fix repair for bkey errors
cover fix repair for bkey errors
bkey validation is done at btree node read time, where repair just means "drop it from the in memory btree node" - persisting that repair means kicking off an async btree node rewrite, which wasn't always happening.
this fixes, among other things, the "disk accounting key with version=0" errors that kept popping up despite the log saying it'd been fixed
Kent Overstreet (4):
bcachefs: list_pop_entry()
bcachefs: bkey_fsck_err now respects errors_silent
bcachefs: If we did repair on a btree node, make sure we rewrite it
bcachefs: bch2_async_btree_node_rewrites_flush()
fs/bcachefs/bcachefs.h | 7 +- fs/bcachefs/btree_io.c | 19 ++-- fs/bcachefs/btree_update_interior.c | 153 ++++++++++++++++------------ fs/bcachefs/btree_update_interior.h | 1 + fs/bcachefs/ec.c | 6 +- fs/bcachefs/error.c | 12 ++- fs/bcachefs/io_write.c | 4 +- fs/bcachefs/recovery.c | 2 + fs/bcachefs/util.h | 13 +++ 9 files changed, 135 insertions(+), 82 deletions(-)
Signature · 1 line
2.45.2
PATCH 1/4 bcachefs: list_pop_entry()
--- fs/bcachefs/ec.c | 6 ++---- fs/bcachefs/io_write.c | 4 +--- fs/bcachefs/util.h | 13 +++++++++++++ 3 files changed, 16 insertions(+), 7 deletions(-)
fs/bcachefs/ec.c +2 -4
@@ -2465,11 +2465,9 @@ void bch2_fs_ec_exit(struct bch_fs *c)while (1 ) {mutex_lock (&c->ec_stripe_head_lock); - h =list_first_entry_or_null (&c->ec_stripe_head_list, - struct ec_stripe_head, list); -if (h) -list_del (&h->list); + h = list_pop_entry(&c->ec_stripe_head_list, struct ec_stripe_head, list);mutex_unlock (&c->ec_stripe_head_lock); +if (!h)break ;
fs/bcachefs/io_write.c +1 -3
@@ -637,9 +637,7 @@ void bch2_write_point_do_index_updates(struct work_struct *work)while (1 ) {spin_lock_irq (&wp->writes_lock); - op = list_first_entry_or_null(&wp->writes, struct bch_write_op, wp_list); -if (op) -list_del (&op->wp_list); + op = list_pop_entry(&wp->writes, struct bch_write_op, wp_list);wp_update_state (wp, op !=NULL );spin_unlock_irq (&wp->writes_lock);
fs/bcachefs/util.h +13 -0
@@ -317,6 +317,19 @@ do { \ _ptr ?container_of (_ptr, type, member) :NULL ; \ }) +static inline struct list_head *list_pop (struct list_head *head) +{ +if (list_empty (head)) +return NULL ; + +struct list_head *ret = head->next; +list_del_init (ret); +return ret; +} + +#define list_pop_entry (head, type, member) \ + container_of_or_null(list_pop(head), type, member) +/* Does linear interpolation between powers of two */ static inline unsigned fract_exp_two (unsigned x,unsigned fract_bits) {
Signature · 1 line
2.45.2
PATCH 2/4 bcachefs: bkey_fsck_err now respects errors_silent
In reply to Kent Overstreet--- fs/bcachefs/error.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-)
fs/bcachefs/error.c +10 -2
@@ -476,11 +476,16 @@ int __bch2_bkey_fsck_err(struct bch_fs *c,return -BCH_ERR_fsck_delete_bkey;unsigned fsck_flags =0 ; -if (!(from.flags & (BCH_VALIDATE_write|BCH_VALIDATE_commit))) +if (!(from.flags & (BCH_VALIDATE_write|BCH_VALIDATE_commit))) { +if (test_bit (err, c->sb.errors_silent)) +return -BCH_ERR_fsck_delete_bkey; + fsck_flags |=FSCK_AUTOFIX |FSCK_CAN_FIX ; + } +if (!WARN_ON (err >=ARRAY_SIZE (fsck_flags_extra))) + fsck_flags |= fsck_flags_extra[err];struct printbuf buf =PRINTBUF ; -va_list args;prt_printf (&buf,"invalid bkey in %s btree=" , bch2_bkey_validate_contexts[from.from]);
@@ -489,9 +494,12 @@ int __bch2_bkey_fsck_err(struct bch_fs *c,bch2_bkey_val_to_text (&buf, c, k);prt_str (&buf,"\n " ); + +va_list args;va_start (args, fmt);prt_vprintf (&buf, fmt, args);va_end (args); +prt_str (&buf,": delete?" );int ret =__bch2_fsck_err (c,NULL , fsck_flags, err,"%s" , buf.buf);
Signature · 1 line
2.45.2
PATCH 3/4 bcachefs: If we did repair on a btree node, make sure we rewrite it
In reply to Kent OverstreetEnsure that "invalid bkey" repair gets persisted, so that it doesn't repeatedly spam the logs.
--- fs/bcachefs/btree_io.c | 19 ++++++++++++------- 1 file changed, 12 insertions(+), 7 deletions(-)
fs/bcachefs/btree_io.c +12 -7
@@ -997,6 +997,7 @@ static int validate_bset_keys(struct bch_fs *c, struct btree *b, got_good_key:le16_add_cpu (&i->u64s, -next_good_key);memmove_u64s_down (k,bkey_p_next (k), (u64 *)vstruct_end (i) - (u64 *) k); +set_btree_node_need_rewrite (b); } fsck_err:printbuf_exit (&buf);
@@ -1258,6 +1259,7 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,memmove_u64s_down (k,bkey_p_next (k), (u64 *)vstruct_end (i) - (u64 *) k);set_btree_bset_end (b, b->set); +set_btree_node_need_rewrite (b);continue ; }if (ret)
@@ -1371,15 +1373,18 @@ static void btree_node_read_work(struct work_struct *work) rb->start_time);bio_put (&rb->bio); -if (saw_error && +if ((saw_error || +btree_node_need_rewrite (b)) && !btree_node_read_error (b) && c->curr_recovery_pass != BCH_RECOVERY_PASS_scan_for_btree_nodes) { -printbuf_reset (&buf); -bch2_btree_id_level_to_text (&buf, b->c.btree_id, b->c.level); -prt_str (&buf," " ); -bch2_bkey_val_to_text (&buf, c,bkey_i_to_s_c (&b->key)); -bch_err_ratelimited (c,"%s: rewriting btree node at due to error\n %s" , - __func__, buf.buf); +if (saw_error) { +printbuf_reset (&buf); +bch2_btree_id_level_to_text (&buf, b->c.btree_id, b->c.level); +prt_str (&buf," " ); +bch2_bkey_val_to_text (&buf, c,bkey_i_to_s_c (&b->key)); +bch_err_ratelimited (c,"%s: rewriting btree node at due to error\n %s" , + __func__, buf.buf); + }bch2_btree_node_rewrite_async (c, b); }
Signature · 1 line
2.45.2
PATCH 4/4 bcachefs: bch2_async_btree_node_rewrites_flush()
In reply to Kent OverstreetAdd a method to flush btree node rewrites at the end of recovery, to ensure that corrected errors are persisted.
--- fs/bcachefs/bcachefs.h | 7 +- fs/bcachefs/btree_update_interior.c | 153 ++++++++++++++++------------ fs/bcachefs/btree_update_interior.h | 1 + fs/bcachefs/error.c | 4 +- fs/bcachefs/recovery.c | 2 + 5 files changed, 99 insertions(+), 68 deletions(-)
fs/bcachefs/bcachefs.h +4 -3
@@ -829,9 +829,10 @@ struct bch_fs {struct work_struct btree_interior_update_work;struct workqueue_struct *btree_node_rewrite_worker; - -struct list_head pending_node_rewrites; -struct mutex pending_node_rewrites_lock; +struct list_head btree_node_rewrites; +struct list_head btree_node_rewrites_pending; +spinlock_t btree_node_rewrites_lock; +struct closure_waitlist btree_node_rewrites_wait;/* btree_io.c: */ spinlock_t btree_write_error_lock;
fs/bcachefs/btree_update_interior.c +90 -63
227 lines not on this page · raw message
fs/bcachefs/btree_update_interior.h +1 -0
@@ -334,6 +334,7 @@ void bch2_journal_entry_to_btree_root(struct bch_fs *, struct jset_entry *);struct jset_entry *bch2_btree_roots_to_journal_entries (struct bch_fs *,struct jset_entry *,unsigned long ); +void bch2_async_btree_node_rewrites_flush (struct bch_fs *);void bch2_do_pending_node_rewrites (struct bch_fs *);void bch2_free_pending_node_rewrites (struct bch_fs *);
fs/bcachefs/error.c +2 -2
@@ -477,8 +477,8 @@ int __bch2_bkey_fsck_err(struct bch_fs *c,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; +//if (test_bit(err, c->sb.errors_silent)) +// return -BCH_ERR_fsck_delete_bkey; fsck_flags |=FSCK_AUTOFIX |FSCK_CAN_FIX ; }
fs/bcachefs/recovery.c +2 -0
@@ -931,6 +931,8 @@ int bch2_fs_recovery(struct bch_fs *c)/* in case we don't run journal replay, i.e. norecovery mode */ set_bit (BCH_FS_accounting_replay_done, &c->flags); +bch2_async_btree_node_rewrites_flush (c); +/* fsync if we fixed errors */ if (test_bit (BCH_FS_errors_fixed, &c->flags)) {bch2_journal_flush_all_pins (&c->journal);
Signature · 1 line
2.45.2