Merge tag 'bcachefs-2024-08-08' of git://evilpiepirate.org/bcachefs

Pull bcachefs fixes from Kent Overstreet:
 "Assorted little stuff:

   - lockdep fixup for lockdep_set_notrack_class()

   - we can now remove a device when using erasure coding without
     deadlocking, though we still hit other issues

   - the 'allocator stuck' timeout is now configurable, and messages are
     ratelimited. The default timeout has been increased from 10 seconds
     to 30"

* tag 'bcachefs-2024-08-08' of git://evilpiepirate.org/bcachefs:
  bcachefs: Use bch2_wait_on_allocator() in btree node alloc path
  bcachefs: Make allocator stuck timeout configurable, ratelimit messages
  bcachefs: Add missing path_traverse() to btree_iter_next_node()
  bcachefs: ec should not allocate from ro devs
  bcachefs: Improved allocator debugging for ec
  bcachefs: Add missing bch2_trans_begin() call
  bcachefs: Add a comment for bucket helper types
  bcachefs: Don't rely on implicit unsigned -> signed integer conversion
  lockdep: Fix lockdep_set_notrack_class() for CONFIG_LOCK_STAT
  bcachefs: Fix double free of ca->buckets_nouse

fs/bcachefs/alloc_background.h

@@ -82,6 +82,14 @@ static inline bool bucket_data_type_mismatch(enum bch_data_type bucket,
 		bucket_data_type(bucket) != bucket_data_type(ptr);
 }
 
+/*
+ * It is my general preference to use unsigned types for unsigned quantities -
+ * however, these helpers are used in disk accounting calculations run by
+ * triggers where the output will be negated and added to an s64. unsigned is
+ * right out even though all these quantities will fit in 32 bits, since it
+ * won't be sign extended correctly; u64 will negate "correctly", but s64 is the
+ * simpler option here.
+ */
 static inline s64 bch2_bucket_sectors_total(struct bch_alloc_v4 a)
 {
 	return a.stripe_sectors + a.dirty_sectors + a.cached_sectors;
@@ -166,8 +174,8 @@ static inline u64 alloc_lru_idx_fragmentation(struct bch_alloc_v4 a,
 	 * avoid overflowing LRU_TIME_BITS on a corrupted fs, when
 	 * bucket_sectors_dirty is (much) bigger than bucket_size
 	 */
-	u64 d = min(bch2_bucket_sectors_dirty(a),
+	u64 d = min_t(s64, bch2_bucket_sectors_dirty(a),
-		    ca->mi.bucket_size);
+		      ca->mi.bucket_size);
 
 	return div_u64(d * (1ULL << 31), ca->mi.bucket_size);
 }

fs/bcachefs/alloc_foreground.c

@@ -1603,7 +1603,8 @@ void bch2_open_bucket_to_text(struct printbuf *out, struct bch_fs *c, struct ope
 	prt_newline(out);
 }
 
-void bch2_open_buckets_to_text(struct printbuf *out, struct bch_fs *c)
+void bch2_open_buckets_to_text(struct printbuf *out, struct bch_fs *c,
+			       struct bch_dev *ca)
 {
 	struct open_bucket *ob;
 
@@ -1613,7 +1614,8 @@ void bch2_open_buckets_to_text(struct printbuf *out, struct bch_fs *c)
 	     ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
 	     ob++) {
 		spin_lock(&ob->lock);
-		if (ob->valid && !ob->on_partial_list)
+		if (ob->valid && !ob->on_partial_list &&
+		    (!ca || ob->dev == ca->dev_idx))
 			bch2_open_bucket_to_text(out, c, ob);
 		spin_unlock(&ob->lock);
 	}
@@ -1756,11 +1758,12 @@ void bch2_dev_alloc_debug_to_text(struct printbuf *out, struct bch_dev *ca)
 	prt_printf(out, "buckets to invalidate\t%llu\r\n",	should_invalidate_buckets(ca, stats));
 }
 
-void bch2_print_allocator_stuck(struct bch_fs *c)
+static noinline void bch2_print_allocator_stuck(struct bch_fs *c)
 {
 	struct printbuf buf = PRINTBUF;
 
-	prt_printf(&buf, "Allocator stuck? Waited for 10 seconds\n");
+	prt_printf(&buf, "Allocator stuck? Waited for %u seconds\n",
+		   c->opts.allocator_stuck_timeout);
 
 	prt_printf(&buf, "Allocator debug:\n");
 	printbuf_indent_add(&buf, 2);
@@ -1790,3 +1793,24 @@ void bch2_print_allocator_stuck(struct bch_fs *c)
 	bch2_print_string_as_lines(KERN_ERR, buf.buf);
 	printbuf_exit(&buf);
 }
+
+static inline unsigned allocator_wait_timeout(struct bch_fs *c)
+{
+	if (c->allocator_last_stuck &&
+	    time_after(c->allocator_last_stuck + HZ * 60 * 2, jiffies))
+		return 0;
+
+	return c->opts.allocator_stuck_timeout * HZ;
+}
+
+void __bch2_wait_on_allocator(struct bch_fs *c, struct closure *cl)
+{
+	unsigned t = allocator_wait_timeout(c);
+
+	if (t && closure_sync_timeout(cl, t)) {
+		c->allocator_last_stuck = jiffies;
+		bch2_print_allocator_stuck(c);
+	}
+
+	closure_sync(cl);
+}

fs/bcachefs/alloc_foreground.h

@@ -223,7 +223,7 @@ static inline struct write_point_specifier writepoint_ptr(struct write_point *wp
 void bch2_fs_allocator_foreground_init(struct bch_fs *);
 
 void bch2_open_bucket_to_text(struct printbuf *, struct bch_fs *, struct open_bucket *);
-void bch2_open_buckets_to_text(struct printbuf *, struct bch_fs *);
+void bch2_open_buckets_to_text(struct printbuf *, struct bch_fs *, struct bch_dev *);
 void bch2_open_buckets_partial_to_text(struct printbuf *, struct bch_fs *);
 
 void bch2_write_points_to_text(struct printbuf *, struct bch_fs *);
@@ -231,6 +231,11 @@ void bch2_write_points_to_text(struct printbuf *, struct bch_fs *);
 void bch2_fs_alloc_debug_to_text(struct printbuf *, struct bch_fs *);
 void bch2_dev_alloc_debug_to_text(struct printbuf *, struct bch_dev *);
 
-void bch2_print_allocator_stuck(struct bch_fs *);
+void __bch2_wait_on_allocator(struct bch_fs *, struct closure *);
+static inline void bch2_wait_on_allocator(struct bch_fs *c, struct closure *cl)
+{
+	if (cl->closure_get_happened)
+		__bch2_wait_on_allocator(c, cl);
+}
 
 #endif /* _BCACHEFS_ALLOC_FOREGROUND_H */

fs/bcachefs/bcachefs.h

@@ -893,6 +893,8 @@ struct bch_fs {
 	struct bch_fs_usage_base __percpu *usage;
 	u64 __percpu		*online_reserved;
 
+	unsigned long		allocator_last_stuck;
+
 	struct io_clock		io_clock[2];
 
 	/* JOURNAL SEQ BLACKLIST */

fs/bcachefs/bcachefs_format.h

@@ -836,6 +836,8 @@ LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
 
 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
 					struct bch_sb, flags[5],  0, 16);
+LE64_BITMASK(BCH_SB_ALLOCATOR_STUCK_TIMEOUT,
+					struct bch_sb, flags[5], 16, 32);
 
 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
 {

fs/bcachefs/btree_iter.c

@@ -1921,6 +1921,11 @@ struct btree *bch2_btree_iter_next_node(struct btree_iter *iter)
 	bch2_trans_verify_not_in_restart(trans);
 	bch2_btree_iter_verify(iter);
 
+	ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
+	if (ret)
+		goto err;
+
+
 	struct btree_path *path = btree_iter_path(trans, iter);
 
 	/* already at end? */

fs/bcachefs/btree_update_interior.c

@@ -1264,7 +1264,7 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 			ret = bch2_btree_reserve_get(trans, as, nr_nodes, flags, &cl);
 
 			bch2_trans_unlock(trans);
-			closure_sync(&cl);
+			bch2_wait_on_allocator(c, &cl);
 		} while (bch2_err_matches(ret, BCH_ERR_operation_blocked));
 	}
 

fs/bcachefs/ec.c

@@ -1809,6 +1809,9 @@ static int new_stripe_alloc_buckets(struct btree_trans *trans, struct ec_stripe_
 	BUG_ON(v->nr_blocks	!= h->s->nr_data + h->s->nr_parity);
 	BUG_ON(v->nr_redundant	!= h->s->nr_parity);
 
+	/* * We bypass the sector allocator which normally does this: */
+	bitmap_and(devs.d, devs.d, c->rw_devs[BCH_DATA_user].d, BCH_SB_MEMBERS_MAX);
+
 	for_each_set_bit(i, h->s->blocks_gotten, v->nr_blocks) {
 		__clear_bit(v->ptrs[i].dev, devs.d);
 		if (i < h->s->nr_data)
@@ -2235,6 +2238,23 @@ void bch2_stripes_heap_to_text(struct printbuf *out, struct bch_fs *c)
 	mutex_unlock(&c->ec_stripes_heap_lock);
 }
 
+static void bch2_new_stripe_to_text(struct printbuf *out, struct bch_fs *c,
+				    struct ec_stripe_new *s)
+{
+	prt_printf(out, "\tidx %llu blocks %u+%u allocated %u ref %u %u %s obs",
+		   s->idx, s->nr_data, s->nr_parity,
+		   bitmap_weight(s->blocks_allocated, s->nr_data),
+		   atomic_read(&s->ref[STRIPE_REF_io]),
+		   atomic_read(&s->ref[STRIPE_REF_stripe]),
+		   bch2_watermarks[s->h->watermark]);
+
+	struct bch_stripe *v = &bkey_i_to_stripe(&s->new_stripe.key)->v;
+	unsigned i;
+	for_each_set_bit(i, s->blocks_gotten, v->nr_blocks)
+		prt_printf(out, " %u", s->blocks[i]);
+	prt_newline(out);
+}
+
 void bch2_new_stripes_to_text(struct printbuf *out, struct bch_fs *c)
 {
 	struct ec_stripe_head *h;
@@ -2247,23 +2267,15 @@ void bch2_new_stripes_to_text(struct printbuf *out, struct bch_fs *c)
 		       bch2_watermarks[h->watermark]);
 
 		if (h->s)
-			prt_printf(out, "\tidx %llu blocks %u+%u allocated %u\n",
+			bch2_new_stripe_to_text(out, c, h->s);
-			       h->s->idx, h->s->nr_data, h->s->nr_parity,
-			       bitmap_weight(h->s->blocks_allocated,
-					     h->s->nr_data));
 	}
 	mutex_unlock(&c->ec_stripe_head_lock);
 
 	prt_printf(out, "in flight:\n");
 
 	mutex_lock(&c->ec_stripe_new_lock);
-	list_for_each_entry(s, &c->ec_stripe_new_list, list) {
+	list_for_each_entry(s, &c->ec_stripe_new_list, list)
-		prt_printf(out, "\tidx %llu blocks %u+%u ref %u %u %s\n",
+		bch2_new_stripe_to_text(out, c, s);
-			   s->idx, s->nr_data, s->nr_parity,
-			   atomic_read(&s->ref[STRIPE_REF_io]),
-			   atomic_read(&s->ref[STRIPE_REF_stripe]),
-			   bch2_watermarks[s->h->watermark]);
-	}
 	mutex_unlock(&c->ec_stripe_new_lock);
 }
 

fs/bcachefs/io_misc.c

@@ -126,11 +126,7 @@ int bch2_extent_fallocate(struct btree_trans *trans,
 
 	if (closure_nr_remaining(&cl) != 1) {
 		bch2_trans_unlock_long(trans);
-
+		bch2_wait_on_allocator(c, &cl);
-		if (closure_sync_timeout(&cl, HZ * 10)) {
-			bch2_print_allocator_stuck(c);
-			closure_sync(&cl);
-		}
 	}
 
 	return ret;

fs/bcachefs/io_read.c

@@ -406,6 +406,7 @@ static void bch2_read_retry_nodecode(struct bch_fs *c, struct bch_read_bio *rbio
 	bch2_trans_iter_init(trans, &iter, rbio->data_btree,
 			     rbio->read_pos, BTREE_ITER_slots);
 retry:
+	bch2_trans_begin(trans);
 	rbio->bio.bi_status = 0;
 
 	k = bch2_btree_iter_peek_slot(&iter);

fs/bcachefs/io_write.c

@@ -1503,10 +1503,7 @@ static void __bch2_write(struct bch_write_op *op)
 	if ((op->flags & BCH_WRITE_SYNC) ||
 	    (!(op->flags & BCH_WRITE_SUBMITTED) &&
 	     !(op->flags & BCH_WRITE_IN_WORKER))) {
-		if (closure_sync_timeout(&op->cl, HZ * 10)) {
+		bch2_wait_on_allocator(c, &op->cl);
-			bch2_print_allocator_stuck(c);
-			closure_sync(&op->cl);
-		}
 
 		__bch2_write_index(op);
 

fs/bcachefs/opts.h

@@ -391,6 +391,11 @@ enum fsck_err_opts {
 	  OPT_BOOL(),							\
 	  BCH_SB_JOURNAL_TRANSACTION_NAMES, true,			\
 	  NULL,		"Log transaction function names in journal")	\
+	x(allocator_stuck_timeout,	u16,				\
+	  OPT_FS|OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME,			\
+	  OPT_UINT(0, U16_MAX),						\
+	  BCH_SB_ALLOCATOR_STUCK_TIMEOUT, 30,				\
+	  NULL,		"Default timeout in seconds for stuck allocator messages")\
 	x(noexcl,			u8,				\
 	  OPT_FS|OPT_MOUNT,						\
 	  OPT_BOOL(),							\

fs/bcachefs/super-io.c

@@ -414,6 +414,10 @@ static int bch2_sb_validate(struct bch_sb_handle *disk_sb,
 
 		if (!BCH_SB_VERSION_UPGRADE_COMPLETE(sb))
 			SET_BCH_SB_VERSION_UPGRADE_COMPLETE(sb, le16_to_cpu(sb->version));
+
+		if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_disk_accounting_v2 &&
+		    !BCH_SB_ALLOCATOR_STUCK_TIMEOUT(sb))
+			SET_BCH_SB_ALLOCATOR_STUCK_TIMEOUT(sb, 30);
 	}
 
 	for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {

fs/bcachefs/super.c

@@ -1193,7 +1193,6 @@ static void bch2_dev_free(struct bch_dev *ca)
 	if (ca->kobj.state_in_sysfs)
 		kobject_del(&ca->kobj);
 
-	kfree(ca->buckets_nouse);
 	bch2_free_super(&ca->disk_sb);
 	bch2_dev_allocator_background_exit(ca);
 	bch2_dev_journal_exit(ca);

fs/bcachefs/sysfs.c

@@ -367,7 +367,7 @@ SHOW(bch2_fs)
 		bch2_stripes_heap_to_text(out, c);
 
 	if (attr == &sysfs_open_buckets)
-		bch2_open_buckets_to_text(out, c);
+		bch2_open_buckets_to_text(out, c, NULL);
 
 	if (attr == &sysfs_open_buckets_partial)
 		bch2_open_buckets_partial_to_text(out, c);
@@ -811,6 +811,9 @@ SHOW(bch2_dev)
 	if (attr == &sysfs_alloc_debug)
 		bch2_dev_alloc_debug_to_text(out, ca);
 
+	if (attr == &sysfs_open_buckets)
+		bch2_open_buckets_to_text(out, c, ca);
+
 	return 0;
 }
 
@@ -892,6 +895,7 @@ struct attribute *bch2_dev_files[] = {
 
 	/* debug: */
 	&sysfs_alloc_debug,
+	&sysfs_open_buckets,
 	NULL
 };
 

kernel/locking/lockdep.c

@@ -5936,6 +5936,9 @@ __lock_contended(struct lockdep_map *lock, unsigned long ip)
 	if (DEBUG_LOCKS_WARN_ON(!depth))
 		return;
 
+	if (unlikely(lock->key == &__lockdep_no_track__))
+		return;
+
 	hlock = find_held_lock(curr, lock, depth, &i);
 	if (!hlock) {
 		print_lock_contention_bug(curr, lock, ip);
@@ -5978,6 +5981,9 @@ __lock_acquired(struct lockdep_map *lock, unsigned long ip)
 	if (DEBUG_LOCKS_WARN_ON(!depth))
 		return;
 
+	if (unlikely(lock->key == &__lockdep_no_track__))
+		return;
+
 	hlock = find_held_lock(curr, lock, depth, &i);
 	if (!hlock) {
 		print_lock_contention_bug(curr, lock, _RET_IP_);