Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) STRATO AG 2011. All rights reserved.
4 */
5
6/*
7 * This module can be used to catch cases when the btrfs kernel
8 * code executes write requests to the disk that bring the file
9 * system in an inconsistent state. In such a state, a power-loss
10 * or kernel panic event would cause that the data on disk is
11 * lost or at least damaged.
12 *
13 * Code is added that examines all block write requests during
14 * runtime (including writes of the super block). Three rules
15 * are verified and an error is printed on violation of the
16 * rules:
17 * 1. It is not allowed to write a disk block which is
18 * currently referenced by the super block (either directly
19 * or indirectly).
20 * 2. When a super block is written, it is verified that all
21 * referenced (directly or indirectly) blocks fulfill the
22 * following requirements:
23 * 2a. All referenced blocks have either been present when
24 * the file system was mounted, (i.e., they have been
25 * referenced by the super block) or they have been
26 * written since then and the write completion callback
27 * was called and no write error was indicated and a
28 * FLUSH request to the device where these blocks are
29 * located was received and completed.
30 * 2b. All referenced blocks need to have a generation
31 * number which is equal to the parent's number.
32 *
33 * One issue that was found using this module was that the log
34 * tree on disk became temporarily corrupted because disk blocks
35 * that had been in use for the log tree had been freed and
36 * reused too early, while being referenced by the written super
37 * block.
38 *
39 * The search term in the kernel log that can be used to filter
40 * on the existence of detected integrity issues is
41 * "btrfs: attempt".
42 *
43 * The integrity check is enabled via mount options. These
44 * mount options are only supported if the integrity check
45 * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY.
46 *
47 * Example #1, apply integrity checks to all metadata:
48 * mount /dev/sdb1 /mnt -o check_int
49 *
50 * Example #2, apply integrity checks to all metadata and
51 * to data extents:
52 * mount /dev/sdb1 /mnt -o check_int_data
53 *
54 * Example #3, apply integrity checks to all metadata and dump
55 * the tree that the super block references to kernel messages
56 * each time after a super block was written:
57 * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263
58 *
59 * If the integrity check tool is included and activated in
60 * the mount options, plenty of kernel memory is used, and
61 * plenty of additional CPU cycles are spent. Enabling this
62 * functionality is not intended for normal use. In most
63 * cases, unless you are a btrfs developer who needs to verify
64 * the integrity of (super)-block write requests, do not
65 * enable the config option BTRFS_FS_CHECK_INTEGRITY to
66 * include and compile the integrity check tool.
67 *
68 * Expect millions of lines of information in the kernel log with an
69 * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the
70 * kernel config to at least 26 (which is 64MB). Usually the value is
71 * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be
72 * changed like this before LOG_BUF_SHIFT can be set to a high value:
73 * config LOG_BUF_SHIFT
74 * int "Kernel log buffer size (16 => 64KB, 17 => 128KB)"
75 * range 12 30
76 */
77
78#include <linux/sched.h>
79#include <linux/slab.h>
80#include <linux/mutex.h>
81#include <linux/blkdev.h>
82#include <linux/mm.h>
83#include <linux/string.h>
84#include <crypto/hash.h>
85#include "messages.h"
86#include "ctree.h"
87#include "disk-io.h"
88#include "transaction.h"
89#include "extent_io.h"
90#include "volumes.h"
91#include "print-tree.h"
92#include "locking.h"
93#include "check-integrity.h"
94#include "rcu-string.h"
95#include "compression.h"
96#include "accessors.h"
97
98#define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
99#define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
100#define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
101#define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
102#define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
103#define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
104#define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
105#define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6) /* in characters,
106 * excluding " [...]" */
107#define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
108
109/*
110 * The definition of the bitmask fields for the print_mask.
111 * They are specified with the mount option check_integrity_print_mask.
112 */
113#define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE 0x00000001
114#define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION 0x00000002
115#define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE 0x00000004
116#define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE 0x00000008
117#define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH 0x00000010
118#define BTRFSIC_PRINT_MASK_END_IO_BIO_BH 0x00000020
119#define BTRFSIC_PRINT_MASK_VERBOSE 0x00000040
120#define BTRFSIC_PRINT_MASK_VERY_VERBOSE 0x00000080
121#define BTRFSIC_PRINT_MASK_INITIAL_TREE 0x00000100
122#define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES 0x00000200
123#define BTRFSIC_PRINT_MASK_INITIAL_DATABASE 0x00000400
124#define BTRFSIC_PRINT_MASK_NUM_COPIES 0x00000800
125#define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS 0x00001000
126#define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE 0x00002000
127
128struct btrfsic_dev_state;
129struct btrfsic_state;
130
131struct btrfsic_block {
132 u32 magic_num; /* only used for debug purposes */
133 unsigned int is_metadata:1; /* if it is meta-data, not data-data */
134 unsigned int is_superblock:1; /* if it is one of the superblocks */
135 unsigned int is_iodone:1; /* if is done by lower subsystem */
136 unsigned int iodone_w_error:1; /* error was indicated to endio */
137 unsigned int never_written:1; /* block was added because it was
138 * referenced, not because it was
139 * written */
140 unsigned int mirror_num; /* large enough to hold
141 * BTRFS_SUPER_MIRROR_MAX */
142 struct btrfsic_dev_state *dev_state;
143 u64 dev_bytenr; /* key, physical byte num on disk */
144 u64 logical_bytenr; /* logical byte num on disk */
145 u64 generation;
146 struct btrfs_disk_key disk_key; /* extra info to print in case of
147 * issues, will not always be correct */
148 struct list_head collision_resolving_node; /* list node */
149 struct list_head all_blocks_node; /* list node */
150
151 /* the following two lists contain block_link items */
152 struct list_head ref_to_list; /* list */
153 struct list_head ref_from_list; /* list */
154 struct btrfsic_block *next_in_same_bio;
155 void *orig_bio_private;
156 bio_end_io_t *orig_bio_end_io;
157 blk_opf_t submit_bio_bh_rw;
158 u64 flush_gen; /* only valid if !never_written */
159};
160
161/*
162 * Elements of this type are allocated dynamically and required because
163 * each block object can refer to and can be ref from multiple blocks.
164 * The key to lookup them in the hashtable is the dev_bytenr of
165 * the block ref to plus the one from the block referred from.
166 * The fact that they are searchable via a hashtable and that a
167 * ref_cnt is maintained is not required for the btrfs integrity
168 * check algorithm itself, it is only used to make the output more
169 * beautiful in case that an error is detected (an error is defined
170 * as a write operation to a block while that block is still referenced).
171 */
172struct btrfsic_block_link {
173 u32 magic_num; /* only used for debug purposes */
174 u32 ref_cnt;
175 struct list_head node_ref_to; /* list node */
176 struct list_head node_ref_from; /* list node */
177 struct list_head collision_resolving_node; /* list node */
178 struct btrfsic_block *block_ref_to;
179 struct btrfsic_block *block_ref_from;
180 u64 parent_generation;
181};
182
183struct btrfsic_dev_state {
184 u32 magic_num; /* only used for debug purposes */
185 struct block_device *bdev;
186 struct btrfsic_state *state;
187 struct list_head collision_resolving_node; /* list node */
188 struct btrfsic_block dummy_block_for_bio_bh_flush;
189 u64 last_flush_gen;
190};
191
192struct btrfsic_block_hashtable {
193 struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
194};
195
196struct btrfsic_block_link_hashtable {
197 struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
198};
199
200struct btrfsic_dev_state_hashtable {
201 struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
202};
203
204struct btrfsic_block_data_ctx {
205 u64 start; /* virtual bytenr */
206 u64 dev_bytenr; /* physical bytenr on device */
207 u32 len;
208 struct btrfsic_dev_state *dev;
209 char **datav;
210 struct page **pagev;
211 void *mem_to_free;
212};
213
214/* This structure is used to implement recursion without occupying
215 * any stack space, refer to btrfsic_process_metablock() */
216struct btrfsic_stack_frame {
217 u32 magic;
218 u32 nr;
219 int error;
220 int i;
221 int limit_nesting;
222 int num_copies;
223 int mirror_num;
224 struct btrfsic_block *block;
225 struct btrfsic_block_data_ctx *block_ctx;
226 struct btrfsic_block *next_block;
227 struct btrfsic_block_data_ctx next_block_ctx;
228 struct btrfs_header *hdr;
229 struct btrfsic_stack_frame *prev;
230};
231
232/* Some state per mounted filesystem */
233struct btrfsic_state {
234 u32 print_mask;
235 int include_extent_data;
236 struct list_head all_blocks_list;
237 struct btrfsic_block_hashtable block_hashtable;
238 struct btrfsic_block_link_hashtable block_link_hashtable;
239 struct btrfs_fs_info *fs_info;
240 u64 max_superblock_generation;
241 struct btrfsic_block *latest_superblock;
242 u32 metablock_size;
243 u32 datablock_size;
244};
245
246static int btrfsic_process_metablock(struct btrfsic_state *state,
247 struct btrfsic_block *block,
248 struct btrfsic_block_data_ctx *block_ctx,
249 int limit_nesting, int force_iodone_flag);
250static void btrfsic_read_from_block_data(
251 struct btrfsic_block_data_ctx *block_ctx,
252 void *dst, u32 offset, size_t len);
253static int btrfsic_create_link_to_next_block(
254 struct btrfsic_state *state,
255 struct btrfsic_block *block,
256 struct btrfsic_block_data_ctx
257 *block_ctx, u64 next_bytenr,
258 int limit_nesting,
259 struct btrfsic_block_data_ctx *next_block_ctx,
260 struct btrfsic_block **next_blockp,
261 int force_iodone_flag,
262 int *num_copiesp, int *mirror_nump,
263 struct btrfs_disk_key *disk_key,
264 u64 parent_generation);
265static int btrfsic_handle_extent_data(struct btrfsic_state *state,
266 struct btrfsic_block *block,
267 struct btrfsic_block_data_ctx *block_ctx,
268 u32 item_offset, int force_iodone_flag);
269static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
270 struct btrfsic_block_data_ctx *block_ctx_out,
271 int mirror_num);
272static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
273static int btrfsic_read_block(struct btrfsic_state *state,
274 struct btrfsic_block_data_ctx *block_ctx);
275static int btrfsic_process_written_superblock(
276 struct btrfsic_state *state,
277 struct btrfsic_block *const block,
278 struct btrfs_super_block *const super_hdr);
279static void btrfsic_bio_end_io(struct bio *bp);
280static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
281 const struct btrfsic_block *block,
282 int recursion_level);
283static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
284 struct btrfsic_block *const block,
285 int recursion_level);
286static void btrfsic_print_add_link(const struct btrfsic_state *state,
287 const struct btrfsic_block_link *l);
288static void btrfsic_print_rem_link(const struct btrfsic_state *state,
289 const struct btrfsic_block_link *l);
290static char btrfsic_get_block_type(const struct btrfsic_state *state,
291 const struct btrfsic_block *block);
292static void btrfsic_dump_tree(const struct btrfsic_state *state);
293static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
294 const struct btrfsic_block *block,
295 int indent_level);
296static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
297 struct btrfsic_state *state,
298 struct btrfsic_block_data_ctx *next_block_ctx,
299 struct btrfsic_block *next_block,
300 struct btrfsic_block *from_block,
301 u64 parent_generation);
302static struct btrfsic_block *btrfsic_block_lookup_or_add(
303 struct btrfsic_state *state,
304 struct btrfsic_block_data_ctx *block_ctx,
305 const char *additional_string,
306 int is_metadata,
307 int is_iodone,
308 int never_written,
309 int mirror_num,
310 int *was_created);
311static int btrfsic_process_superblock_dev_mirror(
312 struct btrfsic_state *state,
313 struct btrfsic_dev_state *dev_state,
314 struct btrfs_device *device,
315 int superblock_mirror_num,
316 struct btrfsic_dev_state **selected_dev_state,
317 struct btrfs_super_block *selected_super);
318static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev);
319static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
320 u64 bytenr,
321 struct btrfsic_dev_state *dev_state,
322 u64 dev_bytenr);
323
324static struct mutex btrfsic_mutex;
325static int btrfsic_is_initialized;
326static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
327
328
329static void btrfsic_block_init(struct btrfsic_block *b)
330{
331 b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
332 b->dev_state = NULL;
333 b->dev_bytenr = 0;
334 b->logical_bytenr = 0;
335 b->generation = BTRFSIC_GENERATION_UNKNOWN;
336 b->disk_key.objectid = 0;
337 b->disk_key.type = 0;
338 b->disk_key.offset = 0;
339 b->is_metadata = 0;
340 b->is_superblock = 0;
341 b->is_iodone = 0;
342 b->iodone_w_error = 0;
343 b->never_written = 0;
344 b->mirror_num = 0;
345 b->next_in_same_bio = NULL;
346 b->orig_bio_private = NULL;
347 b->orig_bio_end_io = NULL;
348 INIT_LIST_HEAD(&b->collision_resolving_node);
349 INIT_LIST_HEAD(&b->all_blocks_node);
350 INIT_LIST_HEAD(&b->ref_to_list);
351 INIT_LIST_HEAD(&b->ref_from_list);
352 b->submit_bio_bh_rw = 0;
353 b->flush_gen = 0;
354}
355
356static struct btrfsic_block *btrfsic_block_alloc(void)
357{
358 struct btrfsic_block *b;
359
360 b = kzalloc(sizeof(*b), GFP_NOFS);
361 if (NULL != b)
362 btrfsic_block_init(b);
363
364 return b;
365}
366
367static void btrfsic_block_free(struct btrfsic_block *b)
368{
369 BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
370 kfree(b);
371}
372
373static void btrfsic_block_link_init(struct btrfsic_block_link *l)
374{
375 l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
376 l->ref_cnt = 1;
377 INIT_LIST_HEAD(&l->node_ref_to);
378 INIT_LIST_HEAD(&l->node_ref_from);
379 INIT_LIST_HEAD(&l->collision_resolving_node);
380 l->block_ref_to = NULL;
381 l->block_ref_from = NULL;
382}
383
384static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
385{
386 struct btrfsic_block_link *l;
387
388 l = kzalloc(sizeof(*l), GFP_NOFS);
389 if (NULL != l)
390 btrfsic_block_link_init(l);
391
392 return l;
393}
394
395static void btrfsic_block_link_free(struct btrfsic_block_link *l)
396{
397 BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
398 kfree(l);
399}
400
401static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
402{
403 ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
404 ds->bdev = NULL;
405 ds->state = NULL;
406 INIT_LIST_HEAD(&ds->collision_resolving_node);
407 ds->last_flush_gen = 0;
408 btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
409 ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
410 ds->dummy_block_for_bio_bh_flush.dev_state = ds;
411}
412
413static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
414{
415 struct btrfsic_dev_state *ds;
416
417 ds = kzalloc(sizeof(*ds), GFP_NOFS);
418 if (NULL != ds)
419 btrfsic_dev_state_init(ds);
420
421 return ds;
422}
423
424static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
425{
426 BUG_ON(!(NULL == ds ||
427 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
428 kfree(ds);
429}
430
431static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
432{
433 int i;
434
435 for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
436 INIT_LIST_HEAD(h->table + i);
437}
438
439static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
440 struct btrfsic_block_hashtable *h)
441{
442 const unsigned int hashval =
443 (((unsigned int)(b->dev_bytenr >> 16)) ^
444 ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
445 (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
446
447 list_add(&b->collision_resolving_node, h->table + hashval);
448}
449
450static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
451{
452 list_del(&b->collision_resolving_node);
453}
454
455static struct btrfsic_block *btrfsic_block_hashtable_lookup(
456 struct block_device *bdev,
457 u64 dev_bytenr,
458 struct btrfsic_block_hashtable *h)
459{
460 const unsigned int hashval =
461 (((unsigned int)(dev_bytenr >> 16)) ^
462 ((unsigned int)((uintptr_t)bdev))) &
463 (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
464 struct btrfsic_block *b;
465
466 list_for_each_entry(b, h->table + hashval, collision_resolving_node) {
467 if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
468 return b;
469 }
470
471 return NULL;
472}
473
474static void btrfsic_block_link_hashtable_init(
475 struct btrfsic_block_link_hashtable *h)
476{
477 int i;
478
479 for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
480 INIT_LIST_HEAD(h->table + i);
481}
482
483static void btrfsic_block_link_hashtable_add(
484 struct btrfsic_block_link *l,
485 struct btrfsic_block_link_hashtable *h)
486{
487 const unsigned int hashval =
488 (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
489 ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
490 ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
491 ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
492 & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
493
494 BUG_ON(NULL == l->block_ref_to);
495 BUG_ON(NULL == l->block_ref_from);
496 list_add(&l->collision_resolving_node, h->table + hashval);
497}
498
499static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
500{
501 list_del(&l->collision_resolving_node);
502}
503
504static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
505 struct block_device *bdev_ref_to,
506 u64 dev_bytenr_ref_to,
507 struct block_device *bdev_ref_from,
508 u64 dev_bytenr_ref_from,
509 struct btrfsic_block_link_hashtable *h)
510{
511 const unsigned int hashval =
512 (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
513 ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
514 ((unsigned int)((uintptr_t)bdev_ref_to)) ^
515 ((unsigned int)((uintptr_t)bdev_ref_from))) &
516 (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
517 struct btrfsic_block_link *l;
518
519 list_for_each_entry(l, h->table + hashval, collision_resolving_node) {
520 BUG_ON(NULL == l->block_ref_to);
521 BUG_ON(NULL == l->block_ref_from);
522 if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
523 l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
524 l->block_ref_from->dev_state->bdev == bdev_ref_from &&
525 l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
526 return l;
527 }
528
529 return NULL;
530}
531
532static void btrfsic_dev_state_hashtable_init(
533 struct btrfsic_dev_state_hashtable *h)
534{
535 int i;
536
537 for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
538 INIT_LIST_HEAD(h->table + i);
539}
540
541static void btrfsic_dev_state_hashtable_add(
542 struct btrfsic_dev_state *ds,
543 struct btrfsic_dev_state_hashtable *h)
544{
545 const unsigned int hashval =
546 (((unsigned int)((uintptr_t)ds->bdev->bd_dev)) &
547 (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
548
549 list_add(&ds->collision_resolving_node, h->table + hashval);
550}
551
552static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
553{
554 list_del(&ds->collision_resolving_node);
555}
556
557static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev,
558 struct btrfsic_dev_state_hashtable *h)
559{
560 const unsigned int hashval =
561 dev & (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1);
562 struct btrfsic_dev_state *ds;
563
564 list_for_each_entry(ds, h->table + hashval, collision_resolving_node) {
565 if (ds->bdev->bd_dev == dev)
566 return ds;
567 }
568
569 return NULL;
570}
571
572static int btrfsic_process_superblock(struct btrfsic_state *state,
573 struct btrfs_fs_devices *fs_devices)
574{
575 struct btrfs_super_block *selected_super;
576 struct list_head *dev_head = &fs_devices->devices;
577 struct btrfs_device *device;
578 struct btrfsic_dev_state *selected_dev_state = NULL;
579 int ret = 0;
580 int pass;
581
582 selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
583 if (!selected_super)
584 return -ENOMEM;
585
586 list_for_each_entry(device, dev_head, dev_list) {
587 int i;
588 struct btrfsic_dev_state *dev_state;
589
590 if (!device->bdev || !device->name)
591 continue;
592
593 dev_state = btrfsic_dev_state_lookup(device->bdev->bd_dev);
594 BUG_ON(NULL == dev_state);
595 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
596 ret = btrfsic_process_superblock_dev_mirror(
597 state, dev_state, device, i,
598 &selected_dev_state, selected_super);
599 if (0 != ret && 0 == i) {
600 kfree(selected_super);
601 return ret;
602 }
603 }
604 }
605
606 if (NULL == state->latest_superblock) {
607 pr_info("btrfsic: no superblock found!\n");
608 kfree(selected_super);
609 return -1;
610 }
611
612 for (pass = 0; pass < 3; pass++) {
613 int num_copies;
614 int mirror_num;
615 u64 next_bytenr;
616
617 switch (pass) {
618 case 0:
619 next_bytenr = btrfs_super_root(selected_super);
620 if (state->print_mask &
621 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
622 pr_info("root@%llu\n", next_bytenr);
623 break;
624 case 1:
625 next_bytenr = btrfs_super_chunk_root(selected_super);
626 if (state->print_mask &
627 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
628 pr_info("chunk@%llu\n", next_bytenr);
629 break;
630 case 2:
631 next_bytenr = btrfs_super_log_root(selected_super);
632 if (0 == next_bytenr)
633 continue;
634 if (state->print_mask &
635 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
636 pr_info("log@%llu\n", next_bytenr);
637 break;
638 }
639
640 num_copies = btrfs_num_copies(state->fs_info, next_bytenr,
641 state->metablock_size);
642 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
643 pr_info("num_copies(log_bytenr=%llu) = %d\n",
644 next_bytenr, num_copies);
645
646 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
647 struct btrfsic_block *next_block;
648 struct btrfsic_block_data_ctx tmp_next_block_ctx;
649 struct btrfsic_block_link *l;
650
651 ret = btrfsic_map_block(state, next_bytenr,
652 state->metablock_size,
653 &tmp_next_block_ctx,
654 mirror_num);
655 if (ret) {
656 pr_info("btrfsic: btrfsic_map_block(root @%llu, mirror %d) failed!\n",
657 next_bytenr, mirror_num);
658 kfree(selected_super);
659 return -1;
660 }
661
662 next_block = btrfsic_block_hashtable_lookup(
663 tmp_next_block_ctx.dev->bdev,
664 tmp_next_block_ctx.dev_bytenr,
665 &state->block_hashtable);
666 BUG_ON(NULL == next_block);
667
668 l = btrfsic_block_link_hashtable_lookup(
669 tmp_next_block_ctx.dev->bdev,
670 tmp_next_block_ctx.dev_bytenr,
671 state->latest_superblock->dev_state->
672 bdev,
673 state->latest_superblock->dev_bytenr,
674 &state->block_link_hashtable);
675 BUG_ON(NULL == l);
676
677 ret = btrfsic_read_block(state, &tmp_next_block_ctx);
678 if (ret < (int)PAGE_SIZE) {
679 pr_info("btrfsic: read @logical %llu failed!\n",
680 tmp_next_block_ctx.start);
681 btrfsic_release_block_ctx(&tmp_next_block_ctx);
682 kfree(selected_super);
683 return -1;
684 }
685
686 ret = btrfsic_process_metablock(state,
687 next_block,
688 &tmp_next_block_ctx,
689 BTRFS_MAX_LEVEL + 3, 1);
690 btrfsic_release_block_ctx(&tmp_next_block_ctx);
691 }
692 }
693
694 kfree(selected_super);
695 return ret;
696}
697
698static int btrfsic_process_superblock_dev_mirror(
699 struct btrfsic_state *state,
700 struct btrfsic_dev_state *dev_state,
701 struct btrfs_device *device,
702 int superblock_mirror_num,
703 struct btrfsic_dev_state **selected_dev_state,
704 struct btrfs_super_block *selected_super)
705{
706 struct btrfs_fs_info *fs_info = state->fs_info;
707 struct btrfs_super_block *super_tmp;
708 u64 dev_bytenr;
709 struct btrfsic_block *superblock_tmp;
710 int pass;
711 struct block_device *const superblock_bdev = device->bdev;
712 struct page *page;
713 struct address_space *mapping = superblock_bdev->bd_inode->i_mapping;
714 int ret = 0;
715
716 /* super block bytenr is always the unmapped device bytenr */
717 dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
718 if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->commit_total_bytes)
719 return -1;
720
721 page = read_cache_page_gfp(mapping, dev_bytenr >> PAGE_SHIFT, GFP_NOFS);
722 if (IS_ERR(page))
723 return -1;
724
725 super_tmp = page_address(page);
726
727 if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
728 btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
729 memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
730 btrfs_super_nodesize(super_tmp) != state->metablock_size ||
731 btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
732 ret = 0;
733 goto out;
734 }
735
736 superblock_tmp =
737 btrfsic_block_hashtable_lookup(superblock_bdev,
738 dev_bytenr,
739 &state->block_hashtable);
740 if (NULL == superblock_tmp) {
741 superblock_tmp = btrfsic_block_alloc();
742 if (NULL == superblock_tmp) {
743 ret = -1;
744 goto out;
745 }
746 /* for superblock, only the dev_bytenr makes sense */
747 superblock_tmp->dev_bytenr = dev_bytenr;
748 superblock_tmp->dev_state = dev_state;
749 superblock_tmp->logical_bytenr = dev_bytenr;
750 superblock_tmp->generation = btrfs_super_generation(super_tmp);
751 superblock_tmp->is_metadata = 1;
752 superblock_tmp->is_superblock = 1;
753 superblock_tmp->is_iodone = 1;
754 superblock_tmp->never_written = 0;
755 superblock_tmp->mirror_num = 1 + superblock_mirror_num;
756 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
757 btrfs_info_in_rcu(fs_info,
758 "new initial S-block (bdev %p, %s) @%llu (%pg/%llu/%d)",
759 superblock_bdev,
760 btrfs_dev_name(device), dev_bytenr,
761 dev_state->bdev, dev_bytenr,
762 superblock_mirror_num);
763 list_add(&superblock_tmp->all_blocks_node,
764 &state->all_blocks_list);
765 btrfsic_block_hashtable_add(superblock_tmp,
766 &state->block_hashtable);
767 }
768
769 /* select the one with the highest generation field */
770 if (btrfs_super_generation(super_tmp) >
771 state->max_superblock_generation ||
772 0 == state->max_superblock_generation) {
773 memcpy(selected_super, super_tmp, sizeof(*selected_super));
774 *selected_dev_state = dev_state;
775 state->max_superblock_generation =
776 btrfs_super_generation(super_tmp);
777 state->latest_superblock = superblock_tmp;
778 }
779
780 for (pass = 0; pass < 3; pass++) {
781 u64 next_bytenr;
782 int num_copies;
783 int mirror_num;
784 const char *additional_string = NULL;
785 struct btrfs_disk_key tmp_disk_key;
786
787 tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
788 tmp_disk_key.offset = 0;
789 switch (pass) {
790 case 0:
791 btrfs_set_disk_key_objectid(&tmp_disk_key,
792 BTRFS_ROOT_TREE_OBJECTID);
793 additional_string = "initial root ";
794 next_bytenr = btrfs_super_root(super_tmp);
795 break;
796 case 1:
797 btrfs_set_disk_key_objectid(&tmp_disk_key,
798 BTRFS_CHUNK_TREE_OBJECTID);
799 additional_string = "initial chunk ";
800 next_bytenr = btrfs_super_chunk_root(super_tmp);
801 break;
802 case 2:
803 btrfs_set_disk_key_objectid(&tmp_disk_key,
804 BTRFS_TREE_LOG_OBJECTID);
805 additional_string = "initial log ";
806 next_bytenr = btrfs_super_log_root(super_tmp);
807 if (0 == next_bytenr)
808 continue;
809 break;
810 }
811
812 num_copies = btrfs_num_copies(fs_info, next_bytenr,
813 state->metablock_size);
814 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
815 pr_info("num_copies(log_bytenr=%llu) = %d\n",
816 next_bytenr, num_copies);
817 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
818 struct btrfsic_block *next_block;
819 struct btrfsic_block_data_ctx tmp_next_block_ctx;
820 struct btrfsic_block_link *l;
821
822 if (btrfsic_map_block(state, next_bytenr,
823 state->metablock_size,
824 &tmp_next_block_ctx,
825 mirror_num)) {
826 pr_info("btrfsic: btrfsic_map_block(bytenr @%llu, mirror %d) failed!\n",
827 next_bytenr, mirror_num);
828 ret = -1;
829 goto out;
830 }
831
832 next_block = btrfsic_block_lookup_or_add(
833 state, &tmp_next_block_ctx,
834 additional_string, 1, 1, 0,
835 mirror_num, NULL);
836 if (NULL == next_block) {
837 btrfsic_release_block_ctx(&tmp_next_block_ctx);
838 ret = -1;
839 goto out;
840 }
841
842 next_block->disk_key = tmp_disk_key;
843 next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
844 l = btrfsic_block_link_lookup_or_add(
845 state, &tmp_next_block_ctx,
846 next_block, superblock_tmp,
847 BTRFSIC_GENERATION_UNKNOWN);
848 btrfsic_release_block_ctx(&tmp_next_block_ctx);
849 if (NULL == l) {
850 ret = -1;
851 goto out;
852 }
853 }
854 }
855 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
856 btrfsic_dump_tree_sub(state, superblock_tmp, 0);
857
858out:
859 put_page(page);
860 return ret;
861}
862
863static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
864{
865 struct btrfsic_stack_frame *sf;
866
867 sf = kzalloc(sizeof(*sf), GFP_NOFS);
868 if (sf)
869 sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
870 return sf;
871}
872
873static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
874{
875 BUG_ON(!(NULL == sf ||
876 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
877 kfree(sf);
878}
879
880static noinline_for_stack int btrfsic_process_metablock(
881 struct btrfsic_state *state,
882 struct btrfsic_block *const first_block,
883 struct btrfsic_block_data_ctx *const first_block_ctx,
884 int first_limit_nesting, int force_iodone_flag)
885{
886 struct btrfsic_stack_frame initial_stack_frame = { 0 };
887 struct btrfsic_stack_frame *sf;
888 struct btrfsic_stack_frame *next_stack;
889 struct btrfs_header *const first_hdr =
890 (struct btrfs_header *)first_block_ctx->datav[0];
891
892 BUG_ON(!first_hdr);
893 sf = &initial_stack_frame;
894 sf->error = 0;
895 sf->i = -1;
896 sf->limit_nesting = first_limit_nesting;
897 sf->block = first_block;
898 sf->block_ctx = first_block_ctx;
899 sf->next_block = NULL;
900 sf->hdr = first_hdr;
901 sf->prev = NULL;
902
903continue_with_new_stack_frame:
904 sf->block->generation = btrfs_stack_header_generation(sf->hdr);
905 if (0 == sf->hdr->level) {
906 struct btrfs_leaf *const leafhdr =
907 (struct btrfs_leaf *)sf->hdr;
908
909 if (-1 == sf->i) {
910 sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
911
912 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
913 pr_info("leaf %llu items %d generation %llu owner %llu\n",
914 sf->block_ctx->start, sf->nr,
915 btrfs_stack_header_generation(
916 &leafhdr->header),
917 btrfs_stack_header_owner(
918 &leafhdr->header));
919 }
920
921continue_with_current_leaf_stack_frame:
922 if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
923 sf->i++;
924 sf->num_copies = 0;
925 }
926
927 if (sf->i < sf->nr) {
928 struct btrfs_item disk_item;
929 u32 disk_item_offset =
930 (uintptr_t)(leafhdr->items + sf->i) -
931 (uintptr_t)leafhdr;
932 struct btrfs_disk_key *disk_key;
933 u8 type;
934 u32 item_offset;
935 u32 item_size;
936
937 if (disk_item_offset + sizeof(struct btrfs_item) >
938 sf->block_ctx->len) {
939leaf_item_out_of_bounce_error:
940 pr_info(
941 "btrfsic: leaf item out of bounce at logical %llu, dev %pg\n",
942 sf->block_ctx->start,
943 sf->block_ctx->dev->bdev);
944 goto one_stack_frame_backwards;
945 }
946 btrfsic_read_from_block_data(sf->block_ctx,
947 &disk_item,
948 disk_item_offset,
949 sizeof(struct btrfs_item));
950 item_offset = btrfs_stack_item_offset(&disk_item);
951 item_size = btrfs_stack_item_size(&disk_item);
952 disk_key = &disk_item.key;
953 type = btrfs_disk_key_type(disk_key);
954
955 if (BTRFS_ROOT_ITEM_KEY == type) {
956 struct btrfs_root_item root_item;
957 u32 root_item_offset;
958 u64 next_bytenr;
959
960 root_item_offset = item_offset +
961 offsetof(struct btrfs_leaf, items);
962 if (root_item_offset + item_size >
963 sf->block_ctx->len)
964 goto leaf_item_out_of_bounce_error;
965 btrfsic_read_from_block_data(
966 sf->block_ctx, &root_item,
967 root_item_offset,
968 item_size);
969 next_bytenr = btrfs_root_bytenr(&root_item);
970
971 sf->error =
972 btrfsic_create_link_to_next_block(
973 state,
974 sf->block,
975 sf->block_ctx,
976 next_bytenr,
977 sf->limit_nesting,
978 &sf->next_block_ctx,
979 &sf->next_block,
980 force_iodone_flag,
981 &sf->num_copies,
982 &sf->mirror_num,
983 disk_key,
984 btrfs_root_generation(
985 &root_item));
986 if (sf->error)
987 goto one_stack_frame_backwards;
988
989 if (NULL != sf->next_block) {
990 struct btrfs_header *const next_hdr =
991 (struct btrfs_header *)
992 sf->next_block_ctx.datav[0];
993
994 next_stack =
995 btrfsic_stack_frame_alloc();
996 if (NULL == next_stack) {
997 sf->error = -1;
998 btrfsic_release_block_ctx(
999 &sf->
1000 next_block_ctx);
1001 goto one_stack_frame_backwards;
1002 }
1003
1004 next_stack->i = -1;
1005 next_stack->block = sf->next_block;
1006 next_stack->block_ctx =
1007 &sf->next_block_ctx;
1008 next_stack->next_block = NULL;
1009 next_stack->hdr = next_hdr;
1010 next_stack->limit_nesting =
1011 sf->limit_nesting - 1;
1012 next_stack->prev = sf;
1013 sf = next_stack;
1014 goto continue_with_new_stack_frame;
1015 }
1016 } else if (BTRFS_EXTENT_DATA_KEY == type &&
1017 state->include_extent_data) {
1018 sf->error = btrfsic_handle_extent_data(
1019 state,
1020 sf->block,
1021 sf->block_ctx,
1022 item_offset,
1023 force_iodone_flag);
1024 if (sf->error)
1025 goto one_stack_frame_backwards;
1026 }
1027
1028 goto continue_with_current_leaf_stack_frame;
1029 }
1030 } else {
1031 struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1032
1033 if (-1 == sf->i) {
1034 sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1035
1036 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1037 pr_info("node %llu level %d items %d generation %llu owner %llu\n",
1038 sf->block_ctx->start,
1039 nodehdr->header.level, sf->nr,
1040 btrfs_stack_header_generation(
1041 &nodehdr->header),
1042 btrfs_stack_header_owner(
1043 &nodehdr->header));
1044 }
1045
1046continue_with_current_node_stack_frame:
1047 if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1048 sf->i++;
1049 sf->num_copies = 0;
1050 }
1051
1052 if (sf->i < sf->nr) {
1053 struct btrfs_key_ptr key_ptr;
1054 u32 key_ptr_offset;
1055 u64 next_bytenr;
1056
1057 key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1058 (uintptr_t)nodehdr;
1059 if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1060 sf->block_ctx->len) {
1061 pr_info(
1062 "btrfsic: node item out of bounce at logical %llu, dev %pg\n",
1063 sf->block_ctx->start,
1064 sf->block_ctx->dev->bdev);
1065 goto one_stack_frame_backwards;
1066 }
1067 btrfsic_read_from_block_data(
1068 sf->block_ctx, &key_ptr, key_ptr_offset,
1069 sizeof(struct btrfs_key_ptr));
1070 next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1071
1072 sf->error = btrfsic_create_link_to_next_block(
1073 state,
1074 sf->block,
1075 sf->block_ctx,
1076 next_bytenr,
1077 sf->limit_nesting,
1078 &sf->next_block_ctx,
1079 &sf->next_block,
1080 force_iodone_flag,
1081 &sf->num_copies,
1082 &sf->mirror_num,
1083 &key_ptr.key,
1084 btrfs_stack_key_generation(&key_ptr));
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 = btrfsic_stack_frame_alloc();
1094 if (NULL == next_stack) {
1095 sf->error = -1;
1096 goto one_stack_frame_backwards;
1097 }
1098
1099 next_stack->i = -1;
1100 next_stack->block = sf->next_block;
1101 next_stack->block_ctx = &sf->next_block_ctx;
1102 next_stack->next_block = NULL;
1103 next_stack->hdr = next_hdr;
1104 next_stack->limit_nesting =
1105 sf->limit_nesting - 1;
1106 next_stack->prev = sf;
1107 sf = next_stack;
1108 goto continue_with_new_stack_frame;
1109 }
1110
1111 goto continue_with_current_node_stack_frame;
1112 }
1113 }
1114
1115one_stack_frame_backwards:
1116 if (NULL != sf->prev) {
1117 struct btrfsic_stack_frame *const prev = sf->prev;
1118
1119 /* the one for the initial block is freed in the caller */
1120 btrfsic_release_block_ctx(sf->block_ctx);
1121
1122 if (sf->error) {
1123 prev->error = sf->error;
1124 btrfsic_stack_frame_free(sf);
1125 sf = prev;
1126 goto one_stack_frame_backwards;
1127 }
1128
1129 btrfsic_stack_frame_free(sf);
1130 sf = prev;
1131 goto continue_with_new_stack_frame;
1132 } else {
1133 BUG_ON(&initial_stack_frame != sf);
1134 }
1135
1136 return sf->error;
1137}
1138
1139static void btrfsic_read_from_block_data(
1140 struct btrfsic_block_data_ctx *block_ctx,
1141 void *dstv, u32 offset, size_t len)
1142{
1143 size_t cur;
1144 size_t pgoff;
1145 char *kaddr;
1146 char *dst = (char *)dstv;
1147 size_t start_offset = offset_in_page(block_ctx->start);
1148 unsigned long i = (start_offset + offset) >> PAGE_SHIFT;
1149
1150 WARN_ON(offset + len > block_ctx->len);
1151 pgoff = offset_in_page(start_offset + offset);
1152
1153 while (len > 0) {
1154 cur = min(len, ((size_t)PAGE_SIZE - pgoff));
1155 BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_SIZE));
1156 kaddr = block_ctx->datav[i];
1157 memcpy(dst, kaddr + pgoff, cur);
1158
1159 dst += cur;
1160 len -= cur;
1161 pgoff = 0;
1162 i++;
1163 }
1164}
1165
1166static int btrfsic_create_link_to_next_block(
1167 struct btrfsic_state *state,
1168 struct btrfsic_block *block,
1169 struct btrfsic_block_data_ctx *block_ctx,
1170 u64 next_bytenr,
1171 int limit_nesting,
1172 struct btrfsic_block_data_ctx *next_block_ctx,
1173 struct btrfsic_block **next_blockp,
1174 int force_iodone_flag,
1175 int *num_copiesp, int *mirror_nump,
1176 struct btrfs_disk_key *disk_key,
1177 u64 parent_generation)
1178{
1179 struct btrfs_fs_info *fs_info = state->fs_info;
1180 struct btrfsic_block *next_block = NULL;
1181 int ret;
1182 struct btrfsic_block_link *l;
1183 int did_alloc_block_link;
1184 int block_was_created;
1185
1186 *next_blockp = NULL;
1187 if (0 == *num_copiesp) {
1188 *num_copiesp = btrfs_num_copies(fs_info, next_bytenr,
1189 state->metablock_size);
1190 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1191 pr_info("num_copies(log_bytenr=%llu) = %d\n",
1192 next_bytenr, *num_copiesp);
1193 *mirror_nump = 1;
1194 }
1195
1196 if (*mirror_nump > *num_copiesp)
1197 return 0;
1198
1199 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1200 pr_info("btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1201 *mirror_nump);
1202 ret = btrfsic_map_block(state, next_bytenr,
1203 state->metablock_size,
1204 next_block_ctx, *mirror_nump);
1205 if (ret) {
1206 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1207 next_bytenr, *mirror_nump);
1208 btrfsic_release_block_ctx(next_block_ctx);
1209 *next_blockp = NULL;
1210 return -1;
1211 }
1212
1213 next_block = btrfsic_block_lookup_or_add(state,
1214 next_block_ctx, "referenced ",
1215 1, force_iodone_flag,
1216 !force_iodone_flag,
1217 *mirror_nump,
1218 &block_was_created);
1219 if (NULL == next_block) {
1220 btrfsic_release_block_ctx(next_block_ctx);
1221 *next_blockp = NULL;
1222 return -1;
1223 }
1224 if (block_was_created) {
1225 l = NULL;
1226 next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1227 } else {
1228 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1229 if (next_block->logical_bytenr != next_bytenr &&
1230 !(!next_block->is_metadata &&
1231 0 == next_block->logical_bytenr))
1232 pr_info(
1233"referenced block @%llu (%pg/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu)\n",
1234 next_bytenr, next_block_ctx->dev->bdev,
1235 next_block_ctx->dev_bytenr, *mirror_nump,
1236 btrfsic_get_block_type(state,
1237 next_block),
1238 next_block->logical_bytenr);
1239 else
1240 pr_info(
1241 "referenced block @%llu (%pg/%llu/%d) found in hash table, %c\n",
1242 next_bytenr, next_block_ctx->dev->bdev,
1243 next_block_ctx->dev_bytenr, *mirror_nump,
1244 btrfsic_get_block_type(state,
1245 next_block));
1246 }
1247 next_block->logical_bytenr = next_bytenr;
1248
1249 next_block->mirror_num = *mirror_nump;
1250 l = btrfsic_block_link_hashtable_lookup(
1251 next_block_ctx->dev->bdev,
1252 next_block_ctx->dev_bytenr,
1253 block_ctx->dev->bdev,
1254 block_ctx->dev_bytenr,
1255 &state->block_link_hashtable);
1256 }
1257
1258 next_block->disk_key = *disk_key;
1259 if (NULL == l) {
1260 l = btrfsic_block_link_alloc();
1261 if (NULL == l) {
1262 btrfsic_release_block_ctx(next_block_ctx);
1263 *next_blockp = NULL;
1264 return -1;
1265 }
1266
1267 did_alloc_block_link = 1;
1268 l->block_ref_to = next_block;
1269 l->block_ref_from = block;
1270 l->ref_cnt = 1;
1271 l->parent_generation = parent_generation;
1272
1273 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1274 btrfsic_print_add_link(state, l);
1275
1276 list_add(&l->node_ref_to, &block->ref_to_list);
1277 list_add(&l->node_ref_from, &next_block->ref_from_list);
1278
1279 btrfsic_block_link_hashtable_add(l,
1280 &state->block_link_hashtable);
1281 } else {
1282 did_alloc_block_link = 0;
1283 if (0 == limit_nesting) {
1284 l->ref_cnt++;
1285 l->parent_generation = parent_generation;
1286 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1287 btrfsic_print_add_link(state, l);
1288 }
1289 }
1290
1291 if (limit_nesting > 0 && did_alloc_block_link) {
1292 ret = btrfsic_read_block(state, next_block_ctx);
1293 if (ret < (int)next_block_ctx->len) {
1294 pr_info("btrfsic: read block @logical %llu failed!\n",
1295 next_bytenr);
1296 btrfsic_release_block_ctx(next_block_ctx);
1297 *next_blockp = NULL;
1298 return -1;
1299 }
1300
1301 *next_blockp = next_block;
1302 } else {
1303 *next_blockp = NULL;
1304 }
1305 (*mirror_nump)++;
1306
1307 return 0;
1308}
1309
1310static int btrfsic_handle_extent_data(
1311 struct btrfsic_state *state,
1312 struct btrfsic_block *block,
1313 struct btrfsic_block_data_ctx *block_ctx,
1314 u32 item_offset, int force_iodone_flag)
1315{
1316 struct btrfs_fs_info *fs_info = state->fs_info;
1317 struct btrfs_file_extent_item file_extent_item;
1318 u64 file_extent_item_offset;
1319 u64 next_bytenr;
1320 u64 num_bytes;
1321 u64 generation;
1322 struct btrfsic_block_link *l;
1323 int ret;
1324
1325 file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1326 item_offset;
1327 if (file_extent_item_offset +
1328 offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1329 block_ctx->len) {
1330 pr_info("btrfsic: file item out of bounce at logical %llu, dev %pg\n",
1331 block_ctx->start, block_ctx->dev->bdev);
1332 return -1;
1333 }
1334
1335 btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1336 file_extent_item_offset,
1337 offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1338 if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1339 btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1340 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1341 pr_info("extent_data: type %u, disk_bytenr = %llu\n",
1342 file_extent_item.type,
1343 btrfs_stack_file_extent_disk_bytenr(
1344 &file_extent_item));
1345 return 0;
1346 }
1347
1348 if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1349 block_ctx->len) {
1350 pr_info("btrfsic: file item out of bounce at logical %llu, dev %pg\n",
1351 block_ctx->start, block_ctx->dev->bdev);
1352 return -1;
1353 }
1354 btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1355 file_extent_item_offset,
1356 sizeof(struct btrfs_file_extent_item));
1357 next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item);
1358 if (btrfs_stack_file_extent_compression(&file_extent_item) ==
1359 BTRFS_COMPRESS_NONE) {
1360 next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item);
1361 num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1362 } else {
1363 num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item);
1364 }
1365 generation = btrfs_stack_file_extent_generation(&file_extent_item);
1366
1367 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1368 pr_info("extent_data: type %u, disk_bytenr = %llu, offset = %llu, num_bytes = %llu\n",
1369 file_extent_item.type,
1370 btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1371 btrfs_stack_file_extent_offset(&file_extent_item),
1372 num_bytes);
1373 while (num_bytes > 0) {
1374 u32 chunk_len;
1375 int num_copies;
1376 int mirror_num;
1377
1378 if (num_bytes > state->datablock_size)
1379 chunk_len = state->datablock_size;
1380 else
1381 chunk_len = num_bytes;
1382
1383 num_copies = btrfs_num_copies(fs_info, next_bytenr,
1384 state->datablock_size);
1385 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1386 pr_info("num_copies(log_bytenr=%llu) = %d\n",
1387 next_bytenr, num_copies);
1388 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1389 struct btrfsic_block_data_ctx next_block_ctx;
1390 struct btrfsic_block *next_block;
1391 int block_was_created;
1392
1393 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1394 pr_info("btrfsic_handle_extent_data(mirror_num=%d)\n",
1395 mirror_num);
1396 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1397 pr_info("\tdisk_bytenr = %llu, num_bytes %u\n",
1398 next_bytenr, chunk_len);
1399 ret = btrfsic_map_block(state, next_bytenr,
1400 chunk_len, &next_block_ctx,
1401 mirror_num);
1402 if (ret) {
1403 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1404 next_bytenr, mirror_num);
1405 return -1;
1406 }
1407
1408 next_block = btrfsic_block_lookup_or_add(
1409 state,
1410 &next_block_ctx,
1411 "referenced ",
1412 0,
1413 force_iodone_flag,
1414 !force_iodone_flag,
1415 mirror_num,
1416 &block_was_created);
1417 if (NULL == next_block) {
1418 btrfsic_release_block_ctx(&next_block_ctx);
1419 return -1;
1420 }
1421 if (!block_was_created) {
1422 if ((state->print_mask &
1423 BTRFSIC_PRINT_MASK_VERBOSE) &&
1424 next_block->logical_bytenr != next_bytenr &&
1425 !(!next_block->is_metadata &&
1426 0 == next_block->logical_bytenr)) {
1427 pr_info(
1428"referenced block @%llu (%pg/%llu/%d) found in hash table, D, bytenr mismatch (!= stored %llu)\n",
1429 next_bytenr,
1430 next_block_ctx.dev->bdev,
1431 next_block_ctx.dev_bytenr,
1432 mirror_num,
1433 next_block->logical_bytenr);
1434 }
1435 next_block->logical_bytenr = next_bytenr;
1436 next_block->mirror_num = mirror_num;
1437 }
1438
1439 l = btrfsic_block_link_lookup_or_add(state,
1440 &next_block_ctx,
1441 next_block, block,
1442 generation);
1443 btrfsic_release_block_ctx(&next_block_ctx);
1444 if (NULL == l)
1445 return -1;
1446 }
1447
1448 next_bytenr += chunk_len;
1449 num_bytes -= chunk_len;
1450 }
1451
1452 return 0;
1453}
1454
1455static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1456 struct btrfsic_block_data_ctx *block_ctx_out,
1457 int mirror_num)
1458{
1459 struct btrfs_fs_info *fs_info = state->fs_info;
1460 int ret;
1461 u64 length;
1462 struct btrfs_io_context *multi = NULL;
1463 struct btrfs_device *device;
1464
1465 length = len;
1466 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ,
1467 bytenr, &length, &multi, mirror_num);
1468
1469 if (ret) {
1470 block_ctx_out->start = 0;
1471 block_ctx_out->dev_bytenr = 0;
1472 block_ctx_out->len = 0;
1473 block_ctx_out->dev = NULL;
1474 block_ctx_out->datav = NULL;
1475 block_ctx_out->pagev = NULL;
1476 block_ctx_out->mem_to_free = NULL;
1477
1478 return ret;
1479 }
1480
1481 device = multi->stripes[0].dev;
1482 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
1483 !device->bdev || !device->name)
1484 block_ctx_out->dev = NULL;
1485 else
1486 block_ctx_out->dev = btrfsic_dev_state_lookup(
1487 device->bdev->bd_dev);
1488 block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1489 block_ctx_out->start = bytenr;
1490 block_ctx_out->len = len;
1491 block_ctx_out->datav = NULL;
1492 block_ctx_out->pagev = NULL;
1493 block_ctx_out->mem_to_free = NULL;
1494
1495 kfree(multi);
1496 if (NULL == block_ctx_out->dev) {
1497 ret = -ENXIO;
1498 pr_info("btrfsic: error, cannot lookup dev (#1)!\n");
1499 }
1500
1501 return ret;
1502}
1503
1504static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1505{
1506 if (block_ctx->mem_to_free) {
1507 unsigned int num_pages;
1508
1509 BUG_ON(!block_ctx->datav);
1510 BUG_ON(!block_ctx->pagev);
1511 num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1512 PAGE_SHIFT;
1513 /* Pages must be unmapped in reverse order */
1514 while (num_pages > 0) {
1515 num_pages--;
1516 if (block_ctx->datav[num_pages])
1517 block_ctx->datav[num_pages] = NULL;
1518 if (block_ctx->pagev[num_pages]) {
1519 __free_page(block_ctx->pagev[num_pages]);
1520 block_ctx->pagev[num_pages] = NULL;
1521 }
1522 }
1523
1524 kfree(block_ctx->mem_to_free);
1525 block_ctx->mem_to_free = NULL;
1526 block_ctx->pagev = NULL;
1527 block_ctx->datav = NULL;
1528 }
1529}
1530
1531static int btrfsic_read_block(struct btrfsic_state *state,
1532 struct btrfsic_block_data_ctx *block_ctx)
1533{
1534 unsigned int num_pages;
1535 unsigned int i;
1536 size_t size;
1537 u64 dev_bytenr;
1538 int ret;
1539
1540 BUG_ON(block_ctx->datav);
1541 BUG_ON(block_ctx->pagev);
1542 BUG_ON(block_ctx->mem_to_free);
1543 if (!PAGE_ALIGNED(block_ctx->dev_bytenr)) {
1544 pr_info("btrfsic: read_block() with unaligned bytenr %llu\n",
1545 block_ctx->dev_bytenr);
1546 return -1;
1547 }
1548
1549 num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1550 PAGE_SHIFT;
1551 size = sizeof(*block_ctx->datav) + sizeof(*block_ctx->pagev);
1552 block_ctx->mem_to_free = kcalloc(num_pages, size, GFP_NOFS);
1553 if (!block_ctx->mem_to_free)
1554 return -ENOMEM;
1555 block_ctx->datav = block_ctx->mem_to_free;
1556 block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1557 ret = btrfs_alloc_page_array(num_pages, block_ctx->pagev);
1558 if (ret)
1559 return ret;
1560
1561 dev_bytenr = block_ctx->dev_bytenr;
1562 for (i = 0; i < num_pages;) {
1563 struct bio *bio;
1564 unsigned int j;
1565
1566 bio = bio_alloc(block_ctx->dev->bdev, num_pages - i,
1567 REQ_OP_READ, GFP_NOFS);
1568 bio->bi_iter.bi_sector = dev_bytenr >> 9;
1569
1570 for (j = i; j < num_pages; j++) {
1571 ret = bio_add_page(bio, block_ctx->pagev[j],
1572 PAGE_SIZE, 0);
1573 if (PAGE_SIZE != ret)
1574 break;
1575 }
1576 if (j == i) {
1577 pr_info("btrfsic: error, failed to add a single page!\n");
1578 return -1;
1579 }
1580 if (submit_bio_wait(bio)) {
1581 pr_info("btrfsic: read error at logical %llu dev %pg!\n",
1582 block_ctx->start, block_ctx->dev->bdev);
1583 bio_put(bio);
1584 return -1;
1585 }
1586 bio_put(bio);
1587 dev_bytenr += (j - i) * PAGE_SIZE;
1588 i = j;
1589 }
1590 for (i = 0; i < num_pages; i++)
1591 block_ctx->datav[i] = page_address(block_ctx->pagev[i]);
1592
1593 return block_ctx->len;
1594}
1595
1596static void btrfsic_dump_database(struct btrfsic_state *state)
1597{
1598 const struct btrfsic_block *b_all;
1599
1600 BUG_ON(NULL == state);
1601
1602 pr_info("all_blocks_list:\n");
1603 list_for_each_entry(b_all, &state->all_blocks_list, all_blocks_node) {
1604 const struct btrfsic_block_link *l;
1605
1606 pr_info("%c-block @%llu (%pg/%llu/%d)\n",
1607 btrfsic_get_block_type(state, b_all),
1608 b_all->logical_bytenr, b_all->dev_state->bdev,
1609 b_all->dev_bytenr, b_all->mirror_num);
1610
1611 list_for_each_entry(l, &b_all->ref_to_list, node_ref_to) {
1612 pr_info(
1613 " %c @%llu (%pg/%llu/%d) refers %u* to %c @%llu (%pg/%llu/%d)\n",
1614 btrfsic_get_block_type(state, b_all),
1615 b_all->logical_bytenr, b_all->dev_state->bdev,
1616 b_all->dev_bytenr, b_all->mirror_num,
1617 l->ref_cnt,
1618 btrfsic_get_block_type(state, l->block_ref_to),
1619 l->block_ref_to->logical_bytenr,
1620 l->block_ref_to->dev_state->bdev,
1621 l->block_ref_to->dev_bytenr,
1622 l->block_ref_to->mirror_num);
1623 }
1624
1625 list_for_each_entry(l, &b_all->ref_from_list, node_ref_from) {
1626 pr_info(
1627 " %c @%llu (%pg/%llu/%d) is ref %u* from %c @%llu (%pg/%llu/%d)\n",
1628 btrfsic_get_block_type(state, b_all),
1629 b_all->logical_bytenr, b_all->dev_state->bdev,
1630 b_all->dev_bytenr, b_all->mirror_num,
1631 l->ref_cnt,
1632 btrfsic_get_block_type(state, l->block_ref_from),
1633 l->block_ref_from->logical_bytenr,
1634 l->block_ref_from->dev_state->bdev,
1635 l->block_ref_from->dev_bytenr,
1636 l->block_ref_from->mirror_num);
1637 }
1638
1639 pr_info("\n");
1640 }
1641}
1642
1643/*
1644 * Test whether the disk block contains a tree block (leaf or node)
1645 * (note that this test fails for the super block)
1646 */
1647static noinline_for_stack int btrfsic_test_for_metadata(
1648 struct btrfsic_state *state,
1649 char **datav, unsigned int num_pages)
1650{
1651 struct btrfs_fs_info *fs_info = state->fs_info;
1652 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
1653 struct btrfs_header *h;
1654 u8 csum[BTRFS_CSUM_SIZE];
1655 unsigned int i;
1656
1657 if (num_pages * PAGE_SIZE < state->metablock_size)
1658 return 1; /* not metadata */
1659 num_pages = state->metablock_size >> PAGE_SHIFT;
1660 h = (struct btrfs_header *)datav[0];
1661
1662 if (memcmp(h->fsid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE))
1663 return 1;
1664
1665 shash->tfm = fs_info->csum_shash;
1666 crypto_shash_init(shash);
1667
1668 for (i = 0; i < num_pages; i++) {
1669 u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1670 size_t sublen = i ? PAGE_SIZE :
1671 (PAGE_SIZE - BTRFS_CSUM_SIZE);
1672
1673 crypto_shash_update(shash, data, sublen);
1674 }
1675 crypto_shash_final(shash, csum);
1676 if (memcmp(csum, h->csum, fs_info->csum_size))
1677 return 1;
1678
1679 return 0; /* is metadata */
1680}
1681
1682static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1683 u64 dev_bytenr, char **mapped_datav,
1684 unsigned int num_pages,
1685 struct bio *bio, int *bio_is_patched,
1686 blk_opf_t submit_bio_bh_rw)
1687{
1688 int is_metadata;
1689 struct btrfsic_block *block;
1690 struct btrfsic_block_data_ctx block_ctx;
1691 int ret;
1692 struct btrfsic_state *state = dev_state->state;
1693 struct block_device *bdev = dev_state->bdev;
1694 unsigned int processed_len;
1695
1696 if (NULL != bio_is_patched)
1697 *bio_is_patched = 0;
1698
1699again:
1700 if (num_pages == 0)
1701 return;
1702
1703 processed_len = 0;
1704 is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1705 num_pages));
1706
1707 block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1708 &state->block_hashtable);
1709 if (NULL != block) {
1710 u64 bytenr = 0;
1711 struct btrfsic_block_link *l, *tmp;
1712
1713 if (block->is_superblock) {
1714 bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1715 mapped_datav[0]);
1716 if (num_pages * PAGE_SIZE <
1717 BTRFS_SUPER_INFO_SIZE) {
1718 pr_info("btrfsic: cannot work with too short bios!\n");
1719 return;
1720 }
1721 is_metadata = 1;
1722 BUG_ON(!PAGE_ALIGNED(BTRFS_SUPER_INFO_SIZE));
1723 processed_len = BTRFS_SUPER_INFO_SIZE;
1724 if (state->print_mask &
1725 BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1726 pr_info("[before new superblock is written]:\n");
1727 btrfsic_dump_tree_sub(state, block, 0);
1728 }
1729 }
1730 if (is_metadata) {
1731 if (!block->is_superblock) {
1732 if (num_pages * PAGE_SIZE <
1733 state->metablock_size) {
1734 pr_info("btrfsic: cannot work with too short bios!\n");
1735 return;
1736 }
1737 processed_len = state->metablock_size;
1738 bytenr = btrfs_stack_header_bytenr(
1739 (struct btrfs_header *)
1740 mapped_datav[0]);
1741 btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1742 dev_state,
1743 dev_bytenr);
1744 }
1745 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1746 if (block->logical_bytenr != bytenr &&
1747 !(!block->is_metadata &&
1748 block->logical_bytenr == 0))
1749 pr_info(
1750"written block @%llu (%pg/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu)\n",
1751 bytenr, dev_state->bdev,
1752 dev_bytenr,
1753 block->mirror_num,
1754 btrfsic_get_block_type(state,
1755 block),
1756 block->logical_bytenr);
1757 else
1758 pr_info(
1759 "written block @%llu (%pg/%llu/%d) found in hash table, %c\n",
1760 bytenr, dev_state->bdev,
1761 dev_bytenr, block->mirror_num,
1762 btrfsic_get_block_type(state,
1763 block));
1764 }
1765 block->logical_bytenr = bytenr;
1766 } else {
1767 if (num_pages * PAGE_SIZE <
1768 state->datablock_size) {
1769 pr_info("btrfsic: cannot work with too short bios!\n");
1770 return;
1771 }
1772 processed_len = state->datablock_size;
1773 bytenr = block->logical_bytenr;
1774 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1775 pr_info(
1776 "written block @%llu (%pg/%llu/%d) found in hash table, %c\n",
1777 bytenr, dev_state->bdev, dev_bytenr,
1778 block->mirror_num,
1779 btrfsic_get_block_type(state, block));
1780 }
1781
1782 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1783 pr_info("ref_to_list: %cE, ref_from_list: %cE\n",
1784 list_empty(&block->ref_to_list) ? ' ' : '!',
1785 list_empty(&block->ref_from_list) ? ' ' : '!');
1786 if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1787 pr_info(
1788"btrfs: attempt to overwrite %c-block @%llu (%pg/%llu/%d), old(gen=%llu, objectid=%llu, type=%d, offset=%llu), new(gen=%llu), which is referenced by most recent superblock (superblockgen=%llu)!\n",
1789 btrfsic_get_block_type(state, block), bytenr,
1790 dev_state->bdev, dev_bytenr, block->mirror_num,
1791 block->generation,
1792 btrfs_disk_key_objectid(&block->disk_key),
1793 block->disk_key.type,
1794 btrfs_disk_key_offset(&block->disk_key),
1795 btrfs_stack_header_generation(
1796 (struct btrfs_header *) mapped_datav[0]),
1797 state->max_superblock_generation);
1798 btrfsic_dump_tree(state);
1799 }
1800
1801 if (!block->is_iodone && !block->never_written) {
1802 pr_info(
1803"btrfs: attempt to overwrite %c-block @%llu (%pg/%llu/%d), oldgen=%llu, newgen=%llu, which is not yet iodone!\n",
1804 btrfsic_get_block_type(state, block), bytenr,
1805 dev_state->bdev, dev_bytenr, block->mirror_num,
1806 block->generation,
1807 btrfs_stack_header_generation(
1808 (struct btrfs_header *)
1809 mapped_datav[0]));
1810 /* it would not be safe to go on */
1811 btrfsic_dump_tree(state);
1812 goto continue_loop;
1813 }
1814
1815 /*
1816 * Clear all references of this block. Do not free
1817 * the block itself even if is not referenced anymore
1818 * because it still carries valuable information
1819 * like whether it was ever written and IO completed.
1820 */
1821 list_for_each_entry_safe(l, tmp, &block->ref_to_list,
1822 node_ref_to) {
1823 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1824 btrfsic_print_rem_link(state, l);
1825 l->ref_cnt--;
1826 if (0 == l->ref_cnt) {
1827 list_del(&l->node_ref_to);
1828 list_del(&l->node_ref_from);
1829 btrfsic_block_link_hashtable_remove(l);
1830 btrfsic_block_link_free(l);
1831 }
1832 }
1833
1834 block_ctx.dev = dev_state;
1835 block_ctx.dev_bytenr = dev_bytenr;
1836 block_ctx.start = bytenr;
1837 block_ctx.len = processed_len;
1838 block_ctx.pagev = NULL;
1839 block_ctx.mem_to_free = NULL;
1840 block_ctx.datav = mapped_datav;
1841
1842 if (is_metadata || state->include_extent_data) {
1843 block->never_written = 0;
1844 block->iodone_w_error = 0;
1845 if (NULL != bio) {
1846 block->is_iodone = 0;
1847 BUG_ON(NULL == bio_is_patched);
1848 if (!*bio_is_patched) {
1849 block->orig_bio_private =
1850 bio->bi_private;
1851 block->orig_bio_end_io =
1852 bio->bi_end_io;
1853 block->next_in_same_bio = NULL;
1854 bio->bi_private = block;
1855 bio->bi_end_io = btrfsic_bio_end_io;
1856 *bio_is_patched = 1;
1857 } else {
1858 struct btrfsic_block *chained_block =
1859 (struct btrfsic_block *)
1860 bio->bi_private;
1861
1862 BUG_ON(NULL == chained_block);
1863 block->orig_bio_private =
1864 chained_block->orig_bio_private;
1865 block->orig_bio_end_io =
1866 chained_block->orig_bio_end_io;
1867 block->next_in_same_bio = chained_block;
1868 bio->bi_private = block;
1869 }
1870 } else {
1871 block->is_iodone = 1;
1872 block->orig_bio_private = NULL;
1873 block->orig_bio_end_io = NULL;
1874 block->next_in_same_bio = NULL;
1875 }
1876 }
1877
1878 block->flush_gen = dev_state->last_flush_gen + 1;
1879 block->submit_bio_bh_rw = submit_bio_bh_rw;
1880 if (is_metadata) {
1881 block->logical_bytenr = bytenr;
1882 block->is_metadata = 1;
1883 if (block->is_superblock) {
1884 BUG_ON(PAGE_SIZE !=
1885 BTRFS_SUPER_INFO_SIZE);
1886 ret = btrfsic_process_written_superblock(
1887 state,
1888 block,
1889 (struct btrfs_super_block *)
1890 mapped_datav[0]);
1891 if (state->print_mask &
1892 BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
1893 pr_info("[after new superblock is written]:\n");
1894 btrfsic_dump_tree_sub(state, block, 0);
1895 }
1896 } else {
1897 block->mirror_num = 0; /* unknown */
1898 ret = btrfsic_process_metablock(
1899 state,
1900 block,
1901 &block_ctx,
1902 0, 0);
1903 }
1904 if (ret)
1905 pr_info("btrfsic: btrfsic_process_metablock(root @%llu) failed!\n",
1906 dev_bytenr);
1907 } else {
1908 block->is_metadata = 0;
1909 block->mirror_num = 0; /* unknown */
1910 block->generation = BTRFSIC_GENERATION_UNKNOWN;
1911 if (!state->include_extent_data
1912 && list_empty(&block->ref_from_list)) {
1913 /*
1914 * disk block is overwritten with extent
1915 * data (not meta data) and we are configured
1916 * to not include extent data: take the
1917 * chance and free the block's memory
1918 */
1919 btrfsic_block_hashtable_remove(block);
1920 list_del(&block->all_blocks_node);
1921 btrfsic_block_free(block);
1922 }
1923 }
1924 btrfsic_release_block_ctx(&block_ctx);
1925 } else {
1926 /* block has not been found in hash table */
1927 u64 bytenr;
1928
1929 if (!is_metadata) {
1930 processed_len = state->datablock_size;
1931 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1932 pr_info(
1933 "written block (%pg/%llu/?) !found in hash table, D\n",
1934 dev_state->bdev, dev_bytenr);
1935 if (!state->include_extent_data) {
1936 /* ignore that written D block */
1937 goto continue_loop;
1938 }
1939
1940 /* this is getting ugly for the
1941 * include_extent_data case... */
1942 bytenr = 0; /* unknown */
1943 } else {
1944 processed_len = state->metablock_size;
1945 bytenr = btrfs_stack_header_bytenr(
1946 (struct btrfs_header *)
1947 mapped_datav[0]);
1948 btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
1949 dev_bytenr);
1950 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1951 pr_info(
1952 "written block @%llu (%pg/%llu/?) !found in hash table, M\n",
1953 bytenr, dev_state->bdev, dev_bytenr);
1954 }
1955
1956 block_ctx.dev = dev_state;
1957 block_ctx.dev_bytenr = dev_bytenr;
1958 block_ctx.start = bytenr;
1959 block_ctx.len = processed_len;
1960 block_ctx.pagev = NULL;
1961 block_ctx.mem_to_free = NULL;
1962 block_ctx.datav = mapped_datav;
1963
1964 block = btrfsic_block_alloc();
1965 if (NULL == block) {
1966 btrfsic_release_block_ctx(&block_ctx);
1967 goto continue_loop;
1968 }
1969 block->dev_state = dev_state;
1970 block->dev_bytenr = dev_bytenr;
1971 block->logical_bytenr = bytenr;
1972 block->is_metadata = is_metadata;
1973 block->never_written = 0;
1974 block->iodone_w_error = 0;
1975 block->mirror_num = 0; /* unknown */
1976 block->flush_gen = dev_state->last_flush_gen + 1;
1977 block->submit_bio_bh_rw = submit_bio_bh_rw;
1978 if (NULL != bio) {
1979 block->is_iodone = 0;
1980 BUG_ON(NULL == bio_is_patched);
1981 if (!*bio_is_patched) {
1982 block->orig_bio_private = bio->bi_private;
1983 block->orig_bio_end_io = bio->bi_end_io;
1984 block->next_in_same_bio = NULL;
1985 bio->bi_private = block;
1986 bio->bi_end_io = btrfsic_bio_end_io;
1987 *bio_is_patched = 1;
1988 } else {
1989 struct btrfsic_block *chained_block =
1990 (struct btrfsic_block *)
1991 bio->bi_private;
1992
1993 BUG_ON(NULL == chained_block);
1994 block->orig_bio_private =
1995 chained_block->orig_bio_private;
1996 block->orig_bio_end_io =
1997 chained_block->orig_bio_end_io;
1998 block->next_in_same_bio = chained_block;
1999 bio->bi_private = block;
2000 }
2001 } else {
2002 block->is_iodone = 1;
2003 block->orig_bio_private = NULL;
2004 block->orig_bio_end_io = NULL;
2005 block->next_in_same_bio = NULL;
2006 }
2007 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2008 pr_info("new written %c-block @%llu (%pg/%llu/%d)\n",
2009 is_metadata ? 'M' : 'D',
2010 block->logical_bytenr, block->dev_state->bdev,
2011 block->dev_bytenr, block->mirror_num);
2012 list_add(&block->all_blocks_node, &state->all_blocks_list);
2013 btrfsic_block_hashtable_add(block, &state->block_hashtable);
2014
2015 if (is_metadata) {
2016 ret = btrfsic_process_metablock(state, block,
2017 &block_ctx, 0, 0);
2018 if (ret)
2019 pr_info("btrfsic: process_metablock(root @%llu) failed!\n",
2020 dev_bytenr);
2021 }
2022 btrfsic_release_block_ctx(&block_ctx);
2023 }
2024
2025continue_loop:
2026 BUG_ON(!processed_len);
2027 dev_bytenr += processed_len;
2028 mapped_datav += processed_len >> PAGE_SHIFT;
2029 num_pages -= processed_len >> PAGE_SHIFT;
2030 goto again;
2031}
2032
2033static void btrfsic_bio_end_io(struct bio *bp)
2034{
2035 struct btrfsic_block *block = bp->bi_private;
2036 int iodone_w_error;
2037
2038 /* mutex is not held! This is not save if IO is not yet completed
2039 * on umount */
2040 iodone_w_error = 0;
2041 if (bp->bi_status)
2042 iodone_w_error = 1;
2043
2044 BUG_ON(NULL == block);
2045 bp->bi_private = block->orig_bio_private;
2046 bp->bi_end_io = block->orig_bio_end_io;
2047
2048 do {
2049 struct btrfsic_block *next_block;
2050 struct btrfsic_dev_state *const dev_state = block->dev_state;
2051
2052 if ((dev_state->state->print_mask &
2053 BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2054 pr_info("bio_end_io(err=%d) for %c @%llu (%pg/%llu/%d)\n",
2055 bp->bi_status,
2056 btrfsic_get_block_type(dev_state->state, block),
2057 block->logical_bytenr, dev_state->bdev,
2058 block->dev_bytenr, block->mirror_num);
2059 next_block = block->next_in_same_bio;
2060 block->iodone_w_error = iodone_w_error;
2061 if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2062 dev_state->last_flush_gen++;
2063 if ((dev_state->state->print_mask &
2064 BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2065 pr_info("bio_end_io() new %pg flush_gen=%llu\n",
2066 dev_state->bdev,
2067 dev_state->last_flush_gen);
2068 }
2069 if (block->submit_bio_bh_rw & REQ_FUA)
2070 block->flush_gen = 0; /* FUA completed means block is
2071 * on disk */
2072 block->is_iodone = 1; /* for FLUSH, this releases the block */
2073 block = next_block;
2074 } while (NULL != block);
2075
2076 bp->bi_end_io(bp);
2077}
2078
2079static int btrfsic_process_written_superblock(
2080 struct btrfsic_state *state,
2081 struct btrfsic_block *const superblock,
2082 struct btrfs_super_block *const super_hdr)
2083{
2084 struct btrfs_fs_info *fs_info = state->fs_info;
2085 int pass;
2086
2087 superblock->generation = btrfs_super_generation(super_hdr);
2088 if (!(superblock->generation > state->max_superblock_generation ||
2089 0 == state->max_superblock_generation)) {
2090 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2091 pr_info(
2092 "btrfsic: superblock @%llu (%pg/%llu/%d) with old gen %llu <= %llu\n",
2093 superblock->logical_bytenr,
2094 superblock->dev_state->bdev,
2095 superblock->dev_bytenr, superblock->mirror_num,
2096 btrfs_super_generation(super_hdr),
2097 state->max_superblock_generation);
2098 } else {
2099 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2100 pr_info(
2101 "btrfsic: got new superblock @%llu (%pg/%llu/%d) with new gen %llu > %llu\n",
2102 superblock->logical_bytenr,
2103 superblock->dev_state->bdev,
2104 superblock->dev_bytenr, superblock->mirror_num,
2105 btrfs_super_generation(super_hdr),
2106 state->max_superblock_generation);
2107
2108 state->max_superblock_generation =
2109 btrfs_super_generation(super_hdr);
2110 state->latest_superblock = superblock;
2111 }
2112
2113 for (pass = 0; pass < 3; pass++) {
2114 int ret;
2115 u64 next_bytenr;
2116 struct btrfsic_block *next_block;
2117 struct btrfsic_block_data_ctx tmp_next_block_ctx;
2118 struct btrfsic_block_link *l;
2119 int num_copies;
2120 int mirror_num;
2121 const char *additional_string = NULL;
2122 struct btrfs_disk_key tmp_disk_key = {0};
2123
2124 btrfs_set_disk_key_objectid(&tmp_disk_key,
2125 BTRFS_ROOT_ITEM_KEY);
2126 btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2127
2128 switch (pass) {
2129 case 0:
2130 btrfs_set_disk_key_objectid(&tmp_disk_key,
2131 BTRFS_ROOT_TREE_OBJECTID);
2132 additional_string = "root ";
2133 next_bytenr = btrfs_super_root(super_hdr);
2134 if (state->print_mask &
2135 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2136 pr_info("root@%llu\n", next_bytenr);
2137 break;
2138 case 1:
2139 btrfs_set_disk_key_objectid(&tmp_disk_key,
2140 BTRFS_CHUNK_TREE_OBJECTID);
2141 additional_string = "chunk ";
2142 next_bytenr = btrfs_super_chunk_root(super_hdr);
2143 if (state->print_mask &
2144 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2145 pr_info("chunk@%llu\n", next_bytenr);
2146 break;
2147 case 2:
2148 btrfs_set_disk_key_objectid(&tmp_disk_key,
2149 BTRFS_TREE_LOG_OBJECTID);
2150 additional_string = "log ";
2151 next_bytenr = btrfs_super_log_root(super_hdr);
2152 if (0 == next_bytenr)
2153 continue;
2154 if (state->print_mask &
2155 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2156 pr_info("log@%llu\n", next_bytenr);
2157 break;
2158 }
2159
2160 num_copies = btrfs_num_copies(fs_info, next_bytenr,
2161 BTRFS_SUPER_INFO_SIZE);
2162 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2163 pr_info("num_copies(log_bytenr=%llu) = %d\n",
2164 next_bytenr, num_copies);
2165 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2166 int was_created;
2167
2168 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2169 pr_info("btrfsic_process_written_superblock(mirror_num=%d)\n", mirror_num);
2170 ret = btrfsic_map_block(state, next_bytenr,
2171 BTRFS_SUPER_INFO_SIZE,
2172 &tmp_next_block_ctx,
2173 mirror_num);
2174 if (ret) {
2175 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
2176 next_bytenr, mirror_num);
2177 return -1;
2178 }
2179
2180 next_block = btrfsic_block_lookup_or_add(
2181 state,
2182 &tmp_next_block_ctx,
2183 additional_string,
2184 1, 0, 1,
2185 mirror_num,
2186 &was_created);
2187 if (NULL == next_block) {
2188 btrfsic_release_block_ctx(&tmp_next_block_ctx);
2189 return -1;
2190 }
2191
2192 next_block->disk_key = tmp_disk_key;
2193 if (was_created)
2194 next_block->generation =
2195 BTRFSIC_GENERATION_UNKNOWN;
2196 l = btrfsic_block_link_lookup_or_add(
2197 state,
2198 &tmp_next_block_ctx,
2199 next_block,
2200 superblock,
2201 BTRFSIC_GENERATION_UNKNOWN);
2202 btrfsic_release_block_ctx(&tmp_next_block_ctx);
2203 if (NULL == l)
2204 return -1;
2205 }
2206 }
2207
2208 if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2209 btrfsic_dump_tree(state);
2210
2211 return 0;
2212}
2213
2214static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2215 struct btrfsic_block *const block,
2216 int recursion_level)
2217{
2218 const struct btrfsic_block_link *l;
2219 int ret = 0;
2220
2221 if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2222 /*
2223 * Note that this situation can happen and does not
2224 * indicate an error in regular cases. It happens
2225 * when disk blocks are freed and later reused.
2226 * The check-integrity module is not aware of any
2227 * block free operations, it just recognizes block
2228 * write operations. Therefore it keeps the linkage
2229 * information for a block until a block is
2230 * rewritten. This can temporarily cause incorrect
2231 * and even circular linkage information. This
2232 * causes no harm unless such blocks are referenced
2233 * by the most recent super block.
2234 */
2235 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2236 pr_info("btrfsic: abort cyclic linkage (case 1).\n");
2237
2238 return ret;
2239 }
2240
2241 /*
2242 * This algorithm is recursive because the amount of used stack
2243 * space is very small and the max recursion depth is limited.
2244 */
2245 list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2246 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2247 pr_info(
2248 "rl=%d, %c @%llu (%pg/%llu/%d) %u* refers to %c @%llu (%pg/%llu/%d)\n",
2249 recursion_level,
2250 btrfsic_get_block_type(state, block),
2251 block->logical_bytenr, block->dev_state->bdev,
2252 block->dev_bytenr, block->mirror_num,
2253 l->ref_cnt,
2254 btrfsic_get_block_type(state, l->block_ref_to),
2255 l->block_ref_to->logical_bytenr,
2256 l->block_ref_to->dev_state->bdev,
2257 l->block_ref_to->dev_bytenr,
2258 l->block_ref_to->mirror_num);
2259 if (l->block_ref_to->never_written) {
2260 pr_info(
2261"btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) which is never written!\n",
2262 btrfsic_get_block_type(state, l->block_ref_to),
2263 l->block_ref_to->logical_bytenr,
2264 l->block_ref_to->dev_state->bdev,
2265 l->block_ref_to->dev_bytenr,
2266 l->block_ref_to->mirror_num);
2267 ret = -1;
2268 } else if (!l->block_ref_to->is_iodone) {
2269 pr_info(
2270"btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) which is not yet iodone!\n",
2271 btrfsic_get_block_type(state, l->block_ref_to),
2272 l->block_ref_to->logical_bytenr,
2273 l->block_ref_to->dev_state->bdev,
2274 l->block_ref_to->dev_bytenr,
2275 l->block_ref_to->mirror_num);
2276 ret = -1;
2277 } else if (l->block_ref_to->iodone_w_error) {
2278 pr_info(
2279"btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) which has write error!\n",
2280 btrfsic_get_block_type(state, l->block_ref_to),
2281 l->block_ref_to->logical_bytenr,
2282 l->block_ref_to->dev_state->bdev,
2283 l->block_ref_to->dev_bytenr,
2284 l->block_ref_to->mirror_num);
2285 ret = -1;
2286 } else if (l->parent_generation !=
2287 l->block_ref_to->generation &&
2288 BTRFSIC_GENERATION_UNKNOWN !=
2289 l->parent_generation &&
2290 BTRFSIC_GENERATION_UNKNOWN !=
2291 l->block_ref_to->generation) {
2292 pr_info(
2293"btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) with generation %llu != parent generation %llu!\n",
2294 btrfsic_get_block_type(state, l->block_ref_to),
2295 l->block_ref_to->logical_bytenr,
2296 l->block_ref_to->dev_state->bdev,
2297 l->block_ref_to->dev_bytenr,
2298 l->block_ref_to->mirror_num,
2299 l->block_ref_to->generation,
2300 l->parent_generation);
2301 ret = -1;
2302 } else if (l->block_ref_to->flush_gen >
2303 l->block_ref_to->dev_state->last_flush_gen) {
2304 pr_info(
2305"btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) which is not flushed out of disk's write cache (block flush_gen=%llu, dev->flush_gen=%llu)!\n",
2306 btrfsic_get_block_type(state, l->block_ref_to),
2307 l->block_ref_to->logical_bytenr,
2308 l->block_ref_to->dev_state->bdev,
2309 l->block_ref_to->dev_bytenr,
2310 l->block_ref_to->mirror_num, block->flush_gen,
2311 l->block_ref_to->dev_state->last_flush_gen);
2312 ret = -1;
2313 } else if (-1 == btrfsic_check_all_ref_blocks(state,
2314 l->block_ref_to,
2315 recursion_level +
2316 1)) {
2317 ret = -1;
2318 }
2319 }
2320
2321 return ret;
2322}
2323
2324static int btrfsic_is_block_ref_by_superblock(
2325 const struct btrfsic_state *state,
2326 const struct btrfsic_block *block,
2327 int recursion_level)
2328{
2329 const struct btrfsic_block_link *l;
2330
2331 if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2332 /* refer to comment at "abort cyclic linkage (case 1)" */
2333 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2334 pr_info("btrfsic: abort cyclic linkage (case 2).\n");
2335
2336 return 0;
2337 }
2338
2339 /*
2340 * This algorithm is recursive because the amount of used stack space
2341 * is very small and the max recursion depth is limited.
2342 */
2343 list_for_each_entry(l, &block->ref_from_list, node_ref_from) {
2344 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2345 pr_info(
2346 "rl=%d, %c @%llu (%pg/%llu/%d) is ref %u* from %c @%llu (%pg/%llu/%d)\n",
2347 recursion_level,
2348 btrfsic_get_block_type(state, block),
2349 block->logical_bytenr, block->dev_state->bdev,
2350 block->dev_bytenr, block->mirror_num,
2351 l->ref_cnt,
2352 btrfsic_get_block_type(state, l->block_ref_from),
2353 l->block_ref_from->logical_bytenr,
2354 l->block_ref_from->dev_state->bdev,
2355 l->block_ref_from->dev_bytenr,
2356 l->block_ref_from->mirror_num);
2357 if (l->block_ref_from->is_superblock &&
2358 state->latest_superblock->dev_bytenr ==
2359 l->block_ref_from->dev_bytenr &&
2360 state->latest_superblock->dev_state->bdev ==
2361 l->block_ref_from->dev_state->bdev)
2362 return 1;
2363 else if (btrfsic_is_block_ref_by_superblock(state,
2364 l->block_ref_from,
2365 recursion_level +
2366 1))
2367 return 1;
2368 }
2369
2370 return 0;
2371}
2372
2373static void btrfsic_print_add_link(const struct btrfsic_state *state,
2374 const struct btrfsic_block_link *l)
2375{
2376 pr_info("add %u* link from %c @%llu (%pg/%llu/%d) to %c @%llu (%pg/%llu/%d)\n",
2377 l->ref_cnt,
2378 btrfsic_get_block_type(state, l->block_ref_from),
2379 l->block_ref_from->logical_bytenr,
2380 l->block_ref_from->dev_state->bdev,
2381 l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2382 btrfsic_get_block_type(state, l->block_ref_to),
2383 l->block_ref_to->logical_bytenr,
2384 l->block_ref_to->dev_state->bdev, l->block_ref_to->dev_bytenr,
2385 l->block_ref_to->mirror_num);
2386}
2387
2388static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2389 const struct btrfsic_block_link *l)
2390{
2391 pr_info("rem %u* link from %c @%llu (%pg/%llu/%d) to %c @%llu (%pg/%llu/%d)\n",
2392 l->ref_cnt,
2393 btrfsic_get_block_type(state, l->block_ref_from),
2394 l->block_ref_from->logical_bytenr,
2395 l->block_ref_from->dev_state->bdev,
2396 l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2397 btrfsic_get_block_type(state, l->block_ref_to),
2398 l->block_ref_to->logical_bytenr,
2399 l->block_ref_to->dev_state->bdev, l->block_ref_to->dev_bytenr,
2400 l->block_ref_to->mirror_num);
2401}
2402
2403static char btrfsic_get_block_type(const struct btrfsic_state *state,
2404 const struct btrfsic_block *block)
2405{
2406 if (block->is_superblock &&
2407 state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2408 state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2409 return 'S';
2410 else if (block->is_superblock)
2411 return 's';
2412 else if (block->is_metadata)
2413 return 'M';
2414 else
2415 return 'D';
2416}
2417
2418static void btrfsic_dump_tree(const struct btrfsic_state *state)
2419{
2420 btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2421}
2422
2423static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2424 const struct btrfsic_block *block,
2425 int indent_level)
2426{
2427 const struct btrfsic_block_link *l;
2428 int indent_add;
2429 static char buf[80];
2430 int cursor_position;
2431
2432 /*
2433 * Should better fill an on-stack buffer with a complete line and
2434 * dump it at once when it is time to print a newline character.
2435 */
2436
2437 /*
2438 * This algorithm is recursive because the amount of used stack space
2439 * is very small and the max recursion depth is limited.
2440 */
2441 indent_add = sprintf(buf, "%c-%llu(%pg/%llu/%u)",
2442 btrfsic_get_block_type(state, block),
2443 block->logical_bytenr, block->dev_state->bdev,
2444 block->dev_bytenr, block->mirror_num);
2445 if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2446 printk("[...]\n");
2447 return;
2448 }
2449 printk(buf);
2450 indent_level += indent_add;
2451 if (list_empty(&block->ref_to_list)) {
2452 printk("\n");
2453 return;
2454 }
2455 if (block->mirror_num > 1 &&
2456 !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2457 printk(" [...]\n");
2458 return;
2459 }
2460
2461 cursor_position = indent_level;
2462 list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2463 while (cursor_position < indent_level) {
2464 printk(" ");
2465 cursor_position++;
2466 }
2467 if (l->ref_cnt > 1)
2468 indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2469 else
2470 indent_add = sprintf(buf, " --> ");
2471 if (indent_level + indent_add >
2472 BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2473 printk("[...]\n");
2474 cursor_position = 0;
2475 continue;
2476 }
2477
2478 printk(buf);
2479
2480 btrfsic_dump_tree_sub(state, l->block_ref_to,
2481 indent_level + indent_add);
2482 cursor_position = 0;
2483 }
2484}
2485
2486static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2487 struct btrfsic_state *state,
2488 struct btrfsic_block_data_ctx *next_block_ctx,
2489 struct btrfsic_block *next_block,
2490 struct btrfsic_block *from_block,
2491 u64 parent_generation)
2492{
2493 struct btrfsic_block_link *l;
2494
2495 l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2496 next_block_ctx->dev_bytenr,
2497 from_block->dev_state->bdev,
2498 from_block->dev_bytenr,
2499 &state->block_link_hashtable);
2500 if (NULL == l) {
2501 l = btrfsic_block_link_alloc();
2502 if (!l)
2503 return NULL;
2504
2505 l->block_ref_to = next_block;
2506 l->block_ref_from = from_block;
2507 l->ref_cnt = 1;
2508 l->parent_generation = parent_generation;
2509
2510 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2511 btrfsic_print_add_link(state, l);
2512
2513 list_add(&l->node_ref_to, &from_block->ref_to_list);
2514 list_add(&l->node_ref_from, &next_block->ref_from_list);
2515
2516 btrfsic_block_link_hashtable_add(l,
2517 &state->block_link_hashtable);
2518 } else {
2519 l->ref_cnt++;
2520 l->parent_generation = parent_generation;
2521 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2522 btrfsic_print_add_link(state, l);
2523 }
2524
2525 return l;
2526}
2527
2528static struct btrfsic_block *btrfsic_block_lookup_or_add(
2529 struct btrfsic_state *state,
2530 struct btrfsic_block_data_ctx *block_ctx,
2531 const char *additional_string,
2532 int is_metadata,
2533 int is_iodone,
2534 int never_written,
2535 int mirror_num,
2536 int *was_created)
2537{
2538 struct btrfsic_block *block;
2539
2540 block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2541 block_ctx->dev_bytenr,
2542 &state->block_hashtable);
2543 if (NULL == block) {
2544 struct btrfsic_dev_state *dev_state;
2545
2546 block = btrfsic_block_alloc();
2547 if (!block)
2548 return NULL;
2549
2550 dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev->bd_dev);
2551 if (NULL == dev_state) {
2552 pr_info("btrfsic: error, lookup dev_state failed!\n");
2553 btrfsic_block_free(block);
2554 return NULL;
2555 }
2556 block->dev_state = dev_state;
2557 block->dev_bytenr = block_ctx->dev_bytenr;
2558 block->logical_bytenr = block_ctx->start;
2559 block->is_metadata = is_metadata;
2560 block->is_iodone = is_iodone;
2561 block->never_written = never_written;
2562 block->mirror_num = mirror_num;
2563 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2564 pr_info("New %s%c-block @%llu (%pg/%llu/%d)\n",
2565 additional_string,
2566 btrfsic_get_block_type(state, block),
2567 block->logical_bytenr, dev_state->bdev,
2568 block->dev_bytenr, mirror_num);
2569 list_add(&block->all_blocks_node, &state->all_blocks_list);
2570 btrfsic_block_hashtable_add(block, &state->block_hashtable);
2571 if (NULL != was_created)
2572 *was_created = 1;
2573 } else {
2574 if (NULL != was_created)
2575 *was_created = 0;
2576 }
2577
2578 return block;
2579}
2580
2581static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2582 u64 bytenr,
2583 struct btrfsic_dev_state *dev_state,
2584 u64 dev_bytenr)
2585{
2586 struct btrfs_fs_info *fs_info = state->fs_info;
2587 struct btrfsic_block_data_ctx block_ctx;
2588 int num_copies;
2589 int mirror_num;
2590 int match = 0;
2591 int ret;
2592
2593 num_copies = btrfs_num_copies(fs_info, bytenr, state->metablock_size);
2594
2595 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2596 ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2597 &block_ctx, mirror_num);
2598 if (ret) {
2599 pr_info("btrfsic: btrfsic_map_block(logical @%llu, mirror %d) failed!\n",
2600 bytenr, mirror_num);
2601 continue;
2602 }
2603
2604 if (dev_state->bdev == block_ctx.dev->bdev &&
2605 dev_bytenr == block_ctx.dev_bytenr) {
2606 match++;
2607 btrfsic_release_block_ctx(&block_ctx);
2608 break;
2609 }
2610 btrfsic_release_block_ctx(&block_ctx);
2611 }
2612
2613 if (WARN_ON(!match)) {
2614 pr_info(
2615"btrfs: attempt to write M-block which contains logical bytenr that doesn't map to dev+physical bytenr of submit_bio, buffer->log_bytenr=%llu, submit_bio(bdev=%pg, phys_bytenr=%llu)!\n",
2616 bytenr, dev_state->bdev, dev_bytenr);
2617 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2618 ret = btrfsic_map_block(state, bytenr,
2619 state->metablock_size,
2620 &block_ctx, mirror_num);
2621 if (ret)
2622 continue;
2623
2624 pr_info("read logical bytenr @%llu maps to (%pg/%llu/%d)\n",
2625 bytenr, block_ctx.dev->bdev,
2626 block_ctx.dev_bytenr, mirror_num);
2627 }
2628 }
2629}
2630
2631static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev)
2632{
2633 return btrfsic_dev_state_hashtable_lookup(dev,
2634 &btrfsic_dev_state_hashtable);
2635}
2636
2637static void btrfsic_check_write_bio(struct bio *bio, struct btrfsic_dev_state *dev_state)
2638{
2639 unsigned int segs = bio_segments(bio);
2640 u64 dev_bytenr = 512 * bio->bi_iter.bi_sector;
2641 u64 cur_bytenr = dev_bytenr;
2642 struct bvec_iter iter;
2643 struct bio_vec bvec;
2644 char **mapped_datav;
2645 int bio_is_patched = 0;
2646 int i = 0;
2647
2648 if (dev_state->state->print_mask & BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2649 pr_info(
2650"submit_bio(rw=%d,0x%x, bi_vcnt=%u, bi_sector=%llu (bytenr %llu), bi_bdev=%p)\n",
2651 bio_op(bio), bio->bi_opf, segs,
2652 bio->bi_iter.bi_sector, dev_bytenr, bio->bi_bdev);
2653
2654 mapped_datav = kmalloc_array(segs, sizeof(*mapped_datav), GFP_NOFS);
2655 if (!mapped_datav)
2656 return;
2657
2658 bio_for_each_segment(bvec, bio, iter) {
2659 BUG_ON(bvec.bv_len != PAGE_SIZE);
2660 mapped_datav[i] = page_address(bvec.bv_page);
2661 i++;
2662
2663 if (dev_state->state->print_mask &
2664 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
2665 pr_info("#%u: bytenr=%llu, len=%u, offset=%u\n",
2666 i, cur_bytenr, bvec.bv_len, bvec.bv_offset);
2667 cur_bytenr += bvec.bv_len;
2668 }
2669
2670 btrfsic_process_written_block(dev_state, dev_bytenr, mapped_datav, segs,
2671 bio, &bio_is_patched, bio->bi_opf);
2672 kfree(mapped_datav);
2673}
2674
2675static void btrfsic_check_flush_bio(struct bio *bio, struct btrfsic_dev_state *dev_state)
2676{
2677 if (dev_state->state->print_mask & BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2678 pr_info("submit_bio(rw=%d,0x%x FLUSH, bdev=%p)\n",
2679 bio_op(bio), bio->bi_opf, bio->bi_bdev);
2680
2681 if (dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2682 struct btrfsic_block *const block =
2683 &dev_state->dummy_block_for_bio_bh_flush;
2684
2685 block->is_iodone = 0;
2686 block->never_written = 0;
2687 block->iodone_w_error = 0;
2688 block->flush_gen = dev_state->last_flush_gen + 1;
2689 block->submit_bio_bh_rw = bio->bi_opf;
2690 block->orig_bio_private = bio->bi_private;
2691 block->orig_bio_end_io = bio->bi_end_io;
2692 block->next_in_same_bio = NULL;
2693 bio->bi_private = block;
2694 bio->bi_end_io = btrfsic_bio_end_io;
2695 } else if ((dev_state->state->print_mask &
2696 (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2697 BTRFSIC_PRINT_MASK_VERBOSE))) {
2698 pr_info(
2699"btrfsic_submit_bio(%pg) with FLUSH but dummy block already in use (ignored)!\n",
2700 dev_state->bdev);
2701 }
2702}
2703
2704void btrfsic_check_bio(struct bio *bio)
2705{
2706 struct btrfsic_dev_state *dev_state;
2707
2708 if (!btrfsic_is_initialized)
2709 return;
2710
2711 /*
2712 * We can be called before btrfsic_mount, so there might not be a
2713 * dev_state.
2714 */
2715 dev_state = btrfsic_dev_state_lookup(bio->bi_bdev->bd_dev);
2716 mutex_lock(&btrfsic_mutex);
2717 if (dev_state) {
2718 if (bio_op(bio) == REQ_OP_WRITE && bio_has_data(bio))
2719 btrfsic_check_write_bio(bio, dev_state);
2720 else if (bio->bi_opf & REQ_PREFLUSH)
2721 btrfsic_check_flush_bio(bio, dev_state);
2722 }
2723 mutex_unlock(&btrfsic_mutex);
2724}
2725
2726int btrfsic_mount(struct btrfs_fs_info *fs_info,
2727 struct btrfs_fs_devices *fs_devices,
2728 int including_extent_data, u32 print_mask)
2729{
2730 int ret;
2731 struct btrfsic_state *state;
2732 struct list_head *dev_head = &fs_devices->devices;
2733 struct btrfs_device *device;
2734
2735 if (!PAGE_ALIGNED(fs_info->nodesize)) {
2736 pr_info("btrfsic: cannot handle nodesize %d not being a multiple of PAGE_SIZE %ld!\n",
2737 fs_info->nodesize, PAGE_SIZE);
2738 return -1;
2739 }
2740 if (!PAGE_ALIGNED(fs_info->sectorsize)) {
2741 pr_info("btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_SIZE %ld!\n",
2742 fs_info->sectorsize, PAGE_SIZE);
2743 return -1;
2744 }
2745 state = kvzalloc(sizeof(*state), GFP_KERNEL);
2746 if (!state)
2747 return -ENOMEM;
2748
2749 if (!btrfsic_is_initialized) {
2750 mutex_init(&btrfsic_mutex);
2751 btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
2752 btrfsic_is_initialized = 1;
2753 }
2754 mutex_lock(&btrfsic_mutex);
2755 state->fs_info = fs_info;
2756 state->print_mask = print_mask;
2757 state->include_extent_data = including_extent_data;
2758 state->metablock_size = fs_info->nodesize;
2759 state->datablock_size = fs_info->sectorsize;
2760 INIT_LIST_HEAD(&state->all_blocks_list);
2761 btrfsic_block_hashtable_init(&state->block_hashtable);
2762 btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
2763 state->max_superblock_generation = 0;
2764 state->latest_superblock = NULL;
2765
2766 list_for_each_entry(device, dev_head, dev_list) {
2767 struct btrfsic_dev_state *ds;
2768
2769 if (!device->bdev || !device->name)
2770 continue;
2771
2772 ds = btrfsic_dev_state_alloc();
2773 if (NULL == ds) {
2774 mutex_unlock(&btrfsic_mutex);
2775 return -ENOMEM;
2776 }
2777 ds->bdev = device->bdev;
2778 ds->state = state;
2779 btrfsic_dev_state_hashtable_add(ds,
2780 &btrfsic_dev_state_hashtable);
2781 }
2782
2783 ret = btrfsic_process_superblock(state, fs_devices);
2784 if (0 != ret) {
2785 mutex_unlock(&btrfsic_mutex);
2786 btrfsic_unmount(fs_devices);
2787 return ret;
2788 }
2789
2790 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
2791 btrfsic_dump_database(state);
2792 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
2793 btrfsic_dump_tree(state);
2794
2795 mutex_unlock(&btrfsic_mutex);
2796 return 0;
2797}
2798
2799void btrfsic_unmount(struct btrfs_fs_devices *fs_devices)
2800{
2801 struct btrfsic_block *b_all, *tmp_all;
2802 struct btrfsic_state *state;
2803 struct list_head *dev_head = &fs_devices->devices;
2804 struct btrfs_device *device;
2805
2806 if (!btrfsic_is_initialized)
2807 return;
2808
2809 mutex_lock(&btrfsic_mutex);
2810
2811 state = NULL;
2812 list_for_each_entry(device, dev_head, dev_list) {
2813 struct btrfsic_dev_state *ds;
2814
2815 if (!device->bdev || !device->name)
2816 continue;
2817
2818 ds = btrfsic_dev_state_hashtable_lookup(
2819 device->bdev->bd_dev,
2820 &btrfsic_dev_state_hashtable);
2821 if (NULL != ds) {
2822 state = ds->state;
2823 btrfsic_dev_state_hashtable_remove(ds);
2824 btrfsic_dev_state_free(ds);
2825 }
2826 }
2827
2828 if (NULL == state) {
2829 pr_info("btrfsic: error, cannot find state information on umount!\n");
2830 mutex_unlock(&btrfsic_mutex);
2831 return;
2832 }
2833
2834 /*
2835 * Don't care about keeping the lists' state up to date,
2836 * just free all memory that was allocated dynamically.
2837 * Free the blocks and the block_links.
2838 */
2839 list_for_each_entry_safe(b_all, tmp_all, &state->all_blocks_list,
2840 all_blocks_node) {
2841 struct btrfsic_block_link *l, *tmp;
2842
2843 list_for_each_entry_safe(l, tmp, &b_all->ref_to_list,
2844 node_ref_to) {
2845 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2846 btrfsic_print_rem_link(state, l);
2847
2848 l->ref_cnt--;
2849 if (0 == l->ref_cnt)
2850 btrfsic_block_link_free(l);
2851 }
2852
2853 if (b_all->is_iodone || b_all->never_written)
2854 btrfsic_block_free(b_all);
2855 else
2856 pr_info(
2857"btrfs: attempt to free %c-block @%llu (%pg/%llu/%d) on umount which is not yet iodone!\n",
2858 btrfsic_get_block_type(state, b_all),
2859 b_all->logical_bytenr, b_all->dev_state->bdev,
2860 b_all->dev_bytenr, b_all->mirror_num);
2861 }
2862
2863 mutex_unlock(&btrfsic_mutex);
2864
2865 kvfree(state);
2866}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) STRATO AG 2011. All rights reserved.
4 */
5
6/*
7 * This module can be used to catch cases when the btrfs kernel
8 * code executes write requests to the disk that bring the file
9 * system in an inconsistent state. In such a state, a power-loss
10 * or kernel panic event would cause that the data on disk is
11 * lost or at least damaged.
12 *
13 * Code is added that examines all block write requests during
14 * runtime (including writes of the super block). Three rules
15 * are verified and an error is printed on violation of the
16 * rules:
17 * 1. It is not allowed to write a disk block which is
18 * currently referenced by the super block (either directly
19 * or indirectly).
20 * 2. When a super block is written, it is verified that all
21 * referenced (directly or indirectly) blocks fulfill the
22 * following requirements:
23 * 2a. All referenced blocks have either been present when
24 * the file system was mounted, (i.e., they have been
25 * referenced by the super block) or they have been
26 * written since then and the write completion callback
27 * was called and no write error was indicated and a
28 * FLUSH request to the device where these blocks are
29 * located was received and completed.
30 * 2b. All referenced blocks need to have a generation
31 * number which is equal to the parent's number.
32 *
33 * One issue that was found using this module was that the log
34 * tree on disk became temporarily corrupted because disk blocks
35 * that had been in use for the log tree had been freed and
36 * reused too early, while being referenced by the written super
37 * block.
38 *
39 * The search term in the kernel log that can be used to filter
40 * on the existence of detected integrity issues is
41 * "btrfs: attempt".
42 *
43 * The integrity check is enabled via mount options. These
44 * mount options are only supported if the integrity check
45 * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY.
46 *
47 * Example #1, apply integrity checks to all metadata:
48 * mount /dev/sdb1 /mnt -o check_int
49 *
50 * Example #2, apply integrity checks to all metadata and
51 * to data extents:
52 * mount /dev/sdb1 /mnt -o check_int_data
53 *
54 * Example #3, apply integrity checks to all metadata and dump
55 * the tree that the super block references to kernel messages
56 * each time after a super block was written:
57 * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263
58 *
59 * If the integrity check tool is included and activated in
60 * the mount options, plenty of kernel memory is used, and
61 * plenty of additional CPU cycles are spent. Enabling this
62 * functionality is not intended for normal use. In most
63 * cases, unless you are a btrfs developer who needs to verify
64 * the integrity of (super)-block write requests, do not
65 * enable the config option BTRFS_FS_CHECK_INTEGRITY to
66 * include and compile the integrity check tool.
67 *
68 * Expect millions of lines of information in the kernel log with an
69 * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the
70 * kernel config to at least 26 (which is 64MB). Usually the value is
71 * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be
72 * changed like this before LOG_BUF_SHIFT can be set to a high value:
73 * config LOG_BUF_SHIFT
74 * int "Kernel log buffer size (16 => 64KB, 17 => 128KB)"
75 * range 12 30
76 */
77
78#include <linux/sched.h>
79#include <linux/slab.h>
80#include <linux/buffer_head.h>
81#include <linux/mutex.h>
82#include <linux/genhd.h>
83#include <linux/blkdev.h>
84#include <linux/mm.h>
85#include <linux/string.h>
86#include <crypto/hash.h>
87#include "ctree.h"
88#include "disk-io.h"
89#include "transaction.h"
90#include "extent_io.h"
91#include "volumes.h"
92#include "print-tree.h"
93#include "locking.h"
94#include "check-integrity.h"
95#include "rcu-string.h"
96#include "compression.h"
97
98#define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
99#define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
100#define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
101#define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
102#define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
103#define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
104#define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
105#define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6) /* in characters,
106 * excluding " [...]" */
107#define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
108
109/*
110 * The definition of the bitmask fields for the print_mask.
111 * They are specified with the mount option check_integrity_print_mask.
112 */
113#define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE 0x00000001
114#define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION 0x00000002
115#define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE 0x00000004
116#define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE 0x00000008
117#define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH 0x00000010
118#define BTRFSIC_PRINT_MASK_END_IO_BIO_BH 0x00000020
119#define BTRFSIC_PRINT_MASK_VERBOSE 0x00000040
120#define BTRFSIC_PRINT_MASK_VERY_VERBOSE 0x00000080
121#define BTRFSIC_PRINT_MASK_INITIAL_TREE 0x00000100
122#define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES 0x00000200
123#define BTRFSIC_PRINT_MASK_INITIAL_DATABASE 0x00000400
124#define BTRFSIC_PRINT_MASK_NUM_COPIES 0x00000800
125#define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS 0x00001000
126#define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE 0x00002000
127
128struct btrfsic_dev_state;
129struct btrfsic_state;
130
131struct btrfsic_block {
132 u32 magic_num; /* only used for debug purposes */
133 unsigned int is_metadata:1; /* if it is meta-data, not data-data */
134 unsigned int is_superblock:1; /* if it is one of the superblocks */
135 unsigned int is_iodone:1; /* if is done by lower subsystem */
136 unsigned int iodone_w_error:1; /* error was indicated to endio */
137 unsigned int never_written:1; /* block was added because it was
138 * referenced, not because it was
139 * written */
140 unsigned int mirror_num; /* large enough to hold
141 * BTRFS_SUPER_MIRROR_MAX */
142 struct btrfsic_dev_state *dev_state;
143 u64 dev_bytenr; /* key, physical byte num on disk */
144 u64 logical_bytenr; /* logical byte num on disk */
145 u64 generation;
146 struct btrfs_disk_key disk_key; /* extra info to print in case of
147 * issues, will not always be correct */
148 struct list_head collision_resolving_node; /* list node */
149 struct list_head all_blocks_node; /* list node */
150
151 /* the following two lists contain block_link items */
152 struct list_head ref_to_list; /* list */
153 struct list_head ref_from_list; /* list */
154 struct btrfsic_block *next_in_same_bio;
155 void *orig_bio_bh_private;
156 union {
157 bio_end_io_t *bio;
158 bh_end_io_t *bh;
159 } orig_bio_bh_end_io;
160 int submit_bio_bh_rw;
161 u64 flush_gen; /* only valid if !never_written */
162};
163
164/*
165 * Elements of this type are allocated dynamically and required because
166 * each block object can refer to and can be ref from multiple blocks.
167 * The key to lookup them in the hashtable is the dev_bytenr of
168 * the block ref to plus the one from the block referred from.
169 * The fact that they are searchable via a hashtable and that a
170 * ref_cnt is maintained is not required for the btrfs integrity
171 * check algorithm itself, it is only used to make the output more
172 * beautiful in case that an error is detected (an error is defined
173 * as a write operation to a block while that block is still referenced).
174 */
175struct btrfsic_block_link {
176 u32 magic_num; /* only used for debug purposes */
177 u32 ref_cnt;
178 struct list_head node_ref_to; /* list node */
179 struct list_head node_ref_from; /* list node */
180 struct list_head collision_resolving_node; /* list node */
181 struct btrfsic_block *block_ref_to;
182 struct btrfsic_block *block_ref_from;
183 u64 parent_generation;
184};
185
186struct btrfsic_dev_state {
187 u32 magic_num; /* only used for debug purposes */
188 struct block_device *bdev;
189 struct btrfsic_state *state;
190 struct list_head collision_resolving_node; /* list node */
191 struct btrfsic_block dummy_block_for_bio_bh_flush;
192 u64 last_flush_gen;
193 char name[BDEVNAME_SIZE];
194};
195
196struct btrfsic_block_hashtable {
197 struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
198};
199
200struct btrfsic_block_link_hashtable {
201 struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
202};
203
204struct btrfsic_dev_state_hashtable {
205 struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
206};
207
208struct btrfsic_block_data_ctx {
209 u64 start; /* virtual bytenr */
210 u64 dev_bytenr; /* physical bytenr on device */
211 u32 len;
212 struct btrfsic_dev_state *dev;
213 char **datav;
214 struct page **pagev;
215 void *mem_to_free;
216};
217
218/* This structure is used to implement recursion without occupying
219 * any stack space, refer to btrfsic_process_metablock() */
220struct btrfsic_stack_frame {
221 u32 magic;
222 u32 nr;
223 int error;
224 int i;
225 int limit_nesting;
226 int num_copies;
227 int mirror_num;
228 struct btrfsic_block *block;
229 struct btrfsic_block_data_ctx *block_ctx;
230 struct btrfsic_block *next_block;
231 struct btrfsic_block_data_ctx next_block_ctx;
232 struct btrfs_header *hdr;
233 struct btrfsic_stack_frame *prev;
234};
235
236/* Some state per mounted filesystem */
237struct btrfsic_state {
238 u32 print_mask;
239 int include_extent_data;
240 int csum_size;
241 struct list_head all_blocks_list;
242 struct btrfsic_block_hashtable block_hashtable;
243 struct btrfsic_block_link_hashtable block_link_hashtable;
244 struct btrfs_fs_info *fs_info;
245 u64 max_superblock_generation;
246 struct btrfsic_block *latest_superblock;
247 u32 metablock_size;
248 u32 datablock_size;
249};
250
251static void btrfsic_block_init(struct btrfsic_block *b);
252static struct btrfsic_block *btrfsic_block_alloc(void);
253static void btrfsic_block_free(struct btrfsic_block *b);
254static void btrfsic_block_link_init(struct btrfsic_block_link *n);
255static struct btrfsic_block_link *btrfsic_block_link_alloc(void);
256static void btrfsic_block_link_free(struct btrfsic_block_link *n);
257static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds);
258static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void);
259static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds);
260static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h);
261static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
262 struct btrfsic_block_hashtable *h);
263static void btrfsic_block_hashtable_remove(struct btrfsic_block *b);
264static struct btrfsic_block *btrfsic_block_hashtable_lookup(
265 struct block_device *bdev,
266 u64 dev_bytenr,
267 struct btrfsic_block_hashtable *h);
268static void btrfsic_block_link_hashtable_init(
269 struct btrfsic_block_link_hashtable *h);
270static void btrfsic_block_link_hashtable_add(
271 struct btrfsic_block_link *l,
272 struct btrfsic_block_link_hashtable *h);
273static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l);
274static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
275 struct block_device *bdev_ref_to,
276 u64 dev_bytenr_ref_to,
277 struct block_device *bdev_ref_from,
278 u64 dev_bytenr_ref_from,
279 struct btrfsic_block_link_hashtable *h);
280static void btrfsic_dev_state_hashtable_init(
281 struct btrfsic_dev_state_hashtable *h);
282static void btrfsic_dev_state_hashtable_add(
283 struct btrfsic_dev_state *ds,
284 struct btrfsic_dev_state_hashtable *h);
285static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds);
286static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev,
287 struct btrfsic_dev_state_hashtable *h);
288static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void);
289static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf);
290static int btrfsic_process_superblock(struct btrfsic_state *state,
291 struct btrfs_fs_devices *fs_devices);
292static int btrfsic_process_metablock(struct btrfsic_state *state,
293 struct btrfsic_block *block,
294 struct btrfsic_block_data_ctx *block_ctx,
295 int limit_nesting, int force_iodone_flag);
296static void btrfsic_read_from_block_data(
297 struct btrfsic_block_data_ctx *block_ctx,
298 void *dst, u32 offset, size_t len);
299static int btrfsic_create_link_to_next_block(
300 struct btrfsic_state *state,
301 struct btrfsic_block *block,
302 struct btrfsic_block_data_ctx
303 *block_ctx, u64 next_bytenr,
304 int limit_nesting,
305 struct btrfsic_block_data_ctx *next_block_ctx,
306 struct btrfsic_block **next_blockp,
307 int force_iodone_flag,
308 int *num_copiesp, int *mirror_nump,
309 struct btrfs_disk_key *disk_key,
310 u64 parent_generation);
311static int btrfsic_handle_extent_data(struct btrfsic_state *state,
312 struct btrfsic_block *block,
313 struct btrfsic_block_data_ctx *block_ctx,
314 u32 item_offset, int force_iodone_flag);
315static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
316 struct btrfsic_block_data_ctx *block_ctx_out,
317 int mirror_num);
318static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
319static int btrfsic_read_block(struct btrfsic_state *state,
320 struct btrfsic_block_data_ctx *block_ctx);
321static void btrfsic_dump_database(struct btrfsic_state *state);
322static int btrfsic_test_for_metadata(struct btrfsic_state *state,
323 char **datav, unsigned int num_pages);
324static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
325 u64 dev_bytenr, char **mapped_datav,
326 unsigned int num_pages,
327 struct bio *bio, int *bio_is_patched,
328 struct buffer_head *bh,
329 int submit_bio_bh_rw);
330static int btrfsic_process_written_superblock(
331 struct btrfsic_state *state,
332 struct btrfsic_block *const block,
333 struct btrfs_super_block *const super_hdr);
334static void btrfsic_bio_end_io(struct bio *bp);
335static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate);
336static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
337 const struct btrfsic_block *block,
338 int recursion_level);
339static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
340 struct btrfsic_block *const block,
341 int recursion_level);
342static void btrfsic_print_add_link(const struct btrfsic_state *state,
343 const struct btrfsic_block_link *l);
344static void btrfsic_print_rem_link(const struct btrfsic_state *state,
345 const struct btrfsic_block_link *l);
346static char btrfsic_get_block_type(const struct btrfsic_state *state,
347 const struct btrfsic_block *block);
348static void btrfsic_dump_tree(const struct btrfsic_state *state);
349static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
350 const struct btrfsic_block *block,
351 int indent_level);
352static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
353 struct btrfsic_state *state,
354 struct btrfsic_block_data_ctx *next_block_ctx,
355 struct btrfsic_block *next_block,
356 struct btrfsic_block *from_block,
357 u64 parent_generation);
358static struct btrfsic_block *btrfsic_block_lookup_or_add(
359 struct btrfsic_state *state,
360 struct btrfsic_block_data_ctx *block_ctx,
361 const char *additional_string,
362 int is_metadata,
363 int is_iodone,
364 int never_written,
365 int mirror_num,
366 int *was_created);
367static int btrfsic_process_superblock_dev_mirror(
368 struct btrfsic_state *state,
369 struct btrfsic_dev_state *dev_state,
370 struct btrfs_device *device,
371 int superblock_mirror_num,
372 struct btrfsic_dev_state **selected_dev_state,
373 struct btrfs_super_block *selected_super);
374static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev);
375static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
376 u64 bytenr,
377 struct btrfsic_dev_state *dev_state,
378 u64 dev_bytenr);
379
380static struct mutex btrfsic_mutex;
381static int btrfsic_is_initialized;
382static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
383
384
385static void btrfsic_block_init(struct btrfsic_block *b)
386{
387 b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
388 b->dev_state = NULL;
389 b->dev_bytenr = 0;
390 b->logical_bytenr = 0;
391 b->generation = BTRFSIC_GENERATION_UNKNOWN;
392 b->disk_key.objectid = 0;
393 b->disk_key.type = 0;
394 b->disk_key.offset = 0;
395 b->is_metadata = 0;
396 b->is_superblock = 0;
397 b->is_iodone = 0;
398 b->iodone_w_error = 0;
399 b->never_written = 0;
400 b->mirror_num = 0;
401 b->next_in_same_bio = NULL;
402 b->orig_bio_bh_private = NULL;
403 b->orig_bio_bh_end_io.bio = NULL;
404 INIT_LIST_HEAD(&b->collision_resolving_node);
405 INIT_LIST_HEAD(&b->all_blocks_node);
406 INIT_LIST_HEAD(&b->ref_to_list);
407 INIT_LIST_HEAD(&b->ref_from_list);
408 b->submit_bio_bh_rw = 0;
409 b->flush_gen = 0;
410}
411
412static struct btrfsic_block *btrfsic_block_alloc(void)
413{
414 struct btrfsic_block *b;
415
416 b = kzalloc(sizeof(*b), GFP_NOFS);
417 if (NULL != b)
418 btrfsic_block_init(b);
419
420 return b;
421}
422
423static void btrfsic_block_free(struct btrfsic_block *b)
424{
425 BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
426 kfree(b);
427}
428
429static void btrfsic_block_link_init(struct btrfsic_block_link *l)
430{
431 l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
432 l->ref_cnt = 1;
433 INIT_LIST_HEAD(&l->node_ref_to);
434 INIT_LIST_HEAD(&l->node_ref_from);
435 INIT_LIST_HEAD(&l->collision_resolving_node);
436 l->block_ref_to = NULL;
437 l->block_ref_from = NULL;
438}
439
440static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
441{
442 struct btrfsic_block_link *l;
443
444 l = kzalloc(sizeof(*l), GFP_NOFS);
445 if (NULL != l)
446 btrfsic_block_link_init(l);
447
448 return l;
449}
450
451static void btrfsic_block_link_free(struct btrfsic_block_link *l)
452{
453 BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
454 kfree(l);
455}
456
457static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
458{
459 ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
460 ds->bdev = NULL;
461 ds->state = NULL;
462 ds->name[0] = '\0';
463 INIT_LIST_HEAD(&ds->collision_resolving_node);
464 ds->last_flush_gen = 0;
465 btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
466 ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
467 ds->dummy_block_for_bio_bh_flush.dev_state = ds;
468}
469
470static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
471{
472 struct btrfsic_dev_state *ds;
473
474 ds = kzalloc(sizeof(*ds), GFP_NOFS);
475 if (NULL != ds)
476 btrfsic_dev_state_init(ds);
477
478 return ds;
479}
480
481static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
482{
483 BUG_ON(!(NULL == ds ||
484 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
485 kfree(ds);
486}
487
488static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
489{
490 int i;
491
492 for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
493 INIT_LIST_HEAD(h->table + i);
494}
495
496static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
497 struct btrfsic_block_hashtable *h)
498{
499 const unsigned int hashval =
500 (((unsigned int)(b->dev_bytenr >> 16)) ^
501 ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
502 (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
503
504 list_add(&b->collision_resolving_node, h->table + hashval);
505}
506
507static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
508{
509 list_del(&b->collision_resolving_node);
510}
511
512static struct btrfsic_block *btrfsic_block_hashtable_lookup(
513 struct block_device *bdev,
514 u64 dev_bytenr,
515 struct btrfsic_block_hashtable *h)
516{
517 const unsigned int hashval =
518 (((unsigned int)(dev_bytenr >> 16)) ^
519 ((unsigned int)((uintptr_t)bdev))) &
520 (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
521 struct btrfsic_block *b;
522
523 list_for_each_entry(b, h->table + hashval, collision_resolving_node) {
524 if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
525 return b;
526 }
527
528 return NULL;
529}
530
531static void btrfsic_block_link_hashtable_init(
532 struct btrfsic_block_link_hashtable *h)
533{
534 int i;
535
536 for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
537 INIT_LIST_HEAD(h->table + i);
538}
539
540static void btrfsic_block_link_hashtable_add(
541 struct btrfsic_block_link *l,
542 struct btrfsic_block_link_hashtable *h)
543{
544 const unsigned int hashval =
545 (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
546 ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
547 ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
548 ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
549 & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
550
551 BUG_ON(NULL == l->block_ref_to);
552 BUG_ON(NULL == l->block_ref_from);
553 list_add(&l->collision_resolving_node, h->table + hashval);
554}
555
556static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
557{
558 list_del(&l->collision_resolving_node);
559}
560
561static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
562 struct block_device *bdev_ref_to,
563 u64 dev_bytenr_ref_to,
564 struct block_device *bdev_ref_from,
565 u64 dev_bytenr_ref_from,
566 struct btrfsic_block_link_hashtable *h)
567{
568 const unsigned int hashval =
569 (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
570 ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
571 ((unsigned int)((uintptr_t)bdev_ref_to)) ^
572 ((unsigned int)((uintptr_t)bdev_ref_from))) &
573 (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
574 struct btrfsic_block_link *l;
575
576 list_for_each_entry(l, h->table + hashval, collision_resolving_node) {
577 BUG_ON(NULL == l->block_ref_to);
578 BUG_ON(NULL == l->block_ref_from);
579 if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
580 l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
581 l->block_ref_from->dev_state->bdev == bdev_ref_from &&
582 l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
583 return l;
584 }
585
586 return NULL;
587}
588
589static void btrfsic_dev_state_hashtable_init(
590 struct btrfsic_dev_state_hashtable *h)
591{
592 int i;
593
594 for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
595 INIT_LIST_HEAD(h->table + i);
596}
597
598static void btrfsic_dev_state_hashtable_add(
599 struct btrfsic_dev_state *ds,
600 struct btrfsic_dev_state_hashtable *h)
601{
602 const unsigned int hashval =
603 (((unsigned int)((uintptr_t)ds->bdev->bd_dev)) &
604 (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
605
606 list_add(&ds->collision_resolving_node, h->table + hashval);
607}
608
609static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
610{
611 list_del(&ds->collision_resolving_node);
612}
613
614static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev,
615 struct btrfsic_dev_state_hashtable *h)
616{
617 const unsigned int hashval =
618 dev & (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1);
619 struct btrfsic_dev_state *ds;
620
621 list_for_each_entry(ds, h->table + hashval, collision_resolving_node) {
622 if (ds->bdev->bd_dev == dev)
623 return ds;
624 }
625
626 return NULL;
627}
628
629static int btrfsic_process_superblock(struct btrfsic_state *state,
630 struct btrfs_fs_devices *fs_devices)
631{
632 struct btrfs_fs_info *fs_info = state->fs_info;
633 struct btrfs_super_block *selected_super;
634 struct list_head *dev_head = &fs_devices->devices;
635 struct btrfs_device *device;
636 struct btrfsic_dev_state *selected_dev_state = NULL;
637 int ret = 0;
638 int pass;
639
640 BUG_ON(NULL == state);
641 selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
642 if (NULL == selected_super) {
643 pr_info("btrfsic: error, kmalloc failed!\n");
644 return -ENOMEM;
645 }
646
647 list_for_each_entry(device, dev_head, dev_list) {
648 int i;
649 struct btrfsic_dev_state *dev_state;
650
651 if (!device->bdev || !device->name)
652 continue;
653
654 dev_state = btrfsic_dev_state_lookup(device->bdev->bd_dev);
655 BUG_ON(NULL == dev_state);
656 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
657 ret = btrfsic_process_superblock_dev_mirror(
658 state, dev_state, device, i,
659 &selected_dev_state, selected_super);
660 if (0 != ret && 0 == i) {
661 kfree(selected_super);
662 return ret;
663 }
664 }
665 }
666
667 if (NULL == state->latest_superblock) {
668 pr_info("btrfsic: no superblock found!\n");
669 kfree(selected_super);
670 return -1;
671 }
672
673 state->csum_size = btrfs_super_csum_size(selected_super);
674
675 for (pass = 0; pass < 3; pass++) {
676 int num_copies;
677 int mirror_num;
678 u64 next_bytenr;
679
680 switch (pass) {
681 case 0:
682 next_bytenr = btrfs_super_root(selected_super);
683 if (state->print_mask &
684 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
685 pr_info("root@%llu\n", next_bytenr);
686 break;
687 case 1:
688 next_bytenr = btrfs_super_chunk_root(selected_super);
689 if (state->print_mask &
690 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
691 pr_info("chunk@%llu\n", next_bytenr);
692 break;
693 case 2:
694 next_bytenr = btrfs_super_log_root(selected_super);
695 if (0 == next_bytenr)
696 continue;
697 if (state->print_mask &
698 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
699 pr_info("log@%llu\n", next_bytenr);
700 break;
701 }
702
703 num_copies = btrfs_num_copies(fs_info, next_bytenr,
704 state->metablock_size);
705 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
706 pr_info("num_copies(log_bytenr=%llu) = %d\n",
707 next_bytenr, num_copies);
708
709 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
710 struct btrfsic_block *next_block;
711 struct btrfsic_block_data_ctx tmp_next_block_ctx;
712 struct btrfsic_block_link *l;
713
714 ret = btrfsic_map_block(state, next_bytenr,
715 state->metablock_size,
716 &tmp_next_block_ctx,
717 mirror_num);
718 if (ret) {
719 pr_info("btrfsic: btrfsic_map_block(root @%llu, mirror %d) failed!\n",
720 next_bytenr, mirror_num);
721 kfree(selected_super);
722 return -1;
723 }
724
725 next_block = btrfsic_block_hashtable_lookup(
726 tmp_next_block_ctx.dev->bdev,
727 tmp_next_block_ctx.dev_bytenr,
728 &state->block_hashtable);
729 BUG_ON(NULL == next_block);
730
731 l = btrfsic_block_link_hashtable_lookup(
732 tmp_next_block_ctx.dev->bdev,
733 tmp_next_block_ctx.dev_bytenr,
734 state->latest_superblock->dev_state->
735 bdev,
736 state->latest_superblock->dev_bytenr,
737 &state->block_link_hashtable);
738 BUG_ON(NULL == l);
739
740 ret = btrfsic_read_block(state, &tmp_next_block_ctx);
741 if (ret < (int)PAGE_SIZE) {
742 pr_info("btrfsic: read @logical %llu failed!\n",
743 tmp_next_block_ctx.start);
744 btrfsic_release_block_ctx(&tmp_next_block_ctx);
745 kfree(selected_super);
746 return -1;
747 }
748
749 ret = btrfsic_process_metablock(state,
750 next_block,
751 &tmp_next_block_ctx,
752 BTRFS_MAX_LEVEL + 3, 1);
753 btrfsic_release_block_ctx(&tmp_next_block_ctx);
754 }
755 }
756
757 kfree(selected_super);
758 return ret;
759}
760
761static int btrfsic_process_superblock_dev_mirror(
762 struct btrfsic_state *state,
763 struct btrfsic_dev_state *dev_state,
764 struct btrfs_device *device,
765 int superblock_mirror_num,
766 struct btrfsic_dev_state **selected_dev_state,
767 struct btrfs_super_block *selected_super)
768{
769 struct btrfs_fs_info *fs_info = state->fs_info;
770 struct btrfs_super_block *super_tmp;
771 u64 dev_bytenr;
772 struct buffer_head *bh;
773 struct btrfsic_block *superblock_tmp;
774 int pass;
775 struct block_device *const superblock_bdev = device->bdev;
776
777 /* super block bytenr is always the unmapped device bytenr */
778 dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
779 if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->commit_total_bytes)
780 return -1;
781 bh = __bread(superblock_bdev, dev_bytenr / BTRFS_BDEV_BLOCKSIZE,
782 BTRFS_SUPER_INFO_SIZE);
783 if (NULL == bh)
784 return -1;
785 super_tmp = (struct btrfs_super_block *)
786 (bh->b_data + (dev_bytenr & (BTRFS_BDEV_BLOCKSIZE - 1)));
787
788 if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
789 btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
790 memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
791 btrfs_super_nodesize(super_tmp) != state->metablock_size ||
792 btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
793 brelse(bh);
794 return 0;
795 }
796
797 superblock_tmp =
798 btrfsic_block_hashtable_lookup(superblock_bdev,
799 dev_bytenr,
800 &state->block_hashtable);
801 if (NULL == superblock_tmp) {
802 superblock_tmp = btrfsic_block_alloc();
803 if (NULL == superblock_tmp) {
804 pr_info("btrfsic: error, kmalloc failed!\n");
805 brelse(bh);
806 return -1;
807 }
808 /* for superblock, only the dev_bytenr makes sense */
809 superblock_tmp->dev_bytenr = dev_bytenr;
810 superblock_tmp->dev_state = dev_state;
811 superblock_tmp->logical_bytenr = dev_bytenr;
812 superblock_tmp->generation = btrfs_super_generation(super_tmp);
813 superblock_tmp->is_metadata = 1;
814 superblock_tmp->is_superblock = 1;
815 superblock_tmp->is_iodone = 1;
816 superblock_tmp->never_written = 0;
817 superblock_tmp->mirror_num = 1 + superblock_mirror_num;
818 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
819 btrfs_info_in_rcu(fs_info,
820 "new initial S-block (bdev %p, %s) @%llu (%s/%llu/%d)",
821 superblock_bdev,
822 rcu_str_deref(device->name), dev_bytenr,
823 dev_state->name, dev_bytenr,
824 superblock_mirror_num);
825 list_add(&superblock_tmp->all_blocks_node,
826 &state->all_blocks_list);
827 btrfsic_block_hashtable_add(superblock_tmp,
828 &state->block_hashtable);
829 }
830
831 /* select the one with the highest generation field */
832 if (btrfs_super_generation(super_tmp) >
833 state->max_superblock_generation ||
834 0 == state->max_superblock_generation) {
835 memcpy(selected_super, super_tmp, sizeof(*selected_super));
836 *selected_dev_state = dev_state;
837 state->max_superblock_generation =
838 btrfs_super_generation(super_tmp);
839 state->latest_superblock = superblock_tmp;
840 }
841
842 for (pass = 0; pass < 3; pass++) {
843 u64 next_bytenr;
844 int num_copies;
845 int mirror_num;
846 const char *additional_string = NULL;
847 struct btrfs_disk_key tmp_disk_key;
848
849 tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
850 tmp_disk_key.offset = 0;
851 switch (pass) {
852 case 0:
853 btrfs_set_disk_key_objectid(&tmp_disk_key,
854 BTRFS_ROOT_TREE_OBJECTID);
855 additional_string = "initial root ";
856 next_bytenr = btrfs_super_root(super_tmp);
857 break;
858 case 1:
859 btrfs_set_disk_key_objectid(&tmp_disk_key,
860 BTRFS_CHUNK_TREE_OBJECTID);
861 additional_string = "initial chunk ";
862 next_bytenr = btrfs_super_chunk_root(super_tmp);
863 break;
864 case 2:
865 btrfs_set_disk_key_objectid(&tmp_disk_key,
866 BTRFS_TREE_LOG_OBJECTID);
867 additional_string = "initial log ";
868 next_bytenr = btrfs_super_log_root(super_tmp);
869 if (0 == next_bytenr)
870 continue;
871 break;
872 }
873
874 num_copies = btrfs_num_copies(fs_info, next_bytenr,
875 state->metablock_size);
876 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
877 pr_info("num_copies(log_bytenr=%llu) = %d\n",
878 next_bytenr, num_copies);
879 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
880 struct btrfsic_block *next_block;
881 struct btrfsic_block_data_ctx tmp_next_block_ctx;
882 struct btrfsic_block_link *l;
883
884 if (btrfsic_map_block(state, next_bytenr,
885 state->metablock_size,
886 &tmp_next_block_ctx,
887 mirror_num)) {
888 pr_info("btrfsic: btrfsic_map_block(bytenr @%llu, mirror %d) failed!\n",
889 next_bytenr, mirror_num);
890 brelse(bh);
891 return -1;
892 }
893
894 next_block = btrfsic_block_lookup_or_add(
895 state, &tmp_next_block_ctx,
896 additional_string, 1, 1, 0,
897 mirror_num, NULL);
898 if (NULL == next_block) {
899 btrfsic_release_block_ctx(&tmp_next_block_ctx);
900 brelse(bh);
901 return -1;
902 }
903
904 next_block->disk_key = tmp_disk_key;
905 next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
906 l = btrfsic_block_link_lookup_or_add(
907 state, &tmp_next_block_ctx,
908 next_block, superblock_tmp,
909 BTRFSIC_GENERATION_UNKNOWN);
910 btrfsic_release_block_ctx(&tmp_next_block_ctx);
911 if (NULL == l) {
912 brelse(bh);
913 return -1;
914 }
915 }
916 }
917 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
918 btrfsic_dump_tree_sub(state, superblock_tmp, 0);
919
920 brelse(bh);
921 return 0;
922}
923
924static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
925{
926 struct btrfsic_stack_frame *sf;
927
928 sf = kzalloc(sizeof(*sf), GFP_NOFS);
929 if (NULL == sf)
930 pr_info("btrfsic: alloc memory failed!\n");
931 else
932 sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
933 return sf;
934}
935
936static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
937{
938 BUG_ON(!(NULL == sf ||
939 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
940 kfree(sf);
941}
942
943static noinline_for_stack int btrfsic_process_metablock(
944 struct btrfsic_state *state,
945 struct btrfsic_block *const first_block,
946 struct btrfsic_block_data_ctx *const first_block_ctx,
947 int first_limit_nesting, int force_iodone_flag)
948{
949 struct btrfsic_stack_frame initial_stack_frame = { 0 };
950 struct btrfsic_stack_frame *sf;
951 struct btrfsic_stack_frame *next_stack;
952 struct btrfs_header *const first_hdr =
953 (struct btrfs_header *)first_block_ctx->datav[0];
954
955 BUG_ON(!first_hdr);
956 sf = &initial_stack_frame;
957 sf->error = 0;
958 sf->i = -1;
959 sf->limit_nesting = first_limit_nesting;
960 sf->block = first_block;
961 sf->block_ctx = first_block_ctx;
962 sf->next_block = NULL;
963 sf->hdr = first_hdr;
964 sf->prev = NULL;
965
966continue_with_new_stack_frame:
967 sf->block->generation = le64_to_cpu(sf->hdr->generation);
968 if (0 == sf->hdr->level) {
969 struct btrfs_leaf *const leafhdr =
970 (struct btrfs_leaf *)sf->hdr;
971
972 if (-1 == sf->i) {
973 sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
974
975 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
976 pr_info("leaf %llu items %d generation %llu owner %llu\n",
977 sf->block_ctx->start, sf->nr,
978 btrfs_stack_header_generation(
979 &leafhdr->header),
980 btrfs_stack_header_owner(
981 &leafhdr->header));
982 }
983
984continue_with_current_leaf_stack_frame:
985 if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
986 sf->i++;
987 sf->num_copies = 0;
988 }
989
990 if (sf->i < sf->nr) {
991 struct btrfs_item disk_item;
992 u32 disk_item_offset =
993 (uintptr_t)(leafhdr->items + sf->i) -
994 (uintptr_t)leafhdr;
995 struct btrfs_disk_key *disk_key;
996 u8 type;
997 u32 item_offset;
998 u32 item_size;
999
1000 if (disk_item_offset + sizeof(struct btrfs_item) >
1001 sf->block_ctx->len) {
1002leaf_item_out_of_bounce_error:
1003 pr_info("btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
1004 sf->block_ctx->start,
1005 sf->block_ctx->dev->name);
1006 goto one_stack_frame_backwards;
1007 }
1008 btrfsic_read_from_block_data(sf->block_ctx,
1009 &disk_item,
1010 disk_item_offset,
1011 sizeof(struct btrfs_item));
1012 item_offset = btrfs_stack_item_offset(&disk_item);
1013 item_size = btrfs_stack_item_size(&disk_item);
1014 disk_key = &disk_item.key;
1015 type = btrfs_disk_key_type(disk_key);
1016
1017 if (BTRFS_ROOT_ITEM_KEY == type) {
1018 struct btrfs_root_item root_item;
1019 u32 root_item_offset;
1020 u64 next_bytenr;
1021
1022 root_item_offset = item_offset +
1023 offsetof(struct btrfs_leaf, items);
1024 if (root_item_offset + item_size >
1025 sf->block_ctx->len)
1026 goto leaf_item_out_of_bounce_error;
1027 btrfsic_read_from_block_data(
1028 sf->block_ctx, &root_item,
1029 root_item_offset,
1030 item_size);
1031 next_bytenr = btrfs_root_bytenr(&root_item);
1032
1033 sf->error =
1034 btrfsic_create_link_to_next_block(
1035 state,
1036 sf->block,
1037 sf->block_ctx,
1038 next_bytenr,
1039 sf->limit_nesting,
1040 &sf->next_block_ctx,
1041 &sf->next_block,
1042 force_iodone_flag,
1043 &sf->num_copies,
1044 &sf->mirror_num,
1045 disk_key,
1046 btrfs_root_generation(
1047 &root_item));
1048 if (sf->error)
1049 goto one_stack_frame_backwards;
1050
1051 if (NULL != sf->next_block) {
1052 struct btrfs_header *const next_hdr =
1053 (struct btrfs_header *)
1054 sf->next_block_ctx.datav[0];
1055
1056 next_stack =
1057 btrfsic_stack_frame_alloc();
1058 if (NULL == next_stack) {
1059 sf->error = -1;
1060 btrfsic_release_block_ctx(
1061 &sf->
1062 next_block_ctx);
1063 goto one_stack_frame_backwards;
1064 }
1065
1066 next_stack->i = -1;
1067 next_stack->block = sf->next_block;
1068 next_stack->block_ctx =
1069 &sf->next_block_ctx;
1070 next_stack->next_block = NULL;
1071 next_stack->hdr = next_hdr;
1072 next_stack->limit_nesting =
1073 sf->limit_nesting - 1;
1074 next_stack->prev = sf;
1075 sf = next_stack;
1076 goto continue_with_new_stack_frame;
1077 }
1078 } else if (BTRFS_EXTENT_DATA_KEY == type &&
1079 state->include_extent_data) {
1080 sf->error = btrfsic_handle_extent_data(
1081 state,
1082 sf->block,
1083 sf->block_ctx,
1084 item_offset,
1085 force_iodone_flag);
1086 if (sf->error)
1087 goto one_stack_frame_backwards;
1088 }
1089
1090 goto continue_with_current_leaf_stack_frame;
1091 }
1092 } else {
1093 struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1094
1095 if (-1 == sf->i) {
1096 sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1097
1098 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1099 pr_info("node %llu level %d items %d generation %llu owner %llu\n",
1100 sf->block_ctx->start,
1101 nodehdr->header.level, sf->nr,
1102 btrfs_stack_header_generation(
1103 &nodehdr->header),
1104 btrfs_stack_header_owner(
1105 &nodehdr->header));
1106 }
1107
1108continue_with_current_node_stack_frame:
1109 if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1110 sf->i++;
1111 sf->num_copies = 0;
1112 }
1113
1114 if (sf->i < sf->nr) {
1115 struct btrfs_key_ptr key_ptr;
1116 u32 key_ptr_offset;
1117 u64 next_bytenr;
1118
1119 key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1120 (uintptr_t)nodehdr;
1121 if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1122 sf->block_ctx->len) {
1123 pr_info("btrfsic: node item out of bounce at logical %llu, dev %s\n",
1124 sf->block_ctx->start,
1125 sf->block_ctx->dev->name);
1126 goto one_stack_frame_backwards;
1127 }
1128 btrfsic_read_from_block_data(
1129 sf->block_ctx, &key_ptr, key_ptr_offset,
1130 sizeof(struct btrfs_key_ptr));
1131 next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1132
1133 sf->error = btrfsic_create_link_to_next_block(
1134 state,
1135 sf->block,
1136 sf->block_ctx,
1137 next_bytenr,
1138 sf->limit_nesting,
1139 &sf->next_block_ctx,
1140 &sf->next_block,
1141 force_iodone_flag,
1142 &sf->num_copies,
1143 &sf->mirror_num,
1144 &key_ptr.key,
1145 btrfs_stack_key_generation(&key_ptr));
1146 if (sf->error)
1147 goto one_stack_frame_backwards;
1148
1149 if (NULL != sf->next_block) {
1150 struct btrfs_header *const next_hdr =
1151 (struct btrfs_header *)
1152 sf->next_block_ctx.datav[0];
1153
1154 next_stack = btrfsic_stack_frame_alloc();
1155 if (NULL == next_stack) {
1156 sf->error = -1;
1157 goto one_stack_frame_backwards;
1158 }
1159
1160 next_stack->i = -1;
1161 next_stack->block = sf->next_block;
1162 next_stack->block_ctx = &sf->next_block_ctx;
1163 next_stack->next_block = NULL;
1164 next_stack->hdr = next_hdr;
1165 next_stack->limit_nesting =
1166 sf->limit_nesting - 1;
1167 next_stack->prev = sf;
1168 sf = next_stack;
1169 goto continue_with_new_stack_frame;
1170 }
1171
1172 goto continue_with_current_node_stack_frame;
1173 }
1174 }
1175
1176one_stack_frame_backwards:
1177 if (NULL != sf->prev) {
1178 struct btrfsic_stack_frame *const prev = sf->prev;
1179
1180 /* the one for the initial block is freed in the caller */
1181 btrfsic_release_block_ctx(sf->block_ctx);
1182
1183 if (sf->error) {
1184 prev->error = sf->error;
1185 btrfsic_stack_frame_free(sf);
1186 sf = prev;
1187 goto one_stack_frame_backwards;
1188 }
1189
1190 btrfsic_stack_frame_free(sf);
1191 sf = prev;
1192 goto continue_with_new_stack_frame;
1193 } else {
1194 BUG_ON(&initial_stack_frame != sf);
1195 }
1196
1197 return sf->error;
1198}
1199
1200static void btrfsic_read_from_block_data(
1201 struct btrfsic_block_data_ctx *block_ctx,
1202 void *dstv, u32 offset, size_t len)
1203{
1204 size_t cur;
1205 size_t pgoff;
1206 char *kaddr;
1207 char *dst = (char *)dstv;
1208 size_t start_offset = offset_in_page(block_ctx->start);
1209 unsigned long i = (start_offset + offset) >> PAGE_SHIFT;
1210
1211 WARN_ON(offset + len > block_ctx->len);
1212 pgoff = offset_in_page(start_offset + offset);
1213
1214 while (len > 0) {
1215 cur = min(len, ((size_t)PAGE_SIZE - pgoff));
1216 BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_SIZE));
1217 kaddr = block_ctx->datav[i];
1218 memcpy(dst, kaddr + pgoff, cur);
1219
1220 dst += cur;
1221 len -= cur;
1222 pgoff = 0;
1223 i++;
1224 }
1225}
1226
1227static int btrfsic_create_link_to_next_block(
1228 struct btrfsic_state *state,
1229 struct btrfsic_block *block,
1230 struct btrfsic_block_data_ctx *block_ctx,
1231 u64 next_bytenr,
1232 int limit_nesting,
1233 struct btrfsic_block_data_ctx *next_block_ctx,
1234 struct btrfsic_block **next_blockp,
1235 int force_iodone_flag,
1236 int *num_copiesp, int *mirror_nump,
1237 struct btrfs_disk_key *disk_key,
1238 u64 parent_generation)
1239{
1240 struct btrfs_fs_info *fs_info = state->fs_info;
1241 struct btrfsic_block *next_block = NULL;
1242 int ret;
1243 struct btrfsic_block_link *l;
1244 int did_alloc_block_link;
1245 int block_was_created;
1246
1247 *next_blockp = NULL;
1248 if (0 == *num_copiesp) {
1249 *num_copiesp = btrfs_num_copies(fs_info, next_bytenr,
1250 state->metablock_size);
1251 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1252 pr_info("num_copies(log_bytenr=%llu) = %d\n",
1253 next_bytenr, *num_copiesp);
1254 *mirror_nump = 1;
1255 }
1256
1257 if (*mirror_nump > *num_copiesp)
1258 return 0;
1259
1260 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1261 pr_info("btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1262 *mirror_nump);
1263 ret = btrfsic_map_block(state, next_bytenr,
1264 state->metablock_size,
1265 next_block_ctx, *mirror_nump);
1266 if (ret) {
1267 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1268 next_bytenr, *mirror_nump);
1269 btrfsic_release_block_ctx(next_block_ctx);
1270 *next_blockp = NULL;
1271 return -1;
1272 }
1273
1274 next_block = btrfsic_block_lookup_or_add(state,
1275 next_block_ctx, "referenced ",
1276 1, force_iodone_flag,
1277 !force_iodone_flag,
1278 *mirror_nump,
1279 &block_was_created);
1280 if (NULL == next_block) {
1281 btrfsic_release_block_ctx(next_block_ctx);
1282 *next_blockp = NULL;
1283 return -1;
1284 }
1285 if (block_was_created) {
1286 l = NULL;
1287 next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1288 } else {
1289 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1290 if (next_block->logical_bytenr != next_bytenr &&
1291 !(!next_block->is_metadata &&
1292 0 == next_block->logical_bytenr))
1293 pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1294 next_bytenr, next_block_ctx->dev->name,
1295 next_block_ctx->dev_bytenr, *mirror_nump,
1296 btrfsic_get_block_type(state,
1297 next_block),
1298 next_block->logical_bytenr);
1299 else
1300 pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1301 next_bytenr, next_block_ctx->dev->name,
1302 next_block_ctx->dev_bytenr, *mirror_nump,
1303 btrfsic_get_block_type(state,
1304 next_block));
1305 }
1306 next_block->logical_bytenr = next_bytenr;
1307
1308 next_block->mirror_num = *mirror_nump;
1309 l = btrfsic_block_link_hashtable_lookup(
1310 next_block_ctx->dev->bdev,
1311 next_block_ctx->dev_bytenr,
1312 block_ctx->dev->bdev,
1313 block_ctx->dev_bytenr,
1314 &state->block_link_hashtable);
1315 }
1316
1317 next_block->disk_key = *disk_key;
1318 if (NULL == l) {
1319 l = btrfsic_block_link_alloc();
1320 if (NULL == l) {
1321 pr_info("btrfsic: error, kmalloc failed!\n");
1322 btrfsic_release_block_ctx(next_block_ctx);
1323 *next_blockp = NULL;
1324 return -1;
1325 }
1326
1327 did_alloc_block_link = 1;
1328 l->block_ref_to = next_block;
1329 l->block_ref_from = block;
1330 l->ref_cnt = 1;
1331 l->parent_generation = parent_generation;
1332
1333 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1334 btrfsic_print_add_link(state, l);
1335
1336 list_add(&l->node_ref_to, &block->ref_to_list);
1337 list_add(&l->node_ref_from, &next_block->ref_from_list);
1338
1339 btrfsic_block_link_hashtable_add(l,
1340 &state->block_link_hashtable);
1341 } else {
1342 did_alloc_block_link = 0;
1343 if (0 == limit_nesting) {
1344 l->ref_cnt++;
1345 l->parent_generation = parent_generation;
1346 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1347 btrfsic_print_add_link(state, l);
1348 }
1349 }
1350
1351 if (limit_nesting > 0 && did_alloc_block_link) {
1352 ret = btrfsic_read_block(state, next_block_ctx);
1353 if (ret < (int)next_block_ctx->len) {
1354 pr_info("btrfsic: read block @logical %llu failed!\n",
1355 next_bytenr);
1356 btrfsic_release_block_ctx(next_block_ctx);
1357 *next_blockp = NULL;
1358 return -1;
1359 }
1360
1361 *next_blockp = next_block;
1362 } else {
1363 *next_blockp = NULL;
1364 }
1365 (*mirror_nump)++;
1366
1367 return 0;
1368}
1369
1370static int btrfsic_handle_extent_data(
1371 struct btrfsic_state *state,
1372 struct btrfsic_block *block,
1373 struct btrfsic_block_data_ctx *block_ctx,
1374 u32 item_offset, int force_iodone_flag)
1375{
1376 struct btrfs_fs_info *fs_info = state->fs_info;
1377 struct btrfs_file_extent_item file_extent_item;
1378 u64 file_extent_item_offset;
1379 u64 next_bytenr;
1380 u64 num_bytes;
1381 u64 generation;
1382 struct btrfsic_block_link *l;
1383 int ret;
1384
1385 file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1386 item_offset;
1387 if (file_extent_item_offset +
1388 offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1389 block_ctx->len) {
1390 pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1391 block_ctx->start, block_ctx->dev->name);
1392 return -1;
1393 }
1394
1395 btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1396 file_extent_item_offset,
1397 offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1398 if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1399 btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1400 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1401 pr_info("extent_data: type %u, disk_bytenr = %llu\n",
1402 file_extent_item.type,
1403 btrfs_stack_file_extent_disk_bytenr(
1404 &file_extent_item));
1405 return 0;
1406 }
1407
1408 if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1409 block_ctx->len) {
1410 pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1411 block_ctx->start, block_ctx->dev->name);
1412 return -1;
1413 }
1414 btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1415 file_extent_item_offset,
1416 sizeof(struct btrfs_file_extent_item));
1417 next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item);
1418 if (btrfs_stack_file_extent_compression(&file_extent_item) ==
1419 BTRFS_COMPRESS_NONE) {
1420 next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item);
1421 num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1422 } else {
1423 num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item);
1424 }
1425 generation = btrfs_stack_file_extent_generation(&file_extent_item);
1426
1427 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1428 pr_info("extent_data: type %u, disk_bytenr = %llu, offset = %llu, num_bytes = %llu\n",
1429 file_extent_item.type,
1430 btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1431 btrfs_stack_file_extent_offset(&file_extent_item),
1432 num_bytes);
1433 while (num_bytes > 0) {
1434 u32 chunk_len;
1435 int num_copies;
1436 int mirror_num;
1437
1438 if (num_bytes > state->datablock_size)
1439 chunk_len = state->datablock_size;
1440 else
1441 chunk_len = num_bytes;
1442
1443 num_copies = btrfs_num_copies(fs_info, next_bytenr,
1444 state->datablock_size);
1445 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1446 pr_info("num_copies(log_bytenr=%llu) = %d\n",
1447 next_bytenr, num_copies);
1448 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1449 struct btrfsic_block_data_ctx next_block_ctx;
1450 struct btrfsic_block *next_block;
1451 int block_was_created;
1452
1453 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1454 pr_info("btrfsic_handle_extent_data(mirror_num=%d)\n",
1455 mirror_num);
1456 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1457 pr_info("\tdisk_bytenr = %llu, num_bytes %u\n",
1458 next_bytenr, chunk_len);
1459 ret = btrfsic_map_block(state, next_bytenr,
1460 chunk_len, &next_block_ctx,
1461 mirror_num);
1462 if (ret) {
1463 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1464 next_bytenr, mirror_num);
1465 return -1;
1466 }
1467
1468 next_block = btrfsic_block_lookup_or_add(
1469 state,
1470 &next_block_ctx,
1471 "referenced ",
1472 0,
1473 force_iodone_flag,
1474 !force_iodone_flag,
1475 mirror_num,
1476 &block_was_created);
1477 if (NULL == next_block) {
1478 pr_info("btrfsic: error, kmalloc failed!\n");
1479 btrfsic_release_block_ctx(&next_block_ctx);
1480 return -1;
1481 }
1482 if (!block_was_created) {
1483 if ((state->print_mask &
1484 BTRFSIC_PRINT_MASK_VERBOSE) &&
1485 next_block->logical_bytenr != next_bytenr &&
1486 !(!next_block->is_metadata &&
1487 0 == next_block->logical_bytenr)) {
1488 pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, D, bytenr mismatch (!= stored %llu).\n",
1489 next_bytenr,
1490 next_block_ctx.dev->name,
1491 next_block_ctx.dev_bytenr,
1492 mirror_num,
1493 next_block->logical_bytenr);
1494 }
1495 next_block->logical_bytenr = next_bytenr;
1496 next_block->mirror_num = mirror_num;
1497 }
1498
1499 l = btrfsic_block_link_lookup_or_add(state,
1500 &next_block_ctx,
1501 next_block, block,
1502 generation);
1503 btrfsic_release_block_ctx(&next_block_ctx);
1504 if (NULL == l)
1505 return -1;
1506 }
1507
1508 next_bytenr += chunk_len;
1509 num_bytes -= chunk_len;
1510 }
1511
1512 return 0;
1513}
1514
1515static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1516 struct btrfsic_block_data_ctx *block_ctx_out,
1517 int mirror_num)
1518{
1519 struct btrfs_fs_info *fs_info = state->fs_info;
1520 int ret;
1521 u64 length;
1522 struct btrfs_bio *multi = NULL;
1523 struct btrfs_device *device;
1524
1525 length = len;
1526 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ,
1527 bytenr, &length, &multi, mirror_num);
1528
1529 if (ret) {
1530 block_ctx_out->start = 0;
1531 block_ctx_out->dev_bytenr = 0;
1532 block_ctx_out->len = 0;
1533 block_ctx_out->dev = NULL;
1534 block_ctx_out->datav = NULL;
1535 block_ctx_out->pagev = NULL;
1536 block_ctx_out->mem_to_free = NULL;
1537
1538 return ret;
1539 }
1540
1541 device = multi->stripes[0].dev;
1542 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
1543 !device->bdev || !device->name)
1544 block_ctx_out->dev = NULL;
1545 else
1546 block_ctx_out->dev = btrfsic_dev_state_lookup(
1547 device->bdev->bd_dev);
1548 block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1549 block_ctx_out->start = bytenr;
1550 block_ctx_out->len = len;
1551 block_ctx_out->datav = NULL;
1552 block_ctx_out->pagev = NULL;
1553 block_ctx_out->mem_to_free = NULL;
1554
1555 kfree(multi);
1556 if (NULL == block_ctx_out->dev) {
1557 ret = -ENXIO;
1558 pr_info("btrfsic: error, cannot lookup dev (#1)!\n");
1559 }
1560
1561 return ret;
1562}
1563
1564static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1565{
1566 if (block_ctx->mem_to_free) {
1567 unsigned int num_pages;
1568
1569 BUG_ON(!block_ctx->datav);
1570 BUG_ON(!block_ctx->pagev);
1571 num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1572 PAGE_SHIFT;
1573 while (num_pages > 0) {
1574 num_pages--;
1575 if (block_ctx->datav[num_pages]) {
1576 kunmap(block_ctx->pagev[num_pages]);
1577 block_ctx->datav[num_pages] = NULL;
1578 }
1579 if (block_ctx->pagev[num_pages]) {
1580 __free_page(block_ctx->pagev[num_pages]);
1581 block_ctx->pagev[num_pages] = NULL;
1582 }
1583 }
1584
1585 kfree(block_ctx->mem_to_free);
1586 block_ctx->mem_to_free = NULL;
1587 block_ctx->pagev = NULL;
1588 block_ctx->datav = NULL;
1589 }
1590}
1591
1592static int btrfsic_read_block(struct btrfsic_state *state,
1593 struct btrfsic_block_data_ctx *block_ctx)
1594{
1595 unsigned int num_pages;
1596 unsigned int i;
1597 size_t size;
1598 u64 dev_bytenr;
1599 int ret;
1600
1601 BUG_ON(block_ctx->datav);
1602 BUG_ON(block_ctx->pagev);
1603 BUG_ON(block_ctx->mem_to_free);
1604 if (!PAGE_ALIGNED(block_ctx->dev_bytenr)) {
1605 pr_info("btrfsic: read_block() with unaligned bytenr %llu\n",
1606 block_ctx->dev_bytenr);
1607 return -1;
1608 }
1609
1610 num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1611 PAGE_SHIFT;
1612 size = sizeof(*block_ctx->datav) + sizeof(*block_ctx->pagev);
1613 block_ctx->mem_to_free = kcalloc(num_pages, size, GFP_NOFS);
1614 if (!block_ctx->mem_to_free)
1615 return -ENOMEM;
1616 block_ctx->datav = block_ctx->mem_to_free;
1617 block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1618 for (i = 0; i < num_pages; i++) {
1619 block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1620 if (!block_ctx->pagev[i])
1621 return -1;
1622 }
1623
1624 dev_bytenr = block_ctx->dev_bytenr;
1625 for (i = 0; i < num_pages;) {
1626 struct bio *bio;
1627 unsigned int j;
1628
1629 bio = btrfs_io_bio_alloc(num_pages - i);
1630 bio_set_dev(bio, block_ctx->dev->bdev);
1631 bio->bi_iter.bi_sector = dev_bytenr >> 9;
1632 bio->bi_opf = REQ_OP_READ;
1633
1634 for (j = i; j < num_pages; j++) {
1635 ret = bio_add_page(bio, block_ctx->pagev[j],
1636 PAGE_SIZE, 0);
1637 if (PAGE_SIZE != ret)
1638 break;
1639 }
1640 if (j == i) {
1641 pr_info("btrfsic: error, failed to add a single page!\n");
1642 return -1;
1643 }
1644 if (submit_bio_wait(bio)) {
1645 pr_info("btrfsic: read error at logical %llu dev %s!\n",
1646 block_ctx->start, block_ctx->dev->name);
1647 bio_put(bio);
1648 return -1;
1649 }
1650 bio_put(bio);
1651 dev_bytenr += (j - i) * PAGE_SIZE;
1652 i = j;
1653 }
1654 for (i = 0; i < num_pages; i++)
1655 block_ctx->datav[i] = kmap(block_ctx->pagev[i]);
1656
1657 return block_ctx->len;
1658}
1659
1660static void btrfsic_dump_database(struct btrfsic_state *state)
1661{
1662 const struct btrfsic_block *b_all;
1663
1664 BUG_ON(NULL == state);
1665
1666 pr_info("all_blocks_list:\n");
1667 list_for_each_entry(b_all, &state->all_blocks_list, all_blocks_node) {
1668 const struct btrfsic_block_link *l;
1669
1670 pr_info("%c-block @%llu (%s/%llu/%d)\n",
1671 btrfsic_get_block_type(state, b_all),
1672 b_all->logical_bytenr, b_all->dev_state->name,
1673 b_all->dev_bytenr, b_all->mirror_num);
1674
1675 list_for_each_entry(l, &b_all->ref_to_list, node_ref_to) {
1676 pr_info(" %c @%llu (%s/%llu/%d) refers %u* to %c @%llu (%s/%llu/%d)\n",
1677 btrfsic_get_block_type(state, b_all),
1678 b_all->logical_bytenr, b_all->dev_state->name,
1679 b_all->dev_bytenr, b_all->mirror_num,
1680 l->ref_cnt,
1681 btrfsic_get_block_type(state, l->block_ref_to),
1682 l->block_ref_to->logical_bytenr,
1683 l->block_ref_to->dev_state->name,
1684 l->block_ref_to->dev_bytenr,
1685 l->block_ref_to->mirror_num);
1686 }
1687
1688 list_for_each_entry(l, &b_all->ref_from_list, node_ref_from) {
1689 pr_info(" %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
1690 btrfsic_get_block_type(state, b_all),
1691 b_all->logical_bytenr, b_all->dev_state->name,
1692 b_all->dev_bytenr, b_all->mirror_num,
1693 l->ref_cnt,
1694 btrfsic_get_block_type(state, l->block_ref_from),
1695 l->block_ref_from->logical_bytenr,
1696 l->block_ref_from->dev_state->name,
1697 l->block_ref_from->dev_bytenr,
1698 l->block_ref_from->mirror_num);
1699 }
1700
1701 pr_info("\n");
1702 }
1703}
1704
1705/*
1706 * Test whether the disk block contains a tree block (leaf or node)
1707 * (note that this test fails for the super block)
1708 */
1709static noinline_for_stack int btrfsic_test_for_metadata(
1710 struct btrfsic_state *state,
1711 char **datav, unsigned int num_pages)
1712{
1713 struct btrfs_fs_info *fs_info = state->fs_info;
1714 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
1715 struct btrfs_header *h;
1716 u8 csum[BTRFS_CSUM_SIZE];
1717 unsigned int i;
1718
1719 if (num_pages * PAGE_SIZE < state->metablock_size)
1720 return 1; /* not metadata */
1721 num_pages = state->metablock_size >> PAGE_SHIFT;
1722 h = (struct btrfs_header *)datav[0];
1723
1724 if (memcmp(h->fsid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE))
1725 return 1;
1726
1727 shash->tfm = fs_info->csum_shash;
1728 crypto_shash_init(shash);
1729
1730 for (i = 0; i < num_pages; i++) {
1731 u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1732 size_t sublen = i ? PAGE_SIZE :
1733 (PAGE_SIZE - BTRFS_CSUM_SIZE);
1734
1735 crypto_shash_update(shash, data, sublen);
1736 }
1737 crypto_shash_final(shash, csum);
1738 if (memcmp(csum, h->csum, state->csum_size))
1739 return 1;
1740
1741 return 0; /* is metadata */
1742}
1743
1744static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1745 u64 dev_bytenr, char **mapped_datav,
1746 unsigned int num_pages,
1747 struct bio *bio, int *bio_is_patched,
1748 struct buffer_head *bh,
1749 int submit_bio_bh_rw)
1750{
1751 int is_metadata;
1752 struct btrfsic_block *block;
1753 struct btrfsic_block_data_ctx block_ctx;
1754 int ret;
1755 struct btrfsic_state *state = dev_state->state;
1756 struct block_device *bdev = dev_state->bdev;
1757 unsigned int processed_len;
1758
1759 if (NULL != bio_is_patched)
1760 *bio_is_patched = 0;
1761
1762again:
1763 if (num_pages == 0)
1764 return;
1765
1766 processed_len = 0;
1767 is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1768 num_pages));
1769
1770 block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1771 &state->block_hashtable);
1772 if (NULL != block) {
1773 u64 bytenr = 0;
1774 struct btrfsic_block_link *l, *tmp;
1775
1776 if (block->is_superblock) {
1777 bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1778 mapped_datav[0]);
1779 if (num_pages * PAGE_SIZE <
1780 BTRFS_SUPER_INFO_SIZE) {
1781 pr_info("btrfsic: cannot work with too short bios!\n");
1782 return;
1783 }
1784 is_metadata = 1;
1785 BUG_ON(!PAGE_ALIGNED(BTRFS_SUPER_INFO_SIZE));
1786 processed_len = BTRFS_SUPER_INFO_SIZE;
1787 if (state->print_mask &
1788 BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1789 pr_info("[before new superblock is written]:\n");
1790 btrfsic_dump_tree_sub(state, block, 0);
1791 }
1792 }
1793 if (is_metadata) {
1794 if (!block->is_superblock) {
1795 if (num_pages * PAGE_SIZE <
1796 state->metablock_size) {
1797 pr_info("btrfsic: cannot work with too short bios!\n");
1798 return;
1799 }
1800 processed_len = state->metablock_size;
1801 bytenr = btrfs_stack_header_bytenr(
1802 (struct btrfs_header *)
1803 mapped_datav[0]);
1804 btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1805 dev_state,
1806 dev_bytenr);
1807 }
1808 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1809 if (block->logical_bytenr != bytenr &&
1810 !(!block->is_metadata &&
1811 block->logical_bytenr == 0))
1812 pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1813 bytenr, dev_state->name,
1814 dev_bytenr,
1815 block->mirror_num,
1816 btrfsic_get_block_type(state,
1817 block),
1818 block->logical_bytenr);
1819 else
1820 pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1821 bytenr, dev_state->name,
1822 dev_bytenr, block->mirror_num,
1823 btrfsic_get_block_type(state,
1824 block));
1825 }
1826 block->logical_bytenr = bytenr;
1827 } else {
1828 if (num_pages * PAGE_SIZE <
1829 state->datablock_size) {
1830 pr_info("btrfsic: cannot work with too short bios!\n");
1831 return;
1832 }
1833 processed_len = state->datablock_size;
1834 bytenr = block->logical_bytenr;
1835 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1836 pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1837 bytenr, dev_state->name, dev_bytenr,
1838 block->mirror_num,
1839 btrfsic_get_block_type(state, block));
1840 }
1841
1842 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1843 pr_info("ref_to_list: %cE, ref_from_list: %cE\n",
1844 list_empty(&block->ref_to_list) ? ' ' : '!',
1845 list_empty(&block->ref_from_list) ? ' ' : '!');
1846 if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1847 pr_info("btrfs: attempt to overwrite %c-block @%llu (%s/%llu/%d), old(gen=%llu, objectid=%llu, type=%d, offset=%llu), new(gen=%llu), which is referenced by most recent superblock (superblockgen=%llu)!\n",
1848 btrfsic_get_block_type(state, block), bytenr,
1849 dev_state->name, dev_bytenr, block->mirror_num,
1850 block->generation,
1851 btrfs_disk_key_objectid(&block->disk_key),
1852 block->disk_key.type,
1853 btrfs_disk_key_offset(&block->disk_key),
1854 btrfs_stack_header_generation(
1855 (struct btrfs_header *) mapped_datav[0]),
1856 state->max_superblock_generation);
1857 btrfsic_dump_tree(state);
1858 }
1859
1860 if (!block->is_iodone && !block->never_written) {
1861 pr_info("btrfs: attempt to overwrite %c-block @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu, which is not yet iodone!\n",
1862 btrfsic_get_block_type(state, block), bytenr,
1863 dev_state->name, dev_bytenr, block->mirror_num,
1864 block->generation,
1865 btrfs_stack_header_generation(
1866 (struct btrfs_header *)
1867 mapped_datav[0]));
1868 /* it would not be safe to go on */
1869 btrfsic_dump_tree(state);
1870 goto continue_loop;
1871 }
1872
1873 /*
1874 * Clear all references of this block. Do not free
1875 * the block itself even if is not referenced anymore
1876 * because it still carries valuable information
1877 * like whether it was ever written and IO completed.
1878 */
1879 list_for_each_entry_safe(l, tmp, &block->ref_to_list,
1880 node_ref_to) {
1881 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1882 btrfsic_print_rem_link(state, l);
1883 l->ref_cnt--;
1884 if (0 == l->ref_cnt) {
1885 list_del(&l->node_ref_to);
1886 list_del(&l->node_ref_from);
1887 btrfsic_block_link_hashtable_remove(l);
1888 btrfsic_block_link_free(l);
1889 }
1890 }
1891
1892 block_ctx.dev = dev_state;
1893 block_ctx.dev_bytenr = dev_bytenr;
1894 block_ctx.start = bytenr;
1895 block_ctx.len = processed_len;
1896 block_ctx.pagev = NULL;
1897 block_ctx.mem_to_free = NULL;
1898 block_ctx.datav = mapped_datav;
1899
1900 if (is_metadata || state->include_extent_data) {
1901 block->never_written = 0;
1902 block->iodone_w_error = 0;
1903 if (NULL != bio) {
1904 block->is_iodone = 0;
1905 BUG_ON(NULL == bio_is_patched);
1906 if (!*bio_is_patched) {
1907 block->orig_bio_bh_private =
1908 bio->bi_private;
1909 block->orig_bio_bh_end_io.bio =
1910 bio->bi_end_io;
1911 block->next_in_same_bio = NULL;
1912 bio->bi_private = block;
1913 bio->bi_end_io = btrfsic_bio_end_io;
1914 *bio_is_patched = 1;
1915 } else {
1916 struct btrfsic_block *chained_block =
1917 (struct btrfsic_block *)
1918 bio->bi_private;
1919
1920 BUG_ON(NULL == chained_block);
1921 block->orig_bio_bh_private =
1922 chained_block->orig_bio_bh_private;
1923 block->orig_bio_bh_end_io.bio =
1924 chained_block->orig_bio_bh_end_io.
1925 bio;
1926 block->next_in_same_bio = chained_block;
1927 bio->bi_private = block;
1928 }
1929 } else if (NULL != bh) {
1930 block->is_iodone = 0;
1931 block->orig_bio_bh_private = bh->b_private;
1932 block->orig_bio_bh_end_io.bh = bh->b_end_io;
1933 block->next_in_same_bio = NULL;
1934 bh->b_private = block;
1935 bh->b_end_io = btrfsic_bh_end_io;
1936 } else {
1937 block->is_iodone = 1;
1938 block->orig_bio_bh_private = NULL;
1939 block->orig_bio_bh_end_io.bio = NULL;
1940 block->next_in_same_bio = NULL;
1941 }
1942 }
1943
1944 block->flush_gen = dev_state->last_flush_gen + 1;
1945 block->submit_bio_bh_rw = submit_bio_bh_rw;
1946 if (is_metadata) {
1947 block->logical_bytenr = bytenr;
1948 block->is_metadata = 1;
1949 if (block->is_superblock) {
1950 BUG_ON(PAGE_SIZE !=
1951 BTRFS_SUPER_INFO_SIZE);
1952 ret = btrfsic_process_written_superblock(
1953 state,
1954 block,
1955 (struct btrfs_super_block *)
1956 mapped_datav[0]);
1957 if (state->print_mask &
1958 BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
1959 pr_info("[after new superblock is written]:\n");
1960 btrfsic_dump_tree_sub(state, block, 0);
1961 }
1962 } else {
1963 block->mirror_num = 0; /* unknown */
1964 ret = btrfsic_process_metablock(
1965 state,
1966 block,
1967 &block_ctx,
1968 0, 0);
1969 }
1970 if (ret)
1971 pr_info("btrfsic: btrfsic_process_metablock(root @%llu) failed!\n",
1972 dev_bytenr);
1973 } else {
1974 block->is_metadata = 0;
1975 block->mirror_num = 0; /* unknown */
1976 block->generation = BTRFSIC_GENERATION_UNKNOWN;
1977 if (!state->include_extent_data
1978 && list_empty(&block->ref_from_list)) {
1979 /*
1980 * disk block is overwritten with extent
1981 * data (not meta data) and we are configured
1982 * to not include extent data: take the
1983 * chance and free the block's memory
1984 */
1985 btrfsic_block_hashtable_remove(block);
1986 list_del(&block->all_blocks_node);
1987 btrfsic_block_free(block);
1988 }
1989 }
1990 btrfsic_release_block_ctx(&block_ctx);
1991 } else {
1992 /* block has not been found in hash table */
1993 u64 bytenr;
1994
1995 if (!is_metadata) {
1996 processed_len = state->datablock_size;
1997 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1998 pr_info("Written block (%s/%llu/?) !found in hash table, D.\n",
1999 dev_state->name, dev_bytenr);
2000 if (!state->include_extent_data) {
2001 /* ignore that written D block */
2002 goto continue_loop;
2003 }
2004
2005 /* this is getting ugly for the
2006 * include_extent_data case... */
2007 bytenr = 0; /* unknown */
2008 } else {
2009 processed_len = state->metablock_size;
2010 bytenr = btrfs_stack_header_bytenr(
2011 (struct btrfs_header *)
2012 mapped_datav[0]);
2013 btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
2014 dev_bytenr);
2015 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2016 pr_info("Written block @%llu (%s/%llu/?) !found in hash table, M.\n",
2017 bytenr, dev_state->name, dev_bytenr);
2018 }
2019
2020 block_ctx.dev = dev_state;
2021 block_ctx.dev_bytenr = dev_bytenr;
2022 block_ctx.start = bytenr;
2023 block_ctx.len = processed_len;
2024 block_ctx.pagev = NULL;
2025 block_ctx.mem_to_free = NULL;
2026 block_ctx.datav = mapped_datav;
2027
2028 block = btrfsic_block_alloc();
2029 if (NULL == block) {
2030 pr_info("btrfsic: error, kmalloc failed!\n");
2031 btrfsic_release_block_ctx(&block_ctx);
2032 goto continue_loop;
2033 }
2034 block->dev_state = dev_state;
2035 block->dev_bytenr = dev_bytenr;
2036 block->logical_bytenr = bytenr;
2037 block->is_metadata = is_metadata;
2038 block->never_written = 0;
2039 block->iodone_w_error = 0;
2040 block->mirror_num = 0; /* unknown */
2041 block->flush_gen = dev_state->last_flush_gen + 1;
2042 block->submit_bio_bh_rw = submit_bio_bh_rw;
2043 if (NULL != bio) {
2044 block->is_iodone = 0;
2045 BUG_ON(NULL == bio_is_patched);
2046 if (!*bio_is_patched) {
2047 block->orig_bio_bh_private = bio->bi_private;
2048 block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2049 block->next_in_same_bio = NULL;
2050 bio->bi_private = block;
2051 bio->bi_end_io = btrfsic_bio_end_io;
2052 *bio_is_patched = 1;
2053 } else {
2054 struct btrfsic_block *chained_block =
2055 (struct btrfsic_block *)
2056 bio->bi_private;
2057
2058 BUG_ON(NULL == chained_block);
2059 block->orig_bio_bh_private =
2060 chained_block->orig_bio_bh_private;
2061 block->orig_bio_bh_end_io.bio =
2062 chained_block->orig_bio_bh_end_io.bio;
2063 block->next_in_same_bio = chained_block;
2064 bio->bi_private = block;
2065 }
2066 } else if (NULL != bh) {
2067 block->is_iodone = 0;
2068 block->orig_bio_bh_private = bh->b_private;
2069 block->orig_bio_bh_end_io.bh = bh->b_end_io;
2070 block->next_in_same_bio = NULL;
2071 bh->b_private = block;
2072 bh->b_end_io = btrfsic_bh_end_io;
2073 } else {
2074 block->is_iodone = 1;
2075 block->orig_bio_bh_private = NULL;
2076 block->orig_bio_bh_end_io.bio = NULL;
2077 block->next_in_same_bio = NULL;
2078 }
2079 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2080 pr_info("New written %c-block @%llu (%s/%llu/%d)\n",
2081 is_metadata ? 'M' : 'D',
2082 block->logical_bytenr, block->dev_state->name,
2083 block->dev_bytenr, block->mirror_num);
2084 list_add(&block->all_blocks_node, &state->all_blocks_list);
2085 btrfsic_block_hashtable_add(block, &state->block_hashtable);
2086
2087 if (is_metadata) {
2088 ret = btrfsic_process_metablock(state, block,
2089 &block_ctx, 0, 0);
2090 if (ret)
2091 pr_info("btrfsic: process_metablock(root @%llu) failed!\n",
2092 dev_bytenr);
2093 }
2094 btrfsic_release_block_ctx(&block_ctx);
2095 }
2096
2097continue_loop:
2098 BUG_ON(!processed_len);
2099 dev_bytenr += processed_len;
2100 mapped_datav += processed_len >> PAGE_SHIFT;
2101 num_pages -= processed_len >> PAGE_SHIFT;
2102 goto again;
2103}
2104
2105static void btrfsic_bio_end_io(struct bio *bp)
2106{
2107 struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2108 int iodone_w_error;
2109
2110 /* mutex is not held! This is not save if IO is not yet completed
2111 * on umount */
2112 iodone_w_error = 0;
2113 if (bp->bi_status)
2114 iodone_w_error = 1;
2115
2116 BUG_ON(NULL == block);
2117 bp->bi_private = block->orig_bio_bh_private;
2118 bp->bi_end_io = block->orig_bio_bh_end_io.bio;
2119
2120 do {
2121 struct btrfsic_block *next_block;
2122 struct btrfsic_dev_state *const dev_state = block->dev_state;
2123
2124 if ((dev_state->state->print_mask &
2125 BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2126 pr_info("bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2127 bp->bi_status,
2128 btrfsic_get_block_type(dev_state->state, block),
2129 block->logical_bytenr, dev_state->name,
2130 block->dev_bytenr, block->mirror_num);
2131 next_block = block->next_in_same_bio;
2132 block->iodone_w_error = iodone_w_error;
2133 if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2134 dev_state->last_flush_gen++;
2135 if ((dev_state->state->print_mask &
2136 BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2137 pr_info("bio_end_io() new %s flush_gen=%llu\n",
2138 dev_state->name,
2139 dev_state->last_flush_gen);
2140 }
2141 if (block->submit_bio_bh_rw & REQ_FUA)
2142 block->flush_gen = 0; /* FUA completed means block is
2143 * on disk */
2144 block->is_iodone = 1; /* for FLUSH, this releases the block */
2145 block = next_block;
2146 } while (NULL != block);
2147
2148 bp->bi_end_io(bp);
2149}
2150
2151static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate)
2152{
2153 struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private;
2154 int iodone_w_error = !uptodate;
2155 struct btrfsic_dev_state *dev_state;
2156
2157 BUG_ON(NULL == block);
2158 dev_state = block->dev_state;
2159 if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2160 pr_info("bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n",
2161 iodone_w_error,
2162 btrfsic_get_block_type(dev_state->state, block),
2163 block->logical_bytenr, block->dev_state->name,
2164 block->dev_bytenr, block->mirror_num);
2165
2166 block->iodone_w_error = iodone_w_error;
2167 if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2168 dev_state->last_flush_gen++;
2169 if ((dev_state->state->print_mask &
2170 BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2171 pr_info("bh_end_io() new %s flush_gen=%llu\n",
2172 dev_state->name, dev_state->last_flush_gen);
2173 }
2174 if (block->submit_bio_bh_rw & REQ_FUA)
2175 block->flush_gen = 0; /* FUA completed means block is on disk */
2176
2177 bh->b_private = block->orig_bio_bh_private;
2178 bh->b_end_io = block->orig_bio_bh_end_io.bh;
2179 block->is_iodone = 1; /* for FLUSH, this releases the block */
2180 bh->b_end_io(bh, uptodate);
2181}
2182
2183static int btrfsic_process_written_superblock(
2184 struct btrfsic_state *state,
2185 struct btrfsic_block *const superblock,
2186 struct btrfs_super_block *const super_hdr)
2187{
2188 struct btrfs_fs_info *fs_info = state->fs_info;
2189 int pass;
2190
2191 superblock->generation = btrfs_super_generation(super_hdr);
2192 if (!(superblock->generation > state->max_superblock_generation ||
2193 0 == state->max_superblock_generation)) {
2194 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2195 pr_info("btrfsic: superblock @%llu (%s/%llu/%d) with old gen %llu <= %llu\n",
2196 superblock->logical_bytenr,
2197 superblock->dev_state->name,
2198 superblock->dev_bytenr, superblock->mirror_num,
2199 btrfs_super_generation(super_hdr),
2200 state->max_superblock_generation);
2201 } else {
2202 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2203 pr_info("btrfsic: got new superblock @%llu (%s/%llu/%d) with new gen %llu > %llu\n",
2204 superblock->logical_bytenr,
2205 superblock->dev_state->name,
2206 superblock->dev_bytenr, superblock->mirror_num,
2207 btrfs_super_generation(super_hdr),
2208 state->max_superblock_generation);
2209
2210 state->max_superblock_generation =
2211 btrfs_super_generation(super_hdr);
2212 state->latest_superblock = superblock;
2213 }
2214
2215 for (pass = 0; pass < 3; pass++) {
2216 int ret;
2217 u64 next_bytenr;
2218 struct btrfsic_block *next_block;
2219 struct btrfsic_block_data_ctx tmp_next_block_ctx;
2220 struct btrfsic_block_link *l;
2221 int num_copies;
2222 int mirror_num;
2223 const char *additional_string = NULL;
2224 struct btrfs_disk_key tmp_disk_key = {0};
2225
2226 btrfs_set_disk_key_objectid(&tmp_disk_key,
2227 BTRFS_ROOT_ITEM_KEY);
2228 btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2229
2230 switch (pass) {
2231 case 0:
2232 btrfs_set_disk_key_objectid(&tmp_disk_key,
2233 BTRFS_ROOT_TREE_OBJECTID);
2234 additional_string = "root ";
2235 next_bytenr = btrfs_super_root(super_hdr);
2236 if (state->print_mask &
2237 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2238 pr_info("root@%llu\n", next_bytenr);
2239 break;
2240 case 1:
2241 btrfs_set_disk_key_objectid(&tmp_disk_key,
2242 BTRFS_CHUNK_TREE_OBJECTID);
2243 additional_string = "chunk ";
2244 next_bytenr = btrfs_super_chunk_root(super_hdr);
2245 if (state->print_mask &
2246 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2247 pr_info("chunk@%llu\n", next_bytenr);
2248 break;
2249 case 2:
2250 btrfs_set_disk_key_objectid(&tmp_disk_key,
2251 BTRFS_TREE_LOG_OBJECTID);
2252 additional_string = "log ";
2253 next_bytenr = btrfs_super_log_root(super_hdr);
2254 if (0 == next_bytenr)
2255 continue;
2256 if (state->print_mask &
2257 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2258 pr_info("log@%llu\n", next_bytenr);
2259 break;
2260 }
2261
2262 num_copies = btrfs_num_copies(fs_info, next_bytenr,
2263 BTRFS_SUPER_INFO_SIZE);
2264 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2265 pr_info("num_copies(log_bytenr=%llu) = %d\n",
2266 next_bytenr, num_copies);
2267 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2268 int was_created;
2269
2270 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2271 pr_info("btrfsic_process_written_superblock(mirror_num=%d)\n", mirror_num);
2272 ret = btrfsic_map_block(state, next_bytenr,
2273 BTRFS_SUPER_INFO_SIZE,
2274 &tmp_next_block_ctx,
2275 mirror_num);
2276 if (ret) {
2277 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
2278 next_bytenr, mirror_num);
2279 return -1;
2280 }
2281
2282 next_block = btrfsic_block_lookup_or_add(
2283 state,
2284 &tmp_next_block_ctx,
2285 additional_string,
2286 1, 0, 1,
2287 mirror_num,
2288 &was_created);
2289 if (NULL == next_block) {
2290 pr_info("btrfsic: error, kmalloc failed!\n");
2291 btrfsic_release_block_ctx(&tmp_next_block_ctx);
2292 return -1;
2293 }
2294
2295 next_block->disk_key = tmp_disk_key;
2296 if (was_created)
2297 next_block->generation =
2298 BTRFSIC_GENERATION_UNKNOWN;
2299 l = btrfsic_block_link_lookup_or_add(
2300 state,
2301 &tmp_next_block_ctx,
2302 next_block,
2303 superblock,
2304 BTRFSIC_GENERATION_UNKNOWN);
2305 btrfsic_release_block_ctx(&tmp_next_block_ctx);
2306 if (NULL == l)
2307 return -1;
2308 }
2309 }
2310
2311 if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2312 btrfsic_dump_tree(state);
2313
2314 return 0;
2315}
2316
2317static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2318 struct btrfsic_block *const block,
2319 int recursion_level)
2320{
2321 const struct btrfsic_block_link *l;
2322 int ret = 0;
2323
2324 if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2325 /*
2326 * Note that this situation can happen and does not
2327 * indicate an error in regular cases. It happens
2328 * when disk blocks are freed and later reused.
2329 * The check-integrity module is not aware of any
2330 * block free operations, it just recognizes block
2331 * write operations. Therefore it keeps the linkage
2332 * information for a block until a block is
2333 * rewritten. This can temporarily cause incorrect
2334 * and even circular linkage information. This
2335 * causes no harm unless such blocks are referenced
2336 * by the most recent super block.
2337 */
2338 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2339 pr_info("btrfsic: abort cyclic linkage (case 1).\n");
2340
2341 return ret;
2342 }
2343
2344 /*
2345 * This algorithm is recursive because the amount of used stack
2346 * space is very small and the max recursion depth is limited.
2347 */
2348 list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2349 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2350 pr_info("rl=%d, %c @%llu (%s/%llu/%d) %u* refers to %c @%llu (%s/%llu/%d)\n",
2351 recursion_level,
2352 btrfsic_get_block_type(state, block),
2353 block->logical_bytenr, block->dev_state->name,
2354 block->dev_bytenr, block->mirror_num,
2355 l->ref_cnt,
2356 btrfsic_get_block_type(state, l->block_ref_to),
2357 l->block_ref_to->logical_bytenr,
2358 l->block_ref_to->dev_state->name,
2359 l->block_ref_to->dev_bytenr,
2360 l->block_ref_to->mirror_num);
2361 if (l->block_ref_to->never_written) {
2362 pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is never written!\n",
2363 btrfsic_get_block_type(state, l->block_ref_to),
2364 l->block_ref_to->logical_bytenr,
2365 l->block_ref_to->dev_state->name,
2366 l->block_ref_to->dev_bytenr,
2367 l->block_ref_to->mirror_num);
2368 ret = -1;
2369 } else if (!l->block_ref_to->is_iodone) {
2370 pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is not yet iodone!\n",
2371 btrfsic_get_block_type(state, l->block_ref_to),
2372 l->block_ref_to->logical_bytenr,
2373 l->block_ref_to->dev_state->name,
2374 l->block_ref_to->dev_bytenr,
2375 l->block_ref_to->mirror_num);
2376 ret = -1;
2377 } else if (l->block_ref_to->iodone_w_error) {
2378 pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which has write error!\n",
2379 btrfsic_get_block_type(state, l->block_ref_to),
2380 l->block_ref_to->logical_bytenr,
2381 l->block_ref_to->dev_state->name,
2382 l->block_ref_to->dev_bytenr,
2383 l->block_ref_to->mirror_num);
2384 ret = -1;
2385 } else if (l->parent_generation !=
2386 l->block_ref_to->generation &&
2387 BTRFSIC_GENERATION_UNKNOWN !=
2388 l->parent_generation &&
2389 BTRFSIC_GENERATION_UNKNOWN !=
2390 l->block_ref_to->generation) {
2391 pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) with generation %llu != parent generation %llu!\n",
2392 btrfsic_get_block_type(state, l->block_ref_to),
2393 l->block_ref_to->logical_bytenr,
2394 l->block_ref_to->dev_state->name,
2395 l->block_ref_to->dev_bytenr,
2396 l->block_ref_to->mirror_num,
2397 l->block_ref_to->generation,
2398 l->parent_generation);
2399 ret = -1;
2400 } else if (l->block_ref_to->flush_gen >
2401 l->block_ref_to->dev_state->last_flush_gen) {
2402 pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is not flushed out of disk's write cache (block flush_gen=%llu, dev->flush_gen=%llu)!\n",
2403 btrfsic_get_block_type(state, l->block_ref_to),
2404 l->block_ref_to->logical_bytenr,
2405 l->block_ref_to->dev_state->name,
2406 l->block_ref_to->dev_bytenr,
2407 l->block_ref_to->mirror_num, block->flush_gen,
2408 l->block_ref_to->dev_state->last_flush_gen);
2409 ret = -1;
2410 } else if (-1 == btrfsic_check_all_ref_blocks(state,
2411 l->block_ref_to,
2412 recursion_level +
2413 1)) {
2414 ret = -1;
2415 }
2416 }
2417
2418 return ret;
2419}
2420
2421static int btrfsic_is_block_ref_by_superblock(
2422 const struct btrfsic_state *state,
2423 const struct btrfsic_block *block,
2424 int recursion_level)
2425{
2426 const struct btrfsic_block_link *l;
2427
2428 if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2429 /* refer to comment at "abort cyclic linkage (case 1)" */
2430 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2431 pr_info("btrfsic: abort cyclic linkage (case 2).\n");
2432
2433 return 0;
2434 }
2435
2436 /*
2437 * This algorithm is recursive because the amount of used stack space
2438 * is very small and the max recursion depth is limited.
2439 */
2440 list_for_each_entry(l, &block->ref_from_list, node_ref_from) {
2441 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2442 pr_info("rl=%d, %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
2443 recursion_level,
2444 btrfsic_get_block_type(state, block),
2445 block->logical_bytenr, block->dev_state->name,
2446 block->dev_bytenr, block->mirror_num,
2447 l->ref_cnt,
2448 btrfsic_get_block_type(state, l->block_ref_from),
2449 l->block_ref_from->logical_bytenr,
2450 l->block_ref_from->dev_state->name,
2451 l->block_ref_from->dev_bytenr,
2452 l->block_ref_from->mirror_num);
2453 if (l->block_ref_from->is_superblock &&
2454 state->latest_superblock->dev_bytenr ==
2455 l->block_ref_from->dev_bytenr &&
2456 state->latest_superblock->dev_state->bdev ==
2457 l->block_ref_from->dev_state->bdev)
2458 return 1;
2459 else if (btrfsic_is_block_ref_by_superblock(state,
2460 l->block_ref_from,
2461 recursion_level +
2462 1))
2463 return 1;
2464 }
2465
2466 return 0;
2467}
2468
2469static void btrfsic_print_add_link(const struct btrfsic_state *state,
2470 const struct btrfsic_block_link *l)
2471{
2472 pr_info("Add %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2473 l->ref_cnt,
2474 btrfsic_get_block_type(state, l->block_ref_from),
2475 l->block_ref_from->logical_bytenr,
2476 l->block_ref_from->dev_state->name,
2477 l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2478 btrfsic_get_block_type(state, l->block_ref_to),
2479 l->block_ref_to->logical_bytenr,
2480 l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2481 l->block_ref_to->mirror_num);
2482}
2483
2484static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2485 const struct btrfsic_block_link *l)
2486{
2487 pr_info("Rem %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2488 l->ref_cnt,
2489 btrfsic_get_block_type(state, l->block_ref_from),
2490 l->block_ref_from->logical_bytenr,
2491 l->block_ref_from->dev_state->name,
2492 l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2493 btrfsic_get_block_type(state, l->block_ref_to),
2494 l->block_ref_to->logical_bytenr,
2495 l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2496 l->block_ref_to->mirror_num);
2497}
2498
2499static char btrfsic_get_block_type(const struct btrfsic_state *state,
2500 const struct btrfsic_block *block)
2501{
2502 if (block->is_superblock &&
2503 state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2504 state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2505 return 'S';
2506 else if (block->is_superblock)
2507 return 's';
2508 else if (block->is_metadata)
2509 return 'M';
2510 else
2511 return 'D';
2512}
2513
2514static void btrfsic_dump_tree(const struct btrfsic_state *state)
2515{
2516 btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2517}
2518
2519static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2520 const struct btrfsic_block *block,
2521 int indent_level)
2522{
2523 const struct btrfsic_block_link *l;
2524 int indent_add;
2525 static char buf[80];
2526 int cursor_position;
2527
2528 /*
2529 * Should better fill an on-stack buffer with a complete line and
2530 * dump it at once when it is time to print a newline character.
2531 */
2532
2533 /*
2534 * This algorithm is recursive because the amount of used stack space
2535 * is very small and the max recursion depth is limited.
2536 */
2537 indent_add = sprintf(buf, "%c-%llu(%s/%llu/%u)",
2538 btrfsic_get_block_type(state, block),
2539 block->logical_bytenr, block->dev_state->name,
2540 block->dev_bytenr, block->mirror_num);
2541 if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2542 printk("[...]\n");
2543 return;
2544 }
2545 printk(buf);
2546 indent_level += indent_add;
2547 if (list_empty(&block->ref_to_list)) {
2548 printk("\n");
2549 return;
2550 }
2551 if (block->mirror_num > 1 &&
2552 !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2553 printk(" [...]\n");
2554 return;
2555 }
2556
2557 cursor_position = indent_level;
2558 list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2559 while (cursor_position < indent_level) {
2560 printk(" ");
2561 cursor_position++;
2562 }
2563 if (l->ref_cnt > 1)
2564 indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2565 else
2566 indent_add = sprintf(buf, " --> ");
2567 if (indent_level + indent_add >
2568 BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2569 printk("[...]\n");
2570 cursor_position = 0;
2571 continue;
2572 }
2573
2574 printk(buf);
2575
2576 btrfsic_dump_tree_sub(state, l->block_ref_to,
2577 indent_level + indent_add);
2578 cursor_position = 0;
2579 }
2580}
2581
2582static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2583 struct btrfsic_state *state,
2584 struct btrfsic_block_data_ctx *next_block_ctx,
2585 struct btrfsic_block *next_block,
2586 struct btrfsic_block *from_block,
2587 u64 parent_generation)
2588{
2589 struct btrfsic_block_link *l;
2590
2591 l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2592 next_block_ctx->dev_bytenr,
2593 from_block->dev_state->bdev,
2594 from_block->dev_bytenr,
2595 &state->block_link_hashtable);
2596 if (NULL == l) {
2597 l = btrfsic_block_link_alloc();
2598 if (NULL == l) {
2599 pr_info("btrfsic: error, kmalloc failed!\n");
2600 return NULL;
2601 }
2602
2603 l->block_ref_to = next_block;
2604 l->block_ref_from = from_block;
2605 l->ref_cnt = 1;
2606 l->parent_generation = parent_generation;
2607
2608 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2609 btrfsic_print_add_link(state, l);
2610
2611 list_add(&l->node_ref_to, &from_block->ref_to_list);
2612 list_add(&l->node_ref_from, &next_block->ref_from_list);
2613
2614 btrfsic_block_link_hashtable_add(l,
2615 &state->block_link_hashtable);
2616 } else {
2617 l->ref_cnt++;
2618 l->parent_generation = parent_generation;
2619 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2620 btrfsic_print_add_link(state, l);
2621 }
2622
2623 return l;
2624}
2625
2626static struct btrfsic_block *btrfsic_block_lookup_or_add(
2627 struct btrfsic_state *state,
2628 struct btrfsic_block_data_ctx *block_ctx,
2629 const char *additional_string,
2630 int is_metadata,
2631 int is_iodone,
2632 int never_written,
2633 int mirror_num,
2634 int *was_created)
2635{
2636 struct btrfsic_block *block;
2637
2638 block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2639 block_ctx->dev_bytenr,
2640 &state->block_hashtable);
2641 if (NULL == block) {
2642 struct btrfsic_dev_state *dev_state;
2643
2644 block = btrfsic_block_alloc();
2645 if (NULL == block) {
2646 pr_info("btrfsic: error, kmalloc failed!\n");
2647 return NULL;
2648 }
2649 dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev->bd_dev);
2650 if (NULL == dev_state) {
2651 pr_info("btrfsic: error, lookup dev_state failed!\n");
2652 btrfsic_block_free(block);
2653 return NULL;
2654 }
2655 block->dev_state = dev_state;
2656 block->dev_bytenr = block_ctx->dev_bytenr;
2657 block->logical_bytenr = block_ctx->start;
2658 block->is_metadata = is_metadata;
2659 block->is_iodone = is_iodone;
2660 block->never_written = never_written;
2661 block->mirror_num = mirror_num;
2662 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2663 pr_info("New %s%c-block @%llu (%s/%llu/%d)\n",
2664 additional_string,
2665 btrfsic_get_block_type(state, block),
2666 block->logical_bytenr, dev_state->name,
2667 block->dev_bytenr, mirror_num);
2668 list_add(&block->all_blocks_node, &state->all_blocks_list);
2669 btrfsic_block_hashtable_add(block, &state->block_hashtable);
2670 if (NULL != was_created)
2671 *was_created = 1;
2672 } else {
2673 if (NULL != was_created)
2674 *was_created = 0;
2675 }
2676
2677 return block;
2678}
2679
2680static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2681 u64 bytenr,
2682 struct btrfsic_dev_state *dev_state,
2683 u64 dev_bytenr)
2684{
2685 struct btrfs_fs_info *fs_info = state->fs_info;
2686 struct btrfsic_block_data_ctx block_ctx;
2687 int num_copies;
2688 int mirror_num;
2689 int match = 0;
2690 int ret;
2691
2692 num_copies = btrfs_num_copies(fs_info, bytenr, state->metablock_size);
2693
2694 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2695 ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2696 &block_ctx, mirror_num);
2697 if (ret) {
2698 pr_info("btrfsic: btrfsic_map_block(logical @%llu, mirror %d) failed!\n",
2699 bytenr, mirror_num);
2700 continue;
2701 }
2702
2703 if (dev_state->bdev == block_ctx.dev->bdev &&
2704 dev_bytenr == block_ctx.dev_bytenr) {
2705 match++;
2706 btrfsic_release_block_ctx(&block_ctx);
2707 break;
2708 }
2709 btrfsic_release_block_ctx(&block_ctx);
2710 }
2711
2712 if (WARN_ON(!match)) {
2713 pr_info("btrfs: attempt to write M-block which contains logical bytenr that doesn't map to dev+physical bytenr of submit_bio, buffer->log_bytenr=%llu, submit_bio(bdev=%s, phys_bytenr=%llu)!\n",
2714 bytenr, dev_state->name, dev_bytenr);
2715 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2716 ret = btrfsic_map_block(state, bytenr,
2717 state->metablock_size,
2718 &block_ctx, mirror_num);
2719 if (ret)
2720 continue;
2721
2722 pr_info("Read logical bytenr @%llu maps to (%s/%llu/%d)\n",
2723 bytenr, block_ctx.dev->name,
2724 block_ctx.dev_bytenr, mirror_num);
2725 }
2726 }
2727}
2728
2729static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev)
2730{
2731 return btrfsic_dev_state_hashtable_lookup(dev,
2732 &btrfsic_dev_state_hashtable);
2733}
2734
2735int btrfsic_submit_bh(int op, int op_flags, struct buffer_head *bh)
2736{
2737 struct btrfsic_dev_state *dev_state;
2738
2739 if (!btrfsic_is_initialized)
2740 return submit_bh(op, op_flags, bh);
2741
2742 mutex_lock(&btrfsic_mutex);
2743 /* since btrfsic_submit_bh() might also be called before
2744 * btrfsic_mount(), this might return NULL */
2745 dev_state = btrfsic_dev_state_lookup(bh->b_bdev->bd_dev);
2746
2747 /* Only called to write the superblock (incl. FLUSH/FUA) */
2748 if (NULL != dev_state &&
2749 (op == REQ_OP_WRITE) && bh->b_size > 0) {
2750 u64 dev_bytenr;
2751
2752 dev_bytenr = BTRFS_BDEV_BLOCKSIZE * bh->b_blocknr;
2753 if (dev_state->state->print_mask &
2754 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2755 pr_info("submit_bh(op=0x%x,0x%x, blocknr=%llu (bytenr %llu), size=%zu, data=%p, bdev=%p)\n",
2756 op, op_flags, (unsigned long long)bh->b_blocknr,
2757 dev_bytenr, bh->b_size, bh->b_data, bh->b_bdev);
2758 btrfsic_process_written_block(dev_state, dev_bytenr,
2759 &bh->b_data, 1, NULL,
2760 NULL, bh, op_flags);
2761 } else if (NULL != dev_state && (op_flags & REQ_PREFLUSH)) {
2762 if (dev_state->state->print_mask &
2763 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2764 pr_info("submit_bh(op=0x%x,0x%x FLUSH, bdev=%p)\n",
2765 op, op_flags, bh->b_bdev);
2766 if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2767 if ((dev_state->state->print_mask &
2768 (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2769 BTRFSIC_PRINT_MASK_VERBOSE)))
2770 pr_info("btrfsic_submit_bh(%s) with FLUSH but dummy block already in use (ignored)!\n",
2771 dev_state->name);
2772 } else {
2773 struct btrfsic_block *const block =
2774 &dev_state->dummy_block_for_bio_bh_flush;
2775
2776 block->is_iodone = 0;
2777 block->never_written = 0;
2778 block->iodone_w_error = 0;
2779 block->flush_gen = dev_state->last_flush_gen + 1;
2780 block->submit_bio_bh_rw = op_flags;
2781 block->orig_bio_bh_private = bh->b_private;
2782 block->orig_bio_bh_end_io.bh = bh->b_end_io;
2783 block->next_in_same_bio = NULL;
2784 bh->b_private = block;
2785 bh->b_end_io = btrfsic_bh_end_io;
2786 }
2787 }
2788 mutex_unlock(&btrfsic_mutex);
2789 return submit_bh(op, op_flags, bh);
2790}
2791
2792static void __btrfsic_submit_bio(struct bio *bio)
2793{
2794 struct btrfsic_dev_state *dev_state;
2795
2796 if (!btrfsic_is_initialized)
2797 return;
2798
2799 mutex_lock(&btrfsic_mutex);
2800 /* since btrfsic_submit_bio() is also called before
2801 * btrfsic_mount(), this might return NULL */
2802 dev_state = btrfsic_dev_state_lookup(bio_dev(bio) + bio->bi_partno);
2803 if (NULL != dev_state &&
2804 (bio_op(bio) == REQ_OP_WRITE) && bio_has_data(bio)) {
2805 unsigned int i = 0;
2806 u64 dev_bytenr;
2807 u64 cur_bytenr;
2808 struct bio_vec bvec;
2809 struct bvec_iter iter;
2810 int bio_is_patched;
2811 char **mapped_datav;
2812 unsigned int segs = bio_segments(bio);
2813
2814 dev_bytenr = 512 * bio->bi_iter.bi_sector;
2815 bio_is_patched = 0;
2816 if (dev_state->state->print_mask &
2817 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2818 pr_info("submit_bio(rw=%d,0x%x, bi_vcnt=%u, bi_sector=%llu (bytenr %llu), bi_disk=%p)\n",
2819 bio_op(bio), bio->bi_opf, segs,
2820 (unsigned long long)bio->bi_iter.bi_sector,
2821 dev_bytenr, bio->bi_disk);
2822
2823 mapped_datav = kmalloc_array(segs,
2824 sizeof(*mapped_datav), GFP_NOFS);
2825 if (!mapped_datav)
2826 goto leave;
2827 cur_bytenr = dev_bytenr;
2828
2829 bio_for_each_segment(bvec, bio, iter) {
2830 BUG_ON(bvec.bv_len != PAGE_SIZE);
2831 mapped_datav[i] = kmap(bvec.bv_page);
2832 i++;
2833
2834 if (dev_state->state->print_mask &
2835 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
2836 pr_info("#%u: bytenr=%llu, len=%u, offset=%u\n",
2837 i, cur_bytenr, bvec.bv_len, bvec.bv_offset);
2838 cur_bytenr += bvec.bv_len;
2839 }
2840 btrfsic_process_written_block(dev_state, dev_bytenr,
2841 mapped_datav, segs,
2842 bio, &bio_is_patched,
2843 NULL, bio->bi_opf);
2844 bio_for_each_segment(bvec, bio, iter)
2845 kunmap(bvec.bv_page);
2846 kfree(mapped_datav);
2847 } else if (NULL != dev_state && (bio->bi_opf & REQ_PREFLUSH)) {
2848 if (dev_state->state->print_mask &
2849 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2850 pr_info("submit_bio(rw=%d,0x%x FLUSH, disk=%p)\n",
2851 bio_op(bio), bio->bi_opf, bio->bi_disk);
2852 if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2853 if ((dev_state->state->print_mask &
2854 (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2855 BTRFSIC_PRINT_MASK_VERBOSE)))
2856 pr_info("btrfsic_submit_bio(%s) with FLUSH but dummy block already in use (ignored)!\n",
2857 dev_state->name);
2858 } else {
2859 struct btrfsic_block *const block =
2860 &dev_state->dummy_block_for_bio_bh_flush;
2861
2862 block->is_iodone = 0;
2863 block->never_written = 0;
2864 block->iodone_w_error = 0;
2865 block->flush_gen = dev_state->last_flush_gen + 1;
2866 block->submit_bio_bh_rw = bio->bi_opf;
2867 block->orig_bio_bh_private = bio->bi_private;
2868 block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2869 block->next_in_same_bio = NULL;
2870 bio->bi_private = block;
2871 bio->bi_end_io = btrfsic_bio_end_io;
2872 }
2873 }
2874leave:
2875 mutex_unlock(&btrfsic_mutex);
2876}
2877
2878void btrfsic_submit_bio(struct bio *bio)
2879{
2880 __btrfsic_submit_bio(bio);
2881 submit_bio(bio);
2882}
2883
2884int btrfsic_submit_bio_wait(struct bio *bio)
2885{
2886 __btrfsic_submit_bio(bio);
2887 return submit_bio_wait(bio);
2888}
2889
2890int btrfsic_mount(struct btrfs_fs_info *fs_info,
2891 struct btrfs_fs_devices *fs_devices,
2892 int including_extent_data, u32 print_mask)
2893{
2894 int ret;
2895 struct btrfsic_state *state;
2896 struct list_head *dev_head = &fs_devices->devices;
2897 struct btrfs_device *device;
2898
2899 if (!PAGE_ALIGNED(fs_info->nodesize)) {
2900 pr_info("btrfsic: cannot handle nodesize %d not being a multiple of PAGE_SIZE %ld!\n",
2901 fs_info->nodesize, PAGE_SIZE);
2902 return -1;
2903 }
2904 if (!PAGE_ALIGNED(fs_info->sectorsize)) {
2905 pr_info("btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_SIZE %ld!\n",
2906 fs_info->sectorsize, PAGE_SIZE);
2907 return -1;
2908 }
2909 state = kvzalloc(sizeof(*state), GFP_KERNEL);
2910 if (!state) {
2911 pr_info("btrfs check-integrity: allocation failed!\n");
2912 return -ENOMEM;
2913 }
2914
2915 if (!btrfsic_is_initialized) {
2916 mutex_init(&btrfsic_mutex);
2917 btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
2918 btrfsic_is_initialized = 1;
2919 }
2920 mutex_lock(&btrfsic_mutex);
2921 state->fs_info = fs_info;
2922 state->print_mask = print_mask;
2923 state->include_extent_data = including_extent_data;
2924 state->csum_size = 0;
2925 state->metablock_size = fs_info->nodesize;
2926 state->datablock_size = fs_info->sectorsize;
2927 INIT_LIST_HEAD(&state->all_blocks_list);
2928 btrfsic_block_hashtable_init(&state->block_hashtable);
2929 btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
2930 state->max_superblock_generation = 0;
2931 state->latest_superblock = NULL;
2932
2933 list_for_each_entry(device, dev_head, dev_list) {
2934 struct btrfsic_dev_state *ds;
2935 const char *p;
2936
2937 if (!device->bdev || !device->name)
2938 continue;
2939
2940 ds = btrfsic_dev_state_alloc();
2941 if (NULL == ds) {
2942 pr_info("btrfs check-integrity: kmalloc() failed!\n");
2943 mutex_unlock(&btrfsic_mutex);
2944 return -ENOMEM;
2945 }
2946 ds->bdev = device->bdev;
2947 ds->state = state;
2948 bdevname(ds->bdev, ds->name);
2949 ds->name[BDEVNAME_SIZE - 1] = '\0';
2950 p = kbasename(ds->name);
2951 strlcpy(ds->name, p, sizeof(ds->name));
2952 btrfsic_dev_state_hashtable_add(ds,
2953 &btrfsic_dev_state_hashtable);
2954 }
2955
2956 ret = btrfsic_process_superblock(state, fs_devices);
2957 if (0 != ret) {
2958 mutex_unlock(&btrfsic_mutex);
2959 btrfsic_unmount(fs_devices);
2960 return ret;
2961 }
2962
2963 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
2964 btrfsic_dump_database(state);
2965 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
2966 btrfsic_dump_tree(state);
2967
2968 mutex_unlock(&btrfsic_mutex);
2969 return 0;
2970}
2971
2972void btrfsic_unmount(struct btrfs_fs_devices *fs_devices)
2973{
2974 struct btrfsic_block *b_all, *tmp_all;
2975 struct btrfsic_state *state;
2976 struct list_head *dev_head = &fs_devices->devices;
2977 struct btrfs_device *device;
2978
2979 if (!btrfsic_is_initialized)
2980 return;
2981
2982 mutex_lock(&btrfsic_mutex);
2983
2984 state = NULL;
2985 list_for_each_entry(device, dev_head, dev_list) {
2986 struct btrfsic_dev_state *ds;
2987
2988 if (!device->bdev || !device->name)
2989 continue;
2990
2991 ds = btrfsic_dev_state_hashtable_lookup(
2992 device->bdev->bd_dev,
2993 &btrfsic_dev_state_hashtable);
2994 if (NULL != ds) {
2995 state = ds->state;
2996 btrfsic_dev_state_hashtable_remove(ds);
2997 btrfsic_dev_state_free(ds);
2998 }
2999 }
3000
3001 if (NULL == state) {
3002 pr_info("btrfsic: error, cannot find state information on umount!\n");
3003 mutex_unlock(&btrfsic_mutex);
3004 return;
3005 }
3006
3007 /*
3008 * Don't care about keeping the lists' state up to date,
3009 * just free all memory that was allocated dynamically.
3010 * Free the blocks and the block_links.
3011 */
3012 list_for_each_entry_safe(b_all, tmp_all, &state->all_blocks_list,
3013 all_blocks_node) {
3014 struct btrfsic_block_link *l, *tmp;
3015
3016 list_for_each_entry_safe(l, tmp, &b_all->ref_to_list,
3017 node_ref_to) {
3018 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
3019 btrfsic_print_rem_link(state, l);
3020
3021 l->ref_cnt--;
3022 if (0 == l->ref_cnt)
3023 btrfsic_block_link_free(l);
3024 }
3025
3026 if (b_all->is_iodone || b_all->never_written)
3027 btrfsic_block_free(b_all);
3028 else
3029 pr_info("btrfs: attempt to free %c-block @%llu (%s/%llu/%d) on umount which is not yet iodone!\n",
3030 btrfsic_get_block_type(state, b_all),
3031 b_all->logical_bytenr, b_all->dev_state->name,
3032 b_all->dev_bytenr, b_all->mirror_num);
3033 }
3034
3035 mutex_unlock(&btrfsic_mutex);
3036
3037 kvfree(state);
3038}