Commit | Line | Data |
---|---|---|
1c6fdbd8 KO |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | #include "bcachefs.h" | |
7b3f84ea | 3 | #include "alloc_foreground.h" |
39fb2983 | 4 | #include "btree_io.h" |
00b8ccf7 | 5 | #include "btree_update_interior.h" |
1c6fdbd8 KO |
6 | #include "buckets.h" |
7 | #include "checksum.h" | |
d042b040 | 8 | #include "disk_groups.h" |
1c6fdbd8 | 9 | #include "error.h" |
63b214e7 | 10 | #include "io.h" |
1c6fdbd8 KO |
11 | #include "journal.h" |
12 | #include "journal_io.h" | |
13 | #include "journal_reclaim.h" | |
adbcada4 | 14 | #include "journal_seq_blacklist.h" |
1c6fdbd8 KO |
15 | #include "replicas.h" |
16 | #include "trace.h" | |
17 | ||
adbcada4 KO |
18 | static void __journal_replay_free(struct journal_replay *i) |
19 | { | |
20 | list_del(&i->list); | |
21 | kvpfree(i, offsetof(struct journal_replay, j) + | |
22 | vstruct_bytes(&i->j)); | |
23 | ||
24 | } | |
25 | ||
26 | static void journal_replay_free(struct bch_fs *c, struct journal_replay *i) | |
27 | { | |
28 | i->ignore = true; | |
29 | ||
30 | if (!c->opts.read_entire_journal) | |
31 | __journal_replay_free(i); | |
32 | } | |
33 | ||
1c6fdbd8 KO |
34 | struct journal_list { |
35 | struct closure cl; | |
36 | struct mutex lock; | |
37 | struct list_head *head; | |
38 | int ret; | |
39 | }; | |
40 | ||
41 | #define JOURNAL_ENTRY_ADD_OK 0 | |
42 | #define JOURNAL_ENTRY_ADD_OUT_OF_RANGE 5 | |
43 | ||
44 | /* | |
45 | * Given a journal entry we just read, add it to the list of journal entries to | |
46 | * be replayed: | |
47 | */ | |
48 | static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca, | |
e4c3f386 | 49 | struct bch_extent_ptr entry_ptr, |
ca73852a KO |
50 | struct journal_list *jlist, struct jset *j, |
51 | bool bad) | |
1c6fdbd8 | 52 | { |
e4c3f386 KO |
53 | struct journal_replay *i, *pos, *dup = NULL; |
54 | struct bch_extent_ptr *ptr; | |
1c6fdbd8 KO |
55 | struct list_head *where; |
56 | size_t bytes = vstruct_bytes(j); | |
adbcada4 | 57 | u64 last_seq = 0; |
e4c3f386 | 58 | int ret = JOURNAL_ENTRY_ADD_OK; |
1c6fdbd8 | 59 | |
adbcada4 KO |
60 | list_for_each_entry_reverse(i, jlist->head, list) { |
61 | if (!JSET_NO_FLUSH(&i->j)) { | |
62 | last_seq = le64_to_cpu(i->j.last_seq); | |
63 | break; | |
7fffc85b | 64 | } |
adbcada4 | 65 | } |
1c6fdbd8 | 66 | |
adbcada4 KO |
67 | /* Is this entry older than the range we need? */ |
68 | if (!c->opts.read_entire_journal && | |
69 | le64_to_cpu(j->seq) < last_seq) { | |
70 | ret = JOURNAL_ENTRY_ADD_OUT_OF_RANGE; | |
71 | goto out; | |
72 | } | |
73 | ||
74 | /* Drop entries we don't need anymore */ | |
75 | if (!JSET_NO_FLUSH(j)) { | |
7fffc85b KO |
76 | list_for_each_entry_safe(i, pos, jlist->head, list) { |
77 | if (le64_to_cpu(i->j.seq) >= le64_to_cpu(j->last_seq)) | |
78 | break; | |
adbcada4 | 79 | journal_replay_free(c, i); |
7fffc85b | 80 | } |
1c6fdbd8 KO |
81 | } |
82 | ||
83 | list_for_each_entry_reverse(i, jlist->head, list) { | |
ca73852a KO |
84 | if (le64_to_cpu(j->seq) > le64_to_cpu(i->j.seq)) { |
85 | where = &i->list; | |
86 | goto add; | |
87 | } | |
88 | } | |
89 | ||
90 | where = jlist->head; | |
91 | add: | |
e4c3f386 | 92 | dup = where->next != jlist->head |
ca73852a KO |
93 | ? container_of(where->next, struct journal_replay, list) |
94 | : NULL; | |
95 | ||
e4c3f386 KO |
96 | if (dup && le64_to_cpu(j->seq) != le64_to_cpu(dup->j.seq)) |
97 | dup = NULL; | |
98 | ||
ca73852a KO |
99 | /* |
100 | * Duplicate journal entries? If so we want the one that didn't have a | |
101 | * checksum error: | |
102 | */ | |
e4c3f386 KO |
103 | if (dup) { |
104 | if (dup->bad) { | |
105 | /* we'll replace @dup: */ | |
ca73852a | 106 | } else if (bad) { |
e4c3f386 | 107 | i = dup; |
ca73852a KO |
108 | goto found; |
109 | } else { | |
e4c3f386 KO |
110 | fsck_err_on(bytes != vstruct_bytes(&dup->j) || |
111 | memcmp(j, &dup->j, bytes), c, | |
1c6fdbd8 KO |
112 | "found duplicate but non identical journal entries (seq %llu)", |
113 | le64_to_cpu(j->seq)); | |
e4c3f386 | 114 | i = dup; |
1c6fdbd8 KO |
115 | goto found; |
116 | } | |
1c6fdbd8 KO |
117 | } |
118 | ||
1c6fdbd8 KO |
119 | i = kvpmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL); |
120 | if (!i) { | |
121 | ret = -ENOMEM; | |
122 | goto out; | |
123 | } | |
124 | ||
e4c3f386 KO |
125 | i->nr_ptrs = 0; |
126 | i->bad = bad; | |
127 | i->ignore = false; | |
1c6fdbd8 | 128 | unsafe_memcpy(&i->j, j, bytes, "embedded variable length struct"); |
e4c3f386 KO |
129 | |
130 | if (dup) { | |
131 | i->nr_ptrs = dup->nr_ptrs; | |
132 | memcpy(i->ptrs, dup->ptrs, sizeof(dup->ptrs)); | |
133 | __journal_replay_free(dup); | |
134 | } | |
135 | ||
136 | list_add(&i->list, where); | |
1c6fdbd8 | 137 | found: |
e4c3f386 KO |
138 | for (ptr = i->ptrs; ptr < i->ptrs + i->nr_ptrs; ptr++) { |
139 | if (ptr->dev == ca->dev_idx) { | |
140 | bch_err(c, "duplicate journal entry %llu on same device", | |
141 | le64_to_cpu(i->j.seq)); | |
142 | goto out; | |
143 | } | |
144 | } | |
145 | ||
146 | if (i->nr_ptrs >= ARRAY_SIZE(i->ptrs)) { | |
147 | bch_err(c, "found too many copies of journal entry %llu", | |
148 | le64_to_cpu(i->j.seq)); | |
149 | goto out; | |
150 | } | |
151 | ||
152 | i->ptrs[i->nr_ptrs++] = entry_ptr; | |
1c6fdbd8 KO |
153 | out: |
154 | fsck_err: | |
155 | return ret; | |
156 | } | |
157 | ||
158 | static struct nonce journal_nonce(const struct jset *jset) | |
159 | { | |
160 | return (struct nonce) {{ | |
161 | [0] = 0, | |
162 | [1] = ((__le32 *) &jset->seq)[0], | |
163 | [2] = ((__le32 *) &jset->seq)[1], | |
164 | [3] = BCH_NONCE_JOURNAL, | |
165 | }}; | |
166 | } | |
167 | ||
168 | /* this fills in a range with empty jset_entries: */ | |
169 | static void journal_entry_null_range(void *start, void *end) | |
170 | { | |
171 | struct jset_entry *entry; | |
172 | ||
173 | for (entry = start; entry != end; entry = vstruct_next(entry)) | |
174 | memset(entry, 0, sizeof(*entry)); | |
175 | } | |
176 | ||
177 | #define JOURNAL_ENTRY_REREAD 5 | |
178 | #define JOURNAL_ENTRY_NONE 6 | |
179 | #define JOURNAL_ENTRY_BAD 7 | |
180 | ||
181 | #define journal_entry_err(c, msg, ...) \ | |
182 | ({ \ | |
183 | switch (write) { \ | |
184 | case READ: \ | |
185 | mustfix_fsck_err(c, msg, ##__VA_ARGS__); \ | |
186 | break; \ | |
187 | case WRITE: \ | |
188 | bch_err(c, "corrupt metadata before write:\n" \ | |
189 | msg, ##__VA_ARGS__); \ | |
190 | if (bch2_fs_inconsistent(c)) { \ | |
191 | ret = BCH_FSCK_ERRORS_NOT_FIXED; \ | |
192 | goto fsck_err; \ | |
193 | } \ | |
194 | break; \ | |
195 | } \ | |
196 | true; \ | |
197 | }) | |
198 | ||
199 | #define journal_entry_err_on(cond, c, msg, ...) \ | |
200 | ((cond) ? journal_entry_err(c, msg, ##__VA_ARGS__) : false) | |
201 | ||
4d54337c KO |
202 | #define FSCK_DELETED_KEY 5 |
203 | ||
7d6f07ed | 204 | static int journal_validate_key(struct bch_fs *c, const char *where, |
1c6fdbd8 | 205 | struct jset_entry *entry, |
39fb2983 | 206 | unsigned level, enum btree_id btree_id, |
7d6f07ed KO |
207 | struct bkey_i *k, const char *type, |
208 | unsigned version, int big_endian, int write) | |
1c6fdbd8 KO |
209 | { |
210 | void *next = vstruct_next(entry); | |
211 | const char *invalid; | |
1c6fdbd8 KO |
212 | int ret = 0; |
213 | ||
214 | if (journal_entry_err_on(!k->k.u64s, c, | |
7d6f07ed KO |
215 | "invalid %s in %s entry offset %zi/%u: k->u64s 0", |
216 | type, where, | |
4d54337c KO |
217 | (u64 *) k - entry->_data, |
218 | le16_to_cpu(entry->u64s))) { | |
1c6fdbd8 KO |
219 | entry->u64s = cpu_to_le16((u64 *) k - entry->_data); |
220 | journal_entry_null_range(vstruct_next(entry), next); | |
4d54337c | 221 | return FSCK_DELETED_KEY; |
1c6fdbd8 KO |
222 | } |
223 | ||
224 | if (journal_entry_err_on((void *) bkey_next(k) > | |
225 | (void *) vstruct_next(entry), c, | |
7d6f07ed KO |
226 | "invalid %s in %s entry offset %zi/%u: extends past end of journal entry", |
227 | type, where, | |
4d54337c KO |
228 | (u64 *) k - entry->_data, |
229 | le16_to_cpu(entry->u64s))) { | |
1c6fdbd8 KO |
230 | entry->u64s = cpu_to_le16((u64 *) k - entry->_data); |
231 | journal_entry_null_range(vstruct_next(entry), next); | |
4d54337c | 232 | return FSCK_DELETED_KEY; |
1c6fdbd8 KO |
233 | } |
234 | ||
235 | if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT, c, | |
7d6f07ed KO |
236 | "invalid %s in %s entry offset %zi/%u: bad format %u", |
237 | type, where, | |
4d54337c KO |
238 | (u64 *) k - entry->_data, |
239 | le16_to_cpu(entry->u64s), | |
ed0d631f | 240 | k->k.format)) { |
4d54337c | 241 | le16_add_cpu(&entry->u64s, -((u16) k->k.u64s)); |
1c6fdbd8 KO |
242 | memmove(k, bkey_next(k), next - (void *) bkey_next(k)); |
243 | journal_entry_null_range(vstruct_next(entry), next); | |
4d54337c | 244 | return FSCK_DELETED_KEY; |
1c6fdbd8 KO |
245 | } |
246 | ||
39fb2983 | 247 | if (!write) |
7d6f07ed KO |
248 | bch2_bkey_compat(level, btree_id, version, big_endian, |
249 | write, NULL, bkey_to_packed(k)); | |
26609b61 | 250 | |
39fb2983 KO |
251 | invalid = bch2_bkey_invalid(c, bkey_i_to_s_c(k), |
252 | __btree_node_type(level, btree_id)); | |
1c6fdbd8 | 253 | if (invalid) { |
319f9ac3 KO |
254 | char buf[160]; |
255 | ||
26609b61 | 256 | bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(k)); |
7d6f07ed KO |
257 | mustfix_fsck_err(c, "invalid %s in %s entry offset %zi/%u: %s\n%s", |
258 | type, where, | |
4d54337c KO |
259 | (u64 *) k - entry->_data, |
260 | le16_to_cpu(entry->u64s), | |
ed0d631f | 261 | invalid, buf); |
1c6fdbd8 | 262 | |
4d54337c | 263 | le16_add_cpu(&entry->u64s, -((u16) k->k.u64s)); |
1c6fdbd8 KO |
264 | memmove(k, bkey_next(k), next - (void *) bkey_next(k)); |
265 | journal_entry_null_range(vstruct_next(entry), next); | |
4d54337c | 266 | return FSCK_DELETED_KEY; |
1c6fdbd8 | 267 | } |
26609b61 | 268 | |
39fb2983 | 269 | if (write) |
7d6f07ed KO |
270 | bch2_bkey_compat(level, btree_id, version, big_endian, |
271 | write, NULL, bkey_to_packed(k)); | |
1c6fdbd8 KO |
272 | fsck_err: |
273 | return ret; | |
274 | } | |
275 | ||
276 | static int journal_entry_validate_btree_keys(struct bch_fs *c, | |
7d6f07ed | 277 | const char *where, |
1c6fdbd8 | 278 | struct jset_entry *entry, |
7d6f07ed | 279 | unsigned version, int big_endian, int write) |
1c6fdbd8 | 280 | { |
4d54337c | 281 | struct bkey_i *k = entry->start; |
1c6fdbd8 | 282 | |
4d54337c | 283 | while (k != vstruct_last(entry)) { |
7d6f07ed | 284 | int ret = journal_validate_key(c, where, entry, |
39fb2983 KO |
285 | entry->level, |
286 | entry->btree_id, | |
7d6f07ed | 287 | k, "key", version, big_endian, write); |
4d54337c KO |
288 | if (ret == FSCK_DELETED_KEY) |
289 | continue; | |
290 | ||
291 | k = bkey_next(k); | |
1c6fdbd8 KO |
292 | } |
293 | ||
294 | return 0; | |
295 | } | |
296 | ||
297 | static int journal_entry_validate_btree_root(struct bch_fs *c, | |
7d6f07ed | 298 | const char *where, |
1c6fdbd8 | 299 | struct jset_entry *entry, |
7d6f07ed | 300 | unsigned version, int big_endian, int write) |
1c6fdbd8 KO |
301 | { |
302 | struct bkey_i *k = entry->start; | |
303 | int ret = 0; | |
304 | ||
305 | if (journal_entry_err_on(!entry->u64s || | |
306 | le16_to_cpu(entry->u64s) != k->k.u64s, c, | |
307 | "invalid btree root journal entry: wrong number of keys")) { | |
308 | void *next = vstruct_next(entry); | |
309 | /* | |
310 | * we don't want to null out this jset_entry, | |
311 | * just the contents, so that later we can tell | |
312 | * we were _supposed_ to have a btree root | |
313 | */ | |
314 | entry->u64s = 0; | |
315 | journal_entry_null_range(vstruct_next(entry), next); | |
316 | return 0; | |
317 | } | |
318 | ||
7d6f07ed KO |
319 | return journal_validate_key(c, where, entry, 1, entry->btree_id, k, |
320 | "btree root", version, big_endian, write); | |
1c6fdbd8 KO |
321 | fsck_err: |
322 | return ret; | |
323 | } | |
324 | ||
325 | static int journal_entry_validate_prio_ptrs(struct bch_fs *c, | |
7d6f07ed | 326 | const char *where, |
1c6fdbd8 | 327 | struct jset_entry *entry, |
7d6f07ed | 328 | unsigned version, int big_endian, int write) |
1c6fdbd8 KO |
329 | { |
330 | /* obsolete, don't care: */ | |
331 | return 0; | |
332 | } | |
333 | ||
334 | static int journal_entry_validate_blacklist(struct bch_fs *c, | |
7d6f07ed | 335 | const char *where, |
1c6fdbd8 | 336 | struct jset_entry *entry, |
7d6f07ed | 337 | unsigned version, int big_endian, int write) |
1c6fdbd8 KO |
338 | { |
339 | int ret = 0; | |
340 | ||
341 | if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 1, c, | |
342 | "invalid journal seq blacklist entry: bad size")) { | |
343 | journal_entry_null_range(entry, vstruct_next(entry)); | |
344 | } | |
345 | fsck_err: | |
346 | return ret; | |
347 | } | |
348 | ||
349 | static int journal_entry_validate_blacklist_v2(struct bch_fs *c, | |
7d6f07ed | 350 | const char *where, |
1c6fdbd8 | 351 | struct jset_entry *entry, |
7d6f07ed | 352 | unsigned version, int big_endian, int write) |
1c6fdbd8 KO |
353 | { |
354 | struct jset_entry_blacklist_v2 *bl_entry; | |
355 | int ret = 0; | |
356 | ||
357 | if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 2, c, | |
358 | "invalid journal seq blacklist entry: bad size")) { | |
359 | journal_entry_null_range(entry, vstruct_next(entry)); | |
2c5af169 | 360 | goto out; |
1c6fdbd8 KO |
361 | } |
362 | ||
363 | bl_entry = container_of(entry, struct jset_entry_blacklist_v2, entry); | |
364 | ||
365 | if (journal_entry_err_on(le64_to_cpu(bl_entry->start) > | |
366 | le64_to_cpu(bl_entry->end), c, | |
367 | "invalid journal seq blacklist entry: start > end")) { | |
368 | journal_entry_null_range(entry, vstruct_next(entry)); | |
369 | } | |
2c5af169 KO |
370 | out: |
371 | fsck_err: | |
372 | return ret; | |
373 | } | |
374 | ||
375 | static int journal_entry_validate_usage(struct bch_fs *c, | |
7d6f07ed | 376 | const char *where, |
2c5af169 | 377 | struct jset_entry *entry, |
7d6f07ed | 378 | unsigned version, int big_endian, int write) |
2c5af169 KO |
379 | { |
380 | struct jset_entry_usage *u = | |
381 | container_of(entry, struct jset_entry_usage, entry); | |
382 | unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64); | |
383 | int ret = 0; | |
384 | ||
3577df5f KO |
385 | if (journal_entry_err_on(bytes < sizeof(*u), |
386 | c, | |
387 | "invalid journal entry usage: bad size")) { | |
388 | journal_entry_null_range(entry, vstruct_next(entry)); | |
389 | return ret; | |
390 | } | |
391 | ||
392 | fsck_err: | |
393 | return ret; | |
394 | } | |
395 | ||
396 | static int journal_entry_validate_data_usage(struct bch_fs *c, | |
7d6f07ed | 397 | const char *where, |
3577df5f | 398 | struct jset_entry *entry, |
7d6f07ed | 399 | unsigned version, int big_endian, int write) |
3577df5f KO |
400 | { |
401 | struct jset_entry_data_usage *u = | |
402 | container_of(entry, struct jset_entry_data_usage, entry); | |
403 | unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64); | |
404 | int ret = 0; | |
405 | ||
2c5af169 KO |
406 | if (journal_entry_err_on(bytes < sizeof(*u) || |
407 | bytes < sizeof(*u) + u->r.nr_devs, | |
408 | c, | |
409 | "invalid journal entry usage: bad size")) { | |
410 | journal_entry_null_range(entry, vstruct_next(entry)); | |
411 | return ret; | |
412 | } | |
1c6fdbd8 KO |
413 | |
414 | fsck_err: | |
415 | return ret; | |
416 | } | |
417 | ||
2abe5420 | 418 | static int journal_entry_validate_clock(struct bch_fs *c, |
7d6f07ed | 419 | const char *where, |
2abe5420 | 420 | struct jset_entry *entry, |
7d6f07ed | 421 | unsigned version, int big_endian, int write) |
2abe5420 KO |
422 | { |
423 | struct jset_entry_clock *clock = | |
424 | container_of(entry, struct jset_entry_clock, entry); | |
425 | unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64); | |
426 | int ret = 0; | |
427 | ||
428 | if (journal_entry_err_on(bytes != sizeof(*clock), | |
429 | c, "invalid journal entry clock: bad size")) { | |
430 | journal_entry_null_range(entry, vstruct_next(entry)); | |
431 | return ret; | |
432 | } | |
433 | ||
434 | if (journal_entry_err_on(clock->rw > 1, | |
435 | c, "invalid journal entry clock: bad rw")) { | |
436 | journal_entry_null_range(entry, vstruct_next(entry)); | |
437 | return ret; | |
438 | } | |
439 | ||
440 | fsck_err: | |
441 | return ret; | |
442 | } | |
443 | ||
180fb49d | 444 | static int journal_entry_validate_dev_usage(struct bch_fs *c, |
7d6f07ed | 445 | const char *where, |
180fb49d | 446 | struct jset_entry *entry, |
7d6f07ed | 447 | unsigned version, int big_endian, int write) |
180fb49d KO |
448 | { |
449 | struct jset_entry_dev_usage *u = | |
450 | container_of(entry, struct jset_entry_dev_usage, entry); | |
451 | unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64); | |
45c2e33f | 452 | unsigned expected = sizeof(*u); |
180fb49d KO |
453 | unsigned dev; |
454 | int ret = 0; | |
455 | ||
456 | if (journal_entry_err_on(bytes < expected, | |
457 | c, "invalid journal entry dev usage: bad size (%u < %u)", | |
458 | bytes, expected)) { | |
459 | journal_entry_null_range(entry, vstruct_next(entry)); | |
460 | return ret; | |
461 | } | |
462 | ||
463 | dev = le32_to_cpu(u->dev); | |
464 | ||
465 | if (journal_entry_err_on(!bch2_dev_exists2(c, dev), | |
466 | c, "invalid journal entry dev usage: bad dev")) { | |
467 | journal_entry_null_range(entry, vstruct_next(entry)); | |
468 | return ret; | |
469 | } | |
470 | ||
471 | if (journal_entry_err_on(u->pad, | |
472 | c, "invalid journal entry dev usage: bad pad")) { | |
473 | journal_entry_null_range(entry, vstruct_next(entry)); | |
474 | return ret; | |
475 | } | |
476 | ||
477 | fsck_err: | |
478 | return ret; | |
479 | } | |
480 | ||
fb64f3fd KO |
481 | static int journal_entry_validate_log(struct bch_fs *c, |
482 | const char *where, | |
483 | struct jset_entry *entry, | |
484 | unsigned version, int big_endian, int write) | |
485 | { | |
486 | return 0; | |
487 | } | |
488 | ||
1c6fdbd8 | 489 | struct jset_entry_ops { |
7d6f07ed KO |
490 | int (*validate)(struct bch_fs *, const char *, |
491 | struct jset_entry *, unsigned, int, int); | |
1c6fdbd8 KO |
492 | }; |
493 | ||
494 | static const struct jset_entry_ops bch2_jset_entry_ops[] = { | |
495 | #define x(f, nr) \ | |
496 | [BCH_JSET_ENTRY_##f] = (struct jset_entry_ops) { \ | |
497 | .validate = journal_entry_validate_##f, \ | |
498 | }, | |
499 | BCH_JSET_ENTRY_TYPES() | |
500 | #undef x | |
501 | }; | |
502 | ||
7d6f07ed KO |
503 | int bch2_journal_entry_validate(struct bch_fs *c, const char *where, |
504 | struct jset_entry *entry, | |
505 | unsigned version, int big_endian, int write) | |
1c6fdbd8 | 506 | { |
2c5af169 | 507 | return entry->type < BCH_JSET_ENTRY_NR |
7d6f07ed KO |
508 | ? bch2_jset_entry_ops[entry->type].validate(c, where, entry, |
509 | version, big_endian, write) | |
2c5af169 | 510 | : 0; |
1c6fdbd8 KO |
511 | } |
512 | ||
513 | static int jset_validate_entries(struct bch_fs *c, struct jset *jset, | |
514 | int write) | |
515 | { | |
7d6f07ed | 516 | char buf[100]; |
1c6fdbd8 KO |
517 | struct jset_entry *entry; |
518 | int ret = 0; | |
519 | ||
520 | vstruct_for_each(jset, entry) { | |
7d6f07ed KO |
521 | scnprintf(buf, sizeof(buf), "jset %llu entry offset %zi/%u", |
522 | le64_to_cpu(jset->seq), | |
523 | (u64 *) entry - jset->_data, | |
524 | le32_to_cpu(jset->u64s)); | |
525 | ||
1c6fdbd8 KO |
526 | if (journal_entry_err_on(vstruct_next(entry) > |
527 | vstruct_last(jset), c, | |
528 | "journal entry extends past end of jset")) { | |
529 | jset->u64s = cpu_to_le32((u64 *) entry - jset->_data); | |
530 | break; | |
531 | } | |
532 | ||
7d6f07ed KO |
533 | ret = bch2_journal_entry_validate(c, buf, entry, |
534 | le32_to_cpu(jset->version), | |
535 | JSET_BIG_ENDIAN(jset), write); | |
1c6fdbd8 KO |
536 | if (ret) |
537 | break; | |
538 | } | |
539 | fsck_err: | |
540 | return ret; | |
541 | } | |
542 | ||
543 | static int jset_validate(struct bch_fs *c, | |
ca73852a | 544 | struct bch_dev *ca, |
1c6fdbd8 KO |
545 | struct jset *jset, u64 sector, |
546 | unsigned bucket_sectors_left, | |
547 | unsigned sectors_read, | |
548 | int write) | |
549 | { | |
550 | size_t bytes = vstruct_bytes(jset); | |
551 | struct bch_csum csum; | |
26609b61 | 552 | unsigned version; |
1c6fdbd8 KO |
553 | int ret = 0; |
554 | ||
555 | if (le64_to_cpu(jset->magic) != jset_magic(c)) | |
556 | return JOURNAL_ENTRY_NONE; | |
557 | ||
26609b61 | 558 | version = le32_to_cpu(jset->version); |
ca73852a KO |
559 | if (journal_entry_err_on((version != BCH_JSET_VERSION_OLD && |
560 | version < bcachefs_metadata_version_min) || | |
561 | version >= bcachefs_metadata_version_max, c, | |
562 | "%s sector %llu seq %llu: unknown journal entry version %u", | |
ed9d58a2 KO |
563 | ca ? ca->name : c->name, |
564 | sector, le64_to_cpu(jset->seq), | |
ca73852a | 565 | version)) { |
35ef6df5 KO |
566 | /* don't try to continue: */ |
567 | return EINVAL; | |
1c6fdbd8 KO |
568 | } |
569 | ||
35ef6df5 KO |
570 | if (bytes > (sectors_read << 9) && |
571 | sectors_read < bucket_sectors_left) | |
572 | return JOURNAL_ENTRY_REREAD; | |
573 | ||
1c6fdbd8 | 574 | if (journal_entry_err_on(bytes > bucket_sectors_left << 9, c, |
ca73852a | 575 | "%s sector %llu seq %llu: journal entry too big (%zu bytes)", |
ed9d58a2 KO |
576 | ca ? ca->name : c->name, |
577 | sector, le64_to_cpu(jset->seq), bytes)) { | |
35ef6df5 KO |
578 | ret = JOURNAL_ENTRY_BAD; |
579 | le32_add_cpu(&jset->u64s, | |
580 | -((bytes - (bucket_sectors_left << 9)) / 8)); | |
1c6fdbd8 KO |
581 | } |
582 | ||
ed9d58a2 | 583 | if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)), c, |
ca73852a | 584 | "%s sector %llu seq %llu: journal entry with unknown csum type %llu", |
ed9d58a2 KO |
585 | ca ? ca->name : c->name, |
586 | sector, le64_to_cpu(jset->seq), | |
35ef6df5 KO |
587 | JSET_CSUM_TYPE(jset))) { |
588 | ret = JOURNAL_ENTRY_BAD; | |
ed9d58a2 | 589 | goto csum_done; |
35ef6df5 | 590 | } |
1c6fdbd8 | 591 | |
ed9d58a2 KO |
592 | if (write) |
593 | goto csum_done; | |
594 | ||
1c6fdbd8 KO |
595 | csum = csum_vstruct(c, JSET_CSUM_TYPE(jset), journal_nonce(jset), jset); |
596 | if (journal_entry_err_on(bch2_crc_cmp(csum, jset->csum), c, | |
ca73852a | 597 | "%s sector %llu seq %llu: journal checksum bad", |
ed9d58a2 KO |
598 | ca ? ca->name : c->name, |
599 | sector, le64_to_cpu(jset->seq))) | |
35ef6df5 | 600 | ret = JOURNAL_ENTRY_BAD; |
1c6fdbd8 KO |
601 | |
602 | bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset), | |
603 | jset->encrypted_start, | |
604 | vstruct_end(jset) - (void *) jset->encrypted_start); | |
ed9d58a2 KO |
605 | csum_done: |
606 | /* last_seq is ignored when JSET_NO_FLUSH is true */ | |
607 | if (journal_entry_err_on(!JSET_NO_FLUSH(jset) && | |
608 | le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq), c, | |
609 | "invalid journal entry: last_seq > seq (%llu > %llu)", | |
610 | le64_to_cpu(jset->last_seq), | |
611 | le64_to_cpu(jset->seq))) { | |
1c6fdbd8 | 612 | jset->last_seq = jset->seq; |
ca73852a KO |
613 | return JOURNAL_ENTRY_BAD; |
614 | } | |
1c6fdbd8 KO |
615 | fsck_err: |
616 | return ret; | |
617 | } | |
618 | ||
ed9d58a2 KO |
619 | static int jset_validate_for_write(struct bch_fs *c, struct jset *jset) |
620 | { | |
621 | unsigned sectors = vstruct_sectors(jset, c->block_bits); | |
622 | ||
623 | return jset_validate(c, NULL, jset, 0, sectors, sectors, WRITE) ?: | |
624 | jset_validate_entries(c, jset, WRITE); | |
625 | } | |
626 | ||
1c6fdbd8 KO |
627 | struct journal_read_buf { |
628 | void *data; | |
629 | size_t size; | |
630 | }; | |
631 | ||
632 | static int journal_read_buf_realloc(struct journal_read_buf *b, | |
633 | size_t new_size) | |
634 | { | |
635 | void *n; | |
636 | ||
637 | /* the bios are sized for this many pages, max: */ | |
638 | if (new_size > JOURNAL_ENTRY_SIZE_MAX) | |
639 | return -ENOMEM; | |
640 | ||
641 | new_size = roundup_pow_of_two(new_size); | |
642 | n = kvpmalloc(new_size, GFP_KERNEL); | |
643 | if (!n) | |
644 | return -ENOMEM; | |
645 | ||
646 | kvpfree(b->data, b->size); | |
647 | b->data = n; | |
648 | b->size = new_size; | |
649 | return 0; | |
650 | } | |
651 | ||
652 | static int journal_read_bucket(struct bch_dev *ca, | |
653 | struct journal_read_buf *buf, | |
654 | struct journal_list *jlist, | |
a9ec3454 | 655 | unsigned bucket) |
1c6fdbd8 KO |
656 | { |
657 | struct bch_fs *c = ca->fs; | |
658 | struct journal_device *ja = &ca->journal; | |
1c6fdbd8 KO |
659 | struct jset *j = NULL; |
660 | unsigned sectors, sectors_read = 0; | |
661 | u64 offset = bucket_to_sector(ca, ja->buckets[bucket]), | |
662 | end = offset + ca->mi.bucket_size; | |
663 | bool saw_bad = false; | |
664 | int ret = 0; | |
665 | ||
666 | pr_debug("reading %u", bucket); | |
667 | ||
668 | while (offset < end) { | |
669 | if (!sectors_read) { | |
ac10a961 KO |
670 | struct bio *bio; |
671 | unsigned nr_bvecs; | |
672 | reread: | |
673 | sectors_read = min_t(unsigned, | |
1c6fdbd8 | 674 | end - offset, buf->size >> 9); |
ac10a961 KO |
675 | nr_bvecs = buf_pages(buf->data, sectors_read << 9); |
676 | ||
677 | bio = bio_kmalloc(nr_bvecs, GFP_KERNEL); | |
678 | bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ); | |
1c6fdbd8 | 679 | |
885678f6 KO |
680 | bio->bi_iter.bi_sector = offset; |
681 | bch2_bio_map(bio, buf->data, sectors_read << 9); | |
1c6fdbd8 KO |
682 | |
683 | ret = submit_bio_wait(bio); | |
ac10a961 | 684 | kfree(bio); |
1c6fdbd8 KO |
685 | |
686 | if (bch2_dev_io_err_on(ret, ca, | |
0fefe8d8 | 687 | "journal read error: sector %llu", |
1c6fdbd8 | 688 | offset) || |
29d90f61 KO |
689 | bch2_meta_read_fault("journal")) { |
690 | /* | |
691 | * We don't error out of the recovery process | |
692 | * here, since the relevant journal entry may be | |
693 | * found on a different device, and missing or | |
694 | * no journal entries will be handled later | |
695 | */ | |
696 | return 0; | |
697 | } | |
1c6fdbd8 KO |
698 | |
699 | j = buf->data; | |
700 | } | |
701 | ||
ca73852a | 702 | ret = jset_validate(c, ca, j, offset, |
1c6fdbd8 KO |
703 | end - offset, sectors_read, |
704 | READ); | |
705 | switch (ret) { | |
706 | case BCH_FSCK_OK: | |
ca73852a | 707 | sectors = vstruct_sectors(j, c->block_bits); |
1c6fdbd8 KO |
708 | break; |
709 | case JOURNAL_ENTRY_REREAD: | |
710 | if (vstruct_bytes(j) > buf->size) { | |
711 | ret = journal_read_buf_realloc(buf, | |
712 | vstruct_bytes(j)); | |
713 | if (ret) | |
714 | return ret; | |
715 | } | |
716 | goto reread; | |
717 | case JOURNAL_ENTRY_NONE: | |
718 | if (!saw_bad) | |
719 | return 0; | |
8244f320 | 720 | sectors = block_sectors(c); |
1c6fdbd8 KO |
721 | goto next_block; |
722 | case JOURNAL_ENTRY_BAD: | |
723 | saw_bad = true; | |
ca73852a KO |
724 | /* |
725 | * On checksum error we don't really trust the size | |
726 | * field of the journal entry we read, so try reading | |
727 | * again at next block boundary: | |
728 | */ | |
8244f320 | 729 | sectors = block_sectors(c); |
ca73852a | 730 | break; |
1c6fdbd8 KO |
731 | default: |
732 | return ret; | |
733 | } | |
734 | ||
735 | /* | |
736 | * This happens sometimes if we don't have discards on - | |
737 | * when we've partially overwritten a bucket with new | |
738 | * journal entries. We don't need the rest of the | |
739 | * bucket: | |
740 | */ | |
741 | if (le64_to_cpu(j->seq) < ja->bucket_seq[bucket]) | |
742 | return 0; | |
743 | ||
744 | ja->bucket_seq[bucket] = le64_to_cpu(j->seq); | |
745 | ||
746 | mutex_lock(&jlist->lock); | |
e4c3f386 KO |
747 | ret = journal_entry_add(c, ca, (struct bch_extent_ptr) { |
748 | .dev = ca->dev_idx, | |
749 | .offset = offset, | |
750 | }, jlist, j, ret != 0); | |
1c6fdbd8 KO |
751 | mutex_unlock(&jlist->lock); |
752 | ||
753 | switch (ret) { | |
754 | case JOURNAL_ENTRY_ADD_OK: | |
1c6fdbd8 KO |
755 | break; |
756 | case JOURNAL_ENTRY_ADD_OUT_OF_RANGE: | |
757 | break; | |
758 | default: | |
759 | return ret; | |
760 | } | |
1c6fdbd8 KO |
761 | next_block: |
762 | pr_debug("next"); | |
763 | offset += sectors; | |
764 | sectors_read -= sectors; | |
765 | j = ((void *) j) + (sectors << 9); | |
766 | } | |
767 | ||
768 | return 0; | |
769 | } | |
770 | ||
771 | static void bch2_journal_read_device(struct closure *cl) | |
772 | { | |
1c6fdbd8 KO |
773 | struct journal_device *ja = |
774 | container_of(cl, struct journal_device, read); | |
775 | struct bch_dev *ca = container_of(ja, struct bch_dev, journal); | |
776 | struct journal_list *jlist = | |
777 | container_of(cl->parent, struct journal_list, cl); | |
1c6fdbd8 | 778 | struct journal_read_buf buf = { NULL, 0 }; |
a9ec3454 KO |
779 | u64 min_seq = U64_MAX; |
780 | unsigned i; | |
1c6fdbd8 KO |
781 | int ret; |
782 | ||
783 | if (!ja->nr) | |
784 | goto out; | |
785 | ||
1c6fdbd8 KO |
786 | ret = journal_read_buf_realloc(&buf, PAGE_SIZE); |
787 | if (ret) | |
788 | goto err; | |
789 | ||
790 | pr_debug("%u journal buckets", ja->nr); | |
791 | ||
1c6fdbd8 | 792 | for (i = 0; i < ja->nr; i++) { |
a9ec3454 KO |
793 | ret = journal_read_bucket(ca, &buf, jlist, i); |
794 | if (ret) | |
795 | goto err; | |
1c6fdbd8 KO |
796 | } |
797 | ||
a9ec3454 KO |
798 | /* Find the journal bucket with the highest sequence number: */ |
799 | for (i = 0; i < ja->nr; i++) { | |
800 | if (ja->bucket_seq[i] > ja->bucket_seq[ja->cur_idx]) | |
801 | ja->cur_idx = i; | |
1c6fdbd8 | 802 | |
a9ec3454 | 803 | min_seq = min(ja->bucket_seq[i], min_seq); |
1c6fdbd8 KO |
804 | } |
805 | ||
1c6fdbd8 | 806 | /* |
1c6fdbd8 KO |
807 | * If there's duplicate journal entries in multiple buckets (which |
808 | * definitely isn't supposed to happen, but...) - make sure to start | |
809 | * cur_idx at the last of those buckets, so we don't deadlock trying to | |
810 | * allocate | |
811 | */ | |
a9ec3454 KO |
812 | while (ja->bucket_seq[ja->cur_idx] > min_seq && |
813 | ja->bucket_seq[ja->cur_idx] > | |
814 | ja->bucket_seq[(ja->cur_idx + 1) % ja->nr]) | |
a36d3685 | 815 | ja->cur_idx = (ja->cur_idx + 1) % ja->nr; |
a9ec3454 KO |
816 | |
817 | ja->sectors_free = 0; | |
1c6fdbd8 KO |
818 | |
819 | /* | |
0ce2dbbe | 820 | * Set dirty_idx to indicate the entire journal is full and needs to be |
1c6fdbd8 KO |
821 | * reclaimed - journal reclaim will immediately reclaim whatever isn't |
822 | * pinned when it first runs: | |
823 | */ | |
0ce2dbbe KO |
824 | ja->discard_idx = ja->dirty_idx_ondisk = |
825 | ja->dirty_idx = (ja->cur_idx + 1) % ja->nr; | |
1c6fdbd8 KO |
826 | out: |
827 | kvpfree(buf.data, buf.size); | |
1c6fdbd8 KO |
828 | percpu_ref_put(&ca->io_ref); |
829 | closure_return(cl); | |
830 | return; | |
831 | err: | |
832 | mutex_lock(&jlist->lock); | |
833 | jlist->ret = ret; | |
834 | mutex_unlock(&jlist->lock); | |
835 | goto out; | |
1c6fdbd8 KO |
836 | } |
837 | ||
e4c3f386 KO |
838 | static void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c, |
839 | struct journal_replay *j) | |
840 | { | |
841 | unsigned i; | |
842 | ||
843 | for (i = 0; i < j->nr_ptrs; i++) { | |
c0ebe3e4 | 844 | struct bch_dev *ca = bch_dev_bkey_exists(c, j->ptrs[i].dev); |
514852c2 KO |
845 | u64 offset; |
846 | ||
847 | div64_u64_rem(j->ptrs[i].offset, ca->mi.bucket_size, &offset); | |
e4c3f386 KO |
848 | |
849 | if (i) | |
850 | pr_buf(out, " "); | |
851 | pr_buf(out, "%u:%llu (offset %llu)", | |
852 | j->ptrs[i].dev, | |
514852c2 | 853 | (u64) j->ptrs[i].offset, offset); |
e4c3f386 KO |
854 | } |
855 | } | |
856 | ||
adbcada4 KO |
857 | int bch2_journal_read(struct bch_fs *c, struct list_head *list, |
858 | u64 *blacklist_seq, u64 *start_seq) | |
1c6fdbd8 | 859 | { |
1c6fdbd8 | 860 | struct journal_list jlist; |
adbcada4 | 861 | struct journal_replay *i, *t; |
1c6fdbd8 | 862 | struct bch_dev *ca; |
1c6fdbd8 KO |
863 | unsigned iter; |
864 | size_t keys = 0, entries = 0; | |
865 | bool degraded = false; | |
adbcada4 | 866 | u64 seq, last_seq = 0; |
1c6fdbd8 KO |
867 | int ret = 0; |
868 | ||
869 | closure_init_stack(&jlist.cl); | |
870 | mutex_init(&jlist.lock); | |
871 | jlist.head = list; | |
872 | jlist.ret = 0; | |
873 | ||
874 | for_each_member_device(ca, c, iter) { | |
6bdbfa87 | 875 | if (!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) && |
89fd25be | 876 | !(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_journal))) |
1c6fdbd8 KO |
877 | continue; |
878 | ||
2436cb9f KO |
879 | if ((ca->mi.state == BCH_MEMBER_STATE_rw || |
880 | ca->mi.state == BCH_MEMBER_STATE_ro) && | |
1c6fdbd8 KO |
881 | percpu_ref_tryget(&ca->io_ref)) |
882 | closure_call(&ca->journal.read, | |
883 | bch2_journal_read_device, | |
884 | system_unbound_wq, | |
885 | &jlist.cl); | |
886 | else | |
887 | degraded = true; | |
888 | } | |
889 | ||
890 | closure_sync(&jlist.cl); | |
891 | ||
892 | if (jlist.ret) | |
893 | return jlist.ret; | |
894 | ||
adbcada4 KO |
895 | if (list_empty(list)) { |
896 | bch_info(c, "journal read done, but no entries found"); | |
897 | return 0; | |
898 | } | |
899 | ||
900 | i = list_last_entry(list, struct journal_replay, list); | |
901 | *start_seq = le64_to_cpu(i->j.seq) + 1; | |
902 | ||
903 | /* | |
904 | * Find most recent flush entry, and ignore newer non flush entries - | |
905 | * those entries will be blacklisted: | |
906 | */ | |
907 | list_for_each_entry_safe_reverse(i, t, list, list) { | |
908 | if (i->ignore) | |
909 | continue; | |
910 | ||
911 | if (!JSET_NO_FLUSH(&i->j)) { | |
912 | last_seq = le64_to_cpu(i->j.last_seq); | |
913 | *blacklist_seq = le64_to_cpu(i->j.seq) + 1; | |
914 | break; | |
915 | } | |
916 | ||
917 | journal_replay_free(c, i); | |
918 | } | |
919 | ||
920 | if (!last_seq) { | |
921 | fsck_err(c, "journal read done, but no entries found after dropping non-flushes"); | |
922 | return -1; | |
923 | } | |
924 | ||
925 | /* Drop blacklisted entries and entries older than last_seq: */ | |
926 | list_for_each_entry_safe(i, t, list, list) { | |
927 | if (i->ignore) | |
928 | continue; | |
929 | ||
930 | seq = le64_to_cpu(i->j.seq); | |
931 | if (seq < last_seq) { | |
932 | journal_replay_free(c, i); | |
933 | continue; | |
934 | } | |
935 | ||
936 | if (bch2_journal_seq_is_blacklisted(c, seq, true)) { | |
937 | fsck_err_on(!JSET_NO_FLUSH(&i->j), c, | |
938 | "found blacklisted journal entry %llu", seq); | |
939 | ||
940 | journal_replay_free(c, i); | |
941 | } | |
942 | } | |
943 | ||
944 | /* Check for missing entries: */ | |
945 | seq = last_seq; | |
946 | list_for_each_entry(i, list, list) { | |
947 | if (i->ignore) | |
948 | continue; | |
949 | ||
950 | BUG_ON(seq > le64_to_cpu(i->j.seq)); | |
951 | ||
952 | while (seq < le64_to_cpu(i->j.seq)) { | |
953 | u64 missing_start, missing_end; | |
e4c3f386 | 954 | char buf1[200], buf2[200]; |
adbcada4 KO |
955 | |
956 | while (seq < le64_to_cpu(i->j.seq) && | |
957 | bch2_journal_seq_is_blacklisted(c, seq, false)) | |
958 | seq++; | |
959 | ||
960 | if (seq == le64_to_cpu(i->j.seq)) | |
961 | break; | |
962 | ||
963 | missing_start = seq; | |
964 | ||
965 | while (seq < le64_to_cpu(i->j.seq) && | |
966 | !bch2_journal_seq_is_blacklisted(c, seq, false)) | |
967 | seq++; | |
968 | ||
e4c3f386 KO |
969 | if (i->list.prev != list) { |
970 | struct printbuf out = PBUF(buf1); | |
971 | struct journal_replay *p = list_prev_entry(i, list); | |
972 | ||
973 | bch2_journal_ptrs_to_text(&out, c, p); | |
974 | pr_buf(&out, " size %llu", vstruct_sectors(&p->j, c->block_bits)); | |
975 | } else | |
976 | sprintf(buf1, "(none)"); | |
977 | bch2_journal_ptrs_to_text(&PBUF(buf2), c, i); | |
978 | ||
adbcada4 | 979 | missing_end = seq - 1; |
e4c3f386 KO |
980 | fsck_err(c, "journal entries %llu-%llu missing! (replaying %llu-%llu)\n" |
981 | " prev at %s\n" | |
982 | " next at %s", | |
adbcada4 | 983 | missing_start, missing_end, |
e4c3f386 KO |
984 | last_seq, *blacklist_seq - 1, |
985 | buf1, buf2); | |
adbcada4 KO |
986 | } |
987 | ||
988 | seq++; | |
989 | } | |
990 | ||
1c6fdbd8 | 991 | list_for_each_entry(i, list, list) { |
1dd7f9d9 KO |
992 | struct jset_entry *entry; |
993 | struct bkey_i *k, *_n; | |
e4c3f386 KO |
994 | struct bch_replicas_padded replicas = { |
995 | .e.data_type = BCH_DATA_journal, | |
996 | .e.nr_required = 1, | |
997 | }; | |
998 | unsigned ptr; | |
7ef2a73a KO |
999 | char buf[80]; |
1000 | ||
adbcada4 KO |
1001 | if (i->ignore) |
1002 | continue; | |
1003 | ||
1c6fdbd8 KO |
1004 | ret = jset_validate_entries(c, &i->j, READ); |
1005 | if (ret) | |
1006 | goto fsck_err; | |
1007 | ||
e4c3f386 KO |
1008 | for (ptr = 0; ptr < i->nr_ptrs; ptr++) |
1009 | replicas.e.devs[replicas.e.nr_devs++] = i->ptrs[ptr].dev; | |
1010 | ||
26452d1d KO |
1011 | bch2_replicas_entry_sort(&replicas.e); |
1012 | ||
1c6fdbd8 KO |
1013 | /* |
1014 | * If we're mounting in degraded mode - if we didn't read all | |
1015 | * the devices - this is wrong: | |
1016 | */ | |
1017 | ||
1018 | if (!degraded && | |
1019 | (test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) || | |
988e98cf | 1020 | fsck_err_on(!bch2_replicas_marked(c, &replicas.e), c, |
7ef2a73a KO |
1021 | "superblock not marked as containing replicas %s", |
1022 | (bch2_replicas_entry_to_text(&PBUF(buf), | |
1023 | &replicas.e), buf)))) { | |
1024 | ret = bch2_mark_replicas(c, &replicas.e); | |
1c6fdbd8 KO |
1025 | if (ret) |
1026 | return ret; | |
1027 | } | |
1c6fdbd8 KO |
1028 | |
1029 | for_each_jset_key(k, _n, entry, &i->j) | |
1030 | keys++; | |
1031 | entries++; | |
1032 | } | |
1033 | ||
adbcada4 KO |
1034 | bch_info(c, "journal read done, %zu keys in %zu entries, seq %llu", |
1035 | keys, entries, *start_seq); | |
1dd7f9d9 | 1036 | |
adbcada4 KO |
1037 | if (*start_seq != *blacklist_seq) |
1038 | bch_info(c, "dropped unflushed entries %llu-%llu", | |
1039 | *blacklist_seq, *start_seq - 1); | |
1c6fdbd8 KO |
1040 | fsck_err: |
1041 | return ret; | |
1042 | } | |
1043 | ||
1c6fdbd8 KO |
1044 | /* journal write: */ |
1045 | ||
a9ec3454 KO |
1046 | static void __journal_write_alloc(struct journal *j, |
1047 | struct journal_buf *w, | |
1048 | struct dev_alloc_list *devs_sorted, | |
1049 | unsigned sectors, | |
1050 | unsigned *replicas, | |
1051 | unsigned replicas_want) | |
1c6fdbd8 KO |
1052 | { |
1053 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
1c6fdbd8 KO |
1054 | struct journal_device *ja; |
1055 | struct bch_dev *ca; | |
a9ec3454 | 1056 | unsigned i; |
a2753581 | 1057 | |
a9ec3454 KO |
1058 | if (*replicas >= replicas_want) |
1059 | return; | |
1c6fdbd8 | 1060 | |
a9ec3454 KO |
1061 | for (i = 0; i < devs_sorted->nr; i++) { |
1062 | ca = rcu_dereference(c->devs[devs_sorted->devs[i]]); | |
1c6fdbd8 KO |
1063 | if (!ca) |
1064 | continue; | |
1065 | ||
1c6fdbd8 | 1066 | ja = &ca->journal; |
1c6fdbd8 KO |
1067 | |
1068 | /* | |
1069 | * Check that we can use this device, and aren't already using | |
1070 | * it: | |
1071 | */ | |
a9ec3454 | 1072 | if (!ca->mi.durability || |
2436cb9f | 1073 | ca->mi.state != BCH_MEMBER_STATE_rw || |
a9ec3454 | 1074 | !ja->nr || |
26609b61 KO |
1075 | bch2_bkey_has_device(bkey_i_to_s_c(&w->key), |
1076 | ca->dev_idx) || | |
a9ec3454 | 1077 | sectors > ja->sectors_free) |
1c6fdbd8 KO |
1078 | continue; |
1079 | ||
3d080aa5 | 1080 | bch2_dev_stripe_increment(ca, &j->wp.stripe); |
1c6fdbd8 | 1081 | |
26609b61 | 1082 | bch2_bkey_append_ptr(&w->key, |
1c6fdbd8 KO |
1083 | (struct bch_extent_ptr) { |
1084 | .offset = bucket_to_sector(ca, | |
a9ec3454 KO |
1085 | ja->buckets[ja->cur_idx]) + |
1086 | ca->mi.bucket_size - | |
1087 | ja->sectors_free, | |
1c6fdbd8 KO |
1088 | .dev = ca->dev_idx, |
1089 | }); | |
1090 | ||
a9ec3454 KO |
1091 | ja->sectors_free -= sectors; |
1092 | ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq); | |
1093 | ||
1094 | *replicas += ca->mi.durability; | |
1095 | ||
1096 | if (*replicas >= replicas_want) | |
1097 | break; | |
1c6fdbd8 | 1098 | } |
a9ec3454 | 1099 | } |
1c6fdbd8 | 1100 | |
a9ec3454 KO |
1101 | /** |
1102 | * journal_next_bucket - move on to the next journal bucket if possible | |
1103 | */ | |
1104 | static int journal_write_alloc(struct journal *j, struct journal_buf *w, | |
1105 | unsigned sectors) | |
1106 | { | |
1107 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
d042b040 | 1108 | struct bch_devs_mask devs; |
a9ec3454 KO |
1109 | struct journal_device *ja; |
1110 | struct bch_dev *ca; | |
1111 | struct dev_alloc_list devs_sorted; | |
d042b040 KO |
1112 | unsigned target = c->opts.metadata_target ?: |
1113 | c->opts.foreground_target; | |
a9ec3454 KO |
1114 | unsigned i, replicas = 0, replicas_want = |
1115 | READ_ONCE(c->opts.metadata_replicas); | |
1c6fdbd8 | 1116 | |
a9ec3454 | 1117 | rcu_read_lock(); |
d042b040 KO |
1118 | retry: |
1119 | devs = target_rw_devs(c, BCH_DATA_journal, target); | |
1c6fdbd8 | 1120 | |
d042b040 | 1121 | devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe, &devs); |
1c6fdbd8 | 1122 | |
a9ec3454 KO |
1123 | __journal_write_alloc(j, w, &devs_sorted, |
1124 | sectors, &replicas, replicas_want); | |
1c6fdbd8 | 1125 | |
a9ec3454 KO |
1126 | if (replicas >= replicas_want) |
1127 | goto done; | |
1128 | ||
1129 | for (i = 0; i < devs_sorted.nr; i++) { | |
1130 | ca = rcu_dereference(c->devs[devs_sorted.devs[i]]); | |
1131 | if (!ca) | |
1132 | continue; | |
1133 | ||
1134 | ja = &ca->journal; | |
1135 | ||
1136 | if (sectors > ja->sectors_free && | |
1137 | sectors <= ca->mi.bucket_size && | |
03d5eaed KO |
1138 | bch2_journal_dev_buckets_available(j, ja, |
1139 | journal_space_discarded)) { | |
a9ec3454 KO |
1140 | ja->cur_idx = (ja->cur_idx + 1) % ja->nr; |
1141 | ja->sectors_free = ca->mi.bucket_size; | |
68ef94a6 KO |
1142 | |
1143 | /* | |
1144 | * ja->bucket_seq[ja->cur_idx] must always have | |
1145 | * something sensible: | |
1146 | */ | |
1147 | ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq); | |
a9ec3454 KO |
1148 | } |
1149 | } | |
1150 | ||
1151 | __journal_write_alloc(j, w, &devs_sorted, | |
1152 | sectors, &replicas, replicas_want); | |
d042b040 KO |
1153 | |
1154 | if (replicas < replicas_want && target) { | |
1155 | /* Retry from all devices: */ | |
1156 | target = 0; | |
1157 | goto retry; | |
1158 | } | |
a9ec3454 | 1159 | done: |
a9ec3454 KO |
1160 | rcu_read_unlock(); |
1161 | ||
07a1006a KO |
1162 | BUG_ON(bkey_val_u64s(&w->key.k) > BCH_REPLICAS_MAX); |
1163 | ||
57cb2142 | 1164 | return replicas >= c->opts.metadata_replicas_required ? 0 : -EROFS; |
1c6fdbd8 KO |
1165 | } |
1166 | ||
1167 | static void journal_write_compact(struct jset *jset) | |
1168 | { | |
1169 | struct jset_entry *i, *next, *prev = NULL; | |
1170 | ||
1171 | /* | |
1172 | * Simple compaction, dropping empty jset_entries (from journal | |
1173 | * reservations that weren't fully used) and merging jset_entries that | |
1174 | * can be. | |
1175 | * | |
1176 | * If we wanted to be really fancy here, we could sort all the keys in | |
1177 | * the jset and drop keys that were overwritten - probably not worth it: | |
1178 | */ | |
1179 | vstruct_for_each_safe(jset, i, next) { | |
1180 | unsigned u64s = le16_to_cpu(i->u64s); | |
1181 | ||
1182 | /* Empty entry: */ | |
1183 | if (!u64s) | |
1184 | continue; | |
1185 | ||
1186 | /* Can we merge with previous entry? */ | |
1187 | if (prev && | |
1188 | i->btree_id == prev->btree_id && | |
1189 | i->level == prev->level && | |
1190 | i->type == prev->type && | |
1191 | i->type == BCH_JSET_ENTRY_btree_keys && | |
1192 | le16_to_cpu(prev->u64s) + u64s <= U16_MAX) { | |
1193 | memmove_u64s_down(vstruct_next(prev), | |
1194 | i->_data, | |
1195 | u64s); | |
1196 | le16_add_cpu(&prev->u64s, u64s); | |
1197 | continue; | |
1198 | } | |
1199 | ||
1200 | /* Couldn't merge, move i into new position (after prev): */ | |
1201 | prev = prev ? vstruct_next(prev) : jset->start; | |
1202 | if (i != prev) | |
1203 | memmove_u64s_down(prev, i, jset_u64s(u64s)); | |
1204 | } | |
1205 | ||
1206 | prev = prev ? vstruct_next(prev) : jset->start; | |
1207 | jset->u64s = cpu_to_le32((u64 *) prev - jset->_data); | |
1208 | } | |
1209 | ||
1210 | static void journal_buf_realloc(struct journal *j, struct journal_buf *buf) | |
1211 | { | |
1212 | /* we aren't holding j->lock: */ | |
1213 | unsigned new_size = READ_ONCE(j->buf_size_want); | |
1214 | void *new_buf; | |
1215 | ||
d16b4a77 | 1216 | if (buf->buf_size >= new_size) |
1c6fdbd8 KO |
1217 | return; |
1218 | ||
1219 | new_buf = kvpmalloc(new_size, GFP_NOIO|__GFP_NOWARN); | |
1220 | if (!new_buf) | |
1221 | return; | |
1222 | ||
d16b4a77 | 1223 | memcpy(new_buf, buf->data, buf->buf_size); |
c859430b KO |
1224 | |
1225 | spin_lock(&j->lock); | |
1226 | swap(buf->data, new_buf); | |
1227 | swap(buf->buf_size, new_size); | |
1228 | spin_unlock(&j->lock); | |
1229 | ||
1230 | kvpfree(new_buf, new_size); | |
1c6fdbd8 KO |
1231 | } |
1232 | ||
ebb84d09 KO |
1233 | static inline struct journal_buf *journal_last_unwritten_buf(struct journal *j) |
1234 | { | |
1235 | return j->buf + j->reservations.unwritten_idx; | |
1236 | } | |
1237 | ||
1c6fdbd8 KO |
1238 | static void journal_write_done(struct closure *cl) |
1239 | { | |
1240 | struct journal *j = container_of(cl, struct journal, io); | |
1241 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
ebb84d09 | 1242 | struct journal_buf *w = journal_last_unwritten_buf(j); |
7ef2a73a | 1243 | struct bch_replicas_padded replicas; |
ebb84d09 | 1244 | union journal_res_state old, new; |
1784d43a | 1245 | u64 v, seq; |
158eecb8 | 1246 | int err = 0; |
1c6fdbd8 | 1247 | |
991ba021 KO |
1248 | bch2_time_stats_update(!JSET_NO_FLUSH(w->data) |
1249 | ? j->flush_write_time | |
1250 | : j->noflush_write_time, j->write_start_time); | |
9c859dc9 | 1251 | |
d797ca3d | 1252 | if (!w->devs_written.nr) { |
1c6fdbd8 | 1253 | bch_err(c, "unable to write journal to sufficient devices"); |
158eecb8 KO |
1254 | err = -EIO; |
1255 | } else { | |
d797ca3d KO |
1256 | bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal, |
1257 | w->devs_written); | |
158eecb8 KO |
1258 | if (bch2_mark_replicas(c, &replicas.e)) |
1259 | err = -EIO; | |
1c6fdbd8 KO |
1260 | } |
1261 | ||
158eecb8 KO |
1262 | if (err) |
1263 | bch2_fatal_error(c); | |
1c6fdbd8 KO |
1264 | |
1265 | spin_lock(&j->lock); | |
ed9d58a2 | 1266 | seq = le64_to_cpu(w->data->seq); |
ed9d58a2 | 1267 | |
1c6fdbd8 | 1268 | if (seq >= j->pin.front) |
d797ca3d | 1269 | journal_seq_pin(j, seq)->devs = w->devs_written; |
1c6fdbd8 | 1270 | |
9be1efe9 KO |
1271 | if (!err) { |
1272 | j->seq_ondisk = seq; | |
adbcada4 | 1273 | |
9be1efe9 KO |
1274 | if (!JSET_NO_FLUSH(w->data)) { |
1275 | j->flushed_seq_ondisk = seq; | |
1276 | j->last_seq_ondisk = w->last_seq; | |
1277 | } | |
1278 | } else if (!j->err_seq || seq < j->err_seq) | |
1279 | j->err_seq = seq; | |
0ce2dbbe | 1280 | |
1c6fdbd8 KO |
1281 | /* |
1282 | * Updating last_seq_ondisk may let bch2_journal_reclaim_work() discard | |
1283 | * more buckets: | |
1284 | * | |
1285 | * Must come before signaling write completion, for | |
1286 | * bch2_fs_journal_stop(): | |
1287 | */ | |
b7a9bbfc | 1288 | journal_reclaim_kick(&c->journal); |
158eecb8 | 1289 | |
1c6fdbd8 KO |
1290 | /* also must come before signalling write completion: */ |
1291 | closure_debug_destroy(cl); | |
1292 | ||
ebb84d09 KO |
1293 | v = atomic64_read(&j->reservations.counter); |
1294 | do { | |
1295 | old.v = new.v = v; | |
1296 | BUG_ON(new.idx == new.unwritten_idx); | |
1297 | ||
1298 | new.unwritten_idx++; | |
1299 | } while ((v = atomic64_cmpxchg(&j->reservations.counter, | |
1300 | old.v, new.v)) != old.v); | |
1c6fdbd8 | 1301 | |
5d32c5bb KO |
1302 | bch2_journal_space_available(j); |
1303 | ||
1c6fdbd8 KO |
1304 | closure_wake_up(&w->wait); |
1305 | journal_wake(j); | |
1306 | ||
1307 | if (test_bit(JOURNAL_NEED_WRITE, &j->flags)) | |
731bdd2e | 1308 | mod_delayed_work(c->io_complete_wq, &j->write_work, 0); |
1c6fdbd8 | 1309 | spin_unlock(&j->lock); |
ebb84d09 KO |
1310 | |
1311 | if (new.unwritten_idx != new.idx && | |
1312 | !journal_state_count(new, new.unwritten_idx)) | |
731bdd2e | 1313 | closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL); |
1c6fdbd8 KO |
1314 | } |
1315 | ||
1316 | static void journal_write_endio(struct bio *bio) | |
1317 | { | |
1318 | struct bch_dev *ca = bio->bi_private; | |
1319 | struct journal *j = &ca->fs->journal; | |
d797ca3d KO |
1320 | struct journal_buf *w = journal_last_unwritten_buf(j); |
1321 | unsigned long flags; | |
1c6fdbd8 | 1322 | |
d797ca3d KO |
1323 | if (bch2_dev_io_err_on(bio->bi_status, ca, "error writing journal entry %llu: %s", |
1324 | le64_to_cpu(w->data->seq), | |
63b214e7 | 1325 | bch2_blk_status_to_str(bio->bi_status)) || |
1c6fdbd8 | 1326 | bch2_meta_write_fault("journal")) { |
1c6fdbd8 | 1327 | spin_lock_irqsave(&j->err_lock, flags); |
d797ca3d | 1328 | bch2_dev_list_drop_dev(&w->devs_written, ca->dev_idx); |
1c6fdbd8 KO |
1329 | spin_unlock_irqrestore(&j->err_lock, flags); |
1330 | } | |
1331 | ||
1332 | closure_put(&j->io); | |
1333 | percpu_ref_put(&ca->io_ref); | |
1334 | } | |
1335 | ||
280249b9 KO |
1336 | static void do_journal_write(struct closure *cl) |
1337 | { | |
1338 | struct journal *j = container_of(cl, struct journal, io); | |
1339 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
1340 | struct bch_dev *ca; | |
1341 | struct journal_buf *w = journal_last_unwritten_buf(j); | |
1342 | struct bch_extent_ptr *ptr; | |
1343 | struct bio *bio; | |
1344 | unsigned sectors = vstruct_sectors(w->data, c->block_bits); | |
1345 | ||
1346 | extent_for_each_ptr(bkey_i_to_s_extent(&w->key), ptr) { | |
1347 | ca = bch_dev_bkey_exists(c, ptr->dev); | |
1348 | if (!percpu_ref_tryget(&ca->io_ref)) { | |
1349 | /* XXX: fix this */ | |
1350 | bch_err(c, "missing device for journal write\n"); | |
1351 | continue; | |
1352 | } | |
1353 | ||
1354 | this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_journal], | |
1355 | sectors); | |
1356 | ||
1357 | bio = ca->journal.bio; | |
1358 | bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META); | |
1359 | bio->bi_iter.bi_sector = ptr->offset; | |
1360 | bio->bi_end_io = journal_write_endio; | |
1361 | bio->bi_private = ca; | |
1362 | ||
a28bd48a KO |
1363 | BUG_ON(bio->bi_iter.bi_sector == ca->prev_journal_sector); |
1364 | ca->prev_journal_sector = bio->bi_iter.bi_sector; | |
1365 | ||
280249b9 KO |
1366 | if (!JSET_NO_FLUSH(w->data)) |
1367 | bio->bi_opf |= REQ_FUA; | |
1368 | if (!JSET_NO_FLUSH(w->data) && !w->separate_flush) | |
1369 | bio->bi_opf |= REQ_PREFLUSH; | |
1370 | ||
1371 | bch2_bio_map(bio, w->data, sectors << 9); | |
1372 | ||
1373 | trace_journal_write(bio); | |
1374 | closure_bio_submit(bio, cl); | |
1375 | ||
1376 | ca->journal.bucket_seq[ca->journal.cur_idx] = | |
1377 | le64_to_cpu(w->data->seq); | |
1378 | } | |
1379 | ||
731bdd2e | 1380 | continue_at(cl, journal_write_done, c->io_complete_wq); |
280249b9 KO |
1381 | return; |
1382 | } | |
1383 | ||
1c6fdbd8 KO |
1384 | void bch2_journal_write(struct closure *cl) |
1385 | { | |
1386 | struct journal *j = container_of(cl, struct journal, io); | |
1387 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
1388 | struct bch_dev *ca; | |
ebb84d09 | 1389 | struct journal_buf *w = journal_last_unwritten_buf(j); |
3ccc5c50 | 1390 | struct jset_entry *start, *end; |
1c6fdbd8 KO |
1391 | struct jset *jset; |
1392 | struct bio *bio; | |
85674154 | 1393 | char *journal_debug_buf = NULL; |
26609b61 | 1394 | bool validate_before_checksum = false; |
280249b9 | 1395 | unsigned i, sectors, bytes, u64s, nr_rw_members = 0; |
e5a66496 KO |
1396 | int ret; |
1397 | ||
b7a9bbfc KO |
1398 | BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb)); |
1399 | ||
1c6fdbd8 KO |
1400 | journal_buf_realloc(j, w); |
1401 | jset = w->data; | |
1402 | ||
1403 | j->write_start_time = local_clock(); | |
1c6fdbd8 | 1404 | |
adbcada4 KO |
1405 | spin_lock(&j->lock); |
1406 | if (c->sb.features & (1ULL << BCH_FEATURE_journal_no_flush) && | |
5b2e599f KO |
1407 | (w->noflush || |
1408 | (!w->must_flush && | |
1409 | (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) && | |
1410 | test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags)))) { | |
adbcada4 KO |
1411 | w->noflush = true; |
1412 | SET_JSET_NO_FLUSH(jset, true); | |
c0ebe3e4 KO |
1413 | jset->last_seq = 0; |
1414 | w->last_seq = 0; | |
adbcada4 KO |
1415 | |
1416 | j->nr_noflush_writes++; | |
1417 | } else { | |
1418 | j->last_flush_write = jiffies; | |
1419 | j->nr_flush_writes++; | |
1420 | } | |
1421 | spin_unlock(&j->lock); | |
1422 | ||
00b8ccf7 KO |
1423 | /* |
1424 | * New btree roots are set by journalling them; when the journal entry | |
1425 | * gets written we have to propagate them to c->btree_roots | |
1426 | * | |
1427 | * But, every journal entry we write has to contain all the btree roots | |
1428 | * (at least for now); so after we copy btree roots to c->btree_roots we | |
1429 | * have to get any missing btree roots and add them to this journal | |
1430 | * entry: | |
1431 | */ | |
1432 | ||
1433 | bch2_journal_entries_to_btree_roots(c, jset); | |
1434 | ||
1435 | start = end = vstruct_last(jset); | |
1436 | ||
1437 | end = bch2_btree_roots_to_journal_entries(c, jset->start, end); | |
1438 | ||
2abe5420 KO |
1439 | bch2_journal_super_entries_add_common(c, &end, |
1440 | le64_to_cpu(jset->seq)); | |
3ccc5c50 KO |
1441 | u64s = (u64 *) end - (u64 *) start; |
1442 | BUG_ON(u64s > j->entry_u64s_reserved); | |
1443 | ||
d16b4a77 KO |
1444 | le32_add_cpu(&jset->u64s, u64s); |
1445 | BUG_ON(vstruct_sectors(jset, c->block_bits) > w->sectors); | |
1c6fdbd8 KO |
1446 | |
1447 | journal_write_compact(jset); | |
1448 | ||
1c6fdbd8 | 1449 | jset->magic = cpu_to_le64(jset_magic(c)); |
26609b61 KO |
1450 | jset->version = c->sb.version < bcachefs_metadata_version_new_versioning |
1451 | ? cpu_to_le32(BCH_JSET_VERSION_OLD) | |
1452 | : cpu_to_le32(c->sb.version); | |
1c6fdbd8 KO |
1453 | |
1454 | SET_JSET_BIG_ENDIAN(jset, CPU_BIG_ENDIAN); | |
1455 | SET_JSET_CSUM_TYPE(jset, bch2_meta_checksum_type(c)); | |
1456 | ||
4141fde0 | 1457 | if (!JSET_NO_FLUSH(jset) && journal_entry_empty(jset)) |
158eecb8 KO |
1458 | j->last_empty_seq = le64_to_cpu(jset->seq); |
1459 | ||
26609b61 KO |
1460 | if (bch2_csum_type_is_encryption(JSET_CSUM_TYPE(jset))) |
1461 | validate_before_checksum = true; | |
1462 | ||
e751c01a | 1463 | if (le32_to_cpu(jset->version) < bcachefs_metadata_version_current) |
26609b61 KO |
1464 | validate_before_checksum = true; |
1465 | ||
1466 | if (validate_before_checksum && | |
ed9d58a2 | 1467 | jset_validate_for_write(c, jset)) |
1c6fdbd8 KO |
1468 | goto err; |
1469 | ||
1470 | bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset), | |
1471 | jset->encrypted_start, | |
1472 | vstruct_end(jset) - (void *) jset->encrypted_start); | |
1473 | ||
1474 | jset->csum = csum_vstruct(c, JSET_CSUM_TYPE(jset), | |
1475 | journal_nonce(jset), jset); | |
1476 | ||
26609b61 | 1477 | if (!validate_before_checksum && |
ed9d58a2 | 1478 | jset_validate_for_write(c, jset)) |
1c6fdbd8 KO |
1479 | goto err; |
1480 | ||
1481 | sectors = vstruct_sectors(jset, c->block_bits); | |
d16b4a77 | 1482 | BUG_ON(sectors > w->sectors); |
1c6fdbd8 | 1483 | |
d16b4a77 KO |
1484 | bytes = vstruct_bytes(jset); |
1485 | memset((void *) jset + bytes, 0, (sectors << 9) - bytes); | |
1c6fdbd8 | 1486 | |
c18dade6 | 1487 | retry_alloc: |
e5a66496 KO |
1488 | spin_lock(&j->lock); |
1489 | ret = journal_write_alloc(j, w, sectors); | |
1490 | ||
c18dade6 KO |
1491 | if (ret && j->can_discard) { |
1492 | spin_unlock(&j->lock); | |
1493 | bch2_journal_do_discards(j); | |
1494 | goto retry_alloc; | |
1495 | } | |
1496 | ||
85674154 KO |
1497 | if (ret) { |
1498 | journal_debug_buf = kmalloc(4096, GFP_ATOMIC); | |
1499 | if (journal_debug_buf) | |
1500 | __bch2_journal_debug_to_text(&_PBUF(journal_debug_buf, 4096), j); | |
1501 | } | |
1502 | ||
e5a66496 KO |
1503 | /* |
1504 | * write is allocated, no longer need to account for it in | |
1505 | * bch2_journal_space_available(): | |
1506 | */ | |
1507 | w->sectors = 0; | |
1508 | ||
1509 | /* | |
1510 | * journal entry has been compacted and allocated, recalculate space | |
1511 | * available: | |
1512 | */ | |
1513 | bch2_journal_space_available(j); | |
1514 | spin_unlock(&j->lock); | |
1515 | ||
1516 | if (ret) { | |
85674154 KO |
1517 | bch_err(c, "Unable to allocate journal write:\n%s", |
1518 | journal_debug_buf); | |
1519 | kfree(journal_debug_buf); | |
1c6fdbd8 | 1520 | bch2_fatal_error(c); |
731bdd2e | 1521 | continue_at(cl, journal_write_done, c->io_complete_wq); |
1c6fdbd8 KO |
1522 | return; |
1523 | } | |
1524 | ||
d797ca3d KO |
1525 | w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key)); |
1526 | ||
fae1157d | 1527 | if (test_bit(JOURNAL_NOCHANGES, &j->flags)) |
1c6fdbd8 KO |
1528 | goto no_io; |
1529 | ||
280249b9 KO |
1530 | for_each_rw_member(ca, c, i) |
1531 | nr_rw_members++; | |
1c6fdbd8 | 1532 | |
280249b9 KO |
1533 | if (nr_rw_members > 1) |
1534 | w->separate_flush = true; | |
1c6fdbd8 | 1535 | |
280249b9 KO |
1536 | if (!JSET_NO_FLUSH(jset) && w->separate_flush) { |
1537 | for_each_rw_member(ca, c, i) { | |
1538 | percpu_ref_get(&ca->io_ref); | |
1c6fdbd8 | 1539 | |
280249b9 KO |
1540 | bio = ca->journal.bio; |
1541 | bio_reset(bio, ca->disk_sb.bdev, REQ_OP_FLUSH); | |
1542 | bio->bi_end_io = journal_write_endio; | |
1543 | bio->bi_private = ca; | |
1544 | closure_bio_submit(bio, cl); | |
1545 | } | |
1c6fdbd8 KO |
1546 | } |
1547 | ||
280249b9 KO |
1548 | bch2_bucket_seq_cleanup(c); |
1549 | ||
731bdd2e | 1550 | continue_at(cl, do_journal_write, c->io_complete_wq); |
280249b9 | 1551 | return; |
1c6fdbd8 | 1552 | no_io: |
c6923995 KO |
1553 | bch2_bucket_seq_cleanup(c); |
1554 | ||
731bdd2e | 1555 | continue_at(cl, journal_write_done, c->io_complete_wq); |
1c6fdbd8 KO |
1556 | return; |
1557 | err: | |
1558 | bch2_inconsistent_error(c); | |
731bdd2e | 1559 | continue_at(cl, journal_write_done, c->io_complete_wq); |
1c6fdbd8 | 1560 | } |