1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) STRATO AG 2011. All rights reserved. 4 */ 5 6 /* 7 * This module can be used to catch cases when the btrfs kernel 8 * code executes write requests to the disk that bring the file 9 * system in an inconsistent state. In such a state, a power-loss 10 * or kernel panic event would cause that the data on disk is 11 * lost or at least damaged. 12 * 13 * Code is added that examines all block write requests during 14 * runtime (including writes of the super block). Three rules 15 * are verified and an error is printed on violation of the 16 * rules: 17 * 1. It is not allowed to write a disk block which is 18 * currently referenced by the super block (either directly 19 * or indirectly). 20 * 2. When a super block is written, it is verified that all 21 * referenced (directly or indirectly) blocks fulfill the 22 * following requirements: 23 * 2a. All referenced blocks have either been present when 24 * the file system was mounted, (i.e., they have been 25 * referenced by the super block) or they have been 26 * written since then and the write completion callback 27 * was called and no write error was indicated and a 28 * FLUSH request to the device where these blocks are 29 * located was received and completed. 30 * 2b. All referenced blocks need to have a generation 31 * number which is equal to the parent's number. 32 * 33 * One issue that was found using this module was that the log 34 * tree on disk became temporarily corrupted because disk blocks 35 * that had been in use for the log tree had been freed and 36 * reused too early, while being referenced by the written super 37 * block. 38 * 39 * The search term in the kernel log that can be used to filter 40 * on the existence of detected integrity issues is 41 * "btrfs: attempt". 42 * 43 * The integrity check is enabled via mount options. These 44 * mount options are only supported if the integrity check 45 * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY. 46 * 47 * Example #1, apply integrity checks to all metadata: 48 * mount /dev/sdb1 /mnt -o check_int 49 * 50 * Example #2, apply integrity checks to all metadata and 51 * to data extents: 52 * mount /dev/sdb1 /mnt -o check_int_data 53 * 54 * Example #3, apply integrity checks to all metadata and dump 55 * the tree that the super block references to kernel messages 56 * each time after a super block was written: 57 * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263 58 * 59 * If the integrity check tool is included and activated in 60 * the mount options, plenty of kernel memory is used, and 61 * plenty of additional CPU cycles are spent. Enabling this 62 * functionality is not intended for normal use. In most 63 * cases, unless you are a btrfs developer who needs to verify 64 * the integrity of (super)-block write requests, do not 65 * enable the config option BTRFS_FS_CHECK_INTEGRITY to 66 * include and compile the integrity check tool. 67 * 68 * Expect millions of lines of information in the kernel log with an 69 * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the 70 * kernel config to at least 26 (which is 64MB). Usually the value is 71 * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be 72 * changed like this before LOG_BUF_SHIFT can be set to a high value: 73 * config LOG_BUF_SHIFT 74 * int "Kernel log buffer size (16 => 64KB, 17 => 128KB)" 75 * range 12 30 76 */ 77 78 #include <linux/sched.h> 79 #include <linux/slab.h> 80 #include <linux/mutex.h> 81 #include <linux/blkdev.h> 82 #include <linux/mm.h> 83 #include <linux/string.h> 84 #include <crypto/hash.h> 85 #include "ctree.h" 86 #include "disk-io.h" 87 #include "transaction.h" 88 #include "extent_io.h" 89 #include "volumes.h" 90 #include "print-tree.h" 91 #include "locking.h" 92 #include "check-integrity.h" 93 #include "rcu-string.h" 94 #include "compression.h" 95 96 #define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000 97 #define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000 98 #define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100 99 #define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051 100 #define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807 101 #define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530 102 #define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300 103 #define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6) /* in characters, 104 * excluding " [...]" */ 105 #define BTRFSIC_GENERATION_UNKNOWN ((u64)-1) 106 107 /* 108 * The definition of the bitmask fields for the print_mask. 109 * They are specified with the mount option check_integrity_print_mask. 110 */ 111 #define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE 0x00000001 112 #define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION 0x00000002 113 #define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE 0x00000004 114 #define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE 0x00000008 115 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH 0x00000010 116 #define BTRFSIC_PRINT_MASK_END_IO_BIO_BH 0x00000020 117 #define BTRFSIC_PRINT_MASK_VERBOSE 0x00000040 118 #define BTRFSIC_PRINT_MASK_VERY_VERBOSE 0x00000080 119 #define BTRFSIC_PRINT_MASK_INITIAL_TREE 0x00000100 120 #define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES 0x00000200 121 #define BTRFSIC_PRINT_MASK_INITIAL_DATABASE 0x00000400 122 #define BTRFSIC_PRINT_MASK_NUM_COPIES 0x00000800 123 #define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS 0x00001000 124 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE 0x00002000 125 126 struct btrfsic_dev_state; 127 struct btrfsic_state; 128 129 struct btrfsic_block { 130 u32 magic_num; /* only used for debug purposes */ 131 unsigned int is_metadata:1; /* if it is meta-data, not data-data */ 132 unsigned int is_superblock:1; /* if it is one of the superblocks */ 133 unsigned int is_iodone:1; /* if is done by lower subsystem */ 134 unsigned int iodone_w_error:1; /* error was indicated to endio */ 135 unsigned int never_written:1; /* block was added because it was 136 * referenced, not because it was 137 * written */ 138 unsigned int mirror_num; /* large enough to hold 139 * BTRFS_SUPER_MIRROR_MAX */ 140 struct btrfsic_dev_state *dev_state; 141 u64 dev_bytenr; /* key, physical byte num on disk */ 142 u64 logical_bytenr; /* logical byte num on disk */ 143 u64 generation; 144 struct btrfs_disk_key disk_key; /* extra info to print in case of 145 * issues, will not always be correct */ 146 struct list_head collision_resolving_node; /* list node */ 147 struct list_head all_blocks_node; /* list node */ 148 149 /* the following two lists contain block_link items */ 150 struct list_head ref_to_list; /* list */ 151 struct list_head ref_from_list; /* list */ 152 struct btrfsic_block *next_in_same_bio; 153 void *orig_bio_private; 154 bio_end_io_t *orig_bio_end_io; 155 blk_opf_t submit_bio_bh_rw; 156 u64 flush_gen; /* only valid if !never_written */ 157 }; 158 159 /* 160 * Elements of this type are allocated dynamically and required because 161 * each block object can refer to and can be ref from multiple blocks. 162 * The key to lookup them in the hashtable is the dev_bytenr of 163 * the block ref to plus the one from the block referred from. 164 * The fact that they are searchable via a hashtable and that a 165 * ref_cnt is maintained is not required for the btrfs integrity 166 * check algorithm itself, it is only used to make the output more 167 * beautiful in case that an error is detected (an error is defined 168 * as a write operation to a block while that block is still referenced). 169 */ 170 struct btrfsic_block_link { 171 u32 magic_num; /* only used for debug purposes */ 172 u32 ref_cnt; 173 struct list_head node_ref_to; /* list node */ 174 struct list_head node_ref_from; /* list node */ 175 struct list_head collision_resolving_node; /* list node */ 176 struct btrfsic_block *block_ref_to; 177 struct btrfsic_block *block_ref_from; 178 u64 parent_generation; 179 }; 180 181 struct btrfsic_dev_state { 182 u32 magic_num; /* only used for debug purposes */ 183 struct block_device *bdev; 184 struct btrfsic_state *state; 185 struct list_head collision_resolving_node; /* list node */ 186 struct btrfsic_block dummy_block_for_bio_bh_flush; 187 u64 last_flush_gen; 188 }; 189 190 struct btrfsic_block_hashtable { 191 struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE]; 192 }; 193 194 struct btrfsic_block_link_hashtable { 195 struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE]; 196 }; 197 198 struct btrfsic_dev_state_hashtable { 199 struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE]; 200 }; 201 202 struct btrfsic_block_data_ctx { 203 u64 start; /* virtual bytenr */ 204 u64 dev_bytenr; /* physical bytenr on device */ 205 u32 len; 206 struct btrfsic_dev_state *dev; 207 char **datav; 208 struct page **pagev; 209 void *mem_to_free; 210 }; 211 212 /* This structure is used to implement recursion without occupying 213 * any stack space, refer to btrfsic_process_metablock() */ 214 struct btrfsic_stack_frame { 215 u32 magic; 216 u32 nr; 217 int error; 218 int i; 219 int limit_nesting; 220 int num_copies; 221 int mirror_num; 222 struct btrfsic_block *block; 223 struct btrfsic_block_data_ctx *block_ctx; 224 struct btrfsic_block *next_block; 225 struct btrfsic_block_data_ctx next_block_ctx; 226 struct btrfs_header *hdr; 227 struct btrfsic_stack_frame *prev; 228 }; 229 230 /* Some state per mounted filesystem */ 231 struct btrfsic_state { 232 u32 print_mask; 233 int include_extent_data; 234 struct list_head all_blocks_list; 235 struct btrfsic_block_hashtable block_hashtable; 236 struct btrfsic_block_link_hashtable block_link_hashtable; 237 struct btrfs_fs_info *fs_info; 238 u64 max_superblock_generation; 239 struct btrfsic_block *latest_superblock; 240 u32 metablock_size; 241 u32 datablock_size; 242 }; 243 244 static int btrfsic_process_metablock(struct btrfsic_state *state, 245 struct btrfsic_block *block, 246 struct btrfsic_block_data_ctx *block_ctx, 247 int limit_nesting, int force_iodone_flag); 248 static void btrfsic_read_from_block_data( 249 struct btrfsic_block_data_ctx *block_ctx, 250 void *dst, u32 offset, size_t len); 251 static int btrfsic_create_link_to_next_block( 252 struct btrfsic_state *state, 253 struct btrfsic_block *block, 254 struct btrfsic_block_data_ctx 255 *block_ctx, u64 next_bytenr, 256 int limit_nesting, 257 struct btrfsic_block_data_ctx *next_block_ctx, 258 struct btrfsic_block **next_blockp, 259 int force_iodone_flag, 260 int *num_copiesp, int *mirror_nump, 261 struct btrfs_disk_key *disk_key, 262 u64 parent_generation); 263 static int btrfsic_handle_extent_data(struct btrfsic_state *state, 264 struct btrfsic_block *block, 265 struct btrfsic_block_data_ctx *block_ctx, 266 u32 item_offset, int force_iodone_flag); 267 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len, 268 struct btrfsic_block_data_ctx *block_ctx_out, 269 int mirror_num); 270 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx); 271 static int btrfsic_read_block(struct btrfsic_state *state, 272 struct btrfsic_block_data_ctx *block_ctx); 273 static int btrfsic_process_written_superblock( 274 struct btrfsic_state *state, 275 struct btrfsic_block *const block, 276 struct btrfs_super_block *const super_hdr); 277 static void btrfsic_bio_end_io(struct bio *bp); 278 static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state, 279 const struct btrfsic_block *block, 280 int recursion_level); 281 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state, 282 struct btrfsic_block *const block, 283 int recursion_level); 284 static void btrfsic_print_add_link(const struct btrfsic_state *state, 285 const struct btrfsic_block_link *l); 286 static void btrfsic_print_rem_link(const struct btrfsic_state *state, 287 const struct btrfsic_block_link *l); 288 static char btrfsic_get_block_type(const struct btrfsic_state *state, 289 const struct btrfsic_block *block); 290 static void btrfsic_dump_tree(const struct btrfsic_state *state); 291 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state, 292 const struct btrfsic_block *block, 293 int indent_level); 294 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add( 295 struct btrfsic_state *state, 296 struct btrfsic_block_data_ctx *next_block_ctx, 297 struct btrfsic_block *next_block, 298 struct btrfsic_block *from_block, 299 u64 parent_generation); 300 static struct btrfsic_block *btrfsic_block_lookup_or_add( 301 struct btrfsic_state *state, 302 struct btrfsic_block_data_ctx *block_ctx, 303 const char *additional_string, 304 int is_metadata, 305 int is_iodone, 306 int never_written, 307 int mirror_num, 308 int *was_created); 309 static int btrfsic_process_superblock_dev_mirror( 310 struct btrfsic_state *state, 311 struct btrfsic_dev_state *dev_state, 312 struct btrfs_device *device, 313 int superblock_mirror_num, 314 struct btrfsic_dev_state **selected_dev_state, 315 struct btrfs_super_block *selected_super); 316 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev); 317 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state, 318 u64 bytenr, 319 struct btrfsic_dev_state *dev_state, 320 u64 dev_bytenr); 321 322 static struct mutex btrfsic_mutex; 323 static int btrfsic_is_initialized; 324 static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable; 325 326 327 static void btrfsic_block_init(struct btrfsic_block *b) 328 { 329 b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER; 330 b->dev_state = NULL; 331 b->dev_bytenr = 0; 332 b->logical_bytenr = 0; 333 b->generation = BTRFSIC_GENERATION_UNKNOWN; 334 b->disk_key.objectid = 0; 335 b->disk_key.type = 0; 336 b->disk_key.offset = 0; 337 b->is_metadata = 0; 338 b->is_superblock = 0; 339 b->is_iodone = 0; 340 b->iodone_w_error = 0; 341 b->never_written = 0; 342 b->mirror_num = 0; 343 b->next_in_same_bio = NULL; 344 b->orig_bio_private = NULL; 345 b->orig_bio_end_io = NULL; 346 INIT_LIST_HEAD(&b->collision_resolving_node); 347 INIT_LIST_HEAD(&b->all_blocks_node); 348 INIT_LIST_HEAD(&b->ref_to_list); 349 INIT_LIST_HEAD(&b->ref_from_list); 350 b->submit_bio_bh_rw = 0; 351 b->flush_gen = 0; 352 } 353 354 static struct btrfsic_block *btrfsic_block_alloc(void) 355 { 356 struct btrfsic_block *b; 357 358 b = kzalloc(sizeof(*b), GFP_NOFS); 359 if (NULL != b) 360 btrfsic_block_init(b); 361 362 return b; 363 } 364 365 static void btrfsic_block_free(struct btrfsic_block *b) 366 { 367 BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num)); 368 kfree(b); 369 } 370 371 static void btrfsic_block_link_init(struct btrfsic_block_link *l) 372 { 373 l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER; 374 l->ref_cnt = 1; 375 INIT_LIST_HEAD(&l->node_ref_to); 376 INIT_LIST_HEAD(&l->node_ref_from); 377 INIT_LIST_HEAD(&l->collision_resolving_node); 378 l->block_ref_to = NULL; 379 l->block_ref_from = NULL; 380 } 381 382 static struct btrfsic_block_link *btrfsic_block_link_alloc(void) 383 { 384 struct btrfsic_block_link *l; 385 386 l = kzalloc(sizeof(*l), GFP_NOFS); 387 if (NULL != l) 388 btrfsic_block_link_init(l); 389 390 return l; 391 } 392 393 static void btrfsic_block_link_free(struct btrfsic_block_link *l) 394 { 395 BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num)); 396 kfree(l); 397 } 398 399 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds) 400 { 401 ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER; 402 ds->bdev = NULL; 403 ds->state = NULL; 404 INIT_LIST_HEAD(&ds->collision_resolving_node); 405 ds->last_flush_gen = 0; 406 btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush); 407 ds->dummy_block_for_bio_bh_flush.is_iodone = 1; 408 ds->dummy_block_for_bio_bh_flush.dev_state = ds; 409 } 410 411 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void) 412 { 413 struct btrfsic_dev_state *ds; 414 415 ds = kzalloc(sizeof(*ds), GFP_NOFS); 416 if (NULL != ds) 417 btrfsic_dev_state_init(ds); 418 419 return ds; 420 } 421 422 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds) 423 { 424 BUG_ON(!(NULL == ds || 425 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num)); 426 kfree(ds); 427 } 428 429 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h) 430 { 431 int i; 432 433 for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++) 434 INIT_LIST_HEAD(h->table + i); 435 } 436 437 static void btrfsic_block_hashtable_add(struct btrfsic_block *b, 438 struct btrfsic_block_hashtable *h) 439 { 440 const unsigned int hashval = 441 (((unsigned int)(b->dev_bytenr >> 16)) ^ 442 ((unsigned int)((uintptr_t)b->dev_state->bdev))) & 443 (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1); 444 445 list_add(&b->collision_resolving_node, h->table + hashval); 446 } 447 448 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b) 449 { 450 list_del(&b->collision_resolving_node); 451 } 452 453 static struct btrfsic_block *btrfsic_block_hashtable_lookup( 454 struct block_device *bdev, 455 u64 dev_bytenr, 456 struct btrfsic_block_hashtable *h) 457 { 458 const unsigned int hashval = 459 (((unsigned int)(dev_bytenr >> 16)) ^ 460 ((unsigned int)((uintptr_t)bdev))) & 461 (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1); 462 struct btrfsic_block *b; 463 464 list_for_each_entry(b, h->table + hashval, collision_resolving_node) { 465 if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr) 466 return b; 467 } 468 469 return NULL; 470 } 471 472 static void btrfsic_block_link_hashtable_init( 473 struct btrfsic_block_link_hashtable *h) 474 { 475 int i; 476 477 for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++) 478 INIT_LIST_HEAD(h->table + i); 479 } 480 481 static void btrfsic_block_link_hashtable_add( 482 struct btrfsic_block_link *l, 483 struct btrfsic_block_link_hashtable *h) 484 { 485 const unsigned int hashval = 486 (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^ 487 ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^ 488 ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^ 489 ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev))) 490 & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1); 491 492 BUG_ON(NULL == l->block_ref_to); 493 BUG_ON(NULL == l->block_ref_from); 494 list_add(&l->collision_resolving_node, h->table + hashval); 495 } 496 497 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l) 498 { 499 list_del(&l->collision_resolving_node); 500 } 501 502 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup( 503 struct block_device *bdev_ref_to, 504 u64 dev_bytenr_ref_to, 505 struct block_device *bdev_ref_from, 506 u64 dev_bytenr_ref_from, 507 struct btrfsic_block_link_hashtable *h) 508 { 509 const unsigned int hashval = 510 (((unsigned int)(dev_bytenr_ref_to >> 16)) ^ 511 ((unsigned int)(dev_bytenr_ref_from >> 16)) ^ 512 ((unsigned int)((uintptr_t)bdev_ref_to)) ^ 513 ((unsigned int)((uintptr_t)bdev_ref_from))) & 514 (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1); 515 struct btrfsic_block_link *l; 516 517 list_for_each_entry(l, h->table + hashval, collision_resolving_node) { 518 BUG_ON(NULL == l->block_ref_to); 519 BUG_ON(NULL == l->block_ref_from); 520 if (l->block_ref_to->dev_state->bdev == bdev_ref_to && 521 l->block_ref_to->dev_bytenr == dev_bytenr_ref_to && 522 l->block_ref_from->dev_state->bdev == bdev_ref_from && 523 l->block_ref_from->dev_bytenr == dev_bytenr_ref_from) 524 return l; 525 } 526 527 return NULL; 528 } 529 530 static void btrfsic_dev_state_hashtable_init( 531 struct btrfsic_dev_state_hashtable *h) 532 { 533 int i; 534 535 for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++) 536 INIT_LIST_HEAD(h->table + i); 537 } 538 539 static void btrfsic_dev_state_hashtable_add( 540 struct btrfsic_dev_state *ds, 541 struct btrfsic_dev_state_hashtable *h) 542 { 543 const unsigned int hashval = 544 (((unsigned int)((uintptr_t)ds->bdev->bd_dev)) & 545 (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1)); 546 547 list_add(&ds->collision_resolving_node, h->table + hashval); 548 } 549 550 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds) 551 { 552 list_del(&ds->collision_resolving_node); 553 } 554 555 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev, 556 struct btrfsic_dev_state_hashtable *h) 557 { 558 const unsigned int hashval = 559 dev & (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1); 560 struct btrfsic_dev_state *ds; 561 562 list_for_each_entry(ds, h->table + hashval, collision_resolving_node) { 563 if (ds->bdev->bd_dev == dev) 564 return ds; 565 } 566 567 return NULL; 568 } 569 570 static int btrfsic_process_superblock(struct btrfsic_state *state, 571 struct btrfs_fs_devices *fs_devices) 572 { 573 struct btrfs_super_block *selected_super; 574 struct list_head *dev_head = &fs_devices->devices; 575 struct btrfs_device *device; 576 struct btrfsic_dev_state *selected_dev_state = NULL; 577 int ret = 0; 578 int pass; 579 580 selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS); 581 if (!selected_super) 582 return -ENOMEM; 583 584 list_for_each_entry(device, dev_head, dev_list) { 585 int i; 586 struct btrfsic_dev_state *dev_state; 587 588 if (!device->bdev || !device->name) 589 continue; 590 591 dev_state = btrfsic_dev_state_lookup(device->bdev->bd_dev); 592 BUG_ON(NULL == dev_state); 593 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { 594 ret = btrfsic_process_superblock_dev_mirror( 595 state, dev_state, device, i, 596 &selected_dev_state, selected_super); 597 if (0 != ret && 0 == i) { 598 kfree(selected_super); 599 return ret; 600 } 601 } 602 } 603 604 if (NULL == state->latest_superblock) { 605 pr_info("btrfsic: no superblock found!\n"); 606 kfree(selected_super); 607 return -1; 608 } 609 610 for (pass = 0; pass < 3; pass++) { 611 int num_copies; 612 int mirror_num; 613 u64 next_bytenr; 614 615 switch (pass) { 616 case 0: 617 next_bytenr = btrfs_super_root(selected_super); 618 if (state->print_mask & 619 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION) 620 pr_info("root@%llu\n", next_bytenr); 621 break; 622 case 1: 623 next_bytenr = btrfs_super_chunk_root(selected_super); 624 if (state->print_mask & 625 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION) 626 pr_info("chunk@%llu\n", next_bytenr); 627 break; 628 case 2: 629 next_bytenr = btrfs_super_log_root(selected_super); 630 if (0 == next_bytenr) 631 continue; 632 if (state->print_mask & 633 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION) 634 pr_info("log@%llu\n", next_bytenr); 635 break; 636 } 637 638 num_copies = btrfs_num_copies(state->fs_info, next_bytenr, 639 state->metablock_size); 640 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES) 641 pr_info("num_copies(log_bytenr=%llu) = %d\n", 642 next_bytenr, num_copies); 643 644 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) { 645 struct btrfsic_block *next_block; 646 struct btrfsic_block_data_ctx tmp_next_block_ctx; 647 struct btrfsic_block_link *l; 648 649 ret = btrfsic_map_block(state, next_bytenr, 650 state->metablock_size, 651 &tmp_next_block_ctx, 652 mirror_num); 653 if (ret) { 654 pr_info("btrfsic: btrfsic_map_block(root @%llu, mirror %d) failed!\n", 655 next_bytenr, mirror_num); 656 kfree(selected_super); 657 return -1; 658 } 659 660 next_block = btrfsic_block_hashtable_lookup( 661 tmp_next_block_ctx.dev->bdev, 662 tmp_next_block_ctx.dev_bytenr, 663 &state->block_hashtable); 664 BUG_ON(NULL == next_block); 665 666 l = btrfsic_block_link_hashtable_lookup( 667 tmp_next_block_ctx.dev->bdev, 668 tmp_next_block_ctx.dev_bytenr, 669 state->latest_superblock->dev_state-> 670 bdev, 671 state->latest_superblock->dev_bytenr, 672 &state->block_link_hashtable); 673 BUG_ON(NULL == l); 674 675 ret = btrfsic_read_block(state, &tmp_next_block_ctx); 676 if (ret < (int)PAGE_SIZE) { 677 pr_info("btrfsic: read @logical %llu failed!\n", 678 tmp_next_block_ctx.start); 679 btrfsic_release_block_ctx(&tmp_next_block_ctx); 680 kfree(selected_super); 681 return -1; 682 } 683 684 ret = btrfsic_process_metablock(state, 685 next_block, 686 &tmp_next_block_ctx, 687 BTRFS_MAX_LEVEL + 3, 1); 688 btrfsic_release_block_ctx(&tmp_next_block_ctx); 689 } 690 } 691 692 kfree(selected_super); 693 return ret; 694 } 695 696 static int btrfsic_process_superblock_dev_mirror( 697 struct btrfsic_state *state, 698 struct btrfsic_dev_state *dev_state, 699 struct btrfs_device *device, 700 int superblock_mirror_num, 701 struct btrfsic_dev_state **selected_dev_state, 702 struct btrfs_super_block *selected_super) 703 { 704 struct btrfs_fs_info *fs_info = state->fs_info; 705 struct btrfs_super_block *super_tmp; 706 u64 dev_bytenr; 707 struct btrfsic_block *superblock_tmp; 708 int pass; 709 struct block_device *const superblock_bdev = device->bdev; 710 struct page *page; 711 struct address_space *mapping = superblock_bdev->bd_inode->i_mapping; 712 int ret = 0; 713 714 /* super block bytenr is always the unmapped device bytenr */ 715 dev_bytenr = btrfs_sb_offset(superblock_mirror_num); 716 if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->commit_total_bytes) 717 return -1; 718 719 page = read_cache_page_gfp(mapping, dev_bytenr >> PAGE_SHIFT, GFP_NOFS); 720 if (IS_ERR(page)) 721 return -1; 722 723 super_tmp = page_address(page); 724 725 if (btrfs_super_bytenr(super_tmp) != dev_bytenr || 726 btrfs_super_magic(super_tmp) != BTRFS_MAGIC || 727 memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) || 728 btrfs_super_nodesize(super_tmp) != state->metablock_size || 729 btrfs_super_sectorsize(super_tmp) != state->datablock_size) { 730 ret = 0; 731 goto out; 732 } 733 734 superblock_tmp = 735 btrfsic_block_hashtable_lookup(superblock_bdev, 736 dev_bytenr, 737 &state->block_hashtable); 738 if (NULL == superblock_tmp) { 739 superblock_tmp = btrfsic_block_alloc(); 740 if (NULL == superblock_tmp) { 741 ret = -1; 742 goto out; 743 } 744 /* for superblock, only the dev_bytenr makes sense */ 745 superblock_tmp->dev_bytenr = dev_bytenr; 746 superblock_tmp->dev_state = dev_state; 747 superblock_tmp->logical_bytenr = dev_bytenr; 748 superblock_tmp->generation = btrfs_super_generation(super_tmp); 749 superblock_tmp->is_metadata = 1; 750 superblock_tmp->is_superblock = 1; 751 superblock_tmp->is_iodone = 1; 752 superblock_tmp->never_written = 0; 753 superblock_tmp->mirror_num = 1 + superblock_mirror_num; 754 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE) 755 btrfs_info_in_rcu(fs_info, 756 "new initial S-block (bdev %p, %s) @%llu (%pg/%llu/%d)", 757 superblock_bdev, 758 rcu_str_deref(device->name), dev_bytenr, 759 dev_state->bdev, dev_bytenr, 760 superblock_mirror_num); 761 list_add(&superblock_tmp->all_blocks_node, 762 &state->all_blocks_list); 763 btrfsic_block_hashtable_add(superblock_tmp, 764 &state->block_hashtable); 765 } 766 767 /* select the one with the highest generation field */ 768 if (btrfs_super_generation(super_tmp) > 769 state->max_superblock_generation || 770 0 == state->max_superblock_generation) { 771 memcpy(selected_super, super_tmp, sizeof(*selected_super)); 772 *selected_dev_state = dev_state; 773 state->max_superblock_generation = 774 btrfs_super_generation(super_tmp); 775 state->latest_superblock = superblock_tmp; 776 } 777 778 for (pass = 0; pass < 3; pass++) { 779 u64 next_bytenr; 780 int num_copies; 781 int mirror_num; 782 const char *additional_string = NULL; 783 struct btrfs_disk_key tmp_disk_key; 784 785 tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY; 786 tmp_disk_key.offset = 0; 787 switch (pass) { 788 case 0: 789 btrfs_set_disk_key_objectid(&tmp_disk_key, 790 BTRFS_ROOT_TREE_OBJECTID); 791 additional_string = "initial root "; 792 next_bytenr = btrfs_super_root(super_tmp); 793 break; 794 case 1: 795 btrfs_set_disk_key_objectid(&tmp_disk_key, 796 BTRFS_CHUNK_TREE_OBJECTID); 797 additional_string = "initial chunk "; 798 next_bytenr = btrfs_super_chunk_root(super_tmp); 799 break; 800 case 2: 801 btrfs_set_disk_key_objectid(&tmp_disk_key, 802 BTRFS_TREE_LOG_OBJECTID); 803 additional_string = "initial log "; 804 next_bytenr = btrfs_super_log_root(super_tmp); 805 if (0 == next_bytenr) 806 continue; 807 break; 808 } 809 810 num_copies = btrfs_num_copies(fs_info, next_bytenr, 811 state->metablock_size); 812 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES) 813 pr_info("num_copies(log_bytenr=%llu) = %d\n", 814 next_bytenr, num_copies); 815 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) { 816 struct btrfsic_block *next_block; 817 struct btrfsic_block_data_ctx tmp_next_block_ctx; 818 struct btrfsic_block_link *l; 819 820 if (btrfsic_map_block(state, next_bytenr, 821 state->metablock_size, 822 &tmp_next_block_ctx, 823 mirror_num)) { 824 pr_info("btrfsic: btrfsic_map_block(bytenr @%llu, mirror %d) failed!\n", 825 next_bytenr, mirror_num); 826 ret = -1; 827 goto out; 828 } 829 830 next_block = btrfsic_block_lookup_or_add( 831 state, &tmp_next_block_ctx, 832 additional_string, 1, 1, 0, 833 mirror_num, NULL); 834 if (NULL == next_block) { 835 btrfsic_release_block_ctx(&tmp_next_block_ctx); 836 ret = -1; 837 goto out; 838 } 839 840 next_block->disk_key = tmp_disk_key; 841 next_block->generation = BTRFSIC_GENERATION_UNKNOWN; 842 l = btrfsic_block_link_lookup_or_add( 843 state, &tmp_next_block_ctx, 844 next_block, superblock_tmp, 845 BTRFSIC_GENERATION_UNKNOWN); 846 btrfsic_release_block_ctx(&tmp_next_block_ctx); 847 if (NULL == l) { 848 ret = -1; 849 goto out; 850 } 851 } 852 } 853 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES) 854 btrfsic_dump_tree_sub(state, superblock_tmp, 0); 855 856 out: 857 put_page(page); 858 return ret; 859 } 860 861 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void) 862 { 863 struct btrfsic_stack_frame *sf; 864 865 sf = kzalloc(sizeof(*sf), GFP_NOFS); 866 if (sf) 867 sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER; 868 return sf; 869 } 870 871 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf) 872 { 873 BUG_ON(!(NULL == sf || 874 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic)); 875 kfree(sf); 876 } 877 878 static noinline_for_stack int btrfsic_process_metablock( 879 struct btrfsic_state *state, 880 struct btrfsic_block *const first_block, 881 struct btrfsic_block_data_ctx *const first_block_ctx, 882 int first_limit_nesting, int force_iodone_flag) 883 { 884 struct btrfsic_stack_frame initial_stack_frame = { 0 }; 885 struct btrfsic_stack_frame *sf; 886 struct btrfsic_stack_frame *next_stack; 887 struct btrfs_header *const first_hdr = 888 (struct btrfs_header *)first_block_ctx->datav[0]; 889 890 BUG_ON(!first_hdr); 891 sf = &initial_stack_frame; 892 sf->error = 0; 893 sf->i = -1; 894 sf->limit_nesting = first_limit_nesting; 895 sf->block = first_block; 896 sf->block_ctx = first_block_ctx; 897 sf->next_block = NULL; 898 sf->hdr = first_hdr; 899 sf->prev = NULL; 900 901 continue_with_new_stack_frame: 902 sf->block->generation = btrfs_stack_header_generation(sf->hdr); 903 if (0 == sf->hdr->level) { 904 struct btrfs_leaf *const leafhdr = 905 (struct btrfs_leaf *)sf->hdr; 906 907 if (-1 == sf->i) { 908 sf->nr = btrfs_stack_header_nritems(&leafhdr->header); 909 910 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 911 pr_info("leaf %llu items %d generation %llu owner %llu\n", 912 sf->block_ctx->start, sf->nr, 913 btrfs_stack_header_generation( 914 &leafhdr->header), 915 btrfs_stack_header_owner( 916 &leafhdr->header)); 917 } 918 919 continue_with_current_leaf_stack_frame: 920 if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) { 921 sf->i++; 922 sf->num_copies = 0; 923 } 924 925 if (sf->i < sf->nr) { 926 struct btrfs_item disk_item; 927 u32 disk_item_offset = 928 (uintptr_t)(leafhdr->items + sf->i) - 929 (uintptr_t)leafhdr; 930 struct btrfs_disk_key *disk_key; 931 u8 type; 932 u32 item_offset; 933 u32 item_size; 934 935 if (disk_item_offset + sizeof(struct btrfs_item) > 936 sf->block_ctx->len) { 937 leaf_item_out_of_bounce_error: 938 pr_info( 939 "btrfsic: leaf item out of bounce at logical %llu, dev %pg\n", 940 sf->block_ctx->start, 941 sf->block_ctx->dev->bdev); 942 goto one_stack_frame_backwards; 943 } 944 btrfsic_read_from_block_data(sf->block_ctx, 945 &disk_item, 946 disk_item_offset, 947 sizeof(struct btrfs_item)); 948 item_offset = btrfs_stack_item_offset(&disk_item); 949 item_size = btrfs_stack_item_size(&disk_item); 950 disk_key = &disk_item.key; 951 type = btrfs_disk_key_type(disk_key); 952 953 if (BTRFS_ROOT_ITEM_KEY == type) { 954 struct btrfs_root_item root_item; 955 u32 root_item_offset; 956 u64 next_bytenr; 957 958 root_item_offset = item_offset + 959 offsetof(struct btrfs_leaf, items); 960 if (root_item_offset + item_size > 961 sf->block_ctx->len) 962 goto leaf_item_out_of_bounce_error; 963 btrfsic_read_from_block_data( 964 sf->block_ctx, &root_item, 965 root_item_offset, 966 item_size); 967 next_bytenr = btrfs_root_bytenr(&root_item); 968 969 sf->error = 970 btrfsic_create_link_to_next_block( 971 state, 972 sf->block, 973 sf->block_ctx, 974 next_bytenr, 975 sf->limit_nesting, 976 &sf->next_block_ctx, 977 &sf->next_block, 978 force_iodone_flag, 979 &sf->num_copies, 980 &sf->mirror_num, 981 disk_key, 982 btrfs_root_generation( 983 &root_item)); 984 if (sf->error) 985 goto one_stack_frame_backwards; 986 987 if (NULL != sf->next_block) { 988 struct btrfs_header *const next_hdr = 989 (struct btrfs_header *) 990 sf->next_block_ctx.datav[0]; 991 992 next_stack = 993 btrfsic_stack_frame_alloc(); 994 if (NULL == next_stack) { 995 sf->error = -1; 996 btrfsic_release_block_ctx( 997 &sf-> 998 next_block_ctx); 999 goto one_stack_frame_backwards; 1000 } 1001 1002 next_stack->i = -1; 1003 next_stack->block = sf->next_block; 1004 next_stack->block_ctx = 1005 &sf->next_block_ctx; 1006 next_stack->next_block = NULL; 1007 next_stack->hdr = next_hdr; 1008 next_stack->limit_nesting = 1009 sf->limit_nesting - 1; 1010 next_stack->prev = sf; 1011 sf = next_stack; 1012 goto continue_with_new_stack_frame; 1013 } 1014 } else if (BTRFS_EXTENT_DATA_KEY == type && 1015 state->include_extent_data) { 1016 sf->error = btrfsic_handle_extent_data( 1017 state, 1018 sf->block, 1019 sf->block_ctx, 1020 item_offset, 1021 force_iodone_flag); 1022 if (sf->error) 1023 goto one_stack_frame_backwards; 1024 } 1025 1026 goto continue_with_current_leaf_stack_frame; 1027 } 1028 } else { 1029 struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr; 1030 1031 if (-1 == sf->i) { 1032 sf->nr = btrfs_stack_header_nritems(&nodehdr->header); 1033 1034 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1035 pr_info("node %llu level %d items %d generation %llu owner %llu\n", 1036 sf->block_ctx->start, 1037 nodehdr->header.level, sf->nr, 1038 btrfs_stack_header_generation( 1039 &nodehdr->header), 1040 btrfs_stack_header_owner( 1041 &nodehdr->header)); 1042 } 1043 1044 continue_with_current_node_stack_frame: 1045 if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) { 1046 sf->i++; 1047 sf->num_copies = 0; 1048 } 1049 1050 if (sf->i < sf->nr) { 1051 struct btrfs_key_ptr key_ptr; 1052 u32 key_ptr_offset; 1053 u64 next_bytenr; 1054 1055 key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) - 1056 (uintptr_t)nodehdr; 1057 if (key_ptr_offset + sizeof(struct btrfs_key_ptr) > 1058 sf->block_ctx->len) { 1059 pr_info( 1060 "btrfsic: node item out of bounce at logical %llu, dev %pg\n", 1061 sf->block_ctx->start, 1062 sf->block_ctx->dev->bdev); 1063 goto one_stack_frame_backwards; 1064 } 1065 btrfsic_read_from_block_data( 1066 sf->block_ctx, &key_ptr, key_ptr_offset, 1067 sizeof(struct btrfs_key_ptr)); 1068 next_bytenr = btrfs_stack_key_blockptr(&key_ptr); 1069 1070 sf->error = btrfsic_create_link_to_next_block( 1071 state, 1072 sf->block, 1073 sf->block_ctx, 1074 next_bytenr, 1075 sf->limit_nesting, 1076 &sf->next_block_ctx, 1077 &sf->next_block, 1078 force_iodone_flag, 1079 &sf->num_copies, 1080 &sf->mirror_num, 1081 &key_ptr.key, 1082 btrfs_stack_key_generation(&key_ptr)); 1083 if (sf->error) 1084 goto one_stack_frame_backwards; 1085 1086 if (NULL != sf->next_block) { 1087 struct btrfs_header *const next_hdr = 1088 (struct btrfs_header *) 1089 sf->next_block_ctx.datav[0]; 1090 1091 next_stack = btrfsic_stack_frame_alloc(); 1092 if (NULL == next_stack) { 1093 sf->error = -1; 1094 goto one_stack_frame_backwards; 1095 } 1096 1097 next_stack->i = -1; 1098 next_stack->block = sf->next_block; 1099 next_stack->block_ctx = &sf->next_block_ctx; 1100 next_stack->next_block = NULL; 1101 next_stack->hdr = next_hdr; 1102 next_stack->limit_nesting = 1103 sf->limit_nesting - 1; 1104 next_stack->prev = sf; 1105 sf = next_stack; 1106 goto continue_with_new_stack_frame; 1107 } 1108 1109 goto continue_with_current_node_stack_frame; 1110 } 1111 } 1112 1113 one_stack_frame_backwards: 1114 if (NULL != sf->prev) { 1115 struct btrfsic_stack_frame *const prev = sf->prev; 1116 1117 /* the one for the initial block is freed in the caller */ 1118 btrfsic_release_block_ctx(sf->block_ctx); 1119 1120 if (sf->error) { 1121 prev->error = sf->error; 1122 btrfsic_stack_frame_free(sf); 1123 sf = prev; 1124 goto one_stack_frame_backwards; 1125 } 1126 1127 btrfsic_stack_frame_free(sf); 1128 sf = prev; 1129 goto continue_with_new_stack_frame; 1130 } else { 1131 BUG_ON(&initial_stack_frame != sf); 1132 } 1133 1134 return sf->error; 1135 } 1136 1137 static void btrfsic_read_from_block_data( 1138 struct btrfsic_block_data_ctx *block_ctx, 1139 void *dstv, u32 offset, size_t len) 1140 { 1141 size_t cur; 1142 size_t pgoff; 1143 char *kaddr; 1144 char *dst = (char *)dstv; 1145 size_t start_offset = offset_in_page(block_ctx->start); 1146 unsigned long i = (start_offset + offset) >> PAGE_SHIFT; 1147 1148 WARN_ON(offset + len > block_ctx->len); 1149 pgoff = offset_in_page(start_offset + offset); 1150 1151 while (len > 0) { 1152 cur = min(len, ((size_t)PAGE_SIZE - pgoff)); 1153 BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_SIZE)); 1154 kaddr = block_ctx->datav[i]; 1155 memcpy(dst, kaddr + pgoff, cur); 1156 1157 dst += cur; 1158 len -= cur; 1159 pgoff = 0; 1160 i++; 1161 } 1162 } 1163 1164 static int btrfsic_create_link_to_next_block( 1165 struct btrfsic_state *state, 1166 struct btrfsic_block *block, 1167 struct btrfsic_block_data_ctx *block_ctx, 1168 u64 next_bytenr, 1169 int limit_nesting, 1170 struct btrfsic_block_data_ctx *next_block_ctx, 1171 struct btrfsic_block **next_blockp, 1172 int force_iodone_flag, 1173 int *num_copiesp, int *mirror_nump, 1174 struct btrfs_disk_key *disk_key, 1175 u64 parent_generation) 1176 { 1177 struct btrfs_fs_info *fs_info = state->fs_info; 1178 struct btrfsic_block *next_block = NULL; 1179 int ret; 1180 struct btrfsic_block_link *l; 1181 int did_alloc_block_link; 1182 int block_was_created; 1183 1184 *next_blockp = NULL; 1185 if (0 == *num_copiesp) { 1186 *num_copiesp = btrfs_num_copies(fs_info, next_bytenr, 1187 state->metablock_size); 1188 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES) 1189 pr_info("num_copies(log_bytenr=%llu) = %d\n", 1190 next_bytenr, *num_copiesp); 1191 *mirror_nump = 1; 1192 } 1193 1194 if (*mirror_nump > *num_copiesp) 1195 return 0; 1196 1197 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1198 pr_info("btrfsic_create_link_to_next_block(mirror_num=%d)\n", 1199 *mirror_nump); 1200 ret = btrfsic_map_block(state, next_bytenr, 1201 state->metablock_size, 1202 next_block_ctx, *mirror_nump); 1203 if (ret) { 1204 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n", 1205 next_bytenr, *mirror_nump); 1206 btrfsic_release_block_ctx(next_block_ctx); 1207 *next_blockp = NULL; 1208 return -1; 1209 } 1210 1211 next_block = btrfsic_block_lookup_or_add(state, 1212 next_block_ctx, "referenced ", 1213 1, force_iodone_flag, 1214 !force_iodone_flag, 1215 *mirror_nump, 1216 &block_was_created); 1217 if (NULL == next_block) { 1218 btrfsic_release_block_ctx(next_block_ctx); 1219 *next_blockp = NULL; 1220 return -1; 1221 } 1222 if (block_was_created) { 1223 l = NULL; 1224 next_block->generation = BTRFSIC_GENERATION_UNKNOWN; 1225 } else { 1226 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) { 1227 if (next_block->logical_bytenr != next_bytenr && 1228 !(!next_block->is_metadata && 1229 0 == next_block->logical_bytenr)) 1230 pr_info( 1231 "referenced block @%llu (%pg/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu)\n", 1232 next_bytenr, next_block_ctx->dev->bdev, 1233 next_block_ctx->dev_bytenr, *mirror_nump, 1234 btrfsic_get_block_type(state, 1235 next_block), 1236 next_block->logical_bytenr); 1237 else 1238 pr_info( 1239 "referenced block @%llu (%pg/%llu/%d) found in hash table, %c\n", 1240 next_bytenr, next_block_ctx->dev->bdev, 1241 next_block_ctx->dev_bytenr, *mirror_nump, 1242 btrfsic_get_block_type(state, 1243 next_block)); 1244 } 1245 next_block->logical_bytenr = next_bytenr; 1246 1247 next_block->mirror_num = *mirror_nump; 1248 l = btrfsic_block_link_hashtable_lookup( 1249 next_block_ctx->dev->bdev, 1250 next_block_ctx->dev_bytenr, 1251 block_ctx->dev->bdev, 1252 block_ctx->dev_bytenr, 1253 &state->block_link_hashtable); 1254 } 1255 1256 next_block->disk_key = *disk_key; 1257 if (NULL == l) { 1258 l = btrfsic_block_link_alloc(); 1259 if (NULL == l) { 1260 btrfsic_release_block_ctx(next_block_ctx); 1261 *next_blockp = NULL; 1262 return -1; 1263 } 1264 1265 did_alloc_block_link = 1; 1266 l->block_ref_to = next_block; 1267 l->block_ref_from = block; 1268 l->ref_cnt = 1; 1269 l->parent_generation = parent_generation; 1270 1271 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1272 btrfsic_print_add_link(state, l); 1273 1274 list_add(&l->node_ref_to, &block->ref_to_list); 1275 list_add(&l->node_ref_from, &next_block->ref_from_list); 1276 1277 btrfsic_block_link_hashtable_add(l, 1278 &state->block_link_hashtable); 1279 } else { 1280 did_alloc_block_link = 0; 1281 if (0 == limit_nesting) { 1282 l->ref_cnt++; 1283 l->parent_generation = parent_generation; 1284 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1285 btrfsic_print_add_link(state, l); 1286 } 1287 } 1288 1289 if (limit_nesting > 0 && did_alloc_block_link) { 1290 ret = btrfsic_read_block(state, next_block_ctx); 1291 if (ret < (int)next_block_ctx->len) { 1292 pr_info("btrfsic: read block @logical %llu failed!\n", 1293 next_bytenr); 1294 btrfsic_release_block_ctx(next_block_ctx); 1295 *next_blockp = NULL; 1296 return -1; 1297 } 1298 1299 *next_blockp = next_block; 1300 } else { 1301 *next_blockp = NULL; 1302 } 1303 (*mirror_nump)++; 1304 1305 return 0; 1306 } 1307 1308 static int btrfsic_handle_extent_data( 1309 struct btrfsic_state *state, 1310 struct btrfsic_block *block, 1311 struct btrfsic_block_data_ctx *block_ctx, 1312 u32 item_offset, int force_iodone_flag) 1313 { 1314 struct btrfs_fs_info *fs_info = state->fs_info; 1315 struct btrfs_file_extent_item file_extent_item; 1316 u64 file_extent_item_offset; 1317 u64 next_bytenr; 1318 u64 num_bytes; 1319 u64 generation; 1320 struct btrfsic_block_link *l; 1321 int ret; 1322 1323 file_extent_item_offset = offsetof(struct btrfs_leaf, items) + 1324 item_offset; 1325 if (file_extent_item_offset + 1326 offsetof(struct btrfs_file_extent_item, disk_num_bytes) > 1327 block_ctx->len) { 1328 pr_info("btrfsic: file item out of bounce at logical %llu, dev %pg\n", 1329 block_ctx->start, block_ctx->dev->bdev); 1330 return -1; 1331 } 1332 1333 btrfsic_read_from_block_data(block_ctx, &file_extent_item, 1334 file_extent_item_offset, 1335 offsetof(struct btrfs_file_extent_item, disk_num_bytes)); 1336 if (BTRFS_FILE_EXTENT_REG != file_extent_item.type || 1337 btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) { 1338 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE) 1339 pr_info("extent_data: type %u, disk_bytenr = %llu\n", 1340 file_extent_item.type, 1341 btrfs_stack_file_extent_disk_bytenr( 1342 &file_extent_item)); 1343 return 0; 1344 } 1345 1346 if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) > 1347 block_ctx->len) { 1348 pr_info("btrfsic: file item out of bounce at logical %llu, dev %pg\n", 1349 block_ctx->start, block_ctx->dev->bdev); 1350 return -1; 1351 } 1352 btrfsic_read_from_block_data(block_ctx, &file_extent_item, 1353 file_extent_item_offset, 1354 sizeof(struct btrfs_file_extent_item)); 1355 next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item); 1356 if (btrfs_stack_file_extent_compression(&file_extent_item) == 1357 BTRFS_COMPRESS_NONE) { 1358 next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item); 1359 num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item); 1360 } else { 1361 num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item); 1362 } 1363 generation = btrfs_stack_file_extent_generation(&file_extent_item); 1364 1365 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE) 1366 pr_info("extent_data: type %u, disk_bytenr = %llu, offset = %llu, num_bytes = %llu\n", 1367 file_extent_item.type, 1368 btrfs_stack_file_extent_disk_bytenr(&file_extent_item), 1369 btrfs_stack_file_extent_offset(&file_extent_item), 1370 num_bytes); 1371 while (num_bytes > 0) { 1372 u32 chunk_len; 1373 int num_copies; 1374 int mirror_num; 1375 1376 if (num_bytes > state->datablock_size) 1377 chunk_len = state->datablock_size; 1378 else 1379 chunk_len = num_bytes; 1380 1381 num_copies = btrfs_num_copies(fs_info, next_bytenr, 1382 state->datablock_size); 1383 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES) 1384 pr_info("num_copies(log_bytenr=%llu) = %d\n", 1385 next_bytenr, num_copies); 1386 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) { 1387 struct btrfsic_block_data_ctx next_block_ctx; 1388 struct btrfsic_block *next_block; 1389 int block_was_created; 1390 1391 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1392 pr_info("btrfsic_handle_extent_data(mirror_num=%d)\n", 1393 mirror_num); 1394 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE) 1395 pr_info("\tdisk_bytenr = %llu, num_bytes %u\n", 1396 next_bytenr, chunk_len); 1397 ret = btrfsic_map_block(state, next_bytenr, 1398 chunk_len, &next_block_ctx, 1399 mirror_num); 1400 if (ret) { 1401 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n", 1402 next_bytenr, mirror_num); 1403 return -1; 1404 } 1405 1406 next_block = btrfsic_block_lookup_or_add( 1407 state, 1408 &next_block_ctx, 1409 "referenced ", 1410 0, 1411 force_iodone_flag, 1412 !force_iodone_flag, 1413 mirror_num, 1414 &block_was_created); 1415 if (NULL == next_block) { 1416 btrfsic_release_block_ctx(&next_block_ctx); 1417 return -1; 1418 } 1419 if (!block_was_created) { 1420 if ((state->print_mask & 1421 BTRFSIC_PRINT_MASK_VERBOSE) && 1422 next_block->logical_bytenr != next_bytenr && 1423 !(!next_block->is_metadata && 1424 0 == next_block->logical_bytenr)) { 1425 pr_info( 1426 "referenced block @%llu (%pg/%llu/%d) found in hash table, D, bytenr mismatch (!= stored %llu)\n", 1427 next_bytenr, 1428 next_block_ctx.dev->bdev, 1429 next_block_ctx.dev_bytenr, 1430 mirror_num, 1431 next_block->logical_bytenr); 1432 } 1433 next_block->logical_bytenr = next_bytenr; 1434 next_block->mirror_num = mirror_num; 1435 } 1436 1437 l = btrfsic_block_link_lookup_or_add(state, 1438 &next_block_ctx, 1439 next_block, block, 1440 generation); 1441 btrfsic_release_block_ctx(&next_block_ctx); 1442 if (NULL == l) 1443 return -1; 1444 } 1445 1446 next_bytenr += chunk_len; 1447 num_bytes -= chunk_len; 1448 } 1449 1450 return 0; 1451 } 1452 1453 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len, 1454 struct btrfsic_block_data_ctx *block_ctx_out, 1455 int mirror_num) 1456 { 1457 struct btrfs_fs_info *fs_info = state->fs_info; 1458 int ret; 1459 u64 length; 1460 struct btrfs_io_context *multi = NULL; 1461 struct btrfs_device *device; 1462 1463 length = len; 1464 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, 1465 bytenr, &length, &multi, mirror_num); 1466 1467 if (ret) { 1468 block_ctx_out->start = 0; 1469 block_ctx_out->dev_bytenr = 0; 1470 block_ctx_out->len = 0; 1471 block_ctx_out->dev = NULL; 1472 block_ctx_out->datav = NULL; 1473 block_ctx_out->pagev = NULL; 1474 block_ctx_out->mem_to_free = NULL; 1475 1476 return ret; 1477 } 1478 1479 device = multi->stripes[0].dev; 1480 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) || 1481 !device->bdev || !device->name) 1482 block_ctx_out->dev = NULL; 1483 else 1484 block_ctx_out->dev = btrfsic_dev_state_lookup( 1485 device->bdev->bd_dev); 1486 block_ctx_out->dev_bytenr = multi->stripes[0].physical; 1487 block_ctx_out->start = bytenr; 1488 block_ctx_out->len = len; 1489 block_ctx_out->datav = NULL; 1490 block_ctx_out->pagev = NULL; 1491 block_ctx_out->mem_to_free = NULL; 1492 1493 kfree(multi); 1494 if (NULL == block_ctx_out->dev) { 1495 ret = -ENXIO; 1496 pr_info("btrfsic: error, cannot lookup dev (#1)!\n"); 1497 } 1498 1499 return ret; 1500 } 1501 1502 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx) 1503 { 1504 if (block_ctx->mem_to_free) { 1505 unsigned int num_pages; 1506 1507 BUG_ON(!block_ctx->datav); 1508 BUG_ON(!block_ctx->pagev); 1509 num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >> 1510 PAGE_SHIFT; 1511 /* Pages must be unmapped in reverse order */ 1512 while (num_pages > 0) { 1513 num_pages--; 1514 if (block_ctx->datav[num_pages]) 1515 block_ctx->datav[num_pages] = NULL; 1516 if (block_ctx->pagev[num_pages]) { 1517 __free_page(block_ctx->pagev[num_pages]); 1518 block_ctx->pagev[num_pages] = NULL; 1519 } 1520 } 1521 1522 kfree(block_ctx->mem_to_free); 1523 block_ctx->mem_to_free = NULL; 1524 block_ctx->pagev = NULL; 1525 block_ctx->datav = NULL; 1526 } 1527 } 1528 1529 static int btrfsic_read_block(struct btrfsic_state *state, 1530 struct btrfsic_block_data_ctx *block_ctx) 1531 { 1532 unsigned int num_pages; 1533 unsigned int i; 1534 size_t size; 1535 u64 dev_bytenr; 1536 int ret; 1537 1538 BUG_ON(block_ctx->datav); 1539 BUG_ON(block_ctx->pagev); 1540 BUG_ON(block_ctx->mem_to_free); 1541 if (!PAGE_ALIGNED(block_ctx->dev_bytenr)) { 1542 pr_info("btrfsic: read_block() with unaligned bytenr %llu\n", 1543 block_ctx->dev_bytenr); 1544 return -1; 1545 } 1546 1547 num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >> 1548 PAGE_SHIFT; 1549 size = sizeof(*block_ctx->datav) + sizeof(*block_ctx->pagev); 1550 block_ctx->mem_to_free = kcalloc(num_pages, size, GFP_NOFS); 1551 if (!block_ctx->mem_to_free) 1552 return -ENOMEM; 1553 block_ctx->datav = block_ctx->mem_to_free; 1554 block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages); 1555 ret = btrfs_alloc_page_array(num_pages, block_ctx->pagev); 1556 if (ret) 1557 return ret; 1558 1559 dev_bytenr = block_ctx->dev_bytenr; 1560 for (i = 0; i < num_pages;) { 1561 struct bio *bio; 1562 unsigned int j; 1563 1564 bio = bio_alloc(block_ctx->dev->bdev, num_pages - i, 1565 REQ_OP_READ, GFP_NOFS); 1566 bio->bi_iter.bi_sector = dev_bytenr >> 9; 1567 1568 for (j = i; j < num_pages; j++) { 1569 ret = bio_add_page(bio, block_ctx->pagev[j], 1570 PAGE_SIZE, 0); 1571 if (PAGE_SIZE != ret) 1572 break; 1573 } 1574 if (j == i) { 1575 pr_info("btrfsic: error, failed to add a single page!\n"); 1576 return -1; 1577 } 1578 if (submit_bio_wait(bio)) { 1579 pr_info("btrfsic: read error at logical %llu dev %pg!\n", 1580 block_ctx->start, block_ctx->dev->bdev); 1581 bio_put(bio); 1582 return -1; 1583 } 1584 bio_put(bio); 1585 dev_bytenr += (j - i) * PAGE_SIZE; 1586 i = j; 1587 } 1588 for (i = 0; i < num_pages; i++) 1589 block_ctx->datav[i] = page_address(block_ctx->pagev[i]); 1590 1591 return block_ctx->len; 1592 } 1593 1594 static void btrfsic_dump_database(struct btrfsic_state *state) 1595 { 1596 const struct btrfsic_block *b_all; 1597 1598 BUG_ON(NULL == state); 1599 1600 pr_info("all_blocks_list:\n"); 1601 list_for_each_entry(b_all, &state->all_blocks_list, all_blocks_node) { 1602 const struct btrfsic_block_link *l; 1603 1604 pr_info("%c-block @%llu (%pg/%llu/%d)\n", 1605 btrfsic_get_block_type(state, b_all), 1606 b_all->logical_bytenr, b_all->dev_state->bdev, 1607 b_all->dev_bytenr, b_all->mirror_num); 1608 1609 list_for_each_entry(l, &b_all->ref_to_list, node_ref_to) { 1610 pr_info( 1611 " %c @%llu (%pg/%llu/%d) refers %u* to %c @%llu (%pg/%llu/%d)\n", 1612 btrfsic_get_block_type(state, b_all), 1613 b_all->logical_bytenr, b_all->dev_state->bdev, 1614 b_all->dev_bytenr, b_all->mirror_num, 1615 l->ref_cnt, 1616 btrfsic_get_block_type(state, l->block_ref_to), 1617 l->block_ref_to->logical_bytenr, 1618 l->block_ref_to->dev_state->bdev, 1619 l->block_ref_to->dev_bytenr, 1620 l->block_ref_to->mirror_num); 1621 } 1622 1623 list_for_each_entry(l, &b_all->ref_from_list, node_ref_from) { 1624 pr_info( 1625 " %c @%llu (%pg/%llu/%d) is ref %u* from %c @%llu (%pg/%llu/%d)\n", 1626 btrfsic_get_block_type(state, b_all), 1627 b_all->logical_bytenr, b_all->dev_state->bdev, 1628 b_all->dev_bytenr, b_all->mirror_num, 1629 l->ref_cnt, 1630 btrfsic_get_block_type(state, l->block_ref_from), 1631 l->block_ref_from->logical_bytenr, 1632 l->block_ref_from->dev_state->bdev, 1633 l->block_ref_from->dev_bytenr, 1634 l->block_ref_from->mirror_num); 1635 } 1636 1637 pr_info("\n"); 1638 } 1639 } 1640 1641 /* 1642 * Test whether the disk block contains a tree block (leaf or node) 1643 * (note that this test fails for the super block) 1644 */ 1645 static noinline_for_stack int btrfsic_test_for_metadata( 1646 struct btrfsic_state *state, 1647 char **datav, unsigned int num_pages) 1648 { 1649 struct btrfs_fs_info *fs_info = state->fs_info; 1650 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 1651 struct btrfs_header *h; 1652 u8 csum[BTRFS_CSUM_SIZE]; 1653 unsigned int i; 1654 1655 if (num_pages * PAGE_SIZE < state->metablock_size) 1656 return 1; /* not metadata */ 1657 num_pages = state->metablock_size >> PAGE_SHIFT; 1658 h = (struct btrfs_header *)datav[0]; 1659 1660 if (memcmp(h->fsid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE)) 1661 return 1; 1662 1663 shash->tfm = fs_info->csum_shash; 1664 crypto_shash_init(shash); 1665 1666 for (i = 0; i < num_pages; i++) { 1667 u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE); 1668 size_t sublen = i ? PAGE_SIZE : 1669 (PAGE_SIZE - BTRFS_CSUM_SIZE); 1670 1671 crypto_shash_update(shash, data, sublen); 1672 } 1673 crypto_shash_final(shash, csum); 1674 if (memcmp(csum, h->csum, fs_info->csum_size)) 1675 return 1; 1676 1677 return 0; /* is metadata */ 1678 } 1679 1680 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state, 1681 u64 dev_bytenr, char **mapped_datav, 1682 unsigned int num_pages, 1683 struct bio *bio, int *bio_is_patched, 1684 blk_opf_t submit_bio_bh_rw) 1685 { 1686 int is_metadata; 1687 struct btrfsic_block *block; 1688 struct btrfsic_block_data_ctx block_ctx; 1689 int ret; 1690 struct btrfsic_state *state = dev_state->state; 1691 struct block_device *bdev = dev_state->bdev; 1692 unsigned int processed_len; 1693 1694 if (NULL != bio_is_patched) 1695 *bio_is_patched = 0; 1696 1697 again: 1698 if (num_pages == 0) 1699 return; 1700 1701 processed_len = 0; 1702 is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav, 1703 num_pages)); 1704 1705 block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr, 1706 &state->block_hashtable); 1707 if (NULL != block) { 1708 u64 bytenr = 0; 1709 struct btrfsic_block_link *l, *tmp; 1710 1711 if (block->is_superblock) { 1712 bytenr = btrfs_super_bytenr((struct btrfs_super_block *) 1713 mapped_datav[0]); 1714 if (num_pages * PAGE_SIZE < 1715 BTRFS_SUPER_INFO_SIZE) { 1716 pr_info("btrfsic: cannot work with too short bios!\n"); 1717 return; 1718 } 1719 is_metadata = 1; 1720 BUG_ON(!PAGE_ALIGNED(BTRFS_SUPER_INFO_SIZE)); 1721 processed_len = BTRFS_SUPER_INFO_SIZE; 1722 if (state->print_mask & 1723 BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) { 1724 pr_info("[before new superblock is written]:\n"); 1725 btrfsic_dump_tree_sub(state, block, 0); 1726 } 1727 } 1728 if (is_metadata) { 1729 if (!block->is_superblock) { 1730 if (num_pages * PAGE_SIZE < 1731 state->metablock_size) { 1732 pr_info("btrfsic: cannot work with too short bios!\n"); 1733 return; 1734 } 1735 processed_len = state->metablock_size; 1736 bytenr = btrfs_stack_header_bytenr( 1737 (struct btrfs_header *) 1738 mapped_datav[0]); 1739 btrfsic_cmp_log_and_dev_bytenr(state, bytenr, 1740 dev_state, 1741 dev_bytenr); 1742 } 1743 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) { 1744 if (block->logical_bytenr != bytenr && 1745 !(!block->is_metadata && 1746 block->logical_bytenr == 0)) 1747 pr_info( 1748 "written block @%llu (%pg/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu)\n", 1749 bytenr, dev_state->bdev, 1750 dev_bytenr, 1751 block->mirror_num, 1752 btrfsic_get_block_type(state, 1753 block), 1754 block->logical_bytenr); 1755 else 1756 pr_info( 1757 "written block @%llu (%pg/%llu/%d) found in hash table, %c\n", 1758 bytenr, dev_state->bdev, 1759 dev_bytenr, block->mirror_num, 1760 btrfsic_get_block_type(state, 1761 block)); 1762 } 1763 block->logical_bytenr = bytenr; 1764 } else { 1765 if (num_pages * PAGE_SIZE < 1766 state->datablock_size) { 1767 pr_info("btrfsic: cannot work with too short bios!\n"); 1768 return; 1769 } 1770 processed_len = state->datablock_size; 1771 bytenr = block->logical_bytenr; 1772 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1773 pr_info( 1774 "written block @%llu (%pg/%llu/%d) found in hash table, %c\n", 1775 bytenr, dev_state->bdev, dev_bytenr, 1776 block->mirror_num, 1777 btrfsic_get_block_type(state, block)); 1778 } 1779 1780 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1781 pr_info("ref_to_list: %cE, ref_from_list: %cE\n", 1782 list_empty(&block->ref_to_list) ? ' ' : '!', 1783 list_empty(&block->ref_from_list) ? ' ' : '!'); 1784 if (btrfsic_is_block_ref_by_superblock(state, block, 0)) { 1785 pr_info( 1786 "btrfs: attempt to overwrite %c-block @%llu (%pg/%llu/%d), old(gen=%llu, objectid=%llu, type=%d, offset=%llu), new(gen=%llu), which is referenced by most recent superblock (superblockgen=%llu)!\n", 1787 btrfsic_get_block_type(state, block), bytenr, 1788 dev_state->bdev, dev_bytenr, block->mirror_num, 1789 block->generation, 1790 btrfs_disk_key_objectid(&block->disk_key), 1791 block->disk_key.type, 1792 btrfs_disk_key_offset(&block->disk_key), 1793 btrfs_stack_header_generation( 1794 (struct btrfs_header *) mapped_datav[0]), 1795 state->max_superblock_generation); 1796 btrfsic_dump_tree(state); 1797 } 1798 1799 if (!block->is_iodone && !block->never_written) { 1800 pr_info( 1801 "btrfs: attempt to overwrite %c-block @%llu (%pg/%llu/%d), oldgen=%llu, newgen=%llu, which is not yet iodone!\n", 1802 btrfsic_get_block_type(state, block), bytenr, 1803 dev_state->bdev, dev_bytenr, block->mirror_num, 1804 block->generation, 1805 btrfs_stack_header_generation( 1806 (struct btrfs_header *) 1807 mapped_datav[0])); 1808 /* it would not be safe to go on */ 1809 btrfsic_dump_tree(state); 1810 goto continue_loop; 1811 } 1812 1813 /* 1814 * Clear all references of this block. Do not free 1815 * the block itself even if is not referenced anymore 1816 * because it still carries valuable information 1817 * like whether it was ever written and IO completed. 1818 */ 1819 list_for_each_entry_safe(l, tmp, &block->ref_to_list, 1820 node_ref_to) { 1821 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1822 btrfsic_print_rem_link(state, l); 1823 l->ref_cnt--; 1824 if (0 == l->ref_cnt) { 1825 list_del(&l->node_ref_to); 1826 list_del(&l->node_ref_from); 1827 btrfsic_block_link_hashtable_remove(l); 1828 btrfsic_block_link_free(l); 1829 } 1830 } 1831 1832 block_ctx.dev = dev_state; 1833 block_ctx.dev_bytenr = dev_bytenr; 1834 block_ctx.start = bytenr; 1835 block_ctx.len = processed_len; 1836 block_ctx.pagev = NULL; 1837 block_ctx.mem_to_free = NULL; 1838 block_ctx.datav = mapped_datav; 1839 1840 if (is_metadata || state->include_extent_data) { 1841 block->never_written = 0; 1842 block->iodone_w_error = 0; 1843 if (NULL != bio) { 1844 block->is_iodone = 0; 1845 BUG_ON(NULL == bio_is_patched); 1846 if (!*bio_is_patched) { 1847 block->orig_bio_private = 1848 bio->bi_private; 1849 block->orig_bio_end_io = 1850 bio->bi_end_io; 1851 block->next_in_same_bio = NULL; 1852 bio->bi_private = block; 1853 bio->bi_end_io = btrfsic_bio_end_io; 1854 *bio_is_patched = 1; 1855 } else { 1856 struct btrfsic_block *chained_block = 1857 (struct btrfsic_block *) 1858 bio->bi_private; 1859 1860 BUG_ON(NULL == chained_block); 1861 block->orig_bio_private = 1862 chained_block->orig_bio_private; 1863 block->orig_bio_end_io = 1864 chained_block->orig_bio_end_io; 1865 block->next_in_same_bio = chained_block; 1866 bio->bi_private = block; 1867 } 1868 } else { 1869 block->is_iodone = 1; 1870 block->orig_bio_private = NULL; 1871 block->orig_bio_end_io = NULL; 1872 block->next_in_same_bio = NULL; 1873 } 1874 } 1875 1876 block->flush_gen = dev_state->last_flush_gen + 1; 1877 block->submit_bio_bh_rw = submit_bio_bh_rw; 1878 if (is_metadata) { 1879 block->logical_bytenr = bytenr; 1880 block->is_metadata = 1; 1881 if (block->is_superblock) { 1882 BUG_ON(PAGE_SIZE != 1883 BTRFS_SUPER_INFO_SIZE); 1884 ret = btrfsic_process_written_superblock( 1885 state, 1886 block, 1887 (struct btrfs_super_block *) 1888 mapped_datav[0]); 1889 if (state->print_mask & 1890 BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) { 1891 pr_info("[after new superblock is written]:\n"); 1892 btrfsic_dump_tree_sub(state, block, 0); 1893 } 1894 } else { 1895 block->mirror_num = 0; /* unknown */ 1896 ret = btrfsic_process_metablock( 1897 state, 1898 block, 1899 &block_ctx, 1900 0, 0); 1901 } 1902 if (ret) 1903 pr_info("btrfsic: btrfsic_process_metablock(root @%llu) failed!\n", 1904 dev_bytenr); 1905 } else { 1906 block->is_metadata = 0; 1907 block->mirror_num = 0; /* unknown */ 1908 block->generation = BTRFSIC_GENERATION_UNKNOWN; 1909 if (!state->include_extent_data 1910 && list_empty(&block->ref_from_list)) { 1911 /* 1912 * disk block is overwritten with extent 1913 * data (not meta data) and we are configured 1914 * to not include extent data: take the 1915 * chance and free the block's memory 1916 */ 1917 btrfsic_block_hashtable_remove(block); 1918 list_del(&block->all_blocks_node); 1919 btrfsic_block_free(block); 1920 } 1921 } 1922 btrfsic_release_block_ctx(&block_ctx); 1923 } else { 1924 /* block has not been found in hash table */ 1925 u64 bytenr; 1926 1927 if (!is_metadata) { 1928 processed_len = state->datablock_size; 1929 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1930 pr_info( 1931 "written block (%pg/%llu/?) !found in hash table, D\n", 1932 dev_state->bdev, dev_bytenr); 1933 if (!state->include_extent_data) { 1934 /* ignore that written D block */ 1935 goto continue_loop; 1936 } 1937 1938 /* this is getting ugly for the 1939 * include_extent_data case... */ 1940 bytenr = 0; /* unknown */ 1941 } else { 1942 processed_len = state->metablock_size; 1943 bytenr = btrfs_stack_header_bytenr( 1944 (struct btrfs_header *) 1945 mapped_datav[0]); 1946 btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state, 1947 dev_bytenr); 1948 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1949 pr_info( 1950 "written block @%llu (%pg/%llu/?) !found in hash table, M\n", 1951 bytenr, dev_state->bdev, dev_bytenr); 1952 } 1953 1954 block_ctx.dev = dev_state; 1955 block_ctx.dev_bytenr = dev_bytenr; 1956 block_ctx.start = bytenr; 1957 block_ctx.len = processed_len; 1958 block_ctx.pagev = NULL; 1959 block_ctx.mem_to_free = NULL; 1960 block_ctx.datav = mapped_datav; 1961 1962 block = btrfsic_block_alloc(); 1963 if (NULL == block) { 1964 btrfsic_release_block_ctx(&block_ctx); 1965 goto continue_loop; 1966 } 1967 block->dev_state = dev_state; 1968 block->dev_bytenr = dev_bytenr; 1969 block->logical_bytenr = bytenr; 1970 block->is_metadata = is_metadata; 1971 block->never_written = 0; 1972 block->iodone_w_error = 0; 1973 block->mirror_num = 0; /* unknown */ 1974 block->flush_gen = dev_state->last_flush_gen + 1; 1975 block->submit_bio_bh_rw = submit_bio_bh_rw; 1976 if (NULL != bio) { 1977 block->is_iodone = 0; 1978 BUG_ON(NULL == bio_is_patched); 1979 if (!*bio_is_patched) { 1980 block->orig_bio_private = bio->bi_private; 1981 block->orig_bio_end_io = bio->bi_end_io; 1982 block->next_in_same_bio = NULL; 1983 bio->bi_private = block; 1984 bio->bi_end_io = btrfsic_bio_end_io; 1985 *bio_is_patched = 1; 1986 } else { 1987 struct btrfsic_block *chained_block = 1988 (struct btrfsic_block *) 1989 bio->bi_private; 1990 1991 BUG_ON(NULL == chained_block); 1992 block->orig_bio_private = 1993 chained_block->orig_bio_private; 1994 block->orig_bio_end_io = 1995 chained_block->orig_bio_end_io; 1996 block->next_in_same_bio = chained_block; 1997 bio->bi_private = block; 1998 } 1999 } else { 2000 block->is_iodone = 1; 2001 block->orig_bio_private = NULL; 2002 block->orig_bio_end_io = NULL; 2003 block->next_in_same_bio = NULL; 2004 } 2005 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2006 pr_info("new written %c-block @%llu (%pg/%llu/%d)\n", 2007 is_metadata ? 'M' : 'D', 2008 block->logical_bytenr, block->dev_state->bdev, 2009 block->dev_bytenr, block->mirror_num); 2010 list_add(&block->all_blocks_node, &state->all_blocks_list); 2011 btrfsic_block_hashtable_add(block, &state->block_hashtable); 2012 2013 if (is_metadata) { 2014 ret = btrfsic_process_metablock(state, block, 2015 &block_ctx, 0, 0); 2016 if (ret) 2017 pr_info("btrfsic: process_metablock(root @%llu) failed!\n", 2018 dev_bytenr); 2019 } 2020 btrfsic_release_block_ctx(&block_ctx); 2021 } 2022 2023 continue_loop: 2024 BUG_ON(!processed_len); 2025 dev_bytenr += processed_len; 2026 mapped_datav += processed_len >> PAGE_SHIFT; 2027 num_pages -= processed_len >> PAGE_SHIFT; 2028 goto again; 2029 } 2030 2031 static void btrfsic_bio_end_io(struct bio *bp) 2032 { 2033 struct btrfsic_block *block = bp->bi_private; 2034 int iodone_w_error; 2035 2036 /* mutex is not held! This is not save if IO is not yet completed 2037 * on umount */ 2038 iodone_w_error = 0; 2039 if (bp->bi_status) 2040 iodone_w_error = 1; 2041 2042 BUG_ON(NULL == block); 2043 bp->bi_private = block->orig_bio_private; 2044 bp->bi_end_io = block->orig_bio_end_io; 2045 2046 do { 2047 struct btrfsic_block *next_block; 2048 struct btrfsic_dev_state *const dev_state = block->dev_state; 2049 2050 if ((dev_state->state->print_mask & 2051 BTRFSIC_PRINT_MASK_END_IO_BIO_BH)) 2052 pr_info("bio_end_io(err=%d) for %c @%llu (%pg/%llu/%d)\n", 2053 bp->bi_status, 2054 btrfsic_get_block_type(dev_state->state, block), 2055 block->logical_bytenr, dev_state->bdev, 2056 block->dev_bytenr, block->mirror_num); 2057 next_block = block->next_in_same_bio; 2058 block->iodone_w_error = iodone_w_error; 2059 if (block->submit_bio_bh_rw & REQ_PREFLUSH) { 2060 dev_state->last_flush_gen++; 2061 if ((dev_state->state->print_mask & 2062 BTRFSIC_PRINT_MASK_END_IO_BIO_BH)) 2063 pr_info("bio_end_io() new %pg flush_gen=%llu\n", 2064 dev_state->bdev, 2065 dev_state->last_flush_gen); 2066 } 2067 if (block->submit_bio_bh_rw & REQ_FUA) 2068 block->flush_gen = 0; /* FUA completed means block is 2069 * on disk */ 2070 block->is_iodone = 1; /* for FLUSH, this releases the block */ 2071 block = next_block; 2072 } while (NULL != block); 2073 2074 bp->bi_end_io(bp); 2075 } 2076 2077 static int btrfsic_process_written_superblock( 2078 struct btrfsic_state *state, 2079 struct btrfsic_block *const superblock, 2080 struct btrfs_super_block *const super_hdr) 2081 { 2082 struct btrfs_fs_info *fs_info = state->fs_info; 2083 int pass; 2084 2085 superblock->generation = btrfs_super_generation(super_hdr); 2086 if (!(superblock->generation > state->max_superblock_generation || 2087 0 == state->max_superblock_generation)) { 2088 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE) 2089 pr_info( 2090 "btrfsic: superblock @%llu (%pg/%llu/%d) with old gen %llu <= %llu\n", 2091 superblock->logical_bytenr, 2092 superblock->dev_state->bdev, 2093 superblock->dev_bytenr, superblock->mirror_num, 2094 btrfs_super_generation(super_hdr), 2095 state->max_superblock_generation); 2096 } else { 2097 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE) 2098 pr_info( 2099 "btrfsic: got new superblock @%llu (%pg/%llu/%d) with new gen %llu > %llu\n", 2100 superblock->logical_bytenr, 2101 superblock->dev_state->bdev, 2102 superblock->dev_bytenr, superblock->mirror_num, 2103 btrfs_super_generation(super_hdr), 2104 state->max_superblock_generation); 2105 2106 state->max_superblock_generation = 2107 btrfs_super_generation(super_hdr); 2108 state->latest_superblock = superblock; 2109 } 2110 2111 for (pass = 0; pass < 3; pass++) { 2112 int ret; 2113 u64 next_bytenr; 2114 struct btrfsic_block *next_block; 2115 struct btrfsic_block_data_ctx tmp_next_block_ctx; 2116 struct btrfsic_block_link *l; 2117 int num_copies; 2118 int mirror_num; 2119 const char *additional_string = NULL; 2120 struct btrfs_disk_key tmp_disk_key = {0}; 2121 2122 btrfs_set_disk_key_objectid(&tmp_disk_key, 2123 BTRFS_ROOT_ITEM_KEY); 2124 btrfs_set_disk_key_objectid(&tmp_disk_key, 0); 2125 2126 switch (pass) { 2127 case 0: 2128 btrfs_set_disk_key_objectid(&tmp_disk_key, 2129 BTRFS_ROOT_TREE_OBJECTID); 2130 additional_string = "root "; 2131 next_bytenr = btrfs_super_root(super_hdr); 2132 if (state->print_mask & 2133 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION) 2134 pr_info("root@%llu\n", next_bytenr); 2135 break; 2136 case 1: 2137 btrfs_set_disk_key_objectid(&tmp_disk_key, 2138 BTRFS_CHUNK_TREE_OBJECTID); 2139 additional_string = "chunk "; 2140 next_bytenr = btrfs_super_chunk_root(super_hdr); 2141 if (state->print_mask & 2142 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION) 2143 pr_info("chunk@%llu\n", next_bytenr); 2144 break; 2145 case 2: 2146 btrfs_set_disk_key_objectid(&tmp_disk_key, 2147 BTRFS_TREE_LOG_OBJECTID); 2148 additional_string = "log "; 2149 next_bytenr = btrfs_super_log_root(super_hdr); 2150 if (0 == next_bytenr) 2151 continue; 2152 if (state->print_mask & 2153 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION) 2154 pr_info("log@%llu\n", next_bytenr); 2155 break; 2156 } 2157 2158 num_copies = btrfs_num_copies(fs_info, next_bytenr, 2159 BTRFS_SUPER_INFO_SIZE); 2160 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES) 2161 pr_info("num_copies(log_bytenr=%llu) = %d\n", 2162 next_bytenr, num_copies); 2163 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) { 2164 int was_created; 2165 2166 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2167 pr_info("btrfsic_process_written_superblock(mirror_num=%d)\n", mirror_num); 2168 ret = btrfsic_map_block(state, next_bytenr, 2169 BTRFS_SUPER_INFO_SIZE, 2170 &tmp_next_block_ctx, 2171 mirror_num); 2172 if (ret) { 2173 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n", 2174 next_bytenr, mirror_num); 2175 return -1; 2176 } 2177 2178 next_block = btrfsic_block_lookup_or_add( 2179 state, 2180 &tmp_next_block_ctx, 2181 additional_string, 2182 1, 0, 1, 2183 mirror_num, 2184 &was_created); 2185 if (NULL == next_block) { 2186 btrfsic_release_block_ctx(&tmp_next_block_ctx); 2187 return -1; 2188 } 2189 2190 next_block->disk_key = tmp_disk_key; 2191 if (was_created) 2192 next_block->generation = 2193 BTRFSIC_GENERATION_UNKNOWN; 2194 l = btrfsic_block_link_lookup_or_add( 2195 state, 2196 &tmp_next_block_ctx, 2197 next_block, 2198 superblock, 2199 BTRFSIC_GENERATION_UNKNOWN); 2200 btrfsic_release_block_ctx(&tmp_next_block_ctx); 2201 if (NULL == l) 2202 return -1; 2203 } 2204 } 2205 2206 if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0))) 2207 btrfsic_dump_tree(state); 2208 2209 return 0; 2210 } 2211 2212 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state, 2213 struct btrfsic_block *const block, 2214 int recursion_level) 2215 { 2216 const struct btrfsic_block_link *l; 2217 int ret = 0; 2218 2219 if (recursion_level >= 3 + BTRFS_MAX_LEVEL) { 2220 /* 2221 * Note that this situation can happen and does not 2222 * indicate an error in regular cases. It happens 2223 * when disk blocks are freed and later reused. 2224 * The check-integrity module is not aware of any 2225 * block free operations, it just recognizes block 2226 * write operations. Therefore it keeps the linkage 2227 * information for a block until a block is 2228 * rewritten. This can temporarily cause incorrect 2229 * and even circular linkage information. This 2230 * causes no harm unless such blocks are referenced 2231 * by the most recent super block. 2232 */ 2233 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2234 pr_info("btrfsic: abort cyclic linkage (case 1).\n"); 2235 2236 return ret; 2237 } 2238 2239 /* 2240 * This algorithm is recursive because the amount of used stack 2241 * space is very small and the max recursion depth is limited. 2242 */ 2243 list_for_each_entry(l, &block->ref_to_list, node_ref_to) { 2244 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2245 pr_info( 2246 "rl=%d, %c @%llu (%pg/%llu/%d) %u* refers to %c @%llu (%pg/%llu/%d)\n", 2247 recursion_level, 2248 btrfsic_get_block_type(state, block), 2249 block->logical_bytenr, block->dev_state->bdev, 2250 block->dev_bytenr, block->mirror_num, 2251 l->ref_cnt, 2252 btrfsic_get_block_type(state, l->block_ref_to), 2253 l->block_ref_to->logical_bytenr, 2254 l->block_ref_to->dev_state->bdev, 2255 l->block_ref_to->dev_bytenr, 2256 l->block_ref_to->mirror_num); 2257 if (l->block_ref_to->never_written) { 2258 pr_info( 2259 "btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) which is never written!\n", 2260 btrfsic_get_block_type(state, l->block_ref_to), 2261 l->block_ref_to->logical_bytenr, 2262 l->block_ref_to->dev_state->bdev, 2263 l->block_ref_to->dev_bytenr, 2264 l->block_ref_to->mirror_num); 2265 ret = -1; 2266 } else if (!l->block_ref_to->is_iodone) { 2267 pr_info( 2268 "btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) which is not yet iodone!\n", 2269 btrfsic_get_block_type(state, l->block_ref_to), 2270 l->block_ref_to->logical_bytenr, 2271 l->block_ref_to->dev_state->bdev, 2272 l->block_ref_to->dev_bytenr, 2273 l->block_ref_to->mirror_num); 2274 ret = -1; 2275 } else if (l->block_ref_to->iodone_w_error) { 2276 pr_info( 2277 "btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) which has write error!\n", 2278 btrfsic_get_block_type(state, l->block_ref_to), 2279 l->block_ref_to->logical_bytenr, 2280 l->block_ref_to->dev_state->bdev, 2281 l->block_ref_to->dev_bytenr, 2282 l->block_ref_to->mirror_num); 2283 ret = -1; 2284 } else if (l->parent_generation != 2285 l->block_ref_to->generation && 2286 BTRFSIC_GENERATION_UNKNOWN != 2287 l->parent_generation && 2288 BTRFSIC_GENERATION_UNKNOWN != 2289 l->block_ref_to->generation) { 2290 pr_info( 2291 "btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) with generation %llu != parent generation %llu!\n", 2292 btrfsic_get_block_type(state, l->block_ref_to), 2293 l->block_ref_to->logical_bytenr, 2294 l->block_ref_to->dev_state->bdev, 2295 l->block_ref_to->dev_bytenr, 2296 l->block_ref_to->mirror_num, 2297 l->block_ref_to->generation, 2298 l->parent_generation); 2299 ret = -1; 2300 } else if (l->block_ref_to->flush_gen > 2301 l->block_ref_to->dev_state->last_flush_gen) { 2302 pr_info( 2303 "btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) which is not flushed out of disk's write cache (block flush_gen=%llu, dev->flush_gen=%llu)!\n", 2304 btrfsic_get_block_type(state, l->block_ref_to), 2305 l->block_ref_to->logical_bytenr, 2306 l->block_ref_to->dev_state->bdev, 2307 l->block_ref_to->dev_bytenr, 2308 l->block_ref_to->mirror_num, block->flush_gen, 2309 l->block_ref_to->dev_state->last_flush_gen); 2310 ret = -1; 2311 } else if (-1 == btrfsic_check_all_ref_blocks(state, 2312 l->block_ref_to, 2313 recursion_level + 2314 1)) { 2315 ret = -1; 2316 } 2317 } 2318 2319 return ret; 2320 } 2321 2322 static int btrfsic_is_block_ref_by_superblock( 2323 const struct btrfsic_state *state, 2324 const struct btrfsic_block *block, 2325 int recursion_level) 2326 { 2327 const struct btrfsic_block_link *l; 2328 2329 if (recursion_level >= 3 + BTRFS_MAX_LEVEL) { 2330 /* refer to comment at "abort cyclic linkage (case 1)" */ 2331 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2332 pr_info("btrfsic: abort cyclic linkage (case 2).\n"); 2333 2334 return 0; 2335 } 2336 2337 /* 2338 * This algorithm is recursive because the amount of used stack space 2339 * is very small and the max recursion depth is limited. 2340 */ 2341 list_for_each_entry(l, &block->ref_from_list, node_ref_from) { 2342 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2343 pr_info( 2344 "rl=%d, %c @%llu (%pg/%llu/%d) is ref %u* from %c @%llu (%pg/%llu/%d)\n", 2345 recursion_level, 2346 btrfsic_get_block_type(state, block), 2347 block->logical_bytenr, block->dev_state->bdev, 2348 block->dev_bytenr, block->mirror_num, 2349 l->ref_cnt, 2350 btrfsic_get_block_type(state, l->block_ref_from), 2351 l->block_ref_from->logical_bytenr, 2352 l->block_ref_from->dev_state->bdev, 2353 l->block_ref_from->dev_bytenr, 2354 l->block_ref_from->mirror_num); 2355 if (l->block_ref_from->is_superblock && 2356 state->latest_superblock->dev_bytenr == 2357 l->block_ref_from->dev_bytenr && 2358 state->latest_superblock->dev_state->bdev == 2359 l->block_ref_from->dev_state->bdev) 2360 return 1; 2361 else if (btrfsic_is_block_ref_by_superblock(state, 2362 l->block_ref_from, 2363 recursion_level + 2364 1)) 2365 return 1; 2366 } 2367 2368 return 0; 2369 } 2370 2371 static void btrfsic_print_add_link(const struct btrfsic_state *state, 2372 const struct btrfsic_block_link *l) 2373 { 2374 pr_info("add %u* link from %c @%llu (%pg/%llu/%d) to %c @%llu (%pg/%llu/%d)\n", 2375 l->ref_cnt, 2376 btrfsic_get_block_type(state, l->block_ref_from), 2377 l->block_ref_from->logical_bytenr, 2378 l->block_ref_from->dev_state->bdev, 2379 l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num, 2380 btrfsic_get_block_type(state, l->block_ref_to), 2381 l->block_ref_to->logical_bytenr, 2382 l->block_ref_to->dev_state->bdev, l->block_ref_to->dev_bytenr, 2383 l->block_ref_to->mirror_num); 2384 } 2385 2386 static void btrfsic_print_rem_link(const struct btrfsic_state *state, 2387 const struct btrfsic_block_link *l) 2388 { 2389 pr_info("rem %u* link from %c @%llu (%pg/%llu/%d) to %c @%llu (%pg/%llu/%d)\n", 2390 l->ref_cnt, 2391 btrfsic_get_block_type(state, l->block_ref_from), 2392 l->block_ref_from->logical_bytenr, 2393 l->block_ref_from->dev_state->bdev, 2394 l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num, 2395 btrfsic_get_block_type(state, l->block_ref_to), 2396 l->block_ref_to->logical_bytenr, 2397 l->block_ref_to->dev_state->bdev, l->block_ref_to->dev_bytenr, 2398 l->block_ref_to->mirror_num); 2399 } 2400 2401 static char btrfsic_get_block_type(const struct btrfsic_state *state, 2402 const struct btrfsic_block *block) 2403 { 2404 if (block->is_superblock && 2405 state->latest_superblock->dev_bytenr == block->dev_bytenr && 2406 state->latest_superblock->dev_state->bdev == block->dev_state->bdev) 2407 return 'S'; 2408 else if (block->is_superblock) 2409 return 's'; 2410 else if (block->is_metadata) 2411 return 'M'; 2412 else 2413 return 'D'; 2414 } 2415 2416 static void btrfsic_dump_tree(const struct btrfsic_state *state) 2417 { 2418 btrfsic_dump_tree_sub(state, state->latest_superblock, 0); 2419 } 2420 2421 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state, 2422 const struct btrfsic_block *block, 2423 int indent_level) 2424 { 2425 const struct btrfsic_block_link *l; 2426 int indent_add; 2427 static char buf[80]; 2428 int cursor_position; 2429 2430 /* 2431 * Should better fill an on-stack buffer with a complete line and 2432 * dump it at once when it is time to print a newline character. 2433 */ 2434 2435 /* 2436 * This algorithm is recursive because the amount of used stack space 2437 * is very small and the max recursion depth is limited. 2438 */ 2439 indent_add = sprintf(buf, "%c-%llu(%pg/%llu/%u)", 2440 btrfsic_get_block_type(state, block), 2441 block->logical_bytenr, block->dev_state->bdev, 2442 block->dev_bytenr, block->mirror_num); 2443 if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) { 2444 printk("[...]\n"); 2445 return; 2446 } 2447 printk(buf); 2448 indent_level += indent_add; 2449 if (list_empty(&block->ref_to_list)) { 2450 printk("\n"); 2451 return; 2452 } 2453 if (block->mirror_num > 1 && 2454 !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) { 2455 printk(" [...]\n"); 2456 return; 2457 } 2458 2459 cursor_position = indent_level; 2460 list_for_each_entry(l, &block->ref_to_list, node_ref_to) { 2461 while (cursor_position < indent_level) { 2462 printk(" "); 2463 cursor_position++; 2464 } 2465 if (l->ref_cnt > 1) 2466 indent_add = sprintf(buf, " %d*--> ", l->ref_cnt); 2467 else 2468 indent_add = sprintf(buf, " --> "); 2469 if (indent_level + indent_add > 2470 BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) { 2471 printk("[...]\n"); 2472 cursor_position = 0; 2473 continue; 2474 } 2475 2476 printk(buf); 2477 2478 btrfsic_dump_tree_sub(state, l->block_ref_to, 2479 indent_level + indent_add); 2480 cursor_position = 0; 2481 } 2482 } 2483 2484 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add( 2485 struct btrfsic_state *state, 2486 struct btrfsic_block_data_ctx *next_block_ctx, 2487 struct btrfsic_block *next_block, 2488 struct btrfsic_block *from_block, 2489 u64 parent_generation) 2490 { 2491 struct btrfsic_block_link *l; 2492 2493 l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev, 2494 next_block_ctx->dev_bytenr, 2495 from_block->dev_state->bdev, 2496 from_block->dev_bytenr, 2497 &state->block_link_hashtable); 2498 if (NULL == l) { 2499 l = btrfsic_block_link_alloc(); 2500 if (!l) 2501 return NULL; 2502 2503 l->block_ref_to = next_block; 2504 l->block_ref_from = from_block; 2505 l->ref_cnt = 1; 2506 l->parent_generation = parent_generation; 2507 2508 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2509 btrfsic_print_add_link(state, l); 2510 2511 list_add(&l->node_ref_to, &from_block->ref_to_list); 2512 list_add(&l->node_ref_from, &next_block->ref_from_list); 2513 2514 btrfsic_block_link_hashtable_add(l, 2515 &state->block_link_hashtable); 2516 } else { 2517 l->ref_cnt++; 2518 l->parent_generation = parent_generation; 2519 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2520 btrfsic_print_add_link(state, l); 2521 } 2522 2523 return l; 2524 } 2525 2526 static struct btrfsic_block *btrfsic_block_lookup_or_add( 2527 struct btrfsic_state *state, 2528 struct btrfsic_block_data_ctx *block_ctx, 2529 const char *additional_string, 2530 int is_metadata, 2531 int is_iodone, 2532 int never_written, 2533 int mirror_num, 2534 int *was_created) 2535 { 2536 struct btrfsic_block *block; 2537 2538 block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev, 2539 block_ctx->dev_bytenr, 2540 &state->block_hashtable); 2541 if (NULL == block) { 2542 struct btrfsic_dev_state *dev_state; 2543 2544 block = btrfsic_block_alloc(); 2545 if (!block) 2546 return NULL; 2547 2548 dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev->bd_dev); 2549 if (NULL == dev_state) { 2550 pr_info("btrfsic: error, lookup dev_state failed!\n"); 2551 btrfsic_block_free(block); 2552 return NULL; 2553 } 2554 block->dev_state = dev_state; 2555 block->dev_bytenr = block_ctx->dev_bytenr; 2556 block->logical_bytenr = block_ctx->start; 2557 block->is_metadata = is_metadata; 2558 block->is_iodone = is_iodone; 2559 block->never_written = never_written; 2560 block->mirror_num = mirror_num; 2561 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2562 pr_info("New %s%c-block @%llu (%pg/%llu/%d)\n", 2563 additional_string, 2564 btrfsic_get_block_type(state, block), 2565 block->logical_bytenr, dev_state->bdev, 2566 block->dev_bytenr, mirror_num); 2567 list_add(&block->all_blocks_node, &state->all_blocks_list); 2568 btrfsic_block_hashtable_add(block, &state->block_hashtable); 2569 if (NULL != was_created) 2570 *was_created = 1; 2571 } else { 2572 if (NULL != was_created) 2573 *was_created = 0; 2574 } 2575 2576 return block; 2577 } 2578 2579 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state, 2580 u64 bytenr, 2581 struct btrfsic_dev_state *dev_state, 2582 u64 dev_bytenr) 2583 { 2584 struct btrfs_fs_info *fs_info = state->fs_info; 2585 struct btrfsic_block_data_ctx block_ctx; 2586 int num_copies; 2587 int mirror_num; 2588 int match = 0; 2589 int ret; 2590 2591 num_copies = btrfs_num_copies(fs_info, bytenr, state->metablock_size); 2592 2593 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) { 2594 ret = btrfsic_map_block(state, bytenr, state->metablock_size, 2595 &block_ctx, mirror_num); 2596 if (ret) { 2597 pr_info("btrfsic: btrfsic_map_block(logical @%llu, mirror %d) failed!\n", 2598 bytenr, mirror_num); 2599 continue; 2600 } 2601 2602 if (dev_state->bdev == block_ctx.dev->bdev && 2603 dev_bytenr == block_ctx.dev_bytenr) { 2604 match++; 2605 btrfsic_release_block_ctx(&block_ctx); 2606 break; 2607 } 2608 btrfsic_release_block_ctx(&block_ctx); 2609 } 2610 2611 if (WARN_ON(!match)) { 2612 pr_info( 2613 "btrfs: attempt to write M-block which contains logical bytenr that doesn't map to dev+physical bytenr of submit_bio, buffer->log_bytenr=%llu, submit_bio(bdev=%pg, phys_bytenr=%llu)!\n", 2614 bytenr, dev_state->bdev, dev_bytenr); 2615 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) { 2616 ret = btrfsic_map_block(state, bytenr, 2617 state->metablock_size, 2618 &block_ctx, mirror_num); 2619 if (ret) 2620 continue; 2621 2622 pr_info("read logical bytenr @%llu maps to (%pg/%llu/%d)\n", 2623 bytenr, block_ctx.dev->bdev, 2624 block_ctx.dev_bytenr, mirror_num); 2625 } 2626 } 2627 } 2628 2629 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev) 2630 { 2631 return btrfsic_dev_state_hashtable_lookup(dev, 2632 &btrfsic_dev_state_hashtable); 2633 } 2634 2635 static void btrfsic_check_write_bio(struct bio *bio, struct btrfsic_dev_state *dev_state) 2636 { 2637 unsigned int segs = bio_segments(bio); 2638 u64 dev_bytenr = 512 * bio->bi_iter.bi_sector; 2639 u64 cur_bytenr = dev_bytenr; 2640 struct bvec_iter iter; 2641 struct bio_vec bvec; 2642 char **mapped_datav; 2643 int bio_is_patched = 0; 2644 int i = 0; 2645 2646 if (dev_state->state->print_mask & BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH) 2647 pr_info( 2648 "submit_bio(rw=%d,0x%x, bi_vcnt=%u, bi_sector=%llu (bytenr %llu), bi_bdev=%p)\n", 2649 bio_op(bio), bio->bi_opf, segs, 2650 bio->bi_iter.bi_sector, dev_bytenr, bio->bi_bdev); 2651 2652 mapped_datav = kmalloc_array(segs, sizeof(*mapped_datav), GFP_NOFS); 2653 if (!mapped_datav) 2654 return; 2655 2656 bio_for_each_segment(bvec, bio, iter) { 2657 BUG_ON(bvec.bv_len != PAGE_SIZE); 2658 mapped_datav[i] = page_address(bvec.bv_page); 2659 i++; 2660 2661 if (dev_state->state->print_mask & 2662 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE) 2663 pr_info("#%u: bytenr=%llu, len=%u, offset=%u\n", 2664 i, cur_bytenr, bvec.bv_len, bvec.bv_offset); 2665 cur_bytenr += bvec.bv_len; 2666 } 2667 2668 btrfsic_process_written_block(dev_state, dev_bytenr, mapped_datav, segs, 2669 bio, &bio_is_patched, bio->bi_opf); 2670 kfree(mapped_datav); 2671 } 2672 2673 static void btrfsic_check_flush_bio(struct bio *bio, struct btrfsic_dev_state *dev_state) 2674 { 2675 if (dev_state->state->print_mask & BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH) 2676 pr_info("submit_bio(rw=%d,0x%x FLUSH, bdev=%p)\n", 2677 bio_op(bio), bio->bi_opf, bio->bi_bdev); 2678 2679 if (dev_state->dummy_block_for_bio_bh_flush.is_iodone) { 2680 struct btrfsic_block *const block = 2681 &dev_state->dummy_block_for_bio_bh_flush; 2682 2683 block->is_iodone = 0; 2684 block->never_written = 0; 2685 block->iodone_w_error = 0; 2686 block->flush_gen = dev_state->last_flush_gen + 1; 2687 block->submit_bio_bh_rw = bio->bi_opf; 2688 block->orig_bio_private = bio->bi_private; 2689 block->orig_bio_end_io = bio->bi_end_io; 2690 block->next_in_same_bio = NULL; 2691 bio->bi_private = block; 2692 bio->bi_end_io = btrfsic_bio_end_io; 2693 } else if ((dev_state->state->print_mask & 2694 (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH | 2695 BTRFSIC_PRINT_MASK_VERBOSE))) { 2696 pr_info( 2697 "btrfsic_submit_bio(%pg) with FLUSH but dummy block already in use (ignored)!\n", 2698 dev_state->bdev); 2699 } 2700 } 2701 2702 void btrfsic_check_bio(struct bio *bio) 2703 { 2704 struct btrfsic_dev_state *dev_state; 2705 2706 if (!btrfsic_is_initialized) 2707 return; 2708 2709 /* 2710 * We can be called before btrfsic_mount, so there might not be a 2711 * dev_state. 2712 */ 2713 dev_state = btrfsic_dev_state_lookup(bio->bi_bdev->bd_dev); 2714 mutex_lock(&btrfsic_mutex); 2715 if (dev_state) { 2716 if (bio_op(bio) == REQ_OP_WRITE && bio_has_data(bio)) 2717 btrfsic_check_write_bio(bio, dev_state); 2718 else if (bio->bi_opf & REQ_PREFLUSH) 2719 btrfsic_check_flush_bio(bio, dev_state); 2720 } 2721 mutex_unlock(&btrfsic_mutex); 2722 } 2723 2724 int btrfsic_mount(struct btrfs_fs_info *fs_info, 2725 struct btrfs_fs_devices *fs_devices, 2726 int including_extent_data, u32 print_mask) 2727 { 2728 int ret; 2729 struct btrfsic_state *state; 2730 struct list_head *dev_head = &fs_devices->devices; 2731 struct btrfs_device *device; 2732 2733 if (!PAGE_ALIGNED(fs_info->nodesize)) { 2734 pr_info("btrfsic: cannot handle nodesize %d not being a multiple of PAGE_SIZE %ld!\n", 2735 fs_info->nodesize, PAGE_SIZE); 2736 return -1; 2737 } 2738 if (!PAGE_ALIGNED(fs_info->sectorsize)) { 2739 pr_info("btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_SIZE %ld!\n", 2740 fs_info->sectorsize, PAGE_SIZE); 2741 return -1; 2742 } 2743 state = kvzalloc(sizeof(*state), GFP_KERNEL); 2744 if (!state) 2745 return -ENOMEM; 2746 2747 if (!btrfsic_is_initialized) { 2748 mutex_init(&btrfsic_mutex); 2749 btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable); 2750 btrfsic_is_initialized = 1; 2751 } 2752 mutex_lock(&btrfsic_mutex); 2753 state->fs_info = fs_info; 2754 state->print_mask = print_mask; 2755 state->include_extent_data = including_extent_data; 2756 state->metablock_size = fs_info->nodesize; 2757 state->datablock_size = fs_info->sectorsize; 2758 INIT_LIST_HEAD(&state->all_blocks_list); 2759 btrfsic_block_hashtable_init(&state->block_hashtable); 2760 btrfsic_block_link_hashtable_init(&state->block_link_hashtable); 2761 state->max_superblock_generation = 0; 2762 state->latest_superblock = NULL; 2763 2764 list_for_each_entry(device, dev_head, dev_list) { 2765 struct btrfsic_dev_state *ds; 2766 2767 if (!device->bdev || !device->name) 2768 continue; 2769 2770 ds = btrfsic_dev_state_alloc(); 2771 if (NULL == ds) { 2772 mutex_unlock(&btrfsic_mutex); 2773 return -ENOMEM; 2774 } 2775 ds->bdev = device->bdev; 2776 ds->state = state; 2777 btrfsic_dev_state_hashtable_add(ds, 2778 &btrfsic_dev_state_hashtable); 2779 } 2780 2781 ret = btrfsic_process_superblock(state, fs_devices); 2782 if (0 != ret) { 2783 mutex_unlock(&btrfsic_mutex); 2784 btrfsic_unmount(fs_devices); 2785 return ret; 2786 } 2787 2788 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE) 2789 btrfsic_dump_database(state); 2790 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE) 2791 btrfsic_dump_tree(state); 2792 2793 mutex_unlock(&btrfsic_mutex); 2794 return 0; 2795 } 2796 2797 void btrfsic_unmount(struct btrfs_fs_devices *fs_devices) 2798 { 2799 struct btrfsic_block *b_all, *tmp_all; 2800 struct btrfsic_state *state; 2801 struct list_head *dev_head = &fs_devices->devices; 2802 struct btrfs_device *device; 2803 2804 if (!btrfsic_is_initialized) 2805 return; 2806 2807 mutex_lock(&btrfsic_mutex); 2808 2809 state = NULL; 2810 list_for_each_entry(device, dev_head, dev_list) { 2811 struct btrfsic_dev_state *ds; 2812 2813 if (!device->bdev || !device->name) 2814 continue; 2815 2816 ds = btrfsic_dev_state_hashtable_lookup( 2817 device->bdev->bd_dev, 2818 &btrfsic_dev_state_hashtable); 2819 if (NULL != ds) { 2820 state = ds->state; 2821 btrfsic_dev_state_hashtable_remove(ds); 2822 btrfsic_dev_state_free(ds); 2823 } 2824 } 2825 2826 if (NULL == state) { 2827 pr_info("btrfsic: error, cannot find state information on umount!\n"); 2828 mutex_unlock(&btrfsic_mutex); 2829 return; 2830 } 2831 2832 /* 2833 * Don't care about keeping the lists' state up to date, 2834 * just free all memory that was allocated dynamically. 2835 * Free the blocks and the block_links. 2836 */ 2837 list_for_each_entry_safe(b_all, tmp_all, &state->all_blocks_list, 2838 all_blocks_node) { 2839 struct btrfsic_block_link *l, *tmp; 2840 2841 list_for_each_entry_safe(l, tmp, &b_all->ref_to_list, 2842 node_ref_to) { 2843 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2844 btrfsic_print_rem_link(state, l); 2845 2846 l->ref_cnt--; 2847 if (0 == l->ref_cnt) 2848 btrfsic_block_link_free(l); 2849 } 2850 2851 if (b_all->is_iodone || b_all->never_written) 2852 btrfsic_block_free(b_all); 2853 else 2854 pr_info( 2855 "btrfs: attempt to free %c-block @%llu (%pg/%llu/%d) on umount which is not yet iodone!\n", 2856 btrfsic_get_block_type(state, b_all), 2857 b_all->logical_bytenr, b_all->dev_state->bdev, 2858 b_all->dev_bytenr, b_all->mirror_num); 2859 } 2860 2861 mutex_unlock(&btrfsic_mutex); 2862 2863 kvfree(state); 2864 } 2865