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 if (vstruct_end(entry) > vstruct_end(&clean->field)) {
33 bch_err(c, "journal entry (u64s %u) overran end of superblock clean section (u64s %u) by %zu",
34 le16_to_cpu(entry->u64s), le32_to_cpu(clean->field.u64s),
35 (u64 *) vstruct_end(entry) - (u64 *) vstruct_end(&clean->field));
36 bch2_sb_error_count(c, BCH_FSCK_ERR_sb_clean_entry_overrun);
37 return -BCH_ERR_fsck_repair_unimplemented;
38 }
39
40 ret = bch2_journal_entry_validate(c, NULL, entry,
41 le16_to_cpu(c->disk_sb.sb->version),
42 BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
43 write);
44 if (ret)
45 return ret;
46 }
47
48 return 0;
49 }
50
btree_root_find(struct bch_fs * c,struct bch_sb_field_clean * clean,struct jset * j,enum btree_id id,unsigned * level)51 static struct bkey_i *btree_root_find(struct bch_fs *c,
52 struct bch_sb_field_clean *clean,
53 struct jset *j,
54 enum btree_id id, unsigned *level)
55 {
56 struct bkey_i *k;
57 struct jset_entry *entry, *start, *end;
58
59 if (clean) {
60 start = clean->start;
61 end = vstruct_end(&clean->field);
62 } else {
63 start = j->start;
64 end = vstruct_last(j);
65 }
66
67 for (entry = start; entry < end; entry = vstruct_next(entry))
68 if (entry->type == BCH_JSET_ENTRY_btree_root &&
69 entry->btree_id == id)
70 goto found;
71
72 return NULL;
73 found:
74 if (!entry->u64s)
75 return ERR_PTR(-EINVAL);
76
77 k = entry->start;
78 *level = entry->level;
79 return k;
80 }
81
bch2_verify_superblock_clean(struct bch_fs * c,struct bch_sb_field_clean ** cleanp,struct jset * j)82 int bch2_verify_superblock_clean(struct bch_fs *c,
83 struct bch_sb_field_clean **cleanp,
84 struct jset *j)
85 {
86 unsigned i;
87 struct bch_sb_field_clean *clean = *cleanp;
88 struct printbuf buf1 = PRINTBUF;
89 struct printbuf buf2 = PRINTBUF;
90 int ret = 0;
91
92 if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
93 sb_clean_journal_seq_mismatch,
94 "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
95 le64_to_cpu(clean->journal_seq),
96 le64_to_cpu(j->seq))) {
97 kfree(clean);
98 *cleanp = NULL;
99 return 0;
100 }
101
102 for (i = 0; i < BTREE_ID_NR; i++) {
103 struct bkey_i *k1, *k2;
104 unsigned l1 = 0, l2 = 0;
105
106 k1 = btree_root_find(c, clean, NULL, i, &l1);
107 k2 = btree_root_find(c, NULL, j, i, &l2);
108
109 if (!k1 && !k2)
110 continue;
111
112 printbuf_reset(&buf1);
113 printbuf_reset(&buf2);
114
115 if (k1)
116 bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
117 else
118 prt_printf(&buf1, "(none)");
119
120 if (k2)
121 bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
122 else
123 prt_printf(&buf2, "(none)");
124
125 mustfix_fsck_err_on(!k1 || !k2 ||
126 IS_ERR(k1) ||
127 IS_ERR(k2) ||
128 k1->k.u64s != k2->k.u64s ||
129 memcmp(k1, k2, bkey_bytes(&k1->k)) ||
130 l1 != l2, c,
131 sb_clean_btree_root_mismatch,
132 "superblock btree root %u doesn't match journal after clean shutdown\n"
133 "sb: l=%u %s\n"
134 "journal: l=%u %s\n", i,
135 l1, buf1.buf,
136 l2, buf2.buf);
137 }
138 fsck_err:
139 printbuf_exit(&buf2);
140 printbuf_exit(&buf1);
141 return ret;
142 }
143
bch2_read_superblock_clean(struct bch_fs * c)144 struct bch_sb_field_clean *bch2_read_superblock_clean(struct bch_fs *c)
145 {
146 struct bch_sb_field_clean *clean, *sb_clean;
147 int ret;
148
149 mutex_lock(&c->sb_lock);
150 sb_clean = bch2_sb_field_get(c->disk_sb.sb, clean);
151
152 if (fsck_err_on(!sb_clean, c,
153 sb_clean_missing,
154 "superblock marked clean but clean section not present")) {
155 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
156 c->sb.clean = false;
157 mutex_unlock(&c->sb_lock);
158 return ERR_PTR(-BCH_ERR_invalid_sb_clean);
159 }
160
161 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
162 GFP_KERNEL);
163 if (!clean) {
164 mutex_unlock(&c->sb_lock);
165 return ERR_PTR(-BCH_ERR_ENOMEM_read_superblock_clean);
166 }
167
168 ret = bch2_sb_clean_validate_late(c, clean, READ);
169 if (ret) {
170 kfree(clean);
171 mutex_unlock(&c->sb_lock);
172 return ERR_PTR(ret);
173 }
174
175 mutex_unlock(&c->sb_lock);
176
177 return clean;
178 fsck_err:
179 mutex_unlock(&c->sb_lock);
180 return ERR_PTR(ret);
181 }
182
bch2_journal_super_entries_add_common(struct bch_fs * c,struct jset_entry ** end,u64 journal_seq)183 void bch2_journal_super_entries_add_common(struct bch_fs *c,
184 struct jset_entry **end,
185 u64 journal_seq)
186 {
187 {
188 struct jset_entry_usage *u =
189 container_of(jset_entry_init(end, sizeof(*u)),
190 struct jset_entry_usage, entry);
191
192 u->entry.type = BCH_JSET_ENTRY_usage;
193 u->entry.btree_id = BCH_FS_USAGE_key_version;
194 u->v = cpu_to_le64(atomic64_read(&c->key_version));
195 }
196
197 for (unsigned i = 0; i < 2; i++) {
198 struct jset_entry_clock *clock =
199 container_of(jset_entry_init(end, sizeof(*clock)),
200 struct jset_entry_clock, entry);
201
202 clock->entry.type = BCH_JSET_ENTRY_clock;
203 clock->rw = i;
204 clock->time = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
205 }
206 }
207
bch2_sb_clean_validate(struct bch_sb * sb,struct bch_sb_field * f,enum bch_validate_flags flags,struct printbuf * err)208 static int bch2_sb_clean_validate(struct bch_sb *sb, struct bch_sb_field *f,
209 enum bch_validate_flags flags, struct printbuf *err)
210 {
211 struct bch_sb_field_clean *clean = field_to_type(f, clean);
212
213 if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
214 prt_printf(err, "wrong size (got %zu should be %zu)",
215 vstruct_bytes(&clean->field), sizeof(*clean));
216 return -BCH_ERR_invalid_sb_clean;
217 }
218
219 for (struct jset_entry *entry = clean->start;
220 entry != vstruct_end(&clean->field);
221 entry = vstruct_next(entry)) {
222 if ((void *) vstruct_next(entry) > vstruct_end(&clean->field)) {
223 prt_str(err, "entry type ");
224 bch2_prt_jset_entry_type(err, entry->type);
225 prt_str(err, " overruns end of section");
226 return -BCH_ERR_invalid_sb_clean;
227 }
228 }
229
230 return 0;
231 }
232
bch2_sb_clean_to_text(struct printbuf * out,struct bch_sb * sb,struct bch_sb_field * f)233 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
234 struct bch_sb_field *f)
235 {
236 struct bch_sb_field_clean *clean = field_to_type(f, clean);
237 struct jset_entry *entry;
238
239 prt_printf(out, "flags: %x\n", le32_to_cpu(clean->flags));
240 prt_printf(out, "journal_seq: %llu\n", le64_to_cpu(clean->journal_seq));
241
242 for (entry = clean->start;
243 entry != vstruct_end(&clean->field);
244 entry = vstruct_next(entry)) {
245 if ((void *) vstruct_next(entry) > vstruct_end(&clean->field))
246 break;
247
248 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
249 !entry->u64s)
250 continue;
251
252 bch2_journal_entry_to_text(out, NULL, entry);
253 prt_newline(out);
254 }
255 }
256
257 const struct bch_sb_field_ops bch_sb_field_ops_clean = {
258 .validate = bch2_sb_clean_validate,
259 .to_text = bch2_sb_clean_to_text,
260 };
261
bch2_fs_mark_dirty(struct bch_fs * c)262 int bch2_fs_mark_dirty(struct bch_fs *c)
263 {
264 int ret;
265
266 /*
267 * Unconditionally write superblock, to verify it hasn't changed before
268 * we go rw:
269 */
270
271 mutex_lock(&c->sb_lock);
272 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
273 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
274
275 ret = bch2_write_super(c);
276 mutex_unlock(&c->sb_lock);
277
278 return ret;
279 }
280
bch2_fs_mark_clean(struct bch_fs * c)281 void bch2_fs_mark_clean(struct bch_fs *c)
282 {
283 struct bch_sb_field_clean *sb_clean;
284 struct jset_entry *entry;
285 unsigned u64s;
286 int ret;
287
288 mutex_lock(&c->sb_lock);
289 if (BCH_SB_CLEAN(c->disk_sb.sb))
290 goto out;
291
292 SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
293
294 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
295 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
296 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
297 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
298
299 u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
300
301 sb_clean = bch2_sb_field_resize(&c->disk_sb, clean, u64s);
302 if (!sb_clean) {
303 bch_err(c, "error resizing superblock while setting filesystem clean");
304 goto out;
305 }
306
307 sb_clean->flags = 0;
308 sb_clean->journal_seq = cpu_to_le64(atomic64_read(&c->journal.seq));
309
310 /* Trying to catch outstanding bug: */
311 BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
312
313 entry = sb_clean->start;
314 bch2_journal_super_entries_add_common(c, &entry, 0);
315 entry = bch2_btree_roots_to_journal_entries(c, entry, 0);
316 BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
317
318 memset(entry, 0,
319 vstruct_end(&sb_clean->field) - (void *) entry);
320
321 /*
322 * this should be in the write path, and we should be validating every
323 * superblock section:
324 */
325 ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
326 if (ret) {
327 bch_err(c, "error writing marking filesystem clean: validate error");
328 goto out;
329 }
330
331 bch2_journal_pos_from_member_info_set(c);
332
333 bch2_write_super(c);
334 out:
335 mutex_unlock(&c->sb_lock);
336 }
337