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