1 // SPDX-License-Identifier: GPL-2.0
3 #include "alloc_background.h"
4 #include "alloc_foreground.h"
6 #include "btree_update_interior.h"
7 #include "btree_write_buffer.h"
10 #include "disk_groups.h"
13 #include "journal_io.h"
14 #include "journal_reclaim.h"
15 #include "journal_seq_blacklist.h"
20 void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
21 struct journal_replay *j)
23 darray_for_each(j->ptrs, i) {
24 struct bch_dev *ca = bch_dev_bkey_exists(c, i->dev);
27 div64_u64_rem(i->sector, ca->mi.bucket_size, &offset);
29 if (i != j->ptrs.data)
31 prt_printf(out, "%u:%u:%u (sector %llu)",
32 i->dev, i->bucket, i->bucket_offset, i->sector);
36 static void bch2_journal_replay_to_text(struct printbuf *out, struct bch_fs *c,
37 struct journal_replay *j)
39 prt_printf(out, "seq %llu ", le64_to_cpu(j->j.seq));
41 bch2_journal_ptrs_to_text(out, c, j);
43 for_each_jset_entry_type(entry, &j->j, BCH_JSET_ENTRY_datetime) {
44 struct jset_entry_datetime *datetime =
45 container_of(entry, struct jset_entry_datetime, entry);
46 bch2_prt_datetime(out, le64_to_cpu(datetime->seconds));
51 static struct nonce journal_nonce(const struct jset *jset)
53 return (struct nonce) {{
55 [1] = ((__le32 *) &jset->seq)[0],
56 [2] = ((__le32 *) &jset->seq)[1],
57 [3] = BCH_NONCE_JOURNAL,
61 static bool jset_csum_good(struct bch_fs *c, struct jset *j, struct bch_csum *csum)
63 if (!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(j))) {
64 *csum = (struct bch_csum) {};
68 *csum = csum_vstruct(c, JSET_CSUM_TYPE(j), journal_nonce(j), j);
69 return !bch2_crc_cmp(j->csum, *csum);
72 static inline u32 journal_entry_radix_idx(struct bch_fs *c, u64 seq)
74 return (seq - c->journal_entries_base_seq) & (~0U >> 1);
77 static void __journal_replay_free(struct bch_fs *c,
78 struct journal_replay *i)
80 struct journal_replay **p =
81 genradix_ptr(&c->journal_entries,
82 journal_entry_radix_idx(c, le64_to_cpu(i->j.seq)));
89 static void journal_replay_free(struct bch_fs *c, struct journal_replay *i, bool blacklisted)
92 i->ignore_blacklisted = true;
94 i->ignore_not_dirty = true;
96 if (!c->opts.read_entire_journal)
97 __journal_replay_free(c, i);
100 struct journal_list {
107 #define JOURNAL_ENTRY_ADD_OK 0
108 #define JOURNAL_ENTRY_ADD_OUT_OF_RANGE 5
111 * Given a journal entry we just read, add it to the list of journal entries to
114 static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca,
115 struct journal_ptr entry_ptr,
116 struct journal_list *jlist, struct jset *j)
118 struct genradix_iter iter;
119 struct journal_replay **_i, *i, *dup;
120 size_t bytes = vstruct_bytes(j);
121 u64 last_seq = !JSET_NO_FLUSH(j) ? le64_to_cpu(j->last_seq) : 0;
122 struct printbuf buf = PRINTBUF;
123 int ret = JOURNAL_ENTRY_ADD_OK;
125 /* Is this entry older than the range we need? */
126 if (!c->opts.read_entire_journal &&
127 le64_to_cpu(j->seq) < jlist->last_seq)
128 return JOURNAL_ENTRY_ADD_OUT_OF_RANGE;
131 * genradixes are indexed by a ulong, not a u64, so we can't index them
132 * by sequence number directly: Assume instead that they will all fall
133 * within the range of +-2billion of the filrst one we find.
135 if (!c->journal_entries_base_seq)
136 c->journal_entries_base_seq = max_t(s64, 1, le64_to_cpu(j->seq) - S32_MAX);
138 /* Drop entries we don't need anymore */
139 if (last_seq > jlist->last_seq && !c->opts.read_entire_journal) {
140 genradix_for_each_from(&c->journal_entries, iter, _i,
141 journal_entry_radix_idx(c, jlist->last_seq)) {
144 if (journal_replay_ignore(i))
147 if (le64_to_cpu(i->j.seq) >= last_seq)
150 journal_replay_free(c, i, false);
154 jlist->last_seq = max(jlist->last_seq, last_seq);
156 _i = genradix_ptr_alloc(&c->journal_entries,
157 journal_entry_radix_idx(c, le64_to_cpu(j->seq)),
160 return -BCH_ERR_ENOMEM_journal_entry_add;
163 * Duplicate journal entries? If so we want the one that didn't have a
168 bool identical = bytes == vstruct_bytes(&dup->j) &&
169 !memcmp(j, &dup->j, bytes);
170 bool not_identical = !identical &&
171 entry_ptr.csum_good &&
174 bool same_device = false;
175 darray_for_each(dup->ptrs, ptr)
176 if (ptr->dev == ca->dev_idx)
179 ret = darray_push(&dup->ptrs, entry_ptr);
183 bch2_journal_replay_to_text(&buf, c, dup);
185 fsck_err_on(same_device,
186 c, journal_entry_dup_same_device,
187 "duplicate journal entry on same device\n %s",
190 fsck_err_on(not_identical,
191 c, journal_entry_replicas_data_mismatch,
192 "found duplicate but non identical journal entries\n %s",
195 if (entry_ptr.csum_good && !identical)
201 i = kvmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL);
203 return -BCH_ERR_ENOMEM_journal_entry_add;
205 darray_init(&i->ptrs);
206 i->csum_good = entry_ptr.csum_good;
207 i->ignore_blacklisted = false;
208 i->ignore_not_dirty = false;
209 unsafe_memcpy(&i->j, j, bytes, "embedded variable length struct");
212 /* The first ptr should represent the jset we kept: */
213 darray_for_each(dup->ptrs, ptr)
214 darray_push(&i->ptrs, *ptr);
215 __journal_replay_free(c, dup);
217 darray_push(&i->ptrs, entry_ptr);
227 /* this fills in a range with empty jset_entries: */
228 static void journal_entry_null_range(void *start, void *end)
230 struct jset_entry *entry;
232 for (entry = start; entry != end; entry = vstruct_next(entry))
233 memset(entry, 0, sizeof(*entry));
236 #define JOURNAL_ENTRY_REREAD 5
237 #define JOURNAL_ENTRY_NONE 6
238 #define JOURNAL_ENTRY_BAD 7
240 static void journal_entry_err_msg(struct printbuf *out,
243 struct jset_entry *entry)
245 prt_str(out, "invalid journal entry, version=");
246 bch2_version_to_text(out, version);
249 prt_str(out, " type=");
250 prt_str(out, bch2_jset_entry_types[entry->type]);
254 prt_printf(out, " in superblock");
257 prt_printf(out, " seq=%llu", le64_to_cpu(jset->seq));
260 prt_printf(out, " offset=%zi/%u",
261 (u64 *) entry - jset->_data,
262 le32_to_cpu(jset->u64s));
268 #define journal_entry_err(c, version, jset, entry, _err, msg, ...) \
270 struct printbuf _buf = PRINTBUF; \
272 journal_entry_err_msg(&_buf, version, jset, entry); \
273 prt_printf(&_buf, msg, ##__VA_ARGS__); \
275 switch (flags & BKEY_INVALID_WRITE) { \
277 mustfix_fsck_err(c, _err, "%s", _buf.buf); \
280 bch2_sb_error_count(c, BCH_FSCK_ERR_##_err); \
281 bch_err(c, "corrupt metadata before write: %s\n", _buf.buf);\
282 if (bch2_fs_inconsistent(c)) { \
283 ret = -BCH_ERR_fsck_errors_not_fixed; \
289 printbuf_exit(&_buf); \
293 #define journal_entry_err_on(cond, ...) \
294 ((cond) ? journal_entry_err(__VA_ARGS__) : false)
296 #define FSCK_DELETED_KEY 5
298 static int journal_validate_key(struct bch_fs *c,
300 struct jset_entry *entry,
301 unsigned level, enum btree_id btree_id,
303 unsigned version, int big_endian,
304 enum bkey_invalid_flags flags)
306 int write = flags & BKEY_INVALID_WRITE;
307 void *next = vstruct_next(entry);
308 struct printbuf buf = PRINTBUF;
311 if (journal_entry_err_on(!k->k.u64s,
312 c, version, jset, entry,
313 journal_entry_bkey_u64s_0,
315 entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
316 journal_entry_null_range(vstruct_next(entry), next);
317 return FSCK_DELETED_KEY;
320 if (journal_entry_err_on((void *) bkey_next(k) >
321 (void *) vstruct_next(entry),
322 c, version, jset, entry,
323 journal_entry_bkey_past_end,
324 "extends past end of journal entry")) {
325 entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
326 journal_entry_null_range(vstruct_next(entry), next);
327 return FSCK_DELETED_KEY;
330 if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT,
331 c, version, jset, entry,
332 journal_entry_bkey_bad_format,
333 "bad format %u", k->k.format)) {
334 le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
335 memmove(k, bkey_next(k), next - (void *) bkey_next(k));
336 journal_entry_null_range(vstruct_next(entry), next);
337 return FSCK_DELETED_KEY;
341 bch2_bkey_compat(level, btree_id, version, big_endian,
342 write, NULL, bkey_to_packed(k));
344 if (bch2_bkey_invalid(c, bkey_i_to_s_c(k),
345 __btree_node_type(level, btree_id), write, &buf)) {
346 printbuf_reset(&buf);
347 journal_entry_err_msg(&buf, version, jset, entry);
349 printbuf_indent_add(&buf, 2);
351 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k));
353 bch2_bkey_invalid(c, bkey_i_to_s_c(k),
354 __btree_node_type(level, btree_id), write, &buf);
356 mustfix_fsck_err(c, journal_entry_bkey_invalid,
359 le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
360 memmove(k, bkey_next(k), next - (void *) bkey_next(k));
361 journal_entry_null_range(vstruct_next(entry), next);
364 return FSCK_DELETED_KEY;
368 bch2_bkey_compat(level, btree_id, version, big_endian,
369 write, NULL, bkey_to_packed(k));
375 static int journal_entry_btree_keys_validate(struct bch_fs *c,
377 struct jset_entry *entry,
378 unsigned version, int big_endian,
379 enum bkey_invalid_flags flags)
381 struct bkey_i *k = entry->start;
383 while (k != vstruct_last(entry)) {
384 int ret = journal_validate_key(c, jset, entry,
387 k, version, big_endian,
388 flags|BKEY_INVALID_JOURNAL);
389 if (ret == FSCK_DELETED_KEY)
398 static void journal_entry_btree_keys_to_text(struct printbuf *out, struct bch_fs *c,
399 struct jset_entry *entry)
403 jset_entry_for_each_key(entry, k) {
406 prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]);
408 prt_printf(out, "btree=%s l=%u ", bch2_btree_id_str(entry->btree_id), entry->level);
409 bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(k));
414 static int journal_entry_btree_root_validate(struct bch_fs *c,
416 struct jset_entry *entry,
417 unsigned version, int big_endian,
418 enum bkey_invalid_flags flags)
420 struct bkey_i *k = entry->start;
423 if (journal_entry_err_on(!entry->u64s ||
424 le16_to_cpu(entry->u64s) != k->k.u64s,
425 c, version, jset, entry,
426 journal_entry_btree_root_bad_size,
427 "invalid btree root journal entry: wrong number of keys")) {
428 void *next = vstruct_next(entry);
430 * we don't want to null out this jset_entry,
431 * just the contents, so that later we can tell
432 * we were _supposed_ to have a btree root
435 journal_entry_null_range(vstruct_next(entry), next);
439 ret = journal_validate_key(c, jset, entry, 1, entry->btree_id, k,
440 version, big_endian, flags);
441 if (ret == FSCK_DELETED_KEY)
447 static void journal_entry_btree_root_to_text(struct printbuf *out, struct bch_fs *c,
448 struct jset_entry *entry)
450 journal_entry_btree_keys_to_text(out, c, entry);
453 static int journal_entry_prio_ptrs_validate(struct bch_fs *c,
455 struct jset_entry *entry,
456 unsigned version, int big_endian,
457 enum bkey_invalid_flags flags)
459 /* obsolete, don't care: */
463 static void journal_entry_prio_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
464 struct jset_entry *entry)
468 static int journal_entry_blacklist_validate(struct bch_fs *c,
470 struct jset_entry *entry,
471 unsigned version, int big_endian,
472 enum bkey_invalid_flags flags)
476 if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 1,
477 c, version, jset, entry,
478 journal_entry_blacklist_bad_size,
479 "invalid journal seq blacklist entry: bad size")) {
480 journal_entry_null_range(entry, vstruct_next(entry));
486 static void journal_entry_blacklist_to_text(struct printbuf *out, struct bch_fs *c,
487 struct jset_entry *entry)
489 struct jset_entry_blacklist *bl =
490 container_of(entry, struct jset_entry_blacklist, entry);
492 prt_printf(out, "seq=%llu", le64_to_cpu(bl->seq));
495 static int journal_entry_blacklist_v2_validate(struct bch_fs *c,
497 struct jset_entry *entry,
498 unsigned version, int big_endian,
499 enum bkey_invalid_flags flags)
501 struct jset_entry_blacklist_v2 *bl_entry;
504 if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 2,
505 c, version, jset, entry,
506 journal_entry_blacklist_v2_bad_size,
507 "invalid journal seq blacklist entry: bad size")) {
508 journal_entry_null_range(entry, vstruct_next(entry));
512 bl_entry = container_of(entry, struct jset_entry_blacklist_v2, entry);
514 if (journal_entry_err_on(le64_to_cpu(bl_entry->start) >
515 le64_to_cpu(bl_entry->end),
516 c, version, jset, entry,
517 journal_entry_blacklist_v2_start_past_end,
518 "invalid journal seq blacklist entry: start > end")) {
519 journal_entry_null_range(entry, vstruct_next(entry));
526 static void journal_entry_blacklist_v2_to_text(struct printbuf *out, struct bch_fs *c,
527 struct jset_entry *entry)
529 struct jset_entry_blacklist_v2 *bl =
530 container_of(entry, struct jset_entry_blacklist_v2, entry);
532 prt_printf(out, "start=%llu end=%llu",
533 le64_to_cpu(bl->start),
534 le64_to_cpu(bl->end));
537 static int journal_entry_usage_validate(struct bch_fs *c,
539 struct jset_entry *entry,
540 unsigned version, int big_endian,
541 enum bkey_invalid_flags flags)
543 struct jset_entry_usage *u =
544 container_of(entry, struct jset_entry_usage, entry);
545 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
548 if (journal_entry_err_on(bytes < sizeof(*u),
549 c, version, jset, entry,
550 journal_entry_usage_bad_size,
551 "invalid journal entry usage: bad size")) {
552 journal_entry_null_range(entry, vstruct_next(entry));
560 static void journal_entry_usage_to_text(struct printbuf *out, struct bch_fs *c,
561 struct jset_entry *entry)
563 struct jset_entry_usage *u =
564 container_of(entry, struct jset_entry_usage, entry);
566 prt_printf(out, "type=%s v=%llu",
567 bch2_fs_usage_types[u->entry.btree_id],
571 static int journal_entry_data_usage_validate(struct bch_fs *c,
573 struct jset_entry *entry,
574 unsigned version, int big_endian,
575 enum bkey_invalid_flags flags)
577 struct jset_entry_data_usage *u =
578 container_of(entry, struct jset_entry_data_usage, entry);
579 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
580 struct printbuf err = PRINTBUF;
583 if (journal_entry_err_on(bytes < sizeof(*u) ||
584 bytes < sizeof(*u) + u->r.nr_devs,
585 c, version, jset, entry,
586 journal_entry_data_usage_bad_size,
587 "invalid journal entry usage: bad size")) {
588 journal_entry_null_range(entry, vstruct_next(entry));
592 if (journal_entry_err_on(bch2_replicas_entry_validate(&u->r, c->disk_sb.sb, &err),
593 c, version, jset, entry,
594 journal_entry_data_usage_bad_size,
595 "invalid journal entry usage: %s", err.buf)) {
596 journal_entry_null_range(entry, vstruct_next(entry));
605 static void journal_entry_data_usage_to_text(struct printbuf *out, struct bch_fs *c,
606 struct jset_entry *entry)
608 struct jset_entry_data_usage *u =
609 container_of(entry, struct jset_entry_data_usage, entry);
611 bch2_replicas_entry_to_text(out, &u->r);
612 prt_printf(out, "=%llu", le64_to_cpu(u->v));
615 static int journal_entry_clock_validate(struct bch_fs *c,
617 struct jset_entry *entry,
618 unsigned version, int big_endian,
619 enum bkey_invalid_flags flags)
621 struct jset_entry_clock *clock =
622 container_of(entry, struct jset_entry_clock, entry);
623 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
626 if (journal_entry_err_on(bytes != sizeof(*clock),
627 c, version, jset, entry,
628 journal_entry_clock_bad_size,
630 journal_entry_null_range(entry, vstruct_next(entry));
634 if (journal_entry_err_on(clock->rw > 1,
635 c, version, jset, entry,
636 journal_entry_clock_bad_rw,
638 journal_entry_null_range(entry, vstruct_next(entry));
646 static void journal_entry_clock_to_text(struct printbuf *out, struct bch_fs *c,
647 struct jset_entry *entry)
649 struct jset_entry_clock *clock =
650 container_of(entry, struct jset_entry_clock, entry);
652 prt_printf(out, "%s=%llu", clock->rw ? "write" : "read", le64_to_cpu(clock->time));
655 static int journal_entry_dev_usage_validate(struct bch_fs *c,
657 struct jset_entry *entry,
658 unsigned version, int big_endian,
659 enum bkey_invalid_flags flags)
661 struct jset_entry_dev_usage *u =
662 container_of(entry, struct jset_entry_dev_usage, entry);
663 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
664 unsigned expected = sizeof(*u);
668 if (journal_entry_err_on(bytes < expected,
669 c, version, jset, entry,
670 journal_entry_dev_usage_bad_size,
671 "bad size (%u < %u)",
673 journal_entry_null_range(entry, vstruct_next(entry));
677 dev = le32_to_cpu(u->dev);
679 if (journal_entry_err_on(!bch2_dev_exists2(c, dev),
680 c, version, jset, entry,
681 journal_entry_dev_usage_bad_dev,
683 journal_entry_null_range(entry, vstruct_next(entry));
687 if (journal_entry_err_on(u->pad,
688 c, version, jset, entry,
689 journal_entry_dev_usage_bad_pad,
691 journal_entry_null_range(entry, vstruct_next(entry));
699 static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs *c,
700 struct jset_entry *entry)
702 struct jset_entry_dev_usage *u =
703 container_of(entry, struct jset_entry_dev_usage, entry);
704 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
706 prt_printf(out, "dev=%u", le32_to_cpu(u->dev));
708 for (i = 0; i < nr_types; i++) {
709 bch2_prt_data_type(out, i);
710 prt_printf(out, ": buckets=%llu sectors=%llu fragmented=%llu",
711 le64_to_cpu(u->d[i].buckets),
712 le64_to_cpu(u->d[i].sectors),
713 le64_to_cpu(u->d[i].fragmented));
717 static int journal_entry_log_validate(struct bch_fs *c,
719 struct jset_entry *entry,
720 unsigned version, int big_endian,
721 enum bkey_invalid_flags flags)
726 static void journal_entry_log_to_text(struct printbuf *out, struct bch_fs *c,
727 struct jset_entry *entry)
729 struct jset_entry_log *l = container_of(entry, struct jset_entry_log, entry);
730 unsigned bytes = vstruct_bytes(entry) - offsetof(struct jset_entry_log, d);
732 prt_printf(out, "%.*s", bytes, l->d);
735 static int journal_entry_overwrite_validate(struct bch_fs *c,
737 struct jset_entry *entry,
738 unsigned version, int big_endian,
739 enum bkey_invalid_flags flags)
741 return journal_entry_btree_keys_validate(c, jset, entry,
742 version, big_endian, READ);
745 static void journal_entry_overwrite_to_text(struct printbuf *out, struct bch_fs *c,
746 struct jset_entry *entry)
748 journal_entry_btree_keys_to_text(out, c, entry);
751 static int journal_entry_write_buffer_keys_validate(struct bch_fs *c,
753 struct jset_entry *entry,
754 unsigned version, int big_endian,
755 enum bkey_invalid_flags flags)
757 return journal_entry_btree_keys_validate(c, jset, entry,
758 version, big_endian, READ);
761 static void journal_entry_write_buffer_keys_to_text(struct printbuf *out, struct bch_fs *c,
762 struct jset_entry *entry)
764 journal_entry_btree_keys_to_text(out, c, entry);
767 static int journal_entry_datetime_validate(struct bch_fs *c,
769 struct jset_entry *entry,
770 unsigned version, int big_endian,
771 enum bkey_invalid_flags flags)
773 unsigned bytes = vstruct_bytes(entry);
774 unsigned expected = 16;
777 if (journal_entry_err_on(vstruct_bytes(entry) < expected,
778 c, version, jset, entry,
779 journal_entry_dev_usage_bad_size,
780 "bad size (%u < %u)",
782 journal_entry_null_range(entry, vstruct_next(entry));
789 static void journal_entry_datetime_to_text(struct printbuf *out, struct bch_fs *c,
790 struct jset_entry *entry)
792 struct jset_entry_datetime *datetime =
793 container_of(entry, struct jset_entry_datetime, entry);
795 bch2_prt_datetime(out, le64_to_cpu(datetime->seconds));
798 struct jset_entry_ops {
799 int (*validate)(struct bch_fs *, struct jset *,
800 struct jset_entry *, unsigned, int,
801 enum bkey_invalid_flags);
802 void (*to_text)(struct printbuf *, struct bch_fs *, struct jset_entry *);
805 static const struct jset_entry_ops bch2_jset_entry_ops[] = {
807 [BCH_JSET_ENTRY_##f] = (struct jset_entry_ops) { \
808 .validate = journal_entry_##f##_validate, \
809 .to_text = journal_entry_##f##_to_text, \
811 BCH_JSET_ENTRY_TYPES()
815 int bch2_journal_entry_validate(struct bch_fs *c,
817 struct jset_entry *entry,
818 unsigned version, int big_endian,
819 enum bkey_invalid_flags flags)
821 return entry->type < BCH_JSET_ENTRY_NR
822 ? bch2_jset_entry_ops[entry->type].validate(c, jset, entry,
823 version, big_endian, flags)
827 void bch2_journal_entry_to_text(struct printbuf *out, struct bch_fs *c,
828 struct jset_entry *entry)
830 if (entry->type < BCH_JSET_ENTRY_NR) {
831 prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]);
832 bch2_jset_entry_ops[entry->type].to_text(out, c, entry);
834 prt_printf(out, "(unknown type %u)", entry->type);
838 static int jset_validate_entries(struct bch_fs *c, struct jset *jset,
839 enum bkey_invalid_flags flags)
841 unsigned version = le32_to_cpu(jset->version);
844 vstruct_for_each(jset, entry) {
845 if (journal_entry_err_on(vstruct_next(entry) > vstruct_last(jset),
846 c, version, jset, entry,
847 journal_entry_past_jset_end,
848 "journal entry extends past end of jset")) {
849 jset->u64s = cpu_to_le32((u64 *) entry - jset->_data);
853 ret = bch2_journal_entry_validate(c, jset, entry,
854 version, JSET_BIG_ENDIAN(jset), flags);
862 static int jset_validate(struct bch_fs *c,
864 struct jset *jset, u64 sector,
865 enum bkey_invalid_flags flags)
870 if (le64_to_cpu(jset->magic) != jset_magic(c))
871 return JOURNAL_ENTRY_NONE;
873 version = le32_to_cpu(jset->version);
874 if (journal_entry_err_on(!bch2_version_compatible(version),
875 c, version, jset, NULL,
876 jset_unsupported_version,
877 "%s sector %llu seq %llu: incompatible journal entry version %u.%u",
878 ca ? ca->name : c->name,
879 sector, le64_to_cpu(jset->seq),
880 BCH_VERSION_MAJOR(version),
881 BCH_VERSION_MINOR(version))) {
882 /* don't try to continue: */
886 if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)),
887 c, version, jset, NULL,
889 "%s sector %llu seq %llu: journal entry with unknown csum type %llu",
890 ca ? ca->name : c->name,
891 sector, le64_to_cpu(jset->seq),
892 JSET_CSUM_TYPE(jset)))
893 ret = JOURNAL_ENTRY_BAD;
895 /* last_seq is ignored when JSET_NO_FLUSH is true */
896 if (journal_entry_err_on(!JSET_NO_FLUSH(jset) &&
897 le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq),
898 c, version, jset, NULL,
899 jset_last_seq_newer_than_seq,
900 "invalid journal entry: last_seq > seq (%llu > %llu)",
901 le64_to_cpu(jset->last_seq),
902 le64_to_cpu(jset->seq))) {
903 jset->last_seq = jset->seq;
904 return JOURNAL_ENTRY_BAD;
907 ret = jset_validate_entries(c, jset, flags);
912 static int jset_validate_early(struct bch_fs *c,
914 struct jset *jset, u64 sector,
915 unsigned bucket_sectors_left,
916 unsigned sectors_read)
918 size_t bytes = vstruct_bytes(jset);
920 enum bkey_invalid_flags flags = BKEY_INVALID_JOURNAL;
923 if (le64_to_cpu(jset->magic) != jset_magic(c))
924 return JOURNAL_ENTRY_NONE;
926 version = le32_to_cpu(jset->version);
927 if (journal_entry_err_on(!bch2_version_compatible(version),
928 c, version, jset, NULL,
929 jset_unsupported_version,
930 "%s sector %llu seq %llu: unknown journal entry version %u.%u",
931 ca ? ca->name : c->name,
932 sector, le64_to_cpu(jset->seq),
933 BCH_VERSION_MAJOR(version),
934 BCH_VERSION_MINOR(version))) {
935 /* don't try to continue: */
939 if (bytes > (sectors_read << 9) &&
940 sectors_read < bucket_sectors_left)
941 return JOURNAL_ENTRY_REREAD;
943 if (journal_entry_err_on(bytes > bucket_sectors_left << 9,
944 c, version, jset, NULL,
945 jset_past_bucket_end,
946 "%s sector %llu seq %llu: journal entry too big (%zu bytes)",
947 ca ? ca->name : c->name,
948 sector, le64_to_cpu(jset->seq), bytes))
949 le32_add_cpu(&jset->u64s,
950 -((bytes - (bucket_sectors_left << 9)) / 8));
955 struct journal_read_buf {
960 static int journal_read_buf_realloc(struct journal_read_buf *b,
965 /* the bios are sized for this many pages, max: */
966 if (new_size > JOURNAL_ENTRY_SIZE_MAX)
967 return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
969 new_size = roundup_pow_of_two(new_size);
970 n = kvmalloc(new_size, GFP_KERNEL);
972 return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
980 static int journal_read_bucket(struct bch_dev *ca,
981 struct journal_read_buf *buf,
982 struct journal_list *jlist,
985 struct bch_fs *c = ca->fs;
986 struct journal_device *ja = &ca->journal;
987 struct jset *j = NULL;
988 unsigned sectors, sectors_read = 0;
989 u64 offset = bucket_to_sector(ca, ja->buckets[bucket]),
990 end = offset + ca->mi.bucket_size;
991 bool saw_bad = false, csum_good;
992 struct printbuf err = PRINTBUF;
995 pr_debug("reading %u", bucket);
997 while (offset < end) {
1002 sectors_read = min_t(unsigned,
1003 end - offset, buf->size >> 9);
1004 nr_bvecs = buf_pages(buf->data, sectors_read << 9);
1006 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
1007 bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ);
1009 bio->bi_iter.bi_sector = offset;
1010 bch2_bio_map(bio, buf->data, sectors_read << 9);
1012 ret = submit_bio_wait(bio);
1015 if (bch2_dev_io_err_on(ret, ca, BCH_MEMBER_ERROR_read,
1016 "journal read error: sector %llu",
1018 bch2_meta_read_fault("journal")) {
1020 * We don't error out of the recovery process
1021 * here, since the relevant journal entry may be
1022 * found on a different device, and missing or
1023 * no journal entries will be handled later
1031 ret = jset_validate_early(c, ca, j, offset,
1032 end - offset, sectors_read);
1035 sectors = vstruct_sectors(j, c->block_bits);
1037 case JOURNAL_ENTRY_REREAD:
1038 if (vstruct_bytes(j) > buf->size) {
1039 ret = journal_read_buf_realloc(buf,
1045 case JOURNAL_ENTRY_NONE:
1049 * On checksum error we don't really trust the size
1050 * field of the journal entry we read, so try reading
1051 * again at next block boundary:
1053 sectors = block_sectors(c);
1060 * This happens sometimes if we don't have discards on -
1061 * when we've partially overwritten a bucket with new
1062 * journal entries. We don't need the rest of the
1065 if (le64_to_cpu(j->seq) < ja->bucket_seq[bucket])
1068 ja->bucket_seq[bucket] = le64_to_cpu(j->seq);
1070 enum bch_csum_type csum_type = JSET_CSUM_TYPE(j);
1071 struct bch_csum csum;
1072 csum_good = jset_csum_good(c, j, &csum);
1074 if (bch2_dev_io_err_on(!csum_good, ca, BCH_MEMBER_ERROR_checksum,
1076 (printbuf_reset(&err),
1077 prt_str(&err, "journal "),
1078 bch2_csum_err_msg(&err, csum_type, j->csum, csum),
1082 ret = bch2_encrypt(c, JSET_CSUM_TYPE(j), journal_nonce(j),
1084 vstruct_end(j) - (void *) j->encrypted_start);
1085 bch2_fs_fatal_err_on(ret, c,
1086 "error decrypting journal entry: %s",
1089 mutex_lock(&jlist->lock);
1090 ret = journal_entry_add(c, ca, (struct journal_ptr) {
1091 .csum_good = csum_good,
1094 .bucket_offset = offset -
1095 bucket_to_sector(ca, ja->buckets[bucket]),
1098 mutex_unlock(&jlist->lock);
1101 case JOURNAL_ENTRY_ADD_OK:
1103 case JOURNAL_ENTRY_ADD_OUT_OF_RANGE:
1111 sectors_read -= sectors;
1112 j = ((void *) j) + (sectors << 9);
1118 printbuf_exit(&err);
1122 static CLOSURE_CALLBACK(bch2_journal_read_device)
1124 closure_type(ja, struct journal_device, read);
1125 struct bch_dev *ca = container_of(ja, struct bch_dev, journal);
1126 struct bch_fs *c = ca->fs;
1127 struct journal_list *jlist =
1128 container_of(cl->parent, struct journal_list, cl);
1129 struct journal_replay *r, **_r;
1130 struct genradix_iter iter;
1131 struct journal_read_buf buf = { NULL, 0 };
1138 ret = journal_read_buf_realloc(&buf, PAGE_SIZE);
1142 pr_debug("%u journal buckets", ja->nr);
1144 for (i = 0; i < ja->nr; i++) {
1145 ret = journal_read_bucket(ca, &buf, jlist, i);
1150 ja->sectors_free = ca->mi.bucket_size;
1152 mutex_lock(&jlist->lock);
1153 genradix_for_each_reverse(&c->journal_entries, iter, _r) {
1159 darray_for_each(r->ptrs, i)
1160 if (i->dev == ca->dev_idx) {
1161 unsigned wrote = bucket_remainder(ca, i->sector) +
1162 vstruct_sectors(&r->j, c->block_bits);
1164 ja->cur_idx = i->bucket;
1165 ja->sectors_free = ca->mi.bucket_size - wrote;
1170 mutex_unlock(&jlist->lock);
1172 if (ja->bucket_seq[ja->cur_idx] &&
1173 ja->sectors_free == ca->mi.bucket_size) {
1176 * Debug code for ZNS support, where we (probably) want to be
1177 * correlated where we stopped in the journal to the zone write
1180 bch_err(c, "ja->sectors_free == ca->mi.bucket_size");
1181 bch_err(c, "cur_idx %u/%u", ja->cur_idx, ja->nr);
1182 for (i = 0; i < 3; i++) {
1183 unsigned idx = (ja->cur_idx + ja->nr - 1 + i) % ja->nr;
1185 bch_err(c, "bucket_seq[%u] = %llu", idx, ja->bucket_seq[idx]);
1188 ja->sectors_free = 0;
1192 * Set dirty_idx to indicate the entire journal is full and needs to be
1193 * reclaimed - journal reclaim will immediately reclaim whatever isn't
1194 * pinned when it first runs:
1196 ja->discard_idx = ja->dirty_idx_ondisk =
1197 ja->dirty_idx = (ja->cur_idx + 1) % ja->nr;
1199 bch_verbose(c, "journal read done on device %s, ret %i", ca->name, ret);
1201 percpu_ref_put(&ca->io_ref);
1205 mutex_lock(&jlist->lock);
1207 mutex_unlock(&jlist->lock);
1211 int bch2_journal_read(struct bch_fs *c,
1216 struct journal_list jlist;
1217 struct journal_replay *i, **_i, *prev = NULL;
1218 struct genradix_iter radix_iter;
1219 struct printbuf buf = PRINTBUF;
1220 bool degraded = false, last_write_torn = false;
1224 closure_init_stack(&jlist.cl);
1225 mutex_init(&jlist.lock);
1229 for_each_member_device(c, ca) {
1230 if (!c->opts.fsck &&
1231 !(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_journal)))
1234 if ((ca->mi.state == BCH_MEMBER_STATE_rw ||
1235 ca->mi.state == BCH_MEMBER_STATE_ro) &&
1236 percpu_ref_tryget(&ca->io_ref))
1237 closure_call(&ca->journal.read,
1238 bch2_journal_read_device,
1245 closure_sync(&jlist.cl);
1255 * Find most recent flush entry, and ignore newer non flush entries -
1256 * those entries will be blacklisted:
1258 genradix_for_each_reverse(&c->journal_entries, radix_iter, _i) {
1259 enum bkey_invalid_flags flags = BKEY_INVALID_JOURNAL;
1263 if (journal_replay_ignore(i))
1267 *blacklist_seq = *start_seq = le64_to_cpu(i->j.seq) + 1;
1269 if (JSET_NO_FLUSH(&i->j)) {
1270 i->ignore_blacklisted = true;
1274 if (!last_write_torn && !i->csum_good) {
1275 last_write_torn = true;
1276 i->ignore_blacklisted = true;
1280 if (journal_entry_err_on(le64_to_cpu(i->j.last_seq) > le64_to_cpu(i->j.seq),
1281 c, le32_to_cpu(i->j.version), &i->j, NULL,
1282 jset_last_seq_newer_than_seq,
1283 "invalid journal entry: last_seq > seq (%llu > %llu)",
1284 le64_to_cpu(i->j.last_seq),
1285 le64_to_cpu(i->j.seq)))
1286 i->j.last_seq = i->j.seq;
1288 *last_seq = le64_to_cpu(i->j.last_seq);
1289 *blacklist_seq = le64_to_cpu(i->j.seq) + 1;
1294 bch_info(c, "journal read done, but no entries found");
1299 fsck_err(c, dirty_but_no_journal_entries_post_drop_nonflushes,
1300 "journal read done, but no entries found after dropping non-flushes");
1304 bch_info(c, "journal read done, replaying entries %llu-%llu",
1305 *last_seq, *blacklist_seq - 1);
1307 if (*start_seq != *blacklist_seq)
1308 bch_info(c, "dropped unflushed entries %llu-%llu",
1309 *blacklist_seq, *start_seq - 1);
1311 /* Drop blacklisted entries and entries older than last_seq: */
1312 genradix_for_each(&c->journal_entries, radix_iter, _i) {
1315 if (journal_replay_ignore(i))
1318 seq = le64_to_cpu(i->j.seq);
1319 if (seq < *last_seq) {
1320 journal_replay_free(c, i, false);
1324 if (bch2_journal_seq_is_blacklisted(c, seq, true)) {
1325 fsck_err_on(!JSET_NO_FLUSH(&i->j), c,
1326 jset_seq_blacklisted,
1327 "found blacklisted journal entry %llu", seq);
1328 i->ignore_blacklisted = true;
1332 /* Check for missing entries: */
1334 genradix_for_each(&c->journal_entries, radix_iter, _i) {
1337 if (journal_replay_ignore(i))
1340 BUG_ON(seq > le64_to_cpu(i->j.seq));
1342 while (seq < le64_to_cpu(i->j.seq)) {
1343 u64 missing_start, missing_end;
1344 struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
1346 while (seq < le64_to_cpu(i->j.seq) &&
1347 bch2_journal_seq_is_blacklisted(c, seq, false))
1350 if (seq == le64_to_cpu(i->j.seq))
1353 missing_start = seq;
1355 while (seq < le64_to_cpu(i->j.seq) &&
1356 !bch2_journal_seq_is_blacklisted(c, seq, false))
1360 bch2_journal_ptrs_to_text(&buf1, c, prev);
1361 prt_printf(&buf1, " size %zu", vstruct_sectors(&prev->j, c->block_bits));
1363 prt_printf(&buf1, "(none)");
1364 bch2_journal_ptrs_to_text(&buf2, c, i);
1366 missing_end = seq - 1;
1367 fsck_err(c, journal_entries_missing,
1368 "journal entries %llu-%llu missing! (replaying %llu-%llu)\n"
1371 missing_start, missing_end,
1372 *last_seq, *blacklist_seq - 1,
1373 buf1.buf, buf2.buf);
1375 printbuf_exit(&buf1);
1376 printbuf_exit(&buf2);
1383 genradix_for_each(&c->journal_entries, radix_iter, _i) {
1384 struct bch_replicas_padded replicas = {
1385 .e.data_type = BCH_DATA_journal,
1390 if (journal_replay_ignore(i))
1393 darray_for_each(i->ptrs, ptr) {
1394 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
1396 if (!ptr->csum_good)
1397 bch_err_dev_offset(ca, ptr->sector,
1398 "invalid journal checksum, seq %llu%s",
1399 le64_to_cpu(i->j.seq),
1400 i->csum_good ? " (had good copy on another device)" : "");
1403 ret = jset_validate(c,
1404 bch_dev_bkey_exists(c, i->ptrs.data[0].dev),
1406 i->ptrs.data[0].sector,
1411 darray_for_each(i->ptrs, ptr)
1412 replicas.e.devs[replicas.e.nr_devs++] = ptr->dev;
1414 bch2_replicas_entry_sort(&replicas.e);
1416 printbuf_reset(&buf);
1417 bch2_replicas_entry_to_text(&buf, &replicas.e);
1420 !bch2_replicas_marked(c, &replicas.e) &&
1421 (le64_to_cpu(i->j.seq) == *last_seq ||
1422 fsck_err(c, journal_entry_replicas_not_marked,
1423 "superblock not marked as containing replicas for journal entry %llu\n %s",
1424 le64_to_cpu(i->j.seq), buf.buf))) {
1425 ret = bch2_mark_replicas(c, &replicas.e);
1432 printbuf_exit(&buf);
1436 /* journal write: */
1438 static void __journal_write_alloc(struct journal *j,
1439 struct journal_buf *w,
1440 struct dev_alloc_list *devs_sorted,
1443 unsigned replicas_want)
1445 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1446 struct journal_device *ja;
1450 if (*replicas >= replicas_want)
1453 for (i = 0; i < devs_sorted->nr; i++) {
1454 ca = rcu_dereference(c->devs[devs_sorted->devs[i]]);
1461 * Check that we can use this device, and aren't already using
1464 if (!ca->mi.durability ||
1465 ca->mi.state != BCH_MEMBER_STATE_rw ||
1467 bch2_bkey_has_device_c(bkey_i_to_s_c(&w->key), ca->dev_idx) ||
1468 sectors > ja->sectors_free)
1471 bch2_dev_stripe_increment(ca, &j->wp.stripe);
1473 bch2_bkey_append_ptr(&w->key,
1474 (struct bch_extent_ptr) {
1475 .offset = bucket_to_sector(ca,
1476 ja->buckets[ja->cur_idx]) +
1477 ca->mi.bucket_size -
1482 ja->sectors_free -= sectors;
1483 ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1485 *replicas += ca->mi.durability;
1487 if (*replicas >= replicas_want)
1493 * journal_write_alloc - decide where to write next journal entry
1495 * @j: journal object
1496 * @w: journal buf (entry to be written)
1498 * Returns: 0 on success, or -EROFS on failure
1500 static int journal_write_alloc(struct journal *j, struct journal_buf *w)
1502 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1503 struct bch_devs_mask devs;
1504 struct journal_device *ja;
1506 struct dev_alloc_list devs_sorted;
1507 unsigned sectors = vstruct_sectors(w->data, c->block_bits);
1508 unsigned target = c->opts.metadata_target ?:
1509 c->opts.foreground_target;
1510 unsigned i, replicas = 0, replicas_want =
1511 READ_ONCE(c->opts.metadata_replicas);
1512 unsigned replicas_need = min_t(unsigned, replicas_want,
1513 READ_ONCE(c->opts.metadata_replicas_required));
1517 devs = target_rw_devs(c, BCH_DATA_journal, target);
1519 devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe, &devs);
1521 __journal_write_alloc(j, w, &devs_sorted,
1522 sectors, &replicas, replicas_want);
1524 if (replicas >= replicas_want)
1527 for (i = 0; i < devs_sorted.nr; i++) {
1528 ca = rcu_dereference(c->devs[devs_sorted.devs[i]]);
1534 if (sectors > ja->sectors_free &&
1535 sectors <= ca->mi.bucket_size &&
1536 bch2_journal_dev_buckets_available(j, ja,
1537 journal_space_discarded)) {
1538 ja->cur_idx = (ja->cur_idx + 1) % ja->nr;
1539 ja->sectors_free = ca->mi.bucket_size;
1542 * ja->bucket_seq[ja->cur_idx] must always have
1543 * something sensible:
1545 ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1549 __journal_write_alloc(j, w, &devs_sorted,
1550 sectors, &replicas, replicas_want);
1552 if (replicas < replicas_want && target) {
1553 /* Retry from all devices: */
1560 BUG_ON(bkey_val_u64s(&w->key.k) > BCH_REPLICAS_MAX);
1562 return replicas >= replicas_need ? 0 : -EROFS;
1565 static void journal_buf_realloc(struct journal *j, struct journal_buf *buf)
1567 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1569 /* we aren't holding j->lock: */
1570 unsigned new_size = READ_ONCE(j->buf_size_want);
1573 if (buf->buf_size >= new_size)
1576 size_t btree_write_buffer_size = new_size / 64;
1578 if (bch2_btree_write_buffer_resize(c, btree_write_buffer_size))
1581 new_buf = kvmalloc(new_size, GFP_NOFS|__GFP_NOWARN);
1585 memcpy(new_buf, buf->data, buf->buf_size);
1587 spin_lock(&j->lock);
1588 swap(buf->data, new_buf);
1589 swap(buf->buf_size, new_size);
1590 spin_unlock(&j->lock);
1595 static inline struct journal_buf *journal_last_unwritten_buf(struct journal *j)
1597 return j->buf + (journal_last_unwritten_seq(j) & JOURNAL_BUF_MASK);
1600 static CLOSURE_CALLBACK(journal_write_done)
1602 closure_type(w, struct journal_buf, io);
1603 struct journal *j = container_of(w, struct journal, buf[w->idx]);
1604 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1605 struct bch_replicas_padded replicas;
1606 union journal_res_state old, new;
1607 u64 v, seq = le64_to_cpu(w->data->seq);
1610 bch2_time_stats_update(!JSET_NO_FLUSH(w->data)
1611 ? j->flush_write_time
1612 : j->noflush_write_time, j->write_start_time);
1614 if (!w->devs_written.nr) {
1615 bch_err(c, "unable to write journal to sufficient devices");
1618 bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
1620 if (bch2_mark_replicas(c, &replicas.e))
1625 bch2_fatal_error(c);
1627 closure_debug_destroy(cl);
1629 spin_lock(&j->lock);
1630 if (seq >= j->pin.front)
1631 journal_seq_pin(j, seq)->devs = w->devs_written;
1632 if (err && (!j->err_seq || seq < j->err_seq))
1634 w->write_done = true;
1636 bool completed = false;
1638 for (seq = journal_last_unwritten_seq(j);
1639 seq <= journal_cur_seq(j);
1641 w = j->buf + (seq & JOURNAL_BUF_MASK);
1645 if (!j->err_seq && !JSET_NO_FLUSH(w->data)) {
1646 j->flushed_seq_ondisk = seq;
1647 j->last_seq_ondisk = w->last_seq;
1649 bch2_do_discards(c);
1650 closure_wake_up(&c->freelist_wait);
1651 bch2_reset_alloc_cursors(c);
1654 j->seq_ondisk = seq;
1657 * Updating last_seq_ondisk may let bch2_journal_reclaim_work() discard
1660 * Must come before signaling write completion, for
1661 * bch2_fs_journal_stop():
1663 if (j->watermark != BCH_WATERMARK_stripe)
1664 journal_reclaim_kick(&c->journal);
1666 v = atomic64_read(&j->reservations.counter);
1669 BUG_ON(journal_state_count(new, new.unwritten_idx));
1670 BUG_ON(new.unwritten_idx != (seq & JOURNAL_BUF_MASK));
1672 new.unwritten_idx++;
1673 } while ((v = atomic64_cmpxchg(&j->reservations.counter, old.v, new.v)) != old.v);
1675 closure_wake_up(&w->wait);
1680 bch2_journal_reclaim_fast(j);
1681 bch2_journal_space_available(j);
1683 track_event_change(&c->times[BCH_TIME_blocked_journal_max_in_flight], false);
1688 if (journal_last_unwritten_seq(j) == journal_cur_seq(j) &&
1689 new.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL) {
1690 struct journal_buf *buf = journal_cur_buf(j);
1691 long delta = buf->expires - jiffies;
1694 * We don't close a journal entry to write it while there's
1695 * previous entries still in flight - the current journal entry
1696 * might want to be written now:
1698 mod_delayed_work(j->wq, &j->write_work, max(0L, delta));
1701 spin_unlock(&j->lock);
1704 static void journal_write_endio(struct bio *bio)
1706 struct journal_bio *jbio = container_of(bio, struct journal_bio, bio);
1707 struct bch_dev *ca = jbio->ca;
1708 struct journal *j = &ca->fs->journal;
1709 struct journal_buf *w = j->buf + jbio->buf_idx;
1711 if (bch2_dev_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_write,
1712 "error writing journal entry %llu: %s",
1713 le64_to_cpu(w->data->seq),
1714 bch2_blk_status_to_str(bio->bi_status)) ||
1715 bch2_meta_write_fault("journal")) {
1716 unsigned long flags;
1718 spin_lock_irqsave(&j->err_lock, flags);
1719 bch2_dev_list_drop_dev(&w->devs_written, ca->dev_idx);
1720 spin_unlock_irqrestore(&j->err_lock, flags);
1723 closure_put(&w->io);
1724 percpu_ref_put(&ca->io_ref);
1727 static CLOSURE_CALLBACK(do_journal_write)
1729 closure_type(w, struct journal_buf, io);
1730 struct journal *j = container_of(w, struct journal, buf[w->idx]);
1731 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1732 unsigned sectors = vstruct_sectors(w->data, c->block_bits);
1734 extent_for_each_ptr(bkey_i_to_s_extent(&w->key), ptr) {
1735 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
1736 struct journal_device *ja = &ca->journal;
1738 if (!percpu_ref_tryget(&ca->io_ref)) {
1740 bch_err(c, "missing device for journal write\n");
1744 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_journal],
1747 struct bio *bio = &ja->bio[w->idx]->bio;
1748 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
1749 bio->bi_iter.bi_sector = ptr->offset;
1750 bio->bi_end_io = journal_write_endio;
1751 bio->bi_private = ca;
1753 BUG_ON(bio->bi_iter.bi_sector == ca->prev_journal_sector);
1754 ca->prev_journal_sector = bio->bi_iter.bi_sector;
1756 if (!JSET_NO_FLUSH(w->data))
1757 bio->bi_opf |= REQ_FUA;
1758 if (!JSET_NO_FLUSH(w->data) && !w->separate_flush)
1759 bio->bi_opf |= REQ_PREFLUSH;
1761 bch2_bio_map(bio, w->data, sectors << 9);
1763 trace_and_count(c, journal_write, bio);
1764 closure_bio_submit(bio, cl);
1766 ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1769 continue_at(cl, journal_write_done, j->wq);
1772 static int bch2_journal_write_prep(struct journal *j, struct journal_buf *w)
1774 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1775 struct jset_entry *start, *end;
1776 struct jset *jset = w->data;
1777 struct journal_keys_to_wb wb = { NULL };
1778 unsigned sectors, bytes, u64s;
1779 unsigned long btree_roots_have = 0;
1780 bool validate_before_checksum = false;
1781 u64 seq = le64_to_cpu(jset->seq);
1785 * Simple compaction, dropping empty jset_entries (from journal
1786 * reservations that weren't fully used) and merging jset_entries that
1789 * If we wanted to be really fancy here, we could sort all the keys in
1790 * the jset and drop keys that were overwritten - probably not worth it:
1792 vstruct_for_each(jset, i) {
1793 unsigned u64s = le16_to_cpu(i->u64s);
1800 * New btree roots are set by journalling them; when the journal
1801 * entry gets written we have to propagate them to
1804 * But, every journal entry we write has to contain all the
1805 * btree roots (at least for now); so after we copy btree roots
1806 * to c->btree_roots we have to get any missing btree roots and
1807 * add them to this journal entry:
1810 case BCH_JSET_ENTRY_btree_root:
1811 bch2_journal_entry_to_btree_root(c, i);
1812 __set_bit(i->btree_id, &btree_roots_have);
1814 case BCH_JSET_ENTRY_write_buffer_keys:
1815 EBUG_ON(!w->need_flush_to_write_buffer);
1818 bch2_journal_keys_to_write_buffer_start(c, &wb, seq);
1820 jset_entry_for_each_key(i, k) {
1821 ret = bch2_journal_key_to_wb(c, &wb, i->btree_id, k);
1823 bch2_fs_fatal_error(c, "-ENOMEM flushing journal keys to btree write buffer");
1824 bch2_journal_keys_to_write_buffer_end(c, &wb);
1828 i->type = BCH_JSET_ENTRY_btree_keys;
1834 bch2_journal_keys_to_write_buffer_end(c, &wb);
1836 spin_lock(&c->journal.lock);
1837 w->need_flush_to_write_buffer = false;
1838 spin_unlock(&c->journal.lock);
1840 start = end = vstruct_last(jset);
1842 end = bch2_btree_roots_to_journal_entries(c, end, btree_roots_have);
1844 struct jset_entry_datetime *d =
1845 container_of(jset_entry_init(&end, sizeof(*d)), struct jset_entry_datetime, entry);
1846 d->entry.type = BCH_JSET_ENTRY_datetime;
1847 d->seconds = cpu_to_le64(ktime_get_real_seconds());
1849 bch2_journal_super_entries_add_common(c, &end, seq);
1850 u64s = (u64 *) end - (u64 *) start;
1851 BUG_ON(u64s > j->entry_u64s_reserved);
1853 le32_add_cpu(&jset->u64s, u64s);
1855 sectors = vstruct_sectors(jset, c->block_bits);
1856 bytes = vstruct_bytes(jset);
1858 if (sectors > w->sectors) {
1859 bch2_fs_fatal_error(c, "aieeee! journal write overran available space, %zu > %u (extra %u reserved %u/%u)",
1860 vstruct_bytes(jset), w->sectors << 9,
1861 u64s, w->u64s_reserved, j->entry_u64s_reserved);
1865 jset->magic = cpu_to_le64(jset_magic(c));
1866 jset->version = cpu_to_le32(c->sb.version);
1868 SET_JSET_BIG_ENDIAN(jset, CPU_BIG_ENDIAN);
1869 SET_JSET_CSUM_TYPE(jset, bch2_meta_checksum_type(c));
1871 if (!JSET_NO_FLUSH(jset) && journal_entry_empty(jset))
1872 j->last_empty_seq = seq;
1874 if (bch2_csum_type_is_encryption(JSET_CSUM_TYPE(jset)))
1875 validate_before_checksum = true;
1877 if (le32_to_cpu(jset->version) < bcachefs_metadata_version_current)
1878 validate_before_checksum = true;
1880 if (validate_before_checksum &&
1881 (ret = jset_validate(c, NULL, jset, 0, WRITE)))
1884 ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset),
1885 jset->encrypted_start,
1886 vstruct_end(jset) - (void *) jset->encrypted_start);
1887 if (bch2_fs_fatal_err_on(ret, c,
1888 "error decrypting journal entry: %i", ret))
1891 jset->csum = csum_vstruct(c, JSET_CSUM_TYPE(jset),
1892 journal_nonce(jset), jset);
1894 if (!validate_before_checksum &&
1895 (ret = jset_validate(c, NULL, jset, 0, WRITE)))
1898 memset((void *) jset + bytes, 0, (sectors << 9) - bytes);
1902 static int bch2_journal_write_pick_flush(struct journal *j, struct journal_buf *w)
1904 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1905 int error = bch2_journal_error(j);
1908 * If the journal is in an error state - we did an emergency shutdown -
1909 * we prefer to continue doing journal writes. We just mark them as
1910 * noflush so they'll never be used, but they'll still be visible by the
1911 * list_journal tool - this helps in debugging.
1913 * There's a caveat: the first journal write after marking the
1914 * superblock dirty must always be a flush write, because on startup
1915 * from a clean shutdown we didn't necessarily read the journal and the
1916 * new journal write might overwrite whatever was in the journal
1917 * previously - we can't leave the journal without any flush writes in
1920 * So if we're in an error state, and we're still starting up, we don't
1921 * write anything at all.
1923 if (error && test_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags))
1929 (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) &&
1930 test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags))) {
1932 SET_JSET_NO_FLUSH(w->data, true);
1933 w->data->last_seq = 0;
1936 j->nr_noflush_writes++;
1938 w->must_flush = true;
1939 j->last_flush_write = jiffies;
1940 j->nr_flush_writes++;
1941 clear_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags);
1947 CLOSURE_CALLBACK(bch2_journal_write)
1949 closure_type(w, struct journal_buf, io);
1950 struct journal *j = container_of(w, struct journal, buf[w->idx]);
1951 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1952 struct bch_replicas_padded replicas;
1953 struct printbuf journal_debug_buf = PRINTBUF;
1954 unsigned nr_rw_members = 0;
1957 for_each_rw_member(c, ca)
1960 BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
1961 BUG_ON(!w->write_started);
1962 BUG_ON(w->write_allocated);
1963 BUG_ON(w->write_done);
1965 j->write_start_time = local_clock();
1967 spin_lock(&j->lock);
1968 if (nr_rw_members > 1)
1969 w->separate_flush = true;
1971 ret = bch2_journal_write_pick_flush(j, w);
1972 spin_unlock(&j->lock);
1976 mutex_lock(&j->buf_lock);
1977 journal_buf_realloc(j, w);
1979 ret = bch2_journal_write_prep(j, w);
1980 mutex_unlock(&j->buf_lock);
1984 j->entry_bytes_written += vstruct_bytes(w->data);
1987 spin_lock(&j->lock);
1988 ret = journal_write_alloc(j, w);
1989 if (!ret || !j->can_discard)
1992 spin_unlock(&j->lock);
1993 bch2_journal_do_discards(j);
1997 __bch2_journal_debug_to_text(&journal_debug_buf, j);
1998 spin_unlock(&j->lock);
1999 bch_err(c, "Unable to allocate journal write:\n%s",
2000 journal_debug_buf.buf);
2001 printbuf_exit(&journal_debug_buf);
2006 * write is allocated, no longer need to account for it in
2007 * bch2_journal_space_available():
2010 w->write_allocated = true;
2013 * journal entry has been compacted and allocated, recalculate space
2016 bch2_journal_space_available(j);
2017 bch2_journal_do_writes(j);
2018 spin_unlock(&j->lock);
2020 w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key));
2022 if (c->opts.nochanges)
2026 * Mark journal replicas before we submit the write to guarantee
2027 * recovery will find the journal entries after a crash.
2029 bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
2031 ret = bch2_mark_replicas(c, &replicas.e);
2035 if (!JSET_NO_FLUSH(w->data))
2036 closure_wait_event(&j->async_wait, j->seq_ondisk + 1 == le64_to_cpu(w->data->seq));
2038 if (!JSET_NO_FLUSH(w->data) && w->separate_flush) {
2039 for_each_rw_member(c, ca) {
2040 percpu_ref_get(&ca->io_ref);
2042 struct journal_device *ja = &ca->journal;
2043 struct bio *bio = &ja->bio[w->idx]->bio;
2044 bio_reset(bio, ca->disk_sb.bdev,
2045 REQ_OP_WRITE|REQ_SYNC|REQ_META|REQ_PREFLUSH);
2046 bio->bi_end_io = journal_write_endio;
2047 bio->bi_private = ca;
2048 closure_bio_submit(bio, cl);
2052 continue_at(cl, do_journal_write, j->wq);
2055 continue_at(cl, journal_write_done, j->wq);
2058 bch2_fatal_error(c);
2059 continue_at(cl, journal_write_done, j->wq);