xref: /openbmc/linux/fs/btrfs/check-integrity.c (revision 15e47304)
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 a FLUSH request to the device where
41  *        these blocks are located was received and completed.
42  *    2b. All referenced blocks need to have a generation
43  *        number which is equal to the parent's number.
44  *
45  * One issue that was found using this module was that the log
46  * tree on disk became temporarily corrupted because disk blocks
47  * that had been in use for the log tree had been freed and
48  * reused too early, while being referenced by the written super
49  * block.
50  *
51  * The search term in the kernel log that can be used to filter
52  * on the existence of detected integrity issues is
53  * "btrfs: attempt".
54  *
55  * The integrity check is enabled via mount options. These
56  * mount options are only supported if the integrity check
57  * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY.
58  *
59  * Example #1, apply integrity checks to all metadata:
60  * mount /dev/sdb1 /mnt -o check_int
61  *
62  * Example #2, apply integrity checks to all metadata and
63  * to data extents:
64  * mount /dev/sdb1 /mnt -o check_int_data
65  *
66  * Example #3, apply integrity checks to all metadata and dump
67  * the tree that the super block references to kernel messages
68  * each time after a super block was written:
69  * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263
70  *
71  * If the integrity check tool is included and activated in
72  * the mount options, plenty of kernel memory is used, and
73  * plenty of additional CPU cycles are spent. Enabling this
74  * functionality is not intended for normal use. In most
75  * cases, unless you are a btrfs developer who needs to verify
76  * the integrity of (super)-block write requests, do not
77  * enable the config option BTRFS_FS_CHECK_INTEGRITY to
78  * include and compile the integrity check tool.
79  */
80 
81 #include <linux/sched.h>
82 #include <linux/slab.h>
83 #include <linux/buffer_head.h>
84 #include <linux/mutex.h>
85 #include <linux/crc32c.h>
86 #include <linux/genhd.h>
87 #include <linux/blkdev.h>
88 #include "ctree.h"
89 #include "disk-io.h"
90 #include "transaction.h"
91 #include "extent_io.h"
92 #include "volumes.h"
93 #include "print-tree.h"
94 #include "locking.h"
95 #include "check-integrity.h"
96 #include "rcu-string.h"
97 
98 #define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
99 #define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
100 #define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
101 #define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
102 #define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
103 #define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
104 #define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
105 #define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6)	/* in characters,
106 							 * excluding " [...]" */
107 #define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
108 
109 /*
110  * The definition of the bitmask fields for the print_mask.
111  * They are specified with the mount option check_integrity_print_mask.
112  */
113 #define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE			0x00000001
114 #define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION		0x00000002
115 #define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE			0x00000004
116 #define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE			0x00000008
117 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH			0x00000010
118 #define BTRFSIC_PRINT_MASK_END_IO_BIO_BH			0x00000020
119 #define BTRFSIC_PRINT_MASK_VERBOSE				0x00000040
120 #define BTRFSIC_PRINT_MASK_VERY_VERBOSE				0x00000080
121 #define BTRFSIC_PRINT_MASK_INITIAL_TREE				0x00000100
122 #define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES			0x00000200
123 #define BTRFSIC_PRINT_MASK_INITIAL_DATABASE			0x00000400
124 #define BTRFSIC_PRINT_MASK_NUM_COPIES				0x00000800
125 #define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS		0x00001000
126 
127 struct btrfsic_dev_state;
128 struct btrfsic_state;
129 
130 struct btrfsic_block {
131 	u32 magic_num;		/* only used for debug purposes */
132 	unsigned int is_metadata:1;	/* if it is meta-data, not data-data */
133 	unsigned int is_superblock:1;	/* if it is one of the superblocks */
134 	unsigned int is_iodone:1;	/* if is done by lower subsystem */
135 	unsigned int iodone_w_error:1;	/* error was indicated to endio */
136 	unsigned int never_written:1;	/* block was added because it was
137 					 * referenced, not because it was
138 					 * written */
139 	unsigned int mirror_num:2;	/* large enough to hold
140 					 * BTRFS_SUPER_MIRROR_MAX */
141 	struct btrfsic_dev_state *dev_state;
142 	u64 dev_bytenr;		/* key, physical byte num on disk */
143 	u64 logical_bytenr;	/* logical byte num on disk */
144 	u64 generation;
145 	struct btrfs_disk_key disk_key;	/* extra info to print in case of
146 					 * issues, will not always be correct */
147 	struct list_head collision_resolving_node;	/* list node */
148 	struct list_head all_blocks_node;	/* list node */
149 
150 	/* the following two lists contain block_link items */
151 	struct list_head ref_to_list;	/* list */
152 	struct list_head ref_from_list;	/* list */
153 	struct btrfsic_block *next_in_same_bio;
154 	void *orig_bio_bh_private;
155 	union {
156 		bio_end_io_t *bio;
157 		bh_end_io_t *bh;
158 	} orig_bio_bh_end_io;
159 	int submit_bio_bh_rw;
160 	u64 flush_gen; /* only valid if !never_written */
161 };
162 
163 /*
164  * Elements of this type are allocated dynamically and required because
165  * each block object can refer to and can be ref from multiple blocks.
166  * The key to lookup them in the hashtable is the dev_bytenr of
167  * the block ref to plus the one from the block refered from.
168  * The fact that they are searchable via a hashtable and that a
169  * ref_cnt is maintained is not required for the btrfs integrity
170  * check algorithm itself, it is only used to make the output more
171  * beautiful in case that an error is detected (an error is defined
172  * as a write operation to a block while that block is still referenced).
173  */
174 struct btrfsic_block_link {
175 	u32 magic_num;		/* only used for debug purposes */
176 	u32 ref_cnt;
177 	struct list_head node_ref_to;	/* list node */
178 	struct list_head node_ref_from;	/* list node */
179 	struct list_head collision_resolving_node;	/* list node */
180 	struct btrfsic_block *block_ref_to;
181 	struct btrfsic_block *block_ref_from;
182 	u64 parent_generation;
183 };
184 
185 struct btrfsic_dev_state {
186 	u32 magic_num;		/* only used for debug purposes */
187 	struct block_device *bdev;
188 	struct btrfsic_state *state;
189 	struct list_head collision_resolving_node;	/* list node */
190 	struct btrfsic_block dummy_block_for_bio_bh_flush;
191 	u64 last_flush_gen;
192 	char name[BDEVNAME_SIZE];
193 };
194 
195 struct btrfsic_block_hashtable {
196 	struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
197 };
198 
199 struct btrfsic_block_link_hashtable {
200 	struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
201 };
202 
203 struct btrfsic_dev_state_hashtable {
204 	struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
205 };
206 
207 struct btrfsic_block_data_ctx {
208 	u64 start;		/* virtual bytenr */
209 	u64 dev_bytenr;		/* physical bytenr on device */
210 	u32 len;
211 	struct btrfsic_dev_state *dev;
212 	char **datav;
213 	struct page **pagev;
214 	void *mem_to_free;
215 };
216 
217 /* This structure is used to implement recursion without occupying
218  * any stack space, refer to btrfsic_process_metablock() */
219 struct btrfsic_stack_frame {
220 	u32 magic;
221 	u32 nr;
222 	int error;
223 	int i;
224 	int limit_nesting;
225 	int num_copies;
226 	int mirror_num;
227 	struct btrfsic_block *block;
228 	struct btrfsic_block_data_ctx *block_ctx;
229 	struct btrfsic_block *next_block;
230 	struct btrfsic_block_data_ctx next_block_ctx;
231 	struct btrfs_header *hdr;
232 	struct btrfsic_stack_frame *prev;
233 };
234 
235 /* Some state per mounted filesystem */
236 struct btrfsic_state {
237 	u32 print_mask;
238 	int include_extent_data;
239 	int csum_size;
240 	struct list_head all_blocks_list;
241 	struct btrfsic_block_hashtable block_hashtable;
242 	struct btrfsic_block_link_hashtable block_link_hashtable;
243 	struct btrfs_root *root;
244 	u64 max_superblock_generation;
245 	struct btrfsic_block *latest_superblock;
246 	u32 metablock_size;
247 	u32 datablock_size;
248 };
249 
250 static void btrfsic_block_init(struct btrfsic_block *b);
251 static struct btrfsic_block *btrfsic_block_alloc(void);
252 static void btrfsic_block_free(struct btrfsic_block *b);
253 static void btrfsic_block_link_init(struct btrfsic_block_link *n);
254 static struct btrfsic_block_link *btrfsic_block_link_alloc(void);
255 static void btrfsic_block_link_free(struct btrfsic_block_link *n);
256 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds);
257 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void);
258 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds);
259 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h);
260 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
261 					struct btrfsic_block_hashtable *h);
262 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b);
263 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
264 		struct block_device *bdev,
265 		u64 dev_bytenr,
266 		struct btrfsic_block_hashtable *h);
267 static void btrfsic_block_link_hashtable_init(
268 		struct btrfsic_block_link_hashtable *h);
269 static void btrfsic_block_link_hashtable_add(
270 		struct btrfsic_block_link *l,
271 		struct btrfsic_block_link_hashtable *h);
272 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l);
273 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
274 		struct block_device *bdev_ref_to,
275 		u64 dev_bytenr_ref_to,
276 		struct block_device *bdev_ref_from,
277 		u64 dev_bytenr_ref_from,
278 		struct btrfsic_block_link_hashtable *h);
279 static void btrfsic_dev_state_hashtable_init(
280 		struct btrfsic_dev_state_hashtable *h);
281 static void btrfsic_dev_state_hashtable_add(
282 		struct btrfsic_dev_state *ds,
283 		struct btrfsic_dev_state_hashtable *h);
284 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds);
285 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(
286 		struct block_device *bdev,
287 		struct btrfsic_dev_state_hashtable *h);
288 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void);
289 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf);
290 static int btrfsic_process_superblock(struct btrfsic_state *state,
291 				      struct btrfs_fs_devices *fs_devices);
292 static int btrfsic_process_metablock(struct btrfsic_state *state,
293 				     struct btrfsic_block *block,
294 				     struct btrfsic_block_data_ctx *block_ctx,
295 				     int limit_nesting, int force_iodone_flag);
296 static void btrfsic_read_from_block_data(
297 	struct btrfsic_block_data_ctx *block_ctx,
298 	void *dst, u32 offset, size_t len);
299 static int btrfsic_create_link_to_next_block(
300 		struct btrfsic_state *state,
301 		struct btrfsic_block *block,
302 		struct btrfsic_block_data_ctx
303 		*block_ctx, u64 next_bytenr,
304 		int limit_nesting,
305 		struct btrfsic_block_data_ctx *next_block_ctx,
306 		struct btrfsic_block **next_blockp,
307 		int force_iodone_flag,
308 		int *num_copiesp, int *mirror_nump,
309 		struct btrfs_disk_key *disk_key,
310 		u64 parent_generation);
311 static int btrfsic_handle_extent_data(struct btrfsic_state *state,
312 				      struct btrfsic_block *block,
313 				      struct btrfsic_block_data_ctx *block_ctx,
314 				      u32 item_offset, int force_iodone_flag);
315 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
316 			     struct btrfsic_block_data_ctx *block_ctx_out,
317 			     int mirror_num);
318 static int btrfsic_map_superblock(struct btrfsic_state *state, u64 bytenr,
319 				  u32 len, struct block_device *bdev,
320 				  struct btrfsic_block_data_ctx *block_ctx_out);
321 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
322 static int btrfsic_read_block(struct btrfsic_state *state,
323 			      struct btrfsic_block_data_ctx *block_ctx);
324 static void btrfsic_dump_database(struct btrfsic_state *state);
325 static void btrfsic_complete_bio_end_io(struct bio *bio, int err);
326 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
327 				     char **datav, unsigned int num_pages);
328 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
329 					  u64 dev_bytenr, char **mapped_datav,
330 					  unsigned int num_pages,
331 					  struct bio *bio, int *bio_is_patched,
332 					  struct buffer_head *bh,
333 					  int submit_bio_bh_rw);
334 static int btrfsic_process_written_superblock(
335 		struct btrfsic_state *state,
336 		struct btrfsic_block *const block,
337 		struct btrfs_super_block *const super_hdr);
338 static void btrfsic_bio_end_io(struct bio *bp, int bio_error_status);
339 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate);
340 static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
341 					      const struct btrfsic_block *block,
342 					      int recursion_level);
343 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
344 					struct btrfsic_block *const block,
345 					int recursion_level);
346 static void btrfsic_print_add_link(const struct btrfsic_state *state,
347 				   const struct btrfsic_block_link *l);
348 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
349 				   const struct btrfsic_block_link *l);
350 static char btrfsic_get_block_type(const struct btrfsic_state *state,
351 				   const struct btrfsic_block *block);
352 static void btrfsic_dump_tree(const struct btrfsic_state *state);
353 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
354 				  const struct btrfsic_block *block,
355 				  int indent_level);
356 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
357 		struct btrfsic_state *state,
358 		struct btrfsic_block_data_ctx *next_block_ctx,
359 		struct btrfsic_block *next_block,
360 		struct btrfsic_block *from_block,
361 		u64 parent_generation);
362 static struct btrfsic_block *btrfsic_block_lookup_or_add(
363 		struct btrfsic_state *state,
364 		struct btrfsic_block_data_ctx *block_ctx,
365 		const char *additional_string,
366 		int is_metadata,
367 		int is_iodone,
368 		int never_written,
369 		int mirror_num,
370 		int *was_created);
371 static int btrfsic_process_superblock_dev_mirror(
372 		struct btrfsic_state *state,
373 		struct btrfsic_dev_state *dev_state,
374 		struct btrfs_device *device,
375 		int superblock_mirror_num,
376 		struct btrfsic_dev_state **selected_dev_state,
377 		struct btrfs_super_block *selected_super);
378 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
379 		struct block_device *bdev);
380 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
381 					   u64 bytenr,
382 					   struct btrfsic_dev_state *dev_state,
383 					   u64 dev_bytenr);
384 
385 static struct mutex btrfsic_mutex;
386 static int btrfsic_is_initialized;
387 static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
388 
389 
390 static void btrfsic_block_init(struct btrfsic_block *b)
391 {
392 	b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
393 	b->dev_state = NULL;
394 	b->dev_bytenr = 0;
395 	b->logical_bytenr = 0;
396 	b->generation = BTRFSIC_GENERATION_UNKNOWN;
397 	b->disk_key.objectid = 0;
398 	b->disk_key.type = 0;
399 	b->disk_key.offset = 0;
400 	b->is_metadata = 0;
401 	b->is_superblock = 0;
402 	b->is_iodone = 0;
403 	b->iodone_w_error = 0;
404 	b->never_written = 0;
405 	b->mirror_num = 0;
406 	b->next_in_same_bio = NULL;
407 	b->orig_bio_bh_private = NULL;
408 	b->orig_bio_bh_end_io.bio = NULL;
409 	INIT_LIST_HEAD(&b->collision_resolving_node);
410 	INIT_LIST_HEAD(&b->all_blocks_node);
411 	INIT_LIST_HEAD(&b->ref_to_list);
412 	INIT_LIST_HEAD(&b->ref_from_list);
413 	b->submit_bio_bh_rw = 0;
414 	b->flush_gen = 0;
415 }
416 
417 static struct btrfsic_block *btrfsic_block_alloc(void)
418 {
419 	struct btrfsic_block *b;
420 
421 	b = kzalloc(sizeof(*b), GFP_NOFS);
422 	if (NULL != b)
423 		btrfsic_block_init(b);
424 
425 	return b;
426 }
427 
428 static void btrfsic_block_free(struct btrfsic_block *b)
429 {
430 	BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
431 	kfree(b);
432 }
433 
434 static void btrfsic_block_link_init(struct btrfsic_block_link *l)
435 {
436 	l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
437 	l->ref_cnt = 1;
438 	INIT_LIST_HEAD(&l->node_ref_to);
439 	INIT_LIST_HEAD(&l->node_ref_from);
440 	INIT_LIST_HEAD(&l->collision_resolving_node);
441 	l->block_ref_to = NULL;
442 	l->block_ref_from = NULL;
443 }
444 
445 static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
446 {
447 	struct btrfsic_block_link *l;
448 
449 	l = kzalloc(sizeof(*l), GFP_NOFS);
450 	if (NULL != l)
451 		btrfsic_block_link_init(l);
452 
453 	return l;
454 }
455 
456 static void btrfsic_block_link_free(struct btrfsic_block_link *l)
457 {
458 	BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
459 	kfree(l);
460 }
461 
462 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
463 {
464 	ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
465 	ds->bdev = NULL;
466 	ds->state = NULL;
467 	ds->name[0] = '\0';
468 	INIT_LIST_HEAD(&ds->collision_resolving_node);
469 	ds->last_flush_gen = 0;
470 	btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
471 	ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
472 	ds->dummy_block_for_bio_bh_flush.dev_state = ds;
473 }
474 
475 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
476 {
477 	struct btrfsic_dev_state *ds;
478 
479 	ds = kzalloc(sizeof(*ds), GFP_NOFS);
480 	if (NULL != ds)
481 		btrfsic_dev_state_init(ds);
482 
483 	return ds;
484 }
485 
486 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
487 {
488 	BUG_ON(!(NULL == ds ||
489 		 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
490 	kfree(ds);
491 }
492 
493 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
494 {
495 	int i;
496 
497 	for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
498 		INIT_LIST_HEAD(h->table + i);
499 }
500 
501 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
502 					struct btrfsic_block_hashtable *h)
503 {
504 	const unsigned int hashval =
505 	    (((unsigned int)(b->dev_bytenr >> 16)) ^
506 	     ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
507 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
508 
509 	list_add(&b->collision_resolving_node, h->table + hashval);
510 }
511 
512 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
513 {
514 	list_del(&b->collision_resolving_node);
515 }
516 
517 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
518 		struct block_device *bdev,
519 		u64 dev_bytenr,
520 		struct btrfsic_block_hashtable *h)
521 {
522 	const unsigned int hashval =
523 	    (((unsigned int)(dev_bytenr >> 16)) ^
524 	     ((unsigned int)((uintptr_t)bdev))) &
525 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
526 	struct list_head *elem;
527 
528 	list_for_each(elem, h->table + hashval) {
529 		struct btrfsic_block *const b =
530 		    list_entry(elem, struct btrfsic_block,
531 			       collision_resolving_node);
532 
533 		if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
534 			return b;
535 	}
536 
537 	return NULL;
538 }
539 
540 static void btrfsic_block_link_hashtable_init(
541 		struct btrfsic_block_link_hashtable *h)
542 {
543 	int i;
544 
545 	for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
546 		INIT_LIST_HEAD(h->table + i);
547 }
548 
549 static void btrfsic_block_link_hashtable_add(
550 		struct btrfsic_block_link *l,
551 		struct btrfsic_block_link_hashtable *h)
552 {
553 	const unsigned int hashval =
554 	    (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
555 	     ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
556 	     ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
557 	     ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
558 	     & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
559 
560 	BUG_ON(NULL == l->block_ref_to);
561 	BUG_ON(NULL == l->block_ref_from);
562 	list_add(&l->collision_resolving_node, h->table + hashval);
563 }
564 
565 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
566 {
567 	list_del(&l->collision_resolving_node);
568 }
569 
570 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
571 		struct block_device *bdev_ref_to,
572 		u64 dev_bytenr_ref_to,
573 		struct block_device *bdev_ref_from,
574 		u64 dev_bytenr_ref_from,
575 		struct btrfsic_block_link_hashtable *h)
576 {
577 	const unsigned int hashval =
578 	    (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
579 	     ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
580 	     ((unsigned int)((uintptr_t)bdev_ref_to)) ^
581 	     ((unsigned int)((uintptr_t)bdev_ref_from))) &
582 	     (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
583 	struct list_head *elem;
584 
585 	list_for_each(elem, h->table + hashval) {
586 		struct btrfsic_block_link *const l =
587 		    list_entry(elem, struct btrfsic_block_link,
588 			       collision_resolving_node);
589 
590 		BUG_ON(NULL == l->block_ref_to);
591 		BUG_ON(NULL == l->block_ref_from);
592 		if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
593 		    l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
594 		    l->block_ref_from->dev_state->bdev == bdev_ref_from &&
595 		    l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
596 			return l;
597 	}
598 
599 	return NULL;
600 }
601 
602 static void btrfsic_dev_state_hashtable_init(
603 		struct btrfsic_dev_state_hashtable *h)
604 {
605 	int i;
606 
607 	for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
608 		INIT_LIST_HEAD(h->table + i);
609 }
610 
611 static void btrfsic_dev_state_hashtable_add(
612 		struct btrfsic_dev_state *ds,
613 		struct btrfsic_dev_state_hashtable *h)
614 {
615 	const unsigned int hashval =
616 	    (((unsigned int)((uintptr_t)ds->bdev)) &
617 	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
618 
619 	list_add(&ds->collision_resolving_node, h->table + hashval);
620 }
621 
622 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
623 {
624 	list_del(&ds->collision_resolving_node);
625 }
626 
627 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(
628 		struct block_device *bdev,
629 		struct btrfsic_dev_state_hashtable *h)
630 {
631 	const unsigned int hashval =
632 	    (((unsigned int)((uintptr_t)bdev)) &
633 	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
634 	struct list_head *elem;
635 
636 	list_for_each(elem, h->table + hashval) {
637 		struct btrfsic_dev_state *const ds =
638 		    list_entry(elem, struct btrfsic_dev_state,
639 			       collision_resolving_node);
640 
641 		if (ds->bdev == bdev)
642 			return ds;
643 	}
644 
645 	return NULL;
646 }
647 
648 static int btrfsic_process_superblock(struct btrfsic_state *state,
649 				      struct btrfs_fs_devices *fs_devices)
650 {
651 	int ret = 0;
652 	struct btrfs_super_block *selected_super;
653 	struct list_head *dev_head = &fs_devices->devices;
654 	struct btrfs_device *device;
655 	struct btrfsic_dev_state *selected_dev_state = NULL;
656 	int pass;
657 
658 	BUG_ON(NULL == state);
659 	selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
660 	if (NULL == selected_super) {
661 		printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
662 		return -1;
663 	}
664 
665 	list_for_each_entry(device, dev_head, dev_list) {
666 		int i;
667 		struct btrfsic_dev_state *dev_state;
668 
669 		if (!device->bdev || !device->name)
670 			continue;
671 
672 		dev_state = btrfsic_dev_state_lookup(device->bdev);
673 		BUG_ON(NULL == dev_state);
674 		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
675 			ret = btrfsic_process_superblock_dev_mirror(
676 					state, dev_state, device, i,
677 					&selected_dev_state, selected_super);
678 			if (0 != ret && 0 == i) {
679 				kfree(selected_super);
680 				return ret;
681 			}
682 		}
683 	}
684 
685 	if (NULL == state->latest_superblock) {
686 		printk(KERN_INFO "btrfsic: no superblock found!\n");
687 		kfree(selected_super);
688 		return -1;
689 	}
690 
691 	state->csum_size = btrfs_super_csum_size(selected_super);
692 
693 	for (pass = 0; pass < 3; pass++) {
694 		int num_copies;
695 		int mirror_num;
696 		u64 next_bytenr;
697 
698 		switch (pass) {
699 		case 0:
700 			next_bytenr = btrfs_super_root(selected_super);
701 			if (state->print_mask &
702 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
703 				printk(KERN_INFO "root@%llu\n",
704 				       (unsigned long long)next_bytenr);
705 			break;
706 		case 1:
707 			next_bytenr = btrfs_super_chunk_root(selected_super);
708 			if (state->print_mask &
709 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
710 				printk(KERN_INFO "chunk@%llu\n",
711 				       (unsigned long long)next_bytenr);
712 			break;
713 		case 2:
714 			next_bytenr = btrfs_super_log_root(selected_super);
715 			if (0 == next_bytenr)
716 				continue;
717 			if (state->print_mask &
718 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
719 				printk(KERN_INFO "log@%llu\n",
720 				       (unsigned long long)next_bytenr);
721 			break;
722 		}
723 
724 		num_copies =
725 		    btrfs_num_copies(&state->root->fs_info->mapping_tree,
726 				     next_bytenr, state->metablock_size);
727 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
728 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
729 			       (unsigned long long)next_bytenr, num_copies);
730 
731 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
732 			struct btrfsic_block *next_block;
733 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
734 			struct btrfsic_block_link *l;
735 
736 			ret = btrfsic_map_block(state, next_bytenr,
737 						state->metablock_size,
738 						&tmp_next_block_ctx,
739 						mirror_num);
740 			if (ret) {
741 				printk(KERN_INFO "btrfsic:"
742 				       " btrfsic_map_block(root @%llu,"
743 				       " mirror %d) failed!\n",
744 				       (unsigned long long)next_bytenr,
745 				       mirror_num);
746 				kfree(selected_super);
747 				return -1;
748 			}
749 
750 			next_block = btrfsic_block_hashtable_lookup(
751 					tmp_next_block_ctx.dev->bdev,
752 					tmp_next_block_ctx.dev_bytenr,
753 					&state->block_hashtable);
754 			BUG_ON(NULL == next_block);
755 
756 			l = btrfsic_block_link_hashtable_lookup(
757 					tmp_next_block_ctx.dev->bdev,
758 					tmp_next_block_ctx.dev_bytenr,
759 					state->latest_superblock->dev_state->
760 					bdev,
761 					state->latest_superblock->dev_bytenr,
762 					&state->block_link_hashtable);
763 			BUG_ON(NULL == l);
764 
765 			ret = btrfsic_read_block(state, &tmp_next_block_ctx);
766 			if (ret < (int)PAGE_CACHE_SIZE) {
767 				printk(KERN_INFO
768 				       "btrfsic: read @logical %llu failed!\n",
769 				       (unsigned long long)
770 				       tmp_next_block_ctx.start);
771 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
772 				kfree(selected_super);
773 				return -1;
774 			}
775 
776 			ret = btrfsic_process_metablock(state,
777 							next_block,
778 							&tmp_next_block_ctx,
779 							BTRFS_MAX_LEVEL + 3, 1);
780 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
781 		}
782 	}
783 
784 	kfree(selected_super);
785 	return ret;
786 }
787 
788 static int btrfsic_process_superblock_dev_mirror(
789 		struct btrfsic_state *state,
790 		struct btrfsic_dev_state *dev_state,
791 		struct btrfs_device *device,
792 		int superblock_mirror_num,
793 		struct btrfsic_dev_state **selected_dev_state,
794 		struct btrfs_super_block *selected_super)
795 {
796 	struct btrfs_super_block *super_tmp;
797 	u64 dev_bytenr;
798 	struct buffer_head *bh;
799 	struct btrfsic_block *superblock_tmp;
800 	int pass;
801 	struct block_device *const superblock_bdev = device->bdev;
802 
803 	/* super block bytenr is always the unmapped device bytenr */
804 	dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
805 	if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->total_bytes)
806 		return -1;
807 	bh = __bread(superblock_bdev, dev_bytenr / 4096,
808 		     BTRFS_SUPER_INFO_SIZE);
809 	if (NULL == bh)
810 		return -1;
811 	super_tmp = (struct btrfs_super_block *)
812 	    (bh->b_data + (dev_bytenr & 4095));
813 
814 	if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
815 	    strncmp((char *)(&(super_tmp->magic)), BTRFS_MAGIC,
816 		    sizeof(super_tmp->magic)) ||
817 	    memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
818 	    btrfs_super_nodesize(super_tmp) != state->metablock_size ||
819 	    btrfs_super_leafsize(super_tmp) != state->metablock_size ||
820 	    btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
821 		brelse(bh);
822 		return 0;
823 	}
824 
825 	superblock_tmp =
826 	    btrfsic_block_hashtable_lookup(superblock_bdev,
827 					   dev_bytenr,
828 					   &state->block_hashtable);
829 	if (NULL == superblock_tmp) {
830 		superblock_tmp = btrfsic_block_alloc();
831 		if (NULL == superblock_tmp) {
832 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
833 			brelse(bh);
834 			return -1;
835 		}
836 		/* for superblock, only the dev_bytenr makes sense */
837 		superblock_tmp->dev_bytenr = dev_bytenr;
838 		superblock_tmp->dev_state = dev_state;
839 		superblock_tmp->logical_bytenr = dev_bytenr;
840 		superblock_tmp->generation = btrfs_super_generation(super_tmp);
841 		superblock_tmp->is_metadata = 1;
842 		superblock_tmp->is_superblock = 1;
843 		superblock_tmp->is_iodone = 1;
844 		superblock_tmp->never_written = 0;
845 		superblock_tmp->mirror_num = 1 + superblock_mirror_num;
846 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
847 			printk_in_rcu(KERN_INFO "New initial S-block (bdev %p, %s)"
848 				     " @%llu (%s/%llu/%d)\n",
849 				     superblock_bdev,
850 				     rcu_str_deref(device->name),
851 				     (unsigned long long)dev_bytenr,
852 				     dev_state->name,
853 				     (unsigned long long)dev_bytenr,
854 				     superblock_mirror_num);
855 		list_add(&superblock_tmp->all_blocks_node,
856 			 &state->all_blocks_list);
857 		btrfsic_block_hashtable_add(superblock_tmp,
858 					    &state->block_hashtable);
859 	}
860 
861 	/* select the one with the highest generation field */
862 	if (btrfs_super_generation(super_tmp) >
863 	    state->max_superblock_generation ||
864 	    0 == state->max_superblock_generation) {
865 		memcpy(selected_super, super_tmp, sizeof(*selected_super));
866 		*selected_dev_state = dev_state;
867 		state->max_superblock_generation =
868 		    btrfs_super_generation(super_tmp);
869 		state->latest_superblock = superblock_tmp;
870 	}
871 
872 	for (pass = 0; pass < 3; pass++) {
873 		u64 next_bytenr;
874 		int num_copies;
875 		int mirror_num;
876 		const char *additional_string = NULL;
877 		struct btrfs_disk_key tmp_disk_key;
878 
879 		tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
880 		tmp_disk_key.offset = 0;
881 		switch (pass) {
882 		case 0:
883 			tmp_disk_key.objectid =
884 			    cpu_to_le64(BTRFS_ROOT_TREE_OBJECTID);
885 			additional_string = "initial root ";
886 			next_bytenr = btrfs_super_root(super_tmp);
887 			break;
888 		case 1:
889 			tmp_disk_key.objectid =
890 			    cpu_to_le64(BTRFS_CHUNK_TREE_OBJECTID);
891 			additional_string = "initial chunk ";
892 			next_bytenr = btrfs_super_chunk_root(super_tmp);
893 			break;
894 		case 2:
895 			tmp_disk_key.objectid =
896 			    cpu_to_le64(BTRFS_TREE_LOG_OBJECTID);
897 			additional_string = "initial log ";
898 			next_bytenr = btrfs_super_log_root(super_tmp);
899 			if (0 == next_bytenr)
900 				continue;
901 			break;
902 		}
903 
904 		num_copies =
905 		    btrfs_num_copies(&state->root->fs_info->mapping_tree,
906 				     next_bytenr, state->metablock_size);
907 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
908 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
909 			       (unsigned long long)next_bytenr, num_copies);
910 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
911 			struct btrfsic_block *next_block;
912 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
913 			struct btrfsic_block_link *l;
914 
915 			if (btrfsic_map_block(state, next_bytenr,
916 					      state->metablock_size,
917 					      &tmp_next_block_ctx,
918 					      mirror_num)) {
919 				printk(KERN_INFO "btrfsic: btrfsic_map_block("
920 				       "bytenr @%llu, mirror %d) failed!\n",
921 				       (unsigned long long)next_bytenr,
922 				       mirror_num);
923 				brelse(bh);
924 				return -1;
925 			}
926 
927 			next_block = btrfsic_block_lookup_or_add(
928 					state, &tmp_next_block_ctx,
929 					additional_string, 1, 1, 0,
930 					mirror_num, NULL);
931 			if (NULL == next_block) {
932 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
933 				brelse(bh);
934 				return -1;
935 			}
936 
937 			next_block->disk_key = tmp_disk_key;
938 			next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
939 			l = btrfsic_block_link_lookup_or_add(
940 					state, &tmp_next_block_ctx,
941 					next_block, superblock_tmp,
942 					BTRFSIC_GENERATION_UNKNOWN);
943 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
944 			if (NULL == l) {
945 				brelse(bh);
946 				return -1;
947 			}
948 		}
949 	}
950 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
951 		btrfsic_dump_tree_sub(state, superblock_tmp, 0);
952 
953 	brelse(bh);
954 	return 0;
955 }
956 
957 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
958 {
959 	struct btrfsic_stack_frame *sf;
960 
961 	sf = kzalloc(sizeof(*sf), GFP_NOFS);
962 	if (NULL == sf)
963 		printk(KERN_INFO "btrfsic: alloc memory failed!\n");
964 	else
965 		sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
966 	return sf;
967 }
968 
969 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
970 {
971 	BUG_ON(!(NULL == sf ||
972 		 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
973 	kfree(sf);
974 }
975 
976 static int btrfsic_process_metablock(
977 		struct btrfsic_state *state,
978 		struct btrfsic_block *const first_block,
979 		struct btrfsic_block_data_ctx *const first_block_ctx,
980 		int first_limit_nesting, int force_iodone_flag)
981 {
982 	struct btrfsic_stack_frame initial_stack_frame = { 0 };
983 	struct btrfsic_stack_frame *sf;
984 	struct btrfsic_stack_frame *next_stack;
985 	struct btrfs_header *const first_hdr =
986 		(struct btrfs_header *)first_block_ctx->datav[0];
987 
988 	BUG_ON(!first_hdr);
989 	sf = &initial_stack_frame;
990 	sf->error = 0;
991 	sf->i = -1;
992 	sf->limit_nesting = first_limit_nesting;
993 	sf->block = first_block;
994 	sf->block_ctx = first_block_ctx;
995 	sf->next_block = NULL;
996 	sf->hdr = first_hdr;
997 	sf->prev = NULL;
998 
999 continue_with_new_stack_frame:
1000 	sf->block->generation = le64_to_cpu(sf->hdr->generation);
1001 	if (0 == sf->hdr->level) {
1002 		struct btrfs_leaf *const leafhdr =
1003 		    (struct btrfs_leaf *)sf->hdr;
1004 
1005 		if (-1 == sf->i) {
1006 			sf->nr = le32_to_cpu(leafhdr->header.nritems);
1007 
1008 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1009 				printk(KERN_INFO
1010 				       "leaf %llu items %d generation %llu"
1011 				       " owner %llu\n",
1012 				       (unsigned long long)
1013 				       sf->block_ctx->start,
1014 				       sf->nr,
1015 				       (unsigned long long)
1016 				       le64_to_cpu(leafhdr->header.generation),
1017 				       (unsigned long long)
1018 				       le64_to_cpu(leafhdr->header.owner));
1019 		}
1020 
1021 continue_with_current_leaf_stack_frame:
1022 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1023 			sf->i++;
1024 			sf->num_copies = 0;
1025 		}
1026 
1027 		if (sf->i < sf->nr) {
1028 			struct btrfs_item disk_item;
1029 			u32 disk_item_offset =
1030 				(uintptr_t)(leafhdr->items + sf->i) -
1031 				(uintptr_t)leafhdr;
1032 			struct btrfs_disk_key *disk_key;
1033 			u8 type;
1034 			u32 item_offset;
1035 			u32 item_size;
1036 
1037 			if (disk_item_offset + sizeof(struct btrfs_item) >
1038 			    sf->block_ctx->len) {
1039 leaf_item_out_of_bounce_error:
1040 				printk(KERN_INFO
1041 				       "btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
1042 				       sf->block_ctx->start,
1043 				       sf->block_ctx->dev->name);
1044 				goto one_stack_frame_backwards;
1045 			}
1046 			btrfsic_read_from_block_data(sf->block_ctx,
1047 						     &disk_item,
1048 						     disk_item_offset,
1049 						     sizeof(struct btrfs_item));
1050 			item_offset = le32_to_cpu(disk_item.offset);
1051 			item_size = le32_to_cpu(disk_item.size);
1052 			disk_key = &disk_item.key;
1053 			type = disk_key->type;
1054 
1055 			if (BTRFS_ROOT_ITEM_KEY == type) {
1056 				struct btrfs_root_item root_item;
1057 				u32 root_item_offset;
1058 				u64 next_bytenr;
1059 
1060 				root_item_offset = item_offset +
1061 					offsetof(struct btrfs_leaf, items);
1062 				if (root_item_offset + item_size >
1063 				    sf->block_ctx->len)
1064 					goto leaf_item_out_of_bounce_error;
1065 				btrfsic_read_from_block_data(
1066 					sf->block_ctx, &root_item,
1067 					root_item_offset,
1068 					item_size);
1069 				next_bytenr = le64_to_cpu(root_item.bytenr);
1070 
1071 				sf->error =
1072 				    btrfsic_create_link_to_next_block(
1073 						state,
1074 						sf->block,
1075 						sf->block_ctx,
1076 						next_bytenr,
1077 						sf->limit_nesting,
1078 						&sf->next_block_ctx,
1079 						&sf->next_block,
1080 						force_iodone_flag,
1081 						&sf->num_copies,
1082 						&sf->mirror_num,
1083 						disk_key,
1084 						le64_to_cpu(root_item.
1085 						generation));
1086 				if (sf->error)
1087 					goto one_stack_frame_backwards;
1088 
1089 				if (NULL != sf->next_block) {
1090 					struct btrfs_header *const next_hdr =
1091 					    (struct btrfs_header *)
1092 					    sf->next_block_ctx.datav[0];
1093 
1094 					next_stack =
1095 					    btrfsic_stack_frame_alloc();
1096 					if (NULL == next_stack) {
1097 						btrfsic_release_block_ctx(
1098 								&sf->
1099 								next_block_ctx);
1100 						goto one_stack_frame_backwards;
1101 					}
1102 
1103 					next_stack->i = -1;
1104 					next_stack->block = sf->next_block;
1105 					next_stack->block_ctx =
1106 					    &sf->next_block_ctx;
1107 					next_stack->next_block = NULL;
1108 					next_stack->hdr = next_hdr;
1109 					next_stack->limit_nesting =
1110 					    sf->limit_nesting - 1;
1111 					next_stack->prev = sf;
1112 					sf = next_stack;
1113 					goto continue_with_new_stack_frame;
1114 				}
1115 			} else if (BTRFS_EXTENT_DATA_KEY == type &&
1116 				   state->include_extent_data) {
1117 				sf->error = btrfsic_handle_extent_data(
1118 						state,
1119 						sf->block,
1120 						sf->block_ctx,
1121 						item_offset,
1122 						force_iodone_flag);
1123 				if (sf->error)
1124 					goto one_stack_frame_backwards;
1125 			}
1126 
1127 			goto continue_with_current_leaf_stack_frame;
1128 		}
1129 	} else {
1130 		struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1131 
1132 		if (-1 == sf->i) {
1133 			sf->nr = le32_to_cpu(nodehdr->header.nritems);
1134 
1135 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1136 				printk(KERN_INFO "node %llu level %d items %d"
1137 				       " generation %llu owner %llu\n",
1138 				       (unsigned long long)
1139 				       sf->block_ctx->start,
1140 				       nodehdr->header.level, sf->nr,
1141 				       (unsigned long long)
1142 				       le64_to_cpu(nodehdr->header.generation),
1143 				       (unsigned long long)
1144 				       le64_to_cpu(nodehdr->header.owner));
1145 		}
1146 
1147 continue_with_current_node_stack_frame:
1148 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1149 			sf->i++;
1150 			sf->num_copies = 0;
1151 		}
1152 
1153 		if (sf->i < sf->nr) {
1154 			struct btrfs_key_ptr key_ptr;
1155 			u32 key_ptr_offset;
1156 			u64 next_bytenr;
1157 
1158 			key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1159 					  (uintptr_t)nodehdr;
1160 			if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1161 			    sf->block_ctx->len) {
1162 				printk(KERN_INFO
1163 				       "btrfsic: node item out of bounce at logical %llu, dev %s\n",
1164 				       sf->block_ctx->start,
1165 				       sf->block_ctx->dev->name);
1166 				goto one_stack_frame_backwards;
1167 			}
1168 			btrfsic_read_from_block_data(
1169 				sf->block_ctx, &key_ptr, key_ptr_offset,
1170 				sizeof(struct btrfs_key_ptr));
1171 			next_bytenr = le64_to_cpu(key_ptr.blockptr);
1172 
1173 			sf->error = btrfsic_create_link_to_next_block(
1174 					state,
1175 					sf->block,
1176 					sf->block_ctx,
1177 					next_bytenr,
1178 					sf->limit_nesting,
1179 					&sf->next_block_ctx,
1180 					&sf->next_block,
1181 					force_iodone_flag,
1182 					&sf->num_copies,
1183 					&sf->mirror_num,
1184 					&key_ptr.key,
1185 					le64_to_cpu(key_ptr.generation));
1186 			if (sf->error)
1187 				goto one_stack_frame_backwards;
1188 
1189 			if (NULL != sf->next_block) {
1190 				struct btrfs_header *const next_hdr =
1191 				    (struct btrfs_header *)
1192 				    sf->next_block_ctx.datav[0];
1193 
1194 				next_stack = btrfsic_stack_frame_alloc();
1195 				if (NULL == next_stack)
1196 					goto one_stack_frame_backwards;
1197 
1198 				next_stack->i = -1;
1199 				next_stack->block = sf->next_block;
1200 				next_stack->block_ctx = &sf->next_block_ctx;
1201 				next_stack->next_block = NULL;
1202 				next_stack->hdr = next_hdr;
1203 				next_stack->limit_nesting =
1204 				    sf->limit_nesting - 1;
1205 				next_stack->prev = sf;
1206 				sf = next_stack;
1207 				goto continue_with_new_stack_frame;
1208 			}
1209 
1210 			goto continue_with_current_node_stack_frame;
1211 		}
1212 	}
1213 
1214 one_stack_frame_backwards:
1215 	if (NULL != sf->prev) {
1216 		struct btrfsic_stack_frame *const prev = sf->prev;
1217 
1218 		/* the one for the initial block is freed in the caller */
1219 		btrfsic_release_block_ctx(sf->block_ctx);
1220 
1221 		if (sf->error) {
1222 			prev->error = sf->error;
1223 			btrfsic_stack_frame_free(sf);
1224 			sf = prev;
1225 			goto one_stack_frame_backwards;
1226 		}
1227 
1228 		btrfsic_stack_frame_free(sf);
1229 		sf = prev;
1230 		goto continue_with_new_stack_frame;
1231 	} else {
1232 		BUG_ON(&initial_stack_frame != sf);
1233 	}
1234 
1235 	return sf->error;
1236 }
1237 
1238 static void btrfsic_read_from_block_data(
1239 	struct btrfsic_block_data_ctx *block_ctx,
1240 	void *dstv, u32 offset, size_t len)
1241 {
1242 	size_t cur;
1243 	size_t offset_in_page;
1244 	char *kaddr;
1245 	char *dst = (char *)dstv;
1246 	size_t start_offset = block_ctx->start & ((u64)PAGE_CACHE_SIZE - 1);
1247 	unsigned long i = (start_offset + offset) >> PAGE_CACHE_SHIFT;
1248 
1249 	WARN_ON(offset + len > block_ctx->len);
1250 	offset_in_page = (start_offset + offset) &
1251 			 ((unsigned long)PAGE_CACHE_SIZE - 1);
1252 
1253 	while (len > 0) {
1254 		cur = min(len, ((size_t)PAGE_CACHE_SIZE - offset_in_page));
1255 		BUG_ON(i >= (block_ctx->len + PAGE_CACHE_SIZE - 1) >>
1256 			    PAGE_CACHE_SHIFT);
1257 		kaddr = block_ctx->datav[i];
1258 		memcpy(dst, kaddr + offset_in_page, cur);
1259 
1260 		dst += cur;
1261 		len -= cur;
1262 		offset_in_page = 0;
1263 		i++;
1264 	}
1265 }
1266 
1267 static int btrfsic_create_link_to_next_block(
1268 		struct btrfsic_state *state,
1269 		struct btrfsic_block *block,
1270 		struct btrfsic_block_data_ctx *block_ctx,
1271 		u64 next_bytenr,
1272 		int limit_nesting,
1273 		struct btrfsic_block_data_ctx *next_block_ctx,
1274 		struct btrfsic_block **next_blockp,
1275 		int force_iodone_flag,
1276 		int *num_copiesp, int *mirror_nump,
1277 		struct btrfs_disk_key *disk_key,
1278 		u64 parent_generation)
1279 {
1280 	struct btrfsic_block *next_block = NULL;
1281 	int ret;
1282 	struct btrfsic_block_link *l;
1283 	int did_alloc_block_link;
1284 	int block_was_created;
1285 
1286 	*next_blockp = NULL;
1287 	if (0 == *num_copiesp) {
1288 		*num_copiesp =
1289 		    btrfs_num_copies(&state->root->fs_info->mapping_tree,
1290 				     next_bytenr, state->metablock_size);
1291 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1292 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
1293 			       (unsigned long long)next_bytenr, *num_copiesp);
1294 		*mirror_nump = 1;
1295 	}
1296 
1297 	if (*mirror_nump > *num_copiesp)
1298 		return 0;
1299 
1300 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1301 		printk(KERN_INFO
1302 		       "btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1303 		       *mirror_nump);
1304 	ret = btrfsic_map_block(state, next_bytenr,
1305 				state->metablock_size,
1306 				next_block_ctx, *mirror_nump);
1307 	if (ret) {
1308 		printk(KERN_INFO
1309 		       "btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1310 		       (unsigned long long)next_bytenr, *mirror_nump);
1311 		btrfsic_release_block_ctx(next_block_ctx);
1312 		*next_blockp = NULL;
1313 		return -1;
1314 	}
1315 
1316 	next_block = btrfsic_block_lookup_or_add(state,
1317 						 next_block_ctx, "referenced ",
1318 						 1, force_iodone_flag,
1319 						 !force_iodone_flag,
1320 						 *mirror_nump,
1321 						 &block_was_created);
1322 	if (NULL == next_block) {
1323 		btrfsic_release_block_ctx(next_block_ctx);
1324 		*next_blockp = NULL;
1325 		return -1;
1326 	}
1327 	if (block_was_created) {
1328 		l = NULL;
1329 		next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1330 	} else {
1331 		if (next_block->logical_bytenr != next_bytenr &&
1332 		    !(!next_block->is_metadata &&
1333 		      0 == next_block->logical_bytenr)) {
1334 			printk(KERN_INFO
1335 			       "Referenced block @%llu (%s/%llu/%d)"
1336 			       " found in hash table, %c,"
1337 			       " bytenr mismatch (!= stored %llu).\n",
1338 			       (unsigned long long)next_bytenr,
1339 			       next_block_ctx->dev->name,
1340 			       (unsigned long long)next_block_ctx->dev_bytenr,
1341 			       *mirror_nump,
1342 			       btrfsic_get_block_type(state, next_block),
1343 			       (unsigned long long)next_block->logical_bytenr);
1344 		} else if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1345 			printk(KERN_INFO
1346 			       "Referenced block @%llu (%s/%llu/%d)"
1347 			       " found in hash table, %c.\n",
1348 			       (unsigned long long)next_bytenr,
1349 			       next_block_ctx->dev->name,
1350 			       (unsigned long long)next_block_ctx->dev_bytenr,
1351 			       *mirror_nump,
1352 			       btrfsic_get_block_type(state, next_block));
1353 		next_block->logical_bytenr = next_bytenr;
1354 
1355 		next_block->mirror_num = *mirror_nump;
1356 		l = btrfsic_block_link_hashtable_lookup(
1357 				next_block_ctx->dev->bdev,
1358 				next_block_ctx->dev_bytenr,
1359 				block_ctx->dev->bdev,
1360 				block_ctx->dev_bytenr,
1361 				&state->block_link_hashtable);
1362 	}
1363 
1364 	next_block->disk_key = *disk_key;
1365 	if (NULL == l) {
1366 		l = btrfsic_block_link_alloc();
1367 		if (NULL == l) {
1368 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
1369 			btrfsic_release_block_ctx(next_block_ctx);
1370 			*next_blockp = NULL;
1371 			return -1;
1372 		}
1373 
1374 		did_alloc_block_link = 1;
1375 		l->block_ref_to = next_block;
1376 		l->block_ref_from = block;
1377 		l->ref_cnt = 1;
1378 		l->parent_generation = parent_generation;
1379 
1380 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1381 			btrfsic_print_add_link(state, l);
1382 
1383 		list_add(&l->node_ref_to, &block->ref_to_list);
1384 		list_add(&l->node_ref_from, &next_block->ref_from_list);
1385 
1386 		btrfsic_block_link_hashtable_add(l,
1387 						 &state->block_link_hashtable);
1388 	} else {
1389 		did_alloc_block_link = 0;
1390 		if (0 == limit_nesting) {
1391 			l->ref_cnt++;
1392 			l->parent_generation = parent_generation;
1393 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1394 				btrfsic_print_add_link(state, l);
1395 		}
1396 	}
1397 
1398 	if (limit_nesting > 0 && did_alloc_block_link) {
1399 		ret = btrfsic_read_block(state, next_block_ctx);
1400 		if (ret < (int)next_block_ctx->len) {
1401 			printk(KERN_INFO
1402 			       "btrfsic: read block @logical %llu failed!\n",
1403 			       (unsigned long long)next_bytenr);
1404 			btrfsic_release_block_ctx(next_block_ctx);
1405 			*next_blockp = NULL;
1406 			return -1;
1407 		}
1408 
1409 		*next_blockp = next_block;
1410 	} else {
1411 		*next_blockp = NULL;
1412 	}
1413 	(*mirror_nump)++;
1414 
1415 	return 0;
1416 }
1417 
1418 static int btrfsic_handle_extent_data(
1419 		struct btrfsic_state *state,
1420 		struct btrfsic_block *block,
1421 		struct btrfsic_block_data_ctx *block_ctx,
1422 		u32 item_offset, int force_iodone_flag)
1423 {
1424 	int ret;
1425 	struct btrfs_file_extent_item file_extent_item;
1426 	u64 file_extent_item_offset;
1427 	u64 next_bytenr;
1428 	u64 num_bytes;
1429 	u64 generation;
1430 	struct btrfsic_block_link *l;
1431 
1432 	file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1433 				  item_offset;
1434 	if (file_extent_item_offset +
1435 	    offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1436 	    block_ctx->len) {
1437 		printk(KERN_INFO
1438 		       "btrfsic: file item out of bounce at logical %llu, dev %s\n",
1439 		       block_ctx->start, block_ctx->dev->name);
1440 		return -1;
1441 	}
1442 
1443 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1444 		file_extent_item_offset,
1445 		offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1446 	if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1447 	    ((u64)0) == le64_to_cpu(file_extent_item.disk_bytenr)) {
1448 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1449 			printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu\n",
1450 			       file_extent_item.type,
1451 			       (unsigned long long)
1452 			       le64_to_cpu(file_extent_item.disk_bytenr));
1453 		return 0;
1454 	}
1455 
1456 	if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1457 	    block_ctx->len) {
1458 		printk(KERN_INFO
1459 		       "btrfsic: file item out of bounce at logical %llu, dev %s\n",
1460 		       block_ctx->start, block_ctx->dev->name);
1461 		return -1;
1462 	}
1463 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1464 				     file_extent_item_offset,
1465 				     sizeof(struct btrfs_file_extent_item));
1466 	next_bytenr = le64_to_cpu(file_extent_item.disk_bytenr) +
1467 		      le64_to_cpu(file_extent_item.offset);
1468 	generation = le64_to_cpu(file_extent_item.generation);
1469 	num_bytes = le64_to_cpu(file_extent_item.num_bytes);
1470 	generation = le64_to_cpu(file_extent_item.generation);
1471 
1472 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1473 		printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu,"
1474 		       " offset = %llu, num_bytes = %llu\n",
1475 		       file_extent_item.type,
1476 		       (unsigned long long)
1477 		       le64_to_cpu(file_extent_item.disk_bytenr),
1478 		       (unsigned long long)le64_to_cpu(file_extent_item.offset),
1479 		       (unsigned long long)num_bytes);
1480 	while (num_bytes > 0) {
1481 		u32 chunk_len;
1482 		int num_copies;
1483 		int mirror_num;
1484 
1485 		if (num_bytes > state->datablock_size)
1486 			chunk_len = state->datablock_size;
1487 		else
1488 			chunk_len = num_bytes;
1489 
1490 		num_copies =
1491 		    btrfs_num_copies(&state->root->fs_info->mapping_tree,
1492 				     next_bytenr, state->datablock_size);
1493 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1494 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
1495 			       (unsigned long long)next_bytenr, num_copies);
1496 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1497 			struct btrfsic_block_data_ctx next_block_ctx;
1498 			struct btrfsic_block *next_block;
1499 			int block_was_created;
1500 
1501 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1502 				printk(KERN_INFO "btrfsic_handle_extent_data("
1503 				       "mirror_num=%d)\n", mirror_num);
1504 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1505 				printk(KERN_INFO
1506 				       "\tdisk_bytenr = %llu, num_bytes %u\n",
1507 				       (unsigned long long)next_bytenr,
1508 				       chunk_len);
1509 			ret = btrfsic_map_block(state, next_bytenr,
1510 						chunk_len, &next_block_ctx,
1511 						mirror_num);
1512 			if (ret) {
1513 				printk(KERN_INFO
1514 				       "btrfsic: btrfsic_map_block(@%llu,"
1515 				       " mirror=%d) failed!\n",
1516 				       (unsigned long long)next_bytenr,
1517 				       mirror_num);
1518 				return -1;
1519 			}
1520 
1521 			next_block = btrfsic_block_lookup_or_add(
1522 					state,
1523 					&next_block_ctx,
1524 					"referenced ",
1525 					0,
1526 					force_iodone_flag,
1527 					!force_iodone_flag,
1528 					mirror_num,
1529 					&block_was_created);
1530 			if (NULL == next_block) {
1531 				printk(KERN_INFO
1532 				       "btrfsic: error, kmalloc failed!\n");
1533 				btrfsic_release_block_ctx(&next_block_ctx);
1534 				return -1;
1535 			}
1536 			if (!block_was_created) {
1537 				if (next_block->logical_bytenr != next_bytenr &&
1538 				    !(!next_block->is_metadata &&
1539 				      0 == next_block->logical_bytenr)) {
1540 					printk(KERN_INFO
1541 					       "Referenced block"
1542 					       " @%llu (%s/%llu/%d)"
1543 					       " found in hash table, D,"
1544 					       " bytenr mismatch"
1545 					       " (!= stored %llu).\n",
1546 					       (unsigned long long)next_bytenr,
1547 					       next_block_ctx.dev->name,
1548 					       (unsigned long long)
1549 					       next_block_ctx.dev_bytenr,
1550 					       mirror_num,
1551 					       (unsigned long long)
1552 					       next_block->logical_bytenr);
1553 				}
1554 				next_block->logical_bytenr = next_bytenr;
1555 				next_block->mirror_num = mirror_num;
1556 			}
1557 
1558 			l = btrfsic_block_link_lookup_or_add(state,
1559 							     &next_block_ctx,
1560 							     next_block, block,
1561 							     generation);
1562 			btrfsic_release_block_ctx(&next_block_ctx);
1563 			if (NULL == l)
1564 				return -1;
1565 		}
1566 
1567 		next_bytenr += chunk_len;
1568 		num_bytes -= chunk_len;
1569 	}
1570 
1571 	return 0;
1572 }
1573 
1574 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1575 			     struct btrfsic_block_data_ctx *block_ctx_out,
1576 			     int mirror_num)
1577 {
1578 	int ret;
1579 	u64 length;
1580 	struct btrfs_bio *multi = NULL;
1581 	struct btrfs_device *device;
1582 
1583 	length = len;
1584 	ret = btrfs_map_block(&state->root->fs_info->mapping_tree, READ,
1585 			      bytenr, &length, &multi, mirror_num);
1586 
1587 	device = multi->stripes[0].dev;
1588 	block_ctx_out->dev = btrfsic_dev_state_lookup(device->bdev);
1589 	block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1590 	block_ctx_out->start = bytenr;
1591 	block_ctx_out->len = len;
1592 	block_ctx_out->datav = NULL;
1593 	block_ctx_out->pagev = NULL;
1594 	block_ctx_out->mem_to_free = NULL;
1595 
1596 	if (0 == ret)
1597 		kfree(multi);
1598 	if (NULL == block_ctx_out->dev) {
1599 		ret = -ENXIO;
1600 		printk(KERN_INFO "btrfsic: error, cannot lookup dev (#1)!\n");
1601 	}
1602 
1603 	return ret;
1604 }
1605 
1606 static int btrfsic_map_superblock(struct btrfsic_state *state, u64 bytenr,
1607 				  u32 len, struct block_device *bdev,
1608 				  struct btrfsic_block_data_ctx *block_ctx_out)
1609 {
1610 	block_ctx_out->dev = btrfsic_dev_state_lookup(bdev);
1611 	block_ctx_out->dev_bytenr = bytenr;
1612 	block_ctx_out->start = bytenr;
1613 	block_ctx_out->len = len;
1614 	block_ctx_out->datav = NULL;
1615 	block_ctx_out->pagev = NULL;
1616 	block_ctx_out->mem_to_free = NULL;
1617 	if (NULL != block_ctx_out->dev) {
1618 		return 0;
1619 	} else {
1620 		printk(KERN_INFO "btrfsic: error, cannot lookup dev (#2)!\n");
1621 		return -ENXIO;
1622 	}
1623 }
1624 
1625 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1626 {
1627 	if (block_ctx->mem_to_free) {
1628 		unsigned int num_pages;
1629 
1630 		BUG_ON(!block_ctx->datav);
1631 		BUG_ON(!block_ctx->pagev);
1632 		num_pages = (block_ctx->len + (u64)PAGE_CACHE_SIZE - 1) >>
1633 			    PAGE_CACHE_SHIFT;
1634 		while (num_pages > 0) {
1635 			num_pages--;
1636 			if (block_ctx->datav[num_pages]) {
1637 				kunmap(block_ctx->pagev[num_pages]);
1638 				block_ctx->datav[num_pages] = NULL;
1639 			}
1640 			if (block_ctx->pagev[num_pages]) {
1641 				__free_page(block_ctx->pagev[num_pages]);
1642 				block_ctx->pagev[num_pages] = NULL;
1643 			}
1644 		}
1645 
1646 		kfree(block_ctx->mem_to_free);
1647 		block_ctx->mem_to_free = NULL;
1648 		block_ctx->pagev = NULL;
1649 		block_ctx->datav = NULL;
1650 	}
1651 }
1652 
1653 static int btrfsic_read_block(struct btrfsic_state *state,
1654 			      struct btrfsic_block_data_ctx *block_ctx)
1655 {
1656 	unsigned int num_pages;
1657 	unsigned int i;
1658 	u64 dev_bytenr;
1659 	int ret;
1660 
1661 	BUG_ON(block_ctx->datav);
1662 	BUG_ON(block_ctx->pagev);
1663 	BUG_ON(block_ctx->mem_to_free);
1664 	if (block_ctx->dev_bytenr & ((u64)PAGE_CACHE_SIZE - 1)) {
1665 		printk(KERN_INFO
1666 		       "btrfsic: read_block() with unaligned bytenr %llu\n",
1667 		       (unsigned long long)block_ctx->dev_bytenr);
1668 		return -1;
1669 	}
1670 
1671 	num_pages = (block_ctx->len + (u64)PAGE_CACHE_SIZE - 1) >>
1672 		    PAGE_CACHE_SHIFT;
1673 	block_ctx->mem_to_free = kzalloc((sizeof(*block_ctx->datav) +
1674 					  sizeof(*block_ctx->pagev)) *
1675 					 num_pages, GFP_NOFS);
1676 	if (!block_ctx->mem_to_free)
1677 		return -1;
1678 	block_ctx->datav = block_ctx->mem_to_free;
1679 	block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1680 	for (i = 0; i < num_pages; i++) {
1681 		block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1682 		if (!block_ctx->pagev[i])
1683 			return -1;
1684 	}
1685 
1686 	dev_bytenr = block_ctx->dev_bytenr;
1687 	for (i = 0; i < num_pages;) {
1688 		struct bio *bio;
1689 		unsigned int j;
1690 		DECLARE_COMPLETION_ONSTACK(complete);
1691 
1692 		bio = bio_alloc(GFP_NOFS, num_pages - i);
1693 		if (!bio) {
1694 			printk(KERN_INFO
1695 			       "btrfsic: bio_alloc() for %u pages failed!\n",
1696 			       num_pages - i);
1697 			return -1;
1698 		}
1699 		bio->bi_bdev = block_ctx->dev->bdev;
1700 		bio->bi_sector = dev_bytenr >> 9;
1701 		bio->bi_end_io = btrfsic_complete_bio_end_io;
1702 		bio->bi_private = &complete;
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 		submit_bio(READ, bio);
1716 
1717 		/* this will also unplug the queue */
1718 		wait_for_completion(&complete);
1719 
1720 		if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1721 			printk(KERN_INFO
1722 			       "btrfsic: read error at logical %llu dev %s!\n",
1723 			       block_ctx->start, block_ctx->dev->name);
1724 			bio_put(bio);
1725 			return -1;
1726 		}
1727 		bio_put(bio);
1728 		dev_bytenr += (j - i) * PAGE_CACHE_SIZE;
1729 		i = j;
1730 	}
1731 	for (i = 0; i < num_pages; i++) {
1732 		block_ctx->datav[i] = kmap(block_ctx->pagev[i]);
1733 		if (!block_ctx->datav[i]) {
1734 			printk(KERN_INFO "btrfsic: kmap() failed (dev %s)!\n",
1735 			       block_ctx->dev->name);
1736 			return -1;
1737 		}
1738 	}
1739 
1740 	return block_ctx->len;
1741 }
1742 
1743 static void btrfsic_complete_bio_end_io(struct bio *bio, int err)
1744 {
1745 	complete((struct completion *)bio->bi_private);
1746 }
1747 
1748 static void btrfsic_dump_database(struct btrfsic_state *state)
1749 {
1750 	struct list_head *elem_all;
1751 
1752 	BUG_ON(NULL == state);
1753 
1754 	printk(KERN_INFO "all_blocks_list:\n");
1755 	list_for_each(elem_all, &state->all_blocks_list) {
1756 		const struct btrfsic_block *const b_all =
1757 		    list_entry(elem_all, struct btrfsic_block,
1758 			       all_blocks_node);
1759 		struct list_head *elem_ref_to;
1760 		struct list_head *elem_ref_from;
1761 
1762 		printk(KERN_INFO "%c-block @%llu (%s/%llu/%d)\n",
1763 		       btrfsic_get_block_type(state, b_all),
1764 		       (unsigned long long)b_all->logical_bytenr,
1765 		       b_all->dev_state->name,
1766 		       (unsigned long long)b_all->dev_bytenr,
1767 		       b_all->mirror_num);
1768 
1769 		list_for_each(elem_ref_to, &b_all->ref_to_list) {
1770 			const struct btrfsic_block_link *const l =
1771 			    list_entry(elem_ref_to,
1772 				       struct btrfsic_block_link,
1773 				       node_ref_to);
1774 
1775 			printk(KERN_INFO " %c @%llu (%s/%llu/%d)"
1776 			       " refers %u* to"
1777 			       " %c @%llu (%s/%llu/%d)\n",
1778 			       btrfsic_get_block_type(state, b_all),
1779 			       (unsigned long long)b_all->logical_bytenr,
1780 			       b_all->dev_state->name,
1781 			       (unsigned long long)b_all->dev_bytenr,
1782 			       b_all->mirror_num,
1783 			       l->ref_cnt,
1784 			       btrfsic_get_block_type(state, l->block_ref_to),
1785 			       (unsigned long long)
1786 			       l->block_ref_to->logical_bytenr,
1787 			       l->block_ref_to->dev_state->name,
1788 			       (unsigned long long)l->block_ref_to->dev_bytenr,
1789 			       l->block_ref_to->mirror_num);
1790 		}
1791 
1792 		list_for_each(elem_ref_from, &b_all->ref_from_list) {
1793 			const struct btrfsic_block_link *const l =
1794 			    list_entry(elem_ref_from,
1795 				       struct btrfsic_block_link,
1796 				       node_ref_from);
1797 
1798 			printk(KERN_INFO " %c @%llu (%s/%llu/%d)"
1799 			       " is ref %u* from"
1800 			       " %c @%llu (%s/%llu/%d)\n",
1801 			       btrfsic_get_block_type(state, b_all),
1802 			       (unsigned long long)b_all->logical_bytenr,
1803 			       b_all->dev_state->name,
1804 			       (unsigned long long)b_all->dev_bytenr,
1805 			       b_all->mirror_num,
1806 			       l->ref_cnt,
1807 			       btrfsic_get_block_type(state, l->block_ref_from),
1808 			       (unsigned long long)
1809 			       l->block_ref_from->logical_bytenr,
1810 			       l->block_ref_from->dev_state->name,
1811 			       (unsigned long long)
1812 			       l->block_ref_from->dev_bytenr,
1813 			       l->block_ref_from->mirror_num);
1814 		}
1815 
1816 		printk(KERN_INFO "\n");
1817 	}
1818 }
1819 
1820 /*
1821  * Test whether the disk block contains a tree block (leaf or node)
1822  * (note that this test fails for the super block)
1823  */
1824 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
1825 				     char **datav, unsigned int num_pages)
1826 {
1827 	struct btrfs_header *h;
1828 	u8 csum[BTRFS_CSUM_SIZE];
1829 	u32 crc = ~(u32)0;
1830 	unsigned int i;
1831 
1832 	if (num_pages * PAGE_CACHE_SIZE < state->metablock_size)
1833 		return 1; /* not metadata */
1834 	num_pages = state->metablock_size >> PAGE_CACHE_SHIFT;
1835 	h = (struct btrfs_header *)datav[0];
1836 
1837 	if (memcmp(h->fsid, state->root->fs_info->fsid, BTRFS_UUID_SIZE))
1838 		return 1;
1839 
1840 	for (i = 0; i < num_pages; i++) {
1841 		u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1842 		size_t sublen = i ? PAGE_CACHE_SIZE :
1843 				    (PAGE_CACHE_SIZE - BTRFS_CSUM_SIZE);
1844 
1845 		crc = crc32c(crc, data, sublen);
1846 	}
1847 	btrfs_csum_final(crc, csum);
1848 	if (memcmp(csum, h->csum, state->csum_size))
1849 		return 1;
1850 
1851 	return 0; /* is metadata */
1852 }
1853 
1854 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1855 					  u64 dev_bytenr, char **mapped_datav,
1856 					  unsigned int num_pages,
1857 					  struct bio *bio, int *bio_is_patched,
1858 					  struct buffer_head *bh,
1859 					  int submit_bio_bh_rw)
1860 {
1861 	int is_metadata;
1862 	struct btrfsic_block *block;
1863 	struct btrfsic_block_data_ctx block_ctx;
1864 	int ret;
1865 	struct btrfsic_state *state = dev_state->state;
1866 	struct block_device *bdev = dev_state->bdev;
1867 	unsigned int processed_len;
1868 
1869 	if (NULL != bio_is_patched)
1870 		*bio_is_patched = 0;
1871 
1872 again:
1873 	if (num_pages == 0)
1874 		return;
1875 
1876 	processed_len = 0;
1877 	is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1878 						      num_pages));
1879 
1880 	block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1881 					       &state->block_hashtable);
1882 	if (NULL != block) {
1883 		u64 bytenr = 0;
1884 		struct list_head *elem_ref_to;
1885 		struct list_head *tmp_ref_to;
1886 
1887 		if (block->is_superblock) {
1888 			bytenr = le64_to_cpu(((struct btrfs_super_block *)
1889 					      mapped_datav[0])->bytenr);
1890 			if (num_pages * PAGE_CACHE_SIZE <
1891 			    BTRFS_SUPER_INFO_SIZE) {
1892 				printk(KERN_INFO
1893 				       "btrfsic: cannot work with too short bios!\n");
1894 				return;
1895 			}
1896 			is_metadata = 1;
1897 			BUG_ON(BTRFS_SUPER_INFO_SIZE & (PAGE_CACHE_SIZE - 1));
1898 			processed_len = BTRFS_SUPER_INFO_SIZE;
1899 			if (state->print_mask &
1900 			    BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1901 				printk(KERN_INFO
1902 				       "[before new superblock is written]:\n");
1903 				btrfsic_dump_tree_sub(state, block, 0);
1904 			}
1905 		}
1906 		if (is_metadata) {
1907 			if (!block->is_superblock) {
1908 				if (num_pages * PAGE_CACHE_SIZE <
1909 				    state->metablock_size) {
1910 					printk(KERN_INFO
1911 					       "btrfsic: cannot work with too short bios!\n");
1912 					return;
1913 				}
1914 				processed_len = state->metablock_size;
1915 				bytenr = le64_to_cpu(((struct btrfs_header *)
1916 						      mapped_datav[0])->bytenr);
1917 				btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1918 							       dev_state,
1919 							       dev_bytenr);
1920 			}
1921 			if (block->logical_bytenr != bytenr) {
1922 				printk(KERN_INFO
1923 				       "Written block @%llu (%s/%llu/%d)"
1924 				       " found in hash table, %c,"
1925 				       " bytenr mismatch"
1926 				       " (!= stored %llu).\n",
1927 				       (unsigned long long)bytenr,
1928 				       dev_state->name,
1929 				       (unsigned long long)dev_bytenr,
1930 				       block->mirror_num,
1931 				       btrfsic_get_block_type(state, block),
1932 				       (unsigned long long)
1933 				       block->logical_bytenr);
1934 				block->logical_bytenr = bytenr;
1935 			} else if (state->print_mask &
1936 				   BTRFSIC_PRINT_MASK_VERBOSE)
1937 				printk(KERN_INFO
1938 				       "Written block @%llu (%s/%llu/%d)"
1939 				       " found in hash table, %c.\n",
1940 				       (unsigned long long)bytenr,
1941 				       dev_state->name,
1942 				       (unsigned long long)dev_bytenr,
1943 				       block->mirror_num,
1944 				       btrfsic_get_block_type(state, block));
1945 		} else {
1946 			if (num_pages * PAGE_CACHE_SIZE <
1947 			    state->datablock_size) {
1948 				printk(KERN_INFO
1949 				       "btrfsic: cannot work with too short bios!\n");
1950 				return;
1951 			}
1952 			processed_len = state->datablock_size;
1953 			bytenr = block->logical_bytenr;
1954 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1955 				printk(KERN_INFO
1956 				       "Written block @%llu (%s/%llu/%d)"
1957 				       " found in hash table, %c.\n",
1958 				       (unsigned long long)bytenr,
1959 				       dev_state->name,
1960 				       (unsigned long long)dev_bytenr,
1961 				       block->mirror_num,
1962 				       btrfsic_get_block_type(state, block));
1963 		}
1964 
1965 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1966 			printk(KERN_INFO
1967 			       "ref_to_list: %cE, ref_from_list: %cE\n",
1968 			       list_empty(&block->ref_to_list) ? ' ' : '!',
1969 			       list_empty(&block->ref_from_list) ? ' ' : '!');
1970 		if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1971 			printk(KERN_INFO "btrfs: attempt to overwrite %c-block"
1972 			       " @%llu (%s/%llu/%d), old(gen=%llu,"
1973 			       " objectid=%llu, type=%d, offset=%llu),"
1974 			       " new(gen=%llu),"
1975 			       " which is referenced by most recent superblock"
1976 			       " (superblockgen=%llu)!\n",
1977 			       btrfsic_get_block_type(state, block),
1978 			       (unsigned long long)bytenr,
1979 			       dev_state->name,
1980 			       (unsigned long long)dev_bytenr,
1981 			       block->mirror_num,
1982 			       (unsigned long long)block->generation,
1983 			       (unsigned long long)
1984 			       le64_to_cpu(block->disk_key.objectid),
1985 			       block->disk_key.type,
1986 			       (unsigned long long)
1987 			       le64_to_cpu(block->disk_key.offset),
1988 			       (unsigned long long)
1989 			       le64_to_cpu(((struct btrfs_header *)
1990 					    mapped_datav[0])->generation),
1991 			       (unsigned long long)
1992 			       state->max_superblock_generation);
1993 			btrfsic_dump_tree(state);
1994 		}
1995 
1996 		if (!block->is_iodone && !block->never_written) {
1997 			printk(KERN_INFO "btrfs: attempt to overwrite %c-block"
1998 			       " @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu,"
1999 			       " which is not yet iodone!\n",
2000 			       btrfsic_get_block_type(state, block),
2001 			       (unsigned long long)bytenr,
2002 			       dev_state->name,
2003 			       (unsigned long long)dev_bytenr,
2004 			       block->mirror_num,
2005 			       (unsigned long long)block->generation,
2006 			       (unsigned long long)
2007 			       le64_to_cpu(((struct btrfs_header *)
2008 					    mapped_datav[0])->generation));
2009 			/* it would not be safe to go on */
2010 			btrfsic_dump_tree(state);
2011 			goto continue_loop;
2012 		}
2013 
2014 		/*
2015 		 * Clear all references of this block. Do not free
2016 		 * the block itself even if is not referenced anymore
2017 		 * because it still carries valueable information
2018 		 * like whether it was ever written and IO completed.
2019 		 */
2020 		list_for_each_safe(elem_ref_to, tmp_ref_to,
2021 				   &block->ref_to_list) {
2022 			struct btrfsic_block_link *const l =
2023 			    list_entry(elem_ref_to,
2024 				       struct btrfsic_block_link,
2025 				       node_ref_to);
2026 
2027 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2028 				btrfsic_print_rem_link(state, l);
2029 			l->ref_cnt--;
2030 			if (0 == l->ref_cnt) {
2031 				list_del(&l->node_ref_to);
2032 				list_del(&l->node_ref_from);
2033 				btrfsic_block_link_hashtable_remove(l);
2034 				btrfsic_block_link_free(l);
2035 			}
2036 		}
2037 
2038 		if (block->is_superblock)
2039 			ret = btrfsic_map_superblock(state, bytenr,
2040 						     processed_len,
2041 						     bdev, &block_ctx);
2042 		else
2043 			ret = btrfsic_map_block(state, bytenr, processed_len,
2044 						&block_ctx, 0);
2045 		if (ret) {
2046 			printk(KERN_INFO
2047 			       "btrfsic: btrfsic_map_block(root @%llu)"
2048 			       " failed!\n", (unsigned long long)bytenr);
2049 			goto continue_loop;
2050 		}
2051 		block_ctx.datav = mapped_datav;
2052 		/* the following is required in case of writes to mirrors,
2053 		 * use the same that was used for the lookup */
2054 		block_ctx.dev = dev_state;
2055 		block_ctx.dev_bytenr = dev_bytenr;
2056 
2057 		if (is_metadata || state->include_extent_data) {
2058 			block->never_written = 0;
2059 			block->iodone_w_error = 0;
2060 			if (NULL != bio) {
2061 				block->is_iodone = 0;
2062 				BUG_ON(NULL == bio_is_patched);
2063 				if (!*bio_is_patched) {
2064 					block->orig_bio_bh_private =
2065 					    bio->bi_private;
2066 					block->orig_bio_bh_end_io.bio =
2067 					    bio->bi_end_io;
2068 					block->next_in_same_bio = NULL;
2069 					bio->bi_private = block;
2070 					bio->bi_end_io = btrfsic_bio_end_io;
2071 					*bio_is_patched = 1;
2072 				} else {
2073 					struct btrfsic_block *chained_block =
2074 					    (struct btrfsic_block *)
2075 					    bio->bi_private;
2076 
2077 					BUG_ON(NULL == chained_block);
2078 					block->orig_bio_bh_private =
2079 					    chained_block->orig_bio_bh_private;
2080 					block->orig_bio_bh_end_io.bio =
2081 					    chained_block->orig_bio_bh_end_io.
2082 					    bio;
2083 					block->next_in_same_bio = chained_block;
2084 					bio->bi_private = block;
2085 				}
2086 			} else if (NULL != bh) {
2087 				block->is_iodone = 0;
2088 				block->orig_bio_bh_private = bh->b_private;
2089 				block->orig_bio_bh_end_io.bh = bh->b_end_io;
2090 				block->next_in_same_bio = NULL;
2091 				bh->b_private = block;
2092 				bh->b_end_io = btrfsic_bh_end_io;
2093 			} else {
2094 				block->is_iodone = 1;
2095 				block->orig_bio_bh_private = NULL;
2096 				block->orig_bio_bh_end_io.bio = NULL;
2097 				block->next_in_same_bio = NULL;
2098 			}
2099 		}
2100 
2101 		block->flush_gen = dev_state->last_flush_gen + 1;
2102 		block->submit_bio_bh_rw = submit_bio_bh_rw;
2103 		if (is_metadata) {
2104 			block->logical_bytenr = bytenr;
2105 			block->is_metadata = 1;
2106 			if (block->is_superblock) {
2107 				BUG_ON(PAGE_CACHE_SIZE !=
2108 				       BTRFS_SUPER_INFO_SIZE);
2109 				ret = btrfsic_process_written_superblock(
2110 						state,
2111 						block,
2112 						(struct btrfs_super_block *)
2113 						mapped_datav[0]);
2114 				if (state->print_mask &
2115 				    BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
2116 					printk(KERN_INFO
2117 					"[after new superblock is written]:\n");
2118 					btrfsic_dump_tree_sub(state, block, 0);
2119 				}
2120 			} else {
2121 				block->mirror_num = 0;	/* unknown */
2122 				ret = btrfsic_process_metablock(
2123 						state,
2124 						block,
2125 						&block_ctx,
2126 						0, 0);
2127 			}
2128 			if (ret)
2129 				printk(KERN_INFO
2130 				       "btrfsic: btrfsic_process_metablock"
2131 				       "(root @%llu) failed!\n",
2132 				       (unsigned long long)dev_bytenr);
2133 		} else {
2134 			block->is_metadata = 0;
2135 			block->mirror_num = 0;	/* unknown */
2136 			block->generation = BTRFSIC_GENERATION_UNKNOWN;
2137 			if (!state->include_extent_data
2138 			    && list_empty(&block->ref_from_list)) {
2139 				/*
2140 				 * disk block is overwritten with extent
2141 				 * data (not meta data) and we are configured
2142 				 * to not include extent data: take the
2143 				 * chance and free the block's memory
2144 				 */
2145 				btrfsic_block_hashtable_remove(block);
2146 				list_del(&block->all_blocks_node);
2147 				btrfsic_block_free(block);
2148 			}
2149 		}
2150 		btrfsic_release_block_ctx(&block_ctx);
2151 	} else {
2152 		/* block has not been found in hash table */
2153 		u64 bytenr;
2154 
2155 		if (!is_metadata) {
2156 			processed_len = state->datablock_size;
2157 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2158 				printk(KERN_INFO "Written block (%s/%llu/?)"
2159 				       " !found in hash table, D.\n",
2160 				       dev_state->name,
2161 				       (unsigned long long)dev_bytenr);
2162 			if (!state->include_extent_data) {
2163 				/* ignore that written D block */
2164 				goto continue_loop;
2165 			}
2166 
2167 			/* this is getting ugly for the
2168 			 * include_extent_data case... */
2169 			bytenr = 0;	/* unknown */
2170 			block_ctx.start = bytenr;
2171 			block_ctx.len = processed_len;
2172 			block_ctx.mem_to_free = NULL;
2173 			block_ctx.pagev = NULL;
2174 		} else {
2175 			processed_len = state->metablock_size;
2176 			bytenr = le64_to_cpu(((struct btrfs_header *)
2177 					      mapped_datav[0])->bytenr);
2178 			btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
2179 						       dev_bytenr);
2180 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2181 				printk(KERN_INFO
2182 				       "Written block @%llu (%s/%llu/?)"
2183 				       " !found in hash table, M.\n",
2184 				       (unsigned long long)bytenr,
2185 				       dev_state->name,
2186 				       (unsigned long long)dev_bytenr);
2187 
2188 			ret = btrfsic_map_block(state, bytenr, processed_len,
2189 						&block_ctx, 0);
2190 			if (ret) {
2191 				printk(KERN_INFO
2192 				       "btrfsic: btrfsic_map_block(root @%llu)"
2193 				       " failed!\n",
2194 				       (unsigned long long)dev_bytenr);
2195 				goto continue_loop;
2196 			}
2197 		}
2198 		block_ctx.datav = mapped_datav;
2199 		/* the following is required in case of writes to mirrors,
2200 		 * use the same that was used for the lookup */
2201 		block_ctx.dev = dev_state;
2202 		block_ctx.dev_bytenr = dev_bytenr;
2203 
2204 		block = btrfsic_block_alloc();
2205 		if (NULL == block) {
2206 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
2207 			btrfsic_release_block_ctx(&block_ctx);
2208 			goto continue_loop;
2209 		}
2210 		block->dev_state = dev_state;
2211 		block->dev_bytenr = dev_bytenr;
2212 		block->logical_bytenr = bytenr;
2213 		block->is_metadata = is_metadata;
2214 		block->never_written = 0;
2215 		block->iodone_w_error = 0;
2216 		block->mirror_num = 0;	/* unknown */
2217 		block->flush_gen = dev_state->last_flush_gen + 1;
2218 		block->submit_bio_bh_rw = submit_bio_bh_rw;
2219 		if (NULL != bio) {
2220 			block->is_iodone = 0;
2221 			BUG_ON(NULL == bio_is_patched);
2222 			if (!*bio_is_patched) {
2223 				block->orig_bio_bh_private = bio->bi_private;
2224 				block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2225 				block->next_in_same_bio = NULL;
2226 				bio->bi_private = block;
2227 				bio->bi_end_io = btrfsic_bio_end_io;
2228 				*bio_is_patched = 1;
2229 			} else {
2230 				struct btrfsic_block *chained_block =
2231 				    (struct btrfsic_block *)
2232 				    bio->bi_private;
2233 
2234 				BUG_ON(NULL == chained_block);
2235 				block->orig_bio_bh_private =
2236 				    chained_block->orig_bio_bh_private;
2237 				block->orig_bio_bh_end_io.bio =
2238 				    chained_block->orig_bio_bh_end_io.bio;
2239 				block->next_in_same_bio = chained_block;
2240 				bio->bi_private = block;
2241 			}
2242 		} else if (NULL != bh) {
2243 			block->is_iodone = 0;
2244 			block->orig_bio_bh_private = bh->b_private;
2245 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2246 			block->next_in_same_bio = NULL;
2247 			bh->b_private = block;
2248 			bh->b_end_io = btrfsic_bh_end_io;
2249 		} else {
2250 			block->is_iodone = 1;
2251 			block->orig_bio_bh_private = NULL;
2252 			block->orig_bio_bh_end_io.bio = NULL;
2253 			block->next_in_same_bio = NULL;
2254 		}
2255 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2256 			printk(KERN_INFO
2257 			       "New written %c-block @%llu (%s/%llu/%d)\n",
2258 			       is_metadata ? 'M' : 'D',
2259 			       (unsigned long long)block->logical_bytenr,
2260 			       block->dev_state->name,
2261 			       (unsigned long long)block->dev_bytenr,
2262 			       block->mirror_num);
2263 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2264 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2265 
2266 		if (is_metadata) {
2267 			ret = btrfsic_process_metablock(state, block,
2268 							&block_ctx, 0, 0);
2269 			if (ret)
2270 				printk(KERN_INFO
2271 				       "btrfsic: process_metablock(root @%llu)"
2272 				       " failed!\n",
2273 				       (unsigned long long)dev_bytenr);
2274 		}
2275 		btrfsic_release_block_ctx(&block_ctx);
2276 	}
2277 
2278 continue_loop:
2279 	BUG_ON(!processed_len);
2280 	dev_bytenr += processed_len;
2281 	mapped_datav += processed_len >> PAGE_CACHE_SHIFT;
2282 	num_pages -= processed_len >> PAGE_CACHE_SHIFT;
2283 	goto again;
2284 }
2285 
2286 static void btrfsic_bio_end_io(struct bio *bp, int bio_error_status)
2287 {
2288 	struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2289 	int iodone_w_error;
2290 
2291 	/* mutex is not held! This is not save if IO is not yet completed
2292 	 * on umount */
2293 	iodone_w_error = 0;
2294 	if (bio_error_status)
2295 		iodone_w_error = 1;
2296 
2297 	BUG_ON(NULL == block);
2298 	bp->bi_private = block->orig_bio_bh_private;
2299 	bp->bi_end_io = block->orig_bio_bh_end_io.bio;
2300 
2301 	do {
2302 		struct btrfsic_block *next_block;
2303 		struct btrfsic_dev_state *const dev_state = block->dev_state;
2304 
2305 		if ((dev_state->state->print_mask &
2306 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2307 			printk(KERN_INFO
2308 			       "bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2309 			       bio_error_status,
2310 			       btrfsic_get_block_type(dev_state->state, block),
2311 			       (unsigned long long)block->logical_bytenr,
2312 			       dev_state->name,
2313 			       (unsigned long long)block->dev_bytenr,
2314 			       block->mirror_num);
2315 		next_block = block->next_in_same_bio;
2316 		block->iodone_w_error = iodone_w_error;
2317 		if (block->submit_bio_bh_rw & REQ_FLUSH) {
2318 			dev_state->last_flush_gen++;
2319 			if ((dev_state->state->print_mask &
2320 			     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2321 				printk(KERN_INFO
2322 				       "bio_end_io() new %s flush_gen=%llu\n",
2323 				       dev_state->name,
2324 				       (unsigned long long)
2325 				       dev_state->last_flush_gen);
2326 		}
2327 		if (block->submit_bio_bh_rw & REQ_FUA)
2328 			block->flush_gen = 0; /* FUA completed means block is
2329 					       * on disk */
2330 		block->is_iodone = 1; /* for FLUSH, this releases the block */
2331 		block = next_block;
2332 	} while (NULL != block);
2333 
2334 	bp->bi_end_io(bp, bio_error_status);
2335 }
2336 
2337 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate)
2338 {
2339 	struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private;
2340 	int iodone_w_error = !uptodate;
2341 	struct btrfsic_dev_state *dev_state;
2342 
2343 	BUG_ON(NULL == block);
2344 	dev_state = block->dev_state;
2345 	if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2346 		printk(KERN_INFO
2347 		       "bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n",
2348 		       iodone_w_error,
2349 		       btrfsic_get_block_type(dev_state->state, block),
2350 		       (unsigned long long)block->logical_bytenr,
2351 		       block->dev_state->name,
2352 		       (unsigned long long)block->dev_bytenr,
2353 		       block->mirror_num);
2354 
2355 	block->iodone_w_error = iodone_w_error;
2356 	if (block->submit_bio_bh_rw & REQ_FLUSH) {
2357 		dev_state->last_flush_gen++;
2358 		if ((dev_state->state->print_mask &
2359 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2360 			printk(KERN_INFO
2361 			       "bh_end_io() new %s flush_gen=%llu\n",
2362 			       dev_state->name,
2363 			       (unsigned long long)dev_state->last_flush_gen);
2364 	}
2365 	if (block->submit_bio_bh_rw & REQ_FUA)
2366 		block->flush_gen = 0; /* FUA completed means block is on disk */
2367 
2368 	bh->b_private = block->orig_bio_bh_private;
2369 	bh->b_end_io = block->orig_bio_bh_end_io.bh;
2370 	block->is_iodone = 1; /* for FLUSH, this releases the block */
2371 	bh->b_end_io(bh, uptodate);
2372 }
2373 
2374 static int btrfsic_process_written_superblock(
2375 		struct btrfsic_state *state,
2376 		struct btrfsic_block *const superblock,
2377 		struct btrfs_super_block *const super_hdr)
2378 {
2379 	int pass;
2380 
2381 	superblock->generation = btrfs_super_generation(super_hdr);
2382 	if (!(superblock->generation > state->max_superblock_generation ||
2383 	      0 == state->max_superblock_generation)) {
2384 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2385 			printk(KERN_INFO
2386 			       "btrfsic: superblock @%llu (%s/%llu/%d)"
2387 			       " with old gen %llu <= %llu\n",
2388 			       (unsigned long long)superblock->logical_bytenr,
2389 			       superblock->dev_state->name,
2390 			       (unsigned long long)superblock->dev_bytenr,
2391 			       superblock->mirror_num,
2392 			       (unsigned long long)
2393 			       btrfs_super_generation(super_hdr),
2394 			       (unsigned long long)
2395 			       state->max_superblock_generation);
2396 	} else {
2397 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2398 			printk(KERN_INFO
2399 			       "btrfsic: got new superblock @%llu (%s/%llu/%d)"
2400 			       " with new gen %llu > %llu\n",
2401 			       (unsigned long long)superblock->logical_bytenr,
2402 			       superblock->dev_state->name,
2403 			       (unsigned long long)superblock->dev_bytenr,
2404 			       superblock->mirror_num,
2405 			       (unsigned long long)
2406 			       btrfs_super_generation(super_hdr),
2407 			       (unsigned long long)
2408 			       state->max_superblock_generation);
2409 
2410 		state->max_superblock_generation =
2411 		    btrfs_super_generation(super_hdr);
2412 		state->latest_superblock = superblock;
2413 	}
2414 
2415 	for (pass = 0; pass < 3; pass++) {
2416 		int ret;
2417 		u64 next_bytenr;
2418 		struct btrfsic_block *next_block;
2419 		struct btrfsic_block_data_ctx tmp_next_block_ctx;
2420 		struct btrfsic_block_link *l;
2421 		int num_copies;
2422 		int mirror_num;
2423 		const char *additional_string = NULL;
2424 		struct btrfs_disk_key tmp_disk_key;
2425 
2426 		tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
2427 		tmp_disk_key.offset = 0;
2428 
2429 		switch (pass) {
2430 		case 0:
2431 			tmp_disk_key.objectid =
2432 			    cpu_to_le64(BTRFS_ROOT_TREE_OBJECTID);
2433 			additional_string = "root ";
2434 			next_bytenr = btrfs_super_root(super_hdr);
2435 			if (state->print_mask &
2436 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2437 				printk(KERN_INFO "root@%llu\n",
2438 				       (unsigned long long)next_bytenr);
2439 			break;
2440 		case 1:
2441 			tmp_disk_key.objectid =
2442 			    cpu_to_le64(BTRFS_CHUNK_TREE_OBJECTID);
2443 			additional_string = "chunk ";
2444 			next_bytenr = btrfs_super_chunk_root(super_hdr);
2445 			if (state->print_mask &
2446 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2447 				printk(KERN_INFO "chunk@%llu\n",
2448 				       (unsigned long long)next_bytenr);
2449 			break;
2450 		case 2:
2451 			tmp_disk_key.objectid =
2452 			    cpu_to_le64(BTRFS_TREE_LOG_OBJECTID);
2453 			additional_string = "log ";
2454 			next_bytenr = btrfs_super_log_root(super_hdr);
2455 			if (0 == next_bytenr)
2456 				continue;
2457 			if (state->print_mask &
2458 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2459 				printk(KERN_INFO "log@%llu\n",
2460 				       (unsigned long long)next_bytenr);
2461 			break;
2462 		}
2463 
2464 		num_copies =
2465 		    btrfs_num_copies(&state->root->fs_info->mapping_tree,
2466 				     next_bytenr, BTRFS_SUPER_INFO_SIZE);
2467 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2468 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
2469 			       (unsigned long long)next_bytenr, num_copies);
2470 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2471 			int was_created;
2472 
2473 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2474 				printk(KERN_INFO
2475 				       "btrfsic_process_written_superblock("
2476 				       "mirror_num=%d)\n", mirror_num);
2477 			ret = btrfsic_map_block(state, next_bytenr,
2478 						BTRFS_SUPER_INFO_SIZE,
2479 						&tmp_next_block_ctx,
2480 						mirror_num);
2481 			if (ret) {
2482 				printk(KERN_INFO
2483 				       "btrfsic: btrfsic_map_block(@%llu,"
2484 				       " mirror=%d) failed!\n",
2485 				       (unsigned long long)next_bytenr,
2486 				       mirror_num);
2487 				return -1;
2488 			}
2489 
2490 			next_block = btrfsic_block_lookup_or_add(
2491 					state,
2492 					&tmp_next_block_ctx,
2493 					additional_string,
2494 					1, 0, 1,
2495 					mirror_num,
2496 					&was_created);
2497 			if (NULL == next_block) {
2498 				printk(KERN_INFO
2499 				       "btrfsic: error, kmalloc failed!\n");
2500 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
2501 				return -1;
2502 			}
2503 
2504 			next_block->disk_key = tmp_disk_key;
2505 			if (was_created)
2506 				next_block->generation =
2507 				    BTRFSIC_GENERATION_UNKNOWN;
2508 			l = btrfsic_block_link_lookup_or_add(
2509 					state,
2510 					&tmp_next_block_ctx,
2511 					next_block,
2512 					superblock,
2513 					BTRFSIC_GENERATION_UNKNOWN);
2514 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
2515 			if (NULL == l)
2516 				return -1;
2517 		}
2518 	}
2519 
2520 	if (-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)) {
2521 		WARN_ON(1);
2522 		btrfsic_dump_tree(state);
2523 	}
2524 
2525 	return 0;
2526 }
2527 
2528 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2529 					struct btrfsic_block *const block,
2530 					int recursion_level)
2531 {
2532 	struct list_head *elem_ref_to;
2533 	int ret = 0;
2534 
2535 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2536 		/*
2537 		 * Note that this situation can happen and does not
2538 		 * indicate an error in regular cases. It happens
2539 		 * when disk blocks are freed and later reused.
2540 		 * The check-integrity module is not aware of any
2541 		 * block free operations, it just recognizes block
2542 		 * write operations. Therefore it keeps the linkage
2543 		 * information for a block until a block is
2544 		 * rewritten. This can temporarily cause incorrect
2545 		 * and even circular linkage informations. This
2546 		 * causes no harm unless such blocks are referenced
2547 		 * by the most recent super block.
2548 		 */
2549 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2550 			printk(KERN_INFO
2551 			       "btrfsic: abort cyclic linkage (case 1).\n");
2552 
2553 		return ret;
2554 	}
2555 
2556 	/*
2557 	 * This algorithm is recursive because the amount of used stack
2558 	 * space is very small and the max recursion depth is limited.
2559 	 */
2560 	list_for_each(elem_ref_to, &block->ref_to_list) {
2561 		const struct btrfsic_block_link *const l =
2562 		    list_entry(elem_ref_to, struct btrfsic_block_link,
2563 			       node_ref_to);
2564 
2565 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2566 			printk(KERN_INFO
2567 			       "rl=%d, %c @%llu (%s/%llu/%d)"
2568 			       " %u* refers to %c @%llu (%s/%llu/%d)\n",
2569 			       recursion_level,
2570 			       btrfsic_get_block_type(state, block),
2571 			       (unsigned long long)block->logical_bytenr,
2572 			       block->dev_state->name,
2573 			       (unsigned long long)block->dev_bytenr,
2574 			       block->mirror_num,
2575 			       l->ref_cnt,
2576 			       btrfsic_get_block_type(state, l->block_ref_to),
2577 			       (unsigned long long)
2578 			       l->block_ref_to->logical_bytenr,
2579 			       l->block_ref_to->dev_state->name,
2580 			       (unsigned long long)l->block_ref_to->dev_bytenr,
2581 			       l->block_ref_to->mirror_num);
2582 		if (l->block_ref_to->never_written) {
2583 			printk(KERN_INFO "btrfs: attempt to write superblock"
2584 			       " which references block %c @%llu (%s/%llu/%d)"
2585 			       " which is never written!\n",
2586 			       btrfsic_get_block_type(state, l->block_ref_to),
2587 			       (unsigned long long)
2588 			       l->block_ref_to->logical_bytenr,
2589 			       l->block_ref_to->dev_state->name,
2590 			       (unsigned long long)l->block_ref_to->dev_bytenr,
2591 			       l->block_ref_to->mirror_num);
2592 			ret = -1;
2593 		} else if (!l->block_ref_to->is_iodone) {
2594 			printk(KERN_INFO "btrfs: attempt to write superblock"
2595 			       " which references block %c @%llu (%s/%llu/%d)"
2596 			       " which is not yet iodone!\n",
2597 			       btrfsic_get_block_type(state, l->block_ref_to),
2598 			       (unsigned long long)
2599 			       l->block_ref_to->logical_bytenr,
2600 			       l->block_ref_to->dev_state->name,
2601 			       (unsigned long long)l->block_ref_to->dev_bytenr,
2602 			       l->block_ref_to->mirror_num);
2603 			ret = -1;
2604 		} else if (l->parent_generation !=
2605 			   l->block_ref_to->generation &&
2606 			   BTRFSIC_GENERATION_UNKNOWN !=
2607 			   l->parent_generation &&
2608 			   BTRFSIC_GENERATION_UNKNOWN !=
2609 			   l->block_ref_to->generation) {
2610 			printk(KERN_INFO "btrfs: attempt to write superblock"
2611 			       " which references block %c @%llu (%s/%llu/%d)"
2612 			       " with generation %llu !="
2613 			       " parent generation %llu!\n",
2614 			       btrfsic_get_block_type(state, l->block_ref_to),
2615 			       (unsigned long long)
2616 			       l->block_ref_to->logical_bytenr,
2617 			       l->block_ref_to->dev_state->name,
2618 			       (unsigned long long)l->block_ref_to->dev_bytenr,
2619 			       l->block_ref_to->mirror_num,
2620 			       (unsigned long long)l->block_ref_to->generation,
2621 			       (unsigned long long)l->parent_generation);
2622 			ret = -1;
2623 		} else if (l->block_ref_to->flush_gen >
2624 			   l->block_ref_to->dev_state->last_flush_gen) {
2625 			printk(KERN_INFO "btrfs: attempt to write superblock"
2626 			       " which references block %c @%llu (%s/%llu/%d)"
2627 			       " which is not flushed out of disk's write cache"
2628 			       " (block flush_gen=%llu,"
2629 			       " dev->flush_gen=%llu)!\n",
2630 			       btrfsic_get_block_type(state, l->block_ref_to),
2631 			       (unsigned long long)
2632 			       l->block_ref_to->logical_bytenr,
2633 			       l->block_ref_to->dev_state->name,
2634 			       (unsigned long long)l->block_ref_to->dev_bytenr,
2635 			       l->block_ref_to->mirror_num,
2636 			       (unsigned long long)block->flush_gen,
2637 			       (unsigned long long)
2638 			       l->block_ref_to->dev_state->last_flush_gen);
2639 			ret = -1;
2640 		} else if (-1 == btrfsic_check_all_ref_blocks(state,
2641 							      l->block_ref_to,
2642 							      recursion_level +
2643 							      1)) {
2644 			ret = -1;
2645 		}
2646 	}
2647 
2648 	return ret;
2649 }
2650 
2651 static int btrfsic_is_block_ref_by_superblock(
2652 		const struct btrfsic_state *state,
2653 		const struct btrfsic_block *block,
2654 		int recursion_level)
2655 {
2656 	struct list_head *elem_ref_from;
2657 
2658 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2659 		/* refer to comment at "abort cyclic linkage (case 1)" */
2660 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2661 			printk(KERN_INFO
2662 			       "btrfsic: abort cyclic linkage (case 2).\n");
2663 
2664 		return 0;
2665 	}
2666 
2667 	/*
2668 	 * This algorithm is recursive because the amount of used stack space
2669 	 * is very small and the max recursion depth is limited.
2670 	 */
2671 	list_for_each(elem_ref_from, &block->ref_from_list) {
2672 		const struct btrfsic_block_link *const l =
2673 		    list_entry(elem_ref_from, struct btrfsic_block_link,
2674 			       node_ref_from);
2675 
2676 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2677 			printk(KERN_INFO
2678 			       "rl=%d, %c @%llu (%s/%llu/%d)"
2679 			       " is ref %u* from %c @%llu (%s/%llu/%d)\n",
2680 			       recursion_level,
2681 			       btrfsic_get_block_type(state, block),
2682 			       (unsigned long long)block->logical_bytenr,
2683 			       block->dev_state->name,
2684 			       (unsigned long long)block->dev_bytenr,
2685 			       block->mirror_num,
2686 			       l->ref_cnt,
2687 			       btrfsic_get_block_type(state, l->block_ref_from),
2688 			       (unsigned long long)
2689 			       l->block_ref_from->logical_bytenr,
2690 			       l->block_ref_from->dev_state->name,
2691 			       (unsigned long long)
2692 			       l->block_ref_from->dev_bytenr,
2693 			       l->block_ref_from->mirror_num);
2694 		if (l->block_ref_from->is_superblock &&
2695 		    state->latest_superblock->dev_bytenr ==
2696 		    l->block_ref_from->dev_bytenr &&
2697 		    state->latest_superblock->dev_state->bdev ==
2698 		    l->block_ref_from->dev_state->bdev)
2699 			return 1;
2700 		else if (btrfsic_is_block_ref_by_superblock(state,
2701 							    l->block_ref_from,
2702 							    recursion_level +
2703 							    1))
2704 			return 1;
2705 	}
2706 
2707 	return 0;
2708 }
2709 
2710 static void btrfsic_print_add_link(const struct btrfsic_state *state,
2711 				   const struct btrfsic_block_link *l)
2712 {
2713 	printk(KERN_INFO
2714 	       "Add %u* link from %c @%llu (%s/%llu/%d)"
2715 	       " to %c @%llu (%s/%llu/%d).\n",
2716 	       l->ref_cnt,
2717 	       btrfsic_get_block_type(state, l->block_ref_from),
2718 	       (unsigned long long)l->block_ref_from->logical_bytenr,
2719 	       l->block_ref_from->dev_state->name,
2720 	       (unsigned long long)l->block_ref_from->dev_bytenr,
2721 	       l->block_ref_from->mirror_num,
2722 	       btrfsic_get_block_type(state, l->block_ref_to),
2723 	       (unsigned long long)l->block_ref_to->logical_bytenr,
2724 	       l->block_ref_to->dev_state->name,
2725 	       (unsigned long long)l->block_ref_to->dev_bytenr,
2726 	       l->block_ref_to->mirror_num);
2727 }
2728 
2729 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2730 				   const struct btrfsic_block_link *l)
2731 {
2732 	printk(KERN_INFO
2733 	       "Rem %u* link from %c @%llu (%s/%llu/%d)"
2734 	       " to %c @%llu (%s/%llu/%d).\n",
2735 	       l->ref_cnt,
2736 	       btrfsic_get_block_type(state, l->block_ref_from),
2737 	       (unsigned long long)l->block_ref_from->logical_bytenr,
2738 	       l->block_ref_from->dev_state->name,
2739 	       (unsigned long long)l->block_ref_from->dev_bytenr,
2740 	       l->block_ref_from->mirror_num,
2741 	       btrfsic_get_block_type(state, l->block_ref_to),
2742 	       (unsigned long long)l->block_ref_to->logical_bytenr,
2743 	       l->block_ref_to->dev_state->name,
2744 	       (unsigned long long)l->block_ref_to->dev_bytenr,
2745 	       l->block_ref_to->mirror_num);
2746 }
2747 
2748 static char btrfsic_get_block_type(const struct btrfsic_state *state,
2749 				   const struct btrfsic_block *block)
2750 {
2751 	if (block->is_superblock &&
2752 	    state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2753 	    state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2754 		return 'S';
2755 	else if (block->is_superblock)
2756 		return 's';
2757 	else if (block->is_metadata)
2758 		return 'M';
2759 	else
2760 		return 'D';
2761 }
2762 
2763 static void btrfsic_dump_tree(const struct btrfsic_state *state)
2764 {
2765 	btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2766 }
2767 
2768 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2769 				  const struct btrfsic_block *block,
2770 				  int indent_level)
2771 {
2772 	struct list_head *elem_ref_to;
2773 	int indent_add;
2774 	static char buf[80];
2775 	int cursor_position;
2776 
2777 	/*
2778 	 * Should better fill an on-stack buffer with a complete line and
2779 	 * dump it at once when it is time to print a newline character.
2780 	 */
2781 
2782 	/*
2783 	 * This algorithm is recursive because the amount of used stack space
2784 	 * is very small and the max recursion depth is limited.
2785 	 */
2786 	indent_add = sprintf(buf, "%c-%llu(%s/%llu/%d)",
2787 			     btrfsic_get_block_type(state, block),
2788 			     (unsigned long long)block->logical_bytenr,
2789 			     block->dev_state->name,
2790 			     (unsigned long long)block->dev_bytenr,
2791 			     block->mirror_num);
2792 	if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2793 		printk("[...]\n");
2794 		return;
2795 	}
2796 	printk(buf);
2797 	indent_level += indent_add;
2798 	if (list_empty(&block->ref_to_list)) {
2799 		printk("\n");
2800 		return;
2801 	}
2802 	if (block->mirror_num > 1 &&
2803 	    !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2804 		printk(" [...]\n");
2805 		return;
2806 	}
2807 
2808 	cursor_position = indent_level;
2809 	list_for_each(elem_ref_to, &block->ref_to_list) {
2810 		const struct btrfsic_block_link *const l =
2811 		    list_entry(elem_ref_to, struct btrfsic_block_link,
2812 			       node_ref_to);
2813 
2814 		while (cursor_position < indent_level) {
2815 			printk(" ");
2816 			cursor_position++;
2817 		}
2818 		if (l->ref_cnt > 1)
2819 			indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2820 		else
2821 			indent_add = sprintf(buf, " --> ");
2822 		if (indent_level + indent_add >
2823 		    BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2824 			printk("[...]\n");
2825 			cursor_position = 0;
2826 			continue;
2827 		}
2828 
2829 		printk(buf);
2830 
2831 		btrfsic_dump_tree_sub(state, l->block_ref_to,
2832 				      indent_level + indent_add);
2833 		cursor_position = 0;
2834 	}
2835 }
2836 
2837 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2838 		struct btrfsic_state *state,
2839 		struct btrfsic_block_data_ctx *next_block_ctx,
2840 		struct btrfsic_block *next_block,
2841 		struct btrfsic_block *from_block,
2842 		u64 parent_generation)
2843 {
2844 	struct btrfsic_block_link *l;
2845 
2846 	l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2847 						next_block_ctx->dev_bytenr,
2848 						from_block->dev_state->bdev,
2849 						from_block->dev_bytenr,
2850 						&state->block_link_hashtable);
2851 	if (NULL == l) {
2852 		l = btrfsic_block_link_alloc();
2853 		if (NULL == l) {
2854 			printk(KERN_INFO
2855 			       "btrfsic: error, kmalloc" " failed!\n");
2856 			return NULL;
2857 		}
2858 
2859 		l->block_ref_to = next_block;
2860 		l->block_ref_from = from_block;
2861 		l->ref_cnt = 1;
2862 		l->parent_generation = parent_generation;
2863 
2864 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2865 			btrfsic_print_add_link(state, l);
2866 
2867 		list_add(&l->node_ref_to, &from_block->ref_to_list);
2868 		list_add(&l->node_ref_from, &next_block->ref_from_list);
2869 
2870 		btrfsic_block_link_hashtable_add(l,
2871 						 &state->block_link_hashtable);
2872 	} else {
2873 		l->ref_cnt++;
2874 		l->parent_generation = parent_generation;
2875 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2876 			btrfsic_print_add_link(state, l);
2877 	}
2878 
2879 	return l;
2880 }
2881 
2882 static struct btrfsic_block *btrfsic_block_lookup_or_add(
2883 		struct btrfsic_state *state,
2884 		struct btrfsic_block_data_ctx *block_ctx,
2885 		const char *additional_string,
2886 		int is_metadata,
2887 		int is_iodone,
2888 		int never_written,
2889 		int mirror_num,
2890 		int *was_created)
2891 {
2892 	struct btrfsic_block *block;
2893 
2894 	block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2895 					       block_ctx->dev_bytenr,
2896 					       &state->block_hashtable);
2897 	if (NULL == block) {
2898 		struct btrfsic_dev_state *dev_state;
2899 
2900 		block = btrfsic_block_alloc();
2901 		if (NULL == block) {
2902 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
2903 			return NULL;
2904 		}
2905 		dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev);
2906 		if (NULL == dev_state) {
2907 			printk(KERN_INFO
2908 			       "btrfsic: error, lookup dev_state failed!\n");
2909 			btrfsic_block_free(block);
2910 			return NULL;
2911 		}
2912 		block->dev_state = dev_state;
2913 		block->dev_bytenr = block_ctx->dev_bytenr;
2914 		block->logical_bytenr = block_ctx->start;
2915 		block->is_metadata = is_metadata;
2916 		block->is_iodone = is_iodone;
2917 		block->never_written = never_written;
2918 		block->mirror_num = mirror_num;
2919 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2920 			printk(KERN_INFO
2921 			       "New %s%c-block @%llu (%s/%llu/%d)\n",
2922 			       additional_string,
2923 			       btrfsic_get_block_type(state, block),
2924 			       (unsigned long long)block->logical_bytenr,
2925 			       dev_state->name,
2926 			       (unsigned long long)block->dev_bytenr,
2927 			       mirror_num);
2928 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2929 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2930 		if (NULL != was_created)
2931 			*was_created = 1;
2932 	} else {
2933 		if (NULL != was_created)
2934 			*was_created = 0;
2935 	}
2936 
2937 	return block;
2938 }
2939 
2940 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2941 					   u64 bytenr,
2942 					   struct btrfsic_dev_state *dev_state,
2943 					   u64 dev_bytenr)
2944 {
2945 	int num_copies;
2946 	int mirror_num;
2947 	int ret;
2948 	struct btrfsic_block_data_ctx block_ctx;
2949 	int match = 0;
2950 
2951 	num_copies = btrfs_num_copies(&state->root->fs_info->mapping_tree,
2952 				      bytenr, state->metablock_size);
2953 
2954 	for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2955 		ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2956 					&block_ctx, mirror_num);
2957 		if (ret) {
2958 			printk(KERN_INFO "btrfsic:"
2959 			       " btrfsic_map_block(logical @%llu,"
2960 			       " mirror %d) failed!\n",
2961 			       (unsigned long long)bytenr, mirror_num);
2962 			continue;
2963 		}
2964 
2965 		if (dev_state->bdev == block_ctx.dev->bdev &&
2966 		    dev_bytenr == block_ctx.dev_bytenr) {
2967 			match++;
2968 			btrfsic_release_block_ctx(&block_ctx);
2969 			break;
2970 		}
2971 		btrfsic_release_block_ctx(&block_ctx);
2972 	}
2973 
2974 	if (!match) {
2975 		printk(KERN_INFO "btrfs: attempt to write M-block which contains logical bytenr that doesn't map to dev+physical bytenr of submit_bio,"
2976 		       " buffer->log_bytenr=%llu, submit_bio(bdev=%s,"
2977 		       " phys_bytenr=%llu)!\n",
2978 		       (unsigned long long)bytenr, dev_state->name,
2979 		       (unsigned long long)dev_bytenr);
2980 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2981 			ret = btrfsic_map_block(state, bytenr,
2982 						state->metablock_size,
2983 						&block_ctx, mirror_num);
2984 			if (ret)
2985 				continue;
2986 
2987 			printk(KERN_INFO "Read logical bytenr @%llu maps to"
2988 			       " (%s/%llu/%d)\n",
2989 			       (unsigned long long)bytenr,
2990 			       block_ctx.dev->name,
2991 			       (unsigned long long)block_ctx.dev_bytenr,
2992 			       mirror_num);
2993 		}
2994 		WARN_ON(1);
2995 	}
2996 }
2997 
2998 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
2999 		struct block_device *bdev)
3000 {
3001 	struct btrfsic_dev_state *ds;
3002 
3003 	ds = btrfsic_dev_state_hashtable_lookup(bdev,
3004 						&btrfsic_dev_state_hashtable);
3005 	return ds;
3006 }
3007 
3008 int btrfsic_submit_bh(int rw, struct buffer_head *bh)
3009 {
3010 	struct btrfsic_dev_state *dev_state;
3011 
3012 	if (!btrfsic_is_initialized)
3013 		return submit_bh(rw, bh);
3014 
3015 	mutex_lock(&btrfsic_mutex);
3016 	/* since btrfsic_submit_bh() might also be called before
3017 	 * btrfsic_mount(), this might return NULL */
3018 	dev_state = btrfsic_dev_state_lookup(bh->b_bdev);
3019 
3020 	/* Only called to write the superblock (incl. FLUSH/FUA) */
3021 	if (NULL != dev_state &&
3022 	    (rw & WRITE) && bh->b_size > 0) {
3023 		u64 dev_bytenr;
3024 
3025 		dev_bytenr = 4096 * bh->b_blocknr;
3026 		if (dev_state->state->print_mask &
3027 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
3028 			printk(KERN_INFO
3029 			       "submit_bh(rw=0x%x, blocknr=%lu (bytenr %llu),"
3030 			       " size=%lu, data=%p, bdev=%p)\n",
3031 			       rw, (unsigned long)bh->b_blocknr,
3032 			       (unsigned long long)dev_bytenr,
3033 			       (unsigned long)bh->b_size, bh->b_data,
3034 			       bh->b_bdev);
3035 		btrfsic_process_written_block(dev_state, dev_bytenr,
3036 					      &bh->b_data, 1, NULL,
3037 					      NULL, bh, rw);
3038 	} else if (NULL != dev_state && (rw & REQ_FLUSH)) {
3039 		if (dev_state->state->print_mask &
3040 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
3041 			printk(KERN_INFO
3042 			       "submit_bh(rw=0x%x FLUSH, bdev=%p)\n",
3043 			       rw, bh->b_bdev);
3044 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
3045 			if ((dev_state->state->print_mask &
3046 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
3047 			      BTRFSIC_PRINT_MASK_VERBOSE)))
3048 				printk(KERN_INFO
3049 				       "btrfsic_submit_bh(%s) with FLUSH"
3050 				       " but dummy block already in use"
3051 				       " (ignored)!\n",
3052 				       dev_state->name);
3053 		} else {
3054 			struct btrfsic_block *const block =
3055 				&dev_state->dummy_block_for_bio_bh_flush;
3056 
3057 			block->is_iodone = 0;
3058 			block->never_written = 0;
3059 			block->iodone_w_error = 0;
3060 			block->flush_gen = dev_state->last_flush_gen + 1;
3061 			block->submit_bio_bh_rw = rw;
3062 			block->orig_bio_bh_private = bh->b_private;
3063 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
3064 			block->next_in_same_bio = NULL;
3065 			bh->b_private = block;
3066 			bh->b_end_io = btrfsic_bh_end_io;
3067 		}
3068 	}
3069 	mutex_unlock(&btrfsic_mutex);
3070 	return submit_bh(rw, bh);
3071 }
3072 
3073 void btrfsic_submit_bio(int rw, struct bio *bio)
3074 {
3075 	struct btrfsic_dev_state *dev_state;
3076 
3077 	if (!btrfsic_is_initialized) {
3078 		submit_bio(rw, bio);
3079 		return;
3080 	}
3081 
3082 	mutex_lock(&btrfsic_mutex);
3083 	/* since btrfsic_submit_bio() is also called before
3084 	 * btrfsic_mount(), this might return NULL */
3085 	dev_state = btrfsic_dev_state_lookup(bio->bi_bdev);
3086 	if (NULL != dev_state &&
3087 	    (rw & WRITE) && NULL != bio->bi_io_vec) {
3088 		unsigned int i;
3089 		u64 dev_bytenr;
3090 		int bio_is_patched;
3091 		char **mapped_datav;
3092 
3093 		dev_bytenr = 512 * bio->bi_sector;
3094 		bio_is_patched = 0;
3095 		if (dev_state->state->print_mask &
3096 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
3097 			printk(KERN_INFO
3098 			       "submit_bio(rw=0x%x, bi_vcnt=%u,"
3099 			       " bi_sector=%lu (bytenr %llu), bi_bdev=%p)\n",
3100 			       rw, bio->bi_vcnt, (unsigned long)bio->bi_sector,
3101 			       (unsigned long long)dev_bytenr,
3102 			       bio->bi_bdev);
3103 
3104 		mapped_datav = kmalloc(sizeof(*mapped_datav) * bio->bi_vcnt,
3105 				       GFP_NOFS);
3106 		if (!mapped_datav)
3107 			goto leave;
3108 		for (i = 0; i < bio->bi_vcnt; i++) {
3109 			BUG_ON(bio->bi_io_vec[i].bv_len != PAGE_CACHE_SIZE);
3110 			mapped_datav[i] = kmap(bio->bi_io_vec[i].bv_page);
3111 			if (!mapped_datav[i]) {
3112 				while (i > 0) {
3113 					i--;
3114 					kunmap(bio->bi_io_vec[i].bv_page);
3115 				}
3116 				kfree(mapped_datav);
3117 				goto leave;
3118 			}
3119 			if ((BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
3120 			     BTRFSIC_PRINT_MASK_VERBOSE) ==
3121 			    (dev_state->state->print_mask &
3122 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
3123 			      BTRFSIC_PRINT_MASK_VERBOSE)))
3124 				printk(KERN_INFO
3125 				       "#%u: page=%p, len=%u, offset=%u\n",
3126 				       i, bio->bi_io_vec[i].bv_page,
3127 				       bio->bi_io_vec[i].bv_len,
3128 				       bio->bi_io_vec[i].bv_offset);
3129 		}
3130 		btrfsic_process_written_block(dev_state, dev_bytenr,
3131 					      mapped_datav, bio->bi_vcnt,
3132 					      bio, &bio_is_patched,
3133 					      NULL, rw);
3134 		while (i > 0) {
3135 			i--;
3136 			kunmap(bio->bi_io_vec[i].bv_page);
3137 		}
3138 		kfree(mapped_datav);
3139 	} else if (NULL != dev_state && (rw & REQ_FLUSH)) {
3140 		if (dev_state->state->print_mask &
3141 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
3142 			printk(KERN_INFO
3143 			       "submit_bio(rw=0x%x FLUSH, bdev=%p)\n",
3144 			       rw, bio->bi_bdev);
3145 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
3146 			if ((dev_state->state->print_mask &
3147 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
3148 			      BTRFSIC_PRINT_MASK_VERBOSE)))
3149 				printk(KERN_INFO
3150 				       "btrfsic_submit_bio(%s) with FLUSH"
3151 				       " but dummy block already in use"
3152 				       " (ignored)!\n",
3153 				       dev_state->name);
3154 		} else {
3155 			struct btrfsic_block *const block =
3156 				&dev_state->dummy_block_for_bio_bh_flush;
3157 
3158 			block->is_iodone = 0;
3159 			block->never_written = 0;
3160 			block->iodone_w_error = 0;
3161 			block->flush_gen = dev_state->last_flush_gen + 1;
3162 			block->submit_bio_bh_rw = rw;
3163 			block->orig_bio_bh_private = bio->bi_private;
3164 			block->orig_bio_bh_end_io.bio = bio->bi_end_io;
3165 			block->next_in_same_bio = NULL;
3166 			bio->bi_private = block;
3167 			bio->bi_end_io = btrfsic_bio_end_io;
3168 		}
3169 	}
3170 leave:
3171 	mutex_unlock(&btrfsic_mutex);
3172 
3173 	submit_bio(rw, bio);
3174 }
3175 
3176 int btrfsic_mount(struct btrfs_root *root,
3177 		  struct btrfs_fs_devices *fs_devices,
3178 		  int including_extent_data, u32 print_mask)
3179 {
3180 	int ret;
3181 	struct btrfsic_state *state;
3182 	struct list_head *dev_head = &fs_devices->devices;
3183 	struct btrfs_device *device;
3184 
3185 	if (root->nodesize != root->leafsize) {
3186 		printk(KERN_INFO
3187 		       "btrfsic: cannot handle nodesize %d != leafsize %d!\n",
3188 		       root->nodesize, root->leafsize);
3189 		return -1;
3190 	}
3191 	if (root->nodesize & ((u64)PAGE_CACHE_SIZE - 1)) {
3192 		printk(KERN_INFO
3193 		       "btrfsic: cannot handle nodesize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3194 		       root->nodesize, (unsigned long)PAGE_CACHE_SIZE);
3195 		return -1;
3196 	}
3197 	if (root->leafsize & ((u64)PAGE_CACHE_SIZE - 1)) {
3198 		printk(KERN_INFO
3199 		       "btrfsic: cannot handle leafsize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3200 		       root->leafsize, (unsigned long)PAGE_CACHE_SIZE);
3201 		return -1;
3202 	}
3203 	if (root->sectorsize & ((u64)PAGE_CACHE_SIZE - 1)) {
3204 		printk(KERN_INFO
3205 		       "btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3206 		       root->sectorsize, (unsigned long)PAGE_CACHE_SIZE);
3207 		return -1;
3208 	}
3209 	state = kzalloc(sizeof(*state), GFP_NOFS);
3210 	if (NULL == state) {
3211 		printk(KERN_INFO "btrfs check-integrity: kmalloc() failed!\n");
3212 		return -1;
3213 	}
3214 
3215 	if (!btrfsic_is_initialized) {
3216 		mutex_init(&btrfsic_mutex);
3217 		btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
3218 		btrfsic_is_initialized = 1;
3219 	}
3220 	mutex_lock(&btrfsic_mutex);
3221 	state->root = root;
3222 	state->print_mask = print_mask;
3223 	state->include_extent_data = including_extent_data;
3224 	state->csum_size = 0;
3225 	state->metablock_size = root->nodesize;
3226 	state->datablock_size = root->sectorsize;
3227 	INIT_LIST_HEAD(&state->all_blocks_list);
3228 	btrfsic_block_hashtable_init(&state->block_hashtable);
3229 	btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
3230 	state->max_superblock_generation = 0;
3231 	state->latest_superblock = NULL;
3232 
3233 	list_for_each_entry(device, dev_head, dev_list) {
3234 		struct btrfsic_dev_state *ds;
3235 		char *p;
3236 
3237 		if (!device->bdev || !device->name)
3238 			continue;
3239 
3240 		ds = btrfsic_dev_state_alloc();
3241 		if (NULL == ds) {
3242 			printk(KERN_INFO
3243 			       "btrfs check-integrity: kmalloc() failed!\n");
3244 			mutex_unlock(&btrfsic_mutex);
3245 			return -1;
3246 		}
3247 		ds->bdev = device->bdev;
3248 		ds->state = state;
3249 		bdevname(ds->bdev, ds->name);
3250 		ds->name[BDEVNAME_SIZE - 1] = '\0';
3251 		for (p = ds->name; *p != '\0'; p++);
3252 		while (p > ds->name && *p != '/')
3253 			p--;
3254 		if (*p == '/')
3255 			p++;
3256 		strlcpy(ds->name, p, sizeof(ds->name));
3257 		btrfsic_dev_state_hashtable_add(ds,
3258 						&btrfsic_dev_state_hashtable);
3259 	}
3260 
3261 	ret = btrfsic_process_superblock(state, fs_devices);
3262 	if (0 != ret) {
3263 		mutex_unlock(&btrfsic_mutex);
3264 		btrfsic_unmount(root, fs_devices);
3265 		return ret;
3266 	}
3267 
3268 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
3269 		btrfsic_dump_database(state);
3270 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
3271 		btrfsic_dump_tree(state);
3272 
3273 	mutex_unlock(&btrfsic_mutex);
3274 	return 0;
3275 }
3276 
3277 void btrfsic_unmount(struct btrfs_root *root,
3278 		     struct btrfs_fs_devices *fs_devices)
3279 {
3280 	struct list_head *elem_all;
3281 	struct list_head *tmp_all;
3282 	struct btrfsic_state *state;
3283 	struct list_head *dev_head = &fs_devices->devices;
3284 	struct btrfs_device *device;
3285 
3286 	if (!btrfsic_is_initialized)
3287 		return;
3288 
3289 	mutex_lock(&btrfsic_mutex);
3290 
3291 	state = NULL;
3292 	list_for_each_entry(device, dev_head, dev_list) {
3293 		struct btrfsic_dev_state *ds;
3294 
3295 		if (!device->bdev || !device->name)
3296 			continue;
3297 
3298 		ds = btrfsic_dev_state_hashtable_lookup(
3299 				device->bdev,
3300 				&btrfsic_dev_state_hashtable);
3301 		if (NULL != ds) {
3302 			state = ds->state;
3303 			btrfsic_dev_state_hashtable_remove(ds);
3304 			btrfsic_dev_state_free(ds);
3305 		}
3306 	}
3307 
3308 	if (NULL == state) {
3309 		printk(KERN_INFO
3310 		       "btrfsic: error, cannot find state information"
3311 		       " on umount!\n");
3312 		mutex_unlock(&btrfsic_mutex);
3313 		return;
3314 	}
3315 
3316 	/*
3317 	 * Don't care about keeping the lists' state up to date,
3318 	 * just free all memory that was allocated dynamically.
3319 	 * Free the blocks and the block_links.
3320 	 */
3321 	list_for_each_safe(elem_all, tmp_all, &state->all_blocks_list) {
3322 		struct btrfsic_block *const b_all =
3323 		    list_entry(elem_all, struct btrfsic_block,
3324 			       all_blocks_node);
3325 		struct list_head *elem_ref_to;
3326 		struct list_head *tmp_ref_to;
3327 
3328 		list_for_each_safe(elem_ref_to, tmp_ref_to,
3329 				   &b_all->ref_to_list) {
3330 			struct btrfsic_block_link *const l =
3331 			    list_entry(elem_ref_to,
3332 				       struct btrfsic_block_link,
3333 				       node_ref_to);
3334 
3335 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
3336 				btrfsic_print_rem_link(state, l);
3337 
3338 			l->ref_cnt--;
3339 			if (0 == l->ref_cnt)
3340 				btrfsic_block_link_free(l);
3341 		}
3342 
3343 		if (b_all->is_iodone || b_all->never_written)
3344 			btrfsic_block_free(b_all);
3345 		else
3346 			printk(KERN_INFO "btrfs: attempt to free %c-block"
3347 			       " @%llu (%s/%llu/%d) on umount which is"
3348 			       " not yet iodone!\n",
3349 			       btrfsic_get_block_type(state, b_all),
3350 			       (unsigned long long)b_all->logical_bytenr,
3351 			       b_all->dev_state->name,
3352 			       (unsigned long long)b_all->dev_bytenr,
3353 			       b_all->mirror_num);
3354 	}
3355 
3356 	mutex_unlock(&btrfsic_mutex);
3357 
3358 	kfree(state);
3359 }
3360