Merge tag 'bcachefs-2024-06-28' of https://evilpiepirate.org/git/bcachefs

Pull bcachefs fixes from Kent Overstreet:
 "Simple stuff:

   - NULL ptr/err ptr deref fixes

   - fix for getting wedged on shutdown after journal error

   - fix missing recalc_capacity() call, capacity now changes correctly
     after a device goes read only

     however: our capacity calculation still doesn't take into account
     when we have mixed ro/rw devices and the ro devices have data on
     them, that's going to be a more involved fix to separate accounting
     for "capacity used on ro devices" and "capacity used on rw devices"

   - boring syzbot stuff

  Slightly more involved:

   - discard, invalidate workers are now per device

     this has the effect of simplifying how we take device refs in these
     paths, and the device ref cleanup fixes a longstanding race between
     the device removal path and the discard path

   - fixes for how the debugfs code takes refs on btree_trans objects we
     have debugfs code that prints in use btree_trans objects.

     It uses closure_get() on trans->ref, which is mainly for the cycle
     detector, but the debugfs code was using it on a closure that may
     have hit 0, which is not allowed; for performance reasons we cannot
     avoid having not-in-use transactions on the global list.

     Introduce some new primitives to fix this and make the
     synchronization here a whole lot saner"

* tag 'bcachefs-2024-06-28' of https://evilpiepirate.org/git/bcachefs:
  bcachefs: Fix kmalloc bug in __snapshot_t_mut
  bcachefs: Discard, invalidate workers are now per device
  bcachefs: Fix shift-out-of-bounds in bch2_blacklist_entries_gc
  bcachefs: slab-use-after-free Read in bch2_sb_errors_from_cpu
  bcachefs: Add missing bch2_journal_do_writes() call
  bcachefs: Fix null ptr deref in journal_pins_to_text()
  bcachefs: Add missing recalc_capacity() call
  bcachefs: Fix btree_trans list ordering
  bcachefs: Fix race between trans_put() and btree_transactions_read()
  closures: closure_get_not_zero(), closure_return_sync()
  bcachefs: Make btree_deadlock_to_text() clearer
  bcachefs: fix seqmutex_relock()
  bcachefs: Fix freeing of error pointers

14 files changed, 273 insertions, 201 deletions

diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c
index 8dec2c6cbb7e..1de9fac3bcf4 100644
--- a/fs/bcachefs/alloc_background.c
+++ b/fs/bcachefs/alloc_background.c
@@ -29,7 +29,7 @@
 #include <linux/sched/task.h>
 #include <linux/sort.h>
 
-static void bch2_discard_one_bucket_fast(struct bch_fs *c, struct bpos bucket);
+static void bch2_discard_one_bucket_fast(struct bch_dev *, u64);
 
 /* Persistent alloc info: */
 
@@ -893,12 +893,12 @@ int bch2_trigger_alloc(struct btree_trans *trans,
 		if (statechange(a->data_type == BCH_DATA_need_discard) &&
 		    !bch2_bucket_is_open_safe(c, new.k->p.inode, new.k->p.offset) &&
 		    bucket_flushed(new_a))
-			bch2_discard_one_bucket_fast(c, new.k->p);
+			bch2_discard_one_bucket_fast(ca, new.k->p.offset);
 
 		if (statechange(a->data_type == BCH_DATA_cached) &&
 		    !bch2_bucket_is_open(c, new.k->p.inode, new.k->p.offset) &&
 		    should_invalidate_buckets(ca, bch2_dev_usage_read(ca)))
-			bch2_do_invalidates(c);
+			bch2_dev_do_invalidates(ca);
 
 		if (statechange(a->data_type == BCH_DATA_need_gc_gens))
 			bch2_gc_gens_async(c);
@@ -1636,34 +1636,38 @@ int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
 	return ret;
 }
 
-static int discard_in_flight_add(struct bch_fs *c, struct bpos bucket)
+static int discard_in_flight_add(struct bch_dev *ca, u64 bucket, bool in_progress)
 {
 	int ret;
 
-	mutex_lock(&c->discard_buckets_in_flight_lock);
-	darray_for_each(c->discard_buckets_in_flight, i)
-		if (bkey_eq(*i, bucket)) {
+	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;
 		}
 
-	ret = darray_push(&c->discard_buckets_in_flight, bucket);
+	ret = darray_push(&ca->discard_buckets_in_flight, ((struct discard_in_flight) {
+			   .in_progress = in_progress,
+			   .bucket	= bucket,
+	}));
 out:
-	mutex_unlock(&c->discard_buckets_in_flight_lock);
+	mutex_unlock(&ca->discard_buckets_in_flight_lock);
 	return ret;
 }
 
-static void discard_in_flight_remove(struct bch_fs *c, struct bpos bucket)
+static void discard_in_flight_remove(struct bch_dev *ca, u64 bucket)
 {
-	mutex_lock(&c->discard_buckets_in_flight_lock);
-	darray_for_each(c->discard_buckets_in_flight, i)
-		if (bkey_eq(*i, bucket)) {
-			darray_remove_item(&c->discard_buckets_in_flight, i);
+	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:
-	mutex_unlock(&c->discard_buckets_in_flight_lock);
+	mutex_unlock(&ca->discard_buckets_in_flight_lock);
 }
 
 struct discard_buckets_state {
@@ -1671,26 +1675,11 @@ struct discard_buckets_state {
 	u64		open;
 	u64		need_journal_commit;
 	u64		discarded;
-	struct bch_dev	*ca;
 	u64		need_journal_commit_this_dev;
 };
 
-static void discard_buckets_next_dev(struct bch_fs *c, struct discard_buckets_state *s, struct bch_dev *ca)
-{
-	if (s->ca == ca)
-		return;
-
-	if (s->ca && s->need_journal_commit_this_dev >
-	    bch2_dev_usage_read(s->ca).d[BCH_DATA_free].buckets)
-		bch2_journal_flush_async(&c->journal, NULL);
-
-	if (s->ca)
-		percpu_ref_put(&s->ca->io_ref);
-	s->ca = ca;
-	s->need_journal_commit_this_dev = 0;
-}
-
 static int bch2_discard_one_bucket(struct btree_trans *trans,
+				   struct bch_dev *ca,
 				   struct btree_iter *need_discard_iter,
 				   struct bpos *discard_pos_done,
 				   struct discard_buckets_state *s)
@@ -1704,16 +1693,6 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 	bool discard_locked = false;
 	int ret = 0;
 
-	struct bch_dev *ca = s->ca && s->ca->dev_idx == pos.inode
-		? s->ca
-		: bch2_dev_get_ioref(c, pos.inode, WRITE);
-	if (!ca) {
-		bch2_btree_iter_set_pos(need_discard_iter, POS(pos.inode + 1, 0));
-		return 0;
-	}
-
-	discard_buckets_next_dev(c, s, ca);
-
 	if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) {
 		s->open++;
 		goto out;
@@ -1773,7 +1752,7 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 		goto out;
 	}
 
-	if (discard_in_flight_add(c, SPOS(iter.pos.inode, iter.pos.offset, true)))
+	if (discard_in_flight_add(ca, iter.pos.offset, true))
 		goto out;
 
 	discard_locked = true;
@@ -1811,7 +1790,7 @@ write:
 	s->discarded++;
 out:
 	if (discard_locked)
-		discard_in_flight_remove(c, iter.pos);
+		discard_in_flight_remove(ca, iter.pos.offset);
 	s->seen++;
 	bch2_trans_iter_exit(trans, &iter);
 	printbuf_exit(&buf);
@@ -1820,7 +1799,8 @@ out:
 
 static void bch2_do_discards_work(struct work_struct *work)
 {
-	struct bch_fs *c = container_of(work, struct bch_fs, discard_work);
+	struct bch_dev *ca = container_of(work, struct bch_dev, discard_work);
+	struct bch_fs *c = ca->fs;
 	struct discard_buckets_state s = {};
 	struct bpos discard_pos_done = POS_MAX;
 	int ret;
@@ -1831,23 +1811,41 @@ static void bch2_do_discards_work(struct work_struct *work)
 	 * successful commit:
 	 */
 	ret = bch2_trans_run(c,
-		for_each_btree_key(trans, iter,
-				   BTREE_ID_need_discard, POS_MIN, 0, k,
-			bch2_discard_one_bucket(trans, &iter, &discard_pos_done, &s)));
-
-	discard_buckets_next_dev(c, &s, NULL);
+		for_each_btree_key_upto(trans, iter,
+				   BTREE_ID_need_discard,
+				   POS(ca->dev_idx, 0),
+				   POS(ca->dev_idx, U64_MAX), 0, k,
+			bch2_discard_one_bucket(trans, ca, &iter, &discard_pos_done, &s)));
 
 	trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded,
 			      bch2_err_str(ret));
 
 	bch2_write_ref_put(c, BCH_WRITE_REF_discard);
+	percpu_ref_put(&ca->io_ref);
+}
+
+void bch2_dev_do_discards(struct bch_dev *ca)
+{
+	struct bch_fs *c = ca->fs;
+
+	if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE))
+		return;
+
+	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard))
+		goto put_ioref;
+
+	if (queue_work(c->write_ref_wq, &ca->discard_work))
+		return;
+
+	bch2_write_ref_put(c, BCH_WRITE_REF_discard);
+put_ioref:
+	percpu_ref_put(&ca->io_ref);
 }
 
 void bch2_do_discards(struct bch_fs *c)
 {
-	if (bch2_write_ref_tryget(c, BCH_WRITE_REF_discard) &&
-	    !queue_work(c->write_ref_wq, &c->discard_work))
-		bch2_write_ref_put(c, BCH_WRITE_REF_discard);
+	for_each_member_device(c, ca)
+		bch2_dev_do_discards(ca);
 }
 
 static int bch2_clear_bucket_needs_discard(struct btree_trans *trans, struct bpos bucket)
@@ -1876,68 +1874,69 @@ err:
 
 static void bch2_do_discards_fast_work(struct work_struct *work)
 {
-	struct bch_fs *c = container_of(work, struct bch_fs, discard_fast_work);
+	struct bch_dev *ca = container_of(work, struct bch_dev, discard_fast_work);
+	struct bch_fs *c = ca->fs;
 
 	while (1) {
 		bool got_bucket = false;
-		struct bpos bucket;
-		struct bch_dev *ca;
+		u64 bucket;
 
-		mutex_lock(&c->discard_buckets_in_flight_lock);
-		darray_for_each(c->discard_buckets_in_flight, i) {
-			if (i->snapshot)
+		mutex_lock(&ca->discard_buckets_in_flight_lock);
+		darray_for_each(ca->discard_buckets_in_flight, i) {
+			if (i->in_progress)
 				continue;
 
-			ca = bch2_dev_get_ioref(c, i->inode, WRITE);
-			if (!ca) {
-				darray_remove_item(&c->discard_buckets_in_flight, i);
-				continue;
-			}
-
 			got_bucket = true;
-			bucket = *i;
-			i->snapshot = true;
+			bucket = i->bucket;
+			i->in_progress = true;
 			break;
 		}
-		mutex_unlock(&c->discard_buckets_in_flight_lock);
+		mutex_unlock(&ca->discard_buckets_in_flight_lock);
 
 		if (!got_bucket)
 			break;
 
 		if (ca->mi.discard && !c->opts.nochanges)
 			blkdev_issue_discard(ca->disk_sb.bdev,
-					     bucket.offset * ca->mi.bucket_size,
+					     bucket_to_sector(ca, bucket),
 					     ca->mi.bucket_size,
 					     GFP_KERNEL);
 
 		int ret = bch2_trans_do(c, NULL, NULL,
-					BCH_WATERMARK_btree|
-					BCH_TRANS_COMMIT_no_enospc,
-					bch2_clear_bucket_needs_discard(trans, bucket));
+			BCH_WATERMARK_btree|
+			BCH_TRANS_COMMIT_no_enospc,
+			bch2_clear_bucket_needs_discard(trans, POS(ca->dev_idx, bucket)));
 		bch_err_fn(c, ret);
 
-		percpu_ref_put(&ca->io_ref);
-		discard_in_flight_remove(c, bucket);
+		discard_in_flight_remove(ca, bucket);
 
 		if (ret)
 			break;
 	}
 
 	bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast);
+	percpu_ref_put(&ca->io_ref);
 }
 
-static void bch2_discard_one_bucket_fast(struct bch_fs *c, struct bpos bucket)
+static void bch2_discard_one_bucket_fast(struct bch_dev *ca, u64 bucket)
 {
-	rcu_read_lock();
-	struct bch_dev *ca = bch2_dev_rcu(c, bucket.inode);
-	bool dead = !ca || percpu_ref_is_dying(&ca->io_ref);
-	rcu_read_unlock();
+	struct bch_fs *c = ca->fs;
+
+	if (discard_in_flight_add(ca, bucket, false))
+		return;
+
+	if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE))
+		return;
+
+	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard_fast))
+		goto put_ioref;
+
+	if (queue_work(c->write_ref_wq, &ca->discard_fast_work))
+		return;
 
-	if (!dead &&
-	    !discard_in_flight_add(c, bucket) &&
-	    bch2_write_ref_tryget(c, BCH_WRITE_REF_discard_fast) &&
-	    !queue_work(c->write_ref_wq, &c->discard_fast_work))
-		bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast);
+	bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast);
+put_ioref:
+	percpu_ref_put(&ca->io_ref);
 }
 
 static int invalidate_one_bucket(struct btree_trans *trans,
@@ -2038,7 +2037,8 @@ again:
 
 static void bch2_do_invalidates_work(struct work_struct *work)
 {
-	struct bch_fs *c = container_of(work, struct bch_fs, invalidate_work);
+	struct bch_dev *ca = container_of(work, struct bch_dev, invalidate_work);
+	struct bch_fs *c = ca->fs;
 	struct btree_trans *trans = bch2_trans_get(c);
 	int ret = 0;
 
@@ -2046,52 +2046,63 @@ static void bch2_do_invalidates_work(struct work_struct *work)
 	if (ret)
 		goto err;
 
-	for_each_member_device(c, ca) {
-		s64 nr_to_invalidate =
-			should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
-		struct btree_iter iter;
-		bool wrapped = false;
-
-		bch2_trans_iter_init(trans, &iter, BTREE_ID_lru,
-				     lru_pos(ca->dev_idx, 0,
-					     ((bch2_current_io_time(c, READ) + U32_MAX) &
-					      LRU_TIME_MAX)), 0);
-
-		while (true) {
-			bch2_trans_begin(trans);
-
-			struct bkey_s_c k = next_lru_key(trans, &iter, ca, &wrapped);
-			ret = bkey_err(k);
-			if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
-				continue;
-			if (ret)
-				break;
-			if (!k.k)
-				break;
+	s64 nr_to_invalidate =
+		should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
+	struct btree_iter iter;
+	bool wrapped = false;
 
-			ret = invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate);
-			if (ret)
-				break;
+	bch2_trans_iter_init(trans, &iter, BTREE_ID_lru,
+			     lru_pos(ca->dev_idx, 0,
+				     ((bch2_current_io_time(c, READ) + U32_MAX) &
+				      LRU_TIME_MAX)), 0);
 
-			bch2_btree_iter_advance(&iter);
-		}
-		bch2_trans_iter_exit(trans, &iter);
+	while (true) {
+		bch2_trans_begin(trans);
 
-		if (ret < 0) {
-			bch2_dev_put(ca);
+		struct bkey_s_c k = next_lru_key(trans, &iter, ca, &wrapped);
+		ret = bkey_err(k);
+		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+			continue;
+		if (ret)
 			break;
-		}
+		if (!k.k)
+			break;
+
+		ret = invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate);
+		if (ret)
+			break;
+
+		bch2_btree_iter_advance(&iter);
 	}
+	bch2_trans_iter_exit(trans, &iter);
 err:
 	bch2_trans_put(trans);
 	bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
+	percpu_ref_put(&ca->io_ref);
+}
+
+void bch2_dev_do_invalidates(struct bch_dev *ca)
+{
+	struct bch_fs *c = ca->fs;
+
+	if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE))
+		return;
+
+	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate))
+		goto put_ioref;
+
+	if (queue_work(c->write_ref_wq, &ca->invalidate_work))
+		return;
+
+	bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
+put_ioref:
+	percpu_ref_put(&ca->io_ref);
 }
 
 void bch2_do_invalidates(struct bch_fs *c)
 {
-	if (bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate) &&
-	    !queue_work(c->write_ref_wq, &c->invalidate_work))
-		bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
+	for_each_member_device(c, ca)
+		bch2_dev_do_invalidates(ca);
 }
 
 int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
@@ -2407,16 +2418,20 @@ void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca)
 			set_bit(ca->dev_idx, c->rw_devs[i].d);
 }
 
-void bch2_fs_allocator_background_exit(struct bch_fs *c)
+void bch2_dev_allocator_background_exit(struct bch_dev *ca)
+{
+	darray_exit(&ca->discard_buckets_in_flight);
+}
+
+void bch2_dev_allocator_background_init(struct bch_dev *ca)
 {
-	darray_exit(&c->discard_buckets_in_flight);
+	mutex_init(&ca->discard_buckets_in_flight_lock);
+	INIT_WORK(&ca->discard_work, bch2_do_discards_work);
+	INIT_WORK(&ca->discard_fast_work, bch2_do_discards_fast_work);
+	INIT_WORK(&ca->invalidate_work, bch2_do_invalidates_work);
 }
 
 void bch2_fs_allocator_background_init(struct bch_fs *c)
 {
 	spin_lock_init(&c->freelist_lock);
-	mutex_init(&c->discard_buckets_in_flight_lock);
-	INIT_WORK(&c->discard_work, bch2_do_discards_work);
-	INIT_WORK(&c->discard_fast_work, bch2_do_discards_fast_work);
-	INIT_WORK(&c->invalidate_work, bch2_do_invalidates_work);
 }
diff --git a/fs/bcachefs/alloc_background.h b/fs/bcachefs/alloc_background.h
index c3cc3c5ba5b6..ba2c5557a3f0 100644
--- a/fs/bcachefs/alloc_background.h
+++ b/fs/bcachefs/alloc_background.h
@@ -275,6 +275,7 @@ int bch2_trigger_alloc(struct btree_trans *, enum btree_id, unsigned,
 		       enum btree_iter_update_trigger_flags);
 int bch2_check_alloc_info(struct bch_fs *);
 int bch2_check_alloc_to_lru_refs(struct bch_fs *);
+void bch2_dev_do_discards(struct bch_dev *);
 void bch2_do_discards(struct bch_fs *);
 
 static inline u64 should_invalidate_buckets(struct bch_dev *ca,
@@ -289,6 +290,7 @@ static inline u64 should_invalidate_buckets(struct bch_dev *ca,
 	return clamp_t(s64, want_free - free, 0, u.d[BCH_DATA_cached].buckets);
 }
 
+void bch2_dev_do_invalidates(struct bch_dev *);
 void bch2_do_invalidates(struct bch_fs *);
 
 static inline struct bch_backpointer *alloc_v4_backpointers(struct bch_alloc_v4 *a)
@@ -312,7 +314,9 @@ u64 bch2_min_rw_member_capacity(struct bch_fs *);
 void bch2_dev_allocator_remove(struct bch_fs *, struct bch_dev *);
 void bch2_dev_allocator_add(struct bch_fs *, struct bch_dev *);
 
-void bch2_fs_allocator_background_exit(struct bch_fs *);
+void bch2_dev_allocator_background_exit(struct bch_dev *);
+void bch2_dev_allocator_background_init(struct bch_dev *);
+
 void bch2_fs_allocator_background_init(struct bch_fs *);
 
 #endif /* _BCACHEFS_ALLOC_BACKGROUND_H */
diff --git a/fs/bcachefs/alloc_foreground.c b/fs/bcachefs/alloc_foreground.c
index 927a5f300b30..9d3d64746a5b 100644
--- a/fs/bcachefs/alloc_foreground.c
+++ b/fs/bcachefs/alloc_foreground.c
@@ -621,13 +621,13 @@ again:
 	avail = dev_buckets_free(ca, *usage, watermark);
 
 	if (usage->d[BCH_DATA_need_discard].buckets > avail)
-		bch2_do_discards(c);
+		bch2_dev_do_discards(ca);
 
 	if (usage->d[BCH_DATA_need_gc_gens].buckets > avail)
 		bch2_gc_gens_async(c);
 
 	if (should_invalidate_buckets(ca, *usage))
-		bch2_do_invalidates(c);
+		bch2_dev_do_invalidates(ca);
 
 	if (!avail) {
 		if (cl && !waiting) {
diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h
index a6b83ecab7ce..1106fec6e155 100644
--- a/fs/bcachefs/bcachefs.h
+++ b/fs/bcachefs/bcachefs.h
@@ -493,6 +493,11 @@ struct io_count {
 	u64			sectors[2][BCH_DATA_NR];
 };
 
+struct discard_in_flight {
+	bool			in_progress:1;
+	u64			bucket:63;
+};
+
 struct bch_dev {
 	struct kobject		kobj;
 #ifdef CONFIG_BCACHEFS_DEBUG
@@ -554,6 +559,12 @@ struct bch_dev {
 	size_t			inc_gen_really_needs_gc;
 	size_t			buckets_waiting_on_journal;
 
+	struct work_struct	invalidate_work;
+	struct work_struct	discard_work;
+	struct mutex		discard_buckets_in_flight_lock;
+	DARRAY(struct discard_in_flight)	discard_buckets_in_flight;
+	struct work_struct	discard_fast_work;
+
 	atomic64_t		rebalance_work;
 
 	struct journal_device	journal;
@@ -915,11 +926,6 @@ struct bch_fs {
 	unsigned		write_points_nr;
 
 	struct buckets_waiting_for_journal buckets_waiting_for_journal;
-	struct work_struct	invalidate_work;
-	struct work_struct	discard_work;
-	struct mutex		discard_buckets_in_flight_lock;
-	DARRAY(struct bpos)	discard_buckets_in_flight;
-	struct work_struct	discard_fast_work;
 
 	/* GARBAGE COLLECTION */
 	struct work_struct	gc_gens_work;
diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c
index 3a1419d17888..0ed9e6574fcd 100644
--- a/fs/bcachefs/btree_iter.c
+++ b/fs/bcachefs/btree_iter.c
@@ -3130,7 +3130,6 @@ struct btree_trans *__bch2_trans_get(struct bch_fs *c, unsigned fn_idx)
 
 	trans = mempool_alloc(&c->btree_trans_pool, GFP_NOFS);
 	memset(trans, 0, sizeof(*trans));
-	closure_init_stack(&trans->ref);
 
 	seqmutex_lock(&c->btree_trans_lock);
 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
@@ -3150,18 +3149,12 @@ struct btree_trans *__bch2_trans_get(struct bch_fs *c, unsigned fn_idx)
 			BUG_ON(pos_task &&
 			       pid == pos_task->pid &&
 			       pos->locked);
-
-			if (pos_task && pid < pos_task->pid) {
-				list_add_tail(&trans->list, &pos->list);
-				goto list_add_done;
-			}
 		}
 	}
-	list_add_tail(&trans->list, &c->btree_trans_list);
-list_add_done:
+
+	list_add(&trans->list, &c->btree_trans_list);
 	seqmutex_unlock(&c->btree_trans_lock);
 got_trans:
-	trans->ref.closure_get_happened = false;
 	trans->c		= c;
 	trans->last_begin_time	= local_clock();
 	trans->fn_idx		= fn_idx;
@@ -3200,6 +3193,8 @@ got_trans:
 	trans->srcu_idx		= srcu_read_lock(&c->btree_trans_barrier);
 	trans->srcu_lock_time	= jiffies;
 	trans->srcu_held	= true;
+
+	closure_init_stack_release(&trans->ref);
 	return trans;
 }
 
@@ -3257,10 +3252,10 @@ void bch2_trans_put(struct btree_trans *trans)
 		bch2_journal_keys_put(c);
 
 	/*
-	 * trans->ref protects trans->locking_wait.task, btree_paths arary; used
+	 * trans->ref protects trans->locking_wait.task, btree_paths array; used
 	 * by cycle detector
 	 */
-	closure_sync(&trans->ref);
+	closure_return_sync(&trans->ref);
 	trans->locking_wait.task = NULL;
 
 	unsigned long *paths_allocated = trans->paths_allocated;
@@ -3385,8 +3380,6 @@ void bch2_fs_btree_iter_exit(struct bch_fs *c)
 				per_cpu_ptr(c->btree_trans_bufs, cpu)->trans;
 
 			if (trans) {
-				closure_sync(&trans->ref);
-
 				seqmutex_lock(&c->btree_trans_lock);
 				list_del(&trans->list);
 				seqmutex_unlock(&c->btree_trans_lock);
diff --git a/fs/bcachefs/chardev.c b/fs/bcachefs/chardev.c
index 9e54323f0f5f..6d82e1165adc 100644
--- a/fs/bcachefs/chardev.c
+++ b/fs/bcachefs/chardev.c
@@ -216,7 +216,8 @@ static long bch2_ioctl_fsck_offline(struct bch_ioctl_fsck_offline __user *user_a
 
 		ret =   PTR_ERR_OR_ZERO(optstr) ?:
 			bch2_parse_mount_opts(NULL, &thr->opts, optstr);
-		kfree(optstr);
+		if (!IS_ERR(optstr))
+			kfree(optstr);
 
 		if (ret)
 			goto err;
@@ -319,7 +320,8 @@ static long bch2_ioctl_disk_add(struct bch_fs *c, struct bch_ioctl_disk arg)
 		return ret;
 
 	ret = bch2_dev_add(c, path);
-	kfree(path);
+	if (!IS_ERR(path))
+		kfree(path);
 
 	return ret;
 }
@@ -850,7 +852,8 @@ static long bch2_ioctl_fsck_online(struct bch_fs *c,
 
 		ret =   PTR_ERR_OR_ZERO(optstr) ?:
 			bch2_parse_mount_opts(c, &thr->opts, optstr);
-		kfree(optstr);
+		if (!IS_ERR(optstr))
+			kfree(optstr);
 
 		if (ret)
 			goto err;
diff --git a/fs/bcachefs/debug.c b/fs/bcachefs/debug.c
index 51cbf3928361..f0d4727c4dc2 100644
--- a/fs/bcachefs/debug.c
+++ b/fs/bcachefs/debug.c
@@ -568,6 +568,32 @@ static const struct file_operations cached_btree_nodes_ops = {
 	.read		= bch2_cached_btree_nodes_read,
 };
 
+typedef int (*list_cmp_fn)(const struct list_head *l, const struct list_head *r);
+
+static void list_sort(struct list_head *head, list_cmp_fn cmp)
+{
+	struct list_head *pos;
+
+	list_for_each(pos, head)
+		while (!list_is_last(pos, head) &&
+		       cmp(pos, pos->next) > 0) {
+			struct list_head *pos2, *next = pos->next;
+
+			list_del(next);
+			list_for_each(pos2, head)
+				if (cmp(next, pos2) < 0)
+					goto pos_found;
+			BUG();
+pos_found:
+			list_add_tail(next, pos2);
+		}
+}
+
+static int list_ptr_order_cmp(const struct list_head *l, const struct list_head *r)
+{
+	return cmp_int(l, r);
+}
+
 static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
 					    size_t size, loff_t *ppos)
 {
@@ -575,41 +601,39 @@ static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
 	struct bch_fs *c = i->c;
 	struct btree_trans *trans;
 	ssize_t ret = 0;
-	u32 seq;
 
 	i->ubuf = buf;
 	i->size	= size;
 	i->ret	= 0;
 restart:
 	seqmutex_lock(&c->btree_trans_lock);
-	list_for_each_entry(trans, &c->btree_trans_list, list) {
-		struct task_struct *task = READ_ONCE(trans->locking_wait.task);
+	list_sort(&c->btree_trans_list, list_ptr_order_cmp);
 
-		if (!task || task->pid <= i->iter)
+	list_for_each_entry(trans, &c->btree_trans_list, list) {
+		if ((ulong) trans < i->iter)
 			continue;
 
-		closure_get(&trans->ref);
-		seq = seqmutex_seq(&c->btree_trans_lock);
-		seqmutex_unlock(&c->btree_trans_lock);
+		i->iter = (ulong) trans;
 
-		ret = flush_buf(i);
-		if (ret) {
-			closure_put(&trans->ref);
-			goto unlocked;
-		}
+		if (!closure_get_not_zero(&trans->ref))
+			continue;
+
+		u32 seq = seqmutex_unlock(&c->btree_trans_lock);
 
 		bch2_btree_trans_to_text(&i->buf, trans);
 
 		prt_printf(&i->buf, "backtrace:\n");
 		printbuf_indent_add(&i->buf, 2);
-		bch2_prt_task_backtrace(&i->buf, task, 0, GFP_KERNEL);
+		bch2_prt_task_backtrace(&i->buf, trans->locking_wait.task, 0, GFP_KERNEL);
 		printbuf_indent_sub(&i->buf, 2);
 		prt_newline(&i->buf);
 
-		i->iter = task->pid;
-
 		closure_put(&trans->ref);
 
+		ret = flush_buf(i);
+		if (ret)
+			goto unlocked;
+
 		if (!seqmutex_relock(&c->btree_trans_lock, seq))
 			goto restart;
 	}
@@ -804,50 +828,55 @@ static const struct file_operations btree_transaction_stats_op = {
 	.read		= btree_transaction_stats_read,
 };
 
-static ssize_t bch2_btree_deadlock_read(struct file *file, char __user *buf,
-					    size_t size, loff_t *ppos)
+/* walk btree transactions until we find a deadlock and print it */
+static void btree_deadlock_to_text(struct printbuf *out, struct bch_fs *c)
 {
-	struct dump_iter *i = file->private_data;
-	struct bch_fs *c = i->c;
 	struct btree_trans *trans;
-	ssize_t ret = 0;
-	u32 seq;
-
-	i->ubuf = buf;
-	i->size	= size;
-	i->ret	= 0;
-
-	if (i->iter)
-		goto out;
+	pid_t iter = 0;
 restart:
 	seqmutex_lock(&c->btree_trans_lock);
 	list_for_each_entry(trans, &c->btree_trans_list, list) {
 		struct task_struct *task = READ_ONCE(trans->locking_wait.task);
 
-		if (!task || task->pid <= i->iter)
+		if (!task || task->pid <= iter)
 			continue;
 
-		closure_get(&trans->ref);
-		seq = seqmutex_seq(&c->btree_trans_lock);
-		seqmutex_unlock(&c->btree_trans_lock);
+		iter = task->pid;
 
-		ret = flush_buf(i);
-		if (ret) {
-			closure_put(&trans->ref);
-			goto out;
-		}
+		if (!closure_get_not_zero(&trans->ref))
+			continue;
 
-		bch2_check_for_deadlock(trans, &i->buf);
+		u32 seq = seqmutex_unlock(&c->btree_trans_lock);
 
-		i->iter = task->pid;
+		bool found = bch2_check_for_deadlock(trans, out) != 0;
 
 		closure_put(&trans->ref);
 
+		if (found)
+			return;
+
 		if (!seqmutex_relock(&c->btree_trans_lock, seq))
 			goto restart;
 	}
 	seqmutex_unlock(&c->btree_trans_lock);
-out:
+}
+
+static ssize_t bch2_btree_deadlock_read(struct file *file, char __user *buf,
+					    size_t size, loff_t *ppos)
+{
+	struct dump_iter *i = file->private_data;
+	struct bch_fs *c = i->c;
+	ssize_t ret = 0;
+
+	i->ubuf = buf;
+	i->size	= size;
+	i->ret	= 0;
+
+	if (!i->iter) {
+		btree_deadlock_to_text(&i->buf, c);
+		i->iter++;
+	}
+
 	if (i->buf.allocation_failure)
 		ret = -ENOMEM;
 
diff --git a/fs/bcachefs/journal.c b/fs/bcachefs/journal.c
index dac2f498ae8b..13669dd0e375 100644
--- a/fs/bcachefs/journal.c
+++ b/fs/bcachefs/journal.c
@@ -1521,6 +1521,11 @@ bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64
 	struct journal_entry_pin *pin;
 
 	spin_lock(&j->lock);
+	if (!test_bit(JOURNAL_running, &j->flags)) {
+		spin_unlock(&j->lock);
+		return true;
+	}
+
 	*seq = max(*seq, j->pin.front);
 
 	if (*seq >= j->pin.back) {
diff --git a/fs/bcachefs/journal_io.c b/fs/bcachefs/journal_io.c
index 492426c8d869..db24ce21b2ac 100644
--- a/fs/bcachefs/journal_io.c
+++ b/fs/bcachefs/journal_io.c
@@ -1677,6 +1677,13 @@ static CLOSURE_CALLBACK(journal_write_done)
 		mod_delayed_work(j->wq, &j->write_work, max(0L, delta));
 	}
 
+	/*
+	 * We don't typically trigger journal writes from her - the next journal
+	 * write will be triggered immediately after the previous one is
+	 * allocated, in bch2_journal_write() - but the journal write error path
+	 * is special:
+	 */
+	bch2_journal_do_writes(j);
 	spin_unlock(&j->lock);
 }
 
diff --git a/fs/bcachefs/journal_seq_blacklist.c b/fs/bcachefs/journal_seq_blacklist.c
index ed4846709611..1f25c111c54c 100644
--- a/fs/bcachefs/journal_seq_blacklist.c
+++ b/fs/bcachefs/journal_seq_blacklist.c
@@ -232,7 +232,7 @@ bool bch2_blacklist_entries_gc(struct bch_fs *c)
 	BUG_ON(nr != t->nr);
 
 	unsigned i;
-	for (src = bl->start, i = eytzinger0_first(t->nr);
+	for (src = bl->start, i = t->nr == 0 ? 0 : eytzinger0_first(t->nr);
 	     src < bl->start + nr;
 	     src++, i = eytzinger0_next(i, nr)) {
 		BUG_ON(t->entries[i].start	!= le64_to_cpu(src->start));
diff --git a/fs/bcachefs/sb-errors.c b/fs/bcachefs/sb-errors.c
index bda33e59e226..c1270d790e43 100644
--- a/fs/bcachefs/sb-errors.c
+++ b/fs/bcachefs/sb-errors.c
@@ -110,19 +110,25 @@ out:
 void bch2_sb_errors_from_cpu(struct bch_fs *c)
 {
 	bch_sb_errors_cpu *src = &c->fsck_error_counts;
-	struct bch_sb_field_errors *dst =
-		bch2_sb_field_resize(&c->disk_sb, errors,
-				     bch2_sb_field_errors_u64s(src->nr));
+	struct bch_sb_field_errors *dst;
 	unsigned i;
 
+	mutex_lock(&c->fsck_error_counts_lock);
+
+	dst = bch2_sb_field_resize(&c->disk_sb, errors,
+				   bch2_sb_field_errors_u64s(src->nr));
+
 	if (!dst)
-		return;
+		goto err;
 
 	for (i = 0; i < src->nr; i++) {
 		SET_BCH_SB_ERROR_ENTRY_ID(&dst->entries[i], src->data[i].id);
 		SET_BCH_SB_ERROR_ENTRY_NR(&dst->entries[i], src->data[i].nr);
 		dst->entries[i].last_error_time = cpu_to_le64(src->data[i].last_error_time);
 	}
+
+err:
+	mutex_unlock(&c->fsck_error_counts_lock);
 }
 
 static int bch2_sb_errors_to_cpu(struct bch_fs *c)
diff --git a/fs/bcachefs/seqmutex.h b/fs/bcachefs/seqmutex.h
index c1860d8163fb..c4b3d8d3f414 100644
--- a/fs/bcachefs/seqmutex.h
+++ b/fs/bcachefs/seqmutex.h
@@ -19,17 +19,14 @@ static inline bool seqmutex_trylock(struct seqmutex *lock)
 static inline void seqmutex_lock(struct seqmutex *lock)
 {
 	mutex_lock(&lock->lock);
-}
-
-static inline void seqmutex_unlock(struct seqmutex *lock)
-{
 	lock->seq++;
-	mutex_unlock(&lock->lock);
 }
 
-static inline u32 seqmutex_seq(struct seqmutex *lock)
+static inline u32 seqmutex_unlock(struct seqmutex *lock)
 {
-	return lock->seq;
+	u32 seq = lock->seq;
+	mutex_unlock(&lock->lock);
+	return seq;
 }
 
 static inline bool seqmutex_relock(struct seqmutex *lock, u32 seq)
diff --git a/fs/bcachefs/snapshot.c b/fs/bcachefs/snapshot.c
index 4ef98e696673..24023d6a9698 100644
--- a/fs/bcachefs/snapshot.c
+++ b/fs/bcachefs/snapshot.c
@@ -168,6 +168,9 @@ static noinline struct snapshot_t *__snapshot_t_mut(struct bch_fs *c, u32 id)
 	size_t new_bytes = kmalloc_size_roundup(struct_size(new, s, idx + 1));
 	size_t new_size = (new_bytes - sizeof(*new)) / sizeof(new->s[0]);
 
+	if (unlikely(new_bytes > INT_MAX))
+		return NULL;
+
 	new = kvzalloc(new_bytes, GFP_KERNEL);
 	if (!new)
 		return NULL;
diff --git a/fs/bcachefs/super.c b/fs/bcachefs/super.c
index 9083df82073a..fb906467201e 100644
--- a/fs/bcachefs/super.c
+++ b/fs/bcachefs/super.c
@@ -536,7 +536,6 @@ static void __bch2_fs_free(struct bch_fs *c)
 
 	bch2_find_btree_nodes_exit(&c->found_btree_nodes);
 	bch2_free_pending_node_rewrites(c);
-	bch2_fs_allocator_background_exit(c);
 	bch2_fs_sb_errors_exit(c);
 	bch2_fs_counters_exit(c);
 	bch2_fs_snapshots_exit(c);
@@ -1195,6 +1194,7 @@ static void bch2_dev_free(struct bch_dev *ca)
 
 	kfree(ca->buckets_nouse);
 	bch2_free_super(&ca->disk_sb);
+	bch2_dev_allocator_background_exit(ca);
 	bch2_dev_journal_exit(ca);
 
 	free_percpu(ca->io_done);
@@ -1317,6 +1317,8 @@ static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
 	atomic_long_set(&ca->ref, 1);
 #endif
 
+	bch2_dev_allocator_background_init(ca);
+
 	if (percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
 	    !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
@@ -1529,6 +1531,7 @@ static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
 	 * The allocator thread itself allocates btree nodes, so stop it first:
 	 */
 	bch2_dev_allocator_remove(c, ca);
+	bch2_recalc_capacity(c);
 	bch2_dev_journal_stop(&c->journal, ca);
 }
 
@@ -1540,6 +1543,7 @@ static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
 
 	bch2_dev_allocator_add(c, ca);
 	bch2_recalc_capacity(c);
+	bch2_dev_do_discards(ca);
 }
 
 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,