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