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 | ||
17fe3b64 KO |
19 | static struct nonce journal_nonce(const struct jset *jset) |
20 | { | |
21 | return (struct nonce) {{ | |
22 | [0] = 0, | |
23 | [1] = ((__le32 *) &jset->seq)[0], | |
24 | [2] = ((__le32 *) &jset->seq)[1], | |
25 | [3] = BCH_NONCE_JOURNAL, | |
26 | }}; | |
27 | } | |
28 | ||
29 | static bool jset_csum_good(struct bch_fs *c, struct jset *j) | |
30 | { | |
31 | return bch2_checksum_type_valid(c, JSET_CSUM_TYPE(j)) && | |
32 | !bch2_crc_cmp(j->csum, | |
33 | csum_vstruct(c, JSET_CSUM_TYPE(j), journal_nonce(j), j)); | |
34 | } | |
35 | ||
ec7ccbde | 36 | static inline u32 journal_entry_radix_idx(struct bch_fs *c, u64 seq) |
adbcada4 | 37 | { |
ec7ccbde | 38 | return (seq - c->journal_entries_base_seq) & (~0U >> 1); |
ce6201c4 KO |
39 | } |
40 | ||
41 | static void __journal_replay_free(struct bch_fs *c, | |
42 | struct journal_replay *i) | |
43 | { | |
44 | struct journal_replay **p = | |
ec7ccbde KO |
45 | genradix_ptr(&c->journal_entries, |
46 | journal_entry_radix_idx(c, le64_to_cpu(i->j.seq))); | |
ce6201c4 KO |
47 | |
48 | BUG_ON(*p != i); | |
49 | *p = NULL; | |
adbcada4 KO |
50 | kvpfree(i, offsetof(struct journal_replay, j) + |
51 | vstruct_bytes(&i->j)); | |
adbcada4 KO |
52 | } |
53 | ||
54 | static void journal_replay_free(struct bch_fs *c, struct journal_replay *i) | |
55 | { | |
56 | i->ignore = true; | |
57 | ||
58 | if (!c->opts.read_entire_journal) | |
ce6201c4 | 59 | __journal_replay_free(c, i); |
adbcada4 KO |
60 | } |
61 | ||
1c6fdbd8 KO |
62 | struct journal_list { |
63 | struct closure cl; | |
ec7ccbde | 64 | u64 last_seq; |
1c6fdbd8 | 65 | struct mutex lock; |
1c6fdbd8 KO |
66 | int ret; |
67 | }; | |
68 | ||
69 | #define JOURNAL_ENTRY_ADD_OK 0 | |
70 | #define JOURNAL_ENTRY_ADD_OUT_OF_RANGE 5 | |
71 | ||
72 | /* | |
73 | * Given a journal entry we just read, add it to the list of journal entries to | |
74 | * be replayed: | |
75 | */ | |
76 | static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca, | |
72b7d633 | 77 | struct journal_ptr entry_ptr, |
17fe3b64 | 78 | struct journal_list *jlist, struct jset *j) |
1c6fdbd8 | 79 | { |
ce6201c4 KO |
80 | struct genradix_iter iter; |
81 | struct journal_replay **_i, *i, *dup; | |
72b7d633 | 82 | struct journal_ptr *ptr; |
1c6fdbd8 | 83 | size_t bytes = vstruct_bytes(j); |
ec7ccbde | 84 | u64 last_seq = !JSET_NO_FLUSH(j) ? le64_to_cpu(j->last_seq) : 0; |
e4c3f386 | 85 | int ret = JOURNAL_ENTRY_ADD_OK; |
1c6fdbd8 | 86 | |
ec7ccbde KO |
87 | /* Is this entry older than the range we need? */ |
88 | if (!c->opts.read_entire_journal && | |
89 | le64_to_cpu(j->seq) < jlist->last_seq) | |
90 | return JOURNAL_ENTRY_ADD_OUT_OF_RANGE; | |
91 | ||
ce6201c4 | 92 | /* |
ec7ccbde KO |
93 | * genradixes are indexed by a ulong, not a u64, so we can't index them |
94 | * by sequence number directly: Assume instead that they will all fall | |
95 | * within the range of +-2billion of the filrst one we find. | |
ce6201c4 KO |
96 | */ |
97 | if (!c->journal_entries_base_seq) | |
98 | c->journal_entries_base_seq = max_t(s64, 1, le64_to_cpu(j->seq) - S32_MAX); | |
99 | ||
adbcada4 | 100 | /* Drop entries we don't need anymore */ |
ec7ccbde KO |
101 | if (last_seq > jlist->last_seq && !c->opts.read_entire_journal) { |
102 | genradix_for_each_from(&c->journal_entries, iter, _i, | |
103 | journal_entry_radix_idx(c, jlist->last_seq)) { | |
ce6201c4 KO |
104 | i = *_i; |
105 | ||
ec7ccbde | 106 | if (!i || i->ignore) |
ce6201c4 KO |
107 | continue; |
108 | ||
ec7ccbde | 109 | if (le64_to_cpu(i->j.seq) >= last_seq) |
7fffc85b | 110 | break; |
adbcada4 | 111 | journal_replay_free(c, i); |
7fffc85b | 112 | } |
1c6fdbd8 KO |
113 | } |
114 | ||
ec7ccbde KO |
115 | jlist->last_seq = max(jlist->last_seq, last_seq); |
116 | ||
117 | _i = genradix_ptr_alloc(&c->journal_entries, | |
118 | journal_entry_radix_idx(c, le64_to_cpu(j->seq)), | |
119 | GFP_KERNEL); | |
120 | if (!_i) | |
121 | return -ENOMEM; | |
e4c3f386 | 122 | |
ca73852a KO |
123 | /* |
124 | * Duplicate journal entries? If so we want the one that didn't have a | |
125 | * checksum error: | |
126 | */ | |
ec7ccbde | 127 | dup = *_i; |
e4c3f386 | 128 | if (dup) { |
17fe3b64 KO |
129 | if (bytes == vstruct_bytes(&dup->j) && |
130 | !memcmp(j, &dup->j, bytes)) { | |
e4c3f386 | 131 | i = dup; |
ca73852a | 132 | goto found; |
17fe3b64 KO |
133 | } |
134 | ||
135 | if (!entry_ptr.csum_good) { | |
e4c3f386 | 136 | i = dup; |
1c6fdbd8 KO |
137 | goto found; |
138 | } | |
1c6fdbd8 | 139 | |
17fe3b64 KO |
140 | if (!dup->csum_good) |
141 | goto replace; | |
142 | ||
143 | fsck_err(c, "found duplicate but non identical journal entries (seq %llu)", | |
144 | le64_to_cpu(j->seq)); | |
145 | i = dup; | |
146 | goto found; | |
147 | } | |
148 | replace: | |
1c6fdbd8 | 149 | i = kvpmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL); |
ec7ccbde KO |
150 | if (!i) |
151 | return -ENOMEM; | |
1c6fdbd8 | 152 | |
17fe3b64 KO |
153 | i->nr_ptrs = 0; |
154 | i->csum_good = entry_ptr.csum_good; | |
e4c3f386 | 155 | i->ignore = false; |
1c6fdbd8 | 156 | unsafe_memcpy(&i->j, j, bytes, "embedded variable length struct"); |
17fe3b64 | 157 | i->ptrs[i->nr_ptrs++] = entry_ptr; |
e4c3f386 KO |
158 | |
159 | if (dup) { | |
17fe3b64 KO |
160 | if (dup->nr_ptrs >= ARRAY_SIZE(dup->ptrs)) { |
161 | bch_err(c, "found too many copies of journal entry %llu", | |
162 | le64_to_cpu(i->j.seq)); | |
163 | dup->nr_ptrs = ARRAY_SIZE(dup->ptrs) - 1; | |
164 | } | |
165 | ||
166 | /* The first ptr should represent the jset we kept: */ | |
167 | memcpy(i->ptrs + i->nr_ptrs, | |
168 | dup->ptrs, | |
169 | sizeof(dup->ptrs[0]) * dup->nr_ptrs); | |
170 | i->nr_ptrs += dup->nr_ptrs; | |
ce6201c4 | 171 | __journal_replay_free(c, dup); |
e4c3f386 KO |
172 | } |
173 | ||
ce6201c4 | 174 | *_i = i; |
17fe3b64 | 175 | return 0; |
1c6fdbd8 | 176 | found: |
e4c3f386 KO |
177 | for (ptr = i->ptrs; ptr < i->ptrs + i->nr_ptrs; ptr++) { |
178 | if (ptr->dev == ca->dev_idx) { | |
179 | bch_err(c, "duplicate journal entry %llu on same device", | |
180 | le64_to_cpu(i->j.seq)); | |
181 | goto out; | |
182 | } | |
183 | } | |
184 | ||
185 | if (i->nr_ptrs >= ARRAY_SIZE(i->ptrs)) { | |
186 | bch_err(c, "found too many copies of journal entry %llu", | |
187 | le64_to_cpu(i->j.seq)); | |
188 | goto out; | |
189 | } | |
190 | ||
191 | i->ptrs[i->nr_ptrs++] = entry_ptr; | |
1c6fdbd8 KO |
192 | out: |
193 | fsck_err: | |
194 | return ret; | |
195 | } | |
196 | ||
1c6fdbd8 KO |
197 | /* this fills in a range with empty jset_entries: */ |
198 | static void journal_entry_null_range(void *start, void *end) | |
199 | { | |
200 | struct jset_entry *entry; | |
201 | ||
202 | for (entry = start; entry != end; entry = vstruct_next(entry)) | |
203 | memset(entry, 0, sizeof(*entry)); | |
204 | } | |
205 | ||
206 | #define JOURNAL_ENTRY_REREAD 5 | |
207 | #define JOURNAL_ENTRY_NONE 6 | |
208 | #define JOURNAL_ENTRY_BAD 7 | |
209 | ||
c23a9e08 KO |
210 | static void journal_entry_err_msg(struct printbuf *out, |
211 | struct jset *jset, | |
212 | struct jset_entry *entry) | |
213 | { | |
214 | prt_str(out, "invalid journal entry "); | |
215 | if (entry) | |
216 | prt_printf(out, "%s ", bch2_jset_entry_types[entry->type]); | |
217 | ||
218 | if (!jset) | |
219 | prt_printf(out, "in superblock"); | |
220 | else if (!entry) | |
221 | prt_printf(out, "at seq %llu", le64_to_cpu(jset->seq)); | |
222 | else | |
223 | prt_printf(out, "at offset %zi/%u seq %llu", | |
224 | (u64 *) entry - jset->_data, | |
225 | le32_to_cpu(jset->u64s), | |
226 | le64_to_cpu(jset->seq)); | |
227 | prt_str(out, ": "); | |
228 | } | |
229 | ||
230 | #define journal_entry_err(c, jset, entry, msg, ...) \ | |
1c6fdbd8 | 231 | ({ \ |
c23a9e08 KO |
232 | struct printbuf buf = PRINTBUF; \ |
233 | \ | |
234 | journal_entry_err_msg(&buf, jset, entry); \ | |
235 | prt_printf(&buf, msg, ##__VA_ARGS__); \ | |
236 | \ | |
1c6fdbd8 KO |
237 | switch (write) { \ |
238 | case READ: \ | |
c23a9e08 | 239 | mustfix_fsck_err(c, "%s", buf.buf); \ |
1c6fdbd8 KO |
240 | break; \ |
241 | case WRITE: \ | |
c23a9e08 | 242 | bch_err(c, "corrupt metadata before write: %s\n", buf.buf);\ |
1c6fdbd8 | 243 | if (bch2_fs_inconsistent(c)) { \ |
1ed0a5d2 | 244 | ret = -BCH_ERR_fsck_errors_not_fixed; \ |
1c6fdbd8 KO |
245 | goto fsck_err; \ |
246 | } \ | |
247 | break; \ | |
248 | } \ | |
c23a9e08 KO |
249 | \ |
250 | printbuf_exit(&buf); \ | |
1c6fdbd8 KO |
251 | true; \ |
252 | }) | |
253 | ||
c23a9e08 KO |
254 | #define journal_entry_err_on(cond, c, jset, entry, msg, ...) \ |
255 | ((cond) ? journal_entry_err(c, jset, entry, msg, ##__VA_ARGS__) : false) | |
1c6fdbd8 | 256 | |
4d54337c KO |
257 | #define FSCK_DELETED_KEY 5 |
258 | ||
c23a9e08 KO |
259 | static int journal_validate_key(struct bch_fs *c, |
260 | struct jset *jset, | |
1c6fdbd8 | 261 | struct jset_entry *entry, |
39fb2983 | 262 | unsigned level, enum btree_id btree_id, |
cb685ce7 | 263 | struct bkey_i *k, |
7d6f07ed | 264 | unsigned version, int big_endian, int write) |
1c6fdbd8 KO |
265 | { |
266 | void *next = vstruct_next(entry); | |
f0ac7df2 | 267 | struct printbuf buf = PRINTBUF; |
1c6fdbd8 KO |
268 | int ret = 0; |
269 | ||
c23a9e08 | 270 | if (journal_entry_err_on(!k->k.u64s, c, jset, entry, "k->u64s 0")) { |
1c6fdbd8 KO |
271 | entry->u64s = cpu_to_le16((u64 *) k - entry->_data); |
272 | journal_entry_null_range(vstruct_next(entry), next); | |
4d54337c | 273 | return FSCK_DELETED_KEY; |
1c6fdbd8 KO |
274 | } |
275 | ||
276 | if (journal_entry_err_on((void *) bkey_next(k) > | |
c23a9e08 KO |
277 | (void *) vstruct_next(entry), |
278 | c, jset, entry, | |
279 | "extends past end of journal entry")) { | |
1c6fdbd8 KO |
280 | entry->u64s = cpu_to_le16((u64 *) k - entry->_data); |
281 | journal_entry_null_range(vstruct_next(entry), next); | |
4d54337c | 282 | return FSCK_DELETED_KEY; |
1c6fdbd8 KO |
283 | } |
284 | ||
c23a9e08 KO |
285 | if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT, |
286 | c, jset, entry, | |
287 | "bad format %u", k->k.format)) { | |
4d54337c | 288 | le16_add_cpu(&entry->u64s, -((u16) k->k.u64s)); |
1c6fdbd8 KO |
289 | memmove(k, bkey_next(k), next - (void *) bkey_next(k)); |
290 | journal_entry_null_range(vstruct_next(entry), next); | |
4d54337c | 291 | return FSCK_DELETED_KEY; |
1c6fdbd8 KO |
292 | } |
293 | ||
39fb2983 | 294 | if (!write) |
7d6f07ed KO |
295 | bch2_bkey_compat(level, btree_id, version, big_endian, |
296 | write, NULL, bkey_to_packed(k)); | |
26609b61 | 297 | |
f0ac7df2 | 298 | if (bch2_bkey_invalid(c, bkey_i_to_s_c(k), |
275c8426 | 299 | __btree_node_type(level, btree_id), write, &buf)) { |
f0ac7df2 | 300 | printbuf_reset(&buf); |
c23a9e08 KO |
301 | prt_printf(&buf, "invalid journal entry %s at offset %zi/%u seq %llu:", |
302 | bch2_jset_entry_types[entry->type], | |
303 | (u64 *) entry - jset->_data, | |
304 | le32_to_cpu(jset->u64s), | |
305 | le64_to_cpu(jset->seq)); | |
401ec4db KO |
306 | prt_newline(&buf); |
307 | printbuf_indent_add(&buf, 2); | |
319f9ac3 | 308 | |
fa8e94fa | 309 | bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k)); |
401ec4db | 310 | prt_newline(&buf); |
f0ac7df2 | 311 | bch2_bkey_invalid(c, bkey_i_to_s_c(k), |
275c8426 | 312 | __btree_node_type(level, btree_id), write, &buf); |
f0ac7df2 KO |
313 | |
314 | mustfix_fsck_err(c, "%s", buf.buf); | |
1c6fdbd8 | 315 | |
4d54337c | 316 | le16_add_cpu(&entry->u64s, -((u16) k->k.u64s)); |
1c6fdbd8 KO |
317 | memmove(k, bkey_next(k), next - (void *) bkey_next(k)); |
318 | journal_entry_null_range(vstruct_next(entry), next); | |
f0ac7df2 KO |
319 | |
320 | printbuf_exit(&buf); | |
4d54337c | 321 | return FSCK_DELETED_KEY; |
1c6fdbd8 | 322 | } |
26609b61 | 323 | |
39fb2983 | 324 | if (write) |
7d6f07ed KO |
325 | bch2_bkey_compat(level, btree_id, version, big_endian, |
326 | write, NULL, bkey_to_packed(k)); | |
1c6fdbd8 | 327 | fsck_err: |
f0ac7df2 | 328 | printbuf_exit(&buf); |
1c6fdbd8 KO |
329 | return ret; |
330 | } | |
331 | ||
528b18e6 | 332 | static int journal_entry_btree_keys_validate(struct bch_fs *c, |
c23a9e08 | 333 | struct jset *jset, |
1c6fdbd8 | 334 | struct jset_entry *entry, |
7d6f07ed | 335 | unsigned version, int big_endian, int write) |
1c6fdbd8 | 336 | { |
4d54337c | 337 | struct bkey_i *k = entry->start; |
1c6fdbd8 | 338 | |
4d54337c | 339 | while (k != vstruct_last(entry)) { |
c23a9e08 | 340 | int ret = journal_validate_key(c, jset, entry, |
39fb2983 KO |
341 | entry->level, |
342 | entry->btree_id, | |
cb685ce7 | 343 | k, version, big_endian, write); |
4d54337c KO |
344 | if (ret == FSCK_DELETED_KEY) |
345 | continue; | |
346 | ||
347 | k = bkey_next(k); | |
1c6fdbd8 KO |
348 | } |
349 | ||
350 | return 0; | |
351 | } | |
352 | ||
528b18e6 KO |
353 | static void journal_entry_btree_keys_to_text(struct printbuf *out, struct bch_fs *c, |
354 | struct jset_entry *entry) | |
355 | { | |
356 | struct bkey_i *k; | |
e7bc7cdf | 357 | bool first = true; |
528b18e6 | 358 | |
e7bc7cdf KO |
359 | vstruct_for_each(entry, k) { |
360 | if (!first) { | |
401ec4db KO |
361 | prt_newline(out); |
362 | prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]); | |
e7bc7cdf | 363 | } |
401ec4db | 364 | prt_printf(out, "btree=%s l=%u ", bch2_btree_ids[entry->btree_id], entry->level); |
528b18e6 | 365 | bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(k)); |
e7bc7cdf KO |
366 | first = false; |
367 | } | |
528b18e6 KO |
368 | } |
369 | ||
370 | static int journal_entry_btree_root_validate(struct bch_fs *c, | |
c23a9e08 | 371 | struct jset *jset, |
1c6fdbd8 | 372 | struct jset_entry *entry, |
7d6f07ed | 373 | unsigned version, int big_endian, int write) |
1c6fdbd8 KO |
374 | { |
375 | struct bkey_i *k = entry->start; | |
376 | int ret = 0; | |
377 | ||
378 | if (journal_entry_err_on(!entry->u64s || | |
c23a9e08 KO |
379 | le16_to_cpu(entry->u64s) != k->k.u64s, |
380 | c, jset, entry, | |
1c6fdbd8 KO |
381 | "invalid btree root journal entry: wrong number of keys")) { |
382 | void *next = vstruct_next(entry); | |
383 | /* | |
384 | * we don't want to null out this jset_entry, | |
385 | * just the contents, so that later we can tell | |
386 | * we were _supposed_ to have a btree root | |
387 | */ | |
388 | entry->u64s = 0; | |
389 | journal_entry_null_range(vstruct_next(entry), next); | |
390 | return 0; | |
391 | } | |
392 | ||
c23a9e08 | 393 | return journal_validate_key(c, jset, entry, 1, entry->btree_id, k, |
cb685ce7 | 394 | version, big_endian, write); |
1c6fdbd8 KO |
395 | fsck_err: |
396 | return ret; | |
397 | } | |
398 | ||
528b18e6 KO |
399 | static void journal_entry_btree_root_to_text(struct printbuf *out, struct bch_fs *c, |
400 | struct jset_entry *entry) | |
401 | { | |
402 | journal_entry_btree_keys_to_text(out, c, entry); | |
403 | } | |
404 | ||
405 | static int journal_entry_prio_ptrs_validate(struct bch_fs *c, | |
c23a9e08 | 406 | struct jset *jset, |
1c6fdbd8 | 407 | struct jset_entry *entry, |
7d6f07ed | 408 | unsigned version, int big_endian, int write) |
1c6fdbd8 KO |
409 | { |
410 | /* obsolete, don't care: */ | |
411 | return 0; | |
412 | } | |
413 | ||
528b18e6 KO |
414 | static void journal_entry_prio_ptrs_to_text(struct printbuf *out, struct bch_fs *c, |
415 | struct jset_entry *entry) | |
416 | { | |
417 | } | |
418 | ||
419 | static int journal_entry_blacklist_validate(struct bch_fs *c, | |
c23a9e08 | 420 | struct jset *jset, |
1c6fdbd8 | 421 | struct jset_entry *entry, |
7d6f07ed | 422 | unsigned version, int big_endian, int write) |
1c6fdbd8 KO |
423 | { |
424 | int ret = 0; | |
425 | ||
c23a9e08 KO |
426 | if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 1, |
427 | c, jset, entry, | |
1c6fdbd8 KO |
428 | "invalid journal seq blacklist entry: bad size")) { |
429 | journal_entry_null_range(entry, vstruct_next(entry)); | |
430 | } | |
431 | fsck_err: | |
432 | return ret; | |
433 | } | |
434 | ||
528b18e6 KO |
435 | static void journal_entry_blacklist_to_text(struct printbuf *out, struct bch_fs *c, |
436 | struct jset_entry *entry) | |
437 | { | |
438 | struct jset_entry_blacklist *bl = | |
439 | container_of(entry, struct jset_entry_blacklist, entry); | |
440 | ||
401ec4db | 441 | prt_printf(out, "seq=%llu", le64_to_cpu(bl->seq)); |
528b18e6 KO |
442 | } |
443 | ||
444 | static int journal_entry_blacklist_v2_validate(struct bch_fs *c, | |
c23a9e08 | 445 | struct jset *jset, |
1c6fdbd8 | 446 | struct jset_entry *entry, |
7d6f07ed | 447 | unsigned version, int big_endian, int write) |
1c6fdbd8 KO |
448 | { |
449 | struct jset_entry_blacklist_v2 *bl_entry; | |
450 | int ret = 0; | |
451 | ||
c23a9e08 KO |
452 | if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 2, |
453 | c, jset, entry, | |
1c6fdbd8 KO |
454 | "invalid journal seq blacklist entry: bad size")) { |
455 | journal_entry_null_range(entry, vstruct_next(entry)); | |
2c5af169 | 456 | goto out; |
1c6fdbd8 KO |
457 | } |
458 | ||
459 | bl_entry = container_of(entry, struct jset_entry_blacklist_v2, entry); | |
460 | ||
461 | if (journal_entry_err_on(le64_to_cpu(bl_entry->start) > | |
c23a9e08 KO |
462 | le64_to_cpu(bl_entry->end), |
463 | c, jset, entry, | |
1c6fdbd8 KO |
464 | "invalid journal seq blacklist entry: start > end")) { |
465 | journal_entry_null_range(entry, vstruct_next(entry)); | |
466 | } | |
2c5af169 KO |
467 | out: |
468 | fsck_err: | |
469 | return ret; | |
470 | } | |
471 | ||
528b18e6 KO |
472 | static void journal_entry_blacklist_v2_to_text(struct printbuf *out, struct bch_fs *c, |
473 | struct jset_entry *entry) | |
474 | { | |
475 | struct jset_entry_blacklist_v2 *bl = | |
476 | container_of(entry, struct jset_entry_blacklist_v2, entry); | |
477 | ||
401ec4db | 478 | prt_printf(out, "start=%llu end=%llu", |
528b18e6 KO |
479 | le64_to_cpu(bl->start), |
480 | le64_to_cpu(bl->end)); | |
481 | } | |
482 | ||
483 | static int journal_entry_usage_validate(struct bch_fs *c, | |
c23a9e08 | 484 | struct jset *jset, |
2c5af169 | 485 | struct jset_entry *entry, |
7d6f07ed | 486 | unsigned version, int big_endian, int write) |
2c5af169 KO |
487 | { |
488 | struct jset_entry_usage *u = | |
489 | container_of(entry, struct jset_entry_usage, entry); | |
490 | unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64); | |
491 | int ret = 0; | |
492 | ||
3577df5f | 493 | if (journal_entry_err_on(bytes < sizeof(*u), |
c23a9e08 | 494 | c, jset, entry, |
3577df5f KO |
495 | "invalid journal entry usage: bad size")) { |
496 | journal_entry_null_range(entry, vstruct_next(entry)); | |
497 | return ret; | |
498 | } | |
499 | ||
500 | fsck_err: | |
501 | return ret; | |
502 | } | |
503 | ||
528b18e6 KO |
504 | static void journal_entry_usage_to_text(struct printbuf *out, struct bch_fs *c, |
505 | struct jset_entry *entry) | |
506 | { | |
507 | struct jset_entry_usage *u = | |
508 | container_of(entry, struct jset_entry_usage, entry); | |
509 | ||
401ec4db | 510 | prt_printf(out, "type=%s v=%llu", |
528b18e6 KO |
511 | bch2_fs_usage_types[u->entry.btree_id], |
512 | le64_to_cpu(u->v)); | |
513 | } | |
514 | ||
515 | static int journal_entry_data_usage_validate(struct bch_fs *c, | |
c23a9e08 | 516 | struct jset *jset, |
3577df5f | 517 | struct jset_entry *entry, |
7d6f07ed | 518 | unsigned version, int big_endian, int write) |
3577df5f KO |
519 | { |
520 | struct jset_entry_data_usage *u = | |
521 | container_of(entry, struct jset_entry_data_usage, entry); | |
522 | unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64); | |
523 | int ret = 0; | |
524 | ||
2c5af169 KO |
525 | if (journal_entry_err_on(bytes < sizeof(*u) || |
526 | bytes < sizeof(*u) + u->r.nr_devs, | |
c23a9e08 | 527 | c, jset, entry, |
2c5af169 KO |
528 | "invalid journal entry usage: bad size")) { |
529 | journal_entry_null_range(entry, vstruct_next(entry)); | |
530 | return ret; | |
531 | } | |
1c6fdbd8 KO |
532 | |
533 | fsck_err: | |
534 | return ret; | |
535 | } | |
536 | ||
528b18e6 KO |
537 | static void journal_entry_data_usage_to_text(struct printbuf *out, struct bch_fs *c, |
538 | struct jset_entry *entry) | |
539 | { | |
540 | struct jset_entry_data_usage *u = | |
541 | container_of(entry, struct jset_entry_data_usage, entry); | |
542 | ||
543 | bch2_replicas_entry_to_text(out, &u->r); | |
401ec4db | 544 | prt_printf(out, "=%llu", le64_to_cpu(u->v)); |
528b18e6 KO |
545 | } |
546 | ||
547 | static int journal_entry_clock_validate(struct bch_fs *c, | |
c23a9e08 | 548 | struct jset *jset, |
2abe5420 | 549 | struct jset_entry *entry, |
7d6f07ed | 550 | unsigned version, int big_endian, int write) |
2abe5420 KO |
551 | { |
552 | struct jset_entry_clock *clock = | |
553 | container_of(entry, struct jset_entry_clock, entry); | |
554 | unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64); | |
555 | int ret = 0; | |
556 | ||
557 | if (journal_entry_err_on(bytes != sizeof(*clock), | |
c23a9e08 | 558 | c, jset, entry, "bad size")) { |
2abe5420 KO |
559 | journal_entry_null_range(entry, vstruct_next(entry)); |
560 | return ret; | |
561 | } | |
562 | ||
563 | if (journal_entry_err_on(clock->rw > 1, | |
c23a9e08 | 564 | c, jset, entry, "bad rw")) { |
2abe5420 KO |
565 | journal_entry_null_range(entry, vstruct_next(entry)); |
566 | return ret; | |
567 | } | |
568 | ||
569 | fsck_err: | |
570 | return ret; | |
571 | } | |
572 | ||
528b18e6 KO |
573 | static void journal_entry_clock_to_text(struct printbuf *out, struct bch_fs *c, |
574 | struct jset_entry *entry) | |
575 | { | |
576 | struct jset_entry_clock *clock = | |
577 | container_of(entry, struct jset_entry_clock, entry); | |
578 | ||
401ec4db | 579 | prt_printf(out, "%s=%llu", clock->rw ? "write" : "read", le64_to_cpu(clock->time)); |
528b18e6 KO |
580 | } |
581 | ||
582 | static int journal_entry_dev_usage_validate(struct bch_fs *c, | |
c23a9e08 | 583 | struct jset *jset, |
180fb49d | 584 | struct jset_entry *entry, |
7d6f07ed | 585 | unsigned version, int big_endian, int write) |
180fb49d KO |
586 | { |
587 | struct jset_entry_dev_usage *u = | |
588 | container_of(entry, struct jset_entry_dev_usage, entry); | |
589 | unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64); | |
45c2e33f | 590 | unsigned expected = sizeof(*u); |
180fb49d KO |
591 | unsigned dev; |
592 | int ret = 0; | |
593 | ||
594 | if (journal_entry_err_on(bytes < expected, | |
c23a9e08 | 595 | c, jset, entry, "bad size (%u < %u)", |
180fb49d KO |
596 | bytes, expected)) { |
597 | journal_entry_null_range(entry, vstruct_next(entry)); | |
598 | return ret; | |
599 | } | |
600 | ||
601 | dev = le32_to_cpu(u->dev); | |
602 | ||
603 | if (journal_entry_err_on(!bch2_dev_exists2(c, dev), | |
c23a9e08 | 604 | c, jset, entry, "bad dev")) { |
180fb49d KO |
605 | journal_entry_null_range(entry, vstruct_next(entry)); |
606 | return ret; | |
607 | } | |
608 | ||
609 | if (journal_entry_err_on(u->pad, | |
c23a9e08 | 610 | c, jset, entry, "bad pad")) { |
180fb49d KO |
611 | journal_entry_null_range(entry, vstruct_next(entry)); |
612 | return ret; | |
613 | } | |
614 | ||
615 | fsck_err: | |
616 | return ret; | |
617 | } | |
618 | ||
528b18e6 KO |
619 | static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs *c, |
620 | struct jset_entry *entry) | |
621 | { | |
622 | struct jset_entry_dev_usage *u = | |
623 | container_of(entry, struct jset_entry_dev_usage, entry); | |
624 | unsigned i, nr_types = jset_entry_dev_usage_nr_types(u); | |
625 | ||
401ec4db | 626 | prt_printf(out, "dev=%u", le32_to_cpu(u->dev)); |
528b18e6 KO |
627 | |
628 | for (i = 0; i < nr_types; i++) { | |
629 | if (i < BCH_DATA_NR) | |
401ec4db | 630 | prt_printf(out, " %s", bch2_data_types[i]); |
528b18e6 | 631 | else |
401ec4db KO |
632 | prt_printf(out, " (unknown data type %u)", i); |
633 | prt_printf(out, ": buckets=%llu sectors=%llu fragmented=%llu", | |
528b18e6 KO |
634 | le64_to_cpu(u->d[i].buckets), |
635 | le64_to_cpu(u->d[i].sectors), | |
636 | le64_to_cpu(u->d[i].fragmented)); | |
637 | } | |
638 | ||
401ec4db | 639 | prt_printf(out, " buckets_ec: %llu", le64_to_cpu(u->buckets_ec)); |
528b18e6 KO |
640 | } |
641 | ||
642 | static int journal_entry_log_validate(struct bch_fs *c, | |
c23a9e08 | 643 | struct jset *jset, |
fb64f3fd KO |
644 | struct jset_entry *entry, |
645 | unsigned version, int big_endian, int write) | |
646 | { | |
647 | return 0; | |
648 | } | |
649 | ||
528b18e6 KO |
650 | static void journal_entry_log_to_text(struct printbuf *out, struct bch_fs *c, |
651 | struct jset_entry *entry) | |
652 | { | |
653 | struct jset_entry_log *l = container_of(entry, struct jset_entry_log, entry); | |
654 | unsigned bytes = vstruct_bytes(entry) - offsetof(struct jset_entry_log, d); | |
655 | ||
401ec4db | 656 | prt_printf(out, "%.*s", bytes, l->d); |
528b18e6 KO |
657 | } |
658 | ||
c23a9e08 KO |
659 | static int journal_entry_overwrite_validate(struct bch_fs *c, |
660 | struct jset *jset, | |
cb685ce7 KO |
661 | struct jset_entry *entry, |
662 | unsigned version, int big_endian, int write) | |
663 | { | |
c23a9e08 | 664 | return journal_entry_btree_keys_validate(c, jset, entry, version, big_endian, write); |
cb685ce7 KO |
665 | } |
666 | ||
667 | static void journal_entry_overwrite_to_text(struct printbuf *out, struct bch_fs *c, | |
668 | struct jset_entry *entry) | |
669 | { | |
670 | journal_entry_btree_keys_to_text(out, c, entry); | |
671 | } | |
672 | ||
1c6fdbd8 | 673 | struct jset_entry_ops { |
c23a9e08 | 674 | int (*validate)(struct bch_fs *, struct jset *, |
7d6f07ed | 675 | struct jset_entry *, unsigned, int, int); |
528b18e6 | 676 | void (*to_text)(struct printbuf *, struct bch_fs *, struct jset_entry *); |
1c6fdbd8 KO |
677 | }; |
678 | ||
679 | static const struct jset_entry_ops bch2_jset_entry_ops[] = { | |
680 | #define x(f, nr) \ | |
681 | [BCH_JSET_ENTRY_##f] = (struct jset_entry_ops) { \ | |
528b18e6 KO |
682 | .validate = journal_entry_##f##_validate, \ |
683 | .to_text = journal_entry_##f##_to_text, \ | |
1c6fdbd8 KO |
684 | }, |
685 | BCH_JSET_ENTRY_TYPES() | |
686 | #undef x | |
687 | }; | |
688 | ||
c23a9e08 KO |
689 | int bch2_journal_entry_validate(struct bch_fs *c, |
690 | struct jset *jset, | |
7d6f07ed KO |
691 | struct jset_entry *entry, |
692 | unsigned version, int big_endian, int write) | |
1c6fdbd8 | 693 | { |
2c5af169 | 694 | return entry->type < BCH_JSET_ENTRY_NR |
c23a9e08 | 695 | ? bch2_jset_entry_ops[entry->type].validate(c, jset, entry, |
7d6f07ed | 696 | version, big_endian, write) |
2c5af169 | 697 | : 0; |
1c6fdbd8 KO |
698 | } |
699 | ||
528b18e6 KO |
700 | void bch2_journal_entry_to_text(struct printbuf *out, struct bch_fs *c, |
701 | struct jset_entry *entry) | |
702 | { | |
703 | if (entry->type < BCH_JSET_ENTRY_NR) { | |
401ec4db | 704 | prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]); |
528b18e6 KO |
705 | bch2_jset_entry_ops[entry->type].to_text(out, c, entry); |
706 | } else { | |
401ec4db | 707 | prt_printf(out, "(unknown type %u)", entry->type); |
528b18e6 KO |
708 | } |
709 | } | |
710 | ||
1c6fdbd8 KO |
711 | static int jset_validate_entries(struct bch_fs *c, struct jset *jset, |
712 | int write) | |
713 | { | |
714 | struct jset_entry *entry; | |
715 | int ret = 0; | |
716 | ||
717 | vstruct_for_each(jset, entry) { | |
718 | if (journal_entry_err_on(vstruct_next(entry) > | |
c23a9e08 | 719 | vstruct_last(jset), c, jset, entry, |
1c6fdbd8 KO |
720 | "journal entry extends past end of jset")) { |
721 | jset->u64s = cpu_to_le32((u64 *) entry - jset->_data); | |
722 | break; | |
723 | } | |
724 | ||
c23a9e08 | 725 | ret = bch2_journal_entry_validate(c, jset, entry, |
7d6f07ed KO |
726 | le32_to_cpu(jset->version), |
727 | JSET_BIG_ENDIAN(jset), write); | |
1c6fdbd8 KO |
728 | if (ret) |
729 | break; | |
730 | } | |
731 | fsck_err: | |
732 | return ret; | |
733 | } | |
734 | ||
735 | static int jset_validate(struct bch_fs *c, | |
ca73852a | 736 | struct bch_dev *ca, |
1c6fdbd8 KO |
737 | struct jset *jset, u64 sector, |
738 | unsigned bucket_sectors_left, | |
739 | unsigned sectors_read, | |
740 | int write) | |
741 | { | |
742 | size_t bytes = vstruct_bytes(jset); | |
743 | struct bch_csum csum; | |
26609b61 | 744 | unsigned version; |
1c6fdbd8 KO |
745 | int ret = 0; |
746 | ||
747 | if (le64_to_cpu(jset->magic) != jset_magic(c)) | |
748 | return JOURNAL_ENTRY_NONE; | |
749 | ||
26609b61 | 750 | version = le32_to_cpu(jset->version); |
ca73852a KO |
751 | if (journal_entry_err_on((version != BCH_JSET_VERSION_OLD && |
752 | version < bcachefs_metadata_version_min) || | |
c23a9e08 KO |
753 | version >= bcachefs_metadata_version_max, |
754 | c, jset, NULL, | |
ca73852a | 755 | "%s sector %llu seq %llu: unknown journal entry version %u", |
ed9d58a2 KO |
756 | ca ? ca->name : c->name, |
757 | sector, le64_to_cpu(jset->seq), | |
ca73852a | 758 | version)) { |
35ef6df5 KO |
759 | /* don't try to continue: */ |
760 | return EINVAL; | |
1c6fdbd8 KO |
761 | } |
762 | ||
35ef6df5 KO |
763 | if (bytes > (sectors_read << 9) && |
764 | sectors_read < bucket_sectors_left) | |
765 | return JOURNAL_ENTRY_REREAD; | |
766 | ||
c23a9e08 KO |
767 | if (journal_entry_err_on(bytes > bucket_sectors_left << 9, |
768 | c, jset, NULL, | |
ca73852a | 769 | "%s sector %llu seq %llu: journal entry too big (%zu bytes)", |
ed9d58a2 KO |
770 | ca ? ca->name : c->name, |
771 | sector, le64_to_cpu(jset->seq), bytes)) { | |
35ef6df5 KO |
772 | ret = JOURNAL_ENTRY_BAD; |
773 | le32_add_cpu(&jset->u64s, | |
774 | -((bytes - (bucket_sectors_left << 9)) / 8)); | |
1c6fdbd8 KO |
775 | } |
776 | ||
c23a9e08 KO |
777 | if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)), |
778 | c, jset, NULL, | |
ca73852a | 779 | "%s sector %llu seq %llu: journal entry with unknown csum type %llu", |
ed9d58a2 KO |
780 | ca ? ca->name : c->name, |
781 | sector, le64_to_cpu(jset->seq), | |
35ef6df5 KO |
782 | JSET_CSUM_TYPE(jset))) { |
783 | ret = JOURNAL_ENTRY_BAD; | |
ed9d58a2 | 784 | goto csum_done; |
35ef6df5 | 785 | } |
1c6fdbd8 | 786 | |
ed9d58a2 KO |
787 | if (write) |
788 | goto csum_done; | |
789 | ||
1c6fdbd8 | 790 | csum = csum_vstruct(c, JSET_CSUM_TYPE(jset), journal_nonce(jset), jset); |
c23a9e08 KO |
791 | if (journal_entry_err_on(bch2_crc_cmp(csum, jset->csum), |
792 | c, jset, NULL, | |
ca73852a | 793 | "%s sector %llu seq %llu: journal checksum bad", |
ed9d58a2 KO |
794 | ca ? ca->name : c->name, |
795 | sector, le64_to_cpu(jset->seq))) | |
35ef6df5 | 796 | ret = JOURNAL_ENTRY_BAD; |
1c6fdbd8 | 797 | |
a9de137b | 798 | ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset), |
1c6fdbd8 KO |
799 | jset->encrypted_start, |
800 | vstruct_end(jset) - (void *) jset->encrypted_start); | |
a9de137b KO |
801 | bch2_fs_fatal_err_on(ret, c, |
802 | "error decrypting journal entry: %i", ret); | |
ed9d58a2 KO |
803 | csum_done: |
804 | /* last_seq is ignored when JSET_NO_FLUSH is true */ | |
805 | if (journal_entry_err_on(!JSET_NO_FLUSH(jset) && | |
c23a9e08 KO |
806 | le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq), |
807 | c, jset, NULL, | |
ed9d58a2 KO |
808 | "invalid journal entry: last_seq > seq (%llu > %llu)", |
809 | le64_to_cpu(jset->last_seq), | |
810 | le64_to_cpu(jset->seq))) { | |
1c6fdbd8 | 811 | jset->last_seq = jset->seq; |
ca73852a KO |
812 | return JOURNAL_ENTRY_BAD; |
813 | } | |
1c6fdbd8 KO |
814 | fsck_err: |
815 | return ret; | |
816 | } | |
817 | ||
ed9d58a2 KO |
818 | static int jset_validate_for_write(struct bch_fs *c, struct jset *jset) |
819 | { | |
820 | unsigned sectors = vstruct_sectors(jset, c->block_bits); | |
821 | ||
822 | return jset_validate(c, NULL, jset, 0, sectors, sectors, WRITE) ?: | |
823 | jset_validate_entries(c, jset, WRITE); | |
824 | } | |
825 | ||
1c6fdbd8 KO |
826 | struct journal_read_buf { |
827 | void *data; | |
828 | size_t size; | |
829 | }; | |
830 | ||
831 | static int journal_read_buf_realloc(struct journal_read_buf *b, | |
832 | size_t new_size) | |
833 | { | |
834 | void *n; | |
835 | ||
836 | /* the bios are sized for this many pages, max: */ | |
837 | if (new_size > JOURNAL_ENTRY_SIZE_MAX) | |
838 | return -ENOMEM; | |
839 | ||
840 | new_size = roundup_pow_of_two(new_size); | |
841 | n = kvpmalloc(new_size, GFP_KERNEL); | |
842 | if (!n) | |
843 | return -ENOMEM; | |
844 | ||
845 | kvpfree(b->data, b->size); | |
846 | b->data = n; | |
847 | b->size = new_size; | |
848 | return 0; | |
849 | } | |
850 | ||
851 | static int journal_read_bucket(struct bch_dev *ca, | |
852 | struct journal_read_buf *buf, | |
853 | struct journal_list *jlist, | |
a9ec3454 | 854 | unsigned bucket) |
1c6fdbd8 KO |
855 | { |
856 | struct bch_fs *c = ca->fs; | |
857 | struct journal_device *ja = &ca->journal; | |
1c6fdbd8 KO |
858 | struct jset *j = NULL; |
859 | unsigned sectors, sectors_read = 0; | |
860 | u64 offset = bucket_to_sector(ca, ja->buckets[bucket]), | |
861 | end = offset + ca->mi.bucket_size; | |
17fe3b64 | 862 | bool saw_bad = false, csum_good; |
1c6fdbd8 KO |
863 | int ret = 0; |
864 | ||
865 | pr_debug("reading %u", bucket); | |
866 | ||
867 | while (offset < end) { | |
868 | if (!sectors_read) { | |
ac10a961 KO |
869 | struct bio *bio; |
870 | unsigned nr_bvecs; | |
871 | reread: | |
872 | sectors_read = min_t(unsigned, | |
1c6fdbd8 | 873 | end - offset, buf->size >> 9); |
ac10a961 KO |
874 | nr_bvecs = buf_pages(buf->data, sectors_read << 9); |
875 | ||
876 | bio = bio_kmalloc(nr_bvecs, GFP_KERNEL); | |
877 | bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ); | |
1c6fdbd8 | 878 | |
885678f6 KO |
879 | bio->bi_iter.bi_sector = offset; |
880 | bch2_bio_map(bio, buf->data, sectors_read << 9); | |
1c6fdbd8 KO |
881 | |
882 | ret = submit_bio_wait(bio); | |
ac10a961 | 883 | kfree(bio); |
1c6fdbd8 KO |
884 | |
885 | if (bch2_dev_io_err_on(ret, ca, | |
0fefe8d8 | 886 | "journal read error: sector %llu", |
1c6fdbd8 | 887 | offset) || |
29d90f61 KO |
888 | bch2_meta_read_fault("journal")) { |
889 | /* | |
890 | * We don't error out of the recovery process | |
891 | * here, since the relevant journal entry may be | |
892 | * found on a different device, and missing or | |
893 | * no journal entries will be handled later | |
894 | */ | |
895 | return 0; | |
896 | } | |
1c6fdbd8 KO |
897 | |
898 | j = buf->data; | |
899 | } | |
900 | ||
ca73852a | 901 | ret = jset_validate(c, ca, j, offset, |
1c6fdbd8 KO |
902 | end - offset, sectors_read, |
903 | READ); | |
904 | switch (ret) { | |
1ed0a5d2 | 905 | case 0: |
ca73852a | 906 | sectors = vstruct_sectors(j, c->block_bits); |
1c6fdbd8 KO |
907 | break; |
908 | case JOURNAL_ENTRY_REREAD: | |
909 | if (vstruct_bytes(j) > buf->size) { | |
910 | ret = journal_read_buf_realloc(buf, | |
911 | vstruct_bytes(j)); | |
912 | if (ret) | |
913 | return ret; | |
914 | } | |
915 | goto reread; | |
916 | case JOURNAL_ENTRY_NONE: | |
917 | if (!saw_bad) | |
918 | return 0; | |
8244f320 | 919 | sectors = block_sectors(c); |
1c6fdbd8 KO |
920 | goto next_block; |
921 | case JOURNAL_ENTRY_BAD: | |
922 | saw_bad = true; | |
ca73852a KO |
923 | /* |
924 | * On checksum error we don't really trust the size | |
925 | * field of the journal entry we read, so try reading | |
926 | * again at next block boundary: | |
927 | */ | |
8244f320 | 928 | sectors = block_sectors(c); |
ca73852a | 929 | break; |
1c6fdbd8 KO |
930 | default: |
931 | return ret; | |
932 | } | |
933 | ||
934 | /* | |
935 | * This happens sometimes if we don't have discards on - | |
936 | * when we've partially overwritten a bucket with new | |
937 | * journal entries. We don't need the rest of the | |
938 | * bucket: | |
939 | */ | |
940 | if (le64_to_cpu(j->seq) < ja->bucket_seq[bucket]) | |
941 | return 0; | |
942 | ||
943 | ja->bucket_seq[bucket] = le64_to_cpu(j->seq); | |
944 | ||
17fe3b64 KO |
945 | csum_good = jset_csum_good(c, j); |
946 | if (!csum_good) | |
947 | saw_bad = true; | |
948 | ||
1c6fdbd8 | 949 | mutex_lock(&jlist->lock); |
72b7d633 | 950 | ret = journal_entry_add(c, ca, (struct journal_ptr) { |
17fe3b64 | 951 | .csum_good = csum_good, |
72b7d633 KO |
952 | .dev = ca->dev_idx, |
953 | .bucket = bucket, | |
954 | .bucket_offset = offset - | |
955 | bucket_to_sector(ca, ja->buckets[bucket]), | |
956 | .sector = offset, | |
17fe3b64 | 957 | }, jlist, j); |
1c6fdbd8 KO |
958 | mutex_unlock(&jlist->lock); |
959 | ||
960 | switch (ret) { | |
961 | case JOURNAL_ENTRY_ADD_OK: | |
1c6fdbd8 KO |
962 | break; |
963 | case JOURNAL_ENTRY_ADD_OUT_OF_RANGE: | |
964 | break; | |
965 | default: | |
966 | return ret; | |
967 | } | |
1c6fdbd8 KO |
968 | next_block: |
969 | pr_debug("next"); | |
970 | offset += sectors; | |
971 | sectors_read -= sectors; | |
972 | j = ((void *) j) + (sectors << 9); | |
973 | } | |
974 | ||
975 | return 0; | |
976 | } | |
977 | ||
978 | static void bch2_journal_read_device(struct closure *cl) | |
979 | { | |
1c6fdbd8 KO |
980 | struct journal_device *ja = |
981 | container_of(cl, struct journal_device, read); | |
982 | struct bch_dev *ca = container_of(ja, struct bch_dev, journal); | |
365f64f3 | 983 | struct bch_fs *c = ca->fs; |
1c6fdbd8 KO |
984 | struct journal_list *jlist = |
985 | container_of(cl->parent, struct journal_list, cl); | |
ce6201c4 KO |
986 | struct journal_replay *r, **_r; |
987 | struct genradix_iter iter; | |
1c6fdbd8 | 988 | struct journal_read_buf buf = { NULL, 0 }; |
a9ec3454 KO |
989 | u64 min_seq = U64_MAX; |
990 | unsigned i; | |
9714baaa | 991 | int ret = 0; |
1c6fdbd8 KO |
992 | |
993 | if (!ja->nr) | |
994 | goto out; | |
995 | ||
1c6fdbd8 KO |
996 | ret = journal_read_buf_realloc(&buf, PAGE_SIZE); |
997 | if (ret) | |
998 | goto err; | |
999 | ||
1000 | pr_debug("%u journal buckets", ja->nr); | |
1001 | ||
1c6fdbd8 | 1002 | for (i = 0; i < ja->nr; i++) { |
a9ec3454 KO |
1003 | ret = journal_read_bucket(ca, &buf, jlist, i); |
1004 | if (ret) | |
1005 | goto err; | |
1c6fdbd8 KO |
1006 | } |
1007 | ||
a9ec3454 KO |
1008 | /* Find the journal bucket with the highest sequence number: */ |
1009 | for (i = 0; i < ja->nr; i++) { | |
1010 | if (ja->bucket_seq[i] > ja->bucket_seq[ja->cur_idx]) | |
1011 | ja->cur_idx = i; | |
1c6fdbd8 | 1012 | |
a9ec3454 | 1013 | min_seq = min(ja->bucket_seq[i], min_seq); |
1c6fdbd8 KO |
1014 | } |
1015 | ||
1c6fdbd8 | 1016 | /* |
1c6fdbd8 KO |
1017 | * If there's duplicate journal entries in multiple buckets (which |
1018 | * definitely isn't supposed to happen, but...) - make sure to start | |
1019 | * cur_idx at the last of those buckets, so we don't deadlock trying to | |
1020 | * allocate | |
1021 | */ | |
a9ec3454 | 1022 | while (ja->bucket_seq[ja->cur_idx] > min_seq && |
062afcba | 1023 | ja->bucket_seq[ja->cur_idx] == |
a9ec3454 | 1024 | ja->bucket_seq[(ja->cur_idx + 1) % ja->nr]) |
a36d3685 | 1025 | ja->cur_idx = (ja->cur_idx + 1) % ja->nr; |
a9ec3454 | 1026 | |
062afcba KO |
1027 | ja->sectors_free = ca->mi.bucket_size; |
1028 | ||
1029 | mutex_lock(&jlist->lock); | |
ce6201c4 KO |
1030 | genradix_for_each(&c->journal_entries, iter, _r) { |
1031 | r = *_r; | |
1032 | ||
1033 | if (!r) | |
1034 | continue; | |
1035 | ||
062afcba KO |
1036 | for (i = 0; i < r->nr_ptrs; i++) { |
1037 | if (r->ptrs[i].dev == ca->dev_idx && | |
1038 | sector_to_bucket(ca, r->ptrs[i].sector) == ja->buckets[ja->cur_idx]) { | |
502f973d | 1039 | unsigned wrote = bucket_remainder(ca, r->ptrs[i].sector) + |
062afcba KO |
1040 | vstruct_sectors(&r->j, c->block_bits); |
1041 | ||
1042 | ja->sectors_free = min(ja->sectors_free, | |
1043 | ca->mi.bucket_size - wrote); | |
1044 | } | |
1045 | } | |
1046 | } | |
1047 | mutex_unlock(&jlist->lock); | |
1048 | ||
b0be2fcf KO |
1049 | if (ja->bucket_seq[ja->cur_idx] && |
1050 | ja->sectors_free == ca->mi.bucket_size) { | |
1051 | bch_err(c, "ja->sectors_free == ca->mi.bucket_size"); | |
1052 | bch_err(c, "cur_idx %u/%u", ja->cur_idx, ja->nr); | |
1053 | for (i = 0; i < 3; i++) { | |
1054 | unsigned idx = (ja->cur_idx + ja->nr - 1 + i) % ja->nr; | |
1055 | bch_err(c, "bucket_seq[%u] = %llu", idx, ja->bucket_seq[idx]); | |
1056 | } | |
1057 | ja->sectors_free = 0; | |
1058 | } | |
1c6fdbd8 KO |
1059 | |
1060 | /* | |
0ce2dbbe | 1061 | * Set dirty_idx to indicate the entire journal is full and needs to be |
1c6fdbd8 KO |
1062 | * reclaimed - journal reclaim will immediately reclaim whatever isn't |
1063 | * pinned when it first runs: | |
1064 | */ | |
0ce2dbbe KO |
1065 | ja->discard_idx = ja->dirty_idx_ondisk = |
1066 | ja->dirty_idx = (ja->cur_idx + 1) % ja->nr; | |
1c6fdbd8 | 1067 | out: |
365f64f3 | 1068 | bch_verbose(c, "journal read done on device %s, ret %i", ca->name, ret); |
1c6fdbd8 | 1069 | kvpfree(buf.data, buf.size); |
1c6fdbd8 KO |
1070 | percpu_ref_put(&ca->io_ref); |
1071 | closure_return(cl); | |
1072 | return; | |
1073 | err: | |
1074 | mutex_lock(&jlist->lock); | |
1075 | jlist->ret = ret; | |
1076 | mutex_unlock(&jlist->lock); | |
1077 | goto out; | |
1c6fdbd8 KO |
1078 | } |
1079 | ||
72b7d633 KO |
1080 | void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c, |
1081 | struct journal_replay *j) | |
e4c3f386 KO |
1082 | { |
1083 | unsigned i; | |
1084 | ||
1085 | for (i = 0; i < j->nr_ptrs; i++) { | |
c0ebe3e4 | 1086 | struct bch_dev *ca = bch_dev_bkey_exists(c, j->ptrs[i].dev); |
514852c2 KO |
1087 | u64 offset; |
1088 | ||
72b7d633 | 1089 | div64_u64_rem(j->ptrs[i].sector, ca->mi.bucket_size, &offset); |
e4c3f386 KO |
1090 | |
1091 | if (i) | |
401ec4db KO |
1092 | prt_printf(out, " "); |
1093 | prt_printf(out, "%u:%u:%u (sector %llu)", | |
e4c3f386 | 1094 | j->ptrs[i].dev, |
72b7d633 KO |
1095 | j->ptrs[i].bucket, |
1096 | j->ptrs[i].bucket_offset, | |
1097 | j->ptrs[i].sector); | |
e4c3f386 KO |
1098 | } |
1099 | } | |
1100 | ||
ce6201c4 | 1101 | int bch2_journal_read(struct bch_fs *c, u64 *blacklist_seq, u64 *start_seq) |
1c6fdbd8 | 1102 | { |
1c6fdbd8 | 1103 | struct journal_list jlist; |
ce6201c4 KO |
1104 | struct journal_replay *i, **_i, *prev = NULL; |
1105 | struct genradix_iter radix_iter; | |
1c6fdbd8 | 1106 | struct bch_dev *ca; |
1c6fdbd8 | 1107 | unsigned iter; |
fa8e94fa | 1108 | struct printbuf buf = PRINTBUF; |
1c6fdbd8 KO |
1109 | size_t keys = 0, entries = 0; |
1110 | bool degraded = false; | |
adbcada4 | 1111 | u64 seq, last_seq = 0; |
1c6fdbd8 KO |
1112 | int ret = 0; |
1113 | ||
1114 | closure_init_stack(&jlist.cl); | |
1115 | mutex_init(&jlist.lock); | |
ec7ccbde | 1116 | jlist.last_seq = 0; |
1c6fdbd8 KO |
1117 | jlist.ret = 0; |
1118 | ||
1119 | for_each_member_device(ca, c, iter) { | |
75c8d030 | 1120 | if (!c->opts.fsck && |
89fd25be | 1121 | !(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_journal))) |
1c6fdbd8 KO |
1122 | continue; |
1123 | ||
2436cb9f KO |
1124 | if ((ca->mi.state == BCH_MEMBER_STATE_rw || |
1125 | ca->mi.state == BCH_MEMBER_STATE_ro) && | |
1c6fdbd8 KO |
1126 | percpu_ref_tryget(&ca->io_ref)) |
1127 | closure_call(&ca->journal.read, | |
1128 | bch2_journal_read_device, | |
1129 | system_unbound_wq, | |
1130 | &jlist.cl); | |
1131 | else | |
1132 | degraded = true; | |
1133 | } | |
1134 | ||
1135 | closure_sync(&jlist.cl); | |
1136 | ||
1137 | if (jlist.ret) | |
1138 | return jlist.ret; | |
1139 | ||
ce6201c4 | 1140 | *start_seq = 0; |
adbcada4 KO |
1141 | |
1142 | /* | |
1143 | * Find most recent flush entry, and ignore newer non flush entries - | |
1144 | * those entries will be blacklisted: | |
1145 | */ | |
ce6201c4 KO |
1146 | genradix_for_each_reverse(&c->journal_entries, radix_iter, _i) { |
1147 | i = *_i; | |
1148 | ||
1149 | if (!i || i->ignore) | |
adbcada4 KO |
1150 | continue; |
1151 | ||
ce6201c4 KO |
1152 | if (!*start_seq) |
1153 | *start_seq = le64_to_cpu(i->j.seq) + 1; | |
1154 | ||
adbcada4 KO |
1155 | if (!JSET_NO_FLUSH(&i->j)) { |
1156 | last_seq = le64_to_cpu(i->j.last_seq); | |
1157 | *blacklist_seq = le64_to_cpu(i->j.seq) + 1; | |
1158 | break; | |
1159 | } | |
1160 | ||
1161 | journal_replay_free(c, i); | |
1162 | } | |
1163 | ||
ce6201c4 KO |
1164 | if (!*start_seq) { |
1165 | bch_info(c, "journal read done, but no entries found"); | |
1166 | return 0; | |
1167 | } | |
1168 | ||
adbcada4 KO |
1169 | if (!last_seq) { |
1170 | fsck_err(c, "journal read done, but no entries found after dropping non-flushes"); | |
fa8e94fa KO |
1171 | ret = -1; |
1172 | goto err; | |
adbcada4 KO |
1173 | } |
1174 | ||
1175 | /* Drop blacklisted entries and entries older than last_seq: */ | |
ce6201c4 KO |
1176 | genradix_for_each(&c->journal_entries, radix_iter, _i) { |
1177 | i = *_i; | |
1178 | ||
1179 | if (!i || i->ignore) | |
adbcada4 KO |
1180 | continue; |
1181 | ||
1182 | seq = le64_to_cpu(i->j.seq); | |
1183 | if (seq < last_seq) { | |
1184 | journal_replay_free(c, i); | |
1185 | continue; | |
1186 | } | |
1187 | ||
1188 | if (bch2_journal_seq_is_blacklisted(c, seq, true)) { | |
1189 | fsck_err_on(!JSET_NO_FLUSH(&i->j), c, | |
1190 | "found blacklisted journal entry %llu", seq); | |
1191 | ||
1192 | journal_replay_free(c, i); | |
1193 | } | |
1194 | } | |
1195 | ||
1196 | /* Check for missing entries: */ | |
1197 | seq = last_seq; | |
ce6201c4 KO |
1198 | genradix_for_each(&c->journal_entries, radix_iter, _i) { |
1199 | i = *_i; | |
1200 | ||
1201 | if (!i || i->ignore) | |
adbcada4 KO |
1202 | continue; |
1203 | ||
1204 | BUG_ON(seq > le64_to_cpu(i->j.seq)); | |
1205 | ||
1206 | while (seq < le64_to_cpu(i->j.seq)) { | |
1207 | u64 missing_start, missing_end; | |
fa8e94fa | 1208 | struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF; |
adbcada4 KO |
1209 | |
1210 | while (seq < le64_to_cpu(i->j.seq) && | |
1211 | bch2_journal_seq_is_blacklisted(c, seq, false)) | |
1212 | seq++; | |
1213 | ||
1214 | if (seq == le64_to_cpu(i->j.seq)) | |
1215 | break; | |
1216 | ||
1217 | missing_start = seq; | |
1218 | ||
1219 | while (seq < le64_to_cpu(i->j.seq) && | |
1220 | !bch2_journal_seq_is_blacklisted(c, seq, false)) | |
1221 | seq++; | |
1222 | ||
ce6201c4 KO |
1223 | if (prev) { |
1224 | bch2_journal_ptrs_to_text(&buf1, c, prev); | |
401ec4db | 1225 | prt_printf(&buf1, " size %zu", vstruct_sectors(&prev->j, c->block_bits)); |
e4c3f386 | 1226 | } else |
401ec4db | 1227 | prt_printf(&buf1, "(none)"); |
fa8e94fa | 1228 | bch2_journal_ptrs_to_text(&buf2, c, i); |
e4c3f386 | 1229 | |
adbcada4 | 1230 | missing_end = seq - 1; |
e4c3f386 KO |
1231 | fsck_err(c, "journal entries %llu-%llu missing! (replaying %llu-%llu)\n" |
1232 | " prev at %s\n" | |
1233 | " next at %s", | |
adbcada4 | 1234 | missing_start, missing_end, |
e4c3f386 | 1235 | last_seq, *blacklist_seq - 1, |
fa8e94fa KO |
1236 | buf1.buf, buf2.buf); |
1237 | ||
1238 | printbuf_exit(&buf1); | |
1239 | printbuf_exit(&buf2); | |
adbcada4 KO |
1240 | } |
1241 | ||
ce6201c4 | 1242 | prev = i; |
adbcada4 KO |
1243 | seq++; |
1244 | } | |
1245 | ||
ce6201c4 | 1246 | genradix_for_each(&c->journal_entries, radix_iter, _i) { |
1dd7f9d9 KO |
1247 | struct jset_entry *entry; |
1248 | struct bkey_i *k, *_n; | |
e4c3f386 KO |
1249 | struct bch_replicas_padded replicas = { |
1250 | .e.data_type = BCH_DATA_journal, | |
1251 | .e.nr_required = 1, | |
1252 | }; | |
1253 | unsigned ptr; | |
7ef2a73a | 1254 | |
ce6201c4 KO |
1255 | i = *_i; |
1256 | if (!i || i->ignore) | |
adbcada4 KO |
1257 | continue; |
1258 | ||
1c6fdbd8 KO |
1259 | ret = jset_validate_entries(c, &i->j, READ); |
1260 | if (ret) | |
fa8e94fa | 1261 | goto err; |
1c6fdbd8 | 1262 | |
e4c3f386 KO |
1263 | for (ptr = 0; ptr < i->nr_ptrs; ptr++) |
1264 | replicas.e.devs[replicas.e.nr_devs++] = i->ptrs[ptr].dev; | |
1265 | ||
26452d1d KO |
1266 | bch2_replicas_entry_sort(&replicas.e); |
1267 | ||
1c6fdbd8 KO |
1268 | /* |
1269 | * If we're mounting in degraded mode - if we didn't read all | |
1270 | * the devices - this is wrong: | |
1271 | */ | |
1272 | ||
fa8e94fa KO |
1273 | printbuf_reset(&buf); |
1274 | bch2_replicas_entry_to_text(&buf, &replicas.e); | |
1275 | ||
1c6fdbd8 | 1276 | if (!degraded && |
75c8d030 KO |
1277 | fsck_err_on(!bch2_replicas_marked(c, &replicas.e), c, |
1278 | "superblock not marked as containing replicas %s", | |
1279 | buf.buf)) { | |
7ef2a73a | 1280 | ret = bch2_mark_replicas(c, &replicas.e); |
1c6fdbd8 | 1281 | if (ret) |
fa8e94fa | 1282 | goto err; |
1c6fdbd8 | 1283 | } |
1c6fdbd8 KO |
1284 | |
1285 | for_each_jset_key(k, _n, entry, &i->j) | |
1286 | keys++; | |
1287 | entries++; | |
1288 | } | |
1289 | ||
adbcada4 KO |
1290 | bch_info(c, "journal read done, %zu keys in %zu entries, seq %llu", |
1291 | keys, entries, *start_seq); | |
1dd7f9d9 | 1292 | |
adbcada4 KO |
1293 | if (*start_seq != *blacklist_seq) |
1294 | bch_info(c, "dropped unflushed entries %llu-%llu", | |
1295 | *blacklist_seq, *start_seq - 1); | |
fa8e94fa | 1296 | err: |
1c6fdbd8 | 1297 | fsck_err: |
fa8e94fa | 1298 | printbuf_exit(&buf); |
1c6fdbd8 KO |
1299 | return ret; |
1300 | } | |
1301 | ||
1c6fdbd8 KO |
1302 | /* journal write: */ |
1303 | ||
a9ec3454 KO |
1304 | static void __journal_write_alloc(struct journal *j, |
1305 | struct journal_buf *w, | |
1306 | struct dev_alloc_list *devs_sorted, | |
1307 | unsigned sectors, | |
1308 | unsigned *replicas, | |
1309 | unsigned replicas_want) | |
1c6fdbd8 KO |
1310 | { |
1311 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
1c6fdbd8 KO |
1312 | struct journal_device *ja; |
1313 | struct bch_dev *ca; | |
a9ec3454 | 1314 | unsigned i; |
a2753581 | 1315 | |
a9ec3454 KO |
1316 | if (*replicas >= replicas_want) |
1317 | return; | |
1c6fdbd8 | 1318 | |
a9ec3454 KO |
1319 | for (i = 0; i < devs_sorted->nr; i++) { |
1320 | ca = rcu_dereference(c->devs[devs_sorted->devs[i]]); | |
1c6fdbd8 KO |
1321 | if (!ca) |
1322 | continue; | |
1323 | ||
1c6fdbd8 | 1324 | ja = &ca->journal; |
1c6fdbd8 KO |
1325 | |
1326 | /* | |
1327 | * Check that we can use this device, and aren't already using | |
1328 | * it: | |
1329 | */ | |
a9ec3454 | 1330 | if (!ca->mi.durability || |
2436cb9f | 1331 | ca->mi.state != BCH_MEMBER_STATE_rw || |
a9ec3454 | 1332 | !ja->nr || |
26609b61 KO |
1333 | bch2_bkey_has_device(bkey_i_to_s_c(&w->key), |
1334 | ca->dev_idx) || | |
a9ec3454 | 1335 | sectors > ja->sectors_free) |
1c6fdbd8 KO |
1336 | continue; |
1337 | ||
3d080aa5 | 1338 | bch2_dev_stripe_increment(ca, &j->wp.stripe); |
1c6fdbd8 | 1339 | |
26609b61 | 1340 | bch2_bkey_append_ptr(&w->key, |
1c6fdbd8 KO |
1341 | (struct bch_extent_ptr) { |
1342 | .offset = bucket_to_sector(ca, | |
a9ec3454 KO |
1343 | ja->buckets[ja->cur_idx]) + |
1344 | ca->mi.bucket_size - | |
1345 | ja->sectors_free, | |
1c6fdbd8 KO |
1346 | .dev = ca->dev_idx, |
1347 | }); | |
1348 | ||
a9ec3454 KO |
1349 | ja->sectors_free -= sectors; |
1350 | ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq); | |
1351 | ||
1352 | *replicas += ca->mi.durability; | |
1353 | ||
1354 | if (*replicas >= replicas_want) | |
1355 | break; | |
1c6fdbd8 | 1356 | } |
a9ec3454 | 1357 | } |
1c6fdbd8 | 1358 | |
a9ec3454 KO |
1359 | /** |
1360 | * journal_next_bucket - move on to the next journal bucket if possible | |
1361 | */ | |
1362 | static int journal_write_alloc(struct journal *j, struct journal_buf *w, | |
1363 | unsigned sectors) | |
1364 | { | |
1365 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
d042b040 | 1366 | struct bch_devs_mask devs; |
a9ec3454 KO |
1367 | struct journal_device *ja; |
1368 | struct bch_dev *ca; | |
1369 | struct dev_alloc_list devs_sorted; | |
d042b040 KO |
1370 | unsigned target = c->opts.metadata_target ?: |
1371 | c->opts.foreground_target; | |
a9ec3454 KO |
1372 | unsigned i, replicas = 0, replicas_want = |
1373 | READ_ONCE(c->opts.metadata_replicas); | |
1c6fdbd8 | 1374 | |
a9ec3454 | 1375 | rcu_read_lock(); |
d042b040 KO |
1376 | retry: |
1377 | devs = target_rw_devs(c, BCH_DATA_journal, target); | |
1c6fdbd8 | 1378 | |
d042b040 | 1379 | devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe, &devs); |
1c6fdbd8 | 1380 | |
a9ec3454 KO |
1381 | __journal_write_alloc(j, w, &devs_sorted, |
1382 | sectors, &replicas, replicas_want); | |
1c6fdbd8 | 1383 | |
a9ec3454 KO |
1384 | if (replicas >= replicas_want) |
1385 | goto done; | |
1386 | ||
1387 | for (i = 0; i < devs_sorted.nr; i++) { | |
1388 | ca = rcu_dereference(c->devs[devs_sorted.devs[i]]); | |
1389 | if (!ca) | |
1390 | continue; | |
1391 | ||
1392 | ja = &ca->journal; | |
1393 | ||
1394 | if (sectors > ja->sectors_free && | |
1395 | sectors <= ca->mi.bucket_size && | |
03d5eaed KO |
1396 | bch2_journal_dev_buckets_available(j, ja, |
1397 | journal_space_discarded)) { | |
a9ec3454 KO |
1398 | ja->cur_idx = (ja->cur_idx + 1) % ja->nr; |
1399 | ja->sectors_free = ca->mi.bucket_size; | |
68ef94a6 KO |
1400 | |
1401 | /* | |
1402 | * ja->bucket_seq[ja->cur_idx] must always have | |
1403 | * something sensible: | |
1404 | */ | |
1405 | ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq); | |
a9ec3454 KO |
1406 | } |
1407 | } | |
1408 | ||
1409 | __journal_write_alloc(j, w, &devs_sorted, | |
1410 | sectors, &replicas, replicas_want); | |
d042b040 KO |
1411 | |
1412 | if (replicas < replicas_want && target) { | |
1413 | /* Retry from all devices: */ | |
1414 | target = 0; | |
1415 | goto retry; | |
1416 | } | |
a9ec3454 | 1417 | done: |
a9ec3454 KO |
1418 | rcu_read_unlock(); |
1419 | ||
07a1006a KO |
1420 | BUG_ON(bkey_val_u64s(&w->key.k) > BCH_REPLICAS_MAX); |
1421 | ||
57cb2142 | 1422 | return replicas >= c->opts.metadata_replicas_required ? 0 : -EROFS; |
1c6fdbd8 KO |
1423 | } |
1424 | ||
1c6fdbd8 KO |
1425 | static void journal_buf_realloc(struct journal *j, struct journal_buf *buf) |
1426 | { | |
1427 | /* we aren't holding j->lock: */ | |
1428 | unsigned new_size = READ_ONCE(j->buf_size_want); | |
1429 | void *new_buf; | |
1430 | ||
d16b4a77 | 1431 | if (buf->buf_size >= new_size) |
1c6fdbd8 KO |
1432 | return; |
1433 | ||
1434 | new_buf = kvpmalloc(new_size, GFP_NOIO|__GFP_NOWARN); | |
1435 | if (!new_buf) | |
1436 | return; | |
1437 | ||
d16b4a77 | 1438 | memcpy(new_buf, buf->data, buf->buf_size); |
c859430b KO |
1439 | |
1440 | spin_lock(&j->lock); | |
1441 | swap(buf->data, new_buf); | |
1442 | swap(buf->buf_size, new_size); | |
1443 | spin_unlock(&j->lock); | |
1444 | ||
1445 | kvpfree(new_buf, new_size); | |
1c6fdbd8 KO |
1446 | } |
1447 | ||
ebb84d09 KO |
1448 | static inline struct journal_buf *journal_last_unwritten_buf(struct journal *j) |
1449 | { | |
30ef633a | 1450 | return j->buf + (journal_last_unwritten_seq(j) & JOURNAL_BUF_MASK); |
ebb84d09 KO |
1451 | } |
1452 | ||
1c6fdbd8 KO |
1453 | static void journal_write_done(struct closure *cl) |
1454 | { | |
1455 | struct journal *j = container_of(cl, struct journal, io); | |
1456 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
ebb84d09 | 1457 | struct journal_buf *w = journal_last_unwritten_buf(j); |
7ef2a73a | 1458 | struct bch_replicas_padded replicas; |
ebb84d09 | 1459 | union journal_res_state old, new; |
1784d43a | 1460 | u64 v, seq; |
158eecb8 | 1461 | int err = 0; |
1c6fdbd8 | 1462 | |
991ba021 KO |
1463 | bch2_time_stats_update(!JSET_NO_FLUSH(w->data) |
1464 | ? j->flush_write_time | |
1465 | : j->noflush_write_time, j->write_start_time); | |
9c859dc9 | 1466 | |
d797ca3d | 1467 | if (!w->devs_written.nr) { |
1c6fdbd8 | 1468 | bch_err(c, "unable to write journal to sufficient devices"); |
158eecb8 KO |
1469 | err = -EIO; |
1470 | } else { | |
d797ca3d KO |
1471 | bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal, |
1472 | w->devs_written); | |
158eecb8 KO |
1473 | if (bch2_mark_replicas(c, &replicas.e)) |
1474 | err = -EIO; | |
1c6fdbd8 KO |
1475 | } |
1476 | ||
158eecb8 KO |
1477 | if (err) |
1478 | bch2_fatal_error(c); | |
1c6fdbd8 KO |
1479 | |
1480 | spin_lock(&j->lock); | |
ed9d58a2 | 1481 | seq = le64_to_cpu(w->data->seq); |
ed9d58a2 | 1482 | |
1c6fdbd8 | 1483 | if (seq >= j->pin.front) |
d797ca3d | 1484 | journal_seq_pin(j, seq)->devs = w->devs_written; |
1c6fdbd8 | 1485 | |
9be1efe9 | 1486 | if (!err) { |
9be1efe9 KO |
1487 | if (!JSET_NO_FLUSH(w->data)) { |
1488 | j->flushed_seq_ondisk = seq; | |
1489 | j->last_seq_ondisk = w->last_seq; | |
f25d8215 | 1490 | |
59cc38b8 | 1491 | bch2_do_discards(c); |
f25d8215 KO |
1492 | closure_wake_up(&c->freelist_wait); |
1493 | ||
1494 | bch2_reset_alloc_cursors(c); | |
9be1efe9 KO |
1495 | } |
1496 | } else if (!j->err_seq || seq < j->err_seq) | |
1497 | j->err_seq = seq; | |
0ce2dbbe | 1498 | |
f0a3a2cc KO |
1499 | j->seq_ondisk = seq; |
1500 | ||
1c6fdbd8 KO |
1501 | /* |
1502 | * Updating last_seq_ondisk may let bch2_journal_reclaim_work() discard | |
1503 | * more buckets: | |
1504 | * | |
1505 | * Must come before signaling write completion, for | |
1506 | * bch2_fs_journal_stop(): | |
1507 | */ | |
8cc052db KO |
1508 | if (j->watermark) |
1509 | journal_reclaim_kick(&c->journal); | |
158eecb8 | 1510 | |
1c6fdbd8 KO |
1511 | /* also must come before signalling write completion: */ |
1512 | closure_debug_destroy(cl); | |
1513 | ||
ebb84d09 KO |
1514 | v = atomic64_read(&j->reservations.counter); |
1515 | do { | |
1516 | old.v = new.v = v; | |
24a3d53b | 1517 | BUG_ON(journal_state_count(new, new.unwritten_idx)); |
ebb84d09 KO |
1518 | |
1519 | new.unwritten_idx++; | |
1520 | } while ((v = atomic64_cmpxchg(&j->reservations.counter, | |
1521 | old.v, new.v)) != old.v); | |
1c6fdbd8 | 1522 | |
5d32c5bb KO |
1523 | bch2_journal_space_available(j); |
1524 | ||
1c6fdbd8 KO |
1525 | closure_wake_up(&w->wait); |
1526 | journal_wake(j); | |
1527 | ||
24a3d53b KO |
1528 | if (!journal_state_count(new, new.unwritten_idx) && |
1529 | journal_last_unwritten_seq(j) <= journal_cur_seq(j)) { | |
1530 | closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL); | |
1531 | } else if (journal_last_unwritten_seq(j) == journal_cur_seq(j) && | |
1532 | new.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL) { | |
fbec3b88 KO |
1533 | struct journal_buf *buf = journal_cur_buf(j); |
1534 | long delta = buf->expires - jiffies; | |
ebb84d09 | 1535 | |
24a3d53b KO |
1536 | /* |
1537 | * We don't close a journal entry to write it while there's | |
1538 | * previous entries still in flight - the current journal entry | |
1539 | * might want to be written now: | |
1540 | */ | |
1541 | ||
fbec3b88 | 1542 | mod_delayed_work(c->io_complete_wq, &j->write_work, max(0L, delta)); |
24a3d53b | 1543 | } |
fbec3b88 KO |
1544 | |
1545 | spin_unlock(&j->lock); | |
1c6fdbd8 KO |
1546 | } |
1547 | ||
1548 | static void journal_write_endio(struct bio *bio) | |
1549 | { | |
1550 | struct bch_dev *ca = bio->bi_private; | |
1551 | struct journal *j = &ca->fs->journal; | |
d797ca3d KO |
1552 | struct journal_buf *w = journal_last_unwritten_buf(j); |
1553 | unsigned long flags; | |
1c6fdbd8 | 1554 | |
d797ca3d KO |
1555 | if (bch2_dev_io_err_on(bio->bi_status, ca, "error writing journal entry %llu: %s", |
1556 | le64_to_cpu(w->data->seq), | |
63b214e7 | 1557 | bch2_blk_status_to_str(bio->bi_status)) || |
1c6fdbd8 | 1558 | bch2_meta_write_fault("journal")) { |
1c6fdbd8 | 1559 | spin_lock_irqsave(&j->err_lock, flags); |
d797ca3d | 1560 | bch2_dev_list_drop_dev(&w->devs_written, ca->dev_idx); |
1c6fdbd8 KO |
1561 | spin_unlock_irqrestore(&j->err_lock, flags); |
1562 | } | |
1563 | ||
1564 | closure_put(&j->io); | |
1565 | percpu_ref_put(&ca->io_ref); | |
1566 | } | |
1567 | ||
280249b9 KO |
1568 | static void do_journal_write(struct closure *cl) |
1569 | { | |
1570 | struct journal *j = container_of(cl, struct journal, io); | |
1571 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
1572 | struct bch_dev *ca; | |
1573 | struct journal_buf *w = journal_last_unwritten_buf(j); | |
1574 | struct bch_extent_ptr *ptr; | |
1575 | struct bio *bio; | |
1576 | unsigned sectors = vstruct_sectors(w->data, c->block_bits); | |
1577 | ||
1578 | extent_for_each_ptr(bkey_i_to_s_extent(&w->key), ptr) { | |
1579 | ca = bch_dev_bkey_exists(c, ptr->dev); | |
1580 | if (!percpu_ref_tryget(&ca->io_ref)) { | |
1581 | /* XXX: fix this */ | |
1582 | bch_err(c, "missing device for journal write\n"); | |
1583 | continue; | |
1584 | } | |
1585 | ||
1586 | this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_journal], | |
1587 | sectors); | |
1588 | ||
1589 | bio = ca->journal.bio; | |
1590 | bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META); | |
1591 | bio->bi_iter.bi_sector = ptr->offset; | |
1592 | bio->bi_end_io = journal_write_endio; | |
1593 | bio->bi_private = ca; | |
1594 | ||
a28bd48a KO |
1595 | BUG_ON(bio->bi_iter.bi_sector == ca->prev_journal_sector); |
1596 | ca->prev_journal_sector = bio->bi_iter.bi_sector; | |
1597 | ||
280249b9 KO |
1598 | if (!JSET_NO_FLUSH(w->data)) |
1599 | bio->bi_opf |= REQ_FUA; | |
1600 | if (!JSET_NO_FLUSH(w->data) && !w->separate_flush) | |
1601 | bio->bi_opf |= REQ_PREFLUSH; | |
1602 | ||
1603 | bch2_bio_map(bio, w->data, sectors << 9); | |
1604 | ||
674cfc26 | 1605 | trace_and_count(c, journal_write, bio); |
280249b9 KO |
1606 | closure_bio_submit(bio, cl); |
1607 | ||
1608 | ca->journal.bucket_seq[ca->journal.cur_idx] = | |
1609 | le64_to_cpu(w->data->seq); | |
1610 | } | |
1611 | ||
731bdd2e | 1612 | continue_at(cl, journal_write_done, c->io_complete_wq); |
280249b9 KO |
1613 | return; |
1614 | } | |
1615 | ||
1c6fdbd8 KO |
1616 | void bch2_journal_write(struct closure *cl) |
1617 | { | |
1618 | struct journal *j = container_of(cl, struct journal, io); | |
1619 | struct bch_fs *c = container_of(j, struct bch_fs, journal); | |
1620 | struct bch_dev *ca; | |
ebb84d09 | 1621 | struct journal_buf *w = journal_last_unwritten_buf(j); |
3ccc5c50 | 1622 | struct jset_entry *start, *end; |
1c6fdbd8 KO |
1623 | struct jset *jset; |
1624 | struct bio *bio; | |
fa8e94fa | 1625 | struct printbuf journal_debug_buf = PRINTBUF; |
26609b61 | 1626 | bool validate_before_checksum = false; |
280249b9 | 1627 | unsigned i, sectors, bytes, u64s, nr_rw_members = 0; |
e5a66496 KO |
1628 | int ret; |
1629 | ||
b7a9bbfc KO |
1630 | BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb)); |
1631 | ||
1c6fdbd8 KO |
1632 | journal_buf_realloc(j, w); |
1633 | jset = w->data; | |
1634 | ||
1635 | j->write_start_time = local_clock(); | |
1c6fdbd8 | 1636 | |
adbcada4 | 1637 | spin_lock(&j->lock); |
e0c014e7 KO |
1638 | if (bch2_journal_error(j) || |
1639 | w->noflush || | |
1640 | (!w->must_flush && | |
1641 | (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) && | |
1642 | test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags))) { | |
adbcada4 KO |
1643 | w->noflush = true; |
1644 | SET_JSET_NO_FLUSH(jset, true); | |
c0ebe3e4 KO |
1645 | jset->last_seq = 0; |
1646 | w->last_seq = 0; | |
adbcada4 KO |
1647 | |
1648 | j->nr_noflush_writes++; | |
1649 | } else { | |
1650 | j->last_flush_write = jiffies; | |
1651 | j->nr_flush_writes++; | |
1652 | } | |
1653 | spin_unlock(&j->lock); | |
1654 | ||
00b8ccf7 KO |
1655 | /* |
1656 | * New btree roots are set by journalling them; when the journal entry | |
1657 | * gets written we have to propagate them to c->btree_roots | |
1658 | * | |
1659 | * But, every journal entry we write has to contain all the btree roots | |
1660 | * (at least for now); so after we copy btree roots to c->btree_roots we | |
1661 | * have to get any missing btree roots and add them to this journal | |
1662 | * entry: | |
1663 | */ | |
1664 | ||
1665 | bch2_journal_entries_to_btree_roots(c, jset); | |
1666 | ||
1667 | start = end = vstruct_last(jset); | |
1668 | ||
1669 | end = bch2_btree_roots_to_journal_entries(c, jset->start, end); | |
1670 | ||
2abe5420 KO |
1671 | bch2_journal_super_entries_add_common(c, &end, |
1672 | le64_to_cpu(jset->seq)); | |
3ccc5c50 KO |
1673 | u64s = (u64 *) end - (u64 *) start; |
1674 | BUG_ON(u64s > j->entry_u64s_reserved); | |
1675 | ||
d16b4a77 KO |
1676 | le32_add_cpu(&jset->u64s, u64s); |
1677 | BUG_ON(vstruct_sectors(jset, c->block_bits) > w->sectors); | |
1c6fdbd8 | 1678 | |
1c6fdbd8 | 1679 | jset->magic = cpu_to_le64(jset_magic(c)); |
74b33393 | 1680 | jset->version = c->sb.version < bcachefs_metadata_version_bkey_renumber |
26609b61 KO |
1681 | ? cpu_to_le32(BCH_JSET_VERSION_OLD) |
1682 | : cpu_to_le32(c->sb.version); | |
1c6fdbd8 KO |
1683 | |
1684 | SET_JSET_BIG_ENDIAN(jset, CPU_BIG_ENDIAN); | |
1685 | SET_JSET_CSUM_TYPE(jset, bch2_meta_checksum_type(c)); | |
1686 | ||
4141fde0 | 1687 | if (!JSET_NO_FLUSH(jset) && journal_entry_empty(jset)) |
158eecb8 KO |
1688 | j->last_empty_seq = le64_to_cpu(jset->seq); |
1689 | ||
26609b61 KO |
1690 | if (bch2_csum_type_is_encryption(JSET_CSUM_TYPE(jset))) |
1691 | validate_before_checksum = true; | |
1692 | ||
e751c01a | 1693 | if (le32_to_cpu(jset->version) < bcachefs_metadata_version_current) |
26609b61 KO |
1694 | validate_before_checksum = true; |
1695 | ||
1696 | if (validate_before_checksum && | |
ed9d58a2 | 1697 | jset_validate_for_write(c, jset)) |
1c6fdbd8 KO |
1698 | goto err; |
1699 | ||
a9de137b | 1700 | ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset), |
1c6fdbd8 KO |
1701 | jset->encrypted_start, |
1702 | vstruct_end(jset) - (void *) jset->encrypted_start); | |
a9de137b KO |
1703 | if (bch2_fs_fatal_err_on(ret, c, |
1704 | "error decrypting journal entry: %i", ret)) | |
1705 | goto err; | |
1c6fdbd8 KO |
1706 | |
1707 | jset->csum = csum_vstruct(c, JSET_CSUM_TYPE(jset), | |
1708 | journal_nonce(jset), jset); | |
1709 | ||
26609b61 | 1710 | if (!validate_before_checksum && |
ed9d58a2 | 1711 | jset_validate_for_write(c, jset)) |
1c6fdbd8 KO |
1712 | goto err; |
1713 | ||
1714 | sectors = vstruct_sectors(jset, c->block_bits); | |
d16b4a77 | 1715 | BUG_ON(sectors > w->sectors); |
1c6fdbd8 | 1716 | |
d16b4a77 KO |
1717 | bytes = vstruct_bytes(jset); |
1718 | memset((void *) jset + bytes, 0, (sectors << 9) - bytes); | |
1c6fdbd8 | 1719 | |
c18dade6 | 1720 | retry_alloc: |
e5a66496 KO |
1721 | spin_lock(&j->lock); |
1722 | ret = journal_write_alloc(j, w, sectors); | |
1723 | ||
c18dade6 KO |
1724 | if (ret && j->can_discard) { |
1725 | spin_unlock(&j->lock); | |
1726 | bch2_journal_do_discards(j); | |
1727 | goto retry_alloc; | |
1728 | } | |
1729 | ||
fa8e94fa KO |
1730 | if (ret) |
1731 | __bch2_journal_debug_to_text(&journal_debug_buf, j); | |
85674154 | 1732 | |
e5a66496 KO |
1733 | /* |
1734 | * write is allocated, no longer need to account for it in | |
1735 | * bch2_journal_space_available(): | |
1736 | */ | |
1737 | w->sectors = 0; | |
1738 | ||
1739 | /* | |
1740 | * journal entry has been compacted and allocated, recalculate space | |
1741 | * available: | |
1742 | */ | |
1743 | bch2_journal_space_available(j); | |
1744 | spin_unlock(&j->lock); | |
1745 | ||
1746 | if (ret) { | |
85674154 | 1747 | bch_err(c, "Unable to allocate journal write:\n%s", |
fa8e94fa KO |
1748 | journal_debug_buf.buf); |
1749 | printbuf_exit(&journal_debug_buf); | |
1c6fdbd8 | 1750 | bch2_fatal_error(c); |
731bdd2e | 1751 | continue_at(cl, journal_write_done, c->io_complete_wq); |
1c6fdbd8 KO |
1752 | return; |
1753 | } | |
1754 | ||
d797ca3d KO |
1755 | w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key)); |
1756 | ||
b66b2bc0 | 1757 | if (c->opts.nochanges) |
1c6fdbd8 KO |
1758 | goto no_io; |
1759 | ||
280249b9 KO |
1760 | for_each_rw_member(ca, c, i) |
1761 | nr_rw_members++; | |
1c6fdbd8 | 1762 | |
280249b9 KO |
1763 | if (nr_rw_members > 1) |
1764 | w->separate_flush = true; | |
1c6fdbd8 | 1765 | |
280249b9 KO |
1766 | if (!JSET_NO_FLUSH(jset) && w->separate_flush) { |
1767 | for_each_rw_member(ca, c, i) { | |
1768 | percpu_ref_get(&ca->io_ref); | |
1c6fdbd8 | 1769 | |
280249b9 KO |
1770 | bio = ca->journal.bio; |
1771 | bio_reset(bio, ca->disk_sb.bdev, REQ_OP_FLUSH); | |
1772 | bio->bi_end_io = journal_write_endio; | |
1773 | bio->bi_private = ca; | |
1774 | closure_bio_submit(bio, cl); | |
1775 | } | |
1c6fdbd8 KO |
1776 | } |
1777 | ||
731bdd2e | 1778 | continue_at(cl, do_journal_write, c->io_complete_wq); |
280249b9 | 1779 | return; |
1c6fdbd8 | 1780 | no_io: |
731bdd2e | 1781 | continue_at(cl, journal_write_done, c->io_complete_wq); |
1c6fdbd8 KO |
1782 | return; |
1783 | err: | |
b74b147d | 1784 | bch2_fatal_error(c); |
731bdd2e | 1785 | continue_at(cl, journal_write_done, c->io_complete_wq); |
1c6fdbd8 | 1786 | } |