• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) STRATO AG 2011.  All rights reserved.
4  */
5 
6 /*
7  * This module can be used to catch cases when the btrfs kernel
8  * code executes write requests to the disk that bring the file
9  * system in an inconsistent state. In such a state, a power-loss
10  * or kernel panic event would cause that the data on disk is
11  * lost or at least damaged.
12  *
13  * Code is added that examines all block write requests during
14  * runtime (including writes of the super block). Three rules
15  * are verified and an error is printed on violation of the
16  * rules:
17  * 1. It is not allowed to write a disk block which is
18  *    currently referenced by the super block (either directly
19  *    or indirectly).
20  * 2. When a super block is written, it is verified that all
21  *    referenced (directly or indirectly) blocks fulfill the
22  *    following requirements:
23  *    2a. All referenced blocks have either been present when
24  *        the file system was mounted, (i.e., they have been
25  *        referenced by the super block) or they have been
26  *        written since then and the write completion callback
27  *        was called and no write error was indicated and a
28  *        FLUSH request to the device where these blocks are
29  *        located was received and completed.
30  *    2b. All referenced blocks need to have a generation
31  *        number which is equal to the parent's number.
32  *
33  * One issue that was found using this module was that the log
34  * tree on disk became temporarily corrupted because disk blocks
35  * that had been in use for the log tree had been freed and
36  * reused too early, while being referenced by the written super
37  * block.
38  *
39  * The search term in the kernel log that can be used to filter
40  * on the existence of detected integrity issues is
41  * "btrfs: attempt".
42  *
43  * The integrity check is enabled via mount options. These
44  * mount options are only supported if the integrity check
45  * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY.
46  *
47  * Example #1, apply integrity checks to all metadata:
48  * mount /dev/sdb1 /mnt -o check_int
49  *
50  * Example #2, apply integrity checks to all metadata and
51  * to data extents:
52  * mount /dev/sdb1 /mnt -o check_int_data
53  *
54  * Example #3, apply integrity checks to all metadata and dump
55  * the tree that the super block references to kernel messages
56  * each time after a super block was written:
57  * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263
58  *
59  * If the integrity check tool is included and activated in
60  * the mount options, plenty of kernel memory is used, and
61  * plenty of additional CPU cycles are spent. Enabling this
62  * functionality is not intended for normal use. In most
63  * cases, unless you are a btrfs developer who needs to verify
64  * the integrity of (super)-block write requests, do not
65  * enable the config option BTRFS_FS_CHECK_INTEGRITY to
66  * include and compile the integrity check tool.
67  *
68  * Expect millions of lines of information in the kernel log with an
69  * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the
70  * kernel config to at least 26 (which is 64MB). Usually the value is
71  * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be
72  * changed like this before LOG_BUF_SHIFT can be set to a high value:
73  * config LOG_BUF_SHIFT
74  *       int "Kernel log buffer size (16 => 64KB, 17 => 128KB)"
75  *       range 12 30
76  */
77 
78 #include <linux/sched.h>
79 #include <linux/slab.h>
80 #include <linux/mutex.h>
81 #include <linux/genhd.h>
82 #include <linux/blkdev.h>
83 #include <linux/mm.h>
84 #include <linux/string.h>
85 #include <crypto/hash.h>
86 #include "ctree.h"
87 #include "disk-io.h"
88 #include "transaction.h"
89 #include "extent_io.h"
90 #include "volumes.h"
91 #include "print-tree.h"
92 #include "locking.h"
93 #include "check-integrity.h"
94 #include "rcu-string.h"
95 #include "compression.h"
96 
97 #define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
98 #define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
99 #define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
100 #define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
101 #define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
102 #define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
103 #define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
104 #define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6)	/* in characters,
105 							 * excluding " [...]" */
106 #define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
107 
108 /*
109  * The definition of the bitmask fields for the print_mask.
110  * They are specified with the mount option check_integrity_print_mask.
111  */
112 #define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE			0x00000001
113 #define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION		0x00000002
114 #define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE			0x00000004
115 #define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE			0x00000008
116 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH			0x00000010
117 #define BTRFSIC_PRINT_MASK_END_IO_BIO_BH			0x00000020
118 #define BTRFSIC_PRINT_MASK_VERBOSE				0x00000040
119 #define BTRFSIC_PRINT_MASK_VERY_VERBOSE				0x00000080
120 #define BTRFSIC_PRINT_MASK_INITIAL_TREE				0x00000100
121 #define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES			0x00000200
122 #define BTRFSIC_PRINT_MASK_INITIAL_DATABASE			0x00000400
123 #define BTRFSIC_PRINT_MASK_NUM_COPIES				0x00000800
124 #define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS		0x00001000
125 #define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE		0x00002000
126 
127 struct btrfsic_dev_state;
128 struct btrfsic_state;
129 
130 struct btrfsic_block {
131 	u32 magic_num;		/* only used for debug purposes */
132 	unsigned int is_metadata:1;	/* if it is meta-data, not data-data */
133 	unsigned int is_superblock:1;	/* if it is one of the superblocks */
134 	unsigned int is_iodone:1;	/* if is done by lower subsystem */
135 	unsigned int iodone_w_error:1;	/* error was indicated to endio */
136 	unsigned int never_written:1;	/* block was added because it was
137 					 * referenced, not because it was
138 					 * written */
139 	unsigned int mirror_num;	/* large enough to hold
140 					 * BTRFS_SUPER_MIRROR_MAX */
141 	struct btrfsic_dev_state *dev_state;
142 	u64 dev_bytenr;		/* key, physical byte num on disk */
143 	u64 logical_bytenr;	/* logical byte num on disk */
144 	u64 generation;
145 	struct btrfs_disk_key disk_key;	/* extra info to print in case of
146 					 * issues, will not always be correct */
147 	struct list_head collision_resolving_node;	/* list node */
148 	struct list_head all_blocks_node;	/* list node */
149 
150 	/* the following two lists contain block_link items */
151 	struct list_head ref_to_list;	/* list */
152 	struct list_head ref_from_list;	/* list */
153 	struct btrfsic_block *next_in_same_bio;
154 	void *orig_bio_private;
155 	bio_end_io_t *orig_bio_end_io;
156 	int submit_bio_bh_rw;
157 	u64 flush_gen; /* only valid if !never_written */
158 };
159 
160 /*
161  * Elements of this type are allocated dynamically and required because
162  * each block object can refer to and can be ref from multiple blocks.
163  * The key to lookup them in the hashtable is the dev_bytenr of
164  * the block ref to plus the one from the block referred from.
165  * The fact that they are searchable via a hashtable and that a
166  * ref_cnt is maintained is not required for the btrfs integrity
167  * check algorithm itself, it is only used to make the output more
168  * beautiful in case that an error is detected (an error is defined
169  * as a write operation to a block while that block is still referenced).
170  */
171 struct btrfsic_block_link {
172 	u32 magic_num;		/* only used for debug purposes */
173 	u32 ref_cnt;
174 	struct list_head node_ref_to;	/* list node */
175 	struct list_head node_ref_from;	/* list node */
176 	struct list_head collision_resolving_node;	/* list node */
177 	struct btrfsic_block *block_ref_to;
178 	struct btrfsic_block *block_ref_from;
179 	u64 parent_generation;
180 };
181 
182 struct btrfsic_dev_state {
183 	u32 magic_num;		/* only used for debug purposes */
184 	struct block_device *bdev;
185 	struct btrfsic_state *state;
186 	struct list_head collision_resolving_node;	/* list node */
187 	struct btrfsic_block dummy_block_for_bio_bh_flush;
188 	u64 last_flush_gen;
189 	char name[BDEVNAME_SIZE];
190 };
191 
192 struct btrfsic_block_hashtable {
193 	struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
194 };
195 
196 struct btrfsic_block_link_hashtable {
197 	struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
198 };
199 
200 struct btrfsic_dev_state_hashtable {
201 	struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
202 };
203 
204 struct btrfsic_block_data_ctx {
205 	u64 start;		/* virtual bytenr */
206 	u64 dev_bytenr;		/* physical bytenr on device */
207 	u32 len;
208 	struct btrfsic_dev_state *dev;
209 	char **datav;
210 	struct page **pagev;
211 	void *mem_to_free;
212 };
213 
214 /* This structure is used to implement recursion without occupying
215  * any stack space, refer to btrfsic_process_metablock() */
216 struct btrfsic_stack_frame {
217 	u32 magic;
218 	u32 nr;
219 	int error;
220 	int i;
221 	int limit_nesting;
222 	int num_copies;
223 	int mirror_num;
224 	struct btrfsic_block *block;
225 	struct btrfsic_block_data_ctx *block_ctx;
226 	struct btrfsic_block *next_block;
227 	struct btrfsic_block_data_ctx next_block_ctx;
228 	struct btrfs_header *hdr;
229 	struct btrfsic_stack_frame *prev;
230 };
231 
232 /* Some state per mounted filesystem */
233 struct btrfsic_state {
234 	u32 print_mask;
235 	int include_extent_data;
236 	struct list_head all_blocks_list;
237 	struct btrfsic_block_hashtable block_hashtable;
238 	struct btrfsic_block_link_hashtable block_link_hashtable;
239 	struct btrfs_fs_info *fs_info;
240 	u64 max_superblock_generation;
241 	struct btrfsic_block *latest_superblock;
242 	u32 metablock_size;
243 	u32 datablock_size;
244 };
245 
246 static int btrfsic_process_metablock(struct btrfsic_state *state,
247 				     struct btrfsic_block *block,
248 				     struct btrfsic_block_data_ctx *block_ctx,
249 				     int limit_nesting, int force_iodone_flag);
250 static void btrfsic_read_from_block_data(
251 	struct btrfsic_block_data_ctx *block_ctx,
252 	void *dst, u32 offset, size_t len);
253 static int btrfsic_create_link_to_next_block(
254 		struct btrfsic_state *state,
255 		struct btrfsic_block *block,
256 		struct btrfsic_block_data_ctx
257 		*block_ctx, u64 next_bytenr,
258 		int limit_nesting,
259 		struct btrfsic_block_data_ctx *next_block_ctx,
260 		struct btrfsic_block **next_blockp,
261 		int force_iodone_flag,
262 		int *num_copiesp, int *mirror_nump,
263 		struct btrfs_disk_key *disk_key,
264 		u64 parent_generation);
265 static int btrfsic_handle_extent_data(struct btrfsic_state *state,
266 				      struct btrfsic_block *block,
267 				      struct btrfsic_block_data_ctx *block_ctx,
268 				      u32 item_offset, int force_iodone_flag);
269 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
270 			     struct btrfsic_block_data_ctx *block_ctx_out,
271 			     int mirror_num);
272 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
273 static int btrfsic_read_block(struct btrfsic_state *state,
274 			      struct btrfsic_block_data_ctx *block_ctx);
275 static int btrfsic_process_written_superblock(
276 		struct btrfsic_state *state,
277 		struct btrfsic_block *const block,
278 		struct btrfs_super_block *const super_hdr);
279 static void btrfsic_bio_end_io(struct bio *bp);
280 static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
281 					      const struct btrfsic_block *block,
282 					      int recursion_level);
283 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
284 					struct btrfsic_block *const block,
285 					int recursion_level);
286 static void btrfsic_print_add_link(const struct btrfsic_state *state,
287 				   const struct btrfsic_block_link *l);
288 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
289 				   const struct btrfsic_block_link *l);
290 static char btrfsic_get_block_type(const struct btrfsic_state *state,
291 				   const struct btrfsic_block *block);
292 static void btrfsic_dump_tree(const struct btrfsic_state *state);
293 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
294 				  const struct btrfsic_block *block,
295 				  int indent_level);
296 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
297 		struct btrfsic_state *state,
298 		struct btrfsic_block_data_ctx *next_block_ctx,
299 		struct btrfsic_block *next_block,
300 		struct btrfsic_block *from_block,
301 		u64 parent_generation);
302 static struct btrfsic_block *btrfsic_block_lookup_or_add(
303 		struct btrfsic_state *state,
304 		struct btrfsic_block_data_ctx *block_ctx,
305 		const char *additional_string,
306 		int is_metadata,
307 		int is_iodone,
308 		int never_written,
309 		int mirror_num,
310 		int *was_created);
311 static int btrfsic_process_superblock_dev_mirror(
312 		struct btrfsic_state *state,
313 		struct btrfsic_dev_state *dev_state,
314 		struct btrfs_device *device,
315 		int superblock_mirror_num,
316 		struct btrfsic_dev_state **selected_dev_state,
317 		struct btrfs_super_block *selected_super);
318 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev);
319 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
320 					   u64 bytenr,
321 					   struct btrfsic_dev_state *dev_state,
322 					   u64 dev_bytenr);
323 
324 static struct mutex btrfsic_mutex;
325 static int btrfsic_is_initialized;
326 static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
327 
328 
btrfsic_block_init(struct btrfsic_block * b)329 static void btrfsic_block_init(struct btrfsic_block *b)
330 {
331 	b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
332 	b->dev_state = NULL;
333 	b->dev_bytenr = 0;
334 	b->logical_bytenr = 0;
335 	b->generation = BTRFSIC_GENERATION_UNKNOWN;
336 	b->disk_key.objectid = 0;
337 	b->disk_key.type = 0;
338 	b->disk_key.offset = 0;
339 	b->is_metadata = 0;
340 	b->is_superblock = 0;
341 	b->is_iodone = 0;
342 	b->iodone_w_error = 0;
343 	b->never_written = 0;
344 	b->mirror_num = 0;
345 	b->next_in_same_bio = NULL;
346 	b->orig_bio_private = NULL;
347 	b->orig_bio_end_io = NULL;
348 	INIT_LIST_HEAD(&b->collision_resolving_node);
349 	INIT_LIST_HEAD(&b->all_blocks_node);
350 	INIT_LIST_HEAD(&b->ref_to_list);
351 	INIT_LIST_HEAD(&b->ref_from_list);
352 	b->submit_bio_bh_rw = 0;
353 	b->flush_gen = 0;
354 }
355 
btrfsic_block_alloc(void)356 static struct btrfsic_block *btrfsic_block_alloc(void)
357 {
358 	struct btrfsic_block *b;
359 
360 	b = kzalloc(sizeof(*b), GFP_NOFS);
361 	if (NULL != b)
362 		btrfsic_block_init(b);
363 
364 	return b;
365 }
366 
btrfsic_block_free(struct btrfsic_block * b)367 static void btrfsic_block_free(struct btrfsic_block *b)
368 {
369 	BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
370 	kfree(b);
371 }
372 
btrfsic_block_link_init(struct btrfsic_block_link * l)373 static void btrfsic_block_link_init(struct btrfsic_block_link *l)
374 {
375 	l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
376 	l->ref_cnt = 1;
377 	INIT_LIST_HEAD(&l->node_ref_to);
378 	INIT_LIST_HEAD(&l->node_ref_from);
379 	INIT_LIST_HEAD(&l->collision_resolving_node);
380 	l->block_ref_to = NULL;
381 	l->block_ref_from = NULL;
382 }
383 
btrfsic_block_link_alloc(void)384 static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
385 {
386 	struct btrfsic_block_link *l;
387 
388 	l = kzalloc(sizeof(*l), GFP_NOFS);
389 	if (NULL != l)
390 		btrfsic_block_link_init(l);
391 
392 	return l;
393 }
394 
btrfsic_block_link_free(struct btrfsic_block_link * l)395 static void btrfsic_block_link_free(struct btrfsic_block_link *l)
396 {
397 	BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
398 	kfree(l);
399 }
400 
btrfsic_dev_state_init(struct btrfsic_dev_state * ds)401 static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
402 {
403 	ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
404 	ds->bdev = NULL;
405 	ds->state = NULL;
406 	ds->name[0] = '\0';
407 	INIT_LIST_HEAD(&ds->collision_resolving_node);
408 	ds->last_flush_gen = 0;
409 	btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
410 	ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
411 	ds->dummy_block_for_bio_bh_flush.dev_state = ds;
412 }
413 
btrfsic_dev_state_alloc(void)414 static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
415 {
416 	struct btrfsic_dev_state *ds;
417 
418 	ds = kzalloc(sizeof(*ds), GFP_NOFS);
419 	if (NULL != ds)
420 		btrfsic_dev_state_init(ds);
421 
422 	return ds;
423 }
424 
btrfsic_dev_state_free(struct btrfsic_dev_state * ds)425 static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
426 {
427 	BUG_ON(!(NULL == ds ||
428 		 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
429 	kfree(ds);
430 }
431 
btrfsic_block_hashtable_init(struct btrfsic_block_hashtable * h)432 static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
433 {
434 	int i;
435 
436 	for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
437 		INIT_LIST_HEAD(h->table + i);
438 }
439 
btrfsic_block_hashtable_add(struct btrfsic_block * b,struct btrfsic_block_hashtable * h)440 static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
441 					struct btrfsic_block_hashtable *h)
442 {
443 	const unsigned int hashval =
444 	    (((unsigned int)(b->dev_bytenr >> 16)) ^
445 	     ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
446 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
447 
448 	list_add(&b->collision_resolving_node, h->table + hashval);
449 }
450 
btrfsic_block_hashtable_remove(struct btrfsic_block * b)451 static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
452 {
453 	list_del(&b->collision_resolving_node);
454 }
455 
btrfsic_block_hashtable_lookup(struct block_device * bdev,u64 dev_bytenr,struct btrfsic_block_hashtable * h)456 static struct btrfsic_block *btrfsic_block_hashtable_lookup(
457 		struct block_device *bdev,
458 		u64 dev_bytenr,
459 		struct btrfsic_block_hashtable *h)
460 {
461 	const unsigned int hashval =
462 	    (((unsigned int)(dev_bytenr >> 16)) ^
463 	     ((unsigned int)((uintptr_t)bdev))) &
464 	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
465 	struct btrfsic_block *b;
466 
467 	list_for_each_entry(b, h->table + hashval, collision_resolving_node) {
468 		if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
469 			return b;
470 	}
471 
472 	return NULL;
473 }
474 
btrfsic_block_link_hashtable_init(struct btrfsic_block_link_hashtable * h)475 static void btrfsic_block_link_hashtable_init(
476 		struct btrfsic_block_link_hashtable *h)
477 {
478 	int i;
479 
480 	for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
481 		INIT_LIST_HEAD(h->table + i);
482 }
483 
btrfsic_block_link_hashtable_add(struct btrfsic_block_link * l,struct btrfsic_block_link_hashtable * h)484 static void btrfsic_block_link_hashtable_add(
485 		struct btrfsic_block_link *l,
486 		struct btrfsic_block_link_hashtable *h)
487 {
488 	const unsigned int hashval =
489 	    (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
490 	     ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
491 	     ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
492 	     ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
493 	     & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
494 
495 	BUG_ON(NULL == l->block_ref_to);
496 	BUG_ON(NULL == l->block_ref_from);
497 	list_add(&l->collision_resolving_node, h->table + hashval);
498 }
499 
btrfsic_block_link_hashtable_remove(struct btrfsic_block_link * l)500 static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
501 {
502 	list_del(&l->collision_resolving_node);
503 }
504 
btrfsic_block_link_hashtable_lookup(struct block_device * bdev_ref_to,u64 dev_bytenr_ref_to,struct block_device * bdev_ref_from,u64 dev_bytenr_ref_from,struct btrfsic_block_link_hashtable * h)505 static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
506 		struct block_device *bdev_ref_to,
507 		u64 dev_bytenr_ref_to,
508 		struct block_device *bdev_ref_from,
509 		u64 dev_bytenr_ref_from,
510 		struct btrfsic_block_link_hashtable *h)
511 {
512 	const unsigned int hashval =
513 	    (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
514 	     ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
515 	     ((unsigned int)((uintptr_t)bdev_ref_to)) ^
516 	     ((unsigned int)((uintptr_t)bdev_ref_from))) &
517 	     (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
518 	struct btrfsic_block_link *l;
519 
520 	list_for_each_entry(l, h->table + hashval, collision_resolving_node) {
521 		BUG_ON(NULL == l->block_ref_to);
522 		BUG_ON(NULL == l->block_ref_from);
523 		if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
524 		    l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
525 		    l->block_ref_from->dev_state->bdev == bdev_ref_from &&
526 		    l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
527 			return l;
528 	}
529 
530 	return NULL;
531 }
532 
btrfsic_dev_state_hashtable_init(struct btrfsic_dev_state_hashtable * h)533 static void btrfsic_dev_state_hashtable_init(
534 		struct btrfsic_dev_state_hashtable *h)
535 {
536 	int i;
537 
538 	for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
539 		INIT_LIST_HEAD(h->table + i);
540 }
541 
btrfsic_dev_state_hashtable_add(struct btrfsic_dev_state * ds,struct btrfsic_dev_state_hashtable * h)542 static void btrfsic_dev_state_hashtable_add(
543 		struct btrfsic_dev_state *ds,
544 		struct btrfsic_dev_state_hashtable *h)
545 {
546 	const unsigned int hashval =
547 	    (((unsigned int)((uintptr_t)ds->bdev->bd_dev)) &
548 	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
549 
550 	list_add(&ds->collision_resolving_node, h->table + hashval);
551 }
552 
btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state * ds)553 static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
554 {
555 	list_del(&ds->collision_resolving_node);
556 }
557 
btrfsic_dev_state_hashtable_lookup(dev_t dev,struct btrfsic_dev_state_hashtable * h)558 static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev,
559 		struct btrfsic_dev_state_hashtable *h)
560 {
561 	const unsigned int hashval =
562 		dev & (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1);
563 	struct btrfsic_dev_state *ds;
564 
565 	list_for_each_entry(ds, h->table + hashval, collision_resolving_node) {
566 		if (ds->bdev->bd_dev == dev)
567 			return ds;
568 	}
569 
570 	return NULL;
571 }
572 
btrfsic_process_superblock(struct btrfsic_state * state,struct btrfs_fs_devices * fs_devices)573 static int btrfsic_process_superblock(struct btrfsic_state *state,
574 				      struct btrfs_fs_devices *fs_devices)
575 {
576 	struct btrfs_super_block *selected_super;
577 	struct list_head *dev_head = &fs_devices->devices;
578 	struct btrfs_device *device;
579 	struct btrfsic_dev_state *selected_dev_state = NULL;
580 	int ret = 0;
581 	int pass;
582 
583 	selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
584 	if (!selected_super)
585 		return -ENOMEM;
586 
587 	list_for_each_entry(device, dev_head, dev_list) {
588 		int i;
589 		struct btrfsic_dev_state *dev_state;
590 
591 		if (!device->bdev || !device->name)
592 			continue;
593 
594 		dev_state = btrfsic_dev_state_lookup(device->bdev->bd_dev);
595 		BUG_ON(NULL == dev_state);
596 		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
597 			ret = btrfsic_process_superblock_dev_mirror(
598 					state, dev_state, device, i,
599 					&selected_dev_state, selected_super);
600 			if (0 != ret && 0 == i) {
601 				kfree(selected_super);
602 				return ret;
603 			}
604 		}
605 	}
606 
607 	if (NULL == state->latest_superblock) {
608 		pr_info("btrfsic: no superblock found!\n");
609 		kfree(selected_super);
610 		return -1;
611 	}
612 
613 	for (pass = 0; pass < 3; pass++) {
614 		int num_copies;
615 		int mirror_num;
616 		u64 next_bytenr;
617 
618 		switch (pass) {
619 		case 0:
620 			next_bytenr = btrfs_super_root(selected_super);
621 			if (state->print_mask &
622 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
623 				pr_info("root@%llu\n", next_bytenr);
624 			break;
625 		case 1:
626 			next_bytenr = btrfs_super_chunk_root(selected_super);
627 			if (state->print_mask &
628 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
629 				pr_info("chunk@%llu\n", next_bytenr);
630 			break;
631 		case 2:
632 			next_bytenr = btrfs_super_log_root(selected_super);
633 			if (0 == next_bytenr)
634 				continue;
635 			if (state->print_mask &
636 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
637 				pr_info("log@%llu\n", next_bytenr);
638 			break;
639 		}
640 
641 		num_copies = btrfs_num_copies(state->fs_info, next_bytenr,
642 					      state->metablock_size);
643 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
644 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
645 			       next_bytenr, num_copies);
646 
647 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
648 			struct btrfsic_block *next_block;
649 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
650 			struct btrfsic_block_link *l;
651 
652 			ret = btrfsic_map_block(state, next_bytenr,
653 						state->metablock_size,
654 						&tmp_next_block_ctx,
655 						mirror_num);
656 			if (ret) {
657 				pr_info("btrfsic: btrfsic_map_block(root @%llu, mirror %d) failed!\n",
658 				       next_bytenr, mirror_num);
659 				kfree(selected_super);
660 				return -1;
661 			}
662 
663 			next_block = btrfsic_block_hashtable_lookup(
664 					tmp_next_block_ctx.dev->bdev,
665 					tmp_next_block_ctx.dev_bytenr,
666 					&state->block_hashtable);
667 			BUG_ON(NULL == next_block);
668 
669 			l = btrfsic_block_link_hashtable_lookup(
670 					tmp_next_block_ctx.dev->bdev,
671 					tmp_next_block_ctx.dev_bytenr,
672 					state->latest_superblock->dev_state->
673 					bdev,
674 					state->latest_superblock->dev_bytenr,
675 					&state->block_link_hashtable);
676 			BUG_ON(NULL == l);
677 
678 			ret = btrfsic_read_block(state, &tmp_next_block_ctx);
679 			if (ret < (int)PAGE_SIZE) {
680 				pr_info("btrfsic: read @logical %llu failed!\n",
681 				       tmp_next_block_ctx.start);
682 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
683 				kfree(selected_super);
684 				return -1;
685 			}
686 
687 			ret = btrfsic_process_metablock(state,
688 							next_block,
689 							&tmp_next_block_ctx,
690 							BTRFS_MAX_LEVEL + 3, 1);
691 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
692 		}
693 	}
694 
695 	kfree(selected_super);
696 	return ret;
697 }
698 
btrfsic_process_superblock_dev_mirror(struct btrfsic_state * state,struct btrfsic_dev_state * dev_state,struct btrfs_device * device,int superblock_mirror_num,struct btrfsic_dev_state ** selected_dev_state,struct btrfs_super_block * selected_super)699 static int btrfsic_process_superblock_dev_mirror(
700 		struct btrfsic_state *state,
701 		struct btrfsic_dev_state *dev_state,
702 		struct btrfs_device *device,
703 		int superblock_mirror_num,
704 		struct btrfsic_dev_state **selected_dev_state,
705 		struct btrfs_super_block *selected_super)
706 {
707 	struct btrfs_fs_info *fs_info = state->fs_info;
708 	struct btrfs_super_block *super_tmp;
709 	u64 dev_bytenr;
710 	struct btrfsic_block *superblock_tmp;
711 	int pass;
712 	struct block_device *const superblock_bdev = device->bdev;
713 	struct page *page;
714 	struct address_space *mapping = superblock_bdev->bd_inode->i_mapping;
715 	int ret = 0;
716 
717 	/* super block bytenr is always the unmapped device bytenr */
718 	dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
719 	if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->commit_total_bytes)
720 		return -1;
721 
722 	page = read_cache_page_gfp(mapping, dev_bytenr >> PAGE_SHIFT, GFP_NOFS);
723 	if (IS_ERR(page))
724 		return -1;
725 
726 	super_tmp = page_address(page);
727 
728 	if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
729 	    btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
730 	    memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
731 	    btrfs_super_nodesize(super_tmp) != state->metablock_size ||
732 	    btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
733 		ret = 0;
734 		goto out;
735 	}
736 
737 	superblock_tmp =
738 	    btrfsic_block_hashtable_lookup(superblock_bdev,
739 					   dev_bytenr,
740 					   &state->block_hashtable);
741 	if (NULL == superblock_tmp) {
742 		superblock_tmp = btrfsic_block_alloc();
743 		if (NULL == superblock_tmp) {
744 			ret = -1;
745 			goto out;
746 		}
747 		/* for superblock, only the dev_bytenr makes sense */
748 		superblock_tmp->dev_bytenr = dev_bytenr;
749 		superblock_tmp->dev_state = dev_state;
750 		superblock_tmp->logical_bytenr = dev_bytenr;
751 		superblock_tmp->generation = btrfs_super_generation(super_tmp);
752 		superblock_tmp->is_metadata = 1;
753 		superblock_tmp->is_superblock = 1;
754 		superblock_tmp->is_iodone = 1;
755 		superblock_tmp->never_written = 0;
756 		superblock_tmp->mirror_num = 1 + superblock_mirror_num;
757 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
758 			btrfs_info_in_rcu(fs_info,
759 				"new initial S-block (bdev %p, %s) @%llu (%s/%llu/%d)",
760 				     superblock_bdev,
761 				     rcu_str_deref(device->name), dev_bytenr,
762 				     dev_state->name, dev_bytenr,
763 				     superblock_mirror_num);
764 		list_add(&superblock_tmp->all_blocks_node,
765 			 &state->all_blocks_list);
766 		btrfsic_block_hashtable_add(superblock_tmp,
767 					    &state->block_hashtable);
768 	}
769 
770 	/* select the one with the highest generation field */
771 	if (btrfs_super_generation(super_tmp) >
772 	    state->max_superblock_generation ||
773 	    0 == state->max_superblock_generation) {
774 		memcpy(selected_super, super_tmp, sizeof(*selected_super));
775 		*selected_dev_state = dev_state;
776 		state->max_superblock_generation =
777 		    btrfs_super_generation(super_tmp);
778 		state->latest_superblock = superblock_tmp;
779 	}
780 
781 	for (pass = 0; pass < 3; pass++) {
782 		u64 next_bytenr;
783 		int num_copies;
784 		int mirror_num;
785 		const char *additional_string = NULL;
786 		struct btrfs_disk_key tmp_disk_key;
787 
788 		tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
789 		tmp_disk_key.offset = 0;
790 		switch (pass) {
791 		case 0:
792 			btrfs_set_disk_key_objectid(&tmp_disk_key,
793 						    BTRFS_ROOT_TREE_OBJECTID);
794 			additional_string = "initial root ";
795 			next_bytenr = btrfs_super_root(super_tmp);
796 			break;
797 		case 1:
798 			btrfs_set_disk_key_objectid(&tmp_disk_key,
799 						    BTRFS_CHUNK_TREE_OBJECTID);
800 			additional_string = "initial chunk ";
801 			next_bytenr = btrfs_super_chunk_root(super_tmp);
802 			break;
803 		case 2:
804 			btrfs_set_disk_key_objectid(&tmp_disk_key,
805 						    BTRFS_TREE_LOG_OBJECTID);
806 			additional_string = "initial log ";
807 			next_bytenr = btrfs_super_log_root(super_tmp);
808 			if (0 == next_bytenr)
809 				continue;
810 			break;
811 		}
812 
813 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
814 					      state->metablock_size);
815 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
816 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
817 			       next_bytenr, num_copies);
818 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
819 			struct btrfsic_block *next_block;
820 			struct btrfsic_block_data_ctx tmp_next_block_ctx;
821 			struct btrfsic_block_link *l;
822 
823 			if (btrfsic_map_block(state, next_bytenr,
824 					      state->metablock_size,
825 					      &tmp_next_block_ctx,
826 					      mirror_num)) {
827 				pr_info("btrfsic: btrfsic_map_block(bytenr @%llu, mirror %d) failed!\n",
828 				       next_bytenr, mirror_num);
829 				ret = -1;
830 				goto out;
831 			}
832 
833 			next_block = btrfsic_block_lookup_or_add(
834 					state, &tmp_next_block_ctx,
835 					additional_string, 1, 1, 0,
836 					mirror_num, NULL);
837 			if (NULL == next_block) {
838 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
839 				ret = -1;
840 				goto out;
841 			}
842 
843 			next_block->disk_key = tmp_disk_key;
844 			next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
845 			l = btrfsic_block_link_lookup_or_add(
846 					state, &tmp_next_block_ctx,
847 					next_block, superblock_tmp,
848 					BTRFSIC_GENERATION_UNKNOWN);
849 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
850 			if (NULL == l) {
851 				ret = -1;
852 				goto out;
853 			}
854 		}
855 	}
856 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
857 		btrfsic_dump_tree_sub(state, superblock_tmp, 0);
858 
859 out:
860 	put_page(page);
861 	return ret;
862 }
863 
btrfsic_stack_frame_alloc(void)864 static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
865 {
866 	struct btrfsic_stack_frame *sf;
867 
868 	sf = kzalloc(sizeof(*sf), GFP_NOFS);
869 	if (sf)
870 		sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
871 	return sf;
872 }
873 
btrfsic_stack_frame_free(struct btrfsic_stack_frame * sf)874 static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
875 {
876 	BUG_ON(!(NULL == sf ||
877 		 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
878 	kfree(sf);
879 }
880 
btrfsic_process_metablock(struct btrfsic_state * state,struct btrfsic_block * const first_block,struct btrfsic_block_data_ctx * const first_block_ctx,int first_limit_nesting,int force_iodone_flag)881 static noinline_for_stack int btrfsic_process_metablock(
882 		struct btrfsic_state *state,
883 		struct btrfsic_block *const first_block,
884 		struct btrfsic_block_data_ctx *const first_block_ctx,
885 		int first_limit_nesting, int force_iodone_flag)
886 {
887 	struct btrfsic_stack_frame initial_stack_frame = { 0 };
888 	struct btrfsic_stack_frame *sf;
889 	struct btrfsic_stack_frame *next_stack;
890 	struct btrfs_header *const first_hdr =
891 		(struct btrfs_header *)first_block_ctx->datav[0];
892 
893 	BUG_ON(!first_hdr);
894 	sf = &initial_stack_frame;
895 	sf->error = 0;
896 	sf->i = -1;
897 	sf->limit_nesting = first_limit_nesting;
898 	sf->block = first_block;
899 	sf->block_ctx = first_block_ctx;
900 	sf->next_block = NULL;
901 	sf->hdr = first_hdr;
902 	sf->prev = NULL;
903 
904 continue_with_new_stack_frame:
905 	sf->block->generation = btrfs_stack_header_generation(sf->hdr);
906 	if (0 == sf->hdr->level) {
907 		struct btrfs_leaf *const leafhdr =
908 		    (struct btrfs_leaf *)sf->hdr;
909 
910 		if (-1 == sf->i) {
911 			sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
912 
913 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
914 				pr_info("leaf %llu items %d generation %llu owner %llu\n",
915 				       sf->block_ctx->start, sf->nr,
916 				       btrfs_stack_header_generation(
917 					       &leafhdr->header),
918 				       btrfs_stack_header_owner(
919 					       &leafhdr->header));
920 		}
921 
922 continue_with_current_leaf_stack_frame:
923 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
924 			sf->i++;
925 			sf->num_copies = 0;
926 		}
927 
928 		if (sf->i < sf->nr) {
929 			struct btrfs_item disk_item;
930 			u32 disk_item_offset =
931 				(uintptr_t)(leafhdr->items + sf->i) -
932 				(uintptr_t)leafhdr;
933 			struct btrfs_disk_key *disk_key;
934 			u8 type;
935 			u32 item_offset;
936 			u32 item_size;
937 
938 			if (disk_item_offset + sizeof(struct btrfs_item) >
939 			    sf->block_ctx->len) {
940 leaf_item_out_of_bounce_error:
941 				pr_info("btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
942 				       sf->block_ctx->start,
943 				       sf->block_ctx->dev->name);
944 				goto one_stack_frame_backwards;
945 			}
946 			btrfsic_read_from_block_data(sf->block_ctx,
947 						     &disk_item,
948 						     disk_item_offset,
949 						     sizeof(struct btrfs_item));
950 			item_offset = btrfs_stack_item_offset(&disk_item);
951 			item_size = btrfs_stack_item_size(&disk_item);
952 			disk_key = &disk_item.key;
953 			type = btrfs_disk_key_type(disk_key);
954 
955 			if (BTRFS_ROOT_ITEM_KEY == type) {
956 				struct btrfs_root_item root_item;
957 				u32 root_item_offset;
958 				u64 next_bytenr;
959 
960 				root_item_offset = item_offset +
961 					offsetof(struct btrfs_leaf, items);
962 				if (root_item_offset + item_size >
963 				    sf->block_ctx->len)
964 					goto leaf_item_out_of_bounce_error;
965 				btrfsic_read_from_block_data(
966 					sf->block_ctx, &root_item,
967 					root_item_offset,
968 					item_size);
969 				next_bytenr = btrfs_root_bytenr(&root_item);
970 
971 				sf->error =
972 				    btrfsic_create_link_to_next_block(
973 						state,
974 						sf->block,
975 						sf->block_ctx,
976 						next_bytenr,
977 						sf->limit_nesting,
978 						&sf->next_block_ctx,
979 						&sf->next_block,
980 						force_iodone_flag,
981 						&sf->num_copies,
982 						&sf->mirror_num,
983 						disk_key,
984 						btrfs_root_generation(
985 						&root_item));
986 				if (sf->error)
987 					goto one_stack_frame_backwards;
988 
989 				if (NULL != sf->next_block) {
990 					struct btrfs_header *const next_hdr =
991 					    (struct btrfs_header *)
992 					    sf->next_block_ctx.datav[0];
993 
994 					next_stack =
995 					    btrfsic_stack_frame_alloc();
996 					if (NULL == next_stack) {
997 						sf->error = -1;
998 						btrfsic_release_block_ctx(
999 								&sf->
1000 								next_block_ctx);
1001 						goto one_stack_frame_backwards;
1002 					}
1003 
1004 					next_stack->i = -1;
1005 					next_stack->block = sf->next_block;
1006 					next_stack->block_ctx =
1007 					    &sf->next_block_ctx;
1008 					next_stack->next_block = NULL;
1009 					next_stack->hdr = next_hdr;
1010 					next_stack->limit_nesting =
1011 					    sf->limit_nesting - 1;
1012 					next_stack->prev = sf;
1013 					sf = next_stack;
1014 					goto continue_with_new_stack_frame;
1015 				}
1016 			} else if (BTRFS_EXTENT_DATA_KEY == type &&
1017 				   state->include_extent_data) {
1018 				sf->error = btrfsic_handle_extent_data(
1019 						state,
1020 						sf->block,
1021 						sf->block_ctx,
1022 						item_offset,
1023 						force_iodone_flag);
1024 				if (sf->error)
1025 					goto one_stack_frame_backwards;
1026 			}
1027 
1028 			goto continue_with_current_leaf_stack_frame;
1029 		}
1030 	} else {
1031 		struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1032 
1033 		if (-1 == sf->i) {
1034 			sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1035 
1036 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1037 				pr_info("node %llu level %d items %d generation %llu owner %llu\n",
1038 				       sf->block_ctx->start,
1039 				       nodehdr->header.level, sf->nr,
1040 				       btrfs_stack_header_generation(
1041 				       &nodehdr->header),
1042 				       btrfs_stack_header_owner(
1043 				       &nodehdr->header));
1044 		}
1045 
1046 continue_with_current_node_stack_frame:
1047 		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1048 			sf->i++;
1049 			sf->num_copies = 0;
1050 		}
1051 
1052 		if (sf->i < sf->nr) {
1053 			struct btrfs_key_ptr key_ptr;
1054 			u32 key_ptr_offset;
1055 			u64 next_bytenr;
1056 
1057 			key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1058 					  (uintptr_t)nodehdr;
1059 			if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1060 			    sf->block_ctx->len) {
1061 				pr_info("btrfsic: node item out of bounce at logical %llu, dev %s\n",
1062 				       sf->block_ctx->start,
1063 				       sf->block_ctx->dev->name);
1064 				goto one_stack_frame_backwards;
1065 			}
1066 			btrfsic_read_from_block_data(
1067 				sf->block_ctx, &key_ptr, key_ptr_offset,
1068 				sizeof(struct btrfs_key_ptr));
1069 			next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1070 
1071 			sf->error = btrfsic_create_link_to_next_block(
1072 					state,
1073 					sf->block,
1074 					sf->block_ctx,
1075 					next_bytenr,
1076 					sf->limit_nesting,
1077 					&sf->next_block_ctx,
1078 					&sf->next_block,
1079 					force_iodone_flag,
1080 					&sf->num_copies,
1081 					&sf->mirror_num,
1082 					&key_ptr.key,
1083 					btrfs_stack_key_generation(&key_ptr));
1084 			if (sf->error)
1085 				goto one_stack_frame_backwards;
1086 
1087 			if (NULL != sf->next_block) {
1088 				struct btrfs_header *const next_hdr =
1089 				    (struct btrfs_header *)
1090 				    sf->next_block_ctx.datav[0];
1091 
1092 				next_stack = btrfsic_stack_frame_alloc();
1093 				if (NULL == next_stack) {
1094 					sf->error = -1;
1095 					goto one_stack_frame_backwards;
1096 				}
1097 
1098 				next_stack->i = -1;
1099 				next_stack->block = sf->next_block;
1100 				next_stack->block_ctx = &sf->next_block_ctx;
1101 				next_stack->next_block = NULL;
1102 				next_stack->hdr = next_hdr;
1103 				next_stack->limit_nesting =
1104 				    sf->limit_nesting - 1;
1105 				next_stack->prev = sf;
1106 				sf = next_stack;
1107 				goto continue_with_new_stack_frame;
1108 			}
1109 
1110 			goto continue_with_current_node_stack_frame;
1111 		}
1112 	}
1113 
1114 one_stack_frame_backwards:
1115 	if (NULL != sf->prev) {
1116 		struct btrfsic_stack_frame *const prev = sf->prev;
1117 
1118 		/* the one for the initial block is freed in the caller */
1119 		btrfsic_release_block_ctx(sf->block_ctx);
1120 
1121 		if (sf->error) {
1122 			prev->error = sf->error;
1123 			btrfsic_stack_frame_free(sf);
1124 			sf = prev;
1125 			goto one_stack_frame_backwards;
1126 		}
1127 
1128 		btrfsic_stack_frame_free(sf);
1129 		sf = prev;
1130 		goto continue_with_new_stack_frame;
1131 	} else {
1132 		BUG_ON(&initial_stack_frame != sf);
1133 	}
1134 
1135 	return sf->error;
1136 }
1137 
btrfsic_read_from_block_data(struct btrfsic_block_data_ctx * block_ctx,void * dstv,u32 offset,size_t len)1138 static void btrfsic_read_from_block_data(
1139 	struct btrfsic_block_data_ctx *block_ctx,
1140 	void *dstv, u32 offset, size_t len)
1141 {
1142 	size_t cur;
1143 	size_t pgoff;
1144 	char *kaddr;
1145 	char *dst = (char *)dstv;
1146 	size_t start_offset = offset_in_page(block_ctx->start);
1147 	unsigned long i = (start_offset + offset) >> PAGE_SHIFT;
1148 
1149 	WARN_ON(offset + len > block_ctx->len);
1150 	pgoff = offset_in_page(start_offset + offset);
1151 
1152 	while (len > 0) {
1153 		cur = min(len, ((size_t)PAGE_SIZE - pgoff));
1154 		BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_SIZE));
1155 		kaddr = block_ctx->datav[i];
1156 		memcpy(dst, kaddr + pgoff, cur);
1157 
1158 		dst += cur;
1159 		len -= cur;
1160 		pgoff = 0;
1161 		i++;
1162 	}
1163 }
1164 
btrfsic_create_link_to_next_block(struct btrfsic_state * state,struct btrfsic_block * block,struct btrfsic_block_data_ctx * block_ctx,u64 next_bytenr,int limit_nesting,struct btrfsic_block_data_ctx * next_block_ctx,struct btrfsic_block ** next_blockp,int force_iodone_flag,int * num_copiesp,int * mirror_nump,struct btrfs_disk_key * disk_key,u64 parent_generation)1165 static int btrfsic_create_link_to_next_block(
1166 		struct btrfsic_state *state,
1167 		struct btrfsic_block *block,
1168 		struct btrfsic_block_data_ctx *block_ctx,
1169 		u64 next_bytenr,
1170 		int limit_nesting,
1171 		struct btrfsic_block_data_ctx *next_block_ctx,
1172 		struct btrfsic_block **next_blockp,
1173 		int force_iodone_flag,
1174 		int *num_copiesp, int *mirror_nump,
1175 		struct btrfs_disk_key *disk_key,
1176 		u64 parent_generation)
1177 {
1178 	struct btrfs_fs_info *fs_info = state->fs_info;
1179 	struct btrfsic_block *next_block = NULL;
1180 	int ret;
1181 	struct btrfsic_block_link *l;
1182 	int did_alloc_block_link;
1183 	int block_was_created;
1184 
1185 	*next_blockp = NULL;
1186 	if (0 == *num_copiesp) {
1187 		*num_copiesp = btrfs_num_copies(fs_info, next_bytenr,
1188 						state->metablock_size);
1189 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1190 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
1191 			       next_bytenr, *num_copiesp);
1192 		*mirror_nump = 1;
1193 	}
1194 
1195 	if (*mirror_nump > *num_copiesp)
1196 		return 0;
1197 
1198 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1199 		pr_info("btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1200 		       *mirror_nump);
1201 	ret = btrfsic_map_block(state, next_bytenr,
1202 				state->metablock_size,
1203 				next_block_ctx, *mirror_nump);
1204 	if (ret) {
1205 		pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1206 		       next_bytenr, *mirror_nump);
1207 		btrfsic_release_block_ctx(next_block_ctx);
1208 		*next_blockp = NULL;
1209 		return -1;
1210 	}
1211 
1212 	next_block = btrfsic_block_lookup_or_add(state,
1213 						 next_block_ctx, "referenced ",
1214 						 1, force_iodone_flag,
1215 						 !force_iodone_flag,
1216 						 *mirror_nump,
1217 						 &block_was_created);
1218 	if (NULL == next_block) {
1219 		btrfsic_release_block_ctx(next_block_ctx);
1220 		*next_blockp = NULL;
1221 		return -1;
1222 	}
1223 	if (block_was_created) {
1224 		l = NULL;
1225 		next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1226 	} else {
1227 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1228 			if (next_block->logical_bytenr != next_bytenr &&
1229 			    !(!next_block->is_metadata &&
1230 			      0 == next_block->logical_bytenr))
1231 				pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1232 				       next_bytenr, next_block_ctx->dev->name,
1233 				       next_block_ctx->dev_bytenr, *mirror_nump,
1234 				       btrfsic_get_block_type(state,
1235 							      next_block),
1236 				       next_block->logical_bytenr);
1237 			else
1238 				pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1239 				       next_bytenr, next_block_ctx->dev->name,
1240 				       next_block_ctx->dev_bytenr, *mirror_nump,
1241 				       btrfsic_get_block_type(state,
1242 							      next_block));
1243 		}
1244 		next_block->logical_bytenr = next_bytenr;
1245 
1246 		next_block->mirror_num = *mirror_nump;
1247 		l = btrfsic_block_link_hashtable_lookup(
1248 				next_block_ctx->dev->bdev,
1249 				next_block_ctx->dev_bytenr,
1250 				block_ctx->dev->bdev,
1251 				block_ctx->dev_bytenr,
1252 				&state->block_link_hashtable);
1253 	}
1254 
1255 	next_block->disk_key = *disk_key;
1256 	if (NULL == l) {
1257 		l = btrfsic_block_link_alloc();
1258 		if (NULL == l) {
1259 			btrfsic_release_block_ctx(next_block_ctx);
1260 			*next_blockp = NULL;
1261 			return -1;
1262 		}
1263 
1264 		did_alloc_block_link = 1;
1265 		l->block_ref_to = next_block;
1266 		l->block_ref_from = block;
1267 		l->ref_cnt = 1;
1268 		l->parent_generation = parent_generation;
1269 
1270 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1271 			btrfsic_print_add_link(state, l);
1272 
1273 		list_add(&l->node_ref_to, &block->ref_to_list);
1274 		list_add(&l->node_ref_from, &next_block->ref_from_list);
1275 
1276 		btrfsic_block_link_hashtable_add(l,
1277 						 &state->block_link_hashtable);
1278 	} else {
1279 		did_alloc_block_link = 0;
1280 		if (0 == limit_nesting) {
1281 			l->ref_cnt++;
1282 			l->parent_generation = parent_generation;
1283 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1284 				btrfsic_print_add_link(state, l);
1285 		}
1286 	}
1287 
1288 	if (limit_nesting > 0 && did_alloc_block_link) {
1289 		ret = btrfsic_read_block(state, next_block_ctx);
1290 		if (ret < (int)next_block_ctx->len) {
1291 			pr_info("btrfsic: read block @logical %llu failed!\n",
1292 			       next_bytenr);
1293 			btrfsic_release_block_ctx(next_block_ctx);
1294 			*next_blockp = NULL;
1295 			return -1;
1296 		}
1297 
1298 		*next_blockp = next_block;
1299 	} else {
1300 		*next_blockp = NULL;
1301 	}
1302 	(*mirror_nump)++;
1303 
1304 	return 0;
1305 }
1306 
btrfsic_handle_extent_data(struct btrfsic_state * state,struct btrfsic_block * block,struct btrfsic_block_data_ctx * block_ctx,u32 item_offset,int force_iodone_flag)1307 static int btrfsic_handle_extent_data(
1308 		struct btrfsic_state *state,
1309 		struct btrfsic_block *block,
1310 		struct btrfsic_block_data_ctx *block_ctx,
1311 		u32 item_offset, int force_iodone_flag)
1312 {
1313 	struct btrfs_fs_info *fs_info = state->fs_info;
1314 	struct btrfs_file_extent_item file_extent_item;
1315 	u64 file_extent_item_offset;
1316 	u64 next_bytenr;
1317 	u64 num_bytes;
1318 	u64 generation;
1319 	struct btrfsic_block_link *l;
1320 	int ret;
1321 
1322 	file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1323 				  item_offset;
1324 	if (file_extent_item_offset +
1325 	    offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1326 	    block_ctx->len) {
1327 		pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1328 		       block_ctx->start, block_ctx->dev->name);
1329 		return -1;
1330 	}
1331 
1332 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1333 		file_extent_item_offset,
1334 		offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1335 	if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1336 	    btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1337 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1338 			pr_info("extent_data: type %u, disk_bytenr = %llu\n",
1339 			       file_extent_item.type,
1340 			       btrfs_stack_file_extent_disk_bytenr(
1341 			       &file_extent_item));
1342 		return 0;
1343 	}
1344 
1345 	if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1346 	    block_ctx->len) {
1347 		pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1348 		       block_ctx->start, block_ctx->dev->name);
1349 		return -1;
1350 	}
1351 	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1352 				     file_extent_item_offset,
1353 				     sizeof(struct btrfs_file_extent_item));
1354 	next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item);
1355 	if (btrfs_stack_file_extent_compression(&file_extent_item) ==
1356 	    BTRFS_COMPRESS_NONE) {
1357 		next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item);
1358 		num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1359 	} else {
1360 		num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item);
1361 	}
1362 	generation = btrfs_stack_file_extent_generation(&file_extent_item);
1363 
1364 	if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1365 		pr_info("extent_data: type %u, disk_bytenr = %llu, offset = %llu, num_bytes = %llu\n",
1366 		       file_extent_item.type,
1367 		       btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1368 		       btrfs_stack_file_extent_offset(&file_extent_item),
1369 		       num_bytes);
1370 	while (num_bytes > 0) {
1371 		u32 chunk_len;
1372 		int num_copies;
1373 		int mirror_num;
1374 
1375 		if (num_bytes > state->datablock_size)
1376 			chunk_len = state->datablock_size;
1377 		else
1378 			chunk_len = num_bytes;
1379 
1380 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
1381 					      state->datablock_size);
1382 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1383 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
1384 			       next_bytenr, num_copies);
1385 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1386 			struct btrfsic_block_data_ctx next_block_ctx;
1387 			struct btrfsic_block *next_block;
1388 			int block_was_created;
1389 
1390 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1391 				pr_info("btrfsic_handle_extent_data(mirror_num=%d)\n",
1392 					mirror_num);
1393 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1394 				pr_info("\tdisk_bytenr = %llu, num_bytes %u\n",
1395 				       next_bytenr, chunk_len);
1396 			ret = btrfsic_map_block(state, next_bytenr,
1397 						chunk_len, &next_block_ctx,
1398 						mirror_num);
1399 			if (ret) {
1400 				pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1401 				       next_bytenr, mirror_num);
1402 				return -1;
1403 			}
1404 
1405 			next_block = btrfsic_block_lookup_or_add(
1406 					state,
1407 					&next_block_ctx,
1408 					"referenced ",
1409 					0,
1410 					force_iodone_flag,
1411 					!force_iodone_flag,
1412 					mirror_num,
1413 					&block_was_created);
1414 			if (NULL == next_block) {
1415 				btrfsic_release_block_ctx(&next_block_ctx);
1416 				return -1;
1417 			}
1418 			if (!block_was_created) {
1419 				if ((state->print_mask &
1420 				     BTRFSIC_PRINT_MASK_VERBOSE) &&
1421 				    next_block->logical_bytenr != next_bytenr &&
1422 				    !(!next_block->is_metadata &&
1423 				      0 == next_block->logical_bytenr)) {
1424 					pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, D, bytenr mismatch (!= stored %llu).\n",
1425 					       next_bytenr,
1426 					       next_block_ctx.dev->name,
1427 					       next_block_ctx.dev_bytenr,
1428 					       mirror_num,
1429 					       next_block->logical_bytenr);
1430 				}
1431 				next_block->logical_bytenr = next_bytenr;
1432 				next_block->mirror_num = mirror_num;
1433 			}
1434 
1435 			l = btrfsic_block_link_lookup_or_add(state,
1436 							     &next_block_ctx,
1437 							     next_block, block,
1438 							     generation);
1439 			btrfsic_release_block_ctx(&next_block_ctx);
1440 			if (NULL == l)
1441 				return -1;
1442 		}
1443 
1444 		next_bytenr += chunk_len;
1445 		num_bytes -= chunk_len;
1446 	}
1447 
1448 	return 0;
1449 }
1450 
btrfsic_map_block(struct btrfsic_state * state,u64 bytenr,u32 len,struct btrfsic_block_data_ctx * block_ctx_out,int mirror_num)1451 static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1452 			     struct btrfsic_block_data_ctx *block_ctx_out,
1453 			     int mirror_num)
1454 {
1455 	struct btrfs_fs_info *fs_info = state->fs_info;
1456 	int ret;
1457 	u64 length;
1458 	struct btrfs_io_context *multi = NULL;
1459 	struct btrfs_device *device;
1460 
1461 	length = len;
1462 	ret = btrfs_map_block(fs_info, BTRFS_MAP_READ,
1463 			      bytenr, &length, &multi, mirror_num);
1464 
1465 	if (ret) {
1466 		block_ctx_out->start = 0;
1467 		block_ctx_out->dev_bytenr = 0;
1468 		block_ctx_out->len = 0;
1469 		block_ctx_out->dev = NULL;
1470 		block_ctx_out->datav = NULL;
1471 		block_ctx_out->pagev = NULL;
1472 		block_ctx_out->mem_to_free = NULL;
1473 
1474 		return ret;
1475 	}
1476 
1477 	device = multi->stripes[0].dev;
1478 	if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
1479 	    !device->bdev || !device->name)
1480 		block_ctx_out->dev = NULL;
1481 	else
1482 		block_ctx_out->dev = btrfsic_dev_state_lookup(
1483 							device->bdev->bd_dev);
1484 	block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1485 	block_ctx_out->start = bytenr;
1486 	block_ctx_out->len = len;
1487 	block_ctx_out->datav = NULL;
1488 	block_ctx_out->pagev = NULL;
1489 	block_ctx_out->mem_to_free = NULL;
1490 
1491 	kfree(multi);
1492 	if (NULL == block_ctx_out->dev) {
1493 		ret = -ENXIO;
1494 		pr_info("btrfsic: error, cannot lookup dev (#1)!\n");
1495 	}
1496 
1497 	return ret;
1498 }
1499 
btrfsic_release_block_ctx(struct btrfsic_block_data_ctx * block_ctx)1500 static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1501 {
1502 	if (block_ctx->mem_to_free) {
1503 		unsigned int num_pages;
1504 
1505 		BUG_ON(!block_ctx->datav);
1506 		BUG_ON(!block_ctx->pagev);
1507 		num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1508 			    PAGE_SHIFT;
1509 		/* Pages must be unmapped in reverse order */
1510 		while (num_pages > 0) {
1511 			num_pages--;
1512 			if (block_ctx->datav[num_pages])
1513 				block_ctx->datav[num_pages] = NULL;
1514 			if (block_ctx->pagev[num_pages]) {
1515 				__free_page(block_ctx->pagev[num_pages]);
1516 				block_ctx->pagev[num_pages] = NULL;
1517 			}
1518 		}
1519 
1520 		kfree(block_ctx->mem_to_free);
1521 		block_ctx->mem_to_free = NULL;
1522 		block_ctx->pagev = NULL;
1523 		block_ctx->datav = NULL;
1524 	}
1525 }
1526 
btrfsic_read_block(struct btrfsic_state * state,struct btrfsic_block_data_ctx * block_ctx)1527 static int btrfsic_read_block(struct btrfsic_state *state,
1528 			      struct btrfsic_block_data_ctx *block_ctx)
1529 {
1530 	unsigned int num_pages;
1531 	unsigned int i;
1532 	size_t size;
1533 	u64 dev_bytenr;
1534 	int ret;
1535 
1536 	BUG_ON(block_ctx->datav);
1537 	BUG_ON(block_ctx->pagev);
1538 	BUG_ON(block_ctx->mem_to_free);
1539 	if (!PAGE_ALIGNED(block_ctx->dev_bytenr)) {
1540 		pr_info("btrfsic: read_block() with unaligned bytenr %llu\n",
1541 		       block_ctx->dev_bytenr);
1542 		return -1;
1543 	}
1544 
1545 	num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1546 		    PAGE_SHIFT;
1547 	size = sizeof(*block_ctx->datav) + sizeof(*block_ctx->pagev);
1548 	block_ctx->mem_to_free = kcalloc(num_pages, size, GFP_NOFS);
1549 	if (!block_ctx->mem_to_free)
1550 		return -ENOMEM;
1551 	block_ctx->datav = block_ctx->mem_to_free;
1552 	block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1553 	for (i = 0; i < num_pages; i++) {
1554 		block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1555 		if (!block_ctx->pagev[i])
1556 			return -1;
1557 	}
1558 
1559 	dev_bytenr = block_ctx->dev_bytenr;
1560 	for (i = 0; i < num_pages;) {
1561 		struct bio *bio;
1562 		unsigned int j;
1563 
1564 		bio = btrfs_io_bio_alloc(num_pages - i);
1565 		bio_set_dev(bio, block_ctx->dev->bdev);
1566 		bio->bi_iter.bi_sector = dev_bytenr >> 9;
1567 		bio->bi_opf = REQ_OP_READ;
1568 
1569 		for (j = i; j < num_pages; j++) {
1570 			ret = bio_add_page(bio, block_ctx->pagev[j],
1571 					   PAGE_SIZE, 0);
1572 			if (PAGE_SIZE != ret)
1573 				break;
1574 		}
1575 		if (j == i) {
1576 			pr_info("btrfsic: error, failed to add a single page!\n");
1577 			return -1;
1578 		}
1579 		if (submit_bio_wait(bio)) {
1580 			pr_info("btrfsic: read error at logical %llu dev %s!\n",
1581 			       block_ctx->start, block_ctx->dev->name);
1582 			bio_put(bio);
1583 			return -1;
1584 		}
1585 		bio_put(bio);
1586 		dev_bytenr += (j - i) * PAGE_SIZE;
1587 		i = j;
1588 	}
1589 	for (i = 0; i < num_pages; i++)
1590 		block_ctx->datav[i] = page_address(block_ctx->pagev[i]);
1591 
1592 	return block_ctx->len;
1593 }
1594 
btrfsic_dump_database(struct btrfsic_state * state)1595 static void btrfsic_dump_database(struct btrfsic_state *state)
1596 {
1597 	const struct btrfsic_block *b_all;
1598 
1599 	BUG_ON(NULL == state);
1600 
1601 	pr_info("all_blocks_list:\n");
1602 	list_for_each_entry(b_all, &state->all_blocks_list, all_blocks_node) {
1603 		const struct btrfsic_block_link *l;
1604 
1605 		pr_info("%c-block @%llu (%s/%llu/%d)\n",
1606 		       btrfsic_get_block_type(state, b_all),
1607 		       b_all->logical_bytenr, b_all->dev_state->name,
1608 		       b_all->dev_bytenr, b_all->mirror_num);
1609 
1610 		list_for_each_entry(l, &b_all->ref_to_list, node_ref_to) {
1611 			pr_info(" %c @%llu (%s/%llu/%d) refers %u* to %c @%llu (%s/%llu/%d)\n",
1612 			       btrfsic_get_block_type(state, b_all),
1613 			       b_all->logical_bytenr, b_all->dev_state->name,
1614 			       b_all->dev_bytenr, b_all->mirror_num,
1615 			       l->ref_cnt,
1616 			       btrfsic_get_block_type(state, l->block_ref_to),
1617 			       l->block_ref_to->logical_bytenr,
1618 			       l->block_ref_to->dev_state->name,
1619 			       l->block_ref_to->dev_bytenr,
1620 			       l->block_ref_to->mirror_num);
1621 		}
1622 
1623 		list_for_each_entry(l, &b_all->ref_from_list, node_ref_from) {
1624 			pr_info(" %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
1625 			       btrfsic_get_block_type(state, b_all),
1626 			       b_all->logical_bytenr, b_all->dev_state->name,
1627 			       b_all->dev_bytenr, b_all->mirror_num,
1628 			       l->ref_cnt,
1629 			       btrfsic_get_block_type(state, l->block_ref_from),
1630 			       l->block_ref_from->logical_bytenr,
1631 			       l->block_ref_from->dev_state->name,
1632 			       l->block_ref_from->dev_bytenr,
1633 			       l->block_ref_from->mirror_num);
1634 		}
1635 
1636 		pr_info("\n");
1637 	}
1638 }
1639 
1640 /*
1641  * Test whether the disk block contains a tree block (leaf or node)
1642  * (note that this test fails for the super block)
1643  */
btrfsic_test_for_metadata(struct btrfsic_state * state,char ** datav,unsigned int num_pages)1644 static noinline_for_stack int btrfsic_test_for_metadata(
1645 		struct btrfsic_state *state,
1646 		char **datav, unsigned int num_pages)
1647 {
1648 	struct btrfs_fs_info *fs_info = state->fs_info;
1649 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
1650 	struct btrfs_header *h;
1651 	u8 csum[BTRFS_CSUM_SIZE];
1652 	unsigned int i;
1653 
1654 	if (num_pages * PAGE_SIZE < state->metablock_size)
1655 		return 1; /* not metadata */
1656 	num_pages = state->metablock_size >> PAGE_SHIFT;
1657 	h = (struct btrfs_header *)datav[0];
1658 
1659 	if (memcmp(h->fsid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE))
1660 		return 1;
1661 
1662 	shash->tfm = fs_info->csum_shash;
1663 	crypto_shash_init(shash);
1664 
1665 	for (i = 0; i < num_pages; i++) {
1666 		u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1667 		size_t sublen = i ? PAGE_SIZE :
1668 				    (PAGE_SIZE - BTRFS_CSUM_SIZE);
1669 
1670 		crypto_shash_update(shash, data, sublen);
1671 	}
1672 	crypto_shash_final(shash, csum);
1673 	if (memcmp(csum, h->csum, fs_info->csum_size))
1674 		return 1;
1675 
1676 	return 0; /* is metadata */
1677 }
1678 
btrfsic_process_written_block(struct btrfsic_dev_state * dev_state,u64 dev_bytenr,char ** mapped_datav,unsigned int num_pages,struct bio * bio,int * bio_is_patched,int submit_bio_bh_rw)1679 static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1680 					  u64 dev_bytenr, char **mapped_datav,
1681 					  unsigned int num_pages,
1682 					  struct bio *bio, int *bio_is_patched,
1683 					  int submit_bio_bh_rw)
1684 {
1685 	int is_metadata;
1686 	struct btrfsic_block *block;
1687 	struct btrfsic_block_data_ctx block_ctx;
1688 	int ret;
1689 	struct btrfsic_state *state = dev_state->state;
1690 	struct block_device *bdev = dev_state->bdev;
1691 	unsigned int processed_len;
1692 
1693 	if (NULL != bio_is_patched)
1694 		*bio_is_patched = 0;
1695 
1696 again:
1697 	if (num_pages == 0)
1698 		return;
1699 
1700 	processed_len = 0;
1701 	is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1702 						      num_pages));
1703 
1704 	block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1705 					       &state->block_hashtable);
1706 	if (NULL != block) {
1707 		u64 bytenr = 0;
1708 		struct btrfsic_block_link *l, *tmp;
1709 
1710 		if (block->is_superblock) {
1711 			bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1712 						    mapped_datav[0]);
1713 			if (num_pages * PAGE_SIZE <
1714 			    BTRFS_SUPER_INFO_SIZE) {
1715 				pr_info("btrfsic: cannot work with too short bios!\n");
1716 				return;
1717 			}
1718 			is_metadata = 1;
1719 			BUG_ON(!PAGE_ALIGNED(BTRFS_SUPER_INFO_SIZE));
1720 			processed_len = BTRFS_SUPER_INFO_SIZE;
1721 			if (state->print_mask &
1722 			    BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1723 				pr_info("[before new superblock is written]:\n");
1724 				btrfsic_dump_tree_sub(state, block, 0);
1725 			}
1726 		}
1727 		if (is_metadata) {
1728 			if (!block->is_superblock) {
1729 				if (num_pages * PAGE_SIZE <
1730 				    state->metablock_size) {
1731 					pr_info("btrfsic: cannot work with too short bios!\n");
1732 					return;
1733 				}
1734 				processed_len = state->metablock_size;
1735 				bytenr = btrfs_stack_header_bytenr(
1736 						(struct btrfs_header *)
1737 						mapped_datav[0]);
1738 				btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1739 							       dev_state,
1740 							       dev_bytenr);
1741 			}
1742 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1743 				if (block->logical_bytenr != bytenr &&
1744 				    !(!block->is_metadata &&
1745 				      block->logical_bytenr == 0))
1746 					pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1747 					       bytenr, dev_state->name,
1748 					       dev_bytenr,
1749 					       block->mirror_num,
1750 					       btrfsic_get_block_type(state,
1751 								      block),
1752 					       block->logical_bytenr);
1753 				else
1754 					pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1755 					       bytenr, dev_state->name,
1756 					       dev_bytenr, block->mirror_num,
1757 					       btrfsic_get_block_type(state,
1758 								      block));
1759 			}
1760 			block->logical_bytenr = bytenr;
1761 		} else {
1762 			if (num_pages * PAGE_SIZE <
1763 			    state->datablock_size) {
1764 				pr_info("btrfsic: cannot work with too short bios!\n");
1765 				return;
1766 			}
1767 			processed_len = state->datablock_size;
1768 			bytenr = block->logical_bytenr;
1769 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1770 				pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1771 				       bytenr, dev_state->name, dev_bytenr,
1772 				       block->mirror_num,
1773 				       btrfsic_get_block_type(state, block));
1774 		}
1775 
1776 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1777 			pr_info("ref_to_list: %cE, ref_from_list: %cE\n",
1778 			       list_empty(&block->ref_to_list) ? ' ' : '!',
1779 			       list_empty(&block->ref_from_list) ? ' ' : '!');
1780 		if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1781 			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",
1782 			       btrfsic_get_block_type(state, block), bytenr,
1783 			       dev_state->name, dev_bytenr, block->mirror_num,
1784 			       block->generation,
1785 			       btrfs_disk_key_objectid(&block->disk_key),
1786 			       block->disk_key.type,
1787 			       btrfs_disk_key_offset(&block->disk_key),
1788 			       btrfs_stack_header_generation(
1789 				       (struct btrfs_header *) mapped_datav[0]),
1790 			       state->max_superblock_generation);
1791 			btrfsic_dump_tree(state);
1792 		}
1793 
1794 		if (!block->is_iodone && !block->never_written) {
1795 			pr_info("btrfs: attempt to overwrite %c-block @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu, which is not yet iodone!\n",
1796 			       btrfsic_get_block_type(state, block), bytenr,
1797 			       dev_state->name, dev_bytenr, block->mirror_num,
1798 			       block->generation,
1799 			       btrfs_stack_header_generation(
1800 				       (struct btrfs_header *)
1801 				       mapped_datav[0]));
1802 			/* it would not be safe to go on */
1803 			btrfsic_dump_tree(state);
1804 			goto continue_loop;
1805 		}
1806 
1807 		/*
1808 		 * Clear all references of this block. Do not free
1809 		 * the block itself even if is not referenced anymore
1810 		 * because it still carries valuable information
1811 		 * like whether it was ever written and IO completed.
1812 		 */
1813 		list_for_each_entry_safe(l, tmp, &block->ref_to_list,
1814 					 node_ref_to) {
1815 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1816 				btrfsic_print_rem_link(state, l);
1817 			l->ref_cnt--;
1818 			if (0 == l->ref_cnt) {
1819 				list_del(&l->node_ref_to);
1820 				list_del(&l->node_ref_from);
1821 				btrfsic_block_link_hashtable_remove(l);
1822 				btrfsic_block_link_free(l);
1823 			}
1824 		}
1825 
1826 		block_ctx.dev = dev_state;
1827 		block_ctx.dev_bytenr = dev_bytenr;
1828 		block_ctx.start = bytenr;
1829 		block_ctx.len = processed_len;
1830 		block_ctx.pagev = NULL;
1831 		block_ctx.mem_to_free = NULL;
1832 		block_ctx.datav = mapped_datav;
1833 
1834 		if (is_metadata || state->include_extent_data) {
1835 			block->never_written = 0;
1836 			block->iodone_w_error = 0;
1837 			if (NULL != bio) {
1838 				block->is_iodone = 0;
1839 				BUG_ON(NULL == bio_is_patched);
1840 				if (!*bio_is_patched) {
1841 					block->orig_bio_private =
1842 					    bio->bi_private;
1843 					block->orig_bio_end_io =
1844 					    bio->bi_end_io;
1845 					block->next_in_same_bio = NULL;
1846 					bio->bi_private = block;
1847 					bio->bi_end_io = btrfsic_bio_end_io;
1848 					*bio_is_patched = 1;
1849 				} else {
1850 					struct btrfsic_block *chained_block =
1851 					    (struct btrfsic_block *)
1852 					    bio->bi_private;
1853 
1854 					BUG_ON(NULL == chained_block);
1855 					block->orig_bio_private =
1856 					    chained_block->orig_bio_private;
1857 					block->orig_bio_end_io =
1858 					    chained_block->orig_bio_end_io;
1859 					block->next_in_same_bio = chained_block;
1860 					bio->bi_private = block;
1861 				}
1862 			} else {
1863 				block->is_iodone = 1;
1864 				block->orig_bio_private = NULL;
1865 				block->orig_bio_end_io = NULL;
1866 				block->next_in_same_bio = NULL;
1867 			}
1868 		}
1869 
1870 		block->flush_gen = dev_state->last_flush_gen + 1;
1871 		block->submit_bio_bh_rw = submit_bio_bh_rw;
1872 		if (is_metadata) {
1873 			block->logical_bytenr = bytenr;
1874 			block->is_metadata = 1;
1875 			if (block->is_superblock) {
1876 				BUG_ON(PAGE_SIZE !=
1877 				       BTRFS_SUPER_INFO_SIZE);
1878 				ret = btrfsic_process_written_superblock(
1879 						state,
1880 						block,
1881 						(struct btrfs_super_block *)
1882 						mapped_datav[0]);
1883 				if (state->print_mask &
1884 				    BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
1885 					pr_info("[after new superblock is written]:\n");
1886 					btrfsic_dump_tree_sub(state, block, 0);
1887 				}
1888 			} else {
1889 				block->mirror_num = 0;	/* unknown */
1890 				ret = btrfsic_process_metablock(
1891 						state,
1892 						block,
1893 						&block_ctx,
1894 						0, 0);
1895 			}
1896 			if (ret)
1897 				pr_info("btrfsic: btrfsic_process_metablock(root @%llu) failed!\n",
1898 				       dev_bytenr);
1899 		} else {
1900 			block->is_metadata = 0;
1901 			block->mirror_num = 0;	/* unknown */
1902 			block->generation = BTRFSIC_GENERATION_UNKNOWN;
1903 			if (!state->include_extent_data
1904 			    && list_empty(&block->ref_from_list)) {
1905 				/*
1906 				 * disk block is overwritten with extent
1907 				 * data (not meta data) and we are configured
1908 				 * to not include extent data: take the
1909 				 * chance and free the block's memory
1910 				 */
1911 				btrfsic_block_hashtable_remove(block);
1912 				list_del(&block->all_blocks_node);
1913 				btrfsic_block_free(block);
1914 			}
1915 		}
1916 		btrfsic_release_block_ctx(&block_ctx);
1917 	} else {
1918 		/* block has not been found in hash table */
1919 		u64 bytenr;
1920 
1921 		if (!is_metadata) {
1922 			processed_len = state->datablock_size;
1923 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1924 				pr_info("Written block (%s/%llu/?) !found in hash table, D.\n",
1925 				       dev_state->name, dev_bytenr);
1926 			if (!state->include_extent_data) {
1927 				/* ignore that written D block */
1928 				goto continue_loop;
1929 			}
1930 
1931 			/* this is getting ugly for the
1932 			 * include_extent_data case... */
1933 			bytenr = 0;	/* unknown */
1934 		} else {
1935 			processed_len = state->metablock_size;
1936 			bytenr = btrfs_stack_header_bytenr(
1937 					(struct btrfs_header *)
1938 					mapped_datav[0]);
1939 			btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
1940 						       dev_bytenr);
1941 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1942 				pr_info("Written block @%llu (%s/%llu/?) !found in hash table, M.\n",
1943 				       bytenr, dev_state->name, dev_bytenr);
1944 		}
1945 
1946 		block_ctx.dev = dev_state;
1947 		block_ctx.dev_bytenr = dev_bytenr;
1948 		block_ctx.start = bytenr;
1949 		block_ctx.len = processed_len;
1950 		block_ctx.pagev = NULL;
1951 		block_ctx.mem_to_free = NULL;
1952 		block_ctx.datav = mapped_datav;
1953 
1954 		block = btrfsic_block_alloc();
1955 		if (NULL == block) {
1956 			btrfsic_release_block_ctx(&block_ctx);
1957 			goto continue_loop;
1958 		}
1959 		block->dev_state = dev_state;
1960 		block->dev_bytenr = dev_bytenr;
1961 		block->logical_bytenr = bytenr;
1962 		block->is_metadata = is_metadata;
1963 		block->never_written = 0;
1964 		block->iodone_w_error = 0;
1965 		block->mirror_num = 0;	/* unknown */
1966 		block->flush_gen = dev_state->last_flush_gen + 1;
1967 		block->submit_bio_bh_rw = submit_bio_bh_rw;
1968 		if (NULL != bio) {
1969 			block->is_iodone = 0;
1970 			BUG_ON(NULL == bio_is_patched);
1971 			if (!*bio_is_patched) {
1972 				block->orig_bio_private = bio->bi_private;
1973 				block->orig_bio_end_io = bio->bi_end_io;
1974 				block->next_in_same_bio = NULL;
1975 				bio->bi_private = block;
1976 				bio->bi_end_io = btrfsic_bio_end_io;
1977 				*bio_is_patched = 1;
1978 			} else {
1979 				struct btrfsic_block *chained_block =
1980 				    (struct btrfsic_block *)
1981 				    bio->bi_private;
1982 
1983 				BUG_ON(NULL == chained_block);
1984 				block->orig_bio_private =
1985 				    chained_block->orig_bio_private;
1986 				block->orig_bio_end_io =
1987 				    chained_block->orig_bio_end_io;
1988 				block->next_in_same_bio = chained_block;
1989 				bio->bi_private = block;
1990 			}
1991 		} else {
1992 			block->is_iodone = 1;
1993 			block->orig_bio_private = NULL;
1994 			block->orig_bio_end_io = NULL;
1995 			block->next_in_same_bio = NULL;
1996 		}
1997 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1998 			pr_info("New written %c-block @%llu (%s/%llu/%d)\n",
1999 			       is_metadata ? 'M' : 'D',
2000 			       block->logical_bytenr, block->dev_state->name,
2001 			       block->dev_bytenr, block->mirror_num);
2002 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2003 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2004 
2005 		if (is_metadata) {
2006 			ret = btrfsic_process_metablock(state, block,
2007 							&block_ctx, 0, 0);
2008 			if (ret)
2009 				pr_info("btrfsic: process_metablock(root @%llu) failed!\n",
2010 				       dev_bytenr);
2011 		}
2012 		btrfsic_release_block_ctx(&block_ctx);
2013 	}
2014 
2015 continue_loop:
2016 	BUG_ON(!processed_len);
2017 	dev_bytenr += processed_len;
2018 	mapped_datav += processed_len >> PAGE_SHIFT;
2019 	num_pages -= processed_len >> PAGE_SHIFT;
2020 	goto again;
2021 }
2022 
btrfsic_bio_end_io(struct bio * bp)2023 static void btrfsic_bio_end_io(struct bio *bp)
2024 {
2025 	struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2026 	int iodone_w_error;
2027 
2028 	/* mutex is not held! This is not save if IO is not yet completed
2029 	 * on umount */
2030 	iodone_w_error = 0;
2031 	if (bp->bi_status)
2032 		iodone_w_error = 1;
2033 
2034 	BUG_ON(NULL == block);
2035 	bp->bi_private = block->orig_bio_private;
2036 	bp->bi_end_io = block->orig_bio_end_io;
2037 
2038 	do {
2039 		struct btrfsic_block *next_block;
2040 		struct btrfsic_dev_state *const dev_state = block->dev_state;
2041 
2042 		if ((dev_state->state->print_mask &
2043 		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2044 			pr_info("bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2045 			       bp->bi_status,
2046 			       btrfsic_get_block_type(dev_state->state, block),
2047 			       block->logical_bytenr, dev_state->name,
2048 			       block->dev_bytenr, block->mirror_num);
2049 		next_block = block->next_in_same_bio;
2050 		block->iodone_w_error = iodone_w_error;
2051 		if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2052 			dev_state->last_flush_gen++;
2053 			if ((dev_state->state->print_mask &
2054 			     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2055 				pr_info("bio_end_io() new %s flush_gen=%llu\n",
2056 				       dev_state->name,
2057 				       dev_state->last_flush_gen);
2058 		}
2059 		if (block->submit_bio_bh_rw & REQ_FUA)
2060 			block->flush_gen = 0; /* FUA completed means block is
2061 					       * on disk */
2062 		block->is_iodone = 1; /* for FLUSH, this releases the block */
2063 		block = next_block;
2064 	} while (NULL != block);
2065 
2066 	bp->bi_end_io(bp);
2067 }
2068 
btrfsic_process_written_superblock(struct btrfsic_state * state,struct btrfsic_block * const superblock,struct btrfs_super_block * const super_hdr)2069 static int btrfsic_process_written_superblock(
2070 		struct btrfsic_state *state,
2071 		struct btrfsic_block *const superblock,
2072 		struct btrfs_super_block *const super_hdr)
2073 {
2074 	struct btrfs_fs_info *fs_info = state->fs_info;
2075 	int pass;
2076 
2077 	superblock->generation = btrfs_super_generation(super_hdr);
2078 	if (!(superblock->generation > state->max_superblock_generation ||
2079 	      0 == state->max_superblock_generation)) {
2080 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2081 			pr_info("btrfsic: superblock @%llu (%s/%llu/%d) with old gen %llu <= %llu\n",
2082 			       superblock->logical_bytenr,
2083 			       superblock->dev_state->name,
2084 			       superblock->dev_bytenr, superblock->mirror_num,
2085 			       btrfs_super_generation(super_hdr),
2086 			       state->max_superblock_generation);
2087 	} else {
2088 		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2089 			pr_info("btrfsic: got new superblock @%llu (%s/%llu/%d) with new gen %llu > %llu\n",
2090 			       superblock->logical_bytenr,
2091 			       superblock->dev_state->name,
2092 			       superblock->dev_bytenr, superblock->mirror_num,
2093 			       btrfs_super_generation(super_hdr),
2094 			       state->max_superblock_generation);
2095 
2096 		state->max_superblock_generation =
2097 		    btrfs_super_generation(super_hdr);
2098 		state->latest_superblock = superblock;
2099 	}
2100 
2101 	for (pass = 0; pass < 3; pass++) {
2102 		int ret;
2103 		u64 next_bytenr;
2104 		struct btrfsic_block *next_block;
2105 		struct btrfsic_block_data_ctx tmp_next_block_ctx;
2106 		struct btrfsic_block_link *l;
2107 		int num_copies;
2108 		int mirror_num;
2109 		const char *additional_string = NULL;
2110 		struct btrfs_disk_key tmp_disk_key = {0};
2111 
2112 		btrfs_set_disk_key_objectid(&tmp_disk_key,
2113 					    BTRFS_ROOT_ITEM_KEY);
2114 		btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2115 
2116 		switch (pass) {
2117 		case 0:
2118 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2119 						    BTRFS_ROOT_TREE_OBJECTID);
2120 			additional_string = "root ";
2121 			next_bytenr = btrfs_super_root(super_hdr);
2122 			if (state->print_mask &
2123 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2124 				pr_info("root@%llu\n", next_bytenr);
2125 			break;
2126 		case 1:
2127 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2128 						    BTRFS_CHUNK_TREE_OBJECTID);
2129 			additional_string = "chunk ";
2130 			next_bytenr = btrfs_super_chunk_root(super_hdr);
2131 			if (state->print_mask &
2132 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2133 				pr_info("chunk@%llu\n", next_bytenr);
2134 			break;
2135 		case 2:
2136 			btrfs_set_disk_key_objectid(&tmp_disk_key,
2137 						    BTRFS_TREE_LOG_OBJECTID);
2138 			additional_string = "log ";
2139 			next_bytenr = btrfs_super_log_root(super_hdr);
2140 			if (0 == next_bytenr)
2141 				continue;
2142 			if (state->print_mask &
2143 			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2144 				pr_info("log@%llu\n", next_bytenr);
2145 			break;
2146 		}
2147 
2148 		num_copies = btrfs_num_copies(fs_info, next_bytenr,
2149 					      BTRFS_SUPER_INFO_SIZE);
2150 		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2151 			pr_info("num_copies(log_bytenr=%llu) = %d\n",
2152 			       next_bytenr, num_copies);
2153 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2154 			int was_created;
2155 
2156 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2157 				pr_info("btrfsic_process_written_superblock(mirror_num=%d)\n", mirror_num);
2158 			ret = btrfsic_map_block(state, next_bytenr,
2159 						BTRFS_SUPER_INFO_SIZE,
2160 						&tmp_next_block_ctx,
2161 						mirror_num);
2162 			if (ret) {
2163 				pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
2164 				       next_bytenr, mirror_num);
2165 				return -1;
2166 			}
2167 
2168 			next_block = btrfsic_block_lookup_or_add(
2169 					state,
2170 					&tmp_next_block_ctx,
2171 					additional_string,
2172 					1, 0, 1,
2173 					mirror_num,
2174 					&was_created);
2175 			if (NULL == next_block) {
2176 				btrfsic_release_block_ctx(&tmp_next_block_ctx);
2177 				return -1;
2178 			}
2179 
2180 			next_block->disk_key = tmp_disk_key;
2181 			if (was_created)
2182 				next_block->generation =
2183 				    BTRFSIC_GENERATION_UNKNOWN;
2184 			l = btrfsic_block_link_lookup_or_add(
2185 					state,
2186 					&tmp_next_block_ctx,
2187 					next_block,
2188 					superblock,
2189 					BTRFSIC_GENERATION_UNKNOWN);
2190 			btrfsic_release_block_ctx(&tmp_next_block_ctx);
2191 			if (NULL == l)
2192 				return -1;
2193 		}
2194 	}
2195 
2196 	if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2197 		btrfsic_dump_tree(state);
2198 
2199 	return 0;
2200 }
2201 
btrfsic_check_all_ref_blocks(struct btrfsic_state * state,struct btrfsic_block * const block,int recursion_level)2202 static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2203 					struct btrfsic_block *const block,
2204 					int recursion_level)
2205 {
2206 	const struct btrfsic_block_link *l;
2207 	int ret = 0;
2208 
2209 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2210 		/*
2211 		 * Note that this situation can happen and does not
2212 		 * indicate an error in regular cases. It happens
2213 		 * when disk blocks are freed and later reused.
2214 		 * The check-integrity module is not aware of any
2215 		 * block free operations, it just recognizes block
2216 		 * write operations. Therefore it keeps the linkage
2217 		 * information for a block until a block is
2218 		 * rewritten. This can temporarily cause incorrect
2219 		 * and even circular linkage information. This
2220 		 * causes no harm unless such blocks are referenced
2221 		 * by the most recent super block.
2222 		 */
2223 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2224 			pr_info("btrfsic: abort cyclic linkage (case 1).\n");
2225 
2226 		return ret;
2227 	}
2228 
2229 	/*
2230 	 * This algorithm is recursive because the amount of used stack
2231 	 * space is very small and the max recursion depth is limited.
2232 	 */
2233 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2234 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2235 			pr_info("rl=%d, %c @%llu (%s/%llu/%d) %u* refers to %c @%llu (%s/%llu/%d)\n",
2236 			       recursion_level,
2237 			       btrfsic_get_block_type(state, block),
2238 			       block->logical_bytenr, block->dev_state->name,
2239 			       block->dev_bytenr, block->mirror_num,
2240 			       l->ref_cnt,
2241 			       btrfsic_get_block_type(state, l->block_ref_to),
2242 			       l->block_ref_to->logical_bytenr,
2243 			       l->block_ref_to->dev_state->name,
2244 			       l->block_ref_to->dev_bytenr,
2245 			       l->block_ref_to->mirror_num);
2246 		if (l->block_ref_to->never_written) {
2247 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is never written!\n",
2248 			       btrfsic_get_block_type(state, l->block_ref_to),
2249 			       l->block_ref_to->logical_bytenr,
2250 			       l->block_ref_to->dev_state->name,
2251 			       l->block_ref_to->dev_bytenr,
2252 			       l->block_ref_to->mirror_num);
2253 			ret = -1;
2254 		} else if (!l->block_ref_to->is_iodone) {
2255 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is not yet iodone!\n",
2256 			       btrfsic_get_block_type(state, l->block_ref_to),
2257 			       l->block_ref_to->logical_bytenr,
2258 			       l->block_ref_to->dev_state->name,
2259 			       l->block_ref_to->dev_bytenr,
2260 			       l->block_ref_to->mirror_num);
2261 			ret = -1;
2262 		} else if (l->block_ref_to->iodone_w_error) {
2263 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which has write error!\n",
2264 			       btrfsic_get_block_type(state, l->block_ref_to),
2265 			       l->block_ref_to->logical_bytenr,
2266 			       l->block_ref_to->dev_state->name,
2267 			       l->block_ref_to->dev_bytenr,
2268 			       l->block_ref_to->mirror_num);
2269 			ret = -1;
2270 		} else if (l->parent_generation !=
2271 			   l->block_ref_to->generation &&
2272 			   BTRFSIC_GENERATION_UNKNOWN !=
2273 			   l->parent_generation &&
2274 			   BTRFSIC_GENERATION_UNKNOWN !=
2275 			   l->block_ref_to->generation) {
2276 			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) with generation %llu != parent generation %llu!\n",
2277 			       btrfsic_get_block_type(state, l->block_ref_to),
2278 			       l->block_ref_to->logical_bytenr,
2279 			       l->block_ref_to->dev_state->name,
2280 			       l->block_ref_to->dev_bytenr,
2281 			       l->block_ref_to->mirror_num,
2282 			       l->block_ref_to->generation,
2283 			       l->parent_generation);
2284 			ret = -1;
2285 		} else if (l->block_ref_to->flush_gen >
2286 			   l->block_ref_to->dev_state->last_flush_gen) {
2287 			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",
2288 			       btrfsic_get_block_type(state, l->block_ref_to),
2289 			       l->block_ref_to->logical_bytenr,
2290 			       l->block_ref_to->dev_state->name,
2291 			       l->block_ref_to->dev_bytenr,
2292 			       l->block_ref_to->mirror_num, block->flush_gen,
2293 			       l->block_ref_to->dev_state->last_flush_gen);
2294 			ret = -1;
2295 		} else if (-1 == btrfsic_check_all_ref_blocks(state,
2296 							      l->block_ref_to,
2297 							      recursion_level +
2298 							      1)) {
2299 			ret = -1;
2300 		}
2301 	}
2302 
2303 	return ret;
2304 }
2305 
btrfsic_is_block_ref_by_superblock(const struct btrfsic_state * state,const struct btrfsic_block * block,int recursion_level)2306 static int btrfsic_is_block_ref_by_superblock(
2307 		const struct btrfsic_state *state,
2308 		const struct btrfsic_block *block,
2309 		int recursion_level)
2310 {
2311 	const struct btrfsic_block_link *l;
2312 
2313 	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2314 		/* refer to comment at "abort cyclic linkage (case 1)" */
2315 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2316 			pr_info("btrfsic: abort cyclic linkage (case 2).\n");
2317 
2318 		return 0;
2319 	}
2320 
2321 	/*
2322 	 * This algorithm is recursive because the amount of used stack space
2323 	 * is very small and the max recursion depth is limited.
2324 	 */
2325 	list_for_each_entry(l, &block->ref_from_list, node_ref_from) {
2326 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2327 			pr_info("rl=%d, %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
2328 			       recursion_level,
2329 			       btrfsic_get_block_type(state, block),
2330 			       block->logical_bytenr, block->dev_state->name,
2331 			       block->dev_bytenr, block->mirror_num,
2332 			       l->ref_cnt,
2333 			       btrfsic_get_block_type(state, l->block_ref_from),
2334 			       l->block_ref_from->logical_bytenr,
2335 			       l->block_ref_from->dev_state->name,
2336 			       l->block_ref_from->dev_bytenr,
2337 			       l->block_ref_from->mirror_num);
2338 		if (l->block_ref_from->is_superblock &&
2339 		    state->latest_superblock->dev_bytenr ==
2340 		    l->block_ref_from->dev_bytenr &&
2341 		    state->latest_superblock->dev_state->bdev ==
2342 		    l->block_ref_from->dev_state->bdev)
2343 			return 1;
2344 		else if (btrfsic_is_block_ref_by_superblock(state,
2345 							    l->block_ref_from,
2346 							    recursion_level +
2347 							    1))
2348 			return 1;
2349 	}
2350 
2351 	return 0;
2352 }
2353 
btrfsic_print_add_link(const struct btrfsic_state * state,const struct btrfsic_block_link * l)2354 static void btrfsic_print_add_link(const struct btrfsic_state *state,
2355 				   const struct btrfsic_block_link *l)
2356 {
2357 	pr_info("Add %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2358 	       l->ref_cnt,
2359 	       btrfsic_get_block_type(state, l->block_ref_from),
2360 	       l->block_ref_from->logical_bytenr,
2361 	       l->block_ref_from->dev_state->name,
2362 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2363 	       btrfsic_get_block_type(state, l->block_ref_to),
2364 	       l->block_ref_to->logical_bytenr,
2365 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2366 	       l->block_ref_to->mirror_num);
2367 }
2368 
btrfsic_print_rem_link(const struct btrfsic_state * state,const struct btrfsic_block_link * l)2369 static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2370 				   const struct btrfsic_block_link *l)
2371 {
2372 	pr_info("Rem %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2373 	       l->ref_cnt,
2374 	       btrfsic_get_block_type(state, l->block_ref_from),
2375 	       l->block_ref_from->logical_bytenr,
2376 	       l->block_ref_from->dev_state->name,
2377 	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2378 	       btrfsic_get_block_type(state, l->block_ref_to),
2379 	       l->block_ref_to->logical_bytenr,
2380 	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2381 	       l->block_ref_to->mirror_num);
2382 }
2383 
btrfsic_get_block_type(const struct btrfsic_state * state,const struct btrfsic_block * block)2384 static char btrfsic_get_block_type(const struct btrfsic_state *state,
2385 				   const struct btrfsic_block *block)
2386 {
2387 	if (block->is_superblock &&
2388 	    state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2389 	    state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2390 		return 'S';
2391 	else if (block->is_superblock)
2392 		return 's';
2393 	else if (block->is_metadata)
2394 		return 'M';
2395 	else
2396 		return 'D';
2397 }
2398 
btrfsic_dump_tree(const struct btrfsic_state * state)2399 static void btrfsic_dump_tree(const struct btrfsic_state *state)
2400 {
2401 	btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2402 }
2403 
btrfsic_dump_tree_sub(const struct btrfsic_state * state,const struct btrfsic_block * block,int indent_level)2404 static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2405 				  const struct btrfsic_block *block,
2406 				  int indent_level)
2407 {
2408 	const struct btrfsic_block_link *l;
2409 	int indent_add;
2410 	static char buf[80];
2411 	int cursor_position;
2412 
2413 	/*
2414 	 * Should better fill an on-stack buffer with a complete line and
2415 	 * dump it at once when it is time to print a newline character.
2416 	 */
2417 
2418 	/*
2419 	 * This algorithm is recursive because the amount of used stack space
2420 	 * is very small and the max recursion depth is limited.
2421 	 */
2422 	indent_add = sprintf(buf, "%c-%llu(%s/%llu/%u)",
2423 			     btrfsic_get_block_type(state, block),
2424 			     block->logical_bytenr, block->dev_state->name,
2425 			     block->dev_bytenr, block->mirror_num);
2426 	if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2427 		printk("[...]\n");
2428 		return;
2429 	}
2430 	printk(buf);
2431 	indent_level += indent_add;
2432 	if (list_empty(&block->ref_to_list)) {
2433 		printk("\n");
2434 		return;
2435 	}
2436 	if (block->mirror_num > 1 &&
2437 	    !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2438 		printk(" [...]\n");
2439 		return;
2440 	}
2441 
2442 	cursor_position = indent_level;
2443 	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2444 		while (cursor_position < indent_level) {
2445 			printk(" ");
2446 			cursor_position++;
2447 		}
2448 		if (l->ref_cnt > 1)
2449 			indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2450 		else
2451 			indent_add = sprintf(buf, " --> ");
2452 		if (indent_level + indent_add >
2453 		    BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2454 			printk("[...]\n");
2455 			cursor_position = 0;
2456 			continue;
2457 		}
2458 
2459 		printk(buf);
2460 
2461 		btrfsic_dump_tree_sub(state, l->block_ref_to,
2462 				      indent_level + indent_add);
2463 		cursor_position = 0;
2464 	}
2465 }
2466 
btrfsic_block_link_lookup_or_add(struct btrfsic_state * state,struct btrfsic_block_data_ctx * next_block_ctx,struct btrfsic_block * next_block,struct btrfsic_block * from_block,u64 parent_generation)2467 static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2468 		struct btrfsic_state *state,
2469 		struct btrfsic_block_data_ctx *next_block_ctx,
2470 		struct btrfsic_block *next_block,
2471 		struct btrfsic_block *from_block,
2472 		u64 parent_generation)
2473 {
2474 	struct btrfsic_block_link *l;
2475 
2476 	l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2477 						next_block_ctx->dev_bytenr,
2478 						from_block->dev_state->bdev,
2479 						from_block->dev_bytenr,
2480 						&state->block_link_hashtable);
2481 	if (NULL == l) {
2482 		l = btrfsic_block_link_alloc();
2483 		if (!l)
2484 			return NULL;
2485 
2486 		l->block_ref_to = next_block;
2487 		l->block_ref_from = from_block;
2488 		l->ref_cnt = 1;
2489 		l->parent_generation = parent_generation;
2490 
2491 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2492 			btrfsic_print_add_link(state, l);
2493 
2494 		list_add(&l->node_ref_to, &from_block->ref_to_list);
2495 		list_add(&l->node_ref_from, &next_block->ref_from_list);
2496 
2497 		btrfsic_block_link_hashtable_add(l,
2498 						 &state->block_link_hashtable);
2499 	} else {
2500 		l->ref_cnt++;
2501 		l->parent_generation = parent_generation;
2502 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2503 			btrfsic_print_add_link(state, l);
2504 	}
2505 
2506 	return l;
2507 }
2508 
btrfsic_block_lookup_or_add(struct btrfsic_state * state,struct btrfsic_block_data_ctx * block_ctx,const char * additional_string,int is_metadata,int is_iodone,int never_written,int mirror_num,int * was_created)2509 static struct btrfsic_block *btrfsic_block_lookup_or_add(
2510 		struct btrfsic_state *state,
2511 		struct btrfsic_block_data_ctx *block_ctx,
2512 		const char *additional_string,
2513 		int is_metadata,
2514 		int is_iodone,
2515 		int never_written,
2516 		int mirror_num,
2517 		int *was_created)
2518 {
2519 	struct btrfsic_block *block;
2520 
2521 	block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2522 					       block_ctx->dev_bytenr,
2523 					       &state->block_hashtable);
2524 	if (NULL == block) {
2525 		struct btrfsic_dev_state *dev_state;
2526 
2527 		block = btrfsic_block_alloc();
2528 		if (!block)
2529 			return NULL;
2530 
2531 		dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev->bd_dev);
2532 		if (NULL == dev_state) {
2533 			pr_info("btrfsic: error, lookup dev_state failed!\n");
2534 			btrfsic_block_free(block);
2535 			return NULL;
2536 		}
2537 		block->dev_state = dev_state;
2538 		block->dev_bytenr = block_ctx->dev_bytenr;
2539 		block->logical_bytenr = block_ctx->start;
2540 		block->is_metadata = is_metadata;
2541 		block->is_iodone = is_iodone;
2542 		block->never_written = never_written;
2543 		block->mirror_num = mirror_num;
2544 		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2545 			pr_info("New %s%c-block @%llu (%s/%llu/%d)\n",
2546 			       additional_string,
2547 			       btrfsic_get_block_type(state, block),
2548 			       block->logical_bytenr, dev_state->name,
2549 			       block->dev_bytenr, mirror_num);
2550 		list_add(&block->all_blocks_node, &state->all_blocks_list);
2551 		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2552 		if (NULL != was_created)
2553 			*was_created = 1;
2554 	} else {
2555 		if (NULL != was_created)
2556 			*was_created = 0;
2557 	}
2558 
2559 	return block;
2560 }
2561 
btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state * state,u64 bytenr,struct btrfsic_dev_state * dev_state,u64 dev_bytenr)2562 static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2563 					   u64 bytenr,
2564 					   struct btrfsic_dev_state *dev_state,
2565 					   u64 dev_bytenr)
2566 {
2567 	struct btrfs_fs_info *fs_info = state->fs_info;
2568 	struct btrfsic_block_data_ctx block_ctx;
2569 	int num_copies;
2570 	int mirror_num;
2571 	int match = 0;
2572 	int ret;
2573 
2574 	num_copies = btrfs_num_copies(fs_info, bytenr, state->metablock_size);
2575 
2576 	for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2577 		ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2578 					&block_ctx, mirror_num);
2579 		if (ret) {
2580 			pr_info("btrfsic: btrfsic_map_block(logical @%llu, mirror %d) failed!\n",
2581 			       bytenr, mirror_num);
2582 			continue;
2583 		}
2584 
2585 		if (dev_state->bdev == block_ctx.dev->bdev &&
2586 		    dev_bytenr == block_ctx.dev_bytenr) {
2587 			match++;
2588 			btrfsic_release_block_ctx(&block_ctx);
2589 			break;
2590 		}
2591 		btrfsic_release_block_ctx(&block_ctx);
2592 	}
2593 
2594 	if (WARN_ON(!match)) {
2595 		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",
2596 		       bytenr, dev_state->name, dev_bytenr);
2597 		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2598 			ret = btrfsic_map_block(state, bytenr,
2599 						state->metablock_size,
2600 						&block_ctx, mirror_num);
2601 			if (ret)
2602 				continue;
2603 
2604 			pr_info("Read logical bytenr @%llu maps to (%s/%llu/%d)\n",
2605 			       bytenr, block_ctx.dev->name,
2606 			       block_ctx.dev_bytenr, mirror_num);
2607 		}
2608 	}
2609 }
2610 
btrfsic_dev_state_lookup(dev_t dev)2611 static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev)
2612 {
2613 	return btrfsic_dev_state_hashtable_lookup(dev,
2614 						  &btrfsic_dev_state_hashtable);
2615 }
2616 
__btrfsic_submit_bio(struct bio * bio)2617 static void __btrfsic_submit_bio(struct bio *bio)
2618 {
2619 	struct btrfsic_dev_state *dev_state;
2620 
2621 	if (!btrfsic_is_initialized)
2622 		return;
2623 
2624 	mutex_lock(&btrfsic_mutex);
2625 	/* since btrfsic_submit_bio() is also called before
2626 	 * btrfsic_mount(), this might return NULL */
2627 	dev_state = btrfsic_dev_state_lookup(bio->bi_bdev->bd_dev);
2628 	if (NULL != dev_state &&
2629 	    (bio_op(bio) == REQ_OP_WRITE) && bio_has_data(bio)) {
2630 		int i = 0;
2631 		u64 dev_bytenr;
2632 		u64 cur_bytenr;
2633 		struct bio_vec bvec;
2634 		struct bvec_iter iter;
2635 		int bio_is_patched;
2636 		char **mapped_datav;
2637 		unsigned int segs = bio_segments(bio);
2638 
2639 		dev_bytenr = 512 * bio->bi_iter.bi_sector;
2640 		bio_is_patched = 0;
2641 		if (dev_state->state->print_mask &
2642 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2643 			pr_info("submit_bio(rw=%d,0x%x, bi_vcnt=%u, bi_sector=%llu (bytenr %llu), bi_bdev=%p)\n",
2644 			       bio_op(bio), bio->bi_opf, segs,
2645 			       bio->bi_iter.bi_sector, dev_bytenr, bio->bi_bdev);
2646 
2647 		mapped_datav = kmalloc_array(segs,
2648 					     sizeof(*mapped_datav), GFP_NOFS);
2649 		if (!mapped_datav)
2650 			goto leave;
2651 		cur_bytenr = dev_bytenr;
2652 
2653 		bio_for_each_segment(bvec, bio, iter) {
2654 			BUG_ON(bvec.bv_len != PAGE_SIZE);
2655 			mapped_datav[i] = page_address(bvec.bv_page);
2656 			i++;
2657 
2658 			if (dev_state->state->print_mask &
2659 			    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
2660 				pr_info("#%u: bytenr=%llu, len=%u, offset=%u\n",
2661 				       i, cur_bytenr, bvec.bv_len, bvec.bv_offset);
2662 			cur_bytenr += bvec.bv_len;
2663 		}
2664 		btrfsic_process_written_block(dev_state, dev_bytenr,
2665 					      mapped_datav, segs,
2666 					      bio, &bio_is_patched,
2667 					      bio->bi_opf);
2668 		kfree(mapped_datav);
2669 	} else if (NULL != dev_state && (bio->bi_opf & REQ_PREFLUSH)) {
2670 		if (dev_state->state->print_mask &
2671 		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2672 			pr_info("submit_bio(rw=%d,0x%x FLUSH, bdev=%p)\n",
2673 			       bio_op(bio), bio->bi_opf, bio->bi_bdev);
2674 		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2675 			if ((dev_state->state->print_mask &
2676 			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2677 			      BTRFSIC_PRINT_MASK_VERBOSE)))
2678 				pr_info("btrfsic_submit_bio(%s) with FLUSH but dummy block already in use (ignored)!\n",
2679 				       dev_state->name);
2680 		} else {
2681 			struct btrfsic_block *const block =
2682 				&dev_state->dummy_block_for_bio_bh_flush;
2683 
2684 			block->is_iodone = 0;
2685 			block->never_written = 0;
2686 			block->iodone_w_error = 0;
2687 			block->flush_gen = dev_state->last_flush_gen + 1;
2688 			block->submit_bio_bh_rw = bio->bi_opf;
2689 			block->orig_bio_private = bio->bi_private;
2690 			block->orig_bio_end_io = bio->bi_end_io;
2691 			block->next_in_same_bio = NULL;
2692 			bio->bi_private = block;
2693 			bio->bi_end_io = btrfsic_bio_end_io;
2694 		}
2695 	}
2696 leave:
2697 	mutex_unlock(&btrfsic_mutex);
2698 }
2699 
btrfsic_submit_bio(struct bio * bio)2700 void btrfsic_submit_bio(struct bio *bio)
2701 {
2702 	__btrfsic_submit_bio(bio);
2703 	submit_bio(bio);
2704 }
2705 
btrfsic_submit_bio_wait(struct bio * bio)2706 int btrfsic_submit_bio_wait(struct bio *bio)
2707 {
2708 	__btrfsic_submit_bio(bio);
2709 	return submit_bio_wait(bio);
2710 }
2711 
btrfsic_mount(struct btrfs_fs_info * fs_info,struct btrfs_fs_devices * fs_devices,int including_extent_data,u32 print_mask)2712 int btrfsic_mount(struct btrfs_fs_info *fs_info,
2713 		  struct btrfs_fs_devices *fs_devices,
2714 		  int including_extent_data, u32 print_mask)
2715 {
2716 	int ret;
2717 	struct btrfsic_state *state;
2718 	struct list_head *dev_head = &fs_devices->devices;
2719 	struct btrfs_device *device;
2720 
2721 	if (!PAGE_ALIGNED(fs_info->nodesize)) {
2722 		pr_info("btrfsic: cannot handle nodesize %d not being a multiple of PAGE_SIZE %ld!\n",
2723 		       fs_info->nodesize, PAGE_SIZE);
2724 		return -1;
2725 	}
2726 	if (!PAGE_ALIGNED(fs_info->sectorsize)) {
2727 		pr_info("btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_SIZE %ld!\n",
2728 		       fs_info->sectorsize, PAGE_SIZE);
2729 		return -1;
2730 	}
2731 	state = kvzalloc(sizeof(*state), GFP_KERNEL);
2732 	if (!state)
2733 		return -ENOMEM;
2734 
2735 	if (!btrfsic_is_initialized) {
2736 		mutex_init(&btrfsic_mutex);
2737 		btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
2738 		btrfsic_is_initialized = 1;
2739 	}
2740 	mutex_lock(&btrfsic_mutex);
2741 	state->fs_info = fs_info;
2742 	state->print_mask = print_mask;
2743 	state->include_extent_data = including_extent_data;
2744 	state->metablock_size = fs_info->nodesize;
2745 	state->datablock_size = fs_info->sectorsize;
2746 	INIT_LIST_HEAD(&state->all_blocks_list);
2747 	btrfsic_block_hashtable_init(&state->block_hashtable);
2748 	btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
2749 	state->max_superblock_generation = 0;
2750 	state->latest_superblock = NULL;
2751 
2752 	list_for_each_entry(device, dev_head, dev_list) {
2753 		struct btrfsic_dev_state *ds;
2754 		const char *p;
2755 
2756 		if (!device->bdev || !device->name)
2757 			continue;
2758 
2759 		ds = btrfsic_dev_state_alloc();
2760 		if (NULL == ds) {
2761 			mutex_unlock(&btrfsic_mutex);
2762 			return -ENOMEM;
2763 		}
2764 		ds->bdev = device->bdev;
2765 		ds->state = state;
2766 		bdevname(ds->bdev, ds->name);
2767 		ds->name[BDEVNAME_SIZE - 1] = '\0';
2768 		p = kbasename(ds->name);
2769 		strlcpy(ds->name, p, sizeof(ds->name));
2770 		btrfsic_dev_state_hashtable_add(ds,
2771 						&btrfsic_dev_state_hashtable);
2772 	}
2773 
2774 	ret = btrfsic_process_superblock(state, fs_devices);
2775 	if (0 != ret) {
2776 		mutex_unlock(&btrfsic_mutex);
2777 		btrfsic_unmount(fs_devices);
2778 		return ret;
2779 	}
2780 
2781 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
2782 		btrfsic_dump_database(state);
2783 	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
2784 		btrfsic_dump_tree(state);
2785 
2786 	mutex_unlock(&btrfsic_mutex);
2787 	return 0;
2788 }
2789 
btrfsic_unmount(struct btrfs_fs_devices * fs_devices)2790 void btrfsic_unmount(struct btrfs_fs_devices *fs_devices)
2791 {
2792 	struct btrfsic_block *b_all, *tmp_all;
2793 	struct btrfsic_state *state;
2794 	struct list_head *dev_head = &fs_devices->devices;
2795 	struct btrfs_device *device;
2796 
2797 	if (!btrfsic_is_initialized)
2798 		return;
2799 
2800 	mutex_lock(&btrfsic_mutex);
2801 
2802 	state = NULL;
2803 	list_for_each_entry(device, dev_head, dev_list) {
2804 		struct btrfsic_dev_state *ds;
2805 
2806 		if (!device->bdev || !device->name)
2807 			continue;
2808 
2809 		ds = btrfsic_dev_state_hashtable_lookup(
2810 				device->bdev->bd_dev,
2811 				&btrfsic_dev_state_hashtable);
2812 		if (NULL != ds) {
2813 			state = ds->state;
2814 			btrfsic_dev_state_hashtable_remove(ds);
2815 			btrfsic_dev_state_free(ds);
2816 		}
2817 	}
2818 
2819 	if (NULL == state) {
2820 		pr_info("btrfsic: error, cannot find state information on umount!\n");
2821 		mutex_unlock(&btrfsic_mutex);
2822 		return;
2823 	}
2824 
2825 	/*
2826 	 * Don't care about keeping the lists' state up to date,
2827 	 * just free all memory that was allocated dynamically.
2828 	 * Free the blocks and the block_links.
2829 	 */
2830 	list_for_each_entry_safe(b_all, tmp_all, &state->all_blocks_list,
2831 				 all_blocks_node) {
2832 		struct btrfsic_block_link *l, *tmp;
2833 
2834 		list_for_each_entry_safe(l, tmp, &b_all->ref_to_list,
2835 					 node_ref_to) {
2836 			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2837 				btrfsic_print_rem_link(state, l);
2838 
2839 			l->ref_cnt--;
2840 			if (0 == l->ref_cnt)
2841 				btrfsic_block_link_free(l);
2842 		}
2843 
2844 		if (b_all->is_iodone || b_all->never_written)
2845 			btrfsic_block_free(b_all);
2846 		else
2847 			pr_info("btrfs: attempt to free %c-block @%llu (%s/%llu/%d) on umount which is not yet iodone!\n",
2848 			       btrfsic_get_block_type(state, b_all),
2849 			       b_all->logical_bytenr, b_all->dev_state->name,
2850 			       b_all->dev_bytenr, b_all->mirror_num);
2851 	}
2852 
2853 	mutex_unlock(&btrfsic_mutex);
2854 
2855 	kvfree(state);
2856 }
2857