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 struct bch2_metadata_version {
29 	u16		version;
30 	const char	*name;
31 };
32 
33 static const struct bch2_metadata_version bch2_metadata_versions[] = {
34 #define x(n, v) {		\
35 	.version = v,				\
36 	.name = #n,				\
37 },
38 	BCH_METADATA_VERSIONS()
39 #undef x
40 };
41 
bch2_version_to_text(struct printbuf * out,enum bcachefs_metadata_version v)42 void bch2_version_to_text(struct printbuf *out, enum bcachefs_metadata_version v)
43 {
44 	const char *str = "(unknown version)";
45 
46 	for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
47 		if (bch2_metadata_versions[i].version == v) {
48 			str = bch2_metadata_versions[i].name;
49 			break;
50 		}
51 
52 	prt_printf(out, "%u.%u: %s", BCH_VERSION_MAJOR(v), BCH_VERSION_MINOR(v), str);
53 }
54 
bch2_latest_compatible_version(enum bcachefs_metadata_version v)55 enum bcachefs_metadata_version bch2_latest_compatible_version(enum bcachefs_metadata_version v)
56 {
57 	if (!BCH_VERSION_MAJOR(v))
58 		return v;
59 
60 	for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
61 		if (bch2_metadata_versions[i].version > v &&
62 		    BCH_VERSION_MAJOR(bch2_metadata_versions[i].version) ==
63 		    BCH_VERSION_MAJOR(v))
64 			v = bch2_metadata_versions[i].version;
65 
66 	return v;
67 }
68 
bch2_set_version_incompat(struct bch_fs * c,enum bcachefs_metadata_version version)69 int bch2_set_version_incompat(struct bch_fs *c, enum bcachefs_metadata_version version)
70 {
71 	int ret = ((c->sb.features & BIT_ULL(BCH_FEATURE_incompat_version_field)) &&
72 		   version <= c->sb.version_incompat_allowed)
73 		? 0
74 		: -BCH_ERR_may_not_use_incompat_feature;
75 
76 	mutex_lock(&c->sb_lock);
77 	if (!ret) {
78 		SET_BCH_SB_VERSION_INCOMPAT(c->disk_sb.sb,
79 			max(BCH_SB_VERSION_INCOMPAT(c->disk_sb.sb), version));
80 		bch2_write_super(c);
81 	} else {
82 		darray_for_each(c->incompat_versions_requested, i)
83 			if (version == *i)
84 				goto out;
85 
86 		darray_push(&c->incompat_versions_requested, version);
87 		struct printbuf buf = PRINTBUF;
88 		prt_str(&buf, "requested incompat feature ");
89 		bch2_version_to_text(&buf, version);
90 		prt_str(&buf, " currently not enabled");
91 		prt_printf(&buf, "\n  set version_upgrade=incompat to enable");
92 
93 		bch_notice(c, "%s", buf.buf);
94 		printbuf_exit(&buf);
95 	}
96 
97 out:
98 	mutex_unlock(&c->sb_lock);
99 
100 	return ret;
101 }
102 
103 const char * const bch2_sb_fields[] = {
104 #define x(name, nr)	#name,
105 	BCH_SB_FIELDS()
106 #undef x
107 	NULL
108 };
109 
110 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
111 				  enum bch_validate_flags, struct printbuf *);
112 
bch2_sb_field_get_id(struct bch_sb * sb,enum bch_sb_field_type type)113 struct bch_sb_field *bch2_sb_field_get_id(struct bch_sb *sb,
114 				      enum bch_sb_field_type type)
115 {
116 	/* XXX: need locking around superblock to access optional fields */
117 
118 	vstruct_for_each(sb, f)
119 		if (le32_to_cpu(f->type) == type)
120 			return f;
121 	return NULL;
122 }
123 
__bch2_sb_field_resize(struct bch_sb_handle * sb,struct bch_sb_field * f,unsigned u64s)124 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
125 						   struct bch_sb_field *f,
126 						   unsigned u64s)
127 {
128 	unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
129 	unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
130 
131 	BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
132 
133 	if (!f && !u64s) {
134 		/* nothing to do: */
135 	} else if (!f) {
136 		f = vstruct_last(sb->sb);
137 		memset(f, 0, sizeof(u64) * u64s);
138 		f->u64s = cpu_to_le32(u64s);
139 		f->type = 0;
140 	} else {
141 		void *src, *dst;
142 
143 		src = vstruct_end(f);
144 
145 		if (u64s) {
146 			f->u64s = cpu_to_le32(u64s);
147 			dst = vstruct_end(f);
148 		} else {
149 			dst = f;
150 		}
151 
152 		memmove(dst, src, vstruct_end(sb->sb) - src);
153 
154 		if (dst > src)
155 			memset(src, 0, dst - src);
156 	}
157 
158 	sb->sb->u64s = cpu_to_le32(sb_u64s);
159 
160 	return u64s ? f : NULL;
161 }
162 
bch2_sb_field_delete(struct bch_sb_handle * sb,enum bch_sb_field_type type)163 void bch2_sb_field_delete(struct bch_sb_handle *sb,
164 			  enum bch_sb_field_type type)
165 {
166 	struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
167 
168 	if (f)
169 		__bch2_sb_field_resize(sb, f, 0);
170 }
171 
172 /* Superblock realloc/free: */
173 
bch2_free_super(struct bch_sb_handle * sb)174 void bch2_free_super(struct bch_sb_handle *sb)
175 {
176 	kfree(sb->bio);
177 	if (!IS_ERR_OR_NULL(sb->s_bdev_file))
178 		bdev_fput(sb->s_bdev_file);
179 	kfree(sb->holder);
180 	kfree(sb->sb_name);
181 
182 	kfree(sb->sb);
183 	memset(sb, 0, sizeof(*sb));
184 }
185 
bch2_sb_realloc(struct bch_sb_handle * sb,unsigned u64s)186 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
187 {
188 	size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
189 	size_t new_buffer_size;
190 	struct bch_sb *new_sb;
191 	struct bio *bio;
192 
193 	if (sb->bdev)
194 		new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
195 
196 	new_buffer_size = roundup_pow_of_two(new_bytes);
197 
198 	if (sb->sb && sb->buffer_size >= new_buffer_size)
199 		return 0;
200 
201 	if (sb->sb && sb->have_layout) {
202 		u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
203 
204 		if (new_bytes > max_bytes) {
205 			struct printbuf buf = PRINTBUF;
206 
207 			prt_bdevname(&buf, sb->bdev);
208 			prt_printf(&buf, ": superblock too big: want %zu but have %llu", new_bytes, max_bytes);
209 			pr_err("%s", buf.buf);
210 			printbuf_exit(&buf);
211 			return -BCH_ERR_ENOSPC_sb;
212 		}
213 	}
214 
215 	if (sb->buffer_size >= new_buffer_size && sb->sb)
216 		return 0;
217 
218 	if (dynamic_fault("bcachefs:add:super_realloc"))
219 		return -BCH_ERR_ENOMEM_sb_realloc_injected;
220 
221 	new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
222 	if (!new_sb)
223 		return -BCH_ERR_ENOMEM_sb_buf_realloc;
224 
225 	sb->sb = new_sb;
226 
227 	if (sb->have_bio) {
228 		unsigned nr_bvecs = buf_pages(sb->sb, new_buffer_size);
229 
230 		bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
231 		if (!bio)
232 			return -BCH_ERR_ENOMEM_sb_bio_realloc;
233 
234 		bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
235 
236 		kfree(sb->bio);
237 		sb->bio = bio;
238 	}
239 
240 	sb->buffer_size = new_buffer_size;
241 
242 	return 0;
243 }
244 
bch2_sb_field_resize_id(struct bch_sb_handle * sb,enum bch_sb_field_type type,unsigned u64s)245 struct bch_sb_field *bch2_sb_field_resize_id(struct bch_sb_handle *sb,
246 					  enum bch_sb_field_type type,
247 					  unsigned u64s)
248 {
249 	struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
250 	ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
251 	ssize_t d = -old_u64s + u64s;
252 
253 	if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
254 		return NULL;
255 
256 	if (sb->fs_sb) {
257 		struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
258 
259 		lockdep_assert_held(&c->sb_lock);
260 
261 		/* XXX: we're not checking that offline device have enough space */
262 
263 		for_each_online_member(c, ca) {
264 			struct bch_sb_handle *dev_sb = &ca->disk_sb;
265 
266 			if (bch2_sb_realloc(dev_sb, le32_to_cpu(dev_sb->sb->u64s) + d)) {
267 				percpu_ref_put(&ca->io_ref[READ]);
268 				return NULL;
269 			}
270 		}
271 	}
272 
273 	f = bch2_sb_field_get_id(sb->sb, type);
274 	f = __bch2_sb_field_resize(sb, f, u64s);
275 	if (f)
276 		f->type = cpu_to_le32(type);
277 	return f;
278 }
279 
bch2_sb_field_get_minsize_id(struct bch_sb_handle * sb,enum bch_sb_field_type type,unsigned u64s)280 struct bch_sb_field *bch2_sb_field_get_minsize_id(struct bch_sb_handle *sb,
281 						  enum bch_sb_field_type type,
282 						  unsigned u64s)
283 {
284 	struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
285 
286 	if (!f || le32_to_cpu(f->u64s) < u64s)
287 		f = bch2_sb_field_resize_id(sb, type, u64s);
288 	return f;
289 }
290 
291 /* Superblock validate: */
292 
validate_sb_layout(struct bch_sb_layout * layout,struct printbuf * out)293 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
294 {
295 	u64 offset, prev_offset, max_sectors;
296 	unsigned i;
297 
298 	BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
299 
300 	if (!uuid_equal(&layout->magic, &BCACHE_MAGIC) &&
301 	    !uuid_equal(&layout->magic, &BCHFS_MAGIC)) {
302 		prt_printf(out, "Not a bcachefs superblock layout");
303 		return -BCH_ERR_invalid_sb_layout;
304 	}
305 
306 	if (layout->layout_type != 0) {
307 		prt_printf(out, "Invalid superblock layout type %u",
308 		       layout->layout_type);
309 		return -BCH_ERR_invalid_sb_layout_type;
310 	}
311 
312 	if (!layout->nr_superblocks) {
313 		prt_printf(out, "Invalid superblock layout: no superblocks");
314 		return -BCH_ERR_invalid_sb_layout_nr_superblocks;
315 	}
316 
317 	if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
318 		prt_printf(out, "Invalid superblock layout: too many superblocks");
319 		return -BCH_ERR_invalid_sb_layout_nr_superblocks;
320 	}
321 
322 	if (layout->sb_max_size_bits > BCH_SB_LAYOUT_SIZE_BITS_MAX) {
323 		prt_printf(out, "Invalid superblock layout: max_size_bits too high");
324 		return -BCH_ERR_invalid_sb_layout_sb_max_size_bits;
325 	}
326 
327 	max_sectors = 1 << layout->sb_max_size_bits;
328 
329 	prev_offset = le64_to_cpu(layout->sb_offset[0]);
330 
331 	for (i = 1; i < layout->nr_superblocks; i++) {
332 		offset = le64_to_cpu(layout->sb_offset[i]);
333 
334 		if (offset < prev_offset + max_sectors) {
335 			prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
336 			       "  (sb %u ends at %llu next starts at %llu",
337 			       i - 1, prev_offset + max_sectors, offset);
338 			return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
339 		}
340 		prev_offset = offset;
341 	}
342 
343 	return 0;
344 }
345 
bch2_sb_compatible(struct bch_sb * sb,struct printbuf * out)346 static int bch2_sb_compatible(struct bch_sb *sb, struct printbuf *out)
347 {
348 	u16 version		= le16_to_cpu(sb->version);
349 	u16 version_min		= le16_to_cpu(sb->version_min);
350 
351 	if (!bch2_version_compatible(version)) {
352 		prt_str(out, "Unsupported superblock version ");
353 		bch2_version_to_text(out, version);
354 		prt_str(out, " (min ");
355 		bch2_version_to_text(out, bcachefs_metadata_version_min);
356 		prt_str(out, ", max ");
357 		bch2_version_to_text(out, bcachefs_metadata_version_current);
358 		prt_str(out, ")");
359 		return -BCH_ERR_invalid_sb_version;
360 	}
361 
362 	if (!bch2_version_compatible(version_min)) {
363 		prt_str(out, "Unsupported superblock version_min ");
364 		bch2_version_to_text(out, version_min);
365 		prt_str(out, " (min ");
366 		bch2_version_to_text(out, bcachefs_metadata_version_min);
367 		prt_str(out, ", max ");
368 		bch2_version_to_text(out, bcachefs_metadata_version_current);
369 		prt_str(out, ")");
370 		return -BCH_ERR_invalid_sb_version;
371 	}
372 
373 	if (version_min > version) {
374 		prt_str(out, "Bad minimum version ");
375 		bch2_version_to_text(out, version_min);
376 		prt_str(out, ", greater than version field ");
377 		bch2_version_to_text(out, version);
378 		return -BCH_ERR_invalid_sb_version;
379 	}
380 
381 	return 0;
382 }
383 
bch2_sb_validate(struct bch_sb * sb,u64 read_offset,enum bch_validate_flags flags,struct printbuf * out)384 int bch2_sb_validate(struct bch_sb *sb, u64 read_offset,
385 		     enum bch_validate_flags flags, struct printbuf *out)
386 {
387 	struct bch_sb_field_members_v1 *mi;
388 	enum bch_opt_id opt_id;
389 	int ret;
390 
391 	ret = bch2_sb_compatible(sb, out);
392 	if (ret)
393 		return ret;
394 
395 	u64 incompat = le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR);
396 	unsigned incompat_bit = 0;
397 	if (incompat)
398 		incompat_bit = __ffs64(incompat);
399 	else if (sb->features[1])
400 		incompat_bit = 64 + __ffs64(le64_to_cpu(sb->features[1]));
401 
402 	if (incompat_bit) {
403 		prt_printf(out, "Filesystem has incompatible feature bit %u, highest supported %s (%u)",
404 			   incompat_bit,
405 			   bch2_sb_features[BCH_FEATURE_NR - 1],
406 			   BCH_FEATURE_NR - 1);
407 		return -BCH_ERR_invalid_sb_features;
408 	}
409 
410 	if (BCH_VERSION_MAJOR(le16_to_cpu(sb->version)) > BCH_VERSION_MAJOR(bcachefs_metadata_version_current) ||
411 	    BCH_SB_VERSION_INCOMPAT(sb) > bcachefs_metadata_version_current) {
412 		prt_str(out, "Filesystem has incompatible version ");
413 		bch2_version_to_text(out, le16_to_cpu(sb->version));
414 		prt_str(out, ", current version ");
415 		bch2_version_to_text(out, bcachefs_metadata_version_current);
416 		return -BCH_ERR_invalid_sb_features;
417 	}
418 
419 	if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid))) {
420 		prt_printf(out, "Bad user UUID (got zeroes)");
421 		return -BCH_ERR_invalid_sb_uuid;
422 	}
423 
424 	if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid))) {
425 		prt_printf(out, "Bad internal UUID (got zeroes)");
426 		return -BCH_ERR_invalid_sb_uuid;
427 	}
428 
429 	if (!(flags & BCH_VALIDATE_write) &&
430 	    le64_to_cpu(sb->offset) != read_offset) {
431 		prt_printf(out, "Bad sb offset (got %llu, read from %llu)",
432 			   le64_to_cpu(sb->offset), read_offset);
433 		return -BCH_ERR_invalid_sb_offset;
434 	}
435 
436 	if (!sb->nr_devices ||
437 	    sb->nr_devices > BCH_SB_MEMBERS_MAX) {
438 		prt_printf(out, "Bad number of member devices %u (max %u)",
439 		       sb->nr_devices, BCH_SB_MEMBERS_MAX);
440 		return -BCH_ERR_invalid_sb_too_many_members;
441 	}
442 
443 	if (sb->dev_idx >= sb->nr_devices) {
444 		prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
445 		       sb->dev_idx, sb->nr_devices);
446 		return -BCH_ERR_invalid_sb_dev_idx;
447 	}
448 
449 	if (!sb->time_precision ||
450 	    le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
451 		prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
452 		       le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
453 		return -BCH_ERR_invalid_sb_time_precision;
454 	}
455 
456 	/* old versions didn't know to downgrade this field */
457 	if (BCH_SB_VERSION_INCOMPAT_ALLOWED(sb) > le16_to_cpu(sb->version))
458 		SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(sb, le16_to_cpu(sb->version));
459 
460 	if (BCH_SB_VERSION_INCOMPAT(sb) > BCH_SB_VERSION_INCOMPAT_ALLOWED(sb)) {
461 		prt_printf(out, "Invalid version_incompat ");
462 		bch2_version_to_text(out, BCH_SB_VERSION_INCOMPAT(sb));
463 		prt_str(out, " > incompat_allowed ");
464 		bch2_version_to_text(out, BCH_SB_VERSION_INCOMPAT_ALLOWED(sb));
465 		if (flags & BCH_VALIDATE_write)
466 			return -BCH_ERR_invalid_sb_version;
467 		else
468 			SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(sb, BCH_SB_VERSION_INCOMPAT(sb));
469 	}
470 
471 	if (!flags) {
472 		/*
473 		 * Been seeing a bug where these are getting inexplicably
474 		 * zeroed, so we're now validating them, but we have to be
475 		 * careful not to preven people's filesystems from mounting:
476 		 */
477 		if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
478 			SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
479 		if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
480 			SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
481 
482 		if (!BCH_SB_VERSION_UPGRADE_COMPLETE(sb))
483 			SET_BCH_SB_VERSION_UPGRADE_COMPLETE(sb, le16_to_cpu(sb->version));
484 
485 		if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_disk_accounting_v2 &&
486 		    !BCH_SB_ALLOCATOR_STUCK_TIMEOUT(sb))
487 			SET_BCH_SB_ALLOCATOR_STUCK_TIMEOUT(sb, 30);
488 
489 		if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_disk_accounting_v2)
490 			SET_BCH_SB_PROMOTE_WHOLE_EXTENTS(sb, true);
491 
492 		if (!BCH_SB_WRITE_ERROR_TIMEOUT(sb))
493 			SET_BCH_SB_WRITE_ERROR_TIMEOUT(sb, 30);
494 
495 		if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_extent_flags &&
496 		    !BCH_SB_CSUM_ERR_RETRY_NR(sb))
497 			SET_BCH_SB_CSUM_ERR_RETRY_NR(sb, 3);
498 	}
499 
500 #ifdef __KERNEL__
501 	if (!BCH_SB_SHARD_INUMS_NBITS(sb))
502 		SET_BCH_SB_SHARD_INUMS_NBITS(sb, ilog2(roundup_pow_of_two(num_online_cpus())));
503 #endif
504 
505 	for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
506 		const struct bch_option *opt = bch2_opt_table + opt_id;
507 
508 		if (opt->get_sb) {
509 			u64 v = bch2_opt_from_sb(sb, opt_id, -1);
510 
511 			prt_printf(out, "Invalid option ");
512 			ret = bch2_opt_validate(opt, v, out);
513 			if (ret)
514 				return ret;
515 
516 			printbuf_reset(out);
517 		}
518 	}
519 
520 	/* validate layout */
521 	ret = validate_sb_layout(&sb->layout, out);
522 	if (ret)
523 		return ret;
524 
525 	vstruct_for_each(sb, f) {
526 		if (!f->u64s) {
527 			prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
528 			       le32_to_cpu(f->type));
529 			return -BCH_ERR_invalid_sb_field_size;
530 		}
531 
532 		if (vstruct_next(f) > vstruct_last(sb)) {
533 			prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
534 			       le32_to_cpu(f->type));
535 			return -BCH_ERR_invalid_sb_field_size;
536 		}
537 	}
538 
539 	/* members must be validated first: */
540 	mi = bch2_sb_field_get(sb, members_v1);
541 	if (!mi) {
542 		prt_printf(out, "Invalid superblock: member info area missing");
543 		return -BCH_ERR_invalid_sb_members_missing;
544 	}
545 
546 	ret = bch2_sb_field_validate(sb, &mi->field, flags, out);
547 	if (ret)
548 		return ret;
549 
550 	vstruct_for_each(sb, f) {
551 		if (le32_to_cpu(f->type) == BCH_SB_FIELD_members_v1)
552 			continue;
553 
554 		ret = bch2_sb_field_validate(sb, f, flags, out);
555 		if (ret)
556 			return ret;
557 	}
558 
559 	if ((flags & BCH_VALIDATE_write) &&
560 	    bch2_sb_member_get(sb, sb->dev_idx).seq != sb->seq) {
561 		prt_printf(out, "Invalid superblock: member seq %llu != sb seq %llu",
562 			   le64_to_cpu(bch2_sb_member_get(sb, sb->dev_idx).seq),
563 			   le64_to_cpu(sb->seq));
564 		return -BCH_ERR_invalid_sb_members_missing;
565 	}
566 
567 	return 0;
568 }
569 
570 /* device open: */
571 
le_ulong_to_cpu(unsigned long v)572 static unsigned long le_ulong_to_cpu(unsigned long v)
573 {
574 	return sizeof(unsigned long) == 8
575 		? le64_to_cpu(v)
576 		: le32_to_cpu(v);
577 }
578 
le_bitvector_to_cpu(unsigned long * dst,unsigned long * src,unsigned nr)579 static void le_bitvector_to_cpu(unsigned long *dst, unsigned long *src, unsigned nr)
580 {
581 	BUG_ON(nr & (BITS_PER_TYPE(long) - 1));
582 
583 	for (unsigned i = 0; i < BITS_TO_LONGS(nr); i++)
584 		dst[i] = le_ulong_to_cpu(src[i]);
585 }
586 
bch2_sb_update(struct bch_fs * c)587 static void bch2_sb_update(struct bch_fs *c)
588 {
589 	struct bch_sb *src = c->disk_sb.sb;
590 
591 	lockdep_assert_held(&c->sb_lock);
592 
593 	c->sb.uuid		= src->uuid;
594 	c->sb.user_uuid		= src->user_uuid;
595 	c->sb.version		= le16_to_cpu(src->version);
596 	c->sb.version_incompat	= BCH_SB_VERSION_INCOMPAT(src);
597 	c->sb.version_incompat_allowed
598 				= BCH_SB_VERSION_INCOMPAT_ALLOWED(src);
599 	c->sb.version_min	= le16_to_cpu(src->version_min);
600 	c->sb.version_upgrade_complete = BCH_SB_VERSION_UPGRADE_COMPLETE(src);
601 	c->sb.nr_devices	= src->nr_devices;
602 	c->sb.clean		= BCH_SB_CLEAN(src);
603 	c->sb.encryption_type	= BCH_SB_ENCRYPTION_TYPE(src);
604 
605 	c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
606 	c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
607 
608 	/* XXX this is wrong, we need a 96 or 128 bit integer type */
609 	c->sb.time_base_lo	= div_u64(le64_to_cpu(src->time_base_lo),
610 					  c->sb.nsec_per_time_unit);
611 	c->sb.time_base_hi	= le32_to_cpu(src->time_base_hi);
612 
613 	c->sb.features		= le64_to_cpu(src->features[0]);
614 	c->sb.compat		= le64_to_cpu(src->compat[0]);
615 
616 	memset(c->sb.errors_silent, 0, sizeof(c->sb.errors_silent));
617 
618 	struct bch_sb_field_ext *ext = bch2_sb_field_get(src, ext);
619 	if (ext) {
620 		le_bitvector_to_cpu(c->sb.errors_silent, (void *) ext->errors_silent,
621 				    sizeof(c->sb.errors_silent) * 8);
622 		c->sb.btrees_lost_data = le64_to_cpu(ext->btrees_lost_data);
623 	}
624 
625 	for_each_member_device(c, ca) {
626 		struct bch_member m = bch2_sb_member_get(src, ca->dev_idx);
627 		ca->mi = bch2_mi_to_cpu(&m);
628 	}
629 }
630 
__copy_super(struct bch_sb_handle * dst_handle,struct bch_sb * src)631 static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
632 {
633 	struct bch_sb_field *src_f, *dst_f;
634 	struct bch_sb *dst = dst_handle->sb;
635 	unsigned i;
636 
637 	dst->version		= src->version;
638 	dst->version_min	= src->version_min;
639 	dst->seq		= src->seq;
640 	dst->uuid		= src->uuid;
641 	dst->user_uuid		= src->user_uuid;
642 	memcpy(dst->label,	src->label, sizeof(dst->label));
643 
644 	dst->block_size		= src->block_size;
645 	dst->nr_devices		= src->nr_devices;
646 
647 	dst->time_base_lo	= src->time_base_lo;
648 	dst->time_base_hi	= src->time_base_hi;
649 	dst->time_precision	= src->time_precision;
650 	dst->write_time		= src->write_time;
651 
652 	memcpy(dst->flags,	src->flags,	sizeof(dst->flags));
653 	memcpy(dst->features,	src->features,	sizeof(dst->features));
654 	memcpy(dst->compat,	src->compat,	sizeof(dst->compat));
655 
656 	for (i = 0; i < BCH_SB_FIELD_NR; i++) {
657 		int d;
658 
659 		if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
660 			continue;
661 
662 		src_f = bch2_sb_field_get_id(src, i);
663 		dst_f = bch2_sb_field_get_id(dst, i);
664 
665 		d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
666 		    (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
667 		if (d > 0) {
668 			int ret = bch2_sb_realloc(dst_handle,
669 					le32_to_cpu(dst_handle->sb->u64s) + d);
670 
671 			if (ret)
672 				return ret;
673 
674 			dst = dst_handle->sb;
675 			dst_f = bch2_sb_field_get_id(dst, i);
676 		}
677 
678 		dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
679 				src_f ? le32_to_cpu(src_f->u64s) : 0);
680 
681 		if (src_f)
682 			memcpy(dst_f, src_f, vstruct_bytes(src_f));
683 	}
684 
685 	return 0;
686 }
687 
bch2_sb_to_fs(struct bch_fs * c,struct bch_sb * src)688 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
689 {
690 	int ret;
691 
692 	lockdep_assert_held(&c->sb_lock);
693 
694 	ret =   bch2_sb_realloc(&c->disk_sb, 0) ?:
695 		__copy_super(&c->disk_sb, src) ?:
696 		bch2_sb_replicas_to_cpu_replicas(c) ?:
697 		bch2_sb_disk_groups_to_cpu(c);
698 	if (ret)
699 		return ret;
700 
701 	bch2_sb_update(c);
702 	return 0;
703 }
704 
bch2_sb_from_fs(struct bch_fs * c,struct bch_dev * ca)705 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
706 {
707 	return __copy_super(&ca->disk_sb, c->disk_sb.sb);
708 }
709 
710 /* read superblock: */
711 
read_one_super(struct bch_sb_handle * sb,u64 offset,struct printbuf * err)712 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
713 {
714 	size_t bytes;
715 	int ret;
716 reread:
717 	bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
718 	sb->bio->bi_iter.bi_sector = offset;
719 	bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
720 
721 	ret = submit_bio_wait(sb->bio);
722 	if (ret) {
723 		prt_printf(err, "IO error: %i", ret);
724 		return ret;
725 	}
726 
727 	if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
728 	    !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC)) {
729 		prt_str(err, "Not a bcachefs superblock (got magic ");
730 		pr_uuid(err, sb->sb->magic.b);
731 		prt_str(err, ")");
732 		return -BCH_ERR_invalid_sb_magic;
733 	}
734 
735 	ret = bch2_sb_compatible(sb->sb, err);
736 	if (ret)
737 		return ret;
738 
739 	bytes = vstruct_bytes(sb->sb);
740 
741 	u64 sb_size = 512ULL << min(BCH_SB_LAYOUT_SIZE_BITS_MAX, sb->sb->layout.sb_max_size_bits);
742 	if (bytes > sb_size) {
743 		prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %llu)",
744 			   bytes, sb_size);
745 		return -BCH_ERR_invalid_sb_too_big;
746 	}
747 
748 	if (bytes > sb->buffer_size) {
749 		ret = bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s));
750 		if (ret)
751 			return ret;
752 		goto reread;
753 	}
754 
755 	enum bch_csum_type csum_type = BCH_SB_CSUM_TYPE(sb->sb);
756 	if (csum_type >= BCH_CSUM_NR ||
757 	    bch2_csum_type_is_encryption(csum_type)) {
758 		prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
759 		return -BCH_ERR_invalid_sb_csum_type;
760 	}
761 
762 	/* XXX: verify MACs */
763 	struct bch_csum csum = csum_vstruct(NULL, csum_type, null_nonce(), sb->sb);
764 	if (bch2_crc_cmp(csum, sb->sb->csum)) {
765 		bch2_csum_err_msg(err, csum_type, sb->sb->csum, csum);
766 		return -BCH_ERR_invalid_sb_csum;
767 	}
768 
769 	sb->seq = le64_to_cpu(sb->sb->seq);
770 
771 	return 0;
772 }
773 
__bch2_read_super(const char * path,struct bch_opts * opts,struct bch_sb_handle * sb,bool ignore_notbchfs_msg)774 static int __bch2_read_super(const char *path, struct bch_opts *opts,
775 		    struct bch_sb_handle *sb, bool ignore_notbchfs_msg)
776 {
777 	u64 offset = opt_get(*opts, sb);
778 	struct bch_sb_layout layout;
779 	struct printbuf err = PRINTBUF;
780 	struct printbuf err2 = PRINTBUF;
781 	__le64 *i;
782 	int ret;
783 #ifndef __KERNEL__
784 retry:
785 #endif
786 	memset(sb, 0, sizeof(*sb));
787 	sb->mode	= BLK_OPEN_READ;
788 	sb->have_bio	= true;
789 	sb->holder	= kzalloc(sizeof(*sb->holder), GFP_KERNEL);
790 	if (!sb->holder)
791 		return -ENOMEM;
792 
793 	sb->sb_name = kstrdup(path, GFP_KERNEL);
794 	if (!sb->sb_name) {
795 		ret = -ENOMEM;
796 		prt_printf(&err, "error allocating memory for sb_name");
797 		goto err;
798 	}
799 
800 #ifndef __KERNEL__
801 	if (opt_get(*opts, direct_io) == false)
802 		sb->mode |= BLK_OPEN_BUFFERED;
803 #endif
804 
805 	if (!opt_get(*opts, noexcl))
806 		sb->mode |= BLK_OPEN_EXCL;
807 
808 	if (!opt_get(*opts, nochanges))
809 		sb->mode |= BLK_OPEN_WRITE;
810 
811 	sb->s_bdev_file = bdev_file_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
812 	if (IS_ERR(sb->s_bdev_file) &&
813 	    PTR_ERR(sb->s_bdev_file) == -EACCES &&
814 	    opt_get(*opts, read_only)) {
815 		sb->mode &= ~BLK_OPEN_WRITE;
816 
817 		sb->s_bdev_file = bdev_file_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
818 		if (!IS_ERR(sb->s_bdev_file))
819 			opt_set(*opts, nochanges, true);
820 	}
821 
822 	if (IS_ERR(sb->s_bdev_file)) {
823 		ret = PTR_ERR(sb->s_bdev_file);
824 		prt_printf(&err, "error opening %s: %s", path, bch2_err_str(ret));
825 		goto err;
826 	}
827 	sb->bdev = file_bdev(sb->s_bdev_file);
828 
829 	ret = bch2_sb_realloc(sb, 0);
830 	if (ret) {
831 		prt_printf(&err, "error allocating memory for superblock");
832 		goto err;
833 	}
834 
835 	if (bch2_fs_init_fault("read_super")) {
836 		prt_printf(&err, "dynamic fault");
837 		ret = -EFAULT;
838 		goto err;
839 	}
840 
841 	ret = read_one_super(sb, offset, &err);
842 	if (!ret)
843 		goto got_super;
844 
845 	if (opt_defined(*opts, sb))
846 		goto err;
847 
848 	prt_printf(&err2, "bcachefs (%s): error reading default superblock: %s\n",
849 	       path, err.buf);
850 	if (ret == -BCH_ERR_invalid_sb_magic && ignore_notbchfs_msg)
851 		bch2_print_opts(opts, KERN_INFO "%s", err2.buf);
852 	else
853 		bch2_print_opts(opts, KERN_ERR "%s", err2.buf);
854 
855 	printbuf_exit(&err2);
856 	printbuf_reset(&err);
857 
858 	/*
859 	 * Error reading primary superblock - read location of backup
860 	 * superblocks:
861 	 */
862 	bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
863 	sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
864 	/*
865 	 * use sb buffer to read layout, since sb buffer is page aligned but
866 	 * layout won't be:
867 	 */
868 	bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
869 
870 	ret = submit_bio_wait(sb->bio);
871 	if (ret) {
872 		prt_printf(&err, "IO error: %i", ret);
873 		goto err;
874 	}
875 
876 	memcpy(&layout, sb->sb, sizeof(layout));
877 	ret = validate_sb_layout(&layout, &err);
878 	if (ret)
879 		goto err;
880 
881 	for (i = layout.sb_offset;
882 	     i < layout.sb_offset + layout.nr_superblocks; i++) {
883 		offset = le64_to_cpu(*i);
884 
885 		if (offset == opt_get(*opts, sb)) {
886 			ret = -BCH_ERR_invalid;
887 			continue;
888 		}
889 
890 		ret = read_one_super(sb, offset, &err);
891 		if (!ret)
892 			goto got_super;
893 	}
894 
895 	goto err;
896 
897 got_super:
898 	if (le16_to_cpu(sb->sb->block_size) << 9 <
899 	    bdev_logical_block_size(sb->bdev) &&
900 	    opt_get(*opts, direct_io)) {
901 #ifndef __KERNEL__
902 		opt_set(*opts, direct_io, false);
903 		bch2_free_super(sb);
904 		goto retry;
905 #endif
906 		prt_printf(&err, "block size (%u) smaller than device block size (%u)",
907 		       le16_to_cpu(sb->sb->block_size) << 9,
908 		       bdev_logical_block_size(sb->bdev));
909 		ret = -BCH_ERR_block_size_too_small;
910 		goto err;
911 	}
912 
913 	sb->have_layout = true;
914 
915 	ret = bch2_sb_validate(sb->sb, offset, 0, &err);
916 	if (ret) {
917 		bch2_print_opts(opts, KERN_ERR "bcachefs (%s): error validating superblock: %s\n",
918 				path, err.buf);
919 		goto err_no_print;
920 	}
921 out:
922 	printbuf_exit(&err);
923 	return ret;
924 err:
925 	bch2_print_opts(opts, KERN_ERR "bcachefs (%s): error reading superblock: %s\n",
926 			path, err.buf);
927 err_no_print:
928 	bch2_free_super(sb);
929 	goto out;
930 }
931 
bch2_read_super(const char * path,struct bch_opts * opts,struct bch_sb_handle * sb)932 int bch2_read_super(const char *path, struct bch_opts *opts,
933 		    struct bch_sb_handle *sb)
934 {
935 	return __bch2_read_super(path, opts, sb, false);
936 }
937 
938 /* provide a silenced version for mount.bcachefs */
939 
bch2_read_super_silent(const char * path,struct bch_opts * opts,struct bch_sb_handle * sb)940 int bch2_read_super_silent(const char *path, struct bch_opts *opts,
941 		    struct bch_sb_handle *sb)
942 {
943 	return __bch2_read_super(path, opts, sb, true);
944 }
945 
946 /* write superblock: */
947 
write_super_endio(struct bio * bio)948 static void write_super_endio(struct bio *bio)
949 {
950 	struct bch_dev *ca = bio->bi_private;
951 
952 	bch2_account_io_success_fail(ca, bio_data_dir(bio), !bio->bi_status);
953 
954 	/* XXX: return errors directly */
955 
956 	if (bio->bi_status) {
957 		bch_err_dev_ratelimited(ca, "superblock %s error: %s",
958 			       str_write_read(bio_data_dir(bio)),
959 			       bch2_blk_status_to_str(bio->bi_status));
960 		ca->sb_write_error = 1;
961 	}
962 
963 	closure_put(&ca->fs->sb_write);
964 	percpu_ref_put(&ca->io_ref[READ]);
965 }
966 
read_back_super(struct bch_fs * c,struct bch_dev * ca)967 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
968 {
969 	struct bch_sb *sb = ca->disk_sb.sb;
970 	struct bio *bio = ca->disk_sb.bio;
971 
972 	memset(ca->sb_read_scratch, 0, BCH_SB_READ_SCRATCH_BUF_SIZE);
973 
974 	bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
975 	bio->bi_iter.bi_sector	= le64_to_cpu(sb->layout.sb_offset[0]);
976 	bio->bi_end_io		= write_super_endio;
977 	bio->bi_private		= ca;
978 	bch2_bio_map(bio, ca->sb_read_scratch, BCH_SB_READ_SCRATCH_BUF_SIZE);
979 
980 	this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb], bio_sectors(bio));
981 
982 	percpu_ref_get(&ca->io_ref[READ]);
983 	closure_bio_submit(bio, &c->sb_write);
984 }
985 
write_one_super(struct bch_fs * c,struct bch_dev * ca,unsigned idx)986 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
987 {
988 	struct bch_sb *sb = ca->disk_sb.sb;
989 	struct bio *bio = ca->disk_sb.bio;
990 
991 	sb->offset = sb->layout.sb_offset[idx];
992 
993 	SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
994 	sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
995 				null_nonce(), sb);
996 
997 	bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
998 	bio->bi_iter.bi_sector	= le64_to_cpu(sb->offset);
999 	bio->bi_end_io		= write_super_endio;
1000 	bio->bi_private		= ca;
1001 	bch2_bio_map(bio, sb,
1002 		     roundup((size_t) vstruct_bytes(sb),
1003 			     bdev_logical_block_size(ca->disk_sb.bdev)));
1004 
1005 	this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
1006 		     bio_sectors(bio));
1007 
1008 	percpu_ref_get(&ca->io_ref[READ]);
1009 	closure_bio_submit(bio, &c->sb_write);
1010 }
1011 
bch2_write_super(struct bch_fs * c)1012 int bch2_write_super(struct bch_fs *c)
1013 {
1014 	struct closure *cl = &c->sb_write;
1015 	struct printbuf err = PRINTBUF;
1016 	unsigned sb = 0, nr_wrote;
1017 	struct bch_devs_mask sb_written;
1018 	bool wrote, can_mount_without_written, can_mount_with_written;
1019 	unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
1020 	DARRAY(struct bch_dev *) online_devices = {};
1021 	int ret = 0;
1022 
1023 	trace_and_count(c, write_super, c, _RET_IP_);
1024 
1025 	if (c->opts.very_degraded)
1026 		degraded_flags |= BCH_FORCE_IF_LOST;
1027 
1028 	lockdep_assert_held(&c->sb_lock);
1029 
1030 	closure_init_stack(cl);
1031 	memset(&sb_written, 0, sizeof(sb_written));
1032 
1033 	/*
1034 	 * Note: we do writes to RO devices here, and we might want to change
1035 	 * that in the future.
1036 	 *
1037 	 * For now, we expect to be able to call write_super() when we're not
1038 	 * yet RW:
1039 	 */
1040 	for_each_online_member(c, ca) {
1041 		ret = darray_push(&online_devices, ca);
1042 		if (bch2_fs_fatal_err_on(ret, c, "%s: error allocating online devices", __func__)) {
1043 			percpu_ref_put(&ca->io_ref[READ]);
1044 			goto out;
1045 		}
1046 		percpu_ref_get(&ca->io_ref[READ]);
1047 	}
1048 
1049 	/* Make sure we're using the new magic numbers: */
1050 	c->disk_sb.sb->magic = BCHFS_MAGIC;
1051 	c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
1052 
1053 	le64_add_cpu(&c->disk_sb.sb->seq, 1);
1054 
1055 	struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1056 	darray_for_each(online_devices, ca)
1057 		__bch2_members_v2_get_mut(mi, (*ca)->dev_idx)->seq = c->disk_sb.sb->seq;
1058 	c->disk_sb.sb->write_time = cpu_to_le64(ktime_get_real_seconds());
1059 
1060 	if (test_bit(BCH_FS_error, &c->flags))
1061 		SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
1062 	if (test_bit(BCH_FS_topology_error, &c->flags))
1063 		SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
1064 
1065 	SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
1066 
1067 	bch2_sb_counters_from_cpu(c);
1068 	bch2_sb_members_from_cpu(c);
1069 	bch2_sb_members_cpy_v2_v1(&c->disk_sb);
1070 	bch2_sb_errors_from_cpu(c);
1071 	bch2_sb_downgrade_update(c);
1072 
1073 	darray_for_each(online_devices, ca)
1074 		bch2_sb_from_fs(c, (*ca));
1075 
1076 	darray_for_each(online_devices, ca) {
1077 		printbuf_reset(&err);
1078 
1079 		ret = bch2_sb_validate((*ca)->disk_sb.sb, 0, BCH_VALIDATE_write, &err);
1080 		if (ret) {
1081 			bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
1082 			goto out;
1083 		}
1084 	}
1085 
1086 	if (c->opts.nochanges)
1087 		goto out;
1088 
1089 	/*
1090 	 * Defer writing the superblock until filesystem initialization is
1091 	 * complete - don't write out a partly initialized superblock:
1092 	 */
1093 	if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
1094 		goto out;
1095 
1096 	if (le16_to_cpu(c->disk_sb.sb->version) > bcachefs_metadata_version_current) {
1097 		struct printbuf buf = PRINTBUF;
1098 		prt_printf(&buf, "attempting to write superblock that wasn't version downgraded (");
1099 		bch2_version_to_text(&buf, le16_to_cpu(c->disk_sb.sb->version));
1100 		prt_str(&buf, " > ");
1101 		bch2_version_to_text(&buf, bcachefs_metadata_version_current);
1102 		prt_str(&buf, ")");
1103 		bch2_fs_fatal_error(c, ": %s", buf.buf);
1104 		printbuf_exit(&buf);
1105 		ret = -BCH_ERR_sb_not_downgraded;
1106 		goto out;
1107 	}
1108 
1109 	darray_for_each(online_devices, ca) {
1110 		__set_bit((*ca)->dev_idx, sb_written.d);
1111 		(*ca)->sb_write_error = 0;
1112 	}
1113 
1114 	darray_for_each(online_devices, ca)
1115 		read_back_super(c, *ca);
1116 	closure_sync(cl);
1117 
1118 	darray_for_each(online_devices, cap) {
1119 		struct bch_dev *ca = *cap;
1120 
1121 		if (ca->sb_write_error)
1122 			continue;
1123 
1124 		if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
1125 			struct printbuf buf = PRINTBUF;
1126 			prt_char(&buf, ' ');
1127 			prt_bdevname(&buf, ca->disk_sb.bdev);
1128 			prt_printf(&buf,
1129 				": Superblock write was silently dropped! (seq %llu expected %llu)",
1130 				le64_to_cpu(ca->sb_read_scratch->seq),
1131 				ca->disk_sb.seq);
1132 
1133 			if (c->opts.errors != BCH_ON_ERROR_continue &&
1134 			    c->opts.errors != BCH_ON_ERROR_fix_safe) {
1135 				ret = -BCH_ERR_erofs_sb_err;
1136 				bch2_fs_fatal_error(c, "%s", buf.buf);
1137 			} else {
1138 				bch_err(c, "%s", buf.buf);
1139 			}
1140 
1141 			printbuf_exit(&buf);
1142 		}
1143 
1144 		if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
1145 			struct printbuf buf = PRINTBUF;
1146 			prt_char(&buf, ' ');
1147 			prt_bdevname(&buf, ca->disk_sb.bdev);
1148 			prt_printf(&buf,
1149 				": Superblock modified by another process (seq %llu expected %llu)",
1150 				le64_to_cpu(ca->sb_read_scratch->seq),
1151 				ca->disk_sb.seq);
1152 			bch2_fs_fatal_error(c, "%s", buf.buf);
1153 			printbuf_exit(&buf);
1154 			ret = -BCH_ERR_erofs_sb_err;
1155 		}
1156 	}
1157 
1158 	if (ret)
1159 		goto out;
1160 
1161 	do {
1162 		wrote = false;
1163 		darray_for_each(online_devices, cap) {
1164 			struct bch_dev *ca = *cap;
1165 			if (!ca->sb_write_error &&
1166 			    sb < ca->disk_sb.sb->layout.nr_superblocks) {
1167 				write_one_super(c, ca, sb);
1168 				wrote = true;
1169 			}
1170 		}
1171 		closure_sync(cl);
1172 		sb++;
1173 	} while (wrote);
1174 
1175 	darray_for_each(online_devices, cap) {
1176 		struct bch_dev *ca = *cap;
1177 		if (ca->sb_write_error)
1178 			__clear_bit(ca->dev_idx, sb_written.d);
1179 		else
1180 			ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
1181 	}
1182 
1183 	nr_wrote = dev_mask_nr(&sb_written);
1184 
1185 	can_mount_with_written =
1186 		bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1187 
1188 	for (unsigned i = 0; i < ARRAY_SIZE(sb_written.d); i++)
1189 		sb_written.d[i] = ~sb_written.d[i];
1190 
1191 	can_mount_without_written =
1192 		bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1193 
1194 	/*
1195 	 * If we would be able to mount _without_ the devices we successfully
1196 	 * wrote superblocks to, we weren't able to write to enough devices:
1197 	 *
1198 	 * Exception: if we can mount without the successes because we haven't
1199 	 * written anything (new filesystem), we continue if we'd be able to
1200 	 * mount with the devices we did successfully write to:
1201 	 */
1202 	if (bch2_fs_fatal_err_on(!nr_wrote ||
1203 				 !can_mount_with_written ||
1204 				 (can_mount_without_written &&
1205 				  !can_mount_with_written), c,
1206 		": Unable to write superblock to sufficient devices (from %ps)",
1207 		(void *) _RET_IP_))
1208 		ret = -BCH_ERR_erofs_sb_err;
1209 out:
1210 	/* Make new options visible after they're persistent: */
1211 	bch2_sb_update(c);
1212 	darray_for_each(online_devices, ca)
1213 		percpu_ref_put(&(*ca)->io_ref[READ]);
1214 	darray_exit(&online_devices);
1215 	printbuf_exit(&err);
1216 	return ret;
1217 }
1218 
__bch2_check_set_feature(struct bch_fs * c,unsigned feat)1219 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
1220 {
1221 	mutex_lock(&c->sb_lock);
1222 	if (!(c->sb.features & (1ULL << feat))) {
1223 		c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
1224 
1225 		bch2_write_super(c);
1226 	}
1227 	mutex_unlock(&c->sb_lock);
1228 }
1229 
1230 /* Downgrade if superblock is at a higher version than currently supported: */
bch2_check_version_downgrade(struct bch_fs * c)1231 bool bch2_check_version_downgrade(struct bch_fs *c)
1232 {
1233 	bool ret = bcachefs_metadata_version_current < c->sb.version;
1234 
1235 	lockdep_assert_held(&c->sb_lock);
1236 
1237 	/*
1238 	 * Downgrade, if superblock is at a higher version than currently
1239 	 * supported:
1240 	 *
1241 	 * c->sb will be checked before we write the superblock, so update it as
1242 	 * well:
1243 	 */
1244 	if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current)
1245 		SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1246 	if (BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb) > bcachefs_metadata_version_current)
1247 		SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb, bcachefs_metadata_version_current);
1248 	if (c->sb.version > bcachefs_metadata_version_current)
1249 		c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1250 	if (c->sb.version_min > bcachefs_metadata_version_current)
1251 		c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
1252 	c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1253 	return ret;
1254 }
1255 
bch2_sb_upgrade(struct bch_fs * c,unsigned new_version,bool incompat)1256 void bch2_sb_upgrade(struct bch_fs *c, unsigned new_version, bool incompat)
1257 {
1258 	lockdep_assert_held(&c->sb_lock);
1259 
1260 	if (BCH_VERSION_MAJOR(new_version) >
1261 	    BCH_VERSION_MAJOR(le16_to_cpu(c->disk_sb.sb->version)))
1262 		bch2_sb_field_resize(&c->disk_sb, downgrade, 0);
1263 
1264 	c->disk_sb.sb->version = cpu_to_le16(new_version);
1265 
1266 	if (incompat) {
1267 		c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1268 		SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb,
1269 			max(BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb), new_version));
1270 	}
1271 }
1272 
bch2_sb_ext_validate(struct bch_sb * sb,struct bch_sb_field * f,enum bch_validate_flags flags,struct printbuf * err)1273 static int bch2_sb_ext_validate(struct bch_sb *sb, struct bch_sb_field *f,
1274 				enum bch_validate_flags flags, struct printbuf *err)
1275 {
1276 	if (vstruct_bytes(f) < 88) {
1277 		prt_printf(err, "field too small (%zu < %u)", vstruct_bytes(f), 88);
1278 		return -BCH_ERR_invalid_sb_ext;
1279 	}
1280 
1281 	return 0;
1282 }
1283 
bch2_sb_ext_to_text(struct printbuf * out,struct bch_sb * sb,struct bch_sb_field * f)1284 static void bch2_sb_ext_to_text(struct printbuf *out, struct bch_sb *sb,
1285 				struct bch_sb_field *f)
1286 {
1287 	struct bch_sb_field_ext *e = field_to_type(f, ext);
1288 
1289 	prt_printf(out, "Recovery passes required:\t");
1290 	prt_bitflags(out, bch2_recovery_passes,
1291 		     bch2_recovery_passes_from_stable(le64_to_cpu(e->recovery_passes_required[0])));
1292 	prt_newline(out);
1293 
1294 	unsigned long *errors_silent = kmalloc(sizeof(e->errors_silent), GFP_KERNEL);
1295 	if (errors_silent) {
1296 		le_bitvector_to_cpu(errors_silent, (void *) e->errors_silent, sizeof(e->errors_silent) * 8);
1297 
1298 		prt_printf(out, "Errors to silently fix:\t");
1299 		prt_bitflags_vector(out, bch2_sb_error_strs, errors_silent,
1300 				    min(BCH_FSCK_ERR_MAX, sizeof(e->errors_silent) * 8));
1301 		prt_newline(out);
1302 
1303 		kfree(errors_silent);
1304 	}
1305 
1306 	prt_printf(out, "Btrees with missing data:\t");
1307 	prt_bitflags(out, __bch2_btree_ids, le64_to_cpu(e->btrees_lost_data));
1308 	prt_newline(out);
1309 }
1310 
1311 static const struct bch_sb_field_ops bch_sb_field_ops_ext = {
1312 	.validate	= bch2_sb_ext_validate,
1313 	.to_text	= bch2_sb_ext_to_text,
1314 };
1315 
1316 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1317 #define x(f, nr)					\
1318 	[BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1319 	BCH_SB_FIELDS()
1320 #undef x
1321 };
1322 
1323 static const struct bch_sb_field_ops bch2_sb_field_null_ops;
1324 
bch2_sb_field_type_ops(unsigned type)1325 static const struct bch_sb_field_ops *bch2_sb_field_type_ops(unsigned type)
1326 {
1327 	return likely(type < ARRAY_SIZE(bch2_sb_field_ops))
1328 		? bch2_sb_field_ops[type]
1329 		: &bch2_sb_field_null_ops;
1330 }
1331 
bch2_sb_field_validate(struct bch_sb * sb,struct bch_sb_field * f,enum bch_validate_flags flags,struct printbuf * err)1332 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1333 				  enum bch_validate_flags flags, struct printbuf *err)
1334 {
1335 	unsigned type = le32_to_cpu(f->type);
1336 	struct printbuf field_err = PRINTBUF;
1337 	const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1338 	int ret;
1339 
1340 	ret = ops->validate ? ops->validate(sb, f, flags, &field_err) : 0;
1341 	if (ret) {
1342 		prt_printf(err, "Invalid superblock section %s: %s",
1343 			   bch2_sb_fields[type], field_err.buf);
1344 		prt_newline(err);
1345 		bch2_sb_field_to_text(err, sb, f);
1346 	}
1347 
1348 	printbuf_exit(&field_err);
1349 	return ret;
1350 }
1351 
__bch2_sb_field_to_text(struct printbuf * out,struct bch_sb * sb,struct bch_sb_field * f)1352 void __bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1353 			     struct bch_sb_field *f)
1354 {
1355 	unsigned type = le32_to_cpu(f->type);
1356 	const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1357 
1358 	if (!out->nr_tabstops)
1359 		printbuf_tabstop_push(out, 32);
1360 
1361 	if (ops->to_text)
1362 		ops->to_text(out, sb, f);
1363 }
1364 
bch2_sb_field_to_text(struct printbuf * out,struct bch_sb * sb,struct bch_sb_field * f)1365 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1366 			   struct bch_sb_field *f)
1367 {
1368 	unsigned type = le32_to_cpu(f->type);
1369 
1370 	if (type < BCH_SB_FIELD_NR)
1371 		prt_printf(out, "%s", bch2_sb_fields[type]);
1372 	else
1373 		prt_printf(out, "(unknown field %u)", type);
1374 
1375 	prt_printf(out, " (size %zu):", vstruct_bytes(f));
1376 	prt_newline(out);
1377 
1378 	__bch2_sb_field_to_text(out, sb, f);
1379 }
1380 
bch2_sb_layout_to_text(struct printbuf * out,struct bch_sb_layout * l)1381 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1382 {
1383 	unsigned i;
1384 
1385 	prt_printf(out, "Type:                    %u", l->layout_type);
1386 	prt_newline(out);
1387 
1388 	prt_str(out, "Superblock max size:     ");
1389 	prt_units_u64(out, 512 << l->sb_max_size_bits);
1390 	prt_newline(out);
1391 
1392 	prt_printf(out, "Nr superblocks:          %u", l->nr_superblocks);
1393 	prt_newline(out);
1394 
1395 	prt_str(out, "Offsets:                 ");
1396 	for (i = 0; i < l->nr_superblocks; i++) {
1397 		if (i)
1398 			prt_str(out, ", ");
1399 		prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1400 	}
1401 	prt_newline(out);
1402 }
1403 
bch2_sb_to_text(struct printbuf * out,struct bch_sb * sb,bool print_layout,unsigned fields)1404 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1405 		     bool print_layout, unsigned fields)
1406 {
1407 	if (!out->nr_tabstops)
1408 		printbuf_tabstop_push(out, 44);
1409 
1410 	prt_printf(out, "External UUID:\t");
1411 	pr_uuid(out, sb->user_uuid.b);
1412 	prt_newline(out);
1413 
1414 	prt_printf(out, "Internal UUID:\t");
1415 	pr_uuid(out, sb->uuid.b);
1416 	prt_newline(out);
1417 
1418 	prt_printf(out, "Magic number:\t");
1419 	pr_uuid(out, sb->magic.b);
1420 	prt_newline(out);
1421 
1422 	prt_printf(out, "Device index:\t%u\n", sb->dev_idx);
1423 
1424 	prt_printf(out, "Label:\t");
1425 	if (!strlen(sb->label))
1426 		prt_printf(out, "(none)");
1427 	else
1428 		prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1429 	prt_newline(out);
1430 
1431 	prt_printf(out, "Version:\t");
1432 	bch2_version_to_text(out, le16_to_cpu(sb->version));
1433 	prt_newline(out);
1434 
1435 	prt_printf(out, "Incompatible features allowed:\t");
1436 	bch2_version_to_text(out, BCH_SB_VERSION_INCOMPAT_ALLOWED(sb));
1437 	prt_newline(out);
1438 
1439 	prt_printf(out, "Incompatible features in use:\t");
1440 	bch2_version_to_text(out, BCH_SB_VERSION_INCOMPAT(sb));
1441 	prt_newline(out);
1442 
1443 	prt_printf(out, "Version upgrade complete:\t");
1444 	bch2_version_to_text(out, BCH_SB_VERSION_UPGRADE_COMPLETE(sb));
1445 	prt_newline(out);
1446 
1447 	prt_printf(out, "Oldest version on disk:\t");
1448 	bch2_version_to_text(out, le16_to_cpu(sb->version_min));
1449 	prt_newline(out);
1450 
1451 	prt_printf(out, "Created:\t");
1452 	if (sb->time_base_lo)
1453 		bch2_prt_datetime(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1454 	else
1455 		prt_printf(out, "(not set)");
1456 	prt_newline(out);
1457 
1458 	prt_printf(out, "Sequence number:\t");
1459 	prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1460 	prt_newline(out);
1461 
1462 	prt_printf(out, "Time of last write:\t");
1463 	bch2_prt_datetime(out, le64_to_cpu(sb->write_time));
1464 	prt_newline(out);
1465 
1466 	prt_printf(out, "Superblock size:\t");
1467 	prt_units_u64(out, vstruct_bytes(sb));
1468 	prt_str(out, "/");
1469 	prt_units_u64(out, 512ULL << sb->layout.sb_max_size_bits);
1470 	prt_newline(out);
1471 
1472 	prt_printf(out, "Clean:\t%llu\n", BCH_SB_CLEAN(sb));
1473 	prt_printf(out, "Devices:\t%u\n", bch2_sb_nr_devices(sb));
1474 
1475 	prt_printf(out, "Sections:\t");
1476 	u64 fields_have = 0;
1477 	vstruct_for_each(sb, f)
1478 		fields_have |= 1 << le32_to_cpu(f->type);
1479 	prt_bitflags(out, bch2_sb_fields, fields_have);
1480 	prt_newline(out);
1481 
1482 	prt_printf(out, "Features:\t");
1483 	prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1484 	prt_newline(out);
1485 
1486 	prt_printf(out, "Compat features:\t");
1487 	prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1488 	prt_newline(out);
1489 
1490 	prt_newline(out);
1491 	prt_printf(out, "Options:");
1492 	prt_newline(out);
1493 	printbuf_indent_add(out, 2);
1494 	{
1495 		enum bch_opt_id id;
1496 
1497 		for (id = 0; id < bch2_opts_nr; id++) {
1498 			const struct bch_option *opt = bch2_opt_table + id;
1499 
1500 			if (opt->get_sb) {
1501 				u64 v = bch2_opt_from_sb(sb, id, -1);
1502 
1503 				prt_printf(out, "%s:\t", opt->attr.name);
1504 				bch2_opt_to_text(out, NULL, sb, opt, v,
1505 						 OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1506 				prt_newline(out);
1507 			}
1508 		}
1509 	}
1510 
1511 	printbuf_indent_sub(out, 2);
1512 
1513 	if (print_layout) {
1514 		prt_newline(out);
1515 		prt_printf(out, "layout:");
1516 		prt_newline(out);
1517 		printbuf_indent_add(out, 2);
1518 		bch2_sb_layout_to_text(out, &sb->layout);
1519 		printbuf_indent_sub(out, 2);
1520 	}
1521 
1522 	vstruct_for_each(sb, f)
1523 		if (fields & (1 << le32_to_cpu(f->type))) {
1524 			prt_newline(out);
1525 			bch2_sb_field_to_text(out, sb, f);
1526 		}
1527 }
1528