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"
13 #include "journal_io.h"
14 #include "journal_reclaim.h"
15 #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,
212 struct jset_entry *entry)
214 prt_str(out, "invalid journal entry ");
216 prt_printf(out, "%s ", bch2_jset_entry_types[entry->type]);
219 prt_printf(out, "in superblock");
221 prt_printf(out, "at seq %llu", le64_to_cpu(jset->seq));
223 prt_printf(out, "at offset %zi/%u seq %llu",
224 (u64 *) entry - jset->_data,
225 le32_to_cpu(jset->u64s),
226 le64_to_cpu(jset->seq));
230 #define journal_entry_err(c, jset, entry, msg, ...) \
232 struct printbuf buf = PRINTBUF; \
234 journal_entry_err_msg(&buf, jset, entry); \
235 prt_printf(&buf, msg, ##__VA_ARGS__); \
239 mustfix_fsck_err(c, "%s", buf.buf); \
242 bch_err(c, "corrupt metadata before write: %s\n", buf.buf);\
243 if (bch2_fs_inconsistent(c)) { \
244 ret = -BCH_ERR_fsck_errors_not_fixed; \
250 printbuf_exit(&buf); \
254 #define journal_entry_err_on(cond, c, jset, entry, msg, ...) \
255 ((cond) ? journal_entry_err(c, jset, entry, msg, ##__VA_ARGS__) : false)
257 #define FSCK_DELETED_KEY 5
259 static int journal_validate_key(struct bch_fs *c,
261 struct jset_entry *entry,
262 unsigned level, enum btree_id btree_id,
264 unsigned version, int big_endian, int write)
266 void *next = vstruct_next(entry);
267 struct printbuf buf = PRINTBUF;
270 if (journal_entry_err_on(!k->k.u64s, c, jset, entry, "k->u64s 0")) {
271 entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
272 journal_entry_null_range(vstruct_next(entry), next);
273 return FSCK_DELETED_KEY;
276 if (journal_entry_err_on((void *) bkey_next(k) >
277 (void *) vstruct_next(entry),
279 "extends past end of journal entry")) {
280 entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
281 journal_entry_null_range(vstruct_next(entry), next);
282 return FSCK_DELETED_KEY;
285 if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT,
287 "bad format %u", k->k.format)) {
288 le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
289 memmove(k, bkey_next(k), next - (void *) bkey_next(k));
290 journal_entry_null_range(vstruct_next(entry), next);
291 return FSCK_DELETED_KEY;
295 bch2_bkey_compat(level, btree_id, version, big_endian,
296 write, NULL, bkey_to_packed(k));
298 if (bch2_bkey_invalid(c, bkey_i_to_s_c(k),
299 __btree_node_type(level, btree_id), write, &buf)) {
300 printbuf_reset(&buf);
301 prt_printf(&buf, "invalid journal entry %s at offset %zi/%u seq %llu:",
302 bch2_jset_entry_types[entry->type],
303 (u64 *) entry - jset->_data,
304 le32_to_cpu(jset->u64s),
305 le64_to_cpu(jset->seq));
307 printbuf_indent_add(&buf, 2);
309 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k));
311 bch2_bkey_invalid(c, bkey_i_to_s_c(k),
312 __btree_node_type(level, btree_id), write, &buf);
314 mustfix_fsck_err(c, "%s", buf.buf);
316 le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
317 memmove(k, bkey_next(k), next - (void *) bkey_next(k));
318 journal_entry_null_range(vstruct_next(entry), next);
321 return FSCK_DELETED_KEY;
325 bch2_bkey_compat(level, btree_id, version, big_endian,
326 write, NULL, bkey_to_packed(k));
332 static int journal_entry_btree_keys_validate(struct bch_fs *c,
334 struct jset_entry *entry,
335 unsigned version, int big_endian, int write)
337 struct bkey_i *k = entry->start;
339 while (k != vstruct_last(entry)) {
340 int ret = journal_validate_key(c, jset, entry,
343 k, version, big_endian, write|BKEY_INVALID_FROM_JOURNAL);
344 if (ret == FSCK_DELETED_KEY)
353 static void journal_entry_btree_keys_to_text(struct printbuf *out, struct bch_fs *c,
354 struct jset_entry *entry)
359 jset_entry_for_each_key(entry, k) {
362 prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]);
364 prt_printf(out, "btree=%s l=%u ", bch2_btree_ids[entry->btree_id], entry->level);
365 bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(k));
370 static int journal_entry_btree_root_validate(struct bch_fs *c,
372 struct jset_entry *entry,
373 unsigned version, int big_endian, int write)
375 struct bkey_i *k = entry->start;
378 if (journal_entry_err_on(!entry->u64s ||
379 le16_to_cpu(entry->u64s) != k->k.u64s,
381 "invalid btree root journal entry: wrong number of keys")) {
382 void *next = vstruct_next(entry);
384 * we don't want to null out this jset_entry,
385 * just the contents, so that later we can tell
386 * we were _supposed_ to have a btree root
389 journal_entry_null_range(vstruct_next(entry), next);
393 return journal_validate_key(c, jset, entry, 1, entry->btree_id, k,
394 version, big_endian, write);
399 static void journal_entry_btree_root_to_text(struct printbuf *out, struct bch_fs *c,
400 struct jset_entry *entry)
402 journal_entry_btree_keys_to_text(out, c, entry);
405 static int journal_entry_prio_ptrs_validate(struct bch_fs *c,
407 struct jset_entry *entry,
408 unsigned version, int big_endian, int write)
410 /* obsolete, don't care: */
414 static void journal_entry_prio_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
415 struct jset_entry *entry)
419 static int journal_entry_blacklist_validate(struct bch_fs *c,
421 struct jset_entry *entry,
422 unsigned version, int big_endian, int write)
426 if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 1,
428 "invalid journal seq blacklist entry: bad size")) {
429 journal_entry_null_range(entry, vstruct_next(entry));
435 static void journal_entry_blacklist_to_text(struct printbuf *out, struct bch_fs *c,
436 struct jset_entry *entry)
438 struct jset_entry_blacklist *bl =
439 container_of(entry, struct jset_entry_blacklist, entry);
441 prt_printf(out, "seq=%llu", le64_to_cpu(bl->seq));
444 static int journal_entry_blacklist_v2_validate(struct bch_fs *c,
446 struct jset_entry *entry,
447 unsigned version, int big_endian, int write)
449 struct jset_entry_blacklist_v2 *bl_entry;
452 if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 2,
454 "invalid journal seq blacklist entry: bad size")) {
455 journal_entry_null_range(entry, vstruct_next(entry));
459 bl_entry = container_of(entry, struct jset_entry_blacklist_v2, entry);
461 if (journal_entry_err_on(le64_to_cpu(bl_entry->start) >
462 le64_to_cpu(bl_entry->end),
464 "invalid journal seq blacklist entry: start > end")) {
465 journal_entry_null_range(entry, vstruct_next(entry));
472 static void journal_entry_blacklist_v2_to_text(struct printbuf *out, struct bch_fs *c,
473 struct jset_entry *entry)
475 struct jset_entry_blacklist_v2 *bl =
476 container_of(entry, struct jset_entry_blacklist_v2, entry);
478 prt_printf(out, "start=%llu end=%llu",
479 le64_to_cpu(bl->start),
480 le64_to_cpu(bl->end));
483 static int journal_entry_usage_validate(struct bch_fs *c,
485 struct jset_entry *entry,
486 unsigned version, int big_endian, int write)
488 struct jset_entry_usage *u =
489 container_of(entry, struct jset_entry_usage, entry);
490 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
493 if (journal_entry_err_on(bytes < sizeof(*u),
495 "invalid journal entry usage: bad size")) {
496 journal_entry_null_range(entry, vstruct_next(entry));
504 static void journal_entry_usage_to_text(struct printbuf *out, struct bch_fs *c,
505 struct jset_entry *entry)
507 struct jset_entry_usage *u =
508 container_of(entry, struct jset_entry_usage, entry);
510 prt_printf(out, "type=%s v=%llu",
511 bch2_fs_usage_types[u->entry.btree_id],
515 static int journal_entry_data_usage_validate(struct bch_fs *c,
517 struct jset_entry *entry,
518 unsigned version, int big_endian, int write)
520 struct jset_entry_data_usage *u =
521 container_of(entry, struct jset_entry_data_usage, entry);
522 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
525 if (journal_entry_err_on(bytes < sizeof(*u) ||
526 bytes < sizeof(*u) + u->r.nr_devs,
528 "invalid journal entry usage: bad size")) {
529 journal_entry_null_range(entry, vstruct_next(entry));
537 static void journal_entry_data_usage_to_text(struct printbuf *out, struct bch_fs *c,
538 struct jset_entry *entry)
540 struct jset_entry_data_usage *u =
541 container_of(entry, struct jset_entry_data_usage, entry);
543 bch2_replicas_entry_to_text(out, &u->r);
544 prt_printf(out, "=%llu", le64_to_cpu(u->v));
547 static int journal_entry_clock_validate(struct bch_fs *c,
549 struct jset_entry *entry,
550 unsigned version, int big_endian, int write)
552 struct jset_entry_clock *clock =
553 container_of(entry, struct jset_entry_clock, entry);
554 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
557 if (journal_entry_err_on(bytes != sizeof(*clock),
558 c, jset, entry, "bad size")) {
559 journal_entry_null_range(entry, vstruct_next(entry));
563 if (journal_entry_err_on(clock->rw > 1,
564 c, jset, entry, "bad rw")) {
565 journal_entry_null_range(entry, vstruct_next(entry));
573 static void journal_entry_clock_to_text(struct printbuf *out, struct bch_fs *c,
574 struct jset_entry *entry)
576 struct jset_entry_clock *clock =
577 container_of(entry, struct jset_entry_clock, entry);
579 prt_printf(out, "%s=%llu", clock->rw ? "write" : "read", le64_to_cpu(clock->time));
582 static int journal_entry_dev_usage_validate(struct bch_fs *c,
584 struct jset_entry *entry,
585 unsigned version, int big_endian, int write)
587 struct jset_entry_dev_usage *u =
588 container_of(entry, struct jset_entry_dev_usage, entry);
589 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
590 unsigned expected = sizeof(*u);
594 if (journal_entry_err_on(bytes < expected,
595 c, jset, entry, "bad size (%u < %u)",
597 journal_entry_null_range(entry, vstruct_next(entry));
601 dev = le32_to_cpu(u->dev);
603 if (journal_entry_err_on(!bch2_dev_exists2(c, dev),
604 c, jset, entry, "bad dev")) {
605 journal_entry_null_range(entry, vstruct_next(entry));
609 if (journal_entry_err_on(u->pad,
610 c, jset, entry, "bad pad")) {
611 journal_entry_null_range(entry, vstruct_next(entry));
619 static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs *c,
620 struct jset_entry *entry)
622 struct jset_entry_dev_usage *u =
623 container_of(entry, struct jset_entry_dev_usage, entry);
624 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
626 prt_printf(out, "dev=%u", le32_to_cpu(u->dev));
628 for (i = 0; i < nr_types; i++) {
630 prt_printf(out, " %s", bch2_data_types[i]);
632 prt_printf(out, " (unknown data type %u)", i);
633 prt_printf(out, ": buckets=%llu sectors=%llu fragmented=%llu",
634 le64_to_cpu(u->d[i].buckets),
635 le64_to_cpu(u->d[i].sectors),
636 le64_to_cpu(u->d[i].fragmented));
639 prt_printf(out, " buckets_ec: %llu", le64_to_cpu(u->buckets_ec));
642 static int journal_entry_log_validate(struct bch_fs *c,
644 struct jset_entry *entry,
645 unsigned version, int big_endian, int write)
650 static void journal_entry_log_to_text(struct printbuf *out, struct bch_fs *c,
651 struct jset_entry *entry)
653 struct jset_entry_log *l = container_of(entry, struct jset_entry_log, entry);
654 unsigned bytes = vstruct_bytes(entry) - offsetof(struct jset_entry_log, d);
656 prt_printf(out, "%.*s", bytes, l->d);
659 static int journal_entry_overwrite_validate(struct bch_fs *c,
661 struct jset_entry *entry,
662 unsigned version, int big_endian, int write)
664 return journal_entry_btree_keys_validate(c, jset, entry,
665 version, big_endian, READ);
668 static void journal_entry_overwrite_to_text(struct printbuf *out, struct bch_fs *c,
669 struct jset_entry *entry)
671 journal_entry_btree_keys_to_text(out, c, entry);
674 struct jset_entry_ops {
675 int (*validate)(struct bch_fs *, struct jset *,
676 struct jset_entry *, unsigned, int, int);
677 void (*to_text)(struct printbuf *, struct bch_fs *, struct jset_entry *);
680 static const struct jset_entry_ops bch2_jset_entry_ops[] = {
682 [BCH_JSET_ENTRY_##f] = (struct jset_entry_ops) { \
683 .validate = journal_entry_##f##_validate, \
684 .to_text = journal_entry_##f##_to_text, \
686 BCH_JSET_ENTRY_TYPES()
690 int bch2_journal_entry_validate(struct bch_fs *c,
692 struct jset_entry *entry,
693 unsigned version, int big_endian, int write)
695 return entry->type < BCH_JSET_ENTRY_NR
696 ? bch2_jset_entry_ops[entry->type].validate(c, jset, entry,
697 version, big_endian, write)
701 void bch2_journal_entry_to_text(struct printbuf *out, struct bch_fs *c,
702 struct jset_entry *entry)
704 if (entry->type < BCH_JSET_ENTRY_NR) {
705 prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]);
706 bch2_jset_entry_ops[entry->type].to_text(out, c, entry);
708 prt_printf(out, "(unknown type %u)", entry->type);
712 static int jset_validate_entries(struct bch_fs *c, struct jset *jset,
715 struct jset_entry *entry;
718 vstruct_for_each(jset, entry) {
719 if (journal_entry_err_on(vstruct_next(entry) >
720 vstruct_last(jset), c, jset, entry,
721 "journal entry extends past end of jset")) {
722 jset->u64s = cpu_to_le32((u64 *) entry - jset->_data);
726 ret = bch2_journal_entry_validate(c, jset, entry,
727 le32_to_cpu(jset->version),
728 JSET_BIG_ENDIAN(jset), write);
736 static int jset_validate(struct bch_fs *c,
738 struct jset *jset, u64 sector,
744 if (le64_to_cpu(jset->magic) != jset_magic(c))
745 return JOURNAL_ENTRY_NONE;
747 version = le32_to_cpu(jset->version);
748 if (journal_entry_err_on((version != BCH_JSET_VERSION_OLD &&
749 version < bcachefs_metadata_version_min) ||
750 version >= bcachefs_metadata_version_max,
752 "%s sector %llu seq %llu: unknown journal entry version %u",
753 ca ? ca->name : c->name,
754 sector, le64_to_cpu(jset->seq),
756 /* don't try to continue: */
760 if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)),
762 "%s sector %llu seq %llu: journal entry with unknown csum type %llu",
763 ca ? ca->name : c->name,
764 sector, le64_to_cpu(jset->seq),
765 JSET_CSUM_TYPE(jset)))
766 ret = JOURNAL_ENTRY_BAD;
768 /* last_seq is ignored when JSET_NO_FLUSH is true */
769 if (journal_entry_err_on(!JSET_NO_FLUSH(jset) &&
770 le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq),
772 "invalid journal entry: last_seq > seq (%llu > %llu)",
773 le64_to_cpu(jset->last_seq),
774 le64_to_cpu(jset->seq))) {
775 jset->last_seq = jset->seq;
776 return JOURNAL_ENTRY_BAD;
779 ret = jset_validate_entries(c, jset, write);
784 static int jset_validate_early(struct bch_fs *c,
786 struct jset *jset, u64 sector,
787 unsigned bucket_sectors_left,
788 unsigned sectors_read)
790 size_t bytes = vstruct_bytes(jset);
795 if (le64_to_cpu(jset->magic) != jset_magic(c))
796 return JOURNAL_ENTRY_NONE;
798 version = le32_to_cpu(jset->version);
799 if (journal_entry_err_on((version != BCH_JSET_VERSION_OLD &&
800 version < bcachefs_metadata_version_min) ||
801 version >= bcachefs_metadata_version_max,
803 "%s sector %llu seq %llu: unknown journal entry version %u",
804 ca ? ca->name : c->name,
805 sector, le64_to_cpu(jset->seq),
807 /* don't try to continue: */
811 if (bytes > (sectors_read << 9) &&
812 sectors_read < bucket_sectors_left)
813 return JOURNAL_ENTRY_REREAD;
815 if (journal_entry_err_on(bytes > bucket_sectors_left << 9,
817 "%s sector %llu seq %llu: journal entry too big (%zu bytes)",
818 ca ? ca->name : c->name,
819 sector, le64_to_cpu(jset->seq), bytes))
820 le32_add_cpu(&jset->u64s,
821 -((bytes - (bucket_sectors_left << 9)) / 8));
826 struct journal_read_buf {
831 static int journal_read_buf_realloc(struct journal_read_buf *b,
836 /* the bios are sized for this many pages, max: */
837 if (new_size > JOURNAL_ENTRY_SIZE_MAX)
838 return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
840 new_size = roundup_pow_of_two(new_size);
841 n = kvpmalloc(new_size, GFP_KERNEL);
843 return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
845 kvpfree(b->data, b->size);
851 static int journal_read_bucket(struct bch_dev *ca,
852 struct journal_read_buf *buf,
853 struct journal_list *jlist,
856 struct bch_fs *c = ca->fs;
857 struct journal_device *ja = &ca->journal;
858 struct jset *j = NULL;
859 unsigned sectors, sectors_read = 0;
860 u64 offset = bucket_to_sector(ca, ja->buckets[bucket]),
861 end = offset + ca->mi.bucket_size;
862 bool saw_bad = false, csum_good;
865 pr_debug("reading %u", bucket);
867 while (offset < end) {
872 sectors_read = min_t(unsigned,
873 end - offset, buf->size >> 9);
874 nr_bvecs = buf_pages(buf->data, sectors_read << 9);
876 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
877 bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ);
879 bio->bi_iter.bi_sector = offset;
880 bch2_bio_map(bio, buf->data, sectors_read << 9);
882 ret = submit_bio_wait(bio);
885 if (bch2_dev_io_err_on(ret, ca,
886 "journal read error: sector %llu",
888 bch2_meta_read_fault("journal")) {
890 * We don't error out of the recovery process
891 * here, since the relevant journal entry may be
892 * found on a different device, and missing or
893 * no journal entries will be handled later
901 ret = jset_validate_early(c, ca, j, offset,
902 end - offset, sectors_read);
905 sectors = vstruct_sectors(j, c->block_bits);
907 case JOURNAL_ENTRY_REREAD:
908 if (vstruct_bytes(j) > buf->size) {
909 ret = journal_read_buf_realloc(buf,
915 case JOURNAL_ENTRY_NONE:
919 * On checksum error we don't really trust the size
920 * field of the journal entry we read, so try reading
921 * again at next block boundary:
923 sectors = block_sectors(c);
930 * This happens sometimes if we don't have discards on -
931 * when we've partially overwritten a bucket with new
932 * journal entries. We don't need the rest of the
935 if (le64_to_cpu(j->seq) < ja->bucket_seq[bucket])
938 ja->bucket_seq[bucket] = le64_to_cpu(j->seq);
940 csum_good = jset_csum_good(c, j);
944 ret = bch2_encrypt(c, JSET_CSUM_TYPE(j), journal_nonce(j),
946 vstruct_end(j) - (void *) j->encrypted_start);
947 bch2_fs_fatal_err_on(ret, c,
948 "error decrypting journal entry: %i", ret);
950 mutex_lock(&jlist->lock);
951 ret = journal_entry_add(c, ca, (struct journal_ptr) {
952 .csum_good = csum_good,
955 .bucket_offset = offset -
956 bucket_to_sector(ca, ja->buckets[bucket]),
959 mutex_unlock(&jlist->lock);
962 case JOURNAL_ENTRY_ADD_OK:
964 case JOURNAL_ENTRY_ADD_OUT_OF_RANGE:
972 sectors_read -= sectors;
973 j = ((void *) j) + (sectors << 9);
979 static void bch2_journal_read_device(struct closure *cl)
981 struct journal_device *ja =
982 container_of(cl, struct journal_device, read);
983 struct bch_dev *ca = container_of(ja, struct bch_dev, journal);
984 struct bch_fs *c = ca->fs;
985 struct journal_list *jlist =
986 container_of(cl->parent, struct journal_list, cl);
987 struct journal_replay *r, **_r;
988 struct genradix_iter iter;
989 struct journal_read_buf buf = { NULL, 0 };
996 ret = journal_read_buf_realloc(&buf, PAGE_SIZE);
1000 pr_debug("%u journal buckets", ja->nr);
1002 for (i = 0; i < ja->nr; i++) {
1003 ret = journal_read_bucket(ca, &buf, jlist, i);
1008 ja->sectors_free = ca->mi.bucket_size;
1010 mutex_lock(&jlist->lock);
1011 genradix_for_each_reverse(&c->journal_entries, iter, _r) {
1017 for (i = 0; i < r->nr_ptrs; i++) {
1018 if (r->ptrs[i].dev == ca->dev_idx) {
1019 unsigned wrote = bucket_remainder(ca, r->ptrs[i].sector) +
1020 vstruct_sectors(&r->j, c->block_bits);
1022 ja->cur_idx = r->ptrs[i].bucket;
1023 ja->sectors_free = ca->mi.bucket_size - wrote;
1029 mutex_unlock(&jlist->lock);
1031 if (ja->bucket_seq[ja->cur_idx] &&
1032 ja->sectors_free == ca->mi.bucket_size) {
1033 bch_err(c, "ja->sectors_free == ca->mi.bucket_size");
1034 bch_err(c, "cur_idx %u/%u", ja->cur_idx, ja->nr);
1035 for (i = 0; i < 3; i++) {
1036 unsigned idx = (ja->cur_idx + ja->nr - 1 + i) % ja->nr;
1037 bch_err(c, "bucket_seq[%u] = %llu", idx, ja->bucket_seq[idx]);
1039 ja->sectors_free = 0;
1043 * Set dirty_idx to indicate the entire journal is full and needs to be
1044 * reclaimed - journal reclaim will immediately reclaim whatever isn't
1045 * pinned when it first runs:
1047 ja->discard_idx = ja->dirty_idx_ondisk =
1048 ja->dirty_idx = (ja->cur_idx + 1) % ja->nr;
1050 bch_verbose(c, "journal read done on device %s, ret %i", ca->name, ret);
1051 kvpfree(buf.data, buf.size);
1052 percpu_ref_put(&ca->io_ref);
1056 mutex_lock(&jlist->lock);
1058 mutex_unlock(&jlist->lock);
1062 void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
1063 struct journal_replay *j)
1067 for (i = 0; i < j->nr_ptrs; i++) {
1068 struct bch_dev *ca = bch_dev_bkey_exists(c, j->ptrs[i].dev);
1071 div64_u64_rem(j->ptrs[i].sector, ca->mi.bucket_size, &offset);
1074 prt_printf(out, " ");
1075 prt_printf(out, "%u:%u:%u (sector %llu)",
1078 j->ptrs[i].bucket_offset,
1083 int bch2_journal_read(struct bch_fs *c,
1088 struct journal_list jlist;
1089 struct journal_replay *i, **_i, *prev = NULL;
1090 struct genradix_iter radix_iter;
1093 struct printbuf buf = PRINTBUF;
1094 bool degraded = false, last_write_torn = false;
1098 closure_init_stack(&jlist.cl);
1099 mutex_init(&jlist.lock);
1103 for_each_member_device(ca, c, iter) {
1104 if (!c->opts.fsck &&
1105 !(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_journal)))
1108 if ((ca->mi.state == BCH_MEMBER_STATE_rw ||
1109 ca->mi.state == BCH_MEMBER_STATE_ro) &&
1110 percpu_ref_tryget(&ca->io_ref))
1111 closure_call(&ca->journal.read,
1112 bch2_journal_read_device,
1119 closure_sync(&jlist.cl);
1129 * Find most recent flush entry, and ignore newer non flush entries -
1130 * those entries will be blacklisted:
1132 genradix_for_each_reverse(&c->journal_entries, radix_iter, _i) {
1137 if (!i || i->ignore)
1141 *blacklist_seq = *start_seq = le64_to_cpu(i->j.seq) + 1;
1143 if (JSET_NO_FLUSH(&i->j)) {
1148 if (!last_write_torn && !i->csum_good) {
1149 last_write_torn = true;
1154 if (journal_entry_err_on(le64_to_cpu(i->j.last_seq) > le64_to_cpu(i->j.seq),
1156 "invalid journal entry: last_seq > seq (%llu > %llu)",
1157 le64_to_cpu(i->j.last_seq),
1158 le64_to_cpu(i->j.seq)))
1159 i->j.last_seq = i->j.seq;
1161 *last_seq = le64_to_cpu(i->j.last_seq);
1162 *blacklist_seq = le64_to_cpu(i->j.seq) + 1;
1167 bch_info(c, "journal read done, but no entries found");
1172 fsck_err(c, "journal read done, but no entries found after dropping non-flushes");
1176 bch_info(c, "journal read done, replaying entries %llu-%llu",
1177 *last_seq, *blacklist_seq - 1);
1179 if (*start_seq != *blacklist_seq)
1180 bch_info(c, "dropped unflushed entries %llu-%llu",
1181 *blacklist_seq, *start_seq - 1);
1183 /* Drop blacklisted entries and entries older than last_seq: */
1184 genradix_for_each(&c->journal_entries, radix_iter, _i) {
1187 if (!i || i->ignore)
1190 seq = le64_to_cpu(i->j.seq);
1191 if (seq < *last_seq) {
1192 journal_replay_free(c, i);
1196 if (bch2_journal_seq_is_blacklisted(c, seq, true)) {
1197 fsck_err_on(!JSET_NO_FLUSH(&i->j), c,
1198 "found blacklisted journal entry %llu", seq);
1203 /* Check for missing entries: */
1205 genradix_for_each(&c->journal_entries, radix_iter, _i) {
1208 if (!i || i->ignore)
1211 BUG_ON(seq > le64_to_cpu(i->j.seq));
1213 while (seq < le64_to_cpu(i->j.seq)) {
1214 u64 missing_start, missing_end;
1215 struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
1217 while (seq < le64_to_cpu(i->j.seq) &&
1218 bch2_journal_seq_is_blacklisted(c, seq, false))
1221 if (seq == le64_to_cpu(i->j.seq))
1224 missing_start = seq;
1226 while (seq < le64_to_cpu(i->j.seq) &&
1227 !bch2_journal_seq_is_blacklisted(c, seq, false))
1231 bch2_journal_ptrs_to_text(&buf1, c, prev);
1232 prt_printf(&buf1, " size %zu", vstruct_sectors(&prev->j, c->block_bits));
1234 prt_printf(&buf1, "(none)");
1235 bch2_journal_ptrs_to_text(&buf2, c, i);
1237 missing_end = seq - 1;
1238 fsck_err(c, "journal entries %llu-%llu missing! (replaying %llu-%llu)\n"
1241 missing_start, missing_end,
1242 *last_seq, *blacklist_seq - 1,
1243 buf1.buf, buf2.buf);
1245 printbuf_exit(&buf1);
1246 printbuf_exit(&buf2);
1253 genradix_for_each(&c->journal_entries, radix_iter, _i) {
1254 struct bch_replicas_padded replicas = {
1255 .e.data_type = BCH_DATA_journal,
1261 if (!i || i->ignore)
1264 for (ptr = 0; ptr < i->nr_ptrs; ptr++) {
1265 struct bch_dev *ca = bch_dev_bkey_exists(c, i->ptrs[ptr].dev);
1267 if (!i->ptrs[ptr].csum_good)
1268 bch_err_dev_offset(ca, i->ptrs[ptr].sector,
1269 "invalid journal checksum, seq %llu%s",
1270 le64_to_cpu(i->j.seq),
1271 i->csum_good ? " (had good copy on another device)" : "");
1274 ret = jset_validate(c,
1275 bch_dev_bkey_exists(c, i->ptrs[0].dev),
1282 for (ptr = 0; ptr < i->nr_ptrs; ptr++)
1283 replicas.e.devs[replicas.e.nr_devs++] = i->ptrs[ptr].dev;
1285 bch2_replicas_entry_sort(&replicas.e);
1288 * If we're mounting in degraded mode - if we didn't read all
1289 * the devices - this is wrong:
1292 printbuf_reset(&buf);
1293 bch2_replicas_entry_to_text(&buf, &replicas.e);
1296 fsck_err_on(!bch2_replicas_marked(c, &replicas.e), c,
1297 "superblock not marked as containing replicas %s",
1299 ret = bch2_mark_replicas(c, &replicas.e);
1306 printbuf_exit(&buf);
1310 /* journal write: */
1312 static void __journal_write_alloc(struct journal *j,
1313 struct journal_buf *w,
1314 struct dev_alloc_list *devs_sorted,
1317 unsigned replicas_want)
1319 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1320 struct journal_device *ja;
1324 if (*replicas >= replicas_want)
1327 for (i = 0; i < devs_sorted->nr; i++) {
1328 ca = rcu_dereference(c->devs[devs_sorted->devs[i]]);
1335 * Check that we can use this device, and aren't already using
1338 if (!ca->mi.durability ||
1339 ca->mi.state != BCH_MEMBER_STATE_rw ||
1341 bch2_bkey_has_device_c(bkey_i_to_s_c(&w->key), ca->dev_idx) ||
1342 sectors > ja->sectors_free)
1345 bch2_dev_stripe_increment(ca, &j->wp.stripe);
1347 bch2_bkey_append_ptr(&w->key,
1348 (struct bch_extent_ptr) {
1349 .offset = bucket_to_sector(ca,
1350 ja->buckets[ja->cur_idx]) +
1351 ca->mi.bucket_size -
1356 ja->sectors_free -= sectors;
1357 ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1359 *replicas += ca->mi.durability;
1361 if (*replicas >= replicas_want)
1367 * journal_next_bucket - move on to the next journal bucket if possible
1369 static int journal_write_alloc(struct journal *j, struct journal_buf *w,
1372 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1373 struct bch_devs_mask devs;
1374 struct journal_device *ja;
1376 struct dev_alloc_list devs_sorted;
1377 unsigned target = c->opts.metadata_target ?:
1378 c->opts.foreground_target;
1379 unsigned i, replicas = 0, replicas_want =
1380 READ_ONCE(c->opts.metadata_replicas);
1384 devs = target_rw_devs(c, BCH_DATA_journal, target);
1386 devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe, &devs);
1388 __journal_write_alloc(j, w, &devs_sorted,
1389 sectors, &replicas, replicas_want);
1391 if (replicas >= replicas_want)
1394 for (i = 0; i < devs_sorted.nr; i++) {
1395 ca = rcu_dereference(c->devs[devs_sorted.devs[i]]);
1401 if (sectors > ja->sectors_free &&
1402 sectors <= ca->mi.bucket_size &&
1403 bch2_journal_dev_buckets_available(j, ja,
1404 journal_space_discarded)) {
1405 ja->cur_idx = (ja->cur_idx + 1) % ja->nr;
1406 ja->sectors_free = ca->mi.bucket_size;
1409 * ja->bucket_seq[ja->cur_idx] must always have
1410 * something sensible:
1412 ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1416 __journal_write_alloc(j, w, &devs_sorted,
1417 sectors, &replicas, replicas_want);
1419 if (replicas < replicas_want && target) {
1420 /* Retry from all devices: */
1427 BUG_ON(bkey_val_u64s(&w->key.k) > BCH_REPLICAS_MAX);
1429 return replicas >= c->opts.metadata_replicas_required ? 0 : -EROFS;
1432 static void journal_buf_realloc(struct journal *j, struct journal_buf *buf)
1434 /* we aren't holding j->lock: */
1435 unsigned new_size = READ_ONCE(j->buf_size_want);
1438 if (buf->buf_size >= new_size)
1441 new_buf = kvpmalloc(new_size, GFP_NOIO|__GFP_NOWARN);
1445 memcpy(new_buf, buf->data, buf->buf_size);
1447 spin_lock(&j->lock);
1448 swap(buf->data, new_buf);
1449 swap(buf->buf_size, new_size);
1450 spin_unlock(&j->lock);
1452 kvpfree(new_buf, new_size);
1455 static inline struct journal_buf *journal_last_unwritten_buf(struct journal *j)
1457 return j->buf + (journal_last_unwritten_seq(j) & JOURNAL_BUF_MASK);
1460 static void journal_write_done(struct closure *cl)
1462 struct journal *j = container_of(cl, struct journal, io);
1463 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1464 struct journal_buf *w = journal_last_unwritten_buf(j);
1465 union journal_res_state old, new;
1469 bch2_time_stats_update(!JSET_NO_FLUSH(w->data)
1470 ? j->flush_write_time
1471 : j->noflush_write_time, j->write_start_time);
1473 if (!w->devs_written.nr) {
1474 bch_err(c, "unable to write journal to sufficient devices");
1478 bch2_fatal_error(c);
1480 spin_lock(&j->lock);
1481 seq = le64_to_cpu(w->data->seq);
1483 if (seq >= j->pin.front)
1484 journal_seq_pin(j, seq)->devs = w->devs_written;
1487 if (!JSET_NO_FLUSH(w->data)) {
1488 j->flushed_seq_ondisk = seq;
1489 j->last_seq_ondisk = w->last_seq;
1491 bch2_do_discards(c);
1492 closure_wake_up(&c->freelist_wait);
1494 bch2_reset_alloc_cursors(c);
1496 } else if (!j->err_seq || seq < j->err_seq)
1499 j->seq_ondisk = seq;
1502 * Updating last_seq_ondisk may let bch2_journal_reclaim_work() discard
1505 * Must come before signaling write completion, for
1506 * bch2_fs_journal_stop():
1509 journal_reclaim_kick(&c->journal);
1511 /* also must come before signalling write completion: */
1512 closure_debug_destroy(cl);
1514 v = atomic64_read(&j->reservations.counter);
1517 BUG_ON(journal_state_count(new, new.unwritten_idx));
1519 new.unwritten_idx++;
1520 } while ((v = atomic64_cmpxchg(&j->reservations.counter,
1521 old.v, new.v)) != old.v);
1523 bch2_journal_space_available(j);
1525 closure_wake_up(&w->wait);
1528 if (!journal_state_count(new, new.unwritten_idx) &&
1529 journal_last_unwritten_seq(j) <= journal_cur_seq(j)) {
1530 closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
1531 } else if (journal_last_unwritten_seq(j) == journal_cur_seq(j) &&
1532 new.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL) {
1533 struct journal_buf *buf = journal_cur_buf(j);
1534 long delta = buf->expires - jiffies;
1537 * We don't close a journal entry to write it while there's
1538 * previous entries still in flight - the current journal entry
1539 * might want to be written now:
1542 mod_delayed_work(c->io_complete_wq, &j->write_work, max(0L, delta));
1545 spin_unlock(&j->lock);
1548 static void journal_write_endio(struct bio *bio)
1550 struct bch_dev *ca = bio->bi_private;
1551 struct journal *j = &ca->fs->journal;
1552 struct journal_buf *w = journal_last_unwritten_buf(j);
1553 unsigned long flags;
1555 if (bch2_dev_io_err_on(bio->bi_status, ca, "error writing journal entry %llu: %s",
1556 le64_to_cpu(w->data->seq),
1557 bch2_blk_status_to_str(bio->bi_status)) ||
1558 bch2_meta_write_fault("journal")) {
1559 spin_lock_irqsave(&j->err_lock, flags);
1560 bch2_dev_list_drop_dev(&w->devs_written, ca->dev_idx);
1561 spin_unlock_irqrestore(&j->err_lock, flags);
1564 closure_put(&j->io);
1565 percpu_ref_put(&ca->io_ref);
1568 static void do_journal_write(struct closure *cl)
1570 struct journal *j = container_of(cl, struct journal, io);
1571 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1573 struct journal_buf *w = journal_last_unwritten_buf(j);
1574 struct bch_extent_ptr *ptr;
1576 unsigned sectors = vstruct_sectors(w->data, c->block_bits);
1578 extent_for_each_ptr(bkey_i_to_s_extent(&w->key), ptr) {
1579 ca = bch_dev_bkey_exists(c, ptr->dev);
1580 if (!percpu_ref_tryget(&ca->io_ref)) {
1582 bch_err(c, "missing device for journal write\n");
1586 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_journal],
1589 bio = ca->journal.bio;
1590 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
1591 bio->bi_iter.bi_sector = ptr->offset;
1592 bio->bi_end_io = journal_write_endio;
1593 bio->bi_private = ca;
1595 BUG_ON(bio->bi_iter.bi_sector == ca->prev_journal_sector);
1596 ca->prev_journal_sector = bio->bi_iter.bi_sector;
1598 if (!JSET_NO_FLUSH(w->data))
1599 bio->bi_opf |= REQ_FUA;
1600 if (!JSET_NO_FLUSH(w->data) && !w->separate_flush)
1601 bio->bi_opf |= REQ_PREFLUSH;
1603 bch2_bio_map(bio, w->data, sectors << 9);
1605 trace_and_count(c, journal_write, bio);
1606 closure_bio_submit(bio, cl);
1608 ca->journal.bucket_seq[ca->journal.cur_idx] =
1609 le64_to_cpu(w->data->seq);
1612 continue_at(cl, journal_write_done, c->io_complete_wq);
1616 static void bch2_journal_entries_postprocess(struct bch_fs *c, struct jset *jset)
1618 struct jset_entry *i, *next, *prev = NULL;
1621 * Simple compaction, dropping empty jset_entries (from journal
1622 * reservations that weren't fully used) and merging jset_entries that
1625 * If we wanted to be really fancy here, we could sort all the keys in
1626 * the jset and drop keys that were overwritten - probably not worth it:
1628 vstruct_for_each_safe(jset, i, next) {
1629 unsigned u64s = le16_to_cpu(i->u64s);
1635 if (i->type == BCH_JSET_ENTRY_btree_root)
1636 bch2_journal_entry_to_btree_root(c, i);
1638 /* Can we merge with previous entry? */
1640 i->btree_id == prev->btree_id &&
1641 i->level == prev->level &&
1642 i->type == prev->type &&
1643 i->type == BCH_JSET_ENTRY_btree_keys &&
1644 le16_to_cpu(prev->u64s) + u64s <= U16_MAX) {
1645 memmove_u64s_down(vstruct_next(prev),
1648 le16_add_cpu(&prev->u64s, u64s);
1652 /* Couldn't merge, move i into new position (after prev): */
1653 prev = prev ? vstruct_next(prev) : jset->start;
1655 memmove_u64s_down(prev, i, jset_u64s(u64s));
1658 prev = prev ? vstruct_next(prev) : jset->start;
1659 jset->u64s = cpu_to_le32((u64 *) prev - jset->_data);
1662 void bch2_journal_write(struct closure *cl)
1664 struct journal *j = container_of(cl, struct journal, io);
1665 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1667 struct journal_buf *w = journal_last_unwritten_buf(j);
1668 struct bch_replicas_padded replicas;
1669 struct jset_entry *start, *end;
1672 struct printbuf journal_debug_buf = PRINTBUF;
1673 bool validate_before_checksum = false;
1674 unsigned i, sectors, bytes, u64s, nr_rw_members = 0;
1677 BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
1679 journal_buf_realloc(j, w);
1682 j->write_start_time = local_clock();
1684 spin_lock(&j->lock);
1687 * If the journal is in an error state - we did an emergency shutdown -
1688 * we prefer to continue doing journal writes. We just mark them as
1689 * noflush so they'll never be used, but they'll still be visible by the
1690 * list_journal tool - this helps in debugging.
1692 * There's a caveat: the first journal write after marking the
1693 * superblock dirty must always be a flush write, because on startup
1694 * from a clean shutdown we didn't necessarily read the journal and the
1695 * new journal write might overwrite whatever was in the journal
1696 * previously - we can't leave the journal without any flush writes in
1699 * So if we're in an error state, and we're still starting up, we don't
1700 * write anything at all.
1702 if (!test_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags) &&
1703 (bch2_journal_error(j) ||
1706 (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) &&
1707 test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags)))) {
1709 SET_JSET_NO_FLUSH(jset, true);
1713 j->nr_noflush_writes++;
1714 } else if (!bch2_journal_error(j)) {
1715 j->last_flush_write = jiffies;
1716 j->nr_flush_writes++;
1717 clear_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags);
1719 spin_unlock(&j->lock);
1722 spin_unlock(&j->lock);
1725 * New btree roots are set by journalling them; when the journal entry
1726 * gets written we have to propagate them to c->btree_roots
1728 * But, every journal entry we write has to contain all the btree roots
1729 * (at least for now); so after we copy btree roots to c->btree_roots we
1730 * have to get any missing btree roots and add them to this journal
1734 bch2_journal_entries_postprocess(c, jset);
1736 start = end = vstruct_last(jset);
1738 end = bch2_btree_roots_to_journal_entries(c, jset->start, end);
1740 bch2_journal_super_entries_add_common(c, &end,
1741 le64_to_cpu(jset->seq));
1742 u64s = (u64 *) end - (u64 *) start;
1743 BUG_ON(u64s > j->entry_u64s_reserved);
1745 le32_add_cpu(&jset->u64s, u64s);
1746 BUG_ON(vstruct_sectors(jset, c->block_bits) > w->sectors);
1748 jset->magic = cpu_to_le64(jset_magic(c));
1749 jset->version = c->sb.version < bcachefs_metadata_version_bkey_renumber
1750 ? cpu_to_le32(BCH_JSET_VERSION_OLD)
1751 : cpu_to_le32(c->sb.version);
1753 SET_JSET_BIG_ENDIAN(jset, CPU_BIG_ENDIAN);
1754 SET_JSET_CSUM_TYPE(jset, bch2_meta_checksum_type(c));
1756 if (!JSET_NO_FLUSH(jset) && journal_entry_empty(jset))
1757 j->last_empty_seq = le64_to_cpu(jset->seq);
1759 if (bch2_csum_type_is_encryption(JSET_CSUM_TYPE(jset)))
1760 validate_before_checksum = true;
1762 if (le32_to_cpu(jset->version) < bcachefs_metadata_version_current)
1763 validate_before_checksum = true;
1765 if (validate_before_checksum &&
1766 jset_validate(c, NULL, jset, 0, WRITE))
1769 ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset),
1770 jset->encrypted_start,
1771 vstruct_end(jset) - (void *) jset->encrypted_start);
1772 if (bch2_fs_fatal_err_on(ret, c,
1773 "error decrypting journal entry: %i", ret))
1776 jset->csum = csum_vstruct(c, JSET_CSUM_TYPE(jset),
1777 journal_nonce(jset), jset);
1779 if (!validate_before_checksum &&
1780 jset_validate(c, NULL, jset, 0, WRITE))
1783 sectors = vstruct_sectors(jset, c->block_bits);
1784 BUG_ON(sectors > w->sectors);
1786 bytes = vstruct_bytes(jset);
1787 memset((void *) jset + bytes, 0, (sectors << 9) - bytes);
1790 spin_lock(&j->lock);
1791 ret = journal_write_alloc(j, w, sectors);
1793 if (ret && j->can_discard) {
1794 spin_unlock(&j->lock);
1795 bch2_journal_do_discards(j);
1800 __bch2_journal_debug_to_text(&journal_debug_buf, j);
1803 * write is allocated, no longer need to account for it in
1804 * bch2_journal_space_available():
1809 * journal entry has been compacted and allocated, recalculate space
1812 bch2_journal_space_available(j);
1813 spin_unlock(&j->lock);
1816 bch_err(c, "Unable to allocate journal write:\n%s",
1817 journal_debug_buf.buf);
1818 printbuf_exit(&journal_debug_buf);
1822 w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key));
1824 if (c->opts.nochanges)
1827 for_each_rw_member(ca, c, i)
1830 if (nr_rw_members > 1)
1831 w->separate_flush = true;
1834 * Mark journal replicas before we submit the write to guarantee
1835 * recovery will find the journal entries after a crash.
1837 bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
1839 ret = bch2_mark_replicas(c, &replicas.e);
1843 if (!JSET_NO_FLUSH(jset) && w->separate_flush) {
1844 for_each_rw_member(ca, c, i) {
1845 percpu_ref_get(&ca->io_ref);
1847 bio = ca->journal.bio;
1848 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_FLUSH);
1849 bio->bi_end_io = journal_write_endio;
1850 bio->bi_private = ca;
1851 closure_bio_submit(bio, cl);
1855 continue_at(cl, do_journal_write, c->io_complete_wq);
1858 continue_at(cl, journal_write_done, c->io_complete_wq);
1861 bch2_fatal_error(c);
1862 continue_at(cl, journal_write_done, c->io_complete_wq);