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