btrfs: open-code bio_set_op_attrs
[sfrench/cifs-2.6.git] / fs / btrfs / check-integrity.c
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
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
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
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
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
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
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
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
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
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
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
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
507 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
508 {
509         list_del(&b->collision_resolving_node);
510 }
511
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
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
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
556 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
557 {
558         list_del(&l->collision_resolving_node);
559 }
560
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
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
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
609 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
610 {
611         list_del(&ds->collision_resolving_node);
612 }
613
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
629 static int btrfsic_process_superblock(struct btrfsic_state *state,
630                                       struct btrfs_fs_devices *fs_devices)
631 {
632         struct btrfs_fs_info *fs_info = state->fs_info;
633         struct btrfs_super_block *selected_super;
634         struct list_head *dev_head = &fs_devices->devices;
635         struct btrfs_device *device;
636         struct btrfsic_dev_state *selected_dev_state = NULL;
637         int ret = 0;
638         int pass;
639
640         BUG_ON(NULL == state);
641         selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
642         if (NULL == selected_super) {
643                 pr_info("btrfsic: error, kmalloc failed!\n");
644                 return -ENOMEM;
645         }
646
647         list_for_each_entry(device, dev_head, dev_list) {
648                 int i;
649                 struct btrfsic_dev_state *dev_state;
650
651                 if (!device->bdev || !device->name)
652                         continue;
653
654                 dev_state = btrfsic_dev_state_lookup(device->bdev->bd_dev);
655                 BUG_ON(NULL == dev_state);
656                 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
657                         ret = btrfsic_process_superblock_dev_mirror(
658                                         state, dev_state, device, i,
659                                         &selected_dev_state, selected_super);
660                         if (0 != ret && 0 == i) {
661                                 kfree(selected_super);
662                                 return ret;
663                         }
664                 }
665         }
666
667         if (NULL == state->latest_superblock) {
668                 pr_info("btrfsic: no superblock found!\n");
669                 kfree(selected_super);
670                 return -1;
671         }
672
673         state->csum_size = btrfs_super_csum_size(selected_super);
674
675         for (pass = 0; pass < 3; pass++) {
676                 int num_copies;
677                 int mirror_num;
678                 u64 next_bytenr;
679
680                 switch (pass) {
681                 case 0:
682                         next_bytenr = btrfs_super_root(selected_super);
683                         if (state->print_mask &
684                             BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
685                                 pr_info("root@%llu\n", next_bytenr);
686                         break;
687                 case 1:
688                         next_bytenr = btrfs_super_chunk_root(selected_super);
689                         if (state->print_mask &
690                             BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
691                                 pr_info("chunk@%llu\n", next_bytenr);
692                         break;
693                 case 2:
694                         next_bytenr = btrfs_super_log_root(selected_super);
695                         if (0 == next_bytenr)
696                                 continue;
697                         if (state->print_mask &
698                             BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
699                                 pr_info("log@%llu\n", next_bytenr);
700                         break;
701                 }
702
703                 num_copies = btrfs_num_copies(fs_info, next_bytenr,
704                                               state->metablock_size);
705                 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
706                         pr_info("num_copies(log_bytenr=%llu) = %d\n",
707                                next_bytenr, num_copies);
708
709                 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
710                         struct btrfsic_block *next_block;
711                         struct btrfsic_block_data_ctx tmp_next_block_ctx;
712                         struct btrfsic_block_link *l;
713
714                         ret = btrfsic_map_block(state, next_bytenr,
715                                                 state->metablock_size,
716                                                 &tmp_next_block_ctx,
717                                                 mirror_num);
718                         if (ret) {
719                                 pr_info("btrfsic: btrfsic_map_block(root @%llu, mirror %d) failed!\n",
720                                        next_bytenr, mirror_num);
721                                 kfree(selected_super);
722                                 return -1;
723                         }
724
725                         next_block = btrfsic_block_hashtable_lookup(
726                                         tmp_next_block_ctx.dev->bdev,
727                                         tmp_next_block_ctx.dev_bytenr,
728                                         &state->block_hashtable);
729                         BUG_ON(NULL == next_block);
730
731                         l = btrfsic_block_link_hashtable_lookup(
732                                         tmp_next_block_ctx.dev->bdev,
733                                         tmp_next_block_ctx.dev_bytenr,
734                                         state->latest_superblock->dev_state->
735                                         bdev,
736                                         state->latest_superblock->dev_bytenr,
737                                         &state->block_link_hashtable);
738                         BUG_ON(NULL == l);
739
740                         ret = btrfsic_read_block(state, &tmp_next_block_ctx);
741                         if (ret < (int)PAGE_SIZE) {
742                                 pr_info("btrfsic: read @logical %llu failed!\n",
743                                        tmp_next_block_ctx.start);
744                                 btrfsic_release_block_ctx(&tmp_next_block_ctx);
745                                 kfree(selected_super);
746                                 return -1;
747                         }
748
749                         ret = btrfsic_process_metablock(state,
750                                                         next_block,
751                                                         &tmp_next_block_ctx,
752                                                         BTRFS_MAX_LEVEL + 3, 1);
753                         btrfsic_release_block_ctx(&tmp_next_block_ctx);
754                 }
755         }
756
757         kfree(selected_super);
758         return ret;
759 }
760
761 static int btrfsic_process_superblock_dev_mirror(
762                 struct btrfsic_state *state,
763                 struct btrfsic_dev_state *dev_state,
764                 struct btrfs_device *device,
765                 int superblock_mirror_num,
766                 struct btrfsic_dev_state **selected_dev_state,
767                 struct btrfs_super_block *selected_super)
768 {
769         struct btrfs_fs_info *fs_info = state->fs_info;
770         struct btrfs_super_block *super_tmp;
771         u64 dev_bytenr;
772         struct buffer_head *bh;
773         struct btrfsic_block *superblock_tmp;
774         int pass;
775         struct block_device *const superblock_bdev = device->bdev;
776
777         /* super block bytenr is always the unmapped device bytenr */
778         dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
779         if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->commit_total_bytes)
780                 return -1;
781         bh = __bread(superblock_bdev, dev_bytenr / BTRFS_BDEV_BLOCKSIZE,
782                      BTRFS_SUPER_INFO_SIZE);
783         if (NULL == bh)
784                 return -1;
785         super_tmp = (struct btrfs_super_block *)
786             (bh->b_data + (dev_bytenr & (BTRFS_BDEV_BLOCKSIZE - 1)));
787
788         if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
789             btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
790             memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
791             btrfs_super_nodesize(super_tmp) != state->metablock_size ||
792             btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
793                 brelse(bh);
794                 return 0;
795         }
796
797         superblock_tmp =
798             btrfsic_block_hashtable_lookup(superblock_bdev,
799                                            dev_bytenr,
800                                            &state->block_hashtable);
801         if (NULL == superblock_tmp) {
802                 superblock_tmp = btrfsic_block_alloc();
803                 if (NULL == superblock_tmp) {
804                         pr_info("btrfsic: error, kmalloc failed!\n");
805                         brelse(bh);
806                         return -1;
807                 }
808                 /* for superblock, only the dev_bytenr makes sense */
809                 superblock_tmp->dev_bytenr = dev_bytenr;
810                 superblock_tmp->dev_state = dev_state;
811                 superblock_tmp->logical_bytenr = dev_bytenr;
812                 superblock_tmp->generation = btrfs_super_generation(super_tmp);
813                 superblock_tmp->is_metadata = 1;
814                 superblock_tmp->is_superblock = 1;
815                 superblock_tmp->is_iodone = 1;
816                 superblock_tmp->never_written = 0;
817                 superblock_tmp->mirror_num = 1 + superblock_mirror_num;
818                 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
819                         btrfs_info_in_rcu(fs_info,
820                                 "new initial S-block (bdev %p, %s) @%llu (%s/%llu/%d)",
821                                      superblock_bdev,
822                                      rcu_str_deref(device->name), dev_bytenr,
823                                      dev_state->name, dev_bytenr,
824                                      superblock_mirror_num);
825                 list_add(&superblock_tmp->all_blocks_node,
826                          &state->all_blocks_list);
827                 btrfsic_block_hashtable_add(superblock_tmp,
828                                             &state->block_hashtable);
829         }
830
831         /* select the one with the highest generation field */
832         if (btrfs_super_generation(super_tmp) >
833             state->max_superblock_generation ||
834             0 == state->max_superblock_generation) {
835                 memcpy(selected_super, super_tmp, sizeof(*selected_super));
836                 *selected_dev_state = dev_state;
837                 state->max_superblock_generation =
838                     btrfs_super_generation(super_tmp);
839                 state->latest_superblock = superblock_tmp;
840         }
841
842         for (pass = 0; pass < 3; pass++) {
843                 u64 next_bytenr;
844                 int num_copies;
845                 int mirror_num;
846                 const char *additional_string = NULL;
847                 struct btrfs_disk_key tmp_disk_key;
848
849                 tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
850                 tmp_disk_key.offset = 0;
851                 switch (pass) {
852                 case 0:
853                         btrfs_set_disk_key_objectid(&tmp_disk_key,
854                                                     BTRFS_ROOT_TREE_OBJECTID);
855                         additional_string = "initial root ";
856                         next_bytenr = btrfs_super_root(super_tmp);
857                         break;
858                 case 1:
859                         btrfs_set_disk_key_objectid(&tmp_disk_key,
860                                                     BTRFS_CHUNK_TREE_OBJECTID);
861                         additional_string = "initial chunk ";
862                         next_bytenr = btrfs_super_chunk_root(super_tmp);
863                         break;
864                 case 2:
865                         btrfs_set_disk_key_objectid(&tmp_disk_key,
866                                                     BTRFS_TREE_LOG_OBJECTID);
867                         additional_string = "initial log ";
868                         next_bytenr = btrfs_super_log_root(super_tmp);
869                         if (0 == next_bytenr)
870                                 continue;
871                         break;
872                 }
873
874                 num_copies = btrfs_num_copies(fs_info, next_bytenr,
875                                               state->metablock_size);
876                 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
877                         pr_info("num_copies(log_bytenr=%llu) = %d\n",
878                                next_bytenr, num_copies);
879                 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
880                         struct btrfsic_block *next_block;
881                         struct btrfsic_block_data_ctx tmp_next_block_ctx;
882                         struct btrfsic_block_link *l;
883
884                         if (btrfsic_map_block(state, next_bytenr,
885                                               state->metablock_size,
886                                               &tmp_next_block_ctx,
887                                               mirror_num)) {
888                                 pr_info("btrfsic: btrfsic_map_block(bytenr @%llu, mirror %d) failed!\n",
889                                        next_bytenr, mirror_num);
890                                 brelse(bh);
891                                 return -1;
892                         }
893
894                         next_block = btrfsic_block_lookup_or_add(
895                                         state, &tmp_next_block_ctx,
896                                         additional_string, 1, 1, 0,
897                                         mirror_num, NULL);
898                         if (NULL == next_block) {
899                                 btrfsic_release_block_ctx(&tmp_next_block_ctx);
900                                 brelse(bh);
901                                 return -1;
902                         }
903
904                         next_block->disk_key = tmp_disk_key;
905                         next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
906                         l = btrfsic_block_link_lookup_or_add(
907                                         state, &tmp_next_block_ctx,
908                                         next_block, superblock_tmp,
909                                         BTRFSIC_GENERATION_UNKNOWN);
910                         btrfsic_release_block_ctx(&tmp_next_block_ctx);
911                         if (NULL == l) {
912                                 brelse(bh);
913                                 return -1;
914                         }
915                 }
916         }
917         if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
918                 btrfsic_dump_tree_sub(state, superblock_tmp, 0);
919
920         brelse(bh);
921         return 0;
922 }
923
924 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
925 {
926         struct btrfsic_stack_frame *sf;
927
928         sf = kzalloc(sizeof(*sf), GFP_NOFS);
929         if (NULL == sf)
930                 pr_info("btrfsic: alloc memory failed!\n");
931         else
932                 sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
933         return sf;
934 }
935
936 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
937 {
938         BUG_ON(!(NULL == sf ||
939                  BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
940         kfree(sf);
941 }
942
943 static int btrfsic_process_metablock(
944                 struct btrfsic_state *state,
945                 struct btrfsic_block *const first_block,
946                 struct btrfsic_block_data_ctx *const first_block_ctx,
947                 int first_limit_nesting, int force_iodone_flag)
948 {
949         struct btrfsic_stack_frame initial_stack_frame = { 0 };
950         struct btrfsic_stack_frame *sf;
951         struct btrfsic_stack_frame *next_stack;
952         struct btrfs_header *const first_hdr =
953                 (struct btrfs_header *)first_block_ctx->datav[0];
954
955         BUG_ON(!first_hdr);
956         sf = &initial_stack_frame;
957         sf->error = 0;
958         sf->i = -1;
959         sf->limit_nesting = first_limit_nesting;
960         sf->block = first_block;
961         sf->block_ctx = first_block_ctx;
962         sf->next_block = NULL;
963         sf->hdr = first_hdr;
964         sf->prev = NULL;
965
966 continue_with_new_stack_frame:
967         sf->block->generation = le64_to_cpu(sf->hdr->generation);
968         if (0 == sf->hdr->level) {
969                 struct btrfs_leaf *const leafhdr =
970                     (struct btrfs_leaf *)sf->hdr;
971
972                 if (-1 == sf->i) {
973                         sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
974
975                         if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
976                                 pr_info("leaf %llu items %d generation %llu owner %llu\n",
977                                        sf->block_ctx->start, sf->nr,
978                                        btrfs_stack_header_generation(
979                                                &leafhdr->header),
980                                        btrfs_stack_header_owner(
981                                                &leafhdr->header));
982                 }
983
984 continue_with_current_leaf_stack_frame:
985                 if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
986                         sf->i++;
987                         sf->num_copies = 0;
988                 }
989
990                 if (sf->i < sf->nr) {
991                         struct btrfs_item disk_item;
992                         u32 disk_item_offset =
993                                 (uintptr_t)(leafhdr->items + sf->i) -
994                                 (uintptr_t)leafhdr;
995                         struct btrfs_disk_key *disk_key;
996                         u8 type;
997                         u32 item_offset;
998                         u32 item_size;
999
1000                         if (disk_item_offset + sizeof(struct btrfs_item) >
1001                             sf->block_ctx->len) {
1002 leaf_item_out_of_bounce_error:
1003                                 pr_info("btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
1004                                        sf->block_ctx->start,
1005                                        sf->block_ctx->dev->name);
1006                                 goto one_stack_frame_backwards;
1007                         }
1008                         btrfsic_read_from_block_data(sf->block_ctx,
1009                                                      &disk_item,
1010                                                      disk_item_offset,
1011                                                      sizeof(struct btrfs_item));
1012                         item_offset = btrfs_stack_item_offset(&disk_item);
1013                         item_size = btrfs_stack_item_size(&disk_item);
1014                         disk_key = &disk_item.key;
1015                         type = btrfs_disk_key_type(disk_key);
1016
1017                         if (BTRFS_ROOT_ITEM_KEY == type) {
1018                                 struct btrfs_root_item root_item;
1019                                 u32 root_item_offset;
1020                                 u64 next_bytenr;
1021
1022                                 root_item_offset = item_offset +
1023                                         offsetof(struct btrfs_leaf, items);
1024                                 if (root_item_offset + item_size >
1025                                     sf->block_ctx->len)
1026                                         goto leaf_item_out_of_bounce_error;
1027                                 btrfsic_read_from_block_data(
1028                                         sf->block_ctx, &root_item,
1029                                         root_item_offset,
1030                                         item_size);
1031                                 next_bytenr = btrfs_root_bytenr(&root_item);
1032
1033                                 sf->error =
1034                                     btrfsic_create_link_to_next_block(
1035                                                 state,
1036                                                 sf->block,
1037                                                 sf->block_ctx,
1038                                                 next_bytenr,
1039                                                 sf->limit_nesting,
1040                                                 &sf->next_block_ctx,
1041                                                 &sf->next_block,
1042                                                 force_iodone_flag,
1043                                                 &sf->num_copies,
1044                                                 &sf->mirror_num,
1045                                                 disk_key,
1046                                                 btrfs_root_generation(
1047                                                 &root_item));
1048                                 if (sf->error)
1049                                         goto one_stack_frame_backwards;
1050
1051                                 if (NULL != sf->next_block) {
1052                                         struct btrfs_header *const next_hdr =
1053                                             (struct btrfs_header *)
1054                                             sf->next_block_ctx.datav[0];
1055
1056                                         next_stack =
1057                                             btrfsic_stack_frame_alloc();
1058                                         if (NULL == next_stack) {
1059                                                 sf->error = -1;
1060                                                 btrfsic_release_block_ctx(
1061                                                                 &sf->
1062                                                                 next_block_ctx);
1063                                                 goto one_stack_frame_backwards;
1064                                         }
1065
1066                                         next_stack->i = -1;
1067                                         next_stack->block = sf->next_block;
1068                                         next_stack->block_ctx =
1069                                             &sf->next_block_ctx;
1070                                         next_stack->next_block = NULL;
1071                                         next_stack->hdr = next_hdr;
1072                                         next_stack->limit_nesting =
1073                                             sf->limit_nesting - 1;
1074                                         next_stack->prev = sf;
1075                                         sf = next_stack;
1076                                         goto continue_with_new_stack_frame;
1077                                 }
1078                         } else if (BTRFS_EXTENT_DATA_KEY == type &&
1079                                    state->include_extent_data) {
1080                                 sf->error = btrfsic_handle_extent_data(
1081                                                 state,
1082                                                 sf->block,
1083                                                 sf->block_ctx,
1084                                                 item_offset,
1085                                                 force_iodone_flag);
1086                                 if (sf->error)
1087                                         goto one_stack_frame_backwards;
1088                         }
1089
1090                         goto continue_with_current_leaf_stack_frame;
1091                 }
1092         } else {
1093                 struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1094
1095                 if (-1 == sf->i) {
1096                         sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1097
1098                         if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1099                                 pr_info("node %llu level %d items %d generation %llu owner %llu\n",
1100                                        sf->block_ctx->start,
1101                                        nodehdr->header.level, sf->nr,
1102                                        btrfs_stack_header_generation(
1103                                        &nodehdr->header),
1104                                        btrfs_stack_header_owner(
1105                                        &nodehdr->header));
1106                 }
1107
1108 continue_with_current_node_stack_frame:
1109                 if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1110                         sf->i++;
1111                         sf->num_copies = 0;
1112                 }
1113
1114                 if (sf->i < sf->nr) {
1115                         struct btrfs_key_ptr key_ptr;
1116                         u32 key_ptr_offset;
1117                         u64 next_bytenr;
1118
1119                         key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1120                                           (uintptr_t)nodehdr;
1121                         if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1122                             sf->block_ctx->len) {
1123                                 pr_info("btrfsic: node item out of bounce at logical %llu, dev %s\n",
1124                                        sf->block_ctx->start,
1125                                        sf->block_ctx->dev->name);
1126                                 goto one_stack_frame_backwards;
1127                         }
1128                         btrfsic_read_from_block_data(
1129                                 sf->block_ctx, &key_ptr, key_ptr_offset,
1130                                 sizeof(struct btrfs_key_ptr));
1131                         next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1132
1133                         sf->error = btrfsic_create_link_to_next_block(
1134                                         state,
1135                                         sf->block,
1136                                         sf->block_ctx,
1137                                         next_bytenr,
1138                                         sf->limit_nesting,
1139                                         &sf->next_block_ctx,
1140                                         &sf->next_block,
1141                                         force_iodone_flag,
1142                                         &sf->num_copies,
1143                                         &sf->mirror_num,
1144                                         &key_ptr.key,
1145                                         btrfs_stack_key_generation(&key_ptr));
1146                         if (sf->error)
1147                                 goto one_stack_frame_backwards;
1148
1149                         if (NULL != sf->next_block) {
1150                                 struct btrfs_header *const next_hdr =
1151                                     (struct btrfs_header *)
1152                                     sf->next_block_ctx.datav[0];
1153
1154                                 next_stack = btrfsic_stack_frame_alloc();
1155                                 if (NULL == next_stack) {
1156                                         sf->error = -1;
1157                                         goto one_stack_frame_backwards;
1158                                 }
1159
1160                                 next_stack->i = -1;
1161                                 next_stack->block = sf->next_block;
1162                                 next_stack->block_ctx = &sf->next_block_ctx;
1163                                 next_stack->next_block = NULL;
1164                                 next_stack->hdr = next_hdr;
1165                                 next_stack->limit_nesting =
1166                                     sf->limit_nesting - 1;
1167                                 next_stack->prev = sf;
1168                                 sf = next_stack;
1169                                 goto continue_with_new_stack_frame;
1170                         }
1171
1172                         goto continue_with_current_node_stack_frame;
1173                 }
1174         }
1175
1176 one_stack_frame_backwards:
1177         if (NULL != sf->prev) {
1178                 struct btrfsic_stack_frame *const prev = sf->prev;
1179
1180                 /* the one for the initial block is freed in the caller */
1181                 btrfsic_release_block_ctx(sf->block_ctx);
1182
1183                 if (sf->error) {
1184                         prev->error = sf->error;
1185                         btrfsic_stack_frame_free(sf);
1186                         sf = prev;
1187                         goto one_stack_frame_backwards;
1188                 }
1189
1190                 btrfsic_stack_frame_free(sf);
1191                 sf = prev;
1192                 goto continue_with_new_stack_frame;
1193         } else {
1194                 BUG_ON(&initial_stack_frame != sf);
1195         }
1196
1197         return sf->error;
1198 }
1199
1200 static void btrfsic_read_from_block_data(
1201         struct btrfsic_block_data_ctx *block_ctx,
1202         void *dstv, u32 offset, size_t len)
1203 {
1204         size_t cur;
1205         size_t offset_in_page;
1206         char *kaddr;
1207         char *dst = (char *)dstv;
1208         size_t start_offset = block_ctx->start & ((u64)PAGE_SIZE - 1);
1209         unsigned long i = (start_offset + offset) >> PAGE_SHIFT;
1210
1211         WARN_ON(offset + len > block_ctx->len);
1212         offset_in_page = (start_offset + offset) & (PAGE_SIZE - 1);
1213
1214         while (len > 0) {
1215                 cur = min(len, ((size_t)PAGE_SIZE - offset_in_page));
1216                 BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_SIZE));
1217                 kaddr = block_ctx->datav[i];
1218                 memcpy(dst, kaddr + offset_in_page, cur);
1219
1220                 dst += cur;
1221                 len -= cur;
1222                 offset_in_page = 0;
1223                 i++;
1224         }
1225 }
1226
1227 static int btrfsic_create_link_to_next_block(
1228                 struct btrfsic_state *state,
1229                 struct btrfsic_block *block,
1230                 struct btrfsic_block_data_ctx *block_ctx,
1231                 u64 next_bytenr,
1232                 int limit_nesting,
1233                 struct btrfsic_block_data_ctx *next_block_ctx,
1234                 struct btrfsic_block **next_blockp,
1235                 int force_iodone_flag,
1236                 int *num_copiesp, int *mirror_nump,
1237                 struct btrfs_disk_key *disk_key,
1238                 u64 parent_generation)
1239 {
1240         struct btrfs_fs_info *fs_info = state->fs_info;
1241         struct btrfsic_block *next_block = NULL;
1242         int ret;
1243         struct btrfsic_block_link *l;
1244         int did_alloc_block_link;
1245         int block_was_created;
1246
1247         *next_blockp = NULL;
1248         if (0 == *num_copiesp) {
1249                 *num_copiesp = btrfs_num_copies(fs_info, next_bytenr,
1250                                                 state->metablock_size);
1251                 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1252                         pr_info("num_copies(log_bytenr=%llu) = %d\n",
1253                                next_bytenr, *num_copiesp);
1254                 *mirror_nump = 1;
1255         }
1256
1257         if (*mirror_nump > *num_copiesp)
1258                 return 0;
1259
1260         if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1261                 pr_info("btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1262                        *mirror_nump);
1263         ret = btrfsic_map_block(state, next_bytenr,
1264                                 state->metablock_size,
1265                                 next_block_ctx, *mirror_nump);
1266         if (ret) {
1267                 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1268                        next_bytenr, *mirror_nump);
1269                 btrfsic_release_block_ctx(next_block_ctx);
1270                 *next_blockp = NULL;
1271                 return -1;
1272         }
1273
1274         next_block = btrfsic_block_lookup_or_add(state,
1275                                                  next_block_ctx, "referenced ",
1276                                                  1, force_iodone_flag,
1277                                                  !force_iodone_flag,
1278                                                  *mirror_nump,
1279                                                  &block_was_created);
1280         if (NULL == next_block) {
1281                 btrfsic_release_block_ctx(next_block_ctx);
1282                 *next_blockp = NULL;
1283                 return -1;
1284         }
1285         if (block_was_created) {
1286                 l = NULL;
1287                 next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1288         } else {
1289                 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1290                         if (next_block->logical_bytenr != next_bytenr &&
1291                             !(!next_block->is_metadata &&
1292                               0 == next_block->logical_bytenr))
1293                                 pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1294                                        next_bytenr, next_block_ctx->dev->name,
1295                                        next_block_ctx->dev_bytenr, *mirror_nump,
1296                                        btrfsic_get_block_type(state,
1297                                                               next_block),
1298                                        next_block->logical_bytenr);
1299                         else
1300                                 pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1301                                        next_bytenr, next_block_ctx->dev->name,
1302                                        next_block_ctx->dev_bytenr, *mirror_nump,
1303                                        btrfsic_get_block_type(state,
1304                                                               next_block));
1305                 }
1306                 next_block->logical_bytenr = next_bytenr;
1307
1308                 next_block->mirror_num = *mirror_nump;
1309                 l = btrfsic_block_link_hashtable_lookup(
1310                                 next_block_ctx->dev->bdev,
1311                                 next_block_ctx->dev_bytenr,
1312                                 block_ctx->dev->bdev,
1313                                 block_ctx->dev_bytenr,
1314                                 &state->block_link_hashtable);
1315         }
1316
1317         next_block->disk_key = *disk_key;
1318         if (NULL == l) {
1319                 l = btrfsic_block_link_alloc();
1320                 if (NULL == l) {
1321                         pr_info("btrfsic: error, kmalloc failed!\n");
1322                         btrfsic_release_block_ctx(next_block_ctx);
1323                         *next_blockp = NULL;
1324                         return -1;
1325                 }
1326
1327                 did_alloc_block_link = 1;
1328                 l->block_ref_to = next_block;
1329                 l->block_ref_from = block;
1330                 l->ref_cnt = 1;
1331                 l->parent_generation = parent_generation;
1332
1333                 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1334                         btrfsic_print_add_link(state, l);
1335
1336                 list_add(&l->node_ref_to, &block->ref_to_list);
1337                 list_add(&l->node_ref_from, &next_block->ref_from_list);
1338
1339                 btrfsic_block_link_hashtable_add(l,
1340                                                  &state->block_link_hashtable);
1341         } else {
1342                 did_alloc_block_link = 0;
1343                 if (0 == limit_nesting) {
1344                         l->ref_cnt++;
1345                         l->parent_generation = parent_generation;
1346                         if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1347                                 btrfsic_print_add_link(state, l);
1348                 }
1349         }
1350
1351         if (limit_nesting > 0 && did_alloc_block_link) {
1352                 ret = btrfsic_read_block(state, next_block_ctx);
1353                 if (ret < (int)next_block_ctx->len) {
1354                         pr_info("btrfsic: read block @logical %llu failed!\n",
1355                                next_bytenr);
1356                         btrfsic_release_block_ctx(next_block_ctx);
1357                         *next_blockp = NULL;
1358                         return -1;
1359                 }
1360
1361                 *next_blockp = next_block;
1362         } else {
1363                 *next_blockp = NULL;
1364         }
1365         (*mirror_nump)++;
1366
1367         return 0;
1368 }
1369
1370 static int btrfsic_handle_extent_data(
1371                 struct btrfsic_state *state,
1372                 struct btrfsic_block *block,
1373                 struct btrfsic_block_data_ctx *block_ctx,
1374                 u32 item_offset, int force_iodone_flag)
1375 {
1376         struct btrfs_fs_info *fs_info = state->fs_info;
1377         struct btrfs_file_extent_item file_extent_item;
1378         u64 file_extent_item_offset;
1379         u64 next_bytenr;
1380         u64 num_bytes;
1381         u64 generation;
1382         struct btrfsic_block_link *l;
1383         int ret;
1384
1385         file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1386                                   item_offset;
1387         if (file_extent_item_offset +
1388             offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1389             block_ctx->len) {
1390                 pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1391                        block_ctx->start, block_ctx->dev->name);
1392                 return -1;
1393         }
1394
1395         btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1396                 file_extent_item_offset,
1397                 offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1398         if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1399             btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1400                 if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1401                         pr_info("extent_data: type %u, disk_bytenr = %llu\n",
1402                                file_extent_item.type,
1403                                btrfs_stack_file_extent_disk_bytenr(
1404                                &file_extent_item));
1405                 return 0;
1406         }
1407
1408         if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1409             block_ctx->len) {
1410                 pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1411                        block_ctx->start, block_ctx->dev->name);
1412                 return -1;
1413         }
1414         btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1415                                      file_extent_item_offset,
1416                                      sizeof(struct btrfs_file_extent_item));
1417         next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item);
1418         if (btrfs_stack_file_extent_compression(&file_extent_item) ==
1419             BTRFS_COMPRESS_NONE) {
1420                 next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item);
1421                 num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1422         } else {
1423                 num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item);
1424         }
1425         generation = btrfs_stack_file_extent_generation(&file_extent_item);
1426
1427         if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1428                 pr_info("extent_data: type %u, disk_bytenr = %llu, offset = %llu, num_bytes = %llu\n",
1429                        file_extent_item.type,
1430                        btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1431                        btrfs_stack_file_extent_offset(&file_extent_item),
1432                        num_bytes);
1433         while (num_bytes > 0) {
1434                 u32 chunk_len;
1435                 int num_copies;
1436                 int mirror_num;
1437
1438                 if (num_bytes > state->datablock_size)
1439                         chunk_len = state->datablock_size;
1440                 else
1441                         chunk_len = num_bytes;
1442
1443                 num_copies = btrfs_num_copies(fs_info, next_bytenr,
1444                                               state->datablock_size);
1445                 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1446                         pr_info("num_copies(log_bytenr=%llu) = %d\n",
1447                                next_bytenr, num_copies);
1448                 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1449                         struct btrfsic_block_data_ctx next_block_ctx;
1450                         struct btrfsic_block *next_block;
1451                         int block_was_created;
1452
1453                         if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1454                                 pr_info("btrfsic_handle_extent_data(mirror_num=%d)\n",
1455                                         mirror_num);
1456                         if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1457                                 pr_info("\tdisk_bytenr = %llu, num_bytes %u\n",
1458                                        next_bytenr, chunk_len);
1459                         ret = btrfsic_map_block(state, next_bytenr,
1460                                                 chunk_len, &next_block_ctx,
1461                                                 mirror_num);
1462                         if (ret) {
1463                                 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1464                                        next_bytenr, mirror_num);
1465                                 return -1;
1466                         }
1467
1468                         next_block = btrfsic_block_lookup_or_add(
1469                                         state,
1470                                         &next_block_ctx,
1471                                         "referenced ",
1472                                         0,
1473                                         force_iodone_flag,
1474                                         !force_iodone_flag,
1475                                         mirror_num,
1476                                         &block_was_created);
1477                         if (NULL == next_block) {
1478                                 pr_info("btrfsic: error, kmalloc failed!\n");
1479                                 btrfsic_release_block_ctx(&next_block_ctx);
1480                                 return -1;
1481                         }
1482                         if (!block_was_created) {
1483                                 if ((state->print_mask &
1484                                      BTRFSIC_PRINT_MASK_VERBOSE) &&
1485                                     next_block->logical_bytenr != next_bytenr &&
1486                                     !(!next_block->is_metadata &&
1487                                       0 == next_block->logical_bytenr)) {
1488                                         pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, D, bytenr mismatch (!= stored %llu).\n",
1489                                                next_bytenr,
1490                                                next_block_ctx.dev->name,
1491                                                next_block_ctx.dev_bytenr,
1492                                                mirror_num,
1493                                                next_block->logical_bytenr);
1494                                 }
1495                                 next_block->logical_bytenr = next_bytenr;
1496                                 next_block->mirror_num = mirror_num;
1497                         }
1498
1499                         l = btrfsic_block_link_lookup_or_add(state,
1500                                                              &next_block_ctx,
1501                                                              next_block, block,
1502                                                              generation);
1503                         btrfsic_release_block_ctx(&next_block_ctx);
1504                         if (NULL == l)
1505                                 return -1;
1506                 }
1507
1508                 next_bytenr += chunk_len;
1509                 num_bytes -= chunk_len;
1510         }
1511
1512         return 0;
1513 }
1514
1515 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1516                              struct btrfsic_block_data_ctx *block_ctx_out,
1517                              int mirror_num)
1518 {
1519         struct btrfs_fs_info *fs_info = state->fs_info;
1520         int ret;
1521         u64 length;
1522         struct btrfs_bio *multi = NULL;
1523         struct btrfs_device *device;
1524
1525         length = len;
1526         ret = btrfs_map_block(fs_info, BTRFS_MAP_READ,
1527                               bytenr, &length, &multi, mirror_num);
1528
1529         if (ret) {
1530                 block_ctx_out->start = 0;
1531                 block_ctx_out->dev_bytenr = 0;
1532                 block_ctx_out->len = 0;
1533                 block_ctx_out->dev = NULL;
1534                 block_ctx_out->datav = NULL;
1535                 block_ctx_out->pagev = NULL;
1536                 block_ctx_out->mem_to_free = NULL;
1537
1538                 return ret;
1539         }
1540
1541         device = multi->stripes[0].dev;
1542         if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
1543             !device->bdev || !device->name)
1544                 block_ctx_out->dev = NULL;
1545         else
1546                 block_ctx_out->dev = btrfsic_dev_state_lookup(
1547                                                         device->bdev->bd_dev);
1548         block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1549         block_ctx_out->start = bytenr;
1550         block_ctx_out->len = len;
1551         block_ctx_out->datav = NULL;
1552         block_ctx_out->pagev = NULL;
1553         block_ctx_out->mem_to_free = NULL;
1554
1555         kfree(multi);
1556         if (NULL == block_ctx_out->dev) {
1557                 ret = -ENXIO;
1558                 pr_info("btrfsic: error, cannot lookup dev (#1)!\n");
1559         }
1560
1561         return ret;
1562 }
1563
1564 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1565 {
1566         if (block_ctx->mem_to_free) {
1567                 unsigned int num_pages;
1568
1569                 BUG_ON(!block_ctx->datav);
1570                 BUG_ON(!block_ctx->pagev);
1571                 num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1572                             PAGE_SHIFT;
1573                 while (num_pages > 0) {
1574                         num_pages--;
1575                         if (block_ctx->datav[num_pages]) {
1576                                 kunmap(block_ctx->pagev[num_pages]);
1577                                 block_ctx->datav[num_pages] = NULL;
1578                         }
1579                         if (block_ctx->pagev[num_pages]) {
1580                                 __free_page(block_ctx->pagev[num_pages]);
1581                                 block_ctx->pagev[num_pages] = NULL;
1582                         }
1583                 }
1584
1585                 kfree(block_ctx->mem_to_free);
1586                 block_ctx->mem_to_free = NULL;
1587                 block_ctx->pagev = NULL;
1588                 block_ctx->datav = NULL;
1589         }
1590 }
1591
1592 static int btrfsic_read_block(struct btrfsic_state *state,
1593                               struct btrfsic_block_data_ctx *block_ctx)
1594 {
1595         unsigned int num_pages;
1596         unsigned int i;
1597         u64 dev_bytenr;
1598         int ret;
1599
1600         BUG_ON(block_ctx->datav);
1601         BUG_ON(block_ctx->pagev);
1602         BUG_ON(block_ctx->mem_to_free);
1603         if (block_ctx->dev_bytenr & ((u64)PAGE_SIZE - 1)) {
1604                 pr_info("btrfsic: read_block() with unaligned bytenr %llu\n",
1605                        block_ctx->dev_bytenr);
1606                 return -1;
1607         }
1608
1609         num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1610                     PAGE_SHIFT;
1611         block_ctx->mem_to_free = kcalloc(sizeof(*block_ctx->datav) +
1612                                                 sizeof(*block_ctx->pagev),
1613                                          num_pages, GFP_NOFS);
1614         if (!block_ctx->mem_to_free)
1615                 return -ENOMEM;
1616         block_ctx->datav = block_ctx->mem_to_free;
1617         block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1618         for (i = 0; i < num_pages; i++) {
1619                 block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1620                 if (!block_ctx->pagev[i])
1621                         return -1;
1622         }
1623
1624         dev_bytenr = block_ctx->dev_bytenr;
1625         for (i = 0; i < num_pages;) {
1626                 struct bio *bio;
1627                 unsigned int j;
1628
1629                 bio = btrfs_io_bio_alloc(num_pages - i);
1630                 bio_set_dev(bio, block_ctx->dev->bdev);
1631                 bio->bi_iter.bi_sector = dev_bytenr >> 9;
1632                 bio->bi_opf = REQ_OP_READ;
1633
1634                 for (j = i; j < num_pages; j++) {
1635                         ret = bio_add_page(bio, block_ctx->pagev[j],
1636                                            PAGE_SIZE, 0);
1637                         if (PAGE_SIZE != ret)
1638                                 break;
1639                 }
1640                 if (j == i) {
1641                         pr_info("btrfsic: error, failed to add a single page!\n");
1642                         return -1;
1643                 }
1644                 if (submit_bio_wait(bio)) {
1645                         pr_info("btrfsic: read error at logical %llu dev %s!\n",
1646                                block_ctx->start, block_ctx->dev->name);
1647                         bio_put(bio);
1648                         return -1;
1649                 }
1650                 bio_put(bio);
1651                 dev_bytenr += (j - i) * PAGE_SIZE;
1652                 i = j;
1653         }
1654         for (i = 0; i < num_pages; i++)
1655                 block_ctx->datav[i] = kmap(block_ctx->pagev[i]);
1656
1657         return block_ctx->len;
1658 }
1659
1660 static void btrfsic_dump_database(struct btrfsic_state *state)
1661 {
1662         const struct btrfsic_block *b_all;
1663
1664         BUG_ON(NULL == state);
1665
1666         pr_info("all_blocks_list:\n");
1667         list_for_each_entry(b_all, &state->all_blocks_list, all_blocks_node) {
1668                 const struct btrfsic_block_link *l;
1669
1670                 pr_info("%c-block @%llu (%s/%llu/%d)\n",
1671                        btrfsic_get_block_type(state, b_all),
1672                        b_all->logical_bytenr, b_all->dev_state->name,
1673                        b_all->dev_bytenr, b_all->mirror_num);
1674
1675                 list_for_each_entry(l, &b_all->ref_to_list, node_ref_to) {
1676                         pr_info(" %c @%llu (%s/%llu/%d) refers %u* to %c @%llu (%s/%llu/%d)\n",
1677                                btrfsic_get_block_type(state, b_all),
1678                                b_all->logical_bytenr, b_all->dev_state->name,
1679                                b_all->dev_bytenr, b_all->mirror_num,
1680                                l->ref_cnt,
1681                                btrfsic_get_block_type(state, l->block_ref_to),
1682                                l->block_ref_to->logical_bytenr,
1683                                l->block_ref_to->dev_state->name,
1684                                l->block_ref_to->dev_bytenr,
1685                                l->block_ref_to->mirror_num);
1686                 }
1687
1688                 list_for_each_entry(l, &b_all->ref_from_list, node_ref_from) {
1689                         pr_info(" %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
1690                                btrfsic_get_block_type(state, b_all),
1691                                b_all->logical_bytenr, b_all->dev_state->name,
1692                                b_all->dev_bytenr, b_all->mirror_num,
1693                                l->ref_cnt,
1694                                btrfsic_get_block_type(state, l->block_ref_from),
1695                                l->block_ref_from->logical_bytenr,
1696                                l->block_ref_from->dev_state->name,
1697                                l->block_ref_from->dev_bytenr,
1698                                l->block_ref_from->mirror_num);
1699                 }
1700
1701                 pr_info("\n");
1702         }
1703 }
1704
1705 /*
1706  * Test whether the disk block contains a tree block (leaf or node)
1707  * (note that this test fails for the super block)
1708  */
1709 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
1710                                      char **datav, unsigned int num_pages)
1711 {
1712         struct btrfs_fs_info *fs_info = state->fs_info;
1713         struct btrfs_header *h;
1714         u8 csum[BTRFS_CSUM_SIZE];
1715         u32 crc = ~(u32)0;
1716         unsigned int i;
1717
1718         if (num_pages * PAGE_SIZE < state->metablock_size)
1719                 return 1; /* not metadata */
1720         num_pages = state->metablock_size >> PAGE_SHIFT;
1721         h = (struct btrfs_header *)datav[0];
1722
1723         if (memcmp(h->fsid, fs_info->fsid, BTRFS_FSID_SIZE))
1724                 return 1;
1725
1726         for (i = 0; i < num_pages; i++) {
1727                 u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1728                 size_t sublen = i ? PAGE_SIZE :
1729                                     (PAGE_SIZE - BTRFS_CSUM_SIZE);
1730
1731                 crc = crc32c(crc, data, sublen);
1732         }
1733         btrfs_csum_final(crc, csum);
1734         if (memcmp(csum, h->csum, state->csum_size))
1735                 return 1;
1736
1737         return 0; /* is metadata */
1738 }
1739
1740 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1741                                           u64 dev_bytenr, char **mapped_datav,
1742                                           unsigned int num_pages,
1743                                           struct bio *bio, int *bio_is_patched,
1744                                           struct buffer_head *bh,
1745                                           int submit_bio_bh_rw)
1746 {
1747         int is_metadata;
1748         struct btrfsic_block *block;
1749         struct btrfsic_block_data_ctx block_ctx;
1750         int ret;
1751         struct btrfsic_state *state = dev_state->state;
1752         struct block_device *bdev = dev_state->bdev;
1753         unsigned int processed_len;
1754
1755         if (NULL != bio_is_patched)
1756                 *bio_is_patched = 0;
1757
1758 again:
1759         if (num_pages == 0)
1760                 return;
1761
1762         processed_len = 0;
1763         is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1764                                                       num_pages));
1765
1766         block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1767                                                &state->block_hashtable);
1768         if (NULL != block) {
1769                 u64 bytenr = 0;
1770                 struct btrfsic_block_link *l, *tmp;
1771
1772                 if (block->is_superblock) {
1773                         bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1774                                                     mapped_datav[0]);
1775                         if (num_pages * PAGE_SIZE <
1776                             BTRFS_SUPER_INFO_SIZE) {
1777                                 pr_info("btrfsic: cannot work with too short bios!\n");
1778                                 return;
1779                         }
1780                         is_metadata = 1;
1781                         BUG_ON(BTRFS_SUPER_INFO_SIZE & (PAGE_SIZE - 1));
1782                         processed_len = BTRFS_SUPER_INFO_SIZE;
1783                         if (state->print_mask &
1784                             BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1785                                 pr_info("[before new superblock is written]:\n");
1786                                 btrfsic_dump_tree_sub(state, block, 0);
1787                         }
1788                 }
1789                 if (is_metadata) {
1790                         if (!block->is_superblock) {
1791                                 if (num_pages * PAGE_SIZE <
1792                                     state->metablock_size) {
1793                                         pr_info("btrfsic: cannot work with too short bios!\n");
1794                                         return;
1795                                 }
1796                                 processed_len = state->metablock_size;
1797                                 bytenr = btrfs_stack_header_bytenr(
1798                                                 (struct btrfs_header *)
1799                                                 mapped_datav[0]);
1800                                 btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1801                                                                dev_state,
1802                                                                dev_bytenr);
1803                         }
1804                         if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1805                                 if (block->logical_bytenr != bytenr &&
1806                                     !(!block->is_metadata &&
1807                                       block->logical_bytenr == 0))
1808                                         pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1809                                                bytenr, dev_state->name,
1810                                                dev_bytenr,
1811                                                block->mirror_num,
1812                                                btrfsic_get_block_type(state,
1813                                                                       block),
1814                                                block->logical_bytenr);
1815                                 else
1816                                         pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1817                                                bytenr, dev_state->name,
1818                                                dev_bytenr, block->mirror_num,
1819                                                btrfsic_get_block_type(state,
1820                                                                       block));
1821                         }
1822                         block->logical_bytenr = bytenr;
1823                 } else {
1824                         if (num_pages * PAGE_SIZE <
1825                             state->datablock_size) {
1826                                 pr_info("btrfsic: cannot work with too short bios!\n");
1827                                 return;
1828                         }
1829                         processed_len = state->datablock_size;
1830                         bytenr = block->logical_bytenr;
1831                         if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1832                                 pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1833                                        bytenr, dev_state->name, dev_bytenr,
1834                                        block->mirror_num,
1835                                        btrfsic_get_block_type(state, block));
1836                 }
1837
1838                 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1839                         pr_info("ref_to_list: %cE, ref_from_list: %cE\n",
1840                                list_empty(&block->ref_to_list) ? ' ' : '!',
1841                                list_empty(&block->ref_from_list) ? ' ' : '!');
1842                 if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1843                         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",
1844                                btrfsic_get_block_type(state, block), bytenr,
1845                                dev_state->name, dev_bytenr, block->mirror_num,
1846                                block->generation,
1847                                btrfs_disk_key_objectid(&block->disk_key),
1848                                block->disk_key.type,
1849                                btrfs_disk_key_offset(&block->disk_key),
1850                                btrfs_stack_header_generation(
1851                                        (struct btrfs_header *) mapped_datav[0]),
1852                                state->max_superblock_generation);
1853                         btrfsic_dump_tree(state);
1854                 }
1855
1856                 if (!block->is_iodone && !block->never_written) {
1857                         pr_info("btrfs: attempt to overwrite %c-block @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu, which is not yet iodone!\n",
1858                                btrfsic_get_block_type(state, block), bytenr,
1859                                dev_state->name, dev_bytenr, block->mirror_num,
1860                                block->generation,
1861                                btrfs_stack_header_generation(
1862                                        (struct btrfs_header *)
1863                                        mapped_datav[0]));
1864                         /* it would not be safe to go on */
1865                         btrfsic_dump_tree(state);
1866                         goto continue_loop;
1867                 }
1868
1869                 /*
1870                  * Clear all references of this block. Do not free
1871                  * the block itself even if is not referenced anymore
1872                  * because it still carries valuable information
1873                  * like whether it was ever written and IO completed.
1874                  */
1875                 list_for_each_entry_safe(l, tmp, &block->ref_to_list,
1876                                          node_ref_to) {
1877                         if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1878                                 btrfsic_print_rem_link(state, l);
1879                         l->ref_cnt--;
1880                         if (0 == l->ref_cnt) {
1881                                 list_del(&l->node_ref_to);
1882                                 list_del(&l->node_ref_from);
1883                                 btrfsic_block_link_hashtable_remove(l);
1884                                 btrfsic_block_link_free(l);
1885                         }
1886                 }
1887
1888                 block_ctx.dev = dev_state;
1889                 block_ctx.dev_bytenr = dev_bytenr;
1890                 block_ctx.start = bytenr;
1891                 block_ctx.len = processed_len;
1892                 block_ctx.pagev = NULL;
1893                 block_ctx.mem_to_free = NULL;
1894                 block_ctx.datav = mapped_datav;
1895
1896                 if (is_metadata || state->include_extent_data) {
1897                         block->never_written = 0;
1898                         block->iodone_w_error = 0;
1899                         if (NULL != bio) {
1900                                 block->is_iodone = 0;
1901                                 BUG_ON(NULL == bio_is_patched);
1902                                 if (!*bio_is_patched) {
1903                                         block->orig_bio_bh_private =
1904                                             bio->bi_private;
1905                                         block->orig_bio_bh_end_io.bio =
1906                                             bio->bi_end_io;
1907                                         block->next_in_same_bio = NULL;
1908                                         bio->bi_private = block;
1909                                         bio->bi_end_io = btrfsic_bio_end_io;
1910                                         *bio_is_patched = 1;
1911                                 } else {
1912                                         struct btrfsic_block *chained_block =
1913                                             (struct btrfsic_block *)
1914                                             bio->bi_private;
1915
1916                                         BUG_ON(NULL == chained_block);
1917                                         block->orig_bio_bh_private =
1918                                             chained_block->orig_bio_bh_private;
1919                                         block->orig_bio_bh_end_io.bio =
1920                                             chained_block->orig_bio_bh_end_io.
1921                                             bio;
1922                                         block->next_in_same_bio = chained_block;
1923                                         bio->bi_private = block;
1924                                 }
1925                         } else if (NULL != bh) {
1926                                 block->is_iodone = 0;
1927                                 block->orig_bio_bh_private = bh->b_private;
1928                                 block->orig_bio_bh_end_io.bh = bh->b_end_io;
1929                                 block->next_in_same_bio = NULL;
1930                                 bh->b_private = block;
1931                                 bh->b_end_io = btrfsic_bh_end_io;
1932                         } else {
1933                                 block->is_iodone = 1;
1934                                 block->orig_bio_bh_private = NULL;
1935                                 block->orig_bio_bh_end_io.bio = NULL;
1936                                 block->next_in_same_bio = NULL;
1937                         }
1938                 }
1939
1940                 block->flush_gen = dev_state->last_flush_gen + 1;
1941                 block->submit_bio_bh_rw = submit_bio_bh_rw;
1942                 if (is_metadata) {
1943                         block->logical_bytenr = bytenr;
1944                         block->is_metadata = 1;
1945                         if (block->is_superblock) {
1946                                 BUG_ON(PAGE_SIZE !=
1947                                        BTRFS_SUPER_INFO_SIZE);
1948                                 ret = btrfsic_process_written_superblock(
1949                                                 state,
1950                                                 block,
1951                                                 (struct btrfs_super_block *)
1952                                                 mapped_datav[0]);
1953                                 if (state->print_mask &
1954                                     BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
1955                                         pr_info("[after new superblock is written]:\n");
1956                                         btrfsic_dump_tree_sub(state, block, 0);
1957                                 }
1958                         } else {
1959                                 block->mirror_num = 0;  /* unknown */
1960                                 ret = btrfsic_process_metablock(
1961                                                 state,
1962                                                 block,
1963                                                 &block_ctx,
1964                                                 0, 0);
1965                         }
1966                         if (ret)
1967                                 pr_info("btrfsic: btrfsic_process_metablock(root @%llu) failed!\n",
1968                                        dev_bytenr);
1969                 } else {
1970                         block->is_metadata = 0;
1971                         block->mirror_num = 0;  /* unknown */
1972                         block->generation = BTRFSIC_GENERATION_UNKNOWN;
1973                         if (!state->include_extent_data
1974                             && list_empty(&block->ref_from_list)) {
1975                                 /*
1976                                  * disk block is overwritten with extent
1977                                  * data (not meta data) and we are configured
1978                                  * to not include extent data: take the
1979                                  * chance and free the block's memory
1980                                  */
1981                                 btrfsic_block_hashtable_remove(block);
1982                                 list_del(&block->all_blocks_node);
1983                                 btrfsic_block_free(block);
1984                         }
1985                 }
1986                 btrfsic_release_block_ctx(&block_ctx);
1987         } else {
1988                 /* block has not been found in hash table */
1989                 u64 bytenr;
1990
1991                 if (!is_metadata) {
1992                         processed_len = state->datablock_size;
1993                         if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1994                                 pr_info("Written block (%s/%llu/?) !found in hash table, D.\n",
1995                                        dev_state->name, dev_bytenr);
1996                         if (!state->include_extent_data) {
1997                                 /* ignore that written D block */
1998                                 goto continue_loop;
1999                         }
2000
2001                         /* this is getting ugly for the
2002                          * include_extent_data case... */
2003                         bytenr = 0;     /* unknown */
2004                 } else {
2005                         processed_len = state->metablock_size;
2006                         bytenr = btrfs_stack_header_bytenr(
2007                                         (struct btrfs_header *)
2008                                         mapped_datav[0]);
2009                         btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
2010                                                        dev_bytenr);
2011                         if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2012                                 pr_info("Written block @%llu (%s/%llu/?) !found in hash table, M.\n",
2013                                        bytenr, dev_state->name, dev_bytenr);
2014                 }
2015
2016                 block_ctx.dev = dev_state;
2017                 block_ctx.dev_bytenr = dev_bytenr;
2018                 block_ctx.start = bytenr;
2019                 block_ctx.len = processed_len;
2020                 block_ctx.pagev = NULL;
2021                 block_ctx.mem_to_free = NULL;
2022                 block_ctx.datav = mapped_datav;
2023
2024                 block = btrfsic_block_alloc();
2025                 if (NULL == block) {
2026                         pr_info("btrfsic: error, kmalloc failed!\n");
2027                         btrfsic_release_block_ctx(&block_ctx);
2028                         goto continue_loop;
2029                 }
2030                 block->dev_state = dev_state;
2031                 block->dev_bytenr = dev_bytenr;
2032                 block->logical_bytenr = bytenr;
2033                 block->is_metadata = is_metadata;
2034                 block->never_written = 0;
2035                 block->iodone_w_error = 0;
2036                 block->mirror_num = 0;  /* unknown */
2037                 block->flush_gen = dev_state->last_flush_gen + 1;
2038                 block->submit_bio_bh_rw = submit_bio_bh_rw;
2039                 if (NULL != bio) {
2040                         block->is_iodone = 0;
2041                         BUG_ON(NULL == bio_is_patched);
2042                         if (!*bio_is_patched) {
2043                                 block->orig_bio_bh_private = bio->bi_private;
2044                                 block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2045                                 block->next_in_same_bio = NULL;
2046                                 bio->bi_private = block;
2047                                 bio->bi_end_io = btrfsic_bio_end_io;
2048                                 *bio_is_patched = 1;
2049                         } else {
2050                                 struct btrfsic_block *chained_block =
2051                                     (struct btrfsic_block *)
2052                                     bio->bi_private;
2053
2054                                 BUG_ON(NULL == chained_block);
2055                                 block->orig_bio_bh_private =
2056                                     chained_block->orig_bio_bh_private;
2057                                 block->orig_bio_bh_end_io.bio =
2058                                     chained_block->orig_bio_bh_end_io.bio;
2059                                 block->next_in_same_bio = chained_block;
2060                                 bio->bi_private = block;
2061                         }
2062                 } else if (NULL != bh) {
2063                         block->is_iodone = 0;
2064                         block->orig_bio_bh_private = bh->b_private;
2065                         block->orig_bio_bh_end_io.bh = bh->b_end_io;
2066                         block->next_in_same_bio = NULL;
2067                         bh->b_private = block;
2068                         bh->b_end_io = btrfsic_bh_end_io;
2069                 } else {
2070                         block->is_iodone = 1;
2071                         block->orig_bio_bh_private = NULL;
2072                         block->orig_bio_bh_end_io.bio = NULL;
2073                         block->next_in_same_bio = NULL;
2074                 }
2075                 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2076                         pr_info("New written %c-block @%llu (%s/%llu/%d)\n",
2077                                is_metadata ? 'M' : 'D',
2078                                block->logical_bytenr, block->dev_state->name,
2079                                block->dev_bytenr, block->mirror_num);
2080                 list_add(&block->all_blocks_node, &state->all_blocks_list);
2081                 btrfsic_block_hashtable_add(block, &state->block_hashtable);
2082
2083                 if (is_metadata) {
2084                         ret = btrfsic_process_metablock(state, block,
2085                                                         &block_ctx, 0, 0);
2086                         if (ret)
2087                                 pr_info("btrfsic: process_metablock(root @%llu) failed!\n",
2088                                        dev_bytenr);
2089                 }
2090                 btrfsic_release_block_ctx(&block_ctx);
2091         }
2092
2093 continue_loop:
2094         BUG_ON(!processed_len);
2095         dev_bytenr += processed_len;
2096         mapped_datav += processed_len >> PAGE_SHIFT;
2097         num_pages -= processed_len >> PAGE_SHIFT;
2098         goto again;
2099 }
2100
2101 static void btrfsic_bio_end_io(struct bio *bp)
2102 {
2103         struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2104         int iodone_w_error;
2105
2106         /* mutex is not held! This is not save if IO is not yet completed
2107          * on umount */
2108         iodone_w_error = 0;
2109         if (bp->bi_status)
2110                 iodone_w_error = 1;
2111
2112         BUG_ON(NULL == block);
2113         bp->bi_private = block->orig_bio_bh_private;
2114         bp->bi_end_io = block->orig_bio_bh_end_io.bio;
2115
2116         do {
2117                 struct btrfsic_block *next_block;
2118                 struct btrfsic_dev_state *const dev_state = block->dev_state;
2119
2120                 if ((dev_state->state->print_mask &
2121                      BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2122                         pr_info("bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2123                                bp->bi_status,
2124                                btrfsic_get_block_type(dev_state->state, block),
2125                                block->logical_bytenr, dev_state->name,
2126                                block->dev_bytenr, block->mirror_num);
2127                 next_block = block->next_in_same_bio;
2128                 block->iodone_w_error = iodone_w_error;
2129                 if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2130                         dev_state->last_flush_gen++;
2131                         if ((dev_state->state->print_mask &
2132                              BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2133                                 pr_info("bio_end_io() new %s flush_gen=%llu\n",
2134                                        dev_state->name,
2135                                        dev_state->last_flush_gen);
2136                 }
2137                 if (block->submit_bio_bh_rw & REQ_FUA)
2138                         block->flush_gen = 0; /* FUA completed means block is
2139                                                * on disk */
2140                 block->is_iodone = 1; /* for FLUSH, this releases the block */
2141                 block = next_block;
2142         } while (NULL != block);
2143
2144         bp->bi_end_io(bp);
2145 }
2146
2147 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate)
2148 {
2149         struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private;
2150         int iodone_w_error = !uptodate;
2151         struct btrfsic_dev_state *dev_state;
2152
2153         BUG_ON(NULL == block);
2154         dev_state = block->dev_state;
2155         if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2156                 pr_info("bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n",
2157                        iodone_w_error,
2158                        btrfsic_get_block_type(dev_state->state, block),
2159                        block->logical_bytenr, block->dev_state->name,
2160                        block->dev_bytenr, block->mirror_num);
2161
2162         block->iodone_w_error = iodone_w_error;
2163         if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2164                 dev_state->last_flush_gen++;
2165                 if ((dev_state->state->print_mask &
2166                      BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2167                         pr_info("bh_end_io() new %s flush_gen=%llu\n",
2168                                dev_state->name, dev_state->last_flush_gen);
2169         }
2170         if (block->submit_bio_bh_rw & REQ_FUA)
2171                 block->flush_gen = 0; /* FUA completed means block is on disk */
2172
2173         bh->b_private = block->orig_bio_bh_private;
2174         bh->b_end_io = block->orig_bio_bh_end_io.bh;
2175         block->is_iodone = 1; /* for FLUSH, this releases the block */
2176         bh->b_end_io(bh, uptodate);
2177 }
2178
2179 static int btrfsic_process_written_superblock(
2180                 struct btrfsic_state *state,
2181                 struct btrfsic_block *const superblock,
2182                 struct btrfs_super_block *const super_hdr)
2183 {
2184         struct btrfs_fs_info *fs_info = state->fs_info;
2185         int pass;
2186
2187         superblock->generation = btrfs_super_generation(super_hdr);
2188         if (!(superblock->generation > state->max_superblock_generation ||
2189               0 == state->max_superblock_generation)) {
2190                 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2191                         pr_info("btrfsic: superblock @%llu (%s/%llu/%d) with old gen %llu <= %llu\n",
2192                                superblock->logical_bytenr,
2193                                superblock->dev_state->name,
2194                                superblock->dev_bytenr, superblock->mirror_num,
2195                                btrfs_super_generation(super_hdr),
2196                                state->max_superblock_generation);
2197         } else {
2198                 if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2199                         pr_info("btrfsic: got new superblock @%llu (%s/%llu/%d) with new gen %llu > %llu\n",
2200                                superblock->logical_bytenr,
2201                                superblock->dev_state->name,
2202                                superblock->dev_bytenr, superblock->mirror_num,
2203                                btrfs_super_generation(super_hdr),
2204                                state->max_superblock_generation);
2205
2206                 state->max_superblock_generation =
2207                     btrfs_super_generation(super_hdr);
2208                 state->latest_superblock = superblock;
2209         }
2210
2211         for (pass = 0; pass < 3; pass++) {
2212                 int ret;
2213                 u64 next_bytenr;
2214                 struct btrfsic_block *next_block;
2215                 struct btrfsic_block_data_ctx tmp_next_block_ctx;
2216                 struct btrfsic_block_link *l;
2217                 int num_copies;
2218                 int mirror_num;
2219                 const char *additional_string = NULL;
2220                 struct btrfs_disk_key tmp_disk_key = {0};
2221
2222                 btrfs_set_disk_key_objectid(&tmp_disk_key,
2223                                             BTRFS_ROOT_ITEM_KEY);
2224                 btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2225
2226                 switch (pass) {
2227                 case 0:
2228                         btrfs_set_disk_key_objectid(&tmp_disk_key,
2229                                                     BTRFS_ROOT_TREE_OBJECTID);
2230                         additional_string = "root ";
2231                         next_bytenr = btrfs_super_root(super_hdr);
2232                         if (state->print_mask &
2233                             BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2234                                 pr_info("root@%llu\n", next_bytenr);
2235                         break;
2236                 case 1:
2237                         btrfs_set_disk_key_objectid(&tmp_disk_key,
2238                                                     BTRFS_CHUNK_TREE_OBJECTID);
2239                         additional_string = "chunk ";
2240                         next_bytenr = btrfs_super_chunk_root(super_hdr);
2241                         if (state->print_mask &
2242                             BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2243                                 pr_info("chunk@%llu\n", next_bytenr);
2244                         break;
2245                 case 2:
2246                         btrfs_set_disk_key_objectid(&tmp_disk_key,
2247                                                     BTRFS_TREE_LOG_OBJECTID);
2248                         additional_string = "log ";
2249                         next_bytenr = btrfs_super_log_root(super_hdr);
2250                         if (0 == next_bytenr)
2251                                 continue;
2252                         if (state->print_mask &
2253                             BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2254                                 pr_info("log@%llu\n", next_bytenr);
2255                         break;
2256                 }
2257
2258                 num_copies = btrfs_num_copies(fs_info, next_bytenr,
2259                                               BTRFS_SUPER_INFO_SIZE);
2260                 if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2261                         pr_info("num_copies(log_bytenr=%llu) = %d\n",
2262                                next_bytenr, num_copies);
2263                 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2264                         int was_created;
2265
2266                         if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2267                                 pr_info("btrfsic_process_written_superblock(mirror_num=%d)\n", mirror_num);
2268                         ret = btrfsic_map_block(state, next_bytenr,
2269                                                 BTRFS_SUPER_INFO_SIZE,
2270                                                 &tmp_next_block_ctx,
2271                                                 mirror_num);
2272                         if (ret) {
2273                                 pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
2274                                        next_bytenr, mirror_num);
2275                                 return -1;
2276                         }
2277
2278                         next_block = btrfsic_block_lookup_or_add(
2279                                         state,
2280                                         &tmp_next_block_ctx,
2281                                         additional_string,
2282                                         1, 0, 1,
2283                                         mirror_num,
2284                                         &was_created);
2285                         if (NULL == next_block) {
2286                                 pr_info("btrfsic: error, kmalloc failed!\n");
2287                                 btrfsic_release_block_ctx(&tmp_next_block_ctx);
2288                                 return -1;
2289                         }
2290
2291                         next_block->disk_key = tmp_disk_key;
2292                         if (was_created)
2293                                 next_block->generation =
2294                                     BTRFSIC_GENERATION_UNKNOWN;
2295                         l = btrfsic_block_link_lookup_or_add(
2296                                         state,
2297                                         &tmp_next_block_ctx,
2298                                         next_block,
2299                                         superblock,
2300                                         BTRFSIC_GENERATION_UNKNOWN);
2301                         btrfsic_release_block_ctx(&tmp_next_block_ctx);
2302                         if (NULL == l)
2303                                 return -1;
2304                 }
2305         }
2306
2307         if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2308                 btrfsic_dump_tree(state);
2309
2310         return 0;
2311 }
2312
2313 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2314                                         struct btrfsic_block *const block,
2315                                         int recursion_level)
2316 {
2317         const struct btrfsic_block_link *l;
2318         int ret = 0;
2319
2320         if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2321                 /*
2322                  * Note that this situation can happen and does not
2323                  * indicate an error in regular cases. It happens
2324                  * when disk blocks are freed and later reused.
2325                  * The check-integrity module is not aware of any
2326                  * block free operations, it just recognizes block
2327                  * write operations. Therefore it keeps the linkage
2328                  * information for a block until a block is
2329                  * rewritten. This can temporarily cause incorrect
2330                  * and even circular linkage informations. This
2331                  * causes no harm unless such blocks are referenced
2332                  * by the most recent super block.
2333                  */
2334                 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2335                         pr_info("btrfsic: abort cyclic linkage (case 1).\n");
2336
2337                 return ret;
2338         }
2339
2340         /*
2341          * This algorithm is recursive because the amount of used stack
2342          * space is very small and the max recursion depth is limited.
2343          */
2344         list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2345                 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2346                         pr_info("rl=%d, %c @%llu (%s/%llu/%d) %u* refers to %c @%llu (%s/%llu/%d)\n",
2347                                recursion_level,
2348                                btrfsic_get_block_type(state, block),
2349                                block->logical_bytenr, block->dev_state->name,
2350                                block->dev_bytenr, block->mirror_num,
2351                                l->ref_cnt,
2352                                btrfsic_get_block_type(state, l->block_ref_to),
2353                                l->block_ref_to->logical_bytenr,
2354                                l->block_ref_to->dev_state->name,
2355                                l->block_ref_to->dev_bytenr,
2356                                l->block_ref_to->mirror_num);
2357                 if (l->block_ref_to->never_written) {
2358                         pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is never written!\n",
2359                                btrfsic_get_block_type(state, l->block_ref_to),
2360                                l->block_ref_to->logical_bytenr,
2361                                l->block_ref_to->dev_state->name,
2362                                l->block_ref_to->dev_bytenr,
2363                                l->block_ref_to->mirror_num);
2364                         ret = -1;
2365                 } else if (!l->block_ref_to->is_iodone) {
2366                         pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is not yet iodone!\n",
2367                                btrfsic_get_block_type(state, l->block_ref_to),
2368                                l->block_ref_to->logical_bytenr,
2369                                l->block_ref_to->dev_state->name,
2370                                l->block_ref_to->dev_bytenr,
2371                                l->block_ref_to->mirror_num);
2372                         ret = -1;
2373                 } else if (l->block_ref_to->iodone_w_error) {
2374                         pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which has write error!\n",
2375                                btrfsic_get_block_type(state, l->block_ref_to),
2376                                l->block_ref_to->logical_bytenr,
2377                                l->block_ref_to->dev_state->name,
2378                                l->block_ref_to->dev_bytenr,
2379                                l->block_ref_to->mirror_num);
2380                         ret = -1;
2381                 } else if (l->parent_generation !=
2382                            l->block_ref_to->generation &&
2383                            BTRFSIC_GENERATION_UNKNOWN !=
2384                            l->parent_generation &&
2385                            BTRFSIC_GENERATION_UNKNOWN !=
2386                            l->block_ref_to->generation) {
2387                         pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) with generation %llu != parent generation %llu!\n",
2388                                btrfsic_get_block_type(state, l->block_ref_to),
2389                                l->block_ref_to->logical_bytenr,
2390                                l->block_ref_to->dev_state->name,
2391                                l->block_ref_to->dev_bytenr,
2392                                l->block_ref_to->mirror_num,
2393                                l->block_ref_to->generation,
2394                                l->parent_generation);
2395                         ret = -1;
2396                 } else if (l->block_ref_to->flush_gen >
2397                            l->block_ref_to->dev_state->last_flush_gen) {
2398                         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",
2399                                btrfsic_get_block_type(state, l->block_ref_to),
2400                                l->block_ref_to->logical_bytenr,
2401                                l->block_ref_to->dev_state->name,
2402                                l->block_ref_to->dev_bytenr,
2403                                l->block_ref_to->mirror_num, block->flush_gen,
2404                                l->block_ref_to->dev_state->last_flush_gen);
2405                         ret = -1;
2406                 } else if (-1 == btrfsic_check_all_ref_blocks(state,
2407                                                               l->block_ref_to,
2408                                                               recursion_level +
2409                                                               1)) {
2410                         ret = -1;
2411                 }
2412         }
2413
2414         return ret;
2415 }
2416
2417 static int btrfsic_is_block_ref_by_superblock(
2418                 const struct btrfsic_state *state,
2419                 const struct btrfsic_block *block,
2420                 int recursion_level)
2421 {
2422         const struct btrfsic_block_link *l;
2423
2424         if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2425                 /* refer to comment at "abort cyclic linkage (case 1)" */
2426                 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2427                         pr_info("btrfsic: abort cyclic linkage (case 2).\n");
2428
2429                 return 0;
2430         }
2431
2432         /*
2433          * This algorithm is recursive because the amount of used stack space
2434          * is very small and the max recursion depth is limited.
2435          */
2436         list_for_each_entry(l, &block->ref_from_list, node_ref_from) {
2437                 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2438                         pr_info("rl=%d, %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
2439                                recursion_level,
2440                                btrfsic_get_block_type(state, block),
2441                                block->logical_bytenr, block->dev_state->name,
2442                                block->dev_bytenr, block->mirror_num,
2443                                l->ref_cnt,
2444                                btrfsic_get_block_type(state, l->block_ref_from),
2445                                l->block_ref_from->logical_bytenr,
2446                                l->block_ref_from->dev_state->name,
2447                                l->block_ref_from->dev_bytenr,
2448                                l->block_ref_from->mirror_num);
2449                 if (l->block_ref_from->is_superblock &&
2450                     state->latest_superblock->dev_bytenr ==
2451                     l->block_ref_from->dev_bytenr &&
2452                     state->latest_superblock->dev_state->bdev ==
2453                     l->block_ref_from->dev_state->bdev)
2454                         return 1;
2455                 else if (btrfsic_is_block_ref_by_superblock(state,
2456                                                             l->block_ref_from,
2457                                                             recursion_level +
2458                                                             1))
2459                         return 1;
2460         }
2461
2462         return 0;
2463 }
2464
2465 static void btrfsic_print_add_link(const struct btrfsic_state *state,
2466                                    const struct btrfsic_block_link *l)
2467 {
2468         pr_info("Add %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2469                l->ref_cnt,
2470                btrfsic_get_block_type(state, l->block_ref_from),
2471                l->block_ref_from->logical_bytenr,
2472                l->block_ref_from->dev_state->name,
2473                l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2474                btrfsic_get_block_type(state, l->block_ref_to),
2475                l->block_ref_to->logical_bytenr,
2476                l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2477                l->block_ref_to->mirror_num);
2478 }
2479
2480 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2481                                    const struct btrfsic_block_link *l)
2482 {
2483         pr_info("Rem %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2484                l->ref_cnt,
2485                btrfsic_get_block_type(state, l->block_ref_from),
2486                l->block_ref_from->logical_bytenr,
2487                l->block_ref_from->dev_state->name,
2488                l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2489                btrfsic_get_block_type(state, l->block_ref_to),
2490                l->block_ref_to->logical_bytenr,
2491                l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2492                l->block_ref_to->mirror_num);
2493 }
2494
2495 static char btrfsic_get_block_type(const struct btrfsic_state *state,
2496                                    const struct btrfsic_block *block)
2497 {
2498         if (block->is_superblock &&
2499             state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2500             state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2501                 return 'S';
2502         else if (block->is_superblock)
2503                 return 's';
2504         else if (block->is_metadata)
2505                 return 'M';
2506         else
2507                 return 'D';
2508 }
2509
2510 static void btrfsic_dump_tree(const struct btrfsic_state *state)
2511 {
2512         btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2513 }
2514
2515 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2516                                   const struct btrfsic_block *block,
2517                                   int indent_level)
2518 {
2519         const struct btrfsic_block_link *l;
2520         int indent_add;
2521         static char buf[80];
2522         int cursor_position;
2523
2524         /*
2525          * Should better fill an on-stack buffer with a complete line and
2526          * dump it at once when it is time to print a newline character.
2527          */
2528
2529         /*
2530          * This algorithm is recursive because the amount of used stack space
2531          * is very small and the max recursion depth is limited.
2532          */
2533         indent_add = sprintf(buf, "%c-%llu(%s/%llu/%u)",
2534                              btrfsic_get_block_type(state, block),
2535                              block->logical_bytenr, block->dev_state->name,
2536                              block->dev_bytenr, block->mirror_num);
2537         if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2538                 printk("[...]\n");
2539                 return;
2540         }
2541         printk(buf);
2542         indent_level += indent_add;
2543         if (list_empty(&block->ref_to_list)) {
2544                 printk("\n");
2545                 return;
2546         }
2547         if (block->mirror_num > 1 &&
2548             !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2549                 printk(" [...]\n");
2550                 return;
2551         }
2552
2553         cursor_position = indent_level;
2554         list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2555                 while (cursor_position < indent_level) {
2556                         printk(" ");
2557                         cursor_position++;
2558                 }
2559                 if (l->ref_cnt > 1)
2560                         indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2561                 else
2562                         indent_add = sprintf(buf, " --> ");
2563                 if (indent_level + indent_add >
2564                     BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2565                         printk("[...]\n");
2566                         cursor_position = 0;
2567                         continue;
2568                 }
2569
2570                 printk(buf);
2571
2572                 btrfsic_dump_tree_sub(state, l->block_ref_to,
2573                                       indent_level + indent_add);
2574                 cursor_position = 0;
2575         }
2576 }
2577
2578 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2579                 struct btrfsic_state *state,
2580                 struct btrfsic_block_data_ctx *next_block_ctx,
2581                 struct btrfsic_block *next_block,
2582                 struct btrfsic_block *from_block,
2583                 u64 parent_generation)
2584 {
2585         struct btrfsic_block_link *l;
2586
2587         l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2588                                                 next_block_ctx->dev_bytenr,
2589                                                 from_block->dev_state->bdev,
2590                                                 from_block->dev_bytenr,
2591                                                 &state->block_link_hashtable);
2592         if (NULL == l) {
2593                 l = btrfsic_block_link_alloc();
2594                 if (NULL == l) {
2595                         pr_info("btrfsic: error, kmalloc failed!\n");
2596                         return NULL;
2597                 }
2598
2599                 l->block_ref_to = next_block;
2600                 l->block_ref_from = from_block;
2601                 l->ref_cnt = 1;
2602                 l->parent_generation = parent_generation;
2603
2604                 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2605                         btrfsic_print_add_link(state, l);
2606
2607                 list_add(&l->node_ref_to, &from_block->ref_to_list);
2608                 list_add(&l->node_ref_from, &next_block->ref_from_list);
2609
2610                 btrfsic_block_link_hashtable_add(l,
2611                                                  &state->block_link_hashtable);
2612         } else {
2613                 l->ref_cnt++;
2614                 l->parent_generation = parent_generation;
2615                 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2616                         btrfsic_print_add_link(state, l);
2617         }
2618
2619         return l;
2620 }
2621
2622 static struct btrfsic_block *btrfsic_block_lookup_or_add(
2623                 struct btrfsic_state *state,
2624                 struct btrfsic_block_data_ctx *block_ctx,
2625                 const char *additional_string,
2626                 int is_metadata,
2627                 int is_iodone,
2628                 int never_written,
2629                 int mirror_num,
2630                 int *was_created)
2631 {
2632         struct btrfsic_block *block;
2633
2634         block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2635                                                block_ctx->dev_bytenr,
2636                                                &state->block_hashtable);
2637         if (NULL == block) {
2638                 struct btrfsic_dev_state *dev_state;
2639
2640                 block = btrfsic_block_alloc();
2641                 if (NULL == block) {
2642                         pr_info("btrfsic: error, kmalloc failed!\n");
2643                         return NULL;
2644                 }
2645                 dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev->bd_dev);
2646                 if (NULL == dev_state) {
2647                         pr_info("btrfsic: error, lookup dev_state failed!\n");
2648                         btrfsic_block_free(block);
2649                         return NULL;
2650                 }
2651                 block->dev_state = dev_state;
2652                 block->dev_bytenr = block_ctx->dev_bytenr;
2653                 block->logical_bytenr = block_ctx->start;
2654                 block->is_metadata = is_metadata;
2655                 block->is_iodone = is_iodone;
2656                 block->never_written = never_written;
2657                 block->mirror_num = mirror_num;
2658                 if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2659                         pr_info("New %s%c-block @%llu (%s/%llu/%d)\n",
2660                                additional_string,
2661                                btrfsic_get_block_type(state, block),
2662                                block->logical_bytenr, dev_state->name,
2663                                block->dev_bytenr, mirror_num);
2664                 list_add(&block->all_blocks_node, &state->all_blocks_list);
2665                 btrfsic_block_hashtable_add(block, &state->block_hashtable);
2666                 if (NULL != was_created)
2667                         *was_created = 1;
2668         } else {
2669                 if (NULL != was_created)
2670                         *was_created = 0;
2671         }
2672
2673         return block;
2674 }
2675
2676 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2677                                            u64 bytenr,
2678                                            struct btrfsic_dev_state *dev_state,
2679                                            u64 dev_bytenr)
2680 {
2681         struct btrfs_fs_info *fs_info = state->fs_info;
2682         struct btrfsic_block_data_ctx block_ctx;
2683         int num_copies;
2684         int mirror_num;
2685         int match = 0;
2686         int ret;
2687
2688         num_copies = btrfs_num_copies(fs_info, bytenr, state->metablock_size);
2689
2690         for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2691                 ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2692                                         &block_ctx, mirror_num);
2693                 if (ret) {
2694                         pr_info("btrfsic: btrfsic_map_block(logical @%llu, mirror %d) failed!\n",
2695                                bytenr, mirror_num);
2696                         continue;
2697                 }
2698
2699                 if (dev_state->bdev == block_ctx.dev->bdev &&
2700                     dev_bytenr == block_ctx.dev_bytenr) {
2701                         match++;
2702                         btrfsic_release_block_ctx(&block_ctx);
2703                         break;
2704                 }
2705                 btrfsic_release_block_ctx(&block_ctx);
2706         }
2707
2708         if (WARN_ON(!match)) {
2709                 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",
2710                        bytenr, dev_state->name, dev_bytenr);
2711                 for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2712                         ret = btrfsic_map_block(state, bytenr,
2713                                                 state->metablock_size,
2714                                                 &block_ctx, mirror_num);
2715                         if (ret)
2716                                 continue;
2717
2718                         pr_info("Read logical bytenr @%llu maps to (%s/%llu/%d)\n",
2719                                bytenr, block_ctx.dev->name,
2720                                block_ctx.dev_bytenr, mirror_num);
2721                 }
2722         }
2723 }
2724
2725 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev)
2726 {
2727         return btrfsic_dev_state_hashtable_lookup(dev,
2728                                                   &btrfsic_dev_state_hashtable);
2729 }
2730
2731 int btrfsic_submit_bh(int op, int op_flags, struct buffer_head *bh)
2732 {
2733         struct btrfsic_dev_state *dev_state;
2734
2735         if (!btrfsic_is_initialized)
2736                 return submit_bh(op, op_flags, bh);
2737
2738         mutex_lock(&btrfsic_mutex);
2739         /* since btrfsic_submit_bh() might also be called before
2740          * btrfsic_mount(), this might return NULL */
2741         dev_state = btrfsic_dev_state_lookup(bh->b_bdev->bd_dev);
2742
2743         /* Only called to write the superblock (incl. FLUSH/FUA) */
2744         if (NULL != dev_state &&
2745             (op == REQ_OP_WRITE) && bh->b_size > 0) {
2746                 u64 dev_bytenr;
2747
2748                 dev_bytenr = BTRFS_BDEV_BLOCKSIZE * bh->b_blocknr;
2749                 if (dev_state->state->print_mask &
2750                     BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2751                         pr_info("submit_bh(op=0x%x,0x%x, blocknr=%llu (bytenr %llu), size=%zu, data=%p, bdev=%p)\n",
2752                                op, op_flags, (unsigned long long)bh->b_blocknr,
2753                                dev_bytenr, bh->b_size, bh->b_data, bh->b_bdev);
2754                 btrfsic_process_written_block(dev_state, dev_bytenr,
2755                                               &bh->b_data, 1, NULL,
2756                                               NULL, bh, op_flags);
2757         } else if (NULL != dev_state && (op_flags & REQ_PREFLUSH)) {
2758                 if (dev_state->state->print_mask &
2759                     BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2760                         pr_info("submit_bh(op=0x%x,0x%x FLUSH, bdev=%p)\n",
2761                                op, op_flags, bh->b_bdev);
2762                 if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2763                         if ((dev_state->state->print_mask &
2764                              (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2765                               BTRFSIC_PRINT_MASK_VERBOSE)))
2766                                 pr_info("btrfsic_submit_bh(%s) with FLUSH but dummy block already in use (ignored)!\n",
2767                                        dev_state->name);
2768                 } else {
2769                         struct btrfsic_block *const block =
2770                                 &dev_state->dummy_block_for_bio_bh_flush;
2771
2772                         block->is_iodone = 0;
2773                         block->never_written = 0;
2774                         block->iodone_w_error = 0;
2775                         block->flush_gen = dev_state->last_flush_gen + 1;
2776                         block->submit_bio_bh_rw = op_flags;
2777                         block->orig_bio_bh_private = bh->b_private;
2778                         block->orig_bio_bh_end_io.bh = bh->b_end_io;
2779                         block->next_in_same_bio = NULL;
2780             &n