xref: /openbmc/linux/fs/btrfs/check-integrity.c (revision 78700c0a)
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 <linux/string.h>
99 #include "ctree.h"
100 #include "disk-io.h"
101 #include "hash.h"
102 #include "transaction.h"
103 #include "extent_io.h"
104 #include "volumes.h"
105 #include "print-tree.h"
106 #include "locking.h"
107 #include "check-integrity.h"
108 #include "rcu-string.h"
109 #include "compression.h"
110 
111 #define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
112 #define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
113 #define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
114 #define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
115 #define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
116 #define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
117 #define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
118 #define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6)	/* in characters,
119 							 * excluding " [...]" */
120 #define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
121 
122 /*
123  * The definition of the bitmask fields for the print_mask.
124  * They are specified with the mount option check_integrity_print_mask.
125  */
126 #define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE			0x00000001
127 #define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION		0x00000002
128 #define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE			0x00000004
129 #define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE			0x00000008
130 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH			0x00000010
131 #define BTRFSIC_PRINT_MASK_END_IO_BIO_BH			0x00000020
132 #define BTRFSIC_PRINT_MASK_VERBOSE				0x00000040
133 #define BTRFSIC_PRINT_MASK_VERY_VERBOSE				0x00000080
134 #define BTRFSIC_PRINT_MASK_INITIAL_TREE				0x00000100
135 #define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES			0x00000200
136 #define BTRFSIC_PRINT_MASK_INITIAL_DATABASE			0x00000400
137 #define BTRFSIC_PRINT_MASK_NUM_COPIES				0x00000800
138 #define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS		0x00001000
139 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE		0x00002000
140 
141 struct btrfsic_dev_state;
142 struct btrfsic_state;
143 
144 struct btrfsic_block {
145 	u32 magic_num;		/* only used for debug purposes */
146 	unsigned int is_metadata:1;	/* if it is meta-data, not data-data */
147 	unsigned int is_superblock:1;	/* if it is one of the superblocks */
148 	unsigned int is_iodone:1;	/* if is done by lower subsystem */
149 	unsigned int iodone_w_error:1;	/* error was indicated to endio */
150 	unsigned int never_written:1;	/* block was added because it was
151 					 * referenced, not because it was
152 					 * written */
153 	unsigned int mirror_num;	/* large enough to hold
154 					 * BTRFS_SUPER_MIRROR_MAX */
155 	struct btrfsic_dev_state *dev_state;
156 	u64 dev_bytenr;		/* key, physical byte num on disk */
157 	u64 logical_bytenr;	/* logical byte num on disk */
158 	u64 generation;
159 	struct btrfs_disk_key disk_key;	/* extra info to print in case of
160 					 * issues, will not always be correct */
161 	struct list_head collision_resolving_node;	/* list node */
162 	struct list_head all_blocks_node;	/* list node */
163 
164 	/* the following two lists contain block_link items */
165 	struct list_head ref_to_list;	/* list */
166 	struct list_head ref_from_list;	/* list */
167 	struct btrfsic_block *next_in_same_bio;
168 	void *orig_bio_bh_private;
169 	union {
170 		bio_end_io_t *bio;
171 		bh_end_io_t *bh;
172 	} orig_bio_bh_end_io;
173 	int submit_bio_bh_rw;
174 	u64 flush_gen; /* only valid if !never_written */
175 };
176 
177 /*
178  * Elements of this type are allocated dynamically and required because
179  * each block object can refer to and can be ref from multiple blocks.
180  * The key to lookup them in the hashtable is the dev_bytenr of
181  * the block ref to plus the one from the block referred from.
182  * The fact that they are searchable via a hashtable and that a
183  * ref_cnt is maintained is not required for the btrfs integrity
184  * check algorithm itself, it is only used to make the output more
185  * beautiful in case that an error is detected (an error is defined
186  * as a write operation to a block while that block is still referenced).
187  */
188 struct btrfsic_block_link {
189 	u32 magic_num;		/* only used for debug purposes */
190 	u32 ref_cnt;
191 	struct list_head node_ref_to;	/* list node */
192 	struct list_head node_ref_from;	/* list node */
193 	struct list_head collision_resolving_node;	/* list node */
194 	struct btrfsic_block *block_ref_to;
195 	struct btrfsic_block *block_ref_from;
196 	u64 parent_generation;
197 };
198 
199 struct btrfsic_dev_state {
200 	u32 magic_num;		/* only used for debug purposes */
201 	struct block_device *bdev;
202 	struct btrfsic_state *state;
203 	struct list_head collision_resolving_node;	/* list node */
204 	struct btrfsic_block dummy_block_for_bio_bh_flush;
205 	u64 last_flush_gen;
206 	char name[BDEVNAME_SIZE];
207 };
208 
209 struct btrfsic_block_hashtable {
210 	struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
211 };
212 
213 struct btrfsic_block_link_hashtable {
214 	struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
215 };
216 
217 struct btrfsic_dev_state_hashtable {
218 	struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
219 };
220 
221 struct btrfsic_block_data_ctx {
222 	u64 start;		/* virtual bytenr */
223 	u64 dev_bytenr;		/* physical bytenr on device */
224 	u32 len;
225 	struct btrfsic_dev_state *dev;
226 	char **datav;
227 	struct page **pagev;
228 	void *mem_to_free;
229 };
230 
231 /* This structure is used to implement recursion without occupying
232  * any stack space, refer to btrfsic_process_metablock() */
233 struct btrfsic_stack_frame {
234 	u32 magic;
235 	u32 nr;
236 	int error;
237 	int i;
238 	int limit_nesting;
239 	int num_copies;
240 	int mirror_num;
241 	struct btrfsic_block *block;
242 	struct btrfsic_block_data_ctx *block_ctx;
243 	struct btrfsic_block *next_block;
244 	struct btrfsic_block_data_ctx next_block_ctx;
245 	struct btrfs_header *hdr;
246 	struct btrfsic_stack_frame *prev;
247 };
248 
249 /* Some state per mounted filesystem */
250 struct btrfsic_state {
251 	u32 print_mask;
252 	int include_extent_data;
253 	int csum_size;
254 	struct list_head all_blocks_list;
255 	struct btrfsic_block_hashtable block_hashtable;
256 	struct btrfsic_block_link_hashtable block_link_hashtable;
257 	struct btrfs_root *root;
258 	u64 max_superblock_generation;
259 	struct btrfsic_block *latest_superblock;
260 	u32 metablock_size;
261 	u32 datablock_size;
262 };
263 
264 static void btrfsic_block_init(struct btrfsic_block *b);
265 static struct btrfsic_block *btrfsic_block_alloc(void);
266 static void btrfsic_block_free(struct btrfsic_block *b);
267 static void btrfsic_block_link_init(struct btrfsic_block_link *n);
268 static struct btrfsic_block_link *btrfsic_block_link_alloc(void);
269 static void btrfsic_block_link_free(struct btrfsic_block_link *n);
270 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds);
271 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void);
272 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds);
273 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h);
274 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
275 					struct btrfsic_block_hashtable *h);
276 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b);
277 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
278 		struct block_device *bdev,
279 		u64 dev_bytenr,
280 		struct btrfsic_block_hashtable *h);
281 static void btrfsic_block_link_hashtable_init(
282 		struct btrfsic_block_link_hashtable *h);
283 static void btrfsic_block_link_hashtable_add(
284 		struct btrfsic_block_link *l,
285 		struct btrfsic_block_link_hashtable *h);
286 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l);
287 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
288 		struct block_device *bdev_ref_to,
289 		u64 dev_bytenr_ref_to,
290 		struct block_device *bdev_ref_from,
291 		u64 dev_bytenr_ref_from,
292 		struct btrfsic_block_link_hashtable *h);
293 static void btrfsic_dev_state_hashtable_init(
294 		struct btrfsic_dev_state_hashtable *h);
295 static void btrfsic_dev_state_hashtable_add(
296 		struct btrfsic_dev_state *ds,
297 		struct btrfsic_dev_state_hashtable *h);
298 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds);
299 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(
300 		struct block_device *bdev,
301 		struct btrfsic_dev_state_hashtable *h);
302 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void);
303 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf);
304 static int btrfsic_process_superblock(struct btrfsic_state *state,
305 				      struct btrfs_fs_devices *fs_devices);
306 static int btrfsic_process_metablock(struct btrfsic_state *state,
307 				     struct btrfsic_block *block,
308 				     struct btrfsic_block_data_ctx *block_ctx,
309 				     int limit_nesting, int force_iodone_flag);
310 static void btrfsic_read_from_block_data(
311 	struct btrfsic_block_data_ctx *block_ctx,
312 	void *dst, u32 offset, size_t len);
313 static int btrfsic_create_link_to_next_block(
314 		struct btrfsic_state *state,
315 		struct btrfsic_block *block,
316 		struct btrfsic_block_data_ctx
317 		*block_ctx, u64 next_bytenr,
318 		int limit_nesting,
319 		struct btrfsic_block_data_ctx *next_block_ctx,
320 		struct btrfsic_block **next_blockp,
321 		int force_iodone_flag,
322 		int *num_copiesp, int *mirror_nump,
323 		struct btrfs_disk_key *disk_key,
324 		u64 parent_generation);
325 static int btrfsic_handle_extent_data(struct btrfsic_state *state,
326 				      struct btrfsic_block *block,
327 				      struct btrfsic_block_data_ctx *block_ctx,
328 				      u32 item_offset, int force_iodone_flag);
329 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
330 			     struct btrfsic_block_data_ctx *block_ctx_out,
331 			     int mirror_num);
332 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
333 static int btrfsic_read_block(struct btrfsic_state *state,
334 			      struct btrfsic_block_data_ctx *block_ctx);
335 static void btrfsic_dump_database(struct btrfsic_state *state);
336 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
337 				     char **datav, unsigned int num_pages);
338 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
339 					  u64 dev_bytenr, char **mapped_datav,
340 					  unsigned int num_pages,
341 					  struct bio *bio, int *bio_is_patched,
342 					  struct buffer_head *bh,
343 					  int submit_bio_bh_rw);
344 static int btrfsic_process_written_superblock(
345 		struct btrfsic_state *state,
346 		struct btrfsic_block *const block,
347 		struct btrfs_super_block *const super_hdr);
348 static void btrfsic_bio_end_io(struct bio *bp);
349 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate);
350 static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
351 					      const struct btrfsic_block *block,
352 					      int recursion_level);
353 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
354 					struct btrfsic_block *const block,
355 					int recursion_level);
356 static void btrfsic_print_add_link(const struct btrfsic_state *state,
357 				   const struct btrfsic_block_link *l);
358 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
359 				   const struct btrfsic_block_link *l);
360 static char btrfsic_get_block_type(const struct btrfsic_state *state,
361 				   const struct btrfsic_block *block);
362 static void btrfsic_dump_tree(const struct btrfsic_state *state);
363 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
364 				  const struct btrfsic_block *block,
365 				  int indent_level);
366 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
367 		struct btrfsic_state *state,
368 		struct btrfsic_block_data_ctx *next_block_ctx,
369 		struct btrfsic_block *next_block,
370 		struct btrfsic_block *from_block,
371 		u64 parent_generation);
372 static struct btrfsic_block *btrfsic_block_lookup_or_add(
373 		struct btrfsic_state *state,
374 		struct btrfsic_block_data_ctx *block_ctx,
375 		const char *additional_string,
376 		int is_metadata,
377 		int is_iodone,
378 		int never_written,
379 		int mirror_num,
380 		int *was_created);
381 static int btrfsic_process_superblock_dev_mirror(
382 		struct btrfsic_state *state,
383 		struct btrfsic_dev_state *dev_state,
384 		struct btrfs_device *device,
385 		int superblock_mirror_num,
386 		struct btrfsic_dev_state **selected_dev_state,
387 		struct btrfs_super_block *selected_super);
388 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
389 		struct block_device *bdev);
390 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
391 					   u64 bytenr,
392 					   struct btrfsic_dev_state *dev_state,
393 					   u64 dev_bytenr);
394 
395 static struct mutex btrfsic_mutex;
396 static int btrfsic_is_initialized;
397 static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
398 
399 
400 static void btrfsic_block_init(struct btrfsic_block *b)
401 {
402 	b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
403 	b->dev_state = NULL;
404 	b->dev_bytenr = 0;
405 	b->logical_bytenr = 0;
406 	b->generation = BTRFSIC_GENERATION_UNKNOWN;
407 	b->disk_key.objectid = 0;
408 	b->disk_key.type = 0;
409 	b->disk_key.offset = 0;
410 	b->is_metadata = 0;
411 	b->is_superblock = 0;
412 	b->is_iodone = 0;
413 	b->iodone_w_error = 0;
414 	b->never_written = 0;
415 	b->mirror_num = 0;
416 	b->next_in_same_bio = NULL;
417 	b->orig_bio_bh_private = NULL;
418 	b->orig_bio_bh_end_io.bio = NULL;
419 	INIT_LIST_HEAD(&b->collision_resolving_node);
420 	INIT_LIST_HEAD(&b->all_blocks_node);
421 	INIT_LIST_HEAD(&b->ref_to_list);
422 	INIT_LIST_HEAD(&b->ref_from_list);
423 	b->submit_bio_bh_rw = 0;
424 	b->flush_gen = 0;
425 }
426 
427 static struct btrfsic_block *btrfsic_block_alloc(void)
428 {
429 	struct btrfsic_block *b;
430 
431 	b = kzalloc(sizeof(*b), GFP_NOFS);
432 	if (NULL != b)
433 		btrfsic_block_init(b);
434 
435 	return b;
436 }
437 
438 static void btrfsic_block_free(struct btrfsic_block *b)
439 {
440 	BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
441 	kfree(b);
442 }
443 
444 static void btrfsic_block_link_init(struct btrfsic_block_link *l)
445 {
446 	l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
447 	l->ref_cnt = 1;
448 	INIT_LIST_HEAD(&l->node_ref_to);
449 	INIT_LIST_HEAD(&l->node_ref_from);
450 	INIT_LIST_HEAD(&l->collision_resolving_node);
451 	l->block_ref_to = NULL;
452 	l->block_ref_from = NULL;
453 }
454 
455 static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
456 {
457 	struct btrfsic_block_link *l;
458 
459 	l = kzalloc(sizeof(*l), GFP_NOFS);
460 	if (NULL != l)
461 		btrfsic_block_link_init(l);
462 
463 	return l;
464 }
465 
466 static void btrfsic_block_link_free(struct btrfsic_block_link *l)
467 {
468 	BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
469 	kfree(l);
470 }
471 
472 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
473 {
474 	ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
475 	ds->bdev = NULL;
476 	ds->state = NULL;
477 	ds->name[0] = '\0';
478 	INIT_LIST_HEAD(&ds->collision_resolving_node);
479 	ds->last_flush_gen = 0;
480 	btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
481 	ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
482 	ds->dummy_block_for_bio_bh_flush.dev_state = ds;
483 }
484 
485 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
486 {
487 	struct btrfsic_dev_state *ds;
488 
489 	ds = kzalloc(sizeof(*ds), GFP_NOFS);
490 	if (NULL != ds)
491 		btrfsic_dev_state_init(ds);
492 
493 	return ds;
494 }
495 
496 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
497 {
498 	BUG_ON(!(NULL == ds ||
499 		 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
500 	kfree(ds);
501 }
502 
503 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
504 {
505 	int i;
506 
507 	for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
508 		INIT_LIST_HEAD(h->table + i);
509 }
510 
511 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
512 					struct btrfsic_block_hashtable *h)
513 {
514 	const unsigned int hashval =
515 	    (((unsigned int)(b->dev_bytenr >> 16)) ^
516 	     ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
517 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
518 
519 	list_add(&b->collision_resolving_node, h->table + hashval);
520 }
521 
522 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
523 {
524 	list_del(&b->collision_resolving_node);
525 }
526 
527 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
528 		struct block_device *bdev,
529 		u64 dev_bytenr,
530 		struct btrfsic_block_hashtable *h)
531 {
532 	const unsigned int hashval =
533 	    (((unsigned int)(dev_bytenr >> 16)) ^
534 	     ((unsigned int)((uintptr_t)bdev))) &
535 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
536 	struct btrfsic_block *b;
537 
538 	list_for_each_entry(b, h->table + hashval, collision_resolving_node) {
539 		if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
540 			return b;
541 	}
542 
543 	return NULL;
544 }
545 
546 static void btrfsic_block_link_hashtable_init(
547 		struct btrfsic_block_link_hashtable *h)
548 {
549 	int i;
550 
551 	for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
552 		INIT_LIST_HEAD(h->table + i);
553 }
554 
555 static void btrfsic_block_link_hashtable_add(
556 		struct btrfsic_block_link *l,
557 		struct btrfsic_block_link_hashtable *h)
558 {
559 	const unsigned int hashval =
560 	    (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
561 	     ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
562 	     ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
563 	     ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
564 	     & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
565 
566 	BUG_ON(NULL == l->block_ref_to);
567 	BUG_ON(NULL == l->block_ref_from);
568 	list_add(&l->collision_resolving_node, h->table + hashval);
569 }
570 
571 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
572 {
573 	list_del(&l->collision_resolving_node);
574 }
575 
576 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
577 		struct block_device *bdev_ref_to,
578 		u64 dev_bytenr_ref_to,
579 		struct block_device *bdev_ref_from,
580 		u64 dev_bytenr_ref_from,
581 		struct btrfsic_block_link_hashtable *h)
582 {
583 	const unsigned int hashval =
584 	    (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
585 	     ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
586 	     ((unsigned int)((uintptr_t)bdev_ref_to)) ^
587 	     ((unsigned int)((uintptr_t)bdev_ref_from))) &
588 	     (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
589 	struct btrfsic_block_link *l;
590 
591 	list_for_each_entry(l, h->table + hashval, collision_resolving_node) {
592 		BUG_ON(NULL == l->block_ref_to);
593 		BUG_ON(NULL == l->block_ref_from);
594 		if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
595 		    l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
596 		    l->block_ref_from->dev_state->bdev == bdev_ref_from &&
597 		    l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
598 			return l;
599 	}
600 
601 	return NULL;
602 }
603 
604 static void btrfsic_dev_state_hashtable_init(
605 		struct btrfsic_dev_state_hashtable *h)
606 {
607 	int i;
608 
609 	for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
610 		INIT_LIST_HEAD(h->table + i);
611 }
612 
613 static void btrfsic_dev_state_hashtable_add(
614 		struct btrfsic_dev_state *ds,
615 		struct btrfsic_dev_state_hashtable *h)
616 {
617 	const unsigned int hashval =
618 	    (((unsigned int)((uintptr_t)ds->bdev)) &
619 	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
620 
621 	list_add(&ds->collision_resolving_node, h->table + hashval);
622 }
623 
624 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
625 {
626 	list_del(&ds->collision_resolving_node);
627 }
628 
629 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(
630 		struct block_device *bdev,
631 		struct btrfsic_dev_state_hashtable *h)
632 {
633 	const unsigned int hashval =
634 	    (((unsigned int)((uintptr_t)bdev)) &
635 	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
636 	struct btrfsic_dev_state *ds;
637 
638 	list_for_each_entry(ds, h->table + hashval, collision_resolving_node) {
639 		if (ds->bdev == bdev)
640 			return ds;
641 	}
642 
643 	return NULL;
644 }
645 
646 static int btrfsic_process_superblock(struct btrfsic_state *state,
647 				      struct btrfs_fs_devices *fs_devices)
648 {
649 	int ret = 0;
650 	struct btrfs_super_block *selected_super;
651 	struct list_head *dev_head = &fs_devices->devices;
652 	struct btrfs_device *device;
653 	struct btrfsic_dev_state *selected_dev_state = NULL;
654 	int pass;
655 
656 	BUG_ON(NULL == state);
657 	selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
658 	if (NULL == selected_super) {
659 		printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
660 		return -ENOMEM;
661 	}
662 
663 	list_for_each_entry(device, dev_head, dev_list) {
664 		int i;
665 		struct btrfsic_dev_state *dev_state;
666 
667 		if (!device->bdev || !device->name)
668 			continue;
669 
670 		dev_state = btrfsic_dev_state_lookup(device->bdev);
671 		BUG_ON(NULL == dev_state);
672 		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
673 			ret = btrfsic_process_superblock_dev_mirror(
674 					state, dev_state, device, i,
675 					&selected_dev_state, selected_super);
676 			if (0 != ret && 0 == i) {
677 				kfree(selected_super);
678 				return ret;
679 			}
680 		}
681 	}
682 
683 	if (NULL == state->latest_superblock) {
684 		printk(KERN_INFO "btrfsic: no superblock found!\n");
685 		kfree(selected_super);
686 		return -1;
687 	}
688 
689 	state->csum_size = btrfs_super_csum_size(selected_super);
690 
691 	for (pass = 0; pass < 3; pass++) {
692 		int num_copies;
693 		int mirror_num;
694 		u64 next_bytenr;
695 
696 		switch (pass) {
697 		case 0:
698 			next_bytenr = btrfs_super_root(selected_super);
699 			if (state->print_mask &
700 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
701 				printk(KERN_INFO "root@%llu\n", next_bytenr);
702 			break;
703 		case 1:
704 			next_bytenr = btrfs_super_chunk_root(selected_super);
705 			if (state->print_mask &
706 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
707 				printk(KERN_INFO "chunk@%llu\n", next_bytenr);
708 			break;
709 		case 2:
710 			next_bytenr = btrfs_super_log_root(selected_super);
711 			if (0 == next_bytenr)
712 				continue;
713 			if (state->print_mask &
714 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
715 				printk(KERN_INFO "log@%llu\n", next_bytenr);
716 			break;
717 		}
718 
719 		num_copies =
720 		    btrfs_num_copies(state->root->fs_info,
721 				     next_bytenr, state->metablock_size);
722 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
723 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
724 			       next_bytenr, num_copies);
725 
726 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
727 			struct btrfsic_block *next_block;
728 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
729 			struct btrfsic_block_link *l;
730 
731 			ret = btrfsic_map_block(state, next_bytenr,
732 						state->metablock_size,
733 						&tmp_next_block_ctx,
734 						mirror_num);
735 			if (ret) {
736 				printk(KERN_INFO "btrfsic:"
737 				       " btrfsic_map_block(root @%llu,"
738 				       " mirror %d) failed!\n",
739 				       next_bytenr, mirror_num);
740 				kfree(selected_super);
741 				return -1;
742 			}
743 
744 			next_block = btrfsic_block_hashtable_lookup(
745 					tmp_next_block_ctx.dev->bdev,
746 					tmp_next_block_ctx.dev_bytenr,
747 					&state->block_hashtable);
748 			BUG_ON(NULL == next_block);
749 
750 			l = btrfsic_block_link_hashtable_lookup(
751 					tmp_next_block_ctx.dev->bdev,
752 					tmp_next_block_ctx.dev_bytenr,
753 					state->latest_superblock->dev_state->
754 					bdev,
755 					state->latest_superblock->dev_bytenr,
756 					&state->block_link_hashtable);
757 			BUG_ON(NULL == l);
758 
759 			ret = btrfsic_read_block(state, &tmp_next_block_ctx);
760 			if (ret < (int)PAGE_SIZE) {
761 				printk(KERN_INFO
762 				       "btrfsic: read @logical %llu failed!\n",
763 				       tmp_next_block_ctx.start);
764 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
765 				kfree(selected_super);
766 				return -1;
767 			}
768 
769 			ret = btrfsic_process_metablock(state,
770 							next_block,
771 							&tmp_next_block_ctx,
772 							BTRFS_MAX_LEVEL + 3, 1);
773 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
774 		}
775 	}
776 
777 	kfree(selected_super);
778 	return ret;
779 }
780 
781 static int btrfsic_process_superblock_dev_mirror(
782 		struct btrfsic_state *state,
783 		struct btrfsic_dev_state *dev_state,
784 		struct btrfs_device *device,
785 		int superblock_mirror_num,
786 		struct btrfsic_dev_state **selected_dev_state,
787 		struct btrfs_super_block *selected_super)
788 {
789 	struct btrfs_super_block *super_tmp;
790 	u64 dev_bytenr;
791 	struct buffer_head *bh;
792 	struct btrfsic_block *superblock_tmp;
793 	int pass;
794 	struct block_device *const superblock_bdev = device->bdev;
795 
796 	/* super block bytenr is always the unmapped device bytenr */
797 	dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
798 	if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->commit_total_bytes)
799 		return -1;
800 	bh = __bread(superblock_bdev, dev_bytenr / 4096,
801 		     BTRFS_SUPER_INFO_SIZE);
802 	if (NULL == bh)
803 		return -1;
804 	super_tmp = (struct btrfs_super_block *)
805 	    (bh->b_data + (dev_bytenr & 4095));
806 
807 	if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
808 	    btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
809 	    memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
810 	    btrfs_super_nodesize(super_tmp) != state->metablock_size ||
811 	    btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
812 		brelse(bh);
813 		return 0;
814 	}
815 
816 	superblock_tmp =
817 	    btrfsic_block_hashtable_lookup(superblock_bdev,
818 					   dev_bytenr,
819 					   &state->block_hashtable);
820 	if (NULL == superblock_tmp) {
821 		superblock_tmp = btrfsic_block_alloc();
822 		if (NULL == superblock_tmp) {
823 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
824 			brelse(bh);
825 			return -1;
826 		}
827 		/* for superblock, only the dev_bytenr makes sense */
828 		superblock_tmp->dev_bytenr = dev_bytenr;
829 		superblock_tmp->dev_state = dev_state;
830 		superblock_tmp->logical_bytenr = dev_bytenr;
831 		superblock_tmp->generation = btrfs_super_generation(super_tmp);
832 		superblock_tmp->is_metadata = 1;
833 		superblock_tmp->is_superblock = 1;
834 		superblock_tmp->is_iodone = 1;
835 		superblock_tmp->never_written = 0;
836 		superblock_tmp->mirror_num = 1 + superblock_mirror_num;
837 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
838 			btrfs_info_in_rcu(device->dev_root->fs_info,
839 				"new initial S-block (bdev %p, %s) @%llu (%s/%llu/%d)",
840 				     superblock_bdev,
841 				     rcu_str_deref(device->name), dev_bytenr,
842 				     dev_state->name, dev_bytenr,
843 				     superblock_mirror_num);
844 		list_add(&superblock_tmp->all_blocks_node,
845 			 &state->all_blocks_list);
846 		btrfsic_block_hashtable_add(superblock_tmp,
847 					    &state->block_hashtable);
848 	}
849 
850 	/* select the one with the highest generation field */
851 	if (btrfs_super_generation(super_tmp) >
852 	    state->max_superblock_generation ||
853 	    0 == state->max_superblock_generation) {
854 		memcpy(selected_super, super_tmp, sizeof(*selected_super));
855 		*selected_dev_state = dev_state;
856 		state->max_superblock_generation =
857 		    btrfs_super_generation(super_tmp);
858 		state->latest_superblock = superblock_tmp;
859 	}
860 
861 	for (pass = 0; pass < 3; pass++) {
862 		u64 next_bytenr;
863 		int num_copies;
864 		int mirror_num;
865 		const char *additional_string = NULL;
866 		struct btrfs_disk_key tmp_disk_key;
867 
868 		tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
869 		tmp_disk_key.offset = 0;
870 		switch (pass) {
871 		case 0:
872 			btrfs_set_disk_key_objectid(&tmp_disk_key,
873 						    BTRFS_ROOT_TREE_OBJECTID);
874 			additional_string = "initial root ";
875 			next_bytenr = btrfs_super_root(super_tmp);
876 			break;
877 		case 1:
878 			btrfs_set_disk_key_objectid(&tmp_disk_key,
879 						    BTRFS_CHUNK_TREE_OBJECTID);
880 			additional_string = "initial chunk ";
881 			next_bytenr = btrfs_super_chunk_root(super_tmp);
882 			break;
883 		case 2:
884 			btrfs_set_disk_key_objectid(&tmp_disk_key,
885 						    BTRFS_TREE_LOG_OBJECTID);
886 			additional_string = "initial log ";
887 			next_bytenr = btrfs_super_log_root(super_tmp);
888 			if (0 == next_bytenr)
889 				continue;
890 			break;
891 		}
892 
893 		num_copies =
894 		    btrfs_num_copies(state->root->fs_info,
895 				     next_bytenr, state->metablock_size);
896 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
897 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
898 			       next_bytenr, num_copies);
899 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
900 			struct btrfsic_block *next_block;
901 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
902 			struct btrfsic_block_link *l;
903 
904 			if (btrfsic_map_block(state, next_bytenr,
905 					      state->metablock_size,
906 					      &tmp_next_block_ctx,
907 					      mirror_num)) {
908 				printk(KERN_INFO "btrfsic: btrfsic_map_block("
909 				       "bytenr @%llu, mirror %d) failed!\n",
910 				       next_bytenr, mirror_num);
911 				brelse(bh);
912 				return -1;
913 			}
914 
915 			next_block = btrfsic_block_lookup_or_add(
916 					state, &tmp_next_block_ctx,
917 					additional_string, 1, 1, 0,
918 					mirror_num, NULL);
919 			if (NULL == next_block) {
920 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
921 				brelse(bh);
922 				return -1;
923 			}
924 
925 			next_block->disk_key = tmp_disk_key;
926 			next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
927 			l = btrfsic_block_link_lookup_or_add(
928 					state, &tmp_next_block_ctx,
929 					next_block, superblock_tmp,
930 					BTRFSIC_GENERATION_UNKNOWN);
931 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
932 			if (NULL == l) {
933 				brelse(bh);
934 				return -1;
935 			}
936 		}
937 	}
938 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
939 		btrfsic_dump_tree_sub(state, superblock_tmp, 0);
940 
941 	brelse(bh);
942 	return 0;
943 }
944 
945 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
946 {
947 	struct btrfsic_stack_frame *sf;
948 
949 	sf = kzalloc(sizeof(*sf), GFP_NOFS);
950 	if (NULL == sf)
951 		printk(KERN_INFO "btrfsic: alloc memory failed!\n");
952 	else
953 		sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
954 	return sf;
955 }
956 
957 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
958 {
959 	BUG_ON(!(NULL == sf ||
960 		 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
961 	kfree(sf);
962 }
963 
964 static int btrfsic_process_metablock(
965 		struct btrfsic_state *state,
966 		struct btrfsic_block *const first_block,
967 		struct btrfsic_block_data_ctx *const first_block_ctx,
968 		int first_limit_nesting, int force_iodone_flag)
969 {
970 	struct btrfsic_stack_frame initial_stack_frame = { 0 };
971 	struct btrfsic_stack_frame *sf;
972 	struct btrfsic_stack_frame *next_stack;
973 	struct btrfs_header *const first_hdr =
974 		(struct btrfs_header *)first_block_ctx->datav[0];
975 
976 	BUG_ON(!first_hdr);
977 	sf = &initial_stack_frame;
978 	sf->error = 0;
979 	sf->i = -1;
980 	sf->limit_nesting = first_limit_nesting;
981 	sf->block = first_block;
982 	sf->block_ctx = first_block_ctx;
983 	sf->next_block = NULL;
984 	sf->hdr = first_hdr;
985 	sf->prev = NULL;
986 
987 continue_with_new_stack_frame:
988 	sf->block->generation = le64_to_cpu(sf->hdr->generation);
989 	if (0 == sf->hdr->level) {
990 		struct btrfs_leaf *const leafhdr =
991 		    (struct btrfs_leaf *)sf->hdr;
992 
993 		if (-1 == sf->i) {
994 			sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
995 
996 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
997 				printk(KERN_INFO
998 				       "leaf %llu items %d generation %llu"
999 				       " owner %llu\n",
1000 				       sf->block_ctx->start, sf->nr,
1001 				       btrfs_stack_header_generation(
1002 					       &leafhdr->header),
1003 				       btrfs_stack_header_owner(
1004 					       &leafhdr->header));
1005 		}
1006 
1007 continue_with_current_leaf_stack_frame:
1008 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1009 			sf->i++;
1010 			sf->num_copies = 0;
1011 		}
1012 
1013 		if (sf->i < sf->nr) {
1014 			struct btrfs_item disk_item;
1015 			u32 disk_item_offset =
1016 				(uintptr_t)(leafhdr->items + sf->i) -
1017 				(uintptr_t)leafhdr;
1018 			struct btrfs_disk_key *disk_key;
1019 			u8 type;
1020 			u32 item_offset;
1021 			u32 item_size;
1022 
1023 			if (disk_item_offset + sizeof(struct btrfs_item) >
1024 			    sf->block_ctx->len) {
1025 leaf_item_out_of_bounce_error:
1026 				printk(KERN_INFO
1027 				       "btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
1028 				       sf->block_ctx->start,
1029 				       sf->block_ctx->dev->name);
1030 				goto one_stack_frame_backwards;
1031 			}
1032 			btrfsic_read_from_block_data(sf->block_ctx,
1033 						     &disk_item,
1034 						     disk_item_offset,
1035 						     sizeof(struct btrfs_item));
1036 			item_offset = btrfs_stack_item_offset(&disk_item);
1037 			item_size = btrfs_stack_item_size(&disk_item);
1038 			disk_key = &disk_item.key;
1039 			type = btrfs_disk_key_type(disk_key);
1040 
1041 			if (BTRFS_ROOT_ITEM_KEY == type) {
1042 				struct btrfs_root_item root_item;
1043 				u32 root_item_offset;
1044 				u64 next_bytenr;
1045 
1046 				root_item_offset = item_offset +
1047 					offsetof(struct btrfs_leaf, items);
1048 				if (root_item_offset + item_size >
1049 				    sf->block_ctx->len)
1050 					goto leaf_item_out_of_bounce_error;
1051 				btrfsic_read_from_block_data(
1052 					sf->block_ctx, &root_item,
1053 					root_item_offset,
1054 					item_size);
1055 				next_bytenr = btrfs_root_bytenr(&root_item);
1056 
1057 				sf->error =
1058 				    btrfsic_create_link_to_next_block(
1059 						state,
1060 						sf->block,
1061 						sf->block_ctx,
1062 						next_bytenr,
1063 						sf->limit_nesting,
1064 						&sf->next_block_ctx,
1065 						&sf->next_block,
1066 						force_iodone_flag,
1067 						&sf->num_copies,
1068 						&sf->mirror_num,
1069 						disk_key,
1070 						btrfs_root_generation(
1071 						&root_item));
1072 				if (sf->error)
1073 					goto one_stack_frame_backwards;
1074 
1075 				if (NULL != sf->next_block) {
1076 					struct btrfs_header *const next_hdr =
1077 					    (struct btrfs_header *)
1078 					    sf->next_block_ctx.datav[0];
1079 
1080 					next_stack =
1081 					    btrfsic_stack_frame_alloc();
1082 					if (NULL == next_stack) {
1083 						sf->error = -1;
1084 						btrfsic_release_block_ctx(
1085 								&sf->
1086 								next_block_ctx);
1087 						goto one_stack_frame_backwards;
1088 					}
1089 
1090 					next_stack->i = -1;
1091 					next_stack->block = sf->next_block;
1092 					next_stack->block_ctx =
1093 					    &sf->next_block_ctx;
1094 					next_stack->next_block = NULL;
1095 					next_stack->hdr = next_hdr;
1096 					next_stack->limit_nesting =
1097 					    sf->limit_nesting - 1;
1098 					next_stack->prev = sf;
1099 					sf = next_stack;
1100 					goto continue_with_new_stack_frame;
1101 				}
1102 			} else if (BTRFS_EXTENT_DATA_KEY == type &&
1103 				   state->include_extent_data) {
1104 				sf->error = btrfsic_handle_extent_data(
1105 						state,
1106 						sf->block,
1107 						sf->block_ctx,
1108 						item_offset,
1109 						force_iodone_flag);
1110 				if (sf->error)
1111 					goto one_stack_frame_backwards;
1112 			}
1113 
1114 			goto continue_with_current_leaf_stack_frame;
1115 		}
1116 	} else {
1117 		struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1118 
1119 		if (-1 == sf->i) {
1120 			sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1121 
1122 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1123 				printk(KERN_INFO "node %llu level %d items %d"
1124 				       " generation %llu owner %llu\n",
1125 				       sf->block_ctx->start,
1126 				       nodehdr->header.level, sf->nr,
1127 				       btrfs_stack_header_generation(
1128 				       &nodehdr->header),
1129 				       btrfs_stack_header_owner(
1130 				       &nodehdr->header));
1131 		}
1132 
1133 continue_with_current_node_stack_frame:
1134 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1135 			sf->i++;
1136 			sf->num_copies = 0;
1137 		}
1138 
1139 		if (sf->i < sf->nr) {
1140 			struct btrfs_key_ptr key_ptr;
1141 			u32 key_ptr_offset;
1142 			u64 next_bytenr;
1143 
1144 			key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1145 					  (uintptr_t)nodehdr;
1146 			if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1147 			    sf->block_ctx->len) {
1148 				printk(KERN_INFO
1149 				       "btrfsic: node item out of bounce at logical %llu, dev %s\n",
1150 				       sf->block_ctx->start,
1151 				       sf->block_ctx->dev->name);
1152 				goto one_stack_frame_backwards;
1153 			}
1154 			btrfsic_read_from_block_data(
1155 				sf->block_ctx, &key_ptr, key_ptr_offset,
1156 				sizeof(struct btrfs_key_ptr));
1157 			next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1158 
1159 			sf->error = btrfsic_create_link_to_next_block(
1160 					state,
1161 					sf->block,
1162 					sf->block_ctx,
1163 					next_bytenr,
1164 					sf->limit_nesting,
1165 					&sf->next_block_ctx,
1166 					&sf->next_block,
1167 					force_iodone_flag,
1168 					&sf->num_copies,
1169 					&sf->mirror_num,
1170 					&key_ptr.key,
1171 					btrfs_stack_key_generation(&key_ptr));
1172 			if (sf->error)
1173 				goto one_stack_frame_backwards;
1174 
1175 			if (NULL != sf->next_block) {
1176 				struct btrfs_header *const next_hdr =
1177 				    (struct btrfs_header *)
1178 				    sf->next_block_ctx.datav[0];
1179 
1180 				next_stack = btrfsic_stack_frame_alloc();
1181 				if (NULL == next_stack) {
1182 					sf->error = -1;
1183 					goto one_stack_frame_backwards;
1184 				}
1185 
1186 				next_stack->i = -1;
1187 				next_stack->block = sf->next_block;
1188 				next_stack->block_ctx = &sf->next_block_ctx;
1189 				next_stack->next_block = NULL;
1190 				next_stack->hdr = next_hdr;
1191 				next_stack->limit_nesting =
1192 				    sf->limit_nesting - 1;
1193 				next_stack->prev = sf;
1194 				sf = next_stack;
1195 				goto continue_with_new_stack_frame;
1196 			}
1197 
1198 			goto continue_with_current_node_stack_frame;
1199 		}
1200 	}
1201 
1202 one_stack_frame_backwards:
1203 	if (NULL != sf->prev) {
1204 		struct btrfsic_stack_frame *const prev = sf->prev;
1205 
1206 		/* the one for the initial block is freed in the caller */
1207 		btrfsic_release_block_ctx(sf->block_ctx);
1208 
1209 		if (sf->error) {
1210 			prev->error = sf->error;
1211 			btrfsic_stack_frame_free(sf);
1212 			sf = prev;
1213 			goto one_stack_frame_backwards;
1214 		}
1215 
1216 		btrfsic_stack_frame_free(sf);
1217 		sf = prev;
1218 		goto continue_with_new_stack_frame;
1219 	} else {
1220 		BUG_ON(&initial_stack_frame != sf);
1221 	}
1222 
1223 	return sf->error;
1224 }
1225 
1226 static void btrfsic_read_from_block_data(
1227 	struct btrfsic_block_data_ctx *block_ctx,
1228 	void *dstv, u32 offset, size_t len)
1229 {
1230 	size_t cur;
1231 	size_t offset_in_page;
1232 	char *kaddr;
1233 	char *dst = (char *)dstv;
1234 	size_t start_offset = block_ctx->start & ((u64)PAGE_SIZE - 1);
1235 	unsigned long i = (start_offset + offset) >> PAGE_SHIFT;
1236 
1237 	WARN_ON(offset + len > block_ctx->len);
1238 	offset_in_page = (start_offset + offset) & (PAGE_SIZE - 1);
1239 
1240 	while (len > 0) {
1241 		cur = min(len, ((size_t)PAGE_SIZE - offset_in_page));
1242 		BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_SIZE));
1243 		kaddr = block_ctx->datav[i];
1244 		memcpy(dst, kaddr + offset_in_page, cur);
1245 
1246 		dst += cur;
1247 		len -= cur;
1248 		offset_in_page = 0;
1249 		i++;
1250 	}
1251 }
1252 
1253 static int btrfsic_create_link_to_next_block(
1254 		struct btrfsic_state *state,
1255 		struct btrfsic_block *block,
1256 		struct btrfsic_block_data_ctx *block_ctx,
1257 		u64 next_bytenr,
1258 		int limit_nesting,
1259 		struct btrfsic_block_data_ctx *next_block_ctx,
1260 		struct btrfsic_block **next_blockp,
1261 		int force_iodone_flag,
1262 		int *num_copiesp, int *mirror_nump,
1263 		struct btrfs_disk_key *disk_key,
1264 		u64 parent_generation)
1265 {
1266 	struct btrfsic_block *next_block = NULL;
1267 	int ret;
1268 	struct btrfsic_block_link *l;
1269 	int did_alloc_block_link;
1270 	int block_was_created;
1271 
1272 	*next_blockp = NULL;
1273 	if (0 == *num_copiesp) {
1274 		*num_copiesp =
1275 		    btrfs_num_copies(state->root->fs_info,
1276 				     next_bytenr, state->metablock_size);
1277 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1278 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
1279 			       next_bytenr, *num_copiesp);
1280 		*mirror_nump = 1;
1281 	}
1282 
1283 	if (*mirror_nump > *num_copiesp)
1284 		return 0;
1285 
1286 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1287 		printk(KERN_INFO
1288 		       "btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1289 		       *mirror_nump);
1290 	ret = btrfsic_map_block(state, next_bytenr,
1291 				state->metablock_size,
1292 				next_block_ctx, *mirror_nump);
1293 	if (ret) {
1294 		printk(KERN_INFO
1295 		       "btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1296 		       next_bytenr, *mirror_nump);
1297 		btrfsic_release_block_ctx(next_block_ctx);
1298 		*next_blockp = NULL;
1299 		return -1;
1300 	}
1301 
1302 	next_block = btrfsic_block_lookup_or_add(state,
1303 						 next_block_ctx, "referenced ",
1304 						 1, force_iodone_flag,
1305 						 !force_iodone_flag,
1306 						 *mirror_nump,
1307 						 &block_was_created);
1308 	if (NULL == next_block) {
1309 		btrfsic_release_block_ctx(next_block_ctx);
1310 		*next_blockp = NULL;
1311 		return -1;
1312 	}
1313 	if (block_was_created) {
1314 		l = NULL;
1315 		next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1316 	} else {
1317 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1318 			if (next_block->logical_bytenr != next_bytenr &&
1319 			    !(!next_block->is_metadata &&
1320 			      0 == next_block->logical_bytenr))
1321 				printk(KERN_INFO
1322 				       "Referenced block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1323 				       next_bytenr, next_block_ctx->dev->name,
1324 				       next_block_ctx->dev_bytenr, *mirror_nump,
1325 				       btrfsic_get_block_type(state,
1326 							      next_block),
1327 				       next_block->logical_bytenr);
1328 			else
1329 				printk(KERN_INFO
1330 				       "Referenced block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1331 				       next_bytenr, next_block_ctx->dev->name,
1332 				       next_block_ctx->dev_bytenr, *mirror_nump,
1333 				       btrfsic_get_block_type(state,
1334 							      next_block));
1335 		}
1336 		next_block->logical_bytenr = next_bytenr;
1337 
1338 		next_block->mirror_num = *mirror_nump;
1339 		l = btrfsic_block_link_hashtable_lookup(
1340 				next_block_ctx->dev->bdev,
1341 				next_block_ctx->dev_bytenr,
1342 				block_ctx->dev->bdev,
1343 				block_ctx->dev_bytenr,
1344 				&state->block_link_hashtable);
1345 	}
1346 
1347 	next_block->disk_key = *disk_key;
1348 	if (NULL == l) {
1349 		l = btrfsic_block_link_alloc();
1350 		if (NULL == l) {
1351 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
1352 			btrfsic_release_block_ctx(next_block_ctx);
1353 			*next_blockp = NULL;
1354 			return -1;
1355 		}
1356 
1357 		did_alloc_block_link = 1;
1358 		l->block_ref_to = next_block;
1359 		l->block_ref_from = block;
1360 		l->ref_cnt = 1;
1361 		l->parent_generation = parent_generation;
1362 
1363 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1364 			btrfsic_print_add_link(state, l);
1365 
1366 		list_add(&l->node_ref_to, &block->ref_to_list);
1367 		list_add(&l->node_ref_from, &next_block->ref_from_list);
1368 
1369 		btrfsic_block_link_hashtable_add(l,
1370 						 &state->block_link_hashtable);
1371 	} else {
1372 		did_alloc_block_link = 0;
1373 		if (0 == limit_nesting) {
1374 			l->ref_cnt++;
1375 			l->parent_generation = parent_generation;
1376 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1377 				btrfsic_print_add_link(state, l);
1378 		}
1379 	}
1380 
1381 	if (limit_nesting > 0 && did_alloc_block_link) {
1382 		ret = btrfsic_read_block(state, next_block_ctx);
1383 		if (ret < (int)next_block_ctx->len) {
1384 			printk(KERN_INFO
1385 			       "btrfsic: read block @logical %llu failed!\n",
1386 			       next_bytenr);
1387 			btrfsic_release_block_ctx(next_block_ctx);
1388 			*next_blockp = NULL;
1389 			return -1;
1390 		}
1391 
1392 		*next_blockp = next_block;
1393 	} else {
1394 		*next_blockp = NULL;
1395 	}
1396 	(*mirror_nump)++;
1397 
1398 	return 0;
1399 }
1400 
1401 static int btrfsic_handle_extent_data(
1402 		struct btrfsic_state *state,
1403 		struct btrfsic_block *block,
1404 		struct btrfsic_block_data_ctx *block_ctx,
1405 		u32 item_offset, int force_iodone_flag)
1406 {
1407 	int ret;
1408 	struct btrfs_file_extent_item file_extent_item;
1409 	u64 file_extent_item_offset;
1410 	u64 next_bytenr;
1411 	u64 num_bytes;
1412 	u64 generation;
1413 	struct btrfsic_block_link *l;
1414 
1415 	file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1416 				  item_offset;
1417 	if (file_extent_item_offset +
1418 	    offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1419 	    block_ctx->len) {
1420 		printk(KERN_INFO
1421 		       "btrfsic: file item out of bounce at logical %llu, dev %s\n",
1422 		       block_ctx->start, block_ctx->dev->name);
1423 		return -1;
1424 	}
1425 
1426 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1427 		file_extent_item_offset,
1428 		offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1429 	if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1430 	    btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1431 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1432 			printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu\n",
1433 			       file_extent_item.type,
1434 			       btrfs_stack_file_extent_disk_bytenr(
1435 			       &file_extent_item));
1436 		return 0;
1437 	}
1438 
1439 	if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1440 	    block_ctx->len) {
1441 		printk(KERN_INFO
1442 		       "btrfsic: file item out of bounce at logical %llu, dev %s\n",
1443 		       block_ctx->start, block_ctx->dev->name);
1444 		return -1;
1445 	}
1446 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1447 				     file_extent_item_offset,
1448 				     sizeof(struct btrfs_file_extent_item));
1449 	next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item);
1450 	if (btrfs_stack_file_extent_compression(&file_extent_item) ==
1451 	    BTRFS_COMPRESS_NONE) {
1452 		next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item);
1453 		num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1454 	} else {
1455 		num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item);
1456 	}
1457 	generation = btrfs_stack_file_extent_generation(&file_extent_item);
1458 
1459 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1460 		printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu,"
1461 		       " offset = %llu, num_bytes = %llu\n",
1462 		       file_extent_item.type,
1463 		       btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1464 		       btrfs_stack_file_extent_offset(&file_extent_item),
1465 		       num_bytes);
1466 	while (num_bytes > 0) {
1467 		u32 chunk_len;
1468 		int num_copies;
1469 		int mirror_num;
1470 
1471 		if (num_bytes > state->datablock_size)
1472 			chunk_len = state->datablock_size;
1473 		else
1474 			chunk_len = num_bytes;
1475 
1476 		num_copies =
1477 		    btrfs_num_copies(state->root->fs_info,
1478 				     next_bytenr, state->datablock_size);
1479 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1480 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
1481 			       next_bytenr, num_copies);
1482 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1483 			struct btrfsic_block_data_ctx next_block_ctx;
1484 			struct btrfsic_block *next_block;
1485 			int block_was_created;
1486 
1487 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1488 				printk(KERN_INFO "btrfsic_handle_extent_data("
1489 				       "mirror_num=%d)\n", mirror_num);
1490 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1491 				printk(KERN_INFO
1492 				       "\tdisk_bytenr = %llu, num_bytes %u\n",
1493 				       next_bytenr, chunk_len);
1494 			ret = btrfsic_map_block(state, next_bytenr,
1495 						chunk_len, &next_block_ctx,
1496 						mirror_num);
1497 			if (ret) {
1498 				printk(KERN_INFO
1499 				       "btrfsic: btrfsic_map_block(@%llu,"
1500 				       " mirror=%d) failed!\n",
1501 				       next_bytenr, mirror_num);
1502 				return -1;
1503 			}
1504 
1505 			next_block = btrfsic_block_lookup_or_add(
1506 					state,
1507 					&next_block_ctx,
1508 					"referenced ",
1509 					0,
1510 					force_iodone_flag,
1511 					!force_iodone_flag,
1512 					mirror_num,
1513 					&block_was_created);
1514 			if (NULL == next_block) {
1515 				printk(KERN_INFO
1516 				       "btrfsic: error, kmalloc failed!\n");
1517 				btrfsic_release_block_ctx(&next_block_ctx);
1518 				return -1;
1519 			}
1520 			if (!block_was_created) {
1521 				if ((state->print_mask &
1522 				     BTRFSIC_PRINT_MASK_VERBOSE) &&
1523 				    next_block->logical_bytenr != next_bytenr &&
1524 				    !(!next_block->is_metadata &&
1525 				      0 == next_block->logical_bytenr)) {
1526 					printk(KERN_INFO
1527 					       "Referenced block"
1528 					       " @%llu (%s/%llu/%d)"
1529 					       " found in hash table, D,"
1530 					       " bytenr mismatch"
1531 					       " (!= stored %llu).\n",
1532 					       next_bytenr,
1533 					       next_block_ctx.dev->name,
1534 					       next_block_ctx.dev_bytenr,
1535 					       mirror_num,
1536 					       next_block->logical_bytenr);
1537 				}
1538 				next_block->logical_bytenr = next_bytenr;
1539 				next_block->mirror_num = mirror_num;
1540 			}
1541 
1542 			l = btrfsic_block_link_lookup_or_add(state,
1543 							     &next_block_ctx,
1544 							     next_block, block,
1545 							     generation);
1546 			btrfsic_release_block_ctx(&next_block_ctx);
1547 			if (NULL == l)
1548 				return -1;
1549 		}
1550 
1551 		next_bytenr += chunk_len;
1552 		num_bytes -= chunk_len;
1553 	}
1554 
1555 	return 0;
1556 }
1557 
1558 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1559 			     struct btrfsic_block_data_ctx *block_ctx_out,
1560 			     int mirror_num)
1561 {
1562 	int ret;
1563 	u64 length;
1564 	struct btrfs_bio *multi = NULL;
1565 	struct btrfs_device *device;
1566 
1567 	length = len;
1568 	ret = btrfs_map_block(state->root->fs_info, READ,
1569 			      bytenr, &length, &multi, mirror_num);
1570 
1571 	if (ret) {
1572 		block_ctx_out->start = 0;
1573 		block_ctx_out->dev_bytenr = 0;
1574 		block_ctx_out->len = 0;
1575 		block_ctx_out->dev = NULL;
1576 		block_ctx_out->datav = NULL;
1577 		block_ctx_out->pagev = NULL;
1578 		block_ctx_out->mem_to_free = NULL;
1579 
1580 		return ret;
1581 	}
1582 
1583 	device = multi->stripes[0].dev;
1584 	block_ctx_out->dev = btrfsic_dev_state_lookup(device->bdev);
1585 	block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1586 	block_ctx_out->start = bytenr;
1587 	block_ctx_out->len = len;
1588 	block_ctx_out->datav = NULL;
1589 	block_ctx_out->pagev = NULL;
1590 	block_ctx_out->mem_to_free = NULL;
1591 
1592 	kfree(multi);
1593 	if (NULL == block_ctx_out->dev) {
1594 		ret = -ENXIO;
1595 		printk(KERN_INFO "btrfsic: error, cannot lookup dev (#1)!\n");
1596 	}
1597 
1598 	return ret;
1599 }
1600 
1601 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1602 {
1603 	if (block_ctx->mem_to_free) {
1604 		unsigned int num_pages;
1605 
1606 		BUG_ON(!block_ctx->datav);
1607 		BUG_ON(!block_ctx->pagev);
1608 		num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1609 			    PAGE_SHIFT;
1610 		while (num_pages > 0) {
1611 			num_pages--;
1612 			if (block_ctx->datav[num_pages]) {
1613 				kunmap(block_ctx->pagev[num_pages]);
1614 				block_ctx->datav[num_pages] = NULL;
1615 			}
1616 			if (block_ctx->pagev[num_pages]) {
1617 				__free_page(block_ctx->pagev[num_pages]);
1618 				block_ctx->pagev[num_pages] = NULL;
1619 			}
1620 		}
1621 
1622 		kfree(block_ctx->mem_to_free);
1623 		block_ctx->mem_to_free = NULL;
1624 		block_ctx->pagev = NULL;
1625 		block_ctx->datav = NULL;
1626 	}
1627 }
1628 
1629 static int btrfsic_read_block(struct btrfsic_state *state,
1630 			      struct btrfsic_block_data_ctx *block_ctx)
1631 {
1632 	unsigned int num_pages;
1633 	unsigned int i;
1634 	u64 dev_bytenr;
1635 	int ret;
1636 
1637 	BUG_ON(block_ctx->datav);
1638 	BUG_ON(block_ctx->pagev);
1639 	BUG_ON(block_ctx->mem_to_free);
1640 	if (block_ctx->dev_bytenr & ((u64)PAGE_SIZE - 1)) {
1641 		printk(KERN_INFO
1642 		       "btrfsic: read_block() with unaligned bytenr %llu\n",
1643 		       block_ctx->dev_bytenr);
1644 		return -1;
1645 	}
1646 
1647 	num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1648 		    PAGE_SHIFT;
1649 	block_ctx->mem_to_free = kzalloc((sizeof(*block_ctx->datav) +
1650 					  sizeof(*block_ctx->pagev)) *
1651 					 num_pages, GFP_NOFS);
1652 	if (!block_ctx->mem_to_free)
1653 		return -ENOMEM;
1654 	block_ctx->datav = block_ctx->mem_to_free;
1655 	block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1656 	for (i = 0; i < num_pages; i++) {
1657 		block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1658 		if (!block_ctx->pagev[i])
1659 			return -1;
1660 	}
1661 
1662 	dev_bytenr = block_ctx->dev_bytenr;
1663 	for (i = 0; i < num_pages;) {
1664 		struct bio *bio;
1665 		unsigned int j;
1666 
1667 		bio = btrfs_io_bio_alloc(GFP_NOFS, num_pages - i);
1668 		if (!bio) {
1669 			printk(KERN_INFO
1670 			       "btrfsic: bio_alloc() for %u pages failed!\n",
1671 			       num_pages - i);
1672 			return -1;
1673 		}
1674 		bio->bi_bdev = block_ctx->dev->bdev;
1675 		bio->bi_iter.bi_sector = dev_bytenr >> 9;
1676 
1677 		for (j = i; j < num_pages; j++) {
1678 			ret = bio_add_page(bio, block_ctx->pagev[j],
1679 					   PAGE_SIZE, 0);
1680 			if (PAGE_SIZE != ret)
1681 				break;
1682 		}
1683 		if (j == i) {
1684 			printk(KERN_INFO
1685 			       "btrfsic: error, failed to add a single page!\n");
1686 			return -1;
1687 		}
1688 		if (submit_bio_wait(READ, bio)) {
1689 			printk(KERN_INFO
1690 			       "btrfsic: read error at logical %llu dev %s!\n",
1691 			       block_ctx->start, block_ctx->dev->name);
1692 			bio_put(bio);
1693 			return -1;
1694 		}
1695 		bio_put(bio);
1696 		dev_bytenr += (j - i) * PAGE_SIZE;
1697 		i = j;
1698 	}
1699 	for (i = 0; i < num_pages; i++) {
1700 		block_ctx->datav[i] = kmap(block_ctx->pagev[i]);
1701 		if (!block_ctx->datav[i]) {
1702 			printk(KERN_INFO "btrfsic: kmap() failed (dev %s)!\n",
1703 			       block_ctx->dev->name);
1704 			return -1;
1705 		}
1706 	}
1707 
1708 	return block_ctx->len;
1709 }
1710 
1711 static void btrfsic_dump_database(struct btrfsic_state *state)
1712 {
1713 	const struct btrfsic_block *b_all;
1714 
1715 	BUG_ON(NULL == state);
1716 
1717 	printk(KERN_INFO "all_blocks_list:\n");
1718 	list_for_each_entry(b_all, &state->all_blocks_list, all_blocks_node) {
1719 		const struct btrfsic_block_link *l;
1720 
1721 		printk(KERN_INFO "%c-block @%llu (%s/%llu/%d)\n",
1722 		       btrfsic_get_block_type(state, b_all),
1723 		       b_all->logical_bytenr, b_all->dev_state->name,
1724 		       b_all->dev_bytenr, b_all->mirror_num);
1725 
1726 		list_for_each_entry(l, &b_all->ref_to_list, node_ref_to) {
1727 			printk(KERN_INFO " %c @%llu (%s/%llu/%d)"
1728 			       " refers %u* to"
1729 			       " %c @%llu (%s/%llu/%d)\n",
1730 			       btrfsic_get_block_type(state, b_all),
1731 			       b_all->logical_bytenr, b_all->dev_state->name,
1732 			       b_all->dev_bytenr, b_all->mirror_num,
1733 			       l->ref_cnt,
1734 			       btrfsic_get_block_type(state, l->block_ref_to),
1735 			       l->block_ref_to->logical_bytenr,
1736 			       l->block_ref_to->dev_state->name,
1737 			       l->block_ref_to->dev_bytenr,
1738 			       l->block_ref_to->mirror_num);
1739 		}
1740 
1741 		list_for_each_entry(l, &b_all->ref_from_list, node_ref_from) {
1742 			printk(KERN_INFO " %c @%llu (%s/%llu/%d)"
1743 			       " is ref %u* from"
1744 			       " %c @%llu (%s/%llu/%d)\n",
1745 			       btrfsic_get_block_type(state, b_all),
1746 			       b_all->logical_bytenr, b_all->dev_state->name,
1747 			       b_all->dev_bytenr, b_all->mirror_num,
1748 			       l->ref_cnt,
1749 			       btrfsic_get_block_type(state, l->block_ref_from),
1750 			       l->block_ref_from->logical_bytenr,
1751 			       l->block_ref_from->dev_state->name,
1752 			       l->block_ref_from->dev_bytenr,
1753 			       l->block_ref_from->mirror_num);
1754 		}
1755 
1756 		printk(KERN_INFO "\n");
1757 	}
1758 }
1759 
1760 /*
1761  * Test whether the disk block contains a tree block (leaf or node)
1762  * (note that this test fails for the super block)
1763  */
1764 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
1765 				     char **datav, unsigned int num_pages)
1766 {
1767 	struct btrfs_header *h;
1768 	u8 csum[BTRFS_CSUM_SIZE];
1769 	u32 crc = ~(u32)0;
1770 	unsigned int i;
1771 
1772 	if (num_pages * PAGE_SIZE < state->metablock_size)
1773 		return 1; /* not metadata */
1774 	num_pages = state->metablock_size >> PAGE_SHIFT;
1775 	h = (struct btrfs_header *)datav[0];
1776 
1777 	if (memcmp(h->fsid, state->root->fs_info->fsid, BTRFS_UUID_SIZE))
1778 		return 1;
1779 
1780 	for (i = 0; i < num_pages; i++) {
1781 		u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1782 		size_t sublen = i ? PAGE_SIZE :
1783 				    (PAGE_SIZE - BTRFS_CSUM_SIZE);
1784 
1785 		crc = btrfs_crc32c(crc, data, sublen);
1786 	}
1787 	btrfs_csum_final(crc, csum);
1788 	if (memcmp(csum, h->csum, state->csum_size))
1789 		return 1;
1790 
1791 	return 0; /* is metadata */
1792 }
1793 
1794 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1795 					  u64 dev_bytenr, char **mapped_datav,
1796 					  unsigned int num_pages,
1797 					  struct bio *bio, int *bio_is_patched,
1798 					  struct buffer_head *bh,
1799 					  int submit_bio_bh_rw)
1800 {
1801 	int is_metadata;
1802 	struct btrfsic_block *block;
1803 	struct btrfsic_block_data_ctx block_ctx;
1804 	int ret;
1805 	struct btrfsic_state *state = dev_state->state;
1806 	struct block_device *bdev = dev_state->bdev;
1807 	unsigned int processed_len;
1808 
1809 	if (NULL != bio_is_patched)
1810 		*bio_is_patched = 0;
1811 
1812 again:
1813 	if (num_pages == 0)
1814 		return;
1815 
1816 	processed_len = 0;
1817 	is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1818 						      num_pages));
1819 
1820 	block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1821 					       &state->block_hashtable);
1822 	if (NULL != block) {
1823 		u64 bytenr = 0;
1824 		struct btrfsic_block_link *l, *tmp;
1825 
1826 		if (block->is_superblock) {
1827 			bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1828 						    mapped_datav[0]);
1829 			if (num_pages * PAGE_SIZE <
1830 			    BTRFS_SUPER_INFO_SIZE) {
1831 				printk(KERN_INFO
1832 				       "btrfsic: cannot work with too short bios!\n");
1833 				return;
1834 			}
1835 			is_metadata = 1;
1836 			BUG_ON(BTRFS_SUPER_INFO_SIZE & (PAGE_SIZE - 1));
1837 			processed_len = BTRFS_SUPER_INFO_SIZE;
1838 			if (state->print_mask &
1839 			    BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1840 				printk(KERN_INFO
1841 				       "[before new superblock is written]:\n");
1842 				btrfsic_dump_tree_sub(state, block, 0);
1843 			}
1844 		}
1845 		if (is_metadata) {
1846 			if (!block->is_superblock) {
1847 				if (num_pages * PAGE_SIZE <
1848 				    state->metablock_size) {
1849 					printk(KERN_INFO
1850 					       "btrfsic: cannot work with too short bios!\n");
1851 					return;
1852 				}
1853 				processed_len = state->metablock_size;
1854 				bytenr = btrfs_stack_header_bytenr(
1855 						(struct btrfs_header *)
1856 						mapped_datav[0]);
1857 				btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1858 							       dev_state,
1859 							       dev_bytenr);
1860 			}
1861 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1862 				if (block->logical_bytenr != bytenr &&
1863 				    !(!block->is_metadata &&
1864 				      block->logical_bytenr == 0))
1865 					printk(KERN_INFO
1866 					       "Written block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1867 					       bytenr, dev_state->name,
1868 					       dev_bytenr,
1869 					       block->mirror_num,
1870 					       btrfsic_get_block_type(state,
1871 								      block),
1872 					       block->logical_bytenr);
1873 				else
1874 					printk(KERN_INFO
1875 					       "Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1876 					       bytenr, dev_state->name,
1877 					       dev_bytenr, block->mirror_num,
1878 					       btrfsic_get_block_type(state,
1879 								      block));
1880 			}
1881 			block->logical_bytenr = bytenr;
1882 		} else {
1883 			if (num_pages * PAGE_SIZE <
1884 			    state->datablock_size) {
1885 				printk(KERN_INFO
1886 				       "btrfsic: cannot work with too short bios!\n");
1887 				return;
1888 			}
1889 			processed_len = state->datablock_size;
1890 			bytenr = block->logical_bytenr;
1891 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1892 				printk(KERN_INFO
1893 				       "Written block @%llu (%s/%llu/%d)"
1894 				       " found in hash table, %c.\n",
1895 				       bytenr, dev_state->name, dev_bytenr,
1896 				       block->mirror_num,
1897 				       btrfsic_get_block_type(state, block));
1898 		}
1899 
1900 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1901 			printk(KERN_INFO
1902 			       "ref_to_list: %cE, ref_from_list: %cE\n",
1903 			       list_empty(&block->ref_to_list) ? ' ' : '!',
1904 			       list_empty(&block->ref_from_list) ? ' ' : '!');
1905 		if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1906 			printk(KERN_INFO "btrfs: attempt to overwrite %c-block"
1907 			       " @%llu (%s/%llu/%d), old(gen=%llu,"
1908 			       " objectid=%llu, type=%d, offset=%llu),"
1909 			       " new(gen=%llu),"
1910 			       " which is referenced by most recent superblock"
1911 			       " (superblockgen=%llu)!\n",
1912 			       btrfsic_get_block_type(state, block), bytenr,
1913 			       dev_state->name, dev_bytenr, block->mirror_num,
1914 			       block->generation,
1915 			       btrfs_disk_key_objectid(&block->disk_key),
1916 			       block->disk_key.type,
1917 			       btrfs_disk_key_offset(&block->disk_key),
1918 			       btrfs_stack_header_generation(
1919 				       (struct btrfs_header *) mapped_datav[0]),
1920 			       state->max_superblock_generation);
1921 			btrfsic_dump_tree(state);
1922 		}
1923 
1924 		if (!block->is_iodone && !block->never_written) {
1925 			printk(KERN_INFO "btrfs: attempt to overwrite %c-block"
1926 			       " @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu,"
1927 			       " which is not yet iodone!\n",
1928 			       btrfsic_get_block_type(state, block), bytenr,
1929 			       dev_state->name, dev_bytenr, block->mirror_num,
1930 			       block->generation,
1931 			       btrfs_stack_header_generation(
1932 				       (struct btrfs_header *)
1933 				       mapped_datav[0]));
1934 			/* it would not be safe to go on */
1935 			btrfsic_dump_tree(state);
1936 			goto continue_loop;
1937 		}
1938 
1939 		/*
1940 		 * Clear all references of this block. Do not free
1941 		 * the block itself even if is not referenced anymore
1942 		 * because it still carries valuable information
1943 		 * like whether it was ever written and IO completed.
1944 		 */
1945 		list_for_each_entry_safe(l, tmp, &block->ref_to_list,
1946 					 node_ref_to) {
1947 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1948 				btrfsic_print_rem_link(state, l);
1949 			l->ref_cnt--;
1950 			if (0 == l->ref_cnt) {
1951 				list_del(&l->node_ref_to);
1952 				list_del(&l->node_ref_from);
1953 				btrfsic_block_link_hashtable_remove(l);
1954 				btrfsic_block_link_free(l);
1955 			}
1956 		}
1957 
1958 		block_ctx.dev = dev_state;
1959 		block_ctx.dev_bytenr = dev_bytenr;
1960 		block_ctx.start = bytenr;
1961 		block_ctx.len = processed_len;
1962 		block_ctx.pagev = NULL;
1963 		block_ctx.mem_to_free = NULL;
1964 		block_ctx.datav = mapped_datav;
1965 
1966 		if (is_metadata || state->include_extent_data) {
1967 			block->never_written = 0;
1968 			block->iodone_w_error = 0;
1969 			if (NULL != bio) {
1970 				block->is_iodone = 0;
1971 				BUG_ON(NULL == bio_is_patched);
1972 				if (!*bio_is_patched) {
1973 					block->orig_bio_bh_private =
1974 					    bio->bi_private;
1975 					block->orig_bio_bh_end_io.bio =
1976 					    bio->bi_end_io;
1977 					block->next_in_same_bio = NULL;
1978 					bio->bi_private = block;
1979 					bio->bi_end_io = btrfsic_bio_end_io;
1980 					*bio_is_patched = 1;
1981 				} else {
1982 					struct btrfsic_block *chained_block =
1983 					    (struct btrfsic_block *)
1984 					    bio->bi_private;
1985 
1986 					BUG_ON(NULL == chained_block);
1987 					block->orig_bio_bh_private =
1988 					    chained_block->orig_bio_bh_private;
1989 					block->orig_bio_bh_end_io.bio =
1990 					    chained_block->orig_bio_bh_end_io.
1991 					    bio;
1992 					block->next_in_same_bio = chained_block;
1993 					bio->bi_private = block;
1994 				}
1995 			} else if (NULL != bh) {
1996 				block->is_iodone = 0;
1997 				block->orig_bio_bh_private = bh->b_private;
1998 				block->orig_bio_bh_end_io.bh = bh->b_end_io;
1999 				block->next_in_same_bio = NULL;
2000 				bh->b_private = block;
2001 				bh->b_end_io = btrfsic_bh_end_io;
2002 			} else {
2003 				block->is_iodone = 1;
2004 				block->orig_bio_bh_private = NULL;
2005 				block->orig_bio_bh_end_io.bio = NULL;
2006 				block->next_in_same_bio = NULL;
2007 			}
2008 		}
2009 
2010 		block->flush_gen = dev_state->last_flush_gen + 1;
2011 		block->submit_bio_bh_rw = submit_bio_bh_rw;
2012 		if (is_metadata) {
2013 			block->logical_bytenr = bytenr;
2014 			block->is_metadata = 1;
2015 			if (block->is_superblock) {
2016 				BUG_ON(PAGE_SIZE !=
2017 				       BTRFS_SUPER_INFO_SIZE);
2018 				ret = btrfsic_process_written_superblock(
2019 						state,
2020 						block,
2021 						(struct btrfs_super_block *)
2022 						mapped_datav[0]);
2023 				if (state->print_mask &
2024 				    BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
2025 					printk(KERN_INFO
2026 					"[after new superblock is written]:\n");
2027 					btrfsic_dump_tree_sub(state, block, 0);
2028 				}
2029 			} else {
2030 				block->mirror_num = 0;	/* unknown */
2031 				ret = btrfsic_process_metablock(
2032 						state,
2033 						block,
2034 						&block_ctx,
2035 						0, 0);
2036 			}
2037 			if (ret)
2038 				printk(KERN_INFO
2039 				       "btrfsic: btrfsic_process_metablock"
2040 				       "(root @%llu) failed!\n",
2041 				       dev_bytenr);
2042 		} else {
2043 			block->is_metadata = 0;
2044 			block->mirror_num = 0;	/* unknown */
2045 			block->generation = BTRFSIC_GENERATION_UNKNOWN;
2046 			if (!state->include_extent_data
2047 			    && list_empty(&block->ref_from_list)) {
2048 				/*
2049 				 * disk block is overwritten with extent
2050 				 * data (not meta data) and we are configured
2051 				 * to not include extent data: take the
2052 				 * chance and free the block's memory
2053 				 */
2054 				btrfsic_block_hashtable_remove(block);
2055 				list_del(&block->all_blocks_node);
2056 				btrfsic_block_free(block);
2057 			}
2058 		}
2059 		btrfsic_release_block_ctx(&block_ctx);
2060 	} else {
2061 		/* block has not been found in hash table */
2062 		u64 bytenr;
2063 
2064 		if (!is_metadata) {
2065 			processed_len = state->datablock_size;
2066 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2067 				printk(KERN_INFO "Written block (%s/%llu/?)"
2068 				       " !found in hash table, D.\n",
2069 				       dev_state->name, dev_bytenr);
2070 			if (!state->include_extent_data) {
2071 				/* ignore that written D block */
2072 				goto continue_loop;
2073 			}
2074 
2075 			/* this is getting ugly for the
2076 			 * include_extent_data case... */
2077 			bytenr = 0;	/* unknown */
2078 		} else {
2079 			processed_len = state->metablock_size;
2080 			bytenr = btrfs_stack_header_bytenr(
2081 					(struct btrfs_header *)
2082 					mapped_datav[0]);
2083 			btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
2084 						       dev_bytenr);
2085 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2086 				printk(KERN_INFO
2087 				       "Written block @%llu (%s/%llu/?)"
2088 				       " !found in hash table, M.\n",
2089 				       bytenr, dev_state->name, dev_bytenr);
2090 		}
2091 
2092 		block_ctx.dev = dev_state;
2093 		block_ctx.dev_bytenr = dev_bytenr;
2094 		block_ctx.start = bytenr;
2095 		block_ctx.len = processed_len;
2096 		block_ctx.pagev = NULL;
2097 		block_ctx.mem_to_free = NULL;
2098 		block_ctx.datav = mapped_datav;
2099 
2100 		block = btrfsic_block_alloc();
2101 		if (NULL == block) {
2102 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
2103 			btrfsic_release_block_ctx(&block_ctx);
2104 			goto continue_loop;
2105 		}
2106 		block->dev_state = dev_state;
2107 		block->dev_bytenr = dev_bytenr;
2108 		block->logical_bytenr = bytenr;
2109 		block->is_metadata = is_metadata;
2110 		block->never_written = 0;
2111 		block->iodone_w_error = 0;
2112 		block->mirror_num = 0;	/* unknown */
2113 		block->flush_gen = dev_state->last_flush_gen + 1;
2114 		block->submit_bio_bh_rw = submit_bio_bh_rw;
2115 		if (NULL != bio) {
2116 			block->is_iodone = 0;
2117 			BUG_ON(NULL == bio_is_patched);
2118 			if (!*bio_is_patched) {
2119 				block->orig_bio_bh_private = bio->bi_private;
2120 				block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2121 				block->next_in_same_bio = NULL;
2122 				bio->bi_private = block;
2123 				bio->bi_end_io = btrfsic_bio_end_io;
2124 				*bio_is_patched = 1;
2125 			} else {
2126 				struct btrfsic_block *chained_block =
2127 				    (struct btrfsic_block *)
2128 				    bio->bi_private;
2129 
2130 				BUG_ON(NULL == chained_block);
2131 				block->orig_bio_bh_private =
2132 				    chained_block->orig_bio_bh_private;
2133 				block->orig_bio_bh_end_io.bio =
2134 				    chained_block->orig_bio_bh_end_io.bio;
2135 				block->next_in_same_bio = chained_block;
2136 				bio->bi_private = block;
2137 			}
2138 		} else if (NULL != bh) {
2139 			block->is_iodone = 0;
2140 			block->orig_bio_bh_private = bh->b_private;
2141 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2142 			block->next_in_same_bio = NULL;
2143 			bh->b_private = block;
2144 			bh->b_end_io = btrfsic_bh_end_io;
2145 		} else {
2146 			block->is_iodone = 1;
2147 			block->orig_bio_bh_private = NULL;
2148 			block->orig_bio_bh_end_io.bio = NULL;
2149 			block->next_in_same_bio = NULL;
2150 		}
2151 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2152 			printk(KERN_INFO
2153 			       "New written %c-block @%llu (%s/%llu/%d)\n",
2154 			       is_metadata ? 'M' : 'D',
2155 			       block->logical_bytenr, block->dev_state->name,
2156 			       block->dev_bytenr, block->mirror_num);
2157 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2158 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2159 
2160 		if (is_metadata) {
2161 			ret = btrfsic_process_metablock(state, block,
2162 							&block_ctx, 0, 0);
2163 			if (ret)
2164 				printk(KERN_INFO
2165 				       "btrfsic: process_metablock(root @%llu)"
2166 				       " failed!\n",
2167 				       dev_bytenr);
2168 		}
2169 		btrfsic_release_block_ctx(&block_ctx);
2170 	}
2171 
2172 continue_loop:
2173 	BUG_ON(!processed_len);
2174 	dev_bytenr += processed_len;
2175 	mapped_datav += processed_len >> PAGE_SHIFT;
2176 	num_pages -= processed_len >> PAGE_SHIFT;
2177 	goto again;
2178 }
2179 
2180 static void btrfsic_bio_end_io(struct bio *bp)
2181 {
2182 	struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2183 	int iodone_w_error;
2184 
2185 	/* mutex is not held! This is not save if IO is not yet completed
2186 	 * on umount */
2187 	iodone_w_error = 0;
2188 	if (bp->bi_error)
2189 		iodone_w_error = 1;
2190 
2191 	BUG_ON(NULL == block);
2192 	bp->bi_private = block->orig_bio_bh_private;
2193 	bp->bi_end_io = block->orig_bio_bh_end_io.bio;
2194 
2195 	do {
2196 		struct btrfsic_block *next_block;
2197 		struct btrfsic_dev_state *const dev_state = block->dev_state;
2198 
2199 		if ((dev_state->state->print_mask &
2200 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2201 			printk(KERN_INFO
2202 			       "bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2203 			       bp->bi_error,
2204 			       btrfsic_get_block_type(dev_state->state, block),
2205 			       block->logical_bytenr, dev_state->name,
2206 			       block->dev_bytenr, block->mirror_num);
2207 		next_block = block->next_in_same_bio;
2208 		block->iodone_w_error = iodone_w_error;
2209 		if (block->submit_bio_bh_rw & REQ_FLUSH) {
2210 			dev_state->last_flush_gen++;
2211 			if ((dev_state->state->print_mask &
2212 			     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2213 				printk(KERN_INFO
2214 				       "bio_end_io() new %s flush_gen=%llu\n",
2215 				       dev_state->name,
2216 				       dev_state->last_flush_gen);
2217 		}
2218 		if (block->submit_bio_bh_rw & REQ_FUA)
2219 			block->flush_gen = 0; /* FUA completed means block is
2220 					       * on disk */
2221 		block->is_iodone = 1; /* for FLUSH, this releases the block */
2222 		block = next_block;
2223 	} while (NULL != block);
2224 
2225 	bp->bi_end_io(bp);
2226 }
2227 
2228 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate)
2229 {
2230 	struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private;
2231 	int iodone_w_error = !uptodate;
2232 	struct btrfsic_dev_state *dev_state;
2233 
2234 	BUG_ON(NULL == block);
2235 	dev_state = block->dev_state;
2236 	if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2237 		printk(KERN_INFO
2238 		       "bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n",
2239 		       iodone_w_error,
2240 		       btrfsic_get_block_type(dev_state->state, block),
2241 		       block->logical_bytenr, block->dev_state->name,
2242 		       block->dev_bytenr, block->mirror_num);
2243 
2244 	block->iodone_w_error = iodone_w_error;
2245 	if (block->submit_bio_bh_rw & REQ_FLUSH) {
2246 		dev_state->last_flush_gen++;
2247 		if ((dev_state->state->print_mask &
2248 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2249 			printk(KERN_INFO
2250 			       "bh_end_io() new %s flush_gen=%llu\n",
2251 			       dev_state->name, dev_state->last_flush_gen);
2252 	}
2253 	if (block->submit_bio_bh_rw & REQ_FUA)
2254 		block->flush_gen = 0; /* FUA completed means block is on disk */
2255 
2256 	bh->b_private = block->orig_bio_bh_private;
2257 	bh->b_end_io = block->orig_bio_bh_end_io.bh;
2258 	block->is_iodone = 1; /* for FLUSH, this releases the block */
2259 	bh->b_end_io(bh, uptodate);
2260 }
2261 
2262 static int btrfsic_process_written_superblock(
2263 		struct btrfsic_state *state,
2264 		struct btrfsic_block *const superblock,
2265 		struct btrfs_super_block *const super_hdr)
2266 {
2267 	int pass;
2268 
2269 	superblock->generation = btrfs_super_generation(super_hdr);
2270 	if (!(superblock->generation > state->max_superblock_generation ||
2271 	      0 == state->max_superblock_generation)) {
2272 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2273 			printk(KERN_INFO
2274 			       "btrfsic: superblock @%llu (%s/%llu/%d)"
2275 			       " with old gen %llu <= %llu\n",
2276 			       superblock->logical_bytenr,
2277 			       superblock->dev_state->name,
2278 			       superblock->dev_bytenr, superblock->mirror_num,
2279 			       btrfs_super_generation(super_hdr),
2280 			       state->max_superblock_generation);
2281 	} else {
2282 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2283 			printk(KERN_INFO
2284 			       "btrfsic: got new superblock @%llu (%s/%llu/%d)"
2285 			       " with new gen %llu > %llu\n",
2286 			       superblock->logical_bytenr,
2287 			       superblock->dev_state->name,
2288 			       superblock->dev_bytenr, superblock->mirror_num,
2289 			       btrfs_super_generation(super_hdr),
2290 			       state->max_superblock_generation);
2291 
2292 		state->max_superblock_generation =
2293 		    btrfs_super_generation(super_hdr);
2294 		state->latest_superblock = superblock;
2295 	}
2296 
2297 	for (pass = 0; pass < 3; pass++) {
2298 		int ret;
2299 		u64 next_bytenr;
2300 		struct btrfsic_block *next_block;
2301 		struct btrfsic_block_data_ctx tmp_next_block_ctx;
2302 		struct btrfsic_block_link *l;
2303 		int num_copies;
2304 		int mirror_num;
2305 		const char *additional_string = NULL;
2306 		struct btrfs_disk_key tmp_disk_key = {0};
2307 
2308 		btrfs_set_disk_key_objectid(&tmp_disk_key,
2309 					    BTRFS_ROOT_ITEM_KEY);
2310 		btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2311 
2312 		switch (pass) {
2313 		case 0:
2314 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2315 						    BTRFS_ROOT_TREE_OBJECTID);
2316 			additional_string = "root ";
2317 			next_bytenr = btrfs_super_root(super_hdr);
2318 			if (state->print_mask &
2319 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2320 				printk(KERN_INFO "root@%llu\n", next_bytenr);
2321 			break;
2322 		case 1:
2323 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2324 						    BTRFS_CHUNK_TREE_OBJECTID);
2325 			additional_string = "chunk ";
2326 			next_bytenr = btrfs_super_chunk_root(super_hdr);
2327 			if (state->print_mask &
2328 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2329 				printk(KERN_INFO "chunk@%llu\n", next_bytenr);
2330 			break;
2331 		case 2:
2332 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2333 						    BTRFS_TREE_LOG_OBJECTID);
2334 			additional_string = "log ";
2335 			next_bytenr = btrfs_super_log_root(super_hdr);
2336 			if (0 == next_bytenr)
2337 				continue;
2338 			if (state->print_mask &
2339 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2340 				printk(KERN_INFO "log@%llu\n", next_bytenr);
2341 			break;
2342 		}
2343 
2344 		num_copies =
2345 		    btrfs_num_copies(state->root->fs_info,
2346 				     next_bytenr, BTRFS_SUPER_INFO_SIZE);
2347 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2348 			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
2349 			       next_bytenr, num_copies);
2350 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2351 			int was_created;
2352 
2353 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2354 				printk(KERN_INFO
2355 				       "btrfsic_process_written_superblock("
2356 				       "mirror_num=%d)\n", mirror_num);
2357 			ret = btrfsic_map_block(state, next_bytenr,
2358 						BTRFS_SUPER_INFO_SIZE,
2359 						&tmp_next_block_ctx,
2360 						mirror_num);
2361 			if (ret) {
2362 				printk(KERN_INFO
2363 				       "btrfsic: btrfsic_map_block(@%llu,"
2364 				       " mirror=%d) failed!\n",
2365 				       next_bytenr, mirror_num);
2366 				return -1;
2367 			}
2368 
2369 			next_block = btrfsic_block_lookup_or_add(
2370 					state,
2371 					&tmp_next_block_ctx,
2372 					additional_string,
2373 					1, 0, 1,
2374 					mirror_num,
2375 					&was_created);
2376 			if (NULL == next_block) {
2377 				printk(KERN_INFO
2378 				       "btrfsic: error, kmalloc failed!\n");
2379 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
2380 				return -1;
2381 			}
2382 
2383 			next_block->disk_key = tmp_disk_key;
2384 			if (was_created)
2385 				next_block->generation =
2386 				    BTRFSIC_GENERATION_UNKNOWN;
2387 			l = btrfsic_block_link_lookup_or_add(
2388 					state,
2389 					&tmp_next_block_ctx,
2390 					next_block,
2391 					superblock,
2392 					BTRFSIC_GENERATION_UNKNOWN);
2393 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
2394 			if (NULL == l)
2395 				return -1;
2396 		}
2397 	}
2398 
2399 	if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2400 		btrfsic_dump_tree(state);
2401 
2402 	return 0;
2403 }
2404 
2405 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2406 					struct btrfsic_block *const block,
2407 					int recursion_level)
2408 {
2409 	const struct btrfsic_block_link *l;
2410 	int ret = 0;
2411 
2412 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2413 		/*
2414 		 * Note that this situation can happen and does not
2415 		 * indicate an error in regular cases. It happens
2416 		 * when disk blocks are freed and later reused.
2417 		 * The check-integrity module is not aware of any
2418 		 * block free operations, it just recognizes block
2419 		 * write operations. Therefore it keeps the linkage
2420 		 * information for a block until a block is
2421 		 * rewritten. This can temporarily cause incorrect
2422 		 * and even circular linkage informations. This
2423 		 * causes no harm unless such blocks are referenced
2424 		 * by the most recent super block.
2425 		 */
2426 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2427 			printk(KERN_INFO
2428 			       "btrfsic: abort cyclic linkage (case 1).\n");
2429 
2430 		return ret;
2431 	}
2432 
2433 	/*
2434 	 * This algorithm is recursive because the amount of used stack
2435 	 * space is very small and the max recursion depth is limited.
2436 	 */
2437 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2438 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2439 			printk(KERN_INFO
2440 			       "rl=%d, %c @%llu (%s/%llu/%d)"
2441 			       " %u* refers to %c @%llu (%s/%llu/%d)\n",
2442 			       recursion_level,
2443 			       btrfsic_get_block_type(state, block),
2444 			       block->logical_bytenr, block->dev_state->name,
2445 			       block->dev_bytenr, block->mirror_num,
2446 			       l->ref_cnt,
2447 			       btrfsic_get_block_type(state, l->block_ref_to),
2448 			       l->block_ref_to->logical_bytenr,
2449 			       l->block_ref_to->dev_state->name,
2450 			       l->block_ref_to->dev_bytenr,
2451 			       l->block_ref_to->mirror_num);
2452 		if (l->block_ref_to->never_written) {
2453 			printk(KERN_INFO "btrfs: attempt to write superblock"
2454 			       " which references block %c @%llu (%s/%llu/%d)"
2455 			       " which is never written!\n",
2456 			       btrfsic_get_block_type(state, l->block_ref_to),
2457 			       l->block_ref_to->logical_bytenr,
2458 			       l->block_ref_to->dev_state->name,
2459 			       l->block_ref_to->dev_bytenr,
2460 			       l->block_ref_to->mirror_num);
2461 			ret = -1;
2462 		} else if (!l->block_ref_to->is_iodone) {
2463 			printk(KERN_INFO "btrfs: attempt to write superblock"
2464 			       " which references block %c @%llu (%s/%llu/%d)"
2465 			       " which is not yet iodone!\n",
2466 			       btrfsic_get_block_type(state, l->block_ref_to),
2467 			       l->block_ref_to->logical_bytenr,
2468 			       l->block_ref_to->dev_state->name,
2469 			       l->block_ref_to->dev_bytenr,
2470 			       l->block_ref_to->mirror_num);
2471 			ret = -1;
2472 		} else if (l->block_ref_to->iodone_w_error) {
2473 			printk(KERN_INFO "btrfs: attempt to write superblock"
2474 			       " which references block %c @%llu (%s/%llu/%d)"
2475 			       " which has write error!\n",
2476 			       btrfsic_get_block_type(state, l->block_ref_to),
2477 			       l->block_ref_to->logical_bytenr,
2478 			       l->block_ref_to->dev_state->name,
2479 			       l->block_ref_to->dev_bytenr,
2480 			       l->block_ref_to->mirror_num);
2481 			ret = -1;
2482 		} else if (l->parent_generation !=
2483 			   l->block_ref_to->generation &&
2484 			   BTRFSIC_GENERATION_UNKNOWN !=
2485 			   l->parent_generation &&
2486 			   BTRFSIC_GENERATION_UNKNOWN !=
2487 			   l->block_ref_to->generation) {
2488 			printk(KERN_INFO "btrfs: attempt to write superblock"
2489 			       " which references block %c @%llu (%s/%llu/%d)"
2490 			       " with generation %llu !="
2491 			       " parent generation %llu!\n",
2492 			       btrfsic_get_block_type(state, l->block_ref_to),
2493 			       l->block_ref_to->logical_bytenr,
2494 			       l->block_ref_to->dev_state->name,
2495 			       l->block_ref_to->dev_bytenr,
2496 			       l->block_ref_to->mirror_num,
2497 			       l->block_ref_to->generation,
2498 			       l->parent_generation);
2499 			ret = -1;
2500 		} else if (l->block_ref_to->flush_gen >
2501 			   l->block_ref_to->dev_state->last_flush_gen) {
2502 			printk(KERN_INFO "btrfs: attempt to write superblock"
2503 			       " which references block %c @%llu (%s/%llu/%d)"
2504 			       " which is not flushed out of disk's write cache"
2505 			       " (block flush_gen=%llu,"
2506 			       " dev->flush_gen=%llu)!\n",
2507 			       btrfsic_get_block_type(state, l->block_ref_to),
2508 			       l->block_ref_to->logical_bytenr,
2509 			       l->block_ref_to->dev_state->name,
2510 			       l->block_ref_to->dev_bytenr,
2511 			       l->block_ref_to->mirror_num, block->flush_gen,
2512 			       l->block_ref_to->dev_state->last_flush_gen);
2513 			ret = -1;
2514 		} else if (-1 == btrfsic_check_all_ref_blocks(state,
2515 							      l->block_ref_to,
2516 							      recursion_level +
2517 							      1)) {
2518 			ret = -1;
2519 		}
2520 	}
2521 
2522 	return ret;
2523 }
2524 
2525 static int btrfsic_is_block_ref_by_superblock(
2526 		const struct btrfsic_state *state,
2527 		const struct btrfsic_block *block,
2528 		int recursion_level)
2529 {
2530 	const struct btrfsic_block_link *l;
2531 
2532 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2533 		/* refer to comment at "abort cyclic linkage (case 1)" */
2534 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2535 			printk(KERN_INFO
2536 			       "btrfsic: abort cyclic linkage (case 2).\n");
2537 
2538 		return 0;
2539 	}
2540 
2541 	/*
2542 	 * This algorithm is recursive because the amount of used stack space
2543 	 * is very small and the max recursion depth is limited.
2544 	 */
2545 	list_for_each_entry(l, &block->ref_from_list, node_ref_from) {
2546 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2547 			printk(KERN_INFO
2548 			       "rl=%d, %c @%llu (%s/%llu/%d)"
2549 			       " is ref %u* from %c @%llu (%s/%llu/%d)\n",
2550 			       recursion_level,
2551 			       btrfsic_get_block_type(state, block),
2552 			       block->logical_bytenr, block->dev_state->name,
2553 			       block->dev_bytenr, block->mirror_num,
2554 			       l->ref_cnt,
2555 			       btrfsic_get_block_type(state, l->block_ref_from),
2556 			       l->block_ref_from->logical_bytenr,
2557 			       l->block_ref_from->dev_state->name,
2558 			       l->block_ref_from->dev_bytenr,
2559 			       l->block_ref_from->mirror_num);
2560 		if (l->block_ref_from->is_superblock &&
2561 		    state->latest_superblock->dev_bytenr ==
2562 		    l->block_ref_from->dev_bytenr &&
2563 		    state->latest_superblock->dev_state->bdev ==
2564 		    l->block_ref_from->dev_state->bdev)
2565 			return 1;
2566 		else if (btrfsic_is_block_ref_by_superblock(state,
2567 							    l->block_ref_from,
2568 							    recursion_level +
2569 							    1))
2570 			return 1;
2571 	}
2572 
2573 	return 0;
2574 }
2575 
2576 static void btrfsic_print_add_link(const struct btrfsic_state *state,
2577 				   const struct btrfsic_block_link *l)
2578 {
2579 	printk(KERN_INFO
2580 	       "Add %u* link from %c @%llu (%s/%llu/%d)"
2581 	       " to %c @%llu (%s/%llu/%d).\n",
2582 	       l->ref_cnt,
2583 	       btrfsic_get_block_type(state, l->block_ref_from),
2584 	       l->block_ref_from->logical_bytenr,
2585 	       l->block_ref_from->dev_state->name,
2586 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2587 	       btrfsic_get_block_type(state, l->block_ref_to),
2588 	       l->block_ref_to->logical_bytenr,
2589 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2590 	       l->block_ref_to->mirror_num);
2591 }
2592 
2593 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2594 				   const struct btrfsic_block_link *l)
2595 {
2596 	printk(KERN_INFO
2597 	       "Rem %u* link from %c @%llu (%s/%llu/%d)"
2598 	       " to %c @%llu (%s/%llu/%d).\n",
2599 	       l->ref_cnt,
2600 	       btrfsic_get_block_type(state, l->block_ref_from),
2601 	       l->block_ref_from->logical_bytenr,
2602 	       l->block_ref_from->dev_state->name,
2603 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2604 	       btrfsic_get_block_type(state, l->block_ref_to),
2605 	       l->block_ref_to->logical_bytenr,
2606 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2607 	       l->block_ref_to->mirror_num);
2608 }
2609 
2610 static char btrfsic_get_block_type(const struct btrfsic_state *state,
2611 				   const struct btrfsic_block *block)
2612 {
2613 	if (block->is_superblock &&
2614 	    state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2615 	    state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2616 		return 'S';
2617 	else if (block->is_superblock)
2618 		return 's';
2619 	else if (block->is_metadata)
2620 		return 'M';
2621 	else
2622 		return 'D';
2623 }
2624 
2625 static void btrfsic_dump_tree(const struct btrfsic_state *state)
2626 {
2627 	btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2628 }
2629 
2630 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2631 				  const struct btrfsic_block *block,
2632 				  int indent_level)
2633 {
2634 	const struct btrfsic_block_link *l;
2635 	int indent_add;
2636 	static char buf[80];
2637 	int cursor_position;
2638 
2639 	/*
2640 	 * Should better fill an on-stack buffer with a complete line and
2641 	 * dump it at once when it is time to print a newline character.
2642 	 */
2643 
2644 	/*
2645 	 * This algorithm is recursive because the amount of used stack space
2646 	 * is very small and the max recursion depth is limited.
2647 	 */
2648 	indent_add = sprintf(buf, "%c-%llu(%s/%llu/%d)",
2649 			     btrfsic_get_block_type(state, block),
2650 			     block->logical_bytenr, block->dev_state->name,
2651 			     block->dev_bytenr, block->mirror_num);
2652 	if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2653 		printk("[...]\n");
2654 		return;
2655 	}
2656 	printk(buf);
2657 	indent_level += indent_add;
2658 	if (list_empty(&block->ref_to_list)) {
2659 		printk("\n");
2660 		return;
2661 	}
2662 	if (block->mirror_num > 1 &&
2663 	    !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2664 		printk(" [...]\n");
2665 		return;
2666 	}
2667 
2668 	cursor_position = indent_level;
2669 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2670 		while (cursor_position < indent_level) {
2671 			printk(" ");
2672 			cursor_position++;
2673 		}
2674 		if (l->ref_cnt > 1)
2675 			indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2676 		else
2677 			indent_add = sprintf(buf, " --> ");
2678 		if (indent_level + indent_add >
2679 		    BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2680 			printk("[...]\n");
2681 			cursor_position = 0;
2682 			continue;
2683 		}
2684 
2685 		printk(buf);
2686 
2687 		btrfsic_dump_tree_sub(state, l->block_ref_to,
2688 				      indent_level + indent_add);
2689 		cursor_position = 0;
2690 	}
2691 }
2692 
2693 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2694 		struct btrfsic_state *state,
2695 		struct btrfsic_block_data_ctx *next_block_ctx,
2696 		struct btrfsic_block *next_block,
2697 		struct btrfsic_block *from_block,
2698 		u64 parent_generation)
2699 {
2700 	struct btrfsic_block_link *l;
2701 
2702 	l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2703 						next_block_ctx->dev_bytenr,
2704 						from_block->dev_state->bdev,
2705 						from_block->dev_bytenr,
2706 						&state->block_link_hashtable);
2707 	if (NULL == l) {
2708 		l = btrfsic_block_link_alloc();
2709 		if (NULL == l) {
2710 			printk(KERN_INFO
2711 			       "btrfsic: error, kmalloc" " failed!\n");
2712 			return NULL;
2713 		}
2714 
2715 		l->block_ref_to = next_block;
2716 		l->block_ref_from = from_block;
2717 		l->ref_cnt = 1;
2718 		l->parent_generation = parent_generation;
2719 
2720 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2721 			btrfsic_print_add_link(state, l);
2722 
2723 		list_add(&l->node_ref_to, &from_block->ref_to_list);
2724 		list_add(&l->node_ref_from, &next_block->ref_from_list);
2725 
2726 		btrfsic_block_link_hashtable_add(l,
2727 						 &state->block_link_hashtable);
2728 	} else {
2729 		l->ref_cnt++;
2730 		l->parent_generation = parent_generation;
2731 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2732 			btrfsic_print_add_link(state, l);
2733 	}
2734 
2735 	return l;
2736 }
2737 
2738 static struct btrfsic_block *btrfsic_block_lookup_or_add(
2739 		struct btrfsic_state *state,
2740 		struct btrfsic_block_data_ctx *block_ctx,
2741 		const char *additional_string,
2742 		int is_metadata,
2743 		int is_iodone,
2744 		int never_written,
2745 		int mirror_num,
2746 		int *was_created)
2747 {
2748 	struct btrfsic_block *block;
2749 
2750 	block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2751 					       block_ctx->dev_bytenr,
2752 					       &state->block_hashtable);
2753 	if (NULL == block) {
2754 		struct btrfsic_dev_state *dev_state;
2755 
2756 		block = btrfsic_block_alloc();
2757 		if (NULL == block) {
2758 			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
2759 			return NULL;
2760 		}
2761 		dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev);
2762 		if (NULL == dev_state) {
2763 			printk(KERN_INFO
2764 			       "btrfsic: error, lookup dev_state failed!\n");
2765 			btrfsic_block_free(block);
2766 			return NULL;
2767 		}
2768 		block->dev_state = dev_state;
2769 		block->dev_bytenr = block_ctx->dev_bytenr;
2770 		block->logical_bytenr = block_ctx->start;
2771 		block->is_metadata = is_metadata;
2772 		block->is_iodone = is_iodone;
2773 		block->never_written = never_written;
2774 		block->mirror_num = mirror_num;
2775 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2776 			printk(KERN_INFO
2777 			       "New %s%c-block @%llu (%s/%llu/%d)\n",
2778 			       additional_string,
2779 			       btrfsic_get_block_type(state, block),
2780 			       block->logical_bytenr, dev_state->name,
2781 			       block->dev_bytenr, mirror_num);
2782 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2783 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2784 		if (NULL != was_created)
2785 			*was_created = 1;
2786 	} else {
2787 		if (NULL != was_created)
2788 			*was_created = 0;
2789 	}
2790 
2791 	return block;
2792 }
2793 
2794 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2795 					   u64 bytenr,
2796 					   struct btrfsic_dev_state *dev_state,
2797 					   u64 dev_bytenr)
2798 {
2799 	int num_copies;
2800 	int mirror_num;
2801 	int ret;
2802 	struct btrfsic_block_data_ctx block_ctx;
2803 	int match = 0;
2804 
2805 	num_copies = btrfs_num_copies(state->root->fs_info,
2806 				      bytenr, state->metablock_size);
2807 
2808 	for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2809 		ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2810 					&block_ctx, mirror_num);
2811 		if (ret) {
2812 			printk(KERN_INFO "btrfsic:"
2813 			       " btrfsic_map_block(logical @%llu,"
2814 			       " mirror %d) failed!\n",
2815 			       bytenr, mirror_num);
2816 			continue;
2817 		}
2818 
2819 		if (dev_state->bdev == block_ctx.dev->bdev &&
2820 		    dev_bytenr == block_ctx.dev_bytenr) {
2821 			match++;
2822 			btrfsic_release_block_ctx(&block_ctx);
2823 			break;
2824 		}
2825 		btrfsic_release_block_ctx(&block_ctx);
2826 	}
2827 
2828 	if (WARN_ON(!match)) {
2829 		printk(KERN_INFO "btrfs: attempt to write M-block which contains logical bytenr that doesn't map to dev+physical bytenr of submit_bio,"
2830 		       " buffer->log_bytenr=%llu, submit_bio(bdev=%s,"
2831 		       " phys_bytenr=%llu)!\n",
2832 		       bytenr, dev_state->name, dev_bytenr);
2833 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2834 			ret = btrfsic_map_block(state, bytenr,
2835 						state->metablock_size,
2836 						&block_ctx, mirror_num);
2837 			if (ret)
2838 				continue;
2839 
2840 			printk(KERN_INFO "Read logical bytenr @%llu maps to"
2841 			       " (%s/%llu/%d)\n",
2842 			       bytenr, block_ctx.dev->name,
2843 			       block_ctx.dev_bytenr, mirror_num);
2844 		}
2845 	}
2846 }
2847 
2848 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
2849 		struct block_device *bdev)
2850 {
2851 	struct btrfsic_dev_state *ds;
2852 
2853 	ds = btrfsic_dev_state_hashtable_lookup(bdev,
2854 						&btrfsic_dev_state_hashtable);
2855 	return ds;
2856 }
2857 
2858 int btrfsic_submit_bh(int rw, struct buffer_head *bh)
2859 {
2860 	struct btrfsic_dev_state *dev_state;
2861 
2862 	if (!btrfsic_is_initialized)
2863 		return submit_bh(rw, bh);
2864 
2865 	mutex_lock(&btrfsic_mutex);
2866 	/* since btrfsic_submit_bh() might also be called before
2867 	 * btrfsic_mount(), this might return NULL */
2868 	dev_state = btrfsic_dev_state_lookup(bh->b_bdev);
2869 
2870 	/* Only called to write the superblock (incl. FLUSH/FUA) */
2871 	if (NULL != dev_state &&
2872 	    (rw & WRITE) && bh->b_size > 0) {
2873 		u64 dev_bytenr;
2874 
2875 		dev_bytenr = 4096 * bh->b_blocknr;
2876 		if (dev_state->state->print_mask &
2877 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2878 			printk(KERN_INFO
2879 			       "submit_bh(rw=0x%x, blocknr=%llu (bytenr %llu),"
2880 			       " size=%zu, data=%p, bdev=%p)\n",
2881 			       rw, (unsigned long long)bh->b_blocknr,
2882 			       dev_bytenr, bh->b_size, bh->b_data, bh->b_bdev);
2883 		btrfsic_process_written_block(dev_state, dev_bytenr,
2884 					      &bh->b_data, 1, NULL,
2885 					      NULL, bh, rw);
2886 	} else if (NULL != dev_state && (rw & REQ_FLUSH)) {
2887 		if (dev_state->state->print_mask &
2888 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2889 			printk(KERN_INFO
2890 			       "submit_bh(rw=0x%x FLUSH, bdev=%p)\n",
2891 			       rw, bh->b_bdev);
2892 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2893 			if ((dev_state->state->print_mask &
2894 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2895 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2896 				printk(KERN_INFO
2897 				       "btrfsic_submit_bh(%s) with FLUSH"
2898 				       " but dummy block already in use"
2899 				       " (ignored)!\n",
2900 				       dev_state->name);
2901 		} else {
2902 			struct btrfsic_block *const block =
2903 				&dev_state->dummy_block_for_bio_bh_flush;
2904 
2905 			block->is_iodone = 0;
2906 			block->never_written = 0;
2907 			block->iodone_w_error = 0;
2908 			block->flush_gen = dev_state->last_flush_gen + 1;
2909 			block->submit_bio_bh_rw = rw;
2910 			block->orig_bio_bh_private = bh->b_private;
2911 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2912 			block->next_in_same_bio = NULL;
2913 			bh->b_private = block;
2914 			bh->b_end_io = btrfsic_bh_end_io;
2915 		}
2916 	}
2917 	mutex_unlock(&btrfsic_mutex);
2918 	return submit_bh(rw, bh);
2919 }
2920 
2921 static void __btrfsic_submit_bio(int rw, struct bio *bio)
2922 {
2923 	struct btrfsic_dev_state *dev_state;
2924 
2925 	if (!btrfsic_is_initialized)
2926 		return;
2927 
2928 	mutex_lock(&btrfsic_mutex);
2929 	/* since btrfsic_submit_bio() is also called before
2930 	 * btrfsic_mount(), this might return NULL */
2931 	dev_state = btrfsic_dev_state_lookup(bio->bi_bdev);
2932 	if (NULL != dev_state &&
2933 	    (rw & WRITE) && NULL != bio->bi_io_vec) {
2934 		unsigned int i;
2935 		u64 dev_bytenr;
2936 		u64 cur_bytenr;
2937 		int bio_is_patched;
2938 		char **mapped_datav;
2939 
2940 		dev_bytenr = 512 * bio->bi_iter.bi_sector;
2941 		bio_is_patched = 0;
2942 		if (dev_state->state->print_mask &
2943 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2944 			printk(KERN_INFO
2945 			       "submit_bio(rw=0x%x, bi_vcnt=%u,"
2946 			       " bi_sector=%llu (bytenr %llu), bi_bdev=%p)\n",
2947 			       rw, bio->bi_vcnt,
2948 			       (unsigned long long)bio->bi_iter.bi_sector,
2949 			       dev_bytenr, bio->bi_bdev);
2950 
2951 		mapped_datav = kmalloc_array(bio->bi_vcnt,
2952 					     sizeof(*mapped_datav), GFP_NOFS);
2953 		if (!mapped_datav)
2954 			goto leave;
2955 		cur_bytenr = dev_bytenr;
2956 		for (i = 0; i < bio->bi_vcnt; i++) {
2957 			BUG_ON(bio->bi_io_vec[i].bv_len != PAGE_SIZE);
2958 			mapped_datav[i] = kmap(bio->bi_io_vec[i].bv_page);
2959 			if (!mapped_datav[i]) {
2960 				while (i > 0) {
2961 					i--;
2962 					kunmap(bio->bi_io_vec[i].bv_page);
2963 				}
2964 				kfree(mapped_datav);
2965 				goto leave;
2966 			}
2967 			if (dev_state->state->print_mask &
2968 			    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
2969 				printk(KERN_INFO
2970 				       "#%u: bytenr=%llu, len=%u, offset=%u\n",
2971 				       i, cur_bytenr, bio->bi_io_vec[i].bv_len,
2972 				       bio->bi_io_vec[i].bv_offset);
2973 			cur_bytenr += bio->bi_io_vec[i].bv_len;
2974 		}
2975 		btrfsic_process_written_block(dev_state, dev_bytenr,
2976 					      mapped_datav, bio->bi_vcnt,
2977 					      bio, &bio_is_patched,
2978 					      NULL, rw);
2979 		while (i > 0) {
2980 			i--;
2981 			kunmap(bio->bi_io_vec[i].bv_page);
2982 		}
2983 		kfree(mapped_datav);
2984 	} else if (NULL != dev_state && (rw & REQ_FLUSH)) {
2985 		if (dev_state->state->print_mask &
2986 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2987 			printk(KERN_INFO
2988 			       "submit_bio(rw=0x%x FLUSH, bdev=%p)\n",
2989 			       rw, bio->bi_bdev);
2990 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2991 			if ((dev_state->state->print_mask &
2992 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2993 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2994 				printk(KERN_INFO
2995 				       "btrfsic_submit_bio(%s) with FLUSH"
2996 				       " but dummy block already in use"
2997 				       " (ignored)!\n",
2998 				       dev_state->name);
2999 		} else {
3000 			struct btrfsic_block *const block =
3001 				&dev_state->dummy_block_for_bio_bh_flush;
3002 
3003 			block->is_iodone = 0;
3004 			block->never_written = 0;
3005 			block->iodone_w_error = 0;
3006 			block->flush_gen = dev_state->last_flush_gen + 1;
3007 			block->submit_bio_bh_rw = rw;
3008 			block->orig_bio_bh_private = bio->bi_private;
3009 			block->orig_bio_bh_end_io.bio = bio->bi_end_io;
3010 			block->next_in_same_bio = NULL;
3011 			bio->bi_private = block;
3012 			bio->bi_end_io = btrfsic_bio_end_io;
3013 		}
3014 	}
3015 leave:
3016 	mutex_unlock(&btrfsic_mutex);
3017 }
3018 
3019 void btrfsic_submit_bio(int rw, struct bio *bio)
3020 {
3021 	__btrfsic_submit_bio(rw, bio);
3022 	submit_bio(rw, bio);
3023 }
3024 
3025 int btrfsic_submit_bio_wait(int rw, struct bio *bio)
3026 {
3027 	__btrfsic_submit_bio(rw, bio);
3028 	return submit_bio_wait(rw, bio);
3029 }
3030 
3031 int btrfsic_mount(struct btrfs_root *root,
3032 		  struct btrfs_fs_devices *fs_devices,
3033 		  int including_extent_data, u32 print_mask)
3034 {
3035 	int ret;
3036 	struct btrfsic_state *state;
3037 	struct list_head *dev_head = &fs_devices->devices;
3038 	struct btrfs_device *device;
3039 
3040 	if (root->nodesize & ((u64)PAGE_SIZE - 1)) {
3041 		printk(KERN_INFO
3042 		       "btrfsic: cannot handle nodesize %d not being a multiple of PAGE_SIZE %ld!\n",
3043 		       root->nodesize, PAGE_SIZE);
3044 		return -1;
3045 	}
3046 	if (root->sectorsize & ((u64)PAGE_SIZE - 1)) {
3047 		printk(KERN_INFO
3048 		       "btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_SIZE %ld!\n",
3049 		       root->sectorsize, PAGE_SIZE);
3050 		return -1;
3051 	}
3052 	state = kzalloc(sizeof(*state), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
3053 	if (!state) {
3054 		state = vzalloc(sizeof(*state));
3055 		if (!state) {
3056 			printk(KERN_INFO "btrfs check-integrity: vzalloc() failed!\n");
3057 			return -1;
3058 		}
3059 	}
3060 
3061 	if (!btrfsic_is_initialized) {
3062 		mutex_init(&btrfsic_mutex);
3063 		btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
3064 		btrfsic_is_initialized = 1;
3065 	}
3066 	mutex_lock(&btrfsic_mutex);
3067 	state->root = root;
3068 	state->print_mask = print_mask;
3069 	state->include_extent_data = including_extent_data;
3070 	state->csum_size = 0;
3071 	state->metablock_size = root->nodesize;
3072 	state->datablock_size = root->sectorsize;
3073 	INIT_LIST_HEAD(&state->all_blocks_list);
3074 	btrfsic_block_hashtable_init(&state->block_hashtable);
3075 	btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
3076 	state->max_superblock_generation = 0;
3077 	state->latest_superblock = NULL;
3078 
3079 	list_for_each_entry(device, dev_head, dev_list) {
3080 		struct btrfsic_dev_state *ds;
3081 		const char *p;
3082 
3083 		if (!device->bdev || !device->name)
3084 			continue;
3085 
3086 		ds = btrfsic_dev_state_alloc();
3087 		if (NULL == ds) {
3088 			printk(KERN_INFO
3089 			       "btrfs check-integrity: kmalloc() failed!\n");
3090 			mutex_unlock(&btrfsic_mutex);
3091 			return -1;
3092 		}
3093 		ds->bdev = device->bdev;
3094 		ds->state = state;
3095 		bdevname(ds->bdev, ds->name);
3096 		ds->name[BDEVNAME_SIZE - 1] = '\0';
3097 		p = kbasename(ds->name);
3098 		strlcpy(ds->name, p, sizeof(ds->name));
3099 		btrfsic_dev_state_hashtable_add(ds,
3100 						&btrfsic_dev_state_hashtable);
3101 	}
3102 
3103 	ret = btrfsic_process_superblock(state, fs_devices);
3104 	if (0 != ret) {
3105 		mutex_unlock(&btrfsic_mutex);
3106 		btrfsic_unmount(root, fs_devices);
3107 		return ret;
3108 	}
3109 
3110 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
3111 		btrfsic_dump_database(state);
3112 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
3113 		btrfsic_dump_tree(state);
3114 
3115 	mutex_unlock(&btrfsic_mutex);
3116 	return 0;
3117 }
3118 
3119 void btrfsic_unmount(struct btrfs_root *root,
3120 		     struct btrfs_fs_devices *fs_devices)
3121 {
3122 	struct btrfsic_block *b_all, *tmp_all;
3123 	struct btrfsic_state *state;
3124 	struct list_head *dev_head = &fs_devices->devices;
3125 	struct btrfs_device *device;
3126 
3127 	if (!btrfsic_is_initialized)
3128 		return;
3129 
3130 	mutex_lock(&btrfsic_mutex);
3131 
3132 	state = NULL;
3133 	list_for_each_entry(device, dev_head, dev_list) {
3134 		struct btrfsic_dev_state *ds;
3135 
3136 		if (!device->bdev || !device->name)
3137 			continue;
3138 
3139 		ds = btrfsic_dev_state_hashtable_lookup(
3140 				device->bdev,
3141 				&btrfsic_dev_state_hashtable);
3142 		if (NULL != ds) {
3143 			state = ds->state;
3144 			btrfsic_dev_state_hashtable_remove(ds);
3145 			btrfsic_dev_state_free(ds);
3146 		}
3147 	}
3148 
3149 	if (NULL == state) {
3150 		printk(KERN_INFO
3151 		       "btrfsic: error, cannot find state information"
3152 		       " on umount!\n");
3153 		mutex_unlock(&btrfsic_mutex);
3154 		return;
3155 	}
3156 
3157 	/*
3158 	 * Don't care about keeping the lists' state up to date,
3159 	 * just free all memory that was allocated dynamically.
3160 	 * Free the blocks and the block_links.
3161 	 */
3162 	list_for_each_entry_safe(b_all, tmp_all, &state->all_blocks_list,
3163 				 all_blocks_node) {
3164 		struct btrfsic_block_link *l, *tmp;
3165 
3166 		list_for_each_entry_safe(l, tmp, &b_all->ref_to_list,
3167 					 node_ref_to) {
3168 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
3169 				btrfsic_print_rem_link(state, l);
3170 
3171 			l->ref_cnt--;
3172 			if (0 == l->ref_cnt)
3173 				btrfsic_block_link_free(l);
3174 		}
3175 
3176 		if (b_all->is_iodone || b_all->never_written)
3177 			btrfsic_block_free(b_all);
3178 		else
3179 			printk(KERN_INFO "btrfs: attempt to free %c-block"
3180 			       " @%llu (%s/%llu/%d) on umount which is"
3181 			       " not yet iodone!\n",
3182 			       btrfsic_get_block_type(state, b_all),
3183 			       b_all->logical_bytenr, b_all->dev_state->name,
3184 			       b_all->dev_bytenr, b_all->mirror_num);
3185 	}
3186 
3187 	mutex_unlock(&btrfsic_mutex);
3188 
3189 	kvfree(state);
3190 }
3191