xref: /linux/fs/bcachefs/super-io.c (revision a19ce320c379e0519b68178c596e43d1d5dda03b)
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 
27 static const struct blk_holder_ops bch2_sb_handle_bdev_ops = {
28 };
29 
30 struct bch2_metadata_version {
31 	u16		version;
32 	const char	*name;
33 };
34 
35 static const struct bch2_metadata_version bch2_metadata_versions[] = {
36 #define x(n, v) {		\
37 	.version = v,				\
38 	.name = #n,				\
39 },
40 	BCH_METADATA_VERSIONS()
41 #undef x
42 };
43 
44 void bch2_version_to_text(struct printbuf *out, unsigned v)
45 {
46 	const char *str = "(unknown version)";
47 
48 	for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
49 		if (bch2_metadata_versions[i].version == v) {
50 			str = bch2_metadata_versions[i].name;
51 			break;
52 		}
53 
54 	prt_printf(out, "%u.%u: %s", BCH_VERSION_MAJOR(v), BCH_VERSION_MINOR(v), str);
55 }
56 
57 unsigned bch2_latest_compatible_version(unsigned v)
58 {
59 	if (!BCH_VERSION_MAJOR(v))
60 		return v;
61 
62 	for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
63 		if (bch2_metadata_versions[i].version > v &&
64 		    BCH_VERSION_MAJOR(bch2_metadata_versions[i].version) ==
65 		    BCH_VERSION_MAJOR(v))
66 			v = bch2_metadata_versions[i].version;
67 
68 	return v;
69 }
70 
71 const char * const bch2_sb_fields[] = {
72 #define x(name, nr)	#name,
73 	BCH_SB_FIELDS()
74 #undef x
75 	NULL
76 };
77 
78 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
79 				  enum bch_validate_flags, struct printbuf *);
80 
81 struct bch_sb_field *bch2_sb_field_get_id(struct bch_sb *sb,
82 				      enum bch_sb_field_type type)
83 {
84 	/* XXX: need locking around superblock to access optional fields */
85 
86 	vstruct_for_each(sb, f)
87 		if (le32_to_cpu(f->type) == type)
88 			return f;
89 	return NULL;
90 }
91 
92 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
93 						   struct bch_sb_field *f,
94 						   unsigned u64s)
95 {
96 	unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
97 	unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
98 
99 	BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
100 
101 	if (!f && !u64s) {
102 		/* nothing to do: */
103 	} else if (!f) {
104 		f = vstruct_last(sb->sb);
105 		memset(f, 0, sizeof(u64) * u64s);
106 		f->u64s = cpu_to_le32(u64s);
107 		f->type = 0;
108 	} else {
109 		void *src, *dst;
110 
111 		src = vstruct_end(f);
112 
113 		if (u64s) {
114 			f->u64s = cpu_to_le32(u64s);
115 			dst = vstruct_end(f);
116 		} else {
117 			dst = f;
118 		}
119 
120 		memmove(dst, src, vstruct_end(sb->sb) - src);
121 
122 		if (dst > src)
123 			memset(src, 0, dst - src);
124 	}
125 
126 	sb->sb->u64s = cpu_to_le32(sb_u64s);
127 
128 	return u64s ? f : NULL;
129 }
130 
131 void bch2_sb_field_delete(struct bch_sb_handle *sb,
132 			  enum bch_sb_field_type type)
133 {
134 	struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
135 
136 	if (f)
137 		__bch2_sb_field_resize(sb, f, 0);
138 }
139 
140 /* Superblock realloc/free: */
141 
142 void bch2_free_super(struct bch_sb_handle *sb)
143 {
144 	kfree(sb->bio);
145 	if (!IS_ERR_OR_NULL(sb->s_bdev_file))
146 		bdev_fput(sb->s_bdev_file);
147 	kfree(sb->holder);
148 	kfree(sb->sb_name);
149 
150 	kfree(sb->sb);
151 	memset(sb, 0, sizeof(*sb));
152 }
153 
154 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
155 {
156 	size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
157 	size_t new_buffer_size;
158 	struct bch_sb *new_sb;
159 	struct bio *bio;
160 
161 	if (sb->bdev)
162 		new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
163 
164 	new_buffer_size = roundup_pow_of_two(new_bytes);
165 
166 	if (sb->sb && sb->buffer_size >= new_buffer_size)
167 		return 0;
168 
169 	if (sb->sb && sb->have_layout) {
170 		u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
171 
172 		if (new_bytes > max_bytes) {
173 			struct printbuf buf = PRINTBUF;
174 
175 			prt_bdevname(&buf, sb->bdev);
176 			prt_printf(&buf, ": superblock too big: want %zu but have %llu", new_bytes, max_bytes);
177 			pr_err("%s", buf.buf);
178 			printbuf_exit(&buf);
179 			return -BCH_ERR_ENOSPC_sb;
180 		}
181 	}
182 
183 	if (sb->buffer_size >= new_buffer_size && sb->sb)
184 		return 0;
185 
186 	if (dynamic_fault("bcachefs:add:super_realloc"))
187 		return -BCH_ERR_ENOMEM_sb_realloc_injected;
188 
189 	new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
190 	if (!new_sb)
191 		return -BCH_ERR_ENOMEM_sb_buf_realloc;
192 
193 	sb->sb = new_sb;
194 
195 	if (sb->have_bio) {
196 		unsigned nr_bvecs = buf_pages(sb->sb, new_buffer_size);
197 
198 		bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
199 		if (!bio)
200 			return -BCH_ERR_ENOMEM_sb_bio_realloc;
201 
202 		bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
203 
204 		kfree(sb->bio);
205 		sb->bio = bio;
206 	}
207 
208 	sb->buffer_size = new_buffer_size;
209 
210 	return 0;
211 }
212 
213 struct bch_sb_field *bch2_sb_field_resize_id(struct bch_sb_handle *sb,
214 					  enum bch_sb_field_type type,
215 					  unsigned u64s)
216 {
217 	struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
218 	ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
219 	ssize_t d = -old_u64s + u64s;
220 
221 	if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
222 		return NULL;
223 
224 	if (sb->fs_sb) {
225 		struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
226 
227 		lockdep_assert_held(&c->sb_lock);
228 
229 		/* XXX: we're not checking that offline device have enough space */
230 
231 		for_each_online_member(c, ca) {
232 			struct bch_sb_handle *dev_sb = &ca->disk_sb;
233 
234 			if (bch2_sb_realloc(dev_sb, le32_to_cpu(dev_sb->sb->u64s) + d)) {
235 				percpu_ref_put(&ca->io_ref);
236 				return NULL;
237 			}
238 		}
239 	}
240 
241 	f = bch2_sb_field_get_id(sb->sb, type);
242 	f = __bch2_sb_field_resize(sb, f, u64s);
243 	if (f)
244 		f->type = cpu_to_le32(type);
245 	return f;
246 }
247 
248 struct bch_sb_field *bch2_sb_field_get_minsize_id(struct bch_sb_handle *sb,
249 						  enum bch_sb_field_type type,
250 						  unsigned u64s)
251 {
252 	struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
253 
254 	if (!f || le32_to_cpu(f->u64s) < u64s)
255 		f = bch2_sb_field_resize_id(sb, type, u64s);
256 	return f;
257 }
258 
259 /* Superblock validate: */
260 
261 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
262 {
263 	u64 offset, prev_offset, max_sectors;
264 	unsigned i;
265 
266 	BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
267 
268 	if (!uuid_equal(&layout->magic, &BCACHE_MAGIC) &&
269 	    !uuid_equal(&layout->magic, &BCHFS_MAGIC)) {
270 		prt_printf(out, "Not a bcachefs superblock layout");
271 		return -BCH_ERR_invalid_sb_layout;
272 	}
273 
274 	if (layout->layout_type != 0) {
275 		prt_printf(out, "Invalid superblock layout type %u",
276 		       layout->layout_type);
277 		return -BCH_ERR_invalid_sb_layout_type;
278 	}
279 
280 	if (!layout->nr_superblocks) {
281 		prt_printf(out, "Invalid superblock layout: no superblocks");
282 		return -BCH_ERR_invalid_sb_layout_nr_superblocks;
283 	}
284 
285 	if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
286 		prt_printf(out, "Invalid superblock layout: too many superblocks");
287 		return -BCH_ERR_invalid_sb_layout_nr_superblocks;
288 	}
289 
290 	max_sectors = 1 << layout->sb_max_size_bits;
291 
292 	prev_offset = le64_to_cpu(layout->sb_offset[0]);
293 
294 	for (i = 1; i < layout->nr_superblocks; i++) {
295 		offset = le64_to_cpu(layout->sb_offset[i]);
296 
297 		if (offset < prev_offset + max_sectors) {
298 			prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
299 			       "  (sb %u ends at %llu next starts at %llu",
300 			       i - 1, prev_offset + max_sectors, offset);
301 			return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
302 		}
303 		prev_offset = offset;
304 	}
305 
306 	return 0;
307 }
308 
309 static int bch2_sb_compatible(struct bch_sb *sb, struct printbuf *out)
310 {
311 	u16 version		= le16_to_cpu(sb->version);
312 	u16 version_min		= le16_to_cpu(sb->version_min);
313 
314 	if (!bch2_version_compatible(version)) {
315 		prt_str(out, "Unsupported superblock version ");
316 		bch2_version_to_text(out, version);
317 		prt_str(out, " (min ");
318 		bch2_version_to_text(out, bcachefs_metadata_version_min);
319 		prt_str(out, ", max ");
320 		bch2_version_to_text(out, bcachefs_metadata_version_current);
321 		prt_str(out, ")");
322 		return -BCH_ERR_invalid_sb_version;
323 	}
324 
325 	if (!bch2_version_compatible(version_min)) {
326 		prt_str(out, "Unsupported superblock version_min ");
327 		bch2_version_to_text(out, version_min);
328 		prt_str(out, " (min ");
329 		bch2_version_to_text(out, bcachefs_metadata_version_min);
330 		prt_str(out, ", max ");
331 		bch2_version_to_text(out, bcachefs_metadata_version_current);
332 		prt_str(out, ")");
333 		return -BCH_ERR_invalid_sb_version;
334 	}
335 
336 	if (version_min > version) {
337 		prt_str(out, "Bad minimum version ");
338 		bch2_version_to_text(out, version_min);
339 		prt_str(out, ", greater than version field ");
340 		bch2_version_to_text(out, version);
341 		return -BCH_ERR_invalid_sb_version;
342 	}
343 
344 	return 0;
345 }
346 
347 static int bch2_sb_validate(struct bch_sb_handle *disk_sb,
348 			    enum bch_validate_flags flags, struct printbuf *out)
349 {
350 	struct bch_sb *sb = disk_sb->sb;
351 	struct bch_sb_field_members_v1 *mi;
352 	enum bch_opt_id opt_id;
353 	u16 block_size;
354 	int ret;
355 
356 	ret = bch2_sb_compatible(sb, out);
357 	if (ret)
358 		return ret;
359 
360 	if (sb->features[1] ||
361 	    (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
362 		prt_printf(out, "Filesystem has incompatible features");
363 		return -BCH_ERR_invalid_sb_features;
364 	}
365 
366 	block_size = le16_to_cpu(sb->block_size);
367 
368 	if (block_size > PAGE_SECTORS) {
369 		prt_printf(out, "Block size too big (got %u, max %u)",
370 		       block_size, PAGE_SECTORS);
371 		return -BCH_ERR_invalid_sb_block_size;
372 	}
373 
374 	if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid))) {
375 		prt_printf(out, "Bad user UUID (got zeroes)");
376 		return -BCH_ERR_invalid_sb_uuid;
377 	}
378 
379 	if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid))) {
380 		prt_printf(out, "Bad internal UUID (got zeroes)");
381 		return -BCH_ERR_invalid_sb_uuid;
382 	}
383 
384 	if (!sb->nr_devices ||
385 	    sb->nr_devices > BCH_SB_MEMBERS_MAX) {
386 		prt_printf(out, "Bad number of member devices %u (max %u)",
387 		       sb->nr_devices, BCH_SB_MEMBERS_MAX);
388 		return -BCH_ERR_invalid_sb_too_many_members;
389 	}
390 
391 	if (sb->dev_idx >= sb->nr_devices) {
392 		prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
393 		       sb->dev_idx, sb->nr_devices);
394 		return -BCH_ERR_invalid_sb_dev_idx;
395 	}
396 
397 	if (!sb->time_precision ||
398 	    le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
399 		prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
400 		       le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
401 		return -BCH_ERR_invalid_sb_time_precision;
402 	}
403 
404 	if (!flags) {
405 		/*
406 		 * Been seeing a bug where these are getting inexplicably
407 		 * zeroed, so we're now validating them, but we have to be
408 		 * careful not to preven people's filesystems from mounting:
409 		 */
410 		if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
411 			SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
412 		if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
413 			SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
414 
415 		if (!BCH_SB_VERSION_UPGRADE_COMPLETE(sb))
416 			SET_BCH_SB_VERSION_UPGRADE_COMPLETE(sb, le16_to_cpu(sb->version));
417 	}
418 
419 	for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
420 		const struct bch_option *opt = bch2_opt_table + opt_id;
421 
422 		if (opt->get_sb != BCH2_NO_SB_OPT) {
423 			u64 v = bch2_opt_from_sb(sb, opt_id);
424 
425 			prt_printf(out, "Invalid option ");
426 			ret = bch2_opt_validate(opt, v, out);
427 			if (ret)
428 				return ret;
429 
430 			printbuf_reset(out);
431 		}
432 	}
433 
434 	/* validate layout */
435 	ret = validate_sb_layout(&sb->layout, out);
436 	if (ret)
437 		return ret;
438 
439 	vstruct_for_each(sb, f) {
440 		if (!f->u64s) {
441 			prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
442 			       le32_to_cpu(f->type));
443 			return -BCH_ERR_invalid_sb_field_size;
444 		}
445 
446 		if (vstruct_next(f) > vstruct_last(sb)) {
447 			prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
448 			       le32_to_cpu(f->type));
449 			return -BCH_ERR_invalid_sb_field_size;
450 		}
451 	}
452 
453 	/* members must be validated first: */
454 	mi = bch2_sb_field_get(sb, members_v1);
455 	if (!mi) {
456 		prt_printf(out, "Invalid superblock: member info area missing");
457 		return -BCH_ERR_invalid_sb_members_missing;
458 	}
459 
460 	ret = bch2_sb_field_validate(sb, &mi->field, flags, out);
461 	if (ret)
462 		return ret;
463 
464 	vstruct_for_each(sb, f) {
465 		if (le32_to_cpu(f->type) == BCH_SB_FIELD_members_v1)
466 			continue;
467 
468 		ret = bch2_sb_field_validate(sb, f, flags, out);
469 		if (ret)
470 			return ret;
471 	}
472 
473 	if ((flags & BCH_VALIDATE_write) &&
474 	    bch2_sb_member_get(sb, sb->dev_idx).seq != sb->seq) {
475 		prt_printf(out, "Invalid superblock: member seq %llu != sb seq %llu",
476 			   le64_to_cpu(bch2_sb_member_get(sb, sb->dev_idx).seq),
477 			   le64_to_cpu(sb->seq));
478 		return -BCH_ERR_invalid_sb_members_missing;
479 	}
480 
481 	return 0;
482 }
483 
484 /* device open: */
485 
486 static unsigned long le_ulong_to_cpu(unsigned long v)
487 {
488 	return sizeof(unsigned long) == 8
489 		? le64_to_cpu(v)
490 		: le32_to_cpu(v);
491 }
492 
493 static void le_bitvector_to_cpu(unsigned long *dst, unsigned long *src, unsigned nr)
494 {
495 	BUG_ON(nr & (BITS_PER_TYPE(long) - 1));
496 
497 	for (unsigned i = 0; i < BITS_TO_LONGS(nr); i++)
498 		dst[i] = le_ulong_to_cpu(src[i]);
499 }
500 
501 static void bch2_sb_update(struct bch_fs *c)
502 {
503 	struct bch_sb *src = c->disk_sb.sb;
504 
505 	lockdep_assert_held(&c->sb_lock);
506 
507 	c->sb.uuid		= src->uuid;
508 	c->sb.user_uuid		= src->user_uuid;
509 	c->sb.version		= le16_to_cpu(src->version);
510 	c->sb.version_min	= le16_to_cpu(src->version_min);
511 	c->sb.version_upgrade_complete = BCH_SB_VERSION_UPGRADE_COMPLETE(src);
512 	c->sb.nr_devices	= src->nr_devices;
513 	c->sb.clean		= BCH_SB_CLEAN(src);
514 	c->sb.encryption_type	= BCH_SB_ENCRYPTION_TYPE(src);
515 
516 	c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
517 	c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
518 
519 	/* XXX this is wrong, we need a 96 or 128 bit integer type */
520 	c->sb.time_base_lo	= div_u64(le64_to_cpu(src->time_base_lo),
521 					  c->sb.nsec_per_time_unit);
522 	c->sb.time_base_hi	= le32_to_cpu(src->time_base_hi);
523 
524 	c->sb.features		= le64_to_cpu(src->features[0]);
525 	c->sb.compat		= le64_to_cpu(src->compat[0]);
526 
527 	memset(c->sb.errors_silent, 0, sizeof(c->sb.errors_silent));
528 
529 	struct bch_sb_field_ext *ext = bch2_sb_field_get(src, ext);
530 	if (ext) {
531 		le_bitvector_to_cpu(c->sb.errors_silent, (void *) ext->errors_silent,
532 				    sizeof(c->sb.errors_silent) * 8);
533 		c->sb.btrees_lost_data = le64_to_cpu(ext->btrees_lost_data);
534 	}
535 
536 	for_each_member_device(c, ca) {
537 		struct bch_member m = bch2_sb_member_get(src, ca->dev_idx);
538 		ca->mi = bch2_mi_to_cpu(&m);
539 	}
540 }
541 
542 static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
543 {
544 	struct bch_sb_field *src_f, *dst_f;
545 	struct bch_sb *dst = dst_handle->sb;
546 	unsigned i;
547 
548 	dst->version		= src->version;
549 	dst->version_min	= src->version_min;
550 	dst->seq		= src->seq;
551 	dst->uuid		= src->uuid;
552 	dst->user_uuid		= src->user_uuid;
553 	memcpy(dst->label,	src->label, sizeof(dst->label));
554 
555 	dst->block_size		= src->block_size;
556 	dst->nr_devices		= src->nr_devices;
557 
558 	dst->time_base_lo	= src->time_base_lo;
559 	dst->time_base_hi	= src->time_base_hi;
560 	dst->time_precision	= src->time_precision;
561 	dst->write_time		= src->write_time;
562 
563 	memcpy(dst->flags,	src->flags,	sizeof(dst->flags));
564 	memcpy(dst->features,	src->features,	sizeof(dst->features));
565 	memcpy(dst->compat,	src->compat,	sizeof(dst->compat));
566 
567 	for (i = 0; i < BCH_SB_FIELD_NR; i++) {
568 		int d;
569 
570 		if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
571 			continue;
572 
573 		src_f = bch2_sb_field_get_id(src, i);
574 		dst_f = bch2_sb_field_get_id(dst, i);
575 
576 		d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
577 		    (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
578 		if (d > 0) {
579 			int ret = bch2_sb_realloc(dst_handle,
580 					le32_to_cpu(dst_handle->sb->u64s) + d);
581 
582 			if (ret)
583 				return ret;
584 
585 			dst = dst_handle->sb;
586 			dst_f = bch2_sb_field_get_id(dst, i);
587 		}
588 
589 		dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
590 				src_f ? le32_to_cpu(src_f->u64s) : 0);
591 
592 		if (src_f)
593 			memcpy(dst_f, src_f, vstruct_bytes(src_f));
594 	}
595 
596 	return 0;
597 }
598 
599 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
600 {
601 	int ret;
602 
603 	lockdep_assert_held(&c->sb_lock);
604 
605 	ret =   bch2_sb_realloc(&c->disk_sb, 0) ?:
606 		__copy_super(&c->disk_sb, src) ?:
607 		bch2_sb_replicas_to_cpu_replicas(c) ?:
608 		bch2_sb_disk_groups_to_cpu(c);
609 	if (ret)
610 		return ret;
611 
612 	bch2_sb_update(c);
613 	return 0;
614 }
615 
616 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
617 {
618 	return __copy_super(&ca->disk_sb, c->disk_sb.sb);
619 }
620 
621 /* read superblock: */
622 
623 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
624 {
625 	size_t bytes;
626 	int ret;
627 reread:
628 	bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
629 	sb->bio->bi_iter.bi_sector = offset;
630 	bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
631 
632 	ret = submit_bio_wait(sb->bio);
633 	if (ret) {
634 		prt_printf(err, "IO error: %i", ret);
635 		return ret;
636 	}
637 
638 	if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
639 	    !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC)) {
640 		prt_str(err, "Not a bcachefs superblock (got magic ");
641 		pr_uuid(err, sb->sb->magic.b);
642 		prt_str(err, ")");
643 		return -BCH_ERR_invalid_sb_magic;
644 	}
645 
646 	ret = bch2_sb_compatible(sb->sb, err);
647 	if (ret)
648 		return ret;
649 
650 	bytes = vstruct_bytes(sb->sb);
651 
652 	u64 sb_size = 512ULL << min(BCH_SB_LAYOUT_SIZE_BITS_MAX, sb->sb->layout.sb_max_size_bits);
653 	if (bytes > sb_size) {
654 		prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %llu)",
655 			   bytes, sb_size);
656 		return -BCH_ERR_invalid_sb_too_big;
657 	}
658 
659 	if (bytes > sb->buffer_size) {
660 		ret = bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s));
661 		if (ret)
662 			return ret;
663 		goto reread;
664 	}
665 
666 	enum bch_csum_type csum_type = BCH_SB_CSUM_TYPE(sb->sb);
667 	if (csum_type >= BCH_CSUM_NR) {
668 		prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
669 		return -BCH_ERR_invalid_sb_csum_type;
670 	}
671 
672 	/* XXX: verify MACs */
673 	struct bch_csum csum = csum_vstruct(NULL, csum_type, null_nonce(), sb->sb);
674 	if (bch2_crc_cmp(csum, sb->sb->csum)) {
675 		bch2_csum_err_msg(err, csum_type, sb->sb->csum, csum);
676 		return -BCH_ERR_invalid_sb_csum;
677 	}
678 
679 	sb->seq = le64_to_cpu(sb->sb->seq);
680 
681 	return 0;
682 }
683 
684 static int __bch2_read_super(const char *path, struct bch_opts *opts,
685 		    struct bch_sb_handle *sb, bool ignore_notbchfs_msg)
686 {
687 	u64 offset = opt_get(*opts, sb);
688 	struct bch_sb_layout layout;
689 	struct printbuf err = PRINTBUF;
690 	struct printbuf err2 = PRINTBUF;
691 	__le64 *i;
692 	int ret;
693 #ifndef __KERNEL__
694 retry:
695 #endif
696 	memset(sb, 0, sizeof(*sb));
697 	sb->mode	= BLK_OPEN_READ;
698 	sb->have_bio	= true;
699 	sb->holder	= kmalloc(1, GFP_KERNEL);
700 	if (!sb->holder)
701 		return -ENOMEM;
702 
703 	sb->sb_name = kstrdup(path, GFP_KERNEL);
704 	if (!sb->sb_name) {
705 		ret = -ENOMEM;
706 		prt_printf(&err, "error allocating memory for sb_name");
707 		goto err;
708 	}
709 
710 #ifndef __KERNEL__
711 	if (opt_get(*opts, direct_io) == false)
712 		sb->mode |= BLK_OPEN_BUFFERED;
713 #endif
714 
715 	if (!opt_get(*opts, noexcl))
716 		sb->mode |= BLK_OPEN_EXCL;
717 
718 	if (!opt_get(*opts, nochanges))
719 		sb->mode |= BLK_OPEN_WRITE;
720 
721 	sb->s_bdev_file = bdev_file_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
722 	if (IS_ERR(sb->s_bdev_file) &&
723 	    PTR_ERR(sb->s_bdev_file) == -EACCES &&
724 	    opt_get(*opts, read_only)) {
725 		sb->mode &= ~BLK_OPEN_WRITE;
726 
727 		sb->s_bdev_file = bdev_file_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
728 		if (!IS_ERR(sb->s_bdev_file))
729 			opt_set(*opts, nochanges, true);
730 	}
731 
732 	if (IS_ERR(sb->s_bdev_file)) {
733 		ret = PTR_ERR(sb->s_bdev_file);
734 		prt_printf(&err, "error opening %s: %s", path, bch2_err_str(ret));
735 		goto err;
736 	}
737 	sb->bdev = file_bdev(sb->s_bdev_file);
738 
739 	ret = bch2_sb_realloc(sb, 0);
740 	if (ret) {
741 		prt_printf(&err, "error allocating memory for superblock");
742 		goto err;
743 	}
744 
745 	if (bch2_fs_init_fault("read_super")) {
746 		prt_printf(&err, "dynamic fault");
747 		ret = -EFAULT;
748 		goto err;
749 	}
750 
751 	ret = read_one_super(sb, offset, &err);
752 	if (!ret)
753 		goto got_super;
754 
755 	if (opt_defined(*opts, sb))
756 		goto err;
757 
758 	prt_printf(&err2, "bcachefs (%s): error reading default superblock: %s\n",
759 	       path, err.buf);
760 	if (ret == -BCH_ERR_invalid_sb_magic && ignore_notbchfs_msg)
761 		bch2_print_opts(opts, KERN_INFO "%s", err2.buf);
762 	else
763 		bch2_print_opts(opts, KERN_ERR "%s", err2.buf);
764 
765 	printbuf_exit(&err2);
766 	printbuf_reset(&err);
767 
768 	/*
769 	 * Error reading primary superblock - read location of backup
770 	 * superblocks:
771 	 */
772 	bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
773 	sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
774 	/*
775 	 * use sb buffer to read layout, since sb buffer is page aligned but
776 	 * layout won't be:
777 	 */
778 	bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
779 
780 	ret = submit_bio_wait(sb->bio);
781 	if (ret) {
782 		prt_printf(&err, "IO error: %i", ret);
783 		goto err;
784 	}
785 
786 	memcpy(&layout, sb->sb, sizeof(layout));
787 	ret = validate_sb_layout(&layout, &err);
788 	if (ret)
789 		goto err;
790 
791 	for (i = layout.sb_offset;
792 	     i < layout.sb_offset + layout.nr_superblocks; i++) {
793 		offset = le64_to_cpu(*i);
794 
795 		if (offset == opt_get(*opts, sb))
796 			continue;
797 
798 		ret = read_one_super(sb, offset, &err);
799 		if (!ret)
800 			goto got_super;
801 	}
802 
803 	goto err;
804 
805 got_super:
806 	if (le16_to_cpu(sb->sb->block_size) << 9 <
807 	    bdev_logical_block_size(sb->bdev) &&
808 	    opt_get(*opts, direct_io)) {
809 #ifndef __KERNEL__
810 		opt_set(*opts, direct_io, false);
811 		bch2_free_super(sb);
812 		goto retry;
813 #endif
814 		prt_printf(&err, "block size (%u) smaller than device block size (%u)",
815 		       le16_to_cpu(sb->sb->block_size) << 9,
816 		       bdev_logical_block_size(sb->bdev));
817 		ret = -BCH_ERR_block_size_too_small;
818 		goto err;
819 	}
820 
821 	sb->have_layout = true;
822 
823 	ret = bch2_sb_validate(sb, 0, &err);
824 	if (ret) {
825 		bch2_print_opts(opts, KERN_ERR "bcachefs (%s): error validating superblock: %s\n",
826 				path, err.buf);
827 		goto err_no_print;
828 	}
829 out:
830 	printbuf_exit(&err);
831 	return ret;
832 err:
833 	bch2_print_opts(opts, KERN_ERR "bcachefs (%s): error reading superblock: %s\n",
834 			path, err.buf);
835 err_no_print:
836 	bch2_free_super(sb);
837 	goto out;
838 }
839 
840 int bch2_read_super(const char *path, struct bch_opts *opts,
841 		    struct bch_sb_handle *sb)
842 {
843 	return __bch2_read_super(path, opts, sb, false);
844 }
845 
846 /* provide a silenced version for mount.bcachefs */
847 
848 int bch2_read_super_silent(const char *path, struct bch_opts *opts,
849 		    struct bch_sb_handle *sb)
850 {
851 	return __bch2_read_super(path, opts, sb, true);
852 }
853 
854 /* write superblock: */
855 
856 static void write_super_endio(struct bio *bio)
857 {
858 	struct bch_dev *ca = bio->bi_private;
859 
860 	/* XXX: return errors directly */
861 
862 	if (bch2_dev_io_err_on(bio->bi_status, ca,
863 			       bio_data_dir(bio)
864 			       ? BCH_MEMBER_ERROR_write
865 			       : BCH_MEMBER_ERROR_read,
866 			       "superblock %s error: %s",
867 			       bio_data_dir(bio) ? "write" : "read",
868 			       bch2_blk_status_to_str(bio->bi_status)))
869 		ca->sb_write_error = 1;
870 
871 	closure_put(&ca->fs->sb_write);
872 	percpu_ref_put(&ca->io_ref);
873 }
874 
875 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
876 {
877 	struct bch_sb *sb = ca->disk_sb.sb;
878 	struct bio *bio = ca->disk_sb.bio;
879 
880 	bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
881 	bio->bi_iter.bi_sector	= le64_to_cpu(sb->layout.sb_offset[0]);
882 	bio->bi_end_io		= write_super_endio;
883 	bio->bi_private		= ca;
884 	bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
885 
886 	this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
887 		     bio_sectors(bio));
888 
889 	percpu_ref_get(&ca->io_ref);
890 	closure_bio_submit(bio, &c->sb_write);
891 }
892 
893 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
894 {
895 	struct bch_sb *sb = ca->disk_sb.sb;
896 	struct bio *bio = ca->disk_sb.bio;
897 
898 	sb->offset = sb->layout.sb_offset[idx];
899 
900 	SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
901 	sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
902 				null_nonce(), sb);
903 
904 	bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
905 	bio->bi_iter.bi_sector	= le64_to_cpu(sb->offset);
906 	bio->bi_end_io		= write_super_endio;
907 	bio->bi_private		= ca;
908 	bch2_bio_map(bio, sb,
909 		     roundup((size_t) vstruct_bytes(sb),
910 			     bdev_logical_block_size(ca->disk_sb.bdev)));
911 
912 	this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
913 		     bio_sectors(bio));
914 
915 	percpu_ref_get(&ca->io_ref);
916 	closure_bio_submit(bio, &c->sb_write);
917 }
918 
919 int bch2_write_super(struct bch_fs *c)
920 {
921 	struct closure *cl = &c->sb_write;
922 	struct printbuf err = PRINTBUF;
923 	unsigned sb = 0, nr_wrote;
924 	struct bch_devs_mask sb_written;
925 	bool wrote, can_mount_without_written, can_mount_with_written;
926 	unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
927 	DARRAY(struct bch_dev *) online_devices = {};
928 	int ret = 0;
929 
930 	trace_and_count(c, write_super, c, _RET_IP_);
931 
932 	if (c->opts.very_degraded)
933 		degraded_flags |= BCH_FORCE_IF_LOST;
934 
935 	lockdep_assert_held(&c->sb_lock);
936 
937 	closure_init_stack(cl);
938 	memset(&sb_written, 0, sizeof(sb_written));
939 
940 	for_each_online_member(c, ca) {
941 		ret = darray_push(&online_devices, ca);
942 		if (bch2_fs_fatal_err_on(ret, c, "%s: error allocating online devices", __func__)) {
943 			percpu_ref_put(&ca->io_ref);
944 			goto out;
945 		}
946 		percpu_ref_get(&ca->io_ref);
947 	}
948 
949 	/* Make sure we're using the new magic numbers: */
950 	c->disk_sb.sb->magic = BCHFS_MAGIC;
951 	c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
952 
953 	le64_add_cpu(&c->disk_sb.sb->seq, 1);
954 
955 	struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
956 	darray_for_each(online_devices, ca)
957 		__bch2_members_v2_get_mut(mi, (*ca)->dev_idx)->seq = c->disk_sb.sb->seq;
958 	c->disk_sb.sb->write_time = cpu_to_le64(ktime_get_real_seconds());
959 
960 	if (test_bit(BCH_FS_error, &c->flags))
961 		SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
962 	if (test_bit(BCH_FS_topology_error, &c->flags))
963 		SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
964 
965 	SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
966 
967 	bch2_sb_counters_from_cpu(c);
968 	bch2_sb_members_from_cpu(c);
969 	bch2_sb_members_cpy_v2_v1(&c->disk_sb);
970 	bch2_sb_errors_from_cpu(c);
971 	bch2_sb_downgrade_update(c);
972 
973 	darray_for_each(online_devices, ca)
974 		bch2_sb_from_fs(c, (*ca));
975 
976 	darray_for_each(online_devices, ca) {
977 		printbuf_reset(&err);
978 
979 		ret = bch2_sb_validate(&(*ca)->disk_sb, BCH_VALIDATE_write, &err);
980 		if (ret) {
981 			bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
982 			goto out;
983 		}
984 	}
985 
986 	if (c->opts.nochanges)
987 		goto out;
988 
989 	/*
990 	 * Defer writing the superblock until filesystem initialization is
991 	 * complete - don't write out a partly initialized superblock:
992 	 */
993 	if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
994 		goto out;
995 
996 	if (le16_to_cpu(c->disk_sb.sb->version) > bcachefs_metadata_version_current) {
997 		struct printbuf buf = PRINTBUF;
998 		prt_printf(&buf, "attempting to write superblock that wasn't version downgraded (");
999 		bch2_version_to_text(&buf, le16_to_cpu(c->disk_sb.sb->version));
1000 		prt_str(&buf, " > ");
1001 		bch2_version_to_text(&buf, bcachefs_metadata_version_current);
1002 		prt_str(&buf, ")");
1003 		bch2_fs_fatal_error(c, ": %s", buf.buf);
1004 		printbuf_exit(&buf);
1005 		return -BCH_ERR_sb_not_downgraded;
1006 	}
1007 
1008 	darray_for_each(online_devices, ca) {
1009 		__set_bit((*ca)->dev_idx, sb_written.d);
1010 		(*ca)->sb_write_error = 0;
1011 	}
1012 
1013 	darray_for_each(online_devices, ca)
1014 		read_back_super(c, *ca);
1015 	closure_sync(cl);
1016 
1017 	darray_for_each(online_devices, cap) {
1018 		struct bch_dev *ca = *cap;
1019 
1020 		if (ca->sb_write_error)
1021 			continue;
1022 
1023 		if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
1024 			struct printbuf buf = PRINTBUF;
1025 			prt_char(&buf, ' ');
1026 			prt_bdevname(&buf, ca->disk_sb.bdev);
1027 			prt_printf(&buf,
1028 				": Superblock write was silently dropped! (seq %llu expected %llu)",
1029 				le64_to_cpu(ca->sb_read_scratch->seq),
1030 				ca->disk_sb.seq);
1031 			bch2_fs_fatal_error(c, "%s", buf.buf);
1032 			printbuf_exit(&buf);
1033 			ret = -BCH_ERR_erofs_sb_err;
1034 		}
1035 
1036 		if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
1037 			struct printbuf buf = PRINTBUF;
1038 			prt_char(&buf, ' ');
1039 			prt_bdevname(&buf, ca->disk_sb.bdev);
1040 			prt_printf(&buf,
1041 				": Superblock modified by another process (seq %llu expected %llu)",
1042 				le64_to_cpu(ca->sb_read_scratch->seq),
1043 				ca->disk_sb.seq);
1044 			bch2_fs_fatal_error(c, "%s", buf.buf);
1045 			printbuf_exit(&buf);
1046 			ret = -BCH_ERR_erofs_sb_err;
1047 		}
1048 	}
1049 
1050 	if (ret)
1051 		goto out;
1052 
1053 	do {
1054 		wrote = false;
1055 		darray_for_each(online_devices, cap) {
1056 			struct bch_dev *ca = *cap;
1057 			if (!ca->sb_write_error &&
1058 			    sb < ca->disk_sb.sb->layout.nr_superblocks) {
1059 				write_one_super(c, ca, sb);
1060 				wrote = true;
1061 			}
1062 		}
1063 		closure_sync(cl);
1064 		sb++;
1065 	} while (wrote);
1066 
1067 	darray_for_each(online_devices, cap) {
1068 		struct bch_dev *ca = *cap;
1069 		if (ca->sb_write_error)
1070 			__clear_bit(ca->dev_idx, sb_written.d);
1071 		else
1072 			ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
1073 	}
1074 
1075 	nr_wrote = dev_mask_nr(&sb_written);
1076 
1077 	can_mount_with_written =
1078 		bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1079 
1080 	for (unsigned i = 0; i < ARRAY_SIZE(sb_written.d); i++)
1081 		sb_written.d[i] = ~sb_written.d[i];
1082 
1083 	can_mount_without_written =
1084 		bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1085 
1086 	/*
1087 	 * If we would be able to mount _without_ the devices we successfully
1088 	 * wrote superblocks to, we weren't able to write to enough devices:
1089 	 *
1090 	 * Exception: if we can mount without the successes because we haven't
1091 	 * written anything (new filesystem), we continue if we'd be able to
1092 	 * mount with the devices we did successfully write to:
1093 	 */
1094 	if (bch2_fs_fatal_err_on(!nr_wrote ||
1095 				 !can_mount_with_written ||
1096 				 (can_mount_without_written &&
1097 				  !can_mount_with_written), c,
1098 		": Unable to write superblock to sufficient devices (from %ps)",
1099 		(void *) _RET_IP_))
1100 		ret = -1;
1101 out:
1102 	/* Make new options visible after they're persistent: */
1103 	bch2_sb_update(c);
1104 	darray_for_each(online_devices, ca)
1105 		percpu_ref_put(&(*ca)->io_ref);
1106 	darray_exit(&online_devices);
1107 	printbuf_exit(&err);
1108 	return ret;
1109 }
1110 
1111 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
1112 {
1113 	mutex_lock(&c->sb_lock);
1114 	if (!(c->sb.features & (1ULL << feat))) {
1115 		c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
1116 
1117 		bch2_write_super(c);
1118 	}
1119 	mutex_unlock(&c->sb_lock);
1120 }
1121 
1122 /* Downgrade if superblock is at a higher version than currently supported: */
1123 bool bch2_check_version_downgrade(struct bch_fs *c)
1124 {
1125 	bool ret = bcachefs_metadata_version_current < c->sb.version;
1126 
1127 	lockdep_assert_held(&c->sb_lock);
1128 
1129 	/*
1130 	 * Downgrade, if superblock is at a higher version than currently
1131 	 * supported:
1132 	 *
1133 	 * c->sb will be checked before we write the superblock, so update it as
1134 	 * well:
1135 	 */
1136 	if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current)
1137 		SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1138 	if (c->sb.version > bcachefs_metadata_version_current)
1139 		c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1140 	if (c->sb.version_min > bcachefs_metadata_version_current)
1141 		c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
1142 	c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1143 	return ret;
1144 }
1145 
1146 void bch2_sb_upgrade(struct bch_fs *c, unsigned new_version)
1147 {
1148 	lockdep_assert_held(&c->sb_lock);
1149 
1150 	if (BCH_VERSION_MAJOR(new_version) >
1151 	    BCH_VERSION_MAJOR(le16_to_cpu(c->disk_sb.sb->version)))
1152 		bch2_sb_field_resize(&c->disk_sb, downgrade, 0);
1153 
1154 	c->disk_sb.sb->version = cpu_to_le16(new_version);
1155 	c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1156 }
1157 
1158 static int bch2_sb_ext_validate(struct bch_sb *sb, struct bch_sb_field *f,
1159 				enum bch_validate_flags flags, struct printbuf *err)
1160 {
1161 	if (vstruct_bytes(f) < 88) {
1162 		prt_printf(err, "field too small (%zu < %u)", vstruct_bytes(f), 88);
1163 		return -BCH_ERR_invalid_sb_ext;
1164 	}
1165 
1166 	return 0;
1167 }
1168 
1169 static void bch2_sb_ext_to_text(struct printbuf *out, struct bch_sb *sb,
1170 				struct bch_sb_field *f)
1171 {
1172 	struct bch_sb_field_ext *e = field_to_type(f, ext);
1173 
1174 	prt_printf(out, "Recovery passes required:\t");
1175 	prt_bitflags(out, bch2_recovery_passes,
1176 		     bch2_recovery_passes_from_stable(le64_to_cpu(e->recovery_passes_required[0])));
1177 	prt_newline(out);
1178 
1179 	unsigned long *errors_silent = kmalloc(sizeof(e->errors_silent), GFP_KERNEL);
1180 	if (errors_silent) {
1181 		le_bitvector_to_cpu(errors_silent, (void *) e->errors_silent, sizeof(e->errors_silent) * 8);
1182 
1183 		prt_printf(out, "Errors to silently fix:\t");
1184 		prt_bitflags_vector(out, bch2_sb_error_strs, errors_silent, sizeof(e->errors_silent) * 8);
1185 		prt_newline(out);
1186 
1187 		kfree(errors_silent);
1188 	}
1189 
1190 	prt_printf(out, "Btrees with missing data:\t");
1191 	prt_bitflags(out, __bch2_btree_ids, le64_to_cpu(e->btrees_lost_data));
1192 	prt_newline(out);
1193 }
1194 
1195 static const struct bch_sb_field_ops bch_sb_field_ops_ext = {
1196 	.validate	= bch2_sb_ext_validate,
1197 	.to_text	= bch2_sb_ext_to_text,
1198 };
1199 
1200 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1201 #define x(f, nr)					\
1202 	[BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1203 	BCH_SB_FIELDS()
1204 #undef x
1205 };
1206 
1207 static const struct bch_sb_field_ops bch2_sb_field_null_ops;
1208 
1209 static const struct bch_sb_field_ops *bch2_sb_field_type_ops(unsigned type)
1210 {
1211 	return likely(type < ARRAY_SIZE(bch2_sb_field_ops))
1212 		? bch2_sb_field_ops[type]
1213 		: &bch2_sb_field_null_ops;
1214 }
1215 
1216 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1217 				  enum bch_validate_flags flags, struct printbuf *err)
1218 {
1219 	unsigned type = le32_to_cpu(f->type);
1220 	struct printbuf field_err = PRINTBUF;
1221 	const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1222 	int ret;
1223 
1224 	ret = ops->validate ? ops->validate(sb, f, flags, &field_err) : 0;
1225 	if (ret) {
1226 		prt_printf(err, "Invalid superblock section %s: %s",
1227 			   bch2_sb_fields[type], field_err.buf);
1228 		prt_newline(err);
1229 		bch2_sb_field_to_text(err, sb, f);
1230 	}
1231 
1232 	printbuf_exit(&field_err);
1233 	return ret;
1234 }
1235 
1236 void __bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1237 			     struct bch_sb_field *f)
1238 {
1239 	unsigned type = le32_to_cpu(f->type);
1240 	const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1241 
1242 	if (!out->nr_tabstops)
1243 		printbuf_tabstop_push(out, 32);
1244 
1245 	if (ops->to_text)
1246 		ops->to_text(out, sb, f);
1247 }
1248 
1249 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1250 			   struct bch_sb_field *f)
1251 {
1252 	unsigned type = le32_to_cpu(f->type);
1253 
1254 	if (type < BCH_SB_FIELD_NR)
1255 		prt_printf(out, "%s", bch2_sb_fields[type]);
1256 	else
1257 		prt_printf(out, "(unknown field %u)", type);
1258 
1259 	prt_printf(out, " (size %zu):", vstruct_bytes(f));
1260 	prt_newline(out);
1261 
1262 	__bch2_sb_field_to_text(out, sb, f);
1263 }
1264 
1265 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1266 {
1267 	unsigned i;
1268 
1269 	prt_printf(out, "Type:                    %u", l->layout_type);
1270 	prt_newline(out);
1271 
1272 	prt_str(out, "Superblock max size:     ");
1273 	prt_units_u64(out, 512 << l->sb_max_size_bits);
1274 	prt_newline(out);
1275 
1276 	prt_printf(out, "Nr superblocks:          %u", l->nr_superblocks);
1277 	prt_newline(out);
1278 
1279 	prt_str(out, "Offsets:                 ");
1280 	for (i = 0; i < l->nr_superblocks; i++) {
1281 		if (i)
1282 			prt_str(out, ", ");
1283 		prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1284 	}
1285 	prt_newline(out);
1286 }
1287 
1288 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1289 		     bool print_layout, unsigned fields)
1290 {
1291 	u64 fields_have = 0;
1292 	unsigned nr_devices = 0;
1293 
1294 	if (!out->nr_tabstops)
1295 		printbuf_tabstop_push(out, 44);
1296 
1297 	for (int i = 0; i < sb->nr_devices; i++)
1298 		nr_devices += bch2_member_exists(sb, i);
1299 
1300 	prt_printf(out, "External UUID:\t");
1301 	pr_uuid(out, sb->user_uuid.b);
1302 	prt_newline(out);
1303 
1304 	prt_printf(out, "Internal UUID:\t");
1305 	pr_uuid(out, sb->uuid.b);
1306 	prt_newline(out);
1307 
1308 	prt_printf(out, "Magic number:\t");
1309 	pr_uuid(out, sb->magic.b);
1310 	prt_newline(out);
1311 
1312 	prt_printf(out, "Device index:\t%u\n", sb->dev_idx);
1313 
1314 	prt_printf(out, "Label:\t");
1315 	prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1316 	prt_newline(out);
1317 
1318 	prt_printf(out, "Version:\t");
1319 	bch2_version_to_text(out, le16_to_cpu(sb->version));
1320 	prt_newline(out);
1321 
1322 	prt_printf(out, "Version upgrade complete:\t");
1323 	bch2_version_to_text(out, BCH_SB_VERSION_UPGRADE_COMPLETE(sb));
1324 	prt_newline(out);
1325 
1326 	prt_printf(out, "Oldest version on disk:\t");
1327 	bch2_version_to_text(out, le16_to_cpu(sb->version_min));
1328 	prt_newline(out);
1329 
1330 	prt_printf(out, "Created:\t");
1331 	if (sb->time_base_lo)
1332 		bch2_prt_datetime(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1333 	else
1334 		prt_printf(out, "(not set)");
1335 	prt_newline(out);
1336 
1337 	prt_printf(out, "Sequence number:\t");
1338 	prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1339 	prt_newline(out);
1340 
1341 	prt_printf(out, "Time of last write:\t");
1342 	bch2_prt_datetime(out, le64_to_cpu(sb->write_time));
1343 	prt_newline(out);
1344 
1345 	prt_printf(out, "Superblock size:\t");
1346 	prt_units_u64(out, vstruct_bytes(sb));
1347 	prt_str(out, "/");
1348 	prt_units_u64(out, 512ULL << sb->layout.sb_max_size_bits);
1349 	prt_newline(out);
1350 
1351 	prt_printf(out, "Clean:\t%llu\n", BCH_SB_CLEAN(sb));
1352 	prt_printf(out, "Devices:\t%u\n", nr_devices);
1353 
1354 	prt_printf(out, "Sections:\t");
1355 	vstruct_for_each(sb, f)
1356 		fields_have |= 1 << le32_to_cpu(f->type);
1357 	prt_bitflags(out, bch2_sb_fields, fields_have);
1358 	prt_newline(out);
1359 
1360 	prt_printf(out, "Features:\t");
1361 	prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1362 	prt_newline(out);
1363 
1364 	prt_printf(out, "Compat features:\t");
1365 	prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1366 	prt_newline(out);
1367 
1368 	prt_newline(out);
1369 	prt_printf(out, "Options:");
1370 	prt_newline(out);
1371 	printbuf_indent_add(out, 2);
1372 	{
1373 		enum bch_opt_id id;
1374 
1375 		for (id = 0; id < bch2_opts_nr; id++) {
1376 			const struct bch_option *opt = bch2_opt_table + id;
1377 
1378 			if (opt->get_sb != BCH2_NO_SB_OPT) {
1379 				u64 v = bch2_opt_from_sb(sb, id);
1380 
1381 				prt_printf(out, "%s:\t", opt->attr.name);
1382 				bch2_opt_to_text(out, NULL, sb, opt, v,
1383 						 OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1384 				prt_newline(out);
1385 			}
1386 		}
1387 	}
1388 
1389 	printbuf_indent_sub(out, 2);
1390 
1391 	if (print_layout) {
1392 		prt_newline(out);
1393 		prt_printf(out, "layout:");
1394 		prt_newline(out);
1395 		printbuf_indent_add(out, 2);
1396 		bch2_sb_layout_to_text(out, &sb->layout);
1397 		printbuf_indent_sub(out, 2);
1398 	}
1399 
1400 	vstruct_for_each(sb, f)
1401 		if (fields & (1 << le32_to_cpu(f->type))) {
1402 			prt_newline(out);
1403 			bch2_sb_field_to_text(out, sb, f);
1404 		}
1405 }
1406