xref: /openbmc/linux/fs/btrfs/check-integrity.c (revision b34e08d5)
1 /*
2  * Copyright (C) STRATO AG 2011.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 /*
20  * This module can be used to catch cases when the btrfs kernel
21  * code executes write requests to the disk that bring the file
22  * system in an inconsistent state. In such a state, a power-loss
23  * or kernel panic event would cause that the data on disk is
24  * lost or at least damaged.
25  *
26  * Code is added that examines all block write requests during
27  * runtime (including writes of the super block). Three rules
28  * are verified and an error is printed on violation of the
29  * rules:
30  * 1. It is not allowed to write a disk block which is
31  *    currently referenced by the super block (either directly
32  *    or indirectly).
33  * 2. When a super block is written, it is verified that all
34  *    referenced (directly or indirectly) blocks fulfill the
35  *    following requirements:
36  *    2a. All referenced blocks have either been present when
37  *        the file system was mounted, (i.e., they have been
38  *        referenced by the super block) or they have been
39  *        written since then and the write completion callback
40  *        was called and no write error was indicated and a
41  *        FLUSH request to the device where these blocks are
42  *        located was received and completed.
43  *    2b. All referenced blocks need to have a generation
44  *        number which is equal to the parent's number.
45  *
46  * One issue that was found using this module was that the log
47  * tree on disk became temporarily corrupted because disk blocks
48  * that had been in use for the log tree had been freed and
49  * reused too early, while being referenced by the written super
50  * block.
51  *
52  * The search term in the kernel log that can be used to filter
53  * on the existence of detected integrity issues is
54  * "btrfs: attempt".
55  *
56  * The integrity check is enabled via mount options. These
57  * mount options are only supported if the integrity check
58  * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY.
59  *
60  * Example #1, apply integrity checks to all metadata:
61  * mount /dev/sdb1 /mnt -o check_int
62  *
63  * Example #2, apply integrity checks to all metadata and
64  * to data extents:
65  * mount /dev/sdb1 /mnt -o check_int_data
66  *
67  * Example #3, apply integrity checks to all metadata and dump
68  * the tree that the super block references to kernel messages
69  * each time after a super block was written:
70  * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263
71  *
72  * If the integrity check tool is included and activated in
73  * the mount options, plenty of kernel memory is used, and
74  * plenty of additional CPU cycles are spent. Enabling this
75  * functionality is not intended for normal use. In most
76  * cases, unless you are a btrfs developer who needs to verify
77  * the integrity of (super)-block write requests, do not
78  * enable the config option BTRFS_FS_CHECK_INTEGRITY to
79  * include and compile the integrity check tool.
80  *
81  * Expect millions of lines of information in the kernel log with an
82  * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the
83  * kernel config to at least 26 (which is 64MB). Usually the value is
84  * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be
85  * changed like this before LOG_BUF_SHIFT can be set to a high value:
86  * config LOG_BUF_SHIFT
87  *       int "Kernel log buffer size (16 => 64KB, 17 => 128KB)"
88  *       range 12 30
89  */
90 
91 #include <linux/sched.h>
92 #include <linux/slab.h>
93 #include <linux/buffer_head.h>
94 #include <linux/mutex.h>
95 #include <linux/genhd.h>
96 #include <linux/blkdev.h>
97 #include "ctree.h"
98 #include "disk-io.h"
99 #include "hash.h"
100 #include "transaction.h"
101 #include "extent_io.h"
102 #include "volumes.h"
103 #include "print-tree.h"
104 #include "locking.h"
105 #include "check-integrity.h"
106 #include "rcu-string.h"
107 
108 #define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
109 #define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
110 #define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
111 #define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
112 #define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
113 #define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
114 #define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
115 #define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6)	/* in characters,
116 							 * excluding " [...]" */
117 #define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
118 
119 /*
120  * The definition of the bitmask fields for the print_mask.
121  * They are specified with the mount option check_integrity_print_mask.
122  */
123 #define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE			0x00000001
124 #define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION		0x00000002
125 #define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE			0x00000004
126 #define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE			0x00000008
127 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH			0x00000010
128 #define BTRFSIC_PRINT_MASK_END_IO_BIO_BH			0x00000020
129 #define BTRFSIC_PRINT_MASK_VERBOSE				0x00000040
130 #define BTRFSIC_PRINT_MASK_VERY_VERBOSE				0x00000080
131 #define BTRFSIC_PRINT_MASK_INITIAL_TREE				0x00000100
132 #define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES			0x00000200
133 #define BTRFSIC_PRINT_MASK_INITIAL_DATABASE			0x00000400
134 #define BTRFSIC_PRINT_MASK_NUM_COPIES				0x00000800
135 #define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS		0x00001000
136 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE		0x00002000
137 
138 struct btrfsic_dev_state;
139 struct btrfsic_state;
140 
141 struct btrfsic_block {
142 	u32 magic_num;		/* only used for debug purposes */
143 	unsigned int is_metadata:1;	/* if it is meta-data, not data-data */
144 	unsigned int is_superblock:1;	/* if it is one of the superblocks */
145 	unsigned int is_iodone:1;	/* if is done by lower subsystem */
146 	unsigned int iodone_w_error:1;	/* error was indicated to endio */
147 	unsigned int never_written:1;	/* block was added because it was
148 					 * referenced, not because it was
149 					 * written */
150 	unsigned int mirror_num;	/* large enough to hold
151 					 * BTRFS_SUPER_MIRROR_MAX */
152 	struct btrfsic_dev_state *dev_state;
153 	u64 dev_bytenr;		/* key, physical byte num on disk */
154 	u64 logical_bytenr;	/* logical byte num on disk */
155 	u64 generation;
156 	struct btrfs_disk_key disk_key;	/* extra info to print in case of
157 					 * issues, will not always be correct */
158 	struct list_head collision_resolving_node;	/* list node */
159 	struct list_head all_blocks_node;	/* list node */
160 
161 	/* the following two lists contain block_link items */
162 	struct list_head ref_to_list;	/* list */
163 	struct list_head ref_from_list;	/* list */
164 	struct btrfsic_block *next_in_same_bio;
165 	void *orig_bio_bh_private;
166 	union {
167 		bio_end_io_t *bio;
168 		bh_end_io_t *bh;
169 	} orig_bio_bh_end_io;
170 	int submit_bio_bh_rw;
171 	u64 flush_gen; /* only valid if !never_written */
172 };
173 
174 /*
175  * Elements of this type are allocated dynamically and required because
176  * each block object can refer to and can be ref from multiple blocks.
177  * The key to lookup them in the hashtable is the dev_bytenr of
178  * the block ref to plus the one from the block refered from.
179  * The fact that they are searchable via a hashtable and that a
180  * ref_cnt is maintained is not required for the btrfs integrity
181  * check algorithm itself, it is only used to make the output more
182  * beautiful in case that an error is detected (an error is defined
183  * as a write operation to a block while that block is still referenced).
184  */
185 struct btrfsic_block_link {
186 	u32 magic_num;		/* only used for debug purposes */
187 	u32 ref_cnt;
188 	struct list_head node_ref_to;	/* list node */
189 	struct list_head node_ref_from;	/* list node */
190 	struct list_head collision_resolving_node;	/* list node */
191 	struct btrfsic_block *block_ref_to;
192 	struct btrfsic_block *block_ref_from;
193 	u64 parent_generation;
194 };
195 
196 struct btrfsic_dev_state {
197 	u32 magic_num;		/* only used for debug purposes */
198 	struct block_device *bdev;
199 	struct btrfsic_state *state;
200 	struct list_head collision_resolving_node;	/* list node */
201 	struct btrfsic_block dummy_block_for_bio_bh_flush;
202 	u64 last_flush_gen;
203 	char name[BDEVNAME_SIZE];
204 };
205 
206 struct btrfsic_block_hashtable {
207 	struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
208 };
209 
210 struct btrfsic_block_link_hashtable {
211 	struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
212 };
213 
214 struct btrfsic_dev_state_hashtable {
215 	struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
216 };
217 
218 struct btrfsic_block_data_ctx {
219 	u64 start;		/* virtual bytenr */
220 	u64 dev_bytenr;		/* physical bytenr on device */
221 	u32 len;
222 	struct btrfsic_dev_state *dev;
223 	char **datav;
224 	struct page **pagev;
225 	void *mem_to_free;
226 };
227 
228 /* This structure is used to implement recursion without occupying
229  * any stack space, refer to btrfsic_process_metablock() */
230 struct btrfsic_stack_frame {
231 	u32 magic;
232 	u32 nr;
233 	int error;
234 	int i;
235 	int limit_nesting;
236 	int num_copies;
237 	int mirror_num;
238 	struct btrfsic_block *block;
239 	struct btrfsic_block_data_ctx *block_ctx;
240 	struct btrfsic_block *next_block;
241 	struct btrfsic_block_data_ctx next_block_ctx;
242 	struct btrfs_header *hdr;
243 	struct btrfsic_stack_frame *prev;
244 };
245 
246 /* Some state per mounted filesystem */
247 struct btrfsic_state {
248 	u32 print_mask;
249 	int include_extent_data;
250 	int csum_size;
251 	struct list_head all_blocks_list;
252 	struct btrfsic_block_hashtable block_hashtable;
253 	struct btrfsic_block_link_hashtable block_link_hashtable;
254 	struct btrfs_root *root;
255 	u64 max_superblock_generation;
256 	struct btrfsic_block *latest_superblock;
257 	u32 metablock_size;
258 	u32 datablock_size;
259 };
260 
261 static void btrfsic_block_init(struct btrfsic_block *b);
262 static struct btrfsic_block *btrfsic_block_alloc(void);
263 static void btrfsic_block_free(struct btrfsic_block *b);
264 static void btrfsic_block_link_init(struct btrfsic_block_link *n);
265 static struct btrfsic_block_link *btrfsic_block_link_alloc(void);
266 static void btrfsic_block_link_free(struct btrfsic_block_link *n);
267 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds);
268 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void);
269 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds);
270 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h);
271 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
272 					struct btrfsic_block_hashtable *h);
273 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b);
274 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
275 		struct block_device *bdev,
276 		u64 dev_bytenr,
277 		struct btrfsic_block_hashtable *h);
278 static void btrfsic_block_link_hashtable_init(
279 		struct btrfsic_block_link_hashtable *h);
280 static void btrfsic_block_link_hashtable_add(
281 		struct btrfsic_block_link *l,
282 		struct btrfsic_block_link_hashtable *h);
283 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l);
284 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
285 		struct block_device *bdev_ref_to,
286 		u64 dev_bytenr_ref_to,
287 		struct block_device *bdev_ref_from,
288 		u64 dev_bytenr_ref_from,
289 		struct btrfsic_block_link_hashtable *h);
290 static void btrfsic_dev_state_hashtable_init(
291 		struct btrfsic_dev_state_hashtable *h);
292 static void btrfsic_dev_state_hashtable_add(
293 		struct btrfsic_dev_state *ds,
294 		struct btrfsic_dev_state_hashtable *h);
295 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds);
296 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(
297 		struct block_device *bdev,
298 		struct btrfsic_dev_state_hashtable *h);
299 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void);
300 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf);
301 static int btrfsic_process_superblock(struct btrfsic_state *state,
302 				      struct btrfs_fs_devices *fs_devices);
303 static int btrfsic_process_metablock(struct btrfsic_state *state,
304 				     struct btrfsic_block *block,
305 				     struct btrfsic_block_data_ctx *block_ctx,
306 				     int limit_nesting, int force_iodone_flag);
307 static void btrfsic_read_from_block_data(
308 	struct btrfsic_block_data_ctx *block_ctx,
309 	void *dst, u32 offset, size_t len);
310 static int btrfsic_create_link_to_next_block(
311 		struct btrfsic_state *state,
312 		struct btrfsic_block *block,
313 		struct btrfsic_block_data_ctx
314 		*block_ctx, u64 next_bytenr,
315 		int limit_nesting,
316 		struct btrfsic_block_data_ctx *next_block_ctx,
317 		struct btrfsic_block **next_blockp,
318 		int force_iodone_flag,
319 		int *num_copiesp, int *mirror_nump,
320 		struct btrfs_disk_key *disk_key,
321 		u64 parent_generation);
322 static int btrfsic_handle_extent_data(struct btrfsic_state *state,
323 				      struct btrfsic_block *block,
324 				      struct btrfsic_block_data_ctx *block_ctx,
325 				      u32 item_offset, int force_iodone_flag);
326 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
327 			     struct btrfsic_block_data_ctx *block_ctx_out,
328 			     int mirror_num);
329 static int btrfsic_map_superblock(struct btrfsic_state *state, u64 bytenr,
330 				  u32 len, struct block_device *bdev,
331 				  struct btrfsic_block_data_ctx *block_ctx_out);
332 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
333 static int btrfsic_read_block(struct btrfsic_state *state,
334 			      struct btrfsic_block_data_ctx *block_ctx);
335 static void btrfsic_dump_database(struct btrfsic_state *state);
336 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
337 				     char **datav, unsigned int num_pages);
338 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
339 					  u64 dev_bytenr, char **mapped_datav,
340 					  unsigned int num_pages,
341 					  struct bio *bio, int *bio_is_patched,
342 					  struct buffer_head *bh,
343 					  int submit_bio_bh_rw);
344 static int btrfsic_process_written_superblock(
345 		struct btrfsic_state *state,
346 		struct btrfsic_block *const block,
347 		struct btrfs_super_block *const super_hdr);
348 static void btrfsic_bio_end_io(struct bio *bp, int bio_error_status);
349 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate);
350 static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
351 					      const struct btrfsic_block *block,
352 					      int recursion_level);
353 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
354 					struct btrfsic_block *const block,
355 					int recursion_level);
356 static void btrfsic_print_add_link(const struct btrfsic_state *state,
357 				   const struct btrfsic_block_link *l);
358 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
359 				   const struct btrfsic_block_link *l);
360 static char btrfsic_get_block_type(const struct btrfsic_state *state,
361 				   const struct btrfsic_block *block);
362 static void btrfsic_dump_tree(const struct btrfsic_state *state);
363 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
364 				  const struct btrfsic_block *block,
365 				  int indent_level);
366 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
367 		struct btrfsic_state *state,
368 		struct btrfsic_block_data_ctx *next_block_ctx,
369 		struct btrfsic_block *next_block,
370 		struct btrfsic_block *from_block,
371 		u64 parent_generation);
372 static struct btrfsic_block *btrfsic_block_lookup_or_add(
373 		struct btrfsic_state *state,
374 		struct btrfsic_block_data_ctx *block_ctx,
375 		const char *additional_string,
376 		int is_metadata,
377 		int is_iodone,
378 		int never_written,
379 		int mirror_num,
380 		int *was_created);
381 static int btrfsic_process_superblock_dev_mirror(
382 		struct btrfsic_state *state,
383 		struct btrfsic_dev_state *dev_state,
384 		struct btrfs_device *device,
385 		int superblock_mirror_num,
386 		struct btrfsic_dev_state **selected_dev_state,
387 		struct btrfs_super_block *selected_super);
388 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
389 		struct block_device *bdev);
390 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
391 					   u64 bytenr,
392 					   struct btrfsic_dev_state *dev_state,
393 					   u64 dev_bytenr);
394 
395 static struct mutex btrfsic_mutex;
396 static int btrfsic_is_initialized;
397 static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
398 
399 
400 static void btrfsic_block_init(struct btrfsic_block *b)
401 {
402 	b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
403 	b->dev_state = NULL;
404 	b->dev_bytenr = 0;
405 	b->logical_bytenr = 0;
406 	b->generation = BTRFSIC_GENERATION_UNKNOWN;
407 	b->disk_key.objectid = 0;
408 	b->disk_key.type = 0;
409 	b->disk_key.offset = 0;
410 	b->is_metadata = 0;
411 	b->is_superblock = 0;
412 	b->is_iodone = 0;
413 	b->iodone_w_error = 0;
414 	b->never_written = 0;
415 	b->mirror_num = 0;
416 	b->next_in_same_bio = NULL;
417 	b->orig_bio_bh_private = NULL;
418 	b->orig_bio_bh_end_io.bio = NULL;
419 	INIT_LIST_HEAD(&b->collision_resolving_node);
420 	INIT_LIST_HEAD(&b->all_blocks_node);
421 	INIT_LIST_HEAD(&b->ref_to_list);
422 	INIT_LIST_HEAD(&b->ref_from_list);
423 	b->submit_bio_bh_rw = 0;
424 	b->flush_gen = 0;
425 }
426 
427 static struct btrfsic_block *btrfsic_block_alloc(void)
428 {
429 	struct btrfsic_block *b;
430 
431 	b = kzalloc(sizeof(*b), GFP_NOFS);
432 	if (NULL != b)
433 		btrfsic_block_init(b);
434 
435 	return b;
436 }
437 
438 static void btrfsic_block_free(struct btrfsic_block *b)
439 {
440 	BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
441 	kfree(b);
442 }
443 
444 static void btrfsic_block_link_init(struct btrfsic_block_link *l)
445 {
446 	l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
447 	l->ref_cnt = 1;
448 	INIT_LIST_HEAD(&l->node_ref_to);
449 	INIT_LIST_HEAD(&l->node_ref_from);
450 	INIT_LIST_HEAD(&l->collision_resolving_node);
451 	l->block_ref_to = NULL;
452 	l->block_ref_from = NULL;
453 }
454 
455 static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
456 {
457 	struct btrfsic_block_link *l;
458 
459 	l = kzalloc(sizeof(*l), GFP_NOFS);
460 	if (NULL != l)
461 		btrfsic_block_link_init(l);
462 
463 	return l;
464 }
465 
466 static void btrfsic_block_link_free(struct btrfsic_block_link *l)
467 {
468 	BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
469 	kfree(l);
470 }
471 
472 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
473 {
474 	ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
475 	ds->bdev = NULL;
476 	ds->state = NULL;
477 	ds->name[0] = '\0';
478 	INIT_LIST_HEAD(&ds->collision_resolving_node);
479 	ds->last_flush_gen = 0;
480 	btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
481 	ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
482 	ds->dummy_block_for_bio_bh_flush.dev_state = ds;
483 }
484 
485 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
486 {
487 	struct btrfsic_dev_state *ds;
488 
489 	ds = kzalloc(sizeof(*ds), GFP_NOFS);
490 	if (NULL != ds)
491 		btrfsic_dev_state_init(ds);
492 
493 	return ds;
494 }
495 
496 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
497 {
498 	BUG_ON(!(NULL == ds ||
499 		 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
500 	kfree(ds);
501 }
502 
503 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
504 {
505 	int i;
506 
507 	for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
508 		INIT_LIST_HEAD(h->table + i);
509 }
510 
511 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
512 					struct btrfsic_block_hashtable *h)
513 {
514 	const unsigned int hashval =
515 	    (((unsigned int)(b->dev_bytenr >> 16)) ^
516 	     ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
517 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
518 
519 	list_add(&b->collision_resolving_node, h->table + hashval);
520 }
521 
522 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
523 {
524 	list_del(&b->collision_resolving_node);
525 }
526 
527 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
528 		struct block_device *bdev,
529 		u64 dev_bytenr,
530 		struct btrfsic_block_hashtable *h)
531 {
532 	const unsigned int hashval =
533 	    (((unsigned int)(dev_bytenr >> 16)) ^
534 	     ((unsigned int)((uintptr_t)bdev))) &
535 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
536 	struct list_head *elem;
537 
538 	list_for_each(elem, h->table + hashval) {
539 		struct btrfsic_block *const b =
540 		    list_entry(elem, struct btrfsic_block,
541 			       collision_resolving_node);
542 
543 		if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
544 			return b;
545 	}
546 
547 	return NULL;
548 }
549 
550 static void btrfsic_block_link_hashtable_init(
551 		struct btrfsic_block_link_hashtable *h)
552 {
553 	int i;
554 
555 	for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
556 		INIT_LIST_HEAD(h->table + i);
557 }
558 
559 static void btrfsic_block_link_hashtable_add(
560 		struct btrfsic_block_link *l,
561 		struct btrfsic_block_link_hashtable *h)
562 {
563 	const unsigned int hashval =
564 	    (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
565 	     ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
566 	     ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
567 	     ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
568 	     & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
569 
570 	BUG_ON(NULL == l->block_ref_to);
571 	BUG_ON(NULL == l->block_ref_from);
572 	list_add(&l->collision_resolving_node, h->table + hashval);
573 }
574 
575 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
576 {
577 	list_del(&l->collision_resolving_node);
578 }
579 
580 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
581 		struct block_device *bdev_ref_to,
582 		u64 dev_bytenr_ref_to,
583 		struct block_device *bdev_ref_from,
584 		u64 dev_bytenr_ref_from,
585 		struct btrfsic_block_link_hashtable *h)
586 {
587 	const unsigned int hashval =
588 	    (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
589 	     ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
590 	     ((unsigned int)((uintptr_t)bdev_ref_to)) ^
591 	     ((unsigned int)((uintptr_t)bdev_ref_from))) &
592 	     (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
593 	struct list_head *elem;
594 
595 	list_for_each(elem, h->table + hashval) {
596 		struct btrfsic_block_link *const l =
597 		    list_entry(elem, struct btrfsic_block_link,
598 			       collision_resolving_node);
599 
600 		BUG_ON(NULL == l->block_ref_to);
601 		BUG_ON(NULL == l->block_ref_from);
602 		if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
603 		    l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
604 		    l->block_ref_from->dev_state->bdev == bdev_ref_from &&
605 		    l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
606 			return l;
607 	}
608 
609 	return NULL;
610 }
611 
612 static void btrfsic_dev_state_hashtable_init(
613 		struct btrfsic_dev_state_hashtable *h)
614 {
615 	int i;
616 
617 	for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
618 		INIT_LIST_HEAD(h->table + i);
619 }
620 
621 static void btrfsic_dev_state_hashtable_add(
622 		struct btrfsic_dev_state *ds,
623 		struct btrfsic_dev_state_hashtable *h)
624 {
625 	const unsigned int hashval =
626 	    (((unsigned int)((uintptr_t)ds->bdev)) &
627 	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
628 
629 	list_add(&ds->collision_resolving_node, h->table + hashval);
630 }
631 
632 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
633 {
634 	list_del(&ds->collision_resolving_node);
635 }
636 
637 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(
638 		struct block_device *bdev,
639 		struct btrfsic_dev_state_hashtable *h)
640 {
641 	const unsigned int hashval =
642 	    (((unsigned int)((uintptr_t)bdev)) &
643 	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
644 	struct list_head *elem;
645 
646 	list_for_each(elem, h->table + hashval) {
647 		struct btrfsic_dev_state *const ds =
648 		    list_entry(elem, struct btrfsic_dev_state,
649 			       collision_resolving_node);
650 
651 		if (ds->bdev == bdev)
652 			return ds;
653 	}
654 
655 	return NULL;
656 }
657 
658 static int btrfsic_process_superblock(struct btrfsic_state *state,
659 				      struct btrfs_fs_devices *fs_devices)
660 {
661 	int ret = 0;
662 	struct btrfs_super_block *selected_super;
663 	struct list_head *dev_head = &fs_devices->devices;
664 	struct btrfs_device *device;
665 	struct btrfsic_dev_state *selected_dev_state = NULL;
666 	int pass;
667 
668 	BUG_ON(NULL == state);
669 	selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
670 	if (NULL == selected_super) {
671 		printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
672 		return -1;
673 	}
674 
675 	list_for_each_entry(device, dev_head, dev_list) {
676 		int i;
677 		struct btrfsic_dev_state *dev_state;
678 
679 		if (!device->bdev || !device->name)
680 			continue;
681 
682 		dev_state = btrfsic_dev_state_lookup(device->bdev);
683 		BUG_ON(NULL == dev_state);
684 		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
685 			ret = btrfsic_process_superblock_dev_mirror(
686 					state, dev_state, device, i,
687 					&selected_dev_state, selected_super);
688 			if (0 != ret && 0 == i) {
689 				kfree(selected_super);
690 				return ret;
691 			}
692 		}
693 	}
694 
695 	if (NULL == state->latest_superblock) {
696 		printk(KERN_INFO "btrfsic: no superblock found!\n");
697 		kfree(selected_super);
698 		return -1;
699 	}
700 
701 	state->csum_size = btrfs_super_csum_size(selected_super);
702 
703 	for (pass = 0; pass < 3; pass++) {
704 		int num_copies;
705 		int mirror_num;
706 		u64 next_bytenr;
707 
708 		switch (pass) {
709 		case 0:
710 			next_bytenr = btrfs_super_root(selected_super);
711 			if (state->print_mask &
712 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
713 				printk(KERN_INFO "root@%llu\n", next_bytenr);
714 			break;
715 		case 1:
716 			next_bytenr = btrfs_super_chunk_root(selected_super);
717 			if (state->print_mask &
718 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
719 				printk(KERN_INFO "chunk@%llu\n", next_bytenr);
720 			break;
721 		case 2:
722 			next_bytenr = btrfs_super_log_root(selected_super);
723 			if (0 == next_bytenr)
724 				continue;
725 			if (state->print_mask &
726 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
727 				printk(KERN_INFO "log@%llu\n", next_bytenr);
728 			break;
729 		}
730 
731 		num_copies =
732 		    btrfs_num_copies(state->root->fs_info,
733 				     next_bytenr, state->metablock_size);
734 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
735 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
736 			       next_bytenr, num_copies);
737 
738 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
739 			struct btrfsic_block *next_block;
740 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
741 			struct btrfsic_block_link *l;
742 
743 			ret = btrfsic_map_block(state, next_bytenr,
744 						state->metablock_size,
745 						&tmp_next_block_ctx,
746 						mirror_num);
747 			if (ret) {
748 				printk(KERN_INFO "btrfsic:"
749 				       " btrfsic_map_block(root @%llu,"
750 				       " mirror %d) failed!\n",
751 				       next_bytenr, mirror_num);
752 				kfree(selected_super);
753 				return -1;
754 			}
755 
756 			next_block = btrfsic_block_hashtable_lookup(
757 					tmp_next_block_ctx.dev->bdev,
758 					tmp_next_block_ctx.dev_bytenr,
759 					&state->block_hashtable);
760 			BUG_ON(NULL == next_block);
761 
762 			l = btrfsic_block_link_hashtable_lookup(
763 					tmp_next_block_ctx.dev->bdev,
764 					tmp_next_block_ctx.dev_bytenr,
765 					state->latest_superblock->dev_state->
766 					bdev,
767 					state->latest_superblock->dev_bytenr,
768 					&state->block_link_hashtable);
769 			BUG_ON(NULL == l);
770 
771 			ret = btrfsic_read_block(state, &tmp_next_block_ctx);
772 			if (ret < (int)PAGE_CACHE_SIZE) {
773 				printk(KERN_INFO
774 				       "btrfsic: read @logical %llu failed!\n",
775 				       tmp_next_block_ctx.start);
776 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
777 				kfree(selected_super);
778 				return -1;
779 			}
780 
781 			ret = btrfsic_process_metablock(state,
782 							next_block,
783 							&tmp_next_block_ctx,
784 							BTRFS_MAX_LEVEL + 3, 1);
785 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
786 		}
787 	}
788 
789 	kfree(selected_super);
790 	return ret;
791 }
792 
793 static int btrfsic_process_superblock_dev_mirror(
794 		struct btrfsic_state *state,
795 		struct btrfsic_dev_state *dev_state,
796 		struct btrfs_device *device,
797 		int superblock_mirror_num,
798 		struct btrfsic_dev_state **selected_dev_state,
799 		struct btrfs_super_block *selected_super)
800 {
801 	struct btrfs_super_block *super_tmp;
802 	u64 dev_bytenr;
803 	struct buffer_head *bh;
804 	struct btrfsic_block *superblock_tmp;
805 	int pass;
806 	struct block_device *const superblock_bdev = device->bdev;
807 
808 	/* super block bytenr is always the unmapped device bytenr */
809 	dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
810 	if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->total_bytes)
811 		return -1;
812 	bh = __bread(superblock_bdev, dev_bytenr / 4096,
813 		     BTRFS_SUPER_INFO_SIZE);
814 	if (NULL == bh)
815 		return -1;
816 	super_tmp = (struct btrfs_super_block *)
817 	    (bh->b_data + (dev_bytenr & 4095));
818 
819 	if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
820 	    btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
821 	    memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
822 	    btrfs_super_nodesize(super_tmp) != state->metablock_size ||
823 	    btrfs_super_leafsize(super_tmp) != state->metablock_size ||
824 	    btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
825 		brelse(bh);
826 		return 0;
827 	}
828 
829 	superblock_tmp =
830 	    btrfsic_block_hashtable_lookup(superblock_bdev,
831 					   dev_bytenr,
832 					   &state->block_hashtable);
833 	if (NULL == superblock_tmp) {
834 		superblock_tmp = btrfsic_block_alloc();
835 		if (NULL == superblock_tmp) {
836 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
837 			brelse(bh);
838 			return -1;
839 		}
840 		/* for superblock, only the dev_bytenr makes sense */
841 		superblock_tmp->dev_bytenr = dev_bytenr;
842 		superblock_tmp->dev_state = dev_state;
843 		superblock_tmp->logical_bytenr = dev_bytenr;
844 		superblock_tmp->generation = btrfs_super_generation(super_tmp);
845 		superblock_tmp->is_metadata = 1;
846 		superblock_tmp->is_superblock = 1;
847 		superblock_tmp->is_iodone = 1;
848 		superblock_tmp->never_written = 0;
849 		superblock_tmp->mirror_num = 1 + superblock_mirror_num;
850 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
851 			printk_in_rcu(KERN_INFO "New initial S-block (bdev %p, %s)"
852 				     " @%llu (%s/%llu/%d)\n",
853 				     superblock_bdev,
854 				     rcu_str_deref(device->name), dev_bytenr,
855 				     dev_state->name, dev_bytenr,
856 				     superblock_mirror_num);
857 		list_add(&superblock_tmp->all_blocks_node,
858 			 &state->all_blocks_list);
859 		btrfsic_block_hashtable_add(superblock_tmp,
860 					    &state->block_hashtable);
861 	}
862 
863 	/* select the one with the highest generation field */
864 	if (btrfs_super_generation(super_tmp) >
865 	    state->max_superblock_generation ||
866 	    0 == state->max_superblock_generation) {
867 		memcpy(selected_super, super_tmp, sizeof(*selected_super));
868 		*selected_dev_state = dev_state;
869 		state->max_superblock_generation =
870 		    btrfs_super_generation(super_tmp);
871 		state->latest_superblock = superblock_tmp;
872 	}
873 
874 	for (pass = 0; pass < 3; pass++) {
875 		u64 next_bytenr;
876 		int num_copies;
877 		int mirror_num;
878 		const char *additional_string = NULL;
879 		struct btrfs_disk_key tmp_disk_key;
880 
881 		tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
882 		tmp_disk_key.offset = 0;
883 		switch (pass) {
884 		case 0:
885 			btrfs_set_disk_key_objectid(&tmp_disk_key,
886 						    BTRFS_ROOT_TREE_OBJECTID);
887 			additional_string = "initial root ";
888 			next_bytenr = btrfs_super_root(super_tmp);
889 			break;
890 		case 1:
891 			btrfs_set_disk_key_objectid(&tmp_disk_key,
892 						    BTRFS_CHUNK_TREE_OBJECTID);
893 			additional_string = "initial chunk ";
894 			next_bytenr = btrfs_super_chunk_root(super_tmp);
895 			break;
896 		case 2:
897 			btrfs_set_disk_key_objectid(&tmp_disk_key,
898 						    BTRFS_TREE_LOG_OBJECTID);
899 			additional_string = "initial log ";
900 			next_bytenr = btrfs_super_log_root(super_tmp);
901 			if (0 == next_bytenr)
902 				continue;
903 			break;
904 		}
905 
906 		num_copies =
907 		    btrfs_num_copies(state->root->fs_info,
908 				     next_bytenr, state->metablock_size);
909 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
910 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
911 			       next_bytenr, num_copies);
912 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
913 			struct btrfsic_block *next_block;
914 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
915 			struct btrfsic_block_link *l;
916 
917 			if (btrfsic_map_block(state, next_bytenr,
918 					      state->metablock_size,
919 					      &tmp_next_block_ctx,
920 					      mirror_num)) {
921 				printk(KERN_INFO "btrfsic: btrfsic_map_block("
922 				       "bytenr @%llu, mirror %d) failed!\n",
923 				       next_bytenr, mirror_num);
924 				brelse(bh);
925 				return -1;
926 			}
927 
928 			next_block = btrfsic_block_lookup_or_add(
929 					state, &tmp_next_block_ctx,
930 					additional_string, 1, 1, 0,
931 					mirror_num, NULL);
932 			if (NULL == next_block) {
933 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
934 				brelse(bh);
935 				return -1;
936 			}
937 
938 			next_block->disk_key = tmp_disk_key;
939 			next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
940 			l = btrfsic_block_link_lookup_or_add(
941 					state, &tmp_next_block_ctx,
942 					next_block, superblock_tmp,
943 					BTRFSIC_GENERATION_UNKNOWN);
944 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
945 			if (NULL == l) {
946 				brelse(bh);
947 				return -1;
948 			}
949 		}
950 	}
951 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
952 		btrfsic_dump_tree_sub(state, superblock_tmp, 0);
953 
954 	brelse(bh);
955 	return 0;
956 }
957 
958 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
959 {
960 	struct btrfsic_stack_frame *sf;
961 
962 	sf = kzalloc(sizeof(*sf), GFP_NOFS);
963 	if (NULL == sf)
964 		printk(KERN_INFO "btrfsic: alloc memory failed!\n");
965 	else
966 		sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
967 	return sf;
968 }
969 
970 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
971 {
972 	BUG_ON(!(NULL == sf ||
973 		 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
974 	kfree(sf);
975 }
976 
977 static int btrfsic_process_metablock(
978 		struct btrfsic_state *state,
979 		struct btrfsic_block *const first_block,
980 		struct btrfsic_block_data_ctx *const first_block_ctx,
981 		int first_limit_nesting, int force_iodone_flag)
982 {
983 	struct btrfsic_stack_frame initial_stack_frame = { 0 };
984 	struct btrfsic_stack_frame *sf;
985 	struct btrfsic_stack_frame *next_stack;
986 	struct btrfs_header *const first_hdr =
987 		(struct btrfs_header *)first_block_ctx->datav[0];
988 
989 	BUG_ON(!first_hdr);
990 	sf = &initial_stack_frame;
991 	sf->error = 0;
992 	sf->i = -1;
993 	sf->limit_nesting = first_limit_nesting;
994 	sf->block = first_block;
995 	sf->block_ctx = first_block_ctx;
996 	sf->next_block = NULL;
997 	sf->hdr = first_hdr;
998 	sf->prev = NULL;
999 
1000 continue_with_new_stack_frame:
1001 	sf->block->generation = le64_to_cpu(sf->hdr->generation);
1002 	if (0 == sf->hdr->level) {
1003 		struct btrfs_leaf *const leafhdr =
1004 		    (struct btrfs_leaf *)sf->hdr;
1005 
1006 		if (-1 == sf->i) {
1007 			sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
1008 
1009 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1010 				printk(KERN_INFO
1011 				       "leaf %llu items %d generation %llu"
1012 				       " owner %llu\n",
1013 				       sf->block_ctx->start, sf->nr,
1014 				       btrfs_stack_header_generation(
1015 					       &leafhdr->header),
1016 				       btrfs_stack_header_owner(
1017 					       &leafhdr->header));
1018 		}
1019 
1020 continue_with_current_leaf_stack_frame:
1021 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1022 			sf->i++;
1023 			sf->num_copies = 0;
1024 		}
1025 
1026 		if (sf->i < sf->nr) {
1027 			struct btrfs_item disk_item;
1028 			u32 disk_item_offset =
1029 				(uintptr_t)(leafhdr->items + sf->i) -
1030 				(uintptr_t)leafhdr;
1031 			struct btrfs_disk_key *disk_key;
1032 			u8 type;
1033 			u32 item_offset;
1034 			u32 item_size;
1035 
1036 			if (disk_item_offset + sizeof(struct btrfs_item) >
1037 			    sf->block_ctx->len) {
1038 leaf_item_out_of_bounce_error:
1039 				printk(KERN_INFO
1040 				       "btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
1041 				       sf->block_ctx->start,
1042 				       sf->block_ctx->dev->name);
1043 				goto one_stack_frame_backwards;
1044 			}
1045 			btrfsic_read_from_block_data(sf->block_ctx,
1046 						     &disk_item,
1047 						     disk_item_offset,
1048 						     sizeof(struct btrfs_item));
1049 			item_offset = btrfs_stack_item_offset(&disk_item);
1050 			item_size = btrfs_stack_item_size(&disk_item);
1051 			disk_key = &disk_item.key;
1052 			type = btrfs_disk_key_type(disk_key);
1053 
1054 			if (BTRFS_ROOT_ITEM_KEY == type) {
1055 				struct btrfs_root_item root_item;
1056 				u32 root_item_offset;
1057 				u64 next_bytenr;
1058 
1059 				root_item_offset = item_offset +
1060 					offsetof(struct btrfs_leaf, items);
1061 				if (root_item_offset + item_size >
1062 				    sf->block_ctx->len)
1063 					goto leaf_item_out_of_bounce_error;
1064 				btrfsic_read_from_block_data(
1065 					sf->block_ctx, &root_item,
1066 					root_item_offset,
1067 					item_size);
1068 				next_bytenr = btrfs_root_bytenr(&root_item);
1069 
1070 				sf->error =
1071 				    btrfsic_create_link_to_next_block(
1072 						state,
1073 						sf->block,
1074 						sf->block_ctx,
1075 						next_bytenr,
1076 						sf->limit_nesting,
1077 						&sf->next_block_ctx,
1078 						&sf->next_block,
1079 						force_iodone_flag,
1080 						&sf->num_copies,
1081 						&sf->mirror_num,
1082 						disk_key,
1083 						btrfs_root_generation(
1084 						&root_item));
1085 				if (sf->error)
1086 					goto one_stack_frame_backwards;
1087 
1088 				if (NULL != sf->next_block) {
1089 					struct btrfs_header *const next_hdr =
1090 					    (struct btrfs_header *)
1091 					    sf->next_block_ctx.datav[0];
1092 
1093 					next_stack =
1094 					    btrfsic_stack_frame_alloc();
1095 					if (NULL == next_stack) {
1096 						btrfsic_release_block_ctx(
1097 								&sf->
1098 								next_block_ctx);
1099 						goto one_stack_frame_backwards;
1100 					}
1101 
1102 					next_stack->i = -1;
1103 					next_stack->block = sf->next_block;
1104 					next_stack->block_ctx =
1105 					    &sf->next_block_ctx;
1106 					next_stack->next_block = NULL;
1107 					next_stack->hdr = next_hdr;
1108 					next_stack->limit_nesting =
1109 					    sf->limit_nesting - 1;
1110 					next_stack->prev = sf;
1111 					sf = next_stack;
1112 					goto continue_with_new_stack_frame;
1113 				}
1114 			} else if (BTRFS_EXTENT_DATA_KEY == type &&
1115 				   state->include_extent_data) {
1116 				sf->error = btrfsic_handle_extent_data(
1117 						state,
1118 						sf->block,
1119 						sf->block_ctx,
1120 						item_offset,
1121 						force_iodone_flag);
1122 				if (sf->error)
1123 					goto one_stack_frame_backwards;
1124 			}
1125 
1126 			goto continue_with_current_leaf_stack_frame;
1127 		}
1128 	} else {
1129 		struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1130 
1131 		if (-1 == sf->i) {
1132 			sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1133 
1134 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1135 				printk(KERN_INFO "node %llu level %d items %d"
1136 				       " generation %llu owner %llu\n",
1137 				       sf->block_ctx->start,
1138 				       nodehdr->header.level, sf->nr,
1139 				       btrfs_stack_header_generation(
1140 				       &nodehdr->header),
1141 				       btrfs_stack_header_owner(
1142 				       &nodehdr->header));
1143 		}
1144 
1145 continue_with_current_node_stack_frame:
1146 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1147 			sf->i++;
1148 			sf->num_copies = 0;
1149 		}
1150 
1151 		if (sf->i < sf->nr) {
1152 			struct btrfs_key_ptr key_ptr;
1153 			u32 key_ptr_offset;
1154 			u64 next_bytenr;
1155 
1156 			key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1157 					  (uintptr_t)nodehdr;
1158 			if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1159 			    sf->block_ctx->len) {
1160 				printk(KERN_INFO
1161 				       "btrfsic: node item out of bounce at logical %llu, dev %s\n",
1162 				       sf->block_ctx->start,
1163 				       sf->block_ctx->dev->name);
1164 				goto one_stack_frame_backwards;
1165 			}
1166 			btrfsic_read_from_block_data(
1167 				sf->block_ctx, &key_ptr, key_ptr_offset,
1168 				sizeof(struct btrfs_key_ptr));
1169 			next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1170 
1171 			sf->error = btrfsic_create_link_to_next_block(
1172 					state,
1173 					sf->block,
1174 					sf->block_ctx,
1175 					next_bytenr,
1176 					sf->limit_nesting,
1177 					&sf->next_block_ctx,
1178 					&sf->next_block,
1179 					force_iodone_flag,
1180 					&sf->num_copies,
1181 					&sf->mirror_num,
1182 					&key_ptr.key,
1183 					btrfs_stack_key_generation(&key_ptr));
1184 			if (sf->error)
1185 				goto one_stack_frame_backwards;
1186 
1187 			if (NULL != sf->next_block) {
1188 				struct btrfs_header *const next_hdr =
1189 				    (struct btrfs_header *)
1190 				    sf->next_block_ctx.datav[0];
1191 
1192 				next_stack = btrfsic_stack_frame_alloc();
1193 				if (NULL == next_stack)
1194 					goto one_stack_frame_backwards;
1195 
1196 				next_stack->i = -1;
1197 				next_stack->block = sf->next_block;
1198 				next_stack->block_ctx = &sf->next_block_ctx;
1199 				next_stack->next_block = NULL;
1200 				next_stack->hdr = next_hdr;
1201 				next_stack->limit_nesting =
1202 				    sf->limit_nesting - 1;
1203 				next_stack->prev = sf;
1204 				sf = next_stack;
1205 				goto continue_with_new_stack_frame;
1206 			}
1207 
1208 			goto continue_with_current_node_stack_frame;
1209 		}
1210 	}
1211 
1212 one_stack_frame_backwards:
1213 	if (NULL != sf->prev) {
1214 		struct btrfsic_stack_frame *const prev = sf->prev;
1215 
1216 		/* the one for the initial block is freed in the caller */
1217 		btrfsic_release_block_ctx(sf->block_ctx);
1218 
1219 		if (sf->error) {
1220 			prev->error = sf->error;
1221 			btrfsic_stack_frame_free(sf);
1222 			sf = prev;
1223 			goto one_stack_frame_backwards;
1224 		}
1225 
1226 		btrfsic_stack_frame_free(sf);
1227 		sf = prev;
1228 		goto continue_with_new_stack_frame;
1229 	} else {
1230 		BUG_ON(&initial_stack_frame != sf);
1231 	}
1232 
1233 	return sf->error;
1234 }
1235 
1236 static void btrfsic_read_from_block_data(
1237 	struct btrfsic_block_data_ctx *block_ctx,
1238 	void *dstv, u32 offset, size_t len)
1239 {
1240 	size_t cur;
1241 	size_t offset_in_page;
1242 	char *kaddr;
1243 	char *dst = (char *)dstv;
1244 	size_t start_offset = block_ctx->start & ((u64)PAGE_CACHE_SIZE - 1);
1245 	unsigned long i = (start_offset + offset) >> PAGE_CACHE_SHIFT;
1246 
1247 	WARN_ON(offset + len > block_ctx->len);
1248 	offset_in_page = (start_offset + offset) & (PAGE_CACHE_SIZE - 1);
1249 
1250 	while (len > 0) {
1251 		cur = min(len, ((size_t)PAGE_CACHE_SIZE - offset_in_page));
1252 		BUG_ON(i >= (block_ctx->len + PAGE_CACHE_SIZE - 1) >>
1253 			    PAGE_CACHE_SHIFT);
1254 		kaddr = block_ctx->datav[i];
1255 		memcpy(dst, kaddr + offset_in_page, cur);
1256 
1257 		dst += cur;
1258 		len -= cur;
1259 		offset_in_page = 0;
1260 		i++;
1261 	}
1262 }
1263 
1264 static int btrfsic_create_link_to_next_block(
1265 		struct btrfsic_state *state,
1266 		struct btrfsic_block *block,
1267 		struct btrfsic_block_data_ctx *block_ctx,
1268 		u64 next_bytenr,
1269 		int limit_nesting,
1270 		struct btrfsic_block_data_ctx *next_block_ctx,
1271 		struct btrfsic_block **next_blockp,
1272 		int force_iodone_flag,
1273 		int *num_copiesp, int *mirror_nump,
1274 		struct btrfs_disk_key *disk_key,
1275 		u64 parent_generation)
1276 {
1277 	struct btrfsic_block *next_block = NULL;
1278 	int ret;
1279 	struct btrfsic_block_link *l;
1280 	int did_alloc_block_link;
1281 	int block_was_created;
1282 
1283 	*next_blockp = NULL;
1284 	if (0 == *num_copiesp) {
1285 		*num_copiesp =
1286 		    btrfs_num_copies(state->root->fs_info,
1287 				     next_bytenr, state->metablock_size);
1288 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1289 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
1290 			       next_bytenr, *num_copiesp);
1291 		*mirror_nump = 1;
1292 	}
1293 
1294 	if (*mirror_nump > *num_copiesp)
1295 		return 0;
1296 
1297 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1298 		printk(KERN_INFO
1299 		       "btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1300 		       *mirror_nump);
1301 	ret = btrfsic_map_block(state, next_bytenr,
1302 				state->metablock_size,
1303 				next_block_ctx, *mirror_nump);
1304 	if (ret) {
1305 		printk(KERN_INFO
1306 		       "btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1307 		       next_bytenr, *mirror_nump);
1308 		btrfsic_release_block_ctx(next_block_ctx);
1309 		*next_blockp = NULL;
1310 		return -1;
1311 	}
1312 
1313 	next_block = btrfsic_block_lookup_or_add(state,
1314 						 next_block_ctx, "referenced ",
1315 						 1, force_iodone_flag,
1316 						 !force_iodone_flag,
1317 						 *mirror_nump,
1318 						 &block_was_created);
1319 	if (NULL == next_block) {
1320 		btrfsic_release_block_ctx(next_block_ctx);
1321 		*next_blockp = NULL;
1322 		return -1;
1323 	}
1324 	if (block_was_created) {
1325 		l = NULL;
1326 		next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1327 	} else {
1328 		if (next_block->logical_bytenr != next_bytenr &&
1329 		    !(!next_block->is_metadata &&
1330 		      0 == next_block->logical_bytenr)) {
1331 			printk(KERN_INFO
1332 			       "Referenced block @%llu (%s/%llu/%d)"
1333 			       " found in hash table, %c,"
1334 			       " bytenr mismatch (!= stored %llu).\n",
1335 			       next_bytenr, next_block_ctx->dev->name,
1336 			       next_block_ctx->dev_bytenr, *mirror_nump,
1337 			       btrfsic_get_block_type(state, next_block),
1338 			       next_block->logical_bytenr);
1339 		} else if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1340 			printk(KERN_INFO
1341 			       "Referenced block @%llu (%s/%llu/%d)"
1342 			       " found in hash table, %c.\n",
1343 			       next_bytenr, next_block_ctx->dev->name,
1344 			       next_block_ctx->dev_bytenr, *mirror_nump,
1345 			       btrfsic_get_block_type(state, next_block));
1346 		next_block->logical_bytenr = next_bytenr;
1347 
1348 		next_block->mirror_num = *mirror_nump;
1349 		l = btrfsic_block_link_hashtable_lookup(
1350 				next_block_ctx->dev->bdev,
1351 				next_block_ctx->dev_bytenr,
1352 				block_ctx->dev->bdev,
1353 				block_ctx->dev_bytenr,
1354 				&state->block_link_hashtable);
1355 	}
1356 
1357 	next_block->disk_key = *disk_key;
1358 	if (NULL == l) {
1359 		l = btrfsic_block_link_alloc();
1360 		if (NULL == l) {
1361 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
1362 			btrfsic_release_block_ctx(next_block_ctx);
1363 			*next_blockp = NULL;
1364 			return -1;
1365 		}
1366 
1367 		did_alloc_block_link = 1;
1368 		l->block_ref_to = next_block;
1369 		l->block_ref_from = block;
1370 		l->ref_cnt = 1;
1371 		l->parent_generation = parent_generation;
1372 
1373 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1374 			btrfsic_print_add_link(state, l);
1375 
1376 		list_add(&l->node_ref_to, &block->ref_to_list);
1377 		list_add(&l->node_ref_from, &next_block->ref_from_list);
1378 
1379 		btrfsic_block_link_hashtable_add(l,
1380 						 &state->block_link_hashtable);
1381 	} else {
1382 		did_alloc_block_link = 0;
1383 		if (0 == limit_nesting) {
1384 			l->ref_cnt++;
1385 			l->parent_generation = parent_generation;
1386 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1387 				btrfsic_print_add_link(state, l);
1388 		}
1389 	}
1390 
1391 	if (limit_nesting > 0 && did_alloc_block_link) {
1392 		ret = btrfsic_read_block(state, next_block_ctx);
1393 		if (ret < (int)next_block_ctx->len) {
1394 			printk(KERN_INFO
1395 			       "btrfsic: read block @logical %llu failed!\n",
1396 			       next_bytenr);
1397 			btrfsic_release_block_ctx(next_block_ctx);
1398 			*next_blockp = NULL;
1399 			return -1;
1400 		}
1401 
1402 		*next_blockp = next_block;
1403 	} else {
1404 		*next_blockp = NULL;
1405 	}
1406 	(*mirror_nump)++;
1407 
1408 	return 0;
1409 }
1410 
1411 static int btrfsic_handle_extent_data(
1412 		struct btrfsic_state *state,
1413 		struct btrfsic_block *block,
1414 		struct btrfsic_block_data_ctx *block_ctx,
1415 		u32 item_offset, int force_iodone_flag)
1416 {
1417 	int ret;
1418 	struct btrfs_file_extent_item file_extent_item;
1419 	u64 file_extent_item_offset;
1420 	u64 next_bytenr;
1421 	u64 num_bytes;
1422 	u64 generation;
1423 	struct btrfsic_block_link *l;
1424 
1425 	file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1426 				  item_offset;
1427 	if (file_extent_item_offset +
1428 	    offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1429 	    block_ctx->len) {
1430 		printk(KERN_INFO
1431 		       "btrfsic: file item out of bounce at logical %llu, dev %s\n",
1432 		       block_ctx->start, block_ctx->dev->name);
1433 		return -1;
1434 	}
1435 
1436 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1437 		file_extent_item_offset,
1438 		offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1439 	if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1440 	    btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1441 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1442 			printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu\n",
1443 			       file_extent_item.type,
1444 			       btrfs_stack_file_extent_disk_bytenr(
1445 			       &file_extent_item));
1446 		return 0;
1447 	}
1448 
1449 	if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1450 	    block_ctx->len) {
1451 		printk(KERN_INFO
1452 		       "btrfsic: file item out of bounce at logical %llu, dev %s\n",
1453 		       block_ctx->start, block_ctx->dev->name);
1454 		return -1;
1455 	}
1456 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1457 				     file_extent_item_offset,
1458 				     sizeof(struct btrfs_file_extent_item));
1459 	next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item);
1460 	if (btrfs_stack_file_extent_compression(&file_extent_item) ==
1461 	    BTRFS_COMPRESS_NONE) {
1462 		next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item);
1463 		num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1464 	} else {
1465 		num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item);
1466 	}
1467 	generation = btrfs_stack_file_extent_generation(&file_extent_item);
1468 
1469 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1470 		printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu,"
1471 		       " offset = %llu, num_bytes = %llu\n",
1472 		       file_extent_item.type,
1473 		       btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1474 		       btrfs_stack_file_extent_offset(&file_extent_item),
1475 		       num_bytes);
1476 	while (num_bytes > 0) {
1477 		u32 chunk_len;
1478 		int num_copies;
1479 		int mirror_num;
1480 
1481 		if (num_bytes > state->datablock_size)
1482 			chunk_len = state->datablock_size;
1483 		else
1484 			chunk_len = num_bytes;
1485 
1486 		num_copies =
1487 		    btrfs_num_copies(state->root->fs_info,
1488 				     next_bytenr, state->datablock_size);
1489 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1490 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
1491 			       next_bytenr, num_copies);
1492 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1493 			struct btrfsic_block_data_ctx next_block_ctx;
1494 			struct btrfsic_block *next_block;
1495 			int block_was_created;
1496 
1497 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1498 				printk(KERN_INFO "btrfsic_handle_extent_data("
1499 				       "mirror_num=%d)\n", mirror_num);
1500 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1501 				printk(KERN_INFO
1502 				       "\tdisk_bytenr = %llu, num_bytes %u\n",
1503 				       next_bytenr, chunk_len);
1504 			ret = btrfsic_map_block(state, next_bytenr,
1505 						chunk_len, &next_block_ctx,
1506 						mirror_num);
1507 			if (ret) {
1508 				printk(KERN_INFO
1509 				       "btrfsic: btrfsic_map_block(@%llu,"
1510 				       " mirror=%d) failed!\n",
1511 				       next_bytenr, mirror_num);
1512 				return -1;
1513 			}
1514 
1515 			next_block = btrfsic_block_lookup_or_add(
1516 					state,
1517 					&next_block_ctx,
1518 					"referenced ",
1519 					0,
1520 					force_iodone_flag,
1521 					!force_iodone_flag,
1522 					mirror_num,
1523 					&block_was_created);
1524 			if (NULL == next_block) {
1525 				printk(KERN_INFO
1526 				       "btrfsic: error, kmalloc failed!\n");
1527 				btrfsic_release_block_ctx(&next_block_ctx);
1528 				return -1;
1529 			}
1530 			if (!block_was_created) {
1531 				if (next_block->logical_bytenr != next_bytenr &&
1532 				    !(!next_block->is_metadata &&
1533 				      0 == next_block->logical_bytenr)) {
1534 					printk(KERN_INFO
1535 					       "Referenced block"
1536 					       " @%llu (%s/%llu/%d)"
1537 					       " found in hash table, D,"
1538 					       " bytenr mismatch"
1539 					       " (!= stored %llu).\n",
1540 					       next_bytenr,
1541 					       next_block_ctx.dev->name,
1542 					       next_block_ctx.dev_bytenr,
1543 					       mirror_num,
1544 					       next_block->logical_bytenr);
1545 				}
1546 				next_block->logical_bytenr = next_bytenr;
1547 				next_block->mirror_num = mirror_num;
1548 			}
1549 
1550 			l = btrfsic_block_link_lookup_or_add(state,
1551 							     &next_block_ctx,
1552 							     next_block, block,
1553 							     generation);
1554 			btrfsic_release_block_ctx(&next_block_ctx);
1555 			if (NULL == l)
1556 				return -1;
1557 		}
1558 
1559 		next_bytenr += chunk_len;
1560 		num_bytes -= chunk_len;
1561 	}
1562 
1563 	return 0;
1564 }
1565 
1566 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1567 			     struct btrfsic_block_data_ctx *block_ctx_out,
1568 			     int mirror_num)
1569 {
1570 	int ret;
1571 	u64 length;
1572 	struct btrfs_bio *multi = NULL;
1573 	struct btrfs_device *device;
1574 
1575 	length = len;
1576 	ret = btrfs_map_block(state->root->fs_info, READ,
1577 			      bytenr, &length, &multi, mirror_num);
1578 
1579 	if (ret) {
1580 		block_ctx_out->start = 0;
1581 		block_ctx_out->dev_bytenr = 0;
1582 		block_ctx_out->len = 0;
1583 		block_ctx_out->dev = NULL;
1584 		block_ctx_out->datav = NULL;
1585 		block_ctx_out->pagev = NULL;
1586 		block_ctx_out->mem_to_free = NULL;
1587 
1588 		return ret;
1589 	}
1590 
1591 	device = multi->stripes[0].dev;
1592 	block_ctx_out->dev = btrfsic_dev_state_lookup(device->bdev);
1593 	block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1594 	block_ctx_out->start = bytenr;
1595 	block_ctx_out->len = len;
1596 	block_ctx_out->datav = NULL;
1597 	block_ctx_out->pagev = NULL;
1598 	block_ctx_out->mem_to_free = NULL;
1599 
1600 	kfree(multi);
1601 	if (NULL == block_ctx_out->dev) {
1602 		ret = -ENXIO;
1603 		printk(KERN_INFO "btrfsic: error, cannot lookup dev (#1)!\n");
1604 	}
1605 
1606 	return ret;
1607 }
1608 
1609 static int btrfsic_map_superblock(struct btrfsic_state *state, u64 bytenr,
1610 				  u32 len, struct block_device *bdev,
1611 				  struct btrfsic_block_data_ctx *block_ctx_out)
1612 {
1613 	block_ctx_out->dev = btrfsic_dev_state_lookup(bdev);
1614 	block_ctx_out->dev_bytenr = bytenr;
1615 	block_ctx_out->start = bytenr;
1616 	block_ctx_out->len = len;
1617 	block_ctx_out->datav = NULL;
1618 	block_ctx_out->pagev = NULL;
1619 	block_ctx_out->mem_to_free = NULL;
1620 	if (NULL != block_ctx_out->dev) {
1621 		return 0;
1622 	} else {
1623 		printk(KERN_INFO "btrfsic: error, cannot lookup dev (#2)!\n");
1624 		return -ENXIO;
1625 	}
1626 }
1627 
1628 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1629 {
1630 	if (block_ctx->mem_to_free) {
1631 		unsigned int num_pages;
1632 
1633 		BUG_ON(!block_ctx->datav);
1634 		BUG_ON(!block_ctx->pagev);
1635 		num_pages = (block_ctx->len + (u64)PAGE_CACHE_SIZE - 1) >>
1636 			    PAGE_CACHE_SHIFT;
1637 		while (num_pages > 0) {
1638 			num_pages--;
1639 			if (block_ctx->datav[num_pages]) {
1640 				kunmap(block_ctx->pagev[num_pages]);
1641 				block_ctx->datav[num_pages] = NULL;
1642 			}
1643 			if (block_ctx->pagev[num_pages]) {
1644 				__free_page(block_ctx->pagev[num_pages]);
1645 				block_ctx->pagev[num_pages] = NULL;
1646 			}
1647 		}
1648 
1649 		kfree(block_ctx->mem_to_free);
1650 		block_ctx->mem_to_free = NULL;
1651 		block_ctx->pagev = NULL;
1652 		block_ctx->datav = NULL;
1653 	}
1654 }
1655 
1656 static int btrfsic_read_block(struct btrfsic_state *state,
1657 			      struct btrfsic_block_data_ctx *block_ctx)
1658 {
1659 	unsigned int num_pages;
1660 	unsigned int i;
1661 	u64 dev_bytenr;
1662 	int ret;
1663 
1664 	BUG_ON(block_ctx->datav);
1665 	BUG_ON(block_ctx->pagev);
1666 	BUG_ON(block_ctx->mem_to_free);
1667 	if (block_ctx->dev_bytenr & ((u64)PAGE_CACHE_SIZE - 1)) {
1668 		printk(KERN_INFO
1669 		       "btrfsic: read_block() with unaligned bytenr %llu\n",
1670 		       block_ctx->dev_bytenr);
1671 		return -1;
1672 	}
1673 
1674 	num_pages = (block_ctx->len + (u64)PAGE_CACHE_SIZE - 1) >>
1675 		    PAGE_CACHE_SHIFT;
1676 	block_ctx->mem_to_free = kzalloc((sizeof(*block_ctx->datav) +
1677 					  sizeof(*block_ctx->pagev)) *
1678 					 num_pages, GFP_NOFS);
1679 	if (!block_ctx->mem_to_free)
1680 		return -1;
1681 	block_ctx->datav = block_ctx->mem_to_free;
1682 	block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1683 	for (i = 0; i < num_pages; i++) {
1684 		block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1685 		if (!block_ctx->pagev[i])
1686 			return -1;
1687 	}
1688 
1689 	dev_bytenr = block_ctx->dev_bytenr;
1690 	for (i = 0; i < num_pages;) {
1691 		struct bio *bio;
1692 		unsigned int j;
1693 
1694 		bio = btrfs_io_bio_alloc(GFP_NOFS, num_pages - i);
1695 		if (!bio) {
1696 			printk(KERN_INFO
1697 			       "btrfsic: bio_alloc() for %u pages failed!\n",
1698 			       num_pages - i);
1699 			return -1;
1700 		}
1701 		bio->bi_bdev = block_ctx->dev->bdev;
1702 		bio->bi_iter.bi_sector = dev_bytenr >> 9;
1703 
1704 		for (j = i; j < num_pages; j++) {
1705 			ret = bio_add_page(bio, block_ctx->pagev[j],
1706 					   PAGE_CACHE_SIZE, 0);
1707 			if (PAGE_CACHE_SIZE != ret)
1708 				break;
1709 		}
1710 		if (j == i) {
1711 			printk(KERN_INFO
1712 			       "btrfsic: error, failed to add a single page!\n");
1713 			return -1;
1714 		}
1715 		if (submit_bio_wait(READ, bio)) {
1716 			printk(KERN_INFO
1717 			       "btrfsic: read error at logical %llu dev %s!\n",
1718 			       block_ctx->start, block_ctx->dev->name);
1719 			bio_put(bio);
1720 			return -1;
1721 		}
1722 		bio_put(bio);
1723 		dev_bytenr += (j - i) * PAGE_CACHE_SIZE;
1724 		i = j;
1725 	}
1726 	for (i = 0; i < num_pages; i++) {
1727 		block_ctx->datav[i] = kmap(block_ctx->pagev[i]);
1728 		if (!block_ctx->datav[i]) {
1729 			printk(KERN_INFO "btrfsic: kmap() failed (dev %s)!\n",
1730 			       block_ctx->dev->name);
1731 			return -1;
1732 		}
1733 	}
1734 
1735 	return block_ctx->len;
1736 }
1737 
1738 static void btrfsic_dump_database(struct btrfsic_state *state)
1739 {
1740 	struct list_head *elem_all;
1741 
1742 	BUG_ON(NULL == state);
1743 
1744 	printk(KERN_INFO "all_blocks_list:\n");
1745 	list_for_each(elem_all, &state->all_blocks_list) {
1746 		const struct btrfsic_block *const b_all =
1747 		    list_entry(elem_all, struct btrfsic_block,
1748 			       all_blocks_node);
1749 		struct list_head *elem_ref_to;
1750 		struct list_head *elem_ref_from;
1751 
1752 		printk(KERN_INFO "%c-block @%llu (%s/%llu/%d)\n",
1753 		       btrfsic_get_block_type(state, b_all),
1754 		       b_all->logical_bytenr, b_all->dev_state->name,
1755 		       b_all->dev_bytenr, b_all->mirror_num);
1756 
1757 		list_for_each(elem_ref_to, &b_all->ref_to_list) {
1758 			const struct btrfsic_block_link *const l =
1759 			    list_entry(elem_ref_to,
1760 				       struct btrfsic_block_link,
1761 				       node_ref_to);
1762 
1763 			printk(KERN_INFO " %c @%llu (%s/%llu/%d)"
1764 			       " refers %u* to"
1765 			       " %c @%llu (%s/%llu/%d)\n",
1766 			       btrfsic_get_block_type(state, b_all),
1767 			       b_all->logical_bytenr, b_all->dev_state->name,
1768 			       b_all->dev_bytenr, b_all->mirror_num,
1769 			       l->ref_cnt,
1770 			       btrfsic_get_block_type(state, l->block_ref_to),
1771 			       l->block_ref_to->logical_bytenr,
1772 			       l->block_ref_to->dev_state->name,
1773 			       l->block_ref_to->dev_bytenr,
1774 			       l->block_ref_to->mirror_num);
1775 		}
1776 
1777 		list_for_each(elem_ref_from, &b_all->ref_from_list) {
1778 			const struct btrfsic_block_link *const l =
1779 			    list_entry(elem_ref_from,
1780 				       struct btrfsic_block_link,
1781 				       node_ref_from);
1782 
1783 			printk(KERN_INFO " %c @%llu (%s/%llu/%d)"
1784 			       " is ref %u* from"
1785 			       " %c @%llu (%s/%llu/%d)\n",
1786 			       btrfsic_get_block_type(state, b_all),
1787 			       b_all->logical_bytenr, b_all->dev_state->name,
1788 			       b_all->dev_bytenr, b_all->mirror_num,
1789 			       l->ref_cnt,
1790 			       btrfsic_get_block_type(state, l->block_ref_from),
1791 			       l->block_ref_from->logical_bytenr,
1792 			       l->block_ref_from->dev_state->name,
1793 			       l->block_ref_from->dev_bytenr,
1794 			       l->block_ref_from->mirror_num);
1795 		}
1796 
1797 		printk(KERN_INFO "\n");
1798 	}
1799 }
1800 
1801 /*
1802  * Test whether the disk block contains a tree block (leaf or node)
1803  * (note that this test fails for the super block)
1804  */
1805 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
1806 				     char **datav, unsigned int num_pages)
1807 {
1808 	struct btrfs_header *h;
1809 	u8 csum[BTRFS_CSUM_SIZE];
1810 	u32 crc = ~(u32)0;
1811 	unsigned int i;
1812 
1813 	if (num_pages * PAGE_CACHE_SIZE < state->metablock_size)
1814 		return 1; /* not metadata */
1815 	num_pages = state->metablock_size >> PAGE_CACHE_SHIFT;
1816 	h = (struct btrfs_header *)datav[0];
1817 
1818 	if (memcmp(h->fsid, state->root->fs_info->fsid, BTRFS_UUID_SIZE))
1819 		return 1;
1820 
1821 	for (i = 0; i < num_pages; i++) {
1822 		u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1823 		size_t sublen = i ? PAGE_CACHE_SIZE :
1824 				    (PAGE_CACHE_SIZE - BTRFS_CSUM_SIZE);
1825 
1826 		crc = btrfs_crc32c(crc, data, sublen);
1827 	}
1828 	btrfs_csum_final(crc, csum);
1829 	if (memcmp(csum, h->csum, state->csum_size))
1830 		return 1;
1831 
1832 	return 0; /* is metadata */
1833 }
1834 
1835 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1836 					  u64 dev_bytenr, char **mapped_datav,
1837 					  unsigned int num_pages,
1838 					  struct bio *bio, int *bio_is_patched,
1839 					  struct buffer_head *bh,
1840 					  int submit_bio_bh_rw)
1841 {
1842 	int is_metadata;
1843 	struct btrfsic_block *block;
1844 	struct btrfsic_block_data_ctx block_ctx;
1845 	int ret;
1846 	struct btrfsic_state *state = dev_state->state;
1847 	struct block_device *bdev = dev_state->bdev;
1848 	unsigned int processed_len;
1849 
1850 	if (NULL != bio_is_patched)
1851 		*bio_is_patched = 0;
1852 
1853 again:
1854 	if (num_pages == 0)
1855 		return;
1856 
1857 	processed_len = 0;
1858 	is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1859 						      num_pages));
1860 
1861 	block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1862 					       &state->block_hashtable);
1863 	if (NULL != block) {
1864 		u64 bytenr = 0;
1865 		struct list_head *elem_ref_to;
1866 		struct list_head *tmp_ref_to;
1867 
1868 		if (block->is_superblock) {
1869 			bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1870 						    mapped_datav[0]);
1871 			if (num_pages * PAGE_CACHE_SIZE <
1872 			    BTRFS_SUPER_INFO_SIZE) {
1873 				printk(KERN_INFO
1874 				       "btrfsic: cannot work with too short bios!\n");
1875 				return;
1876 			}
1877 			is_metadata = 1;
1878 			BUG_ON(BTRFS_SUPER_INFO_SIZE & (PAGE_CACHE_SIZE - 1));
1879 			processed_len = BTRFS_SUPER_INFO_SIZE;
1880 			if (state->print_mask &
1881 			    BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1882 				printk(KERN_INFO
1883 				       "[before new superblock is written]:\n");
1884 				btrfsic_dump_tree_sub(state, block, 0);
1885 			}
1886 		}
1887 		if (is_metadata) {
1888 			if (!block->is_superblock) {
1889 				if (num_pages * PAGE_CACHE_SIZE <
1890 				    state->metablock_size) {
1891 					printk(KERN_INFO
1892 					       "btrfsic: cannot work with too short bios!\n");
1893 					return;
1894 				}
1895 				processed_len = state->metablock_size;
1896 				bytenr = btrfs_stack_header_bytenr(
1897 						(struct btrfs_header *)
1898 						mapped_datav[0]);
1899 				btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1900 							       dev_state,
1901 							       dev_bytenr);
1902 			}
1903 			if (block->logical_bytenr != bytenr &&
1904 			    !(!block->is_metadata &&
1905 			      block->logical_bytenr == 0))
1906 				printk(KERN_INFO
1907 				       "Written block @%llu (%s/%llu/%d)"
1908 				       " found in hash table, %c,"
1909 				       " bytenr mismatch"
1910 				       " (!= stored %llu).\n",
1911 				       bytenr, dev_state->name, dev_bytenr,
1912 				       block->mirror_num,
1913 				       btrfsic_get_block_type(state, block),
1914 				       block->logical_bytenr);
1915 			else if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1916 				printk(KERN_INFO
1917 				       "Written block @%llu (%s/%llu/%d)"
1918 				       " found in hash table, %c.\n",
1919 				       bytenr, dev_state->name, dev_bytenr,
1920 				       block->mirror_num,
1921 				       btrfsic_get_block_type(state, block));
1922 			block->logical_bytenr = bytenr;
1923 		} else {
1924 			if (num_pages * PAGE_CACHE_SIZE <
1925 			    state->datablock_size) {
1926 				printk(KERN_INFO
1927 				       "btrfsic: cannot work with too short bios!\n");
1928 				return;
1929 			}
1930 			processed_len = state->datablock_size;
1931 			bytenr = block->logical_bytenr;
1932 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1933 				printk(KERN_INFO
1934 				       "Written block @%llu (%s/%llu/%d)"
1935 				       " found in hash table, %c.\n",
1936 				       bytenr, dev_state->name, dev_bytenr,
1937 				       block->mirror_num,
1938 				       btrfsic_get_block_type(state, block));
1939 		}
1940 
1941 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1942 			printk(KERN_INFO
1943 			       "ref_to_list: %cE, ref_from_list: %cE\n",
1944 			       list_empty(&block->ref_to_list) ? ' ' : '!',
1945 			       list_empty(&block->ref_from_list) ? ' ' : '!');
1946 		if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1947 			printk(KERN_INFO "btrfs: attempt to overwrite %c-block"
1948 			       " @%llu (%s/%llu/%d), old(gen=%llu,"
1949 			       " objectid=%llu, type=%d, offset=%llu),"
1950 			       " new(gen=%llu),"
1951 			       " which is referenced by most recent superblock"
1952 			       " (superblockgen=%llu)!\n",
1953 			       btrfsic_get_block_type(state, block), bytenr,
1954 			       dev_state->name, dev_bytenr, block->mirror_num,
1955 			       block->generation,
1956 			       btrfs_disk_key_objectid(&block->disk_key),
1957 			       block->disk_key.type,
1958 			       btrfs_disk_key_offset(&block->disk_key),
1959 			       btrfs_stack_header_generation(
1960 				       (struct btrfs_header *) mapped_datav[0]),
1961 			       state->max_superblock_generation);
1962 			btrfsic_dump_tree(state);
1963 		}
1964 
1965 		if (!block->is_iodone && !block->never_written) {
1966 			printk(KERN_INFO "btrfs: attempt to overwrite %c-block"
1967 			       " @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu,"
1968 			       " which is not yet iodone!\n",
1969 			       btrfsic_get_block_type(state, block), bytenr,
1970 			       dev_state->name, dev_bytenr, block->mirror_num,
1971 			       block->generation,
1972 			       btrfs_stack_header_generation(
1973 				       (struct btrfs_header *)
1974 				       mapped_datav[0]));
1975 			/* it would not be safe to go on */
1976 			btrfsic_dump_tree(state);
1977 			goto continue_loop;
1978 		}
1979 
1980 		/*
1981 		 * Clear all references of this block. Do not free
1982 		 * the block itself even if is not referenced anymore
1983 		 * because it still carries valueable information
1984 		 * like whether it was ever written and IO completed.
1985 		 */
1986 		list_for_each_safe(elem_ref_to, tmp_ref_to,
1987 				   &block->ref_to_list) {
1988 			struct btrfsic_block_link *const l =
1989 			    list_entry(elem_ref_to,
1990 				       struct btrfsic_block_link,
1991 				       node_ref_to);
1992 
1993 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1994 				btrfsic_print_rem_link(state, l);
1995 			l->ref_cnt--;
1996 			if (0 == l->ref_cnt) {
1997 				list_del(&l->node_ref_to);
1998 				list_del(&l->node_ref_from);
1999 				btrfsic_block_link_hashtable_remove(l);
2000 				btrfsic_block_link_free(l);
2001 			}
2002 		}
2003 
2004 		if (block->is_superblock)
2005 			ret = btrfsic_map_superblock(state, bytenr,
2006 						     processed_len,
2007 						     bdev, &block_ctx);
2008 		else
2009 			ret = btrfsic_map_block(state, bytenr, processed_len,
2010 						&block_ctx, 0);
2011 		if (ret) {
2012 			printk(KERN_INFO
2013 			       "btrfsic: btrfsic_map_block(root @%llu)"
2014 			       " failed!\n", bytenr);
2015 			goto continue_loop;
2016 		}
2017 		block_ctx.datav = mapped_datav;
2018 		/* the following is required in case of writes to mirrors,
2019 		 * use the same that was used for the lookup */
2020 		block_ctx.dev = dev_state;
2021 		block_ctx.dev_bytenr = dev_bytenr;
2022 
2023 		if (is_metadata || state->include_extent_data) {
2024 			block->never_written = 0;
2025 			block->iodone_w_error = 0;
2026 			if (NULL != bio) {
2027 				block->is_iodone = 0;
2028 				BUG_ON(NULL == bio_is_patched);
2029 				if (!*bio_is_patched) {
2030 					block->orig_bio_bh_private =
2031 					    bio->bi_private;
2032 					block->orig_bio_bh_end_io.bio =
2033 					    bio->bi_end_io;
2034 					block->next_in_same_bio = NULL;
2035 					bio->bi_private = block;
2036 					bio->bi_end_io = btrfsic_bio_end_io;
2037 					*bio_is_patched = 1;
2038 				} else {
2039 					struct btrfsic_block *chained_block =
2040 					    (struct btrfsic_block *)
2041 					    bio->bi_private;
2042 
2043 					BUG_ON(NULL == chained_block);
2044 					block->orig_bio_bh_private =
2045 					    chained_block->orig_bio_bh_private;
2046 					block->orig_bio_bh_end_io.bio =
2047 					    chained_block->orig_bio_bh_end_io.
2048 					    bio;
2049 					block->next_in_same_bio = chained_block;
2050 					bio->bi_private = block;
2051 				}
2052 			} else if (NULL != bh) {
2053 				block->is_iodone = 0;
2054 				block->orig_bio_bh_private = bh->b_private;
2055 				block->orig_bio_bh_end_io.bh = bh->b_end_io;
2056 				block->next_in_same_bio = NULL;
2057 				bh->b_private = block;
2058 				bh->b_end_io = btrfsic_bh_end_io;
2059 			} else {
2060 				block->is_iodone = 1;
2061 				block->orig_bio_bh_private = NULL;
2062 				block->orig_bio_bh_end_io.bio = NULL;
2063 				block->next_in_same_bio = NULL;
2064 			}
2065 		}
2066 
2067 		block->flush_gen = dev_state->last_flush_gen + 1;
2068 		block->submit_bio_bh_rw = submit_bio_bh_rw;
2069 		if (is_metadata) {
2070 			block->logical_bytenr = bytenr;
2071 			block->is_metadata = 1;
2072 			if (block->is_superblock) {
2073 				BUG_ON(PAGE_CACHE_SIZE !=
2074 				       BTRFS_SUPER_INFO_SIZE);
2075 				ret = btrfsic_process_written_superblock(
2076 						state,
2077 						block,
2078 						(struct btrfs_super_block *)
2079 						mapped_datav[0]);
2080 				if (state->print_mask &
2081 				    BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
2082 					printk(KERN_INFO
2083 					"[after new superblock is written]:\n");
2084 					btrfsic_dump_tree_sub(state, block, 0);
2085 				}
2086 			} else {
2087 				block->mirror_num = 0;	/* unknown */
2088 				ret = btrfsic_process_metablock(
2089 						state,
2090 						block,
2091 						&block_ctx,
2092 						0, 0);
2093 			}
2094 			if (ret)
2095 				printk(KERN_INFO
2096 				       "btrfsic: btrfsic_process_metablock"
2097 				       "(root @%llu) failed!\n",
2098 				       dev_bytenr);
2099 		} else {
2100 			block->is_metadata = 0;
2101 			block->mirror_num = 0;	/* unknown */
2102 			block->generation = BTRFSIC_GENERATION_UNKNOWN;
2103 			if (!state->include_extent_data
2104 			    && list_empty(&block->ref_from_list)) {
2105 				/*
2106 				 * disk block is overwritten with extent
2107 				 * data (not meta data) and we are configured
2108 				 * to not include extent data: take the
2109 				 * chance and free the block's memory
2110 				 */
2111 				btrfsic_block_hashtable_remove(block);
2112 				list_del(&block->all_blocks_node);
2113 				btrfsic_block_free(block);
2114 			}
2115 		}
2116 		btrfsic_release_block_ctx(&block_ctx);
2117 	} else {
2118 		/* block has not been found in hash table */
2119 		u64 bytenr;
2120 
2121 		if (!is_metadata) {
2122 			processed_len = state->datablock_size;
2123 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2124 				printk(KERN_INFO "Written block (%s/%llu/?)"
2125 				       " !found in hash table, D.\n",
2126 				       dev_state->name, dev_bytenr);
2127 			if (!state->include_extent_data) {
2128 				/* ignore that written D block */
2129 				goto continue_loop;
2130 			}
2131 
2132 			/* this is getting ugly for the
2133 			 * include_extent_data case... */
2134 			bytenr = 0;	/* unknown */
2135 			block_ctx.start = bytenr;
2136 			block_ctx.len = processed_len;
2137 			block_ctx.mem_to_free = NULL;
2138 			block_ctx.pagev = NULL;
2139 		} else {
2140 			processed_len = state->metablock_size;
2141 			bytenr = btrfs_stack_header_bytenr(
2142 					(struct btrfs_header *)
2143 					mapped_datav[0]);
2144 			btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
2145 						       dev_bytenr);
2146 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2147 				printk(KERN_INFO
2148 				       "Written block @%llu (%s/%llu/?)"
2149 				       " !found in hash table, M.\n",
2150 				       bytenr, dev_state->name, dev_bytenr);
2151 
2152 			ret = btrfsic_map_block(state, bytenr, processed_len,
2153 						&block_ctx, 0);
2154 			if (ret) {
2155 				printk(KERN_INFO
2156 				       "btrfsic: btrfsic_map_block(root @%llu)"
2157 				       " failed!\n",
2158 				       dev_bytenr);
2159 				goto continue_loop;
2160 			}
2161 		}
2162 		block_ctx.datav = mapped_datav;
2163 		/* the following is required in case of writes to mirrors,
2164 		 * use the same that was used for the lookup */
2165 		block_ctx.dev = dev_state;
2166 		block_ctx.dev_bytenr = dev_bytenr;
2167 
2168 		block = btrfsic_block_alloc();
2169 		if (NULL == block) {
2170 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
2171 			btrfsic_release_block_ctx(&block_ctx);
2172 			goto continue_loop;
2173 		}
2174 		block->dev_state = dev_state;
2175 		block->dev_bytenr = dev_bytenr;
2176 		block->logical_bytenr = bytenr;
2177 		block->is_metadata = is_metadata;
2178 		block->never_written = 0;
2179 		block->iodone_w_error = 0;
2180 		block->mirror_num = 0;	/* unknown */
2181 		block->flush_gen = dev_state->last_flush_gen + 1;
2182 		block->submit_bio_bh_rw = submit_bio_bh_rw;
2183 		if (NULL != bio) {
2184 			block->is_iodone = 0;
2185 			BUG_ON(NULL == bio_is_patched);
2186 			if (!*bio_is_patched) {
2187 				block->orig_bio_bh_private = bio->bi_private;
2188 				block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2189 				block->next_in_same_bio = NULL;
2190 				bio->bi_private = block;
2191 				bio->bi_end_io = btrfsic_bio_end_io;
2192 				*bio_is_patched = 1;
2193 			} else {
2194 				struct btrfsic_block *chained_block =
2195 				    (struct btrfsic_block *)
2196 				    bio->bi_private;
2197 
2198 				BUG_ON(NULL == chained_block);
2199 				block->orig_bio_bh_private =
2200 				    chained_block->orig_bio_bh_private;
2201 				block->orig_bio_bh_end_io.bio =
2202 				    chained_block->orig_bio_bh_end_io.bio;
2203 				block->next_in_same_bio = chained_block;
2204 				bio->bi_private = block;
2205 			}
2206 		} else if (NULL != bh) {
2207 			block->is_iodone = 0;
2208 			block->orig_bio_bh_private = bh->b_private;
2209 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2210 			block->next_in_same_bio = NULL;
2211 			bh->b_private = block;
2212 			bh->b_end_io = btrfsic_bh_end_io;
2213 		} else {
2214 			block->is_iodone = 1;
2215 			block->orig_bio_bh_private = NULL;
2216 			block->orig_bio_bh_end_io.bio = NULL;
2217 			block->next_in_same_bio = NULL;
2218 		}
2219 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2220 			printk(KERN_INFO
2221 			       "New written %c-block @%llu (%s/%llu/%d)\n",
2222 			       is_metadata ? 'M' : 'D',
2223 			       block->logical_bytenr, block->dev_state->name,
2224 			       block->dev_bytenr, block->mirror_num);
2225 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2226 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2227 
2228 		if (is_metadata) {
2229 			ret = btrfsic_process_metablock(state, block,
2230 							&block_ctx, 0, 0);
2231 			if (ret)
2232 				printk(KERN_INFO
2233 				       "btrfsic: process_metablock(root @%llu)"
2234 				       " failed!\n",
2235 				       dev_bytenr);
2236 		}
2237 		btrfsic_release_block_ctx(&block_ctx);
2238 	}
2239 
2240 continue_loop:
2241 	BUG_ON(!processed_len);
2242 	dev_bytenr += processed_len;
2243 	mapped_datav += processed_len >> PAGE_CACHE_SHIFT;
2244 	num_pages -= processed_len >> PAGE_CACHE_SHIFT;
2245 	goto again;
2246 }
2247 
2248 static void btrfsic_bio_end_io(struct bio *bp, int bio_error_status)
2249 {
2250 	struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2251 	int iodone_w_error;
2252 
2253 	/* mutex is not held! This is not save if IO is not yet completed
2254 	 * on umount */
2255 	iodone_w_error = 0;
2256 	if (bio_error_status)
2257 		iodone_w_error = 1;
2258 
2259 	BUG_ON(NULL == block);
2260 	bp->bi_private = block->orig_bio_bh_private;
2261 	bp->bi_end_io = block->orig_bio_bh_end_io.bio;
2262 
2263 	do {
2264 		struct btrfsic_block *next_block;
2265 		struct btrfsic_dev_state *const dev_state = block->dev_state;
2266 
2267 		if ((dev_state->state->print_mask &
2268 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2269 			printk(KERN_INFO
2270 			       "bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2271 			       bio_error_status,
2272 			       btrfsic_get_block_type(dev_state->state, block),
2273 			       block->logical_bytenr, dev_state->name,
2274 			       block->dev_bytenr, block->mirror_num);
2275 		next_block = block->next_in_same_bio;
2276 		block->iodone_w_error = iodone_w_error;
2277 		if (block->submit_bio_bh_rw & REQ_FLUSH) {
2278 			dev_state->last_flush_gen++;
2279 			if ((dev_state->state->print_mask &
2280 			     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2281 				printk(KERN_INFO
2282 				       "bio_end_io() new %s flush_gen=%llu\n",
2283 				       dev_state->name,
2284 				       dev_state->last_flush_gen);
2285 		}
2286 		if (block->submit_bio_bh_rw & REQ_FUA)
2287 			block->flush_gen = 0; /* FUA completed means block is
2288 					       * on disk */
2289 		block->is_iodone = 1; /* for FLUSH, this releases the block */
2290 		block = next_block;
2291 	} while (NULL != block);
2292 
2293 	bp->bi_end_io(bp, bio_error_status);
2294 }
2295 
2296 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate)
2297 {
2298 	struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private;
2299 	int iodone_w_error = !uptodate;
2300 	struct btrfsic_dev_state *dev_state;
2301 
2302 	BUG_ON(NULL == block);
2303 	dev_state = block->dev_state;
2304 	if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2305 		printk(KERN_INFO
2306 		       "bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n",
2307 		       iodone_w_error,
2308 		       btrfsic_get_block_type(dev_state->state, block),
2309 		       block->logical_bytenr, block->dev_state->name,
2310 		       block->dev_bytenr, block->mirror_num);
2311 
2312 	block->iodone_w_error = iodone_w_error;
2313 	if (block->submit_bio_bh_rw & REQ_FLUSH) {
2314 		dev_state->last_flush_gen++;
2315 		if ((dev_state->state->print_mask &
2316 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2317 			printk(KERN_INFO
2318 			       "bh_end_io() new %s flush_gen=%llu\n",
2319 			       dev_state->name, dev_state->last_flush_gen);
2320 	}
2321 	if (block->submit_bio_bh_rw & REQ_FUA)
2322 		block->flush_gen = 0; /* FUA completed means block is on disk */
2323 
2324 	bh->b_private = block->orig_bio_bh_private;
2325 	bh->b_end_io = block->orig_bio_bh_end_io.bh;
2326 	block->is_iodone = 1; /* for FLUSH, this releases the block */
2327 	bh->b_end_io(bh, uptodate);
2328 }
2329 
2330 static int btrfsic_process_written_superblock(
2331 		struct btrfsic_state *state,
2332 		struct btrfsic_block *const superblock,
2333 		struct btrfs_super_block *const super_hdr)
2334 {
2335 	int pass;
2336 
2337 	superblock->generation = btrfs_super_generation(super_hdr);
2338 	if (!(superblock->generation > state->max_superblock_generation ||
2339 	      0 == state->max_superblock_generation)) {
2340 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2341 			printk(KERN_INFO
2342 			       "btrfsic: superblock @%llu (%s/%llu/%d)"
2343 			       " with old gen %llu <= %llu\n",
2344 			       superblock->logical_bytenr,
2345 			       superblock->dev_state->name,
2346 			       superblock->dev_bytenr, superblock->mirror_num,
2347 			       btrfs_super_generation(super_hdr),
2348 			       state->max_superblock_generation);
2349 	} else {
2350 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2351 			printk(KERN_INFO
2352 			       "btrfsic: got new superblock @%llu (%s/%llu/%d)"
2353 			       " with new gen %llu > %llu\n",
2354 			       superblock->logical_bytenr,
2355 			       superblock->dev_state->name,
2356 			       superblock->dev_bytenr, superblock->mirror_num,
2357 			       btrfs_super_generation(super_hdr),
2358 			       state->max_superblock_generation);
2359 
2360 		state->max_superblock_generation =
2361 		    btrfs_super_generation(super_hdr);
2362 		state->latest_superblock = superblock;
2363 	}
2364 
2365 	for (pass = 0; pass < 3; pass++) {
2366 		int ret;
2367 		u64 next_bytenr;
2368 		struct btrfsic_block *next_block;
2369 		struct btrfsic_block_data_ctx tmp_next_block_ctx;
2370 		struct btrfsic_block_link *l;
2371 		int num_copies;
2372 		int mirror_num;
2373 		const char *additional_string = NULL;
2374 		struct btrfs_disk_key tmp_disk_key = {0};
2375 
2376 		btrfs_set_disk_key_objectid(&tmp_disk_key,
2377 					    BTRFS_ROOT_ITEM_KEY);
2378 		btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2379 
2380 		switch (pass) {
2381 		case 0:
2382 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2383 						    BTRFS_ROOT_TREE_OBJECTID);
2384 			additional_string = "root ";
2385 			next_bytenr = btrfs_super_root(super_hdr);
2386 			if (state->print_mask &
2387 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2388 				printk(KERN_INFO "root@%llu\n", next_bytenr);
2389 			break;
2390 		case 1:
2391 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2392 						    BTRFS_CHUNK_TREE_OBJECTID);
2393 			additional_string = "chunk ";
2394 			next_bytenr = btrfs_super_chunk_root(super_hdr);
2395 			if (state->print_mask &
2396 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2397 				printk(KERN_INFO "chunk@%llu\n", next_bytenr);
2398 			break;
2399 		case 2:
2400 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2401 						    BTRFS_TREE_LOG_OBJECTID);
2402 			additional_string = "log ";
2403 			next_bytenr = btrfs_super_log_root(super_hdr);
2404 			if (0 == next_bytenr)
2405 				continue;
2406 			if (state->print_mask &
2407 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2408 				printk(KERN_INFO "log@%llu\n", next_bytenr);
2409 			break;
2410 		}
2411 
2412 		num_copies =
2413 		    btrfs_num_copies(state->root->fs_info,
2414 				     next_bytenr, BTRFS_SUPER_INFO_SIZE);
2415 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2416 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
2417 			       next_bytenr, num_copies);
2418 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2419 			int was_created;
2420 
2421 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2422 				printk(KERN_INFO
2423 				       "btrfsic_process_written_superblock("
2424 				       "mirror_num=%d)\n", mirror_num);
2425 			ret = btrfsic_map_block(state, next_bytenr,
2426 						BTRFS_SUPER_INFO_SIZE,
2427 						&tmp_next_block_ctx,
2428 						mirror_num);
2429 			if (ret) {
2430 				printk(KERN_INFO
2431 				       "btrfsic: btrfsic_map_block(@%llu,"
2432 				       " mirror=%d) failed!\n",
2433 				       next_bytenr, mirror_num);
2434 				return -1;
2435 			}
2436 
2437 			next_block = btrfsic_block_lookup_or_add(
2438 					state,
2439 					&tmp_next_block_ctx,
2440 					additional_string,
2441 					1, 0, 1,
2442 					mirror_num,
2443 					&was_created);
2444 			if (NULL == next_block) {
2445 				printk(KERN_INFO
2446 				       "btrfsic: error, kmalloc failed!\n");
2447 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
2448 				return -1;
2449 			}
2450 
2451 			next_block->disk_key = tmp_disk_key;
2452 			if (was_created)
2453 				next_block->generation =
2454 				    BTRFSIC_GENERATION_UNKNOWN;
2455 			l = btrfsic_block_link_lookup_or_add(
2456 					state,
2457 					&tmp_next_block_ctx,
2458 					next_block,
2459 					superblock,
2460 					BTRFSIC_GENERATION_UNKNOWN);
2461 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
2462 			if (NULL == l)
2463 				return -1;
2464 		}
2465 	}
2466 
2467 	if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2468 		btrfsic_dump_tree(state);
2469 
2470 	return 0;
2471 }
2472 
2473 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2474 					struct btrfsic_block *const block,
2475 					int recursion_level)
2476 {
2477 	struct list_head *elem_ref_to;
2478 	int ret = 0;
2479 
2480 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2481 		/*
2482 		 * Note that this situation can happen and does not
2483 		 * indicate an error in regular cases. It happens
2484 		 * when disk blocks are freed and later reused.
2485 		 * The check-integrity module is not aware of any
2486 		 * block free operations, it just recognizes block
2487 		 * write operations. Therefore it keeps the linkage
2488 		 * information for a block until a block is
2489 		 * rewritten. This can temporarily cause incorrect
2490 		 * and even circular linkage informations. This
2491 		 * causes no harm unless such blocks are referenced
2492 		 * by the most recent super block.
2493 		 */
2494 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2495 			printk(KERN_INFO
2496 			       "btrfsic: abort cyclic linkage (case 1).\n");
2497 
2498 		return ret;
2499 	}
2500 
2501 	/*
2502 	 * This algorithm is recursive because the amount of used stack
2503 	 * space is very small and the max recursion depth is limited.
2504 	 */
2505 	list_for_each(elem_ref_to, &block->ref_to_list) {
2506 		const struct btrfsic_block_link *const l =
2507 		    list_entry(elem_ref_to, struct btrfsic_block_link,
2508 			       node_ref_to);
2509 
2510 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2511 			printk(KERN_INFO
2512 			       "rl=%d, %c @%llu (%s/%llu/%d)"
2513 			       " %u* refers to %c @%llu (%s/%llu/%d)\n",
2514 			       recursion_level,
2515 			       btrfsic_get_block_type(state, block),
2516 			       block->logical_bytenr, block->dev_state->name,
2517 			       block->dev_bytenr, block->mirror_num,
2518 			       l->ref_cnt,
2519 			       btrfsic_get_block_type(state, l->block_ref_to),
2520 			       l->block_ref_to->logical_bytenr,
2521 			       l->block_ref_to->dev_state->name,
2522 			       l->block_ref_to->dev_bytenr,
2523 			       l->block_ref_to->mirror_num);
2524 		if (l->block_ref_to->never_written) {
2525 			printk(KERN_INFO "btrfs: attempt to write superblock"
2526 			       " which references block %c @%llu (%s/%llu/%d)"
2527 			       " which is never written!\n",
2528 			       btrfsic_get_block_type(state, l->block_ref_to),
2529 			       l->block_ref_to->logical_bytenr,
2530 			       l->block_ref_to->dev_state->name,
2531 			       l->block_ref_to->dev_bytenr,
2532 			       l->block_ref_to->mirror_num);
2533 			ret = -1;
2534 		} else if (!l->block_ref_to->is_iodone) {
2535 			printk(KERN_INFO "btrfs: attempt to write superblock"
2536 			       " which references block %c @%llu (%s/%llu/%d)"
2537 			       " which is not yet iodone!\n",
2538 			       btrfsic_get_block_type(state, l->block_ref_to),
2539 			       l->block_ref_to->logical_bytenr,
2540 			       l->block_ref_to->dev_state->name,
2541 			       l->block_ref_to->dev_bytenr,
2542 			       l->block_ref_to->mirror_num);
2543 			ret = -1;
2544 		} else if (l->block_ref_to->iodone_w_error) {
2545 			printk(KERN_INFO "btrfs: attempt to write superblock"
2546 			       " which references block %c @%llu (%s/%llu/%d)"
2547 			       " which has write error!\n",
2548 			       btrfsic_get_block_type(state, l->block_ref_to),
2549 			       l->block_ref_to->logical_bytenr,
2550 			       l->block_ref_to->dev_state->name,
2551 			       l->block_ref_to->dev_bytenr,
2552 			       l->block_ref_to->mirror_num);
2553 			ret = -1;
2554 		} else if (l->parent_generation !=
2555 			   l->block_ref_to->generation &&
2556 			   BTRFSIC_GENERATION_UNKNOWN !=
2557 			   l->parent_generation &&
2558 			   BTRFSIC_GENERATION_UNKNOWN !=
2559 			   l->block_ref_to->generation) {
2560 			printk(KERN_INFO "btrfs: attempt to write superblock"
2561 			       " which references block %c @%llu (%s/%llu/%d)"
2562 			       " with generation %llu !="
2563 			       " parent generation %llu!\n",
2564 			       btrfsic_get_block_type(state, l->block_ref_to),
2565 			       l->block_ref_to->logical_bytenr,
2566 			       l->block_ref_to->dev_state->name,
2567 			       l->block_ref_to->dev_bytenr,
2568 			       l->block_ref_to->mirror_num,
2569 			       l->block_ref_to->generation,
2570 			       l->parent_generation);
2571 			ret = -1;
2572 		} else if (l->block_ref_to->flush_gen >
2573 			   l->block_ref_to->dev_state->last_flush_gen) {
2574 			printk(KERN_INFO "btrfs: attempt to write superblock"
2575 			       " which references block %c @%llu (%s/%llu/%d)"
2576 			       " which is not flushed out of disk's write cache"
2577 			       " (block flush_gen=%llu,"
2578 			       " dev->flush_gen=%llu)!\n",
2579 			       btrfsic_get_block_type(state, l->block_ref_to),
2580 			       l->block_ref_to->logical_bytenr,
2581 			       l->block_ref_to->dev_state->name,
2582 			       l->block_ref_to->dev_bytenr,
2583 			       l->block_ref_to->mirror_num, block->flush_gen,
2584 			       l->block_ref_to->dev_state->last_flush_gen);
2585 			ret = -1;
2586 		} else if (-1 == btrfsic_check_all_ref_blocks(state,
2587 							      l->block_ref_to,
2588 							      recursion_level +
2589 							      1)) {
2590 			ret = -1;
2591 		}
2592 	}
2593 
2594 	return ret;
2595 }
2596 
2597 static int btrfsic_is_block_ref_by_superblock(
2598 		const struct btrfsic_state *state,
2599 		const struct btrfsic_block *block,
2600 		int recursion_level)
2601 {
2602 	struct list_head *elem_ref_from;
2603 
2604 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2605 		/* refer to comment at "abort cyclic linkage (case 1)" */
2606 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2607 			printk(KERN_INFO
2608 			       "btrfsic: abort cyclic linkage (case 2).\n");
2609 
2610 		return 0;
2611 	}
2612 
2613 	/*
2614 	 * This algorithm is recursive because the amount of used stack space
2615 	 * is very small and the max recursion depth is limited.
2616 	 */
2617 	list_for_each(elem_ref_from, &block->ref_from_list) {
2618 		const struct btrfsic_block_link *const l =
2619 		    list_entry(elem_ref_from, struct btrfsic_block_link,
2620 			       node_ref_from);
2621 
2622 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2623 			printk(KERN_INFO
2624 			       "rl=%d, %c @%llu (%s/%llu/%d)"
2625 			       " is ref %u* from %c @%llu (%s/%llu/%d)\n",
2626 			       recursion_level,
2627 			       btrfsic_get_block_type(state, block),
2628 			       block->logical_bytenr, block->dev_state->name,
2629 			       block->dev_bytenr, block->mirror_num,
2630 			       l->ref_cnt,
2631 			       btrfsic_get_block_type(state, l->block_ref_from),
2632 			       l->block_ref_from->logical_bytenr,
2633 			       l->block_ref_from->dev_state->name,
2634 			       l->block_ref_from->dev_bytenr,
2635 			       l->block_ref_from->mirror_num);
2636 		if (l->block_ref_from->is_superblock &&
2637 		    state->latest_superblock->dev_bytenr ==
2638 		    l->block_ref_from->dev_bytenr &&
2639 		    state->latest_superblock->dev_state->bdev ==
2640 		    l->block_ref_from->dev_state->bdev)
2641 			return 1;
2642 		else if (btrfsic_is_block_ref_by_superblock(state,
2643 							    l->block_ref_from,
2644 							    recursion_level +
2645 							    1))
2646 			return 1;
2647 	}
2648 
2649 	return 0;
2650 }
2651 
2652 static void btrfsic_print_add_link(const struct btrfsic_state *state,
2653 				   const struct btrfsic_block_link *l)
2654 {
2655 	printk(KERN_INFO
2656 	       "Add %u* link from %c @%llu (%s/%llu/%d)"
2657 	       " to %c @%llu (%s/%llu/%d).\n",
2658 	       l->ref_cnt,
2659 	       btrfsic_get_block_type(state, l->block_ref_from),
2660 	       l->block_ref_from->logical_bytenr,
2661 	       l->block_ref_from->dev_state->name,
2662 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2663 	       btrfsic_get_block_type(state, l->block_ref_to),
2664 	       l->block_ref_to->logical_bytenr,
2665 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2666 	       l->block_ref_to->mirror_num);
2667 }
2668 
2669 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2670 				   const struct btrfsic_block_link *l)
2671 {
2672 	printk(KERN_INFO
2673 	       "Rem %u* link from %c @%llu (%s/%llu/%d)"
2674 	       " to %c @%llu (%s/%llu/%d).\n",
2675 	       l->ref_cnt,
2676 	       btrfsic_get_block_type(state, l->block_ref_from),
2677 	       l->block_ref_from->logical_bytenr,
2678 	       l->block_ref_from->dev_state->name,
2679 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2680 	       btrfsic_get_block_type(state, l->block_ref_to),
2681 	       l->block_ref_to->logical_bytenr,
2682 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2683 	       l->block_ref_to->mirror_num);
2684 }
2685 
2686 static char btrfsic_get_block_type(const struct btrfsic_state *state,
2687 				   const struct btrfsic_block *block)
2688 {
2689 	if (block->is_superblock &&
2690 	    state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2691 	    state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2692 		return 'S';
2693 	else if (block->is_superblock)
2694 		return 's';
2695 	else if (block->is_metadata)
2696 		return 'M';
2697 	else
2698 		return 'D';
2699 }
2700 
2701 static void btrfsic_dump_tree(const struct btrfsic_state *state)
2702 {
2703 	btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2704 }
2705 
2706 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2707 				  const struct btrfsic_block *block,
2708 				  int indent_level)
2709 {
2710 	struct list_head *elem_ref_to;
2711 	int indent_add;
2712 	static char buf[80];
2713 	int cursor_position;
2714 
2715 	/*
2716 	 * Should better fill an on-stack buffer with a complete line and
2717 	 * dump it at once when it is time to print a newline character.
2718 	 */
2719 
2720 	/*
2721 	 * This algorithm is recursive because the amount of used stack space
2722 	 * is very small and the max recursion depth is limited.
2723 	 */
2724 	indent_add = sprintf(buf, "%c-%llu(%s/%llu/%d)",
2725 			     btrfsic_get_block_type(state, block),
2726 			     block->logical_bytenr, block->dev_state->name,
2727 			     block->dev_bytenr, block->mirror_num);
2728 	if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2729 		printk("[...]\n");
2730 		return;
2731 	}
2732 	printk(buf);
2733 	indent_level += indent_add;
2734 	if (list_empty(&block->ref_to_list)) {
2735 		printk("\n");
2736 		return;
2737 	}
2738 	if (block->mirror_num > 1 &&
2739 	    !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2740 		printk(" [...]\n");
2741 		return;
2742 	}
2743 
2744 	cursor_position = indent_level;
2745 	list_for_each(elem_ref_to, &block->ref_to_list) {
2746 		const struct btrfsic_block_link *const l =
2747 		    list_entry(elem_ref_to, struct btrfsic_block_link,
2748 			       node_ref_to);
2749 
2750 		while (cursor_position < indent_level) {
2751 			printk(" ");
2752 			cursor_position++;
2753 		}
2754 		if (l->ref_cnt > 1)
2755 			indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2756 		else
2757 			indent_add = sprintf(buf, " --> ");
2758 		if (indent_level + indent_add >
2759 		    BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2760 			printk("[...]\n");
2761 			cursor_position = 0;
2762 			continue;
2763 		}
2764 
2765 		printk(buf);
2766 
2767 		btrfsic_dump_tree_sub(state, l->block_ref_to,
2768 				      indent_level + indent_add);
2769 		cursor_position = 0;
2770 	}
2771 }
2772 
2773 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2774 		struct btrfsic_state *state,
2775 		struct btrfsic_block_data_ctx *next_block_ctx,
2776 		struct btrfsic_block *next_block,
2777 		struct btrfsic_block *from_block,
2778 		u64 parent_generation)
2779 {
2780 	struct btrfsic_block_link *l;
2781 
2782 	l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2783 						next_block_ctx->dev_bytenr,
2784 						from_block->dev_state->bdev,
2785 						from_block->dev_bytenr,
2786 						&state->block_link_hashtable);
2787 	if (NULL == l) {
2788 		l = btrfsic_block_link_alloc();
2789 		if (NULL == l) {
2790 			printk(KERN_INFO
2791 			       "btrfsic: error, kmalloc" " failed!\n");
2792 			return NULL;
2793 		}
2794 
2795 		l->block_ref_to = next_block;
2796 		l->block_ref_from = from_block;
2797 		l->ref_cnt = 1;
2798 		l->parent_generation = parent_generation;
2799 
2800 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2801 			btrfsic_print_add_link(state, l);
2802 
2803 		list_add(&l->node_ref_to, &from_block->ref_to_list);
2804 		list_add(&l->node_ref_from, &next_block->ref_from_list);
2805 
2806 		btrfsic_block_link_hashtable_add(l,
2807 						 &state->block_link_hashtable);
2808 	} else {
2809 		l->ref_cnt++;
2810 		l->parent_generation = parent_generation;
2811 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2812 			btrfsic_print_add_link(state, l);
2813 	}
2814 
2815 	return l;
2816 }
2817 
2818 static struct btrfsic_block *btrfsic_block_lookup_or_add(
2819 		struct btrfsic_state *state,
2820 		struct btrfsic_block_data_ctx *block_ctx,
2821 		const char *additional_string,
2822 		int is_metadata,
2823 		int is_iodone,
2824 		int never_written,
2825 		int mirror_num,
2826 		int *was_created)
2827 {
2828 	struct btrfsic_block *block;
2829 
2830 	block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2831 					       block_ctx->dev_bytenr,
2832 					       &state->block_hashtable);
2833 	if (NULL == block) {
2834 		struct btrfsic_dev_state *dev_state;
2835 
2836 		block = btrfsic_block_alloc();
2837 		if (NULL == block) {
2838 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
2839 			return NULL;
2840 		}
2841 		dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev);
2842 		if (NULL == dev_state) {
2843 			printk(KERN_INFO
2844 			       "btrfsic: error, lookup dev_state failed!\n");
2845 			btrfsic_block_free(block);
2846 			return NULL;
2847 		}
2848 		block->dev_state = dev_state;
2849 		block->dev_bytenr = block_ctx->dev_bytenr;
2850 		block->logical_bytenr = block_ctx->start;
2851 		block->is_metadata = is_metadata;
2852 		block->is_iodone = is_iodone;
2853 		block->never_written = never_written;
2854 		block->mirror_num = mirror_num;
2855 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2856 			printk(KERN_INFO
2857 			       "New %s%c-block @%llu (%s/%llu/%d)\n",
2858 			       additional_string,
2859 			       btrfsic_get_block_type(state, block),
2860 			       block->logical_bytenr, dev_state->name,
2861 			       block->dev_bytenr, mirror_num);
2862 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2863 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2864 		if (NULL != was_created)
2865 			*was_created = 1;
2866 	} else {
2867 		if (NULL != was_created)
2868 			*was_created = 0;
2869 	}
2870 
2871 	return block;
2872 }
2873 
2874 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2875 					   u64 bytenr,
2876 					   struct btrfsic_dev_state *dev_state,
2877 					   u64 dev_bytenr)
2878 {
2879 	int num_copies;
2880 	int mirror_num;
2881 	int ret;
2882 	struct btrfsic_block_data_ctx block_ctx;
2883 	int match = 0;
2884 
2885 	num_copies = btrfs_num_copies(state->root->fs_info,
2886 				      bytenr, state->metablock_size);
2887 
2888 	for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2889 		ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2890 					&block_ctx, mirror_num);
2891 		if (ret) {
2892 			printk(KERN_INFO "btrfsic:"
2893 			       " btrfsic_map_block(logical @%llu,"
2894 			       " mirror %d) failed!\n",
2895 			       bytenr, mirror_num);
2896 			continue;
2897 		}
2898 
2899 		if (dev_state->bdev == block_ctx.dev->bdev &&
2900 		    dev_bytenr == block_ctx.dev_bytenr) {
2901 			match++;
2902 			btrfsic_release_block_ctx(&block_ctx);
2903 			break;
2904 		}
2905 		btrfsic_release_block_ctx(&block_ctx);
2906 	}
2907 
2908 	if (WARN_ON(!match)) {
2909 		printk(KERN_INFO "btrfs: attempt to write M-block which contains logical bytenr that doesn't map to dev+physical bytenr of submit_bio,"
2910 		       " buffer->log_bytenr=%llu, submit_bio(bdev=%s,"
2911 		       " phys_bytenr=%llu)!\n",
2912 		       bytenr, dev_state->name, dev_bytenr);
2913 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2914 			ret = btrfsic_map_block(state, bytenr,
2915 						state->metablock_size,
2916 						&block_ctx, mirror_num);
2917 			if (ret)
2918 				continue;
2919 
2920 			printk(KERN_INFO "Read logical bytenr @%llu maps to"
2921 			       " (%s/%llu/%d)\n",
2922 			       bytenr, block_ctx.dev->name,
2923 			       block_ctx.dev_bytenr, mirror_num);
2924 		}
2925 	}
2926 }
2927 
2928 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
2929 		struct block_device *bdev)
2930 {
2931 	struct btrfsic_dev_state *ds;
2932 
2933 	ds = btrfsic_dev_state_hashtable_lookup(bdev,
2934 						&btrfsic_dev_state_hashtable);
2935 	return ds;
2936 }
2937 
2938 int btrfsic_submit_bh(int rw, struct buffer_head *bh)
2939 {
2940 	struct btrfsic_dev_state *dev_state;
2941 
2942 	if (!btrfsic_is_initialized)
2943 		return submit_bh(rw, bh);
2944 
2945 	mutex_lock(&btrfsic_mutex);
2946 	/* since btrfsic_submit_bh() might also be called before
2947 	 * btrfsic_mount(), this might return NULL */
2948 	dev_state = btrfsic_dev_state_lookup(bh->b_bdev);
2949 
2950 	/* Only called to write the superblock (incl. FLUSH/FUA) */
2951 	if (NULL != dev_state &&
2952 	    (rw & WRITE) && bh->b_size > 0) {
2953 		u64 dev_bytenr;
2954 
2955 		dev_bytenr = 4096 * bh->b_blocknr;
2956 		if (dev_state->state->print_mask &
2957 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2958 			printk(KERN_INFO
2959 			       "submit_bh(rw=0x%x, blocknr=%llu (bytenr %llu),"
2960 			       " size=%zu, data=%p, bdev=%p)\n",
2961 			       rw, (unsigned long long)bh->b_blocknr,
2962 			       dev_bytenr, bh->b_size, bh->b_data, bh->b_bdev);
2963 		btrfsic_process_written_block(dev_state, dev_bytenr,
2964 					      &bh->b_data, 1, NULL,
2965 					      NULL, bh, rw);
2966 	} else if (NULL != dev_state && (rw & REQ_FLUSH)) {
2967 		if (dev_state->state->print_mask &
2968 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2969 			printk(KERN_INFO
2970 			       "submit_bh(rw=0x%x FLUSH, bdev=%p)\n",
2971 			       rw, bh->b_bdev);
2972 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2973 			if ((dev_state->state->print_mask &
2974 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2975 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2976 				printk(KERN_INFO
2977 				       "btrfsic_submit_bh(%s) with FLUSH"
2978 				       " but dummy block already in use"
2979 				       " (ignored)!\n",
2980 				       dev_state->name);
2981 		} else {
2982 			struct btrfsic_block *const block =
2983 				&dev_state->dummy_block_for_bio_bh_flush;
2984 
2985 			block->is_iodone = 0;
2986 			block->never_written = 0;
2987 			block->iodone_w_error = 0;
2988 			block->flush_gen = dev_state->last_flush_gen + 1;
2989 			block->submit_bio_bh_rw = rw;
2990 			block->orig_bio_bh_private = bh->b_private;
2991 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2992 			block->next_in_same_bio = NULL;
2993 			bh->b_private = block;
2994 			bh->b_end_io = btrfsic_bh_end_io;
2995 		}
2996 	}
2997 	mutex_unlock(&btrfsic_mutex);
2998 	return submit_bh(rw, bh);
2999 }
3000 
3001 static void __btrfsic_submit_bio(int rw, struct bio *bio)
3002 {
3003 	struct btrfsic_dev_state *dev_state;
3004 
3005 	if (!btrfsic_is_initialized)
3006 		return;
3007 
3008 	mutex_lock(&btrfsic_mutex);
3009 	/* since btrfsic_submit_bio() is also called before
3010 	 * btrfsic_mount(), this might return NULL */
3011 	dev_state = btrfsic_dev_state_lookup(bio->bi_bdev);
3012 	if (NULL != dev_state &&
3013 	    (rw & WRITE) && NULL != bio->bi_io_vec) {
3014 		unsigned int i;
3015 		u64 dev_bytenr;
3016 		u64 cur_bytenr;
3017 		int bio_is_patched;
3018 		char **mapped_datav;
3019 
3020 		dev_bytenr = 512 * bio->bi_iter.bi_sector;
3021 		bio_is_patched = 0;
3022 		if (dev_state->state->print_mask &
3023 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
3024 			printk(KERN_INFO
3025 			       "submit_bio(rw=0x%x, bi_vcnt=%u,"
3026 			       " bi_sector=%llu (bytenr %llu), bi_bdev=%p)\n",
3027 			       rw, bio->bi_vcnt,
3028 			       (unsigned long long)bio->bi_iter.bi_sector,
3029 			       dev_bytenr, bio->bi_bdev);
3030 
3031 		mapped_datav = kmalloc(sizeof(*mapped_datav) * bio->bi_vcnt,
3032 				       GFP_NOFS);
3033 		if (!mapped_datav)
3034 			goto leave;
3035 		cur_bytenr = dev_bytenr;
3036 		for (i = 0; i < bio->bi_vcnt; i++) {
3037 			BUG_ON(bio->bi_io_vec[i].bv_len != PAGE_CACHE_SIZE);
3038 			mapped_datav[i] = kmap(bio->bi_io_vec[i].bv_page);
3039 			if (!mapped_datav[i]) {
3040 				while (i > 0) {
3041 					i--;
3042 					kunmap(bio->bi_io_vec[i].bv_page);
3043 				}
3044 				kfree(mapped_datav);
3045 				goto leave;
3046 			}
3047 			if (dev_state->state->print_mask &
3048 			    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
3049 				printk(KERN_INFO
3050 				       "#%u: bytenr=%llu, len=%u, offset=%u\n",
3051 				       i, cur_bytenr, bio->bi_io_vec[i].bv_len,
3052 				       bio->bi_io_vec[i].bv_offset);
3053 			cur_bytenr += bio->bi_io_vec[i].bv_len;
3054 		}
3055 		btrfsic_process_written_block(dev_state, dev_bytenr,
3056 					      mapped_datav, bio->bi_vcnt,
3057 					      bio, &bio_is_patched,
3058 					      NULL, rw);
3059 		while (i > 0) {
3060 			i--;
3061 			kunmap(bio->bi_io_vec[i].bv_page);
3062 		}
3063 		kfree(mapped_datav);
3064 	} else if (NULL != dev_state && (rw & REQ_FLUSH)) {
3065 		if (dev_state->state->print_mask &
3066 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
3067 			printk(KERN_INFO
3068 			       "submit_bio(rw=0x%x FLUSH, bdev=%p)\n",
3069 			       rw, bio->bi_bdev);
3070 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
3071 			if ((dev_state->state->print_mask &
3072 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
3073 			      BTRFSIC_PRINT_MASK_VERBOSE)))
3074 				printk(KERN_INFO
3075 				       "btrfsic_submit_bio(%s) with FLUSH"
3076 				       " but dummy block already in use"
3077 				       " (ignored)!\n",
3078 				       dev_state->name);
3079 		} else {
3080 			struct btrfsic_block *const block =
3081 				&dev_state->dummy_block_for_bio_bh_flush;
3082 
3083 			block->is_iodone = 0;
3084 			block->never_written = 0;
3085 			block->iodone_w_error = 0;
3086 			block->flush_gen = dev_state->last_flush_gen + 1;
3087 			block->submit_bio_bh_rw = rw;
3088 			block->orig_bio_bh_private = bio->bi_private;
3089 			block->orig_bio_bh_end_io.bio = bio->bi_end_io;
3090 			block->next_in_same_bio = NULL;
3091 			bio->bi_private = block;
3092 			bio->bi_end_io = btrfsic_bio_end_io;
3093 		}
3094 	}
3095 leave:
3096 	mutex_unlock(&btrfsic_mutex);
3097 }
3098 
3099 void btrfsic_submit_bio(int rw, struct bio *bio)
3100 {
3101 	__btrfsic_submit_bio(rw, bio);
3102 	submit_bio(rw, bio);
3103 }
3104 
3105 int btrfsic_submit_bio_wait(int rw, struct bio *bio)
3106 {
3107 	__btrfsic_submit_bio(rw, bio);
3108 	return submit_bio_wait(rw, bio);
3109 }
3110 
3111 int btrfsic_mount(struct btrfs_root *root,
3112 		  struct btrfs_fs_devices *fs_devices,
3113 		  int including_extent_data, u32 print_mask)
3114 {
3115 	int ret;
3116 	struct btrfsic_state *state;
3117 	struct list_head *dev_head = &fs_devices->devices;
3118 	struct btrfs_device *device;
3119 
3120 	if (root->nodesize != root->leafsize) {
3121 		printk(KERN_INFO
3122 		       "btrfsic: cannot handle nodesize %d != leafsize %d!\n",
3123 		       root->nodesize, root->leafsize);
3124 		return -1;
3125 	}
3126 	if (root->nodesize & ((u64)PAGE_CACHE_SIZE - 1)) {
3127 		printk(KERN_INFO
3128 		       "btrfsic: cannot handle nodesize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3129 		       root->nodesize, PAGE_CACHE_SIZE);
3130 		return -1;
3131 	}
3132 	if (root->leafsize & ((u64)PAGE_CACHE_SIZE - 1)) {
3133 		printk(KERN_INFO
3134 		       "btrfsic: cannot handle leafsize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3135 		       root->leafsize, PAGE_CACHE_SIZE);
3136 		return -1;
3137 	}
3138 	if (root->sectorsize & ((u64)PAGE_CACHE_SIZE - 1)) {
3139 		printk(KERN_INFO
3140 		       "btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3141 		       root->sectorsize, PAGE_CACHE_SIZE);
3142 		return -1;
3143 	}
3144 	state = kzalloc(sizeof(*state), GFP_NOFS);
3145 	if (NULL == state) {
3146 		printk(KERN_INFO "btrfs check-integrity: kmalloc() failed!\n");
3147 		return -1;
3148 	}
3149 
3150 	if (!btrfsic_is_initialized) {
3151 		mutex_init(&btrfsic_mutex);
3152 		btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
3153 		btrfsic_is_initialized = 1;
3154 	}
3155 	mutex_lock(&btrfsic_mutex);
3156 	state->root = root;
3157 	state->print_mask = print_mask;
3158 	state->include_extent_data = including_extent_data;
3159 	state->csum_size = 0;
3160 	state->metablock_size = root->nodesize;
3161 	state->datablock_size = root->sectorsize;
3162 	INIT_LIST_HEAD(&state->all_blocks_list);
3163 	btrfsic_block_hashtable_init(&state->block_hashtable);
3164 	btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
3165 	state->max_superblock_generation = 0;
3166 	state->latest_superblock = NULL;
3167 
3168 	list_for_each_entry(device, dev_head, dev_list) {
3169 		struct btrfsic_dev_state *ds;
3170 		char *p;
3171 
3172 		if (!device->bdev || !device->name)
3173 			continue;
3174 
3175 		ds = btrfsic_dev_state_alloc();
3176 		if (NULL == ds) {
3177 			printk(KERN_INFO
3178 			       "btrfs check-integrity: kmalloc() failed!\n");
3179 			mutex_unlock(&btrfsic_mutex);
3180 			return -1;
3181 		}
3182 		ds->bdev = device->bdev;
3183 		ds->state = state;
3184 		bdevname(ds->bdev, ds->name);
3185 		ds->name[BDEVNAME_SIZE - 1] = '\0';
3186 		for (p = ds->name; *p != '\0'; p++);
3187 		while (p > ds->name && *p != '/')
3188 			p--;
3189 		if (*p == '/')
3190 			p++;
3191 		strlcpy(ds->name, p, sizeof(ds->name));
3192 		btrfsic_dev_state_hashtable_add(ds,
3193 						&btrfsic_dev_state_hashtable);
3194 	}
3195 
3196 	ret = btrfsic_process_superblock(state, fs_devices);
3197 	if (0 != ret) {
3198 		mutex_unlock(&btrfsic_mutex);
3199 		btrfsic_unmount(root, fs_devices);
3200 		return ret;
3201 	}
3202 
3203 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
3204 		btrfsic_dump_database(state);
3205 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
3206 		btrfsic_dump_tree(state);
3207 
3208 	mutex_unlock(&btrfsic_mutex);
3209 	return 0;
3210 }
3211 
3212 void btrfsic_unmount(struct btrfs_root *root,
3213 		     struct btrfs_fs_devices *fs_devices)
3214 {
3215 	struct list_head *elem_all;
3216 	struct list_head *tmp_all;
3217 	struct btrfsic_state *state;
3218 	struct list_head *dev_head = &fs_devices->devices;
3219 	struct btrfs_device *device;
3220 
3221 	if (!btrfsic_is_initialized)
3222 		return;
3223 
3224 	mutex_lock(&btrfsic_mutex);
3225 
3226 	state = NULL;
3227 	list_for_each_entry(device, dev_head, dev_list) {
3228 		struct btrfsic_dev_state *ds;
3229 
3230 		if (!device->bdev || !device->name)
3231 			continue;
3232 
3233 		ds = btrfsic_dev_state_hashtable_lookup(
3234 				device->bdev,
3235 				&btrfsic_dev_state_hashtable);
3236 		if (NULL != ds) {
3237 			state = ds->state;
3238 			btrfsic_dev_state_hashtable_remove(ds);
3239 			btrfsic_dev_state_free(ds);
3240 		}
3241 	}
3242 
3243 	if (NULL == state) {
3244 		printk(KERN_INFO
3245 		       "btrfsic: error, cannot find state information"
3246 		       " on umount!\n");
3247 		mutex_unlock(&btrfsic_mutex);
3248 		return;
3249 	}
3250 
3251 	/*
3252 	 * Don't care about keeping the lists' state up to date,
3253 	 * just free all memory that was allocated dynamically.
3254 	 * Free the blocks and the block_links.
3255 	 */
3256 	list_for_each_safe(elem_all, tmp_all, &state->all_blocks_list) {
3257 		struct btrfsic_block *const b_all =
3258 		    list_entry(elem_all, struct btrfsic_block,
3259 			       all_blocks_node);
3260 		struct list_head *elem_ref_to;
3261 		struct list_head *tmp_ref_to;
3262 
3263 		list_for_each_safe(elem_ref_to, tmp_ref_to,
3264 				   &b_all->ref_to_list) {
3265 			struct btrfsic_block_link *const l =
3266 			    list_entry(elem_ref_to,
3267 				       struct btrfsic_block_link,
3268 				       node_ref_to);
3269 
3270 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
3271 				btrfsic_print_rem_link(state, l);
3272 
3273 			l->ref_cnt--;
3274 			if (0 == l->ref_cnt)
3275 				btrfsic_block_link_free(l);
3276 		}
3277 
3278 		if (b_all->is_iodone || b_all->never_written)
3279 			btrfsic_block_free(b_all);
3280 		else
3281 			printk(KERN_INFO "btrfs: attempt to free %c-block"
3282 			       " @%llu (%s/%llu/%d) on umount which is"
3283 			       " not yet iodone!\n",
3284 			       btrfsic_get_block_type(state, b_all),
3285 			       b_all->logical_bytenr, b_all->dev_state->name,
3286 			       b_all->dev_bytenr, b_all->mirror_num);
3287 	}
3288 
3289 	mutex_unlock(&btrfsic_mutex);
3290 
3291 	kfree(state);
3292 }
3293