1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) STRATO AG 2011. All rights reserved.
4 */
5
6 /*
7 * This module can be used to catch cases when the btrfs kernel
8 * code executes write requests to the disk that bring the file
9 * system in an inconsistent state. In such a state, a power-loss
10 * or kernel panic event would cause that the data on disk is
11 * lost or at least damaged.
12 *
13 * Code is added that examines all block write requests during
14 * runtime (including writes of the super block). Three rules
15 * are verified and an error is printed on violation of the
16 * rules:
17 * 1. It is not allowed to write a disk block which is
18 * currently referenced by the super block (either directly
19 * or indirectly).
20 * 2. When a super block is written, it is verified that all
21 * referenced (directly or indirectly) blocks fulfill the
22 * following requirements:
23 * 2a. All referenced blocks have either been present when
24 * the file system was mounted, (i.e., they have been
25 * referenced by the super block) or they have been
26 * written since then and the write completion callback
27 * was called and no write error was indicated and a
28 * FLUSH request to the device where these blocks are
29 * located was received and completed.
30 * 2b. All referenced blocks need to have a generation
31 * number which is equal to the parent's number.
32 *
33 * One issue that was found using this module was that the log
34 * tree on disk became temporarily corrupted because disk blocks
35 * that had been in use for the log tree had been freed and
36 * reused too early, while being referenced by the written super
37 * block.
38 *
39 * The search term in the kernel log that can be used to filter
40 * on the existence of detected integrity issues is
41 * "btrfs: attempt".
42 *
43 * The integrity check is enabled via mount options. These
44 * mount options are only supported if the integrity check
45 * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY.
46 *
47 * Example #1, apply integrity checks to all metadata:
48 * mount /dev/sdb1 /mnt -o check_int
49 *
50 * Example #2, apply integrity checks to all metadata and
51 * to data extents:
52 * mount /dev/sdb1 /mnt -o check_int_data
53 *
54 * Example #3, apply integrity checks to all metadata and dump
55 * the tree that the super block references to kernel messages
56 * each time after a super block was written:
57 * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263
58 *
59 * If the integrity check tool is included and activated in
60 * the mount options, plenty of kernel memory is used, and
61 * plenty of additional CPU cycles are spent. Enabling this
62 * functionality is not intended for normal use. In most
63 * cases, unless you are a btrfs developer who needs to verify
64 * the integrity of (super)-block write requests, do not
65 * enable the config option BTRFS_FS_CHECK_INTEGRITY to
66 * include and compile the integrity check tool.
67 *
68 * Expect millions of lines of information in the kernel log with an
69 * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the
70 * kernel config to at least 26 (which is 64MB). Usually the value is
71 * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be
72 * changed like this before LOG_BUF_SHIFT can be set to a high value:
73 * config LOG_BUF_SHIFT
74 * int "Kernel log buffer size (16 => 64KB, 17 => 128KB)"
75 * range 12 30
76 */
77
78 #include <linux/sched.h>
79 #include <linux/slab.h>
80 #include <linux/buffer_head.h>
81 #include <linux/mutex.h>
82 #include <linux/genhd.h>
83 #include <linux/blkdev.h>
84 #include <linux/mm.h>
85 #include <linux/string.h>
86 #include <linux/crc32c.h>
87 #include "ctree.h"
88 #include "disk-io.h"
89 #include "transaction.h"
90 #include "extent_io.h"
91 #include "volumes.h"
92 #include "print-tree.h"
93 #include "locking.h"
94 #include "check-integrity.h"
95 #include "rcu-string.h"
96 #include "compression.h"
97
98 #define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
99 #define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
100 #define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
101 #define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
102 #define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
103 #define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
104 #define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
105 #define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6) /* in characters,
106 * excluding " [...]" */
107 #define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
108
109 /*
110 * The definition of the bitmask fields for the print_mask.
111 * They are specified with the mount option check_integrity_print_mask.
112 */
113 #define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE 0x00000001
114 #define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION 0x00000002
115 #define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE 0x00000004
116 #define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE 0x00000008
117 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH 0x00000010
118 #define BTRFSIC_PRINT_MASK_END_IO_BIO_BH 0x00000020
119 #define BTRFSIC_PRINT_MASK_VERBOSE 0x00000040
120 #define BTRFSIC_PRINT_MASK_VERY_VERBOSE 0x00000080
121 #define BTRFSIC_PRINT_MASK_INITIAL_TREE 0x00000100
122 #define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES 0x00000200
123 #define BTRFSIC_PRINT_MASK_INITIAL_DATABASE 0x00000400
124 #define BTRFSIC_PRINT_MASK_NUM_COPIES 0x00000800
125 #define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS 0x00001000
126 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE 0x00002000
127
128 struct btrfsic_dev_state;
129 struct btrfsic_state;
130
131 struct btrfsic_block {
132 u32 magic_num; /* only used for debug purposes */
133 unsigned int is_metadata:1; /* if it is meta-data, not data-data */
134 unsigned int is_superblock:1; /* if it is one of the superblocks */
135 unsigned int is_iodone:1; /* if is done by lower subsystem */
136 unsigned int iodone_w_error:1; /* error was indicated to endio */
137 unsigned int never_written:1; /* block was added because it was
138 * referenced, not because it was
139 * written */
140 unsigned int mirror_num; /* large enough to hold
141 * BTRFS_SUPER_MIRROR_MAX */
142 struct btrfsic_dev_state *dev_state;
143 u64 dev_bytenr; /* key, physical byte num on disk */
144 u64 logical_bytenr; /* logical byte num on disk */
145 u64 generation;
146 struct btrfs_disk_key disk_key; /* extra info to print in case of
147 * issues, will not always be correct */
148 struct list_head collision_resolving_node; /* list node */
149 struct list_head all_blocks_node; /* list node */
150
151 /* the following two lists contain block_link items */
152 struct list_head ref_to_list; /* list */
153 struct list_head ref_from_list; /* list */
154 struct btrfsic_block *next_in_same_bio;
155 void *orig_bio_bh_private;
156 union {
157 bio_end_io_t *bio;
158 bh_end_io_t *bh;
159 } orig_bio_bh_end_io;
160 int submit_bio_bh_rw;
161 u64 flush_gen; /* only valid if !never_written */
162 };
163
164 /*
165 * Elements of this type are allocated dynamically and required because
166 * each block object can refer to and can be ref from multiple blocks.
167 * The key to lookup them in the hashtable is the dev_bytenr of
168 * the block ref to plus the one from the block referred from.
169 * The fact that they are searchable via a hashtable and that a
170 * ref_cnt is maintained is not required for the btrfs integrity
171 * check algorithm itself, it is only used to make the output more
172 * beautiful in case that an error is detected (an error is defined
173 * as a write operation to a block while that block is still referenced).
174 */
175 struct btrfsic_block_link {
176 u32 magic_num; /* only used for debug purposes */
177 u32 ref_cnt;
178 struct list_head node_ref_to; /* list node */
179 struct list_head node_ref_from; /* list node */
180 struct list_head collision_resolving_node; /* list node */
181 struct btrfsic_block *block_ref_to;
182 struct btrfsic_block *block_ref_from;
183 u64 parent_generation;
184 };
185
186 struct btrfsic_dev_state {
187 u32 magic_num; /* only used for debug purposes */
188 struct block_device *bdev;
189 struct btrfsic_state *state;
190 struct list_head collision_resolving_node; /* list node */
191 struct btrfsic_block dummy_block_for_bio_bh_flush;
192 u64 last_flush_gen;
193 char name[BDEVNAME_SIZE];
194 };
195
196 struct btrfsic_block_hashtable {
197 struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
198 };
199
200 struct btrfsic_block_link_hashtable {
201 struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
202 };
203
204 struct btrfsic_dev_state_hashtable {
205 struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
206 };
207
208 struct btrfsic_block_data_ctx {
209 u64 start; /* virtual bytenr */
210 u64 dev_bytenr; /* physical bytenr on device */
211 u32 len;
212 struct btrfsic_dev_state *dev;
213 char **datav;
214 struct page **pagev;
215 void *mem_to_free;
216 };
217
218 /* This structure is used to implement recursion without occupying
219 * any stack space, refer to btrfsic_process_metablock() */
220 struct btrfsic_stack_frame {
221 u32 magic;
222 u32 nr;
223 int error;
224 int i;
225 int limit_nesting;
226 int num_copies;
227 int mirror_num;
228 struct btrfsic_block *block;
229 struct btrfsic_block_data_ctx *block_ctx;
230 struct btrfsic_block *next_block;
231 struct btrfsic_block_data_ctx next_block_ctx;
232 struct btrfs_header *hdr;
233 struct btrfsic_stack_frame *prev;
234 };
235
236 /* Some state per mounted filesystem */
237 struct btrfsic_state {
238 u32 print_mask;
239 int include_extent_data;
240 int csum_size;
241 struct list_head all_blocks_list;
242 struct btrfsic_block_hashtable block_hashtable;
243 struct btrfsic_block_link_hashtable block_link_hashtable;
244 struct btrfs_fs_info *fs_info;
245 u64 max_superblock_generation;
246 struct btrfsic_block *latest_superblock;
247 u32 metablock_size;
248 u32 datablock_size;
249 };
250
251 static void btrfsic_block_init(struct btrfsic_block *b);
252 static struct btrfsic_block *btrfsic_block_alloc(void);
253 static void btrfsic_block_free(struct btrfsic_block *b);
254 static void btrfsic_block_link_init(struct btrfsic_block_link *n);
255 static struct btrfsic_block_link *btrfsic_block_link_alloc(void);
256 static void btrfsic_block_link_free(struct btrfsic_block_link *n);
257 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds);
258 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void);
259 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds);
260 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h);
261 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
262 struct btrfsic_block_hashtable *h);
263 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b);
264 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
265 struct block_device *bdev,
266 u64 dev_bytenr,
267 struct btrfsic_block_hashtable *h);
268 static void btrfsic_block_link_hashtable_init(
269 struct btrfsic_block_link_hashtable *h);
270 static void btrfsic_block_link_hashtable_add(
271 struct btrfsic_block_link *l,
272 struct btrfsic_block_link_hashtable *h);
273 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l);
274 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
275 struct block_device *bdev_ref_to,
276 u64 dev_bytenr_ref_to,
277 struct block_device *bdev_ref_from,
278 u64 dev_bytenr_ref_from,
279 struct btrfsic_block_link_hashtable *h);
280 static void btrfsic_dev_state_hashtable_init(
281 struct btrfsic_dev_state_hashtable *h);
282 static void btrfsic_dev_state_hashtable_add(
283 struct btrfsic_dev_state *ds,
284 struct btrfsic_dev_state_hashtable *h);
285 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds);
286 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev,
287 struct btrfsic_dev_state_hashtable *h);
288 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void);
289 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf);
290 static int btrfsic_process_superblock(struct btrfsic_state *state,
291 struct btrfs_fs_devices *fs_devices);
292 static int btrfsic_process_metablock(struct btrfsic_state *state,
293 struct btrfsic_block *block,
294 struct btrfsic_block_data_ctx *block_ctx,
295 int limit_nesting, int force_iodone_flag);
296 static void btrfsic_read_from_block_data(
297 struct btrfsic_block_data_ctx *block_ctx,
298 void *dst, u32 offset, size_t len);
299 static int btrfsic_create_link_to_next_block(
300 struct btrfsic_state *state,
301 struct btrfsic_block *block,
302 struct btrfsic_block_data_ctx
303 *block_ctx, u64 next_bytenr,
304 int limit_nesting,
305 struct btrfsic_block_data_ctx *next_block_ctx,
306 struct btrfsic_block **next_blockp,
307 int force_iodone_flag,
308 int *num_copiesp, int *mirror_nump,
309 struct btrfs_disk_key *disk_key,
310 u64 parent_generation);
311 static int btrfsic_handle_extent_data(struct btrfsic_state *state,
312 struct btrfsic_block *block,
313 struct btrfsic_block_data_ctx *block_ctx,
314 u32 item_offset, int force_iodone_flag);
315 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
316 struct btrfsic_block_data_ctx *block_ctx_out,
317 int mirror_num);
318 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
319 static int btrfsic_read_block(struct btrfsic_state *state,
320 struct btrfsic_block_data_ctx *block_ctx);
321 static void btrfsic_dump_database(struct btrfsic_state *state);
322 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
323 char **datav, unsigned int num_pages);
324 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
325 u64 dev_bytenr, char **mapped_datav,
326 unsigned int num_pages,
327 struct bio *bio, int *bio_is_patched,
328 struct buffer_head *bh,
329 int submit_bio_bh_rw);
330 static int btrfsic_process_written_superblock(
331 struct btrfsic_state *state,
332 struct btrfsic_block *const block,
333 struct btrfs_super_block *const super_hdr);
334 static void btrfsic_bio_end_io(struct bio *bp);
335 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate);
336 static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
337 const struct btrfsic_block *block,
338 int recursion_level);
339 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
340 struct btrfsic_block *const block,
341 int recursion_level);
342 static void btrfsic_print_add_link(const struct btrfsic_state *state,
343 const struct btrfsic_block_link *l);
344 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
345 const struct btrfsic_block_link *l);
346 static char btrfsic_get_block_type(const struct btrfsic_state *state,
347 const struct btrfsic_block *block);
348 static void btrfsic_dump_tree(const struct btrfsic_state *state);
349 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
350 const struct btrfsic_block *block,
351 int indent_level);
352 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
353 struct btrfsic_state *state,
354 struct btrfsic_block_data_ctx *next_block_ctx,
355 struct btrfsic_block *next_block,
356 struct btrfsic_block *from_block,
357 u64 parent_generation);
358 static struct btrfsic_block *btrfsic_block_lookup_or_add(
359 struct btrfsic_state *state,
360 struct btrfsic_block_data_ctx *block_ctx,
361 const char *additional_string,
362 int is_metadata,
363 int is_iodone,
364 int never_written,
365 int mirror_num,
366 int *was_created);
367 static int btrfsic_process_superblock_dev_mirror(
368 struct btrfsic_state *state,
369 struct btrfsic_dev_state *dev_state,
370 struct btrfs_device *device,
371 int superblock_mirror_num,
372 struct btrfsic_dev_state **selected_dev_state,
373 struct btrfs_super_block *selected_super);
374 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev);
375 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
376 u64 bytenr,
377 struct btrfsic_dev_state *dev_state,
378 u64 dev_bytenr);
379
380 static struct mutex btrfsic_mutex;
381 static int btrfsic_is_initialized;
382 static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
383
384
btrfsic_block_init(struct btrfsic_block * b)385 static void btrfsic_block_init(struct btrfsic_block *b)
386 {
387 b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
388 b->dev_state = NULL;
389 b->dev_bytenr = 0;
390 b->logical_bytenr = 0;
391 b->generation = BTRFSIC_GENERATION_UNKNOWN;
392 b->disk_key.objectid = 0;
393 b->disk_key.type = 0;
394 b->disk_key.offset = 0;
395 b->is_metadata = 0;
396 b->is_superblock = 0;
397 b->is_iodone = 0;
398 b->iodone_w_error = 0;
399 b->never_written = 0;
400 b->mirror_num = 0;
401 b->next_in_same_bio = NULL;
402 b->orig_bio_bh_private = NULL;
403 b->orig_bio_bh_end_io.bio = NULL;
404 INIT_LIST_HEAD(&b->collision_resolving_node);
405 INIT_LIST_HEAD(&b->all_blocks_node);
406 INIT_LIST_HEAD(&b->ref_to_list);
407 INIT_LIST_HEAD(&b->ref_from_list);
408 b->submit_bio_bh_rw = 0;
409 b->flush_gen = 0;
410 }
411
btrfsic_block_alloc(void)412 static struct btrfsic_block *btrfsic_block_alloc(void)
413 {
414 struct btrfsic_block *b;
415
416 b = kzalloc(sizeof(*b), GFP_NOFS);
417 if (NULL != b)
418 btrfsic_block_init(b);
419
420 return b;
421 }
422
btrfsic_block_free(struct btrfsic_block * b)423 static void btrfsic_block_free(struct btrfsic_block *b)
424 {
425 BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
426 kfree(b);
427 }
428
btrfsic_block_link_init(struct btrfsic_block_link * l)429 static void btrfsic_block_link_init(struct btrfsic_block_link *l)
430 {
431 l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
432 l->ref_cnt = 1;
433 INIT_LIST_HEAD(&l->node_ref_to);
434 INIT_LIST_HEAD(&l->node_ref_from);
435 INIT_LIST_HEAD(&l->collision_resolving_node);
436 l->block_ref_to = NULL;
437 l->block_ref_from = NULL;
438 }
439
btrfsic_block_link_alloc(void)440 static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
441 {
442 struct btrfsic_block_link *l;
443
444 l = kzalloc(sizeof(*l), GFP_NOFS);
445 if (NULL != l)
446 btrfsic_block_link_init(l);
447
448 return l;
449 }
450
btrfsic_block_link_free(struct btrfsic_block_link * l)451 static void btrfsic_block_link_free(struct btrfsic_block_link *l)
452 {
453 BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
454 kfree(l);
455 }
456
btrfsic_dev_state_init(struct btrfsic_dev_state * ds)457 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
458 {
459 ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
460 ds->bdev = NULL;
461 ds->state = NULL;
462 ds->name[0] = '\0';
463 INIT_LIST_HEAD(&ds->collision_resolving_node);
464 ds->last_flush_gen = 0;
465 btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
466 ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
467 ds->dummy_block_for_bio_bh_flush.dev_state = ds;
468 }
469
btrfsic_dev_state_alloc(void)470 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
471 {
472 struct btrfsic_dev_state *ds;
473
474 ds = kzalloc(sizeof(*ds), GFP_NOFS);
475 if (NULL != ds)
476 btrfsic_dev_state_init(ds);
477
478 return ds;
479 }
480
btrfsic_dev_state_free(struct btrfsic_dev_state * ds)481 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
482 {
483 BUG_ON(!(NULL == ds ||
484 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
485 kfree(ds);
486 }
487
btrfsic_block_hashtable_init(struct btrfsic_block_hashtable * h)488 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
489 {
490 int i;
491
492 for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
493 INIT_LIST_HEAD(h->table + i);
494 }
495
btrfsic_block_hashtable_add(struct btrfsic_block * b,struct btrfsic_block_hashtable * h)496 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
497 struct btrfsic_block_hashtable *h)
498 {
499 const unsigned int hashval =
500 (((unsigned int)(b->dev_bytenr >> 16)) ^
501 ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
502 (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
503
504 list_add(&b->collision_resolving_node, h->table + hashval);
505 }
506
btrfsic_block_hashtable_remove(struct btrfsic_block * b)507 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
508 {
509 list_del(&b->collision_resolving_node);
510 }
511
btrfsic_block_hashtable_lookup(struct block_device * bdev,u64 dev_bytenr,struct btrfsic_block_hashtable * h)512 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
513 struct block_device *bdev,
514 u64 dev_bytenr,
515 struct btrfsic_block_hashtable *h)
516 {
517 const unsigned int hashval =
518 (((unsigned int)(dev_bytenr >> 16)) ^
519 ((unsigned int)((uintptr_t)bdev))) &
520 (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
521 struct btrfsic_block *b;
522
523 list_for_each_entry(b, h->table + hashval, collision_resolving_node) {
524 if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
525 return b;
526 }
527
528 return NULL;
529 }
530
btrfsic_block_link_hashtable_init(struct btrfsic_block_link_hashtable * h)531 static void btrfsic_block_link_hashtable_init(
532 struct btrfsic_block_link_hashtable *h)
533 {
534 int i;
535
536 for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
537 INIT_LIST_HEAD(h->table + i);
538 }
539
btrfsic_block_link_hashtable_add(struct btrfsic_block_link * l,struct btrfsic_block_link_hashtable * h)540 static void btrfsic_block_link_hashtable_add(
541 struct btrfsic_block_link *l,
542 struct btrfsic_block_link_hashtable *h)
543 {
544 const unsigned int hashval =
545 (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
546 ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
547 ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
548 ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
549 & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
550
551 BUG_ON(NULL == l->block_ref_to);
552 BUG_ON(NULL == l->block_ref_from);
553 list_add(&l->collision_resolving_node, h->table + hashval);
554 }
555
btrfsic_block_link_hashtable_remove(struct btrfsic_block_link * l)556 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
557 {
558 list_del(&l->collision_resolving_node);
559 }
560
btrfsic_block_link_hashtable_lookup(struct block_device * bdev_ref_to,u64 dev_bytenr_ref_to,struct block_device * bdev_ref_from,u64 dev_bytenr_ref_from,struct btrfsic_block_link_hashtable * h)561 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
562 struct block_device *bdev_ref_to,
563 u64 dev_bytenr_ref_to,
564 struct block_device *bdev_ref_from,
565 u64 dev_bytenr_ref_from,
566 struct btrfsic_block_link_hashtable *h)
567 {
568 const unsigned int hashval =
569 (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
570 ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
571 ((unsigned int)((uintptr_t)bdev_ref_to)) ^
572 ((unsigned int)((uintptr_t)bdev_ref_from))) &
573 (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
574 struct btrfsic_block_link *l;
575
576 list_for_each_entry(l, h->table + hashval, collision_resolving_node) {
577 BUG_ON(NULL == l->block_ref_to);
578 BUG_ON(NULL == l->block_ref_from);
579 if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
580 l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
581 l->block_ref_from->dev_state->bdev == bdev_ref_from &&
582 l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
583 return l;
584 }
585
586 return NULL;
587 }
588
btrfsic_dev_state_hashtable_init(struct btrfsic_dev_state_hashtable * h)589 static void btrfsic_dev_state_hashtable_init(
590 struct btrfsic_dev_state_hashtable *h)
591 {
592 int i;
593
594 for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
595 INIT_LIST_HEAD(h->table + i);
596 }
597
btrfsic_dev_state_hashtable_add(struct btrfsic_dev_state * ds,struct btrfsic_dev_state_hashtable * h)598 static void btrfsic_dev_state_hashtable_add(
599 struct btrfsic_dev_state *ds,
600 struct btrfsic_dev_state_hashtable *h)
601 {
602 const unsigned int hashval =
603 (((unsigned int)((uintptr_t)ds->bdev->bd_dev)) &
604 (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
605
606 list_add(&ds->collision_resolving_node, h->table + hashval);
607 }
608
btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state * ds)609 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
610 {
611 list_del(&ds->collision_resolving_node);
612 }
613
btrfsic_dev_state_hashtable_lookup(dev_t dev,struct btrfsic_dev_state_hashtable * h)614 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev,
615 struct btrfsic_dev_state_hashtable *h)
616 {
617 const unsigned int hashval =
618 dev & (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1);
619 struct btrfsic_dev_state *ds;
620
621 list_for_each_entry(ds, h->table + hashval, collision_resolving_node) {
622 if (ds->bdev->bd_dev == dev)
623 return ds;
624 }
625
626 return NULL;
627 }
628
btrfsic_process_superblock(struct btrfsic_state * state,struct btrfs_fs_devices * fs_devices)629 static int btrfsic_process_superblock(struct btrfsic_state *state,
630 struct btrfs_fs_devices *fs_devices)
631 {
632 struct btrfs_super_block *selected_super;
633 struct list_head *dev_head = &fs_devices->devices;
634 struct btrfs_device *device;
635 struct btrfsic_dev_state *selected_dev_state = NULL;
636 int ret = 0;
637 int pass;
638
639 BUG_ON(NULL == state);
640 selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
641 if (NULL == selected_super) {
642 pr_info("btrfsic: error, kmalloc failed!\n");
643 return -ENOMEM;
644 }
645
646 list_for_each_entry(device, dev_head, dev_list) {
647 int i;
648 struct btrfsic_dev_state *dev_state;
649
650 if (!device->bdev || !device->name)
651 continue;
652
653 dev_state = btrfsic_dev_state_lookup(device->bdev->bd_dev);
654 BUG_ON(NULL == dev_state);
655 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
656 ret = btrfsic_process_superblock_dev_mirror(
657 state, dev_state, device, i,
658 &selected_dev_state, selected_super);
659 if (0 != ret && 0 == i) {
660 kfree(selected_super);
661 return ret;
662 }
663 }
664 }
665
666 if (NULL == state->latest_superblock) {
667 pr_info("btrfsic: no superblock found!\n");
668 kfree(selected_super);
669 return -1;
670 }
671
672 state->csum_size = btrfs_super_csum_size(selected_super);
673
674 for (pass = 0; pass < 3; pass++) {
675 int num_copies;
676 int mirror_num;
677 u64 next_bytenr;
678
679 switch (pass) {
680 case 0:
681 next_bytenr = btrfs_super_root(selected_super);
682 if (state->print_mask &
683 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
684 pr_info("root@%llu\n", next_bytenr);
685 break;
686 case 1:
687 next_bytenr = btrfs_super_chunk_root(selected_super);
688 if (state->print_mask &
689 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
690 pr_info("chunk@%llu\n", next_bytenr);
691 break;
692 case 2:
693 next_bytenr = btrfs_super_log_root(selected_super);
694 if (0 == next_bytenr)
695 continue;
696 if (state->print_mask &
697 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
698 pr_info("log@%llu\n", next_bytenr);
699 break;
700 }
701
702 num_copies = btrfs_num_copies(state->fs_info, next_bytenr,
703 state->metablock_size);
704 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
705 pr_info("num_copies(log_bytenr=%llu) = %d\n",
706 next_bytenr, num_copies);
707
708 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
709 struct btrfsic_block *next_block;
710 struct btrfsic_block_data_ctx tmp_next_block_ctx;
711 struct btrfsic_block_link *l;
712
713 ret = btrfsic_map_block(state, next_bytenr,
714 state->metablock_size,
715 &tmp_next_block_ctx,
716 mirror_num);
717 if (ret) {
718 pr_info("btrfsic: btrfsic_map_block(root @%llu, mirror %d) failed!\n",
719 next_bytenr, mirror_num);
720 kfree(selected_super);
721 return -1;
722 }
723
724 next_block = btrfsic_block_hashtable_lookup(
725 tmp_next_block_ctx.dev->bdev,
726 tmp_next_block_ctx.dev_bytenr,
727 &state->block_hashtable);
728 BUG_ON(NULL == next_block);
729
730 l = btrfsic_block_link_hashtable_lookup(
731 tmp_next_block_ctx.dev->bdev,
732 tmp_next_block_ctx.dev_bytenr,
733 state->latest_superblock->dev_state->
734 bdev,
735 state->latest_superblock->dev_bytenr,
736 &state->block_link_hashtable);
737 BUG_ON(NULL == l);
738
739 ret = btrfsic_read_block(state, &tmp_next_block_ctx);
740 if (ret < (int)PAGE_SIZE) {
741 pr_info("btrfsic: read @logical %llu failed!\n",
742 tmp_next_block_ctx.start);
743 btrfsic_release_block_ctx(&tmp_next_block_ctx);
744 kfree(selected_super);
745 return -1;
746 }
747
748 ret = btrfsic_process_metablock(state,
749 next_block,
750 &tmp_next_block_ctx,
751 BTRFS_MAX_LEVEL + 3, 1);
752 btrfsic_release_block_ctx(&tmp_next_block_ctx);
753 }
754 }
755
756 kfree(selected_super);
757 return ret;
758 }
759
btrfsic_process_superblock_dev_mirror(struct btrfsic_state * state,struct btrfsic_dev_state * dev_state,struct btrfs_device * device,int superblock_mirror_num,struct btrfsic_dev_state ** selected_dev_state,struct btrfs_super_block * selected_super)760 static int btrfsic_process_superblock_dev_mirror(
761 struct btrfsic_state *state,
762 struct btrfsic_dev_state *dev_state,
763 struct btrfs_device *device,
764 int superblock_mirror_num,
765 struct btrfsic_dev_state **selected_dev_state,
766 struct btrfs_super_block *selected_super)
767 {
768 struct btrfs_fs_info *fs_info = state->fs_info;
769 struct btrfs_super_block *super_tmp;
770 u64 dev_bytenr;
771 struct buffer_head *bh;
772 struct btrfsic_block *superblock_tmp;
773 int pass;
774 struct block_device *const superblock_bdev = device->bdev;
775
776 /* super block bytenr is always the unmapped device bytenr */
777 dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
778 if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->commit_total_bytes)
779 return -1;
780 bh = __bread(superblock_bdev, dev_bytenr / BTRFS_BDEV_BLOCKSIZE,
781 BTRFS_SUPER_INFO_SIZE);
782 if (NULL == bh)
783 return -1;
784 super_tmp = (struct btrfs_super_block *)
785 (bh->b_data + (dev_bytenr & (BTRFS_BDEV_BLOCKSIZE - 1)));
786
787 if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
788 btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
789 memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
790 btrfs_super_nodesize(super_tmp) != state->metablock_size ||
791 btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
792 brelse(bh);
793 return 0;
794 }
795
796 superblock_tmp =
797 btrfsic_block_hashtable_lookup(superblock_bdev,
798 dev_bytenr,
799 &state->block_hashtable);
800 if (NULL == superblock_tmp) {
801 superblock_tmp = btrfsic_block_alloc();
802 if (NULL == superblock_tmp) {
803 pr_info("btrfsic: error, kmalloc failed!\n");
804 brelse(bh);
805 return -1;
806 }
807 /* for superblock, only the dev_bytenr makes sense */
808 superblock_tmp->dev_bytenr = dev_bytenr;
809 superblock_tmp->dev_state = dev_state;
810 superblock_tmp->logical_bytenr = dev_bytenr;
811 superblock_tmp->generation = btrfs_super_generation(super_tmp);
812 superblock_tmp->is_metadata = 1;
813 superblock_tmp->is_superblock = 1;
814 superblock_tmp->is_iodone = 1;
815 superblock_tmp->never_written = 0;
816 superblock_tmp->mirror_num = 1 + superblock_mirror_num;
817 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
818 btrfs_info_in_rcu(fs_info,
819 "new initial S-block (bdev %p, %s) @%llu (%s/%llu/%d)",
820 superblock_bdev,
821 rcu_str_deref(device->name), dev_bytenr,
822 dev_state->name, dev_bytenr,
823 superblock_mirror_num);
824 list_add(&superblock_tmp->all_blocks_node,
825 &state->all_blocks_list);
826 btrfsic_block_hashtable_add(superblock_tmp,
827 &state->block_hashtable);
828 }
829
830 /* select the one with the highest generation field */
831 if (btrfs_super_generation(super_tmp) >
832 state->max_superblock_generation ||
833 0 == state->max_superblock_generation) {
834 memcpy(selected_super, super_tmp, sizeof(*selected_super));
835 *selected_dev_state = dev_state;
836 state->max_superblock_generation =
837 btrfs_super_generation(super_tmp);
838 state->latest_superblock = superblock_tmp;
839 }
840
841 for (pass = 0; pass < 3; pass++) {
842 u64 next_bytenr;
843 int num_copies;
844 int mirror_num;
845 const char *additional_string = NULL;
846 struct btrfs_disk_key tmp_disk_key;
847
848 tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
849 tmp_disk_key.offset = 0;
850 switch (pass) {
851 case 0:
852 btrfs_set_disk_key_objectid(&tmp_disk_key,
853 BTRFS_ROOT_TREE_OBJECTID);
854 additional_string = "initial root ";
855 next_bytenr = btrfs_super_root(super_tmp);
856 break;
857 case 1:
858 btrfs_set_disk_key_objectid(&tmp_disk_key,
859 BTRFS_CHUNK_TREE_OBJECTID);
860 additional_string = "initial chunk ";
861 next_bytenr = btrfs_super_chunk_root(super_tmp);
862 break;
863 case 2:
864 btrfs_set_disk_key_objectid(&tmp_disk_key,
865 BTRFS_TREE_LOG_OBJECTID);
866 additional_string = "initial log ";
867 next_bytenr = btrfs_super_log_root(super_tmp);
868 if (0 == next_bytenr)
869 continue;
870 break;
871 }
872
873 num_copies = btrfs_num_copies(fs_info, next_bytenr,
874 state->metablock_size);
875 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
876 pr_info("num_copies(log_bytenr=%llu) = %d\n",
877 next_bytenr, num_copies);
878 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
879 struct btrfsic_block *next_block;
880 struct btrfsic_block_data_ctx tmp_next_block_ctx;
881 struct btrfsic_block_link *l;
882
883 if (btrfsic_map_block(state, next_bytenr,
884 state->metablock_size,
885 &tmp_next_block_ctx,
886 mirror_num)) {
887 pr_info("btrfsic: btrfsic_map_block(bytenr @%llu, mirror %d) failed!\n",
888 next_bytenr, mirror_num);
889 brelse(bh);
890 return -1;
891 }
892
893 next_block = btrfsic_block_lookup_or_add(
894 state, &tmp_next_block_ctx,
895 additional_string, 1, 1, 0,
896 mirror_num, NULL);
897 if (NULL == next_block) {
898 btrfsic_release_block_ctx(&tmp_next_block_ctx);
899 brelse(bh);
900 return -1;
901 }
902
903 next_block->disk_key = tmp_disk_key;
904 next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
905 l = btrfsic_block_link_lookup_or_add(
906 state, &tmp_next_block_ctx,
907 next_block, superblock_tmp,
908 BTRFSIC_GENERATION_UNKNOWN);
909 btrfsic_release_block_ctx(&tmp_next_block_ctx);
910 if (NULL == l) {
911 brelse(bh);
912 return -1;
913 }
914 }
915 }
916 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
917 btrfsic_dump_tree_sub(state, superblock_tmp, 0);
918
919 brelse(bh);
920 return 0;
921 }
922
btrfsic_stack_frame_alloc(void)923 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
924 {
925 struct btrfsic_stack_frame *sf;
926
927 sf = kzalloc(sizeof(*sf), GFP_NOFS);
928 if (NULL == sf)
929 pr_info("btrfsic: alloc memory failed!\n");
930 else
931 sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
932 return sf;
933 }
934
btrfsic_stack_frame_free(struct btrfsic_stack_frame * sf)935 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
936 {
937 BUG_ON(!(NULL == sf ||
938 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
939 kfree(sf);
940 }
941
btrfsic_process_metablock(struct btrfsic_state * state,struct btrfsic_block * const first_block,struct btrfsic_block_data_ctx * const first_block_ctx,int first_limit_nesting,int force_iodone_flag)942 static int btrfsic_process_metablock(
943 struct btrfsic_state *state,
944 struct btrfsic_block *const first_block,
945 struct btrfsic_block_data_ctx *const first_block_ctx,
946 int first_limit_nesting, int force_iodone_flag)
947 {
948 struct btrfsic_stack_frame initial_stack_frame = { 0 };
949 struct btrfsic_stack_frame *sf;
950 struct btrfsic_stack_frame *next_stack;
951 struct btrfs_header *const first_hdr =
952 (struct btrfs_header *)first_block_ctx->datav[0];
953
954 BUG_ON(!first_hdr);
955 sf = &initial_stack_frame;
956 sf->error = 0;
957 sf->i = -1;
958 sf->limit_nesting = first_limit_nesting;
959 sf->block = first_block;
960 sf->block_ctx = first_block_ctx;
961 sf->next_block = NULL;
962 sf->hdr = first_hdr;
963 sf->prev = NULL;
964
965 continue_with_new_stack_frame:
966 sf->block->generation = le64_to_cpu(sf->hdr->generation);
967 if (0 == sf->hdr->level) {
968 struct btrfs_leaf *const leafhdr =
969 (struct btrfs_leaf *)sf->hdr;
970
971 if (-1 == sf->i) {
972 sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
973
974 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
975 pr_info("leaf %llu items %d generation %llu owner %llu\n",
976 sf->block_ctx->start, sf->nr,
977 btrfs_stack_header_generation(
978 &leafhdr->header),
979 btrfs_stack_header_owner(
980 &leafhdr->header));
981 }
982
983 continue_with_current_leaf_stack_frame:
984 if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
985 sf->i++;
986 sf->num_copies = 0;
987 }
988
989 if (sf->i < sf->nr) {
990 struct btrfs_item disk_item;
991 u32 disk_item_offset =
992 (uintptr_t)(leafhdr->items + sf->i) -
993 (uintptr_t)leafhdr;
994 struct btrfs_disk_key *disk_key;
995 u8 type;
996 u32 item_offset;
997 u32 item_size;
998
999 if (disk_item_offset + sizeof(struct btrfs_item) >
1000 sf->block_ctx->len) {
1001 leaf_item_out_of_bounce_error:
1002 pr_info("btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
1003 sf->block_ctx->start,
1004 sf->block_ctx->dev->name);
1005 goto one_stack_frame_backwards;
1006 }
1007 btrfsic_read_from_block_data(sf->block_ctx,
1008 &disk_item,
1009 disk_item_offset,
1010 sizeof(struct btrfs_item));
1011 item_offset = btrfs_stack_item_offset(&disk_item);
1012 item_size = btrfs_stack_item_size(&disk_item);
1013 disk_key = &disk_item.key;
1014 type = btrfs_disk_key_type(disk_key);
1015
1016 if (BTRFS_ROOT_ITEM_KEY == type) {
1017 struct btrfs_root_item root_item;
1018 u32 root_item_offset;
1019 u64 next_bytenr;
1020
1021 root_item_offset = item_offset +
1022 offsetof(struct btrfs_leaf, items);
1023 if (root_item_offset + item_size >
1024 sf->block_ctx->len)
1025 goto leaf_item_out_of_bounce_error;
1026 btrfsic_read_from_block_data(
1027 sf->block_ctx, &root_item,
1028 root_item_offset,
1029 item_size);
1030 next_bytenr = btrfs_root_bytenr(&root_item);
1031
1032 sf->error =
1033 btrfsic_create_link_to_next_block(
1034 state,
1035 sf->block,
1036 sf->block_ctx,
1037 next_bytenr,
1038 sf->limit_nesting,
1039 &sf->next_block_ctx,
1040 &sf->next_block,
1041 force_iodone_flag,
1042 &sf->num_copies,
1043 &sf->mirror_num,
1044 disk_key,
1045 btrfs_root_generation(
1046 &root_item));
1047 if (sf->error)
1048 goto one_stack_frame_backwards;
1049
1050 if (NULL != sf->next_block) {
1051 struct btrfs_header *const next_hdr =
1052 (struct btrfs_header *)
1053 sf->next_block_ctx.datav[0];
1054
1055 next_stack =
1056 btrfsic_stack_frame_alloc();
1057 if (NULL == next_stack) {
1058 sf->error = -1;
1059 btrfsic_release_block_ctx(
1060 &sf->
1061 next_block_ctx);
1062 goto one_stack_frame_backwards;
1063 }
1064
1065 next_stack->i = -1;
1066 next_stack->block = sf->next_block;
1067 next_stack->block_ctx =
1068 &sf->next_block_ctx;
1069 next_stack->next_block = NULL;
1070 next_stack->hdr = next_hdr;
1071 next_stack->limit_nesting =
1072 sf->limit_nesting - 1;
1073 next_stack->prev = sf;
1074 sf = next_stack;
1075 goto continue_with_new_stack_frame;
1076 }
1077 } else if (BTRFS_EXTENT_DATA_KEY == type &&
1078 state->include_extent_data) {
1079 sf->error = btrfsic_handle_extent_data(
1080 state,
1081 sf->block,
1082 sf->block_ctx,
1083 item_offset,
1084 force_iodone_flag);
1085 if (sf->error)
1086 goto one_stack_frame_backwards;
1087 }
1088
1089 goto continue_with_current_leaf_stack_frame;
1090 }
1091 } else {
1092 struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1093
1094 if (-1 == sf->i) {
1095 sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1096
1097 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1098 pr_info("node %llu level %d items %d generation %llu owner %llu\n",
1099 sf->block_ctx->start,
1100 nodehdr->header.level, sf->nr,
1101 btrfs_stack_header_generation(
1102 &nodehdr->header),
1103 btrfs_stack_header_owner(
1104 &nodehdr->header));
1105 }
1106
1107 continue_with_current_node_stack_frame:
1108 if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1109 sf->i++;
1110 sf->num_copies = 0;
1111 }
1112
1113 if (sf->i < sf->nr) {
1114 struct btrfs_key_ptr key_ptr;
1115 u32 key_ptr_offset;
1116 u64 next_bytenr;
1117
1118 key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1119 (uintptr_t)nodehdr;
1120 if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1121 sf->block_ctx->len) {
1122 pr_info("btrfsic: node item out of bounce at logical %llu, dev %s\n",
1123 sf->block_ctx->start,
1124 sf->block_ctx->dev->name);
1125 goto one_stack_frame_backwards;
1126 }
1127 btrfsic_read_from_block_data(
1128 sf->block_ctx, &key_ptr, key_ptr_offset,
1129 sizeof(struct btrfs_key_ptr));
1130 next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1131
1132 sf->error = btrfsic_create_link_to_next_block(
1133 state,
1134 sf->block,
1135 sf->block_ctx,
1136 next_bytenr,
1137 sf->limit_nesting,
1138 &sf->next_block_ctx,
1139 &sf->next_block,
1140 force_iodone_flag,
1141 &sf->num_copies,
1142 &sf->mirror_num,
1143 &key_ptr.key,
1144 btrfs_stack_key_generation(&key_ptr));
1145 if (sf->error)
1146 goto one_stack_frame_backwards;
1147
1148 if (NULL != sf->next_block) {
1149 struct btrfs_header *const next_hdr =
1150 (struct btrfs_header *)
1151 sf->next_block_ctx.datav[0];
1152
1153 next_stack = btrfsic_stack_frame_alloc();
1154 if (NULL == next_stack) {
1155 sf->error = -1;
1156 goto one_stack_frame_backwards;
1157 }
1158
1159 next_stack->i = -1;
1160 next_stack->block = sf->next_block;
1161 next_stack->block_ctx = &sf->next_block_ctx;
1162 next_stack->next_block = NULL;
1163 next_stack->hdr = next_hdr;
1164 next_stack->limit_nesting =
1165 sf->limit_nesting - 1;
1166 next_stack->prev = sf;
1167 sf = next_stack;
1168 goto continue_with_new_stack_frame;
1169 }
1170
1171 goto continue_with_current_node_stack_frame;
1172 }
1173 }
1174
1175 one_stack_frame_backwards:
1176 if (NULL != sf->prev) {
1177 struct btrfsic_stack_frame *const prev = sf->prev;
1178
1179 /* the one for the initial block is freed in the caller */
1180 btrfsic_release_block_ctx(sf->block_ctx);
1181
1182 if (sf->error) {
1183 prev->error = sf->error;
1184 btrfsic_stack_frame_free(sf);
1185 sf = prev;
1186 goto one_stack_frame_backwards;
1187 }
1188
1189 btrfsic_stack_frame_free(sf);
1190 sf = prev;
1191 goto continue_with_new_stack_frame;
1192 } else {
1193 BUG_ON(&initial_stack_frame != sf);
1194 }
1195
1196 return sf->error;
1197 }
1198
btrfsic_read_from_block_data(struct btrfsic_block_data_ctx * block_ctx,void * dstv,u32 offset,size_t len)1199 static void btrfsic_read_from_block_data(
1200 struct btrfsic_block_data_ctx *block_ctx,
1201 void *dstv, u32 offset, size_t len)
1202 {
1203 size_t cur;
1204 size_t offset_in_page;
1205 char *kaddr;
1206 char *dst = (char *)dstv;
1207 size_t start_offset = block_ctx->start & ((u64)PAGE_SIZE - 1);
1208 unsigned long i = (start_offset + offset) >> PAGE_SHIFT;
1209
1210 WARN_ON(offset + len > block_ctx->len);
1211 offset_in_page = (start_offset + offset) & (PAGE_SIZE - 1);
1212
1213 while (len > 0) {
1214 cur = min(len, ((size_t)PAGE_SIZE - offset_in_page));
1215 BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_SIZE));
1216 kaddr = block_ctx->datav[i];
1217 memcpy(dst, kaddr + offset_in_page, cur);
1218
1219 dst += cur;
1220 len -= cur;
1221 offset_in_page = 0;
1222 i++;
1223 }
1224 }
1225
btrfsic_create_link_to_next_block(struct btrfsic_state * state,struct btrfsic_block * block,struct btrfsic_block_data_ctx * block_ctx,u64 next_bytenr,int limit_nesting,struct btrfsic_block_data_ctx * next_block_ctx,struct btrfsic_block ** next_blockp,int force_iodone_flag,int * num_copiesp,int * mirror_nump,struct btrfs_disk_key * disk_key,u64 parent_generation)1226 static int btrfsic_create_link_to_next_block(
1227 struct btrfsic_state *state,
1228 struct btrfsic_block *block,
1229 struct btrfsic_block_data_ctx *block_ctx,
1230 u64 next_bytenr,
1231 int limit_nesting,
1232 struct btrfsic_block_data_ctx *next_block_ctx,
1233 struct btrfsic_block **next_blockp,
1234 int force_iodone_flag,
1235 int *num_copiesp, int *mirror_nump,
1236 struct btrfs_disk_key *disk_key,
1237 u64 parent_generation)
1238 {
1239 struct btrfs_fs_info *fs_info = state->fs_info;
1240 struct btrfsic_block *next_block = NULL;
1241 int ret;
1242 struct btrfsic_block_link *l;
1243 int did_alloc_block_link;
1244 int block_was_created;
1245
1246 *next_blockp = NULL;
1247 if (0 == *num_copiesp) {
1248 *num_copiesp = btrfs_num_copies(fs_info, next_bytenr,
1249 state->metablock_size);
1250 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1251 pr_info("num_copies(log_bytenr=%llu) = %d\n",
1252 next_bytenr, *num_copiesp);
1253 *mirror_nump = 1;
1254 }
1255
1256 if (*mirror_nump > *num_copiesp)
1257 return 0;
1258
1259 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1260 pr_info("btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1261 *mirror_nump);
1262 ret = btrfsic_map_block(state, next_bytenr,
1263 state->metablock_size,
1264 next_block_ctx, *mirror_nump);
1265 if (ret) {
1266 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1267 next_bytenr, *mirror_nump);
1268 btrfsic_release_block_ctx(next_block_ctx);
1269 *next_blockp = NULL;
1270 return -1;
1271 }
1272
1273 next_block = btrfsic_block_lookup_or_add(state,
1274 next_block_ctx, "referenced ",
1275 1, force_iodone_flag,
1276 !force_iodone_flag,
1277 *mirror_nump,
1278 &block_was_created);
1279 if (NULL == next_block) {
1280 btrfsic_release_block_ctx(next_block_ctx);
1281 *next_blockp = NULL;
1282 return -1;
1283 }
1284 if (block_was_created) {
1285 l = NULL;
1286 next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1287 } else {
1288 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1289 if (next_block->logical_bytenr != next_bytenr &&
1290 !(!next_block->is_metadata &&
1291 0 == next_block->logical_bytenr))
1292 pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1293 next_bytenr, next_block_ctx->dev->name,
1294 next_block_ctx->dev_bytenr, *mirror_nump,
1295 btrfsic_get_block_type(state,
1296 next_block),
1297 next_block->logical_bytenr);
1298 else
1299 pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1300 next_bytenr, next_block_ctx->dev->name,
1301 next_block_ctx->dev_bytenr, *mirror_nump,
1302 btrfsic_get_block_type(state,
1303 next_block));
1304 }
1305 next_block->logical_bytenr = next_bytenr;
1306
1307 next_block->mirror_num = *mirror_nump;
1308 l = btrfsic_block_link_hashtable_lookup(
1309 next_block_ctx->dev->bdev,
1310 next_block_ctx->dev_bytenr,
1311 block_ctx->dev->bdev,
1312 block_ctx->dev_bytenr,
1313 &state->block_link_hashtable);
1314 }
1315
1316 next_block->disk_key = *disk_key;
1317 if (NULL == l) {
1318 l = btrfsic_block_link_alloc();
1319 if (NULL == l) {
1320 pr_info("btrfsic: error, kmalloc failed!\n");
1321 btrfsic_release_block_ctx(next_block_ctx);
1322 *next_blockp = NULL;
1323 return -1;
1324 }
1325
1326 did_alloc_block_link = 1;
1327 l->block_ref_to = next_block;
1328 l->block_ref_from = block;
1329 l->ref_cnt = 1;
1330 l->parent_generation = parent_generation;
1331
1332 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1333 btrfsic_print_add_link(state, l);
1334
1335 list_add(&l->node_ref_to, &block->ref_to_list);
1336 list_add(&l->node_ref_from, &next_block->ref_from_list);
1337
1338 btrfsic_block_link_hashtable_add(l,
1339 &state->block_link_hashtable);
1340 } else {
1341 did_alloc_block_link = 0;
1342 if (0 == limit_nesting) {
1343 l->ref_cnt++;
1344 l->parent_generation = parent_generation;
1345 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1346 btrfsic_print_add_link(state, l);
1347 }
1348 }
1349
1350 if (limit_nesting > 0 && did_alloc_block_link) {
1351 ret = btrfsic_read_block(state, next_block_ctx);
1352 if (ret < (int)next_block_ctx->len) {
1353 pr_info("btrfsic: read block @logical %llu failed!\n",
1354 next_bytenr);
1355 btrfsic_release_block_ctx(next_block_ctx);
1356 *next_blockp = NULL;
1357 return -1;
1358 }
1359
1360 *next_blockp = next_block;
1361 } else {
1362 *next_blockp = NULL;
1363 }
1364 (*mirror_nump)++;
1365
1366 return 0;
1367 }
1368
btrfsic_handle_extent_data(struct btrfsic_state * state,struct btrfsic_block * block,struct btrfsic_block_data_ctx * block_ctx,u32 item_offset,int force_iodone_flag)1369 static int btrfsic_handle_extent_data(
1370 struct btrfsic_state *state,
1371 struct btrfsic_block *block,
1372 struct btrfsic_block_data_ctx *block_ctx,
1373 u32 item_offset, int force_iodone_flag)
1374 {
1375 struct btrfs_fs_info *fs_info = state->fs_info;
1376 struct btrfs_file_extent_item file_extent_item;
1377 u64 file_extent_item_offset;
1378 u64 next_bytenr;
1379 u64 num_bytes;
1380 u64 generation;
1381 struct btrfsic_block_link *l;
1382 int ret;
1383
1384 file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1385 item_offset;
1386 if (file_extent_item_offset +
1387 offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1388 block_ctx->len) {
1389 pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1390 block_ctx->start, block_ctx->dev->name);
1391 return -1;
1392 }
1393
1394 btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1395 file_extent_item_offset,
1396 offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1397 if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1398 btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1399 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1400 pr_info("extent_data: type %u, disk_bytenr = %llu\n",
1401 file_extent_item.type,
1402 btrfs_stack_file_extent_disk_bytenr(
1403 &file_extent_item));
1404 return 0;
1405 }
1406
1407 if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1408 block_ctx->len) {
1409 pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1410 block_ctx->start, block_ctx->dev->name);
1411 return -1;
1412 }
1413 btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1414 file_extent_item_offset,
1415 sizeof(struct btrfs_file_extent_item));
1416 next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item);
1417 if (btrfs_stack_file_extent_compression(&file_extent_item) ==
1418 BTRFS_COMPRESS_NONE) {
1419 next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item);
1420 num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1421 } else {
1422 num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item);
1423 }
1424 generation = btrfs_stack_file_extent_generation(&file_extent_item);
1425
1426 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1427 pr_info("extent_data: type %u, disk_bytenr = %llu, offset = %llu, num_bytes = %llu\n",
1428 file_extent_item.type,
1429 btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1430 btrfs_stack_file_extent_offset(&file_extent_item),
1431 num_bytes);
1432 while (num_bytes > 0) {
1433 u32 chunk_len;
1434 int num_copies;
1435 int mirror_num;
1436
1437 if (num_bytes > state->datablock_size)
1438 chunk_len = state->datablock_size;
1439 else
1440 chunk_len = num_bytes;
1441
1442 num_copies = btrfs_num_copies(fs_info, next_bytenr,
1443 state->datablock_size);
1444 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1445 pr_info("num_copies(log_bytenr=%llu) = %d\n",
1446 next_bytenr, num_copies);
1447 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1448 struct btrfsic_block_data_ctx next_block_ctx;
1449 struct btrfsic_block *next_block;
1450 int block_was_created;
1451
1452 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1453 pr_info("btrfsic_handle_extent_data(mirror_num=%d)\n",
1454 mirror_num);
1455 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1456 pr_info("\tdisk_bytenr = %llu, num_bytes %u\n",
1457 next_bytenr, chunk_len);
1458 ret = btrfsic_map_block(state, next_bytenr,
1459 chunk_len, &next_block_ctx,
1460 mirror_num);
1461 if (ret) {
1462 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1463 next_bytenr, mirror_num);
1464 return -1;
1465 }
1466
1467 next_block = btrfsic_block_lookup_or_add(
1468 state,
1469 &next_block_ctx,
1470 "referenced ",
1471 0,
1472 force_iodone_flag,
1473 !force_iodone_flag,
1474 mirror_num,
1475 &block_was_created);
1476 if (NULL == next_block) {
1477 pr_info("btrfsic: error, kmalloc failed!\n");
1478 btrfsic_release_block_ctx(&next_block_ctx);
1479 return -1;
1480 }
1481 if (!block_was_created) {
1482 if ((state->print_mask &
1483 BTRFSIC_PRINT_MASK_VERBOSE) &&
1484 next_block->logical_bytenr != next_bytenr &&
1485 !(!next_block->is_metadata &&
1486 0 == next_block->logical_bytenr)) {
1487 pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, D, bytenr mismatch (!= stored %llu).\n",
1488 next_bytenr,
1489 next_block_ctx.dev->name,
1490 next_block_ctx.dev_bytenr,
1491 mirror_num,
1492 next_block->logical_bytenr);
1493 }
1494 next_block->logical_bytenr = next_bytenr;
1495 next_block->mirror_num = mirror_num;
1496 }
1497
1498 l = btrfsic_block_link_lookup_or_add(state,
1499 &next_block_ctx,
1500 next_block, block,
1501 generation);
1502 btrfsic_release_block_ctx(&next_block_ctx);
1503 if (NULL == l)
1504 return -1;
1505 }
1506
1507 next_bytenr += chunk_len;
1508 num_bytes -= chunk_len;
1509 }
1510
1511 return 0;
1512 }
1513
btrfsic_map_block(struct btrfsic_state * state,u64 bytenr,u32 len,struct btrfsic_block_data_ctx * block_ctx_out,int mirror_num)1514 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1515 struct btrfsic_block_data_ctx *block_ctx_out,
1516 int mirror_num)
1517 {
1518 struct btrfs_fs_info *fs_info = state->fs_info;
1519 int ret;
1520 u64 length;
1521 struct btrfs_bio *multi = NULL;
1522 struct btrfs_device *device;
1523
1524 length = len;
1525 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ,
1526 bytenr, &length, &multi, mirror_num);
1527
1528 if (ret) {
1529 block_ctx_out->start = 0;
1530 block_ctx_out->dev_bytenr = 0;
1531 block_ctx_out->len = 0;
1532 block_ctx_out->dev = NULL;
1533 block_ctx_out->datav = NULL;
1534 block_ctx_out->pagev = NULL;
1535 block_ctx_out->mem_to_free = NULL;
1536
1537 return ret;
1538 }
1539
1540 device = multi->stripes[0].dev;
1541 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
1542 !device->bdev || !device->name)
1543 block_ctx_out->dev = NULL;
1544 else
1545 block_ctx_out->dev = btrfsic_dev_state_lookup(
1546 device->bdev->bd_dev);
1547 block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1548 block_ctx_out->start = bytenr;
1549 block_ctx_out->len = len;
1550 block_ctx_out->datav = NULL;
1551 block_ctx_out->pagev = NULL;
1552 block_ctx_out->mem_to_free = NULL;
1553
1554 kfree(multi);
1555 if (NULL == block_ctx_out->dev) {
1556 ret = -ENXIO;
1557 pr_info("btrfsic: error, cannot lookup dev (#1)!\n");
1558 }
1559
1560 return ret;
1561 }
1562
btrfsic_release_block_ctx(struct btrfsic_block_data_ctx * block_ctx)1563 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1564 {
1565 if (block_ctx->mem_to_free) {
1566 unsigned int num_pages;
1567
1568 BUG_ON(!block_ctx->datav);
1569 BUG_ON(!block_ctx->pagev);
1570 num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1571 PAGE_SHIFT;
1572 while (num_pages > 0) {
1573 num_pages--;
1574 if (block_ctx->datav[num_pages]) {
1575 kunmap(block_ctx->pagev[num_pages]);
1576 block_ctx->datav[num_pages] = NULL;
1577 }
1578 if (block_ctx->pagev[num_pages]) {
1579 __free_page(block_ctx->pagev[num_pages]);
1580 block_ctx->pagev[num_pages] = NULL;
1581 }
1582 }
1583
1584 kfree(block_ctx->mem_to_free);
1585 block_ctx->mem_to_free = NULL;
1586 block_ctx->pagev = NULL;
1587 block_ctx->datav = NULL;
1588 }
1589 }
1590
btrfsic_read_block(struct btrfsic_state * state,struct btrfsic_block_data_ctx * block_ctx)1591 static int btrfsic_read_block(struct btrfsic_state *state,
1592 struct btrfsic_block_data_ctx *block_ctx)
1593 {
1594 unsigned int num_pages;
1595 unsigned int i;
1596 u64 dev_bytenr;
1597 int ret;
1598
1599 BUG_ON(block_ctx->datav);
1600 BUG_ON(block_ctx->pagev);
1601 BUG_ON(block_ctx->mem_to_free);
1602 if (block_ctx->dev_bytenr & ((u64)PAGE_SIZE - 1)) {
1603 pr_info("btrfsic: read_block() with unaligned bytenr %llu\n",
1604 block_ctx->dev_bytenr);
1605 return -1;
1606 }
1607
1608 num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1609 PAGE_SHIFT;
1610 block_ctx->mem_to_free = kcalloc(sizeof(*block_ctx->datav) +
1611 sizeof(*block_ctx->pagev),
1612 num_pages, GFP_NOFS);
1613 if (!block_ctx->mem_to_free)
1614 return -ENOMEM;
1615 block_ctx->datav = block_ctx->mem_to_free;
1616 block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1617 for (i = 0; i < num_pages; i++) {
1618 block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1619 if (!block_ctx->pagev[i])
1620 return -1;
1621 }
1622
1623 dev_bytenr = block_ctx->dev_bytenr;
1624 for (i = 0; i < num_pages;) {
1625 struct bio *bio;
1626 unsigned int j;
1627
1628 bio = btrfs_io_bio_alloc(num_pages - i);
1629 bio_set_dev(bio, block_ctx->dev->bdev);
1630 bio->bi_iter.bi_sector = dev_bytenr >> 9;
1631 bio->bi_opf = REQ_OP_READ;
1632
1633 for (j = i; j < num_pages; j++) {
1634 ret = bio_add_page(bio, block_ctx->pagev[j],
1635 PAGE_SIZE, 0);
1636 if (PAGE_SIZE != ret)
1637 break;
1638 }
1639 if (j == i) {
1640 pr_info("btrfsic: error, failed to add a single page!\n");
1641 return -1;
1642 }
1643 if (submit_bio_wait(bio)) {
1644 pr_info("btrfsic: read error at logical %llu dev %s!\n",
1645 block_ctx->start, block_ctx->dev->name);
1646 bio_put(bio);
1647 return -1;
1648 }
1649 bio_put(bio);
1650 dev_bytenr += (j - i) * PAGE_SIZE;
1651 i = j;
1652 }
1653 for (i = 0; i < num_pages; i++)
1654 block_ctx->datav[i] = kmap(block_ctx->pagev[i]);
1655
1656 return block_ctx->len;
1657 }
1658
btrfsic_dump_database(struct btrfsic_state * state)1659 static void btrfsic_dump_database(struct btrfsic_state *state)
1660 {
1661 const struct btrfsic_block *b_all;
1662
1663 BUG_ON(NULL == state);
1664
1665 pr_info("all_blocks_list:\n");
1666 list_for_each_entry(b_all, &state->all_blocks_list, all_blocks_node) {
1667 const struct btrfsic_block_link *l;
1668
1669 pr_info("%c-block @%llu (%s/%llu/%d)\n",
1670 btrfsic_get_block_type(state, b_all),
1671 b_all->logical_bytenr, b_all->dev_state->name,
1672 b_all->dev_bytenr, b_all->mirror_num);
1673
1674 list_for_each_entry(l, &b_all->ref_to_list, node_ref_to) {
1675 pr_info(" %c @%llu (%s/%llu/%d) refers %u* to %c @%llu (%s/%llu/%d)\n",
1676 btrfsic_get_block_type(state, b_all),
1677 b_all->logical_bytenr, b_all->dev_state->name,
1678 b_all->dev_bytenr, b_all->mirror_num,
1679 l->ref_cnt,
1680 btrfsic_get_block_type(state, l->block_ref_to),
1681 l->block_ref_to->logical_bytenr,
1682 l->block_ref_to->dev_state->name,
1683 l->block_ref_to->dev_bytenr,
1684 l->block_ref_to->mirror_num);
1685 }
1686
1687 list_for_each_entry(l, &b_all->ref_from_list, node_ref_from) {
1688 pr_info(" %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
1689 btrfsic_get_block_type(state, b_all),
1690 b_all->logical_bytenr, b_all->dev_state->name,
1691 b_all->dev_bytenr, b_all->mirror_num,
1692 l->ref_cnt,
1693 btrfsic_get_block_type(state, l->block_ref_from),
1694 l->block_ref_from->logical_bytenr,
1695 l->block_ref_from->dev_state->name,
1696 l->block_ref_from->dev_bytenr,
1697 l->block_ref_from->mirror_num);
1698 }
1699
1700 pr_info("\n");
1701 }
1702 }
1703
1704 /*
1705 * Test whether the disk block contains a tree block (leaf or node)
1706 * (note that this test fails for the super block)
1707 */
btrfsic_test_for_metadata(struct btrfsic_state * state,char ** datav,unsigned int num_pages)1708 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
1709 char **datav, unsigned int num_pages)
1710 {
1711 struct btrfs_fs_info *fs_info = state->fs_info;
1712 struct btrfs_header *h;
1713 u8 csum[BTRFS_CSUM_SIZE];
1714 u32 crc = ~(u32)0;
1715 unsigned int i;
1716
1717 if (num_pages * PAGE_SIZE < state->metablock_size)
1718 return 1; /* not metadata */
1719 num_pages = state->metablock_size >> PAGE_SHIFT;
1720 h = (struct btrfs_header *)datav[0];
1721
1722 if (memcmp(h->fsid, fs_info->fsid, BTRFS_FSID_SIZE))
1723 return 1;
1724
1725 for (i = 0; i < num_pages; i++) {
1726 u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1727 size_t sublen = i ? PAGE_SIZE :
1728 (PAGE_SIZE - BTRFS_CSUM_SIZE);
1729
1730 crc = crc32c(crc, data, sublen);
1731 }
1732 btrfs_csum_final(crc, csum);
1733 if (memcmp(csum, h->csum, state->csum_size))
1734 return 1;
1735
1736 return 0; /* is metadata */
1737 }
1738
btrfsic_process_written_block(struct btrfsic_dev_state * dev_state,u64 dev_bytenr,char ** mapped_datav,unsigned int num_pages,struct bio * bio,int * bio_is_patched,struct buffer_head * bh,int submit_bio_bh_rw)1739 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1740 u64 dev_bytenr, char **mapped_datav,
1741 unsigned int num_pages,
1742 struct bio *bio, int *bio_is_patched,
1743 struct buffer_head *bh,
1744 int submit_bio_bh_rw)
1745 {
1746 int is_metadata;
1747 struct btrfsic_block *block;
1748 struct btrfsic_block_data_ctx block_ctx;
1749 int ret;
1750 struct btrfsic_state *state = dev_state->state;
1751 struct block_device *bdev = dev_state->bdev;
1752 unsigned int processed_len;
1753
1754 if (NULL != bio_is_patched)
1755 *bio_is_patched = 0;
1756
1757 again:
1758 if (num_pages == 0)
1759 return;
1760
1761 processed_len = 0;
1762 is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1763 num_pages));
1764
1765 block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1766 &state->block_hashtable);
1767 if (NULL != block) {
1768 u64 bytenr = 0;
1769 struct btrfsic_block_link *l, *tmp;
1770
1771 if (block->is_superblock) {
1772 bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1773 mapped_datav[0]);
1774 if (num_pages * PAGE_SIZE <
1775 BTRFS_SUPER_INFO_SIZE) {
1776 pr_info("btrfsic: cannot work with too short bios!\n");
1777 return;
1778 }
1779 is_metadata = 1;
1780 BUG_ON(BTRFS_SUPER_INFO_SIZE & (PAGE_SIZE - 1));
1781 processed_len = BTRFS_SUPER_INFO_SIZE;
1782 if (state->print_mask &
1783 BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1784 pr_info("[before new superblock is written]:\n");
1785 btrfsic_dump_tree_sub(state, block, 0);
1786 }
1787 }
1788 if (is_metadata) {
1789 if (!block->is_superblock) {
1790 if (num_pages * PAGE_SIZE <
1791 state->metablock_size) {
1792 pr_info("btrfsic: cannot work with too short bios!\n");
1793 return;
1794 }
1795 processed_len = state->metablock_size;
1796 bytenr = btrfs_stack_header_bytenr(
1797 (struct btrfs_header *)
1798 mapped_datav[0]);
1799 btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1800 dev_state,
1801 dev_bytenr);
1802 }
1803 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1804 if (block->logical_bytenr != bytenr &&
1805 !(!block->is_metadata &&
1806 block->logical_bytenr == 0))
1807 pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1808 bytenr, dev_state->name,
1809 dev_bytenr,
1810 block->mirror_num,
1811 btrfsic_get_block_type(state,
1812 block),
1813 block->logical_bytenr);
1814 else
1815 pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1816 bytenr, dev_state->name,
1817 dev_bytenr, block->mirror_num,
1818 btrfsic_get_block_type(state,
1819 block));
1820 }
1821 block->logical_bytenr = bytenr;
1822 } else {
1823 if (num_pages * PAGE_SIZE <
1824 state->datablock_size) {
1825 pr_info("btrfsic: cannot work with too short bios!\n");
1826 return;
1827 }
1828 processed_len = state->datablock_size;
1829 bytenr = block->logical_bytenr;
1830 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1831 pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1832 bytenr, dev_state->name, dev_bytenr,
1833 block->mirror_num,
1834 btrfsic_get_block_type(state, block));
1835 }
1836
1837 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1838 pr_info("ref_to_list: %cE, ref_from_list: %cE\n",
1839 list_empty(&block->ref_to_list) ? ' ' : '!',
1840 list_empty(&block->ref_from_list) ? ' ' : '!');
1841 if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1842 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",
1843 btrfsic_get_block_type(state, block), bytenr,
1844 dev_state->name, dev_bytenr, block->mirror_num,
1845 block->generation,
1846 btrfs_disk_key_objectid(&block->disk_key),
1847 block->disk_key.type,
1848 btrfs_disk_key_offset(&block->disk_key),
1849 btrfs_stack_header_generation(
1850 (struct btrfs_header *) mapped_datav[0]),
1851 state->max_superblock_generation);
1852 btrfsic_dump_tree(state);
1853 }
1854
1855 if (!block->is_iodone && !block->never_written) {
1856 pr_info("btrfs: attempt to overwrite %c-block @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu, which is not yet iodone!\n",
1857 btrfsic_get_block_type(state, block), bytenr,
1858 dev_state->name, dev_bytenr, block->mirror_num,
1859 block->generation,
1860 btrfs_stack_header_generation(
1861 (struct btrfs_header *)
1862 mapped_datav[0]));
1863 /* it would not be safe to go on */
1864 btrfsic_dump_tree(state);
1865 goto continue_loop;
1866 }
1867
1868 /*
1869 * Clear all references of this block. Do not free
1870 * the block itself even if is not referenced anymore
1871 * because it still carries valuable information
1872 * like whether it was ever written and IO completed.
1873 */
1874 list_for_each_entry_safe(l, tmp, &block->ref_to_list,
1875 node_ref_to) {
1876 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1877 btrfsic_print_rem_link(state, l);
1878 l->ref_cnt--;
1879 if (0 == l->ref_cnt) {
1880 list_del(&l->node_ref_to);
1881 list_del(&l->node_ref_from);
1882 btrfsic_block_link_hashtable_remove(l);
1883 btrfsic_block_link_free(l);
1884 }
1885 }
1886
1887 block_ctx.dev = dev_state;
1888 block_ctx.dev_bytenr = dev_bytenr;
1889 block_ctx.start = bytenr;
1890 block_ctx.len = processed_len;
1891 block_ctx.pagev = NULL;
1892 block_ctx.mem_to_free = NULL;
1893 block_ctx.datav = mapped_datav;
1894
1895 if (is_metadata || state->include_extent_data) {
1896 block->never_written = 0;
1897 block->iodone_w_error = 0;
1898 if (NULL != bio) {
1899 block->is_iodone = 0;
1900 BUG_ON(NULL == bio_is_patched);
1901 if (!*bio_is_patched) {
1902 block->orig_bio_bh_private =
1903 bio->bi_private;
1904 block->orig_bio_bh_end_io.bio =
1905 bio->bi_end_io;
1906 block->next_in_same_bio = NULL;
1907 bio->bi_private = block;
1908 bio->bi_end_io = btrfsic_bio_end_io;
1909 *bio_is_patched = 1;
1910 } else {
1911 struct btrfsic_block *chained_block =
1912 (struct btrfsic_block *)
1913 bio->bi_private;
1914
1915 BUG_ON(NULL == chained_block);
1916 block->orig_bio_bh_private =
1917 chained_block->orig_bio_bh_private;
1918 block->orig_bio_bh_end_io.bio =
1919 chained_block->orig_bio_bh_end_io.
1920 bio;
1921 block->next_in_same_bio = chained_block;
1922 bio->bi_private = block;
1923 }
1924 } else if (NULL != bh) {
1925 block->is_iodone = 0;
1926 block->orig_bio_bh_private = bh->b_private;
1927 block->orig_bio_bh_end_io.bh = bh->b_end_io;
1928 block->next_in_same_bio = NULL;
1929 bh->b_private = block;
1930 bh->b_end_io = btrfsic_bh_end_io;
1931 } else {
1932 block->is_iodone = 1;
1933 block->orig_bio_bh_private = NULL;
1934 block->orig_bio_bh_end_io.bio = NULL;
1935 block->next_in_same_bio = NULL;
1936 }
1937 }
1938
1939 block->flush_gen = dev_state->last_flush_gen + 1;
1940 block->submit_bio_bh_rw = submit_bio_bh_rw;
1941 if (is_metadata) {
1942 block->logical_bytenr = bytenr;
1943 block->is_metadata = 1;
1944 if (block->is_superblock) {
1945 BUG_ON(PAGE_SIZE !=
1946 BTRFS_SUPER_INFO_SIZE);
1947 ret = btrfsic_process_written_superblock(
1948 state,
1949 block,
1950 (struct btrfs_super_block *)
1951 mapped_datav[0]);
1952 if (state->print_mask &
1953 BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
1954 pr_info("[after new superblock is written]:\n");
1955 btrfsic_dump_tree_sub(state, block, 0);
1956 }
1957 } else {
1958 block->mirror_num = 0; /* unknown */
1959 ret = btrfsic_process_metablock(
1960 state,
1961 block,
1962 &block_ctx,
1963 0, 0);
1964 }
1965 if (ret)
1966 pr_info("btrfsic: btrfsic_process_metablock(root @%llu) failed!\n",
1967 dev_bytenr);
1968 } else {
1969 block->is_metadata = 0;
1970 block->mirror_num = 0; /* unknown */
1971 block->generation = BTRFSIC_GENERATION_UNKNOWN;
1972 if (!state->include_extent_data
1973 && list_empty(&block->ref_from_list)) {
1974 /*
1975 * disk block is overwritten with extent
1976 * data (not meta data) and we are configured
1977 * to not include extent data: take the
1978 * chance and free the block's memory
1979 */
1980 btrfsic_block_hashtable_remove(block);
1981 list_del(&block->all_blocks_node);
1982 btrfsic_block_free(block);
1983 }
1984 }
1985 btrfsic_release_block_ctx(&block_ctx);
1986 } else {
1987 /* block has not been found in hash table */
1988 u64 bytenr;
1989
1990 if (!is_metadata) {
1991 processed_len = state->datablock_size;
1992 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1993 pr_info("Written block (%s/%llu/?) !found in hash table, D.\n",
1994 dev_state->name, dev_bytenr);
1995 if (!state->include_extent_data) {
1996 /* ignore that written D block */
1997 goto continue_loop;
1998 }
1999
2000 /* this is getting ugly for the
2001 * include_extent_data case... */
2002 bytenr = 0; /* unknown */
2003 } else {
2004 processed_len = state->metablock_size;
2005 bytenr = btrfs_stack_header_bytenr(
2006 (struct btrfs_header *)
2007 mapped_datav[0]);
2008 btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
2009 dev_bytenr);
2010 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2011 pr_info("Written block @%llu (%s/%llu/?) !found in hash table, M.\n",
2012 bytenr, dev_state->name, dev_bytenr);
2013 }
2014
2015 block_ctx.dev = dev_state;
2016 block_ctx.dev_bytenr = dev_bytenr;
2017 block_ctx.start = bytenr;
2018 block_ctx.len = processed_len;
2019 block_ctx.pagev = NULL;
2020 block_ctx.mem_to_free = NULL;
2021 block_ctx.datav = mapped_datav;
2022
2023 block = btrfsic_block_alloc();
2024 if (NULL == block) {
2025 pr_info("btrfsic: error, kmalloc failed!\n");
2026 btrfsic_release_block_ctx(&block_ctx);
2027 goto continue_loop;
2028 }
2029 block->dev_state = dev_state;
2030 block->dev_bytenr = dev_bytenr;
2031 block->logical_bytenr = bytenr;
2032 block->is_metadata = is_metadata;
2033 block->never_written = 0;
2034 block->iodone_w_error = 0;
2035 block->mirror_num = 0; /* unknown */
2036 block->flush_gen = dev_state->last_flush_gen + 1;
2037 block->submit_bio_bh_rw = submit_bio_bh_rw;
2038 if (NULL != bio) {
2039 block->is_iodone = 0;
2040 BUG_ON(NULL == bio_is_patched);
2041 if (!*bio_is_patched) {
2042 block->orig_bio_bh_private = bio->bi_private;
2043 block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2044 block->next_in_same_bio = NULL;
2045 bio->bi_private = block;
2046 bio->bi_end_io = btrfsic_bio_end_io;
2047 *bio_is_patched = 1;
2048 } else {
2049 struct btrfsic_block *chained_block =
2050 (struct btrfsic_block *)
2051 bio->bi_private;
2052
2053 BUG_ON(NULL == chained_block);
2054 block->orig_bio_bh_private =
2055 chained_block->orig_bio_bh_private;
2056 block->orig_bio_bh_end_io.bio =
2057 chained_block->orig_bio_bh_end_io.bio;
2058 block->next_in_same_bio = chained_block;
2059 bio->bi_private = block;
2060 }
2061 } else if (NULL != bh) {
2062 block->is_iodone = 0;
2063 block->orig_bio_bh_private = bh->b_private;
2064 block->orig_bio_bh_end_io.bh = bh->b_end_io;
2065 block->next_in_same_bio = NULL;
2066 bh->b_private = block;
2067 bh->b_end_io = btrfsic_bh_end_io;
2068 } else {
2069 block->is_iodone = 1;
2070 block->orig_bio_bh_private = NULL;
2071 block->orig_bio_bh_end_io.bio = NULL;
2072 block->next_in_same_bio = NULL;
2073 }
2074 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2075 pr_info("New written %c-block @%llu (%s/%llu/%d)\n",
2076 is_metadata ? 'M' : 'D',
2077 block->logical_bytenr, block->dev_state->name,
2078 block->dev_bytenr, block->mirror_num);
2079 list_add(&block->all_blocks_node, &state->all_blocks_list);
2080 btrfsic_block_hashtable_add(block, &state->block_hashtable);
2081
2082 if (is_metadata) {
2083 ret = btrfsic_process_metablock(state, block,
2084 &block_ctx, 0, 0);
2085 if (ret)
2086 pr_info("btrfsic: process_metablock(root @%llu) failed!\n",
2087 dev_bytenr);
2088 }
2089 btrfsic_release_block_ctx(&block_ctx);
2090 }
2091
2092 continue_loop:
2093 BUG_ON(!processed_len);
2094 dev_bytenr += processed_len;
2095 mapped_datav += processed_len >> PAGE_SHIFT;
2096 num_pages -= processed_len >> PAGE_SHIFT;
2097 goto again;
2098 }
2099
btrfsic_bio_end_io(struct bio * bp)2100 static void btrfsic_bio_end_io(struct bio *bp)
2101 {
2102 struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2103 int iodone_w_error;
2104
2105 /* mutex is not held! This is not save if IO is not yet completed
2106 * on umount */
2107 iodone_w_error = 0;
2108 if (bp->bi_status)
2109 iodone_w_error = 1;
2110
2111 BUG_ON(NULL == block);
2112 bp->bi_private = block->orig_bio_bh_private;
2113 bp->bi_end_io = block->orig_bio_bh_end_io.bio;
2114
2115 do {
2116 struct btrfsic_block *next_block;
2117 struct btrfsic_dev_state *const dev_state = block->dev_state;
2118
2119 if ((dev_state->state->print_mask &
2120 BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2121 pr_info("bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2122 bp->bi_status,
2123 btrfsic_get_block_type(dev_state->state, block),
2124 block->logical_bytenr, dev_state->name,
2125 block->dev_bytenr, block->mirror_num);
2126 next_block = block->next_in_same_bio;
2127 block->iodone_w_error = iodone_w_error;
2128 if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2129 dev_state->last_flush_gen++;
2130 if ((dev_state->state->print_mask &
2131 BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2132 pr_info("bio_end_io() new %s flush_gen=%llu\n",
2133 dev_state->name,
2134 dev_state->last_flush_gen);
2135 }
2136 if (block->submit_bio_bh_rw & REQ_FUA)
2137 block->flush_gen = 0; /* FUA completed means block is
2138 * on disk */
2139 block->is_iodone = 1; /* for FLUSH, this releases the block */
2140 block = next_block;
2141 } while (NULL != block);
2142
2143 bp->bi_end_io(bp);
2144 }
2145
btrfsic_bh_end_io(struct buffer_head * bh,int uptodate)2146 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate)
2147 {
2148 struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private;
2149 int iodone_w_error = !uptodate;
2150 struct btrfsic_dev_state *dev_state;
2151
2152 BUG_ON(NULL == block);
2153 dev_state = block->dev_state;
2154 if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2155 pr_info("bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n",
2156 iodone_w_error,
2157 btrfsic_get_block_type(dev_state->state, block),
2158 block->logical_bytenr, block->dev_state->name,
2159 block->dev_bytenr, block->mirror_num);
2160
2161 block->iodone_w_error = iodone_w_error;
2162 if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2163 dev_state->last_flush_gen++;
2164 if ((dev_state->state->print_mask &
2165 BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2166 pr_info("bh_end_io() new %s flush_gen=%llu\n",
2167 dev_state->name, dev_state->last_flush_gen);
2168 }
2169 if (block->submit_bio_bh_rw & REQ_FUA)
2170 block->flush_gen = 0; /* FUA completed means block is on disk */
2171
2172 bh->b_private = block->orig_bio_bh_private;
2173 bh->b_end_io = block->orig_bio_bh_end_io.bh;
2174 block->is_iodone = 1; /* for FLUSH, this releases the block */
2175 bh->b_end_io(bh, uptodate);
2176 }
2177
btrfsic_process_written_superblock(struct btrfsic_state * state,struct btrfsic_block * const superblock,struct btrfs_super_block * const super_hdr)2178 static int btrfsic_process_written_superblock(
2179 struct btrfsic_state *state,
2180 struct btrfsic_block *const superblock,
2181 struct btrfs_super_block *const super_hdr)
2182 {
2183 struct btrfs_fs_info *fs_info = state->fs_info;
2184 int pass;
2185
2186 superblock->generation = btrfs_super_generation(super_hdr);
2187 if (!(superblock->generation > state->max_superblock_generation ||
2188 0 == state->max_superblock_generation)) {
2189 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2190 pr_info("btrfsic: superblock @%llu (%s/%llu/%d) with old gen %llu <= %llu\n",
2191 superblock->logical_bytenr,
2192 superblock->dev_state->name,
2193 superblock->dev_bytenr, superblock->mirror_num,
2194 btrfs_super_generation(super_hdr),
2195 state->max_superblock_generation);
2196 } else {
2197 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2198 pr_info("btrfsic: got new superblock @%llu (%s/%llu/%d) with new gen %llu > %llu\n",
2199 superblock->logical_bytenr,
2200 superblock->dev_state->name,
2201 superblock->dev_bytenr, superblock->mirror_num,
2202 btrfs_super_generation(super_hdr),
2203 state->max_superblock_generation);
2204
2205 state->max_superblock_generation =
2206 btrfs_super_generation(super_hdr);
2207 state->latest_superblock = superblock;
2208 }
2209
2210 for (pass = 0; pass < 3; pass++) {
2211 int ret;
2212 u64 next_bytenr;
2213 struct btrfsic_block *next_block;
2214 struct btrfsic_block_data_ctx tmp_next_block_ctx;
2215 struct btrfsic_block_link *l;
2216 int num_copies;
2217 int mirror_num;
2218 const char *additional_string = NULL;
2219 struct btrfs_disk_key tmp_disk_key = {0};
2220
2221 btrfs_set_disk_key_objectid(&tmp_disk_key,
2222 BTRFS_ROOT_ITEM_KEY);
2223 btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2224
2225 switch (pass) {
2226 case 0:
2227 btrfs_set_disk_key_objectid(&tmp_disk_key,
2228 BTRFS_ROOT_TREE_OBJECTID);
2229 additional_string = "root ";
2230 next_bytenr = btrfs_super_root(super_hdr);
2231 if (state->print_mask &
2232 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2233 pr_info("root@%llu\n", next_bytenr);
2234 break;
2235 case 1:
2236 btrfs_set_disk_key_objectid(&tmp_disk_key,
2237 BTRFS_CHUNK_TREE_OBJECTID);
2238 additional_string = "chunk ";
2239 next_bytenr = btrfs_super_chunk_root(super_hdr);
2240 if (state->print_mask &
2241 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2242 pr_info("chunk@%llu\n", next_bytenr);
2243 break;
2244 case 2:
2245 btrfs_set_disk_key_objectid(&tmp_disk_key,
2246 BTRFS_TREE_LOG_OBJECTID);
2247 additional_string = "log ";
2248 next_bytenr = btrfs_super_log_root(super_hdr);
2249 if (0 == next_bytenr)
2250 continue;
2251 if (state->print_mask &
2252 BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2253 pr_info("log@%llu\n", next_bytenr);
2254 break;
2255 }
2256
2257 num_copies = btrfs_num_copies(fs_info, next_bytenr,
2258 BTRFS_SUPER_INFO_SIZE);
2259 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2260 pr_info("num_copies(log_bytenr=%llu) = %d\n",
2261 next_bytenr, num_copies);
2262 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2263 int was_created;
2264
2265 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2266 pr_info("btrfsic_process_written_superblock(mirror_num=%d)\n", mirror_num);
2267 ret = btrfsic_map_block(state, next_bytenr,
2268 BTRFS_SUPER_INFO_SIZE,
2269 &tmp_next_block_ctx,
2270 mirror_num);
2271 if (ret) {
2272 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
2273 next_bytenr, mirror_num);
2274 return -1;
2275 }
2276
2277 next_block = btrfsic_block_lookup_or_add(
2278 state,
2279 &tmp_next_block_ctx,
2280 additional_string,
2281 1, 0, 1,
2282 mirror_num,
2283 &was_created);
2284 if (NULL == next_block) {
2285 pr_info("btrfsic: error, kmalloc failed!\n");
2286 btrfsic_release_block_ctx(&tmp_next_block_ctx);
2287 return -1;
2288 }
2289
2290 next_block->disk_key = tmp_disk_key;
2291 if (was_created)
2292 next_block->generation =
2293 BTRFSIC_GENERATION_UNKNOWN;
2294 l = btrfsic_block_link_lookup_or_add(
2295 state,
2296 &tmp_next_block_ctx,
2297 next_block,
2298 superblock,
2299 BTRFSIC_GENERATION_UNKNOWN);
2300 btrfsic_release_block_ctx(&tmp_next_block_ctx);
2301 if (NULL == l)
2302 return -1;
2303 }
2304 }
2305
2306 if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2307 btrfsic_dump_tree(state);
2308
2309 return 0;
2310 }
2311
btrfsic_check_all_ref_blocks(struct btrfsic_state * state,struct btrfsic_block * const block,int recursion_level)2312 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2313 struct btrfsic_block *const block,
2314 int recursion_level)
2315 {
2316 const struct btrfsic_block_link *l;
2317 int ret = 0;
2318
2319 if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2320 /*
2321 * Note that this situation can happen and does not
2322 * indicate an error in regular cases. It happens
2323 * when disk blocks are freed and later reused.
2324 * The check-integrity module is not aware of any
2325 * block free operations, it just recognizes block
2326 * write operations. Therefore it keeps the linkage
2327 * information for a block until a block is
2328 * rewritten. This can temporarily cause incorrect
2329 * and even circular linkage informations. This
2330 * causes no harm unless such blocks are referenced
2331 * by the most recent super block.
2332 */
2333 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2334 pr_info("btrfsic: abort cyclic linkage (case 1).\n");
2335
2336 return ret;
2337 }
2338
2339 /*
2340 * This algorithm is recursive because the amount of used stack
2341 * space is very small and the max recursion depth is limited.
2342 */
2343 list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2344 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2345 pr_info("rl=%d, %c @%llu (%s/%llu/%d) %u* refers to %c @%llu (%s/%llu/%d)\n",
2346 recursion_level,
2347 btrfsic_get_block_type(state, block),
2348 block->logical_bytenr, block->dev_state->name,
2349 block->dev_bytenr, block->mirror_num,
2350 l->ref_cnt,
2351 btrfsic_get_block_type(state, l->block_ref_to),
2352 l->block_ref_to->logical_bytenr,
2353 l->block_ref_to->dev_state->name,
2354 l->block_ref_to->dev_bytenr,
2355 l->block_ref_to->mirror_num);
2356 if (l->block_ref_to->never_written) {
2357 pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is never written!\n",
2358 btrfsic_get_block_type(state, l->block_ref_to),
2359 l->block_ref_to->logical_bytenr,
2360 l->block_ref_to->dev_state->name,
2361 l->block_ref_to->dev_bytenr,
2362 l->block_ref_to->mirror_num);
2363 ret = -1;
2364 } else if (!l->block_ref_to->is_iodone) {
2365 pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is not yet iodone!\n",
2366 btrfsic_get_block_type(state, l->block_ref_to),
2367 l->block_ref_to->logical_bytenr,
2368 l->block_ref_to->dev_state->name,
2369 l->block_ref_to->dev_bytenr,
2370 l->block_ref_to->mirror_num);
2371 ret = -1;
2372 } else if (l->block_ref_to->iodone_w_error) {
2373 pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which has write error!\n",
2374 btrfsic_get_block_type(state, l->block_ref_to),
2375 l->block_ref_to->logical_bytenr,
2376 l->block_ref_to->dev_state->name,
2377 l->block_ref_to->dev_bytenr,
2378 l->block_ref_to->mirror_num);
2379 ret = -1;
2380 } else if (l->parent_generation !=
2381 l->block_ref_to->generation &&
2382 BTRFSIC_GENERATION_UNKNOWN !=
2383 l->parent_generation &&
2384 BTRFSIC_GENERATION_UNKNOWN !=
2385 l->block_ref_to->generation) {
2386 pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) with generation %llu != parent generation %llu!\n",
2387 btrfsic_get_block_type(state, l->block_ref_to),
2388 l->block_ref_to->logical_bytenr,
2389 l->block_ref_to->dev_state->name,
2390 l->block_ref_to->dev_bytenr,
2391 l->block_ref_to->mirror_num,
2392 l->block_ref_to->generation,
2393 l->parent_generation);
2394 ret = -1;
2395 } else if (l->block_ref_to->flush_gen >
2396 l->block_ref_to->dev_state->last_flush_gen) {
2397 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",
2398 btrfsic_get_block_type(state, l->block_ref_to),
2399 l->block_ref_to->logical_bytenr,
2400 l->block_ref_to->dev_state->name,
2401 l->block_ref_to->dev_bytenr,
2402 l->block_ref_to->mirror_num, block->flush_gen,
2403 l->block_ref_to->dev_state->last_flush_gen);
2404 ret = -1;
2405 } else if (-1 == btrfsic_check_all_ref_blocks(state,
2406 l->block_ref_to,
2407 recursion_level +
2408 1)) {
2409 ret = -1;
2410 }
2411 }
2412
2413 return ret;
2414 }
2415
btrfsic_is_block_ref_by_superblock(const struct btrfsic_state * state,const struct btrfsic_block * block,int recursion_level)2416 static int btrfsic_is_block_ref_by_superblock(
2417 const struct btrfsic_state *state,
2418 const struct btrfsic_block *block,
2419 int recursion_level)
2420 {
2421 const struct btrfsic_block_link *l;
2422
2423 if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2424 /* refer to comment at "abort cyclic linkage (case 1)" */
2425 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2426 pr_info("btrfsic: abort cyclic linkage (case 2).\n");
2427
2428 return 0;
2429 }
2430
2431 /*
2432 * This algorithm is recursive because the amount of used stack space
2433 * is very small and the max recursion depth is limited.
2434 */
2435 list_for_each_entry(l, &block->ref_from_list, node_ref_from) {
2436 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2437 pr_info("rl=%d, %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
2438 recursion_level,
2439 btrfsic_get_block_type(state, block),
2440 block->logical_bytenr, block->dev_state->name,
2441 block->dev_bytenr, block->mirror_num,
2442 l->ref_cnt,
2443 btrfsic_get_block_type(state, l->block_ref_from),
2444 l->block_ref_from->logical_bytenr,
2445 l->block_ref_from->dev_state->name,
2446 l->block_ref_from->dev_bytenr,
2447 l->block_ref_from->mirror_num);
2448 if (l->block_ref_from->is_superblock &&
2449 state->latest_superblock->dev_bytenr ==
2450 l->block_ref_from->dev_bytenr &&
2451 state->latest_superblock->dev_state->bdev ==
2452 l->block_ref_from->dev_state->bdev)
2453 return 1;
2454 else if (btrfsic_is_block_ref_by_superblock(state,
2455 l->block_ref_from,
2456 recursion_level +
2457 1))
2458 return 1;
2459 }
2460
2461 return 0;
2462 }
2463
btrfsic_print_add_link(const struct btrfsic_state * state,const struct btrfsic_block_link * l)2464 static void btrfsic_print_add_link(const struct btrfsic_state *state,
2465 const struct btrfsic_block_link *l)
2466 {
2467 pr_info("Add %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2468 l->ref_cnt,
2469 btrfsic_get_block_type(state, l->block_ref_from),
2470 l->block_ref_from->logical_bytenr,
2471 l->block_ref_from->dev_state->name,
2472 l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2473 btrfsic_get_block_type(state, l->block_ref_to),
2474 l->block_ref_to->logical_bytenr,
2475 l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2476 l->block_ref_to->mirror_num);
2477 }
2478
btrfsic_print_rem_link(const struct btrfsic_state * state,const struct btrfsic_block_link * l)2479 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2480 const struct btrfsic_block_link *l)
2481 {
2482 pr_info("Rem %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2483 l->ref_cnt,
2484 btrfsic_get_block_type(state, l->block_ref_from),
2485 l->block_ref_from->logical_bytenr,
2486 l->block_ref_from->dev_state->name,
2487 l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2488 btrfsic_get_block_type(state, l->block_ref_to),
2489 l->block_ref_to->logical_bytenr,
2490 l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2491 l->block_ref_to->mirror_num);
2492 }
2493
btrfsic_get_block_type(const struct btrfsic_state * state,const struct btrfsic_block * block)2494 static char btrfsic_get_block_type(const struct btrfsic_state *state,
2495 const struct btrfsic_block *block)
2496 {
2497 if (block->is_superblock &&
2498 state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2499 state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2500 return 'S';
2501 else if (block->is_superblock)
2502 return 's';
2503 else if (block->is_metadata)
2504 return 'M';
2505 else
2506 return 'D';
2507 }
2508
btrfsic_dump_tree(const struct btrfsic_state * state)2509 static void btrfsic_dump_tree(const struct btrfsic_state *state)
2510 {
2511 btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2512 }
2513
btrfsic_dump_tree_sub(const struct btrfsic_state * state,const struct btrfsic_block * block,int indent_level)2514 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2515 const struct btrfsic_block *block,
2516 int indent_level)
2517 {
2518 const struct btrfsic_block_link *l;
2519 int indent_add;
2520 static char buf[80];
2521 int cursor_position;
2522
2523 /*
2524 * Should better fill an on-stack buffer with a complete line and
2525 * dump it at once when it is time to print a newline character.
2526 */
2527
2528 /*
2529 * This algorithm is recursive because the amount of used stack space
2530 * is very small and the max recursion depth is limited.
2531 */
2532 indent_add = sprintf(buf, "%c-%llu(%s/%llu/%u)",
2533 btrfsic_get_block_type(state, block),
2534 block->logical_bytenr, block->dev_state->name,
2535 block->dev_bytenr, block->mirror_num);
2536 if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2537 printk("[...]\n");
2538 return;
2539 }
2540 printk(buf);
2541 indent_level += indent_add;
2542 if (list_empty(&block->ref_to_list)) {
2543 printk("\n");
2544 return;
2545 }
2546 if (block->mirror_num > 1 &&
2547 !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2548 printk(" [...]\n");
2549 return;
2550 }
2551
2552 cursor_position = indent_level;
2553 list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2554 while (cursor_position < indent_level) {
2555 printk(" ");
2556 cursor_position++;
2557 }
2558 if (l->ref_cnt > 1)
2559 indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2560 else
2561 indent_add = sprintf(buf, " --> ");
2562 if (indent_level + indent_add >
2563 BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2564 printk("[...]\n");
2565 cursor_position = 0;
2566 continue;
2567 }
2568
2569 printk(buf);
2570
2571 btrfsic_dump_tree_sub(state, l->block_ref_to,
2572 indent_level + indent_add);
2573 cursor_position = 0;
2574 }
2575 }
2576
btrfsic_block_link_lookup_or_add(struct btrfsic_state * state,struct btrfsic_block_data_ctx * next_block_ctx,struct btrfsic_block * next_block,struct btrfsic_block * from_block,u64 parent_generation)2577 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2578 struct btrfsic_state *state,
2579 struct btrfsic_block_data_ctx *next_block_ctx,
2580 struct btrfsic_block *next_block,
2581 struct btrfsic_block *from_block,
2582 u64 parent_generation)
2583 {
2584 struct btrfsic_block_link *l;
2585
2586 l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2587 next_block_ctx->dev_bytenr,
2588 from_block->dev_state->bdev,
2589 from_block->dev_bytenr,
2590 &state->block_link_hashtable);
2591 if (NULL == l) {
2592 l = btrfsic_block_link_alloc();
2593 if (NULL == l) {
2594 pr_info("btrfsic: error, kmalloc failed!\n");
2595 return NULL;
2596 }
2597
2598 l->block_ref_to = next_block;
2599 l->block_ref_from = from_block;
2600 l->ref_cnt = 1;
2601 l->parent_generation = parent_generation;
2602
2603 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2604 btrfsic_print_add_link(state, l);
2605
2606 list_add(&l->node_ref_to, &from_block->ref_to_list);
2607 list_add(&l->node_ref_from, &next_block->ref_from_list);
2608
2609 btrfsic_block_link_hashtable_add(l,
2610 &state->block_link_hashtable);
2611 } else {
2612 l->ref_cnt++;
2613 l->parent_generation = parent_generation;
2614 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2615 btrfsic_print_add_link(state, l);
2616 }
2617
2618 return l;
2619 }
2620
btrfsic_block_lookup_or_add(struct btrfsic_state * state,struct btrfsic_block_data_ctx * block_ctx,const char * additional_string,int is_metadata,int is_iodone,int never_written,int mirror_num,int * was_created)2621 static struct btrfsic_block *btrfsic_block_lookup_or_add(
2622 struct btrfsic_state *state,
2623 struct btrfsic_block_data_ctx *block_ctx,
2624 const char *additional_string,
2625 int is_metadata,
2626 int is_iodone,
2627 int never_written,
2628 int mirror_num,
2629 int *was_created)
2630 {
2631 struct btrfsic_block *block;
2632
2633 block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2634 block_ctx->dev_bytenr,
2635 &state->block_hashtable);
2636 if (NULL == block) {
2637 struct btrfsic_dev_state *dev_state;
2638
2639 block = btrfsic_block_alloc();
2640 if (NULL == block) {
2641 pr_info("btrfsic: error, kmalloc failed!\n");
2642 return NULL;
2643 }
2644 dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev->bd_dev);
2645 if (NULL == dev_state) {
2646 pr_info("btrfsic: error, lookup dev_state failed!\n");
2647 btrfsic_block_free(block);
2648 return NULL;
2649 }
2650 block->dev_state = dev_state;
2651 block->dev_bytenr = block_ctx->dev_bytenr;
2652 block->logical_bytenr = block_ctx->start;
2653 block->is_metadata = is_metadata;
2654 block->is_iodone = is_iodone;
2655 block->never_written = never_written;
2656 block->mirror_num = mirror_num;
2657 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2658 pr_info("New %s%c-block @%llu (%s/%llu/%d)\n",
2659 additional_string,
2660 btrfsic_get_block_type(state, block),
2661 block->logical_bytenr, dev_state->name,
2662 block->dev_bytenr, mirror_num);
2663 list_add(&block->all_blocks_node, &state->all_blocks_list);
2664 btrfsic_block_hashtable_add(block, &state->block_hashtable);
2665 if (NULL != was_created)
2666 *was_created = 1;
2667 } else {
2668 if (NULL != was_created)
2669 *was_created = 0;
2670 }
2671
2672 return block;
2673 }
2674
btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state * state,u64 bytenr,struct btrfsic_dev_state * dev_state,u64 dev_bytenr)2675 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2676 u64 bytenr,
2677 struct btrfsic_dev_state *dev_state,
2678 u64 dev_bytenr)
2679 {
2680 struct btrfs_fs_info *fs_info = state->fs_info;
2681 struct btrfsic_block_data_ctx block_ctx;
2682 int num_copies;
2683 int mirror_num;
2684 int match = 0;
2685 int ret;
2686
2687 num_copies = btrfs_num_copies(fs_info, bytenr, state->metablock_size);
2688
2689 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2690 ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2691 &block_ctx, mirror_num);
2692 if (ret) {
2693 pr_info("btrfsic: btrfsic_map_block(logical @%llu, mirror %d) failed!\n",
2694 bytenr, mirror_num);
2695 continue;
2696 }
2697
2698 if (dev_state->bdev == block_ctx.dev->bdev &&
2699 dev_bytenr == block_ctx.dev_bytenr) {
2700 match++;
2701 btrfsic_release_block_ctx(&block_ctx);
2702 break;
2703 }
2704 btrfsic_release_block_ctx(&block_ctx);
2705 }
2706
2707 if (WARN_ON(!match)) {
2708 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",
2709 bytenr, dev_state->name, dev_bytenr);
2710 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2711 ret = btrfsic_map_block(state, bytenr,
2712 state->metablock_size,
2713 &block_ctx, mirror_num);
2714 if (ret)
2715 continue;
2716
2717 pr_info("Read logical bytenr @%llu maps to (%s/%llu/%d)\n",
2718 bytenr, block_ctx.dev->name,
2719 block_ctx.dev_bytenr, mirror_num);
2720 }
2721 }
2722 }
2723
btrfsic_dev_state_lookup(dev_t dev)2724 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev)
2725 {
2726 return btrfsic_dev_state_hashtable_lookup(dev,
2727 &btrfsic_dev_state_hashtable);
2728 }
2729
btrfsic_submit_bh(int op,int op_flags,struct buffer_head * bh)2730 int btrfsic_submit_bh(int op, int op_flags, struct buffer_head *bh)
2731 {
2732 struct btrfsic_dev_state *dev_state;
2733
2734 if (!btrfsic_is_initialized)
2735 return submit_bh(op, op_flags, bh);
2736
2737 mutex_lock(&btrfsic_mutex);
2738 /* since btrfsic_submit_bh() might also be called before
2739 * btrfsic_mount(), this might return NULL */
2740 dev_state = btrfsic_dev_state_lookup(bh->b_bdev->bd_dev);
2741
2742 /* Only called to write the superblock (incl. FLUSH/FUA) */
2743 if (NULL != dev_state &&
2744 (op == REQ_OP_WRITE) && bh->b_size > 0) {
2745 u64 dev_bytenr;
2746
2747 dev_bytenr = BTRFS_BDEV_BLOCKSIZE * bh->b_blocknr;
2748 if (dev_state->state->print_mask &
2749 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2750 pr_info("submit_bh(op=0x%x,0x%x, blocknr=%llu (bytenr %llu), size=%zu, data=%p, bdev=%p)\n",
2751 op, op_flags, (unsigned long long)bh->b_blocknr,
2752 dev_bytenr, bh->b_size, bh->b_data, bh->b_bdev);
2753 btrfsic_process_written_block(dev_state, dev_bytenr,
2754 &bh->b_data, 1, NULL,
2755 NULL, bh, op_flags);
2756 } else if (NULL != dev_state && (op_flags & REQ_PREFLUSH)) {
2757 if (dev_state->state->print_mask &
2758 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2759 pr_info("submit_bh(op=0x%x,0x%x FLUSH, bdev=%p)\n",
2760 op, op_flags, bh->b_bdev);
2761 if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2762 if ((dev_state->state->print_mask &
2763 (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2764 BTRFSIC_PRINT_MASK_VERBOSE)))
2765 pr_info("btrfsic_submit_bh(%s) with FLUSH but dummy block already in use (ignored)!\n",
2766 dev_state->name);
2767 } else {
2768 struct btrfsic_block *const block =
2769 &dev_state->dummy_block_for_bio_bh_flush;
2770
2771 block->is_iodone = 0;
2772 block->never_written = 0;
2773 block->iodone_w_error = 0;
2774 block->flush_gen = dev_state->last_flush_gen + 1;
2775 block->submit_bio_bh_rw = op_flags;
2776 block->orig_bio_bh_private = bh->b_private;
2777 block->orig_bio_bh_end_io.bh = bh->b_end_io;
2778 block->next_in_same_bio = NULL;
2779 bh->b_private = block;
2780 bh->b_end_io = btrfsic_bh_end_io;
2781 }
2782 }
2783 mutex_unlock(&btrfsic_mutex);
2784 return submit_bh(op, op_flags, bh);
2785 }
2786
__btrfsic_submit_bio(struct bio * bio)2787 static void __btrfsic_submit_bio(struct bio *bio)
2788 {
2789 struct btrfsic_dev_state *dev_state;
2790
2791 if (!btrfsic_is_initialized)
2792 return;
2793
2794 mutex_lock(&btrfsic_mutex);
2795 /* since btrfsic_submit_bio() is also called before
2796 * btrfsic_mount(), this might return NULL */
2797 dev_state = btrfsic_dev_state_lookup(bio_dev(bio) + bio->bi_partno);
2798 if (NULL != dev_state &&
2799 (bio_op(bio) == REQ_OP_WRITE) && bio_has_data(bio)) {
2800 unsigned int i = 0;
2801 u64 dev_bytenr;
2802 u64 cur_bytenr;
2803 struct bio_vec bvec;
2804 struct bvec_iter iter;
2805 int bio_is_patched;
2806 char **mapped_datav;
2807 unsigned int segs = bio_segments(bio);
2808
2809 dev_bytenr = 512 * bio->bi_iter.bi_sector;
2810 bio_is_patched = 0;
2811 if (dev_state->state->print_mask &
2812 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2813 pr_info("submit_bio(rw=%d,0x%x, bi_vcnt=%u, bi_sector=%llu (bytenr %llu), bi_disk=%p)\n",
2814 bio_op(bio), bio->bi_opf, segs,
2815 (unsigned long long)bio->bi_iter.bi_sector,
2816 dev_bytenr, bio->bi_disk);
2817
2818 mapped_datav = kmalloc_array(segs,
2819 sizeof(*mapped_datav), GFP_NOFS);
2820 if (!mapped_datav)
2821 goto leave;
2822 cur_bytenr = dev_bytenr;
2823
2824 bio_for_each_segment(bvec, bio, iter) {
2825 BUG_ON(bvec.bv_len != PAGE_SIZE);
2826 mapped_datav[i] = kmap(bvec.bv_page);
2827 i++;
2828
2829 if (dev_state->state->print_mask &
2830 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
2831 pr_info("#%u: bytenr=%llu, len=%u, offset=%u\n",
2832 i, cur_bytenr, bvec.bv_len, bvec.bv_offset);
2833 cur_bytenr += bvec.bv_len;
2834 }
2835 btrfsic_process_written_block(dev_state, dev_bytenr,
2836 mapped_datav, segs,
2837 bio, &bio_is_patched,
2838 NULL, bio->bi_opf);
2839 bio_for_each_segment(bvec, bio, iter)
2840 kunmap(bvec.bv_page);
2841 kfree(mapped_datav);
2842 } else if (NULL != dev_state && (bio->bi_opf & REQ_PREFLUSH)) {
2843 if (dev_state->state->print_mask &
2844 BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2845 pr_info("submit_bio(rw=%d,0x%x FLUSH, disk=%p)\n",
2846 bio_op(bio), bio->bi_opf, bio->bi_disk);
2847 if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2848 if ((dev_state->state->print_mask &
2849 (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2850 BTRFSIC_PRINT_MASK_VERBOSE)))
2851 pr_info("btrfsic_submit_bio(%s) with FLUSH but dummy block already in use (ignored)!\n",
2852 dev_state->name);
2853 } else {
2854 struct btrfsic_block *const block =
2855 &dev_state->dummy_block_for_bio_bh_flush;
2856
2857 block->is_iodone = 0;
2858 block->never_written = 0;
2859 block->iodone_w_error = 0;
2860 block->flush_gen = dev_state->last_flush_gen + 1;
2861 block->submit_bio_bh_rw = bio->bi_opf;
2862 block->orig_bio_bh_private = bio->bi_private;
2863 block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2864 block->next_in_same_bio = NULL;
2865 bio->bi_private = block;
2866 bio->bi_end_io = btrfsic_bio_end_io;
2867 }
2868 }
2869 leave:
2870 mutex_unlock(&btrfsic_mutex);
2871 }
2872
btrfsic_submit_bio(struct bio * bio)2873 void btrfsic_submit_bio(struct bio *bio)
2874 {
2875 __btrfsic_submit_bio(bio);
2876 submit_bio(bio);
2877 }
2878
btrfsic_submit_bio_wait(struct bio * bio)2879 int btrfsic_submit_bio_wait(struct bio *bio)
2880 {
2881 __btrfsic_submit_bio(bio);
2882 return submit_bio_wait(bio);
2883 }
2884
btrfsic_mount(struct btrfs_fs_info * fs_info,struct btrfs_fs_devices * fs_devices,int including_extent_data,u32 print_mask)2885 int btrfsic_mount(struct btrfs_fs_info *fs_info,
2886 struct btrfs_fs_devices *fs_devices,
2887 int including_extent_data, u32 print_mask)
2888 {
2889 int ret;
2890 struct btrfsic_state *state;
2891 struct list_head *dev_head = &fs_devices->devices;
2892 struct btrfs_device *device;
2893
2894 if (fs_info->nodesize & ((u64)PAGE_SIZE - 1)) {
2895 pr_info("btrfsic: cannot handle nodesize %d not being a multiple of PAGE_SIZE %ld!\n",
2896 fs_info->nodesize, PAGE_SIZE);
2897 return -1;
2898 }
2899 if (fs_info->sectorsize & ((u64)PAGE_SIZE - 1)) {
2900 pr_info("btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_SIZE %ld!\n",
2901 fs_info->sectorsize, PAGE_SIZE);
2902 return -1;
2903 }
2904 state = kvzalloc(sizeof(*state), GFP_KERNEL);
2905 if (!state) {
2906 pr_info("btrfs check-integrity: allocation failed!\n");
2907 return -ENOMEM;
2908 }
2909
2910 if (!btrfsic_is_initialized) {
2911 mutex_init(&btrfsic_mutex);
2912 btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
2913 btrfsic_is_initialized = 1;
2914 }
2915 mutex_lock(&btrfsic_mutex);
2916 state->fs_info = fs_info;
2917 state->print_mask = print_mask;
2918 state->include_extent_data = including_extent_data;
2919 state->csum_size = 0;
2920 state->metablock_size = fs_info->nodesize;
2921 state->datablock_size = fs_info->sectorsize;
2922 INIT_LIST_HEAD(&state->all_blocks_list);
2923 btrfsic_block_hashtable_init(&state->block_hashtable);
2924 btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
2925 state->max_superblock_generation = 0;
2926 state->latest_superblock = NULL;
2927
2928 list_for_each_entry(device, dev_head, dev_list) {
2929 struct btrfsic_dev_state *ds;
2930 const char *p;
2931
2932 if (!device->bdev || !device->name)
2933 continue;
2934
2935 ds = btrfsic_dev_state_alloc();
2936 if (NULL == ds) {
2937 pr_info("btrfs check-integrity: kmalloc() failed!\n");
2938 mutex_unlock(&btrfsic_mutex);
2939 return -ENOMEM;
2940 }
2941 ds->bdev = device->bdev;
2942 ds->state = state;
2943 bdevname(ds->bdev, ds->name);
2944 ds->name[BDEVNAME_SIZE - 1] = '\0';
2945 p = kbasename(ds->name);
2946 strlcpy(ds->name, p, sizeof(ds->name));
2947 btrfsic_dev_state_hashtable_add(ds,
2948 &btrfsic_dev_state_hashtable);
2949 }
2950
2951 ret = btrfsic_process_superblock(state, fs_devices);
2952 if (0 != ret) {
2953 mutex_unlock(&btrfsic_mutex);
2954 btrfsic_unmount(fs_devices);
2955 return ret;
2956 }
2957
2958 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
2959 btrfsic_dump_database(state);
2960 if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
2961 btrfsic_dump_tree(state);
2962
2963 mutex_unlock(&btrfsic_mutex);
2964 return 0;
2965 }
2966
btrfsic_unmount(struct btrfs_fs_devices * fs_devices)2967 void btrfsic_unmount(struct btrfs_fs_devices *fs_devices)
2968 {
2969 struct btrfsic_block *b_all, *tmp_all;
2970 struct btrfsic_state *state;
2971 struct list_head *dev_head = &fs_devices->devices;
2972 struct btrfs_device *device;
2973
2974 if (!btrfsic_is_initialized)
2975 return;
2976
2977 mutex_lock(&btrfsic_mutex);
2978
2979 state = NULL;
2980 list_for_each_entry(device, dev_head, dev_list) {
2981 struct btrfsic_dev_state *ds;
2982
2983 if (!device->bdev || !device->name)
2984 continue;
2985
2986 ds = btrfsic_dev_state_hashtable_lookup(
2987 device->bdev->bd_dev,
2988 &btrfsic_dev_state_hashtable);
2989 if (NULL != ds) {
2990 state = ds->state;
2991 btrfsic_dev_state_hashtable_remove(ds);
2992 btrfsic_dev_state_free(ds);
2993 }
2994 }
2995
2996 if (NULL == state) {
2997 pr_info("btrfsic: error, cannot find state information on umount!\n");
2998 mutex_unlock(&btrfsic_mutex);
2999 return;
3000 }
3001
3002 /*
3003 * Don't care about keeping the lists' state up to date,
3004 * just free all memory that was allocated dynamically.
3005 * Free the blocks and the block_links.
3006 */
3007 list_for_each_entry_safe(b_all, tmp_all, &state->all_blocks_list,
3008 all_blocks_node) {
3009 struct btrfsic_block_link *l, *tmp;
3010
3011 list_for_each_entry_safe(l, tmp, &b_all->ref_to_list,
3012 node_ref_to) {
3013 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
3014 btrfsic_print_rem_link(state, l);
3015
3016 l->ref_cnt--;
3017 if (0 == l->ref_cnt)
3018 btrfsic_block_link_free(l);
3019 }
3020
3021 if (b_all->is_iodone || b_all->never_written)
3022 btrfsic_block_free(b_all);
3023 else
3024 pr_info("btrfs: attempt to free %c-block @%llu (%s/%llu/%d) on umount which is not yet iodone!\n",
3025 btrfsic_get_block_type(state, b_all),
3026 b_all->logical_bytenr, b_all->dev_state->name,
3027 b_all->dev_bytenr, b_all->mirror_num);
3028 }
3029
3030 mutex_unlock(&btrfsic_mutex);
3031
3032 kvfree(state);
3033 }
3034