xref: /openbmc/linux/fs/btrfs/check-integrity.c (revision b830f94f)
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 <crypto/hash.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 pgoff;
1206 	char *kaddr;
1207 	char *dst = (char *)dstv;
1208 	size_t start_offset = offset_in_page(block_ctx->start);
1209 	unsigned long i = (start_offset + offset) >> PAGE_SHIFT;
1210 
1211 	WARN_ON(offset + len > block_ctx->len);
1212 	pgoff = offset_in_page(start_offset + offset);
1213 
1214 	while (len > 0) {
1215 		cur = min(len, ((size_t)PAGE_SIZE - pgoff));
1216 		BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_SIZE));
1217 		kaddr = block_ctx->datav[i];
1218 		memcpy(dst, kaddr + pgoff, cur);
1219 
1220 		dst += cur;
1221 		len -= cur;
1222 		pgoff = 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 	size_t size;
1598 	u64 dev_bytenr;
1599 	int ret;
1600 
1601 	BUG_ON(block_ctx->datav);
1602 	BUG_ON(block_ctx->pagev);
1603 	BUG_ON(block_ctx->mem_to_free);
1604 	if (!PAGE_ALIGNED(block_ctx->dev_bytenr)) {
1605 		pr_info("btrfsic: read_block() with unaligned bytenr %llu\n",
1606 		       block_ctx->dev_bytenr);
1607 		return -1;
1608 	}
1609 
1610 	num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1611 		    PAGE_SHIFT;
1612 	size = sizeof(*block_ctx->datav) + sizeof(*block_ctx->pagev);
1613 	block_ctx->mem_to_free = kcalloc(num_pages, size, 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 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
1714 	struct btrfs_header *h;
1715 	u8 csum[BTRFS_CSUM_SIZE];
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->fs_devices->fsid, BTRFS_FSID_SIZE))
1724 		return 1;
1725 
1726 	shash->tfm = fs_info->csum_shash;
1727 	crypto_shash_init(shash);
1728 
1729 	for (i = 0; i < num_pages; i++) {
1730 		u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1731 		size_t sublen = i ? PAGE_SIZE :
1732 				    (PAGE_SIZE - BTRFS_CSUM_SIZE);
1733 
1734 		crypto_shash_update(shash, data, sublen);
1735 	}
1736 	crypto_shash_final(shash, csum);
1737 	if (memcmp(csum, h->csum, state->csum_size))
1738 		return 1;
1739 
1740 	return 0; /* is metadata */
1741 }
1742 
1743 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1744 					  u64 dev_bytenr, char **mapped_datav,
1745 					  unsigned int num_pages,
1746 					  struct bio *bio, int *bio_is_patched,
1747 					  struct buffer_head *bh,
1748 					  int submit_bio_bh_rw)
1749 {
1750 	int is_metadata;
1751 	struct btrfsic_block *block;
1752 	struct btrfsic_block_data_ctx block_ctx;
1753 	int ret;
1754 	struct btrfsic_state *state = dev_state->state;
1755 	struct block_device *bdev = dev_state->bdev;
1756 	unsigned int processed_len;
1757 
1758 	if (NULL != bio_is_patched)
1759 		*bio_is_patched = 0;
1760 
1761 again:
1762 	if (num_pages == 0)
1763 		return;
1764 
1765 	processed_len = 0;
1766 	is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1767 						      num_pages));
1768 
1769 	block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1770 					       &state->block_hashtable);
1771 	if (NULL != block) {
1772 		u64 bytenr = 0;
1773 		struct btrfsic_block_link *l, *tmp;
1774 
1775 		if (block->is_superblock) {
1776 			bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1777 						    mapped_datav[0]);
1778 			if (num_pages * PAGE_SIZE <
1779 			    BTRFS_SUPER_INFO_SIZE) {
1780 				pr_info("btrfsic: cannot work with too short bios!\n");
1781 				return;
1782 			}
1783 			is_metadata = 1;
1784 			BUG_ON(!PAGE_ALIGNED(BTRFS_SUPER_INFO_SIZE));
1785 			processed_len = BTRFS_SUPER_INFO_SIZE;
1786 			if (state->print_mask &
1787 			    BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1788 				pr_info("[before new superblock is written]:\n");
1789 				btrfsic_dump_tree_sub(state, block, 0);
1790 			}
1791 		}
1792 		if (is_metadata) {
1793 			if (!block->is_superblock) {
1794 				if (num_pages * PAGE_SIZE <
1795 				    state->metablock_size) {
1796 					pr_info("btrfsic: cannot work with too short bios!\n");
1797 					return;
1798 				}
1799 				processed_len = state->metablock_size;
1800 				bytenr = btrfs_stack_header_bytenr(
1801 						(struct btrfs_header *)
1802 						mapped_datav[0]);
1803 				btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1804 							       dev_state,
1805 							       dev_bytenr);
1806 			}
1807 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1808 				if (block->logical_bytenr != bytenr &&
1809 				    !(!block->is_metadata &&
1810 				      block->logical_bytenr == 0))
1811 					pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1812 					       bytenr, dev_state->name,
1813 					       dev_bytenr,
1814 					       block->mirror_num,
1815 					       btrfsic_get_block_type(state,
1816 								      block),
1817 					       block->logical_bytenr);
1818 				else
1819 					pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1820 					       bytenr, dev_state->name,
1821 					       dev_bytenr, block->mirror_num,
1822 					       btrfsic_get_block_type(state,
1823 								      block));
1824 			}
1825 			block->logical_bytenr = bytenr;
1826 		} else {
1827 			if (num_pages * PAGE_SIZE <
1828 			    state->datablock_size) {
1829 				pr_info("btrfsic: cannot work with too short bios!\n");
1830 				return;
1831 			}
1832 			processed_len = state->datablock_size;
1833 			bytenr = block->logical_bytenr;
1834 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1835 				pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1836 				       bytenr, dev_state->name, dev_bytenr,
1837 				       block->mirror_num,
1838 				       btrfsic_get_block_type(state, block));
1839 		}
1840 
1841 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1842 			pr_info("ref_to_list: %cE, ref_from_list: %cE\n",
1843 			       list_empty(&block->ref_to_list) ? ' ' : '!',
1844 			       list_empty(&block->ref_from_list) ? ' ' : '!');
1845 		if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1846 			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",
1847 			       btrfsic_get_block_type(state, block), bytenr,
1848 			       dev_state->name, dev_bytenr, block->mirror_num,
1849 			       block->generation,
1850 			       btrfs_disk_key_objectid(&block->disk_key),
1851 			       block->disk_key.type,
1852 			       btrfs_disk_key_offset(&block->disk_key),
1853 			       btrfs_stack_header_generation(
1854 				       (struct btrfs_header *) mapped_datav[0]),
1855 			       state->max_superblock_generation);
1856 			btrfsic_dump_tree(state);
1857 		}
1858 
1859 		if (!block->is_iodone && !block->never_written) {
1860 			pr_info("btrfs: attempt to overwrite %c-block @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu, which is not yet iodone!\n",
1861 			       btrfsic_get_block_type(state, block), bytenr,
1862 			       dev_state->name, dev_bytenr, block->mirror_num,
1863 			       block->generation,
1864 			       btrfs_stack_header_generation(
1865 				       (struct btrfs_header *)
1866 				       mapped_datav[0]));
1867 			/* it would not be safe to go on */
1868 			btrfsic_dump_tree(state);
1869 			goto continue_loop;
1870 		}
1871 
1872 		/*
1873 		 * Clear all references of this block. Do not free
1874 		 * the block itself even if is not referenced anymore
1875 		 * because it still carries valuable information
1876 		 * like whether it was ever written and IO completed.
1877 		 */
1878 		list_for_each_entry_safe(l, tmp, &block->ref_to_list,
1879 					 node_ref_to) {
1880 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1881 				btrfsic_print_rem_link(state, l);
1882 			l->ref_cnt--;
1883 			if (0 == l->ref_cnt) {
1884 				list_del(&l->node_ref_to);
1885 				list_del(&l->node_ref_from);
1886 				btrfsic_block_link_hashtable_remove(l);
1887 				btrfsic_block_link_free(l);
1888 			}
1889 		}
1890 
1891 		block_ctx.dev = dev_state;
1892 		block_ctx.dev_bytenr = dev_bytenr;
1893 		block_ctx.start = bytenr;
1894 		block_ctx.len = processed_len;
1895 		block_ctx.pagev = NULL;
1896 		block_ctx.mem_to_free = NULL;
1897 		block_ctx.datav = mapped_datav;
1898 
1899 		if (is_metadata || state->include_extent_data) {
1900 			block->never_written = 0;
1901 			block->iodone_w_error = 0;
1902 			if (NULL != bio) {
1903 				block->is_iodone = 0;
1904 				BUG_ON(NULL == bio_is_patched);
1905 				if (!*bio_is_patched) {
1906 					block->orig_bio_bh_private =
1907 					    bio->bi_private;
1908 					block->orig_bio_bh_end_io.bio =
1909 					    bio->bi_end_io;
1910 					block->next_in_same_bio = NULL;
1911 					bio->bi_private = block;
1912 					bio->bi_end_io = btrfsic_bio_end_io;
1913 					*bio_is_patched = 1;
1914 				} else {
1915 					struct btrfsic_block *chained_block =
1916 					    (struct btrfsic_block *)
1917 					    bio->bi_private;
1918 
1919 					BUG_ON(NULL == chained_block);
1920 					block->orig_bio_bh_private =
1921 					    chained_block->orig_bio_bh_private;
1922 					block->orig_bio_bh_end_io.bio =
1923 					    chained_block->orig_bio_bh_end_io.
1924 					    bio;
1925 					block->next_in_same_bio = chained_block;
1926 					bio->bi_private = block;
1927 				}
1928 			} else if (NULL != bh) {
1929 				block->is_iodone = 0;
1930 				block->orig_bio_bh_private = bh->b_private;
1931 				block->orig_bio_bh_end_io.bh = bh->b_end_io;
1932 				block->next_in_same_bio = NULL;
1933 				bh->b_private = block;
1934 				bh->b_end_io = btrfsic_bh_end_io;
1935 			} else {
1936 				block->is_iodone = 1;
1937 				block->orig_bio_bh_private = NULL;
1938 				block->orig_bio_bh_end_io.bio = NULL;
1939 				block->next_in_same_bio = NULL;
1940 			}
1941 		}
1942 
1943 		block->flush_gen = dev_state->last_flush_gen + 1;
1944 		block->submit_bio_bh_rw = submit_bio_bh_rw;
1945 		if (is_metadata) {
1946 			block->logical_bytenr = bytenr;
1947 			block->is_metadata = 1;
1948 			if (block->is_superblock) {
1949 				BUG_ON(PAGE_SIZE !=
1950 				       BTRFS_SUPER_INFO_SIZE);
1951 				ret = btrfsic_process_written_superblock(
1952 						state,
1953 						block,
1954 						(struct btrfs_super_block *)
1955 						mapped_datav[0]);
1956 				if (state->print_mask &
1957 				    BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
1958 					pr_info("[after new superblock is written]:\n");
1959 					btrfsic_dump_tree_sub(state, block, 0);
1960 				}
1961 			} else {
1962 				block->mirror_num = 0;	/* unknown */
1963 				ret = btrfsic_process_metablock(
1964 						state,
1965 						block,
1966 						&block_ctx,
1967 						0, 0);
1968 			}
1969 			if (ret)
1970 				pr_info("btrfsic: btrfsic_process_metablock(root @%llu) failed!\n",
1971 				       dev_bytenr);
1972 		} else {
1973 			block->is_metadata = 0;
1974 			block->mirror_num = 0;	/* unknown */
1975 			block->generation = BTRFSIC_GENERATION_UNKNOWN;
1976 			if (!state->include_extent_data
1977 			    && list_empty(&block->ref_from_list)) {
1978 				/*
1979 				 * disk block is overwritten with extent
1980 				 * data (not meta data) and we are configured
1981 				 * to not include extent data: take the
1982 				 * chance and free the block's memory
1983 				 */
1984 				btrfsic_block_hashtable_remove(block);
1985 				list_del(&block->all_blocks_node);
1986 				btrfsic_block_free(block);
1987 			}
1988 		}
1989 		btrfsic_release_block_ctx(&block_ctx);
1990 	} else {
1991 		/* block has not been found in hash table */
1992 		u64 bytenr;
1993 
1994 		if (!is_metadata) {
1995 			processed_len = state->datablock_size;
1996 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1997 				pr_info("Written block (%s/%llu/?) !found in hash table, D.\n",
1998 				       dev_state->name, dev_bytenr);
1999 			if (!state->include_extent_data) {
2000 				/* ignore that written D block */
2001 				goto continue_loop;
2002 			}
2003 
2004 			/* this is getting ugly for the
2005 			 * include_extent_data case... */
2006 			bytenr = 0;	/* unknown */
2007 		} else {
2008 			processed_len = state->metablock_size;
2009 			bytenr = btrfs_stack_header_bytenr(
2010 					(struct btrfs_header *)
2011 					mapped_datav[0]);
2012 			btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
2013 						       dev_bytenr);
2014 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2015 				pr_info("Written block @%llu (%s/%llu/?) !found in hash table, M.\n",
2016 				       bytenr, dev_state->name, dev_bytenr);
2017 		}
2018 
2019 		block_ctx.dev = dev_state;
2020 		block_ctx.dev_bytenr = dev_bytenr;
2021 		block_ctx.start = bytenr;
2022 		block_ctx.len = processed_len;
2023 		block_ctx.pagev = NULL;
2024 		block_ctx.mem_to_free = NULL;
2025 		block_ctx.datav = mapped_datav;
2026 
2027 		block = btrfsic_block_alloc();
2028 		if (NULL == block) {
2029 			pr_info("btrfsic: error, kmalloc failed!\n");
2030 			btrfsic_release_block_ctx(&block_ctx);
2031 			goto continue_loop;
2032 		}
2033 		block->dev_state = dev_state;
2034 		block->dev_bytenr = dev_bytenr;
2035 		block->logical_bytenr = bytenr;
2036 		block->is_metadata = is_metadata;
2037 		block->never_written = 0;
2038 		block->iodone_w_error = 0;
2039 		block->mirror_num = 0;	/* unknown */
2040 		block->flush_gen = dev_state->last_flush_gen + 1;
2041 		block->submit_bio_bh_rw = submit_bio_bh_rw;
2042 		if (NULL != bio) {
2043 			block->is_iodone = 0;
2044 			BUG_ON(NULL == bio_is_patched);
2045 			if (!*bio_is_patched) {
2046 				block->orig_bio_bh_private = bio->bi_private;
2047 				block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2048 				block->next_in_same_bio = NULL;
2049 				bio->bi_private = block;
2050 				bio->bi_end_io = btrfsic_bio_end_io;
2051 				*bio_is_patched = 1;
2052 			} else {
2053 				struct btrfsic_block *chained_block =
2054 				    (struct btrfsic_block *)
2055 				    bio->bi_private;
2056 
2057 				BUG_ON(NULL == chained_block);
2058 				block->orig_bio_bh_private =
2059 				    chained_block->orig_bio_bh_private;
2060 				block->orig_bio_bh_end_io.bio =
2061 				    chained_block->orig_bio_bh_end_io.bio;
2062 				block->next_in_same_bio = chained_block;
2063 				bio->bi_private = block;
2064 			}
2065 		} else if (NULL != bh) {
2066 			block->is_iodone = 0;
2067 			block->orig_bio_bh_private = bh->b_private;
2068 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2069 			block->next_in_same_bio = NULL;
2070 			bh->b_private = block;
2071 			bh->b_end_io = btrfsic_bh_end_io;
2072 		} else {
2073 			block->is_iodone = 1;
2074 			block->orig_bio_bh_private = NULL;
2075 			block->orig_bio_bh_end_io.bio = NULL;
2076 			block->next_in_same_bio = NULL;
2077 		}
2078 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2079 			pr_info("New written %c-block @%llu (%s/%llu/%d)\n",
2080 			       is_metadata ? 'M' : 'D',
2081 			       block->logical_bytenr, block->dev_state->name,
2082 			       block->dev_bytenr, block->mirror_num);
2083 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2084 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2085 
2086 		if (is_metadata) {
2087 			ret = btrfsic_process_metablock(state, block,
2088 							&block_ctx, 0, 0);
2089 			if (ret)
2090 				pr_info("btrfsic: process_metablock(root @%llu) failed!\n",
2091 				       dev_bytenr);
2092 		}
2093 		btrfsic_release_block_ctx(&block_ctx);
2094 	}
2095 
2096 continue_loop:
2097 	BUG_ON(!processed_len);
2098 	dev_bytenr += processed_len;
2099 	mapped_datav += processed_len >> PAGE_SHIFT;
2100 	num_pages -= processed_len >> PAGE_SHIFT;
2101 	goto again;
2102 }
2103 
2104 static void btrfsic_bio_end_io(struct bio *bp)
2105 {
2106 	struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2107 	int iodone_w_error;
2108 
2109 	/* mutex is not held! This is not save if IO is not yet completed
2110 	 * on umount */
2111 	iodone_w_error = 0;
2112 	if (bp->bi_status)
2113 		iodone_w_error = 1;
2114 
2115 	BUG_ON(NULL == block);
2116 	bp->bi_private = block->orig_bio_bh_private;
2117 	bp->bi_end_io = block->orig_bio_bh_end_io.bio;
2118 
2119 	do {
2120 		struct btrfsic_block *next_block;
2121 		struct btrfsic_dev_state *const dev_state = block->dev_state;
2122 
2123 		if ((dev_state->state->print_mask &
2124 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2125 			pr_info("bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2126 			       bp->bi_status,
2127 			       btrfsic_get_block_type(dev_state->state, block),
2128 			       block->logical_bytenr, dev_state->name,
2129 			       block->dev_bytenr, block->mirror_num);
2130 		next_block = block->next_in_same_bio;
2131 		block->iodone_w_error = iodone_w_error;
2132 		if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2133 			dev_state->last_flush_gen++;
2134 			if ((dev_state->state->print_mask &
2135 			     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2136 				pr_info("bio_end_io() new %s flush_gen=%llu\n",
2137 				       dev_state->name,
2138 				       dev_state->last_flush_gen);
2139 		}
2140 		if (block->submit_bio_bh_rw & REQ_FUA)
2141 			block->flush_gen = 0; /* FUA completed means block is
2142 					       * on disk */
2143 		block->is_iodone = 1; /* for FLUSH, this releases the block */
2144 		block = next_block;
2145 	} while (NULL != block);
2146 
2147 	bp->bi_end_io(bp);
2148 }
2149 
2150 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate)
2151 {
2152 	struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private;
2153 	int iodone_w_error = !uptodate;
2154 	struct btrfsic_dev_state *dev_state;
2155 
2156 	BUG_ON(NULL == block);
2157 	dev_state = block->dev_state;
2158 	if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2159 		pr_info("bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n",
2160 		       iodone_w_error,
2161 		       btrfsic_get_block_type(dev_state->state, block),
2162 		       block->logical_bytenr, block->dev_state->name,
2163 		       block->dev_bytenr, block->mirror_num);
2164 
2165 	block->iodone_w_error = iodone_w_error;
2166 	if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2167 		dev_state->last_flush_gen++;
2168 		if ((dev_state->state->print_mask &
2169 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2170 			pr_info("bh_end_io() new %s flush_gen=%llu\n",
2171 			       dev_state->name, dev_state->last_flush_gen);
2172 	}
2173 	if (block->submit_bio_bh_rw & REQ_FUA)
2174 		block->flush_gen = 0; /* FUA completed means block is on disk */
2175 
2176 	bh->b_private = block->orig_bio_bh_private;
2177 	bh->b_end_io = block->orig_bio_bh_end_io.bh;
2178 	block->is_iodone = 1; /* for FLUSH, this releases the block */
2179 	bh->b_end_io(bh, uptodate);
2180 }
2181 
2182 static int btrfsic_process_written_superblock(
2183 		struct btrfsic_state *state,
2184 		struct btrfsic_block *const superblock,
2185 		struct btrfs_super_block *const super_hdr)
2186 {
2187 	struct btrfs_fs_info *fs_info = state->fs_info;
2188 	int pass;
2189 
2190 	superblock->generation = btrfs_super_generation(super_hdr);
2191 	if (!(superblock->generation > state->max_superblock_generation ||
2192 	      0 == state->max_superblock_generation)) {
2193 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2194 			pr_info("btrfsic: superblock @%llu (%s/%llu/%d) with old gen %llu <= %llu\n",
2195 			       superblock->logical_bytenr,
2196 			       superblock->dev_state->name,
2197 			       superblock->dev_bytenr, superblock->mirror_num,
2198 			       btrfs_super_generation(super_hdr),
2199 			       state->max_superblock_generation);
2200 	} else {
2201 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2202 			pr_info("btrfsic: got new superblock @%llu (%s/%llu/%d) with new gen %llu > %llu\n",
2203 			       superblock->logical_bytenr,
2204 			       superblock->dev_state->name,
2205 			       superblock->dev_bytenr, superblock->mirror_num,
2206 			       btrfs_super_generation(super_hdr),
2207 			       state->max_superblock_generation);
2208 
2209 		state->max_superblock_generation =
2210 		    btrfs_super_generation(super_hdr);
2211 		state->latest_superblock = superblock;
2212 	}
2213 
2214 	for (pass = 0; pass < 3; pass++) {
2215 		int ret;
2216 		u64 next_bytenr;
2217 		struct btrfsic_block *next_block;
2218 		struct btrfsic_block_data_ctx tmp_next_block_ctx;
2219 		struct btrfsic_block_link *l;
2220 		int num_copies;
2221 		int mirror_num;
2222 		const char *additional_string = NULL;
2223 		struct btrfs_disk_key tmp_disk_key = {0};
2224 
2225 		btrfs_set_disk_key_objectid(&tmp_disk_key,
2226 					    BTRFS_ROOT_ITEM_KEY);
2227 		btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2228 
2229 		switch (pass) {
2230 		case 0:
2231 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2232 						    BTRFS_ROOT_TREE_OBJECTID);
2233 			additional_string = "root ";
2234 			next_bytenr = btrfs_super_root(super_hdr);
2235 			if (state->print_mask &
2236 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2237 				pr_info("root@%llu\n", next_bytenr);
2238 			break;
2239 		case 1:
2240 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2241 						    BTRFS_CHUNK_TREE_OBJECTID);
2242 			additional_string = "chunk ";
2243 			next_bytenr = btrfs_super_chunk_root(super_hdr);
2244 			if (state->print_mask &
2245 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2246 				pr_info("chunk@%llu\n", next_bytenr);
2247 			break;
2248 		case 2:
2249 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2250 						    BTRFS_TREE_LOG_OBJECTID);
2251 			additional_string = "log ";
2252 			next_bytenr = btrfs_super_log_root(super_hdr);
2253 			if (0 == next_bytenr)
2254 				continue;
2255 			if (state->print_mask &
2256 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2257 				pr_info("log@%llu\n", next_bytenr);
2258 			break;
2259 		}
2260 
2261 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
2262 					      BTRFS_SUPER_INFO_SIZE);
2263 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2264 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
2265 			       next_bytenr, num_copies);
2266 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2267 			int was_created;
2268 
2269 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2270 				pr_info("btrfsic_process_written_superblock(mirror_num=%d)\n", mirror_num);
2271 			ret = btrfsic_map_block(state, next_bytenr,
2272 						BTRFS_SUPER_INFO_SIZE,
2273 						&tmp_next_block_ctx,
2274 						mirror_num);
2275 			if (ret) {
2276 				pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
2277 				       next_bytenr, mirror_num);
2278 				return -1;
2279 			}
2280 
2281 			next_block = btrfsic_block_lookup_or_add(
2282 					state,
2283 					&tmp_next_block_ctx,
2284 					additional_string,
2285 					1, 0, 1,
2286 					mirror_num,
2287 					&was_created);
2288 			if (NULL == next_block) {
2289 				pr_info("btrfsic: error, kmalloc failed!\n");
2290 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
2291 				return -1;
2292 			}
2293 
2294 			next_block->disk_key = tmp_disk_key;
2295 			if (was_created)
2296 				next_block->generation =
2297 				    BTRFSIC_GENERATION_UNKNOWN;
2298 			l = btrfsic_block_link_lookup_or_add(
2299 					state,
2300 					&tmp_next_block_ctx,
2301 					next_block,
2302 					superblock,
2303 					BTRFSIC_GENERATION_UNKNOWN);
2304 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
2305 			if (NULL == l)
2306 				return -1;
2307 		}
2308 	}
2309 
2310 	if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2311 		btrfsic_dump_tree(state);
2312 
2313 	return 0;
2314 }
2315 
2316 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2317 					struct btrfsic_block *const block,
2318 					int recursion_level)
2319 {
2320 	const struct btrfsic_block_link *l;
2321 	int ret = 0;
2322 
2323 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2324 		/*
2325 		 * Note that this situation can happen and does not
2326 		 * indicate an error in regular cases. It happens
2327 		 * when disk blocks are freed and later reused.
2328 		 * The check-integrity module is not aware of any
2329 		 * block free operations, it just recognizes block
2330 		 * write operations. Therefore it keeps the linkage
2331 		 * information for a block until a block is
2332 		 * rewritten. This can temporarily cause incorrect
2333 		 * and even circular linkage information. This
2334 		 * causes no harm unless such blocks are referenced
2335 		 * by the most recent super block.
2336 		 */
2337 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2338 			pr_info("btrfsic: abort cyclic linkage (case 1).\n");
2339 
2340 		return ret;
2341 	}
2342 
2343 	/*
2344 	 * This algorithm is recursive because the amount of used stack
2345 	 * space is very small and the max recursion depth is limited.
2346 	 */
2347 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2348 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2349 			pr_info("rl=%d, %c @%llu (%s/%llu/%d) %u* refers to %c @%llu (%s/%llu/%d)\n",
2350 			       recursion_level,
2351 			       btrfsic_get_block_type(state, block),
2352 			       block->logical_bytenr, block->dev_state->name,
2353 			       block->dev_bytenr, block->mirror_num,
2354 			       l->ref_cnt,
2355 			       btrfsic_get_block_type(state, l->block_ref_to),
2356 			       l->block_ref_to->logical_bytenr,
2357 			       l->block_ref_to->dev_state->name,
2358 			       l->block_ref_to->dev_bytenr,
2359 			       l->block_ref_to->mirror_num);
2360 		if (l->block_ref_to->never_written) {
2361 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is never written!\n",
2362 			       btrfsic_get_block_type(state, l->block_ref_to),
2363 			       l->block_ref_to->logical_bytenr,
2364 			       l->block_ref_to->dev_state->name,
2365 			       l->block_ref_to->dev_bytenr,
2366 			       l->block_ref_to->mirror_num);
2367 			ret = -1;
2368 		} else if (!l->block_ref_to->is_iodone) {
2369 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is not yet iodone!\n",
2370 			       btrfsic_get_block_type(state, l->block_ref_to),
2371 			       l->block_ref_to->logical_bytenr,
2372 			       l->block_ref_to->dev_state->name,
2373 			       l->block_ref_to->dev_bytenr,
2374 			       l->block_ref_to->mirror_num);
2375 			ret = -1;
2376 		} else if (l->block_ref_to->iodone_w_error) {
2377 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which has write error!\n",
2378 			       btrfsic_get_block_type(state, l->block_ref_to),
2379 			       l->block_ref_to->logical_bytenr,
2380 			       l->block_ref_to->dev_state->name,
2381 			       l->block_ref_to->dev_bytenr,
2382 			       l->block_ref_to->mirror_num);
2383 			ret = -1;
2384 		} else if (l->parent_generation !=
2385 			   l->block_ref_to->generation &&
2386 			   BTRFSIC_GENERATION_UNKNOWN !=
2387 			   l->parent_generation &&
2388 			   BTRFSIC_GENERATION_UNKNOWN !=
2389 			   l->block_ref_to->generation) {
2390 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) with generation %llu != parent generation %llu!\n",
2391 			       btrfsic_get_block_type(state, l->block_ref_to),
2392 			       l->block_ref_to->logical_bytenr,
2393 			       l->block_ref_to->dev_state->name,
2394 			       l->block_ref_to->dev_bytenr,
2395 			       l->block_ref_to->mirror_num,
2396 			       l->block_ref_to->generation,
2397 			       l->parent_generation);
2398 			ret = -1;
2399 		} else if (l->block_ref_to->flush_gen >
2400 			   l->block_ref_to->dev_state->last_flush_gen) {
2401 			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",
2402 			       btrfsic_get_block_type(state, l->block_ref_to),
2403 			       l->block_ref_to->logical_bytenr,
2404 			       l->block_ref_to->dev_state->name,
2405 			       l->block_ref_to->dev_bytenr,
2406 			       l->block_ref_to->mirror_num, block->flush_gen,
2407 			       l->block_ref_to->dev_state->last_flush_gen);
2408 			ret = -1;
2409 		} else if (-1 == btrfsic_check_all_ref_blocks(state,
2410 							      l->block_ref_to,
2411 							      recursion_level +
2412 							      1)) {
2413 			ret = -1;
2414 		}
2415 	}
2416 
2417 	return ret;
2418 }
2419 
2420 static int btrfsic_is_block_ref_by_superblock(
2421 		const struct btrfsic_state *state,
2422 		const struct btrfsic_block *block,
2423 		int recursion_level)
2424 {
2425 	const struct btrfsic_block_link *l;
2426 
2427 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2428 		/* refer to comment at "abort cyclic linkage (case 1)" */
2429 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2430 			pr_info("btrfsic: abort cyclic linkage (case 2).\n");
2431 
2432 		return 0;
2433 	}
2434 
2435 	/*
2436 	 * This algorithm is recursive because the amount of used stack space
2437 	 * is very small and the max recursion depth is limited.
2438 	 */
2439 	list_for_each_entry(l, &block->ref_from_list, node_ref_from) {
2440 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2441 			pr_info("rl=%d, %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
2442 			       recursion_level,
2443 			       btrfsic_get_block_type(state, block),
2444 			       block->logical_bytenr, block->dev_state->name,
2445 			       block->dev_bytenr, block->mirror_num,
2446 			       l->ref_cnt,
2447 			       btrfsic_get_block_type(state, l->block_ref_from),
2448 			       l->block_ref_from->logical_bytenr,
2449 			       l->block_ref_from->dev_state->name,
2450 			       l->block_ref_from->dev_bytenr,
2451 			       l->block_ref_from->mirror_num);
2452 		if (l->block_ref_from->is_superblock &&
2453 		    state->latest_superblock->dev_bytenr ==
2454 		    l->block_ref_from->dev_bytenr &&
2455 		    state->latest_superblock->dev_state->bdev ==
2456 		    l->block_ref_from->dev_state->bdev)
2457 			return 1;
2458 		else if (btrfsic_is_block_ref_by_superblock(state,
2459 							    l->block_ref_from,
2460 							    recursion_level +
2461 							    1))
2462 			return 1;
2463 	}
2464 
2465 	return 0;
2466 }
2467 
2468 static void btrfsic_print_add_link(const struct btrfsic_state *state,
2469 				   const struct btrfsic_block_link *l)
2470 {
2471 	pr_info("Add %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2472 	       l->ref_cnt,
2473 	       btrfsic_get_block_type(state, l->block_ref_from),
2474 	       l->block_ref_from->logical_bytenr,
2475 	       l->block_ref_from->dev_state->name,
2476 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2477 	       btrfsic_get_block_type(state, l->block_ref_to),
2478 	       l->block_ref_to->logical_bytenr,
2479 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2480 	       l->block_ref_to->mirror_num);
2481 }
2482 
2483 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2484 				   const struct btrfsic_block_link *l)
2485 {
2486 	pr_info("Rem %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2487 	       l->ref_cnt,
2488 	       btrfsic_get_block_type(state, l->block_ref_from),
2489 	       l->block_ref_from->logical_bytenr,
2490 	       l->block_ref_from->dev_state->name,
2491 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2492 	       btrfsic_get_block_type(state, l->block_ref_to),
2493 	       l->block_ref_to->logical_bytenr,
2494 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2495 	       l->block_ref_to->mirror_num);
2496 }
2497 
2498 static char btrfsic_get_block_type(const struct btrfsic_state *state,
2499 				   const struct btrfsic_block *block)
2500 {
2501 	if (block->is_superblock &&
2502 	    state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2503 	    state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2504 		return 'S';
2505 	else if (block->is_superblock)
2506 		return 's';
2507 	else if (block->is_metadata)
2508 		return 'M';
2509 	else
2510 		return 'D';
2511 }
2512 
2513 static void btrfsic_dump_tree(const struct btrfsic_state *state)
2514 {
2515 	btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2516 }
2517 
2518 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2519 				  const struct btrfsic_block *block,
2520 				  int indent_level)
2521 {
2522 	const struct btrfsic_block_link *l;
2523 	int indent_add;
2524 	static char buf[80];
2525 	int cursor_position;
2526 
2527 	/*
2528 	 * Should better fill an on-stack buffer with a complete line and
2529 	 * dump it at once when it is time to print a newline character.
2530 	 */
2531 
2532 	/*
2533 	 * This algorithm is recursive because the amount of used stack space
2534 	 * is very small and the max recursion depth is limited.
2535 	 */
2536 	indent_add = sprintf(buf, "%c-%llu(%s/%llu/%u)",
2537 			     btrfsic_get_block_type(state, block),
2538 			     block->logical_bytenr, block->dev_state->name,
2539 			     block->dev_bytenr, block->mirror_num);
2540 	if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2541 		printk("[...]\n");
2542 		return;
2543 	}
2544 	printk(buf);
2545 	indent_level += indent_add;
2546 	if (list_empty(&block->ref_to_list)) {
2547 		printk("\n");
2548 		return;
2549 	}
2550 	if (block->mirror_num > 1 &&
2551 	    !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2552 		printk(" [...]\n");
2553 		return;
2554 	}
2555 
2556 	cursor_position = indent_level;
2557 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2558 		while (cursor_position < indent_level) {
2559 			printk(" ");
2560 			cursor_position++;
2561 		}
2562 		if (l->ref_cnt > 1)
2563 			indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2564 		else
2565 			indent_add = sprintf(buf, " --> ");
2566 		if (indent_level + indent_add >
2567 		    BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2568 			printk("[...]\n");
2569 			cursor_position = 0;
2570 			continue;
2571 		}
2572 
2573 		printk(buf);
2574 
2575 		btrfsic_dump_tree_sub(state, l->block_ref_to,
2576 				      indent_level + indent_add);
2577 		cursor_position = 0;
2578 	}
2579 }
2580 
2581 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2582 		struct btrfsic_state *state,
2583 		struct btrfsic_block_data_ctx *next_block_ctx,
2584 		struct btrfsic_block *next_block,
2585 		struct btrfsic_block *from_block,
2586 		u64 parent_generation)
2587 {
2588 	struct btrfsic_block_link *l;
2589 
2590 	l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2591 						next_block_ctx->dev_bytenr,
2592 						from_block->dev_state->bdev,
2593 						from_block->dev_bytenr,
2594 						&state->block_link_hashtable);
2595 	if (NULL == l) {
2596 		l = btrfsic_block_link_alloc();
2597 		if (NULL == l) {
2598 			pr_info("btrfsic: error, kmalloc failed!\n");
2599 			return NULL;
2600 		}
2601 
2602 		l->block_ref_to = next_block;
2603 		l->block_ref_from = from_block;
2604 		l->ref_cnt = 1;
2605 		l->parent_generation = parent_generation;
2606 
2607 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2608 			btrfsic_print_add_link(state, l);
2609 
2610 		list_add(&l->node_ref_to, &from_block->ref_to_list);
2611 		list_add(&l->node_ref_from, &next_block->ref_from_list);
2612 
2613 		btrfsic_block_link_hashtable_add(l,
2614 						 &state->block_link_hashtable);
2615 	} else {
2616 		l->ref_cnt++;
2617 		l->parent_generation = parent_generation;
2618 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2619 			btrfsic_print_add_link(state, l);
2620 	}
2621 
2622 	return l;
2623 }
2624 
2625 static struct btrfsic_block *btrfsic_block_lookup_or_add(
2626 		struct btrfsic_state *state,
2627 		struct btrfsic_block_data_ctx *block_ctx,
2628 		const char *additional_string,
2629 		int is_metadata,
2630 		int is_iodone,
2631 		int never_written,
2632 		int mirror_num,
2633 		int *was_created)
2634 {
2635 	struct btrfsic_block *block;
2636 
2637 	block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2638 					       block_ctx->dev_bytenr,
2639 					       &state->block_hashtable);
2640 	if (NULL == block) {
2641 		struct btrfsic_dev_state *dev_state;
2642 
2643 		block = btrfsic_block_alloc();
2644 		if (NULL == block) {
2645 			pr_info("btrfsic: error, kmalloc failed!\n");
2646 			return NULL;
2647 		}
2648 		dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev->bd_dev);
2649 		if (NULL == dev_state) {
2650 			pr_info("btrfsic: error, lookup dev_state failed!\n");
2651 			btrfsic_block_free(block);
2652 			return NULL;
2653 		}
2654 		block->dev_state = dev_state;
2655 		block->dev_bytenr = block_ctx->dev_bytenr;
2656 		block->logical_bytenr = block_ctx->start;
2657 		block->is_metadata = is_metadata;
2658 		block->is_iodone = is_iodone;
2659 		block->never_written = never_written;
2660 		block->mirror_num = mirror_num;
2661 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2662 			pr_info("New %s%c-block @%llu (%s/%llu/%d)\n",
2663 			       additional_string,
2664 			       btrfsic_get_block_type(state, block),
2665 			       block->logical_bytenr, dev_state->name,
2666 			       block->dev_bytenr, mirror_num);
2667 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2668 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2669 		if (NULL != was_created)
2670 			*was_created = 1;
2671 	} else {
2672 		if (NULL != was_created)
2673 			*was_created = 0;
2674 	}
2675 
2676 	return block;
2677 }
2678 
2679 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2680 					   u64 bytenr,
2681 					   struct btrfsic_dev_state *dev_state,
2682 					   u64 dev_bytenr)
2683 {
2684 	struct btrfs_fs_info *fs_info = state->fs_info;
2685 	struct btrfsic_block_data_ctx block_ctx;
2686 	int num_copies;
2687 	int mirror_num;
2688 	int match = 0;
2689 	int ret;
2690 
2691 	num_copies = btrfs_num_copies(fs_info, bytenr, state->metablock_size);
2692 
2693 	for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2694 		ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2695 					&block_ctx, mirror_num);
2696 		if (ret) {
2697 			pr_info("btrfsic: btrfsic_map_block(logical @%llu, mirror %d) failed!\n",
2698 			       bytenr, mirror_num);
2699 			continue;
2700 		}
2701 
2702 		if (dev_state->bdev == block_ctx.dev->bdev &&
2703 		    dev_bytenr == block_ctx.dev_bytenr) {
2704 			match++;
2705 			btrfsic_release_block_ctx(&block_ctx);
2706 			break;
2707 		}
2708 		btrfsic_release_block_ctx(&block_ctx);
2709 	}
2710 
2711 	if (WARN_ON(!match)) {
2712 		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",
2713 		       bytenr, dev_state->name, dev_bytenr);
2714 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2715 			ret = btrfsic_map_block(state, bytenr,
2716 						state->metablock_size,
2717 						&block_ctx, mirror_num);
2718 			if (ret)
2719 				continue;
2720 
2721 			pr_info("Read logical bytenr @%llu maps to (%s/%llu/%d)\n",
2722 			       bytenr, block_ctx.dev->name,
2723 			       block_ctx.dev_bytenr, mirror_num);
2724 		}
2725 	}
2726 }
2727 
2728 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev)
2729 {
2730 	return btrfsic_dev_state_hashtable_lookup(dev,
2731 						  &btrfsic_dev_state_hashtable);
2732 }
2733 
2734 int btrfsic_submit_bh(int op, int op_flags, struct buffer_head *bh)
2735 {
2736 	struct btrfsic_dev_state *dev_state;
2737 
2738 	if (!btrfsic_is_initialized)
2739 		return submit_bh(op, op_flags, bh);
2740 
2741 	mutex_lock(&btrfsic_mutex);
2742 	/* since btrfsic_submit_bh() might also be called before
2743 	 * btrfsic_mount(), this might return NULL */
2744 	dev_state = btrfsic_dev_state_lookup(bh->b_bdev->bd_dev);
2745 
2746 	/* Only called to write the superblock (incl. FLUSH/FUA) */
2747 	if (NULL != dev_state &&
2748 	    (op == REQ_OP_WRITE) && bh->b_size > 0) {
2749 		u64 dev_bytenr;
2750 
2751 		dev_bytenr = BTRFS_BDEV_BLOCKSIZE * bh->b_blocknr;
2752 		if (dev_state->state->print_mask &
2753 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2754 			pr_info("submit_bh(op=0x%x,0x%x, blocknr=%llu (bytenr %llu), size=%zu, data=%p, bdev=%p)\n",
2755 			       op, op_flags, (unsigned long long)bh->b_blocknr,
2756 			       dev_bytenr, bh->b_size, bh->b_data, bh->b_bdev);
2757 		btrfsic_process_written_block(dev_state, dev_bytenr,
2758 					      &bh->b_data, 1, NULL,
2759 					      NULL, bh, op_flags);
2760 	} else if (NULL != dev_state && (op_flags & REQ_PREFLUSH)) {
2761 		if (dev_state->state->print_mask &
2762 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2763 			pr_info("submit_bh(op=0x%x,0x%x FLUSH, bdev=%p)\n",
2764 			       op, op_flags, bh->b_bdev);
2765 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2766 			if ((dev_state->state->print_mask &
2767 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2768 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2769 				pr_info("btrfsic_submit_bh(%s) with FLUSH but dummy block already in use (ignored)!\n",
2770 				       dev_state->name);
2771 		} else {
2772 			struct btrfsic_block *const block =
2773 				&dev_state->dummy_block_for_bio_bh_flush;
2774 
2775 			block->is_iodone = 0;
2776 			block->never_written = 0;
2777 			block->iodone_w_error = 0;
2778 			block->flush_gen = dev_state->last_flush_gen + 1;
2779 			block->submit_bio_bh_rw = op_flags;
2780 			block->orig_bio_bh_private = bh->b_private;
2781 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2782 			block->next_in_same_bio = NULL;
2783 			bh->b_private = block;
2784 			bh->b_end_io = btrfsic_bh_end_io;
2785 		}
2786 	}
2787 	mutex_unlock(&btrfsic_mutex);
2788 	return submit_bh(op, op_flags, bh);
2789 }
2790 
2791 static void __btrfsic_submit_bio(struct bio *bio)
2792 {
2793 	struct btrfsic_dev_state *dev_state;
2794 
2795 	if (!btrfsic_is_initialized)
2796 		return;
2797 
2798 	mutex_lock(&btrfsic_mutex);
2799 	/* since btrfsic_submit_bio() is also called before
2800 	 * btrfsic_mount(), this might return NULL */
2801 	dev_state = btrfsic_dev_state_lookup(bio_dev(bio) + bio->bi_partno);
2802 	if (NULL != dev_state &&
2803 	    (bio_op(bio) == REQ_OP_WRITE) && bio_has_data(bio)) {
2804 		unsigned int i = 0;
2805 		u64 dev_bytenr;
2806 		u64 cur_bytenr;
2807 		struct bio_vec bvec;
2808 		struct bvec_iter iter;
2809 		int bio_is_patched;
2810 		char **mapped_datav;
2811 		unsigned int segs = bio_segments(bio);
2812 
2813 		dev_bytenr = 512 * bio->bi_iter.bi_sector;
2814 		bio_is_patched = 0;
2815 		if (dev_state->state->print_mask &
2816 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2817 			pr_info("submit_bio(rw=%d,0x%x, bi_vcnt=%u, bi_sector=%llu (bytenr %llu), bi_disk=%p)\n",
2818 			       bio_op(bio), bio->bi_opf, segs,
2819 			       (unsigned long long)bio->bi_iter.bi_sector,
2820 			       dev_bytenr, bio->bi_disk);
2821 
2822 		mapped_datav = kmalloc_array(segs,
2823 					     sizeof(*mapped_datav), GFP_NOFS);
2824 		if (!mapped_datav)
2825 			goto leave;
2826 		cur_bytenr = dev_bytenr;
2827 
2828 		bio_for_each_segment(bvec, bio, iter) {
2829 			BUG_ON(bvec.bv_len != PAGE_SIZE);
2830 			mapped_datav[i] = kmap(bvec.bv_page);
2831 			i++;
2832 
2833 			if (dev_state->state->print_mask &
2834 			    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
2835 				pr_info("#%u: bytenr=%llu, len=%u, offset=%u\n",
2836 				       i, cur_bytenr, bvec.bv_len, bvec.bv_offset);
2837 			cur_bytenr += bvec.bv_len;
2838 		}
2839 		btrfsic_process_written_block(dev_state, dev_bytenr,
2840 					      mapped_datav, segs,
2841 					      bio, &bio_is_patched,
2842 					      NULL, bio->bi_opf);
2843 		bio_for_each_segment(bvec, bio, iter)
2844 			kunmap(bvec.bv_page);
2845 		kfree(mapped_datav);
2846 	} else if (NULL != dev_state && (bio->bi_opf & REQ_PREFLUSH)) {
2847 		if (dev_state->state->print_mask &
2848 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2849 			pr_info("submit_bio(rw=%d,0x%x FLUSH, disk=%p)\n",
2850 			       bio_op(bio), bio->bi_opf, bio->bi_disk);
2851 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2852 			if ((dev_state->state->print_mask &
2853 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2854 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2855 				pr_info("btrfsic_submit_bio(%s) with FLUSH but dummy block already in use (ignored)!\n",
2856 				       dev_state->name);
2857 		} else {
2858 			struct btrfsic_block *const block =
2859 				&dev_state->dummy_block_for_bio_bh_flush;
2860 
2861 			block->is_iodone = 0;
2862 			block->never_written = 0;
2863 			block->iodone_w_error = 0;
2864 			block->flush_gen = dev_state->last_flush_gen + 1;
2865 			block->submit_bio_bh_rw = bio->bi_opf;
2866 			block->orig_bio_bh_private = bio->bi_private;
2867 			block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2868 			block->next_in_same_bio = NULL;
2869 			bio->bi_private = block;
2870 			bio->bi_end_io = btrfsic_bio_end_io;
2871 		}
2872 	}
2873 leave:
2874 	mutex_unlock(&btrfsic_mutex);
2875 }
2876 
2877 void btrfsic_submit_bio(struct bio *bio)
2878 {
2879 	__btrfsic_submit_bio(bio);
2880 	submit_bio(bio);
2881 }
2882 
2883 int btrfsic_submit_bio_wait(struct bio *bio)
2884 {
2885 	__btrfsic_submit_bio(bio);
2886 	return submit_bio_wait(bio);
2887 }
2888 
2889 int btrfsic_mount(struct btrfs_fs_info *fs_info,
2890 		  struct btrfs_fs_devices *fs_devices,
2891 		  int including_extent_data, u32 print_mask)
2892 {
2893 	int ret;
2894 	struct btrfsic_state *state;
2895 	struct list_head *dev_head = &fs_devices->devices;
2896 	struct btrfs_device *device;
2897 
2898 	if (!PAGE_ALIGNED(fs_info->nodesize)) {
2899 		pr_info("btrfsic: cannot handle nodesize %d not being a multiple of PAGE_SIZE %ld!\n",
2900 		       fs_info->nodesize, PAGE_SIZE);
2901 		return -1;
2902 	}
2903 	if (!PAGE_ALIGNED(fs_info->sectorsize)) {
2904 		pr_info("btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_SIZE %ld!\n",
2905 		       fs_info->sectorsize, PAGE_SIZE);
2906 		return -1;
2907 	}
2908 	state = kvzalloc(sizeof(*state), GFP_KERNEL);
2909 	if (!state) {
2910 		pr_info("btrfs check-integrity: allocation failed!\n");
2911 		return -ENOMEM;
2912 	}
2913 
2914 	if (!btrfsic_is_initialized) {
2915 		mutex_init(&btrfsic_mutex);
2916 		btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
2917 		btrfsic_is_initialized = 1;
2918 	}
2919 	mutex_lock(&btrfsic_mutex);
2920 	state->fs_info = fs_info;
2921 	state->print_mask = print_mask;
2922 	state->include_extent_data = including_extent_data;
2923 	state->csum_size = 0;
2924 	state->metablock_size = fs_info->nodesize;
2925 	state->datablock_size = fs_info->sectorsize;
2926 	INIT_LIST_HEAD(&state->all_blocks_list);
2927 	btrfsic_block_hashtable_init(&state->block_hashtable);
2928 	btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
2929 	state->max_superblock_generation = 0;
2930 	state->latest_superblock = NULL;
2931 
2932 	list_for_each_entry(device, dev_head, dev_list) {
2933 		struct btrfsic_dev_state *ds;
2934 		const char *p;
2935 
2936 		if (!device->bdev || !device->name)
2937 			continue;
2938 
2939 		ds = btrfsic_dev_state_alloc();
2940 		if (NULL == ds) {
2941 			pr_info("btrfs check-integrity: kmalloc() failed!\n");
2942 			mutex_unlock(&btrfsic_mutex);
2943 			return -ENOMEM;
2944 		}
2945 		ds->bdev = device->bdev;
2946 		ds->state = state;
2947 		bdevname(ds->bdev, ds->name);
2948 		ds->name[BDEVNAME_SIZE - 1] = '\0';
2949 		p = kbasename(ds->name);
2950 		strlcpy(ds->name, p, sizeof(ds->name));
2951 		btrfsic_dev_state_hashtable_add(ds,
2952 						&btrfsic_dev_state_hashtable);
2953 	}
2954 
2955 	ret = btrfsic_process_superblock(state, fs_devices);
2956 	if (0 != ret) {
2957 		mutex_unlock(&btrfsic_mutex);
2958 		btrfsic_unmount(fs_devices);
2959 		return ret;
2960 	}
2961 
2962 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
2963 		btrfsic_dump_database(state);
2964 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
2965 		btrfsic_dump_tree(state);
2966 
2967 	mutex_unlock(&btrfsic_mutex);
2968 	return 0;
2969 }
2970 
2971 void btrfsic_unmount(struct btrfs_fs_devices *fs_devices)
2972 {
2973 	struct btrfsic_block *b_all, *tmp_all;
2974 	struct btrfsic_state *state;
2975 	struct list_head *dev_head = &fs_devices->devices;
2976 	struct btrfs_device *device;
2977 
2978 	if (!btrfsic_is_initialized)
2979 		return;
2980 
2981 	mutex_lock(&btrfsic_mutex);
2982 
2983 	state = NULL;
2984 	list_for_each_entry(device, dev_head, dev_list) {
2985 		struct btrfsic_dev_state *ds;
2986 
2987 		if (!device->bdev || !device->name)
2988 			continue;
2989 
2990 		ds = btrfsic_dev_state_hashtable_lookup(
2991 				device->bdev->bd_dev,
2992 				&btrfsic_dev_state_hashtable);
2993 		if (NULL != ds) {
2994 			state = ds->state;
2995 			btrfsic_dev_state_hashtable_remove(ds);
2996 			btrfsic_dev_state_free(ds);
2997 		}
2998 	}
2999 
3000 	if (NULL == state) {
3001 		pr_info("btrfsic: error, cannot find state information on umount!\n");
3002 		mutex_unlock(&btrfsic_mutex);
3003 		return;
3004 	}
3005 
3006 	/*
3007 	 * Don't care about keeping the lists' state up to date,
3008 	 * just free all memory that was allocated dynamically.
3009 	 * Free the blocks and the block_links.
3010 	 */
3011 	list_for_each_entry_safe(b_all, tmp_all, &state->all_blocks_list,
3012 				 all_blocks_node) {
3013 		struct btrfsic_block_link *l, *tmp;
3014 
3015 		list_for_each_entry_safe(l, tmp, &b_all->ref_to_list,
3016 					 node_ref_to) {
3017 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
3018 				btrfsic_print_rem_link(state, l);
3019 
3020 			l->ref_cnt--;
3021 			if (0 == l->ref_cnt)
3022 				btrfsic_block_link_free(l);
3023 		}
3024 
3025 		if (b_all->is_iodone || b_all->never_written)
3026 			btrfsic_block_free(b_all);
3027 		else
3028 			pr_info("btrfs: attempt to free %c-block @%llu (%s/%llu/%d) on umount which is not yet iodone!\n",
3029 			       btrfsic_get_block_type(state, b_all),
3030 			       b_all->logical_bytenr, b_all->dev_state->name,
3031 			       b_all->dev_bytenr, b_all->mirror_num);
3032 	}
3033 
3034 	mutex_unlock(&btrfsic_mutex);
3035 
3036 	kvfree(state);
3037 }
3038