Update bcachefs sources to efd3df255ba5 bcachefs: Btree node reads no longer kick off rewrites for degraded nodes

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

.bcachefs_revision

@@ -1 +1 @@
-fba763d22acfb4feaacc45e88803f8b90c9740aa
+efd3df255ba56d795750510e79d8d79f7812a029

libbcachefs/bcachefs.h

@@ -350,12 +350,14 @@ do {									\
 #define bch_verbose(c, ...)		bch_log(c, KERN_DEBUG, __VA_ARGS__)
 #define bch_verbose_ratelimited(c, ...)	bch_log_ratelimited(c, KERN_DEBUG, __VA_ARGS__)
 
-#define bch_info_dev(ca, fmt, ...) \
+#define bch_dev_log(ca, loglevel, fmt, ...) \
-	bch2_print(c, KERN_INFO bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
+	bch2_print(ca->fs, loglevel bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
-#define bch_notice_dev(ca, fmt, ...) \
+
-	bch2_print(c, KERN_NOTICE bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
+#define bch_err_dev(ca, ...)		bch_dev_log(ca, KERN_ERR, __VA_ARGS__)
-#define bch_err_dev(ca, fmt, ...) \
+#define bch_notice_dev(ca, ...)		bch_dev_log(ca, KERN_NOTICE, __VA_ARGS__)
-	bch2_print(c, KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
+#define bch_info_dev(ca, ...)		bch_dev_log(ca, KERN_INFO, __VA_ARGS__)
+#define bch_verbose_dev(ca, ...)	bch_dev_log(ca, KERN_DEBUG, __VA_ARGS__)
+
 #define bch_err_dev_offset(ca, _offset, fmt, ...) \
 	bch2_print(c, KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
 #define bch_err_inum(c, _inum, fmt, ...) \

libbcachefs/btree/read.c

@@ -623,6 +623,20 @@ fsck_err:
 	return ret;
 }
 
+static bool btree_node_degraded(struct bch_fs *c, struct btree *b)
+{
+	guard(rcu)();
+	bkey_for_each_ptr(bch2_bkey_ptrs(bkey_i_to_s(&b->key)), ptr) {
+		if (ptr->dev == BCH_SB_MEMBER_INVALID)
+			continue;
+
+		struct bch_dev *ca = bch2_dev_rcu(c, ptr->dev);
+		if (!ca || ca->mi.state != BCH_MEMBER_STATE_rw)
+			return true;
+	}
+	return false;
+}
+
 int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
 			      struct btree *b,
 			      struct bch_io_failures *failed,
@@ -912,43 +926,6 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
 	if (updated_range)
 		bch2_btree_node_drop_keys_outside_node(b);
 
-	/*
-	 * XXX:
-	 *
-	 * We deadlock if too many btree updates require node rewrites while
-	 * we're still in journal replay.
-	 *
-	 * This is because btree node rewrites generate more updates for the
-	 * interior updates (alloc, backpointers), and if those updates touch
-	 * new nodes and generate more rewrites - well, you see the problem.
-	 *
-	 * The biggest cause is that we don't use the btree write buffer (for
-	 * the backpointer updates - this needs some real thought on locking in
-	 * order to fix.
-	 *
-	 * The problem with this workaround (not doing the rewrite for degraded
-	 * nodes in journal replay) is that those degraded nodes persist, and we
-	 * don't want that (this is a real bug when a btree node write completes
-	 * with fewer replicas than we wanted and leaves a degraded node due to
-	 * device _removal_, i.e. the device went away mid write).
-	 *
-	 * It's less of a bug here, but still a problem because we don't yet
-	 * have a way of tracking degraded data - we another index (all
-	 * extents/btree nodes, by replicas entry) in order to fix properly
-	 * (re-replicate degraded data at the earliest possible time).
-	 */
-	if (c->recovery.passes_complete & BIT_ULL(BCH_RECOVERY_PASS_journal_replay)) {
-		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);
-					set_btree_node_need_rewrite_degraded(b);
-				}
-			}
-	}
-
 	if (!ptr_written) {
 		set_btree_node_need_rewrite(b);
 		set_btree_node_need_rewrite_ptr_written_zero(b);
@@ -1052,6 +1029,16 @@ start:
 	if (ret || failed.nr)
 		bch2_print_str_ratelimited(c, KERN_ERR, buf.buf);
 
+	/*
+	 * Do this late; unlike other btree_node_need_rewrite() cases if a node
+	 * is merely degraded we should rewrite it before we update it, but we
+	 * don't need to kick off an async rewrite now:
+	 */
+	if (btree_node_degraded(c, b)) {
+		set_btree_node_need_rewrite(b);
+		set_btree_node_need_rewrite_degraded(b);
+	}
+
 	async_object_list_del(c, btree_read_bio, rb->list_idx);
 	bch2_time_stats_update(&c->times[BCH_TIME_btree_node_read],
 			       rb->start_time);

libbcachefs/btree/update.h

@@ -210,8 +210,7 @@ int bch2_btree_write_buffer_insert_err(struct bch_fs *, enum btree_id, struct bk
 static inline int bch2_btree_write_buffer_insert_checks(struct bch_fs *c, enum btree_id btree,
 							struct bkey_i *k)
 {
-	if (unlikely(!btree_type_uses_write_buffer(btree) ||
+	if (unlikely(!btree_type_uses_write_buffer(btree)))
-		     k->k.u64s > BTREE_WRITE_BUFERED_U64s_MAX))
 		try(bch2_btree_write_buffer_insert_err(c, btree, k));
 
 	return 0;

libbcachefs/btree/write_buffer.c

@@ -57,12 +57,14 @@ static inline bool wb_key_ref_cmp(const struct wb_key_ref *l, const struct wb_ke
 #endif
 }
 
-static int wb_key_seq_cmp(const void *_l, const void *_r)
+static int wb_key_seq_cmp(const void *_l, const void *_r, const void *priv)
 {
-	const struct btree_write_buffered_key *l = _l;
+	const struct btree_write_buffer_keys *keys = priv;
-	const struct btree_write_buffered_key *r = _r;
+	const struct wb_key_ref *l = _l;
+	const struct wb_key_ref *r = _r;
 
-	return cmp_int(l->journal_seq, r->journal_seq);
+	return cmp_int(wb_keys_idx(keys, l->idx)->journal_seq,
+		       wb_keys_idx(keys, r->idx)->journal_seq);
 }
 
 /* Compare excluding idx, the low 24 bits: */
@@ -227,7 +229,7 @@ static inline int wb_flush_one(struct btree_trans *trans, struct btree_iter *ite
  */
 static int
 btree_write_buffered_insert(struct btree_trans *trans,
-			  struct btree_write_buffered_key *wb)
+			    struct btree_write_buffered_key *wb)
 {
 	CLASS(btree_iter, iter)(trans, wb->btree, bkey_start_pos(&wb->k.k),
 				BTREE_ITER_cached|BTREE_ITER_intent);
@@ -247,7 +249,7 @@ static void move_keys_from_inc_to_flushing(struct btree_write_buffer *wb)
 	if (!wb->inc.keys.nr)
 		return;
 
-	bch2_journal_pin_add(j, wb->inc.keys.data[0].journal_seq, &wb->flushing.pin,
+	bch2_journal_pin_add(j, wb_keys_start(&wb->inc)->journal_seq, &wb->flushing.pin,
 			     bch2_btree_write_buffer_journal_flush);
 
 	darray_resize(&wb->flushing.keys, min_t(size_t, 1U << 20, wb->flushing.keys.nr + wb->inc.keys.nr));
@@ -276,7 +278,7 @@ out:
 	if (!wb->inc.keys.nr)
 		bch2_journal_pin_drop(j, &wb->inc.pin);
 	else
-		bch2_journal_pin_update(j, wb->inc.keys.data[0].journal_seq, &wb->inc.pin,
+		bch2_journal_pin_update(j, wb_keys_start(&wb->inc)->journal_seq, &wb->inc.pin,
 					bch2_btree_write_buffer_journal_flush);
 
 	if (j->watermark) {
@@ -326,12 +328,15 @@ static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 	u64 start_time = local_clock();
 	u64 nr_flushing = wb->flushing.keys.nr;
 
-	for (size_t i = 0; i < wb->flushing.keys.nr; i++) {
+	wb->sorted.nr = 0;
-		wb->sorted.data[i].idx = i;
+	wb_keys_for_each(&wb->flushing, k) {
-		wb->sorted.data[i].btree = wb->flushing.keys.data[i].btree;
+		struct wb_key_ref *dst = &darray_top(wb->sorted);
-		memcpy(&wb->sorted.data[i].pos, &wb->flushing.keys.data[i].k.k.p, sizeof(struct bpos));
+		wb->sorted.nr++;
+
+		dst->idx	= (u64 *) k - wb->flushing.keys.data;
+		dst->btree	= k->btree;
+		memcpy(&dst->pos, &k->k.k.p, sizeof(struct bpos));
 	}
-	wb->sorted.nr = wb->flushing.keys.nr;
 
 	/*
 	 * We first sort so that we can detect and skip redundant updates, and
@@ -350,7 +355,7 @@ static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 	wb_sort(wb->sorted.data, wb->sorted.nr);
 
 	darray_for_each(wb->sorted, i) {
-		struct btree_write_buffered_key *k = &wb->flushing.keys.data[i->idx];
+		struct btree_write_buffered_key *k = wb_keys_idx(&wb->flushing, i->idx);
 
 		ret = bch2_btree_write_buffer_insert_checks(c, k->btree, &k->k);
 		if (unlikely(ret))
@@ -369,7 +374,7 @@ static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 
 		if (i + 1 < &darray_top(wb->sorted) &&
 		    wb_key_eq(i, i + 1)) {
-			struct btree_write_buffered_key *n = &wb->flushing.keys.data[i[1].idx];
+			struct btree_write_buffered_key *n = wb_keys_idx(&wb->flushing, i[1].idx);
 
 			if (k->k.k.type == KEY_TYPE_accounting &&
 			    n->k.k.type == KEY_TYPE_accounting)
@@ -439,23 +444,25 @@ static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 		 */
 		trace_and_count(c, write_buffer_flush_slowpath, trans, slowpath, wb->flushing.keys.nr);
 
-		sort_nonatomic(wb->flushing.keys.data,
+		sort_r_nonatomic(wb->sorted.data,
-			       wb->flushing.keys.nr,
+				 wb->sorted.nr,
-			       sizeof(wb->flushing.keys.data[0]),
+				 sizeof(wb->sorted.data[0]),
-			       wb_key_seq_cmp, NULL);
+				 wb_key_seq_cmp, NULL,
+				 &wb->flushing);
 
-		darray_for_each(wb->flushing.keys, i) {
+		darray_for_each(wb->sorted, i) {
-			if (!i->journal_seq)
+			struct btree_write_buffered_key *k = wb_keys_idx(&wb->flushing, i->idx);
+			if (!k->journal_seq)
 				continue;
 
 			if (!accounting_replay_done &&
-			    i->k.k.type == KEY_TYPE_accounting) {
+			    k->k.k.type == KEY_TYPE_accounting) {
 				could_not_insert++;
 				continue;
 			}
 
 			if (!could_not_insert)
-				bch2_journal_pin_update(j, i->journal_seq, &wb->flushing.pin,
+				bch2_journal_pin_update(j, k->journal_seq, &wb->flushing.pin,
 							bch2_btree_write_buffer_journal_flush);
 
 			bch2_trans_begin(trans);
@@ -466,11 +473,11 @@ static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 					BCH_TRANS_COMMIT_no_check_rw|
 					BCH_TRANS_COMMIT_no_enospc|
 					BCH_TRANS_COMMIT_no_journal_res ,
-					btree_write_buffered_insert(trans, i));
+					btree_write_buffered_insert(trans, k));
 			if (ret)
 				goto err;
 
-			i->journal_seq = 0;
+			k->journal_seq = 0;
 		}
 
 		/*
@@ -492,12 +499,14 @@ static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 		 * distinct counters touched somehow was very large.
 		 */
 		if (could_not_insert) {
-			struct btree_write_buffered_key *dst = wb->flushing.keys.data;
+			struct btree_write_buffered_key *dst = wb_keys_start(&wb->flushing);
 
-			darray_for_each(wb->flushing.keys, i)
+			wb_keys_for_each_safe(&wb->flushing, i)
-				if (i->journal_seq)
+				if (i->journal_seq) {
-					*dst++ = *i;
+					memmove_u64s_down(dst, i, wb_key_u64s(&i->k));
-			wb->flushing.keys.nr = dst - wb->flushing.keys.data;
+					dst = wb_key_next(dst);
+				}
+			wb->flushing.keys.nr = (u64 *) dst - wb->flushing.keys.data;
 		}
 	}
 err:
@@ -745,9 +754,10 @@ int bch2_journal_key_to_wb_slowpath(struct bch_fs *c,
 			     enum btree_id btree, struct bkey_i *k)
 {
 	struct btree_write_buffer *wb = &c->btree_write_buffer;
+	unsigned u64s = wb_key_u64s(k);
 	int ret;
 retry:
-	ret = darray_make_room_gfp(&dst->wb->keys, 1, GFP_KERNEL);
+	ret = darray_make_room_gfp(&dst->wb->keys, u64s, GFP_KERNEL);
 	if (!ret && dst->wb == &wb->flushing)
 		ret = darray_resize(&wb->sorted, wb->flushing.keys.size);
 
@@ -766,15 +776,10 @@ retry:
 	dst->room = darray_room(dst->wb->keys);
 	if (dst->wb == &wb->flushing)
 		dst->room = min(dst->room, wb->sorted.size - wb->flushing.keys.nr);
-	BUG_ON(!dst->room);
+	BUG_ON(dst->room < u64s);
 	BUG_ON(!dst->seq);
 
-	struct btree_write_buffered_key *wb_k = &darray_top(dst->wb->keys);
+	bch2_journal_key_to_wb_reserved(c, dst, btree, k);
-	wb_k->journal_seq	= dst->seq;
-	wb_k->btree		= btree;
-	bkey_copy(&wb_k->k, k);
-	dst->wb->keys.nr++;
-	dst->room--;
 	return 0;
 }
 

libbcachefs/btree/write_buffer.h

@@ -93,19 +93,63 @@ int bch2_journal_key_to_wb_slowpath(struct bch_fs *,
 			     struct journal_keys_to_wb *,
 			     enum btree_id, struct bkey_i *);
 
+static inline unsigned wb_key_u64s(const struct bkey_i *k)
+{
+	return k->k.u64s + offsetof(struct btree_write_buffered_key, k) / sizeof(u64);
+}
+
+static inline struct btree_write_buffered_key *wb_keys_start(const struct btree_write_buffer_keys *keys)
+{
+	return (struct btree_write_buffered_key *) &darray_first(keys->keys);
+}
+
+static inline struct btree_write_buffered_key *wb_keys_end(const struct btree_write_buffer_keys *keys)
+{
+	return (struct btree_write_buffered_key *) &darray_top(keys->keys);
+}
+
+static inline struct btree_write_buffered_key *wb_keys_idx(const struct btree_write_buffer_keys *keys,
+							   unsigned idx)
+{
+	return (struct btree_write_buffered_key *) &keys->keys.data[idx];
+}
+
+static inline struct btree_write_buffered_key *wb_key_next(const struct btree_write_buffered_key *k)
+{
+	return (struct btree_write_buffered_key *) ((u64 *) k + wb_key_u64s(&k->k));
+}
+
+#define wb_keys_for_each(_keys, _k)							\
+	for (struct btree_write_buffered_key *_k = wb_keys_start(_keys);		\
+	     _k != wb_keys_end(_keys);							\
+	     _k = wb_key_next(_k))
+
+#define wb_keys_for_each_safe(_keys, _k)						\
+	for (struct btree_write_buffered_key *_next, *_k = wb_keys_start(_keys);	\
+	     _k != wb_keys_end(_keys) && (_next = wb_key_next(_k), true);		\
+	     _k = _next)
+
+static inline void bch2_journal_key_to_wb_reserved(struct bch_fs *c,
+			     struct journal_keys_to_wb *dst,
+			     enum btree_id btree, struct bkey_i *k)
+{
+	unsigned u64s = wb_key_u64s(k);
+	struct btree_write_buffered_key *wb_k = wb_keys_end(dst->wb);
+	wb_k->journal_seq	= dst->seq;
+	wb_k->btree		= btree;
+	bkey_copy(&wb_k->k, k);
+	dst->wb->keys.nr += u64s;
+	dst->room -= u64s;
+}
+
 static inline int __bch2_journal_key_to_wb(struct bch_fs *c,
 			     struct journal_keys_to_wb *dst,
 			     enum btree_id btree, struct bkey_i *k)
 {
-	if (unlikely(!dst->room))
+	if (unlikely(dst->room < wb_key_u64s(k)))
 		return bch2_journal_key_to_wb_slowpath(c, dst, btree, k);
 
-	struct btree_write_buffered_key *wb_k = &darray_top(dst->wb->keys);
+	bch2_journal_key_to_wb_reserved(c, dst, btree, k);
-	wb_k->journal_seq	= dst->seq;
-	wb_k->btree		= btree;
-	bkey_copy(&wb_k->k, k);
-	dst->wb->keys.nr++;
-	dst->room--;
 	return 0;
 }
 

libbcachefs/btree/write_buffer_types.h

@@ -6,7 +6,6 @@
 #include "journal/types.h"
 
 #define BTREE_WRITE_BUFERED_VAL_U64s_MAX	4
-#define BTREE_WRITE_BUFERED_U64s_MAX	(BKEY_U64s + BTREE_WRITE_BUFERED_VAL_U64s_MAX)
 
 struct wb_key_ref {
 union {
@@ -38,11 +37,13 @@ union {
 struct btree_write_buffered_key {
 	enum btree_id			btree:8;
 	u64				journal_seq:56;
+
+	/* BTREE_WRITE_BUFERED_VAL_U64s_MAX only applies to accounting keys */
 	__BKEY_PADDED(k, BTREE_WRITE_BUFERED_VAL_U64s_MAX);
 };
 
 struct btree_write_buffer_keys {
-	DARRAY(struct btree_write_buffered_key) keys;
+	darray_u64			keys;
 	struct journal_entry_pin	pin;
 	struct mutex			lock;
 };

libbcachefs/data/reconcile.c

@@ -2239,6 +2239,8 @@ static int check_reconcile_work_btrees(struct btree_trans *trans)
 	struct bch_fs *c = trans->c;
 
 	CLASS(disk_reservation, res)(c);
+	struct progress_indicator progress;
+	bch2_progress_init_inner(&progress, c, 0, ~0ULL);
 
 	for (enum btree_id btree = 0; btree < btree_id_nr_alive(c); btree++) {
 		if (!bch2_btree_id_root(c, btree)->b)
@@ -2252,6 +2254,7 @@ static int check_reconcile_work_btrees(struct btree_trans *trans)
 
 			try(for_each_btree_key_continue(trans, iter, 0, k, ({
 				bch2_disk_reservation_put(c, &res.r);
+				progress_update_iter(trans, &progress, &iter) ?:
 				check_reconcile_work_btree_key(trans, &iter, k) ?:
 				bch2_trans_commit(trans, &res.r, NULL, BCH_TRANS_COMMIT_no_enospc);
 			})));
@@ -2274,10 +2277,15 @@ static int check_reconcile_btree_bp(struct btree_trans *trans, struct bkey_s_c k
 noinline_for_stack
 static int check_reconcile_btree_bps(struct btree_trans *trans)
 {
+	struct progress_indicator progress;
+	bch2_progress_init(&progress, trans->c, BIT_ULL(BTREE_ID_reconcile_scan));
+
 	return for_each_btree_key_max(trans, iter, BTREE_ID_reconcile_scan,
 				      POS(1, 0), POS(1, U64_MAX),
-				      BTREE_ITER_prefetch, k,
+				      BTREE_ITER_prefetch, k, ({
-		check_reconcile_btree_bp(trans, k));
+		progress_update_iter(trans, &progress, &iter) ?:
+		check_reconcile_btree_bp(trans, k);
+	}));
 }
 
 int bch2_check_reconcile_work(struct bch_fs *c)

libbcachefs/journal/read.c

@@ -1233,7 +1233,11 @@ static CLOSURE_CALLBACK(bch2_journal_read_device)
 	ja->discard_idx = ja->dirty_idx_ondisk =
 		ja->dirty_idx = (ja->cur_idx + 1) % ja->nr;
 out:
-	bch_verbose(c, "journal read done on device %s, ret %i", ca->name, ret);
+	if (!ret)
+		bch_verbose_dev(ca, "journal read done");
+	else
+		bch_err_dev(ca, "journal read error %s", bch2_err_str(ret));
+
 	kvfree(buf.data);
 	enumerated_ref_put(&ca->io_ref[READ], BCH_DEV_READ_REF_journal_read);
 	closure_return(cl);