+			struct jset_entry *e = bch2_trans_jset_entry_alloc(trans,
+					       jset_u64s(new->k.u64s));
+			ret = PTR_ERR_OR_ZERO(e);
+			if (ret)
+				goto err;
+
+			journal_entry_set(e,
+					  BCH_JSET_ENTRY_btree_root,
+					  btree, level - 1,
+					  new, new->k.u64s);
+
+			/*
+			 * no locking, we're single threaded and not rw yet, see
+			 * the big assertino above that we repeat here:
+			 */
+			BUG_ON(test_bit(BCH_FS_rw, &c->flags));
+
+			struct btree *b = bch2_btree_id_root(c, btree)->b;
+			bkey_copy(&b->key, new);
+		}
 	}
 err:
 	printbuf_exit(&buf);
···
 	if (insert) {
 		bch2_trans_updates_to_text(buf, trans);
 		__bch2_inconsistent_error(c, buf);
-		ret = -BCH_ERR_bucket_ref_update;
+		/*
+		 * If we're in recovery, run_explicit_recovery_pass might give
+		 * us an error code for rewinding recovery
+		 */
+		if (!ret)
+			ret = bch_err_throw(c, bucket_ref_update);
+	} else {
+		/* Always ignore overwrite errors, so that deletion works */
+		ret = 0;
 	}
 
 	if (print || insert)
···
 	struct bch_dev *ca = bch2_dev_tryget(c, p.ptr.dev);
 	if (unlikely(!ca)) {
 		if (insert && p.ptr.dev != BCH_SB_MEMBER_INVALID)
-			ret = -BCH_ERR_trigger_pointer;
+			ret = bch_err_throw(c, trigger_pointer);
 		goto err;
 	}
 
···
 	if (!bucket_valid(ca, bucket.offset)) {
 		if (insert) {
 			bch2_dev_bucket_missing(ca, bucket.offset);
-			ret = -BCH_ERR_trigger_pointer;
+			ret = bch_err_throw(c, trigger_pointer);
 		}
 		goto err;
 	}
···
 		if (bch2_fs_inconsistent_on(!g, c, "reference to invalid bucket on device %u\n  %s",
 					    p.ptr.dev,
 					    (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
-			ret = -BCH_ERR_trigger_pointer;
+			ret = bch_err_throw(c, trigger_pointer);
 			goto err;
 		}
 
···
 				s64 sectors,
 				enum btree_iter_update_trigger_flags flags)
 {
+	struct bch_fs *c = trans->c;
+
 	if (flags & BTREE_TRIGGER_transactional) {
 		struct btree_iter iter;
 		struct bkey_i_stripe *s = bch2_bkey_get_mut_typed(trans, &iter,
···
 			bch2_trans_inconsistent(trans,
 				"stripe pointer doesn't match stripe %llu",
 				(u64) p.ec.idx);
-			ret = -BCH_ERR_trigger_stripe_pointer;
+			ret = bch_err_throw(c, trigger_stripe_pointer);
 			goto err;
 		}
 
···
 	}
 
 	if (flags & BTREE_TRIGGER_gc) {
-		struct bch_fs *c = trans->c;
-
 		struct gc_stripe *m = genradix_ptr_alloc(&c->gc_stripes, p.ec.idx, GFP_KERNEL);
 		if (!m) {
 			bch_err(c, "error allocating memory for gc_stripes, idx %llu",
 				(u64) p.ec.idx);
-			return -BCH_ERR_ENOMEM_mark_stripe_ptr;
+			return bch_err_throw(c, ENOMEM_mark_stripe_ptr);
 		}
 
 		gc_stripe_lock(m);
···
 			__bch2_inconsistent_error(c, &buf);
 			bch2_print_str(c, KERN_ERR, buf.buf);
 			printbuf_exit(&buf);
-			return -BCH_ERR_trigger_stripe_pointer;
+			return bch_err_throw(c, trigger_stripe_pointer);
 		}
 
 		m->block_sectors[p.ec.block] += sectors;
···
 static int __trigger_extent(struct btree_trans *trans,
 			    enum btree_id btree_id, unsigned level,
 			    struct bkey_s_c k,
-			    enum btree_iter_update_trigger_flags flags,
-			    s64 *replicas_sectors)
+			    enum btree_iter_update_trigger_flags flags)
 {
 	bool gc = flags & BTREE_TRIGGER_gc;
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
···
 		? BCH_DATA_btree
 		: BCH_DATA_user;
 	int ret = 0;
+
+	s64 replicas_sectors = 0;
 
 	struct disk_accounting_pos acc_replicas_key;
 	memset(&acc_replicas_key, 0, sizeof(acc_replicas_key));
···
 			if (ret)
 				return ret;
 		} else if (!p.has_ec) {
-			*replicas_sectors       += disk_sectors;
+			replicas_sectors       += disk_sectors;
 			replicas_entry_add_dev(&acc_replicas_key.replicas, p.ptr.dev);
 		} else {
 			ret = bch2_trigger_stripe_ptr(trans, k, p, data_type, disk_sectors, flags);
···
 	}
 
 	if (acc_replicas_key.replicas.nr_devs) {
-		ret = bch2_disk_accounting_mod(trans, &acc_replicas_key, replicas_sectors, 1, gc);
+		ret = bch2_disk_accounting_mod(trans, &acc_replicas_key, &replicas_sectors, 1, gc);
 		if (ret)
 			return ret;
 	}
 
 	if (acc_replicas_key.replicas.nr_devs && !level && k.k->p.snapshot) {
-		ret = bch2_disk_accounting_mod2_nr(trans, gc, replicas_sectors, 1, snapshot, k.k->p.snapshot);
+		ret = bch2_disk_accounting_mod2_nr(trans, gc, &replicas_sectors, 1, snapshot, k.k->p.snapshot);
 		if (ret)
 			return ret;
 	}
···
 	}
 
 	if (level) {
-		ret = bch2_disk_accounting_mod2_nr(trans, gc, replicas_sectors, 1, btree, btree_id);
+		ret = bch2_disk_accounting_mod2_nr(trans, gc, &replicas_sectors, 1, btree, btree_id);
 		if (ret)
 			return ret;
 	} else {
···
 		s64 v[3] = {
 			insert ? 1 : -1,
 			insert ? k.k->size : -((s64) k.k->size),
-			*replicas_sectors,
+			replicas_sectors,
 		};
 		ret = bch2_disk_accounting_mod2(trans, gc, v, inum, k.k->p.inode);
 		if (ret)
···
 		return 0;
 
 	if (flags & (BTREE_TRIGGER_transactional|BTREE_TRIGGER_gc)) {
-		s64 old_replicas_sectors = 0, new_replicas_sectors = 0;
-
 		if (old.k->type) {
 			int ret = __trigger_extent(trans, btree, level, old,
-						   flags & ~BTREE_TRIGGER_insert,
-						   &old_replicas_sectors);
+						   flags & ~BTREE_TRIGGER_insert);
 			if (ret)
 				return ret;
 		}
 
 		if (new.k->type) {
 			int ret = __trigger_extent(trans, btree, level, new.s_c,
-						   flags & ~BTREE_TRIGGER_overwrite,
-						   &new_replicas_sectors);
+						   flags & ~BTREE_TRIGGER_overwrite);
 			if (ret)
 				return ret;
 		}
···
 			   bch2_data_type_str(type),
 			   bch2_data_type_str(type));
 
-		bool print = bch2_count_fsck_err(c, bucket_metadata_type_mismatch, &buf);
+		bch2_count_fsck_err(c, bucket_metadata_type_mismatch, &buf);
 
-		bch2_run_explicit_recovery_pass(c, &buf,
+		ret = bch2_run_explicit_recovery_pass(c, &buf,
 					BCH_RECOVERY_PASS_check_allocations, 0);
 
-		if (print)
-			bch2_print_str(c, KERN_ERR, buf.buf);
+		/* Always print, this is always fatal */
+		bch2_print_str(c, KERN_ERR, buf.buf);
 		printbuf_exit(&buf);
-		ret = -BCH_ERR_metadata_bucket_inconsistency;
+		if (!ret)
+			ret = bch_err_throw(c, metadata_bucket_inconsistency);
 		goto err;
 	}
 
···
 err_unlock:
 	bucket_unlock(g);
 err:
-	return -BCH_ERR_metadata_bucket_inconsistency;
+	return bch_err_throw(c, metadata_bucket_inconsistency);
 }
 
 int bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
···
 		ret = 0;
 	} else {
 		atomic64_set(&c->sectors_available, sectors_available);
-		ret = -BCH_ERR_ENOSPC_disk_reservation;
+		ret = bch_err_throw(c, ENOSPC_disk_reservation);
 	}
 
 	mutex_unlock(&c->sectors_available_lock);
···
 					    GFP_KERNEL|__GFP_ZERO);
 		if (!ca->buckets_nouse) {
 			bch2_dev_put(ca);
-			return -BCH_ERR_ENOMEM_buckets_nouse;
+			return bch_err_throw(c, ENOMEM_buckets_nouse);
 		}
 	}
 
···
 		lockdep_assert_held(&c->state_lock);
 
 	if (resize && ca->buckets_nouse)
-		return -BCH_ERR_no_resize_with_buckets_nouse;
+		return bch_err_throw(c, no_resize_with_buckets_nouse);
 
 	bucket_gens = bch2_kvmalloc(struct_size(bucket_gens, b, nbuckets),
 				    GFP_KERNEL|__GFP_ZERO);
 	if (!bucket_gens) {
-		ret = -BCH_ERR_ENOMEM_bucket_gens;
+		ret = bch_err_throw(c, ENOMEM_bucket_gens);
 		goto err;
 	}
 
···
 		       sizeof(bucket_gens->b[0]) * copy);
 	}
 
-	ret =   bch2_bucket_bitmap_resize(&ca->bucket_backpointer_mismatch,
+	ret =   bch2_bucket_bitmap_resize(ca, &ca->bucket_backpointer_mismatch,
 					  ca->mi.nbuckets, nbuckets) ?:
-		bch2_bucket_bitmap_resize(&ca->bucket_backpointer_empty,
+		bch2_bucket_bitmap_resize(ca, &ca->bucket_backpointer_empty,
 					  ca->mi.nbuckets, nbuckets);
 
 	rcu_assign_pointer(ca->bucket_gens, bucket_gens);
···
 {
 	ca->usage = alloc_percpu(struct bch_dev_usage_full);
 	if (!ca->usage)
-		return -BCH_ERR_ENOMEM_usage_init;
+		return bch_err_throw(c, ENOMEM_usage_init);
 
 	return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets);
 }

fs/bcachefs/buckets.h

···
 
 static inline int bucket_gen_get(struct bch_dev *ca, size_t b)
 {
-	rcu_read_lock();
-	int ret = bucket_gen_get_rcu(ca, b);
-	rcu_read_unlock();
-	return ret;
+	guard(rcu)();
+	return bucket_gen_get_rcu(ca, b);
 }
 
 static inline size_t PTR_BUCKET_NR(const struct bch_dev *ca,
···
  */
 static inline int dev_ptr_stale(struct bch_dev *ca, const struct bch_extent_ptr *ptr)
 {
-	rcu_read_lock();
-	int ret = dev_ptr_stale_rcu(ca, ptr);
-	rcu_read_unlock();
-	return ret;
+	guard(rcu)();
+	return dev_ptr_stale_rcu(ca, ptr);
 }
 
 /* Device usage: */

fs/bcachefs/buckets_waiting_for_journal.c

···
 realloc:
 	n = kvmalloc(sizeof(*n) + (sizeof(n->d[0]) << new_bits), GFP_KERNEL);
 	if (!n) {
-		ret = -BCH_ERR_ENOMEM_buckets_waiting_for_journal_set;
+		struct bch_fs *c = container_of(b, struct bch_fs, buckets_waiting_for_journal);
+		ret = bch_err_throw(c, ENOMEM_buckets_waiting_for_journal_set);
 		goto out;
 	}
 

fs/bcachefs/chardev.c

···
 	if (!dev)
 		return -EINVAL;
 
-	rcu_read_lock();
+	guard(rcu)();
 	for_each_online_member_rcu(c, ca)
-		if (ca->dev == dev) {
-			rcu_read_unlock();
+		if (ca->dev == dev)
 			return ca->dev_idx;
-		}
-	rcu_read_unlock();
 
-	return -BCH_ERR_ENOENT_dev_idx_not_found;
+	return bch_err_throw(c, ENOENT_dev_idx_not_found);
 }
 
 static long bch2_ioctl_disk_resize(struct bch_fs *c,

fs/bcachefs/checksum.c

···
 
 	if (bch2_fs_inconsistent_on(!c->chacha20_key_set,
 				    c, "attempting to encrypt without encryption key"))
-		return -BCH_ERR_no_encryption_key;
+		return bch_err_throw(c, no_encryption_key);
 
 	bch2_chacha20(&c->chacha20_key, nonce, data, len);
 	return 0;
···
 
 	if (bch2_fs_inconsistent_on(!c->chacha20_key_set,
 				    c, "attempting to encrypt without encryption key"))
-		return -BCH_ERR_no_encryption_key;
+		return bch_err_throw(c, no_encryption_key);
 
 	bch2_chacha20_init(&chacha_state, &c->chacha20_key, nonce);
 
···
 		prt_str(&buf, ")");
 		WARN_RATELIMIT(1, "%s", buf.buf);
 		printbuf_exit(&buf);
-		return -BCH_ERR_recompute_checksum;
+		return bch_err_throw(c, recompute_checksum);
 	}
 
 	for (i = splits; i < splits + ARRAY_SIZE(splits); i++) {
···
 	crypt = bch2_sb_field_resize(&c->disk_sb, crypt,
 				     sizeof(*crypt) / sizeof(u64));
 	if (!crypt) {
-		ret = -BCH_ERR_ENOSPC_sb_crypt;
+		ret = bch_err_throw(c, ENOSPC_sb_crypt);
 		goto err;
 	}
 

fs/bcachefs/clock.c

···
 
 struct io_clock_wait {
 	struct io_timer		io_timer;
-	struct timer_list	cpu_timer;
 	struct task_struct	*task;
 	int			expired;
 };
···
 {
 	struct io_clock_wait *wait = container_of(timer,
 				struct io_clock_wait, io_timer);
-
-	wait->expired = 1;
-	wake_up_process(wait->task);
-}
-
-static void io_clock_cpu_timeout(struct timer_list *timer)
-{
-	struct io_clock_wait *wait = container_of(timer,
-				struct io_clock_wait, cpu_timer);
 
 	wait->expired = 1;
 	wake_up_process(wait->task);
···
 	bch2_io_timer_del(clock, &wait.io_timer);
 }
 
-void bch2_kthread_io_clock_wait(struct io_clock *clock,
-				u64 io_until, unsigned long cpu_timeout)
+unsigned long bch2_kthread_io_clock_wait_once(struct io_clock *clock,
+				     u64 io_until, unsigned long cpu_timeout)
 {
 	bool kthread = (current->flags & PF_KTHREAD) != 0;
 	struct io_clock_wait wait = {
···
 
 	bch2_io_timer_add(clock, &wait.io_timer);
 
-	timer_setup_on_stack(&wait.cpu_timer, io_clock_cpu_timeout, 0);
-
-	if (cpu_timeout != MAX_SCHEDULE_TIMEOUT)
-		mod_timer(&wait.cpu_timer, cpu_timeout + jiffies);
-
-	do {
-		set_current_state(TASK_INTERRUPTIBLE);
-		if (kthread && kthread_should_stop())
-			break;
-
-		if (wait.expired)
-			break;
-
-		schedule();
+	set_current_state(TASK_INTERRUPTIBLE);
+	if (!(kthread && kthread_should_stop())) {
+		cpu_timeout = schedule_timeout(cpu_timeout);
 		try_to_freeze();
-	} while (0);
+	}
 
 	__set_current_state(TASK_RUNNING);
-	timer_delete_sync(&wait.cpu_timer);
-	timer_destroy_on_stack(&wait.cpu_timer);
 	bch2_io_timer_del(clock, &wait.io_timer);
+	return cpu_timeout;
+}
+
+void bch2_kthread_io_clock_wait(struct io_clock *clock,
+				u64 io_until, unsigned long cpu_timeout)
+{
+	bool kthread = (current->flags & PF_KTHREAD) != 0;
+
+	while (!(kthread && kthread_should_stop()) &&
+	       cpu_timeout &&
+	       atomic64_read(&clock->now) < io_until)
+		cpu_timeout = bch2_kthread_io_clock_wait_once(clock, io_until, cpu_timeout);
 }
 
 static struct io_timer *get_expired_timer(struct io_clock *clock, u64 now)

fs/bcachefs/clock.h

···
 
 void bch2_io_timer_add(struct io_clock *, struct io_timer *);
 void bch2_io_timer_del(struct io_clock *, struct io_timer *);
+unsigned long bch2_kthread_io_clock_wait_once(struct io_clock *, u64, unsigned long);
 void bch2_kthread_io_clock_wait(struct io_clock *, u64, unsigned long);
 
 void __bch2_increment_clock(struct io_clock *, u64);

fs/bcachefs/compress.c

···
 			     __bch2_compression_types[crc.compression_type]))
 			ret = bch2_check_set_has_compressed_data(c, opt);
 		else
-			ret = -BCH_ERR_compression_workspace_not_initialized;
+			ret = bch_err_throw(c, compression_workspace_not_initialized);
 		if (ret)
 			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;
+			ret = bch_err_throw(c, decompress_lz4);
 		break;
 	case BCH_COMPRESSION_TYPE_gzip: {
 		z_stream strm = {
···
 		mempool_free(workspace, workspace_pool);
 
 		if (ret2 != Z_STREAM_END)
-			ret = -BCH_ERR_decompress_gzip;
+			ret = bch_err_throw(c, decompress_gzip);
 		break;
 	}
 	case BCH_COMPRESSION_TYPE_zstd: {
···
 		size_t real_src_len = le32_to_cpup(src_data.b);
 
 		if (real_src_len > src_len - 4) {
-			ret = -BCH_ERR_decompress_zstd_src_len_bad;
+			ret = bch_err_throw(c, decompress_zstd_src_len_bad);
 			goto err;
 		}
 
···
 		mempool_free(workspace, workspace_pool);
 
 		if (ret2 != dst_len)
-			ret = -BCH_ERR_decompress_zstd;
+			ret = bch_err_throw(c, decompress_zstd);
 		break;
 	}
 	default:
···
 		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;
+		return bch_err_throw(c, decompress_exceeded_max_encoded_extent);
 	}
 
 	data = __bounce_alloc(c, dst_len, WRITE);
···
 
 	if (crc.uncompressed_size << 9	> c->opts.encoded_extent_max ||
 	    crc.compressed_size << 9	> c->opts.encoded_extent_max)
-		return -BCH_ERR_decompress_exceeded_max_encoded_extent;
+		return bch_err_throw(c, decompress_exceeded_max_encoded_extent);
 
 	dst_data = dst_len == dst_iter.bi_size
 		? __bio_map_or_bounce(c, dst, dst_iter, WRITE)
···
 	if (!mempool_initialized(&c->compression_bounce[READ]) &&
 	    mempool_init_kvmalloc_pool(&c->compression_bounce[READ],
 				       1, c->opts.encoded_extent_max))
-		return -BCH_ERR_ENOMEM_compression_bounce_read_init;
+		return bch_err_throw(c, ENOMEM_compression_bounce_read_init);
 
 	if (!mempool_initialized(&c->compression_bounce[WRITE]) &&
 	    mempool_init_kvmalloc_pool(&c->compression_bounce[WRITE],
 				       1, c->opts.encoded_extent_max))
-		return -BCH_ERR_ENOMEM_compression_bounce_write_init;
+		return bch_err_throw(c, ENOMEM_compression_bounce_write_init);
 
 	for (i = compression_types;
 	     i < compression_types + ARRAY_SIZE(compression_types);
···
 		if (mempool_init_kvmalloc_pool(
 				&c->compress_workspace[i->type],
 				1, i->compress_workspace))
-			return -BCH_ERR_ENOMEM_compression_workspace_init;
+			return bch_err_throw(c, ENOMEM_compression_workspace_init);
 	}
 
 	return 0;

fs/bcachefs/darray.h

···
  * Inspired by CCAN's darray
  */
 
+#include <linux/cleanup.h>
 #include <linux/slab.h>
 
 #define DARRAY_PREALLOCATED(_type, _nr)					\
···
 #define darray_remove_item(_d, _pos)					\
 	array_remove_item((_d)->data, (_d)->nr, (_pos) - (_d)->data)
 
-#define __darray_for_each(_d, _i)						\
+#define darray_find_p(_d, _i, cond)					\
+({									\
+	typeof((_d).data) _ret = NULL;					\
+									\
+	darray_for_each(_d, _i)						\
+		if (cond) {						\
+			_ret = _i;					\
+			break;						\
+		}							\
+	_ret;								\
+})
+
+#define darray_find(_d, _item)	darray_find_p(_d, _i, *_i == _item)
+
+/* Iteration: */
+
+#define __darray_for_each(_d, _i)					\
 	for ((_i) = (_d).data; _i < (_d).data + (_d).nr; _i++)
 
 #define darray_for_each(_d, _i)						\
···
 
 #define darray_for_each_reverse(_d, _i)					\
 	for (typeof(&(_d).data[0]) _i = (_d).data + (_d).nr - 1; _i >= (_d).data && (_d).nr; --_i)
+
+/* Init/exit */
 
 #define darray_init(_d)							\
 do {									\
···
 		kvfree((_d)->data);					\
 	darray_init(_d);						\
 } while (0)
+
+#define DEFINE_DARRAY_CLASS(_type)					\
+DEFINE_CLASS(_type, _type, darray_exit(&(_T)), (_type) {}, void)
+
+#define DEFINE_DARRAY(_type)						\
+typedef DARRAY(_type)	darray_##_type;					\
+DEFINE_DARRAY_CLASS(darray_##_type)
+
+#define DEFINE_DARRAY_NAMED(_name, _type)				\
+typedef DARRAY(_type)	_name;						\
+DEFINE_DARRAY_CLASS(_name)
+
+DEFINE_DARRAY_CLASS(darray_char);
+DEFINE_DARRAY_CLASS(darray_str)
+DEFINE_DARRAY_CLASS(darray_const_str)
+
+DEFINE_DARRAY_CLASS(darray_u8)
+DEFINE_DARRAY_CLASS(darray_u16)
+DEFINE_DARRAY_CLASS(darray_u32)
+DEFINE_DARRAY_CLASS(darray_u64)
+
+DEFINE_DARRAY_CLASS(darray_s8)
+DEFINE_DARRAY_CLASS(darray_s16)
+DEFINE_DARRAY_CLASS(darray_s32)
+DEFINE_DARRAY_CLASS(darray_s64)
 
 #endif /* _BCACHEFS_DARRAY_H */

fs/bcachefs/data_update.c

···
 	}
 }
 
-static bool bkey_nocow_lock(struct bch_fs *c, struct moving_context *ctxt, struct bkey_s_c k)
+static noinline_for_stack
+bool __bkey_nocow_lock(struct bch_fs *c, struct moving_context *ctxt, struct bkey_ptrs_c ptrs,
+		       const struct bch_extent_ptr *start)
 {
-	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+	if (!ctxt) {
+		bkey_for_each_ptr(ptrs, ptr) {
+			if (ptr == start)
+				break;
 
+			struct bch_dev *ca = bch2_dev_have_ref(c, ptr->dev);
+			struct bpos bucket = PTR_BUCKET_POS(ca, ptr);
+			bch2_bucket_nocow_unlock(&c->nocow_locks, bucket, 0);
+		}
+		return false;
+	}
+
+	__bkey_for_each_ptr(start, ptrs.end, ptr) {
+		struct bch_dev *ca = bch2_dev_have_ref(c, ptr->dev);
+		struct bpos bucket = PTR_BUCKET_POS(ca, ptr);
+
+		bool locked;
+		move_ctxt_wait_event(ctxt,
+				     (locked = bch2_bucket_nocow_trylock(&c->nocow_locks, bucket, 0)) ||
+				     list_empty(&ctxt->ios));
+		if (!locked)
+			bch2_bucket_nocow_lock(&c->nocow_locks, bucket, 0);
+	}
+	return true;
+}
+
+static bool bkey_nocow_lock(struct bch_fs *c, struct moving_context *ctxt, struct bkey_ptrs_c ptrs)
+{
 	bkey_for_each_ptr(ptrs, ptr) {
 		struct bch_dev *ca = bch2_dev_have_ref(c, ptr->dev);
 		struct bpos bucket = PTR_BUCKET_POS(ca, ptr);
 
-		if (ctxt) {
-			bool locked;
-
-			move_ctxt_wait_event(ctxt,
-				(locked = bch2_bucket_nocow_trylock(&c->nocow_locks, bucket, 0)) ||
-				list_empty(&ctxt->ios));
-
-			if (!locked)
-				bch2_bucket_nocow_lock(&c->nocow_locks, bucket, 0);
-		} else {
-			if (!bch2_bucket_nocow_trylock(&c->nocow_locks, bucket, 0)) {
-				bkey_for_each_ptr(ptrs, ptr2) {
-					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);
-				}
-				return false;
-			}
-		}
+		if (!bch2_bucket_nocow_trylock(&c->nocow_locks, bucket, 0))
+			return __bkey_nocow_lock(c, ctxt, ptrs, ptr);
 	}
+
 	return true;
 }
 
···
 	bch2_print_str(c, KERN_ERR, buf.buf);
 	printbuf_exit(&buf);
 
-	return -BCH_ERR_invalid_bkey;
+	return bch_err_throw(c, invalid_bkey);
 }
 
 static int __bch2_data_update_index_update(struct btree_trans *trans,
···
 			bch2_bkey_durability(c, bkey_i_to_s_c(&new->k_i));
 
 		/* Now, drop excess replicas: */
-		rcu_read_lock();
+		scoped_guard(rcu) {
 restart_drop_extra_replicas:
-		bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs(bkey_i_to_s(insert)), p, entry) {
-			unsigned ptr_durability = bch2_extent_ptr_durability(c, &p);
+			bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs(bkey_i_to_s(insert)), p, entry) {
+				unsigned ptr_durability = bch2_extent_ptr_durability(c, &p);
 
-			if (!p.ptr.cached &&
-			    durability - ptr_durability >= m->op.opts.data_replicas) {
-				durability -= ptr_durability;
+				if (!p.ptr.cached &&
+				    durability - ptr_durability >= m->op.opts.data_replicas) {
+					durability -= ptr_durability;
 
-				bch2_extent_ptr_set_cached(c, &m->op.opts,
-							   bkey_i_to_s(insert), &entry->ptr);
-				goto restart_drop_extra_replicas;
+					bch2_extent_ptr_set_cached(c, &m->op.opts,
+								   bkey_i_to_s(insert), &entry->ptr);
+					goto restart_drop_extra_replicas;
+				}
 			}
 		}
-		rcu_read_unlock();
 
 		/* Finally, add the pointers we just wrote: */
 		extent_for_each_ptr_decode(extent_i_to_s(new), p, entry)
···
 	bch2_bkey_buf_exit(&update->k, c);
 }
 
-static int bch2_update_unwritten_extent(struct btree_trans *trans,
-					struct data_update *update)
+static noinline_for_stack
+int bch2_update_unwritten_extent(struct btree_trans *trans,
+				 struct data_update *update)
 {
 	struct bch_fs *c = update->op.c;
 	struct bkey_i_extent *e;
···
 		bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
 }
 
-int bch2_data_update_bios_init(struct data_update *m, struct bch_fs *c,
-			       struct bch_io_opts *io_opts)
+static int __bch2_data_update_bios_init(struct data_update *m, struct bch_fs *c,
+					struct bch_io_opts *io_opts,
+					unsigned buf_bytes)
 {
-	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);
···
 	return 0;
 }
 
+int bch2_data_update_bios_init(struct data_update *m, struct bch_fs *c,
+			       struct bch_io_opts *io_opts)
+{
+	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);
+
+	return __bch2_data_update_bios_init(m, c, io_opts, buf_bytes);
+}
+
 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;
+		return bch_err_throw(c, data_update_done_would_block);
 
 	unsigned target = m->op.flags & BCH_WRITE_only_specified_devs
 		? m->op.target
···
 	darray_for_each(m->op.devs_have, i)
 		__clear_bit(*i, devs.d);
 
-	rcu_read_lock();
+	guard(rcu)();
+
 	unsigned nr_replicas = 0, i;
 	for_each_set_bit(i, devs.d, BCH_SB_MEMBERS_MAX) {
 		struct bch_dev *ca = bch2_dev_rcu_noerror(c, i);
···
 		if (nr_replicas >= m->op.nr_replicas)
 			break;
 	}
-	rcu_read_unlock();
 
 	if (!nr_replicas)
-		return -BCH_ERR_data_update_done_no_rw_devs;
+		return bch_err_throw(c, data_update_done_no_rw_devs);
 	if (nr_replicas < m->op.nr_replicas)
-		return -BCH_ERR_insufficient_devices;
+		return bch_err_throw(c, insufficient_devices);
 	return 0;
 }
 
···
 			  struct bkey_s_c k)
 {
 	struct bch_fs *c = trans->c;
-	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
-	const union bch_extent_entry *entry;
-	struct extent_ptr_decoded p;
-	unsigned reserve_sectors = k.k->size * data_opts.extra_replicas;
 	int ret = 0;
 
-	/*
-	 * fs is corrupt  we have a key for a snapshot node that doesn't exist,
-	 * and we have to check for this because we go rw before repairing the
-	 * snapshots table - just skip it, we can move it later.
-	 */
-	if (unlikely(k.k->p.snapshot && !bch2_snapshot_exists(c, k.k->p.snapshot)))
-		return -BCH_ERR_data_update_done_no_snapshot;
+	if (k.k->p.snapshot) {
+		ret = bch2_check_key_has_snapshot(trans, iter, k);
+		if (bch2_err_matches(ret, BCH_ERR_recovery_will_run)) {
+			/* Can't repair yet, waiting on other recovery passes */
+			return bch_err_throw(c, data_update_done_no_snapshot);
+		}
+		if (ret < 0)
+			return ret;
+		if (ret) /* key was deleted */
+			return bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc) ?:
+				bch_err_throw(c, data_update_done_no_snapshot);
+		ret = 0;
+	}
 
 	bch2_bkey_buf_init(&m->k);
 	bch2_bkey_buf_reassemble(&m->k, c, k);
···
 
 	unsigned durability_have = 0, durability_removing = 0;
 
+	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;
+	unsigned reserve_sectors = k.k->size * data_opts.extra_replicas;
+	unsigned buf_bytes = 0;
+	bool unwritten = false;
+
 	unsigned ptr_bit = 1;
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 		if (!p.ptr.cached) {
-			rcu_read_lock();
+			guard(rcu)();
 			if (ptr_bit & m->data_opts.rewrite_ptrs) {
 				if (crc_is_compressed(p.crc))
 					reserve_sectors += k.k->size;
···
 				bch2_dev_list_add_dev(&m->op.devs_have, p.ptr.dev);
 				durability_have += bch2_extent_ptr_durability(c, &p);
 			}
-			rcu_read_unlock();
 		}
 
 		/*
···
 
 		if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
 			m->op.incompressible = true;
+
+		buf_bytes = max_t(unsigned, buf_bytes, p.crc.uncompressed_size << 9);
+		unwritten |= p.ptr.unwritten;
 
 		ptr_bit <<= 1;
 	}
···
 		if (iter)
 			ret = bch2_extent_drop_ptrs(trans, iter, k, io_opts, &m->data_opts);
 		if (!ret)
-			ret = -BCH_ERR_data_update_done_no_writes_needed;
+			ret = bch_err_throw(c, data_update_done_no_writes_needed);
 		goto out_bkey_buf_exit;
 	}
 
···
 	}
 
 	if (!bkey_get_dev_refs(c, k)) {
-		ret = -BCH_ERR_data_update_done_no_dev_refs;
+		ret = bch_err_throw(c, data_update_done_no_dev_refs);
 		goto out_put_disk_res;
 	}
 
 	if (c->opts.nocow_enabled &&
-	    !bkey_nocow_lock(c, ctxt, k)) {
-		ret = -BCH_ERR_nocow_lock_blocked;
+	    !bkey_nocow_lock(c, ctxt, ptrs)) {
+		ret = bch_err_throw(c, nocow_lock_blocked);
 		goto out_put_dev_refs;
 	}
 
-	if (bkey_extent_is_unwritten(k)) {
+	if (unwritten) {
 		ret = bch2_update_unwritten_extent(trans, m) ?:
-			-BCH_ERR_data_update_done_unwritten;
+			bch_err_throw(c, data_update_done_unwritten);
 		goto out_nocow_unlock;
 	}
 
-	ret = bch2_data_update_bios_init(m, c, io_opts);
+	bch2_trans_unlock(trans);
+
+	ret = __bch2_data_update_bios_init(m, c, io_opts, buf_bytes);
 	if (ret)
 		goto out_nocow_unlock;
 

fs/bcachefs/debug.c

···
 	prt_printf(out, "journal pin %px:\t%llu\n",
 		   &b->writes[1].journal, b->writes[1].journal.seq);
 
+	prt_printf(out, "ob:\t%u\n", b->ob.nr);
+
 	printbuf_indent_sub(out, 2);
 }
 
···
 	i->ret	= 0;
 
 	do {
-		struct bucket_table *tbl;
-		struct rhash_head *pos;
-		struct btree *b;
-
 		ret = bch2_debugfs_flush_buf(i);
 		if (ret)
 			return ret;
 
-		rcu_read_lock();
 		i->buf.atomic++;
-		tbl = rht_dereference_rcu(c->btree_cache.table.tbl,
-					  &c->btree_cache.table);
-		if (i->iter < tbl->size) {
-			rht_for_each_entry_rcu(b, pos, tbl, i->iter, hash)
-				bch2_cached_btree_node_to_text(&i->buf, c, b);
-			i->iter++;
-		} else {
-			done = true;
+		scoped_guard(rcu) {
+			struct bucket_table *tbl =
+				rht_dereference_rcu(c->btree_cache.table.tbl,
+						    &c->btree_cache.table);
+			if (i->iter < tbl->size) {
+				struct rhash_head *pos;
+				struct btree *b;
+
+				rht_for_each_entry_rcu(b, pos, tbl, i->iter, hash)
+					bch2_cached_btree_node_to_text(&i->buf, c, b);
+				i->iter++;
+			} else {
+				done = true;
+			}
 		}
 		--i->buf.atomic;
-		rcu_read_unlock();
 	} while (!done);
 
 	if (i->buf.allocation_failure)

fs/bcachefs/dirent.c

···
 	prt_printf(out, " type %s", bch2_d_type_str(d.v->d_type));
 }
 
-static struct bkey_i_dirent *dirent_alloc_key(struct btree_trans *trans,
+int bch2_dirent_init_name(struct bkey_i_dirent *dirent,
+			  const struct bch_hash_info *hash_info,
+			  const struct qstr *name,
+			  const struct qstr *cf_name)
+{
+	EBUG_ON(hash_info->cf_encoding == NULL && cf_name);
+	int cf_len = 0;
+
+	if (name->len > BCH_NAME_MAX)
+		return -ENAMETOOLONG;
+
+	dirent->v.d_casefold = hash_info->cf_encoding != NULL;
+
+	if (!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);
+	} else {
+#ifdef CONFIG_UNICODE
+		memcpy(&dirent->v.d_cf_name_block.d_names[0], name->name, name->len);
+
+		char *cf_out = &dirent->v.d_cf_name_block.d_names[name->len];
+
+		if (cf_name) {
+			cf_len = cf_name->len;
+
+			memcpy(cf_out, cf_name->name, cf_name->len);
+		} else {
+			cf_len = utf8_casefold(hash_info->cf_encoding, name,
+					       cf_out,
+					       bkey_val_end(bkey_i_to_s(&dirent->k_i)) - (void *) cf_out);
+			if (cf_len <= 0)
+				return cf_len;
+		}
+
+		memset(&dirent->v.d_cf_name_block.d_names[name->len + cf_len], 0,
+		       bkey_val_bytes(&dirent->k) -
+		       offsetof(struct bch_dirent, d_cf_name_block.d_names) -
+		       name->len + cf_len);
+
+		dirent->v.d_cf_name_block.d_name_len = cpu_to_le16(name->len);
+		dirent->v.d_cf_name_block.d_cf_name_len = cpu_to_le16(cf_len);
+
+		EBUG_ON(bch2_dirent_get_casefold_name(dirent_i_to_s_c(dirent)).len != cf_len);
+#else
+	return -EOPNOTSUPP;
+#endif
+	}
+
+	unsigned u64s = dirent_val_u64s(name->len, cf_len);
+	BUG_ON(u64s > bkey_val_u64s(&dirent->k));
+	set_bkey_val_u64s(&dirent->k, u64s);
+	return 0;
+}
+
+struct bkey_i_dirent *bch2_dirent_create_key(struct btree_trans *trans,
+				const struct bch_hash_info *hash_info,
 				subvol_inum dir,
 				u8 type,
-				int name_len, int cf_name_len,
+				const struct qstr *name,
+				const struct qstr *cf_name,
 				u64 dst)
 {
-	struct bkey_i_dirent *dirent;
-	unsigned u64s = BKEY_U64s + dirent_val_u64s(name_len, cf_name_len);
-
-	BUG_ON(u64s > U8_MAX);
-
-	dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
+	struct bkey_i_dirent *dirent = bch2_trans_kmalloc(trans, BKEY_U64s_MAX * sizeof(u64));
 	if (IS_ERR(dirent))
 		return dirent;
 
 	bkey_dirent_init(&dirent->k_i);
-	dirent->k.u64s = u64s;
+	dirent->k.u64s = BKEY_U64s_MAX;
 
 	if (type != DT_SUBVOL) {
 		dirent->v.d_inum = cpu_to_le64(dst);
···
 
 	dirent->v.d_type = type;
 	dirent->v.d_unused = 0;
-	dirent->v.d_casefold = cf_name_len ? 1 : 0;
 
-	return dirent;
-}
-
-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 = cpu_to_le16(name->len);
-	dirent->v.d_cf_name_block.d_cf_name_len = cpu_to_le16(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,
-				const struct bch_hash_info *hash_info,
-				subvol_inum dir,
-				u8 type,
-				const struct qstr *name,
-				const struct qstr *cf_name,
-				u64 dst)
-{
-	struct bkey_i_dirent *dirent;
-	struct qstr _cf_name;
-
-	if (name->len > BCH_NAME_MAX)
-		return ERR_PTR(-ENAMETOOLONG);
-
-	if (hash_info->cf_encoding && !cf_name) {
-		int ret = bch2_casefold(trans, hash_info, name, &_cf_name);
-		if (ret)
-			return ERR_PTR(ret);
-
-		cf_name = &_cf_name;
-	}
-
-	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);
+	int ret = bch2_dirent_init_name(dirent, hash_info, name, cf_name);
+	if (ret)
+		return ERR_PTR(ret);
 
 	EBUG_ON(bch2_dirent_get_name(dirent_i_to_s_c(dirent)).len != name->len);
-
 	return dirent;
 }
 
···
 	struct bkey_i_dirent *dirent;
 	int ret;
 
-	dirent = dirent_create_key(trans, hash_info, dir_inum, type, name, NULL, dst_inum);
+	dirent = bch2_dirent_create_key(trans, hash_info, dir_inum, type, name, NULL, dst_inum);
 	ret = PTR_ERR_OR_ZERO(dirent);
 	if (ret)
 		return ret;
···
 	struct bkey_i_dirent *dirent;
 	int ret;
 
-	dirent = dirent_create_key(trans, hash_info, dir, type, name, NULL, dst_inum);
+	dirent = bch2_dirent_create_key(trans, hash_info, dir, type, name, NULL, dst_inum);
 	ret = PTR_ERR_OR_ZERO(dirent);
 	if (ret)
 		return ret;
···
 }
 
 int bch2_dirent_rename(struct btree_trans *trans,
-		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,
+		subvol_inum src_dir, struct bch_hash_info *src_hash,
+		subvol_inum dst_dir, struct bch_hash_info *dst_hash,
 		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)
···
 		*src_offset = dst_iter.pos.offset;
 
 	/* Create new dst key: */
-	new_dst = dirent_create_key(trans, dst_hash, dst_dir, 0, dst_name,
-				    dst_hash->cf_encoding ? &dst_name_lookup : NULL, 0);
+	new_dst = bch2_dirent_create_key(trans, dst_hash, 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;
···
 
 	/* Create new src key: */
 	if (mode == BCH_RENAME_EXCHANGE) {
-		new_src = dirent_create_key(trans, src_hash, src_dir, 0, src_name,
-					    src_hash->cf_encoding ? &src_name_lookup : NULL, 0);
+		new_src = bch2_dirent_create_key(trans, src_hash, 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;
···
 	if ((mode == BCH_RENAME_EXCHANGE) &&
 	    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)
···
 			struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
 			if (d.v->d_type == DT_SUBVOL && le32_to_cpu(d.v->d_parent_subvol) != subvol)
 				continue;
-			ret = -BCH_ERR_ENOTEMPTY_dir_not_empty;
+			ret = bch_err_throw(trans->c, ENOTEMPTY_dir_not_empty);
 			break;
 		}
 	bch2_trans_iter_exit(trans, &iter);
···
 	return !ret;
 }
 
-int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
+int bch2_readdir(struct bch_fs *c, subvol_inum inum,
+		 struct bch_hash_info *hash_info,
+		 struct dir_context *ctx)
 {
 	struct bkey_buf sk;
 	bch2_bkey_buf_init(&sk);
···
 			struct bkey_s_c_dirent dirent = bkey_i_to_s_c_dirent(sk.k);
 
 			subvol_inum target;
-			int ret2 = bch2_dirent_read_target(trans, inum, dirent, &target);
+
+			bool need_second_pass = false;
+			int ret2 = bch2_str_hash_check_key(trans, NULL, &bch2_dirent_hash_desc,
+							   hash_info, &iter, k, &need_second_pass) ?:
+				bch2_dirent_read_target(trans, inum, dirent, &target);
 			if (ret2 > 0)
 				continue;
 
···
 		ret = bch2_inode_unpack(k, inode);
 		goto found;
 	}
-	ret = -BCH_ERR_ENOENT_inode;
+	ret = bch_err_throw(trans->c, ENOENT_inode);
 found:
 	bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
 	bch2_trans_iter_exit(trans, &iter);

fs/bcachefs/dirent.h

···
 	}
 }
 
-struct qstr bch2_dirent_get_name(struct bkey_s_c_dirent d);
+struct qstr bch2_dirent_get_name(struct bkey_s_c_dirent);
 
 static inline unsigned dirent_val_u64s(unsigned len, unsigned cf_len)
 {
···
 	dst->v.d_inum = src.v->d_inum;
 	dst->v.d_type = src.v->d_type;
 }
+
+int bch2_dirent_init_name(struct bkey_i_dirent *,
+			  const struct bch_hash_info *,
+			  const struct qstr *,
+			  const struct qstr *);
+struct bkey_i_dirent *bch2_dirent_create_key(struct btree_trans *,
+				const struct bch_hash_info *, subvol_inum, u8,
+				const struct qstr *, const struct qstr *, u64);
 
 int bch2_dirent_create_snapshot(struct btree_trans *, u32, u64, u32,
 			const struct bch_hash_info *, u8,
···
 };
 
 int bch2_dirent_rename(struct btree_trans *,
-		       subvol_inum, struct bch_hash_info *, u64 *,
-		       subvol_inum, struct bch_hash_info *, u64 *,
+		       subvol_inum, struct bch_hash_info *,
+		       subvol_inum, struct bch_hash_info *,
 		       const struct qstr *, subvol_inum *, u64 *,
 		       const struct qstr *, subvol_inum *, u64 *,
 		       enum bch_rename_mode);
···
 
 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_readdir(struct bch_fs *, subvol_inum, struct bch_hash_info *, struct dir_context *);
 
 int bch2_fsck_remove_dirent(struct btree_trans *, struct bpos);
 

fs/bcachefs/disk_accounting.c

···
 err:
 	free_percpu(n.v[1]);
 	free_percpu(n.v[0]);
-	return -BCH_ERR_ENOMEM_disk_accounting;
+	return bch_err_throw(c, ENOMEM_disk_accounting);
 }
 
 int bch2_accounting_mem_insert(struct bch_fs *c, struct bkey_s_c_accounting a,
···
 	if (mode != BCH_ACCOUNTING_read &&
 	    accounting_to_replicas(&r.e, a.k->p) &&
 	    !bch2_replicas_marked_locked(c, &r.e))
-		return -BCH_ERR_btree_insert_need_mark_replicas;
+		return bch_err_throw(c, btree_insert_need_mark_replicas);
 
 	percpu_up_read(&c->mark_lock);
 	percpu_down_write(&c->mark_lock);
···
 	if (mode != BCH_ACCOUNTING_read &&
 	    accounting_to_replicas(&r.e, a.k->p) &&
 	    !bch2_replicas_marked_locked(c, &r.e))
-		return -BCH_ERR_btree_insert_need_mark_replicas;
+		return bch_err_throw(c, btree_insert_need_mark_replicas);
 
 	return __bch2_accounting_mem_insert(c, a);
 }
···
 					     sizeof(u64), GFP_KERNEL);
 		if (!e->v[1]) {
 			bch2_accounting_free_counters(acc, true);
-			ret = -BCH_ERR_ENOMEM_disk_accounting;
+			ret = bch_err_throw(c, ENOMEM_disk_accounting);
 			break;
 		}
 	}
···
 				bch2_disk_accounting_mod(trans, acc, v, nr, false)) ?:
 			-BCH_ERR_remove_disk_accounting_entry;
 	} else {
-		ret = -BCH_ERR_remove_disk_accounting_entry;
+		ret = bch_err_throw(c, remove_disk_accounting_entry);
 	}
 	goto fsck_err;
 }
···
 		case BCH_DISK_ACCOUNTING_replicas:
 			fs_usage_data_type_to_base(usage, k.replicas.data_type, v[0]);
 			break;
-		case BCH_DISK_ACCOUNTING_dev_data_type:
-			rcu_read_lock();
+		case BCH_DISK_ACCOUNTING_dev_data_type: {
+			guard(rcu)();
 			struct bch_dev *ca = bch2_dev_rcu_noerror(c, k.dev_data_type.dev);
 			if (ca) {
 				struct bch_dev_usage_type __percpu *d = &ca->usage->d[k.dev_data_type.data_type];
···
 				    k.dev_data_type.data_type == BCH_DATA_journal)
 					usage->hidden += v[0] * ca->mi.bucket_size;
 			}
-			rcu_read_unlock();
 			break;
+		}
 		}
 	}
 	preempt_enable();
···
 			case BCH_DISK_ACCOUNTING_replicas:
 				fs_usage_data_type_to_base(&base, acc_k.replicas.data_type, a.v->d[0]);
 				break;
-			case BCH_DISK_ACCOUNTING_dev_data_type: {
-				rcu_read_lock();
-				struct bch_dev *ca = bch2_dev_rcu_noerror(c, acc_k.dev_data_type.dev);
-				if (!ca) {
-					rcu_read_unlock();
-					continue;
-				}
+			case BCH_DISK_ACCOUNTING_dev_data_type:
+				{
+					guard(rcu)(); /* scoped guard is a loop, and doesn't play nicely with continue */
+					struct bch_dev *ca = bch2_dev_rcu_noerror(c, acc_k.dev_data_type.dev);
+					if (!ca)
+						continue;
 
-				v[0] = percpu_u64_get(&ca->usage->d[acc_k.dev_data_type.data_type].buckets);
-				v[1] = percpu_u64_get(&ca->usage->d[acc_k.dev_data_type.data_type].sectors);
-				v[2] = percpu_u64_get(&ca->usage->d[acc_k.dev_data_type.data_type].fragmented);
-				rcu_read_unlock();
+					v[0] = percpu_u64_get(&ca->usage->d[acc_k.dev_data_type.data_type].buckets);
+					v[1] = percpu_u64_get(&ca->usage->d[acc_k.dev_data_type.data_type].sectors);
+					v[2] = percpu_u64_get(&ca->usage->d[acc_k.dev_data_type.data_type].fragmented);
+				}
 
 				if (memcmp(a.v->d, v, 3 * sizeof(u64))) {
 					struct printbuf buf = PRINTBUF;
···
 					printbuf_exit(&buf);
 					mismatch = true;
 				}
-			}
 			}
 
 			0;

fs/bcachefs/disk_accounting.h

···
 		case BCH_DISK_ACCOUNTING_replicas:
 			fs_usage_data_type_to_base(&trans->fs_usage_delta, acc_k.replicas.data_type, a.v->d[0]);
 			break;
-		case BCH_DISK_ACCOUNTING_dev_data_type:
-			rcu_read_lock();
+		case BCH_DISK_ACCOUNTING_dev_data_type: {
+			guard(rcu)();
 			struct bch_dev *ca = bch2_dev_rcu_noerror(c, acc_k.dev_data_type.dev);
 			if (ca) {
 				this_cpu_add(ca->usage->d[acc_k.dev_data_type.data_type].buckets, a.v->d[0]);
 				this_cpu_add(ca->usage->d[acc_k.dev_data_type.data_type].sectors, a.v->d[1]);
 				this_cpu_add(ca->usage->d[acc_k.dev_data_type.data_type].fragmented, a.v->d[2]);
 			}
-			rcu_read_unlock();
 			break;
+		}
 		}
 	}
 

fs/bcachefs/disk_groups.c

···
 
 	cpu_g = kzalloc(struct_size(cpu_g, entries, nr_groups), GFP_KERNEL);
 	if (!cpu_g)
-		return -BCH_ERR_ENOMEM_disk_groups_to_cpu;
+		return bch_err_throw(c, ENOMEM_disk_groups_to_cpu);
 
 	cpu_g->nr = nr_groups;
 
···
 const struct bch_devs_mask *bch2_target_to_mask(struct bch_fs *c, unsigned target)
 {
 	struct target t = target_decode(target);
-	struct bch_devs_mask *devs;
 
-	rcu_read_lock();
+	guard(rcu)();
 
 	switch (t.type) {
 	case TARGET_NULL:
-		devs = NULL;
-		break;
+		return NULL;
 	case TARGET_DEV: {
 		struct bch_dev *ca = t.dev < c->sb.nr_devices
 			? rcu_dereference(c->devs[t.dev])
 			: NULL;
-		devs = ca ? &ca->self : NULL;
-		break;
+		return ca ? &ca->self : NULL;
 	}
 	case TARGET_GROUP: {
 		struct bch_disk_groups_cpu *g = rcu_dereference(c->disk_groups);
 
-		devs = g && t.group < g->nr && !g->entries[t.group].deleted
+		return g && t.group < g->nr && !g->entries[t.group].deleted
 			? &g->entries[t.group].devs
 			: NULL;
-		break;
 	}
 	default:
 		BUG();
 	}
-
-	rcu_read_unlock();
-
-	return devs;
 }
 
 bool bch2_dev_in_target(struct bch_fs *c, unsigned dev, unsigned target)
···
 	bch2_printbuf_make_room(out, 4096);
 
 	out->atomic++;
-	rcu_read_lock();
+	guard(rcu)();
 	struct bch_disk_groups_cpu *g = rcu_dereference(c->disk_groups);
 
 	for (unsigned i = 0; i < (g ? g->nr : 0); i++) {
···
 		prt_newline(out);
 	}
 
-	rcu_read_unlock();
 	out->atomic--;
 }
 
 void bch2_disk_path_to_text(struct printbuf *out, struct bch_fs *c, unsigned v)
 {
 	out->atomic++;
-	rcu_read_lock();
+	guard(rcu)();
 	__bch2_disk_path_to_text(out, rcu_dereference(c->disk_groups), v),
-	rcu_read_unlock();
 	--out->atomic;
 }
 
···
 	switch (t.type) {
 	case TARGET_NULL:
 		prt_printf(out, "none");
-		break;
+		return;
 	case TARGET_DEV: {
-		struct bch_dev *ca;
-
 		out->atomic++;
-		rcu_read_lock();
-		ca = t.dev < c->sb.nr_devices
+		guard(rcu)();
+		struct bch_dev *ca = t.dev < c->sb.nr_devices
 			? rcu_dereference(c->devs[t.dev])
 			: NULL;
 
···
 		else
 			prt_printf(out, "invalid device %u", t.dev);
 
-		rcu_read_unlock();
 		out->atomic--;
-		break;
+		return;
 	}
 	case TARGET_GROUP:
 		bch2_disk_path_to_text(out, c, t.group);
-		break;
+		return;
 	default:
 		BUG();
 	}

fs/bcachefs/ec.c

···
 				a->dirty_sectors,
 				a->stripe, s.k->p.offset,
 				(bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
-			ret = -BCH_ERR_mark_stripe;
+			ret = bch_err_throw(c, mark_stripe);
 			goto err;
 		}
 
···
 				a->dirty_sectors,
 				a->cached_sectors,
 				(bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
-			ret = -BCH_ERR_mark_stripe;
+			ret = bch_err_throw(c, mark_stripe);
 			goto err;
 		}
 	} else {
···
 				bucket.inode, bucket.offset, a->gen,
 				a->stripe,
 				(bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
-			ret = -BCH_ERR_mark_stripe;
+			ret = bch_err_throw(c, mark_stripe);
 			goto err;
 		}
 
···
 				bch2_data_type_str(a->data_type),
 				bch2_data_type_str(data_type),
 				(bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
-			ret = -BCH_ERR_mark_stripe;
+			ret = bch_err_throw(c, mark_stripe);
 			goto err;
 		}
 
···
 				a->dirty_sectors,
 				a->cached_sectors,
 				(bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
-			ret = -BCH_ERR_mark_stripe;
+			ret = bch_err_throw(c, mark_stripe);
 			goto err;
 		}
 	}
···
 	struct bch_dev *ca = bch2_dev_tryget(c, ptr->dev);
 	if (unlikely(!ca)) {
 		if (ptr->dev != BCH_SB_MEMBER_INVALID && !(flags & BTREE_TRIGGER_overwrite))
-			ret = -BCH_ERR_mark_stripe;
+			ret = bch_err_throw(c, mark_stripe);
 		goto err;
 	}
 
···
 		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;
+			ret = bch_err_throw(c, mark_stripe);
 			goto err;
 		}
 
···
 			gc = genradix_ptr_alloc(&c->gc_stripes, idx, GFP_KERNEL);
 			if (!gc) {
 				bch_err(c, "error allocating memory for gc_stripes, idx %llu", idx);
-				return -BCH_ERR_ENOMEM_mark_stripe;
+				return bch_err_throw(c, ENOMEM_mark_stripe);
 			}
 
 			/*
···
 }
 
 /* XXX: this is a non-mempoolified memory allocation: */
-static int ec_stripe_buf_init(struct ec_stripe_buf *buf,
+static int ec_stripe_buf_init(struct bch_fs *c,
+			      struct ec_stripe_buf *buf,
 			      unsigned offset, unsigned size)
 {
 	struct bch_stripe *v = &bkey_i_to_stripe(&buf->key)->v;
···
 	return 0;
 err:
 	ec_stripe_buf_exit(buf);
-	return -BCH_ERR_ENOMEM_stripe_buf;
+	return bch_err_throw(c, ENOMEM_stripe_buf);
 }
 
 /* Checksumming: */
···
 
 	buf = kzalloc(sizeof(*buf), GFP_NOFS);
 	if (!buf)
-		return -BCH_ERR_ENOMEM_ec_read_extent;
+		return bch_err_throw(c, ENOMEM_ec_read_extent);
 
 	ret = lockrestart_do(trans, get_stripe_key_trans(trans, rbio->pick.ec.idx, buf));
 	if (ret) {
···
 		goto err;
 	}
 
-	ret = ec_stripe_buf_init(buf, offset, bio_sectors(&rbio->bio));
+	ret = ec_stripe_buf_init(c, buf, offset, bio_sectors(&rbio->bio));
 	if (ret) {
 		msg = "-ENOMEM";
 		goto err;
···
 	bch_err_ratelimited(c,
 			    "error doing reconstruct read: %s\n  %s", msg, msgbuf.buf);
 	printbuf_exit(&msgbuf);
-	ret = -BCH_ERR_stripe_reconstruct;
+	ret = bch_err_throw(c, stripe_reconstruct);
 	goto out;
 }
 
···
 {
 	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;
+		return bch_err_throw(c, ENOMEM_ec_stripe_mem_alloc);
 
 	return 0;
 }
···
 
 		bch2_fs_inconsistent(c, "%s", buf.buf);
 		printbuf_exit(&buf);
-		return -BCH_ERR_erasure_coding_found_btree_node;
+		return bch_err_throw(c, erasure_coding_found_btree_node);
 	}
 
 	k = bch2_backpointer_get_key(trans, bp, &iter, BTREE_ITER_intent, last_flushed);
···
 
 	struct bch_dev *ca = bch2_dev_tryget(c, ptr.dev);
 	if (!ca)
-		return -BCH_ERR_ENOENT_dev_not_found;
+		return bch_err_throw(c, ENOENT_dev_not_found);
 
 	struct bpos bucket_pos = PTR_BUCKET_POS(ca, &ptr);
 
···
 	struct bch_dev *ca = bch2_dev_get_ioref(c, ob->dev, WRITE,
 				BCH_DEV_WRITE_REF_ec_bucket_zero);
 	if (!ca) {
-		s->err = -BCH_ERR_erofs_no_writes;
+		s->err = bch_err_throw(c, erofs_no_writes);
 		return;
 	}
 
···
 
 		if (ec_do_recov(c, &s->existing_stripe)) {
 			bch_err(c, "error creating stripe: error reading existing stripe");
-			ret = -BCH_ERR_ec_block_read;
+			ret = bch_err_throw(c, ec_block_read);
 			goto err;
 		}
 
···
 
 	if (ec_nr_failed(&s->new_stripe)) {
 		bch_err(c, "error creating stripe: error writing redundancy buckets");
-		ret = -BCH_ERR_ec_block_write;
+		ret = bch_err_throw(c, ec_block_write);
 		goto err;
 	}
 
···
 static void ec_stripe_head_devs_update(struct bch_fs *c, struct ec_stripe_head *h)
 {
 	struct bch_devs_mask devs = h->devs;
+	unsigned nr_devs, nr_devs_with_durability;
 
-	rcu_read_lock();
-	h->devs = target_rw_devs(c, BCH_DATA_user, h->disk_label
-				 ? group_to_target(h->disk_label - 1)
-				 : 0);
-	unsigned nr_devs = dev_mask_nr(&h->devs);
+	scoped_guard(rcu) {
+		h->devs = target_rw_devs(c, BCH_DATA_user, h->disk_label
+					 ? group_to_target(h->disk_label - 1)
+					 : 0);
+		nr_devs = dev_mask_nr(&h->devs);
 
-	for_each_member_device_rcu(c, ca, &h->devs)
-		if (!ca->mi.durability)
-			__clear_bit(ca->dev_idx, h->devs.d);
-	unsigned nr_devs_with_durability = dev_mask_nr(&h->devs);
+		for_each_member_device_rcu(c, ca, &h->devs)
+			if (!ca->mi.durability)
+				__clear_bit(ca->dev_idx, h->devs.d);
+		nr_devs_with_durability = dev_mask_nr(&h->devs);
 
-	h->blocksize = pick_blocksize(c, &h->devs);
+		h->blocksize = pick_blocksize(c, &h->devs);
 
-	h->nr_active_devs = 0;
-	for_each_member_device_rcu(c, ca, &h->devs)
-		if (ca->mi.bucket_size == h->blocksize)
-			h->nr_active_devs++;
-
-	rcu_read_unlock();
+		h->nr_active_devs = 0;
+		for_each_member_device_rcu(c, ca, &h->devs)
+			if (ca->mi.bucket_size == h->blocksize)
+				h->nr_active_devs++;
+	}
 
 	/*
 	 * If we only have redundancy + 1 devices, we're better off with just
···
 	s->nr_data = existing_v->nr_blocks -
 		existing_v->nr_redundant;
 
-	int ret = ec_stripe_buf_init(&s->existing_stripe, 0, le16_to_cpu(existing_v->sectors));
+	int ret = ec_stripe_buf_init(c, &s->existing_stripe, 0, le16_to_cpu(existing_v->sectors));
 	if (ret) {
 		bch2_stripe_close(c, s);
 		return ret;
···
 	}
 	bch2_trans_iter_exit(trans, &lru_iter);
 	if (!ret)
-		ret = -BCH_ERR_stripe_alloc_blocked;
+		ret = bch_err_throw(c, stripe_alloc_blocked);
 	if (ret == 1)
 		ret = 0;
 	if (ret)
···
 				continue;
 			}
 
-			ret = -BCH_ERR_ENOSPC_stripe_create;
+			ret = bch_err_throw(c, ENOSPC_stripe_create);
 			break;
 		}
 
···
 	if (!h->s) {
 		h->s = ec_new_stripe_alloc(c, h);
 		if (!h->s) {
-			ret = -BCH_ERR_ENOMEM_ec_new_stripe_alloc;
+			ret = bch_err_throw(c, ENOMEM_ec_new_stripe_alloc);
 			bch_err(c, "failed to allocate new stripe");
 			goto err;
 		}
···
 		goto err;
 
 allocate_buf:
-	ret = ec_stripe_buf_init(&s->new_stripe, 0, h->blocksize);
+	ret = ec_stripe_buf_init(c, &s->new_stripe, 0, h->blocksize);
 	if (ret)
 		goto err;
 
···
 	if (k.k->type != KEY_TYPE_stripe)
 		return 0;
 
+	struct bch_fs *c = trans->c;
 	struct bkey_i_stripe *s =
 		bch2_bkey_make_mut_typed(trans, iter, &k, 0, stripe);
 	int ret = PTR_ERR_OR_ZERO(s);
···
 
 	unsigned nr_good = 0;
 
-	rcu_read_lock();
-	bkey_for_each_ptr(ptrs, ptr) {
-		if (ptr->dev == dev_idx)
-			ptr->dev = BCH_SB_MEMBER_INVALID;
+	scoped_guard(rcu)
+		bkey_for_each_ptr(ptrs, ptr) {
+			if (ptr->dev == dev_idx)
+				ptr->dev = BCH_SB_MEMBER_INVALID;
 
-		struct bch_dev *ca = bch2_dev_rcu(trans->c, ptr->dev);
-		nr_good += ca && ca->mi.state != BCH_MEMBER_STATE_failed;
-	}
-	rcu_read_unlock();
+			struct bch_dev *ca = bch2_dev_rcu(c, ptr->dev);
+			nr_good += ca && ca->mi.state != BCH_MEMBER_STATE_failed;
+		}
 
 	if (nr_good < s->v.nr_blocks && !(flags & BCH_FORCE_IF_DATA_DEGRADED))
-		return -BCH_ERR_remove_would_lose_data;
+		return bch_err_throw(c, remove_would_lose_data);
 
 	unsigned nr_data = s->v.nr_blocks - s->v.nr_redundant;
 
 	if (nr_good < nr_data && !(flags & BCH_FORCE_IF_DATA_LOST))
-		return -BCH_ERR_remove_would_lose_data;
+		return bch_err_throw(c, remove_would_lose_data);
 
 	sectors = -sectors;
 
···
 		return 0;
 
 	if (a->stripe_sectors) {
-		bch_err(trans->c, "trying to invalidate device in stripe when bucket has stripe data");
-		return -BCH_ERR_invalidate_stripe_to_dev;
+		struct bch_fs *c = trans->c;
+		bch_err(c, "trying to invalidate device in stripe when bucket has stripe data");
+		return bch_err_throw(c, invalidate_stripe_to_dev);
 	}
 
 	struct btree_iter iter;
 	struct bkey_s_c_stripe s =
 		bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_stripes, POS(0, a->stripe),
-					BTREE_ITER_slots, stripe);
+					 BTREE_ITER_slots, stripe);
 	int ret = bkey_err(s);
 	if (ret)
 		return ret;

fs/bcachefs/errcode.c

···
 	NULL
 };
 
-static unsigned bch2_errcode_parents[] = {
+static const unsigned bch2_errcode_parents[] = {
 #define x(class, err) [BCH_ERR_##err - BCH_ERR_START] = class,
 	BCH_ERRCODES()
 #undef x
 };
 
+__attribute__((const))
 const char *bch2_err_str(int err)
 {
 	const char *errstr;
···
 	return errstr ?: "(Invalid error)";
 }
 
+__attribute__((const))
 bool __bch2_err_matches(int err, int class)
 {
 	err	= abs(err);

fs/bcachefs/errcode.h

···
 	x(BCH_ERR_fsck,			fsck_errors_not_fixed)			\
 	x(BCH_ERR_fsck,			fsck_repair_unimplemented)		\
 	x(BCH_ERR_fsck,			fsck_repair_impossible)			\
-	x(EINVAL,			restart_recovery)			\
-	x(EINVAL,			cannot_rewind_recovery)			\
+	x(EINVAL,			recovery_will_run)			\
+	x(BCH_ERR_recovery_will_run,	restart_recovery)			\
+	x(BCH_ERR_recovery_will_run,	cannot_rewind_recovery)			\
+	x(BCH_ERR_recovery_will_run,	recovery_pass_will_run)			\
 	x(0,				data_update_done)			\
+	x(0,				bkey_was_deleted)			\
 	x(BCH_ERR_data_update_done,	data_update_done_would_block)		\
 	x(BCH_ERR_data_update_done,	data_update_done_unwritten)		\
 	x(BCH_ERR_data_update_done,	data_update_done_no_writes_needed)	\
···
 	x(EINVAL,			remove_would_lose_data)			\
 	x(EINVAL,			no_resize_with_buckets_nouse)		\
 	x(EINVAL,			inode_unpack_error)			\
+	x(EINVAL,			inode_not_unlinked)			\
+	x(EINVAL,			inode_has_child_snapshot)		\
 	x(EINVAL,			varint_decode_error)			\
 	x(EINVAL,			erasure_coding_found_btree_node)	\
 	x(EINVAL,			option_negative)			\
···
 	BCH_ERR_MAX
 };
 
-const char *bch2_err_str(int);
-bool __bch2_err_matches(int, int);
+__attribute__((const)) const char *bch2_err_str(int);
 
+__attribute__((const)) bool __bch2_err_matches(int, int);
+
+__attribute__((const))
 static inline bool _bch2_err_matches(int err, int class)
 {
 	return err < 0 && __bch2_err_matches(err, class);

fs/bcachefs/error.c

···
 	set_bit(BCH_FS_topology_error, &c->flags);
 	if (!test_bit(BCH_FS_in_recovery, &c->flags)) {
 		__bch2_inconsistent_error(c, out);
-		return -BCH_ERR_btree_need_topology_repair;
+		return bch_err_throw(c, btree_need_topology_repair);
 	} else {
 		return bch2_run_explicit_recovery_pass(c, out, BCH_RECOVERY_PASS_check_topology, 0) ?:
-			-BCH_ERR_btree_node_read_validate_error;
+			bch_err_throw(c, btree_node_read_validate_error);
 	}
 }
 
···
 
 	if (test_bit(BCH_FS_in_fsck, &c->flags)) {
 		if (!(flags & (FSCK_CAN_FIX|FSCK_CAN_IGNORE)))
-			return -BCH_ERR_fsck_repair_unimplemented;
+			return bch_err_throw(c, fsck_repair_unimplemented);
 
 		switch (c->opts.fix_errors) {
 		case FSCK_FIX_exit:
-			return -BCH_ERR_fsck_errors_not_fixed;
+			return bch_err_throw(c, fsck_errors_not_fixed);
 		case FSCK_FIX_yes:
 			if (flags & FSCK_CAN_FIX)
-				return -BCH_ERR_fsck_fix;
+				return bch_err_throw(c, fsck_fix);
 			fallthrough;
 		case FSCK_FIX_no:
 			if (flags & FSCK_CAN_IGNORE)
-				return -BCH_ERR_fsck_ignore;
-			return -BCH_ERR_fsck_errors_not_fixed;
+				return bch_err_throw(c, fsck_ignore);
+			return bch_err_throw(c, fsck_errors_not_fixed);
 		case FSCK_FIX_ask:
 			if (flags & FSCK_AUTOFIX)
-				return -BCH_ERR_fsck_fix;
-			return -BCH_ERR_fsck_ask;
+				return bch_err_throw(c, fsck_fix);
+			return bch_err_throw(c, fsck_ask);
 		default:
 			BUG();
 		}
···
 		if ((flags & FSCK_AUTOFIX) &&
 		    (c->opts.errors == BCH_ON_ERROR_continue ||
 		     c->opts.errors == BCH_ON_ERROR_fix_safe))
-			return -BCH_ERR_fsck_fix;
+			return bch_err_throw(c, fsck_fix);
 
 		if (c->opts.errors == BCH_ON_ERROR_continue &&
 		    (flags & FSCK_CAN_IGNORE))
-			return -BCH_ERR_fsck_ignore;
-		return -BCH_ERR_fsck_errors_not_fixed;
+			return bch_err_throw(c, fsck_ignore);
+		return bch_err_throw(c, fsck_errors_not_fixed);
 	}
 }
 
···
 {
 	va_list args;
 	struct printbuf buf = PRINTBUF, *out = &buf;
-	int ret = -BCH_ERR_fsck_ignore;
+	int ret = 0;
 	const char *action_orig = "fix?", *action = action_orig;
 
 	might_sleep();
···
 
 	if (test_bit(err, c->sb.errors_silent))
 		return flags & FSCK_CAN_FIX
-			? -BCH_ERR_fsck_fix
-			: -BCH_ERR_fsck_ignore;
+			? bch_err_throw(c, fsck_fix)
+			: bch_err_throw(c, fsck_ignore);
 
 	printbuf_indent_add_nextline(out, 2);
 
···
 		prt_str(out, ", ");
 		if (flags & FSCK_CAN_FIX) {
 			prt_actioning(out, action);
-			ret = -BCH_ERR_fsck_fix;
+			ret = bch_err_throw(c, fsck_fix);
 		} else {
 			prt_str(out, ", continuing");
-			ret = -BCH_ERR_fsck_ignore;
+			ret = bch_err_throw(c, fsck_ignore);
 		}
 
 		goto print;
···
 					 "run fsck, and forward to devs so error can be marked for self-healing");
 			inconsistent = true;
 			print = true;
-			ret = -BCH_ERR_fsck_errors_not_fixed;
+			ret = bch_err_throw(c, fsck_errors_not_fixed);
 		} else if (flags & FSCK_CAN_FIX) {
 			prt_str(out, ", ");
 			prt_actioning(out, action);
-			ret = -BCH_ERR_fsck_fix;
+			ret = bch_err_throw(c, fsck_fix);
 		} else {
 			prt_str(out, ", continuing");
-			ret = -BCH_ERR_fsck_ignore;
+			ret = bch_err_throw(c, fsck_ignore);
 		}
 	} else if (c->opts.fix_errors == FSCK_FIX_exit) {
 		prt_str(out, ", exiting");
-		ret = -BCH_ERR_fsck_errors_not_fixed;
+		ret = bch_err_throw(c, fsck_errors_not_fixed);
 	} else if (flags & FSCK_CAN_FIX) {
 		int fix = s && s->fix
 			? s->fix
···
 					: FSCK_FIX_yes;
 
 			ret = ret & 1
-				? -BCH_ERR_fsck_fix
-				: -BCH_ERR_fsck_ignore;
+				? bch_err_throw(c, fsck_fix)
+				: bch_err_throw(c, fsck_ignore);
 		} else if (fix == FSCK_FIX_yes ||
 			   (c->opts.nochanges &&
 			    !(flags & FSCK_CAN_IGNORE))) {
 			prt_str(out, ", ");
 			prt_actioning(out, action);
-			ret = -BCH_ERR_fsck_fix;
+			ret = bch_err_throw(c, fsck_fix);
 		} else {
 			prt_str(out, ", not ");
 			prt_actioning(out, action);
+			ret = bch_err_throw(c, fsck_ignore);
 		}
-	} else if (!(flags & FSCK_CAN_IGNORE)) {
-		prt_str(out, " (repair unimplemented)");
+	} else {
+		if (flags & FSCK_CAN_IGNORE) {
+			prt_str(out, ", continuing");
+			ret = bch_err_throw(c, fsck_ignore);
+		} else {
+			prt_str(out, " (repair unimplemented)");
+			ret = bch_err_throw(c, fsck_repair_unimplemented);
+		}
 	}
 
-	if (ret == -BCH_ERR_fsck_ignore &&
+	if (bch2_err_matches(ret, BCH_ERR_fsck_ignore) &&
 	    (c->opts.fix_errors == FSCK_FIX_exit ||
 	     !(flags & FSCK_CAN_IGNORE)))
-		ret = -BCH_ERR_fsck_errors_not_fixed;
+		ret = bch_err_throw(c, fsck_errors_not_fixed);
 
 	if (test_bit(BCH_FS_in_fsck, &c->flags) &&
-	    (ret != -BCH_ERR_fsck_fix &&
-	     ret != -BCH_ERR_fsck_ignore)) {
+	    (!bch2_err_matches(ret, BCH_ERR_fsck_fix) &&
+	     !bch2_err_matches(ret, BCH_ERR_fsck_ignore))) {
 		exiting = true;
 		print = true;
 	}
···
 
 	if (s)
 		s->ret = ret;
-
+err_unlock:
+	mutex_unlock(&c->fsck_error_msgs_lock);
+err:
 	/*
 	 * 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
 	 * which recovery pass corrects it, to get the fsck exit status correct:
 	 */
-	if (flags & FSCK_CAN_FIX) {
-		if (ret == -BCH_ERR_fsck_fix) {
-			set_bit(BCH_FS_errors_fixed, &c->flags);
-		} else {
-			set_bit(BCH_FS_errors_not_fixed, &c->flags);
-			set_bit(BCH_FS_error, &c->flags);
-		}
+	if (bch2_err_matches(ret, BCH_ERR_fsck_fix)) {
+		set_bit(BCH_FS_errors_fixed, &c->flags);
+	} else {
+		set_bit(BCH_FS_errors_not_fixed, &c->flags);
+		set_bit(BCH_FS_error, &c->flags);
 	}
-err_unlock:
-	mutex_unlock(&c->fsck_error_msgs_lock);
-err:
+
 	if (action != action_orig)
 		kfree(action);
 	printbuf_exit(&buf);
+
+	BUG_ON(!ret);
 	return ret;
 }
 
···
 			 const char *fmt, ...)
 {
 	if (from.flags & BCH_VALIDATE_silent)
-		return -BCH_ERR_fsck_delete_bkey;
+		return bch_err_throw(c, fsck_delete_bkey);
 
 	unsigned fsck_flags = 0;
 	if (!(from.flags & (BCH_VALIDATE_write|BCH_VALIDATE_commit))) {
 		if (test_bit(err, c->sb.errors_silent))
-			return -BCH_ERR_fsck_delete_bkey;
+			return bch_err_throw(c, fsck_delete_bkey);
 
 		fsck_flags |= FSCK_AUTOFIX|FSCK_CAN_FIX;
 	}

fs/bcachefs/error.h

···
 #define fsck_err_wrap(_do)						\
 ({									\
 	int _ret = _do;							\
-	if (_ret != -BCH_ERR_fsck_fix &&				\
-	    _ret != -BCH_ERR_fsck_ignore) {				\
+	if (!bch2_err_matches(_ret, BCH_ERR_fsck_fix) &&		\
+	    !bch2_err_matches(_ret, BCH_ERR_fsck_ignore)) {		\
 		ret = _ret;						\
 		goto fsck_err;						\
 	}								\
 									\
-	_ret == -BCH_ERR_fsck_fix;					\
+	bch2_err_matches(_ret, BCH_ERR_fsck_fix);			\
 })
 
 #define __fsck_err(...)		fsck_err_wrap(bch2_fsck_err(__VA_ARGS__))
···
 	int _ret = __bch2_bkey_fsck_err(c, k, from,			\
 				BCH_FSCK_ERR_##_err_type,		\
 				_err_msg, ##__VA_ARGS__);		\
-	if (_ret != -BCH_ERR_fsck_fix &&				\
-	    _ret != -BCH_ERR_fsck_ignore)				\
+	if (!bch2_err_matches(_ret, BCH_ERR_fsck_fix) &&		\
+	    !bch2_err_matches(_ret, BCH_ERR_fsck_ignore))		\
 		ret = _ret;						\
-	ret = -BCH_ERR_fsck_delete_bkey;				\
+	ret = bch_err_throw(c, fsck_delete_bkey);			\
 	goto fsck_err;							\
 } while (0)
 

fs/bcachefs/extents.c

···
 			continue;
 
 		bch2_printbuf_make_room(out, 1024);
-		rcu_read_lock();
 		out->atomic++;
-		struct bch_dev *ca = bch2_dev_rcu_noerror(c, f->dev);
-		if (ca)
-			prt_str(out, ca->name);
-		else
-			prt_printf(out, "(invalid device %u)", f->dev);
+		scoped_guard(rcu) {
+			struct bch_dev *ca = bch2_dev_rcu_noerror(c, f->dev);
+			if (ca)
+				prt_str(out, ca->name);
+			else
+				prt_printf(out, "(invalid device %u)", f->dev);
+		}
 		--out->atomic;
-		rcu_read_unlock();
 
 		prt_char(out, ' ');
 
···
 	bool have_dirty_ptrs = false, have_pick = false;
 
 	if (k.k->type == KEY_TYPE_error)
-		return -BCH_ERR_key_type_error;
+		return bch_err_throw(c, key_type_error);
 
 	rcu_read_lock();
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
···
 	if (!have_dirty_ptrs)
 		return 0;
 	if (have_missing_devs)
-		return -BCH_ERR_no_device_to_read_from;
+		return bch_err_throw(c, no_device_to_read_from);
 	if (have_csum_errors)
-		return -BCH_ERR_data_read_csum_err;
+		return bch_err_throw(c, data_read_csum_err);
 	if (have_io_errors)
-		return -BCH_ERR_data_read_io_err;
+		return bch_err_throw(c, 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;
+	return bch_err_throw(c, no_devices_valid);
 }
 
 /* KEY_TYPE_btree_ptr: */
···
 	lp.crc = bch2_extent_crc_unpack(l.k, NULL);
 	rp.crc = bch2_extent_crc_unpack(r.k, NULL);
 
+	guard(rcu)();
+
 	while (__bkey_ptr_next_decode(l.k, l_ptrs.end, lp, en_l) &&
 	       __bkey_ptr_next_decode(r.k, r_ptrs.end, rp, en_r)) {
 		if (lp.ptr.offset + lp.crc.offset + lp.crc.live_size !=
···
 			return false;
 
 		/* Extents may not straddle buckets: */
-		rcu_read_lock();
 		struct bch_dev *ca = bch2_dev_rcu(c, lp.ptr.dev);
 		bool same_bucket = ca && PTR_BUCKET_NR(ca, &lp.ptr) == PTR_BUCKET_NR(ca, &rp.ptr);
-		rcu_read_unlock();
 
 		if (!same_bucket)
 			return false;
···
 	struct extent_ptr_decoded p;
 	unsigned durability = 0;
 
-	rcu_read_lock();
+	guard(rcu)();
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 		durability += bch2_extent_ptr_durability(c, &p);
-	rcu_read_unlock();
-
 	return durability;
 }
 
···
 	struct extent_ptr_decoded p;
 	unsigned durability = 0;
 
-	rcu_read_lock();
+	guard(rcu)();
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 		if (p.ptr.dev < c->sb.nr_devices && c->devs[p.ptr.dev])
 			durability += bch2_extent_ptr_durability(c, &p);
-	rcu_read_unlock();
-
 	return durability;
 }
 
···
 {
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	struct bch_dev *ca;
-	bool ret = false;
 
-	rcu_read_lock();
+	guard(rcu)();
 	bkey_for_each_ptr(ptrs, ptr)
 		if (bch2_dev_in_target(c, ptr->dev, target) &&
 		    (ca = bch2_dev_rcu(c, ptr->dev)) &&
 		    (!ptr->cached ||
-		     !dev_ptr_stale_rcu(ca, ptr))) {
-			ret = true;
-			break;
-		}
-	rcu_read_unlock();
+		     !dev_ptr_stale_rcu(ca, ptr)))
+			return true;
 
-	return ret;
+	return false;
 }
 
 bool bch2_bkey_matches_ptr(struct bch_fs *c, struct bkey_s_c k,
···
 	bool have_cached_ptr;
 	unsigned drop_dev = ptr->dev;
 
-	rcu_read_lock();
+	guard(rcu)();
 restart_drop_ptrs:
 	ptrs = bch2_bkey_ptrs(k);
 	have_cached_ptr = false;
···
 		goto drop;
 
 	ptr->cached = true;
-	rcu_read_unlock();
 	return;
 drop:
-	rcu_read_unlock();
 	bch2_bkey_drop_ptr_noerror(k, ptr);
 }
 
···
 {
 	struct bch_dev *ca;
 
-	rcu_read_lock();
+	guard(rcu)();
 	bch2_bkey_drop_ptrs(k, ptr,
 		ptr->cached &&
 		(!(ca = bch2_dev_rcu(c, ptr->dev)) ||
 		 dev_ptr_stale_rcu(ca, ptr) > 0));
-	rcu_read_unlock();
 
 	return bkey_deleted(k.k);
 }
···
 	struct bkey_ptrs ptrs;
 	bool have_cached_ptr;
 
-	rcu_read_lock();
+	guard(rcu)();
 restart_drop_ptrs:
 	ptrs = bch2_bkey_ptrs(k);
 	have_cached_ptr = false;
···
 			}
 			have_cached_ptr = true;
 		}
-	rcu_read_unlock();
 
 	return bkey_deleted(k.k);
 }
···
 void bch2_extent_ptr_to_text(struct printbuf *out, struct bch_fs *c, const struct bch_extent_ptr *ptr)
 {
 	out->atomic++;
-	rcu_read_lock();
+	guard(rcu)();
 	struct bch_dev *ca = bch2_dev_rcu_noerror(c, ptr->dev);
 	if (!ca) {
 		prt_printf(out, "ptr: %u:%llu gen %u%s", ptr->dev,
···
 		else if (stale)
 			prt_printf(out, " invalid");
 	}
-	rcu_read_unlock();
 	--out->atomic;
 }
 
···
 				struct bch_compression_opt opt = __bch2_compression_decode(r->compression);
 				prt_printf(err, "invalid compression opt %u:%u",
 					   opt.type, opt.level);
-				return -BCH_ERR_invalid_bkey;
+				return bch_err_throw(c, invalid_bkey);
 			}
 #endif
 			break;

fs/bcachefs/fs-io-buffered.c

···
 	struct bch_io_opts	opts;
 	struct bch_folio_sector	*tmp;
 	unsigned		tmp_sectors;
+	struct blk_plug		plug;
 };
-
-static inline struct bch_writepage_state bch_writepage_state_init(struct bch_fs *c,
-								  struct bch_inode_info *inode)
-{
-	struct bch_writepage_state ret = { 0 };
-
-	bch2_inode_opts_get(&ret.opts, c, &inode->ei_inode);
-	return ret;
-}
 
 /*
  * Determine when a writepage io is full. We have to limit writepage bios to a
···
 int bch2_writepages(struct address_space *mapping, struct writeback_control *wbc)
 {
 	struct bch_fs *c = mapping->host->i_sb->s_fs_info;
-	struct bch_writepage_state w =
-		bch_writepage_state_init(c, to_bch_ei(mapping->host));
-	struct blk_plug plug;
-	int ret;
+	struct bch_writepage_state *w = kzalloc(sizeof(*w), GFP_NOFS|__GFP_NOFAIL);
 
-	blk_start_plug(&plug);
-	ret = write_cache_pages(mapping, wbc, __bch2_writepage, &w);
-	if (w.io)
-		bch2_writepage_do_io(&w);
-	blk_finish_plug(&plug);
-	kfree(w.tmp);
+	bch2_inode_opts_get(&w->opts, c, &to_bch_ei(mapping->host)->ei_inode);
+
+	blk_start_plug(&w->plug);
+	int ret = write_cache_pages(mapping, wbc, __bch2_writepage, w);
+	if (w->io)
+		bch2_writepage_do_io(w);
+	blk_finish_plug(&w->plug);
+	kfree(w->tmp);
+	kfree(w);
 	return bch2_err_class(ret);
 }
 

fs/bcachefs/fs-io-pagecache.c

···
 
 					if (!reserved) {
 						bch2_disk_reservation_put(c, &disk_res);
-						return -BCH_ERR_ENOSPC_disk_reservation;
+						return bch_err_throw(c, ENOSPC_disk_reservation);
 					}
 					break;
 				}

fs/bcachefs/fs-io.c

···
 	memset(&inode->ei_devs_need_flush, 0, sizeof(inode->ei_devs_need_flush));
 
 	for_each_set_bit(dev, devs.d, BCH_SB_MEMBERS_MAX) {
-		rcu_read_lock();
-		ca = rcu_dereference(c->devs[dev]);
-		if (ca && !enumerated_ref_tryget(&ca->io_ref[WRITE],
-					BCH_DEV_WRITE_REF_nocow_flush))
-			ca = NULL;
-		rcu_read_unlock();
+		scoped_guard(rcu) {
+			ca = rcu_dereference(c->devs[dev]);
+			if (ca && !enumerated_ref_tryget(&ca->io_ref[WRITE],
+							 BCH_DEV_WRITE_REF_nocow_flush))
+				ca = NULL;
+		}
 
 		if (!ca)
 			continue;

fs/bcachefs/fs-ioctl.c

···
 	}
 
 	if (dst_dentry->d_inode) {
-		error = -BCH_ERR_EEXIST_subvolume_create;
+		error = bch_err_throw(c, EEXIST_subvolume_create);
 		goto err3;
 	}
 
 	dir = dst_path.dentry->d_inode;
 	if (IS_DEADDIR(dir)) {
-		error = -BCH_ERR_ENOENT_directory_dead;
+		error = bch_err_throw(c, ENOENT_directory_dead);
 		goto err3;
 	}
 

fs/bcachefs/fs.c

···
 		goto err;
 
 	struct bch_extent_rebalance new_r = bch2_inode_rebalance_opts_get(c, &inode_u);
+	bool rebalance_changed = memcmp(&old_r, &new_r, sizeof(new_r));
 
-	if (memcmp(&old_r, &new_r, sizeof(new_r))) {
+	if (rebalance_changed) {
 		ret = bch2_set_rebalance_needs_scan_trans(trans, inode_u.bi_inum);
 		if (ret)
 			goto err;
···
 
 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 		goto retry;
+
+	if (rebalance_changed)
+		bch2_rebalance_wakeup(c);
 
 	bch2_fs_fatal_err_on(bch2_err_matches(ret, ENOENT), c,
 			     "%s: inode %llu:%llu not found when updating",
···
 {
 	struct bch_inode_info *inode = file_bch_inode(file);
 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
+	struct bch_hash_info hash = bch2_hash_info_init(c, &inode->ei_inode);
 
 	if (!dir_emit_dots(file, ctx))
 		return 0;
 
-	int ret = bch2_readdir(c, inode_inum(inode), ctx);
+	int ret = bch2_readdir(c, inode_inum(inode), &hash, ctx);
 
 	bch_err_fn(c, ret);
 	return bch2_err_class(ret);
···
 			goto err;
 
 		if (k.k->type != KEY_TYPE_dirent) {
-			ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
+			ret = bch_err_throw(c, ENOENT_dirent_doesnt_match_inode);
 			goto err;
 		}
 
 		d = bkey_s_c_to_dirent(k);
 		ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
 		if (ret > 0)
-			ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
+			ret = bch_err_throw(c, ENOENT_dirent_doesnt_match_inode);
 		if (ret)
 			goto err;
 
···
 				KEY_TYPE_QUOTA_WARN);
 		bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
 				KEY_TYPE_QUOTA_WARN);
-		bch2_inode_rm(c, inode_inum(inode));
+		int ret = bch2_inode_rm(c, inode_inum(inode));
+		if (ret && !bch2_err_matches(ret, EROFS)) {
+			bch_err_msg(c, ret, "VFS incorrectly tried to delete inode %llu:%llu",
+				    inode->ei_inum.subvol,
+				    inode->ei_inum.inum);
+			bch2_sb_error_count(c, BCH_FSCK_ERR_vfs_bad_inode_rm);
+		}
 
 		/*
 		 * If we are deleting, we need it present in the vfs hash table
···
 	struct bch_fs *c = root->d_sb->s_fs_info;
 	bool first = true;
 
-	rcu_read_lock();
+	guard(rcu)();
 	for_each_online_member_rcu(c, ca) {
 		if (!first)
 			seq_putc(seq, ':');
 		first = false;
 		seq_puts(seq, ca->disk_sb.sb_name);
 	}
-	rcu_read_unlock();
 
 	return 0;
 }
···
 
 	sb->s_bdi->ra_pages		= VM_READAHEAD_PAGES;
 
-	rcu_read_lock();
-	for_each_online_member_rcu(c, ca) {
-		struct block_device *bdev = ca->disk_sb.bdev;
+	scoped_guard(rcu) {
+		for_each_online_member_rcu(c, ca) {
+			struct block_device *bdev = ca->disk_sb.bdev;
 
-		/* XXX: create an anonymous device for multi device filesystems */
-		sb->s_bdev	= bdev;
-		sb->s_dev	= bdev->bd_dev;
-		break;
+			/* XXX: create an anonymous device for multi device filesystems */
+			sb->s_bdev	= bdev;
+			sb->s_dev	= bdev->bd_dev;
+			break;
+		}
 	}
-	rcu_read_unlock();
 
 	c->dev = sb->s_dev;
 

fs/bcachefs/fsck.c

···
 #include <linux/bsearch.h>
 #include <linux/dcache.h> /* struct qstr */
 
-static int dirent_points_to_inode_nowarn(struct bkey_s_c_dirent d,
+static int dirent_points_to_inode_nowarn(struct bch_fs *c,
+					 struct bkey_s_c_dirent d,
 					 struct bch_inode_unpacked *inode)
 {
 	if (d.v->d_type == DT_SUBVOL
 	    ? le32_to_cpu(d.v->d_child_subvol)	== inode->bi_subvol
 	    : le64_to_cpu(d.v->d_inum)		== inode->bi_inum)
 		return 0;
-	return -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
+	return bch_err_throw(c, ENOENT_dirent_doesnt_match_inode);
 }
 
 static void dirent_inode_mismatch_msg(struct printbuf *out,
···
 				  struct bkey_s_c_dirent dirent,
 				  struct bch_inode_unpacked *inode)
 {
-	int ret = dirent_points_to_inode_nowarn(dirent, inode);
+	int ret = dirent_points_to_inode_nowarn(c, dirent, inode);
 	if (ret) {
 		struct printbuf buf = PRINTBUF;
 		dirent_inode_mismatch_msg(&buf, c, dirent, inode);
···
 			goto found;
 		}
 	}
-	ret = -BCH_ERR_ENOENT_no_snapshot_tree_subvol;
+	ret = bch_err_throw(trans->c, ENOENT_no_snapshot_tree_subvol);
 found:
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
···
 
 	if (d_type != DT_DIR) {
 		bch_err(c, "error looking up lost+found: not a directory");
-		return -BCH_ERR_ENOENT_not_directory;
+		return bch_err_throw(c, ENOENT_not_directory);
 	}
 
 	/*
···
 
 	if (!bch2_snapshot_is_leaf(c, snapshotid)) {
 		bch_err(c, "need to reconstruct subvol, but have interior node snapshot");
-		return -BCH_ERR_fsck_repair_unimplemented;
+		return bch_err_throw(c, fsck_repair_unimplemented);
 	}
 
 	/*
···
 
 	return __bch2_fsck_write_inode(trans, &new_inode);
 }
-
-struct snapshots_seen {
-	struct bpos			pos;
-	snapshot_id_list		ids;
-};
 
 static inline void snapshots_seen_exit(struct snapshots_seen *s)
 {
···
 {
 	struct bch_fs *c = trans->c;
 
-	struct inode_walker_entry *i;
-	__darray_for_each(w->inodes, i)
-		if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, i->inode.bi_snapshot))
-			goto found;
+	struct inode_walker_entry *i = darray_find_p(w->inodes, i,
+		    bch2_snapshot_is_ancestor(c, k.k->p.snapshot, i->inode.bi_snapshot));
 
-	return NULL;
-found:
-	BUG_ON(k.k->p.snapshot > i->inode.bi_snapshot);
+	if (!i)
+		return NULL;
 
 	struct printbuf buf = PRINTBUF;
 	int ret = 0;
···
 		if (ret)
 			goto fsck_err;
 
-		ret = -BCH_ERR_transaction_restart_nested;
+		ret = bch_err_throw(c, transaction_restart_nested);
 		goto fsck_err;
 	}
 
···
 	int ret = 0;
 
 	if (d->v.d_type == DT_SUBVOL) {
-		BUG();
+		bch_err(trans->c, "%s does not support DT_SUBVOL", __func__);
+		ret = -BCH_ERR_fsck_repair_unimplemented;
 	} else {
 		ret = get_visible_inodes(trans, &target, s, le64_to_cpu(d->v.d_inum));
 		if (ret)
···
 	if (ret && !bch2_err_matches(ret, ENOENT))
 		return ret;
 
-	if ((ret || dirent_points_to_inode_nowarn(d, inode)) &&
+	if ((ret || dirent_points_to_inode_nowarn(c, d, inode)) &&
 	    inode->bi_subvol &&
 	    (inode->bi_flags & BCH_INODE_has_child_snapshot)) {
 		/* Older version of a renamed subvolume root: we won't have a
···
 			trans, inode_points_to_missing_dirent,
 			"inode points to missing dirent\n%s",
 			(bch2_inode_unpacked_to_text(&buf, inode), buf.buf)) ||
-	    fsck_err_on(!ret && dirent_points_to_inode_nowarn(d, inode),
+	    fsck_err_on(!ret && dirent_points_to_inode_nowarn(c, d, inode),
 			trans, inode_points_to_wrong_dirent,
 			"%s",
 			(printbuf_reset(&buf),
···
 		u.bi_flags &= ~BCH_INODE_unlinked;
 		do_update = true;
 		ret = 0;
+	}
+
+	if (fsck_err_on(S_ISDIR(u.bi_mode) && u.bi_size,
+			trans, inode_dir_has_nonzero_i_size,
+			"directory %llu:%u with nonzero i_size %lli",
+			u.bi_inum, u.bi_snapshot, u.bi_size)) {
+		u.bi_size = 0;
+		do_update = true;
 	}
 
 	ret = bch2_inode_has_child_snapshots(trans, k.k->p);
···
 			goto err;
 
 		inode->last_pos.inode--;
-		ret = -BCH_ERR_transaction_restart_nested;
+		ret = bch_err_throw(c, transaction_restart_nested);
 		goto err;
 	}
 
···
 			      sizeof(seen->ids.data[0]) * seen->ids.size,
 			      GFP_KERNEL);
 	if (!n.seen.ids.data)
-		return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
+		return bch_err_throw(c, ENOMEM_fsck_extent_ends_at);
 
 	__darray_for_each(extent_ends->e, i) {
 		if (i->snapshot == k.k->p.snapshot) {
···
 
 		bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
 			__func__, buf.buf);
-		ret = -BCH_ERR_internal_fsck_err;
+		ret = bch_err_throw(c, internal_fsck_err);
 		goto err;
 	}
 
···
 	    pos2.size != k2.k->size) {
 		bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
 			__func__, buf.buf);
-		ret = -BCH_ERR_internal_fsck_err;
+		ret = bch_err_throw(c, internal_fsck_err);
 		goto err;
 	}
 
···
 			 * We overwrote the second extent - restart
 			 * check_extent() from the top:
 			 */
-			ret = -BCH_ERR_transaction_restart_nested;
+			ret = bch_err_throw(c, transaction_restart_nested);
 		}
 	}
 fsck_err:
···
 			(bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
 		if (!new_parent_subvol) {
 			bch_err(c, "could not find a subvol for snapshot %u", d.k->p.snapshot);
-			return -BCH_ERR_fsck_repair_unimplemented;
+			return bch_err_throw(c, fsck_repair_unimplemented);
 		}
 
 		struct bkey_i_dirent *new_dirent = bch2_bkey_make_mut_typed(trans, iter, &d.s_c, 0, dirent);
···
 
 	if (ret) {
 		bch_err(c, "subvol %u points to missing inode root %llu", target_subvol, target_inum);
-		ret = -BCH_ERR_fsck_repair_unimplemented;
+		ret = bch_err_throw(c, fsck_repair_unimplemented);
 		goto err;
 	}
 
···
 			struct bch_hash_info *hash_info,
 			struct inode_walker *dir,
 			struct inode_walker *target,
-			struct snapshots_seen *s)
+			struct snapshots_seen *s,
+			bool *need_second_pass)
 {
 	struct bch_fs *c = trans->c;
 	struct inode_walker_entry *i;
···
 		*hash_info = bch2_hash_info_init(c, &i->inode);
 	dir->first_this_inode = false;
 
-	ret = bch2_str_hash_check_key(trans, s, &bch2_dirent_hash_desc, hash_info, iter, k);
+#ifdef CONFIG_UNICODE
+	hash_info->cf_encoding = bch2_inode_casefold(c, &i->inode) ? c->cf_encoding : NULL;
+#endif
+
+	ret = bch2_str_hash_check_key(trans, s, &bch2_dirent_hash_desc, hash_info,
+				      iter, k, need_second_pass);
 	if (ret < 0)
 		goto err;
 	if (ret) {
···
 			(printbuf_reset(&buf),
 			 bch2_bkey_val_to_text(&buf, c, k),
 			 buf.buf))) {
-		struct qstr name = bch2_dirent_get_name(d);
-		u32 subvol = d.v->d_type == DT_SUBVOL
-			? le32_to_cpu(d.v->d_parent_subvol)
-			: 0;
+		subvol_inum dir_inum = { .subvol = d.v->d_type == DT_SUBVOL
+				? le32_to_cpu(d.v->d_parent_subvol)
+				: 0,
+		};
 		u64 target = d.v->d_type == DT_SUBVOL
 			? le32_to_cpu(d.v->d_child_subvol)
 			: le64_to_cpu(d.v->d_inum);
-		u64 dir_offset;
+		struct qstr name = bch2_dirent_get_name(d);
 
-		ret =   bch2_hash_delete_at(trans,
+		struct bkey_i_dirent *new_d =
+			bch2_dirent_create_key(trans, hash_info, dir_inum,
+					       d.v->d_type, &name, NULL, target);
+		ret = PTR_ERR_OR_ZERO(new_d);
+		if (ret)
+			goto out;
+
+		new_d->k.p.inode	= d.k->p.inode;
+		new_d->k.p.snapshot	= d.k->p.snapshot;
+
+		struct btree_iter dup_iter = {};
+		ret =	bch2_hash_delete_at(trans,
 					    bch2_dirent_hash_desc, hash_info, iter,
 					    BTREE_UPDATE_internal_snapshot_node) ?:
-			bch2_dirent_create_snapshot(trans, subvol,
-						    d.k->p.inode, d.k->p.snapshot,
-						    hash_info,
-						    d.v->d_type,
-						    &name,
-						    target,
-						    &dir_offset,
-						    BTREE_ITER_with_updates|
-						    BTREE_UPDATE_internal_snapshot_node|
-						    STR_HASH_must_create) ?:
-			bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
-
-		/* might need another check_dirents pass */
+			bch2_str_hash_repair_key(trans, s,
+						 &bch2_dirent_hash_desc, hash_info,
+						 iter, bkey_i_to_s_c(&new_d->k_i),
+						 &dup_iter, bkey_s_c_null,
+						 need_second_pass);
 		goto out;
 	}
 
···
 err:
 fsck_err:
 	printbuf_exit(&buf);
-	bch_err_fn(c, ret);
 	return ret;
 }
 
···
 	struct inode_walker target = inode_walker_init();
 	struct snapshots_seen s;
 	struct bch_hash_info hash_info;
+	bool need_second_pass = false, did_second_pass = false;
+	int ret;
 
 	snapshots_seen_init(&s);
-
-	int ret = bch2_trans_run(c,
-		for_each_btree_key(trans, iter, BTREE_ID_dirents,
+again:
+	ret = bch2_trans_run(c,
+		for_each_btree_key_commit(trans, iter, BTREE_ID_dirents,
 				POS(BCACHEFS_ROOT_INO, 0),
 				BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
-			check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s)) ?:
+				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+			check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s,
+				     &need_second_pass)) ?:
 		check_subdir_count_notnested(trans, &dir));
+
+	if (!ret && need_second_pass && !did_second_pass) {
+		bch_info(c, "check_dirents requires second pass");
+		swap(did_second_pass, need_second_pass);
+		goto again;
+	}
+
+	if (!ret && need_second_pass) {
+		bch_err(c, "dirents not repairing");
+		ret = -EINVAL;
+	}
 
 	snapshots_seen_exit(&s);
 	inode_walker_exit(&dir);
···
 		       struct inode_walker *inode)
 {
 	struct bch_fs *c = trans->c;
-	struct inode_walker_entry *i;
-	int ret;
 
-	ret = bch2_check_key_has_snapshot(trans, iter, k);
+	int ret = bch2_check_key_has_snapshot(trans, iter, k);
 	if (ret < 0)
 		return ret;
 	if (ret)
 		return 0;
 
-	i = walk_inode(trans, inode, k);
+	struct inode_walker_entry *i = walk_inode(trans, inode, k);
 	ret = PTR_ERR_OR_ZERO(i);
 	if (ret)
 		return ret;
···
 		*hash_info = bch2_hash_info_init(c, &i->inode);
 	inode->first_this_inode = false;
 
-	ret = bch2_str_hash_check_key(trans, NULL, &bch2_xattr_hash_desc, hash_info, iter, k);
-	bch_err_fn(c, ret);
-	return ret;
+	bool need_second_pass = false;
+	return bch2_str_hash_check_key(trans, NULL, &bch2_xattr_hash_desc, hash_info,
+				      iter, k, &need_second_pass);
 }
 
 /*
···
 		if (!d) {
 			bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
 				new_size);
-			return -BCH_ERR_ENOMEM_fsck_add_nlink;
+			return bch_err_throw(c, ENOMEM_fsck_add_nlink);
 		}
 
 		if (t->d)

fs/bcachefs/fsck.h

···
 
 #include "str_hash.h"
 
+/* recoverds snapshot IDs of overwrites at @pos */
+struct snapshots_seen {
+	struct bpos			pos;
+	snapshot_id_list		ids;
+};
+
 int bch2_fsck_update_backpointers(struct btree_trans *,
 				  struct snapshots_seen *,
 				  const struct bch_hash_desc,

fs/bcachefs/inode.c

···
 #undef  x
 
 static int delete_ancestor_snapshot_inodes(struct btree_trans *, struct bpos);
+static int may_delete_deleted_inum(struct btree_trans *, subvol_inum);
 
 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
 
···
 		goto found_slot;
 
 	if (!ret && start == min)
-		ret = -BCH_ERR_ENOSPC_inode_create;
+		ret = bch_err_throw(trans->c, ENOSPC_inode_create);
 
 	if (ret) {
 		bch2_trans_iter_exit(trans, iter);
···
 	u32 snapshot;
 	int ret;
 
+	ret = lockrestart_do(trans, may_delete_deleted_inum(trans, inum));
+	if (ret)
+		goto err2;
+
 	/*
 	 * If this was a directory, there shouldn't be any real dirents left -
 	 * but there could be whiteouts (from hash collisions) that we should
 	 * delete:
 	 *
-	 * XXX: the dirent could ideally would delete whiteouts when they're no
+	 * XXX: the dirent code ideally would delete whiteouts when they're no
 	 * longer needed
 	 */
 	ret   = bch2_inode_delete_keys(trans, inum, BTREE_ID_extents) ?:
 		bch2_inode_delete_keys(trans, inum, BTREE_ID_xattrs) ?:
 		bch2_inode_delete_keys(trans, inum, BTREE_ID_dirents);
 	if (ret)
-		goto err;
+		goto err2;
 retry:
 	bch2_trans_begin(trans);
 
···
 		bch2_fs_inconsistent(c,
 				     "inode %llu:%u not found when deleting",
 				     inum.inum, snapshot);
-		ret = -BCH_ERR_ENOENT_inode;
+		ret = bch_err_throw(c, ENOENT_inode);
 		goto err;
 	}
 
···
 		bch2_fs_inconsistent(c,
 				     "inode %llu:%u not found when deleting",
 				     inum, snapshot);
-		ret = -BCH_ERR_ENOENT_inode;
+		ret = bch_err_throw(c, ENOENT_inode);
 		goto err;
 	}
 
···
 		delete_ancestor_snapshot_inodes(trans, SPOS(0, inum, snapshot));
 }
 
-static int may_delete_deleted_inode(struct btree_trans *trans,
-				    struct btree_iter *iter,
-				    struct bpos pos,
-				    bool *need_another_pass)
+static int may_delete_deleted_inode(struct btree_trans *trans, struct bpos pos,
+				    bool from_deleted_inodes)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_iter inode_iter;
···
 	if (ret)
 		return ret;
 
-	ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
-	if (fsck_err_on(!bkey_is_inode(k.k),
+	ret = bkey_is_inode(k.k) ? 0 : bch_err_throw(c, ENOENT_inode);
+	if (fsck_err_on(from_deleted_inodes && ret,
 			trans, deleted_inode_missing,
 			"nonexistent inode %llu:%u in deleted_inodes btree",
 			pos.offset, pos.snapshot))
 		goto delete;
+	if (ret)
+		goto out;
 
 	ret = bch2_inode_unpack(k, &inode);
 	if (ret)
···
 
 	if (S_ISDIR(inode.bi_mode)) {
 		ret = bch2_empty_dir_snapshot(trans, pos.offset, 0, pos.snapshot);
-		if (fsck_err_on(bch2_err_matches(ret, ENOTEMPTY),
+		if (fsck_err_on(from_deleted_inodes &&
+				bch2_err_matches(ret, ENOTEMPTY),
 				trans, deleted_inode_is_dir,
 				"non empty directory %llu:%u in deleted_inodes btree",
 				pos.offset, pos.snapshot))
···
 			goto out;
 	}
 
-	if (fsck_err_on(!(inode.bi_flags & BCH_INODE_unlinked),
+	ret = inode.bi_flags & BCH_INODE_unlinked ? 0 : bch_err_throw(c, inode_not_unlinked);
+	if (fsck_err_on(from_deleted_inodes && ret,
 			trans, deleted_inode_not_unlinked,
 			"non-deleted inode %llu:%u in deleted_inodes btree",
 			pos.offset, pos.snapshot))
 		goto delete;
+	if (ret)
+		goto out;
 
-	if (fsck_err_on(inode.bi_flags & BCH_INODE_has_child_snapshot,
+	ret = !(inode.bi_flags & BCH_INODE_has_child_snapshot)
+		? 0 : bch_err_throw(c, inode_has_child_snapshot);
+
+	if (fsck_err_on(from_deleted_inodes && ret,
 			trans, deleted_inode_has_child_snapshots,
 			"inode with child snapshots %llu:%u in deleted_inodes btree",
 			pos.offset, pos.snapshot))
 		goto delete;
+	if (ret)
+		goto out;
 
 	ret = bch2_inode_has_child_snapshots(trans, k.k->p);
 	if (ret < 0)
···
 			if (ret)
 				goto out;
 		}
+
+		if (!from_deleted_inodes) {
+			ret =   bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc) ?:
+				bch_err_throw(c, inode_has_child_snapshot);
+			goto out;
+		}
+
 		goto delete;
 
 	}
 
-	if (test_bit(BCH_FS_clean_recovery, &c->flags) &&
-	    !fsck_err(trans, deleted_inode_but_clean,
-		      "filesystem marked as clean but have deleted inode %llu:%u",
-		      pos.offset, pos.snapshot)) {
-		ret = 0;
-		goto out;
-	}
+	if (from_deleted_inodes) {
+		if (test_bit(BCH_FS_clean_recovery, &c->flags) &&
+		    !fsck_err(trans, deleted_inode_but_clean,
+			      "filesystem marked as clean but have deleted inode %llu:%u",
+			      pos.offset, pos.snapshot)) {
+			ret = 0;
+			goto out;
+		}
 
-	ret = 1;
+		ret = 1;
+	}
 out:
 fsck_err:
 	bch2_trans_iter_exit(trans, &inode_iter);
···
 	goto out;
 }
 
+static int may_delete_deleted_inum(struct btree_trans *trans, subvol_inum inum)
+{
+	u32 snapshot;
+
+	return bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot) ?:
+		may_delete_deleted_inode(trans, SPOS(0, inum.inum, snapshot), false);
+}
+
 int bch2_delete_dead_inodes(struct bch_fs *c)
 {
 	struct btree_trans *trans = bch2_trans_get(c);
-	bool need_another_pass;
 	int ret;
-again:
+
 	/*
 	 * if we ran check_inodes() unlinked inodes will have already been
 	 * cleaned up but the write buffer will be out of sync; therefore we
···
 	ret = bch2_btree_write_buffer_flush_sync(trans);
 	if (ret)
 		goto err;
-
-	need_another_pass = false;
 
 	/*
 	 * Weird transaction restart handling here because on successful delete,
···
 	ret = for_each_btree_key_commit(trans, iter, BTREE_ID_deleted_inodes, POS_MIN,
 					BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
 					NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
-		ret = may_delete_deleted_inode(trans, &iter, k.k->p, &need_another_pass);
+		ret = may_delete_deleted_inode(trans, k.k->p, true);
 		if (ret > 0) {
 			bch_verbose_ratelimited(c, "deleting unlinked inode %llu:%u",
 						k.k->p.offset, k.k->p.snapshot);
···
 
 		ret;
 	}));
-
-	if (!ret && need_another_pass)
-		goto again;
 err:
 	bch2_trans_put(trans);
+	bch_err_fn(c, ret);
 	return ret;
 }

fs/bcachefs/inode.h

···
 int bch2_inode_nlink_inc(struct bch_inode_unpacked *);
 void bch2_inode_nlink_dec(struct btree_trans *, struct bch_inode_unpacked *);
 
-static inline bool bch2_inode_should_have_single_bp(struct bch_inode_unpacked *inode)
-{
-	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);
-}
-
 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *);
 void bch2_inode_opts_get(struct bch_io_opts *, struct bch_fs *,
 			 struct bch_inode_unpacked *);

fs/bcachefs/io_misc.c

···
 				opts.data_replicas,
 				BCH_WATERMARK_normal, 0, &cl, &wp);
 		if (bch2_err_matches(ret, BCH_ERR_operation_blocked))
-			ret = -BCH_ERR_transaction_restart_nested;
+			ret = bch_err_throw(c, transaction_restart_nested);
 		if (ret)
 			goto err;
 

fs/bcachefs/io_read.c

···
 	if (!target)
 		return false;
 
-	rcu_read_lock();
+	guard(rcu)();
 	devs = bch2_target_to_mask(c, target) ?:
 		&c->rw_devs[BCH_DATA_user];
 
···
 		total += max(congested, 0LL);
 		nr++;
 	}
-	rcu_read_unlock();
 
 	return get_random_u32_below(nr * CONGESTED_MAX) < total;
 }
···
 		BUG_ON(!opts.promote_target);
 
 		if (!(flags & BCH_READ_may_promote))
-			return -BCH_ERR_nopromote_may_not;
+			return bch_err_throw(c, nopromote_may_not);
 
 		if (bch2_bkey_has_target(c, k, opts.promote_target))
-			return -BCH_ERR_nopromote_already_promoted;
+			return bch_err_throw(c, nopromote_already_promoted);
 
 		if (bkey_extent_is_unwritten(k))
-			return -BCH_ERR_nopromote_unwritten;
+			return bch_err_throw(c, nopromote_unwritten);
 
 		if (bch2_target_congested(c, opts.promote_target))
-			return -BCH_ERR_nopromote_congested;
+			return bch_err_throw(c, nopromote_congested);
 	}
 
 	if (rhashtable_lookup_fast(&c->promote_table, &pos,
 				   bch_promote_params))
-		return -BCH_ERR_nopromote_in_flight;
+		return bch_err_throw(c, nopromote_in_flight);
 
 	return 0;
 }
···
 
 	struct promote_op *op = kzalloc(sizeof(*op), GFP_KERNEL);
 	if (!op) {
-		ret = -BCH_ERR_nopromote_enomem;
+		ret = bch_err_throw(c, nopromote_enomem);
 		goto err_put;
 	}
 
···
 
 	if (rhashtable_lookup_insert_fast(&c->promote_table, &op->hash,
 					  bch_promote_params)) {
-		ret = -BCH_ERR_nopromote_in_flight;
+		ret = bch_err_throw(c, nopromote_in_flight);
 		goto err;
 	}
 
···
 
 	if (!bkey_and_val_eq(k, bkey_i_to_s_c(u->k.k))) {
 		/* extent we wanted to read no longer exists: */
-		rbio->ret = -BCH_ERR_data_read_key_overwritten;
+		rbio->ret = bch_err_throw(trans->c, data_read_key_overwritten);
 		goto err;
 	}
 
···
 
 	if ((bch2_bkey_extent_flags(k) & BIT_ULL(BCH_EXTENT_FLAG_poisoned)) &&
 	    !orig->data_update)
-		return -BCH_ERR_extent_poisoned;
+		return bch_err_throw(c, extent_poisoned);
 retry_pick:
 	ret = bch2_bkey_pick_read_device(c, k, failed, &pick, dev);
 
···
 
 		bch_err_ratelimited(c, "%s", buf.buf);
 		printbuf_exit(&buf);
-		ret = -BCH_ERR_data_read_no_encryption_key;
+		ret = bch_err_throw(c, data_read_no_encryption_key);
 		goto err;
 	}
 
···
 			if (ca)
 				enumerated_ref_put(&ca->io_ref[READ],
 					BCH_DEV_READ_REF_io_read);
-			rbio->ret = -BCH_ERR_data_read_buffer_too_small;
+			rbio->ret = bch_err_throw(c, data_read_buffer_too_small);
 			goto out_read_done;
 		}
 
···
 	 * have to signal that:
 	 */
 	if (u)
-		orig->ret = -BCH_ERR_data_read_key_overwritten;
+		orig->ret = bch_err_throw(c, data_read_key_overwritten);
 
 	zero_fill_bio_iter(&orig->bio, iter);
 out_read_done:
···
 					 c->opts.btree_node_size,
 					 c->opts.encoded_extent_max) /
 				   PAGE_SIZE, 0))
-		return -BCH_ERR_ENOMEM_bio_bounce_pages_init;
+		return bch_err_throw(c, ENOMEM_bio_bounce_pages_init);
 
 	if (bioset_init(&c->bio_read, 1, offsetof(struct bch_read_bio, bio),
 			BIOSET_NEED_BVECS))
-		return -BCH_ERR_ENOMEM_bio_read_init;
+		return bch_err_throw(c, ENOMEM_bio_read_init);
 
 	if (bioset_init(&c->bio_read_split, 1, offsetof(struct bch_read_bio, bio),
 			BIOSET_NEED_BVECS))
-		return -BCH_ERR_ENOMEM_bio_read_split_init;
+		return bch_err_throw(c, ENOMEM_bio_read_split_init);
 
 	if (rhashtable_init(&c->promote_table, &bch_promote_params))
-		return -BCH_ERR_ENOMEM_promote_table_init;
+		return bch_err_throw(c, ENOMEM_promote_table_init);
 
 	return 0;
 }

fs/bcachefs/io_read.h

···
 		return 0;
 
 	*data_btree = BTREE_ID_reflink;
+
+	struct bch_fs *c = trans->c;
 	struct btree_iter iter;
 	struct bkey_s_c k = bch2_lookup_indirect_extent(trans, &iter,
 						offset_into_extent,
···
 
 	if (bkey_deleted(k.k)) {
 		bch2_trans_iter_exit(trans, &iter);
-		return -BCH_ERR_missing_indirect_extent;
+		return bch_err_throw(c, missing_indirect_extent);
 	}
 
-	bch2_bkey_buf_reassemble(extent, trans->c, k);
+	bch2_bkey_buf_reassemble(extent, c, k);
 	bch2_trans_iter_exit(trans, &iter);
 	return 0;
 }

fs/bcachefs/io_write.c

···
 
 static noinline int bch2_write_drop_io_error_ptrs(struct bch_write_op *op)
 {
+	struct bch_fs *c = op->c;
 	struct keylist *keys = &op->insert_keys;
 	struct bkey_i *src, *dst = keys->keys, *n;
 
···
 					    test_bit(ptr->dev, op->failed.d));
 
 			if (!bch2_bkey_nr_ptrs(bkey_i_to_s_c(src)))
-				return -BCH_ERR_data_write_io;
+				return bch_err_throw(c, data_write_io);
 		}
 
 		if (dst != src)
···
 		op->crc.csum_type < BCH_CSUM_NR
 		? __bch2_csum_types[op->crc.csum_type]
 		: "(unknown)");
-	return -BCH_ERR_data_write_csum;
+	return bch_err_throw(c, data_write_csum);
 }
 
 static int bch2_write_extent(struct bch_write_op *op, struct write_point *wp,
···
 
 	e = bkey_s_c_to_extent(k);
 
-	rcu_read_lock();
+	guard(rcu)();
 	extent_for_each_ptr_decode(e, p, entry) {
-		if (crc_is_encoded(p.crc) || p.has_ec) {
-			rcu_read_unlock();
+		if (crc_is_encoded(p.crc) || p.has_ec)
 			return false;
-		}
 
 		replicas += bch2_extent_ptr_durability(c, &p);
 	}
-	rcu_read_unlock();
 
 	return replicas >= op->opts.data_replicas;
 }
···
 static void __bch2_nocow_write_done(struct bch_write_op *op)
 {
 	if (unlikely(op->flags & BCH_WRITE_io_error)) {
-		op->error = -BCH_ERR_data_write_io;
+		op->error = bch_err_throw(op->c, data_write_io);
 	} else if (unlikely(op->flags & BCH_WRITE_convert_unwritten))
 		bch2_nocow_write_convert_unwritten(op);
 }
···
 				    "pointer to invalid bucket in nocow path on device %llu\n  %s",
 				    stale_at->b.inode,
 				    (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
-		ret = -BCH_ERR_data_write_invalid_ptr;
+		ret = bch_err_throw(c, data_write_invalid_ptr);
 	} else {
 		/* We can retry this: */
-		ret = -BCH_ERR_transaction_restart;
+		ret = bch_err_throw(c, transaction_restart);
 	}
 	printbuf_exit(&buf);
 
···
 
 	if (unlikely(bio->bi_iter.bi_size & (c->opts.block_size - 1))) {
 		bch2_write_op_error(op, op->pos.offset, "misaligned write");
-		op->error = -BCH_ERR_data_write_misaligned;
+		op->error = bch_err_throw(c, data_write_misaligned);
 		goto err;
 	}
 
 	if (c->opts.nochanges) {
-		op->error = -BCH_ERR_erofs_no_writes;
+		op->error = bch_err_throw(c, erofs_no_writes);
 		goto err;
 	}
 
 	if (!(op->flags & BCH_WRITE_move) &&
 	    !enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_write)) {
-		op->error = -BCH_ERR_erofs_no_writes;
+		op->error = bch_err_throw(c, erofs_no_writes);
 		goto err;
 	}
 
···
 {
 	if (bioset_init(&c->bio_write,   1, offsetof(struct bch_write_bio, bio), BIOSET_NEED_BVECS) ||
 	    bioset_init(&c->replica_set, 4, offsetof(struct bch_write_bio, bio), 0))
-		return -BCH_ERR_ENOMEM_bio_write_init;
+		return bch_err_throw(c, ENOMEM_bio_write_init);
 
 	return 0;
 }

fs/bcachefs/journal.c

···
 	BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
 
 	if (j->blocked)
-		return -BCH_ERR_journal_blocked;
+		return bch_err_throw(c, journal_blocked);
 
 	if (j->cur_entry_error)
 		return j->cur_entry_error;
···
 		return ret;
 
 	if (!fifo_free(&j->pin))
-		return -BCH_ERR_journal_pin_full;
+		return bch_err_throw(c, journal_pin_full);
 
 	if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf))
-		return -BCH_ERR_journal_max_in_flight;
+		return bch_err_throw(c, journal_max_in_flight);
 
 	if (atomic64_read(&j->seq) - j->seq_write_started == JOURNAL_STATE_BUF_NR)
-		return -BCH_ERR_journal_max_open;
+		return bch_err_throw(c, journal_max_open);
 
 	if (unlikely(journal_cur_seq(j) >= JOURNAL_SEQ_MAX)) {
 		bch_err(c, "cannot start: journal seq overflow");
 		if (bch2_fs_emergency_read_only_locked(c))
 			bch_err(c, "fatal error - emergency read only");
-		return -BCH_ERR_journal_shutdown;
+		return bch_err_throw(c, journal_shutdown);
 	}
 
 	if (!j->free_buf && !buf->data)
-		return -BCH_ERR_journal_buf_enomem; /* will retry after write completion frees up a buf */
+		return bch_err_throw(c, journal_buf_enomem); /* will retry after write completion frees up a buf */
 
 	BUG_ON(!j->cur_entry_sectors);
 
···
 	u64s = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1);
 
 	if (u64s <= (ssize_t) j->early_journal_entries.nr)
-		return -BCH_ERR_journal_full;
+		return bch_err_throw(c, journal_full);
 
 	if (fifo_empty(&j->pin) && j->reclaim_thread)
 		wake_up_process(j->reclaim_thread);
···
 			journal_cur_seq(j));
 		if (bch2_fs_emergency_read_only_locked(c))
 			bch_err(c, "fatal error - emergency read only");
-		return -BCH_ERR_journal_shutdown;
+		return bch_err_throw(c, journal_shutdown);
 	}
 
 	BUG_ON(j->pin.back - 1 != atomic64_read(&j->seq));
···
 		return ret;
 
 	if (j->blocked)
-		return -BCH_ERR_journal_blocked;
+		return bch_err_throw(c, journal_blocked);
 
 	if ((flags & BCH_WATERMARK_MASK) < j->watermark) {
-		ret = -BCH_ERR_journal_full;
+		ret = bch_err_throw(c, journal_full);
 		can_discard = j->can_discard;
 		goto out;
 	}
 
 	if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf) && !journal_entry_is_open(j)) {
-		ret = -BCH_ERR_journal_max_in_flight;
+		ret = bch_err_throw(c, journal_max_in_flight);
 		goto out;
 	}
 
···
 		goto retry;
 
 	if (journal_error_check_stuck(j, ret, flags))
-		ret = -BCH_ERR_journal_stuck;
+		ret = bch_err_throw(c, journal_stuck);
 
 	if (ret == -BCH_ERR_journal_max_in_flight &&
 	    track_event_change(&c->times[BCH_TIME_blocked_journal_max_in_flight], true) &&
···
 {
 	u64 nsecs = 0;
 
-	rcu_read_lock();
+	guard(rcu)();
 	for_each_rw_member_rcu(c, ca)
 		nsecs = max(nsecs, ca->io_latency[WRITE].stats.max_duration);
-	rcu_read_unlock();
 
 	return nsecs_to_jiffies(nsecs);
 }
···
 int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 				 struct closure *parent)
 {
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct journal_buf *buf;
 	int ret = 0;
 
···
 
 	/* Recheck under lock: */
 	if (j->err_seq && seq >= j->err_seq) {
-		ret = -BCH_ERR_journal_flush_err;
+		ret = bch_err_throw(c, journal_flush_err);
 		goto out;
 	}
 
···
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 
 	if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_journal))
-		return -BCH_ERR_erofs_no_writes;
+		return bch_err_throw(c, erofs_no_writes);
 
 	int ret = __bch2_journal_meta(j);
 	enumerated_ref_put(&c->writes, BCH_WRITE_REF_journal);
···
 	new_buckets	= kcalloc(nr, sizeof(u64), GFP_KERNEL);
 	new_bucket_seq	= kcalloc(nr, sizeof(u64), GFP_KERNEL);
 	if (!bu || !ob || !new_buckets || !new_bucket_seq) {
-		ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets;
+		ret = bch_err_throw(c, ENOMEM_set_nr_journal_buckets);
 		goto err_free;
 	}
 
···
 	return ret;
 }
 
+int bch2_dev_journal_bucket_delete(struct bch_dev *ca, u64 b)
+{
+	struct bch_fs *c = ca->fs;
+	struct journal *j = &c->journal;
+	struct journal_device *ja = &ca->journal;
+
+	guard(mutex)(&c->sb_lock);
+	unsigned pos;
+	for (pos = 0; pos < ja->nr; pos++)
+		if (ja->buckets[pos] == b)
+			break;
+
+	if (pos == ja->nr) {
+		bch_err(ca, "journal bucket %llu not found when deleting", b);
+		return -EINVAL;
+	}
+
+	u64 *new_buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);;
+	if (!new_buckets)
+		return bch_err_throw(c, ENOMEM_set_nr_journal_buckets);
+
+	memcpy(new_buckets, ja->buckets, ja->nr * sizeof(u64));
+	memmove(&new_buckets[pos],
+		&new_buckets[pos + 1],
+		(ja->nr - 1 - pos) * sizeof(new_buckets[0]));
+
+	int ret = bch2_journal_buckets_to_sb(c, ca, ja->buckets, ja->nr - 1) ?:
+		bch2_write_super(c);
+	if (ret) {
+		kfree(new_buckets);
+		return ret;
+	}
+
+	scoped_guard(spinlock, &j->lock) {
+		if (pos < ja->discard_idx)
+			--ja->discard_idx;
+		if (pos < ja->dirty_idx_ondisk)
+			--ja->dirty_idx_ondisk;
+		if (pos < ja->dirty_idx)
+			--ja->dirty_idx;
+		if (pos < ja->cur_idx)
+			--ja->cur_idx;
+
+		ja->nr--;
+
+		memmove(&ja->buckets[pos],
+			&ja->buckets[pos + 1],
+			(ja->nr - pos) * sizeof(ja->buckets[0]));
+
+		memmove(&ja->bucket_seq[pos],
+			&ja->bucket_seq[pos + 1],
+			(ja->nr - pos) * sizeof(ja->bucket_seq[0]));
+
+		bch2_journal_space_available(j);
+	}
+
+	kfree(new_buckets);
+	return 0;
+}
+
 int bch2_dev_journal_alloc(struct bch_dev *ca, bool new_fs)
 {
 	struct bch_fs *c = ca->fs;
···
 
 	if (c->sb.features & BIT_ULL(BCH_FEATURE_small_image)) {
 		bch_err(c, "cannot allocate journal, filesystem is an unresized image file");
-		return -BCH_ERR_erofs_filesystem_full;
+		return bch_err_throw(c, erofs_filesystem_full);
 	}
 
 	unsigned nr;
 	int ret;
 
 	if (dynamic_fault("bcachefs:add:journal_alloc")) {
-		ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets;
+		ret = bch_err_throw(c, ENOMEM_set_nr_journal_buckets);
 		goto err;
 	}
 
···
 	init_fifo(&j->pin, roundup_pow_of_two(nr), GFP_KERNEL);
 	if (!j->pin.data) {
 		bch_err(c, "error reallocating journal fifo (%llu open entries)", nr);
-		return -BCH_ERR_ENOMEM_journal_pin_fifo;
+		return bch_err_throw(c, ENOMEM_journal_pin_fifo);
 	}
 
 	j->replay_journal_seq	= last_seq;
···
 
 int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
 {
+	struct bch_fs *c = ca->fs;
 	struct journal_device *ja = &ca->journal;
 	struct bch_sb_field_journal *journal_buckets =
 		bch2_sb_field_get(sb, journal);
···
 
 	ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
 	if (!ja->bucket_seq)
-		return -BCH_ERR_ENOMEM_dev_journal_init;
+		return bch_err_throw(c, ENOMEM_dev_journal_init);
 
 	unsigned nr_bvecs = DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE);
 
···
 		ja->bio[i] = kzalloc(struct_size(ja->bio[i], bio.bi_inline_vecs,
 				     nr_bvecs), GFP_KERNEL);
 		if (!ja->bio[i])
-			return -BCH_ERR_ENOMEM_dev_journal_init;
+			return bch_err_throw(c, ENOMEM_dev_journal_init);
 
 		ja->bio[i]->ca = ca;
 		ja->bio[i]->buf_idx = i;
···
 
 	ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
 	if (!ja->buckets)
-		return -BCH_ERR_ENOMEM_dev_journal_init;
+		return bch_err_throw(c, ENOMEM_dev_journal_init);
 
 	if (journal_buckets_v2) {
 		unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
···
 
 int bch2_fs_journal_init(struct journal *j)
 {
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+
 	j->free_buf_size = j->buf_size_want = JOURNAL_ENTRY_SIZE_MIN;
 	j->free_buf = kvmalloc(j->free_buf_size, GFP_KERNEL);
 	if (!j->free_buf)
-		return -BCH_ERR_ENOMEM_journal_buf;
+		return bch_err_throw(c, ENOMEM_journal_buf);
 
 	for (unsigned i = 0; i < ARRAY_SIZE(j->buf); i++)
 		j->buf[i].idx = i;
···
 	j->wq = alloc_workqueue("bcachefs_journal",
 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512);
 	if (!j->wq)
-		return -BCH_ERR_ENOMEM_fs_other_alloc;
+		return bch_err_throw(c, ENOMEM_fs_other_alloc);
 	return 0;
 }
 
···
 	printbuf_tabstop_push(out, 28);
 	out->atomic++;
 
-	rcu_read_lock();
+	guard(rcu)();
 	s = READ_ONCE(j->reservations);
 
 	prt_printf(out, "flags:\t");
···
 	}
 
 	prt_printf(out, "replicas want %u need %u\n", c->opts.metadata_replicas, c->opts.metadata_replicas_required);
-
-	rcu_read_unlock();
 
 	--out->atomic;
 }

fs/bcachefs/journal.h

···
 void __bch2_journal_debug_to_text(struct printbuf *, struct journal *);
 void bch2_journal_debug_to_text(struct printbuf *, struct journal *);
 
-int bch2_set_nr_journal_buckets(struct bch_fs *, struct bch_dev *,
-				unsigned nr);
+int bch2_set_nr_journal_buckets(struct bch_fs *, struct bch_dev *, unsigned);
+int bch2_dev_journal_bucket_delete(struct bch_dev *, u64);
+
 int bch2_dev_journal_alloc(struct bch_dev *, bool);
 int bch2_fs_journal_alloc(struct bch_fs *);
 

fs/bcachefs/journal_io.c

···
 	mutex_unlock(&c->sb_lock);
 }
 
-void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
-			       struct journal_replay *j)
+static void bch2_journal_ptr_to_text(struct printbuf *out, struct bch_fs *c, struct journal_ptr *p)
+{
+	struct bch_dev *ca = bch2_dev_tryget_noerror(c, p->dev);
+	prt_printf(out, "%s %u:%u:%u (sector %llu)",
+		   ca ? ca->name : "(invalid dev)",
+		   p->dev, p->bucket, p->bucket_offset, p->sector);
+	bch2_dev_put(ca);
+}
+
+void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c, struct journal_replay *j)
 {
 	darray_for_each(j->ptrs, i) {
 		if (i != j->ptrs.data)
 			prt_printf(out, " ");
-		prt_printf(out, "%u:%u:%u (sector %llu)",
-			   i->dev, i->bucket, i->bucket_offset, i->sector);
+		bch2_journal_ptr_to_text(out, c, i);
+	}
+}
+
+static void bch2_journal_datetime_to_text(struct printbuf *out, struct jset *j)
+{
+	for_each_jset_entry_type(entry, j, BCH_JSET_ENTRY_datetime) {
+		struct jset_entry_datetime *datetime =
+			container_of(entry, struct jset_entry_datetime, entry);
+		bch2_prt_datetime(out, le64_to_cpu(datetime->seconds));
+		break;
 	}
 }
 
···
 					struct journal_replay *j)
 {
 	prt_printf(out, "seq %llu ", le64_to_cpu(j->j.seq));
-
+	bch2_journal_datetime_to_text(out, &j->j);
+	prt_char(out, ' ');
 	bch2_journal_ptrs_to_text(out, c, j);
-
-	for_each_jset_entry_type(entry, &j->j, BCH_JSET_ENTRY_datetime) {
-		struct jset_entry_datetime *datetime =
-			container_of(entry, struct jset_entry_datetime, entry);
-		bch2_prt_datetime(out, le64_to_cpu(datetime->seconds));
-		break;
-	}
 }
 
 static struct nonce journal_nonce(const struct jset *jset)
···
 				journal_entry_radix_idx(c, le64_to_cpu(j->seq)),
 				GFP_KERNEL);
 	if (!_i)
-		return -BCH_ERR_ENOMEM_journal_entry_add;
+		return bch_err_throw(c, ENOMEM_journal_entry_add);
 
 	/*
 	 * Duplicate journal entries? If so we want the one that didn't have a
···
 replace:
 	i = kvmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL);
 	if (!i)
-		return -BCH_ERR_ENOMEM_journal_entry_add;
+		return bch_err_throw(c, ENOMEM_journal_entry_add);
 
 	darray_init(&i->ptrs);
 	i->csum_good		= entry_ptr.csum_good;
···
 		bch2_sb_error_count(c, BCH_FSCK_ERR_##_err);		\
 		if (bch2_fs_inconsistent(c,				\
 				"corrupt metadata before write: %s\n", _buf.buf)) {\
-			ret = -BCH_ERR_fsck_errors_not_fixed;		\
+			ret = bch_err_throw(c, fsck_errors_not_fixed);		\
 			goto fsck_err;					\
 		}							\
 		break;							\
···
 	bool first = true;
 
 	jset_entry_for_each_key(entry, k) {
+		/* We may be called on entries that haven't been validated: */
+		if (!k->k.u64s)
+			break;
+
 		if (!first) {
 			prt_newline(out);
 			bch2_prt_jset_entry_type(out, entry->type);
···
 	size_t		size;
 };
 
-static int journal_read_buf_realloc(struct journal_read_buf *b,
+static int journal_read_buf_realloc(struct bch_fs *c, struct journal_read_buf *b,
 				    size_t new_size)
 {
 	void *n;
 
 	/* the bios are sized for this many pages, max: */
 	if (new_size > JOURNAL_ENTRY_SIZE_MAX)
-		return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
+		return bch_err_throw(c, ENOMEM_journal_read_buf_realloc);
 
 	new_size = roundup_pow_of_two(new_size);
 	n = kvmalloc(new_size, GFP_KERNEL);
 	if (!n)
-		return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
+		return bch_err_throw(c, ENOMEM_journal_read_buf_realloc);
 
 	kvfree(b->data);
 	b->data = n;
···
 	u64 offset = bucket_to_sector(ca, ja->buckets[bucket]),
 	    end = offset + ca->mi.bucket_size;
 	bool saw_bad = false, csum_good;
-	struct printbuf err = PRINTBUF;
 	int ret = 0;
 
 	pr_debug("reading %u", bucket);
···
 
 			bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
 			if (!bio)
-				return -BCH_ERR_ENOMEM_journal_read_bucket;
+				return bch_err_throw(c, ENOMEM_journal_read_bucket);
 			bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ);
 
 			bio->bi_iter.bi_sector = offset;
···
 			kfree(bio);
 
 			if (!ret && bch2_meta_read_fault("journal"))
-				ret = -BCH_ERR_EIO_fault_injected;
+				ret = bch_err_throw(c, EIO_fault_injected);
 
 			bch2_account_io_completion(ca, BCH_MEMBER_ERROR_read,
 						   submit_time, !ret);
···
 				 * found on a different device, and missing or
 				 * no journal entries will be handled later
 				 */
-				goto out;
+				return 0;
 			}
 
 			j = buf->data;
···
 			break;
 		case JOURNAL_ENTRY_REREAD:
 			if (vstruct_bytes(j) > buf->size) {
-				ret = journal_read_buf_realloc(buf,
+				ret = journal_read_buf_realloc(c, buf,
 							vstruct_bytes(j));
 				if (ret)
-					goto err;
+					return ret;
 			}
 			goto reread;
 		case JOURNAL_ENTRY_NONE:
 			if (!saw_bad)
-				goto out;
+				return 0;
 			/*
 			 * On checksum error we don't really trust the size
 			 * field of the journal entry we read, so try reading
···
 			sectors = block_sectors(c);
 			goto next_block;
 		default:
-			goto err;
+			return ret;
 		}
 
 		if (le64_to_cpu(j->seq) > ja->highest_seq_found) {
···
 		 * bucket:
 		 */
 		if (le64_to_cpu(j->seq) < ja->bucket_seq[bucket])
-			goto out;
+			return 0;
 
 		ja->bucket_seq[bucket] = le64_to_cpu(j->seq);
 
-		enum bch_csum_type csum_type = JSET_CSUM_TYPE(j);
 		struct bch_csum csum;
 		csum_good = jset_csum_good(c, j, &csum);
 
 		bch2_account_io_completion(ca, BCH_MEMBER_ERROR_checksum, 0, csum_good);
 
 		if (!csum_good) {
-			bch_err_dev_ratelimited(ca, "%s",
-				(printbuf_reset(&err),
-				 prt_str(&err, "journal "),
-				 bch2_csum_err_msg(&err, csum_type, j->csum, csum),
-				 err.buf));
+			/*
+			 * Don't print an error here, we'll print the error
+			 * later if we need this journal entry
+			 */
 			saw_bad = true;
 		}
 
···
 		mutex_lock(&jlist->lock);
 		ret = journal_entry_add(c, ca, (struct journal_ptr) {
 					.csum_good	= csum_good,
+					.csum		= csum,
 					.dev		= ca->dev_idx,
 					.bucket		= bucket,
 					.bucket_offset	= offset -
···
 		case JOURNAL_ENTRY_ADD_OUT_OF_RANGE:
 			break;
 		default:
-			goto err;
+			return ret;
 		}
 next_block:
 		pr_debug("next");
···
 		j = ((void *) j) + (sectors << 9);
 	}
 
-out:
-	ret = 0;
-err:
-	printbuf_exit(&err);
-	return ret;
+	return 0;
 }
 
 static CLOSURE_CALLBACK(bch2_journal_read_device)
···
 	if (!ja->nr)
 		goto out;
 
-	ret = journal_read_buf_realloc(&buf, PAGE_SIZE);
+	ret = journal_read_buf_realloc(c, &buf, PAGE_SIZE);
 	if (ret)
 		goto err;
 
···
 	goto out;
 }
 
+noinline_for_stack
+static void bch2_journal_print_checksum_error(struct bch_fs *c, struct journal_replay *j)
+{
+	struct printbuf buf = PRINTBUF;
+	enum bch_csum_type csum_type = JSET_CSUM_TYPE(&j->j);
+	bool have_good = false;
+
+	prt_printf(&buf, "invalid journal checksum(s) at seq %llu ", le64_to_cpu(j->j.seq));
+	bch2_journal_datetime_to_text(&buf, &j->j);
+	prt_newline(&buf);
+
+	darray_for_each(j->ptrs, ptr)
+		if (!ptr->csum_good) {
+			bch2_journal_ptr_to_text(&buf, c, ptr);
+			prt_char(&buf, ' ');
+			bch2_csum_to_text(&buf, csum_type, ptr->csum);
+			prt_newline(&buf);
+		} else {
+			have_good = true;
+		}
+
+	prt_printf(&buf, "should be ");
+	bch2_csum_to_text(&buf, csum_type, j->j.csum);
+
+	if (have_good)
+		prt_printf(&buf, "\n(had good copy on another device)");
+
+	bch2_print_str(c, KERN_ERR, buf.buf);
+	printbuf_exit(&buf);
+}
+
+noinline_for_stack
+static int bch2_journal_check_for_missing(struct bch_fs *c, u64 start_seq, u64 end_seq)
+{
+	struct printbuf buf = PRINTBUF;
+	int ret = 0;
+
+	struct genradix_iter radix_iter;
+	struct journal_replay *i, **_i, *prev = NULL;
+	u64 seq = start_seq;
+
+	genradix_for_each(&c->journal_entries, radix_iter, _i) {
+		i = *_i;
+
+		if (journal_replay_ignore(i))
+			continue;
+
+		BUG_ON(seq > le64_to_cpu(i->j.seq));
+
+		while (seq < le64_to_cpu(i->j.seq)) {
+			while (seq < le64_to_cpu(i->j.seq) &&
+			       bch2_journal_seq_is_blacklisted(c, seq, false))
+				seq++;
+
+			if (seq == le64_to_cpu(i->j.seq))
+				break;
+
+			u64 missing_start = seq;
+
+			while (seq < le64_to_cpu(i->j.seq) &&
+			       !bch2_journal_seq_is_blacklisted(c, seq, false))
+				seq++;
+
+			u64 missing_end = seq - 1;
+
+			printbuf_reset(&buf);
+			prt_printf(&buf, "journal entries %llu-%llu missing! (replaying %llu-%llu)",
+				   missing_start, missing_end,
+				   start_seq, end_seq);
+
+			prt_printf(&buf, "\nprev at ");
+			if (prev) {
+				bch2_journal_ptrs_to_text(&buf, c, prev);
+				prt_printf(&buf, " size %zu", vstruct_sectors(&prev->j, c->block_bits));
+			} else
+				prt_printf(&buf, "(none)");
+
+			prt_printf(&buf, "\nnext at ");
+			bch2_journal_ptrs_to_text(&buf, c, i);
+			prt_printf(&buf, ", continue?");
+
+			fsck_err(c, journal_entries_missing, "%s", buf.buf);
+		}
+
+		prev = i;
+		seq++;
+	}
+fsck_err:
+	printbuf_exit(&buf);
+	return ret;
+}
+
 int bch2_journal_read(struct bch_fs *c,
 		      u64 *last_seq,
 		      u64 *blacklist_seq,
 		      u64 *start_seq)
 {
 	struct journal_list jlist;
-	struct journal_replay *i, **_i, *prev = NULL;
+	struct journal_replay *i, **_i;
 	struct genradix_iter radix_iter;
 	struct printbuf buf = PRINTBUF;
 	bool degraded = false, last_write_torn = false;
···
 		return 0;
 	}
 
-	bch_info(c, "journal read done, replaying entries %llu-%llu",
-		 *last_seq, *blacklist_seq - 1);
-
+	printbuf_reset(&buf);
+	prt_printf(&buf, "journal read done, replaying entries %llu-%llu",
+		   *last_seq, *blacklist_seq - 1);
 	if (*start_seq != *blacklist_seq)
-		bch_info(c, "dropped unflushed entries %llu-%llu",
-			 *blacklist_seq, *start_seq - 1);
+		prt_printf(&buf, " (unflushed %llu-%llu)", *blacklist_seq, *start_seq - 1);
+	bch_info(c, "%s", buf.buf);
 
 	/* Drop blacklisted entries and entries older than last_seq: */
 	genradix_for_each(&c->journal_entries, radix_iter, _i) {
···
 		}
 	}
 
-	/* Check for missing entries: */
-	seq = *last_seq;
-	genradix_for_each(&c->journal_entries, radix_iter, _i) {
-		i = *_i;
-
-		if (journal_replay_ignore(i))
-			continue;
-
-		BUG_ON(seq > le64_to_cpu(i->j.seq));
-
-		while (seq < le64_to_cpu(i->j.seq)) {
-			u64 missing_start, missing_end;
-			struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
-
-			while (seq < le64_to_cpu(i->j.seq) &&
-			       bch2_journal_seq_is_blacklisted(c, seq, false))
-				seq++;
-
-			if (seq == le64_to_cpu(i->j.seq))
-				break;
-
-			missing_start = seq;
-
-			while (seq < le64_to_cpu(i->j.seq) &&
-			       !bch2_journal_seq_is_blacklisted(c, seq, false))
-				seq++;
-
-			if (prev) {
-				bch2_journal_ptrs_to_text(&buf1, c, prev);
-				prt_printf(&buf1, " size %zu", vstruct_sectors(&prev->j, c->block_bits));
-			} else
-				prt_printf(&buf1, "(none)");
-			bch2_journal_ptrs_to_text(&buf2, c, i);
-
-			missing_end = seq - 1;
-			fsck_err(c, journal_entries_missing,
-				 "journal entries %llu-%llu missing! (replaying %llu-%llu)\n"
-				 "prev at %s\n"
-				 "next at %s, continue?",
-				 missing_start, missing_end,
-				 *last_seq, *blacklist_seq - 1,
-				 buf1.buf, buf2.buf);
-
-			printbuf_exit(&buf1);
-			printbuf_exit(&buf2);
-		}
-
-		prev = i;
-		seq++;
-	}
+	ret = bch2_journal_check_for_missing(c, *last_seq, *blacklist_seq - 1);
+	if (ret)
+		goto err;
 
 	genradix_for_each(&c->journal_entries, radix_iter, _i) {
 		union bch_replicas_padded replicas = {
···
 		if (journal_replay_ignore(i))
 			continue;
 
-		darray_for_each(i->ptrs, ptr) {
-			struct bch_dev *ca = bch2_dev_have_ref(c, ptr->dev);
-
-			if (!ptr->csum_good)
-				bch_err_dev_offset(ca, ptr->sector,
-						   "invalid journal checksum, seq %llu%s",
-						   le64_to_cpu(i->j.seq),
-						   i->csum_good ? " (had good copy on another device)" : "");
-		}
+		/*
+		 * Don't print checksum errors until we know we're going to use
+		 * a given journal entry:
+		 */
+		darray_for_each(i->ptrs, ptr)
+			if (!ptr->csum_good) {
+				bch2_journal_print_checksum_error(c, i);
+				break;
+			}
 
 		ret = jset_validate(c,
 				    bch2_dev_have_ref(c, i->ptrs.data[0].dev),
···
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 
-	rcu_read_lock();
+	guard(rcu)();
 	darray_for_each(*devs, i) {
 		struct bch_dev *ca = rcu_dereference(c->devs[*i]);
 		if (!ca)
···
 			ja->bucket_seq[ja->cur_idx] = le64_to_cpu(seq);
 		}
 	}
-	rcu_read_unlock();
 }
 
 static void __journal_write_alloc(struct journal *j,
···
 
 retry_target:
 	devs = target_rw_devs(c, BCH_DATA_journal, target);
-	devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe, &devs);
+	bch2_dev_alloc_list(c, &j->wp.stripe, &devs, &devs_sorted);
 retry_alloc:
 	__journal_write_alloc(j, w, &devs_sorted, sectors, replicas, replicas_want);
 
···
 	}
 done:
 	BUG_ON(bkey_val_u64s(&w->key.k) > BCH_REPLICAS_MAX);
+
+#if 0
+	/*
+	 * XXX: we need a way to alert the user when we go degraded for any
+	 * reason
+	 */
+	if (*replicas < min(replicas_want,
+			    dev_mask_nr(&c->rw_devs[BCH_DATA_free]))) {
+	}
+#endif
 
 	return *replicas >= replicas_need ? 0 : -BCH_ERR_insufficient_journal_devices;
 }
···
 			       : j->noflush_write_time, j->write_start_time);
 
 	if (!w->devs_written.nr) {
-		err = -BCH_ERR_journal_write_err;
+		err = bch_err_throw(c, journal_write_err);
 	} else {
 		bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
 					 w->devs_written);
···
 	struct journal *j = container_of(w, struct journal, buf[w->idx]);
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	union bch_replicas_padded replicas;
-	unsigned nr_rw_members = dev_mask_nr(&c->rw_devs[BCH_DATA_journal]);
+	unsigned nr_rw_members = dev_mask_nr(&c->rw_devs[BCH_DATA_free]);
 	int ret;
 
 	BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));

fs/bcachefs/journal_io.h

···
 
 struct journal_ptr {
 	bool		csum_good;
+	struct bch_csum	csum;
 	u8		dev;
 	u32		bucket;
 	u32		bucket_offset;

fs/bcachefs/journal_reclaim.c

···
 journal_dev_space_available(struct journal *j, struct bch_dev *ca,
 			    enum journal_space_from from)
 {
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct journal_device *ja = &ca->journal;
 	unsigned sectors, buckets, unwritten;
+	unsigned bucket_size_aligned = round_down(ca->mi.bucket_size, block_sectors(c));
 	u64 seq;
 
 	if (from == journal_space_total)
 		return (struct journal_space) {
-			.next_entry	= ca->mi.bucket_size,
-			.total		= ca->mi.bucket_size * ja->nr,
+			.next_entry	= bucket_size_aligned,
+			.total		= bucket_size_aligned * ja->nr,
 		};
 
 	buckets = bch2_journal_dev_buckets_available(j, ja, from);
-	sectors = ja->sectors_free;
+	sectors = round_down(ja->sectors_free, block_sectors(c));
 
 	/*
 	 * We that we don't allocate the space for a journal entry
···
 			continue;
 
 		/* entry won't fit on this device, skip: */
-		if (unwritten > ca->mi.bucket_size)
+		if (unwritten > bucket_size_aligned)
 			continue;
 
 		if (unwritten >= sectors) {
···
 			}
 
 			buckets--;
-			sectors = ca->mi.bucket_size;
+			sectors = bucket_size_aligned;
 		}
 
 		sectors -= unwritten;
···
 
 	if (sectors < ca->mi.bucket_size && buckets) {
 		buckets--;
-		sectors = ca->mi.bucket_size;
+		sectors = bucket_size_aligned;
 	}
 
 	return (struct journal_space) {
 		.next_entry	= sectors,
-		.total		= sectors + buckets * ca->mi.bucket_size,
+		.total		= sectors + buckets * bucket_size_aligned,
 	};
 }
 
···
 
 	BUG_ON(nr_devs_want > ARRAY_SIZE(dev_space));
 
-	rcu_read_lock();
 	for_each_member_device_rcu(c, ca, &c->rw_devs[BCH_DATA_journal]) {
 		if (!ca->journal.nr ||
 		    !ca->mi.durability)
···
 
 		array_insert_item(dev_space, nr_devs, pos, space);
 	}
-	rcu_read_unlock();
 
 	if (nr_devs < nr_devs_want)
 		return (struct journal_space) { 0, 0 };
···
 	int ret = 0;
 
 	lockdep_assert_held(&j->lock);
+	guard(rcu)();
 
-	rcu_read_lock();
 	for_each_member_device_rcu(c, ca, &c->rw_devs[BCH_DATA_journal]) {
 		struct journal_device *ja = &ca->journal;
 
···
 		max_entry_size = min_t(unsigned, max_entry_size, ca->mi.bucket_size);
 		nr_online++;
 	}
-	rcu_read_unlock();
 
 	j->can_discard = can_discard;
 
···
 			prt_printf(&buf, "insufficient writeable journal devices available: have %u, need %u\n"
 				   "rw journal devs:", nr_online, metadata_replicas_required(c));
 
-			rcu_read_lock();
 			for_each_member_device_rcu(c, ca, &c->rw_devs[BCH_DATA_journal])
 				prt_printf(&buf, " %s", ca->name);
-			rcu_read_unlock();
 
 			bch_err(c, "%s", buf.buf);
 			printbuf_exit(&buf);
 		}
-		ret = -BCH_ERR_insufficient_journal_devices;
+		ret = bch_err_throw(c, insufficient_journal_devices);
 		goto out;
 	}
 
···
 	total		= j->space[journal_space_total].total;
 
 	if (!j->space[journal_space_discarded].next_entry)
-		ret = -BCH_ERR_journal_full;
+		ret = bch_err_throw(c, journal_full);
 
 	if ((j->space[journal_space_clean_ondisk].next_entry <
 	     j->space[journal_space_clean_ondisk].total) &&
···
 	bch2_journal_set_watermark(j);
 out:
 	j->cur_entry_sectors	= !ret
-		? round_down(j->space[journal_space_discarded].next_entry,
-			     block_sectors(c))
+		? j->space[journal_space_discarded].next_entry
 		: 0;
 	j->cur_entry_error	= ret;
 
···
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	u64 seq_to_flush = 0;
 
-	spin_lock(&j->lock);
+	guard(spinlock)(&j->lock);
+	guard(rcu)();
 
-	rcu_read_lock();
 	for_each_rw_member_rcu(c, ca) {
 		struct journal_device *ja = &ca->journal;
 		unsigned nr_buckets, bucket_to_flush;
···
 		seq_to_flush = max(seq_to_flush,
 				   ja->bucket_seq[bucket_to_flush]);
 	}
-	rcu_read_unlock();
 
 	/* Also flush if the pin fifo is more than half full */
-	seq_to_flush = max_t(s64, seq_to_flush,
-			     (s64) journal_cur_seq(j) -
-			     (j->pin.size >> 1));
-	spin_unlock(&j->lock);
-
-	return seq_to_flush;
+	return max_t(s64, seq_to_flush,
+		     (s64) journal_cur_seq(j) -
+		     (j->pin.size >> 1));
 }
 
 /**

fs/bcachefs/journal_sb.c

···
 	j = bch2_sb_field_resize(&ca->disk_sb, journal_v2,
 			 (sizeof(*j) + sizeof(j->d[0]) * nr_compacted) / sizeof(u64));
 	if (!j)
-		return -BCH_ERR_ENOSPC_sb_journal;
+		return bch_err_throw(c, ENOSPC_sb_journal);
 
 	bch2_sb_field_delete(&ca->disk_sb, BCH_SB_FIELD_journal);
 

fs/bcachefs/journal_seq_blacklist.c

···
 	bl = bch2_sb_field_resize(&c->disk_sb, journal_seq_blacklist,
 				  sb_blacklist_u64s(nr + 1));
 	if (!bl) {
-		ret = -BCH_ERR_ENOSPC_sb_journal_seq_blacklist;
+		ret = bch_err_throw(c, ENOSPC_sb_journal_seq_blacklist);
 		goto out;
 	}
 
···
 
 	t = kzalloc(struct_size(t, entries, nr), GFP_KERNEL);
 	if (!t)
-		return -BCH_ERR_ENOMEM_blacklist_table_init;
+		return bch_err_throw(c, ENOMEM_blacklist_table_init);
 
 	t->nr = nr;
 

fs/bcachefs/lru.c

···
 	case BCH_LRU_fragmentation: {
 		a = bch2_alloc_to_v4(k, &a_convert);
 
-		rcu_read_lock();
+		guard(rcu)();
 		struct bch_dev *ca = bch2_dev_rcu_noerror(c, k.k->p.inode);
-		u64 idx = ca
+		return ca
 			? alloc_lru_idx_fragmentation(*a, ca)
 			: 0;
-		rcu_read_unlock();
-		return idx;
 	}
 	case BCH_LRU_stripes:
 		return k.k->type == KEY_TYPE_stripe

fs/bcachefs/migrate.c

···
 	nr_good = bch2_bkey_durability(c, k.s_c);
 	if ((!nr_good && !(flags & lost)) ||
 	    (nr_good < replicas && !(flags & degraded)))
-		return -BCH_ERR_remove_would_lose_data;
+		return bch_err_throw(c, remove_would_lose_data);
 
 	return 0;
 }
···
 
 	/* don't handle this yet: */
 	if (flags & BCH_FORCE_IF_METADATA_LOST)
-		return -BCH_ERR_remove_with_metadata_missing_unimplemented;
+		return bch_err_throw(c, remove_with_metadata_missing_unimplemented);
 
 	trans = bch2_trans_get(c);
 	bch2_bkey_buf_init(&k);

fs/bcachefs/move.c

···
 	NULL
 };
 
-static void trace_io_move2(struct bch_fs *c, struct bkey_s_c k,
-			       struct bch_io_opts *io_opts,
-			       struct data_update_opts *data_opts)
-{
-	if (trace_io_move_enabled()) {
-		struct printbuf buf = PRINTBUF;
+struct evacuate_bucket_arg {
+	struct bpos		bucket;
+	int			gen;
+	struct data_update_opts	data_opts;
+};
 
-		bch2_bkey_val_to_text(&buf, c, k);
-		prt_newline(&buf);
-		bch2_data_update_opts_to_text(&buf, c, io_opts, data_opts);
-		trace_io_move(c, buf.buf);
-		printbuf_exit(&buf);
-	}
+static bool evacuate_bucket_pred(struct bch_fs *, void *,
+				 enum btree_id, struct bkey_s_c,
+				 struct bch_io_opts *,
+				 struct data_update_opts *);
+
+static noinline void
+trace_io_move2(struct bch_fs *c, struct bkey_s_c k,
+	       struct bch_io_opts *io_opts,
+	       struct data_update_opts *data_opts)
+{
+	struct printbuf buf = PRINTBUF;
+
+	bch2_bkey_val_to_text(&buf, c, k);
+	prt_newline(&buf);
+	bch2_data_update_opts_to_text(&buf, c, io_opts, data_opts);
+	trace_io_move(c, buf.buf);
+	printbuf_exit(&buf);
 }
 
-static void trace_io_move_read2(struct bch_fs *c, struct bkey_s_c k)
+static noinline void trace_io_move_read2(struct bch_fs *c, struct bkey_s_c k)
 {
-	if (trace_io_move_read_enabled()) {
-		struct printbuf buf = PRINTBUF;
+	struct printbuf buf = PRINTBUF;
 
-		bch2_bkey_val_to_text(&buf, c, k);
-		trace_io_move_read(c, buf.buf);
-		printbuf_exit(&buf);
+	bch2_bkey_val_to_text(&buf, c, k);
+	trace_io_move_read(c, buf.buf);
+	printbuf_exit(&buf);
+}
+
+static noinline void
+trace_io_move_pred2(struct bch_fs *c, struct bkey_s_c k,
+		    struct bch_io_opts *io_opts,
+		    struct data_update_opts *data_opts,
+		    move_pred_fn pred, void *_arg, bool p)
+{
+	struct printbuf buf = PRINTBUF;
+
+	prt_printf(&buf, "%ps: %u", pred, p);
+
+	if (pred == evacuate_bucket_pred) {
+		struct evacuate_bucket_arg *arg = _arg;
+		prt_printf(&buf, " gen=%u", arg->gen);
 	}
+
+	prt_newline(&buf);
+	bch2_bkey_val_to_text(&buf, c, k);
+	prt_newline(&buf);
+	bch2_data_update_opts_to_text(&buf, c, io_opts, data_opts);
+	trace_io_move_pred(c, buf.buf);
+	printbuf_exit(&buf);
+}
+
+static noinline void
+trace_io_move_evacuate_bucket2(struct bch_fs *c, struct bpos bucket, int gen)
+{
+	struct printbuf buf = PRINTBUF;
+
+	prt_printf(&buf, "bucket: ");
+	bch2_bpos_to_text(&buf, bucket);
+	prt_printf(&buf, " gen: %i\n", gen);
+
+	trace_io_move_evacuate_bucket(c, buf.buf);
+	printbuf_exit(&buf);
 }
 
 struct moving_io {
···
 	struct bch_fs *c = trans->c;
 	int ret = -ENOMEM;
 
-	trace_io_move2(c, k, &io_opts, &data_opts);
+	if (trace_io_move_enabled())
+		trace_io_move2(c, k, &io_opts, &data_opts);
 	this_cpu_add(c->counters[BCH_COUNTER_io_move], k.k->size);
 
 	if (ctxt->stats)
···
 		return 0;
 	}
 
-	/*
-	 * Before memory allocations & taking nocow locks in
-	 * bch2_data_update_init():
-	 */
-	bch2_trans_unlock(trans);
-
-	struct moving_io *io = kzalloc(sizeof(struct moving_io), GFP_KERNEL);
+	struct moving_io *io = allocate_dropping_locks(trans, ret,
+				kzalloc(sizeof(struct moving_io), _gfp));
 	if (!io)
 		goto err;
+
+	if (ret)
+		goto err_free;
 
 	INIT_LIST_HEAD(&io->io_list);
 	io->write.ctxt		= ctxt;
···
 
 		io->write.op.c		= c;
 		io->write.data_opts	= data_opts;
+
+		bch2_trans_unlock(trans);
 
 		ret = bch2_data_update_bios_init(&io->write, c, &io_opts);
 		if (ret)
···
 		atomic_inc(&io->b->count);
 	}
 
-	trace_io_move_read2(c, k);
+	if (trace_io_move_read_enabled())
+		trace_io_move_read2(c, k);
 
 	mutex_lock(&ctxt->lock);
 	atomic_add(io->read_sectors, &ctxt->read_sectors);
···
 err_free:
 	kfree(io);
 err:
-	if (bch2_err_matches(ret, BCH_ERR_data_update_done))
-		return 0;
-
 	if (bch2_err_matches(ret, EROFS) ||
 	    bch2_err_matches(ret, BCH_ERR_transaction_restart))
 		return ret;
···
 		trace_io_move_start_fail(c, buf.buf);
 		printbuf_exit(&buf);
 	}
+
+	if (bch2_err_matches(ret, BCH_ERR_data_update_done))
+		return 0;
 	return ret;
 }
 
···
 		bch2_inode_opts_get(io_opts, c, &inode);
 	}
 	bch2_trans_iter_exit(trans, &inode_iter);
+	/* seem to be spinning here? */
 out:
 	return bch2_get_update_rebalance_opts(trans, io_opts, extent_iter, extent_k);
 }
···
 		}
 
 		struct data_update_opts data_opts = {};
-		if (!pred(c, arg, bp.v->btree_id, k, &io_opts, &data_opts)) {
+		bool p = pred(c, arg, bp.v->btree_id, k, &io_opts, &data_opts);
+
+		if (trace_io_move_pred_enabled())
+			trace_io_move_pred2(c, k, &io_opts, &data_opts,
+					    pred, arg, p);
+
+		if (!p) {
 			bch2_trans_iter_exit(trans, &iter);
 			goto next;
 		}
···
 		if (data_opts.scrub &&
 		    !bch2_dev_idx_is_online(c, data_opts.read_dev)) {
 			bch2_trans_iter_exit(trans, &iter);
-			ret = -BCH_ERR_device_offline;
+			ret = bch_err_throw(c, device_offline);
 			break;
 		}
 
···
 	return ret;
 }
 
-struct evacuate_bucket_arg {
-	struct bpos		bucket;
-	int			gen;
-	struct data_update_opts	data_opts;
-};
-
 static bool evacuate_bucket_pred(struct bch_fs *c, void *_arg,
 				 enum btree_id btree, struct bkey_s_c k,
 				 struct bch_io_opts *io_opts,
···
 			 struct bpos bucket, int gen,
 			 struct data_update_opts data_opts)
 {
+	struct bch_fs *c = ctxt->trans->c;
 	struct evacuate_bucket_arg arg = { bucket, gen, data_opts, };
+
+	count_event(c, io_move_evacuate_bucket);
+	if (trace_io_move_evacuate_bucket_enabled())
+		trace_io_move_evacuate_bucket2(c, bucket, gen);
 
 	return __bch2_move_data_phys(ctxt, bucket_in_flight,
 				   bucket.inode,
···
 		? c->opts.metadata_replicas
 		: io_opts->data_replicas;
 
-	rcu_read_lock();
+	guard(rcu)();
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	unsigned i = 0;
 	bkey_for_each_ptr(ptrs, ptr) {
···
 			data_opts->kill_ptrs |= BIT(i);
 		i++;
 	}
-	rcu_read_unlock();
 
 	if (!data_opts->kill_ptrs &&
 	    (!nr_good || nr_good >= replicas))
···
 	struct extent_ptr_decoded p;
 	unsigned i = 0;
 
-	rcu_read_lock();
+	guard(rcu)();
 	bkey_for_each_ptr_decode(k.k, bch2_bkey_ptrs_c(k), p, entry) {
 		unsigned d = bch2_extent_ptr_durability(c, &p);
 
···
 
 		i++;
 	}
-	rcu_read_unlock();
 
 	return data_opts->kill_ptrs != 0;
 }

fs/bcachefs/movinggc.c

···
 {
 	u64 wait = U64_MAX;
 
-	rcu_read_lock();
+	guard(rcu)();
 	for_each_rw_member_rcu(c, ca)
 		wait = min(wait, bch2_copygc_dev_wait_amount(ca));
-	rcu_read_unlock();
-
 	return wait;
 }
 
···
 
 	bch2_printbuf_make_room(out, 4096);
 
-	rcu_read_lock();
+	struct task_struct *t;
 	out->atomic++;
+	scoped_guard(rcu) {
+		prt_printf(out, "Currently calculated wait:\n");
+		for_each_rw_member_rcu(c, ca) {
+			prt_printf(out, "  %s:\t", ca->name);
+			prt_human_readable_u64(out, bch2_copygc_dev_wait_amount(ca));
+			prt_newline(out);
+		}
 
-	prt_printf(out, "Currently calculated wait:\n");
-	for_each_rw_member_rcu(c, ca) {
-		prt_printf(out, "  %s:\t", ca->name);
-		prt_human_readable_u64(out, bch2_copygc_dev_wait_amount(ca));
-		prt_newline(out);
+		t = rcu_dereference(c->copygc_thread);
+		if (t)
+			get_task_struct(t);
 	}
-
-	struct task_struct *t = rcu_dereference(c->copygc_thread);
-	if (t)
-		get_task_struct(t);
 	--out->atomic;
-	rcu_read_unlock();
 
 	if (t) {
 		bch2_prt_task_backtrace(out, t, 0, GFP_KERNEL);

fs/bcachefs/movinggc.h

···
 
 static inline void bch2_copygc_wakeup(struct bch_fs *c)
 {
-	rcu_read_lock();
+	guard(rcu)();
 	struct task_struct *p = rcu_dereference(c->copygc_thread);
 	if (p)
 		wake_up_process(p);
-	rcu_read_unlock();
 }
 
 void bch2_copygc_stop(struct bch_fs *);

fs/bcachefs/namei.c

···
 	}
 
 	if (deleting_subvol && !inode_u->bi_subvol) {
-		ret = -BCH_ERR_ENOENT_not_subvol;
+		ret = bch_err_throw(c, ENOENT_not_subvol);
 		goto err;
 	}
 
···
 	}
 
 	ret = bch2_dirent_rename(trans,
-				 src_dir, &src_hash, &src_dir_u->bi_size,
-				 dst_dir, &dst_hash, &dst_dir_u->bi_size,
+				 src_dir, &src_hash,
+				 dst_dir, &dst_hash,
 				 src_name, &src_inum, &src_offset,
 				 dst_name, &dst_inum, &dst_offset,
 				 mode);
···
 			break;
 
 		if (!inode.bi_dir && !inode.bi_dir_offset) {
-			ret = -BCH_ERR_ENOENT_inode_no_backpointer;
+			ret = bch_err_throw(trans->c, ENOENT_inode_no_backpointer);
 			goto disconnected;
 		}
 
···
 		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_newline(&buf),
-		       bch2_inode_unpacked_to_text(&buf, target),
-		       buf.buf)))
-		goto err;
-
 	struct bkey_s_c_dirent bp_dirent =
 		bch2_bkey_get_iter_typed(trans, &bp_iter, BTREE_ID_dirents,
 			      SPOS(target->bi_dir, target->bi_dir_offset, target->bi_snapshot),
···
 			ret = __bch2_fsck_write_inode(trans, target);
 		}
 	} else {
+		printbuf_reset(&buf);
 		bch2_bkey_val_to_text(&buf, c, d.s_c);
 		prt_newline(&buf);
 		bch2_bkey_val_to_text(&buf, c, bp_dirent.s_c);
···
 			n->v.d_inum = cpu_to_le64(target->bi_inum);
 		}
 
-		ret = bch2_trans_update(trans, dirent_iter, &n->k_i, 0);
+		ret = bch2_trans_update(trans, dirent_iter, &n->k_i,
+					BTREE_UPDATE_internal_snapshot_node);
 		if (ret)
 			goto err;
 	}

fs/bcachefs/printbuf.h

···
 	.size	= _size,				\
 })
 
+static inline struct printbuf bch2_printbuf_init(void)
+{
+	return PRINTBUF;
+}
+
+DEFINE_CLASS(printbuf, struct printbuf,
+	     bch2_printbuf_exit(&_T), bch2_printbuf_init(), void)
+
 /*
  * Returns size remaining of output buffer:
  */

fs/bcachefs/quota.c

···
 	struct bch_sb_field_quota *sb_quota = bch2_sb_get_or_create_quota(&c->disk_sb);
 	if (!sb_quota) {
 		mutex_unlock(&c->sb_lock);
-		return -BCH_ERR_ENOSPC_sb_quota;
+		return bch_err_throw(c, ENOSPC_sb_quota);
 	}
 
 	bch2_sb_quota_read(c);
···
 	mutex_lock(&c->sb_lock);
 	sb_quota = bch2_sb_get_or_create_quota(&c->disk_sb);
 	if (!sb_quota) {
-		ret = -BCH_ERR_ENOSPC_sb_quota;
+		ret = bch_err_throw(c, ENOSPC_sb_quota);
 		goto unlock;
 	}
 
···
 	mutex_lock(&c->sb_lock);
 	sb_quota = bch2_sb_get_or_create_quota(&c->disk_sb);
 	if (!sb_quota) {
-		ret = -BCH_ERR_ENOSPC_sb_quota;
+		ret = bch_err_throw(c, ENOSPC_sb_quota);
 		goto unlock;
 	}
 

fs/bcachefs/rebalance.c

···
 	unsigned ptr_bit = 1;
 	unsigned rewrite_ptrs = 0;
 
-	rcu_read_lock();
+	guard(rcu)();
 	bkey_for_each_ptr(ptrs, ptr) {
 		if (!ptr->cached && !bch2_dev_in_target(c, ptr->dev, opts->background_target))
 			rewrite_ptrs |= ptr_bit;
 		ptr_bit <<= 1;
 	}
-	rcu_read_unlock();
 
 	return rewrite_ptrs;
 }
···
 	}
 incompressible:
 	if (opts->background_target) {
-		rcu_read_lock();
+		guard(rcu)();
 		bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 			if (!p.ptr.cached &&
 			    !bch2_dev_in_target(c, p.ptr.dev, opts->background_target))
 				sectors += p.crc.compressed_size;
-		rcu_read_unlock();
 	}
 
 	return sectors;
···
 		if (bch2_err_matches(ret, ENOMEM)) {
 			/* memory allocation failure, wait for some IO to finish */
 			bch2_move_ctxt_wait_for_io(ctxt);
-			ret = -BCH_ERR_transaction_restart_nested;
+			ret = bch_err_throw(c, transaction_restart_nested);
 		}
 
 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
···
 		r->state		= BCH_REBALANCE_waiting;
 	}
 
-	bch2_kthread_io_clock_wait(clock, r->wait_iotime_end, MAX_SCHEDULE_TIMEOUT);
+	bch2_kthread_io_clock_wait_once(clock, r->wait_iotime_end, MAX_SCHEDULE_TIMEOUT);
 }
 
 static bool bch2_rebalance_enabled(struct bch_fs *c)
···
 	struct bch_fs_rebalance *r = &c->rebalance;
 	struct btree_iter rebalance_work_iter, extent_iter = {};
 	struct bkey_s_c k;
+	u32 kick = r->kick;
 	int ret = 0;
 
 	bch2_trans_begin(trans);
···
 	if (!ret &&
 	    !kthread_should_stop() &&
 	    !atomic64_read(&r->work_stats.sectors_seen) &&
-	    !atomic64_read(&r->scan_stats.sectors_seen)) {
+	    !atomic64_read(&r->scan_stats.sectors_seen) &&
+	    kick == r->kick) {
 		bch2_moving_ctxt_flush_all(ctxt);
 		bch2_trans_unlock_long(trans);
 		rebalance_wait(c);
···
 	}
 	prt_newline(out);
 
-	rcu_read_lock();
-	struct task_struct *t = rcu_dereference(c->rebalance.thread);
-	if (t)
-		get_task_struct(t);
-	rcu_read_unlock();
+	struct task_struct *t;
+	scoped_guard(rcu) {
+		t = rcu_dereference(c->rebalance.thread);
+		if (t)
+			get_task_struct(t);
+	}
 
 	if (t) {
 		bch2_prt_task_backtrace(out, t, 0, GFP_KERNEL);
···
 				     BTREE_ID_extents, POS_MIN,
 				     BTREE_ITER_prefetch|
 				     BTREE_ITER_all_snapshots);
-		return -BCH_ERR_transaction_restart_nested;
+		return bch_err_throw(c, transaction_restart_nested);
 	}
 
 	if (!extent_k.k && !rebalance_k.k)

fs/bcachefs/rebalance.h

···
 
 static inline void bch2_rebalance_wakeup(struct bch_fs *c)
 {
-	struct task_struct *p;
-
-	rcu_read_lock();
-	p = rcu_dereference(c->rebalance.thread);
+	c->rebalance.kick++;
+	guard(rcu)();
+	struct task_struct *p = rcu_dereference(c->rebalance.thread);
 	if (p)
 		wake_up_process(p);
-	rcu_read_unlock();
 }
 
 void bch2_rebalance_status_to_text(struct printbuf *, struct bch_fs *);

fs/bcachefs/rebalance_types.h

···
 
 struct bch_fs_rebalance {
 	struct task_struct __rcu	*thread;
+	u32				kick;
 	struct bch_pd_controller pd;
 
 	enum bch_rebalance_states	state;

fs/bcachefs/recovery.c

···
 use_clean:
 		if (!clean) {
 			bch_err(c, "no superblock clean section found");
-			ret = -BCH_ERR_fsck_repair_impossible;
+			ret = bch_err_throw(c, fsck_repair_impossible);
 			goto err;
 
 		}
···
 out:
 	bch2_flush_fsck_errs(c);
 
-	if (!c->opts.retain_recovery_info) {
-		bch2_journal_keys_put_initial(c);
-		bch2_find_btree_nodes_exit(&c->found_btree_nodes);
-	}
 	if (!IS_ERR(clean))
 		kfree(clean);
 

fs/bcachefs/recovery_passes.c

···
 		prt_tab(out);
 
 		bch2_pr_time_units(out, le32_to_cpu(i->last_runtime) * NSEC_PER_SEC);
+
+		if (BCH_RECOVERY_PASS_NO_RATELIMIT(i))
+			prt_str(out, " (no ratelimit)");
+
 		prt_newline(out);
 	}
 }
 
-static void bch2_sb_recovery_pass_complete(struct bch_fs *c,
-					   enum bch_recovery_pass pass,
-					   s64 start_time)
+static struct recovery_pass_entry *bch2_sb_recovery_pass_entry(struct bch_fs *c,
+							       enum bch_recovery_pass pass)
 {
 	enum bch_recovery_pass_stable stable = bch2_recovery_pass_to_stable(pass);
-	s64 end_time = ktime_get_real_seconds();
 
-	mutex_lock(&c->sb_lock);
-	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
-	__clear_bit_le64(stable, ext->recovery_passes_required);
+	lockdep_assert_held(&c->sb_lock);
 
 	struct bch_sb_field_recovery_passes *r =
 		bch2_sb_field_get(c->disk_sb.sb, recovery_passes);
···
 		r = bch2_sb_field_resize(&c->disk_sb, recovery_passes, u64s);
 		if (!r) {
 			bch_err(c, "error creating recovery_passes sb section");
-			goto out;
+			return NULL;
 		}
 	}
 
-	r->start[stable].last_run	= cpu_to_le64(end_time);
-	r->start[stable].last_runtime	= cpu_to_le32(max(0, end_time - start_time));
-out:
+	return r->start + stable;
+}
+
+static void bch2_sb_recovery_pass_complete(struct bch_fs *c,
+					   enum bch_recovery_pass pass,
+					   s64 start_time)
+{
+	guard(mutex)(&c->sb_lock);
+	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
+	__clear_bit_le64(bch2_recovery_pass_to_stable(pass),
+			 ext->recovery_passes_required);
+
+	struct recovery_pass_entry *e = bch2_sb_recovery_pass_entry(c, pass);
+	if (e) {
+		s64 end_time	= ktime_get_real_seconds();
+		e->last_run	= cpu_to_le64(end_time);
+		e->last_runtime	= cpu_to_le32(max(0, end_time - start_time));
+		SET_BCH_RECOVERY_PASS_NO_RATELIMIT(e, false);
+	}
+
 	bch2_write_super(c);
-	mutex_unlock(&c->sb_lock);
+}
+
+void bch2_recovery_pass_set_no_ratelimit(struct bch_fs *c,
+					 enum bch_recovery_pass pass)
+{
+	guard(mutex)(&c->sb_lock);
+
+	struct recovery_pass_entry *e = bch2_sb_recovery_pass_entry(c, pass);
+	if (e && !BCH_RECOVERY_PASS_NO_RATELIMIT(e)) {
+		SET_BCH_RECOVERY_PASS_NO_RATELIMIT(e, false);
+		bch2_write_super(c);
+	}
 }
 
 static bool bch2_recovery_pass_want_ratelimit(struct bch_fs *c, enum bch_recovery_pass pass)
···
 		 */
 		ret = (u64) le32_to_cpu(i->last_runtime) * 100 >
 			ktime_get_real_seconds() - le64_to_cpu(i->last_run);
+
+		if (BCH_RECOVERY_PASS_NO_RATELIMIT(i))
+			ret = false;
 	}
 
 	return ret;
···
 		goto out;
 
 	bool in_recovery = test_bit(BCH_FS_in_recovery, &c->flags);
-	bool rewind = in_recovery && r->curr_pass > pass;
+	bool rewind = in_recovery &&
+		r->curr_pass > pass &&
+		!(r->passes_complete & BIT_ULL(pass));
 	bool ratelimit = flags & RUN_RECOVERY_PASS_ratelimit;
 
 	if (!(in_recovery && (flags & RUN_RECOVERY_PASS_nopersistent))) {
···
 	    (!in_recovery || r->curr_pass >= BCH_RECOVERY_PASS_set_may_go_rw)) {
 		prt_printf(out, "need recovery pass %s (%u), but already rw\n",
 			   bch2_recovery_passes[pass], pass);
-		ret = -BCH_ERR_cannot_rewind_recovery;
+		ret = bch_err_throw(c, cannot_rewind_recovery);
 		goto out;
 	}
 
···
 		if (rewind) {
 			r->next_pass = pass;
 			r->passes_complete &= (1ULL << pass) >> 1;
-			ret = -BCH_ERR_restart_recovery;
+			ret = bch_err_throw(c, restart_recovery);
 		}
 	} else {
 		prt_printf(out, "scheduling recovery pass %s (%u)%s\n",
···
 	}
 
 	return ret;
+}
+
+/*
+ * Returns 0 if @pass has run recently, otherwise one of
+ * -BCH_ERR_restart_recovery
+ * -BCH_ERR_recovery_pass_will_run
+ */
+int bch2_require_recovery_pass(struct bch_fs *c,
+			       struct printbuf *out,
+			       enum bch_recovery_pass pass)
+{
+	if (test_bit(BCH_FS_in_recovery, &c->flags) &&
+	    c->recovery.passes_complete & BIT_ULL(pass))
+		return 0;
+
+	guard(mutex)(&c->sb_lock);
+
+	if (bch2_recovery_pass_want_ratelimit(c, pass))
+		return 0;
+
+	enum bch_run_recovery_pass_flags flags = 0;
+	int ret = 0;
+
+	if (recovery_pass_needs_set(c, pass, &flags)) {
+		ret = __bch2_run_explicit_recovery_pass(c, out, pass, flags);
+		bch2_write_super(c);
+	}
+
+	return ret ?: bch_err_throw(c, recovery_pass_will_run);
 }
 
 int bch2_run_print_explicit_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)

fs/bcachefs/recovery_passes.h

···
 
 u64 bch2_fsck_recovery_passes(void);
 
+void bch2_recovery_pass_set_no_ratelimit(struct bch_fs *, enum bch_recovery_pass);
+
 enum bch_run_recovery_pass_flags {
 	RUN_RECOVERY_PASS_nopersistent	= BIT(0),
 	RUN_RECOVERY_PASS_ratelimit	= BIT(1),
···
 int bch2_run_explicit_recovery_pass(struct bch_fs *, struct printbuf *,
 				    enum bch_recovery_pass,
 				    enum bch_run_recovery_pass_flags);
+
+int bch2_require_recovery_pass(struct bch_fs *, struct printbuf *,
+			       enum bch_recovery_pass);
 
 int bch2_run_online_recovery_passes(struct bch_fs *, u64);
 int bch2_run_recovery_passes(struct bch_fs *, enum bch_recovery_pass);

fs/bcachefs/recovery_passes_format.h

···
 	__le32			flags;
 };
 
+LE32_BITMASK(BCH_RECOVERY_PASS_NO_RATELIMIT,	struct recovery_pass_entry, flags, 0, 1)
+
 struct bch_sb_field_recovery_passes {
 	struct bch_sb_field	field;
 	struct recovery_pass_entry start[];

fs/bcachefs/reflink.c

···
 
 	if (!bkey_refcount_c(k)) {
 		if (!(flags & BTREE_TRIGGER_overwrite))
-			ret = -BCH_ERR_missing_indirect_extent;
+			ret = bch_err_throw(c, missing_indirect_extent);
 		goto next;
 	}
 
···
 	int ret = 0, ret2 = 0;
 
 	if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_reflink))
-		return -BCH_ERR_erofs_no_writes;
+		return bch_err_throw(c, erofs_no_writes);
 
 	bch2_check_set_feature(c, BCH_FEATURE_reflink);
 
···
 			SET_REFLINK_P_IDX(&dst_p->v, offset);
 
 			if (reflink_p_may_update_opts_field &&
-			    may_change_src_io_path_opts)
+			    may_change_src_io_path_opts &&
+			    REFLINK_P_MAY_UPDATE_OPTIONS(src_p.v))
 				SET_REFLINK_P_MAY_UPDATE_OPTIONS(&dst_p->v, true);
 		} else {
 			BUG();
···
 			struct reflink_gc *r = genradix_ptr_alloc(&c->reflink_gc_table,
 							c->reflink_gc_nr++, GFP_KERNEL);
 			if (!r) {
-				ret = -BCH_ERR_ENOMEM_gc_reflink_start;
+				ret = bch_err_throw(c, ENOMEM_gc_reflink_start);
 				break;
 			}
 

fs/bcachefs/replicas.c

···
 	return 0;
 bad:
 	bch2_replicas_entry_to_text(err, r);
-	return -BCH_ERR_invalid_replicas_entry;
+	return bch_err_throw(c, invalid_replicas_entry);
 }
 
 void bch2_cpu_replicas_to_text(struct printbuf *out,
···
 	    !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 		new_gc = cpu_replicas_add_entry(c, &c->replicas_gc, new_entry);
 		if (!new_gc.entries) {
-			ret = -BCH_ERR_ENOMEM_cpu_replicas;
+			ret = bch_err_throw(c, ENOMEM_cpu_replicas);
 			goto err;
 		}
 	}
···
 	if (!__replicas_has_entry(&c->replicas, new_entry)) {
 		new_r = cpu_replicas_add_entry(c, &c->replicas, new_entry);
 		if (!new_r.entries) {
-			ret = -BCH_ERR_ENOMEM_cpu_replicas;
+			ret = bch_err_throw(c, ENOMEM_cpu_replicas);
 			goto err;
 		}
 
···
 	if (!c->replicas_gc.entries) {
 		mutex_unlock(&c->sb_lock);
 		bch_err(c, "error allocating c->replicas_gc");
-		return -BCH_ERR_ENOMEM_replicas_gc;
+		return bch_err_throw(c, ENOMEM_replicas_gc);
 	}
 
 	for_each_cpu_replicas_entry(&c->replicas, e)
···
 	new.entries	= kcalloc(nr, new.entry_size, GFP_KERNEL);
 	if (!new.entries) {
 		bch_err(c, "error allocating c->replicas_gc");
-		return -BCH_ERR_ENOMEM_replicas_gc;
+		return bch_err_throw(c, ENOMEM_replicas_gc);
 	}
 
 	mutex_lock(&c->sb_lock);
···
 	sb_r = bch2_sb_field_resize(&c->disk_sb, replicas_v0,
 			DIV_ROUND_UP(bytes, sizeof(u64)));
 	if (!sb_r)
-		return -BCH_ERR_ENOSPC_sb_replicas;
+		return bch_err_throw(c, ENOSPC_sb_replicas);
 
 	bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 	sb_r = bch2_sb_field_get(c->disk_sb.sb, replicas_v0);
···
 	sb_r = bch2_sb_field_resize(&c->disk_sb, replicas,
 			DIV_ROUND_UP(bytes, sizeof(u64)));
 	if (!sb_r)
-		return -BCH_ERR_ENOSPC_sb_replicas;
+		return bch_err_throw(c, ENOSPC_sb_replicas);
 
 	bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 	sb_r = bch2_sb_field_get(c->disk_sb.sb, replicas);
···
 		if (e->data_type == BCH_DATA_cached)
 			continue;
 
-		rcu_read_lock();
-		for (unsigned i = 0; i < e->nr_devs; i++) {
-			if (e->devs[i] == BCH_SB_MEMBER_INVALID) {
-				nr_failed++;
-				continue;
+		scoped_guard(rcu)
+			for (unsigned i = 0; i < e->nr_devs; i++) {
+				if (e->devs[i] == BCH_SB_MEMBER_INVALID) {
+					nr_failed++;
+					continue;
+				}
+
+				nr_online += test_bit(e->devs[i], devs.d);
+
+				struct bch_dev *ca = bch2_dev_rcu_noerror(c, e->devs[i]);
+				nr_failed += !ca || ca->mi.state == BCH_MEMBER_STATE_failed;
 			}
-
-			nr_online += test_bit(e->devs[i], devs.d);
-
-			struct bch_dev *ca = bch2_dev_rcu_noerror(c, e->devs[i]);
-			nr_failed += !ca || ca->mi.state == BCH_MEMBER_STATE_failed;
-		}
-		rcu_read_unlock();
 
 		if (nr_online + nr_failed == e->nr_devs)
 			continue;

fs/bcachefs/sb-counters_format.h

···
 	x(io_move_write_fail,				82,	TYPE_COUNTER)	\
 	x(io_move_start_fail,				39,	TYPE_COUNTER)	\
 	x(io_move_created_rebalance,			83,	TYPE_COUNTER)	\
+	x(io_move_evacuate_bucket,			84,	TYPE_COUNTER)	\
 	x(bucket_invalidate,				3,	TYPE_COUNTER)	\
 	x(bucket_discard,				4,	TYPE_COUNTER)	\
 	x(bucket_discard_fast,				79,	TYPE_COUNTER)	\

fs/bcachefs/sb-downgrade.c

···
 
 	d = bch2_sb_field_resize(&c->disk_sb, downgrade, sb_u64s);
 	if (!d) {
-		ret = -BCH_ERR_ENOSPC_sb_downgrade;
+		ret = bch_err_throw(c, ENOSPC_sb_downgrade);
 		goto out;
 	}
 

fs/bcachefs/sb-errors.c

···
 	.to_text	= bch2_sb_errors_to_text,
 };
 
+void bch2_fs_errors_to_text(struct printbuf *out, struct bch_fs *c)
+{
+	if (out->nr_tabstops < 1)
+		printbuf_tabstop_push(out, 48);
+	if (out->nr_tabstops < 2)
+		printbuf_tabstop_push(out, 8);
+	if (out->nr_tabstops < 3)
+		printbuf_tabstop_push(out, 16);
+
+	guard(mutex)(&c->fsck_error_counts_lock);
+
+	bch_sb_errors_cpu *e = &c->fsck_error_counts;
+	darray_for_each(*e, i) {
+		bch2_sb_error_id_to_text(out, i->id);
+		prt_tab(out);
+		prt_u64(out, i->nr);
+		prt_tab(out);
+		bch2_prt_datetime(out, i->last_error_time);
+		prt_newline(out);
+	}
+}
+
 void bch2_sb_error_count(struct bch_fs *c, enum bch_sb_error_id err)
 {
 	bch_sb_errors_cpu *e = &c->fsck_error_counts;

fs/bcachefs/sb-errors.h

···
 extern const char * const bch2_sb_error_strs[];
 
 void bch2_sb_error_id_to_text(struct printbuf *, enum bch_sb_error_id);
+void bch2_fs_errors_to_text(struct printbuf *, struct bch_fs *);
 
 extern const struct bch_sb_field_ops bch_sb_field_ops_errors;
 

fs/bcachefs/sb-errors_format.h

···
 	x(inode_dir_multiple_links,				206,	FSCK_AUTOFIX)	\
 	x(inode_dir_missing_backpointer,			284,	FSCK_AUTOFIX)	\
 	x(inode_dir_unlinked_but_not_empty,			286,	FSCK_AUTOFIX)	\
+	x(inode_dir_has_nonzero_i_size,				319,	FSCK_AUTOFIX)	\
 	x(inode_multiple_links_but_nlink_0,			207,	FSCK_AUTOFIX)	\
 	x(inode_wrong_backpointer,				208,	FSCK_AUTOFIX)	\
 	x(inode_wrong_nlink,					209,	FSCK_AUTOFIX)	\
···
 	x(inode_parent_has_case_insensitive_not_set,		317,	FSCK_AUTOFIX)	\
 	x(vfs_inode_i_blocks_underflow,				311,	FSCK_AUTOFIX)	\
 	x(vfs_inode_i_blocks_not_zero_at_truncate,		313,	FSCK_AUTOFIX)	\
+	x(vfs_bad_inode_rm,					320,	0)		\
 	x(deleted_inode_but_clean,				211,	FSCK_AUTOFIX)	\
 	x(deleted_inode_missing,				212,	FSCK_AUTOFIX)	\
 	x(deleted_inode_is_dir,					213,	FSCK_AUTOFIX)	\
···
 	x(dirent_stray_data_after_cf_name,			305,	0)		\
 	x(rebalance_work_incorrectly_set,			309,	FSCK_AUTOFIX)	\
 	x(rebalance_work_incorrectly_unset,			310,	FSCK_AUTOFIX)	\
-	x(MAX,							319,	0)
+	x(MAX,							321,	0)
 
 enum bch_sb_error_id {
 #define x(t, n, ...) BCH_FSCK_ERR_##t = n,

fs/bcachefs/sb-members.c

···
 
 		mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
 		if (!mi)
-			return -BCH_ERR_ENOSPC_sb_members_v2;
+			return bch_err_throw(c, ENOSPC_sb_members_v2);
 
 		for (int i = c->disk_sb.sb->nr_devices - 1; i >= 0; --i) {
 			void *dst = (void *) mi->_members + (i * sizeof(struct bch_member));
···
 {
 	struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
 
-	rcu_read_lock();
+	guard(rcu)();
 	for_each_member_device_rcu(c, ca, NULL) {
 		struct bch_member *m = __bch2_members_v2_get_mut(mi, ca->dev_idx);
 
 		for (unsigned e = 0; e < BCH_MEMBER_ERROR_NR; e++)
 			m->errors[e] = cpu_to_le64(atomic64_read(&ca->errors[e]));
 	}
-	rcu_read_unlock();
 }
 
 void bch2_dev_io_errors_to_text(struct printbuf *out, struct bch_dev *ca)
···
 
 bool bch2_dev_btree_bitmap_marked(struct bch_fs *c, struct bkey_s_c k)
 {
-	bool ret = true;
-	rcu_read_lock();
+	guard(rcu)();
 	bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
 		struct bch_dev *ca = bch2_dev_rcu(c, ptr->dev);
-		if (!ca)
-			continue;
-
-		if (!bch2_dev_btree_bitmap_marked_sectors(ca, ptr->offset, btree_sectors(c))) {
-			ret = false;
-			break;
-		}
+		if (ca &&
+		    !bch2_dev_btree_bitmap_marked_sectors(ca, ptr->offset, btree_sectors(c)))
+			return false;
 	}
-	rcu_read_unlock();
-	return ret;
+	return true;
 }
 
 static void __bch2_dev_btree_bitmap_mark(struct bch_sb_field_members_v2 *mi, unsigned dev,

fs/bcachefs/sb-members.h

···
 
 static inline bool bch2_dev_idx_is_online(struct bch_fs *c, unsigned dev)
 {
-	rcu_read_lock();
+	guard(rcu)();
 	struct bch_dev *ca = bch2_dev_rcu(c, dev);
-	bool ret = ca && bch2_dev_is_online(ca);
-	rcu_read_unlock();
-
-	return ret;
+	return ca && bch2_dev_is_online(ca);
 }
 
 static inline bool bch2_dev_is_healthy(struct bch_dev *ca)
···
 
 static inline struct bch_dev *bch2_get_next_dev(struct bch_fs *c, struct bch_dev *ca)
 {
-	rcu_read_lock();
+	guard(rcu)();
 	bch2_dev_put(ca);
 	if ((ca = __bch2_next_dev(c, ca, NULL)))
 		bch2_dev_get(ca);
-	rcu_read_unlock();
-
 	return ca;
 }
 
···
 						       unsigned state_mask,
 						       int rw, unsigned ref_idx)
 {
-	rcu_read_lock();
+	guard(rcu)();
 	if (ca)
 		enumerated_ref_put(&ca->io_ref[rw], ref_idx);
 
···
 	       (!((1 << ca->mi.state) & state_mask) ||
 		!enumerated_ref_tryget(&ca->io_ref[rw], ref_idx)))
 		;
-	rcu_read_unlock();
 
 	return ca;
 }
···
 
 static inline struct bch_dev *bch2_dev_tryget_noerror(struct bch_fs *c, unsigned dev)
 {
-	rcu_read_lock();
+	guard(rcu)();
 	struct bch_dev *ca = bch2_dev_rcu_noerror(c, dev);
 	if (ca)
 		bch2_dev_get(ca);
-	rcu_read_unlock();
 	return ca;
 }
 
···
 {
 	might_sleep();
 
-	rcu_read_lock();
+	guard(rcu)();
 	struct bch_dev *ca = bch2_dev_rcu(c, dev);
-	if (ca && !enumerated_ref_tryget(&ca->io_ref[rw], ref_idx))
-		ca = NULL;
-	rcu_read_unlock();
+	if (!ca || !enumerated_ref_tryget(&ca->io_ref[rw], ref_idx))
+		return NULL;
 
-	if (ca &&
-	    (ca->mi.state == BCH_MEMBER_STATE_rw ||
-	    (ca->mi.state == BCH_MEMBER_STATE_ro && rw == READ)))
+	if (ca->mi.state == BCH_MEMBER_STATE_rw ||
+	    (ca->mi.state == BCH_MEMBER_STATE_ro && rw == READ))
 		return ca;
 
-	if (ca)
-		enumerated_ref_put(&ca->io_ref[rw], ref_idx);
+	enumerated_ref_put(&ca->io_ref[rw], ref_idx);
 	return NULL;
 }
 

fs/bcachefs/six.c

···
 	 * acquiring the lock and setting the owner field. If we're an RT task
 	 * that will live-lock because we won't let the owner complete.
 	 */
-	rcu_read_lock();
+	guard(rcu)();
 	struct task_struct *owner = READ_ONCE(lock->owner);
-	bool ret = owner ? owner_on_cpu(owner) : !rt_or_dl_task(current);
-	rcu_read_unlock();
-
-	return ret;
+	return owner ? owner_on_cpu(owner) : !rt_or_dl_task(current);
 }
 
 static inline bool six_optimistic_spin(struct six_lock *lock,

fs/bcachefs/snapshot.c

···
 					  BTREE_ITER_with_updates, snapshot_tree, s);
 
 	if (bch2_err_matches(ret, ENOENT))
-		ret = -BCH_ERR_ENOENT_snapshot_tree;
+		ret = bch_err_throw(trans->c, ENOENT_snapshot_tree);
 	return ret;
 }
 
···
 	struct bkey_i_snapshot_tree *s_t;
 
 	if (ret == -BCH_ERR_ENOSPC_btree_slot)
-		ret = -BCH_ERR_ENOSPC_snapshot_tree;
+		ret = bch_err_throw(trans->c, ENOSPC_snapshot_tree);
 	if (ret)
 		return ERR_PTR(ret);
 
···
 
 static bool bch2_snapshot_is_ancestor_early(struct bch_fs *c, u32 id, u32 ancestor)
 {
-	rcu_read_lock();
-	bool ret = __bch2_snapshot_is_ancestor_early(rcu_dereference(c->snapshots), id, ancestor);
-	rcu_read_unlock();
-
-	return ret;
+	guard(rcu)();
+	return __bch2_snapshot_is_ancestor_early(rcu_dereference(c->snapshots), id, ancestor);
 }
 
 static inline u32 get_ancestor_below(struct snapshot_table *t, u32 id, u32 ancestor)
···
 {
 	bool ret;
 
-	rcu_read_lock();
+	guard(rcu)();
 	struct snapshot_table *t = rcu_dereference(c->snapshots);
 
-	if (unlikely(c->recovery.pass_done < BCH_RECOVERY_PASS_check_snapshots)) {
-		ret = __bch2_snapshot_is_ancestor_early(t, id, ancestor);
-		goto out;
-	}
+	if (unlikely(c->recovery.pass_done < BCH_RECOVERY_PASS_check_snapshots))
+		return __bch2_snapshot_is_ancestor_early(t, id, ancestor);
 
 	if (likely(ancestor >= IS_ANCESTOR_BITMAP))
 		while (id && id < ancestor - IS_ANCESTOR_BITMAP)
···
 		: id == ancestor;
 
 	EBUG_ON(ret != __bch2_snapshot_is_ancestor_early(t, id, ancestor));
-out:
-	rcu_read_unlock();
-
 	return ret;
 }
 
···
 	mutex_lock(&c->snapshot_table_lock);
 	int ret = snapshot_t_mut(c, id)
 		? 0
-		: -BCH_ERR_ENOMEM_mark_snapshot;
+		: bch_err_throw(c, ENOMEM_mark_snapshot);
 	mutex_unlock(&c->snapshot_table_lock);
 	return ret;
 }
···
 
 	t = snapshot_t_mut(c, id);
 	if (!t) {
-		ret = -BCH_ERR_ENOMEM_mark_snapshot;
+		ret = bch_err_throw(c, ENOMEM_mark_snapshot);
 		goto err;
 	}
 
···
 u32 bch2_snapshot_oldest_subvol(struct bch_fs *c, u32 snapshot_root,
 				snapshot_id_list *skip)
 {
+	guard(rcu)();
 	u32 id, subvol = 0, s;
 retry:
 	id = snapshot_root;
-	rcu_read_lock();
 	while (id && bch2_snapshot_exists(c, id)) {
 		if (!(skip && snapshot_list_has_id(skip, id))) {
 			s = snapshot_t(c, id)->subvol;
···
 		if (id == snapshot_root)
 			break;
 	}
-	rcu_read_unlock();
 
 	if (!subvol && skip) {
 		skip = NULL;
···
 
 u32 bch2_snapshot_skiplist_get(struct bch_fs *c, u32 id)
 {
-	const struct snapshot_t *s;
-
 	if (!id)
 		return 0;
 
-	rcu_read_lock();
-	s = snapshot_t(c, id);
-	if (s->parent)
-		id = bch2_snapshot_nth_parent(c, id, get_random_u32_below(s->depth));
-	rcu_read_unlock();
-
-	return id;
+	guard(rcu)();
+	const struct snapshot_t *s = snapshot_t(c, id);
+	return s->parent
+		? bch2_snapshot_nth_parent(c, id, get_random_u32_below(s->depth))
+		: id;
 }
 
 static int snapshot_skiplist_good(struct btree_trans *trans, u32 id, struct bch_snapshot s)
···
 
 static inline bool snapshot_id_lists_have_common(snapshot_id_list *l, snapshot_id_list *r)
 {
-	darray_for_each(*l, i)
-		if (snapshot_list_has_id(r, *i))
-			return true;
-	return false;
+	return darray_find_p(*l, i, snapshot_list_has_id(r, *i)) != NULL;
 }
 
 static void snapshot_id_list_to_text(struct printbuf *out, snapshot_id_list *s)
···
 					"snapshot node %u from tree %s missing, recreate?", *id, buf.buf)) {
 				if (t->nr > 1) {
 					bch_err(c, "cannot reconstruct snapshot trees with multiple nodes");
-					ret = -BCH_ERR_fsck_repair_unimplemented;
+					ret = bch_err_throw(c, fsck_repair_unimplemented);
 					goto err;
 				}
 
···
 		ret = bch2_btree_delete_at(trans, iter,
 					   BTREE_UPDATE_internal_snapshot_node) ?: 1;
 
-	/*
-	 * Snapshot missing: we should have caught this with btree_lost_data and
-	 * kicked off reconstruct_snapshots, so if we end up here we have no
-	 * idea what happened:
-	 */
-	if (fsck_err_on(state == SNAPSHOT_ID_empty,
-			trans, bkey_in_missing_snapshot,
-			"key in missing snapshot %s, delete?",
-			(bch2_btree_id_to_text(&buf, iter->btree_id),
-			 prt_char(&buf, ' '),
-			 bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
-		ret = bch2_btree_delete_at(trans, iter,
-					   BTREE_UPDATE_internal_snapshot_node) ?: 1;
+	if (state == SNAPSHOT_ID_empty) {
+		/*
+		 * Snapshot missing: we should have caught this with btree_lost_data and
+		 * kicked off reconstruct_snapshots, so if we end up here we have no
+		 * idea what happened.
+		 *
+		 * Do not delete unless we know that subvolumes and snapshots
+		 * are consistent:
+		 *
+		 * XXX:
+		 *
+		 * We could be smarter here, and instead of using the generic
+		 * recovery pass ratelimiting, track if there have been any
+		 * changes to the snapshots or inodes btrees since those passes
+		 * last ran.
+		 */
+		ret = bch2_require_recovery_pass(c, &buf, BCH_RECOVERY_PASS_check_snapshots) ?: ret;
+		ret = bch2_require_recovery_pass(c, &buf, BCH_RECOVERY_PASS_check_subvols) ?: ret;
+
+		if (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_snapshots))
+			ret = bch2_require_recovery_pass(c, &buf, BCH_RECOVERY_PASS_reconstruct_snapshots) ?: ret;
+
+		unsigned repair_flags = FSCK_CAN_IGNORE | (!ret ? FSCK_CAN_FIX : 0);
+
+		if (__fsck_err(trans, repair_flags, bkey_in_missing_snapshot,
+			     "key in missing snapshot %s, delete?",
+			     (bch2_btree_id_to_text(&buf, iter->btree_id),
+			      prt_char(&buf, ' '),
+			      bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+			ret = bch2_btree_delete_at(trans, iter,
+						   BTREE_UPDATE_internal_snapshot_node) ?: 1;
+		}
+	}
 fsck_err:
 	printbuf_exit(&buf);
+	return ret;
+}
+
+int __bch2_get_snapshot_overwrites(struct btree_trans *trans,
+				   enum btree_id btree, struct bpos pos,
+				   snapshot_id_list *s)
+{
+	struct bch_fs *c = trans->c;
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	int ret = 0;
+
+	for_each_btree_key_reverse_norestart(trans, iter, btree, bpos_predecessor(pos),
+					     BTREE_ITER_all_snapshots, k, ret) {
+		if (!bkey_eq(k.k->p, pos))
+			break;
+
+		if (!bch2_snapshot_is_ancestor(c, k.k->p.snapshot, pos.snapshot) ||
+		    snapshot_list_has_ancestor(c, s, k.k->p.snapshot))
+			continue;
+
+		ret = snapshot_list_add(c, s, k.k->p.snapshot);
+		if (ret)
+			break;
+	}
+	bch2_trans_iter_exit(trans, &iter);
+	if (ret)
+		darray_exit(s);
+
 	return ret;
 }
 
···
 			goto err;
 
 		if (!k.k || !k.k->p.offset) {
-			ret = -BCH_ERR_ENOSPC_snapshot_create;
+			ret = bch_err_throw(c, ENOSPC_snapshot_create);
 			goto err;
 		}
 
···
 
 static inline u32 interior_delete_has_id(interior_delete_list *l, u32 id)
 {
-	darray_for_each(*l, i)
-		if (i->id == id)
-			return i->live_child;
-	return 0;
+	struct snapshot_interior_delete *i = darray_find_p(*l, i, i->id == id);
+	return i ? i->live_child : 0;
 }
 
 static unsigned __live_child(struct snapshot_table *t, u32 id,
···
 {
 	struct snapshot_delete *d = &c->snapshot_delete;
 
-	rcu_read_lock();
-	u32 ret = __live_child(rcu_dereference(c->snapshots), id,
-			       &d->delete_leaves, &d->delete_interior);
-	rcu_read_unlock();
-	return ret;
+	guard(rcu)();
+	return __live_child(rcu_dereference(c->snapshots), id,
+			    &d->delete_leaves, &d->delete_interior);
 }
 
 static bool snapshot_id_dying(struct snapshot_delete *d, unsigned id)
···
 static inline u32 bch2_snapshot_nth_parent_skip(struct bch_fs *c, u32 id, u32 n,
 						interior_delete_list *skip)
 {
-	rcu_read_lock();
+	guard(rcu)();
 	while (interior_delete_has_id(skip, id))
 		id = __bch2_snapshot_parent(c, id);
 
···
 			id = __bch2_snapshot_parent(c, id);
 		} while (interior_delete_has_id(skip, id));
 	}
-	rcu_read_unlock();
 
 	return id;
 }
···
 	d->running = false;
 	mutex_unlock(&d->progress_lock);
 	bch2_trans_put(trans);
+
+	bch2_recovery_pass_set_no_ratelimit(c, BCH_RECOVERY_PASS_check_snapshots);
 out_unlock:
 	mutex_unlock(&d->lock);
 	if (!bch2_err_matches(ret, EROFS))
···
 
 	BUG_ON(!test_bit(BCH_FS_may_go_rw, &c->flags));
 
-	if (!queue_work(c->write_ref_wq, &c->snapshot_delete.work))
+	if (!queue_work(system_long_wq, &c->snapshot_delete.work))
 		enumerated_ref_put(&c->writes, BCH_WRITE_REF_delete_dead_snapshots);
 }
 

fs/bcachefs/snapshot.h

···
 
 static inline u32 bch2_snapshot_tree(struct bch_fs *c, u32 id)
 {
-	rcu_read_lock();
+	guard(rcu)();
 	const struct snapshot_t *s = snapshot_t(c, id);
-	id = s ? s->tree : 0;
-	rcu_read_unlock();
-
-	return id;
+	return s ? s->tree : 0;
 }
 
 static inline u32 __bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
···
 
 static inline u32 bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
 {
-	rcu_read_lock();
-	id = __bch2_snapshot_parent_early(c, id);
-	rcu_read_unlock();
-
-	return id;
+	guard(rcu)();
+	return __bch2_snapshot_parent_early(c, id);
 }
 
 static inline u32 __bch2_snapshot_parent(struct bch_fs *c, u32 id)
···
 
 static inline u32 bch2_snapshot_parent(struct bch_fs *c, u32 id)
 {
-	rcu_read_lock();
-	id = __bch2_snapshot_parent(c, id);
-	rcu_read_unlock();
-
-	return id;
+	guard(rcu)();
+	return __bch2_snapshot_parent(c, id);
 }
 
 static inline u32 bch2_snapshot_nth_parent(struct bch_fs *c, u32 id, u32 n)
 {
-	rcu_read_lock();
+	guard(rcu)();
 	while (n--)
 		id = __bch2_snapshot_parent(c, id);
-	rcu_read_unlock();
-
 	return id;
 }
 
···
 
 static inline u32 bch2_snapshot_root(struct bch_fs *c, u32 id)
 {
-	u32 parent;
+	guard(rcu)();
 
-	rcu_read_lock();
+	u32 parent;
 	while ((parent = __bch2_snapshot_parent(c, id)))
 		id = parent;
-	rcu_read_unlock();
-
 	return id;
 }
 
···
 
 static inline enum snapshot_id_state bch2_snapshot_id_state(struct bch_fs *c, u32 id)
 {
-	rcu_read_lock();
-	enum snapshot_id_state ret = __bch2_snapshot_id_state(c, id);
-	rcu_read_unlock();
-
-	return ret;
+	guard(rcu)();
+	return __bch2_snapshot_id_state(c, id);
 }
 
 static inline bool bch2_snapshot_exists(struct bch_fs *c, u32 id)
···
 
 static inline int bch2_snapshot_is_internal_node(struct bch_fs *c, u32 id)
 {
-	rcu_read_lock();
+	guard(rcu)();
 	const struct snapshot_t *s = snapshot_t(c, id);
-	int ret = s ? s->children[0] : -BCH_ERR_invalid_snapshot_node;
-	rcu_read_unlock();
-
-	return ret;
+	return s ? s->children[0] : -BCH_ERR_invalid_snapshot_node;
 }
 
 static inline int bch2_snapshot_is_leaf(struct bch_fs *c, u32 id)
···
 
 static inline u32 bch2_snapshot_depth(struct bch_fs *c, u32 parent)
 {
-	u32 depth;
-
-	rcu_read_lock();
-	depth = parent ? snapshot_t(c, parent)->depth + 1 : 0;
-	rcu_read_unlock();
-
-	return depth;
+	guard(rcu)();
+	return parent ? snapshot_t(c, parent)->depth + 1 : 0;
 }
 
 bool __bch2_snapshot_is_ancestor(struct bch_fs *, u32, u32);
···
 
 static inline bool bch2_snapshot_has_children(struct bch_fs *c, u32 id)
 {
-	rcu_read_lock();
+	guard(rcu)();
 	const struct snapshot_t *t = snapshot_t(c, id);
-	bool ret = t && (t->children[0]|t->children[1]) != 0;
-	rcu_read_unlock();
-
-	return ret;
+	return t && (t->children[0]|t->children[1]) != 0;
 }
 
 static inline bool snapshot_list_has_id(snapshot_id_list *s, u32 id)
 {
-	darray_for_each(*s, i)
-		if (*i == id)
-			return true;
-	return false;
+	return darray_find(*s, id) != NULL;
 }
 
 static inline bool snapshot_list_has_ancestor(struct bch_fs *c, snapshot_id_list *s, u32 id)
···
 	return likely(bch2_snapshot_exists(trans->c, k.k->p.snapshot))
 		? 0
 		: __bch2_check_key_has_snapshot(trans, iter, k);
+}
+
+int __bch2_get_snapshot_overwrites(struct btree_trans *,
+				   enum btree_id, struct bpos,
+				   snapshot_id_list *);
+
+/*
+ * Get a list of snapshot IDs that have overwritten a given key:
+ */
+static inline int bch2_get_snapshot_overwrites(struct btree_trans *trans,
+					       enum btree_id btree, struct bpos pos,
+					       snapshot_id_list *s)
+{
+	darray_init(s);
+
+	return bch2_snapshot_has_children(trans->c, pos.snapshot)
+		? __bch2_get_snapshot_overwrites(trans, btree, pos, s)
+		: 0;
+
 }
 
 int bch2_snapshot_node_set_deleted(struct btree_trans *, u32);

fs/bcachefs/str_hash.c

···
 	}
 }
 
-static noinline 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)
+static int bch2_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,
+				   bool *updated_before_k_pos)
 {
 	struct qstr old_name = bch2_dirent_get_name(old);
-	struct bkey_i_dirent *new = bch2_trans_kmalloc(trans, bkey_bytes(old.k) + 32);
+	struct bkey_i_dirent *new = bch2_trans_kmalloc(trans, BKEY_U64s_MAX * sizeof(u64));
 	int ret = PTR_ERR_OR_ZERO(new);
 	if (ret)
 		return ret;
···
 	dirent_copy_target(new, old);
 	new->k.p = old.k->p;
 
+	char *renamed_buf = bch2_trans_kmalloc(trans, old_name.len + 20);
+	ret = PTR_ERR_OR_ZERO(renamed_buf);
+	if (ret)
+		return ret;
+
 	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, 0);
+		new->k.u64s = BKEY_U64s_MAX;
 
-		if (u64s > U8_MAX)
-			return -EINVAL;
+		struct qstr renamed_name = (struct qstr) QSTR_INIT(renamed_buf,
+					sprintf(renamed_buf, "%.*s.fsck_renamed-%u",
+						old_name.len, old_name.name, i));
 
-		new->k.u64s = u64s;
+		ret = bch2_dirent_init_name(new, hash_info, &renamed_name, NULL);
+		if (ret)
+			return ret;
 
 		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))
+						BTREE_UPDATE_internal_snapshot_node|
+						STR_HASH_must_create);
+		if (ret && !bch2_err_matches(ret, EEXIST))
 			break;
+		if (!ret) {
+			if (bpos_lt(new->k.p, old.k->p))
+				*updated_before_k_pos = true;
+			break;
+		}
 	}
 
-	if (ret)
-		return ret;
-
-	return bch2_fsck_update_backpointers(trans, s, desc, hash_info, &new->k_i);
+	ret = ret ?: bch2_fsck_update_backpointers(trans, s, desc, hash_info, &new->k_i);
+	bch_err_fn(trans->c, ret);
+	return ret;
 }
 
 static noinline int hash_pick_winner(struct btree_trans *trans,
···
 #endif
 		bch2_print_str(c, KERN_ERR, buf.buf);
 		printbuf_exit(&buf);
-		ret = -BCH_ERR_fsck_repair_unimplemented;
+		ret = bch_err_throw(c, fsck_repair_unimplemented);
 		goto err;
 	}
 
···
 	return ret;
 }
 
+/* Put a str_hash key in its proper location, checking for duplicates */
+int bch2_str_hash_repair_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 k,
+			     struct btree_iter *dup_iter, struct bkey_s_c dup_k,
+			     bool *updated_before_k_pos)
+{
+	struct bch_fs *c = trans->c;
+	struct printbuf buf = PRINTBUF;
+	bool free_snapshots_seen = false;
+	int ret = 0;
+
+	if (!s) {
+		s = bch2_trans_kmalloc(trans, sizeof(*s));
+		ret = PTR_ERR_OR_ZERO(s);
+		if (ret)
+			goto out;
+
+		s->pos = k_iter->pos;
+		darray_init(&s->ids);
+
+		ret = bch2_get_snapshot_overwrites(trans, desc->btree_id, k_iter->pos, &s->ids);
+		if (ret)
+			goto out;
+
+		free_snapshots_seen = true;
+	}
+
+	if (!dup_k.k) {
+		struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k);
+		ret = PTR_ERR_OR_ZERO(new);
+		if (ret)
+			goto out;
+
+		dup_k = bch2_hash_set_or_get_in_snapshot(trans, dup_iter, *desc, hash_info,
+				       (subvol_inum) { 0, new->k.p.inode },
+				       new->k.p.snapshot, new,
+				       STR_HASH_must_create|
+				       BTREE_ITER_with_updates|
+				       BTREE_UPDATE_internal_snapshot_node);
+		ret = bkey_err(dup_k);
+		if (ret)
+			goto out;
+		if (dup_k.k)
+			goto duplicate_entries;
+
+		if (bpos_lt(new->k.p, k.k->p))
+			*updated_before_k_pos = true;
+
+		ret =   bch2_insert_snapshot_whiteouts(trans, desc->btree_id,
+						       k_iter->pos, new->k.p) ?:
+			bch2_hash_delete_at(trans, *desc, hash_info, k_iter,
+					    BTREE_ITER_with_updates|
+					    BTREE_UPDATE_internal_snapshot_node) ?:
+			bch2_fsck_update_backpointers(trans, s, *desc, hash_info, new) ?:
+			bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc) ?:
+			-BCH_ERR_transaction_restart_commit;
+	} else {
+duplicate_entries:
+		ret = hash_pick_winner(trans, *desc, hash_info, k, dup_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, k),
+			       prt_newline(&buf),
+			       bch2_bkey_val_to_text(&buf, c, dup_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, dup_iter, 0);
+			break;
+		case 2:
+			ret = bch2_fsck_rename_dirent(trans, s, *desc, hash_info,
+						      bkey_s_c_to_dirent(k),
+						      updated_before_k_pos) ?:
+				bch2_hash_delete_at(trans, *desc, hash_info, k_iter,
+						    BTREE_ITER_with_updates);
+			goto out;
+		}
+
+		ret = bch2_trans_commit(trans, NULL, NULL, 0) ?:
+			-BCH_ERR_transaction_restart_commit;
+	}
+out:
+fsck_err:
+	bch2_trans_iter_exit(trans, dup_iter);
+	printbuf_exit(&buf);
+	if (free_snapshots_seen)
+		darray_exit(&s->ids);
+	return ret;
+}
+
 int __bch2_str_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)
+			      struct btree_iter *k_iter, struct bkey_s_c hash_k,
+			      bool *updated_before_k_pos)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_iter iter = {};
···
 
 	for_each_btree_key_norestart(trans, iter, desc->btree_id,
 				     SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
-				     BTREE_ITER_slots, k, ret) {
+				     BTREE_ITER_slots|
+				     BTREE_ITER_with_updates, k, ret) {
 		if (bkey_eq(k.k->p, hash_k.k->p))
 			break;
 
 		if (k.k->type == desc->key_type &&
-		    !desc->cmp_bkey(k, hash_k))
-			goto duplicate_entries;
-
-		if (bkey_deleted(k.k)) {
-			bch2_trans_iter_exit(trans, &iter);
-			goto bad_hash;
+		    !desc->cmp_bkey(k, hash_k)) {
+			ret =	check_inode_hash_info_matches_root(trans, hash_k.k->p.inode,
+								   hash_info) ?:
+				bch2_str_hash_repair_key(trans, s, desc, hash_info,
+							 k_iter, hash_k,
+							 &iter, k, updated_before_k_pos);
+			break;
 		}
+
+		if (bkey_deleted(k.k))
+			goto bad_hash;
 	}
-out:
 	bch2_trans_iter_exit(trans, &iter);
+out:
+fsck_err:
 	printbuf_exit(&buf);
 	return ret;
 bad_hash:
+	bch2_trans_iter_exit(trans, &iter);
 	/*
 	 * Before doing any repair, check hash_info itself:
 	 */
···
 		goto out;
 
 	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",
-		     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))) {
-		struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, hash_k);
-		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)
-			goto out;
-		if (k.k)
-			goto duplicate_entries;
-
-		ret =   bch2_hash_delete_at(trans, *desc, hash_info, k_iter,
-					    BTREE_UPDATE_internal_snapshot_node) ?:
-			bch2_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;
+		     "hash table key at wrong offset: should be at %llu\n%s",
+		     hash,
+		     (bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf)))
+		ret = bch2_str_hash_repair_key(trans, s, desc, hash_info,
+					       k_iter, hash_k,
+					       &iter, bkey_s_c_null,
+					       updated_before_k_pos);
 	goto out;
 }

fs/bcachefs/str_hash.h

···
 			   struct bkey_i *insert,
 			   enum btree_iter_update_trigger_flags flags)
 {
+	struct bch_fs *c = trans->c;
 	struct btree_iter slot = {};
 	struct bkey_s_c k;
 	bool found = false;
···
 	}
 
 	if (!ret)
-		ret = -BCH_ERR_ENOSPC_str_hash_create;
+		ret = bch_err_throw(c, ENOSPC_str_hash_create);
 out:
 	bch2_trans_iter_exit(trans, &slot);
 	bch2_trans_iter_exit(trans, iter);
···
 		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;
+		ret = bch_err_throw(c, ENOENT_str_hash_set_must_replace);
 	} else {
 		if (!found && slot.path)
 			swap(*iter, slot);
···
 		return ret;
 	if (k.k) {
 		bch2_trans_iter_exit(trans, &iter);
-		return -BCH_ERR_EEXIST_str_hash_set;
+		return bch_err_throw(trans->c, EEXIST_str_hash_set);
 	}
 
 	return 0;
···
 int bch2_repair_inode_hash_info(struct btree_trans *, struct bch_inode_unpacked *);
 
 struct snapshots_seen;
+int bch2_str_hash_repair_key(struct btree_trans *,
+			     struct snapshots_seen *,
+			     const struct bch_hash_desc *,
+			     struct bch_hash_info *,
+			     struct btree_iter *, struct bkey_s_c,
+			     struct btree_iter *, struct bkey_s_c,
+			     bool *);
+
 int __bch2_str_hash_check_key(struct btree_trans *,
 			      struct snapshots_seen *,
 			      const struct bch_hash_desc *,
 			      struct bch_hash_info *,
-			      struct btree_iter *, struct bkey_s_c);
+			      struct btree_iter *, struct bkey_s_c,
+			      bool *);
 
 static inline int bch2_str_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)
+			    struct btree_iter *k_iter, struct bkey_s_c hash_k,
+			    bool *updated_before_k_pos)
 {
 	if (hash_k.k->type != desc->key_type)
 		return 0;
···
 	if (likely(desc->hash_bkey(hash_info, hash_k) == hash_k.k->p.offset))
 		return 0;
 
-	return __bch2_str_hash_check_key(trans, s, desc, hash_info, k_iter, hash_k);
+	return __bch2_str_hash_check_key(trans, s, desc, hash_info, k_iter, hash_k,
+					 updated_before_k_pos);
 }
 
 #endif /* _BCACHEFS_STR_HASH_H */

fs/bcachefs/subvolume.c

···
 			     "subvolume %llu points to missing subvolume root %llu:%u",
 			     k.k->p.offset, le64_to_cpu(subvol.v->inode),
 			     le32_to_cpu(subvol.v->snapshot))) {
-			ret = bch2_subvolume_delete(trans, iter->pos.offset);
-			bch_err_msg(c, ret, "deleting subvolume %llu", iter->pos.offset);
-			ret = ret ?: -BCH_ERR_transaction_restart_nested;
-			goto err;
+			/*
+			 * Recreate - any contents that are still disconnected
+			 * will then get reattached under lost+found
+			 */
+			bch2_inode_init_early(c, &inode);
+			bch2_inode_init_late(c, &inode, bch2_current_time(c),
+					     0, 0, S_IFDIR|0700, 0, NULL);
+			inode.bi_inum			= le64_to_cpu(subvol.v->inode);
+			inode.bi_snapshot		= le32_to_cpu(subvol.v->snapshot);
+			inode.bi_subvol			= k.k->p.offset;
+			inode.bi_parent_subvol		= le32_to_cpu(subvol.v->fs_path_parent);
+			ret = __bch2_fsck_write_inode(trans, &inode);
+			if (ret)
+				goto err;
 		}
 	} else {
 		goto err;
···
 
 	if (!BCH_SUBVOLUME_SNAP(subvol.v)) {
 		u32 snapshot_root = bch2_snapshot_root(c, le32_to_cpu(subvol.v->snapshot));
-		u32 snapshot_tree;
+		u32 snapshot_tree = bch2_snapshot_tree(c, snapshot_root);
+
 		struct bch_snapshot_tree st;
-
-		rcu_read_lock();
-		snapshot_tree = snapshot_t(c, snapshot_root)->tree;
-		rcu_read_unlock();
-
 		ret = bch2_snapshot_tree_lookup(trans, snapshot_tree, &st);
 
 		bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
···
 		prt_printf(out, " creation_parent %u", le32_to_cpu(s.v->creation_parent));
 		prt_printf(out, " fs_parent %u", le32_to_cpu(s.v->fs_path_parent));
 	}
+
+	if (BCH_SUBVOLUME_RO(s.v))
+		prt_printf(out, " ro");
+	if (BCH_SUBVOLUME_SNAP(s.v))
+		prt_printf(out, " snapshot");
+	if (BCH_SUBVOLUME_UNLINKED(s.v))
+		prt_printf(out, " unlinked");
 }
 
 static int subvolume_children_mod(struct btree_trans *trans, struct bpos pos, bool set)
···
 
 static int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
 {
-	return bch2_subvolumes_reparent(trans, subvolid) ?:
+	int ret = bch2_subvolumes_reparent(trans, subvolid) ?:
 		commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 			  __bch2_subvolume_delete(trans, subvolid));
+
+	bch2_recovery_pass_set_no_ratelimit(trans->c, BCH_RECOVERY_PASS_check_subvols);
+	return ret;
 }
 
 static void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
···
 	ret = bch2_bkey_get_empty_slot(trans, &dst_iter,
 				BTREE_ID_subvolumes, POS(0, U32_MAX));
 	if (ret == -BCH_ERR_ENOSPC_btree_slot)
-		ret = -BCH_ERR_ENOSPC_subvolume_create;
+		ret = bch_err_throw(c, ENOSPC_subvolume_create);
 	if (ret)
 		return ret;
 
···
 		return ret;
 
 	if (!bkey_is_inode(k.k)) {
-		bch_err(trans->c, "root inode not found");
-		ret = -BCH_ERR_ENOENT_inode;
+		struct bch_fs *c = trans->c;
+		bch_err(c, "root inode not found");
+		ret = bch_err_throw(c, ENOENT_inode);
 		goto err;
 	}
 

fs/bcachefs/super-io.c

···
 		prt_str(&buf, ")");
 		bch2_fs_fatal_error(c, ": %s", buf.buf);
 		printbuf_exit(&buf);
-		ret = -BCH_ERR_sb_not_downgraded;
+		ret = bch_err_throw(c, sb_not_downgraded);
 		goto out;
 	}
 
···
 
 			if (c->opts.errors != BCH_ON_ERROR_continue &&
 			    c->opts.errors != BCH_ON_ERROR_fix_safe) {
-				ret = -BCH_ERR_erofs_sb_err;
+				ret = bch_err_throw(c, erofs_sb_err);
 				bch2_fs_fatal_error(c, "%s", buf.buf);
 			} else {
 				bch_err(c, "%s", buf.buf);
···
 				ca->disk_sb.seq);
 			bch2_fs_fatal_error(c, "%s", buf.buf);
 			printbuf_exit(&buf);
-			ret = -BCH_ERR_erofs_sb_err;
+			ret = bch_err_throw(c, erofs_sb_err);
 		}
 	}
 
···
 				  !can_mount_with_written), c,
 		": Unable to write superblock to sufficient devices (from %ps)",
 		(void *) _RET_IP_))
-		ret = -BCH_ERR_erofs_sb_err;
+		ret = bch_err_throw(c, erofs_sb_err);
 out:
 	/* Make new options visible after they're persistent: */
 	bch2_sb_update(c);

fs/bcachefs/super.c

···
 
 struct bch_fs *bch2_dev_to_fs(dev_t dev)
 {
+	guard(mutex)(&bch_fs_list_lock);
+	guard(rcu)();
+
 	struct bch_fs *c;
-
-	mutex_lock(&bch_fs_list_lock);
-	rcu_read_lock();
-
 	list_for_each_entry(c, &bch_fs_list, list)
 		for_each_member_device_rcu(c, ca, NULL)
 			if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
 				closure_get(&c->cl);
-				goto found;
+				return c;
 			}
-	c = NULL;
-found:
-	rcu_read_unlock();
-	mutex_unlock(&bch_fs_list_lock);
-
-	return c;
+	return NULL;
 }
 
 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
···
 	BUG_ON(!test_bit(BCH_FS_may_go_rw, &c->flags));
 
 	if (WARN_ON(c->sb.features & BIT_ULL(BCH_FEATURE_no_alloc_info)))
-		return -BCH_ERR_erofs_no_alloc_info;
+		return bch_err_throw(c, erofs_no_alloc_info);
 
 	if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
 		bch_err(c, "cannot go rw, unfixed btree errors");
-		return -BCH_ERR_erofs_unfixed_errors;
+		return bch_err_throw(c, erofs_unfixed_errors);
 	}
 
 	if (c->sb.features & BIT_ULL(BCH_FEATURE_small_image)) {
 		bch_err(c, "cannot go rw, filesystem is an unresized image file");
-		return -BCH_ERR_erofs_filesystem_full;
+		return bch_err_throw(c, erofs_filesystem_full);
 	}
 
 	if (test_bit(BCH_FS_rw, &c->flags))
···
 
 	clear_bit(BCH_FS_clean_shutdown, &c->flags);
 
-	rcu_read_lock();
-	for_each_online_member_rcu(c, ca)
-		if (ca->mi.state == BCH_MEMBER_STATE_rw) {
-			bch2_dev_allocator_add(c, ca);
-			enumerated_ref_start(&ca->io_ref[WRITE]);
-		}
-	rcu_read_unlock();
+	scoped_guard(rcu)
+		for_each_online_member_rcu(c, ca)
+			if (ca->mi.state == BCH_MEMBER_STATE_rw) {
+				bch2_dev_allocator_add(c, ca);
+				enumerated_ref_start(&ca->io_ref[WRITE]);
+			}
 
 	bch2_recalc_capacity(c);
 
···
 {
 	if (c->opts.recovery_pass_last &&
 	    c->opts.recovery_pass_last < BCH_RECOVERY_PASS_journal_replay)
-		return -BCH_ERR_erofs_norecovery;
+		return bch_err_throw(c, erofs_norecovery);
 
 	if (c->opts.nochanges)
-		return -BCH_ERR_erofs_nochanges;
+		return bch_err_throw(c, erofs_nochanges);
 
 	if (c->sb.features & BIT_ULL(BCH_FEATURE_no_alloc_info))
-		return -BCH_ERR_erofs_no_alloc_info;
+		return bch_err_throw(c, erofs_no_alloc_info);
 
 	return __bch2_fs_read_write(c, false);
 }
···
 	if (c->sb.multi_device &&
 	    __bch2_uuid_to_fs(c->sb.uuid)) {
 		bch_err(c, "filesystem UUID already open");
-		return -BCH_ERR_filesystem_uuid_already_open;
+		return bch_err_throw(c, filesystem_uuid_already_open);
 	}
 
 	ret = bch2_fs_chardev_init(c);
···
 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
 	    !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
 				WQ_FREEZABLE, 0)))
-		return -BCH_ERR_ENOMEM_fs_other_alloc;
+		return bch_err_throw(c, ENOMEM_fs_other_alloc);
 
 	int ret = bch2_fs_btree_interior_update_init(c) ?:
 		bch2_fs_btree_write_buffer_init(c) ?:
···
 	    mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1,
 				       c->opts.btree_node_size) ||
 	    mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048)) {
-		ret = -BCH_ERR_ENOMEM_fs_other_alloc;
+		ret = bch_err_throw(c, ENOMEM_fs_other_alloc);
 		goto err;
 	}
 
···
 		ret = -EINVAL;
 		goto err;
 	}
-	bch_info(c, "Using encoding defined by superblock: utf8-%u.%u.%u",
-		 unicode_major(BCH_FS_DEFAULT_UTF8_ENCODING),
-		 unicode_minor(BCH_FS_DEFAULT_UTF8_ENCODING),
-		 unicode_rev(BCH_FS_DEFAULT_UTF8_ENCODING));
 #else
 	if (c->sb.features & BIT_ULL(BCH_FEATURE_casefolding)) {
 		printk(KERN_ERR "Cannot mount a filesystem with casefolding on a kernel without CONFIG_UNICODE\n");
···
 
 	print_mount_opts(c);
 
+#ifdef CONFIG_UNICODE
+	bch_info(c, "Using encoding defined by superblock: utf8-%u.%u.%u",
+		 unicode_major(BCH_FS_DEFAULT_UTF8_ENCODING),
+		 unicode_minor(BCH_FS_DEFAULT_UTF8_ENCODING),
+		 unicode_rev(BCH_FS_DEFAULT_UTF8_ENCODING));
+#endif
+
 	if (!bch2_fs_may_start(c))
-		return -BCH_ERR_insufficient_devices_to_start;
+		return bch_err_throw(c, insufficient_devices_to_start);
 
 	down_write(&c->state_lock);
 	mutex_lock(&c->sb_lock);
···
 			sizeof(struct bch_sb_field_ext) / sizeof(u64))) {
 		mutex_unlock(&c->sb_lock);
 		up_write(&c->state_lock);
-		ret = -BCH_ERR_ENOSPC_sb;
+		ret = bch_err_throw(c, ENOSPC_sb);
 		goto err;
 	}
 
···
 		goto err;
 	}
 
-	rcu_read_lock();
-	for_each_online_member_rcu(c, ca)
-		bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
-		cpu_to_le64(now);
-	rcu_read_unlock();
+	scoped_guard(rcu)
+		for_each_online_member_rcu(c, ca)
+			bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
+			cpu_to_le64(now);
 
 	/*
 	 * Dno't write superblock yet: recovery might have to downgrade
 	 */
 	mutex_unlock(&c->sb_lock);
 
-	rcu_read_lock();
-	for_each_online_member_rcu(c, ca)
-		if (ca->mi.state == BCH_MEMBER_STATE_rw)
-			bch2_dev_allocator_add(c, ca);
-	rcu_read_unlock();
+	scoped_guard(rcu)
+		for_each_online_member_rcu(c, ca)
+			if (ca->mi.state == BCH_MEMBER_STATE_rw)
+				bch2_dev_allocator_add(c, ca);
 	bch2_recalc_capacity(c);
 	up_write(&c->state_lock);
 
···
 		goto err;
 
 	if (bch2_fs_init_fault("fs_start")) {
-		ret = -BCH_ERR_injected_fs_start;
+		ret = bch_err_throw(c, injected_fs_start);
 		goto err;
 	}
 
···
 	struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
 
 	if (le16_to_cpu(sb->block_size) != block_sectors(c))
-		return -BCH_ERR_mismatched_block_size;
+		return bch_err_throw(c, mismatched_block_size);
 
 	if (le16_to_cpu(m.bucket_size) <
 	    BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
-		return -BCH_ERR_bucket_size_too_small;
+		return bch_err_throw(c, bucket_size_too_small);
 
 	return 0;
 }
···
 	bch2_dev_attach(c, ca, dev_idx);
 	return 0;
 err:
-	return -BCH_ERR_ENOMEM_dev_alloc;
+	return bch_err_throw(c, ENOMEM_dev_alloc);
 }
 
 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
···
 	if (bch2_dev_is_online(ca)) {
 		bch_err(ca, "already have device online in slot %u",
 			sb->sb->dev_idx);
-		return -BCH_ERR_device_already_online;
+		return bch_err_throw(ca->fs, device_already_online);
 	}
 
 	if (get_capacity(sb->bdev->bd_disk) <
 	    ca->mi.bucket_size * ca->mi.nbuckets) {
 		bch_err(ca, "cannot online: device too small");
-		return -BCH_ERR_device_size_too_small;
+		return bch_err_throw(ca->fs, device_size_too_small);
 	}
 
 	BUG_ON(!enumerated_ref_is_zero(&ca->io_ref[READ]));
···
 		return 0;
 
 	if (!bch2_dev_state_allowed(c, ca, new_state, flags))
-		return -BCH_ERR_device_state_not_allowed;
+		return bch_err_throw(c, device_state_not_allowed);
 
 	if (new_state != BCH_MEMBER_STATE_rw)
 		__bch2_dev_read_only(c, ca);
···
 
 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
 		bch_err(ca, "Cannot remove without losing data");
-		ret = -BCH_ERR_device_state_not_allowed;
+		ret = bch_err_throw(c, device_state_not_allowed);
 		goto err;
 	}
 
···
 	if (list_empty(&c->list)) {
 		mutex_lock(&bch_fs_list_lock);
 		if (__bch2_uuid_to_fs(c->sb.uuid))
-			ret = -BCH_ERR_filesystem_uuid_already_open;
+			ret = bch_err_throw(c, filesystem_uuid_already_open);
 		else
 			list_add(&c->list, &bch_fs_list);
 		mutex_unlock(&bch_fs_list_lock);
···
 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
 		bch_err(ca, "Cannot offline required disk");
 		up_write(&c->state_lock);
-		return -BCH_ERR_device_state_not_allowed;
+		return bch_err_throw(c, device_state_not_allowed);
 	}
 
 	__bch2_dev_offline(c, ca);
···
 	if (nbuckets > BCH_MEMBER_NBUCKETS_MAX) {
 		bch_err(ca, "New device size too big (%llu greater than max %u)",
 			nbuckets, BCH_MEMBER_NBUCKETS_MAX);
-		ret = -BCH_ERR_device_size_too_big;
+		ret = bch_err_throw(c, device_size_too_big);
 		goto err;
 	}
 
···
 	    get_capacity(ca->disk_sb.bdev->bd_disk) <
 	    ca->mi.bucket_size * nbuckets) {
 		bch_err(ca, "New size larger than device");
-		ret = -BCH_ERR_device_size_too_small;
+		ret = bch_err_throw(c, device_size_too_small);
 		goto err;
 	}
 
···
 	}
 
 	if (opts->nochanges && !opts->read_only) {
-		ret = -BCH_ERR_erofs_nochanges;
+		ret = bch_err_throw(c, erofs_nochanges);
 		goto err_print;
 	}
 

fs/bcachefs/sysfs.c

···
 #include "disk_groups.h"
 #include "ec.h"
 #include "enumerated_ref.h"
+#include "error.h"
 #include "inode.h"
 #include "journal.h"
 #include "journal_reclaim.h"
···
 #include "rebalance.h"
 #include "recovery_passes.h"
 #include "replicas.h"
+#include "sb-errors.h"
 #include "super-io.h"
 #include "tests.h"
 
···
 write_attribute(trigger_gc);
 write_attribute(trigger_discards);
 write_attribute(trigger_invalidates);
+write_attribute(trigger_journal_commit);
 write_attribute(trigger_journal_flush);
 write_attribute(trigger_journal_writes);
 write_attribute(trigger_btree_cache_shrink);
···
 write_attribute(trigger_freelist_wakeup);
 write_attribute(trigger_recalc_capacity);
 write_attribute(trigger_delete_dead_snapshots);
+write_attribute(trigger_emergency_read_only);
 read_attribute(gc_gens_pos);
 
 read_attribute(uuid);
···
 
 read_attribute(btree_cache_size);
 read_attribute(compression_stats);
+read_attribute(errors);
 read_attribute(journal_debug);
 read_attribute(btree_cache);
 read_attribute(btree_key_cache);
···
 	if (attr == &sysfs_compression_stats)
 		bch2_compression_stats_to_text(out, c);
 
+	if (attr == &sysfs_errors)
+		bch2_fs_errors_to_text(out, c);
+
 	if (attr == &sysfs_new_stripes)
 		bch2_new_stripes_to_text(out, c);
 
···
 	if (attr == &sysfs_trigger_invalidates)
 		bch2_do_invalidates(c);
 
+	if (attr == &sysfs_trigger_journal_commit)
+		bch2_journal_flush(&c->journal);
+
 	if (attr == &sysfs_trigger_journal_flush) {
 		bch2_journal_flush_all_pins(&c->journal);
 		bch2_journal_meta(&c->journal);
···
 
 	if (attr == &sysfs_trigger_delete_dead_snapshots)
 		__bch2_delete_dead_snapshots(c);
+
+	if (attr == &sysfs_trigger_emergency_read_only) {
+		struct printbuf buf = PRINTBUF;
+		bch2_log_msg_start(c, &buf);
+
+		prt_printf(&buf, "shutdown by sysfs\n");
+		bch2_fs_emergency_read_only2(c, &buf);
+		bch2_print_str(c, KERN_ERR, buf.buf);
+		printbuf_exit(&buf);
+	}
 
 #ifdef CONFIG_BCACHEFS_TESTS
 	if (attr == &sysfs_perf_test) {
···
 	&sysfs_recovery_status,
 
 	&sysfs_compression_stats,
+	&sysfs_errors,
 
 #ifdef CONFIG_BCACHEFS_TESTS
 	&sysfs_perf_test,
···
 	&sysfs_trigger_gc,
 	&sysfs_trigger_discards,
 	&sysfs_trigger_invalidates,
+	&sysfs_trigger_journal_commit,
 	&sysfs_trigger_journal_flush,
 	&sysfs_trigger_journal_writes,
 	&sysfs_trigger_btree_cache_shrink,
···
 	&sysfs_trigger_freelist_wakeup,
 	&sysfs_trigger_recalc_capacity,
 	&sysfs_trigger_delete_dead_snapshots,
+	&sysfs_trigger_emergency_read_only,
 
 	&sysfs_gc_gens_pos,
 

fs/bcachefs/trace.h

···
 		  (unsigned long long)__entry->sector, __entry->nr_sector)
 );
 
+/* errors */
+
+TRACE_EVENT(error_throw,
+	TP_PROTO(struct bch_fs *c, int bch_err, unsigned long ip),
+	TP_ARGS(c, bch_err, ip),
+
+	TP_STRUCT__entry(
+		__field(dev_t,		dev			)
+		__field(int,		err			)
+		__array(char,		err_str, 32		)
+		__array(char,		ip, 32			)
+	),
+
+	TP_fast_assign(
+		__entry->dev		= c->dev;
+		__entry->err		= bch_err;
+		strscpy(__entry->err_str, bch2_err_str(bch_err), sizeof(__entry->err_str));
+		snprintf(__entry->ip, sizeof(__entry->ip), "%ps", (void *) ip);
+	),
+
+	TP_printk("%d,%d %s ret %s", MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->ip, __entry->err_str)
+);
+
+TRACE_EVENT(error_downcast,
+	TP_PROTO(int bch_err, int std_err, unsigned long ip),
+	TP_ARGS(bch_err, std_err, ip),
+
+	TP_STRUCT__entry(
+		__array(char,		bch_err, 32		)
+		__array(char,		std_err, 32		)
+		__array(char,		ip, 32			)
+	),
+
+	TP_fast_assign(
+		strscpy(__entry->bch_err, bch2_err_str(bch_err), sizeof(__entry->bch_err));
+		strscpy(__entry->std_err, bch2_err_str(std_err), sizeof(__entry->std_err));
+		snprintf(__entry->ip, sizeof(__entry->ip), "%ps", (void *) ip);
+	),
+
+	TP_printk("%s ret %s -> %s %s", __entry->ip,
+		  __entry->bch_err, __entry->std_err, __entry->ip)
+);
+
 /* disk_accounting.c */
 
 TRACE_EVENT(accounting_mem_insert,
···
 	TP_ARGS(c, str)
 );
 
+DEFINE_EVENT(fs_str, io_move_pred,
+	TP_PROTO(struct bch_fs *c, const char *str),
+	TP_ARGS(c, str)
+);
+
 DEFINE_EVENT(fs_str, io_move_created_rebalance,
 	TP_PROTO(struct bch_fs *c, const char *str),
 	TP_ARGS(c, str)
 );
 
-TRACE_EVENT(error_downcast,
-	TP_PROTO(int bch_err, int std_err, unsigned long ip),
-	TP_ARGS(bch_err, std_err, ip),
-
-	TP_STRUCT__entry(
-		__array(char,		bch_err, 32		)
-		__array(char,		std_err, 32		)
-		__array(char,		ip, 32			)
-	),
-
-	TP_fast_assign(
-		strscpy(__entry->bch_err, bch2_err_str(bch_err), sizeof(__entry->bch_err));
-		strscpy(__entry->std_err, bch2_err_str(std_err), sizeof(__entry->std_err));
-		snprintf(__entry->ip, sizeof(__entry->ip), "%ps", (void *) ip);
-	),
-
-	TP_printk("%s -> %s %s", __entry->bch_err, __entry->std_err, __entry->ip)
+DEFINE_EVENT(fs_str, io_move_evacuate_bucket,
+	TP_PROTO(struct bch_fs *c, const char *str),
+	TP_ARGS(c, str)
 );
 
 #ifdef CONFIG_BCACHEFS_PATH_TRACEPOINTS