xref: /openbmc/linux/fs/btrfs/check-integrity.c (revision a06c488d)
1 /*
2  * Copyright (C) STRATO AG 2011.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 /*
20  * This module can be used to catch cases when the btrfs kernel
21  * code executes write requests to the disk that bring the file
22  * system in an inconsistent state. In such a state, a power-loss
23  * or kernel panic event would cause that the data on disk is
24  * lost or at least damaged.
25  *
26  * Code is added that examines all block write requests during
27  * runtime (including writes of the super block). Three rules
28  * are verified and an error is printed on violation of the
29  * rules:
30  * 1. It is not allowed to write a disk block which is
31  *    currently referenced by the super block (either directly
32  *    or indirectly).
33  * 2. When a super block is written, it is verified that all
34  *    referenced (directly or indirectly) blocks fulfill the
35  *    following requirements:
36  *    2a. All referenced blocks have either been present when
37  *        the file system was mounted, (i.e., they have been
38  *        referenced by the super block) or they have been
39  *        written since then and the write completion callback
40  *        was called and no write error was indicated and a
41  *        FLUSH request to the device where these blocks are
42  *        located was received and completed.
43  *    2b. All referenced blocks need to have a generation
44  *        number which is equal to the parent's number.
45  *
46  * One issue that was found using this module was that the log
47  * tree on disk became temporarily corrupted because disk blocks
48  * that had been in use for the log tree had been freed and
49  * reused too early, while being referenced by the written super
50  * block.
51  *
52  * The search term in the kernel log that can be used to filter
53  * on the existence of detected integrity issues is
54  * "btrfs: attempt".
55  *
56  * The integrity check is enabled via mount options. These
57  * mount options are only supported if the integrity check
58  * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY.
59  *
60  * Example #1, apply integrity checks to all metadata:
61  * mount /dev/sdb1 /mnt -o check_int
62  *
63  * Example #2, apply integrity checks to all metadata and
64  * to data extents:
65  * mount /dev/sdb1 /mnt -o check_int_data
66  *
67  * Example #3, apply integrity checks to all metadata and dump
68  * the tree that the super block references to kernel messages
69  * each time after a super block was written:
70  * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263
71  *
72  * If the integrity check tool is included and activated in
73  * the mount options, plenty of kernel memory is used, and
74  * plenty of additional CPU cycles are spent. Enabling this
75  * functionality is not intended for normal use. In most
76  * cases, unless you are a btrfs developer who needs to verify
77  * the integrity of (super)-block write requests, do not
78  * enable the config option BTRFS_FS_CHECK_INTEGRITY to
79  * include and compile the integrity check tool.
80  *
81  * Expect millions of lines of information in the kernel log with an
82  * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the
83  * kernel config to at least 26 (which is 64MB). Usually the value is
84  * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be
85  * changed like this before LOG_BUF_SHIFT can be set to a high value:
86  * config LOG_BUF_SHIFT
87  *       int "Kernel log buffer size (16 => 64KB, 17 => 128KB)"
88  *       range 12 30
89  */
90 
91 #include <linux/sched.h>
92 #include <linux/slab.h>
93 #include <linux/buffer_head.h>
94 #include <linux/mutex.h>
95 #include <linux/genhd.h>
96 #include <linux/blkdev.h>
97 #include <linux/vmalloc.h>
98 #include "ctree.h"
99 #include "disk-io.h"
100 #include "hash.h"
101 #include "transaction.h"
102 #include "extent_io.h"
103 #include "volumes.h"
104 #include "print-tree.h"
105 #include "locking.h"
106 #include "check-integrity.h"
107 #include "rcu-string.h"
108 
109 #define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
110 #define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
111 #define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
112 #define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
113 #define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
114 #define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
115 #define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
116 #define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6)	/* in characters,
117 							 * excluding " [...]" */
118 #define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
119 
120 /*
121  * The definition of the bitmask fields for the print_mask.
122  * They are specified with the mount option check_integrity_print_mask.
123  */
124 #define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE			0x00000001
125 #define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION		0x00000002
126 #define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE			0x00000004
127 #define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE			0x00000008
128 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH			0x00000010
129 #define BTRFSIC_PRINT_MASK_END_IO_BIO_BH			0x00000020
130 #define BTRFSIC_PRINT_MASK_VERBOSE				0x00000040
131 #define BTRFSIC_PRINT_MASK_VERY_VERBOSE				0x00000080
132 #define BTRFSIC_PRINT_MASK_INITIAL_TREE				0x00000100
133 #define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES			0x00000200
134 #define BTRFSIC_PRINT_MASK_INITIAL_DATABASE			0x00000400
135 #define BTRFSIC_PRINT_MASK_NUM_COPIES				0x00000800
136 #define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS		0x00001000
137 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE		0x00002000
138 
139 struct btrfsic_dev_state;
140 struct btrfsic_state;
141 
142 struct btrfsic_block {
143 	u32 magic_num;		/* only used for debug purposes */
144 	unsigned int is_metadata:1;	/* if it is meta-data, not data-data */
145 	unsigned int is_superblock:1;	/* if it is one of the superblocks */
146 	unsigned int is_iodone:1;	/* if is done by lower subsystem */
147 	unsigned int iodone_w_error:1;	/* error was indicated to endio */
148 	unsigned int never_written:1;	/* block was added because it was
149 					 * referenced, not because it was
150 					 * written */
151 	unsigned int mirror_num;	/* large enough to hold
152 					 * BTRFS_SUPER_MIRROR_MAX */
153 	struct btrfsic_dev_state *dev_state;
154 	u64 dev_bytenr;		/* key, physical byte num on disk */
155 	u64 logical_bytenr;	/* logical byte num on disk */
156 	u64 generation;
157 	struct btrfs_disk_key disk_key;	/* extra info to print in case of
158 					 * issues, will not always be correct */
159 	struct list_head collision_resolving_node;	/* list node */
160 	struct list_head all_blocks_node;	/* list node */
161 
162 	/* the following two lists contain block_link items */
163 	struct list_head ref_to_list;	/* list */
164 	struct list_head ref_from_list;	/* list */
165 	struct btrfsic_block *next_in_same_bio;
166 	void *orig_bio_bh_private;
167 	union {
168 		bio_end_io_t *bio;
169 		bh_end_io_t *bh;
170 	} orig_bio_bh_end_io;
171 	int submit_bio_bh_rw;
172 	u64 flush_gen; /* only valid if !never_written */
173 };
174 
175 /*
176  * Elements of this type are allocated dynamically and required because
177  * each block object can refer to and can be ref from multiple blocks.
178  * The key to lookup them in the hashtable is the dev_bytenr of
179  * the block ref to plus the one from the block refered from.
180  * The fact that they are searchable via a hashtable and that a
181  * ref_cnt is maintained is not required for the btrfs integrity
182  * check algorithm itself, it is only used to make the output more
183  * beautiful in case that an error is detected (an error is defined
184  * as a write operation to a block while that block is still referenced).
185  */
186 struct btrfsic_block_link {
187 	u32 magic_num;		/* only used for debug purposes */
188 	u32 ref_cnt;
189 	struct list_head node_ref_to;	/* list node */
190 	struct list_head node_ref_from;	/* list node */
191 	struct list_head collision_resolving_node;	/* list node */
192 	struct btrfsic_block *block_ref_to;
193 	struct btrfsic_block *block_ref_from;
194 	u64 parent_generation;
195 };
196 
197 struct btrfsic_dev_state {
198 	u32 magic_num;		/* only used for debug purposes */
199 	struct block_device *bdev;
200 	struct btrfsic_state *state;
201 	struct list_head collision_resolving_node;	/* list node */
202 	struct btrfsic_block dummy_block_for_bio_bh_flush;
203 	u64 last_flush_gen;
204 	char name[BDEVNAME_SIZE];
205 };
206 
207 struct btrfsic_block_hashtable {
208 	struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
209 };
210 
211 struct btrfsic_block_link_hashtable {
212 	struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
213 };
214 
215 struct btrfsic_dev_state_hashtable {
216 	struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
217 };
218 
219 struct btrfsic_block_data_ctx {
220 	u64 start;		/* virtual bytenr */
221 	u64 dev_bytenr;		/* physical bytenr on device */
222 	u32 len;
223 	struct btrfsic_dev_state *dev;
224 	char **datav;
225 	struct page **pagev;
226 	void *mem_to_free;
227 };
228 
229 /* This structure is used to implement recursion without occupying
230  * any stack space, refer to btrfsic_process_metablock() */
231 struct btrfsic_stack_frame {
232 	u32 magic;
233 	u32 nr;
234 	int error;
235 	int i;
236 	int limit_nesting;
237 	int num_copies;
238 	int mirror_num;
239 	struct btrfsic_block *block;
240 	struct btrfsic_block_data_ctx *block_ctx;
241 	struct btrfsic_block *next_block;
242 	struct btrfsic_block_data_ctx next_block_ctx;
243 	struct btrfs_header *hdr;
244 	struct btrfsic_stack_frame *prev;
245 };
246 
247 /* Some state per mounted filesystem */
248 struct btrfsic_state {
249 	u32 print_mask;
250 	int include_extent_data;
251 	int csum_size;
252 	struct list_head all_blocks_list;
253 	struct btrfsic_block_hashtable block_hashtable;
254 	struct btrfsic_block_link_hashtable block_link_hashtable;
255 	struct btrfs_root *root;
256 	u64 max_superblock_generation;
257 	struct btrfsic_block *latest_superblock;
258 	u32 metablock_size;
259 	u32 datablock_size;
260 };
261 
262 static void btrfsic_block_init(struct btrfsic_block *b);
263 static struct btrfsic_block *btrfsic_block_alloc(void);
264 static void btrfsic_block_free(struct btrfsic_block *b);
265 static void btrfsic_block_link_init(struct btrfsic_block_link *n);
266 static struct btrfsic_block_link *btrfsic_block_link_alloc(void);
267 static void btrfsic_block_link_free(struct btrfsic_block_link *n);
268 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds);
269 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void);
270 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds);
271 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h);
272 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
273 					struct btrfsic_block_hashtable *h);
274 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b);
275 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
276 		struct block_device *bdev,
277 		u64 dev_bytenr,
278 		struct btrfsic_block_hashtable *h);
279 static void btrfsic_block_link_hashtable_init(
280 		struct btrfsic_block_link_hashtable *h);
281 static void btrfsic_block_link_hashtable_add(
282 		struct btrfsic_block_link *l,
283 		struct btrfsic_block_link_hashtable *h);
284 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l);
285 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
286 		struct block_device *bdev_ref_to,
287 		u64 dev_bytenr_ref_to,
288 		struct block_device *bdev_ref_from,
289 		u64 dev_bytenr_ref_from,
290 		struct btrfsic_block_link_hashtable *h);
291 static void btrfsic_dev_state_hashtable_init(
292 		struct btrfsic_dev_state_hashtable *h);
293 static void btrfsic_dev_state_hashtable_add(
294 		struct btrfsic_dev_state *ds,
295 		struct btrfsic_dev_state_hashtable *h);
296 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds);
297 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(
298 		struct block_device *bdev,
299 		struct btrfsic_dev_state_hashtable *h);
300 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void);
301 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf);
302 static int btrfsic_process_superblock(struct btrfsic_state *state,
303 				      struct btrfs_fs_devices *fs_devices);
304 static int btrfsic_process_metablock(struct btrfsic_state *state,
305 				     struct btrfsic_block *block,
306 				     struct btrfsic_block_data_ctx *block_ctx,
307 				     int limit_nesting, int force_iodone_flag);
308 static void btrfsic_read_from_block_data(
309 	struct btrfsic_block_data_ctx *block_ctx,
310 	void *dst, u32 offset, size_t len);
311 static int btrfsic_create_link_to_next_block(
312 		struct btrfsic_state *state,
313 		struct btrfsic_block *block,
314 		struct btrfsic_block_data_ctx
315 		*block_ctx, u64 next_bytenr,
316 		int limit_nesting,
317 		struct btrfsic_block_data_ctx *next_block_ctx,
318 		struct btrfsic_block **next_blockp,
319 		int force_iodone_flag,
320 		int *num_copiesp, int *mirror_nump,
321 		struct btrfs_disk_key *disk_key,
322 		u64 parent_generation);
323 static int btrfsic_handle_extent_data(struct btrfsic_state *state,
324 				      struct btrfsic_block *block,
325 				      struct btrfsic_block_data_ctx *block_ctx,
326 				      u32 item_offset, int force_iodone_flag);
327 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
328 			     struct btrfsic_block_data_ctx *block_ctx_out,
329 			     int mirror_num);
330 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
331 static int btrfsic_read_block(struct btrfsic_state *state,
332 			      struct btrfsic_block_data_ctx *block_ctx);
333 static void btrfsic_dump_database(struct btrfsic_state *state);
334 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
335 				     char **datav, unsigned int num_pages);
336 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
337 					  u64 dev_bytenr, char **mapped_datav,
338 					  unsigned int num_pages,
339 					  struct bio *bio, int *bio_is_patched,
340 					  struct buffer_head *bh,
341 					  int submit_bio_bh_rw);
342 static int btrfsic_process_written_superblock(
343 		struct btrfsic_state *state,
344 		struct btrfsic_block *const block,
345 		struct btrfs_super_block *const super_hdr);
346 static void btrfsic_bio_end_io(struct bio *bp);
347 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate);
348 static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
349 					      const struct btrfsic_block *block,
350 					      int recursion_level);
351 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
352 					struct btrfsic_block *const block,
353 					int recursion_level);
354 static void btrfsic_print_add_link(const struct btrfsic_state *state,
355 				   const struct btrfsic_block_link *l);
356 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
357 				   const struct btrfsic_block_link *l);
358 static char btrfsic_get_block_type(const struct btrfsic_state *state,
359 				   const struct btrfsic_block *block);
360 static void btrfsic_dump_tree(const struct btrfsic_state *state);
361 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
362 				  const struct btrfsic_block *block,
363 				  int indent_level);
364 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
365 		struct btrfsic_state *state,
366 		struct btrfsic_block_data_ctx *next_block_ctx,
367 		struct btrfsic_block *next_block,
368 		struct btrfsic_block *from_block,
369 		u64 parent_generation);
370 static struct btrfsic_block *btrfsic_block_lookup_or_add(
371 		struct btrfsic_state *state,
372 		struct btrfsic_block_data_ctx *block_ctx,
373 		const char *additional_string,
374 		int is_metadata,
375 		int is_iodone,
376 		int never_written,
377 		int mirror_num,
378 		int *was_created);
379 static int btrfsic_process_superblock_dev_mirror(
380 		struct btrfsic_state *state,
381 		struct btrfsic_dev_state *dev_state,
382 		struct btrfs_device *device,
383 		int superblock_mirror_num,
384 		struct btrfsic_dev_state **selected_dev_state,
385 		struct btrfs_super_block *selected_super);
386 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
387 		struct block_device *bdev);
388 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
389 					   u64 bytenr,
390 					   struct btrfsic_dev_state *dev_state,
391 					   u64 dev_bytenr);
392 
393 static struct mutex btrfsic_mutex;
394 static int btrfsic_is_initialized;
395 static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
396 
397 
398 static void btrfsic_block_init(struct btrfsic_block *b)
399 {
400 	b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
401 	b->dev_state = NULL;
402 	b->dev_bytenr = 0;
403 	b->logical_bytenr = 0;
404 	b->generation = BTRFSIC_GENERATION_UNKNOWN;
405 	b->disk_key.objectid = 0;
406 	b->disk_key.type = 0;
407 	b->disk_key.offset = 0;
408 	b->is_metadata = 0;
409 	b->is_superblock = 0;
410 	b->is_iodone = 0;
411 	b->iodone_w_error = 0;
412 	b->never_written = 0;
413 	b->mirror_num = 0;
414 	b->next_in_same_bio = NULL;
415 	b->orig_bio_bh_private = NULL;
416 	b->orig_bio_bh_end_io.bio = NULL;
417 	INIT_LIST_HEAD(&b->collision_resolving_node);
418 	INIT_LIST_HEAD(&b->all_blocks_node);
419 	INIT_LIST_HEAD(&b->ref_to_list);
420 	INIT_LIST_HEAD(&b->ref_from_list);
421 	b->submit_bio_bh_rw = 0;
422 	b->flush_gen = 0;
423 }
424 
425 static struct btrfsic_block *btrfsic_block_alloc(void)
426 {
427 	struct btrfsic_block *b;
428 
429 	b = kzalloc(sizeof(*b), GFP_NOFS);
430 	if (NULL != b)
431 		btrfsic_block_init(b);
432 
433 	return b;
434 }
435 
436 static void btrfsic_block_free(struct btrfsic_block *b)
437 {
438 	BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
439 	kfree(b);
440 }
441 
442 static void btrfsic_block_link_init(struct btrfsic_block_link *l)
443 {
444 	l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
445 	l->ref_cnt = 1;
446 	INIT_LIST_HEAD(&l->node_ref_to);
447 	INIT_LIST_HEAD(&l->node_ref_from);
448 	INIT_LIST_HEAD(&l->collision_resolving_node);
449 	l->block_ref_to = NULL;
450 	l->block_ref_from = NULL;
451 }
452 
453 static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
454 {
455 	struct btrfsic_block_link *l;
456 
457 	l = kzalloc(sizeof(*l), GFP_NOFS);
458 	if (NULL != l)
459 		btrfsic_block_link_init(l);
460 
461 	return l;
462 }
463 
464 static void btrfsic_block_link_free(struct btrfsic_block_link *l)
465 {
466 	BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
467 	kfree(l);
468 }
469 
470 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
471 {
472 	ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
473 	ds->bdev = NULL;
474 	ds->state = NULL;
475 	ds->name[0] = '\0';
476 	INIT_LIST_HEAD(&ds->collision_resolving_node);
477 	ds->last_flush_gen = 0;
478 	btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
479 	ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
480 	ds->dummy_block_for_bio_bh_flush.dev_state = ds;
481 }
482 
483 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
484 {
485 	struct btrfsic_dev_state *ds;
486 
487 	ds = kzalloc(sizeof(*ds), GFP_NOFS);
488 	if (NULL != ds)
489 		btrfsic_dev_state_init(ds);
490 
491 	return ds;
492 }
493 
494 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
495 {
496 	BUG_ON(!(NULL == ds ||
497 		 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
498 	kfree(ds);
499 }
500 
501 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
502 {
503 	int i;
504 
505 	for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
506 		INIT_LIST_HEAD(h->table + i);
507 }
508 
509 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
510 					struct btrfsic_block_hashtable *h)
511 {
512 	const unsigned int hashval =
513 	    (((unsigned int)(b->dev_bytenr >> 16)) ^
514 	     ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
515 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
516 
517 	list_add(&b->collision_resolving_node, h->table + hashval);
518 }
519 
520 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
521 {
522 	list_del(&b->collision_resolving_node);
523 }
524 
525 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
526 		struct block_device *bdev,
527 		u64 dev_bytenr,
528 		struct btrfsic_block_hashtable *h)
529 {
530 	const unsigned int hashval =
531 	    (((unsigned int)(dev_bytenr >> 16)) ^
532 	     ((unsigned int)((uintptr_t)bdev))) &
533 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
534 	struct btrfsic_block *b;
535 
536 	list_for_each_entry(b, h->table + hashval, collision_resolving_node) {
537 		if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
538 			return b;
539 	}
540 
541 	return NULL;
542 }
543 
544 static void btrfsic_block_link_hashtable_init(
545 		struct btrfsic_block_link_hashtable *h)
546 {
547 	int i;
548 
549 	for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
550 		INIT_LIST_HEAD(h->table + i);
551 }
552 
553 static void btrfsic_block_link_hashtable_add(
554 		struct btrfsic_block_link *l,
555 		struct btrfsic_block_link_hashtable *h)
556 {
557 	const unsigned int hashval =
558 	    (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
559 	     ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
560 	     ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
561 	     ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
562 	     & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
563 
564 	BUG_ON(NULL == l->block_ref_to);
565 	BUG_ON(NULL == l->block_ref_from);
566 	list_add(&l->collision_resolving_node, h->table + hashval);
567 }
568 
569 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
570 {
571 	list_del(&l->collision_resolving_node);
572 }
573 
574 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
575 		struct block_device *bdev_ref_to,
576 		u64 dev_bytenr_ref_to,
577 		struct block_device *bdev_ref_from,
578 		u64 dev_bytenr_ref_from,
579 		struct btrfsic_block_link_hashtable *h)
580 {
581 	const unsigned int hashval =
582 	    (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
583 	     ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
584 	     ((unsigned int)((uintptr_t)bdev_ref_to)) ^
585 	     ((unsigned int)((uintptr_t)bdev_ref_from))) &
586 	     (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
587 	struct btrfsic_block_link *l;
588 
589 	list_for_each_entry(l, h->table + hashval, collision_resolving_node) {
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 btrfsic_dev_state *ds;
635 
636 	list_for_each_entry(ds, h->table + hashval, collision_resolving_node) {
637 		if (ds->bdev == bdev)
638 			return ds;
639 	}
640 
641 	return NULL;
642 }
643 
644 static int btrfsic_process_superblock(struct btrfsic_state *state,
645 				      struct btrfs_fs_devices *fs_devices)
646 {
647 	int ret = 0;
648 	struct btrfs_super_block *selected_super;
649 	struct list_head *dev_head = &fs_devices->devices;
650 	struct btrfs_device *device;
651 	struct btrfsic_dev_state *selected_dev_state = NULL;
652 	int pass;
653 
654 	BUG_ON(NULL == state);
655 	selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
656 	if (NULL == selected_super) {
657 		printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
658 		return -ENOMEM;
659 	}
660 
661 	list_for_each_entry(device, dev_head, dev_list) {
662 		int i;
663 		struct btrfsic_dev_state *dev_state;
664 
665 		if (!device->bdev || !device->name)
666 			continue;
667 
668 		dev_state = btrfsic_dev_state_lookup(device->bdev);
669 		BUG_ON(NULL == dev_state);
670 		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
671 			ret = btrfsic_process_superblock_dev_mirror(
672 					state, dev_state, device, i,
673 					&selected_dev_state, selected_super);
674 			if (0 != ret && 0 == i) {
675 				kfree(selected_super);
676 				return ret;
677 			}
678 		}
679 	}
680 
681 	if (NULL == state->latest_superblock) {
682 		printk(KERN_INFO "btrfsic: no superblock found!\n");
683 		kfree(selected_super);
684 		return -1;
685 	}
686 
687 	state->csum_size = btrfs_super_csum_size(selected_super);
688 
689 	for (pass = 0; pass < 3; pass++) {
690 		int num_copies;
691 		int mirror_num;
692 		u64 next_bytenr;
693 
694 		switch (pass) {
695 		case 0:
696 			next_bytenr = btrfs_super_root(selected_super);
697 			if (state->print_mask &
698 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
699 				printk(KERN_INFO "root@%llu\n", next_bytenr);
700 			break;
701 		case 1:
702 			next_bytenr = btrfs_super_chunk_root(selected_super);
703 			if (state->print_mask &
704 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
705 				printk(KERN_INFO "chunk@%llu\n", next_bytenr);
706 			break;
707 		case 2:
708 			next_bytenr = btrfs_super_log_root(selected_super);
709 			if (0 == next_bytenr)
710 				continue;
711 			if (state->print_mask &
712 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
713 				printk(KERN_INFO "log@%llu\n", next_bytenr);
714 			break;
715 		}
716 
717 		num_copies =
718 		    btrfs_num_copies(state->root->fs_info,
719 				     next_bytenr, state->metablock_size);
720 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
721 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
722 			       next_bytenr, num_copies);
723 
724 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
725 			struct btrfsic_block *next_block;
726 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
727 			struct btrfsic_block_link *l;
728 
729 			ret = btrfsic_map_block(state, next_bytenr,
730 						state->metablock_size,
731 						&tmp_next_block_ctx,
732 						mirror_num);
733 			if (ret) {
734 				printk(KERN_INFO "btrfsic:"
735 				       " btrfsic_map_block(root @%llu,"
736 				       " mirror %d) failed!\n",
737 				       next_bytenr, mirror_num);
738 				kfree(selected_super);
739 				return -1;
740 			}
741 
742 			next_block = btrfsic_block_hashtable_lookup(
743 					tmp_next_block_ctx.dev->bdev,
744 					tmp_next_block_ctx.dev_bytenr,
745 					&state->block_hashtable);
746 			BUG_ON(NULL == next_block);
747 
748 			l = btrfsic_block_link_hashtable_lookup(
749 					tmp_next_block_ctx.dev->bdev,
750 					tmp_next_block_ctx.dev_bytenr,
751 					state->latest_superblock->dev_state->
752 					bdev,
753 					state->latest_superblock->dev_bytenr,
754 					&state->block_link_hashtable);
755 			BUG_ON(NULL == l);
756 
757 			ret = btrfsic_read_block(state, &tmp_next_block_ctx);
758 			if (ret < (int)PAGE_CACHE_SIZE) {
759 				printk(KERN_INFO
760 				       "btrfsic: read @logical %llu failed!\n",
761 				       tmp_next_block_ctx.start);
762 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
763 				kfree(selected_super);
764 				return -1;
765 			}
766 
767 			ret = btrfsic_process_metablock(state,
768 							next_block,
769 							&tmp_next_block_ctx,
770 							BTRFS_MAX_LEVEL + 3, 1);
771 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
772 		}
773 	}
774 
775 	kfree(selected_super);
776 	return ret;
777 }
778 
779 static int btrfsic_process_superblock_dev_mirror(
780 		struct btrfsic_state *state,
781 		struct btrfsic_dev_state *dev_state,
782 		struct btrfs_device *device,
783 		int superblock_mirror_num,
784 		struct btrfsic_dev_state **selected_dev_state,
785 		struct btrfs_super_block *selected_super)
786 {
787 	struct btrfs_super_block *super_tmp;
788 	u64 dev_bytenr;
789 	struct buffer_head *bh;
790 	struct btrfsic_block *superblock_tmp;
791 	int pass;
792 	struct block_device *const superblock_bdev = device->bdev;
793 
794 	/* super block bytenr is always the unmapped device bytenr */
795 	dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
796 	if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->commit_total_bytes)
797 		return -1;
798 	bh = __bread(superblock_bdev, dev_bytenr / 4096,
799 		     BTRFS_SUPER_INFO_SIZE);
800 	if (NULL == bh)
801 		return -1;
802 	super_tmp = (struct btrfs_super_block *)
803 	    (bh->b_data + (dev_bytenr & 4095));
804 
805 	if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
806 	    btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
807 	    memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
808 	    btrfs_super_nodesize(super_tmp) != state->metablock_size ||
809 	    btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
810 		brelse(bh);
811 		return 0;
812 	}
813 
814 	superblock_tmp =
815 	    btrfsic_block_hashtable_lookup(superblock_bdev,
816 					   dev_bytenr,
817 					   &state->block_hashtable);
818 	if (NULL == superblock_tmp) {
819 		superblock_tmp = btrfsic_block_alloc();
820 		if (NULL == superblock_tmp) {
821 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
822 			brelse(bh);
823 			return -1;
824 		}
825 		/* for superblock, only the dev_bytenr makes sense */
826 		superblock_tmp->dev_bytenr = dev_bytenr;
827 		superblock_tmp->dev_state = dev_state;
828 		superblock_tmp->logical_bytenr = dev_bytenr;
829 		superblock_tmp->generation = btrfs_super_generation(super_tmp);
830 		superblock_tmp->is_metadata = 1;
831 		superblock_tmp->is_superblock = 1;
832 		superblock_tmp->is_iodone = 1;
833 		superblock_tmp->never_written = 0;
834 		superblock_tmp->mirror_num = 1 + superblock_mirror_num;
835 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
836 			btrfs_info_in_rcu(device->dev_root->fs_info,
837 				"new initial S-block (bdev %p, %s) @%llu (%s/%llu/%d)",
838 				     superblock_bdev,
839 				     rcu_str_deref(device->name), dev_bytenr,
840 				     dev_state->name, dev_bytenr,
841 				     superblock_mirror_num);
842 		list_add(&superblock_tmp->all_blocks_node,
843 			 &state->all_blocks_list);
844 		btrfsic_block_hashtable_add(superblock_tmp,
845 					    &state->block_hashtable);
846 	}
847 
848 	/* select the one with the highest generation field */
849 	if (btrfs_super_generation(super_tmp) >
850 	    state->max_superblock_generation ||
851 	    0 == state->max_superblock_generation) {
852 		memcpy(selected_super, super_tmp, sizeof(*selected_super));
853 		*selected_dev_state = dev_state;
854 		state->max_superblock_generation =
855 		    btrfs_super_generation(super_tmp);
856 		state->latest_superblock = superblock_tmp;
857 	}
858 
859 	for (pass = 0; pass < 3; pass++) {
860 		u64 next_bytenr;
861 		int num_copies;
862 		int mirror_num;
863 		const char *additional_string = NULL;
864 		struct btrfs_disk_key tmp_disk_key;
865 
866 		tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
867 		tmp_disk_key.offset = 0;
868 		switch (pass) {
869 		case 0:
870 			btrfs_set_disk_key_objectid(&tmp_disk_key,
871 						    BTRFS_ROOT_TREE_OBJECTID);
872 			additional_string = "initial root ";
873 			next_bytenr = btrfs_super_root(super_tmp);
874 			break;
875 		case 1:
876 			btrfs_set_disk_key_objectid(&tmp_disk_key,
877 						    BTRFS_CHUNK_TREE_OBJECTID);
878 			additional_string = "initial chunk ";
879 			next_bytenr = btrfs_super_chunk_root(super_tmp);
880 			break;
881 		case 2:
882 			btrfs_set_disk_key_objectid(&tmp_disk_key,
883 						    BTRFS_TREE_LOG_OBJECTID);
884 			additional_string = "initial log ";
885 			next_bytenr = btrfs_super_log_root(super_tmp);
886 			if (0 == next_bytenr)
887 				continue;
888 			break;
889 		}
890 
891 		num_copies =
892 		    btrfs_num_copies(state->root->fs_info,
893 				     next_bytenr, state->metablock_size);
894 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
895 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
896 			       next_bytenr, num_copies);
897 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
898 			struct btrfsic_block *next_block;
899 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
900 			struct btrfsic_block_link *l;
901 
902 			if (btrfsic_map_block(state, next_bytenr,
903 					      state->metablock_size,
904 					      &tmp_next_block_ctx,
905 					      mirror_num)) {
906 				printk(KERN_INFO "btrfsic: btrfsic_map_block("
907 				       "bytenr @%llu, mirror %d) failed!\n",
908 				       next_bytenr, mirror_num);
909 				brelse(bh);
910 				return -1;
911 			}
912 
913 			next_block = btrfsic_block_lookup_or_add(
914 					state, &tmp_next_block_ctx,
915 					additional_string, 1, 1, 0,
916 					mirror_num, NULL);
917 			if (NULL == next_block) {
918 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
919 				brelse(bh);
920 				return -1;
921 			}
922 
923 			next_block->disk_key = tmp_disk_key;
924 			next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
925 			l = btrfsic_block_link_lookup_or_add(
926 					state, &tmp_next_block_ctx,
927 					next_block, superblock_tmp,
928 					BTRFSIC_GENERATION_UNKNOWN);
929 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
930 			if (NULL == l) {
931 				brelse(bh);
932 				return -1;
933 			}
934 		}
935 	}
936 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
937 		btrfsic_dump_tree_sub(state, superblock_tmp, 0);
938 
939 	brelse(bh);
940 	return 0;
941 }
942 
943 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
944 {
945 	struct btrfsic_stack_frame *sf;
946 
947 	sf = kzalloc(sizeof(*sf), GFP_NOFS);
948 	if (NULL == sf)
949 		printk(KERN_INFO "btrfsic: alloc memory failed!\n");
950 	else
951 		sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
952 	return sf;
953 }
954 
955 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
956 {
957 	BUG_ON(!(NULL == sf ||
958 		 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
959 	kfree(sf);
960 }
961 
962 static int btrfsic_process_metablock(
963 		struct btrfsic_state *state,
964 		struct btrfsic_block *const first_block,
965 		struct btrfsic_block_data_ctx *const first_block_ctx,
966 		int first_limit_nesting, int force_iodone_flag)
967 {
968 	struct btrfsic_stack_frame initial_stack_frame = { 0 };
969 	struct btrfsic_stack_frame *sf;
970 	struct btrfsic_stack_frame *next_stack;
971 	struct btrfs_header *const first_hdr =
972 		(struct btrfs_header *)first_block_ctx->datav[0];
973 
974 	BUG_ON(!first_hdr);
975 	sf = &initial_stack_frame;
976 	sf->error = 0;
977 	sf->i = -1;
978 	sf->limit_nesting = first_limit_nesting;
979 	sf->block = first_block;
980 	sf->block_ctx = first_block_ctx;
981 	sf->next_block = NULL;
982 	sf->hdr = first_hdr;
983 	sf->prev = NULL;
984 
985 continue_with_new_stack_frame:
986 	sf->block->generation = le64_to_cpu(sf->hdr->generation);
987 	if (0 == sf->hdr->level) {
988 		struct btrfs_leaf *const leafhdr =
989 		    (struct btrfs_leaf *)sf->hdr;
990 
991 		if (-1 == sf->i) {
992 			sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
993 
994 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
995 				printk(KERN_INFO
996 				       "leaf %llu items %d generation %llu"
997 				       " owner %llu\n",
998 				       sf->block_ctx->start, sf->nr,
999 				       btrfs_stack_header_generation(
1000 					       &leafhdr->header),
1001 				       btrfs_stack_header_owner(
1002 					       &leafhdr->header));
1003 		}
1004 
1005 continue_with_current_leaf_stack_frame:
1006 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1007 			sf->i++;
1008 			sf->num_copies = 0;
1009 		}
1010 
1011 		if (sf->i < sf->nr) {
1012 			struct btrfs_item disk_item;
1013 			u32 disk_item_offset =
1014 				(uintptr_t)(leafhdr->items + sf->i) -
1015 				(uintptr_t)leafhdr;
1016 			struct btrfs_disk_key *disk_key;
1017 			u8 type;
1018 			u32 item_offset;
1019 			u32 item_size;
1020 
1021 			if (disk_item_offset + sizeof(struct btrfs_item) >
1022 			    sf->block_ctx->len) {
1023 leaf_item_out_of_bounce_error:
1024 				printk(KERN_INFO
1025 				       "btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
1026 				       sf->block_ctx->start,
1027 				       sf->block_ctx->dev->name);
1028 				goto one_stack_frame_backwards;
1029 			}
1030 			btrfsic_read_from_block_data(sf->block_ctx,
1031 						     &disk_item,
1032 						     disk_item_offset,
1033 						     sizeof(struct btrfs_item));
1034 			item_offset = btrfs_stack_item_offset(&disk_item);
1035 			item_size = btrfs_stack_item_size(&disk_item);
1036 			disk_key = &disk_item.key;
1037 			type = btrfs_disk_key_type(disk_key);
1038 
1039 			if (BTRFS_ROOT_ITEM_KEY == type) {
1040 				struct btrfs_root_item root_item;
1041 				u32 root_item_offset;
1042 				u64 next_bytenr;
1043 
1044 				root_item_offset = item_offset +
1045 					offsetof(struct btrfs_leaf, items);
1046 				if (root_item_offset + item_size >
1047 				    sf->block_ctx->len)
1048 					goto leaf_item_out_of_bounce_error;
1049 				btrfsic_read_from_block_data(
1050 					sf->block_ctx, &root_item,
1051 					root_item_offset,
1052 					item_size);
1053 				next_bytenr = btrfs_root_bytenr(&root_item);
1054 
1055 				sf->error =
1056 				    btrfsic_create_link_to_next_block(
1057 						state,
1058 						sf->block,
1059 						sf->block_ctx,
1060 						next_bytenr,
1061 						sf->limit_nesting,
1062 						&sf->next_block_ctx,
1063 						&sf->next_block,
1064 						force_iodone_flag,
1065 						&sf->num_copies,
1066 						&sf->mirror_num,
1067 						disk_key,
1068 						btrfs_root_generation(
1069 						&root_item));
1070 				if (sf->error)
1071 					goto one_stack_frame_backwards;
1072 
1073 				if (NULL != sf->next_block) {
1074 					struct btrfs_header *const next_hdr =
1075 					    (struct btrfs_header *)
1076 					    sf->next_block_ctx.datav[0];
1077 
1078 					next_stack =
1079 					    btrfsic_stack_frame_alloc();
1080 					if (NULL == next_stack) {
1081 						sf->error = -1;
1082 						btrfsic_release_block_ctx(
1083 								&sf->
1084 								next_block_ctx);
1085 						goto one_stack_frame_backwards;
1086 					}
1087 
1088 					next_stack->i = -1;
1089 					next_stack->block = sf->next_block;
1090 					next_stack->block_ctx =
1091 					    &sf->next_block_ctx;
1092 					next_stack->next_block = NULL;
1093 					next_stack->hdr = next_hdr;
1094 					next_stack->limit_nesting =
1095 					    sf->limit_nesting - 1;
1096 					next_stack->prev = sf;
1097 					sf = next_stack;
1098 					goto continue_with_new_stack_frame;
1099 				}
1100 			} else if (BTRFS_EXTENT_DATA_KEY == type &&
1101 				   state->include_extent_data) {
1102 				sf->error = btrfsic_handle_extent_data(
1103 						state,
1104 						sf->block,
1105 						sf->block_ctx,
1106 						item_offset,
1107 						force_iodone_flag);
1108 				if (sf->error)
1109 					goto one_stack_frame_backwards;
1110 			}
1111 
1112 			goto continue_with_current_leaf_stack_frame;
1113 		}
1114 	} else {
1115 		struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1116 
1117 		if (-1 == sf->i) {
1118 			sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1119 
1120 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1121 				printk(KERN_INFO "node %llu level %d items %d"
1122 				       " generation %llu owner %llu\n",
1123 				       sf->block_ctx->start,
1124 				       nodehdr->header.level, sf->nr,
1125 				       btrfs_stack_header_generation(
1126 				       &nodehdr->header),
1127 				       btrfs_stack_header_owner(
1128 				       &nodehdr->header));
1129 		}
1130 
1131 continue_with_current_node_stack_frame:
1132 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1133 			sf->i++;
1134 			sf->num_copies = 0;
1135 		}
1136 
1137 		if (sf->i < sf->nr) {
1138 			struct btrfs_key_ptr key_ptr;
1139 			u32 key_ptr_offset;
1140 			u64 next_bytenr;
1141 
1142 			key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1143 					  (uintptr_t)nodehdr;
1144 			if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1145 			    sf->block_ctx->len) {
1146 				printk(KERN_INFO
1147 				       "btrfsic: node item out of bounce at logical %llu, dev %s\n",
1148 				       sf->block_ctx->start,
1149 				       sf->block_ctx->dev->name);
1150 				goto one_stack_frame_backwards;
1151 			}
1152 			btrfsic_read_from_block_data(
1153 				sf->block_ctx, &key_ptr, key_ptr_offset,
1154 				sizeof(struct btrfs_key_ptr));
1155 			next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1156 
1157 			sf->error = btrfsic_create_link_to_next_block(
1158 					state,
1159 					sf->block,
1160 					sf->block_ctx,
1161 					next_bytenr,
1162 					sf->limit_nesting,
1163 					&sf->next_block_ctx,
1164 					&sf->next_block,
1165 					force_iodone_flag,
1166 					&sf->num_copies,
1167 					&sf->mirror_num,
1168 					&key_ptr.key,
1169 					btrfs_stack_key_generation(&key_ptr));
1170 			if (sf->error)
1171 				goto one_stack_frame_backwards;
1172 
1173 			if (NULL != sf->next_block) {
1174 				struct btrfs_header *const next_hdr =
1175 				    (struct btrfs_header *)
1176 				    sf->next_block_ctx.datav[0];
1177 
1178 				next_stack = btrfsic_stack_frame_alloc();
1179 				if (NULL == next_stack) {
1180 					sf->error = -1;
1181 					goto one_stack_frame_backwards;
1182 				}
1183 
1184 				next_stack->i = -1;
1185 				next_stack->block = sf->next_block;
1186 				next_stack->block_ctx = &sf->next_block_ctx;
1187 				next_stack->next_block = NULL;
1188 				next_stack->hdr = next_hdr;
1189 				next_stack->limit_nesting =
1190 				    sf->limit_nesting - 1;
1191 				next_stack->prev = sf;
1192 				sf = next_stack;
1193 				goto continue_with_new_stack_frame;
1194 			}
1195 
1196 			goto continue_with_current_node_stack_frame;
1197 		}
1198 	}
1199 
1200 one_stack_frame_backwards:
1201 	if (NULL != sf->prev) {
1202 		struct btrfsic_stack_frame *const prev = sf->prev;
1203 
1204 		/* the one for the initial block is freed in the caller */
1205 		btrfsic_release_block_ctx(sf->block_ctx);
1206 
1207 		if (sf->error) {
1208 			prev->error = sf->error;
1209 			btrfsic_stack_frame_free(sf);
1210 			sf = prev;
1211 			goto one_stack_frame_backwards;
1212 		}
1213 
1214 		btrfsic_stack_frame_free(sf);
1215 		sf = prev;
1216 		goto continue_with_new_stack_frame;
1217 	} else {
1218 		BUG_ON(&initial_stack_frame != sf);
1219 	}
1220 
1221 	return sf->error;
1222 }
1223 
1224 static void btrfsic_read_from_block_data(
1225 	struct btrfsic_block_data_ctx *block_ctx,
1226 	void *dstv, u32 offset, size_t len)
1227 {
1228 	size_t cur;
1229 	size_t offset_in_page;
1230 	char *kaddr;
1231 	char *dst = (char *)dstv;
1232 	size_t start_offset = block_ctx->start & ((u64)PAGE_CACHE_SIZE - 1);
1233 	unsigned long i = (start_offset + offset) >> PAGE_CACHE_SHIFT;
1234 
1235 	WARN_ON(offset + len > block_ctx->len);
1236 	offset_in_page = (start_offset + offset) & (PAGE_CACHE_SIZE - 1);
1237 
1238 	while (len > 0) {
1239 		cur = min(len, ((size_t)PAGE_CACHE_SIZE - offset_in_page));
1240 		BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_CACHE_SIZE));
1241 		kaddr = block_ctx->datav[i];
1242 		memcpy(dst, kaddr + offset_in_page, cur);
1243 
1244 		dst += cur;
1245 		len -= cur;
1246 		offset_in_page = 0;
1247 		i++;
1248 	}
1249 }
1250 
1251 static int btrfsic_create_link_to_next_block(
1252 		struct btrfsic_state *state,
1253 		struct btrfsic_block *block,
1254 		struct btrfsic_block_data_ctx *block_ctx,
1255 		u64 next_bytenr,
1256 		int limit_nesting,
1257 		struct btrfsic_block_data_ctx *next_block_ctx,
1258 		struct btrfsic_block **next_blockp,
1259 		int force_iodone_flag,
1260 		int *num_copiesp, int *mirror_nump,
1261 		struct btrfs_disk_key *disk_key,
1262 		u64 parent_generation)
1263 {
1264 	struct btrfsic_block *next_block = NULL;
1265 	int ret;
1266 	struct btrfsic_block_link *l;
1267 	int did_alloc_block_link;
1268 	int block_was_created;
1269 
1270 	*next_blockp = NULL;
1271 	if (0 == *num_copiesp) {
1272 		*num_copiesp =
1273 		    btrfs_num_copies(state->root->fs_info,
1274 				     next_bytenr, state->metablock_size);
1275 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1276 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
1277 			       next_bytenr, *num_copiesp);
1278 		*mirror_nump = 1;
1279 	}
1280 
1281 	if (*mirror_nump > *num_copiesp)
1282 		return 0;
1283 
1284 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1285 		printk(KERN_INFO
1286 		       "btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1287 		       *mirror_nump);
1288 	ret = btrfsic_map_block(state, next_bytenr,
1289 				state->metablock_size,
1290 				next_block_ctx, *mirror_nump);
1291 	if (ret) {
1292 		printk(KERN_INFO
1293 		       "btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1294 		       next_bytenr, *mirror_nump);
1295 		btrfsic_release_block_ctx(next_block_ctx);
1296 		*next_blockp = NULL;
1297 		return -1;
1298 	}
1299 
1300 	next_block = btrfsic_block_lookup_or_add(state,
1301 						 next_block_ctx, "referenced ",
1302 						 1, force_iodone_flag,
1303 						 !force_iodone_flag,
1304 						 *mirror_nump,
1305 						 &block_was_created);
1306 	if (NULL == next_block) {
1307 		btrfsic_release_block_ctx(next_block_ctx);
1308 		*next_blockp = NULL;
1309 		return -1;
1310 	}
1311 	if (block_was_created) {
1312 		l = NULL;
1313 		next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1314 	} else {
1315 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1316 			if (next_block->logical_bytenr != next_bytenr &&
1317 			    !(!next_block->is_metadata &&
1318 			      0 == next_block->logical_bytenr))
1319 				printk(KERN_INFO
1320 				       "Referenced block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1321 				       next_bytenr, next_block_ctx->dev->name,
1322 				       next_block_ctx->dev_bytenr, *mirror_nump,
1323 				       btrfsic_get_block_type(state,
1324 							      next_block),
1325 				       next_block->logical_bytenr);
1326 			else
1327 				printk(KERN_INFO
1328 				       "Referenced block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1329 				       next_bytenr, next_block_ctx->dev->name,
1330 				       next_block_ctx->dev_bytenr, *mirror_nump,
1331 				       btrfsic_get_block_type(state,
1332 							      next_block));
1333 		}
1334 		next_block->logical_bytenr = next_bytenr;
1335 
1336 		next_block->mirror_num = *mirror_nump;
1337 		l = btrfsic_block_link_hashtable_lookup(
1338 				next_block_ctx->dev->bdev,
1339 				next_block_ctx->dev_bytenr,
1340 				block_ctx->dev->bdev,
1341 				block_ctx->dev_bytenr,
1342 				&state->block_link_hashtable);
1343 	}
1344 
1345 	next_block->disk_key = *disk_key;
1346 	if (NULL == l) {
1347 		l = btrfsic_block_link_alloc();
1348 		if (NULL == l) {
1349 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
1350 			btrfsic_release_block_ctx(next_block_ctx);
1351 			*next_blockp = NULL;
1352 			return -1;
1353 		}
1354 
1355 		did_alloc_block_link = 1;
1356 		l->block_ref_to = next_block;
1357 		l->block_ref_from = block;
1358 		l->ref_cnt = 1;
1359 		l->parent_generation = parent_generation;
1360 
1361 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1362 			btrfsic_print_add_link(state, l);
1363 
1364 		list_add(&l->node_ref_to, &block->ref_to_list);
1365 		list_add(&l->node_ref_from, &next_block->ref_from_list);
1366 
1367 		btrfsic_block_link_hashtable_add(l,
1368 						 &state->block_link_hashtable);
1369 	} else {
1370 		did_alloc_block_link = 0;
1371 		if (0 == limit_nesting) {
1372 			l->ref_cnt++;
1373 			l->parent_generation = parent_generation;
1374 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1375 				btrfsic_print_add_link(state, l);
1376 		}
1377 	}
1378 
1379 	if (limit_nesting > 0 && did_alloc_block_link) {
1380 		ret = btrfsic_read_block(state, next_block_ctx);
1381 		if (ret < (int)next_block_ctx->len) {
1382 			printk(KERN_INFO
1383 			       "btrfsic: read block @logical %llu failed!\n",
1384 			       next_bytenr);
1385 			btrfsic_release_block_ctx(next_block_ctx);
1386 			*next_blockp = NULL;
1387 			return -1;
1388 		}
1389 
1390 		*next_blockp = next_block;
1391 	} else {
1392 		*next_blockp = NULL;
1393 	}
1394 	(*mirror_nump)++;
1395 
1396 	return 0;
1397 }
1398 
1399 static int btrfsic_handle_extent_data(
1400 		struct btrfsic_state *state,
1401 		struct btrfsic_block *block,
1402 		struct btrfsic_block_data_ctx *block_ctx,
1403 		u32 item_offset, int force_iodone_flag)
1404 {
1405 	int ret;
1406 	struct btrfs_file_extent_item file_extent_item;
1407 	u64 file_extent_item_offset;
1408 	u64 next_bytenr;
1409 	u64 num_bytes;
1410 	u64 generation;
1411 	struct btrfsic_block_link *l;
1412 
1413 	file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1414 				  item_offset;
1415 	if (file_extent_item_offset +
1416 	    offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1417 	    block_ctx->len) {
1418 		printk(KERN_INFO
1419 		       "btrfsic: file item out of bounce at logical %llu, dev %s\n",
1420 		       block_ctx->start, block_ctx->dev->name);
1421 		return -1;
1422 	}
1423 
1424 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1425 		file_extent_item_offset,
1426 		offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1427 	if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1428 	    btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1429 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1430 			printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu\n",
1431 			       file_extent_item.type,
1432 			       btrfs_stack_file_extent_disk_bytenr(
1433 			       &file_extent_item));
1434 		return 0;
1435 	}
1436 
1437 	if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1438 	    block_ctx->len) {
1439 		printk(KERN_INFO
1440 		       "btrfsic: file item out of bounce at logical %llu, dev %s\n",
1441 		       block_ctx->start, block_ctx->dev->name);
1442 		return -1;
1443 	}
1444 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1445 				     file_extent_item_offset,
1446 				     sizeof(struct btrfs_file_extent_item));
1447 	next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item);
1448 	if (btrfs_stack_file_extent_compression(&file_extent_item) ==
1449 	    BTRFS_COMPRESS_NONE) {
1450 		next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item);
1451 		num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1452 	} else {
1453 		num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item);
1454 	}
1455 	generation = btrfs_stack_file_extent_generation(&file_extent_item);
1456 
1457 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1458 		printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu,"
1459 		       " offset = %llu, num_bytes = %llu\n",
1460 		       file_extent_item.type,
1461 		       btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1462 		       btrfs_stack_file_extent_offset(&file_extent_item),
1463 		       num_bytes);
1464 	while (num_bytes > 0) {
1465 		u32 chunk_len;
1466 		int num_copies;
1467 		int mirror_num;
1468 
1469 		if (num_bytes > state->datablock_size)
1470 			chunk_len = state->datablock_size;
1471 		else
1472 			chunk_len = num_bytes;
1473 
1474 		num_copies =
1475 		    btrfs_num_copies(state->root->fs_info,
1476 				     next_bytenr, state->datablock_size);
1477 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1478 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
1479 			       next_bytenr, num_copies);
1480 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1481 			struct btrfsic_block_data_ctx next_block_ctx;
1482 			struct btrfsic_block *next_block;
1483 			int block_was_created;
1484 
1485 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1486 				printk(KERN_INFO "btrfsic_handle_extent_data("
1487 				       "mirror_num=%d)\n", mirror_num);
1488 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1489 				printk(KERN_INFO
1490 				       "\tdisk_bytenr = %llu, num_bytes %u\n",
1491 				       next_bytenr, chunk_len);
1492 			ret = btrfsic_map_block(state, next_bytenr,
1493 						chunk_len, &next_block_ctx,
1494 						mirror_num);
1495 			if (ret) {
1496 				printk(KERN_INFO
1497 				       "btrfsic: btrfsic_map_block(@%llu,"
1498 				       " mirror=%d) failed!\n",
1499 				       next_bytenr, mirror_num);
1500 				return -1;
1501 			}
1502 
1503 			next_block = btrfsic_block_lookup_or_add(
1504 					state,
1505 					&next_block_ctx,
1506 					"referenced ",
1507 					0,
1508 					force_iodone_flag,
1509 					!force_iodone_flag,
1510 					mirror_num,
1511 					&block_was_created);
1512 			if (NULL == next_block) {
1513 				printk(KERN_INFO
1514 				       "btrfsic: error, kmalloc failed!\n");
1515 				btrfsic_release_block_ctx(&next_block_ctx);
1516 				return -1;
1517 			}
1518 			if (!block_was_created) {
1519 				if ((state->print_mask &
1520 				     BTRFSIC_PRINT_MASK_VERBOSE) &&
1521 				    next_block->logical_bytenr != next_bytenr &&
1522 				    !(!next_block->is_metadata &&
1523 				      0 == next_block->logical_bytenr)) {
1524 					printk(KERN_INFO
1525 					       "Referenced block"
1526 					       " @%llu (%s/%llu/%d)"
1527 					       " found in hash table, D,"
1528 					       " bytenr mismatch"
1529 					       " (!= stored %llu).\n",
1530 					       next_bytenr,
1531 					       next_block_ctx.dev->name,
1532 					       next_block_ctx.dev_bytenr,
1533 					       mirror_num,
1534 					       next_block->logical_bytenr);
1535 				}
1536 				next_block->logical_bytenr = next_bytenr;
1537 				next_block->mirror_num = mirror_num;
1538 			}
1539 
1540 			l = btrfsic_block_link_lookup_or_add(state,
1541 							     &next_block_ctx,
1542 							     next_block, block,
1543 							     generation);
1544 			btrfsic_release_block_ctx(&next_block_ctx);
1545 			if (NULL == l)
1546 				return -1;
1547 		}
1548 
1549 		next_bytenr += chunk_len;
1550 		num_bytes -= chunk_len;
1551 	}
1552 
1553 	return 0;
1554 }
1555 
1556 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1557 			     struct btrfsic_block_data_ctx *block_ctx_out,
1558 			     int mirror_num)
1559 {
1560 	int ret;
1561 	u64 length;
1562 	struct btrfs_bio *multi = NULL;
1563 	struct btrfs_device *device;
1564 
1565 	length = len;
1566 	ret = btrfs_map_block(state->root->fs_info, READ,
1567 			      bytenr, &length, &multi, mirror_num);
1568 
1569 	if (ret) {
1570 		block_ctx_out->start = 0;
1571 		block_ctx_out->dev_bytenr = 0;
1572 		block_ctx_out->len = 0;
1573 		block_ctx_out->dev = NULL;
1574 		block_ctx_out->datav = NULL;
1575 		block_ctx_out->pagev = NULL;
1576 		block_ctx_out->mem_to_free = NULL;
1577 
1578 		return ret;
1579 	}
1580 
1581 	device = multi->stripes[0].dev;
1582 	block_ctx_out->dev = btrfsic_dev_state_lookup(device->bdev);
1583 	block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1584 	block_ctx_out->start = bytenr;
1585 	block_ctx_out->len = len;
1586 	block_ctx_out->datav = NULL;
1587 	block_ctx_out->pagev = NULL;
1588 	block_ctx_out->mem_to_free = NULL;
1589 
1590 	kfree(multi);
1591 	if (NULL == block_ctx_out->dev) {
1592 		ret = -ENXIO;
1593 		printk(KERN_INFO "btrfsic: error, cannot lookup dev (#1)!\n");
1594 	}
1595 
1596 	return ret;
1597 }
1598 
1599 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1600 {
1601 	if (block_ctx->mem_to_free) {
1602 		unsigned int num_pages;
1603 
1604 		BUG_ON(!block_ctx->datav);
1605 		BUG_ON(!block_ctx->pagev);
1606 		num_pages = (block_ctx->len + (u64)PAGE_CACHE_SIZE - 1) >>
1607 			    PAGE_CACHE_SHIFT;
1608 		while (num_pages > 0) {
1609 			num_pages--;
1610 			if (block_ctx->datav[num_pages]) {
1611 				kunmap(block_ctx->pagev[num_pages]);
1612 				block_ctx->datav[num_pages] = NULL;
1613 			}
1614 			if (block_ctx->pagev[num_pages]) {
1615 				__free_page(block_ctx->pagev[num_pages]);
1616 				block_ctx->pagev[num_pages] = NULL;
1617 			}
1618 		}
1619 
1620 		kfree(block_ctx->mem_to_free);
1621 		block_ctx->mem_to_free = NULL;
1622 		block_ctx->pagev = NULL;
1623 		block_ctx->datav = NULL;
1624 	}
1625 }
1626 
1627 static int btrfsic_read_block(struct btrfsic_state *state,
1628 			      struct btrfsic_block_data_ctx *block_ctx)
1629 {
1630 	unsigned int num_pages;
1631 	unsigned int i;
1632 	u64 dev_bytenr;
1633 	int ret;
1634 
1635 	BUG_ON(block_ctx->datav);
1636 	BUG_ON(block_ctx->pagev);
1637 	BUG_ON(block_ctx->mem_to_free);
1638 	if (block_ctx->dev_bytenr & ((u64)PAGE_CACHE_SIZE - 1)) {
1639 		printk(KERN_INFO
1640 		       "btrfsic: read_block() with unaligned bytenr %llu\n",
1641 		       block_ctx->dev_bytenr);
1642 		return -1;
1643 	}
1644 
1645 	num_pages = (block_ctx->len + (u64)PAGE_CACHE_SIZE - 1) >>
1646 		    PAGE_CACHE_SHIFT;
1647 	block_ctx->mem_to_free = kzalloc((sizeof(*block_ctx->datav) +
1648 					  sizeof(*block_ctx->pagev)) *
1649 					 num_pages, GFP_NOFS);
1650 	if (!block_ctx->mem_to_free)
1651 		return -ENOMEM;
1652 	block_ctx->datav = block_ctx->mem_to_free;
1653 	block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1654 	for (i = 0; i < num_pages; i++) {
1655 		block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1656 		if (!block_ctx->pagev[i])
1657 			return -1;
1658 	}
1659 
1660 	dev_bytenr = block_ctx->dev_bytenr;
1661 	for (i = 0; i < num_pages;) {
1662 		struct bio *bio;
1663 		unsigned int j;
1664 
1665 		bio = btrfs_io_bio_alloc(GFP_NOFS, num_pages - i);
1666 		if (!bio) {
1667 			printk(KERN_INFO
1668 			       "btrfsic: bio_alloc() for %u pages failed!\n",
1669 			       num_pages - i);
1670 			return -1;
1671 		}
1672 		bio->bi_bdev = block_ctx->dev->bdev;
1673 		bio->bi_iter.bi_sector = dev_bytenr >> 9;
1674 
1675 		for (j = i; j < num_pages; j++) {
1676 			ret = bio_add_page(bio, block_ctx->pagev[j],
1677 					   PAGE_CACHE_SIZE, 0);
1678 			if (PAGE_CACHE_SIZE != ret)
1679 				break;
1680 		}
1681 		if (j == i) {
1682 			printk(KERN_INFO
1683 			       "btrfsic: error, failed to add a single page!\n");
1684 			return -1;
1685 		}
1686 		if (submit_bio_wait(READ, bio)) {
1687 			printk(KERN_INFO
1688 			       "btrfsic: read error at logical %llu dev %s!\n",
1689 			       block_ctx->start, block_ctx->dev->name);
1690 			bio_put(bio);
1691 			return -1;
1692 		}
1693 		bio_put(bio);
1694 		dev_bytenr += (j - i) * PAGE_CACHE_SIZE;
1695 		i = j;
1696 	}
1697 	for (i = 0; i < num_pages; i++) {
1698 		block_ctx->datav[i] = kmap(block_ctx->pagev[i]);
1699 		if (!block_ctx->datav[i]) {
1700 			printk(KERN_INFO "btrfsic: kmap() failed (dev %s)!\n",
1701 			       block_ctx->dev->name);
1702 			return -1;
1703 		}
1704 	}
1705 
1706 	return block_ctx->len;
1707 }
1708 
1709 static void btrfsic_dump_database(struct btrfsic_state *state)
1710 {
1711 	const struct btrfsic_block *b_all;
1712 
1713 	BUG_ON(NULL == state);
1714 
1715 	printk(KERN_INFO "all_blocks_list:\n");
1716 	list_for_each_entry(b_all, &state->all_blocks_list, all_blocks_node) {
1717 		const struct btrfsic_block_link *l;
1718 
1719 		printk(KERN_INFO "%c-block @%llu (%s/%llu/%d)\n",
1720 		       btrfsic_get_block_type(state, b_all),
1721 		       b_all->logical_bytenr, b_all->dev_state->name,
1722 		       b_all->dev_bytenr, b_all->mirror_num);
1723 
1724 		list_for_each_entry(l, &b_all->ref_to_list, node_ref_to) {
1725 			printk(KERN_INFO " %c @%llu (%s/%llu/%d)"
1726 			       " refers %u* to"
1727 			       " %c @%llu (%s/%llu/%d)\n",
1728 			       btrfsic_get_block_type(state, b_all),
1729 			       b_all->logical_bytenr, b_all->dev_state->name,
1730 			       b_all->dev_bytenr, b_all->mirror_num,
1731 			       l->ref_cnt,
1732 			       btrfsic_get_block_type(state, l->block_ref_to),
1733 			       l->block_ref_to->logical_bytenr,
1734 			       l->block_ref_to->dev_state->name,
1735 			       l->block_ref_to->dev_bytenr,
1736 			       l->block_ref_to->mirror_num);
1737 		}
1738 
1739 		list_for_each_entry(l, &b_all->ref_from_list, node_ref_from) {
1740 			printk(KERN_INFO " %c @%llu (%s/%llu/%d)"
1741 			       " is ref %u* from"
1742 			       " %c @%llu (%s/%llu/%d)\n",
1743 			       btrfsic_get_block_type(state, b_all),
1744 			       b_all->logical_bytenr, b_all->dev_state->name,
1745 			       b_all->dev_bytenr, b_all->mirror_num,
1746 			       l->ref_cnt,
1747 			       btrfsic_get_block_type(state, l->block_ref_from),
1748 			       l->block_ref_from->logical_bytenr,
1749 			       l->block_ref_from->dev_state->name,
1750 			       l->block_ref_from->dev_bytenr,
1751 			       l->block_ref_from->mirror_num);
1752 		}
1753 
1754 		printk(KERN_INFO "\n");
1755 	}
1756 }
1757 
1758 /*
1759  * Test whether the disk block contains a tree block (leaf or node)
1760  * (note that this test fails for the super block)
1761  */
1762 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
1763 				     char **datav, unsigned int num_pages)
1764 {
1765 	struct btrfs_header *h;
1766 	u8 csum[BTRFS_CSUM_SIZE];
1767 	u32 crc = ~(u32)0;
1768 	unsigned int i;
1769 
1770 	if (num_pages * PAGE_CACHE_SIZE < state->metablock_size)
1771 		return 1; /* not metadata */
1772 	num_pages = state->metablock_size >> PAGE_CACHE_SHIFT;
1773 	h = (struct btrfs_header *)datav[0];
1774 
1775 	if (memcmp(h->fsid, state->root->fs_info->fsid, BTRFS_UUID_SIZE))
1776 		return 1;
1777 
1778 	for (i = 0; i < num_pages; i++) {
1779 		u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1780 		size_t sublen = i ? PAGE_CACHE_SIZE :
1781 				    (PAGE_CACHE_SIZE - BTRFS_CSUM_SIZE);
1782 
1783 		crc = btrfs_crc32c(crc, data, sublen);
1784 	}
1785 	btrfs_csum_final(crc, csum);
1786 	if (memcmp(csum, h->csum, state->csum_size))
1787 		return 1;
1788 
1789 	return 0; /* is metadata */
1790 }
1791 
1792 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1793 					  u64 dev_bytenr, char **mapped_datav,
1794 					  unsigned int num_pages,
1795 					  struct bio *bio, int *bio_is_patched,
1796 					  struct buffer_head *bh,
1797 					  int submit_bio_bh_rw)
1798 {
1799 	int is_metadata;
1800 	struct btrfsic_block *block;
1801 	struct btrfsic_block_data_ctx block_ctx;
1802 	int ret;
1803 	struct btrfsic_state *state = dev_state->state;
1804 	struct block_device *bdev = dev_state->bdev;
1805 	unsigned int processed_len;
1806 
1807 	if (NULL != bio_is_patched)
1808 		*bio_is_patched = 0;
1809 
1810 again:
1811 	if (num_pages == 0)
1812 		return;
1813 
1814 	processed_len = 0;
1815 	is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1816 						      num_pages));
1817 
1818 	block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1819 					       &state->block_hashtable);
1820 	if (NULL != block) {
1821 		u64 bytenr = 0;
1822 		struct btrfsic_block_link *l, *tmp;
1823 
1824 		if (block->is_superblock) {
1825 			bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1826 						    mapped_datav[0]);
1827 			if (num_pages * PAGE_CACHE_SIZE <
1828 			    BTRFS_SUPER_INFO_SIZE) {
1829 				printk(KERN_INFO
1830 				       "btrfsic: cannot work with too short bios!\n");
1831 				return;
1832 			}
1833 			is_metadata = 1;
1834 			BUG_ON(BTRFS_SUPER_INFO_SIZE & (PAGE_CACHE_SIZE - 1));
1835 			processed_len = BTRFS_SUPER_INFO_SIZE;
1836 			if (state->print_mask &
1837 			    BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1838 				printk(KERN_INFO
1839 				       "[before new superblock is written]:\n");
1840 				btrfsic_dump_tree_sub(state, block, 0);
1841 			}
1842 		}
1843 		if (is_metadata) {
1844 			if (!block->is_superblock) {
1845 				if (num_pages * PAGE_CACHE_SIZE <
1846 				    state->metablock_size) {
1847 					printk(KERN_INFO
1848 					       "btrfsic: cannot work with too short bios!\n");
1849 					return;
1850 				}
1851 				processed_len = state->metablock_size;
1852 				bytenr = btrfs_stack_header_bytenr(
1853 						(struct btrfs_header *)
1854 						mapped_datav[0]);
1855 				btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1856 							       dev_state,
1857 							       dev_bytenr);
1858 			}
1859 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1860 				if (block->logical_bytenr != bytenr &&
1861 				    !(!block->is_metadata &&
1862 				      block->logical_bytenr == 0))
1863 					printk(KERN_INFO
1864 					       "Written block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1865 					       bytenr, dev_state->name,
1866 					       dev_bytenr,
1867 					       block->mirror_num,
1868 					       btrfsic_get_block_type(state,
1869 								      block),
1870 					       block->logical_bytenr);
1871 				else
1872 					printk(KERN_INFO
1873 					       "Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1874 					       bytenr, dev_state->name,
1875 					       dev_bytenr, block->mirror_num,
1876 					       btrfsic_get_block_type(state,
1877 								      block));
1878 			}
1879 			block->logical_bytenr = bytenr;
1880 		} else {
1881 			if (num_pages * PAGE_CACHE_SIZE <
1882 			    state->datablock_size) {
1883 				printk(KERN_INFO
1884 				       "btrfsic: cannot work with too short bios!\n");
1885 				return;
1886 			}
1887 			processed_len = state->datablock_size;
1888 			bytenr = block->logical_bytenr;
1889 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1890 				printk(KERN_INFO
1891 				       "Written block @%llu (%s/%llu/%d)"
1892 				       " found in hash table, %c.\n",
1893 				       bytenr, dev_state->name, dev_bytenr,
1894 				       block->mirror_num,
1895 				       btrfsic_get_block_type(state, block));
1896 		}
1897 
1898 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1899 			printk(KERN_INFO
1900 			       "ref_to_list: %cE, ref_from_list: %cE\n",
1901 			       list_empty(&block->ref_to_list) ? ' ' : '!',
1902 			       list_empty(&block->ref_from_list) ? ' ' : '!');
1903 		if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1904 			printk(KERN_INFO "btrfs: attempt to overwrite %c-block"
1905 			       " @%llu (%s/%llu/%d), old(gen=%llu,"
1906 			       " objectid=%llu, type=%d, offset=%llu),"
1907 			       " new(gen=%llu),"
1908 			       " which is referenced by most recent superblock"
1909 			       " (superblockgen=%llu)!\n",
1910 			       btrfsic_get_block_type(state, block), bytenr,
1911 			       dev_state->name, dev_bytenr, block->mirror_num,
1912 			       block->generation,
1913 			       btrfs_disk_key_objectid(&block->disk_key),
1914 			       block->disk_key.type,
1915 			       btrfs_disk_key_offset(&block->disk_key),
1916 			       btrfs_stack_header_generation(
1917 				       (struct btrfs_header *) mapped_datav[0]),
1918 			       state->max_superblock_generation);
1919 			btrfsic_dump_tree(state);
1920 		}
1921 
1922 		if (!block->is_iodone && !block->never_written) {
1923 			printk(KERN_INFO "btrfs: attempt to overwrite %c-block"
1924 			       " @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu,"
1925 			       " which is not yet iodone!\n",
1926 			       btrfsic_get_block_type(state, block), bytenr,
1927 			       dev_state->name, dev_bytenr, block->mirror_num,
1928 			       block->generation,
1929 			       btrfs_stack_header_generation(
1930 				       (struct btrfs_header *)
1931 				       mapped_datav[0]));
1932 			/* it would not be safe to go on */
1933 			btrfsic_dump_tree(state);
1934 			goto continue_loop;
1935 		}
1936 
1937 		/*
1938 		 * Clear all references of this block. Do not free
1939 		 * the block itself even if is not referenced anymore
1940 		 * because it still carries valueable information
1941 		 * like whether it was ever written and IO completed.
1942 		 */
1943 		list_for_each_entry_safe(l, tmp, &block->ref_to_list,
1944 					 node_ref_to) {
1945 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1946 				btrfsic_print_rem_link(state, l);
1947 			l->ref_cnt--;
1948 			if (0 == l->ref_cnt) {
1949 				list_del(&l->node_ref_to);
1950 				list_del(&l->node_ref_from);
1951 				btrfsic_block_link_hashtable_remove(l);
1952 				btrfsic_block_link_free(l);
1953 			}
1954 		}
1955 
1956 		block_ctx.dev = dev_state;
1957 		block_ctx.dev_bytenr = dev_bytenr;
1958 		block_ctx.start = bytenr;
1959 		block_ctx.len = processed_len;
1960 		block_ctx.pagev = NULL;
1961 		block_ctx.mem_to_free = NULL;
1962 		block_ctx.datav = mapped_datav;
1963 
1964 		if (is_metadata || state->include_extent_data) {
1965 			block->never_written = 0;
1966 			block->iodone_w_error = 0;
1967 			if (NULL != bio) {
1968 				block->is_iodone = 0;
1969 				BUG_ON(NULL == bio_is_patched);
1970 				if (!*bio_is_patched) {
1971 					block->orig_bio_bh_private =
1972 					    bio->bi_private;
1973 					block->orig_bio_bh_end_io.bio =
1974 					    bio->bi_end_io;
1975 					block->next_in_same_bio = NULL;
1976 					bio->bi_private = block;
1977 					bio->bi_end_io = btrfsic_bio_end_io;
1978 					*bio_is_patched = 1;
1979 				} else {
1980 					struct btrfsic_block *chained_block =
1981 					    (struct btrfsic_block *)
1982 					    bio->bi_private;
1983 
1984 					BUG_ON(NULL == chained_block);
1985 					block->orig_bio_bh_private =
1986 					    chained_block->orig_bio_bh_private;
1987 					block->orig_bio_bh_end_io.bio =
1988 					    chained_block->orig_bio_bh_end_io.
1989 					    bio;
1990 					block->next_in_same_bio = chained_block;
1991 					bio->bi_private = block;
1992 				}
1993 			} else if (NULL != bh) {
1994 				block->is_iodone = 0;
1995 				block->orig_bio_bh_private = bh->b_private;
1996 				block->orig_bio_bh_end_io.bh = bh->b_end_io;
1997 				block->next_in_same_bio = NULL;
1998 				bh->b_private = block;
1999 				bh->b_end_io = btrfsic_bh_end_io;
2000 			} else {
2001 				block->is_iodone = 1;
2002 				block->orig_bio_bh_private = NULL;
2003 				block->orig_bio_bh_end_io.bio = NULL;
2004 				block->next_in_same_bio = NULL;
2005 			}
2006 		}
2007 
2008 		block->flush_gen = dev_state->last_flush_gen + 1;
2009 		block->submit_bio_bh_rw = submit_bio_bh_rw;
2010 		if (is_metadata) {
2011 			block->logical_bytenr = bytenr;
2012 			block->is_metadata = 1;
2013 			if (block->is_superblock) {
2014 				BUG_ON(PAGE_CACHE_SIZE !=
2015 				       BTRFS_SUPER_INFO_SIZE);
2016 				ret = btrfsic_process_written_superblock(
2017 						state,
2018 						block,
2019 						(struct btrfs_super_block *)
2020 						mapped_datav[0]);
2021 				if (state->print_mask &
2022 				    BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
2023 					printk(KERN_INFO
2024 					"[after new superblock is written]:\n");
2025 					btrfsic_dump_tree_sub(state, block, 0);
2026 				}
2027 			} else {
2028 				block->mirror_num = 0;	/* unknown */
2029 				ret = btrfsic_process_metablock(
2030 						state,
2031 						block,
2032 						&block_ctx,
2033 						0, 0);
2034 			}
2035 			if (ret)
2036 				printk(KERN_INFO
2037 				       "btrfsic: btrfsic_process_metablock"
2038 				       "(root @%llu) failed!\n",
2039 				       dev_bytenr);
2040 		} else {
2041 			block->is_metadata = 0;
2042 			block->mirror_num = 0;	/* unknown */
2043 			block->generation = BTRFSIC_GENERATION_UNKNOWN;
2044 			if (!state->include_extent_data
2045 			    && list_empty(&block->ref_from_list)) {
2046 				/*
2047 				 * disk block is overwritten with extent
2048 				 * data (not meta data) and we are configured
2049 				 * to not include extent data: take the
2050 				 * chance and free the block's memory
2051 				 */
2052 				btrfsic_block_hashtable_remove(block);
2053 				list_del(&block->all_blocks_node);
2054 				btrfsic_block_free(block);
2055 			}
2056 		}
2057 		btrfsic_release_block_ctx(&block_ctx);
2058 	} else {
2059 		/* block has not been found in hash table */
2060 		u64 bytenr;
2061 
2062 		if (!is_metadata) {
2063 			processed_len = state->datablock_size;
2064 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2065 				printk(KERN_INFO "Written block (%s/%llu/?)"
2066 				       " !found in hash table, D.\n",
2067 				       dev_state->name, dev_bytenr);
2068 			if (!state->include_extent_data) {
2069 				/* ignore that written D block */
2070 				goto continue_loop;
2071 			}
2072 
2073 			/* this is getting ugly for the
2074 			 * include_extent_data case... */
2075 			bytenr = 0;	/* unknown */
2076 		} else {
2077 			processed_len = state->metablock_size;
2078 			bytenr = btrfs_stack_header_bytenr(
2079 					(struct btrfs_header *)
2080 					mapped_datav[0]);
2081 			btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
2082 						       dev_bytenr);
2083 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2084 				printk(KERN_INFO
2085 				       "Written block @%llu (%s/%llu/?)"
2086 				       " !found in hash table, M.\n",
2087 				       bytenr, dev_state->name, dev_bytenr);
2088 		}
2089 
2090 		block_ctx.dev = dev_state;
2091 		block_ctx.dev_bytenr = dev_bytenr;
2092 		block_ctx.start = bytenr;
2093 		block_ctx.len = processed_len;
2094 		block_ctx.pagev = NULL;
2095 		block_ctx.mem_to_free = NULL;
2096 		block_ctx.datav = mapped_datav;
2097 
2098 		block = btrfsic_block_alloc();
2099 		if (NULL == block) {
2100 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
2101 			btrfsic_release_block_ctx(&block_ctx);
2102 			goto continue_loop;
2103 		}
2104 		block->dev_state = dev_state;
2105 		block->dev_bytenr = dev_bytenr;
2106 		block->logical_bytenr = bytenr;
2107 		block->is_metadata = is_metadata;
2108 		block->never_written = 0;
2109 		block->iodone_w_error = 0;
2110 		block->mirror_num = 0;	/* unknown */
2111 		block->flush_gen = dev_state->last_flush_gen + 1;
2112 		block->submit_bio_bh_rw = submit_bio_bh_rw;
2113 		if (NULL != bio) {
2114 			block->is_iodone = 0;
2115 			BUG_ON(NULL == bio_is_patched);
2116 			if (!*bio_is_patched) {
2117 				block->orig_bio_bh_private = bio->bi_private;
2118 				block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2119 				block->next_in_same_bio = NULL;
2120 				bio->bi_private = block;
2121 				bio->bi_end_io = btrfsic_bio_end_io;
2122 				*bio_is_patched = 1;
2123 			} else {
2124 				struct btrfsic_block *chained_block =
2125 				    (struct btrfsic_block *)
2126 				    bio->bi_private;
2127 
2128 				BUG_ON(NULL == chained_block);
2129 				block->orig_bio_bh_private =
2130 				    chained_block->orig_bio_bh_private;
2131 				block->orig_bio_bh_end_io.bio =
2132 				    chained_block->orig_bio_bh_end_io.bio;
2133 				block->next_in_same_bio = chained_block;
2134 				bio->bi_private = block;
2135 			}
2136 		} else if (NULL != bh) {
2137 			block->is_iodone = 0;
2138 			block->orig_bio_bh_private = bh->b_private;
2139 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2140 			block->next_in_same_bio = NULL;
2141 			bh->b_private = block;
2142 			bh->b_end_io = btrfsic_bh_end_io;
2143 		} else {
2144 			block->is_iodone = 1;
2145 			block->orig_bio_bh_private = NULL;
2146 			block->orig_bio_bh_end_io.bio = NULL;
2147 			block->next_in_same_bio = NULL;
2148 		}
2149 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2150 			printk(KERN_INFO
2151 			       "New written %c-block @%llu (%s/%llu/%d)\n",
2152 			       is_metadata ? 'M' : 'D',
2153 			       block->logical_bytenr, block->dev_state->name,
2154 			       block->dev_bytenr, block->mirror_num);
2155 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2156 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2157 
2158 		if (is_metadata) {
2159 			ret = btrfsic_process_metablock(state, block,
2160 							&block_ctx, 0, 0);
2161 			if (ret)
2162 				printk(KERN_INFO
2163 				       "btrfsic: process_metablock(root @%llu)"
2164 				       " failed!\n",
2165 				       dev_bytenr);
2166 		}
2167 		btrfsic_release_block_ctx(&block_ctx);
2168 	}
2169 
2170 continue_loop:
2171 	BUG_ON(!processed_len);
2172 	dev_bytenr += processed_len;
2173 	mapped_datav += processed_len >> PAGE_CACHE_SHIFT;
2174 	num_pages -= processed_len >> PAGE_CACHE_SHIFT;
2175 	goto again;
2176 }
2177 
2178 static void btrfsic_bio_end_io(struct bio *bp)
2179 {
2180 	struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2181 	int iodone_w_error;
2182 
2183 	/* mutex is not held! This is not save if IO is not yet completed
2184 	 * on umount */
2185 	iodone_w_error = 0;
2186 	if (bp->bi_error)
2187 		iodone_w_error = 1;
2188 
2189 	BUG_ON(NULL == block);
2190 	bp->bi_private = block->orig_bio_bh_private;
2191 	bp->bi_end_io = block->orig_bio_bh_end_io.bio;
2192 
2193 	do {
2194 		struct btrfsic_block *next_block;
2195 		struct btrfsic_dev_state *const dev_state = block->dev_state;
2196 
2197 		if ((dev_state->state->print_mask &
2198 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2199 			printk(KERN_INFO
2200 			       "bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2201 			       bp->bi_error,
2202 			       btrfsic_get_block_type(dev_state->state, block),
2203 			       block->logical_bytenr, dev_state->name,
2204 			       block->dev_bytenr, block->mirror_num);
2205 		next_block = block->next_in_same_bio;
2206 		block->iodone_w_error = iodone_w_error;
2207 		if (block->submit_bio_bh_rw & REQ_FLUSH) {
2208 			dev_state->last_flush_gen++;
2209 			if ((dev_state->state->print_mask &
2210 			     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2211 				printk(KERN_INFO
2212 				       "bio_end_io() new %s flush_gen=%llu\n",
2213 				       dev_state->name,
2214 				       dev_state->last_flush_gen);
2215 		}
2216 		if (block->submit_bio_bh_rw & REQ_FUA)
2217 			block->flush_gen = 0; /* FUA completed means block is
2218 					       * on disk */
2219 		block->is_iodone = 1; /* for FLUSH, this releases the block */
2220 		block = next_block;
2221 	} while (NULL != block);
2222 
2223 	bp->bi_end_io(bp);
2224 }
2225 
2226 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate)
2227 {
2228 	struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private;
2229 	int iodone_w_error = !uptodate;
2230 	struct btrfsic_dev_state *dev_state;
2231 
2232 	BUG_ON(NULL == block);
2233 	dev_state = block->dev_state;
2234 	if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2235 		printk(KERN_INFO
2236 		       "bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n",
2237 		       iodone_w_error,
2238 		       btrfsic_get_block_type(dev_state->state, block),
2239 		       block->logical_bytenr, block->dev_state->name,
2240 		       block->dev_bytenr, block->mirror_num);
2241 
2242 	block->iodone_w_error = iodone_w_error;
2243 	if (block->submit_bio_bh_rw & REQ_FLUSH) {
2244 		dev_state->last_flush_gen++;
2245 		if ((dev_state->state->print_mask &
2246 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2247 			printk(KERN_INFO
2248 			       "bh_end_io() new %s flush_gen=%llu\n",
2249 			       dev_state->name, dev_state->last_flush_gen);
2250 	}
2251 	if (block->submit_bio_bh_rw & REQ_FUA)
2252 		block->flush_gen = 0; /* FUA completed means block is on disk */
2253 
2254 	bh->b_private = block->orig_bio_bh_private;
2255 	bh->b_end_io = block->orig_bio_bh_end_io.bh;
2256 	block->is_iodone = 1; /* for FLUSH, this releases the block */
2257 	bh->b_end_io(bh, uptodate);
2258 }
2259 
2260 static int btrfsic_process_written_superblock(
2261 		struct btrfsic_state *state,
2262 		struct btrfsic_block *const superblock,
2263 		struct btrfs_super_block *const super_hdr)
2264 {
2265 	int pass;
2266 
2267 	superblock->generation = btrfs_super_generation(super_hdr);
2268 	if (!(superblock->generation > state->max_superblock_generation ||
2269 	      0 == state->max_superblock_generation)) {
2270 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2271 			printk(KERN_INFO
2272 			       "btrfsic: superblock @%llu (%s/%llu/%d)"
2273 			       " with old gen %llu <= %llu\n",
2274 			       superblock->logical_bytenr,
2275 			       superblock->dev_state->name,
2276 			       superblock->dev_bytenr, superblock->mirror_num,
2277 			       btrfs_super_generation(super_hdr),
2278 			       state->max_superblock_generation);
2279 	} else {
2280 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2281 			printk(KERN_INFO
2282 			       "btrfsic: got new superblock @%llu (%s/%llu/%d)"
2283 			       " with new gen %llu > %llu\n",
2284 			       superblock->logical_bytenr,
2285 			       superblock->dev_state->name,
2286 			       superblock->dev_bytenr, superblock->mirror_num,
2287 			       btrfs_super_generation(super_hdr),
2288 			       state->max_superblock_generation);
2289 
2290 		state->max_superblock_generation =
2291 		    btrfs_super_generation(super_hdr);
2292 		state->latest_superblock = superblock;
2293 	}
2294 
2295 	for (pass = 0; pass < 3; pass++) {
2296 		int ret;
2297 		u64 next_bytenr;
2298 		struct btrfsic_block *next_block;
2299 		struct btrfsic_block_data_ctx tmp_next_block_ctx;
2300 		struct btrfsic_block_link *l;
2301 		int num_copies;
2302 		int mirror_num;
2303 		const char *additional_string = NULL;
2304 		struct btrfs_disk_key tmp_disk_key = {0};
2305 
2306 		btrfs_set_disk_key_objectid(&tmp_disk_key,
2307 					    BTRFS_ROOT_ITEM_KEY);
2308 		btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2309 
2310 		switch (pass) {
2311 		case 0:
2312 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2313 						    BTRFS_ROOT_TREE_OBJECTID);
2314 			additional_string = "root ";
2315 			next_bytenr = btrfs_super_root(super_hdr);
2316 			if (state->print_mask &
2317 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2318 				printk(KERN_INFO "root@%llu\n", next_bytenr);
2319 			break;
2320 		case 1:
2321 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2322 						    BTRFS_CHUNK_TREE_OBJECTID);
2323 			additional_string = "chunk ";
2324 			next_bytenr = btrfs_super_chunk_root(super_hdr);
2325 			if (state->print_mask &
2326 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2327 				printk(KERN_INFO "chunk@%llu\n", next_bytenr);
2328 			break;
2329 		case 2:
2330 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2331 						    BTRFS_TREE_LOG_OBJECTID);
2332 			additional_string = "log ";
2333 			next_bytenr = btrfs_super_log_root(super_hdr);
2334 			if (0 == next_bytenr)
2335 				continue;
2336 			if (state->print_mask &
2337 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2338 				printk(KERN_INFO "log@%llu\n", next_bytenr);
2339 			break;
2340 		}
2341 
2342 		num_copies =
2343 		    btrfs_num_copies(state->root->fs_info,
2344 				     next_bytenr, BTRFS_SUPER_INFO_SIZE);
2345 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2346 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
2347 			       next_bytenr, num_copies);
2348 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2349 			int was_created;
2350 
2351 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2352 				printk(KERN_INFO
2353 				       "btrfsic_process_written_superblock("
2354 				       "mirror_num=%d)\n", mirror_num);
2355 			ret = btrfsic_map_block(state, next_bytenr,
2356 						BTRFS_SUPER_INFO_SIZE,
2357 						&tmp_next_block_ctx,
2358 						mirror_num);
2359 			if (ret) {
2360 				printk(KERN_INFO
2361 				       "btrfsic: btrfsic_map_block(@%llu,"
2362 				       " mirror=%d) failed!\n",
2363 				       next_bytenr, mirror_num);
2364 				return -1;
2365 			}
2366 
2367 			next_block = btrfsic_block_lookup_or_add(
2368 					state,
2369 					&tmp_next_block_ctx,
2370 					additional_string,
2371 					1, 0, 1,
2372 					mirror_num,
2373 					&was_created);
2374 			if (NULL == next_block) {
2375 				printk(KERN_INFO
2376 				       "btrfsic: error, kmalloc failed!\n");
2377 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
2378 				return -1;
2379 			}
2380 
2381 			next_block->disk_key = tmp_disk_key;
2382 			if (was_created)
2383 				next_block->generation =
2384 				    BTRFSIC_GENERATION_UNKNOWN;
2385 			l = btrfsic_block_link_lookup_or_add(
2386 					state,
2387 					&tmp_next_block_ctx,
2388 					next_block,
2389 					superblock,
2390 					BTRFSIC_GENERATION_UNKNOWN);
2391 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
2392 			if (NULL == l)
2393 				return -1;
2394 		}
2395 	}
2396 
2397 	if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2398 		btrfsic_dump_tree(state);
2399 
2400 	return 0;
2401 }
2402 
2403 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2404 					struct btrfsic_block *const block,
2405 					int recursion_level)
2406 {
2407 	const struct btrfsic_block_link *l;
2408 	int ret = 0;
2409 
2410 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2411 		/*
2412 		 * Note that this situation can happen and does not
2413 		 * indicate an error in regular cases. It happens
2414 		 * when disk blocks are freed and later reused.
2415 		 * The check-integrity module is not aware of any
2416 		 * block free operations, it just recognizes block
2417 		 * write operations. Therefore it keeps the linkage
2418 		 * information for a block until a block is
2419 		 * rewritten. This can temporarily cause incorrect
2420 		 * and even circular linkage informations. This
2421 		 * causes no harm unless such blocks are referenced
2422 		 * by the most recent super block.
2423 		 */
2424 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2425 			printk(KERN_INFO
2426 			       "btrfsic: abort cyclic linkage (case 1).\n");
2427 
2428 		return ret;
2429 	}
2430 
2431 	/*
2432 	 * This algorithm is recursive because the amount of used stack
2433 	 * space is very small and the max recursion depth is limited.
2434 	 */
2435 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2436 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2437 			printk(KERN_INFO
2438 			       "rl=%d, %c @%llu (%s/%llu/%d)"
2439 			       " %u* refers to %c @%llu (%s/%llu/%d)\n",
2440 			       recursion_level,
2441 			       btrfsic_get_block_type(state, block),
2442 			       block->logical_bytenr, block->dev_state->name,
2443 			       block->dev_bytenr, block->mirror_num,
2444 			       l->ref_cnt,
2445 			       btrfsic_get_block_type(state, l->block_ref_to),
2446 			       l->block_ref_to->logical_bytenr,
2447 			       l->block_ref_to->dev_state->name,
2448 			       l->block_ref_to->dev_bytenr,
2449 			       l->block_ref_to->mirror_num);
2450 		if (l->block_ref_to->never_written) {
2451 			printk(KERN_INFO "btrfs: attempt to write superblock"
2452 			       " which references block %c @%llu (%s/%llu/%d)"
2453 			       " which is never written!\n",
2454 			       btrfsic_get_block_type(state, l->block_ref_to),
2455 			       l->block_ref_to->logical_bytenr,
2456 			       l->block_ref_to->dev_state->name,
2457 			       l->block_ref_to->dev_bytenr,
2458 			       l->block_ref_to->mirror_num);
2459 			ret = -1;
2460 		} else if (!l->block_ref_to->is_iodone) {
2461 			printk(KERN_INFO "btrfs: attempt to write superblock"
2462 			       " which references block %c @%llu (%s/%llu/%d)"
2463 			       " which is not yet iodone!\n",
2464 			       btrfsic_get_block_type(state, l->block_ref_to),
2465 			       l->block_ref_to->logical_bytenr,
2466 			       l->block_ref_to->dev_state->name,
2467 			       l->block_ref_to->dev_bytenr,
2468 			       l->block_ref_to->mirror_num);
2469 			ret = -1;
2470 		} else if (l->block_ref_to->iodone_w_error) {
2471 			printk(KERN_INFO "btrfs: attempt to write superblock"
2472 			       " which references block %c @%llu (%s/%llu/%d)"
2473 			       " which has write error!\n",
2474 			       btrfsic_get_block_type(state, l->block_ref_to),
2475 			       l->block_ref_to->logical_bytenr,
2476 			       l->block_ref_to->dev_state->name,
2477 			       l->block_ref_to->dev_bytenr,
2478 			       l->block_ref_to->mirror_num);
2479 			ret = -1;
2480 		} else if (l->parent_generation !=
2481 			   l->block_ref_to->generation &&
2482 			   BTRFSIC_GENERATION_UNKNOWN !=
2483 			   l->parent_generation &&
2484 			   BTRFSIC_GENERATION_UNKNOWN !=
2485 			   l->block_ref_to->generation) {
2486 			printk(KERN_INFO "btrfs: attempt to write superblock"
2487 			       " which references block %c @%llu (%s/%llu/%d)"
2488 			       " with generation %llu !="
2489 			       " parent generation %llu!\n",
2490 			       btrfsic_get_block_type(state, l->block_ref_to),
2491 			       l->block_ref_to->logical_bytenr,
2492 			       l->block_ref_to->dev_state->name,
2493 			       l->block_ref_to->dev_bytenr,
2494 			       l->block_ref_to->mirror_num,
2495 			       l->block_ref_to->generation,
2496 			       l->parent_generation);
2497 			ret = -1;
2498 		} else if (l->block_ref_to->flush_gen >
2499 			   l->block_ref_to->dev_state->last_flush_gen) {
2500 			printk(KERN_INFO "btrfs: attempt to write superblock"
2501 			       " which references block %c @%llu (%s/%llu/%d)"
2502 			       " which is not flushed out of disk's write cache"
2503 			       " (block flush_gen=%llu,"
2504 			       " dev->flush_gen=%llu)!\n",
2505 			       btrfsic_get_block_type(state, l->block_ref_to),
2506 			       l->block_ref_to->logical_bytenr,
2507 			       l->block_ref_to->dev_state->name,
2508 			       l->block_ref_to->dev_bytenr,
2509 			       l->block_ref_to->mirror_num, block->flush_gen,
2510 			       l->block_ref_to->dev_state->last_flush_gen);
2511 			ret = -1;
2512 		} else if (-1 == btrfsic_check_all_ref_blocks(state,
2513 							      l->block_ref_to,
2514 							      recursion_level +
2515 							      1)) {
2516 			ret = -1;
2517 		}
2518 	}
2519 
2520 	return ret;
2521 }
2522 
2523 static int btrfsic_is_block_ref_by_superblock(
2524 		const struct btrfsic_state *state,
2525 		const struct btrfsic_block *block,
2526 		int recursion_level)
2527 {
2528 	const struct btrfsic_block_link *l;
2529 
2530 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2531 		/* refer to comment at "abort cyclic linkage (case 1)" */
2532 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2533 			printk(KERN_INFO
2534 			       "btrfsic: abort cyclic linkage (case 2).\n");
2535 
2536 		return 0;
2537 	}
2538 
2539 	/*
2540 	 * This algorithm is recursive because the amount of used stack space
2541 	 * is very small and the max recursion depth is limited.
2542 	 */
2543 	list_for_each_entry(l, &block->ref_from_list, node_ref_from) {
2544 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2545 			printk(KERN_INFO
2546 			       "rl=%d, %c @%llu (%s/%llu/%d)"
2547 			       " is ref %u* from %c @%llu (%s/%llu/%d)\n",
2548 			       recursion_level,
2549 			       btrfsic_get_block_type(state, block),
2550 			       block->logical_bytenr, block->dev_state->name,
2551 			       block->dev_bytenr, block->mirror_num,
2552 			       l->ref_cnt,
2553 			       btrfsic_get_block_type(state, l->block_ref_from),
2554 			       l->block_ref_from->logical_bytenr,
2555 			       l->block_ref_from->dev_state->name,
2556 			       l->block_ref_from->dev_bytenr,
2557 			       l->block_ref_from->mirror_num);
2558 		if (l->block_ref_from->is_superblock &&
2559 		    state->latest_superblock->dev_bytenr ==
2560 		    l->block_ref_from->dev_bytenr &&
2561 		    state->latest_superblock->dev_state->bdev ==
2562 		    l->block_ref_from->dev_state->bdev)
2563 			return 1;
2564 		else if (btrfsic_is_block_ref_by_superblock(state,
2565 							    l->block_ref_from,
2566 							    recursion_level +
2567 							    1))
2568 			return 1;
2569 	}
2570 
2571 	return 0;
2572 }
2573 
2574 static void btrfsic_print_add_link(const struct btrfsic_state *state,
2575 				   const struct btrfsic_block_link *l)
2576 {
2577 	printk(KERN_INFO
2578 	       "Add %u* link from %c @%llu (%s/%llu/%d)"
2579 	       " to %c @%llu (%s/%llu/%d).\n",
2580 	       l->ref_cnt,
2581 	       btrfsic_get_block_type(state, l->block_ref_from),
2582 	       l->block_ref_from->logical_bytenr,
2583 	       l->block_ref_from->dev_state->name,
2584 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2585 	       btrfsic_get_block_type(state, l->block_ref_to),
2586 	       l->block_ref_to->logical_bytenr,
2587 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2588 	       l->block_ref_to->mirror_num);
2589 }
2590 
2591 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2592 				   const struct btrfsic_block_link *l)
2593 {
2594 	printk(KERN_INFO
2595 	       "Rem %u* link from %c @%llu (%s/%llu/%d)"
2596 	       " to %c @%llu (%s/%llu/%d).\n",
2597 	       l->ref_cnt,
2598 	       btrfsic_get_block_type(state, l->block_ref_from),
2599 	       l->block_ref_from->logical_bytenr,
2600 	       l->block_ref_from->dev_state->name,
2601 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2602 	       btrfsic_get_block_type(state, l->block_ref_to),
2603 	       l->block_ref_to->logical_bytenr,
2604 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2605 	       l->block_ref_to->mirror_num);
2606 }
2607 
2608 static char btrfsic_get_block_type(const struct btrfsic_state *state,
2609 				   const struct btrfsic_block *block)
2610 {
2611 	if (block->is_superblock &&
2612 	    state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2613 	    state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2614 		return 'S';
2615 	else if (block->is_superblock)
2616 		return 's';
2617 	else if (block->is_metadata)
2618 		return 'M';
2619 	else
2620 		return 'D';
2621 }
2622 
2623 static void btrfsic_dump_tree(const struct btrfsic_state *state)
2624 {
2625 	btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2626 }
2627 
2628 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2629 				  const struct btrfsic_block *block,
2630 				  int indent_level)
2631 {
2632 	const struct btrfsic_block_link *l;
2633 	int indent_add;
2634 	static char buf[80];
2635 	int cursor_position;
2636 
2637 	/*
2638 	 * Should better fill an on-stack buffer with a complete line and
2639 	 * dump it at once when it is time to print a newline character.
2640 	 */
2641 
2642 	/*
2643 	 * This algorithm is recursive because the amount of used stack space
2644 	 * is very small and the max recursion depth is limited.
2645 	 */
2646 	indent_add = sprintf(buf, "%c-%llu(%s/%llu/%d)",
2647 			     btrfsic_get_block_type(state, block),
2648 			     block->logical_bytenr, block->dev_state->name,
2649 			     block->dev_bytenr, block->mirror_num);
2650 	if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2651 		printk("[...]\n");
2652 		return;
2653 	}
2654 	printk(buf);
2655 	indent_level += indent_add;
2656 	if (list_empty(&block->ref_to_list)) {
2657 		printk("\n");
2658 		return;
2659 	}
2660 	if (block->mirror_num > 1 &&
2661 	    !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2662 		printk(" [...]\n");
2663 		return;
2664 	}
2665 
2666 	cursor_position = indent_level;
2667 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2668 		while (cursor_position < indent_level) {
2669 			printk(" ");
2670 			cursor_position++;
2671 		}
2672 		if (l->ref_cnt > 1)
2673 			indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2674 		else
2675 			indent_add = sprintf(buf, " --> ");
2676 		if (indent_level + indent_add >
2677 		    BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2678 			printk("[...]\n");
2679 			cursor_position = 0;
2680 			continue;
2681 		}
2682 
2683 		printk(buf);
2684 
2685 		btrfsic_dump_tree_sub(state, l->block_ref_to,
2686 				      indent_level + indent_add);
2687 		cursor_position = 0;
2688 	}
2689 }
2690 
2691 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2692 		struct btrfsic_state *state,
2693 		struct btrfsic_block_data_ctx *next_block_ctx,
2694 		struct btrfsic_block *next_block,
2695 		struct btrfsic_block *from_block,
2696 		u64 parent_generation)
2697 {
2698 	struct btrfsic_block_link *l;
2699 
2700 	l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2701 						next_block_ctx->dev_bytenr,
2702 						from_block->dev_state->bdev,
2703 						from_block->dev_bytenr,
2704 						&state->block_link_hashtable);
2705 	if (NULL == l) {
2706 		l = btrfsic_block_link_alloc();
2707 		if (NULL == l) {
2708 			printk(KERN_INFO
2709 			       "btrfsic: error, kmalloc" " failed!\n");
2710 			return NULL;
2711 		}
2712 
2713 		l->block_ref_to = next_block;
2714 		l->block_ref_from = from_block;
2715 		l->ref_cnt = 1;
2716 		l->parent_generation = parent_generation;
2717 
2718 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2719 			btrfsic_print_add_link(state, l);
2720 
2721 		list_add(&l->node_ref_to, &from_block->ref_to_list);
2722 		list_add(&l->node_ref_from, &next_block->ref_from_list);
2723 
2724 		btrfsic_block_link_hashtable_add(l,
2725 						 &state->block_link_hashtable);
2726 	} else {
2727 		l->ref_cnt++;
2728 		l->parent_generation = parent_generation;
2729 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2730 			btrfsic_print_add_link(state, l);
2731 	}
2732 
2733 	return l;
2734 }
2735 
2736 static struct btrfsic_block *btrfsic_block_lookup_or_add(
2737 		struct btrfsic_state *state,
2738 		struct btrfsic_block_data_ctx *block_ctx,
2739 		const char *additional_string,
2740 		int is_metadata,
2741 		int is_iodone,
2742 		int never_written,
2743 		int mirror_num,
2744 		int *was_created)
2745 {
2746 	struct btrfsic_block *block;
2747 
2748 	block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2749 					       block_ctx->dev_bytenr,
2750 					       &state->block_hashtable);
2751 	if (NULL == block) {
2752 		struct btrfsic_dev_state *dev_state;
2753 
2754 		block = btrfsic_block_alloc();
2755 		if (NULL == block) {
2756 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
2757 			return NULL;
2758 		}
2759 		dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev);
2760 		if (NULL == dev_state) {
2761 			printk(KERN_INFO
2762 			       "btrfsic: error, lookup dev_state failed!\n");
2763 			btrfsic_block_free(block);
2764 			return NULL;
2765 		}
2766 		block->dev_state = dev_state;
2767 		block->dev_bytenr = block_ctx->dev_bytenr;
2768 		block->logical_bytenr = block_ctx->start;
2769 		block->is_metadata = is_metadata;
2770 		block->is_iodone = is_iodone;
2771 		block->never_written = never_written;
2772 		block->mirror_num = mirror_num;
2773 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2774 			printk(KERN_INFO
2775 			       "New %s%c-block @%llu (%s/%llu/%d)\n",
2776 			       additional_string,
2777 			       btrfsic_get_block_type(state, block),
2778 			       block->logical_bytenr, dev_state->name,
2779 			       block->dev_bytenr, mirror_num);
2780 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2781 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2782 		if (NULL != was_created)
2783 			*was_created = 1;
2784 	} else {
2785 		if (NULL != was_created)
2786 			*was_created = 0;
2787 	}
2788 
2789 	return block;
2790 }
2791 
2792 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2793 					   u64 bytenr,
2794 					   struct btrfsic_dev_state *dev_state,
2795 					   u64 dev_bytenr)
2796 {
2797 	int num_copies;
2798 	int mirror_num;
2799 	int ret;
2800 	struct btrfsic_block_data_ctx block_ctx;
2801 	int match = 0;
2802 
2803 	num_copies = btrfs_num_copies(state->root->fs_info,
2804 				      bytenr, state->metablock_size);
2805 
2806 	for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2807 		ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2808 					&block_ctx, mirror_num);
2809 		if (ret) {
2810 			printk(KERN_INFO "btrfsic:"
2811 			       " btrfsic_map_block(logical @%llu,"
2812 			       " mirror %d) failed!\n",
2813 			       bytenr, mirror_num);
2814 			continue;
2815 		}
2816 
2817 		if (dev_state->bdev == block_ctx.dev->bdev &&
2818 		    dev_bytenr == block_ctx.dev_bytenr) {
2819 			match++;
2820 			btrfsic_release_block_ctx(&block_ctx);
2821 			break;
2822 		}
2823 		btrfsic_release_block_ctx(&block_ctx);
2824 	}
2825 
2826 	if (WARN_ON(!match)) {
2827 		printk(KERN_INFO "btrfs: attempt to write M-block which contains logical bytenr that doesn't map to dev+physical bytenr of submit_bio,"
2828 		       " buffer->log_bytenr=%llu, submit_bio(bdev=%s,"
2829 		       " phys_bytenr=%llu)!\n",
2830 		       bytenr, dev_state->name, dev_bytenr);
2831 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2832 			ret = btrfsic_map_block(state, bytenr,
2833 						state->metablock_size,
2834 						&block_ctx, mirror_num);
2835 			if (ret)
2836 				continue;
2837 
2838 			printk(KERN_INFO "Read logical bytenr @%llu maps to"
2839 			       " (%s/%llu/%d)\n",
2840 			       bytenr, block_ctx.dev->name,
2841 			       block_ctx.dev_bytenr, mirror_num);
2842 		}
2843 	}
2844 }
2845 
2846 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
2847 		struct block_device *bdev)
2848 {
2849 	struct btrfsic_dev_state *ds;
2850 
2851 	ds = btrfsic_dev_state_hashtable_lookup(bdev,
2852 						&btrfsic_dev_state_hashtable);
2853 	return ds;
2854 }
2855 
2856 int btrfsic_submit_bh(int rw, struct buffer_head *bh)
2857 {
2858 	struct btrfsic_dev_state *dev_state;
2859 
2860 	if (!btrfsic_is_initialized)
2861 		return submit_bh(rw, bh);
2862 
2863 	mutex_lock(&btrfsic_mutex);
2864 	/* since btrfsic_submit_bh() might also be called before
2865 	 * btrfsic_mount(), this might return NULL */
2866 	dev_state = btrfsic_dev_state_lookup(bh->b_bdev);
2867 
2868 	/* Only called to write the superblock (incl. FLUSH/FUA) */
2869 	if (NULL != dev_state &&
2870 	    (rw & WRITE) && bh->b_size > 0) {
2871 		u64 dev_bytenr;
2872 
2873 		dev_bytenr = 4096 * bh->b_blocknr;
2874 		if (dev_state->state->print_mask &
2875 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2876 			printk(KERN_INFO
2877 			       "submit_bh(rw=0x%x, blocknr=%llu (bytenr %llu),"
2878 			       " size=%zu, data=%p, bdev=%p)\n",
2879 			       rw, (unsigned long long)bh->b_blocknr,
2880 			       dev_bytenr, bh->b_size, bh->b_data, bh->b_bdev);
2881 		btrfsic_process_written_block(dev_state, dev_bytenr,
2882 					      &bh->b_data, 1, NULL,
2883 					      NULL, bh, rw);
2884 	} else if (NULL != dev_state && (rw & REQ_FLUSH)) {
2885 		if (dev_state->state->print_mask &
2886 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2887 			printk(KERN_INFO
2888 			       "submit_bh(rw=0x%x FLUSH, bdev=%p)\n",
2889 			       rw, bh->b_bdev);
2890 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2891 			if ((dev_state->state->print_mask &
2892 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2893 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2894 				printk(KERN_INFO
2895 				       "btrfsic_submit_bh(%s) with FLUSH"
2896 				       " but dummy block already in use"
2897 				       " (ignored)!\n",
2898 				       dev_state->name);
2899 		} else {
2900 			struct btrfsic_block *const block =
2901 				&dev_state->dummy_block_for_bio_bh_flush;
2902 
2903 			block->is_iodone = 0;
2904 			block->never_written = 0;
2905 			block->iodone_w_error = 0;
2906 			block->flush_gen = dev_state->last_flush_gen + 1;
2907 			block->submit_bio_bh_rw = rw;
2908 			block->orig_bio_bh_private = bh->b_private;
2909 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2910 			block->next_in_same_bio = NULL;
2911 			bh->b_private = block;
2912 			bh->b_end_io = btrfsic_bh_end_io;
2913 		}
2914 	}
2915 	mutex_unlock(&btrfsic_mutex);
2916 	return submit_bh(rw, bh);
2917 }
2918 
2919 static void __btrfsic_submit_bio(int rw, struct bio *bio)
2920 {
2921 	struct btrfsic_dev_state *dev_state;
2922 
2923 	if (!btrfsic_is_initialized)
2924 		return;
2925 
2926 	mutex_lock(&btrfsic_mutex);
2927 	/* since btrfsic_submit_bio() is also called before
2928 	 * btrfsic_mount(), this might return NULL */
2929 	dev_state = btrfsic_dev_state_lookup(bio->bi_bdev);
2930 	if (NULL != dev_state &&
2931 	    (rw & WRITE) && NULL != bio->bi_io_vec) {
2932 		unsigned int i;
2933 		u64 dev_bytenr;
2934 		u64 cur_bytenr;
2935 		int bio_is_patched;
2936 		char **mapped_datav;
2937 
2938 		dev_bytenr = 512 * bio->bi_iter.bi_sector;
2939 		bio_is_patched = 0;
2940 		if (dev_state->state->print_mask &
2941 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2942 			printk(KERN_INFO
2943 			       "submit_bio(rw=0x%x, bi_vcnt=%u,"
2944 			       " bi_sector=%llu (bytenr %llu), bi_bdev=%p)\n",
2945 			       rw, bio->bi_vcnt,
2946 			       (unsigned long long)bio->bi_iter.bi_sector,
2947 			       dev_bytenr, bio->bi_bdev);
2948 
2949 		mapped_datav = kmalloc_array(bio->bi_vcnt,
2950 					     sizeof(*mapped_datav), GFP_NOFS);
2951 		if (!mapped_datav)
2952 			goto leave;
2953 		cur_bytenr = dev_bytenr;
2954 		for (i = 0; i < bio->bi_vcnt; i++) {
2955 			BUG_ON(bio->bi_io_vec[i].bv_len != PAGE_CACHE_SIZE);
2956 			mapped_datav[i] = kmap(bio->bi_io_vec[i].bv_page);
2957 			if (!mapped_datav[i]) {
2958 				while (i > 0) {
2959 					i--;
2960 					kunmap(bio->bi_io_vec[i].bv_page);
2961 				}
2962 				kfree(mapped_datav);
2963 				goto leave;
2964 			}
2965 			if (dev_state->state->print_mask &
2966 			    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
2967 				printk(KERN_INFO
2968 				       "#%u: bytenr=%llu, len=%u, offset=%u\n",
2969 				       i, cur_bytenr, bio->bi_io_vec[i].bv_len,
2970 				       bio->bi_io_vec[i].bv_offset);
2971 			cur_bytenr += bio->bi_io_vec[i].bv_len;
2972 		}
2973 		btrfsic_process_written_block(dev_state, dev_bytenr,
2974 					      mapped_datav, bio->bi_vcnt,
2975 					      bio, &bio_is_patched,
2976 					      NULL, rw);
2977 		while (i > 0) {
2978 			i--;
2979 			kunmap(bio->bi_io_vec[i].bv_page);
2980 		}
2981 		kfree(mapped_datav);
2982 	} else if (NULL != dev_state && (rw & REQ_FLUSH)) {
2983 		if (dev_state->state->print_mask &
2984 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2985 			printk(KERN_INFO
2986 			       "submit_bio(rw=0x%x FLUSH, bdev=%p)\n",
2987 			       rw, bio->bi_bdev);
2988 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2989 			if ((dev_state->state->print_mask &
2990 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2991 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2992 				printk(KERN_INFO
2993 				       "btrfsic_submit_bio(%s) with FLUSH"
2994 				       " but dummy block already in use"
2995 				       " (ignored)!\n",
2996 				       dev_state->name);
2997 		} else {
2998 			struct btrfsic_block *const block =
2999 				&dev_state->dummy_block_for_bio_bh_flush;
3000 
3001 			block->is_iodone = 0;
3002 			block->never_written = 0;
3003 			block->iodone_w_error = 0;
3004 			block->flush_gen = dev_state->last_flush_gen + 1;
3005 			block->submit_bio_bh_rw = rw;
3006 			block->orig_bio_bh_private = bio->bi_private;
3007 			block->orig_bio_bh_end_io.bio = bio->bi_end_io;
3008 			block->next_in_same_bio = NULL;
3009 			bio->bi_private = block;
3010 			bio->bi_end_io = btrfsic_bio_end_io;
3011 		}
3012 	}
3013 leave:
3014 	mutex_unlock(&btrfsic_mutex);
3015 }
3016 
3017 void btrfsic_submit_bio(int rw, struct bio *bio)
3018 {
3019 	__btrfsic_submit_bio(rw, bio);
3020 	submit_bio(rw, bio);
3021 }
3022 
3023 int btrfsic_submit_bio_wait(int rw, struct bio *bio)
3024 {
3025 	__btrfsic_submit_bio(rw, bio);
3026 	return submit_bio_wait(rw, bio);
3027 }
3028 
3029 int btrfsic_mount(struct btrfs_root *root,
3030 		  struct btrfs_fs_devices *fs_devices,
3031 		  int including_extent_data, u32 print_mask)
3032 {
3033 	int ret;
3034 	struct btrfsic_state *state;
3035 	struct list_head *dev_head = &fs_devices->devices;
3036 	struct btrfs_device *device;
3037 
3038 	if (root->nodesize & ((u64)PAGE_CACHE_SIZE - 1)) {
3039 		printk(KERN_INFO
3040 		       "btrfsic: cannot handle nodesize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3041 		       root->nodesize, PAGE_CACHE_SIZE);
3042 		return -1;
3043 	}
3044 	if (root->sectorsize & ((u64)PAGE_CACHE_SIZE - 1)) {
3045 		printk(KERN_INFO
3046 		       "btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3047 		       root->sectorsize, PAGE_CACHE_SIZE);
3048 		return -1;
3049 	}
3050 	state = kzalloc(sizeof(*state), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
3051 	if (!state) {
3052 		state = vzalloc(sizeof(*state));
3053 		if (!state) {
3054 			printk(KERN_INFO "btrfs check-integrity: vzalloc() failed!\n");
3055 			return -1;
3056 		}
3057 	}
3058 
3059 	if (!btrfsic_is_initialized) {
3060 		mutex_init(&btrfsic_mutex);
3061 		btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
3062 		btrfsic_is_initialized = 1;
3063 	}
3064 	mutex_lock(&btrfsic_mutex);
3065 	state->root = root;
3066 	state->print_mask = print_mask;
3067 	state->include_extent_data = including_extent_data;
3068 	state->csum_size = 0;
3069 	state->metablock_size = root->nodesize;
3070 	state->datablock_size = root->sectorsize;
3071 	INIT_LIST_HEAD(&state->all_blocks_list);
3072 	btrfsic_block_hashtable_init(&state->block_hashtable);
3073 	btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
3074 	state->max_superblock_generation = 0;
3075 	state->latest_superblock = NULL;
3076 
3077 	list_for_each_entry(device, dev_head, dev_list) {
3078 		struct btrfsic_dev_state *ds;
3079 		char *p;
3080 
3081 		if (!device->bdev || !device->name)
3082 			continue;
3083 
3084 		ds = btrfsic_dev_state_alloc();
3085 		if (NULL == ds) {
3086 			printk(KERN_INFO
3087 			       "btrfs check-integrity: kmalloc() failed!\n");
3088 			mutex_unlock(&btrfsic_mutex);
3089 			return -1;
3090 		}
3091 		ds->bdev = device->bdev;
3092 		ds->state = state;
3093 		bdevname(ds->bdev, ds->name);
3094 		ds->name[BDEVNAME_SIZE - 1] = '\0';
3095 		for (p = ds->name; *p != '\0'; p++);
3096 		while (p > ds->name && *p != '/')
3097 			p--;
3098 		if (*p == '/')
3099 			p++;
3100 		strlcpy(ds->name, p, sizeof(ds->name));
3101 		btrfsic_dev_state_hashtable_add(ds,
3102 						&btrfsic_dev_state_hashtable);
3103 	}
3104 
3105 	ret = btrfsic_process_superblock(state, fs_devices);
3106 	if (0 != ret) {
3107 		mutex_unlock(&btrfsic_mutex);
3108 		btrfsic_unmount(root, fs_devices);
3109 		return ret;
3110 	}
3111 
3112 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
3113 		btrfsic_dump_database(state);
3114 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
3115 		btrfsic_dump_tree(state);
3116 
3117 	mutex_unlock(&btrfsic_mutex);
3118 	return 0;
3119 }
3120 
3121 void btrfsic_unmount(struct btrfs_root *root,
3122 		     struct btrfs_fs_devices *fs_devices)
3123 {
3124 	struct btrfsic_block *b_all, *tmp_all;
3125 	struct btrfsic_state *state;
3126 	struct list_head *dev_head = &fs_devices->devices;
3127 	struct btrfs_device *device;
3128 
3129 	if (!btrfsic_is_initialized)
3130 		return;
3131 
3132 	mutex_lock(&btrfsic_mutex);
3133 
3134 	state = NULL;
3135 	list_for_each_entry(device, dev_head, dev_list) {
3136 		struct btrfsic_dev_state *ds;
3137 
3138 		if (!device->bdev || !device->name)
3139 			continue;
3140 
3141 		ds = btrfsic_dev_state_hashtable_lookup(
3142 				device->bdev,
3143 				&btrfsic_dev_state_hashtable);
3144 		if (NULL != ds) {
3145 			state = ds->state;
3146 			btrfsic_dev_state_hashtable_remove(ds);
3147 			btrfsic_dev_state_free(ds);
3148 		}
3149 	}
3150 
3151 	if (NULL == state) {
3152 		printk(KERN_INFO
3153 		       "btrfsic: error, cannot find state information"
3154 		       " on umount!\n");
3155 		mutex_unlock(&btrfsic_mutex);
3156 		return;
3157 	}
3158 
3159 	/*
3160 	 * Don't care about keeping the lists' state up to date,
3161 	 * just free all memory that was allocated dynamically.
3162 	 * Free the blocks and the block_links.
3163 	 */
3164 	list_for_each_entry_safe(b_all, tmp_all, &state->all_blocks_list,
3165 				 all_blocks_node) {
3166 		struct btrfsic_block_link *l, *tmp;
3167 
3168 		list_for_each_entry_safe(l, tmp, &b_all->ref_to_list,
3169 					 node_ref_to) {
3170 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
3171 				btrfsic_print_rem_link(state, l);
3172 
3173 			l->ref_cnt--;
3174 			if (0 == l->ref_cnt)
3175 				btrfsic_block_link_free(l);
3176 		}
3177 
3178 		if (b_all->is_iodone || b_all->never_written)
3179 			btrfsic_block_free(b_all);
3180 		else
3181 			printk(KERN_INFO "btrfs: attempt to free %c-block"
3182 			       " @%llu (%s/%llu/%d) on umount which is"
3183 			       " not yet iodone!\n",
3184 			       btrfsic_get_block_type(state, b_all),
3185 			       b_all->logical_bytenr, b_all->dev_state->name,
3186 			       b_all->dev_bytenr, b_all->mirror_num);
3187 	}
3188 
3189 	mutex_unlock(&btrfsic_mutex);
3190 
3191 	kvfree(state);
3192 }
3193