xref: /openbmc/linux/fs/btrfs/check-integrity.c (revision 4f6cce39)
1 /*
2  * Copyright (C) STRATO AG 2011.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 /*
20  * This module can be used to catch cases when the btrfs kernel
21  * code executes write requests to the disk that bring the file
22  * system in an inconsistent state. In such a state, a power-loss
23  * or kernel panic event would cause that the data on disk is
24  * lost or at least damaged.
25  *
26  * Code is added that examines all block write requests during
27  * runtime (including writes of the super block). Three rules
28  * are verified and an error is printed on violation of the
29  * rules:
30  * 1. It is not allowed to write a disk block which is
31  *    currently referenced by the super block (either directly
32  *    or indirectly).
33  * 2. When a super block is written, it is verified that all
34  *    referenced (directly or indirectly) blocks fulfill the
35  *    following requirements:
36  *    2a. All referenced blocks have either been present when
37  *        the file system was mounted, (i.e., they have been
38  *        referenced by the super block) or they have been
39  *        written since then and the write completion callback
40  *        was called and no write error was indicated and a
41  *        FLUSH request to the device where these blocks are
42  *        located was received and completed.
43  *    2b. All referenced blocks need to have a generation
44  *        number which is equal to the parent's number.
45  *
46  * One issue that was found using this module was that the log
47  * tree on disk became temporarily corrupted because disk blocks
48  * that had been in use for the log tree had been freed and
49  * reused too early, while being referenced by the written super
50  * block.
51  *
52  * The search term in the kernel log that can be used to filter
53  * on the existence of detected integrity issues is
54  * "btrfs: attempt".
55  *
56  * The integrity check is enabled via mount options. These
57  * mount options are only supported if the integrity check
58  * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY.
59  *
60  * Example #1, apply integrity checks to all metadata:
61  * mount /dev/sdb1 /mnt -o check_int
62  *
63  * Example #2, apply integrity checks to all metadata and
64  * to data extents:
65  * mount /dev/sdb1 /mnt -o check_int_data
66  *
67  * Example #3, apply integrity checks to all metadata and dump
68  * the tree that the super block references to kernel messages
69  * each time after a super block was written:
70  * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263
71  *
72  * If the integrity check tool is included and activated in
73  * the mount options, plenty of kernel memory is used, and
74  * plenty of additional CPU cycles are spent. Enabling this
75  * functionality is not intended for normal use. In most
76  * cases, unless you are a btrfs developer who needs to verify
77  * the integrity of (super)-block write requests, do not
78  * enable the config option BTRFS_FS_CHECK_INTEGRITY to
79  * include and compile the integrity check tool.
80  *
81  * Expect millions of lines of information in the kernel log with an
82  * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the
83  * kernel config to at least 26 (which is 64MB). Usually the value is
84  * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be
85  * changed like this before LOG_BUF_SHIFT can be set to a high value:
86  * config LOG_BUF_SHIFT
87  *       int "Kernel log buffer size (16 => 64KB, 17 => 128KB)"
88  *       range 12 30
89  */
90 
91 #include <linux/sched.h>
92 #include <linux/slab.h>
93 #include <linux/buffer_head.h>
94 #include <linux/mutex.h>
95 #include <linux/genhd.h>
96 #include <linux/blkdev.h>
97 #include <linux/vmalloc.h>
98 #include <linux/string.h>
99 #include "ctree.h"
100 #include "disk-io.h"
101 #include "hash.h"
102 #include "transaction.h"
103 #include "extent_io.h"
104 #include "volumes.h"
105 #include "print-tree.h"
106 #include "locking.h"
107 #include "check-integrity.h"
108 #include "rcu-string.h"
109 #include "compression.h"
110 
111 #define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
112 #define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
113 #define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
114 #define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
115 #define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
116 #define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
117 #define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
118 #define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6)	/* in characters,
119 							 * excluding " [...]" */
120 #define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
121 
122 /*
123  * The definition of the bitmask fields for the print_mask.
124  * They are specified with the mount option check_integrity_print_mask.
125  */
126 #define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE			0x00000001
127 #define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION		0x00000002
128 #define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE			0x00000004
129 #define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE			0x00000008
130 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH			0x00000010
131 #define BTRFSIC_PRINT_MASK_END_IO_BIO_BH			0x00000020
132 #define BTRFSIC_PRINT_MASK_VERBOSE				0x00000040
133 #define BTRFSIC_PRINT_MASK_VERY_VERBOSE				0x00000080
134 #define BTRFSIC_PRINT_MASK_INITIAL_TREE				0x00000100
135 #define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES			0x00000200
136 #define BTRFSIC_PRINT_MASK_INITIAL_DATABASE			0x00000400
137 #define BTRFSIC_PRINT_MASK_NUM_COPIES				0x00000800
138 #define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS		0x00001000
139 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE		0x00002000
140 
141 struct btrfsic_dev_state;
142 struct btrfsic_state;
143 
144 struct btrfsic_block {
145 	u32 magic_num;		/* only used for debug purposes */
146 	unsigned int is_metadata:1;	/* if it is meta-data, not data-data */
147 	unsigned int is_superblock:1;	/* if it is one of the superblocks */
148 	unsigned int is_iodone:1;	/* if is done by lower subsystem */
149 	unsigned int iodone_w_error:1;	/* error was indicated to endio */
150 	unsigned int never_written:1;	/* block was added because it was
151 					 * referenced, not because it was
152 					 * written */
153 	unsigned int mirror_num;	/* large enough to hold
154 					 * BTRFS_SUPER_MIRROR_MAX */
155 	struct btrfsic_dev_state *dev_state;
156 	u64 dev_bytenr;		/* key, physical byte num on disk */
157 	u64 logical_bytenr;	/* logical byte num on disk */
158 	u64 generation;
159 	struct btrfs_disk_key disk_key;	/* extra info to print in case of
160 					 * issues, will not always be correct */
161 	struct list_head collision_resolving_node;	/* list node */
162 	struct list_head all_blocks_node;	/* list node */
163 
164 	/* the following two lists contain block_link items */
165 	struct list_head ref_to_list;	/* list */
166 	struct list_head ref_from_list;	/* list */
167 	struct btrfsic_block *next_in_same_bio;
168 	void *orig_bio_bh_private;
169 	union {
170 		bio_end_io_t *bio;
171 		bh_end_io_t *bh;
172 	} orig_bio_bh_end_io;
173 	int submit_bio_bh_rw;
174 	u64 flush_gen; /* only valid if !never_written */
175 };
176 
177 /*
178  * Elements of this type are allocated dynamically and required because
179  * each block object can refer to and can be ref from multiple blocks.
180  * The key to lookup them in the hashtable is the dev_bytenr of
181  * the block ref to plus the one from the block referred from.
182  * The fact that they are searchable via a hashtable and that a
183  * ref_cnt is maintained is not required for the btrfs integrity
184  * check algorithm itself, it is only used to make the output more
185  * beautiful in case that an error is detected (an error is defined
186  * as a write operation to a block while that block is still referenced).
187  */
188 struct btrfsic_block_link {
189 	u32 magic_num;		/* only used for debug purposes */
190 	u32 ref_cnt;
191 	struct list_head node_ref_to;	/* list node */
192 	struct list_head node_ref_from;	/* list node */
193 	struct list_head collision_resolving_node;	/* list node */
194 	struct btrfsic_block *block_ref_to;
195 	struct btrfsic_block *block_ref_from;
196 	u64 parent_generation;
197 };
198 
199 struct btrfsic_dev_state {
200 	u32 magic_num;		/* only used for debug purposes */
201 	struct block_device *bdev;
202 	struct btrfsic_state *state;
203 	struct list_head collision_resolving_node;	/* list node */
204 	struct btrfsic_block dummy_block_for_bio_bh_flush;
205 	u64 last_flush_gen;
206 	char name[BDEVNAME_SIZE];
207 };
208 
209 struct btrfsic_block_hashtable {
210 	struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
211 };
212 
213 struct btrfsic_block_link_hashtable {
214 	struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
215 };
216 
217 struct btrfsic_dev_state_hashtable {
218 	struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
219 };
220 
221 struct btrfsic_block_data_ctx {
222 	u64 start;		/* virtual bytenr */
223 	u64 dev_bytenr;		/* physical bytenr on device */
224 	u32 len;
225 	struct btrfsic_dev_state *dev;
226 	char **datav;
227 	struct page **pagev;
228 	void *mem_to_free;
229 };
230 
231 /* This structure is used to implement recursion without occupying
232  * any stack space, refer to btrfsic_process_metablock() */
233 struct btrfsic_stack_frame {
234 	u32 magic;
235 	u32 nr;
236 	int error;
237 	int i;
238 	int limit_nesting;
239 	int num_copies;
240 	int mirror_num;
241 	struct btrfsic_block *block;
242 	struct btrfsic_block_data_ctx *block_ctx;
243 	struct btrfsic_block *next_block;
244 	struct btrfsic_block_data_ctx next_block_ctx;
245 	struct btrfs_header *hdr;
246 	struct btrfsic_stack_frame *prev;
247 };
248 
249 /* Some state per mounted filesystem */
250 struct btrfsic_state {
251 	u32 print_mask;
252 	int include_extent_data;
253 	int csum_size;
254 	struct list_head all_blocks_list;
255 	struct btrfsic_block_hashtable block_hashtable;
256 	struct btrfsic_block_link_hashtable block_link_hashtable;
257 	struct btrfs_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(
300 		struct block_device *bdev,
301 		struct btrfsic_dev_state_hashtable *h);
302 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void);
303 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf);
304 static int btrfsic_process_superblock(struct btrfsic_state *state,
305 				      struct btrfs_fs_devices *fs_devices);
306 static int btrfsic_process_metablock(struct btrfsic_state *state,
307 				     struct btrfsic_block *block,
308 				     struct btrfsic_block_data_ctx *block_ctx,
309 				     int limit_nesting, int force_iodone_flag);
310 static void btrfsic_read_from_block_data(
311 	struct btrfsic_block_data_ctx *block_ctx,
312 	void *dst, u32 offset, size_t len);
313 static int btrfsic_create_link_to_next_block(
314 		struct btrfsic_state *state,
315 		struct btrfsic_block *block,
316 		struct btrfsic_block_data_ctx
317 		*block_ctx, u64 next_bytenr,
318 		int limit_nesting,
319 		struct btrfsic_block_data_ctx *next_block_ctx,
320 		struct btrfsic_block **next_blockp,
321 		int force_iodone_flag,
322 		int *num_copiesp, int *mirror_nump,
323 		struct btrfs_disk_key *disk_key,
324 		u64 parent_generation);
325 static int btrfsic_handle_extent_data(struct btrfsic_state *state,
326 				      struct btrfsic_block *block,
327 				      struct btrfsic_block_data_ctx *block_ctx,
328 				      u32 item_offset, int force_iodone_flag);
329 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
330 			     struct btrfsic_block_data_ctx *block_ctx_out,
331 			     int mirror_num);
332 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
333 static int btrfsic_read_block(struct btrfsic_state *state,
334 			      struct btrfsic_block_data_ctx *block_ctx);
335 static void btrfsic_dump_database(struct btrfsic_state *state);
336 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
337 				     char **datav, unsigned int num_pages);
338 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
339 					  u64 dev_bytenr, char **mapped_datav,
340 					  unsigned int num_pages,
341 					  struct bio *bio, int *bio_is_patched,
342 					  struct buffer_head *bh,
343 					  int submit_bio_bh_rw);
344 static int btrfsic_process_written_superblock(
345 		struct btrfsic_state *state,
346 		struct btrfsic_block *const block,
347 		struct btrfs_super_block *const super_hdr);
348 static void btrfsic_bio_end_io(struct bio *bp);
349 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate);
350 static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
351 					      const struct btrfsic_block *block,
352 					      int recursion_level);
353 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
354 					struct btrfsic_block *const block,
355 					int recursion_level);
356 static void btrfsic_print_add_link(const struct btrfsic_state *state,
357 				   const struct btrfsic_block_link *l);
358 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
359 				   const struct btrfsic_block_link *l);
360 static char btrfsic_get_block_type(const struct btrfsic_state *state,
361 				   const struct btrfsic_block *block);
362 static void btrfsic_dump_tree(const struct btrfsic_state *state);
363 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
364 				  const struct btrfsic_block *block,
365 				  int indent_level);
366 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
367 		struct btrfsic_state *state,
368 		struct btrfsic_block_data_ctx *next_block_ctx,
369 		struct btrfsic_block *next_block,
370 		struct btrfsic_block *from_block,
371 		u64 parent_generation);
372 static struct btrfsic_block *btrfsic_block_lookup_or_add(
373 		struct btrfsic_state *state,
374 		struct btrfsic_block_data_ctx *block_ctx,
375 		const char *additional_string,
376 		int is_metadata,
377 		int is_iodone,
378 		int never_written,
379 		int mirror_num,
380 		int *was_created);
381 static int btrfsic_process_superblock_dev_mirror(
382 		struct btrfsic_state *state,
383 		struct btrfsic_dev_state *dev_state,
384 		struct btrfs_device *device,
385 		int superblock_mirror_num,
386 		struct btrfsic_dev_state **selected_dev_state,
387 		struct btrfs_super_block *selected_super);
388 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
389 		struct block_device *bdev);
390 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
391 					   u64 bytenr,
392 					   struct btrfsic_dev_state *dev_state,
393 					   u64 dev_bytenr);
394 
395 static struct mutex btrfsic_mutex;
396 static int btrfsic_is_initialized;
397 static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
398 
399 
400 static void btrfsic_block_init(struct btrfsic_block *b)
401 {
402 	b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
403 	b->dev_state = NULL;
404 	b->dev_bytenr = 0;
405 	b->logical_bytenr = 0;
406 	b->generation = BTRFSIC_GENERATION_UNKNOWN;
407 	b->disk_key.objectid = 0;
408 	b->disk_key.type = 0;
409 	b->disk_key.offset = 0;
410 	b->is_metadata = 0;
411 	b->is_superblock = 0;
412 	b->is_iodone = 0;
413 	b->iodone_w_error = 0;
414 	b->never_written = 0;
415 	b->mirror_num = 0;
416 	b->next_in_same_bio = NULL;
417 	b->orig_bio_bh_private = NULL;
418 	b->orig_bio_bh_end_io.bio = NULL;
419 	INIT_LIST_HEAD(&b->collision_resolving_node);
420 	INIT_LIST_HEAD(&b->all_blocks_node);
421 	INIT_LIST_HEAD(&b->ref_to_list);
422 	INIT_LIST_HEAD(&b->ref_from_list);
423 	b->submit_bio_bh_rw = 0;
424 	b->flush_gen = 0;
425 }
426 
427 static struct btrfsic_block *btrfsic_block_alloc(void)
428 {
429 	struct btrfsic_block *b;
430 
431 	b = kzalloc(sizeof(*b), GFP_NOFS);
432 	if (NULL != b)
433 		btrfsic_block_init(b);
434 
435 	return b;
436 }
437 
438 static void btrfsic_block_free(struct btrfsic_block *b)
439 {
440 	BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
441 	kfree(b);
442 }
443 
444 static void btrfsic_block_link_init(struct btrfsic_block_link *l)
445 {
446 	l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
447 	l->ref_cnt = 1;
448 	INIT_LIST_HEAD(&l->node_ref_to);
449 	INIT_LIST_HEAD(&l->node_ref_from);
450 	INIT_LIST_HEAD(&l->collision_resolving_node);
451 	l->block_ref_to = NULL;
452 	l->block_ref_from = NULL;
453 }
454 
455 static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
456 {
457 	struct btrfsic_block_link *l;
458 
459 	l = kzalloc(sizeof(*l), GFP_NOFS);
460 	if (NULL != l)
461 		btrfsic_block_link_init(l);
462 
463 	return l;
464 }
465 
466 static void btrfsic_block_link_free(struct btrfsic_block_link *l)
467 {
468 	BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
469 	kfree(l);
470 }
471 
472 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
473 {
474 	ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
475 	ds->bdev = NULL;
476 	ds->state = NULL;
477 	ds->name[0] = '\0';
478 	INIT_LIST_HEAD(&ds->collision_resolving_node);
479 	ds->last_flush_gen = 0;
480 	btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
481 	ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
482 	ds->dummy_block_for_bio_bh_flush.dev_state = ds;
483 }
484 
485 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
486 {
487 	struct btrfsic_dev_state *ds;
488 
489 	ds = kzalloc(sizeof(*ds), GFP_NOFS);
490 	if (NULL != ds)
491 		btrfsic_dev_state_init(ds);
492 
493 	return ds;
494 }
495 
496 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
497 {
498 	BUG_ON(!(NULL == ds ||
499 		 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
500 	kfree(ds);
501 }
502 
503 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
504 {
505 	int i;
506 
507 	for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
508 		INIT_LIST_HEAD(h->table + i);
509 }
510 
511 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
512 					struct btrfsic_block_hashtable *h)
513 {
514 	const unsigned int hashval =
515 	    (((unsigned int)(b->dev_bytenr >> 16)) ^
516 	     ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
517 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
518 
519 	list_add(&b->collision_resolving_node, h->table + hashval);
520 }
521 
522 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
523 {
524 	list_del(&b->collision_resolving_node);
525 }
526 
527 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
528 		struct block_device *bdev,
529 		u64 dev_bytenr,
530 		struct btrfsic_block_hashtable *h)
531 {
532 	const unsigned int hashval =
533 	    (((unsigned int)(dev_bytenr >> 16)) ^
534 	     ((unsigned int)((uintptr_t)bdev))) &
535 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
536 	struct btrfsic_block *b;
537 
538 	list_for_each_entry(b, h->table + hashval, collision_resolving_node) {
539 		if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
540 			return b;
541 	}
542 
543 	return NULL;
544 }
545 
546 static void btrfsic_block_link_hashtable_init(
547 		struct btrfsic_block_link_hashtable *h)
548 {
549 	int i;
550 
551 	for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
552 		INIT_LIST_HEAD(h->table + i);
553 }
554 
555 static void btrfsic_block_link_hashtable_add(
556 		struct btrfsic_block_link *l,
557 		struct btrfsic_block_link_hashtable *h)
558 {
559 	const unsigned int hashval =
560 	    (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
561 	     ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
562 	     ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
563 	     ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
564 	     & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
565 
566 	BUG_ON(NULL == l->block_ref_to);
567 	BUG_ON(NULL == l->block_ref_from);
568 	list_add(&l->collision_resolving_node, h->table + hashval);
569 }
570 
571 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
572 {
573 	list_del(&l->collision_resolving_node);
574 }
575 
576 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
577 		struct block_device *bdev_ref_to,
578 		u64 dev_bytenr_ref_to,
579 		struct block_device *bdev_ref_from,
580 		u64 dev_bytenr_ref_from,
581 		struct btrfsic_block_link_hashtable *h)
582 {
583 	const unsigned int hashval =
584 	    (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
585 	     ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
586 	     ((unsigned int)((uintptr_t)bdev_ref_to)) ^
587 	     ((unsigned int)((uintptr_t)bdev_ref_from))) &
588 	     (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
589 	struct btrfsic_block_link *l;
590 
591 	list_for_each_entry(l, h->table + hashval, collision_resolving_node) {
592 		BUG_ON(NULL == l->block_ref_to);
593 		BUG_ON(NULL == l->block_ref_from);
594 		if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
595 		    l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
596 		    l->block_ref_from->dev_state->bdev == bdev_ref_from &&
597 		    l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
598 			return l;
599 	}
600 
601 	return NULL;
602 }
603 
604 static void btrfsic_dev_state_hashtable_init(
605 		struct btrfsic_dev_state_hashtable *h)
606 {
607 	int i;
608 
609 	for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
610 		INIT_LIST_HEAD(h->table + i);
611 }
612 
613 static void btrfsic_dev_state_hashtable_add(
614 		struct btrfsic_dev_state *ds,
615 		struct btrfsic_dev_state_hashtable *h)
616 {
617 	const unsigned int hashval =
618 	    (((unsigned int)((uintptr_t)ds->bdev)) &
619 	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
620 
621 	list_add(&ds->collision_resolving_node, h->table + hashval);
622 }
623 
624 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
625 {
626 	list_del(&ds->collision_resolving_node);
627 }
628 
629 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(
630 		struct block_device *bdev,
631 		struct btrfsic_dev_state_hashtable *h)
632 {
633 	const unsigned int hashval =
634 	    (((unsigned int)((uintptr_t)bdev)) &
635 	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
636 	struct btrfsic_dev_state *ds;
637 
638 	list_for_each_entry(ds, h->table + hashval, collision_resolving_node) {
639 		if (ds->bdev == bdev)
640 			return ds;
641 	}
642 
643 	return NULL;
644 }
645 
646 static int btrfsic_process_superblock(struct btrfsic_state *state,
647 				      struct btrfs_fs_devices *fs_devices)
648 {
649 	struct btrfs_fs_info *fs_info = state->fs_info;
650 	struct btrfs_super_block *selected_super;
651 	struct list_head *dev_head = &fs_devices->devices;
652 	struct btrfs_device *device;
653 	struct btrfsic_dev_state *selected_dev_state = NULL;
654 	int ret = 0;
655 	int pass;
656 
657 	BUG_ON(NULL == state);
658 	selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
659 	if (NULL == selected_super) {
660 		pr_info("btrfsic: error, kmalloc failed!\n");
661 		return -ENOMEM;
662 	}
663 
664 	list_for_each_entry(device, dev_head, dev_list) {
665 		int i;
666 		struct btrfsic_dev_state *dev_state;
667 
668 		if (!device->bdev || !device->name)
669 			continue;
670 
671 		dev_state = btrfsic_dev_state_lookup(device->bdev);
672 		BUG_ON(NULL == dev_state);
673 		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
674 			ret = btrfsic_process_superblock_dev_mirror(
675 					state, dev_state, device, i,
676 					&selected_dev_state, selected_super);
677 			if (0 != ret && 0 == i) {
678 				kfree(selected_super);
679 				return ret;
680 			}
681 		}
682 	}
683 
684 	if (NULL == state->latest_superblock) {
685 		pr_info("btrfsic: no superblock found!\n");
686 		kfree(selected_super);
687 		return -1;
688 	}
689 
690 	state->csum_size = btrfs_super_csum_size(selected_super);
691 
692 	for (pass = 0; pass < 3; pass++) {
693 		int num_copies;
694 		int mirror_num;
695 		u64 next_bytenr;
696 
697 		switch (pass) {
698 		case 0:
699 			next_bytenr = btrfs_super_root(selected_super);
700 			if (state->print_mask &
701 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
702 				pr_info("root@%llu\n", next_bytenr);
703 			break;
704 		case 1:
705 			next_bytenr = btrfs_super_chunk_root(selected_super);
706 			if (state->print_mask &
707 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
708 				pr_info("chunk@%llu\n", next_bytenr);
709 			break;
710 		case 2:
711 			next_bytenr = btrfs_super_log_root(selected_super);
712 			if (0 == next_bytenr)
713 				continue;
714 			if (state->print_mask &
715 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
716 				pr_info("log@%llu\n", next_bytenr);
717 			break;
718 		}
719 
720 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
721 					      state->metablock_size);
722 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
723 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
724 			       next_bytenr, num_copies);
725 
726 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
727 			struct btrfsic_block *next_block;
728 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
729 			struct btrfsic_block_link *l;
730 
731 			ret = btrfsic_map_block(state, next_bytenr,
732 						state->metablock_size,
733 						&tmp_next_block_ctx,
734 						mirror_num);
735 			if (ret) {
736 				pr_info("btrfsic: btrfsic_map_block(root @%llu, mirror %d) failed!\n",
737 				       next_bytenr, mirror_num);
738 				kfree(selected_super);
739 				return -1;
740 			}
741 
742 			next_block = btrfsic_block_hashtable_lookup(
743 					tmp_next_block_ctx.dev->bdev,
744 					tmp_next_block_ctx.dev_bytenr,
745 					&state->block_hashtable);
746 			BUG_ON(NULL == next_block);
747 
748 			l = btrfsic_block_link_hashtable_lookup(
749 					tmp_next_block_ctx.dev->bdev,
750 					tmp_next_block_ctx.dev_bytenr,
751 					state->latest_superblock->dev_state->
752 					bdev,
753 					state->latest_superblock->dev_bytenr,
754 					&state->block_link_hashtable);
755 			BUG_ON(NULL == l);
756 
757 			ret = btrfsic_read_block(state, &tmp_next_block_ctx);
758 			if (ret < (int)PAGE_SIZE) {
759 				pr_info("btrfsic: read @logical %llu failed!\n",
760 				       tmp_next_block_ctx.start);
761 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
762 				kfree(selected_super);
763 				return -1;
764 			}
765 
766 			ret = btrfsic_process_metablock(state,
767 							next_block,
768 							&tmp_next_block_ctx,
769 							BTRFS_MAX_LEVEL + 3, 1);
770 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
771 		}
772 	}
773 
774 	kfree(selected_super);
775 	return ret;
776 }
777 
778 static int btrfsic_process_superblock_dev_mirror(
779 		struct btrfsic_state *state,
780 		struct btrfsic_dev_state *dev_state,
781 		struct btrfs_device *device,
782 		int superblock_mirror_num,
783 		struct btrfsic_dev_state **selected_dev_state,
784 		struct btrfs_super_block *selected_super)
785 {
786 	struct btrfs_fs_info *fs_info = state->fs_info;
787 	struct btrfs_super_block *super_tmp;
788 	u64 dev_bytenr;
789 	struct buffer_head *bh;
790 	struct btrfsic_block *superblock_tmp;
791 	int pass;
792 	struct block_device *const superblock_bdev = device->bdev;
793 
794 	/* super block bytenr is always the unmapped device bytenr */
795 	dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
796 	if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->commit_total_bytes)
797 		return -1;
798 	bh = __bread(superblock_bdev, dev_bytenr / 4096,
799 		     BTRFS_SUPER_INFO_SIZE);
800 	if (NULL == bh)
801 		return -1;
802 	super_tmp = (struct btrfs_super_block *)
803 	    (bh->b_data + (dev_bytenr & 4095));
804 
805 	if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
806 	    btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
807 	    memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
808 	    btrfs_super_nodesize(super_tmp) != state->metablock_size ||
809 	    btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
810 		brelse(bh);
811 		return 0;
812 	}
813 
814 	superblock_tmp =
815 	    btrfsic_block_hashtable_lookup(superblock_bdev,
816 					   dev_bytenr,
817 					   &state->block_hashtable);
818 	if (NULL == superblock_tmp) {
819 		superblock_tmp = btrfsic_block_alloc();
820 		if (NULL == superblock_tmp) {
821 			pr_info("btrfsic: error, kmalloc failed!\n");
822 			brelse(bh);
823 			return -1;
824 		}
825 		/* for superblock, only the dev_bytenr makes sense */
826 		superblock_tmp->dev_bytenr = dev_bytenr;
827 		superblock_tmp->dev_state = dev_state;
828 		superblock_tmp->logical_bytenr = dev_bytenr;
829 		superblock_tmp->generation = btrfs_super_generation(super_tmp);
830 		superblock_tmp->is_metadata = 1;
831 		superblock_tmp->is_superblock = 1;
832 		superblock_tmp->is_iodone = 1;
833 		superblock_tmp->never_written = 0;
834 		superblock_tmp->mirror_num = 1 + superblock_mirror_num;
835 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
836 			btrfs_info_in_rcu(fs_info,
837 				"new initial S-block (bdev %p, %s) @%llu (%s/%llu/%d)",
838 				     superblock_bdev,
839 				     rcu_str_deref(device->name), dev_bytenr,
840 				     dev_state->name, dev_bytenr,
841 				     superblock_mirror_num);
842 		list_add(&superblock_tmp->all_blocks_node,
843 			 &state->all_blocks_list);
844 		btrfsic_block_hashtable_add(superblock_tmp,
845 					    &state->block_hashtable);
846 	}
847 
848 	/* select the one with the highest generation field */
849 	if (btrfs_super_generation(super_tmp) >
850 	    state->max_superblock_generation ||
851 	    0 == state->max_superblock_generation) {
852 		memcpy(selected_super, super_tmp, sizeof(*selected_super));
853 		*selected_dev_state = dev_state;
854 		state->max_superblock_generation =
855 		    btrfs_super_generation(super_tmp);
856 		state->latest_superblock = superblock_tmp;
857 	}
858 
859 	for (pass = 0; pass < 3; pass++) {
860 		u64 next_bytenr;
861 		int num_copies;
862 		int mirror_num;
863 		const char *additional_string = NULL;
864 		struct btrfs_disk_key tmp_disk_key;
865 
866 		tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
867 		tmp_disk_key.offset = 0;
868 		switch (pass) {
869 		case 0:
870 			btrfs_set_disk_key_objectid(&tmp_disk_key,
871 						    BTRFS_ROOT_TREE_OBJECTID);
872 			additional_string = "initial root ";
873 			next_bytenr = btrfs_super_root(super_tmp);
874 			break;
875 		case 1:
876 			btrfs_set_disk_key_objectid(&tmp_disk_key,
877 						    BTRFS_CHUNK_TREE_OBJECTID);
878 			additional_string = "initial chunk ";
879 			next_bytenr = btrfs_super_chunk_root(super_tmp);
880 			break;
881 		case 2:
882 			btrfs_set_disk_key_objectid(&tmp_disk_key,
883 						    BTRFS_TREE_LOG_OBJECTID);
884 			additional_string = "initial log ";
885 			next_bytenr = btrfs_super_log_root(super_tmp);
886 			if (0 == next_bytenr)
887 				continue;
888 			break;
889 		}
890 
891 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
892 					      state->metablock_size);
893 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
894 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
895 			       next_bytenr, num_copies);
896 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
897 			struct btrfsic_block *next_block;
898 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
899 			struct btrfsic_block_link *l;
900 
901 			if (btrfsic_map_block(state, next_bytenr,
902 					      state->metablock_size,
903 					      &tmp_next_block_ctx,
904 					      mirror_num)) {
905 				pr_info("btrfsic: btrfsic_map_block(bytenr @%llu, mirror %d) failed!\n",
906 				       next_bytenr, mirror_num);
907 				brelse(bh);
908 				return -1;
909 			}
910 
911 			next_block = btrfsic_block_lookup_or_add(
912 					state, &tmp_next_block_ctx,
913 					additional_string, 1, 1, 0,
914 					mirror_num, NULL);
915 			if (NULL == next_block) {
916 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
917 				brelse(bh);
918 				return -1;
919 			}
920 
921 			next_block->disk_key = tmp_disk_key;
922 			next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
923 			l = btrfsic_block_link_lookup_or_add(
924 					state, &tmp_next_block_ctx,
925 					next_block, superblock_tmp,
926 					BTRFSIC_GENERATION_UNKNOWN);
927 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
928 			if (NULL == l) {
929 				brelse(bh);
930 				return -1;
931 			}
932 		}
933 	}
934 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
935 		btrfsic_dump_tree_sub(state, superblock_tmp, 0);
936 
937 	brelse(bh);
938 	return 0;
939 }
940 
941 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
942 {
943 	struct btrfsic_stack_frame *sf;
944 
945 	sf = kzalloc(sizeof(*sf), GFP_NOFS);
946 	if (NULL == sf)
947 		pr_info("btrfsic: alloc memory failed!\n");
948 	else
949 		sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
950 	return sf;
951 }
952 
953 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
954 {
955 	BUG_ON(!(NULL == sf ||
956 		 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
957 	kfree(sf);
958 }
959 
960 static int btrfsic_process_metablock(
961 		struct btrfsic_state *state,
962 		struct btrfsic_block *const first_block,
963 		struct btrfsic_block_data_ctx *const first_block_ctx,
964 		int first_limit_nesting, int force_iodone_flag)
965 {
966 	struct btrfsic_stack_frame initial_stack_frame = { 0 };
967 	struct btrfsic_stack_frame *sf;
968 	struct btrfsic_stack_frame *next_stack;
969 	struct btrfs_header *const first_hdr =
970 		(struct btrfs_header *)first_block_ctx->datav[0];
971 
972 	BUG_ON(!first_hdr);
973 	sf = &initial_stack_frame;
974 	sf->error = 0;
975 	sf->i = -1;
976 	sf->limit_nesting = first_limit_nesting;
977 	sf->block = first_block;
978 	sf->block_ctx = first_block_ctx;
979 	sf->next_block = NULL;
980 	sf->hdr = first_hdr;
981 	sf->prev = NULL;
982 
983 continue_with_new_stack_frame:
984 	sf->block->generation = le64_to_cpu(sf->hdr->generation);
985 	if (0 == sf->hdr->level) {
986 		struct btrfs_leaf *const leafhdr =
987 		    (struct btrfs_leaf *)sf->hdr;
988 
989 		if (-1 == sf->i) {
990 			sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
991 
992 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
993 				pr_info("leaf %llu items %d generation %llu owner %llu\n",
994 				       sf->block_ctx->start, sf->nr,
995 				       btrfs_stack_header_generation(
996 					       &leafhdr->header),
997 				       btrfs_stack_header_owner(
998 					       &leafhdr->header));
999 		}
1000 
1001 continue_with_current_leaf_stack_frame:
1002 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1003 			sf->i++;
1004 			sf->num_copies = 0;
1005 		}
1006 
1007 		if (sf->i < sf->nr) {
1008 			struct btrfs_item disk_item;
1009 			u32 disk_item_offset =
1010 				(uintptr_t)(leafhdr->items + sf->i) -
1011 				(uintptr_t)leafhdr;
1012 			struct btrfs_disk_key *disk_key;
1013 			u8 type;
1014 			u32 item_offset;
1015 			u32 item_size;
1016 
1017 			if (disk_item_offset + sizeof(struct btrfs_item) >
1018 			    sf->block_ctx->len) {
1019 leaf_item_out_of_bounce_error:
1020 				pr_info("btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
1021 				       sf->block_ctx->start,
1022 				       sf->block_ctx->dev->name);
1023 				goto one_stack_frame_backwards;
1024 			}
1025 			btrfsic_read_from_block_data(sf->block_ctx,
1026 						     &disk_item,
1027 						     disk_item_offset,
1028 						     sizeof(struct btrfs_item));
1029 			item_offset = btrfs_stack_item_offset(&disk_item);
1030 			item_size = btrfs_stack_item_size(&disk_item);
1031 			disk_key = &disk_item.key;
1032 			type = btrfs_disk_key_type(disk_key);
1033 
1034 			if (BTRFS_ROOT_ITEM_KEY == type) {
1035 				struct btrfs_root_item root_item;
1036 				u32 root_item_offset;
1037 				u64 next_bytenr;
1038 
1039 				root_item_offset = item_offset +
1040 					offsetof(struct btrfs_leaf, items);
1041 				if (root_item_offset + item_size >
1042 				    sf->block_ctx->len)
1043 					goto leaf_item_out_of_bounce_error;
1044 				btrfsic_read_from_block_data(
1045 					sf->block_ctx, &root_item,
1046 					root_item_offset,
1047 					item_size);
1048 				next_bytenr = btrfs_root_bytenr(&root_item);
1049 
1050 				sf->error =
1051 				    btrfsic_create_link_to_next_block(
1052 						state,
1053 						sf->block,
1054 						sf->block_ctx,
1055 						next_bytenr,
1056 						sf->limit_nesting,
1057 						&sf->next_block_ctx,
1058 						&sf->next_block,
1059 						force_iodone_flag,
1060 						&sf->num_copies,
1061 						&sf->mirror_num,
1062 						disk_key,
1063 						btrfs_root_generation(
1064 						&root_item));
1065 				if (sf->error)
1066 					goto one_stack_frame_backwards;
1067 
1068 				if (NULL != sf->next_block) {
1069 					struct btrfs_header *const next_hdr =
1070 					    (struct btrfs_header *)
1071 					    sf->next_block_ctx.datav[0];
1072 
1073 					next_stack =
1074 					    btrfsic_stack_frame_alloc();
1075 					if (NULL == next_stack) {
1076 						sf->error = -1;
1077 						btrfsic_release_block_ctx(
1078 								&sf->
1079 								next_block_ctx);
1080 						goto one_stack_frame_backwards;
1081 					}
1082 
1083 					next_stack->i = -1;
1084 					next_stack->block = sf->next_block;
1085 					next_stack->block_ctx =
1086 					    &sf->next_block_ctx;
1087 					next_stack->next_block = NULL;
1088 					next_stack->hdr = next_hdr;
1089 					next_stack->limit_nesting =
1090 					    sf->limit_nesting - 1;
1091 					next_stack->prev = sf;
1092 					sf = next_stack;
1093 					goto continue_with_new_stack_frame;
1094 				}
1095 			} else if (BTRFS_EXTENT_DATA_KEY == type &&
1096 				   state->include_extent_data) {
1097 				sf->error = btrfsic_handle_extent_data(
1098 						state,
1099 						sf->block,
1100 						sf->block_ctx,
1101 						item_offset,
1102 						force_iodone_flag);
1103 				if (sf->error)
1104 					goto one_stack_frame_backwards;
1105 			}
1106 
1107 			goto continue_with_current_leaf_stack_frame;
1108 		}
1109 	} else {
1110 		struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1111 
1112 		if (-1 == sf->i) {
1113 			sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1114 
1115 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1116 				pr_info("node %llu level %d items %d generation %llu owner %llu\n",
1117 				       sf->block_ctx->start,
1118 				       nodehdr->header.level, sf->nr,
1119 				       btrfs_stack_header_generation(
1120 				       &nodehdr->header),
1121 				       btrfs_stack_header_owner(
1122 				       &nodehdr->header));
1123 		}
1124 
1125 continue_with_current_node_stack_frame:
1126 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1127 			sf->i++;
1128 			sf->num_copies = 0;
1129 		}
1130 
1131 		if (sf->i < sf->nr) {
1132 			struct btrfs_key_ptr key_ptr;
1133 			u32 key_ptr_offset;
1134 			u64 next_bytenr;
1135 
1136 			key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1137 					  (uintptr_t)nodehdr;
1138 			if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1139 			    sf->block_ctx->len) {
1140 				pr_info("btrfsic: node item out of bounce at logical %llu, dev %s\n",
1141 				       sf->block_ctx->start,
1142 				       sf->block_ctx->dev->name);
1143 				goto one_stack_frame_backwards;
1144 			}
1145 			btrfsic_read_from_block_data(
1146 				sf->block_ctx, &key_ptr, key_ptr_offset,
1147 				sizeof(struct btrfs_key_ptr));
1148 			next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1149 
1150 			sf->error = btrfsic_create_link_to_next_block(
1151 					state,
1152 					sf->block,
1153 					sf->block_ctx,
1154 					next_bytenr,
1155 					sf->limit_nesting,
1156 					&sf->next_block_ctx,
1157 					&sf->next_block,
1158 					force_iodone_flag,
1159 					&sf->num_copies,
1160 					&sf->mirror_num,
1161 					&key_ptr.key,
1162 					btrfs_stack_key_generation(&key_ptr));
1163 			if (sf->error)
1164 				goto one_stack_frame_backwards;
1165 
1166 			if (NULL != sf->next_block) {
1167 				struct btrfs_header *const next_hdr =
1168 				    (struct btrfs_header *)
1169 				    sf->next_block_ctx.datav[0];
1170 
1171 				next_stack = btrfsic_stack_frame_alloc();
1172 				if (NULL == next_stack) {
1173 					sf->error = -1;
1174 					goto one_stack_frame_backwards;
1175 				}
1176 
1177 				next_stack->i = -1;
1178 				next_stack->block = sf->next_block;
1179 				next_stack->block_ctx = &sf->next_block_ctx;
1180 				next_stack->next_block = NULL;
1181 				next_stack->hdr = next_hdr;
1182 				next_stack->limit_nesting =
1183 				    sf->limit_nesting - 1;
1184 				next_stack->prev = sf;
1185 				sf = next_stack;
1186 				goto continue_with_new_stack_frame;
1187 			}
1188 
1189 			goto continue_with_current_node_stack_frame;
1190 		}
1191 	}
1192 
1193 one_stack_frame_backwards:
1194 	if (NULL != sf->prev) {
1195 		struct btrfsic_stack_frame *const prev = sf->prev;
1196 
1197 		/* the one for the initial block is freed in the caller */
1198 		btrfsic_release_block_ctx(sf->block_ctx);
1199 
1200 		if (sf->error) {
1201 			prev->error = sf->error;
1202 			btrfsic_stack_frame_free(sf);
1203 			sf = prev;
1204 			goto one_stack_frame_backwards;
1205 		}
1206 
1207 		btrfsic_stack_frame_free(sf);
1208 		sf = prev;
1209 		goto continue_with_new_stack_frame;
1210 	} else {
1211 		BUG_ON(&initial_stack_frame != sf);
1212 	}
1213 
1214 	return sf->error;
1215 }
1216 
1217 static void btrfsic_read_from_block_data(
1218 	struct btrfsic_block_data_ctx *block_ctx,
1219 	void *dstv, u32 offset, size_t len)
1220 {
1221 	size_t cur;
1222 	size_t offset_in_page;
1223 	char *kaddr;
1224 	char *dst = (char *)dstv;
1225 	size_t start_offset = block_ctx->start & ((u64)PAGE_SIZE - 1);
1226 	unsigned long i = (start_offset + offset) >> PAGE_SHIFT;
1227 
1228 	WARN_ON(offset + len > block_ctx->len);
1229 	offset_in_page = (start_offset + offset) & (PAGE_SIZE - 1);
1230 
1231 	while (len > 0) {
1232 		cur = min(len, ((size_t)PAGE_SIZE - offset_in_page));
1233 		BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_SIZE));
1234 		kaddr = block_ctx->datav[i];
1235 		memcpy(dst, kaddr + offset_in_page, cur);
1236 
1237 		dst += cur;
1238 		len -= cur;
1239 		offset_in_page = 0;
1240 		i++;
1241 	}
1242 }
1243 
1244 static int btrfsic_create_link_to_next_block(
1245 		struct btrfsic_state *state,
1246 		struct btrfsic_block *block,
1247 		struct btrfsic_block_data_ctx *block_ctx,
1248 		u64 next_bytenr,
1249 		int limit_nesting,
1250 		struct btrfsic_block_data_ctx *next_block_ctx,
1251 		struct btrfsic_block **next_blockp,
1252 		int force_iodone_flag,
1253 		int *num_copiesp, int *mirror_nump,
1254 		struct btrfs_disk_key *disk_key,
1255 		u64 parent_generation)
1256 {
1257 	struct btrfs_fs_info *fs_info = state->fs_info;
1258 	struct btrfsic_block *next_block = NULL;
1259 	int ret;
1260 	struct btrfsic_block_link *l;
1261 	int did_alloc_block_link;
1262 	int block_was_created;
1263 
1264 	*next_blockp = NULL;
1265 	if (0 == *num_copiesp) {
1266 		*num_copiesp = btrfs_num_copies(fs_info, next_bytenr,
1267 						state->metablock_size);
1268 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1269 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
1270 			       next_bytenr, *num_copiesp);
1271 		*mirror_nump = 1;
1272 	}
1273 
1274 	if (*mirror_nump > *num_copiesp)
1275 		return 0;
1276 
1277 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1278 		pr_info("btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1279 		       *mirror_nump);
1280 	ret = btrfsic_map_block(state, next_bytenr,
1281 				state->metablock_size,
1282 				next_block_ctx, *mirror_nump);
1283 	if (ret) {
1284 		pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1285 		       next_bytenr, *mirror_nump);
1286 		btrfsic_release_block_ctx(next_block_ctx);
1287 		*next_blockp = NULL;
1288 		return -1;
1289 	}
1290 
1291 	next_block = btrfsic_block_lookup_or_add(state,
1292 						 next_block_ctx, "referenced ",
1293 						 1, force_iodone_flag,
1294 						 !force_iodone_flag,
1295 						 *mirror_nump,
1296 						 &block_was_created);
1297 	if (NULL == next_block) {
1298 		btrfsic_release_block_ctx(next_block_ctx);
1299 		*next_blockp = NULL;
1300 		return -1;
1301 	}
1302 	if (block_was_created) {
1303 		l = NULL;
1304 		next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1305 	} else {
1306 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1307 			if (next_block->logical_bytenr != next_bytenr &&
1308 			    !(!next_block->is_metadata &&
1309 			      0 == next_block->logical_bytenr))
1310 				pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1311 				       next_bytenr, next_block_ctx->dev->name,
1312 				       next_block_ctx->dev_bytenr, *mirror_nump,
1313 				       btrfsic_get_block_type(state,
1314 							      next_block),
1315 				       next_block->logical_bytenr);
1316 			else
1317 				pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1318 				       next_bytenr, next_block_ctx->dev->name,
1319 				       next_block_ctx->dev_bytenr, *mirror_nump,
1320 				       btrfsic_get_block_type(state,
1321 							      next_block));
1322 		}
1323 		next_block->logical_bytenr = next_bytenr;
1324 
1325 		next_block->mirror_num = *mirror_nump;
1326 		l = btrfsic_block_link_hashtable_lookup(
1327 				next_block_ctx->dev->bdev,
1328 				next_block_ctx->dev_bytenr,
1329 				block_ctx->dev->bdev,
1330 				block_ctx->dev_bytenr,
1331 				&state->block_link_hashtable);
1332 	}
1333 
1334 	next_block->disk_key = *disk_key;
1335 	if (NULL == l) {
1336 		l = btrfsic_block_link_alloc();
1337 		if (NULL == l) {
1338 			pr_info("btrfsic: error, kmalloc failed!\n");
1339 			btrfsic_release_block_ctx(next_block_ctx);
1340 			*next_blockp = NULL;
1341 			return -1;
1342 		}
1343 
1344 		did_alloc_block_link = 1;
1345 		l->block_ref_to = next_block;
1346 		l->block_ref_from = block;
1347 		l->ref_cnt = 1;
1348 		l->parent_generation = parent_generation;
1349 
1350 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1351 			btrfsic_print_add_link(state, l);
1352 
1353 		list_add(&l->node_ref_to, &block->ref_to_list);
1354 		list_add(&l->node_ref_from, &next_block->ref_from_list);
1355 
1356 		btrfsic_block_link_hashtable_add(l,
1357 						 &state->block_link_hashtable);
1358 	} else {
1359 		did_alloc_block_link = 0;
1360 		if (0 == limit_nesting) {
1361 			l->ref_cnt++;
1362 			l->parent_generation = parent_generation;
1363 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1364 				btrfsic_print_add_link(state, l);
1365 		}
1366 	}
1367 
1368 	if (limit_nesting > 0 && did_alloc_block_link) {
1369 		ret = btrfsic_read_block(state, next_block_ctx);
1370 		if (ret < (int)next_block_ctx->len) {
1371 			pr_info("btrfsic: read block @logical %llu failed!\n",
1372 			       next_bytenr);
1373 			btrfsic_release_block_ctx(next_block_ctx);
1374 			*next_blockp = NULL;
1375 			return -1;
1376 		}
1377 
1378 		*next_blockp = next_block;
1379 	} else {
1380 		*next_blockp = NULL;
1381 	}
1382 	(*mirror_nump)++;
1383 
1384 	return 0;
1385 }
1386 
1387 static int btrfsic_handle_extent_data(
1388 		struct btrfsic_state *state,
1389 		struct btrfsic_block *block,
1390 		struct btrfsic_block_data_ctx *block_ctx,
1391 		u32 item_offset, int force_iodone_flag)
1392 {
1393 	struct btrfs_fs_info *fs_info = state->fs_info;
1394 	struct btrfs_file_extent_item file_extent_item;
1395 	u64 file_extent_item_offset;
1396 	u64 next_bytenr;
1397 	u64 num_bytes;
1398 	u64 generation;
1399 	struct btrfsic_block_link *l;
1400 	int ret;
1401 
1402 	file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1403 				  item_offset;
1404 	if (file_extent_item_offset +
1405 	    offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1406 	    block_ctx->len) {
1407 		pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1408 		       block_ctx->start, block_ctx->dev->name);
1409 		return -1;
1410 	}
1411 
1412 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1413 		file_extent_item_offset,
1414 		offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1415 	if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1416 	    btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1417 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1418 			pr_info("extent_data: type %u, disk_bytenr = %llu\n",
1419 			       file_extent_item.type,
1420 			       btrfs_stack_file_extent_disk_bytenr(
1421 			       &file_extent_item));
1422 		return 0;
1423 	}
1424 
1425 	if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1426 	    block_ctx->len) {
1427 		pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1428 		       block_ctx->start, block_ctx->dev->name);
1429 		return -1;
1430 	}
1431 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1432 				     file_extent_item_offset,
1433 				     sizeof(struct btrfs_file_extent_item));
1434 	next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item);
1435 	if (btrfs_stack_file_extent_compression(&file_extent_item) ==
1436 	    BTRFS_COMPRESS_NONE) {
1437 		next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item);
1438 		num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1439 	} else {
1440 		num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item);
1441 	}
1442 	generation = btrfs_stack_file_extent_generation(&file_extent_item);
1443 
1444 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1445 		pr_info("extent_data: type %u, disk_bytenr = %llu, offset = %llu, num_bytes = %llu\n",
1446 		       file_extent_item.type,
1447 		       btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1448 		       btrfs_stack_file_extent_offset(&file_extent_item),
1449 		       num_bytes);
1450 	while (num_bytes > 0) {
1451 		u32 chunk_len;
1452 		int num_copies;
1453 		int mirror_num;
1454 
1455 		if (num_bytes > state->datablock_size)
1456 			chunk_len = state->datablock_size;
1457 		else
1458 			chunk_len = num_bytes;
1459 
1460 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
1461 					      state->datablock_size);
1462 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1463 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
1464 			       next_bytenr, num_copies);
1465 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1466 			struct btrfsic_block_data_ctx next_block_ctx;
1467 			struct btrfsic_block *next_block;
1468 			int block_was_created;
1469 
1470 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1471 				pr_info("btrfsic_handle_extent_data(mirror_num=%d)\n",
1472 					mirror_num);
1473 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1474 				pr_info("\tdisk_bytenr = %llu, num_bytes %u\n",
1475 				       next_bytenr, chunk_len);
1476 			ret = btrfsic_map_block(state, next_bytenr,
1477 						chunk_len, &next_block_ctx,
1478 						mirror_num);
1479 			if (ret) {
1480 				pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1481 				       next_bytenr, mirror_num);
1482 				return -1;
1483 			}
1484 
1485 			next_block = btrfsic_block_lookup_or_add(
1486 					state,
1487 					&next_block_ctx,
1488 					"referenced ",
1489 					0,
1490 					force_iodone_flag,
1491 					!force_iodone_flag,
1492 					mirror_num,
1493 					&block_was_created);
1494 			if (NULL == next_block) {
1495 				pr_info("btrfsic: error, kmalloc failed!\n");
1496 				btrfsic_release_block_ctx(&next_block_ctx);
1497 				return -1;
1498 			}
1499 			if (!block_was_created) {
1500 				if ((state->print_mask &
1501 				     BTRFSIC_PRINT_MASK_VERBOSE) &&
1502 				    next_block->logical_bytenr != next_bytenr &&
1503 				    !(!next_block->is_metadata &&
1504 				      0 == next_block->logical_bytenr)) {
1505 					pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, D, bytenr mismatch (!= stored %llu).\n",
1506 					       next_bytenr,
1507 					       next_block_ctx.dev->name,
1508 					       next_block_ctx.dev_bytenr,
1509 					       mirror_num,
1510 					       next_block->logical_bytenr);
1511 				}
1512 				next_block->logical_bytenr = next_bytenr;
1513 				next_block->mirror_num = mirror_num;
1514 			}
1515 
1516 			l = btrfsic_block_link_lookup_or_add(state,
1517 							     &next_block_ctx,
1518 							     next_block, block,
1519 							     generation);
1520 			btrfsic_release_block_ctx(&next_block_ctx);
1521 			if (NULL == l)
1522 				return -1;
1523 		}
1524 
1525 		next_bytenr += chunk_len;
1526 		num_bytes -= chunk_len;
1527 	}
1528 
1529 	return 0;
1530 }
1531 
1532 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1533 			     struct btrfsic_block_data_ctx *block_ctx_out,
1534 			     int mirror_num)
1535 {
1536 	struct btrfs_fs_info *fs_info = state->fs_info;
1537 	int ret;
1538 	u64 length;
1539 	struct btrfs_bio *multi = NULL;
1540 	struct btrfs_device *device;
1541 
1542 	length = len;
1543 	ret = btrfs_map_block(fs_info, BTRFS_MAP_READ,
1544 			      bytenr, &length, &multi, mirror_num);
1545 
1546 	if (ret) {
1547 		block_ctx_out->start = 0;
1548 		block_ctx_out->dev_bytenr = 0;
1549 		block_ctx_out->len = 0;
1550 		block_ctx_out->dev = NULL;
1551 		block_ctx_out->datav = NULL;
1552 		block_ctx_out->pagev = NULL;
1553 		block_ctx_out->mem_to_free = NULL;
1554 
1555 		return ret;
1556 	}
1557 
1558 	device = multi->stripes[0].dev;
1559 	block_ctx_out->dev = btrfsic_dev_state_lookup(device->bdev);
1560 	block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1561 	block_ctx_out->start = bytenr;
1562 	block_ctx_out->len = len;
1563 	block_ctx_out->datav = NULL;
1564 	block_ctx_out->pagev = NULL;
1565 	block_ctx_out->mem_to_free = NULL;
1566 
1567 	kfree(multi);
1568 	if (NULL == block_ctx_out->dev) {
1569 		ret = -ENXIO;
1570 		pr_info("btrfsic: error, cannot lookup dev (#1)!\n");
1571 	}
1572 
1573 	return ret;
1574 }
1575 
1576 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1577 {
1578 	if (block_ctx->mem_to_free) {
1579 		unsigned int num_pages;
1580 
1581 		BUG_ON(!block_ctx->datav);
1582 		BUG_ON(!block_ctx->pagev);
1583 		num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1584 			    PAGE_SHIFT;
1585 		while (num_pages > 0) {
1586 			num_pages--;
1587 			if (block_ctx->datav[num_pages]) {
1588 				kunmap(block_ctx->pagev[num_pages]);
1589 				block_ctx->datav[num_pages] = NULL;
1590 			}
1591 			if (block_ctx->pagev[num_pages]) {
1592 				__free_page(block_ctx->pagev[num_pages]);
1593 				block_ctx->pagev[num_pages] = NULL;
1594 			}
1595 		}
1596 
1597 		kfree(block_ctx->mem_to_free);
1598 		block_ctx->mem_to_free = NULL;
1599 		block_ctx->pagev = NULL;
1600 		block_ctx->datav = NULL;
1601 	}
1602 }
1603 
1604 static int btrfsic_read_block(struct btrfsic_state *state,
1605 			      struct btrfsic_block_data_ctx *block_ctx)
1606 {
1607 	unsigned int num_pages;
1608 	unsigned int i;
1609 	u64 dev_bytenr;
1610 	int ret;
1611 
1612 	BUG_ON(block_ctx->datav);
1613 	BUG_ON(block_ctx->pagev);
1614 	BUG_ON(block_ctx->mem_to_free);
1615 	if (block_ctx->dev_bytenr & ((u64)PAGE_SIZE - 1)) {
1616 		pr_info("btrfsic: read_block() with unaligned bytenr %llu\n",
1617 		       block_ctx->dev_bytenr);
1618 		return -1;
1619 	}
1620 
1621 	num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1622 		    PAGE_SHIFT;
1623 	block_ctx->mem_to_free = kzalloc((sizeof(*block_ctx->datav) +
1624 					  sizeof(*block_ctx->pagev)) *
1625 					 num_pages, GFP_NOFS);
1626 	if (!block_ctx->mem_to_free)
1627 		return -ENOMEM;
1628 	block_ctx->datav = block_ctx->mem_to_free;
1629 	block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1630 	for (i = 0; i < num_pages; i++) {
1631 		block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1632 		if (!block_ctx->pagev[i])
1633 			return -1;
1634 	}
1635 
1636 	dev_bytenr = block_ctx->dev_bytenr;
1637 	for (i = 0; i < num_pages;) {
1638 		struct bio *bio;
1639 		unsigned int j;
1640 
1641 		bio = btrfs_io_bio_alloc(GFP_NOFS, num_pages - i);
1642 		if (!bio) {
1643 			pr_info("btrfsic: bio_alloc() for %u pages failed!\n",
1644 			       num_pages - i);
1645 			return -1;
1646 		}
1647 		bio->bi_bdev = block_ctx->dev->bdev;
1648 		bio->bi_iter.bi_sector = dev_bytenr >> 9;
1649 		bio_set_op_attrs(bio, REQ_OP_READ, 0);
1650 
1651 		for (j = i; j < num_pages; j++) {
1652 			ret = bio_add_page(bio, block_ctx->pagev[j],
1653 					   PAGE_SIZE, 0);
1654 			if (PAGE_SIZE != ret)
1655 				break;
1656 		}
1657 		if (j == i) {
1658 			pr_info("btrfsic: error, failed to add a single page!\n");
1659 			return -1;
1660 		}
1661 		if (submit_bio_wait(bio)) {
1662 			pr_info("btrfsic: read error at logical %llu dev %s!\n",
1663 			       block_ctx->start, block_ctx->dev->name);
1664 			bio_put(bio);
1665 			return -1;
1666 		}
1667 		bio_put(bio);
1668 		dev_bytenr += (j - i) * PAGE_SIZE;
1669 		i = j;
1670 	}
1671 	for (i = 0; i < num_pages; i++) {
1672 		block_ctx->datav[i] = kmap(block_ctx->pagev[i]);
1673 		if (!block_ctx->datav[i]) {
1674 			pr_info("btrfsic: kmap() failed (dev %s)!\n",
1675 			       block_ctx->dev->name);
1676 			return -1;
1677 		}
1678 	}
1679 
1680 	return block_ctx->len;
1681 }
1682 
1683 static void btrfsic_dump_database(struct btrfsic_state *state)
1684 {
1685 	const struct btrfsic_block *b_all;
1686 
1687 	BUG_ON(NULL == state);
1688 
1689 	pr_info("all_blocks_list:\n");
1690 	list_for_each_entry(b_all, &state->all_blocks_list, all_blocks_node) {
1691 		const struct btrfsic_block_link *l;
1692 
1693 		pr_info("%c-block @%llu (%s/%llu/%d)\n",
1694 		       btrfsic_get_block_type(state, b_all),
1695 		       b_all->logical_bytenr, b_all->dev_state->name,
1696 		       b_all->dev_bytenr, b_all->mirror_num);
1697 
1698 		list_for_each_entry(l, &b_all->ref_to_list, node_ref_to) {
1699 			pr_info(" %c @%llu (%s/%llu/%d) refers %u* to %c @%llu (%s/%llu/%d)\n",
1700 			       btrfsic_get_block_type(state, b_all),
1701 			       b_all->logical_bytenr, b_all->dev_state->name,
1702 			       b_all->dev_bytenr, b_all->mirror_num,
1703 			       l->ref_cnt,
1704 			       btrfsic_get_block_type(state, l->block_ref_to),
1705 			       l->block_ref_to->logical_bytenr,
1706 			       l->block_ref_to->dev_state->name,
1707 			       l->block_ref_to->dev_bytenr,
1708 			       l->block_ref_to->mirror_num);
1709 		}
1710 
1711 		list_for_each_entry(l, &b_all->ref_from_list, node_ref_from) {
1712 			pr_info(" %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
1713 			       btrfsic_get_block_type(state, b_all),
1714 			       b_all->logical_bytenr, b_all->dev_state->name,
1715 			       b_all->dev_bytenr, b_all->mirror_num,
1716 			       l->ref_cnt,
1717 			       btrfsic_get_block_type(state, l->block_ref_from),
1718 			       l->block_ref_from->logical_bytenr,
1719 			       l->block_ref_from->dev_state->name,
1720 			       l->block_ref_from->dev_bytenr,
1721 			       l->block_ref_from->mirror_num);
1722 		}
1723 
1724 		pr_info("\n");
1725 	}
1726 }
1727 
1728 /*
1729  * Test whether the disk block contains a tree block (leaf or node)
1730  * (note that this test fails for the super block)
1731  */
1732 static int btrfsic_test_for_metadata(struct btrfsic_state *state,
1733 				     char **datav, unsigned int num_pages)
1734 {
1735 	struct btrfs_fs_info *fs_info = state->fs_info;
1736 	struct btrfs_header *h;
1737 	u8 csum[BTRFS_CSUM_SIZE];
1738 	u32 crc = ~(u32)0;
1739 	unsigned int i;
1740 
1741 	if (num_pages * PAGE_SIZE < state->metablock_size)
1742 		return 1; /* not metadata */
1743 	num_pages = state->metablock_size >> PAGE_SHIFT;
1744 	h = (struct btrfs_header *)datav[0];
1745 
1746 	if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
1747 		return 1;
1748 
1749 	for (i = 0; i < num_pages; i++) {
1750 		u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1751 		size_t sublen = i ? PAGE_SIZE :
1752 				    (PAGE_SIZE - BTRFS_CSUM_SIZE);
1753 
1754 		crc = btrfs_crc32c(crc, data, sublen);
1755 	}
1756 	btrfs_csum_final(crc, csum);
1757 	if (memcmp(csum, h->csum, state->csum_size))
1758 		return 1;
1759 
1760 	return 0; /* is metadata */
1761 }
1762 
1763 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1764 					  u64 dev_bytenr, char **mapped_datav,
1765 					  unsigned int num_pages,
1766 					  struct bio *bio, int *bio_is_patched,
1767 					  struct buffer_head *bh,
1768 					  int submit_bio_bh_rw)
1769 {
1770 	int is_metadata;
1771 	struct btrfsic_block *block;
1772 	struct btrfsic_block_data_ctx block_ctx;
1773 	int ret;
1774 	struct btrfsic_state *state = dev_state->state;
1775 	struct block_device *bdev = dev_state->bdev;
1776 	unsigned int processed_len;
1777 
1778 	if (NULL != bio_is_patched)
1779 		*bio_is_patched = 0;
1780 
1781 again:
1782 	if (num_pages == 0)
1783 		return;
1784 
1785 	processed_len = 0;
1786 	is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1787 						      num_pages));
1788 
1789 	block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1790 					       &state->block_hashtable);
1791 	if (NULL != block) {
1792 		u64 bytenr = 0;
1793 		struct btrfsic_block_link *l, *tmp;
1794 
1795 		if (block->is_superblock) {
1796 			bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1797 						    mapped_datav[0]);
1798 			if (num_pages * PAGE_SIZE <
1799 			    BTRFS_SUPER_INFO_SIZE) {
1800 				pr_info("btrfsic: cannot work with too short bios!\n");
1801 				return;
1802 			}
1803 			is_metadata = 1;
1804 			BUG_ON(BTRFS_SUPER_INFO_SIZE & (PAGE_SIZE - 1));
1805 			processed_len = BTRFS_SUPER_INFO_SIZE;
1806 			if (state->print_mask &
1807 			    BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1808 				pr_info("[before new superblock is written]:\n");
1809 				btrfsic_dump_tree_sub(state, block, 0);
1810 			}
1811 		}
1812 		if (is_metadata) {
1813 			if (!block->is_superblock) {
1814 				if (num_pages * PAGE_SIZE <
1815 				    state->metablock_size) {
1816 					pr_info("btrfsic: cannot work with too short bios!\n");
1817 					return;
1818 				}
1819 				processed_len = state->metablock_size;
1820 				bytenr = btrfs_stack_header_bytenr(
1821 						(struct btrfs_header *)
1822 						mapped_datav[0]);
1823 				btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1824 							       dev_state,
1825 							       dev_bytenr);
1826 			}
1827 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1828 				if (block->logical_bytenr != bytenr &&
1829 				    !(!block->is_metadata &&
1830 				      block->logical_bytenr == 0))
1831 					pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1832 					       bytenr, dev_state->name,
1833 					       dev_bytenr,
1834 					       block->mirror_num,
1835 					       btrfsic_get_block_type(state,
1836 								      block),
1837 					       block->logical_bytenr);
1838 				else
1839 					pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1840 					       bytenr, dev_state->name,
1841 					       dev_bytenr, block->mirror_num,
1842 					       btrfsic_get_block_type(state,
1843 								      block));
1844 			}
1845 			block->logical_bytenr = bytenr;
1846 		} else {
1847 			if (num_pages * PAGE_SIZE <
1848 			    state->datablock_size) {
1849 				pr_info("btrfsic: cannot work with too short bios!\n");
1850 				return;
1851 			}
1852 			processed_len = state->datablock_size;
1853 			bytenr = block->logical_bytenr;
1854 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1855 				pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1856 				       bytenr, dev_state->name, dev_bytenr,
1857 				       block->mirror_num,
1858 				       btrfsic_get_block_type(state, block));
1859 		}
1860 
1861 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1862 			pr_info("ref_to_list: %cE, ref_from_list: %cE\n",
1863 			       list_empty(&block->ref_to_list) ? ' ' : '!',
1864 			       list_empty(&block->ref_from_list) ? ' ' : '!');
1865 		if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1866 			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",
1867 			       btrfsic_get_block_type(state, block), bytenr,
1868 			       dev_state->name, dev_bytenr, block->mirror_num,
1869 			       block->generation,
1870 			       btrfs_disk_key_objectid(&block->disk_key),
1871 			       block->disk_key.type,
1872 			       btrfs_disk_key_offset(&block->disk_key),
1873 			       btrfs_stack_header_generation(
1874 				       (struct btrfs_header *) mapped_datav[0]),
1875 			       state->max_superblock_generation);
1876 			btrfsic_dump_tree(state);
1877 		}
1878 
1879 		if (!block->is_iodone && !block->never_written) {
1880 			pr_info("btrfs: attempt to overwrite %c-block @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu, which is not yet iodone!\n",
1881 			       btrfsic_get_block_type(state, block), bytenr,
1882 			       dev_state->name, dev_bytenr, block->mirror_num,
1883 			       block->generation,
1884 			       btrfs_stack_header_generation(
1885 				       (struct btrfs_header *)
1886 				       mapped_datav[0]));
1887 			/* it would not be safe to go on */
1888 			btrfsic_dump_tree(state);
1889 			goto continue_loop;
1890 		}
1891 
1892 		/*
1893 		 * Clear all references of this block. Do not free
1894 		 * the block itself even if is not referenced anymore
1895 		 * because it still carries valuable information
1896 		 * like whether it was ever written and IO completed.
1897 		 */
1898 		list_for_each_entry_safe(l, tmp, &block->ref_to_list,
1899 					 node_ref_to) {
1900 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1901 				btrfsic_print_rem_link(state, l);
1902 			l->ref_cnt--;
1903 			if (0 == l->ref_cnt) {
1904 				list_del(&l->node_ref_to);
1905 				list_del(&l->node_ref_from);
1906 				btrfsic_block_link_hashtable_remove(l);
1907 				btrfsic_block_link_free(l);
1908 			}
1909 		}
1910 
1911 		block_ctx.dev = dev_state;
1912 		block_ctx.dev_bytenr = dev_bytenr;
1913 		block_ctx.start = bytenr;
1914 		block_ctx.len = processed_len;
1915 		block_ctx.pagev = NULL;
1916 		block_ctx.mem_to_free = NULL;
1917 		block_ctx.datav = mapped_datav;
1918 
1919 		if (is_metadata || state->include_extent_data) {
1920 			block->never_written = 0;
1921 			block->iodone_w_error = 0;
1922 			if (NULL != bio) {
1923 				block->is_iodone = 0;
1924 				BUG_ON(NULL == bio_is_patched);
1925 				if (!*bio_is_patched) {
1926 					block->orig_bio_bh_private =
1927 					    bio->bi_private;
1928 					block->orig_bio_bh_end_io.bio =
1929 					    bio->bi_end_io;
1930 					block->next_in_same_bio = NULL;
1931 					bio->bi_private = block;
1932 					bio->bi_end_io = btrfsic_bio_end_io;
1933 					*bio_is_patched = 1;
1934 				} else {
1935 					struct btrfsic_block *chained_block =
1936 					    (struct btrfsic_block *)
1937 					    bio->bi_private;
1938 
1939 					BUG_ON(NULL == chained_block);
1940 					block->orig_bio_bh_private =
1941 					    chained_block->orig_bio_bh_private;
1942 					block->orig_bio_bh_end_io.bio =
1943 					    chained_block->orig_bio_bh_end_io.
1944 					    bio;
1945 					block->next_in_same_bio = chained_block;
1946 					bio->bi_private = block;
1947 				}
1948 			} else if (NULL != bh) {
1949 				block->is_iodone = 0;
1950 				block->orig_bio_bh_private = bh->b_private;
1951 				block->orig_bio_bh_end_io.bh = bh->b_end_io;
1952 				block->next_in_same_bio = NULL;
1953 				bh->b_private = block;
1954 				bh->b_end_io = btrfsic_bh_end_io;
1955 			} else {
1956 				block->is_iodone = 1;
1957 				block->orig_bio_bh_private = NULL;
1958 				block->orig_bio_bh_end_io.bio = NULL;
1959 				block->next_in_same_bio = NULL;
1960 			}
1961 		}
1962 
1963 		block->flush_gen = dev_state->last_flush_gen + 1;
1964 		block->submit_bio_bh_rw = submit_bio_bh_rw;
1965 		if (is_metadata) {
1966 			block->logical_bytenr = bytenr;
1967 			block->is_metadata = 1;
1968 			if (block->is_superblock) {
1969 				BUG_ON(PAGE_SIZE !=
1970 				       BTRFS_SUPER_INFO_SIZE);
1971 				ret = btrfsic_process_written_superblock(
1972 						state,
1973 						block,
1974 						(struct btrfs_super_block *)
1975 						mapped_datav[0]);
1976 				if (state->print_mask &
1977 				    BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
1978 					pr_info("[after new superblock is written]:\n");
1979 					btrfsic_dump_tree_sub(state, block, 0);
1980 				}
1981 			} else {
1982 				block->mirror_num = 0;	/* unknown */
1983 				ret = btrfsic_process_metablock(
1984 						state,
1985 						block,
1986 						&block_ctx,
1987 						0, 0);
1988 			}
1989 			if (ret)
1990 				pr_info("btrfsic: btrfsic_process_metablock(root @%llu) failed!\n",
1991 				       dev_bytenr);
1992 		} else {
1993 			block->is_metadata = 0;
1994 			block->mirror_num = 0;	/* unknown */
1995 			block->generation = BTRFSIC_GENERATION_UNKNOWN;
1996 			if (!state->include_extent_data
1997 			    && list_empty(&block->ref_from_list)) {
1998 				/*
1999 				 * disk block is overwritten with extent
2000 				 * data (not meta data) and we are configured
2001 				 * to not include extent data: take the
2002 				 * chance and free the block's memory
2003 				 */
2004 				btrfsic_block_hashtable_remove(block);
2005 				list_del(&block->all_blocks_node);
2006 				btrfsic_block_free(block);
2007 			}
2008 		}
2009 		btrfsic_release_block_ctx(&block_ctx);
2010 	} else {
2011 		/* block has not been found in hash table */
2012 		u64 bytenr;
2013 
2014 		if (!is_metadata) {
2015 			processed_len = state->datablock_size;
2016 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2017 				pr_info("Written block (%s/%llu/?) !found in hash table, D.\n",
2018 				       dev_state->name, dev_bytenr);
2019 			if (!state->include_extent_data) {
2020 				/* ignore that written D block */
2021 				goto continue_loop;
2022 			}
2023 
2024 			/* this is getting ugly for the
2025 			 * include_extent_data case... */
2026 			bytenr = 0;	/* unknown */
2027 		} else {
2028 			processed_len = state->metablock_size;
2029 			bytenr = btrfs_stack_header_bytenr(
2030 					(struct btrfs_header *)
2031 					mapped_datav[0]);
2032 			btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
2033 						       dev_bytenr);
2034 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2035 				pr_info("Written block @%llu (%s/%llu/?) !found in hash table, M.\n",
2036 				       bytenr, dev_state->name, dev_bytenr);
2037 		}
2038 
2039 		block_ctx.dev = dev_state;
2040 		block_ctx.dev_bytenr = dev_bytenr;
2041 		block_ctx.start = bytenr;
2042 		block_ctx.len = processed_len;
2043 		block_ctx.pagev = NULL;
2044 		block_ctx.mem_to_free = NULL;
2045 		block_ctx.datav = mapped_datav;
2046 
2047 		block = btrfsic_block_alloc();
2048 		if (NULL == block) {
2049 			pr_info("btrfsic: error, kmalloc failed!\n");
2050 			btrfsic_release_block_ctx(&block_ctx);
2051 			goto continue_loop;
2052 		}
2053 		block->dev_state = dev_state;
2054 		block->dev_bytenr = dev_bytenr;
2055 		block->logical_bytenr = bytenr;
2056 		block->is_metadata = is_metadata;
2057 		block->never_written = 0;
2058 		block->iodone_w_error = 0;
2059 		block->mirror_num = 0;	/* unknown */
2060 		block->flush_gen = dev_state->last_flush_gen + 1;
2061 		block->submit_bio_bh_rw = submit_bio_bh_rw;
2062 		if (NULL != bio) {
2063 			block->is_iodone = 0;
2064 			BUG_ON(NULL == bio_is_patched);
2065 			if (!*bio_is_patched) {
2066 				block->orig_bio_bh_private = bio->bi_private;
2067 				block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2068 				block->next_in_same_bio = NULL;
2069 				bio->bi_private = block;
2070 				bio->bi_end_io = btrfsic_bio_end_io;
2071 				*bio_is_patched = 1;
2072 			} else {
2073 				struct btrfsic_block *chained_block =
2074 				    (struct btrfsic_block *)
2075 				    bio->bi_private;
2076 
2077 				BUG_ON(NULL == chained_block);
2078 				block->orig_bio_bh_private =
2079 				    chained_block->orig_bio_bh_private;
2080 				block->orig_bio_bh_end_io.bio =
2081 				    chained_block->orig_bio_bh_end_io.bio;
2082 				block->next_in_same_bio = chained_block;
2083 				bio->bi_private = block;
2084 			}
2085 		} else if (NULL != bh) {
2086 			block->is_iodone = 0;
2087 			block->orig_bio_bh_private = bh->b_private;
2088 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2089 			block->next_in_same_bio = NULL;
2090 			bh->b_private = block;
2091 			bh->b_end_io = btrfsic_bh_end_io;
2092 		} else {
2093 			block->is_iodone = 1;
2094 			block->orig_bio_bh_private = NULL;
2095 			block->orig_bio_bh_end_io.bio = NULL;
2096 			block->next_in_same_bio = NULL;
2097 		}
2098 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2099 			pr_info("New written %c-block @%llu (%s/%llu/%d)\n",
2100 			       is_metadata ? 'M' : 'D',
2101 			       block->logical_bytenr, block->dev_state->name,
2102 			       block->dev_bytenr, block->mirror_num);
2103 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2104 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2105 
2106 		if (is_metadata) {
2107 			ret = btrfsic_process_metablock(state, block,
2108 							&block_ctx, 0, 0);
2109 			if (ret)
2110 				pr_info("btrfsic: process_metablock(root @%llu) failed!\n",
2111 				       dev_bytenr);
2112 		}
2113 		btrfsic_release_block_ctx(&block_ctx);
2114 	}
2115 
2116 continue_loop:
2117 	BUG_ON(!processed_len);
2118 	dev_bytenr += processed_len;
2119 	mapped_datav += processed_len >> PAGE_SHIFT;
2120 	num_pages -= processed_len >> PAGE_SHIFT;
2121 	goto again;
2122 }
2123 
2124 static void btrfsic_bio_end_io(struct bio *bp)
2125 {
2126 	struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2127 	int iodone_w_error;
2128 
2129 	/* mutex is not held! This is not save if IO is not yet completed
2130 	 * on umount */
2131 	iodone_w_error = 0;
2132 	if (bp->bi_error)
2133 		iodone_w_error = 1;
2134 
2135 	BUG_ON(NULL == block);
2136 	bp->bi_private = block->orig_bio_bh_private;
2137 	bp->bi_end_io = block->orig_bio_bh_end_io.bio;
2138 
2139 	do {
2140 		struct btrfsic_block *next_block;
2141 		struct btrfsic_dev_state *const dev_state = block->dev_state;
2142 
2143 		if ((dev_state->state->print_mask &
2144 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2145 			pr_info("bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2146 			       bp->bi_error,
2147 			       btrfsic_get_block_type(dev_state->state, block),
2148 			       block->logical_bytenr, dev_state->name,
2149 			       block->dev_bytenr, block->mirror_num);
2150 		next_block = block->next_in_same_bio;
2151 		block->iodone_w_error = iodone_w_error;
2152 		if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2153 			dev_state->last_flush_gen++;
2154 			if ((dev_state->state->print_mask &
2155 			     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2156 				pr_info("bio_end_io() new %s flush_gen=%llu\n",
2157 				       dev_state->name,
2158 				       dev_state->last_flush_gen);
2159 		}
2160 		if (block->submit_bio_bh_rw & REQ_FUA)
2161 			block->flush_gen = 0; /* FUA completed means block is
2162 					       * on disk */
2163 		block->is_iodone = 1; /* for FLUSH, this releases the block */
2164 		block = next_block;
2165 	} while (NULL != block);
2166 
2167 	bp->bi_end_io(bp);
2168 }
2169 
2170 static void btrfsic_bh_end_io(struct buffer_head *bh, int uptodate)
2171 {
2172 	struct btrfsic_block *block = (struct btrfsic_block *)bh->b_private;
2173 	int iodone_w_error = !uptodate;
2174 	struct btrfsic_dev_state *dev_state;
2175 
2176 	BUG_ON(NULL == block);
2177 	dev_state = block->dev_state;
2178 	if ((dev_state->state->print_mask & BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2179 		pr_info("bh_end_io(error=%d) for %c @%llu (%s/%llu/%d)\n",
2180 		       iodone_w_error,
2181 		       btrfsic_get_block_type(dev_state->state, block),
2182 		       block->logical_bytenr, block->dev_state->name,
2183 		       block->dev_bytenr, block->mirror_num);
2184 
2185 	block->iodone_w_error = iodone_w_error;
2186 	if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2187 		dev_state->last_flush_gen++;
2188 		if ((dev_state->state->print_mask &
2189 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2190 			pr_info("bh_end_io() new %s flush_gen=%llu\n",
2191 			       dev_state->name, dev_state->last_flush_gen);
2192 	}
2193 	if (block->submit_bio_bh_rw & REQ_FUA)
2194 		block->flush_gen = 0; /* FUA completed means block is on disk */
2195 
2196 	bh->b_private = block->orig_bio_bh_private;
2197 	bh->b_end_io = block->orig_bio_bh_end_io.bh;
2198 	block->is_iodone = 1; /* for FLUSH, this releases the block */
2199 	bh->b_end_io(bh, uptodate);
2200 }
2201 
2202 static int btrfsic_process_written_superblock(
2203 		struct btrfsic_state *state,
2204 		struct btrfsic_block *const superblock,
2205 		struct btrfs_super_block *const super_hdr)
2206 {
2207 	struct btrfs_fs_info *fs_info = state->fs_info;
2208 	int pass;
2209 
2210 	superblock->generation = btrfs_super_generation(super_hdr);
2211 	if (!(superblock->generation > state->max_superblock_generation ||
2212 	      0 == state->max_superblock_generation)) {
2213 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2214 			pr_info("btrfsic: superblock @%llu (%s/%llu/%d) with old gen %llu <= %llu\n",
2215 			       superblock->logical_bytenr,
2216 			       superblock->dev_state->name,
2217 			       superblock->dev_bytenr, superblock->mirror_num,
2218 			       btrfs_super_generation(super_hdr),
2219 			       state->max_superblock_generation);
2220 	} else {
2221 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2222 			pr_info("btrfsic: got new superblock @%llu (%s/%llu/%d) with new gen %llu > %llu\n",
2223 			       superblock->logical_bytenr,
2224 			       superblock->dev_state->name,
2225 			       superblock->dev_bytenr, superblock->mirror_num,
2226 			       btrfs_super_generation(super_hdr),
2227 			       state->max_superblock_generation);
2228 
2229 		state->max_superblock_generation =
2230 		    btrfs_super_generation(super_hdr);
2231 		state->latest_superblock = superblock;
2232 	}
2233 
2234 	for (pass = 0; pass < 3; pass++) {
2235 		int ret;
2236 		u64 next_bytenr;
2237 		struct btrfsic_block *next_block;
2238 		struct btrfsic_block_data_ctx tmp_next_block_ctx;
2239 		struct btrfsic_block_link *l;
2240 		int num_copies;
2241 		int mirror_num;
2242 		const char *additional_string = NULL;
2243 		struct btrfs_disk_key tmp_disk_key = {0};
2244 
2245 		btrfs_set_disk_key_objectid(&tmp_disk_key,
2246 					    BTRFS_ROOT_ITEM_KEY);
2247 		btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2248 
2249 		switch (pass) {
2250 		case 0:
2251 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2252 						    BTRFS_ROOT_TREE_OBJECTID);
2253 			additional_string = "root ";
2254 			next_bytenr = btrfs_super_root(super_hdr);
2255 			if (state->print_mask &
2256 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2257 				pr_info("root@%llu\n", next_bytenr);
2258 			break;
2259 		case 1:
2260 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2261 						    BTRFS_CHUNK_TREE_OBJECTID);
2262 			additional_string = "chunk ";
2263 			next_bytenr = btrfs_super_chunk_root(super_hdr);
2264 			if (state->print_mask &
2265 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2266 				pr_info("chunk@%llu\n", next_bytenr);
2267 			break;
2268 		case 2:
2269 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2270 						    BTRFS_TREE_LOG_OBJECTID);
2271 			additional_string = "log ";
2272 			next_bytenr = btrfs_super_log_root(super_hdr);
2273 			if (0 == next_bytenr)
2274 				continue;
2275 			if (state->print_mask &
2276 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2277 				pr_info("log@%llu\n", next_bytenr);
2278 			break;
2279 		}
2280 
2281 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
2282 					      BTRFS_SUPER_INFO_SIZE);
2283 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2284 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
2285 			       next_bytenr, num_copies);
2286 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2287 			int was_created;
2288 
2289 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2290 				pr_info("btrfsic_process_written_superblock(mirror_num=%d)\n", mirror_num);
2291 			ret = btrfsic_map_block(state, next_bytenr,
2292 						BTRFS_SUPER_INFO_SIZE,
2293 						&tmp_next_block_ctx,
2294 						mirror_num);
2295 			if (ret) {
2296 				pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
2297 				       next_bytenr, mirror_num);
2298 				return -1;
2299 			}
2300 
2301 			next_block = btrfsic_block_lookup_or_add(
2302 					state,
2303 					&tmp_next_block_ctx,
2304 					additional_string,
2305 					1, 0, 1,
2306 					mirror_num,
2307 					&was_created);
2308 			if (NULL == next_block) {
2309 				pr_info("btrfsic: error, kmalloc failed!\n");
2310 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
2311 				return -1;
2312 			}
2313 
2314 			next_block->disk_key = tmp_disk_key;
2315 			if (was_created)
2316 				next_block->generation =
2317 				    BTRFSIC_GENERATION_UNKNOWN;
2318 			l = btrfsic_block_link_lookup_or_add(
2319 					state,
2320 					&tmp_next_block_ctx,
2321 					next_block,
2322 					superblock,
2323 					BTRFSIC_GENERATION_UNKNOWN);
2324 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
2325 			if (NULL == l)
2326 				return -1;
2327 		}
2328 	}
2329 
2330 	if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2331 		btrfsic_dump_tree(state);
2332 
2333 	return 0;
2334 }
2335 
2336 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2337 					struct btrfsic_block *const block,
2338 					int recursion_level)
2339 {
2340 	const struct btrfsic_block_link *l;
2341 	int ret = 0;
2342 
2343 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2344 		/*
2345 		 * Note that this situation can happen and does not
2346 		 * indicate an error in regular cases. It happens
2347 		 * when disk blocks are freed and later reused.
2348 		 * The check-integrity module is not aware of any
2349 		 * block free operations, it just recognizes block
2350 		 * write operations. Therefore it keeps the linkage
2351 		 * information for a block until a block is
2352 		 * rewritten. This can temporarily cause incorrect
2353 		 * and even circular linkage informations. This
2354 		 * causes no harm unless such blocks are referenced
2355 		 * by the most recent super block.
2356 		 */
2357 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2358 			pr_info("btrfsic: abort cyclic linkage (case 1).\n");
2359 
2360 		return ret;
2361 	}
2362 
2363 	/*
2364 	 * This algorithm is recursive because the amount of used stack
2365 	 * space is very small and the max recursion depth is limited.
2366 	 */
2367 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2368 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2369 			pr_info("rl=%d, %c @%llu (%s/%llu/%d) %u* refers to %c @%llu (%s/%llu/%d)\n",
2370 			       recursion_level,
2371 			       btrfsic_get_block_type(state, block),
2372 			       block->logical_bytenr, block->dev_state->name,
2373 			       block->dev_bytenr, block->mirror_num,
2374 			       l->ref_cnt,
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 		if (l->block_ref_to->never_written) {
2381 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is never written!\n",
2382 			       btrfsic_get_block_type(state, l->block_ref_to),
2383 			       l->block_ref_to->logical_bytenr,
2384 			       l->block_ref_to->dev_state->name,
2385 			       l->block_ref_to->dev_bytenr,
2386 			       l->block_ref_to->mirror_num);
2387 			ret = -1;
2388 		} else if (!l->block_ref_to->is_iodone) {
2389 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is not yet iodone!\n",
2390 			       btrfsic_get_block_type(state, l->block_ref_to),
2391 			       l->block_ref_to->logical_bytenr,
2392 			       l->block_ref_to->dev_state->name,
2393 			       l->block_ref_to->dev_bytenr,
2394 			       l->block_ref_to->mirror_num);
2395 			ret = -1;
2396 		} else if (l->block_ref_to->iodone_w_error) {
2397 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which has write error!\n",
2398 			       btrfsic_get_block_type(state, l->block_ref_to),
2399 			       l->block_ref_to->logical_bytenr,
2400 			       l->block_ref_to->dev_state->name,
2401 			       l->block_ref_to->dev_bytenr,
2402 			       l->block_ref_to->mirror_num);
2403 			ret = -1;
2404 		} else if (l->parent_generation !=
2405 			   l->block_ref_to->generation &&
2406 			   BTRFSIC_GENERATION_UNKNOWN !=
2407 			   l->parent_generation &&
2408 			   BTRFSIC_GENERATION_UNKNOWN !=
2409 			   l->block_ref_to->generation) {
2410 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) with generation %llu != parent generation %llu!\n",
2411 			       btrfsic_get_block_type(state, l->block_ref_to),
2412 			       l->block_ref_to->logical_bytenr,
2413 			       l->block_ref_to->dev_state->name,
2414 			       l->block_ref_to->dev_bytenr,
2415 			       l->block_ref_to->mirror_num,
2416 			       l->block_ref_to->generation,
2417 			       l->parent_generation);
2418 			ret = -1;
2419 		} else if (l->block_ref_to->flush_gen >
2420 			   l->block_ref_to->dev_state->last_flush_gen) {
2421 			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",
2422 			       btrfsic_get_block_type(state, l->block_ref_to),
2423 			       l->block_ref_to->logical_bytenr,
2424 			       l->block_ref_to->dev_state->name,
2425 			       l->block_ref_to->dev_bytenr,
2426 			       l->block_ref_to->mirror_num, block->flush_gen,
2427 			       l->block_ref_to->dev_state->last_flush_gen);
2428 			ret = -1;
2429 		} else if (-1 == btrfsic_check_all_ref_blocks(state,
2430 							      l->block_ref_to,
2431 							      recursion_level +
2432 							      1)) {
2433 			ret = -1;
2434 		}
2435 	}
2436 
2437 	return ret;
2438 }
2439 
2440 static int btrfsic_is_block_ref_by_superblock(
2441 		const struct btrfsic_state *state,
2442 		const struct btrfsic_block *block,
2443 		int recursion_level)
2444 {
2445 	const struct btrfsic_block_link *l;
2446 
2447 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2448 		/* refer to comment at "abort cyclic linkage (case 1)" */
2449 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2450 			pr_info("btrfsic: abort cyclic linkage (case 2).\n");
2451 
2452 		return 0;
2453 	}
2454 
2455 	/*
2456 	 * This algorithm is recursive because the amount of used stack space
2457 	 * is very small and the max recursion depth is limited.
2458 	 */
2459 	list_for_each_entry(l, &block->ref_from_list, node_ref_from) {
2460 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2461 			pr_info("rl=%d, %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
2462 			       recursion_level,
2463 			       btrfsic_get_block_type(state, block),
2464 			       block->logical_bytenr, block->dev_state->name,
2465 			       block->dev_bytenr, block->mirror_num,
2466 			       l->ref_cnt,
2467 			       btrfsic_get_block_type(state, l->block_ref_from),
2468 			       l->block_ref_from->logical_bytenr,
2469 			       l->block_ref_from->dev_state->name,
2470 			       l->block_ref_from->dev_bytenr,
2471 			       l->block_ref_from->mirror_num);
2472 		if (l->block_ref_from->is_superblock &&
2473 		    state->latest_superblock->dev_bytenr ==
2474 		    l->block_ref_from->dev_bytenr &&
2475 		    state->latest_superblock->dev_state->bdev ==
2476 		    l->block_ref_from->dev_state->bdev)
2477 			return 1;
2478 		else if (btrfsic_is_block_ref_by_superblock(state,
2479 							    l->block_ref_from,
2480 							    recursion_level +
2481 							    1))
2482 			return 1;
2483 	}
2484 
2485 	return 0;
2486 }
2487 
2488 static void btrfsic_print_add_link(const struct btrfsic_state *state,
2489 				   const struct btrfsic_block_link *l)
2490 {
2491 	pr_info("Add %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 void btrfsic_print_rem_link(const struct btrfsic_state *state,
2504 				   const struct btrfsic_block_link *l)
2505 {
2506 	pr_info("Rem %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2507 	       l->ref_cnt,
2508 	       btrfsic_get_block_type(state, l->block_ref_from),
2509 	       l->block_ref_from->logical_bytenr,
2510 	       l->block_ref_from->dev_state->name,
2511 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2512 	       btrfsic_get_block_type(state, l->block_ref_to),
2513 	       l->block_ref_to->logical_bytenr,
2514 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2515 	       l->block_ref_to->mirror_num);
2516 }
2517 
2518 static char btrfsic_get_block_type(const struct btrfsic_state *state,
2519 				   const struct btrfsic_block *block)
2520 {
2521 	if (block->is_superblock &&
2522 	    state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2523 	    state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2524 		return 'S';
2525 	else if (block->is_superblock)
2526 		return 's';
2527 	else if (block->is_metadata)
2528 		return 'M';
2529 	else
2530 		return 'D';
2531 }
2532 
2533 static void btrfsic_dump_tree(const struct btrfsic_state *state)
2534 {
2535 	btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2536 }
2537 
2538 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2539 				  const struct btrfsic_block *block,
2540 				  int indent_level)
2541 {
2542 	const struct btrfsic_block_link *l;
2543 	int indent_add;
2544 	static char buf[80];
2545 	int cursor_position;
2546 
2547 	/*
2548 	 * Should better fill an on-stack buffer with a complete line and
2549 	 * dump it at once when it is time to print a newline character.
2550 	 */
2551 
2552 	/*
2553 	 * This algorithm is recursive because the amount of used stack space
2554 	 * is very small and the max recursion depth is limited.
2555 	 */
2556 	indent_add = sprintf(buf, "%c-%llu(%s/%llu/%u)",
2557 			     btrfsic_get_block_type(state, block),
2558 			     block->logical_bytenr, block->dev_state->name,
2559 			     block->dev_bytenr, block->mirror_num);
2560 	if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2561 		printk("[...]\n");
2562 		return;
2563 	}
2564 	printk(buf);
2565 	indent_level += indent_add;
2566 	if (list_empty(&block->ref_to_list)) {
2567 		printk("\n");
2568 		return;
2569 	}
2570 	if (block->mirror_num > 1 &&
2571 	    !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2572 		printk(" [...]\n");
2573 		return;
2574 	}
2575 
2576 	cursor_position = indent_level;
2577 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2578 		while (cursor_position < indent_level) {
2579 			printk(" ");
2580 			cursor_position++;
2581 		}
2582 		if (l->ref_cnt > 1)
2583 			indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2584 		else
2585 			indent_add = sprintf(buf, " --> ");
2586 		if (indent_level + indent_add >
2587 		    BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2588 			printk("[...]\n");
2589 			cursor_position = 0;
2590 			continue;
2591 		}
2592 
2593 		printk(buf);
2594 
2595 		btrfsic_dump_tree_sub(state, l->block_ref_to,
2596 				      indent_level + indent_add);
2597 		cursor_position = 0;
2598 	}
2599 }
2600 
2601 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2602 		struct btrfsic_state *state,
2603 		struct btrfsic_block_data_ctx *next_block_ctx,
2604 		struct btrfsic_block *next_block,
2605 		struct btrfsic_block *from_block,
2606 		u64 parent_generation)
2607 {
2608 	struct btrfsic_block_link *l;
2609 
2610 	l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2611 						next_block_ctx->dev_bytenr,
2612 						from_block->dev_state->bdev,
2613 						from_block->dev_bytenr,
2614 						&state->block_link_hashtable);
2615 	if (NULL == l) {
2616 		l = btrfsic_block_link_alloc();
2617 		if (NULL == l) {
2618 			pr_info("btrfsic: error, kmalloc failed!\n");
2619 			return NULL;
2620 		}
2621 
2622 		l->block_ref_to = next_block;
2623 		l->block_ref_from = from_block;
2624 		l->ref_cnt = 1;
2625 		l->parent_generation = parent_generation;
2626 
2627 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2628 			btrfsic_print_add_link(state, l);
2629 
2630 		list_add(&l->node_ref_to, &from_block->ref_to_list);
2631 		list_add(&l->node_ref_from, &next_block->ref_from_list);
2632 
2633 		btrfsic_block_link_hashtable_add(l,
2634 						 &state->block_link_hashtable);
2635 	} else {
2636 		l->ref_cnt++;
2637 		l->parent_generation = parent_generation;
2638 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2639 			btrfsic_print_add_link(state, l);
2640 	}
2641 
2642 	return l;
2643 }
2644 
2645 static struct btrfsic_block *btrfsic_block_lookup_or_add(
2646 		struct btrfsic_state *state,
2647 		struct btrfsic_block_data_ctx *block_ctx,
2648 		const char *additional_string,
2649 		int is_metadata,
2650 		int is_iodone,
2651 		int never_written,
2652 		int mirror_num,
2653 		int *was_created)
2654 {
2655 	struct btrfsic_block *block;
2656 
2657 	block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2658 					       block_ctx->dev_bytenr,
2659 					       &state->block_hashtable);
2660 	if (NULL == block) {
2661 		struct btrfsic_dev_state *dev_state;
2662 
2663 		block = btrfsic_block_alloc();
2664 		if (NULL == block) {
2665 			pr_info("btrfsic: error, kmalloc failed!\n");
2666 			return NULL;
2667 		}
2668 		dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev);
2669 		if (NULL == dev_state) {
2670 			pr_info("btrfsic: error, lookup dev_state failed!\n");
2671 			btrfsic_block_free(block);
2672 			return NULL;
2673 		}
2674 		block->dev_state = dev_state;
2675 		block->dev_bytenr = block_ctx->dev_bytenr;
2676 		block->logical_bytenr = block_ctx->start;
2677 		block->is_metadata = is_metadata;
2678 		block->is_iodone = is_iodone;
2679 		block->never_written = never_written;
2680 		block->mirror_num = mirror_num;
2681 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2682 			pr_info("New %s%c-block @%llu (%s/%llu/%d)\n",
2683 			       additional_string,
2684 			       btrfsic_get_block_type(state, block),
2685 			       block->logical_bytenr, dev_state->name,
2686 			       block->dev_bytenr, mirror_num);
2687 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2688 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2689 		if (NULL != was_created)
2690 			*was_created = 1;
2691 	} else {
2692 		if (NULL != was_created)
2693 			*was_created = 0;
2694 	}
2695 
2696 	return block;
2697 }
2698 
2699 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2700 					   u64 bytenr,
2701 					   struct btrfsic_dev_state *dev_state,
2702 					   u64 dev_bytenr)
2703 {
2704 	struct btrfs_fs_info *fs_info = state->fs_info;
2705 	struct btrfsic_block_data_ctx block_ctx;
2706 	int num_copies;
2707 	int mirror_num;
2708 	int match = 0;
2709 	int ret;
2710 
2711 	num_copies = btrfs_num_copies(fs_info, bytenr, state->metablock_size);
2712 
2713 	for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2714 		ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2715 					&block_ctx, mirror_num);
2716 		if (ret) {
2717 			pr_info("btrfsic: btrfsic_map_block(logical @%llu, mirror %d) failed!\n",
2718 			       bytenr, mirror_num);
2719 			continue;
2720 		}
2721 
2722 		if (dev_state->bdev == block_ctx.dev->bdev &&
2723 		    dev_bytenr == block_ctx.dev_bytenr) {
2724 			match++;
2725 			btrfsic_release_block_ctx(&block_ctx);
2726 			break;
2727 		}
2728 		btrfsic_release_block_ctx(&block_ctx);
2729 	}
2730 
2731 	if (WARN_ON(!match)) {
2732 		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",
2733 		       bytenr, dev_state->name, dev_bytenr);
2734 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2735 			ret = btrfsic_map_block(state, bytenr,
2736 						state->metablock_size,
2737 						&block_ctx, mirror_num);
2738 			if (ret)
2739 				continue;
2740 
2741 			pr_info("Read logical bytenr @%llu maps to (%s/%llu/%d)\n",
2742 			       bytenr, block_ctx.dev->name,
2743 			       block_ctx.dev_bytenr, mirror_num);
2744 		}
2745 	}
2746 }
2747 
2748 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(
2749 		struct block_device *bdev)
2750 {
2751 	return btrfsic_dev_state_hashtable_lookup(bdev,
2752 						  &btrfsic_dev_state_hashtable);
2753 }
2754 
2755 int btrfsic_submit_bh(int op, int op_flags, struct buffer_head *bh)
2756 {
2757 	struct btrfsic_dev_state *dev_state;
2758 
2759 	if (!btrfsic_is_initialized)
2760 		return submit_bh(op, op_flags, bh);
2761 
2762 	mutex_lock(&btrfsic_mutex);
2763 	/* since btrfsic_submit_bh() might also be called before
2764 	 * btrfsic_mount(), this might return NULL */
2765 	dev_state = btrfsic_dev_state_lookup(bh->b_bdev);
2766 
2767 	/* Only called to write the superblock (incl. FLUSH/FUA) */
2768 	if (NULL != dev_state &&
2769 	    (op == REQ_OP_WRITE) && bh->b_size > 0) {
2770 		u64 dev_bytenr;
2771 
2772 		dev_bytenr = 4096 * bh->b_blocknr;
2773 		if (dev_state->state->print_mask &
2774 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2775 			pr_info("submit_bh(op=0x%x,0x%x, blocknr=%llu (bytenr %llu), size=%zu, data=%p, bdev=%p)\n",
2776 			       op, op_flags, (unsigned long long)bh->b_blocknr,
2777 			       dev_bytenr, bh->b_size, bh->b_data, bh->b_bdev);
2778 		btrfsic_process_written_block(dev_state, dev_bytenr,
2779 					      &bh->b_data, 1, NULL,
2780 					      NULL, bh, op_flags);
2781 	} else if (NULL != dev_state && (op_flags & REQ_PREFLUSH)) {
2782 		if (dev_state->state->print_mask &
2783 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2784 			pr_info("submit_bh(op=0x%x,0x%x FLUSH, bdev=%p)\n",
2785 			       op, op_flags, bh->b_bdev);
2786 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2787 			if ((dev_state->state->print_mask &
2788 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2789 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2790 				pr_info("btrfsic_submit_bh(%s) with FLUSH but dummy block already in use (ignored)!\n",
2791 				       dev_state->name);
2792 		} else {
2793 			struct btrfsic_block *const block =
2794 				&dev_state->dummy_block_for_bio_bh_flush;
2795 
2796 			block->is_iodone = 0;
2797 			block->never_written = 0;
2798 			block->iodone_w_error = 0;
2799 			block->flush_gen = dev_state->last_flush_gen + 1;
2800 			block->submit_bio_bh_rw = op_flags;
2801 			block->orig_bio_bh_private = bh->b_private;
2802 			block->orig_bio_bh_end_io.bh = bh->b_end_io;
2803 			block->next_in_same_bio = NULL;
2804 			bh->b_private = block;
2805 			bh->b_end_io = btrfsic_bh_end_io;
2806 		}
2807 	}
2808 	mutex_unlock(&btrfsic_mutex);
2809 	return submit_bh(op, op_flags, bh);
2810 }
2811 
2812 static void __btrfsic_submit_bio(struct bio *bio)
2813 {
2814 	struct btrfsic_dev_state *dev_state;
2815 
2816 	if (!btrfsic_is_initialized)
2817 		return;
2818 
2819 	mutex_lock(&btrfsic_mutex);
2820 	/* since btrfsic_submit_bio() is also called before
2821 	 * btrfsic_mount(), this might return NULL */
2822 	dev_state = btrfsic_dev_state_lookup(bio->bi_bdev);
2823 	if (NULL != dev_state &&
2824 	    (bio_op(bio) == REQ_OP_WRITE) && bio_has_data(bio)) {
2825 		unsigned int i;
2826 		u64 dev_bytenr;
2827 		u64 cur_bytenr;
2828 		struct bio_vec *bvec;
2829 		int bio_is_patched;
2830 		char **mapped_datav;
2831 
2832 		dev_bytenr = 512 * bio->bi_iter.bi_sector;
2833 		bio_is_patched = 0;
2834 		if (dev_state->state->print_mask &
2835 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2836 			pr_info("submit_bio(rw=%d,0x%x, bi_vcnt=%u, bi_sector=%llu (bytenr %llu), bi_bdev=%p)\n",
2837 			       bio_op(bio), bio->bi_opf, bio->bi_vcnt,
2838 			       (unsigned long long)bio->bi_iter.bi_sector,
2839 			       dev_bytenr, bio->bi_bdev);
2840 
2841 		mapped_datav = kmalloc_array(bio->bi_vcnt,
2842 					     sizeof(*mapped_datav), GFP_NOFS);
2843 		if (!mapped_datav)
2844 			goto leave;
2845 		cur_bytenr = dev_bytenr;
2846 
2847 		bio_for_each_segment_all(bvec, bio, i) {
2848 			BUG_ON(bvec->bv_len != PAGE_SIZE);
2849 			mapped_datav[i] = kmap(bvec->bv_page);
2850 
2851 			if (dev_state->state->print_mask &
2852 			    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
2853 				pr_info("#%u: bytenr=%llu, len=%u, offset=%u\n",
2854 				       i, cur_bytenr, bvec->bv_len, bvec->bv_offset);
2855 			cur_bytenr += bvec->bv_len;
2856 		}
2857 		btrfsic_process_written_block(dev_state, dev_bytenr,
2858 					      mapped_datav, bio->bi_vcnt,
2859 					      bio, &bio_is_patched,
2860 					      NULL, bio->bi_opf);
2861 		bio_for_each_segment_all(bvec, bio, i)
2862 			kunmap(bvec->bv_page);
2863 		kfree(mapped_datav);
2864 	} else if (NULL != dev_state && (bio->bi_opf & REQ_PREFLUSH)) {
2865 		if (dev_state->state->print_mask &
2866 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2867 			pr_info("submit_bio(rw=%d,0x%x FLUSH, bdev=%p)\n",
2868 			       bio_op(bio), bio->bi_opf, bio->bi_bdev);
2869 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2870 			if ((dev_state->state->print_mask &
2871 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2872 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2873 				pr_info("btrfsic_submit_bio(%s) with FLUSH but dummy block already in use (ignored)!\n",
2874 				       dev_state->name);
2875 		} else {
2876 			struct btrfsic_block *const block =
2877 				&dev_state->dummy_block_for_bio_bh_flush;
2878 
2879 			block->is_iodone = 0;
2880 			block->never_written = 0;
2881 			block->iodone_w_error = 0;
2882 			block->flush_gen = dev_state->last_flush_gen + 1;
2883 			block->submit_bio_bh_rw = bio->bi_opf;
2884 			block->orig_bio_bh_private = bio->bi_private;
2885 			block->orig_bio_bh_end_io.bio = bio->bi_end_io;
2886 			block->next_in_same_bio = NULL;
2887 			bio->bi_private = block;
2888 			bio->bi_end_io = btrfsic_bio_end_io;
2889 		}
2890 	}
2891 leave:
2892 	mutex_unlock(&btrfsic_mutex);
2893 }
2894 
2895 void btrfsic_submit_bio(struct bio *bio)
2896 {
2897 	__btrfsic_submit_bio(bio);
2898 	submit_bio(bio);
2899 }
2900 
2901 int btrfsic_submit_bio_wait(struct bio *bio)
2902 {
2903 	__btrfsic_submit_bio(bio);
2904 	return submit_bio_wait(bio);
2905 }
2906 
2907 int btrfsic_mount(struct btrfs_fs_info *fs_info,
2908 		  struct btrfs_fs_devices *fs_devices,
2909 		  int including_extent_data, u32 print_mask)
2910 {
2911 	int ret;
2912 	struct btrfsic_state *state;
2913 	struct list_head *dev_head = &fs_devices->devices;
2914 	struct btrfs_device *device;
2915 
2916 	if (fs_info->nodesize & ((u64)PAGE_SIZE - 1)) {
2917 		pr_info("btrfsic: cannot handle nodesize %d not being a multiple of PAGE_SIZE %ld!\n",
2918 		       fs_info->nodesize, PAGE_SIZE);
2919 		return -1;
2920 	}
2921 	if (fs_info->sectorsize & ((u64)PAGE_SIZE - 1)) {
2922 		pr_info("btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_SIZE %ld!\n",
2923 		       fs_info->sectorsize, PAGE_SIZE);
2924 		return -1;
2925 	}
2926 	state = kzalloc(sizeof(*state), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
2927 	if (!state) {
2928 		state = vzalloc(sizeof(*state));
2929 		if (!state) {
2930 			pr_info("btrfs check-integrity: vzalloc() failed!\n");
2931 			return -1;
2932 		}
2933 	}
2934 
2935 	if (!btrfsic_is_initialized) {
2936 		mutex_init(&btrfsic_mutex);
2937 		btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
2938 		btrfsic_is_initialized = 1;
2939 	}
2940 	mutex_lock(&btrfsic_mutex);
2941 	state->fs_info = fs_info;
2942 	state->print_mask = print_mask;
2943 	state->include_extent_data = including_extent_data;
2944 	state->csum_size = 0;
2945 	state->metablock_size = fs_info->nodesize;
2946 	state->datablock_size = fs_info->sectorsize;
2947 	INIT_LIST_HEAD(&state->all_blocks_list);
2948 	btrfsic_block_hashtable_init(&state->block_hashtable);
2949 	btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
2950 	state->max_superblock_generation = 0;
2951 	state->latest_superblock = NULL;
2952 
2953 	list_for_each_entry(device, dev_head, dev_list) {
2954 		struct btrfsic_dev_state *ds;
2955 		const char *p;
2956 
2957 		if (!device->bdev || !device->name)
2958 			continue;
2959 
2960 		ds = btrfsic_dev_state_alloc();
2961 		if (NULL == ds) {
2962 			pr_info("btrfs check-integrity: kmalloc() failed!\n");
2963 			mutex_unlock(&btrfsic_mutex);
2964 			return -1;
2965 		}
2966 		ds->bdev = device->bdev;
2967 		ds->state = state;
2968 		bdevname(ds->bdev, ds->name);
2969 		ds->name[BDEVNAME_SIZE - 1] = '\0';
2970 		p = kbasename(ds->name);
2971 		strlcpy(ds->name, p, sizeof(ds->name));
2972 		btrfsic_dev_state_hashtable_add(ds,
2973 						&btrfsic_dev_state_hashtable);
2974 	}
2975 
2976 	ret = btrfsic_process_superblock(state, fs_devices);
2977 	if (0 != ret) {
2978 		mutex_unlock(&btrfsic_mutex);
2979 		btrfsic_unmount(fs_devices);
2980 		return ret;
2981 	}
2982 
2983 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
2984 		btrfsic_dump_database(state);
2985 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
2986 		btrfsic_dump_tree(state);
2987 
2988 	mutex_unlock(&btrfsic_mutex);
2989 	return 0;
2990 }
2991 
2992 void btrfsic_unmount(struct btrfs_fs_devices *fs_devices)
2993 {
2994 	struct btrfsic_block *b_all, *tmp_all;
2995 	struct btrfsic_state *state;
2996 	struct list_head *dev_head = &fs_devices->devices;
2997 	struct btrfs_device *device;
2998 
2999 	if (!btrfsic_is_initialized)
3000 		return;
3001 
3002 	mutex_lock(&btrfsic_mutex);
3003 
3004 	state = NULL;
3005 	list_for_each_entry(device, dev_head, dev_list) {
3006 		struct btrfsic_dev_state *ds;
3007 
3008 		if (!device->bdev || !device->name)
3009 			continue;
3010 
3011 		ds = btrfsic_dev_state_hashtable_lookup(
3012 				device->bdev,
3013 				&btrfsic_dev_state_hashtable);
3014 		if (NULL != ds) {
3015 			state = ds->state;
3016 			btrfsic_dev_state_hashtable_remove(ds);
3017 			btrfsic_dev_state_free(ds);
3018 		}
3019 	}
3020 
3021 	if (NULL == state) {
3022 		pr_info("btrfsic: error, cannot find state information on umount!\n");
3023 		mutex_unlock(&btrfsic_mutex);
3024 		return;
3025 	}
3026 
3027 	/*
3028 	 * Don't care about keeping the lists' state up to date,
3029 	 * just free all memory that was allocated dynamically.
3030 	 * Free the blocks and the block_links.
3031 	 */
3032 	list_for_each_entry_safe(b_all, tmp_all, &state->all_blocks_list,
3033 				 all_blocks_node) {
3034 		struct btrfsic_block_link *l, *tmp;
3035 
3036 		list_for_each_entry_safe(l, tmp, &b_all->ref_to_list,
3037 					 node_ref_to) {
3038 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
3039 				btrfsic_print_rem_link(state, l);
3040 
3041 			l->ref_cnt--;
3042 			if (0 == l->ref_cnt)
3043 				btrfsic_block_link_free(l);
3044 		}
3045 
3046 		if (b_all->is_iodone || b_all->never_written)
3047 			btrfsic_block_free(b_all);
3048 		else
3049 			pr_info("btrfs: attempt to free %c-block @%llu (%s/%llu/%d) on umount which is not yet iodone!\n",
3050 			       btrfsic_get_block_type(state, b_all),
3051 			       b_all->logical_bytenr, b_all->dev_state->name,
3052 			       b_all->dev_bytenr, b_all->mirror_num);
3053 	}
3054 
3055 	mutex_unlock(&btrfsic_mutex);
3056 
3057 	kvfree(state);
3058 }
3059