1 // SPDX-License-Identifier: GPL-2.0
2
3 #include "bcachefs.h"
4 #include "btree_update_interior.h"
5 #include "buckets.h"
6 #include "error.h"
7 #include "journal_io.h"
8 #include "replicas.h"
9 #include "sb-clean.h"
10 #include "super-io.h"
11
12 /*
13 * BCH_SB_FIELD_clean:
14 *
15 * Btree roots, and a few other things, are recovered from the journal after an
16 * unclean shutdown - but after a clean shutdown, to avoid having to read the
17 * journal, we can store them in the superblock.
18 *
19 * bch_sb_field_clean simply contains a list of journal entries, stored exactly
20 * as they would be in the journal:
21 */
22
bch2_sb_clean_validate_late(struct bch_fs * c,struct bch_sb_field_clean * clean,int write)23 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean,
24 int write)
25 {
26 struct jset_entry *entry;
27 int ret;
28
29 for (entry = clean->start;
30 entry < (struct jset_entry *) vstruct_end(&clean->field);
31 entry = vstruct_next(entry)) {
32 ret = bch2_journal_entry_validate(c, NULL, entry,
33 le16_to_cpu(c->disk_sb.sb->version),
34 BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
35 write);
36 if (ret)
37 return ret;
38 }
39
40 return 0;
41 }
42
btree_root_find(struct bch_fs * c,struct bch_sb_field_clean * clean,struct jset * j,enum btree_id id,unsigned * level)43 static struct bkey_i *btree_root_find(struct bch_fs *c,
44 struct bch_sb_field_clean *clean,
45 struct jset *j,
46 enum btree_id id, unsigned *level)
47 {
48 struct bkey_i *k;
49 struct jset_entry *entry, *start, *end;
50
51 if (clean) {
52 start = clean->start;
53 end = vstruct_end(&clean->field);
54 } else {
55 start = j->start;
56 end = vstruct_last(j);
57 }
58
59 for (entry = start; entry < end; entry = vstruct_next(entry))
60 if (entry->type == BCH_JSET_ENTRY_btree_root &&
61 entry->btree_id == id)
62 goto found;
63
64 return NULL;
65 found:
66 if (!entry->u64s)
67 return ERR_PTR(-EINVAL);
68
69 k = entry->start;
70 *level = entry->level;
71 return k;
72 }
73
bch2_verify_superblock_clean(struct bch_fs * c,struct bch_sb_field_clean ** cleanp,struct jset * j)74 int bch2_verify_superblock_clean(struct bch_fs *c,
75 struct bch_sb_field_clean **cleanp,
76 struct jset *j)
77 {
78 unsigned i;
79 struct bch_sb_field_clean *clean = *cleanp;
80 struct printbuf buf1 = PRINTBUF;
81 struct printbuf buf2 = PRINTBUF;
82 int ret = 0;
83
84 if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
85 sb_clean_journal_seq_mismatch,
86 "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
87 le64_to_cpu(clean->journal_seq),
88 le64_to_cpu(j->seq))) {
89 kfree(clean);
90 *cleanp = NULL;
91 return 0;
92 }
93
94 for (i = 0; i < BTREE_ID_NR; i++) {
95 struct bkey_i *k1, *k2;
96 unsigned l1 = 0, l2 = 0;
97
98 k1 = btree_root_find(c, clean, NULL, i, &l1);
99 k2 = btree_root_find(c, NULL, j, i, &l2);
100
101 if (!k1 && !k2)
102 continue;
103
104 printbuf_reset(&buf1);
105 printbuf_reset(&buf2);
106
107 if (k1)
108 bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
109 else
110 prt_printf(&buf1, "(none)");
111
112 if (k2)
113 bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
114 else
115 prt_printf(&buf2, "(none)");
116
117 mustfix_fsck_err_on(!k1 || !k2 ||
118 IS_ERR(k1) ||
119 IS_ERR(k2) ||
120 k1->k.u64s != k2->k.u64s ||
121 memcmp(k1, k2, bkey_bytes(&k1->k)) ||
122 l1 != l2, c,
123 sb_clean_btree_root_mismatch,
124 "superblock btree root %u doesn't match journal after clean shutdown\n"
125 "sb: l=%u %s\n"
126 "journal: l=%u %s\n", i,
127 l1, buf1.buf,
128 l2, buf2.buf);
129 }
130 fsck_err:
131 printbuf_exit(&buf2);
132 printbuf_exit(&buf1);
133 return ret;
134 }
135
bch2_read_superblock_clean(struct bch_fs * c)136 struct bch_sb_field_clean *bch2_read_superblock_clean(struct bch_fs *c)
137 {
138 struct bch_sb_field_clean *clean, *sb_clean;
139 int ret;
140
141 mutex_lock(&c->sb_lock);
142 sb_clean = bch2_sb_field_get(c->disk_sb.sb, clean);
143
144 if (fsck_err_on(!sb_clean, c,
145 sb_clean_missing,
146 "superblock marked clean but clean section not present")) {
147 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
148 c->sb.clean = false;
149 mutex_unlock(&c->sb_lock);
150 return NULL;
151 }
152
153 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
154 GFP_KERNEL);
155 if (!clean) {
156 mutex_unlock(&c->sb_lock);
157 return ERR_PTR(-BCH_ERR_ENOMEM_read_superblock_clean);
158 }
159
160 ret = bch2_sb_clean_validate_late(c, clean, READ);
161 if (ret) {
162 mutex_unlock(&c->sb_lock);
163 return ERR_PTR(ret);
164 }
165
166 mutex_unlock(&c->sb_lock);
167
168 return clean;
169 fsck_err:
170 mutex_unlock(&c->sb_lock);
171 return ERR_PTR(ret);
172 }
173
jset_entry_init(struct jset_entry ** end,size_t size)174 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
175 {
176 struct jset_entry *entry = *end;
177 unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
178
179 memset(entry, 0, u64s * sizeof(u64));
180 /*
181 * The u64s field counts from the start of data, ignoring the shared
182 * fields.
183 */
184 entry->u64s = cpu_to_le16(u64s - 1);
185
186 *end = vstruct_next(*end);
187 return entry;
188 }
189
bch2_journal_super_entries_add_common(struct bch_fs * c,struct jset_entry ** end,u64 journal_seq)190 void bch2_journal_super_entries_add_common(struct bch_fs *c,
191 struct jset_entry **end,
192 u64 journal_seq)
193 {
194 percpu_down_read(&c->mark_lock);
195
196 if (!journal_seq) {
197 for (unsigned i = 0; i < ARRAY_SIZE(c->usage); i++)
198 bch2_fs_usage_acc_to_base(c, i);
199 } else {
200 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
201 }
202
203 {
204 struct jset_entry_usage *u =
205 container_of(jset_entry_init(end, sizeof(*u)),
206 struct jset_entry_usage, entry);
207
208 u->entry.type = BCH_JSET_ENTRY_usage;
209 u->entry.btree_id = BCH_FS_USAGE_inodes;
210 u->v = cpu_to_le64(c->usage_base->b.nr_inodes);
211 }
212
213 {
214 struct jset_entry_usage *u =
215 container_of(jset_entry_init(end, sizeof(*u)),
216 struct jset_entry_usage, entry);
217
218 u->entry.type = BCH_JSET_ENTRY_usage;
219 u->entry.btree_id = BCH_FS_USAGE_key_version;
220 u->v = cpu_to_le64(atomic64_read(&c->key_version));
221 }
222
223 for (unsigned i = 0; i < BCH_REPLICAS_MAX; i++) {
224 struct jset_entry_usage *u =
225 container_of(jset_entry_init(end, sizeof(*u)),
226 struct jset_entry_usage, entry);
227
228 u->entry.type = BCH_JSET_ENTRY_usage;
229 u->entry.btree_id = BCH_FS_USAGE_reserved;
230 u->entry.level = i;
231 u->v = cpu_to_le64(c->usage_base->persistent_reserved[i]);
232 }
233
234 for (unsigned i = 0; i < c->replicas.nr; i++) {
235 struct bch_replicas_entry_v1 *e =
236 cpu_replicas_entry(&c->replicas, i);
237 struct jset_entry_data_usage *u =
238 container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
239 struct jset_entry_data_usage, entry);
240
241 u->entry.type = BCH_JSET_ENTRY_data_usage;
242 u->v = cpu_to_le64(c->usage_base->replicas[i]);
243 unsafe_memcpy(&u->r, e, replicas_entry_bytes(e),
244 "embedded variable length struct");
245 }
246
247 for_each_member_device(c, ca) {
248 unsigned b = sizeof(struct jset_entry_dev_usage) +
249 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
250 struct jset_entry_dev_usage *u =
251 container_of(jset_entry_init(end, b),
252 struct jset_entry_dev_usage, entry);
253
254 u->entry.type = BCH_JSET_ENTRY_dev_usage;
255 u->dev = cpu_to_le32(ca->dev_idx);
256
257 for (unsigned i = 0; i < BCH_DATA_NR; i++) {
258 u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
259 u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
260 u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
261 }
262 }
263
264 percpu_up_read(&c->mark_lock);
265
266 for (unsigned i = 0; i < 2; i++) {
267 struct jset_entry_clock *clock =
268 container_of(jset_entry_init(end, sizeof(*clock)),
269 struct jset_entry_clock, entry);
270
271 clock->entry.type = BCH_JSET_ENTRY_clock;
272 clock->rw = i;
273 clock->time = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
274 }
275 }
276
bch2_sb_clean_validate(struct bch_sb * sb,struct bch_sb_field * f,struct printbuf * err)277 static int bch2_sb_clean_validate(struct bch_sb *sb,
278 struct bch_sb_field *f,
279 struct printbuf *err)
280 {
281 struct bch_sb_field_clean *clean = field_to_type(f, clean);
282
283 if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
284 prt_printf(err, "wrong size (got %zu should be %zu)",
285 vstruct_bytes(&clean->field), sizeof(*clean));
286 return -BCH_ERR_invalid_sb_clean;
287 }
288
289 return 0;
290 }
291
bch2_sb_clean_to_text(struct printbuf * out,struct bch_sb * sb,struct bch_sb_field * f)292 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
293 struct bch_sb_field *f)
294 {
295 struct bch_sb_field_clean *clean = field_to_type(f, clean);
296 struct jset_entry *entry;
297
298 prt_printf(out, "flags: %x", le32_to_cpu(clean->flags));
299 prt_newline(out);
300 prt_printf(out, "journal_seq: %llu", le64_to_cpu(clean->journal_seq));
301 prt_newline(out);
302
303 for (entry = clean->start;
304 entry != vstruct_end(&clean->field);
305 entry = vstruct_next(entry)) {
306 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
307 !entry->u64s)
308 continue;
309
310 bch2_journal_entry_to_text(out, NULL, entry);
311 prt_newline(out);
312 }
313 }
314
315 const struct bch_sb_field_ops bch_sb_field_ops_clean = {
316 .validate = bch2_sb_clean_validate,
317 .to_text = bch2_sb_clean_to_text,
318 };
319
bch2_fs_mark_dirty(struct bch_fs * c)320 int bch2_fs_mark_dirty(struct bch_fs *c)
321 {
322 int ret;
323
324 /*
325 * Unconditionally write superblock, to verify it hasn't changed before
326 * we go rw:
327 */
328
329 mutex_lock(&c->sb_lock);
330 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
331 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
332
333 ret = bch2_write_super(c);
334 mutex_unlock(&c->sb_lock);
335
336 return ret;
337 }
338
bch2_fs_mark_clean(struct bch_fs * c)339 void bch2_fs_mark_clean(struct bch_fs *c)
340 {
341 struct bch_sb_field_clean *sb_clean;
342 struct jset_entry *entry;
343 unsigned u64s;
344 int ret;
345
346 mutex_lock(&c->sb_lock);
347 if (BCH_SB_CLEAN(c->disk_sb.sb))
348 goto out;
349
350 SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
351
352 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
353 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
354 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
355 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
356
357 u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
358
359 sb_clean = bch2_sb_field_resize(&c->disk_sb, clean, u64s);
360 if (!sb_clean) {
361 bch_err(c, "error resizing superblock while setting filesystem clean");
362 goto out;
363 }
364
365 sb_clean->flags = 0;
366 sb_clean->journal_seq = cpu_to_le64(atomic64_read(&c->journal.seq));
367
368 /* Trying to catch outstanding bug: */
369 BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
370
371 entry = sb_clean->start;
372 bch2_journal_super_entries_add_common(c, &entry, 0);
373 entry = bch2_btree_roots_to_journal_entries(c, entry, 0);
374 BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
375
376 memset(entry, 0,
377 vstruct_end(&sb_clean->field) - (void *) entry);
378
379 /*
380 * this should be in the write path, and we should be validating every
381 * superblock section:
382 */
383 ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
384 if (ret) {
385 bch_err(c, "error writing marking filesystem clean: validate error");
386 goto out;
387 }
388
389 bch2_write_super(c);
390 out:
391 mutex_unlock(&c->sb_lock);
392 }
393