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 bkey_validate_context from = {
27 .flags = write,
28 .from = BKEY_VALIDATE_superblock,
29 };
30 struct jset_entry *entry;
31 int ret;
32
33 for (entry = clean->start;
34 entry < (struct jset_entry *) vstruct_end(&clean->field);
35 entry = vstruct_next(entry)) {
36 if (vstruct_end(entry) > vstruct_end(&clean->field)) {
37 bch_err(c, "journal entry (u64s %u) overran end of superblock clean section (u64s %u) by %zu",
38 le16_to_cpu(entry->u64s), le32_to_cpu(clean->field.u64s),
39 (u64 *) vstruct_end(entry) - (u64 *) vstruct_end(&clean->field));
40 bch2_sb_error_count(c, BCH_FSCK_ERR_sb_clean_entry_overrun);
41 return -BCH_ERR_fsck_repair_unimplemented;
42 }
43
44 ret = bch2_journal_entry_validate(c, NULL, entry,
45 le16_to_cpu(c->disk_sb.sb->version),
46 BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
47 from);
48 if (ret)
49 return ret;
50 }
51
52 return 0;
53 }
54
btree_root_find(struct bch_fs * c,struct bch_sb_field_clean * clean,struct jset * j,enum btree_id id,unsigned * level)55 static struct bkey_i *btree_root_find(struct bch_fs *c,
56 struct bch_sb_field_clean *clean,
57 struct jset *j,
58 enum btree_id id, unsigned *level)
59 {
60 struct bkey_i *k;
61 struct jset_entry *entry, *start, *end;
62
63 if (clean) {
64 start = clean->start;
65 end = vstruct_end(&clean->field);
66 } else {
67 start = j->start;
68 end = vstruct_last(j);
69 }
70
71 for (entry = start; entry < end; entry = vstruct_next(entry))
72 if (entry->type == BCH_JSET_ENTRY_btree_root &&
73 entry->btree_id == id)
74 goto found;
75
76 return NULL;
77 found:
78 if (!entry->u64s)
79 return ERR_PTR(-EINVAL);
80
81 k = entry->start;
82 *level = entry->level;
83 return k;
84 }
85
bch2_verify_superblock_clean(struct bch_fs * c,struct bch_sb_field_clean ** cleanp,struct jset * j)86 int bch2_verify_superblock_clean(struct bch_fs *c,
87 struct bch_sb_field_clean **cleanp,
88 struct jset *j)
89 {
90 unsigned i;
91 struct bch_sb_field_clean *clean = *cleanp;
92 struct printbuf buf1 = PRINTBUF;
93 struct printbuf buf2 = PRINTBUF;
94 int ret = 0;
95
96 if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
97 sb_clean_journal_seq_mismatch,
98 "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
99 le64_to_cpu(clean->journal_seq),
100 le64_to_cpu(j->seq))) {
101 kfree(clean);
102 *cleanp = NULL;
103 return 0;
104 }
105
106 for (i = 0; i < BTREE_ID_NR; i++) {
107 struct bkey_i *k1, *k2;
108 unsigned l1 = 0, l2 = 0;
109
110 k1 = btree_root_find(c, clean, NULL, i, &l1);
111 k2 = btree_root_find(c, NULL, j, i, &l2);
112
113 if (!k1 && !k2)
114 continue;
115
116 printbuf_reset(&buf1);
117 printbuf_reset(&buf2);
118
119 if (k1)
120 bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
121 else
122 prt_printf(&buf1, "(none)");
123
124 if (k2)
125 bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
126 else
127 prt_printf(&buf2, "(none)");
128
129 mustfix_fsck_err_on(!k1 || !k2 ||
130 IS_ERR(k1) ||
131 IS_ERR(k2) ||
132 k1->k.u64s != k2->k.u64s ||
133 memcmp(k1, k2, bkey_bytes(&k1->k)) ||
134 l1 != l2, c,
135 sb_clean_btree_root_mismatch,
136 "superblock btree root %u doesn't match journal after clean shutdown\n"
137 "sb: l=%u %s\n"
138 "journal: l=%u %s\n", i,
139 l1, buf1.buf,
140 l2, buf2.buf);
141 }
142 fsck_err:
143 printbuf_exit(&buf2);
144 printbuf_exit(&buf1);
145 return ret;
146 }
147
bch2_read_superblock_clean(struct bch_fs * c)148 struct bch_sb_field_clean *bch2_read_superblock_clean(struct bch_fs *c)
149 {
150 struct bch_sb_field_clean *clean, *sb_clean;
151 int ret;
152
153 mutex_lock(&c->sb_lock);
154 sb_clean = bch2_sb_field_get(c->disk_sb.sb, clean);
155
156 if (fsck_err_on(!sb_clean, c,
157 sb_clean_missing,
158 "superblock marked clean but clean section not present")) {
159 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
160 c->sb.clean = false;
161 mutex_unlock(&c->sb_lock);
162 return ERR_PTR(-BCH_ERR_invalid_sb_clean);
163 }
164
165 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
166 GFP_KERNEL);
167 if (!clean) {
168 mutex_unlock(&c->sb_lock);
169 return ERR_PTR(-BCH_ERR_ENOMEM_read_superblock_clean);
170 }
171
172 ret = bch2_sb_clean_validate_late(c, clean, READ);
173 if (ret) {
174 kfree(clean);
175 mutex_unlock(&c->sb_lock);
176 return ERR_PTR(ret);
177 }
178
179 mutex_unlock(&c->sb_lock);
180
181 return clean;
182 fsck_err:
183 mutex_unlock(&c->sb_lock);
184 return ERR_PTR(ret);
185 }
186
bch2_journal_super_entries_add_common(struct bch_fs * c,struct jset_entry ** end,u64 journal_seq)187 void bch2_journal_super_entries_add_common(struct bch_fs *c,
188 struct jset_entry **end,
189 u64 journal_seq)
190 {
191 {
192 struct jset_entry_usage *u =
193 container_of(jset_entry_init(end, sizeof(*u)),
194 struct jset_entry_usage, entry);
195
196 u->entry.type = BCH_JSET_ENTRY_usage;
197 u->entry.btree_id = BCH_FS_USAGE_key_version;
198 u->v = cpu_to_le64(atomic64_read(&c->key_version));
199 }
200
201 for (unsigned i = 0; i < 2; i++) {
202 struct jset_entry_clock *clock =
203 container_of(jset_entry_init(end, sizeof(*clock)),
204 struct jset_entry_clock, entry);
205
206 clock->entry.type = BCH_JSET_ENTRY_clock;
207 clock->rw = i;
208 clock->time = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
209 }
210 }
211
bch2_sb_clean_validate(struct bch_sb * sb,struct bch_sb_field * f,enum bch_validate_flags flags,struct printbuf * err)212 static int bch2_sb_clean_validate(struct bch_sb *sb, struct bch_sb_field *f,
213 enum bch_validate_flags flags, struct printbuf *err)
214 {
215 struct bch_sb_field_clean *clean = field_to_type(f, clean);
216
217 if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
218 prt_printf(err, "wrong size (got %zu should be %zu)",
219 vstruct_bytes(&clean->field), sizeof(*clean));
220 return -BCH_ERR_invalid_sb_clean;
221 }
222
223 for (struct jset_entry *entry = clean->start;
224 entry != vstruct_end(&clean->field);
225 entry = vstruct_next(entry)) {
226 if ((void *) vstruct_next(entry) > vstruct_end(&clean->field)) {
227 prt_str(err, "entry type ");
228 bch2_prt_jset_entry_type(err, entry->type);
229 prt_str(err, " overruns end of section");
230 return -BCH_ERR_invalid_sb_clean;
231 }
232 }
233
234 return 0;
235 }
236
bch2_sb_clean_to_text(struct printbuf * out,struct bch_sb * sb,struct bch_sb_field * f)237 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
238 struct bch_sb_field *f)
239 {
240 struct bch_sb_field_clean *clean = field_to_type(f, clean);
241 struct jset_entry *entry;
242
243 prt_printf(out, "flags: %x\n", le32_to_cpu(clean->flags));
244 prt_printf(out, "journal_seq: %llu\n", le64_to_cpu(clean->journal_seq));
245
246 for (entry = clean->start;
247 entry != vstruct_end(&clean->field);
248 entry = vstruct_next(entry)) {
249 if ((void *) vstruct_next(entry) > vstruct_end(&clean->field))
250 break;
251
252 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
253 !entry->u64s)
254 continue;
255
256 bch2_journal_entry_to_text(out, NULL, entry);
257 prt_newline(out);
258 }
259 }
260
261 const struct bch_sb_field_ops bch_sb_field_ops_clean = {
262 .validate = bch2_sb_clean_validate,
263 .to_text = bch2_sb_clean_to_text,
264 };
265
bch2_fs_mark_dirty(struct bch_fs * c)266 int bch2_fs_mark_dirty(struct bch_fs *c)
267 {
268 int ret;
269
270 /*
271 * Unconditionally write superblock, to verify it hasn't changed before
272 * we go rw:
273 */
274
275 mutex_lock(&c->sb_lock);
276 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
277 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
278
279 ret = bch2_write_super(c);
280 mutex_unlock(&c->sb_lock);
281
282 return ret;
283 }
284
bch2_fs_mark_clean(struct bch_fs * c)285 void bch2_fs_mark_clean(struct bch_fs *c)
286 {
287 struct bch_sb_field_clean *sb_clean;
288 struct jset_entry *entry;
289 unsigned u64s;
290 int ret;
291
292 mutex_lock(&c->sb_lock);
293 if (BCH_SB_CLEAN(c->disk_sb.sb))
294 goto out;
295
296 SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
297
298 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
299 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
300 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
301 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
302
303 u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
304
305 sb_clean = bch2_sb_field_resize(&c->disk_sb, clean, u64s);
306 if (!sb_clean) {
307 bch_err(c, "error resizing superblock while setting filesystem clean");
308 goto out;
309 }
310
311 sb_clean->flags = 0;
312 sb_clean->journal_seq = cpu_to_le64(atomic64_read(&c->journal.seq));
313
314 /* Trying to catch outstanding bug: */
315 BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
316
317 entry = sb_clean->start;
318 bch2_journal_super_entries_add_common(c, &entry, 0);
319 entry = bch2_btree_roots_to_journal_entries(c, entry, 0);
320 BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
321
322 memset(entry, 0,
323 vstruct_end(&sb_clean->field) - (void *) entry);
324
325 /*
326 * this should be in the write path, and we should be validating every
327 * superblock section:
328 */
329 ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
330 if (ret) {
331 bch_err(c, "error writing marking filesystem clean: validate error");
332 goto out;
333 }
334
335 bch2_journal_pos_from_member_info_set(c);
336
337 bch2_write_super(c);
338 out:
339 mutex_unlock(&c->sb_lock);
340 }
341