1 // SPDX-License-Identifier: GPL-2.0
2 #include "bcachefs.h"
3 #include "alloc_background.h"
4 #include "alloc_foreground.h"
5 #include "btree_io.h"
6 #include "btree_update_interior.h"
7 #include "btree_write_buffer.h"
8 #include "buckets.h"
9 #include "checksum.h"
10 #include "disk_groups.h"
11 #include "error.h"
12 #include "journal.h"
13 #include "journal_io.h"
14 #include "journal_reclaim.h"
15 #include "journal_seq_blacklist.h"
16 #include "replicas.h"
17 #include "sb-clean.h"
18 #include "trace.h"
19
bch2_journal_pos_from_member_info_set(struct bch_fs * c)20 void bch2_journal_pos_from_member_info_set(struct bch_fs *c)
21 {
22 lockdep_assert_held(&c->sb_lock);
23
24 for_each_member_device(c, ca) {
25 struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
26
27 m->last_journal_bucket = cpu_to_le32(ca->journal.cur_idx);
28 m->last_journal_bucket_offset = cpu_to_le32(ca->mi.bucket_size - ca->journal.sectors_free);
29 }
30 }
31
bch2_journal_pos_from_member_info_resume(struct bch_fs * c)32 void bch2_journal_pos_from_member_info_resume(struct bch_fs *c)
33 {
34 mutex_lock(&c->sb_lock);
35 for_each_member_device(c, ca) {
36 struct bch_member m = bch2_sb_member_get(c->disk_sb.sb, ca->dev_idx);
37
38 unsigned idx = le32_to_cpu(m.last_journal_bucket);
39 if (idx < ca->journal.nr)
40 ca->journal.cur_idx = idx;
41 unsigned offset = le32_to_cpu(m.last_journal_bucket_offset);
42 if (offset <= ca->mi.bucket_size)
43 ca->journal.sectors_free = ca->mi.bucket_size - offset;
44 }
45 mutex_unlock(&c->sb_lock);
46 }
47
bch2_journal_ptrs_to_text(struct printbuf * out,struct bch_fs * c,struct journal_replay * j)48 void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
49 struct journal_replay *j)
50 {
51 darray_for_each(j->ptrs, i) {
52 if (i != j->ptrs.data)
53 prt_printf(out, " ");
54 prt_printf(out, "%u:%u:%u (sector %llu)",
55 i->dev, i->bucket, i->bucket_offset, i->sector);
56 }
57 }
58
bch2_journal_replay_to_text(struct printbuf * out,struct bch_fs * c,struct journal_replay * j)59 static void bch2_journal_replay_to_text(struct printbuf *out, struct bch_fs *c,
60 struct journal_replay *j)
61 {
62 prt_printf(out, "seq %llu ", le64_to_cpu(j->j.seq));
63
64 bch2_journal_ptrs_to_text(out, c, j);
65
66 for_each_jset_entry_type(entry, &j->j, BCH_JSET_ENTRY_datetime) {
67 struct jset_entry_datetime *datetime =
68 container_of(entry, struct jset_entry_datetime, entry);
69 bch2_prt_datetime(out, le64_to_cpu(datetime->seconds));
70 break;
71 }
72 }
73
journal_nonce(const struct jset * jset)74 static struct nonce journal_nonce(const struct jset *jset)
75 {
76 return (struct nonce) {{
77 [0] = 0,
78 [1] = ((__le32 *) &jset->seq)[0],
79 [2] = ((__le32 *) &jset->seq)[1],
80 [3] = BCH_NONCE_JOURNAL,
81 }};
82 }
83
jset_csum_good(struct bch_fs * c,struct jset * j,struct bch_csum * csum)84 static bool jset_csum_good(struct bch_fs *c, struct jset *j, struct bch_csum *csum)
85 {
86 if (!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(j))) {
87 *csum = (struct bch_csum) {};
88 return false;
89 }
90
91 *csum = csum_vstruct(c, JSET_CSUM_TYPE(j), journal_nonce(j), j);
92 return !bch2_crc_cmp(j->csum, *csum);
93 }
94
journal_entry_radix_idx(struct bch_fs * c,u64 seq)95 static inline u32 journal_entry_radix_idx(struct bch_fs *c, u64 seq)
96 {
97 return (seq - c->journal_entries_base_seq) & (~0U >> 1);
98 }
99
__journal_replay_free(struct bch_fs * c,struct journal_replay * i)100 static void __journal_replay_free(struct bch_fs *c,
101 struct journal_replay *i)
102 {
103 struct journal_replay **p =
104 genradix_ptr(&c->journal_entries,
105 journal_entry_radix_idx(c, le64_to_cpu(i->j.seq)));
106
107 BUG_ON(*p != i);
108 *p = NULL;
109 kvfree(i);
110 }
111
journal_replay_free(struct bch_fs * c,struct journal_replay * i,bool blacklisted)112 static void journal_replay_free(struct bch_fs *c, struct journal_replay *i, bool blacklisted)
113 {
114 if (blacklisted)
115 i->ignore_blacklisted = true;
116 else
117 i->ignore_not_dirty = true;
118
119 if (!c->opts.read_entire_journal)
120 __journal_replay_free(c, i);
121 }
122
123 struct journal_list {
124 struct closure cl;
125 u64 last_seq;
126 struct mutex lock;
127 int ret;
128 };
129
130 #define JOURNAL_ENTRY_ADD_OK 0
131 #define JOURNAL_ENTRY_ADD_OUT_OF_RANGE 5
132
133 /*
134 * Given a journal entry we just read, add it to the list of journal entries to
135 * be replayed:
136 */
journal_entry_add(struct bch_fs * c,struct bch_dev * ca,struct journal_ptr entry_ptr,struct journal_list * jlist,struct jset * j)137 static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca,
138 struct journal_ptr entry_ptr,
139 struct journal_list *jlist, struct jset *j)
140 {
141 struct genradix_iter iter;
142 struct journal_replay **_i, *i, *dup;
143 size_t bytes = vstruct_bytes(j);
144 u64 last_seq = !JSET_NO_FLUSH(j) ? le64_to_cpu(j->last_seq) : 0;
145 struct printbuf buf = PRINTBUF;
146 int ret = JOURNAL_ENTRY_ADD_OK;
147
148 if (!c->journal.oldest_seq_found_ondisk ||
149 le64_to_cpu(j->seq) < c->journal.oldest_seq_found_ondisk)
150 c->journal.oldest_seq_found_ondisk = le64_to_cpu(j->seq);
151
152 /* Is this entry older than the range we need? */
153 if (!c->opts.read_entire_journal &&
154 le64_to_cpu(j->seq) < jlist->last_seq)
155 return JOURNAL_ENTRY_ADD_OUT_OF_RANGE;
156
157 /*
158 * genradixes are indexed by a ulong, not a u64, so we can't index them
159 * by sequence number directly: Assume instead that they will all fall
160 * within the range of +-2billion of the filrst one we find.
161 */
162 if (!c->journal_entries_base_seq)
163 c->journal_entries_base_seq = max_t(s64, 1, le64_to_cpu(j->seq) - S32_MAX);
164
165 /* Drop entries we don't need anymore */
166 if (last_seq > jlist->last_seq && !c->opts.read_entire_journal) {
167 genradix_for_each_from(&c->journal_entries, iter, _i,
168 journal_entry_radix_idx(c, jlist->last_seq)) {
169 i = *_i;
170
171 if (journal_replay_ignore(i))
172 continue;
173
174 if (le64_to_cpu(i->j.seq) >= last_seq)
175 break;
176
177 journal_replay_free(c, i, false);
178 }
179 }
180
181 jlist->last_seq = max(jlist->last_seq, last_seq);
182
183 _i = genradix_ptr_alloc(&c->journal_entries,
184 journal_entry_radix_idx(c, le64_to_cpu(j->seq)),
185 GFP_KERNEL);
186 if (!_i)
187 return -BCH_ERR_ENOMEM_journal_entry_add;
188
189 /*
190 * Duplicate journal entries? If so we want the one that didn't have a
191 * checksum error:
192 */
193 dup = *_i;
194 if (dup) {
195 bool identical = bytes == vstruct_bytes(&dup->j) &&
196 !memcmp(j, &dup->j, bytes);
197 bool not_identical = !identical &&
198 entry_ptr.csum_good &&
199 dup->csum_good;
200
201 bool same_device = false;
202 darray_for_each(dup->ptrs, ptr)
203 if (ptr->dev == ca->dev_idx)
204 same_device = true;
205
206 ret = darray_push(&dup->ptrs, entry_ptr);
207 if (ret)
208 goto out;
209
210 bch2_journal_replay_to_text(&buf, c, dup);
211
212 fsck_err_on(same_device,
213 c, journal_entry_dup_same_device,
214 "duplicate journal entry on same device\n %s",
215 buf.buf);
216
217 fsck_err_on(not_identical,
218 c, journal_entry_replicas_data_mismatch,
219 "found duplicate but non identical journal entries\n %s",
220 buf.buf);
221
222 if (entry_ptr.csum_good && !identical)
223 goto replace;
224
225 goto out;
226 }
227 replace:
228 i = kvmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL);
229 if (!i)
230 return -BCH_ERR_ENOMEM_journal_entry_add;
231
232 darray_init(&i->ptrs);
233 i->csum_good = entry_ptr.csum_good;
234 i->ignore_blacklisted = false;
235 i->ignore_not_dirty = false;
236 unsafe_memcpy(&i->j, j, bytes, "embedded variable length struct");
237
238 if (dup) {
239 /* The first ptr should represent the jset we kept: */
240 darray_for_each(dup->ptrs, ptr)
241 darray_push(&i->ptrs, *ptr);
242 __journal_replay_free(c, dup);
243 } else {
244 darray_push(&i->ptrs, entry_ptr);
245 }
246
247 *_i = i;
248 out:
249 fsck_err:
250 printbuf_exit(&buf);
251 return ret;
252 }
253
254 /* this fills in a range with empty jset_entries: */
journal_entry_null_range(void * start,void * end)255 static void journal_entry_null_range(void *start, void *end)
256 {
257 struct jset_entry *entry;
258
259 for (entry = start; entry != end; entry = vstruct_next(entry))
260 memset(entry, 0, sizeof(*entry));
261 }
262
263 #define JOURNAL_ENTRY_REREAD 5
264 #define JOURNAL_ENTRY_NONE 6
265 #define JOURNAL_ENTRY_BAD 7
266
journal_entry_err_msg(struct printbuf * out,u32 version,struct jset * jset,struct jset_entry * entry)267 static void journal_entry_err_msg(struct printbuf *out,
268 u32 version,
269 struct jset *jset,
270 struct jset_entry *entry)
271 {
272 prt_str(out, "invalid journal entry, version=");
273 bch2_version_to_text(out, version);
274
275 if (entry) {
276 prt_str(out, " type=");
277 bch2_prt_jset_entry_type(out, entry->type);
278 }
279
280 if (!jset) {
281 prt_printf(out, " in superblock");
282 } else {
283
284 prt_printf(out, " seq=%llu", le64_to_cpu(jset->seq));
285
286 if (entry)
287 prt_printf(out, " offset=%zi/%u",
288 (u64 *) entry - jset->_data,
289 le32_to_cpu(jset->u64s));
290 }
291
292 prt_str(out, ": ");
293 }
294
295 #define journal_entry_err(c, version, jset, entry, _err, msg, ...) \
296 ({ \
297 struct printbuf _buf = PRINTBUF; \
298 \
299 journal_entry_err_msg(&_buf, version, jset, entry); \
300 prt_printf(&_buf, msg, ##__VA_ARGS__); \
301 \
302 switch (flags & BCH_VALIDATE_write) { \
303 case READ: \
304 mustfix_fsck_err(c, _err, "%s", _buf.buf); \
305 break; \
306 case WRITE: \
307 bch2_sb_error_count(c, BCH_FSCK_ERR_##_err); \
308 bch_err(c, "corrupt metadata before write: %s\n", _buf.buf);\
309 if (bch2_fs_inconsistent(c)) { \
310 ret = -BCH_ERR_fsck_errors_not_fixed; \
311 goto fsck_err; \
312 } \
313 break; \
314 } \
315 \
316 printbuf_exit(&_buf); \
317 true; \
318 })
319
320 #define journal_entry_err_on(cond, ...) \
321 ((cond) ? journal_entry_err(__VA_ARGS__) : false)
322
323 #define FSCK_DELETED_KEY 5
324
journal_validate_key(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned level,enum btree_id btree_id,struct bkey_i * k,unsigned version,int big_endian,enum bch_validate_flags flags)325 static int journal_validate_key(struct bch_fs *c,
326 struct jset *jset,
327 struct jset_entry *entry,
328 unsigned level, enum btree_id btree_id,
329 struct bkey_i *k,
330 unsigned version, int big_endian,
331 enum bch_validate_flags flags)
332 {
333 int write = flags & BCH_VALIDATE_write;
334 void *next = vstruct_next(entry);
335 int ret = 0;
336
337 if (journal_entry_err_on(!k->k.u64s,
338 c, version, jset, entry,
339 journal_entry_bkey_u64s_0,
340 "k->u64s 0")) {
341 entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
342 journal_entry_null_range(vstruct_next(entry), next);
343 return FSCK_DELETED_KEY;
344 }
345
346 if (journal_entry_err_on((void *) bkey_next(k) >
347 (void *) vstruct_next(entry),
348 c, version, jset, entry,
349 journal_entry_bkey_past_end,
350 "extends past end of journal entry")) {
351 entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
352 journal_entry_null_range(vstruct_next(entry), next);
353 return FSCK_DELETED_KEY;
354 }
355
356 if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT,
357 c, version, jset, entry,
358 journal_entry_bkey_bad_format,
359 "bad format %u", k->k.format)) {
360 le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
361 memmove(k, bkey_next(k), next - (void *) bkey_next(k));
362 journal_entry_null_range(vstruct_next(entry), next);
363 return FSCK_DELETED_KEY;
364 }
365
366 if (!write)
367 bch2_bkey_compat(level, btree_id, version, big_endian,
368 write, NULL, bkey_to_packed(k));
369
370 ret = bch2_bkey_validate(c, bkey_i_to_s_c(k),
371 __btree_node_type(level, btree_id), write);
372 if (ret == -BCH_ERR_fsck_delete_bkey) {
373 le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
374 memmove(k, bkey_next(k), next - (void *) bkey_next(k));
375 journal_entry_null_range(vstruct_next(entry), next);
376 return FSCK_DELETED_KEY;
377 }
378 if (ret)
379 goto fsck_err;
380
381 if (write)
382 bch2_bkey_compat(level, btree_id, version, big_endian,
383 write, NULL, bkey_to_packed(k));
384 fsck_err:
385 return ret;
386 }
387
journal_entry_btree_keys_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)388 static int journal_entry_btree_keys_validate(struct bch_fs *c,
389 struct jset *jset,
390 struct jset_entry *entry,
391 unsigned version, int big_endian,
392 enum bch_validate_flags flags)
393 {
394 struct bkey_i *k = entry->start;
395
396 while (k != vstruct_last(entry)) {
397 int ret = journal_validate_key(c, jset, entry,
398 entry->level,
399 entry->btree_id,
400 k, version, big_endian,
401 flags|BCH_VALIDATE_journal);
402 if (ret == FSCK_DELETED_KEY)
403 continue;
404 else if (ret)
405 return ret;
406
407 k = bkey_next(k);
408 }
409
410 return 0;
411 }
412
journal_entry_btree_keys_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)413 static void journal_entry_btree_keys_to_text(struct printbuf *out, struct bch_fs *c,
414 struct jset_entry *entry)
415 {
416 bool first = true;
417
418 jset_entry_for_each_key(entry, k) {
419 if (!first) {
420 prt_newline(out);
421 bch2_prt_jset_entry_type(out, entry->type);
422 prt_str(out, ": ");
423 }
424 prt_printf(out, "btree=%s l=%u ", bch2_btree_id_str(entry->btree_id), entry->level);
425 bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(k));
426 first = false;
427 }
428 }
429
journal_entry_btree_root_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)430 static int journal_entry_btree_root_validate(struct bch_fs *c,
431 struct jset *jset,
432 struct jset_entry *entry,
433 unsigned version, int big_endian,
434 enum bch_validate_flags flags)
435 {
436 struct bkey_i *k = entry->start;
437 int ret = 0;
438
439 if (journal_entry_err_on(!entry->u64s ||
440 le16_to_cpu(entry->u64s) != k->k.u64s,
441 c, version, jset, entry,
442 journal_entry_btree_root_bad_size,
443 "invalid btree root journal entry: wrong number of keys")) {
444 void *next = vstruct_next(entry);
445 /*
446 * we don't want to null out this jset_entry,
447 * just the contents, so that later we can tell
448 * we were _supposed_ to have a btree root
449 */
450 entry->u64s = 0;
451 journal_entry_null_range(vstruct_next(entry), next);
452 return 0;
453 }
454
455 ret = journal_validate_key(c, jset, entry, 1, entry->btree_id, k,
456 version, big_endian, flags);
457 if (ret == FSCK_DELETED_KEY)
458 ret = 0;
459 fsck_err:
460 return ret;
461 }
462
journal_entry_btree_root_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)463 static void journal_entry_btree_root_to_text(struct printbuf *out, struct bch_fs *c,
464 struct jset_entry *entry)
465 {
466 journal_entry_btree_keys_to_text(out, c, entry);
467 }
468
journal_entry_prio_ptrs_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)469 static int journal_entry_prio_ptrs_validate(struct bch_fs *c,
470 struct jset *jset,
471 struct jset_entry *entry,
472 unsigned version, int big_endian,
473 enum bch_validate_flags flags)
474 {
475 /* obsolete, don't care: */
476 return 0;
477 }
478
journal_entry_prio_ptrs_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)479 static void journal_entry_prio_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
480 struct jset_entry *entry)
481 {
482 }
483
journal_entry_blacklist_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)484 static int journal_entry_blacklist_validate(struct bch_fs *c,
485 struct jset *jset,
486 struct jset_entry *entry,
487 unsigned version, int big_endian,
488 enum bch_validate_flags flags)
489 {
490 int ret = 0;
491
492 if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 1,
493 c, version, jset, entry,
494 journal_entry_blacklist_bad_size,
495 "invalid journal seq blacklist entry: bad size")) {
496 journal_entry_null_range(entry, vstruct_next(entry));
497 }
498 fsck_err:
499 return ret;
500 }
501
journal_entry_blacklist_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)502 static void journal_entry_blacklist_to_text(struct printbuf *out, struct bch_fs *c,
503 struct jset_entry *entry)
504 {
505 struct jset_entry_blacklist *bl =
506 container_of(entry, struct jset_entry_blacklist, entry);
507
508 prt_printf(out, "seq=%llu", le64_to_cpu(bl->seq));
509 }
510
journal_entry_blacklist_v2_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)511 static int journal_entry_blacklist_v2_validate(struct bch_fs *c,
512 struct jset *jset,
513 struct jset_entry *entry,
514 unsigned version, int big_endian,
515 enum bch_validate_flags flags)
516 {
517 struct jset_entry_blacklist_v2 *bl_entry;
518 int ret = 0;
519
520 if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 2,
521 c, version, jset, entry,
522 journal_entry_blacklist_v2_bad_size,
523 "invalid journal seq blacklist entry: bad size")) {
524 journal_entry_null_range(entry, vstruct_next(entry));
525 goto out;
526 }
527
528 bl_entry = container_of(entry, struct jset_entry_blacklist_v2, entry);
529
530 if (journal_entry_err_on(le64_to_cpu(bl_entry->start) >
531 le64_to_cpu(bl_entry->end),
532 c, version, jset, entry,
533 journal_entry_blacklist_v2_start_past_end,
534 "invalid journal seq blacklist entry: start > end")) {
535 journal_entry_null_range(entry, vstruct_next(entry));
536 }
537 out:
538 fsck_err:
539 return ret;
540 }
541
journal_entry_blacklist_v2_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)542 static void journal_entry_blacklist_v2_to_text(struct printbuf *out, struct bch_fs *c,
543 struct jset_entry *entry)
544 {
545 struct jset_entry_blacklist_v2 *bl =
546 container_of(entry, struct jset_entry_blacklist_v2, entry);
547
548 prt_printf(out, "start=%llu end=%llu",
549 le64_to_cpu(bl->start),
550 le64_to_cpu(bl->end));
551 }
552
journal_entry_usage_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)553 static int journal_entry_usage_validate(struct bch_fs *c,
554 struct jset *jset,
555 struct jset_entry *entry,
556 unsigned version, int big_endian,
557 enum bch_validate_flags flags)
558 {
559 struct jset_entry_usage *u =
560 container_of(entry, struct jset_entry_usage, entry);
561 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
562 int ret = 0;
563
564 if (journal_entry_err_on(bytes < sizeof(*u),
565 c, version, jset, entry,
566 journal_entry_usage_bad_size,
567 "invalid journal entry usage: bad size")) {
568 journal_entry_null_range(entry, vstruct_next(entry));
569 return ret;
570 }
571
572 fsck_err:
573 return ret;
574 }
575
journal_entry_usage_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)576 static void journal_entry_usage_to_text(struct printbuf *out, struct bch_fs *c,
577 struct jset_entry *entry)
578 {
579 struct jset_entry_usage *u =
580 container_of(entry, struct jset_entry_usage, entry);
581
582 prt_str(out, "type=");
583 bch2_prt_fs_usage_type(out, u->entry.btree_id);
584 prt_printf(out, " v=%llu", le64_to_cpu(u->v));
585 }
586
journal_entry_data_usage_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)587 static int journal_entry_data_usage_validate(struct bch_fs *c,
588 struct jset *jset,
589 struct jset_entry *entry,
590 unsigned version, int big_endian,
591 enum bch_validate_flags flags)
592 {
593 struct jset_entry_data_usage *u =
594 container_of(entry, struct jset_entry_data_usage, entry);
595 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
596 struct printbuf err = PRINTBUF;
597 int ret = 0;
598
599 if (journal_entry_err_on(bytes < sizeof(*u) ||
600 bytes < sizeof(*u) + u->r.nr_devs,
601 c, version, jset, entry,
602 journal_entry_data_usage_bad_size,
603 "invalid journal entry usage: bad size")) {
604 journal_entry_null_range(entry, vstruct_next(entry));
605 goto out;
606 }
607
608 if (journal_entry_err_on(bch2_replicas_entry_validate(&u->r, c->disk_sb.sb, &err),
609 c, version, jset, entry,
610 journal_entry_data_usage_bad_size,
611 "invalid journal entry usage: %s", err.buf)) {
612 journal_entry_null_range(entry, vstruct_next(entry));
613 goto out;
614 }
615 out:
616 fsck_err:
617 printbuf_exit(&err);
618 return ret;
619 }
620
journal_entry_data_usage_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)621 static void journal_entry_data_usage_to_text(struct printbuf *out, struct bch_fs *c,
622 struct jset_entry *entry)
623 {
624 struct jset_entry_data_usage *u =
625 container_of(entry, struct jset_entry_data_usage, entry);
626
627 bch2_replicas_entry_to_text(out, &u->r);
628 prt_printf(out, "=%llu", le64_to_cpu(u->v));
629 }
630
journal_entry_clock_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)631 static int journal_entry_clock_validate(struct bch_fs *c,
632 struct jset *jset,
633 struct jset_entry *entry,
634 unsigned version, int big_endian,
635 enum bch_validate_flags flags)
636 {
637 struct jset_entry_clock *clock =
638 container_of(entry, struct jset_entry_clock, entry);
639 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
640 int ret = 0;
641
642 if (journal_entry_err_on(bytes != sizeof(*clock),
643 c, version, jset, entry,
644 journal_entry_clock_bad_size,
645 "bad size")) {
646 journal_entry_null_range(entry, vstruct_next(entry));
647 return ret;
648 }
649
650 if (journal_entry_err_on(clock->rw > 1,
651 c, version, jset, entry,
652 journal_entry_clock_bad_rw,
653 "bad rw")) {
654 journal_entry_null_range(entry, vstruct_next(entry));
655 return ret;
656 }
657
658 fsck_err:
659 return ret;
660 }
661
journal_entry_clock_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)662 static void journal_entry_clock_to_text(struct printbuf *out, struct bch_fs *c,
663 struct jset_entry *entry)
664 {
665 struct jset_entry_clock *clock =
666 container_of(entry, struct jset_entry_clock, entry);
667
668 prt_printf(out, "%s=%llu", clock->rw ? "write" : "read", le64_to_cpu(clock->time));
669 }
670
journal_entry_dev_usage_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)671 static int journal_entry_dev_usage_validate(struct bch_fs *c,
672 struct jset *jset,
673 struct jset_entry *entry,
674 unsigned version, int big_endian,
675 enum bch_validate_flags flags)
676 {
677 struct jset_entry_dev_usage *u =
678 container_of(entry, struct jset_entry_dev_usage, entry);
679 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
680 unsigned expected = sizeof(*u);
681 int ret = 0;
682
683 if (journal_entry_err_on(bytes < expected,
684 c, version, jset, entry,
685 journal_entry_dev_usage_bad_size,
686 "bad size (%u < %u)",
687 bytes, expected)) {
688 journal_entry_null_range(entry, vstruct_next(entry));
689 return ret;
690 }
691
692 if (journal_entry_err_on(u->pad,
693 c, version, jset, entry,
694 journal_entry_dev_usage_bad_pad,
695 "bad pad")) {
696 journal_entry_null_range(entry, vstruct_next(entry));
697 return ret;
698 }
699
700 fsck_err:
701 return ret;
702 }
703
journal_entry_dev_usage_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)704 static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs *c,
705 struct jset_entry *entry)
706 {
707 struct jset_entry_dev_usage *u =
708 container_of(entry, struct jset_entry_dev_usage, entry);
709 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
710
711 prt_printf(out, "dev=%u", le32_to_cpu(u->dev));
712
713 printbuf_indent_add(out, 2);
714 for (i = 0; i < nr_types; i++) {
715 prt_newline(out);
716 bch2_prt_data_type(out, i);
717 prt_printf(out, ": buckets=%llu sectors=%llu fragmented=%llu",
718 le64_to_cpu(u->d[i].buckets),
719 le64_to_cpu(u->d[i].sectors),
720 le64_to_cpu(u->d[i].fragmented));
721 }
722 printbuf_indent_sub(out, 2);
723 }
724
journal_entry_log_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)725 static int journal_entry_log_validate(struct bch_fs *c,
726 struct jset *jset,
727 struct jset_entry *entry,
728 unsigned version, int big_endian,
729 enum bch_validate_flags flags)
730 {
731 return 0;
732 }
733
journal_entry_log_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)734 static void journal_entry_log_to_text(struct printbuf *out, struct bch_fs *c,
735 struct jset_entry *entry)
736 {
737 struct jset_entry_log *l = container_of(entry, struct jset_entry_log, entry);
738 unsigned bytes = vstruct_bytes(entry) - offsetof(struct jset_entry_log, d);
739
740 prt_printf(out, "%.*s", bytes, l->d);
741 }
742
journal_entry_overwrite_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)743 static int journal_entry_overwrite_validate(struct bch_fs *c,
744 struct jset *jset,
745 struct jset_entry *entry,
746 unsigned version, int big_endian,
747 enum bch_validate_flags flags)
748 {
749 return journal_entry_btree_keys_validate(c, jset, entry,
750 version, big_endian, READ);
751 }
752
journal_entry_overwrite_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)753 static void journal_entry_overwrite_to_text(struct printbuf *out, struct bch_fs *c,
754 struct jset_entry *entry)
755 {
756 journal_entry_btree_keys_to_text(out, c, entry);
757 }
758
journal_entry_write_buffer_keys_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)759 static int journal_entry_write_buffer_keys_validate(struct bch_fs *c,
760 struct jset *jset,
761 struct jset_entry *entry,
762 unsigned version, int big_endian,
763 enum bch_validate_flags flags)
764 {
765 return journal_entry_btree_keys_validate(c, jset, entry,
766 version, big_endian, READ);
767 }
768
journal_entry_write_buffer_keys_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)769 static void journal_entry_write_buffer_keys_to_text(struct printbuf *out, struct bch_fs *c,
770 struct jset_entry *entry)
771 {
772 journal_entry_btree_keys_to_text(out, c, entry);
773 }
774
journal_entry_datetime_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)775 static int journal_entry_datetime_validate(struct bch_fs *c,
776 struct jset *jset,
777 struct jset_entry *entry,
778 unsigned version, int big_endian,
779 enum bch_validate_flags flags)
780 {
781 unsigned bytes = vstruct_bytes(entry);
782 unsigned expected = 16;
783 int ret = 0;
784
785 if (journal_entry_err_on(vstruct_bytes(entry) < expected,
786 c, version, jset, entry,
787 journal_entry_dev_usage_bad_size,
788 "bad size (%u < %u)",
789 bytes, expected)) {
790 journal_entry_null_range(entry, vstruct_next(entry));
791 return ret;
792 }
793 fsck_err:
794 return ret;
795 }
796
journal_entry_datetime_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)797 static void journal_entry_datetime_to_text(struct printbuf *out, struct bch_fs *c,
798 struct jset_entry *entry)
799 {
800 struct jset_entry_datetime *datetime =
801 container_of(entry, struct jset_entry_datetime, entry);
802
803 bch2_prt_datetime(out, le64_to_cpu(datetime->seconds));
804 }
805
806 struct jset_entry_ops {
807 int (*validate)(struct bch_fs *, struct jset *,
808 struct jset_entry *, unsigned, int,
809 enum bch_validate_flags);
810 void (*to_text)(struct printbuf *, struct bch_fs *, struct jset_entry *);
811 };
812
813 static const struct jset_entry_ops bch2_jset_entry_ops[] = {
814 #define x(f, nr) \
815 [BCH_JSET_ENTRY_##f] = (struct jset_entry_ops) { \
816 .validate = journal_entry_##f##_validate, \
817 .to_text = journal_entry_##f##_to_text, \
818 },
819 BCH_JSET_ENTRY_TYPES()
820 #undef x
821 };
822
bch2_journal_entry_validate(struct bch_fs * c,struct jset * jset,struct jset_entry * entry,unsigned version,int big_endian,enum bch_validate_flags flags)823 int bch2_journal_entry_validate(struct bch_fs *c,
824 struct jset *jset,
825 struct jset_entry *entry,
826 unsigned version, int big_endian,
827 enum bch_validate_flags flags)
828 {
829 return entry->type < BCH_JSET_ENTRY_NR
830 ? bch2_jset_entry_ops[entry->type].validate(c, jset, entry,
831 version, big_endian, flags)
832 : 0;
833 }
834
bch2_journal_entry_to_text(struct printbuf * out,struct bch_fs * c,struct jset_entry * entry)835 void bch2_journal_entry_to_text(struct printbuf *out, struct bch_fs *c,
836 struct jset_entry *entry)
837 {
838 bch2_prt_jset_entry_type(out, entry->type);
839
840 if (entry->type < BCH_JSET_ENTRY_NR) {
841 prt_str(out, ": ");
842 bch2_jset_entry_ops[entry->type].to_text(out, c, entry);
843 }
844 }
845
jset_validate_entries(struct bch_fs * c,struct jset * jset,enum bch_validate_flags flags)846 static int jset_validate_entries(struct bch_fs *c, struct jset *jset,
847 enum bch_validate_flags flags)
848 {
849 unsigned version = le32_to_cpu(jset->version);
850 int ret = 0;
851
852 vstruct_for_each(jset, entry) {
853 if (journal_entry_err_on(vstruct_next(entry) > vstruct_last(jset),
854 c, version, jset, entry,
855 journal_entry_past_jset_end,
856 "journal entry extends past end of jset")) {
857 jset->u64s = cpu_to_le32((u64 *) entry - jset->_data);
858 break;
859 }
860
861 ret = bch2_journal_entry_validate(c, jset, entry,
862 version, JSET_BIG_ENDIAN(jset), flags);
863 if (ret)
864 break;
865 }
866 fsck_err:
867 return ret;
868 }
869
jset_validate(struct bch_fs * c,struct bch_dev * ca,struct jset * jset,u64 sector,enum bch_validate_flags flags)870 static int jset_validate(struct bch_fs *c,
871 struct bch_dev *ca,
872 struct jset *jset, u64 sector,
873 enum bch_validate_flags flags)
874 {
875 unsigned version;
876 int ret = 0;
877
878 if (le64_to_cpu(jset->magic) != jset_magic(c))
879 return JOURNAL_ENTRY_NONE;
880
881 version = le32_to_cpu(jset->version);
882 if (journal_entry_err_on(!bch2_version_compatible(version),
883 c, version, jset, NULL,
884 jset_unsupported_version,
885 "%s sector %llu seq %llu: incompatible journal entry version %u.%u",
886 ca ? ca->name : c->name,
887 sector, le64_to_cpu(jset->seq),
888 BCH_VERSION_MAJOR(version),
889 BCH_VERSION_MINOR(version))) {
890 /* don't try to continue: */
891 return -EINVAL;
892 }
893
894 if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)),
895 c, version, jset, NULL,
896 jset_unknown_csum,
897 "%s sector %llu seq %llu: journal entry with unknown csum type %llu",
898 ca ? ca->name : c->name,
899 sector, le64_to_cpu(jset->seq),
900 JSET_CSUM_TYPE(jset)))
901 ret = JOURNAL_ENTRY_BAD;
902
903 /* last_seq is ignored when JSET_NO_FLUSH is true */
904 if (journal_entry_err_on(!JSET_NO_FLUSH(jset) &&
905 le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq),
906 c, version, jset, NULL,
907 jset_last_seq_newer_than_seq,
908 "invalid journal entry: last_seq > seq (%llu > %llu)",
909 le64_to_cpu(jset->last_seq),
910 le64_to_cpu(jset->seq))) {
911 jset->last_seq = jset->seq;
912 return JOURNAL_ENTRY_BAD;
913 }
914
915 ret = jset_validate_entries(c, jset, flags);
916 fsck_err:
917 return ret;
918 }
919
jset_validate_early(struct bch_fs * c,struct bch_dev * ca,struct jset * jset,u64 sector,unsigned bucket_sectors_left,unsigned sectors_read)920 static int jset_validate_early(struct bch_fs *c,
921 struct bch_dev *ca,
922 struct jset *jset, u64 sector,
923 unsigned bucket_sectors_left,
924 unsigned sectors_read)
925 {
926 size_t bytes = vstruct_bytes(jset);
927 unsigned version;
928 enum bch_validate_flags flags = BCH_VALIDATE_journal;
929 int ret = 0;
930
931 if (le64_to_cpu(jset->magic) != jset_magic(c))
932 return JOURNAL_ENTRY_NONE;
933
934 version = le32_to_cpu(jset->version);
935 if (journal_entry_err_on(!bch2_version_compatible(version),
936 c, version, jset, NULL,
937 jset_unsupported_version,
938 "%s sector %llu seq %llu: unknown journal entry version %u.%u",
939 ca ? ca->name : c->name,
940 sector, le64_to_cpu(jset->seq),
941 BCH_VERSION_MAJOR(version),
942 BCH_VERSION_MINOR(version))) {
943 /* don't try to continue: */
944 return -EINVAL;
945 }
946
947 if (bytes > (sectors_read << 9) &&
948 sectors_read < bucket_sectors_left)
949 return JOURNAL_ENTRY_REREAD;
950
951 if (journal_entry_err_on(bytes > bucket_sectors_left << 9,
952 c, version, jset, NULL,
953 jset_past_bucket_end,
954 "%s sector %llu seq %llu: journal entry too big (%zu bytes)",
955 ca ? ca->name : c->name,
956 sector, le64_to_cpu(jset->seq), bytes))
957 le32_add_cpu(&jset->u64s,
958 -((bytes - (bucket_sectors_left << 9)) / 8));
959 fsck_err:
960 return ret;
961 }
962
963 struct journal_read_buf {
964 void *data;
965 size_t size;
966 };
967
journal_read_buf_realloc(struct journal_read_buf * b,size_t new_size)968 static int journal_read_buf_realloc(struct journal_read_buf *b,
969 size_t new_size)
970 {
971 void *n;
972
973 /* the bios are sized for this many pages, max: */
974 if (new_size > JOURNAL_ENTRY_SIZE_MAX)
975 return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
976
977 new_size = roundup_pow_of_two(new_size);
978 n = kvmalloc(new_size, GFP_KERNEL);
979 if (!n)
980 return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
981
982 kvfree(b->data);
983 b->data = n;
984 b->size = new_size;
985 return 0;
986 }
987
journal_read_bucket(struct bch_dev * ca,struct journal_read_buf * buf,struct journal_list * jlist,unsigned bucket)988 static int journal_read_bucket(struct bch_dev *ca,
989 struct journal_read_buf *buf,
990 struct journal_list *jlist,
991 unsigned bucket)
992 {
993 struct bch_fs *c = ca->fs;
994 struct journal_device *ja = &ca->journal;
995 struct jset *j = NULL;
996 unsigned sectors, sectors_read = 0;
997 u64 offset = bucket_to_sector(ca, ja->buckets[bucket]),
998 end = offset + ca->mi.bucket_size;
999 bool saw_bad = false, csum_good;
1000 struct printbuf err = PRINTBUF;
1001 int ret = 0;
1002
1003 pr_debug("reading %u", bucket);
1004
1005 while (offset < end) {
1006 if (!sectors_read) {
1007 struct bio *bio;
1008 unsigned nr_bvecs;
1009 reread:
1010 sectors_read = min_t(unsigned,
1011 end - offset, buf->size >> 9);
1012 nr_bvecs = buf_pages(buf->data, sectors_read << 9);
1013
1014 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
1015 bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ);
1016
1017 bio->bi_iter.bi_sector = offset;
1018 bch2_bio_map(bio, buf->data, sectors_read << 9);
1019
1020 ret = submit_bio_wait(bio);
1021 kfree(bio);
1022
1023 if (bch2_dev_io_err_on(ret, ca, BCH_MEMBER_ERROR_read,
1024 "journal read error: sector %llu",
1025 offset) ||
1026 bch2_meta_read_fault("journal")) {
1027 /*
1028 * We don't error out of the recovery process
1029 * here, since the relevant journal entry may be
1030 * found on a different device, and missing or
1031 * no journal entries will be handled later
1032 */
1033 goto out;
1034 }
1035
1036 j = buf->data;
1037 }
1038
1039 ret = jset_validate_early(c, ca, j, offset,
1040 end - offset, sectors_read);
1041 switch (ret) {
1042 case 0:
1043 sectors = vstruct_sectors(j, c->block_bits);
1044 break;
1045 case JOURNAL_ENTRY_REREAD:
1046 if (vstruct_bytes(j) > buf->size) {
1047 ret = journal_read_buf_realloc(buf,
1048 vstruct_bytes(j));
1049 if (ret)
1050 goto err;
1051 }
1052 goto reread;
1053 case JOURNAL_ENTRY_NONE:
1054 if (!saw_bad)
1055 goto out;
1056 /*
1057 * On checksum error we don't really trust the size
1058 * field of the journal entry we read, so try reading
1059 * again at next block boundary:
1060 */
1061 sectors = block_sectors(c);
1062 goto next_block;
1063 default:
1064 goto err;
1065 }
1066
1067 if (le64_to_cpu(j->seq) > ja->highest_seq_found) {
1068 ja->highest_seq_found = le64_to_cpu(j->seq);
1069 ja->cur_idx = bucket;
1070 ja->sectors_free = ca->mi.bucket_size -
1071 bucket_remainder(ca, offset) - sectors;
1072 }
1073
1074 /*
1075 * This happens sometimes if we don't have discards on -
1076 * when we've partially overwritten a bucket with new
1077 * journal entries. We don't need the rest of the
1078 * bucket:
1079 */
1080 if (le64_to_cpu(j->seq) < ja->bucket_seq[bucket])
1081 goto out;
1082
1083 ja->bucket_seq[bucket] = le64_to_cpu(j->seq);
1084
1085 enum bch_csum_type csum_type = JSET_CSUM_TYPE(j);
1086 struct bch_csum csum;
1087 csum_good = jset_csum_good(c, j, &csum);
1088
1089 if (bch2_dev_io_err_on(!csum_good, ca, BCH_MEMBER_ERROR_checksum,
1090 "%s",
1091 (printbuf_reset(&err),
1092 prt_str(&err, "journal "),
1093 bch2_csum_err_msg(&err, csum_type, j->csum, csum),
1094 err.buf)))
1095 saw_bad = true;
1096
1097 ret = bch2_encrypt(c, JSET_CSUM_TYPE(j), journal_nonce(j),
1098 j->encrypted_start,
1099 vstruct_end(j) - (void *) j->encrypted_start);
1100 bch2_fs_fatal_err_on(ret, c, "decrypting journal entry: %s", bch2_err_str(ret));
1101
1102 mutex_lock(&jlist->lock);
1103 ret = journal_entry_add(c, ca, (struct journal_ptr) {
1104 .csum_good = csum_good,
1105 .dev = ca->dev_idx,
1106 .bucket = bucket,
1107 .bucket_offset = offset -
1108 bucket_to_sector(ca, ja->buckets[bucket]),
1109 .sector = offset,
1110 }, jlist, j);
1111 mutex_unlock(&jlist->lock);
1112
1113 switch (ret) {
1114 case JOURNAL_ENTRY_ADD_OK:
1115 break;
1116 case JOURNAL_ENTRY_ADD_OUT_OF_RANGE:
1117 break;
1118 default:
1119 goto err;
1120 }
1121 next_block:
1122 pr_debug("next");
1123 offset += sectors;
1124 sectors_read -= sectors;
1125 j = ((void *) j) + (sectors << 9);
1126 }
1127
1128 out:
1129 ret = 0;
1130 err:
1131 printbuf_exit(&err);
1132 return ret;
1133 }
1134
CLOSURE_CALLBACK(bch2_journal_read_device)1135 static CLOSURE_CALLBACK(bch2_journal_read_device)
1136 {
1137 closure_type(ja, struct journal_device, read);
1138 struct bch_dev *ca = container_of(ja, struct bch_dev, journal);
1139 struct bch_fs *c = ca->fs;
1140 struct journal_list *jlist =
1141 container_of(cl->parent, struct journal_list, cl);
1142 struct journal_read_buf buf = { NULL, 0 };
1143 unsigned i;
1144 int ret = 0;
1145
1146 if (!ja->nr)
1147 goto out;
1148
1149 ret = journal_read_buf_realloc(&buf, PAGE_SIZE);
1150 if (ret)
1151 goto err;
1152
1153 pr_debug("%u journal buckets", ja->nr);
1154
1155 for (i = 0; i < ja->nr; i++) {
1156 ret = journal_read_bucket(ca, &buf, jlist, i);
1157 if (ret)
1158 goto err;
1159 }
1160
1161 /*
1162 * Set dirty_idx to indicate the entire journal is full and needs to be
1163 * reclaimed - journal reclaim will immediately reclaim whatever isn't
1164 * pinned when it first runs:
1165 */
1166 ja->discard_idx = ja->dirty_idx_ondisk =
1167 ja->dirty_idx = (ja->cur_idx + 1) % ja->nr;
1168 out:
1169 bch_verbose(c, "journal read done on device %s, ret %i", ca->name, ret);
1170 kvfree(buf.data);
1171 percpu_ref_put(&ca->io_ref);
1172 closure_return(cl);
1173 return;
1174 err:
1175 mutex_lock(&jlist->lock);
1176 jlist->ret = ret;
1177 mutex_unlock(&jlist->lock);
1178 goto out;
1179 }
1180
bch2_journal_read(struct bch_fs * c,u64 * last_seq,u64 * blacklist_seq,u64 * start_seq)1181 int bch2_journal_read(struct bch_fs *c,
1182 u64 *last_seq,
1183 u64 *blacklist_seq,
1184 u64 *start_seq)
1185 {
1186 struct journal_list jlist;
1187 struct journal_replay *i, **_i, *prev = NULL;
1188 struct genradix_iter radix_iter;
1189 struct printbuf buf = PRINTBUF;
1190 bool degraded = false, last_write_torn = false;
1191 u64 seq;
1192 int ret = 0;
1193
1194 closure_init_stack(&jlist.cl);
1195 mutex_init(&jlist.lock);
1196 jlist.last_seq = 0;
1197 jlist.ret = 0;
1198
1199 for_each_member_device(c, ca) {
1200 if (!c->opts.fsck &&
1201 !(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_journal)))
1202 continue;
1203
1204 if ((ca->mi.state == BCH_MEMBER_STATE_rw ||
1205 ca->mi.state == BCH_MEMBER_STATE_ro) &&
1206 percpu_ref_tryget(&ca->io_ref))
1207 closure_call(&ca->journal.read,
1208 bch2_journal_read_device,
1209 system_unbound_wq,
1210 &jlist.cl);
1211 else
1212 degraded = true;
1213 }
1214
1215 closure_sync(&jlist.cl);
1216
1217 if (jlist.ret)
1218 return jlist.ret;
1219
1220 *last_seq = 0;
1221 *start_seq = 0;
1222 *blacklist_seq = 0;
1223
1224 /*
1225 * Find most recent flush entry, and ignore newer non flush entries -
1226 * those entries will be blacklisted:
1227 */
1228 genradix_for_each_reverse(&c->journal_entries, radix_iter, _i) {
1229 enum bch_validate_flags flags = BCH_VALIDATE_journal;
1230
1231 i = *_i;
1232
1233 if (journal_replay_ignore(i))
1234 continue;
1235
1236 if (!*start_seq)
1237 *blacklist_seq = *start_seq = le64_to_cpu(i->j.seq) + 1;
1238
1239 if (JSET_NO_FLUSH(&i->j)) {
1240 i->ignore_blacklisted = true;
1241 continue;
1242 }
1243
1244 if (!last_write_torn && !i->csum_good) {
1245 last_write_torn = true;
1246 i->ignore_blacklisted = true;
1247 continue;
1248 }
1249
1250 if (journal_entry_err_on(le64_to_cpu(i->j.last_seq) > le64_to_cpu(i->j.seq),
1251 c, le32_to_cpu(i->j.version), &i->j, NULL,
1252 jset_last_seq_newer_than_seq,
1253 "invalid journal entry: last_seq > seq (%llu > %llu)",
1254 le64_to_cpu(i->j.last_seq),
1255 le64_to_cpu(i->j.seq)))
1256 i->j.last_seq = i->j.seq;
1257
1258 *last_seq = le64_to_cpu(i->j.last_seq);
1259 *blacklist_seq = le64_to_cpu(i->j.seq) + 1;
1260 break;
1261 }
1262
1263 if (!*start_seq) {
1264 bch_info(c, "journal read done, but no entries found");
1265 return 0;
1266 }
1267
1268 if (!*last_seq) {
1269 fsck_err(c, dirty_but_no_journal_entries_post_drop_nonflushes,
1270 "journal read done, but no entries found after dropping non-flushes");
1271 return 0;
1272 }
1273
1274 bch_info(c, "journal read done, replaying entries %llu-%llu",
1275 *last_seq, *blacklist_seq - 1);
1276
1277 if (*start_seq != *blacklist_seq)
1278 bch_info(c, "dropped unflushed entries %llu-%llu",
1279 *blacklist_seq, *start_seq - 1);
1280
1281 /* Drop blacklisted entries and entries older than last_seq: */
1282 genradix_for_each(&c->journal_entries, radix_iter, _i) {
1283 i = *_i;
1284
1285 if (journal_replay_ignore(i))
1286 continue;
1287
1288 seq = le64_to_cpu(i->j.seq);
1289 if (seq < *last_seq) {
1290 journal_replay_free(c, i, false);
1291 continue;
1292 }
1293
1294 if (bch2_journal_seq_is_blacklisted(c, seq, true)) {
1295 fsck_err_on(!JSET_NO_FLUSH(&i->j), c,
1296 jset_seq_blacklisted,
1297 "found blacklisted journal entry %llu", seq);
1298 i->ignore_blacklisted = true;
1299 }
1300 }
1301
1302 /* Check for missing entries: */
1303 seq = *last_seq;
1304 genradix_for_each(&c->journal_entries, radix_iter, _i) {
1305 i = *_i;
1306
1307 if (journal_replay_ignore(i))
1308 continue;
1309
1310 BUG_ON(seq > le64_to_cpu(i->j.seq));
1311
1312 while (seq < le64_to_cpu(i->j.seq)) {
1313 u64 missing_start, missing_end;
1314 struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
1315
1316 while (seq < le64_to_cpu(i->j.seq) &&
1317 bch2_journal_seq_is_blacklisted(c, seq, false))
1318 seq++;
1319
1320 if (seq == le64_to_cpu(i->j.seq))
1321 break;
1322
1323 missing_start = seq;
1324
1325 while (seq < le64_to_cpu(i->j.seq) &&
1326 !bch2_journal_seq_is_blacklisted(c, seq, false))
1327 seq++;
1328
1329 if (prev) {
1330 bch2_journal_ptrs_to_text(&buf1, c, prev);
1331 prt_printf(&buf1, " size %zu", vstruct_sectors(&prev->j, c->block_bits));
1332 } else
1333 prt_printf(&buf1, "(none)");
1334 bch2_journal_ptrs_to_text(&buf2, c, i);
1335
1336 missing_end = seq - 1;
1337 fsck_err(c, journal_entries_missing,
1338 "journal entries %llu-%llu missing! (replaying %llu-%llu)\n"
1339 " prev at %s\n"
1340 " next at %s, continue?",
1341 missing_start, missing_end,
1342 *last_seq, *blacklist_seq - 1,
1343 buf1.buf, buf2.buf);
1344
1345 printbuf_exit(&buf1);
1346 printbuf_exit(&buf2);
1347 }
1348
1349 prev = i;
1350 seq++;
1351 }
1352
1353 genradix_for_each(&c->journal_entries, radix_iter, _i) {
1354 struct bch_replicas_padded replicas = {
1355 .e.data_type = BCH_DATA_journal,
1356 .e.nr_required = 1,
1357 };
1358
1359 i = *_i;
1360 if (journal_replay_ignore(i))
1361 continue;
1362
1363 darray_for_each(i->ptrs, ptr) {
1364 struct bch_dev *ca = bch2_dev_have_ref(c, ptr->dev);
1365
1366 if (!ptr->csum_good)
1367 bch_err_dev_offset(ca, ptr->sector,
1368 "invalid journal checksum, seq %llu%s",
1369 le64_to_cpu(i->j.seq),
1370 i->csum_good ? " (had good copy on another device)" : "");
1371 }
1372
1373 ret = jset_validate(c,
1374 bch2_dev_have_ref(c, i->ptrs.data[0].dev),
1375 &i->j,
1376 i->ptrs.data[0].sector,
1377 READ);
1378 if (ret)
1379 goto err;
1380
1381 darray_for_each(i->ptrs, ptr)
1382 replicas.e.devs[replicas.e.nr_devs++] = ptr->dev;
1383
1384 bch2_replicas_entry_sort(&replicas.e);
1385
1386 printbuf_reset(&buf);
1387 bch2_replicas_entry_to_text(&buf, &replicas.e);
1388
1389 if (!degraded &&
1390 !bch2_replicas_marked(c, &replicas.e) &&
1391 (le64_to_cpu(i->j.seq) == *last_seq ||
1392 fsck_err(c, journal_entry_replicas_not_marked,
1393 "superblock not marked as containing replicas for journal entry %llu\n %s",
1394 le64_to_cpu(i->j.seq), buf.buf))) {
1395 ret = bch2_mark_replicas(c, &replicas.e);
1396 if (ret)
1397 goto err;
1398 }
1399 }
1400 err:
1401 fsck_err:
1402 printbuf_exit(&buf);
1403 return ret;
1404 }
1405
1406 /* journal write: */
1407
__journal_write_alloc(struct journal * j,struct journal_buf * w,struct dev_alloc_list * devs_sorted,unsigned sectors,unsigned * replicas,unsigned replicas_want)1408 static void __journal_write_alloc(struct journal *j,
1409 struct journal_buf *w,
1410 struct dev_alloc_list *devs_sorted,
1411 unsigned sectors,
1412 unsigned *replicas,
1413 unsigned replicas_want)
1414 {
1415 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1416 struct journal_device *ja;
1417 struct bch_dev *ca;
1418 unsigned i;
1419
1420 if (*replicas >= replicas_want)
1421 return;
1422
1423 for (i = 0; i < devs_sorted->nr; i++) {
1424 ca = rcu_dereference(c->devs[devs_sorted->devs[i]]);
1425 if (!ca)
1426 continue;
1427
1428 ja = &ca->journal;
1429
1430 /*
1431 * Check that we can use this device, and aren't already using
1432 * it:
1433 */
1434 if (!ca->mi.durability ||
1435 ca->mi.state != BCH_MEMBER_STATE_rw ||
1436 !ja->nr ||
1437 bch2_bkey_has_device_c(bkey_i_to_s_c(&w->key), ca->dev_idx) ||
1438 sectors > ja->sectors_free)
1439 continue;
1440
1441 bch2_dev_stripe_increment(ca, &j->wp.stripe);
1442
1443 bch2_bkey_append_ptr(&w->key,
1444 (struct bch_extent_ptr) {
1445 .offset = bucket_to_sector(ca,
1446 ja->buckets[ja->cur_idx]) +
1447 ca->mi.bucket_size -
1448 ja->sectors_free,
1449 .dev = ca->dev_idx,
1450 });
1451
1452 ja->sectors_free -= sectors;
1453 ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1454
1455 *replicas += ca->mi.durability;
1456
1457 if (*replicas >= replicas_want)
1458 break;
1459 }
1460 }
1461
1462 /**
1463 * journal_write_alloc - decide where to write next journal entry
1464 *
1465 * @j: journal object
1466 * @w: journal buf (entry to be written)
1467 *
1468 * Returns: 0 on success, or -EROFS on failure
1469 */
journal_write_alloc(struct journal * j,struct journal_buf * w)1470 static int journal_write_alloc(struct journal *j, struct journal_buf *w)
1471 {
1472 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1473 struct bch_devs_mask devs;
1474 struct journal_device *ja;
1475 struct bch_dev *ca;
1476 struct dev_alloc_list devs_sorted;
1477 unsigned sectors = vstruct_sectors(w->data, c->block_bits);
1478 unsigned target = c->opts.metadata_target ?:
1479 c->opts.foreground_target;
1480 unsigned i, replicas = 0, replicas_want =
1481 READ_ONCE(c->opts.metadata_replicas);
1482 unsigned replicas_need = min_t(unsigned, replicas_want,
1483 READ_ONCE(c->opts.metadata_replicas_required));
1484
1485 rcu_read_lock();
1486 retry:
1487 devs = target_rw_devs(c, BCH_DATA_journal, target);
1488
1489 devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe, &devs);
1490
1491 __journal_write_alloc(j, w, &devs_sorted,
1492 sectors, &replicas, replicas_want);
1493
1494 if (replicas >= replicas_want)
1495 goto done;
1496
1497 for (i = 0; i < devs_sorted.nr; i++) {
1498 ca = rcu_dereference(c->devs[devs_sorted.devs[i]]);
1499 if (!ca)
1500 continue;
1501
1502 ja = &ca->journal;
1503
1504 if (sectors > ja->sectors_free &&
1505 sectors <= ca->mi.bucket_size &&
1506 bch2_journal_dev_buckets_available(j, ja,
1507 journal_space_discarded)) {
1508 ja->cur_idx = (ja->cur_idx + 1) % ja->nr;
1509 ja->sectors_free = ca->mi.bucket_size;
1510
1511 /*
1512 * ja->bucket_seq[ja->cur_idx] must always have
1513 * something sensible:
1514 */
1515 ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1516 }
1517 }
1518
1519 __journal_write_alloc(j, w, &devs_sorted,
1520 sectors, &replicas, replicas_want);
1521
1522 if (replicas < replicas_want && target) {
1523 /* Retry from all devices: */
1524 target = 0;
1525 goto retry;
1526 }
1527 done:
1528 rcu_read_unlock();
1529
1530 BUG_ON(bkey_val_u64s(&w->key.k) > BCH_REPLICAS_MAX);
1531
1532 return replicas >= replicas_need ? 0 : -EROFS;
1533 }
1534
journal_buf_realloc(struct journal * j,struct journal_buf * buf)1535 static void journal_buf_realloc(struct journal *j, struct journal_buf *buf)
1536 {
1537 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1538
1539 /* we aren't holding j->lock: */
1540 unsigned new_size = READ_ONCE(j->buf_size_want);
1541 void *new_buf;
1542
1543 if (buf->buf_size >= new_size)
1544 return;
1545
1546 size_t btree_write_buffer_size = new_size / 64;
1547
1548 if (bch2_btree_write_buffer_resize(c, btree_write_buffer_size))
1549 return;
1550
1551 new_buf = kvmalloc(new_size, GFP_NOFS|__GFP_NOWARN);
1552 if (!new_buf)
1553 return;
1554
1555 memcpy(new_buf, buf->data, buf->buf_size);
1556
1557 spin_lock(&j->lock);
1558 swap(buf->data, new_buf);
1559 swap(buf->buf_size, new_size);
1560 spin_unlock(&j->lock);
1561
1562 kvfree(new_buf);
1563 }
1564
journal_last_unwritten_buf(struct journal * j)1565 static inline struct journal_buf *journal_last_unwritten_buf(struct journal *j)
1566 {
1567 return j->buf + (journal_last_unwritten_seq(j) & JOURNAL_BUF_MASK);
1568 }
1569
CLOSURE_CALLBACK(journal_write_done)1570 static CLOSURE_CALLBACK(journal_write_done)
1571 {
1572 closure_type(w, struct journal_buf, io);
1573 struct journal *j = container_of(w, struct journal, buf[w->idx]);
1574 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1575 struct bch_replicas_padded replicas;
1576 union journal_res_state old, new;
1577 u64 seq = le64_to_cpu(w->data->seq);
1578 int err = 0;
1579
1580 bch2_time_stats_update(!JSET_NO_FLUSH(w->data)
1581 ? j->flush_write_time
1582 : j->noflush_write_time, j->write_start_time);
1583
1584 if (!w->devs_written.nr) {
1585 bch_err(c, "unable to write journal to sufficient devices");
1586 err = -EIO;
1587 } else {
1588 bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
1589 w->devs_written);
1590 if (bch2_mark_replicas(c, &replicas.e))
1591 err = -EIO;
1592 }
1593
1594 if (err)
1595 bch2_fatal_error(c);
1596
1597 closure_debug_destroy(cl);
1598
1599 spin_lock(&j->lock);
1600 if (seq >= j->pin.front)
1601 journal_seq_pin(j, seq)->devs = w->devs_written;
1602 if (err && (!j->err_seq || seq < j->err_seq))
1603 j->err_seq = seq;
1604 w->write_done = true;
1605
1606 bool completed = false;
1607
1608 for (seq = journal_last_unwritten_seq(j);
1609 seq <= journal_cur_seq(j);
1610 seq++) {
1611 w = j->buf + (seq & JOURNAL_BUF_MASK);
1612 if (!w->write_done)
1613 break;
1614
1615 if (!j->err_seq && !JSET_NO_FLUSH(w->data)) {
1616 j->flushed_seq_ondisk = seq;
1617 j->last_seq_ondisk = w->last_seq;
1618
1619 bch2_do_discards(c);
1620 closure_wake_up(&c->freelist_wait);
1621 bch2_reset_alloc_cursors(c);
1622 }
1623
1624 j->seq_ondisk = seq;
1625
1626 /*
1627 * Updating last_seq_ondisk may let bch2_journal_reclaim_work() discard
1628 * more buckets:
1629 *
1630 * Must come before signaling write completion, for
1631 * bch2_fs_journal_stop():
1632 */
1633 if (j->watermark != BCH_WATERMARK_stripe)
1634 journal_reclaim_kick(&c->journal);
1635
1636 old.v = atomic64_read(&j->reservations.counter);
1637 do {
1638 new.v = old.v;
1639 BUG_ON(journal_state_count(new, new.unwritten_idx));
1640 BUG_ON(new.unwritten_idx != (seq & JOURNAL_BUF_MASK));
1641
1642 new.unwritten_idx++;
1643 } while (!atomic64_try_cmpxchg(&j->reservations.counter,
1644 &old.v, new.v));
1645
1646 closure_wake_up(&w->wait);
1647 completed = true;
1648 }
1649
1650 if (completed) {
1651 bch2_journal_reclaim_fast(j);
1652 bch2_journal_space_available(j);
1653
1654 track_event_change(&c->times[BCH_TIME_blocked_journal_max_in_flight], false);
1655
1656 journal_wake(j);
1657 }
1658
1659 if (journal_last_unwritten_seq(j) == journal_cur_seq(j) &&
1660 new.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL) {
1661 struct journal_buf *buf = journal_cur_buf(j);
1662 long delta = buf->expires - jiffies;
1663
1664 /*
1665 * We don't close a journal entry to write it while there's
1666 * previous entries still in flight - the current journal entry
1667 * might want to be written now:
1668 */
1669 mod_delayed_work(j->wq, &j->write_work, max(0L, delta));
1670 }
1671
1672 /*
1673 * We don't typically trigger journal writes from her - the next journal
1674 * write will be triggered immediately after the previous one is
1675 * allocated, in bch2_journal_write() - but the journal write error path
1676 * is special:
1677 */
1678 bch2_journal_do_writes(j);
1679 spin_unlock(&j->lock);
1680 }
1681
journal_write_endio(struct bio * bio)1682 static void journal_write_endio(struct bio *bio)
1683 {
1684 struct journal_bio *jbio = container_of(bio, struct journal_bio, bio);
1685 struct bch_dev *ca = jbio->ca;
1686 struct journal *j = &ca->fs->journal;
1687 struct journal_buf *w = j->buf + jbio->buf_idx;
1688
1689 if (bch2_dev_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_write,
1690 "error writing journal entry %llu: %s",
1691 le64_to_cpu(w->data->seq),
1692 bch2_blk_status_to_str(bio->bi_status)) ||
1693 bch2_meta_write_fault("journal")) {
1694 unsigned long flags;
1695
1696 spin_lock_irqsave(&j->err_lock, flags);
1697 bch2_dev_list_drop_dev(&w->devs_written, ca->dev_idx);
1698 spin_unlock_irqrestore(&j->err_lock, flags);
1699 }
1700
1701 closure_put(&w->io);
1702 percpu_ref_put(&ca->io_ref);
1703 }
1704
CLOSURE_CALLBACK(journal_write_submit)1705 static CLOSURE_CALLBACK(journal_write_submit)
1706 {
1707 closure_type(w, struct journal_buf, io);
1708 struct journal *j = container_of(w, struct journal, buf[w->idx]);
1709 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1710 unsigned sectors = vstruct_sectors(w->data, c->block_bits);
1711
1712 extent_for_each_ptr(bkey_i_to_s_extent(&w->key), ptr) {
1713 struct bch_dev *ca = bch2_dev_get_ioref(c, ptr->dev, WRITE);
1714 if (!ca) {
1715 /* XXX: fix this */
1716 bch_err(c, "missing device for journal write\n");
1717 continue;
1718 }
1719
1720 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_journal],
1721 sectors);
1722
1723 struct journal_device *ja = &ca->journal;
1724 struct bio *bio = &ja->bio[w->idx]->bio;
1725 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
1726 bio->bi_iter.bi_sector = ptr->offset;
1727 bio->bi_end_io = journal_write_endio;
1728 bio->bi_private = ca;
1729
1730 BUG_ON(bio->bi_iter.bi_sector == ca->prev_journal_sector);
1731 ca->prev_journal_sector = bio->bi_iter.bi_sector;
1732
1733 if (!JSET_NO_FLUSH(w->data))
1734 bio->bi_opf |= REQ_FUA;
1735 if (!JSET_NO_FLUSH(w->data) && !w->separate_flush)
1736 bio->bi_opf |= REQ_PREFLUSH;
1737
1738 bch2_bio_map(bio, w->data, sectors << 9);
1739
1740 trace_and_count(c, journal_write, bio);
1741 closure_bio_submit(bio, cl);
1742
1743 ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1744 }
1745
1746 continue_at(cl, journal_write_done, j->wq);
1747 }
1748
CLOSURE_CALLBACK(journal_write_preflush)1749 static CLOSURE_CALLBACK(journal_write_preflush)
1750 {
1751 closure_type(w, struct journal_buf, io);
1752 struct journal *j = container_of(w, struct journal, buf[w->idx]);
1753 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1754
1755 if (j->seq_ondisk + 1 != le64_to_cpu(w->data->seq)) {
1756 spin_lock(&j->lock);
1757 if (j->seq_ondisk + 1 != le64_to_cpu(w->data->seq)) {
1758 closure_wait(&j->async_wait, cl);
1759 spin_unlock(&j->lock);
1760 continue_at(cl, journal_write_preflush, j->wq);
1761 return;
1762 }
1763 spin_unlock(&j->lock);
1764 }
1765
1766 if (w->separate_flush) {
1767 for_each_rw_member(c, ca) {
1768 percpu_ref_get(&ca->io_ref);
1769
1770 struct journal_device *ja = &ca->journal;
1771 struct bio *bio = &ja->bio[w->idx]->bio;
1772 bio_reset(bio, ca->disk_sb.bdev,
1773 REQ_OP_WRITE|REQ_SYNC|REQ_META|REQ_PREFLUSH);
1774 bio->bi_end_io = journal_write_endio;
1775 bio->bi_private = ca;
1776 closure_bio_submit(bio, cl);
1777 }
1778
1779 continue_at(cl, journal_write_submit, j->wq);
1780 } else {
1781 /*
1782 * no need to punt to another work item if we're not waiting on
1783 * preflushes
1784 */
1785 journal_write_submit(&cl->work);
1786 }
1787 }
1788
bch2_journal_write_prep(struct journal * j,struct journal_buf * w)1789 static int bch2_journal_write_prep(struct journal *j, struct journal_buf *w)
1790 {
1791 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1792 struct jset_entry *start, *end;
1793 struct jset *jset = w->data;
1794 struct journal_keys_to_wb wb = { NULL };
1795 unsigned sectors, bytes, u64s;
1796 unsigned long btree_roots_have = 0;
1797 bool validate_before_checksum = false;
1798 u64 seq = le64_to_cpu(jset->seq);
1799 int ret;
1800
1801 /*
1802 * Simple compaction, dropping empty jset_entries (from journal
1803 * reservations that weren't fully used) and merging jset_entries that
1804 * can be.
1805 *
1806 * If we wanted to be really fancy here, we could sort all the keys in
1807 * the jset and drop keys that were overwritten - probably not worth it:
1808 */
1809 vstruct_for_each(jset, i) {
1810 unsigned u64s = le16_to_cpu(i->u64s);
1811
1812 /* Empty entry: */
1813 if (!u64s)
1814 continue;
1815
1816 /*
1817 * New btree roots are set by journalling them; when the journal
1818 * entry gets written we have to propagate them to
1819 * c->btree_roots
1820 *
1821 * But, every journal entry we write has to contain all the
1822 * btree roots (at least for now); so after we copy btree roots
1823 * to c->btree_roots we have to get any missing btree roots and
1824 * add them to this journal entry:
1825 */
1826 switch (i->type) {
1827 case BCH_JSET_ENTRY_btree_root:
1828 bch2_journal_entry_to_btree_root(c, i);
1829 __set_bit(i->btree_id, &btree_roots_have);
1830 break;
1831 case BCH_JSET_ENTRY_write_buffer_keys:
1832 EBUG_ON(!w->need_flush_to_write_buffer);
1833
1834 if (!wb.wb)
1835 bch2_journal_keys_to_write_buffer_start(c, &wb, seq);
1836
1837 jset_entry_for_each_key(i, k) {
1838 ret = bch2_journal_key_to_wb(c, &wb, i->btree_id, k);
1839 if (ret) {
1840 bch2_fs_fatal_error(c, "flushing journal keys to btree write buffer: %s",
1841 bch2_err_str(ret));
1842 bch2_journal_keys_to_write_buffer_end(c, &wb);
1843 return ret;
1844 }
1845 }
1846 i->type = BCH_JSET_ENTRY_btree_keys;
1847 break;
1848 }
1849 }
1850
1851 if (wb.wb) {
1852 ret = bch2_journal_keys_to_write_buffer_end(c, &wb);
1853 if (ret) {
1854 bch2_fs_fatal_error(c, "error flushing journal keys to btree write buffer: %s",
1855 bch2_err_str(ret));
1856 return ret;
1857 }
1858 }
1859
1860 spin_lock(&c->journal.lock);
1861 w->need_flush_to_write_buffer = false;
1862 spin_unlock(&c->journal.lock);
1863
1864 start = end = vstruct_last(jset);
1865
1866 end = bch2_btree_roots_to_journal_entries(c, end, btree_roots_have);
1867
1868 struct jset_entry_datetime *d =
1869 container_of(jset_entry_init(&end, sizeof(*d)), struct jset_entry_datetime, entry);
1870 d->entry.type = BCH_JSET_ENTRY_datetime;
1871 d->seconds = cpu_to_le64(ktime_get_real_seconds());
1872
1873 bch2_journal_super_entries_add_common(c, &end, seq);
1874 u64s = (u64 *) end - (u64 *) start;
1875
1876 WARN_ON(u64s > j->entry_u64s_reserved);
1877
1878 le32_add_cpu(&jset->u64s, u64s);
1879
1880 sectors = vstruct_sectors(jset, c->block_bits);
1881 bytes = vstruct_bytes(jset);
1882
1883 if (sectors > w->sectors) {
1884 bch2_fs_fatal_error(c, ": journal write overran available space, %zu > %u (extra %u reserved %u/%u)",
1885 vstruct_bytes(jset), w->sectors << 9,
1886 u64s, w->u64s_reserved, j->entry_u64s_reserved);
1887 return -EINVAL;
1888 }
1889
1890 jset->magic = cpu_to_le64(jset_magic(c));
1891 jset->version = cpu_to_le32(c->sb.version);
1892
1893 SET_JSET_BIG_ENDIAN(jset, CPU_BIG_ENDIAN);
1894 SET_JSET_CSUM_TYPE(jset, bch2_meta_checksum_type(c));
1895
1896 if (!JSET_NO_FLUSH(jset) && journal_entry_empty(jset))
1897 j->last_empty_seq = seq;
1898
1899 if (bch2_csum_type_is_encryption(JSET_CSUM_TYPE(jset)))
1900 validate_before_checksum = true;
1901
1902 if (le32_to_cpu(jset->version) < bcachefs_metadata_version_current)
1903 validate_before_checksum = true;
1904
1905 if (validate_before_checksum &&
1906 (ret = jset_validate(c, NULL, jset, 0, WRITE)))
1907 return ret;
1908
1909 ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset),
1910 jset->encrypted_start,
1911 vstruct_end(jset) - (void *) jset->encrypted_start);
1912 if (bch2_fs_fatal_err_on(ret, c, "decrypting journal entry: %s", bch2_err_str(ret)))
1913 return ret;
1914
1915 jset->csum = csum_vstruct(c, JSET_CSUM_TYPE(jset),
1916 journal_nonce(jset), jset);
1917
1918 if (!validate_before_checksum &&
1919 (ret = jset_validate(c, NULL, jset, 0, WRITE)))
1920 return ret;
1921
1922 memset((void *) jset + bytes, 0, (sectors << 9) - bytes);
1923 return 0;
1924 }
1925
bch2_journal_write_pick_flush(struct journal * j,struct journal_buf * w)1926 static int bch2_journal_write_pick_flush(struct journal *j, struct journal_buf *w)
1927 {
1928 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1929 int error = bch2_journal_error(j);
1930
1931 /*
1932 * If the journal is in an error state - we did an emergency shutdown -
1933 * we prefer to continue doing journal writes. We just mark them as
1934 * noflush so they'll never be used, but they'll still be visible by the
1935 * list_journal tool - this helps in debugging.
1936 *
1937 * There's a caveat: the first journal write after marking the
1938 * superblock dirty must always be a flush write, because on startup
1939 * from a clean shutdown we didn't necessarily read the journal and the
1940 * new journal write might overwrite whatever was in the journal
1941 * previously - we can't leave the journal without any flush writes in
1942 * it.
1943 *
1944 * So if we're in an error state, and we're still starting up, we don't
1945 * write anything at all.
1946 */
1947 if (error && test_bit(JOURNAL_need_flush_write, &j->flags))
1948 return -EIO;
1949
1950 if (error ||
1951 w->noflush ||
1952 (!w->must_flush &&
1953 (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) &&
1954 test_bit(JOURNAL_may_skip_flush, &j->flags))) {
1955 w->noflush = true;
1956 SET_JSET_NO_FLUSH(w->data, true);
1957 w->data->last_seq = 0;
1958 w->last_seq = 0;
1959
1960 j->nr_noflush_writes++;
1961 } else {
1962 w->must_flush = true;
1963 j->last_flush_write = jiffies;
1964 j->nr_flush_writes++;
1965 clear_bit(JOURNAL_need_flush_write, &j->flags);
1966 }
1967
1968 return 0;
1969 }
1970
CLOSURE_CALLBACK(bch2_journal_write)1971 CLOSURE_CALLBACK(bch2_journal_write)
1972 {
1973 closure_type(w, struct journal_buf, io);
1974 struct journal *j = container_of(w, struct journal, buf[w->idx]);
1975 struct bch_fs *c = container_of(j, struct bch_fs, journal);
1976 struct bch_replicas_padded replicas;
1977 unsigned nr_rw_members = 0;
1978 int ret;
1979
1980 for_each_rw_member(c, ca)
1981 nr_rw_members++;
1982
1983 BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
1984 BUG_ON(!w->write_started);
1985 BUG_ON(w->write_allocated);
1986 BUG_ON(w->write_done);
1987
1988 j->write_start_time = local_clock();
1989
1990 spin_lock(&j->lock);
1991 if (nr_rw_members > 1)
1992 w->separate_flush = true;
1993
1994 ret = bch2_journal_write_pick_flush(j, w);
1995 spin_unlock(&j->lock);
1996 if (ret)
1997 goto err;
1998
1999 mutex_lock(&j->buf_lock);
2000 journal_buf_realloc(j, w);
2001
2002 ret = bch2_journal_write_prep(j, w);
2003 mutex_unlock(&j->buf_lock);
2004 if (ret)
2005 goto err;
2006
2007 j->entry_bytes_written += vstruct_bytes(w->data);
2008
2009 while (1) {
2010 spin_lock(&j->lock);
2011 ret = journal_write_alloc(j, w);
2012 if (!ret || !j->can_discard)
2013 break;
2014
2015 spin_unlock(&j->lock);
2016 bch2_journal_do_discards(j);
2017 }
2018
2019 if (ret) {
2020 struct printbuf buf = PRINTBUF;
2021 buf.atomic++;
2022
2023 prt_printf(&buf, bch2_fmt(c, "Unable to allocate journal write at seq %llu: %s"),
2024 le64_to_cpu(w->data->seq),
2025 bch2_err_str(ret));
2026 __bch2_journal_debug_to_text(&buf, j);
2027 spin_unlock(&j->lock);
2028 bch2_print_string_as_lines(KERN_ERR, buf.buf);
2029 printbuf_exit(&buf);
2030 goto err;
2031 }
2032
2033 /*
2034 * write is allocated, no longer need to account for it in
2035 * bch2_journal_space_available():
2036 */
2037 w->sectors = 0;
2038 w->write_allocated = true;
2039
2040 /*
2041 * journal entry has been compacted and allocated, recalculate space
2042 * available:
2043 */
2044 bch2_journal_space_available(j);
2045 bch2_journal_do_writes(j);
2046 spin_unlock(&j->lock);
2047
2048 w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key));
2049
2050 if (c->opts.nochanges)
2051 goto no_io;
2052
2053 /*
2054 * Mark journal replicas before we submit the write to guarantee
2055 * recovery will find the journal entries after a crash.
2056 */
2057 bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
2058 w->devs_written);
2059 ret = bch2_mark_replicas(c, &replicas.e);
2060 if (ret)
2061 goto err;
2062
2063 if (!JSET_NO_FLUSH(w->data))
2064 continue_at(cl, journal_write_preflush, j->wq);
2065 else
2066 continue_at(cl, journal_write_submit, j->wq);
2067 return;
2068 no_io:
2069 continue_at(cl, journal_write_done, j->wq);
2070 return;
2071 err:
2072 bch2_fatal_error(c);
2073 continue_at(cl, journal_write_done, j->wq);
2074 }
2075