xref: /openbmc/linux/fs/btrfs/check-integrity.c (revision d28e649a)
1 /*
2  * Copyright (C) STRATO AG 2011.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 /*
20  * This module can be used to catch cases when the btrfs kernel
21  * code executes write requests to the disk that bring the file
22  * system in an inconsistent state. In such a state, a power-loss
23  * or kernel panic event would cause that the data on disk is
24  * lost or at least damaged.
25  *
26  * Code is added that examines all block write requests during
27  * runtime (including writes of the super block). Three rules
28  * are verified and an error is printed on violation of the
29  * rules:
30  * 1. It is not allowed to write a disk block which is
31  *    currently referenced by the super block (either directly
32  *    or indirectly).
33  * 2. When a super block is written, it is verified that all
34  *    referenced (directly or indirectly) blocks fulfill the
35  *    following requirements:
36  *    2a. All referenced blocks have either been present when
37  *        the file system was mounted, (i.e., they have been
38  *        referenced by the super block) or they have been
39  *        written since then and the write completion callback
40  *        was called and no write error was indicated and a
41  *        FLUSH request to the device where these blocks are
42  *        located was received and completed.
43  *    2b. All referenced blocks need to have a generation
44  *        number which is equal to the parent's number.
45  *
46  * One issue that was found using this module was that the log
47  * tree on disk became temporarily corrupted because disk blocks
48  * that had been in use for the log tree had been freed and
49  * reused too early, while being referenced by the written super
50  * block.
51  *
52  * The search term in the kernel log that can be used to filter
53  * on the existence of detected integrity issues is
54  * "btrfs: attempt".
55  *
56  * The integrity check is enabled via mount options. These
57  * mount options are only supported if the integrity check
58  * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY.
59  *
60  * Example #1, apply integrity checks to all metadata:
61  * mount /dev/sdb1 /mnt -o check_int
62  *
63  * Example #2, apply integrity checks to all metadata and
64  * to data extents:
65  * mount /dev/sdb1 /mnt -o check_int_data
66  *
67  * Example #3, apply integrity checks to all metadata and dump
68  * the tree that the super block references to kernel messages
69  * each time after a super block was written:
70  * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263
71  *
72  * If the integrity check tool is included and activated in
73  * the mount options, plenty of kernel memory is used, and
74  * plenty of additional CPU cycles are spent. Enabling this
75  * functionality is not intended for normal use. In most
76  * cases, unless you are a btrfs developer who needs to verify
77  * the integrity of (super)-block write requests, do not
78  * enable the config option BTRFS_FS_CHECK_INTEGRITY to
79  * include and compile the integrity check tool.
80  *
81  * Expect millions of lines of information in the kernel log with an
82  * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the
83  * kernel config to at least 26 (which is 64MB). Usually the value is
84  * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be
85  * changed like this before LOG_BUF_SHIFT can be set to a high value:
86  * config LOG_BUF_SHIFT
87  *       int "Kernel log buffer size (16 => 64KB, 17 => 128KB)"
88  *       range 12 30
89  */
90 
91 #include <linux/sched.h>
92 #include <linux/slab.h>
93 #include <linux/buffer_head.h>
94 #include <linux/mutex.h>
95 #include <linux/genhd.h>
96 #include <linux/blkdev.h>
97 #include <linux/mm.h>
98 #include <linux/string.h>
99 #include "ctree.h"
100 #include "disk-io.h"
101 #include "hash.h"
102 #include "transaction.h"
103 #include "extent_io.h"
104 #include "volumes.h"
105 #include "print-tree.h"
106 #include "locking.h"
107 #include "check-integrity.h"
108 #include "rcu-string.h"
109 #include "compression.h"
110 
111 #define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
112 #define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
113 #define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
114 #define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
115 #define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
116 #define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
117 #define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
118 #define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6)	/* in characters,
119 							 * excluding " [...]" */
120 #define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
121 
122 /*
123  * The definition of the bitmask fields for the print_mask.
124  * They are specified with the mount option check_integrity_print_mask.
125  */
126 #define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE			0x00000001
127 #define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION		0x00000002
128 #define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE			0x00000004
129 #define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE			0x00000008
130 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH			0x00000010
131 #define BTRFSIC_PRINT_MASK_END_IO_BIO_BH			0x00000020
132 #define BTRFSIC_PRINT_MASK_VERBOSE				0x00000040
133 #define BTRFSIC_PRINT_MASK_VERY_VERBOSE				0x00000080
134 #define BTRFSIC_PRINT_MASK_INITIAL_TREE				0x00000100
135 #define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES			0x00000200
136 #define BTRFSIC_PRINT_MASK_INITIAL_DATABASE			0x00000400
137 #define BTRFSIC_PRINT_MASK_NUM_COPIES				0x00000800
138 #define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS		0x00001000
139 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE		0x00002000
140 
141 struct btrfsic_dev_state;
142 struct btrfsic_state;
143 
144 struct btrfsic_block {
145 	u32 magic_num;		/* only used for debug purposes */
146 	unsigned int is_metadata:1;	/* if it is meta-data, not data-data */
147 	unsigned int is_superblock:1;	/* if it is one of the superblocks */
148 	unsigned int is_iodone:1;	/* if is done by lower subsystem */
149 	unsigned int iodone_w_error:1;	/* error was indicated to endio */
150 	unsigned int never_written:1;	/* block was added because it was
151 					 * referenced, not because it was
152 					 * written */
153 	unsigned int mirror_num;	/* large enough to hold
154 					 * BTRFS_SUPER_MIRROR_MAX */
155 	struct btrfsic_dev_state *dev_state;
156 	u64 dev_bytenr;		/* key, physical byte num on disk */
157 	u64 logical_bytenr;	/* logical byte num on disk */
158 	u64 generation;
159 	struct btrfs_disk_key disk_key;	/* extra info to print in case of
160 					 * issues, will not always be correct */
161 	struct list_head collision_resolving_node;	/* list node */
162 	struct list_head all_blocks_node;	/* list node */
163 
164 	/* the following two lists contain block_link items */
165 	struct list_head ref_to_list;	/* list */
166 	struct list_head ref_from_list;	/* list */
167 	struct btrfsic_block *next_in_same_bio;
168 	void *orig_bio_bh_private;
169 	union {
170 		bio_end_io_t *bio;
171 		bh_end_io_t *bh;
172 	} orig_bio_bh_end_io;
173 	int submit_bio_bh_rw;
174 	u64 flush_gen; /* only valid if !never_written */
175 };
176 
177 /*
178  * Elements of this type are allocated dynamically and required because
179  * each block object can refer to and can be ref from multiple blocks.
180  * The key to lookup them in the hashtable is the dev_bytenr of
181  * the block ref to plus the one from the block referred from.
182  * The fact that they are searchable via a hashtable and that a
183  * ref_cnt is maintained is not required for the btrfs integrity
184  * check algorithm itself, it is only used to make the output more
185  * beautiful in case that an error is detected (an error is defined
186  * as a write operation to a block while that block is still referenced).
187  */
188 struct btrfsic_block_link {
189 	u32 magic_num;		/* only used for debug purposes */
190 	u32 ref_cnt;
191 	struct list_head node_ref_to;	/* list node */
192 	struct list_head node_ref_from;	/* list node */
193 	struct list_head collision_resolving_node;	/* list node */
194 	struct btrfsic_block *block_ref_to;
195 	struct btrfsic_block *block_ref_from;
196 	u64 parent_generation;
197 };
198 
199 struct btrfsic_dev_state {
200 	u32 magic_num;		/* only used for debug purposes */
201 	struct block_device *bdev;
202 	struct btrfsic_state *state;
203 	struct list_head collision_resolving_node;	/* list node */
204 	struct btrfsic_block dummy_block_for_bio_bh_flush;
205 	u64 last_flush_gen;
206 	char name[BDEVNAME_SIZE];
207 };
208 
209 struct btrfsic_block_hashtable {
210 	struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
211 };
212 
213 struct btrfsic_block_link_hashtable {
214 	struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
215 };
216 
217 struct btrfsic_dev_state_hashtable {
218 	struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
219 };
220 
221 struct btrfsic_block_data_ctx {
222 	u64 start;		/* virtual bytenr */
223 	u64 dev_bytenr;		/* physical bytenr on device */
224 	u32 len;
225 	struct btrfsic_dev_state *dev;
226 	char **datav;
227 	struct page **pagev;
228 	void *mem_to_free;
229 };
230 
231 /* This structure is used to implement recursion without occupying
232  * any stack space, refer to btrfsic_process_metablock() */
233 struct btrfsic_stack_frame {
234 	u32 magic;
235 	u32 nr;
236 	int error;
237 	int i;
238 	int limit_nesting;
239 	int num_copies;
240 	int mirror_num;
241 	struct btrfsic_block *block;
242 	struct btrfsic_block_data_ctx *block_ctx;
243 	struct btrfsic_block *next_block;
244 	struct btrfsic_block_data_ctx next_block_ctx;
245 	struct btrfs_header *hdr;
246 	struct btrfsic_stack_frame *prev;
247 };
248 
249 /* Some state per mounted filesystem */
250 struct btrfsic_state {
251 	u32 print_mask;
252 	int include_extent_data;
253 	int csum_size;
254 	struct list_head all_blocks_list;
255 	struct btrfsic_block_hashtable block_hashtable;
256 	struct btrfsic_block_link_hashtable block_link_hashtable;
257 	struct btrfs_fs_info *fs_info;
258 	u64 max_superblock_generation;
259 	struct btrfsic_block *latest_superblock;
260 	u32 metablock_size;
261 	u32 datablock_size;
262 };
263 
264 static void btrfsic_block_init(struct btrfsic_block *b);
265 static struct btrfsic_block *btrfsic_block_alloc(void);
266 static void btrfsic_block_free(struct btrfsic_block *b);
267 static void btrfsic_block_link_init(struct btrfsic_block_link *n);
268 static struct btrfsic_block_link *btrfsic_block_link_alloc(void);
269 static void btrfsic_block_link_free(struct btrfsic_block_link *n);
270 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds);
271 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void);
272 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds);
273 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h);
274 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
275 					struct btrfsic_block_hashtable *h);
276 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b);
277 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
278 		struct block_device *bdev,
279 		u64 dev_bytenr,
280 		struct btrfsic_block_hashtable *h);
281 static void btrfsic_block_link_hashtable_init(
282 		struct btrfsic_block_link_hashtable *h);
283 static void btrfsic_block_link_hashtable_add(
284 		struct btrfsic_block_link *l,
285 		struct btrfsic_block_link_hashtable *h);
286 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l);
287 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
288 		struct block_device *bdev_ref_to,
289 		u64 dev_bytenr_ref_to,
290 		struct block_device *bdev_ref_from,
291 		u64 dev_bytenr_ref_from,
292 		struct btrfsic_block_link_hashtable *h);
293 static void btrfsic_dev_state_hashtable_init(
294 		struct btrfsic_dev_state_hashtable *h);
295 static void btrfsic_dev_state_hashtable_add(
296 		struct btrfsic_dev_state *ds,
297 		struct btrfsic_dev_state_hashtable *h);
298 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds);
299 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev,
300 		struct btrfsic_dev_state_hashtable *h);
301 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void);
302 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf);
303 static int btrfsic_process_superblock(struct btrfsic_state *state,
304 				      struct btrfs_fs_devices *fs_devices);
305 static int btrfsic_process_metablock(struct btrfsic_state *state,
306 				     struct btrfsic_block *block,
307 				     struct btrfsic_block_data_ctx *block_ctx,
308 				     int limit_nesting, int force_iodone_flag);
309 static void btrfsic_read_from_block_data(
310 	struct btrfsic_block_data_ctx *block_ctx,
311 	void *dst, u32 offset, size_t len);
312 static int btrfsic_create_link_to_next_block(
313 		struct btrfsic_state *state,
314 		struct btrfsic_block *block,
315 		struct btrfsic_block_data_ctx
316 		*block_ctx, u64 next_bytenr,
317 		int limit_nesting,
318 		struct btrfsic_block_data_ctx *next_block_ctx,
319 		struct btrfsic_block **next_blockp,
320 		int force_iodone_flag,
321 		int *num_copiesp, int *mirror_nump,
322 		struct btrfs_disk_key *disk_key,
323 		u64 parent_generation);
324 static int btrfsic_handle_extent_data(struct btrfsic_state *state,
325 				      struct btrfsic_block *block,
326 				      struct btrfsic_block_data_ctx *block_ctx,
327 				      u32 item_offset, int force_iodone_flag);
328 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
329 			     struct btrfsic_block_data_ctx *block_ctx_out,
330 			     int mirror_num);
331 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
332 static int btrfsic_read_block(struct btrfsic_state *state,
333 			      struct btrfsic_block_data_ctx *block_ctx);
334 static void btrfsic_dump_database(struct btrfsic_state *state);
335 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
336 				     char **datav, unsigned int num_pages);
337 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
338 					  u64 dev_bytenr, char **mapped_datav,
339 					  unsigned int num_pages,
340 					  struct bio *bio, int *bio_is_patched,
341 					  struct buffer_head *bh,
342 					  int submit_bio_bh_rw);
343 static int btrfsic_process_written_superblock(
344 		struct btrfsic_state *state,
345 		struct btrfsic_block *const block,
346 		struct btrfs_super_block *const super_hdr);
347 static void btrfsic_bio_end_io(struct bio *bp);
348 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate);
349 static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
350 					      const struct btrfsic_block *block,
351 					      int recursion_level);
352 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
353 					struct btrfsic_block *const block,
354 					int recursion_level);
355 static void btrfsic_print_add_link(const struct btrfsic_state *state,
356 				   const struct btrfsic_block_link *l);
357 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
358 				   const struct btrfsic_block_link *l);
359 static char btrfsic_get_block_type(const struct btrfsic_state *state,
360 				   const struct btrfsic_block *block);
361 static void btrfsic_dump_tree(const struct btrfsic_state *state);
362 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
363 				  const struct btrfsic_block *block,
364 				  int indent_level);
365 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
366 		struct btrfsic_state *state,
367 		struct btrfsic_block_data_ctx *next_block_ctx,
368 		struct btrfsic_block *next_block,
369 		struct btrfsic_block *from_block,
370 		u64 parent_generation);
371 static struct btrfsic_block *btrfsic_block_lookup_or_add(
372 		struct btrfsic_state *state,
373 		struct btrfsic_block_data_ctx *block_ctx,
374 		const char *additional_string,
375 		int is_metadata,
376 		int is_iodone,
377 		int never_written,
378 		int mirror_num,
379 		int *was_created);
380 static int btrfsic_process_superblock_dev_mirror(
381 		struct btrfsic_state *state,
382 		struct btrfsic_dev_state *dev_state,
383 		struct btrfs_device *device,
384 		int superblock_mirror_num,
385 		struct btrfsic_dev_state **selected_dev_state,
386 		struct btrfs_super_block *selected_super);
387 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev);
388 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
389 					   u64 bytenr,
390 					   struct btrfsic_dev_state *dev_state,
391 					   u64 dev_bytenr);
392 
393 static struct mutex btrfsic_mutex;
394 static int btrfsic_is_initialized;
395 static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
396 
397 
398 static void btrfsic_block_init(struct btrfsic_block *b)
399 {
400 	b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
401 	b->dev_state = NULL;
402 	b->dev_bytenr = 0;
403 	b->logical_bytenr = 0;
404 	b->generation = BTRFSIC_GENERATION_UNKNOWN;
405 	b->disk_key.objectid = 0;
406 	b->disk_key.type = 0;
407 	b->disk_key.offset = 0;
408 	b->is_metadata = 0;
409 	b->is_superblock = 0;
410 	b->is_iodone = 0;
411 	b->iodone_w_error = 0;
412 	b->never_written = 0;
413 	b->mirror_num = 0;
414 	b->next_in_same_bio = NULL;
415 	b->orig_bio_bh_private = NULL;
416 	b->orig_bio_bh_end_io.bio = NULL;
417 	INIT_LIST_HEAD(&b->collision_resolving_node);
418 	INIT_LIST_HEAD(&b->all_blocks_node);
419 	INIT_LIST_HEAD(&b->ref_to_list);
420 	INIT_LIST_HEAD(&b->ref_from_list);
421 	b->submit_bio_bh_rw = 0;
422 	b->flush_gen = 0;
423 }
424 
425 static struct btrfsic_block *btrfsic_block_alloc(void)
426 {
427 	struct btrfsic_block *b;
428 
429 	b = kzalloc(sizeof(*b), GFP_NOFS);
430 	if (NULL != b)
431 		btrfsic_block_init(b);
432 
433 	return b;
434 }
435 
436 static void btrfsic_block_free(struct btrfsic_block *b)
437 {
438 	BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
439 	kfree(b);
440 }
441 
442 static void btrfsic_block_link_init(struct btrfsic_block_link *l)
443 {
444 	l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
445 	l->ref_cnt = 1;
446 	INIT_LIST_HEAD(&l->node_ref_to);
447 	INIT_LIST_HEAD(&l->node_ref_from);
448 	INIT_LIST_HEAD(&l->collision_resolving_node);
449 	l->block_ref_to = NULL;
450 	l->block_ref_from = NULL;
451 }
452 
453 static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
454 {
455 	struct btrfsic_block_link *l;
456 
457 	l = kzalloc(sizeof(*l), GFP_NOFS);
458 	if (NULL != l)
459 		btrfsic_block_link_init(l);
460 
461 	return l;
462 }
463 
464 static void btrfsic_block_link_free(struct btrfsic_block_link *l)
465 {
466 	BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
467 	kfree(l);
468 }
469 
470 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
471 {
472 	ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
473 	ds->bdev = NULL;
474 	ds->state = NULL;
475 	ds->name[0] = '\0';
476 	INIT_LIST_HEAD(&ds->collision_resolving_node);
477 	ds->last_flush_gen = 0;
478 	btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
479 	ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
480 	ds->dummy_block_for_bio_bh_flush.dev_state = ds;
481 }
482 
483 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
484 {
485 	struct btrfsic_dev_state *ds;
486 
487 	ds = kzalloc(sizeof(*ds), GFP_NOFS);
488 	if (NULL != ds)
489 		btrfsic_dev_state_init(ds);
490 
491 	return ds;
492 }
493 
494 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
495 {
496 	BUG_ON(!(NULL == ds ||
497 		 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
498 	kfree(ds);
499 }
500 
501 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
502 {
503 	int i;
504 
505 	for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
506 		INIT_LIST_HEAD(h->table + i);
507 }
508 
509 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
510 					struct btrfsic_block_hashtable *h)
511 {
512 	const unsigned int hashval =
513 	    (((unsigned int)(b->dev_bytenr >> 16)) ^
514 	     ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
515 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
516 
517 	list_add(&b->collision_resolving_node, h->table + hashval);
518 }
519 
520 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
521 {
522 	list_del(&b->collision_resolving_node);
523 }
524 
525 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
526 		struct block_device *bdev,
527 		u64 dev_bytenr,
528 		struct btrfsic_block_hashtable *h)
529 {
530 	const unsigned int hashval =
531 	    (((unsigned int)(dev_bytenr >> 16)) ^
532 	     ((unsigned int)((uintptr_t)bdev))) &
533 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
534 	struct btrfsic_block *b;
535 
536 	list_for_each_entry(b, h->table + hashval, collision_resolving_node) {
537 		if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
538 			return b;
539 	}
540 
541 	return NULL;
542 }
543 
544 static void btrfsic_block_link_hashtable_init(
545 		struct btrfsic_block_link_hashtable *h)
546 {
547 	int i;
548 
549 	for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
550 		INIT_LIST_HEAD(h->table + i);
551 }
552 
553 static void btrfsic_block_link_hashtable_add(
554 		struct btrfsic_block_link *l,
555 		struct btrfsic_block_link_hashtable *h)
556 {
557 	const unsigned int hashval =
558 	    (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
559 	     ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
560 	     ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
561 	     ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
562 	     & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
563 
564 	BUG_ON(NULL == l->block_ref_to);
565 	BUG_ON(NULL == l->block_ref_from);
566 	list_add(&l->collision_resolving_node, h->table + hashval);
567 }
568 
569 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
570 {
571 	list_del(&l->collision_resolving_node);
572 }
573 
574 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
575 		struct block_device *bdev_ref_to,
576 		u64 dev_bytenr_ref_to,
577 		struct block_device *bdev_ref_from,
578 		u64 dev_bytenr_ref_from,
579 		struct btrfsic_block_link_hashtable *h)
580 {
581 	const unsigned int hashval =
582 	    (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
583 	     ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
584 	     ((unsigned int)((uintptr_t)bdev_ref_to)) ^
585 	     ((unsigned int)((uintptr_t)bdev_ref_from))) &
586 	     (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
587 	struct btrfsic_block_link *l;
588 
589 	list_for_each_entry(l, h->table + hashval, collision_resolving_node) {
590 		BUG_ON(NULL == l->block_ref_to);
591 		BUG_ON(NULL == l->block_ref_from);
592 		if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
593 		    l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
594 		    l->block_ref_from->dev_state->bdev == bdev_ref_from &&
595 		    l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
596 			return l;
597 	}
598 
599 	return NULL;
600 }
601 
602 static void btrfsic_dev_state_hashtable_init(
603 		struct btrfsic_dev_state_hashtable *h)
604 {
605 	int i;
606 
607 	for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
608 		INIT_LIST_HEAD(h->table + i);
609 }
610 
611 static void btrfsic_dev_state_hashtable_add(
612 		struct btrfsic_dev_state *ds,
613 		struct btrfsic_dev_state_hashtable *h)
614 {
615 	const unsigned int hashval =
616 	    (((unsigned int)((uintptr_t)ds->bdev->bd_dev)) &
617 	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
618 
619 	list_add(&ds->collision_resolving_node, h->table + hashval);
620 }
621 
622 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
623 {
624 	list_del(&ds->collision_resolving_node);
625 }
626 
627 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev,
628 		struct btrfsic_dev_state_hashtable *h)
629 {
630 	const unsigned int hashval =
631 		dev & (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1);
632 	struct btrfsic_dev_state *ds;
633 
634 	list_for_each_entry(ds, h->table + hashval, collision_resolving_node) {
635 		if (ds->bdev->bd_dev == dev)
636 			return ds;
637 	}
638 
639 	return NULL;
640 }
641 
642 static int btrfsic_process_superblock(struct btrfsic_state *state,
643 				      struct btrfs_fs_devices *fs_devices)
644 {
645 	struct btrfs_fs_info *fs_info = state->fs_info;
646 	struct btrfs_super_block *selected_super;
647 	struct list_head *dev_head = &fs_devices->devices;
648 	struct btrfs_device *device;
649 	struct btrfsic_dev_state *selected_dev_state = NULL;
650 	int ret = 0;
651 	int pass;
652 
653 	BUG_ON(NULL == state);
654 	selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
655 	if (NULL == selected_super) {
656 		pr_info("btrfsic: error, kmalloc failed!\n");
657 		return -ENOMEM;
658 	}
659 
660 	list_for_each_entry(device, dev_head, dev_list) {
661 		int i;
662 		struct btrfsic_dev_state *dev_state;
663 
664 		if (!device->bdev || !device->name)
665 			continue;
666 
667 		dev_state = btrfsic_dev_state_lookup(device->bdev->bd_dev);
668 		BUG_ON(NULL == dev_state);
669 		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
670 			ret = btrfsic_process_superblock_dev_mirror(
671 					state, dev_state, device, i,
672 					&selected_dev_state, selected_super);
673 			if (0 != ret && 0 == i) {
674 				kfree(selected_super);
675 				return ret;
676 			}
677 		}
678 	}
679 
680 	if (NULL == state->latest_superblock) {
681 		pr_info("btrfsic: no superblock found!\n");
682 		kfree(selected_super);
683 		return -1;
684 	}
685 
686 	state->csum_size = btrfs_super_csum_size(selected_super);
687 
688 	for (pass = 0; pass < 3; pass++) {
689 		int num_copies;
690 		int mirror_num;
691 		u64 next_bytenr;
692 
693 		switch (pass) {
694 		case 0:
695 			next_bytenr = btrfs_super_root(selected_super);
696 			if (state->print_mask &
697 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
698 				pr_info("root@%llu\n", next_bytenr);
699 			break;
700 		case 1:
701 			next_bytenr = btrfs_super_chunk_root(selected_super);
702 			if (state->print_mask &
703 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
704 				pr_info("chunk@%llu\n", next_bytenr);
705 			break;
706 		case 2:
707 			next_bytenr = btrfs_super_log_root(selected_super);
708 			if (0 == next_bytenr)
709 				continue;
710 			if (state->print_mask &
711 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
712 				pr_info("log@%llu\n", next_bytenr);
713 			break;
714 		}
715 
716 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
717 					      state->metablock_size);
718 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
719 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
720 			       next_bytenr, num_copies);
721 
722 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
723 			struct btrfsic_block *next_block;
724 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
725 			struct btrfsic_block_link *l;
726 
727 			ret = btrfsic_map_block(state, next_bytenr,
728 						state->metablock_size,
729 						&tmp_next_block_ctx,
730 						mirror_num);
731 			if (ret) {
732 				pr_info("btrfsic: btrfsic_map_block(root @%llu, mirror %d) failed!\n",
733 				       next_bytenr, mirror_num);
734 				kfree(selected_super);
735 				return -1;
736 			}
737 
738 			next_block = btrfsic_block_hashtable_lookup(
739 					tmp_next_block_ctx.dev->bdev,
740 					tmp_next_block_ctx.dev_bytenr,
741 					&state->block_hashtable);
742 			BUG_ON(NULL == next_block);
743 
744 			l = btrfsic_block_link_hashtable_lookup(
745 					tmp_next_block_ctx.dev->bdev,
746 					tmp_next_block_ctx.dev_bytenr,
747 					state->latest_superblock->dev_state->
748 					bdev,
749 					state->latest_superblock->dev_bytenr,
750 					&state->block_link_hashtable);
751 			BUG_ON(NULL == l);
752 
753 			ret = btrfsic_read_block(state, &tmp_next_block_ctx);
754 			if (ret < (int)PAGE_SIZE) {
755 				pr_info("btrfsic: read @logical %llu failed!\n",
756 				       tmp_next_block_ctx.start);
757 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
758 				kfree(selected_super);
759 				return -1;
760 			}
761 
762 			ret = btrfsic_process_metablock(state,
763 							next_block,
764 							&tmp_next_block_ctx,
765 							BTRFS_MAX_LEVEL + 3, 1);
766 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
767 		}
768 	}
769 
770 	kfree(selected_super);
771 	return ret;
772 }
773 
774 static int btrfsic_process_superblock_dev_mirror(
775 		struct btrfsic_state *state,
776 		struct btrfsic_dev_state *dev_state,
777 		struct btrfs_device *device,
778 		int superblock_mirror_num,
779 		struct btrfsic_dev_state **selected_dev_state,
780 		struct btrfs_super_block *selected_super)
781 {
782 	struct btrfs_fs_info *fs_info = state->fs_info;
783 	struct btrfs_super_block *super_tmp;
784 	u64 dev_bytenr;
785 	struct buffer_head *bh;
786 	struct btrfsic_block *superblock_tmp;
787 	int pass;
788 	struct block_device *const superblock_bdev = device->bdev;
789 
790 	/* super block bytenr is always the unmapped device bytenr */
791 	dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
792 	if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->commit_total_bytes)
793 		return -1;
794 	bh = __bread(superblock_bdev, dev_bytenr / BTRFS_BDEV_BLOCKSIZE,
795 		     BTRFS_SUPER_INFO_SIZE);
796 	if (NULL == bh)
797 		return -1;
798 	super_tmp = (struct btrfs_super_block *)
799 	    (bh->b_data + (dev_bytenr & (BTRFS_BDEV_BLOCKSIZE - 1)));
800 
801 	if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
802 	    btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
803 	    memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
804 	    btrfs_super_nodesize(super_tmp) != state->metablock_size ||
805 	    btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
806 		brelse(bh);
807 		return 0;
808 	}
809 
810 	superblock_tmp =
811 	    btrfsic_block_hashtable_lookup(superblock_bdev,
812 					   dev_bytenr,
813 					   &state->block_hashtable);
814 	if (NULL == superblock_tmp) {
815 		superblock_tmp = btrfsic_block_alloc();
816 		if (NULL == superblock_tmp) {
817 			pr_info("btrfsic: error, kmalloc failed!\n");
818 			brelse(bh);
819 			return -1;
820 		}
821 		/* for superblock, only the dev_bytenr makes sense */
822 		superblock_tmp->dev_bytenr = dev_bytenr;
823 		superblock_tmp->dev_state = dev_state;
824 		superblock_tmp->logical_bytenr = dev_bytenr;
825 		superblock_tmp->generation = btrfs_super_generation(super_tmp);
826 		superblock_tmp->is_metadata = 1;
827 		superblock_tmp->is_superblock = 1;
828 		superblock_tmp->is_iodone = 1;
829 		superblock_tmp->never_written = 0;
830 		superblock_tmp->mirror_num = 1 + superblock_mirror_num;
831 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
832 			btrfs_info_in_rcu(fs_info,
833 				"new initial S-block (bdev %p, %s) @%llu (%s/%llu/%d)",
834 				     superblock_bdev,
835 				     rcu_str_deref(device->name), dev_bytenr,
836 				     dev_state->name, dev_bytenr,
837 				     superblock_mirror_num);
838 		list_add(&superblock_tmp->all_blocks_node,
839 			 &state->all_blocks_list);
840 		btrfsic_block_hashtable_add(superblock_tmp,
841 					    &state->block_hashtable);
842 	}
843 
844 	/* select the one with the highest generation field */
845 	if (btrfs_super_generation(super_tmp) >
846 	    state->max_superblock_generation ||
847 	    0 == state->max_superblock_generation) {
848 		memcpy(selected_super, super_tmp, sizeof(*selected_super));
849 		*selected_dev_state = dev_state;
850 		state->max_superblock_generation =
851 		    btrfs_super_generation(super_tmp);
852 		state->latest_superblock = superblock_tmp;
853 	}
854 
855 	for (pass = 0; pass < 3; pass++) {
856 		u64 next_bytenr;
857 		int num_copies;
858 		int mirror_num;
859 		const char *additional_string = NULL;
860 		struct btrfs_disk_key tmp_disk_key;
861 
862 		tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
863 		tmp_disk_key.offset = 0;
864 		switch (pass) {
865 		case 0:
866 			btrfs_set_disk_key_objectid(&tmp_disk_key,
867 						    BTRFS_ROOT_TREE_OBJECTID);
868 			additional_string = "initial root ";
869 			next_bytenr = btrfs_super_root(super_tmp);
870 			break;
871 		case 1:
872 			btrfs_set_disk_key_objectid(&tmp_disk_key,
873 						    BTRFS_CHUNK_TREE_OBJECTID);
874 			additional_string = "initial chunk ";
875 			next_bytenr = btrfs_super_chunk_root(super_tmp);
876 			break;
877 		case 2:
878 			btrfs_set_disk_key_objectid(&tmp_disk_key,
879 						    BTRFS_TREE_LOG_OBJECTID);
880 			additional_string = "initial log ";
881 			next_bytenr = btrfs_super_log_root(super_tmp);
882 			if (0 == next_bytenr)
883 				continue;
884 			break;
885 		}
886 
887 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
888 					      state->metablock_size);
889 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
890 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
891 			       next_bytenr, num_copies);
892 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
893 			struct btrfsic_block *next_block;
894 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
895 			struct btrfsic_block_link *l;
896 
897 			if (btrfsic_map_block(state, next_bytenr,
898 					      state->metablock_size,
899 					      &tmp_next_block_ctx,
900 					      mirror_num)) {
901 				pr_info("btrfsic: btrfsic_map_block(bytenr @%llu, mirror %d) failed!\n",
902 				       next_bytenr, mirror_num);
903 				brelse(bh);
904 				return -1;
905 			}
906 
907 			next_block = btrfsic_block_lookup_or_add(
908 					state, &tmp_next_block_ctx,
909 					additional_string, 1, 1, 0,
910 					mirror_num, NULL);
911 			if (NULL == next_block) {
912 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
913 				brelse(bh);
914 				return -1;
915 			}
916 
917 			next_block->disk_key = tmp_disk_key;
918 			next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
919 			l = btrfsic_block_link_lookup_or_add(
920 					state, &tmp_next_block_ctx,
921 					next_block, superblock_tmp,
922 					BTRFSIC_GENERATION_UNKNOWN);
923 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
924 			if (NULL == l) {
925 				brelse(bh);
926 				return -1;
927 			}
928 		}
929 	}
930 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
931 		btrfsic_dump_tree_sub(state, superblock_tmp, 0);
932 
933 	brelse(bh);
934 	return 0;
935 }
936 
937 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
938 {
939 	struct btrfsic_stack_frame *sf;
940 
941 	sf = kzalloc(sizeof(*sf), GFP_NOFS);
942 	if (NULL == sf)
943 		pr_info("btrfsic: alloc memory failed!\n");
944 	else
945 		sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
946 	return sf;
947 }
948 
949 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
950 {
951 	BUG_ON(!(NULL == sf ||
952 		 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
953 	kfree(sf);
954 }
955 
956 static int btrfsic_process_metablock(
957 		struct btrfsic_state *state,
958 		struct btrfsic_block *const first_block,
959 		struct btrfsic_block_data_ctx *const first_block_ctx,
960 		int first_limit_nesting, int force_iodone_flag)
961 {
962 	struct btrfsic_stack_frame initial_stack_frame = { 0 };
963 	struct btrfsic_stack_frame *sf;
964 	struct btrfsic_stack_frame *next_stack;
965 	struct btrfs_header *const first_hdr =
966 		(struct btrfs_header *)first_block_ctx->datav[0];
967 
968 	BUG_ON(!first_hdr);
969 	sf = &initial_stack_frame;
970 	sf->error = 0;
971 	sf->i = -1;
972 	sf->limit_nesting = first_limit_nesting;
973 	sf->block = first_block;
974 	sf->block_ctx = first_block_ctx;
975 	sf->next_block = NULL;
976 	sf->hdr = first_hdr;
977 	sf->prev = NULL;
978 
979 continue_with_new_stack_frame:
980 	sf->block->generation = le64_to_cpu(sf->hdr->generation);
981 	if (0 == sf->hdr->level) {
982 		struct btrfs_leaf *const leafhdr =
983 		    (struct btrfs_leaf *)sf->hdr;
984 
985 		if (-1 == sf->i) {
986 			sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
987 
988 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
989 				pr_info("leaf %llu items %d generation %llu owner %llu\n",
990 				       sf->block_ctx->start, sf->nr,
991 				       btrfs_stack_header_generation(
992 					       &leafhdr->header),
993 				       btrfs_stack_header_owner(
994 					       &leafhdr->header));
995 		}
996 
997 continue_with_current_leaf_stack_frame:
998 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
999 			sf->i++;
1000 			sf->num_copies = 0;
1001 		}
1002 
1003 		if (sf->i < sf->nr) {
1004 			struct btrfs_item disk_item;
1005 			u32 disk_item_offset =
1006 				(uintptr_t)(leafhdr->items + sf->i) -
1007 				(uintptr_t)leafhdr;
1008 			struct btrfs_disk_key *disk_key;
1009 			u8 type;
1010 			u32 item_offset;
1011 			u32 item_size;
1012 
1013 			if (disk_item_offset + sizeof(struct btrfs_item) >
1014 			    sf->block_ctx->len) {
1015 leaf_item_out_of_bounce_error:
1016 				pr_info("btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
1017 				       sf->block_ctx->start,
1018 				       sf->block_ctx->dev->name);
1019 				goto one_stack_frame_backwards;
1020 			}
1021 			btrfsic_read_from_block_data(sf->block_ctx,
1022 						     &disk_item,
1023 						     disk_item_offset,
1024 						     sizeof(struct btrfs_item));
1025 			item_offset = btrfs_stack_item_offset(&disk_item);
1026 			item_size = btrfs_stack_item_size(&disk_item);
1027 			disk_key = &disk_item.key;
1028 			type = btrfs_disk_key_type(disk_key);
1029 
1030 			if (BTRFS_ROOT_ITEM_KEY == type) {
1031 				struct btrfs_root_item root_item;
1032 				u32 root_item_offset;
1033 				u64 next_bytenr;
1034 
1035 				root_item_offset = item_offset +
1036 					offsetof(struct btrfs_leaf, items);
1037 				if (root_item_offset + item_size >
1038 				    sf->block_ctx->len)
1039 					goto leaf_item_out_of_bounce_error;
1040 				btrfsic_read_from_block_data(
1041 					sf->block_ctx, &root_item,
1042 					root_item_offset,
1043 					item_size);
1044 				next_bytenr = btrfs_root_bytenr(&root_item);
1045 
1046 				sf->error =
1047 				    btrfsic_create_link_to_next_block(
1048 						state,
1049 						sf->block,
1050 						sf->block_ctx,
1051 						next_bytenr,
1052 						sf->limit_nesting,
1053 						&sf->next_block_ctx,
1054 						&sf->next_block,
1055 						force_iodone_flag,
1056 						&sf->num_copies,
1057 						&sf->mirror_num,
1058 						disk_key,
1059 						btrfs_root_generation(
1060 						&root_item));
1061 				if (sf->error)
1062 					goto one_stack_frame_backwards;
1063 
1064 				if (NULL != sf->next_block) {
1065 					struct btrfs_header *const next_hdr =
1066 					    (struct btrfs_header *)
1067 					    sf->next_block_ctx.datav[0];
1068 
1069 					next_stack =
1070 					    btrfsic_stack_frame_alloc();
1071 					if (NULL == next_stack) {
1072 						sf->error = -1;
1073 						btrfsic_release_block_ctx(
1074 								&sf->
1075 								next_block_ctx);
1076 						goto one_stack_frame_backwards;
1077 					}
1078 
1079 					next_stack->i = -1;
1080 					next_stack->block = sf->next_block;
1081 					next_stack->block_ctx =
1082 					    &sf->next_block_ctx;
1083 					next_stack->next_block = NULL;
1084 					next_stack->hdr = next_hdr;
1085 					next_stack->limit_nesting =
1086 					    sf->limit_nesting - 1;
1087 					next_stack->prev = sf;
1088 					sf = next_stack;
1089 					goto continue_with_new_stack_frame;
1090 				}
1091 			} else if (BTRFS_EXTENT_DATA_KEY == type &&
1092 				   state->include_extent_data) {
1093 				sf->error = btrfsic_handle_extent_data(
1094 						state,
1095 						sf->block,
1096 						sf->block_ctx,
1097 						item_offset,
1098 						force_iodone_flag);
1099 				if (sf->error)
1100 					goto one_stack_frame_backwards;
1101 			}
1102 
1103 			goto continue_with_current_leaf_stack_frame;
1104 		}
1105 	} else {
1106 		struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1107 
1108 		if (-1 == sf->i) {
1109 			sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1110 
1111 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1112 				pr_info("node %llu level %d items %d generation %llu owner %llu\n",
1113 				       sf->block_ctx->start,
1114 				       nodehdr->header.level, sf->nr,
1115 				       btrfs_stack_header_generation(
1116 				       &nodehdr->header),
1117 				       btrfs_stack_header_owner(
1118 				       &nodehdr->header));
1119 		}
1120 
1121 continue_with_current_node_stack_frame:
1122 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1123 			sf->i++;
1124 			sf->num_copies = 0;
1125 		}
1126 
1127 		if (sf->i < sf->nr) {
1128 			struct btrfs_key_ptr key_ptr;
1129 			u32 key_ptr_offset;
1130 			u64 next_bytenr;
1131 
1132 			key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1133 					  (uintptr_t)nodehdr;
1134 			if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1135 			    sf->block_ctx->len) {
1136 				pr_info("btrfsic: node item out of bounce at logical %llu, dev %s\n",
1137 				       sf->block_ctx->start,
1138 				       sf->block_ctx->dev->name);
1139 				goto one_stack_frame_backwards;
1140 			}
1141 			btrfsic_read_from_block_data(
1142 				sf->block_ctx, &key_ptr, key_ptr_offset,
1143 				sizeof(struct btrfs_key_ptr));
1144 			next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1145 
1146 			sf->error = btrfsic_create_link_to_next_block(
1147 					state,
1148 					sf->block,
1149 					sf->block_ctx,
1150 					next_bytenr,
1151 					sf->limit_nesting,
1152 					&sf->next_block_ctx,
1153 					&sf->next_block,
1154 					force_iodone_flag,
1155 					&sf->num_copies,
1156 					&sf->mirror_num,
1157 					&key_ptr.key,
1158 					btrfs_stack_key_generation(&key_ptr));
1159 			if (sf->error)
1160 				goto one_stack_frame_backwards;
1161 
1162 			if (NULL != sf->next_block) {
1163 				struct btrfs_header *const next_hdr =
1164 				    (struct btrfs_header *)
1165 				    sf->next_block_ctx.datav[0];
1166 
1167 				next_stack = btrfsic_stack_frame_alloc();
1168 				if (NULL == next_stack) {
1169 					sf->error = -1;
1170 					goto one_stack_frame_backwards;
1171 				}
1172 
1173 				next_stack->i = -1;
1174 				next_stack->block = sf->next_block;
1175 				next_stack->block_ctx = &sf->next_block_ctx;
1176 				next_stack->next_block = NULL;
1177 				next_stack->hdr = next_hdr;
1178 				next_stack->limit_nesting =
1179 				    sf->limit_nesting - 1;
1180 				next_stack->prev = sf;
1181 				sf = next_stack;
1182 				goto continue_with_new_stack_frame;
1183 			}
1184 
1185 			goto continue_with_current_node_stack_frame;
1186 		}
1187 	}
1188 
1189 one_stack_frame_backwards:
1190 	if (NULL != sf->prev) {
1191 		struct btrfsic_stack_frame *const prev = sf->prev;
1192 
1193 		/* the one for the initial block is freed in the caller */
1194 		btrfsic_release_block_ctx(sf->block_ctx);
1195 
1196 		if (sf->error) {
1197 			prev->error = sf->error;
1198 			btrfsic_stack_frame_free(sf);
1199 			sf = prev;
1200 			goto one_stack_frame_backwards;
1201 		}
1202 
1203 		btrfsic_stack_frame_free(sf);
1204 		sf = prev;
1205 		goto continue_with_new_stack_frame;
1206 	} else {
1207 		BUG_ON(&initial_stack_frame != sf);
1208 	}
1209 
1210 	return sf->error;
1211 }
1212 
1213 static void btrfsic_read_from_block_data(
1214 	struct btrfsic_block_data_ctx *block_ctx,
1215 	void *dstv, u32 offset, size_t len)
1216 {
1217 	size_t cur;
1218 	size_t offset_in_page;
1219 	char *kaddr;
1220 	char *dst = (char *)dstv;
1221 	size_t start_offset = block_ctx->start & ((u64)PAGE_SIZE - 1);
1222 	unsigned long i = (start_offset + offset) >> PAGE_SHIFT;
1223 
1224 	WARN_ON(offset + len > block_ctx->len);
1225 	offset_in_page = (start_offset + offset) & (PAGE_SIZE - 1);
1226 
1227 	while (len > 0) {
1228 		cur = min(len, ((size_t)PAGE_SIZE - offset_in_page));
1229 		BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_SIZE));
1230 		kaddr = block_ctx->datav[i];
1231 		memcpy(dst, kaddr + offset_in_page, cur);
1232 
1233 		dst += cur;
1234 		len -= cur;
1235 		offset_in_page = 0;
1236 		i++;
1237 	}
1238 }
1239 
1240 static int btrfsic_create_link_to_next_block(
1241 		struct btrfsic_state *state,
1242 		struct btrfsic_block *block,
1243 		struct btrfsic_block_data_ctx *block_ctx,
1244 		u64 next_bytenr,
1245 		int limit_nesting,
1246 		struct btrfsic_block_data_ctx *next_block_ctx,
1247 		struct btrfsic_block **next_blockp,
1248 		int force_iodone_flag,
1249 		int *num_copiesp, int *mirror_nump,
1250 		struct btrfs_disk_key *disk_key,
1251 		u64 parent_generation)
1252 {
1253 	struct btrfs_fs_info *fs_info = state->fs_info;
1254 	struct btrfsic_block *next_block = NULL;
1255 	int ret;
1256 	struct btrfsic_block_link *l;
1257 	int did_alloc_block_link;
1258 	int block_was_created;
1259 
1260 	*next_blockp = NULL;
1261 	if (0 == *num_copiesp) {
1262 		*num_copiesp = btrfs_num_copies(fs_info, next_bytenr,
1263 						state->metablock_size);
1264 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1265 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
1266 			       next_bytenr, *num_copiesp);
1267 		*mirror_nump = 1;
1268 	}
1269 
1270 	if (*mirror_nump > *num_copiesp)
1271 		return 0;
1272 
1273 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1274 		pr_info("btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1275 		       *mirror_nump);
1276 	ret = btrfsic_map_block(state, next_bytenr,
1277 				state->metablock_size,
1278 				next_block_ctx, *mirror_nump);
1279 	if (ret) {
1280 		pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1281 		       next_bytenr, *mirror_nump);
1282 		btrfsic_release_block_ctx(next_block_ctx);
1283 		*next_blockp = NULL;
1284 		return -1;
1285 	}
1286 
1287 	next_block = btrfsic_block_lookup_or_add(state,
1288 						 next_block_ctx, "referenced ",
1289 						 1, force_iodone_flag,
1290 						 !force_iodone_flag,
1291 						 *mirror_nump,
1292 						 &block_was_created);
1293 	if (NULL == next_block) {
1294 		btrfsic_release_block_ctx(next_block_ctx);
1295 		*next_blockp = NULL;
1296 		return -1;
1297 	}
1298 	if (block_was_created) {
1299 		l = NULL;
1300 		next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1301 	} else {
1302 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1303 			if (next_block->logical_bytenr != next_bytenr &&
1304 			    !(!next_block->is_metadata &&
1305 			      0 == next_block->logical_bytenr))
1306 				pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1307 				       next_bytenr, next_block_ctx->dev->name,
1308 				       next_block_ctx->dev_bytenr, *mirror_nump,
1309 				       btrfsic_get_block_type(state,
1310 							      next_block),
1311 				       next_block->logical_bytenr);
1312 			else
1313 				pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1314 				       next_bytenr, next_block_ctx->dev->name,
1315 				       next_block_ctx->dev_bytenr, *mirror_nump,
1316 				       btrfsic_get_block_type(state,
1317 							      next_block));
1318 		}
1319 		next_block->logical_bytenr = next_bytenr;
1320 
1321 		next_block->mirror_num = *mirror_nump;
1322 		l = btrfsic_block_link_hashtable_lookup(
1323 				next_block_ctx->dev->bdev,
1324 				next_block_ctx->dev_bytenr,
1325 				block_ctx->dev->bdev,
1326 				block_ctx->dev_bytenr,
1327 				&state->block_link_hashtable);
1328 	}
1329 
1330 	next_block->disk_key = *disk_key;
1331 	if (NULL == l) {
1332 		l = btrfsic_block_link_alloc();
1333 		if (NULL == l) {
1334 			pr_info("btrfsic: error, kmalloc failed!\n");
1335 			btrfsic_release_block_ctx(next_block_ctx);
1336 			*next_blockp = NULL;
1337 			return -1;
1338 		}
1339 
1340 		did_alloc_block_link = 1;
1341 		l->block_ref_to = next_block;
1342 		l->block_ref_from = block;
1343 		l->ref_cnt = 1;
1344 		l->parent_generation = parent_generation;
1345 
1346 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1347 			btrfsic_print_add_link(state, l);
1348 
1349 		list_add(&l->node_ref_to, &block->ref_to_list);
1350 		list_add(&l->node_ref_from, &next_block->ref_from_list);
1351 
1352 		btrfsic_block_link_hashtable_add(l,
1353 						 &state->block_link_hashtable);
1354 	} else {
1355 		did_alloc_block_link = 0;
1356 		if (0 == limit_nesting) {
1357 			l->ref_cnt++;
1358 			l->parent_generation = parent_generation;
1359 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1360 				btrfsic_print_add_link(state, l);
1361 		}
1362 	}
1363 
1364 	if (limit_nesting > 0 && did_alloc_block_link) {
1365 		ret = btrfsic_read_block(state, next_block_ctx);
1366 		if (ret < (int)next_block_ctx->len) {
1367 			pr_info("btrfsic: read block @logical %llu failed!\n",
1368 			       next_bytenr);
1369 			btrfsic_release_block_ctx(next_block_ctx);
1370 			*next_blockp = NULL;
1371 			return -1;
1372 		}
1373 
1374 		*next_blockp = next_block;
1375 	} else {
1376 		*next_blockp = NULL;
1377 	}
1378 	(*mirror_nump)++;
1379 
1380 	return 0;
1381 }
1382 
1383 static int btrfsic_handle_extent_data(
1384 		struct btrfsic_state *state,
1385 		struct btrfsic_block *block,
1386 		struct btrfsic_block_data_ctx *block_ctx,
1387 		u32 item_offset, int force_iodone_flag)
1388 {
1389 	struct btrfs_fs_info *fs_info = state->fs_info;
1390 	struct btrfs_file_extent_item file_extent_item;
1391 	u64 file_extent_item_offset;
1392 	u64 next_bytenr;
1393 	u64 num_bytes;
1394 	u64 generation;
1395 	struct btrfsic_block_link *l;
1396 	int ret;
1397 
1398 	file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1399 				  item_offset;
1400 	if (file_extent_item_offset +
1401 	    offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1402 	    block_ctx->len) {
1403 		pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1404 		       block_ctx->start, block_ctx->dev->name);
1405 		return -1;
1406 	}
1407 
1408 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1409 		file_extent_item_offset,
1410 		offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1411 	if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1412 	    btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1413 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1414 			pr_info("extent_data: type %u, disk_bytenr = %llu\n",
1415 			       file_extent_item.type,
1416 			       btrfs_stack_file_extent_disk_bytenr(
1417 			       &file_extent_item));
1418 		return 0;
1419 	}
1420 
1421 	if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1422 	    block_ctx->len) {
1423 		pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1424 		       block_ctx->start, block_ctx->dev->name);
1425 		return -1;
1426 	}
1427 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1428 				     file_extent_item_offset,
1429 				     sizeof(struct btrfs_file_extent_item));
1430 	next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item);
1431 	if (btrfs_stack_file_extent_compression(&file_extent_item) ==
1432 	    BTRFS_COMPRESS_NONE) {
1433 		next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item);
1434 		num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1435 	} else {
1436 		num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item);
1437 	}
1438 	generation = btrfs_stack_file_extent_generation(&file_extent_item);
1439 
1440 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1441 		pr_info("extent_data: type %u, disk_bytenr = %llu, offset = %llu, num_bytes = %llu\n",
1442 		       file_extent_item.type,
1443 		       btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1444 		       btrfs_stack_file_extent_offset(&file_extent_item),
1445 		       num_bytes);
1446 	while (num_bytes > 0) {
1447 		u32 chunk_len;
1448 		int num_copies;
1449 		int mirror_num;
1450 
1451 		if (num_bytes > state->datablock_size)
1452 			chunk_len = state->datablock_size;
1453 		else
1454 			chunk_len = num_bytes;
1455 
1456 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
1457 					      state->datablock_size);
1458 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1459 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
1460 			       next_bytenr, num_copies);
1461 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1462 			struct btrfsic_block_data_ctx next_block_ctx;
1463 			struct btrfsic_block *next_block;
1464 			int block_was_created;
1465 
1466 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1467 				pr_info("btrfsic_handle_extent_data(mirror_num=%d)\n",
1468 					mirror_num);
1469 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1470 				pr_info("\tdisk_bytenr = %llu, num_bytes %u\n",
1471 				       next_bytenr, chunk_len);
1472 			ret = btrfsic_map_block(state, next_bytenr,
1473 						chunk_len, &next_block_ctx,
1474 						mirror_num);
1475 			if (ret) {
1476 				pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1477 				       next_bytenr, mirror_num);
1478 				return -1;
1479 			}
1480 
1481 			next_block = btrfsic_block_lookup_or_add(
1482 					state,
1483 					&next_block_ctx,
1484 					"referenced ",
1485 					0,
1486 					force_iodone_flag,
1487 					!force_iodone_flag,
1488 					mirror_num,
1489 					&block_was_created);
1490 			if (NULL == next_block) {
1491 				pr_info("btrfsic: error, kmalloc failed!\n");
1492 				btrfsic_release_block_ctx(&next_block_ctx);
1493 				return -1;
1494 			}
1495 			if (!block_was_created) {
1496 				if ((state->print_mask &
1497 				     BTRFSIC_PRINT_MASK_VERBOSE) &&
1498 				    next_block->logical_bytenr != next_bytenr &&
1499 				    !(!next_block->is_metadata &&
1500 				      0 == next_block->logical_bytenr)) {
1501 					pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, D, bytenr mismatch (!= stored %llu).\n",
1502 					       next_bytenr,
1503 					       next_block_ctx.dev->name,
1504 					       next_block_ctx.dev_bytenr,
1505 					       mirror_num,
1506 					       next_block->logical_bytenr);
1507 				}
1508 				next_block->logical_bytenr = next_bytenr;
1509 				next_block->mirror_num = mirror_num;
1510 			}
1511 
1512 			l = btrfsic_block_link_lookup_or_add(state,
1513 							     &next_block_ctx,
1514 							     next_block, block,
1515 							     generation);
1516 			btrfsic_release_block_ctx(&next_block_ctx);
1517 			if (NULL == l)
1518 				return -1;
1519 		}
1520 
1521 		next_bytenr += chunk_len;
1522 		num_bytes -= chunk_len;
1523 	}
1524 
1525 	return 0;
1526 }
1527 
1528 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1529 			     struct btrfsic_block_data_ctx *block_ctx_out,
1530 			     int mirror_num)
1531 {
1532 	struct btrfs_fs_info *fs_info = state->fs_info;
1533 	int ret;
1534 	u64 length;
1535 	struct btrfs_bio *multi = NULL;
1536 	struct btrfs_device *device;
1537 
1538 	length = len;
1539 	ret = btrfs_map_block(fs_info, BTRFS_MAP_READ,
1540 			      bytenr, &length, &multi, mirror_num);
1541 
1542 	if (ret) {
1543 		block_ctx_out->start = 0;
1544 		block_ctx_out->dev_bytenr = 0;
1545 		block_ctx_out->len = 0;
1546 		block_ctx_out->dev = NULL;
1547 		block_ctx_out->datav = NULL;
1548 		block_ctx_out->pagev = NULL;
1549 		block_ctx_out->mem_to_free = NULL;
1550 
1551 		return ret;
1552 	}
1553 
1554 	device = multi->stripes[0].dev;
1555 	block_ctx_out->dev = btrfsic_dev_state_lookup(device->bdev->bd_dev);
1556 	block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1557 	block_ctx_out->start = bytenr;
1558 	block_ctx_out->len = len;
1559 	block_ctx_out->datav = NULL;
1560 	block_ctx_out->pagev = NULL;
1561 	block_ctx_out->mem_to_free = NULL;
1562 
1563 	kfree(multi);
1564 	if (NULL == block_ctx_out->dev) {
1565 		ret = -ENXIO;
1566 		pr_info("btrfsic: error, cannot lookup dev (#1)!\n");
1567 	}
1568 
1569 	return ret;
1570 }
1571 
1572 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1573 {
1574 	if (block_ctx->mem_to_free) {
1575 		unsigned int num_pages;
1576 
1577 		BUG_ON(!block_ctx->datav);
1578 		BUG_ON(!block_ctx->pagev);
1579 		num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1580 			    PAGE_SHIFT;
1581 		while (num_pages > 0) {
1582 			num_pages--;
1583 			if (block_ctx->datav[num_pages]) {
1584 				kunmap(block_ctx->pagev[num_pages]);
1585 				block_ctx->datav[num_pages] = NULL;
1586 			}
1587 			if (block_ctx->pagev[num_pages]) {
1588 				__free_page(block_ctx->pagev[num_pages]);
1589 				block_ctx->pagev[num_pages] = NULL;
1590 			}
1591 		}
1592 
1593 		kfree(block_ctx->mem_to_free);
1594 		block_ctx->mem_to_free = NULL;
1595 		block_ctx->pagev = NULL;
1596 		block_ctx->datav = NULL;
1597 	}
1598 }
1599 
1600 static int btrfsic_read_block(struct btrfsic_state *state,
1601 			      struct btrfsic_block_data_ctx *block_ctx)
1602 {
1603 	unsigned int num_pages;
1604 	unsigned int i;
1605 	u64 dev_bytenr;
1606 	int ret;
1607 
1608 	BUG_ON(block_ctx->datav);
1609 	BUG_ON(block_ctx->pagev);
1610 	BUG_ON(block_ctx->mem_to_free);
1611 	if (block_ctx->dev_bytenr & ((u64)PAGE_SIZE - 1)) {
1612 		pr_info("btrfsic: read_block() with unaligned bytenr %llu\n",
1613 		       block_ctx->dev_bytenr);
1614 		return -1;
1615 	}
1616 
1617 	num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1618 		    PAGE_SHIFT;
1619 	block_ctx->mem_to_free = kzalloc((sizeof(*block_ctx->datav) +
1620 					  sizeof(*block_ctx->pagev)) *
1621 					 num_pages, GFP_NOFS);
1622 	if (!block_ctx->mem_to_free)
1623 		return -ENOMEM;
1624 	block_ctx->datav = block_ctx->mem_to_free;
1625 	block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1626 	for (i = 0; i < num_pages; i++) {
1627 		block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1628 		if (!block_ctx->pagev[i])
1629 			return -1;
1630 	}
1631 
1632 	dev_bytenr = block_ctx->dev_bytenr;
1633 	for (i = 0; i < num_pages;) {
1634 		struct bio *bio;
1635 		unsigned int j;
1636 
1637 		bio = btrfs_io_bio_alloc(num_pages - i);
1638 		bio_set_dev(bio, block_ctx->dev->bdev);
1639 		bio->bi_iter.bi_sector = dev_bytenr >> 9;
1640 		bio_set_op_attrs(bio, REQ_OP_READ, 0);
1641 
1642 		for (j = i; j < num_pages; j++) {
1643 			ret = bio_add_page(bio, block_ctx->pagev[j],
1644 					   PAGE_SIZE, 0);
1645 			if (PAGE_SIZE != ret)
1646 				break;
1647 		}
1648 		if (j == i) {
1649 			pr_info("btrfsic: error, failed to add a single page!\n");
1650 			return -1;
1651 		}
1652 		if (submit_bio_wait(bio)) {
1653 			pr_info("btrfsic: read error at logical %llu dev %s!\n",
1654 			       block_ctx->start, block_ctx->dev->name);
1655 			bio_put(bio);
1656 			return -1;
1657 		}
1658 		bio_put(bio);
1659 		dev_bytenr += (j - i) * PAGE_SIZE;
1660 		i = j;
1661 	}
1662 	for (i = 0; i < num_pages; i++)
1663 		block_ctx->datav[i] = kmap(block_ctx->pagev[i]);
1664 
1665 	return block_ctx->len;
1666 }
1667 
1668 static void btrfsic_dump_database(struct btrfsic_state *state)
1669 {
1670 	const struct btrfsic_block *b_all;
1671 
1672 	BUG_ON(NULL == state);
1673 
1674 	pr_info("all_blocks_list:\n");
1675 	list_for_each_entry(b_all, &state->all_blocks_list, all_blocks_node) {
1676 		const struct btrfsic_block_link *l;
1677 
1678 		pr_info("%c-block @%llu (%s/%llu/%d)\n",
1679 		       btrfsic_get_block_type(state, b_all),
1680 		       b_all->logical_bytenr, b_all->dev_state->name,
1681 		       b_all->dev_bytenr, b_all->mirror_num);
1682 
1683 		list_for_each_entry(l, &b_all->ref_to_list, node_ref_to) {
1684 			pr_info(" %c @%llu (%s/%llu/%d) refers %u* to %c @%llu (%s/%llu/%d)\n",
1685 			       btrfsic_get_block_type(state, b_all),
1686 			       b_all->logical_bytenr, b_all->dev_state->name,
1687 			       b_all->dev_bytenr, b_all->mirror_num,
1688 			       l->ref_cnt,
1689 			       btrfsic_get_block_type(state, l->block_ref_to),
1690 			       l->block_ref_to->logical_bytenr,
1691 			       l->block_ref_to->dev_state->name,
1692 			       l->block_ref_to->dev_bytenr,
1693 			       l->block_ref_to->mirror_num);
1694 		}
1695 
1696 		list_for_each_entry(l, &b_all->ref_from_list, node_ref_from) {
1697 			pr_info(" %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
1698 			       btrfsic_get_block_type(state, b_all),
1699 			       b_all->logical_bytenr, b_all->dev_state->name,
1700 			       b_all->dev_bytenr, b_all->mirror_num,
1701 			       l->ref_cnt,
1702 			       btrfsic_get_block_type(state, l->block_ref_from),
1703 			       l->block_ref_from->logical_bytenr,
1704 			       l->block_ref_from->dev_state->name,
1705 			       l->block_ref_from->dev_bytenr,
1706 			       l->block_ref_from->mirror_num);
1707 		}
1708 
1709 		pr_info("\n");
1710 	}
1711 }
1712 
1713 /*
1714  * Test whether the disk block contains a tree block (leaf or node)
1715  * (note that this test fails for the super block)
1716  */
1717 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
1718 				     char **datav, unsigned int num_pages)
1719 {
1720 	struct btrfs_fs_info *fs_info = state->fs_info;
1721 	struct btrfs_header *h;
1722 	u8 csum[BTRFS_CSUM_SIZE];
1723 	u32 crc = ~(u32)0;
1724 	unsigned int i;
1725 
1726 	if (num_pages * PAGE_SIZE < state->metablock_size)
1727 		return 1; /* not metadata */
1728 	num_pages = state->metablock_size >> PAGE_SHIFT;
1729 	h = (struct btrfs_header *)datav[0];
1730 
1731 	if (memcmp(h->fsid, fs_info->fsid, BTRFS_FSID_SIZE))
1732 		return 1;
1733 
1734 	for (i = 0; i < num_pages; i++) {
1735 		u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1736 		size_t sublen = i ? PAGE_SIZE :
1737 				    (PAGE_SIZE - BTRFS_CSUM_SIZE);
1738 
1739 		crc = btrfs_crc32c(crc, data, sublen);
1740 	}
1741 	btrfs_csum_final(crc, csum);
1742 	if (memcmp(csum, h->csum, state->csum_size))
1743 		return 1;
1744 
1745 	return 0; /* is metadata */
1746 }
1747 
1748 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1749 					  u64 dev_bytenr, char **mapped_datav,
1750 					  unsigned int num_pages,
1751 					  struct bio *bio, int *bio_is_patched,
1752 					  struct buffer_head *bh,
1753 					  int submit_bio_bh_rw)
1754 {
1755 	int is_metadata;
1756 	struct btrfsic_block *block;
1757 	struct btrfsic_block_data_ctx block_ctx;
1758 	int ret;
1759 	struct btrfsic_state *state = dev_state->state;
1760 	struct block_device *bdev = dev_state->bdev;
1761 	unsigned int processed_len;
1762 
1763 	if (NULL != bio_is_patched)
1764 		*bio_is_patched = 0;
1765 
1766 again:
1767 	if (num_pages == 0)
1768 		return;
1769 
1770 	processed_len = 0;
1771 	is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1772 						      num_pages));
1773 
1774 	block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1775 					       &state->block_hashtable);
1776 	if (NULL != block) {
1777 		u64 bytenr = 0;
1778 		struct btrfsic_block_link *l, *tmp;
1779 
1780 		if (block->is_superblock) {
1781 			bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1782 						    mapped_datav[0]);
1783 			if (num_pages * PAGE_SIZE <
1784 			    BTRFS_SUPER_INFO_SIZE) {
1785 				pr_info("btrfsic: cannot work with too short bios!\n");
1786 				return;
1787 			}
1788 			is_metadata = 1;
1789 			BUG_ON(BTRFS_SUPER_INFO_SIZE & (PAGE_SIZE - 1));
1790 			processed_len = BTRFS_SUPER_INFO_SIZE;
1791 			if (state->print_mask &
1792 			    BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1793 				pr_info("[before new superblock is written]:\n");
1794 				btrfsic_dump_tree_sub(state, block, 0);
1795 			}
1796 		}
1797 		if (is_metadata) {
1798 			if (!block->is_superblock) {
1799 				if (num_pages * PAGE_SIZE <
1800 				    state->metablock_size) {
1801 					pr_info("btrfsic: cannot work with too short bios!\n");
1802 					return;
1803 				}
1804 				processed_len = state->metablock_size;
1805 				bytenr = btrfs_stack_header_bytenr(
1806 						(struct btrfs_header *)
1807 						mapped_datav[0]);
1808 				btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1809 							       dev_state,
1810 							       dev_bytenr);
1811 			}
1812 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1813 				if (block->logical_bytenr != bytenr &&
1814 				    !(!block->is_metadata &&
1815 				      block->logical_bytenr == 0))
1816 					pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1817 					       bytenr, dev_state->name,
1818 					       dev_bytenr,
1819 					       block->mirror_num,
1820 					       btrfsic_get_block_type(state,
1821 								      block),
1822 					       block->logical_bytenr);
1823 				else
1824 					pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1825 					       bytenr, dev_state->name,
1826 					       dev_bytenr, block->mirror_num,
1827 					       btrfsic_get_block_type(state,
1828 								      block));
1829 			}
1830 			block->logical_bytenr = bytenr;
1831 		} else {
1832 			if (num_pages * PAGE_SIZE <
1833 			    state->datablock_size) {
1834 				pr_info("btrfsic: cannot work with too short bios!\n");
1835 				return;
1836 			}
1837 			processed_len = state->datablock_size;
1838 			bytenr = block->logical_bytenr;
1839 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1840 				pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1841 				       bytenr, dev_state->name, dev_bytenr,
1842 				       block->mirror_num,
1843 				       btrfsic_get_block_type(state, block));
1844 		}
1845 
1846 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1847 			pr_info("ref_to_list: %cE, ref_from_list: %cE\n",
1848 			       list_empty(&block->ref_to_list) ? ' ' : '!',
1849 			       list_empty(&block->ref_from_list) ? ' ' : '!');
1850 		if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1851 			pr_info("btrfs: attempt to overwrite %c-block @%llu (%s/%llu/%d), old(gen=%llu, objectid=%llu, type=%d, offset=%llu), new(gen=%llu), which is referenced by most recent superblock (superblockgen=%llu)!\n",
1852 			       btrfsic_get_block_type(state, block), bytenr,
1853 			       dev_state->name, dev_bytenr, block->mirror_num,
1854 			       block->generation,
1855 			       btrfs_disk_key_objectid(&block->disk_key),
1856 			       block->disk_key.type,
1857 			       btrfs_disk_key_offset(&block->disk_key),
1858 			       btrfs_stack_header_generation(
1859 				       (struct btrfs_header *) mapped_datav[0]),
1860 			       state->max_superblock_generation);
1861 			btrfsic_dump_tree(state);
1862 		}
1863 
1864 		if (!block->is_iodone && !block->never_written) {
1865 			pr_info("btrfs: attempt to overwrite %c-block @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu, which is not yet iodone!\n",
1866 			       btrfsic_get_block_type(state, block), bytenr,
1867 			       dev_state->name, dev_bytenr, block->mirror_num,
1868 			       block->generation,
1869 			       btrfs_stack_header_generation(
1870 				       (struct btrfs_header *)
1871 				       mapped_datav[0]));
1872 			/* it would not be safe to go on */
1873 			btrfsic_dump_tree(state);
1874 			goto continue_loop;
1875 		}
1876 
1877 		/*
1878 		 * Clear all references of this block. Do not free
1879 		 * the block itself even if is not referenced anymore
1880 		 * because it still carries valuable information
1881 		 * like whether it was ever written and IO completed.
1882 		 */
1883 		list_for_each_entry_safe(l, tmp, &block->ref_to_list,
1884 					 node_ref_to) {
1885 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1886 				btrfsic_print_rem_link(state, l);
1887 			l->ref_cnt--;
1888 			if (0 == l->ref_cnt) {
1889 				list_del(&l->node_ref_to);
1890 				list_del(&l->node_ref_from);
1891 				btrfsic_block_link_hashtable_remove(l);
1892 				btrfsic_block_link_free(l);
1893 			}
1894 		}
1895 
1896 		block_ctx.dev = dev_state;
1897 		block_ctx.dev_bytenr = dev_bytenr;
1898 		block_ctx.start = bytenr;
1899 		block_ctx.len = processed_len;
1900 		block_ctx.pagev = NULL;
1901 		block_ctx.mem_to_free = NULL;
1902 		block_ctx.datav = mapped_datav;
1903 
1904 		if (is_metadata || state->include_extent_data) {
1905 			block->never_written = 0;
1906 			block->iodone_w_error = 0;
1907 			if (NULL != bio) {
1908 				block->is_iodone = 0;
1909 				BUG_ON(NULL == bio_is_patched);
1910 				if (!*bio_is_patched) {
1911 					block->orig_bio_bh_private =
1912 					    bio->bi_private;
1913 					block->orig_bio_bh_end_io.bio =
1914 					    bio->bi_end_io;
1915 					block->next_in_same_bio = NULL;
1916 					bio->bi_private = block;
1917 					bio->bi_end_io = btrfsic_bio_end_io;
1918 					*bio_is_patched = 1;
1919 				} else {
1920 					struct btrfsic_block *chained_block =
1921 					    (struct btrfsic_block *)
1922 					    bio->bi_private;
1923 
1924 					BUG_ON(NULL == chained_block);
1925 					block->orig_bio_bh_private =
1926 					    chained_block->orig_bio_bh_private;
1927 					block->orig_bio_bh_end_io.bio =
1928 					    chained_block->orig_bio_bh_end_io.
1929 					    bio;
1930 					block->next_in_same_bio = chained_block;
1931 					bio->bi_private = block;
1932 				}
1933 			} else if (NULL != bh) {
1934 				block->is_iodone = 0;
1935 				block->orig_bio_bh_private = bh->b_private;
1936 				block->orig_bio_bh_end_io.bh = bh->b_end_io;
1937 				block->next_in_same_bio = NULL;
1938 				bh->b_private = block;
1939 				bh->b_end_io = btrfsic_bh_end_io;
1940 			} else {
1941 				block->is_iodone = 1;
1942 				block->orig_bio_bh_private = NULL;
1943 				block->orig_bio_bh_end_io.bio = NULL;
1944 				block->next_in_same_bio = NULL;
1945 			}
1946 		}
1947 
1948 		block->flush_gen = dev_state->last_flush_gen + 1;
1949 		block->submit_bio_bh_rw = submit_bio_bh_rw;
1950 		if (is_metadata) {
1951 			block->logical_bytenr = bytenr;
1952 			block->is_metadata = 1;
1953 			if (block->is_superblock) {
1954 				BUG_ON(PAGE_SIZE !=
1955 				       BTRFS_SUPER_INFO_SIZE);
1956 				ret = btrfsic_process_written_superblock(
1957 						state,
1958 						block,
1959 						(struct btrfs_super_block *)
1960 						mapped_datav[0]);
1961 				if (state->print_mask &
1962 				    BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
1963 					pr_info("[after new superblock is written]:\n");
1964 					btrfsic_dump_tree_sub(state, block, 0);
1965 				}
1966 			} else {
1967 				block->mirror_num = 0;	/* unknown */
1968 				ret = btrfsic_process_metablock(
1969 						state,
1970 						block,
1971 						&block_ctx,
1972 						0, 0);
1973 			}
1974 			if (ret)
1975 				pr_info("btrfsic: btrfsic_process_metablock(root @%llu) failed!\n",
1976 				       dev_bytenr);
1977 		} else {
1978 			block->is_metadata = 0;
1979 			block->mirror_num = 0;	/* unknown */
1980 			block->generation = BTRFSIC_GENERATION_UNKNOWN;
1981 			if (!state->include_extent_data
1982 			    && list_empty(&block->ref_from_list)) {
1983 				/*
1984 				 * disk block is overwritten with extent
1985 				 * data (not meta data) and we are configured
1986 				 * to not include extent data: take the
1987 				 * chance and free the block's memory
1988 				 */
1989 				btrfsic_block_hashtable_remove(block);
1990 				list_del(&block->all_blocks_node);
1991 				btrfsic_block_free(block);
1992 			}
1993 		}
1994 		btrfsic_release_block_ctx(&block_ctx);
1995 	} else {
1996 		/* block has not been found in hash table */
1997 		u64 bytenr;
1998 
1999 		if (!is_metadata) {
2000 			processed_len = state->datablock_size;
2001 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2002 				pr_info("Written block (%s/%llu/?) !found in hash table, D.\n",
2003 				       dev_state->name, dev_bytenr);
2004 			if (!state->include_extent_data) {
2005 				/* ignore that written D block */
2006 				goto continue_loop;
2007 			}
2008 
2009 			/* this is getting ugly for the
2010 			 * include_extent_data case... */
2011 			bytenr = 0;	/* unknown */
2012 		} else {
2013 			processed_len = state->metablock_size;
2014 			bytenr = btrfs_stack_header_bytenr(
2015 					(struct btrfs_header *)
2016 					mapped_datav[0]);
2017 			btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
2018 						       dev_bytenr);
2019 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2020 				pr_info("Written block @%llu (%s/%llu/?) !found in hash table, M.\n",
2021 				       bytenr, dev_state->name, dev_bytenr);
2022 		}
2023 
2024 		block_ctx.dev = dev_state;
2025 		block_ctx.dev_bytenr = dev_bytenr;
2026 		block_ctx.start = bytenr;
2027 		block_ctx.len = processed_len;
2028 		block_ctx.pagev = NULL;
2029 		block_ctx.mem_to_free = NULL;
2030 		block_ctx.datav = mapped_datav;
2031 
2032 		block = btrfsic_block_alloc();
2033 		if (NULL == block) {
2034 			pr_info("btrfsic: error, kmalloc failed!\n");
2035 			btrfsic_release_block_ctx(&block_ctx);
2036 			goto continue_loop;
2037 		}
2038 		block->dev_state = dev_state;
2039 		block->dev_bytenr = dev_bytenr;
2040 		block->logical_bytenr = bytenr;
2041 		block->is_metadata = is_metadata;
2042 		block->never_written = 0;
2043 		block->iodone_w_error = 0;
2044 		block->mirror_num = 0;	/* unknown */
2045 		block->flush_gen = dev_state->last_flush_gen + 1;
2046 		block->submit_bio_bh_rw = submit_bio_bh_rw;
2047 		if (NULL != bio) {
2048 			block->is_iodone = 0;
2049 			BUG_ON(NULL == bio_is_patched);
2050 			if (!*bio_is_patched) {
2051 				block->orig_bio_bh_private = bio->bi_private;
2052 				block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2053 				block->next_in_same_bio = NULL;
2054 				bio->bi_private = block;
2055 				bio->bi_end_io = btrfsic_bio_end_io;
2056 				*bio_is_patched = 1;
2057 			} else {
2058 				struct btrfsic_block *chained_block =
2059 				    (struct btrfsic_block *)
2060 				    bio->bi_private;
2061 
2062 				BUG_ON(NULL == chained_block);
2063 				block->orig_bio_bh_private =
2064 				    chained_block->orig_bio_bh_private;
2065 				block->orig_bio_bh_end_io.bio =
2066 				    chained_block->orig_bio_bh_end_io.bio;
2067 				block->next_in_same_bio = chained_block;
2068 				bio->bi_private = block;
2069 			}
2070 		} else if (NULL != bh) {
2071 			block->is_iodone = 0;
2072 			block->orig_bio_bh_private = bh->b_private;
2073 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2074 			block->next_in_same_bio = NULL;
2075 			bh->b_private = block;
2076 			bh->b_end_io = btrfsic_bh_end_io;
2077 		} else {
2078 			block->is_iodone = 1;
2079 			block->orig_bio_bh_private = NULL;
2080 			block->orig_bio_bh_end_io.bio = NULL;
2081 			block->next_in_same_bio = NULL;
2082 		}
2083 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2084 			pr_info("New written %c-block @%llu (%s/%llu/%d)\n",
2085 			       is_metadata ? 'M' : 'D',
2086 			       block->logical_bytenr, block->dev_state->name,
2087 			       block->dev_bytenr, block->mirror_num);
2088 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2089 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2090 
2091 		if (is_metadata) {
2092 			ret = btrfsic_process_metablock(state, block,
2093 							&block_ctx, 0, 0);
2094 			if (ret)
2095 				pr_info("btrfsic: process_metablock(root @%llu) failed!\n",
2096 				       dev_bytenr);
2097 		}
2098 		btrfsic_release_block_ctx(&block_ctx);
2099 	}
2100 
2101 continue_loop:
2102 	BUG_ON(!processed_len);
2103 	dev_bytenr += processed_len;
2104 	mapped_datav += processed_len >> PAGE_SHIFT;
2105 	num_pages -= processed_len >> PAGE_SHIFT;
2106 	goto again;
2107 }
2108 
2109 static void btrfsic_bio_end_io(struct bio *bp)
2110 {
2111 	struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2112 	int iodone_w_error;
2113 
2114 	/* mutex is not held! This is not save if IO is not yet completed
2115 	 * on umount */
2116 	iodone_w_error = 0;
2117 	if (bp->bi_status)
2118 		iodone_w_error = 1;
2119 
2120 	BUG_ON(NULL == block);
2121 	bp->bi_private = block->orig_bio_bh_private;
2122 	bp->bi_end_io = block->orig_bio_bh_end_io.bio;
2123 
2124 	do {
2125 		struct btrfsic_block *next_block;
2126 		struct btrfsic_dev_state *const dev_state = block->dev_state;
2127 
2128 		if ((dev_state->state->print_mask &
2129 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2130 			pr_info("bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2131 			       bp->bi_status,
2132 			       btrfsic_get_block_type(dev_state->state, block),
2133 			       block->logical_bytenr, dev_state->name,
2134 			       block->dev_bytenr, block->mirror_num);
2135 		next_block = block->next_in_same_bio;
2136 		block->iodone_w_error = iodone_w_error;
2137 		if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2138 			dev_state->last_flush_gen++;
2139 			if ((dev_state->state->print_mask &
2140 			     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2141 				pr_info("bio_end_io() new %s flush_gen=%llu\n",
2142 				       dev_state->name,
2143 				       dev_state->last_flush_gen);
2144 		}
2145 		if (block->submit_bio_bh_rw & REQ_FUA)
2146 			block->flush_gen = 0; /* FUA completed means block is
2147 					       * on disk */
2148 		block->is_iodone = 1; /* for FLUSH, this releases the block */
2149 		block = next_block;
2150 	} while (NULL != block);
2151 
2152 	bp->bi_end_io(bp);
2153 }
2154 
2155 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate)
2156 {
2157 	struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private;
2158 	int iodone_w_error = !uptodate;
2159 	struct btrfsic_dev_state *dev_state;
2160 
2161 	BUG_ON(NULL == block);
2162 	dev_state = block->dev_state;
2163 	if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2164 		pr_info("bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n",
2165 		       iodone_w_error,
2166 		       btrfsic_get_block_type(dev_state->state, block),
2167 		       block->logical_bytenr, block->dev_state->name,
2168 		       block->dev_bytenr, block->mirror_num);
2169 
2170 	block->iodone_w_error = iodone_w_error;
2171 	if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2172 		dev_state->last_flush_gen++;
2173 		if ((dev_state->state->print_mask &
2174 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2175 			pr_info("bh_end_io() new %s flush_gen=%llu\n",
2176 			       dev_state->name, dev_state->last_flush_gen);
2177 	}
2178 	if (block->submit_bio_bh_rw & REQ_FUA)
2179 		block->flush_gen = 0; /* FUA completed means block is on disk */
2180 
2181 	bh->b_private = block->orig_bio_bh_private;
2182 	bh->b_end_io = block->orig_bio_bh_end_io.bh;
2183 	block->is_iodone = 1; /* for FLUSH, this releases the block */
2184 	bh->b_end_io(bh, uptodate);
2185 }
2186 
2187 static int btrfsic_process_written_superblock(
2188 		struct btrfsic_state *state,
2189 		struct btrfsic_block *const superblock,
2190 		struct btrfs_super_block *const super_hdr)
2191 {
2192 	struct btrfs_fs_info *fs_info = state->fs_info;
2193 	int pass;
2194 
2195 	superblock->generation = btrfs_super_generation(super_hdr);
2196 	if (!(superblock->generation > state->max_superblock_generation ||
2197 	      0 == state->max_superblock_generation)) {
2198 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2199 			pr_info("btrfsic: superblock @%llu (%s/%llu/%d) with old gen %llu <= %llu\n",
2200 			       superblock->logical_bytenr,
2201 			       superblock->dev_state->name,
2202 			       superblock->dev_bytenr, superblock->mirror_num,
2203 			       btrfs_super_generation(super_hdr),
2204 			       state->max_superblock_generation);
2205 	} else {
2206 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2207 			pr_info("btrfsic: got new superblock @%llu (%s/%llu/%d) with new gen %llu > %llu\n",
2208 			       superblock->logical_bytenr,
2209 			       superblock->dev_state->name,
2210 			       superblock->dev_bytenr, superblock->mirror_num,
2211 			       btrfs_super_generation(super_hdr),
2212 			       state->max_superblock_generation);
2213 
2214 		state->max_superblock_generation =
2215 		    btrfs_super_generation(super_hdr);
2216 		state->latest_superblock = superblock;
2217 	}
2218 
2219 	for (pass = 0; pass < 3; pass++) {
2220 		int ret;
2221 		u64 next_bytenr;
2222 		struct btrfsic_block *next_block;
2223 		struct btrfsic_block_data_ctx tmp_next_block_ctx;
2224 		struct btrfsic_block_link *l;
2225 		int num_copies;
2226 		int mirror_num;
2227 		const char *additional_string = NULL;
2228 		struct btrfs_disk_key tmp_disk_key = {0};
2229 
2230 		btrfs_set_disk_key_objectid(&tmp_disk_key,
2231 					    BTRFS_ROOT_ITEM_KEY);
2232 		btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2233 
2234 		switch (pass) {
2235 		case 0:
2236 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2237 						    BTRFS_ROOT_TREE_OBJECTID);
2238 			additional_string = "root ";
2239 			next_bytenr = btrfs_super_root(super_hdr);
2240 			if (state->print_mask &
2241 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2242 				pr_info("root@%llu\n", next_bytenr);
2243 			break;
2244 		case 1:
2245 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2246 						    BTRFS_CHUNK_TREE_OBJECTID);
2247 			additional_string = "chunk ";
2248 			next_bytenr = btrfs_super_chunk_root(super_hdr);
2249 			if (state->print_mask &
2250 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2251 				pr_info("chunk@%llu\n", next_bytenr);
2252 			break;
2253 		case 2:
2254 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2255 						    BTRFS_TREE_LOG_OBJECTID);
2256 			additional_string = "log ";
2257 			next_bytenr = btrfs_super_log_root(super_hdr);
2258 			if (0 == next_bytenr)
2259 				continue;
2260 			if (state->print_mask &
2261 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2262 				pr_info("log@%llu\n", next_bytenr);
2263 			break;
2264 		}
2265 
2266 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
2267 					      BTRFS_SUPER_INFO_SIZE);
2268 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2269 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
2270 			       next_bytenr, num_copies);
2271 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2272 			int was_created;
2273 
2274 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2275 				pr_info("btrfsic_process_written_superblock(mirror_num=%d)\n", mirror_num);
2276 			ret = btrfsic_map_block(state, next_bytenr,
2277 						BTRFS_SUPER_INFO_SIZE,
2278 						&tmp_next_block_ctx,
2279 						mirror_num);
2280 			if (ret) {
2281 				pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
2282 				       next_bytenr, mirror_num);
2283 				return -1;
2284 			}
2285 
2286 			next_block = btrfsic_block_lookup_or_add(
2287 					state,
2288 					&tmp_next_block_ctx,
2289 					additional_string,
2290 					1, 0, 1,
2291 					mirror_num,
2292 					&was_created);
2293 			if (NULL == next_block) {
2294 				pr_info("btrfsic: error, kmalloc failed!\n");
2295 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
2296 				return -1;
2297 			}
2298 
2299 			next_block->disk_key = tmp_disk_key;
2300 			if (was_created)
2301 				next_block->generation =
2302 				    BTRFSIC_GENERATION_UNKNOWN;
2303 			l = btrfsic_block_link_lookup_or_add(
2304 					state,
2305 					&tmp_next_block_ctx,
2306 					next_block,
2307 					superblock,
2308 					BTRFSIC_GENERATION_UNKNOWN);
2309 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
2310 			if (NULL == l)
2311 				return -1;
2312 		}
2313 	}
2314 
2315 	if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2316 		btrfsic_dump_tree(state);
2317 
2318 	return 0;
2319 }
2320 
2321 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2322 					struct btrfsic_block *const block,
2323 					int recursion_level)
2324 {
2325 	const struct btrfsic_block_link *l;
2326 	int ret = 0;
2327 
2328 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2329 		/*
2330 		 * Note that this situation can happen and does not
2331 		 * indicate an error in regular cases. It happens
2332 		 * when disk blocks are freed and later reused.
2333 		 * The check-integrity module is not aware of any
2334 		 * block free operations, it just recognizes block
2335 		 * write operations. Therefore it keeps the linkage
2336 		 * information for a block until a block is
2337 		 * rewritten. This can temporarily cause incorrect
2338 		 * and even circular linkage informations. This
2339 		 * causes no harm unless such blocks are referenced
2340 		 * by the most recent super block.
2341 		 */
2342 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2343 			pr_info("btrfsic: abort cyclic linkage (case 1).\n");
2344 
2345 		return ret;
2346 	}
2347 
2348 	/*
2349 	 * This algorithm is recursive because the amount of used stack
2350 	 * space is very small and the max recursion depth is limited.
2351 	 */
2352 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2353 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2354 			pr_info("rl=%d, %c @%llu (%s/%llu/%d) %u* refers to %c @%llu (%s/%llu/%d)\n",
2355 			       recursion_level,
2356 			       btrfsic_get_block_type(state, block),
2357 			       block->logical_bytenr, block->dev_state->name,
2358 			       block->dev_bytenr, block->mirror_num,
2359 			       l->ref_cnt,
2360 			       btrfsic_get_block_type(state, l->block_ref_to),
2361 			       l->block_ref_to->logical_bytenr,
2362 			       l->block_ref_to->dev_state->name,
2363 			       l->block_ref_to->dev_bytenr,
2364 			       l->block_ref_to->mirror_num);
2365 		if (l->block_ref_to->never_written) {
2366 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is never written!\n",
2367 			       btrfsic_get_block_type(state, l->block_ref_to),
2368 			       l->block_ref_to->logical_bytenr,
2369 			       l->block_ref_to->dev_state->name,
2370 			       l->block_ref_to->dev_bytenr,
2371 			       l->block_ref_to->mirror_num);
2372 			ret = -1;
2373 		} else if (!l->block_ref_to->is_iodone) {
2374 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is not yet iodone!\n",
2375 			       btrfsic_get_block_type(state, l->block_ref_to),
2376 			       l->block_ref_to->logical_bytenr,
2377 			       l->block_ref_to->dev_state->name,
2378 			       l->block_ref_to->dev_bytenr,
2379 			       l->block_ref_to->mirror_num);
2380 			ret = -1;
2381 		} else if (l->block_ref_to->iodone_w_error) {
2382 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which has write error!\n",
2383 			       btrfsic_get_block_type(state, l->block_ref_to),
2384 			       l->block_ref_to->logical_bytenr,
2385 			       l->block_ref_to->dev_state->name,
2386 			       l->block_ref_to->dev_bytenr,
2387 			       l->block_ref_to->mirror_num);
2388 			ret = -1;
2389 		} else if (l->parent_generation !=
2390 			   l->block_ref_to->generation &&
2391 			   BTRFSIC_GENERATION_UNKNOWN !=
2392 			   l->parent_generation &&
2393 			   BTRFSIC_GENERATION_UNKNOWN !=
2394 			   l->block_ref_to->generation) {
2395 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) with generation %llu != parent generation %llu!\n",
2396 			       btrfsic_get_block_type(state, l->block_ref_to),
2397 			       l->block_ref_to->logical_bytenr,
2398 			       l->block_ref_to->dev_state->name,
2399 			       l->block_ref_to->dev_bytenr,
2400 			       l->block_ref_to->mirror_num,
2401 			       l->block_ref_to->generation,
2402 			       l->parent_generation);
2403 			ret = -1;
2404 		} else if (l->block_ref_to->flush_gen >
2405 			   l->block_ref_to->dev_state->last_flush_gen) {
2406 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is not flushed out of disk's write cache (block flush_gen=%llu, dev->flush_gen=%llu)!\n",
2407 			       btrfsic_get_block_type(state, l->block_ref_to),
2408 			       l->block_ref_to->logical_bytenr,
2409 			       l->block_ref_to->dev_state->name,
2410 			       l->block_ref_to->dev_bytenr,
2411 			       l->block_ref_to->mirror_num, block->flush_gen,
2412 			       l->block_ref_to->dev_state->last_flush_gen);
2413 			ret = -1;
2414 		} else if (-1 == btrfsic_check_all_ref_blocks(state,
2415 							      l->block_ref_to,
2416 							      recursion_level +
2417 							      1)) {
2418 			ret = -1;
2419 		}
2420 	}
2421 
2422 	return ret;
2423 }
2424 
2425 static int btrfsic_is_block_ref_by_superblock(
2426 		const struct btrfsic_state *state,
2427 		const struct btrfsic_block *block,
2428 		int recursion_level)
2429 {
2430 	const struct btrfsic_block_link *l;
2431 
2432 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2433 		/* refer to comment at "abort cyclic linkage (case 1)" */
2434 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2435 			pr_info("btrfsic: abort cyclic linkage (case 2).\n");
2436 
2437 		return 0;
2438 	}
2439 
2440 	/*
2441 	 * This algorithm is recursive because the amount of used stack space
2442 	 * is very small and the max recursion depth is limited.
2443 	 */
2444 	list_for_each_entry(l, &block->ref_from_list, node_ref_from) {
2445 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2446 			pr_info("rl=%d, %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
2447 			       recursion_level,
2448 			       btrfsic_get_block_type(state, block),
2449 			       block->logical_bytenr, block->dev_state->name,
2450 			       block->dev_bytenr, block->mirror_num,
2451 			       l->ref_cnt,
2452 			       btrfsic_get_block_type(state, l->block_ref_from),
2453 			       l->block_ref_from->logical_bytenr,
2454 			       l->block_ref_from->dev_state->name,
2455 			       l->block_ref_from->dev_bytenr,
2456 			       l->block_ref_from->mirror_num);
2457 		if (l->block_ref_from->is_superblock &&
2458 		    state->latest_superblock->dev_bytenr ==
2459 		    l->block_ref_from->dev_bytenr &&
2460 		    state->latest_superblock->dev_state->bdev ==
2461 		    l->block_ref_from->dev_state->bdev)
2462 			return 1;
2463 		else if (btrfsic_is_block_ref_by_superblock(state,
2464 							    l->block_ref_from,
2465 							    recursion_level +
2466 							    1))
2467 			return 1;
2468 	}
2469 
2470 	return 0;
2471 }
2472 
2473 static void btrfsic_print_add_link(const struct btrfsic_state *state,
2474 				   const struct btrfsic_block_link *l)
2475 {
2476 	pr_info("Add %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2477 	       l->ref_cnt,
2478 	       btrfsic_get_block_type(state, l->block_ref_from),
2479 	       l->block_ref_from->logical_bytenr,
2480 	       l->block_ref_from->dev_state->name,
2481 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2482 	       btrfsic_get_block_type(state, l->block_ref_to),
2483 	       l->block_ref_to->logical_bytenr,
2484 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2485 	       l->block_ref_to->mirror_num);
2486 }
2487 
2488 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2489 				   const struct btrfsic_block_link *l)
2490 {
2491 	pr_info("Rem %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2492 	       l->ref_cnt,
2493 	       btrfsic_get_block_type(state, l->block_ref_from),
2494 	       l->block_ref_from->logical_bytenr,
2495 	       l->block_ref_from->dev_state->name,
2496 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2497 	       btrfsic_get_block_type(state, l->block_ref_to),
2498 	       l->block_ref_to->logical_bytenr,
2499 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2500 	       l->block_ref_to->mirror_num);
2501 }
2502 
2503 static char btrfsic_get_block_type(const struct btrfsic_state *state,
2504 				   const struct btrfsic_block *block)
2505 {
2506 	if (block->is_superblock &&
2507 	    state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2508 	    state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2509 		return 'S';
2510 	else if (block->is_superblock)
2511 		return 's';
2512 	else if (block->is_metadata)
2513 		return 'M';
2514 	else
2515 		return 'D';
2516 }
2517 
2518 static void btrfsic_dump_tree(const struct btrfsic_state *state)
2519 {
2520 	btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2521 }
2522 
2523 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2524 				  const struct btrfsic_block *block,
2525 				  int indent_level)
2526 {
2527 	const struct btrfsic_block_link *l;
2528 	int indent_add;
2529 	static char buf[80];
2530 	int cursor_position;
2531 
2532 	/*
2533 	 * Should better fill an on-stack buffer with a complete line and
2534 	 * dump it at once when it is time to print a newline character.
2535 	 */
2536 
2537 	/*
2538 	 * This algorithm is recursive because the amount of used stack space
2539 	 * is very small and the max recursion depth is limited.
2540 	 */
2541 	indent_add = sprintf(buf, "%c-%llu(%s/%llu/%u)",
2542 			     btrfsic_get_block_type(state, block),
2543 			     block->logical_bytenr, block->dev_state->name,
2544 			     block->dev_bytenr, block->mirror_num);
2545 	if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2546 		printk("[...]\n");
2547 		return;
2548 	}
2549 	printk(buf);
2550 	indent_level += indent_add;
2551 	if (list_empty(&block->ref_to_list)) {
2552 		printk("\n");
2553 		return;
2554 	}
2555 	if (block->mirror_num > 1 &&
2556 	    !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2557 		printk(" [...]\n");
2558 		return;
2559 	}
2560 
2561 	cursor_position = indent_level;
2562 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2563 		while (cursor_position < indent_level) {
2564 			printk(" ");
2565 			cursor_position++;
2566 		}
2567 		if (l->ref_cnt > 1)
2568 			indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2569 		else
2570 			indent_add = sprintf(buf, " --> ");
2571 		if (indent_level + indent_add >
2572 		    BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2573 			printk("[...]\n");
2574 			cursor_position = 0;
2575 			continue;
2576 		}
2577 
2578 		printk(buf);
2579 
2580 		btrfsic_dump_tree_sub(state, l->block_ref_to,
2581 				      indent_level + indent_add);
2582 		cursor_position = 0;
2583 	}
2584 }
2585 
2586 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2587 		struct btrfsic_state *state,
2588 		struct btrfsic_block_data_ctx *next_block_ctx,
2589 		struct btrfsic_block *next_block,
2590 		struct btrfsic_block *from_block,
2591 		u64 parent_generation)
2592 {
2593 	struct btrfsic_block_link *l;
2594 
2595 	l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2596 						next_block_ctx->dev_bytenr,
2597 						from_block->dev_state->bdev,
2598 						from_block->dev_bytenr,
2599 						&state->block_link_hashtable);
2600 	if (NULL == l) {
2601 		l = btrfsic_block_link_alloc();
2602 		if (NULL == l) {
2603 			pr_info("btrfsic: error, kmalloc failed!\n");
2604 			return NULL;
2605 		}
2606 
2607 		l->block_ref_to = next_block;
2608 		l->block_ref_from = from_block;
2609 		l->ref_cnt = 1;
2610 		l->parent_generation = parent_generation;
2611 
2612 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2613 			btrfsic_print_add_link(state, l);
2614 
2615 		list_add(&l->node_ref_to, &from_block->ref_to_list);
2616 		list_add(&l->node_ref_from, &next_block->ref_from_list);
2617 
2618 		btrfsic_block_link_hashtable_add(l,
2619 						 &state->block_link_hashtable);
2620 	} else {
2621 		l->ref_cnt++;
2622 		l->parent_generation = parent_generation;
2623 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2624 			btrfsic_print_add_link(state, l);
2625 	}
2626 
2627 	return l;
2628 }
2629 
2630 static struct btrfsic_block *btrfsic_block_lookup_or_add(
2631 		struct btrfsic_state *state,
2632 		struct btrfsic_block_data_ctx *block_ctx,
2633 		const char *additional_string,
2634 		int is_metadata,
2635 		int is_iodone,
2636 		int never_written,
2637 		int mirror_num,
2638 		int *was_created)
2639 {
2640 	struct btrfsic_block *block;
2641 
2642 	block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2643 					       block_ctx->dev_bytenr,
2644 					       &state->block_hashtable);
2645 	if (NULL == block) {
2646 		struct btrfsic_dev_state *dev_state;
2647 
2648 		block = btrfsic_block_alloc();
2649 		if (NULL == block) {
2650 			pr_info("btrfsic: error, kmalloc failed!\n");
2651 			return NULL;
2652 		}
2653 		dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev->bd_dev);
2654 		if (NULL == dev_state) {
2655 			pr_info("btrfsic: error, lookup dev_state failed!\n");
2656 			btrfsic_block_free(block);
2657 			return NULL;
2658 		}
2659 		block->dev_state = dev_state;
2660 		block->dev_bytenr = block_ctx->dev_bytenr;
2661 		block->logical_bytenr = block_ctx->start;
2662 		block->is_metadata = is_metadata;
2663 		block->is_iodone = is_iodone;
2664 		block->never_written = never_written;
2665 		block->mirror_num = mirror_num;
2666 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2667 			pr_info("New %s%c-block @%llu (%s/%llu/%d)\n",
2668 			       additional_string,
2669 			       btrfsic_get_block_type(state, block),
2670 			       block->logical_bytenr, dev_state->name,
2671 			       block->dev_bytenr, mirror_num);
2672 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2673 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2674 		if (NULL != was_created)
2675 			*was_created = 1;
2676 	} else {
2677 		if (NULL != was_created)
2678 			*was_created = 0;
2679 	}
2680 
2681 	return block;
2682 }
2683 
2684 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2685 					   u64 bytenr,
2686 					   struct btrfsic_dev_state *dev_state,
2687 					   u64 dev_bytenr)
2688 {
2689 	struct btrfs_fs_info *fs_info = state->fs_info;
2690 	struct btrfsic_block_data_ctx block_ctx;
2691 	int num_copies;
2692 	int mirror_num;
2693 	int match = 0;
2694 	int ret;
2695 
2696 	num_copies = btrfs_num_copies(fs_info, bytenr, state->metablock_size);
2697 
2698 	for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2699 		ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2700 					&block_ctx, mirror_num);
2701 		if (ret) {
2702 			pr_info("btrfsic: btrfsic_map_block(logical @%llu, mirror %d) failed!\n",
2703 			       bytenr, mirror_num);
2704 			continue;
2705 		}
2706 
2707 		if (dev_state->bdev == block_ctx.dev->bdev &&
2708 		    dev_bytenr == block_ctx.dev_bytenr) {
2709 			match++;
2710 			btrfsic_release_block_ctx(&block_ctx);
2711 			break;
2712 		}
2713 		btrfsic_release_block_ctx(&block_ctx);
2714 	}
2715 
2716 	if (WARN_ON(!match)) {
2717 		pr_info("btrfs: attempt to write M-block which contains logical bytenr that doesn't map to dev+physical bytenr of submit_bio, buffer->log_bytenr=%llu, submit_bio(bdev=%s, phys_bytenr=%llu)!\n",
2718 		       bytenr, dev_state->name, dev_bytenr);
2719 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2720 			ret = btrfsic_map_block(state, bytenr,
2721 						state->metablock_size,
2722 						&block_ctx, mirror_num);
2723 			if (ret)
2724 				continue;
2725 
2726 			pr_info("Read logical bytenr @%llu maps to (%s/%llu/%d)\n",
2727 			       bytenr, block_ctx.dev->name,
2728 			       block_ctx.dev_bytenr, mirror_num);
2729 		}
2730 	}
2731 }
2732 
2733 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev)
2734 {
2735 	return btrfsic_dev_state_hashtable_lookup(dev,
2736 						  &btrfsic_dev_state_hashtable);
2737 }
2738 
2739 int btrfsic_submit_bh(int op, int op_flags, struct buffer_head *bh)
2740 {
2741 	struct btrfsic_dev_state *dev_state;
2742 
2743 	if (!btrfsic_is_initialized)
2744 		return submit_bh(op, op_flags, bh);
2745 
2746 	mutex_lock(&btrfsic_mutex);
2747 	/* since btrfsic_submit_bh() might also be called before
2748 	 * btrfsic_mount(), this might return NULL */
2749 	dev_state = btrfsic_dev_state_lookup(bh->b_bdev->bd_dev);
2750 
2751 	/* Only called to write the superblock (incl. FLUSH/FUA) */
2752 	if (NULL != dev_state &&
2753 	    (op == REQ_OP_WRITE) && bh->b_size > 0) {
2754 		u64 dev_bytenr;
2755 
2756 		dev_bytenr = BTRFS_BDEV_BLOCKSIZE * bh->b_blocknr;
2757 		if (dev_state->state->print_mask &
2758 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2759 			pr_info("submit_bh(op=0x%x,0x%x, blocknr=%llu (bytenr %llu), size=%zu, data=%p, bdev=%p)\n",
2760 			       op, op_flags, (unsigned long long)bh->b_blocknr,
2761 			       dev_bytenr, bh->b_size, bh->b_data, bh->b_bdev);
2762 		btrfsic_process_written_block(dev_state, dev_bytenr,
2763 					      &bh->b_data, 1, NULL,
2764 					      NULL, bh, op_flags);
2765 	} else if (NULL != dev_state && (op_flags & REQ_PREFLUSH)) {
2766 		if (dev_state->state->print_mask &
2767 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2768 			pr_info("submit_bh(op=0x%x,0x%x FLUSH, bdev=%p)\n",
2769 			       op, op_flags, bh->b_bdev);
2770 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2771 			if ((dev_state->state->print_mask &
2772 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2773 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2774 				pr_info("btrfsic_submit_bh(%s) with FLUSH but dummy block already in use (ignored)!\n",
2775 				       dev_state->name);
2776 		} else {
2777 			struct btrfsic_block *const block =
2778 				&dev_state->dummy_block_for_bio_bh_flush;
2779 
2780 			block->is_iodone = 0;
2781 			block->never_written = 0;
2782 			block->iodone_w_error = 0;
2783 			block->flush_gen = dev_state->last_flush_gen + 1;
2784 			block->submit_bio_bh_rw = op_flags;
2785 			block->orig_bio_bh_private = bh->b_private;
2786 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2787 			block->next_in_same_bio = NULL;
2788 			bh->b_private = block;
2789 			bh->b_end_io = btrfsic_bh_end_io;
2790 		}
2791 	}
2792 	mutex_unlock(&btrfsic_mutex);
2793 	return submit_bh(op, op_flags, bh);
2794 }
2795 
2796 static void __btrfsic_submit_bio(struct bio *bio)
2797 {
2798 	struct btrfsic_dev_state *dev_state;
2799 
2800 	if (!btrfsic_is_initialized)
2801 		return;
2802 
2803 	mutex_lock(&btrfsic_mutex);
2804 	/* since btrfsic_submit_bio() is also called before
2805 	 * btrfsic_mount(), this might return NULL */
2806 	dev_state = btrfsic_dev_state_lookup(bio_dev(bio) + bio->bi_partno);
2807 	if (NULL != dev_state &&
2808 	    (bio_op(bio) == REQ_OP_WRITE) && bio_has_data(bio)) {
2809 		unsigned int i = 0;
2810 		u64 dev_bytenr;
2811 		u64 cur_bytenr;
2812 		struct bio_vec bvec;
2813 		struct bvec_iter iter;
2814 		int bio_is_patched;
2815 		char **mapped_datav;
2816 		unsigned int segs = bio_segments(bio);
2817 
2818 		dev_bytenr = 512 * bio->bi_iter.bi_sector;
2819 		bio_is_patched = 0;
2820 		if (dev_state->state->print_mask &
2821 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2822 			pr_info("submit_bio(rw=%d,0x%x, bi_vcnt=%u, bi_sector=%llu (bytenr %llu), bi_disk=%p)\n",
2823 			       bio_op(bio), bio->bi_opf, segs,
2824 			       (unsigned long long)bio->bi_iter.bi_sector,
2825 			       dev_bytenr, bio->bi_disk);
2826 
2827 		mapped_datav = kmalloc_array(segs,
2828 					     sizeof(*mapped_datav), GFP_NOFS);
2829 		if (!mapped_datav)
2830 			goto leave;
2831 		cur_bytenr = dev_bytenr;
2832 
2833 		bio_for_each_segment(bvec, bio, iter) {
2834 			BUG_ON(bvec.bv_len != PAGE_SIZE);
2835 			mapped_datav[i] = kmap(bvec.bv_page);
2836 			i++;
2837 
2838 			if (dev_state->state->print_mask &
2839 			    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
2840 				pr_info("#%u: bytenr=%llu, len=%u, offset=%u\n",
2841 				       i, cur_bytenr, bvec.bv_len, bvec.bv_offset);
2842 			cur_bytenr += bvec.bv_len;
2843 		}
2844 		btrfsic_process_written_block(dev_state, dev_bytenr,
2845 					      mapped_datav, segs,
2846 					      bio, &bio_is_patched,
2847 					      NULL, bio->bi_opf);
2848 		bio_for_each_segment(bvec, bio, iter)
2849 			kunmap(bvec.bv_page);
2850 		kfree(mapped_datav);
2851 	} else if (NULL != dev_state && (bio->bi_opf & REQ_PREFLUSH)) {
2852 		if (dev_state->state->print_mask &
2853 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2854 			pr_info("submit_bio(rw=%d,0x%x FLUSH, disk=%p)\n",
2855 			       bio_op(bio), bio->bi_opf, bio->bi_disk);
2856 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2857 			if ((dev_state->state->print_mask &
2858 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2859 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2860 				pr_info("btrfsic_submit_bio(%s) with FLUSH but dummy block already in use (ignored)!\n",
2861 				       dev_state->name);
2862 		} else {
2863 			struct btrfsic_block *const block =
2864 				&dev_state->dummy_block_for_bio_bh_flush;
2865 
2866 			block->is_iodone = 0;
2867 			block->never_written = 0;
2868 			block->iodone_w_error = 0;
2869 			block->flush_gen = dev_state->last_flush_gen + 1;
2870 			block->submit_bio_bh_rw = bio->bi_opf;
2871 			block->orig_bio_bh_private = bio->bi_private;
2872 			block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2873 			block->next_in_same_bio = NULL;
2874 			bio->bi_private = block;
2875 			bio->bi_end_io = btrfsic_bio_end_io;
2876 		}
2877 	}
2878 leave:
2879 	mutex_unlock(&btrfsic_mutex);
2880 }
2881 
2882 void btrfsic_submit_bio(struct bio *bio)
2883 {
2884 	__btrfsic_submit_bio(bio);
2885 	submit_bio(bio);
2886 }
2887 
2888 int btrfsic_submit_bio_wait(struct bio *bio)
2889 {
2890 	__btrfsic_submit_bio(bio);
2891 	return submit_bio_wait(bio);
2892 }
2893 
2894 int btrfsic_mount(struct btrfs_fs_info *fs_info,
2895 		  struct btrfs_fs_devices *fs_devices,
2896 		  int including_extent_data, u32 print_mask)
2897 {
2898 	int ret;
2899 	struct btrfsic_state *state;
2900 	struct list_head *dev_head = &fs_devices->devices;
2901 	struct btrfs_device *device;
2902 
2903 	if (fs_info->nodesize & ((u64)PAGE_SIZE - 1)) {
2904 		pr_info("btrfsic: cannot handle nodesize %d not being a multiple of PAGE_SIZE %ld!\n",
2905 		       fs_info->nodesize, PAGE_SIZE);
2906 		return -1;
2907 	}
2908 	if (fs_info->sectorsize & ((u64)PAGE_SIZE - 1)) {
2909 		pr_info("btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_SIZE %ld!\n",
2910 		       fs_info->sectorsize, PAGE_SIZE);
2911 		return -1;
2912 	}
2913 	state = kvzalloc(sizeof(*state), GFP_KERNEL);
2914 	if (!state) {
2915 		pr_info("btrfs check-integrity: allocation failed!\n");
2916 		return -ENOMEM;
2917 	}
2918 
2919 	if (!btrfsic_is_initialized) {
2920 		mutex_init(&btrfsic_mutex);
2921 		btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
2922 		btrfsic_is_initialized = 1;
2923 	}
2924 	mutex_lock(&btrfsic_mutex);
2925 	state->fs_info = fs_info;
2926 	state->print_mask = print_mask;
2927 	state->include_extent_data = including_extent_data;
2928 	state->csum_size = 0;
2929 	state->metablock_size = fs_info->nodesize;
2930 	state->datablock_size = fs_info->sectorsize;
2931 	INIT_LIST_HEAD(&state->all_blocks_list);
2932 	btrfsic_block_hashtable_init(&state->block_hashtable);
2933 	btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
2934 	state->max_superblock_generation = 0;
2935 	state->latest_superblock = NULL;
2936 
2937 	list_for_each_entry(device, dev_head, dev_list) {
2938 		struct btrfsic_dev_state *ds;
2939 		const char *p;
2940 
2941 		if (!device->bdev || !device->name)
2942 			continue;
2943 
2944 		ds = btrfsic_dev_state_alloc();
2945 		if (NULL == ds) {
2946 			pr_info("btrfs check-integrity: kmalloc() failed!\n");
2947 			mutex_unlock(&btrfsic_mutex);
2948 			return -ENOMEM;
2949 		}
2950 		ds->bdev = device->bdev;
2951 		ds->state = state;
2952 		bdevname(ds->bdev, ds->name);
2953 		ds->name[BDEVNAME_SIZE - 1] = '\0';
2954 		p = kbasename(ds->name);
2955 		strlcpy(ds->name, p, sizeof(ds->name));
2956 		btrfsic_dev_state_hashtable_add(ds,
2957 						&btrfsic_dev_state_hashtable);
2958 	}
2959 
2960 	ret = btrfsic_process_superblock(state, fs_devices);
2961 	if (0 != ret) {
2962 		mutex_unlock(&btrfsic_mutex);
2963 		btrfsic_unmount(fs_devices);
2964 		return ret;
2965 	}
2966 
2967 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
2968 		btrfsic_dump_database(state);
2969 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
2970 		btrfsic_dump_tree(state);
2971 
2972 	mutex_unlock(&btrfsic_mutex);
2973 	return 0;
2974 }
2975 
2976 void btrfsic_unmount(struct btrfs_fs_devices *fs_devices)
2977 {
2978 	struct btrfsic_block *b_all, *tmp_all;
2979 	struct btrfsic_state *state;
2980 	struct list_head *dev_head = &fs_devices->devices;
2981 	struct btrfs_device *device;
2982 
2983 	if (!btrfsic_is_initialized)
2984 		return;
2985 
2986 	mutex_lock(&btrfsic_mutex);
2987 
2988 	state = NULL;
2989 	list_for_each_entry(device, dev_head, dev_list) {
2990 		struct btrfsic_dev_state *ds;
2991 
2992 		if (!device->bdev || !device->name)
2993 			continue;
2994 
2995 		ds = btrfsic_dev_state_hashtable_lookup(
2996 				device->bdev->bd_dev,
2997 				&btrfsic_dev_state_hashtable);
2998 		if (NULL != ds) {
2999 			state = ds->state;
3000 			btrfsic_dev_state_hashtable_remove(ds);
3001 			btrfsic_dev_state_free(ds);
3002 		}
3003 	}
3004 
3005 	if (NULL == state) {
3006 		pr_info("btrfsic: error, cannot find state information on umount!\n");
3007 		mutex_unlock(&btrfsic_mutex);
3008 		return;
3009 	}
3010 
3011 	/*
3012 	 * Don't care about keeping the lists' state up to date,
3013 	 * just free all memory that was allocated dynamically.
3014 	 * Free the blocks and the block_links.
3015 	 */
3016 	list_for_each_entry_safe(b_all, tmp_all, &state->all_blocks_list,
3017 				 all_blocks_node) {
3018 		struct btrfsic_block_link *l, *tmp;
3019 
3020 		list_for_each_entry_safe(l, tmp, &b_all->ref_to_list,
3021 					 node_ref_to) {
3022 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
3023 				btrfsic_print_rem_link(state, l);
3024 
3025 			l->ref_cnt--;
3026 			if (0 == l->ref_cnt)
3027 				btrfsic_block_link_free(l);
3028 		}
3029 
3030 		if (b_all->is_iodone || b_all->never_written)
3031 			btrfsic_block_free(b_all);
3032 		else
3033 			pr_info("btrfs: attempt to free %c-block @%llu (%s/%llu/%d) on umount which is not yet iodone!\n",
3034 			       btrfsic_get_block_type(state, b_all),
3035 			       b_all->logical_bytenr, b_all->dev_state->name,
3036 			       b_all->dev_bytenr, b_all->mirror_num);
3037 	}
3038 
3039 	mutex_unlock(&btrfsic_mutex);
3040 
3041 	kvfree(state);
3042 }
3043