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