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