Loading...
Note: File does not exist in v6.8.
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}