xref: /openbmc/linux/fs/btrfs/check-integrity.c (revision a17922de)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) STRATO AG 2011.  All rights reserved.
4  */
5 
6 /*
7  * This module can be used to catch cases when the btrfs kernel
8  * code executes write requests to the disk that bring the file
9  * system in an inconsistent state. In such a state, a power-loss
10  * or kernel panic event would cause that the data on disk is
11  * lost or at least damaged.
12  *
13  * Code is added that examines all block write requests during
14  * runtime (including writes of the super block). Three rules
15  * are verified and an error is printed on violation of the
16  * rules:
17  * 1. It is not allowed to write a disk block which is
18  *    currently referenced by the super block (either directly
19  *    or indirectly).
20  * 2. When a super block is written, it is verified that all
21  *    referenced (directly or indirectly) blocks fulfill the
22  *    following requirements:
23  *    2a. All referenced blocks have either been present when
24  *        the file system was mounted, (i.e., they have been
25  *        referenced by the super block) or they have been
26  *        written since then and the write completion callback
27  *        was called and no write error was indicated and a
28  *        FLUSH request to the device where these blocks are
29  *        located was received and completed.
30  *    2b. All referenced blocks need to have a generation
31  *        number which is equal to the parent's number.
32  *
33  * One issue that was found using this module was that the log
34  * tree on disk became temporarily corrupted because disk blocks
35  * that had been in use for the log tree had been freed and
36  * reused too early, while being referenced by the written super
37  * block.
38  *
39  * The search term in the kernel log that can be used to filter
40  * on the existence of detected integrity issues is
41  * "btrfs: attempt".
42  *
43  * The integrity check is enabled via mount options. These
44  * mount options are only supported if the integrity check
45  * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY.
46  *
47  * Example #1, apply integrity checks to all metadata:
48  * mount /dev/sdb1 /mnt -o check_int
49  *
50  * Example #2, apply integrity checks to all metadata and
51  * to data extents:
52  * mount /dev/sdb1 /mnt -o check_int_data
53  *
54  * Example #3, apply integrity checks to all metadata and dump
55  * the tree that the super block references to kernel messages
56  * each time after a super block was written:
57  * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263
58  *
59  * If the integrity check tool is included and activated in
60  * the mount options, plenty of kernel memory is used, and
61  * plenty of additional CPU cycles are spent. Enabling this
62  * functionality is not intended for normal use. In most
63  * cases, unless you are a btrfs developer who needs to verify
64  * the integrity of (super)-block write requests, do not
65  * enable the config option BTRFS_FS_CHECK_INTEGRITY to
66  * include and compile the integrity check tool.
67  *
68  * Expect millions of lines of information in the kernel log with an
69  * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the
70  * kernel config to at least 26 (which is 64MB). Usually the value is
71  * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be
72  * changed like this before LOG_BUF_SHIFT can be set to a high value:
73  * config LOG_BUF_SHIFT
74  *       int "Kernel log buffer size (16 => 64KB, 17 => 128KB)"
75  *       range 12 30
76  */
77 
78 #include <linux/sched.h>
79 #include <linux/slab.h>
80 #include <linux/buffer_head.h>
81 #include <linux/mutex.h>
82 #include <linux/genhd.h>
83 #include <linux/blkdev.h>
84 #include <linux/mm.h>
85 #include <linux/string.h>
86 #include <linux/crc32c.h>
87 #include "ctree.h"
88 #include "disk-io.h"
89 #include "transaction.h"
90 #include "extent_io.h"
91 #include "volumes.h"
92 #include "print-tree.h"
93 #include "locking.h"
94 #include "check-integrity.h"
95 #include "rcu-string.h"
96 #include "compression.h"
97 
98 #define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
99 #define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
100 #define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
101 #define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
102 #define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
103 #define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
104 #define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
105 #define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6)	/* in characters,
106 							 * excluding " [...]" */
107 #define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
108 
109 /*
110  * The definition of the bitmask fields for the print_mask.
111  * They are specified with the mount option check_integrity_print_mask.
112  */
113 #define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE			0x00000001
114 #define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION		0x00000002
115 #define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE			0x00000004
116 #define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE			0x00000008
117 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH			0x00000010
118 #define BTRFSIC_PRINT_MASK_END_IO_BIO_BH			0x00000020
119 #define BTRFSIC_PRINT_MASK_VERBOSE				0x00000040
120 #define BTRFSIC_PRINT_MASK_VERY_VERBOSE				0x00000080
121 #define BTRFSIC_PRINT_MASK_INITIAL_TREE				0x00000100
122 #define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES			0x00000200
123 #define BTRFSIC_PRINT_MASK_INITIAL_DATABASE			0x00000400
124 #define BTRFSIC_PRINT_MASK_NUM_COPIES				0x00000800
125 #define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS		0x00001000
126 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE		0x00002000
127 
128 struct btrfsic_dev_state;
129 struct btrfsic_state;
130 
131 struct btrfsic_block {
132 	u32 magic_num;		/* only used for debug purposes */
133 	unsigned int is_metadata:1;	/* if it is meta-data, not data-data */
134 	unsigned int is_superblock:1;	/* if it is one of the superblocks */
135 	unsigned int is_iodone:1;	/* if is done by lower subsystem */
136 	unsigned int iodone_w_error:1;	/* error was indicated to endio */
137 	unsigned int never_written:1;	/* block was added because it was
138 					 * referenced, not because it was
139 					 * written */
140 	unsigned int mirror_num;	/* large enough to hold
141 					 * BTRFS_SUPER_MIRROR_MAX */
142 	struct btrfsic_dev_state *dev_state;
143 	u64 dev_bytenr;		/* key, physical byte num on disk */
144 	u64 logical_bytenr;	/* logical byte num on disk */
145 	u64 generation;
146 	struct btrfs_disk_key disk_key;	/* extra info to print in case of
147 					 * issues, will not always be correct */
148 	struct list_head collision_resolving_node;	/* list node */
149 	struct list_head all_blocks_node;	/* list node */
150 
151 	/* the following two lists contain block_link items */
152 	struct list_head ref_to_list;	/* list */
153 	struct list_head ref_from_list;	/* list */
154 	struct btrfsic_block *next_in_same_bio;
155 	void *orig_bio_bh_private;
156 	union {
157 		bio_end_io_t *bio;
158 		bh_end_io_t *bh;
159 	} orig_bio_bh_end_io;
160 	int submit_bio_bh_rw;
161 	u64 flush_gen; /* only valid if !never_written */
162 };
163 
164 /*
165  * Elements of this type are allocated dynamically and required because
166  * each block object can refer to and can be ref from multiple blocks.
167  * The key to lookup them in the hashtable is the dev_bytenr of
168  * the block ref to plus the one from the block referred from.
169  * The fact that they are searchable via a hashtable and that a
170  * ref_cnt is maintained is not required for the btrfs integrity
171  * check algorithm itself, it is only used to make the output more
172  * beautiful in case that an error is detected (an error is defined
173  * as a write operation to a block while that block is still referenced).
174  */
175 struct btrfsic_block_link {
176 	u32 magic_num;		/* only used for debug purposes */
177 	u32 ref_cnt;
178 	struct list_head node_ref_to;	/* list node */
179 	struct list_head node_ref_from;	/* list node */
180 	struct list_head collision_resolving_node;	/* list node */
181 	struct btrfsic_block *block_ref_to;
182 	struct btrfsic_block *block_ref_from;
183 	u64 parent_generation;
184 };
185 
186 struct btrfsic_dev_state {
187 	u32 magic_num;		/* only used for debug purposes */
188 	struct block_device *bdev;
189 	struct btrfsic_state *state;
190 	struct list_head collision_resolving_node;	/* list node */
191 	struct btrfsic_block dummy_block_for_bio_bh_flush;
192 	u64 last_flush_gen;
193 	char name[BDEVNAME_SIZE];
194 };
195 
196 struct btrfsic_block_hashtable {
197 	struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
198 };
199 
200 struct btrfsic_block_link_hashtable {
201 	struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
202 };
203 
204 struct btrfsic_dev_state_hashtable {
205 	struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
206 };
207 
208 struct btrfsic_block_data_ctx {
209 	u64 start;		/* virtual bytenr */
210 	u64 dev_bytenr;		/* physical bytenr on device */
211 	u32 len;
212 	struct btrfsic_dev_state *dev;
213 	char **datav;
214 	struct page **pagev;
215 	void *mem_to_free;
216 };
217 
218 /* This structure is used to implement recursion without occupying
219  * any stack space, refer to btrfsic_process_metablock() */
220 struct btrfsic_stack_frame {
221 	u32 magic;
222 	u32 nr;
223 	int error;
224 	int i;
225 	int limit_nesting;
226 	int num_copies;
227 	int mirror_num;
228 	struct btrfsic_block *block;
229 	struct btrfsic_block_data_ctx *block_ctx;
230 	struct btrfsic_block *next_block;
231 	struct btrfsic_block_data_ctx next_block_ctx;
232 	struct btrfs_header *hdr;
233 	struct btrfsic_stack_frame *prev;
234 };
235 
236 /* Some state per mounted filesystem */
237 struct btrfsic_state {
238 	u32 print_mask;
239 	int include_extent_data;
240 	int csum_size;
241 	struct list_head all_blocks_list;
242 	struct btrfsic_block_hashtable block_hashtable;
243 	struct btrfsic_block_link_hashtable block_link_hashtable;
244 	struct btrfs_fs_info *fs_info;
245 	u64 max_superblock_generation;
246 	struct btrfsic_block *latest_superblock;
247 	u32 metablock_size;
248 	u32 datablock_size;
249 };
250 
251 static void btrfsic_block_init(struct btrfsic_block *b);
252 static struct btrfsic_block *btrfsic_block_alloc(void);
253 static void btrfsic_block_free(struct btrfsic_block *b);
254 static void btrfsic_block_link_init(struct btrfsic_block_link *n);
255 static struct btrfsic_block_link *btrfsic_block_link_alloc(void);
256 static void btrfsic_block_link_free(struct btrfsic_block_link *n);
257 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds);
258 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void);
259 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds);
260 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h);
261 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
262 					struct btrfsic_block_hashtable *h);
263 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b);
264 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
265 		struct block_device *bdev,
266 		u64 dev_bytenr,
267 		struct btrfsic_block_hashtable *h);
268 static void btrfsic_block_link_hashtable_init(
269 		struct btrfsic_block_link_hashtable *h);
270 static void btrfsic_block_link_hashtable_add(
271 		struct btrfsic_block_link *l,
272 		struct btrfsic_block_link_hashtable *h);
273 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l);
274 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
275 		struct block_device *bdev_ref_to,
276 		u64 dev_bytenr_ref_to,
277 		struct block_device *bdev_ref_from,
278 		u64 dev_bytenr_ref_from,
279 		struct btrfsic_block_link_hashtable *h);
280 static void btrfsic_dev_state_hashtable_init(
281 		struct btrfsic_dev_state_hashtable *h);
282 static void btrfsic_dev_state_hashtable_add(
283 		struct btrfsic_dev_state *ds,
284 		struct btrfsic_dev_state_hashtable *h);
285 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds);
286 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev,
287 		struct btrfsic_dev_state_hashtable *h);
288 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void);
289 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf);
290 static int btrfsic_process_superblock(struct btrfsic_state *state,
291 				      struct btrfs_fs_devices *fs_devices);
292 static int btrfsic_process_metablock(struct btrfsic_state *state,
293 				     struct btrfsic_block *block,
294 				     struct btrfsic_block_data_ctx *block_ctx,
295 				     int limit_nesting, int force_iodone_flag);
296 static void btrfsic_read_from_block_data(
297 	struct btrfsic_block_data_ctx *block_ctx,
298 	void *dst, u32 offset, size_t len);
299 static int btrfsic_create_link_to_next_block(
300 		struct btrfsic_state *state,
301 		struct btrfsic_block *block,
302 		struct btrfsic_block_data_ctx
303 		*block_ctx, u64 next_bytenr,
304 		int limit_nesting,
305 		struct btrfsic_block_data_ctx *next_block_ctx,
306 		struct btrfsic_block **next_blockp,
307 		int force_iodone_flag,
308 		int *num_copiesp, int *mirror_nump,
309 		struct btrfs_disk_key *disk_key,
310 		u64 parent_generation);
311 static int btrfsic_handle_extent_data(struct btrfsic_state *state,
312 				      struct btrfsic_block *block,
313 				      struct btrfsic_block_data_ctx *block_ctx,
314 				      u32 item_offset, int force_iodone_flag);
315 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
316 			     struct btrfsic_block_data_ctx *block_ctx_out,
317 			     int mirror_num);
318 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
319 static int btrfsic_read_block(struct btrfsic_state *state,
320 			      struct btrfsic_block_data_ctx *block_ctx);
321 static void btrfsic_dump_database(struct btrfsic_state *state);
322 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
323 				     char **datav, unsigned int num_pages);
324 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
325 					  u64 dev_bytenr, char **mapped_datav,
326 					  unsigned int num_pages,
327 					  struct bio *bio, int *bio_is_patched,
328 					  struct buffer_head *bh,
329 					  int submit_bio_bh_rw);
330 static int btrfsic_process_written_superblock(
331 		struct btrfsic_state *state,
332 		struct btrfsic_block *const block,
333 		struct btrfs_super_block *const super_hdr);
334 static void btrfsic_bio_end_io(struct bio *bp);
335 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate);
336 static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
337 					      const struct btrfsic_block *block,
338 					      int recursion_level);
339 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
340 					struct btrfsic_block *const block,
341 					int recursion_level);
342 static void btrfsic_print_add_link(const struct btrfsic_state *state,
343 				   const struct btrfsic_block_link *l);
344 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
345 				   const struct btrfsic_block_link *l);
346 static char btrfsic_get_block_type(const struct btrfsic_state *state,
347 				   const struct btrfsic_block *block);
348 static void btrfsic_dump_tree(const struct btrfsic_state *state);
349 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
350 				  const struct btrfsic_block *block,
351 				  int indent_level);
352 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
353 		struct btrfsic_state *state,
354 		struct btrfsic_block_data_ctx *next_block_ctx,
355 		struct btrfsic_block *next_block,
356 		struct btrfsic_block *from_block,
357 		u64 parent_generation);
358 static struct btrfsic_block *btrfsic_block_lookup_or_add(
359 		struct btrfsic_state *state,
360 		struct btrfsic_block_data_ctx *block_ctx,
361 		const char *additional_string,
362 		int is_metadata,
363 		int is_iodone,
364 		int never_written,
365 		int mirror_num,
366 		int *was_created);
367 static int btrfsic_process_superblock_dev_mirror(
368 		struct btrfsic_state *state,
369 		struct btrfsic_dev_state *dev_state,
370 		struct btrfs_device *device,
371 		int superblock_mirror_num,
372 		struct btrfsic_dev_state **selected_dev_state,
373 		struct btrfs_super_block *selected_super);
374 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev);
375 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
376 					   u64 bytenr,
377 					   struct btrfsic_dev_state *dev_state,
378 					   u64 dev_bytenr);
379 
380 static struct mutex btrfsic_mutex;
381 static int btrfsic_is_initialized;
382 static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
383 
384 
385 static void btrfsic_block_init(struct btrfsic_block *b)
386 {
387 	b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
388 	b->dev_state = NULL;
389 	b->dev_bytenr = 0;
390 	b->logical_bytenr = 0;
391 	b->generation = BTRFSIC_GENERATION_UNKNOWN;
392 	b->disk_key.objectid = 0;
393 	b->disk_key.type = 0;
394 	b->disk_key.offset = 0;
395 	b->is_metadata = 0;
396 	b->is_superblock = 0;
397 	b->is_iodone = 0;
398 	b->iodone_w_error = 0;
399 	b->never_written = 0;
400 	b->mirror_num = 0;
401 	b->next_in_same_bio = NULL;
402 	b->orig_bio_bh_private = NULL;
403 	b->orig_bio_bh_end_io.bio = NULL;
404 	INIT_LIST_HEAD(&b->collision_resolving_node);
405 	INIT_LIST_HEAD(&b->all_blocks_node);
406 	INIT_LIST_HEAD(&b->ref_to_list);
407 	INIT_LIST_HEAD(&b->ref_from_list);
408 	b->submit_bio_bh_rw = 0;
409 	b->flush_gen = 0;
410 }
411 
412 static struct btrfsic_block *btrfsic_block_alloc(void)
413 {
414 	struct btrfsic_block *b;
415 
416 	b = kzalloc(sizeof(*b), GFP_NOFS);
417 	if (NULL != b)
418 		btrfsic_block_init(b);
419 
420 	return b;
421 }
422 
423 static void btrfsic_block_free(struct btrfsic_block *b)
424 {
425 	BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
426 	kfree(b);
427 }
428 
429 static void btrfsic_block_link_init(struct btrfsic_block_link *l)
430 {
431 	l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
432 	l->ref_cnt = 1;
433 	INIT_LIST_HEAD(&l->node_ref_to);
434 	INIT_LIST_HEAD(&l->node_ref_from);
435 	INIT_LIST_HEAD(&l->collision_resolving_node);
436 	l->block_ref_to = NULL;
437 	l->block_ref_from = NULL;
438 }
439 
440 static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
441 {
442 	struct btrfsic_block_link *l;
443 
444 	l = kzalloc(sizeof(*l), GFP_NOFS);
445 	if (NULL != l)
446 		btrfsic_block_link_init(l);
447 
448 	return l;
449 }
450 
451 static void btrfsic_block_link_free(struct btrfsic_block_link *l)
452 {
453 	BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
454 	kfree(l);
455 }
456 
457 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
458 {
459 	ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
460 	ds->bdev = NULL;
461 	ds->state = NULL;
462 	ds->name[0] = '\0';
463 	INIT_LIST_HEAD(&ds->collision_resolving_node);
464 	ds->last_flush_gen = 0;
465 	btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
466 	ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
467 	ds->dummy_block_for_bio_bh_flush.dev_state = ds;
468 }
469 
470 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
471 {
472 	struct btrfsic_dev_state *ds;
473 
474 	ds = kzalloc(sizeof(*ds), GFP_NOFS);
475 	if (NULL != ds)
476 		btrfsic_dev_state_init(ds);
477 
478 	return ds;
479 }
480 
481 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
482 {
483 	BUG_ON(!(NULL == ds ||
484 		 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
485 	kfree(ds);
486 }
487 
488 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
489 {
490 	int i;
491 
492 	for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
493 		INIT_LIST_HEAD(h->table + i);
494 }
495 
496 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
497 					struct btrfsic_block_hashtable *h)
498 {
499 	const unsigned int hashval =
500 	    (((unsigned int)(b->dev_bytenr >> 16)) ^
501 	     ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
502 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
503 
504 	list_add(&b->collision_resolving_node, h->table + hashval);
505 }
506 
507 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
508 {
509 	list_del(&b->collision_resolving_node);
510 }
511 
512 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
513 		struct block_device *bdev,
514 		u64 dev_bytenr,
515 		struct btrfsic_block_hashtable *h)
516 {
517 	const unsigned int hashval =
518 	    (((unsigned int)(dev_bytenr >> 16)) ^
519 	     ((unsigned int)((uintptr_t)bdev))) &
520 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
521 	struct btrfsic_block *b;
522 
523 	list_for_each_entry(b, h->table + hashval, collision_resolving_node) {
524 		if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
525 			return b;
526 	}
527 
528 	return NULL;
529 }
530 
531 static void btrfsic_block_link_hashtable_init(
532 		struct btrfsic_block_link_hashtable *h)
533 {
534 	int i;
535 
536 	for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
537 		INIT_LIST_HEAD(h->table + i);
538 }
539 
540 static void btrfsic_block_link_hashtable_add(
541 		struct btrfsic_block_link *l,
542 		struct btrfsic_block_link_hashtable *h)
543 {
544 	const unsigned int hashval =
545 	    (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
546 	     ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
547 	     ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
548 	     ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
549 	     & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
550 
551 	BUG_ON(NULL == l->block_ref_to);
552 	BUG_ON(NULL == l->block_ref_from);
553 	list_add(&l->collision_resolving_node, h->table + hashval);
554 }
555 
556 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
557 {
558 	list_del(&l->collision_resolving_node);
559 }
560 
561 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
562 		struct block_device *bdev_ref_to,
563 		u64 dev_bytenr_ref_to,
564 		struct block_device *bdev_ref_from,
565 		u64 dev_bytenr_ref_from,
566 		struct btrfsic_block_link_hashtable *h)
567 {
568 	const unsigned int hashval =
569 	    (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
570 	     ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
571 	     ((unsigned int)((uintptr_t)bdev_ref_to)) ^
572 	     ((unsigned int)((uintptr_t)bdev_ref_from))) &
573 	     (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
574 	struct btrfsic_block_link *l;
575 
576 	list_for_each_entry(l, h->table + hashval, collision_resolving_node) {
577 		BUG_ON(NULL == l->block_ref_to);
578 		BUG_ON(NULL == l->block_ref_from);
579 		if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
580 		    l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
581 		    l->block_ref_from->dev_state->bdev == bdev_ref_from &&
582 		    l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
583 			return l;
584 	}
585 
586 	return NULL;
587 }
588 
589 static void btrfsic_dev_state_hashtable_init(
590 		struct btrfsic_dev_state_hashtable *h)
591 {
592 	int i;
593 
594 	for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
595 		INIT_LIST_HEAD(h->table + i);
596 }
597 
598 static void btrfsic_dev_state_hashtable_add(
599 		struct btrfsic_dev_state *ds,
600 		struct btrfsic_dev_state_hashtable *h)
601 {
602 	const unsigned int hashval =
603 	    (((unsigned int)((uintptr_t)ds->bdev->bd_dev)) &
604 	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
605 
606 	list_add(&ds->collision_resolving_node, h->table + hashval);
607 }
608 
609 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
610 {
611 	list_del(&ds->collision_resolving_node);
612 }
613 
614 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev,
615 		struct btrfsic_dev_state_hashtable *h)
616 {
617 	const unsigned int hashval =
618 		dev & (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1);
619 	struct btrfsic_dev_state *ds;
620 
621 	list_for_each_entry(ds, h->table + hashval, collision_resolving_node) {
622 		if (ds->bdev->bd_dev == dev)
623 			return ds;
624 	}
625 
626 	return NULL;
627 }
628 
629 static int btrfsic_process_superblock(struct btrfsic_state *state,
630 				      struct btrfs_fs_devices *fs_devices)
631 {
632 	struct btrfs_fs_info *fs_info = state->fs_info;
633 	struct btrfs_super_block *selected_super;
634 	struct list_head *dev_head = &fs_devices->devices;
635 	struct btrfs_device *device;
636 	struct btrfsic_dev_state *selected_dev_state = NULL;
637 	int ret = 0;
638 	int pass;
639 
640 	BUG_ON(NULL == state);
641 	selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
642 	if (NULL == selected_super) {
643 		pr_info("btrfsic: error, kmalloc failed!\n");
644 		return -ENOMEM;
645 	}
646 
647 	list_for_each_entry(device, dev_head, dev_list) {
648 		int i;
649 		struct btrfsic_dev_state *dev_state;
650 
651 		if (!device->bdev || !device->name)
652 			continue;
653 
654 		dev_state = btrfsic_dev_state_lookup(device->bdev->bd_dev);
655 		BUG_ON(NULL == dev_state);
656 		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
657 			ret = btrfsic_process_superblock_dev_mirror(
658 					state, dev_state, device, i,
659 					&selected_dev_state, selected_super);
660 			if (0 != ret && 0 == i) {
661 				kfree(selected_super);
662 				return ret;
663 			}
664 		}
665 	}
666 
667 	if (NULL == state->latest_superblock) {
668 		pr_info("btrfsic: no superblock found!\n");
669 		kfree(selected_super);
670 		return -1;
671 	}
672 
673 	state->csum_size = btrfs_super_csum_size(selected_super);
674 
675 	for (pass = 0; pass < 3; pass++) {
676 		int num_copies;
677 		int mirror_num;
678 		u64 next_bytenr;
679 
680 		switch (pass) {
681 		case 0:
682 			next_bytenr = btrfs_super_root(selected_super);
683 			if (state->print_mask &
684 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
685 				pr_info("root@%llu\n", next_bytenr);
686 			break;
687 		case 1:
688 			next_bytenr = btrfs_super_chunk_root(selected_super);
689 			if (state->print_mask &
690 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
691 				pr_info("chunk@%llu\n", next_bytenr);
692 			break;
693 		case 2:
694 			next_bytenr = btrfs_super_log_root(selected_super);
695 			if (0 == next_bytenr)
696 				continue;
697 			if (state->print_mask &
698 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
699 				pr_info("log@%llu\n", next_bytenr);
700 			break;
701 		}
702 
703 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
704 					      state->metablock_size);
705 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
706 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
707 			       next_bytenr, num_copies);
708 
709 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
710 			struct btrfsic_block *next_block;
711 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
712 			struct btrfsic_block_link *l;
713 
714 			ret = btrfsic_map_block(state, next_bytenr,
715 						state->metablock_size,
716 						&tmp_next_block_ctx,
717 						mirror_num);
718 			if (ret) {
719 				pr_info("btrfsic: btrfsic_map_block(root @%llu, mirror %d) failed!\n",
720 				       next_bytenr, mirror_num);
721 				kfree(selected_super);
722 				return -1;
723 			}
724 
725 			next_block = btrfsic_block_hashtable_lookup(
726 					tmp_next_block_ctx.dev->bdev,
727 					tmp_next_block_ctx.dev_bytenr,
728 					&state->block_hashtable);
729 			BUG_ON(NULL == next_block);
730 
731 			l = btrfsic_block_link_hashtable_lookup(
732 					tmp_next_block_ctx.dev->bdev,
733 					tmp_next_block_ctx.dev_bytenr,
734 					state->latest_superblock->dev_state->
735 					bdev,
736 					state->latest_superblock->dev_bytenr,
737 					&state->block_link_hashtable);
738 			BUG_ON(NULL == l);
739 
740 			ret = btrfsic_read_block(state, &tmp_next_block_ctx);
741 			if (ret < (int)PAGE_SIZE) {
742 				pr_info("btrfsic: read @logical %llu failed!\n",
743 				       tmp_next_block_ctx.start);
744 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
745 				kfree(selected_super);
746 				return -1;
747 			}
748 
749 			ret = btrfsic_process_metablock(state,
750 							next_block,
751 							&tmp_next_block_ctx,
752 							BTRFS_MAX_LEVEL + 3, 1);
753 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
754 		}
755 	}
756 
757 	kfree(selected_super);
758 	return ret;
759 }
760 
761 static int btrfsic_process_superblock_dev_mirror(
762 		struct btrfsic_state *state,
763 		struct btrfsic_dev_state *dev_state,
764 		struct btrfs_device *device,
765 		int superblock_mirror_num,
766 		struct btrfsic_dev_state **selected_dev_state,
767 		struct btrfs_super_block *selected_super)
768 {
769 	struct btrfs_fs_info *fs_info = state->fs_info;
770 	struct btrfs_super_block *super_tmp;
771 	u64 dev_bytenr;
772 	struct buffer_head *bh;
773 	struct btrfsic_block *superblock_tmp;
774 	int pass;
775 	struct block_device *const superblock_bdev = device->bdev;
776 
777 	/* super block bytenr is always the unmapped device bytenr */
778 	dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
779 	if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->commit_total_bytes)
780 		return -1;
781 	bh = __bread(superblock_bdev, dev_bytenr / BTRFS_BDEV_BLOCKSIZE,
782 		     BTRFS_SUPER_INFO_SIZE);
783 	if (NULL == bh)
784 		return -1;
785 	super_tmp = (struct btrfs_super_block *)
786 	    (bh->b_data + (dev_bytenr & (BTRFS_BDEV_BLOCKSIZE - 1)));
787 
788 	if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
789 	    btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
790 	    memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
791 	    btrfs_super_nodesize(super_tmp) != state->metablock_size ||
792 	    btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
793 		brelse(bh);
794 		return 0;
795 	}
796 
797 	superblock_tmp =
798 	    btrfsic_block_hashtable_lookup(superblock_bdev,
799 					   dev_bytenr,
800 					   &state->block_hashtable);
801 	if (NULL == superblock_tmp) {
802 		superblock_tmp = btrfsic_block_alloc();
803 		if (NULL == superblock_tmp) {
804 			pr_info("btrfsic: error, kmalloc failed!\n");
805 			brelse(bh);
806 			return -1;
807 		}
808 		/* for superblock, only the dev_bytenr makes sense */
809 		superblock_tmp->dev_bytenr = dev_bytenr;
810 		superblock_tmp->dev_state = dev_state;
811 		superblock_tmp->logical_bytenr = dev_bytenr;
812 		superblock_tmp->generation = btrfs_super_generation(super_tmp);
813 		superblock_tmp->is_metadata = 1;
814 		superblock_tmp->is_superblock = 1;
815 		superblock_tmp->is_iodone = 1;
816 		superblock_tmp->never_written = 0;
817 		superblock_tmp->mirror_num = 1 + superblock_mirror_num;
818 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
819 			btrfs_info_in_rcu(fs_info,
820 				"new initial S-block (bdev %p, %s) @%llu (%s/%llu/%d)",
821 				     superblock_bdev,
822 				     rcu_str_deref(device->name), dev_bytenr,
823 				     dev_state->name, dev_bytenr,
824 				     superblock_mirror_num);
825 		list_add(&superblock_tmp->all_blocks_node,
826 			 &state->all_blocks_list);
827 		btrfsic_block_hashtable_add(superblock_tmp,
828 					    &state->block_hashtable);
829 	}
830 
831 	/* select the one with the highest generation field */
832 	if (btrfs_super_generation(super_tmp) >
833 	    state->max_superblock_generation ||
834 	    0 == state->max_superblock_generation) {
835 		memcpy(selected_super, super_tmp, sizeof(*selected_super));
836 		*selected_dev_state = dev_state;
837 		state->max_superblock_generation =
838 		    btrfs_super_generation(super_tmp);
839 		state->latest_superblock = superblock_tmp;
840 	}
841 
842 	for (pass = 0; pass < 3; pass++) {
843 		u64 next_bytenr;
844 		int num_copies;
845 		int mirror_num;
846 		const char *additional_string = NULL;
847 		struct btrfs_disk_key tmp_disk_key;
848 
849 		tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
850 		tmp_disk_key.offset = 0;
851 		switch (pass) {
852 		case 0:
853 			btrfs_set_disk_key_objectid(&tmp_disk_key,
854 						    BTRFS_ROOT_TREE_OBJECTID);
855 			additional_string = "initial root ";
856 			next_bytenr = btrfs_super_root(super_tmp);
857 			break;
858 		case 1:
859 			btrfs_set_disk_key_objectid(&tmp_disk_key,
860 						    BTRFS_CHUNK_TREE_OBJECTID);
861 			additional_string = "initial chunk ";
862 			next_bytenr = btrfs_super_chunk_root(super_tmp);
863 			break;
864 		case 2:
865 			btrfs_set_disk_key_objectid(&tmp_disk_key,
866 						    BTRFS_TREE_LOG_OBJECTID);
867 			additional_string = "initial log ";
868 			next_bytenr = btrfs_super_log_root(super_tmp);
869 			if (0 == next_bytenr)
870 				continue;
871 			break;
872 		}
873 
874 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
875 					      state->metablock_size);
876 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
877 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
878 			       next_bytenr, num_copies);
879 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
880 			struct btrfsic_block *next_block;
881 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
882 			struct btrfsic_block_link *l;
883 
884 			if (btrfsic_map_block(state, next_bytenr,
885 					      state->metablock_size,
886 					      &tmp_next_block_ctx,
887 					      mirror_num)) {
888 				pr_info("btrfsic: btrfsic_map_block(bytenr @%llu, mirror %d) failed!\n",
889 				       next_bytenr, mirror_num);
890 				brelse(bh);
891 				return -1;
892 			}
893 
894 			next_block = btrfsic_block_lookup_or_add(
895 					state, &tmp_next_block_ctx,
896 					additional_string, 1, 1, 0,
897 					mirror_num, NULL);
898 			if (NULL == next_block) {
899 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
900 				brelse(bh);
901 				return -1;
902 			}
903 
904 			next_block->disk_key = tmp_disk_key;
905 			next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
906 			l = btrfsic_block_link_lookup_or_add(
907 					state, &tmp_next_block_ctx,
908 					next_block, superblock_tmp,
909 					BTRFSIC_GENERATION_UNKNOWN);
910 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
911 			if (NULL == l) {
912 				brelse(bh);
913 				return -1;
914 			}
915 		}
916 	}
917 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
918 		btrfsic_dump_tree_sub(state, superblock_tmp, 0);
919 
920 	brelse(bh);
921 	return 0;
922 }
923 
924 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
925 {
926 	struct btrfsic_stack_frame *sf;
927 
928 	sf = kzalloc(sizeof(*sf), GFP_NOFS);
929 	if (NULL == sf)
930 		pr_info("btrfsic: alloc memory failed!\n");
931 	else
932 		sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
933 	return sf;
934 }
935 
936 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
937 {
938 	BUG_ON(!(NULL == sf ||
939 		 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
940 	kfree(sf);
941 }
942 
943 static int btrfsic_process_metablock(
944 		struct btrfsic_state *state,
945 		struct btrfsic_block *const first_block,
946 		struct btrfsic_block_data_ctx *const first_block_ctx,
947 		int first_limit_nesting, int force_iodone_flag)
948 {
949 	struct btrfsic_stack_frame initial_stack_frame = { 0 };
950 	struct btrfsic_stack_frame *sf;
951 	struct btrfsic_stack_frame *next_stack;
952 	struct btrfs_header *const first_hdr =
953 		(struct btrfs_header *)first_block_ctx->datav[0];
954 
955 	BUG_ON(!first_hdr);
956 	sf = &initial_stack_frame;
957 	sf->error = 0;
958 	sf->i = -1;
959 	sf->limit_nesting = first_limit_nesting;
960 	sf->block = first_block;
961 	sf->block_ctx = first_block_ctx;
962 	sf->next_block = NULL;
963 	sf->hdr = first_hdr;
964 	sf->prev = NULL;
965 
966 continue_with_new_stack_frame:
967 	sf->block->generation = le64_to_cpu(sf->hdr->generation);
968 	if (0 == sf->hdr->level) {
969 		struct btrfs_leaf *const leafhdr =
970 		    (struct btrfs_leaf *)sf->hdr;
971 
972 		if (-1 == sf->i) {
973 			sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
974 
975 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
976 				pr_info("leaf %llu items %d generation %llu owner %llu\n",
977 				       sf->block_ctx->start, sf->nr,
978 				       btrfs_stack_header_generation(
979 					       &leafhdr->header),
980 				       btrfs_stack_header_owner(
981 					       &leafhdr->header));
982 		}
983 
984 continue_with_current_leaf_stack_frame:
985 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
986 			sf->i++;
987 			sf->num_copies = 0;
988 		}
989 
990 		if (sf->i < sf->nr) {
991 			struct btrfs_item disk_item;
992 			u32 disk_item_offset =
993 				(uintptr_t)(leafhdr->items + sf->i) -
994 				(uintptr_t)leafhdr;
995 			struct btrfs_disk_key *disk_key;
996 			u8 type;
997 			u32 item_offset;
998 			u32 item_size;
999 
1000 			if (disk_item_offset + sizeof(struct btrfs_item) >
1001 			    sf->block_ctx->len) {
1002 leaf_item_out_of_bounce_error:
1003 				pr_info("btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
1004 				       sf->block_ctx->start,
1005 				       sf->block_ctx->dev->name);
1006 				goto one_stack_frame_backwards;
1007 			}
1008 			btrfsic_read_from_block_data(sf->block_ctx,
1009 						     &disk_item,
1010 						     disk_item_offset,
1011 						     sizeof(struct btrfs_item));
1012 			item_offset = btrfs_stack_item_offset(&disk_item);
1013 			item_size = btrfs_stack_item_size(&disk_item);
1014 			disk_key = &disk_item.key;
1015 			type = btrfs_disk_key_type(disk_key);
1016 
1017 			if (BTRFS_ROOT_ITEM_KEY == type) {
1018 				struct btrfs_root_item root_item;
1019 				u32 root_item_offset;
1020 				u64 next_bytenr;
1021 
1022 				root_item_offset = item_offset +
1023 					offsetof(struct btrfs_leaf, items);
1024 				if (root_item_offset + item_size >
1025 				    sf->block_ctx->len)
1026 					goto leaf_item_out_of_bounce_error;
1027 				btrfsic_read_from_block_data(
1028 					sf->block_ctx, &root_item,
1029 					root_item_offset,
1030 					item_size);
1031 				next_bytenr = btrfs_root_bytenr(&root_item);
1032 
1033 				sf->error =
1034 				    btrfsic_create_link_to_next_block(
1035 						state,
1036 						sf->block,
1037 						sf->block_ctx,
1038 						next_bytenr,
1039 						sf->limit_nesting,
1040 						&sf->next_block_ctx,
1041 						&sf->next_block,
1042 						force_iodone_flag,
1043 						&sf->num_copies,
1044 						&sf->mirror_num,
1045 						disk_key,
1046 						btrfs_root_generation(
1047 						&root_item));
1048 				if (sf->error)
1049 					goto one_stack_frame_backwards;
1050 
1051 				if (NULL != sf->next_block) {
1052 					struct btrfs_header *const next_hdr =
1053 					    (struct btrfs_header *)
1054 					    sf->next_block_ctx.datav[0];
1055 
1056 					next_stack =
1057 					    btrfsic_stack_frame_alloc();
1058 					if (NULL == next_stack) {
1059 						sf->error = -1;
1060 						btrfsic_release_block_ctx(
1061 								&sf->
1062 								next_block_ctx);
1063 						goto one_stack_frame_backwards;
1064 					}
1065 
1066 					next_stack->i = -1;
1067 					next_stack->block = sf->next_block;
1068 					next_stack->block_ctx =
1069 					    &sf->next_block_ctx;
1070 					next_stack->next_block = NULL;
1071 					next_stack->hdr = next_hdr;
1072 					next_stack->limit_nesting =
1073 					    sf->limit_nesting - 1;
1074 					next_stack->prev = sf;
1075 					sf = next_stack;
1076 					goto continue_with_new_stack_frame;
1077 				}
1078 			} else if (BTRFS_EXTENT_DATA_KEY == type &&
1079 				   state->include_extent_data) {
1080 				sf->error = btrfsic_handle_extent_data(
1081 						state,
1082 						sf->block,
1083 						sf->block_ctx,
1084 						item_offset,
1085 						force_iodone_flag);
1086 				if (sf->error)
1087 					goto one_stack_frame_backwards;
1088 			}
1089 
1090 			goto continue_with_current_leaf_stack_frame;
1091 		}
1092 	} else {
1093 		struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1094 
1095 		if (-1 == sf->i) {
1096 			sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1097 
1098 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1099 				pr_info("node %llu level %d items %d generation %llu owner %llu\n",
1100 				       sf->block_ctx->start,
1101 				       nodehdr->header.level, sf->nr,
1102 				       btrfs_stack_header_generation(
1103 				       &nodehdr->header),
1104 				       btrfs_stack_header_owner(
1105 				       &nodehdr->header));
1106 		}
1107 
1108 continue_with_current_node_stack_frame:
1109 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1110 			sf->i++;
1111 			sf->num_copies = 0;
1112 		}
1113 
1114 		if (sf->i < sf->nr) {
1115 			struct btrfs_key_ptr key_ptr;
1116 			u32 key_ptr_offset;
1117 			u64 next_bytenr;
1118 
1119 			key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1120 					  (uintptr_t)nodehdr;
1121 			if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1122 			    sf->block_ctx->len) {
1123 				pr_info("btrfsic: node item out of bounce at logical %llu, dev %s\n",
1124 				       sf->block_ctx->start,
1125 				       sf->block_ctx->dev->name);
1126 				goto one_stack_frame_backwards;
1127 			}
1128 			btrfsic_read_from_block_data(
1129 				sf->block_ctx, &key_ptr, key_ptr_offset,
1130 				sizeof(struct btrfs_key_ptr));
1131 			next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1132 
1133 			sf->error = btrfsic_create_link_to_next_block(
1134 					state,
1135 					sf->block,
1136 					sf->block_ctx,
1137 					next_bytenr,
1138 					sf->limit_nesting,
1139 					&sf->next_block_ctx,
1140 					&sf->next_block,
1141 					force_iodone_flag,
1142 					&sf->num_copies,
1143 					&sf->mirror_num,
1144 					&key_ptr.key,
1145 					btrfs_stack_key_generation(&key_ptr));
1146 			if (sf->error)
1147 				goto one_stack_frame_backwards;
1148 
1149 			if (NULL != sf->next_block) {
1150 				struct btrfs_header *const next_hdr =
1151 				    (struct btrfs_header *)
1152 				    sf->next_block_ctx.datav[0];
1153 
1154 				next_stack = btrfsic_stack_frame_alloc();
1155 				if (NULL == next_stack) {
1156 					sf->error = -1;
1157 					goto one_stack_frame_backwards;
1158 				}
1159 
1160 				next_stack->i = -1;
1161 				next_stack->block = sf->next_block;
1162 				next_stack->block_ctx = &sf->next_block_ctx;
1163 				next_stack->next_block = NULL;
1164 				next_stack->hdr = next_hdr;
1165 				next_stack->limit_nesting =
1166 				    sf->limit_nesting - 1;
1167 				next_stack->prev = sf;
1168 				sf = next_stack;
1169 				goto continue_with_new_stack_frame;
1170 			}
1171 
1172 			goto continue_with_current_node_stack_frame;
1173 		}
1174 	}
1175 
1176 one_stack_frame_backwards:
1177 	if (NULL != sf->prev) {
1178 		struct btrfsic_stack_frame *const prev = sf->prev;
1179 
1180 		/* the one for the initial block is freed in the caller */
1181 		btrfsic_release_block_ctx(sf->block_ctx);
1182 
1183 		if (sf->error) {
1184 			prev->error = sf->error;
1185 			btrfsic_stack_frame_free(sf);
1186 			sf = prev;
1187 			goto one_stack_frame_backwards;
1188 		}
1189 
1190 		btrfsic_stack_frame_free(sf);
1191 		sf = prev;
1192 		goto continue_with_new_stack_frame;
1193 	} else {
1194 		BUG_ON(&initial_stack_frame != sf);
1195 	}
1196 
1197 	return sf->error;
1198 }
1199 
1200 static void btrfsic_read_from_block_data(
1201 	struct btrfsic_block_data_ctx *block_ctx,
1202 	void *dstv, u32 offset, size_t len)
1203 {
1204 	size_t cur;
1205 	size_t offset_in_page;
1206 	char *kaddr;
1207 	char *dst = (char *)dstv;
1208 	size_t start_offset = block_ctx->start & ((u64)PAGE_SIZE - 1);
1209 	unsigned long i = (start_offset + offset) >> PAGE_SHIFT;
1210 
1211 	WARN_ON(offset + len > block_ctx->len);
1212 	offset_in_page = (start_offset + offset) & (PAGE_SIZE - 1);
1213 
1214 	while (len > 0) {
1215 		cur = min(len, ((size_t)PAGE_SIZE - offset_in_page));
1216 		BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_SIZE));
1217 		kaddr = block_ctx->datav[i];
1218 		memcpy(dst, kaddr + offset_in_page, cur);
1219 
1220 		dst += cur;
1221 		len -= cur;
1222 		offset_in_page = 0;
1223 		i++;
1224 	}
1225 }
1226 
1227 static int btrfsic_create_link_to_next_block(
1228 		struct btrfsic_state *state,
1229 		struct btrfsic_block *block,
1230 		struct btrfsic_block_data_ctx *block_ctx,
1231 		u64 next_bytenr,
1232 		int limit_nesting,
1233 		struct btrfsic_block_data_ctx *next_block_ctx,
1234 		struct btrfsic_block **next_blockp,
1235 		int force_iodone_flag,
1236 		int *num_copiesp, int *mirror_nump,
1237 		struct btrfs_disk_key *disk_key,
1238 		u64 parent_generation)
1239 {
1240 	struct btrfs_fs_info *fs_info = state->fs_info;
1241 	struct btrfsic_block *next_block = NULL;
1242 	int ret;
1243 	struct btrfsic_block_link *l;
1244 	int did_alloc_block_link;
1245 	int block_was_created;
1246 
1247 	*next_blockp = NULL;
1248 	if (0 == *num_copiesp) {
1249 		*num_copiesp = btrfs_num_copies(fs_info, next_bytenr,
1250 						state->metablock_size);
1251 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1252 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
1253 			       next_bytenr, *num_copiesp);
1254 		*mirror_nump = 1;
1255 	}
1256 
1257 	if (*mirror_nump > *num_copiesp)
1258 		return 0;
1259 
1260 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1261 		pr_info("btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1262 		       *mirror_nump);
1263 	ret = btrfsic_map_block(state, next_bytenr,
1264 				state->metablock_size,
1265 				next_block_ctx, *mirror_nump);
1266 	if (ret) {
1267 		pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1268 		       next_bytenr, *mirror_nump);
1269 		btrfsic_release_block_ctx(next_block_ctx);
1270 		*next_blockp = NULL;
1271 		return -1;
1272 	}
1273 
1274 	next_block = btrfsic_block_lookup_or_add(state,
1275 						 next_block_ctx, "referenced ",
1276 						 1, force_iodone_flag,
1277 						 !force_iodone_flag,
1278 						 *mirror_nump,
1279 						 &block_was_created);
1280 	if (NULL == next_block) {
1281 		btrfsic_release_block_ctx(next_block_ctx);
1282 		*next_blockp = NULL;
1283 		return -1;
1284 	}
1285 	if (block_was_created) {
1286 		l = NULL;
1287 		next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1288 	} else {
1289 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1290 			if (next_block->logical_bytenr != next_bytenr &&
1291 			    !(!next_block->is_metadata &&
1292 			      0 == next_block->logical_bytenr))
1293 				pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1294 				       next_bytenr, next_block_ctx->dev->name,
1295 				       next_block_ctx->dev_bytenr, *mirror_nump,
1296 				       btrfsic_get_block_type(state,
1297 							      next_block),
1298 				       next_block->logical_bytenr);
1299 			else
1300 				pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1301 				       next_bytenr, next_block_ctx->dev->name,
1302 				       next_block_ctx->dev_bytenr, *mirror_nump,
1303 				       btrfsic_get_block_type(state,
1304 							      next_block));
1305 		}
1306 		next_block->logical_bytenr = next_bytenr;
1307 
1308 		next_block->mirror_num = *mirror_nump;
1309 		l = btrfsic_block_link_hashtable_lookup(
1310 				next_block_ctx->dev->bdev,
1311 				next_block_ctx->dev_bytenr,
1312 				block_ctx->dev->bdev,
1313 				block_ctx->dev_bytenr,
1314 				&state->block_link_hashtable);
1315 	}
1316 
1317 	next_block->disk_key = *disk_key;
1318 	if (NULL == l) {
1319 		l = btrfsic_block_link_alloc();
1320 		if (NULL == l) {
1321 			pr_info("btrfsic: error, kmalloc failed!\n");
1322 			btrfsic_release_block_ctx(next_block_ctx);
1323 			*next_blockp = NULL;
1324 			return -1;
1325 		}
1326 
1327 		did_alloc_block_link = 1;
1328 		l->block_ref_to = next_block;
1329 		l->block_ref_from = block;
1330 		l->ref_cnt = 1;
1331 		l->parent_generation = parent_generation;
1332 
1333 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1334 			btrfsic_print_add_link(state, l);
1335 
1336 		list_add(&l->node_ref_to, &block->ref_to_list);
1337 		list_add(&l->node_ref_from, &next_block->ref_from_list);
1338 
1339 		btrfsic_block_link_hashtable_add(l,
1340 						 &state->block_link_hashtable);
1341 	} else {
1342 		did_alloc_block_link = 0;
1343 		if (0 == limit_nesting) {
1344 			l->ref_cnt++;
1345 			l->parent_generation = parent_generation;
1346 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1347 				btrfsic_print_add_link(state, l);
1348 		}
1349 	}
1350 
1351 	if (limit_nesting > 0 && did_alloc_block_link) {
1352 		ret = btrfsic_read_block(state, next_block_ctx);
1353 		if (ret < (int)next_block_ctx->len) {
1354 			pr_info("btrfsic: read block @logical %llu failed!\n",
1355 			       next_bytenr);
1356 			btrfsic_release_block_ctx(next_block_ctx);
1357 			*next_blockp = NULL;
1358 			return -1;
1359 		}
1360 
1361 		*next_blockp = next_block;
1362 	} else {
1363 		*next_blockp = NULL;
1364 	}
1365 	(*mirror_nump)++;
1366 
1367 	return 0;
1368 }
1369 
1370 static int btrfsic_handle_extent_data(
1371 		struct btrfsic_state *state,
1372 		struct btrfsic_block *block,
1373 		struct btrfsic_block_data_ctx *block_ctx,
1374 		u32 item_offset, int force_iodone_flag)
1375 {
1376 	struct btrfs_fs_info *fs_info = state->fs_info;
1377 	struct btrfs_file_extent_item file_extent_item;
1378 	u64 file_extent_item_offset;
1379 	u64 next_bytenr;
1380 	u64 num_bytes;
1381 	u64 generation;
1382 	struct btrfsic_block_link *l;
1383 	int ret;
1384 
1385 	file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1386 				  item_offset;
1387 	if (file_extent_item_offset +
1388 	    offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1389 	    block_ctx->len) {
1390 		pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1391 		       block_ctx->start, block_ctx->dev->name);
1392 		return -1;
1393 	}
1394 
1395 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1396 		file_extent_item_offset,
1397 		offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1398 	if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1399 	    btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1400 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1401 			pr_info("extent_data: type %u, disk_bytenr = %llu\n",
1402 			       file_extent_item.type,
1403 			       btrfs_stack_file_extent_disk_bytenr(
1404 			       &file_extent_item));
1405 		return 0;
1406 	}
1407 
1408 	if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1409 	    block_ctx->len) {
1410 		pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1411 		       block_ctx->start, block_ctx->dev->name);
1412 		return -1;
1413 	}
1414 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1415 				     file_extent_item_offset,
1416 				     sizeof(struct btrfs_file_extent_item));
1417 	next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item);
1418 	if (btrfs_stack_file_extent_compression(&file_extent_item) ==
1419 	    BTRFS_COMPRESS_NONE) {
1420 		next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item);
1421 		num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1422 	} else {
1423 		num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item);
1424 	}
1425 	generation = btrfs_stack_file_extent_generation(&file_extent_item);
1426 
1427 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1428 		pr_info("extent_data: type %u, disk_bytenr = %llu, offset = %llu, num_bytes = %llu\n",
1429 		       file_extent_item.type,
1430 		       btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1431 		       btrfs_stack_file_extent_offset(&file_extent_item),
1432 		       num_bytes);
1433 	while (num_bytes > 0) {
1434 		u32 chunk_len;
1435 		int num_copies;
1436 		int mirror_num;
1437 
1438 		if (num_bytes > state->datablock_size)
1439 			chunk_len = state->datablock_size;
1440 		else
1441 			chunk_len = num_bytes;
1442 
1443 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
1444 					      state->datablock_size);
1445 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1446 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
1447 			       next_bytenr, num_copies);
1448 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1449 			struct btrfsic_block_data_ctx next_block_ctx;
1450 			struct btrfsic_block *next_block;
1451 			int block_was_created;
1452 
1453 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1454 				pr_info("btrfsic_handle_extent_data(mirror_num=%d)\n",
1455 					mirror_num);
1456 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1457 				pr_info("\tdisk_bytenr = %llu, num_bytes %u\n",
1458 				       next_bytenr, chunk_len);
1459 			ret = btrfsic_map_block(state, next_bytenr,
1460 						chunk_len, &next_block_ctx,
1461 						mirror_num);
1462 			if (ret) {
1463 				pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1464 				       next_bytenr, mirror_num);
1465 				return -1;
1466 			}
1467 
1468 			next_block = btrfsic_block_lookup_or_add(
1469 					state,
1470 					&next_block_ctx,
1471 					"referenced ",
1472 					0,
1473 					force_iodone_flag,
1474 					!force_iodone_flag,
1475 					mirror_num,
1476 					&block_was_created);
1477 			if (NULL == next_block) {
1478 				pr_info("btrfsic: error, kmalloc failed!\n");
1479 				btrfsic_release_block_ctx(&next_block_ctx);
1480 				return -1;
1481 			}
1482 			if (!block_was_created) {
1483 				if ((state->print_mask &
1484 				     BTRFSIC_PRINT_MASK_VERBOSE) &&
1485 				    next_block->logical_bytenr != next_bytenr &&
1486 				    !(!next_block->is_metadata &&
1487 				      0 == next_block->logical_bytenr)) {
1488 					pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, D, bytenr mismatch (!= stored %llu).\n",
1489 					       next_bytenr,
1490 					       next_block_ctx.dev->name,
1491 					       next_block_ctx.dev_bytenr,
1492 					       mirror_num,
1493 					       next_block->logical_bytenr);
1494 				}
1495 				next_block->logical_bytenr = next_bytenr;
1496 				next_block->mirror_num = mirror_num;
1497 			}
1498 
1499 			l = btrfsic_block_link_lookup_or_add(state,
1500 							     &next_block_ctx,
1501 							     next_block, block,
1502 							     generation);
1503 			btrfsic_release_block_ctx(&next_block_ctx);
1504 			if (NULL == l)
1505 				return -1;
1506 		}
1507 
1508 		next_bytenr += chunk_len;
1509 		num_bytes -= chunk_len;
1510 	}
1511 
1512 	return 0;
1513 }
1514 
1515 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1516 			     struct btrfsic_block_data_ctx *block_ctx_out,
1517 			     int mirror_num)
1518 {
1519 	struct btrfs_fs_info *fs_info = state->fs_info;
1520 	int ret;
1521 	u64 length;
1522 	struct btrfs_bio *multi = NULL;
1523 	struct btrfs_device *device;
1524 
1525 	length = len;
1526 	ret = btrfs_map_block(fs_info, BTRFS_MAP_READ,
1527 			      bytenr, &length, &multi, mirror_num);
1528 
1529 	if (ret) {
1530 		block_ctx_out->start = 0;
1531 		block_ctx_out->dev_bytenr = 0;
1532 		block_ctx_out->len = 0;
1533 		block_ctx_out->dev = NULL;
1534 		block_ctx_out->datav = NULL;
1535 		block_ctx_out->pagev = NULL;
1536 		block_ctx_out->mem_to_free = NULL;
1537 
1538 		return ret;
1539 	}
1540 
1541 	device = multi->stripes[0].dev;
1542 	block_ctx_out->dev = btrfsic_dev_state_lookup(device->bdev->bd_dev);
1543 	block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1544 	block_ctx_out->start = bytenr;
1545 	block_ctx_out->len = len;
1546 	block_ctx_out->datav = NULL;
1547 	block_ctx_out->pagev = NULL;
1548 	block_ctx_out->mem_to_free = NULL;
1549 
1550 	kfree(multi);
1551 	if (NULL == block_ctx_out->dev) {
1552 		ret = -ENXIO;
1553 		pr_info("btrfsic: error, cannot lookup dev (#1)!\n");
1554 	}
1555 
1556 	return ret;
1557 }
1558 
1559 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1560 {
1561 	if (block_ctx->mem_to_free) {
1562 		unsigned int num_pages;
1563 
1564 		BUG_ON(!block_ctx->datav);
1565 		BUG_ON(!block_ctx->pagev);
1566 		num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1567 			    PAGE_SHIFT;
1568 		while (num_pages > 0) {
1569 			num_pages--;
1570 			if (block_ctx->datav[num_pages]) {
1571 				kunmap(block_ctx->pagev[num_pages]);
1572 				block_ctx->datav[num_pages] = NULL;
1573 			}
1574 			if (block_ctx->pagev[num_pages]) {
1575 				__free_page(block_ctx->pagev[num_pages]);
1576 				block_ctx->pagev[num_pages] = NULL;
1577 			}
1578 		}
1579 
1580 		kfree(block_ctx->mem_to_free);
1581 		block_ctx->mem_to_free = NULL;
1582 		block_ctx->pagev = NULL;
1583 		block_ctx->datav = NULL;
1584 	}
1585 }
1586 
1587 static int btrfsic_read_block(struct btrfsic_state *state,
1588 			      struct btrfsic_block_data_ctx *block_ctx)
1589 {
1590 	unsigned int num_pages;
1591 	unsigned int i;
1592 	u64 dev_bytenr;
1593 	int ret;
1594 
1595 	BUG_ON(block_ctx->datav);
1596 	BUG_ON(block_ctx->pagev);
1597 	BUG_ON(block_ctx->mem_to_free);
1598 	if (block_ctx->dev_bytenr & ((u64)PAGE_SIZE - 1)) {
1599 		pr_info("btrfsic: read_block() with unaligned bytenr %llu\n",
1600 		       block_ctx->dev_bytenr);
1601 		return -1;
1602 	}
1603 
1604 	num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1605 		    PAGE_SHIFT;
1606 	block_ctx->mem_to_free = kcalloc(sizeof(*block_ctx->datav) +
1607 						sizeof(*block_ctx->pagev),
1608 					 num_pages, GFP_NOFS);
1609 	if (!block_ctx->mem_to_free)
1610 		return -ENOMEM;
1611 	block_ctx->datav = block_ctx->mem_to_free;
1612 	block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1613 	for (i = 0; i < num_pages; i++) {
1614 		block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1615 		if (!block_ctx->pagev[i])
1616 			return -1;
1617 	}
1618 
1619 	dev_bytenr = block_ctx->dev_bytenr;
1620 	for (i = 0; i < num_pages;) {
1621 		struct bio *bio;
1622 		unsigned int j;
1623 
1624 		bio = btrfs_io_bio_alloc(num_pages - i);
1625 		bio_set_dev(bio, block_ctx->dev->bdev);
1626 		bio->bi_iter.bi_sector = dev_bytenr >> 9;
1627 		bio_set_op_attrs(bio, REQ_OP_READ, 0);
1628 
1629 		for (j = i; j < num_pages; j++) {
1630 			ret = bio_add_page(bio, block_ctx->pagev[j],
1631 					   PAGE_SIZE, 0);
1632 			if (PAGE_SIZE != ret)
1633 				break;
1634 		}
1635 		if (j == i) {
1636 			pr_info("btrfsic: error, failed to add a single page!\n");
1637 			return -1;
1638 		}
1639 		if (submit_bio_wait(bio)) {
1640 			pr_info("btrfsic: read error at logical %llu dev %s!\n",
1641 			       block_ctx->start, block_ctx->dev->name);
1642 			bio_put(bio);
1643 			return -1;
1644 		}
1645 		bio_put(bio);
1646 		dev_bytenr += (j - i) * PAGE_SIZE;
1647 		i = j;
1648 	}
1649 	for (i = 0; i < num_pages; i++)
1650 		block_ctx->datav[i] = kmap(block_ctx->pagev[i]);
1651 
1652 	return block_ctx->len;
1653 }
1654 
1655 static void btrfsic_dump_database(struct btrfsic_state *state)
1656 {
1657 	const struct btrfsic_block *b_all;
1658 
1659 	BUG_ON(NULL == state);
1660 
1661 	pr_info("all_blocks_list:\n");
1662 	list_for_each_entry(b_all, &state->all_blocks_list, all_blocks_node) {
1663 		const struct btrfsic_block_link *l;
1664 
1665 		pr_info("%c-block @%llu (%s/%llu/%d)\n",
1666 		       btrfsic_get_block_type(state, b_all),
1667 		       b_all->logical_bytenr, b_all->dev_state->name,
1668 		       b_all->dev_bytenr, b_all->mirror_num);
1669 
1670 		list_for_each_entry(l, &b_all->ref_to_list, node_ref_to) {
1671 			pr_info(" %c @%llu (%s/%llu/%d) refers %u* to %c @%llu (%s/%llu/%d)\n",
1672 			       btrfsic_get_block_type(state, b_all),
1673 			       b_all->logical_bytenr, b_all->dev_state->name,
1674 			       b_all->dev_bytenr, b_all->mirror_num,
1675 			       l->ref_cnt,
1676 			       btrfsic_get_block_type(state, l->block_ref_to),
1677 			       l->block_ref_to->logical_bytenr,
1678 			       l->block_ref_to->dev_state->name,
1679 			       l->block_ref_to->dev_bytenr,
1680 			       l->block_ref_to->mirror_num);
1681 		}
1682 
1683 		list_for_each_entry(l, &b_all->ref_from_list, node_ref_from) {
1684 			pr_info(" %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
1685 			       btrfsic_get_block_type(state, b_all),
1686 			       b_all->logical_bytenr, b_all->dev_state->name,
1687 			       b_all->dev_bytenr, b_all->mirror_num,
1688 			       l->ref_cnt,
1689 			       btrfsic_get_block_type(state, l->block_ref_from),
1690 			       l->block_ref_from->logical_bytenr,
1691 			       l->block_ref_from->dev_state->name,
1692 			       l->block_ref_from->dev_bytenr,
1693 			       l->block_ref_from->mirror_num);
1694 		}
1695 
1696 		pr_info("\n");
1697 	}
1698 }
1699 
1700 /*
1701  * Test whether the disk block contains a tree block (leaf or node)
1702  * (note that this test fails for the super block)
1703  */
1704 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
1705 				     char **datav, unsigned int num_pages)
1706 {
1707 	struct btrfs_fs_info *fs_info = state->fs_info;
1708 	struct btrfs_header *h;
1709 	u8 csum[BTRFS_CSUM_SIZE];
1710 	u32 crc = ~(u32)0;
1711 	unsigned int i;
1712 
1713 	if (num_pages * PAGE_SIZE < state->metablock_size)
1714 		return 1; /* not metadata */
1715 	num_pages = state->metablock_size >> PAGE_SHIFT;
1716 	h = (struct btrfs_header *)datav[0];
1717 
1718 	if (memcmp(h->fsid, fs_info->fsid, BTRFS_FSID_SIZE))
1719 		return 1;
1720 
1721 	for (i = 0; i < num_pages; i++) {
1722 		u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1723 		size_t sublen = i ? PAGE_SIZE :
1724 				    (PAGE_SIZE - BTRFS_CSUM_SIZE);
1725 
1726 		crc = crc32c(crc, data, sublen);
1727 	}
1728 	btrfs_csum_final(crc, csum);
1729 	if (memcmp(csum, h->csum, state->csum_size))
1730 		return 1;
1731 
1732 	return 0; /* is metadata */
1733 }
1734 
1735 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1736 					  u64 dev_bytenr, char **mapped_datav,
1737 					  unsigned int num_pages,
1738 					  struct bio *bio, int *bio_is_patched,
1739 					  struct buffer_head *bh,
1740 					  int submit_bio_bh_rw)
1741 {
1742 	int is_metadata;
1743 	struct btrfsic_block *block;
1744 	struct btrfsic_block_data_ctx block_ctx;
1745 	int ret;
1746 	struct btrfsic_state *state = dev_state->state;
1747 	struct block_device *bdev = dev_state->bdev;
1748 	unsigned int processed_len;
1749 
1750 	if (NULL != bio_is_patched)
1751 		*bio_is_patched = 0;
1752 
1753 again:
1754 	if (num_pages == 0)
1755 		return;
1756 
1757 	processed_len = 0;
1758 	is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1759 						      num_pages));
1760 
1761 	block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1762 					       &state->block_hashtable);
1763 	if (NULL != block) {
1764 		u64 bytenr = 0;
1765 		struct btrfsic_block_link *l, *tmp;
1766 
1767 		if (block->is_superblock) {
1768 			bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1769 						    mapped_datav[0]);
1770 			if (num_pages * PAGE_SIZE <
1771 			    BTRFS_SUPER_INFO_SIZE) {
1772 				pr_info("btrfsic: cannot work with too short bios!\n");
1773 				return;
1774 			}
1775 			is_metadata = 1;
1776 			BUG_ON(BTRFS_SUPER_INFO_SIZE & (PAGE_SIZE - 1));
1777 			processed_len = BTRFS_SUPER_INFO_SIZE;
1778 			if (state->print_mask &
1779 			    BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1780 				pr_info("[before new superblock is written]:\n");
1781 				btrfsic_dump_tree_sub(state, block, 0);
1782 			}
1783 		}
1784 		if (is_metadata) {
1785 			if (!block->is_superblock) {
1786 				if (num_pages * PAGE_SIZE <
1787 				    state->metablock_size) {
1788 					pr_info("btrfsic: cannot work with too short bios!\n");
1789 					return;
1790 				}
1791 				processed_len = state->metablock_size;
1792 				bytenr = btrfs_stack_header_bytenr(
1793 						(struct btrfs_header *)
1794 						mapped_datav[0]);
1795 				btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1796 							       dev_state,
1797 							       dev_bytenr);
1798 			}
1799 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1800 				if (block->logical_bytenr != bytenr &&
1801 				    !(!block->is_metadata &&
1802 				      block->logical_bytenr == 0))
1803 					pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1804 					       bytenr, dev_state->name,
1805 					       dev_bytenr,
1806 					       block->mirror_num,
1807 					       btrfsic_get_block_type(state,
1808 								      block),
1809 					       block->logical_bytenr);
1810 				else
1811 					pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1812 					       bytenr, dev_state->name,
1813 					       dev_bytenr, block->mirror_num,
1814 					       btrfsic_get_block_type(state,
1815 								      block));
1816 			}
1817 			block->logical_bytenr = bytenr;
1818 		} else {
1819 			if (num_pages * PAGE_SIZE <
1820 			    state->datablock_size) {
1821 				pr_info("btrfsic: cannot work with too short bios!\n");
1822 				return;
1823 			}
1824 			processed_len = state->datablock_size;
1825 			bytenr = block->logical_bytenr;
1826 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1827 				pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1828 				       bytenr, dev_state->name, dev_bytenr,
1829 				       block->mirror_num,
1830 				       btrfsic_get_block_type(state, block));
1831 		}
1832 
1833 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1834 			pr_info("ref_to_list: %cE, ref_from_list: %cE\n",
1835 			       list_empty(&block->ref_to_list) ? ' ' : '!',
1836 			       list_empty(&block->ref_from_list) ? ' ' : '!');
1837 		if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1838 			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",
1839 			       btrfsic_get_block_type(state, block), bytenr,
1840 			       dev_state->name, dev_bytenr, block->mirror_num,
1841 			       block->generation,
1842 			       btrfs_disk_key_objectid(&block->disk_key),
1843 			       block->disk_key.type,
1844 			       btrfs_disk_key_offset(&block->disk_key),
1845 			       btrfs_stack_header_generation(
1846 				       (struct btrfs_header *) mapped_datav[0]),
1847 			       state->max_superblock_generation);
1848 			btrfsic_dump_tree(state);
1849 		}
1850 
1851 		if (!block->is_iodone && !block->never_written) {
1852 			pr_info("btrfs: attempt to overwrite %c-block @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu, which is not yet iodone!\n",
1853 			       btrfsic_get_block_type(state, block), bytenr,
1854 			       dev_state->name, dev_bytenr, block->mirror_num,
1855 			       block->generation,
1856 			       btrfs_stack_header_generation(
1857 				       (struct btrfs_header *)
1858 				       mapped_datav[0]));
1859 			/* it would not be safe to go on */
1860 			btrfsic_dump_tree(state);
1861 			goto continue_loop;
1862 		}
1863 
1864 		/*
1865 		 * Clear all references of this block. Do not free
1866 		 * the block itself even if is not referenced anymore
1867 		 * because it still carries valuable information
1868 		 * like whether it was ever written and IO completed.
1869 		 */
1870 		list_for_each_entry_safe(l, tmp, &block->ref_to_list,
1871 					 node_ref_to) {
1872 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1873 				btrfsic_print_rem_link(state, l);
1874 			l->ref_cnt--;
1875 			if (0 == l->ref_cnt) {
1876 				list_del(&l->node_ref_to);
1877 				list_del(&l->node_ref_from);
1878 				btrfsic_block_link_hashtable_remove(l);
1879 				btrfsic_block_link_free(l);
1880 			}
1881 		}
1882 
1883 		block_ctx.dev = dev_state;
1884 		block_ctx.dev_bytenr = dev_bytenr;
1885 		block_ctx.start = bytenr;
1886 		block_ctx.len = processed_len;
1887 		block_ctx.pagev = NULL;
1888 		block_ctx.mem_to_free = NULL;
1889 		block_ctx.datav = mapped_datav;
1890 
1891 		if (is_metadata || state->include_extent_data) {
1892 			block->never_written = 0;
1893 			block->iodone_w_error = 0;
1894 			if (NULL != bio) {
1895 				block->is_iodone = 0;
1896 				BUG_ON(NULL == bio_is_patched);
1897 				if (!*bio_is_patched) {
1898 					block->orig_bio_bh_private =
1899 					    bio->bi_private;
1900 					block->orig_bio_bh_end_io.bio =
1901 					    bio->bi_end_io;
1902 					block->next_in_same_bio = NULL;
1903 					bio->bi_private = block;
1904 					bio->bi_end_io = btrfsic_bio_end_io;
1905 					*bio_is_patched = 1;
1906 				} else {
1907 					struct btrfsic_block *chained_block =
1908 					    (struct btrfsic_block *)
1909 					    bio->bi_private;
1910 
1911 					BUG_ON(NULL == chained_block);
1912 					block->orig_bio_bh_private =
1913 					    chained_block->orig_bio_bh_private;
1914 					block->orig_bio_bh_end_io.bio =
1915 					    chained_block->orig_bio_bh_end_io.
1916 					    bio;
1917 					block->next_in_same_bio = chained_block;
1918 					bio->bi_private = block;
1919 				}
1920 			} else if (NULL != bh) {
1921 				block->is_iodone = 0;
1922 				block->orig_bio_bh_private = bh->b_private;
1923 				block->orig_bio_bh_end_io.bh = bh->b_end_io;
1924 				block->next_in_same_bio = NULL;
1925 				bh->b_private = block;
1926 				bh->b_end_io = btrfsic_bh_end_io;
1927 			} else {
1928 				block->is_iodone = 1;
1929 				block->orig_bio_bh_private = NULL;
1930 				block->orig_bio_bh_end_io.bio = NULL;
1931 				block->next_in_same_bio = NULL;
1932 			}
1933 		}
1934 
1935 		block->flush_gen = dev_state->last_flush_gen + 1;
1936 		block->submit_bio_bh_rw = submit_bio_bh_rw;
1937 		if (is_metadata) {
1938 			block->logical_bytenr = bytenr;
1939 			block->is_metadata = 1;
1940 			if (block->is_superblock) {
1941 				BUG_ON(PAGE_SIZE !=
1942 				       BTRFS_SUPER_INFO_SIZE);
1943 				ret = btrfsic_process_written_superblock(
1944 						state,
1945 						block,
1946 						(struct btrfs_super_block *)
1947 						mapped_datav[0]);
1948 				if (state->print_mask &
1949 				    BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
1950 					pr_info("[after new superblock is written]:\n");
1951 					btrfsic_dump_tree_sub(state, block, 0);
1952 				}
1953 			} else {
1954 				block->mirror_num = 0;	/* unknown */
1955 				ret = btrfsic_process_metablock(
1956 						state,
1957 						block,
1958 						&block_ctx,
1959 						0, 0);
1960 			}
1961 			if (ret)
1962 				pr_info("btrfsic: btrfsic_process_metablock(root @%llu) failed!\n",
1963 				       dev_bytenr);
1964 		} else {
1965 			block->is_metadata = 0;
1966 			block->mirror_num = 0;	/* unknown */
1967 			block->generation = BTRFSIC_GENERATION_UNKNOWN;
1968 			if (!state->include_extent_data
1969 			    && list_empty(&block->ref_from_list)) {
1970 				/*
1971 				 * disk block is overwritten with extent
1972 				 * data (not meta data) and we are configured
1973 				 * to not include extent data: take the
1974 				 * chance and free the block's memory
1975 				 */
1976 				btrfsic_block_hashtable_remove(block);
1977 				list_del(&block->all_blocks_node);
1978 				btrfsic_block_free(block);
1979 			}
1980 		}
1981 		btrfsic_release_block_ctx(&block_ctx);
1982 	} else {
1983 		/* block has not been found in hash table */
1984 		u64 bytenr;
1985 
1986 		if (!is_metadata) {
1987 			processed_len = state->datablock_size;
1988 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1989 				pr_info("Written block (%s/%llu/?) !found in hash table, D.\n",
1990 				       dev_state->name, dev_bytenr);
1991 			if (!state->include_extent_data) {
1992 				/* ignore that written D block */
1993 				goto continue_loop;
1994 			}
1995 
1996 			/* this is getting ugly for the
1997 			 * include_extent_data case... */
1998 			bytenr = 0;	/* unknown */
1999 		} else {
2000 			processed_len = state->metablock_size;
2001 			bytenr = btrfs_stack_header_bytenr(
2002 					(struct btrfs_header *)
2003 					mapped_datav[0]);
2004 			btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
2005 						       dev_bytenr);
2006 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2007 				pr_info("Written block @%llu (%s/%llu/?) !found in hash table, M.\n",
2008 				       bytenr, dev_state->name, dev_bytenr);
2009 		}
2010 
2011 		block_ctx.dev = dev_state;
2012 		block_ctx.dev_bytenr = dev_bytenr;
2013 		block_ctx.start = bytenr;
2014 		block_ctx.len = processed_len;
2015 		block_ctx.pagev = NULL;
2016 		block_ctx.mem_to_free = NULL;
2017 		block_ctx.datav = mapped_datav;
2018 
2019 		block = btrfsic_block_alloc();
2020 		if (NULL == block) {
2021 			pr_info("btrfsic: error, kmalloc failed!\n");
2022 			btrfsic_release_block_ctx(&block_ctx);
2023 			goto continue_loop;
2024 		}
2025 		block->dev_state = dev_state;
2026 		block->dev_bytenr = dev_bytenr;
2027 		block->logical_bytenr = bytenr;
2028 		block->is_metadata = is_metadata;
2029 		block->never_written = 0;
2030 		block->iodone_w_error = 0;
2031 		block->mirror_num = 0;	/* unknown */
2032 		block->flush_gen = dev_state->last_flush_gen + 1;
2033 		block->submit_bio_bh_rw = submit_bio_bh_rw;
2034 		if (NULL != bio) {
2035 			block->is_iodone = 0;
2036 			BUG_ON(NULL == bio_is_patched);
2037 			if (!*bio_is_patched) {
2038 				block->orig_bio_bh_private = bio->bi_private;
2039 				block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2040 				block->next_in_same_bio = NULL;
2041 				bio->bi_private = block;
2042 				bio->bi_end_io = btrfsic_bio_end_io;
2043 				*bio_is_patched = 1;
2044 			} else {
2045 				struct btrfsic_block *chained_block =
2046 				    (struct btrfsic_block *)
2047 				    bio->bi_private;
2048 
2049 				BUG_ON(NULL == chained_block);
2050 				block->orig_bio_bh_private =
2051 				    chained_block->orig_bio_bh_private;
2052 				block->orig_bio_bh_end_io.bio =
2053 				    chained_block->orig_bio_bh_end_io.bio;
2054 				block->next_in_same_bio = chained_block;
2055 				bio->bi_private = block;
2056 			}
2057 		} else if (NULL != bh) {
2058 			block->is_iodone = 0;
2059 			block->orig_bio_bh_private = bh->b_private;
2060 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2061 			block->next_in_same_bio = NULL;
2062 			bh->b_private = block;
2063 			bh->b_end_io = btrfsic_bh_end_io;
2064 		} else {
2065 			block->is_iodone = 1;
2066 			block->orig_bio_bh_private = NULL;
2067 			block->orig_bio_bh_end_io.bio = NULL;
2068 			block->next_in_same_bio = NULL;
2069 		}
2070 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2071 			pr_info("New written %c-block @%llu (%s/%llu/%d)\n",
2072 			       is_metadata ? 'M' : 'D',
2073 			       block->logical_bytenr, block->dev_state->name,
2074 			       block->dev_bytenr, block->mirror_num);
2075 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2076 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2077 
2078 		if (is_metadata) {
2079 			ret = btrfsic_process_metablock(state, block,
2080 							&block_ctx, 0, 0);
2081 			if (ret)
2082 				pr_info("btrfsic: process_metablock(root @%llu) failed!\n",
2083 				       dev_bytenr);
2084 		}
2085 		btrfsic_release_block_ctx(&block_ctx);
2086 	}
2087 
2088 continue_loop:
2089 	BUG_ON(!processed_len);
2090 	dev_bytenr += processed_len;
2091 	mapped_datav += processed_len >> PAGE_SHIFT;
2092 	num_pages -= processed_len >> PAGE_SHIFT;
2093 	goto again;
2094 }
2095 
2096 static void btrfsic_bio_end_io(struct bio *bp)
2097 {
2098 	struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2099 	int iodone_w_error;
2100 
2101 	/* mutex is not held! This is not save if IO is not yet completed
2102 	 * on umount */
2103 	iodone_w_error = 0;
2104 	if (bp->bi_status)
2105 		iodone_w_error = 1;
2106 
2107 	BUG_ON(NULL == block);
2108 	bp->bi_private = block->orig_bio_bh_private;
2109 	bp->bi_end_io = block->orig_bio_bh_end_io.bio;
2110 
2111 	do {
2112 		struct btrfsic_block *next_block;
2113 		struct btrfsic_dev_state *const dev_state = block->dev_state;
2114 
2115 		if ((dev_state->state->print_mask &
2116 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2117 			pr_info("bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2118 			       bp->bi_status,
2119 			       btrfsic_get_block_type(dev_state->state, block),
2120 			       block->logical_bytenr, dev_state->name,
2121 			       block->dev_bytenr, block->mirror_num);
2122 		next_block = block->next_in_same_bio;
2123 		block->iodone_w_error = iodone_w_error;
2124 		if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2125 			dev_state->last_flush_gen++;
2126 			if ((dev_state->state->print_mask &
2127 			     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2128 				pr_info("bio_end_io() new %s flush_gen=%llu\n",
2129 				       dev_state->name,
2130 				       dev_state->last_flush_gen);
2131 		}
2132 		if (block->submit_bio_bh_rw & REQ_FUA)
2133 			block->flush_gen = 0; /* FUA completed means block is
2134 					       * on disk */
2135 		block->is_iodone = 1; /* for FLUSH, this releases the block */
2136 		block = next_block;
2137 	} while (NULL != block);
2138 
2139 	bp->bi_end_io(bp);
2140 }
2141 
2142 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate)
2143 {
2144 	struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private;
2145 	int iodone_w_error = !uptodate;
2146 	struct btrfsic_dev_state *dev_state;
2147 
2148 	BUG_ON(NULL == block);
2149 	dev_state = block->dev_state;
2150 	if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2151 		pr_info("bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n",
2152 		       iodone_w_error,
2153 		       btrfsic_get_block_type(dev_state->state, block),
2154 		       block->logical_bytenr, block->dev_state->name,
2155 		       block->dev_bytenr, block->mirror_num);
2156 
2157 	block->iodone_w_error = iodone_w_error;
2158 	if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2159 		dev_state->last_flush_gen++;
2160 		if ((dev_state->state->print_mask &
2161 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2162 			pr_info("bh_end_io() new %s flush_gen=%llu\n",
2163 			       dev_state->name, dev_state->last_flush_gen);
2164 	}
2165 	if (block->submit_bio_bh_rw & REQ_FUA)
2166 		block->flush_gen = 0; /* FUA completed means block is on disk */
2167 
2168 	bh->b_private = block->orig_bio_bh_private;
2169 	bh->b_end_io = block->orig_bio_bh_end_io.bh;
2170 	block->is_iodone = 1; /* for FLUSH, this releases the block */
2171 	bh->b_end_io(bh, uptodate);
2172 }
2173 
2174 static int btrfsic_process_written_superblock(
2175 		struct btrfsic_state *state,
2176 		struct btrfsic_block *const superblock,
2177 		struct btrfs_super_block *const super_hdr)
2178 {
2179 	struct btrfs_fs_info *fs_info = state->fs_info;
2180 	int pass;
2181 
2182 	superblock->generation = btrfs_super_generation(super_hdr);
2183 	if (!(superblock->generation > state->max_superblock_generation ||
2184 	      0 == state->max_superblock_generation)) {
2185 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2186 			pr_info("btrfsic: superblock @%llu (%s/%llu/%d) with old gen %llu <= %llu\n",
2187 			       superblock->logical_bytenr,
2188 			       superblock->dev_state->name,
2189 			       superblock->dev_bytenr, superblock->mirror_num,
2190 			       btrfs_super_generation(super_hdr),
2191 			       state->max_superblock_generation);
2192 	} else {
2193 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2194 			pr_info("btrfsic: got new superblock @%llu (%s/%llu/%d) with new 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 
2201 		state->max_superblock_generation =
2202 		    btrfs_super_generation(super_hdr);
2203 		state->latest_superblock = superblock;
2204 	}
2205 
2206 	for (pass = 0; pass < 3; pass++) {
2207 		int ret;
2208 		u64 next_bytenr;
2209 		struct btrfsic_block *next_block;
2210 		struct btrfsic_block_data_ctx tmp_next_block_ctx;
2211 		struct btrfsic_block_link *l;
2212 		int num_copies;
2213 		int mirror_num;
2214 		const char *additional_string = NULL;
2215 		struct btrfs_disk_key tmp_disk_key = {0};
2216 
2217 		btrfs_set_disk_key_objectid(&tmp_disk_key,
2218 					    BTRFS_ROOT_ITEM_KEY);
2219 		btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2220 
2221 		switch (pass) {
2222 		case 0:
2223 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2224 						    BTRFS_ROOT_TREE_OBJECTID);
2225 			additional_string = "root ";
2226 			next_bytenr = btrfs_super_root(super_hdr);
2227 			if (state->print_mask &
2228 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2229 				pr_info("root@%llu\n", next_bytenr);
2230 			break;
2231 		case 1:
2232 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2233 						    BTRFS_CHUNK_TREE_OBJECTID);
2234 			additional_string = "chunk ";
2235 			next_bytenr = btrfs_super_chunk_root(super_hdr);
2236 			if (state->print_mask &
2237 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2238 				pr_info("chunk@%llu\n", next_bytenr);
2239 			break;
2240 		case 2:
2241 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2242 						    BTRFS_TREE_LOG_OBJECTID);
2243 			additional_string = "log ";
2244 			next_bytenr = btrfs_super_log_root(super_hdr);
2245 			if (0 == next_bytenr)
2246 				continue;
2247 			if (state->print_mask &
2248 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2249 				pr_info("log@%llu\n", next_bytenr);
2250 			break;
2251 		}
2252 
2253 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
2254 					      BTRFS_SUPER_INFO_SIZE);
2255 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2256 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
2257 			       next_bytenr, num_copies);
2258 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2259 			int was_created;
2260 
2261 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2262 				pr_info("btrfsic_process_written_superblock(mirror_num=%d)\n", mirror_num);
2263 			ret = btrfsic_map_block(state, next_bytenr,
2264 						BTRFS_SUPER_INFO_SIZE,
2265 						&tmp_next_block_ctx,
2266 						mirror_num);
2267 			if (ret) {
2268 				pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
2269 				       next_bytenr, mirror_num);
2270 				return -1;
2271 			}
2272 
2273 			next_block = btrfsic_block_lookup_or_add(
2274 					state,
2275 					&tmp_next_block_ctx,
2276 					additional_string,
2277 					1, 0, 1,
2278 					mirror_num,
2279 					&was_created);
2280 			if (NULL == next_block) {
2281 				pr_info("btrfsic: error, kmalloc failed!\n");
2282 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
2283 				return -1;
2284 			}
2285 
2286 			next_block->disk_key = tmp_disk_key;
2287 			if (was_created)
2288 				next_block->generation =
2289 				    BTRFSIC_GENERATION_UNKNOWN;
2290 			l = btrfsic_block_link_lookup_or_add(
2291 					state,
2292 					&tmp_next_block_ctx,
2293 					next_block,
2294 					superblock,
2295 					BTRFSIC_GENERATION_UNKNOWN);
2296 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
2297 			if (NULL == l)
2298 				return -1;
2299 		}
2300 	}
2301 
2302 	if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2303 		btrfsic_dump_tree(state);
2304 
2305 	return 0;
2306 }
2307 
2308 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2309 					struct btrfsic_block *const block,
2310 					int recursion_level)
2311 {
2312 	const struct btrfsic_block_link *l;
2313 	int ret = 0;
2314 
2315 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2316 		/*
2317 		 * Note that this situation can happen and does not
2318 		 * indicate an error in regular cases. It happens
2319 		 * when disk blocks are freed and later reused.
2320 		 * The check-integrity module is not aware of any
2321 		 * block free operations, it just recognizes block
2322 		 * write operations. Therefore it keeps the linkage
2323 		 * information for a block until a block is
2324 		 * rewritten. This can temporarily cause incorrect
2325 		 * and even circular linkage informations. This
2326 		 * causes no harm unless such blocks are referenced
2327 		 * by the most recent super block.
2328 		 */
2329 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2330 			pr_info("btrfsic: abort cyclic linkage (case 1).\n");
2331 
2332 		return ret;
2333 	}
2334 
2335 	/*
2336 	 * This algorithm is recursive because the amount of used stack
2337 	 * space is very small and the max recursion depth is limited.
2338 	 */
2339 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2340 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2341 			pr_info("rl=%d, %c @%llu (%s/%llu/%d) %u* refers to %c @%llu (%s/%llu/%d)\n",
2342 			       recursion_level,
2343 			       btrfsic_get_block_type(state, block),
2344 			       block->logical_bytenr, block->dev_state->name,
2345 			       block->dev_bytenr, block->mirror_num,
2346 			       l->ref_cnt,
2347 			       btrfsic_get_block_type(state, l->block_ref_to),
2348 			       l->block_ref_to->logical_bytenr,
2349 			       l->block_ref_to->dev_state->name,
2350 			       l->block_ref_to->dev_bytenr,
2351 			       l->block_ref_to->mirror_num);
2352 		if (l->block_ref_to->never_written) {
2353 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is never written!\n",
2354 			       btrfsic_get_block_type(state, l->block_ref_to),
2355 			       l->block_ref_to->logical_bytenr,
2356 			       l->block_ref_to->dev_state->name,
2357 			       l->block_ref_to->dev_bytenr,
2358 			       l->block_ref_to->mirror_num);
2359 			ret = -1;
2360 		} else if (!l->block_ref_to->is_iodone) {
2361 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is not yet iodone!\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->iodone_w_error) {
2369 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which has write error!\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->parent_generation !=
2377 			   l->block_ref_to->generation &&
2378 			   BTRFSIC_GENERATION_UNKNOWN !=
2379 			   l->parent_generation &&
2380 			   BTRFSIC_GENERATION_UNKNOWN !=
2381 			   l->block_ref_to->generation) {
2382 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) with generation %llu != parent generation %llu!\n",
2383 			       btrfsic_get_block_type(state, l->block_ref_to),
2384 			       l->block_ref_to->logical_bytenr,
2385 			       l->block_ref_to->dev_state->name,
2386 			       l->block_ref_to->dev_bytenr,
2387 			       l->block_ref_to->mirror_num,
2388 			       l->block_ref_to->generation,
2389 			       l->parent_generation);
2390 			ret = -1;
2391 		} else if (l->block_ref_to->flush_gen >
2392 			   l->block_ref_to->dev_state->last_flush_gen) {
2393 			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",
2394 			       btrfsic_get_block_type(state, l->block_ref_to),
2395 			       l->block_ref_to->logical_bytenr,
2396 			       l->block_ref_to->dev_state->name,
2397 			       l->block_ref_to->dev_bytenr,
2398 			       l->block_ref_to->mirror_num, block->flush_gen,
2399 			       l->block_ref_to->dev_state->last_flush_gen);
2400 			ret = -1;
2401 		} else if (-1 == btrfsic_check_all_ref_blocks(state,
2402 							      l->block_ref_to,
2403 							      recursion_level +
2404 							      1)) {
2405 			ret = -1;
2406 		}
2407 	}
2408 
2409 	return ret;
2410 }
2411 
2412 static int btrfsic_is_block_ref_by_superblock(
2413 		const struct btrfsic_state *state,
2414 		const struct btrfsic_block *block,
2415 		int recursion_level)
2416 {
2417 	const struct btrfsic_block_link *l;
2418 
2419 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2420 		/* refer to comment at "abort cyclic linkage (case 1)" */
2421 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2422 			pr_info("btrfsic: abort cyclic linkage (case 2).\n");
2423 
2424 		return 0;
2425 	}
2426 
2427 	/*
2428 	 * This algorithm is recursive because the amount of used stack space
2429 	 * is very small and the max recursion depth is limited.
2430 	 */
2431 	list_for_each_entry(l, &block->ref_from_list, node_ref_from) {
2432 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2433 			pr_info("rl=%d, %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
2434 			       recursion_level,
2435 			       btrfsic_get_block_type(state, block),
2436 			       block->logical_bytenr, block->dev_state->name,
2437 			       block->dev_bytenr, block->mirror_num,
2438 			       l->ref_cnt,
2439 			       btrfsic_get_block_type(state, l->block_ref_from),
2440 			       l->block_ref_from->logical_bytenr,
2441 			       l->block_ref_from->dev_state->name,
2442 			       l->block_ref_from->dev_bytenr,
2443 			       l->block_ref_from->mirror_num);
2444 		if (l->block_ref_from->is_superblock &&
2445 		    state->latest_superblock->dev_bytenr ==
2446 		    l->block_ref_from->dev_bytenr &&
2447 		    state->latest_superblock->dev_state->bdev ==
2448 		    l->block_ref_from->dev_state->bdev)
2449 			return 1;
2450 		else if (btrfsic_is_block_ref_by_superblock(state,
2451 							    l->block_ref_from,
2452 							    recursion_level +
2453 							    1))
2454 			return 1;
2455 	}
2456 
2457 	return 0;
2458 }
2459 
2460 static void btrfsic_print_add_link(const struct btrfsic_state *state,
2461 				   const struct btrfsic_block_link *l)
2462 {
2463 	pr_info("Add %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2464 	       l->ref_cnt,
2465 	       btrfsic_get_block_type(state, l->block_ref_from),
2466 	       l->block_ref_from->logical_bytenr,
2467 	       l->block_ref_from->dev_state->name,
2468 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2469 	       btrfsic_get_block_type(state, l->block_ref_to),
2470 	       l->block_ref_to->logical_bytenr,
2471 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2472 	       l->block_ref_to->mirror_num);
2473 }
2474 
2475 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2476 				   const struct btrfsic_block_link *l)
2477 {
2478 	pr_info("Rem %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2479 	       l->ref_cnt,
2480 	       btrfsic_get_block_type(state, l->block_ref_from),
2481 	       l->block_ref_from->logical_bytenr,
2482 	       l->block_ref_from->dev_state->name,
2483 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2484 	       btrfsic_get_block_type(state, l->block_ref_to),
2485 	       l->block_ref_to->logical_bytenr,
2486 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2487 	       l->block_ref_to->mirror_num);
2488 }
2489 
2490 static char btrfsic_get_block_type(const struct btrfsic_state *state,
2491 				   const struct btrfsic_block *block)
2492 {
2493 	if (block->is_superblock &&
2494 	    state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2495 	    state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2496 		return 'S';
2497 	else if (block->is_superblock)
2498 		return 's';
2499 	else if (block->is_metadata)
2500 		return 'M';
2501 	else
2502 		return 'D';
2503 }
2504 
2505 static void btrfsic_dump_tree(const struct btrfsic_state *state)
2506 {
2507 	btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2508 }
2509 
2510 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2511 				  const struct btrfsic_block *block,
2512 				  int indent_level)
2513 {
2514 	const struct btrfsic_block_link *l;
2515 	int indent_add;
2516 	static char buf[80];
2517 	int cursor_position;
2518 
2519 	/*
2520 	 * Should better fill an on-stack buffer with a complete line and
2521 	 * dump it at once when it is time to print a newline character.
2522 	 */
2523 
2524 	/*
2525 	 * This algorithm is recursive because the amount of used stack space
2526 	 * is very small and the max recursion depth is limited.
2527 	 */
2528 	indent_add = sprintf(buf, "%c-%llu(%s/%llu/%u)",
2529 			     btrfsic_get_block_type(state, block),
2530 			     block->logical_bytenr, block->dev_state->name,
2531 			     block->dev_bytenr, block->mirror_num);
2532 	if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2533 		printk("[...]\n");
2534 		return;
2535 	}
2536 	printk(buf);
2537 	indent_level += indent_add;
2538 	if (list_empty(&block->ref_to_list)) {
2539 		printk("\n");
2540 		return;
2541 	}
2542 	if (block->mirror_num > 1 &&
2543 	    !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2544 		printk(" [...]\n");
2545 		return;
2546 	}
2547 
2548 	cursor_position = indent_level;
2549 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2550 		while (cursor_position < indent_level) {
2551 			printk(" ");
2552 			cursor_position++;
2553 		}
2554 		if (l->ref_cnt > 1)
2555 			indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2556 		else
2557 			indent_add = sprintf(buf, " --> ");
2558 		if (indent_level + indent_add >
2559 		    BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2560 			printk("[...]\n");
2561 			cursor_position = 0;
2562 			continue;
2563 		}
2564 
2565 		printk(buf);
2566 
2567 		btrfsic_dump_tree_sub(state, l->block_ref_to,
2568 				      indent_level + indent_add);
2569 		cursor_position = 0;
2570 	}
2571 }
2572 
2573 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2574 		struct btrfsic_state *state,
2575 		struct btrfsic_block_data_ctx *next_block_ctx,
2576 		struct btrfsic_block *next_block,
2577 		struct btrfsic_block *from_block,
2578 		u64 parent_generation)
2579 {
2580 	struct btrfsic_block_link *l;
2581 
2582 	l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2583 						next_block_ctx->dev_bytenr,
2584 						from_block->dev_state->bdev,
2585 						from_block->dev_bytenr,
2586 						&state->block_link_hashtable);
2587 	if (NULL == l) {
2588 		l = btrfsic_block_link_alloc();
2589 		if (NULL == l) {
2590 			pr_info("btrfsic: error, kmalloc failed!\n");
2591 			return NULL;
2592 		}
2593 
2594 		l->block_ref_to = next_block;
2595 		l->block_ref_from = from_block;
2596 		l->ref_cnt = 1;
2597 		l->parent_generation = parent_generation;
2598 
2599 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2600 			btrfsic_print_add_link(state, l);
2601 
2602 		list_add(&l->node_ref_to, &from_block->ref_to_list);
2603 		list_add(&l->node_ref_from, &next_block->ref_from_list);
2604 
2605 		btrfsic_block_link_hashtable_add(l,
2606 						 &state->block_link_hashtable);
2607 	} else {
2608 		l->ref_cnt++;
2609 		l->parent_generation = parent_generation;
2610 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2611 			btrfsic_print_add_link(state, l);
2612 	}
2613 
2614 	return l;
2615 }
2616 
2617 static struct btrfsic_block *btrfsic_block_lookup_or_add(
2618 		struct btrfsic_state *state,
2619 		struct btrfsic_block_data_ctx *block_ctx,
2620 		const char *additional_string,
2621 		int is_metadata,
2622 		int is_iodone,
2623 		int never_written,
2624 		int mirror_num,
2625 		int *was_created)
2626 {
2627 	struct btrfsic_block *block;
2628 
2629 	block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2630 					       block_ctx->dev_bytenr,
2631 					       &state->block_hashtable);
2632 	if (NULL == block) {
2633 		struct btrfsic_dev_state *dev_state;
2634 
2635 		block = btrfsic_block_alloc();
2636 		if (NULL == block) {
2637 			pr_info("btrfsic: error, kmalloc failed!\n");
2638 			return NULL;
2639 		}
2640 		dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev->bd_dev);
2641 		if (NULL == dev_state) {
2642 			pr_info("btrfsic: error, lookup dev_state failed!\n");
2643 			btrfsic_block_free(block);
2644 			return NULL;
2645 		}
2646 		block->dev_state = dev_state;
2647 		block->dev_bytenr = block_ctx->dev_bytenr;
2648 		block->logical_bytenr = block_ctx->start;
2649 		block->is_metadata = is_metadata;
2650 		block->is_iodone = is_iodone;
2651 		block->never_written = never_written;
2652 		block->mirror_num = mirror_num;
2653 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2654 			pr_info("New %s%c-block @%llu (%s/%llu/%d)\n",
2655 			       additional_string,
2656 			       btrfsic_get_block_type(state, block),
2657 			       block->logical_bytenr, dev_state->name,
2658 			       block->dev_bytenr, mirror_num);
2659 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2660 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2661 		if (NULL != was_created)
2662 			*was_created = 1;
2663 	} else {
2664 		if (NULL != was_created)
2665 			*was_created = 0;
2666 	}
2667 
2668 	return block;
2669 }
2670 
2671 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2672 					   u64 bytenr,
2673 					   struct btrfsic_dev_state *dev_state,
2674 					   u64 dev_bytenr)
2675 {
2676 	struct btrfs_fs_info *fs_info = state->fs_info;
2677 	struct btrfsic_block_data_ctx block_ctx;
2678 	int num_copies;
2679 	int mirror_num;
2680 	int match = 0;
2681 	int ret;
2682 
2683 	num_copies = btrfs_num_copies(fs_info, bytenr, state->metablock_size);
2684 
2685 	for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2686 		ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2687 					&block_ctx, mirror_num);
2688 		if (ret) {
2689 			pr_info("btrfsic: btrfsic_map_block(logical @%llu, mirror %d) failed!\n",
2690 			       bytenr, mirror_num);
2691 			continue;
2692 		}
2693 
2694 		if (dev_state->bdev == block_ctx.dev->bdev &&
2695 		    dev_bytenr == block_ctx.dev_bytenr) {
2696 			match++;
2697 			btrfsic_release_block_ctx(&block_ctx);
2698 			break;
2699 		}
2700 		btrfsic_release_block_ctx(&block_ctx);
2701 	}
2702 
2703 	if (WARN_ON(!match)) {
2704 		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",
2705 		       bytenr, dev_state->name, dev_bytenr);
2706 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2707 			ret = btrfsic_map_block(state, bytenr,
2708 						state->metablock_size,
2709 						&block_ctx, mirror_num);
2710 			if (ret)
2711 				continue;
2712 
2713 			pr_info("Read logical bytenr @%llu maps to (%s/%llu/%d)\n",
2714 			       bytenr, block_ctx.dev->name,
2715 			       block_ctx.dev_bytenr, mirror_num);
2716 		}
2717 	}
2718 }
2719 
2720 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev)
2721 {
2722 	return btrfsic_dev_state_hashtable_lookup(dev,
2723 						  &btrfsic_dev_state_hashtable);
2724 }
2725 
2726 int btrfsic_submit_bh(int op, int op_flags, struct buffer_head *bh)
2727 {
2728 	struct btrfsic_dev_state *dev_state;
2729 
2730 	if (!btrfsic_is_initialized)
2731 		return submit_bh(op, op_flags, bh);
2732 
2733 	mutex_lock(&btrfsic_mutex);
2734 	/* since btrfsic_submit_bh() might also be called before
2735 	 * btrfsic_mount(), this might return NULL */
2736 	dev_state = btrfsic_dev_state_lookup(bh->b_bdev->bd_dev);
2737 
2738 	/* Only called to write the superblock (incl. FLUSH/FUA) */
2739 	if (NULL != dev_state &&
2740 	    (op == REQ_OP_WRITE) && bh->b_size > 0) {
2741 		u64 dev_bytenr;
2742 
2743 		dev_bytenr = BTRFS_BDEV_BLOCKSIZE * bh->b_blocknr;
2744 		if (dev_state->state->print_mask &
2745 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2746 			pr_info("submit_bh(op=0x%x,0x%x, blocknr=%llu (bytenr %llu), size=%zu, data=%p, bdev=%p)\n",
2747 			       op, op_flags, (unsigned long long)bh->b_blocknr,
2748 			       dev_bytenr, bh->b_size, bh->b_data, bh->b_bdev);
2749 		btrfsic_process_written_block(dev_state, dev_bytenr,
2750 					      &bh->b_data, 1, NULL,
2751 					      NULL, bh, op_flags);
2752 	} else if (NULL != dev_state && (op_flags & REQ_PREFLUSH)) {
2753 		if (dev_state->state->print_mask &
2754 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2755 			pr_info("submit_bh(op=0x%x,0x%x FLUSH, bdev=%p)\n",
2756 			       op, op_flags, bh->b_bdev);
2757 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2758 			if ((dev_state->state->print_mask &
2759 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2760 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2761 				pr_info("btrfsic_submit_bh(%s) with FLUSH but dummy block already in use (ignored)!\n",
2762 				       dev_state->name);
2763 		} else {
2764 			struct btrfsic_block *const block =
2765 				&dev_state->dummy_block_for_bio_bh_flush;
2766 
2767 			block->is_iodone = 0;
2768 			block->never_written = 0;
2769 			block->iodone_w_error = 0;
2770 			block->flush_gen = dev_state->last_flush_gen + 1;
2771 			block->submit_bio_bh_rw = op_flags;
2772 			block->orig_bio_bh_private = bh->b_private;
2773 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2774 			block->next_in_same_bio = NULL;
2775 			bh->b_private = block;
2776 			bh->b_end_io = btrfsic_bh_end_io;
2777 		}
2778 	}
2779 	mutex_unlock(&btrfsic_mutex);
2780 	return submit_bh(op, op_flags, bh);
2781 }
2782 
2783 static void __btrfsic_submit_bio(struct bio *bio)
2784 {
2785 	struct btrfsic_dev_state *dev_state;
2786 
2787 	if (!btrfsic_is_initialized)
2788 		return;
2789 
2790 	mutex_lock(&btrfsic_mutex);
2791 	/* since btrfsic_submit_bio() is also called before
2792 	 * btrfsic_mount(), this might return NULL */
2793 	dev_state = btrfsic_dev_state_lookup(bio_dev(bio) + bio->bi_partno);
2794 	if (NULL != dev_state &&
2795 	    (bio_op(bio) == REQ_OP_WRITE) && bio_has_data(bio)) {
2796 		unsigned int i = 0;
2797 		u64 dev_bytenr;
2798 		u64 cur_bytenr;
2799 		struct bio_vec bvec;
2800 		struct bvec_iter iter;
2801 		int bio_is_patched;
2802 		char **mapped_datav;
2803 		unsigned int segs = bio_segments(bio);
2804 
2805 		dev_bytenr = 512 * bio->bi_iter.bi_sector;
2806 		bio_is_patched = 0;
2807 		if (dev_state->state->print_mask &
2808 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2809 			pr_info("submit_bio(rw=%d,0x%x, bi_vcnt=%u, bi_sector=%llu (bytenr %llu), bi_disk=%p)\n",
2810 			       bio_op(bio), bio->bi_opf, segs,
2811 			       (unsigned long long)bio->bi_iter.bi_sector,
2812 			       dev_bytenr, bio->bi_disk);
2813 
2814 		mapped_datav = kmalloc_array(segs,
2815 					     sizeof(*mapped_datav), GFP_NOFS);
2816 		if (!mapped_datav)
2817 			goto leave;
2818 		cur_bytenr = dev_bytenr;
2819 
2820 		bio_for_each_segment(bvec, bio, iter) {
2821 			BUG_ON(bvec.bv_len != PAGE_SIZE);
2822 			mapped_datav[i] = kmap(bvec.bv_page);
2823 			i++;
2824 
2825 			if (dev_state->state->print_mask &
2826 			    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
2827 				pr_info("#%u: bytenr=%llu, len=%u, offset=%u\n",
2828 				       i, cur_bytenr, bvec.bv_len, bvec.bv_offset);
2829 			cur_bytenr += bvec.bv_len;
2830 		}
2831 		btrfsic_process_written_block(dev_state, dev_bytenr,
2832 					      mapped_datav, segs,
2833 					      bio, &bio_is_patched,
2834 					      NULL, bio->bi_opf);
2835 		bio_for_each_segment(bvec, bio, iter)
2836 			kunmap(bvec.bv_page);
2837 		kfree(mapped_datav);
2838 	} else if (NULL != dev_state && (bio->bi_opf & REQ_PREFLUSH)) {
2839 		if (dev_state->state->print_mask &
2840 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2841 			pr_info("submit_bio(rw=%d,0x%x FLUSH, disk=%p)\n",
2842 			       bio_op(bio), bio->bi_opf, bio->bi_disk);
2843 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2844 			if ((dev_state->state->print_mask &
2845 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2846 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2847 				pr_info("btrfsic_submit_bio(%s) with FLUSH but dummy block already in use (ignored)!\n",
2848 				       dev_state->name);
2849 		} else {
2850 			struct btrfsic_block *const block =
2851 				&dev_state->dummy_block_for_bio_bh_flush;
2852 
2853 			block->is_iodone = 0;
2854 			block->never_written = 0;
2855 			block->iodone_w_error = 0;
2856 			block->flush_gen = dev_state->last_flush_gen + 1;
2857 			block->submit_bio_bh_rw = bio->bi_opf;
2858 			block->orig_bio_bh_private = bio->bi_private;
2859 			block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2860 			block->next_in_same_bio = NULL;
2861 			bio->bi_private = block;
2862 			bio->bi_end_io = btrfsic_bio_end_io;
2863 		}
2864 	}
2865 leave:
2866 	mutex_unlock(&btrfsic_mutex);
2867 }
2868 
2869 void btrfsic_submit_bio(struct bio *bio)
2870 {
2871 	__btrfsic_submit_bio(bio);
2872 	submit_bio(bio);
2873 }
2874 
2875 int btrfsic_submit_bio_wait(struct bio *bio)
2876 {
2877 	__btrfsic_submit_bio(bio);
2878 	return submit_bio_wait(bio);
2879 }
2880 
2881 int btrfsic_mount(struct btrfs_fs_info *fs_info,
2882 		  struct btrfs_fs_devices *fs_devices,
2883 		  int including_extent_data, u32 print_mask)
2884 {
2885 	int ret;
2886 	struct btrfsic_state *state;
2887 	struct list_head *dev_head = &fs_devices->devices;
2888 	struct btrfs_device *device;
2889 
2890 	if (fs_info->nodesize & ((u64)PAGE_SIZE - 1)) {
2891 		pr_info("btrfsic: cannot handle nodesize %d not being a multiple of PAGE_SIZE %ld!\n",
2892 		       fs_info->nodesize, PAGE_SIZE);
2893 		return -1;
2894 	}
2895 	if (fs_info->sectorsize & ((u64)PAGE_SIZE - 1)) {
2896 		pr_info("btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_SIZE %ld!\n",
2897 		       fs_info->sectorsize, PAGE_SIZE);
2898 		return -1;
2899 	}
2900 	state = kvzalloc(sizeof(*state), GFP_KERNEL);
2901 	if (!state) {
2902 		pr_info("btrfs check-integrity: allocation failed!\n");
2903 		return -ENOMEM;
2904 	}
2905 
2906 	if (!btrfsic_is_initialized) {
2907 		mutex_init(&btrfsic_mutex);
2908 		btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
2909 		btrfsic_is_initialized = 1;
2910 	}
2911 	mutex_lock(&btrfsic_mutex);
2912 	state->fs_info = fs_info;
2913 	state->print_mask = print_mask;
2914 	state->include_extent_data = including_extent_data;
2915 	state->csum_size = 0;
2916 	state->metablock_size = fs_info->nodesize;
2917 	state->datablock_size = fs_info->sectorsize;
2918 	INIT_LIST_HEAD(&state->all_blocks_list);
2919 	btrfsic_block_hashtable_init(&state->block_hashtable);
2920 	btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
2921 	state->max_superblock_generation = 0;
2922 	state->latest_superblock = NULL;
2923 
2924 	list_for_each_entry(device, dev_head, dev_list) {
2925 		struct btrfsic_dev_state *ds;
2926 		const char *p;
2927 
2928 		if (!device->bdev || !device->name)
2929 			continue;
2930 
2931 		ds = btrfsic_dev_state_alloc();
2932 		if (NULL == ds) {
2933 			pr_info("btrfs check-integrity: kmalloc() failed!\n");
2934 			mutex_unlock(&btrfsic_mutex);
2935 			return -ENOMEM;
2936 		}
2937 		ds->bdev = device->bdev;
2938 		ds->state = state;
2939 		bdevname(ds->bdev, ds->name);
2940 		ds->name[BDEVNAME_SIZE - 1] = '\0';
2941 		p = kbasename(ds->name);
2942 		strlcpy(ds->name, p, sizeof(ds->name));
2943 		btrfsic_dev_state_hashtable_add(ds,
2944 						&btrfsic_dev_state_hashtable);
2945 	}
2946 
2947 	ret = btrfsic_process_superblock(state, fs_devices);
2948 	if (0 != ret) {
2949 		mutex_unlock(&btrfsic_mutex);
2950 		btrfsic_unmount(fs_devices);
2951 		return ret;
2952 	}
2953 
2954 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
2955 		btrfsic_dump_database(state);
2956 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
2957 		btrfsic_dump_tree(state);
2958 
2959 	mutex_unlock(&btrfsic_mutex);
2960 	return 0;
2961 }
2962 
2963 void btrfsic_unmount(struct btrfs_fs_devices *fs_devices)
2964 {
2965 	struct btrfsic_block *b_all, *tmp_all;
2966 	struct btrfsic_state *state;
2967 	struct list_head *dev_head = &fs_devices->devices;
2968 	struct btrfs_device *device;
2969 
2970 	if (!btrfsic_is_initialized)
2971 		return;
2972 
2973 	mutex_lock(&btrfsic_mutex);
2974 
2975 	state = NULL;
2976 	list_for_each_entry(device, dev_head, dev_list) {
2977 		struct btrfsic_dev_state *ds;
2978 
2979 		if (!device->bdev || !device->name)
2980 			continue;
2981 
2982 		ds = btrfsic_dev_state_hashtable_lookup(
2983 				device->bdev->bd_dev,
2984 				&btrfsic_dev_state_hashtable);
2985 		if (NULL != ds) {
2986 			state = ds->state;
2987 			btrfsic_dev_state_hashtable_remove(ds);
2988 			btrfsic_dev_state_free(ds);
2989 		}
2990 	}
2991 
2992 	if (NULL == state) {
2993 		pr_info("btrfsic: error, cannot find state information on umount!\n");
2994 		mutex_unlock(&btrfsic_mutex);
2995 		return;
2996 	}
2997 
2998 	/*
2999 	 * Don't care about keeping the lists' state up to date,
3000 	 * just free all memory that was allocated dynamically.
3001 	 * Free the blocks and the block_links.
3002 	 */
3003 	list_for_each_entry_safe(b_all, tmp_all, &state->all_blocks_list,
3004 				 all_blocks_node) {
3005 		struct btrfsic_block_link *l, *tmp;
3006 
3007 		list_for_each_entry_safe(l, tmp, &b_all->ref_to_list,
3008 					 node_ref_to) {
3009 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
3010 				btrfsic_print_rem_link(state, l);
3011 
3012 			l->ref_cnt--;
3013 			if (0 == l->ref_cnt)
3014 				btrfsic_block_link_free(l);
3015 		}
3016 
3017 		if (b_all->is_iodone || b_all->never_written)
3018 			btrfsic_block_free(b_all);
3019 		else
3020 			pr_info("btrfs: attempt to free %c-block @%llu (%s/%llu/%d) on umount which is not yet iodone!\n",
3021 			       btrfsic_get_block_type(state, b_all),
3022 			       b_all->logical_bytenr, b_all->dev_state->name,
3023 			       b_all->dev_bytenr, b_all->mirror_num);
3024 	}
3025 
3026 	mutex_unlock(&btrfsic_mutex);
3027 
3028 	kvfree(state);
3029 }
3030