1 // SPDX-License-Identifier: GPL-2.0
3 #include "alloc_background.h"
4 #include "alloc_foreground.h"
6 #include "btree_update_interior.h"
9 #include "disk_groups.h"
12 #include "journal_io.h"
13 #include "journal_reclaim.h"
14 #include "journal_seq_blacklist.h"
19 static struct nonce journal_nonce(const struct jset *jset)
21 return (struct nonce) {{
23 [1] = ((__le32 *) &jset->seq)[0],
24 [2] = ((__le32 *) &jset->seq)[1],
25 [3] = BCH_NONCE_JOURNAL,
29 static bool jset_csum_good(struct bch_fs *c, struct jset *j)
31 return bch2_checksum_type_valid(c, JSET_CSUM_TYPE(j)) &&
32 !bch2_crc_cmp(j->csum,
33 csum_vstruct(c, JSET_CSUM_TYPE(j), journal_nonce(j), j));
36 static inline u32 journal_entry_radix_idx(struct bch_fs *c, u64 seq)
38 return (seq - c->journal_entries_base_seq) & (~0U >> 1);
41 static void __journal_replay_free(struct bch_fs *c,
42 struct journal_replay *i)
44 struct journal_replay **p =
45 genradix_ptr(&c->journal_entries,
46 journal_entry_radix_idx(c, le64_to_cpu(i->j.seq)));
50 kvpfree(i, offsetof(struct journal_replay, j) +
51 vstruct_bytes(&i->j));
54 static void journal_replay_free(struct bch_fs *c, struct journal_replay *i)
58 if (!c->opts.read_entire_journal)
59 __journal_replay_free(c, i);
69 #define JOURNAL_ENTRY_ADD_OK 0
70 #define JOURNAL_ENTRY_ADD_OUT_OF_RANGE 5
73 * Given a journal entry we just read, add it to the list of journal entries to
76 static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca,
77 struct journal_ptr entry_ptr,
78 struct journal_list *jlist, struct jset *j)
80 struct genradix_iter iter;
81 struct journal_replay **_i, *i, *dup;
82 struct journal_ptr *ptr;
83 size_t bytes = vstruct_bytes(j);
84 u64 last_seq = !JSET_NO_FLUSH(j) ? le64_to_cpu(j->last_seq) : 0;
85 int ret = JOURNAL_ENTRY_ADD_OK;
87 /* Is this entry older than the range we need? */
88 if (!c->opts.read_entire_journal &&
89 le64_to_cpu(j->seq) < jlist->last_seq)
90 return JOURNAL_ENTRY_ADD_OUT_OF_RANGE;
93 * genradixes are indexed by a ulong, not a u64, so we can't index them
94 * by sequence number directly: Assume instead that they will all fall
95 * within the range of +-2billion of the filrst one we find.
97 if (!c->journal_entries_base_seq)
98 c->journal_entries_base_seq = max_t(s64, 1, le64_to_cpu(j->seq) - S32_MAX);
100 /* Drop entries we don't need anymore */
101 if (last_seq > jlist->last_seq && !c->opts.read_entire_journal) {
102 genradix_for_each_from(&c->journal_entries, iter, _i,
103 journal_entry_radix_idx(c, jlist->last_seq)) {
109 if (le64_to_cpu(i->j.seq) >= last_seq)
111 journal_replay_free(c, i);
115 jlist->last_seq = max(jlist->last_seq, last_seq);
117 _i = genradix_ptr_alloc(&c->journal_entries,
118 journal_entry_radix_idx(c, le64_to_cpu(j->seq)),
121 return -BCH_ERR_ENOMEM_journal_entry_add;
124 * Duplicate journal entries? If so we want the one that didn't have a
129 if (bytes == vstruct_bytes(&dup->j) &&
130 !memcmp(j, &dup->j, bytes)) {
135 if (!entry_ptr.csum_good) {
143 fsck_err(c, "found duplicate but non identical journal entries (seq %llu)",
144 le64_to_cpu(j->seq));
149 i = kvpmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL);
151 return -BCH_ERR_ENOMEM_journal_entry_add;
154 i->csum_good = entry_ptr.csum_good;
156 unsafe_memcpy(&i->j, j, bytes, "embedded variable length struct");
157 i->ptrs[i->nr_ptrs++] = entry_ptr;
160 if (dup->nr_ptrs >= ARRAY_SIZE(dup->ptrs)) {
161 bch_err(c, "found too many copies of journal entry %llu",
162 le64_to_cpu(i->j.seq));
163 dup->nr_ptrs = ARRAY_SIZE(dup->ptrs) - 1;
166 /* The first ptr should represent the jset we kept: */
167 memcpy(i->ptrs + i->nr_ptrs,
169 sizeof(dup->ptrs[0]) * dup->nr_ptrs);
170 i->nr_ptrs += dup->nr_ptrs;
171 __journal_replay_free(c, dup);
177 for (ptr = i->ptrs; ptr < i->ptrs + i->nr_ptrs; ptr++) {
178 if (ptr->dev == ca->dev_idx) {
179 bch_err(c, "duplicate journal entry %llu on same device",
180 le64_to_cpu(i->j.seq));
185 if (i->nr_ptrs >= ARRAY_SIZE(i->ptrs)) {
186 bch_err(c, "found too many copies of journal entry %llu",
187 le64_to_cpu(i->j.seq));
191 i->ptrs[i->nr_ptrs++] = entry_ptr;
197 /* this fills in a range with empty jset_entries: */
198 static void journal_entry_null_range(void *start, void *end)
200 struct jset_entry *entry;
202 for (entry = start; entry != end; entry = vstruct_next(entry))
203 memset(entry, 0, sizeof(*entry));
206 #define JOURNAL_ENTRY_REREAD 5
207 #define JOURNAL_ENTRY_NONE 6
208 #define JOURNAL_ENTRY_BAD 7
210 static void journal_entry_err_msg(struct printbuf *out,
213 struct jset_entry *entry)
215 prt_str(out, "invalid journal entry, version=");
216 bch2_version_to_text(out, version);
219 prt_str(out, " type=");
220 prt_str(out, bch2_jset_entry_types[entry->type]);
224 prt_printf(out, " in superblock");
227 prt_printf(out, " seq=%llu", le64_to_cpu(jset->seq));
230 prt_printf(out, " offset=%zi/%u",
231 (u64 *) entry - jset->_data,
232 le32_to_cpu(jset->u64s));
238 #define journal_entry_err(c, version, jset, entry, msg, ...) \
240 struct printbuf _buf = PRINTBUF; \
242 journal_entry_err_msg(&_buf, version, jset, entry); \
243 prt_printf(&_buf, msg, ##__VA_ARGS__); \
245 switch (flags & BKEY_INVALID_WRITE) { \
247 mustfix_fsck_err(c, "%s", _buf.buf); \
250 bch_err(c, "corrupt metadata before write: %s\n", _buf.buf);\
251 if (bch2_fs_inconsistent(c)) { \
252 ret = -BCH_ERR_fsck_errors_not_fixed; \
258 printbuf_exit(&_buf); \
262 #define journal_entry_err_on(cond, c, version, jset, entry, msg, ...) \
263 ((cond) ? journal_entry_err(c, version, jset, entry, msg, ##__VA_ARGS__) : false)
265 #define FSCK_DELETED_KEY 5
267 static int journal_validate_key(struct bch_fs *c,
269 struct jset_entry *entry,
270 unsigned level, enum btree_id btree_id,
272 unsigned version, int big_endian,
273 enum bkey_invalid_flags flags)
275 int write = flags & BKEY_INVALID_WRITE;
276 void *next = vstruct_next(entry);
277 struct printbuf buf = PRINTBUF;
280 if (journal_entry_err_on(!k->k.u64s, c, version, jset, entry, "k->u64s 0")) {
281 entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
282 journal_entry_null_range(vstruct_next(entry), next);
283 return FSCK_DELETED_KEY;
286 if (journal_entry_err_on((void *) bkey_next(k) >
287 (void *) vstruct_next(entry),
288 c, version, jset, entry,
289 "extends past end of journal entry")) {
290 entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
291 journal_entry_null_range(vstruct_next(entry), next);
292 return FSCK_DELETED_KEY;
295 if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT,
296 c, version, jset, entry,
297 "bad format %u", k->k.format)) {
298 le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
299 memmove(k, bkey_next(k), next - (void *) bkey_next(k));
300 journal_entry_null_range(vstruct_next(entry), next);
301 return FSCK_DELETED_KEY;
305 bch2_bkey_compat(level, btree_id, version, big_endian,
306 write, NULL, bkey_to_packed(k));
308 if (bch2_bkey_invalid(c, bkey_i_to_s_c(k),
309 __btree_node_type(level, btree_id), write, &buf)) {
310 printbuf_reset(&buf);
311 journal_entry_err_msg(&buf, version, jset, entry);
313 printbuf_indent_add(&buf, 2);
315 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k));
317 bch2_bkey_invalid(c, bkey_i_to_s_c(k),
318 __btree_node_type(level, btree_id), write, &buf);
320 mustfix_fsck_err(c, "%s", buf.buf);
322 le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
323 memmove(k, bkey_next(k), next - (void *) bkey_next(k));
324 journal_entry_null_range(vstruct_next(entry), next);
327 return FSCK_DELETED_KEY;
331 bch2_bkey_compat(level, btree_id, version, big_endian,
332 write, NULL, bkey_to_packed(k));
338 static int journal_entry_btree_keys_validate(struct bch_fs *c,
340 struct jset_entry *entry,
341 unsigned version, int big_endian,
342 enum bkey_invalid_flags flags)
344 struct bkey_i *k = entry->start;
346 while (k != vstruct_last(entry)) {
347 int ret = journal_validate_key(c, jset, entry,
350 k, version, big_endian,
351 flags|BKEY_INVALID_JOURNAL);
352 if (ret == FSCK_DELETED_KEY)
361 static void journal_entry_btree_keys_to_text(struct printbuf *out, struct bch_fs *c,
362 struct jset_entry *entry)
367 jset_entry_for_each_key(entry, k) {
370 prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]);
372 prt_printf(out, "btree=%s l=%u ", bch2_btree_id_str(entry->btree_id), entry->level);
373 bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(k));
378 static int journal_entry_btree_root_validate(struct bch_fs *c,
380 struct jset_entry *entry,
381 unsigned version, int big_endian,
382 enum bkey_invalid_flags flags)
384 struct bkey_i *k = entry->start;
387 if (journal_entry_err_on(!entry->u64s ||
388 le16_to_cpu(entry->u64s) != k->k.u64s,
389 c, version, jset, entry,
390 "invalid btree root journal entry: wrong number of keys")) {
391 void *next = vstruct_next(entry);
393 * we don't want to null out this jset_entry,
394 * just the contents, so that later we can tell
395 * we were _supposed_ to have a btree root
398 journal_entry_null_range(vstruct_next(entry), next);
402 return journal_validate_key(c, jset, entry, 1, entry->btree_id, k,
403 version, big_endian, flags);
408 static void journal_entry_btree_root_to_text(struct printbuf *out, struct bch_fs *c,
409 struct jset_entry *entry)
411 journal_entry_btree_keys_to_text(out, c, entry);
414 static int journal_entry_prio_ptrs_validate(struct bch_fs *c,
416 struct jset_entry *entry,
417 unsigned version, int big_endian,
418 enum bkey_invalid_flags flags)
420 /* obsolete, don't care: */
424 static void journal_entry_prio_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
425 struct jset_entry *entry)
429 static int journal_entry_blacklist_validate(struct bch_fs *c,
431 struct jset_entry *entry,
432 unsigned version, int big_endian,
433 enum bkey_invalid_flags flags)
437 if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 1,
438 c, version, jset, entry,
439 "invalid journal seq blacklist entry: bad size")) {
440 journal_entry_null_range(entry, vstruct_next(entry));
446 static void journal_entry_blacklist_to_text(struct printbuf *out, struct bch_fs *c,
447 struct jset_entry *entry)
449 struct jset_entry_blacklist *bl =
450 container_of(entry, struct jset_entry_blacklist, entry);
452 prt_printf(out, "seq=%llu", le64_to_cpu(bl->seq));
455 static int journal_entry_blacklist_v2_validate(struct bch_fs *c,
457 struct jset_entry *entry,
458 unsigned version, int big_endian,
459 enum bkey_invalid_flags flags)
461 struct jset_entry_blacklist_v2 *bl_entry;
464 if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 2,
465 c, version, jset, entry,
466 "invalid journal seq blacklist entry: bad size")) {
467 journal_entry_null_range(entry, vstruct_next(entry));
471 bl_entry = container_of(entry, struct jset_entry_blacklist_v2, entry);
473 if (journal_entry_err_on(le64_to_cpu(bl_entry->start) >
474 le64_to_cpu(bl_entry->end),
475 c, version, jset, entry,
476 "invalid journal seq blacklist entry: start > end")) {
477 journal_entry_null_range(entry, vstruct_next(entry));
484 static void journal_entry_blacklist_v2_to_text(struct printbuf *out, struct bch_fs *c,
485 struct jset_entry *entry)
487 struct jset_entry_blacklist_v2 *bl =
488 container_of(entry, struct jset_entry_blacklist_v2, entry);
490 prt_printf(out, "start=%llu end=%llu",
491 le64_to_cpu(bl->start),
492 le64_to_cpu(bl->end));
495 static int journal_entry_usage_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_usage *u =
502 container_of(entry, struct jset_entry_usage, entry);
503 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
506 if (journal_entry_err_on(bytes < sizeof(*u),
507 c, version, jset, entry,
508 "invalid journal entry usage: bad size")) {
509 journal_entry_null_range(entry, vstruct_next(entry));
517 static void journal_entry_usage_to_text(struct printbuf *out, struct bch_fs *c,
518 struct jset_entry *entry)
520 struct jset_entry_usage *u =
521 container_of(entry, struct jset_entry_usage, entry);
523 prt_printf(out, "type=%s v=%llu",
524 bch2_fs_usage_types[u->entry.btree_id],
528 static int journal_entry_data_usage_validate(struct bch_fs *c,
530 struct jset_entry *entry,
531 unsigned version, int big_endian,
532 enum bkey_invalid_flags flags)
534 struct jset_entry_data_usage *u =
535 container_of(entry, struct jset_entry_data_usage, entry);
536 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
539 if (journal_entry_err_on(bytes < sizeof(*u) ||
540 bytes < sizeof(*u) + u->r.nr_devs,
541 c, version, jset, entry,
542 "invalid journal entry usage: bad size")) {
543 journal_entry_null_range(entry, vstruct_next(entry));
551 static void journal_entry_data_usage_to_text(struct printbuf *out, struct bch_fs *c,
552 struct jset_entry *entry)
554 struct jset_entry_data_usage *u =
555 container_of(entry, struct jset_entry_data_usage, entry);
557 bch2_replicas_entry_to_text(out, &u->r);
558 prt_printf(out, "=%llu", le64_to_cpu(u->v));
561 static int journal_entry_clock_validate(struct bch_fs *c,
563 struct jset_entry *entry,
564 unsigned version, int big_endian,
565 enum bkey_invalid_flags flags)
567 struct jset_entry_clock *clock =
568 container_of(entry, struct jset_entry_clock, entry);
569 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
572 if (journal_entry_err_on(bytes != sizeof(*clock),
573 c, version, jset, entry, "bad size")) {
574 journal_entry_null_range(entry, vstruct_next(entry));
578 if (journal_entry_err_on(clock->rw > 1,
579 c, version, jset, entry, "bad rw")) {
580 journal_entry_null_range(entry, vstruct_next(entry));
588 static void journal_entry_clock_to_text(struct printbuf *out, struct bch_fs *c,
589 struct jset_entry *entry)
591 struct jset_entry_clock *clock =
592 container_of(entry, struct jset_entry_clock, entry);
594 prt_printf(out, "%s=%llu", clock->rw ? "write" : "read", le64_to_cpu(clock->time));
597 static int journal_entry_dev_usage_validate(struct bch_fs *c,
599 struct jset_entry *entry,
600 unsigned version, int big_endian,
601 enum bkey_invalid_flags flags)
603 struct jset_entry_dev_usage *u =
604 container_of(entry, struct jset_entry_dev_usage, entry);
605 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
606 unsigned expected = sizeof(*u);
610 if (journal_entry_err_on(bytes < expected,
611 c, version, jset, entry, "bad size (%u < %u)",
613 journal_entry_null_range(entry, vstruct_next(entry));
617 dev = le32_to_cpu(u->dev);
619 if (journal_entry_err_on(!bch2_dev_exists2(c, dev),
620 c, version, jset, entry, "bad dev")) {
621 journal_entry_null_range(entry, vstruct_next(entry));
625 if (journal_entry_err_on(u->pad,
626 c, version, jset, entry, "bad pad")) {
627 journal_entry_null_range(entry, vstruct_next(entry));
635 static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs *c,
636 struct jset_entry *entry)
638 struct jset_entry_dev_usage *u =
639 container_of(entry, struct jset_entry_dev_usage, entry);
640 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
642 prt_printf(out, "dev=%u", le32_to_cpu(u->dev));
644 for (i = 0; i < nr_types; i++) {
646 prt_printf(out, " %s", bch2_data_types[i]);
648 prt_printf(out, " (unknown data type %u)", i);
649 prt_printf(out, ": buckets=%llu sectors=%llu fragmented=%llu",
650 le64_to_cpu(u->d[i].buckets),
651 le64_to_cpu(u->d[i].sectors),
652 le64_to_cpu(u->d[i].fragmented));
655 prt_printf(out, " buckets_ec: %llu", le64_to_cpu(u->buckets_ec));
658 static int journal_entry_log_validate(struct bch_fs *c,
660 struct jset_entry *entry,
661 unsigned version, int big_endian,
662 enum bkey_invalid_flags flags)
667 static void journal_entry_log_to_text(struct printbuf *out, struct bch_fs *c,
668 struct jset_entry *entry)
670 struct jset_entry_log *l = container_of(entry, struct jset_entry_log, entry);
671 unsigned bytes = vstruct_bytes(entry) - offsetof(struct jset_entry_log, d);
673 prt_printf(out, "%.*s", bytes, l->d);
676 static int journal_entry_overwrite_validate(struct bch_fs *c,
678 struct jset_entry *entry,
679 unsigned version, int big_endian,
680 enum bkey_invalid_flags flags)
682 return journal_entry_btree_keys_validate(c, jset, entry,
683 version, big_endian, READ);
686 static void journal_entry_overwrite_to_text(struct printbuf *out, struct bch_fs *c,
687 struct jset_entry *entry)
689 journal_entry_btree_keys_to_text(out, c, entry);
692 struct jset_entry_ops {
693 int (*validate)(struct bch_fs *, struct jset *,
694 struct jset_entry *, unsigned, int,
695 enum bkey_invalid_flags);
696 void (*to_text)(struct printbuf *, struct bch_fs *, struct jset_entry *);
699 static const struct jset_entry_ops bch2_jset_entry_ops[] = {
701 [BCH_JSET_ENTRY_##f] = (struct jset_entry_ops) { \
702 .validate = journal_entry_##f##_validate, \
703 .to_text = journal_entry_##f##_to_text, \
705 BCH_JSET_ENTRY_TYPES()
709 int bch2_journal_entry_validate(struct bch_fs *c,
711 struct jset_entry *entry,
712 unsigned version, int big_endian,
713 enum bkey_invalid_flags flags)
715 return entry->type < BCH_JSET_ENTRY_NR
716 ? bch2_jset_entry_ops[entry->type].validate(c, jset, entry,
717 version, big_endian, flags)
721 void bch2_journal_entry_to_text(struct printbuf *out, struct bch_fs *c,
722 struct jset_entry *entry)
724 if (entry->type < BCH_JSET_ENTRY_NR) {
725 prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]);
726 bch2_jset_entry_ops[entry->type].to_text(out, c, entry);
728 prt_printf(out, "(unknown type %u)", entry->type);
732 static int jset_validate_entries(struct bch_fs *c, struct jset *jset,
733 enum bkey_invalid_flags flags)
735 struct jset_entry *entry;
736 unsigned version = le32_to_cpu(jset->version);
739 vstruct_for_each(jset, entry) {
740 if (journal_entry_err_on(vstruct_next(entry) > vstruct_last(jset),
741 c, version, jset, entry,
742 "journal entry extends past end of jset")) {
743 jset->u64s = cpu_to_le32((u64 *) entry - jset->_data);
747 ret = bch2_journal_entry_validate(c, jset, entry,
748 version, JSET_BIG_ENDIAN(jset), flags);
756 static int jset_validate(struct bch_fs *c,
758 struct jset *jset, u64 sector,
759 enum bkey_invalid_flags flags)
764 if (le64_to_cpu(jset->magic) != jset_magic(c))
765 return JOURNAL_ENTRY_NONE;
767 version = le32_to_cpu(jset->version);
768 if (journal_entry_err_on(!bch2_version_compatible(version),
769 c, version, jset, NULL,
770 "%s sector %llu seq %llu: incompatible journal entry version %u.%u",
771 ca ? ca->name : c->name,
772 sector, le64_to_cpu(jset->seq),
773 BCH_VERSION_MAJOR(version),
774 BCH_VERSION_MINOR(version))) {
775 /* don't try to continue: */
779 if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)),
780 c, version, jset, NULL,
781 "%s sector %llu seq %llu: journal entry with unknown csum type %llu",
782 ca ? ca->name : c->name,
783 sector, le64_to_cpu(jset->seq),
784 JSET_CSUM_TYPE(jset)))
785 ret = JOURNAL_ENTRY_BAD;
787 /* last_seq is ignored when JSET_NO_FLUSH is true */
788 if (journal_entry_err_on(!JSET_NO_FLUSH(jset) &&
789 le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq),
790 c, version, jset, NULL,
791 "invalid journal entry: last_seq > seq (%llu > %llu)",
792 le64_to_cpu(jset->last_seq),
793 le64_to_cpu(jset->seq))) {
794 jset->last_seq = jset->seq;
795 return JOURNAL_ENTRY_BAD;
798 ret = jset_validate_entries(c, jset, flags);
803 static int jset_validate_early(struct bch_fs *c,
805 struct jset *jset, u64 sector,
806 unsigned bucket_sectors_left,
807 unsigned sectors_read)
809 size_t bytes = vstruct_bytes(jset);
811 enum bkey_invalid_flags flags = BKEY_INVALID_JOURNAL;
814 if (le64_to_cpu(jset->magic) != jset_magic(c))
815 return JOURNAL_ENTRY_NONE;
817 version = le32_to_cpu(jset->version);
818 if (journal_entry_err_on(!bch2_version_compatible(version),
819 c, version, jset, NULL,
820 "%s sector %llu seq %llu: unknown journal entry version %u.%u",
821 ca ? ca->name : c->name,
822 sector, le64_to_cpu(jset->seq),
823 BCH_VERSION_MAJOR(version),
824 BCH_VERSION_MINOR(version))) {
825 /* don't try to continue: */
829 if (bytes > (sectors_read << 9) &&
830 sectors_read < bucket_sectors_left)
831 return JOURNAL_ENTRY_REREAD;
833 if (journal_entry_err_on(bytes > bucket_sectors_left << 9,
834 c, version, jset, NULL,
835 "%s sector %llu seq %llu: journal entry too big (%zu bytes)",
836 ca ? ca->name : c->name,
837 sector, le64_to_cpu(jset->seq), bytes))
838 le32_add_cpu(&jset->u64s,
839 -((bytes - (bucket_sectors_left << 9)) / 8));
844 struct journal_read_buf {
849 static int journal_read_buf_realloc(struct journal_read_buf *b,
854 /* the bios are sized for this many pages, max: */
855 if (new_size > JOURNAL_ENTRY_SIZE_MAX)
856 return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
858 new_size = roundup_pow_of_two(new_size);
859 n = kvpmalloc(new_size, GFP_KERNEL);
861 return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
863 kvpfree(b->data, b->size);
869 static int journal_read_bucket(struct bch_dev *ca,
870 struct journal_read_buf *buf,
871 struct journal_list *jlist,
874 struct bch_fs *c = ca->fs;
875 struct journal_device *ja = &ca->journal;
876 struct jset *j = NULL;
877 unsigned sectors, sectors_read = 0;
878 u64 offset = bucket_to_sector(ca, ja->buckets[bucket]),
879 end = offset + ca->mi.bucket_size;
880 bool saw_bad = false, csum_good;
883 pr_debug("reading %u", bucket);
885 while (offset < end) {
890 sectors_read = min_t(unsigned,
891 end - offset, buf->size >> 9);
892 nr_bvecs = buf_pages(buf->data, sectors_read << 9);
894 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
895 bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ);
897 bio->bi_iter.bi_sector = offset;
898 bch2_bio_map(bio, buf->data, sectors_read << 9);
900 ret = submit_bio_wait(bio);
903 if (bch2_dev_io_err_on(ret, ca,
904 "journal read error: sector %llu",
906 bch2_meta_read_fault("journal")) {
908 * We don't error out of the recovery process
909 * here, since the relevant journal entry may be
910 * found on a different device, and missing or
911 * no journal entries will be handled later
919 ret = jset_validate_early(c, ca, j, offset,
920 end - offset, sectors_read);
923 sectors = vstruct_sectors(j, c->block_bits);
925 case JOURNAL_ENTRY_REREAD:
926 if (vstruct_bytes(j) > buf->size) {
927 ret = journal_read_buf_realloc(buf,
933 case JOURNAL_ENTRY_NONE:
937 * On checksum error we don't really trust the size
938 * field of the journal entry we read, so try reading
939 * again at next block boundary:
941 sectors = block_sectors(c);
948 * This happens sometimes if we don't have discards on -
949 * when we've partially overwritten a bucket with new
950 * journal entries. We don't need the rest of the
953 if (le64_to_cpu(j->seq) < ja->bucket_seq[bucket])
956 ja->bucket_seq[bucket] = le64_to_cpu(j->seq);
958 csum_good = jset_csum_good(c, j);
962 ret = bch2_encrypt(c, JSET_CSUM_TYPE(j), journal_nonce(j),
964 vstruct_end(j) - (void *) j->encrypted_start);
965 bch2_fs_fatal_err_on(ret, c,
966 "error decrypting journal entry: %i", ret);
968 mutex_lock(&jlist->lock);
969 ret = journal_entry_add(c, ca, (struct journal_ptr) {
970 .csum_good = csum_good,
973 .bucket_offset = offset -
974 bucket_to_sector(ca, ja->buckets[bucket]),
977 mutex_unlock(&jlist->lock);
980 case JOURNAL_ENTRY_ADD_OK:
982 case JOURNAL_ENTRY_ADD_OUT_OF_RANGE:
990 sectors_read -= sectors;
991 j = ((void *) j) + (sectors << 9);
997 static void bch2_journal_read_device(struct closure *cl)
999 struct journal_device *ja =
1000 container_of(cl, struct journal_device, read);
1001 struct bch_dev *ca = container_of(ja, struct bch_dev, journal);
1002 struct bch_fs *c = ca->fs;
1003 struct journal_list *jlist =
1004 container_of(cl->parent, struct journal_list, cl);
1005 struct journal_replay *r, **_r;
1006 struct genradix_iter iter;
1007 struct journal_read_buf buf = { NULL, 0 };
1014 ret = journal_read_buf_realloc(&buf, PAGE_SIZE);
1018 pr_debug("%u journal buckets", ja->nr);
1020 for (i = 0; i < ja->nr; i++) {
1021 ret = journal_read_bucket(ca, &buf, jlist, i);
1026 ja->sectors_free = ca->mi.bucket_size;
1028 mutex_lock(&jlist->lock);
1029 genradix_for_each_reverse(&c->journal_entries, iter, _r) {
1035 for (i = 0; i < r->nr_ptrs; i++) {
1036 if (r->ptrs[i].dev == ca->dev_idx) {
1037 unsigned wrote = bucket_remainder(ca, r->ptrs[i].sector) +
1038 vstruct_sectors(&r->j, c->block_bits);
1040 ja->cur_idx = r->ptrs[i].bucket;
1041 ja->sectors_free = ca->mi.bucket_size - wrote;
1047 mutex_unlock(&jlist->lock);
1049 if (ja->bucket_seq[ja->cur_idx] &&
1050 ja->sectors_free == ca->mi.bucket_size) {
1051 bch_err(c, "ja->sectors_free == ca->mi.bucket_size");
1052 bch_err(c, "cur_idx %u/%u", ja->cur_idx, ja->nr);
1053 for (i = 0; i < 3; i++) {
1054 unsigned idx = (ja->cur_idx + ja->nr - 1 + i) % ja->nr;
1056 bch_err(c, "bucket_seq[%u] = %llu", idx, ja->bucket_seq[idx]);
1058 ja->sectors_free = 0;
1062 * Set dirty_idx to indicate the entire journal is full and needs to be
1063 * reclaimed - journal reclaim will immediately reclaim whatever isn't
1064 * pinned when it first runs:
1066 ja->discard_idx = ja->dirty_idx_ondisk =
1067 ja->dirty_idx = (ja->cur_idx + 1) % ja->nr;
1069 bch_verbose(c, "journal read done on device %s, ret %i", ca->name, ret);
1070 kvpfree(buf.data, buf.size);
1071 percpu_ref_put(&ca->io_ref);
1075 mutex_lock(&jlist->lock);
1077 mutex_unlock(&jlist->lock);
1081 void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
1082 struct journal_replay *j)
1086 for (i = 0; i < j->nr_ptrs; i++) {
1087 struct bch_dev *ca = bch_dev_bkey_exists(c, j->ptrs[i].dev);
1090 div64_u64_rem(j->ptrs[i].sector, ca->mi.bucket_size, &offset);
1093 prt_printf(out, " ");
1094 prt_printf(out, "%u:%u:%u (sector %llu)",
1097 j->ptrs[i].bucket_offset,
1102 int bch2_journal_read(struct bch_fs *c,
1107 struct journal_list jlist;
1108 struct journal_replay *i, **_i, *prev = NULL;
1109 struct genradix_iter radix_iter;
1112 struct printbuf buf = PRINTBUF;
1113 bool degraded = false, last_write_torn = false;
1117 closure_init_stack(&jlist.cl);
1118 mutex_init(&jlist.lock);
1122 for_each_member_device(ca, c, iter) {
1123 if (!c->opts.fsck &&
1124 !(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_journal)))
1127 if ((ca->mi.state == BCH_MEMBER_STATE_rw ||
1128 ca->mi.state == BCH_MEMBER_STATE_ro) &&
1129 percpu_ref_tryget(&ca->io_ref))
1130 closure_call(&ca->journal.read,
1131 bch2_journal_read_device,
1138 closure_sync(&jlist.cl);
1148 * Find most recent flush entry, and ignore newer non flush entries -
1149 * those entries will be blacklisted:
1151 genradix_for_each_reverse(&c->journal_entries, radix_iter, _i) {
1152 enum bkey_invalid_flags flags = BKEY_INVALID_JOURNAL;
1156 if (!i || i->ignore)
1160 *blacklist_seq = *start_seq = le64_to_cpu(i->j.seq) + 1;
1162 if (JSET_NO_FLUSH(&i->j)) {
1167 if (!last_write_torn && !i->csum_good) {
1168 last_write_torn = true;
1173 if (journal_entry_err_on(le64_to_cpu(i->j.last_seq) > le64_to_cpu(i->j.seq),
1174 c, le32_to_cpu(i->j.version), &i->j, NULL,
1175 "invalid journal entry: last_seq > seq (%llu > %llu)",
1176 le64_to_cpu(i->j.last_seq),
1177 le64_to_cpu(i->j.seq)))
1178 i->j.last_seq = i->j.seq;
1180 *last_seq = le64_to_cpu(i->j.last_seq);
1181 *blacklist_seq = le64_to_cpu(i->j.seq) + 1;
1186 bch_info(c, "journal read done, but no entries found");
1191 fsck_err(c, "journal read done, but no entries found after dropping non-flushes");
1195 bch_info(c, "journal read done, replaying entries %llu-%llu",
1196 *last_seq, *blacklist_seq - 1);
1198 if (*start_seq != *blacklist_seq)
1199 bch_info(c, "dropped unflushed entries %llu-%llu",
1200 *blacklist_seq, *start_seq - 1);
1202 /* Drop blacklisted entries and entries older than last_seq: */
1203 genradix_for_each(&c->journal_entries, radix_iter, _i) {
1206 if (!i || i->ignore)
1209 seq = le64_to_cpu(i->j.seq);
1210 if (seq < *last_seq) {
1211 journal_replay_free(c, i);
1215 if (bch2_journal_seq_is_blacklisted(c, seq, true)) {
1216 fsck_err_on(!JSET_NO_FLUSH(&i->j), c,
1217 "found blacklisted journal entry %llu", seq);
1222 /* Check for missing entries: */
1224 genradix_for_each(&c->journal_entries, radix_iter, _i) {
1227 if (!i || i->ignore)
1230 BUG_ON(seq > le64_to_cpu(i->j.seq));
1232 while (seq < le64_to_cpu(i->j.seq)) {
1233 u64 missing_start, missing_end;
1234 struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
1236 while (seq < le64_to_cpu(i->j.seq) &&
1237 bch2_journal_seq_is_blacklisted(c, seq, false))
1240 if (seq == le64_to_cpu(i->j.seq))
1243 missing_start = seq;
1245 while (seq < le64_to_cpu(i->j.seq) &&
1246 !bch2_journal_seq_is_blacklisted(c, seq, false))
1250 bch2_journal_ptrs_to_text(&buf1, c, prev);
1251 prt_printf(&buf1, " size %zu", vstruct_sectors(&prev->j, c->block_bits));
1253 prt_printf(&buf1, "(none)");
1254 bch2_journal_ptrs_to_text(&buf2, c, i);
1256 missing_end = seq - 1;
1257 fsck_err(c, "journal entries %llu-%llu missing! (replaying %llu-%llu)\n"
1260 missing_start, missing_end,
1261 *last_seq, *blacklist_seq - 1,
1262 buf1.buf, buf2.buf);
1264 printbuf_exit(&buf1);
1265 printbuf_exit(&buf2);
1272 genradix_for_each(&c->journal_entries, radix_iter, _i) {
1273 struct bch_replicas_padded replicas = {
1274 .e.data_type = BCH_DATA_journal,
1280 if (!i || i->ignore)
1283 for (ptr = 0; ptr < i->nr_ptrs; ptr++) {
1284 ca = bch_dev_bkey_exists(c, i->ptrs[ptr].dev);
1286 if (!i->ptrs[ptr].csum_good)
1287 bch_err_dev_offset(ca, i->ptrs[ptr].sector,
1288 "invalid journal checksum, seq %llu%s",
1289 le64_to_cpu(i->j.seq),
1290 i->csum_good ? " (had good copy on another device)" : "");
1293 ret = jset_validate(c,
1294 bch_dev_bkey_exists(c, i->ptrs[0].dev),
1301 for (ptr = 0; ptr < i->nr_ptrs; ptr++)
1302 replicas.e.devs[replicas.e.nr_devs++] = i->ptrs[ptr].dev;
1304 bch2_replicas_entry_sort(&replicas.e);
1306 printbuf_reset(&buf);
1307 bch2_replicas_entry_to_text(&buf, &replicas.e);
1310 !bch2_replicas_marked(c, &replicas.e) &&
1311 (le64_to_cpu(i->j.seq) == *last_seq ||
1312 fsck_err(c, "superblock not marked as containing replicas for journal entry %llu\n %s",
1313 le64_to_cpu(i->j.seq), buf.buf))) {
1314 ret = bch2_mark_replicas(c, &replicas.e);
1321 printbuf_exit(&buf);
1325 /* journal write: */
1327 static void __journal_write_alloc(struct journal *j,
1328 struct journal_buf *w,
1329 struct dev_alloc_list *devs_sorted,
1332 unsigned replicas_want)
1334 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1335 struct journal_device *ja;
1339 if (*replicas >= replicas_want)
1342 for (i = 0; i < devs_sorted->nr; i++) {
1343 ca = rcu_dereference(c->devs[devs_sorted->devs[i]]);
1350 * Check that we can use this device, and aren't already using
1353 if (!ca->mi.durability ||
1354 ca->mi.state != BCH_MEMBER_STATE_rw ||
1356 bch2_bkey_has_device_c(bkey_i_to_s_c(&w->key), ca->dev_idx) ||
1357 sectors > ja->sectors_free)
1360 bch2_dev_stripe_increment(ca, &j->wp.stripe);
1362 bch2_bkey_append_ptr(&w->key,
1363 (struct bch_extent_ptr) {
1364 .offset = bucket_to_sector(ca,
1365 ja->buckets[ja->cur_idx]) +
1366 ca->mi.bucket_size -
1371 ja->sectors_free -= sectors;
1372 ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1374 *replicas += ca->mi.durability;
1376 if (*replicas >= replicas_want)
1382 * journal_write_alloc - decide where to write next journal entry
1384 * @j: journal object
1385 * @w: journal buf (entry to be written)
1387 * Returns: 0 on success, or -EROFS on failure
1389 static int journal_write_alloc(struct journal *j, struct journal_buf *w)
1391 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1392 struct bch_devs_mask devs;
1393 struct journal_device *ja;
1395 struct dev_alloc_list devs_sorted;
1396 unsigned sectors = vstruct_sectors(w->data, c->block_bits);
1397 unsigned target = c->opts.metadata_target ?:
1398 c->opts.foreground_target;
1399 unsigned i, replicas = 0, replicas_want =
1400 READ_ONCE(c->opts.metadata_replicas);
1404 devs = target_rw_devs(c, BCH_DATA_journal, target);
1406 devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe, &devs);
1408 __journal_write_alloc(j, w, &devs_sorted,
1409 sectors, &replicas, replicas_want);
1411 if (replicas >= replicas_want)
1414 for (i = 0; i < devs_sorted.nr; i++) {
1415 ca = rcu_dereference(c->devs[devs_sorted.devs[i]]);
1421 if (sectors > ja->sectors_free &&
1422 sectors <= ca->mi.bucket_size &&
1423 bch2_journal_dev_buckets_available(j, ja,
1424 journal_space_discarded)) {
1425 ja->cur_idx = (ja->cur_idx + 1) % ja->nr;
1426 ja->sectors_free = ca->mi.bucket_size;
1429 * ja->bucket_seq[ja->cur_idx] must always have
1430 * something sensible:
1432 ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1436 __journal_write_alloc(j, w, &devs_sorted,
1437 sectors, &replicas, replicas_want);
1439 if (replicas < replicas_want && target) {
1440 /* Retry from all devices: */
1447 BUG_ON(bkey_val_u64s(&w->key.k) > BCH_REPLICAS_MAX);
1449 return replicas >= c->opts.metadata_replicas_required ? 0 : -EROFS;
1452 static void journal_buf_realloc(struct journal *j, struct journal_buf *buf)
1454 /* we aren't holding j->lock: */
1455 unsigned new_size = READ_ONCE(j->buf_size_want);
1458 if (buf->buf_size >= new_size)
1461 new_buf = kvpmalloc(new_size, GFP_NOFS|__GFP_NOWARN);
1465 memcpy(new_buf, buf->data, buf->buf_size);
1467 spin_lock(&j->lock);
1468 swap(buf->data, new_buf);
1469 swap(buf->buf_size, new_size);
1470 spin_unlock(&j->lock);
1472 kvpfree(new_buf, new_size);
1475 static inline struct journal_buf *journal_last_unwritten_buf(struct journal *j)
1477 return j->buf + (journal_last_unwritten_seq(j) & JOURNAL_BUF_MASK);
1480 static void journal_write_done(struct closure *cl)
1482 struct journal *j = container_of(cl, struct journal, io);
1483 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1484 struct journal_buf *w = journal_last_unwritten_buf(j);
1485 struct bch_replicas_padded replicas;
1486 union journal_res_state old, new;
1490 bch2_time_stats_update(!JSET_NO_FLUSH(w->data)
1491 ? j->flush_write_time
1492 : j->noflush_write_time, j->write_start_time);
1494 if (!w->devs_written.nr) {
1495 bch_err(c, "unable to write journal to sufficient devices");
1498 bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
1500 if (bch2_mark_replicas(c, &replicas.e))
1505 bch2_fatal_error(c);
1507 spin_lock(&j->lock);
1508 seq = le64_to_cpu(w->data->seq);
1510 if (seq >= j->pin.front)
1511 journal_seq_pin(j, seq)->devs = w->devs_written;
1514 if (!JSET_NO_FLUSH(w->data)) {
1515 j->flushed_seq_ondisk = seq;
1516 j->last_seq_ondisk = w->last_seq;
1518 bch2_do_discards(c);
1519 closure_wake_up(&c->freelist_wait);
1521 bch2_reset_alloc_cursors(c);
1523 } else if (!j->err_seq || seq < j->err_seq)
1526 j->seq_ondisk = seq;
1529 * Updating last_seq_ondisk may let bch2_journal_reclaim_work() discard
1532 * Must come before signaling write completion, for
1533 * bch2_fs_journal_stop():
1535 if (j->watermark != BCH_WATERMARK_stripe)
1536 journal_reclaim_kick(&c->journal);
1538 /* also must come before signalling write completion: */
1539 closure_debug_destroy(cl);
1541 v = atomic64_read(&j->reservations.counter);
1544 BUG_ON(journal_state_count(new, new.unwritten_idx));
1546 new.unwritten_idx++;
1547 } while ((v = atomic64_cmpxchg(&j->reservations.counter,
1548 old.v, new.v)) != old.v);
1550 bch2_journal_space_available(j);
1552 closure_wake_up(&w->wait);
1555 if (!journal_state_count(new, new.unwritten_idx) &&
1556 journal_last_unwritten_seq(j) <= journal_cur_seq(j)) {
1557 spin_unlock(&j->lock);
1558 closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
1559 } else if (journal_last_unwritten_seq(j) == journal_cur_seq(j) &&
1560 new.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL) {
1561 struct journal_buf *buf = journal_cur_buf(j);
1562 long delta = buf->expires - jiffies;
1565 * We don't close a journal entry to write it while there's
1566 * previous entries still in flight - the current journal entry
1567 * might want to be written now:
1570 spin_unlock(&j->lock);
1571 mod_delayed_work(c->io_complete_wq, &j->write_work, max(0L, delta));
1573 spin_unlock(&j->lock);
1577 static void journal_write_endio(struct bio *bio)
1579 struct bch_dev *ca = bio->bi_private;
1580 struct journal *j = &ca->fs->journal;
1581 struct journal_buf *w = journal_last_unwritten_buf(j);
1582 unsigned long flags;
1584 if (bch2_dev_io_err_on(bio->bi_status, ca, "error writing journal entry %llu: %s",
1585 le64_to_cpu(w->data->seq),
1586 bch2_blk_status_to_str(bio->bi_status)) ||
1587 bch2_meta_write_fault("journal")) {
1588 spin_lock_irqsave(&j->err_lock, flags);
1589 bch2_dev_list_drop_dev(&w->devs_written, ca->dev_idx);
1590 spin_unlock_irqrestore(&j->err_lock, flags);
1593 closure_put(&j->io);
1594 percpu_ref_put(&ca->io_ref);
1597 static void do_journal_write(struct closure *cl)
1599 struct journal *j = container_of(cl, struct journal, io);
1600 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1602 struct journal_buf *w = journal_last_unwritten_buf(j);
1603 struct bch_extent_ptr *ptr;
1605 unsigned sectors = vstruct_sectors(w->data, c->block_bits);
1607 extent_for_each_ptr(bkey_i_to_s_extent(&w->key), ptr) {
1608 ca = bch_dev_bkey_exists(c, ptr->dev);
1609 if (!percpu_ref_tryget(&ca->io_ref)) {
1611 bch_err(c, "missing device for journal write\n");
1615 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_journal],
1618 bio = ca->journal.bio;
1619 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
1620 bio->bi_iter.bi_sector = ptr->offset;
1621 bio->bi_end_io = journal_write_endio;
1622 bio->bi_private = ca;
1624 BUG_ON(bio->bi_iter.bi_sector == ca->prev_journal_sector);
1625 ca->prev_journal_sector = bio->bi_iter.bi_sector;
1627 if (!JSET_NO_FLUSH(w->data))
1628 bio->bi_opf |= REQ_FUA;
1629 if (!JSET_NO_FLUSH(w->data) && !w->separate_flush)
1630 bio->bi_opf |= REQ_PREFLUSH;
1632 bch2_bio_map(bio, w->data, sectors << 9);
1634 trace_and_count(c, journal_write, bio);
1635 closure_bio_submit(bio, cl);
1637 ca->journal.bucket_seq[ca->journal.cur_idx] =
1638 le64_to_cpu(w->data->seq);
1641 continue_at(cl, journal_write_done, c->io_complete_wq);
1644 static void bch2_journal_entries_postprocess(struct bch_fs *c, struct jset *jset)
1646 struct jset_entry *i, *next, *prev = NULL;
1649 * Simple compaction, dropping empty jset_entries (from journal
1650 * reservations that weren't fully used) and merging jset_entries that
1653 * If we wanted to be really fancy here, we could sort all the keys in
1654 * the jset and drop keys that were overwritten - probably not worth it:
1656 vstruct_for_each_safe(jset, i, next) {
1657 unsigned u64s = le16_to_cpu(i->u64s);
1663 if (i->type == BCH_JSET_ENTRY_btree_root)
1664 bch2_journal_entry_to_btree_root(c, i);
1666 /* Can we merge with previous entry? */
1668 i->btree_id == prev->btree_id &&
1669 i->level == prev->level &&
1670 i->type == prev->type &&
1671 i->type == BCH_JSET_ENTRY_btree_keys &&
1672 le16_to_cpu(prev->u64s) + u64s <= U16_MAX) {
1673 memmove_u64s_down(vstruct_next(prev),
1676 le16_add_cpu(&prev->u64s, u64s);
1680 /* Couldn't merge, move i into new position (after prev): */
1681 prev = prev ? vstruct_next(prev) : jset->start;
1683 memmove_u64s_down(prev, i, jset_u64s(u64s));
1686 prev = prev ? vstruct_next(prev) : jset->start;
1687 jset->u64s = cpu_to_le32((u64 *) prev - jset->_data);
1690 void bch2_journal_write(struct closure *cl)
1692 struct journal *j = container_of(cl, struct journal, io);
1693 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1695 struct journal_buf *w = journal_last_unwritten_buf(j);
1696 struct bch_replicas_padded replicas;
1697 struct jset_entry *start, *end;
1700 struct printbuf journal_debug_buf = PRINTBUF;
1701 bool validate_before_checksum = false;
1702 unsigned i, sectors, bytes, u64s, nr_rw_members = 0;
1705 BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
1707 journal_buf_realloc(j, w);
1710 j->write_start_time = local_clock();
1712 spin_lock(&j->lock);
1715 * If the journal is in an error state - we did an emergency shutdown -
1716 * we prefer to continue doing journal writes. We just mark them as
1717 * noflush so they'll never be used, but they'll still be visible by the
1718 * list_journal tool - this helps in debugging.
1720 * There's a caveat: the first journal write after marking the
1721 * superblock dirty must always be a flush write, because on startup
1722 * from a clean shutdown we didn't necessarily read the journal and the
1723 * new journal write might overwrite whatever was in the journal
1724 * previously - we can't leave the journal without any flush writes in
1727 * So if we're in an error state, and we're still starting up, we don't
1728 * write anything at all.
1730 if (!test_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags) &&
1731 (bch2_journal_error(j) ||
1734 (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) &&
1735 test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags)))) {
1737 SET_JSET_NO_FLUSH(jset, true);
1741 j->nr_noflush_writes++;
1742 } else if (!bch2_journal_error(j)) {
1743 j->last_flush_write = jiffies;
1744 j->nr_flush_writes++;
1745 clear_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags);
1747 spin_unlock(&j->lock);
1750 spin_unlock(&j->lock);
1753 * New btree roots are set by journalling them; when the journal entry
1754 * gets written we have to propagate them to c->btree_roots
1756 * But, every journal entry we write has to contain all the btree roots
1757 * (at least for now); so after we copy btree roots to c->btree_roots we
1758 * have to get any missing btree roots and add them to this journal
1762 bch2_journal_entries_postprocess(c, jset);
1764 start = end = vstruct_last(jset);
1766 end = bch2_btree_roots_to_journal_entries(c, jset->start, end);
1768 bch2_journal_super_entries_add_common(c, &end,
1769 le64_to_cpu(jset->seq));
1770 u64s = (u64 *) end - (u64 *) start;
1771 BUG_ON(u64s > j->entry_u64s_reserved);
1773 le32_add_cpu(&jset->u64s, u64s);
1775 sectors = vstruct_sectors(jset, c->block_bits);
1776 bytes = vstruct_bytes(jset);
1778 if (sectors > w->sectors) {
1779 bch2_fs_fatal_error(c, "aieeee! journal write overran available space, %zu > %u (extra %u reserved %u/%u)",
1780 vstruct_bytes(jset), w->sectors << 9,
1781 u64s, w->u64s_reserved, j->entry_u64s_reserved);
1785 jset->magic = cpu_to_le64(jset_magic(c));
1786 jset->version = cpu_to_le32(c->sb.version);
1788 SET_JSET_BIG_ENDIAN(jset, CPU_BIG_ENDIAN);
1789 SET_JSET_CSUM_TYPE(jset, bch2_meta_checksum_type(c));
1791 if (!JSET_NO_FLUSH(jset) && journal_entry_empty(jset))
1792 j->last_empty_seq = le64_to_cpu(jset->seq);
1794 if (bch2_csum_type_is_encryption(JSET_CSUM_TYPE(jset)))
1795 validate_before_checksum = true;
1797 if (le32_to_cpu(jset->version) < bcachefs_metadata_version_current)
1798 validate_before_checksum = true;
1800 if (validate_before_checksum &&
1801 jset_validate(c, NULL, jset, 0, WRITE))
1804 ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset),
1805 jset->encrypted_start,
1806 vstruct_end(jset) - (void *) jset->encrypted_start);
1807 if (bch2_fs_fatal_err_on(ret, c,
1808 "error decrypting journal entry: %i", ret))
1811 jset->csum = csum_vstruct(c, JSET_CSUM_TYPE(jset),
1812 journal_nonce(jset), jset);
1814 if (!validate_before_checksum &&
1815 jset_validate(c, NULL, jset, 0, WRITE))
1818 memset((void *) jset + bytes, 0, (sectors << 9) - bytes);
1821 spin_lock(&j->lock);
1822 ret = journal_write_alloc(j, w);
1824 if (ret && j->can_discard) {
1825 spin_unlock(&j->lock);
1826 bch2_journal_do_discards(j);
1831 __bch2_journal_debug_to_text(&journal_debug_buf, j);
1834 * write is allocated, no longer need to account for it in
1835 * bch2_journal_space_available():
1840 * journal entry has been compacted and allocated, recalculate space
1843 bch2_journal_space_available(j);
1844 spin_unlock(&j->lock);
1847 bch_err(c, "Unable to allocate journal write:\n%s",
1848 journal_debug_buf.buf);
1849 printbuf_exit(&journal_debug_buf);
1853 w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key));
1855 if (c->opts.nochanges)
1858 for_each_rw_member(ca, c, i)
1861 if (nr_rw_members > 1)
1862 w->separate_flush = true;
1865 * Mark journal replicas before we submit the write to guarantee
1866 * recovery will find the journal entries after a crash.
1868 bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
1870 ret = bch2_mark_replicas(c, &replicas.e);
1874 if (!JSET_NO_FLUSH(jset) && w->separate_flush) {
1875 for_each_rw_member(ca, c, i) {
1876 percpu_ref_get(&ca->io_ref);
1878 bio = ca->journal.bio;
1879 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_FLUSH);
1880 bio->bi_end_io = journal_write_endio;
1881 bio->bi_private = ca;
1882 closure_bio_submit(bio, cl);
1886 continue_at(cl, do_journal_write, c->io_complete_wq);
1889 continue_at(cl, journal_write_done, c->io_complete_wq);
1892 bch2_fatal_error(c);
1893 continue_at(cl, journal_write_done, c->io_complete_wq);