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_VERSION_4                    4
  57 #define SB_SECTORS                      8
  58 #define MAX_SECTORS_PER_BLOCK           8
  59
  60 struct superblock {
  61         __u8 magic[8];
  62         __u8 version;
  63         __u8 log2_interleave_sectors;
  64         __u16 integrity_tag_size;
  65         __u32 journal_sections;
  66         __u64 provided_data_sectors;    /* userspace uses this value */
  67         __u32 flags;
  68         __u8 log2_sectors_per_block;
  69         __u8 log2_blocks_per_bitmap_bit;
  70         __u8 pad[2];
  71         __u64 recalc_sector;
  72 };
  73
  74 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  75 #define SB_FLAG_RECALCULATING           0x2
  76 #define SB_FLAG_DIRTY_BITMAP            0x4
  77 #define SB_FLAG_FIXED_PADDING           0x8
  78
  79 #define JOURNAL_ENTRY_ROUNDUP           8
  80
  81 typedef __u64 commit_id_t;
  82 #define JOURNAL_MAC_PER_SECTOR          8
  83
  84 struct journal_entry {
  85         union {
  86                 struct {
  87                         __u32 sector_lo;
  88                         __u32 sector_hi;
  89                 } s;
  90                 __u64 sector;
  91         } u;
  92         commit_id_t last_bytes[0];
  93         /* __u8 tag[0]; */
  94 };
  95
  96 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
  97
  98 #if BITS_PER_LONG == 64
  99 #define journal_entry_set_sector(je, x)         do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
 100 #else
 101 #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)
 102 #endif
 103 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
 104 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
 105 #define journal_entry_set_unused(je)            do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
 106 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
 107 #define journal_entry_set_inprogress(je)        do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
 108
 109 #define JOURNAL_BLOCK_SECTORS           8
 110 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 111 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 112
 113 struct journal_sector {
 114         __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 115         __u8 mac[JOURNAL_MAC_PER_SECTOR];
 116         commit_id_t commit_id;
 117 };
 118
 119 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 120
 121 #define METADATA_PADDING_SECTORS        8
 122
 123 #define N_COMMIT_IDS                    4
 124
 125 static unsigned char prev_commit_seq(unsigned char seq)
 126 {
 127         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 128 }
 129
 130 static unsigned char next_commit_seq(unsigned char seq)
 131 {
 132         return (seq + 1) % N_COMMIT_IDS;
 133 }
 134
 135 /*
 136  * In-memory structures
 137  */
 138
 139 struct journal_node {
 140         struct rb_node node;
 141         sector_t sector;
 142 };
 143
 144 struct alg_spec {
 145         char *alg_string;
 146         char *key_string;
 147         __u8 *key;
 148         unsigned key_size;
 149 };
 150
 151 struct dm_integrity_c {
 152         struct dm_dev *dev;
 153         struct dm_dev *meta_dev;
 154         unsigned tag_size;
 155         __s8 log2_tag_size;
 156         sector_t start;
 157         mempool_t journal_io_mempool;
 158         struct dm_io_client *io;
 159         struct dm_bufio_client *bufio;
 160         struct workqueue_struct *metadata_wq;
 161         struct superblock *sb;
 162         unsigned journal_pages;
 163         unsigned n_bitmap_blocks;
 164
 165         struct page_list *journal;
 166         struct page_list *journal_io;
 167         struct page_list *journal_xor;
 168         struct page_list *recalc_bitmap;
 169         struct page_list *may_write_bitmap;
 170         struct bitmap_block_status *bbs;
 171         unsigned bitmap_flush_interval;
 172         int synchronous_mode;
 173         struct bio_list synchronous_bios;
 174         struct delayed_work bitmap_flush_work;
 175
 176         struct crypto_skcipher *journal_crypt;
 177         struct scatterlist **journal_scatterlist;
 178         struct scatterlist **journal_io_scatterlist;
 179         struct skcipher_request **sk_requests;
 180
 181         struct crypto_shash *journal_mac;
 182
 183         struct journal_node *journal_tree;
 184         struct rb_root journal_tree_root;
 185
 186         sector_t provided_data_sectors;
 187
 188         unsigned short journal_entry_size;
 189         unsigned char journal_entries_per_sector;
 190         unsigned char journal_section_entries;
 191         unsigned short journal_section_sectors;
 192         unsigned journal_sections;
 193         unsigned journal_entries;
 194         sector_t data_device_sectors;
 195         sector_t meta_device_sectors;
 196         unsigned initial_sectors;
 197         unsigned metadata_run;
 198         __s8 log2_metadata_run;
 199         __u8 log2_buffer_sectors;
 200         __u8 sectors_per_block;
 201         __u8 log2_blocks_per_bitmap_bit;
 202
 203         unsigned char mode;
 204         int suspending;
 205
 206         int failed;
 207
 208         struct crypto_shash *internal_hash;
 209
 210         /* these variables are locked with endio_wait.lock */
 211         struct rb_root in_progress;
 212         struct list_head wait_list;
 213         wait_queue_head_t endio_wait;
 214         struct workqueue_struct *wait_wq;
 215
 216         unsigned char commit_seq;
 217         commit_id_t commit_ids[N_COMMIT_IDS];
 218
 219         unsigned committed_section;
 220         unsigned n_committed_sections;
 221
 222         unsigned uncommitted_section;
 223         unsigned n_uncommitted_sections;
 224
 225         unsigned free_section;
 226         unsigned char free_section_entry;
 227         unsigned free_sectors;
 228
 229         unsigned free_sectors_threshold;
 230
 231         struct workqueue_struct *commit_wq;
 232         struct work_struct commit_work;
 233
 234         struct workqueue_struct *writer_wq;
 235         struct work_struct writer_work;
 236
 237         struct workqueue_struct *recalc_wq;
 238         struct work_struct recalc_work;
 239         u8 *recalc_buffer;
 240         u8 *recalc_tags;
 241
 242         struct bio_list flush_bio_list;
 243
 244         unsigned long autocommit_jiffies;
 245         struct timer_list autocommit_timer;
 246         unsigned autocommit_msec;
 247
 248         wait_queue_head_t copy_to_journal_wait;
 249
 250         struct completion crypto_backoff;
 251
 252         bool journal_uptodate;
 253         bool just_formatted;
 254         bool recalculate_flag;
 255         bool fix_padding;
 256
 257         struct alg_spec internal_hash_alg;
 258         struct alg_spec journal_crypt_alg;
 259         struct alg_spec journal_mac_alg;
 260
 261         atomic64_t number_of_mismatches;
 262
 263         struct notifier_block reboot_notifier;
 264 };
 265
 266 struct dm_integrity_range {
 267         sector_t logical_sector;
 268         sector_t n_sectors;
 269         bool waiting;
 270         union {
 271                 struct rb_node node;
 272                 struct {
 273                         struct task_struct *task;
 274                         struct list_head wait_entry;
 275                 };
 276         };
 277 };
 278
 279 struct dm_integrity_io {
 280         struct work_struct work;
 281
 282         struct dm_integrity_c *ic;
 283         bool write;
 284         bool fua;
 285
 286         struct dm_integrity_range range;
 287
 288         sector_t metadata_block;
 289         unsigned metadata_offset;
 290
 291         atomic_t in_flight;
 292         blk_status_t bi_status;
 293
 294         struct completion *completion;
 295
 296         struct gendisk *orig_bi_disk;
 297         u8 orig_bi_partno;
 298         bio_end_io_t *orig_bi_end_io;
 299         struct bio_integrity_payload *orig_bi_integrity;
 300         struct bvec_iter orig_bi_iter;
 301 };
 302
 303 struct journal_completion {
 304         struct dm_integrity_c *ic;
 305         atomic_t in_flight;
 306         struct completion comp;
 307 };
 308
 309 struct journal_io {
 310         struct dm_integrity_range range;
 311         struct journal_completion *comp;
 312 };
 313
 314 struct bitmap_block_status {
 315         struct work_struct work;
 316         struct dm_integrity_c *ic;
 317         unsigned idx;
 318         unsigned long *bitmap;
 319         struct bio_list bio_queue;
 320         spinlock_t bio_queue_lock;
 321
 322 };
 323
 324 static struct kmem_cache *journal_io_cache;
 325
 326 #define JOURNAL_IO_MEMPOOL      32
 327
 328 #ifdef DEBUG_PRINT
 329 #define DEBUG_print(x, ...)     printk(KERN_DEBUG x, ##__VA_ARGS__)
 330 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 331 {
 332         va_list args;
 333         va_start(args, msg);
 334         vprintk(msg, args);
 335         va_end(args);
 336         if (len)
 337                 pr_cont(":");
 338         while (len) {
 339                 pr_cont(" %02x", *bytes);
 340                 bytes++;
 341                 len--;
 342         }
 343         pr_cont("\n");
 344 }
 345 #define DEBUG_bytes(bytes, len, msg, ...)       __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 346 #else
 347 #define DEBUG_print(x, ...)                     do { } while (0)
 348 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
 349 #endif
 350
 351 static void dm_integrity_prepare(struct request *rq)
 352 {
 353 }
 354
 355 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
 356 {
 357 }
 358
 359 /*
 360  * DM Integrity profile, protection is performed layer above (dm-crypt)
 361  */
 362 static const struct blk_integrity_profile dm_integrity_profile = {
 363         .name                   = "DM-DIF-EXT-TAG",
 364         .generate_fn            = NULL,
 365         .verify_fn              = NULL,
 366         .prepare_fn             = dm_integrity_prepare,
 367         .complete_fn            = dm_integrity_complete,
 368 };
 369
 370 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 371 static void integrity_bio_wait(struct work_struct *w);
 372 static void dm_integrity_dtr(struct dm_target *ti);
 373
 374 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 375 {
 376         if (err == -EILSEQ)
 377                 atomic64_inc(&ic->number_of_mismatches);
 378         if (!cmpxchg(&ic->failed, 0, err))
 379                 DMERR("Error on %s: %d", msg, err);
 380 }
 381
 382 static int dm_integrity_failed(struct dm_integrity_c *ic)
 383 {
 384         return READ_ONCE(ic->failed);
 385 }
 386
 387 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 388                                           unsigned j, unsigned char seq)
 389 {
 390         /*
 391          * Xor the number with section and sector, so that if a piece of
 392          * journal is written at wrong place, it is detected.
 393          */
 394         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 395 }
 396
 397 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 398                                 sector_t *area, sector_t *offset)
 399 {
 400         if (!ic->meta_dev) {
 401                 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 402                 *area = data_sector >> log2_interleave_sectors;
 403                 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 404         } else {
 405                 *area = 0;
 406                 *offset = data_sector;
 407         }
 408 }
 409
 410 #define sector_to_block(ic, n)                                          \
 411 do {                                                                    \
 412         BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));          \
 413         (n) >>= (ic)->sb->log2_sectors_per_block;                       \
 414 } while (0)
 415
 416 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 417                                             sector_t offset, unsigned *metadata_offset)
 418 {
 419         __u64 ms;
 420         unsigned mo;
 421
 422         ms = area << ic->sb->log2_interleave_sectors;
 423         if (likely(ic->log2_metadata_run >= 0))
 424                 ms += area << ic->log2_metadata_run;
 425         else
 426                 ms += area * ic->metadata_run;
 427         ms >>= ic->log2_buffer_sectors;
 428
 429         sector_to_block(ic, offset);
 430
 431         if (likely(ic->log2_tag_size >= 0)) {
 432                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 433                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 434         } else {
 435                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 436                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 437         }
 438         *metadata_offset = mo;
 439         return ms;
 440 }
 441
 442 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 443 {
 444         sector_t result;
 445
 446         if (ic->meta_dev)
 447                 return offset;
 448
 449         result = area << ic->sb->log2_interleave_sectors;
 450         if (likely(ic->log2_metadata_run >= 0))
 451                 result += (area + 1) << ic->log2_metadata_run;
 452         else
 453                 result += (area + 1) * ic->metadata_run;
 454
 455         result += (sector_t)ic->initial_sectors + offset;
 456         result += ic->start;
 457
 458         return result;
 459 }
 460
 461 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 462 {
 463         if (unlikely(*sec_ptr >= ic->journal_sections))
 464                 *sec_ptr -= ic->journal_sections;
 465 }
 466
 467 static void sb_set_version(struct dm_integrity_c *ic)
 468 {
 469         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
 470                 ic->sb->version = SB_VERSION_4;
 471         else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
 472                 ic->sb->version = SB_VERSION_3;
 473         else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
 474                 ic->sb->version = SB_VERSION_2;
 475         else
 476                 ic->sb->version = SB_VERSION_1;
 477 }
 478
 479 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
 480 {
 481         struct dm_io_request io_req;
 482         struct dm_io_region io_loc;
 483
 484         io_req.bi_op = op;
 485         io_req.bi_op_flags = op_flags;
 486         io_req.mem.type = DM_IO_KMEM;
 487         io_req.mem.ptr.addr = ic->sb;
 488         io_req.notify.fn = NULL;
 489         io_req.client = ic->io;
 490         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 491         io_loc.sector = ic->start;
 492         io_loc.count = SB_SECTORS;
 493
 494         if (op == REQ_OP_WRITE)
 495                 sb_set_version(ic);
 496
 497         return dm_io(&io_req, 1, &io_loc, NULL);
 498 }
 499
 500 #define BITMAP_OP_TEST_ALL_SET          0
 501 #define BITMAP_OP_TEST_ALL_CLEAR        1
 502 #define BITMAP_OP_SET                   2
 503 #define BITMAP_OP_CLEAR                 3
 504
 505 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
 506                             sector_t sector, sector_t n_sectors, int mode)
 507 {
 508         unsigned long bit, end_bit, this_end_bit, page, end_page;
 509         unsigned long *data;
 510
 511         if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
 512                 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
 513                         (unsigned long long)sector,
 514                         (unsigned long long)n_sectors,
 515                         ic->sb->log2_sectors_per_block,
 516                         ic->log2_blocks_per_bitmap_bit,
 517                         mode);
 518                 BUG();
 519         }
 520
 521         if (unlikely(!n_sectors))
 522                 return true;
 523
 524         bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 525         end_bit = (sector + n_sectors - 1) >>
 526                 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 527
 528         page = bit / (PAGE_SIZE * 8);
 529         bit %= PAGE_SIZE * 8;
 530
 531         end_page = end_bit / (PAGE_SIZE * 8);
 532         end_bit %= PAGE_SIZE * 8;
 533
 534 repeat:
 535         if (page < end_page) {
 536                 this_end_bit = PAGE_SIZE * 8 - 1;
 537         } else {
 538                 this_end_bit = end_bit;
 539         }
 540
 541         data = lowmem_page_address(bitmap[page].page);
 542
 543         if (mode == BITMAP_OP_TEST_ALL_SET) {
 544                 while (bit <= this_end_bit) {
 545                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 546                                 do {
 547                                         if (data[bit / BITS_PER_LONG] != -1)
 548                                                 return false;
 549                                         bit += BITS_PER_LONG;
 550                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 551                                 continue;
 552                         }
 553                         if (!test_bit(bit, data))
 554                                 return false;
 555                         bit++;
 556                 }
 557         } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
 558                 while (bit <= this_end_bit) {
 559                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 560                                 do {
 561                                         if (data[bit / BITS_PER_LONG] != 0)
 562                                                 return false;
 563                                         bit += BITS_PER_LONG;
 564                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 565                                 continue;
 566                         }
 567                         if (test_bit(bit, data))
 568                                 return false;
 569                         bit++;
 570                 }
 571         } else if (mode == BITMAP_OP_SET) {
 572                 while (bit <= this_end_bit) {
 573                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 574                                 do {
 575                                         data[bit / BITS_PER_LONG] = -1;
 576                                         bit += BITS_PER_LONG;
 577                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 578                                 continue;
 579                         }
 580                         __set_bit(bit, data);
 581                         bit++;
 582                 }
 583         } else if (mode == BITMAP_OP_CLEAR) {
 584                 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
 585                         clear_page(data);
 586                 else while (bit <= this_end_bit) {
 587                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 588                                 do {
 589                                         data[bit / BITS_PER_LONG] = 0;
 590                                         bit += BITS_PER_LONG;
 591                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 592                                 continue;
 593                         }
 594                         __clear_bit(bit, data);
 595                         bit++;
 596                 }
 597         } else {
 598                 BUG();
 599         }
 600
 601         if (unlikely(page < end_page)) {
 602                 bit = 0;
 603                 page++;
 604                 goto repeat;
 605         }
 606
 607         return true;
 608 }
 609
 610 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
 611 {
 612         unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
 613         unsigned i;
 614
 615         for (i = 0; i < n_bitmap_pages; i++) {
 616                 unsigned long *dst_data = lowmem_page_address(dst[i].page);
 617                 unsigned long *src_data = lowmem_page_address(src[i].page);
 618                 copy_page(dst_data, src_data);
 619         }
 620 }
 621
 622 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
 623 {
 624         unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 625         unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
 626
 627         BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
 628         return &ic->bbs[bitmap_block];
 629 }
 630
 631 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 632                                  bool e, const char *function)
 633 {
 634 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 635         unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 636
 637         if (unlikely(section >= ic->journal_sections) ||
 638             unlikely(offset >= limit)) {
 639                 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
 640                        function, section, offset, ic->journal_sections, limit);
 641                 BUG();
 642         }
 643 #endif
 644 }
 645
 646 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 647                                unsigned *pl_index, unsigned *pl_offset)
 648 {
 649         unsigned sector;
 650
 651         access_journal_check(ic, section, offset, false, "page_list_location");
 652
 653         sector = section * ic->journal_section_sectors + offset;
 654
 655         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 656         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 657 }
 658
 659 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 660                                                unsigned section, unsigned offset, unsigned *n_sectors)
 661 {
 662         unsigned pl_index, pl_offset;
 663         char *va;
 664
 665         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 666
 667         if (n_sectors)
 668                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 669
 670         va = lowmem_page_address(pl[pl_index].page);
 671
 672         return (struct journal_sector *)(va + pl_offset);
 673 }
 674
 675 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 676 {
 677         return access_page_list(ic, ic->journal, section, offset, NULL);
 678 }
 679
 680 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 681 {
 682         unsigned rel_sector, offset;
 683         struct journal_sector *js;
 684
 685         access_journal_check(ic, section, n, true, "access_journal_entry");
 686
 687         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 688         offset = n / JOURNAL_BLOCK_SECTORS;
 689
 690         js = access_journal(ic, section, rel_sector);
 691         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 692 }
 693
 694 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 695 {
 696         n <<= ic->sb->log2_sectors_per_block;
 697
 698         n += JOURNAL_BLOCK_SECTORS;
 699
 700         access_journal_check(ic, section, n, false, "access_journal_data");
 701
 702         return access_journal(ic, section, n);
 703 }
 704
 705 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 706 {
 707         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 708         int r;
 709         unsigned j, size;
 710
 711         desc->tfm = ic->journal_mac;
 712
 713         r = crypto_shash_init(desc);
 714         if (unlikely(r)) {
 715                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 716                 goto err;
 717         }
 718
 719         for (j = 0; j < ic->journal_section_entries; j++) {
 720                 struct journal_entry *je = access_journal_entry(ic, section, j);
 721                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 722                 if (unlikely(r)) {
 723                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 724                         goto err;
 725                 }
 726         }
 727
 728         size = crypto_shash_digestsize(ic->journal_mac);
 729
 730         if (likely(size <= JOURNAL_MAC_SIZE)) {
 731                 r = crypto_shash_final(desc, result);
 732                 if (unlikely(r)) {
 733                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 734                         goto err;
 735                 }
 736                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 737         } else {
 738                 __u8 digest[HASH_MAX_DIGESTSIZE];
 739
 740                 if (WARN_ON(size > sizeof(digest))) {
 741                         dm_integrity_io_error(ic, "digest_size", -EINVAL);
 742                         goto err;
 743                 }
 744                 r = crypto_shash_final(desc, digest);
 745                 if (unlikely(r)) {
 746                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 747                         goto err;
 748                 }
 749                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 750         }
 751
 752         return;
 753 err:
 754         memset(result, 0, JOURNAL_MAC_SIZE);
 755 }
 756
 757 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 758 {
 759         __u8 result[JOURNAL_MAC_SIZE];
 760         unsigned j;
 761
 762         if (!ic->journal_mac)
 763                 return;
 764
 765         section_mac(ic, section, result);
 766
 767         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 768                 struct journal_sector *js = access_journal(ic, section, j);
 769
 770                 if (likely(wr))
 771                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 772                 else {
 773                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
 774                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 775                 }
 776         }
 777 }
 778
 779 static void complete_journal_op(void *context)
 780 {
 781         struct journal_completion *comp = context;
 782         BUG_ON(!atomic_read(&comp->in_flight));
 783         if (likely(atomic_dec_and_test(&comp->in_flight)))
 784                 complete(&comp->comp);
 785 }
 786
 787 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 788                         unsigned n_sections, struct journal_completion *comp)
 789 {
 790         struct async_submit_ctl submit;
 791         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 792         unsigned pl_index, pl_offset, section_index;
 793         struct page_list *source_pl, *target_pl;
 794
 795         if (likely(encrypt)) {
 796                 source_pl = ic->journal;
 797                 target_pl = ic->journal_io;
 798         } else {
 799                 source_pl = ic->journal_io;
 800                 target_pl = ic->journal;
 801         }
 802
 803         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 804
 805         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 806
 807         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 808
 809         section_index = pl_index;
 810
 811         do {
 812                 size_t this_step;
 813                 struct page *src_pages[2];
 814                 struct page *dst_page;
 815
 816                 while (unlikely(pl_index == section_index)) {
 817                         unsigned dummy;
 818                         if (likely(encrypt))
 819                                 rw_section_mac(ic, section, true);
 820                         section++;
 821                         n_sections--;
 822                         if (!n_sections)
 823                                 break;
 824                         page_list_location(ic, section, 0, &section_index, &dummy);
 825                 }
 826
 827                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 828                 dst_page = target_pl[pl_index].page;
 829                 src_pages[0] = source_pl[pl_index].page;
 830                 src_pages[1] = ic->journal_xor[pl_index].page;
 831
 832                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 833
 834                 pl_index++;
 835                 pl_offset = 0;
 836                 n_bytes -= this_step;
 837         } while (n_bytes);
 838
 839         BUG_ON(n_sections);
 840
 841         async_tx_issue_pending_all();
 842 }
 843
 844 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 845 {
 846         struct journal_completion *comp = req->data;
 847         if (unlikely(err)) {
 848                 if (likely(err == -EINPROGRESS)) {
 849                         complete(&comp->ic->crypto_backoff);
 850                         return;
 851                 }
 852                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 853         }
 854         complete_journal_op(comp);
 855 }
 856
 857 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 858 {
 859         int r;
 860         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 861                                       complete_journal_encrypt, comp);
 862         if (likely(encrypt))
 863                 r = crypto_skcipher_encrypt(req);
 864         else
 865                 r = crypto_skcipher_decrypt(req);
 866         if (likely(!r))
 867                 return false;
 868         if (likely(r == -EINPROGRESS))
 869                 return true;
 870         if (likely(r == -EBUSY)) {
 871                 wait_for_completion(&comp->ic->crypto_backoff);
 872                 reinit_completion(&comp->ic->crypto_backoff);
 873                 return true;
 874         }
 875         dm_integrity_io_error(comp->ic, "encrypt", r);
 876         return false;
 877 }
 878
 879 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 880                           unsigned n_sections, struct journal_completion *comp)
 881 {
 882         struct scatterlist **source_sg;
 883         struct scatterlist **target_sg;
 884
 885         atomic_add(2, &comp->in_flight);
 886
 887         if (likely(encrypt)) {
 888                 source_sg = ic->journal_scatterlist;
 889                 target_sg = ic->journal_io_scatterlist;
 890         } else {
 891                 source_sg = ic->journal_io_scatterlist;
 892                 target_sg = ic->journal_scatterlist;
 893         }
 894
 895         do {
 896                 struct skcipher_request *req;
 897                 unsigned ivsize;
 898                 char *iv;
 899
 900                 if (likely(encrypt))
 901                         rw_section_mac(ic, section, true);
 902
 903                 req = ic->sk_requests[section];
 904                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
 905                 iv = req->iv;
 906
 907                 memcpy(iv, iv + ivsize, ivsize);
 908
 909                 req->src = source_sg[section];
 910                 req->dst = target_sg[section];
 911
 912                 if (unlikely(do_crypt(encrypt, req, comp)))
 913                         atomic_inc(&comp->in_flight);
 914
 915                 section++;
 916                 n_sections--;
 917         } while (n_sections);
 918
 919         atomic_dec(&comp->in_flight);
 920         complete_journal_op(comp);
 921 }
 922
 923 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 924                             unsigned n_sections, struct journal_completion *comp)
 925 {
 926         if (ic->journal_xor)
 927                 return xor_journal(ic, encrypt, section, n_sections, comp);
 928         else
 929                 return crypt_journal(ic, encrypt, section, n_sections, comp);
 930 }
 931
 932 static void complete_journal_io(unsigned long error, void *context)
 933 {
 934         struct journal_completion *comp = context;
 935         if (unlikely(error != 0))
 936                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
 937         complete_journal_op(comp);
 938 }
 939
 940 static void rw_journal_sectors(struct dm_integrity_c *ic, int op, int op_flags,
 941                                unsigned sector, unsigned n_sectors, struct journal_completion *comp)
 942 {
 943         struct dm_io_request io_req;
 944         struct dm_io_region io_loc;
 945         unsigned pl_index, pl_offset;
 946         int r;
 947
 948         if (unlikely(dm_integrity_failed(ic))) {
 949                 if (comp)
 950                         complete_journal_io(-1UL, comp);
 951                 return;
 952         }
 953
 954         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 955         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 956
 957         io_req.bi_op = op;
 958         io_req.bi_op_flags = op_flags;
 959         io_req.mem.type = DM_IO_PAGE_LIST;
 960         if (ic->journal_io)
 961                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
 962         else
 963                 io_req.mem.ptr.pl = &ic->journal[pl_index];
 964         io_req.mem.offset = pl_offset;
 965         if (likely(comp != NULL)) {
 966                 io_req.notify.fn = complete_journal_io;
 967                 io_req.notify.context = comp;
 968         } else {
 969                 io_req.notify.fn = NULL;
 970         }
 971         io_req.client = ic->io;
 972         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 973         io_loc.sector = ic->start + SB_SECTORS + sector;
 974         io_loc.count = n_sectors;
 975
 976         r = dm_io(&io_req, 1, &io_loc, NULL);
 977         if (unlikely(r)) {
 978                 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
 979                 if (comp) {
 980                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 981                         complete_journal_io(-1UL, comp);
 982                 }
 983         }
 984 }
 985
 986 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
 987                        unsigned n_sections, struct journal_completion *comp)
 988 {
 989         unsigned sector, n_sectors;
 990
 991         sector = section * ic->journal_section_sectors;
 992         n_sectors = n_sections * ic->journal_section_sectors;
 993
 994         rw_journal_sectors(ic, op, op_flags, sector, n_sectors, comp);
 995 }
 996
 997 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
 998 {
 999         struct journal_completion io_comp;
1000         struct journal_completion crypt_comp_1;
1001         struct journal_completion crypt_comp_2;
1002         unsigned i;
1003
1004         io_comp.ic = ic;
1005         init_completion(&io_comp.comp);
1006
1007         if (commit_start + commit_sections <= ic->journal_sections) {
1008                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1009                 if (ic->journal_io) {
1010                         crypt_comp_1.ic = ic;
1011                         init_completion(&crypt_comp_1.comp);
1012                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1013                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1014                         wait_for_completion_io(&crypt_comp_1.comp);
1015                 } else {
1016                         for (i = 0; i < commit_sections; i++)
1017                                 rw_section_mac(ic, commit_start + i, true);
1018                 }
1019                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
1020                            commit_sections, &io_comp);
1021         } else {
1022                 unsigned to_end;
1023                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1024                 to_end = ic->journal_sections - commit_start;
1025                 if (ic->journal_io) {
1026                         crypt_comp_1.ic = ic;
1027                         init_completion(&crypt_comp_1.comp);
1028                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1029                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1030                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
1031                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1032                                 reinit_completion(&crypt_comp_1.comp);
1033                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1034                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1035                                 wait_for_completion_io(&crypt_comp_1.comp);
1036                         } else {
1037                                 crypt_comp_2.ic = ic;
1038                                 init_completion(&crypt_comp_2.comp);
1039                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1040                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1041                                 wait_for_completion_io(&crypt_comp_1.comp);
1042                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1043                                 wait_for_completion_io(&crypt_comp_2.comp);
1044                         }
1045                 } else {
1046                         for (i = 0; i < to_end; i++)
1047                                 rw_section_mac(ic, commit_start + i, true);
1048                         rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1049                         for (i = 0; i < commit_sections - to_end; i++)
1050                                 rw_section_mac(ic, i, true);
1051                 }
1052                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
1053         }
1054
1055         wait_for_completion_io(&io_comp.comp);
1056 }
1057
1058 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1059                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1060 {
1061         struct dm_io_request io_req;
1062         struct dm_io_region io_loc;
1063         int r;
1064         unsigned sector, pl_index, pl_offset;
1065
1066         BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1067
1068         if (unlikely(dm_integrity_failed(ic))) {
1069                 fn(-1UL, data);
1070                 return;
1071         }
1072
1073         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1074
1075         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1076         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1077
1078         io_req.bi_op = REQ_OP_WRITE;
1079         io_req.bi_op_flags = 0;
1080         io_req.mem.type = DM_IO_PAGE_LIST;
1081         io_req.mem.ptr.pl = &ic->journal[pl_index];
1082         io_req.mem.offset = pl_offset;
1083         io_req.notify.fn = fn;
1084         io_req.notify.context = data;
1085         io_req.client = ic->io;
1086         io_loc.bdev = ic->dev->bdev;
1087         io_loc.sector = target;
1088         io_loc.count = n_sectors;
1089
1090         r = dm_io(&io_req, 1, &io_loc, NULL);
1091         if (unlikely(r)) {
1092                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1093                 fn(-1UL, data);
1094         }
1095 }
1096
1097 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1098 {
1099         return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1100                range1->logical_sector + range1->n_sectors > range2->logical_sector;
1101 }
1102
1103 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1104 {
1105         struct rb_node **n = &ic->in_progress.rb_node;
1106         struct rb_node *parent;
1107
1108         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1109
1110         if (likely(check_waiting)) {
1111                 struct dm_integrity_range *range;
1112                 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1113                         if (unlikely(ranges_overlap(range, new_range)))
1114                                 return false;
1115                 }
1116         }
1117
1118         parent = NULL;
1119
1120         while (*n) {
1121                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1122
1123                 parent = *n;
1124                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1125                         n = &range->node.rb_left;
1126                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1127                         n = &range->node.rb_right;
1128                 } else {
1129                         return false;
1130                 }
1131         }
1132
1133         rb_link_node(&new_range->node, parent, n);
1134         rb_insert_color(&new_range->node, &ic->in_progress);
1135
1136         return true;
1137 }
1138
1139 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1140 {
1141         rb_erase(&range->node, &ic->in_progress);
1142         while (unlikely(!list_empty(&ic->wait_list))) {
1143                 struct dm_integrity_range *last_range =
1144                         list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1145                 struct task_struct *last_range_task;
1146                 last_range_task = last_range->task;
1147                 list_del(&last_range->wait_entry);
1148                 if (!add_new_range(ic, last_range, false)) {
1149                         last_range->task = last_range_task;
1150                         list_add(&last_range->wait_entry, &ic->wait_list);
1151                         break;
1152                 }
1153                 last_range->waiting = false;
1154                 wake_up_process(last_range_task);
1155         }
1156 }
1157
1158 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1159 {
1160         unsigned long flags;
1161
1162         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1163         remove_range_unlocked(ic, range);
1164         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1165 }
1166
1167 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1168 {
1169         new_range->waiting = true;
1170         list_add_tail(&new_range->wait_entry, &ic->wait_list);
1171         new_range->task = current;
1172         do {
1173                 __set_current_state(TASK_UNINTERRUPTIBLE);
1174                 spin_unlock_irq(&ic->endio_wait.lock);
1175                 io_schedule();
1176                 spin_lock_irq(&ic->endio_wait.lock);
1177         } while (unlikely(new_range->waiting));
1178 }
1179
1180 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1181 {
1182         if (unlikely(!add_new_range(ic, new_range, true)))
1183                 wait_and_add_new_range(ic, new_range);
1184 }
1185
1186 static void init_journal_node(struct journal_node *node)
1187 {
1188         RB_CLEAR_NODE(&node->node);
1189         node->sector = (sector_t)-1;
1190 }
1191
1192 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1193 {
1194         struct rb_node **link;
1195         struct rb_node *parent;
1196
1197         node->sector = sector;
1198         BUG_ON(!RB_EMPTY_NODE(&node->node));
1199
1200         link = &ic->journal_tree_root.rb_node;
1201         parent = NULL;
1202
1203         while (*link) {
1204                 struct journal_node *j;
1205                 parent = *link;
1206                 j = container_of(parent, struct journal_node, node);
1207                 if (sector < j->sector)
1208                         link = &j->node.rb_left;
1209                 else
1210                         link = &j->node.rb_right;
1211         }
1212
1213         rb_link_node(&node->node, parent, link);
1214         rb_insert_color(&node->node, &ic->journal_tree_root);
1215 }
1216
1217 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1218 {
1219         BUG_ON(RB_EMPTY_NODE(&node->node));
1220         rb_erase(&node->node, &ic->journal_tree_root);
1221         init_journal_node(node);
1222 }
1223
1224 #define NOT_FOUND       (-1U)
1225
1226 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1227 {
1228         struct rb_node *n = ic->journal_tree_root.rb_node;
1229         unsigned found = NOT_FOUND;
1230         *next_sector = (sector_t)-1;
1231         while (n) {
1232                 struct journal_node *j = container_of(n, struct journal_node, node);
1233                 if (sector == j->sector) {
1234                         found = j - ic->journal_tree;
1235                 }
1236                 if (sector < j->sector) {
1237                         *next_sector = j->sector;
1238                         n = j->node.rb_left;
1239                 } else {
1240                         n = j->node.rb_right;
1241                 }
1242         }
1243
1244         return found;
1245 }
1246
1247 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1248 {
1249         struct journal_node *node, *next_node;
1250         struct rb_node *next;
1251
1252         if (unlikely(pos >= ic->journal_entries))
1253                 return false;
1254         node = &ic->journal_tree[pos];
1255         if (unlikely(RB_EMPTY_NODE(&node->node)))
1256                 return false;
1257         if (unlikely(node->sector != sector))
1258                 return false;
1259
1260         next = rb_next(&node->node);
1261         if (unlikely(!next))
1262                 return true;
1263
1264         next_node = container_of(next, struct journal_node, node);
1265         return next_node->sector != sector;
1266 }
1267
1268 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1269 {
1270         struct rb_node *next;
1271         struct journal_node *next_node;
1272         unsigned next_section;
1273
1274         BUG_ON(RB_EMPTY_NODE(&node->node));
1275
1276         next = rb_next(&node->node);
1277         if (unlikely(!next))
1278                 return false;
1279
1280         next_node = container_of(next, struct journal_node, node);
1281
1282         if (next_node->sector != node->sector)
1283                 return false;
1284
1285         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1286         if (next_section >= ic->committed_section &&
1287             next_section < ic->committed_section + ic->n_committed_sections)
1288                 return true;
1289         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1290                 return true;
1291
1292         return false;
1293 }
1294
1295 #define TAG_READ        0
1296 #define TAG_WRITE       1
1297 #define TAG_CMP         2
1298
1299 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1300                                unsigned *metadata_offset, unsigned total_size, int op)
1301 {
1302         do {
1303                 unsigned char *data, *dp;
1304                 struct dm_buffer *b;
1305                 unsigned to_copy;
1306                 int r;
1307
1308                 r = dm_integrity_failed(ic);
1309                 if (unlikely(r))
1310                         return r;
1311
1312                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1313                 if (IS_ERR(data))
1314                         return PTR_ERR(data);
1315
1316                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1317                 dp = data + *metadata_offset;
1318                 if (op == TAG_READ) {
1319                         memcpy(tag, dp, to_copy);
1320                 } else if (op == TAG_WRITE) {
1321                         memcpy(dp, tag, to_copy);
1322                         dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1323                 } else  {
1324                         /* e.g.: op == TAG_CMP */
1325                         if (unlikely(memcmp(dp, tag, to_copy))) {
1326                                 unsigned i;
1327
1328                                 for (i = 0; i < to_copy; i++) {
1329                                         if (dp[i] != tag[i])
1330                                                 break;
1331                                         total_size--;
1332                                 }
1333                                 dm_bufio_release(b);
1334                                 return total_size;
1335                         }
1336                 }
1337                 dm_bufio_release(b);
1338
1339                 tag += to_copy;
1340                 *metadata_offset += to_copy;
1341                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1342                         (*metadata_block)++;
1343                         *metadata_offset = 0;
1344                 }
1345                 total_size -= to_copy;
1346         } while (unlikely(total_size));
1347
1348         return 0;
1349 }
1350
1351 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1352 {
1353         int r;
1354         r = dm_bufio_write_dirty_buffers(ic->bufio);
1355         if (unlikely(r))
1356                 dm_integrity_io_error(ic, "writing tags", r);
1357 }
1358
1359 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1360 {
1361         DECLARE_WAITQUEUE(wait, current);
1362         __add_wait_queue(&ic->endio_wait, &wait);
1363         __set_current_state(TASK_UNINTERRUPTIBLE);
1364         spin_unlock_irq(&ic->endio_wait.lock);
1365         io_schedule();
1366         spin_lock_irq(&ic->endio_wait.lock);
1367         __remove_wait_queue(&ic->endio_wait, &wait);
1368 }
1369
1370 static void autocommit_fn(struct timer_list *t)
1371 {
1372         struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1373
1374         if (likely(!dm_integrity_failed(ic)))
1375                 queue_work(ic->commit_wq, &ic->commit_work);
1376 }
1377
1378 static void schedule_autocommit(struct dm_integrity_c *ic)
1379 {
1380         if (!timer_pending(&ic->autocommit_timer))
1381                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1382 }
1383
1384 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1385 {
1386         struct bio *bio;
1387         unsigned long flags;
1388
1389         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1390         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1391         bio_list_add(&ic->flush_bio_list, bio);
1392         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1393
1394         queue_work(ic->commit_wq, &ic->commit_work);
1395 }
1396
1397 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1398 {
1399         int r = dm_integrity_failed(ic);
1400         if (unlikely(r) && !bio->bi_status)
1401                 bio->bi_status = errno_to_blk_status(r);
1402         if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1403                 unsigned long flags;
1404                 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1405                 bio_list_add(&ic->synchronous_bios, bio);
1406                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1407                 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1408                 return;
1409         }
1410         bio_endio(bio);
1411 }
1412
1413 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1414 {
1415         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1416
1417         if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1418                 submit_flush_bio(ic, dio);
1419         else
1420                 do_endio(ic, bio);
1421 }
1422
1423 static void dec_in_flight(struct dm_integrity_io *dio)
1424 {
1425         if (atomic_dec_and_test(&dio->in_flight)) {
1426                 struct dm_integrity_c *ic = dio->ic;
1427                 struct bio *bio;
1428
1429                 remove_range(ic, &dio->range);
1430
1431                 if (unlikely(dio->write))
1432                         schedule_autocommit(ic);
1433
1434                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1435
1436                 if (unlikely(dio->bi_status) && !bio->bi_status)
1437                         bio->bi_status = dio->bi_status;
1438                 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1439                         dio->range.logical_sector += dio->range.n_sectors;
1440                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1441                         INIT_WORK(&dio->work, integrity_bio_wait);
1442                         queue_work(ic->wait_wq, &dio->work);
1443                         return;
1444                 }
1445                 do_endio_flush(ic, dio);
1446         }
1447 }
1448
1449 static void integrity_end_io(struct bio *bio)
1450 {
1451         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1452
1453         bio->bi_iter = dio->orig_bi_iter;
1454         bio->bi_disk = dio->orig_bi_disk;
1455         bio->bi_partno = dio->orig_bi_partno;
1456         if (dio->orig_bi_integrity) {
1457                 bio->bi_integrity = dio->orig_bi_integrity;
1458                 bio->bi_opf |= REQ_INTEGRITY;
1459         }
1460         bio->bi_end_io = dio->orig_bi_end_io;
1461
1462         if (dio->completion)
1463                 complete(dio->completion);
1464
1465         dec_in_flight(dio);
1466 }
1467
1468 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1469                                       const char *data, char *result)
1470 {
1471         __u64 sector_le = cpu_to_le64(sector);
1472         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1473         int r;
1474         unsigned digest_size;
1475
1476         req->tfm = ic->internal_hash;
1477
1478         r = crypto_shash_init(req);
1479         if (unlikely(r < 0)) {
1480                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1481                 goto failed;
1482         }
1483
1484         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1485         if (unlikely(r < 0)) {
1486                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1487                 goto failed;
1488         }
1489
1490         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1491         if (unlikely(r < 0)) {
1492                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1493                 goto failed;
1494         }
1495
1496         r = crypto_shash_final(req, result);
1497         if (unlikely(r < 0)) {
1498                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1499                 goto failed;
1500         }
1501
1502         digest_size = crypto_shash_digestsize(ic->internal_hash);
1503         if (unlikely(digest_size < ic->tag_size))
1504                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1505
1506         return;
1507
1508 failed:
1509         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1510         get_random_bytes(result, ic->tag_size);
1511 }
1512
1513 static void integrity_metadata(struct work_struct *w)
1514 {
1515         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1516         struct dm_integrity_c *ic = dio->ic;
1517
1518         int r;
1519
1520         if (ic->internal_hash) {
1521                 struct bvec_iter iter;
1522                 struct bio_vec bv;
1523                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1524                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1525                 char *checksums;
1526                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1527                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1528                 unsigned sectors_to_process = dio->range.n_sectors;
1529                 sector_t sector = dio->range.logical_sector;
1530
1531                 if (unlikely(ic->mode == 'R'))
1532                         goto skip_io;
1533
1534                 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1535                                     GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1536                 if (!checksums) {
1537                         checksums = checksums_onstack;
1538                         if (WARN_ON(extra_space &&
1539                                     digest_size > sizeof(checksums_onstack))) {
1540                                 r = -EINVAL;
1541                                 goto error;
1542                         }
1543                 }
1544
1545                 __bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
1546                         unsigned pos;
1547                         char *mem, *checksums_ptr;
1548
1549 again:
1550                         mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1551                         pos = 0;
1552                         checksums_ptr = checksums;
1553                         do {
1554                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1555                                 checksums_ptr += ic->tag_size;
1556                                 sectors_to_process -= ic->sectors_per_block;
1557                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1558                                 sector += ic->sectors_per_block;
1559                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1560                         kunmap_atomic(mem);
1561
1562                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1563                                                 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1564                         if (unlikely(r)) {
1565                                 if (r > 0) {
1566                                         DMERR_LIMIT("Checksum failed at sector 0x%llx",
1567                                                     (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1568                                         r = -EILSEQ;
1569                                         atomic64_inc(&ic->number_of_mismatches);
1570                                 }
1571                                 if (likely(checksums != checksums_onstack))
1572                                         kfree(checksums);
1573                                 goto error;
1574                         }
1575
1576                         if (!sectors_to_process)
1577                                 break;
1578
1579                         if (unlikely(pos < bv.bv_len)) {
1580                                 bv.bv_offset += pos;
1581                                 bv.bv_len -= pos;
1582                                 goto again;
1583                         }
1584                 }
1585
1586                 if (likely(checksums != checksums_onstack))
1587                         kfree(checksums);
1588         } else {
1589                 struct bio_integrity_payload *bip = dio->orig_bi_integrity;
1590
1591                 if (bip) {
1592                         struct bio_vec biv;
1593                         struct bvec_iter iter;
1594                         unsigned data_to_process = dio->range.n_sectors;
1595                         sector_to_block(ic, data_to_process);
1596                         data_to_process *= ic->tag_size;
1597
1598                         bip_for_each_vec(biv, bip, iter) {
1599                                 unsigned char *tag;
1600                                 unsigned this_len;
1601
1602                                 BUG_ON(PageHighMem(biv.bv_page));
1603                                 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1604                                 this_len = min(biv.bv_len, data_to_process);
1605                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1606                                                         this_len, !dio->write ? TAG_READ : TAG_WRITE);
1607                                 if (unlikely(r))
1608                                         goto error;
1609                                 data_to_process -= this_len;
1610                                 if (!data_to_process)
1611                                         break;
1612                         }
1613                 }
1614         }
1615 skip_io:
1616         dec_in_flight(dio);
1617         return;
1618 error:
1619         dio->bi_status = errno_to_blk_status(r);
1620         dec_in_flight(dio);
1621 }
1622
1623 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1624 {
1625         struct dm_integrity_c *ic = ti->private;
1626         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1627         struct bio_integrity_payload *bip;
1628
1629         sector_t area, offset;
1630
1631         dio->ic = ic;
1632         dio->bi_status = 0;
1633
1634         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1635                 submit_flush_bio(ic, dio);
1636                 return DM_MAPIO_SUBMITTED;
1637         }
1638
1639         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1640         dio->write = bio_op(bio) == REQ_OP_WRITE;
1641         dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1642         if (unlikely(dio->fua)) {
1643                 /*
1644                  * Don't pass down the FUA flag because we have to flush
1645                  * disk cache anyway.
1646                  */
1647                 bio->bi_opf &= ~REQ_FUA;
1648         }
1649         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1650                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1651                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1652                       (unsigned long long)ic->provided_data_sectors);
1653                 return DM_MAPIO_KILL;
1654         }
1655         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1656                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1657                       ic->sectors_per_block,
1658                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1659                 return DM_MAPIO_KILL;
1660         }
1661
1662         if (ic->sectors_per_block > 1) {
1663                 struct bvec_iter iter;
1664                 struct bio_vec bv;
1665                 bio_for_each_segment(bv, bio, iter) {
1666                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1667                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1668                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1669                                 return DM_MAPIO_KILL;
1670                         }
1671                 }
1672         }
1673
1674         bip = bio_integrity(bio);
1675         if (!ic->internal_hash) {
1676                 if (bip) {
1677                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1678                         if (ic->log2_tag_size >= 0)
1679                                 wanted_tag_size <<= ic->log2_tag_size;
1680                         else
1681                                 wanted_tag_size *= ic->tag_size;
1682                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1683                                 DMERR("Invalid integrity data size %u, expected %u",
1684                                       bip->bip_iter.bi_size, wanted_tag_size);
1685                                 return DM_MAPIO_KILL;
1686                         }
1687                 }
1688         } else {
1689                 if (unlikely(bip != NULL)) {
1690                         DMERR("Unexpected integrity data when using internal hash");
1691                         return DM_MAPIO_KILL;
1692                 }
1693         }
1694
1695         if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1696                 return DM_MAPIO_KILL;
1697
1698         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1699         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1700         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1701
1702         dm_integrity_map_continue(dio, true);
1703         return DM_MAPIO_SUBMITTED;
1704 }
1705
1706 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1707                                  unsigned journal_section, unsigned journal_entry)
1708 {
1709         struct dm_integrity_c *ic = dio->ic;
1710         sector_t logical_sector;
1711         unsigned n_sectors;
1712
1713         logical_sector = dio->range.logical_sector;
1714         n_sectors = dio->range.n_sectors;
1715         do {
1716                 struct bio_vec bv = bio_iovec(bio);
1717                 char *mem;
1718
1719                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1720                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1721                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1722                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1723 retry_kmap:
1724                 mem = kmap_atomic(bv.bv_page);
1725                 if (likely(dio->write))
1726                         flush_dcache_page(bv.bv_page);
1727
1728                 do {
1729                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1730
1731                         if (unlikely(!dio->write)) {
1732                                 struct journal_sector *js;
1733                                 char *mem_ptr;
1734                                 unsigned s;
1735
1736                                 if (unlikely(journal_entry_is_inprogress(je))) {
1737                                         flush_dcache_page(bv.bv_page);
1738                                         kunmap_atomic(mem);
1739
1740                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1741                                         goto retry_kmap;
1742                                 }
1743                                 smp_rmb();
1744                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1745                                 js = access_journal_data(ic, journal_section, journal_entry);
1746                                 mem_ptr = mem + bv.bv_offset;
1747                                 s = 0;
1748                                 do {
1749                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1750                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1751                                         js++;
1752                                         mem_ptr += 1 << SECTOR_SHIFT;
1753                                 } while (++s < ic->sectors_per_block);
1754 #ifdef INTERNAL_VERIFY
1755                                 if (ic->internal_hash) {
1756                                         char checksums_onstack[max(HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1757
1758                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1759                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1760                                                 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1761                                                             (unsigned long long)logical_sector);
1762                                         }
1763                                 }
1764 #endif
1765                         }
1766
1767                         if (!ic->internal_hash) {
1768                                 struct bio_integrity_payload *bip = bio_integrity(bio);
1769                                 unsigned tag_todo = ic->tag_size;
1770                                 char *tag_ptr = journal_entry_tag(ic, je);
1771
1772                                 if (bip) do {
1773                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1774                                         unsigned tag_now = min(biv.bv_len, tag_todo);
1775                                         char *tag_addr;
1776                                         BUG_ON(PageHighMem(biv.bv_page));
1777                                         tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1778                                         if (likely(dio->write))
1779                                                 memcpy(tag_ptr, tag_addr, tag_now);
1780                                         else
1781                                                 memcpy(tag_addr, tag_ptr, tag_now);
1782                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1783                                         tag_ptr += tag_now;
1784                                         tag_todo -= tag_now;
1785                                 } while (unlikely(tag_todo)); else {
1786                                         if (likely(dio->write))
1787                                                 memset(tag_ptr, 0, tag_todo);
1788                                 }
1789                         }
1790
1791                         if (likely(dio->write)) {
1792                                 struct journal_sector *js;
1793                                 unsigned s;
1794
1795                                 js = access_journal_data(ic, journal_section, journal_entry);
1796                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1797
1798                                 s = 0;
1799                                 do {
1800                                         je->last_bytes[s] = js[s].commit_id;
1801                                 } while (++s < ic->sectors_per_block);
1802
1803                                 if (ic->internal_hash) {
1804                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1805                                         if (unlikely(digest_size > ic->tag_size)) {
1806                                                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1807                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1808                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1809                                         } else
1810                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1811                                 }
1812
1813                                 journal_entry_set_sector(je, logical_sector);
1814                         }
1815                         logical_sector += ic->sectors_per_block;
1816
1817                         journal_entry++;
1818                         if (unlikely(journal_entry == ic->journal_section_entries)) {
1819                                 journal_entry = 0;
1820                                 journal_section++;
1821                                 wraparound_section(ic, &journal_section);
1822                         }
1823
1824                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1825                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1826
1827                 if (unlikely(!dio->write))
1828                         flush_dcache_page(bv.bv_page);
1829                 kunmap_atomic(mem);
1830         } while (n_sectors);
1831
1832         if (likely(dio->write)) {
1833                 smp_mb();
1834                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1835                         wake_up(&ic->copy_to_journal_wait);
1836                 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1837                         queue_work(ic->commit_wq, &ic->commit_work);
1838                 } else {
1839                         schedule_autocommit(ic);
1840                 }
1841         } else {
1842                 remove_range(ic, &dio->range);
1843         }
1844
1845         if (unlikely(bio->bi_iter.bi_size)) {
1846                 sector_t area, offset;
1847
1848                 dio->range.logical_sector = logical_sector;
1849                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1850                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1851                 return true;
1852         }
1853
1854         return false;
1855 }
1856
1857 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1858 {
1859         struct dm_integrity_c *ic = dio->ic;
1860         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1861         unsigned journal_section, journal_entry;
1862         unsigned journal_read_pos;
1863         struct completion read_comp;
1864         bool need_sync_io = ic->internal_hash && !dio->write;
1865
1866         if (need_sync_io && from_map) {
1867                 INIT_WORK(&dio->work, integrity_bio_wait);
1868                 queue_work(ic->metadata_wq, &dio->work);
1869                 return;
1870         }
1871
1872 lock_retry:
1873         spin_lock_irq(&ic->endio_wait.lock);
1874 retry:
1875         if (unlikely(dm_integrity_failed(ic))) {
1876                 spin_unlock_irq(&ic->endio_wait.lock);
1877                 do_endio(ic, bio);
1878                 return;
1879         }
1880         dio->range.n_sectors = bio_sectors(bio);
1881         journal_read_pos = NOT_FOUND;
1882         if (likely(ic->mode == 'J')) {
1883                 if (dio->write) {
1884                         unsigned next_entry, i, pos;
1885                         unsigned ws, we, range_sectors;
1886
1887                         dio->range.n_sectors = min(dio->range.n_sectors,
1888                                                    (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
1889                         if (unlikely(!dio->range.n_sectors)) {
1890                                 if (from_map)
1891                                         goto offload_to_thread;
1892                                 sleep_on_endio_wait(ic);
1893                                 goto retry;
1894                         }
1895                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1896                         ic->free_sectors -= range_sectors;
1897                         journal_section = ic->free_section;
1898                         journal_entry = ic->free_section_entry;
1899
1900                         next_entry = ic->free_section_entry + range_sectors;
1901                         ic->free_section_entry = next_entry % ic->journal_section_entries;
1902                         ic->free_section += next_entry / ic->journal_section_entries;
1903                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1904                         wraparound_section(ic, &ic->free_section);
1905
1906                         pos = journal_section * ic->journal_section_entries + journal_entry;
1907                         ws = journal_section;
1908                         we = journal_entry;
1909                         i = 0;
1910                         do {
1911                                 struct journal_entry *je;
1912
1913                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1914                                 pos++;
1915                                 if (unlikely(pos >= ic->journal_entries))
1916                                         pos = 0;
1917
1918                                 je = access_journal_entry(ic, ws, we);
1919                                 BUG_ON(!journal_entry_is_unused(je));
1920                                 journal_entry_set_inprogress(je);
1921                                 we++;
1922                                 if (unlikely(we == ic->journal_section_entries)) {
1923                                         we = 0;
1924                                         ws++;
1925                                         wraparound_section(ic, &ws);
1926                                 }
1927                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1928
1929                         spin_unlock_irq(&ic->endio_wait.lock);
1930                         goto journal_read_write;
1931                 } else {
1932                         sector_t next_sector;
1933                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1934                         if (likely(journal_read_pos == NOT_FOUND)) {
1935                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1936                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
1937                         } else {
1938                                 unsigned i;
1939                                 unsigned jp = journal_read_pos + 1;
1940                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1941                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1942                                                 break;
1943                                 }
1944                                 dio->range.n_sectors = i;
1945                         }
1946                 }
1947         }
1948         if (unlikely(!add_new_range(ic, &dio->range, true))) {
1949                 /*
1950                  * We must not sleep in the request routine because it could
1951                  * stall bios on current->bio_list.
1952                  * So, we offload the bio to a workqueue if we have to sleep.
1953                  */
1954                 if (from_map) {
1955 offload_to_thread:
1956                         spin_unlock_irq(&ic->endio_wait.lock);
1957                         INIT_WORK(&dio->work, integrity_bio_wait);
1958                         queue_work(ic->wait_wq, &dio->work);
1959                         return;
1960                 }
1961                 if (journal_read_pos != NOT_FOUND)
1962                         dio->range.n_sectors = ic->sectors_per_block;
1963                 wait_and_add_new_range(ic, &dio->range);
1964                 /*
1965                  * wait_and_add_new_range drops the spinlock, so the journal
1966                  * may have been changed arbitrarily. We need to recheck.
1967                  * To simplify the code, we restrict I/O size to just one block.
1968                  */
1969                 if (journal_read_pos != NOT_FOUND) {
1970                         sector_t next_sector;
1971                         unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1972                         if (unlikely(new_pos != journal_read_pos)) {
1973                                 remove_range_unlocked(ic, &dio->range);
1974                                 goto retry;
1975                         }
1976                 }
1977         }
1978         spin_unlock_irq(&ic->endio_wait.lock);
1979
1980         if (unlikely(journal_read_pos != NOT_FOUND)) {
1981                 journal_section = journal_read_pos / ic->journal_section_entries;
1982                 journal_entry = journal_read_pos % ic->journal_section_entries;
1983                 goto journal_read_write;
1984         }
1985
1986         if (ic->mode == 'B' && dio->write) {
1987                 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
1988                                      dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
1989                         struct bitmap_block_status *bbs;
1990
1991                         bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
1992                         spin_lock(&bbs->bio_queue_lock);
1993                         bio_list_add(&bbs->bio_queue, bio);
1994                         spin_unlock(&bbs->bio_queue_lock);
1995                         queue_work(ic->writer_wq, &bbs->work);
1996                         return;
1997                 }
1998         }
1999
2000         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2001
2002         if (need_sync_io) {
2003                 init_completion(&read_comp);
2004                 dio->completion = &read_comp;
2005         } else
2006                 dio->completion = NULL;
2007
2008         dio->orig_bi_iter = bio->bi_iter;
2009
2010         dio->orig_bi_disk = bio->bi_disk;
2011         dio->orig_bi_partno = bio->bi_partno;
2012         bio_set_dev(bio, ic->dev->bdev);
2013
2014         dio->orig_bi_integrity = bio_integrity(bio);
2015         bio->bi_integrity = NULL;
2016         bio->bi_opf &= ~REQ_INTEGRITY;
2017
2018         dio->orig_bi_end_io = bio->bi_end_io;
2019         bio->bi_end_io = integrity_end_io;
2020
2021         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2022         generic_make_request(bio);
2023
2024         if (need_sync_io) {
2025                 wait_for_completion_io(&read_comp);
2026                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2027                     dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2028                         goto skip_check;
2029                 if (ic->mode == 'B') {
2030                         if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2031                                              dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2032                                 goto skip_check;
2033                 }
2034
2035                 if (likely(!bio->bi_status))
2036                         integrity_metadata(&dio->work);
2037                 else
2038 skip_check:
2039                         dec_in_flight(dio);
2040
2041         } else {
2042                 INIT_WORK(&dio->work, integrity_metadata);
2043                 queue_work(ic->metadata_wq, &dio->work);
2044         }
2045
2046         return;
2047
2048 journal_read_write:
2049         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2050                 goto lock_retry;
2051
2052         do_endio_flush(ic, dio);
2053 }
2054
2055
2056 static void integrity_bio_wait(struct work_struct *w)
2057 {
2058         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2059
2060         dm_integrity_map_continue(dio, false);
2061 }
2062
2063 static void pad_uncommitted(struct dm_integrity_c *ic)
2064 {
2065         if (ic->free_section_entry) {
2066                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2067                 ic->free_section_entry = 0;
2068                 ic->free_section++;
2069                 wraparound_section(ic, &ic->free_section);
2070                 ic->n_uncommitted_sections++;
2071         }
2072         if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2073                     (ic->n_uncommitted_sections + ic->n_committed_sections) *
2074                     ic->journal_section_entries + ic->free_sectors)) {
2075                 DMCRIT("journal_sections %u, journal_section_entries %u, "
2076                        "n_uncommitted_sections %u, n_committed_sections %u, "
2077                        "journal_section_entries %u, free_sectors %u",
2078                        ic->journal_sections, ic->journal_section_entries,
2079                        ic->n_uncommitted_sections, ic->n_committed_sections,
2080                        ic->journal_section_entries, ic->free_sectors);
2081         }
2082 }
2083
2084 static void integrity_commit(struct work_struct *w)
2085 {
2086         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2087         unsigned commit_start, commit_sections;
2088         unsigned i, j, n;
2089         struct bio *flushes;
2090
2091         del_timer(&ic->autocommit_timer);
2092
2093         spin_lock_irq(&ic->endio_wait.lock);
2094         flushes = bio_list_get(&ic->flush_bio_list);
2095         if (unlikely(ic->mode != 'J')) {
2096                 spin_unlock_irq(&ic->endio_wait.lock);
2097                 dm_integrity_flush_buffers(ic);
2098                 goto release_flush_bios;
2099         }
2100
2101         pad_uncommitted(ic);
2102         commit_start = ic->uncommitted_section;
2103         commit_sections = ic->n_uncommitted_sections;
2104         spin_unlock_irq(&ic->endio_wait.lock);
2105
2106         if (!commit_sections)
2107                 goto release_flush_bios;
2108
2109         i = commit_start;
2110         for (n = 0; n < commit_sections; n++) {
2111                 for (j = 0; j < ic->journal_section_entries; j++) {
2112                         struct journal_entry *je;
2113                         je = access_journal_entry(ic, i, j);
2114                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2115                 }
2116                 for (j = 0; j < ic->journal_section_sectors; j++) {
2117                         struct journal_sector *js;
2118                         js = access_journal(ic, i, j);
2119                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2120                 }
2121                 i++;
2122                 if (unlikely(i >= ic->journal_sections))
2123                         ic->commit_seq = next_commit_seq(ic->commit_seq);
2124                 wraparound_section(ic, &i);
2125         }
2126         smp_rmb();
2127
2128         write_journal(ic, commit_start, commit_sections);
2129
2130         spin_lock_irq(&ic->endio_wait.lock);
2131         ic->uncommitted_section += commit_sections;
2132         wraparound_section(ic, &ic->uncommitted_section);
2133         ic->n_uncommitted_sections -= commit_sections;
2134         ic->n_committed_sections += commit_sections;
2135         spin_unlock_irq(&ic->endio_wait.lock);
2136
2137         if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2138                 queue_work(ic->writer_wq, &ic->writer_work);
2139
2140 release_flush_bios:
2141         while (flushes) {
2142                 struct bio *next = flushes->bi_next;
2143                 flushes->bi_next = NULL;
2144                 do_endio(ic, flushes);
2145                 flushes = next;
2146         }
2147 }
2148
2149 static void complete_copy_from_journal(unsigned long error, void *context)
2150 {
2151         struct journal_io *io = context;
2152         struct journal_completion *comp = io->comp;
2153         struct dm_integrity_c *ic = comp->ic;
2154         remove_range(ic, &io->range);
2155         mempool_free(io, &ic->journal_io_mempool);
2156         if (unlikely(error != 0))
2157                 dm_integrity_io_error(ic, "copying from journal", -EIO);
2158         complete_journal_op(comp);
2159 }
2160
2161 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2162                                struct journal_entry *je)
2163 {
2164         unsigned s = 0;
2165         do {
2166                 js->commit_id = je->last_bytes[s];
2167                 js++;
2168         } while (++s < ic->sectors_per_block);
2169 }
2170
2171 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2172                              unsigned write_sections, bool from_replay)
2173 {
2174         unsigned i, j, n;
2175         struct journal_completion comp;
2176         struct blk_plug plug;
2177
2178         blk_start_plug(&plug);
2179
2180         comp.ic = ic;
2181         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2182         init_completion(&comp.comp);
2183
2184         i = write_start;
2185         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2186 #ifndef INTERNAL_VERIFY
2187                 if (unlikely(from_replay))
2188 #endif
2189                         rw_section_mac(ic, i, false);
2190                 for (j = 0; j < ic->journal_section_entries; j++) {
2191                         struct journal_entry *je = access_journal_entry(ic, i, j);
2192                         sector_t sec, area, offset;
2193                         unsigned k, l, next_loop;
2194                         sector_t metadata_block;
2195                         unsigned metadata_offset;
2196                         struct journal_io *io;
2197
2198                         if (journal_entry_is_unused(je))
2199                                 continue;
2200                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2201                         sec = journal_entry_get_sector(je);
2202                         if (unlikely(from_replay)) {
2203                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2204                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2205                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
2206                                 }
2207                         }
2208                         get_area_and_offset(ic, sec, &area, &offset);
2209                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2210                         for (k = j + 1; k < ic->journal_section_entries; k++) {
2211                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2212                                 sector_t sec2, area2, offset2;
2213                                 if (journal_entry_is_unused(je2))
2214                                         break;
2215                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2216                                 sec2 = journal_entry_get_sector(je2);
2217                                 get_area_and_offset(ic, sec2, &area2, &offset2);
2218                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2219                                         break;
2220                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2221                         }
2222                         next_loop = k - 1;
2223
2224                         io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2225                         io->comp = &comp;
2226                         io->range.logical_sector = sec;
2227                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2228
2229                         spin_lock_irq(&ic->endio_wait.lock);
2230                         add_new_range_and_wait(ic, &io->range);
2231
2232                         if (likely(!from_replay)) {
2233                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2234
2235                                 /* don't write if there is newer committed sector */
2236                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2237                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
2238
2239                                         journal_entry_set_unused(je2);
2240                                         remove_journal_node(ic, &section_node[j]);
2241                                         j++;
2242                                         sec += ic->sectors_per_block;
2243                                         offset += ic->sectors_per_block;
2244                                 }
2245                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2246                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2247
2248                                         journal_entry_set_unused(je2);
2249                                         remove_journal_node(ic, &section_node[k - 1]);
2250                                         k--;
2251                                 }
2252                                 if (j == k) {
2253                                         remove_range_unlocked(ic, &io->range);
2254                                         spin_unlock_irq(&ic->endio_wait.lock);
2255                                         mempool_free(io, &ic->journal_io_mempool);
2256                                         goto skip_io;
2257                                 }
2258                                 for (l = j; l < k; l++) {
2259                                         remove_journal_node(ic, &section_node[l]);
2260                                 }
2261                         }
2262                         spin_unlock_irq(&ic->endio_wait.lock);
2263
2264                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2265                         for (l = j; l < k; l++) {
2266                                 int r;
2267                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2268
2269                                 if (
2270 #ifndef INTERNAL_VERIFY
2271                                     unlikely(from_replay) &&
2272 #endif
2273                                     ic->internal_hash) {
2274                                         char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2275
2276                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2277                                                                   (char *)access_journal_data(ic, i, l), test_tag);
2278                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2279                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2280                                 }
2281
2282                                 journal_entry_set_unused(je2);
2283                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2284                                                         ic->tag_size, TAG_WRITE);
2285                                 if (unlikely(r)) {
2286                                         dm_integrity_io_error(ic, "reading tags", r);
2287                                 }
2288                         }
2289
2290                         atomic_inc(&comp.in_flight);
2291                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2292                                           (k - j) << ic->sb->log2_sectors_per_block,
2293                                           get_data_sector(ic, area, offset),
2294                                           complete_copy_from_journal, io);
2295 skip_io:
2296                         j = next_loop;
2297                 }
2298         }
2299
2300         dm_bufio_write_dirty_buffers_async(ic->bufio);
2301
2302         blk_finish_plug(&plug);
2303
2304         complete_journal_op(&comp);
2305         wait_for_completion_io(&comp.comp);
2306
2307         dm_integrity_flush_buffers(ic);
2308 }
2309
2310 static void integrity_writer(struct work_struct *w)
2311 {
2312         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2313         unsigned write_start, write_sections;
2314
2315         unsigned prev_free_sectors;
2316
2317         /* the following test is not needed, but it tests the replay code */
2318         if (READ_ONCE(ic->suspending) && !ic->meta_dev)
2319                 return;
2320
2321         spin_lock_irq(&ic->endio_wait.lock);
2322         write_start = ic->committed_section;
2323         write_sections = ic->n_committed_sections;
2324         spin_unlock_irq(&ic->endio_wait.lock);
2325
2326         if (!write_sections)
2327                 return;
2328
2329         do_journal_write(ic, write_start, write_sections, false);
2330
2331         spin_lock_irq(&ic->endio_wait.lock);
2332
2333         ic->committed_section += write_sections;
2334         wraparound_section(ic, &ic->committed_section);
2335         ic->n_committed_sections -= write_sections;
2336
2337         prev_free_sectors = ic->free_sectors;
2338         ic->free_sectors += write_sections * ic->journal_section_entries;
2339         if (unlikely(!prev_free_sectors))
2340                 wake_up_locked(&ic->endio_wait);
2341
2342         spin_unlock_irq(&ic->endio_wait.lock);
2343 }
2344
2345 static void recalc_write_super(struct dm_integrity_c *ic)
2346 {
2347         int r;
2348
2349         dm_integrity_flush_buffers(ic);
2350         if (dm_integrity_failed(ic))
2351                 return;
2352
2353         r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2354         if (unlikely(r))
2355                 dm_integrity_io_error(ic, "writing superblock", r);
2356 }
2357
2358 static void integrity_recalc(struct work_struct *w)
2359 {
2360         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2361         struct dm_integrity_range range;
2362         struct dm_io_request io_req;
2363         struct dm_io_region io_loc;
2364         sector_t area, offset;
2365         sector_t metadata_block;
2366         unsigned metadata_offset;
2367         sector_t logical_sector, n_sectors;
2368         __u8 *t;
2369         unsigned i;
2370         int r;
2371         unsigned super_counter = 0;
2372
2373         DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2374
2375         spin_lock_irq(&ic->endio_wait.lock);
2376
2377 next_chunk:
2378
2379         if (unlikely(READ_ONCE(ic->suspending)))
2380                 goto unlock_ret;
2381
2382         range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2383         if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2384                 if (ic->mode == 'B') {
2385                         DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2386                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2387                 }
2388                 goto unlock_ret;
2389         }
2390
2391         get_area_and_offset(ic, range.logical_sector, &area, &offset);
2392         range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2393         if (!ic->meta_dev)
2394                 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2395
2396         add_new_range_and_wait(ic, &range);
2397         spin_unlock_irq(&ic->endio_wait.lock);
2398         logical_sector = range.logical_sector;
2399         n_sectors = range.n_sectors;
2400
2401         if (ic->mode == 'B') {
2402                 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2403                         goto advance_and_next;
2404                 }
2405                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2406                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2407                         logical_sector += ic->sectors_per_block;
2408                         n_sectors -= ic->sectors_per_block;
2409                         cond_resched();
2410                 }
2411                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2412                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2413                         n_sectors -= ic->sectors_per_block;
2414                         cond_resched();
2415                 }
2416                 get_area_and_offset(ic, logical_sector, &area, &offset);
2417         }
2418
2419         DEBUG_print("recalculating: %lx, %lx\n", logical_sector, n_sectors);
2420
2421         if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2422                 recalc_write_super(ic);
2423                 if (ic->mode == 'B') {
2424                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2425                 }
2426                 super_counter = 0;
2427         }
2428
2429         if (unlikely(dm_integrity_failed(ic)))
2430                 goto err;
2431
2432         io_req.bi_op = REQ_OP_READ;
2433         io_req.bi_op_flags = 0;
2434         io_req.mem.type = DM_IO_VMA;
2435         io_req.mem.ptr.addr = ic->recalc_buffer;
2436         io_req.notify.fn = NULL;
2437         io_req.client = ic->io;
2438         io_loc.bdev = ic->dev->bdev;
2439         io_loc.sector = get_data_sector(ic, area, offset);
2440         io_loc.count = n_sectors;
2441
2442         r = dm_io(&io_req, 1, &io_loc, NULL);
2443         if (unlikely(r)) {
2444                 dm_integrity_io_error(ic, "reading data", r);
2445                 goto err;
2446         }
2447
2448         t = ic->recalc_tags;
2449         for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2450                 integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2451                 t += ic->tag_size;
2452         }
2453
2454         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2455
2456         r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2457         if (unlikely(r)) {
2458                 dm_integrity_io_error(ic, "writing tags", r);
2459                 goto err;
2460         }
2461
2462 advance_and_next:
2463         cond_resched();
2464
2465         spin_lock_irq(&ic->endio_wait.lock);
2466         remove_range_unlocked(ic, &range);
2467         ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2468         goto next_chunk;
2469
2470 err:
2471         remove_range(ic, &range);
2472         return;
2473
2474 unlock_ret:
2475         spin_unlock_irq(&ic->endio_wait.lock);
2476
2477         recalc_write_super(ic);
2478 }
2479
2480 static void bitmap_block_work(struct work_struct *w)
2481 {
2482         struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2483         struct dm_integrity_c *ic = bbs->ic;
2484         struct bio *bio;
2485         struct bio_list bio_queue;
2486         struct bio_list waiting;
2487
2488         bio_list_init(&waiting);
2489
2490         spin_lock(&bbs->bio_queue_lock);
2491         bio_queue = bbs->bio_queue;
2492         bio_list_init(&bbs->bio_queue);
2493         spin_unlock(&bbs->bio_queue_lock);
2494
2495         while ((bio = bio_list_pop(&bio_queue))) {
2496                 struct dm_integrity_io *dio;
2497
2498                 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2499
2500                 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2501                                     dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2502                         remove_range(ic, &dio->range);
2503                         INIT_WORK(&dio->work, integrity_bio_wait);
2504                         queue_work(ic->wait_wq, &dio->work);
2505                 } else {
2506                         block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2507                                         dio->range.n_sectors, BITMAP_OP_SET);
2508                         bio_list_add(&waiting, bio);
2509                 }
2510         }
2511
2512         if (bio_list_empty(&waiting))
2513                 return;
2514
2515         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC,
2516                            bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2517                            BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2518
2519         while ((bio = bio_list_pop(&waiting))) {
2520                 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2521
2522                 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2523                                 dio->range.n_sectors, BITMAP_OP_SET);
2524
2525                 remove_range(ic, &dio->range);
2526                 INIT_WORK(&dio->work, integrity_bio_wait);
2527                 queue_work(ic->wait_wq, &dio->work);
2528         }
2529
2530         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2531 }
2532
2533 static void bitmap_flush_work(struct work_struct *work)
2534 {
2535         struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2536         struct dm_integrity_range range;
2537         unsigned long limit;
2538         struct bio *bio;
2539
2540         dm_integrity_flush_buffers(ic);
2541
2542         range.logical_sector = 0;
2543         range.n_sectors = ic->provided_data_sectors;
2544
2545         spin_lock_irq(&ic->endio_wait.lock);
2546         add_new_range_and_wait(ic, &range);
2547         spin_unlock_irq(&ic->endio_wait.lock);
2548
2549         dm_integrity_flush_buffers(ic);
2550         if (ic->meta_dev)
2551                 blkdev_issue_flush(ic->dev->bdev, GFP_NOIO, NULL);
2552
2553         limit = ic->provided_data_sectors;
2554         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2555                 limit = le64_to_cpu(ic->sb->recalc_sector)
2556                         >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2557                         << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2558         }
2559         /*DEBUG_print("zeroing journal\n");*/
2560         block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2561         block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2562
2563         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2564                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2565
2566         spin_lock_irq(&ic->endio_wait.lock);
2567         remove_range_unlocked(ic, &range);
2568         while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2569                 bio_endio(bio);
2570                 spin_unlock_irq(&ic->endio_wait.lock);
2571                 spin_lock_irq(&ic->endio_wait.lock);
2572         }
2573         spin_unlock_irq(&ic->endio_wait.lock);
2574 }
2575
2576
2577 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2578                          unsigned n_sections, unsigned char commit_seq)
2579 {
2580         unsigned i, j, n;
2581
2582         if (!n_sections)
2583                 return;
2584
2585         for (n = 0; n < n_sections; n++) {
2586                 i = start_section + n;
2587                 wraparound_section(ic, &i);
2588                 for (j = 0; j < ic->journal_section_sectors; j++) {
2589                         struct journal_sector *js = access_journal(ic, i, j);
2590                         memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2591                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2592                 }
2593                 for (j = 0; j < ic->journal_section_entries; j++) {
2594                         struct journal_entry *je = access_journal_entry(ic, i, j);
2595                         journal_entry_set_unused(je);
2596                 }
2597         }
2598
2599         write_journal(ic, start_section, n_sections);
2600 }
2601
2602 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2603 {
2604         unsigned char k;
2605         for (k = 0; k < N_COMMIT_IDS; k++) {
2606                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2607                         return k;
2608         }
2609         dm_integrity_io_error(ic, "journal commit id", -EIO);
2610         return -EIO;
2611 }
2612
2613 static void replay_journal(struct dm_integrity_c *ic)
2614 {
2615         unsigned i, j;
2616         bool used_commit_ids[N_COMMIT_IDS];
2617         unsigned max_commit_id_sections[N_COMMIT_IDS];
2618         unsigned write_start, write_sections;
2619         unsigned continue_section;
2620         bool journal_empty;
2621         unsigned char unused, last_used, want_commit_seq;
2622
2623         if (ic->mode == 'R')
2624                 return;
2625
2626         if (ic->journal_uptodate)
2627                 return;
2628
2629         last_used = 0;
2630         write_start = 0;
2631
2632         if (!ic->just_formatted) {
2633                 DEBUG_print("reading journal\n");
2634                 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2635                 if (ic->journal_io)
2636                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2637                 if (ic->journal_io) {
2638                         struct journal_completion crypt_comp;
2639                         crypt_comp.ic = ic;
2640                         init_completion(&crypt_comp.comp);
2641                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2642                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2643                         wait_for_completion(&crypt_comp.comp);
2644                 }
2645                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2646         }
2647
2648         if (dm_integrity_failed(ic))
2649                 goto clear_journal;
2650
2651         journal_empty = true;
2652         memset(used_commit_ids, 0, sizeof used_commit_ids);
2653         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2654         for (i = 0; i < ic->journal_sections; i++) {
2655                 for (j = 0; j < ic->journal_section_sectors; j++) {
2656                         int k;
2657                         struct journal_sector *js = access_journal(ic, i, j);
2658                         k = find_commit_seq(ic, i, j, js->commit_id);
2659                         if (k < 0)
2660                                 goto clear_journal;
2661                         used_commit_ids[k] = true;
2662                         max_commit_id_sections[k] = i;
2663                 }
2664                 if (journal_empty) {
2665                         for (j = 0; j < ic->journal_section_entries; j++) {
2666                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2667                                 if (!journal_entry_is_unused(je)) {
2668                                         journal_empty = false;
2669                                         break;
2670                                 }
2671                         }
2672                 }
2673         }
2674
2675         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2676                 unused = N_COMMIT_IDS - 1;
2677                 while (unused && !used_commit_ids[unused - 1])
2678                         unused--;
2679         } else {
2680                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2681                         if (!used_commit_ids[unused])
2682                                 break;
2683                 if (unused == N_COMMIT_IDS) {
2684                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2685                         goto clear_journal;
2686                 }
2687         }
2688         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2689                     unused, used_commit_ids[0], used_commit_ids[1],
2690                     used_commit_ids[2], used_commit_ids[3]);
2691
2692         last_used = prev_commit_seq(unused);
2693         want_commit_seq = prev_commit_seq(last_used);
2694
2695         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2696                 journal_empty = true;
2697
2698         write_start = max_commit_id_sections[last_used] + 1;
2699         if (unlikely(write_start >= ic->journal_sections))
2700                 want_commit_seq = next_commit_seq(want_commit_seq);
2701         wraparound_section(ic, &write_start);
2702
2703         i = write_start;
2704         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2705                 for (j = 0; j < ic->journal_section_sectors; j++) {
2706                         struct journal_sector *js = access_journal(ic, i, j);
2707
2708                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2709                                 /*
2710                                  * This could be caused by crash during writing.
2711                                  * We won't replay the inconsistent part of the
2712                                  * journal.
2713                                  */
2714                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2715                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2716                                 goto brk;
2717                         }
2718                 }
2719                 i++;
2720                 if (unlikely(i >= ic->journal_sections))
2721                         want_commit_seq = next_commit_seq(want_commit_seq);
2722                 wraparound_section(ic, &i);
2723         }
2724 brk:
2725
2726         if (!journal_empty) {
2727                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2728                             write_sections, write_start, want_commit_seq);
2729                 do_journal_write(ic, write_start, write_sections, true);
2730         }
2731
2732         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2733                 continue_section = write_start;
2734                 ic->commit_seq = want_commit_seq;
2735                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2736         } else {
2737                 unsigned s;
2738                 unsigned char erase_seq;
2739 clear_journal:
2740                 DEBUG_print("clearing journal\n");
2741
2742                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2743                 s = write_start;
2744                 init_journal(ic, s, 1, erase_seq);
2745                 s++;
2746                 wraparound_section(ic, &s);
2747                 if (ic->journal_sections >= 2) {
2748                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2749                         s += ic->journal_sections - 2;
2750                         wraparound_section(ic, &s);
2751                         init_journal(ic, s, 1, erase_seq);
2752                 }
2753
2754                 continue_section = 0;
2755                 ic->commit_seq = next_commit_seq(erase_seq);
2756         }
2757
2758         ic->committed_section = continue_section;
2759         ic->n_committed_sections = 0;
2760
2761         ic->uncommitted_section = continue_section;
2762         ic->n_uncommitted_sections = 0;
2763
2764         ic->free_section = continue_section;
2765         ic->free_section_entry = 0;
2766         ic->free_sectors = ic->journal_entries;
2767
2768         ic->journal_tree_root = RB_ROOT;
2769         for (i = 0; i < ic->journal_entries; i++)
2770                 init_journal_node(&ic->journal_tree[i]);
2771 }
2772
2773 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
2774 {
2775         DEBUG_print("dm_integrity_enter_synchronous_mode\n");
2776
2777         if (ic->mode == 'B') {
2778                 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
2779                 ic->synchronous_mode = 1;
2780
2781                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2782                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2783                 flush_workqueue(ic->commit_wq);
2784         }
2785 }
2786
2787 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
2788 {
2789         struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
2790
2791         DEBUG_print("dm_integrity_reboot\n");
2792
2793         dm_integrity_enter_synchronous_mode(ic);
2794
2795         return NOTIFY_DONE;
2796 }
2797
2798 static void dm_integrity_postsuspend(struct dm_target *ti)
2799 {
2800         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2801         int r;
2802
2803         WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
2804
2805         del_timer_sync(&ic->autocommit_timer);
2806
2807         WRITE_ONCE(ic->suspending, 1);
2808
2809         if (ic->recalc_wq)
2810                 drain_workqueue(ic->recalc_wq);
2811
2812         if (ic->mode == 'B')
2813                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2814
2815         queue_work(ic->commit_wq, &ic->commit_work);
2816         drain_workqueue(ic->commit_wq);
2817
2818         if (ic->mode == 'J') {
2819                 if (ic->meta_dev)
2820                         queue_work(ic->writer_wq, &ic->writer_work);
2821                 drain_workqueue(ic->writer_wq);
2822                 dm_integrity_flush_buffers(ic);
2823         }
2824
2825         if (ic->mode == 'B') {
2826                 dm_integrity_flush_buffers(ic);
2827 #if 1
2828                 /* set to 0 to test bitmap replay code */
2829                 init_journal(ic, 0, ic->journal_sections, 0);
2830                 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2831                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2832                 if (unlikely(r))
2833                         dm_integrity_io_error(ic, "writing superblock", r);
2834 #endif
2835         }
2836
2837         WRITE_ONCE(ic->suspending, 0);
2838
2839         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2840
2841         ic->journal_uptodate = true;
2842 }
2843
2844 static void dm_integrity_resume(struct dm_target *ti)
2845 {
2846         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2847         int r;
2848         DEBUG_print("resume\n");
2849
2850         if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
2851                 DEBUG_print("resume dirty_bitmap\n");
2852                 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
2853                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2854                 if (ic->mode == 'B') {
2855                         if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
2856                                 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
2857                                 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
2858                                 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
2859                                                      BITMAP_OP_TEST_ALL_CLEAR)) {
2860                                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2861                                         ic->sb->recalc_sector = cpu_to_le64(0);
2862                                 }
2863                         } else {
2864                                 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
2865                                             ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
2866                                 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
2867                                 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2868                                 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2869                                 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2870                                 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2871                                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2872                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2873                                 ic->sb->recalc_sector = cpu_to_le64(0);
2874                         }
2875                 } else {
2876                         if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
2877                               block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR))) {
2878                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2879                                 ic->sb->recalc_sector = cpu_to_le64(0);
2880                         }
2881                         init_journal(ic, 0, ic->journal_sections, 0);
2882                         replay_journal(ic);
2883                         ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2884                 }
2885                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2886                 if (unlikely(r))
2887                         dm_integrity_io_error(ic, "writing superblock", r);
2888         } else {
2889                 replay_journal(ic);
2890                 if (ic->mode == 'B') {
2891                         int mode;
2892                         ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2893                         ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
2894                         r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2895                         if (unlikely(r))
2896                                 dm_integrity_io_error(ic, "writing superblock", r);
2897
2898                         mode = ic->recalculate_flag ? BITMAP_OP_SET : BITMAP_OP_CLEAR;
2899                         block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, mode);
2900                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, mode);
2901                         block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, mode);
2902                         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2903                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2904                 }
2905         }
2906
2907         DEBUG_print("testing recalc: %x\n", ic->sb->flags);
2908         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2909                 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
2910                 DEBUG_print("recalc pos: %lx / %lx\n", (long)recalc_pos, ic->provided_data_sectors);
2911                 if (recalc_pos < ic->provided_data_sectors) {
2912                         queue_work(ic->recalc_wq, &ic->recalc_work);
2913                 } else if (recalc_pos > ic->provided_data_sectors) {
2914                         ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
2915                         recalc_write_super(ic);
2916                 }
2917         }
2918
2919         ic->reboot_notifier.notifier_call = dm_integrity_reboot;
2920         ic->reboot_notifier.next = NULL;
2921         ic->reboot_notifier.priority = INT_MAX - 1;     /* be notified after md and before hardware drivers */
2922         WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
2923
2924 #if 0
2925         /* set to 1 to stress test synchronous mode */
2926         dm_integrity_enter_synchronous_mode(ic);
2927 #endif
2928 }
2929
2930 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2931                                 unsigned status_flags, char *result, unsigned maxlen)
2932 {
2933         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2934         unsigned arg_count;
2935         size_t sz = 0;
2936
2937         switch (type) {
2938         case STATUSTYPE_INFO:
2939                 DMEMIT("%llu %llu",
2940                         (unsigned long long)atomic64_read(&ic->number_of_mismatches),
2941                         (unsigned long long)ic->provided_data_sectors);
2942                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2943                         DMEMIT(" %llu", (unsigned long long)le64_to_cpu(ic->sb->recalc_sector));
2944                 else
2945                         DMEMIT(" -");
2946                 break;
2947
2948         case STATUSTYPE_TABLE: {
2949                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2950                 watermark_percentage += ic->journal_entries / 2;
2951                 do_div(watermark_percentage, ic->journal_entries);
2952                 arg_count = 3;
2953                 arg_count += !!ic->meta_dev;
2954                 arg_count += ic->sectors_per_block != 1;
2955                 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
2956                 arg_count += ic->mode == 'J';
2957                 arg_count += ic->mode == 'J';
2958                 arg_count += ic->mode == 'B';
2959                 arg_count += ic->mode == 'B';
2960                 arg_count += !!ic->internal_hash_alg.alg_string;
2961                 arg_count += !!ic->journal_crypt_alg.alg_string;
2962                 arg_count += !!ic->journal_mac_alg.alg_string;
2963                 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
2964                 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2965                        ic->tag_size, ic->mode, arg_count);
2966                 if (ic->meta_dev)
2967                         DMEMIT(" meta_device:%s", ic->meta_dev->name);
2968                 if (ic->sectors_per_block != 1)
2969                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2970                 if (ic->recalculate_flag)
2971                         DMEMIT(" recalculate");
2972                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2973                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2974                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2975                 if (ic->mode == 'J') {
2976                         DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2977                         DMEMIT(" commit_time:%u", ic->autocommit_msec);
2978                 }
2979                 if (ic->mode == 'B') {
2980                         DMEMIT(" sectors_per_bit:%llu", (unsigned long long)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
2981                         DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
2982                 }
2983                 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
2984                         DMEMIT(" fix_padding");
2985
2986 #define EMIT_ALG(a, n)                                                  \
2987                 do {                                                    \
2988                         if (ic->a.alg_string) {                         \
2989                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
2990                                 if (ic->a.key_string)                   \
2991                                         DMEMIT(":%s", ic->a.key_string);\
2992                         }                                               \
2993                 } while (0)
2994                 EMIT_ALG(internal_hash_alg, "internal_hash");
2995                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2996                 EMIT_ALG(journal_mac_alg, "journal_mac");
2997                 break;
2998         }
2999         }
3000 }
3001
3002 static int dm_integrity_iterate_devices(struct dm_target *ti,
3003                                         iterate_devices_callout_fn fn, void *data)
3004 {
3005         struct dm_integrity_c *ic = ti->private;
3006
3007         if (!ic->meta_dev)
3008                 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3009         else
3010                 return fn(ti, ic->dev, 0, ti->len, data);
3011 }
3012
3013 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3014 {
3015         struct dm_integrity_c *ic = ti->private;
3016
3017         if (ic->sectors_per_block > 1) {
3018                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3019                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3020                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3021         }
3022 }
3023
3024 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3025 {
3026         unsigned sector_space = JOURNAL_SECTOR_DATA;
3027
3028         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3029         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3030                                          JOURNAL_ENTRY_ROUNDUP);
3031
3032         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3033                 sector_space -= JOURNAL_MAC_PER_SECTOR;
3034         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3035         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3036         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3037         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3038 }
3039
3040 static int calculate_device_limits(struct dm_integrity_c *ic)
3041 {
3042         __u64 initial_sectors;
3043
3044         calculate_journal_section_size(ic);
3045         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3046         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3047                 return -EINVAL;
3048         ic->initial_sectors = initial_sectors;
3049
3050         if (!ic->meta_dev) {
3051                 sector_t last_sector, last_area, last_offset;
3052
3053                 /* we have to maintain excessive padding for compatibility with existing volumes */
3054                 __u64 metadata_run_padding =
3055                         ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3056                         (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3057                         (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3058
3059                 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3060                                             metadata_run_padding) >> SECTOR_SHIFT;
3061                 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3062                         ic->log2_metadata_run = __ffs(ic->metadata_run);
3063                 else
3064                         ic->log2_metadata_run = -1;
3065
3066                 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3067                 last_sector = get_data_sector(ic, last_area, last_offset);
3068                 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3069                         return -EINVAL;
3070         } else {
3071                 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3072                 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3073                                 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3074                 meta_size <<= ic->log2_buffer_sectors;
3075                 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3076                     ic->initial_sectors + meta_size > ic->meta_device_sectors)
3077                         return -EINVAL;
3078                 ic->metadata_run = 1;
3079                 ic->log2_metadata_run = 0;
3080         }
3081
3082         return 0;
3083 }
3084
3085 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3086 {
3087         unsigned journal_sections;
3088         int test_bit;
3089
3090         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3091         memcpy(ic->sb->magic, SB_MAGIC, 8);
3092         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3093         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3094         if (ic->journal_mac_alg.alg_string)
3095                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3096
3097         calculate_journal_section_size(ic);
3098         journal_sections = journal_sectors / ic->journal_section_sectors;
3099         if (!journal_sections)
3100                 journal_sections = 1;
3101
3102         if (!ic->meta_dev) {
3103                 if (ic->fix_padding)
3104                         ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3105                 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3106                 if (!interleave_sectors)
3107                         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3108                 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3109                 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3110                 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3111
3112                 ic->provided_data_sectors = 0;
3113                 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3114                         __u64 prev_data_sectors = ic->provided_data_sectors;
3115
3116                         ic->provided_data_sectors |= (sector_t)1 << test_bit;
3117                         if (calculate_device_limits(ic))
3118                                 ic->provided_data_sectors = prev_data_sectors;
3119                 }
3120                 if (!ic->provided_data_sectors)
3121                         return -EINVAL;
3122         } else {
3123                 ic->sb->log2_interleave_sectors = 0;
3124                 ic->provided_data_sectors = ic->data_device_sectors;
3125                 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3126
3127 try_smaller_buffer:
3128                 ic->sb->journal_sections = cpu_to_le32(0);
3129                 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3130                         __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3131                         __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3132                         if (test_journal_sections > journal_sections)
3133                                 continue;
3134                         ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3135                         if (calculate_device_limits(ic))
3136                                 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3137
3138                 }
3139                 if (!le32_to_cpu(ic->sb->journal_sections)) {
3140                         if (ic->log2_buffer_sectors > 3) {
3141                                 ic->log2_buffer_sectors--;
3142                                 goto try_smaller_buffer;
3143                         }
3144                         return -EINVAL;
3145                 }
3146         }
3147
3148         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3149
3150         sb_set_version(ic);
3151
3152         return 0;
3153 }
3154
3155 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3156 {
3157         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3158         struct blk_integrity bi;
3159
3160         memset(&bi, 0, sizeof(bi));
3161         bi.profile = &dm_integrity_profile;
3162         bi.tuple_size = ic->tag_size;
3163         bi.tag_size = bi.tuple_size;
3164         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3165
3166         blk_integrity_register(disk, &bi);
3167         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3168 }
3169
3170 static void dm_integrity_free_page_list(struct page_list *pl)
3171 {
3172         unsigned i;
3173
3174         if (!pl)
3175                 return;
3176         for (i = 0; pl[i].page; i++)
3177                 __free_page(pl[i].page);
3178         kvfree(pl);
3179 }
3180
3181 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3182 {
3183         struct page_list *pl;
3184         unsigned i;
3185
3186         pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3187         if (!pl)
3188                 return NULL;
3189
3190         for (i = 0; i < n_pages; i++) {
3191                 pl[i].page = alloc_page(GFP_KERNEL);
3192                 if (!pl[i].page) {
3193                         dm_integrity_free_page_list(pl);
3194                         return NULL;
3195                 }
3196                 if (i)
3197                         pl[i - 1].next = &pl[i];
3198         }
3199         pl[i].page = NULL;
3200         pl[i].next = NULL;
3201
3202         return pl;
3203 }
3204
3205 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3206 {
3207         unsigned i;
3208         for (i = 0; i < ic->journal_sections; i++)
3209                 kvfree(sl[i]);
3210         kvfree(sl);
3211 }
3212
3213 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3214                                                                    struct page_list *pl)
3215 {
3216         struct scatterlist **sl;
3217         unsigned i;
3218
3219         sl = kvmalloc_array(ic->journal_sections,
3220                             sizeof(struct scatterlist *),
3221                             GFP_KERNEL | __GFP_ZERO);
3222         if (!sl)
3223                 return NULL;
3224
3225         for (i = 0; i < ic->journal_sections; i++) {
3226                 struct scatterlist *s;
3227                 unsigned start_index, start_offset;
3228                 unsigned end_index, end_offset;
3229                 unsigned n_pages;
3230                 unsigned idx;
3231
3232                 page_list_location(ic, i, 0, &start_index, &start_offset);
3233                 page_list_location(ic, i, ic->journal_section_sectors - 1,
3234                                    &end_index, &end_offset);
3235
3236                 n_pages = (end_index - start_index + 1);
3237
3238                 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3239                                    GFP_KERNEL);
3240                 if (!s) {
3241                         dm_integrity_free_journal_scatterlist(ic, sl);
3242                         return NULL;
3243                 }
3244
3245                 sg_init_table(s, n_pages);
3246                 for (idx = start_index; idx <= end_index; idx++) {
3247                         char *va = lowmem_page_address(pl[idx].page);
3248                         unsigned start = 0, end = PAGE_SIZE;
3249                         if (idx == start_index)
3250                                 start = start_offset;
3251                         if (idx == end_index)
3252                                 end = end_offset + (1 << SECTOR_SHIFT);
3253                         sg_set_buf(&s[idx - start_index], va + start, end - start);
3254                 }
3255
3256                 sl[i] = s;
3257         }
3258
3259         return sl;
3260 }
3261
3262 static void free_alg(struct alg_spec *a)
3263 {
3264         kzfree(a->alg_string);
3265         kzfree(a->key);
3266         memset(a, 0, sizeof *a);
3267 }
3268
3269 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3270 {
3271         char *k;
3272
3273         free_alg(a);
3274
3275         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3276         if (!a->alg_string)
3277                 goto nomem;
3278
3279         k = strchr(a->alg_string, ':');
3280         if (k) {
3281                 *k = 0;
3282                 a->key_string = k + 1;
3283                 if (strlen(a->key_string) & 1)
3284                         goto inval;
3285
3286                 a->key_size = strlen(a->key_string) / 2;
3287                 a->key = kmalloc(a->key_size, GFP_KERNEL);
3288                 if (!a->key)
3289                         goto nomem;
3290                 if (hex2bin(a->key, a->key_string, a->key_size))
3291                         goto inval;
3292         }
3293
3294         return 0;
3295 inval:
3296         *error = error_inval;
3297         return -EINVAL;
3298 nomem:
3299         *error = "Out of memory for an argument";
3300         return -ENOMEM;
3301 }
3302
3303 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3304                    char *error_alg, char *error_key)
3305 {
3306         int r;
3307
3308         if (a->alg_string) {
3309                 *hash = crypto_alloc_shash(a->alg_string, 0, 0);
3310                 if (IS_ERR(*hash)) {
3311                         *error = error_alg;
3312                         r = PTR_ERR(*hash);
3313                         *hash = NULL;
3314                         return r;
3315                 }
3316
3317                 if (a->key) {
3318                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
3319                         if (r) {
3320                                 *error = error_key;
3321                                 return r;
3322                         }
3323                 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3324                         *error = error_key;
3325                         return -ENOKEY;
3326                 }
3327         }
3328
3329         return 0;
3330 }
3331
3332 static int create_journal(struct dm_integrity_c *ic, char **error)
3333 {
3334         int r = 0;
3335         unsigned i;
3336         __u64 journal_pages, journal_desc_size, journal_tree_size;
3337         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3338         struct skcipher_request *req = NULL;
3339
3340         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3341         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3342         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3343         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3344
3345         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3346                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3347         journal_desc_size = journal_pages * sizeof(struct page_list);
3348         if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3349                 *error = "Journal doesn't fit into memory";
3350                 r = -ENOMEM;
3351                 goto bad;
3352         }
3353         ic->journal_pages = journal_pages;
3354
3355         ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3356         if (!ic->journal) {
3357                 *error = "Could not allocate memory for journal";
3358                 r = -ENOMEM;
3359                 goto bad;
3360         }
3361         if (ic->journal_crypt_alg.alg_string) {
3362                 unsigned ivsize, blocksize;
3363                 struct journal_completion comp;
3364
3365                 comp.ic = ic;
3366                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
3367                 if (IS_ERR(ic->journal_crypt)) {
3368                         *error = "Invalid journal cipher";
3369                         r = PTR_ERR(ic->journal_crypt);
3370                         ic->journal_crypt = NULL;
3371                         goto bad;
3372                 }
3373                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3374                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3375
3376                 if (ic->journal_crypt_alg.key) {
3377                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3378                                                    ic->journal_crypt_alg.key_size);
3379                         if (r) {
3380                                 *error = "Error setting encryption key";
3381                                 goto bad;
3382                         }
3383                 }
3384                 DEBUG_print("cipher %s, block size %u iv size %u\n",
3385                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3386
3387                 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3388                 if (!ic->journal_io) {
3389                         *error = "Could not allocate memory for journal io";
3390                         r = -ENOMEM;
3391                         goto bad;
3392                 }
3393
3394                 if (blocksize == 1) {
3395                         struct scatterlist *sg;
3396
3397                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3398                         if (!req) {
3399                                 *error = "Could not allocate crypt request";
3400                                 r = -ENOMEM;
3401                                 goto bad;
3402                         }
3403
3404                         crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3405                         if (!crypt_iv) {
3406                                 *error = "Could not allocate iv";
3407                                 r = -ENOMEM;
3408                                 goto bad;
3409                         }
3410
3411                         ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3412                         if (!ic->journal_xor) {
3413                                 *error = "Could not allocate memory for journal xor";
3414                                 r = -ENOMEM;
3415                                 goto bad;
3416                         }
3417
3418                         sg = kvmalloc_array(ic->journal_pages + 1,
3419                                             sizeof(struct scatterlist),
3420                                             GFP_KERNEL);
3421                         if (!sg) {
3422                                 *error = "Unable to allocate sg list";
3423                                 r = -ENOMEM;
3424                                 goto bad;
3425                         }
3426                         sg_init_table(sg, ic->journal_pages + 1);
3427                         for (i = 0; i < ic->journal_pages; i++) {
3428                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
3429                                 clear_page(va);
3430                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
3431                         }
3432                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3433
3434                         skcipher_request_set_crypt(req, sg, sg,
3435                                                    PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3436                         init_completion(&comp.comp);
3437                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3438                         if (do_crypt(true, req, &comp))
3439                                 wait_for_completion(&comp.comp);
3440                         kvfree(sg);
3441                         r = dm_integrity_failed(ic);
3442                         if (r) {
3443                                 *error = "Unable to encrypt journal";
3444                                 goto bad;
3445                         }
3446                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3447
3448                         crypto_free_skcipher(ic->journal_crypt);
3449                         ic->journal_crypt = NULL;
3450                 } else {
3451                         unsigned crypt_len = roundup(ivsize, blocksize);
3452
3453                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3454                         if (!req) {
3455                                 *error = "Could not allocate crypt request";
3456                                 r = -ENOMEM;
3457                                 goto bad;
3458                         }
3459
3460                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3461                         if (!crypt_iv) {
3462                                 *error = "Could not allocate iv";
3463                                 r = -ENOMEM;
3464                                 goto bad;
3465                         }
3466
3467                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3468                         if (!crypt_data) {
3469                                 *error = "Unable to allocate crypt data";
3470                                 r = -ENOMEM;
3471                                 goto bad;
3472                         }
3473
3474                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3475                         if (!ic->journal_scatterlist) {
3476                                 *error = "Unable to allocate sg list";
3477                                 r = -ENOMEM;
3478                                 goto bad;
3479                         }
3480                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3481                         if (!ic->journal_io_scatterlist) {
3482                                 *error = "Unable to allocate sg list";
3483                                 r = -ENOMEM;
3484                                 goto bad;
3485                         }
3486                         ic->sk_requests = kvmalloc_array(ic->journal_sections,
3487                                                          sizeof(struct skcipher_request *),
3488                                                          GFP_KERNEL | __GFP_ZERO);
3489                         if (!ic->sk_requests) {
3490                                 *error = "Unable to allocate sk requests";
3491                                 r = -ENOMEM;
3492                                 goto bad;
3493                         }
3494                         for (i = 0; i < ic->journal_sections; i++) {
3495                                 struct scatterlist sg;
3496                                 struct skcipher_request *section_req;
3497                                 __u32 section_le = cpu_to_le32(i);
3498
3499                                 memset(crypt_iv, 0x00, ivsize);
3500                                 memset(crypt_data, 0x00, crypt_len);
3501                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
3502
3503                                 sg_init_one(&sg, crypt_data, crypt_len);
3504                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3505                                 init_completion(&comp.comp);
3506                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3507                                 if (do_crypt(true, req, &comp))
3508                                         wait_for_completion(&comp.comp);
3509
3510                                 r = dm_integrity_failed(ic);
3511                                 if (r) {
3512                                         *error = "Unable to generate iv";
3513                                         goto bad;
3514                                 }
3515
3516                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3517                                 if (!section_req) {
3518                                         *error = "Unable to allocate crypt request";
3519                                         r = -ENOMEM;
3520                                         goto bad;
3521                                 }
3522                                 section_req->iv = kmalloc_array(ivsize, 2,
3523                                                                 GFP_KERNEL);
3524                                 if (!section_req->iv) {
3525                                         skcipher_request_free(section_req);
3526                                         *error = "Unable to allocate iv";
3527                                         r = -ENOMEM;
3528                                         goto bad;
3529                                 }
3530                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3531                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3532                                 ic->sk_requests[i] = section_req;
3533                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3534                         }
3535                 }
3536         }
3537
3538         for (i = 0; i < N_COMMIT_IDS; i++) {
3539                 unsigned j;
3540 retest_commit_id:
3541                 for (j = 0; j < i; j++) {
3542                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
3543                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3544                                 goto retest_commit_id;
3545                         }
3546                 }
3547                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3548         }
3549
3550         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3551         if (journal_tree_size > ULONG_MAX) {
3552                 *error = "Journal doesn't fit into memory";
3553                 r = -ENOMEM;
3554                 goto bad;
3555         }
3556         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3557         if (!ic->journal_tree) {
3558                 *error = "Could not allocate memory for journal tree";
3559                 r = -ENOMEM;
3560         }
3561 bad:
3562         kfree(crypt_data);
3563         kfree(crypt_iv);
3564         skcipher_request_free(req);
3565
3566         return r;
3567 }
3568
3569 /*
3570  * Construct a integrity mapping
3571  *
3572  * Arguments:
3573  *      device
3574  *      offset from the start of the device
3575  *      tag size
3576  *      D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3577  *      number of optional arguments
3578  *      optional arguments:
3579  *              journal_sectors
3580  *              interleave_sectors
3581  *              buffer_sectors
3582  *              journal_watermark
3583  *              commit_time
3584  *              meta_device
3585  *              block_size
3586  *              sectors_per_bit
3587  *              bitmap_flush_interval
3588  *              internal_hash
3589  *              journal_crypt
3590  *              journal_mac
3591  *              recalculate
3592  */
3593 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3594 {
3595         struct dm_integrity_c *ic;
3596         char dummy;
3597         int r;
3598         unsigned extra_args;
3599         struct dm_arg_set as;
3600         static const struct dm_arg _args[] = {
3601                 {0, 9, "Invalid number of feature args"},
3602         };
3603         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3604         bool should_write_sb;
3605         __u64 threshold;
3606         unsigned long long start;
3607         __s8 log2_sectors_per_bitmap_bit = -1;
3608         __s8 log2_blocks_per_bitmap_bit;
3609         __u64 bits_in_journal;
3610         __u64 n_bitmap_bits;
3611
3612 #define DIRECT_ARGUMENTS        4
3613
3614         if (argc <= DIRECT_ARGUMENTS) {
3615                 ti->error = "Invalid argument count";
3616                 return -EINVAL;
3617         }
3618
3619         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3620         if (!ic) {
3621                 ti->error = "Cannot allocate integrity context";
3622                 return -ENOMEM;
3623         }
3624         ti->private = ic;
3625         ti->per_io_data_size = sizeof(struct dm_integrity_io);
3626
3627         ic->in_progress = RB_ROOT;
3628         INIT_LIST_HEAD(&ic->wait_list);
3629         init_waitqueue_head(&ic->endio_wait);
3630         bio_list_init(&ic->flush_bio_list);
3631         init_waitqueue_head(&ic->copy_to_journal_wait);
3632         init_completion(&ic->crypto_backoff);
3633         atomic64_set(&ic->number_of_mismatches, 0);
3634         ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
3635
3636         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3637         if (r) {
3638                 ti->error = "Device lookup failed";
3639                 goto bad;
3640         }
3641
3642         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3643                 ti->error = "Invalid starting offset";
3644                 r = -EINVAL;
3645                 goto bad;
3646         }
3647         ic->start = start;
3648
3649         if (strcmp(argv[2], "-")) {
3650                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3651                         ti->error = "Invalid tag size";
3652                         r = -EINVAL;
3653                         goto bad;
3654                 }
3655         }
3656
3657         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
3658             !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
3659                 ic->mode = argv[3][0];
3660         } else {
3661                 ti->error = "Invalid mode (expecting J, B, D, R)";
3662                 r = -EINVAL;
3663                 goto bad;
3664         }
3665
3666         journal_sectors = 0;
3667         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3668         buffer_sectors = DEFAULT_BUFFER_SECTORS;
3669         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3670         sync_msec = DEFAULT_SYNC_MSEC;
3671         ic->sectors_per_block = 1;
3672
3673         as.argc = argc - DIRECT_ARGUMENTS;
3674         as.argv = argv + DIRECT_ARGUMENTS;
3675         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3676         if (r)
3677                 goto bad;
3678
3679         while (extra_args--) {
3680                 const char *opt_string;
3681                 unsigned val;
3682                 unsigned long long llval;
3683                 opt_string = dm_shift_arg(&as);
3684                 if (!opt_string) {
3685                         r = -EINVAL;
3686                         ti->error = "Not enough feature arguments";
3687                         goto bad;
3688                 }
3689                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3690                         journal_sectors = val ? val : 1;
3691                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3692                         interleave_sectors = val;
3693                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3694                         buffer_sectors = val;
3695                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3696                         journal_watermark = val;
3697                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3698                         sync_msec = val;
3699                 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3700                         if (ic->meta_dev) {
3701                                 dm_put_device(ti, ic->meta_dev);
3702                                 ic->meta_dev = NULL;
3703                         }
3704                         r = dm_get_device(ti, strchr(opt_string, ':') + 1,
3705                                           dm_table_get_mode(ti->table), &ic->meta_dev);
3706                         if (r) {
3707                                 ti->error = "Device lookup failed";
3708                                 goto bad;
3709                         }
3710                 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3711                         if (val < 1 << SECTOR_SHIFT ||
3712                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3713                             (val & (val -1))) {
3714                                 r = -EINVAL;
3715                                 ti->error = "Invalid block_size argument";
3716                                 goto bad;
3717                         }
3718                         ic->sectors_per_block = val >> SECTOR_SHIFT;
3719                 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
3720                         log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
3721                 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
3722                         if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
3723                                 r = -EINVAL;
3724                                 ti->error = "Invalid bitmap_flush_interval argument";
3725                         }
3726                         ic->bitmap_flush_interval = msecs_to_jiffies(val);
3727                 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3728                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3729                                             "Invalid internal_hash argument");
3730                         if (r)
3731                                 goto bad;
3732                 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3733                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3734                                             "Invalid journal_crypt argument");
3735                         if (r)
3736                                 goto bad;
3737                 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3738                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
3739                                             "Invalid journal_mac argument");
3740                         if (r)
3741                                 goto bad;
3742                 } else if (!strcmp(opt_string, "recalculate")) {
3743                         ic->recalculate_flag = true;
3744                 } else if (!strcmp(opt_string, "fix_padding")) {
3745                         ic->fix_padding = true;
3746                 } else {
3747                         r = -EINVAL;
3748                         ti->error = "Invalid argument";
3749                         goto bad;
3750                 }
3751         }
3752
3753         ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3754         if (!ic->meta_dev)
3755                 ic->meta_device_sectors = ic->data_device_sectors;
3756         else
3757                 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3758
3759         if (!journal_sectors) {
3760                 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3761                                       ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3762         }
3763
3764         if (!buffer_sectors)
3765                 buffer_sectors = 1;
3766         ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3767
3768         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3769                     "Invalid internal hash", "Error setting internal hash key");
3770         if (r)
3771                 goto bad;
3772
3773         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3774                     "Invalid journal mac", "Error setting journal mac key");
3775         if (r)
3776                 goto bad;
3777
3778         if (!ic->tag_size) {
3779                 if (!ic->internal_hash) {
3780                         ti->error = "Unknown tag size";
3781                         r = -EINVAL;
3782                         goto bad;
3783                 }
3784                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
3785         }
3786         if (ic->tag_size > MAX_TAG_SIZE) {
3787                 ti->error = "Too big tag size";
3788                 r = -EINVAL;
3789                 goto bad;
3790         }
3791         if (!(ic->tag_size & (ic->tag_size - 1)))
3792                 ic->log2_tag_size = __ffs(ic->tag_size);
3793         else
3794                 ic->log2_tag_size = -1;
3795
3796         if (ic->mode == 'B' && !ic->internal_hash) {
3797                 r = -EINVAL;
3798                 ti->error = "Bitmap mode can be only used with internal hash";
3799                 goto bad;
3800         }
3801
3802         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
3803         ic->autocommit_msec = sync_msec;
3804         timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
3805
3806         ic->io = dm_io_client_create();
3807         if (IS_ERR(ic->io)) {
3808                 r = PTR_ERR(ic->io);
3809                 ic->io = NULL;
3810                 ti->error = "Cannot allocate dm io";
3811                 goto bad;
3812         }
3813
3814         r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
3815         if (r) {
3816                 ti->error = "Cannot allocate mempool";
3817                 goto bad;
3818         }
3819
3820         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
3821                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
3822         if (!ic->metadata_wq) {
3823                 ti->error = "Cannot allocate workqueue";
3824                 r = -ENOMEM;
3825                 goto bad;
3826         }
3827
3828         /*
3829          * If this workqueue were percpu, it would cause bio reordering
3830          * and reduced performance.
3831          */
3832         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3833         if (!ic->wait_wq) {
3834                 ti->error = "Cannot allocate workqueue";
3835                 r = -ENOMEM;
3836                 goto bad;
3837         }
3838
3839         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
3840         if (!ic->commit_wq) {
3841                 ti->error = "Cannot allocate workqueue";
3842                 r = -ENOMEM;
3843                 goto bad;
3844         }
3845         INIT_WORK(&ic->commit_work, integrity_commit);
3846
3847         if (ic->mode == 'J' || ic->mode == 'B') {
3848                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
3849                 if (!ic->writer_wq) {
3850                         ti->error = "Cannot allocate workqueue";
3851                         r = -ENOMEM;
3852                         goto bad;
3853                 }
3854                 INIT_WORK(&ic->writer_work, integrity_writer);
3855         }
3856
3857         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
3858         if (!ic->sb) {
3859                 r = -ENOMEM;
3860                 ti->error = "Cannot allocate superblock area";
3861                 goto bad;
3862         }
3863
3864         r = sync_rw_sb(ic, REQ_OP_READ, 0);
3865         if (r) {
3866                 ti->error = "Error reading superblock";
3867                 goto bad;
3868         }
3869         should_write_sb = false;
3870         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3871                 if (ic->mode != 'R') {
3872                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3873                                 r = -EINVAL;
3874                                 ti->error = "The device is not initialized";
3875                                 goto bad;
3876                         }
3877                 }
3878
3879                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3880                 if (r) {
3881                         ti->error = "Could not initialize superblock";
3882                         goto bad;
3883                 }
3884                 if (ic->mode != 'R')
3885                         should_write_sb = true;
3886         }
3887
3888         if (!ic->sb->version || ic->sb->version > SB_VERSION_4) {
3889                 r = -EINVAL;
3890                 ti->error = "Unknown version";
3891                 goto bad;
3892         }
3893         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3894                 r = -EINVAL;
3895                 ti->error = "Tag size doesn't match the information in superblock";
3896                 goto bad;
3897         }
3898         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3899                 r = -EINVAL;
3900                 ti->error = "Block size doesn't match the information in superblock";
3901                 goto bad;
3902         }
3903         if (!le32_to_cpu(ic->sb->journal_sections)) {
3904                 r = -EINVAL;
3905                 ti->error = "Corrupted superblock, journal_sections is 0";
3906                 goto bad;
3907         }
3908         /* make sure that ti->max_io_len doesn't overflow */
3909         if (!ic->meta_dev) {
3910                 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3911                     ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3912                         r = -EINVAL;
3913                         ti->error = "Invalid interleave_sectors in the superblock";
3914                         goto bad;
3915                 }
3916         } else {
3917                 if (ic->sb->log2_interleave_sectors) {
3918                         r = -EINVAL;
3919                         ti->error = "Invalid interleave_sectors in the superblock";
3920                         goto bad;
3921                 }
3922         }
3923         ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3924         if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3925                 /* test for overflow */
3926                 r = -EINVAL;
3927                 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3928                 goto bad;
3929         }
3930         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3931                 r = -EINVAL;
3932                 ti->error = "Journal mac mismatch";
3933                 goto bad;
3934         }
3935
3936 try_smaller_buffer:
3937         r = calculate_device_limits(ic);
3938         if (r) {
3939                 if (ic->meta_dev) {
3940                         if (ic->log2_buffer_sectors > 3) {
3941                                 ic->log2_buffer_sectors--;
3942                                 goto try_smaller_buffer;
3943                         }
3944                 }
3945                 ti->error = "The device is too small";
3946                 goto bad;
3947         }
3948
3949         if (log2_sectors_per_bitmap_bit < 0)
3950                 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
3951         if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
3952                 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
3953
3954         bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
3955         if (bits_in_journal > UINT_MAX)
3956                 bits_in_journal = UINT_MAX;
3957         while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
3958                 log2_sectors_per_bitmap_bit++;
3959
3960         log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
3961         ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
3962         if (should_write_sb) {
3963                 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
3964         }
3965         n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
3966                                 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
3967         ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
3968
3969         if (!ic->meta_dev)
3970                 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
3971
3972         if (ti->len > ic->provided_data_sectors) {
3973                 r = -EINVAL;
3974                 ti->error = "Not enough provided sectors for requested mapping size";
3975                 goto bad;
3976         }
3977
3978
3979         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3980         threshold += 50;
3981         do_div(threshold, 100);
3982         ic->free_sectors_threshold = threshold;
3983
3984         DEBUG_print("initialized:\n");
3985         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3986         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
3987         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3988         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
3989         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
3990         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3991         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
3992         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3993         DEBUG_print("   data_device_sectors 0x%llx\n", i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT);
3994         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
3995         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
3996         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
3997         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
3998                     (unsigned long long)ic->provided_data_sectors);
3999         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4000         DEBUG_print("   bits_in_journal %llu\n", (unsigned long long)bits_in_journal);
4001
4002         if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4003                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4004                 ic->sb->recalc_sector = cpu_to_le64(0);
4005         }
4006
4007         if (ic->internal_hash) {
4008                 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4009                 if (!ic->recalc_wq ) {
4010                         ti->error = "Cannot allocate workqueue";
4011                         r = -ENOMEM;
4012                         goto bad;
4013                 }
4014                 INIT_WORK(&ic->recalc_work, integrity_recalc);
4015                 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4016                 if (!ic->recalc_buffer) {
4017                         ti->error = "Cannot allocate buffer for recalculating";
4018                         r = -ENOMEM;
4019                         goto bad;
4020                 }
4021                 ic->recalc_tags = kvmalloc_array(RECALC_SECTORS >> ic->sb->log2_sectors_per_block,
4022                                                  ic->tag_size, GFP_KERNEL);
4023                 if (!ic->recalc_tags) {
4024                         ti->error = "Cannot allocate tags for recalculating";
4025                         r = -ENOMEM;
4026                         goto bad;
4027                 }
4028         }
4029
4030         ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4031                         1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
4032         if (IS_ERR(ic->bufio)) {
4033                 r = PTR_ERR(ic->bufio);
4034                 ti->error = "Cannot initialize dm-bufio";
4035                 ic->bufio = NULL;
4036                 goto bad;
4037         }
4038         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4039
4040         if (ic->mode != 'R') {
4041                 r = create_journal(ic, &ti->error);
4042                 if (r)
4043                         goto bad;
4044
4045         }
4046
4047         if (ic->mode == 'B') {
4048                 unsigned i;
4049                 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4050
4051                 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4052                 if (!ic->recalc_bitmap) {
4053                         r = -ENOMEM;
4054                         goto bad;
4055                 }
4056                 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4057                 if (!ic->may_write_bitmap) {
4058                         r = -ENOMEM;
4059                         goto bad;
4060                 }
4061                 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4062                 if (!ic->bbs) {
4063                         r = -ENOMEM;
4064                         goto bad;
4065                 }
4066                 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4067                 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4068                         struct bitmap_block_status *bbs = &ic->bbs[i];
4069                         unsigned sector, pl_index, pl_offset;
4070
4071                         INIT_WORK(&bbs->work, bitmap_block_work);
4072                         bbs->ic = ic;
4073                         bbs->idx = i;
4074                         bio_list_init(&bbs->bio_queue);
4075                         spin_lock_init(&bbs->bio_queue_lock);
4076
4077                         sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4078                         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4079                         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4080
4081                         bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4082                 }
4083         }
4084
4085         if (should_write_sb) {
4086                 int r;
4087
4088                 init_journal(ic, 0, ic->journal_sections, 0);
4089                 r = dm_integrity_failed(ic);
4090                 if (unlikely(r)) {
4091                         ti->error = "Error initializing journal";
4092                         goto bad;
4093                 }
4094                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4095                 if (r) {
4096                         ti->error = "Error initializing superblock";
4097                         goto bad;
4098                 }
4099                 ic->just_formatted = true;
4100         }
4101
4102         if (!ic->meta_dev) {
4103                 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4104                 if (r)
4105                         goto bad;
4106         }
4107         if (ic->mode == 'B') {
4108                 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4109                 if (!max_io_len)
4110                         max_io_len = 1U << 31;
4111                 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4112                 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4113                         r = dm_set_target_max_io_len(ti, max_io_len);
4114                         if (r)
4115                                 goto bad;
4116                 }
4117         }
4118
4119         if (!ic->internal_hash)
4120                 dm_integrity_set(ti, ic);
4121
4122         ti->num_flush_bios = 1;
4123         ti->flush_supported = true;
4124
4125         return 0;
4126
4127 bad:
4128         dm_integrity_dtr(ti);
4129         return r;
4130 }
4131
4132 static void dm_integrity_dtr(struct dm_target *ti)
4133 {
4134         struct dm_integrity_c *ic = ti->private;
4135
4136         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4137         BUG_ON(!list_empty(&ic->wait_list));
4138
4139         if (ic->metadata_wq)
4140                 destroy_workqueue(ic->metadata_wq);
4141         if (ic->wait_wq)
4142                 destroy_workqueue(ic->wait_wq);
4143         if (ic->commit_wq)
4144                 destroy_workqueue(ic->commit_wq);
4145         if (ic->writer_wq)
4146                 destroy_workqueue(ic->writer_wq);
4147         if (ic->recalc_wq)
4148                 destroy_workqueue(ic->recalc_wq);
4149         vfree(ic->recalc_buffer);
4150         kvfree(ic->recalc_tags);
4151         kvfree(ic->bbs);
4152         if (ic->bufio)
4153                 dm_bufio_client_destroy(ic->bufio);
4154         mempool_exit(&ic->journal_io_mempool);
4155         if (ic->io)
4156                 dm_io_client_destroy(ic->io);
4157         if (ic->dev)
4158                 dm_put_device(ti, ic->dev);
4159         if (ic->meta_dev)
4160                 dm_put_device(ti, ic->meta_dev);
4161         dm_integrity_free_page_list(ic->journal);
4162         dm_integrity_free_page_list(ic->journal_io);
4163         dm_integrity_free_page_list(ic->journal_xor);
4164         dm_integrity_free_page_list(ic->recalc_bitmap);
4165         dm_integrity_free_page_list(ic->may_write_bitmap);
4166         if (ic->journal_scatterlist)
4167                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4168         if (ic->journal_io_scatterlist)
4169                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4170         if (ic->sk_requests) {
4171                 unsigned i;
4172
4173                 for (i = 0; i < ic->journal_sections; i++) {
4174                         struct skcipher_request *req = ic->sk_requests[i];
4175                         if (req) {
4176                                 kzfree(req->iv);
4177                                 skcipher_request_free(req);
4178                         }
4179                 }
4180                 kvfree(ic->sk_requests);
4181         }
4182         kvfree(ic->journal_tree);
4183         if (ic->sb)
4184                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4185
4186         if (ic->internal_hash)
4187                 crypto_free_shash(ic->internal_hash);
4188         free_alg(&ic->internal_hash_alg);
4189
4190         if (ic->journal_crypt)
4191                 crypto_free_skcipher(ic->journal_crypt);
4192         free_alg(&ic->journal_crypt_alg);
4193
4194         if (ic->journal_mac)
4195                 crypto_free_shash(ic->journal_mac);
4196         free_alg(&ic->journal_mac_alg);
4197
4198         kfree(ic);
4199 }
4200
4201 static struct target_type integrity_target = {
4202         .name                   = "integrity",
4203         .version                = {1, 4, 0},
4204         .module                 = THIS_MODULE,
4205         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4206         .ctr                    = dm_integrity_ctr,
4207         .dtr                    = dm_integrity_dtr,
4208         .map                    = dm_integrity_map,
4209         .postsuspend            = dm_integrity_postsuspend,
4210         .resume                 = dm_integrity_resume,
4211         .status                 = dm_integrity_status,
4212         .iterate_devices        = dm_integrity_iterate_devices,
4213         .io_hints               = dm_integrity_io_hints,
4214 };
4215
4216 static int __init dm_integrity_init(void)
4217 {
4218         int r;
4219
4220         journal_io_cache = kmem_cache_create("integrity_journal_io",
4221                                              sizeof(struct journal_io), 0, 0, NULL);
4222         if (!journal_io_cache) {
4223                 DMERR("can't allocate journal io cache");
4224                 return -ENOMEM;
4225         }
4226
4227         r = dm_register_target(&integrity_target);
4228
4229         if (r < 0)
4230                 DMERR("register failed %d", r);
4231
4232         return r;
4233 }
4234
4235 static void __exit dm_integrity_exit(void)
4236 {
4237         dm_unregister_target(&integrity_target);
4238         kmem_cache_destroy(journal_io_cache);
4239 }
4240
4241 module_init(dm_integrity_init);
4242 module_exit(dm_integrity_exit);
4243
4244 MODULE_AUTHOR("Milan Broz");
4245 MODULE_AUTHOR("Mikulas Patocka");
4246 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4247 MODULE_LICENSE("GPL");