xref: /linux/fs/bcachefs/sb-downgrade.c (revision e814f3fd16acfb7f9966773953de8f740a1e3202)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 /*
4  * Superblock section that contains a list of recovery passes to run when
5  * downgrading past a given version
6  */
7 
8 #include "bcachefs.h"
9 #include "darray.h"
10 #include "recovery_passes.h"
11 #include "sb-downgrade.h"
12 #include "sb-errors.h"
13 #include "super-io.h"
14 
15 #define RECOVERY_PASS_ALL_FSCK		BIT_ULL(63)
16 
17 /*
18  * Upgrade, downgrade tables - run certain recovery passes, fix certain errors
19  *
20  * x(version, recovery_passes, errors...)
21  */
22 #define UPGRADE_TABLE()						\
23 	x(backpointers,						\
24 	  RECOVERY_PASS_ALL_FSCK)				\
25 	x(inode_v3,						\
26 	  RECOVERY_PASS_ALL_FSCK)				\
27 	x(unwritten_extents,					\
28 	  RECOVERY_PASS_ALL_FSCK)				\
29 	x(bucket_gens,						\
30 	  BIT_ULL(BCH_RECOVERY_PASS_bucket_gens_init)|		\
31 	  RECOVERY_PASS_ALL_FSCK)				\
32 	x(lru_v2,						\
33 	  RECOVERY_PASS_ALL_FSCK)				\
34 	x(fragmentation_lru,					\
35 	  RECOVERY_PASS_ALL_FSCK)				\
36 	x(no_bps_in_alloc_keys,					\
37 	  RECOVERY_PASS_ALL_FSCK)				\
38 	x(snapshot_trees,					\
39 	  RECOVERY_PASS_ALL_FSCK)				\
40 	x(snapshot_skiplists,					\
41 	  BIT_ULL(BCH_RECOVERY_PASS_check_snapshots),		\
42 	  BCH_FSCK_ERR_snapshot_bad_depth,			\
43 	  BCH_FSCK_ERR_snapshot_bad_skiplist)			\
44 	x(deleted_inodes,					\
45 	  BIT_ULL(BCH_RECOVERY_PASS_check_inodes),		\
46 	  BCH_FSCK_ERR_unlinked_inode_not_on_deleted_list)	\
47 	x(rebalance_work,					\
48 	  BIT_ULL(BCH_RECOVERY_PASS_set_fs_needs_rebalance))	\
49 	x(subvolume_fs_parent,					\
50 	  BIT_ULL(BCH_RECOVERY_PASS_check_dirents),		\
51 	  BCH_FSCK_ERR_subvol_fs_path_parent_wrong)		\
52 	x(btree_subvolume_children,				\
53 	  BIT_ULL(BCH_RECOVERY_PASS_check_subvols),		\
54 	  BCH_FSCK_ERR_subvol_children_not_set)			\
55 	x(mi_btree_bitmap,					\
56 	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
57 	  BCH_FSCK_ERR_btree_bitmap_not_marked)			\
58 	x(disk_accounting_v2,					\
59 	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
60 	  BCH_FSCK_ERR_bkey_version_in_future,			\
61 	  BCH_FSCK_ERR_dev_usage_buckets_wrong,			\
62 	  BCH_FSCK_ERR_dev_usage_sectors_wrong,			\
63 	  BCH_FSCK_ERR_dev_usage_fragmented_wrong,		\
64 	  BCH_FSCK_ERR_accounting_mismatch)			\
65 	x(disk_accounting_v3,					\
66 	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
67 	  BCH_FSCK_ERR_bkey_version_in_future,			\
68 	  BCH_FSCK_ERR_dev_usage_buckets_wrong,			\
69 	  BCH_FSCK_ERR_dev_usage_sectors_wrong,			\
70 	  BCH_FSCK_ERR_dev_usage_fragmented_wrong,		\
71 	  BCH_FSCK_ERR_accounting_mismatch,			\
72 	  BCH_FSCK_ERR_accounting_key_replicas_nr_devs_0,	\
73 	  BCH_FSCK_ERR_accounting_key_replicas_nr_required_bad,	\
74 	  BCH_FSCK_ERR_accounting_key_replicas_devs_unsorted,	\
75 	  BCH_FSCK_ERR_accounting_key_junk_at_end)		\
76 	x(disk_accounting_inum,					\
77 	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
78 	  BCH_FSCK_ERR_accounting_mismatch)			\
79 	x(rebalance_work_acct_fix,				\
80 	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
81 	  BCH_FSCK_ERR_accounting_mismatch)			\
82 	x(inode_has_child_snapshots,				\
83 	  BIT_ULL(BCH_RECOVERY_PASS_check_inodes),		\
84 	  BCH_FSCK_ERR_inode_has_child_snapshots_wrong)		\
85 	x(backpointer_bucket_gen,				\
86 	  BIT_ULL(BCH_RECOVERY_PASS_check_extents_to_backpointers),\
87 	  BCH_FSCK_ERR_backpointer_to_missing_ptr,		\
88 	  BCH_FSCK_ERR_ptr_to_missing_backpointer)		\
89 	x(disk_accounting_big_endian,				\
90 	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
91 	  BCH_FSCK_ERR_accounting_mismatch,			\
92 	  BCH_FSCK_ERR_accounting_key_replicas_nr_devs_0,	\
93 	  BCH_FSCK_ERR_accounting_key_junk_at_end)		\
94 	x(directory_size,					\
95 	  BIT_ULL(BCH_RECOVERY_PASS_check_inodes),		\
96 	  BCH_FSCK_ERR_directory_size_mismatch)			\
97 
98 #define DOWNGRADE_TABLE()					\
99 	x(bucket_stripe_sectors,				\
100 	  0)							\
101 	x(disk_accounting_v2,					\
102 	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
103 	  BCH_FSCK_ERR_dev_usage_buckets_wrong,			\
104 	  BCH_FSCK_ERR_dev_usage_sectors_wrong,			\
105 	  BCH_FSCK_ERR_dev_usage_fragmented_wrong,		\
106 	  BCH_FSCK_ERR_fs_usage_hidden_wrong,			\
107 	  BCH_FSCK_ERR_fs_usage_btree_wrong,			\
108 	  BCH_FSCK_ERR_fs_usage_data_wrong,			\
109 	  BCH_FSCK_ERR_fs_usage_cached_wrong,			\
110 	  BCH_FSCK_ERR_fs_usage_reserved_wrong,			\
111 	  BCH_FSCK_ERR_fs_usage_nr_inodes_wrong,		\
112 	  BCH_FSCK_ERR_fs_usage_persistent_reserved_wrong,	\
113 	  BCH_FSCK_ERR_fs_usage_replicas_wrong,			\
114 	  BCH_FSCK_ERR_bkey_version_in_future)			\
115 	x(disk_accounting_v3,					\
116 	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
117 	  BCH_FSCK_ERR_dev_usage_buckets_wrong,			\
118 	  BCH_FSCK_ERR_dev_usage_sectors_wrong,			\
119 	  BCH_FSCK_ERR_dev_usage_fragmented_wrong,		\
120 	  BCH_FSCK_ERR_fs_usage_hidden_wrong,			\
121 	  BCH_FSCK_ERR_fs_usage_btree_wrong,			\
122 	  BCH_FSCK_ERR_fs_usage_data_wrong,			\
123 	  BCH_FSCK_ERR_fs_usage_cached_wrong,			\
124 	  BCH_FSCK_ERR_fs_usage_reserved_wrong,			\
125 	  BCH_FSCK_ERR_fs_usage_nr_inodes_wrong,		\
126 	  BCH_FSCK_ERR_fs_usage_persistent_reserved_wrong,	\
127 	  BCH_FSCK_ERR_fs_usage_replicas_wrong,			\
128 	  BCH_FSCK_ERR_accounting_replicas_not_marked,		\
129 	  BCH_FSCK_ERR_bkey_version_in_future)			\
130 	x(rebalance_work_acct_fix,				\
131 	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
132 	  BCH_FSCK_ERR_accounting_mismatch,			\
133 	  BCH_FSCK_ERR_accounting_key_replicas_nr_devs_0,	\
134 	  BCH_FSCK_ERR_accounting_key_junk_at_end)		\
135 	x(backpointer_bucket_gen,				\
136 	  BIT_ULL(BCH_RECOVERY_PASS_check_extents_to_backpointers),\
137 	  BCH_FSCK_ERR_backpointer_bucket_offset_wrong,		\
138 	  BCH_FSCK_ERR_backpointer_to_missing_ptr,		\
139 	  BCH_FSCK_ERR_ptr_to_missing_backpointer)		\
140 	x(disk_accounting_big_endian,				\
141 	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
142 	  BCH_FSCK_ERR_accounting_mismatch,			\
143 	  BCH_FSCK_ERR_accounting_key_replicas_nr_devs_0,	\
144 	  BCH_FSCK_ERR_accounting_key_junk_at_end)
145 
146 struct upgrade_downgrade_entry {
147 	u64		recovery_passes;
148 	u16		version;
149 	u16		nr_errors;
150 	const u16	*errors;
151 };
152 
153 #define x(ver, passes, ...) static const u16 upgrade_##ver##_errors[] = { __VA_ARGS__ };
154 UPGRADE_TABLE()
155 #undef x
156 
157 static const struct upgrade_downgrade_entry upgrade_table[] = {
158 #define x(ver, passes, ...) {					\
159 	.recovery_passes	= passes,			\
160 	.version		= bcachefs_metadata_version_##ver,\
161 	.nr_errors		= ARRAY_SIZE(upgrade_##ver##_errors),	\
162 	.errors			= upgrade_##ver##_errors,	\
163 },
164 UPGRADE_TABLE()
165 #undef x
166 };
167 
168 static int have_stripes(struct bch_fs *c)
169 {
170 	if (IS_ERR_OR_NULL(c->btree_roots_known[BTREE_ID_stripes].b))
171 		return 0;
172 
173 	return !btree_node_fake(c->btree_roots_known[BTREE_ID_stripes].b);
174 }
175 
176 int bch2_sb_set_upgrade_extra(struct bch_fs *c)
177 {
178 	unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version;
179 	unsigned new_version = c->sb.version;
180 	bool write_sb = false;
181 	int ret = 0;
182 
183 	mutex_lock(&c->sb_lock);
184 	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
185 
186 	if (old_version <  bcachefs_metadata_version_bucket_stripe_sectors &&
187 	    new_version >= bcachefs_metadata_version_bucket_stripe_sectors &&
188 	    (ret = have_stripes(c) > 0)) {
189 		__set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_allocations, ext->recovery_passes_required);
190 		__set_bit_le64(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong, ext->errors_silent);
191 		__set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_sectors_wrong, ext->errors_silent);
192 		write_sb = true;
193 	}
194 
195 	if (write_sb)
196 		bch2_write_super(c);
197 	mutex_unlock(&c->sb_lock);
198 
199 	return ret < 0 ? ret : 0;
200 }
201 
202 void bch2_sb_set_upgrade(struct bch_fs *c,
203 			 unsigned old_version,
204 			 unsigned new_version)
205 {
206 	lockdep_assert_held(&c->sb_lock);
207 
208 	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
209 
210 	for (const struct upgrade_downgrade_entry *i = upgrade_table;
211 	     i < upgrade_table + ARRAY_SIZE(upgrade_table);
212 	     i++)
213 		if (i->version > old_version && i->version <= new_version) {
214 			u64 passes = i->recovery_passes;
215 
216 			if (passes & RECOVERY_PASS_ALL_FSCK)
217 				passes |= bch2_fsck_recovery_passes();
218 			passes &= ~RECOVERY_PASS_ALL_FSCK;
219 
220 			ext->recovery_passes_required[0] |=
221 				cpu_to_le64(bch2_recovery_passes_to_stable(passes));
222 
223 			for (const u16 *e = i->errors; e < i->errors + i->nr_errors; e++)
224 				__set_bit_le64(*e, ext->errors_silent);
225 		}
226 }
227 
228 #define x(ver, passes, ...) static const u16 downgrade_##ver##_errors[] = { __VA_ARGS__ };
229 DOWNGRADE_TABLE()
230 #undef x
231 
232 static const struct upgrade_downgrade_entry downgrade_table[] = {
233 #define x(ver, passes, ...) {					\
234 	.recovery_passes	= passes,			\
235 	.version		= bcachefs_metadata_version_##ver,\
236 	.nr_errors		= ARRAY_SIZE(downgrade_##ver##_errors),	\
237 	.errors			= downgrade_##ver##_errors,	\
238 },
239 DOWNGRADE_TABLE()
240 #undef x
241 };
242 
243 static int downgrade_table_extra(struct bch_fs *c, darray_char *table)
244 {
245 	struct bch_sb_field_downgrade_entry *dst = (void *) &darray_top(*table);
246 	unsigned bytes = sizeof(*dst) + sizeof(dst->errors[0]) * le16_to_cpu(dst->nr_errors);
247 	int ret = 0;
248 
249 	unsigned nr_errors = le16_to_cpu(dst->nr_errors);
250 
251 	switch (le16_to_cpu(dst->version)) {
252 	case bcachefs_metadata_version_bucket_stripe_sectors:
253 		if (have_stripes(c)) {
254 			bytes += sizeof(dst->errors[0]) * 2;
255 
256 			ret = darray_make_room(table, bytes);
257 			if (ret)
258 				return ret;
259 
260 			/* open coded __set_bit_le64, as dst is packed and
261 			 * dst->recovery_passes is misaligned */
262 			unsigned b = BCH_RECOVERY_PASS_STABLE_check_allocations;
263 			dst->recovery_passes[b / 64] |= cpu_to_le64(BIT_ULL(b % 64));
264 
265 			dst->errors[nr_errors++] = cpu_to_le16(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong);
266 		}
267 		break;
268 	}
269 
270 	dst->nr_errors = cpu_to_le16(nr_errors);
271 	return ret;
272 }
273 
274 static inline const struct bch_sb_field_downgrade_entry *
275 downgrade_entry_next_c(const struct bch_sb_field_downgrade_entry *e)
276 {
277 	return (void *) &e->errors[le16_to_cpu(e->nr_errors)];
278 }
279 
280 #define for_each_downgrade_entry(_d, _i)						\
281 	for (const struct bch_sb_field_downgrade_entry *_i = (_d)->entries;		\
282 	     (void *) _i	< vstruct_end(&(_d)->field) &&				\
283 	     (void *) &_i->errors[0] <= vstruct_end(&(_d)->field) &&			\
284 	     (void *) downgrade_entry_next_c(_i) <= vstruct_end(&(_d)->field);		\
285 	     _i = downgrade_entry_next_c(_i))
286 
287 static int bch2_sb_downgrade_validate(struct bch_sb *sb, struct bch_sb_field *f,
288 				      enum bch_validate_flags flags, struct printbuf *err)
289 {
290 	struct bch_sb_field_downgrade *e = field_to_type(f, downgrade);
291 
292 	for (const struct bch_sb_field_downgrade_entry *i = e->entries;
293 	     (void *) i	< vstruct_end(&e->field);
294 	     i = downgrade_entry_next_c(i)) {
295 		/*
296 		 * Careful: sb_field_downgrade_entry is only 2 byte aligned, but
297 		 * section sizes are 8 byte aligned - an empty entry spanning
298 		 * the end of the section is allowed (and ignored):
299 		 */
300 		if ((void *) &i->errors[0] > vstruct_end(&e->field))
301 			break;
302 
303 		if (flags & BCH_VALIDATE_write &&
304 		    (void *) downgrade_entry_next_c(i) > vstruct_end(&e->field)) {
305 			prt_printf(err, "downgrade entry overruns end of superblock section");
306 			return -BCH_ERR_invalid_sb_downgrade;
307 		}
308 
309 		if (BCH_VERSION_MAJOR(le16_to_cpu(i->version)) !=
310 		    BCH_VERSION_MAJOR(le16_to_cpu(sb->version))) {
311 			prt_printf(err, "downgrade entry with mismatched major version (%u != %u)",
312 				   BCH_VERSION_MAJOR(le16_to_cpu(i->version)),
313 				   BCH_VERSION_MAJOR(le16_to_cpu(sb->version)));
314 			return -BCH_ERR_invalid_sb_downgrade;
315 		}
316 	}
317 
318 	return 0;
319 }
320 
321 static void bch2_sb_downgrade_to_text(struct printbuf *out, struct bch_sb *sb,
322 				      struct bch_sb_field *f)
323 {
324 	struct bch_sb_field_downgrade *e = field_to_type(f, downgrade);
325 
326 	if (out->nr_tabstops <= 1)
327 		printbuf_tabstop_push(out, 16);
328 
329 	for_each_downgrade_entry(e, i) {
330 		prt_str(out, "version:\t");
331 		bch2_version_to_text(out, le16_to_cpu(i->version));
332 		prt_newline(out);
333 
334 		prt_str(out, "recovery passes:\t");
335 		prt_bitflags(out, bch2_recovery_passes,
336 			     bch2_recovery_passes_from_stable(le64_to_cpu(i->recovery_passes[0])));
337 		prt_newline(out);
338 
339 		prt_str(out, "errors:\t");
340 		bool first = true;
341 		for (unsigned j = 0; j < le16_to_cpu(i->nr_errors); j++) {
342 			if (!first)
343 				prt_char(out, ',');
344 			first = false;
345 			bch2_sb_error_id_to_text(out, le16_to_cpu(i->errors[j]));
346 		}
347 		prt_newline(out);
348 	}
349 }
350 
351 const struct bch_sb_field_ops bch_sb_field_ops_downgrade = {
352 	.validate	= bch2_sb_downgrade_validate,
353 	.to_text	= bch2_sb_downgrade_to_text,
354 };
355 
356 int bch2_sb_downgrade_update(struct bch_fs *c)
357 {
358 	if (!test_bit(BCH_FS_btree_running, &c->flags))
359 		return 0;
360 
361 	darray_char table = {};
362 	int ret = 0;
363 
364 	for (const struct upgrade_downgrade_entry *src = downgrade_table;
365 	     src < downgrade_table + ARRAY_SIZE(downgrade_table);
366 	     src++) {
367 		if (BCH_VERSION_MAJOR(src->version) != BCH_VERSION_MAJOR(le16_to_cpu(c->disk_sb.sb->version)))
368 			continue;
369 
370 		struct bch_sb_field_downgrade_entry *dst;
371 		unsigned bytes = sizeof(*dst) + sizeof(dst->errors[0]) * src->nr_errors;
372 
373 		ret = darray_make_room(&table, bytes);
374 		if (ret)
375 			goto out;
376 
377 		dst = (void *) &darray_top(table);
378 		dst->version = cpu_to_le16(src->version);
379 		dst->recovery_passes[0]	= cpu_to_le64(bch2_recovery_passes_to_stable(src->recovery_passes));
380 		dst->recovery_passes[1]	= 0;
381 		dst->nr_errors		= cpu_to_le16(src->nr_errors);
382 		for (unsigned i = 0; i < src->nr_errors; i++)
383 			dst->errors[i] = cpu_to_le16(src->errors[i]);
384 
385 		ret = downgrade_table_extra(c, &table);
386 		if (ret)
387 			goto out;
388 
389 		if (!dst->recovery_passes[0] &&
390 		    !dst->recovery_passes[1] &&
391 		    !dst->nr_errors)
392 			continue;
393 
394 		table.nr += sizeof(*dst) + sizeof(dst->errors[0]) * le16_to_cpu(dst->nr_errors);
395 	}
396 
397 	struct bch_sb_field_downgrade *d = bch2_sb_field_get(c->disk_sb.sb, downgrade);
398 
399 	unsigned sb_u64s = DIV_ROUND_UP(sizeof(*d) + table.nr, sizeof(u64));
400 
401 	if (d && le32_to_cpu(d->field.u64s) > sb_u64s)
402 		goto out;
403 
404 	d = bch2_sb_field_resize(&c->disk_sb, downgrade, sb_u64s);
405 	if (!d) {
406 		ret = -BCH_ERR_ENOSPC_sb_downgrade;
407 		goto out;
408 	}
409 
410 	memcpy(d->entries, table.data, table.nr);
411 	memset_u64s_tail(d->entries, 0, table.nr);
412 out:
413 	darray_exit(&table);
414 	return ret;
415 }
416 
417 void bch2_sb_set_downgrade(struct bch_fs *c, unsigned new_minor, unsigned old_minor)
418 {
419 	struct bch_sb_field_downgrade *d = bch2_sb_field_get(c->disk_sb.sb, downgrade);
420 	if (!d)
421 		return;
422 
423 	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
424 
425 	for_each_downgrade_entry(d, i) {
426 		unsigned minor = BCH_VERSION_MINOR(le16_to_cpu(i->version));
427 		if (new_minor < minor && minor <= old_minor) {
428 			ext->recovery_passes_required[0] |= i->recovery_passes[0];
429 			ext->recovery_passes_required[1] |= i->recovery_passes[1];
430 
431 			for (unsigned j = 0; j < le16_to_cpu(i->nr_errors); j++) {
432 				unsigned e = le16_to_cpu(i->errors[j]);
433 				if (e < BCH_FSCK_ERR_MAX)
434 					__set_bit(e, c->sb.errors_silent);
435 				if (e < sizeof(ext->errors_silent) * 8)
436 					__set_bit_le64(e, ext->errors_silent);
437 			}
438 		}
439 	}
440 }
441