1 /* 2 * Copyright (C) STRATO AG 2011. All rights reserved. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public 6 * License v2 as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public 14 * License along with this program; if not, write to the 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 16 * Boston, MA 021110-1307, USA. 17 */ 18 19 /* 20 * This module can be used to catch cases when the btrfs kernel 21 * code executes write requests to the disk that bring the file 22 * system in an inconsistent state. In such a state, a power-loss 23 * or kernel panic event would cause that the data on disk is 24 * lost or at least damaged. 25 * 26 * Code is added that examines all block write requests during 27 * runtime (including writes of the super block). Three rules 28 * are verified and an error is printed on violation of the 29 * rules: 30 * 1. It is not allowed to write a disk block which is 31 * currently referenced by the super block (either directly 32 * or indirectly). 33 * 2. When a super block is written, it is verified that all 34 * referenced (directly or indirectly) blocks fulfill the 35 * following requirements: 36 * 2a. All referenced blocks have either been present when 37 * the file system was mounted, (i.e., they have been 38 * referenced by the super block) or they have been 39 * written since then and the write completion callback 40 * was called and no write error was indicated and a 41 * FLUSH request to the device where these blocks are 42 * located was received and completed. 43 * 2b. All referenced blocks need to have a generation 44 * number which is equal to the parent's number. 45 * 46 * One issue that was found using this module was that the log 47 * tree on disk became temporarily corrupted because disk blocks 48 * that had been in use for the log tree had been freed and 49 * reused too early, while being referenced by the written super 50 * block. 51 * 52 * The search term in the kernel log that can be used to filter 53 * on the existence of detected integrity issues is 54 * "btrfs: attempt". 55 * 56 * The integrity check is enabled via mount options. These 57 * mount options are only supported if the integrity check 58 * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY. 59 * 60 * Example #1, apply integrity checks to all metadata: 61 * mount /dev/sdb1 /mnt -o check_int 62 * 63 * Example #2, apply integrity checks to all metadata and 64 * to data extents: 65 * mount /dev/sdb1 /mnt -o check_int_data 66 * 67 * Example #3, apply integrity checks to all metadata and dump 68 * the tree that the super block references to kernel messages 69 * each time after a super block was written: 70 * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263 71 * 72 * If the integrity check tool is included and activated in 73 * the mount options, plenty of kernel memory is used, and 74 * plenty of additional CPU cycles are spent. Enabling this 75 * functionality is not intended for normal use. In most 76 * cases, unless you are a btrfs developer who needs to verify 77 * the integrity of (super)-block write requests, do not 78 * enable the config option BTRFS_FS_CHECK_INTEGRITY to 79 * include and compile the integrity check tool. 80 * 81 * Expect millions of lines of information in the kernel log with an 82 * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the 83 * kernel config to at least 26 (which is 64MB). Usually the value is 84 * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be 85 * changed like this before LOG_BUF_SHIFT can be set to a high value: 86 * config LOG_BUF_SHIFT 87 * int "Kernel log buffer size (16 => 64KB, 17 => 128KB)" 88 * range 12 30 89 */ 90 91 #include <linux/sched.h> 92 #include <linux/slab.h> 93 #include <linux/buffer_head.h> 94 #include <linux/mutex.h> 95 #include <linux/crc32c.h> 96 #include <linux/genhd.h> 97 #include <linux/blkdev.h> 98 #include "ctree.h" 99 #include "disk-io.h" 100 #include "transaction.h" 101 #include "extent_io.h" 102 #include "volumes.h" 103 #include "print-tree.h" 104 #include "locking.h" 105 #include "check-integrity.h" 106 #include "rcu-string.h" 107 108 #define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000 109 #define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000 110 #define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100 111 #define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051 112 #define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807 113 #define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530 114 #define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300 115 #define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6) /* in characters, 116 * excluding " [...]" */ 117 #define BTRFSIC_GENERATION_UNKNOWN ((u64)-1) 118 119 /* 120 * The definition of the bitmask fields for the print_mask. 121 * They are specified with the mount option check_integrity_print_mask. 122 */ 123 #define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE 0x00000001 124 #define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION 0x00000002 125 #define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE 0x00000004 126 #define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE 0x00000008 127 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH 0x00000010 128 #define BTRFSIC_PRINT_MASK_END_IO_BIO_BH 0x00000020 129 #define BTRFSIC_PRINT_MASK_VERBOSE 0x00000040 130 #define BTRFSIC_PRINT_MASK_VERY_VERBOSE 0x00000080 131 #define BTRFSIC_PRINT_MASK_INITIAL_TREE 0x00000100 132 #define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES 0x00000200 133 #define BTRFSIC_PRINT_MASK_INITIAL_DATABASE 0x00000400 134 #define BTRFSIC_PRINT_MASK_NUM_COPIES 0x00000800 135 #define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS 0x00001000 136 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE 0x00002000 137 138 struct btrfsic_dev_state; 139 struct btrfsic_state; 140 141 struct btrfsic_block { 142 u32 magic_num; /* only used for debug purposes */ 143 unsigned int is_metadata:1; /* if it is meta-data, not data-data */ 144 unsigned int is_superblock:1; /* if it is one of the superblocks */ 145 unsigned int is_iodone:1; /* if is done by lower subsystem */ 146 unsigned int iodone_w_error:1; /* error was indicated to endio */ 147 unsigned int never_written:1; /* block was added because it was 148 * referenced, not because it was 149 * written */ 150 unsigned int mirror_num; /* large enough to hold 151 * BTRFS_SUPER_MIRROR_MAX */ 152 struct btrfsic_dev_state *dev_state; 153 u64 dev_bytenr; /* key, physical byte num on disk */ 154 u64 logical_bytenr; /* logical byte num on disk */ 155 u64 generation; 156 struct btrfs_disk_key disk_key; /* extra info to print in case of 157 * issues, will not always be correct */ 158 struct list_head collision_resolving_node; /* list node */ 159 struct list_head all_blocks_node; /* list node */ 160 161 /* the following two lists contain block_link items */ 162 struct list_head ref_to_list; /* list */ 163 struct list_head ref_from_list; /* list */ 164 struct btrfsic_block *next_in_same_bio; 165 void *orig_bio_bh_private; 166 union { 167 bio_end_io_t *bio; 168 bh_end_io_t *bh; 169 } orig_bio_bh_end_io; 170 int submit_bio_bh_rw; 171 u64 flush_gen; /* only valid if !never_written */ 172 }; 173 174 /* 175 * Elements of this type are allocated dynamically and required because 176 * each block object can refer to and can be ref from multiple blocks. 177 * The key to lookup them in the hashtable is the dev_bytenr of 178 * the block ref to plus the one from the block refered from. 179 * The fact that they are searchable via a hashtable and that a 180 * ref_cnt is maintained is not required for the btrfs integrity 181 * check algorithm itself, it is only used to make the output more 182 * beautiful in case that an error is detected (an error is defined 183 * as a write operation to a block while that block is still referenced). 184 */ 185 struct btrfsic_block_link { 186 u32 magic_num; /* only used for debug purposes */ 187 u32 ref_cnt; 188 struct list_head node_ref_to; /* list node */ 189 struct list_head node_ref_from; /* list node */ 190 struct list_head collision_resolving_node; /* list node */ 191 struct btrfsic_block *block_ref_to; 192 struct btrfsic_block *block_ref_from; 193 u64 parent_generation; 194 }; 195 196 struct btrfsic_dev_state { 197 u32 magic_num; /* only used for debug purposes */ 198 struct block_device *bdev; 199 struct btrfsic_state *state; 200 struct list_head collision_resolving_node; /* list node */ 201 struct btrfsic_block dummy_block_for_bio_bh_flush; 202 u64 last_flush_gen; 203 char name[BDEVNAME_SIZE]; 204 }; 205 206 struct btrfsic_block_hashtable { 207 struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE]; 208 }; 209 210 struct btrfsic_block_link_hashtable { 211 struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE]; 212 }; 213 214 struct btrfsic_dev_state_hashtable { 215 struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE]; 216 }; 217 218 struct btrfsic_block_data_ctx { 219 u64 start; /* virtual bytenr */ 220 u64 dev_bytenr; /* physical bytenr on device */ 221 u32 len; 222 struct btrfsic_dev_state *dev; 223 char **datav; 224 struct page **pagev; 225 void *mem_to_free; 226 }; 227 228 /* This structure is used to implement recursion without occupying 229 * any stack space, refer to btrfsic_process_metablock() */ 230 struct btrfsic_stack_frame { 231 u32 magic; 232 u32 nr; 233 int error; 234 int i; 235 int limit_nesting; 236 int num_copies; 237 int mirror_num; 238 struct btrfsic_block *block; 239 struct btrfsic_block_data_ctx *block_ctx; 240 struct btrfsic_block *next_block; 241 struct btrfsic_block_data_ctx next_block_ctx; 242 struct btrfs_header *hdr; 243 struct btrfsic_stack_frame *prev; 244 }; 245 246 /* Some state per mounted filesystem */ 247 struct btrfsic_state { 248 u32 print_mask; 249 int include_extent_data; 250 int csum_size; 251 struct list_head all_blocks_list; 252 struct btrfsic_block_hashtable block_hashtable; 253 struct btrfsic_block_link_hashtable block_link_hashtable; 254 struct btrfs_root *root; 255 u64 max_superblock_generation; 256 struct btrfsic_block *latest_superblock; 257 u32 metablock_size; 258 u32 datablock_size; 259 }; 260 261 static void btrfsic_block_init(struct btrfsic_block *b); 262 static struct btrfsic_block *btrfsic_block_alloc(void); 263 static void btrfsic_block_free(struct btrfsic_block *b); 264 static void btrfsic_block_link_init(struct btrfsic_block_link *n); 265 static struct btrfsic_block_link *btrfsic_block_link_alloc(void); 266 static void btrfsic_block_link_free(struct btrfsic_block_link *n); 267 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds); 268 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void); 269 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds); 270 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h); 271 static void btrfsic_block_hashtable_add(struct btrfsic_block *b, 272 struct btrfsic_block_hashtable *h); 273 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b); 274 static struct btrfsic_block *btrfsic_block_hashtable_lookup( 275 struct block_device *bdev, 276 u64 dev_bytenr, 277 struct btrfsic_block_hashtable *h); 278 static void btrfsic_block_link_hashtable_init( 279 struct btrfsic_block_link_hashtable *h); 280 static void btrfsic_block_link_hashtable_add( 281 struct btrfsic_block_link *l, 282 struct btrfsic_block_link_hashtable *h); 283 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l); 284 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup( 285 struct block_device *bdev_ref_to, 286 u64 dev_bytenr_ref_to, 287 struct block_device *bdev_ref_from, 288 u64 dev_bytenr_ref_from, 289 struct btrfsic_block_link_hashtable *h); 290 static void btrfsic_dev_state_hashtable_init( 291 struct btrfsic_dev_state_hashtable *h); 292 static void btrfsic_dev_state_hashtable_add( 293 struct btrfsic_dev_state *ds, 294 struct btrfsic_dev_state_hashtable *h); 295 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds); 296 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup( 297 struct block_device *bdev, 298 struct btrfsic_dev_state_hashtable *h); 299 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void); 300 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf); 301 static int btrfsic_process_superblock(struct btrfsic_state *state, 302 struct btrfs_fs_devices *fs_devices); 303 static int btrfsic_process_metablock(struct btrfsic_state *state, 304 struct btrfsic_block *block, 305 struct btrfsic_block_data_ctx *block_ctx, 306 int limit_nesting, int force_iodone_flag); 307 static void btrfsic_read_from_block_data( 308 struct btrfsic_block_data_ctx *block_ctx, 309 void *dst, u32 offset, size_t len); 310 static int btrfsic_create_link_to_next_block( 311 struct btrfsic_state *state, 312 struct btrfsic_block *block, 313 struct btrfsic_block_data_ctx 314 *block_ctx, u64 next_bytenr, 315 int limit_nesting, 316 struct btrfsic_block_data_ctx *next_block_ctx, 317 struct btrfsic_block **next_blockp, 318 int force_iodone_flag, 319 int *num_copiesp, int *mirror_nump, 320 struct btrfs_disk_key *disk_key, 321 u64 parent_generation); 322 static int btrfsic_handle_extent_data(struct btrfsic_state *state, 323 struct btrfsic_block *block, 324 struct btrfsic_block_data_ctx *block_ctx, 325 u32 item_offset, int force_iodone_flag); 326 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len, 327 struct btrfsic_block_data_ctx *block_ctx_out, 328 int mirror_num); 329 static int btrfsic_map_superblock(struct btrfsic_state *state, u64 bytenr, 330 u32 len, struct block_device *bdev, 331 struct btrfsic_block_data_ctx *block_ctx_out); 332 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx); 333 static int btrfsic_read_block(struct btrfsic_state *state, 334 struct btrfsic_block_data_ctx *block_ctx); 335 static void btrfsic_dump_database(struct btrfsic_state *state); 336 static int btrfsic_test_for_metadata(struct btrfsic_state *state, 337 char **datav, unsigned int num_pages); 338 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state, 339 u64 dev_bytenr, char **mapped_datav, 340 unsigned int num_pages, 341 struct bio *bio, int *bio_is_patched, 342 struct buffer_head *bh, 343 int submit_bio_bh_rw); 344 static int btrfsic_process_written_superblock( 345 struct btrfsic_state *state, 346 struct btrfsic_block *const block, 347 struct btrfs_super_block *const super_hdr); 348 static void btrfsic_bio_end_io(struct bio *bp, int bio_error_status); 349 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate); 350 static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state, 351 const struct btrfsic_block *block, 352 int recursion_level); 353 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state, 354 struct btrfsic_block *const block, 355 int recursion_level); 356 static void btrfsic_print_add_link(const struct btrfsic_state *state, 357 const struct btrfsic_block_link *l); 358 static void btrfsic_print_rem_link(const struct btrfsic_state *state, 359 const struct btrfsic_block_link *l); 360 static char btrfsic_get_block_type(const struct btrfsic_state *state, 361 const struct btrfsic_block *block); 362 static void btrfsic_dump_tree(const struct btrfsic_state *state); 363 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state, 364 const struct btrfsic_block *block, 365 int indent_level); 366 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add( 367 struct btrfsic_state *state, 368 struct btrfsic_block_data_ctx *next_block_ctx, 369 struct btrfsic_block *next_block, 370 struct btrfsic_block *from_block, 371 u64 parent_generation); 372 static struct btrfsic_block *btrfsic_block_lookup_or_add( 373 struct btrfsic_state *state, 374 struct btrfsic_block_data_ctx *block_ctx, 375 const char *additional_string, 376 int is_metadata, 377 int is_iodone, 378 int never_written, 379 int mirror_num, 380 int *was_created); 381 static int btrfsic_process_superblock_dev_mirror( 382 struct btrfsic_state *state, 383 struct btrfsic_dev_state *dev_state, 384 struct btrfs_device *device, 385 int superblock_mirror_num, 386 struct btrfsic_dev_state **selected_dev_state, 387 struct btrfs_super_block *selected_super); 388 static struct btrfsic_dev_state *btrfsic_dev_state_lookup( 389 struct block_device *bdev); 390 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state, 391 u64 bytenr, 392 struct btrfsic_dev_state *dev_state, 393 u64 dev_bytenr); 394 395 static struct mutex btrfsic_mutex; 396 static int btrfsic_is_initialized; 397 static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable; 398 399 400 static void btrfsic_block_init(struct btrfsic_block *b) 401 { 402 b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER; 403 b->dev_state = NULL; 404 b->dev_bytenr = 0; 405 b->logical_bytenr = 0; 406 b->generation = BTRFSIC_GENERATION_UNKNOWN; 407 b->disk_key.objectid = 0; 408 b->disk_key.type = 0; 409 b->disk_key.offset = 0; 410 b->is_metadata = 0; 411 b->is_superblock = 0; 412 b->is_iodone = 0; 413 b->iodone_w_error = 0; 414 b->never_written = 0; 415 b->mirror_num = 0; 416 b->next_in_same_bio = NULL; 417 b->orig_bio_bh_private = NULL; 418 b->orig_bio_bh_end_io.bio = NULL; 419 INIT_LIST_HEAD(&b->collision_resolving_node); 420 INIT_LIST_HEAD(&b->all_blocks_node); 421 INIT_LIST_HEAD(&b->ref_to_list); 422 INIT_LIST_HEAD(&b->ref_from_list); 423 b->submit_bio_bh_rw = 0; 424 b->flush_gen = 0; 425 } 426 427 static struct btrfsic_block *btrfsic_block_alloc(void) 428 { 429 struct btrfsic_block *b; 430 431 b = kzalloc(sizeof(*b), GFP_NOFS); 432 if (NULL != b) 433 btrfsic_block_init(b); 434 435 return b; 436 } 437 438 static void btrfsic_block_free(struct btrfsic_block *b) 439 { 440 BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num)); 441 kfree(b); 442 } 443 444 static void btrfsic_block_link_init(struct btrfsic_block_link *l) 445 { 446 l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER; 447 l->ref_cnt = 1; 448 INIT_LIST_HEAD(&l->node_ref_to); 449 INIT_LIST_HEAD(&l->node_ref_from); 450 INIT_LIST_HEAD(&l->collision_resolving_node); 451 l->block_ref_to = NULL; 452 l->block_ref_from = NULL; 453 } 454 455 static struct btrfsic_block_link *btrfsic_block_link_alloc(void) 456 { 457 struct btrfsic_block_link *l; 458 459 l = kzalloc(sizeof(*l), GFP_NOFS); 460 if (NULL != l) 461 btrfsic_block_link_init(l); 462 463 return l; 464 } 465 466 static void btrfsic_block_link_free(struct btrfsic_block_link *l) 467 { 468 BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num)); 469 kfree(l); 470 } 471 472 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds) 473 { 474 ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER; 475 ds->bdev = NULL; 476 ds->state = NULL; 477 ds->name[0] = '\0'; 478 INIT_LIST_HEAD(&ds->collision_resolving_node); 479 ds->last_flush_gen = 0; 480 btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush); 481 ds->dummy_block_for_bio_bh_flush.is_iodone = 1; 482 ds->dummy_block_for_bio_bh_flush.dev_state = ds; 483 } 484 485 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void) 486 { 487 struct btrfsic_dev_state *ds; 488 489 ds = kzalloc(sizeof(*ds), GFP_NOFS); 490 if (NULL != ds) 491 btrfsic_dev_state_init(ds); 492 493 return ds; 494 } 495 496 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds) 497 { 498 BUG_ON(!(NULL == ds || 499 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num)); 500 kfree(ds); 501 } 502 503 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h) 504 { 505 int i; 506 507 for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++) 508 INIT_LIST_HEAD(h->table + i); 509 } 510 511 static void btrfsic_block_hashtable_add(struct btrfsic_block *b, 512 struct btrfsic_block_hashtable *h) 513 { 514 const unsigned int hashval = 515 (((unsigned int)(b->dev_bytenr >> 16)) ^ 516 ((unsigned int)((uintptr_t)b->dev_state->bdev))) & 517 (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1); 518 519 list_add(&b->collision_resolving_node, h->table + hashval); 520 } 521 522 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b) 523 { 524 list_del(&b->collision_resolving_node); 525 } 526 527 static struct btrfsic_block *btrfsic_block_hashtable_lookup( 528 struct block_device *bdev, 529 u64 dev_bytenr, 530 struct btrfsic_block_hashtable *h) 531 { 532 const unsigned int hashval = 533 (((unsigned int)(dev_bytenr >> 16)) ^ 534 ((unsigned int)((uintptr_t)bdev))) & 535 (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1); 536 struct list_head *elem; 537 538 list_for_each(elem, h->table + hashval) { 539 struct btrfsic_block *const b = 540 list_entry(elem, struct btrfsic_block, 541 collision_resolving_node); 542 543 if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr) 544 return b; 545 } 546 547 return NULL; 548 } 549 550 static void btrfsic_block_link_hashtable_init( 551 struct btrfsic_block_link_hashtable *h) 552 { 553 int i; 554 555 for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++) 556 INIT_LIST_HEAD(h->table + i); 557 } 558 559 static void btrfsic_block_link_hashtable_add( 560 struct btrfsic_block_link *l, 561 struct btrfsic_block_link_hashtable *h) 562 { 563 const unsigned int hashval = 564 (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^ 565 ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^ 566 ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^ 567 ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev))) 568 & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1); 569 570 BUG_ON(NULL == l->block_ref_to); 571 BUG_ON(NULL == l->block_ref_from); 572 list_add(&l->collision_resolving_node, h->table + hashval); 573 } 574 575 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l) 576 { 577 list_del(&l->collision_resolving_node); 578 } 579 580 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup( 581 struct block_device *bdev_ref_to, 582 u64 dev_bytenr_ref_to, 583 struct block_device *bdev_ref_from, 584 u64 dev_bytenr_ref_from, 585 struct btrfsic_block_link_hashtable *h) 586 { 587 const unsigned int hashval = 588 (((unsigned int)(dev_bytenr_ref_to >> 16)) ^ 589 ((unsigned int)(dev_bytenr_ref_from >> 16)) ^ 590 ((unsigned int)((uintptr_t)bdev_ref_to)) ^ 591 ((unsigned int)((uintptr_t)bdev_ref_from))) & 592 (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1); 593 struct list_head *elem; 594 595 list_for_each(elem, h->table + hashval) { 596 struct btrfsic_block_link *const l = 597 list_entry(elem, struct btrfsic_block_link, 598 collision_resolving_node); 599 600 BUG_ON(NULL == l->block_ref_to); 601 BUG_ON(NULL == l->block_ref_from); 602 if (l->block_ref_to->dev_state->bdev == bdev_ref_to && 603 l->block_ref_to->dev_bytenr == dev_bytenr_ref_to && 604 l->block_ref_from->dev_state->bdev == bdev_ref_from && 605 l->block_ref_from->dev_bytenr == dev_bytenr_ref_from) 606 return l; 607 } 608 609 return NULL; 610 } 611 612 static void btrfsic_dev_state_hashtable_init( 613 struct btrfsic_dev_state_hashtable *h) 614 { 615 int i; 616 617 for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++) 618 INIT_LIST_HEAD(h->table + i); 619 } 620 621 static void btrfsic_dev_state_hashtable_add( 622 struct btrfsic_dev_state *ds, 623 struct btrfsic_dev_state_hashtable *h) 624 { 625 const unsigned int hashval = 626 (((unsigned int)((uintptr_t)ds->bdev)) & 627 (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1)); 628 629 list_add(&ds->collision_resolving_node, h->table + hashval); 630 } 631 632 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds) 633 { 634 list_del(&ds->collision_resolving_node); 635 } 636 637 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup( 638 struct block_device *bdev, 639 struct btrfsic_dev_state_hashtable *h) 640 { 641 const unsigned int hashval = 642 (((unsigned int)((uintptr_t)bdev)) & 643 (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1)); 644 struct list_head *elem; 645 646 list_for_each(elem, h->table + hashval) { 647 struct btrfsic_dev_state *const ds = 648 list_entry(elem, struct btrfsic_dev_state, 649 collision_resolving_node); 650 651 if (ds->bdev == bdev) 652 return ds; 653 } 654 655 return NULL; 656 } 657 658 static int btrfsic_process_superblock(struct btrfsic_state *state, 659 struct btrfs_fs_devices *fs_devices) 660 { 661 int ret = 0; 662 struct btrfs_super_block *selected_super; 663 struct list_head *dev_head = &fs_devices->devices; 664 struct btrfs_device *device; 665 struct btrfsic_dev_state *selected_dev_state = NULL; 666 int pass; 667 668 BUG_ON(NULL == state); 669 selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS); 670 if (NULL == selected_super) { 671 printk(KERN_INFO "btrfsic: error, kmalloc failed!\n"); 672 return -1; 673 } 674 675 list_for_each_entry(device, dev_head, dev_list) { 676 int i; 677 struct btrfsic_dev_state *dev_state; 678 679 if (!device->bdev || !device->name) 680 continue; 681 682 dev_state = btrfsic_dev_state_lookup(device->bdev); 683 BUG_ON(NULL == dev_state); 684 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { 685 ret = btrfsic_process_superblock_dev_mirror( 686 state, dev_state, device, i, 687 &selected_dev_state, selected_super); 688 if (0 != ret && 0 == i) { 689 kfree(selected_super); 690 return ret; 691 } 692 } 693 } 694 695 if (NULL == state->latest_superblock) { 696 printk(KERN_INFO "btrfsic: no superblock found!\n"); 697 kfree(selected_super); 698 return -1; 699 } 700 701 state->csum_size = btrfs_super_csum_size(selected_super); 702 703 for (pass = 0; pass < 3; pass++) { 704 int num_copies; 705 int mirror_num; 706 u64 next_bytenr; 707 708 switch (pass) { 709 case 0: 710 next_bytenr = btrfs_super_root(selected_super); 711 if (state->print_mask & 712 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION) 713 printk(KERN_INFO "root@%llu\n", next_bytenr); 714 break; 715 case 1: 716 next_bytenr = btrfs_super_chunk_root(selected_super); 717 if (state->print_mask & 718 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION) 719 printk(KERN_INFO "chunk@%llu\n", next_bytenr); 720 break; 721 case 2: 722 next_bytenr = btrfs_super_log_root(selected_super); 723 if (0 == next_bytenr) 724 continue; 725 if (state->print_mask & 726 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION) 727 printk(KERN_INFO "log@%llu\n", next_bytenr); 728 break; 729 } 730 731 num_copies = 732 btrfs_num_copies(state->root->fs_info, 733 next_bytenr, state->metablock_size); 734 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES) 735 printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n", 736 next_bytenr, num_copies); 737 738 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) { 739 struct btrfsic_block *next_block; 740 struct btrfsic_block_data_ctx tmp_next_block_ctx; 741 struct btrfsic_block_link *l; 742 743 ret = btrfsic_map_block(state, next_bytenr, 744 state->metablock_size, 745 &tmp_next_block_ctx, 746 mirror_num); 747 if (ret) { 748 printk(KERN_INFO "btrfsic:" 749 " btrfsic_map_block(root @%llu," 750 " mirror %d) failed!\n", 751 next_bytenr, mirror_num); 752 kfree(selected_super); 753 return -1; 754 } 755 756 next_block = btrfsic_block_hashtable_lookup( 757 tmp_next_block_ctx.dev->bdev, 758 tmp_next_block_ctx.dev_bytenr, 759 &state->block_hashtable); 760 BUG_ON(NULL == next_block); 761 762 l = btrfsic_block_link_hashtable_lookup( 763 tmp_next_block_ctx.dev->bdev, 764 tmp_next_block_ctx.dev_bytenr, 765 state->latest_superblock->dev_state-> 766 bdev, 767 state->latest_superblock->dev_bytenr, 768 &state->block_link_hashtable); 769 BUG_ON(NULL == l); 770 771 ret = btrfsic_read_block(state, &tmp_next_block_ctx); 772 if (ret < (int)PAGE_CACHE_SIZE) { 773 printk(KERN_INFO 774 "btrfsic: read @logical %llu failed!\n", 775 tmp_next_block_ctx.start); 776 btrfsic_release_block_ctx(&tmp_next_block_ctx); 777 kfree(selected_super); 778 return -1; 779 } 780 781 ret = btrfsic_process_metablock(state, 782 next_block, 783 &tmp_next_block_ctx, 784 BTRFS_MAX_LEVEL + 3, 1); 785 btrfsic_release_block_ctx(&tmp_next_block_ctx); 786 } 787 } 788 789 kfree(selected_super); 790 return ret; 791 } 792 793 static int btrfsic_process_superblock_dev_mirror( 794 struct btrfsic_state *state, 795 struct btrfsic_dev_state *dev_state, 796 struct btrfs_device *device, 797 int superblock_mirror_num, 798 struct btrfsic_dev_state **selected_dev_state, 799 struct btrfs_super_block *selected_super) 800 { 801 struct btrfs_super_block *super_tmp; 802 u64 dev_bytenr; 803 struct buffer_head *bh; 804 struct btrfsic_block *superblock_tmp; 805 int pass; 806 struct block_device *const superblock_bdev = device->bdev; 807 808 /* super block bytenr is always the unmapped device bytenr */ 809 dev_bytenr = btrfs_sb_offset(superblock_mirror_num); 810 if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->total_bytes) 811 return -1; 812 bh = __bread(superblock_bdev, dev_bytenr / 4096, 813 BTRFS_SUPER_INFO_SIZE); 814 if (NULL == bh) 815 return -1; 816 super_tmp = (struct btrfs_super_block *) 817 (bh->b_data + (dev_bytenr & 4095)); 818 819 if (btrfs_super_bytenr(super_tmp) != dev_bytenr || 820 btrfs_super_magic(super_tmp) != BTRFS_MAGIC || 821 memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) || 822 btrfs_super_nodesize(super_tmp) != state->metablock_size || 823 btrfs_super_leafsize(super_tmp) != state->metablock_size || 824 btrfs_super_sectorsize(super_tmp) != state->datablock_size) { 825 brelse(bh); 826 return 0; 827 } 828 829 superblock_tmp = 830 btrfsic_block_hashtable_lookup(superblock_bdev, 831 dev_bytenr, 832 &state->block_hashtable); 833 if (NULL == superblock_tmp) { 834 superblock_tmp = btrfsic_block_alloc(); 835 if (NULL == superblock_tmp) { 836 printk(KERN_INFO "btrfsic: error, kmalloc failed!\n"); 837 brelse(bh); 838 return -1; 839 } 840 /* for superblock, only the dev_bytenr makes sense */ 841 superblock_tmp->dev_bytenr = dev_bytenr; 842 superblock_tmp->dev_state = dev_state; 843 superblock_tmp->logical_bytenr = dev_bytenr; 844 superblock_tmp->generation = btrfs_super_generation(super_tmp); 845 superblock_tmp->is_metadata = 1; 846 superblock_tmp->is_superblock = 1; 847 superblock_tmp->is_iodone = 1; 848 superblock_tmp->never_written = 0; 849 superblock_tmp->mirror_num = 1 + superblock_mirror_num; 850 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE) 851 printk_in_rcu(KERN_INFO "New initial S-block (bdev %p, %s)" 852 " @%llu (%s/%llu/%d)\n", 853 superblock_bdev, 854 rcu_str_deref(device->name), dev_bytenr, 855 dev_state->name, dev_bytenr, 856 superblock_mirror_num); 857 list_add(&superblock_tmp->all_blocks_node, 858 &state->all_blocks_list); 859 btrfsic_block_hashtable_add(superblock_tmp, 860 &state->block_hashtable); 861 } 862 863 /* select the one with the highest generation field */ 864 if (btrfs_super_generation(super_tmp) > 865 state->max_superblock_generation || 866 0 == state->max_superblock_generation) { 867 memcpy(selected_super, super_tmp, sizeof(*selected_super)); 868 *selected_dev_state = dev_state; 869 state->max_superblock_generation = 870 btrfs_super_generation(super_tmp); 871 state->latest_superblock = superblock_tmp; 872 } 873 874 for (pass = 0; pass < 3; pass++) { 875 u64 next_bytenr; 876 int num_copies; 877 int mirror_num; 878 const char *additional_string = NULL; 879 struct btrfs_disk_key tmp_disk_key; 880 881 tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY; 882 tmp_disk_key.offset = 0; 883 switch (pass) { 884 case 0: 885 btrfs_set_disk_key_objectid(&tmp_disk_key, 886 BTRFS_ROOT_TREE_OBJECTID); 887 additional_string = "initial root "; 888 next_bytenr = btrfs_super_root(super_tmp); 889 break; 890 case 1: 891 btrfs_set_disk_key_objectid(&tmp_disk_key, 892 BTRFS_CHUNK_TREE_OBJECTID); 893 additional_string = "initial chunk "; 894 next_bytenr = btrfs_super_chunk_root(super_tmp); 895 break; 896 case 2: 897 btrfs_set_disk_key_objectid(&tmp_disk_key, 898 BTRFS_TREE_LOG_OBJECTID); 899 additional_string = "initial log "; 900 next_bytenr = btrfs_super_log_root(super_tmp); 901 if (0 == next_bytenr) 902 continue; 903 break; 904 } 905 906 num_copies = 907 btrfs_num_copies(state->root->fs_info, 908 next_bytenr, state->metablock_size); 909 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES) 910 printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n", 911 next_bytenr, num_copies); 912 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) { 913 struct btrfsic_block *next_block; 914 struct btrfsic_block_data_ctx tmp_next_block_ctx; 915 struct btrfsic_block_link *l; 916 917 if (btrfsic_map_block(state, next_bytenr, 918 state->metablock_size, 919 &tmp_next_block_ctx, 920 mirror_num)) { 921 printk(KERN_INFO "btrfsic: btrfsic_map_block(" 922 "bytenr @%llu, mirror %d) failed!\n", 923 next_bytenr, mirror_num); 924 brelse(bh); 925 return -1; 926 } 927 928 next_block = btrfsic_block_lookup_or_add( 929 state, &tmp_next_block_ctx, 930 additional_string, 1, 1, 0, 931 mirror_num, NULL); 932 if (NULL == next_block) { 933 btrfsic_release_block_ctx(&tmp_next_block_ctx); 934 brelse(bh); 935 return -1; 936 } 937 938 next_block->disk_key = tmp_disk_key; 939 next_block->generation = BTRFSIC_GENERATION_UNKNOWN; 940 l = btrfsic_block_link_lookup_or_add( 941 state, &tmp_next_block_ctx, 942 next_block, superblock_tmp, 943 BTRFSIC_GENERATION_UNKNOWN); 944 btrfsic_release_block_ctx(&tmp_next_block_ctx); 945 if (NULL == l) { 946 brelse(bh); 947 return -1; 948 } 949 } 950 } 951 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES) 952 btrfsic_dump_tree_sub(state, superblock_tmp, 0); 953 954 brelse(bh); 955 return 0; 956 } 957 958 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void) 959 { 960 struct btrfsic_stack_frame *sf; 961 962 sf = kzalloc(sizeof(*sf), GFP_NOFS); 963 if (NULL == sf) 964 printk(KERN_INFO "btrfsic: alloc memory failed!\n"); 965 else 966 sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER; 967 return sf; 968 } 969 970 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf) 971 { 972 BUG_ON(!(NULL == sf || 973 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic)); 974 kfree(sf); 975 } 976 977 static int btrfsic_process_metablock( 978 struct btrfsic_state *state, 979 struct btrfsic_block *const first_block, 980 struct btrfsic_block_data_ctx *const first_block_ctx, 981 int first_limit_nesting, int force_iodone_flag) 982 { 983 struct btrfsic_stack_frame initial_stack_frame = { 0 }; 984 struct btrfsic_stack_frame *sf; 985 struct btrfsic_stack_frame *next_stack; 986 struct btrfs_header *const first_hdr = 987 (struct btrfs_header *)first_block_ctx->datav[0]; 988 989 BUG_ON(!first_hdr); 990 sf = &initial_stack_frame; 991 sf->error = 0; 992 sf->i = -1; 993 sf->limit_nesting = first_limit_nesting; 994 sf->block = first_block; 995 sf->block_ctx = first_block_ctx; 996 sf->next_block = NULL; 997 sf->hdr = first_hdr; 998 sf->prev = NULL; 999 1000 continue_with_new_stack_frame: 1001 sf->block->generation = le64_to_cpu(sf->hdr->generation); 1002 if (0 == sf->hdr->level) { 1003 struct btrfs_leaf *const leafhdr = 1004 (struct btrfs_leaf *)sf->hdr; 1005 1006 if (-1 == sf->i) { 1007 sf->nr = btrfs_stack_header_nritems(&leafhdr->header); 1008 1009 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1010 printk(KERN_INFO 1011 "leaf %llu items %d generation %llu" 1012 " owner %llu\n", 1013 sf->block_ctx->start, sf->nr, 1014 btrfs_stack_header_generation( 1015 &leafhdr->header), 1016 btrfs_stack_header_owner( 1017 &leafhdr->header)); 1018 } 1019 1020 continue_with_current_leaf_stack_frame: 1021 if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) { 1022 sf->i++; 1023 sf->num_copies = 0; 1024 } 1025 1026 if (sf->i < sf->nr) { 1027 struct btrfs_item disk_item; 1028 u32 disk_item_offset = 1029 (uintptr_t)(leafhdr->items + sf->i) - 1030 (uintptr_t)leafhdr; 1031 struct btrfs_disk_key *disk_key; 1032 u8 type; 1033 u32 item_offset; 1034 u32 item_size; 1035 1036 if (disk_item_offset + sizeof(struct btrfs_item) > 1037 sf->block_ctx->len) { 1038 leaf_item_out_of_bounce_error: 1039 printk(KERN_INFO 1040 "btrfsic: leaf item out of bounce at logical %llu, dev %s\n", 1041 sf->block_ctx->start, 1042 sf->block_ctx->dev->name); 1043 goto one_stack_frame_backwards; 1044 } 1045 btrfsic_read_from_block_data(sf->block_ctx, 1046 &disk_item, 1047 disk_item_offset, 1048 sizeof(struct btrfs_item)); 1049 item_offset = btrfs_stack_item_offset(&disk_item); 1050 item_size = btrfs_stack_item_size(&disk_item); 1051 disk_key = &disk_item.key; 1052 type = btrfs_disk_key_type(disk_key); 1053 1054 if (BTRFS_ROOT_ITEM_KEY == type) { 1055 struct btrfs_root_item root_item; 1056 u32 root_item_offset; 1057 u64 next_bytenr; 1058 1059 root_item_offset = item_offset + 1060 offsetof(struct btrfs_leaf, items); 1061 if (root_item_offset + item_size > 1062 sf->block_ctx->len) 1063 goto leaf_item_out_of_bounce_error; 1064 btrfsic_read_from_block_data( 1065 sf->block_ctx, &root_item, 1066 root_item_offset, 1067 item_size); 1068 next_bytenr = btrfs_root_bytenr(&root_item); 1069 1070 sf->error = 1071 btrfsic_create_link_to_next_block( 1072 state, 1073 sf->block, 1074 sf->block_ctx, 1075 next_bytenr, 1076 sf->limit_nesting, 1077 &sf->next_block_ctx, 1078 &sf->next_block, 1079 force_iodone_flag, 1080 &sf->num_copies, 1081 &sf->mirror_num, 1082 disk_key, 1083 btrfs_root_generation( 1084 &root_item)); 1085 if (sf->error) 1086 goto one_stack_frame_backwards; 1087 1088 if (NULL != sf->next_block) { 1089 struct btrfs_header *const next_hdr = 1090 (struct btrfs_header *) 1091 sf->next_block_ctx.datav[0]; 1092 1093 next_stack = 1094 btrfsic_stack_frame_alloc(); 1095 if (NULL == next_stack) { 1096 btrfsic_release_block_ctx( 1097 &sf-> 1098 next_block_ctx); 1099 goto one_stack_frame_backwards; 1100 } 1101 1102 next_stack->i = -1; 1103 next_stack->block = sf->next_block; 1104 next_stack->block_ctx = 1105 &sf->next_block_ctx; 1106 next_stack->next_block = NULL; 1107 next_stack->hdr = next_hdr; 1108 next_stack->limit_nesting = 1109 sf->limit_nesting - 1; 1110 next_stack->prev = sf; 1111 sf = next_stack; 1112 goto continue_with_new_stack_frame; 1113 } 1114 } else if (BTRFS_EXTENT_DATA_KEY == type && 1115 state->include_extent_data) { 1116 sf->error = btrfsic_handle_extent_data( 1117 state, 1118 sf->block, 1119 sf->block_ctx, 1120 item_offset, 1121 force_iodone_flag); 1122 if (sf->error) 1123 goto one_stack_frame_backwards; 1124 } 1125 1126 goto continue_with_current_leaf_stack_frame; 1127 } 1128 } else { 1129 struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr; 1130 1131 if (-1 == sf->i) { 1132 sf->nr = btrfs_stack_header_nritems(&nodehdr->header); 1133 1134 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1135 printk(KERN_INFO "node %llu level %d items %d" 1136 " generation %llu owner %llu\n", 1137 sf->block_ctx->start, 1138 nodehdr->header.level, sf->nr, 1139 btrfs_stack_header_generation( 1140 &nodehdr->header), 1141 btrfs_stack_header_owner( 1142 &nodehdr->header)); 1143 } 1144 1145 continue_with_current_node_stack_frame: 1146 if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) { 1147 sf->i++; 1148 sf->num_copies = 0; 1149 } 1150 1151 if (sf->i < sf->nr) { 1152 struct btrfs_key_ptr key_ptr; 1153 u32 key_ptr_offset; 1154 u64 next_bytenr; 1155 1156 key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) - 1157 (uintptr_t)nodehdr; 1158 if (key_ptr_offset + sizeof(struct btrfs_key_ptr) > 1159 sf->block_ctx->len) { 1160 printk(KERN_INFO 1161 "btrfsic: node item out of bounce at logical %llu, dev %s\n", 1162 sf->block_ctx->start, 1163 sf->block_ctx->dev->name); 1164 goto one_stack_frame_backwards; 1165 } 1166 btrfsic_read_from_block_data( 1167 sf->block_ctx, &key_ptr, key_ptr_offset, 1168 sizeof(struct btrfs_key_ptr)); 1169 next_bytenr = btrfs_stack_key_blockptr(&key_ptr); 1170 1171 sf->error = btrfsic_create_link_to_next_block( 1172 state, 1173 sf->block, 1174 sf->block_ctx, 1175 next_bytenr, 1176 sf->limit_nesting, 1177 &sf->next_block_ctx, 1178 &sf->next_block, 1179 force_iodone_flag, 1180 &sf->num_copies, 1181 &sf->mirror_num, 1182 &key_ptr.key, 1183 btrfs_stack_key_generation(&key_ptr)); 1184 if (sf->error) 1185 goto one_stack_frame_backwards; 1186 1187 if (NULL != sf->next_block) { 1188 struct btrfs_header *const next_hdr = 1189 (struct btrfs_header *) 1190 sf->next_block_ctx.datav[0]; 1191 1192 next_stack = btrfsic_stack_frame_alloc(); 1193 if (NULL == next_stack) 1194 goto one_stack_frame_backwards; 1195 1196 next_stack->i = -1; 1197 next_stack->block = sf->next_block; 1198 next_stack->block_ctx = &sf->next_block_ctx; 1199 next_stack->next_block = NULL; 1200 next_stack->hdr = next_hdr; 1201 next_stack->limit_nesting = 1202 sf->limit_nesting - 1; 1203 next_stack->prev = sf; 1204 sf = next_stack; 1205 goto continue_with_new_stack_frame; 1206 } 1207 1208 goto continue_with_current_node_stack_frame; 1209 } 1210 } 1211 1212 one_stack_frame_backwards: 1213 if (NULL != sf->prev) { 1214 struct btrfsic_stack_frame *const prev = sf->prev; 1215 1216 /* the one for the initial block is freed in the caller */ 1217 btrfsic_release_block_ctx(sf->block_ctx); 1218 1219 if (sf->error) { 1220 prev->error = sf->error; 1221 btrfsic_stack_frame_free(sf); 1222 sf = prev; 1223 goto one_stack_frame_backwards; 1224 } 1225 1226 btrfsic_stack_frame_free(sf); 1227 sf = prev; 1228 goto continue_with_new_stack_frame; 1229 } else { 1230 BUG_ON(&initial_stack_frame != sf); 1231 } 1232 1233 return sf->error; 1234 } 1235 1236 static void btrfsic_read_from_block_data( 1237 struct btrfsic_block_data_ctx *block_ctx, 1238 void *dstv, u32 offset, size_t len) 1239 { 1240 size_t cur; 1241 size_t offset_in_page; 1242 char *kaddr; 1243 char *dst = (char *)dstv; 1244 size_t start_offset = block_ctx->start & ((u64)PAGE_CACHE_SIZE - 1); 1245 unsigned long i = (start_offset + offset) >> PAGE_CACHE_SHIFT; 1246 1247 WARN_ON(offset + len > block_ctx->len); 1248 offset_in_page = (start_offset + offset) & (PAGE_CACHE_SIZE - 1); 1249 1250 while (len > 0) { 1251 cur = min(len, ((size_t)PAGE_CACHE_SIZE - offset_in_page)); 1252 BUG_ON(i >= (block_ctx->len + PAGE_CACHE_SIZE - 1) >> 1253 PAGE_CACHE_SHIFT); 1254 kaddr = block_ctx->datav[i]; 1255 memcpy(dst, kaddr + offset_in_page, cur); 1256 1257 dst += cur; 1258 len -= cur; 1259 offset_in_page = 0; 1260 i++; 1261 } 1262 } 1263 1264 static int btrfsic_create_link_to_next_block( 1265 struct btrfsic_state *state, 1266 struct btrfsic_block *block, 1267 struct btrfsic_block_data_ctx *block_ctx, 1268 u64 next_bytenr, 1269 int limit_nesting, 1270 struct btrfsic_block_data_ctx *next_block_ctx, 1271 struct btrfsic_block **next_blockp, 1272 int force_iodone_flag, 1273 int *num_copiesp, int *mirror_nump, 1274 struct btrfs_disk_key *disk_key, 1275 u64 parent_generation) 1276 { 1277 struct btrfsic_block *next_block = NULL; 1278 int ret; 1279 struct btrfsic_block_link *l; 1280 int did_alloc_block_link; 1281 int block_was_created; 1282 1283 *next_blockp = NULL; 1284 if (0 == *num_copiesp) { 1285 *num_copiesp = 1286 btrfs_num_copies(state->root->fs_info, 1287 next_bytenr, state->metablock_size); 1288 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES) 1289 printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n", 1290 next_bytenr, *num_copiesp); 1291 *mirror_nump = 1; 1292 } 1293 1294 if (*mirror_nump > *num_copiesp) 1295 return 0; 1296 1297 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1298 printk(KERN_INFO 1299 "btrfsic_create_link_to_next_block(mirror_num=%d)\n", 1300 *mirror_nump); 1301 ret = btrfsic_map_block(state, next_bytenr, 1302 state->metablock_size, 1303 next_block_ctx, *mirror_nump); 1304 if (ret) { 1305 printk(KERN_INFO 1306 "btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n", 1307 next_bytenr, *mirror_nump); 1308 btrfsic_release_block_ctx(next_block_ctx); 1309 *next_blockp = NULL; 1310 return -1; 1311 } 1312 1313 next_block = btrfsic_block_lookup_or_add(state, 1314 next_block_ctx, "referenced ", 1315 1, force_iodone_flag, 1316 !force_iodone_flag, 1317 *mirror_nump, 1318 &block_was_created); 1319 if (NULL == next_block) { 1320 btrfsic_release_block_ctx(next_block_ctx); 1321 *next_blockp = NULL; 1322 return -1; 1323 } 1324 if (block_was_created) { 1325 l = NULL; 1326 next_block->generation = BTRFSIC_GENERATION_UNKNOWN; 1327 } else { 1328 if (next_block->logical_bytenr != next_bytenr && 1329 !(!next_block->is_metadata && 1330 0 == next_block->logical_bytenr)) { 1331 printk(KERN_INFO 1332 "Referenced block @%llu (%s/%llu/%d)" 1333 " found in hash table, %c," 1334 " bytenr mismatch (!= stored %llu).\n", 1335 next_bytenr, next_block_ctx->dev->name, 1336 next_block_ctx->dev_bytenr, *mirror_nump, 1337 btrfsic_get_block_type(state, next_block), 1338 next_block->logical_bytenr); 1339 } else if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1340 printk(KERN_INFO 1341 "Referenced block @%llu (%s/%llu/%d)" 1342 " found in hash table, %c.\n", 1343 next_bytenr, next_block_ctx->dev->name, 1344 next_block_ctx->dev_bytenr, *mirror_nump, 1345 btrfsic_get_block_type(state, next_block)); 1346 next_block->logical_bytenr = next_bytenr; 1347 1348 next_block->mirror_num = *mirror_nump; 1349 l = btrfsic_block_link_hashtable_lookup( 1350 next_block_ctx->dev->bdev, 1351 next_block_ctx->dev_bytenr, 1352 block_ctx->dev->bdev, 1353 block_ctx->dev_bytenr, 1354 &state->block_link_hashtable); 1355 } 1356 1357 next_block->disk_key = *disk_key; 1358 if (NULL == l) { 1359 l = btrfsic_block_link_alloc(); 1360 if (NULL == l) { 1361 printk(KERN_INFO "btrfsic: error, kmalloc failed!\n"); 1362 btrfsic_release_block_ctx(next_block_ctx); 1363 *next_blockp = NULL; 1364 return -1; 1365 } 1366 1367 did_alloc_block_link = 1; 1368 l->block_ref_to = next_block; 1369 l->block_ref_from = block; 1370 l->ref_cnt = 1; 1371 l->parent_generation = parent_generation; 1372 1373 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1374 btrfsic_print_add_link(state, l); 1375 1376 list_add(&l->node_ref_to, &block->ref_to_list); 1377 list_add(&l->node_ref_from, &next_block->ref_from_list); 1378 1379 btrfsic_block_link_hashtable_add(l, 1380 &state->block_link_hashtable); 1381 } else { 1382 did_alloc_block_link = 0; 1383 if (0 == limit_nesting) { 1384 l->ref_cnt++; 1385 l->parent_generation = parent_generation; 1386 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1387 btrfsic_print_add_link(state, l); 1388 } 1389 } 1390 1391 if (limit_nesting > 0 && did_alloc_block_link) { 1392 ret = btrfsic_read_block(state, next_block_ctx); 1393 if (ret < (int)next_block_ctx->len) { 1394 printk(KERN_INFO 1395 "btrfsic: read block @logical %llu failed!\n", 1396 next_bytenr); 1397 btrfsic_release_block_ctx(next_block_ctx); 1398 *next_blockp = NULL; 1399 return -1; 1400 } 1401 1402 *next_blockp = next_block; 1403 } else { 1404 *next_blockp = NULL; 1405 } 1406 (*mirror_nump)++; 1407 1408 return 0; 1409 } 1410 1411 static int btrfsic_handle_extent_data( 1412 struct btrfsic_state *state, 1413 struct btrfsic_block *block, 1414 struct btrfsic_block_data_ctx *block_ctx, 1415 u32 item_offset, int force_iodone_flag) 1416 { 1417 int ret; 1418 struct btrfs_file_extent_item file_extent_item; 1419 u64 file_extent_item_offset; 1420 u64 next_bytenr; 1421 u64 num_bytes; 1422 u64 generation; 1423 struct btrfsic_block_link *l; 1424 1425 file_extent_item_offset = offsetof(struct btrfs_leaf, items) + 1426 item_offset; 1427 if (file_extent_item_offset + 1428 offsetof(struct btrfs_file_extent_item, disk_num_bytes) > 1429 block_ctx->len) { 1430 printk(KERN_INFO 1431 "btrfsic: file item out of bounce at logical %llu, dev %s\n", 1432 block_ctx->start, block_ctx->dev->name); 1433 return -1; 1434 } 1435 1436 btrfsic_read_from_block_data(block_ctx, &file_extent_item, 1437 file_extent_item_offset, 1438 offsetof(struct btrfs_file_extent_item, disk_num_bytes)); 1439 if (BTRFS_FILE_EXTENT_REG != file_extent_item.type || 1440 btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) { 1441 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE) 1442 printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu\n", 1443 file_extent_item.type, 1444 btrfs_stack_file_extent_disk_bytenr( 1445 &file_extent_item)); 1446 return 0; 1447 } 1448 1449 if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) > 1450 block_ctx->len) { 1451 printk(KERN_INFO 1452 "btrfsic: file item out of bounce at logical %llu, dev %s\n", 1453 block_ctx->start, block_ctx->dev->name); 1454 return -1; 1455 } 1456 btrfsic_read_from_block_data(block_ctx, &file_extent_item, 1457 file_extent_item_offset, 1458 sizeof(struct btrfs_file_extent_item)); 1459 next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item) + 1460 btrfs_stack_file_extent_offset(&file_extent_item); 1461 generation = btrfs_stack_file_extent_generation(&file_extent_item); 1462 num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item); 1463 generation = btrfs_stack_file_extent_generation(&file_extent_item); 1464 1465 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE) 1466 printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu," 1467 " offset = %llu, num_bytes = %llu\n", 1468 file_extent_item.type, 1469 btrfs_stack_file_extent_disk_bytenr(&file_extent_item), 1470 btrfs_stack_file_extent_offset(&file_extent_item), 1471 num_bytes); 1472 while (num_bytes > 0) { 1473 u32 chunk_len; 1474 int num_copies; 1475 int mirror_num; 1476 1477 if (num_bytes > state->datablock_size) 1478 chunk_len = state->datablock_size; 1479 else 1480 chunk_len = num_bytes; 1481 1482 num_copies = 1483 btrfs_num_copies(state->root->fs_info, 1484 next_bytenr, state->datablock_size); 1485 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES) 1486 printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n", 1487 next_bytenr, num_copies); 1488 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) { 1489 struct btrfsic_block_data_ctx next_block_ctx; 1490 struct btrfsic_block *next_block; 1491 int block_was_created; 1492 1493 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1494 printk(KERN_INFO "btrfsic_handle_extent_data(" 1495 "mirror_num=%d)\n", mirror_num); 1496 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE) 1497 printk(KERN_INFO 1498 "\tdisk_bytenr = %llu, num_bytes %u\n", 1499 next_bytenr, chunk_len); 1500 ret = btrfsic_map_block(state, next_bytenr, 1501 chunk_len, &next_block_ctx, 1502 mirror_num); 1503 if (ret) { 1504 printk(KERN_INFO 1505 "btrfsic: btrfsic_map_block(@%llu," 1506 " mirror=%d) failed!\n", 1507 next_bytenr, mirror_num); 1508 return -1; 1509 } 1510 1511 next_block = btrfsic_block_lookup_or_add( 1512 state, 1513 &next_block_ctx, 1514 "referenced ", 1515 0, 1516 force_iodone_flag, 1517 !force_iodone_flag, 1518 mirror_num, 1519 &block_was_created); 1520 if (NULL == next_block) { 1521 printk(KERN_INFO 1522 "btrfsic: error, kmalloc failed!\n"); 1523 btrfsic_release_block_ctx(&next_block_ctx); 1524 return -1; 1525 } 1526 if (!block_was_created) { 1527 if (next_block->logical_bytenr != next_bytenr && 1528 !(!next_block->is_metadata && 1529 0 == next_block->logical_bytenr)) { 1530 printk(KERN_INFO 1531 "Referenced block" 1532 " @%llu (%s/%llu/%d)" 1533 " found in hash table, D," 1534 " bytenr mismatch" 1535 " (!= stored %llu).\n", 1536 next_bytenr, 1537 next_block_ctx.dev->name, 1538 next_block_ctx.dev_bytenr, 1539 mirror_num, 1540 next_block->logical_bytenr); 1541 } 1542 next_block->logical_bytenr = next_bytenr; 1543 next_block->mirror_num = mirror_num; 1544 } 1545 1546 l = btrfsic_block_link_lookup_or_add(state, 1547 &next_block_ctx, 1548 next_block, block, 1549 generation); 1550 btrfsic_release_block_ctx(&next_block_ctx); 1551 if (NULL == l) 1552 return -1; 1553 } 1554 1555 next_bytenr += chunk_len; 1556 num_bytes -= chunk_len; 1557 } 1558 1559 return 0; 1560 } 1561 1562 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len, 1563 struct btrfsic_block_data_ctx *block_ctx_out, 1564 int mirror_num) 1565 { 1566 int ret; 1567 u64 length; 1568 struct btrfs_bio *multi = NULL; 1569 struct btrfs_device *device; 1570 1571 length = len; 1572 ret = btrfs_map_block(state->root->fs_info, READ, 1573 bytenr, &length, &multi, mirror_num); 1574 1575 if (ret) { 1576 block_ctx_out->start = 0; 1577 block_ctx_out->dev_bytenr = 0; 1578 block_ctx_out->len = 0; 1579 block_ctx_out->dev = NULL; 1580 block_ctx_out->datav = NULL; 1581 block_ctx_out->pagev = NULL; 1582 block_ctx_out->mem_to_free = NULL; 1583 1584 return ret; 1585 } 1586 1587 device = multi->stripes[0].dev; 1588 block_ctx_out->dev = btrfsic_dev_state_lookup(device->bdev); 1589 block_ctx_out->dev_bytenr = multi->stripes[0].physical; 1590 block_ctx_out->start = bytenr; 1591 block_ctx_out->len = len; 1592 block_ctx_out->datav = NULL; 1593 block_ctx_out->pagev = NULL; 1594 block_ctx_out->mem_to_free = NULL; 1595 1596 kfree(multi); 1597 if (NULL == block_ctx_out->dev) { 1598 ret = -ENXIO; 1599 printk(KERN_INFO "btrfsic: error, cannot lookup dev (#1)!\n"); 1600 } 1601 1602 return ret; 1603 } 1604 1605 static int btrfsic_map_superblock(struct btrfsic_state *state, u64 bytenr, 1606 u32 len, struct block_device *bdev, 1607 struct btrfsic_block_data_ctx *block_ctx_out) 1608 { 1609 block_ctx_out->dev = btrfsic_dev_state_lookup(bdev); 1610 block_ctx_out->dev_bytenr = bytenr; 1611 block_ctx_out->start = bytenr; 1612 block_ctx_out->len = len; 1613 block_ctx_out->datav = NULL; 1614 block_ctx_out->pagev = NULL; 1615 block_ctx_out->mem_to_free = NULL; 1616 if (NULL != block_ctx_out->dev) { 1617 return 0; 1618 } else { 1619 printk(KERN_INFO "btrfsic: error, cannot lookup dev (#2)!\n"); 1620 return -ENXIO; 1621 } 1622 } 1623 1624 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx) 1625 { 1626 if (block_ctx->mem_to_free) { 1627 unsigned int num_pages; 1628 1629 BUG_ON(!block_ctx->datav); 1630 BUG_ON(!block_ctx->pagev); 1631 num_pages = (block_ctx->len + (u64)PAGE_CACHE_SIZE - 1) >> 1632 PAGE_CACHE_SHIFT; 1633 while (num_pages > 0) { 1634 num_pages--; 1635 if (block_ctx->datav[num_pages]) { 1636 kunmap(block_ctx->pagev[num_pages]); 1637 block_ctx->datav[num_pages] = NULL; 1638 } 1639 if (block_ctx->pagev[num_pages]) { 1640 __free_page(block_ctx->pagev[num_pages]); 1641 block_ctx->pagev[num_pages] = NULL; 1642 } 1643 } 1644 1645 kfree(block_ctx->mem_to_free); 1646 block_ctx->mem_to_free = NULL; 1647 block_ctx->pagev = NULL; 1648 block_ctx->datav = NULL; 1649 } 1650 } 1651 1652 static int btrfsic_read_block(struct btrfsic_state *state, 1653 struct btrfsic_block_data_ctx *block_ctx) 1654 { 1655 unsigned int num_pages; 1656 unsigned int i; 1657 u64 dev_bytenr; 1658 int ret; 1659 1660 BUG_ON(block_ctx->datav); 1661 BUG_ON(block_ctx->pagev); 1662 BUG_ON(block_ctx->mem_to_free); 1663 if (block_ctx->dev_bytenr & ((u64)PAGE_CACHE_SIZE - 1)) { 1664 printk(KERN_INFO 1665 "btrfsic: read_block() with unaligned bytenr %llu\n", 1666 block_ctx->dev_bytenr); 1667 return -1; 1668 } 1669 1670 num_pages = (block_ctx->len + (u64)PAGE_CACHE_SIZE - 1) >> 1671 PAGE_CACHE_SHIFT; 1672 block_ctx->mem_to_free = kzalloc((sizeof(*block_ctx->datav) + 1673 sizeof(*block_ctx->pagev)) * 1674 num_pages, GFP_NOFS); 1675 if (!block_ctx->mem_to_free) 1676 return -1; 1677 block_ctx->datav = block_ctx->mem_to_free; 1678 block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages); 1679 for (i = 0; i < num_pages; i++) { 1680 block_ctx->pagev[i] = alloc_page(GFP_NOFS); 1681 if (!block_ctx->pagev[i]) 1682 return -1; 1683 } 1684 1685 dev_bytenr = block_ctx->dev_bytenr; 1686 for (i = 0; i < num_pages;) { 1687 struct bio *bio; 1688 unsigned int j; 1689 1690 bio = btrfs_io_bio_alloc(GFP_NOFS, num_pages - i); 1691 if (!bio) { 1692 printk(KERN_INFO 1693 "btrfsic: bio_alloc() for %u pages failed!\n", 1694 num_pages - i); 1695 return -1; 1696 } 1697 bio->bi_bdev = block_ctx->dev->bdev; 1698 bio->bi_sector = dev_bytenr >> 9; 1699 1700 for (j = i; j < num_pages; j++) { 1701 ret = bio_add_page(bio, block_ctx->pagev[j], 1702 PAGE_CACHE_SIZE, 0); 1703 if (PAGE_CACHE_SIZE != ret) 1704 break; 1705 } 1706 if (j == i) { 1707 printk(KERN_INFO 1708 "btrfsic: error, failed to add a single page!\n"); 1709 return -1; 1710 } 1711 if (submit_bio_wait(READ, bio)) { 1712 printk(KERN_INFO 1713 "btrfsic: read error at logical %llu dev %s!\n", 1714 block_ctx->start, block_ctx->dev->name); 1715 bio_put(bio); 1716 return -1; 1717 } 1718 bio_put(bio); 1719 dev_bytenr += (j - i) * PAGE_CACHE_SIZE; 1720 i = j; 1721 } 1722 for (i = 0; i < num_pages; i++) { 1723 block_ctx->datav[i] = kmap(block_ctx->pagev[i]); 1724 if (!block_ctx->datav[i]) { 1725 printk(KERN_INFO "btrfsic: kmap() failed (dev %s)!\n", 1726 block_ctx->dev->name); 1727 return -1; 1728 } 1729 } 1730 1731 return block_ctx->len; 1732 } 1733 1734 static void btrfsic_dump_database(struct btrfsic_state *state) 1735 { 1736 struct list_head *elem_all; 1737 1738 BUG_ON(NULL == state); 1739 1740 printk(KERN_INFO "all_blocks_list:\n"); 1741 list_for_each(elem_all, &state->all_blocks_list) { 1742 const struct btrfsic_block *const b_all = 1743 list_entry(elem_all, struct btrfsic_block, 1744 all_blocks_node); 1745 struct list_head *elem_ref_to; 1746 struct list_head *elem_ref_from; 1747 1748 printk(KERN_INFO "%c-block @%llu (%s/%llu/%d)\n", 1749 btrfsic_get_block_type(state, b_all), 1750 b_all->logical_bytenr, b_all->dev_state->name, 1751 b_all->dev_bytenr, b_all->mirror_num); 1752 1753 list_for_each(elem_ref_to, &b_all->ref_to_list) { 1754 const struct btrfsic_block_link *const l = 1755 list_entry(elem_ref_to, 1756 struct btrfsic_block_link, 1757 node_ref_to); 1758 1759 printk(KERN_INFO " %c @%llu (%s/%llu/%d)" 1760 " refers %u* to" 1761 " %c @%llu (%s/%llu/%d)\n", 1762 btrfsic_get_block_type(state, b_all), 1763 b_all->logical_bytenr, b_all->dev_state->name, 1764 b_all->dev_bytenr, b_all->mirror_num, 1765 l->ref_cnt, 1766 btrfsic_get_block_type(state, l->block_ref_to), 1767 l->block_ref_to->logical_bytenr, 1768 l->block_ref_to->dev_state->name, 1769 l->block_ref_to->dev_bytenr, 1770 l->block_ref_to->mirror_num); 1771 } 1772 1773 list_for_each(elem_ref_from, &b_all->ref_from_list) { 1774 const struct btrfsic_block_link *const l = 1775 list_entry(elem_ref_from, 1776 struct btrfsic_block_link, 1777 node_ref_from); 1778 1779 printk(KERN_INFO " %c @%llu (%s/%llu/%d)" 1780 " is ref %u* from" 1781 " %c @%llu (%s/%llu/%d)\n", 1782 btrfsic_get_block_type(state, b_all), 1783 b_all->logical_bytenr, b_all->dev_state->name, 1784 b_all->dev_bytenr, b_all->mirror_num, 1785 l->ref_cnt, 1786 btrfsic_get_block_type(state, l->block_ref_from), 1787 l->block_ref_from->logical_bytenr, 1788 l->block_ref_from->dev_state->name, 1789 l->block_ref_from->dev_bytenr, 1790 l->block_ref_from->mirror_num); 1791 } 1792 1793 printk(KERN_INFO "\n"); 1794 } 1795 } 1796 1797 /* 1798 * Test whether the disk block contains a tree block (leaf or node) 1799 * (note that this test fails for the super block) 1800 */ 1801 static int btrfsic_test_for_metadata(struct btrfsic_state *state, 1802 char **datav, unsigned int num_pages) 1803 { 1804 struct btrfs_header *h; 1805 u8 csum[BTRFS_CSUM_SIZE]; 1806 u32 crc = ~(u32)0; 1807 unsigned int i; 1808 1809 if (num_pages * PAGE_CACHE_SIZE < state->metablock_size) 1810 return 1; /* not metadata */ 1811 num_pages = state->metablock_size >> PAGE_CACHE_SHIFT; 1812 h = (struct btrfs_header *)datav[0]; 1813 1814 if (memcmp(h->fsid, state->root->fs_info->fsid, BTRFS_UUID_SIZE)) 1815 return 1; 1816 1817 for (i = 0; i < num_pages; i++) { 1818 u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE); 1819 size_t sublen = i ? PAGE_CACHE_SIZE : 1820 (PAGE_CACHE_SIZE - BTRFS_CSUM_SIZE); 1821 1822 crc = crc32c(crc, data, sublen); 1823 } 1824 btrfs_csum_final(crc, csum); 1825 if (memcmp(csum, h->csum, state->csum_size)) 1826 return 1; 1827 1828 return 0; /* is metadata */ 1829 } 1830 1831 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state, 1832 u64 dev_bytenr, char **mapped_datav, 1833 unsigned int num_pages, 1834 struct bio *bio, int *bio_is_patched, 1835 struct buffer_head *bh, 1836 int submit_bio_bh_rw) 1837 { 1838 int is_metadata; 1839 struct btrfsic_block *block; 1840 struct btrfsic_block_data_ctx block_ctx; 1841 int ret; 1842 struct btrfsic_state *state = dev_state->state; 1843 struct block_device *bdev = dev_state->bdev; 1844 unsigned int processed_len; 1845 1846 if (NULL != bio_is_patched) 1847 *bio_is_patched = 0; 1848 1849 again: 1850 if (num_pages == 0) 1851 return; 1852 1853 processed_len = 0; 1854 is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav, 1855 num_pages)); 1856 1857 block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr, 1858 &state->block_hashtable); 1859 if (NULL != block) { 1860 u64 bytenr = 0; 1861 struct list_head *elem_ref_to; 1862 struct list_head *tmp_ref_to; 1863 1864 if (block->is_superblock) { 1865 bytenr = btrfs_super_bytenr((struct btrfs_super_block *) 1866 mapped_datav[0]); 1867 if (num_pages * PAGE_CACHE_SIZE < 1868 BTRFS_SUPER_INFO_SIZE) { 1869 printk(KERN_INFO 1870 "btrfsic: cannot work with too short bios!\n"); 1871 return; 1872 } 1873 is_metadata = 1; 1874 BUG_ON(BTRFS_SUPER_INFO_SIZE & (PAGE_CACHE_SIZE - 1)); 1875 processed_len = BTRFS_SUPER_INFO_SIZE; 1876 if (state->print_mask & 1877 BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) { 1878 printk(KERN_INFO 1879 "[before new superblock is written]:\n"); 1880 btrfsic_dump_tree_sub(state, block, 0); 1881 } 1882 } 1883 if (is_metadata) { 1884 if (!block->is_superblock) { 1885 if (num_pages * PAGE_CACHE_SIZE < 1886 state->metablock_size) { 1887 printk(KERN_INFO 1888 "btrfsic: cannot work with too short bios!\n"); 1889 return; 1890 } 1891 processed_len = state->metablock_size; 1892 bytenr = btrfs_stack_header_bytenr( 1893 (struct btrfs_header *) 1894 mapped_datav[0]); 1895 btrfsic_cmp_log_and_dev_bytenr(state, bytenr, 1896 dev_state, 1897 dev_bytenr); 1898 } 1899 if (block->logical_bytenr != bytenr && 1900 !(!block->is_metadata && 1901 block->logical_bytenr == 0)) 1902 printk(KERN_INFO 1903 "Written block @%llu (%s/%llu/%d)" 1904 " found in hash table, %c," 1905 " bytenr mismatch" 1906 " (!= stored %llu).\n", 1907 bytenr, dev_state->name, dev_bytenr, 1908 block->mirror_num, 1909 btrfsic_get_block_type(state, block), 1910 block->logical_bytenr); 1911 else if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1912 printk(KERN_INFO 1913 "Written block @%llu (%s/%llu/%d)" 1914 " found in hash table, %c.\n", 1915 bytenr, dev_state->name, dev_bytenr, 1916 block->mirror_num, 1917 btrfsic_get_block_type(state, block)); 1918 block->logical_bytenr = bytenr; 1919 } else { 1920 if (num_pages * PAGE_CACHE_SIZE < 1921 state->datablock_size) { 1922 printk(KERN_INFO 1923 "btrfsic: cannot work with too short bios!\n"); 1924 return; 1925 } 1926 processed_len = state->datablock_size; 1927 bytenr = block->logical_bytenr; 1928 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1929 printk(KERN_INFO 1930 "Written block @%llu (%s/%llu/%d)" 1931 " found in hash table, %c.\n", 1932 bytenr, dev_state->name, dev_bytenr, 1933 block->mirror_num, 1934 btrfsic_get_block_type(state, block)); 1935 } 1936 1937 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1938 printk(KERN_INFO 1939 "ref_to_list: %cE, ref_from_list: %cE\n", 1940 list_empty(&block->ref_to_list) ? ' ' : '!', 1941 list_empty(&block->ref_from_list) ? ' ' : '!'); 1942 if (btrfsic_is_block_ref_by_superblock(state, block, 0)) { 1943 printk(KERN_INFO "btrfs: attempt to overwrite %c-block" 1944 " @%llu (%s/%llu/%d), old(gen=%llu," 1945 " objectid=%llu, type=%d, offset=%llu)," 1946 " new(gen=%llu)," 1947 " which is referenced by most recent superblock" 1948 " (superblockgen=%llu)!\n", 1949 btrfsic_get_block_type(state, block), bytenr, 1950 dev_state->name, dev_bytenr, block->mirror_num, 1951 block->generation, 1952 btrfs_disk_key_objectid(&block->disk_key), 1953 block->disk_key.type, 1954 btrfs_disk_key_offset(&block->disk_key), 1955 btrfs_stack_header_generation( 1956 (struct btrfs_header *) mapped_datav[0]), 1957 state->max_superblock_generation); 1958 btrfsic_dump_tree(state); 1959 } 1960 1961 if (!block->is_iodone && !block->never_written) { 1962 printk(KERN_INFO "btrfs: attempt to overwrite %c-block" 1963 " @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu," 1964 " which is not yet iodone!\n", 1965 btrfsic_get_block_type(state, block), bytenr, 1966 dev_state->name, dev_bytenr, block->mirror_num, 1967 block->generation, 1968 btrfs_stack_header_generation( 1969 (struct btrfs_header *) 1970 mapped_datav[0])); 1971 /* it would not be safe to go on */ 1972 btrfsic_dump_tree(state); 1973 goto continue_loop; 1974 } 1975 1976 /* 1977 * Clear all references of this block. Do not free 1978 * the block itself even if is not referenced anymore 1979 * because it still carries valueable information 1980 * like whether it was ever written and IO completed. 1981 */ 1982 list_for_each_safe(elem_ref_to, tmp_ref_to, 1983 &block->ref_to_list) { 1984 struct btrfsic_block_link *const l = 1985 list_entry(elem_ref_to, 1986 struct btrfsic_block_link, 1987 node_ref_to); 1988 1989 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 1990 btrfsic_print_rem_link(state, l); 1991 l->ref_cnt--; 1992 if (0 == l->ref_cnt) { 1993 list_del(&l->node_ref_to); 1994 list_del(&l->node_ref_from); 1995 btrfsic_block_link_hashtable_remove(l); 1996 btrfsic_block_link_free(l); 1997 } 1998 } 1999 2000 if (block->is_superblock) 2001 ret = btrfsic_map_superblock(state, bytenr, 2002 processed_len, 2003 bdev, &block_ctx); 2004 else 2005 ret = btrfsic_map_block(state, bytenr, processed_len, 2006 &block_ctx, 0); 2007 if (ret) { 2008 printk(KERN_INFO 2009 "btrfsic: btrfsic_map_block(root @%llu)" 2010 " failed!\n", bytenr); 2011 goto continue_loop; 2012 } 2013 block_ctx.datav = mapped_datav; 2014 /* the following is required in case of writes to mirrors, 2015 * use the same that was used for the lookup */ 2016 block_ctx.dev = dev_state; 2017 block_ctx.dev_bytenr = dev_bytenr; 2018 2019 if (is_metadata || state->include_extent_data) { 2020 block->never_written = 0; 2021 block->iodone_w_error = 0; 2022 if (NULL != bio) { 2023 block->is_iodone = 0; 2024 BUG_ON(NULL == bio_is_patched); 2025 if (!*bio_is_patched) { 2026 block->orig_bio_bh_private = 2027 bio->bi_private; 2028 block->orig_bio_bh_end_io.bio = 2029 bio->bi_end_io; 2030 block->next_in_same_bio = NULL; 2031 bio->bi_private = block; 2032 bio->bi_end_io = btrfsic_bio_end_io; 2033 *bio_is_patched = 1; 2034 } else { 2035 struct btrfsic_block *chained_block = 2036 (struct btrfsic_block *) 2037 bio->bi_private; 2038 2039 BUG_ON(NULL == chained_block); 2040 block->orig_bio_bh_private = 2041 chained_block->orig_bio_bh_private; 2042 block->orig_bio_bh_end_io.bio = 2043 chained_block->orig_bio_bh_end_io. 2044 bio; 2045 block->next_in_same_bio = chained_block; 2046 bio->bi_private = block; 2047 } 2048 } else if (NULL != bh) { 2049 block->is_iodone = 0; 2050 block->orig_bio_bh_private = bh->b_private; 2051 block->orig_bio_bh_end_io.bh = bh->b_end_io; 2052 block->next_in_same_bio = NULL; 2053 bh->b_private = block; 2054 bh->b_end_io = btrfsic_bh_end_io; 2055 } else { 2056 block->is_iodone = 1; 2057 block->orig_bio_bh_private = NULL; 2058 block->orig_bio_bh_end_io.bio = NULL; 2059 block->next_in_same_bio = NULL; 2060 } 2061 } 2062 2063 block->flush_gen = dev_state->last_flush_gen + 1; 2064 block->submit_bio_bh_rw = submit_bio_bh_rw; 2065 if (is_metadata) { 2066 block->logical_bytenr = bytenr; 2067 block->is_metadata = 1; 2068 if (block->is_superblock) { 2069 BUG_ON(PAGE_CACHE_SIZE != 2070 BTRFS_SUPER_INFO_SIZE); 2071 ret = btrfsic_process_written_superblock( 2072 state, 2073 block, 2074 (struct btrfs_super_block *) 2075 mapped_datav[0]); 2076 if (state->print_mask & 2077 BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) { 2078 printk(KERN_INFO 2079 "[after new superblock is written]:\n"); 2080 btrfsic_dump_tree_sub(state, block, 0); 2081 } 2082 } else { 2083 block->mirror_num = 0; /* unknown */ 2084 ret = btrfsic_process_metablock( 2085 state, 2086 block, 2087 &block_ctx, 2088 0, 0); 2089 } 2090 if (ret) 2091 printk(KERN_INFO 2092 "btrfsic: btrfsic_process_metablock" 2093 "(root @%llu) failed!\n", 2094 dev_bytenr); 2095 } else { 2096 block->is_metadata = 0; 2097 block->mirror_num = 0; /* unknown */ 2098 block->generation = BTRFSIC_GENERATION_UNKNOWN; 2099 if (!state->include_extent_data 2100 && list_empty(&block->ref_from_list)) { 2101 /* 2102 * disk block is overwritten with extent 2103 * data (not meta data) and we are configured 2104 * to not include extent data: take the 2105 * chance and free the block's memory 2106 */ 2107 btrfsic_block_hashtable_remove(block); 2108 list_del(&block->all_blocks_node); 2109 btrfsic_block_free(block); 2110 } 2111 } 2112 btrfsic_release_block_ctx(&block_ctx); 2113 } else { 2114 /* block has not been found in hash table */ 2115 u64 bytenr; 2116 2117 if (!is_metadata) { 2118 processed_len = state->datablock_size; 2119 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2120 printk(KERN_INFO "Written block (%s/%llu/?)" 2121 " !found in hash table, D.\n", 2122 dev_state->name, dev_bytenr); 2123 if (!state->include_extent_data) { 2124 /* ignore that written D block */ 2125 goto continue_loop; 2126 } 2127 2128 /* this is getting ugly for the 2129 * include_extent_data case... */ 2130 bytenr = 0; /* unknown */ 2131 block_ctx.start = bytenr; 2132 block_ctx.len = processed_len; 2133 block_ctx.mem_to_free = NULL; 2134 block_ctx.pagev = NULL; 2135 } else { 2136 processed_len = state->metablock_size; 2137 bytenr = btrfs_stack_header_bytenr( 2138 (struct btrfs_header *) 2139 mapped_datav[0]); 2140 btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state, 2141 dev_bytenr); 2142 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2143 printk(KERN_INFO 2144 "Written block @%llu (%s/%llu/?)" 2145 " !found in hash table, M.\n", 2146 bytenr, dev_state->name, dev_bytenr); 2147 2148 ret = btrfsic_map_block(state, bytenr, processed_len, 2149 &block_ctx, 0); 2150 if (ret) { 2151 printk(KERN_INFO 2152 "btrfsic: btrfsic_map_block(root @%llu)" 2153 " failed!\n", 2154 dev_bytenr); 2155 goto continue_loop; 2156 } 2157 } 2158 block_ctx.datav = mapped_datav; 2159 /* the following is required in case of writes to mirrors, 2160 * use the same that was used for the lookup */ 2161 block_ctx.dev = dev_state; 2162 block_ctx.dev_bytenr = dev_bytenr; 2163 2164 block = btrfsic_block_alloc(); 2165 if (NULL == block) { 2166 printk(KERN_INFO "btrfsic: error, kmalloc failed!\n"); 2167 btrfsic_release_block_ctx(&block_ctx); 2168 goto continue_loop; 2169 } 2170 block->dev_state = dev_state; 2171 block->dev_bytenr = dev_bytenr; 2172 block->logical_bytenr = bytenr; 2173 block->is_metadata = is_metadata; 2174 block->never_written = 0; 2175 block->iodone_w_error = 0; 2176 block->mirror_num = 0; /* unknown */ 2177 block->flush_gen = dev_state->last_flush_gen + 1; 2178 block->submit_bio_bh_rw = submit_bio_bh_rw; 2179 if (NULL != bio) { 2180 block->is_iodone = 0; 2181 BUG_ON(NULL == bio_is_patched); 2182 if (!*bio_is_patched) { 2183 block->orig_bio_bh_private = bio->bi_private; 2184 block->orig_bio_bh_end_io.bio = bio->bi_end_io; 2185 block->next_in_same_bio = NULL; 2186 bio->bi_private = block; 2187 bio->bi_end_io = btrfsic_bio_end_io; 2188 *bio_is_patched = 1; 2189 } else { 2190 struct btrfsic_block *chained_block = 2191 (struct btrfsic_block *) 2192 bio->bi_private; 2193 2194 BUG_ON(NULL == chained_block); 2195 block->orig_bio_bh_private = 2196 chained_block->orig_bio_bh_private; 2197 block->orig_bio_bh_end_io.bio = 2198 chained_block->orig_bio_bh_end_io.bio; 2199 block->next_in_same_bio = chained_block; 2200 bio->bi_private = block; 2201 } 2202 } else if (NULL != bh) { 2203 block->is_iodone = 0; 2204 block->orig_bio_bh_private = bh->b_private; 2205 block->orig_bio_bh_end_io.bh = bh->b_end_io; 2206 block->next_in_same_bio = NULL; 2207 bh->b_private = block; 2208 bh->b_end_io = btrfsic_bh_end_io; 2209 } else { 2210 block->is_iodone = 1; 2211 block->orig_bio_bh_private = NULL; 2212 block->orig_bio_bh_end_io.bio = NULL; 2213 block->next_in_same_bio = NULL; 2214 } 2215 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2216 printk(KERN_INFO 2217 "New written %c-block @%llu (%s/%llu/%d)\n", 2218 is_metadata ? 'M' : 'D', 2219 block->logical_bytenr, block->dev_state->name, 2220 block->dev_bytenr, block->mirror_num); 2221 list_add(&block->all_blocks_node, &state->all_blocks_list); 2222 btrfsic_block_hashtable_add(block, &state->block_hashtable); 2223 2224 if (is_metadata) { 2225 ret = btrfsic_process_metablock(state, block, 2226 &block_ctx, 0, 0); 2227 if (ret) 2228 printk(KERN_INFO 2229 "btrfsic: process_metablock(root @%llu)" 2230 " failed!\n", 2231 dev_bytenr); 2232 } 2233 btrfsic_release_block_ctx(&block_ctx); 2234 } 2235 2236 continue_loop: 2237 BUG_ON(!processed_len); 2238 dev_bytenr += processed_len; 2239 mapped_datav += processed_len >> PAGE_CACHE_SHIFT; 2240 num_pages -= processed_len >> PAGE_CACHE_SHIFT; 2241 goto again; 2242 } 2243 2244 static void btrfsic_bio_end_io(struct bio *bp, int bio_error_status) 2245 { 2246 struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private; 2247 int iodone_w_error; 2248 2249 /* mutex is not held! This is not save if IO is not yet completed 2250 * on umount */ 2251 iodone_w_error = 0; 2252 if (bio_error_status) 2253 iodone_w_error = 1; 2254 2255 BUG_ON(NULL == block); 2256 bp->bi_private = block->orig_bio_bh_private; 2257 bp->bi_end_io = block->orig_bio_bh_end_io.bio; 2258 2259 do { 2260 struct btrfsic_block *next_block; 2261 struct btrfsic_dev_state *const dev_state = block->dev_state; 2262 2263 if ((dev_state->state->print_mask & 2264 BTRFSIC_PRINT_MASK_END_IO_BIO_BH)) 2265 printk(KERN_INFO 2266 "bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n", 2267 bio_error_status, 2268 btrfsic_get_block_type(dev_state->state, block), 2269 block->logical_bytenr, dev_state->name, 2270 block->dev_bytenr, block->mirror_num); 2271 next_block = block->next_in_same_bio; 2272 block->iodone_w_error = iodone_w_error; 2273 if (block->submit_bio_bh_rw & REQ_FLUSH) { 2274 dev_state->last_flush_gen++; 2275 if ((dev_state->state->print_mask & 2276 BTRFSIC_PRINT_MASK_END_IO_BIO_BH)) 2277 printk(KERN_INFO 2278 "bio_end_io() new %s flush_gen=%llu\n", 2279 dev_state->name, 2280 dev_state->last_flush_gen); 2281 } 2282 if (block->submit_bio_bh_rw & REQ_FUA) 2283 block->flush_gen = 0; /* FUA completed means block is 2284 * on disk */ 2285 block->is_iodone = 1; /* for FLUSH, this releases the block */ 2286 block = next_block; 2287 } while (NULL != block); 2288 2289 bp->bi_end_io(bp, bio_error_status); 2290 } 2291 2292 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate) 2293 { 2294 struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private; 2295 int iodone_w_error = !uptodate; 2296 struct btrfsic_dev_state *dev_state; 2297 2298 BUG_ON(NULL == block); 2299 dev_state = block->dev_state; 2300 if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH)) 2301 printk(KERN_INFO 2302 "bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n", 2303 iodone_w_error, 2304 btrfsic_get_block_type(dev_state->state, block), 2305 block->logical_bytenr, block->dev_state->name, 2306 block->dev_bytenr, block->mirror_num); 2307 2308 block->iodone_w_error = iodone_w_error; 2309 if (block->submit_bio_bh_rw & REQ_FLUSH) { 2310 dev_state->last_flush_gen++; 2311 if ((dev_state->state->print_mask & 2312 BTRFSIC_PRINT_MASK_END_IO_BIO_BH)) 2313 printk(KERN_INFO 2314 "bh_end_io() new %s flush_gen=%llu\n", 2315 dev_state->name, dev_state->last_flush_gen); 2316 } 2317 if (block->submit_bio_bh_rw & REQ_FUA) 2318 block->flush_gen = 0; /* FUA completed means block is on disk */ 2319 2320 bh->b_private = block->orig_bio_bh_private; 2321 bh->b_end_io = block->orig_bio_bh_end_io.bh; 2322 block->is_iodone = 1; /* for FLUSH, this releases the block */ 2323 bh->b_end_io(bh, uptodate); 2324 } 2325 2326 static int btrfsic_process_written_superblock( 2327 struct btrfsic_state *state, 2328 struct btrfsic_block *const superblock, 2329 struct btrfs_super_block *const super_hdr) 2330 { 2331 int pass; 2332 2333 superblock->generation = btrfs_super_generation(super_hdr); 2334 if (!(superblock->generation > state->max_superblock_generation || 2335 0 == state->max_superblock_generation)) { 2336 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE) 2337 printk(KERN_INFO 2338 "btrfsic: superblock @%llu (%s/%llu/%d)" 2339 " with old gen %llu <= %llu\n", 2340 superblock->logical_bytenr, 2341 superblock->dev_state->name, 2342 superblock->dev_bytenr, superblock->mirror_num, 2343 btrfs_super_generation(super_hdr), 2344 state->max_superblock_generation); 2345 } else { 2346 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE) 2347 printk(KERN_INFO 2348 "btrfsic: got new superblock @%llu (%s/%llu/%d)" 2349 " with new gen %llu > %llu\n", 2350 superblock->logical_bytenr, 2351 superblock->dev_state->name, 2352 superblock->dev_bytenr, superblock->mirror_num, 2353 btrfs_super_generation(super_hdr), 2354 state->max_superblock_generation); 2355 2356 state->max_superblock_generation = 2357 btrfs_super_generation(super_hdr); 2358 state->latest_superblock = superblock; 2359 } 2360 2361 for (pass = 0; pass < 3; pass++) { 2362 int ret; 2363 u64 next_bytenr; 2364 struct btrfsic_block *next_block; 2365 struct btrfsic_block_data_ctx tmp_next_block_ctx; 2366 struct btrfsic_block_link *l; 2367 int num_copies; 2368 int mirror_num; 2369 const char *additional_string = NULL; 2370 struct btrfs_disk_key tmp_disk_key = {0}; 2371 2372 btrfs_set_disk_key_objectid(&tmp_disk_key, 2373 BTRFS_ROOT_ITEM_KEY); 2374 btrfs_set_disk_key_objectid(&tmp_disk_key, 0); 2375 2376 switch (pass) { 2377 case 0: 2378 btrfs_set_disk_key_objectid(&tmp_disk_key, 2379 BTRFS_ROOT_TREE_OBJECTID); 2380 additional_string = "root "; 2381 next_bytenr = btrfs_super_root(super_hdr); 2382 if (state->print_mask & 2383 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION) 2384 printk(KERN_INFO "root@%llu\n", next_bytenr); 2385 break; 2386 case 1: 2387 btrfs_set_disk_key_objectid(&tmp_disk_key, 2388 BTRFS_CHUNK_TREE_OBJECTID); 2389 additional_string = "chunk "; 2390 next_bytenr = btrfs_super_chunk_root(super_hdr); 2391 if (state->print_mask & 2392 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION) 2393 printk(KERN_INFO "chunk@%llu\n", next_bytenr); 2394 break; 2395 case 2: 2396 btrfs_set_disk_key_objectid(&tmp_disk_key, 2397 BTRFS_TREE_LOG_OBJECTID); 2398 additional_string = "log "; 2399 next_bytenr = btrfs_super_log_root(super_hdr); 2400 if (0 == next_bytenr) 2401 continue; 2402 if (state->print_mask & 2403 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION) 2404 printk(KERN_INFO "log@%llu\n", next_bytenr); 2405 break; 2406 } 2407 2408 num_copies = 2409 btrfs_num_copies(state->root->fs_info, 2410 next_bytenr, BTRFS_SUPER_INFO_SIZE); 2411 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES) 2412 printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n", 2413 next_bytenr, num_copies); 2414 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) { 2415 int was_created; 2416 2417 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2418 printk(KERN_INFO 2419 "btrfsic_process_written_superblock(" 2420 "mirror_num=%d)\n", mirror_num); 2421 ret = btrfsic_map_block(state, next_bytenr, 2422 BTRFS_SUPER_INFO_SIZE, 2423 &tmp_next_block_ctx, 2424 mirror_num); 2425 if (ret) { 2426 printk(KERN_INFO 2427 "btrfsic: btrfsic_map_block(@%llu," 2428 " mirror=%d) failed!\n", 2429 next_bytenr, mirror_num); 2430 return -1; 2431 } 2432 2433 next_block = btrfsic_block_lookup_or_add( 2434 state, 2435 &tmp_next_block_ctx, 2436 additional_string, 2437 1, 0, 1, 2438 mirror_num, 2439 &was_created); 2440 if (NULL == next_block) { 2441 printk(KERN_INFO 2442 "btrfsic: error, kmalloc failed!\n"); 2443 btrfsic_release_block_ctx(&tmp_next_block_ctx); 2444 return -1; 2445 } 2446 2447 next_block->disk_key = tmp_disk_key; 2448 if (was_created) 2449 next_block->generation = 2450 BTRFSIC_GENERATION_UNKNOWN; 2451 l = btrfsic_block_link_lookup_or_add( 2452 state, 2453 &tmp_next_block_ctx, 2454 next_block, 2455 superblock, 2456 BTRFSIC_GENERATION_UNKNOWN); 2457 btrfsic_release_block_ctx(&tmp_next_block_ctx); 2458 if (NULL == l) 2459 return -1; 2460 } 2461 } 2462 2463 if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0))) 2464 btrfsic_dump_tree(state); 2465 2466 return 0; 2467 } 2468 2469 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state, 2470 struct btrfsic_block *const block, 2471 int recursion_level) 2472 { 2473 struct list_head *elem_ref_to; 2474 int ret = 0; 2475 2476 if (recursion_level >= 3 + BTRFS_MAX_LEVEL) { 2477 /* 2478 * Note that this situation can happen and does not 2479 * indicate an error in regular cases. It happens 2480 * when disk blocks are freed and later reused. 2481 * The check-integrity module is not aware of any 2482 * block free operations, it just recognizes block 2483 * write operations. Therefore it keeps the linkage 2484 * information for a block until a block is 2485 * rewritten. This can temporarily cause incorrect 2486 * and even circular linkage informations. This 2487 * causes no harm unless such blocks are referenced 2488 * by the most recent super block. 2489 */ 2490 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2491 printk(KERN_INFO 2492 "btrfsic: abort cyclic linkage (case 1).\n"); 2493 2494 return ret; 2495 } 2496 2497 /* 2498 * This algorithm is recursive because the amount of used stack 2499 * space is very small and the max recursion depth is limited. 2500 */ 2501 list_for_each(elem_ref_to, &block->ref_to_list) { 2502 const struct btrfsic_block_link *const l = 2503 list_entry(elem_ref_to, struct btrfsic_block_link, 2504 node_ref_to); 2505 2506 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2507 printk(KERN_INFO 2508 "rl=%d, %c @%llu (%s/%llu/%d)" 2509 " %u* refers to %c @%llu (%s/%llu/%d)\n", 2510 recursion_level, 2511 btrfsic_get_block_type(state, block), 2512 block->logical_bytenr, block->dev_state->name, 2513 block->dev_bytenr, block->mirror_num, 2514 l->ref_cnt, 2515 btrfsic_get_block_type(state, l->block_ref_to), 2516 l->block_ref_to->logical_bytenr, 2517 l->block_ref_to->dev_state->name, 2518 l->block_ref_to->dev_bytenr, 2519 l->block_ref_to->mirror_num); 2520 if (l->block_ref_to->never_written) { 2521 printk(KERN_INFO "btrfs: attempt to write superblock" 2522 " which references block %c @%llu (%s/%llu/%d)" 2523 " which is never written!\n", 2524 btrfsic_get_block_type(state, l->block_ref_to), 2525 l->block_ref_to->logical_bytenr, 2526 l->block_ref_to->dev_state->name, 2527 l->block_ref_to->dev_bytenr, 2528 l->block_ref_to->mirror_num); 2529 ret = -1; 2530 } else if (!l->block_ref_to->is_iodone) { 2531 printk(KERN_INFO "btrfs: attempt to write superblock" 2532 " which references block %c @%llu (%s/%llu/%d)" 2533 " which is not yet iodone!\n", 2534 btrfsic_get_block_type(state, l->block_ref_to), 2535 l->block_ref_to->logical_bytenr, 2536 l->block_ref_to->dev_state->name, 2537 l->block_ref_to->dev_bytenr, 2538 l->block_ref_to->mirror_num); 2539 ret = -1; 2540 } else if (l->block_ref_to->iodone_w_error) { 2541 printk(KERN_INFO "btrfs: attempt to write superblock" 2542 " which references block %c @%llu (%s/%llu/%d)" 2543 " which has write error!\n", 2544 btrfsic_get_block_type(state, l->block_ref_to), 2545 l->block_ref_to->logical_bytenr, 2546 l->block_ref_to->dev_state->name, 2547 l->block_ref_to->dev_bytenr, 2548 l->block_ref_to->mirror_num); 2549 ret = -1; 2550 } else if (l->parent_generation != 2551 l->block_ref_to->generation && 2552 BTRFSIC_GENERATION_UNKNOWN != 2553 l->parent_generation && 2554 BTRFSIC_GENERATION_UNKNOWN != 2555 l->block_ref_to->generation) { 2556 printk(KERN_INFO "btrfs: attempt to write superblock" 2557 " which references block %c @%llu (%s/%llu/%d)" 2558 " with generation %llu !=" 2559 " parent generation %llu!\n", 2560 btrfsic_get_block_type(state, l->block_ref_to), 2561 l->block_ref_to->logical_bytenr, 2562 l->block_ref_to->dev_state->name, 2563 l->block_ref_to->dev_bytenr, 2564 l->block_ref_to->mirror_num, 2565 l->block_ref_to->generation, 2566 l->parent_generation); 2567 ret = -1; 2568 } else if (l->block_ref_to->flush_gen > 2569 l->block_ref_to->dev_state->last_flush_gen) { 2570 printk(KERN_INFO "btrfs: attempt to write superblock" 2571 " which references block %c @%llu (%s/%llu/%d)" 2572 " which is not flushed out of disk's write cache" 2573 " (block flush_gen=%llu," 2574 " dev->flush_gen=%llu)!\n", 2575 btrfsic_get_block_type(state, l->block_ref_to), 2576 l->block_ref_to->logical_bytenr, 2577 l->block_ref_to->dev_state->name, 2578 l->block_ref_to->dev_bytenr, 2579 l->block_ref_to->mirror_num, block->flush_gen, 2580 l->block_ref_to->dev_state->last_flush_gen); 2581 ret = -1; 2582 } else if (-1 == btrfsic_check_all_ref_blocks(state, 2583 l->block_ref_to, 2584 recursion_level + 2585 1)) { 2586 ret = -1; 2587 } 2588 } 2589 2590 return ret; 2591 } 2592 2593 static int btrfsic_is_block_ref_by_superblock( 2594 const struct btrfsic_state *state, 2595 const struct btrfsic_block *block, 2596 int recursion_level) 2597 { 2598 struct list_head *elem_ref_from; 2599 2600 if (recursion_level >= 3 + BTRFS_MAX_LEVEL) { 2601 /* refer to comment at "abort cyclic linkage (case 1)" */ 2602 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2603 printk(KERN_INFO 2604 "btrfsic: abort cyclic linkage (case 2).\n"); 2605 2606 return 0; 2607 } 2608 2609 /* 2610 * This algorithm is recursive because the amount of used stack space 2611 * is very small and the max recursion depth is limited. 2612 */ 2613 list_for_each(elem_ref_from, &block->ref_from_list) { 2614 const struct btrfsic_block_link *const l = 2615 list_entry(elem_ref_from, struct btrfsic_block_link, 2616 node_ref_from); 2617 2618 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2619 printk(KERN_INFO 2620 "rl=%d, %c @%llu (%s/%llu/%d)" 2621 " is ref %u* from %c @%llu (%s/%llu/%d)\n", 2622 recursion_level, 2623 btrfsic_get_block_type(state, block), 2624 block->logical_bytenr, block->dev_state->name, 2625 block->dev_bytenr, block->mirror_num, 2626 l->ref_cnt, 2627 btrfsic_get_block_type(state, l->block_ref_from), 2628 l->block_ref_from->logical_bytenr, 2629 l->block_ref_from->dev_state->name, 2630 l->block_ref_from->dev_bytenr, 2631 l->block_ref_from->mirror_num); 2632 if (l->block_ref_from->is_superblock && 2633 state->latest_superblock->dev_bytenr == 2634 l->block_ref_from->dev_bytenr && 2635 state->latest_superblock->dev_state->bdev == 2636 l->block_ref_from->dev_state->bdev) 2637 return 1; 2638 else if (btrfsic_is_block_ref_by_superblock(state, 2639 l->block_ref_from, 2640 recursion_level + 2641 1)) 2642 return 1; 2643 } 2644 2645 return 0; 2646 } 2647 2648 static void btrfsic_print_add_link(const struct btrfsic_state *state, 2649 const struct btrfsic_block_link *l) 2650 { 2651 printk(KERN_INFO 2652 "Add %u* link from %c @%llu (%s/%llu/%d)" 2653 " to %c @%llu (%s/%llu/%d).\n", 2654 l->ref_cnt, 2655 btrfsic_get_block_type(state, l->block_ref_from), 2656 l->block_ref_from->logical_bytenr, 2657 l->block_ref_from->dev_state->name, 2658 l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num, 2659 btrfsic_get_block_type(state, l->block_ref_to), 2660 l->block_ref_to->logical_bytenr, 2661 l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr, 2662 l->block_ref_to->mirror_num); 2663 } 2664 2665 static void btrfsic_print_rem_link(const struct btrfsic_state *state, 2666 const struct btrfsic_block_link *l) 2667 { 2668 printk(KERN_INFO 2669 "Rem %u* link from %c @%llu (%s/%llu/%d)" 2670 " to %c @%llu (%s/%llu/%d).\n", 2671 l->ref_cnt, 2672 btrfsic_get_block_type(state, l->block_ref_from), 2673 l->block_ref_from->logical_bytenr, 2674 l->block_ref_from->dev_state->name, 2675 l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num, 2676 btrfsic_get_block_type(state, l->block_ref_to), 2677 l->block_ref_to->logical_bytenr, 2678 l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr, 2679 l->block_ref_to->mirror_num); 2680 } 2681 2682 static char btrfsic_get_block_type(const struct btrfsic_state *state, 2683 const struct btrfsic_block *block) 2684 { 2685 if (block->is_superblock && 2686 state->latest_superblock->dev_bytenr == block->dev_bytenr && 2687 state->latest_superblock->dev_state->bdev == block->dev_state->bdev) 2688 return 'S'; 2689 else if (block->is_superblock) 2690 return 's'; 2691 else if (block->is_metadata) 2692 return 'M'; 2693 else 2694 return 'D'; 2695 } 2696 2697 static void btrfsic_dump_tree(const struct btrfsic_state *state) 2698 { 2699 btrfsic_dump_tree_sub(state, state->latest_superblock, 0); 2700 } 2701 2702 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state, 2703 const struct btrfsic_block *block, 2704 int indent_level) 2705 { 2706 struct list_head *elem_ref_to; 2707 int indent_add; 2708 static char buf[80]; 2709 int cursor_position; 2710 2711 /* 2712 * Should better fill an on-stack buffer with a complete line and 2713 * dump it at once when it is time to print a newline character. 2714 */ 2715 2716 /* 2717 * This algorithm is recursive because the amount of used stack space 2718 * is very small and the max recursion depth is limited. 2719 */ 2720 indent_add = sprintf(buf, "%c-%llu(%s/%llu/%d)", 2721 btrfsic_get_block_type(state, block), 2722 block->logical_bytenr, block->dev_state->name, 2723 block->dev_bytenr, block->mirror_num); 2724 if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) { 2725 printk("[...]\n"); 2726 return; 2727 } 2728 printk(buf); 2729 indent_level += indent_add; 2730 if (list_empty(&block->ref_to_list)) { 2731 printk("\n"); 2732 return; 2733 } 2734 if (block->mirror_num > 1 && 2735 !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) { 2736 printk(" [...]\n"); 2737 return; 2738 } 2739 2740 cursor_position = indent_level; 2741 list_for_each(elem_ref_to, &block->ref_to_list) { 2742 const struct btrfsic_block_link *const l = 2743 list_entry(elem_ref_to, struct btrfsic_block_link, 2744 node_ref_to); 2745 2746 while (cursor_position < indent_level) { 2747 printk(" "); 2748 cursor_position++; 2749 } 2750 if (l->ref_cnt > 1) 2751 indent_add = sprintf(buf, " %d*--> ", l->ref_cnt); 2752 else 2753 indent_add = sprintf(buf, " --> "); 2754 if (indent_level + indent_add > 2755 BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) { 2756 printk("[...]\n"); 2757 cursor_position = 0; 2758 continue; 2759 } 2760 2761 printk(buf); 2762 2763 btrfsic_dump_tree_sub(state, l->block_ref_to, 2764 indent_level + indent_add); 2765 cursor_position = 0; 2766 } 2767 } 2768 2769 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add( 2770 struct btrfsic_state *state, 2771 struct btrfsic_block_data_ctx *next_block_ctx, 2772 struct btrfsic_block *next_block, 2773 struct btrfsic_block *from_block, 2774 u64 parent_generation) 2775 { 2776 struct btrfsic_block_link *l; 2777 2778 l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev, 2779 next_block_ctx->dev_bytenr, 2780 from_block->dev_state->bdev, 2781 from_block->dev_bytenr, 2782 &state->block_link_hashtable); 2783 if (NULL == l) { 2784 l = btrfsic_block_link_alloc(); 2785 if (NULL == l) { 2786 printk(KERN_INFO 2787 "btrfsic: error, kmalloc" " failed!\n"); 2788 return NULL; 2789 } 2790 2791 l->block_ref_to = next_block; 2792 l->block_ref_from = from_block; 2793 l->ref_cnt = 1; 2794 l->parent_generation = parent_generation; 2795 2796 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2797 btrfsic_print_add_link(state, l); 2798 2799 list_add(&l->node_ref_to, &from_block->ref_to_list); 2800 list_add(&l->node_ref_from, &next_block->ref_from_list); 2801 2802 btrfsic_block_link_hashtable_add(l, 2803 &state->block_link_hashtable); 2804 } else { 2805 l->ref_cnt++; 2806 l->parent_generation = parent_generation; 2807 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2808 btrfsic_print_add_link(state, l); 2809 } 2810 2811 return l; 2812 } 2813 2814 static struct btrfsic_block *btrfsic_block_lookup_or_add( 2815 struct btrfsic_state *state, 2816 struct btrfsic_block_data_ctx *block_ctx, 2817 const char *additional_string, 2818 int is_metadata, 2819 int is_iodone, 2820 int never_written, 2821 int mirror_num, 2822 int *was_created) 2823 { 2824 struct btrfsic_block *block; 2825 2826 block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev, 2827 block_ctx->dev_bytenr, 2828 &state->block_hashtable); 2829 if (NULL == block) { 2830 struct btrfsic_dev_state *dev_state; 2831 2832 block = btrfsic_block_alloc(); 2833 if (NULL == block) { 2834 printk(KERN_INFO "btrfsic: error, kmalloc failed!\n"); 2835 return NULL; 2836 } 2837 dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev); 2838 if (NULL == dev_state) { 2839 printk(KERN_INFO 2840 "btrfsic: error, lookup dev_state failed!\n"); 2841 btrfsic_block_free(block); 2842 return NULL; 2843 } 2844 block->dev_state = dev_state; 2845 block->dev_bytenr = block_ctx->dev_bytenr; 2846 block->logical_bytenr = block_ctx->start; 2847 block->is_metadata = is_metadata; 2848 block->is_iodone = is_iodone; 2849 block->never_written = never_written; 2850 block->mirror_num = mirror_num; 2851 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 2852 printk(KERN_INFO 2853 "New %s%c-block @%llu (%s/%llu/%d)\n", 2854 additional_string, 2855 btrfsic_get_block_type(state, block), 2856 block->logical_bytenr, dev_state->name, 2857 block->dev_bytenr, mirror_num); 2858 list_add(&block->all_blocks_node, &state->all_blocks_list); 2859 btrfsic_block_hashtable_add(block, &state->block_hashtable); 2860 if (NULL != was_created) 2861 *was_created = 1; 2862 } else { 2863 if (NULL != was_created) 2864 *was_created = 0; 2865 } 2866 2867 return block; 2868 } 2869 2870 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state, 2871 u64 bytenr, 2872 struct btrfsic_dev_state *dev_state, 2873 u64 dev_bytenr) 2874 { 2875 int num_copies; 2876 int mirror_num; 2877 int ret; 2878 struct btrfsic_block_data_ctx block_ctx; 2879 int match = 0; 2880 2881 num_copies = btrfs_num_copies(state->root->fs_info, 2882 bytenr, state->metablock_size); 2883 2884 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) { 2885 ret = btrfsic_map_block(state, bytenr, state->metablock_size, 2886 &block_ctx, mirror_num); 2887 if (ret) { 2888 printk(KERN_INFO "btrfsic:" 2889 " btrfsic_map_block(logical @%llu," 2890 " mirror %d) failed!\n", 2891 bytenr, mirror_num); 2892 continue; 2893 } 2894 2895 if (dev_state->bdev == block_ctx.dev->bdev && 2896 dev_bytenr == block_ctx.dev_bytenr) { 2897 match++; 2898 btrfsic_release_block_ctx(&block_ctx); 2899 break; 2900 } 2901 btrfsic_release_block_ctx(&block_ctx); 2902 } 2903 2904 if (WARN_ON(!match)) { 2905 printk(KERN_INFO "btrfs: attempt to write M-block which contains logical bytenr that doesn't map to dev+physical bytenr of submit_bio," 2906 " buffer->log_bytenr=%llu, submit_bio(bdev=%s," 2907 " phys_bytenr=%llu)!\n", 2908 bytenr, dev_state->name, dev_bytenr); 2909 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) { 2910 ret = btrfsic_map_block(state, bytenr, 2911 state->metablock_size, 2912 &block_ctx, mirror_num); 2913 if (ret) 2914 continue; 2915 2916 printk(KERN_INFO "Read logical bytenr @%llu maps to" 2917 " (%s/%llu/%d)\n", 2918 bytenr, block_ctx.dev->name, 2919 block_ctx.dev_bytenr, mirror_num); 2920 } 2921 } 2922 } 2923 2924 static struct btrfsic_dev_state *btrfsic_dev_state_lookup( 2925 struct block_device *bdev) 2926 { 2927 struct btrfsic_dev_state *ds; 2928 2929 ds = btrfsic_dev_state_hashtable_lookup(bdev, 2930 &btrfsic_dev_state_hashtable); 2931 return ds; 2932 } 2933 2934 int btrfsic_submit_bh(int rw, struct buffer_head *bh) 2935 { 2936 struct btrfsic_dev_state *dev_state; 2937 2938 if (!btrfsic_is_initialized) 2939 return submit_bh(rw, bh); 2940 2941 mutex_lock(&btrfsic_mutex); 2942 /* since btrfsic_submit_bh() might also be called before 2943 * btrfsic_mount(), this might return NULL */ 2944 dev_state = btrfsic_dev_state_lookup(bh->b_bdev); 2945 2946 /* Only called to write the superblock (incl. FLUSH/FUA) */ 2947 if (NULL != dev_state && 2948 (rw & WRITE) && bh->b_size > 0) { 2949 u64 dev_bytenr; 2950 2951 dev_bytenr = 4096 * bh->b_blocknr; 2952 if (dev_state->state->print_mask & 2953 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH) 2954 printk(KERN_INFO 2955 "submit_bh(rw=0x%x, blocknr=%llu (bytenr %llu)," 2956 " size=%zu, data=%p, bdev=%p)\n", 2957 rw, (unsigned long long)bh->b_blocknr, 2958 dev_bytenr, bh->b_size, bh->b_data, bh->b_bdev); 2959 btrfsic_process_written_block(dev_state, dev_bytenr, 2960 &bh->b_data, 1, NULL, 2961 NULL, bh, rw); 2962 } else if (NULL != dev_state && (rw & REQ_FLUSH)) { 2963 if (dev_state->state->print_mask & 2964 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH) 2965 printk(KERN_INFO 2966 "submit_bh(rw=0x%x FLUSH, bdev=%p)\n", 2967 rw, bh->b_bdev); 2968 if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) { 2969 if ((dev_state->state->print_mask & 2970 (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH | 2971 BTRFSIC_PRINT_MASK_VERBOSE))) 2972 printk(KERN_INFO 2973 "btrfsic_submit_bh(%s) with FLUSH" 2974 " but dummy block already in use" 2975 " (ignored)!\n", 2976 dev_state->name); 2977 } else { 2978 struct btrfsic_block *const block = 2979 &dev_state->dummy_block_for_bio_bh_flush; 2980 2981 block->is_iodone = 0; 2982 block->never_written = 0; 2983 block->iodone_w_error = 0; 2984 block->flush_gen = dev_state->last_flush_gen + 1; 2985 block->submit_bio_bh_rw = rw; 2986 block->orig_bio_bh_private = bh->b_private; 2987 block->orig_bio_bh_end_io.bh = bh->b_end_io; 2988 block->next_in_same_bio = NULL; 2989 bh->b_private = block; 2990 bh->b_end_io = btrfsic_bh_end_io; 2991 } 2992 } 2993 mutex_unlock(&btrfsic_mutex); 2994 return submit_bh(rw, bh); 2995 } 2996 2997 static void __btrfsic_submit_bio(int rw, struct bio *bio) 2998 { 2999 struct btrfsic_dev_state *dev_state; 3000 3001 if (!btrfsic_is_initialized) 3002 return; 3003 3004 mutex_lock(&btrfsic_mutex); 3005 /* since btrfsic_submit_bio() is also called before 3006 * btrfsic_mount(), this might return NULL */ 3007 dev_state = btrfsic_dev_state_lookup(bio->bi_bdev); 3008 if (NULL != dev_state && 3009 (rw & WRITE) && NULL != bio->bi_io_vec) { 3010 unsigned int i; 3011 u64 dev_bytenr; 3012 u64 cur_bytenr; 3013 int bio_is_patched; 3014 char **mapped_datav; 3015 3016 dev_bytenr = 512 * bio->bi_sector; 3017 bio_is_patched = 0; 3018 if (dev_state->state->print_mask & 3019 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH) 3020 printk(KERN_INFO 3021 "submit_bio(rw=0x%x, bi_vcnt=%u," 3022 " bi_sector=%llu (bytenr %llu), bi_bdev=%p)\n", 3023 rw, bio->bi_vcnt, 3024 (unsigned long long)bio->bi_sector, dev_bytenr, 3025 bio->bi_bdev); 3026 3027 mapped_datav = kmalloc(sizeof(*mapped_datav) * bio->bi_vcnt, 3028 GFP_NOFS); 3029 if (!mapped_datav) 3030 goto leave; 3031 cur_bytenr = dev_bytenr; 3032 for (i = 0; i < bio->bi_vcnt; i++) { 3033 BUG_ON(bio->bi_io_vec[i].bv_len != PAGE_CACHE_SIZE); 3034 mapped_datav[i] = kmap(bio->bi_io_vec[i].bv_page); 3035 if (!mapped_datav[i]) { 3036 while (i > 0) { 3037 i--; 3038 kunmap(bio->bi_io_vec[i].bv_page); 3039 } 3040 kfree(mapped_datav); 3041 goto leave; 3042 } 3043 if (dev_state->state->print_mask & 3044 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE) 3045 printk(KERN_INFO 3046 "#%u: bytenr=%llu, len=%u, offset=%u\n", 3047 i, cur_bytenr, bio->bi_io_vec[i].bv_len, 3048 bio->bi_io_vec[i].bv_offset); 3049 cur_bytenr += bio->bi_io_vec[i].bv_len; 3050 } 3051 btrfsic_process_written_block(dev_state, dev_bytenr, 3052 mapped_datav, bio->bi_vcnt, 3053 bio, &bio_is_patched, 3054 NULL, rw); 3055 while (i > 0) { 3056 i--; 3057 kunmap(bio->bi_io_vec[i].bv_page); 3058 } 3059 kfree(mapped_datav); 3060 } else if (NULL != dev_state && (rw & REQ_FLUSH)) { 3061 if (dev_state->state->print_mask & 3062 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH) 3063 printk(KERN_INFO 3064 "submit_bio(rw=0x%x FLUSH, bdev=%p)\n", 3065 rw, bio->bi_bdev); 3066 if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) { 3067 if ((dev_state->state->print_mask & 3068 (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH | 3069 BTRFSIC_PRINT_MASK_VERBOSE))) 3070 printk(KERN_INFO 3071 "btrfsic_submit_bio(%s) with FLUSH" 3072 " but dummy block already in use" 3073 " (ignored)!\n", 3074 dev_state->name); 3075 } else { 3076 struct btrfsic_block *const block = 3077 &dev_state->dummy_block_for_bio_bh_flush; 3078 3079 block->is_iodone = 0; 3080 block->never_written = 0; 3081 block->iodone_w_error = 0; 3082 block->flush_gen = dev_state->last_flush_gen + 1; 3083 block->submit_bio_bh_rw = rw; 3084 block->orig_bio_bh_private = bio->bi_private; 3085 block->orig_bio_bh_end_io.bio = bio->bi_end_io; 3086 block->next_in_same_bio = NULL; 3087 bio->bi_private = block; 3088 bio->bi_end_io = btrfsic_bio_end_io; 3089 } 3090 } 3091 leave: 3092 mutex_unlock(&btrfsic_mutex); 3093 } 3094 3095 void btrfsic_submit_bio(int rw, struct bio *bio) 3096 { 3097 __btrfsic_submit_bio(rw, bio); 3098 submit_bio(rw, bio); 3099 } 3100 3101 int btrfsic_submit_bio_wait(int rw, struct bio *bio) 3102 { 3103 __btrfsic_submit_bio(rw, bio); 3104 return submit_bio_wait(rw, bio); 3105 } 3106 3107 int btrfsic_mount(struct btrfs_root *root, 3108 struct btrfs_fs_devices *fs_devices, 3109 int including_extent_data, u32 print_mask) 3110 { 3111 int ret; 3112 struct btrfsic_state *state; 3113 struct list_head *dev_head = &fs_devices->devices; 3114 struct btrfs_device *device; 3115 3116 if (root->nodesize != root->leafsize) { 3117 printk(KERN_INFO 3118 "btrfsic: cannot handle nodesize %d != leafsize %d!\n", 3119 root->nodesize, root->leafsize); 3120 return -1; 3121 } 3122 if (root->nodesize & ((u64)PAGE_CACHE_SIZE - 1)) { 3123 printk(KERN_INFO 3124 "btrfsic: cannot handle nodesize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n", 3125 root->nodesize, PAGE_CACHE_SIZE); 3126 return -1; 3127 } 3128 if (root->leafsize & ((u64)PAGE_CACHE_SIZE - 1)) { 3129 printk(KERN_INFO 3130 "btrfsic: cannot handle leafsize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n", 3131 root->leafsize, PAGE_CACHE_SIZE); 3132 return -1; 3133 } 3134 if (root->sectorsize & ((u64)PAGE_CACHE_SIZE - 1)) { 3135 printk(KERN_INFO 3136 "btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n", 3137 root->sectorsize, PAGE_CACHE_SIZE); 3138 return -1; 3139 } 3140 state = kzalloc(sizeof(*state), GFP_NOFS); 3141 if (NULL == state) { 3142 printk(KERN_INFO "btrfs check-integrity: kmalloc() failed!\n"); 3143 return -1; 3144 } 3145 3146 if (!btrfsic_is_initialized) { 3147 mutex_init(&btrfsic_mutex); 3148 btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable); 3149 btrfsic_is_initialized = 1; 3150 } 3151 mutex_lock(&btrfsic_mutex); 3152 state->root = root; 3153 state->print_mask = print_mask; 3154 state->include_extent_data = including_extent_data; 3155 state->csum_size = 0; 3156 state->metablock_size = root->nodesize; 3157 state->datablock_size = root->sectorsize; 3158 INIT_LIST_HEAD(&state->all_blocks_list); 3159 btrfsic_block_hashtable_init(&state->block_hashtable); 3160 btrfsic_block_link_hashtable_init(&state->block_link_hashtable); 3161 state->max_superblock_generation = 0; 3162 state->latest_superblock = NULL; 3163 3164 list_for_each_entry(device, dev_head, dev_list) { 3165 struct btrfsic_dev_state *ds; 3166 char *p; 3167 3168 if (!device->bdev || !device->name) 3169 continue; 3170 3171 ds = btrfsic_dev_state_alloc(); 3172 if (NULL == ds) { 3173 printk(KERN_INFO 3174 "btrfs check-integrity: kmalloc() failed!\n"); 3175 mutex_unlock(&btrfsic_mutex); 3176 return -1; 3177 } 3178 ds->bdev = device->bdev; 3179 ds->state = state; 3180 bdevname(ds->bdev, ds->name); 3181 ds->name[BDEVNAME_SIZE - 1] = '\0'; 3182 for (p = ds->name; *p != '\0'; p++); 3183 while (p > ds->name && *p != '/') 3184 p--; 3185 if (*p == '/') 3186 p++; 3187 strlcpy(ds->name, p, sizeof(ds->name)); 3188 btrfsic_dev_state_hashtable_add(ds, 3189 &btrfsic_dev_state_hashtable); 3190 } 3191 3192 ret = btrfsic_process_superblock(state, fs_devices); 3193 if (0 != ret) { 3194 mutex_unlock(&btrfsic_mutex); 3195 btrfsic_unmount(root, fs_devices); 3196 return ret; 3197 } 3198 3199 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE) 3200 btrfsic_dump_database(state); 3201 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE) 3202 btrfsic_dump_tree(state); 3203 3204 mutex_unlock(&btrfsic_mutex); 3205 return 0; 3206 } 3207 3208 void btrfsic_unmount(struct btrfs_root *root, 3209 struct btrfs_fs_devices *fs_devices) 3210 { 3211 struct list_head *elem_all; 3212 struct list_head *tmp_all; 3213 struct btrfsic_state *state; 3214 struct list_head *dev_head = &fs_devices->devices; 3215 struct btrfs_device *device; 3216 3217 if (!btrfsic_is_initialized) 3218 return; 3219 3220 mutex_lock(&btrfsic_mutex); 3221 3222 state = NULL; 3223 list_for_each_entry(device, dev_head, dev_list) { 3224 struct btrfsic_dev_state *ds; 3225 3226 if (!device->bdev || !device->name) 3227 continue; 3228 3229 ds = btrfsic_dev_state_hashtable_lookup( 3230 device->bdev, 3231 &btrfsic_dev_state_hashtable); 3232 if (NULL != ds) { 3233 state = ds->state; 3234 btrfsic_dev_state_hashtable_remove(ds); 3235 btrfsic_dev_state_free(ds); 3236 } 3237 } 3238 3239 if (NULL == state) { 3240 printk(KERN_INFO 3241 "btrfsic: error, cannot find state information" 3242 " on umount!\n"); 3243 mutex_unlock(&btrfsic_mutex); 3244 return; 3245 } 3246 3247 /* 3248 * Don't care about keeping the lists' state up to date, 3249 * just free all memory that was allocated dynamically. 3250 * Free the blocks and the block_links. 3251 */ 3252 list_for_each_safe(elem_all, tmp_all, &state->all_blocks_list) { 3253 struct btrfsic_block *const b_all = 3254 list_entry(elem_all, struct btrfsic_block, 3255 all_blocks_node); 3256 struct list_head *elem_ref_to; 3257 struct list_head *tmp_ref_to; 3258 3259 list_for_each_safe(elem_ref_to, tmp_ref_to, 3260 &b_all->ref_to_list) { 3261 struct btrfsic_block_link *const l = 3262 list_entry(elem_ref_to, 3263 struct btrfsic_block_link, 3264 node_ref_to); 3265 3266 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) 3267 btrfsic_print_rem_link(state, l); 3268 3269 l->ref_cnt--; 3270 if (0 == l->ref_cnt) 3271 btrfsic_block_link_free(l); 3272 } 3273 3274 if (b_all->is_iodone || b_all->never_written) 3275 btrfsic_block_free(b_all); 3276 else 3277 printk(KERN_INFO "btrfs: attempt to free %c-block" 3278 " @%llu (%s/%llu/%d) on umount which is" 3279 " not yet iodone!\n", 3280 btrfsic_get_block_type(state, b_all), 3281 b_all->logical_bytenr, b_all->dev_state->name, 3282 b_all->dev_bytenr, b_all->mirror_num); 3283 } 3284 3285 mutex_unlock(&btrfsic_mutex); 3286 3287 kfree(state); 3288 } 3289