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 66 #define DOWNGRADE_TABLE() \ 67 x(bucket_stripe_sectors, \ 68 0) \ 69 x(disk_accounting_v2, \ 70 BIT_ULL(BCH_RECOVERY_PASS_check_allocations), \ 71 BCH_FSCK_ERR_dev_usage_buckets_wrong, \ 72 BCH_FSCK_ERR_dev_usage_sectors_wrong, \ 73 BCH_FSCK_ERR_dev_usage_fragmented_wrong, \ 74 BCH_FSCK_ERR_fs_usage_hidden_wrong, \ 75 BCH_FSCK_ERR_fs_usage_btree_wrong, \ 76 BCH_FSCK_ERR_fs_usage_data_wrong, \ 77 BCH_FSCK_ERR_fs_usage_cached_wrong, \ 78 BCH_FSCK_ERR_fs_usage_reserved_wrong, \ 79 BCH_FSCK_ERR_fs_usage_nr_inodes_wrong, \ 80 BCH_FSCK_ERR_fs_usage_persistent_reserved_wrong, \ 81 BCH_FSCK_ERR_fs_usage_replicas_wrong, \ 82 BCH_FSCK_ERR_bkey_version_in_future) 83 84 struct upgrade_downgrade_entry { 85 u64 recovery_passes; 86 u16 version; 87 u16 nr_errors; 88 const u16 *errors; 89 }; 90 91 #define x(ver, passes, ...) static const u16 upgrade_##ver##_errors[] = { __VA_ARGS__ }; 92 UPGRADE_TABLE() 93 #undef x 94 95 static const struct upgrade_downgrade_entry upgrade_table[] = { 96 #define x(ver, passes, ...) { \ 97 .recovery_passes = passes, \ 98 .version = bcachefs_metadata_version_##ver,\ 99 .nr_errors = ARRAY_SIZE(upgrade_##ver##_errors), \ 100 .errors = upgrade_##ver##_errors, \ 101 }, 102 UPGRADE_TABLE() 103 #undef x 104 }; 105 106 static int have_stripes(struct bch_fs *c) 107 { 108 return !btree_node_fake(c->btree_roots_known[BTREE_ID_stripes].b); 109 } 110 111 int bch2_sb_set_upgrade_extra(struct bch_fs *c) 112 { 113 unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version; 114 unsigned new_version = c->sb.version; 115 bool write_sb = false; 116 int ret = 0; 117 118 mutex_lock(&c->sb_lock); 119 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext); 120 121 if (old_version < bcachefs_metadata_version_bucket_stripe_sectors && 122 new_version >= bcachefs_metadata_version_bucket_stripe_sectors && 123 (ret = have_stripes(c) > 0)) { 124 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_allocations, ext->recovery_passes_required); 125 __set_bit_le64(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong, ext->errors_silent); 126 __set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_sectors_wrong, ext->errors_silent); 127 write_sb = true; 128 } 129 130 if (write_sb) 131 bch2_write_super(c); 132 mutex_unlock(&c->sb_lock); 133 134 return ret < 0 ? ret : 0; 135 } 136 137 void bch2_sb_set_upgrade(struct bch_fs *c, 138 unsigned old_version, 139 unsigned new_version) 140 { 141 lockdep_assert_held(&c->sb_lock); 142 143 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext); 144 145 for (const struct upgrade_downgrade_entry *i = upgrade_table; 146 i < upgrade_table + ARRAY_SIZE(upgrade_table); 147 i++) 148 if (i->version > old_version && i->version <= new_version) { 149 u64 passes = i->recovery_passes; 150 151 if (passes & RECOVERY_PASS_ALL_FSCK) 152 passes |= bch2_fsck_recovery_passes(); 153 passes &= ~RECOVERY_PASS_ALL_FSCK; 154 155 ext->recovery_passes_required[0] |= 156 cpu_to_le64(bch2_recovery_passes_to_stable(passes)); 157 158 for (const u16 *e = i->errors; e < i->errors + i->nr_errors; e++) 159 __set_bit_le64(*e, ext->errors_silent); 160 } 161 } 162 163 #define x(ver, passes, ...) static const u16 downgrade_##ver##_errors[] = { __VA_ARGS__ }; 164 DOWNGRADE_TABLE() 165 #undef x 166 167 static const struct upgrade_downgrade_entry downgrade_table[] = { 168 #define x(ver, passes, ...) { \ 169 .recovery_passes = passes, \ 170 .version = bcachefs_metadata_version_##ver,\ 171 .nr_errors = ARRAY_SIZE(downgrade_##ver##_errors), \ 172 .errors = downgrade_##ver##_errors, \ 173 }, 174 DOWNGRADE_TABLE() 175 #undef x 176 }; 177 178 static int downgrade_table_extra(struct bch_fs *c, darray_char *table) 179 { 180 struct bch_sb_field_downgrade_entry *dst = (void *) &darray_top(*table); 181 unsigned bytes = sizeof(*dst) + sizeof(dst->errors[0]) * le16_to_cpu(dst->nr_errors); 182 int ret = 0; 183 184 unsigned nr_errors = le16_to_cpu(dst->nr_errors); 185 186 switch (le16_to_cpu(dst->version)) { 187 case bcachefs_metadata_version_bucket_stripe_sectors: 188 if (have_stripes(c)) { 189 bytes += sizeof(dst->errors[0]) * 2; 190 191 ret = darray_make_room(table, bytes); 192 if (ret) 193 return ret; 194 195 /* open coded __set_bit_le64, as dst is packed and 196 * dst->recovery_passes is misaligned */ 197 unsigned b = BCH_RECOVERY_PASS_STABLE_check_allocations; 198 dst->recovery_passes[b / 64] |= cpu_to_le64(BIT_ULL(b % 64)); 199 200 dst->errors[nr_errors++] = cpu_to_le16(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong); 201 } 202 break; 203 } 204 205 dst->nr_errors = cpu_to_le16(nr_errors); 206 return ret; 207 } 208 209 static inline const struct bch_sb_field_downgrade_entry * 210 downgrade_entry_next_c(const struct bch_sb_field_downgrade_entry *e) 211 { 212 return (void *) &e->errors[le16_to_cpu(e->nr_errors)]; 213 } 214 215 #define for_each_downgrade_entry(_d, _i) \ 216 for (const struct bch_sb_field_downgrade_entry *_i = (_d)->entries; \ 217 (void *) _i < vstruct_end(&(_d)->field) && \ 218 (void *) &_i->errors[0] <= vstruct_end(&(_d)->field) && \ 219 (void *) downgrade_entry_next_c(_i) <= vstruct_end(&(_d)->field); \ 220 _i = downgrade_entry_next_c(_i)) 221 222 static int bch2_sb_downgrade_validate(struct bch_sb *sb, struct bch_sb_field *f, 223 enum bch_validate_flags flags, struct printbuf *err) 224 { 225 struct bch_sb_field_downgrade *e = field_to_type(f, downgrade); 226 227 for (const struct bch_sb_field_downgrade_entry *i = e->entries; 228 (void *) i < vstruct_end(&e->field); 229 i = downgrade_entry_next_c(i)) { 230 /* 231 * Careful: sb_field_downgrade_entry is only 2 byte aligned, but 232 * section sizes are 8 byte aligned - an empty entry spanning 233 * the end of the section is allowed (and ignored): 234 */ 235 if ((void *) &i->errors[0] > vstruct_end(&e->field)) 236 break; 237 238 if (flags & BCH_VALIDATE_write && 239 (void *) downgrade_entry_next_c(i) > vstruct_end(&e->field)) { 240 prt_printf(err, "downgrade entry overruns end of superblock section"); 241 return -BCH_ERR_invalid_sb_downgrade; 242 } 243 244 if (BCH_VERSION_MAJOR(le16_to_cpu(i->version)) != 245 BCH_VERSION_MAJOR(le16_to_cpu(sb->version))) { 246 prt_printf(err, "downgrade entry with mismatched major version (%u != %u)", 247 BCH_VERSION_MAJOR(le16_to_cpu(i->version)), 248 BCH_VERSION_MAJOR(le16_to_cpu(sb->version))); 249 return -BCH_ERR_invalid_sb_downgrade; 250 } 251 } 252 253 return 0; 254 } 255 256 static void bch2_sb_downgrade_to_text(struct printbuf *out, struct bch_sb *sb, 257 struct bch_sb_field *f) 258 { 259 struct bch_sb_field_downgrade *e = field_to_type(f, downgrade); 260 261 if (out->nr_tabstops <= 1) 262 printbuf_tabstop_push(out, 16); 263 264 for_each_downgrade_entry(e, i) { 265 prt_str(out, "version:\t"); 266 bch2_version_to_text(out, le16_to_cpu(i->version)); 267 prt_newline(out); 268 269 prt_str(out, "recovery passes:\t"); 270 prt_bitflags(out, bch2_recovery_passes, 271 bch2_recovery_passes_from_stable(le64_to_cpu(i->recovery_passes[0]))); 272 prt_newline(out); 273 274 prt_str(out, "errors:\t"); 275 bool first = true; 276 for (unsigned j = 0; j < le16_to_cpu(i->nr_errors); j++) { 277 if (!first) 278 prt_char(out, ','); 279 first = false; 280 unsigned e = le16_to_cpu(i->errors[j]); 281 prt_str(out, e < BCH_SB_ERR_MAX ? bch2_sb_error_strs[e] : "(unknown)"); 282 } 283 prt_newline(out); 284 } 285 } 286 287 const struct bch_sb_field_ops bch_sb_field_ops_downgrade = { 288 .validate = bch2_sb_downgrade_validate, 289 .to_text = bch2_sb_downgrade_to_text, 290 }; 291 292 int bch2_sb_downgrade_update(struct bch_fs *c) 293 { 294 if (!test_bit(BCH_FS_btree_running, &c->flags)) 295 return 0; 296 297 darray_char table = {}; 298 int ret = 0; 299 300 for (const struct upgrade_downgrade_entry *src = downgrade_table; 301 src < downgrade_table + ARRAY_SIZE(downgrade_table); 302 src++) { 303 if (BCH_VERSION_MAJOR(src->version) != BCH_VERSION_MAJOR(le16_to_cpu(c->disk_sb.sb->version))) 304 continue; 305 306 struct bch_sb_field_downgrade_entry *dst; 307 unsigned bytes = sizeof(*dst) + sizeof(dst->errors[0]) * src->nr_errors; 308 309 ret = darray_make_room(&table, bytes); 310 if (ret) 311 goto out; 312 313 dst = (void *) &darray_top(table); 314 dst->version = cpu_to_le16(src->version); 315 dst->recovery_passes[0] = cpu_to_le64(bch2_recovery_passes_to_stable(src->recovery_passes)); 316 dst->recovery_passes[1] = 0; 317 dst->nr_errors = cpu_to_le16(src->nr_errors); 318 for (unsigned i = 0; i < src->nr_errors; i++) 319 dst->errors[i] = cpu_to_le16(src->errors[i]); 320 321 downgrade_table_extra(c, &table); 322 323 if (!dst->recovery_passes[0] && 324 !dst->recovery_passes[1] && 325 !dst->nr_errors) 326 continue; 327 328 table.nr += sizeof(*dst) + sizeof(dst->errors[0]) * le16_to_cpu(dst->nr_errors); 329 } 330 331 struct bch_sb_field_downgrade *d = bch2_sb_field_get(c->disk_sb.sb, downgrade); 332 333 unsigned sb_u64s = DIV_ROUND_UP(sizeof(*d) + table.nr, sizeof(u64)); 334 335 if (d && le32_to_cpu(d->field.u64s) > sb_u64s) 336 goto out; 337 338 d = bch2_sb_field_resize(&c->disk_sb, downgrade, sb_u64s); 339 if (!d) { 340 ret = -BCH_ERR_ENOSPC_sb_downgrade; 341 goto out; 342 } 343 344 memcpy(d->entries, table.data, table.nr); 345 memset_u64s_tail(d->entries, 0, table.nr); 346 out: 347 darray_exit(&table); 348 return ret; 349 } 350 351 void bch2_sb_set_downgrade(struct bch_fs *c, unsigned new_minor, unsigned old_minor) 352 { 353 struct bch_sb_field_downgrade *d = bch2_sb_field_get(c->disk_sb.sb, downgrade); 354 if (!d) 355 return; 356 357 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext); 358 359 for_each_downgrade_entry(d, i) { 360 unsigned minor = BCH_VERSION_MINOR(le16_to_cpu(i->version)); 361 if (new_minor < minor && minor <= old_minor) { 362 ext->recovery_passes_required[0] |= i->recovery_passes[0]; 363 ext->recovery_passes_required[1] |= i->recovery_passes[1]; 364 365 for (unsigned j = 0; j < le16_to_cpu(i->nr_errors); j++) { 366 unsigned e = le16_to_cpu(i->errors[j]); 367 if (e < BCH_SB_ERR_MAX) 368 __set_bit(e, c->sb.errors_silent); 369 if (e < sizeof(ext->errors_silent) * 8) 370 __set_bit_le64(e, ext->errors_silent); 371 } 372 } 373 } 374 } 375