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