1 // SPDX-License-Identifier: GPL-2.0
2
3 #include "bcachefs.h"
4 #include "checksum.h"
5 #include "disk_groups.h"
6 #include "ec.h"
7 #include "error.h"
8 #include "journal.h"
9 #include "journal_sb.h"
10 #include "journal_seq_blacklist.h"
11 #include "recovery_passes.h"
12 #include "replicas.h"
13 #include "quota.h"
14 #include "sb-clean.h"
15 #include "sb-counters.h"
16 #include "sb-downgrade.h"
17 #include "sb-errors.h"
18 #include "sb-members.h"
19 #include "super-io.h"
20 #include "super.h"
21 #include "trace.h"
22 #include "vstructs.h"
23
24 #include <linux/backing-dev.h>
25 #include <linux/sort.h>
26 #include <linux/string_choices.h>
27
28 static const struct blk_holder_ops bch2_sb_handle_bdev_ops = {
29 };
30
31 struct bch2_metadata_version {
32 u16 version;
33 const char *name;
34 };
35
36 static const struct bch2_metadata_version bch2_metadata_versions[] = {
37 #define x(n, v) { \
38 .version = v, \
39 .name = #n, \
40 },
41 BCH_METADATA_VERSIONS()
42 #undef x
43 };
44
bch2_version_to_text(struct printbuf * out,enum bcachefs_metadata_version v)45 void bch2_version_to_text(struct printbuf *out, enum bcachefs_metadata_version v)
46 {
47 const char *str = "(unknown version)";
48
49 for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
50 if (bch2_metadata_versions[i].version == v) {
51 str = bch2_metadata_versions[i].name;
52 break;
53 }
54
55 prt_printf(out, "%u.%u: %s", BCH_VERSION_MAJOR(v), BCH_VERSION_MINOR(v), str);
56 }
57
bch2_latest_compatible_version(enum bcachefs_metadata_version v)58 enum bcachefs_metadata_version bch2_latest_compatible_version(enum bcachefs_metadata_version v)
59 {
60 if (!BCH_VERSION_MAJOR(v))
61 return v;
62
63 for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
64 if (bch2_metadata_versions[i].version > v &&
65 BCH_VERSION_MAJOR(bch2_metadata_versions[i].version) ==
66 BCH_VERSION_MAJOR(v))
67 v = bch2_metadata_versions[i].version;
68
69 return v;
70 }
71
bch2_set_version_incompat(struct bch_fs * c,enum bcachefs_metadata_version version)72 bool bch2_set_version_incompat(struct bch_fs *c, enum bcachefs_metadata_version version)
73 {
74 bool ret = (c->sb.features & BIT_ULL(BCH_FEATURE_incompat_version_field)) &&
75 version <= c->sb.version_incompat_allowed;
76
77 if (ret) {
78 mutex_lock(&c->sb_lock);
79 SET_BCH_SB_VERSION_INCOMPAT(c->disk_sb.sb,
80 max(BCH_SB_VERSION_INCOMPAT(c->disk_sb.sb), version));
81 bch2_write_super(c);
82 mutex_unlock(&c->sb_lock);
83 }
84
85 return ret;
86 }
87
88 const char * const bch2_sb_fields[] = {
89 #define x(name, nr) #name,
90 BCH_SB_FIELDS()
91 #undef x
92 NULL
93 };
94
95 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
96 enum bch_validate_flags, struct printbuf *);
97
bch2_sb_field_get_id(struct bch_sb * sb,enum bch_sb_field_type type)98 struct bch_sb_field *bch2_sb_field_get_id(struct bch_sb *sb,
99 enum bch_sb_field_type type)
100 {
101 /* XXX: need locking around superblock to access optional fields */
102
103 vstruct_for_each(sb, f)
104 if (le32_to_cpu(f->type) == type)
105 return f;
106 return NULL;
107 }
108
__bch2_sb_field_resize(struct bch_sb_handle * sb,struct bch_sb_field * f,unsigned u64s)109 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
110 struct bch_sb_field *f,
111 unsigned u64s)
112 {
113 unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
114 unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
115
116 BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
117
118 if (!f && !u64s) {
119 /* nothing to do: */
120 } else if (!f) {
121 f = vstruct_last(sb->sb);
122 memset(f, 0, sizeof(u64) * u64s);
123 f->u64s = cpu_to_le32(u64s);
124 f->type = 0;
125 } else {
126 void *src, *dst;
127
128 src = vstruct_end(f);
129
130 if (u64s) {
131 f->u64s = cpu_to_le32(u64s);
132 dst = vstruct_end(f);
133 } else {
134 dst = f;
135 }
136
137 memmove(dst, src, vstruct_end(sb->sb) - src);
138
139 if (dst > src)
140 memset(src, 0, dst - src);
141 }
142
143 sb->sb->u64s = cpu_to_le32(sb_u64s);
144
145 return u64s ? f : NULL;
146 }
147
bch2_sb_field_delete(struct bch_sb_handle * sb,enum bch_sb_field_type type)148 void bch2_sb_field_delete(struct bch_sb_handle *sb,
149 enum bch_sb_field_type type)
150 {
151 struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
152
153 if (f)
154 __bch2_sb_field_resize(sb, f, 0);
155 }
156
157 /* Superblock realloc/free: */
158
bch2_free_super(struct bch_sb_handle * sb)159 void bch2_free_super(struct bch_sb_handle *sb)
160 {
161 kfree(sb->bio);
162 if (!IS_ERR_OR_NULL(sb->s_bdev_file))
163 bdev_fput(sb->s_bdev_file);
164 kfree(sb->holder);
165 kfree(sb->sb_name);
166
167 kfree(sb->sb);
168 memset(sb, 0, sizeof(*sb));
169 }
170
bch2_sb_realloc(struct bch_sb_handle * sb,unsigned u64s)171 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
172 {
173 size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
174 size_t new_buffer_size;
175 struct bch_sb *new_sb;
176 struct bio *bio;
177
178 if (sb->bdev)
179 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
180
181 new_buffer_size = roundup_pow_of_two(new_bytes);
182
183 if (sb->sb && sb->buffer_size >= new_buffer_size)
184 return 0;
185
186 if (sb->sb && sb->have_layout) {
187 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
188
189 if (new_bytes > max_bytes) {
190 struct printbuf buf = PRINTBUF;
191
192 prt_bdevname(&buf, sb->bdev);
193 prt_printf(&buf, ": superblock too big: want %zu but have %llu", new_bytes, max_bytes);
194 pr_err("%s", buf.buf);
195 printbuf_exit(&buf);
196 return -BCH_ERR_ENOSPC_sb;
197 }
198 }
199
200 if (sb->buffer_size >= new_buffer_size && sb->sb)
201 return 0;
202
203 if (dynamic_fault("bcachefs:add:super_realloc"))
204 return -BCH_ERR_ENOMEM_sb_realloc_injected;
205
206 new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
207 if (!new_sb)
208 return -BCH_ERR_ENOMEM_sb_buf_realloc;
209
210 sb->sb = new_sb;
211
212 if (sb->have_bio) {
213 unsigned nr_bvecs = buf_pages(sb->sb, new_buffer_size);
214
215 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
216 if (!bio)
217 return -BCH_ERR_ENOMEM_sb_bio_realloc;
218
219 bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
220
221 kfree(sb->bio);
222 sb->bio = bio;
223 }
224
225 sb->buffer_size = new_buffer_size;
226
227 return 0;
228 }
229
bch2_sb_field_resize_id(struct bch_sb_handle * sb,enum bch_sb_field_type type,unsigned u64s)230 struct bch_sb_field *bch2_sb_field_resize_id(struct bch_sb_handle *sb,
231 enum bch_sb_field_type type,
232 unsigned u64s)
233 {
234 struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
235 ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
236 ssize_t d = -old_u64s + u64s;
237
238 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
239 return NULL;
240
241 if (sb->fs_sb) {
242 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
243
244 lockdep_assert_held(&c->sb_lock);
245
246 /* XXX: we're not checking that offline device have enough space */
247
248 for_each_online_member(c, ca) {
249 struct bch_sb_handle *dev_sb = &ca->disk_sb;
250
251 if (bch2_sb_realloc(dev_sb, le32_to_cpu(dev_sb->sb->u64s) + d)) {
252 percpu_ref_put(&ca->io_ref);
253 return NULL;
254 }
255 }
256 }
257
258 f = bch2_sb_field_get_id(sb->sb, type);
259 f = __bch2_sb_field_resize(sb, f, u64s);
260 if (f)
261 f->type = cpu_to_le32(type);
262 return f;
263 }
264
bch2_sb_field_get_minsize_id(struct bch_sb_handle * sb,enum bch_sb_field_type type,unsigned u64s)265 struct bch_sb_field *bch2_sb_field_get_minsize_id(struct bch_sb_handle *sb,
266 enum bch_sb_field_type type,
267 unsigned u64s)
268 {
269 struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
270
271 if (!f || le32_to_cpu(f->u64s) < u64s)
272 f = bch2_sb_field_resize_id(sb, type, u64s);
273 return f;
274 }
275
276 /* Superblock validate: */
277
validate_sb_layout(struct bch_sb_layout * layout,struct printbuf * out)278 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
279 {
280 u64 offset, prev_offset, max_sectors;
281 unsigned i;
282
283 BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
284
285 if (!uuid_equal(&layout->magic, &BCACHE_MAGIC) &&
286 !uuid_equal(&layout->magic, &BCHFS_MAGIC)) {
287 prt_printf(out, "Not a bcachefs superblock layout");
288 return -BCH_ERR_invalid_sb_layout;
289 }
290
291 if (layout->layout_type != 0) {
292 prt_printf(out, "Invalid superblock layout type %u",
293 layout->layout_type);
294 return -BCH_ERR_invalid_sb_layout_type;
295 }
296
297 if (!layout->nr_superblocks) {
298 prt_printf(out, "Invalid superblock layout: no superblocks");
299 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
300 }
301
302 if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
303 prt_printf(out, "Invalid superblock layout: too many superblocks");
304 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
305 }
306
307 if (layout->sb_max_size_bits > BCH_SB_LAYOUT_SIZE_BITS_MAX) {
308 prt_printf(out, "Invalid superblock layout: max_size_bits too high");
309 return -BCH_ERR_invalid_sb_layout_sb_max_size_bits;
310 }
311
312 max_sectors = 1 << layout->sb_max_size_bits;
313
314 prev_offset = le64_to_cpu(layout->sb_offset[0]);
315
316 for (i = 1; i < layout->nr_superblocks; i++) {
317 offset = le64_to_cpu(layout->sb_offset[i]);
318
319 if (offset < prev_offset + max_sectors) {
320 prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
321 " (sb %u ends at %llu next starts at %llu",
322 i - 1, prev_offset + max_sectors, offset);
323 return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
324 }
325 prev_offset = offset;
326 }
327
328 return 0;
329 }
330
bch2_sb_compatible(struct bch_sb * sb,struct printbuf * out)331 static int bch2_sb_compatible(struct bch_sb *sb, struct printbuf *out)
332 {
333 u16 version = le16_to_cpu(sb->version);
334 u16 version_min = le16_to_cpu(sb->version_min);
335
336 if (!bch2_version_compatible(version)) {
337 prt_str(out, "Unsupported superblock version ");
338 bch2_version_to_text(out, version);
339 prt_str(out, " (min ");
340 bch2_version_to_text(out, bcachefs_metadata_version_min);
341 prt_str(out, ", max ");
342 bch2_version_to_text(out, bcachefs_metadata_version_current);
343 prt_str(out, ")");
344 return -BCH_ERR_invalid_sb_version;
345 }
346
347 if (!bch2_version_compatible(version_min)) {
348 prt_str(out, "Unsupported superblock version_min ");
349 bch2_version_to_text(out, version_min);
350 prt_str(out, " (min ");
351 bch2_version_to_text(out, bcachefs_metadata_version_min);
352 prt_str(out, ", max ");
353 bch2_version_to_text(out, bcachefs_metadata_version_current);
354 prt_str(out, ")");
355 return -BCH_ERR_invalid_sb_version;
356 }
357
358 if (version_min > version) {
359 prt_str(out, "Bad minimum version ");
360 bch2_version_to_text(out, version_min);
361 prt_str(out, ", greater than version field ");
362 bch2_version_to_text(out, version);
363 return -BCH_ERR_invalid_sb_version;
364 }
365
366 return 0;
367 }
368
bch2_sb_validate(struct bch_sb_handle * disk_sb,enum bch_validate_flags flags,struct printbuf * out)369 static int bch2_sb_validate(struct bch_sb_handle *disk_sb,
370 enum bch_validate_flags flags, struct printbuf *out)
371 {
372 struct bch_sb *sb = disk_sb->sb;
373 struct bch_sb_field_members_v1 *mi;
374 enum bch_opt_id opt_id;
375 u16 block_size;
376 int ret;
377
378 ret = bch2_sb_compatible(sb, out);
379 if (ret)
380 return ret;
381
382 if (sb->features[1] ||
383 (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
384 prt_printf(out, "Filesystem has incompatible features");
385 return -BCH_ERR_invalid_sb_features;
386 }
387
388 if (BCH_VERSION_MAJOR(le16_to_cpu(sb->version)) > BCH_VERSION_MAJOR(bcachefs_metadata_version_current) ||
389 BCH_SB_VERSION_INCOMPAT(sb) > bcachefs_metadata_version_current) {
390 prt_printf(out, "Filesystem has incompatible version");
391 return -BCH_ERR_invalid_sb_features;
392 }
393
394 block_size = le16_to_cpu(sb->block_size);
395
396 if (block_size > PAGE_SECTORS) {
397 prt_printf(out, "Block size too big (got %u, max %u)",
398 block_size, PAGE_SECTORS);
399 return -BCH_ERR_invalid_sb_block_size;
400 }
401
402 if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid))) {
403 prt_printf(out, "Bad user UUID (got zeroes)");
404 return -BCH_ERR_invalid_sb_uuid;
405 }
406
407 if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid))) {
408 prt_printf(out, "Bad internal UUID (got zeroes)");
409 return -BCH_ERR_invalid_sb_uuid;
410 }
411
412 if (!sb->nr_devices ||
413 sb->nr_devices > BCH_SB_MEMBERS_MAX) {
414 prt_printf(out, "Bad number of member devices %u (max %u)",
415 sb->nr_devices, BCH_SB_MEMBERS_MAX);
416 return -BCH_ERR_invalid_sb_too_many_members;
417 }
418
419 if (sb->dev_idx >= sb->nr_devices) {
420 prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
421 sb->dev_idx, sb->nr_devices);
422 return -BCH_ERR_invalid_sb_dev_idx;
423 }
424
425 if (!sb->time_precision ||
426 le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
427 prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
428 le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
429 return -BCH_ERR_invalid_sb_time_precision;
430 }
431
432 /* old versions didn't know to downgrade this field */
433 if (BCH_SB_VERSION_INCOMPAT_ALLOWED(sb) > le16_to_cpu(sb->version))
434 SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(sb, le16_to_cpu(sb->version));
435
436 if (BCH_SB_VERSION_INCOMPAT(sb) > BCH_SB_VERSION_INCOMPAT_ALLOWED(sb)) {
437 prt_printf(out, "Invalid version_incompat ");
438 bch2_version_to_text(out, BCH_SB_VERSION_INCOMPAT(sb));
439 prt_str(out, " > incompat_allowed ");
440 bch2_version_to_text(out, BCH_SB_VERSION_INCOMPAT_ALLOWED(sb));
441 if (flags & BCH_VALIDATE_write)
442 return -BCH_ERR_invalid_sb_version;
443 else
444 SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(sb, BCH_SB_VERSION_INCOMPAT(sb));
445 }
446
447 if (!flags) {
448 /*
449 * Been seeing a bug where these are getting inexplicably
450 * zeroed, so we're now validating them, but we have to be
451 * careful not to preven people's filesystems from mounting:
452 */
453 if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
454 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
455 if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
456 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
457
458 if (!BCH_SB_VERSION_UPGRADE_COMPLETE(sb))
459 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(sb, le16_to_cpu(sb->version));
460
461 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_disk_accounting_v2 &&
462 !BCH_SB_ALLOCATOR_STUCK_TIMEOUT(sb))
463 SET_BCH_SB_ALLOCATOR_STUCK_TIMEOUT(sb, 30);
464
465 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_disk_accounting_v2)
466 SET_BCH_SB_PROMOTE_WHOLE_EXTENTS(sb, true);
467 }
468
469 #ifdef __KERNEL__
470 if (!BCH_SB_SHARD_INUMS_NBITS(sb))
471 SET_BCH_SB_SHARD_INUMS_NBITS(sb, ilog2(roundup_pow_of_two(num_online_cpus())));
472 #endif
473
474 for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
475 const struct bch_option *opt = bch2_opt_table + opt_id;
476
477 if (opt->get_sb != BCH2_NO_SB_OPT) {
478 u64 v = bch2_opt_from_sb(sb, opt_id);
479
480 prt_printf(out, "Invalid option ");
481 ret = bch2_opt_validate(opt, v, out);
482 if (ret)
483 return ret;
484
485 printbuf_reset(out);
486 }
487 }
488
489 /* validate layout */
490 ret = validate_sb_layout(&sb->layout, out);
491 if (ret)
492 return ret;
493
494 vstruct_for_each(sb, f) {
495 if (!f->u64s) {
496 prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
497 le32_to_cpu(f->type));
498 return -BCH_ERR_invalid_sb_field_size;
499 }
500
501 if (vstruct_next(f) > vstruct_last(sb)) {
502 prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
503 le32_to_cpu(f->type));
504 return -BCH_ERR_invalid_sb_field_size;
505 }
506 }
507
508 /* members must be validated first: */
509 mi = bch2_sb_field_get(sb, members_v1);
510 if (!mi) {
511 prt_printf(out, "Invalid superblock: member info area missing");
512 return -BCH_ERR_invalid_sb_members_missing;
513 }
514
515 ret = bch2_sb_field_validate(sb, &mi->field, flags, out);
516 if (ret)
517 return ret;
518
519 vstruct_for_each(sb, f) {
520 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members_v1)
521 continue;
522
523 ret = bch2_sb_field_validate(sb, f, flags, out);
524 if (ret)
525 return ret;
526 }
527
528 if ((flags & BCH_VALIDATE_write) &&
529 bch2_sb_member_get(sb, sb->dev_idx).seq != sb->seq) {
530 prt_printf(out, "Invalid superblock: member seq %llu != sb seq %llu",
531 le64_to_cpu(bch2_sb_member_get(sb, sb->dev_idx).seq),
532 le64_to_cpu(sb->seq));
533 return -BCH_ERR_invalid_sb_members_missing;
534 }
535
536 return 0;
537 }
538
539 /* device open: */
540
le_ulong_to_cpu(unsigned long v)541 static unsigned long le_ulong_to_cpu(unsigned long v)
542 {
543 return sizeof(unsigned long) == 8
544 ? le64_to_cpu(v)
545 : le32_to_cpu(v);
546 }
547
le_bitvector_to_cpu(unsigned long * dst,unsigned long * src,unsigned nr)548 static void le_bitvector_to_cpu(unsigned long *dst, unsigned long *src, unsigned nr)
549 {
550 BUG_ON(nr & (BITS_PER_TYPE(long) - 1));
551
552 for (unsigned i = 0; i < BITS_TO_LONGS(nr); i++)
553 dst[i] = le_ulong_to_cpu(src[i]);
554 }
555
bch2_sb_update(struct bch_fs * c)556 static void bch2_sb_update(struct bch_fs *c)
557 {
558 struct bch_sb *src = c->disk_sb.sb;
559
560 lockdep_assert_held(&c->sb_lock);
561
562 c->sb.uuid = src->uuid;
563 c->sb.user_uuid = src->user_uuid;
564 c->sb.version = le16_to_cpu(src->version);
565 c->sb.version_incompat = BCH_SB_VERSION_INCOMPAT(src);
566 c->sb.version_incompat_allowed
567 = BCH_SB_VERSION_INCOMPAT_ALLOWED(src);
568 c->sb.version_min = le16_to_cpu(src->version_min);
569 c->sb.version_upgrade_complete = BCH_SB_VERSION_UPGRADE_COMPLETE(src);
570 c->sb.nr_devices = src->nr_devices;
571 c->sb.clean = BCH_SB_CLEAN(src);
572 c->sb.encryption_type = BCH_SB_ENCRYPTION_TYPE(src);
573
574 c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
575 c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
576
577 /* XXX this is wrong, we need a 96 or 128 bit integer type */
578 c->sb.time_base_lo = div_u64(le64_to_cpu(src->time_base_lo),
579 c->sb.nsec_per_time_unit);
580 c->sb.time_base_hi = le32_to_cpu(src->time_base_hi);
581
582 c->sb.features = le64_to_cpu(src->features[0]);
583 c->sb.compat = le64_to_cpu(src->compat[0]);
584
585 memset(c->sb.errors_silent, 0, sizeof(c->sb.errors_silent));
586
587 struct bch_sb_field_ext *ext = bch2_sb_field_get(src, ext);
588 if (ext) {
589 le_bitvector_to_cpu(c->sb.errors_silent, (void *) ext->errors_silent,
590 sizeof(c->sb.errors_silent) * 8);
591 c->sb.btrees_lost_data = le64_to_cpu(ext->btrees_lost_data);
592 }
593
594 for_each_member_device(c, ca) {
595 struct bch_member m = bch2_sb_member_get(src, ca->dev_idx);
596 ca->mi = bch2_mi_to_cpu(&m);
597 }
598 }
599
__copy_super(struct bch_sb_handle * dst_handle,struct bch_sb * src)600 static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
601 {
602 struct bch_sb_field *src_f, *dst_f;
603 struct bch_sb *dst = dst_handle->sb;
604 unsigned i;
605
606 dst->version = src->version;
607 dst->version_min = src->version_min;
608 dst->seq = src->seq;
609 dst->uuid = src->uuid;
610 dst->user_uuid = src->user_uuid;
611 memcpy(dst->label, src->label, sizeof(dst->label));
612
613 dst->block_size = src->block_size;
614 dst->nr_devices = src->nr_devices;
615
616 dst->time_base_lo = src->time_base_lo;
617 dst->time_base_hi = src->time_base_hi;
618 dst->time_precision = src->time_precision;
619 dst->write_time = src->write_time;
620
621 memcpy(dst->flags, src->flags, sizeof(dst->flags));
622 memcpy(dst->features, src->features, sizeof(dst->features));
623 memcpy(dst->compat, src->compat, sizeof(dst->compat));
624
625 for (i = 0; i < BCH_SB_FIELD_NR; i++) {
626 int d;
627
628 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
629 continue;
630
631 src_f = bch2_sb_field_get_id(src, i);
632 dst_f = bch2_sb_field_get_id(dst, i);
633
634 d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
635 (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
636 if (d > 0) {
637 int ret = bch2_sb_realloc(dst_handle,
638 le32_to_cpu(dst_handle->sb->u64s) + d);
639
640 if (ret)
641 return ret;
642
643 dst = dst_handle->sb;
644 dst_f = bch2_sb_field_get_id(dst, i);
645 }
646
647 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
648 src_f ? le32_to_cpu(src_f->u64s) : 0);
649
650 if (src_f)
651 memcpy(dst_f, src_f, vstruct_bytes(src_f));
652 }
653
654 return 0;
655 }
656
bch2_sb_to_fs(struct bch_fs * c,struct bch_sb * src)657 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
658 {
659 int ret;
660
661 lockdep_assert_held(&c->sb_lock);
662
663 ret = bch2_sb_realloc(&c->disk_sb, 0) ?:
664 __copy_super(&c->disk_sb, src) ?:
665 bch2_sb_replicas_to_cpu_replicas(c) ?:
666 bch2_sb_disk_groups_to_cpu(c);
667 if (ret)
668 return ret;
669
670 bch2_sb_update(c);
671 return 0;
672 }
673
bch2_sb_from_fs(struct bch_fs * c,struct bch_dev * ca)674 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
675 {
676 return __copy_super(&ca->disk_sb, c->disk_sb.sb);
677 }
678
679 /* read superblock: */
680
read_one_super(struct bch_sb_handle * sb,u64 offset,struct printbuf * err)681 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
682 {
683 size_t bytes;
684 int ret;
685 reread:
686 bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
687 sb->bio->bi_iter.bi_sector = offset;
688 bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
689
690 ret = submit_bio_wait(sb->bio);
691 if (ret) {
692 prt_printf(err, "IO error: %i", ret);
693 return ret;
694 }
695
696 if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
697 !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC)) {
698 prt_str(err, "Not a bcachefs superblock (got magic ");
699 pr_uuid(err, sb->sb->magic.b);
700 prt_str(err, ")");
701 return -BCH_ERR_invalid_sb_magic;
702 }
703
704 ret = bch2_sb_compatible(sb->sb, err);
705 if (ret)
706 return ret;
707
708 bytes = vstruct_bytes(sb->sb);
709
710 u64 sb_size = 512ULL << min(BCH_SB_LAYOUT_SIZE_BITS_MAX, sb->sb->layout.sb_max_size_bits);
711 if (bytes > sb_size) {
712 prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %llu)",
713 bytes, sb_size);
714 return -BCH_ERR_invalid_sb_too_big;
715 }
716
717 if (bytes > sb->buffer_size) {
718 ret = bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s));
719 if (ret)
720 return ret;
721 goto reread;
722 }
723
724 enum bch_csum_type csum_type = BCH_SB_CSUM_TYPE(sb->sb);
725 if (csum_type >= BCH_CSUM_NR ||
726 bch2_csum_type_is_encryption(csum_type)) {
727 prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
728 return -BCH_ERR_invalid_sb_csum_type;
729 }
730
731 /* XXX: verify MACs */
732 struct bch_csum csum = csum_vstruct(NULL, csum_type, null_nonce(), sb->sb);
733 if (bch2_crc_cmp(csum, sb->sb->csum)) {
734 bch2_csum_err_msg(err, csum_type, sb->sb->csum, csum);
735 return -BCH_ERR_invalid_sb_csum;
736 }
737
738 sb->seq = le64_to_cpu(sb->sb->seq);
739
740 return 0;
741 }
742
__bch2_read_super(const char * path,struct bch_opts * opts,struct bch_sb_handle * sb,bool ignore_notbchfs_msg)743 static int __bch2_read_super(const char *path, struct bch_opts *opts,
744 struct bch_sb_handle *sb, bool ignore_notbchfs_msg)
745 {
746 u64 offset = opt_get(*opts, sb);
747 struct bch_sb_layout layout;
748 struct printbuf err = PRINTBUF;
749 struct printbuf err2 = PRINTBUF;
750 __le64 *i;
751 int ret;
752 #ifndef __KERNEL__
753 retry:
754 #endif
755 memset(sb, 0, sizeof(*sb));
756 sb->mode = BLK_OPEN_READ;
757 sb->have_bio = true;
758 sb->holder = kmalloc(1, GFP_KERNEL);
759 if (!sb->holder)
760 return -ENOMEM;
761
762 sb->sb_name = kstrdup(path, GFP_KERNEL);
763 if (!sb->sb_name) {
764 ret = -ENOMEM;
765 prt_printf(&err, "error allocating memory for sb_name");
766 goto err;
767 }
768
769 #ifndef __KERNEL__
770 if (opt_get(*opts, direct_io) == false)
771 sb->mode |= BLK_OPEN_BUFFERED;
772 #endif
773
774 if (!opt_get(*opts, noexcl))
775 sb->mode |= BLK_OPEN_EXCL;
776
777 if (!opt_get(*opts, nochanges))
778 sb->mode |= BLK_OPEN_WRITE;
779
780 sb->s_bdev_file = bdev_file_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
781 if (IS_ERR(sb->s_bdev_file) &&
782 PTR_ERR(sb->s_bdev_file) == -EACCES &&
783 opt_get(*opts, read_only)) {
784 sb->mode &= ~BLK_OPEN_WRITE;
785
786 sb->s_bdev_file = bdev_file_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
787 if (!IS_ERR(sb->s_bdev_file))
788 opt_set(*opts, nochanges, true);
789 }
790
791 if (IS_ERR(sb->s_bdev_file)) {
792 ret = PTR_ERR(sb->s_bdev_file);
793 prt_printf(&err, "error opening %s: %s", path, bch2_err_str(ret));
794 goto err;
795 }
796 sb->bdev = file_bdev(sb->s_bdev_file);
797
798 ret = bch2_sb_realloc(sb, 0);
799 if (ret) {
800 prt_printf(&err, "error allocating memory for superblock");
801 goto err;
802 }
803
804 if (bch2_fs_init_fault("read_super")) {
805 prt_printf(&err, "dynamic fault");
806 ret = -EFAULT;
807 goto err;
808 }
809
810 ret = read_one_super(sb, offset, &err);
811 if (!ret)
812 goto got_super;
813
814 if (opt_defined(*opts, sb))
815 goto err;
816
817 prt_printf(&err2, "bcachefs (%s): error reading default superblock: %s\n",
818 path, err.buf);
819 if (ret == -BCH_ERR_invalid_sb_magic && ignore_notbchfs_msg)
820 bch2_print_opts(opts, KERN_INFO "%s", err2.buf);
821 else
822 bch2_print_opts(opts, KERN_ERR "%s", err2.buf);
823
824 printbuf_exit(&err2);
825 printbuf_reset(&err);
826
827 /*
828 * Error reading primary superblock - read location of backup
829 * superblocks:
830 */
831 bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
832 sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
833 /*
834 * use sb buffer to read layout, since sb buffer is page aligned but
835 * layout won't be:
836 */
837 bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
838
839 ret = submit_bio_wait(sb->bio);
840 if (ret) {
841 prt_printf(&err, "IO error: %i", ret);
842 goto err;
843 }
844
845 memcpy(&layout, sb->sb, sizeof(layout));
846 ret = validate_sb_layout(&layout, &err);
847 if (ret)
848 goto err;
849
850 for (i = layout.sb_offset;
851 i < layout.sb_offset + layout.nr_superblocks; i++) {
852 offset = le64_to_cpu(*i);
853
854 if (offset == opt_get(*opts, sb)) {
855 ret = -BCH_ERR_invalid;
856 continue;
857 }
858
859 ret = read_one_super(sb, offset, &err);
860 if (!ret)
861 goto got_super;
862 }
863
864 goto err;
865
866 got_super:
867 if (le16_to_cpu(sb->sb->block_size) << 9 <
868 bdev_logical_block_size(sb->bdev) &&
869 opt_get(*opts, direct_io)) {
870 #ifndef __KERNEL__
871 opt_set(*opts, direct_io, false);
872 bch2_free_super(sb);
873 goto retry;
874 #endif
875 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
876 le16_to_cpu(sb->sb->block_size) << 9,
877 bdev_logical_block_size(sb->bdev));
878 ret = -BCH_ERR_block_size_too_small;
879 goto err;
880 }
881
882 sb->have_layout = true;
883
884 ret = bch2_sb_validate(sb, 0, &err);
885 if (ret) {
886 bch2_print_opts(opts, KERN_ERR "bcachefs (%s): error validating superblock: %s\n",
887 path, err.buf);
888 goto err_no_print;
889 }
890 out:
891 printbuf_exit(&err);
892 return ret;
893 err:
894 bch2_print_opts(opts, KERN_ERR "bcachefs (%s): error reading superblock: %s\n",
895 path, err.buf);
896 err_no_print:
897 bch2_free_super(sb);
898 goto out;
899 }
900
bch2_read_super(const char * path,struct bch_opts * opts,struct bch_sb_handle * sb)901 int bch2_read_super(const char *path, struct bch_opts *opts,
902 struct bch_sb_handle *sb)
903 {
904 return __bch2_read_super(path, opts, sb, false);
905 }
906
907 /* provide a silenced version for mount.bcachefs */
908
bch2_read_super_silent(const char * path,struct bch_opts * opts,struct bch_sb_handle * sb)909 int bch2_read_super_silent(const char *path, struct bch_opts *opts,
910 struct bch_sb_handle *sb)
911 {
912 return __bch2_read_super(path, opts, sb, true);
913 }
914
915 /* write superblock: */
916
write_super_endio(struct bio * bio)917 static void write_super_endio(struct bio *bio)
918 {
919 struct bch_dev *ca = bio->bi_private;
920
921 /* XXX: return errors directly */
922
923 if (bch2_dev_io_err_on(bio->bi_status, ca,
924 bio_data_dir(bio)
925 ? BCH_MEMBER_ERROR_write
926 : BCH_MEMBER_ERROR_read,
927 "superblock %s error: %s",
928 str_write_read(bio_data_dir(bio)),
929 bch2_blk_status_to_str(bio->bi_status)))
930 ca->sb_write_error = 1;
931
932 closure_put(&ca->fs->sb_write);
933 percpu_ref_put(&ca->io_ref);
934 }
935
read_back_super(struct bch_fs * c,struct bch_dev * ca)936 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
937 {
938 struct bch_sb *sb = ca->disk_sb.sb;
939 struct bio *bio = ca->disk_sb.bio;
940
941 memset(ca->sb_read_scratch, 0, BCH_SB_READ_SCRATCH_BUF_SIZE);
942
943 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
944 bio->bi_iter.bi_sector = le64_to_cpu(sb->layout.sb_offset[0]);
945 bio->bi_end_io = write_super_endio;
946 bio->bi_private = ca;
947 bch2_bio_map(bio, ca->sb_read_scratch, BCH_SB_READ_SCRATCH_BUF_SIZE);
948
949 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb], bio_sectors(bio));
950
951 percpu_ref_get(&ca->io_ref);
952 closure_bio_submit(bio, &c->sb_write);
953 }
954
write_one_super(struct bch_fs * c,struct bch_dev * ca,unsigned idx)955 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
956 {
957 struct bch_sb *sb = ca->disk_sb.sb;
958 struct bio *bio = ca->disk_sb.bio;
959
960 sb->offset = sb->layout.sb_offset[idx];
961
962 SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
963 sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
964 null_nonce(), sb);
965
966 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
967 bio->bi_iter.bi_sector = le64_to_cpu(sb->offset);
968 bio->bi_end_io = write_super_endio;
969 bio->bi_private = ca;
970 bch2_bio_map(bio, sb,
971 roundup((size_t) vstruct_bytes(sb),
972 bdev_logical_block_size(ca->disk_sb.bdev)));
973
974 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
975 bio_sectors(bio));
976
977 percpu_ref_get(&ca->io_ref);
978 closure_bio_submit(bio, &c->sb_write);
979 }
980
bch2_write_super(struct bch_fs * c)981 int bch2_write_super(struct bch_fs *c)
982 {
983 struct closure *cl = &c->sb_write;
984 struct printbuf err = PRINTBUF;
985 unsigned sb = 0, nr_wrote;
986 struct bch_devs_mask sb_written;
987 bool wrote, can_mount_without_written, can_mount_with_written;
988 unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
989 DARRAY(struct bch_dev *) online_devices = {};
990 int ret = 0;
991
992 trace_and_count(c, write_super, c, _RET_IP_);
993
994 if (c->opts.very_degraded)
995 degraded_flags |= BCH_FORCE_IF_LOST;
996
997 lockdep_assert_held(&c->sb_lock);
998
999 closure_init_stack(cl);
1000 memset(&sb_written, 0, sizeof(sb_written));
1001
1002 for_each_online_member(c, ca) {
1003 ret = darray_push(&online_devices, ca);
1004 if (bch2_fs_fatal_err_on(ret, c, "%s: error allocating online devices", __func__)) {
1005 percpu_ref_put(&ca->io_ref);
1006 goto out;
1007 }
1008 percpu_ref_get(&ca->io_ref);
1009 }
1010
1011 /* Make sure we're using the new magic numbers: */
1012 c->disk_sb.sb->magic = BCHFS_MAGIC;
1013 c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
1014
1015 le64_add_cpu(&c->disk_sb.sb->seq, 1);
1016
1017 struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1018 darray_for_each(online_devices, ca)
1019 __bch2_members_v2_get_mut(mi, (*ca)->dev_idx)->seq = c->disk_sb.sb->seq;
1020 c->disk_sb.sb->write_time = cpu_to_le64(ktime_get_real_seconds());
1021
1022 if (test_bit(BCH_FS_error, &c->flags))
1023 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
1024 if (test_bit(BCH_FS_topology_error, &c->flags))
1025 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
1026
1027 SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
1028
1029 bch2_sb_counters_from_cpu(c);
1030 bch2_sb_members_from_cpu(c);
1031 bch2_sb_members_cpy_v2_v1(&c->disk_sb);
1032 bch2_sb_errors_from_cpu(c);
1033 bch2_sb_downgrade_update(c);
1034
1035 darray_for_each(online_devices, ca)
1036 bch2_sb_from_fs(c, (*ca));
1037
1038 darray_for_each(online_devices, ca) {
1039 printbuf_reset(&err);
1040
1041 ret = bch2_sb_validate(&(*ca)->disk_sb, BCH_VALIDATE_write, &err);
1042 if (ret) {
1043 bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
1044 goto out;
1045 }
1046 }
1047
1048 if (c->opts.nochanges)
1049 goto out;
1050
1051 /*
1052 * Defer writing the superblock until filesystem initialization is
1053 * complete - don't write out a partly initialized superblock:
1054 */
1055 if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
1056 goto out;
1057
1058 if (le16_to_cpu(c->disk_sb.sb->version) > bcachefs_metadata_version_current) {
1059 struct printbuf buf = PRINTBUF;
1060 prt_printf(&buf, "attempting to write superblock that wasn't version downgraded (");
1061 bch2_version_to_text(&buf, le16_to_cpu(c->disk_sb.sb->version));
1062 prt_str(&buf, " > ");
1063 bch2_version_to_text(&buf, bcachefs_metadata_version_current);
1064 prt_str(&buf, ")");
1065 bch2_fs_fatal_error(c, ": %s", buf.buf);
1066 printbuf_exit(&buf);
1067 return -BCH_ERR_sb_not_downgraded;
1068 }
1069
1070 darray_for_each(online_devices, ca) {
1071 __set_bit((*ca)->dev_idx, sb_written.d);
1072 (*ca)->sb_write_error = 0;
1073 }
1074
1075 darray_for_each(online_devices, ca)
1076 read_back_super(c, *ca);
1077 closure_sync(cl);
1078
1079 darray_for_each(online_devices, cap) {
1080 struct bch_dev *ca = *cap;
1081
1082 if (ca->sb_write_error)
1083 continue;
1084
1085 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
1086 struct printbuf buf = PRINTBUF;
1087 prt_char(&buf, ' ');
1088 prt_bdevname(&buf, ca->disk_sb.bdev);
1089 prt_printf(&buf,
1090 ": Superblock write was silently dropped! (seq %llu expected %llu)",
1091 le64_to_cpu(ca->sb_read_scratch->seq),
1092 ca->disk_sb.seq);
1093
1094 if (c->opts.errors != BCH_ON_ERROR_continue &&
1095 c->opts.errors != BCH_ON_ERROR_fix_safe) {
1096 ret = -BCH_ERR_erofs_sb_err;
1097 bch2_fs_fatal_error(c, "%s", buf.buf);
1098 } else {
1099 bch_err(c, "%s", buf.buf);
1100 }
1101
1102 printbuf_exit(&buf);
1103 }
1104
1105 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
1106 struct printbuf buf = PRINTBUF;
1107 prt_char(&buf, ' ');
1108 prt_bdevname(&buf, ca->disk_sb.bdev);
1109 prt_printf(&buf,
1110 ": Superblock modified by another process (seq %llu expected %llu)",
1111 le64_to_cpu(ca->sb_read_scratch->seq),
1112 ca->disk_sb.seq);
1113 bch2_fs_fatal_error(c, "%s", buf.buf);
1114 printbuf_exit(&buf);
1115 ret = -BCH_ERR_erofs_sb_err;
1116 }
1117 }
1118
1119 if (ret)
1120 goto out;
1121
1122 do {
1123 wrote = false;
1124 darray_for_each(online_devices, cap) {
1125 struct bch_dev *ca = *cap;
1126 if (!ca->sb_write_error &&
1127 sb < ca->disk_sb.sb->layout.nr_superblocks) {
1128 write_one_super(c, ca, sb);
1129 wrote = true;
1130 }
1131 }
1132 closure_sync(cl);
1133 sb++;
1134 } while (wrote);
1135
1136 darray_for_each(online_devices, cap) {
1137 struct bch_dev *ca = *cap;
1138 if (ca->sb_write_error)
1139 __clear_bit(ca->dev_idx, sb_written.d);
1140 else
1141 ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
1142 }
1143
1144 nr_wrote = dev_mask_nr(&sb_written);
1145
1146 can_mount_with_written =
1147 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1148
1149 for (unsigned i = 0; i < ARRAY_SIZE(sb_written.d); i++)
1150 sb_written.d[i] = ~sb_written.d[i];
1151
1152 can_mount_without_written =
1153 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1154
1155 /*
1156 * If we would be able to mount _without_ the devices we successfully
1157 * wrote superblocks to, we weren't able to write to enough devices:
1158 *
1159 * Exception: if we can mount without the successes because we haven't
1160 * written anything (new filesystem), we continue if we'd be able to
1161 * mount with the devices we did successfully write to:
1162 */
1163 if (bch2_fs_fatal_err_on(!nr_wrote ||
1164 !can_mount_with_written ||
1165 (can_mount_without_written &&
1166 !can_mount_with_written), c,
1167 ": Unable to write superblock to sufficient devices (from %ps)",
1168 (void *) _RET_IP_))
1169 ret = -1;
1170 out:
1171 /* Make new options visible after they're persistent: */
1172 bch2_sb_update(c);
1173 darray_for_each(online_devices, ca)
1174 percpu_ref_put(&(*ca)->io_ref);
1175 darray_exit(&online_devices);
1176 printbuf_exit(&err);
1177 return ret;
1178 }
1179
__bch2_check_set_feature(struct bch_fs * c,unsigned feat)1180 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
1181 {
1182 mutex_lock(&c->sb_lock);
1183 if (!(c->sb.features & (1ULL << feat))) {
1184 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
1185
1186 bch2_write_super(c);
1187 }
1188 mutex_unlock(&c->sb_lock);
1189 }
1190
1191 /* Downgrade if superblock is at a higher version than currently supported: */
bch2_check_version_downgrade(struct bch_fs * c)1192 bool bch2_check_version_downgrade(struct bch_fs *c)
1193 {
1194 bool ret = bcachefs_metadata_version_current < c->sb.version;
1195
1196 lockdep_assert_held(&c->sb_lock);
1197
1198 /*
1199 * Downgrade, if superblock is at a higher version than currently
1200 * supported:
1201 *
1202 * c->sb will be checked before we write the superblock, so update it as
1203 * well:
1204 */
1205 if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current)
1206 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1207 if (BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb) > bcachefs_metadata_version_current)
1208 SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb, bcachefs_metadata_version_current);
1209 if (c->sb.version > bcachefs_metadata_version_current)
1210 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1211 if (c->sb.version_min > bcachefs_metadata_version_current)
1212 c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
1213 c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1214 return ret;
1215 }
1216
bch2_sb_upgrade(struct bch_fs * c,unsigned new_version,bool incompat)1217 void bch2_sb_upgrade(struct bch_fs *c, unsigned new_version, bool incompat)
1218 {
1219 lockdep_assert_held(&c->sb_lock);
1220
1221 if (BCH_VERSION_MAJOR(new_version) >
1222 BCH_VERSION_MAJOR(le16_to_cpu(c->disk_sb.sb->version)))
1223 bch2_sb_field_resize(&c->disk_sb, downgrade, 0);
1224
1225 c->disk_sb.sb->version = cpu_to_le16(new_version);
1226 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1227
1228 if (incompat) {
1229 SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb,
1230 max(BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb), new_version));
1231 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_FEATURE_incompat_version_field);
1232 }
1233 }
1234
bch2_sb_ext_validate(struct bch_sb * sb,struct bch_sb_field * f,enum bch_validate_flags flags,struct printbuf * err)1235 static int bch2_sb_ext_validate(struct bch_sb *sb, struct bch_sb_field *f,
1236 enum bch_validate_flags flags, struct printbuf *err)
1237 {
1238 if (vstruct_bytes(f) < 88) {
1239 prt_printf(err, "field too small (%zu < %u)", vstruct_bytes(f), 88);
1240 return -BCH_ERR_invalid_sb_ext;
1241 }
1242
1243 return 0;
1244 }
1245
bch2_sb_ext_to_text(struct printbuf * out,struct bch_sb * sb,struct bch_sb_field * f)1246 static void bch2_sb_ext_to_text(struct printbuf *out, struct bch_sb *sb,
1247 struct bch_sb_field *f)
1248 {
1249 struct bch_sb_field_ext *e = field_to_type(f, ext);
1250
1251 prt_printf(out, "Recovery passes required:\t");
1252 prt_bitflags(out, bch2_recovery_passes,
1253 bch2_recovery_passes_from_stable(le64_to_cpu(e->recovery_passes_required[0])));
1254 prt_newline(out);
1255
1256 unsigned long *errors_silent = kmalloc(sizeof(e->errors_silent), GFP_KERNEL);
1257 if (errors_silent) {
1258 le_bitvector_to_cpu(errors_silent, (void *) e->errors_silent, sizeof(e->errors_silent) * 8);
1259
1260 prt_printf(out, "Errors to silently fix:\t");
1261 prt_bitflags_vector(out, bch2_sb_error_strs, errors_silent,
1262 min(BCH_FSCK_ERR_MAX, sizeof(e->errors_silent) * 8));
1263 prt_newline(out);
1264
1265 kfree(errors_silent);
1266 }
1267
1268 prt_printf(out, "Btrees with missing data:\t");
1269 prt_bitflags(out, __bch2_btree_ids, le64_to_cpu(e->btrees_lost_data));
1270 prt_newline(out);
1271 }
1272
1273 static const struct bch_sb_field_ops bch_sb_field_ops_ext = {
1274 .validate = bch2_sb_ext_validate,
1275 .to_text = bch2_sb_ext_to_text,
1276 };
1277
1278 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1279 #define x(f, nr) \
1280 [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1281 BCH_SB_FIELDS()
1282 #undef x
1283 };
1284
1285 static const struct bch_sb_field_ops bch2_sb_field_null_ops;
1286
bch2_sb_field_type_ops(unsigned type)1287 static const struct bch_sb_field_ops *bch2_sb_field_type_ops(unsigned type)
1288 {
1289 return likely(type < ARRAY_SIZE(bch2_sb_field_ops))
1290 ? bch2_sb_field_ops[type]
1291 : &bch2_sb_field_null_ops;
1292 }
1293
bch2_sb_field_validate(struct bch_sb * sb,struct bch_sb_field * f,enum bch_validate_flags flags,struct printbuf * err)1294 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1295 enum bch_validate_flags flags, struct printbuf *err)
1296 {
1297 unsigned type = le32_to_cpu(f->type);
1298 struct printbuf field_err = PRINTBUF;
1299 const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1300 int ret;
1301
1302 ret = ops->validate ? ops->validate(sb, f, flags, &field_err) : 0;
1303 if (ret) {
1304 prt_printf(err, "Invalid superblock section %s: %s",
1305 bch2_sb_fields[type], field_err.buf);
1306 prt_newline(err);
1307 bch2_sb_field_to_text(err, sb, f);
1308 }
1309
1310 printbuf_exit(&field_err);
1311 return ret;
1312 }
1313
__bch2_sb_field_to_text(struct printbuf * out,struct bch_sb * sb,struct bch_sb_field * f)1314 void __bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1315 struct bch_sb_field *f)
1316 {
1317 unsigned type = le32_to_cpu(f->type);
1318 const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1319
1320 if (!out->nr_tabstops)
1321 printbuf_tabstop_push(out, 32);
1322
1323 if (ops->to_text)
1324 ops->to_text(out, sb, f);
1325 }
1326
bch2_sb_field_to_text(struct printbuf * out,struct bch_sb * sb,struct bch_sb_field * f)1327 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1328 struct bch_sb_field *f)
1329 {
1330 unsigned type = le32_to_cpu(f->type);
1331
1332 if (type < BCH_SB_FIELD_NR)
1333 prt_printf(out, "%s", bch2_sb_fields[type]);
1334 else
1335 prt_printf(out, "(unknown field %u)", type);
1336
1337 prt_printf(out, " (size %zu):", vstruct_bytes(f));
1338 prt_newline(out);
1339
1340 __bch2_sb_field_to_text(out, sb, f);
1341 }
1342
bch2_sb_layout_to_text(struct printbuf * out,struct bch_sb_layout * l)1343 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1344 {
1345 unsigned i;
1346
1347 prt_printf(out, "Type: %u", l->layout_type);
1348 prt_newline(out);
1349
1350 prt_str(out, "Superblock max size: ");
1351 prt_units_u64(out, 512 << l->sb_max_size_bits);
1352 prt_newline(out);
1353
1354 prt_printf(out, "Nr superblocks: %u", l->nr_superblocks);
1355 prt_newline(out);
1356
1357 prt_str(out, "Offsets: ");
1358 for (i = 0; i < l->nr_superblocks; i++) {
1359 if (i)
1360 prt_str(out, ", ");
1361 prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1362 }
1363 prt_newline(out);
1364 }
1365
bch2_sb_to_text(struct printbuf * out,struct bch_sb * sb,bool print_layout,unsigned fields)1366 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1367 bool print_layout, unsigned fields)
1368 {
1369 if (!out->nr_tabstops)
1370 printbuf_tabstop_push(out, 44);
1371
1372 prt_printf(out, "External UUID:\t");
1373 pr_uuid(out, sb->user_uuid.b);
1374 prt_newline(out);
1375
1376 prt_printf(out, "Internal UUID:\t");
1377 pr_uuid(out, sb->uuid.b);
1378 prt_newline(out);
1379
1380 prt_printf(out, "Magic number:\t");
1381 pr_uuid(out, sb->magic.b);
1382 prt_newline(out);
1383
1384 prt_printf(out, "Device index:\t%u\n", sb->dev_idx);
1385
1386 prt_printf(out, "Label:\t");
1387 if (!strlen(sb->label))
1388 prt_printf(out, "(none)");
1389 else
1390 prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1391 prt_newline(out);
1392
1393 prt_printf(out, "Version:\t");
1394 bch2_version_to_text(out, le16_to_cpu(sb->version));
1395 prt_newline(out);
1396
1397 prt_printf(out, "Incompatible features allowed:\t");
1398 bch2_version_to_text(out, BCH_SB_VERSION_INCOMPAT_ALLOWED(sb));
1399 prt_newline(out);
1400
1401 prt_printf(out, "Incompatible features in use:\t");
1402 bch2_version_to_text(out, BCH_SB_VERSION_INCOMPAT(sb));
1403 prt_newline(out);
1404
1405 prt_printf(out, "Version upgrade complete:\t");
1406 bch2_version_to_text(out, BCH_SB_VERSION_UPGRADE_COMPLETE(sb));
1407 prt_newline(out);
1408
1409 prt_printf(out, "Oldest version on disk:\t");
1410 bch2_version_to_text(out, le16_to_cpu(sb->version_min));
1411 prt_newline(out);
1412
1413 prt_printf(out, "Created:\t");
1414 if (sb->time_base_lo)
1415 bch2_prt_datetime(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1416 else
1417 prt_printf(out, "(not set)");
1418 prt_newline(out);
1419
1420 prt_printf(out, "Sequence number:\t");
1421 prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1422 prt_newline(out);
1423
1424 prt_printf(out, "Time of last write:\t");
1425 bch2_prt_datetime(out, le64_to_cpu(sb->write_time));
1426 prt_newline(out);
1427
1428 prt_printf(out, "Superblock size:\t");
1429 prt_units_u64(out, vstruct_bytes(sb));
1430 prt_str(out, "/");
1431 prt_units_u64(out, 512ULL << sb->layout.sb_max_size_bits);
1432 prt_newline(out);
1433
1434 prt_printf(out, "Clean:\t%llu\n", BCH_SB_CLEAN(sb));
1435 prt_printf(out, "Devices:\t%u\n", bch2_sb_nr_devices(sb));
1436
1437 prt_printf(out, "Sections:\t");
1438 u64 fields_have = 0;
1439 vstruct_for_each(sb, f)
1440 fields_have |= 1 << le32_to_cpu(f->type);
1441 prt_bitflags(out, bch2_sb_fields, fields_have);
1442 prt_newline(out);
1443
1444 prt_printf(out, "Features:\t");
1445 prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1446 prt_newline(out);
1447
1448 prt_printf(out, "Compat features:\t");
1449 prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1450 prt_newline(out);
1451
1452 prt_newline(out);
1453 prt_printf(out, "Options:");
1454 prt_newline(out);
1455 printbuf_indent_add(out, 2);
1456 {
1457 enum bch_opt_id id;
1458
1459 for (id = 0; id < bch2_opts_nr; id++) {
1460 const struct bch_option *opt = bch2_opt_table + id;
1461
1462 if (opt->get_sb != BCH2_NO_SB_OPT) {
1463 u64 v = bch2_opt_from_sb(sb, id);
1464
1465 prt_printf(out, "%s:\t", opt->attr.name);
1466 bch2_opt_to_text(out, NULL, sb, opt, v,
1467 OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1468 prt_newline(out);
1469 }
1470 }
1471 }
1472
1473 printbuf_indent_sub(out, 2);
1474
1475 if (print_layout) {
1476 prt_newline(out);
1477 prt_printf(out, "layout:");
1478 prt_newline(out);
1479 printbuf_indent_add(out, 2);
1480 bch2_sb_layout_to_text(out, &sb->layout);
1481 printbuf_indent_sub(out, 2);
1482 }
1483
1484 vstruct_for_each(sb, f)
1485 if (fields & (1 << le32_to_cpu(f->type))) {
1486 prt_newline(out);
1487 bch2_sb_field_to_text(out, sb, f);
1488 }
1489 }
1490