drivers/md/dm-integrity.c

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