check-integrity.c revision 56d140f5f61d3547dcc194de9cd5193b957b8016
1/*
2 * Copyright (C) STRATO AG 2011.  All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19/*
20 * This module can be used to catch cases when the btrfs kernel
21 * code executes write requests to the disk that bring the file
22 * system in an inconsistent state. In such a state, a power-loss
23 * or kernel panic event would cause that the data on disk is
24 * lost or at least damaged.
25 *
26 * Code is added that examines all block write requests during
27 * runtime (including writes of the super block). Three rules
28 * are verified and an error is printed on violation of the
29 * rules:
30 * 1. It is not allowed to write a disk block which is
31 *    currently referenced by the super block (either directly
32 *    or indirectly).
33 * 2. When a super block is written, it is verified that all
34 *    referenced (directly or indirectly) blocks fulfill the
35 *    following requirements:
36 *    2a. All referenced blocks have either been present when
37 *        the file system was mounted, (i.e., they have been
38 *        referenced by the super block) or they have been
39 *        written since then and the write completion callback
40 *        was called and no write error was indicated and a
41 *        FLUSH request to the device where these blocks are
42 *        located was received and completed.
43 *    2b. All referenced blocks need to have a generation
44 *        number which is equal to the parent's number.
45 *
46 * One issue that was found using this module was that the log
47 * tree on disk became temporarily corrupted because disk blocks
48 * that had been in use for the log tree had been freed and
49 * reused too early, while being referenced by the written super
50 * block.
51 *
52 * The search term in the kernel log that can be used to filter
53 * on the existence of detected integrity issues is
54 * "btrfs: attempt".
55 *
56 * The integrity check is enabled via mount options. These
57 * mount options are only supported if the integrity check
58 * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY.
59 *
60 * Example #1, apply integrity checks to all metadata:
61 * mount /dev/sdb1 /mnt -o check_int
62 *
63 * Example #2, apply integrity checks to all metadata and
64 * to data extents:
65 * mount /dev/sdb1 /mnt -o check_int_data
66 *
67 * Example #3, apply integrity checks to all metadata and dump
68 * the tree that the super block references to kernel messages
69 * each time after a super block was written:
70 * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263
71 *
72 * If the integrity check tool is included and activated in
73 * the mount options, plenty of kernel memory is used, and
74 * plenty of additional CPU cycles are spent. Enabling this
75 * functionality is not intended for normal use. In most
76 * cases, unless you are a btrfs developer who needs to verify
77 * the integrity of (super)-block write requests, do not
78 * enable the config option BTRFS_FS_CHECK_INTEGRITY to
79 * include and compile the integrity check tool.
80 *
81 * Expect millions of lines of information in the kernel log with an
82 * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the
83 * kernel config to at least 26 (which is 64MB). Usually the value is
84 * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be
85 * changed like this before LOG_BUF_SHIFT can be set to a high value:
86 * config LOG_BUF_SHIFT
87 *       int "Kernel log buffer size (16 => 64KB, 17 => 128KB)"
88 *       range 12 30
89 */
90
91#include <linux/sched.h>
92#include <linux/slab.h>
93#include <linux/buffer_head.h>
94#include <linux/mutex.h>
95#include <linux/crc32c.h>
96#include <linux/genhd.h>
97#include <linux/blkdev.h>
98#include "ctree.h"
99#include "disk-io.h"
100#include "transaction.h"
101#include "extent_io.h"
102#include "volumes.h"
103#include "print-tree.h"
104#include "locking.h"
105#include "check-integrity.h"
106#include "rcu-string.h"
107
108#define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
109#define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
110#define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
111#define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
112#define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
113#define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
114#define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
115#define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6)	/* in characters,
116							 * excluding " [...]" */
117#define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
118
119/*
120 * The definition of the bitmask fields for the print_mask.
121 * They are specified with the mount option check_integrity_print_mask.
122 */
123#define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE			0x00000001
124#define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION		0x00000002
125#define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE			0x00000004
126#define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE			0x00000008
127#define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH			0x00000010
128#define BTRFSIC_PRINT_MASK_END_IO_BIO_BH			0x00000020
129#define BTRFSIC_PRINT_MASK_VERBOSE				0x00000040
130#define BTRFSIC_PRINT_MASK_VERY_VERBOSE				0x00000080
131#define BTRFSIC_PRINT_MASK_INITIAL_TREE				0x00000100
132#define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES			0x00000200
133#define BTRFSIC_PRINT_MASK_INITIAL_DATABASE			0x00000400
134#define BTRFSIC_PRINT_MASK_NUM_COPIES				0x00000800
135#define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS		0x00001000
136#define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE		0x00002000
137
138struct btrfsic_dev_state;
139struct btrfsic_state;
140
141struct btrfsic_block {
142	u32 magic_num;		/* only used for debug purposes */
143	unsigned int is_metadata:1;	/* if it is meta-data, not data-data */
144	unsigned int is_superblock:1;	/* if it is one of the superblocks */
145	unsigned int is_iodone:1;	/* if is done by lower subsystem */
146	unsigned int iodone_w_error:1;	/* error was indicated to endio */
147	unsigned int never_written:1;	/* block was added because it was
148					 * referenced, not because it was
149					 * written */
150	unsigned int mirror_num;	/* large enough to hold
151					 * BTRFS_SUPER_MIRROR_MAX */
152	struct btrfsic_dev_state *dev_state;
153	u64 dev_bytenr;		/* key, physical byte num on disk */
154	u64 logical_bytenr;	/* logical byte num on disk */
155	u64 generation;
156	struct btrfs_disk_key disk_key;	/* extra info to print in case of
157					 * issues, will not always be correct */
158	struct list_head collision_resolving_node;	/* list node */
159	struct list_head all_blocks_node;	/* list node */
160
161	/* the following two lists contain block_link items */
162	struct list_head ref_to_list;	/* list */
163	struct list_head ref_from_list;	/* list */
164	struct btrfsic_block *next_in_same_bio;
165	void *orig_bio_bh_private;
166	union {
167		bio_end_io_t *bio;
168		bh_end_io_t *bh;
169	} orig_bio_bh_end_io;
170	int submit_bio_bh_rw;
171	u64 flush_gen; /* only valid if !never_written */
172};
173
174/*
175 * Elements of this type are allocated dynamically and required because
176 * each block object can refer to and can be ref from multiple blocks.
177 * The key to lookup them in the hashtable is the dev_bytenr of
178 * the block ref to plus the one from the block refered from.
179 * The fact that they are searchable via a hashtable and that a
180 * ref_cnt is maintained is not required for the btrfs integrity
181 * check algorithm itself, it is only used to make the output more
182 * beautiful in case that an error is detected (an error is defined
183 * as a write operation to a block while that block is still referenced).
184 */
185struct btrfsic_block_link {
186	u32 magic_num;		/* only used for debug purposes */
187	u32 ref_cnt;
188	struct list_head node_ref_to;	/* list node */
189	struct list_head node_ref_from;	/* list node */
190	struct list_head collision_resolving_node;	/* list node */
191	struct btrfsic_block *block_ref_to;
192	struct btrfsic_block *block_ref_from;
193	u64 parent_generation;
194};
195
196struct btrfsic_dev_state {
197	u32 magic_num;		/* only used for debug purposes */
198	struct block_device *bdev;
199	struct btrfsic_state *state;
200	struct list_head collision_resolving_node;	/* list node */
201	struct btrfsic_block dummy_block_for_bio_bh_flush;
202	u64 last_flush_gen;
203	char name[BDEVNAME_SIZE];
204};
205
206struct btrfsic_block_hashtable {
207	struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
208};
209
210struct btrfsic_block_link_hashtable {
211	struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
212};
213
214struct btrfsic_dev_state_hashtable {
215	struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
216};
217
218struct btrfsic_block_data_ctx {
219	u64 start;		/* virtual bytenr */
220	u64 dev_bytenr;		/* physical bytenr on device */
221	u32 len;
222	struct btrfsic_dev_state *dev;
223	char **datav;
224	struct page **pagev;
225	void *mem_to_free;
226};
227
228/* This structure is used to implement recursion without occupying
229 * any stack space, refer to btrfsic_process_metablock() */
230struct btrfsic_stack_frame {
231	u32 magic;
232	u32 nr;
233	int error;
234	int i;
235	int limit_nesting;
236	int num_copies;
237	int mirror_num;
238	struct btrfsic_block *block;
239	struct btrfsic_block_data_ctx *block_ctx;
240	struct btrfsic_block *next_block;
241	struct btrfsic_block_data_ctx next_block_ctx;
242	struct btrfs_header *hdr;
243	struct btrfsic_stack_frame *prev;
244};
245
246/* Some state per mounted filesystem */
247struct btrfsic_state {
248	u32 print_mask;
249	int include_extent_data;
250	int csum_size;
251	struct list_head all_blocks_list;
252	struct btrfsic_block_hashtable block_hashtable;
253	struct btrfsic_block_link_hashtable block_link_hashtable;
254	struct btrfs_root *root;
255	u64 max_superblock_generation;
256	struct btrfsic_block *latest_superblock;
257	u32 metablock_size;
258	u32 datablock_size;
259};
260
261static void btrfsic_block_init(struct btrfsic_block *b);
262static struct btrfsic_block *btrfsic_block_alloc(void);
263static void btrfsic_block_free(struct btrfsic_block *b);
264static void btrfsic_block_link_init(struct btrfsic_block_link *n);
265static struct btrfsic_block_link *btrfsic_block_link_alloc(void);
266static void btrfsic_block_link_free(struct btrfsic_block_link *n);
267static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds);
268static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void);
269static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds);
270static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h);
271static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
272					struct btrfsic_block_hashtable *h);
273static void btrfsic_block_hashtable_remove(struct btrfsic_block *b);
274static struct btrfsic_block *btrfsic_block_hashtable_lookup(
275		struct block_device *bdev,
276		u64 dev_bytenr,
277		struct btrfsic_block_hashtable *h);
278static void btrfsic_block_link_hashtable_init(
279		struct btrfsic_block_link_hashtable *h);
280static void btrfsic_block_link_hashtable_add(
281		struct btrfsic_block_link *l,
282		struct btrfsic_block_link_hashtable *h);
283static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l);
284static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
285		struct block_device *bdev_ref_to,
286		u64 dev_bytenr_ref_to,
287		struct block_device *bdev_ref_from,
288		u64 dev_bytenr_ref_from,
289		struct btrfsic_block_link_hashtable *h);
290static void btrfsic_dev_state_hashtable_init(
291		struct btrfsic_dev_state_hashtable *h);
292static void btrfsic_dev_state_hashtable_add(
293		struct btrfsic_dev_state *ds,
294		struct btrfsic_dev_state_hashtable *h);
295static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds);
296static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(
297		struct block_device *bdev,
298		struct btrfsic_dev_state_hashtable *h);
299static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void);
300static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf);
301static int btrfsic_process_superblock(struct btrfsic_state *state,
302				      struct btrfs_fs_devices *fs_devices);
303static int btrfsic_process_metablock(struct btrfsic_state *state,
304				     struct btrfsic_block *block,
305				     struct btrfsic_block_data_ctx *block_ctx,
306				     int limit_nesting, int force_iodone_flag);
307static void btrfsic_read_from_block_data(
308	struct btrfsic_block_data_ctx *block_ctx,
309	void *dst, u32 offset, size_t len);
310static int btrfsic_create_link_to_next_block(
311		struct btrfsic_state *state,
312		struct btrfsic_block *block,
313		struct btrfsic_block_data_ctx
314		*block_ctx, u64 next_bytenr,
315		int limit_nesting,
316		struct btrfsic_block_data_ctx *next_block_ctx,
317		struct btrfsic_block **next_blockp,
318		int force_iodone_flag,
319		int *num_copiesp, int *mirror_nump,
320		struct btrfs_disk_key *disk_key,
321		u64 parent_generation);
322static int btrfsic_handle_extent_data(struct btrfsic_state *state,
323				      struct btrfsic_block *block,
324				      struct btrfsic_block_data_ctx *block_ctx,
325				      u32 item_offset, int force_iodone_flag);
326static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
327			     struct btrfsic_block_data_ctx *block_ctx_out,
328			     int mirror_num);
329static int btrfsic_map_superblock(struct btrfsic_state *state, u64 bytenr,
330				  u32 len, struct block_device *bdev,
331				  struct btrfsic_block_data_ctx *block_ctx_out);
332static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
333static int btrfsic_read_block(struct btrfsic_state *state,
334			      struct btrfsic_block_data_ctx *block_ctx);
335static void btrfsic_dump_database(struct btrfsic_state *state);
336static void btrfsic_complete_bio_end_io(struct bio *bio, int err);
337static int btrfsic_test_for_metadata(struct btrfsic_state *state,
338				     char **datav, unsigned int num_pages);
339static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
340					  u64 dev_bytenr, char **mapped_datav,
341					  unsigned int num_pages,
342					  struct bio *bio, int *bio_is_patched,
343					  struct buffer_head *bh,
344					  int submit_bio_bh_rw);
345static int btrfsic_process_written_superblock(
346		struct btrfsic_state *state,
347		struct btrfsic_block *const block,
348		struct btrfs_super_block *const super_hdr);
349static void btrfsic_bio_end_io(struct bio *bp, int bio_error_status);
350static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate);
351static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
352					      const struct btrfsic_block *block,
353					      int recursion_level);
354static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
355					struct btrfsic_block *const block,
356					int recursion_level);
357static void btrfsic_print_add_link(const struct btrfsic_state *state,
358				   const struct btrfsic_block_link *l);
359static void btrfsic_print_rem_link(const struct btrfsic_state *state,
360				   const struct btrfsic_block_link *l);
361static char btrfsic_get_block_type(const struct btrfsic_state *state,
362				   const struct btrfsic_block *block);
363static void btrfsic_dump_tree(const struct btrfsic_state *state);
364static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
365				  const struct btrfsic_block *block,
366				  int indent_level);
367static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
368		struct btrfsic_state *state,
369		struct btrfsic_block_data_ctx *next_block_ctx,
370		struct btrfsic_block *next_block,
371		struct btrfsic_block *from_block,
372		u64 parent_generation);
373static struct btrfsic_block *btrfsic_block_lookup_or_add(
374		struct btrfsic_state *state,
375		struct btrfsic_block_data_ctx *block_ctx,
376		const char *additional_string,
377		int is_metadata,
378		int is_iodone,
379		int never_written,
380		int mirror_num,
381		int *was_created);
382static int btrfsic_process_superblock_dev_mirror(
383		struct btrfsic_state *state,
384		struct btrfsic_dev_state *dev_state,
385		struct btrfs_device *device,
386		int superblock_mirror_num,
387		struct btrfsic_dev_state **selected_dev_state,
388		struct btrfs_super_block *selected_super);
389static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
390		struct block_device *bdev);
391static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
392					   u64 bytenr,
393					   struct btrfsic_dev_state *dev_state,
394					   u64 dev_bytenr);
395
396static struct mutex btrfsic_mutex;
397static int btrfsic_is_initialized;
398static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
399
400
401static void btrfsic_block_init(struct btrfsic_block *b)
402{
403	b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
404	b->dev_state = NULL;
405	b->dev_bytenr = 0;
406	b->logical_bytenr = 0;
407	b->generation = BTRFSIC_GENERATION_UNKNOWN;
408	b->disk_key.objectid = 0;
409	b->disk_key.type = 0;
410	b->disk_key.offset = 0;
411	b->is_metadata = 0;
412	b->is_superblock = 0;
413	b->is_iodone = 0;
414	b->iodone_w_error = 0;
415	b->never_written = 0;
416	b->mirror_num = 0;
417	b->next_in_same_bio = NULL;
418	b->orig_bio_bh_private = NULL;
419	b->orig_bio_bh_end_io.bio = NULL;
420	INIT_LIST_HEAD(&b->collision_resolving_node);
421	INIT_LIST_HEAD(&b->all_blocks_node);
422	INIT_LIST_HEAD(&b->ref_to_list);
423	INIT_LIST_HEAD(&b->ref_from_list);
424	b->submit_bio_bh_rw = 0;
425	b->flush_gen = 0;
426}
427
428static struct btrfsic_block *btrfsic_block_alloc(void)
429{
430	struct btrfsic_block *b;
431
432	b = kzalloc(sizeof(*b), GFP_NOFS);
433	if (NULL != b)
434		btrfsic_block_init(b);
435
436	return b;
437}
438
439static void btrfsic_block_free(struct btrfsic_block *b)
440{
441	BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
442	kfree(b);
443}
444
445static void btrfsic_block_link_init(struct btrfsic_block_link *l)
446{
447	l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
448	l->ref_cnt = 1;
449	INIT_LIST_HEAD(&l->node_ref_to);
450	INIT_LIST_HEAD(&l->node_ref_from);
451	INIT_LIST_HEAD(&l->collision_resolving_node);
452	l->block_ref_to = NULL;
453	l->block_ref_from = NULL;
454}
455
456static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
457{
458	struct btrfsic_block_link *l;
459
460	l = kzalloc(sizeof(*l), GFP_NOFS);
461	if (NULL != l)
462		btrfsic_block_link_init(l);
463
464	return l;
465}
466
467static void btrfsic_block_link_free(struct btrfsic_block_link *l)
468{
469	BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
470	kfree(l);
471}
472
473static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
474{
475	ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
476	ds->bdev = NULL;
477	ds->state = NULL;
478	ds->name[0] = '\0';
479	INIT_LIST_HEAD(&ds->collision_resolving_node);
480	ds->last_flush_gen = 0;
481	btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
482	ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
483	ds->dummy_block_for_bio_bh_flush.dev_state = ds;
484}
485
486static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
487{
488	struct btrfsic_dev_state *ds;
489
490	ds = kzalloc(sizeof(*ds), GFP_NOFS);
491	if (NULL != ds)
492		btrfsic_dev_state_init(ds);
493
494	return ds;
495}
496
497static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
498{
499	BUG_ON(!(NULL == ds ||
500		 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
501	kfree(ds);
502}
503
504static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
505{
506	int i;
507
508	for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
509		INIT_LIST_HEAD(h->table + i);
510}
511
512static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
513					struct btrfsic_block_hashtable *h)
514{
515	const unsigned int hashval =
516	    (((unsigned int)(b->dev_bytenr >> 16)) ^
517	     ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
518	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
519
520	list_add(&b->collision_resolving_node, h->table + hashval);
521}
522
523static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
524{
525	list_del(&b->collision_resolving_node);
526}
527
528static struct btrfsic_block *btrfsic_block_hashtable_lookup(
529		struct block_device *bdev,
530		u64 dev_bytenr,
531		struct btrfsic_block_hashtable *h)
532{
533	const unsigned int hashval =
534	    (((unsigned int)(dev_bytenr >> 16)) ^
535	     ((unsigned int)((uintptr_t)bdev))) &
536	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
537	struct list_head *elem;
538
539	list_for_each(elem, h->table + hashval) {
540		struct btrfsic_block *const b =
541		    list_entry(elem, struct btrfsic_block,
542			       collision_resolving_node);
543
544		if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
545			return b;
546	}
547
548	return NULL;
549}
550
551static void btrfsic_block_link_hashtable_init(
552		struct btrfsic_block_link_hashtable *h)
553{
554	int i;
555
556	for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
557		INIT_LIST_HEAD(h->table + i);
558}
559
560static void btrfsic_block_link_hashtable_add(
561		struct btrfsic_block_link *l,
562		struct btrfsic_block_link_hashtable *h)
563{
564	const unsigned int hashval =
565	    (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
566	     ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
567	     ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
568	     ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
569	     & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
570
571	BUG_ON(NULL == l->block_ref_to);
572	BUG_ON(NULL == l->block_ref_from);
573	list_add(&l->collision_resolving_node, h->table + hashval);
574}
575
576static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
577{
578	list_del(&l->collision_resolving_node);
579}
580
581static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
582		struct block_device *bdev_ref_to,
583		u64 dev_bytenr_ref_to,
584		struct block_device *bdev_ref_from,
585		u64 dev_bytenr_ref_from,
586		struct btrfsic_block_link_hashtable *h)
587{
588	const unsigned int hashval =
589	    (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
590	     ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
591	     ((unsigned int)((uintptr_t)bdev_ref_to)) ^
592	     ((unsigned int)((uintptr_t)bdev_ref_from))) &
593	     (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
594	struct list_head *elem;
595
596	list_for_each(elem, h->table + hashval) {
597		struct btrfsic_block_link *const l =
598		    list_entry(elem, struct btrfsic_block_link,
599			       collision_resolving_node);
600
601		BUG_ON(NULL == l->block_ref_to);
602		BUG_ON(NULL == l->block_ref_from);
603		if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
604		    l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
605		    l->block_ref_from->dev_state->bdev == bdev_ref_from &&
606		    l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
607			return l;
608	}
609
610	return NULL;
611}
612
613static void btrfsic_dev_state_hashtable_init(
614		struct btrfsic_dev_state_hashtable *h)
615{
616	int i;
617
618	for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
619		INIT_LIST_HEAD(h->table + i);
620}
621
622static void btrfsic_dev_state_hashtable_add(
623		struct btrfsic_dev_state *ds,
624		struct btrfsic_dev_state_hashtable *h)
625{
626	const unsigned int hashval =
627	    (((unsigned int)((uintptr_t)ds->bdev)) &
628	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
629
630	list_add(&ds->collision_resolving_node, h->table + hashval);
631}
632
633static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
634{
635	list_del(&ds->collision_resolving_node);
636}
637
638static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(
639		struct block_device *bdev,
640		struct btrfsic_dev_state_hashtable *h)
641{
642	const unsigned int hashval =
643	    (((unsigned int)((uintptr_t)bdev)) &
644	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
645	struct list_head *elem;
646
647	list_for_each(elem, h->table + hashval) {
648		struct btrfsic_dev_state *const ds =
649		    list_entry(elem, struct btrfsic_dev_state,
650			       collision_resolving_node);
651
652		if (ds->bdev == bdev)
653			return ds;
654	}
655
656	return NULL;
657}
658
659static int btrfsic_process_superblock(struct btrfsic_state *state,
660				      struct btrfs_fs_devices *fs_devices)
661{
662	int ret = 0;
663	struct btrfs_super_block *selected_super;
664	struct list_head *dev_head = &fs_devices->devices;
665	struct btrfs_device *device;
666	struct btrfsic_dev_state *selected_dev_state = NULL;
667	int pass;
668
669	BUG_ON(NULL == state);
670	selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
671	if (NULL == selected_super) {
672		printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
673		return -1;
674	}
675
676	list_for_each_entry(device, dev_head, dev_list) {
677		int i;
678		struct btrfsic_dev_state *dev_state;
679
680		if (!device->bdev || !device->name)
681			continue;
682
683		dev_state = btrfsic_dev_state_lookup(device->bdev);
684		BUG_ON(NULL == dev_state);
685		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
686			ret = btrfsic_process_superblock_dev_mirror(
687					state, dev_state, device, i,
688					&selected_dev_state, selected_super);
689			if (0 != ret && 0 == i) {
690				kfree(selected_super);
691				return ret;
692			}
693		}
694	}
695
696	if (NULL == state->latest_superblock) {
697		printk(KERN_INFO "btrfsic: no superblock found!\n");
698		kfree(selected_super);
699		return -1;
700	}
701
702	state->csum_size = btrfs_super_csum_size(selected_super);
703
704	for (pass = 0; pass < 3; pass++) {
705		int num_copies;
706		int mirror_num;
707		u64 next_bytenr;
708
709		switch (pass) {
710		case 0:
711			next_bytenr = btrfs_super_root(selected_super);
712			if (state->print_mask &
713			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
714				printk(KERN_INFO "root@%llu\n", next_bytenr);
715			break;
716		case 1:
717			next_bytenr = btrfs_super_chunk_root(selected_super);
718			if (state->print_mask &
719			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
720				printk(KERN_INFO "chunk@%llu\n", next_bytenr);
721			break;
722		case 2:
723			next_bytenr = btrfs_super_log_root(selected_super);
724			if (0 == next_bytenr)
725				continue;
726			if (state->print_mask &
727			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
728				printk(KERN_INFO "log@%llu\n", next_bytenr);
729			break;
730		}
731
732		num_copies =
733		    btrfs_num_copies(state->root->fs_info,
734				     next_bytenr, state->metablock_size);
735		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
736			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
737			       next_bytenr, num_copies);
738
739		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
740			struct btrfsic_block *next_block;
741			struct btrfsic_block_data_ctx tmp_next_block_ctx;
742			struct btrfsic_block_link *l;
743
744			ret = btrfsic_map_block(state, next_bytenr,
745						state->metablock_size,
746						&tmp_next_block_ctx,
747						mirror_num);
748			if (ret) {
749				printk(KERN_INFO "btrfsic:"
750				       " btrfsic_map_block(root @%llu,"
751				       " mirror %d) failed!\n",
752				       next_bytenr, mirror_num);
753				kfree(selected_super);
754				return -1;
755			}
756
757			next_block = btrfsic_block_hashtable_lookup(
758					tmp_next_block_ctx.dev->bdev,
759					tmp_next_block_ctx.dev_bytenr,
760					&state->block_hashtable);
761			BUG_ON(NULL == next_block);
762
763			l = btrfsic_block_link_hashtable_lookup(
764					tmp_next_block_ctx.dev->bdev,
765					tmp_next_block_ctx.dev_bytenr,
766					state->latest_superblock->dev_state->
767					bdev,
768					state->latest_superblock->dev_bytenr,
769					&state->block_link_hashtable);
770			BUG_ON(NULL == l);
771
772			ret = btrfsic_read_block(state, &tmp_next_block_ctx);
773			if (ret < (int)PAGE_CACHE_SIZE) {
774				printk(KERN_INFO
775				       "btrfsic: read @logical %llu failed!\n",
776				       tmp_next_block_ctx.start);
777				btrfsic_release_block_ctx(&tmp_next_block_ctx);
778				kfree(selected_super);
779				return -1;
780			}
781
782			ret = btrfsic_process_metablock(state,
783							next_block,
784							&tmp_next_block_ctx,
785							BTRFS_MAX_LEVEL + 3, 1);
786			btrfsic_release_block_ctx(&tmp_next_block_ctx);
787		}
788	}
789
790	kfree(selected_super);
791	return ret;
792}
793
794static int btrfsic_process_superblock_dev_mirror(
795		struct btrfsic_state *state,
796		struct btrfsic_dev_state *dev_state,
797		struct btrfs_device *device,
798		int superblock_mirror_num,
799		struct btrfsic_dev_state **selected_dev_state,
800		struct btrfs_super_block *selected_super)
801{
802	struct btrfs_super_block *super_tmp;
803	u64 dev_bytenr;
804	struct buffer_head *bh;
805	struct btrfsic_block *superblock_tmp;
806	int pass;
807	struct block_device *const superblock_bdev = device->bdev;
808
809	/* super block bytenr is always the unmapped device bytenr */
810	dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
811	if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->total_bytes)
812		return -1;
813	bh = __bread(superblock_bdev, dev_bytenr / 4096,
814		     BTRFS_SUPER_INFO_SIZE);
815	if (NULL == bh)
816		return -1;
817	super_tmp = (struct btrfs_super_block *)
818	    (bh->b_data + (dev_bytenr & 4095));
819
820	if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
821	    btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
822	    memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
823	    btrfs_super_nodesize(super_tmp) != state->metablock_size ||
824	    btrfs_super_leafsize(super_tmp) != state->metablock_size ||
825	    btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
826		brelse(bh);
827		return 0;
828	}
829
830	superblock_tmp =
831	    btrfsic_block_hashtable_lookup(superblock_bdev,
832					   dev_bytenr,
833					   &state->block_hashtable);
834	if (NULL == superblock_tmp) {
835		superblock_tmp = btrfsic_block_alloc();
836		if (NULL == superblock_tmp) {
837			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
838			brelse(bh);
839			return -1;
840		}
841		/* for superblock, only the dev_bytenr makes sense */
842		superblock_tmp->dev_bytenr = dev_bytenr;
843		superblock_tmp->dev_state = dev_state;
844		superblock_tmp->logical_bytenr = dev_bytenr;
845		superblock_tmp->generation = btrfs_super_generation(super_tmp);
846		superblock_tmp->is_metadata = 1;
847		superblock_tmp->is_superblock = 1;
848		superblock_tmp->is_iodone = 1;
849		superblock_tmp->never_written = 0;
850		superblock_tmp->mirror_num = 1 + superblock_mirror_num;
851		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
852			printk_in_rcu(KERN_INFO "New initial S-block (bdev %p, %s)"
853				     " @%llu (%s/%llu/%d)\n",
854				     superblock_bdev,
855				     rcu_str_deref(device->name), dev_bytenr,
856				     dev_state->name, dev_bytenr,
857				     superblock_mirror_num);
858		list_add(&superblock_tmp->all_blocks_node,
859			 &state->all_blocks_list);
860		btrfsic_block_hashtable_add(superblock_tmp,
861					    &state->block_hashtable);
862	}
863
864	/* select the one with the highest generation field */
865	if (btrfs_super_generation(super_tmp) >
866	    state->max_superblock_generation ||
867	    0 == state->max_superblock_generation) {
868		memcpy(selected_super, super_tmp, sizeof(*selected_super));
869		*selected_dev_state = dev_state;
870		state->max_superblock_generation =
871		    btrfs_super_generation(super_tmp);
872		state->latest_superblock = superblock_tmp;
873	}
874
875	for (pass = 0; pass < 3; pass++) {
876		u64 next_bytenr;
877		int num_copies;
878		int mirror_num;
879		const char *additional_string = NULL;
880		struct btrfs_disk_key tmp_disk_key;
881
882		tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
883		tmp_disk_key.offset = 0;
884		switch (pass) {
885		case 0:
886			btrfs_set_disk_key_objectid(&tmp_disk_key,
887						    BTRFS_ROOT_TREE_OBJECTID);
888			additional_string = "initial root ";
889			next_bytenr = btrfs_super_root(super_tmp);
890			break;
891		case 1:
892			btrfs_set_disk_key_objectid(&tmp_disk_key,
893						    BTRFS_CHUNK_TREE_OBJECTID);
894			additional_string = "initial chunk ";
895			next_bytenr = btrfs_super_chunk_root(super_tmp);
896			break;
897		case 2:
898			btrfs_set_disk_key_objectid(&tmp_disk_key,
899						    BTRFS_TREE_LOG_OBJECTID);
900			additional_string = "initial log ";
901			next_bytenr = btrfs_super_log_root(super_tmp);
902			if (0 == next_bytenr)
903				continue;
904			break;
905		}
906
907		num_copies =
908		    btrfs_num_copies(state->root->fs_info,
909				     next_bytenr, state->metablock_size);
910		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
911			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
912			       next_bytenr, num_copies);
913		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
914			struct btrfsic_block *next_block;
915			struct btrfsic_block_data_ctx tmp_next_block_ctx;
916			struct btrfsic_block_link *l;
917
918			if (btrfsic_map_block(state, next_bytenr,
919					      state->metablock_size,
920					      &tmp_next_block_ctx,
921					      mirror_num)) {
922				printk(KERN_INFO "btrfsic: btrfsic_map_block("
923				       "bytenr @%llu, mirror %d) failed!\n",
924				       next_bytenr, mirror_num);
925				brelse(bh);
926				return -1;
927			}
928
929			next_block = btrfsic_block_lookup_or_add(
930					state, &tmp_next_block_ctx,
931					additional_string, 1, 1, 0,
932					mirror_num, NULL);
933			if (NULL == next_block) {
934				btrfsic_release_block_ctx(&tmp_next_block_ctx);
935				brelse(bh);
936				return -1;
937			}
938
939			next_block->disk_key = tmp_disk_key;
940			next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
941			l = btrfsic_block_link_lookup_or_add(
942					state, &tmp_next_block_ctx,
943					next_block, superblock_tmp,
944					BTRFSIC_GENERATION_UNKNOWN);
945			btrfsic_release_block_ctx(&tmp_next_block_ctx);
946			if (NULL == l) {
947				brelse(bh);
948				return -1;
949			}
950		}
951	}
952	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
953		btrfsic_dump_tree_sub(state, superblock_tmp, 0);
954
955	brelse(bh);
956	return 0;
957}
958
959static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
960{
961	struct btrfsic_stack_frame *sf;
962
963	sf = kzalloc(sizeof(*sf), GFP_NOFS);
964	if (NULL == sf)
965		printk(KERN_INFO "btrfsic: alloc memory failed!\n");
966	else
967		sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
968	return sf;
969}
970
971static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
972{
973	BUG_ON(!(NULL == sf ||
974		 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
975	kfree(sf);
976}
977
978static int btrfsic_process_metablock(
979		struct btrfsic_state *state,
980		struct btrfsic_block *const first_block,
981		struct btrfsic_block_data_ctx *const first_block_ctx,
982		int first_limit_nesting, int force_iodone_flag)
983{
984	struct btrfsic_stack_frame initial_stack_frame = { 0 };
985	struct btrfsic_stack_frame *sf;
986	struct btrfsic_stack_frame *next_stack;
987	struct btrfs_header *const first_hdr =
988		(struct btrfs_header *)first_block_ctx->datav[0];
989
990	BUG_ON(!first_hdr);
991	sf = &initial_stack_frame;
992	sf->error = 0;
993	sf->i = -1;
994	sf->limit_nesting = first_limit_nesting;
995	sf->block = first_block;
996	sf->block_ctx = first_block_ctx;
997	sf->next_block = NULL;
998	sf->hdr = first_hdr;
999	sf->prev = NULL;
1000
1001continue_with_new_stack_frame:
1002	sf->block->generation = le64_to_cpu(sf->hdr->generation);
1003	if (0 == sf->hdr->level) {
1004		struct btrfs_leaf *const leafhdr =
1005		    (struct btrfs_leaf *)sf->hdr;
1006
1007		if (-1 == sf->i) {
1008			sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
1009
1010			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1011				printk(KERN_INFO
1012				       "leaf %llu items %d generation %llu"
1013				       " owner %llu\n",
1014				       sf->block_ctx->start, sf->nr,
1015				       btrfs_stack_header_generation(
1016					       &leafhdr->header),
1017				       btrfs_stack_header_owner(
1018					       &leafhdr->header));
1019		}
1020
1021continue_with_current_leaf_stack_frame:
1022		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1023			sf->i++;
1024			sf->num_copies = 0;
1025		}
1026
1027		if (sf->i < sf->nr) {
1028			struct btrfs_item disk_item;
1029			u32 disk_item_offset =
1030				(uintptr_t)(leafhdr->items + sf->i) -
1031				(uintptr_t)leafhdr;
1032			struct btrfs_disk_key *disk_key;
1033			u8 type;
1034			u32 item_offset;
1035			u32 item_size;
1036
1037			if (disk_item_offset + sizeof(struct btrfs_item) >
1038			    sf->block_ctx->len) {
1039leaf_item_out_of_bounce_error:
1040				printk(KERN_INFO
1041				       "btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
1042				       sf->block_ctx->start,
1043				       sf->block_ctx->dev->name);
1044				goto one_stack_frame_backwards;
1045			}
1046			btrfsic_read_from_block_data(sf->block_ctx,
1047						     &disk_item,
1048						     disk_item_offset,
1049						     sizeof(struct btrfs_item));
1050			item_offset = btrfs_stack_item_offset(&disk_item);
1051			item_size = btrfs_stack_item_size(&disk_item);
1052			disk_key = &disk_item.key;
1053			type = btrfs_disk_key_type(disk_key);
1054
1055			if (BTRFS_ROOT_ITEM_KEY == type) {
1056				struct btrfs_root_item root_item;
1057				u32 root_item_offset;
1058				u64 next_bytenr;
1059
1060				root_item_offset = item_offset +
1061					offsetof(struct btrfs_leaf, items);
1062				if (root_item_offset + item_size >
1063				    sf->block_ctx->len)
1064					goto leaf_item_out_of_bounce_error;
1065				btrfsic_read_from_block_data(
1066					sf->block_ctx, &root_item,
1067					root_item_offset,
1068					item_size);
1069				next_bytenr = btrfs_root_bytenr(&root_item);
1070
1071				sf->error =
1072				    btrfsic_create_link_to_next_block(
1073						state,
1074						sf->block,
1075						sf->block_ctx,
1076						next_bytenr,
1077						sf->limit_nesting,
1078						&sf->next_block_ctx,
1079						&sf->next_block,
1080						force_iodone_flag,
1081						&sf->num_copies,
1082						&sf->mirror_num,
1083						disk_key,
1084						btrfs_root_generation(
1085						&root_item));
1086				if (sf->error)
1087					goto one_stack_frame_backwards;
1088
1089				if (NULL != sf->next_block) {
1090					struct btrfs_header *const next_hdr =
1091					    (struct btrfs_header *)
1092					    sf->next_block_ctx.datav[0];
1093
1094					next_stack =
1095					    btrfsic_stack_frame_alloc();
1096					if (NULL == next_stack) {
1097						btrfsic_release_block_ctx(
1098								&sf->
1099								next_block_ctx);
1100						goto one_stack_frame_backwards;
1101					}
1102
1103					next_stack->i = -1;
1104					next_stack->block = sf->next_block;
1105					next_stack->block_ctx =
1106					    &sf->next_block_ctx;
1107					next_stack->next_block = NULL;
1108					next_stack->hdr = next_hdr;
1109					next_stack->limit_nesting =
1110					    sf->limit_nesting - 1;
1111					next_stack->prev = sf;
1112					sf = next_stack;
1113					goto continue_with_new_stack_frame;
1114				}
1115			} else if (BTRFS_EXTENT_DATA_KEY == type &&
1116				   state->include_extent_data) {
1117				sf->error = btrfsic_handle_extent_data(
1118						state,
1119						sf->block,
1120						sf->block_ctx,
1121						item_offset,
1122						force_iodone_flag);
1123				if (sf->error)
1124					goto one_stack_frame_backwards;
1125			}
1126
1127			goto continue_with_current_leaf_stack_frame;
1128		}
1129	} else {
1130		struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1131
1132		if (-1 == sf->i) {
1133			sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1134
1135			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1136				printk(KERN_INFO "node %llu level %d items %d"
1137				       " generation %llu owner %llu\n",
1138				       sf->block_ctx->start,
1139				       nodehdr->header.level, sf->nr,
1140				       btrfs_stack_header_generation(
1141				       &nodehdr->header),
1142				       btrfs_stack_header_owner(
1143				       &nodehdr->header));
1144		}
1145
1146continue_with_current_node_stack_frame:
1147		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1148			sf->i++;
1149			sf->num_copies = 0;
1150		}
1151
1152		if (sf->i < sf->nr) {
1153			struct btrfs_key_ptr key_ptr;
1154			u32 key_ptr_offset;
1155			u64 next_bytenr;
1156
1157			key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1158					  (uintptr_t)nodehdr;
1159			if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1160			    sf->block_ctx->len) {
1161				printk(KERN_INFO
1162				       "btrfsic: node item out of bounce at logical %llu, dev %s\n",
1163				       sf->block_ctx->start,
1164				       sf->block_ctx->dev->name);
1165				goto one_stack_frame_backwards;
1166			}
1167			btrfsic_read_from_block_data(
1168				sf->block_ctx, &key_ptr, key_ptr_offset,
1169				sizeof(struct btrfs_key_ptr));
1170			next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1171
1172			sf->error = btrfsic_create_link_to_next_block(
1173					state,
1174					sf->block,
1175					sf->block_ctx,
1176					next_bytenr,
1177					sf->limit_nesting,
1178					&sf->next_block_ctx,
1179					&sf->next_block,
1180					force_iodone_flag,
1181					&sf->num_copies,
1182					&sf->mirror_num,
1183					&key_ptr.key,
1184					btrfs_stack_key_generation(&key_ptr));
1185			if (sf->error)
1186				goto one_stack_frame_backwards;
1187
1188			if (NULL != sf->next_block) {
1189				struct btrfs_header *const next_hdr =
1190				    (struct btrfs_header *)
1191				    sf->next_block_ctx.datav[0];
1192
1193				next_stack = btrfsic_stack_frame_alloc();
1194				if (NULL == next_stack)
1195					goto one_stack_frame_backwards;
1196
1197				next_stack->i = -1;
1198				next_stack->block = sf->next_block;
1199				next_stack->block_ctx = &sf->next_block_ctx;
1200				next_stack->next_block = NULL;
1201				next_stack->hdr = next_hdr;
1202				next_stack->limit_nesting =
1203				    sf->limit_nesting - 1;
1204				next_stack->prev = sf;
1205				sf = next_stack;
1206				goto continue_with_new_stack_frame;
1207			}
1208
1209			goto continue_with_current_node_stack_frame;
1210		}
1211	}
1212
1213one_stack_frame_backwards:
1214	if (NULL != sf->prev) {
1215		struct btrfsic_stack_frame *const prev = sf->prev;
1216
1217		/* the one for the initial block is freed in the caller */
1218		btrfsic_release_block_ctx(sf->block_ctx);
1219
1220		if (sf->error) {
1221			prev->error = sf->error;
1222			btrfsic_stack_frame_free(sf);
1223			sf = prev;
1224			goto one_stack_frame_backwards;
1225		}
1226
1227		btrfsic_stack_frame_free(sf);
1228		sf = prev;
1229		goto continue_with_new_stack_frame;
1230	} else {
1231		BUG_ON(&initial_stack_frame != sf);
1232	}
1233
1234	return sf->error;
1235}
1236
1237static void btrfsic_read_from_block_data(
1238	struct btrfsic_block_data_ctx *block_ctx,
1239	void *dstv, u32 offset, size_t len)
1240{
1241	size_t cur;
1242	size_t offset_in_page;
1243	char *kaddr;
1244	char *dst = (char *)dstv;
1245	size_t start_offset = block_ctx->start & ((u64)PAGE_CACHE_SIZE - 1);
1246	unsigned long i = (start_offset + offset) >> PAGE_CACHE_SHIFT;
1247
1248	WARN_ON(offset + len > block_ctx->len);
1249	offset_in_page = (start_offset + offset) & (PAGE_CACHE_SIZE - 1);
1250
1251	while (len > 0) {
1252		cur = min(len, ((size_t)PAGE_CACHE_SIZE - offset_in_page));
1253		BUG_ON(i >= (block_ctx->len + PAGE_CACHE_SIZE - 1) >>
1254			    PAGE_CACHE_SHIFT);
1255		kaddr = block_ctx->datav[i];
1256		memcpy(dst, kaddr + offset_in_page, cur);
1257
1258		dst += cur;
1259		len -= cur;
1260		offset_in_page = 0;
1261		i++;
1262	}
1263}
1264
1265static int btrfsic_create_link_to_next_block(
1266		struct btrfsic_state *state,
1267		struct btrfsic_block *block,
1268		struct btrfsic_block_data_ctx *block_ctx,
1269		u64 next_bytenr,
1270		int limit_nesting,
1271		struct btrfsic_block_data_ctx *next_block_ctx,
1272		struct btrfsic_block **next_blockp,
1273		int force_iodone_flag,
1274		int *num_copiesp, int *mirror_nump,
1275		struct btrfs_disk_key *disk_key,
1276		u64 parent_generation)
1277{
1278	struct btrfsic_block *next_block = NULL;
1279	int ret;
1280	struct btrfsic_block_link *l;
1281	int did_alloc_block_link;
1282	int block_was_created;
1283
1284	*next_blockp = NULL;
1285	if (0 == *num_copiesp) {
1286		*num_copiesp =
1287		    btrfs_num_copies(state->root->fs_info,
1288				     next_bytenr, state->metablock_size);
1289		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1290			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
1291			       next_bytenr, *num_copiesp);
1292		*mirror_nump = 1;
1293	}
1294
1295	if (*mirror_nump > *num_copiesp)
1296		return 0;
1297
1298	if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1299		printk(KERN_INFO
1300		       "btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1301		       *mirror_nump);
1302	ret = btrfsic_map_block(state, next_bytenr,
1303				state->metablock_size,
1304				next_block_ctx, *mirror_nump);
1305	if (ret) {
1306		printk(KERN_INFO
1307		       "btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1308		       next_bytenr, *mirror_nump);
1309		btrfsic_release_block_ctx(next_block_ctx);
1310		*next_blockp = NULL;
1311		return -1;
1312	}
1313
1314	next_block = btrfsic_block_lookup_or_add(state,
1315						 next_block_ctx, "referenced ",
1316						 1, force_iodone_flag,
1317						 !force_iodone_flag,
1318						 *mirror_nump,
1319						 &block_was_created);
1320	if (NULL == next_block) {
1321		btrfsic_release_block_ctx(next_block_ctx);
1322		*next_blockp = NULL;
1323		return -1;
1324	}
1325	if (block_was_created) {
1326		l = NULL;
1327		next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1328	} else {
1329		if (next_block->logical_bytenr != next_bytenr &&
1330		    !(!next_block->is_metadata &&
1331		      0 == next_block->logical_bytenr)) {
1332			printk(KERN_INFO
1333			       "Referenced block @%llu (%s/%llu/%d)"
1334			       " found in hash table, %c,"
1335			       " bytenr mismatch (!= stored %llu).\n",
1336			       next_bytenr, next_block_ctx->dev->name,
1337			       next_block_ctx->dev_bytenr, *mirror_nump,
1338			       btrfsic_get_block_type(state, next_block),
1339			       next_block->logical_bytenr);
1340		} else if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1341			printk(KERN_INFO
1342			       "Referenced block @%llu (%s/%llu/%d)"
1343			       " found in hash table, %c.\n",
1344			       next_bytenr, next_block_ctx->dev->name,
1345			       next_block_ctx->dev_bytenr, *mirror_nump,
1346			       btrfsic_get_block_type(state, next_block));
1347		next_block->logical_bytenr = next_bytenr;
1348
1349		next_block->mirror_num = *mirror_nump;
1350		l = btrfsic_block_link_hashtable_lookup(
1351				next_block_ctx->dev->bdev,
1352				next_block_ctx->dev_bytenr,
1353				block_ctx->dev->bdev,
1354				block_ctx->dev_bytenr,
1355				&state->block_link_hashtable);
1356	}
1357
1358	next_block->disk_key = *disk_key;
1359	if (NULL == l) {
1360		l = btrfsic_block_link_alloc();
1361		if (NULL == l) {
1362			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
1363			btrfsic_release_block_ctx(next_block_ctx);
1364			*next_blockp = NULL;
1365			return -1;
1366		}
1367
1368		did_alloc_block_link = 1;
1369		l->block_ref_to = next_block;
1370		l->block_ref_from = block;
1371		l->ref_cnt = 1;
1372		l->parent_generation = parent_generation;
1373
1374		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1375			btrfsic_print_add_link(state, l);
1376
1377		list_add(&l->node_ref_to, &block->ref_to_list);
1378		list_add(&l->node_ref_from, &next_block->ref_from_list);
1379
1380		btrfsic_block_link_hashtable_add(l,
1381						 &state->block_link_hashtable);
1382	} else {
1383		did_alloc_block_link = 0;
1384		if (0 == limit_nesting) {
1385			l->ref_cnt++;
1386			l->parent_generation = parent_generation;
1387			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1388				btrfsic_print_add_link(state, l);
1389		}
1390	}
1391
1392	if (limit_nesting > 0 && did_alloc_block_link) {
1393		ret = btrfsic_read_block(state, next_block_ctx);
1394		if (ret < (int)next_block_ctx->len) {
1395			printk(KERN_INFO
1396			       "btrfsic: read block @logical %llu failed!\n",
1397			       next_bytenr);
1398			btrfsic_release_block_ctx(next_block_ctx);
1399			*next_blockp = NULL;
1400			return -1;
1401		}
1402
1403		*next_blockp = next_block;
1404	} else {
1405		*next_blockp = NULL;
1406	}
1407	(*mirror_nump)++;
1408
1409	return 0;
1410}
1411
1412static int btrfsic_handle_extent_data(
1413		struct btrfsic_state *state,
1414		struct btrfsic_block *block,
1415		struct btrfsic_block_data_ctx *block_ctx,
1416		u32 item_offset, int force_iodone_flag)
1417{
1418	int ret;
1419	struct btrfs_file_extent_item file_extent_item;
1420	u64 file_extent_item_offset;
1421	u64 next_bytenr;
1422	u64 num_bytes;
1423	u64 generation;
1424	struct btrfsic_block_link *l;
1425
1426	file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1427				  item_offset;
1428	if (file_extent_item_offset +
1429	    offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1430	    block_ctx->len) {
1431		printk(KERN_INFO
1432		       "btrfsic: file item out of bounce at logical %llu, dev %s\n",
1433		       block_ctx->start, block_ctx->dev->name);
1434		return -1;
1435	}
1436
1437	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1438		file_extent_item_offset,
1439		offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1440	if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1441	    btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1442		if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1443			printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu\n",
1444			       file_extent_item.type,
1445			       btrfs_stack_file_extent_disk_bytenr(
1446			       &file_extent_item));
1447		return 0;
1448	}
1449
1450	if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1451	    block_ctx->len) {
1452		printk(KERN_INFO
1453		       "btrfsic: file item out of bounce at logical %llu, dev %s\n",
1454		       block_ctx->start, block_ctx->dev->name);
1455		return -1;
1456	}
1457	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1458				     file_extent_item_offset,
1459				     sizeof(struct btrfs_file_extent_item));
1460	next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item) +
1461		      btrfs_stack_file_extent_offset(&file_extent_item);
1462	generation = btrfs_stack_file_extent_generation(&file_extent_item);
1463	num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1464	generation = btrfs_stack_file_extent_generation(&file_extent_item);
1465
1466	if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1467		printk(KERN_INFO "extent_data: type %u, disk_bytenr = %llu,"
1468		       " offset = %llu, num_bytes = %llu\n",
1469		       file_extent_item.type,
1470		       btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1471		       btrfs_stack_file_extent_offset(&file_extent_item),
1472		       num_bytes);
1473	while (num_bytes > 0) {
1474		u32 chunk_len;
1475		int num_copies;
1476		int mirror_num;
1477
1478		if (num_bytes > state->datablock_size)
1479			chunk_len = state->datablock_size;
1480		else
1481			chunk_len = num_bytes;
1482
1483		num_copies =
1484		    btrfs_num_copies(state->root->fs_info,
1485				     next_bytenr, state->datablock_size);
1486		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1487			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
1488			       next_bytenr, num_copies);
1489		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1490			struct btrfsic_block_data_ctx next_block_ctx;
1491			struct btrfsic_block *next_block;
1492			int block_was_created;
1493
1494			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1495				printk(KERN_INFO "btrfsic_handle_extent_data("
1496				       "mirror_num=%d)\n", mirror_num);
1497			if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1498				printk(KERN_INFO
1499				       "\tdisk_bytenr = %llu, num_bytes %u\n",
1500				       next_bytenr, chunk_len);
1501			ret = btrfsic_map_block(state, next_bytenr,
1502						chunk_len, &next_block_ctx,
1503						mirror_num);
1504			if (ret) {
1505				printk(KERN_INFO
1506				       "btrfsic: btrfsic_map_block(@%llu,"
1507				       " mirror=%d) failed!\n",
1508				       next_bytenr, mirror_num);
1509				return -1;
1510			}
1511
1512			next_block = btrfsic_block_lookup_or_add(
1513					state,
1514					&next_block_ctx,
1515					"referenced ",
1516					0,
1517					force_iodone_flag,
1518					!force_iodone_flag,
1519					mirror_num,
1520					&block_was_created);
1521			if (NULL == next_block) {
1522				printk(KERN_INFO
1523				       "btrfsic: error, kmalloc failed!\n");
1524				btrfsic_release_block_ctx(&next_block_ctx);
1525				return -1;
1526			}
1527			if (!block_was_created) {
1528				if (next_block->logical_bytenr != next_bytenr &&
1529				    !(!next_block->is_metadata &&
1530				      0 == next_block->logical_bytenr)) {
1531					printk(KERN_INFO
1532					       "Referenced block"
1533					       " @%llu (%s/%llu/%d)"
1534					       " found in hash table, D,"
1535					       " bytenr mismatch"
1536					       " (!= stored %llu).\n",
1537					       next_bytenr,
1538					       next_block_ctx.dev->name,
1539					       next_block_ctx.dev_bytenr,
1540					       mirror_num,
1541					       next_block->logical_bytenr);
1542				}
1543				next_block->logical_bytenr = next_bytenr;
1544				next_block->mirror_num = mirror_num;
1545			}
1546
1547			l = btrfsic_block_link_lookup_or_add(state,
1548							     &next_block_ctx,
1549							     next_block, block,
1550							     generation);
1551			btrfsic_release_block_ctx(&next_block_ctx);
1552			if (NULL == l)
1553				return -1;
1554		}
1555
1556		next_bytenr += chunk_len;
1557		num_bytes -= chunk_len;
1558	}
1559
1560	return 0;
1561}
1562
1563static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1564			     struct btrfsic_block_data_ctx *block_ctx_out,
1565			     int mirror_num)
1566{
1567	int ret;
1568	u64 length;
1569	struct btrfs_bio *multi = NULL;
1570	struct btrfs_device *device;
1571
1572	length = len;
1573	ret = btrfs_map_block(state->root->fs_info, READ,
1574			      bytenr, &length, &multi, mirror_num);
1575
1576	if (ret) {
1577		block_ctx_out->start = 0;
1578		block_ctx_out->dev_bytenr = 0;
1579		block_ctx_out->len = 0;
1580		block_ctx_out->dev = NULL;
1581		block_ctx_out->datav = NULL;
1582		block_ctx_out->pagev = NULL;
1583		block_ctx_out->mem_to_free = NULL;
1584
1585		return ret;
1586	}
1587
1588	device = multi->stripes[0].dev;
1589	block_ctx_out->dev = btrfsic_dev_state_lookup(device->bdev);
1590	block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1591	block_ctx_out->start = bytenr;
1592	block_ctx_out->len = len;
1593	block_ctx_out->datav = NULL;
1594	block_ctx_out->pagev = NULL;
1595	block_ctx_out->mem_to_free = NULL;
1596
1597	kfree(multi);
1598	if (NULL == block_ctx_out->dev) {
1599		ret = -ENXIO;
1600		printk(KERN_INFO "btrfsic: error, cannot lookup dev (#1)!\n");
1601	}
1602
1603	return ret;
1604}
1605
1606static int btrfsic_map_superblock(struct btrfsic_state *state, u64 bytenr,
1607				  u32 len, struct block_device *bdev,
1608				  struct btrfsic_block_data_ctx *block_ctx_out)
1609{
1610	block_ctx_out->dev = btrfsic_dev_state_lookup(bdev);
1611	block_ctx_out->dev_bytenr = bytenr;
1612	block_ctx_out->start = bytenr;
1613	block_ctx_out->len = len;
1614	block_ctx_out->datav = NULL;
1615	block_ctx_out->pagev = NULL;
1616	block_ctx_out->mem_to_free = NULL;
1617	if (NULL != block_ctx_out->dev) {
1618		return 0;
1619	} else {
1620		printk(KERN_INFO "btrfsic: error, cannot lookup dev (#2)!\n");
1621		return -ENXIO;
1622	}
1623}
1624
1625static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1626{
1627	if (block_ctx->mem_to_free) {
1628		unsigned int num_pages;
1629
1630		BUG_ON(!block_ctx->datav);
1631		BUG_ON(!block_ctx->pagev);
1632		num_pages = (block_ctx->len + (u64)PAGE_CACHE_SIZE - 1) >>
1633			    PAGE_CACHE_SHIFT;
1634		while (num_pages > 0) {
1635			num_pages--;
1636			if (block_ctx->datav[num_pages]) {
1637				kunmap(block_ctx->pagev[num_pages]);
1638				block_ctx->datav[num_pages] = NULL;
1639			}
1640			if (block_ctx->pagev[num_pages]) {
1641				__free_page(block_ctx->pagev[num_pages]);
1642				block_ctx->pagev[num_pages] = NULL;
1643			}
1644		}
1645
1646		kfree(block_ctx->mem_to_free);
1647		block_ctx->mem_to_free = NULL;
1648		block_ctx->pagev = NULL;
1649		block_ctx->datav = NULL;
1650	}
1651}
1652
1653static int btrfsic_read_block(struct btrfsic_state *state,
1654			      struct btrfsic_block_data_ctx *block_ctx)
1655{
1656	unsigned int num_pages;
1657	unsigned int i;
1658	u64 dev_bytenr;
1659	int ret;
1660
1661	BUG_ON(block_ctx->datav);
1662	BUG_ON(block_ctx->pagev);
1663	BUG_ON(block_ctx->mem_to_free);
1664	if (block_ctx->dev_bytenr & ((u64)PAGE_CACHE_SIZE - 1)) {
1665		printk(KERN_INFO
1666		       "btrfsic: read_block() with unaligned bytenr %llu\n",
1667		       block_ctx->dev_bytenr);
1668		return -1;
1669	}
1670
1671	num_pages = (block_ctx->len + (u64)PAGE_CACHE_SIZE - 1) >>
1672		    PAGE_CACHE_SHIFT;
1673	block_ctx->mem_to_free = kzalloc((sizeof(*block_ctx->datav) +
1674					  sizeof(*block_ctx->pagev)) *
1675					 num_pages, GFP_NOFS);
1676	if (!block_ctx->mem_to_free)
1677		return -1;
1678	block_ctx->datav = block_ctx->mem_to_free;
1679	block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1680	for (i = 0; i < num_pages; i++) {
1681		block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1682		if (!block_ctx->pagev[i])
1683			return -1;
1684	}
1685
1686	dev_bytenr = block_ctx->dev_bytenr;
1687	for (i = 0; i < num_pages;) {
1688		struct bio *bio;
1689		unsigned int j;
1690		DECLARE_COMPLETION_ONSTACK(complete);
1691
1692		bio = btrfs_io_bio_alloc(GFP_NOFS, num_pages - i);
1693		if (!bio) {
1694			printk(KERN_INFO
1695			       "btrfsic: bio_alloc() for %u pages failed!\n",
1696			       num_pages - i);
1697			return -1;
1698		}
1699		bio->bi_bdev = block_ctx->dev->bdev;
1700		bio->bi_sector = dev_bytenr >> 9;
1701		bio->bi_end_io = btrfsic_complete_bio_end_io;
1702		bio->bi_private = &complete;
1703
1704		for (j = i; j < num_pages; j++) {
1705			ret = bio_add_page(bio, block_ctx->pagev[j],
1706					   PAGE_CACHE_SIZE, 0);
1707			if (PAGE_CACHE_SIZE != ret)
1708				break;
1709		}
1710		if (j == i) {
1711			printk(KERN_INFO
1712			       "btrfsic: error, failed to add a single page!\n");
1713			return -1;
1714		}
1715		submit_bio(READ, bio);
1716
1717		/* this will also unplug the queue */
1718		wait_for_completion(&complete);
1719
1720		if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1721			printk(KERN_INFO
1722			       "btrfsic: read error at logical %llu dev %s!\n",
1723			       block_ctx->start, block_ctx->dev->name);
1724			bio_put(bio);
1725			return -1;
1726		}
1727		bio_put(bio);
1728		dev_bytenr += (j - i) * PAGE_CACHE_SIZE;
1729		i = j;
1730	}
1731	for (i = 0; i < num_pages; i++) {
1732		block_ctx->datav[i] = kmap(block_ctx->pagev[i]);
1733		if (!block_ctx->datav[i]) {
1734			printk(KERN_INFO "btrfsic: kmap() failed (dev %s)!\n",
1735			       block_ctx->dev->name);
1736			return -1;
1737		}
1738	}
1739
1740	return block_ctx->len;
1741}
1742
1743static void btrfsic_complete_bio_end_io(struct bio *bio, int err)
1744{
1745	complete((struct completion *)bio->bi_private);
1746}
1747
1748static void btrfsic_dump_database(struct btrfsic_state *state)
1749{
1750	struct list_head *elem_all;
1751
1752	BUG_ON(NULL == state);
1753
1754	printk(KERN_INFO "all_blocks_list:\n");
1755	list_for_each(elem_all, &state->all_blocks_list) {
1756		const struct btrfsic_block *const b_all =
1757		    list_entry(elem_all, struct btrfsic_block,
1758			       all_blocks_node);
1759		struct list_head *elem_ref_to;
1760		struct list_head *elem_ref_from;
1761
1762		printk(KERN_INFO "%c-block @%llu (%s/%llu/%d)\n",
1763		       btrfsic_get_block_type(state, b_all),
1764		       b_all->logical_bytenr, b_all->dev_state->name,
1765		       b_all->dev_bytenr, b_all->mirror_num);
1766
1767		list_for_each(elem_ref_to, &b_all->ref_to_list) {
1768			const struct btrfsic_block_link *const l =
1769			    list_entry(elem_ref_to,
1770				       struct btrfsic_block_link,
1771				       node_ref_to);
1772
1773			printk(KERN_INFO " %c @%llu (%s/%llu/%d)"
1774			       " refers %u* to"
1775			       " %c @%llu (%s/%llu/%d)\n",
1776			       btrfsic_get_block_type(state, b_all),
1777			       b_all->logical_bytenr, b_all->dev_state->name,
1778			       b_all->dev_bytenr, b_all->mirror_num,
1779			       l->ref_cnt,
1780			       btrfsic_get_block_type(state, l->block_ref_to),
1781			       l->block_ref_to->logical_bytenr,
1782			       l->block_ref_to->dev_state->name,
1783			       l->block_ref_to->dev_bytenr,
1784			       l->block_ref_to->mirror_num);
1785		}
1786
1787		list_for_each(elem_ref_from, &b_all->ref_from_list) {
1788			const struct btrfsic_block_link *const l =
1789			    list_entry(elem_ref_from,
1790				       struct btrfsic_block_link,
1791				       node_ref_from);
1792
1793			printk(KERN_INFO " %c @%llu (%s/%llu/%d)"
1794			       " is ref %u* from"
1795			       " %c @%llu (%s/%llu/%d)\n",
1796			       btrfsic_get_block_type(state, b_all),
1797			       b_all->logical_bytenr, b_all->dev_state->name,
1798			       b_all->dev_bytenr, b_all->mirror_num,
1799			       l->ref_cnt,
1800			       btrfsic_get_block_type(state, l->block_ref_from),
1801			       l->block_ref_from->logical_bytenr,
1802			       l->block_ref_from->dev_state->name,
1803			       l->block_ref_from->dev_bytenr,
1804			       l->block_ref_from->mirror_num);
1805		}
1806
1807		printk(KERN_INFO "\n");
1808	}
1809}
1810
1811/*
1812 * Test whether the disk block contains a tree block (leaf or node)
1813 * (note that this test fails for the super block)
1814 */
1815static int btrfsic_test_for_metadata(struct btrfsic_state *state,
1816				     char **datav, unsigned int num_pages)
1817{
1818	struct btrfs_header *h;
1819	u8 csum[BTRFS_CSUM_SIZE];
1820	u32 crc = ~(u32)0;
1821	unsigned int i;
1822
1823	if (num_pages * PAGE_CACHE_SIZE < state->metablock_size)
1824		return 1; /* not metadata */
1825	num_pages = state->metablock_size >> PAGE_CACHE_SHIFT;
1826	h = (struct btrfs_header *)datav[0];
1827
1828	if (memcmp(h->fsid, state->root->fs_info->fsid, BTRFS_UUID_SIZE))
1829		return 1;
1830
1831	for (i = 0; i < num_pages; i++) {
1832		u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1833		size_t sublen = i ? PAGE_CACHE_SIZE :
1834				    (PAGE_CACHE_SIZE - BTRFS_CSUM_SIZE);
1835
1836		crc = crc32c(crc, data, sublen);
1837	}
1838	btrfs_csum_final(crc, csum);
1839	if (memcmp(csum, h->csum, state->csum_size))
1840		return 1;
1841
1842	return 0; /* is metadata */
1843}
1844
1845static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1846					  u64 dev_bytenr, char **mapped_datav,
1847					  unsigned int num_pages,
1848					  struct bio *bio, int *bio_is_patched,
1849					  struct buffer_head *bh,
1850					  int submit_bio_bh_rw)
1851{
1852	int is_metadata;
1853	struct btrfsic_block *block;
1854	struct btrfsic_block_data_ctx block_ctx;
1855	int ret;
1856	struct btrfsic_state *state = dev_state->state;
1857	struct block_device *bdev = dev_state->bdev;
1858	unsigned int processed_len;
1859
1860	if (NULL != bio_is_patched)
1861		*bio_is_patched = 0;
1862
1863again:
1864	if (num_pages == 0)
1865		return;
1866
1867	processed_len = 0;
1868	is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1869						      num_pages));
1870
1871	block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1872					       &state->block_hashtable);
1873	if (NULL != block) {
1874		u64 bytenr = 0;
1875		struct list_head *elem_ref_to;
1876		struct list_head *tmp_ref_to;
1877
1878		if (block->is_superblock) {
1879			bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1880						    mapped_datav[0]);
1881			if (num_pages * PAGE_CACHE_SIZE <
1882			    BTRFS_SUPER_INFO_SIZE) {
1883				printk(KERN_INFO
1884				       "btrfsic: cannot work with too short bios!\n");
1885				return;
1886			}
1887			is_metadata = 1;
1888			BUG_ON(BTRFS_SUPER_INFO_SIZE & (PAGE_CACHE_SIZE - 1));
1889			processed_len = BTRFS_SUPER_INFO_SIZE;
1890			if (state->print_mask &
1891			    BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1892				printk(KERN_INFO
1893				       "[before new superblock is written]:\n");
1894				btrfsic_dump_tree_sub(state, block, 0);
1895			}
1896		}
1897		if (is_metadata) {
1898			if (!block->is_superblock) {
1899				if (num_pages * PAGE_CACHE_SIZE <
1900				    state->metablock_size) {
1901					printk(KERN_INFO
1902					       "btrfsic: cannot work with too short bios!\n");
1903					return;
1904				}
1905				processed_len = state->metablock_size;
1906				bytenr = btrfs_stack_header_bytenr(
1907						(struct btrfs_header *)
1908						mapped_datav[0]);
1909				btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1910							       dev_state,
1911							       dev_bytenr);
1912			}
1913			if (block->logical_bytenr != bytenr &&
1914			    !(!block->is_metadata &&
1915			      block->logical_bytenr == 0))
1916				printk(KERN_INFO
1917				       "Written block @%llu (%s/%llu/%d)"
1918				       " found in hash table, %c,"
1919				       " bytenr mismatch"
1920				       " (!= stored %llu).\n",
1921				       bytenr, dev_state->name, dev_bytenr,
1922				       block->mirror_num,
1923				       btrfsic_get_block_type(state, block),
1924				       block->logical_bytenr);
1925			else if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1926				printk(KERN_INFO
1927				       "Written block @%llu (%s/%llu/%d)"
1928				       " found in hash table, %c.\n",
1929				       bytenr, dev_state->name, dev_bytenr,
1930				       block->mirror_num,
1931				       btrfsic_get_block_type(state, block));
1932			block->logical_bytenr = bytenr;
1933		} else {
1934			if (num_pages * PAGE_CACHE_SIZE <
1935			    state->datablock_size) {
1936				printk(KERN_INFO
1937				       "btrfsic: cannot work with too short bios!\n");
1938				return;
1939			}
1940			processed_len = state->datablock_size;
1941			bytenr = block->logical_bytenr;
1942			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1943				printk(KERN_INFO
1944				       "Written block @%llu (%s/%llu/%d)"
1945				       " found in hash table, %c.\n",
1946				       bytenr, dev_state->name, dev_bytenr,
1947				       block->mirror_num,
1948				       btrfsic_get_block_type(state, block));
1949		}
1950
1951		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1952			printk(KERN_INFO
1953			       "ref_to_list: %cE, ref_from_list: %cE\n",
1954			       list_empty(&block->ref_to_list) ? ' ' : '!',
1955			       list_empty(&block->ref_from_list) ? ' ' : '!');
1956		if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1957			printk(KERN_INFO "btrfs: attempt to overwrite %c-block"
1958			       " @%llu (%s/%llu/%d), old(gen=%llu,"
1959			       " objectid=%llu, type=%d, offset=%llu),"
1960			       " new(gen=%llu),"
1961			       " which is referenced by most recent superblock"
1962			       " (superblockgen=%llu)!\n",
1963			       btrfsic_get_block_type(state, block), bytenr,
1964			       dev_state->name, dev_bytenr, block->mirror_num,
1965			       block->generation,
1966			       btrfs_disk_key_objectid(&block->disk_key),
1967			       block->disk_key.type,
1968			       btrfs_disk_key_offset(&block->disk_key),
1969			       btrfs_stack_header_generation(
1970				       (struct btrfs_header *) mapped_datav[0]),
1971			       state->max_superblock_generation);
1972			btrfsic_dump_tree(state);
1973		}
1974
1975		if (!block->is_iodone && !block->never_written) {
1976			printk(KERN_INFO "btrfs: attempt to overwrite %c-block"
1977			       " @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu,"
1978			       " which is not yet iodone!\n",
1979			       btrfsic_get_block_type(state, block), bytenr,
1980			       dev_state->name, dev_bytenr, block->mirror_num,
1981			       block->generation,
1982			       btrfs_stack_header_generation(
1983				       (struct btrfs_header *)
1984				       mapped_datav[0]));
1985			/* it would not be safe to go on */
1986			btrfsic_dump_tree(state);
1987			goto continue_loop;
1988		}
1989
1990		/*
1991		 * Clear all references of this block. Do not free
1992		 * the block itself even if is not referenced anymore
1993		 * because it still carries valueable information
1994		 * like whether it was ever written and IO completed.
1995		 */
1996		list_for_each_safe(elem_ref_to, tmp_ref_to,
1997				   &block->ref_to_list) {
1998			struct btrfsic_block_link *const l =
1999			    list_entry(elem_ref_to,
2000				       struct btrfsic_block_link,
2001				       node_ref_to);
2002
2003			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2004				btrfsic_print_rem_link(state, l);
2005			l->ref_cnt--;
2006			if (0 == l->ref_cnt) {
2007				list_del(&l->node_ref_to);
2008				list_del(&l->node_ref_from);
2009				btrfsic_block_link_hashtable_remove(l);
2010				btrfsic_block_link_free(l);
2011			}
2012		}
2013
2014		if (block->is_superblock)
2015			ret = btrfsic_map_superblock(state, bytenr,
2016						     processed_len,
2017						     bdev, &block_ctx);
2018		else
2019			ret = btrfsic_map_block(state, bytenr, processed_len,
2020						&block_ctx, 0);
2021		if (ret) {
2022			printk(KERN_INFO
2023			       "btrfsic: btrfsic_map_block(root @%llu)"
2024			       " failed!\n", bytenr);
2025			goto continue_loop;
2026		}
2027		block_ctx.datav = mapped_datav;
2028		/* the following is required in case of writes to mirrors,
2029		 * use the same that was used for the lookup */
2030		block_ctx.dev = dev_state;
2031		block_ctx.dev_bytenr = dev_bytenr;
2032
2033		if (is_metadata || state->include_extent_data) {
2034			block->never_written = 0;
2035			block->iodone_w_error = 0;
2036			if (NULL != bio) {
2037				block->is_iodone = 0;
2038				BUG_ON(NULL == bio_is_patched);
2039				if (!*bio_is_patched) {
2040					block->orig_bio_bh_private =
2041					    bio->bi_private;
2042					block->orig_bio_bh_end_io.bio =
2043					    bio->bi_end_io;
2044					block->next_in_same_bio = NULL;
2045					bio->bi_private = block;
2046					bio->bi_end_io = btrfsic_bio_end_io;
2047					*bio_is_patched = 1;
2048				} else {
2049					struct btrfsic_block *chained_block =
2050					    (struct btrfsic_block *)
2051					    bio->bi_private;
2052
2053					BUG_ON(NULL == chained_block);
2054					block->orig_bio_bh_private =
2055					    chained_block->orig_bio_bh_private;
2056					block->orig_bio_bh_end_io.bio =
2057					    chained_block->orig_bio_bh_end_io.
2058					    bio;
2059					block->next_in_same_bio = chained_block;
2060					bio->bi_private = block;
2061				}
2062			} else if (NULL != bh) {
2063				block->is_iodone = 0;
2064				block->orig_bio_bh_private = bh->b_private;
2065				block->orig_bio_bh_end_io.bh = bh->b_end_io;
2066				block->next_in_same_bio = NULL;
2067				bh->b_private = block;
2068				bh->b_end_io = btrfsic_bh_end_io;
2069			} else {
2070				block->is_iodone = 1;
2071				block->orig_bio_bh_private = NULL;
2072				block->orig_bio_bh_end_io.bio = NULL;
2073				block->next_in_same_bio = NULL;
2074			}
2075		}
2076
2077		block->flush_gen = dev_state->last_flush_gen + 1;
2078		block->submit_bio_bh_rw = submit_bio_bh_rw;
2079		if (is_metadata) {
2080			block->logical_bytenr = bytenr;
2081			block->is_metadata = 1;
2082			if (block->is_superblock) {
2083				BUG_ON(PAGE_CACHE_SIZE !=
2084				       BTRFS_SUPER_INFO_SIZE);
2085				ret = btrfsic_process_written_superblock(
2086						state,
2087						block,
2088						(struct btrfs_super_block *)
2089						mapped_datav[0]);
2090				if (state->print_mask &
2091				    BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
2092					printk(KERN_INFO
2093					"[after new superblock is written]:\n");
2094					btrfsic_dump_tree_sub(state, block, 0);
2095				}
2096			} else {
2097				block->mirror_num = 0;	/* unknown */
2098				ret = btrfsic_process_metablock(
2099						state,
2100						block,
2101						&block_ctx,
2102						0, 0);
2103			}
2104			if (ret)
2105				printk(KERN_INFO
2106				       "btrfsic: btrfsic_process_metablock"
2107				       "(root @%llu) failed!\n",
2108				       dev_bytenr);
2109		} else {
2110			block->is_metadata = 0;
2111			block->mirror_num = 0;	/* unknown */
2112			block->generation = BTRFSIC_GENERATION_UNKNOWN;
2113			if (!state->include_extent_data
2114			    && list_empty(&block->ref_from_list)) {
2115				/*
2116				 * disk block is overwritten with extent
2117				 * data (not meta data) and we are configured
2118				 * to not include extent data: take the
2119				 * chance and free the block's memory
2120				 */
2121				btrfsic_block_hashtable_remove(block);
2122				list_del(&block->all_blocks_node);
2123				btrfsic_block_free(block);
2124			}
2125		}
2126		btrfsic_release_block_ctx(&block_ctx);
2127	} else {
2128		/* block has not been found in hash table */
2129		u64 bytenr;
2130
2131		if (!is_metadata) {
2132			processed_len = state->datablock_size;
2133			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2134				printk(KERN_INFO "Written block (%s/%llu/?)"
2135				       " !found in hash table, D.\n",
2136				       dev_state->name, dev_bytenr);
2137			if (!state->include_extent_data) {
2138				/* ignore that written D block */
2139				goto continue_loop;
2140			}
2141
2142			/* this is getting ugly for the
2143			 * include_extent_data case... */
2144			bytenr = 0;	/* unknown */
2145			block_ctx.start = bytenr;
2146			block_ctx.len = processed_len;
2147			block_ctx.mem_to_free = NULL;
2148			block_ctx.pagev = NULL;
2149		} else {
2150			processed_len = state->metablock_size;
2151			bytenr = btrfs_stack_header_bytenr(
2152					(struct btrfs_header *)
2153					mapped_datav[0]);
2154			btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
2155						       dev_bytenr);
2156			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2157				printk(KERN_INFO
2158				       "Written block @%llu (%s/%llu/?)"
2159				       " !found in hash table, M.\n",
2160				       bytenr, dev_state->name, dev_bytenr);
2161
2162			ret = btrfsic_map_block(state, bytenr, processed_len,
2163						&block_ctx, 0);
2164			if (ret) {
2165				printk(KERN_INFO
2166				       "btrfsic: btrfsic_map_block(root @%llu)"
2167				       " failed!\n",
2168				       dev_bytenr);
2169				goto continue_loop;
2170			}
2171		}
2172		block_ctx.datav = mapped_datav;
2173		/* the following is required in case of writes to mirrors,
2174		 * use the same that was used for the lookup */
2175		block_ctx.dev = dev_state;
2176		block_ctx.dev_bytenr = dev_bytenr;
2177
2178		block = btrfsic_block_alloc();
2179		if (NULL == block) {
2180			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
2181			btrfsic_release_block_ctx(&block_ctx);
2182			goto continue_loop;
2183		}
2184		block->dev_state = dev_state;
2185		block->dev_bytenr = dev_bytenr;
2186		block->logical_bytenr = bytenr;
2187		block->is_metadata = is_metadata;
2188		block->never_written = 0;
2189		block->iodone_w_error = 0;
2190		block->mirror_num = 0;	/* unknown */
2191		block->flush_gen = dev_state->last_flush_gen + 1;
2192		block->submit_bio_bh_rw = submit_bio_bh_rw;
2193		if (NULL != bio) {
2194			block->is_iodone = 0;
2195			BUG_ON(NULL == bio_is_patched);
2196			if (!*bio_is_patched) {
2197				block->orig_bio_bh_private = bio->bi_private;
2198				block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2199				block->next_in_same_bio = NULL;
2200				bio->bi_private = block;
2201				bio->bi_end_io = btrfsic_bio_end_io;
2202				*bio_is_patched = 1;
2203			} else {
2204				struct btrfsic_block *chained_block =
2205				    (struct btrfsic_block *)
2206				    bio->bi_private;
2207
2208				BUG_ON(NULL == chained_block);
2209				block->orig_bio_bh_private =
2210				    chained_block->orig_bio_bh_private;
2211				block->orig_bio_bh_end_io.bio =
2212				    chained_block->orig_bio_bh_end_io.bio;
2213				block->next_in_same_bio = chained_block;
2214				bio->bi_private = block;
2215			}
2216		} else if (NULL != bh) {
2217			block->is_iodone = 0;
2218			block->orig_bio_bh_private = bh->b_private;
2219			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2220			block->next_in_same_bio = NULL;
2221			bh->b_private = block;
2222			bh->b_end_io = btrfsic_bh_end_io;
2223		} else {
2224			block->is_iodone = 1;
2225			block->orig_bio_bh_private = NULL;
2226			block->orig_bio_bh_end_io.bio = NULL;
2227			block->next_in_same_bio = NULL;
2228		}
2229		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2230			printk(KERN_INFO
2231			       "New written %c-block @%llu (%s/%llu/%d)\n",
2232			       is_metadata ? 'M' : 'D',
2233			       block->logical_bytenr, block->dev_state->name,
2234			       block->dev_bytenr, block->mirror_num);
2235		list_add(&block->all_blocks_node, &state->all_blocks_list);
2236		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2237
2238		if (is_metadata) {
2239			ret = btrfsic_process_metablock(state, block,
2240							&block_ctx, 0, 0);
2241			if (ret)
2242				printk(KERN_INFO
2243				       "btrfsic: process_metablock(root @%llu)"
2244				       " failed!\n",
2245				       dev_bytenr);
2246		}
2247		btrfsic_release_block_ctx(&block_ctx);
2248	}
2249
2250continue_loop:
2251	BUG_ON(!processed_len);
2252	dev_bytenr += processed_len;
2253	mapped_datav += processed_len >> PAGE_CACHE_SHIFT;
2254	num_pages -= processed_len >> PAGE_CACHE_SHIFT;
2255	goto again;
2256}
2257
2258static void btrfsic_bio_end_io(struct bio *bp, int bio_error_status)
2259{
2260	struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2261	int iodone_w_error;
2262
2263	/* mutex is not held! This is not save if IO is not yet completed
2264	 * on umount */
2265	iodone_w_error = 0;
2266	if (bio_error_status)
2267		iodone_w_error = 1;
2268
2269	BUG_ON(NULL == block);
2270	bp->bi_private = block->orig_bio_bh_private;
2271	bp->bi_end_io = block->orig_bio_bh_end_io.bio;
2272
2273	do {
2274		struct btrfsic_block *next_block;
2275		struct btrfsic_dev_state *const dev_state = block->dev_state;
2276
2277		if ((dev_state->state->print_mask &
2278		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2279			printk(KERN_INFO
2280			       "bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2281			       bio_error_status,
2282			       btrfsic_get_block_type(dev_state->state, block),
2283			       block->logical_bytenr, dev_state->name,
2284			       block->dev_bytenr, block->mirror_num);
2285		next_block = block->next_in_same_bio;
2286		block->iodone_w_error = iodone_w_error;
2287		if (block->submit_bio_bh_rw & REQ_FLUSH) {
2288			dev_state->last_flush_gen++;
2289			if ((dev_state->state->print_mask &
2290			     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2291				printk(KERN_INFO
2292				       "bio_end_io() new %s flush_gen=%llu\n",
2293				       dev_state->name,
2294				       dev_state->last_flush_gen);
2295		}
2296		if (block->submit_bio_bh_rw & REQ_FUA)
2297			block->flush_gen = 0; /* FUA completed means block is
2298					       * on disk */
2299		block->is_iodone = 1; /* for FLUSH, this releases the block */
2300		block = next_block;
2301	} while (NULL != block);
2302
2303	bp->bi_end_io(bp, bio_error_status);
2304}
2305
2306static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate)
2307{
2308	struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private;
2309	int iodone_w_error = !uptodate;
2310	struct btrfsic_dev_state *dev_state;
2311
2312	BUG_ON(NULL == block);
2313	dev_state = block->dev_state;
2314	if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2315		printk(KERN_INFO
2316		       "bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n",
2317		       iodone_w_error,
2318		       btrfsic_get_block_type(dev_state->state, block),
2319		       block->logical_bytenr, block->dev_state->name,
2320		       block->dev_bytenr, block->mirror_num);
2321
2322	block->iodone_w_error = iodone_w_error;
2323	if (block->submit_bio_bh_rw & REQ_FLUSH) {
2324		dev_state->last_flush_gen++;
2325		if ((dev_state->state->print_mask &
2326		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2327			printk(KERN_INFO
2328			       "bh_end_io() new %s flush_gen=%llu\n",
2329			       dev_state->name, dev_state->last_flush_gen);
2330	}
2331	if (block->submit_bio_bh_rw & REQ_FUA)
2332		block->flush_gen = 0; /* FUA completed means block is on disk */
2333
2334	bh->b_private = block->orig_bio_bh_private;
2335	bh->b_end_io = block->orig_bio_bh_end_io.bh;
2336	block->is_iodone = 1; /* for FLUSH, this releases the block */
2337	bh->b_end_io(bh, uptodate);
2338}
2339
2340static int btrfsic_process_written_superblock(
2341		struct btrfsic_state *state,
2342		struct btrfsic_block *const superblock,
2343		struct btrfs_super_block *const super_hdr)
2344{
2345	int pass;
2346
2347	superblock->generation = btrfs_super_generation(super_hdr);
2348	if (!(superblock->generation > state->max_superblock_generation ||
2349	      0 == state->max_superblock_generation)) {
2350		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2351			printk(KERN_INFO
2352			       "btrfsic: superblock @%llu (%s/%llu/%d)"
2353			       " with old gen %llu <= %llu\n",
2354			       superblock->logical_bytenr,
2355			       superblock->dev_state->name,
2356			       superblock->dev_bytenr, superblock->mirror_num,
2357			       btrfs_super_generation(super_hdr),
2358			       state->max_superblock_generation);
2359	} else {
2360		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2361			printk(KERN_INFO
2362			       "btrfsic: got new superblock @%llu (%s/%llu/%d)"
2363			       " with new gen %llu > %llu\n",
2364			       superblock->logical_bytenr,
2365			       superblock->dev_state->name,
2366			       superblock->dev_bytenr, superblock->mirror_num,
2367			       btrfs_super_generation(super_hdr),
2368			       state->max_superblock_generation);
2369
2370		state->max_superblock_generation =
2371		    btrfs_super_generation(super_hdr);
2372		state->latest_superblock = superblock;
2373	}
2374
2375	for (pass = 0; pass < 3; pass++) {
2376		int ret;
2377		u64 next_bytenr;
2378		struct btrfsic_block *next_block;
2379		struct btrfsic_block_data_ctx tmp_next_block_ctx;
2380		struct btrfsic_block_link *l;
2381		int num_copies;
2382		int mirror_num;
2383		const char *additional_string = NULL;
2384		struct btrfs_disk_key tmp_disk_key = {0};
2385
2386		btrfs_set_disk_key_objectid(&tmp_disk_key,
2387					    BTRFS_ROOT_ITEM_KEY);
2388		btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2389
2390		switch (pass) {
2391		case 0:
2392			btrfs_set_disk_key_objectid(&tmp_disk_key,
2393						    BTRFS_ROOT_TREE_OBJECTID);
2394			additional_string = "root ";
2395			next_bytenr = btrfs_super_root(super_hdr);
2396			if (state->print_mask &
2397			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2398				printk(KERN_INFO "root@%llu\n", next_bytenr);
2399			break;
2400		case 1:
2401			btrfs_set_disk_key_objectid(&tmp_disk_key,
2402						    BTRFS_CHUNK_TREE_OBJECTID);
2403			additional_string = "chunk ";
2404			next_bytenr = btrfs_super_chunk_root(super_hdr);
2405			if (state->print_mask &
2406			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2407				printk(KERN_INFO "chunk@%llu\n", next_bytenr);
2408			break;
2409		case 2:
2410			btrfs_set_disk_key_objectid(&tmp_disk_key,
2411						    BTRFS_TREE_LOG_OBJECTID);
2412			additional_string = "log ";
2413			next_bytenr = btrfs_super_log_root(super_hdr);
2414			if (0 == next_bytenr)
2415				continue;
2416			if (state->print_mask &
2417			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2418				printk(KERN_INFO "log@%llu\n", next_bytenr);
2419			break;
2420		}
2421
2422		num_copies =
2423		    btrfs_num_copies(state->root->fs_info,
2424				     next_bytenr, BTRFS_SUPER_INFO_SIZE);
2425		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2426			printk(KERN_INFO "num_copies(log_bytenr=%llu) = %d\n",
2427			       next_bytenr, num_copies);
2428		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2429			int was_created;
2430
2431			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2432				printk(KERN_INFO
2433				       "btrfsic_process_written_superblock("
2434				       "mirror_num=%d)\n", mirror_num);
2435			ret = btrfsic_map_block(state, next_bytenr,
2436						BTRFS_SUPER_INFO_SIZE,
2437						&tmp_next_block_ctx,
2438						mirror_num);
2439			if (ret) {
2440				printk(KERN_INFO
2441				       "btrfsic: btrfsic_map_block(@%llu,"
2442				       " mirror=%d) failed!\n",
2443				       next_bytenr, mirror_num);
2444				return -1;
2445			}
2446
2447			next_block = btrfsic_block_lookup_or_add(
2448					state,
2449					&tmp_next_block_ctx,
2450					additional_string,
2451					1, 0, 1,
2452					mirror_num,
2453					&was_created);
2454			if (NULL == next_block) {
2455				printk(KERN_INFO
2456				       "btrfsic: error, kmalloc failed!\n");
2457				btrfsic_release_block_ctx(&tmp_next_block_ctx);
2458				return -1;
2459			}
2460
2461			next_block->disk_key = tmp_disk_key;
2462			if (was_created)
2463				next_block->generation =
2464				    BTRFSIC_GENERATION_UNKNOWN;
2465			l = btrfsic_block_link_lookup_or_add(
2466					state,
2467					&tmp_next_block_ctx,
2468					next_block,
2469					superblock,
2470					BTRFSIC_GENERATION_UNKNOWN);
2471			btrfsic_release_block_ctx(&tmp_next_block_ctx);
2472			if (NULL == l)
2473				return -1;
2474		}
2475	}
2476
2477	if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2478		btrfsic_dump_tree(state);
2479
2480	return 0;
2481}
2482
2483static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2484					struct btrfsic_block *const block,
2485					int recursion_level)
2486{
2487	struct list_head *elem_ref_to;
2488	int ret = 0;
2489
2490	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2491		/*
2492		 * Note that this situation can happen and does not
2493		 * indicate an error in regular cases. It happens
2494		 * when disk blocks are freed and later reused.
2495		 * The check-integrity module is not aware of any
2496		 * block free operations, it just recognizes block
2497		 * write operations. Therefore it keeps the linkage
2498		 * information for a block until a block is
2499		 * rewritten. This can temporarily cause incorrect
2500		 * and even circular linkage informations. This
2501		 * causes no harm unless such blocks are referenced
2502		 * by the most recent super block.
2503		 */
2504		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2505			printk(KERN_INFO
2506			       "btrfsic: abort cyclic linkage (case 1).\n");
2507
2508		return ret;
2509	}
2510
2511	/*
2512	 * This algorithm is recursive because the amount of used stack
2513	 * space is very small and the max recursion depth is limited.
2514	 */
2515	list_for_each(elem_ref_to, &block->ref_to_list) {
2516		const struct btrfsic_block_link *const l =
2517		    list_entry(elem_ref_to, struct btrfsic_block_link,
2518			       node_ref_to);
2519
2520		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2521			printk(KERN_INFO
2522			       "rl=%d, %c @%llu (%s/%llu/%d)"
2523			       " %u* refers to %c @%llu (%s/%llu/%d)\n",
2524			       recursion_level,
2525			       btrfsic_get_block_type(state, block),
2526			       block->logical_bytenr, block->dev_state->name,
2527			       block->dev_bytenr, block->mirror_num,
2528			       l->ref_cnt,
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		if (l->block_ref_to->never_written) {
2535			printk(KERN_INFO "btrfs: attempt to write superblock"
2536			       " which references block %c @%llu (%s/%llu/%d)"
2537			       " which is never written!\n",
2538			       btrfsic_get_block_type(state, l->block_ref_to),
2539			       l->block_ref_to->logical_bytenr,
2540			       l->block_ref_to->dev_state->name,
2541			       l->block_ref_to->dev_bytenr,
2542			       l->block_ref_to->mirror_num);
2543			ret = -1;
2544		} else if (!l->block_ref_to->is_iodone) {
2545			printk(KERN_INFO "btrfs: attempt to write superblock"
2546			       " which references block %c @%llu (%s/%llu/%d)"
2547			       " which is not yet iodone!\n",
2548			       btrfsic_get_block_type(state, l->block_ref_to),
2549			       l->block_ref_to->logical_bytenr,
2550			       l->block_ref_to->dev_state->name,
2551			       l->block_ref_to->dev_bytenr,
2552			       l->block_ref_to->mirror_num);
2553			ret = -1;
2554		} else if (l->block_ref_to->iodone_w_error) {
2555			printk(KERN_INFO "btrfs: attempt to write superblock"
2556			       " which references block %c @%llu (%s/%llu/%d)"
2557			       " which has write error!\n",
2558			       btrfsic_get_block_type(state, l->block_ref_to),
2559			       l->block_ref_to->logical_bytenr,
2560			       l->block_ref_to->dev_state->name,
2561			       l->block_ref_to->dev_bytenr,
2562			       l->block_ref_to->mirror_num);
2563			ret = -1;
2564		} else if (l->parent_generation !=
2565			   l->block_ref_to->generation &&
2566			   BTRFSIC_GENERATION_UNKNOWN !=
2567			   l->parent_generation &&
2568			   BTRFSIC_GENERATION_UNKNOWN !=
2569			   l->block_ref_to->generation) {
2570			printk(KERN_INFO "btrfs: attempt to write superblock"
2571			       " which references block %c @%llu (%s/%llu/%d)"
2572			       " with generation %llu !="
2573			       " parent generation %llu!\n",
2574			       btrfsic_get_block_type(state, l->block_ref_to),
2575			       l->block_ref_to->logical_bytenr,
2576			       l->block_ref_to->dev_state->name,
2577			       l->block_ref_to->dev_bytenr,
2578			       l->block_ref_to->mirror_num,
2579			       l->block_ref_to->generation,
2580			       l->parent_generation);
2581			ret = -1;
2582		} else if (l->block_ref_to->flush_gen >
2583			   l->block_ref_to->dev_state->last_flush_gen) {
2584			printk(KERN_INFO "btrfs: attempt to write superblock"
2585			       " which references block %c @%llu (%s/%llu/%d)"
2586			       " which is not flushed out of disk's write cache"
2587			       " (block flush_gen=%llu,"
2588			       " dev->flush_gen=%llu)!\n",
2589			       btrfsic_get_block_type(state, l->block_ref_to),
2590			       l->block_ref_to->logical_bytenr,
2591			       l->block_ref_to->dev_state->name,
2592			       l->block_ref_to->dev_bytenr,
2593			       l->block_ref_to->mirror_num, block->flush_gen,
2594			       l->block_ref_to->dev_state->last_flush_gen);
2595			ret = -1;
2596		} else if (-1 == btrfsic_check_all_ref_blocks(state,
2597							      l->block_ref_to,
2598							      recursion_level +
2599							      1)) {
2600			ret = -1;
2601		}
2602	}
2603
2604	return ret;
2605}
2606
2607static int btrfsic_is_block_ref_by_superblock(
2608		const struct btrfsic_state *state,
2609		const struct btrfsic_block *block,
2610		int recursion_level)
2611{
2612	struct list_head *elem_ref_from;
2613
2614	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2615		/* refer to comment at "abort cyclic linkage (case 1)" */
2616		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2617			printk(KERN_INFO
2618			       "btrfsic: abort cyclic linkage (case 2).\n");
2619
2620		return 0;
2621	}
2622
2623	/*
2624	 * This algorithm is recursive because the amount of used stack space
2625	 * is very small and the max recursion depth is limited.
2626	 */
2627	list_for_each(elem_ref_from, &block->ref_from_list) {
2628		const struct btrfsic_block_link *const l =
2629		    list_entry(elem_ref_from, struct btrfsic_block_link,
2630			       node_ref_from);
2631
2632		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2633			printk(KERN_INFO
2634			       "rl=%d, %c @%llu (%s/%llu/%d)"
2635			       " is ref %u* from %c @%llu (%s/%llu/%d)\n",
2636			       recursion_level,
2637			       btrfsic_get_block_type(state, block),
2638			       block->logical_bytenr, block->dev_state->name,
2639			       block->dev_bytenr, block->mirror_num,
2640			       l->ref_cnt,
2641			       btrfsic_get_block_type(state, l->block_ref_from),
2642			       l->block_ref_from->logical_bytenr,
2643			       l->block_ref_from->dev_state->name,
2644			       l->block_ref_from->dev_bytenr,
2645			       l->block_ref_from->mirror_num);
2646		if (l->block_ref_from->is_superblock &&
2647		    state->latest_superblock->dev_bytenr ==
2648		    l->block_ref_from->dev_bytenr &&
2649		    state->latest_superblock->dev_state->bdev ==
2650		    l->block_ref_from->dev_state->bdev)
2651			return 1;
2652		else if (btrfsic_is_block_ref_by_superblock(state,
2653							    l->block_ref_from,
2654							    recursion_level +
2655							    1))
2656			return 1;
2657	}
2658
2659	return 0;
2660}
2661
2662static void btrfsic_print_add_link(const struct btrfsic_state *state,
2663				   const struct btrfsic_block_link *l)
2664{
2665	printk(KERN_INFO
2666	       "Add %u* link from %c @%llu (%s/%llu/%d)"
2667	       " to %c @%llu (%s/%llu/%d).\n",
2668	       l->ref_cnt,
2669	       btrfsic_get_block_type(state, l->block_ref_from),
2670	       l->block_ref_from->logical_bytenr,
2671	       l->block_ref_from->dev_state->name,
2672	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2673	       btrfsic_get_block_type(state, l->block_ref_to),
2674	       l->block_ref_to->logical_bytenr,
2675	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2676	       l->block_ref_to->mirror_num);
2677}
2678
2679static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2680				   const struct btrfsic_block_link *l)
2681{
2682	printk(KERN_INFO
2683	       "Rem %u* link from %c @%llu (%s/%llu/%d)"
2684	       " to %c @%llu (%s/%llu/%d).\n",
2685	       l->ref_cnt,
2686	       btrfsic_get_block_type(state, l->block_ref_from),
2687	       l->block_ref_from->logical_bytenr,
2688	       l->block_ref_from->dev_state->name,
2689	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2690	       btrfsic_get_block_type(state, l->block_ref_to),
2691	       l->block_ref_to->logical_bytenr,
2692	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2693	       l->block_ref_to->mirror_num);
2694}
2695
2696static char btrfsic_get_block_type(const struct btrfsic_state *state,
2697				   const struct btrfsic_block *block)
2698{
2699	if (block->is_superblock &&
2700	    state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2701	    state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2702		return 'S';
2703	else if (block->is_superblock)
2704		return 's';
2705	else if (block->is_metadata)
2706		return 'M';
2707	else
2708		return 'D';
2709}
2710
2711static void btrfsic_dump_tree(const struct btrfsic_state *state)
2712{
2713	btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2714}
2715
2716static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2717				  const struct btrfsic_block *block,
2718				  int indent_level)
2719{
2720	struct list_head *elem_ref_to;
2721	int indent_add;
2722	static char buf[80];
2723	int cursor_position;
2724
2725	/*
2726	 * Should better fill an on-stack buffer with a complete line and
2727	 * dump it at once when it is time to print a newline character.
2728	 */
2729
2730	/*
2731	 * This algorithm is recursive because the amount of used stack space
2732	 * is very small and the max recursion depth is limited.
2733	 */
2734	indent_add = sprintf(buf, "%c-%llu(%s/%llu/%d)",
2735			     btrfsic_get_block_type(state, block),
2736			     block->logical_bytenr, block->dev_state->name,
2737			     block->dev_bytenr, block->mirror_num);
2738	if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2739		printk("[...]\n");
2740		return;
2741	}
2742	printk(buf);
2743	indent_level += indent_add;
2744	if (list_empty(&block->ref_to_list)) {
2745		printk("\n");
2746		return;
2747	}
2748	if (block->mirror_num > 1 &&
2749	    !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2750		printk(" [...]\n");
2751		return;
2752	}
2753
2754	cursor_position = indent_level;
2755	list_for_each(elem_ref_to, &block->ref_to_list) {
2756		const struct btrfsic_block_link *const l =
2757		    list_entry(elem_ref_to, struct btrfsic_block_link,
2758			       node_ref_to);
2759
2760		while (cursor_position < indent_level) {
2761			printk(" ");
2762			cursor_position++;
2763		}
2764		if (l->ref_cnt > 1)
2765			indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2766		else
2767			indent_add = sprintf(buf, " --> ");
2768		if (indent_level + indent_add >
2769		    BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2770			printk("[...]\n");
2771			cursor_position = 0;
2772			continue;
2773		}
2774
2775		printk(buf);
2776
2777		btrfsic_dump_tree_sub(state, l->block_ref_to,
2778				      indent_level + indent_add);
2779		cursor_position = 0;
2780	}
2781}
2782
2783static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2784		struct btrfsic_state *state,
2785		struct btrfsic_block_data_ctx *next_block_ctx,
2786		struct btrfsic_block *next_block,
2787		struct btrfsic_block *from_block,
2788		u64 parent_generation)
2789{
2790	struct btrfsic_block_link *l;
2791
2792	l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2793						next_block_ctx->dev_bytenr,
2794						from_block->dev_state->bdev,
2795						from_block->dev_bytenr,
2796						&state->block_link_hashtable);
2797	if (NULL == l) {
2798		l = btrfsic_block_link_alloc();
2799		if (NULL == l) {
2800			printk(KERN_INFO
2801			       "btrfsic: error, kmalloc" " failed!\n");
2802			return NULL;
2803		}
2804
2805		l->block_ref_to = next_block;
2806		l->block_ref_from = from_block;
2807		l->ref_cnt = 1;
2808		l->parent_generation = parent_generation;
2809
2810		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2811			btrfsic_print_add_link(state, l);
2812
2813		list_add(&l->node_ref_to, &from_block->ref_to_list);
2814		list_add(&l->node_ref_from, &next_block->ref_from_list);
2815
2816		btrfsic_block_link_hashtable_add(l,
2817						 &state->block_link_hashtable);
2818	} else {
2819		l->ref_cnt++;
2820		l->parent_generation = parent_generation;
2821		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2822			btrfsic_print_add_link(state, l);
2823	}
2824
2825	return l;
2826}
2827
2828static struct btrfsic_block *btrfsic_block_lookup_or_add(
2829		struct btrfsic_state *state,
2830		struct btrfsic_block_data_ctx *block_ctx,
2831		const char *additional_string,
2832		int is_metadata,
2833		int is_iodone,
2834		int never_written,
2835		int mirror_num,
2836		int *was_created)
2837{
2838	struct btrfsic_block *block;
2839
2840	block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2841					       block_ctx->dev_bytenr,
2842					       &state->block_hashtable);
2843	if (NULL == block) {
2844		struct btrfsic_dev_state *dev_state;
2845
2846		block = btrfsic_block_alloc();
2847		if (NULL == block) {
2848			printk(KERN_INFO "btrfsic: error, kmalloc failed!\n");
2849			return NULL;
2850		}
2851		dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev);
2852		if (NULL == dev_state) {
2853			printk(KERN_INFO
2854			       "btrfsic: error, lookup dev_state failed!\n");
2855			btrfsic_block_free(block);
2856			return NULL;
2857		}
2858		block->dev_state = dev_state;
2859		block->dev_bytenr = block_ctx->dev_bytenr;
2860		block->logical_bytenr = block_ctx->start;
2861		block->is_metadata = is_metadata;
2862		block->is_iodone = is_iodone;
2863		block->never_written = never_written;
2864		block->mirror_num = mirror_num;
2865		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2866			printk(KERN_INFO
2867			       "New %s%c-block @%llu (%s/%llu/%d)\n",
2868			       additional_string,
2869			       btrfsic_get_block_type(state, block),
2870			       block->logical_bytenr, dev_state->name,
2871			       block->dev_bytenr, mirror_num);
2872		list_add(&block->all_blocks_node, &state->all_blocks_list);
2873		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2874		if (NULL != was_created)
2875			*was_created = 1;
2876	} else {
2877		if (NULL != was_created)
2878			*was_created = 0;
2879	}
2880
2881	return block;
2882}
2883
2884static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2885					   u64 bytenr,
2886					   struct btrfsic_dev_state *dev_state,
2887					   u64 dev_bytenr)
2888{
2889	int num_copies;
2890	int mirror_num;
2891	int ret;
2892	struct btrfsic_block_data_ctx block_ctx;
2893	int match = 0;
2894
2895	num_copies = btrfs_num_copies(state->root->fs_info,
2896				      bytenr, state->metablock_size);
2897
2898	for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2899		ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2900					&block_ctx, mirror_num);
2901		if (ret) {
2902			printk(KERN_INFO "btrfsic:"
2903			       " btrfsic_map_block(logical @%llu,"
2904			       " mirror %d) failed!\n",
2905			       bytenr, mirror_num);
2906			continue;
2907		}
2908
2909		if (dev_state->bdev == block_ctx.dev->bdev &&
2910		    dev_bytenr == block_ctx.dev_bytenr) {
2911			match++;
2912			btrfsic_release_block_ctx(&block_ctx);
2913			break;
2914		}
2915		btrfsic_release_block_ctx(&block_ctx);
2916	}
2917
2918	if (WARN_ON(!match)) {
2919		printk(KERN_INFO "btrfs: attempt to write M-block which contains logical bytenr that doesn't map to dev+physical bytenr of submit_bio,"
2920		       " buffer->log_bytenr=%llu, submit_bio(bdev=%s,"
2921		       " phys_bytenr=%llu)!\n",
2922		       bytenr, dev_state->name, dev_bytenr);
2923		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2924			ret = btrfsic_map_block(state, bytenr,
2925						state->metablock_size,
2926						&block_ctx, mirror_num);
2927			if (ret)
2928				continue;
2929
2930			printk(KERN_INFO "Read logical bytenr @%llu maps to"
2931			       " (%s/%llu/%d)\n",
2932			       bytenr, block_ctx.dev->name,
2933			       block_ctx.dev_bytenr, mirror_num);
2934		}
2935	}
2936}
2937
2938static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
2939		struct block_device *bdev)
2940{
2941	struct btrfsic_dev_state *ds;
2942
2943	ds = btrfsic_dev_state_hashtable_lookup(bdev,
2944						&btrfsic_dev_state_hashtable);
2945	return ds;
2946}
2947
2948int btrfsic_submit_bh(int rw, struct buffer_head *bh)
2949{
2950	struct btrfsic_dev_state *dev_state;
2951
2952	if (!btrfsic_is_initialized)
2953		return submit_bh(rw, bh);
2954
2955	mutex_lock(&btrfsic_mutex);
2956	/* since btrfsic_submit_bh() might also be called before
2957	 * btrfsic_mount(), this might return NULL */
2958	dev_state = btrfsic_dev_state_lookup(bh->b_bdev);
2959
2960	/* Only called to write the superblock (incl. FLUSH/FUA) */
2961	if (NULL != dev_state &&
2962	    (rw & WRITE) && bh->b_size > 0) {
2963		u64 dev_bytenr;
2964
2965		dev_bytenr = 4096 * bh->b_blocknr;
2966		if (dev_state->state->print_mask &
2967		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2968			printk(KERN_INFO
2969			       "submit_bh(rw=0x%x, blocknr=%llu (bytenr %llu),"
2970			       " size=%zu, data=%p, bdev=%p)\n",
2971			       rw, (unsigned long long)bh->b_blocknr,
2972			       dev_bytenr, bh->b_size, bh->b_data, bh->b_bdev);
2973		btrfsic_process_written_block(dev_state, dev_bytenr,
2974					      &bh->b_data, 1, NULL,
2975					      NULL, bh, rw);
2976	} else if (NULL != dev_state && (rw & REQ_FLUSH)) {
2977		if (dev_state->state->print_mask &
2978		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2979			printk(KERN_INFO
2980			       "submit_bh(rw=0x%x FLUSH, bdev=%p)\n",
2981			       rw, bh->b_bdev);
2982		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2983			if ((dev_state->state->print_mask &
2984			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2985			      BTRFSIC_PRINT_MASK_VERBOSE)))
2986				printk(KERN_INFO
2987				       "btrfsic_submit_bh(%s) with FLUSH"
2988				       " but dummy block already in use"
2989				       " (ignored)!\n",
2990				       dev_state->name);
2991		} else {
2992			struct btrfsic_block *const block =
2993				&dev_state->dummy_block_for_bio_bh_flush;
2994
2995			block->is_iodone = 0;
2996			block->never_written = 0;
2997			block->iodone_w_error = 0;
2998			block->flush_gen = dev_state->last_flush_gen + 1;
2999			block->submit_bio_bh_rw = rw;
3000			block->orig_bio_bh_private = bh->b_private;
3001			block->orig_bio_bh_end_io.bh = bh->b_end_io;
3002			block->next_in_same_bio = NULL;
3003			bh->b_private = block;
3004			bh->b_end_io = btrfsic_bh_end_io;
3005		}
3006	}
3007	mutex_unlock(&btrfsic_mutex);
3008	return submit_bh(rw, bh);
3009}
3010
3011void btrfsic_submit_bio(int rw, struct bio *bio)
3012{
3013	struct btrfsic_dev_state *dev_state;
3014
3015	if (!btrfsic_is_initialized) {
3016		submit_bio(rw, bio);
3017		return;
3018	}
3019
3020	mutex_lock(&btrfsic_mutex);
3021	/* since btrfsic_submit_bio() is also called before
3022	 * btrfsic_mount(), this might return NULL */
3023	dev_state = btrfsic_dev_state_lookup(bio->bi_bdev);
3024	if (NULL != dev_state &&
3025	    (rw & WRITE) && NULL != bio->bi_io_vec) {
3026		unsigned int i;
3027		u64 dev_bytenr;
3028		u64 cur_bytenr;
3029		int bio_is_patched;
3030		char **mapped_datav;
3031
3032		dev_bytenr = 512 * bio->bi_sector;
3033		bio_is_patched = 0;
3034		if (dev_state->state->print_mask &
3035		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
3036			printk(KERN_INFO
3037			       "submit_bio(rw=0x%x, bi_vcnt=%u,"
3038			       " bi_sector=%llu (bytenr %llu), bi_bdev=%p)\n",
3039			       rw, bio->bi_vcnt,
3040			       (unsigned long long)bio->bi_sector, dev_bytenr,
3041			       bio->bi_bdev);
3042
3043		mapped_datav = kmalloc(sizeof(*mapped_datav) * bio->bi_vcnt,
3044				       GFP_NOFS);
3045		if (!mapped_datav)
3046			goto leave;
3047		cur_bytenr = dev_bytenr;
3048		for (i = 0; i < bio->bi_vcnt; i++) {
3049			BUG_ON(bio->bi_io_vec[i].bv_len != PAGE_CACHE_SIZE);
3050			mapped_datav[i] = kmap(bio->bi_io_vec[i].bv_page);
3051			if (!mapped_datav[i]) {
3052				while (i > 0) {
3053					i--;
3054					kunmap(bio->bi_io_vec[i].bv_page);
3055				}
3056				kfree(mapped_datav);
3057				goto leave;
3058			}
3059			if (dev_state->state->print_mask &
3060			    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
3061				printk(KERN_INFO
3062				       "#%u: bytenr=%llu, len=%u, offset=%u\n",
3063				       i, cur_bytenr, bio->bi_io_vec[i].bv_len,
3064				       bio->bi_io_vec[i].bv_offset);
3065			cur_bytenr += bio->bi_io_vec[i].bv_len;
3066		}
3067		btrfsic_process_written_block(dev_state, dev_bytenr,
3068					      mapped_datav, bio->bi_vcnt,
3069					      bio, &bio_is_patched,
3070					      NULL, rw);
3071		while (i > 0) {
3072			i--;
3073			kunmap(bio->bi_io_vec[i].bv_page);
3074		}
3075		kfree(mapped_datav);
3076	} else if (NULL != dev_state && (rw & REQ_FLUSH)) {
3077		if (dev_state->state->print_mask &
3078		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
3079			printk(KERN_INFO
3080			       "submit_bio(rw=0x%x FLUSH, bdev=%p)\n",
3081			       rw, bio->bi_bdev);
3082		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
3083			if ((dev_state->state->print_mask &
3084			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
3085			      BTRFSIC_PRINT_MASK_VERBOSE)))
3086				printk(KERN_INFO
3087				       "btrfsic_submit_bio(%s) with FLUSH"
3088				       " but dummy block already in use"
3089				       " (ignored)!\n",
3090				       dev_state->name);
3091		} else {
3092			struct btrfsic_block *const block =
3093				&dev_state->dummy_block_for_bio_bh_flush;
3094
3095			block->is_iodone = 0;
3096			block->never_written = 0;
3097			block->iodone_w_error = 0;
3098			block->flush_gen = dev_state->last_flush_gen + 1;
3099			block->submit_bio_bh_rw = rw;
3100			block->orig_bio_bh_private = bio->bi_private;
3101			block->orig_bio_bh_end_io.bio = bio->bi_end_io;
3102			block->next_in_same_bio = NULL;
3103			bio->bi_private = block;
3104			bio->bi_end_io = btrfsic_bio_end_io;
3105		}
3106	}
3107leave:
3108	mutex_unlock(&btrfsic_mutex);
3109
3110	submit_bio(rw, bio);
3111}
3112
3113int btrfsic_mount(struct btrfs_root *root,
3114		  struct btrfs_fs_devices *fs_devices,
3115		  int including_extent_data, u32 print_mask)
3116{
3117	int ret;
3118	struct btrfsic_state *state;
3119	struct list_head *dev_head = &fs_devices->devices;
3120	struct btrfs_device *device;
3121
3122	if (root->nodesize != root->leafsize) {
3123		printk(KERN_INFO
3124		       "btrfsic: cannot handle nodesize %d != leafsize %d!\n",
3125		       root->nodesize, root->leafsize);
3126		return -1;
3127	}
3128	if (root->nodesize & ((u64)PAGE_CACHE_SIZE - 1)) {
3129		printk(KERN_INFO
3130		       "btrfsic: cannot handle nodesize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3131		       root->nodesize, PAGE_CACHE_SIZE);
3132		return -1;
3133	}
3134	if (root->leafsize & ((u64)PAGE_CACHE_SIZE - 1)) {
3135		printk(KERN_INFO
3136		       "btrfsic: cannot handle leafsize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3137		       root->leafsize, PAGE_CACHE_SIZE);
3138		return -1;
3139	}
3140	if (root->sectorsize & ((u64)PAGE_CACHE_SIZE - 1)) {
3141		printk(KERN_INFO
3142		       "btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_CACHE_SIZE %ld!\n",
3143		       root->sectorsize, PAGE_CACHE_SIZE);
3144		return -1;
3145	}
3146	state = kzalloc(sizeof(*state), GFP_NOFS);
3147	if (NULL == state) {
3148		printk(KERN_INFO "btrfs check-integrity: kmalloc() failed!\n");
3149		return -1;
3150	}
3151
3152	if (!btrfsic_is_initialized) {
3153		mutex_init(&btrfsic_mutex);
3154		btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
3155		btrfsic_is_initialized = 1;
3156	}
3157	mutex_lock(&btrfsic_mutex);
3158	state->root = root;
3159	state->print_mask = print_mask;
3160	state->include_extent_data = including_extent_data;
3161	state->csum_size = 0;
3162	state->metablock_size = root->nodesize;
3163	state->datablock_size = root->sectorsize;
3164	INIT_LIST_HEAD(&state->all_blocks_list);
3165	btrfsic_block_hashtable_init(&state->block_hashtable);
3166	btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
3167	state->max_superblock_generation = 0;
3168	state->latest_superblock = NULL;
3169
3170	list_for_each_entry(device, dev_head, dev_list) {
3171		struct btrfsic_dev_state *ds;
3172		char *p;
3173
3174		if (!device->bdev || !device->name)
3175			continue;
3176
3177		ds = btrfsic_dev_state_alloc();
3178		if (NULL == ds) {
3179			printk(KERN_INFO
3180			       "btrfs check-integrity: kmalloc() failed!\n");
3181			mutex_unlock(&btrfsic_mutex);
3182			return -1;
3183		}
3184		ds->bdev = device->bdev;
3185		ds->state = state;
3186		bdevname(ds->bdev, ds->name);
3187		ds->name[BDEVNAME_SIZE - 1] = '\0';
3188		for (p = ds->name; *p != '\0'; p++);
3189		while (p > ds->name && *p != '/')
3190			p--;
3191		if (*p == '/')
3192			p++;
3193		strlcpy(ds->name, p, sizeof(ds->name));
3194		btrfsic_dev_state_hashtable_add(ds,
3195						&btrfsic_dev_state_hashtable);
3196	}
3197
3198	ret = btrfsic_process_superblock(state, fs_devices);
3199	if (0 != ret) {
3200		mutex_unlock(&btrfsic_mutex);
3201		btrfsic_unmount(root, fs_devices);
3202		return ret;
3203	}
3204
3205	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
3206		btrfsic_dump_database(state);
3207	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
3208		btrfsic_dump_tree(state);
3209
3210	mutex_unlock(&btrfsic_mutex);
3211	return 0;
3212}
3213
3214void btrfsic_unmount(struct btrfs_root *root,
3215		     struct btrfs_fs_devices *fs_devices)
3216{
3217	struct list_head *elem_all;
3218	struct list_head *tmp_all;
3219	struct btrfsic_state *state;
3220	struct list_head *dev_head = &fs_devices->devices;
3221	struct btrfs_device *device;
3222
3223	if (!btrfsic_is_initialized)
3224		return;
3225
3226	mutex_lock(&btrfsic_mutex);
3227
3228	state = NULL;
3229	list_for_each_entry(device, dev_head, dev_list) {
3230		struct btrfsic_dev_state *ds;
3231
3232		if (!device->bdev || !device->name)
3233			continue;
3234
3235		ds = btrfsic_dev_state_hashtable_lookup(
3236				device->bdev,
3237				&btrfsic_dev_state_hashtable);
3238		if (NULL != ds) {
3239			state = ds->state;
3240			btrfsic_dev_state_hashtable_remove(ds);
3241			btrfsic_dev_state_free(ds);
3242		}
3243	}
3244
3245	if (NULL == state) {
3246		printk(KERN_INFO
3247		       "btrfsic: error, cannot find state information"
3248		       " on umount!\n");
3249		mutex_unlock(&btrfsic_mutex);
3250		return;
3251	}
3252
3253	/*
3254	 * Don't care about keeping the lists' state up to date,
3255	 * just free all memory that was allocated dynamically.
3256	 * Free the blocks and the block_links.
3257	 */
3258	list_for_each_safe(elem_all, tmp_all, &state->all_blocks_list) {
3259		struct btrfsic_block *const b_all =
3260		    list_entry(elem_all, struct btrfsic_block,
3261			       all_blocks_node);
3262		struct list_head *elem_ref_to;
3263		struct list_head *tmp_ref_to;
3264
3265		list_for_each_safe(elem_ref_to, tmp_ref_to,
3266				   &b_all->ref_to_list) {
3267			struct btrfsic_block_link *const l =
3268			    list_entry(elem_ref_to,
3269				       struct btrfsic_block_link,
3270				       node_ref_to);
3271
3272			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
3273				btrfsic_print_rem_link(state, l);
3274
3275			l->ref_cnt--;
3276			if (0 == l->ref_cnt)
3277				btrfsic_block_link_free(l);
3278		}
3279
3280		if (b_all->is_iodone || b_all->never_written)
3281			btrfsic_block_free(b_all);
3282		else
3283			printk(KERN_INFO "btrfs: attempt to free %c-block"
3284			       " @%llu (%s/%llu/%d) on umount which is"
3285			       " not yet iodone!\n",
3286			       btrfsic_get_block_type(state, b_all),
3287			       b_all->logical_bytenr, b_all->dev_state->name,
3288			       b_all->dev_bytenr, b_all->mirror_num);
3289	}
3290
3291	mutex_unlock(&btrfsic_mutex);
3292
3293	kfree(state);
3294}
3295