check-integrity.c revision 8d78eb16638b4e9e57225fa32b0f6595ecbf4424
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
82#include <linux/sched.h>
83#include <linux/slab.h>
84#include <linux/buffer_head.h>
85#include <linux/mutex.h>
86#include <linux/crc32c.h>
87#include <linux/genhd.h>
88#include <linux/blkdev.h>
89#include "ctree.h"
90#include "disk-io.h"
91#include "transaction.h"
92#include "extent_io.h"
93#include "volumes.h"
94#include "print-tree.h"
95#include "locking.h"
96#include "check-integrity.h"
97#include "rcu-string.h"
98
99#define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
100#define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
101#define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
102#define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
103#define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
104#define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
105#define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
106#define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6)	/* in characters,
107							 * excluding " [...]" */
108#define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
109
110/*
111 * The definition of the bitmask fields for the print_mask.
112 * They are specified with the mount option check_integrity_print_mask.
113 */
114#define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE			0x00000001
115#define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION		0x00000002
116#define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE			0x00000004
117#define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE			0x00000008
118#define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH			0x00000010
119#define BTRFSIC_PRINT_MASK_END_IO_BIO_BH			0x00000020
120#define BTRFSIC_PRINT_MASK_VERBOSE				0x00000040
121#define BTRFSIC_PRINT_MASK_VERY_VERBOSE				0x00000080
122#define BTRFSIC_PRINT_MASK_INITIAL_TREE				0x00000100
123#define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES			0x00000200
124#define BTRFSIC_PRINT_MASK_INITIAL_DATABASE			0x00000400
125#define BTRFSIC_PRINT_MASK_NUM_COPIES				0x00000800
126#define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS		0x00001000
127
128struct btrfsic_dev_state;
129struct btrfsic_state;
130
131struct btrfsic_block {
132	u32 magic_num;		/* only used for debug purposes */
133	unsigned int is_metadata:1;	/* if it is meta-data, not data-data */
134	unsigned int is_superblock:1;	/* if it is one of the superblocks */
135	unsigned int is_iodone:1;	/* if is done by lower subsystem */
136	unsigned int iodone_w_error:1;	/* error was indicated to endio */
137	unsigned int never_written:1;	/* block was added because it was
138					 * referenced, not because it was
139					 * written */
140	unsigned int mirror_num;	/* large enough to hold
141					 * BTRFS_SUPER_MIRROR_MAX */
142	struct btrfsic_dev_state *dev_state;
143	u64 dev_bytenr;		/* key, physical byte num on disk */
144	u64 logical_bytenr;	/* logical byte num on disk */
145	u64 generation;
146	struct btrfs_disk_key disk_key;	/* extra info to print in case of
147					 * issues, will not always be correct */
148	struct list_head collision_resolving_node;	/* list node */
149	struct list_head all_blocks_node;	/* list node */
150
151	/* the following two lists contain block_link items */
152	struct list_head ref_to_list;	/* list */
153	struct list_head ref_from_list;	/* list */
154	struct btrfsic_block *next_in_same_bio;
155	void *orig_bio_bh_private;
156	union {
157		bio_end_io_t *bio;
158		bh_end_io_t *bh;
159	} orig_bio_bh_end_io;
160	int submit_bio_bh_rw;
161	u64 flush_gen; /* only valid if !never_written */
162};
163
164/*
165 * Elements of this type are allocated dynamically and required because
166 * each block object can refer to and can be ref from multiple blocks.
167 * The key to lookup them in the hashtable is the dev_bytenr of
168 * the block ref to plus the one from the block refered from.
169 * The fact that they are searchable via a hashtable and that a
170 * ref_cnt is maintained is not required for the btrfs integrity
171 * check algorithm itself, it is only used to make the output more
172 * beautiful in case that an error is detected (an error is defined
173 * as a write operation to a block while that block is still referenced).
174 */
175struct btrfsic_block_link {
176	u32 magic_num;		/* only used for debug purposes */
177	u32 ref_cnt;
178	struct list_head node_ref_to;	/* list node */
179	struct list_head node_ref_from;	/* list node */
180	struct list_head collision_resolving_node;	/* list node */
181	struct btrfsic_block *block_ref_to;
182	struct btrfsic_block *block_ref_from;
183	u64 parent_generation;
184};
185
186struct btrfsic_dev_state {
187	u32 magic_num;		/* only used for debug purposes */
188	struct block_device *bdev;
189	struct btrfsic_state *state;
190	struct list_head collision_resolving_node;	/* list node */
191	struct btrfsic_block dummy_block_for_bio_bh_flush;
192	u64 last_flush_gen;
193	char name[BDEVNAME_SIZE];
194};
195
196struct btrfsic_block_hashtable {
197	struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
198};
199
200struct btrfsic_block_link_hashtable {
201	struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
202};
203
204struct btrfsic_dev_state_hashtable {
205	struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
206};
207
208struct btrfsic_block_data_ctx {
209	u64 start;		/* virtual bytenr */
210	u64 dev_bytenr;		/* physical bytenr on device */
211	u32 len;
212	struct btrfsic_dev_state *dev;
213	char **datav;
214	struct page **pagev;
215	void *mem_to_free;
216};
217
218/* This structure is used to implement recursion without occupying
219 * any stack space, refer to btrfsic_process_metablock() */
220struct btrfsic_stack_frame {
221	u32 magic;
222	u32 nr;
223	int error;
224	int i;
225	int limit_nesting;
226	int num_copies;
227	int mirror_num;
228	struct btrfsic_block *block;
229	struct btrfsic_block_data_ctx *block_ctx;
230	struct btrfsic_block *next_block;
231	struct btrfsic_block_data_ctx next_block_ctx;
232	struct btrfs_header *hdr;
233	struct btrfsic_stack_frame *prev;
234};
235
236/* Some state per mounted filesystem */
237struct btrfsic_state {
238	u32 print_mask;
239	int include_extent_data;
240	int csum_size;
241	struct list_head all_blocks_list;
242	struct btrfsic_block_hashtable block_hashtable;
243	struct btrfsic_block_link_hashtable block_link_hashtable;
244	struct btrfs_root *root;
245	u64 max_superblock_generation;
246	struct btrfsic_block *latest_superblock;
247	u32 metablock_size;
248	u32 datablock_size;
249};
250
251static void btrfsic_block_init(struct btrfsic_block *b);
252static struct btrfsic_block *btrfsic_block_alloc(void);
253static void btrfsic_block_free(struct btrfsic_block *b);
254static void btrfsic_block_link_init(struct btrfsic_block_link *n);
255static struct btrfsic_block_link *btrfsic_block_link_alloc(void);
256static void btrfsic_block_link_free(struct btrfsic_block_link *n);
257static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds);
258static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void);
259static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds);
260static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h);
261static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
262					struct btrfsic_block_hashtable *h);
263static void btrfsic_block_hashtable_remove(struct btrfsic_block *b);
264static struct btrfsic_block *btrfsic_block_hashtable_lookup(
265		struct block_device *bdev,
266		u64 dev_bytenr,
267		struct btrfsic_block_hashtable *h);
268static void btrfsic_block_link_hashtable_init(
269		struct btrfsic_block_link_hashtable *h);
270static void btrfsic_block_link_hashtable_add(
271		struct btrfsic_block_link *l,
272		struct btrfsic_block_link_hashtable *h);
273static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l);
274static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
275		struct block_device *bdev_ref_to,
276		u64 dev_bytenr_ref_to,
277		struct block_device *bdev_ref_from,
278		u64 dev_bytenr_ref_from,
279		struct btrfsic_block_link_hashtable *h);
280static void btrfsic_dev_state_hashtable_init(
281		struct btrfsic_dev_state_hashtable *h);
282static void btrfsic_dev_state_hashtable_add(
283		struct btrfsic_dev_state *ds,
284		struct btrfsic_dev_state_hashtable *h);
285static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds);
286static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(
287		struct block_device *bdev,
288		struct btrfsic_dev_state_hashtable *h);
289static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void);
290static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf);
291static int btrfsic_process_superblock(struct btrfsic_state *state,
292				      struct btrfs_fs_devices *fs_devices);
293static int btrfsic_process_metablock(struct btrfsic_state *state,
294				     struct btrfsic_block *block,
295				     struct btrfsic_block_data_ctx *block_ctx,
296				     int limit_nesting, int force_iodone_flag);
297static void btrfsic_read_from_block_data(
298	struct btrfsic_block_data_ctx *block_ctx,
299	void *dst, u32 offset, size_t len);
300static int btrfsic_create_link_to_next_block(
301		struct btrfsic_state *state,
302		struct btrfsic_block *block,
303		struct btrfsic_block_data_ctx
304		*block_ctx, u64 next_bytenr,
305		int limit_nesting,
306		struct btrfsic_block_data_ctx *next_block_ctx,
307		struct btrfsic_block **next_blockp,
308		int force_iodone_flag,
309		int *num_copiesp, int *mirror_nump,
310		struct btrfs_disk_key *disk_key,
311		u64 parent_generation);
312static int btrfsic_handle_extent_data(struct btrfsic_state *state,
313				      struct btrfsic_block *block,
314				      struct btrfsic_block_data_ctx *block_ctx,
315				      u32 item_offset, int force_iodone_flag);
316static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
317			     struct btrfsic_block_data_ctx *block_ctx_out,
318			     int mirror_num);
319static int btrfsic_map_superblock(struct btrfsic_state *state, u64 bytenr,
320				  u32 len, struct block_device *bdev,
321				  struct btrfsic_block_data_ctx *block_ctx_out);
322static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
323static int btrfsic_read_block(struct btrfsic_state *state,
324			      struct btrfsic_block_data_ctx *block_ctx);
325static void btrfsic_dump_database(struct btrfsic_state *state);
326static void btrfsic_complete_bio_end_io(struct bio *bio, int err);
327static int btrfsic_test_for_metadata(struct btrfsic_state *state,
328				     char **datav, unsigned int num_pages);
329static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
330					  u64 dev_bytenr, char **mapped_datav,
331					  unsigned int num_pages,
332					  struct bio *bio, int *bio_is_patched,
333					  struct buffer_head *bh,
334					  int submit_bio_bh_rw);
335static int btrfsic_process_written_superblock(
336		struct btrfsic_state *state,
337		struct btrfsic_block *const block,
338		struct btrfs_super_block *const super_hdr);
339static void btrfsic_bio_end_io(struct bio *bp, int bio_error_status);
340static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate);
341static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
342					      const struct btrfsic_block *block,
343					      int recursion_level);
344static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
345					struct btrfsic_block *const block,
346					int recursion_level);
347static void btrfsic_print_add_link(const struct btrfsic_state *state,
348				   const struct btrfsic_block_link *l);
349static void btrfsic_print_rem_link(const struct btrfsic_state *state,
350				   const struct btrfsic_block_link *l);
351static char btrfsic_get_block_type(const struct btrfsic_state *state,
352				   const struct btrfsic_block *block);
353static void btrfsic_dump_tree(const struct btrfsic_state *state);
354static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
355				  const struct btrfsic_block *block,
356				  int indent_level);
357static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
358		struct btrfsic_state *state,
359		struct btrfsic_block_data_ctx *next_block_ctx,
360		struct btrfsic_block *next_block,
361		struct btrfsic_block *from_block,
362		u64 parent_generation);
363static struct btrfsic_block *btrfsic_block_lookup_or_add(
364		struct btrfsic_state *state,
365		struct btrfsic_block_data_ctx *block_ctx,
366		const char *additional_string,
367		int is_metadata,
368		int is_iodone,
369		int never_written,
370		int mirror_num,
371		int *was_created);
372static int btrfsic_process_superblock_dev_mirror(
373		struct btrfsic_state *state,
374		struct btrfsic_dev_state *dev_state,
375		struct btrfs_device *device,
376		int superblock_mirror_num,
377		struct btrfsic_dev_state **selected_dev_state,
378		struct btrfs_super_block *selected_super);
379static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
380		struct block_device *bdev);
381static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
382					   u64 bytenr,
383					   struct btrfsic_dev_state *dev_state,
384					   u64 dev_bytenr);
385
386static struct mutex btrfsic_mutex;
387static int btrfsic_is_initialized;
388static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
389
390
391static void btrfsic_block_init(struct btrfsic_block *b)
392{
393	b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
394	b->dev_state = NULL;
395	b->dev_bytenr = 0;
396	b->logical_bytenr = 0;
397	b->generation = BTRFSIC_GENERATION_UNKNOWN;
398	b->disk_key.objectid = 0;
399	b->disk_key.type = 0;
400	b->disk_key.offset = 0;
401	b->is_metadata = 0;
402	b->is_superblock = 0;
403	b->is_iodone = 0;
404	b->iodone_w_error = 0;
405	b->never_written = 0;
406	b->mirror_num = 0;
407	b->next_in_same_bio = NULL;
408	b->orig_bio_bh_private = NULL;
409	b->orig_bio_bh_end_io.bio = NULL;
410	INIT_LIST_HEAD(&b->collision_resolving_node);
411	INIT_LIST_HEAD(&b->all_blocks_node);
412	INIT_LIST_HEAD(&b->ref_to_list);
413	INIT_LIST_HEAD(&b->ref_from_list);
414	b->submit_bio_bh_rw = 0;
415	b->flush_gen = 0;
416}
417
418static struct btrfsic_block *btrfsic_block_alloc(void)
419{
420	struct btrfsic_block *b;
421
422	b = kzalloc(sizeof(*b), GFP_NOFS);
423	if (NULL != b)
424		btrfsic_block_init(b);
425
426	return b;
427}
428
429static void btrfsic_block_free(struct btrfsic_block *b)
430{
431	BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
432	kfree(b);
433}
434
435static void btrfsic_block_link_init(struct btrfsic_block_link *l)
436{
437	l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
438	l->ref_cnt = 1;
439	INIT_LIST_HEAD(&l->node_ref_to);
440	INIT_LIST_HEAD(&l->node_ref_from);
441	INIT_LIST_HEAD(&l->collision_resolving_node);
442	l->block_ref_to = NULL;
443	l->block_ref_from = NULL;
444}
445
446static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
447{
448	struct btrfsic_block_link *l;
449
450	l = kzalloc(sizeof(*l), GFP_NOFS);
451	if (NULL != l)
452		btrfsic_block_link_init(l);
453
454	return l;
455}
456
457static void btrfsic_block_link_free(struct btrfsic_block_link *l)
458{
459	BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
460	kfree(l);
461}
462
463static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
464{
465	ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
466	ds->bdev = NULL;
467	ds->state = NULL;
468	ds->name[0] = '\0';
469	INIT_LIST_HEAD(&ds->collision_resolving_node);
470	ds->last_flush_gen = 0;
471	btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
472	ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
473	ds->dummy_block_for_bio_bh_flush.dev_state = ds;
474}
475
476static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
477{
478	struct btrfsic_dev_state *ds;
479
480	ds = kzalloc(sizeof(*ds), GFP_NOFS);
481	if (NULL != ds)
482		btrfsic_dev_state_init(ds);
483
484	return ds;
485}
486
487static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
488{
489	BUG_ON(!(NULL == ds ||
490		 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
491	kfree(ds);
492}
493
494static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
495{
496	int i;
497
498	for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
499		INIT_LIST_HEAD(h->table + i);
500}
501
502static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
503					struct btrfsic_block_hashtable *h)
504{
505	const unsigned int hashval =
506	    (((unsigned int)(b->dev_bytenr >> 16)) ^
507	     ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
508	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
509
510	list_add(&b->collision_resolving_node, h->table + hashval);
511}
512
513static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
514{
515	list_del(&b->collision_resolving_node);
516}
517
518static struct btrfsic_block *btrfsic_block_hashtable_lookup(
519		struct block_device *bdev,
520		u64 dev_bytenr,
521		struct btrfsic_block_hashtable *h)
522{
523	const unsigned int hashval =
524	    (((unsigned int)(dev_bytenr >> 16)) ^
525	     ((unsigned int)((uintptr_t)bdev))) &
526	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
527	struct list_head *elem;
528
529	list_for_each(elem, h->table + hashval) {
530		struct btrfsic_block *const b =
531		    list_entry(elem, struct btrfsic_block,
532			       collision_resolving_node);
533
534		if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
535			return b;
536	}
537
538	return NULL;
539}
540
541static void btrfsic_block_link_hashtable_init(
542		struct btrfsic_block_link_hashtable *h)
543{
544	int i;
545
546	for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
547		INIT_LIST_HEAD(h->table + i);
548}
549
550static void btrfsic_block_link_hashtable_add(
551		struct btrfsic_block_link *l,
552		struct btrfsic_block_link_hashtable *h)
553{
554	const unsigned int hashval =
555	    (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
556	     ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
557	     ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
558	     ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
559	     & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
560
561	BUG_ON(NULL == l->block_ref_to);
562	BUG_ON(NULL == l->block_ref_from);
563	list_add(&l->collision_resolving_node, h->table + hashval);
564}
565
566static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
567{
568	list_del(&l->collision_resolving_node);
569}
570
571static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
572		struct block_device *bdev_ref_to,
573		u64 dev_bytenr_ref_to,
574		struct block_device *bdev_ref_from,
575		u64 dev_bytenr_ref_from,
576		struct btrfsic_block_link_hashtable *h)
577{
578	const unsigned int hashval =
579	    (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
580	     ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
581	     ((unsigned int)((uintptr_t)bdev_ref_to)) ^
582	     ((unsigned int)((uintptr_t)bdev_ref_from))) &
583	     (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
584	struct list_head *elem;
585
586	list_for_each(elem, h->table + hashval) {
587		struct btrfsic_block_link *const l =
588		    list_entry(elem, struct btrfsic_block_link,
589			       collision_resolving_node);
590
591		BUG_ON(NULL == l->block_ref_to);
592		BUG_ON(NULL == l->block_ref_from);
593		if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
594		    l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
595		    l->block_ref_from->dev_state->bdev == bdev_ref_from &&
596		    l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
597			return l;
598	}
599
600	return NULL;
601}
602
603static void btrfsic_dev_state_hashtable_init(
604		struct btrfsic_dev_state_hashtable *h)
605{
606	int i;
607
608	for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
609		INIT_LIST_HEAD(h->table + i);
610}
611
612static void btrfsic_dev_state_hashtable_add(
613		struct btrfsic_dev_state *ds,
614		struct btrfsic_dev_state_hashtable *h)
615{
616	const unsigned int hashval =
617	    (((unsigned int)((uintptr_t)ds->bdev)) &
618	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
619
620	list_add(&ds->collision_resolving_node, h->table + hashval);
621}
622
623static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
624{
625	list_del(&ds->collision_resolving_node);
626}
627
628static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(
629		struct block_device *bdev,
630		struct btrfsic_dev_state_hashtable *h)
631{
632	const unsigned int hashval =
633	    (((unsigned int)((uintptr_t)bdev)) &
634	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
635	struct list_head *elem;
636
637	list_for_each(elem, h->table + hashval) {
638		struct btrfsic_dev_state *const ds =
639		    list_entry(elem, struct btrfsic_dev_state,
640			       collision_resolving_node);
641
642		if (ds->bdev == bdev)
643			return ds;
644	}
645
646	return NULL;
647}
648
649static int btrfsic_process_superblock(struct btrfsic_state *state,
650				      struct btrfs_fs_devices *fs_devices)
651{
652	int ret = 0;
653	struct btrfs_super_block *selected_super;
654	struct list_head *dev_head = &fs_devices->devices;
655	struct btrfs_device *device;
656	struct btrfsic_dev_state *selected_dev_state = NULL;
657	int pass;
658
659	BUG_ON(NULL == state);
660	selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
661	if (NULL == selected_super) {
662		printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
663		return -1;
664	}
665
666	list_for_each_entry(device, dev_head, dev_list) {
667		int i;
668		struct btrfsic_dev_state *dev_state;
669
670		if (!device->bdev || !device->name)
671			continue;
672
673		dev_state = btrfsic_dev_state_lookup(device->bdev);
674		BUG_ON(NULL == dev_state);
675		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
676			ret = btrfsic_process_superblock_dev_mirror(
677					state, dev_state, device, i,
678					&selected_dev_state, selected_super);
679			if (0 != ret && 0 == i) {
680				kfree(selected_super);
681				return ret;
682			}
683		}
684	}
685
686	if (NULL == state->latest_superblock) {
687		printk(KERN_INFO "btrfsic: no superblock found!\n");
688		kfree(selected_super);
689		return -1;
690	}
691
692	state->csum_size = btrfs_super_csum_size(selected_super);
693
694	for (pass = 0; pass < 3; pass++) {
695		int num_copies;
696		int mirror_num;
697		u64 next_bytenr;
698
699		switch (pass) {
700		case 0:
701			next_bytenr = btrfs_super_root(selected_super);
702			if (state->print_mask &
703			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
704				printk(KERN_INFO "root@%llu\n", next_bytenr);
705			break;
706		case 1:
707			next_bytenr = btrfs_super_chunk_root(selected_super);
708			if (state->print_mask &
709			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
710				printk(KERN_INFO "chunk@%llu\n", next_bytenr);
711			break;
712		case 2:
713			next_bytenr = btrfs_super_log_root(selected_super);
714			if (0 == next_bytenr)
715				continue;
716			if (state->print_mask &
717			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
718				printk(KERN_INFO "log@%llu\n", next_bytenr);
719			break;
720		}
721
722		num_copies =
723		    btrfs_num_copies(state->root->fs_info,
724				     next_bytenr, state->metablock_size);
725		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
726			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
727			       next_bytenr, num_copies);
728
729		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
730			struct btrfsic_block *next_block;
731			struct btrfsic_block_data_ctx tmp_next_block_ctx;
732			struct btrfsic_block_link *l;
733
734			ret = btrfsic_map_block(state, next_bytenr,
735						state->metablock_size,
736						&tmp_next_block_ctx,
737						mirror_num);
738			if (ret) {
739				printk(KERN_INFO "btrfsic:"
740				       " btrfsic_map_block(root @%llu,"
741				       " mirror %d) failed!\n",
742				       next_bytenr, mirror_num);
743				kfree(selected_super);
744				return -1;
745			}
746
747			next_block = btrfsic_block_hashtable_lookup(
748					tmp_next_block_ctx.dev->bdev,
749					tmp_next_block_ctx.dev_bytenr,
750					&state->block_hashtable);
751			BUG_ON(NULL == next_block);
752
753			l = btrfsic_block_link_hashtable_lookup(
754					tmp_next_block_ctx.dev->bdev,
755					tmp_next_block_ctx.dev_bytenr,
756					state->latest_superblock->dev_state->
757					bdev,
758					state->latest_superblock->dev_bytenr,
759					&state->block_link_hashtable);
760			BUG_ON(NULL == l);
761
762			ret = btrfsic_read_block(state, &tmp_next_block_ctx);
763			if (ret < (int)PAGE_CACHE_SIZE) {
764				printk(KERN_INFO
765				       "btrfsic: read @logical %llu failed!\n",
766				       tmp_next_block_ctx.start);
767				btrfsic_release_block_ctx(&tmp_next_block_ctx);
768				kfree(selected_super);
769				return -1;
770			}
771
772			ret = btrfsic_process_metablock(state,
773							next_block,
774							&tmp_next_block_ctx,
775							BTRFS_MAX_LEVEL + 3, 1);
776			btrfsic_release_block_ctx(&tmp_next_block_ctx);
777		}
778	}
779
780	kfree(selected_super);
781	return ret;
782}
783
784static int btrfsic_process_superblock_dev_mirror(
785		struct btrfsic_state *state,
786		struct btrfsic_dev_state *dev_state,
787		struct btrfs_device *device,
788		int superblock_mirror_num,
789		struct btrfsic_dev_state **selected_dev_state,
790		struct btrfs_super_block *selected_super)
791{
792	struct btrfs_super_block *super_tmp;
793	u64 dev_bytenr;
794	struct buffer_head *bh;
795	struct btrfsic_block *superblock_tmp;
796	int pass;
797	struct block_device *const superblock_bdev = device->bdev;
798
799	/* super block bytenr is always the unmapped device bytenr */
800	dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
801	if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->total_bytes)
802		return -1;
803	bh = __bread(superblock_bdev, dev_bytenr / 4096,
804		     BTRFS_SUPER_INFO_SIZE);
805	if (NULL == bh)
806		return -1;
807	super_tmp = (struct btrfs_super_block *)
808	    (bh->b_data + (dev_bytenr & 4095));
809
810	if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
811	    btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
812	    memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
813	    btrfs_super_nodesize(super_tmp) != state->metablock_size ||
814	    btrfs_super_leafsize(super_tmp) != state->metablock_size ||
815	    btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
816		brelse(bh);
817		return 0;
818	}
819
820	superblock_tmp =
821	    btrfsic_block_hashtable_lookup(superblock_bdev,
822					   dev_bytenr,
823					   &state->block_hashtable);
824	if (NULL == superblock_tmp) {
825		superblock_tmp = btrfsic_block_alloc();
826		if (NULL == superblock_tmp) {
827			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
828			brelse(bh);
829			return -1;
830		}
831		/* for superblock, only the dev_bytenr makes sense */
832		superblock_tmp->dev_bytenr = dev_bytenr;
833		superblock_tmp->dev_state = dev_state;
834		superblock_tmp->logical_bytenr = dev_bytenr;
835		superblock_tmp->generation = btrfs_super_generation(super_tmp);
836		superblock_tmp->is_metadata = 1;
837		superblock_tmp->is_superblock = 1;
838		superblock_tmp->is_iodone = 1;
839		superblock_tmp->never_written = 0;
840		superblock_tmp->mirror_num = 1 + superblock_mirror_num;
841		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
842			printk_in_rcu(KERN_INFO "New initial S-block (bdev %p, %s)"
843				     " @%llu (%s/%llu/%d)\n",
844				     superblock_bdev,
845				     rcu_str_deref(device->name), dev_bytenr,
846				     dev_state->name, dev_bytenr,
847				     superblock_mirror_num);
848		list_add(&superblock_tmp->all_blocks_node,
849			 &state->all_blocks_list);
850		btrfsic_block_hashtable_add(superblock_tmp,
851					    &state->block_hashtable);
852	}
853
854	/* select the one with the highest generation field */
855	if (btrfs_super_generation(super_tmp) >
856	    state->max_superblock_generation ||
857	    0 == state->max_superblock_generation) {
858		memcpy(selected_super, super_tmp, sizeof(*selected_super));
859		*selected_dev_state = dev_state;
860		state->max_superblock_generation =
861		    btrfs_super_generation(super_tmp);
862		state->latest_superblock = superblock_tmp;
863	}
864
865	for (pass = 0; pass < 3; pass++) {
866		u64 next_bytenr;
867		int num_copies;
868		int mirror_num;
869		const char *additional_string = NULL;
870		struct btrfs_disk_key tmp_disk_key;
871
872		tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
873		tmp_disk_key.offset = 0;
874		switch (pass) {
875		case 0:
876			btrfs_set_disk_key_objectid(&tmp_disk_key,
877						    BTRFS_ROOT_TREE_OBJECTID);
878			additional_string = "initial root ";
879			next_bytenr = btrfs_super_root(super_tmp);
880			break;
881		case 1:
882			btrfs_set_disk_key_objectid(&tmp_disk_key,
883						    BTRFS_CHUNK_TREE_OBJECTID);
884			additional_string = "initial chunk ";
885			next_bytenr = btrfs_super_chunk_root(super_tmp);
886			break;
887		case 2:
888			btrfs_set_disk_key_objectid(&tmp_disk_key,
889						    BTRFS_TREE_LOG_OBJECTID);
890			additional_string = "initial log ";
891			next_bytenr = btrfs_super_log_root(super_tmp);
892			if (0 == next_bytenr)
893				continue;
894			break;
895		}
896
897		num_copies =
898		    btrfs_num_copies(state->root->fs_info,
899				     next_bytenr, state->metablock_size);
900		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
901			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
902			       next_bytenr, num_copies);
903		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
904			struct btrfsic_block *next_block;
905			struct btrfsic_block_data_ctx tmp_next_block_ctx;
906			struct btrfsic_block_link *l;
907
908			if (btrfsic_map_block(state, next_bytenr,
909					      state->metablock_size,
910					      &tmp_next_block_ctx,
911					      mirror_num)) {
912				printk(KERN_INFO "btrfsic: btrfsic_map_block("
913				       "bytenr @%llu, mirror %d) failed!\n",
914				       next_bytenr, mirror_num);
915				brelse(bh);
916				return -1;
917			}
918
919			next_block = btrfsic_block_lookup_or_add(
920					state, &tmp_next_block_ctx,
921					additional_string, 1, 1, 0,
922					mirror_num, NULL);
923			if (NULL == next_block) {
924				btrfsic_release_block_ctx(&tmp_next_block_ctx);
925				brelse(bh);
926				return -1;
927			}
928
929			next_block->disk_key = tmp_disk_key;
930			next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
931			l = btrfsic_block_link_lookup_or_add(
932					state, &tmp_next_block_ctx,
933					next_block, superblock_tmp,
934					BTRFSIC_GENERATION_UNKNOWN);
935			btrfsic_release_block_ctx(&tmp_next_block_ctx);
936			if (NULL == l) {
937				brelse(bh);
938				return -1;
939			}
940		}
941	}
942	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
943		btrfsic_dump_tree_sub(state, superblock_tmp, 0);
944
945	brelse(bh);
946	return 0;
947}
948
949static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
950{
951	struct btrfsic_stack_frame *sf;
952
953	sf = kzalloc(sizeof(*sf), GFP_NOFS);
954	if (NULL == sf)
955		printk(KERN_INFO "btrfsic: alloc memory failed!\n");
956	else
957		sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
958	return sf;
959}
960
961static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
962{
963	BUG_ON(!(NULL == sf ||
964		 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
965	kfree(sf);
966}
967
968static int btrfsic_process_metablock(
969		struct btrfsic_state *state,
970		struct btrfsic_block *const first_block,
971		struct btrfsic_block_data_ctx *const first_block_ctx,
972		int first_limit_nesting, int force_iodone_flag)
973{
974	struct btrfsic_stack_frame initial_stack_frame = { 0 };
975	struct btrfsic_stack_frame *sf;
976	struct btrfsic_stack_frame *next_stack;
977	struct btrfs_header *const first_hdr =
978		(struct btrfs_header *)first_block_ctx->datav[0];
979
980	BUG_ON(!first_hdr);
981	sf = &initial_stack_frame;
982	sf->error = 0;
983	sf->i = -1;
984	sf->limit_nesting = first_limit_nesting;
985	sf->block = first_block;
986	sf->block_ctx = first_block_ctx;
987	sf->next_block = NULL;
988	sf->hdr = first_hdr;
989	sf->prev = NULL;
990
991continue_with_new_stack_frame:
992	sf->block->generation = le64_to_cpu(sf->hdr->generation);
993	if (0 == sf->hdr->level) {
994		struct btrfs_leaf *const leafhdr =
995		    (struct btrfs_leaf *)sf->hdr;
996
997		if (-1 == sf->i) {
998			sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
999
1000			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1001				printk(KERN_INFO
1002				       "leaf %llu items %d generation %llu"
1003				       " owner %llu\n",
1004				       sf->block_ctx->start, sf->nr,
1005				       btrfs_stack_header_generation(
1006					       &leafhdr->header),
1007				       btrfs_stack_header_owner(
1008					       &leafhdr->header));
1009		}
1010
1011continue_with_current_leaf_stack_frame:
1012		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1013			sf->i++;
1014			sf->num_copies = 0;
1015		}
1016
1017		if (sf->i < sf->nr) {
1018			struct btrfs_item disk_item;
1019			u32 disk_item_offset =
1020				(uintptr_t)(leafhdr->items + sf->i) -
1021				(uintptr_t)leafhdr;
1022			struct btrfs_disk_key *disk_key;
1023			u8 type;
1024			u32 item_offset;
1025			u32 item_size;
1026
1027			if (disk_item_offset + sizeof(struct btrfs_item) >
1028			    sf->block_ctx->len) {
1029leaf_item_out_of_bounce_error:
1030				printk(KERN_INFO
1031				       "btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
1032				       sf->block_ctx->start,
1033				       sf->block_ctx->dev->name);
1034				goto one_stack_frame_backwards;
1035			}
1036			btrfsic_read_from_block_data(sf->block_ctx,
1037						     &disk_item,
1038						     disk_item_offset,
1039						     sizeof(struct btrfs_item));
1040			item_offset = btrfs_stack_item_offset(&disk_item);
1041			item_size = btrfs_stack_item_offset(&disk_item);
1042			disk_key = &disk_item.key;
1043			type = btrfs_disk_key_type(disk_key);
1044
1045			if (BTRFS_ROOT_ITEM_KEY == type) {
1046				struct btrfs_root_item root_item;
1047				u32 root_item_offset;
1048				u64 next_bytenr;
1049
1050				root_item_offset = item_offset +
1051					offsetof(struct btrfs_leaf, items);
1052				if (root_item_offset + item_size >
1053				    sf->block_ctx->len)
1054					goto leaf_item_out_of_bounce_error;
1055				btrfsic_read_from_block_data(
1056					sf->block_ctx, &root_item,
1057					root_item_offset,
1058					item_size);
1059				next_bytenr = btrfs_root_bytenr(&root_item);
1060
1061				sf->error =
1062				    btrfsic_create_link_to_next_block(
1063						state,
1064						sf->block,
1065						sf->block_ctx,
1066						next_bytenr,
1067						sf->limit_nesting,
1068						&sf->next_block_ctx,
1069						&sf->next_block,
1070						force_iodone_flag,
1071						&sf->num_copies,
1072						&sf->mirror_num,
1073						disk_key,
1074						btrfs_root_generation(
1075						&root_item));
1076				if (sf->error)
1077					goto one_stack_frame_backwards;
1078
1079				if (NULL != sf->next_block) {
1080					struct btrfs_header *const next_hdr =
1081					    (struct btrfs_header *)
1082					    sf->next_block_ctx.datav[0];
1083
1084					next_stack =
1085					    btrfsic_stack_frame_alloc();
1086					if (NULL == next_stack) {
1087						btrfsic_release_block_ctx(
1088								&sf->
1089								next_block_ctx);
1090						goto one_stack_frame_backwards;
1091					}
1092
1093					next_stack->i = -1;
1094					next_stack->block = sf->next_block;
1095					next_stack->block_ctx =
1096					    &sf->next_block_ctx;
1097					next_stack->next_block = NULL;
1098					next_stack->hdr = next_hdr;
1099					next_stack->limit_nesting =
1100					    sf->limit_nesting - 1;
1101					next_stack->prev = sf;
1102					sf = next_stack;
1103					goto continue_with_new_stack_frame;
1104				}
1105			} else if (BTRFS_EXTENT_DATA_KEY == type &&
1106				   state->include_extent_data) {
1107				sf->error = btrfsic_handle_extent_data(
1108						state,
1109						sf->block,
1110						sf->block_ctx,
1111						item_offset,
1112						force_iodone_flag);
1113				if (sf->error)
1114					goto one_stack_frame_backwards;
1115			}
1116
1117			goto continue_with_current_leaf_stack_frame;
1118		}
1119	} else {
1120		struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1121
1122		if (-1 == sf->i) {
1123			sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1124
1125			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1126				printk(KERN_INFO "node %llu level %d items %d"
1127				       " generation %llu owner %llu\n",
1128				       sf->block_ctx->start,
1129				       nodehdr->header.level, sf->nr,
1130				       btrfs_stack_header_generation(
1131				       &nodehdr->header),
1132				       btrfs_stack_header_owner(
1133				       &nodehdr->header));
1134		}
1135
1136continue_with_current_node_stack_frame:
1137		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1138			sf->i++;
1139			sf->num_copies = 0;
1140		}
1141
1142		if (sf->i < sf->nr) {
1143			struct btrfs_key_ptr key_ptr;
1144			u32 key_ptr_offset;
1145			u64 next_bytenr;
1146
1147			key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1148					  (uintptr_t)nodehdr;
1149			if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1150			    sf->block_ctx->len) {
1151				printk(KERN_INFO
1152				       "btrfsic: node item out of bounce at logical %llu, dev %s\n",
1153				       sf->block_ctx->start,
1154				       sf->block_ctx->dev->name);
1155				goto one_stack_frame_backwards;
1156			}
1157			btrfsic_read_from_block_data(
1158				sf->block_ctx, &key_ptr, key_ptr_offset,
1159				sizeof(struct btrfs_key_ptr));
1160			next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1161
1162			sf->error = btrfsic_create_link_to_next_block(
1163					state,
1164					sf->block,
1165					sf->block_ctx,
1166					next_bytenr,
1167					sf->limit_nesting,
1168					&sf->next_block_ctx,
1169					&sf->next_block,
1170					force_iodone_flag,
1171					&sf->num_copies,
1172					&sf->mirror_num,
1173					&key_ptr.key,
1174					btrfs_stack_key_generation(&key_ptr));
1175			if (sf->error)
1176				goto one_stack_frame_backwards;
1177
1178			if (NULL != sf->next_block) {
1179				struct btrfs_header *const next_hdr =
1180				    (struct btrfs_header *)
1181				    sf->next_block_ctx.datav[0];
1182
1183				next_stack = btrfsic_stack_frame_alloc();
1184				if (NULL == next_stack)
1185					goto one_stack_frame_backwards;
1186
1187				next_stack->i = -1;
1188				next_stack->block = sf->next_block;
1189				next_stack->block_ctx = &sf->next_block_ctx;
1190				next_stack->next_block = NULL;
1191				next_stack->hdr = next_hdr;
1192				next_stack->limit_nesting =
1193				    sf->limit_nesting - 1;
1194				next_stack->prev = sf;
1195				sf = next_stack;
1196				goto continue_with_new_stack_frame;
1197			}
1198
1199			goto continue_with_current_node_stack_frame;
1200		}
1201	}
1202
1203one_stack_frame_backwards:
1204	if (NULL != sf->prev) {
1205		struct btrfsic_stack_frame *const prev = sf->prev;
1206
1207		/* the one for the initial block is freed in the caller */
1208		btrfsic_release_block_ctx(sf->block_ctx);
1209
1210		if (sf->error) {
1211			prev->error = sf->error;
1212			btrfsic_stack_frame_free(sf);
1213			sf = prev;
1214			goto one_stack_frame_backwards;
1215		}
1216
1217		btrfsic_stack_frame_free(sf);
1218		sf = prev;
1219		goto continue_with_new_stack_frame;
1220	} else {
1221		BUG_ON(&initial_stack_frame != sf);
1222	}
1223
1224	return sf->error;
1225}
1226
1227static void btrfsic_read_from_block_data(
1228	struct btrfsic_block_data_ctx *block_ctx,
1229	void *dstv, u32 offset, size_t len)
1230{
1231	size_t cur;
1232	size_t offset_in_page;
1233	char *kaddr;
1234	char *dst = (char *)dstv;
1235	size_t start_offset = block_ctx->start & ((u64)PAGE_CACHE_SIZE - 1);
1236	unsigned long i = (start_offset + offset) >> PAGE_CACHE_SHIFT;
1237
1238	WARN_ON(offset + len > block_ctx->len);
1239	offset_in_page = (start_offset + offset) & (PAGE_CACHE_SIZE - 1);
1240
1241	while (len > 0) {
1242		cur = min(len, ((size_t)PAGE_CACHE_SIZE - offset_in_page));
1243		BUG_ON(i >= (block_ctx->len + PAGE_CACHE_SIZE - 1) >>
1244			    PAGE_CACHE_SHIFT);
1245		kaddr = block_ctx->datav[i];
1246		memcpy(dst, kaddr + offset_in_page, cur);
1247
1248		dst += cur;
1249		len -= cur;
1250		offset_in_page = 0;
1251		i++;
1252	}
1253}
1254
1255static int btrfsic_create_link_to_next_block(
1256		struct btrfsic_state *state,
1257		struct btrfsic_block *block,
1258		struct btrfsic_block_data_ctx *block_ctx,
1259		u64 next_bytenr,
1260		int limit_nesting,
1261		struct btrfsic_block_data_ctx *next_block_ctx,
1262		struct btrfsic_block **next_blockp,
1263		int force_iodone_flag,
1264		int *num_copiesp, int *mirror_nump,
1265		struct btrfs_disk_key *disk_key,
1266		u64 parent_generation)
1267{
1268	struct btrfsic_block *next_block = NULL;
1269	int ret;
1270	struct btrfsic_block_link *l;
1271	int did_alloc_block_link;
1272	int block_was_created;
1273
1274	*next_blockp = NULL;
1275	if (0 == *num_copiesp) {
1276		*num_copiesp =
1277		    btrfs_num_copies(state->root->fs_info,
1278				     next_bytenr, state->metablock_size);
1279		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1280			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
1281			       next_bytenr, *num_copiesp);
1282		*mirror_nump = 1;
1283	}
1284
1285	if (*mirror_nump > *num_copiesp)
1286		return 0;
1287
1288	if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1289		printk(KERN_INFO
1290		       "btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1291		       *mirror_nump);
1292	ret = btrfsic_map_block(state, next_bytenr,
1293				state->metablock_size,
1294				next_block_ctx, *mirror_nump);
1295	if (ret) {
1296		printk(KERN_INFO
1297		       "btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1298		       next_bytenr, *mirror_nump);
1299		btrfsic_release_block_ctx(next_block_ctx);
1300		*next_blockp = NULL;
1301		return -1;
1302	}
1303
1304	next_block = btrfsic_block_lookup_or_add(state,
1305						 next_block_ctx, "referenced ",
1306						 1, force_iodone_flag,
1307						 !force_iodone_flag,
1308						 *mirror_nump,
1309						 &block_was_created);
1310	if (NULL == next_block) {
1311		btrfsic_release_block_ctx(next_block_ctx);
1312		*next_blockp = NULL;
1313		return -1;
1314	}
1315	if (block_was_created) {
1316		l = NULL;
1317		next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1318	} else {
1319		if (next_block->logical_bytenr != next_bytenr &&
1320		    !(!next_block->is_metadata &&
1321		      0 == next_block->logical_bytenr)) {
1322			printk(KERN_INFO
1323			       "Referenced block @%llu (%s/%llu/%d)"
1324			       " found in hash table, %c,"
1325			       " bytenr mismatch (!= stored %llu).\n",
1326			       next_bytenr, next_block_ctx->dev->name,
1327			       next_block_ctx->dev_bytenr, *mirror_nump,
1328			       btrfsic_get_block_type(state, next_block),
1329			       next_block->logical_bytenr);
1330		} else if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1331			printk(KERN_INFO
1332			       "Referenced block @%llu (%s/%llu/%d)"
1333			       " found in hash table, %c.\n",
1334			       next_bytenr, next_block_ctx->dev->name,
1335			       next_block_ctx->dev_bytenr, *mirror_nump,
1336			       btrfsic_get_block_type(state, next_block));
1337		next_block->logical_bytenr = next_bytenr;
1338
1339		next_block->mirror_num = *mirror_nump;
1340		l = btrfsic_block_link_hashtable_lookup(
1341				next_block_ctx->dev->bdev,
1342				next_block_ctx->dev_bytenr,
1343				block_ctx->dev->bdev,
1344				block_ctx->dev_bytenr,
1345				&state->block_link_hashtable);
1346	}
1347
1348	next_block->disk_key = *disk_key;
1349	if (NULL == l) {
1350		l = btrfsic_block_link_alloc();
1351		if (NULL == l) {
1352			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
1353			btrfsic_release_block_ctx(next_block_ctx);
1354			*next_blockp = NULL;
1355			return -1;
1356		}
1357
1358		did_alloc_block_link = 1;
1359		l->block_ref_to = next_block;
1360		l->block_ref_from = block;
1361		l->ref_cnt = 1;
1362		l->parent_generation = parent_generation;
1363
1364		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1365			btrfsic_print_add_link(state, l);
1366
1367		list_add(&l->node_ref_to, &block->ref_to_list);
1368		list_add(&l->node_ref_from, &next_block->ref_from_list);
1369
1370		btrfsic_block_link_hashtable_add(l,
1371						 &state->block_link_hashtable);
1372	} else {
1373		did_alloc_block_link = 0;
1374		if (0 == limit_nesting) {
1375			l->ref_cnt++;
1376			l->parent_generation = parent_generation;
1377			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1378				btrfsic_print_add_link(state, l);
1379		}
1380	}
1381
1382	if (limit_nesting > 0 && did_alloc_block_link) {
1383		ret = btrfsic_read_block(state, next_block_ctx);
1384		if (ret < (int)next_block_ctx->len) {
1385			printk(KERN_INFO
1386			       "btrfsic: read block @logical %llu failed!\n",
1387			       next_bytenr);
1388			btrfsic_release_block_ctx(next_block_ctx);
1389			*next_blockp = NULL;
1390			return -1;
1391		}
1392
1393		*next_blockp = next_block;
1394	} else {
1395		*next_blockp = NULL;
1396	}
1397	(*mirror_nump)++;
1398
1399	return 0;
1400}
1401
1402static int btrfsic_handle_extent_data(
1403		struct btrfsic_state *state,
1404		struct btrfsic_block *block,
1405		struct btrfsic_block_data_ctx *block_ctx,
1406		u32 item_offset, int force_iodone_flag)
1407{
1408	int ret;
1409	struct btrfs_file_extent_item file_extent_item;
1410	u64 file_extent_item_offset;
1411	u64 next_bytenr;
1412	u64 num_bytes;
1413	u64 generation;
1414	struct btrfsic_block_link *l;
1415
1416	file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1417				  item_offset;
1418	if (file_extent_item_offset +
1419	    offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1420	    block_ctx->len) {
1421		printk(KERN_INFO
1422		       "btrfsic: file item out of bounce at logical %llu, dev %s\n",
1423		       block_ctx->start, block_ctx->dev->name);
1424		return -1;
1425	}
1426
1427	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1428		file_extent_item_offset,
1429		offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1430	if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1431	    btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1432		if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1433			printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu\n",
1434			       file_extent_item.type,
1435			       btrfs_stack_file_extent_disk_bytenr(
1436			       &file_extent_item));
1437		return 0;
1438	}
1439
1440	if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1441	    block_ctx->len) {
1442		printk(KERN_INFO
1443		       "btrfsic: file item out of bounce at logical %llu, dev %s\n",
1444		       block_ctx->start, block_ctx->dev->name);
1445		return -1;
1446	}
1447	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1448				     file_extent_item_offset,
1449				     sizeof(struct btrfs_file_extent_item));
1450	next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item) +
1451		      btrfs_stack_file_extent_offset(&file_extent_item);
1452	generation = btrfs_stack_file_extent_generation(&file_extent_item);
1453	num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1454	generation = btrfs_stack_file_extent_generation(&file_extent_item);
1455
1456	if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1457		printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu,"
1458		       " offset = %llu, num_bytes = %llu\n",
1459		       file_extent_item.type,
1460		       btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1461		       btrfs_stack_file_extent_offset(&file_extent_item),
1462		       num_bytes);
1463	while (num_bytes > 0) {
1464		u32 chunk_len;
1465		int num_copies;
1466		int mirror_num;
1467
1468		if (num_bytes > state->datablock_size)
1469			chunk_len = state->datablock_size;
1470		else
1471			chunk_len = num_bytes;
1472
1473		num_copies =
1474		    btrfs_num_copies(state->root->fs_info,
1475				     next_bytenr, state->datablock_size);
1476		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1477			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
1478			       next_bytenr, num_copies);
1479		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1480			struct btrfsic_block_data_ctx next_block_ctx;
1481			struct btrfsic_block *next_block;
1482			int block_was_created;
1483
1484			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1485				printk(KERN_INFO "btrfsic_handle_extent_data("
1486				       "mirror_num=%d)\n", mirror_num);
1487			if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1488				printk(KERN_INFO
1489				       "\tdisk_bytenr = %llu, num_bytes %u\n",
1490				       next_bytenr, chunk_len);
1491			ret = btrfsic_map_block(state, next_bytenr,
1492						chunk_len, &next_block_ctx,
1493						mirror_num);
1494			if (ret) {
1495				printk(KERN_INFO
1496				       "btrfsic: btrfsic_map_block(@%llu,"
1497				       " mirror=%d) failed!\n",
1498				       next_bytenr, mirror_num);
1499				return -1;
1500			}
1501
1502			next_block = btrfsic_block_lookup_or_add(
1503					state,
1504					&next_block_ctx,
1505					"referenced ",
1506					0,
1507					force_iodone_flag,
1508					!force_iodone_flag,
1509					mirror_num,
1510					&block_was_created);
1511			if (NULL == next_block) {
1512				printk(KERN_INFO
1513				       "btrfsic: error, kmalloc failed!\n");
1514				btrfsic_release_block_ctx(&next_block_ctx);
1515				return -1;
1516			}
1517			if (!block_was_created) {
1518				if (next_block->logical_bytenr != next_bytenr &&
1519				    !(!next_block->is_metadata &&
1520				      0 == next_block->logical_bytenr)) {
1521					printk(KERN_INFO
1522					       "Referenced block"
1523					       " @%llu (%s/%llu/%d)"
1524					       " found in hash table, D,"
1525					       " bytenr mismatch"
1526					       " (!= stored %llu).\n",
1527					       next_bytenr,
1528					       next_block_ctx.dev->name,
1529					       next_block_ctx.dev_bytenr,
1530					       mirror_num,
1531					       next_block->logical_bytenr);
1532				}
1533				next_block->logical_bytenr = next_bytenr;
1534				next_block->mirror_num = mirror_num;
1535			}
1536
1537			l = btrfsic_block_link_lookup_or_add(state,
1538							     &next_block_ctx,
1539							     next_block, block,
1540							     generation);
1541			btrfsic_release_block_ctx(&next_block_ctx);
1542			if (NULL == l)
1543				return -1;
1544		}
1545
1546		next_bytenr += chunk_len;
1547		num_bytes -= chunk_len;
1548	}
1549
1550	return 0;
1551}
1552
1553static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1554			     struct btrfsic_block_data_ctx *block_ctx_out,
1555			     int mirror_num)
1556{
1557	int ret;
1558	u64 length;
1559	struct btrfs_bio *multi = NULL;
1560	struct btrfs_device *device;
1561
1562	length = len;
1563	ret = btrfs_map_block(state->root->fs_info, READ,
1564			      bytenr, &length, &multi, mirror_num);
1565
1566	if (ret) {
1567		block_ctx_out->start = 0;
1568		block_ctx_out->dev_bytenr = 0;
1569		block_ctx_out->len = 0;
1570		block_ctx_out->dev = NULL;
1571		block_ctx_out->datav = NULL;
1572		block_ctx_out->pagev = NULL;
1573		block_ctx_out->mem_to_free = NULL;
1574
1575		return ret;
1576	}
1577
1578	device = multi->stripes[0].dev;
1579	block_ctx_out->dev = btrfsic_dev_state_lookup(device->bdev);
1580	block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1581	block_ctx_out->start = bytenr;
1582	block_ctx_out->len = len;
1583	block_ctx_out->datav = NULL;
1584	block_ctx_out->pagev = NULL;
1585	block_ctx_out->mem_to_free = NULL;
1586
1587	kfree(multi);
1588	if (NULL == block_ctx_out->dev) {
1589		ret = -ENXIO;
1590		printk(KERN_INFO "btrfsic: error, cannot lookup dev (#1)!\n");
1591	}
1592
1593	return ret;
1594}
1595
1596static int btrfsic_map_superblock(struct btrfsic_state *state, u64 bytenr,
1597				  u32 len, struct block_device *bdev,
1598				  struct btrfsic_block_data_ctx *block_ctx_out)
1599{
1600	block_ctx_out->dev = btrfsic_dev_state_lookup(bdev);
1601	block_ctx_out->dev_bytenr = bytenr;
1602	block_ctx_out->start = bytenr;
1603	block_ctx_out->len = len;
1604	block_ctx_out->datav = NULL;
1605	block_ctx_out->pagev = NULL;
1606	block_ctx_out->mem_to_free = NULL;
1607	if (NULL != block_ctx_out->dev) {
1608		return 0;
1609	} else {
1610		printk(KERN_INFO "btrfsic: error, cannot lookup dev (#2)!\n");
1611		return -ENXIO;
1612	}
1613}
1614
1615static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1616{
1617	if (block_ctx->mem_to_free) {
1618		unsigned int num_pages;
1619
1620		BUG_ON(!block_ctx->datav);
1621		BUG_ON(!block_ctx->pagev);
1622		num_pages = (block_ctx->len + (u64)PAGE_CACHE_SIZE - 1) >>
1623			    PAGE_CACHE_SHIFT;
1624		while (num_pages > 0) {
1625			num_pages--;
1626			if (block_ctx->datav[num_pages]) {
1627				kunmap(block_ctx->pagev[num_pages]);
1628				block_ctx->datav[num_pages] = NULL;
1629			}
1630			if (block_ctx->pagev[num_pages]) {
1631				__free_page(block_ctx->pagev[num_pages]);
1632				block_ctx->pagev[num_pages] = NULL;
1633			}
1634		}
1635
1636		kfree(block_ctx->mem_to_free);
1637		block_ctx->mem_to_free = NULL;
1638		block_ctx->pagev = NULL;
1639		block_ctx->datav = NULL;
1640	}
1641}
1642
1643static int btrfsic_read_block(struct btrfsic_state *state,
1644			      struct btrfsic_block_data_ctx *block_ctx)
1645{
1646	unsigned int num_pages;
1647	unsigned int i;
1648	u64 dev_bytenr;
1649	int ret;
1650
1651	BUG_ON(block_ctx->datav);
1652	BUG_ON(block_ctx->pagev);
1653	BUG_ON(block_ctx->mem_to_free);
1654	if (block_ctx->dev_bytenr & ((u64)PAGE_CACHE_SIZE - 1)) {
1655		printk(KERN_INFO
1656		       "btrfsic: read_block() with unaligned bytenr %llu\n",
1657		       block_ctx->dev_bytenr);
1658		return -1;
1659	}
1660
1661	num_pages = (block_ctx->len + (u64)PAGE_CACHE_SIZE - 1) >>
1662		    PAGE_CACHE_SHIFT;
1663	block_ctx->mem_to_free = kzalloc((sizeof(*block_ctx->datav) +
1664					  sizeof(*block_ctx->pagev)) *
1665					 num_pages, GFP_NOFS);
1666	if (!block_ctx->mem_to_free)
1667		return -1;
1668	block_ctx->datav = block_ctx->mem_to_free;
1669	block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1670	for (i = 0; i < num_pages; i++) {
1671		block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1672		if (!block_ctx->pagev[i])
1673			return -1;
1674	}
1675
1676	dev_bytenr = block_ctx->dev_bytenr;
1677	for (i = 0; i < num_pages;) {
1678		struct bio *bio;
1679		unsigned int j;
1680		DECLARE_COMPLETION_ONSTACK(complete);
1681
1682		bio = btrfs_io_bio_alloc(GFP_NOFS, num_pages - i);
1683		if (!bio) {
1684			printk(KERN_INFO
1685			       "btrfsic: bio_alloc() for %u pages failed!\n",
1686			       num_pages - i);
1687			return -1;
1688		}
1689		bio->bi_bdev = block_ctx->dev->bdev;
1690		bio->bi_sector = dev_bytenr >> 9;
1691		bio->bi_end_io = btrfsic_complete_bio_end_io;
1692		bio->bi_private = &complete;
1693
1694		for (j = i; j < num_pages; j++) {
1695			ret = bio_add_page(bio, block_ctx->pagev[j],
1696					   PAGE_CACHE_SIZE, 0);
1697			if (PAGE_CACHE_SIZE != ret)
1698				break;
1699		}
1700		if (j == i) {
1701			printk(KERN_INFO
1702			       "btrfsic: error, failed to add a single page!\n");
1703			return -1;
1704		}
1705		submit_bio(READ, bio);
1706
1707		/* this will also unplug the queue */
1708		wait_for_completion(&complete);
1709
1710		if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1711			printk(KERN_INFO
1712			       "btrfsic: read error at logical %llu dev %s!\n",
1713			       block_ctx->start, block_ctx->dev->name);
1714			bio_put(bio);
1715			return -1;
1716		}
1717		bio_put(bio);
1718		dev_bytenr += (j - i) * PAGE_CACHE_SIZE;
1719		i = j;
1720	}
1721	for (i = 0; i < num_pages; i++) {
1722		block_ctx->datav[i] = kmap(block_ctx->pagev[i]);
1723		if (!block_ctx->datav[i]) {
1724			printk(KERN_INFO "btrfsic: kmap() failed (dev %s)!\n",
1725			       block_ctx->dev->name);
1726			return -1;
1727		}
1728	}
1729
1730	return block_ctx->len;
1731}
1732
1733static void btrfsic_complete_bio_end_io(struct bio *bio, int err)
1734{
1735	complete((struct completion *)bio->bi_private);
1736}
1737
1738static void btrfsic_dump_database(struct btrfsic_state *state)
1739{
1740	struct list_head *elem_all;
1741
1742	BUG_ON(NULL == state);
1743
1744	printk(KERN_INFO "all_blocks_list:\n");
1745	list_for_each(elem_all, &state->all_blocks_list) {
1746		const struct btrfsic_block *const b_all =
1747		    list_entry(elem_all, struct btrfsic_block,
1748			       all_blocks_node);
1749		struct list_head *elem_ref_to;
1750		struct list_head *elem_ref_from;
1751
1752		printk(KERN_INFO "%c-block @%llu (%s/%llu/%d)\n",
1753		       btrfsic_get_block_type(state, b_all),
1754		       b_all->logical_bytenr, b_all->dev_state->name,
1755		       b_all->dev_bytenr, b_all->mirror_num);
1756
1757		list_for_each(elem_ref_to, &b_all->ref_to_list) {
1758			const struct btrfsic_block_link *const l =
1759			    list_entry(elem_ref_to,
1760				       struct btrfsic_block_link,
1761				       node_ref_to);
1762
1763			printk(KERN_INFO " %c @%llu (%s/%llu/%d)"
1764			       " refers %u* to"
1765			       " %c @%llu (%s/%llu/%d)\n",
1766			       btrfsic_get_block_type(state, b_all),
1767			       b_all->logical_bytenr, b_all->dev_state->name,
1768			       b_all->dev_bytenr, b_all->mirror_num,
1769			       l->ref_cnt,
1770			       btrfsic_get_block_type(state, l->block_ref_to),
1771			       l->block_ref_to->logical_bytenr,
1772			       l->block_ref_to->dev_state->name,
1773			       l->block_ref_to->dev_bytenr,
1774			       l->block_ref_to->mirror_num);
1775		}
1776
1777		list_for_each(elem_ref_from, &b_all->ref_from_list) {
1778			const struct btrfsic_block_link *const l =
1779			    list_entry(elem_ref_from,
1780				       struct btrfsic_block_link,
1781				       node_ref_from);
1782
1783			printk(KERN_INFO " %c @%llu (%s/%llu/%d)"
1784			       " is ref %u* from"
1785			       " %c @%llu (%s/%llu/%d)\n",
1786			       btrfsic_get_block_type(state, b_all),
1787			       b_all->logical_bytenr, b_all->dev_state->name,
1788			       b_all->dev_bytenr, b_all->mirror_num,
1789			       l->ref_cnt,
1790			       btrfsic_get_block_type(state, l->block_ref_from),
1791			       l->block_ref_from->logical_bytenr,
1792			       l->block_ref_from->dev_state->name,
1793			       l->block_ref_from->dev_bytenr,
1794			       l->block_ref_from->mirror_num);
1795		}
1796
1797		printk(KERN_INFO "\n");
1798	}
1799}
1800
1801/*
1802 * Test whether the disk block contains a tree block (leaf or node)
1803 * (note that this test fails for the super block)
1804 */
1805static int btrfsic_test_for_metadata(struct btrfsic_state *state,
1806				     char **datav, unsigned int num_pages)
1807{
1808	struct btrfs_header *h;
1809	u8 csum[BTRFS_CSUM_SIZE];
1810	u32 crc = ~(u32)0;
1811	unsigned int i;
1812
1813	if (num_pages * PAGE_CACHE_SIZE < state->metablock_size)
1814		return 1; /* not metadata */
1815	num_pages = state->metablock_size >> PAGE_CACHE_SHIFT;
1816	h = (struct btrfs_header *)datav[0];
1817
1818	if (memcmp(h->fsid, state->root->fs_info->fsid, BTRFS_UUID_SIZE))
1819		return 1;
1820
1821	for (i = 0; i < num_pages; i++) {
1822		u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1823		size_t sublen = i ? PAGE_CACHE_SIZE :
1824				    (PAGE_CACHE_SIZE - BTRFS_CSUM_SIZE);
1825
1826		crc = crc32c(crc, data, sublen);
1827	}
1828	btrfs_csum_final(crc, csum);
1829	if (memcmp(csum, h->csum, state->csum_size))
1830		return 1;
1831
1832	return 0; /* is metadata */
1833}
1834
1835static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1836					  u64 dev_bytenr, char **mapped_datav,
1837					  unsigned int num_pages,
1838					  struct bio *bio, int *bio_is_patched,
1839					  struct buffer_head *bh,
1840					  int submit_bio_bh_rw)
1841{
1842	int is_metadata;
1843	struct btrfsic_block *block;
1844	struct btrfsic_block_data_ctx block_ctx;
1845	int ret;
1846	struct btrfsic_state *state = dev_state->state;
1847	struct block_device *bdev = dev_state->bdev;
1848	unsigned int processed_len;
1849
1850	if (NULL != bio_is_patched)
1851		*bio_is_patched = 0;
1852
1853again:
1854	if (num_pages == 0)
1855		return;
1856
1857	processed_len = 0;
1858	is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1859						      num_pages));
1860
1861	block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1862					       &state->block_hashtable);
1863	if (NULL != block) {
1864		u64 bytenr = 0;
1865		struct list_head *elem_ref_to;
1866		struct list_head *tmp_ref_to;
1867
1868		if (block->is_superblock) {
1869			bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1870						    mapped_datav[0]);
1871			if (num_pages * PAGE_CACHE_SIZE <
1872			    BTRFS_SUPER_INFO_SIZE) {
1873				printk(KERN_INFO
1874				       "btrfsic: cannot work with too short bios!\n");
1875				return;
1876			}
1877			is_metadata = 1;
1878			BUG_ON(BTRFS_SUPER_INFO_SIZE & (PAGE_CACHE_SIZE - 1));
1879			processed_len = BTRFS_SUPER_INFO_SIZE;
1880			if (state->print_mask &
1881			    BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1882				printk(KERN_INFO
1883				       "[before new superblock is written]:\n");
1884				btrfsic_dump_tree_sub(state, block, 0);
1885			}
1886		}
1887		if (is_metadata) {
1888			if (!block->is_superblock) {
1889				if (num_pages * PAGE_CACHE_SIZE <
1890				    state->metablock_size) {
1891					printk(KERN_INFO
1892					       "btrfsic: cannot work with too short bios!\n");
1893					return;
1894				}
1895				processed_len = state->metablock_size;
1896				bytenr = btrfs_stack_header_bytenr(
1897						(struct btrfs_header *)
1898						mapped_datav[0]);
1899				btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1900							       dev_state,
1901							       dev_bytenr);
1902			}
1903			if (block->logical_bytenr != bytenr) {
1904				printk(KERN_INFO
1905				       "Written block @%llu (%s/%llu/%d)"
1906				       " found in hash table, %c,"
1907				       " bytenr mismatch"
1908				       " (!= stored %llu).\n",
1909				       bytenr, dev_state->name, dev_bytenr,
1910				       block->mirror_num,
1911				       btrfsic_get_block_type(state, block),
1912				       block->logical_bytenr);
1913				block->logical_bytenr = bytenr;
1914			} else if (state->print_mask &
1915				   BTRFSIC_PRINT_MASK_VERBOSE)
1916				printk(KERN_INFO
1917				       "Written block @%llu (%s/%llu/%d)"
1918				       " found in hash table, %c.\n",
1919				       bytenr, dev_state->name, dev_bytenr,
1920				       block->mirror_num,
1921				       btrfsic_get_block_type(state, block));
1922		} else {
1923			if (num_pages * PAGE_CACHE_SIZE <
1924			    state->datablock_size) {
1925				printk(KERN_INFO
1926				       "btrfsic: cannot work with too short bios!\n");
1927				return;
1928			}
1929			processed_len = state->datablock_size;
1930			bytenr = block->logical_bytenr;
1931			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1932				printk(KERN_INFO
1933				       "Written block @%llu (%s/%llu/%d)"
1934				       " found in hash table, %c.\n",
1935				       bytenr, dev_state->name, dev_bytenr,
1936				       block->mirror_num,
1937				       btrfsic_get_block_type(state, block));
1938		}
1939
1940		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1941			printk(KERN_INFO
1942			       "ref_to_list: %cE, ref_from_list: %cE\n",
1943			       list_empty(&block->ref_to_list) ? ' ' : '!',
1944			       list_empty(&block->ref_from_list) ? ' ' : '!');
1945		if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1946			printk(KERN_INFO "btrfs: attempt to overwrite %c-block"
1947			       " @%llu (%s/%llu/%d), old(gen=%llu,"
1948			       " objectid=%llu, type=%d, offset=%llu),"
1949			       " new(gen=%llu),"
1950			       " which is referenced by most recent superblock"
1951			       " (superblockgen=%llu)!\n",
1952			       btrfsic_get_block_type(state, block), bytenr,
1953			       dev_state->name, dev_bytenr, block->mirror_num,
1954			       block->generation,
1955			       btrfs_disk_key_objectid(&block->disk_key),
1956			       block->disk_key.type,
1957			       btrfs_disk_key_offset(&block->disk_key),
1958			       btrfs_stack_header_generation(
1959				       (struct btrfs_header *) mapped_datav[0]),
1960			       state->max_superblock_generation);
1961			btrfsic_dump_tree(state);
1962		}
1963
1964		if (!block->is_iodone && !block->never_written) {
1965			printk(KERN_INFO "btrfs: attempt to overwrite %c-block"
1966			       " @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu,"
1967			       " which is not yet iodone!\n",
1968			       btrfsic_get_block_type(state, block), bytenr,
1969			       dev_state->name, dev_bytenr, block->mirror_num,
1970			       block->generation,
1971			       btrfs_stack_header_generation(
1972				       (struct btrfs_header *)
1973				       mapped_datav[0]));
1974			/* it would not be safe to go on */
1975			btrfsic_dump_tree(state);
1976			goto continue_loop;
1977		}
1978
1979		/*
1980		 * Clear all references of this block. Do not free
1981		 * the block itself even if is not referenced anymore
1982		 * because it still carries valueable information
1983		 * like whether it was ever written and IO completed.
1984		 */
1985		list_for_each_safe(elem_ref_to, tmp_ref_to,
1986				   &block->ref_to_list) {
1987			struct btrfsic_block_link *const l =
1988			    list_entry(elem_ref_to,
1989				       struct btrfsic_block_link,
1990				       node_ref_to);
1991
1992			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1993				btrfsic_print_rem_link(state, l);
1994			l->ref_cnt--;
1995			if (0 == l->ref_cnt) {
1996				list_del(&l->node_ref_to);
1997				list_del(&l->node_ref_from);
1998				btrfsic_block_link_hashtable_remove(l);
1999				btrfsic_block_link_free(l);
2000			}
2001		}
2002
2003		if (block->is_superblock)
2004			ret = btrfsic_map_superblock(state, bytenr,
2005						     processed_len,
2006						     bdev, &block_ctx);
2007		else
2008			ret = btrfsic_map_block(state, bytenr, processed_len,
2009						&block_ctx, 0);
2010		if (ret) {
2011			printk(KERN_INFO
2012			       "btrfsic: btrfsic_map_block(root @%llu)"
2013			       " failed!\n", bytenr);
2014			goto continue_loop;
2015		}
2016		block_ctx.datav = mapped_datav;
2017		/* the following is required in case of writes to mirrors,
2018		 * use the same that was used for the lookup */
2019		block_ctx.dev = dev_state;
2020		block_ctx.dev_bytenr = dev_bytenr;
2021
2022		if (is_metadata || state->include_extent_data) {
2023			block->never_written = 0;
2024			block->iodone_w_error = 0;
2025			if (NULL != bio) {
2026				block->is_iodone = 0;
2027				BUG_ON(NULL == bio_is_patched);
2028				if (!*bio_is_patched) {
2029					block->orig_bio_bh_private =
2030					    bio->bi_private;
2031					block->orig_bio_bh_end_io.bio =
2032					    bio->bi_end_io;
2033					block->next_in_same_bio = NULL;
2034					bio->bi_private = block;
2035					bio->bi_end_io = btrfsic_bio_end_io;
2036					*bio_is_patched = 1;
2037				} else {
2038					struct btrfsic_block *chained_block =
2039					    (struct btrfsic_block *)
2040					    bio->bi_private;
2041
2042					BUG_ON(NULL == chained_block);
2043					block->orig_bio_bh_private =
2044					    chained_block->orig_bio_bh_private;
2045					block->orig_bio_bh_end_io.bio =
2046					    chained_block->orig_bio_bh_end_io.
2047					    bio;
2048					block->next_in_same_bio = chained_block;
2049					bio->bi_private = block;
2050				}
2051			} else if (NULL != bh) {
2052				block->is_iodone = 0;
2053				block->orig_bio_bh_private = bh->b_private;
2054				block->orig_bio_bh_end_io.bh = bh->b_end_io;
2055				block->next_in_same_bio = NULL;
2056				bh->b_private = block;
2057				bh->b_end_io = btrfsic_bh_end_io;
2058			} else {
2059				block->is_iodone = 1;
2060				block->orig_bio_bh_private = NULL;
2061				block->orig_bio_bh_end_io.bio = NULL;
2062				block->next_in_same_bio = NULL;
2063			}
2064		}
2065
2066		block->flush_gen = dev_state->last_flush_gen + 1;
2067		block->submit_bio_bh_rw = submit_bio_bh_rw;
2068		if (is_metadata) {
2069			block->logical_bytenr = bytenr;
2070			block->is_metadata = 1;
2071			if (block->is_superblock) {
2072				BUG_ON(PAGE_CACHE_SIZE !=
2073				       BTRFS_SUPER_INFO_SIZE);
2074				ret = btrfsic_process_written_superblock(
2075						state,
2076						block,
2077						(struct btrfs_super_block *)
2078						mapped_datav[0]);
2079				if (state->print_mask &
2080				    BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
2081					printk(KERN_INFO
2082					"[after new superblock is written]:\n");
2083					btrfsic_dump_tree_sub(state, block, 0);
2084				}
2085			} else {
2086				block->mirror_num = 0;	/* unknown */
2087				ret = btrfsic_process_metablock(
2088						state,
2089						block,
2090						&block_ctx,
2091						0, 0);
2092			}
2093			if (ret)
2094				printk(KERN_INFO
2095				       "btrfsic: btrfsic_process_metablock"
2096				       "(root @%llu) failed!\n",
2097				       dev_bytenr);
2098		} else {
2099			block->is_metadata = 0;
2100			block->mirror_num = 0;	/* unknown */
2101			block->generation = BTRFSIC_GENERATION_UNKNOWN;
2102			if (!state->include_extent_data
2103			    && list_empty(&block->ref_from_list)) {
2104				/*
2105				 * disk block is overwritten with extent
2106				 * data (not meta data) and we are configured
2107				 * to not include extent data: take the
2108				 * chance and free the block's memory
2109				 */
2110				btrfsic_block_hashtable_remove(block);
2111				list_del(&block->all_blocks_node);
2112				btrfsic_block_free(block);
2113			}
2114		}
2115		btrfsic_release_block_ctx(&block_ctx);
2116	} else {
2117		/* block has not been found in hash table */
2118		u64 bytenr;
2119
2120		if (!is_metadata) {
2121			processed_len = state->datablock_size;
2122			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2123				printk(KERN_INFO "Written block (%s/%llu/?)"
2124				       " !found in hash table, D.\n",
2125				       dev_state->name, dev_bytenr);
2126			if (!state->include_extent_data) {
2127				/* ignore that written D block */
2128				goto continue_loop;
2129			}
2130
2131			/* this is getting ugly for the
2132			 * include_extent_data case... */
2133			bytenr = 0;	/* unknown */
2134			block_ctx.start = bytenr;
2135			block_ctx.len = processed_len;
2136			block_ctx.mem_to_free = NULL;
2137			block_ctx.pagev = NULL;
2138		} else {
2139			processed_len = state->metablock_size;
2140			bytenr = btrfs_stack_header_bytenr(
2141					(struct btrfs_header *)
2142					mapped_datav[0]);
2143			btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
2144						       dev_bytenr);
2145			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2146				printk(KERN_INFO
2147				       "Written block @%llu (%s/%llu/?)"
2148				       " !found in hash table, M.\n",
2149				       bytenr, dev_state->name, dev_bytenr);
2150
2151			ret = btrfsic_map_block(state, bytenr, processed_len,
2152						&block_ctx, 0);
2153			if (ret) {
2154				printk(KERN_INFO
2155				       "btrfsic: btrfsic_map_block(root @%llu)"
2156				       " failed!\n",
2157				       dev_bytenr);
2158				goto continue_loop;
2159			}
2160		}
2161		block_ctx.datav = mapped_datav;
2162		/* the following is required in case of writes to mirrors,
2163		 * use the same that was used for the lookup */
2164		block_ctx.dev = dev_state;
2165		block_ctx.dev_bytenr = dev_bytenr;
2166
2167		block = btrfsic_block_alloc();
2168		if (NULL == block) {
2169			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
2170			btrfsic_release_block_ctx(&block_ctx);
2171			goto continue_loop;
2172		}
2173		block->dev_state = dev_state;
2174		block->dev_bytenr = dev_bytenr;
2175		block->logical_bytenr = bytenr;
2176		block->is_metadata = is_metadata;
2177		block->never_written = 0;
2178		block->iodone_w_error = 0;
2179		block->mirror_num = 0;	/* unknown */
2180		block->flush_gen = dev_state->last_flush_gen + 1;
2181		block->submit_bio_bh_rw = submit_bio_bh_rw;
2182		if (NULL != bio) {
2183			block->is_iodone = 0;
2184			BUG_ON(NULL == bio_is_patched);
2185			if (!*bio_is_patched) {
2186				block->orig_bio_bh_private = bio->bi_private;
2187				block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2188				block->next_in_same_bio = NULL;
2189				bio->bi_private = block;
2190				bio->bi_end_io = btrfsic_bio_end_io;
2191				*bio_is_patched = 1;
2192			} else {
2193				struct btrfsic_block *chained_block =
2194				    (struct btrfsic_block *)
2195				    bio->bi_private;
2196
2197				BUG_ON(NULL == chained_block);
2198				block->orig_bio_bh_private =
2199				    chained_block->orig_bio_bh_private;
2200				block->orig_bio_bh_end_io.bio =
2201				    chained_block->orig_bio_bh_end_io.bio;
2202				block->next_in_same_bio = chained_block;
2203				bio->bi_private = block;
2204			}
2205		} else if (NULL != bh) {
2206			block->is_iodone = 0;
2207			block->orig_bio_bh_private = bh->b_private;
2208			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2209			block->next_in_same_bio = NULL;
2210			bh->b_private = block;
2211			bh->b_end_io = btrfsic_bh_end_io;
2212		} else {
2213			block->is_iodone = 1;
2214			block->orig_bio_bh_private = NULL;
2215			block->orig_bio_bh_end_io.bio = NULL;
2216			block->next_in_same_bio = NULL;
2217		}
2218		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2219			printk(KERN_INFO
2220			       "New written %c-block @%llu (%s/%llu/%d)\n",
2221			       is_metadata ? 'M' : 'D',
2222			       block->logical_bytenr, block->dev_state->name,
2223			       block->dev_bytenr, block->mirror_num);
2224		list_add(&block->all_blocks_node, &state->all_blocks_list);
2225		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2226
2227		if (is_metadata) {
2228			ret = btrfsic_process_metablock(state, block,
2229							&block_ctx, 0, 0);
2230			if (ret)
2231				printk(KERN_INFO
2232				       "btrfsic: process_metablock(root @%llu)"
2233				       " failed!\n",
2234				       dev_bytenr);
2235		}
2236		btrfsic_release_block_ctx(&block_ctx);
2237	}
2238
2239continue_loop:
2240	BUG_ON(!processed_len);
2241	dev_bytenr += processed_len;
2242	mapped_datav += processed_len >> PAGE_CACHE_SHIFT;
2243	num_pages -= processed_len >> PAGE_CACHE_SHIFT;
2244	goto again;
2245}
2246
2247static void btrfsic_bio_end_io(struct bio *bp, int bio_error_status)
2248{
2249	struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2250	int iodone_w_error;
2251
2252	/* mutex is not held! This is not save if IO is not yet completed
2253	 * on umount */
2254	iodone_w_error = 0;
2255	if (bio_error_status)
2256		iodone_w_error = 1;
2257
2258	BUG_ON(NULL == block);
2259	bp->bi_private = block->orig_bio_bh_private;
2260	bp->bi_end_io = block->orig_bio_bh_end_io.bio;
2261
2262	do {
2263		struct btrfsic_block *next_block;
2264		struct btrfsic_dev_state *const dev_state = block->dev_state;
2265
2266		if ((dev_state->state->print_mask &
2267		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2268			printk(KERN_INFO
2269			       "bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2270			       bio_error_status,
2271			       btrfsic_get_block_type(dev_state->state, block),
2272			       block->logical_bytenr, dev_state->name,
2273			       block->dev_bytenr, block->mirror_num);
2274		next_block = block->next_in_same_bio;
2275		block->iodone_w_error = iodone_w_error;
2276		if (block->submit_bio_bh_rw & REQ_FLUSH) {
2277			dev_state->last_flush_gen++;
2278			if ((dev_state->state->print_mask &
2279			     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2280				printk(KERN_INFO
2281				       "bio_end_io() new %s flush_gen=%llu\n",
2282				       dev_state->name,
2283				       dev_state->last_flush_gen);
2284		}
2285		if (block->submit_bio_bh_rw & REQ_FUA)
2286			block->flush_gen = 0; /* FUA completed means block is
2287					       * on disk */
2288		block->is_iodone = 1; /* for FLUSH, this releases the block */
2289		block = next_block;
2290	} while (NULL != block);
2291
2292	bp->bi_end_io(bp, bio_error_status);
2293}
2294
2295static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate)
2296{
2297	struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private;
2298	int iodone_w_error = !uptodate;
2299	struct btrfsic_dev_state *dev_state;
2300
2301	BUG_ON(NULL == block);
2302	dev_state = block->dev_state;
2303	if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2304		printk(KERN_INFO
2305		       "bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n",
2306		       iodone_w_error,
2307		       btrfsic_get_block_type(dev_state->state, block),
2308		       block->logical_bytenr, block->dev_state->name,
2309		       block->dev_bytenr, block->mirror_num);
2310
2311	block->iodone_w_error = iodone_w_error;
2312	if (block->submit_bio_bh_rw & REQ_FLUSH) {
2313		dev_state->last_flush_gen++;
2314		if ((dev_state->state->print_mask &
2315		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2316			printk(KERN_INFO
2317			       "bh_end_io() new %s flush_gen=%llu\n",
2318			       dev_state->name, dev_state->last_flush_gen);
2319	}
2320	if (block->submit_bio_bh_rw & REQ_FUA)
2321		block->flush_gen = 0; /* FUA completed means block is on disk */
2322
2323	bh->b_private = block->orig_bio_bh_private;
2324	bh->b_end_io = block->orig_bio_bh_end_io.bh;
2325	block->is_iodone = 1; /* for FLUSH, this releases the block */
2326	bh->b_end_io(bh, uptodate);
2327}
2328
2329static int btrfsic_process_written_superblock(
2330		struct btrfsic_state *state,
2331		struct btrfsic_block *const superblock,
2332		struct btrfs_super_block *const super_hdr)
2333{
2334	int pass;
2335
2336	superblock->generation = btrfs_super_generation(super_hdr);
2337	if (!(superblock->generation > state->max_superblock_generation ||
2338	      0 == state->max_superblock_generation)) {
2339		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2340			printk(KERN_INFO
2341			       "btrfsic: superblock @%llu (%s/%llu/%d)"
2342			       " with old gen %llu <= %llu\n",
2343			       superblock->logical_bytenr,
2344			       superblock->dev_state->name,
2345			       superblock->dev_bytenr, superblock->mirror_num,
2346			       btrfs_super_generation(super_hdr),
2347			       state->max_superblock_generation);
2348	} else {
2349		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2350			printk(KERN_INFO
2351			       "btrfsic: got new superblock @%llu (%s/%llu/%d)"
2352			       " with new gen %llu > %llu\n",
2353			       superblock->logical_bytenr,
2354			       superblock->dev_state->name,
2355			       superblock->dev_bytenr, superblock->mirror_num,
2356			       btrfs_super_generation(super_hdr),
2357			       state->max_superblock_generation);
2358
2359		state->max_superblock_generation =
2360		    btrfs_super_generation(super_hdr);
2361		state->latest_superblock = superblock;
2362	}
2363
2364	for (pass = 0; pass < 3; pass++) {
2365		int ret;
2366		u64 next_bytenr;
2367		struct btrfsic_block *next_block;
2368		struct btrfsic_block_data_ctx tmp_next_block_ctx;
2369		struct btrfsic_block_link *l;
2370		int num_copies;
2371		int mirror_num;
2372		const char *additional_string = NULL;
2373		struct btrfs_disk_key tmp_disk_key = {0};
2374
2375		btrfs_set_disk_key_objectid(&tmp_disk_key,
2376					    BTRFS_ROOT_ITEM_KEY);
2377		btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2378
2379		switch (pass) {
2380		case 0:
2381			btrfs_set_disk_key_objectid(&tmp_disk_key,
2382						    BTRFS_ROOT_TREE_OBJECTID);
2383			additional_string = "root ";
2384			next_bytenr = btrfs_super_root(super_hdr);
2385			if (state->print_mask &
2386			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2387				printk(KERN_INFO "root@%llu\n", next_bytenr);
2388			break;
2389		case 1:
2390			btrfs_set_disk_key_objectid(&tmp_disk_key,
2391						    BTRFS_CHUNK_TREE_OBJECTID);
2392			additional_string = "chunk ";
2393			next_bytenr = btrfs_super_chunk_root(super_hdr);
2394			if (state->print_mask &
2395			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2396				printk(KERN_INFO "chunk@%llu\n", next_bytenr);
2397			break;
2398		case 2:
2399			btrfs_set_disk_key_objectid(&tmp_disk_key,
2400						    BTRFS_TREE_LOG_OBJECTID);
2401			additional_string = "log ";
2402			next_bytenr = btrfs_super_log_root(super_hdr);
2403			if (0 == next_bytenr)
2404				continue;
2405			if (state->print_mask &
2406			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2407				printk(KERN_INFO "log@%llu\n", next_bytenr);
2408			break;
2409		}
2410
2411		num_copies =
2412		    btrfs_num_copies(state->root->fs_info,
2413				     next_bytenr, BTRFS_SUPER_INFO_SIZE);
2414		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2415			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
2416			       next_bytenr, num_copies);
2417		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2418			int was_created;
2419
2420			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2421				printk(KERN_INFO
2422				       "btrfsic_process_written_superblock("
2423				       "mirror_num=%d)\n", mirror_num);
2424			ret = btrfsic_map_block(state, next_bytenr,
2425						BTRFS_SUPER_INFO_SIZE,
2426						&tmp_next_block_ctx,
2427						mirror_num);
2428			if (ret) {
2429				printk(KERN_INFO
2430				       "btrfsic: btrfsic_map_block(@%llu,"
2431				       " mirror=%d) failed!\n",
2432				       next_bytenr, mirror_num);
2433				return -1;
2434			}
2435
2436			next_block = btrfsic_block_lookup_or_add(
2437					state,
2438					&tmp_next_block_ctx,
2439					additional_string,
2440					1, 0, 1,
2441					mirror_num,
2442					&was_created);
2443			if (NULL == next_block) {
2444				printk(KERN_INFO
2445				       "btrfsic: error, kmalloc failed!\n");
2446				btrfsic_release_block_ctx(&tmp_next_block_ctx);
2447				return -1;
2448			}
2449
2450			next_block->disk_key = tmp_disk_key;
2451			if (was_created)
2452				next_block->generation =
2453				    BTRFSIC_GENERATION_UNKNOWN;
2454			l = btrfsic_block_link_lookup_or_add(
2455					state,
2456					&tmp_next_block_ctx,
2457					next_block,
2458					superblock,
2459					BTRFSIC_GENERATION_UNKNOWN);
2460			btrfsic_release_block_ctx(&tmp_next_block_ctx);
2461			if (NULL == l)
2462				return -1;
2463		}
2464	}
2465
2466	if (-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)) {
2467		WARN_ON(1);
2468		btrfsic_dump_tree(state);
2469	}
2470
2471	return 0;
2472}
2473
2474static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2475					struct btrfsic_block *const block,
2476					int recursion_level)
2477{
2478	struct list_head *elem_ref_to;
2479	int ret = 0;
2480
2481	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2482		/*
2483		 * Note that this situation can happen and does not
2484		 * indicate an error in regular cases. It happens
2485		 * when disk blocks are freed and later reused.
2486		 * The check-integrity module is not aware of any
2487		 * block free operations, it just recognizes block
2488		 * write operations. Therefore it keeps the linkage
2489		 * information for a block until a block is
2490		 * rewritten. This can temporarily cause incorrect
2491		 * and even circular linkage informations. This
2492		 * causes no harm unless such blocks are referenced
2493		 * by the most recent super block.
2494		 */
2495		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2496			printk(KERN_INFO
2497			       "btrfsic: abort cyclic linkage (case 1).\n");
2498
2499		return ret;
2500	}
2501
2502	/*
2503	 * This algorithm is recursive because the amount of used stack
2504	 * space is very small and the max recursion depth is limited.
2505	 */
2506	list_for_each(elem_ref_to, &block->ref_to_list) {
2507		const struct btrfsic_block_link *const l =
2508		    list_entry(elem_ref_to, struct btrfsic_block_link,
2509			       node_ref_to);
2510
2511		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2512			printk(KERN_INFO
2513			       "rl=%d, %c @%llu (%s/%llu/%d)"
2514			       " %u* refers to %c @%llu (%s/%llu/%d)\n",
2515			       recursion_level,
2516			       btrfsic_get_block_type(state, block),
2517			       block->logical_bytenr, block->dev_state->name,
2518			       block->dev_bytenr, block->mirror_num,
2519			       l->ref_cnt,
2520			       btrfsic_get_block_type(state, l->block_ref_to),
2521			       l->block_ref_to->logical_bytenr,
2522			       l->block_ref_to->dev_state->name,
2523			       l->block_ref_to->dev_bytenr,
2524			       l->block_ref_to->mirror_num);
2525		if (l->block_ref_to->never_written) {
2526			printk(KERN_INFO "btrfs: attempt to write superblock"
2527			       " which references block %c @%llu (%s/%llu/%d)"
2528			       " which is never written!\n",
2529			       btrfsic_get_block_type(state, l->block_ref_to),
2530			       l->block_ref_to->logical_bytenr,
2531			       l->block_ref_to->dev_state->name,
2532			       l->block_ref_to->dev_bytenr,
2533			       l->block_ref_to->mirror_num);
2534			ret = -1;
2535		} else if (!l->block_ref_to->is_iodone) {
2536			printk(KERN_INFO "btrfs: attempt to write superblock"
2537			       " which references block %c @%llu (%s/%llu/%d)"
2538			       " which is not yet iodone!\n",
2539			       btrfsic_get_block_type(state, l->block_ref_to),
2540			       l->block_ref_to->logical_bytenr,
2541			       l->block_ref_to->dev_state->name,
2542			       l->block_ref_to->dev_bytenr,
2543			       l->block_ref_to->mirror_num);
2544			ret = -1;
2545		} else if (l->block_ref_to->iodone_w_error) {
2546			printk(KERN_INFO "btrfs: attempt to write superblock"
2547			       " which references block %c @%llu (%s/%llu/%d)"
2548			       " which has write error!\n",
2549			       btrfsic_get_block_type(state, l->block_ref_to),
2550			       l->block_ref_to->logical_bytenr,
2551			       l->block_ref_to->dev_state->name,
2552			       l->block_ref_to->dev_bytenr,
2553			       l->block_ref_to->mirror_num);
2554			ret = -1;
2555		} else if (l->parent_generation !=
2556			   l->block_ref_to->generation &&
2557			   BTRFSIC_GENERATION_UNKNOWN !=
2558			   l->parent_generation &&
2559			   BTRFSIC_GENERATION_UNKNOWN !=
2560			   l->block_ref_to->generation) {
2561			printk(KERN_INFO "btrfs: attempt to write superblock"
2562			       " which references block %c @%llu (%s/%llu/%d)"
2563			       " with generation %llu !="
2564			       " parent generation %llu!\n",
2565			       btrfsic_get_block_type(state, l->block_ref_to),
2566			       l->block_ref_to->logical_bytenr,
2567			       l->block_ref_to->dev_state->name,
2568			       l->block_ref_to->dev_bytenr,
2569			       l->block_ref_to->mirror_num,
2570			       l->block_ref_to->generation,
2571			       l->parent_generation);
2572			ret = -1;
2573		} else if (l->block_ref_to->flush_gen >
2574			   l->block_ref_to->dev_state->last_flush_gen) {
2575			printk(KERN_INFO "btrfs: attempt to write superblock"
2576			       " which references block %c @%llu (%s/%llu/%d)"
2577			       " which is not flushed out of disk's write cache"
2578			       " (block flush_gen=%llu,"
2579			       " dev->flush_gen=%llu)!\n",
2580			       btrfsic_get_block_type(state, l->block_ref_to),
2581			       l->block_ref_to->logical_bytenr,
2582			       l->block_ref_to->dev_state->name,
2583			       l->block_ref_to->dev_bytenr,
2584			       l->block_ref_to->mirror_num, block->flush_gen,
2585			       l->block_ref_to->dev_state->last_flush_gen);
2586			ret = -1;
2587		} else if (-1 == btrfsic_check_all_ref_blocks(state,
2588							      l->block_ref_to,
2589							      recursion_level +
2590							      1)) {
2591			ret = -1;
2592		}
2593	}
2594
2595	return ret;
2596}
2597
2598static int btrfsic_is_block_ref_by_superblock(
2599		const struct btrfsic_state *state,
2600		const struct btrfsic_block *block,
2601		int recursion_level)
2602{
2603	struct list_head *elem_ref_from;
2604
2605	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2606		/* refer to comment at "abort cyclic linkage (case 1)" */
2607		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2608			printk(KERN_INFO
2609			       "btrfsic: abort cyclic linkage (case 2).\n");
2610
2611		return 0;
2612	}
2613
2614	/*
2615	 * This algorithm is recursive because the amount of used stack space
2616	 * is very small and the max recursion depth is limited.
2617	 */
2618	list_for_each(elem_ref_from, &block->ref_from_list) {
2619		const struct btrfsic_block_link *const l =
2620		    list_entry(elem_ref_from, struct btrfsic_block_link,
2621			       node_ref_from);
2622
2623		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2624			printk(KERN_INFO
2625			       "rl=%d, %c @%llu (%s/%llu/%d)"
2626			       " is ref %u* from %c @%llu (%s/%llu/%d)\n",
2627			       recursion_level,
2628			       btrfsic_get_block_type(state, block),
2629			       block->logical_bytenr, block->dev_state->name,
2630			       block->dev_bytenr, block->mirror_num,
2631			       l->ref_cnt,
2632			       btrfsic_get_block_type(state, l->block_ref_from),
2633			       l->block_ref_from->logical_bytenr,
2634			       l->block_ref_from->dev_state->name,
2635			       l->block_ref_from->dev_bytenr,
2636			       l->block_ref_from->mirror_num);
2637		if (l->block_ref_from->is_superblock &&
2638		    state->latest_superblock->dev_bytenr ==
2639		    l->block_ref_from->dev_bytenr &&
2640		    state->latest_superblock->dev_state->bdev ==
2641		    l->block_ref_from->dev_state->bdev)
2642			return 1;
2643		else if (btrfsic_is_block_ref_by_superblock(state,
2644							    l->block_ref_from,
2645							    recursion_level +
2646							    1))
2647			return 1;
2648	}
2649
2650	return 0;
2651}
2652
2653static void btrfsic_print_add_link(const struct btrfsic_state *state,
2654				   const struct btrfsic_block_link *l)
2655{
2656	printk(KERN_INFO
2657	       "Add %u* link from %c @%llu (%s/%llu/%d)"
2658	       " to %c @%llu (%s/%llu/%d).\n",
2659	       l->ref_cnt,
2660	       btrfsic_get_block_type(state, l->block_ref_from),
2661	       l->block_ref_from->logical_bytenr,
2662	       l->block_ref_from->dev_state->name,
2663	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2664	       btrfsic_get_block_type(state, l->block_ref_to),
2665	       l->block_ref_to->logical_bytenr,
2666	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2667	       l->block_ref_to->mirror_num);
2668}
2669
2670static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2671				   const struct btrfsic_block_link *l)
2672{
2673	printk(KERN_INFO
2674	       "Rem %u* link from %c @%llu (%s/%llu/%d)"
2675	       " to %c @%llu (%s/%llu/%d).\n",
2676	       l->ref_cnt,
2677	       btrfsic_get_block_type(state, l->block_ref_from),
2678	       l->block_ref_from->logical_bytenr,
2679	       l->block_ref_from->dev_state->name,
2680	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2681	       btrfsic_get_block_type(state, l->block_ref_to),
2682	       l->block_ref_to->logical_bytenr,
2683	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2684	       l->block_ref_to->mirror_num);
2685}
2686
2687static char btrfsic_get_block_type(const struct btrfsic_state *state,
2688				   const struct btrfsic_block *block)
2689{
2690	if (block->is_superblock &&
2691	    state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2692	    state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2693		return 'S';
2694	else if (block->is_superblock)
2695		return 's';
2696	else if (block->is_metadata)
2697		return 'M';
2698	else
2699		return 'D';
2700}
2701
2702static void btrfsic_dump_tree(const struct btrfsic_state *state)
2703{
2704	btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2705}
2706
2707static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2708				  const struct btrfsic_block *block,
2709				  int indent_level)
2710{
2711	struct list_head *elem_ref_to;
2712	int indent_add;
2713	static char buf[80];
2714	int cursor_position;
2715
2716	/*
2717	 * Should better fill an on-stack buffer with a complete line and
2718	 * dump it at once when it is time to print a newline character.
2719	 */
2720
2721	/*
2722	 * This algorithm is recursive because the amount of used stack space
2723	 * is very small and the max recursion depth is limited.
2724	 */
2725	indent_add = sprintf(buf, "%c-%llu(%s/%llu/%d)",
2726			     btrfsic_get_block_type(state, block),
2727			     block->logical_bytenr, block->dev_state->name,
2728			     block->dev_bytenr, block->mirror_num);
2729	if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2730		printk("[...]\n");
2731		return;
2732	}
2733	printk(buf);
2734	indent_level += indent_add;
2735	if (list_empty(&block->ref_to_list)) {
2736		printk("\n");
2737		return;
2738	}
2739	if (block->mirror_num > 1 &&
2740	    !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2741		printk(" [...]\n");
2742		return;
2743	}
2744
2745	cursor_position = indent_level;
2746	list_for_each(elem_ref_to, &block->ref_to_list) {
2747		const struct btrfsic_block_link *const l =
2748		    list_entry(elem_ref_to, struct btrfsic_block_link,
2749			       node_ref_to);
2750
2751		while (cursor_position < indent_level) {
2752			printk(" ");
2753			cursor_position++;
2754		}
2755		if (l->ref_cnt > 1)
2756			indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2757		else
2758			indent_add = sprintf(buf, " --> ");
2759		if (indent_level + indent_add >
2760		    BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2761			printk("[...]\n");
2762			cursor_position = 0;
2763			continue;
2764		}
2765
2766		printk(buf);
2767
2768		btrfsic_dump_tree_sub(state, l->block_ref_to,
2769				      indent_level + indent_add);
2770		cursor_position = 0;
2771	}
2772}
2773
2774static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2775		struct btrfsic_state *state,
2776		struct btrfsic_block_data_ctx *next_block_ctx,
2777		struct btrfsic_block *next_block,
2778		struct btrfsic_block *from_block,
2779		u64 parent_generation)
2780{
2781	struct btrfsic_block_link *l;
2782
2783	l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2784						next_block_ctx->dev_bytenr,
2785						from_block->dev_state->bdev,
2786						from_block->dev_bytenr,
2787						&state->block_link_hashtable);
2788	if (NULL == l) {
2789		l = btrfsic_block_link_alloc();
2790		if (NULL == l) {
2791			printk(KERN_INFO
2792			       "btrfsic: error, kmalloc" " failed!\n");
2793			return NULL;
2794		}
2795
2796		l->block_ref_to = next_block;
2797		l->block_ref_from = from_block;
2798		l->ref_cnt = 1;
2799		l->parent_generation = parent_generation;
2800
2801		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2802			btrfsic_print_add_link(state, l);
2803
2804		list_add(&l->node_ref_to, &from_block->ref_to_list);
2805		list_add(&l->node_ref_from, &next_block->ref_from_list);
2806
2807		btrfsic_block_link_hashtable_add(l,
2808						 &state->block_link_hashtable);
2809	} else {
2810		l->ref_cnt++;
2811		l->parent_generation = parent_generation;
2812		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2813			btrfsic_print_add_link(state, l);
2814	}
2815
2816	return l;
2817}
2818
2819static struct btrfsic_block *btrfsic_block_lookup_or_add(
2820		struct btrfsic_state *state,
2821		struct btrfsic_block_data_ctx *block_ctx,
2822		const char *additional_string,
2823		int is_metadata,
2824		int is_iodone,
2825		int never_written,
2826		int mirror_num,
2827		int *was_created)
2828{
2829	struct btrfsic_block *block;
2830
2831	block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2832					       block_ctx->dev_bytenr,
2833					       &state->block_hashtable);
2834	if (NULL == block) {
2835		struct btrfsic_dev_state *dev_state;
2836
2837		block = btrfsic_block_alloc();
2838		if (NULL == block) {
2839			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
2840			return NULL;
2841		}
2842		dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev);
2843		if (NULL == dev_state) {
2844			printk(KERN_INFO
2845			       "btrfsic: error, lookup dev_state failed!\n");
2846			btrfsic_block_free(block);
2847			return NULL;
2848		}
2849		block->dev_state = dev_state;
2850		block->dev_bytenr = block_ctx->dev_bytenr;
2851		block->logical_bytenr = block_ctx->start;
2852		block->is_metadata = is_metadata;
2853		block->is_iodone = is_iodone;
2854		block->never_written = never_written;
2855		block->mirror_num = mirror_num;
2856		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2857			printk(KERN_INFO
2858			       "New %s%c-block @%llu (%s/%llu/%d)\n",
2859			       additional_string,
2860			       btrfsic_get_block_type(state, block),
2861			       block->logical_bytenr, dev_state->name,
2862			       block->dev_bytenr, mirror_num);
2863		list_add(&block->all_blocks_node, &state->all_blocks_list);
2864		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2865		if (NULL != was_created)
2866			*was_created = 1;
2867	} else {
2868		if (NULL != was_created)
2869			*was_created = 0;
2870	}
2871
2872	return block;
2873}
2874
2875static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2876					   u64 bytenr,
2877					   struct btrfsic_dev_state *dev_state,
2878					   u64 dev_bytenr)
2879{
2880	int num_copies;
2881	int mirror_num;
2882	int ret;
2883	struct btrfsic_block_data_ctx block_ctx;
2884	int match = 0;
2885
2886	num_copies = btrfs_num_copies(state->root->fs_info,
2887				      bytenr, state->metablock_size);
2888
2889	for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2890		ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2891					&block_ctx, mirror_num);
2892		if (ret) {
2893			printk(KERN_INFO "btrfsic:"
2894			       " btrfsic_map_block(logical @%llu,"
2895			       " mirror %d) failed!\n",
2896			       bytenr, mirror_num);
2897			continue;
2898		}
2899
2900		if (dev_state->bdev == block_ctx.dev->bdev &&
2901		    dev_bytenr == block_ctx.dev_bytenr) {
2902			match++;
2903			btrfsic_release_block_ctx(&block_ctx);
2904			break;
2905		}
2906		btrfsic_release_block_ctx(&block_ctx);
2907	}
2908
2909	if (!match) {
2910		printk(KERN_INFO "btrfs: attempt to write M-block which contains logical bytenr that doesn't map to dev+physical bytenr of submit_bio,"
2911		       " buffer->log_bytenr=%llu, submit_bio(bdev=%s,"
2912		       " phys_bytenr=%llu)!\n",
2913		       bytenr, dev_state->name, dev_bytenr);
2914		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2915			ret = btrfsic_map_block(state, bytenr,
2916						state->metablock_size,
2917						&block_ctx, mirror_num);
2918			if (ret)
2919				continue;
2920
2921			printk(KERN_INFO "Read logical bytenr @%llu maps to"
2922			       " (%s/%llu/%d)\n",
2923			       bytenr, block_ctx.dev->name,
2924			       block_ctx.dev_bytenr, mirror_num);
2925		}
2926		WARN_ON(1);
2927	}
2928}
2929
2930static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
2931		struct block_device *bdev)
2932{
2933	struct btrfsic_dev_state *ds;
2934
2935	ds = btrfsic_dev_state_hashtable_lookup(bdev,
2936						&btrfsic_dev_state_hashtable);
2937	return ds;
2938}
2939
2940int btrfsic_submit_bh(int rw, struct buffer_head *bh)
2941{
2942	struct btrfsic_dev_state *dev_state;
2943
2944	if (!btrfsic_is_initialized)
2945		return submit_bh(rw, bh);
2946
2947	mutex_lock(&btrfsic_mutex);
2948	/* since btrfsic_submit_bh() might also be called before
2949	 * btrfsic_mount(), this might return NULL */
2950	dev_state = btrfsic_dev_state_lookup(bh->b_bdev);
2951
2952	/* Only called to write the superblock (incl. FLUSH/FUA) */
2953	if (NULL != dev_state &&
2954	    (rw & WRITE) && bh->b_size > 0) {
2955		u64 dev_bytenr;
2956
2957		dev_bytenr = 4096 * bh->b_blocknr;
2958		if (dev_state->state->print_mask &
2959		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2960			printk(KERN_INFO
2961			       "submit_bh(rw=0x%x, blocknr=%llu (bytenr %llu),"
2962			       " size=%zu, data=%p, bdev=%p)\n",
2963			       rw, (unsigned long long)bh->b_blocknr,
2964			       dev_bytenr, bh->b_size, bh->b_data, bh->b_bdev);
2965		btrfsic_process_written_block(dev_state, dev_bytenr,
2966					      &bh->b_data, 1, NULL,
2967					      NULL, bh, rw);
2968	} else if (NULL != dev_state && (rw & REQ_FLUSH)) {
2969		if (dev_state->state->print_mask &
2970		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2971			printk(KERN_INFO
2972			       "submit_bh(rw=0x%x FLUSH, bdev=%p)\n",
2973			       rw, bh->b_bdev);
2974		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2975			if ((dev_state->state->print_mask &
2976			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2977			      BTRFSIC_PRINT_MASK_VERBOSE)))
2978				printk(KERN_INFO
2979				       "btrfsic_submit_bh(%s) with FLUSH"
2980				       " but dummy block already in use"
2981				       " (ignored)!\n",
2982				       dev_state->name);
2983		} else {
2984			struct btrfsic_block *const block =
2985				&dev_state->dummy_block_for_bio_bh_flush;
2986
2987			block->is_iodone = 0;
2988			block->never_written = 0;
2989			block->iodone_w_error = 0;
2990			block->flush_gen = dev_state->last_flush_gen + 1;
2991			block->submit_bio_bh_rw = rw;
2992			block->orig_bio_bh_private = bh->b_private;
2993			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2994			block->next_in_same_bio = NULL;
2995			bh->b_private = block;
2996			bh->b_end_io = btrfsic_bh_end_io;
2997		}
2998	}
2999	mutex_unlock(&btrfsic_mutex);
3000	return submit_bh(rw, bh);
3001}
3002
3003void btrfsic_submit_bio(int rw, struct bio *bio)
3004{
3005	struct btrfsic_dev_state *dev_state;
3006
3007	if (!btrfsic_is_initialized) {
3008		submit_bio(rw, bio);
3009		return;
3010	}
3011
3012	mutex_lock(&btrfsic_mutex);
3013	/* since btrfsic_submit_bio() is also called before
3014	 * btrfsic_mount(), this might return NULL */
3015	dev_state = btrfsic_dev_state_lookup(bio->bi_bdev);
3016	if (NULL != dev_state &&
3017	    (rw & WRITE) && NULL != bio->bi_io_vec) {
3018		unsigned int i;
3019		u64 dev_bytenr;
3020		int bio_is_patched;
3021		char **mapped_datav;
3022
3023		dev_bytenr = 512 * bio->bi_sector;
3024		bio_is_patched = 0;
3025		if (dev_state->state->print_mask &
3026		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
3027			printk(KERN_INFO
3028			       "submit_bio(rw=0x%x, bi_vcnt=%u,"
3029			       " bi_sector=%llu (bytenr %llu), bi_bdev=%p)\n",
3030			       rw, bio->bi_vcnt,
3031			       (unsigned long long)bio->bi_sector, dev_bytenr,
3032			       bio->bi_bdev);
3033
3034		mapped_datav = kmalloc(sizeof(*mapped_datav) * bio->bi_vcnt,
3035				       GFP_NOFS);
3036		if (!mapped_datav)
3037			goto leave;
3038		for (i = 0; i < bio->bi_vcnt; i++) {
3039			BUG_ON(bio->bi_io_vec[i].bv_len != PAGE_CACHE_SIZE);
3040			mapped_datav[i] = kmap(bio->bi_io_vec[i].bv_page);
3041			if (!mapped_datav[i]) {
3042				while (i > 0) {
3043					i--;
3044					kunmap(bio->bi_io_vec[i].bv_page);
3045				}
3046				kfree(mapped_datav);
3047				goto leave;
3048			}
3049			if ((BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
3050			     BTRFSIC_PRINT_MASK_VERBOSE) ==
3051			    (dev_state->state->print_mask &
3052			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
3053			      BTRFSIC_PRINT_MASK_VERBOSE)))
3054				printk(KERN_INFO
3055				       "#%u: page=%p, len=%u, offset=%u\n",
3056				       i, bio->bi_io_vec[i].bv_page,
3057				       bio->bi_io_vec[i].bv_len,
3058				       bio->bi_io_vec[i].bv_offset);
3059		}
3060		btrfsic_process_written_block(dev_state, dev_bytenr,
3061					      mapped_datav, bio->bi_vcnt,
3062					      bio, &bio_is_patched,
3063					      NULL, rw);
3064		while (i > 0) {
3065			i--;
3066			kunmap(bio->bi_io_vec[i].bv_page);
3067		}
3068		kfree(mapped_datav);
3069	} else if (NULL != dev_state && (rw & REQ_FLUSH)) {
3070		if (dev_state->state->print_mask &
3071		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
3072			printk(KERN_INFO
3073			       "submit_bio(rw=0x%x FLUSH, bdev=%p)\n",
3074			       rw, bio->bi_bdev);
3075		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
3076			if ((dev_state->state->print_mask &
3077			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
3078			      BTRFSIC_PRINT_MASK_VERBOSE)))
3079				printk(KERN_INFO
3080				       "btrfsic_submit_bio(%s) with FLUSH"
3081				       " but dummy block already in use"
3082				       " (ignored)!\n",
3083				       dev_state->name);
3084		} else {
3085			struct btrfsic_block *const block =
3086				&dev_state->dummy_block_for_bio_bh_flush;
3087
3088			block->is_iodone = 0;
3089			block->never_written = 0;
3090			block->iodone_w_error = 0;
3091			block->flush_gen = dev_state->last_flush_gen + 1;
3092			block->submit_bio_bh_rw = rw;
3093			block->orig_bio_bh_private = bio->bi_private;
3094			block->orig_bio_bh_end_io.bio = bio->bi_end_io;
3095			block->next_in_same_bio = NULL;
3096			bio->bi_private = block;
3097			bio->bi_end_io = btrfsic_bio_end_io;
3098		}
3099	}
3100leave:
3101	mutex_unlock(&btrfsic_mutex);
3102
3103	submit_bio(rw, bio);
3104}
3105
3106int btrfsic_mount(struct btrfs_root *root,
3107		  struct btrfs_fs_devices *fs_devices,
3108		  int including_extent_data, u32 print_mask)
3109{
3110	int ret;
3111	struct btrfsic_state *state;
3112	struct list_head *dev_head = &fs_devices->devices;
3113	struct btrfs_device *device;
3114
3115	if (root->nodesize != root->leafsize) {
3116		printk(KERN_INFO
3117		       "btrfsic: cannot handle nodesize %d != leafsize %d!\n",
3118		       root->nodesize, root->leafsize);
3119		return -1;
3120	}
3121	if (root->nodesize & ((u64)PAGE_CACHE_SIZE - 1)) {
3122		printk(KERN_INFO
3123		       "btrfsic: cannot handle nodesize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3124		       root->nodesize, PAGE_CACHE_SIZE);
3125		return -1;
3126	}
3127	if (root->leafsize & ((u64)PAGE_CACHE_SIZE - 1)) {
3128		printk(KERN_INFO
3129		       "btrfsic: cannot handle leafsize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3130		       root->leafsize, PAGE_CACHE_SIZE);
3131		return -1;
3132	}
3133	if (root->sectorsize & ((u64)PAGE_CACHE_SIZE - 1)) {
3134		printk(KERN_INFO
3135		       "btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3136		       root->sectorsize, PAGE_CACHE_SIZE);
3137		return -1;
3138	}
3139	state = kzalloc(sizeof(*state), GFP_NOFS);
3140	if (NULL == state) {
3141		printk(KERN_INFO "btrfs check-integrity: kmalloc() failed!\n");
3142		return -1;
3143	}
3144
3145	if (!btrfsic_is_initialized) {
3146		mutex_init(&btrfsic_mutex);
3147		btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
3148		btrfsic_is_initialized = 1;
3149	}
3150	mutex_lock(&btrfsic_mutex);
3151	state->root = root;
3152	state->print_mask = print_mask;
3153	state->include_extent_data = including_extent_data;
3154	state->csum_size = 0;
3155	state->metablock_size = root->nodesize;
3156	state->datablock_size = root->sectorsize;
3157	INIT_LIST_HEAD(&state->all_blocks_list);
3158	btrfsic_block_hashtable_init(&state->block_hashtable);
3159	btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
3160	state->max_superblock_generation = 0;
3161	state->latest_superblock = NULL;
3162
3163	list_for_each_entry(device, dev_head, dev_list) {
3164		struct btrfsic_dev_state *ds;
3165		char *p;
3166
3167		if (!device->bdev || !device->name)
3168			continue;
3169
3170		ds = btrfsic_dev_state_alloc();
3171		if (NULL == ds) {
3172			printk(KERN_INFO
3173			       "btrfs check-integrity: kmalloc() failed!\n");
3174			mutex_unlock(&btrfsic_mutex);
3175			return -1;
3176		}
3177		ds->bdev = device->bdev;
3178		ds->state = state;
3179		bdevname(ds->bdev, ds->name);
3180		ds->name[BDEVNAME_SIZE - 1] = '\0';
3181		for (p = ds->name; *p != '\0'; p++);
3182		while (p > ds->name && *p != '/')
3183			p--;
3184		if (*p == '/')
3185			p++;
3186		strlcpy(ds->name, p, sizeof(ds->name));
3187		btrfsic_dev_state_hashtable_add(ds,
3188						&btrfsic_dev_state_hashtable);
3189	}
3190
3191	ret = btrfsic_process_superblock(state, fs_devices);
3192	if (0 != ret) {
3193		mutex_unlock(&btrfsic_mutex);
3194		btrfsic_unmount(root, fs_devices);
3195		return ret;
3196	}
3197
3198	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
3199		btrfsic_dump_database(state);
3200	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
3201		btrfsic_dump_tree(state);
3202
3203	mutex_unlock(&btrfsic_mutex);
3204	return 0;
3205}
3206
3207void btrfsic_unmount(struct btrfs_root *root,
3208		     struct btrfs_fs_devices *fs_devices)
3209{
3210	struct list_head *elem_all;
3211	struct list_head *tmp_all;
3212	struct btrfsic_state *state;
3213	struct list_head *dev_head = &fs_devices->devices;
3214	struct btrfs_device *device;
3215
3216	if (!btrfsic_is_initialized)
3217		return;
3218
3219	mutex_lock(&btrfsic_mutex);
3220
3221	state = NULL;
3222	list_for_each_entry(device, dev_head, dev_list) {
3223		struct btrfsic_dev_state *ds;
3224
3225		if (!device->bdev || !device->name)
3226			continue;
3227
3228		ds = btrfsic_dev_state_hashtable_lookup(
3229				device->bdev,
3230				&btrfsic_dev_state_hashtable);
3231		if (NULL != ds) {
3232			state = ds->state;
3233			btrfsic_dev_state_hashtable_remove(ds);
3234			btrfsic_dev_state_free(ds);
3235		}
3236	}
3237
3238	if (NULL == state) {
3239		printk(KERN_INFO
3240		       "btrfsic: error, cannot find state information"
3241		       " on umount!\n");
3242		mutex_unlock(&btrfsic_mutex);
3243		return;
3244	}
3245
3246	/*
3247	 * Don't care about keeping the lists' state up to date,
3248	 * just free all memory that was allocated dynamically.
3249	 * Free the blocks and the block_links.
3250	 */
3251	list_for_each_safe(elem_all, tmp_all, &state->all_blocks_list) {
3252		struct btrfsic_block *const b_all =
3253		    list_entry(elem_all, struct btrfsic_block,
3254			       all_blocks_node);
3255		struct list_head *elem_ref_to;
3256		struct list_head *tmp_ref_to;
3257
3258		list_for_each_safe(elem_ref_to, tmp_ref_to,
3259				   &b_all->ref_to_list) {
3260			struct btrfsic_block_link *const l =
3261			    list_entry(elem_ref_to,
3262				       struct btrfsic_block_link,
3263				       node_ref_to);
3264
3265			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
3266				btrfsic_print_rem_link(state, l);
3267
3268			l->ref_cnt--;
3269			if (0 == l->ref_cnt)
3270				btrfsic_block_link_free(l);
3271		}
3272
3273		if (b_all->is_iodone || b_all->never_written)
3274			btrfsic_block_free(b_all);
3275		else
3276			printk(KERN_INFO "btrfs: attempt to free %c-block"
3277			       " @%llu (%s/%llu/%d) on umount which is"
3278			       " not yet iodone!\n",
3279			       btrfsic_get_block_type(state, b_all),
3280			       b_all->logical_bytenr, b_all->dev_state->name,
3281			       b_all->dev_bytenr, b_all->mirror_num);
3282	}
3283
3284	mutex_unlock(&btrfsic_mutex);
3285
3286	kfree(state);
3287}
3288