drivers/md/dm-integrity.c

   1 /*
   2  * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
   3  * Copyright (C) 2016-2017 Milan Broz
   4  * Copyright (C) 2016-2017 Mikulas Patocka
   5  *
   6  * This file is released under the GPL.
   7  */
   8
   9 #include "dm-bio-record.h"
  10
  11 #include <linux/compiler.h>
  12 #include <linux/module.h>
  13 #include <linux/device-mapper.h>
  14 #include <linux/dm-io.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/sort.h>
  17 #include <linux/rbtree.h>
  18 #include <linux/delay.h>
  19 #include <linux/random.h>
  20 #include <linux/reboot.h>
  21 #include <crypto/hash.h>
  22 #include <crypto/skcipher.h>
  23 #include <linux/async_tx.h>
  24 #include <linux/dm-bufio.h>
  25
  26 #define DM_MSG_PREFIX "integrity"
  27
  28 #define DEFAULT_INTERLEAVE_SECTORS      32768
  29 #define DEFAULT_JOURNAL_SIZE_FACTOR     7
  30 #define DEFAULT_SECTORS_PER_BITMAP_BIT  32768
  31 #define DEFAULT_BUFFER_SECTORS          128
  32 #define DEFAULT_JOURNAL_WATERMARK       50
  33 #define DEFAULT_SYNC_MSEC               10000
  34 #define DEFAULT_MAX_JOURNAL_SECTORS     131072
  35 #define MIN_LOG2_INTERLEAVE_SECTORS     3
  36 #define MAX_LOG2_INTERLEAVE_SECTORS     31
  37 #define METADATA_WORKQUEUE_MAX_ACTIVE   16
  38 #define RECALC_SECTORS                  8192
  39 #define RECALC_WRITE_SUPER              16
  40 #define BITMAP_BLOCK_SIZE               4096    /* don't change it */
  41 #define BITMAP_FLUSH_INTERVAL           (10 * HZ)
  42 #define DISCARD_FILLER                  0xf6
  43
  44 /*
  45  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  46  * so it should not be enabled in the official kernel
  47  */
  48 //#define DEBUG_PRINT
  49 //#define INTERNAL_VERIFY
  50
  51 /*
  52  * On disk structures
  53  */
  54
  55 #define SB_MAGIC                        "integrt"
  56 #define SB_VERSION_1                    1
  57 #define SB_VERSION_2                    2
  58 #define SB_VERSION_3                    3
  59 #define SB_VERSION_4                    4
  60 #define SB_SECTORS                      8
  61 #define MAX_SECTORS_PER_BLOCK           8
  62
  63 struct superblock {
  64         __u8 magic[8];
  65         __u8 version;
  66         __u8 log2_interleave_sectors;
  67         __u16 integrity_tag_size;
  68         __u32 journal_sections;
  69         __u64 provided_data_sectors;    /* userspace uses this value */
  70         __u32 flags;
  71         __u8 log2_sectors_per_block;
  72         __u8 log2_blocks_per_bitmap_bit;
  73         __u8 pad[2];
  74         __u64 recalc_sector;
  75 };
  76
  77 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  78 #define SB_FLAG_RECALCULATING           0x2
  79 #define SB_FLAG_DIRTY_BITMAP            0x4
  80 #define SB_FLAG_FIXED_PADDING           0x8
  81
  82 #define JOURNAL_ENTRY_ROUNDUP           8
  83
  84 typedef __u64 commit_id_t;
  85 #define JOURNAL_MAC_PER_SECTOR          8
  86
  87 struct journal_entry {
  88         union {
  89                 struct {
  90                         __u32 sector_lo;
  91                         __u32 sector_hi;
  92                 } s;
  93                 __u64 sector;
  94         } u;
  95         commit_id_t last_bytes[];
  96         /* __u8 tag[0]; */
  97 };
  98
  99 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
 100
 101 #if BITS_PER_LONG == 64
 102 #define journal_entry_set_sector(je, x)         do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
 103 #else
 104 #define journal_entry_set_sector(je, x)         do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32((x) >> 32)); } while (0)
 105 #endif
 106 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
 107 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
 108 #define journal_entry_set_unused(je)            do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
 109 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
 110 #define journal_entry_set_inprogress(je)        do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
 111
 112 #define JOURNAL_BLOCK_SECTORS           8
 113 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 114 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 115
 116 struct journal_sector {
 117         __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 118         __u8 mac[JOURNAL_MAC_PER_SECTOR];
 119         commit_id_t commit_id;
 120 };
 121
 122 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 123
 124 #define METADATA_PADDING_SECTORS        8
 125
 126 #define N_COMMIT_IDS                    4
 127
 128 static unsigned char prev_commit_seq(unsigned char seq)
 129 {
 130         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 131 }
 132
 133 static unsigned char next_commit_seq(unsigned char seq)
 134 {
 135         return (seq + 1) % N_COMMIT_IDS;
 136 }
 137
 138 /*
 139  * In-memory structures
 140  */
 141
 142 struct journal_node {
 143         struct rb_node node;
 144         sector_t sector;
 145 };
 146
 147 struct alg_spec {
 148         char *alg_string;
 149         char *key_string;
 150         __u8 *key;
 151         unsigned key_size;
 152 };
 153
 154 struct dm_integrity_c {
 155         struct dm_dev *dev;
 156         struct dm_dev *meta_dev;
 157         unsigned tag_size;
 158         __s8 log2_tag_size;
 159         sector_t start;
 160         mempool_t journal_io_mempool;
 161         struct dm_io_client *io;
 162         struct dm_bufio_client *bufio;
 163         struct workqueue_struct *metadata_wq;
 164         struct superblock *sb;
 165         unsigned journal_pages;
 166         unsigned n_bitmap_blocks;
 167
 168         struct page_list *journal;
 169         struct page_list *journal_io;
 170         struct page_list *journal_xor;
 171         struct page_list *recalc_bitmap;
 172         struct page_list *may_write_bitmap;
 173         struct bitmap_block_status *bbs;
 174         unsigned bitmap_flush_interval;
 175         int synchronous_mode;
 176         struct bio_list synchronous_bios;
 177         struct delayed_work bitmap_flush_work;
 178
 179         struct crypto_skcipher *journal_crypt;
 180         struct scatterlist **journal_scatterlist;
 181         struct scatterlist **journal_io_scatterlist;
 182         struct skcipher_request **sk_requests;
 183
 184         struct crypto_shash *journal_mac;
 185
 186         struct journal_node *journal_tree;
 187         struct rb_root journal_tree_root;
 188
 189         sector_t provided_data_sectors;
 190
 191         unsigned short journal_entry_size;
 192         unsigned char journal_entries_per_sector;
 193         unsigned char journal_section_entries;
 194         unsigned short journal_section_sectors;
 195         unsigned journal_sections;
 196         unsigned journal_entries;
 197         sector_t data_device_sectors;
 198         sector_t meta_device_sectors;
 199         unsigned initial_sectors;
 200         unsigned metadata_run;
 201         __s8 log2_metadata_run;
 202         __u8 log2_buffer_sectors;
 203         __u8 sectors_per_block;
 204         __u8 log2_blocks_per_bitmap_bit;
 205
 206         unsigned char mode;
 207
 208         int failed;
 209
 210         struct crypto_shash *internal_hash;
 211
 212         struct dm_target *ti;
 213
 214         /* these variables are locked with endio_wait.lock */
 215         struct rb_root in_progress;
 216         struct list_head wait_list;
 217         wait_queue_head_t endio_wait;
 218         struct workqueue_struct *wait_wq;
 219         struct workqueue_struct *offload_wq;
 220
 221         unsigned char commit_seq;
 222         commit_id_t commit_ids[N_COMMIT_IDS];
 223
 224         unsigned committed_section;
 225         unsigned n_committed_sections;
 226
 227         unsigned uncommitted_section;
 228         unsigned n_uncommitted_sections;
 229
 230         unsigned free_section;
 231         unsigned char free_section_entry;
 232         unsigned free_sectors;
 233
 234         unsigned free_sectors_threshold;
 235
 236         struct workqueue_struct *commit_wq;
 237         struct work_struct commit_work;
 238
 239         struct workqueue_struct *writer_wq;
 240         struct work_struct writer_work;
 241
 242         struct workqueue_struct *recalc_wq;
 243         struct work_struct recalc_work;
 244         u8 *recalc_buffer;
 245         u8 *recalc_tags;
 246
 247         struct bio_list flush_bio_list;
 248
 249         unsigned long autocommit_jiffies;
 250         struct timer_list autocommit_timer;
 251         unsigned autocommit_msec;
 252
 253         wait_queue_head_t copy_to_journal_wait;
 254
 255         struct completion crypto_backoff;
 256
 257         bool journal_uptodate;
 258         bool just_formatted;
 259         bool recalculate_flag;
 260         bool discard;
 261         bool fix_padding;
 262         bool legacy_recalculate;
 263
 264         struct alg_spec internal_hash_alg;
 265         struct alg_spec journal_crypt_alg;
 266         struct alg_spec journal_mac_alg;
 267
 268         atomic64_t number_of_mismatches;
 269
 270         struct notifier_block reboot_notifier;
 271 };
 272
 273 struct dm_integrity_range {
 274         sector_t logical_sector;
 275         sector_t n_sectors;
 276         bool waiting;
 277         union {
 278                 struct rb_node node;
 279                 struct {
 280                         struct task_struct *task;
 281                         struct list_head wait_entry;
 282                 };
 283         };
 284 };
 285
 286 struct dm_integrity_io {
 287         struct work_struct work;
 288
 289         struct dm_integrity_c *ic;
 290         enum req_opf op;
 291         bool fua;
 292
 293         struct dm_integrity_range range;
 294
 295         sector_t metadata_block;
 296         unsigned metadata_offset;
 297
 298         atomic_t in_flight;
 299         blk_status_t bi_status;
 300
 301         struct completion *completion;
 302
 303         struct dm_bio_details bio_details;
 304 };
 305
 306 struct journal_completion {
 307         struct dm_integrity_c *ic;
 308         atomic_t in_flight;
 309         struct completion comp;
 310 };
 311
 312 struct journal_io {
 313         struct dm_integrity_range range;
 314         struct journal_completion *comp;
 315 };
 316
 317 struct bitmap_block_status {
 318         struct work_struct work;
 319         struct dm_integrity_c *ic;
 320         unsigned idx;
 321         unsigned long *bitmap;
 322         struct bio_list bio_queue;
 323         spinlock_t bio_queue_lock;
 324
 325 };
 326
 327 static struct kmem_cache *journal_io_cache;
 328
 329 #define JOURNAL_IO_MEMPOOL      32
 330
 331 #ifdef DEBUG_PRINT
 332 #define DEBUG_print(x, ...)     printk(KERN_DEBUG x, ##__VA_ARGS__)
 333 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 334 {
 335         va_list args;
 336         va_start(args, msg);
 337         vprintk(msg, args);
 338         va_end(args);
 339         if (len)
 340                 pr_cont(":");
 341         while (len) {
 342                 pr_cont(" %02x", *bytes);
 343                 bytes++;
 344                 len--;
 345         }
 346         pr_cont("\n");
 347 }
 348 #define DEBUG_bytes(bytes, len, msg, ...)       __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 349 #else
 350 #define DEBUG_print(x, ...)                     do { } while (0)
 351 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
 352 #endif
 353
 354 static void dm_integrity_prepare(struct request *rq)
 355 {
 356 }
 357
 358 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
 359 {
 360 }
 361
 362 /*
 363  * DM Integrity profile, protection is performed layer above (dm-crypt)
 364  */
 365 static const struct blk_integrity_profile dm_integrity_profile = {
 366         .name                   = "DM-DIF-EXT-TAG",
 367         .generate_fn            = NULL,
 368         .verify_fn              = NULL,
 369         .prepare_fn             = dm_integrity_prepare,
 370         .complete_fn            = dm_integrity_complete,
 371 };
 372
 373 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 374 static void integrity_bio_wait(struct work_struct *w);
 375 static void dm_integrity_dtr(struct dm_target *ti);
 376
 377 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 378 {
 379         if (err == -EILSEQ)
 380                 atomic64_inc(&ic->number_of_mismatches);
 381         if (!cmpxchg(&ic->failed, 0, err))
 382                 DMERR("Error on %s: %d", msg, err);
 383 }
 384
 385 static int dm_integrity_failed(struct dm_integrity_c *ic)
 386 {
 387         return READ_ONCE(ic->failed);
 388 }
 389
 390 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
 391 {
 392         if ((ic->internal_hash_alg.key || ic->journal_mac_alg.key) &&
 393             !ic->legacy_recalculate)
 394                 return true;
 395         return false;
 396 }
 397
 398 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 399                                           unsigned j, unsigned char seq)
 400 {
 401         /*
 402          * Xor the number with section and sector, so that if a piece of
 403          * journal is written at wrong place, it is detected.
 404          */
 405         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 406 }
 407
 408 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 409                                 sector_t *area, sector_t *offset)
 410 {
 411         if (!ic->meta_dev) {
 412                 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 413                 *area = data_sector >> log2_interleave_sectors;
 414                 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 415         } else {
 416                 *area = 0;
 417                 *offset = data_sector;
 418         }
 419 }
 420
 421 #define sector_to_block(ic, n)                                          \
 422 do {                                                                    \
 423         BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));          \
 424         (n) >>= (ic)->sb->log2_sectors_per_block;                       \
 425 } while (0)
 426
 427 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 428                                             sector_t offset, unsigned *metadata_offset)
 429 {
 430         __u64 ms;
 431         unsigned mo;
 432
 433         ms = area << ic->sb->log2_interleave_sectors;
 434         if (likely(ic->log2_metadata_run >= 0))
 435                 ms += area << ic->log2_metadata_run;
 436         else
 437                 ms += area * ic->metadata_run;
 438         ms >>= ic->log2_buffer_sectors;
 439
 440         sector_to_block(ic, offset);
 441
 442         if (likely(ic->log2_tag_size >= 0)) {
 443                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 444                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 445         } else {
 446                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 447                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 448         }
 449         *metadata_offset = mo;
 450         return ms;
 451 }
 452
 453 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 454 {
 455         sector_t result;
 456
 457         if (ic->meta_dev)
 458                 return offset;
 459
 460         result = area << ic->sb->log2_interleave_sectors;
 461         if (likely(ic->log2_metadata_run >= 0))
 462                 result += (area + 1) << ic->log2_metadata_run;
 463         else
 464                 result += (area + 1) * ic->metadata_run;
 465
 466         result += (sector_t)ic->initial_sectors + offset;
 467         result += ic->start;
 468
 469         return result;
 470 }
 471
 472 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 473 {
 474         if (unlikely(*sec_ptr >= ic->journal_sections))
 475                 *sec_ptr -= ic->journal_sections;
 476 }
 477
 478 static void sb_set_version(struct dm_integrity_c *ic)
 479 {
 480         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
 481                 ic->sb->version = SB_VERSION_4;
 482         else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
 483                 ic->sb->version = SB_VERSION_3;
 484         else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
 485                 ic->sb->version = SB_VERSION_2;
 486         else
 487                 ic->sb->version = SB_VERSION_1;
 488 }
 489
 490 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
 491 {
 492         struct dm_io_request io_req;
 493         struct dm_io_region io_loc;
 494
 495         io_req.bi_op = op;
 496         io_req.bi_op_flags = op_flags;
 497         io_req.mem.type = DM_IO_KMEM;
 498         io_req.mem.ptr.addr = ic->sb;
 499         io_req.notify.fn = NULL;
 500         io_req.client = ic->io;
 501         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 502         io_loc.sector = ic->start;
 503         io_loc.count = SB_SECTORS;
 504
 505         if (op == REQ_OP_WRITE)
 506                 sb_set_version(ic);
 507
 508         return dm_io(&io_req, 1, &io_loc, NULL);
 509 }
 510
 511 #define BITMAP_OP_TEST_ALL_SET          0
 512 #define BITMAP_OP_TEST_ALL_CLEAR        1
 513 #define BITMAP_OP_SET                   2
 514 #define BITMAP_OP_CLEAR                 3
 515
 516 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
 517                             sector_t sector, sector_t n_sectors, int mode)
 518 {
 519         unsigned long bit, end_bit, this_end_bit, page, end_page;
 520         unsigned long *data;
 521
 522         if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
 523                 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
 524                         sector,
 525                         n_sectors,
 526                         ic->sb->log2_sectors_per_block,
 527                         ic->log2_blocks_per_bitmap_bit,
 528                         mode);
 529                 BUG();
 530         }
 531
 532         if (unlikely(!n_sectors))
 533                 return true;
 534
 535         bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 536         end_bit = (sector + n_sectors - 1) >>
 537                 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 538
 539         page = bit / (PAGE_SIZE * 8);
 540         bit %= PAGE_SIZE * 8;
 541
 542         end_page = end_bit / (PAGE_SIZE * 8);
 543         end_bit %= PAGE_SIZE * 8;
 544
 545 repeat:
 546         if (page < end_page) {
 547                 this_end_bit = PAGE_SIZE * 8 - 1;
 548         } else {
 549                 this_end_bit = end_bit;
 550         }
 551
 552         data = lowmem_page_address(bitmap[page].page);
 553
 554         if (mode == BITMAP_OP_TEST_ALL_SET) {
 555                 while (bit <= this_end_bit) {
 556                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 557                                 do {
 558                                         if (data[bit / BITS_PER_LONG] != -1)
 559                                                 return false;
 560                                         bit += BITS_PER_LONG;
 561                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 562                                 continue;
 563                         }
 564                         if (!test_bit(bit, data))
 565                                 return false;
 566                         bit++;
 567                 }
 568         } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
 569                 while (bit <= this_end_bit) {
 570                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 571                                 do {
 572                                         if (data[bit / BITS_PER_LONG] != 0)
 573                                                 return false;
 574                                         bit += BITS_PER_LONG;
 575                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 576                                 continue;
 577                         }
 578                         if (test_bit(bit, data))
 579                                 return false;
 580                         bit++;
 581                 }
 582         } else if (mode == BITMAP_OP_SET) {
 583                 while (bit <= this_end_bit) {
 584                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 585                                 do {
 586                                         data[bit / BITS_PER_LONG] = -1;
 587                                         bit += BITS_PER_LONG;
 588                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 589                                 continue;
 590                         }
 591                         __set_bit(bit, data);
 592                         bit++;
 593                 }
 594         } else if (mode == BITMAP_OP_CLEAR) {
 595                 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
 596                         clear_page(data);
 597                 else while (bit <= this_end_bit) {
 598                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 599                                 do {
 600                                         data[bit / BITS_PER_LONG] = 0;
 601                                         bit += BITS_PER_LONG;
 602                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 603                                 continue;
 604                         }
 605                         __clear_bit(bit, data);
 606                         bit++;
 607                 }
 608         } else {
 609                 BUG();
 610         }
 611
 612         if (unlikely(page < end_page)) {
 613                 bit = 0;
 614                 page++;
 615                 goto repeat;
 616         }
 617
 618         return true;
 619 }
 620
 621 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
 622 {
 623         unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
 624         unsigned i;
 625
 626         for (i = 0; i < n_bitmap_pages; i++) {
 627                 unsigned long *dst_data = lowmem_page_address(dst[i].page);
 628                 unsigned long *src_data = lowmem_page_address(src[i].page);
 629                 copy_page(dst_data, src_data);
 630         }
 631 }
 632
 633 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
 634 {
 635         unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 636         unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
 637
 638         BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
 639         return &ic->bbs[bitmap_block];
 640 }
 641
 642 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 643                                  bool e, const char *function)
 644 {
 645 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 646         unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 647
 648         if (unlikely(section >= ic->journal_sections) ||
 649             unlikely(offset >= limit)) {
 650                 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
 651                        function, section, offset, ic->journal_sections, limit);
 652                 BUG();
 653         }
 654 #endif
 655 }
 656
 657 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 658                                unsigned *pl_index, unsigned *pl_offset)
 659 {
 660         unsigned sector;
 661
 662         access_journal_check(ic, section, offset, false, "page_list_location");
 663
 664         sector = section * ic->journal_section_sectors + offset;
 665
 666         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 667         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 668 }
 669
 670 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 671                                                unsigned section, unsigned offset, unsigned *n_sectors)
 672 {
 673         unsigned pl_index, pl_offset;
 674         char *va;
 675
 676         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 677
 678         if (n_sectors)
 679                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 680
 681         va = lowmem_page_address(pl[pl_index].page);
 682
 683         return (struct journal_sector *)(va + pl_offset);
 684 }
 685
 686 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 687 {
 688         return access_page_list(ic, ic->journal, section, offset, NULL);
 689 }
 690
 691 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 692 {
 693         unsigned rel_sector, offset;
 694         struct journal_sector *js;
 695
 696         access_journal_check(ic, section, n, true, "access_journal_entry");
 697
 698         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 699         offset = n / JOURNAL_BLOCK_SECTORS;
 700
 701         js = access_journal(ic, section, rel_sector);
 702         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 703 }
 704
 705 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 706 {
 707         n <<= ic->sb->log2_sectors_per_block;
 708
 709         n += JOURNAL_BLOCK_SECTORS;
 710
 711         access_journal_check(ic, section, n, false, "access_journal_data");
 712
 713         return access_journal(ic, section, n);
 714 }
 715
 716 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 717 {
 718         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 719         int r;
 720         unsigned j, size;
 721
 722         desc->tfm = ic->journal_mac;
 723
 724         r = crypto_shash_init(desc);
 725         if (unlikely(r)) {
 726                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 727                 goto err;
 728         }
 729
 730         for (j = 0; j < ic->journal_section_entries; j++) {
 731                 struct journal_entry *je = access_journal_entry(ic, section, j);
 732                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 733                 if (unlikely(r)) {
 734                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 735                         goto err;
 736                 }
 737         }
 738
 739         size = crypto_shash_digestsize(ic->journal_mac);
 740
 741         if (likely(size <= JOURNAL_MAC_SIZE)) {
 742                 r = crypto_shash_final(desc, result);
 743                 if (unlikely(r)) {
 744                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 745                         goto err;
 746                 }
 747                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 748         } else {
 749                 __u8 digest[HASH_MAX_DIGESTSIZE];
 750
 751                 if (WARN_ON(size > sizeof(digest))) {
 752                         dm_integrity_io_error(ic, "digest_size", -EINVAL);
 753                         goto err;
 754                 }
 755                 r = crypto_shash_final(desc, digest);
 756                 if (unlikely(r)) {
 757                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 758                         goto err;
 759                 }
 760                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 761         }
 762
 763         return;
 764 err:
 765         memset(result, 0, JOURNAL_MAC_SIZE);
 766 }
 767
 768 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 769 {
 770         __u8 result[JOURNAL_MAC_SIZE];
 771         unsigned j;
 772
 773         if (!ic->journal_mac)
 774                 return;
 775
 776         section_mac(ic, section, result);
 777
 778         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 779                 struct journal_sector *js = access_journal(ic, section, j);
 780
 781                 if (likely(wr))
 782                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 783                 else {
 784                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
 785                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 786                 }
 787         }
 788 }
 789
 790 static void complete_journal_op(void *context)
 791 {
 792         struct journal_completion *comp = context;
 793         BUG_ON(!atomic_read(&comp->in_flight));
 794         if (likely(atomic_dec_and_test(&comp->in_flight)))
 795                 complete(&comp->comp);
 796 }
 797
 798 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 799                         unsigned n_sections, struct journal_completion *comp)
 800 {
 801         struct async_submit_ctl submit;
 802         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 803         unsigned pl_index, pl_offset, section_index;
 804         struct page_list *source_pl, *target_pl;
 805
 806         if (likely(encrypt)) {
 807                 source_pl = ic->journal;
 808                 target_pl = ic->journal_io;
 809         } else {
 810                 source_pl = ic->journal_io;
 811                 target_pl = ic->journal;
 812         }
 813
 814         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 815
 816         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 817
 818         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 819
 820         section_index = pl_index;
 821
 822         do {
 823                 size_t this_step;
 824                 struct page *src_pages[2];
 825                 struct page *dst_page;
 826
 827                 while (unlikely(pl_index == section_index)) {
 828                         unsigned dummy;
 829                         if (likely(encrypt))
 830                                 rw_section_mac(ic, section, true);
 831                         section++;
 832                         n_sections--;
 833                         if (!n_sections)
 834                                 break;
 835                         page_list_location(ic, section, 0, &section_index, &dummy);
 836                 }
 837
 838                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 839                 dst_page = target_pl[pl_index].page;
 840                 src_pages[0] = source_pl[pl_index].page;
 841                 src_pages[1] = ic->journal_xor[pl_index].page;
 842
 843                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 844
 845                 pl_index++;
 846                 pl_offset = 0;
 847                 n_bytes -= this_step;
 848         } while (n_bytes);
 849
 850         BUG_ON(n_sections);
 851
 852         async_tx_issue_pending_all();
 853 }
 854
 855 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 856 {
 857         struct journal_completion *comp = req->data;
 858         if (unlikely(err)) {
 859                 if (likely(err == -EINPROGRESS)) {
 860                         complete(&comp->ic->crypto_backoff);
 861                         return;
 862                 }
 863                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 864         }
 865         complete_journal_op(comp);
 866 }
 867
 868 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 869 {
 870         int r;
 871         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 872                                       complete_journal_encrypt, comp);
 873         if (likely(encrypt))
 874                 r = crypto_skcipher_encrypt(req);
 875         else
 876                 r = crypto_skcipher_decrypt(req);
 877         if (likely(!r))
 878                 return false;
 879         if (likely(r == -EINPROGRESS))
 880                 return true;
 881         if (likely(r == -EBUSY)) {
 882                 wait_for_completion(&comp->ic->crypto_backoff);
 883                 reinit_completion(&comp->ic->crypto_backoff);
 884                 return true;
 885         }
 886         dm_integrity_io_error(comp->ic, "encrypt", r);
 887         return false;
 888 }
 889
 890 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 891                           unsigned n_sections, struct journal_completion *comp)
 892 {
 893         struct scatterlist **source_sg;
 894         struct scatterlist **target_sg;
 895
 896         atomic_add(2, &comp->in_flight);
 897
 898         if (likely(encrypt)) {
 899                 source_sg = ic->journal_scatterlist;
 900                 target_sg = ic->journal_io_scatterlist;
 901         } else {
 902                 source_sg = ic->journal_io_scatterlist;
 903                 target_sg = ic->journal_scatterlist;
 904         }
 905
 906         do {
 907                 struct skcipher_request *req;
 908                 unsigned ivsize;
 909                 char *iv;
 910
 911                 if (likely(encrypt))
 912                         rw_section_mac(ic, section, true);
 913
 914                 req = ic->sk_requests[section];
 915                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
 916                 iv = req->iv;
 917
 918                 memcpy(iv, iv + ivsize, ivsize);
 919
 920                 req->src = source_sg[section];
 921                 req->dst = target_sg[section];
 922
 923                 if (unlikely(do_crypt(encrypt, req, comp)))
 924                         atomic_inc(&comp->in_flight);
 925
 926                 section++;
 927                 n_sections--;
 928         } while (n_sections);
 929
 930         atomic_dec(&comp->in_flight);
 931         complete_journal_op(comp);
 932 }
 933
 934 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 935                             unsigned n_sections, struct journal_completion *comp)
 936 {
 937         if (ic->journal_xor)
 938                 return xor_journal(ic, encrypt, section, n_sections, comp);
 939         else
 940                 return crypt_journal(ic, encrypt, section, n_sections, comp);
 941 }
 942
 943 static void complete_journal_io(unsigned long error, void *context)
 944 {
 945         struct journal_completion *comp = context;
 946         if (unlikely(error != 0))
 947                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
 948         complete_journal_op(comp);
 949 }
 950
 951 static void rw_journal_sectors(struct dm_integrity_c *ic, int op, int op_flags,
 952                                unsigned sector, unsigned n_sectors, struct journal_completion *comp)
 953 {
 954         struct dm_io_request io_req;
 955         struct dm_io_region io_loc;
 956         unsigned pl_index, pl_offset;
 957         int r;
 958
 959         if (unlikely(dm_integrity_failed(ic))) {
 960                 if (comp)
 961                         complete_journal_io(-1UL, comp);
 962                 return;
 963         }
 964
 965         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 966         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 967
 968         io_req.bi_op = op;
 969         io_req.bi_op_flags = op_flags;
 970         io_req.mem.type = DM_IO_PAGE_LIST;
 971         if (ic->journal_io)
 972                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
 973         else
 974                 io_req.mem.ptr.pl = &ic->journal[pl_index];
 975         io_req.mem.offset = pl_offset;
 976         if (likely(comp != NULL)) {
 977                 io_req.notify.fn = complete_journal_io;
 978                 io_req.notify.context = comp;
 979         } else {
 980                 io_req.notify.fn = NULL;
 981         }
 982         io_req.client = ic->io;
 983         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 984         io_loc.sector = ic->start + SB_SECTORS + sector;
 985         io_loc.count = n_sectors;
 986
 987         r = dm_io(&io_req, 1, &io_loc, NULL);
 988         if (unlikely(r)) {
 989                 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
 990                 if (comp) {
 991                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 992                         complete_journal_io(-1UL, comp);
 993                 }
 994         }
 995 }
 996
 997 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
 998                        unsigned n_sections, struct journal_completion *comp)
 999 {
1000         unsigned sector, n_sectors;
1001
1002         sector = section * ic->journal_section_sectors;
1003         n_sectors = n_sections * ic->journal_section_sectors;
1004
1005         rw_journal_sectors(ic, op, op_flags, sector, n_sectors, comp);
1006 }
1007
1008 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
1009 {
1010         struct journal_completion io_comp;
1011         struct journal_completion crypt_comp_1;
1012         struct journal_completion crypt_comp_2;
1013         unsigned i;
1014
1015         io_comp.ic = ic;
1016         init_completion(&io_comp.comp);
1017
1018         if (commit_start + commit_sections <= ic->journal_sections) {
1019                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1020                 if (ic->journal_io) {
1021                         crypt_comp_1.ic = ic;
1022                         init_completion(&crypt_comp_1.comp);
1023                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1024                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1025                         wait_for_completion_io(&crypt_comp_1.comp);
1026                 } else {
1027                         for (i = 0; i < commit_sections; i++)
1028                                 rw_section_mac(ic, commit_start + i, true);
1029                 }
1030                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
1031                            commit_sections, &io_comp);
1032         } else {
1033                 unsigned to_end;
1034                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1035                 to_end = ic->journal_sections - commit_start;
1036                 if (ic->journal_io) {
1037                         crypt_comp_1.ic = ic;
1038                         init_completion(&crypt_comp_1.comp);
1039                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1040                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1041                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
1042                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1043                                 reinit_completion(&crypt_comp_1.comp);
1044                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1045                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1046                                 wait_for_completion_io(&crypt_comp_1.comp);
1047                         } else {
1048                                 crypt_comp_2.ic = ic;
1049                                 init_completion(&crypt_comp_2.comp);
1050                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1051                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1052                                 wait_for_completion_io(&crypt_comp_1.comp);
1053                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1054                                 wait_for_completion_io(&crypt_comp_2.comp);
1055                         }
1056                 } else {
1057                         for (i = 0; i < to_end; i++)
1058                                 rw_section_mac(ic, commit_start + i, true);
1059                         rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1060                         for (i = 0; i < commit_sections - to_end; i++)
1061                                 rw_section_mac(ic, i, true);
1062                 }
1063                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
1064         }
1065
1066         wait_for_completion_io(&io_comp.comp);
1067 }
1068
1069 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1070                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1071 {
1072         struct dm_io_request io_req;
1073         struct dm_io_region io_loc;
1074         int r;
1075         unsigned sector, pl_index, pl_offset;
1076
1077         BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1078
1079         if (unlikely(dm_integrity_failed(ic))) {
1080                 fn(-1UL, data);
1081                 return;
1082         }
1083
1084         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1085
1086         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1087         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1088
1089         io_req.bi_op = REQ_OP_WRITE;
1090         io_req.bi_op_flags = 0;
1091         io_req.mem.type = DM_IO_PAGE_LIST;
1092         io_req.mem.ptr.pl = &ic->journal[pl_index];
1093         io_req.mem.offset = pl_offset;
1094         io_req.notify.fn = fn;
1095         io_req.notify.context = data;
1096         io_req.client = ic->io;
1097         io_loc.bdev = ic->dev->bdev;
1098         io_loc.sector = target;
1099         io_loc.count = n_sectors;
1100
1101         r = dm_io(&io_req, 1, &io_loc, NULL);
1102         if (unlikely(r)) {
1103                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1104                 fn(-1UL, data);
1105         }
1106 }
1107
1108 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1109 {
1110         return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1111                range1->logical_sector + range1->n_sectors > range2->logical_sector;
1112 }
1113
1114 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1115 {
1116         struct rb_node **n = &ic->in_progress.rb_node;
1117         struct rb_node *parent;
1118
1119         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1120
1121         if (likely(check_waiting)) {
1122                 struct dm_integrity_range *range;
1123                 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1124                         if (unlikely(ranges_overlap(range, new_range)))
1125                                 return false;
1126                 }
1127         }
1128
1129         parent = NULL;
1130
1131         while (*n) {
1132                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1133
1134                 parent = *n;
1135                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1136                         n = &range->node.rb_left;
1137                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1138                         n = &range->node.rb_right;
1139                 } else {
1140                         return false;
1141                 }
1142         }
1143
1144         rb_link_node(&new_range->node, parent, n);
1145         rb_insert_color(&new_range->node, &ic->in_progress);
1146
1147         return true;
1148 }
1149
1150 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1151 {
1152         rb_erase(&range->node, &ic->in_progress);
1153         while (unlikely(!list_empty(&ic->wait_list))) {
1154                 struct dm_integrity_range *last_range =
1155                         list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1156                 struct task_struct *last_range_task;
1157                 last_range_task = last_range->task;
1158                 list_del(&last_range->wait_entry);
1159                 if (!add_new_range(ic, last_range, false)) {
1160                         last_range->task = last_range_task;
1161                         list_add(&last_range->wait_entry, &ic->wait_list);
1162                         break;
1163                 }
1164                 last_range->waiting = false;
1165                 wake_up_process(last_range_task);
1166         }
1167 }
1168
1169 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1170 {
1171         unsigned long flags;
1172
1173         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1174         remove_range_unlocked(ic, range);
1175         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1176 }
1177
1178 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1179 {
1180         new_range->waiting = true;
1181         list_add_tail(&new_range->wait_entry, &ic->wait_list);
1182         new_range->task = current;
1183         do {
1184                 __set_current_state(TASK_UNINTERRUPTIBLE);
1185                 spin_unlock_irq(&ic->endio_wait.lock);
1186                 io_schedule();
1187                 spin_lock_irq(&ic->endio_wait.lock);
1188         } while (unlikely(new_range->waiting));
1189 }
1190
1191 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1192 {
1193         if (unlikely(!add_new_range(ic, new_range, true)))
1194                 wait_and_add_new_range(ic, new_range);
1195 }
1196
1197 static void init_journal_node(struct journal_node *node)
1198 {
1199         RB_CLEAR_NODE(&node->node);
1200         node->sector = (sector_t)-1;
1201 }
1202
1203 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1204 {
1205         struct rb_node **link;
1206         struct rb_node *parent;
1207
1208         node->sector = sector;
1209         BUG_ON(!RB_EMPTY_NODE(&node->node));
1210
1211         link = &ic->journal_tree_root.rb_node;
1212         parent = NULL;
1213
1214         while (*link) {
1215                 struct journal_node *j;
1216                 parent = *link;
1217                 j = container_of(parent, struct journal_node, node);
1218                 if (sector < j->sector)
1219                         link = &j->node.rb_left;
1220                 else
1221                         link = &j->node.rb_right;
1222         }
1223
1224         rb_link_node(&node->node, parent, link);
1225         rb_insert_color(&node->node, &ic->journal_tree_root);
1226 }
1227
1228 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1229 {
1230         BUG_ON(RB_EMPTY_NODE(&node->node));
1231         rb_erase(&node->node, &ic->journal_tree_root);
1232         init_journal_node(node);
1233 }
1234
1235 #define NOT_FOUND       (-1U)
1236
1237 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1238 {
1239         struct rb_node *n = ic->journal_tree_root.rb_node;
1240         unsigned found = NOT_FOUND;
1241         *next_sector = (sector_t)-1;
1242         while (n) {
1243                 struct journal_node *j = container_of(n, struct journal_node, node);
1244                 if (sector == j->sector) {
1245                         found = j - ic->journal_tree;
1246                 }
1247                 if (sector < j->sector) {
1248                         *next_sector = j->sector;
1249                         n = j->node.rb_left;
1250                 } else {
1251                         n = j->node.rb_right;
1252                 }
1253         }
1254
1255         return found;
1256 }
1257
1258 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1259 {
1260         struct journal_node *node, *next_node;
1261         struct rb_node *next;
1262
1263         if (unlikely(pos >= ic->journal_entries))
1264                 return false;
1265         node = &ic->journal_tree[pos];
1266         if (unlikely(RB_EMPTY_NODE(&node->node)))
1267                 return false;
1268         if (unlikely(node->sector != sector))
1269                 return false;
1270
1271         next = rb_next(&node->node);
1272         if (unlikely(!next))
1273                 return true;
1274
1275         next_node = container_of(next, struct journal_node, node);
1276         return next_node->sector != sector;
1277 }
1278
1279 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1280 {
1281         struct rb_node *next;
1282         struct journal_node *next_node;
1283         unsigned next_section;
1284
1285         BUG_ON(RB_EMPTY_NODE(&node->node));
1286
1287         next = rb_next(&node->node);
1288         if (unlikely(!next))
1289                 return false;
1290
1291         next_node = container_of(next, struct journal_node, node);
1292
1293         if (next_node->sector != node->sector)
1294                 return false;
1295
1296         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1297         if (next_section >= ic->committed_section &&
1298             next_section < ic->committed_section + ic->n_committed_sections)
1299                 return true;
1300         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1301                 return true;
1302
1303         return false;
1304 }
1305
1306 #define TAG_READ        0
1307 #define TAG_WRITE       1
1308 #define TAG_CMP         2
1309
1310 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1311                                unsigned *metadata_offset, unsigned total_size, int op)
1312 {
1313 #define MAY_BE_FILLER           1
1314 #define MAY_BE_HASH             2
1315         unsigned hash_offset = 0;
1316         unsigned may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1317
1318         do {
1319                 unsigned char *data, *dp;
1320                 struct dm_buffer *b;
1321                 unsigned to_copy;
1322                 int r;
1323
1324                 r = dm_integrity_failed(ic);
1325                 if (unlikely(r))
1326                         return r;
1327
1328                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1329                 if (IS_ERR(data))
1330                         return PTR_ERR(data);
1331
1332                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1333                 dp = data + *metadata_offset;
1334                 if (op == TAG_READ) {
1335                         memcpy(tag, dp, to_copy);
1336                 } else if (op == TAG_WRITE) {
1337                         memcpy(dp, tag, to_copy);
1338                         dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1339                 } else {
1340                         /* e.g.: op == TAG_CMP */
1341
1342                         if (likely(is_power_of_2(ic->tag_size))) {
1343                                 if (unlikely(memcmp(dp, tag, to_copy)))
1344                                         if (unlikely(!ic->discard) ||
1345                                             unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1346                                                 goto thorough_test;
1347                                 }
1348                         } else {
1349                                 unsigned i, ts;
1350 thorough_test:
1351                                 ts = total_size;
1352
1353                                 for (i = 0; i < to_copy; i++, ts--) {
1354                                         if (unlikely(dp[i] != tag[i]))
1355                                                 may_be &= ~MAY_BE_HASH;
1356                                         if (likely(dp[i] != DISCARD_FILLER))
1357                                                 may_be &= ~MAY_BE_FILLER;
1358                                         hash_offset++;
1359                                         if (unlikely(hash_offset == ic->tag_size)) {
1360                                                 if (unlikely(!may_be)) {
1361                                                         dm_bufio_release(b);
1362                                                         return ts;
1363                                                 }
1364                                                 hash_offset = 0;
1365                                                 may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1366                                         }
1367                                 }
1368                         }
1369                 }
1370                 dm_bufio_release(b);
1371
1372                 tag += to_copy;
1373                 *metadata_offset += to_copy;
1374                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1375                         (*metadata_block)++;
1376                         *metadata_offset = 0;
1377                 }
1378
1379                 if (unlikely(!is_power_of_2(ic->tag_size))) {
1380                         hash_offset = (hash_offset + to_copy) % ic->tag_size;
1381                 }
1382
1383                 total_size -= to_copy;
1384         } while (unlikely(total_size));
1385
1386         return 0;
1387 #undef MAY_BE_FILLER
1388 #undef MAY_BE_HASH
1389 }
1390
1391 struct flush_request {
1392         struct dm_io_request io_req;
1393         struct dm_io_region io_reg;
1394         struct dm_integrity_c *ic;
1395         struct completion comp;
1396 };
1397
1398 static void flush_notify(unsigned long error, void *fr_)
1399 {
1400         struct flush_request *fr = fr_;
1401         if (unlikely(error != 0))
1402                 dm_integrity_io_error(fr->ic, "flusing disk cache", -EIO);
1403         complete(&fr->comp);
1404 }
1405
1406 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1407 {
1408         int r;
1409
1410         struct flush_request fr;
1411
1412         if (!ic->meta_dev)
1413                 flush_data = false;
1414         if (flush_data) {
1415                 fr.io_req.bi_op = REQ_OP_WRITE,
1416                 fr.io_req.bi_op_flags = REQ_PREFLUSH | REQ_SYNC,
1417                 fr.io_req.mem.type = DM_IO_KMEM,
1418                 fr.io_req.mem.ptr.addr = NULL,
1419                 fr.io_req.notify.fn = flush_notify,
1420                 fr.io_req.notify.context = &fr;
1421                 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1422                 fr.io_reg.bdev = ic->dev->bdev,
1423                 fr.io_reg.sector = 0,
1424                 fr.io_reg.count = 0,
1425                 fr.ic = ic;
1426                 init_completion(&fr.comp);
1427                 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1428                 BUG_ON(r);
1429         }
1430
1431         r = dm_bufio_write_dirty_buffers(ic->bufio);
1432         if (unlikely(r))
1433                 dm_integrity_io_error(ic, "writing tags", r);
1434
1435         if (flush_data)
1436                 wait_for_completion(&fr.comp);
1437 }
1438
1439 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1440 {
1441         DECLARE_WAITQUEUE(wait, current);
1442         __add_wait_queue(&ic->endio_wait, &wait);
1443         __set_current_state(TASK_UNINTERRUPTIBLE);
1444         spin_unlock_irq(&ic->endio_wait.lock);
1445         io_schedule();
1446         spin_lock_irq(&ic->endio_wait.lock);
1447         __remove_wait_queue(&ic->endio_wait, &wait);
1448 }
1449
1450 static void autocommit_fn(struct timer_list *t)
1451 {
1452         struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1453
1454         if (likely(!dm_integrity_failed(ic)))
1455                 queue_work(ic->commit_wq, &ic->commit_work);
1456 }
1457
1458 static void schedule_autocommit(struct dm_integrity_c *ic)
1459 {
1460         if (!timer_pending(&ic->autocommit_timer))
1461                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1462 }
1463
1464 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1465 {
1466         struct bio *bio;
1467         unsigned long flags;
1468
1469         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1470         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1471         bio_list_add(&ic->flush_bio_list, bio);
1472         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1473
1474         queue_work(ic->commit_wq, &ic->commit_work);
1475 }
1476
1477 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1478 {
1479         int r = dm_integrity_failed(ic);
1480         if (unlikely(r) && !bio->bi_status)
1481                 bio->bi_status = errno_to_blk_status(r);
1482         if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1483                 unsigned long flags;
1484                 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1485                 bio_list_add(&ic->synchronous_bios, bio);
1486                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1487                 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1488                 return;
1489         }
1490         bio_endio(bio);
1491 }
1492
1493 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1494 {
1495         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1496
1497         if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1498                 submit_flush_bio(ic, dio);
1499         else
1500                 do_endio(ic, bio);
1501 }
1502
1503 static void dec_in_flight(struct dm_integrity_io *dio)
1504 {
1505         if (atomic_dec_and_test(&dio->in_flight)) {
1506                 struct dm_integrity_c *ic = dio->ic;
1507                 struct bio *bio;
1508
1509                 remove_range(ic, &dio->range);
1510
1511                 if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1512                         schedule_autocommit(ic);
1513
1514                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1515
1516                 if (unlikely(dio->bi_status) && !bio->bi_status)
1517                         bio->bi_status = dio->bi_status;
1518                 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1519                         dio->range.logical_sector += dio->range.n_sectors;
1520                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1521                         INIT_WORK(&dio->work, integrity_bio_wait);
1522                         queue_work(ic->offload_wq, &dio->work);
1523                         return;
1524                 }
1525                 do_endio_flush(ic, dio);
1526         }
1527 }
1528
1529 static void integrity_end_io(struct bio *bio)
1530 {
1531         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1532
1533         dm_bio_restore(&dio->bio_details, bio);
1534         if (bio->bi_integrity)
1535                 bio->bi_opf |= REQ_INTEGRITY;
1536
1537         if (dio->completion)
1538                 complete(dio->completion);
1539
1540         dec_in_flight(dio);
1541 }
1542
1543 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1544                                       const char *data, char *result)
1545 {
1546         __u64 sector_le = cpu_to_le64(sector);
1547         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1548         int r;
1549         unsigned digest_size;
1550
1551         req->tfm = ic->internal_hash;
1552
1553         r = crypto_shash_init(req);
1554         if (unlikely(r < 0)) {
1555                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1556                 goto failed;
1557         }
1558
1559         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1560         if (unlikely(r < 0)) {
1561                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1562                 goto failed;
1563         }
1564
1565         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1566         if (unlikely(r < 0)) {
1567                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1568                 goto failed;
1569         }
1570
1571         r = crypto_shash_final(req, result);
1572         if (unlikely(r < 0)) {
1573                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1574                 goto failed;
1575         }
1576
1577         digest_size = crypto_shash_digestsize(ic->internal_hash);
1578         if (unlikely(digest_size < ic->tag_size))
1579                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1580
1581         return;
1582
1583 failed:
1584         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1585         get_random_bytes(result, ic->tag_size);
1586 }
1587
1588 static void integrity_metadata(struct work_struct *w)
1589 {
1590         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1591         struct dm_integrity_c *ic = dio->ic;
1592
1593         int r;
1594
1595         if (ic->internal_hash) {
1596                 struct bvec_iter iter;
1597                 struct bio_vec bv;
1598                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1599                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1600                 char *checksums;
1601                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1602                 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1603                 sector_t sector;
1604                 unsigned sectors_to_process;
1605
1606                 if (unlikely(ic->mode == 'R'))
1607                         goto skip_io;
1608
1609                 if (likely(dio->op != REQ_OP_DISCARD))
1610                         checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1611                                             GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1612                 else
1613                         checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1614                 if (!checksums) {
1615                         checksums = checksums_onstack;
1616                         if (WARN_ON(extra_space &&
1617                                     digest_size > sizeof(checksums_onstack))) {
1618                                 r = -EINVAL;
1619                                 goto error;
1620                         }
1621                 }
1622
1623                 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1624                         sector_t bi_sector = dio->bio_details.bi_iter.bi_sector;
1625                         unsigned bi_size = dio->bio_details.bi_iter.bi_size;
1626                         unsigned max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1627                         unsigned max_blocks = max_size / ic->tag_size;
1628                         memset(checksums, DISCARD_FILLER, max_size);
1629
1630                         while (bi_size) {
1631                                 unsigned this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1632                                 this_step_blocks = min(this_step_blocks, max_blocks);
1633                                 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1634                                                         this_step_blocks * ic->tag_size, TAG_WRITE);
1635                                 if (unlikely(r)) {
1636                                         if (likely(checksums != checksums_onstack))
1637                                                 kfree(checksums);
1638                                         goto error;
1639                                 }
1640
1641                                 /*if (bi_size < this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block)) {
1642                                         printk("BUGG: bi_sector: %llx, bi_size: %u\n", bi_sector, bi_size);
1643                                         printk("BUGG: this_step_blocks: %u\n", this_step_blocks);
1644                                         BUG();
1645                                 }*/
1646                                 bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1647                                 bi_sector += this_step_blocks << ic->sb->log2_sectors_per_block;
1648                         }
1649
1650                         if (likely(checksums != checksums_onstack))
1651                                 kfree(checksums);
1652                         goto skip_io;
1653                 }
1654
1655                 sector = dio->range.logical_sector;
1656                 sectors_to_process = dio->range.n_sectors;
1657
1658                 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1659                         unsigned pos;
1660                         char *mem, *checksums_ptr;
1661
1662 again:
1663                         mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1664                         pos = 0;
1665                         checksums_ptr = checksums;
1666                         do {
1667                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1668                                 checksums_ptr += ic->tag_size;
1669                                 sectors_to_process -= ic->sectors_per_block;
1670                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1671                                 sector += ic->sectors_per_block;
1672                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1673                         kunmap_atomic(mem);
1674
1675                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1676                                                 checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1677                         if (unlikely(r)) {
1678                                 if (r > 0) {
1679                                         char b[BDEVNAME_SIZE];
1680                                         DMERR_LIMIT("%s: Checksum failed at sector 0x%llx", bio_devname(bio, b),
1681                                                     (sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1682                                         r = -EILSEQ;
1683                                         atomic64_inc(&ic->number_of_mismatches);
1684                                 }
1685                                 if (likely(checksums != checksums_onstack))
1686                                         kfree(checksums);
1687                                 goto error;
1688                         }
1689
1690                         if (!sectors_to_process)
1691                                 break;
1692
1693                         if (unlikely(pos < bv.bv_len)) {
1694                                 bv.bv_offset += pos;
1695                                 bv.bv_len -= pos;
1696                                 goto again;
1697                         }
1698                 }
1699
1700                 if (likely(checksums != checksums_onstack))
1701                         kfree(checksums);
1702         } else {
1703                 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1704
1705                 if (bip) {
1706                         struct bio_vec biv;
1707                         struct bvec_iter iter;
1708                         unsigned data_to_process = dio->range.n_sectors;
1709                         sector_to_block(ic, data_to_process);
1710                         data_to_process *= ic->tag_size;
1711
1712                         bip_for_each_vec(biv, bip, iter) {
1713                                 unsigned char *tag;
1714                                 unsigned this_len;
1715
1716                                 BUG_ON(PageHighMem(biv.bv_page));
1717                                 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1718                                 this_len = min(biv.bv_len, data_to_process);
1719                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1720                                                         this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1721                                 if (unlikely(r))
1722                                         goto error;
1723                                 data_to_process -= this_len;
1724                                 if (!data_to_process)
1725                                         break;
1726                         }
1727                 }
1728         }
1729 skip_io:
1730         dec_in_flight(dio);
1731         return;
1732 error:
1733         dio->bi_status = errno_to_blk_status(r);
1734         dec_in_flight(dio);
1735 }
1736
1737 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1738 {
1739         struct dm_integrity_c *ic = ti->private;
1740         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1741         struct bio_integrity_payload *bip;
1742
1743         sector_t area, offset;
1744
1745         dio->ic = ic;
1746         dio->bi_status = 0;
1747         dio->op = bio_op(bio);
1748
1749         if (unlikely(dio->op == REQ_OP_DISCARD)) {
1750                 if (ti->max_io_len) {
1751                         sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1752                         unsigned log2_max_io_len = __fls(ti->max_io_len);
1753                         sector_t start_boundary = sec >> log2_max_io_len;
1754                         sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1755                         if (start_boundary < end_boundary) {
1756                                 sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1757                                 dm_accept_partial_bio(bio, len);
1758                         }
1759                 }
1760         }
1761
1762         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1763                 submit_flush_bio(ic, dio);
1764                 return DM_MAPIO_SUBMITTED;
1765         }
1766
1767         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1768         dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1769         if (unlikely(dio->fua)) {
1770                 /*
1771                  * Don't pass down the FUA flag because we have to flush
1772                  * disk cache anyway.
1773                  */
1774                 bio->bi_opf &= ~REQ_FUA;
1775         }
1776         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1777                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1778                       dio->range.logical_sector, bio_sectors(bio),
1779                       ic->provided_data_sectors);
1780                 return DM_MAPIO_KILL;
1781         }
1782         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1783                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1784                       ic->sectors_per_block,
1785                       dio->range.logical_sector, bio_sectors(bio));
1786                 return DM_MAPIO_KILL;
1787         }
1788
1789         if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1790                 struct bvec_iter iter;
1791                 struct bio_vec bv;
1792                 bio_for_each_segment(bv, bio, iter) {
1793                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1794                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1795                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1796                                 return DM_MAPIO_KILL;
1797                         }
1798                 }
1799         }
1800
1801         bip = bio_integrity(bio);
1802         if (!ic->internal_hash) {
1803                 if (bip) {
1804                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1805                         if (ic->log2_tag_size >= 0)
1806                                 wanted_tag_size <<= ic->log2_tag_size;
1807                         else
1808                                 wanted_tag_size *= ic->tag_size;
1809                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1810                                 DMERR("Invalid integrity data size %u, expected %u",
1811                                       bip->bip_iter.bi_size, wanted_tag_size);
1812                                 return DM_MAPIO_KILL;
1813                         }
1814                 }
1815         } else {
1816                 if (unlikely(bip != NULL)) {
1817                         DMERR("Unexpected integrity data when using internal hash");
1818                         return DM_MAPIO_KILL;
1819                 }
1820         }
1821
1822         if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1823                 return DM_MAPIO_KILL;
1824
1825         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1826         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1827         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1828
1829         dm_integrity_map_continue(dio, true);
1830         return DM_MAPIO_SUBMITTED;
1831 }
1832
1833 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1834                                  unsigned journal_section, unsigned journal_entry)
1835 {
1836         struct dm_integrity_c *ic = dio->ic;
1837         sector_t logical_sector;
1838         unsigned n_sectors;
1839
1840         logical_sector = dio->range.logical_sector;
1841         n_sectors = dio->range.n_sectors;
1842         do {
1843                 struct bio_vec bv = bio_iovec(bio);
1844                 char *mem;
1845
1846                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1847                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1848                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1849                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1850 retry_kmap:
1851                 mem = kmap_atomic(bv.bv_page);
1852                 if (likely(dio->op == REQ_OP_WRITE))
1853                         flush_dcache_page(bv.bv_page);
1854
1855                 do {
1856                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1857
1858                         if (unlikely(dio->op == REQ_OP_READ)) {
1859                                 struct journal_sector *js;
1860                                 char *mem_ptr;
1861                                 unsigned s;
1862
1863                                 if (unlikely(journal_entry_is_inprogress(je))) {
1864                                         flush_dcache_page(bv.bv_page);
1865                                         kunmap_atomic(mem);
1866
1867                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1868                                         goto retry_kmap;
1869                                 }
1870                                 smp_rmb();
1871                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1872                                 js = access_journal_data(ic, journal_section, journal_entry);
1873                                 mem_ptr = mem + bv.bv_offset;
1874                                 s = 0;
1875                                 do {
1876                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1877                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1878                                         js++;
1879                                         mem_ptr += 1 << SECTOR_SHIFT;
1880                                 } while (++s < ic->sectors_per_block);
1881 #ifdef INTERNAL_VERIFY
1882                                 if (ic->internal_hash) {
1883                                         char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1884
1885                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1886                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1887                                                 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1888                                                             logical_sector);
1889                                         }
1890                                 }
1891 #endif
1892                         }
1893
1894                         if (!ic->internal_hash) {
1895                                 struct bio_integrity_payload *bip = bio_integrity(bio);
1896                                 unsigned tag_todo = ic->tag_size;
1897                                 char *tag_ptr = journal_entry_tag(ic, je);
1898
1899                                 if (bip) do {
1900                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1901                                         unsigned tag_now = min(biv.bv_len, tag_todo);
1902                                         char *tag_addr;
1903                                         BUG_ON(PageHighMem(biv.bv_page));
1904                                         tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1905                                         if (likely(dio->op == REQ_OP_WRITE))
1906                                                 memcpy(tag_ptr, tag_addr, tag_now);
1907                                         else
1908                                                 memcpy(tag_addr, tag_ptr, tag_now);
1909                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1910                                         tag_ptr += tag_now;
1911                                         tag_todo -= tag_now;
1912                                 } while (unlikely(tag_todo)); else {
1913                                         if (likely(dio->op == REQ_OP_WRITE))
1914                                                 memset(tag_ptr, 0, tag_todo);
1915                                 }
1916                         }
1917
1918                         if (likely(dio->op == REQ_OP_WRITE)) {
1919                                 struct journal_sector *js;
1920                                 unsigned s;
1921
1922                                 js = access_journal_data(ic, journal_section, journal_entry);
1923                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1924
1925                                 s = 0;
1926                                 do {
1927                                         je->last_bytes[s] = js[s].commit_id;
1928                                 } while (++s < ic->sectors_per_block);
1929
1930                                 if (ic->internal_hash) {
1931                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1932                                         if (unlikely(digest_size > ic->tag_size)) {
1933                                                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1934                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1935                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1936                                         } else
1937                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1938                                 }
1939
1940                                 journal_entry_set_sector(je, logical_sector);
1941                         }
1942                         logical_sector += ic->sectors_per_block;
1943
1944                         journal_entry++;
1945                         if (unlikely(journal_entry == ic->journal_section_entries)) {
1946                                 journal_entry = 0;
1947                                 journal_section++;
1948                                 wraparound_section(ic, &journal_section);
1949                         }
1950
1951                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1952                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1953
1954                 if (unlikely(dio->op == REQ_OP_READ))
1955                         flush_dcache_page(bv.bv_page);
1956                 kunmap_atomic(mem);
1957         } while (n_sectors);
1958
1959         if (likely(dio->op == REQ_OP_WRITE)) {
1960                 smp_mb();
1961                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1962                         wake_up(&ic->copy_to_journal_wait);
1963                 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1964                         queue_work(ic->commit_wq, &ic->commit_work);
1965                 } else {
1966                         schedule_autocommit(ic);
1967                 }
1968         } else {
1969                 remove_range(ic, &dio->range);
1970         }
1971
1972         if (unlikely(bio->bi_iter.bi_size)) {
1973                 sector_t area, offset;
1974
1975                 dio->range.logical_sector = logical_sector;
1976                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1977                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1978                 return true;
1979         }
1980
1981         return false;
1982 }
1983
1984 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1985 {
1986         struct dm_integrity_c *ic = dio->ic;
1987         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1988         unsigned journal_section, journal_entry;
1989         unsigned journal_read_pos;
1990         struct completion read_comp;
1991         bool discard_retried = false;
1992         bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
1993         if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
1994                 need_sync_io = true;
1995
1996         if (need_sync_io && from_map) {
1997                 INIT_WORK(&dio->work, integrity_bio_wait);
1998                 queue_work(ic->offload_wq, &dio->work);
1999                 return;
2000         }
2001
2002 lock_retry:
2003         spin_lock_irq(&ic->endio_wait.lock);
2004 retry:
2005         if (unlikely(dm_integrity_failed(ic))) {
2006                 spin_unlock_irq(&ic->endio_wait.lock);
2007                 do_endio(ic, bio);
2008                 return;
2009         }
2010         dio->range.n_sectors = bio_sectors(bio);
2011         journal_read_pos = NOT_FOUND;
2012         if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2013                 if (dio->op == REQ_OP_WRITE) {
2014                         unsigned next_entry, i, pos;
2015                         unsigned ws, we, range_sectors;
2016
2017                         dio->range.n_sectors = min(dio->range.n_sectors,
2018                                                    (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2019                         if (unlikely(!dio->range.n_sectors)) {
2020                                 if (from_map)
2021                                         goto offload_to_thread;
2022                                 sleep_on_endio_wait(ic);
2023                                 goto retry;
2024                         }
2025                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2026                         ic->free_sectors -= range_sectors;
2027                         journal_section = ic->free_section;
2028                         journal_entry = ic->free_section_entry;
2029
2030                         next_entry = ic->free_section_entry + range_sectors;
2031                         ic->free_section_entry = next_entry % ic->journal_section_entries;
2032                         ic->free_section += next_entry / ic->journal_section_entries;
2033                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2034                         wraparound_section(ic, &ic->free_section);
2035
2036                         pos = journal_section * ic->journal_section_entries + journal_entry;
2037                         ws = journal_section;
2038                         we = journal_entry;
2039                         i = 0;
2040                         do {
2041                                 struct journal_entry *je;
2042
2043                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2044                                 pos++;
2045                                 if (unlikely(pos >= ic->journal_entries))
2046                                         pos = 0;
2047
2048                                 je = access_journal_entry(ic, ws, we);
2049                                 BUG_ON(!journal_entry_is_unused(je));
2050                                 journal_entry_set_inprogress(je);
2051                                 we++;
2052                                 if (unlikely(we == ic->journal_section_entries)) {
2053                                         we = 0;
2054                                         ws++;
2055                                         wraparound_section(ic, &ws);
2056                                 }
2057                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2058
2059                         spin_unlock_irq(&ic->endio_wait.lock);
2060                         goto journal_read_write;
2061                 } else {
2062                         sector_t next_sector;
2063                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2064                         if (likely(journal_read_pos == NOT_FOUND)) {
2065                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2066                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
2067                         } else {
2068                                 unsigned i;
2069                                 unsigned jp = journal_read_pos + 1;
2070                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2071                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2072                                                 break;
2073                                 }
2074                                 dio->range.n_sectors = i;
2075                         }
2076                 }
2077         }
2078         if (unlikely(!add_new_range(ic, &dio->range, true))) {
2079                 /*
2080                  * We must not sleep in the request routine because it could
2081                  * stall bios on current->bio_list.
2082                  * So, we offload the bio to a workqueue if we have to sleep.
2083                  */
2084                 if (from_map) {
2085 offload_to_thread:
2086                         spin_unlock_irq(&ic->endio_wait.lock);
2087                         INIT_WORK(&dio->work, integrity_bio_wait);
2088                         queue_work(ic->wait_wq, &dio->work);
2089                         return;
2090                 }
2091                 if (journal_read_pos != NOT_FOUND)
2092                         dio->range.n_sectors = ic->sectors_per_block;
2093                 wait_and_add_new_range(ic, &dio->range);
2094                 /*
2095                  * wait_and_add_new_range drops the spinlock, so the journal
2096                  * may have been changed arbitrarily. We need to recheck.
2097                  * To simplify the code, we restrict I/O size to just one block.
2098                  */
2099                 if (journal_read_pos != NOT_FOUND) {
2100                         sector_t next_sector;
2101                         unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2102                         if (unlikely(new_pos != journal_read_pos)) {
2103                                 remove_range_unlocked(ic, &dio->range);
2104                                 goto retry;
2105                         }
2106                 }
2107         }
2108         if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2109                 sector_t next_sector;
2110                 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2111                 if (unlikely(new_pos != NOT_FOUND) ||
2112                     unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2113                         remove_range_unlocked(ic, &dio->range);
2114                         spin_unlock_irq(&ic->endio_wait.lock);
2115                         queue_work(ic->commit_wq, &ic->commit_work);
2116                         flush_workqueue(ic->commit_wq);
2117                         queue_work(ic->writer_wq, &ic->writer_work);
2118                         flush_workqueue(ic->writer_wq);
2119                         discard_retried = true;
2120                         goto lock_retry;
2121                 }
2122         }
2123         spin_unlock_irq(&ic->endio_wait.lock);
2124
2125         if (unlikely(journal_read_pos != NOT_FOUND)) {
2126                 journal_section = journal_read_pos / ic->journal_section_entries;
2127                 journal_entry = journal_read_pos % ic->journal_section_entries;
2128                 goto journal_read_write;
2129         }
2130
2131         if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2132                 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2133                                      dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2134                         struct bitmap_block_status *bbs;
2135
2136                         bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2137                         spin_lock(&bbs->bio_queue_lock);
2138                         bio_list_add(&bbs->bio_queue, bio);
2139                         spin_unlock(&bbs->bio_queue_lock);
2140                         queue_work(ic->writer_wq, &bbs->work);
2141                         return;
2142                 }
2143         }
2144
2145         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2146
2147         if (need_sync_io) {
2148                 init_completion(&read_comp);
2149                 dio->completion = &read_comp;
2150         } else
2151                 dio->completion = NULL;
2152
2153         dm_bio_record(&dio->bio_details, bio);
2154         bio_set_dev(bio, ic->dev->bdev);
2155         bio->bi_integrity = NULL;
2156         bio->bi_opf &= ~REQ_INTEGRITY;
2157         bio->bi_end_io = integrity_end_io;
2158         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2159
2160         if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2161                 integrity_metadata(&dio->work);
2162                 dm_integrity_flush_buffers(ic, false);
2163
2164                 dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2165                 dio->completion = NULL;
2166
2167                 submit_bio_noacct(bio);
2168
2169                 return;
2170         }
2171
2172         submit_bio_noacct(bio);
2173
2174         if (need_sync_io) {
2175                 wait_for_completion_io(&read_comp);
2176                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2177                     dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2178                         goto skip_check;
2179                 if (ic->mode == 'B') {
2180                         if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2181                                              dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2182                                 goto skip_check;
2183                 }
2184
2185                 if (likely(!bio->bi_status))
2186                         integrity_metadata(&dio->work);
2187                 else
2188 skip_check:
2189                         dec_in_flight(dio);
2190
2191         } else {
2192                 INIT_WORK(&dio->work, integrity_metadata);
2193                 queue_work(ic->metadata_wq, &dio->work);
2194         }
2195
2196         return;
2197
2198 journal_read_write:
2199         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2200                 goto lock_retry;
2201
2202         do_endio_flush(ic, dio);
2203 }
2204
2205
2206 static void integrity_bio_wait(struct work_struct *w)
2207 {
2208         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2209
2210         dm_integrity_map_continue(dio, false);
2211 }
2212
2213 static void pad_uncommitted(struct dm_integrity_c *ic)
2214 {
2215         if (ic->free_section_entry) {
2216                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2217                 ic->free_section_entry = 0;
2218                 ic->free_section++;
2219                 wraparound_section(ic, &ic->free_section);
2220                 ic->n_uncommitted_sections++;
2221         }
2222         if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2223                     (ic->n_uncommitted_sections + ic->n_committed_sections) *
2224                     ic->journal_section_entries + ic->free_sectors)) {
2225                 DMCRIT("journal_sections %u, journal_section_entries %u, "
2226                        "n_uncommitted_sections %u, n_committed_sections %u, "
2227                        "journal_section_entries %u, free_sectors %u",
2228                        ic->journal_sections, ic->journal_section_entries,
2229                        ic->n_uncommitted_sections, ic->n_committed_sections,
2230                        ic->journal_section_entries, ic->free_sectors);
2231         }
2232 }
2233
2234 static void integrity_commit(struct work_struct *w)
2235 {
2236         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2237         unsigned commit_start, commit_sections;
2238         unsigned i, j, n;
2239         struct bio *flushes;
2240
2241         del_timer(&ic->autocommit_timer);
2242
2243         spin_lock_irq(&ic->endio_wait.lock);
2244         flushes = bio_list_get(&ic->flush_bio_list);
2245         if (unlikely(ic->mode != 'J')) {
2246                 spin_unlock_irq(&ic->endio_wait.lock);
2247                 dm_integrity_flush_buffers(ic, true);
2248                 goto release_flush_bios;
2249         }
2250
2251         pad_uncommitted(ic);
2252         commit_start = ic->uncommitted_section;
2253         commit_sections = ic->n_uncommitted_sections;
2254         spin_unlock_irq(&ic->endio_wait.lock);
2255
2256         if (!commit_sections)
2257                 goto release_flush_bios;
2258
2259         i = commit_start;
2260         for (n = 0; n < commit_sections; n++) {
2261                 for (j = 0; j < ic->journal_section_entries; j++) {
2262                         struct journal_entry *je;
2263                         je = access_journal_entry(ic, i, j);
2264                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2265                 }
2266                 for (j = 0; j < ic->journal_section_sectors; j++) {
2267                         struct journal_sector *js;
2268                         js = access_journal(ic, i, j);
2269                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2270                 }
2271                 i++;
2272                 if (unlikely(i >= ic->journal_sections))
2273                         ic->commit_seq = next_commit_seq(ic->commit_seq);
2274                 wraparound_section(ic, &i);
2275         }
2276         smp_rmb();
2277
2278         write_journal(ic, commit_start, commit_sections);
2279
2280         spin_lock_irq(&ic->endio_wait.lock);
2281         ic->uncommitted_section += commit_sections;
2282         wraparound_section(ic, &ic->uncommitted_section);
2283         ic->n_uncommitted_sections -= commit_sections;
2284         ic->n_committed_sections += commit_sections;
2285         spin_unlock_irq(&ic->endio_wait.lock);
2286
2287         if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2288                 queue_work(ic->writer_wq, &ic->writer_work);
2289
2290 release_flush_bios:
2291         while (flushes) {
2292                 struct bio *next = flushes->bi_next;
2293                 flushes->bi_next = NULL;
2294                 do_endio(ic, flushes);
2295                 flushes = next;
2296         }
2297 }
2298
2299 static void complete_copy_from_journal(unsigned long error, void *context)
2300 {
2301         struct journal_io *io = context;
2302         struct journal_completion *comp = io->comp;
2303         struct dm_integrity_c *ic = comp->ic;
2304         remove_range(ic, &io->range);
2305         mempool_free(io, &ic->journal_io_mempool);
2306         if (unlikely(error != 0))
2307                 dm_integrity_io_error(ic, "copying from journal", -EIO);
2308         complete_journal_op(comp);
2309 }
2310
2311 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2312                                struct journal_entry *je)
2313 {
2314         unsigned s = 0;
2315         do {
2316                 js->commit_id = je->last_bytes[s];
2317                 js++;
2318         } while (++s < ic->sectors_per_block);
2319 }
2320
2321 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2322                              unsigned write_sections, bool from_replay)
2323 {
2324         unsigned i, j, n;
2325         struct journal_completion comp;
2326         struct blk_plug plug;
2327
2328         blk_start_plug(&plug);
2329
2330         comp.ic = ic;
2331         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2332         init_completion(&comp.comp);
2333
2334         i = write_start;
2335         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2336 #ifndef INTERNAL_VERIFY
2337                 if (unlikely(from_replay))
2338 #endif
2339                         rw_section_mac(ic, i, false);
2340                 for (j = 0; j < ic->journal_section_entries; j++) {
2341                         struct journal_entry *je = access_journal_entry(ic, i, j);
2342                         sector_t sec, area, offset;
2343                         unsigned k, l, next_loop;
2344                         sector_t metadata_block;
2345                         unsigned metadata_offset;
2346                         struct journal_io *io;
2347
2348                         if (journal_entry_is_unused(je))
2349                                 continue;
2350                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2351                         sec = journal_entry_get_sector(je);
2352                         if (unlikely(from_replay)) {
2353                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2354                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2355                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
2356                                 }
2357                         }
2358                         if (unlikely(sec >= ic->provided_data_sectors))
2359                                 continue;
2360                         get_area_and_offset(ic, sec, &area, &offset);
2361                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2362                         for (k = j + 1; k < ic->journal_section_entries; k++) {
2363                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2364                                 sector_t sec2, area2, offset2;
2365                                 if (journal_entry_is_unused(je2))
2366                                         break;
2367                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2368                                 sec2 = journal_entry_get_sector(je2);
2369                                 if (unlikely(sec2 >= ic->provided_data_sectors))
2370                                         break;
2371                                 get_area_and_offset(ic, sec2, &area2, &offset2);
2372                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2373                                         break;
2374                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2375                         }
2376                         next_loop = k - 1;
2377
2378                         io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2379                         io->comp = &comp;
2380                         io->range.logical_sector = sec;
2381                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2382
2383                         spin_lock_irq(&ic->endio_wait.lock);
2384                         add_new_range_and_wait(ic, &io->range);
2385
2386                         if (likely(!from_replay)) {
2387                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2388
2389                                 /* don't write if there is newer committed sector */
2390                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2391                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
2392
2393                                         journal_entry_set_unused(je2);
2394                                         remove_journal_node(ic, &section_node[j]);
2395                                         j++;
2396                                         sec += ic->sectors_per_block;
2397                                         offset += ic->sectors_per_block;
2398                                 }
2399                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2400                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2401
2402                                         journal_entry_set_unused(je2);
2403                                         remove_journal_node(ic, &section_node[k - 1]);
2404                                         k--;
2405                                 }
2406                                 if (j == k) {
2407                                         remove_range_unlocked(ic, &io->range);
2408                                         spin_unlock_irq(&ic->endio_wait.lock);
2409                                         mempool_free(io, &ic->journal_io_mempool);
2410                                         goto skip_io;
2411                                 }
2412                                 for (l = j; l < k; l++) {
2413                                         remove_journal_node(ic, &section_node[l]);
2414                                 }
2415                         }
2416                         spin_unlock_irq(&ic->endio_wait.lock);
2417
2418                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2419                         for (l = j; l < k; l++) {
2420                                 int r;
2421                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2422
2423                                 if (
2424 #ifndef INTERNAL_VERIFY
2425                                     unlikely(from_replay) &&
2426 #endif
2427                                     ic->internal_hash) {
2428                                         char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2429
2430                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2431                                                                   (char *)access_journal_data(ic, i, l), test_tag);
2432                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2433                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2434                                 }
2435
2436                                 journal_entry_set_unused(je2);
2437                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2438                                                         ic->tag_size, TAG_WRITE);
2439                                 if (unlikely(r)) {
2440                                         dm_integrity_io_error(ic, "reading tags", r);
2441                                 }
2442                         }
2443
2444                         atomic_inc(&comp.in_flight);
2445                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2446                                           (k - j) << ic->sb->log2_sectors_per_block,
2447                                           get_data_sector(ic, area, offset),
2448                                           complete_copy_from_journal, io);
2449 skip_io:
2450                         j = next_loop;
2451                 }
2452         }
2453
2454         dm_bufio_write_dirty_buffers_async(ic->bufio);
2455
2456         blk_finish_plug(&plug);
2457
2458         complete_journal_op(&comp);
2459         wait_for_completion_io(&comp.comp);
2460
2461         dm_integrity_flush_buffers(ic, true);
2462 }
2463
2464 static void integrity_writer(struct work_struct *w)
2465 {
2466         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2467         unsigned write_start, write_sections;
2468
2469         unsigned prev_free_sectors;
2470
2471         /* the following test is not needed, but it tests the replay code */
2472         if (unlikely(dm_post_suspending(ic->ti)) && !ic->meta_dev)
2473                 return;
2474
2475         spin_lock_irq(&ic->endio_wait.lock);
2476         write_start = ic->committed_section;
2477         write_sections = ic->n_committed_sections;
2478         spin_unlock_irq(&ic->endio_wait.lock);
2479
2480         if (!write_sections)
2481                 return;
2482
2483         do_journal_write(ic, write_start, write_sections, false);
2484
2485         spin_lock_irq(&ic->endio_wait.lock);
2486
2487         ic->committed_section += write_sections;
2488         wraparound_section(ic, &ic->committed_section);
2489         ic->n_committed_sections -= write_sections;
2490
2491         prev_free_sectors = ic->free_sectors;
2492         ic->free_sectors += write_sections * ic->journal_section_entries;
2493         if (unlikely(!prev_free_sectors))
2494                 wake_up_locked(&ic->endio_wait);
2495
2496         spin_unlock_irq(&ic->endio_wait.lock);
2497 }
2498
2499 static void recalc_write_super(struct dm_integrity_c *ic)
2500 {
2501         int r;
2502
2503         dm_integrity_flush_buffers(ic, false);
2504         if (dm_integrity_failed(ic))
2505                 return;
2506
2507         r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2508         if (unlikely(r))
2509                 dm_integrity_io_error(ic, "writing superblock", r);
2510 }
2511
2512 static void integrity_recalc(struct work_struct *w)
2513 {
2514         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2515         struct dm_integrity_range range;
2516         struct dm_io_request io_req;
2517         struct dm_io_region io_loc;
2518         sector_t area, offset;
2519         sector_t metadata_block;
2520         unsigned metadata_offset;
2521         sector_t logical_sector, n_sectors;
2522         __u8 *t;
2523         unsigned i;
2524         int r;
2525         unsigned super_counter = 0;
2526
2527         DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2528
2529         spin_lock_irq(&ic->endio_wait.lock);
2530
2531 next_chunk:
2532
2533         if (unlikely(dm_post_suspending(ic->ti)))
2534                 goto unlock_ret;
2535
2536         range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2537         if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2538                 if (ic->mode == 'B') {
2539                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2540                         DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2541                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2542                 }
2543                 goto unlock_ret;
2544         }
2545
2546         get_area_and_offset(ic, range.logical_sector, &area, &offset);
2547         range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2548         if (!ic->meta_dev)
2549                 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2550
2551         add_new_range_and_wait(ic, &range);
2552         spin_unlock_irq(&ic->endio_wait.lock);
2553         logical_sector = range.logical_sector;
2554         n_sectors = range.n_sectors;
2555
2556         if (ic->mode == 'B') {
2557                 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2558                         goto advance_and_next;
2559                 }
2560                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2561                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2562                         logical_sector += ic->sectors_per_block;
2563                         n_sectors -= ic->sectors_per_block;
2564                         cond_resched();
2565                 }
2566                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2567                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2568                         n_sectors -= ic->sectors_per_block;
2569                         cond_resched();
2570                 }
2571                 get_area_and_offset(ic, logical_sector, &area, &offset);
2572         }
2573
2574         DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2575
2576         if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2577                 recalc_write_super(ic);
2578                 if (ic->mode == 'B') {
2579                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2580                 }
2581                 super_counter = 0;
2582         }
2583
2584         if (unlikely(dm_integrity_failed(ic)))
2585                 goto err;
2586
2587         io_req.bi_op = REQ_OP_READ;
2588         io_req.bi_op_flags = 0;
2589         io_req.mem.type = DM_IO_VMA;
2590         io_req.mem.ptr.addr = ic->recalc_buffer;
2591         io_req.notify.fn = NULL;
2592         io_req.client = ic->io;
2593         io_loc.bdev = ic->dev->bdev;
2594         io_loc.sector = get_data_sector(ic, area, offset);
2595         io_loc.count = n_sectors;
2596
2597         r = dm_io(&io_req, 1, &io_loc, NULL);
2598         if (unlikely(r)) {
2599                 dm_integrity_io_error(ic, "reading data", r);
2600                 goto err;
2601         }
2602
2603         t = ic->recalc_tags;
2604         for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2605                 integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2606                 t += ic->tag_size;
2607         }
2608
2609         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2610
2611         r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2612         if (unlikely(r)) {
2613                 dm_integrity_io_error(ic, "writing tags", r);
2614                 goto err;
2615         }
2616
2617         if (ic->mode == 'B') {
2618                 sector_t start, end;
2619                 start = (range.logical_sector >>
2620                          (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2621                         (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2622                 end = ((range.logical_sector + range.n_sectors) >>
2623                        (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2624                         (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2625                 block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2626         }
2627
2628 advance_and_next:
2629         cond_resched();
2630
2631         spin_lock_irq(&ic->endio_wait.lock);
2632         remove_range_unlocked(ic, &range);
2633         ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2634         goto next_chunk;
2635
2636 err:
2637         remove_range(ic, &range);
2638         return;
2639
2640 unlock_ret:
2641         spin_unlock_irq(&ic->endio_wait.lock);
2642
2643         recalc_write_super(ic);
2644 }
2645
2646 static void bitmap_block_work(struct work_struct *w)
2647 {
2648         struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2649         struct dm_integrity_c *ic = bbs->ic;
2650         struct bio *bio;
2651         struct bio_list bio_queue;
2652         struct bio_list waiting;
2653
2654         bio_list_init(&waiting);
2655
2656         spin_lock(&bbs->bio_queue_lock);
2657         bio_queue = bbs->bio_queue;
2658         bio_list_init(&bbs->bio_queue);
2659         spin_unlock(&bbs->bio_queue_lock);
2660
2661         while ((bio = bio_list_pop(&bio_queue))) {
2662                 struct dm_integrity_io *dio;
2663
2664                 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2665
2666                 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2667                                     dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2668                         remove_range(ic, &dio->range);
2669                         INIT_WORK(&dio->work, integrity_bio_wait);
2670                         queue_work(ic->offload_wq, &dio->work);
2671                 } else {
2672                         block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2673                                         dio->range.n_sectors, BITMAP_OP_SET);
2674                         bio_list_add(&waiting, bio);
2675                 }
2676         }
2677
2678         if (bio_list_empty(&waiting))
2679                 return;
2680
2681         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC,
2682                            bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2683                            BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2684
2685         while ((bio = bio_list_pop(&waiting))) {
2686                 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2687
2688                 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2689                                 dio->range.n_sectors, BITMAP_OP_SET);
2690
2691                 remove_range(ic, &dio->range);
2692                 INIT_WORK(&dio->work, integrity_bio_wait);
2693                 queue_work(ic->offload_wq, &dio->work);
2694         }
2695
2696         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2697 }
2698
2699 static void bitmap_flush_work(struct work_struct *work)
2700 {
2701         struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2702         struct dm_integrity_range range;
2703         unsigned long limit;
2704         struct bio *bio;
2705
2706         dm_integrity_flush_buffers(ic, false);
2707
2708         range.logical_sector = 0;
2709         range.n_sectors = ic->provided_data_sectors;
2710
2711         spin_lock_irq(&ic->endio_wait.lock);
2712         add_new_range_and_wait(ic, &range);
2713         spin_unlock_irq(&ic->endio_wait.lock);
2714
2715         dm_integrity_flush_buffers(ic, true);
2716
2717         limit = ic->provided_data_sectors;
2718         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2719                 limit = le64_to_cpu(ic->sb->recalc_sector)
2720                         >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2721                         << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2722         }
2723         /*DEBUG_print("zeroing journal\n");*/
2724         block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2725         block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2726
2727         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2728                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2729
2730         spin_lock_irq(&ic->endio_wait.lock);
2731         remove_range_unlocked(ic, &range);
2732         while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2733                 bio_endio(bio);
2734                 spin_unlock_irq(&ic->endio_wait.lock);
2735                 spin_lock_irq(&ic->endio_wait.lock);
2736         }
2737         spin_unlock_irq(&ic->endio_wait.lock);
2738 }
2739
2740
2741 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2742                          unsigned n_sections, unsigned char commit_seq)
2743 {
2744         unsigned i, j, n;
2745
2746         if (!n_sections)
2747                 return;
2748
2749         for (n = 0; n < n_sections; n++) {
2750                 i = start_section + n;
2751                 wraparound_section(ic, &i);
2752                 for (j = 0; j < ic->journal_section_sectors; j++) {
2753                         struct journal_sector *js = access_journal(ic, i, j);
2754                         memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2755                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2756                 }
2757                 for (j = 0; j < ic->journal_section_entries; j++) {
2758                         struct journal_entry *je = access_journal_entry(ic, i, j);
2759                         journal_entry_set_unused(je);
2760                 }
2761         }
2762
2763         write_journal(ic, start_section, n_sections);
2764 }
2765
2766 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2767 {
2768         unsigned char k;
2769         for (k = 0; k < N_COMMIT_IDS; k++) {
2770                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2771                         return k;
2772         }
2773         dm_integrity_io_error(ic, "journal commit id", -EIO);
2774         return -EIO;
2775 }
2776
2777 static void replay_journal(struct dm_integrity_c *ic)
2778 {
2779         unsigned i, j;
2780         bool used_commit_ids[N_COMMIT_IDS];
2781         unsigned max_commit_id_sections[N_COMMIT_IDS];
2782         unsigned write_start, write_sections;
2783         unsigned continue_section;
2784         bool journal_empty;
2785         unsigned char unused, last_used, want_commit_seq;
2786
2787         if (ic->mode == 'R')
2788                 return;
2789
2790         if (ic->journal_uptodate)
2791                 return;
2792
2793         last_used = 0;
2794         write_start = 0;
2795
2796         if (!ic->just_formatted) {
2797                 DEBUG_print("reading journal\n");
2798                 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2799                 if (ic->journal_io)
2800                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2801                 if (ic->journal_io) {
2802                         struct journal_completion crypt_comp;
2803                         crypt_comp.ic = ic;
2804                         init_completion(&crypt_comp.comp);
2805                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2806                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2807                         wait_for_completion(&crypt_comp.comp);
2808                 }
2809                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2810         }
2811
2812         if (dm_integrity_failed(ic))
2813                 goto clear_journal;
2814
2815         journal_empty = true;
2816         memset(used_commit_ids, 0, sizeof used_commit_ids);
2817         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2818         for (i = 0; i < ic->journal_sections; i++) {
2819                 for (j = 0; j < ic->journal_section_sectors; j++) {
2820                         int k;
2821                         struct journal_sector *js = access_journal(ic, i, j);
2822                         k = find_commit_seq(ic, i, j, js->commit_id);
2823                         if (k < 0)
2824                                 goto clear_journal;
2825                         used_commit_ids[k] = true;
2826                         max_commit_id_sections[k] = i;
2827                 }
2828                 if (journal_empty) {
2829                         for (j = 0; j < ic->journal_section_entries; j++) {
2830                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2831                                 if (!journal_entry_is_unused(je)) {
2832                                         journal_empty = false;
2833                                         break;
2834                                 }
2835                         }
2836                 }
2837         }
2838
2839         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2840                 unused = N_COMMIT_IDS - 1;
2841                 while (unused && !used_commit_ids[unused - 1])
2842                         unused--;
2843         } else {
2844                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2845                         if (!used_commit_ids[unused])
2846                                 break;
2847                 if (unused == N_COMMIT_IDS) {
2848                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2849                         goto clear_journal;
2850                 }
2851         }
2852         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2853                     unused, used_commit_ids[0], used_commit_ids[1],
2854                     used_commit_ids[2], used_commit_ids[3]);
2855
2856         last_used = prev_commit_seq(unused);
2857         want_commit_seq = prev_commit_seq(last_used);
2858
2859         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2860                 journal_empty = true;
2861
2862         write_start = max_commit_id_sections[last_used] + 1;
2863         if (unlikely(write_start >= ic->journal_sections))
2864                 want_commit_seq = next_commit_seq(want_commit_seq);
2865         wraparound_section(ic, &write_start);
2866
2867         i = write_start;
2868         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2869                 for (j = 0; j < ic->journal_section_sectors; j++) {
2870                         struct journal_sector *js = access_journal(ic, i, j);
2871
2872                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2873                                 /*
2874                                  * This could be caused by crash during writing.
2875                                  * We won't replay the inconsistent part of the
2876                                  * journal.
2877                                  */
2878                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2879                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2880                                 goto brk;
2881                         }
2882                 }
2883                 i++;
2884                 if (unlikely(i >= ic->journal_sections))
2885                         want_commit_seq = next_commit_seq(want_commit_seq);
2886                 wraparound_section(ic, &i);
2887         }
2888 brk:
2889
2890         if (!journal_empty) {
2891                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2892                             write_sections, write_start, want_commit_seq);
2893                 do_journal_write(ic, write_start, write_sections, true);
2894         }
2895
2896         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2897                 continue_section = write_start;
2898                 ic->commit_seq = want_commit_seq;
2899                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2900         } else {
2901                 unsigned s;
2902                 unsigned char erase_seq;
2903 clear_journal:
2904                 DEBUG_print("clearing journal\n");
2905
2906                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2907                 s = write_start;
2908                 init_journal(ic, s, 1, erase_seq);
2909                 s++;
2910                 wraparound_section(ic, &s);
2911                 if (ic->journal_sections >= 2) {
2912                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2913                         s += ic->journal_sections - 2;
2914                         wraparound_section(ic, &s);
2915                         init_journal(ic, s, 1, erase_seq);
2916                 }
2917
2918                 continue_section = 0;
2919                 ic->commit_seq = next_commit_seq(erase_seq);
2920         }
2921
2922         ic->committed_section = continue_section;
2923         ic->n_committed_sections = 0;
2924
2925         ic->uncommitted_section = continue_section;
2926         ic->n_uncommitted_sections = 0;
2927
2928         ic->free_section = continue_section;
2929         ic->free_section_entry = 0;
2930         ic->free_sectors = ic->journal_entries;
2931
2932         ic->journal_tree_root = RB_ROOT;
2933         for (i = 0; i < ic->journal_entries; i++)
2934                 init_journal_node(&ic->journal_tree[i]);
2935 }
2936
2937 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
2938 {
2939         DEBUG_print("dm_integrity_enter_synchronous_mode\n");
2940
2941         if (ic->mode == 'B') {
2942                 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
2943                 ic->synchronous_mode = 1;
2944
2945                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2946                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2947                 flush_workqueue(ic->commit_wq);
2948         }
2949 }
2950
2951 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
2952 {
2953         struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
2954
2955         DEBUG_print("dm_integrity_reboot\n");
2956
2957         dm_integrity_enter_synchronous_mode(ic);
2958
2959         return NOTIFY_DONE;
2960 }
2961
2962 static void dm_integrity_postsuspend(struct dm_target *ti)
2963 {
2964         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2965         int r;
2966
2967         WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
2968
2969         del_timer_sync(&ic->autocommit_timer);
2970
2971         if (ic->recalc_wq)
2972                 drain_workqueue(ic->recalc_wq);
2973
2974         if (ic->mode == 'B')
2975                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2976
2977         queue_work(ic->commit_wq, &ic->commit_work);
2978         drain_workqueue(ic->commit_wq);
2979
2980         if (ic->mode == 'J') {
2981                 if (ic->meta_dev)
2982                         queue_work(ic->writer_wq, &ic->writer_work);
2983                 drain_workqueue(ic->writer_wq);
2984                 dm_integrity_flush_buffers(ic, true);
2985         }
2986
2987         if (ic->mode == 'B') {
2988                 dm_integrity_flush_buffers(ic, true);
2989 #if 1
2990                 /* set to 0 to test bitmap replay code */
2991                 init_journal(ic, 0, ic->journal_sections, 0);
2992                 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2993                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2994                 if (unlikely(r))
2995                         dm_integrity_io_error(ic, "writing superblock", r);
2996 #endif
2997         }
2998
2999         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3000
3001         ic->journal_uptodate = true;
3002 }
3003
3004 static void dm_integrity_resume(struct dm_target *ti)
3005 {
3006         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3007         __u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3008         int r;
3009
3010         DEBUG_print("resume\n");
3011
3012         if (ic->provided_data_sectors != old_provided_data_sectors) {
3013                 if (ic->provided_data_sectors > old_provided_data_sectors &&
3014                     ic->mode == 'B' &&
3015                     ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3016                         rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3017                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3018                         block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3019                                         ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3020                         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3021                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3022                 }
3023
3024                 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3025                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3026                 if (unlikely(r))
3027                         dm_integrity_io_error(ic, "writing superblock", r);
3028         }
3029
3030         if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3031                 DEBUG_print("resume dirty_bitmap\n");
3032                 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3033                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3034                 if (ic->mode == 'B') {
3035                         if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3036                                 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3037                                 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3038                                 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3039                                                      BITMAP_OP_TEST_ALL_CLEAR)) {
3040                                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3041                                         ic->sb->recalc_sector = cpu_to_le64(0);
3042                                 }
3043                         } else {
3044                                 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3045                                             ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3046                                 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3047                                 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3048                                 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3049                                 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3050                                 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3051                                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3052                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3053                                 ic->sb->recalc_sector = cpu_to_le64(0);
3054                         }
3055                 } else {
3056                         if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3057                               block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR))) {
3058                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3059                                 ic->sb->recalc_sector = cpu_to_le64(0);
3060                         }
3061                         init_journal(ic, 0, ic->journal_sections, 0);
3062                         replay_journal(ic);
3063                         ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3064                 }
3065                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3066                 if (unlikely(r))
3067                         dm_integrity_io_error(ic, "writing superblock", r);
3068         } else {
3069                 replay_journal(ic);
3070                 if (ic->mode == 'B') {
3071                         ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3072                         ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3073                         r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3074                         if (unlikely(r))
3075                                 dm_integrity_io_error(ic, "writing superblock", r);
3076
3077                         block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3078                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3079                         block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3080                         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3081                             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3082                                 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3083                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3084                                 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3085                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3086                                 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3087                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3088                         }
3089                         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3090                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3091                 }
3092         }
3093
3094         DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3095         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3096                 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3097                 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3098                 if (recalc_pos < ic->provided_data_sectors) {
3099                         queue_work(ic->recalc_wq, &ic->recalc_work);
3100                 } else if (recalc_pos > ic->provided_data_sectors) {
3101                         ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3102                         recalc_write_super(ic);
3103                 }
3104         }
3105
3106         ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3107         ic->reboot_notifier.next = NULL;
3108         ic->reboot_notifier.priority = INT_MAX - 1;     /* be notified after md and before hardware drivers */
3109         WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3110
3111 #if 0
3112         /* set to 1 to stress test synchronous mode */
3113         dm_integrity_enter_synchronous_mode(ic);
3114 #endif
3115 }
3116
3117 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3118                                 unsigned status_flags, char *result, unsigned maxlen)
3119 {
3120         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3121         unsigned arg_count;
3122         size_t sz = 0;
3123
3124         switch (type) {
3125         case STATUSTYPE_INFO:
3126                 DMEMIT("%llu %llu",
3127                         (unsigned long long)atomic64_read(&ic->number_of_mismatches),
3128                         ic->provided_data_sectors);
3129                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3130                         DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3131                 else
3132                         DMEMIT(" -");
3133                 break;
3134
3135         case STATUSTYPE_TABLE: {
3136                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3137                 watermark_percentage += ic->journal_entries / 2;
3138                 do_div(watermark_percentage, ic->journal_entries);
3139                 arg_count = 3;
3140                 arg_count += !!ic->meta_dev;
3141                 arg_count += ic->sectors_per_block != 1;
3142                 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3143                 arg_count += ic->discard;
3144                 arg_count += ic->mode == 'J';
3145                 arg_count += ic->mode == 'J';
3146                 arg_count += ic->mode == 'B';
3147                 arg_count += ic->mode == 'B';
3148                 arg_count += !!ic->internal_hash_alg.alg_string;
3149                 arg_count += !!ic->journal_crypt_alg.alg_string;
3150                 arg_count += !!ic->journal_mac_alg.alg_string;
3151                 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3152                 arg_count += ic->legacy_recalculate;
3153                 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3154                        ic->tag_size, ic->mode, arg_count);
3155                 if (ic->meta_dev)
3156                         DMEMIT(" meta_device:%s", ic->meta_dev->name);
3157                 if (ic->sectors_per_block != 1)
3158                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3159                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3160                         DMEMIT(" recalculate");
3161                 if (ic->discard)
3162                         DMEMIT(" allow_discards");
3163                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3164                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3165                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3166                 if (ic->mode == 'J') {
3167                         DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
3168                         DMEMIT(" commit_time:%u", ic->autocommit_msec);
3169                 }
3170                 if (ic->mode == 'B') {
3171                         DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3172                         DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3173                 }
3174                 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3175                         DMEMIT(" fix_padding");
3176                 if (ic->legacy_recalculate)
3177                         DMEMIT(" legacy_recalculate");
3178
3179 #define EMIT_ALG(a, n)                                                  \
3180                 do {                                                    \
3181                         if (ic->a.alg_string) {                         \
3182                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
3183                                 if (ic->a.key_string)                   \
3184                                         DMEMIT(":%s", ic->a.key_string);\
3185                         }                                               \
3186                 } while (0)
3187                 EMIT_ALG(internal_hash_alg, "internal_hash");
3188                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3189                 EMIT_ALG(journal_mac_alg, "journal_mac");
3190                 break;
3191         }
3192         }
3193 }
3194
3195 static int dm_integrity_iterate_devices(struct dm_target *ti,
3196                                         iterate_devices_callout_fn fn, void *data)
3197 {
3198         struct dm_integrity_c *ic = ti->private;
3199
3200         if (!ic->meta_dev)
3201                 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3202         else
3203                 return fn(ti, ic->dev, 0, ti->len, data);
3204 }
3205
3206 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3207 {
3208         struct dm_integrity_c *ic = ti->private;
3209
3210         if (ic->sectors_per_block > 1) {
3211                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3212                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3213                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3214         }
3215 }
3216
3217 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3218 {
3219         unsigned sector_space = JOURNAL_SECTOR_DATA;
3220
3221         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3222         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3223                                          JOURNAL_ENTRY_ROUNDUP);
3224
3225         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3226                 sector_space -= JOURNAL_MAC_PER_SECTOR;
3227         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3228         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3229         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3230         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3231 }
3232
3233 static int calculate_device_limits(struct dm_integrity_c *ic)
3234 {
3235         __u64 initial_sectors;
3236
3237         calculate_journal_section_size(ic);
3238         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3239         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3240                 return -EINVAL;
3241         ic->initial_sectors = initial_sectors;
3242
3243         if (!ic->meta_dev) {
3244                 sector_t last_sector, last_area, last_offset;
3245
3246                 /* we have to maintain excessive padding for compatibility with existing volumes */
3247                 __u64 metadata_run_padding =
3248                         ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3249                         (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3250                         (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3251
3252                 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3253                                             metadata_run_padding) >> SECTOR_SHIFT;
3254                 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3255                         ic->log2_metadata_run = __ffs(ic->metadata_run);
3256                 else
3257                         ic->log2_metadata_run = -1;
3258
3259                 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3260                 last_sector = get_data_sector(ic, last_area, last_offset);
3261                 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3262                         return -EINVAL;
3263         } else {
3264                 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3265                 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3266                                 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3267                 meta_size <<= ic->log2_buffer_sectors;
3268                 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3269                     ic->initial_sectors + meta_size > ic->meta_device_sectors)
3270                         return -EINVAL;
3271                 ic->metadata_run = 1;
3272                 ic->log2_metadata_run = 0;
3273         }
3274
3275         return 0;
3276 }
3277
3278 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3279 {
3280         if (!ic->meta_dev) {
3281                 int test_bit;
3282                 ic->provided_data_sectors = 0;
3283                 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3284                         __u64 prev_data_sectors = ic->provided_data_sectors;
3285
3286                         ic->provided_data_sectors |= (sector_t)1 << test_bit;
3287                         if (calculate_device_limits(ic))
3288                                 ic->provided_data_sectors = prev_data_sectors;
3289                 }
3290         } else {
3291                 ic->provided_data_sectors = ic->data_device_sectors;
3292                 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3293         }
3294 }
3295
3296 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3297 {
3298         unsigned journal_sections;
3299         int test_bit;
3300
3301         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3302         memcpy(ic->sb->magic, SB_MAGIC, 8);
3303         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3304         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3305         if (ic->journal_mac_alg.alg_string)
3306                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3307
3308         calculate_journal_section_size(ic);
3309         journal_sections = journal_sectors / ic->journal_section_sectors;
3310         if (!journal_sections)
3311                 journal_sections = 1;
3312
3313         if (!ic->meta_dev) {
3314                 if (ic->fix_padding)
3315                         ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3316                 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3317                 if (!interleave_sectors)
3318                         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3319                 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3320                 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3321                 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3322
3323                 get_provided_data_sectors(ic);
3324                 if (!ic->provided_data_sectors)
3325                         return -EINVAL;
3326         } else {
3327                 ic->sb->log2_interleave_sectors = 0;
3328
3329                 get_provided_data_sectors(ic);
3330                 if (!ic->provided_data_sectors)
3331                         return -EINVAL;
3332
3333 try_smaller_buffer:
3334                 ic->sb->journal_sections = cpu_to_le32(0);
3335                 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3336                         __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3337                         __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3338                         if (test_journal_sections > journal_sections)
3339                                 continue;
3340                         ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3341                         if (calculate_device_limits(ic))
3342                                 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3343
3344                 }
3345                 if (!le32_to_cpu(ic->sb->journal_sections)) {
3346                         if (ic->log2_buffer_sectors > 3) {
3347                                 ic->log2_buffer_sectors--;
3348                                 goto try_smaller_buffer;
3349                         }
3350                         return -EINVAL;
3351                 }
3352         }
3353
3354         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3355
3356         sb_set_version(ic);
3357
3358         return 0;
3359 }
3360
3361 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3362 {
3363         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3364         struct blk_integrity bi;
3365
3366         memset(&bi, 0, sizeof(bi));
3367         bi.profile = &dm_integrity_profile;
3368         bi.tuple_size = ic->tag_size;
3369         bi.tag_size = bi.tuple_size;
3370         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3371
3372         blk_integrity_register(disk, &bi);
3373         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3374 }
3375
3376 static void dm_integrity_free_page_list(struct page_list *pl)
3377 {
3378         unsigned i;
3379
3380         if (!pl)
3381                 return;
3382         for (i = 0; pl[i].page; i++)
3383                 __free_page(pl[i].page);
3384         kvfree(pl);
3385 }
3386
3387 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3388 {
3389         struct page_list *pl;
3390         unsigned i;
3391
3392         pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3393         if (!pl)
3394                 return NULL;
3395
3396         for (i = 0; i < n_pages; i++) {
3397                 pl[i].page = alloc_page(GFP_KERNEL);
3398                 if (!pl[i].page) {
3399                         dm_integrity_free_page_list(pl);
3400                         return NULL;
3401                 }
3402                 if (i)
3403                         pl[i - 1].next = &pl[i];
3404         }
3405         pl[i].page = NULL;
3406         pl[i].next = NULL;
3407
3408         return pl;
3409 }
3410
3411 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3412 {
3413         unsigned i;
3414         for (i = 0; i < ic->journal_sections; i++)
3415                 kvfree(sl[i]);
3416         kvfree(sl);
3417 }
3418
3419 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3420                                                                    struct page_list *pl)
3421 {
3422         struct scatterlist **sl;
3423         unsigned i;
3424
3425         sl = kvmalloc_array(ic->journal_sections,
3426                             sizeof(struct scatterlist *),
3427                             GFP_KERNEL | __GFP_ZERO);
3428         if (!sl)
3429                 return NULL;
3430
3431         for (i = 0; i < ic->journal_sections; i++) {
3432                 struct scatterlist *s;
3433                 unsigned start_index, start_offset;
3434                 unsigned end_index, end_offset;
3435                 unsigned n_pages;
3436                 unsigned idx;
3437
3438                 page_list_location(ic, i, 0, &start_index, &start_offset);
3439                 page_list_location(ic, i, ic->journal_section_sectors - 1,
3440                                    &end_index, &end_offset);
3441
3442                 n_pages = (end_index - start_index + 1);
3443
3444                 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3445                                    GFP_KERNEL);
3446                 if (!s) {
3447                         dm_integrity_free_journal_scatterlist(ic, sl);
3448                         return NULL;
3449                 }
3450
3451                 sg_init_table(s, n_pages);
3452                 for (idx = start_index; idx <= end_index; idx++) {
3453                         char *va = lowmem_page_address(pl[idx].page);
3454                         unsigned start = 0, end = PAGE_SIZE;
3455                         if (idx == start_index)
3456                                 start = start_offset;
3457                         if (idx == end_index)
3458                                 end = end_offset + (1 << SECTOR_SHIFT);
3459                         sg_set_buf(&s[idx - start_index], va + start, end - start);
3460                 }
3461
3462                 sl[i] = s;
3463         }
3464
3465         return sl;
3466 }
3467
3468 static void free_alg(struct alg_spec *a)
3469 {
3470         kfree_sensitive(a->alg_string);
3471         kfree_sensitive(a->key);
3472         memset(a, 0, sizeof *a);
3473 }
3474
3475 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3476 {
3477         char *k;
3478
3479         free_alg(a);
3480
3481         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3482         if (!a->alg_string)
3483                 goto nomem;
3484
3485         k = strchr(a->alg_string, ':');
3486         if (k) {
3487                 *k = 0;
3488                 a->key_string = k + 1;
3489                 if (strlen(a->key_string) & 1)
3490                         goto inval;
3491
3492                 a->key_size = strlen(a->key_string) / 2;
3493                 a->key = kmalloc(a->key_size, GFP_KERNEL);
3494                 if (!a->key)
3495                         goto nomem;
3496                 if (hex2bin(a->key, a->key_string, a->key_size))
3497                         goto inval;
3498         }
3499
3500         return 0;
3501 inval:
3502         *error = error_inval;
3503         return -EINVAL;
3504 nomem:
3505         *error = "Out of memory for an argument";
3506         return -ENOMEM;
3507 }
3508
3509 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3510                    char *error_alg, char *error_key)
3511 {
3512         int r;
3513
3514         if (a->alg_string) {
3515                 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3516                 if (IS_ERR(*hash)) {
3517                         *error = error_alg;
3518                         r = PTR_ERR(*hash);
3519                         *hash = NULL;
3520                         return r;
3521                 }
3522
3523                 if (a->key) {
3524                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
3525                         if (r) {
3526                                 *error = error_key;
3527                                 return r;
3528                         }
3529                 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3530                         *error = error_key;
3531                         return -ENOKEY;
3532                 }
3533         }
3534
3535         return 0;
3536 }
3537
3538 static int create_journal(struct dm_integrity_c *ic, char **error)
3539 {
3540         int r = 0;
3541         unsigned i;
3542         __u64 journal_pages, journal_desc_size, journal_tree_size;
3543         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3544         struct skcipher_request *req = NULL;
3545
3546         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3547         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3548         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3549         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3550
3551         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3552                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3553         journal_desc_size = journal_pages * sizeof(struct page_list);
3554         if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3555                 *error = "Journal doesn't fit into memory";
3556                 r = -ENOMEM;
3557                 goto bad;
3558         }
3559         ic->journal_pages = journal_pages;
3560
3561         ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3562         if (!ic->journal) {
3563                 *error = "Could not allocate memory for journal";
3564                 r = -ENOMEM;
3565                 goto bad;
3566         }
3567         if (ic->journal_crypt_alg.alg_string) {
3568                 unsigned ivsize, blocksize;
3569                 struct journal_completion comp;
3570
3571                 comp.ic = ic;
3572                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3573                 if (IS_ERR(ic->journal_crypt)) {
3574                         *error = "Invalid journal cipher";
3575                         r = PTR_ERR(ic->journal_crypt);
3576                         ic->journal_crypt = NULL;
3577                         goto bad;
3578                 }
3579                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3580                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3581
3582                 if (ic->journal_crypt_alg.key) {
3583                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3584                                                    ic->journal_crypt_alg.key_size);
3585                         if (r) {
3586                                 *error = "Error setting encryption key";
3587                                 goto bad;
3588                         }
3589                 }
3590                 DEBUG_print("cipher %s, block size %u iv size %u\n",
3591                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3592
3593                 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3594                 if (!ic->journal_io) {
3595                         *error = "Could not allocate memory for journal io";
3596                         r = -ENOMEM;
3597                         goto bad;
3598                 }
3599
3600                 if (blocksize == 1) {
3601                         struct scatterlist *sg;
3602
3603                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3604                         if (!req) {
3605                                 *error = "Could not allocate crypt request";
3606                                 r = -ENOMEM;
3607                                 goto bad;
3608                         }
3609
3610                         crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3611                         if (!crypt_iv) {
3612                                 *error = "Could not allocate iv";
3613                                 r = -ENOMEM;
3614                                 goto bad;
3615                         }
3616
3617                         ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3618                         if (!ic->journal_xor) {
3619                                 *error = "Could not allocate memory for journal xor";
3620                                 r = -ENOMEM;
3621                                 goto bad;
3622                         }
3623
3624                         sg = kvmalloc_array(ic->journal_pages + 1,
3625                                             sizeof(struct scatterlist),
3626                                             GFP_KERNEL);
3627                         if (!sg) {
3628                                 *error = "Unable to allocate sg list";
3629                                 r = -ENOMEM;
3630                                 goto bad;
3631                         }
3632                         sg_init_table(sg, ic->journal_pages + 1);
3633                         for (i = 0; i < ic->journal_pages; i++) {
3634                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
3635                                 clear_page(va);
3636                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
3637                         }
3638                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3639
3640                         skcipher_request_set_crypt(req, sg, sg,
3641                                                    PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3642                         init_completion(&comp.comp);
3643                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3644                         if (do_crypt(true, req, &comp))
3645                                 wait_for_completion(&comp.comp);
3646                         kvfree(sg);
3647                         r = dm_integrity_failed(ic);
3648                         if (r) {
3649                                 *error = "Unable to encrypt journal";
3650                                 goto bad;
3651                         }
3652                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3653
3654                         crypto_free_skcipher(ic->journal_crypt);
3655                         ic->journal_crypt = NULL;
3656                 } else {
3657                         unsigned crypt_len = roundup(ivsize, blocksize);
3658
3659                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3660                         if (!req) {
3661                                 *error = "Could not allocate crypt request";
3662                                 r = -ENOMEM;
3663                                 goto bad;
3664                         }
3665
3666                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3667                         if (!crypt_iv) {
3668                                 *error = "Could not allocate iv";
3669                                 r = -ENOMEM;
3670                                 goto bad;
3671                         }
3672
3673                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3674                         if (!crypt_data) {
3675                                 *error = "Unable to allocate crypt data";
3676                                 r = -ENOMEM;
3677                                 goto bad;
3678                         }
3679
3680                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3681                         if (!ic->journal_scatterlist) {
3682                                 *error = "Unable to allocate sg list";
3683                                 r = -ENOMEM;
3684                                 goto bad;
3685                         }
3686                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3687                         if (!ic->journal_io_scatterlist) {
3688                                 *error = "Unable to allocate sg list";
3689                                 r = -ENOMEM;
3690                                 goto bad;
3691                         }
3692                         ic->sk_requests = kvmalloc_array(ic->journal_sections,
3693                                                          sizeof(struct skcipher_request *),
3694                                                          GFP_KERNEL | __GFP_ZERO);
3695                         if (!ic->sk_requests) {
3696                                 *error = "Unable to allocate sk requests";
3697                                 r = -ENOMEM;
3698                                 goto bad;
3699                         }
3700                         for (i = 0; i < ic->journal_sections; i++) {
3701                                 struct scatterlist sg;
3702                                 struct skcipher_request *section_req;
3703                                 __u32 section_le = cpu_to_le32(i);
3704
3705                                 memset(crypt_iv, 0x00, ivsize);
3706                                 memset(crypt_data, 0x00, crypt_len);
3707                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
3708
3709                                 sg_init_one(&sg, crypt_data, crypt_len);
3710                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3711                                 init_completion(&comp.comp);
3712                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3713                                 if (do_crypt(true, req, &comp))
3714                                         wait_for_completion(&comp.comp);
3715
3716                                 r = dm_integrity_failed(ic);
3717                                 if (r) {
3718                                         *error = "Unable to generate iv";
3719                                         goto bad;
3720                                 }
3721
3722                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3723                                 if (!section_req) {
3724                                         *error = "Unable to allocate crypt request";
3725                                         r = -ENOMEM;
3726                                         goto bad;
3727                                 }
3728                                 section_req->iv = kmalloc_array(ivsize, 2,
3729                                                                 GFP_KERNEL);
3730                                 if (!section_req->iv) {
3731                                         skcipher_request_free(section_req);
3732                                         *error = "Unable to allocate iv";
3733                                         r = -ENOMEM;
3734                                         goto bad;
3735                                 }
3736                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3737                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3738                                 ic->sk_requests[i] = section_req;
3739                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3740                         }
3741                 }
3742         }
3743
3744         for (i = 0; i < N_COMMIT_IDS; i++) {
3745                 unsigned j;
3746 retest_commit_id:
3747                 for (j = 0; j < i; j++) {
3748                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
3749                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3750                                 goto retest_commit_id;
3751                         }
3752                 }
3753                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3754         }
3755
3756         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3757         if (journal_tree_size > ULONG_MAX) {
3758                 *error = "Journal doesn't fit into memory";
3759                 r = -ENOMEM;
3760                 goto bad;
3761         }
3762         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3763         if (!ic->journal_tree) {
3764                 *error = "Could not allocate memory for journal tree";
3765                 r = -ENOMEM;
3766         }
3767 bad:
3768         kfree(crypt_data);
3769         kfree(crypt_iv);
3770         skcipher_request_free(req);
3771
3772         return r;
3773 }
3774
3775 /*
3776  * Construct a integrity mapping
3777  *
3778  * Arguments:
3779  *      device
3780  *      offset from the start of the device
3781  *      tag size
3782  *      D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3783  *      number of optional arguments
3784  *      optional arguments:
3785  *              journal_sectors
3786  *              interleave_sectors
3787  *              buffer_sectors
3788  *              journal_watermark
3789  *              commit_time
3790  *              meta_device
3791  *              block_size
3792  *              sectors_per_bit
3793  *              bitmap_flush_interval
3794  *              internal_hash
3795  *              journal_crypt
3796  *              journal_mac
3797  *              recalculate
3798  */
3799 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3800 {
3801         struct dm_integrity_c *ic;
3802         char dummy;
3803         int r;
3804         unsigned extra_args;
3805         struct dm_arg_set as;
3806         static const struct dm_arg _args[] = {
3807                 {0, 16, "Invalid number of feature args"},
3808         };
3809         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3810         bool should_write_sb;
3811         __u64 threshold;
3812         unsigned long long start;
3813         __s8 log2_sectors_per_bitmap_bit = -1;
3814         __s8 log2_blocks_per_bitmap_bit;
3815         __u64 bits_in_journal;
3816         __u64 n_bitmap_bits;
3817
3818 #define DIRECT_ARGUMENTS        4
3819
3820         if (argc <= DIRECT_ARGUMENTS) {
3821                 ti->error = "Invalid argument count";
3822                 return -EINVAL;
3823         }
3824
3825         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3826         if (!ic) {
3827                 ti->error = "Cannot allocate integrity context";
3828                 return -ENOMEM;
3829         }
3830         ti->private = ic;
3831         ti->per_io_data_size = sizeof(struct dm_integrity_io);
3832         ic->ti = ti;
3833
3834         ic->in_progress = RB_ROOT;
3835         INIT_LIST_HEAD(&ic->wait_list);
3836         init_waitqueue_head(&ic->endio_wait);
3837         bio_list_init(&ic->flush_bio_list);
3838         init_waitqueue_head(&ic->copy_to_journal_wait);
3839         init_completion(&ic->crypto_backoff);
3840         atomic64_set(&ic->number_of_mismatches, 0);
3841         ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
3842
3843         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3844         if (r) {
3845                 ti->error = "Device lookup failed";
3846                 goto bad;
3847         }
3848
3849         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3850                 ti->error = "Invalid starting offset";
3851                 r = -EINVAL;
3852                 goto bad;
3853         }
3854         ic->start = start;
3855
3856         if (strcmp(argv[2], "-")) {
3857                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3858                         ti->error = "Invalid tag size";
3859                         r = -EINVAL;
3860                         goto bad;
3861                 }
3862         }
3863
3864         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
3865             !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
3866                 ic->mode = argv[3][0];
3867         } else {
3868                 ti->error = "Invalid mode (expecting J, B, D, R)";
3869                 r = -EINVAL;
3870                 goto bad;
3871         }
3872
3873         journal_sectors = 0;
3874         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3875         buffer_sectors = DEFAULT_BUFFER_SECTORS;
3876         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3877         sync_msec = DEFAULT_SYNC_MSEC;
3878         ic->sectors_per_block = 1;
3879
3880         as.argc = argc - DIRECT_ARGUMENTS;
3881         as.argv = argv + DIRECT_ARGUMENTS;
3882         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3883         if (r)
3884                 goto bad;
3885
3886         while (extra_args--) {
3887                 const char *opt_string;
3888                 unsigned val;
3889                 unsigned long long llval;
3890                 opt_string = dm_shift_arg(&as);
3891                 if (!opt_string) {
3892                         r = -EINVAL;
3893                         ti->error = "Not enough feature arguments";
3894                         goto bad;
3895                 }
3896                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3897                         journal_sectors = val ? val : 1;
3898                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3899                         interleave_sectors = val;
3900                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3901                         buffer_sectors = val;
3902                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3903                         journal_watermark = val;
3904                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3905                         sync_msec = val;
3906                 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3907                         if (ic->meta_dev) {
3908                                 dm_put_device(ti, ic->meta_dev);
3909                                 ic->meta_dev = NULL;
3910                         }
3911                         r = dm_get_device(ti, strchr(opt_string, ':') + 1,
3912                                           dm_table_get_mode(ti->table), &ic->meta_dev);
3913                         if (r) {
3914                                 ti->error = "Device lookup failed";
3915                                 goto bad;
3916                         }
3917                 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3918                         if (val < 1 << SECTOR_SHIFT ||
3919                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3920                             (val & (val -1))) {
3921                                 r = -EINVAL;
3922                                 ti->error = "Invalid block_size argument";
3923                                 goto bad;
3924                         }
3925                         ic->sectors_per_block = val >> SECTOR_SHIFT;
3926                 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
3927                         log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
3928                 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
3929                         if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
3930                                 r = -EINVAL;
3931                                 ti->error = "Invalid bitmap_flush_interval argument";
3932                         }
3933                         ic->bitmap_flush_interval = msecs_to_jiffies(val);
3934                 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3935                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3936                                             "Invalid internal_hash argument");
3937                         if (r)
3938                                 goto bad;
3939                 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3940                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3941                                             "Invalid journal_crypt argument");
3942                         if (r)
3943                                 goto bad;
3944                 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3945                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
3946                                             "Invalid journal_mac argument");
3947                         if (r)
3948                                 goto bad;
3949                 } else if (!strcmp(opt_string, "recalculate")) {
3950                         ic->recalculate_flag = true;
3951                 } else if (!strcmp(opt_string, "allow_discards")) {
3952                         ic->discard = true;
3953                 } else if (!strcmp(opt_string, "fix_padding")) {
3954                         ic->fix_padding = true;
3955                 } else if (!strcmp(opt_string, "legacy_recalculate")) {
3956                         ic->legacy_recalculate = true;
3957                 } else {
3958                         r = -EINVAL;
3959                         ti->error = "Invalid argument";
3960                         goto bad;
3961                 }
3962         }
3963
3964         ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3965         if (!ic->meta_dev)
3966                 ic->meta_device_sectors = ic->data_device_sectors;
3967         else
3968                 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3969
3970         if (!journal_sectors) {
3971                 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3972                                       ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3973         }
3974
3975         if (!buffer_sectors)
3976                 buffer_sectors = 1;
3977         ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3978
3979         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3980                     "Invalid internal hash", "Error setting internal hash key");
3981         if (r)
3982                 goto bad;
3983
3984         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3985                     "Invalid journal mac", "Error setting journal mac key");
3986         if (r)
3987                 goto bad;
3988
3989         if (!ic->tag_size) {
3990                 if (!ic->internal_hash) {
3991                         ti->error = "Unknown tag size";
3992                         r = -EINVAL;
3993                         goto bad;
3994                 }
3995                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
3996         }
3997         if (ic->tag_size > MAX_TAG_SIZE) {
3998                 ti->error = "Too big tag size";
3999                 r = -EINVAL;
4000                 goto bad;
4001         }
4002         if (!(ic->tag_size & (ic->tag_size - 1)))
4003                 ic->log2_tag_size = __ffs(ic->tag_size);
4004         else
4005                 ic->log2_tag_size = -1;
4006
4007         if (ic->mode == 'B' && !ic->internal_hash) {
4008                 r = -EINVAL;
4009                 ti->error = "Bitmap mode can be only used with internal hash";
4010                 goto bad;
4011         }
4012
4013         if (ic->discard && !ic->internal_hash) {
4014                 r = -EINVAL;
4015                 ti->error = "Discard can be only used with internal hash";
4016                 goto bad;
4017         }
4018
4019         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4020         ic->autocommit_msec = sync_msec;
4021         timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4022
4023         ic->io = dm_io_client_create();
4024         if (IS_ERR(ic->io)) {
4025                 r = PTR_ERR(ic->io);
4026                 ic->io = NULL;
4027                 ti->error = "Cannot allocate dm io";
4028                 goto bad;
4029         }
4030
4031         r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4032         if (r) {
4033                 ti->error = "Cannot allocate mempool";
4034                 goto bad;
4035         }
4036
4037         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4038                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4039         if (!ic->metadata_wq) {
4040                 ti->error = "Cannot allocate workqueue";
4041                 r = -ENOMEM;
4042                 goto bad;
4043         }
4044
4045         /*
4046          * If this workqueue were percpu, it would cause bio reordering
4047          * and reduced performance.
4048          */
4049         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4050         if (!ic->wait_wq) {
4051                 ti->error = "Cannot allocate workqueue";
4052                 r = -ENOMEM;
4053                 goto bad;
4054         }
4055
4056         ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4057                                           METADATA_WORKQUEUE_MAX_ACTIVE);
4058         if (!ic->offload_wq) {
4059                 ti->error = "Cannot allocate workqueue";
4060                 r = -ENOMEM;
4061                 goto bad;
4062         }
4063
4064         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4065         if (!ic->commit_wq) {
4066                 ti->error = "Cannot allocate workqueue";
4067                 r = -ENOMEM;
4068                 goto bad;
4069         }
4070         INIT_WORK(&ic->commit_work, integrity_commit);
4071
4072         if (ic->mode == 'J' || ic->mode == 'B') {
4073                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4074                 if (!ic->writer_wq) {
4075                         ti->error = "Cannot allocate workqueue";
4076                         r = -ENOMEM;
4077                         goto bad;
4078                 }
4079                 INIT_WORK(&ic->writer_work, integrity_writer);
4080         }
4081
4082         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4083         if (!ic->sb) {
4084                 r = -ENOMEM;
4085                 ti->error = "Cannot allocate superblock area";
4086                 goto bad;
4087         }
4088
4089         r = sync_rw_sb(ic, REQ_OP_READ, 0);
4090         if (r) {
4091                 ti->error = "Error reading superblock";
4092                 goto bad;
4093         }
4094         should_write_sb = false;
4095         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4096                 if (ic->mode != 'R') {
4097                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4098                                 r = -EINVAL;
4099                                 ti->error = "The device is not initialized";
4100                                 goto bad;
4101                         }
4102                 }
4103
4104                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4105                 if (r) {
4106                         ti->error = "Could not initialize superblock";
4107                         goto bad;
4108                 }
4109                 if (ic->mode != 'R')
4110                         should_write_sb = true;
4111         }
4112
4113         if (!ic->sb->version || ic->sb->version > SB_VERSION_4) {
4114                 r = -EINVAL;
4115                 ti->error = "Unknown version";
4116                 goto bad;
4117         }
4118         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4119                 r = -EINVAL;
4120                 ti->error = "Tag size doesn't match the information in superblock";
4121                 goto bad;
4122         }
4123         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4124                 r = -EINVAL;
4125                 ti->error = "Block size doesn't match the information in superblock";
4126                 goto bad;
4127         }
4128         if (!le32_to_cpu(ic->sb->journal_sections)) {
4129                 r = -EINVAL;
4130                 ti->error = "Corrupted superblock, journal_sections is 0";
4131                 goto bad;
4132         }
4133         /* make sure that ti->max_io_len doesn't overflow */
4134         if (!ic->meta_dev) {
4135                 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4136                     ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4137                         r = -EINVAL;
4138                         ti->error = "Invalid interleave_sectors in the superblock";
4139                         goto bad;
4140                 }
4141         } else {
4142                 if (ic->sb->log2_interleave_sectors) {
4143                         r = -EINVAL;
4144                         ti->error = "Invalid interleave_sectors in the superblock";
4145                         goto bad;
4146                 }
4147         }
4148         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4149                 r = -EINVAL;
4150                 ti->error = "Journal mac mismatch";
4151                 goto bad;
4152         }
4153
4154         get_provided_data_sectors(ic);
4155         if (!ic->provided_data_sectors) {
4156                 r = -EINVAL;
4157                 ti->error = "The device is too small";
4158                 goto bad;
4159         }
4160
4161 try_smaller_buffer:
4162         r = calculate_device_limits(ic);
4163         if (r) {
4164                 if (ic->meta_dev) {
4165                         if (ic->log2_buffer_sectors > 3) {
4166                                 ic->log2_buffer_sectors--;
4167                                 goto try_smaller_buffer;
4168                         }
4169                 }
4170                 ti->error = "The device is too small";
4171                 goto bad;
4172         }
4173
4174         if (log2_sectors_per_bitmap_bit < 0)
4175                 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4176         if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4177                 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4178
4179         bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4180         if (bits_in_journal > UINT_MAX)
4181                 bits_in_journal = UINT_MAX;
4182         while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4183                 log2_sectors_per_bitmap_bit++;
4184
4185         log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4186         ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4187         if (should_write_sb) {
4188                 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4189         }
4190         n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4191                                 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4192         ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4193
4194         if (!ic->meta_dev)
4195                 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4196
4197         if (ti->len > ic->provided_data_sectors) {
4198                 r = -EINVAL;
4199                 ti->error = "Not enough provided sectors for requested mapping size";
4200                 goto bad;
4201         }
4202
4203
4204         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4205         threshold += 50;
4206         do_div(threshold, 100);
4207         ic->free_sectors_threshold = threshold;
4208
4209         DEBUG_print("initialized:\n");
4210         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4211         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
4212         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4213         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
4214         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
4215         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
4216         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
4217         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4218         DEBUG_print("   data_device_sectors 0x%llx\n", i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT);
4219         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
4220         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
4221         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
4222         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4223         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4224         DEBUG_print("   bits_in_journal %llu\n", bits_in_journal);
4225
4226         if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4227                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4228                 ic->sb->recalc_sector = cpu_to_le64(0);
4229         }
4230
4231         if (ic->internal_hash) {
4232                 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4233                 if (!ic->recalc_wq ) {
4234                         ti->error = "Cannot allocate workqueue";
4235                         r = -ENOMEM;
4236                         goto bad;
4237                 }
4238                 INIT_WORK(&ic->recalc_work, integrity_recalc);
4239                 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4240                 if (!ic->recalc_buffer) {
4241                         ti->error = "Cannot allocate buffer for recalculating";
4242                         r = -ENOMEM;
4243                         goto bad;
4244                 }
4245                 ic->recalc_tags = kvmalloc_array(RECALC_SECTORS >> ic->sb->log2_sectors_per_block,
4246                                                  ic->tag_size, GFP_KERNEL);
4247                 if (!ic->recalc_tags) {
4248                         ti->error = "Cannot allocate tags for recalculating";
4249                         r = -ENOMEM;
4250                         goto bad;
4251                 }
4252         } else {
4253                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4254                         ti->error = "Recalculate can only be specified with internal_hash";
4255                         r = -EINVAL;
4256                         goto bad;
4257                 }
4258         }
4259
4260         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4261             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4262             dm_integrity_disable_recalculate(ic)) {
4263                 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4264                 r = -EOPNOTSUPP;
4265                 goto bad;
4266         }
4267
4268         ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4269                         1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
4270         if (IS_ERR(ic->bufio)) {
4271                 r = PTR_ERR(ic->bufio);
4272                 ti->error = "Cannot initialize dm-bufio";
4273                 ic->bufio = NULL;
4274                 goto bad;
4275         }
4276         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4277
4278         if (ic->mode != 'R') {
4279                 r = create_journal(ic, &ti->error);
4280                 if (r)
4281                         goto bad;
4282
4283         }
4284
4285         if (ic->mode == 'B') {
4286                 unsigned i;
4287                 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4288
4289                 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4290                 if (!ic->recalc_bitmap) {
4291                         r = -ENOMEM;
4292                         goto bad;
4293                 }
4294                 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4295                 if (!ic->may_write_bitmap) {
4296                         r = -ENOMEM;
4297                         goto bad;
4298                 }
4299                 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4300                 if (!ic->bbs) {
4301                         r = -ENOMEM;
4302                         goto bad;
4303                 }
4304                 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4305                 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4306                         struct bitmap_block_status *bbs = &ic->bbs[i];
4307                         unsigned sector, pl_index, pl_offset;
4308
4309                         INIT_WORK(&bbs->work, bitmap_block_work);
4310                         bbs->ic = ic;
4311                         bbs->idx = i;
4312                         bio_list_init(&bbs->bio_queue);
4313                         spin_lock_init(&bbs->bio_queue_lock);
4314
4315                         sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4316                         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4317                         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4318
4319                         bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4320                 }
4321         }
4322
4323         if (should_write_sb) {
4324                 int r;
4325
4326                 init_journal(ic, 0, ic->journal_sections, 0);
4327                 r = dm_integrity_failed(ic);
4328                 if (unlikely(r)) {
4329                         ti->error = "Error initializing journal";
4330                         goto bad;
4331                 }
4332                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4333                 if (r) {
4334                         ti->error = "Error initializing superblock";
4335                         goto bad;
4336                 }
4337                 ic->just_formatted = true;
4338         }
4339
4340         if (!ic->meta_dev) {
4341                 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4342                 if (r)
4343                         goto bad;
4344         }
4345         if (ic->mode == 'B') {
4346                 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4347                 if (!max_io_len)
4348                         max_io_len = 1U << 31;
4349                 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4350                 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4351                         r = dm_set_target_max_io_len(ti, max_io_len);
4352                         if (r)
4353                                 goto bad;
4354                 }
4355         }
4356
4357         if (!ic->internal_hash)
4358                 dm_integrity_set(ti, ic);
4359
4360         ti->num_flush_bios = 1;
4361         ti->flush_supported = true;
4362         if (ic->discard)
4363                 ti->num_discard_bios = 1;
4364
4365         return 0;
4366
4367 bad:
4368         dm_integrity_dtr(ti);
4369         return r;
4370 }
4371
4372 static void dm_integrity_dtr(struct dm_target *ti)
4373 {
4374         struct dm_integrity_c *ic = ti->private;
4375
4376         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4377         BUG_ON(!list_empty(&ic->wait_list));
4378
4379         if (ic->metadata_wq)
4380                 destroy_workqueue(ic->metadata_wq);
4381         if (ic->wait_wq)
4382                 destroy_workqueue(ic->wait_wq);
4383         if (ic->offload_wq)
4384                 destroy_workqueue(ic->offload_wq);
4385         if (ic->commit_wq)
4386                 destroy_workqueue(ic->commit_wq);
4387         if (ic->writer_wq)
4388                 destroy_workqueue(ic->writer_wq);
4389         if (ic->recalc_wq)
4390                 destroy_workqueue(ic->recalc_wq);
4391         vfree(ic->recalc_buffer);
4392         kvfree(ic->recalc_tags);
4393         kvfree(ic->bbs);
4394         if (ic->bufio)
4395                 dm_bufio_client_destroy(ic->bufio);
4396         mempool_exit(&ic->journal_io_mempool);
4397         if (ic->io)
4398                 dm_io_client_destroy(ic->io);
4399         if (ic->dev)
4400                 dm_put_device(ti, ic->dev);
4401         if (ic->meta_dev)
4402                 dm_put_device(ti, ic->meta_dev);
4403         dm_integrity_free_page_list(ic->journal);
4404         dm_integrity_free_page_list(ic->journal_io);
4405         dm_integrity_free_page_list(ic->journal_xor);
4406         dm_integrity_free_page_list(ic->recalc_bitmap);
4407         dm_integrity_free_page_list(ic->may_write_bitmap);
4408         if (ic->journal_scatterlist)
4409                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4410         if (ic->journal_io_scatterlist)
4411                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4412         if (ic->sk_requests) {
4413                 unsigned i;
4414
4415                 for (i = 0; i < ic->journal_sections; i++) {
4416                         struct skcipher_request *req = ic->sk_requests[i];
4417                         if (req) {
4418                                 kfree_sensitive(req->iv);
4419                                 skcipher_request_free(req);
4420                         }
4421                 }
4422                 kvfree(ic->sk_requests);
4423         }
4424         kvfree(ic->journal_tree);
4425         if (ic->sb)
4426                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4427
4428         if (ic->internal_hash)
4429                 crypto_free_shash(ic->internal_hash);
4430         free_alg(&ic->internal_hash_alg);
4431
4432         if (ic->journal_crypt)
4433                 crypto_free_skcipher(ic->journal_crypt);
4434         free_alg(&ic->journal_crypt_alg);
4435
4436         if (ic->journal_mac)
4437                 crypto_free_shash(ic->journal_mac);
4438         free_alg(&ic->journal_mac_alg);
4439
4440         kfree(ic);
4441 }
4442
4443 static struct target_type integrity_target = {
4444         .name                   = "integrity",
4445         .version                = {1, 6, 0},
4446         .module                 = THIS_MODULE,
4447         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4448         .ctr                    = dm_integrity_ctr,
4449         .dtr                    = dm_integrity_dtr,
4450         .map                    = dm_integrity_map,
4451         .postsuspend            = dm_integrity_postsuspend,
4452         .resume                 = dm_integrity_resume,
4453         .status                 = dm_integrity_status,
4454         .iterate_devices        = dm_integrity_iterate_devices,
4455         .io_hints               = dm_integrity_io_hints,
4456 };
4457
4458 static int __init dm_integrity_init(void)
4459 {
4460         int r;
4461
4462         journal_io_cache = kmem_cache_create("integrity_journal_io",
4463                                              sizeof(struct journal_io), 0, 0, NULL);
4464         if (!journal_io_cache) {
4465                 DMERR("can't allocate journal io cache");
4466                 return -ENOMEM;
4467         }
4468
4469         r = dm_register_target(&integrity_target);
4470
4471         if (r < 0)
4472                 DMERR("register failed %d", r);
4473
4474         return r;
4475 }
4476
4477 static void __exit dm_integrity_exit(void)
4478 {
4479         dm_unregister_target(&integrity_target);
4480         kmem_cache_destroy(journal_io_cache);
4481 }
4482
4483 module_init(dm_integrity_init);
4484 module_exit(dm_integrity_exit);
4485
4486 MODULE_AUTHOR("Milan Broz");
4487 MODULE_AUTHOR("Mikulas Patocka");
4488 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4489 MODULE_LICENSE("GPL");