Commit | Line | Data |
---|---|---|
1c6fdbd8 KO |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | #include "bcachefs.h" | |
59cc38b8 | 3 | #include "alloc_background.h" |
7b3f84ea | 4 | #include "alloc_foreground.h" |
39fb2983 | 5 | #include "btree_io.h" |
00b8ccf7 | 6 | #include "btree_update_interior.h" |
1c6fdbd8 KO |
7 | #include "buckets.h" |
8 | #include "checksum.h" | |
d042b040 | 9 | #include "disk_groups.h" |
1c6fdbd8 | 10 | #include "error.h" |
63b214e7 | 11 | #include "io.h" |
1c6fdbd8 KO |
12 | #include "journal.h" |
13 | #include "journal_io.h" | |
14 | #include "journal_reclaim.h" | |
adbcada4 | 15 | #include "journal_seq_blacklist.h" |
1c6fdbd8 KO |
16 | #include "replicas.h" |
17 | #include "trace.h" | |
18 | ||
adbcada4 KO |
19 | static void __journal_replay_free(struct journal_replay *i) |
20 | { | |
21 | list_del(&i->list); | |
22 | kvpfree(i, offsetof(struct journal_replay, j) + | |
23 | vstruct_bytes(&i->j)); | |
24 | ||
25 | } | |
26 | ||
27 | static void journal_replay_free(struct bch_fs *c, struct journal_replay *i) | |
28 | { | |
29 | i->ignore = true; | |
30 | ||
31 | if (!c->opts.read_entire_journal) | |
32 | __journal_replay_free(i); | |
33 | } | |
34 | ||
1c6fdbd8 KO |
35 | struct journal_list { |
36 | struct closure cl; | |
37 | struct mutex lock; | |
38 | struct list_head *head; | |
39 | int ret; | |
40 | }; | |
41 | ||
42 | #define JOURNAL_ENTRY_ADD_OK 0 | |
43 | #define JOURNAL_ENTRY_ADD_OUT_OF_RANGE 5 | |
44 | ||
45 | /* | |
46 | * Given a journal entry we just read, add it to the list of journal entries to | |
47 | * be replayed: | |
48 | */ | |
49 | static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca, | |
72b7d633 | 50 | struct journal_ptr entry_ptr, |
ca73852a KO |
51 | struct journal_list *jlist, struct jset *j, |
52 | bool bad) | |
1c6fdbd8 | 53 | { |
e4c3f386 | 54 | struct journal_replay *i, *pos, *dup = NULL; |
72b7d633 | 55 | struct journal_ptr *ptr; |
1c6fdbd8 KO |
56 | struct list_head *where; |
57 | size_t bytes = vstruct_bytes(j); | |
adbcada4 | 58 | u64 last_seq = 0; |
e4c3f386 | 59 | int ret = JOURNAL_ENTRY_ADD_OK; |
1c6fdbd8 | 60 | |
adbcada4 KO |
61 | list_for_each_entry_reverse(i, jlist->head, list) { |
62 | if (!JSET_NO_FLUSH(&i->j)) { | |
63 | last_seq = le64_to_cpu(i->j.last_seq); | |
64 | break; | |
7fffc85b | 65 | } |
adbcada4 | 66 | } |
1c6fdbd8 | 67 | |
adbcada4 KO |
68 | /* Is this entry older than the range we need? */ |
69 | if (!c->opts.read_entire_journal && | |
70 | le64_to_cpu(j->seq) < last_seq) { | |
71 | ret = JOURNAL_ENTRY_ADD_OUT_OF_RANGE; | |
72 | goto out; | |
73 | } | |
74 | ||
75 | /* Drop entries we don't need anymore */ | |
76 | if (!JSET_NO_FLUSH(j)) { | |
7fffc85b KO |
77 | list_for_each_entry_safe(i, pos, jlist->head, list) { |
78 | if (le64_to_cpu(i->j.seq) >= le64_to_cpu(j->last_seq)) | |
79 | break; | |
adbcada4 | 80 | journal_replay_free(c, i); |
7fffc85b | 81 | } |
1c6fdbd8 KO |
82 | } |
83 | ||
84 | list_for_each_entry_reverse(i, jlist->head, list) { | |
ca73852a KO |
85 | if (le64_to_cpu(j->seq) > le64_to_cpu(i->j.seq)) { |
86 | where = &i->list; | |
87 | goto add; | |
88 | } | |
89 | } | |
90 | ||
91 | where = jlist->head; | |
92 | add: | |
e4c3f386 | 93 | dup = where->next != jlist->head |
ca73852a KO |
94 | ? container_of(where->next, struct journal_replay, list) |
95 | : NULL; | |
96 | ||
e4c3f386 KO |
97 | if (dup && le64_to_cpu(j->seq) != le64_to_cpu(dup->j.seq)) |
98 | dup = NULL; | |
99 | ||
ca73852a KO |
100 | /* |
101 | * Duplicate journal entries? If so we want the one that didn't have a | |
102 | * checksum error: | |
103 | */ | |
e4c3f386 KO |
104 | if (dup) { |
105 | if (dup->bad) { | |
106 | /* we'll replace @dup: */ | |
ca73852a | 107 | } else if (bad) { |
e4c3f386 | 108 | i = dup; |
ca73852a KO |
109 | goto found; |
110 | } else { | |
e4c3f386 KO |
111 | fsck_err_on(bytes != vstruct_bytes(&dup->j) || |
112 | memcmp(j, &dup->j, bytes), c, | |
1c6fdbd8 KO |
113 | "found duplicate but non identical journal entries (seq %llu)", |
114 | le64_to_cpu(j->seq)); | |
e4c3f386 | 115 | i = dup; |
1c6fdbd8 KO |
116 | goto found; |
117 | } | |
1c6fdbd8 KO |
118 | } |
119 | ||
1c6fdbd8 KO |
120 | i = kvpmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL); |
121 | if (!i) { | |
122 | ret = -ENOMEM; | |
123 | goto out; | |
124 | } | |
125 | ||
e4c3f386 KO |
126 | i->nr_ptrs = 0; |
127 | i->bad = bad; | |
128 | i->ignore = false; | |
1c6fdbd8 | 129 | unsafe_memcpy(&i->j, j, bytes, "embedded variable length struct"); |
e4c3f386 KO |
130 | |
131 | if (dup) { | |
132 | i->nr_ptrs = dup->nr_ptrs; | |
133 | memcpy(i->ptrs, dup->ptrs, sizeof(dup->ptrs)); | |
134 | __journal_replay_free(dup); | |
135 | } | |
136 | ||
137 | list_add(&i->list, where); | |
1c6fdbd8 | 138 | found: |
e4c3f386 KO |
139 | for (ptr = i->ptrs; ptr < i->ptrs + i->nr_ptrs; ptr++) { |
140 | if (ptr->dev == ca->dev_idx) { | |
141 | bch_err(c, "duplicate journal entry %llu on same device", | |
142 | le64_to_cpu(i->j.seq)); | |
143 | goto out; | |
144 | } | |
145 | } | |
146 | ||
147 | if (i->nr_ptrs >= ARRAY_SIZE(i->ptrs)) { | |
148 | bch_err(c, "found too many copies of journal entry %llu", | |
149 | le64_to_cpu(i->j.seq)); | |
150 | goto out; | |
151 | } | |
152 | ||
153 | i->ptrs[i->nr_ptrs++] = entry_ptr; | |
1c6fdbd8 KO |
154 | out: |
155 | fsck_err: | |
156 | return ret; | |
157 | } | |
158 | ||
159 | static struct nonce journal_nonce(const struct jset *jset) | |
160 | { | |
161 | return (struct nonce) {{ | |
162 | [0] = 0, | |
163 | [1] = ((__le32 *) &jset->seq)[0], | |
164 | [2] = ((__le32 *) &jset->seq)[1], | |
165 | [3] = BCH_NONCE_JOURNAL, | |
166 | }}; | |
167 | } | |
168 | ||
169 | /* this fills in a range with empty jset_entries: */ | |
170 | static void journal_entry_null_range(void *start, void *end) | |
171 | { | |
172 | struct jset_entry *entry; | |
173 | ||
174 | for (entry = start; entry != end; entry = vstruct_next(entry)) | |
175 | memset(entry, 0, sizeof(*entry)); | |
176 | } | |
177 | ||
178 | #define JOURNAL_ENTRY_REREAD 5 | |
179 | #define JOURNAL_ENTRY_NONE 6 | |
180 | #define JOURNAL_ENTRY_BAD 7 | |
181 | ||
182 | #define journal_entry_err(c, msg, ...) \ | |
183 | ({ \ | |
184 | switch (write) { \ | |
185 | case READ: \ | |
186 | mustfix_fsck_err(c, msg, ##__VA_ARGS__); \ | |
187 | break; \ | |
188 | case WRITE: \ | |
189 | bch_err(c, "corrupt metadata before write:\n" \ | |
190 | msg, ##__VA_ARGS__); \ | |
191 | if (bch2_fs_inconsistent(c)) { \ | |
192 | ret = BCH_FSCK_ERRORS_NOT_FIXED; \ | |
193 | goto fsck_err; \ | |
194 | } \ | |
195 | break; \ | |
196 | } \ | |
197 | true; \ | |
198 | }) | |
199 | ||
200 | #define journal_entry_err_on(cond, c, msg, ...) \ | |
201 | ((cond) ? journal_entry_err(c, msg, ##__VA_ARGS__) : false) | |
202 | ||
4d54337c KO |
203 | #define FSCK_DELETED_KEY 5 |
204 | ||
7d6f07ed | 205 | static int journal_validate_key(struct bch_fs *c, const char *where, |
1c6fdbd8 | 206 | struct jset_entry *entry, |
39fb2983 | 207 | unsigned level, enum btree_id btree_id, |
7d6f07ed KO |
208 | struct bkey_i *k, const char *type, |
209 | unsigned version, int big_endian, int write) | |
1c6fdbd8 KO |
210 | { |
211 | void *next = vstruct_next(entry); | |
f0ac7df2 | 212 | struct printbuf buf = PRINTBUF; |
1c6fdbd8 KO |
213 | int ret = 0; |
214 | ||
215 | if (journal_entry_err_on(!k->k.u64s, c, | |
7d6f07ed KO |
216 | "invalid %s in %s entry offset %zi/%u: k->u64s 0", |
217 | type, where, | |
4d54337c KO |
218 | (u64 *) k - entry->_data, |
219 | le16_to_cpu(entry->u64s))) { | |
1c6fdbd8 KO |
220 | entry->u64s = cpu_to_le16((u64 *) k - entry->_data); |
221 | journal_entry_null_range(vstruct_next(entry), next); | |
4d54337c | 222 | return FSCK_DELETED_KEY; |
1c6fdbd8 KO |
223 | } |
224 | ||
225 | if (journal_entry_err_on((void *) bkey_next(k) > | |
226 | (void *) vstruct_next(entry), c, | |
7d6f07ed KO |
227 | "invalid %s in %s entry offset %zi/%u: extends past end of journal entry", |
228 | type, where, | |
4d54337c KO |
229 | (u64 *) k - entry->_data, |
230 | le16_to_cpu(entry->u64s))) { | |
1c6fdbd8 KO |
231 | entry->u64s = cpu_to_le16((u64 *) k - entry->_data); |
232 | journal_entry_null_range(vstruct_next(entry), next); | |
4d54337c | 233 | return FSCK_DELETED_KEY; |
1c6fdbd8 KO |
234 | } |
235 | ||
236 | if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT, c, | |
7d6f07ed KO |
237 | "invalid %s in %s entry offset %zi/%u: bad format %u", |
238 | type, where, | |
4d54337c KO |
239 | (u64 *) k - entry->_data, |
240 | le16_to_cpu(entry->u64s), | |
ed0d631f | 241 | k->k.format)) { |
4d54337c | 242 | le16_add_cpu(&entry->u64s, -((u16) k->k.u64s)); |
1c6fdbd8 KO |
243 | memmove(k, bkey_next(k), next - (void *) bkey_next(k)); |
244 | journal_entry_null_range(vstruct_next(entry), next); | |
4d54337c | 245 | return FSCK_DELETED_KEY; |
1c6fdbd8 KO |
246 | } |
247 | ||
39fb2983 | 248 | if (!write) |
7d6f07ed KO |
249 | bch2_bkey_compat(level, btree_id, version, big_endian, |
250 | write, NULL, bkey_to_packed(k)); | |
26609b61 | 251 | |
f0ac7df2 KO |
252 | if (bch2_bkey_invalid(c, bkey_i_to_s_c(k), |
253 | __btree_node_type(level, btree_id), &buf)) { | |
254 | printbuf_reset(&buf); | |
255 | pr_buf(&buf, "invalid %s in %s entry offset %zi/%u:", | |
256 | type, where, | |
257 | (u64 *) k - entry->_data, | |
258 | le16_to_cpu(entry->u64s)); | |
259 | pr_newline(&buf); | |
260 | pr_indent_push(&buf, 2); | |
319f9ac3 | 261 | |
fa8e94fa | 262 | bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k)); |
f0ac7df2 KO |
263 | pr_newline(&buf); |
264 | bch2_bkey_invalid(c, bkey_i_to_s_c(k), | |
265 | __btree_node_type(level, btree_id), &buf); | |
266 | ||
267 | mustfix_fsck_err(c, "%s", buf.buf); | |
1c6fdbd8 | 268 | |
4d54337c | 269 | le16_add_cpu(&entry->u64s, -((u16) k->k.u64s)); |
1c6fdbd8 KO |
270 | memmove(k, bkey_next(k), next - (void *) bkey_next(k)); |
271 | journal_entry_null_range(vstruct_next(entry), next); | |
f0ac7df2 KO |
272 | |
273 | printbuf_exit(&buf); | |
4d54337c | 274 | return FSCK_DELETED_KEY; |
1c6fdbd8 | 275 | } |
26609b61 | 276 | |
39fb2983 | 277 | if (write) |
7d6f07ed KO |
278 | bch2_bkey_compat(level, btree_id, version, big_endian, |
279 | write, NULL, bkey_to_packed(k)); | |
1c6fdbd8 | 280 | fsck_err: |
f0ac7df2 | 281 | printbuf_exit(&buf); |
1c6fdbd8 KO |
282 | return ret; |
283 | } | |
284 | ||
528b18e6 | 285 | static int journal_entry_btree_keys_validate(struct bch_fs *c, |
7d6f07ed | 286 | const char *where, |
1c6fdbd8 | 287 | struct jset_entry *entry, |
7d6f07ed | 288 | unsigned version, int big_endian, int write) |
1c6fdbd8 | 289 | { |
4d54337c | 290 | struct bkey_i *k = entry->start; |
1c6fdbd8 | 291 | |
4d54337c | 292 | while (k != vstruct_last(entry)) { |
7d6f07ed | 293 | int ret = journal_validate_key(c, where, entry, |
39fb2983 KO |
294 | entry->level, |
295 | entry->btree_id, | |
7d6f07ed | 296 | k, "key", version, big_endian, write); |
4d54337c KO |
297 | if (ret == FSCK_DELETED_KEY) |
298 | continue; | |
299 | ||
300 | k = bkey_next(k); | |
1c6fdbd8 KO |
301 | } |
302 | ||
303 | return 0; | |
304 | } | |
305 | ||
528b18e6 KO |
306 | static void journal_entry_btree_keys_to_text(struct printbuf *out, struct bch_fs *c, |
307 | struct jset_entry *entry) | |
308 | { | |
309 | struct bkey_i *k; | |
e7bc7cdf | 310 | bool first = true; |
528b18e6 | 311 | |
e7bc7cdf KO |
312 | vstruct_for_each(entry, k) { |
313 | if (!first) { | |
12bf93a4 | 314 | pr_newline(out); |
e7bc7cdf KO |
315 | pr_buf(out, "%s: ", bch2_jset_entry_types[entry->type]); |
316 | } | |
317 | pr_buf(out, "btree=%s l=%u ", bch2_btree_ids[entry->btree_id], entry->level); | |
528b18e6 | 318 | bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(k)); |
e7bc7cdf KO |
319 | first = false; |
320 | } | |
528b18e6 KO |
321 | } |
322 | ||
323 | static int journal_entry_btree_root_validate(struct bch_fs *c, | |
7d6f07ed | 324 | const char *where, |
1c6fdbd8 | 325 | struct jset_entry *entry, |
7d6f07ed | 326 | unsigned version, int big_endian, int write) |
1c6fdbd8 KO |
327 | { |
328 | struct bkey_i *k = entry->start; | |
329 | int ret = 0; | |
330 | ||
331 | if (journal_entry_err_on(!entry->u64s || | |
332 | le16_to_cpu(entry->u64s) != k->k.u64s, c, | |
333 | "invalid btree root journal entry: wrong number of keys")) { | |
334 | void *next = vstruct_next(entry); | |
335 | /* | |
336 | * we don't want to null out this jset_entry, | |
337 | * just the contents, so that later we can tell | |
338 | * we were _supposed_ to have a btree root | |
339 | */ | |
340 | entry->u64s = 0; | |
341 | journal_entry_null_range(vstruct_next(entry), next); | |
342 | return 0; | |
343 | } | |
344 | ||
7d6f07ed KO |
345 | return journal_validate_key(c, where, entry, 1, entry->btree_id, k, |
346 | "btree root", version, big_endian, write); | |
1c6fdbd8 KO |
347 | fsck_err: |
348 | return ret; | |
349 | } | |
350 | ||
528b18e6 KO |
351 | static void journal_entry_btree_root_to_text(struct printbuf *out, struct bch_fs *c, |
352 | struct jset_entry *entry) | |
353 | { | |
354 | journal_entry_btree_keys_to_text(out, c, entry); | |
355 | } | |
356 | ||
357 | static int journal_entry_prio_ptrs_validate(struct bch_fs *c, | |
7d6f07ed | 358 | const char *where, |
1c6fdbd8 | 359 | struct jset_entry *entry, |
7d6f07ed | 360 | unsigned version, int big_endian, int write) |
1c6fdbd8 KO |
361 | { |
362 | /* obsolete, don't care: */ | |
363 | return 0; | |
364 | } | |
365 | ||
528b18e6 KO |
366 | static void journal_entry_prio_ptrs_to_text(struct printbuf *out, struct bch_fs *c, |
367 | struct jset_entry *entry) | |
368 | { | |
369 | } | |
370 | ||
371 | static int journal_entry_blacklist_validate(struct bch_fs *c, | |
7d6f07ed | 372 | const char *where, |
1c6fdbd8 | 373 | struct jset_entry *entry, |
7d6f07ed | 374 | unsigned version, int big_endian, int write) |
1c6fdbd8 KO |
375 | { |
376 | int ret = 0; | |
377 | ||
378 | if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 1, c, | |
379 | "invalid journal seq blacklist entry: bad size")) { | |
380 | journal_entry_null_range(entry, vstruct_next(entry)); | |
381 | } | |
382 | fsck_err: | |
383 | return ret; | |
384 | } | |
385 | ||
528b18e6 KO |
386 | static void journal_entry_blacklist_to_text(struct printbuf *out, struct bch_fs *c, |
387 | struct jset_entry *entry) | |
388 | { | |
389 | struct jset_entry_blacklist *bl = | |
390 | container_of(entry, struct jset_entry_blacklist, entry); | |
391 | ||
392 | pr_buf(out, "seq=%llu", le64_to_cpu(bl->seq)); | |
393 | } | |
394 | ||
395 | static int journal_entry_blacklist_v2_validate(struct bch_fs *c, | |
7d6f07ed | 396 | const char *where, |
1c6fdbd8 | 397 | struct jset_entry *entry, |
7d6f07ed | 398 | unsigned version, int big_endian, int write) |
1c6fdbd8 KO |
399 | { |
400 | struct jset_entry_blacklist_v2 *bl_entry; | |
401 | int ret = 0; | |
402 | ||
403 | if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 2, c, | |
404 | "invalid journal seq blacklist entry: bad size")) { | |
405 | journal_entry_null_range(entry, vstruct_next(entry)); | |
2c5af169 | 406 | goto out; |
1c6fdbd8 KO |
407 | } |
408 | ||
409 | bl_entry = container_of(entry, struct jset_entry_blacklist_v2, entry); | |
410 | ||
411 | if (journal_entry_err_on(le64_to_cpu(bl_entry->start) > | |
412 | le64_to_cpu(bl_entry->end), c, | |
413 | "invalid journal seq blacklist entry: start > end")) { | |
414 | journal_entry_null_range(entry, vstruct_next(entry)); | |
415 | } | |
2c5af169 KO |
416 | out: |
417 | fsck_err: | |
418 | return ret; | |
419 | } | |
420 | ||
528b18e6 KO |
421 | static void journal_entry_blacklist_v2_to_text(struct printbuf *out, struct bch_fs *c, |
422 | struct jset_entry *entry) | |
423 | { | |
424 | struct jset_entry_blacklist_v2 *bl = | |
425 | container_of(entry, struct jset_entry_blacklist_v2, entry); | |
426 | ||
427 | pr_buf(out, "start=%llu end=%llu", | |
428 | le64_to_cpu(bl->start), | |
429 | le64_to_cpu(bl->end)); | |
430 | } | |
431 | ||
432 | static int journal_entry_usage_validate(struct bch_fs *c, | |
7d6f07ed | 433 | const char *where, |
2c5af169 | 434 | struct jset_entry *entry, |
7d6f07ed | 435 | unsigned version, int big_endian, int write) |
2c5af169 KO |
436 | { |
437 | struct jset_entry_usage *u = | |
438 | container_of(entry, struct jset_entry_usage, entry); | |
439 | unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64); | |
440 | int ret = 0; | |
441 | ||
3577df5f KO |
442 | if (journal_entry_err_on(bytes < sizeof(*u), |
443 | c, | |
444 | "invalid journal entry usage: bad size")) { | |
445 | journal_entry_null_range(entry, vstruct_next(entry)); | |
446 | return ret; | |
447 | } | |
448 | ||
449 | fsck_err: | |
450 | return ret; | |
451 | } | |
452 | ||
528b18e6 KO |
453 | static void journal_entry_usage_to_text(struct printbuf *out, struct bch_fs *c, |
454 | struct jset_entry *entry) | |
455 | { | |
456 | struct jset_entry_usage *u = | |
457 | container_of(entry, struct jset_entry_usage, entry); | |
458 | ||
459 | pr_buf(out, "type=%s v=%llu", | |
460 | bch2_fs_usage_types[u->entry.btree_id], | |
461 | le64_to_cpu(u->v)); | |
462 | } | |
463 | ||
464 | static int journal_entry_data_usage_validate(struct bch_fs *c, | |
7d6f07ed | 465 | const char *where, |
3577df5f | 466 | struct jset_entry *entry, |
7d6f07ed | 467 | unsigned version, int big_endian, int write) |
3577df5f KO |
468 | { |
469 | struct jset_entry_data_usage *u = | |
470 | container_of(entry, struct jset_entry_data_usage, entry); | |
471 | unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64); | |
472 | int ret = 0; | |
473 | ||
2c5af169 KO |
474 | if (journal_entry_err_on(bytes < sizeof(*u) || |
475 | bytes < sizeof(*u) + u->r.nr_devs, | |
476 | c, | |
477 | "invalid journal entry usage: bad size")) { | |
478 | journal_entry_null_range(entry, vstruct_next(entry)); | |
479 | return ret; | |
480 | } | |
1c6fdbd8 KO |
481 | |
482 | fsck_err: | |
483 | return ret; | |
484 | } | |
485 | ||
528b18e6 KO |
486 | static void journal_entry_data_usage_to_text(struct printbuf *out, struct bch_fs *c, |
487 | struct jset_entry *entry) | |
488 | { | |
489 | struct jset_entry_data_usage *u = | |
490 | container_of(entry, struct jset_entry_data_usage, entry); | |
491 | ||
492 | bch2_replicas_entry_to_text(out, &u->r); | |
493 | pr_buf(out, "=%llu", le64_to_cpu(u->v)); | |
494 | } | |
495 | ||
496 | static int journal_entry_clock_validate(struct bch_fs *c, | |
7d6f07ed | 497 | const char *where, |
2abe5420 | 498 | struct jset_entry *entry, |
7d6f07ed | 499 | unsigned version, int big_endian, int write) |
2abe5420 KO |
500 | { |
501 | struct jset_entry_clock *clock = | |
502 | container_of(entry, struct jset_entry_clock, entry); | |
503 | unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64); | |
504 | int ret = 0; | |
505 | ||
506 | if (journal_entry_err_on(bytes != sizeof(*clock), | |
507 | c, "invalid journal entry clock: bad size")) { | |
508 | journal_entry_null_range(entry, vstruct_next(entry)); | |
509 | return ret; | |
510 | } | |
511 | ||
512 | if (journal_entry_err_on(clock->rw > 1, | |
513 | c, "invalid journal entry clock: bad rw")) { | |
514 | journal_entry_null_range(entry, vstruct_next(entry)); | |
515 | return ret; | |
516 | } | |
517 | ||
518 | fsck_err: | |
519 | return ret; | |
520 | } | |
521 | ||
528b18e6 KO |
522 | static void journal_entry_clock_to_text(struct printbuf *out, struct bch_fs *c, |
523 | struct jset_entry *entry) | |
524 | { | |
525 | struct jset_entry_clock *clock = | |
526 | container_of(entry, struct jset_entry_clock, entry); | |
527 | ||
528 | pr_buf(out, "%s=%llu", clock->rw ? "write" : "read", le64_to_cpu(clock->time)); | |
529 | } | |
530 | ||
531 | static int journal_entry_dev_usage_validate(struct bch_fs *c, | |
7d6f07ed | 532 | const char *where, |
180fb49d | 533 | struct jset_entry *entry, |
7d6f07ed | 534 | unsigned version, int big_endian, int write) |
180fb49d KO |
535 | { |
536 | struct jset_entry_dev_usage *u = | |
537 | container_of(entry, struct jset_entry_dev_usage, entry); | |
538 | unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64); | |
45c2e33f | 539 | unsigned expected = sizeof(*u); |
180fb49d KO |
540 | unsigned dev; |
541 | int ret = 0; | |
542 | ||
543 | if (journal_entry_err_on(bytes < expected, | |
544 | c, "invalid journal entry dev usage: bad size (%u < %u)", | |
545 | bytes, expected)) { | |
546 | journal_entry_null_range(entry, vstruct_next(entry)); | |
547 | return ret; | |
548 | } | |
549 | ||
550 | dev = le32_to_cpu(u->dev); | |
551 | ||
552 | if (journal_entry_err_on(!bch2_dev_exists2(c, dev), | |
553 | c, "invalid journal entry dev usage: bad dev")) { | |
554 | journal_entry_null_range(entry, vstruct_next(entry)); | |
555 | return ret; | |
556 | } | |
557 | ||
558 | if (journal_entry_err_on(u->pad, | |
559 | c, "invalid journal entry dev usage: bad pad")) { | |
560 | journal_entry_null_range(entry, vstruct_next(entry)); | |
561 | return ret; | |
562 | } | |
563 | ||
564 | fsck_err: | |
565 | return ret; | |
566 | } | |
567 | ||
528b18e6 KO |
568 | static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs *c, |
569 | struct jset_entry *entry) | |
570 | { | |
571 | struct jset_entry_dev_usage *u = | |
572 | container_of(entry, struct jset_entry_dev_usage, entry); | |
573 | unsigned i, nr_types = jset_entry_dev_usage_nr_types(u); | |
574 | ||
575 | pr_buf(out, "dev=%u", le32_to_cpu(u->dev)); | |
576 | ||
577 | for (i = 0; i < nr_types; i++) { | |
578 | if (i < BCH_DATA_NR) | |
579 | pr_buf(out, " %s", bch2_data_types[i]); | |
580 | else | |
581 | pr_buf(out, " (unknown data type %u)", i); | |
582 | pr_buf(out, ": buckets=%llu sectors=%llu fragmented=%llu", | |
583 | le64_to_cpu(u->d[i].buckets), | |
584 | le64_to_cpu(u->d[i].sectors), | |
585 | le64_to_cpu(u->d[i].fragmented)); | |
586 | } | |
587 | ||
588 | pr_buf(out, " buckets_ec: %llu buckets_unavailable: %llu", | |
589 | le64_to_cpu(u->buckets_ec), | |
590 | le64_to_cpu(u->buckets_unavailable)); | |
591 | } | |
592 | ||
593 | static int journal_entry_log_validate(struct bch_fs *c, | |
fb64f3fd KO |
594 | const char *where, |
595 | struct jset_entry *entry, | |
596 | unsigned version, int big_endian, int write) | |
597 | { | |
598 | return 0; | |
599 | } | |
600 | ||
528b18e6 KO |
601 | static void journal_entry_log_to_text(struct printbuf *out, struct bch_fs *c, |
602 | struct jset_entry *entry) | |
603 | { | |
604 | struct jset_entry_log *l = container_of(entry, struct jset_entry_log, entry); | |
605 | unsigned bytes = vstruct_bytes(entry) - offsetof(struct jset_entry_log, d); | |
606 | ||
d4b69152 | 607 | pr_buf(out, "%.*s", bytes, l->d); |
528b18e6 KO |
608 | } |
609 | ||
1c6fdbd8 | 610 | struct jset_entry_ops { |
7d6f07ed KO |
611 | int (*validate)(struct bch_fs *, const char *, |
612 | struct jset_entry *, unsigned, int, int); | |
528b18e6 | 613 | void (*to_text)(struct printbuf *, struct bch_fs *, struct jset_entry *); |
1c6fdbd8 KO |
614 | }; |
615 | ||
616 | static const struct jset_entry_ops bch2_jset_entry_ops[] = { | |
617 | #define x(f, nr) \ | |
618 | [BCH_JSET_ENTRY_##f] = (struct jset_entry_ops) { \ | |
528b18e6 KO |
619 | .validate = journal_entry_##f##_validate, \ |
620 | .to_text = journal_entry_##f##_to_text, \ | |
1c6fdbd8 KO |
621 | }, |
622 | BCH_JSET_ENTRY_TYPES() | |
623 | #undef x | |
624 | }; | |
625 | ||
7d6f07ed KO |
626 | int bch2_journal_entry_validate(struct bch_fs *c, const char *where, |
627 | struct jset_entry *entry, | |
628 | unsigned version, int big_endian, int write) | |
1c6fdbd8 | 629 | { |
2c5af169 | 630 | return entry->type < BCH_JSET_ENTRY_NR |
7d6f07ed KO |
631 | ? bch2_jset_entry_ops[entry->type].validate(c, where, entry, |
632 | version, big_endian, write) | |
2c5af169 | 633 | : 0; |
1c6fdbd8 KO |
634 | } |
635 | ||
528b18e6 KO |
636 | void bch2_journal_entry_to_text(struct printbuf *out, struct bch_fs *c, |
637 | struct jset_entry *entry) | |
638 | { | |
639 | if (entry->type < BCH_JSET_ENTRY_NR) { | |
640 | pr_buf(out, "%s: ", bch2_jset_entry_types[entry->type]); | |
641 | bch2_jset_entry_ops[entry->type].to_text(out, c, entry); | |
642 | } else { | |
643 | pr_buf(out, "(unknown type %u)", entry->type); | |
644 | } | |
645 | } | |
646 | ||
1c6fdbd8 KO |
647 | static int jset_validate_entries(struct bch_fs *c, struct jset *jset, |
648 | int write) | |
649 | { | |
7d6f07ed | 650 | char buf[100]; |
1c6fdbd8 KO |
651 | struct jset_entry *entry; |
652 | int ret = 0; | |
653 | ||
654 | vstruct_for_each(jset, entry) { | |
7d6f07ed KO |
655 | scnprintf(buf, sizeof(buf), "jset %llu entry offset %zi/%u", |
656 | le64_to_cpu(jset->seq), | |
657 | (u64 *) entry - jset->_data, | |
658 | le32_to_cpu(jset->u64s)); | |
659 | ||
1c6fdbd8 KO |
660 | if (journal_entry_err_on(vstruct_next(entry) > |
661 | vstruct_last(jset), c, | |
662 | "journal entry extends past end of jset")) { | |
663 | jset->u64s = cpu_to_le32((u64 *) entry - jset->_data); | |
664 | break; | |
665 | } | |
666 | ||
7d6f07ed KO |
667 | ret = bch2_journal_entry_validate(c, buf, entry, |
668 | le32_to_cpu(jset->version), | |
669 | JSET_BIG_ENDIAN(jset), write); | |
1c6fdbd8 KO |
670 | if (ret) |
671 | break; | |
672 | } | |
673 | fsck_err: | |
674 | return ret; | |
675 | } | |
676 | ||
677 | static int jset_validate(struct bch_fs *c, | |
ca73852a | 678 | struct bch_dev *ca, |
1c6fdbd8 KO |
679 | struct jset *jset, u64 sector, |
680 | unsigned bucket_sectors_left, | |
681 | unsigned sectors_read, | |
682 | int write) | |
683 | { | |
684 | size_t bytes = vstruct_bytes(jset); | |
685 | struct bch_csum csum; | |
26609b61 | 686 | unsigned version; |
1c6fdbd8 KO |
687 | int ret = 0; |
688 | ||
689 | if (le64_to_cpu(jset->magic) != jset_magic(c)) | |
690 | return JOURNAL_ENTRY_NONE; | |
691 | ||
26609b61 | 692 | version = le32_to_cpu(jset->version); |
ca73852a KO |
693 | if (journal_entry_err_on((version != BCH_JSET_VERSION_OLD && |
694 | version < bcachefs_metadata_version_min) || | |
695 | version >= bcachefs_metadata_version_max, c, | |
696 | "%s sector %llu seq %llu: unknown journal entry version %u", | |
ed9d58a2 KO |
697 | ca ? ca->name : c->name, |
698 | sector, le64_to_cpu(jset->seq), | |
ca73852a | 699 | version)) { |
35ef6df5 KO |
700 | /* don't try to continue: */ |
701 | return EINVAL; | |
1c6fdbd8 KO |
702 | } |
703 | ||
35ef6df5 KO |
704 | if (bytes > (sectors_read << 9) && |
705 | sectors_read < bucket_sectors_left) | |
706 | return JOURNAL_ENTRY_REREAD; | |
707 | ||
1c6fdbd8 | 708 | if (journal_entry_err_on(bytes > bucket_sectors_left << 9, c, |
ca73852a | 709 | "%s sector %llu seq %llu: journal entry too big (%zu bytes)", |
ed9d58a2 KO |
710 | ca ? ca->name : c->name, |
711 | sector, le64_to_cpu(jset->seq), bytes)) { | |
35ef6df5 KO |
712 | ret = JOURNAL_ENTRY_BAD; |
713 | le32_add_cpu(&jset->u64s, | |
714 | -((bytes - (bucket_sectors_left << 9)) / 8)); | |
1c6fdbd8 KO |
715 | } |
716 | ||
ed9d58a2 | 717 | if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)), c, |
ca73852a | 718 | "%s sector %llu seq %llu: journal entry with unknown csum type %llu", |
ed9d58a2 KO |
719 | ca ? ca->name : c->name, |
720 | sector, le64_to_cpu(jset->seq), | |
35ef6df5 KO |
721 | JSET_CSUM_TYPE(jset))) { |
722 | ret = JOURNAL_ENTRY_BAD; | |
ed9d58a2 | 723 | goto csum_done; |
35ef6df5 | 724 | } |
1c6fdbd8 | 725 | |
ed9d58a2 KO |
726 | if (write) |
727 | goto csum_done; | |
728 | ||
1c6fdbd8 KO |
729 | csum = csum_vstruct(c, JSET_CSUM_TYPE(jset), journal_nonce(jset), jset); |
730 | if (journal_entry_err_on(bch2_crc_cmp(csum, jset->csum), c, | |
ca73852a | 731 | "%s sector %llu seq %llu: journal checksum bad", |
ed9d58a2 KO |
732 | ca ? ca->name : c->name, |
733 | sector, le64_to_cpu(jset->seq))) | |
35ef6df5 | 734 | ret = JOURNAL_ENTRY_BAD; |
1c6fdbd8 | 735 | |
a9de137b | 736 | ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset), |
1c6fdbd8 KO |
737 | jset->encrypted_start, |
738 | vstruct_end(jset) - (void *) jset->encrypted_start); | |
a9de137b KO |
739 | bch2_fs_fatal_err_on(ret, c, |
740 | "error decrypting journal entry: %i", ret); | |
ed9d58a2 KO |
741 | csum_done: |
742 | /* last_seq is ignored when JSET_NO_FLUSH is true */ | |
743 | if (journal_entry_err_on(!JSET_NO_FLUSH(jset) && | |
744 | le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq), c, | |
745 | "invalid journal entry: last_seq > seq (%llu > %llu)", | |
746 | le64_to_cpu(jset->last_seq), | |
747 | le64_to_cpu(jset->seq))) { | |
1c6fdbd8 | 748 | jset->last_seq = jset->seq; |
ca73852a KO |
749 | return JOURNAL_ENTRY_BAD; |
750 | } | |
1c6fdbd8 KO |
751 | fsck_err: |
752 | return ret; | |
753 | } | |
754 | ||
ed9d58a2 KO |
755 | static int jset_validate_for_write(struct bch_fs *c, struct jset *jset) |
756 | { | |
757 | unsigned sectors = vstruct_sectors(jset, c->block_bits); | |
758 | ||
759 | return jset_validate(c, NULL, jset, 0, sectors, sectors, WRITE) ?: | |
760 | jset_validate_entries(c, jset, WRITE); | |
761 | } | |
762 | ||
1c6fdbd8 KO |
763 | struct journal_read_buf { |
764 | void *data; | |
765 | size_t size; | |
766 | }; | |
767 | ||
768 | static int journal_read_buf_realloc(struct journal_read_buf *b, | |
769 | size_t new_size) | |
770 | { | |
771 | void *n; | |
772 | ||
773 | /* the bios are sized for this many pages, max: */ | |
774 | if (new_size > JOURNAL_ENTRY_SIZE_MAX) | |
775 | return -ENOMEM; | |
776 | ||
777 | new_size = roundup_pow_of_two(new_size); | |
778 | n = kvpmalloc(new_size, GFP_KERNEL); | |
779 | if (!n) | |
780 | return -ENOMEM; | |
781 | ||
782 | kvpfree(b->data, b->size); | |
783 | b->data = n; | |
784 | b->size = new_size; | |
785 | return 0; | |
786 | } | |
787 | ||
788 | static int journal_read_bucket(struct bch_dev *ca, | |
789 | struct journal_read_buf *buf, | |
790 | struct journal_list *jlist, | |
a9ec3454 | 791 | unsigned bucket) |
1c6fdbd8 KO |
792 | { |
793 | struct bch_fs *c = ca->fs; | |
794 | struct journal_device *ja = &ca->journal; | |
1c6fdbd8 KO |
795 | struct jset *j = NULL; |
796 | unsigned sectors, sectors_read = 0; | |
797 | u64 offset = bucket_to_sector(ca, ja->buckets[bucket]), | |
798 | end = offset + ca->mi.bucket_size; | |
799 | bool saw_bad = false; | |
800 | int ret = 0; | |
801 | ||
802 | pr_debug("reading %u", bucket); | |
803 | ||
804 | while (offset < end) { | |
805 | if (!sectors_read) { | |
ac10a961 KO |
806 | struct bio *bio; |
807 | unsigned nr_bvecs; | |
808 | reread: | |
809 | sectors_read = min_t(unsigned, | |
1c6fdbd8 | 810 | end - offset, buf->size >> 9); |
ac10a961 KO |
811 | nr_bvecs = buf_pages(buf->data, sectors_read << 9); |
812 | ||
813 | bio = bio_kmalloc(nr_bvecs, GFP_KERNEL); | |
814 | bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ); | |
1c6fdbd8 | 815 | |
885678f6 KO |
816 | bio->bi_iter.bi_sector = offset; |
817 | bch2_bio_map(bio, buf->data, sectors_read << 9); | |
1c6fdbd8 KO |
818 | |
819 | ret = submit_bio_wait(bio); | |
ac10a961 | 820 | kfree(bio); |
1c6fdbd8 KO |
821 | |
822 | if (bch2_dev_io_err_on(ret, ca, | |
0fefe8d8 | 823 | "journal read error: sector %llu", |
1c6fdbd8 | 824 | offset) || |
29d90f61 KO |
825 | bch2_meta_read_fault("journal")) { |
826 | /* | |
827 | * We don't error out of the recovery process | |
828 | * here, since the relevant journal entry may be | |
829 | * found on a different device, and missing or | |
830 | * no journal entries will be handled later | |
831 | */ | |
832 | return 0; | |
833 | } | |
1c6fdbd8 KO |
834 | |
835 | j = buf->data; | |
836 | } | |
837 | ||
ca73852a | 838 | ret = jset_validate(c, ca, j, offset, |
1c6fdbd8 KO |
839 | end - offset, sectors_read, |
840 | READ); | |
841 | switch (ret) { | |
842 | case BCH_FSCK_OK: | |
ca73852a | 843 | sectors = vstruct_sectors(j, c->block_bits); |
1c6fdbd8 KO |
844 | break; |
845 | case JOURNAL_ENTRY_REREAD: | |
846 | if (vstruct_bytes(j) > buf->size) { | |
847 | ret = journal_read_buf_realloc(buf, | |
848 | vstruct_bytes(j)); | |
849 | if (ret) | |
850 | return ret; | |
851 | } | |
852 | goto reread; | |
853 | case JOURNAL_ENTRY_NONE: | |
854 | if (!saw_bad) | |
855 | return 0; | |
8244f320 | 856 | sectors = block_sectors(c); |
1c6fdbd8 KO |
857 | goto next_block; |
858 | case JOURNAL_ENTRY_BAD: | |
859 | saw_bad = true; | |
ca73852a KO |
860 | /* |
861 | * On checksum error we don't really trust the size | |
862 | * field of the journal entry we read, so try reading | |
863 | * again at next block boundary: | |
864 | */ | |
8244f320 | 865 | sectors = block_sectors(c); |
ca73852a | 866 | break; |
1c6fdbd8 KO |
867 | default: |
868 | return ret; | |
869 | } | |
870 | ||
871 | /* | |
872 | * This happens sometimes if we don't have discards on - | |
873 | * when we've partially overwritten a bucket with new | |
874 | * journal entries. We don't need the rest of the | |
875 | * bucket: | |
876 | */ | |
877 | if (le64_to_cpu(j->seq) < ja->bucket_seq[bucket]) | |
878 | return 0; | |
879 | ||
880 | ja->bucket_seq[bucket] = le64_to_cpu(j->seq); | |
881 | ||
882 | mutex_lock(&jlist->lock); | |
72b7d633 KO |
883 | ret = journal_entry_add(c, ca, (struct journal_ptr) { |
884 | .dev = ca->dev_idx, | |
885 | .bucket = bucket, | |
886 | .bucket_offset = offset - | |
887 | bucket_to_sector(ca, ja->buckets[bucket]), | |
888 | .sector = offset, | |
e4c3f386 | 889 | }, jlist, j, ret != 0); |
1c6fdbd8 KO |
890 | mutex_unlock(&jlist->lock); |
891 | ||
892 | switch (ret) { | |
893 | case JOURNAL_ENTRY_ADD_OK: | |
1c6fdbd8 KO |
894 | break; |
895 | case JOURNAL_ENTRY_ADD_OUT_OF_RANGE: | |
896 | break; | |
897 | default: | |
898 | return ret; | |
899 | } | |
1c6fdbd8 KO |
900 | next_block: |
901 | pr_debug("next"); | |
902 | offset += sectors; | |
903 | sectors_read -= sectors; | |
904 | j = ((void *) j) + (sectors << 9); | |
905 | } | |
906 | ||
907 | return 0; | |
908 | } | |
909 | ||
910 | static void bch2_journal_read_device(struct closure *cl) | |
911 | { | |
1c6fdbd8 KO |
912 | struct journal_device *ja = |
913 | container_of(cl, struct journal_device, read); | |
914 | struct bch_dev *ca = container_of(ja, struct bch_dev, journal); | |
365f64f3 | 915 | struct bch_fs *c = ca->fs; |
1c6fdbd8 KO |
916 | struct journal_list *jlist = |
917 | container_of(cl->parent, struct journal_list, cl); | |
062afcba | 918 | struct journal_replay *r; |
1c6fdbd8 | 919 | struct journal_read_buf buf = { NULL, 0 }; |
a9ec3454 KO |
920 | u64 min_seq = U64_MAX; |
921 | unsigned i; | |
9714baaa | 922 | int ret = 0; |
1c6fdbd8 KO |
923 | |
924 | if (!ja->nr) | |
925 | goto out; | |
926 | ||
1c6fdbd8 KO |
927 | ret = journal_read_buf_realloc(&buf, PAGE_SIZE); |
928 | if (ret) | |
929 | goto err; | |
930 | ||
931 | pr_debug("%u journal buckets", ja->nr); | |
932 | ||
1c6fdbd8 | 933 | for (i = 0; i < ja->nr; i++) { |
a9ec3454 KO |
934 | ret = journal_read_bucket(ca, &buf, jlist, i); |
935 | if (ret) | |
936 | goto err; | |
1c6fdbd8 KO |
937 | } |
938 | ||
a9ec3454 KO |
939 | /* Find the journal bucket with the highest sequence number: */ |
940 | for (i = 0; i < ja->nr; i++) { | |
941 | if (ja->bucket_seq[i] > ja->bucket_seq[ja->cur_idx]) | |
942 | ja->cur_idx = i; | |
1c6fdbd8 | 943 | |
a9ec3454 | 944 | min_seq = min(ja->bucket_seq[i], min_seq); |
1c6fdbd8 KO |
945 | } |
946 | ||
1c6fdbd8 | 947 | /* |
1c6fdbd8 KO |
948 | * If there's duplicate journal entries in multiple buckets (which |
949 | * definitely isn't supposed to happen, but...) - make sure to start | |
950 | * cur_idx at the last of those buckets, so we don't deadlock trying to | |
951 | * allocate | |
952 | */ | |
a9ec3454 | 953 | while (ja->bucket_seq[ja->cur_idx] > min_seq && |
062afcba | 954 | ja->bucket_seq[ja->cur_idx] == |
a9ec3454 | 955 | ja->bucket_seq[(ja->cur_idx + 1) % ja->nr]) |
a36d3685 | 956 | ja->cur_idx = (ja->cur_idx + 1) % ja->nr; |
a9ec3454 | 957 | |
062afcba KO |
958 | ja->sectors_free = ca->mi.bucket_size; |
959 | ||
960 | mutex_lock(&jlist->lock); | |
961 | list_for_each_entry(r, jlist->head, list) { | |
962 | for (i = 0; i < r->nr_ptrs; i++) { | |
963 | if (r->ptrs[i].dev == ca->dev_idx && | |
964 | sector_to_bucket(ca, r->ptrs[i].sector) == ja->buckets[ja->cur_idx]) { | |
965 | unsigned wrote = (r->ptrs[i].sector % ca->mi.bucket_size) + | |
966 | vstruct_sectors(&r->j, c->block_bits); | |
967 | ||
968 | ja->sectors_free = min(ja->sectors_free, | |
969 | ca->mi.bucket_size - wrote); | |
970 | } | |
971 | } | |
972 | } | |
973 | mutex_unlock(&jlist->lock); | |
974 | ||
b0be2fcf KO |
975 | if (ja->bucket_seq[ja->cur_idx] && |
976 | ja->sectors_free == ca->mi.bucket_size) { | |
977 | bch_err(c, "ja->sectors_free == ca->mi.bucket_size"); | |
978 | bch_err(c, "cur_idx %u/%u", ja->cur_idx, ja->nr); | |
979 | for (i = 0; i < 3; i++) { | |
980 | unsigned idx = (ja->cur_idx + ja->nr - 1 + i) % ja->nr; | |
981 | bch_err(c, "bucket_seq[%u] = %llu", idx, ja->bucket_seq[idx]); | |
982 | } | |
983 | ja->sectors_free = 0; | |
984 | } | |
1c6fdbd8 KO |
985 | |
986 | /* | |
0ce2dbbe | 987 | * Set dirty_idx to indicate the entire journal is full and needs to be |
1c6fdbd8 KO |
988 | * reclaimed - journal reclaim will immediately reclaim whatever isn't |
989 | * pinned when it first runs: | |
990 | */ | |
0ce2dbbe KO |
991 | ja->discard_idx = ja->dirty_idx_ondisk = |
992 | ja->dirty_idx = (ja->cur_idx + 1) % ja->nr; | |
1c6fdbd8 | 993 | out: |
365f64f3 | 994 | bch_verbose(c, "journal read done on device %s, ret %i", ca->name, ret); |
1c6fdbd8 | 995 | kvpfree(buf.data, buf.size); |
1c6fdbd8 KO |
996 | percpu_ref_put(&ca->io_ref); |
997 | closure_return(cl); | |
998 | return; | |
999 | err: | |
1000 | mutex_lock(&jlist->lock); | |
1001 | jlist->ret = ret; | |
1002 | mutex_unlock(&jlist->lock); | |
1003 | goto out; | |
1c6fdbd8 KO |
1004 | } |
1005 | ||
72b7d633 KO |
1006 | void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c, |
1007 | struct journal_replay *j) | |
e4c3f386 KO |
1008 | { |
1009 | unsigned i; | |
1010 | ||
1011 | for (i = 0; i < j->nr_ptrs; i++) { | |
c0ebe3e4 | 1012 | struct bch_dev *ca = bch_dev_bkey_exists(c, j->ptrs[i].dev); |
514852c2 KO |
1013 | u64 offset; |
1014 | ||
72b7d633 | 1015 | div64_u64_rem(j->ptrs[i].sector, ca->mi.bucket_size, &offset); |
e4c3f386 KO |
1016 | |
1017 | if (i) | |
1018 | pr_buf(out, " "); | |
72b7d633 | 1019 | pr_buf(out, "%u:%u:%u (sector %llu)", |
e4c3f386 | 1020 | j->ptrs[i].dev, |
72b7d633 KO |
1021 | j->ptrs[i].bucket, |
1022 | j->ptrs[i].bucket_offset, | |
1023 | j->ptrs[i].sector); | |
e4c3f386 KO |
1024 | } |
1025 | } | |
1026 | ||
adbcada4 KO |
1027 | int bch2_journal_read(struct bch_fs *c, struct list_head *list, |
1028 | u64 *blacklist_seq, u64 *start_seq) | |
1c6fdbd8 | 1029 | { |
1c6fdbd8 | 1030 | struct journal_list jlist; |
adbcada4 | 1031 | struct journal_replay *i, *t; |
1c6fdbd8 | 1032 | struct bch_dev *ca; |
1c6fdbd8 | 1033 | unsigned iter; |
fa8e94fa | 1034 | struct printbuf buf = PRINTBUF; |
1c6fdbd8 KO |
1035 | size_t keys = 0, entries = 0; |
1036 | bool degraded = false; | |
adbcada4 | 1037 | u64 seq, last_seq = 0; |
1c6fdbd8 KO |
1038 | int ret = 0; |
1039 | ||
1040 | closure_init_stack(&jlist.cl); | |
1041 | mutex_init(&jlist.lock); | |
1042 | jlist.head = list; | |
1043 | jlist.ret = 0; | |
1044 | ||
1045 | for_each_member_device(ca, c, iter) { | |
6bdbfa87 | 1046 | if (!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) && |
89fd25be | 1047 | !(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_journal))) |
1c6fdbd8 KO |
1048 | continue; |
1049 | ||
2436cb9f KO |
1050 | if ((ca->mi.state == BCH_MEMBER_STATE_rw || |
1051 | ca->mi.state == BCH_MEMBER_STATE_ro) && | |
1c6fdbd8 KO |
1052 | percpu_ref_tryget(&ca->io_ref)) |
1053 | closure_call(&ca->journal.read, | |
1054 | bch2_journal_read_device, | |
1055 | system_unbound_wq, | |
1056 | &jlist.cl); | |
1057 | else | |
1058 | degraded = true; | |
1059 | } | |
1060 | ||
1061 | closure_sync(&jlist.cl); | |
1062 | ||
1063 | if (jlist.ret) | |
1064 | return jlist.ret; | |
1065 | ||
adbcada4 KO |
1066 | if (list_empty(list)) { |
1067 | bch_info(c, "journal read done, but no entries found"); | |
1068 | return 0; | |
1069 | } | |
1070 | ||
1071 | i = list_last_entry(list, struct journal_replay, list); | |
1072 | *start_seq = le64_to_cpu(i->j.seq) + 1; | |
1073 | ||
1074 | /* | |
1075 | * Find most recent flush entry, and ignore newer non flush entries - | |
1076 | * those entries will be blacklisted: | |
1077 | */ | |
1078 | list_for_each_entry_safe_reverse(i, t, list, list) { | |
1079 | if (i->ignore) | |
1080 | continue; | |
1081 | ||
1082 | if (!JSET_NO_FLUSH(&i->j)) { | |
1083 | last_seq = le64_to_cpu(i->j.last_seq); | |
1084 | *blacklist_seq = le64_to_cpu(i->j.seq) + 1; | |
1085 | break; | |
1086 | } | |
1087 | ||
1088 | journal_replay_free(c, i); | |
1089 | } | |
1090 | ||
1091 | if (!last_seq) { | |
1092 | fsck_err(c, "journal read done, but no entries found after dropping non-flushes"); | |
fa8e94fa KO |
1093 | ret = -1; |
1094 | goto err; | |
adbcada4 KO |
1095 | } |
1096 | ||
1097 | /* Drop blacklisted entries and entries older than last_seq: */ | |
1098 | list_for_each_entry_safe(i, t, list, list) { | |
1099 | if (i->ignore) | |
1100 | continue; | |
1101 | ||
1102 | seq = le64_to_cpu(i->j.seq); | |
1103 | if (seq < last_seq) { | |
1104 | journal_replay_free(c, i); | |
1105 | continue; | |
1106 | } | |
1107 | ||
1108 | if (bch2_journal_seq_is_blacklisted(c, seq, true)) { | |
1109 | fsck_err_on(!JSET_NO_FLUSH(&i->j), c, | |
1110 | "found blacklisted journal entry %llu", seq); | |
1111 | ||
1112 | journal_replay_free(c, i); | |
1113 | } | |
1114 | } | |
1115 | ||
1116 | /* Check for missing entries: */ | |
1117 | seq = last_seq; | |
1118 | list_for_each_entry(i, list, list) { | |
1119 | if (i->ignore) | |
1120 | continue; | |
1121 | ||
1122 | BUG_ON(seq > le64_to_cpu(i->j.seq)); | |
1123 | ||
1124 | while (seq < le64_to_cpu(i->j.seq)) { | |
1125 | u64 missing_start, missing_end; | |
fa8e94fa | 1126 | struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF; |
adbcada4 KO |
1127 | |
1128 | while (seq < le64_to_cpu(i->j.seq) && | |
1129 | bch2_journal_seq_is_blacklisted(c, seq, false)) | |
1130 | seq++; | |
1131 | ||
1132 | if (seq == le64_to_cpu(i->j.seq)) | |
1133 | break; | |
1134 | ||
1135 | missing_start = seq; | |
1136 | ||
1137 | while (seq < le64_to_cpu(i->j.seq) && | |
1138 | !bch2_journal_seq_is_blacklisted(c, seq, false)) | |
1139 | seq++; | |
1140 | ||
e4c3f386 | 1141 | if (i->list.prev != list) { |
e4c3f386 KO |
1142 | struct journal_replay *p = list_prev_entry(i, list); |
1143 | ||
fa8e94fa KO |
1144 | bch2_journal_ptrs_to_text(&buf1, c, p); |
1145 | pr_buf(&buf1, " size %zu", vstruct_sectors(&p->j, c->block_bits)); | |
e4c3f386 | 1146 | } else |
fa8e94fa KO |
1147 | pr_buf(&buf1, "(none)"); |
1148 | bch2_journal_ptrs_to_text(&buf2, c, i); | |
e4c3f386 | 1149 | |
adbcada4 | 1150 | missing_end = seq - 1; |
e4c3f386 KO |
1151 | fsck_err(c, "journal entries %llu-%llu missing! (replaying %llu-%llu)\n" |
1152 | " prev at %s\n" | |
1153 | " next at %s", | |
adbcada4 | 1154 | missing_start, missing_end, |
e4c3f386 | 1155 | last_seq, *blacklist_seq - 1, |
fa8e94fa KO |
1156 | buf1.buf, buf2.buf); |
1157 | ||
1158 | printbuf_exit(&buf1); | |
1159 | printbuf_exit(&buf2); | |
adbcada4 KO |
1160 | } |
1161 | ||
1162 | seq++; | |
1163 | } | |
1164 | ||
1c6fdbd8 | 1165 | list_for_each_entry(i, list, list) { |
1dd7f9d9 KO |
1166 | struct jset_entry *entry; |
1167 | struct bkey_i *k, *_n; | |
e4c3f386 KO |
1168 | struct bch_replicas_padded replicas = { |
1169 | .e.data_type = BCH_DATA_journal, | |
1170 | .e.nr_required = 1, | |
1171 | }; | |
1172 | unsigned ptr; | |
7ef2a73a | 1173 | |
adbcada4 KO |
1174 | if (i->ignore) |
1175 | continue; | |
1176 | ||
1c6fdbd8 KO |
1177 | ret = jset_validate_entries(c, &i->j, READ); |
1178 | if (ret) | |
fa8e94fa | 1179 | goto err; |
1c6fdbd8 | 1180 | |
e4c3f386 KO |
1181 | for (ptr = 0; ptr < i->nr_ptrs; ptr++) |
1182 | replicas.e.devs[replicas.e.nr_devs++] = i->ptrs[ptr].dev; | |
1183 | ||
26452d1d KO |
1184 | bch2_replicas_entry_sort(&replicas.e); |
1185 | ||
1c6fdbd8 KO |
1186 | /* |
1187 | * If we're mounting in degraded mode - if we didn't read all | |
1188 | * the devices - this is wrong: | |
1189 | */ | |
1190 | ||
fa8e94fa KO |
1191 | printbuf_reset(&buf); |
1192 | bch2_replicas_entry_to_text(&buf, &replicas.e); | |
1193 | ||
1c6fdbd8 KO |
1194 | if (!degraded && |
1195 | (test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) || | |
988e98cf | 1196 | fsck_err_on(!bch2_replicas_marked(c, &replicas.e), c, |
7ef2a73a | 1197 | "superblock not marked as containing replicas %s", |
fa8e94fa | 1198 | buf.buf))) { |
7ef2a73a | 1199 | ret = bch2_mark_replicas(c, &replicas.e); |
1c6fdbd8 | 1200 | if (ret) |
fa8e94fa | 1201 | goto err; |
1c6fdbd8 | 1202 | } |
1c6fdbd8 KO |
1203 | |
1204 | for_each_jset_key(k, _n, entry, &i->j) | |
1205 | keys++; | |
1206 | entries++; | |
1207 | } | |
1208 | ||
adbcada4 KO |
1209 | bch_info(c, "journal read done, %zu keys in %zu entries, seq %llu", |
1210 | keys, entries, *start_seq); | |
1dd7f9d9 | 1211 | |
adbcada4 KO |
1212 | if (*start_seq != *blacklist_seq) |
1213 | bch_info(c, "dropped unflushed entries %llu-%llu", | |
1214 | *blacklist_seq, *start_seq - 1); | |
fa8e94fa | 1215 | err: |
1c6fdbd8 | 1216 | fsck_err: |
fa8e94fa | 1217 | printbuf_exit(&buf); |
1c6fdbd8 KO |
1218 | return ret; |
1219 | } | |
1220 | ||
1c6fdbd8 KO |
1221 | /* journal write: */ |
1222 | ||
a9ec3454 KO |
1223 | static void __journal_write_alloc(struct journal *j, |
1224 | struct journal_buf *w, | |
1225 | struct dev_alloc_list *devs_sorted, | |
1226 | unsigned sectors, | |
1227 | unsigned *replicas, | |
1228 | unsigned replicas_want) | |
1c6fdbd8 KO |
1229 | { |
1230 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
1c6fdbd8 KO |
1231 | struct journal_device *ja; |
1232 | struct bch_dev *ca; | |
a9ec3454 | 1233 | unsigned i; |
a2753581 | 1234 | |
a9ec3454 KO |
1235 | if (*replicas >= replicas_want) |
1236 | return; | |
1c6fdbd8 | 1237 | |
a9ec3454 KO |
1238 | for (i = 0; i < devs_sorted->nr; i++) { |
1239 | ca = rcu_dereference(c->devs[devs_sorted->devs[i]]); | |
1c6fdbd8 KO |
1240 | if (!ca) |
1241 | continue; | |
1242 | ||
1c6fdbd8 | 1243 | ja = &ca->journal; |
1c6fdbd8 KO |
1244 | |
1245 | /* | |
1246 | * Check that we can use this device, and aren't already using | |
1247 | * it: | |
1248 | */ | |
a9ec3454 | 1249 | if (!ca->mi.durability || |
2436cb9f | 1250 | ca->mi.state != BCH_MEMBER_STATE_rw || |
a9ec3454 | 1251 | !ja->nr || |
26609b61 KO |
1252 | bch2_bkey_has_device(bkey_i_to_s_c(&w->key), |
1253 | ca->dev_idx) || | |
a9ec3454 | 1254 | sectors > ja->sectors_free) |
1c6fdbd8 KO |
1255 | continue; |
1256 | ||
3d080aa5 | 1257 | bch2_dev_stripe_increment(ca, &j->wp.stripe); |
1c6fdbd8 | 1258 | |
26609b61 | 1259 | bch2_bkey_append_ptr(&w->key, |
1c6fdbd8 KO |
1260 | (struct bch_extent_ptr) { |
1261 | .offset = bucket_to_sector(ca, | |
a9ec3454 KO |
1262 | ja->buckets[ja->cur_idx]) + |
1263 | ca->mi.bucket_size - | |
1264 | ja->sectors_free, | |
1c6fdbd8 KO |
1265 | .dev = ca->dev_idx, |
1266 | }); | |
1267 | ||
a9ec3454 KO |
1268 | ja->sectors_free -= sectors; |
1269 | ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq); | |
1270 | ||
1271 | *replicas += ca->mi.durability; | |
1272 | ||
1273 | if (*replicas >= replicas_want) | |
1274 | break; | |
1c6fdbd8 | 1275 | } |
a9ec3454 | 1276 | } |
1c6fdbd8 | 1277 | |
a9ec3454 KO |
1278 | /** |
1279 | * journal_next_bucket - move on to the next journal bucket if possible | |
1280 | */ | |
1281 | static int journal_write_alloc(struct journal *j, struct journal_buf *w, | |
1282 | unsigned sectors) | |
1283 | { | |
1284 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
d042b040 | 1285 | struct bch_devs_mask devs; |
a9ec3454 KO |
1286 | struct journal_device *ja; |
1287 | struct bch_dev *ca; | |
1288 | struct dev_alloc_list devs_sorted; | |
d042b040 KO |
1289 | unsigned target = c->opts.metadata_target ?: |
1290 | c->opts.foreground_target; | |
a9ec3454 KO |
1291 | unsigned i, replicas = 0, replicas_want = |
1292 | READ_ONCE(c->opts.metadata_replicas); | |
1c6fdbd8 | 1293 | |
a9ec3454 | 1294 | rcu_read_lock(); |
d042b040 KO |
1295 | retry: |
1296 | devs = target_rw_devs(c, BCH_DATA_journal, target); | |
1c6fdbd8 | 1297 | |
d042b040 | 1298 | devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe, &devs); |
1c6fdbd8 | 1299 | |
a9ec3454 KO |
1300 | __journal_write_alloc(j, w, &devs_sorted, |
1301 | sectors, &replicas, replicas_want); | |
1c6fdbd8 | 1302 | |
a9ec3454 KO |
1303 | if (replicas >= replicas_want) |
1304 | goto done; | |
1305 | ||
1306 | for (i = 0; i < devs_sorted.nr; i++) { | |
1307 | ca = rcu_dereference(c->devs[devs_sorted.devs[i]]); | |
1308 | if (!ca) | |
1309 | continue; | |
1310 | ||
1311 | ja = &ca->journal; | |
1312 | ||
1313 | if (sectors > ja->sectors_free && | |
1314 | sectors <= ca->mi.bucket_size && | |
03d5eaed KO |
1315 | bch2_journal_dev_buckets_available(j, ja, |
1316 | journal_space_discarded)) { | |
a9ec3454 KO |
1317 | ja->cur_idx = (ja->cur_idx + 1) % ja->nr; |
1318 | ja->sectors_free = ca->mi.bucket_size; | |
68ef94a6 KO |
1319 | |
1320 | /* | |
1321 | * ja->bucket_seq[ja->cur_idx] must always have | |
1322 | * something sensible: | |
1323 | */ | |
1324 | ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq); | |
a9ec3454 KO |
1325 | } |
1326 | } | |
1327 | ||
1328 | __journal_write_alloc(j, w, &devs_sorted, | |
1329 | sectors, &replicas, replicas_want); | |
d042b040 KO |
1330 | |
1331 | if (replicas < replicas_want && target) { | |
1332 | /* Retry from all devices: */ | |
1333 | target = 0; | |
1334 | goto retry; | |
1335 | } | |
a9ec3454 | 1336 | done: |
a9ec3454 KO |
1337 | rcu_read_unlock(); |
1338 | ||
07a1006a KO |
1339 | BUG_ON(bkey_val_u64s(&w->key.k) > BCH_REPLICAS_MAX); |
1340 | ||
57cb2142 | 1341 | return replicas >= c->opts.metadata_replicas_required ? 0 : -EROFS; |
1c6fdbd8 KO |
1342 | } |
1343 | ||
1c6fdbd8 KO |
1344 | static void journal_buf_realloc(struct journal *j, struct journal_buf *buf) |
1345 | { | |
1346 | /* we aren't holding j->lock: */ | |
1347 | unsigned new_size = READ_ONCE(j->buf_size_want); | |
1348 | void *new_buf; | |
1349 | ||
d16b4a77 | 1350 | if (buf->buf_size >= new_size) |
1c6fdbd8 KO |
1351 | return; |
1352 | ||
1353 | new_buf = kvpmalloc(new_size, GFP_NOIO|__GFP_NOWARN); | |
1354 | if (!new_buf) | |
1355 | return; | |
1356 | ||
d16b4a77 | 1357 | memcpy(new_buf, buf->data, buf->buf_size); |
c859430b KO |
1358 | |
1359 | spin_lock(&j->lock); | |
1360 | swap(buf->data, new_buf); | |
1361 | swap(buf->buf_size, new_size); | |
1362 | spin_unlock(&j->lock); | |
1363 | ||
1364 | kvpfree(new_buf, new_size); | |
1c6fdbd8 KO |
1365 | } |
1366 | ||
ebb84d09 KO |
1367 | static inline struct journal_buf *journal_last_unwritten_buf(struct journal *j) |
1368 | { | |
30ef633a | 1369 | return j->buf + (journal_last_unwritten_seq(j) & JOURNAL_BUF_MASK); |
ebb84d09 KO |
1370 | } |
1371 | ||
1c6fdbd8 KO |
1372 | static void journal_write_done(struct closure *cl) |
1373 | { | |
1374 | struct journal *j = container_of(cl, struct journal, io); | |
1375 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
ebb84d09 | 1376 | struct journal_buf *w = journal_last_unwritten_buf(j); |
7ef2a73a | 1377 | struct bch_replicas_padded replicas; |
ebb84d09 | 1378 | union journal_res_state old, new; |
1784d43a | 1379 | u64 v, seq; |
158eecb8 | 1380 | int err = 0; |
1c6fdbd8 | 1381 | |
991ba021 KO |
1382 | bch2_time_stats_update(!JSET_NO_FLUSH(w->data) |
1383 | ? j->flush_write_time | |
1384 | : j->noflush_write_time, j->write_start_time); | |
9c859dc9 | 1385 | |
d797ca3d | 1386 | if (!w->devs_written.nr) { |
1c6fdbd8 | 1387 | bch_err(c, "unable to write journal to sufficient devices"); |
158eecb8 KO |
1388 | err = -EIO; |
1389 | } else { | |
d797ca3d KO |
1390 | bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal, |
1391 | w->devs_written); | |
158eecb8 KO |
1392 | if (bch2_mark_replicas(c, &replicas.e)) |
1393 | err = -EIO; | |
1c6fdbd8 KO |
1394 | } |
1395 | ||
158eecb8 KO |
1396 | if (err) |
1397 | bch2_fatal_error(c); | |
1c6fdbd8 KO |
1398 | |
1399 | spin_lock(&j->lock); | |
ed9d58a2 | 1400 | seq = le64_to_cpu(w->data->seq); |
ed9d58a2 | 1401 | |
1c6fdbd8 | 1402 | if (seq >= j->pin.front) |
d797ca3d | 1403 | journal_seq_pin(j, seq)->devs = w->devs_written; |
1c6fdbd8 | 1404 | |
9be1efe9 | 1405 | if (!err) { |
9be1efe9 KO |
1406 | if (!JSET_NO_FLUSH(w->data)) { |
1407 | j->flushed_seq_ondisk = seq; | |
1408 | j->last_seq_ondisk = w->last_seq; | |
f25d8215 | 1409 | |
59cc38b8 | 1410 | bch2_do_discards(c); |
f25d8215 KO |
1411 | closure_wake_up(&c->freelist_wait); |
1412 | ||
1413 | bch2_reset_alloc_cursors(c); | |
9be1efe9 KO |
1414 | } |
1415 | } else if (!j->err_seq || seq < j->err_seq) | |
1416 | j->err_seq = seq; | |
0ce2dbbe | 1417 | |
f0a3a2cc KO |
1418 | j->seq_ondisk = seq; |
1419 | ||
1c6fdbd8 KO |
1420 | /* |
1421 | * Updating last_seq_ondisk may let bch2_journal_reclaim_work() discard | |
1422 | * more buckets: | |
1423 | * | |
1424 | * Must come before signaling write completion, for | |
1425 | * bch2_fs_journal_stop(): | |
1426 | */ | |
b7a9bbfc | 1427 | journal_reclaim_kick(&c->journal); |
158eecb8 | 1428 | |
1c6fdbd8 KO |
1429 | /* also must come before signalling write completion: */ |
1430 | closure_debug_destroy(cl); | |
1431 | ||
ebb84d09 KO |
1432 | v = atomic64_read(&j->reservations.counter); |
1433 | do { | |
1434 | old.v = new.v = v; | |
24a3d53b | 1435 | BUG_ON(journal_state_count(new, new.unwritten_idx)); |
ebb84d09 KO |
1436 | |
1437 | new.unwritten_idx++; | |
1438 | } while ((v = atomic64_cmpxchg(&j->reservations.counter, | |
1439 | old.v, new.v)) != old.v); | |
1c6fdbd8 | 1440 | |
5d32c5bb KO |
1441 | bch2_journal_space_available(j); |
1442 | ||
1c6fdbd8 KO |
1443 | closure_wake_up(&w->wait); |
1444 | journal_wake(j); | |
1445 | ||
24a3d53b KO |
1446 | if (!journal_state_count(new, new.unwritten_idx) && |
1447 | journal_last_unwritten_seq(j) <= journal_cur_seq(j)) { | |
1448 | closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL); | |
1449 | } else if (journal_last_unwritten_seq(j) == journal_cur_seq(j) && | |
1450 | new.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL) { | |
fbec3b88 KO |
1451 | struct journal_buf *buf = journal_cur_buf(j); |
1452 | long delta = buf->expires - jiffies; | |
ebb84d09 | 1453 | |
24a3d53b KO |
1454 | /* |
1455 | * We don't close a journal entry to write it while there's | |
1456 | * previous entries still in flight - the current journal entry | |
1457 | * might want to be written now: | |
1458 | */ | |
1459 | ||
fbec3b88 | 1460 | mod_delayed_work(c->io_complete_wq, &j->write_work, max(0L, delta)); |
24a3d53b | 1461 | } |
fbec3b88 KO |
1462 | |
1463 | spin_unlock(&j->lock); | |
1c6fdbd8 KO |
1464 | } |
1465 | ||
1466 | static void journal_write_endio(struct bio *bio) | |
1467 | { | |
1468 | struct bch_dev *ca = bio->bi_private; | |
1469 | struct journal *j = &ca->fs->journal; | |
d797ca3d KO |
1470 | struct journal_buf *w = journal_last_unwritten_buf(j); |
1471 | unsigned long flags; | |
1c6fdbd8 | 1472 | |
d797ca3d KO |
1473 | if (bch2_dev_io_err_on(bio->bi_status, ca, "error writing journal entry %llu: %s", |
1474 | le64_to_cpu(w->data->seq), | |
63b214e7 | 1475 | bch2_blk_status_to_str(bio->bi_status)) || |
1c6fdbd8 | 1476 | bch2_meta_write_fault("journal")) { |
1c6fdbd8 | 1477 | spin_lock_irqsave(&j->err_lock, flags); |
d797ca3d | 1478 | bch2_dev_list_drop_dev(&w->devs_written, ca->dev_idx); |
1c6fdbd8 KO |
1479 | spin_unlock_irqrestore(&j->err_lock, flags); |
1480 | } | |
1481 | ||
1482 | closure_put(&j->io); | |
1483 | percpu_ref_put(&ca->io_ref); | |
1484 | } | |
1485 | ||
280249b9 KO |
1486 | static void do_journal_write(struct closure *cl) |
1487 | { | |
1488 | struct journal *j = container_of(cl, struct journal, io); | |
1489 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
1490 | struct bch_dev *ca; | |
1491 | struct journal_buf *w = journal_last_unwritten_buf(j); | |
1492 | struct bch_extent_ptr *ptr; | |
1493 | struct bio *bio; | |
1494 | unsigned sectors = vstruct_sectors(w->data, c->block_bits); | |
1495 | ||
1496 | extent_for_each_ptr(bkey_i_to_s_extent(&w->key), ptr) { | |
1497 | ca = bch_dev_bkey_exists(c, ptr->dev); | |
1498 | if (!percpu_ref_tryget(&ca->io_ref)) { | |
1499 | /* XXX: fix this */ | |
1500 | bch_err(c, "missing device for journal write\n"); | |
1501 | continue; | |
1502 | } | |
1503 | ||
1504 | this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_journal], | |
1505 | sectors); | |
1506 | ||
1507 | bio = ca->journal.bio; | |
1508 | bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META); | |
1509 | bio->bi_iter.bi_sector = ptr->offset; | |
1510 | bio->bi_end_io = journal_write_endio; | |
1511 | bio->bi_private = ca; | |
1512 | ||
a28bd48a KO |
1513 | BUG_ON(bio->bi_iter.bi_sector == ca->prev_journal_sector); |
1514 | ca->prev_journal_sector = bio->bi_iter.bi_sector; | |
1515 | ||
280249b9 KO |
1516 | if (!JSET_NO_FLUSH(w->data)) |
1517 | bio->bi_opf |= REQ_FUA; | |
1518 | if (!JSET_NO_FLUSH(w->data) && !w->separate_flush) | |
1519 | bio->bi_opf |= REQ_PREFLUSH; | |
1520 | ||
1521 | bch2_bio_map(bio, w->data, sectors << 9); | |
1522 | ||
1523 | trace_journal_write(bio); | |
1524 | closure_bio_submit(bio, cl); | |
1525 | ||
1526 | ca->journal.bucket_seq[ca->journal.cur_idx] = | |
1527 | le64_to_cpu(w->data->seq); | |
1528 | } | |
1529 | ||
731bdd2e | 1530 | continue_at(cl, journal_write_done, c->io_complete_wq); |
280249b9 KO |
1531 | return; |
1532 | } | |
1533 | ||
1c6fdbd8 KO |
1534 | void bch2_journal_write(struct closure *cl) |
1535 | { | |
1536 | struct journal *j = container_of(cl, struct journal, io); | |
1537 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
1538 | struct bch_dev *ca; | |
ebb84d09 | 1539 | struct journal_buf *w = journal_last_unwritten_buf(j); |
3ccc5c50 | 1540 | struct jset_entry *start, *end; |
1c6fdbd8 KO |
1541 | struct jset *jset; |
1542 | struct bio *bio; | |
fa8e94fa | 1543 | struct printbuf journal_debug_buf = PRINTBUF; |
26609b61 | 1544 | bool validate_before_checksum = false; |
280249b9 | 1545 | unsigned i, sectors, bytes, u64s, nr_rw_members = 0; |
e5a66496 KO |
1546 | int ret; |
1547 | ||
b7a9bbfc KO |
1548 | BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb)); |
1549 | ||
1c6fdbd8 KO |
1550 | journal_buf_realloc(j, w); |
1551 | jset = w->data; | |
1552 | ||
1553 | j->write_start_time = local_clock(); | |
1c6fdbd8 | 1554 | |
adbcada4 | 1555 | spin_lock(&j->lock); |
e0c014e7 KO |
1556 | if (bch2_journal_error(j) || |
1557 | w->noflush || | |
1558 | (!w->must_flush && | |
1559 | (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) && | |
1560 | test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags))) { | |
adbcada4 KO |
1561 | w->noflush = true; |
1562 | SET_JSET_NO_FLUSH(jset, true); | |
c0ebe3e4 KO |
1563 | jset->last_seq = 0; |
1564 | w->last_seq = 0; | |
adbcada4 KO |
1565 | |
1566 | j->nr_noflush_writes++; | |
1567 | } else { | |
1568 | j->last_flush_write = jiffies; | |
1569 | j->nr_flush_writes++; | |
1570 | } | |
1571 | spin_unlock(&j->lock); | |
1572 | ||
00b8ccf7 KO |
1573 | /* |
1574 | * New btree roots are set by journalling them; when the journal entry | |
1575 | * gets written we have to propagate them to c->btree_roots | |
1576 | * | |
1577 | * But, every journal entry we write has to contain all the btree roots | |
1578 | * (at least for now); so after we copy btree roots to c->btree_roots we | |
1579 | * have to get any missing btree roots and add them to this journal | |
1580 | * entry: | |
1581 | */ | |
1582 | ||
1583 | bch2_journal_entries_to_btree_roots(c, jset); | |
1584 | ||
1585 | start = end = vstruct_last(jset); | |
1586 | ||
1587 | end = bch2_btree_roots_to_journal_entries(c, jset->start, end); | |
1588 | ||
2abe5420 KO |
1589 | bch2_journal_super_entries_add_common(c, &end, |
1590 | le64_to_cpu(jset->seq)); | |
3ccc5c50 KO |
1591 | u64s = (u64 *) end - (u64 *) start; |
1592 | BUG_ON(u64s > j->entry_u64s_reserved); | |
1593 | ||
d16b4a77 KO |
1594 | le32_add_cpu(&jset->u64s, u64s); |
1595 | BUG_ON(vstruct_sectors(jset, c->block_bits) > w->sectors); | |
1c6fdbd8 | 1596 | |
1c6fdbd8 | 1597 | jset->magic = cpu_to_le64(jset_magic(c)); |
74b33393 | 1598 | jset->version = c->sb.version < bcachefs_metadata_version_bkey_renumber |
26609b61 KO |
1599 | ? cpu_to_le32(BCH_JSET_VERSION_OLD) |
1600 | : cpu_to_le32(c->sb.version); | |
1c6fdbd8 KO |
1601 | |
1602 | SET_JSET_BIG_ENDIAN(jset, CPU_BIG_ENDIAN); | |
1603 | SET_JSET_CSUM_TYPE(jset, bch2_meta_checksum_type(c)); | |
1604 | ||
4141fde0 | 1605 | if (!JSET_NO_FLUSH(jset) && journal_entry_empty(jset)) |
158eecb8 KO |
1606 | j->last_empty_seq = le64_to_cpu(jset->seq); |
1607 | ||
26609b61 KO |
1608 | if (bch2_csum_type_is_encryption(JSET_CSUM_TYPE(jset))) |
1609 | validate_before_checksum = true; | |
1610 | ||
e751c01a | 1611 | if (le32_to_cpu(jset->version) < bcachefs_metadata_version_current) |
26609b61 KO |
1612 | validate_before_checksum = true; |
1613 | ||
1614 | if (validate_before_checksum && | |
ed9d58a2 | 1615 | jset_validate_for_write(c, jset)) |
1c6fdbd8 KO |
1616 | goto err; |
1617 | ||
a9de137b | 1618 | ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset), |
1c6fdbd8 KO |
1619 | jset->encrypted_start, |
1620 | vstruct_end(jset) - (void *) jset->encrypted_start); | |
a9de137b KO |
1621 | if (bch2_fs_fatal_err_on(ret, c, |
1622 | "error decrypting journal entry: %i", ret)) | |
1623 | goto err; | |
1c6fdbd8 KO |
1624 | |
1625 | jset->csum = csum_vstruct(c, JSET_CSUM_TYPE(jset), | |
1626 | journal_nonce(jset), jset); | |
1627 | ||
26609b61 | 1628 | if (!validate_before_checksum && |
ed9d58a2 | 1629 | jset_validate_for_write(c, jset)) |
1c6fdbd8 KO |
1630 | goto err; |
1631 | ||
1632 | sectors = vstruct_sectors(jset, c->block_bits); | |
d16b4a77 | 1633 | BUG_ON(sectors > w->sectors); |
1c6fdbd8 | 1634 | |
d16b4a77 KO |
1635 | bytes = vstruct_bytes(jset); |
1636 | memset((void *) jset + bytes, 0, (sectors << 9) - bytes); | |
1c6fdbd8 | 1637 | |
c18dade6 | 1638 | retry_alloc: |
e5a66496 KO |
1639 | spin_lock(&j->lock); |
1640 | ret = journal_write_alloc(j, w, sectors); | |
1641 | ||
c18dade6 KO |
1642 | if (ret && j->can_discard) { |
1643 | spin_unlock(&j->lock); | |
1644 | bch2_journal_do_discards(j); | |
1645 | goto retry_alloc; | |
1646 | } | |
1647 | ||
fa8e94fa KO |
1648 | if (ret) |
1649 | __bch2_journal_debug_to_text(&journal_debug_buf, j); | |
85674154 | 1650 | |
e5a66496 KO |
1651 | /* |
1652 | * write is allocated, no longer need to account for it in | |
1653 | * bch2_journal_space_available(): | |
1654 | */ | |
1655 | w->sectors = 0; | |
1656 | ||
1657 | /* | |
1658 | * journal entry has been compacted and allocated, recalculate space | |
1659 | * available: | |
1660 | */ | |
1661 | bch2_journal_space_available(j); | |
1662 | spin_unlock(&j->lock); | |
1663 | ||
1664 | if (ret) { | |
85674154 | 1665 | bch_err(c, "Unable to allocate journal write:\n%s", |
fa8e94fa KO |
1666 | journal_debug_buf.buf); |
1667 | printbuf_exit(&journal_debug_buf); | |
1c6fdbd8 | 1668 | bch2_fatal_error(c); |
731bdd2e | 1669 | continue_at(cl, journal_write_done, c->io_complete_wq); |
1c6fdbd8 KO |
1670 | return; |
1671 | } | |
1672 | ||
d797ca3d KO |
1673 | w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key)); |
1674 | ||
b66b2bc0 | 1675 | if (c->opts.nochanges) |
1c6fdbd8 KO |
1676 | goto no_io; |
1677 | ||
280249b9 KO |
1678 | for_each_rw_member(ca, c, i) |
1679 | nr_rw_members++; | |
1c6fdbd8 | 1680 | |
280249b9 KO |
1681 | if (nr_rw_members > 1) |
1682 | w->separate_flush = true; | |
1c6fdbd8 | 1683 | |
280249b9 KO |
1684 | if (!JSET_NO_FLUSH(jset) && w->separate_flush) { |
1685 | for_each_rw_member(ca, c, i) { | |
1686 | percpu_ref_get(&ca->io_ref); | |
1c6fdbd8 | 1687 | |
280249b9 KO |
1688 | bio = ca->journal.bio; |
1689 | bio_reset(bio, ca->disk_sb.bdev, REQ_OP_FLUSH); | |
1690 | bio->bi_end_io = journal_write_endio; | |
1691 | bio->bi_private = ca; | |
1692 | closure_bio_submit(bio, cl); | |
1693 | } | |
1c6fdbd8 KO |
1694 | } |
1695 | ||
731bdd2e | 1696 | continue_at(cl, do_journal_write, c->io_complete_wq); |
280249b9 | 1697 | return; |
1c6fdbd8 | 1698 | no_io: |
731bdd2e | 1699 | continue_at(cl, journal_write_done, c->io_complete_wq); |
1c6fdbd8 KO |
1700 | return; |
1701 | err: | |
b74b147d | 1702 | bch2_fatal_error(c); |
731bdd2e | 1703 | continue_at(cl, journal_write_done, c->io_complete_wq); |
1c6fdbd8 | 1704 | } |