Linux Audio

Check our new training course

Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) STRATO AG 2011.  All rights reserved.
   4 */
   5
   6/*
   7 * This module can be used to catch cases when the btrfs kernel
   8 * code executes write requests to the disk that bring the file
   9 * system in an inconsistent state. In such a state, a power-loss
  10 * or kernel panic event would cause that the data on disk is
  11 * lost or at least damaged.
  12 *
  13 * Code is added that examines all block write requests during
  14 * runtime (including writes of the super block). Three rules
  15 * are verified and an error is printed on violation of the
  16 * rules:
  17 * 1. It is not allowed to write a disk block which is
  18 *    currently referenced by the super block (either directly
  19 *    or indirectly).
  20 * 2. When a super block is written, it is verified that all
  21 *    referenced (directly or indirectly) blocks fulfill the
  22 *    following requirements:
  23 *    2a. All referenced blocks have either been present when
  24 *        the file system was mounted, (i.e., they have been
  25 *        referenced by the super block) or they have been
  26 *        written since then and the write completion callback
  27 *        was called and no write error was indicated and a
  28 *        FLUSH request to the device where these blocks are
  29 *        located was received and completed.
  30 *    2b. All referenced blocks need to have a generation
  31 *        number which is equal to the parent's number.
  32 *
  33 * One issue that was found using this module was that the log
  34 * tree on disk became temporarily corrupted because disk blocks
  35 * that had been in use for the log tree had been freed and
  36 * reused too early, while being referenced by the written super
  37 * block.
  38 *
  39 * The search term in the kernel log that can be used to filter
  40 * on the existence of detected integrity issues is
  41 * "btrfs: attempt".
  42 *
  43 * The integrity check is enabled via mount options. These
  44 * mount options are only supported if the integrity check
  45 * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY.
  46 *
  47 * Example #1, apply integrity checks to all metadata:
  48 * mount /dev/sdb1 /mnt -o check_int
  49 *
  50 * Example #2, apply integrity checks to all metadata and
  51 * to data extents:
  52 * mount /dev/sdb1 /mnt -o check_int_data
  53 *
  54 * Example #3, apply integrity checks to all metadata and dump
  55 * the tree that the super block references to kernel messages
  56 * each time after a super block was written:
  57 * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263
  58 *
  59 * If the integrity check tool is included and activated in
  60 * the mount options, plenty of kernel memory is used, and
  61 * plenty of additional CPU cycles are spent. Enabling this
  62 * functionality is not intended for normal use. In most
  63 * cases, unless you are a btrfs developer who needs to verify
  64 * the integrity of (super)-block write requests, do not
  65 * enable the config option BTRFS_FS_CHECK_INTEGRITY to
  66 * include and compile the integrity check tool.
  67 *
  68 * Expect millions of lines of information in the kernel log with an
  69 * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the
  70 * kernel config to at least 26 (which is 64MB). Usually the value is
  71 * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be
  72 * changed like this before LOG_BUF_SHIFT can be set to a high value:
  73 * config LOG_BUF_SHIFT
  74 *       int "Kernel log buffer size (16 => 64KB, 17 => 128KB)"
  75 *       range 12 30
  76 */
  77
  78#include <linux/sched.h>
  79#include <linux/slab.h>
  80#include <linux/mutex.h>
 
  81#include <linux/blkdev.h>
  82#include <linux/mm.h>
  83#include <linux/string.h>
  84#include <crypto/hash.h>
  85#include "messages.h"
  86#include "ctree.h"
  87#include "disk-io.h"
  88#include "transaction.h"
  89#include "extent_io.h"
  90#include "volumes.h"
  91#include "print-tree.h"
  92#include "locking.h"
  93#include "check-integrity.h"
  94#include "rcu-string.h"
  95#include "compression.h"
  96#include "accessors.h"
  97
  98#define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
  99#define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
 100#define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
 101#define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
 102#define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
 103#define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
 104#define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
 105#define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6)	/* in characters,
 106							 * excluding " [...]" */
 107#define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
 108
 109/*
 110 * The definition of the bitmask fields for the print_mask.
 111 * They are specified with the mount option check_integrity_print_mask.
 112 */
 113#define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE			0x00000001
 114#define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION		0x00000002
 115#define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE			0x00000004
 116#define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE			0x00000008
 117#define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH			0x00000010
 118#define BTRFSIC_PRINT_MASK_END_IO_BIO_BH			0x00000020
 119#define BTRFSIC_PRINT_MASK_VERBOSE				0x00000040
 120#define BTRFSIC_PRINT_MASK_VERY_VERBOSE				0x00000080
 121#define BTRFSIC_PRINT_MASK_INITIAL_TREE				0x00000100
 122#define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES			0x00000200
 123#define BTRFSIC_PRINT_MASK_INITIAL_DATABASE			0x00000400
 124#define BTRFSIC_PRINT_MASK_NUM_COPIES				0x00000800
 125#define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS		0x00001000
 126#define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE		0x00002000
 127
 128struct btrfsic_dev_state;
 129struct btrfsic_state;
 130
 131struct btrfsic_block {
 132	u32 magic_num;		/* only used for debug purposes */
 133	unsigned int is_metadata:1;	/* if it is meta-data, not data-data */
 134	unsigned int is_superblock:1;	/* if it is one of the superblocks */
 135	unsigned int is_iodone:1;	/* if is done by lower subsystem */
 136	unsigned int iodone_w_error:1;	/* error was indicated to endio */
 137	unsigned int never_written:1;	/* block was added because it was
 138					 * referenced, not because it was
 139					 * written */
 140	unsigned int mirror_num;	/* large enough to hold
 141					 * BTRFS_SUPER_MIRROR_MAX */
 142	struct btrfsic_dev_state *dev_state;
 143	u64 dev_bytenr;		/* key, physical byte num on disk */
 144	u64 logical_bytenr;	/* logical byte num on disk */
 145	u64 generation;
 146	struct btrfs_disk_key disk_key;	/* extra info to print in case of
 147					 * issues, will not always be correct */
 148	struct list_head collision_resolving_node;	/* list node */
 149	struct list_head all_blocks_node;	/* list node */
 150
 151	/* the following two lists contain block_link items */
 152	struct list_head ref_to_list;	/* list */
 153	struct list_head ref_from_list;	/* list */
 154	struct btrfsic_block *next_in_same_bio;
 155	void *orig_bio_private;
 156	bio_end_io_t *orig_bio_end_io;
 157	blk_opf_t submit_bio_bh_rw;
 158	u64 flush_gen; /* only valid if !never_written */
 159};
 160
 161/*
 162 * Elements of this type are allocated dynamically and required because
 163 * each block object can refer to and can be ref from multiple blocks.
 164 * The key to lookup them in the hashtable is the dev_bytenr of
 165 * the block ref to plus the one from the block referred from.
 166 * The fact that they are searchable via a hashtable and that a
 167 * ref_cnt is maintained is not required for the btrfs integrity
 168 * check algorithm itself, it is only used to make the output more
 169 * beautiful in case that an error is detected (an error is defined
 170 * as a write operation to a block while that block is still referenced).
 171 */
 172struct btrfsic_block_link {
 173	u32 magic_num;		/* only used for debug purposes */
 174	u32 ref_cnt;
 175	struct list_head node_ref_to;	/* list node */
 176	struct list_head node_ref_from;	/* list node */
 177	struct list_head collision_resolving_node;	/* list node */
 178	struct btrfsic_block *block_ref_to;
 179	struct btrfsic_block *block_ref_from;
 180	u64 parent_generation;
 181};
 182
 183struct btrfsic_dev_state {
 184	u32 magic_num;		/* only used for debug purposes */
 185	struct block_device *bdev;
 186	struct btrfsic_state *state;
 187	struct list_head collision_resolving_node;	/* list node */
 188	struct btrfsic_block dummy_block_for_bio_bh_flush;
 189	u64 last_flush_gen;
 
 190};
 191
 192struct btrfsic_block_hashtable {
 193	struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
 194};
 195
 196struct btrfsic_block_link_hashtable {
 197	struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
 198};
 199
 200struct btrfsic_dev_state_hashtable {
 201	struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
 202};
 203
 204struct btrfsic_block_data_ctx {
 205	u64 start;		/* virtual bytenr */
 206	u64 dev_bytenr;		/* physical bytenr on device */
 207	u32 len;
 208	struct btrfsic_dev_state *dev;
 209	char **datav;
 210	struct page **pagev;
 211	void *mem_to_free;
 212};
 213
 214/* This structure is used to implement recursion without occupying
 215 * any stack space, refer to btrfsic_process_metablock() */
 216struct btrfsic_stack_frame {
 217	u32 magic;
 218	u32 nr;
 219	int error;
 220	int i;
 221	int limit_nesting;
 222	int num_copies;
 223	int mirror_num;
 224	struct btrfsic_block *block;
 225	struct btrfsic_block_data_ctx *block_ctx;
 226	struct btrfsic_block *next_block;
 227	struct btrfsic_block_data_ctx next_block_ctx;
 228	struct btrfs_header *hdr;
 229	struct btrfsic_stack_frame *prev;
 230};
 231
 232/* Some state per mounted filesystem */
 233struct btrfsic_state {
 234	u32 print_mask;
 235	int include_extent_data;
 
 236	struct list_head all_blocks_list;
 237	struct btrfsic_block_hashtable block_hashtable;
 238	struct btrfsic_block_link_hashtable block_link_hashtable;
 239	struct btrfs_fs_info *fs_info;
 240	u64 max_superblock_generation;
 241	struct btrfsic_block *latest_superblock;
 242	u32 metablock_size;
 243	u32 datablock_size;
 244};
 245
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 246static int btrfsic_process_metablock(struct btrfsic_state *state,
 247				     struct btrfsic_block *block,
 248				     struct btrfsic_block_data_ctx *block_ctx,
 249				     int limit_nesting, int force_iodone_flag);
 250static void btrfsic_read_from_block_data(
 251	struct btrfsic_block_data_ctx *block_ctx,
 252	void *dst, u32 offset, size_t len);
 253static int btrfsic_create_link_to_next_block(
 254		struct btrfsic_state *state,
 255		struct btrfsic_block *block,
 256		struct btrfsic_block_data_ctx
 257		*block_ctx, u64 next_bytenr,
 258		int limit_nesting,
 259		struct btrfsic_block_data_ctx *next_block_ctx,
 260		struct btrfsic_block **next_blockp,
 261		int force_iodone_flag,
 262		int *num_copiesp, int *mirror_nump,
 263		struct btrfs_disk_key *disk_key,
 264		u64 parent_generation);
 265static int btrfsic_handle_extent_data(struct btrfsic_state *state,
 266				      struct btrfsic_block *block,
 267				      struct btrfsic_block_data_ctx *block_ctx,
 268				      u32 item_offset, int force_iodone_flag);
 269static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
 270			     struct btrfsic_block_data_ctx *block_ctx_out,
 271			     int mirror_num);
 272static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
 273static int btrfsic_read_block(struct btrfsic_state *state,
 274			      struct btrfsic_block_data_ctx *block_ctx);
 
 
 
 
 
 
 
 
 275static int btrfsic_process_written_superblock(
 276		struct btrfsic_state *state,
 277		struct btrfsic_block *const block,
 278		struct btrfs_super_block *const super_hdr);
 279static void btrfsic_bio_end_io(struct bio *bp);
 280static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
 281					      const struct btrfsic_block *block,
 282					      int recursion_level);
 283static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
 284					struct btrfsic_block *const block,
 285					int recursion_level);
 286static void btrfsic_print_add_link(const struct btrfsic_state *state,
 287				   const struct btrfsic_block_link *l);
 288static void btrfsic_print_rem_link(const struct btrfsic_state *state,
 289				   const struct btrfsic_block_link *l);
 290static char btrfsic_get_block_type(const struct btrfsic_state *state,
 291				   const struct btrfsic_block *block);
 292static void btrfsic_dump_tree(const struct btrfsic_state *state);
 293static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
 294				  const struct btrfsic_block *block,
 295				  int indent_level);
 296static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
 297		struct btrfsic_state *state,
 298		struct btrfsic_block_data_ctx *next_block_ctx,
 299		struct btrfsic_block *next_block,
 300		struct btrfsic_block *from_block,
 301		u64 parent_generation);
 302static struct btrfsic_block *btrfsic_block_lookup_or_add(
 303		struct btrfsic_state *state,
 304		struct btrfsic_block_data_ctx *block_ctx,
 305		const char *additional_string,
 306		int is_metadata,
 307		int is_iodone,
 308		int never_written,
 309		int mirror_num,
 310		int *was_created);
 311static int btrfsic_process_superblock_dev_mirror(
 312		struct btrfsic_state *state,
 313		struct btrfsic_dev_state *dev_state,
 314		struct btrfs_device *device,
 315		int superblock_mirror_num,
 316		struct btrfsic_dev_state **selected_dev_state,
 317		struct btrfs_super_block *selected_super);
 318static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev);
 319static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
 320					   u64 bytenr,
 321					   struct btrfsic_dev_state *dev_state,
 322					   u64 dev_bytenr);
 323
 324static struct mutex btrfsic_mutex;
 325static int btrfsic_is_initialized;
 326static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
 327
 328
 329static void btrfsic_block_init(struct btrfsic_block *b)
 330{
 331	b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
 332	b->dev_state = NULL;
 333	b->dev_bytenr = 0;
 334	b->logical_bytenr = 0;
 335	b->generation = BTRFSIC_GENERATION_UNKNOWN;
 336	b->disk_key.objectid = 0;
 337	b->disk_key.type = 0;
 338	b->disk_key.offset = 0;
 339	b->is_metadata = 0;
 340	b->is_superblock = 0;
 341	b->is_iodone = 0;
 342	b->iodone_w_error = 0;
 343	b->never_written = 0;
 344	b->mirror_num = 0;
 345	b->next_in_same_bio = NULL;
 346	b->orig_bio_private = NULL;
 347	b->orig_bio_end_io = NULL;
 348	INIT_LIST_HEAD(&b->collision_resolving_node);
 349	INIT_LIST_HEAD(&b->all_blocks_node);
 350	INIT_LIST_HEAD(&b->ref_to_list);
 351	INIT_LIST_HEAD(&b->ref_from_list);
 352	b->submit_bio_bh_rw = 0;
 353	b->flush_gen = 0;
 354}
 355
 356static struct btrfsic_block *btrfsic_block_alloc(void)
 357{
 358	struct btrfsic_block *b;
 359
 360	b = kzalloc(sizeof(*b), GFP_NOFS);
 361	if (NULL != b)
 362		btrfsic_block_init(b);
 363
 364	return b;
 365}
 366
 367static void btrfsic_block_free(struct btrfsic_block *b)
 368{
 369	BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
 370	kfree(b);
 371}
 372
 373static void btrfsic_block_link_init(struct btrfsic_block_link *l)
 374{
 375	l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
 376	l->ref_cnt = 1;
 377	INIT_LIST_HEAD(&l->node_ref_to);
 378	INIT_LIST_HEAD(&l->node_ref_from);
 379	INIT_LIST_HEAD(&l->collision_resolving_node);
 380	l->block_ref_to = NULL;
 381	l->block_ref_from = NULL;
 382}
 383
 384static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
 385{
 386	struct btrfsic_block_link *l;
 387
 388	l = kzalloc(sizeof(*l), GFP_NOFS);
 389	if (NULL != l)
 390		btrfsic_block_link_init(l);
 391
 392	return l;
 393}
 394
 395static void btrfsic_block_link_free(struct btrfsic_block_link *l)
 396{
 397	BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
 398	kfree(l);
 399}
 400
 401static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
 402{
 403	ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
 404	ds->bdev = NULL;
 405	ds->state = NULL;
 
 406	INIT_LIST_HEAD(&ds->collision_resolving_node);
 407	ds->last_flush_gen = 0;
 408	btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
 409	ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
 410	ds->dummy_block_for_bio_bh_flush.dev_state = ds;
 411}
 412
 413static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
 414{
 415	struct btrfsic_dev_state *ds;
 416
 417	ds = kzalloc(sizeof(*ds), GFP_NOFS);
 418	if (NULL != ds)
 419		btrfsic_dev_state_init(ds);
 420
 421	return ds;
 422}
 423
 424static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
 425{
 426	BUG_ON(!(NULL == ds ||
 427		 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
 428	kfree(ds);
 429}
 430
 431static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
 432{
 433	int i;
 434
 435	for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
 436		INIT_LIST_HEAD(h->table + i);
 437}
 438
 439static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
 440					struct btrfsic_block_hashtable *h)
 441{
 442	const unsigned int hashval =
 443	    (((unsigned int)(b->dev_bytenr >> 16)) ^
 444	     ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
 445	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
 446
 447	list_add(&b->collision_resolving_node, h->table + hashval);
 448}
 449
 450static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
 451{
 452	list_del(&b->collision_resolving_node);
 453}
 454
 455static struct btrfsic_block *btrfsic_block_hashtable_lookup(
 456		struct block_device *bdev,
 457		u64 dev_bytenr,
 458		struct btrfsic_block_hashtable *h)
 459{
 460	const unsigned int hashval =
 461	    (((unsigned int)(dev_bytenr >> 16)) ^
 462	     ((unsigned int)((uintptr_t)bdev))) &
 463	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
 464	struct btrfsic_block *b;
 465
 466	list_for_each_entry(b, h->table + hashval, collision_resolving_node) {
 467		if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
 468			return b;
 469	}
 470
 471	return NULL;
 472}
 473
 474static void btrfsic_block_link_hashtable_init(
 475		struct btrfsic_block_link_hashtable *h)
 476{
 477	int i;
 478
 479	for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
 480		INIT_LIST_HEAD(h->table + i);
 481}
 482
 483static void btrfsic_block_link_hashtable_add(
 484		struct btrfsic_block_link *l,
 485		struct btrfsic_block_link_hashtable *h)
 486{
 487	const unsigned int hashval =
 488	    (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
 489	     ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
 490	     ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
 491	     ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
 492	     & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
 493
 494	BUG_ON(NULL == l->block_ref_to);
 495	BUG_ON(NULL == l->block_ref_from);
 496	list_add(&l->collision_resolving_node, h->table + hashval);
 497}
 498
 499static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
 500{
 501	list_del(&l->collision_resolving_node);
 502}
 503
 504static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
 505		struct block_device *bdev_ref_to,
 506		u64 dev_bytenr_ref_to,
 507		struct block_device *bdev_ref_from,
 508		u64 dev_bytenr_ref_from,
 509		struct btrfsic_block_link_hashtable *h)
 510{
 511	const unsigned int hashval =
 512	    (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
 513	     ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
 514	     ((unsigned int)((uintptr_t)bdev_ref_to)) ^
 515	     ((unsigned int)((uintptr_t)bdev_ref_from))) &
 516	     (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
 517	struct btrfsic_block_link *l;
 518
 519	list_for_each_entry(l, h->table + hashval, collision_resolving_node) {
 520		BUG_ON(NULL == l->block_ref_to);
 521		BUG_ON(NULL == l->block_ref_from);
 522		if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
 523		    l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
 524		    l->block_ref_from->dev_state->bdev == bdev_ref_from &&
 525		    l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
 526			return l;
 527	}
 528
 529	return NULL;
 530}
 531
 532static void btrfsic_dev_state_hashtable_init(
 533		struct btrfsic_dev_state_hashtable *h)
 534{
 535	int i;
 536
 537	for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
 538		INIT_LIST_HEAD(h->table + i);
 539}
 540
 541static void btrfsic_dev_state_hashtable_add(
 542		struct btrfsic_dev_state *ds,
 543		struct btrfsic_dev_state_hashtable *h)
 544{
 545	const unsigned int hashval =
 546	    (((unsigned int)((uintptr_t)ds->bdev->bd_dev)) &
 547	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
 548
 549	list_add(&ds->collision_resolving_node, h->table + hashval);
 550}
 551
 552static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
 553{
 554	list_del(&ds->collision_resolving_node);
 555}
 556
 557static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev,
 558		struct btrfsic_dev_state_hashtable *h)
 559{
 560	const unsigned int hashval =
 561		dev & (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1);
 562	struct btrfsic_dev_state *ds;
 563
 564	list_for_each_entry(ds, h->table + hashval, collision_resolving_node) {
 565		if (ds->bdev->bd_dev == dev)
 566			return ds;
 567	}
 568
 569	return NULL;
 570}
 571
 572static int btrfsic_process_superblock(struct btrfsic_state *state,
 573				      struct btrfs_fs_devices *fs_devices)
 574{
 575	struct btrfs_super_block *selected_super;
 576	struct list_head *dev_head = &fs_devices->devices;
 577	struct btrfs_device *device;
 578	struct btrfsic_dev_state *selected_dev_state = NULL;
 579	int ret = 0;
 580	int pass;
 581
 582	selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
 583	if (!selected_super)
 584		return -ENOMEM;
 585
 586	list_for_each_entry(device, dev_head, dev_list) {
 587		int i;
 588		struct btrfsic_dev_state *dev_state;
 589
 590		if (!device->bdev || !device->name)
 591			continue;
 592
 593		dev_state = btrfsic_dev_state_lookup(device->bdev->bd_dev);
 594		BUG_ON(NULL == dev_state);
 595		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 596			ret = btrfsic_process_superblock_dev_mirror(
 597					state, dev_state, device, i,
 598					&selected_dev_state, selected_super);
 599			if (0 != ret && 0 == i) {
 600				kfree(selected_super);
 601				return ret;
 602			}
 603		}
 604	}
 605
 606	if (NULL == state->latest_superblock) {
 607		pr_info("btrfsic: no superblock found!\n");
 608		kfree(selected_super);
 609		return -1;
 610	}
 611
 
 
 612	for (pass = 0; pass < 3; pass++) {
 613		int num_copies;
 614		int mirror_num;
 615		u64 next_bytenr;
 616
 617		switch (pass) {
 618		case 0:
 619			next_bytenr = btrfs_super_root(selected_super);
 620			if (state->print_mask &
 621			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
 622				pr_info("root@%llu\n", next_bytenr);
 623			break;
 624		case 1:
 625			next_bytenr = btrfs_super_chunk_root(selected_super);
 626			if (state->print_mask &
 627			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
 628				pr_info("chunk@%llu\n", next_bytenr);
 629			break;
 630		case 2:
 631			next_bytenr = btrfs_super_log_root(selected_super);
 632			if (0 == next_bytenr)
 633				continue;
 634			if (state->print_mask &
 635			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
 636				pr_info("log@%llu\n", next_bytenr);
 637			break;
 638		}
 639
 640		num_copies = btrfs_num_copies(state->fs_info, next_bytenr,
 641					      state->metablock_size);
 642		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
 643			pr_info("num_copies(log_bytenr=%llu) = %d\n",
 644			       next_bytenr, num_copies);
 645
 646		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
 647			struct btrfsic_block *next_block;
 648			struct btrfsic_block_data_ctx tmp_next_block_ctx;
 649			struct btrfsic_block_link *l;
 650
 651			ret = btrfsic_map_block(state, next_bytenr,
 652						state->metablock_size,
 653						&tmp_next_block_ctx,
 654						mirror_num);
 655			if (ret) {
 656				pr_info("btrfsic: btrfsic_map_block(root @%llu, mirror %d) failed!\n",
 657				       next_bytenr, mirror_num);
 658				kfree(selected_super);
 659				return -1;
 660			}
 661
 662			next_block = btrfsic_block_hashtable_lookup(
 663					tmp_next_block_ctx.dev->bdev,
 664					tmp_next_block_ctx.dev_bytenr,
 665					&state->block_hashtable);
 666			BUG_ON(NULL == next_block);
 667
 668			l = btrfsic_block_link_hashtable_lookup(
 669					tmp_next_block_ctx.dev->bdev,
 670					tmp_next_block_ctx.dev_bytenr,
 671					state->latest_superblock->dev_state->
 672					bdev,
 673					state->latest_superblock->dev_bytenr,
 674					&state->block_link_hashtable);
 675			BUG_ON(NULL == l);
 676
 677			ret = btrfsic_read_block(state, &tmp_next_block_ctx);
 678			if (ret < (int)PAGE_SIZE) {
 679				pr_info("btrfsic: read @logical %llu failed!\n",
 680				       tmp_next_block_ctx.start);
 681				btrfsic_release_block_ctx(&tmp_next_block_ctx);
 682				kfree(selected_super);
 683				return -1;
 684			}
 685
 686			ret = btrfsic_process_metablock(state,
 687							next_block,
 688							&tmp_next_block_ctx,
 689							BTRFS_MAX_LEVEL + 3, 1);
 690			btrfsic_release_block_ctx(&tmp_next_block_ctx);
 691		}
 692	}
 693
 694	kfree(selected_super);
 695	return ret;
 696}
 697
 698static int btrfsic_process_superblock_dev_mirror(
 699		struct btrfsic_state *state,
 700		struct btrfsic_dev_state *dev_state,
 701		struct btrfs_device *device,
 702		int superblock_mirror_num,
 703		struct btrfsic_dev_state **selected_dev_state,
 704		struct btrfs_super_block *selected_super)
 705{
 706	struct btrfs_fs_info *fs_info = state->fs_info;
 707	struct btrfs_super_block *super_tmp;
 708	u64 dev_bytenr;
 709	struct btrfsic_block *superblock_tmp;
 710	int pass;
 711	struct block_device *const superblock_bdev = device->bdev;
 712	struct page *page;
 713	struct address_space *mapping = superblock_bdev->bd_inode->i_mapping;
 714	int ret = 0;
 715
 716	/* super block bytenr is always the unmapped device bytenr */
 717	dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
 718	if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->commit_total_bytes)
 719		return -1;
 720
 721	page = read_cache_page_gfp(mapping, dev_bytenr >> PAGE_SHIFT, GFP_NOFS);
 722	if (IS_ERR(page))
 723		return -1;
 724
 725	super_tmp = page_address(page);
 726
 727	if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
 728	    btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
 729	    memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
 730	    btrfs_super_nodesize(super_tmp) != state->metablock_size ||
 731	    btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
 732		ret = 0;
 733		goto out;
 734	}
 735
 736	superblock_tmp =
 737	    btrfsic_block_hashtable_lookup(superblock_bdev,
 738					   dev_bytenr,
 739					   &state->block_hashtable);
 740	if (NULL == superblock_tmp) {
 741		superblock_tmp = btrfsic_block_alloc();
 742		if (NULL == superblock_tmp) {
 743			ret = -1;
 744			goto out;
 745		}
 746		/* for superblock, only the dev_bytenr makes sense */
 747		superblock_tmp->dev_bytenr = dev_bytenr;
 748		superblock_tmp->dev_state = dev_state;
 749		superblock_tmp->logical_bytenr = dev_bytenr;
 750		superblock_tmp->generation = btrfs_super_generation(super_tmp);
 751		superblock_tmp->is_metadata = 1;
 752		superblock_tmp->is_superblock = 1;
 753		superblock_tmp->is_iodone = 1;
 754		superblock_tmp->never_written = 0;
 755		superblock_tmp->mirror_num = 1 + superblock_mirror_num;
 756		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
 757			btrfs_info_in_rcu(fs_info,
 758			"new initial S-block (bdev %p, %s) @%llu (%pg/%llu/%d)",
 759				     superblock_bdev,
 760				     btrfs_dev_name(device), dev_bytenr,
 761				     dev_state->bdev, dev_bytenr,
 762				     superblock_mirror_num);
 763		list_add(&superblock_tmp->all_blocks_node,
 764			 &state->all_blocks_list);
 765		btrfsic_block_hashtable_add(superblock_tmp,
 766					    &state->block_hashtable);
 767	}
 768
 769	/* select the one with the highest generation field */
 770	if (btrfs_super_generation(super_tmp) >
 771	    state->max_superblock_generation ||
 772	    0 == state->max_superblock_generation) {
 773		memcpy(selected_super, super_tmp, sizeof(*selected_super));
 774		*selected_dev_state = dev_state;
 775		state->max_superblock_generation =
 776		    btrfs_super_generation(super_tmp);
 777		state->latest_superblock = superblock_tmp;
 778	}
 779
 780	for (pass = 0; pass < 3; pass++) {
 781		u64 next_bytenr;
 782		int num_copies;
 783		int mirror_num;
 784		const char *additional_string = NULL;
 785		struct btrfs_disk_key tmp_disk_key;
 786
 787		tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
 788		tmp_disk_key.offset = 0;
 789		switch (pass) {
 790		case 0:
 791			btrfs_set_disk_key_objectid(&tmp_disk_key,
 792						    BTRFS_ROOT_TREE_OBJECTID);
 793			additional_string = "initial root ";
 794			next_bytenr = btrfs_super_root(super_tmp);
 795			break;
 796		case 1:
 797			btrfs_set_disk_key_objectid(&tmp_disk_key,
 798						    BTRFS_CHUNK_TREE_OBJECTID);
 799			additional_string = "initial chunk ";
 800			next_bytenr = btrfs_super_chunk_root(super_tmp);
 801			break;
 802		case 2:
 803			btrfs_set_disk_key_objectid(&tmp_disk_key,
 804						    BTRFS_TREE_LOG_OBJECTID);
 805			additional_string = "initial log ";
 806			next_bytenr = btrfs_super_log_root(super_tmp);
 807			if (0 == next_bytenr)
 808				continue;
 809			break;
 810		}
 811
 812		num_copies = btrfs_num_copies(fs_info, next_bytenr,
 813					      state->metablock_size);
 814		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
 815			pr_info("num_copies(log_bytenr=%llu) = %d\n",
 816			       next_bytenr, num_copies);
 817		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
 818			struct btrfsic_block *next_block;
 819			struct btrfsic_block_data_ctx tmp_next_block_ctx;
 820			struct btrfsic_block_link *l;
 821
 822			if (btrfsic_map_block(state, next_bytenr,
 823					      state->metablock_size,
 824					      &tmp_next_block_ctx,
 825					      mirror_num)) {
 826				pr_info("btrfsic: btrfsic_map_block(bytenr @%llu, mirror %d) failed!\n",
 827				       next_bytenr, mirror_num);
 828				ret = -1;
 829				goto out;
 830			}
 831
 832			next_block = btrfsic_block_lookup_or_add(
 833					state, &tmp_next_block_ctx,
 834					additional_string, 1, 1, 0,
 835					mirror_num, NULL);
 836			if (NULL == next_block) {
 837				btrfsic_release_block_ctx(&tmp_next_block_ctx);
 838				ret = -1;
 839				goto out;
 840			}
 841
 842			next_block->disk_key = tmp_disk_key;
 843			next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
 844			l = btrfsic_block_link_lookup_or_add(
 845					state, &tmp_next_block_ctx,
 846					next_block, superblock_tmp,
 847					BTRFSIC_GENERATION_UNKNOWN);
 848			btrfsic_release_block_ctx(&tmp_next_block_ctx);
 849			if (NULL == l) {
 850				ret = -1;
 851				goto out;
 852			}
 853		}
 854	}
 855	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
 856		btrfsic_dump_tree_sub(state, superblock_tmp, 0);
 857
 858out:
 859	put_page(page);
 860	return ret;
 861}
 862
 863static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
 864{
 865	struct btrfsic_stack_frame *sf;
 866
 867	sf = kzalloc(sizeof(*sf), GFP_NOFS);
 868	if (sf)
 869		sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
 870	return sf;
 871}
 872
 873static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
 874{
 875	BUG_ON(!(NULL == sf ||
 876		 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
 877	kfree(sf);
 878}
 879
 880static noinline_for_stack int btrfsic_process_metablock(
 881		struct btrfsic_state *state,
 882		struct btrfsic_block *const first_block,
 883		struct btrfsic_block_data_ctx *const first_block_ctx,
 884		int first_limit_nesting, int force_iodone_flag)
 885{
 886	struct btrfsic_stack_frame initial_stack_frame = { 0 };
 887	struct btrfsic_stack_frame *sf;
 888	struct btrfsic_stack_frame *next_stack;
 889	struct btrfs_header *const first_hdr =
 890		(struct btrfs_header *)first_block_ctx->datav[0];
 891
 892	BUG_ON(!first_hdr);
 893	sf = &initial_stack_frame;
 894	sf->error = 0;
 895	sf->i = -1;
 896	sf->limit_nesting = first_limit_nesting;
 897	sf->block = first_block;
 898	sf->block_ctx = first_block_ctx;
 899	sf->next_block = NULL;
 900	sf->hdr = first_hdr;
 901	sf->prev = NULL;
 902
 903continue_with_new_stack_frame:
 904	sf->block->generation = btrfs_stack_header_generation(sf->hdr);
 905	if (0 == sf->hdr->level) {
 906		struct btrfs_leaf *const leafhdr =
 907		    (struct btrfs_leaf *)sf->hdr;
 908
 909		if (-1 == sf->i) {
 910			sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
 911
 912			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
 913				pr_info("leaf %llu items %d generation %llu owner %llu\n",
 914				       sf->block_ctx->start, sf->nr,
 915				       btrfs_stack_header_generation(
 916					       &leafhdr->header),
 917				       btrfs_stack_header_owner(
 918					       &leafhdr->header));
 919		}
 920
 921continue_with_current_leaf_stack_frame:
 922		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
 923			sf->i++;
 924			sf->num_copies = 0;
 925		}
 926
 927		if (sf->i < sf->nr) {
 928			struct btrfs_item disk_item;
 929			u32 disk_item_offset =
 930				(uintptr_t)(leafhdr->items + sf->i) -
 931				(uintptr_t)leafhdr;
 932			struct btrfs_disk_key *disk_key;
 933			u8 type;
 934			u32 item_offset;
 935			u32 item_size;
 936
 937			if (disk_item_offset + sizeof(struct btrfs_item) >
 938			    sf->block_ctx->len) {
 939leaf_item_out_of_bounce_error:
 940				pr_info(
 941		"btrfsic: leaf item out of bounce at logical %llu, dev %pg\n",
 942				       sf->block_ctx->start,
 943				       sf->block_ctx->dev->bdev);
 944				goto one_stack_frame_backwards;
 945			}
 946			btrfsic_read_from_block_data(sf->block_ctx,
 947						     &disk_item,
 948						     disk_item_offset,
 949						     sizeof(struct btrfs_item));
 950			item_offset = btrfs_stack_item_offset(&disk_item);
 951			item_size = btrfs_stack_item_size(&disk_item);
 952			disk_key = &disk_item.key;
 953			type = btrfs_disk_key_type(disk_key);
 954
 955			if (BTRFS_ROOT_ITEM_KEY == type) {
 956				struct btrfs_root_item root_item;
 957				u32 root_item_offset;
 958				u64 next_bytenr;
 959
 960				root_item_offset = item_offset +
 961					offsetof(struct btrfs_leaf, items);
 962				if (root_item_offset + item_size >
 963				    sf->block_ctx->len)
 964					goto leaf_item_out_of_bounce_error;
 965				btrfsic_read_from_block_data(
 966					sf->block_ctx, &root_item,
 967					root_item_offset,
 968					item_size);
 969				next_bytenr = btrfs_root_bytenr(&root_item);
 970
 971				sf->error =
 972				    btrfsic_create_link_to_next_block(
 973						state,
 974						sf->block,
 975						sf->block_ctx,
 976						next_bytenr,
 977						sf->limit_nesting,
 978						&sf->next_block_ctx,
 979						&sf->next_block,
 980						force_iodone_flag,
 981						&sf->num_copies,
 982						&sf->mirror_num,
 983						disk_key,
 984						btrfs_root_generation(
 985						&root_item));
 986				if (sf->error)
 987					goto one_stack_frame_backwards;
 988
 989				if (NULL != sf->next_block) {
 990					struct btrfs_header *const next_hdr =
 991					    (struct btrfs_header *)
 992					    sf->next_block_ctx.datav[0];
 993
 994					next_stack =
 995					    btrfsic_stack_frame_alloc();
 996					if (NULL == next_stack) {
 997						sf->error = -1;
 998						btrfsic_release_block_ctx(
 999								&sf->
1000								next_block_ctx);
1001						goto one_stack_frame_backwards;
1002					}
1003
1004					next_stack->i = -1;
1005					next_stack->block = sf->next_block;
1006					next_stack->block_ctx =
1007					    &sf->next_block_ctx;
1008					next_stack->next_block = NULL;
1009					next_stack->hdr = next_hdr;
1010					next_stack->limit_nesting =
1011					    sf->limit_nesting - 1;
1012					next_stack->prev = sf;
1013					sf = next_stack;
1014					goto continue_with_new_stack_frame;
1015				}
1016			} else if (BTRFS_EXTENT_DATA_KEY == type &&
1017				   state->include_extent_data) {
1018				sf->error = btrfsic_handle_extent_data(
1019						state,
1020						sf->block,
1021						sf->block_ctx,
1022						item_offset,
1023						force_iodone_flag);
1024				if (sf->error)
1025					goto one_stack_frame_backwards;
1026			}
1027
1028			goto continue_with_current_leaf_stack_frame;
1029		}
1030	} else {
1031		struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1032
1033		if (-1 == sf->i) {
1034			sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1035
1036			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1037				pr_info("node %llu level %d items %d generation %llu owner %llu\n",
1038				       sf->block_ctx->start,
1039				       nodehdr->header.level, sf->nr,
1040				       btrfs_stack_header_generation(
1041				       &nodehdr->header),
1042				       btrfs_stack_header_owner(
1043				       &nodehdr->header));
1044		}
1045
1046continue_with_current_node_stack_frame:
1047		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1048			sf->i++;
1049			sf->num_copies = 0;
1050		}
1051
1052		if (sf->i < sf->nr) {
1053			struct btrfs_key_ptr key_ptr;
1054			u32 key_ptr_offset;
1055			u64 next_bytenr;
1056
1057			key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1058					  (uintptr_t)nodehdr;
1059			if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1060			    sf->block_ctx->len) {
1061				pr_info(
1062		"btrfsic: node item out of bounce at logical %llu, dev %pg\n",
1063				       sf->block_ctx->start,
1064				       sf->block_ctx->dev->bdev);
1065				goto one_stack_frame_backwards;
1066			}
1067			btrfsic_read_from_block_data(
1068				sf->block_ctx, &key_ptr, key_ptr_offset,
1069				sizeof(struct btrfs_key_ptr));
1070			next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1071
1072			sf->error = btrfsic_create_link_to_next_block(
1073					state,
1074					sf->block,
1075					sf->block_ctx,
1076					next_bytenr,
1077					sf->limit_nesting,
1078					&sf->next_block_ctx,
1079					&sf->next_block,
1080					force_iodone_flag,
1081					&sf->num_copies,
1082					&sf->mirror_num,
1083					&key_ptr.key,
1084					btrfs_stack_key_generation(&key_ptr));
1085			if (sf->error)
1086				goto one_stack_frame_backwards;
1087
1088			if (NULL != sf->next_block) {
1089				struct btrfs_header *const next_hdr =
1090				    (struct btrfs_header *)
1091				    sf->next_block_ctx.datav[0];
1092
1093				next_stack = btrfsic_stack_frame_alloc();
1094				if (NULL == next_stack) {
1095					sf->error = -1;
1096					goto one_stack_frame_backwards;
1097				}
1098
1099				next_stack->i = -1;
1100				next_stack->block = sf->next_block;
1101				next_stack->block_ctx = &sf->next_block_ctx;
1102				next_stack->next_block = NULL;
1103				next_stack->hdr = next_hdr;
1104				next_stack->limit_nesting =
1105				    sf->limit_nesting - 1;
1106				next_stack->prev = sf;
1107				sf = next_stack;
1108				goto continue_with_new_stack_frame;
1109			}
1110
1111			goto continue_with_current_node_stack_frame;
1112		}
1113	}
1114
1115one_stack_frame_backwards:
1116	if (NULL != sf->prev) {
1117		struct btrfsic_stack_frame *const prev = sf->prev;
1118
1119		/* the one for the initial block is freed in the caller */
1120		btrfsic_release_block_ctx(sf->block_ctx);
1121
1122		if (sf->error) {
1123			prev->error = sf->error;
1124			btrfsic_stack_frame_free(sf);
1125			sf = prev;
1126			goto one_stack_frame_backwards;
1127		}
1128
1129		btrfsic_stack_frame_free(sf);
1130		sf = prev;
1131		goto continue_with_new_stack_frame;
1132	} else {
1133		BUG_ON(&initial_stack_frame != sf);
1134	}
1135
1136	return sf->error;
1137}
1138
1139static void btrfsic_read_from_block_data(
1140	struct btrfsic_block_data_ctx *block_ctx,
1141	void *dstv, u32 offset, size_t len)
1142{
1143	size_t cur;
1144	size_t pgoff;
1145	char *kaddr;
1146	char *dst = (char *)dstv;
1147	size_t start_offset = offset_in_page(block_ctx->start);
1148	unsigned long i = (start_offset + offset) >> PAGE_SHIFT;
1149
1150	WARN_ON(offset + len > block_ctx->len);
1151	pgoff = offset_in_page(start_offset + offset);
1152
1153	while (len > 0) {
1154		cur = min(len, ((size_t)PAGE_SIZE - pgoff));
1155		BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_SIZE));
1156		kaddr = block_ctx->datav[i];
1157		memcpy(dst, kaddr + pgoff, cur);
1158
1159		dst += cur;
1160		len -= cur;
1161		pgoff = 0;
1162		i++;
1163	}
1164}
1165
1166static int btrfsic_create_link_to_next_block(
1167		struct btrfsic_state *state,
1168		struct btrfsic_block *block,
1169		struct btrfsic_block_data_ctx *block_ctx,
1170		u64 next_bytenr,
1171		int limit_nesting,
1172		struct btrfsic_block_data_ctx *next_block_ctx,
1173		struct btrfsic_block **next_blockp,
1174		int force_iodone_flag,
1175		int *num_copiesp, int *mirror_nump,
1176		struct btrfs_disk_key *disk_key,
1177		u64 parent_generation)
1178{
1179	struct btrfs_fs_info *fs_info = state->fs_info;
1180	struct btrfsic_block *next_block = NULL;
1181	int ret;
1182	struct btrfsic_block_link *l;
1183	int did_alloc_block_link;
1184	int block_was_created;
1185
1186	*next_blockp = NULL;
1187	if (0 == *num_copiesp) {
1188		*num_copiesp = btrfs_num_copies(fs_info, next_bytenr,
1189						state->metablock_size);
1190		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1191			pr_info("num_copies(log_bytenr=%llu) = %d\n",
1192			       next_bytenr, *num_copiesp);
1193		*mirror_nump = 1;
1194	}
1195
1196	if (*mirror_nump > *num_copiesp)
1197		return 0;
1198
1199	if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1200		pr_info("btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1201		       *mirror_nump);
1202	ret = btrfsic_map_block(state, next_bytenr,
1203				state->metablock_size,
1204				next_block_ctx, *mirror_nump);
1205	if (ret) {
1206		pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1207		       next_bytenr, *mirror_nump);
1208		btrfsic_release_block_ctx(next_block_ctx);
1209		*next_blockp = NULL;
1210		return -1;
1211	}
1212
1213	next_block = btrfsic_block_lookup_or_add(state,
1214						 next_block_ctx, "referenced ",
1215						 1, force_iodone_flag,
1216						 !force_iodone_flag,
1217						 *mirror_nump,
1218						 &block_was_created);
1219	if (NULL == next_block) {
1220		btrfsic_release_block_ctx(next_block_ctx);
1221		*next_blockp = NULL;
1222		return -1;
1223	}
1224	if (block_was_created) {
1225		l = NULL;
1226		next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1227	} else {
1228		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1229			if (next_block->logical_bytenr != next_bytenr &&
1230			    !(!next_block->is_metadata &&
1231			      0 == next_block->logical_bytenr))
1232				pr_info(
1233"referenced block @%llu (%pg/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu)\n",
1234				       next_bytenr, next_block_ctx->dev->bdev,
1235				       next_block_ctx->dev_bytenr, *mirror_nump,
1236				       btrfsic_get_block_type(state,
1237							      next_block),
1238				       next_block->logical_bytenr);
1239			else
1240				pr_info(
1241		"referenced block @%llu (%pg/%llu/%d) found in hash table, %c\n",
1242				       next_bytenr, next_block_ctx->dev->bdev,
1243				       next_block_ctx->dev_bytenr, *mirror_nump,
1244				       btrfsic_get_block_type(state,
1245							      next_block));
1246		}
1247		next_block->logical_bytenr = next_bytenr;
1248
1249		next_block->mirror_num = *mirror_nump;
1250		l = btrfsic_block_link_hashtable_lookup(
1251				next_block_ctx->dev->bdev,
1252				next_block_ctx->dev_bytenr,
1253				block_ctx->dev->bdev,
1254				block_ctx->dev_bytenr,
1255				&state->block_link_hashtable);
1256	}
1257
1258	next_block->disk_key = *disk_key;
1259	if (NULL == l) {
1260		l = btrfsic_block_link_alloc();
1261		if (NULL == l) {
1262			btrfsic_release_block_ctx(next_block_ctx);
1263			*next_blockp = NULL;
1264			return -1;
1265		}
1266
1267		did_alloc_block_link = 1;
1268		l->block_ref_to = next_block;
1269		l->block_ref_from = block;
1270		l->ref_cnt = 1;
1271		l->parent_generation = parent_generation;
1272
1273		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1274			btrfsic_print_add_link(state, l);
1275
1276		list_add(&l->node_ref_to, &block->ref_to_list);
1277		list_add(&l->node_ref_from, &next_block->ref_from_list);
1278
1279		btrfsic_block_link_hashtable_add(l,
1280						 &state->block_link_hashtable);
1281	} else {
1282		did_alloc_block_link = 0;
1283		if (0 == limit_nesting) {
1284			l->ref_cnt++;
1285			l->parent_generation = parent_generation;
1286			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1287				btrfsic_print_add_link(state, l);
1288		}
1289	}
1290
1291	if (limit_nesting > 0 && did_alloc_block_link) {
1292		ret = btrfsic_read_block(state, next_block_ctx);
1293		if (ret < (int)next_block_ctx->len) {
1294			pr_info("btrfsic: read block @logical %llu failed!\n",
1295			       next_bytenr);
1296			btrfsic_release_block_ctx(next_block_ctx);
1297			*next_blockp = NULL;
1298			return -1;
1299		}
1300
1301		*next_blockp = next_block;
1302	} else {
1303		*next_blockp = NULL;
1304	}
1305	(*mirror_nump)++;
1306
1307	return 0;
1308}
1309
1310static int btrfsic_handle_extent_data(
1311		struct btrfsic_state *state,
1312		struct btrfsic_block *block,
1313		struct btrfsic_block_data_ctx *block_ctx,
1314		u32 item_offset, int force_iodone_flag)
1315{
1316	struct btrfs_fs_info *fs_info = state->fs_info;
1317	struct btrfs_file_extent_item file_extent_item;
1318	u64 file_extent_item_offset;
1319	u64 next_bytenr;
1320	u64 num_bytes;
1321	u64 generation;
1322	struct btrfsic_block_link *l;
1323	int ret;
1324
1325	file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1326				  item_offset;
1327	if (file_extent_item_offset +
1328	    offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1329	    block_ctx->len) {
1330		pr_info("btrfsic: file item out of bounce at logical %llu, dev %pg\n",
1331		       block_ctx->start, block_ctx->dev->bdev);
1332		return -1;
1333	}
1334
1335	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1336		file_extent_item_offset,
1337		offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1338	if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1339	    btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1340		if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1341			pr_info("extent_data: type %u, disk_bytenr = %llu\n",
1342			       file_extent_item.type,
1343			       btrfs_stack_file_extent_disk_bytenr(
1344			       &file_extent_item));
1345		return 0;
1346	}
1347
1348	if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1349	    block_ctx->len) {
1350		pr_info("btrfsic: file item out of bounce at logical %llu, dev %pg\n",
1351		       block_ctx->start, block_ctx->dev->bdev);
1352		return -1;
1353	}
1354	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1355				     file_extent_item_offset,
1356				     sizeof(struct btrfs_file_extent_item));
1357	next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item);
1358	if (btrfs_stack_file_extent_compression(&file_extent_item) ==
1359	    BTRFS_COMPRESS_NONE) {
1360		next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item);
1361		num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1362	} else {
1363		num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item);
1364	}
1365	generation = btrfs_stack_file_extent_generation(&file_extent_item);
1366
1367	if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1368		pr_info("extent_data: type %u, disk_bytenr = %llu, offset = %llu, num_bytes = %llu\n",
1369		       file_extent_item.type,
1370		       btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1371		       btrfs_stack_file_extent_offset(&file_extent_item),
1372		       num_bytes);
1373	while (num_bytes > 0) {
1374		u32 chunk_len;
1375		int num_copies;
1376		int mirror_num;
1377
1378		if (num_bytes > state->datablock_size)
1379			chunk_len = state->datablock_size;
1380		else
1381			chunk_len = num_bytes;
1382
1383		num_copies = btrfs_num_copies(fs_info, next_bytenr,
1384					      state->datablock_size);
1385		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1386			pr_info("num_copies(log_bytenr=%llu) = %d\n",
1387			       next_bytenr, num_copies);
1388		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1389			struct btrfsic_block_data_ctx next_block_ctx;
1390			struct btrfsic_block *next_block;
1391			int block_was_created;
1392
1393			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1394				pr_info("btrfsic_handle_extent_data(mirror_num=%d)\n",
1395					mirror_num);
1396			if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1397				pr_info("\tdisk_bytenr = %llu, num_bytes %u\n",
1398				       next_bytenr, chunk_len);
1399			ret = btrfsic_map_block(state, next_bytenr,
1400						chunk_len, &next_block_ctx,
1401						mirror_num);
1402			if (ret) {
1403				pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1404				       next_bytenr, mirror_num);
1405				return -1;
1406			}
1407
1408			next_block = btrfsic_block_lookup_or_add(
1409					state,
1410					&next_block_ctx,
1411					"referenced ",
1412					0,
1413					force_iodone_flag,
1414					!force_iodone_flag,
1415					mirror_num,
1416					&block_was_created);
1417			if (NULL == next_block) {
1418				btrfsic_release_block_ctx(&next_block_ctx);
1419				return -1;
1420			}
1421			if (!block_was_created) {
1422				if ((state->print_mask &
1423				     BTRFSIC_PRINT_MASK_VERBOSE) &&
1424				    next_block->logical_bytenr != next_bytenr &&
1425				    !(!next_block->is_metadata &&
1426				      0 == next_block->logical_bytenr)) {
1427					pr_info(
1428"referenced block @%llu (%pg/%llu/%d) found in hash table, D, bytenr mismatch (!= stored %llu)\n",
1429					       next_bytenr,
1430					       next_block_ctx.dev->bdev,
1431					       next_block_ctx.dev_bytenr,
1432					       mirror_num,
1433					       next_block->logical_bytenr);
1434				}
1435				next_block->logical_bytenr = next_bytenr;
1436				next_block->mirror_num = mirror_num;
1437			}
1438
1439			l = btrfsic_block_link_lookup_or_add(state,
1440							     &next_block_ctx,
1441							     next_block, block,
1442							     generation);
1443			btrfsic_release_block_ctx(&next_block_ctx);
1444			if (NULL == l)
1445				return -1;
1446		}
1447
1448		next_bytenr += chunk_len;
1449		num_bytes -= chunk_len;
1450	}
1451
1452	return 0;
1453}
1454
1455static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1456			     struct btrfsic_block_data_ctx *block_ctx_out,
1457			     int mirror_num)
1458{
1459	struct btrfs_fs_info *fs_info = state->fs_info;
1460	int ret;
1461	u64 length;
1462	struct btrfs_io_context *multi = NULL;
1463	struct btrfs_device *device;
1464
1465	length = len;
1466	ret = btrfs_map_block(fs_info, BTRFS_MAP_READ,
1467			      bytenr, &length, &multi, mirror_num);
1468
1469	if (ret) {
1470		block_ctx_out->start = 0;
1471		block_ctx_out->dev_bytenr = 0;
1472		block_ctx_out->len = 0;
1473		block_ctx_out->dev = NULL;
1474		block_ctx_out->datav = NULL;
1475		block_ctx_out->pagev = NULL;
1476		block_ctx_out->mem_to_free = NULL;
1477
1478		return ret;
1479	}
1480
1481	device = multi->stripes[0].dev;
1482	if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
1483	    !device->bdev || !device->name)
1484		block_ctx_out->dev = NULL;
1485	else
1486		block_ctx_out->dev = btrfsic_dev_state_lookup(
1487							device->bdev->bd_dev);
1488	block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1489	block_ctx_out->start = bytenr;
1490	block_ctx_out->len = len;
1491	block_ctx_out->datav = NULL;
1492	block_ctx_out->pagev = NULL;
1493	block_ctx_out->mem_to_free = NULL;
1494
1495	kfree(multi);
1496	if (NULL == block_ctx_out->dev) {
1497		ret = -ENXIO;
1498		pr_info("btrfsic: error, cannot lookup dev (#1)!\n");
1499	}
1500
1501	return ret;
1502}
1503
1504static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1505{
1506	if (block_ctx->mem_to_free) {
1507		unsigned int num_pages;
1508
1509		BUG_ON(!block_ctx->datav);
1510		BUG_ON(!block_ctx->pagev);
1511		num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1512			    PAGE_SHIFT;
1513		/* Pages must be unmapped in reverse order */
1514		while (num_pages > 0) {
1515			num_pages--;
1516			if (block_ctx->datav[num_pages])
 
1517				block_ctx->datav[num_pages] = NULL;
 
1518			if (block_ctx->pagev[num_pages]) {
1519				__free_page(block_ctx->pagev[num_pages]);
1520				block_ctx->pagev[num_pages] = NULL;
1521			}
1522		}
1523
1524		kfree(block_ctx->mem_to_free);
1525		block_ctx->mem_to_free = NULL;
1526		block_ctx->pagev = NULL;
1527		block_ctx->datav = NULL;
1528	}
1529}
1530
1531static int btrfsic_read_block(struct btrfsic_state *state,
1532			      struct btrfsic_block_data_ctx *block_ctx)
1533{
1534	unsigned int num_pages;
1535	unsigned int i;
1536	size_t size;
1537	u64 dev_bytenr;
1538	int ret;
1539
1540	BUG_ON(block_ctx->datav);
1541	BUG_ON(block_ctx->pagev);
1542	BUG_ON(block_ctx->mem_to_free);
1543	if (!PAGE_ALIGNED(block_ctx->dev_bytenr)) {
1544		pr_info("btrfsic: read_block() with unaligned bytenr %llu\n",
1545		       block_ctx->dev_bytenr);
1546		return -1;
1547	}
1548
1549	num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1550		    PAGE_SHIFT;
1551	size = sizeof(*block_ctx->datav) + sizeof(*block_ctx->pagev);
1552	block_ctx->mem_to_free = kcalloc(num_pages, size, GFP_NOFS);
1553	if (!block_ctx->mem_to_free)
1554		return -ENOMEM;
1555	block_ctx->datav = block_ctx->mem_to_free;
1556	block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1557	ret = btrfs_alloc_page_array(num_pages, block_ctx->pagev);
1558	if (ret)
1559		return ret;
 
 
1560
1561	dev_bytenr = block_ctx->dev_bytenr;
1562	for (i = 0; i < num_pages;) {
1563		struct bio *bio;
1564		unsigned int j;
1565
1566		bio = bio_alloc(block_ctx->dev->bdev, num_pages - i,
1567				REQ_OP_READ, GFP_NOFS);
1568		bio->bi_iter.bi_sector = dev_bytenr >> 9;
 
1569
1570		for (j = i; j < num_pages; j++) {
1571			ret = bio_add_page(bio, block_ctx->pagev[j],
1572					   PAGE_SIZE, 0);
1573			if (PAGE_SIZE != ret)
1574				break;
1575		}
1576		if (j == i) {
1577			pr_info("btrfsic: error, failed to add a single page!\n");
1578			return -1;
1579		}
1580		if (submit_bio_wait(bio)) {
1581			pr_info("btrfsic: read error at logical %llu dev %pg!\n",
1582			       block_ctx->start, block_ctx->dev->bdev);
1583			bio_put(bio);
1584			return -1;
1585		}
1586		bio_put(bio);
1587		dev_bytenr += (j - i) * PAGE_SIZE;
1588		i = j;
1589	}
1590	for (i = 0; i < num_pages; i++)
1591		block_ctx->datav[i] = page_address(block_ctx->pagev[i]);
1592
1593	return block_ctx->len;
1594}
1595
1596static void btrfsic_dump_database(struct btrfsic_state *state)
1597{
1598	const struct btrfsic_block *b_all;
1599
1600	BUG_ON(NULL == state);
1601
1602	pr_info("all_blocks_list:\n");
1603	list_for_each_entry(b_all, &state->all_blocks_list, all_blocks_node) {
1604		const struct btrfsic_block_link *l;
1605
1606		pr_info("%c-block @%llu (%pg/%llu/%d)\n",
1607		       btrfsic_get_block_type(state, b_all),
1608		       b_all->logical_bytenr, b_all->dev_state->bdev,
1609		       b_all->dev_bytenr, b_all->mirror_num);
1610
1611		list_for_each_entry(l, &b_all->ref_to_list, node_ref_to) {
1612			pr_info(
1613		" %c @%llu (%pg/%llu/%d) refers %u* to %c @%llu (%pg/%llu/%d)\n",
1614			       btrfsic_get_block_type(state, b_all),
1615			       b_all->logical_bytenr, b_all->dev_state->bdev,
1616			       b_all->dev_bytenr, b_all->mirror_num,
1617			       l->ref_cnt,
1618			       btrfsic_get_block_type(state, l->block_ref_to),
1619			       l->block_ref_to->logical_bytenr,
1620			       l->block_ref_to->dev_state->bdev,
1621			       l->block_ref_to->dev_bytenr,
1622			       l->block_ref_to->mirror_num);
1623		}
1624
1625		list_for_each_entry(l, &b_all->ref_from_list, node_ref_from) {
1626			pr_info(
1627		" %c @%llu (%pg/%llu/%d) is ref %u* from %c @%llu (%pg/%llu/%d)\n",
1628			       btrfsic_get_block_type(state, b_all),
1629			       b_all->logical_bytenr, b_all->dev_state->bdev,
1630			       b_all->dev_bytenr, b_all->mirror_num,
1631			       l->ref_cnt,
1632			       btrfsic_get_block_type(state, l->block_ref_from),
1633			       l->block_ref_from->logical_bytenr,
1634			       l->block_ref_from->dev_state->bdev,
1635			       l->block_ref_from->dev_bytenr,
1636			       l->block_ref_from->mirror_num);
1637		}
1638
1639		pr_info("\n");
1640	}
1641}
1642
1643/*
1644 * Test whether the disk block contains a tree block (leaf or node)
1645 * (note that this test fails for the super block)
1646 */
1647static noinline_for_stack int btrfsic_test_for_metadata(
1648		struct btrfsic_state *state,
1649		char **datav, unsigned int num_pages)
1650{
1651	struct btrfs_fs_info *fs_info = state->fs_info;
1652	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
1653	struct btrfs_header *h;
1654	u8 csum[BTRFS_CSUM_SIZE];
1655	unsigned int i;
1656
1657	if (num_pages * PAGE_SIZE < state->metablock_size)
1658		return 1; /* not metadata */
1659	num_pages = state->metablock_size >> PAGE_SHIFT;
1660	h = (struct btrfs_header *)datav[0];
1661
1662	if (memcmp(h->fsid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE))
1663		return 1;
1664
1665	shash->tfm = fs_info->csum_shash;
1666	crypto_shash_init(shash);
1667
1668	for (i = 0; i < num_pages; i++) {
1669		u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1670		size_t sublen = i ? PAGE_SIZE :
1671				    (PAGE_SIZE - BTRFS_CSUM_SIZE);
1672
1673		crypto_shash_update(shash, data, sublen);
1674	}
1675	crypto_shash_final(shash, csum);
1676	if (memcmp(csum, h->csum, fs_info->csum_size))
1677		return 1;
1678
1679	return 0; /* is metadata */
1680}
1681
1682static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1683					  u64 dev_bytenr, char **mapped_datav,
1684					  unsigned int num_pages,
1685					  struct bio *bio, int *bio_is_patched,
1686					  blk_opf_t submit_bio_bh_rw)
1687{
1688	int is_metadata;
1689	struct btrfsic_block *block;
1690	struct btrfsic_block_data_ctx block_ctx;
1691	int ret;
1692	struct btrfsic_state *state = dev_state->state;
1693	struct block_device *bdev = dev_state->bdev;
1694	unsigned int processed_len;
1695
1696	if (NULL != bio_is_patched)
1697		*bio_is_patched = 0;
1698
1699again:
1700	if (num_pages == 0)
1701		return;
1702
1703	processed_len = 0;
1704	is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1705						      num_pages));
1706
1707	block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1708					       &state->block_hashtable);
1709	if (NULL != block) {
1710		u64 bytenr = 0;
1711		struct btrfsic_block_link *l, *tmp;
1712
1713		if (block->is_superblock) {
1714			bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1715						    mapped_datav[0]);
1716			if (num_pages * PAGE_SIZE <
1717			    BTRFS_SUPER_INFO_SIZE) {
1718				pr_info("btrfsic: cannot work with too short bios!\n");
1719				return;
1720			}
1721			is_metadata = 1;
1722			BUG_ON(!PAGE_ALIGNED(BTRFS_SUPER_INFO_SIZE));
1723			processed_len = BTRFS_SUPER_INFO_SIZE;
1724			if (state->print_mask &
1725			    BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1726				pr_info("[before new superblock is written]:\n");
1727				btrfsic_dump_tree_sub(state, block, 0);
1728			}
1729		}
1730		if (is_metadata) {
1731			if (!block->is_superblock) {
1732				if (num_pages * PAGE_SIZE <
1733				    state->metablock_size) {
1734					pr_info("btrfsic: cannot work with too short bios!\n");
1735					return;
1736				}
1737				processed_len = state->metablock_size;
1738				bytenr = btrfs_stack_header_bytenr(
1739						(struct btrfs_header *)
1740						mapped_datav[0]);
1741				btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1742							       dev_state,
1743							       dev_bytenr);
1744			}
1745			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1746				if (block->logical_bytenr != bytenr &&
1747				    !(!block->is_metadata &&
1748				      block->logical_bytenr == 0))
1749					pr_info(
1750"written block @%llu (%pg/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu)\n",
1751					       bytenr, dev_state->bdev,
1752					       dev_bytenr,
1753					       block->mirror_num,
1754					       btrfsic_get_block_type(state,
1755								      block),
1756					       block->logical_bytenr);
1757				else
1758					pr_info(
1759		"written block @%llu (%pg/%llu/%d) found in hash table, %c\n",
1760					       bytenr, dev_state->bdev,
1761					       dev_bytenr, block->mirror_num,
1762					       btrfsic_get_block_type(state,
1763								      block));
1764			}
1765			block->logical_bytenr = bytenr;
1766		} else {
1767			if (num_pages * PAGE_SIZE <
1768			    state->datablock_size) {
1769				pr_info("btrfsic: cannot work with too short bios!\n");
1770				return;
1771			}
1772			processed_len = state->datablock_size;
1773			bytenr = block->logical_bytenr;
1774			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1775				pr_info(
1776		"written block @%llu (%pg/%llu/%d) found in hash table, %c\n",
1777				       bytenr, dev_state->bdev, dev_bytenr,
1778				       block->mirror_num,
1779				       btrfsic_get_block_type(state, block));
1780		}
1781
1782		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1783			pr_info("ref_to_list: %cE, ref_from_list: %cE\n",
1784			       list_empty(&block->ref_to_list) ? ' ' : '!',
1785			       list_empty(&block->ref_from_list) ? ' ' : '!');
1786		if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1787			pr_info(
1788"btrfs: attempt to overwrite %c-block @%llu (%pg/%llu/%d), old(gen=%llu, objectid=%llu, type=%d, offset=%llu), new(gen=%llu), which is referenced by most recent superblock (superblockgen=%llu)!\n",
1789			       btrfsic_get_block_type(state, block), bytenr,
1790			       dev_state->bdev, dev_bytenr, block->mirror_num,
1791			       block->generation,
1792			       btrfs_disk_key_objectid(&block->disk_key),
1793			       block->disk_key.type,
1794			       btrfs_disk_key_offset(&block->disk_key),
1795			       btrfs_stack_header_generation(
1796				       (struct btrfs_header *) mapped_datav[0]),
1797			       state->max_superblock_generation);
1798			btrfsic_dump_tree(state);
1799		}
1800
1801		if (!block->is_iodone && !block->never_written) {
1802			pr_info(
1803"btrfs: attempt to overwrite %c-block @%llu (%pg/%llu/%d), oldgen=%llu, newgen=%llu, which is not yet iodone!\n",
1804			       btrfsic_get_block_type(state, block), bytenr,
1805			       dev_state->bdev, dev_bytenr, block->mirror_num,
1806			       block->generation,
1807			       btrfs_stack_header_generation(
1808				       (struct btrfs_header *)
1809				       mapped_datav[0]));
1810			/* it would not be safe to go on */
1811			btrfsic_dump_tree(state);
1812			goto continue_loop;
1813		}
1814
1815		/*
1816		 * Clear all references of this block. Do not free
1817		 * the block itself even if is not referenced anymore
1818		 * because it still carries valuable information
1819		 * like whether it was ever written and IO completed.
1820		 */
1821		list_for_each_entry_safe(l, tmp, &block->ref_to_list,
1822					 node_ref_to) {
1823			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1824				btrfsic_print_rem_link(state, l);
1825			l->ref_cnt--;
1826			if (0 == l->ref_cnt) {
1827				list_del(&l->node_ref_to);
1828				list_del(&l->node_ref_from);
1829				btrfsic_block_link_hashtable_remove(l);
1830				btrfsic_block_link_free(l);
1831			}
1832		}
1833
1834		block_ctx.dev = dev_state;
1835		block_ctx.dev_bytenr = dev_bytenr;
1836		block_ctx.start = bytenr;
1837		block_ctx.len = processed_len;
1838		block_ctx.pagev = NULL;
1839		block_ctx.mem_to_free = NULL;
1840		block_ctx.datav = mapped_datav;
1841
1842		if (is_metadata || state->include_extent_data) {
1843			block->never_written = 0;
1844			block->iodone_w_error = 0;
1845			if (NULL != bio) {
1846				block->is_iodone = 0;
1847				BUG_ON(NULL == bio_is_patched);
1848				if (!*bio_is_patched) {
1849					block->orig_bio_private =
1850					    bio->bi_private;
1851					block->orig_bio_end_io =
1852					    bio->bi_end_io;
1853					block->next_in_same_bio = NULL;
1854					bio->bi_private = block;
1855					bio->bi_end_io = btrfsic_bio_end_io;
1856					*bio_is_patched = 1;
1857				} else {
1858					struct btrfsic_block *chained_block =
1859					    (struct btrfsic_block *)
1860					    bio->bi_private;
1861
1862					BUG_ON(NULL == chained_block);
1863					block->orig_bio_private =
1864					    chained_block->orig_bio_private;
1865					block->orig_bio_end_io =
1866					    chained_block->orig_bio_end_io;
1867					block->next_in_same_bio = chained_block;
1868					bio->bi_private = block;
1869				}
1870			} else {
1871				block->is_iodone = 1;
1872				block->orig_bio_private = NULL;
1873				block->orig_bio_end_io = NULL;
1874				block->next_in_same_bio = NULL;
1875			}
1876		}
1877
1878		block->flush_gen = dev_state->last_flush_gen + 1;
1879		block->submit_bio_bh_rw = submit_bio_bh_rw;
1880		if (is_metadata) {
1881			block->logical_bytenr = bytenr;
1882			block->is_metadata = 1;
1883			if (block->is_superblock) {
1884				BUG_ON(PAGE_SIZE !=
1885				       BTRFS_SUPER_INFO_SIZE);
1886				ret = btrfsic_process_written_superblock(
1887						state,
1888						block,
1889						(struct btrfs_super_block *)
1890						mapped_datav[0]);
1891				if (state->print_mask &
1892				    BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
1893					pr_info("[after new superblock is written]:\n");
1894					btrfsic_dump_tree_sub(state, block, 0);
1895				}
1896			} else {
1897				block->mirror_num = 0;	/* unknown */
1898				ret = btrfsic_process_metablock(
1899						state,
1900						block,
1901						&block_ctx,
1902						0, 0);
1903			}
1904			if (ret)
1905				pr_info("btrfsic: btrfsic_process_metablock(root @%llu) failed!\n",
1906				       dev_bytenr);
1907		} else {
1908			block->is_metadata = 0;
1909			block->mirror_num = 0;	/* unknown */
1910			block->generation = BTRFSIC_GENERATION_UNKNOWN;
1911			if (!state->include_extent_data
1912			    && list_empty(&block->ref_from_list)) {
1913				/*
1914				 * disk block is overwritten with extent
1915				 * data (not meta data) and we are configured
1916				 * to not include extent data: take the
1917				 * chance and free the block's memory
1918				 */
1919				btrfsic_block_hashtable_remove(block);
1920				list_del(&block->all_blocks_node);
1921				btrfsic_block_free(block);
1922			}
1923		}
1924		btrfsic_release_block_ctx(&block_ctx);
1925	} else {
1926		/* block has not been found in hash table */
1927		u64 bytenr;
1928
1929		if (!is_metadata) {
1930			processed_len = state->datablock_size;
1931			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1932				pr_info(
1933			"written block (%pg/%llu/?) !found in hash table, D\n",
1934				       dev_state->bdev, dev_bytenr);
1935			if (!state->include_extent_data) {
1936				/* ignore that written D block */
1937				goto continue_loop;
1938			}
1939
1940			/* this is getting ugly for the
1941			 * include_extent_data case... */
1942			bytenr = 0;	/* unknown */
1943		} else {
1944			processed_len = state->metablock_size;
1945			bytenr = btrfs_stack_header_bytenr(
1946					(struct btrfs_header *)
1947					mapped_datav[0]);
1948			btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
1949						       dev_bytenr);
1950			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1951				pr_info(
1952			"written block @%llu (%pg/%llu/?) !found in hash table, M\n",
1953				       bytenr, dev_state->bdev, dev_bytenr);
1954		}
1955
1956		block_ctx.dev = dev_state;
1957		block_ctx.dev_bytenr = dev_bytenr;
1958		block_ctx.start = bytenr;
1959		block_ctx.len = processed_len;
1960		block_ctx.pagev = NULL;
1961		block_ctx.mem_to_free = NULL;
1962		block_ctx.datav = mapped_datav;
1963
1964		block = btrfsic_block_alloc();
1965		if (NULL == block) {
1966			btrfsic_release_block_ctx(&block_ctx);
1967			goto continue_loop;
1968		}
1969		block->dev_state = dev_state;
1970		block->dev_bytenr = dev_bytenr;
1971		block->logical_bytenr = bytenr;
1972		block->is_metadata = is_metadata;
1973		block->never_written = 0;
1974		block->iodone_w_error = 0;
1975		block->mirror_num = 0;	/* unknown */
1976		block->flush_gen = dev_state->last_flush_gen + 1;
1977		block->submit_bio_bh_rw = submit_bio_bh_rw;
1978		if (NULL != bio) {
1979			block->is_iodone = 0;
1980			BUG_ON(NULL == bio_is_patched);
1981			if (!*bio_is_patched) {
1982				block->orig_bio_private = bio->bi_private;
1983				block->orig_bio_end_io = bio->bi_end_io;
1984				block->next_in_same_bio = NULL;
1985				bio->bi_private = block;
1986				bio->bi_end_io = btrfsic_bio_end_io;
1987				*bio_is_patched = 1;
1988			} else {
1989				struct btrfsic_block *chained_block =
1990				    (struct btrfsic_block *)
1991				    bio->bi_private;
1992
1993				BUG_ON(NULL == chained_block);
1994				block->orig_bio_private =
1995				    chained_block->orig_bio_private;
1996				block->orig_bio_end_io =
1997				    chained_block->orig_bio_end_io;
1998				block->next_in_same_bio = chained_block;
1999				bio->bi_private = block;
2000			}
2001		} else {
2002			block->is_iodone = 1;
2003			block->orig_bio_private = NULL;
2004			block->orig_bio_end_io = NULL;
2005			block->next_in_same_bio = NULL;
2006		}
2007		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2008			pr_info("new written %c-block @%llu (%pg/%llu/%d)\n",
2009			       is_metadata ? 'M' : 'D',
2010			       block->logical_bytenr, block->dev_state->bdev,
2011			       block->dev_bytenr, block->mirror_num);
2012		list_add(&block->all_blocks_node, &state->all_blocks_list);
2013		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2014
2015		if (is_metadata) {
2016			ret = btrfsic_process_metablock(state, block,
2017							&block_ctx, 0, 0);
2018			if (ret)
2019				pr_info("btrfsic: process_metablock(root @%llu) failed!\n",
2020				       dev_bytenr);
2021		}
2022		btrfsic_release_block_ctx(&block_ctx);
2023	}
2024
2025continue_loop:
2026	BUG_ON(!processed_len);
2027	dev_bytenr += processed_len;
2028	mapped_datav += processed_len >> PAGE_SHIFT;
2029	num_pages -= processed_len >> PAGE_SHIFT;
2030	goto again;
2031}
2032
2033static void btrfsic_bio_end_io(struct bio *bp)
2034{
2035	struct btrfsic_block *block = bp->bi_private;
2036	int iodone_w_error;
2037
2038	/* mutex is not held! This is not save if IO is not yet completed
2039	 * on umount */
2040	iodone_w_error = 0;
2041	if (bp->bi_status)
2042		iodone_w_error = 1;
2043
2044	BUG_ON(NULL == block);
2045	bp->bi_private = block->orig_bio_private;
2046	bp->bi_end_io = block->orig_bio_end_io;
2047
2048	do {
2049		struct btrfsic_block *next_block;
2050		struct btrfsic_dev_state *const dev_state = block->dev_state;
2051
2052		if ((dev_state->state->print_mask &
2053		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2054			pr_info("bio_end_io(err=%d) for %c @%llu (%pg/%llu/%d)\n",
2055			       bp->bi_status,
2056			       btrfsic_get_block_type(dev_state->state, block),
2057			       block->logical_bytenr, dev_state->bdev,
2058			       block->dev_bytenr, block->mirror_num);
2059		next_block = block->next_in_same_bio;
2060		block->iodone_w_error = iodone_w_error;
2061		if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2062			dev_state->last_flush_gen++;
2063			if ((dev_state->state->print_mask &
2064			     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2065				pr_info("bio_end_io() new %pg flush_gen=%llu\n",
2066				       dev_state->bdev,
2067				       dev_state->last_flush_gen);
2068		}
2069		if (block->submit_bio_bh_rw & REQ_FUA)
2070			block->flush_gen = 0; /* FUA completed means block is
2071					       * on disk */
2072		block->is_iodone = 1; /* for FLUSH, this releases the block */
2073		block = next_block;
2074	} while (NULL != block);
2075
2076	bp->bi_end_io(bp);
2077}
2078
2079static int btrfsic_process_written_superblock(
2080		struct btrfsic_state *state,
2081		struct btrfsic_block *const superblock,
2082		struct btrfs_super_block *const super_hdr)
2083{
2084	struct btrfs_fs_info *fs_info = state->fs_info;
2085	int pass;
2086
2087	superblock->generation = btrfs_super_generation(super_hdr);
2088	if (!(superblock->generation > state->max_superblock_generation ||
2089	      0 == state->max_superblock_generation)) {
2090		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2091			pr_info(
2092	"btrfsic: superblock @%llu (%pg/%llu/%d) with old gen %llu <= %llu\n",
2093			       superblock->logical_bytenr,
2094			       superblock->dev_state->bdev,
2095			       superblock->dev_bytenr, superblock->mirror_num,
2096			       btrfs_super_generation(super_hdr),
2097			       state->max_superblock_generation);
2098	} else {
2099		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2100			pr_info(
2101	"btrfsic: got new superblock @%llu (%pg/%llu/%d) with new gen %llu > %llu\n",
2102			       superblock->logical_bytenr,
2103			       superblock->dev_state->bdev,
2104			       superblock->dev_bytenr, superblock->mirror_num,
2105			       btrfs_super_generation(super_hdr),
2106			       state->max_superblock_generation);
2107
2108		state->max_superblock_generation =
2109		    btrfs_super_generation(super_hdr);
2110		state->latest_superblock = superblock;
2111	}
2112
2113	for (pass = 0; pass < 3; pass++) {
2114		int ret;
2115		u64 next_bytenr;
2116		struct btrfsic_block *next_block;
2117		struct btrfsic_block_data_ctx tmp_next_block_ctx;
2118		struct btrfsic_block_link *l;
2119		int num_copies;
2120		int mirror_num;
2121		const char *additional_string = NULL;
2122		struct btrfs_disk_key tmp_disk_key = {0};
2123
2124		btrfs_set_disk_key_objectid(&tmp_disk_key,
2125					    BTRFS_ROOT_ITEM_KEY);
2126		btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2127
2128		switch (pass) {
2129		case 0:
2130			btrfs_set_disk_key_objectid(&tmp_disk_key,
2131						    BTRFS_ROOT_TREE_OBJECTID);
2132			additional_string = "root ";
2133			next_bytenr = btrfs_super_root(super_hdr);
2134			if (state->print_mask &
2135			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2136				pr_info("root@%llu\n", next_bytenr);
2137			break;
2138		case 1:
2139			btrfs_set_disk_key_objectid(&tmp_disk_key,
2140						    BTRFS_CHUNK_TREE_OBJECTID);
2141			additional_string = "chunk ";
2142			next_bytenr = btrfs_super_chunk_root(super_hdr);
2143			if (state->print_mask &
2144			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2145				pr_info("chunk@%llu\n", next_bytenr);
2146			break;
2147		case 2:
2148			btrfs_set_disk_key_objectid(&tmp_disk_key,
2149						    BTRFS_TREE_LOG_OBJECTID);
2150			additional_string = "log ";
2151			next_bytenr = btrfs_super_log_root(super_hdr);
2152			if (0 == next_bytenr)
2153				continue;
2154			if (state->print_mask &
2155			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2156				pr_info("log@%llu\n", next_bytenr);
2157			break;
2158		}
2159
2160		num_copies = btrfs_num_copies(fs_info, next_bytenr,
2161					      BTRFS_SUPER_INFO_SIZE);
2162		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2163			pr_info("num_copies(log_bytenr=%llu) = %d\n",
2164			       next_bytenr, num_copies);
2165		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2166			int was_created;
2167
2168			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2169				pr_info("btrfsic_process_written_superblock(mirror_num=%d)\n", mirror_num);
2170			ret = btrfsic_map_block(state, next_bytenr,
2171						BTRFS_SUPER_INFO_SIZE,
2172						&tmp_next_block_ctx,
2173						mirror_num);
2174			if (ret) {
2175				pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
2176				       next_bytenr, mirror_num);
2177				return -1;
2178			}
2179
2180			next_block = btrfsic_block_lookup_or_add(
2181					state,
2182					&tmp_next_block_ctx,
2183					additional_string,
2184					1, 0, 1,
2185					mirror_num,
2186					&was_created);
2187			if (NULL == next_block) {
2188				btrfsic_release_block_ctx(&tmp_next_block_ctx);
2189				return -1;
2190			}
2191
2192			next_block->disk_key = tmp_disk_key;
2193			if (was_created)
2194				next_block->generation =
2195				    BTRFSIC_GENERATION_UNKNOWN;
2196			l = btrfsic_block_link_lookup_or_add(
2197					state,
2198					&tmp_next_block_ctx,
2199					next_block,
2200					superblock,
2201					BTRFSIC_GENERATION_UNKNOWN);
2202			btrfsic_release_block_ctx(&tmp_next_block_ctx);
2203			if (NULL == l)
2204				return -1;
2205		}
2206	}
2207
2208	if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2209		btrfsic_dump_tree(state);
2210
2211	return 0;
2212}
2213
2214static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2215					struct btrfsic_block *const block,
2216					int recursion_level)
2217{
2218	const struct btrfsic_block_link *l;
2219	int ret = 0;
2220
2221	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2222		/*
2223		 * Note that this situation can happen and does not
2224		 * indicate an error in regular cases. It happens
2225		 * when disk blocks are freed and later reused.
2226		 * The check-integrity module is not aware of any
2227		 * block free operations, it just recognizes block
2228		 * write operations. Therefore it keeps the linkage
2229		 * information for a block until a block is
2230		 * rewritten. This can temporarily cause incorrect
2231		 * and even circular linkage information. This
2232		 * causes no harm unless such blocks are referenced
2233		 * by the most recent super block.
2234		 */
2235		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2236			pr_info("btrfsic: abort cyclic linkage (case 1).\n");
2237
2238		return ret;
2239	}
2240
2241	/*
2242	 * This algorithm is recursive because the amount of used stack
2243	 * space is very small and the max recursion depth is limited.
2244	 */
2245	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2246		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2247			pr_info(
2248		"rl=%d, %c @%llu (%pg/%llu/%d) %u* refers to %c @%llu (%pg/%llu/%d)\n",
2249			       recursion_level,
2250			       btrfsic_get_block_type(state, block),
2251			       block->logical_bytenr, block->dev_state->bdev,
2252			       block->dev_bytenr, block->mirror_num,
2253			       l->ref_cnt,
2254			       btrfsic_get_block_type(state, l->block_ref_to),
2255			       l->block_ref_to->logical_bytenr,
2256			       l->block_ref_to->dev_state->bdev,
2257			       l->block_ref_to->dev_bytenr,
2258			       l->block_ref_to->mirror_num);
2259		if (l->block_ref_to->never_written) {
2260			pr_info(
2261"btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) which is never written!\n",
2262			       btrfsic_get_block_type(state, l->block_ref_to),
2263			       l->block_ref_to->logical_bytenr,
2264			       l->block_ref_to->dev_state->bdev,
2265			       l->block_ref_to->dev_bytenr,
2266			       l->block_ref_to->mirror_num);
2267			ret = -1;
2268		} else if (!l->block_ref_to->is_iodone) {
2269			pr_info(
2270"btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) which is not yet iodone!\n",
2271			       btrfsic_get_block_type(state, l->block_ref_to),
2272			       l->block_ref_to->logical_bytenr,
2273			       l->block_ref_to->dev_state->bdev,
2274			       l->block_ref_to->dev_bytenr,
2275			       l->block_ref_to->mirror_num);
2276			ret = -1;
2277		} else if (l->block_ref_to->iodone_w_error) {
2278			pr_info(
2279"btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) which has write error!\n",
2280			       btrfsic_get_block_type(state, l->block_ref_to),
2281			       l->block_ref_to->logical_bytenr,
2282			       l->block_ref_to->dev_state->bdev,
2283			       l->block_ref_to->dev_bytenr,
2284			       l->block_ref_to->mirror_num);
2285			ret = -1;
2286		} else if (l->parent_generation !=
2287			   l->block_ref_to->generation &&
2288			   BTRFSIC_GENERATION_UNKNOWN !=
2289			   l->parent_generation &&
2290			   BTRFSIC_GENERATION_UNKNOWN !=
2291			   l->block_ref_to->generation) {
2292			pr_info(
2293"btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) with generation %llu != parent generation %llu!\n",
2294			       btrfsic_get_block_type(state, l->block_ref_to),
2295			       l->block_ref_to->logical_bytenr,
2296			       l->block_ref_to->dev_state->bdev,
2297			       l->block_ref_to->dev_bytenr,
2298			       l->block_ref_to->mirror_num,
2299			       l->block_ref_to->generation,
2300			       l->parent_generation);
2301			ret = -1;
2302		} else if (l->block_ref_to->flush_gen >
2303			   l->block_ref_to->dev_state->last_flush_gen) {
2304			pr_info(
2305"btrfs: attempt to write superblock which references block %c @%llu (%pg/%llu/%d) which is not flushed out of disk's write cache (block flush_gen=%llu, dev->flush_gen=%llu)!\n",
2306			       btrfsic_get_block_type(state, l->block_ref_to),
2307			       l->block_ref_to->logical_bytenr,
2308			       l->block_ref_to->dev_state->bdev,
2309			       l->block_ref_to->dev_bytenr,
2310			       l->block_ref_to->mirror_num, block->flush_gen,
2311			       l->block_ref_to->dev_state->last_flush_gen);
2312			ret = -1;
2313		} else if (-1 == btrfsic_check_all_ref_blocks(state,
2314							      l->block_ref_to,
2315							      recursion_level +
2316							      1)) {
2317			ret = -1;
2318		}
2319	}
2320
2321	return ret;
2322}
2323
2324static int btrfsic_is_block_ref_by_superblock(
2325		const struct btrfsic_state *state,
2326		const struct btrfsic_block *block,
2327		int recursion_level)
2328{
2329	const struct btrfsic_block_link *l;
2330
2331	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2332		/* refer to comment at "abort cyclic linkage (case 1)" */
2333		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2334			pr_info("btrfsic: abort cyclic linkage (case 2).\n");
2335
2336		return 0;
2337	}
2338
2339	/*
2340	 * This algorithm is recursive because the amount of used stack space
2341	 * is very small and the max recursion depth is limited.
2342	 */
2343	list_for_each_entry(l, &block->ref_from_list, node_ref_from) {
2344		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2345			pr_info(
2346	"rl=%d, %c @%llu (%pg/%llu/%d) is ref %u* from %c @%llu (%pg/%llu/%d)\n",
2347			       recursion_level,
2348			       btrfsic_get_block_type(state, block),
2349			       block->logical_bytenr, block->dev_state->bdev,
2350			       block->dev_bytenr, block->mirror_num,
2351			       l->ref_cnt,
2352			       btrfsic_get_block_type(state, l->block_ref_from),
2353			       l->block_ref_from->logical_bytenr,
2354			       l->block_ref_from->dev_state->bdev,
2355			       l->block_ref_from->dev_bytenr,
2356			       l->block_ref_from->mirror_num);
2357		if (l->block_ref_from->is_superblock &&
2358		    state->latest_superblock->dev_bytenr ==
2359		    l->block_ref_from->dev_bytenr &&
2360		    state->latest_superblock->dev_state->bdev ==
2361		    l->block_ref_from->dev_state->bdev)
2362			return 1;
2363		else if (btrfsic_is_block_ref_by_superblock(state,
2364							    l->block_ref_from,
2365							    recursion_level +
2366							    1))
2367			return 1;
2368	}
2369
2370	return 0;
2371}
2372
2373static void btrfsic_print_add_link(const struct btrfsic_state *state,
2374				   const struct btrfsic_block_link *l)
2375{
2376	pr_info("add %u* link from %c @%llu (%pg/%llu/%d) to %c @%llu (%pg/%llu/%d)\n",
2377	       l->ref_cnt,
2378	       btrfsic_get_block_type(state, l->block_ref_from),
2379	       l->block_ref_from->logical_bytenr,
2380	       l->block_ref_from->dev_state->bdev,
2381	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2382	       btrfsic_get_block_type(state, l->block_ref_to),
2383	       l->block_ref_to->logical_bytenr,
2384	       l->block_ref_to->dev_state->bdev, l->block_ref_to->dev_bytenr,
2385	       l->block_ref_to->mirror_num);
2386}
2387
2388static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2389				   const struct btrfsic_block_link *l)
2390{
2391	pr_info("rem %u* link from %c @%llu (%pg/%llu/%d) to %c @%llu (%pg/%llu/%d)\n",
2392	       l->ref_cnt,
2393	       btrfsic_get_block_type(state, l->block_ref_from),
2394	       l->block_ref_from->logical_bytenr,
2395	       l->block_ref_from->dev_state->bdev,
2396	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2397	       btrfsic_get_block_type(state, l->block_ref_to),
2398	       l->block_ref_to->logical_bytenr,
2399	       l->block_ref_to->dev_state->bdev, l->block_ref_to->dev_bytenr,
2400	       l->block_ref_to->mirror_num);
2401}
2402
2403static char btrfsic_get_block_type(const struct btrfsic_state *state,
2404				   const struct btrfsic_block *block)
2405{
2406	if (block->is_superblock &&
2407	    state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2408	    state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2409		return 'S';
2410	else if (block->is_superblock)
2411		return 's';
2412	else if (block->is_metadata)
2413		return 'M';
2414	else
2415		return 'D';
2416}
2417
2418static void btrfsic_dump_tree(const struct btrfsic_state *state)
2419{
2420	btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2421}
2422
2423static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2424				  const struct btrfsic_block *block,
2425				  int indent_level)
2426{
2427	const struct btrfsic_block_link *l;
2428	int indent_add;
2429	static char buf[80];
2430	int cursor_position;
2431
2432	/*
2433	 * Should better fill an on-stack buffer with a complete line and
2434	 * dump it at once when it is time to print a newline character.
2435	 */
2436
2437	/*
2438	 * This algorithm is recursive because the amount of used stack space
2439	 * is very small and the max recursion depth is limited.
2440	 */
2441	indent_add = sprintf(buf, "%c-%llu(%pg/%llu/%u)",
2442			     btrfsic_get_block_type(state, block),
2443			     block->logical_bytenr, block->dev_state->bdev,
2444			     block->dev_bytenr, block->mirror_num);
2445	if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2446		printk("[...]\n");
2447		return;
2448	}
2449	printk(buf);
2450	indent_level += indent_add;
2451	if (list_empty(&block->ref_to_list)) {
2452		printk("\n");
2453		return;
2454	}
2455	if (block->mirror_num > 1 &&
2456	    !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2457		printk(" [...]\n");
2458		return;
2459	}
2460
2461	cursor_position = indent_level;
2462	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2463		while (cursor_position < indent_level) {
2464			printk(" ");
2465			cursor_position++;
2466		}
2467		if (l->ref_cnt > 1)
2468			indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2469		else
2470			indent_add = sprintf(buf, " --> ");
2471		if (indent_level + indent_add >
2472		    BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2473			printk("[...]\n");
2474			cursor_position = 0;
2475			continue;
2476		}
2477
2478		printk(buf);
2479
2480		btrfsic_dump_tree_sub(state, l->block_ref_to,
2481				      indent_level + indent_add);
2482		cursor_position = 0;
2483	}
2484}
2485
2486static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2487		struct btrfsic_state *state,
2488		struct btrfsic_block_data_ctx *next_block_ctx,
2489		struct btrfsic_block *next_block,
2490		struct btrfsic_block *from_block,
2491		u64 parent_generation)
2492{
2493	struct btrfsic_block_link *l;
2494
2495	l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2496						next_block_ctx->dev_bytenr,
2497						from_block->dev_state->bdev,
2498						from_block->dev_bytenr,
2499						&state->block_link_hashtable);
2500	if (NULL == l) {
2501		l = btrfsic_block_link_alloc();
2502		if (!l)
2503			return NULL;
2504
2505		l->block_ref_to = next_block;
2506		l->block_ref_from = from_block;
2507		l->ref_cnt = 1;
2508		l->parent_generation = parent_generation;
2509
2510		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2511			btrfsic_print_add_link(state, l);
2512
2513		list_add(&l->node_ref_to, &from_block->ref_to_list);
2514		list_add(&l->node_ref_from, &next_block->ref_from_list);
2515
2516		btrfsic_block_link_hashtable_add(l,
2517						 &state->block_link_hashtable);
2518	} else {
2519		l->ref_cnt++;
2520		l->parent_generation = parent_generation;
2521		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2522			btrfsic_print_add_link(state, l);
2523	}
2524
2525	return l;
2526}
2527
2528static struct btrfsic_block *btrfsic_block_lookup_or_add(
2529		struct btrfsic_state *state,
2530		struct btrfsic_block_data_ctx *block_ctx,
2531		const char *additional_string,
2532		int is_metadata,
2533		int is_iodone,
2534		int never_written,
2535		int mirror_num,
2536		int *was_created)
2537{
2538	struct btrfsic_block *block;
2539
2540	block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2541					       block_ctx->dev_bytenr,
2542					       &state->block_hashtable);
2543	if (NULL == block) {
2544		struct btrfsic_dev_state *dev_state;
2545
2546		block = btrfsic_block_alloc();
2547		if (!block)
2548			return NULL;
2549
2550		dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev->bd_dev);
2551		if (NULL == dev_state) {
2552			pr_info("btrfsic: error, lookup dev_state failed!\n");
2553			btrfsic_block_free(block);
2554			return NULL;
2555		}
2556		block->dev_state = dev_state;
2557		block->dev_bytenr = block_ctx->dev_bytenr;
2558		block->logical_bytenr = block_ctx->start;
2559		block->is_metadata = is_metadata;
2560		block->is_iodone = is_iodone;
2561		block->never_written = never_written;
2562		block->mirror_num = mirror_num;
2563		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2564			pr_info("New %s%c-block @%llu (%pg/%llu/%d)\n",
2565			       additional_string,
2566			       btrfsic_get_block_type(state, block),
2567			       block->logical_bytenr, dev_state->bdev,
2568			       block->dev_bytenr, mirror_num);
2569		list_add(&block->all_blocks_node, &state->all_blocks_list);
2570		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2571		if (NULL != was_created)
2572			*was_created = 1;
2573	} else {
2574		if (NULL != was_created)
2575			*was_created = 0;
2576	}
2577
2578	return block;
2579}
2580
2581static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2582					   u64 bytenr,
2583					   struct btrfsic_dev_state *dev_state,
2584					   u64 dev_bytenr)
2585{
2586	struct btrfs_fs_info *fs_info = state->fs_info;
2587	struct btrfsic_block_data_ctx block_ctx;
2588	int num_copies;
2589	int mirror_num;
2590	int match = 0;
2591	int ret;
2592
2593	num_copies = btrfs_num_copies(fs_info, bytenr, state->metablock_size);
2594
2595	for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2596		ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2597					&block_ctx, mirror_num);
2598		if (ret) {
2599			pr_info("btrfsic: btrfsic_map_block(logical @%llu, mirror %d) failed!\n",
2600			       bytenr, mirror_num);
2601			continue;
2602		}
2603
2604		if (dev_state->bdev == block_ctx.dev->bdev &&
2605		    dev_bytenr == block_ctx.dev_bytenr) {
2606			match++;
2607			btrfsic_release_block_ctx(&block_ctx);
2608			break;
2609		}
2610		btrfsic_release_block_ctx(&block_ctx);
2611	}
2612
2613	if (WARN_ON(!match)) {
2614		pr_info(
2615"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=%pg, phys_bytenr=%llu)!\n",
2616		       bytenr, dev_state->bdev, dev_bytenr);
2617		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2618			ret = btrfsic_map_block(state, bytenr,
2619						state->metablock_size,
2620						&block_ctx, mirror_num);
2621			if (ret)
2622				continue;
2623
2624			pr_info("read logical bytenr @%llu maps to (%pg/%llu/%d)\n",
2625			       bytenr, block_ctx.dev->bdev,
2626			       block_ctx.dev_bytenr, mirror_num);
2627		}
2628	}
2629}
2630
2631static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev)
2632{
2633	return btrfsic_dev_state_hashtable_lookup(dev,
2634						  &btrfsic_dev_state_hashtable);
2635}
2636
2637static void btrfsic_check_write_bio(struct bio *bio, struct btrfsic_dev_state *dev_state)
2638{
2639	unsigned int segs = bio_segments(bio);
2640	u64 dev_bytenr = 512 * bio->bi_iter.bi_sector;
2641	u64 cur_bytenr = dev_bytenr;
2642	struct bvec_iter iter;
2643	struct bio_vec bvec;
2644	char **mapped_datav;
2645	int bio_is_patched = 0;
2646	int i = 0;
2647
2648	if (dev_state->state->print_mask & BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2649		pr_info(
2650"submit_bio(rw=%d,0x%x, bi_vcnt=%u, bi_sector=%llu (bytenr %llu), bi_bdev=%p)\n",
2651		       bio_op(bio), bio->bi_opf, segs,
2652		       bio->bi_iter.bi_sector, dev_bytenr, bio->bi_bdev);
2653
2654	mapped_datav = kmalloc_array(segs, sizeof(*mapped_datav), GFP_NOFS);
2655	if (!mapped_datav)
2656		return;
2657
2658	bio_for_each_segment(bvec, bio, iter) {
2659		BUG_ON(bvec.bv_len != PAGE_SIZE);
2660		mapped_datav[i] = page_address(bvec.bv_page);
2661		i++;
 
 
 
 
 
 
 
 
 
 
2662
 
 
2663		if (dev_state->state->print_mask &
2664		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
2665			pr_info("#%u: bytenr=%llu, len=%u, offset=%u\n",
2666			       i, cur_bytenr, bvec.bv_len, bvec.bv_offset);
2667		cur_bytenr += bvec.bv_len;
2668	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2669
2670	btrfsic_process_written_block(dev_state, dev_bytenr, mapped_datav, segs,
2671				      bio, &bio_is_patched, bio->bi_opf);
2672	kfree(mapped_datav);
 
 
 
 
 
 
 
 
 
 
 
2673}
2674
2675static void btrfsic_check_flush_bio(struct bio *bio, struct btrfsic_dev_state *dev_state)
2676{
2677	if (dev_state->state->print_mask & BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2678		pr_info("submit_bio(rw=%d,0x%x FLUSH, bdev=%p)\n",
2679		       bio_op(bio), bio->bi_opf, bio->bi_bdev);
2680
2681	if (dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2682		struct btrfsic_block *const block =
2683			&dev_state->dummy_block_for_bio_bh_flush;
2684
2685		block->is_iodone = 0;
2686		block->never_written = 0;
2687		block->iodone_w_error = 0;
2688		block->flush_gen = dev_state->last_flush_gen + 1;
2689		block->submit_bio_bh_rw = bio->bi_opf;
2690		block->orig_bio_private = bio->bi_private;
2691		block->orig_bio_end_io = bio->bi_end_io;
2692		block->next_in_same_bio = NULL;
2693		bio->bi_private = block;
2694		bio->bi_end_io = btrfsic_bio_end_io;
2695	} else if ((dev_state->state->print_mask &
2696		   (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2697		    BTRFSIC_PRINT_MASK_VERBOSE))) {
2698		pr_info(
2699"btrfsic_submit_bio(%pg) with FLUSH but dummy block already in use (ignored)!\n",
2700		       dev_state->bdev);
2701	}
2702}
2703
2704void btrfsic_check_bio(struct bio *bio)
2705{
2706	struct btrfsic_dev_state *dev_state;
2707
2708	if (!btrfsic_is_initialized)
2709		return;
2710
2711	/*
2712	 * We can be called before btrfsic_mount, so there might not be a
2713	 * dev_state.
2714	 */
2715	dev_state = btrfsic_dev_state_lookup(bio->bi_bdev->bd_dev);
2716	mutex_lock(&btrfsic_mutex);
2717	if (dev_state) {
2718		if (bio_op(bio) == REQ_OP_WRITE && bio_has_data(bio))
2719			btrfsic_check_write_bio(bio, dev_state);
2720		else if (bio->bi_opf & REQ_PREFLUSH)
2721			btrfsic_check_flush_bio(bio, dev_state);
2722	}
2723	mutex_unlock(&btrfsic_mutex);
2724}
2725
2726int btrfsic_mount(struct btrfs_fs_info *fs_info,
2727		  struct btrfs_fs_devices *fs_devices,
2728		  int including_extent_data, u32 print_mask)
2729{
2730	int ret;
2731	struct btrfsic_state *state;
2732	struct list_head *dev_head = &fs_devices->devices;
2733	struct btrfs_device *device;
2734
2735	if (!PAGE_ALIGNED(fs_info->nodesize)) {
2736		pr_info("btrfsic: cannot handle nodesize %d not being a multiple of PAGE_SIZE %ld!\n",
2737		       fs_info->nodesize, PAGE_SIZE);
2738		return -1;
2739	}
2740	if (!PAGE_ALIGNED(fs_info->sectorsize)) {
2741		pr_info("btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_SIZE %ld!\n",
2742		       fs_info->sectorsize, PAGE_SIZE);
2743		return -1;
2744	}
2745	state = kvzalloc(sizeof(*state), GFP_KERNEL);
2746	if (!state)
2747		return -ENOMEM;
2748
2749	if (!btrfsic_is_initialized) {
2750		mutex_init(&btrfsic_mutex);
2751		btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
2752		btrfsic_is_initialized = 1;
2753	}
2754	mutex_lock(&btrfsic_mutex);
2755	state->fs_info = fs_info;
2756	state->print_mask = print_mask;
2757	state->include_extent_data = including_extent_data;
 
2758	state->metablock_size = fs_info->nodesize;
2759	state->datablock_size = fs_info->sectorsize;
2760	INIT_LIST_HEAD(&state->all_blocks_list);
2761	btrfsic_block_hashtable_init(&state->block_hashtable);
2762	btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
2763	state->max_superblock_generation = 0;
2764	state->latest_superblock = NULL;
2765
2766	list_for_each_entry(device, dev_head, dev_list) {
2767		struct btrfsic_dev_state *ds;
 
2768
2769		if (!device->bdev || !device->name)
2770			continue;
2771
2772		ds = btrfsic_dev_state_alloc();
2773		if (NULL == ds) {
2774			mutex_unlock(&btrfsic_mutex);
2775			return -ENOMEM;
2776		}
2777		ds->bdev = device->bdev;
2778		ds->state = state;
 
 
 
 
2779		btrfsic_dev_state_hashtable_add(ds,
2780						&btrfsic_dev_state_hashtable);
2781	}
2782
2783	ret = btrfsic_process_superblock(state, fs_devices);
2784	if (0 != ret) {
2785		mutex_unlock(&btrfsic_mutex);
2786		btrfsic_unmount(fs_devices);
2787		return ret;
2788	}
2789
2790	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
2791		btrfsic_dump_database(state);
2792	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
2793		btrfsic_dump_tree(state);
2794
2795	mutex_unlock(&btrfsic_mutex);
2796	return 0;
2797}
2798
2799void btrfsic_unmount(struct btrfs_fs_devices *fs_devices)
2800{
2801	struct btrfsic_block *b_all, *tmp_all;
2802	struct btrfsic_state *state;
2803	struct list_head *dev_head = &fs_devices->devices;
2804	struct btrfs_device *device;
2805
2806	if (!btrfsic_is_initialized)
2807		return;
2808
2809	mutex_lock(&btrfsic_mutex);
2810
2811	state = NULL;
2812	list_for_each_entry(device, dev_head, dev_list) {
2813		struct btrfsic_dev_state *ds;
2814
2815		if (!device->bdev || !device->name)
2816			continue;
2817
2818		ds = btrfsic_dev_state_hashtable_lookup(
2819				device->bdev->bd_dev,
2820				&btrfsic_dev_state_hashtable);
2821		if (NULL != ds) {
2822			state = ds->state;
2823			btrfsic_dev_state_hashtable_remove(ds);
2824			btrfsic_dev_state_free(ds);
2825		}
2826	}
2827
2828	if (NULL == state) {
2829		pr_info("btrfsic: error, cannot find state information on umount!\n");
2830		mutex_unlock(&btrfsic_mutex);
2831		return;
2832	}
2833
2834	/*
2835	 * Don't care about keeping the lists' state up to date,
2836	 * just free all memory that was allocated dynamically.
2837	 * Free the blocks and the block_links.
2838	 */
2839	list_for_each_entry_safe(b_all, tmp_all, &state->all_blocks_list,
2840				 all_blocks_node) {
2841		struct btrfsic_block_link *l, *tmp;
2842
2843		list_for_each_entry_safe(l, tmp, &b_all->ref_to_list,
2844					 node_ref_to) {
2845			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2846				btrfsic_print_rem_link(state, l);
2847
2848			l->ref_cnt--;
2849			if (0 == l->ref_cnt)
2850				btrfsic_block_link_free(l);
2851		}
2852
2853		if (b_all->is_iodone || b_all->never_written)
2854			btrfsic_block_free(b_all);
2855		else
2856			pr_info(
2857"btrfs: attempt to free %c-block @%llu (%pg/%llu/%d) on umount which is not yet iodone!\n",
2858			       btrfsic_get_block_type(state, b_all),
2859			       b_all->logical_bytenr, b_all->dev_state->bdev,
2860			       b_all->dev_bytenr, b_all->mirror_num);
2861	}
2862
2863	mutex_unlock(&btrfsic_mutex);
2864
2865	kvfree(state);
2866}
v5.9
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) STRATO AG 2011.  All rights reserved.
   4 */
   5
   6/*
   7 * This module can be used to catch cases when the btrfs kernel
   8 * code executes write requests to the disk that bring the file
   9 * system in an inconsistent state. In such a state, a power-loss
  10 * or kernel panic event would cause that the data on disk is
  11 * lost or at least damaged.
  12 *
  13 * Code is added that examines all block write requests during
  14 * runtime (including writes of the super block). Three rules
  15 * are verified and an error is printed on violation of the
  16 * rules:
  17 * 1. It is not allowed to write a disk block which is
  18 *    currently referenced by the super block (either directly
  19 *    or indirectly).
  20 * 2. When a super block is written, it is verified that all
  21 *    referenced (directly or indirectly) blocks fulfill the
  22 *    following requirements:
  23 *    2a. All referenced blocks have either been present when
  24 *        the file system was mounted, (i.e., they have been
  25 *        referenced by the super block) or they have been
  26 *        written since then and the write completion callback
  27 *        was called and no write error was indicated and a
  28 *        FLUSH request to the device where these blocks are
  29 *        located was received and completed.
  30 *    2b. All referenced blocks need to have a generation
  31 *        number which is equal to the parent's number.
  32 *
  33 * One issue that was found using this module was that the log
  34 * tree on disk became temporarily corrupted because disk blocks
  35 * that had been in use for the log tree had been freed and
  36 * reused too early, while being referenced by the written super
  37 * block.
  38 *
  39 * The search term in the kernel log that can be used to filter
  40 * on the existence of detected integrity issues is
  41 * "btrfs: attempt".
  42 *
  43 * The integrity check is enabled via mount options. These
  44 * mount options are only supported if the integrity check
  45 * tool is compiled by defining BTRFS_FS_CHECK_INTEGRITY.
  46 *
  47 * Example #1, apply integrity checks to all metadata:
  48 * mount /dev/sdb1 /mnt -o check_int
  49 *
  50 * Example #2, apply integrity checks to all metadata and
  51 * to data extents:
  52 * mount /dev/sdb1 /mnt -o check_int_data
  53 *
  54 * Example #3, apply integrity checks to all metadata and dump
  55 * the tree that the super block references to kernel messages
  56 * each time after a super block was written:
  57 * mount /dev/sdb1 /mnt -o check_int,check_int_print_mask=263
  58 *
  59 * If the integrity check tool is included and activated in
  60 * the mount options, plenty of kernel memory is used, and
  61 * plenty of additional CPU cycles are spent. Enabling this
  62 * functionality is not intended for normal use. In most
  63 * cases, unless you are a btrfs developer who needs to verify
  64 * the integrity of (super)-block write requests, do not
  65 * enable the config option BTRFS_FS_CHECK_INTEGRITY to
  66 * include and compile the integrity check tool.
  67 *
  68 * Expect millions of lines of information in the kernel log with an
  69 * enabled check_int_print_mask. Therefore set LOG_BUF_SHIFT in the
  70 * kernel config to at least 26 (which is 64MB). Usually the value is
  71 * limited to 21 (which is 2MB) in init/Kconfig. The file needs to be
  72 * changed like this before LOG_BUF_SHIFT can be set to a high value:
  73 * config LOG_BUF_SHIFT
  74 *       int "Kernel log buffer size (16 => 64KB, 17 => 128KB)"
  75 *       range 12 30
  76 */
  77
  78#include <linux/sched.h>
  79#include <linux/slab.h>
  80#include <linux/mutex.h>
  81#include <linux/genhd.h>
  82#include <linux/blkdev.h>
  83#include <linux/mm.h>
  84#include <linux/string.h>
  85#include <crypto/hash.h>
 
  86#include "ctree.h"
  87#include "disk-io.h"
  88#include "transaction.h"
  89#include "extent_io.h"
  90#include "volumes.h"
  91#include "print-tree.h"
  92#include "locking.h"
  93#include "check-integrity.h"
  94#include "rcu-string.h"
  95#include "compression.h"
 
  96
  97#define BTRFSIC_BLOCK_HASHTABLE_SIZE 0x10000
  98#define BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE 0x10000
  99#define BTRFSIC_DEV2STATE_HASHTABLE_SIZE 0x100
 100#define BTRFSIC_BLOCK_MAGIC_NUMBER 0x14491051
 101#define BTRFSIC_BLOCK_LINK_MAGIC_NUMBER 0x11070807
 102#define BTRFSIC_DEV2STATE_MAGIC_NUMBER 0x20111530
 103#define BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER 20111300
 104#define BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL (200 - 6)	/* in characters,
 105							 * excluding " [...]" */
 106#define BTRFSIC_GENERATION_UNKNOWN ((u64)-1)
 107
 108/*
 109 * The definition of the bitmask fields for the print_mask.
 110 * They are specified with the mount option check_integrity_print_mask.
 111 */
 112#define BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE			0x00000001
 113#define BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION		0x00000002
 114#define BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE			0x00000004
 115#define BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE			0x00000008
 116#define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH			0x00000010
 117#define BTRFSIC_PRINT_MASK_END_IO_BIO_BH			0x00000020
 118#define BTRFSIC_PRINT_MASK_VERBOSE				0x00000040
 119#define BTRFSIC_PRINT_MASK_VERY_VERBOSE				0x00000080
 120#define BTRFSIC_PRINT_MASK_INITIAL_TREE				0x00000100
 121#define BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES			0x00000200
 122#define BTRFSIC_PRINT_MASK_INITIAL_DATABASE			0x00000400
 123#define BTRFSIC_PRINT_MASK_NUM_COPIES				0x00000800
 124#define BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS		0x00001000
 125#define BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE		0x00002000
 126
 127struct btrfsic_dev_state;
 128struct btrfsic_state;
 129
 130struct btrfsic_block {
 131	u32 magic_num;		/* only used for debug purposes */
 132	unsigned int is_metadata:1;	/* if it is meta-data, not data-data */
 133	unsigned int is_superblock:1;	/* if it is one of the superblocks */
 134	unsigned int is_iodone:1;	/* if is done by lower subsystem */
 135	unsigned int iodone_w_error:1;	/* error was indicated to endio */
 136	unsigned int never_written:1;	/* block was added because it was
 137					 * referenced, not because it was
 138					 * written */
 139	unsigned int mirror_num;	/* large enough to hold
 140					 * BTRFS_SUPER_MIRROR_MAX */
 141	struct btrfsic_dev_state *dev_state;
 142	u64 dev_bytenr;		/* key, physical byte num on disk */
 143	u64 logical_bytenr;	/* logical byte num on disk */
 144	u64 generation;
 145	struct btrfs_disk_key disk_key;	/* extra info to print in case of
 146					 * issues, will not always be correct */
 147	struct list_head collision_resolving_node;	/* list node */
 148	struct list_head all_blocks_node;	/* list node */
 149
 150	/* the following two lists contain block_link items */
 151	struct list_head ref_to_list;	/* list */
 152	struct list_head ref_from_list;	/* list */
 153	struct btrfsic_block *next_in_same_bio;
 154	void *orig_bio_private;
 155	bio_end_io_t *orig_bio_end_io;
 156	int submit_bio_bh_rw;
 157	u64 flush_gen; /* only valid if !never_written */
 158};
 159
 160/*
 161 * Elements of this type are allocated dynamically and required because
 162 * each block object can refer to and can be ref from multiple blocks.
 163 * The key to lookup them in the hashtable is the dev_bytenr of
 164 * the block ref to plus the one from the block referred from.
 165 * The fact that they are searchable via a hashtable and that a
 166 * ref_cnt is maintained is not required for the btrfs integrity
 167 * check algorithm itself, it is only used to make the output more
 168 * beautiful in case that an error is detected (an error is defined
 169 * as a write operation to a block while that block is still referenced).
 170 */
 171struct btrfsic_block_link {
 172	u32 magic_num;		/* only used for debug purposes */
 173	u32 ref_cnt;
 174	struct list_head node_ref_to;	/* list node */
 175	struct list_head node_ref_from;	/* list node */
 176	struct list_head collision_resolving_node;	/* list node */
 177	struct btrfsic_block *block_ref_to;
 178	struct btrfsic_block *block_ref_from;
 179	u64 parent_generation;
 180};
 181
 182struct btrfsic_dev_state {
 183	u32 magic_num;		/* only used for debug purposes */
 184	struct block_device *bdev;
 185	struct btrfsic_state *state;
 186	struct list_head collision_resolving_node;	/* list node */
 187	struct btrfsic_block dummy_block_for_bio_bh_flush;
 188	u64 last_flush_gen;
 189	char name[BDEVNAME_SIZE];
 190};
 191
 192struct btrfsic_block_hashtable {
 193	struct list_head table[BTRFSIC_BLOCK_HASHTABLE_SIZE];
 194};
 195
 196struct btrfsic_block_link_hashtable {
 197	struct list_head table[BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE];
 198};
 199
 200struct btrfsic_dev_state_hashtable {
 201	struct list_head table[BTRFSIC_DEV2STATE_HASHTABLE_SIZE];
 202};
 203
 204struct btrfsic_block_data_ctx {
 205	u64 start;		/* virtual bytenr */
 206	u64 dev_bytenr;		/* physical bytenr on device */
 207	u32 len;
 208	struct btrfsic_dev_state *dev;
 209	char **datav;
 210	struct page **pagev;
 211	void *mem_to_free;
 212};
 213
 214/* This structure is used to implement recursion without occupying
 215 * any stack space, refer to btrfsic_process_metablock() */
 216struct btrfsic_stack_frame {
 217	u32 magic;
 218	u32 nr;
 219	int error;
 220	int i;
 221	int limit_nesting;
 222	int num_copies;
 223	int mirror_num;
 224	struct btrfsic_block *block;
 225	struct btrfsic_block_data_ctx *block_ctx;
 226	struct btrfsic_block *next_block;
 227	struct btrfsic_block_data_ctx next_block_ctx;
 228	struct btrfs_header *hdr;
 229	struct btrfsic_stack_frame *prev;
 230};
 231
 232/* Some state per mounted filesystem */
 233struct btrfsic_state {
 234	u32 print_mask;
 235	int include_extent_data;
 236	int csum_size;
 237	struct list_head all_blocks_list;
 238	struct btrfsic_block_hashtable block_hashtable;
 239	struct btrfsic_block_link_hashtable block_link_hashtable;
 240	struct btrfs_fs_info *fs_info;
 241	u64 max_superblock_generation;
 242	struct btrfsic_block *latest_superblock;
 243	u32 metablock_size;
 244	u32 datablock_size;
 245};
 246
 247static void btrfsic_block_init(struct btrfsic_block *b);
 248static struct btrfsic_block *btrfsic_block_alloc(void);
 249static void btrfsic_block_free(struct btrfsic_block *b);
 250static void btrfsic_block_link_init(struct btrfsic_block_link *n);
 251static struct btrfsic_block_link *btrfsic_block_link_alloc(void);
 252static void btrfsic_block_link_free(struct btrfsic_block_link *n);
 253static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds);
 254static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void);
 255static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds);
 256static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h);
 257static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
 258					struct btrfsic_block_hashtable *h);
 259static void btrfsic_block_hashtable_remove(struct btrfsic_block *b);
 260static struct btrfsic_block *btrfsic_block_hashtable_lookup(
 261		struct block_device *bdev,
 262		u64 dev_bytenr,
 263		struct btrfsic_block_hashtable *h);
 264static void btrfsic_block_link_hashtable_init(
 265		struct btrfsic_block_link_hashtable *h);
 266static void btrfsic_block_link_hashtable_add(
 267		struct btrfsic_block_link *l,
 268		struct btrfsic_block_link_hashtable *h);
 269static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l);
 270static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
 271		struct block_device *bdev_ref_to,
 272		u64 dev_bytenr_ref_to,
 273		struct block_device *bdev_ref_from,
 274		u64 dev_bytenr_ref_from,
 275		struct btrfsic_block_link_hashtable *h);
 276static void btrfsic_dev_state_hashtable_init(
 277		struct btrfsic_dev_state_hashtable *h);
 278static void btrfsic_dev_state_hashtable_add(
 279		struct btrfsic_dev_state *ds,
 280		struct btrfsic_dev_state_hashtable *h);
 281static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds);
 282static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev,
 283		struct btrfsic_dev_state_hashtable *h);
 284static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void);
 285static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf);
 286static int btrfsic_process_superblock(struct btrfsic_state *state,
 287				      struct btrfs_fs_devices *fs_devices);
 288static int btrfsic_process_metablock(struct btrfsic_state *state,
 289				     struct btrfsic_block *block,
 290				     struct btrfsic_block_data_ctx *block_ctx,
 291				     int limit_nesting, int force_iodone_flag);
 292static void btrfsic_read_from_block_data(
 293	struct btrfsic_block_data_ctx *block_ctx,
 294	void *dst, u32 offset, size_t len);
 295static int btrfsic_create_link_to_next_block(
 296		struct btrfsic_state *state,
 297		struct btrfsic_block *block,
 298		struct btrfsic_block_data_ctx
 299		*block_ctx, u64 next_bytenr,
 300		int limit_nesting,
 301		struct btrfsic_block_data_ctx *next_block_ctx,
 302		struct btrfsic_block **next_blockp,
 303		int force_iodone_flag,
 304		int *num_copiesp, int *mirror_nump,
 305		struct btrfs_disk_key *disk_key,
 306		u64 parent_generation);
 307static int btrfsic_handle_extent_data(struct btrfsic_state *state,
 308				      struct btrfsic_block *block,
 309				      struct btrfsic_block_data_ctx *block_ctx,
 310				      u32 item_offset, int force_iodone_flag);
 311static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
 312			     struct btrfsic_block_data_ctx *block_ctx_out,
 313			     int mirror_num);
 314static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx);
 315static int btrfsic_read_block(struct btrfsic_state *state,
 316			      struct btrfsic_block_data_ctx *block_ctx);
 317static void btrfsic_dump_database(struct btrfsic_state *state);
 318static int btrfsic_test_for_metadata(struct btrfsic_state *state,
 319				     char **datav, unsigned int num_pages);
 320static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
 321					  u64 dev_bytenr, char **mapped_datav,
 322					  unsigned int num_pages,
 323					  struct bio *bio, int *bio_is_patched,
 324					  int submit_bio_bh_rw);
 325static int btrfsic_process_written_superblock(
 326		struct btrfsic_state *state,
 327		struct btrfsic_block *const block,
 328		struct btrfs_super_block *const super_hdr);
 329static void btrfsic_bio_end_io(struct bio *bp);
 330static int btrfsic_is_block_ref_by_superblock(const struct btrfsic_state *state,
 331					      const struct btrfsic_block *block,
 332					      int recursion_level);
 333static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
 334					struct btrfsic_block *const block,
 335					int recursion_level);
 336static void btrfsic_print_add_link(const struct btrfsic_state *state,
 337				   const struct btrfsic_block_link *l);
 338static void btrfsic_print_rem_link(const struct btrfsic_state *state,
 339				   const struct btrfsic_block_link *l);
 340static char btrfsic_get_block_type(const struct btrfsic_state *state,
 341				   const struct btrfsic_block *block);
 342static void btrfsic_dump_tree(const struct btrfsic_state *state);
 343static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
 344				  const struct btrfsic_block *block,
 345				  int indent_level);
 346static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
 347		struct btrfsic_state *state,
 348		struct btrfsic_block_data_ctx *next_block_ctx,
 349		struct btrfsic_block *next_block,
 350		struct btrfsic_block *from_block,
 351		u64 parent_generation);
 352static struct btrfsic_block *btrfsic_block_lookup_or_add(
 353		struct btrfsic_state *state,
 354		struct btrfsic_block_data_ctx *block_ctx,
 355		const char *additional_string,
 356		int is_metadata,
 357		int is_iodone,
 358		int never_written,
 359		int mirror_num,
 360		int *was_created);
 361static int btrfsic_process_superblock_dev_mirror(
 362		struct btrfsic_state *state,
 363		struct btrfsic_dev_state *dev_state,
 364		struct btrfs_device *device,
 365		int superblock_mirror_num,
 366		struct btrfsic_dev_state **selected_dev_state,
 367		struct btrfs_super_block *selected_super);
 368static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev);
 369static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
 370					   u64 bytenr,
 371					   struct btrfsic_dev_state *dev_state,
 372					   u64 dev_bytenr);
 373
 374static struct mutex btrfsic_mutex;
 375static int btrfsic_is_initialized;
 376static struct btrfsic_dev_state_hashtable btrfsic_dev_state_hashtable;
 377
 378
 379static void btrfsic_block_init(struct btrfsic_block *b)
 380{
 381	b->magic_num = BTRFSIC_BLOCK_MAGIC_NUMBER;
 382	b->dev_state = NULL;
 383	b->dev_bytenr = 0;
 384	b->logical_bytenr = 0;
 385	b->generation = BTRFSIC_GENERATION_UNKNOWN;
 386	b->disk_key.objectid = 0;
 387	b->disk_key.type = 0;
 388	b->disk_key.offset = 0;
 389	b->is_metadata = 0;
 390	b->is_superblock = 0;
 391	b->is_iodone = 0;
 392	b->iodone_w_error = 0;
 393	b->never_written = 0;
 394	b->mirror_num = 0;
 395	b->next_in_same_bio = NULL;
 396	b->orig_bio_private = NULL;
 397	b->orig_bio_end_io = NULL;
 398	INIT_LIST_HEAD(&b->collision_resolving_node);
 399	INIT_LIST_HEAD(&b->all_blocks_node);
 400	INIT_LIST_HEAD(&b->ref_to_list);
 401	INIT_LIST_HEAD(&b->ref_from_list);
 402	b->submit_bio_bh_rw = 0;
 403	b->flush_gen = 0;
 404}
 405
 406static struct btrfsic_block *btrfsic_block_alloc(void)
 407{
 408	struct btrfsic_block *b;
 409
 410	b = kzalloc(sizeof(*b), GFP_NOFS);
 411	if (NULL != b)
 412		btrfsic_block_init(b);
 413
 414	return b;
 415}
 416
 417static void btrfsic_block_free(struct btrfsic_block *b)
 418{
 419	BUG_ON(!(NULL == b || BTRFSIC_BLOCK_MAGIC_NUMBER == b->magic_num));
 420	kfree(b);
 421}
 422
 423static void btrfsic_block_link_init(struct btrfsic_block_link *l)
 424{
 425	l->magic_num = BTRFSIC_BLOCK_LINK_MAGIC_NUMBER;
 426	l->ref_cnt = 1;
 427	INIT_LIST_HEAD(&l->node_ref_to);
 428	INIT_LIST_HEAD(&l->node_ref_from);
 429	INIT_LIST_HEAD(&l->collision_resolving_node);
 430	l->block_ref_to = NULL;
 431	l->block_ref_from = NULL;
 432}
 433
 434static struct btrfsic_block_link *btrfsic_block_link_alloc(void)
 435{
 436	struct btrfsic_block_link *l;
 437
 438	l = kzalloc(sizeof(*l), GFP_NOFS);
 439	if (NULL != l)
 440		btrfsic_block_link_init(l);
 441
 442	return l;
 443}
 444
 445static void btrfsic_block_link_free(struct btrfsic_block_link *l)
 446{
 447	BUG_ON(!(NULL == l || BTRFSIC_BLOCK_LINK_MAGIC_NUMBER == l->magic_num));
 448	kfree(l);
 449}
 450
 451static void btrfsic_dev_state_init(struct btrfsic_dev_state *ds)
 452{
 453	ds->magic_num = BTRFSIC_DEV2STATE_MAGIC_NUMBER;
 454	ds->bdev = NULL;
 455	ds->state = NULL;
 456	ds->name[0] = '\0';
 457	INIT_LIST_HEAD(&ds->collision_resolving_node);
 458	ds->last_flush_gen = 0;
 459	btrfsic_block_init(&ds->dummy_block_for_bio_bh_flush);
 460	ds->dummy_block_for_bio_bh_flush.is_iodone = 1;
 461	ds->dummy_block_for_bio_bh_flush.dev_state = ds;
 462}
 463
 464static struct btrfsic_dev_state *btrfsic_dev_state_alloc(void)
 465{
 466	struct btrfsic_dev_state *ds;
 467
 468	ds = kzalloc(sizeof(*ds), GFP_NOFS);
 469	if (NULL != ds)
 470		btrfsic_dev_state_init(ds);
 471
 472	return ds;
 473}
 474
 475static void btrfsic_dev_state_free(struct btrfsic_dev_state *ds)
 476{
 477	BUG_ON(!(NULL == ds ||
 478		 BTRFSIC_DEV2STATE_MAGIC_NUMBER == ds->magic_num));
 479	kfree(ds);
 480}
 481
 482static void btrfsic_block_hashtable_init(struct btrfsic_block_hashtable *h)
 483{
 484	int i;
 485
 486	for (i = 0; i < BTRFSIC_BLOCK_HASHTABLE_SIZE; i++)
 487		INIT_LIST_HEAD(h->table + i);
 488}
 489
 490static void btrfsic_block_hashtable_add(struct btrfsic_block *b,
 491					struct btrfsic_block_hashtable *h)
 492{
 493	const unsigned int hashval =
 494	    (((unsigned int)(b->dev_bytenr >> 16)) ^
 495	     ((unsigned int)((uintptr_t)b->dev_state->bdev))) &
 496	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
 497
 498	list_add(&b->collision_resolving_node, h->table + hashval);
 499}
 500
 501static void btrfsic_block_hashtable_remove(struct btrfsic_block *b)
 502{
 503	list_del(&b->collision_resolving_node);
 504}
 505
 506static struct btrfsic_block *btrfsic_block_hashtable_lookup(
 507		struct block_device *bdev,
 508		u64 dev_bytenr,
 509		struct btrfsic_block_hashtable *h)
 510{
 511	const unsigned int hashval =
 512	    (((unsigned int)(dev_bytenr >> 16)) ^
 513	     ((unsigned int)((uintptr_t)bdev))) &
 514	     (BTRFSIC_BLOCK_HASHTABLE_SIZE - 1);
 515	struct btrfsic_block *b;
 516
 517	list_for_each_entry(b, h->table + hashval, collision_resolving_node) {
 518		if (b->dev_state->bdev == bdev && b->dev_bytenr == dev_bytenr)
 519			return b;
 520	}
 521
 522	return NULL;
 523}
 524
 525static void btrfsic_block_link_hashtable_init(
 526		struct btrfsic_block_link_hashtable *h)
 527{
 528	int i;
 529
 530	for (i = 0; i < BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE; i++)
 531		INIT_LIST_HEAD(h->table + i);
 532}
 533
 534static void btrfsic_block_link_hashtable_add(
 535		struct btrfsic_block_link *l,
 536		struct btrfsic_block_link_hashtable *h)
 537{
 538	const unsigned int hashval =
 539	    (((unsigned int)(l->block_ref_to->dev_bytenr >> 16)) ^
 540	     ((unsigned int)(l->block_ref_from->dev_bytenr >> 16)) ^
 541	     ((unsigned int)((uintptr_t)l->block_ref_to->dev_state->bdev)) ^
 542	     ((unsigned int)((uintptr_t)l->block_ref_from->dev_state->bdev)))
 543	     & (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
 544
 545	BUG_ON(NULL == l->block_ref_to);
 546	BUG_ON(NULL == l->block_ref_from);
 547	list_add(&l->collision_resolving_node, h->table + hashval);
 548}
 549
 550static void btrfsic_block_link_hashtable_remove(struct btrfsic_block_link *l)
 551{
 552	list_del(&l->collision_resolving_node);
 553}
 554
 555static struct btrfsic_block_link *btrfsic_block_link_hashtable_lookup(
 556		struct block_device *bdev_ref_to,
 557		u64 dev_bytenr_ref_to,
 558		struct block_device *bdev_ref_from,
 559		u64 dev_bytenr_ref_from,
 560		struct btrfsic_block_link_hashtable *h)
 561{
 562	const unsigned int hashval =
 563	    (((unsigned int)(dev_bytenr_ref_to >> 16)) ^
 564	     ((unsigned int)(dev_bytenr_ref_from >> 16)) ^
 565	     ((unsigned int)((uintptr_t)bdev_ref_to)) ^
 566	     ((unsigned int)((uintptr_t)bdev_ref_from))) &
 567	     (BTRFSIC_BLOCK_LINK_HASHTABLE_SIZE - 1);
 568	struct btrfsic_block_link *l;
 569
 570	list_for_each_entry(l, h->table + hashval, collision_resolving_node) {
 571		BUG_ON(NULL == l->block_ref_to);
 572		BUG_ON(NULL == l->block_ref_from);
 573		if (l->block_ref_to->dev_state->bdev == bdev_ref_to &&
 574		    l->block_ref_to->dev_bytenr == dev_bytenr_ref_to &&
 575		    l->block_ref_from->dev_state->bdev == bdev_ref_from &&
 576		    l->block_ref_from->dev_bytenr == dev_bytenr_ref_from)
 577			return l;
 578	}
 579
 580	return NULL;
 581}
 582
 583static void btrfsic_dev_state_hashtable_init(
 584		struct btrfsic_dev_state_hashtable *h)
 585{
 586	int i;
 587
 588	for (i = 0; i < BTRFSIC_DEV2STATE_HASHTABLE_SIZE; i++)
 589		INIT_LIST_HEAD(h->table + i);
 590}
 591
 592static void btrfsic_dev_state_hashtable_add(
 593		struct btrfsic_dev_state *ds,
 594		struct btrfsic_dev_state_hashtable *h)
 595{
 596	const unsigned int hashval =
 597	    (((unsigned int)((uintptr_t)ds->bdev->bd_dev)) &
 598	     (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1));
 599
 600	list_add(&ds->collision_resolving_node, h->table + hashval);
 601}
 602
 603static void btrfsic_dev_state_hashtable_remove(struct btrfsic_dev_state *ds)
 604{
 605	list_del(&ds->collision_resolving_node);
 606}
 607
 608static struct btrfsic_dev_state *btrfsic_dev_state_hashtable_lookup(dev_t dev,
 609		struct btrfsic_dev_state_hashtable *h)
 610{
 611	const unsigned int hashval =
 612		dev & (BTRFSIC_DEV2STATE_HASHTABLE_SIZE - 1);
 613	struct btrfsic_dev_state *ds;
 614
 615	list_for_each_entry(ds, h->table + hashval, collision_resolving_node) {
 616		if (ds->bdev->bd_dev == dev)
 617			return ds;
 618	}
 619
 620	return NULL;
 621}
 622
 623static int btrfsic_process_superblock(struct btrfsic_state *state,
 624				      struct btrfs_fs_devices *fs_devices)
 625{
 626	struct btrfs_super_block *selected_super;
 627	struct list_head *dev_head = &fs_devices->devices;
 628	struct btrfs_device *device;
 629	struct btrfsic_dev_state *selected_dev_state = NULL;
 630	int ret = 0;
 631	int pass;
 632
 633	selected_super = kzalloc(sizeof(*selected_super), GFP_NOFS);
 634	if (!selected_super)
 635		return -ENOMEM;
 636
 637	list_for_each_entry(device, dev_head, dev_list) {
 638		int i;
 639		struct btrfsic_dev_state *dev_state;
 640
 641		if (!device->bdev || !device->name)
 642			continue;
 643
 644		dev_state = btrfsic_dev_state_lookup(device->bdev->bd_dev);
 645		BUG_ON(NULL == dev_state);
 646		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 647			ret = btrfsic_process_superblock_dev_mirror(
 648					state, dev_state, device, i,
 649					&selected_dev_state, selected_super);
 650			if (0 != ret && 0 == i) {
 651				kfree(selected_super);
 652				return ret;
 653			}
 654		}
 655	}
 656
 657	if (NULL == state->latest_superblock) {
 658		pr_info("btrfsic: no superblock found!\n");
 659		kfree(selected_super);
 660		return -1;
 661	}
 662
 663	state->csum_size = btrfs_super_csum_size(selected_super);
 664
 665	for (pass = 0; pass < 3; pass++) {
 666		int num_copies;
 667		int mirror_num;
 668		u64 next_bytenr;
 669
 670		switch (pass) {
 671		case 0:
 672			next_bytenr = btrfs_super_root(selected_super);
 673			if (state->print_mask &
 674			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
 675				pr_info("root@%llu\n", next_bytenr);
 676			break;
 677		case 1:
 678			next_bytenr = btrfs_super_chunk_root(selected_super);
 679			if (state->print_mask &
 680			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
 681				pr_info("chunk@%llu\n", next_bytenr);
 682			break;
 683		case 2:
 684			next_bytenr = btrfs_super_log_root(selected_super);
 685			if (0 == next_bytenr)
 686				continue;
 687			if (state->print_mask &
 688			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
 689				pr_info("log@%llu\n", next_bytenr);
 690			break;
 691		}
 692
 693		num_copies = btrfs_num_copies(state->fs_info, next_bytenr,
 694					      state->metablock_size);
 695		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
 696			pr_info("num_copies(log_bytenr=%llu) = %d\n",
 697			       next_bytenr, num_copies);
 698
 699		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
 700			struct btrfsic_block *next_block;
 701			struct btrfsic_block_data_ctx tmp_next_block_ctx;
 702			struct btrfsic_block_link *l;
 703
 704			ret = btrfsic_map_block(state, next_bytenr,
 705						state->metablock_size,
 706						&tmp_next_block_ctx,
 707						mirror_num);
 708			if (ret) {
 709				pr_info("btrfsic: btrfsic_map_block(root @%llu, mirror %d) failed!\n",
 710				       next_bytenr, mirror_num);
 711				kfree(selected_super);
 712				return -1;
 713			}
 714
 715			next_block = btrfsic_block_hashtable_lookup(
 716					tmp_next_block_ctx.dev->bdev,
 717					tmp_next_block_ctx.dev_bytenr,
 718					&state->block_hashtable);
 719			BUG_ON(NULL == next_block);
 720
 721			l = btrfsic_block_link_hashtable_lookup(
 722					tmp_next_block_ctx.dev->bdev,
 723					tmp_next_block_ctx.dev_bytenr,
 724					state->latest_superblock->dev_state->
 725					bdev,
 726					state->latest_superblock->dev_bytenr,
 727					&state->block_link_hashtable);
 728			BUG_ON(NULL == l);
 729
 730			ret = btrfsic_read_block(state, &tmp_next_block_ctx);
 731			if (ret < (int)PAGE_SIZE) {
 732				pr_info("btrfsic: read @logical %llu failed!\n",
 733				       tmp_next_block_ctx.start);
 734				btrfsic_release_block_ctx(&tmp_next_block_ctx);
 735				kfree(selected_super);
 736				return -1;
 737			}
 738
 739			ret = btrfsic_process_metablock(state,
 740							next_block,
 741							&tmp_next_block_ctx,
 742							BTRFS_MAX_LEVEL + 3, 1);
 743			btrfsic_release_block_ctx(&tmp_next_block_ctx);
 744		}
 745	}
 746
 747	kfree(selected_super);
 748	return ret;
 749}
 750
 751static int btrfsic_process_superblock_dev_mirror(
 752		struct btrfsic_state *state,
 753		struct btrfsic_dev_state *dev_state,
 754		struct btrfs_device *device,
 755		int superblock_mirror_num,
 756		struct btrfsic_dev_state **selected_dev_state,
 757		struct btrfs_super_block *selected_super)
 758{
 759	struct btrfs_fs_info *fs_info = state->fs_info;
 760	struct btrfs_super_block *super_tmp;
 761	u64 dev_bytenr;
 762	struct btrfsic_block *superblock_tmp;
 763	int pass;
 764	struct block_device *const superblock_bdev = device->bdev;
 765	struct page *page;
 766	struct address_space *mapping = superblock_bdev->bd_inode->i_mapping;
 767	int ret = 0;
 768
 769	/* super block bytenr is always the unmapped device bytenr */
 770	dev_bytenr = btrfs_sb_offset(superblock_mirror_num);
 771	if (dev_bytenr + BTRFS_SUPER_INFO_SIZE > device->commit_total_bytes)
 772		return -1;
 773
 774	page = read_cache_page_gfp(mapping, dev_bytenr >> PAGE_SHIFT, GFP_NOFS);
 775	if (IS_ERR(page))
 776		return -1;
 777
 778	super_tmp = page_address(page);
 779
 780	if (btrfs_super_bytenr(super_tmp) != dev_bytenr ||
 781	    btrfs_super_magic(super_tmp) != BTRFS_MAGIC ||
 782	    memcmp(device->uuid, super_tmp->dev_item.uuid, BTRFS_UUID_SIZE) ||
 783	    btrfs_super_nodesize(super_tmp) != state->metablock_size ||
 784	    btrfs_super_sectorsize(super_tmp) != state->datablock_size) {
 785		ret = 0;
 786		goto out;
 787	}
 788
 789	superblock_tmp =
 790	    btrfsic_block_hashtable_lookup(superblock_bdev,
 791					   dev_bytenr,
 792					   &state->block_hashtable);
 793	if (NULL == superblock_tmp) {
 794		superblock_tmp = btrfsic_block_alloc();
 795		if (NULL == superblock_tmp) {
 796			ret = -1;
 797			goto out;
 798		}
 799		/* for superblock, only the dev_bytenr makes sense */
 800		superblock_tmp->dev_bytenr = dev_bytenr;
 801		superblock_tmp->dev_state = dev_state;
 802		superblock_tmp->logical_bytenr = dev_bytenr;
 803		superblock_tmp->generation = btrfs_super_generation(super_tmp);
 804		superblock_tmp->is_metadata = 1;
 805		superblock_tmp->is_superblock = 1;
 806		superblock_tmp->is_iodone = 1;
 807		superblock_tmp->never_written = 0;
 808		superblock_tmp->mirror_num = 1 + superblock_mirror_num;
 809		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
 810			btrfs_info_in_rcu(fs_info,
 811				"new initial S-block (bdev %p, %s) @%llu (%s/%llu/%d)",
 812				     superblock_bdev,
 813				     rcu_str_deref(device->name), dev_bytenr,
 814				     dev_state->name, dev_bytenr,
 815				     superblock_mirror_num);
 816		list_add(&superblock_tmp->all_blocks_node,
 817			 &state->all_blocks_list);
 818		btrfsic_block_hashtable_add(superblock_tmp,
 819					    &state->block_hashtable);
 820	}
 821
 822	/* select the one with the highest generation field */
 823	if (btrfs_super_generation(super_tmp) >
 824	    state->max_superblock_generation ||
 825	    0 == state->max_superblock_generation) {
 826		memcpy(selected_super, super_tmp, sizeof(*selected_super));
 827		*selected_dev_state = dev_state;
 828		state->max_superblock_generation =
 829		    btrfs_super_generation(super_tmp);
 830		state->latest_superblock = superblock_tmp;
 831	}
 832
 833	for (pass = 0; pass < 3; pass++) {
 834		u64 next_bytenr;
 835		int num_copies;
 836		int mirror_num;
 837		const char *additional_string = NULL;
 838		struct btrfs_disk_key tmp_disk_key;
 839
 840		tmp_disk_key.type = BTRFS_ROOT_ITEM_KEY;
 841		tmp_disk_key.offset = 0;
 842		switch (pass) {
 843		case 0:
 844			btrfs_set_disk_key_objectid(&tmp_disk_key,
 845						    BTRFS_ROOT_TREE_OBJECTID);
 846			additional_string = "initial root ";
 847			next_bytenr = btrfs_super_root(super_tmp);
 848			break;
 849		case 1:
 850			btrfs_set_disk_key_objectid(&tmp_disk_key,
 851						    BTRFS_CHUNK_TREE_OBJECTID);
 852			additional_string = "initial chunk ";
 853			next_bytenr = btrfs_super_chunk_root(super_tmp);
 854			break;
 855		case 2:
 856			btrfs_set_disk_key_objectid(&tmp_disk_key,
 857						    BTRFS_TREE_LOG_OBJECTID);
 858			additional_string = "initial log ";
 859			next_bytenr = btrfs_super_log_root(super_tmp);
 860			if (0 == next_bytenr)
 861				continue;
 862			break;
 863		}
 864
 865		num_copies = btrfs_num_copies(fs_info, next_bytenr,
 866					      state->metablock_size);
 867		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
 868			pr_info("num_copies(log_bytenr=%llu) = %d\n",
 869			       next_bytenr, num_copies);
 870		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
 871			struct btrfsic_block *next_block;
 872			struct btrfsic_block_data_ctx tmp_next_block_ctx;
 873			struct btrfsic_block_link *l;
 874
 875			if (btrfsic_map_block(state, next_bytenr,
 876					      state->metablock_size,
 877					      &tmp_next_block_ctx,
 878					      mirror_num)) {
 879				pr_info("btrfsic: btrfsic_map_block(bytenr @%llu, mirror %d) failed!\n",
 880				       next_bytenr, mirror_num);
 881				ret = -1;
 882				goto out;
 883			}
 884
 885			next_block = btrfsic_block_lookup_or_add(
 886					state, &tmp_next_block_ctx,
 887					additional_string, 1, 1, 0,
 888					mirror_num, NULL);
 889			if (NULL == next_block) {
 890				btrfsic_release_block_ctx(&tmp_next_block_ctx);
 891				ret = -1;
 892				goto out;
 893			}
 894
 895			next_block->disk_key = tmp_disk_key;
 896			next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
 897			l = btrfsic_block_link_lookup_or_add(
 898					state, &tmp_next_block_ctx,
 899					next_block, superblock_tmp,
 900					BTRFSIC_GENERATION_UNKNOWN);
 901			btrfsic_release_block_ctx(&tmp_next_block_ctx);
 902			if (NULL == l) {
 903				ret = -1;
 904				goto out;
 905			}
 906		}
 907	}
 908	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_ALL_TREES)
 909		btrfsic_dump_tree_sub(state, superblock_tmp, 0);
 910
 911out:
 912	put_page(page);
 913	return ret;
 914}
 915
 916static struct btrfsic_stack_frame *btrfsic_stack_frame_alloc(void)
 917{
 918	struct btrfsic_stack_frame *sf;
 919
 920	sf = kzalloc(sizeof(*sf), GFP_NOFS);
 921	if (sf)
 922		sf->magic = BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER;
 923	return sf;
 924}
 925
 926static void btrfsic_stack_frame_free(struct btrfsic_stack_frame *sf)
 927{
 928	BUG_ON(!(NULL == sf ||
 929		 BTRFSIC_BLOCK_STACK_FRAME_MAGIC_NUMBER == sf->magic));
 930	kfree(sf);
 931}
 932
 933static noinline_for_stack int btrfsic_process_metablock(
 934		struct btrfsic_state *state,
 935		struct btrfsic_block *const first_block,
 936		struct btrfsic_block_data_ctx *const first_block_ctx,
 937		int first_limit_nesting, int force_iodone_flag)
 938{
 939	struct btrfsic_stack_frame initial_stack_frame = { 0 };
 940	struct btrfsic_stack_frame *sf;
 941	struct btrfsic_stack_frame *next_stack;
 942	struct btrfs_header *const first_hdr =
 943		(struct btrfs_header *)first_block_ctx->datav[0];
 944
 945	BUG_ON(!first_hdr);
 946	sf = &initial_stack_frame;
 947	sf->error = 0;
 948	sf->i = -1;
 949	sf->limit_nesting = first_limit_nesting;
 950	sf->block = first_block;
 951	sf->block_ctx = first_block_ctx;
 952	sf->next_block = NULL;
 953	sf->hdr = first_hdr;
 954	sf->prev = NULL;
 955
 956continue_with_new_stack_frame:
 957	sf->block->generation = le64_to_cpu(sf->hdr->generation);
 958	if (0 == sf->hdr->level) {
 959		struct btrfs_leaf *const leafhdr =
 960		    (struct btrfs_leaf *)sf->hdr;
 961
 962		if (-1 == sf->i) {
 963			sf->nr = btrfs_stack_header_nritems(&leafhdr->header);
 964
 965			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
 966				pr_info("leaf %llu items %d generation %llu owner %llu\n",
 967				       sf->block_ctx->start, sf->nr,
 968				       btrfs_stack_header_generation(
 969					       &leafhdr->header),
 970				       btrfs_stack_header_owner(
 971					       &leafhdr->header));
 972		}
 973
 974continue_with_current_leaf_stack_frame:
 975		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
 976			sf->i++;
 977			sf->num_copies = 0;
 978		}
 979
 980		if (sf->i < sf->nr) {
 981			struct btrfs_item disk_item;
 982			u32 disk_item_offset =
 983				(uintptr_t)(leafhdr->items + sf->i) -
 984				(uintptr_t)leafhdr;
 985			struct btrfs_disk_key *disk_key;
 986			u8 type;
 987			u32 item_offset;
 988			u32 item_size;
 989
 990			if (disk_item_offset + sizeof(struct btrfs_item) >
 991			    sf->block_ctx->len) {
 992leaf_item_out_of_bounce_error:
 993				pr_info("btrfsic: leaf item out of bounce at logical %llu, dev %s\n",
 
 994				       sf->block_ctx->start,
 995				       sf->block_ctx->dev->name);
 996				goto one_stack_frame_backwards;
 997			}
 998			btrfsic_read_from_block_data(sf->block_ctx,
 999						     &disk_item,
1000						     disk_item_offset,
1001						     sizeof(struct btrfs_item));
1002			item_offset = btrfs_stack_item_offset(&disk_item);
1003			item_size = btrfs_stack_item_size(&disk_item);
1004			disk_key = &disk_item.key;
1005			type = btrfs_disk_key_type(disk_key);
1006
1007			if (BTRFS_ROOT_ITEM_KEY == type) {
1008				struct btrfs_root_item root_item;
1009				u32 root_item_offset;
1010				u64 next_bytenr;
1011
1012				root_item_offset = item_offset +
1013					offsetof(struct btrfs_leaf, items);
1014				if (root_item_offset + item_size >
1015				    sf->block_ctx->len)
1016					goto leaf_item_out_of_bounce_error;
1017				btrfsic_read_from_block_data(
1018					sf->block_ctx, &root_item,
1019					root_item_offset,
1020					item_size);
1021				next_bytenr = btrfs_root_bytenr(&root_item);
1022
1023				sf->error =
1024				    btrfsic_create_link_to_next_block(
1025						state,
1026						sf->block,
1027						sf->block_ctx,
1028						next_bytenr,
1029						sf->limit_nesting,
1030						&sf->next_block_ctx,
1031						&sf->next_block,
1032						force_iodone_flag,
1033						&sf->num_copies,
1034						&sf->mirror_num,
1035						disk_key,
1036						btrfs_root_generation(
1037						&root_item));
1038				if (sf->error)
1039					goto one_stack_frame_backwards;
1040
1041				if (NULL != sf->next_block) {
1042					struct btrfs_header *const next_hdr =
1043					    (struct btrfs_header *)
1044					    sf->next_block_ctx.datav[0];
1045
1046					next_stack =
1047					    btrfsic_stack_frame_alloc();
1048					if (NULL == next_stack) {
1049						sf->error = -1;
1050						btrfsic_release_block_ctx(
1051								&sf->
1052								next_block_ctx);
1053						goto one_stack_frame_backwards;
1054					}
1055
1056					next_stack->i = -1;
1057					next_stack->block = sf->next_block;
1058					next_stack->block_ctx =
1059					    &sf->next_block_ctx;
1060					next_stack->next_block = NULL;
1061					next_stack->hdr = next_hdr;
1062					next_stack->limit_nesting =
1063					    sf->limit_nesting - 1;
1064					next_stack->prev = sf;
1065					sf = next_stack;
1066					goto continue_with_new_stack_frame;
1067				}
1068			} else if (BTRFS_EXTENT_DATA_KEY == type &&
1069				   state->include_extent_data) {
1070				sf->error = btrfsic_handle_extent_data(
1071						state,
1072						sf->block,
1073						sf->block_ctx,
1074						item_offset,
1075						force_iodone_flag);
1076				if (sf->error)
1077					goto one_stack_frame_backwards;
1078			}
1079
1080			goto continue_with_current_leaf_stack_frame;
1081		}
1082	} else {
1083		struct btrfs_node *const nodehdr = (struct btrfs_node *)sf->hdr;
1084
1085		if (-1 == sf->i) {
1086			sf->nr = btrfs_stack_header_nritems(&nodehdr->header);
1087
1088			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1089				pr_info("node %llu level %d items %d generation %llu owner %llu\n",
1090				       sf->block_ctx->start,
1091				       nodehdr->header.level, sf->nr,
1092				       btrfs_stack_header_generation(
1093				       &nodehdr->header),
1094				       btrfs_stack_header_owner(
1095				       &nodehdr->header));
1096		}
1097
1098continue_with_current_node_stack_frame:
1099		if (0 == sf->num_copies || sf->mirror_num > sf->num_copies) {
1100			sf->i++;
1101			sf->num_copies = 0;
1102		}
1103
1104		if (sf->i < sf->nr) {
1105			struct btrfs_key_ptr key_ptr;
1106			u32 key_ptr_offset;
1107			u64 next_bytenr;
1108
1109			key_ptr_offset = (uintptr_t)(nodehdr->ptrs + sf->i) -
1110					  (uintptr_t)nodehdr;
1111			if (key_ptr_offset + sizeof(struct btrfs_key_ptr) >
1112			    sf->block_ctx->len) {
1113				pr_info("btrfsic: node item out of bounce at logical %llu, dev %s\n",
 
1114				       sf->block_ctx->start,
1115				       sf->block_ctx->dev->name);
1116				goto one_stack_frame_backwards;
1117			}
1118			btrfsic_read_from_block_data(
1119				sf->block_ctx, &key_ptr, key_ptr_offset,
1120				sizeof(struct btrfs_key_ptr));
1121			next_bytenr = btrfs_stack_key_blockptr(&key_ptr);
1122
1123			sf->error = btrfsic_create_link_to_next_block(
1124					state,
1125					sf->block,
1126					sf->block_ctx,
1127					next_bytenr,
1128					sf->limit_nesting,
1129					&sf->next_block_ctx,
1130					&sf->next_block,
1131					force_iodone_flag,
1132					&sf->num_copies,
1133					&sf->mirror_num,
1134					&key_ptr.key,
1135					btrfs_stack_key_generation(&key_ptr));
1136			if (sf->error)
1137				goto one_stack_frame_backwards;
1138
1139			if (NULL != sf->next_block) {
1140				struct btrfs_header *const next_hdr =
1141				    (struct btrfs_header *)
1142				    sf->next_block_ctx.datav[0];
1143
1144				next_stack = btrfsic_stack_frame_alloc();
1145				if (NULL == next_stack) {
1146					sf->error = -1;
1147					goto one_stack_frame_backwards;
1148				}
1149
1150				next_stack->i = -1;
1151				next_stack->block = sf->next_block;
1152				next_stack->block_ctx = &sf->next_block_ctx;
1153				next_stack->next_block = NULL;
1154				next_stack->hdr = next_hdr;
1155				next_stack->limit_nesting =
1156				    sf->limit_nesting - 1;
1157				next_stack->prev = sf;
1158				sf = next_stack;
1159				goto continue_with_new_stack_frame;
1160			}
1161
1162			goto continue_with_current_node_stack_frame;
1163		}
1164	}
1165
1166one_stack_frame_backwards:
1167	if (NULL != sf->prev) {
1168		struct btrfsic_stack_frame *const prev = sf->prev;
1169
1170		/* the one for the initial block is freed in the caller */
1171		btrfsic_release_block_ctx(sf->block_ctx);
1172
1173		if (sf->error) {
1174			prev->error = sf->error;
1175			btrfsic_stack_frame_free(sf);
1176			sf = prev;
1177			goto one_stack_frame_backwards;
1178		}
1179
1180		btrfsic_stack_frame_free(sf);
1181		sf = prev;
1182		goto continue_with_new_stack_frame;
1183	} else {
1184		BUG_ON(&initial_stack_frame != sf);
1185	}
1186
1187	return sf->error;
1188}
1189
1190static void btrfsic_read_from_block_data(
1191	struct btrfsic_block_data_ctx *block_ctx,
1192	void *dstv, u32 offset, size_t len)
1193{
1194	size_t cur;
1195	size_t pgoff;
1196	char *kaddr;
1197	char *dst = (char *)dstv;
1198	size_t start_offset = offset_in_page(block_ctx->start);
1199	unsigned long i = (start_offset + offset) >> PAGE_SHIFT;
1200
1201	WARN_ON(offset + len > block_ctx->len);
1202	pgoff = offset_in_page(start_offset + offset);
1203
1204	while (len > 0) {
1205		cur = min(len, ((size_t)PAGE_SIZE - pgoff));
1206		BUG_ON(i >= DIV_ROUND_UP(block_ctx->len, PAGE_SIZE));
1207		kaddr = block_ctx->datav[i];
1208		memcpy(dst, kaddr + pgoff, cur);
1209
1210		dst += cur;
1211		len -= cur;
1212		pgoff = 0;
1213		i++;
1214	}
1215}
1216
1217static int btrfsic_create_link_to_next_block(
1218		struct btrfsic_state *state,
1219		struct btrfsic_block *block,
1220		struct btrfsic_block_data_ctx *block_ctx,
1221		u64 next_bytenr,
1222		int limit_nesting,
1223		struct btrfsic_block_data_ctx *next_block_ctx,
1224		struct btrfsic_block **next_blockp,
1225		int force_iodone_flag,
1226		int *num_copiesp, int *mirror_nump,
1227		struct btrfs_disk_key *disk_key,
1228		u64 parent_generation)
1229{
1230	struct btrfs_fs_info *fs_info = state->fs_info;
1231	struct btrfsic_block *next_block = NULL;
1232	int ret;
1233	struct btrfsic_block_link *l;
1234	int did_alloc_block_link;
1235	int block_was_created;
1236
1237	*next_blockp = NULL;
1238	if (0 == *num_copiesp) {
1239		*num_copiesp = btrfs_num_copies(fs_info, next_bytenr,
1240						state->metablock_size);
1241		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1242			pr_info("num_copies(log_bytenr=%llu) = %d\n",
1243			       next_bytenr, *num_copiesp);
1244		*mirror_nump = 1;
1245	}
1246
1247	if (*mirror_nump > *num_copiesp)
1248		return 0;
1249
1250	if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1251		pr_info("btrfsic_create_link_to_next_block(mirror_num=%d)\n",
1252		       *mirror_nump);
1253	ret = btrfsic_map_block(state, next_bytenr,
1254				state->metablock_size,
1255				next_block_ctx, *mirror_nump);
1256	if (ret) {
1257		pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1258		       next_bytenr, *mirror_nump);
1259		btrfsic_release_block_ctx(next_block_ctx);
1260		*next_blockp = NULL;
1261		return -1;
1262	}
1263
1264	next_block = btrfsic_block_lookup_or_add(state,
1265						 next_block_ctx, "referenced ",
1266						 1, force_iodone_flag,
1267						 !force_iodone_flag,
1268						 *mirror_nump,
1269						 &block_was_created);
1270	if (NULL == next_block) {
1271		btrfsic_release_block_ctx(next_block_ctx);
1272		*next_blockp = NULL;
1273		return -1;
1274	}
1275	if (block_was_created) {
1276		l = NULL;
1277		next_block->generation = BTRFSIC_GENERATION_UNKNOWN;
1278	} else {
1279		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1280			if (next_block->logical_bytenr != next_bytenr &&
1281			    !(!next_block->is_metadata &&
1282			      0 == next_block->logical_bytenr))
1283				pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1284				       next_bytenr, next_block_ctx->dev->name,
 
1285				       next_block_ctx->dev_bytenr, *mirror_nump,
1286				       btrfsic_get_block_type(state,
1287							      next_block),
1288				       next_block->logical_bytenr);
1289			else
1290				pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1291				       next_bytenr, next_block_ctx->dev->name,
 
1292				       next_block_ctx->dev_bytenr, *mirror_nump,
1293				       btrfsic_get_block_type(state,
1294							      next_block));
1295		}
1296		next_block->logical_bytenr = next_bytenr;
1297
1298		next_block->mirror_num = *mirror_nump;
1299		l = btrfsic_block_link_hashtable_lookup(
1300				next_block_ctx->dev->bdev,
1301				next_block_ctx->dev_bytenr,
1302				block_ctx->dev->bdev,
1303				block_ctx->dev_bytenr,
1304				&state->block_link_hashtable);
1305	}
1306
1307	next_block->disk_key = *disk_key;
1308	if (NULL == l) {
1309		l = btrfsic_block_link_alloc();
1310		if (NULL == l) {
1311			btrfsic_release_block_ctx(next_block_ctx);
1312			*next_blockp = NULL;
1313			return -1;
1314		}
1315
1316		did_alloc_block_link = 1;
1317		l->block_ref_to = next_block;
1318		l->block_ref_from = block;
1319		l->ref_cnt = 1;
1320		l->parent_generation = parent_generation;
1321
1322		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1323			btrfsic_print_add_link(state, l);
1324
1325		list_add(&l->node_ref_to, &block->ref_to_list);
1326		list_add(&l->node_ref_from, &next_block->ref_from_list);
1327
1328		btrfsic_block_link_hashtable_add(l,
1329						 &state->block_link_hashtable);
1330	} else {
1331		did_alloc_block_link = 0;
1332		if (0 == limit_nesting) {
1333			l->ref_cnt++;
1334			l->parent_generation = parent_generation;
1335			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1336				btrfsic_print_add_link(state, l);
1337		}
1338	}
1339
1340	if (limit_nesting > 0 && did_alloc_block_link) {
1341		ret = btrfsic_read_block(state, next_block_ctx);
1342		if (ret < (int)next_block_ctx->len) {
1343			pr_info("btrfsic: read block @logical %llu failed!\n",
1344			       next_bytenr);
1345			btrfsic_release_block_ctx(next_block_ctx);
1346			*next_blockp = NULL;
1347			return -1;
1348		}
1349
1350		*next_blockp = next_block;
1351	} else {
1352		*next_blockp = NULL;
1353	}
1354	(*mirror_nump)++;
1355
1356	return 0;
1357}
1358
1359static int btrfsic_handle_extent_data(
1360		struct btrfsic_state *state,
1361		struct btrfsic_block *block,
1362		struct btrfsic_block_data_ctx *block_ctx,
1363		u32 item_offset, int force_iodone_flag)
1364{
1365	struct btrfs_fs_info *fs_info = state->fs_info;
1366	struct btrfs_file_extent_item file_extent_item;
1367	u64 file_extent_item_offset;
1368	u64 next_bytenr;
1369	u64 num_bytes;
1370	u64 generation;
1371	struct btrfsic_block_link *l;
1372	int ret;
1373
1374	file_extent_item_offset = offsetof(struct btrfs_leaf, items) +
1375				  item_offset;
1376	if (file_extent_item_offset +
1377	    offsetof(struct btrfs_file_extent_item, disk_num_bytes) >
1378	    block_ctx->len) {
1379		pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1380		       block_ctx->start, block_ctx->dev->name);
1381		return -1;
1382	}
1383
1384	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1385		file_extent_item_offset,
1386		offsetof(struct btrfs_file_extent_item, disk_num_bytes));
1387	if (BTRFS_FILE_EXTENT_REG != file_extent_item.type ||
1388	    btrfs_stack_file_extent_disk_bytenr(&file_extent_item) == 0) {
1389		if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1390			pr_info("extent_data: type %u, disk_bytenr = %llu\n",
1391			       file_extent_item.type,
1392			       btrfs_stack_file_extent_disk_bytenr(
1393			       &file_extent_item));
1394		return 0;
1395	}
1396
1397	if (file_extent_item_offset + sizeof(struct btrfs_file_extent_item) >
1398	    block_ctx->len) {
1399		pr_info("btrfsic: file item out of bounce at logical %llu, dev %s\n",
1400		       block_ctx->start, block_ctx->dev->name);
1401		return -1;
1402	}
1403	btrfsic_read_from_block_data(block_ctx, &file_extent_item,
1404				     file_extent_item_offset,
1405				     sizeof(struct btrfs_file_extent_item));
1406	next_bytenr = btrfs_stack_file_extent_disk_bytenr(&file_extent_item);
1407	if (btrfs_stack_file_extent_compression(&file_extent_item) ==
1408	    BTRFS_COMPRESS_NONE) {
1409		next_bytenr += btrfs_stack_file_extent_offset(&file_extent_item);
1410		num_bytes = btrfs_stack_file_extent_num_bytes(&file_extent_item);
1411	} else {
1412		num_bytes = btrfs_stack_file_extent_disk_num_bytes(&file_extent_item);
1413	}
1414	generation = btrfs_stack_file_extent_generation(&file_extent_item);
1415
1416	if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1417		pr_info("extent_data: type %u, disk_bytenr = %llu, offset = %llu, num_bytes = %llu\n",
1418		       file_extent_item.type,
1419		       btrfs_stack_file_extent_disk_bytenr(&file_extent_item),
1420		       btrfs_stack_file_extent_offset(&file_extent_item),
1421		       num_bytes);
1422	while (num_bytes > 0) {
1423		u32 chunk_len;
1424		int num_copies;
1425		int mirror_num;
1426
1427		if (num_bytes > state->datablock_size)
1428			chunk_len = state->datablock_size;
1429		else
1430			chunk_len = num_bytes;
1431
1432		num_copies = btrfs_num_copies(fs_info, next_bytenr,
1433					      state->datablock_size);
1434		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
1435			pr_info("num_copies(log_bytenr=%llu) = %d\n",
1436			       next_bytenr, num_copies);
1437		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
1438			struct btrfsic_block_data_ctx next_block_ctx;
1439			struct btrfsic_block *next_block;
1440			int block_was_created;
1441
1442			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1443				pr_info("btrfsic_handle_extent_data(mirror_num=%d)\n",
1444					mirror_num);
1445			if (state->print_mask & BTRFSIC_PRINT_MASK_VERY_VERBOSE)
1446				pr_info("\tdisk_bytenr = %llu, num_bytes %u\n",
1447				       next_bytenr, chunk_len);
1448			ret = btrfsic_map_block(state, next_bytenr,
1449						chunk_len, &next_block_ctx,
1450						mirror_num);
1451			if (ret) {
1452				pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
1453				       next_bytenr, mirror_num);
1454				return -1;
1455			}
1456
1457			next_block = btrfsic_block_lookup_or_add(
1458					state,
1459					&next_block_ctx,
1460					"referenced ",
1461					0,
1462					force_iodone_flag,
1463					!force_iodone_flag,
1464					mirror_num,
1465					&block_was_created);
1466			if (NULL == next_block) {
1467				btrfsic_release_block_ctx(&next_block_ctx);
1468				return -1;
1469			}
1470			if (!block_was_created) {
1471				if ((state->print_mask &
1472				     BTRFSIC_PRINT_MASK_VERBOSE) &&
1473				    next_block->logical_bytenr != next_bytenr &&
1474				    !(!next_block->is_metadata &&
1475				      0 == next_block->logical_bytenr)) {
1476					pr_info("Referenced block @%llu (%s/%llu/%d) found in hash table, D, bytenr mismatch (!= stored %llu).\n",
 
1477					       next_bytenr,
1478					       next_block_ctx.dev->name,
1479					       next_block_ctx.dev_bytenr,
1480					       mirror_num,
1481					       next_block->logical_bytenr);
1482				}
1483				next_block->logical_bytenr = next_bytenr;
1484				next_block->mirror_num = mirror_num;
1485			}
1486
1487			l = btrfsic_block_link_lookup_or_add(state,
1488							     &next_block_ctx,
1489							     next_block, block,
1490							     generation);
1491			btrfsic_release_block_ctx(&next_block_ctx);
1492			if (NULL == l)
1493				return -1;
1494		}
1495
1496		next_bytenr += chunk_len;
1497		num_bytes -= chunk_len;
1498	}
1499
1500	return 0;
1501}
1502
1503static int btrfsic_map_block(struct btrfsic_state *state, u64 bytenr, u32 len,
1504			     struct btrfsic_block_data_ctx *block_ctx_out,
1505			     int mirror_num)
1506{
1507	struct btrfs_fs_info *fs_info = state->fs_info;
1508	int ret;
1509	u64 length;
1510	struct btrfs_bio *multi = NULL;
1511	struct btrfs_device *device;
1512
1513	length = len;
1514	ret = btrfs_map_block(fs_info, BTRFS_MAP_READ,
1515			      bytenr, &length, &multi, mirror_num);
1516
1517	if (ret) {
1518		block_ctx_out->start = 0;
1519		block_ctx_out->dev_bytenr = 0;
1520		block_ctx_out->len = 0;
1521		block_ctx_out->dev = NULL;
1522		block_ctx_out->datav = NULL;
1523		block_ctx_out->pagev = NULL;
1524		block_ctx_out->mem_to_free = NULL;
1525
1526		return ret;
1527	}
1528
1529	device = multi->stripes[0].dev;
1530	if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
1531	    !device->bdev || !device->name)
1532		block_ctx_out->dev = NULL;
1533	else
1534		block_ctx_out->dev = btrfsic_dev_state_lookup(
1535							device->bdev->bd_dev);
1536	block_ctx_out->dev_bytenr = multi->stripes[0].physical;
1537	block_ctx_out->start = bytenr;
1538	block_ctx_out->len = len;
1539	block_ctx_out->datav = NULL;
1540	block_ctx_out->pagev = NULL;
1541	block_ctx_out->mem_to_free = NULL;
1542
1543	kfree(multi);
1544	if (NULL == block_ctx_out->dev) {
1545		ret = -ENXIO;
1546		pr_info("btrfsic: error, cannot lookup dev (#1)!\n");
1547	}
1548
1549	return ret;
1550}
1551
1552static void btrfsic_release_block_ctx(struct btrfsic_block_data_ctx *block_ctx)
1553{
1554	if (block_ctx->mem_to_free) {
1555		unsigned int num_pages;
1556
1557		BUG_ON(!block_ctx->datav);
1558		BUG_ON(!block_ctx->pagev);
1559		num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1560			    PAGE_SHIFT;
 
1561		while (num_pages > 0) {
1562			num_pages--;
1563			if (block_ctx->datav[num_pages]) {
1564				kunmap(block_ctx->pagev[num_pages]);
1565				block_ctx->datav[num_pages] = NULL;
1566			}
1567			if (block_ctx->pagev[num_pages]) {
1568				__free_page(block_ctx->pagev[num_pages]);
1569				block_ctx->pagev[num_pages] = NULL;
1570			}
1571		}
1572
1573		kfree(block_ctx->mem_to_free);
1574		block_ctx->mem_to_free = NULL;
1575		block_ctx->pagev = NULL;
1576		block_ctx->datav = NULL;
1577	}
1578}
1579
1580static int btrfsic_read_block(struct btrfsic_state *state,
1581			      struct btrfsic_block_data_ctx *block_ctx)
1582{
1583	unsigned int num_pages;
1584	unsigned int i;
1585	size_t size;
1586	u64 dev_bytenr;
1587	int ret;
1588
1589	BUG_ON(block_ctx->datav);
1590	BUG_ON(block_ctx->pagev);
1591	BUG_ON(block_ctx->mem_to_free);
1592	if (!PAGE_ALIGNED(block_ctx->dev_bytenr)) {
1593		pr_info("btrfsic: read_block() with unaligned bytenr %llu\n",
1594		       block_ctx->dev_bytenr);
1595		return -1;
1596	}
1597
1598	num_pages = (block_ctx->len + (u64)PAGE_SIZE - 1) >>
1599		    PAGE_SHIFT;
1600	size = sizeof(*block_ctx->datav) + sizeof(*block_ctx->pagev);
1601	block_ctx->mem_to_free = kcalloc(num_pages, size, GFP_NOFS);
1602	if (!block_ctx->mem_to_free)
1603		return -ENOMEM;
1604	block_ctx->datav = block_ctx->mem_to_free;
1605	block_ctx->pagev = (struct page **)(block_ctx->datav + num_pages);
1606	for (i = 0; i < num_pages; i++) {
1607		block_ctx->pagev[i] = alloc_page(GFP_NOFS);
1608		if (!block_ctx->pagev[i])
1609			return -1;
1610	}
1611
1612	dev_bytenr = block_ctx->dev_bytenr;
1613	for (i = 0; i < num_pages;) {
1614		struct bio *bio;
1615		unsigned int j;
1616
1617		bio = btrfs_io_bio_alloc(num_pages - i);
1618		bio_set_dev(bio, block_ctx->dev->bdev);
1619		bio->bi_iter.bi_sector = dev_bytenr >> 9;
1620		bio->bi_opf = REQ_OP_READ;
1621
1622		for (j = i; j < num_pages; j++) {
1623			ret = bio_add_page(bio, block_ctx->pagev[j],
1624					   PAGE_SIZE, 0);
1625			if (PAGE_SIZE != ret)
1626				break;
1627		}
1628		if (j == i) {
1629			pr_info("btrfsic: error, failed to add a single page!\n");
1630			return -1;
1631		}
1632		if (submit_bio_wait(bio)) {
1633			pr_info("btrfsic: read error at logical %llu dev %s!\n",
1634			       block_ctx->start, block_ctx->dev->name);
1635			bio_put(bio);
1636			return -1;
1637		}
1638		bio_put(bio);
1639		dev_bytenr += (j - i) * PAGE_SIZE;
1640		i = j;
1641	}
1642	for (i = 0; i < num_pages; i++)
1643		block_ctx->datav[i] = kmap(block_ctx->pagev[i]);
1644
1645	return block_ctx->len;
1646}
1647
1648static void btrfsic_dump_database(struct btrfsic_state *state)
1649{
1650	const struct btrfsic_block *b_all;
1651
1652	BUG_ON(NULL == state);
1653
1654	pr_info("all_blocks_list:\n");
1655	list_for_each_entry(b_all, &state->all_blocks_list, all_blocks_node) {
1656		const struct btrfsic_block_link *l;
1657
1658		pr_info("%c-block @%llu (%s/%llu/%d)\n",
1659		       btrfsic_get_block_type(state, b_all),
1660		       b_all->logical_bytenr, b_all->dev_state->name,
1661		       b_all->dev_bytenr, b_all->mirror_num);
1662
1663		list_for_each_entry(l, &b_all->ref_to_list, node_ref_to) {
1664			pr_info(" %c @%llu (%s/%llu/%d) refers %u* to %c @%llu (%s/%llu/%d)\n",
 
1665			       btrfsic_get_block_type(state, b_all),
1666			       b_all->logical_bytenr, b_all->dev_state->name,
1667			       b_all->dev_bytenr, b_all->mirror_num,
1668			       l->ref_cnt,
1669			       btrfsic_get_block_type(state, l->block_ref_to),
1670			       l->block_ref_to->logical_bytenr,
1671			       l->block_ref_to->dev_state->name,
1672			       l->block_ref_to->dev_bytenr,
1673			       l->block_ref_to->mirror_num);
1674		}
1675
1676		list_for_each_entry(l, &b_all->ref_from_list, node_ref_from) {
1677			pr_info(" %c @%llu (%s/%llu/%d) is ref %u* from %c @%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			       l->ref_cnt,
1682			       btrfsic_get_block_type(state, l->block_ref_from),
1683			       l->block_ref_from->logical_bytenr,
1684			       l->block_ref_from->dev_state->name,
1685			       l->block_ref_from->dev_bytenr,
1686			       l->block_ref_from->mirror_num);
1687		}
1688
1689		pr_info("\n");
1690	}
1691}
1692
1693/*
1694 * Test whether the disk block contains a tree block (leaf or node)
1695 * (note that this test fails for the super block)
1696 */
1697static noinline_for_stack int btrfsic_test_for_metadata(
1698		struct btrfsic_state *state,
1699		char **datav, unsigned int num_pages)
1700{
1701	struct btrfs_fs_info *fs_info = state->fs_info;
1702	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
1703	struct btrfs_header *h;
1704	u8 csum[BTRFS_CSUM_SIZE];
1705	unsigned int i;
1706
1707	if (num_pages * PAGE_SIZE < state->metablock_size)
1708		return 1; /* not metadata */
1709	num_pages = state->metablock_size >> PAGE_SHIFT;
1710	h = (struct btrfs_header *)datav[0];
1711
1712	if (memcmp(h->fsid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE))
1713		return 1;
1714
1715	shash->tfm = fs_info->csum_shash;
1716	crypto_shash_init(shash);
1717
1718	for (i = 0; i < num_pages; i++) {
1719		u8 *data = i ? datav[i] : (datav[i] + BTRFS_CSUM_SIZE);
1720		size_t sublen = i ? PAGE_SIZE :
1721				    (PAGE_SIZE - BTRFS_CSUM_SIZE);
1722
1723		crypto_shash_update(shash, data, sublen);
1724	}
1725	crypto_shash_final(shash, csum);
1726	if (memcmp(csum, h->csum, state->csum_size))
1727		return 1;
1728
1729	return 0; /* is metadata */
1730}
1731
1732static void btrfsic_process_written_block(struct btrfsic_dev_state *dev_state,
1733					  u64 dev_bytenr, char **mapped_datav,
1734					  unsigned int num_pages,
1735					  struct bio *bio, int *bio_is_patched,
1736					  int submit_bio_bh_rw)
1737{
1738	int is_metadata;
1739	struct btrfsic_block *block;
1740	struct btrfsic_block_data_ctx block_ctx;
1741	int ret;
1742	struct btrfsic_state *state = dev_state->state;
1743	struct block_device *bdev = dev_state->bdev;
1744	unsigned int processed_len;
1745
1746	if (NULL != bio_is_patched)
1747		*bio_is_patched = 0;
1748
1749again:
1750	if (num_pages == 0)
1751		return;
1752
1753	processed_len = 0;
1754	is_metadata = (0 == btrfsic_test_for_metadata(state, mapped_datav,
1755						      num_pages));
1756
1757	block = btrfsic_block_hashtable_lookup(bdev, dev_bytenr,
1758					       &state->block_hashtable);
1759	if (NULL != block) {
1760		u64 bytenr = 0;
1761		struct btrfsic_block_link *l, *tmp;
1762
1763		if (block->is_superblock) {
1764			bytenr = btrfs_super_bytenr((struct btrfs_super_block *)
1765						    mapped_datav[0]);
1766			if (num_pages * PAGE_SIZE <
1767			    BTRFS_SUPER_INFO_SIZE) {
1768				pr_info("btrfsic: cannot work with too short bios!\n");
1769				return;
1770			}
1771			is_metadata = 1;
1772			BUG_ON(!PAGE_ALIGNED(BTRFS_SUPER_INFO_SIZE));
1773			processed_len = BTRFS_SUPER_INFO_SIZE;
1774			if (state->print_mask &
1775			    BTRFSIC_PRINT_MASK_TREE_BEFORE_SB_WRITE) {
1776				pr_info("[before new superblock is written]:\n");
1777				btrfsic_dump_tree_sub(state, block, 0);
1778			}
1779		}
1780		if (is_metadata) {
1781			if (!block->is_superblock) {
1782				if (num_pages * PAGE_SIZE <
1783				    state->metablock_size) {
1784					pr_info("btrfsic: cannot work with too short bios!\n");
1785					return;
1786				}
1787				processed_len = state->metablock_size;
1788				bytenr = btrfs_stack_header_bytenr(
1789						(struct btrfs_header *)
1790						mapped_datav[0]);
1791				btrfsic_cmp_log_and_dev_bytenr(state, bytenr,
1792							       dev_state,
1793							       dev_bytenr);
1794			}
1795			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE) {
1796				if (block->logical_bytenr != bytenr &&
1797				    !(!block->is_metadata &&
1798				      block->logical_bytenr == 0))
1799					pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c, bytenr mismatch (!= stored %llu).\n",
1800					       bytenr, dev_state->name,
 
1801					       dev_bytenr,
1802					       block->mirror_num,
1803					       btrfsic_get_block_type(state,
1804								      block),
1805					       block->logical_bytenr);
1806				else
1807					pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1808					       bytenr, dev_state->name,
 
1809					       dev_bytenr, block->mirror_num,
1810					       btrfsic_get_block_type(state,
1811								      block));
1812			}
1813			block->logical_bytenr = bytenr;
1814		} else {
1815			if (num_pages * PAGE_SIZE <
1816			    state->datablock_size) {
1817				pr_info("btrfsic: cannot work with too short bios!\n");
1818				return;
1819			}
1820			processed_len = state->datablock_size;
1821			bytenr = block->logical_bytenr;
1822			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1823				pr_info("Written block @%llu (%s/%llu/%d) found in hash table, %c.\n",
1824				       bytenr, dev_state->name, dev_bytenr,
 
1825				       block->mirror_num,
1826				       btrfsic_get_block_type(state, block));
1827		}
1828
1829		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1830			pr_info("ref_to_list: %cE, ref_from_list: %cE\n",
1831			       list_empty(&block->ref_to_list) ? ' ' : '!',
1832			       list_empty(&block->ref_from_list) ? ' ' : '!');
1833		if (btrfsic_is_block_ref_by_superblock(state, block, 0)) {
1834			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",
 
1835			       btrfsic_get_block_type(state, block), bytenr,
1836			       dev_state->name, dev_bytenr, block->mirror_num,
1837			       block->generation,
1838			       btrfs_disk_key_objectid(&block->disk_key),
1839			       block->disk_key.type,
1840			       btrfs_disk_key_offset(&block->disk_key),
1841			       btrfs_stack_header_generation(
1842				       (struct btrfs_header *) mapped_datav[0]),
1843			       state->max_superblock_generation);
1844			btrfsic_dump_tree(state);
1845		}
1846
1847		if (!block->is_iodone && !block->never_written) {
1848			pr_info("btrfs: attempt to overwrite %c-block @%llu (%s/%llu/%d), oldgen=%llu, newgen=%llu, which is not yet iodone!\n",
 
1849			       btrfsic_get_block_type(state, block), bytenr,
1850			       dev_state->name, dev_bytenr, block->mirror_num,
1851			       block->generation,
1852			       btrfs_stack_header_generation(
1853				       (struct btrfs_header *)
1854				       mapped_datav[0]));
1855			/* it would not be safe to go on */
1856			btrfsic_dump_tree(state);
1857			goto continue_loop;
1858		}
1859
1860		/*
1861		 * Clear all references of this block. Do not free
1862		 * the block itself even if is not referenced anymore
1863		 * because it still carries valuable information
1864		 * like whether it was ever written and IO completed.
1865		 */
1866		list_for_each_entry_safe(l, tmp, &block->ref_to_list,
1867					 node_ref_to) {
1868			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1869				btrfsic_print_rem_link(state, l);
1870			l->ref_cnt--;
1871			if (0 == l->ref_cnt) {
1872				list_del(&l->node_ref_to);
1873				list_del(&l->node_ref_from);
1874				btrfsic_block_link_hashtable_remove(l);
1875				btrfsic_block_link_free(l);
1876			}
1877		}
1878
1879		block_ctx.dev = dev_state;
1880		block_ctx.dev_bytenr = dev_bytenr;
1881		block_ctx.start = bytenr;
1882		block_ctx.len = processed_len;
1883		block_ctx.pagev = NULL;
1884		block_ctx.mem_to_free = NULL;
1885		block_ctx.datav = mapped_datav;
1886
1887		if (is_metadata || state->include_extent_data) {
1888			block->never_written = 0;
1889			block->iodone_w_error = 0;
1890			if (NULL != bio) {
1891				block->is_iodone = 0;
1892				BUG_ON(NULL == bio_is_patched);
1893				if (!*bio_is_patched) {
1894					block->orig_bio_private =
1895					    bio->bi_private;
1896					block->orig_bio_end_io =
1897					    bio->bi_end_io;
1898					block->next_in_same_bio = NULL;
1899					bio->bi_private = block;
1900					bio->bi_end_io = btrfsic_bio_end_io;
1901					*bio_is_patched = 1;
1902				} else {
1903					struct btrfsic_block *chained_block =
1904					    (struct btrfsic_block *)
1905					    bio->bi_private;
1906
1907					BUG_ON(NULL == chained_block);
1908					block->orig_bio_private =
1909					    chained_block->orig_bio_private;
1910					block->orig_bio_end_io =
1911					    chained_block->orig_bio_end_io;
1912					block->next_in_same_bio = chained_block;
1913					bio->bi_private = block;
1914				}
1915			} else {
1916				block->is_iodone = 1;
1917				block->orig_bio_private = NULL;
1918				block->orig_bio_end_io = NULL;
1919				block->next_in_same_bio = NULL;
1920			}
1921		}
1922
1923		block->flush_gen = dev_state->last_flush_gen + 1;
1924		block->submit_bio_bh_rw = submit_bio_bh_rw;
1925		if (is_metadata) {
1926			block->logical_bytenr = bytenr;
1927			block->is_metadata = 1;
1928			if (block->is_superblock) {
1929				BUG_ON(PAGE_SIZE !=
1930				       BTRFS_SUPER_INFO_SIZE);
1931				ret = btrfsic_process_written_superblock(
1932						state,
1933						block,
1934						(struct btrfs_super_block *)
1935						mapped_datav[0]);
1936				if (state->print_mask &
1937				    BTRFSIC_PRINT_MASK_TREE_AFTER_SB_WRITE) {
1938					pr_info("[after new superblock is written]:\n");
1939					btrfsic_dump_tree_sub(state, block, 0);
1940				}
1941			} else {
1942				block->mirror_num = 0;	/* unknown */
1943				ret = btrfsic_process_metablock(
1944						state,
1945						block,
1946						&block_ctx,
1947						0, 0);
1948			}
1949			if (ret)
1950				pr_info("btrfsic: btrfsic_process_metablock(root @%llu) failed!\n",
1951				       dev_bytenr);
1952		} else {
1953			block->is_metadata = 0;
1954			block->mirror_num = 0;	/* unknown */
1955			block->generation = BTRFSIC_GENERATION_UNKNOWN;
1956			if (!state->include_extent_data
1957			    && list_empty(&block->ref_from_list)) {
1958				/*
1959				 * disk block is overwritten with extent
1960				 * data (not meta data) and we are configured
1961				 * to not include extent data: take the
1962				 * chance and free the block's memory
1963				 */
1964				btrfsic_block_hashtable_remove(block);
1965				list_del(&block->all_blocks_node);
1966				btrfsic_block_free(block);
1967			}
1968		}
1969		btrfsic_release_block_ctx(&block_ctx);
1970	} else {
1971		/* block has not been found in hash table */
1972		u64 bytenr;
1973
1974		if (!is_metadata) {
1975			processed_len = state->datablock_size;
1976			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1977				pr_info("Written block (%s/%llu/?) !found in hash table, D.\n",
1978				       dev_state->name, dev_bytenr);
 
1979			if (!state->include_extent_data) {
1980				/* ignore that written D block */
1981				goto continue_loop;
1982			}
1983
1984			/* this is getting ugly for the
1985			 * include_extent_data case... */
1986			bytenr = 0;	/* unknown */
1987		} else {
1988			processed_len = state->metablock_size;
1989			bytenr = btrfs_stack_header_bytenr(
1990					(struct btrfs_header *)
1991					mapped_datav[0]);
1992			btrfsic_cmp_log_and_dev_bytenr(state, bytenr, dev_state,
1993						       dev_bytenr);
1994			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
1995				pr_info("Written block @%llu (%s/%llu/?) !found in hash table, M.\n",
1996				       bytenr, dev_state->name, dev_bytenr);
 
1997		}
1998
1999		block_ctx.dev = dev_state;
2000		block_ctx.dev_bytenr = dev_bytenr;
2001		block_ctx.start = bytenr;
2002		block_ctx.len = processed_len;
2003		block_ctx.pagev = NULL;
2004		block_ctx.mem_to_free = NULL;
2005		block_ctx.datav = mapped_datav;
2006
2007		block = btrfsic_block_alloc();
2008		if (NULL == block) {
2009			btrfsic_release_block_ctx(&block_ctx);
2010			goto continue_loop;
2011		}
2012		block->dev_state = dev_state;
2013		block->dev_bytenr = dev_bytenr;
2014		block->logical_bytenr = bytenr;
2015		block->is_metadata = is_metadata;
2016		block->never_written = 0;
2017		block->iodone_w_error = 0;
2018		block->mirror_num = 0;	/* unknown */
2019		block->flush_gen = dev_state->last_flush_gen + 1;
2020		block->submit_bio_bh_rw = submit_bio_bh_rw;
2021		if (NULL != bio) {
2022			block->is_iodone = 0;
2023			BUG_ON(NULL == bio_is_patched);
2024			if (!*bio_is_patched) {
2025				block->orig_bio_private = bio->bi_private;
2026				block->orig_bio_end_io = bio->bi_end_io;
2027				block->next_in_same_bio = NULL;
2028				bio->bi_private = block;
2029				bio->bi_end_io = btrfsic_bio_end_io;
2030				*bio_is_patched = 1;
2031			} else {
2032				struct btrfsic_block *chained_block =
2033				    (struct btrfsic_block *)
2034				    bio->bi_private;
2035
2036				BUG_ON(NULL == chained_block);
2037				block->orig_bio_private =
2038				    chained_block->orig_bio_private;
2039				block->orig_bio_end_io =
2040				    chained_block->orig_bio_end_io;
2041				block->next_in_same_bio = chained_block;
2042				bio->bi_private = block;
2043			}
2044		} else {
2045			block->is_iodone = 1;
2046			block->orig_bio_private = NULL;
2047			block->orig_bio_end_io = NULL;
2048			block->next_in_same_bio = NULL;
2049		}
2050		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2051			pr_info("New written %c-block @%llu (%s/%llu/%d)\n",
2052			       is_metadata ? 'M' : 'D',
2053			       block->logical_bytenr, block->dev_state->name,
2054			       block->dev_bytenr, block->mirror_num);
2055		list_add(&block->all_blocks_node, &state->all_blocks_list);
2056		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2057
2058		if (is_metadata) {
2059			ret = btrfsic_process_metablock(state, block,
2060							&block_ctx, 0, 0);
2061			if (ret)
2062				pr_info("btrfsic: process_metablock(root @%llu) failed!\n",
2063				       dev_bytenr);
2064		}
2065		btrfsic_release_block_ctx(&block_ctx);
2066	}
2067
2068continue_loop:
2069	BUG_ON(!processed_len);
2070	dev_bytenr += processed_len;
2071	mapped_datav += processed_len >> PAGE_SHIFT;
2072	num_pages -= processed_len >> PAGE_SHIFT;
2073	goto again;
2074}
2075
2076static void btrfsic_bio_end_io(struct bio *bp)
2077{
2078	struct btrfsic_block *block = (struct btrfsic_block *)bp->bi_private;
2079	int iodone_w_error;
2080
2081	/* mutex is not held! This is not save if IO is not yet completed
2082	 * on umount */
2083	iodone_w_error = 0;
2084	if (bp->bi_status)
2085		iodone_w_error = 1;
2086
2087	BUG_ON(NULL == block);
2088	bp->bi_private = block->orig_bio_private;
2089	bp->bi_end_io = block->orig_bio_end_io;
2090
2091	do {
2092		struct btrfsic_block *next_block;
2093		struct btrfsic_dev_state *const dev_state = block->dev_state;
2094
2095		if ((dev_state->state->print_mask &
2096		     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2097			pr_info("bio_end_io(err=%d) for %c @%llu (%s/%llu/%d)\n",
2098			       bp->bi_status,
2099			       btrfsic_get_block_type(dev_state->state, block),
2100			       block->logical_bytenr, dev_state->name,
2101			       block->dev_bytenr, block->mirror_num);
2102		next_block = block->next_in_same_bio;
2103		block->iodone_w_error = iodone_w_error;
2104		if (block->submit_bio_bh_rw & REQ_PREFLUSH) {
2105			dev_state->last_flush_gen++;
2106			if ((dev_state->state->print_mask &
2107			     BTRFSIC_PRINT_MASK_END_IO_BIO_BH))
2108				pr_info("bio_end_io() new %s flush_gen=%llu\n",
2109				       dev_state->name,
2110				       dev_state->last_flush_gen);
2111		}
2112		if (block->submit_bio_bh_rw & REQ_FUA)
2113			block->flush_gen = 0; /* FUA completed means block is
2114					       * on disk */
2115		block->is_iodone = 1; /* for FLUSH, this releases the block */
2116		block = next_block;
2117	} while (NULL != block);
2118
2119	bp->bi_end_io(bp);
2120}
2121
2122static int btrfsic_process_written_superblock(
2123		struct btrfsic_state *state,
2124		struct btrfsic_block *const superblock,
2125		struct btrfs_super_block *const super_hdr)
2126{
2127	struct btrfs_fs_info *fs_info = state->fs_info;
2128	int pass;
2129
2130	superblock->generation = btrfs_super_generation(super_hdr);
2131	if (!(superblock->generation > state->max_superblock_generation ||
2132	      0 == state->max_superblock_generation)) {
2133		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2134			pr_info("btrfsic: superblock @%llu (%s/%llu/%d) with old gen %llu <= %llu\n",
 
2135			       superblock->logical_bytenr,
2136			       superblock->dev_state->name,
2137			       superblock->dev_bytenr, superblock->mirror_num,
2138			       btrfs_super_generation(super_hdr),
2139			       state->max_superblock_generation);
2140	} else {
2141		if (state->print_mask & BTRFSIC_PRINT_MASK_SUPERBLOCK_WRITE)
2142			pr_info("btrfsic: got new superblock @%llu (%s/%llu/%d) with new gen %llu > %llu\n",
 
2143			       superblock->logical_bytenr,
2144			       superblock->dev_state->name,
2145			       superblock->dev_bytenr, superblock->mirror_num,
2146			       btrfs_super_generation(super_hdr),
2147			       state->max_superblock_generation);
2148
2149		state->max_superblock_generation =
2150		    btrfs_super_generation(super_hdr);
2151		state->latest_superblock = superblock;
2152	}
2153
2154	for (pass = 0; pass < 3; pass++) {
2155		int ret;
2156		u64 next_bytenr;
2157		struct btrfsic_block *next_block;
2158		struct btrfsic_block_data_ctx tmp_next_block_ctx;
2159		struct btrfsic_block_link *l;
2160		int num_copies;
2161		int mirror_num;
2162		const char *additional_string = NULL;
2163		struct btrfs_disk_key tmp_disk_key = {0};
2164
2165		btrfs_set_disk_key_objectid(&tmp_disk_key,
2166					    BTRFS_ROOT_ITEM_KEY);
2167		btrfs_set_disk_key_objectid(&tmp_disk_key, 0);
2168
2169		switch (pass) {
2170		case 0:
2171			btrfs_set_disk_key_objectid(&tmp_disk_key,
2172						    BTRFS_ROOT_TREE_OBJECTID);
2173			additional_string = "root ";
2174			next_bytenr = btrfs_super_root(super_hdr);
2175			if (state->print_mask &
2176			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2177				pr_info("root@%llu\n", next_bytenr);
2178			break;
2179		case 1:
2180			btrfs_set_disk_key_objectid(&tmp_disk_key,
2181						    BTRFS_CHUNK_TREE_OBJECTID);
2182			additional_string = "chunk ";
2183			next_bytenr = btrfs_super_chunk_root(super_hdr);
2184			if (state->print_mask &
2185			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2186				pr_info("chunk@%llu\n", next_bytenr);
2187			break;
2188		case 2:
2189			btrfs_set_disk_key_objectid(&tmp_disk_key,
2190						    BTRFS_TREE_LOG_OBJECTID);
2191			additional_string = "log ";
2192			next_bytenr = btrfs_super_log_root(super_hdr);
2193			if (0 == next_bytenr)
2194				continue;
2195			if (state->print_mask &
2196			    BTRFSIC_PRINT_MASK_ROOT_CHUNK_LOG_TREE_LOCATION)
2197				pr_info("log@%llu\n", next_bytenr);
2198			break;
2199		}
2200
2201		num_copies = btrfs_num_copies(fs_info, next_bytenr,
2202					      BTRFS_SUPER_INFO_SIZE);
2203		if (state->print_mask & BTRFSIC_PRINT_MASK_NUM_COPIES)
2204			pr_info("num_copies(log_bytenr=%llu) = %d\n",
2205			       next_bytenr, num_copies);
2206		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2207			int was_created;
2208
2209			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2210				pr_info("btrfsic_process_written_superblock(mirror_num=%d)\n", mirror_num);
2211			ret = btrfsic_map_block(state, next_bytenr,
2212						BTRFS_SUPER_INFO_SIZE,
2213						&tmp_next_block_ctx,
2214						mirror_num);
2215			if (ret) {
2216				pr_info("btrfsic: btrfsic_map_block(@%llu, mirror=%d) failed!\n",
2217				       next_bytenr, mirror_num);
2218				return -1;
2219			}
2220
2221			next_block = btrfsic_block_lookup_or_add(
2222					state,
2223					&tmp_next_block_ctx,
2224					additional_string,
2225					1, 0, 1,
2226					mirror_num,
2227					&was_created);
2228			if (NULL == next_block) {
2229				btrfsic_release_block_ctx(&tmp_next_block_ctx);
2230				return -1;
2231			}
2232
2233			next_block->disk_key = tmp_disk_key;
2234			if (was_created)
2235				next_block->generation =
2236				    BTRFSIC_GENERATION_UNKNOWN;
2237			l = btrfsic_block_link_lookup_or_add(
2238					state,
2239					&tmp_next_block_ctx,
2240					next_block,
2241					superblock,
2242					BTRFSIC_GENERATION_UNKNOWN);
2243			btrfsic_release_block_ctx(&tmp_next_block_ctx);
2244			if (NULL == l)
2245				return -1;
2246		}
2247	}
2248
2249	if (WARN_ON(-1 == btrfsic_check_all_ref_blocks(state, superblock, 0)))
2250		btrfsic_dump_tree(state);
2251
2252	return 0;
2253}
2254
2255static int btrfsic_check_all_ref_blocks(struct btrfsic_state *state,
2256					struct btrfsic_block *const block,
2257					int recursion_level)
2258{
2259	const struct btrfsic_block_link *l;
2260	int ret = 0;
2261
2262	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2263		/*
2264		 * Note that this situation can happen and does not
2265		 * indicate an error in regular cases. It happens
2266		 * when disk blocks are freed and later reused.
2267		 * The check-integrity module is not aware of any
2268		 * block free operations, it just recognizes block
2269		 * write operations. Therefore it keeps the linkage
2270		 * information for a block until a block is
2271		 * rewritten. This can temporarily cause incorrect
2272		 * and even circular linkage information. This
2273		 * causes no harm unless such blocks are referenced
2274		 * by the most recent super block.
2275		 */
2276		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2277			pr_info("btrfsic: abort cyclic linkage (case 1).\n");
2278
2279		return ret;
2280	}
2281
2282	/*
2283	 * This algorithm is recursive because the amount of used stack
2284	 * space is very small and the max recursion depth is limited.
2285	 */
2286	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2287		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2288			pr_info("rl=%d, %c @%llu (%s/%llu/%d) %u* refers to %c @%llu (%s/%llu/%d)\n",
 
2289			       recursion_level,
2290			       btrfsic_get_block_type(state, block),
2291			       block->logical_bytenr, block->dev_state->name,
2292			       block->dev_bytenr, block->mirror_num,
2293			       l->ref_cnt,
2294			       btrfsic_get_block_type(state, l->block_ref_to),
2295			       l->block_ref_to->logical_bytenr,
2296			       l->block_ref_to->dev_state->name,
2297			       l->block_ref_to->dev_bytenr,
2298			       l->block_ref_to->mirror_num);
2299		if (l->block_ref_to->never_written) {
2300			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is never written!\n",
 
2301			       btrfsic_get_block_type(state, l->block_ref_to),
2302			       l->block_ref_to->logical_bytenr,
2303			       l->block_ref_to->dev_state->name,
2304			       l->block_ref_to->dev_bytenr,
2305			       l->block_ref_to->mirror_num);
2306			ret = -1;
2307		} else if (!l->block_ref_to->is_iodone) {
2308			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which is not yet iodone!\n",
 
2309			       btrfsic_get_block_type(state, l->block_ref_to),
2310			       l->block_ref_to->logical_bytenr,
2311			       l->block_ref_to->dev_state->name,
2312			       l->block_ref_to->dev_bytenr,
2313			       l->block_ref_to->mirror_num);
2314			ret = -1;
2315		} else if (l->block_ref_to->iodone_w_error) {
2316			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) which has write error!\n",
 
2317			       btrfsic_get_block_type(state, l->block_ref_to),
2318			       l->block_ref_to->logical_bytenr,
2319			       l->block_ref_to->dev_state->name,
2320			       l->block_ref_to->dev_bytenr,
2321			       l->block_ref_to->mirror_num);
2322			ret = -1;
2323		} else if (l->parent_generation !=
2324			   l->block_ref_to->generation &&
2325			   BTRFSIC_GENERATION_UNKNOWN !=
2326			   l->parent_generation &&
2327			   BTRFSIC_GENERATION_UNKNOWN !=
2328			   l->block_ref_to->generation) {
2329			pr_info("btrfs: attempt to write superblock which references block %c @%llu (%s/%llu/%d) with generation %llu != parent generation %llu!\n",
 
2330			       btrfsic_get_block_type(state, l->block_ref_to),
2331			       l->block_ref_to->logical_bytenr,
2332			       l->block_ref_to->dev_state->name,
2333			       l->block_ref_to->dev_bytenr,
2334			       l->block_ref_to->mirror_num,
2335			       l->block_ref_to->generation,
2336			       l->parent_generation);
2337			ret = -1;
2338		} else if (l->block_ref_to->flush_gen >
2339			   l->block_ref_to->dev_state->last_flush_gen) {
2340			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",
 
2341			       btrfsic_get_block_type(state, l->block_ref_to),
2342			       l->block_ref_to->logical_bytenr,
2343			       l->block_ref_to->dev_state->name,
2344			       l->block_ref_to->dev_bytenr,
2345			       l->block_ref_to->mirror_num, block->flush_gen,
2346			       l->block_ref_to->dev_state->last_flush_gen);
2347			ret = -1;
2348		} else if (-1 == btrfsic_check_all_ref_blocks(state,
2349							      l->block_ref_to,
2350							      recursion_level +
2351							      1)) {
2352			ret = -1;
2353		}
2354	}
2355
2356	return ret;
2357}
2358
2359static int btrfsic_is_block_ref_by_superblock(
2360		const struct btrfsic_state *state,
2361		const struct btrfsic_block *block,
2362		int recursion_level)
2363{
2364	const struct btrfsic_block_link *l;
2365
2366	if (recursion_level >= 3 + BTRFS_MAX_LEVEL) {
2367		/* refer to comment at "abort cyclic linkage (case 1)" */
2368		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2369			pr_info("btrfsic: abort cyclic linkage (case 2).\n");
2370
2371		return 0;
2372	}
2373
2374	/*
2375	 * This algorithm is recursive because the amount of used stack space
2376	 * is very small and the max recursion depth is limited.
2377	 */
2378	list_for_each_entry(l, &block->ref_from_list, node_ref_from) {
2379		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2380			pr_info("rl=%d, %c @%llu (%s/%llu/%d) is ref %u* from %c @%llu (%s/%llu/%d)\n",
 
2381			       recursion_level,
2382			       btrfsic_get_block_type(state, block),
2383			       block->logical_bytenr, block->dev_state->name,
2384			       block->dev_bytenr, block->mirror_num,
2385			       l->ref_cnt,
2386			       btrfsic_get_block_type(state, l->block_ref_from),
2387			       l->block_ref_from->logical_bytenr,
2388			       l->block_ref_from->dev_state->name,
2389			       l->block_ref_from->dev_bytenr,
2390			       l->block_ref_from->mirror_num);
2391		if (l->block_ref_from->is_superblock &&
2392		    state->latest_superblock->dev_bytenr ==
2393		    l->block_ref_from->dev_bytenr &&
2394		    state->latest_superblock->dev_state->bdev ==
2395		    l->block_ref_from->dev_state->bdev)
2396			return 1;
2397		else if (btrfsic_is_block_ref_by_superblock(state,
2398							    l->block_ref_from,
2399							    recursion_level +
2400							    1))
2401			return 1;
2402	}
2403
2404	return 0;
2405}
2406
2407static void btrfsic_print_add_link(const struct btrfsic_state *state,
2408				   const struct btrfsic_block_link *l)
2409{
2410	pr_info("Add %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2411	       l->ref_cnt,
2412	       btrfsic_get_block_type(state, l->block_ref_from),
2413	       l->block_ref_from->logical_bytenr,
2414	       l->block_ref_from->dev_state->name,
2415	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2416	       btrfsic_get_block_type(state, l->block_ref_to),
2417	       l->block_ref_to->logical_bytenr,
2418	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2419	       l->block_ref_to->mirror_num);
2420}
2421
2422static void btrfsic_print_rem_link(const struct btrfsic_state *state,
2423				   const struct btrfsic_block_link *l)
2424{
2425	pr_info("Rem %u* link from %c @%llu (%s/%llu/%d) to %c @%llu (%s/%llu/%d).\n",
2426	       l->ref_cnt,
2427	       btrfsic_get_block_type(state, l->block_ref_from),
2428	       l->block_ref_from->logical_bytenr,
2429	       l->block_ref_from->dev_state->name,
2430	       l->block_ref_from->dev_bytenr, l->block_ref_from->mirror_num,
2431	       btrfsic_get_block_type(state, l->block_ref_to),
2432	       l->block_ref_to->logical_bytenr,
2433	       l->block_ref_to->dev_state->name, l->block_ref_to->dev_bytenr,
2434	       l->block_ref_to->mirror_num);
2435}
2436
2437static char btrfsic_get_block_type(const struct btrfsic_state *state,
2438				   const struct btrfsic_block *block)
2439{
2440	if (block->is_superblock &&
2441	    state->latest_superblock->dev_bytenr == block->dev_bytenr &&
2442	    state->latest_superblock->dev_state->bdev == block->dev_state->bdev)
2443		return 'S';
2444	else if (block->is_superblock)
2445		return 's';
2446	else if (block->is_metadata)
2447		return 'M';
2448	else
2449		return 'D';
2450}
2451
2452static void btrfsic_dump_tree(const struct btrfsic_state *state)
2453{
2454	btrfsic_dump_tree_sub(state, state->latest_superblock, 0);
2455}
2456
2457static void btrfsic_dump_tree_sub(const struct btrfsic_state *state,
2458				  const struct btrfsic_block *block,
2459				  int indent_level)
2460{
2461	const struct btrfsic_block_link *l;
2462	int indent_add;
2463	static char buf[80];
2464	int cursor_position;
2465
2466	/*
2467	 * Should better fill an on-stack buffer with a complete line and
2468	 * dump it at once when it is time to print a newline character.
2469	 */
2470
2471	/*
2472	 * This algorithm is recursive because the amount of used stack space
2473	 * is very small and the max recursion depth is limited.
2474	 */
2475	indent_add = sprintf(buf, "%c-%llu(%s/%llu/%u)",
2476			     btrfsic_get_block_type(state, block),
2477			     block->logical_bytenr, block->dev_state->name,
2478			     block->dev_bytenr, block->mirror_num);
2479	if (indent_level + indent_add > BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2480		printk("[...]\n");
2481		return;
2482	}
2483	printk(buf);
2484	indent_level += indent_add;
2485	if (list_empty(&block->ref_to_list)) {
2486		printk("\n");
2487		return;
2488	}
2489	if (block->mirror_num > 1 &&
2490	    !(state->print_mask & BTRFSIC_PRINT_MASK_TREE_WITH_ALL_MIRRORS)) {
2491		printk(" [...]\n");
2492		return;
2493	}
2494
2495	cursor_position = indent_level;
2496	list_for_each_entry(l, &block->ref_to_list, node_ref_to) {
2497		while (cursor_position < indent_level) {
2498			printk(" ");
2499			cursor_position++;
2500		}
2501		if (l->ref_cnt > 1)
2502			indent_add = sprintf(buf, " %d*--> ", l->ref_cnt);
2503		else
2504			indent_add = sprintf(buf, " --> ");
2505		if (indent_level + indent_add >
2506		    BTRFSIC_TREE_DUMP_MAX_INDENT_LEVEL) {
2507			printk("[...]\n");
2508			cursor_position = 0;
2509			continue;
2510		}
2511
2512		printk(buf);
2513
2514		btrfsic_dump_tree_sub(state, l->block_ref_to,
2515				      indent_level + indent_add);
2516		cursor_position = 0;
2517	}
2518}
2519
2520static struct btrfsic_block_link *btrfsic_block_link_lookup_or_add(
2521		struct btrfsic_state *state,
2522		struct btrfsic_block_data_ctx *next_block_ctx,
2523		struct btrfsic_block *next_block,
2524		struct btrfsic_block *from_block,
2525		u64 parent_generation)
2526{
2527	struct btrfsic_block_link *l;
2528
2529	l = btrfsic_block_link_hashtable_lookup(next_block_ctx->dev->bdev,
2530						next_block_ctx->dev_bytenr,
2531						from_block->dev_state->bdev,
2532						from_block->dev_bytenr,
2533						&state->block_link_hashtable);
2534	if (NULL == l) {
2535		l = btrfsic_block_link_alloc();
2536		if (!l)
2537			return NULL;
2538
2539		l->block_ref_to = next_block;
2540		l->block_ref_from = from_block;
2541		l->ref_cnt = 1;
2542		l->parent_generation = parent_generation;
2543
2544		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2545			btrfsic_print_add_link(state, l);
2546
2547		list_add(&l->node_ref_to, &from_block->ref_to_list);
2548		list_add(&l->node_ref_from, &next_block->ref_from_list);
2549
2550		btrfsic_block_link_hashtable_add(l,
2551						 &state->block_link_hashtable);
2552	} else {
2553		l->ref_cnt++;
2554		l->parent_generation = parent_generation;
2555		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2556			btrfsic_print_add_link(state, l);
2557	}
2558
2559	return l;
2560}
2561
2562static struct btrfsic_block *btrfsic_block_lookup_or_add(
2563		struct btrfsic_state *state,
2564		struct btrfsic_block_data_ctx *block_ctx,
2565		const char *additional_string,
2566		int is_metadata,
2567		int is_iodone,
2568		int never_written,
2569		int mirror_num,
2570		int *was_created)
2571{
2572	struct btrfsic_block *block;
2573
2574	block = btrfsic_block_hashtable_lookup(block_ctx->dev->bdev,
2575					       block_ctx->dev_bytenr,
2576					       &state->block_hashtable);
2577	if (NULL == block) {
2578		struct btrfsic_dev_state *dev_state;
2579
2580		block = btrfsic_block_alloc();
2581		if (!block)
2582			return NULL;
2583
2584		dev_state = btrfsic_dev_state_lookup(block_ctx->dev->bdev->bd_dev);
2585		if (NULL == dev_state) {
2586			pr_info("btrfsic: error, lookup dev_state failed!\n");
2587			btrfsic_block_free(block);
2588			return NULL;
2589		}
2590		block->dev_state = dev_state;
2591		block->dev_bytenr = block_ctx->dev_bytenr;
2592		block->logical_bytenr = block_ctx->start;
2593		block->is_metadata = is_metadata;
2594		block->is_iodone = is_iodone;
2595		block->never_written = never_written;
2596		block->mirror_num = mirror_num;
2597		if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2598			pr_info("New %s%c-block @%llu (%s/%llu/%d)\n",
2599			       additional_string,
2600			       btrfsic_get_block_type(state, block),
2601			       block->logical_bytenr, dev_state->name,
2602			       block->dev_bytenr, mirror_num);
2603		list_add(&block->all_blocks_node, &state->all_blocks_list);
2604		btrfsic_block_hashtable_add(block, &state->block_hashtable);
2605		if (NULL != was_created)
2606			*was_created = 1;
2607	} else {
2608		if (NULL != was_created)
2609			*was_created = 0;
2610	}
2611
2612	return block;
2613}
2614
2615static void btrfsic_cmp_log_and_dev_bytenr(struct btrfsic_state *state,
2616					   u64 bytenr,
2617					   struct btrfsic_dev_state *dev_state,
2618					   u64 dev_bytenr)
2619{
2620	struct btrfs_fs_info *fs_info = state->fs_info;
2621	struct btrfsic_block_data_ctx block_ctx;
2622	int num_copies;
2623	int mirror_num;
2624	int match = 0;
2625	int ret;
2626
2627	num_copies = btrfs_num_copies(fs_info, bytenr, state->metablock_size);
2628
2629	for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2630		ret = btrfsic_map_block(state, bytenr, state->metablock_size,
2631					&block_ctx, mirror_num);
2632		if (ret) {
2633			pr_info("btrfsic: btrfsic_map_block(logical @%llu, mirror %d) failed!\n",
2634			       bytenr, mirror_num);
2635			continue;
2636		}
2637
2638		if (dev_state->bdev == block_ctx.dev->bdev &&
2639		    dev_bytenr == block_ctx.dev_bytenr) {
2640			match++;
2641			btrfsic_release_block_ctx(&block_ctx);
2642			break;
2643		}
2644		btrfsic_release_block_ctx(&block_ctx);
2645	}
2646
2647	if (WARN_ON(!match)) {
2648		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",
2649		       bytenr, dev_state->name, dev_bytenr);
 
2650		for (mirror_num = 1; mirror_num <= num_copies; mirror_num++) {
2651			ret = btrfsic_map_block(state, bytenr,
2652						state->metablock_size,
2653						&block_ctx, mirror_num);
2654			if (ret)
2655				continue;
2656
2657			pr_info("Read logical bytenr @%llu maps to (%s/%llu/%d)\n",
2658			       bytenr, block_ctx.dev->name,
2659			       block_ctx.dev_bytenr, mirror_num);
2660		}
2661	}
2662}
2663
2664static struct btrfsic_dev_state *btrfsic_dev_state_lookup(dev_t dev)
2665{
2666	return btrfsic_dev_state_hashtable_lookup(dev,
2667						  &btrfsic_dev_state_hashtable);
2668}
2669
2670static void __btrfsic_submit_bio(struct bio *bio)
2671{
2672	struct btrfsic_dev_state *dev_state;
 
 
 
 
 
 
 
 
 
 
 
 
 
2673
2674	if (!btrfsic_is_initialized)
 
2675		return;
2676
2677	mutex_lock(&btrfsic_mutex);
2678	/* since btrfsic_submit_bio() is also called before
2679	 * btrfsic_mount(), this might return NULL */
2680	dev_state = btrfsic_dev_state_lookup(bio_dev(bio) + bio->bi_partno);
2681	if (NULL != dev_state &&
2682	    (bio_op(bio) == REQ_OP_WRITE) && bio_has_data(bio)) {
2683		unsigned int i = 0;
2684		u64 dev_bytenr;
2685		u64 cur_bytenr;
2686		struct bio_vec bvec;
2687		struct bvec_iter iter;
2688		int bio_is_patched;
2689		char **mapped_datav;
2690		unsigned int segs = bio_segments(bio);
2691
2692		dev_bytenr = 512 * bio->bi_iter.bi_sector;
2693		bio_is_patched = 0;
2694		if (dev_state->state->print_mask &
2695		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2696			pr_info("submit_bio(rw=%d,0x%x, bi_vcnt=%u, bi_sector=%llu (bytenr %llu), bi_disk=%p)\n",
2697			       bio_op(bio), bio->bi_opf, segs,
2698			       (unsigned long long)bio->bi_iter.bi_sector,
2699			       dev_bytenr, bio->bi_disk);
2700
2701		mapped_datav = kmalloc_array(segs,
2702					     sizeof(*mapped_datav), GFP_NOFS);
2703		if (!mapped_datav)
2704			goto leave;
2705		cur_bytenr = dev_bytenr;
2706
2707		bio_for_each_segment(bvec, bio, iter) {
2708			BUG_ON(bvec.bv_len != PAGE_SIZE);
2709			mapped_datav[i] = kmap(bvec.bv_page);
2710			i++;
2711
2712			if (dev_state->state->print_mask &
2713			    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH_VERBOSE)
2714				pr_info("#%u: bytenr=%llu, len=%u, offset=%u\n",
2715				       i, cur_bytenr, bvec.bv_len, bvec.bv_offset);
2716			cur_bytenr += bvec.bv_len;
2717		}
2718		btrfsic_process_written_block(dev_state, dev_bytenr,
2719					      mapped_datav, segs,
2720					      bio, &bio_is_patched,
2721					      bio->bi_opf);
2722		bio_for_each_segment(bvec, bio, iter)
2723			kunmap(bvec.bv_page);
2724		kfree(mapped_datav);
2725	} else if (NULL != dev_state && (bio->bi_opf & REQ_PREFLUSH)) {
2726		if (dev_state->state->print_mask &
2727		    BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH)
2728			pr_info("submit_bio(rw=%d,0x%x FLUSH, disk=%p)\n",
2729			       bio_op(bio), bio->bi_opf, bio->bi_disk);
2730		if (!dev_state->dummy_block_for_bio_bh_flush.is_iodone) {
2731			if ((dev_state->state->print_mask &
2732			     (BTRFSIC_PRINT_MASK_SUBMIT_BIO_BH |
2733			      BTRFSIC_PRINT_MASK_VERBOSE)))
2734				pr_info("btrfsic_submit_bio(%s) with FLUSH but dummy block already in use (ignored)!\n",
2735				       dev_state->name);
2736		} else {
2737			struct btrfsic_block *const block =
2738				&dev_state->dummy_block_for_bio_bh_flush;
2739
2740			block->is_iodone = 0;
2741			block->never_written = 0;
2742			block->iodone_w_error = 0;
2743			block->flush_gen = dev_state->last_flush_gen + 1;
2744			block->submit_bio_bh_rw = bio->bi_opf;
2745			block->orig_bio_private = bio->bi_private;
2746			block->orig_bio_end_io = bio->bi_end_io;
2747			block->next_in_same_bio = NULL;
2748			bio->bi_private = block;
2749			bio->bi_end_io = btrfsic_bio_end_io;
2750		}
2751	}
2752leave:
2753	mutex_unlock(&btrfsic_mutex);
2754}
2755
2756void btrfsic_submit_bio(struct bio *bio)
2757{
2758	__btrfsic_submit_bio(bio);
2759	submit_bio(bio);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2760}
2761
2762int btrfsic_submit_bio_wait(struct bio *bio)
2763{
2764	__btrfsic_submit_bio(bio);
2765	return submit_bio_wait(bio);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2766}
2767
2768int btrfsic_mount(struct btrfs_fs_info *fs_info,
2769		  struct btrfs_fs_devices *fs_devices,
2770		  int including_extent_data, u32 print_mask)
2771{
2772	int ret;
2773	struct btrfsic_state *state;
2774	struct list_head *dev_head = &fs_devices->devices;
2775	struct btrfs_device *device;
2776
2777	if (!PAGE_ALIGNED(fs_info->nodesize)) {
2778		pr_info("btrfsic: cannot handle nodesize %d not being a multiple of PAGE_SIZE %ld!\n",
2779		       fs_info->nodesize, PAGE_SIZE);
2780		return -1;
2781	}
2782	if (!PAGE_ALIGNED(fs_info->sectorsize)) {
2783		pr_info("btrfsic: cannot handle sectorsize %d not being a multiple of PAGE_SIZE %ld!\n",
2784		       fs_info->sectorsize, PAGE_SIZE);
2785		return -1;
2786	}
2787	state = kvzalloc(sizeof(*state), GFP_KERNEL);
2788	if (!state)
2789		return -ENOMEM;
2790
2791	if (!btrfsic_is_initialized) {
2792		mutex_init(&btrfsic_mutex);
2793		btrfsic_dev_state_hashtable_init(&btrfsic_dev_state_hashtable);
2794		btrfsic_is_initialized = 1;
2795	}
2796	mutex_lock(&btrfsic_mutex);
2797	state->fs_info = fs_info;
2798	state->print_mask = print_mask;
2799	state->include_extent_data = including_extent_data;
2800	state->csum_size = 0;
2801	state->metablock_size = fs_info->nodesize;
2802	state->datablock_size = fs_info->sectorsize;
2803	INIT_LIST_HEAD(&state->all_blocks_list);
2804	btrfsic_block_hashtable_init(&state->block_hashtable);
2805	btrfsic_block_link_hashtable_init(&state->block_link_hashtable);
2806	state->max_superblock_generation = 0;
2807	state->latest_superblock = NULL;
2808
2809	list_for_each_entry(device, dev_head, dev_list) {
2810		struct btrfsic_dev_state *ds;
2811		const char *p;
2812
2813		if (!device->bdev || !device->name)
2814			continue;
2815
2816		ds = btrfsic_dev_state_alloc();
2817		if (NULL == ds) {
2818			mutex_unlock(&btrfsic_mutex);
2819			return -ENOMEM;
2820		}
2821		ds->bdev = device->bdev;
2822		ds->state = state;
2823		bdevname(ds->bdev, ds->name);
2824		ds->name[BDEVNAME_SIZE - 1] = '\0';
2825		p = kbasename(ds->name);
2826		strlcpy(ds->name, p, sizeof(ds->name));
2827		btrfsic_dev_state_hashtable_add(ds,
2828						&btrfsic_dev_state_hashtable);
2829	}
2830
2831	ret = btrfsic_process_superblock(state, fs_devices);
2832	if (0 != ret) {
2833		mutex_unlock(&btrfsic_mutex);
2834		btrfsic_unmount(fs_devices);
2835		return ret;
2836	}
2837
2838	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_DATABASE)
2839		btrfsic_dump_database(state);
2840	if (state->print_mask & BTRFSIC_PRINT_MASK_INITIAL_TREE)
2841		btrfsic_dump_tree(state);
2842
2843	mutex_unlock(&btrfsic_mutex);
2844	return 0;
2845}
2846
2847void btrfsic_unmount(struct btrfs_fs_devices *fs_devices)
2848{
2849	struct btrfsic_block *b_all, *tmp_all;
2850	struct btrfsic_state *state;
2851	struct list_head *dev_head = &fs_devices->devices;
2852	struct btrfs_device *device;
2853
2854	if (!btrfsic_is_initialized)
2855		return;
2856
2857	mutex_lock(&btrfsic_mutex);
2858
2859	state = NULL;
2860	list_for_each_entry(device, dev_head, dev_list) {
2861		struct btrfsic_dev_state *ds;
2862
2863		if (!device->bdev || !device->name)
2864			continue;
2865
2866		ds = btrfsic_dev_state_hashtable_lookup(
2867				device->bdev->bd_dev,
2868				&btrfsic_dev_state_hashtable);
2869		if (NULL != ds) {
2870			state = ds->state;
2871			btrfsic_dev_state_hashtable_remove(ds);
2872			btrfsic_dev_state_free(ds);
2873		}
2874	}
2875
2876	if (NULL == state) {
2877		pr_info("btrfsic: error, cannot find state information on umount!\n");
2878		mutex_unlock(&btrfsic_mutex);
2879		return;
2880	}
2881
2882	/*
2883	 * Don't care about keeping the lists' state up to date,
2884	 * just free all memory that was allocated dynamically.
2885	 * Free the blocks and the block_links.
2886	 */
2887	list_for_each_entry_safe(b_all, tmp_all, &state->all_blocks_list,
2888				 all_blocks_node) {
2889		struct btrfsic_block_link *l, *tmp;
2890
2891		list_for_each_entry_safe(l, tmp, &b_all->ref_to_list,
2892					 node_ref_to) {
2893			if (state->print_mask & BTRFSIC_PRINT_MASK_VERBOSE)
2894				btrfsic_print_rem_link(state, l);
2895
2896			l->ref_cnt--;
2897			if (0 == l->ref_cnt)
2898				btrfsic_block_link_free(l);
2899		}
2900
2901		if (b_all->is_iodone || b_all->never_written)
2902			btrfsic_block_free(b_all);
2903		else
2904			pr_info("btrfs: attempt to free %c-block @%llu (%s/%llu/%d) on umount which is not yet iodone!\n",
 
2905			       btrfsic_get_block_type(state, b_all),
2906			       b_all->logical_bytenr, b_all->dev_state->name,
2907			       b_all->dev_bytenr, b_all->mirror_num);
2908	}
2909
2910	mutex_unlock(&btrfsic_mutex);
2911
2912	kvfree(state);
2913}