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[0];
  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))) {
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                 sector_t save_metadata_block;
1557                 unsigned save_metadata_offset;
1558
1559                 if (unlikely(ic->mode == 'R'))
1560                         goto skip_io;
1561
1562                 if (likely(dio->op != REQ_OP_DISCARD))
1563                         checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1564                                             GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1565                 else
1566                         checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1567                 if (!checksums) {
1568                         checksums = checksums_onstack;
1569                         if (WARN_ON(extra_space &&
1570                                     digest_size > sizeof(checksums_onstack))) {
1571                                 r = -EINVAL;
1572                                 goto error;
1573                         }
1574                 }
1575
1576                 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1577                         sector_t bi_sector = dio->bio_details.bi_iter.bi_sector;
1578                         unsigned bi_size = dio->bio_details.bi_iter.bi_size;
1579                         unsigned max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1580                         unsigned max_blocks = max_size / ic->tag_size;
1581                         memset(checksums, DISCARD_FILLER, max_size);
1582
1583                         while (bi_size) {
1584                                 unsigned this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1585                                 this_step_blocks = min(this_step_blocks, max_blocks);
1586                                 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1587                                                         this_step_blocks * ic->tag_size, TAG_WRITE);
1588                                 if (unlikely(r)) {
1589                                         if (likely(checksums != checksums_onstack))
1590                                                 kfree(checksums);
1591                                         goto error;
1592                                 }
1593
1594                                 /*if (bi_size < this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block)) {
1595                                         printk("BUGG: bi_sector: %llx, bi_size: %u\n", bi_sector, bi_size);
1596                                         printk("BUGG: this_step_blocks: %u\n", this_step_blocks);
1597                                         BUG();
1598                                 }*/
1599                                 bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1600                                 bi_sector += this_step_blocks << ic->sb->log2_sectors_per_block;
1601                         }
1602
1603                         if (likely(checksums != checksums_onstack))
1604                                 kfree(checksums);
1605                         goto skip_io;
1606                 }
1607
1608                 save_metadata_block = dio->metadata_block;
1609                 save_metadata_offset = dio->metadata_offset;
1610                 sector = dio->range.logical_sector;
1611                 sectors_to_process = dio->range.n_sectors;
1612
1613                 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1614                         unsigned pos;
1615                         char *mem, *checksums_ptr;
1616
1617 again:
1618                         mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1619                         pos = 0;
1620                         checksums_ptr = checksums;
1621                         do {
1622                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1623                                 checksums_ptr += ic->tag_size;
1624                                 sectors_to_process -= ic->sectors_per_block;
1625                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1626                                 sector += ic->sectors_per_block;
1627                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1628                         kunmap_atomic(mem);
1629
1630                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1631                                                 checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1632                         if (unlikely(r)) {
1633                                 if (r > 0) {
1634                                         char b[BDEVNAME_SIZE];
1635                                         DMERR_LIMIT("%s: Checksum failed at sector 0x%llx", bio_devname(bio, b),
1636                                                     (sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1637                                         r = -EILSEQ;
1638                                         atomic64_inc(&ic->number_of_mismatches);
1639                                 }
1640                                 if (likely(checksums != checksums_onstack))
1641                                         kfree(checksums);
1642                                 goto error;
1643                         }
1644
1645                         if (!sectors_to_process)
1646                                 break;
1647
1648                         if (unlikely(pos < bv.bv_len)) {
1649                                 bv.bv_offset += pos;
1650                                 bv.bv_len -= pos;
1651                                 goto again;
1652                         }
1653                 }
1654
1655                 if (likely(checksums != checksums_onstack))
1656                         kfree(checksums);
1657         } else {
1658                 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1659
1660                 if (bip) {
1661                         struct bio_vec biv;
1662                         struct bvec_iter iter;
1663                         unsigned data_to_process = dio->range.n_sectors;
1664                         sector_to_block(ic, data_to_process);
1665                         data_to_process *= ic->tag_size;
1666
1667                         bip_for_each_vec(biv, bip, iter) {
1668                                 unsigned char *tag;
1669                                 unsigned this_len;
1670
1671                                 BUG_ON(PageHighMem(biv.bv_page));
1672                                 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1673                                 this_len = min(biv.bv_len, data_to_process);
1674                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1675                                                         this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1676                                 if (unlikely(r))
1677                                         goto error;
1678                                 data_to_process -= this_len;
1679                                 if (!data_to_process)
1680                                         break;
1681                         }
1682                 }
1683         }
1684 skip_io:
1685         dec_in_flight(dio);
1686         return;
1687 error:
1688         dio->bi_status = errno_to_blk_status(r);
1689         dec_in_flight(dio);
1690 }
1691
1692 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1693 {
1694         struct dm_integrity_c *ic = ti->private;
1695         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1696         struct bio_integrity_payload *bip;
1697
1698         sector_t area, offset;
1699
1700         dio->ic = ic;
1701         dio->bi_status = 0;
1702         dio->op = bio_op(bio);
1703
1704         if (unlikely(dio->op == REQ_OP_DISCARD)) {
1705                 if (ti->max_io_len) {
1706                         sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1707                         unsigned log2_max_io_len = __fls(ti->max_io_len);
1708                         sector_t start_boundary = sec >> log2_max_io_len;
1709                         sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1710                         if (start_boundary < end_boundary) {
1711                                 sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1712                                 dm_accept_partial_bio(bio, len);
1713                         }
1714                 }
1715         }
1716
1717         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1718                 submit_flush_bio(ic, dio);
1719                 return DM_MAPIO_SUBMITTED;
1720         }
1721
1722         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1723         dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1724         if (unlikely(dio->fua)) {
1725                 /*
1726                  * Don't pass down the FUA flag because we have to flush
1727                  * disk cache anyway.
1728                  */
1729                 bio->bi_opf &= ~REQ_FUA;
1730         }
1731         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1732                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1733                       dio->range.logical_sector, bio_sectors(bio),
1734                       ic->provided_data_sectors);
1735                 return DM_MAPIO_KILL;
1736         }
1737         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1738                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1739                       ic->sectors_per_block,
1740                       dio->range.logical_sector, bio_sectors(bio));
1741                 return DM_MAPIO_KILL;
1742         }
1743
1744         if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1745                 struct bvec_iter iter;
1746                 struct bio_vec bv;
1747                 bio_for_each_segment(bv, bio, iter) {
1748                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1749                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1750                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1751                                 return DM_MAPIO_KILL;
1752                         }
1753                 }
1754         }
1755
1756         bip = bio_integrity(bio);
1757         if (!ic->internal_hash) {
1758                 if (bip) {
1759                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1760                         if (ic->log2_tag_size >= 0)
1761                                 wanted_tag_size <<= ic->log2_tag_size;
1762                         else
1763                                 wanted_tag_size *= ic->tag_size;
1764                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1765                                 DMERR("Invalid integrity data size %u, expected %u",
1766                                       bip->bip_iter.bi_size, wanted_tag_size);
1767                                 return DM_MAPIO_KILL;
1768                         }
1769                 }
1770         } else {
1771                 if (unlikely(bip != NULL)) {
1772                         DMERR("Unexpected integrity data when using internal hash");
1773                         return DM_MAPIO_KILL;
1774                 }
1775         }
1776
1777         if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1778                 return DM_MAPIO_KILL;
1779
1780         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1781         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1782         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1783
1784         dm_integrity_map_continue(dio, true);
1785         return DM_MAPIO_SUBMITTED;
1786 }
1787
1788 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1789                                  unsigned journal_section, unsigned journal_entry)
1790 {
1791         struct dm_integrity_c *ic = dio->ic;
1792         sector_t logical_sector;
1793         unsigned n_sectors;
1794
1795         logical_sector = dio->range.logical_sector;
1796         n_sectors = dio->range.n_sectors;
1797         do {
1798                 struct bio_vec bv = bio_iovec(bio);
1799                 char *mem;
1800
1801                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1802                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1803                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1804                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1805 retry_kmap:
1806                 mem = kmap_atomic(bv.bv_page);
1807                 if (likely(dio->op == REQ_OP_WRITE))
1808                         flush_dcache_page(bv.bv_page);
1809
1810                 do {
1811                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1812
1813                         if (unlikely(dio->op == REQ_OP_READ)) {
1814                                 struct journal_sector *js;
1815                                 char *mem_ptr;
1816                                 unsigned s;
1817
1818                                 if (unlikely(journal_entry_is_inprogress(je))) {
1819                                         flush_dcache_page(bv.bv_page);
1820                                         kunmap_atomic(mem);
1821
1822                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1823                                         goto retry_kmap;
1824                                 }
1825                                 smp_rmb();
1826                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1827                                 js = access_journal_data(ic, journal_section, journal_entry);
1828                                 mem_ptr = mem + bv.bv_offset;
1829                                 s = 0;
1830                                 do {
1831                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1832                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1833                                         js++;
1834                                         mem_ptr += 1 << SECTOR_SHIFT;
1835                                 } while (++s < ic->sectors_per_block);
1836 #ifdef INTERNAL_VERIFY
1837                                 if (ic->internal_hash) {
1838                                         char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1839
1840                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1841                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1842                                                 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1843                                                             logical_sector);
1844                                         }
1845                                 }
1846 #endif
1847                         }
1848
1849                         if (!ic->internal_hash) {
1850                                 struct bio_integrity_payload *bip = bio_integrity(bio);
1851                                 unsigned tag_todo = ic->tag_size;
1852                                 char *tag_ptr = journal_entry_tag(ic, je);
1853
1854                                 if (bip) do {
1855                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1856                                         unsigned tag_now = min(biv.bv_len, tag_todo);
1857                                         char *tag_addr;
1858                                         BUG_ON(PageHighMem(biv.bv_page));
1859                                         tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1860                                         if (likely(dio->op == REQ_OP_WRITE))
1861                                                 memcpy(tag_ptr, tag_addr, tag_now);
1862                                         else
1863                                                 memcpy(tag_addr, tag_ptr, tag_now);
1864                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1865                                         tag_ptr += tag_now;
1866                                         tag_todo -= tag_now;
1867                                 } while (unlikely(tag_todo)); else {
1868                                         if (likely(dio->op == REQ_OP_WRITE))
1869                                                 memset(tag_ptr, 0, tag_todo);
1870                                 }
1871                         }
1872
1873                         if (likely(dio->op == REQ_OP_WRITE)) {
1874                                 struct journal_sector *js;
1875                                 unsigned s;
1876
1877                                 js = access_journal_data(ic, journal_section, journal_entry);
1878                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1879
1880                                 s = 0;
1881                                 do {
1882                                         je->last_bytes[s] = js[s].commit_id;
1883                                 } while (++s < ic->sectors_per_block);
1884
1885                                 if (ic->internal_hash) {
1886                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1887                                         if (unlikely(digest_size > ic->tag_size)) {
1888                                                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1889                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1890                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1891                                         } else
1892                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1893                                 }
1894
1895                                 journal_entry_set_sector(je, logical_sector);
1896                         }
1897                         logical_sector += ic->sectors_per_block;
1898
1899                         journal_entry++;
1900                         if (unlikely(journal_entry == ic->journal_section_entries)) {
1901                                 journal_entry = 0;
1902                                 journal_section++;
1903                                 wraparound_section(ic, &journal_section);
1904                         }
1905
1906                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1907                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1908
1909                 if (unlikely(dio->op == REQ_OP_READ))
1910                         flush_dcache_page(bv.bv_page);
1911                 kunmap_atomic(mem);
1912         } while (n_sectors);
1913
1914         if (likely(dio->op == REQ_OP_WRITE)) {
1915                 smp_mb();
1916                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1917                         wake_up(&ic->copy_to_journal_wait);
1918                 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1919                         queue_work(ic->commit_wq, &ic->commit_work);
1920                 } else {
1921                         schedule_autocommit(ic);
1922                 }
1923         } else {
1924                 remove_range(ic, &dio->range);
1925         }
1926
1927         if (unlikely(bio->bi_iter.bi_size)) {
1928                 sector_t area, offset;
1929
1930                 dio->range.logical_sector = logical_sector;
1931                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1932                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1933                 return true;
1934         }
1935
1936         return false;
1937 }
1938
1939 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1940 {
1941         struct dm_integrity_c *ic = dio->ic;
1942         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1943         unsigned journal_section, journal_entry;
1944         unsigned journal_read_pos;
1945         struct completion read_comp;
1946         bool discard_retried = false;
1947         bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
1948         if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
1949                 need_sync_io = true;
1950
1951         if (need_sync_io && from_map) {
1952                 INIT_WORK(&dio->work, integrity_bio_wait);
1953                 queue_work(ic->offload_wq, &dio->work);
1954                 return;
1955         }
1956
1957 lock_retry:
1958         spin_lock_irq(&ic->endio_wait.lock);
1959 retry:
1960         if (unlikely(dm_integrity_failed(ic))) {
1961                 spin_unlock_irq(&ic->endio_wait.lock);
1962                 do_endio(ic, bio);
1963                 return;
1964         }
1965         dio->range.n_sectors = bio_sectors(bio);
1966         journal_read_pos = NOT_FOUND;
1967         if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
1968                 if (dio->op == REQ_OP_WRITE) {
1969                         unsigned next_entry, i, pos;
1970                         unsigned ws, we, range_sectors;
1971
1972                         dio->range.n_sectors = min(dio->range.n_sectors,
1973                                                    (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
1974                         if (unlikely(!dio->range.n_sectors)) {
1975                                 if (from_map)
1976                                         goto offload_to_thread;
1977                                 sleep_on_endio_wait(ic);
1978                                 goto retry;
1979                         }
1980                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1981                         ic->free_sectors -= range_sectors;
1982                         journal_section = ic->free_section;
1983                         journal_entry = ic->free_section_entry;
1984
1985                         next_entry = ic->free_section_entry + range_sectors;
1986                         ic->free_section_entry = next_entry % ic->journal_section_entries;
1987                         ic->free_section += next_entry / ic->journal_section_entries;
1988                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1989                         wraparound_section(ic, &ic->free_section);
1990
1991                         pos = journal_section * ic->journal_section_entries + journal_entry;
1992                         ws = journal_section;
1993                         we = journal_entry;
1994                         i = 0;
1995                         do {
1996                                 struct journal_entry *je;
1997
1998                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1999                                 pos++;
2000                                 if (unlikely(pos >= ic->journal_entries))
2001                                         pos = 0;
2002
2003                                 je = access_journal_entry(ic, ws, we);
2004                                 BUG_ON(!journal_entry_is_unused(je));
2005                                 journal_entry_set_inprogress(je);
2006                                 we++;
2007                                 if (unlikely(we == ic->journal_section_entries)) {
2008                                         we = 0;
2009                                         ws++;
2010                                         wraparound_section(ic, &ws);
2011                                 }
2012                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2013
2014                         spin_unlock_irq(&ic->endio_wait.lock);
2015                         goto journal_read_write;
2016                 } else {
2017                         sector_t next_sector;
2018                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2019                         if (likely(journal_read_pos == NOT_FOUND)) {
2020                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2021                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
2022                         } else {
2023                                 unsigned i;
2024                                 unsigned jp = journal_read_pos + 1;
2025                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2026                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2027                                                 break;
2028                                 }
2029                                 dio->range.n_sectors = i;
2030                         }
2031                 }
2032         }
2033         if (unlikely(!add_new_range(ic, &dio->range, true))) {
2034                 /*
2035                  * We must not sleep in the request routine because it could
2036                  * stall bios on current->bio_list.
2037                  * So, we offload the bio to a workqueue if we have to sleep.
2038                  */
2039                 if (from_map) {
2040 offload_to_thread:
2041                         spin_unlock_irq(&ic->endio_wait.lock);
2042                         INIT_WORK(&dio->work, integrity_bio_wait);
2043                         queue_work(ic->wait_wq, &dio->work);
2044                         return;
2045                 }
2046                 if (journal_read_pos != NOT_FOUND)
2047                         dio->range.n_sectors = ic->sectors_per_block;
2048                 wait_and_add_new_range(ic, &dio->range);
2049                 /*
2050                  * wait_and_add_new_range drops the spinlock, so the journal
2051                  * may have been changed arbitrarily. We need to recheck.
2052                  * To simplify the code, we restrict I/O size to just one block.
2053                  */
2054                 if (journal_read_pos != NOT_FOUND) {
2055                         sector_t next_sector;
2056                         unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2057                         if (unlikely(new_pos != journal_read_pos)) {
2058                                 remove_range_unlocked(ic, &dio->range);
2059                                 goto retry;
2060                         }
2061                 }
2062         }
2063         if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2064                 sector_t next_sector;
2065                 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2066                 if (unlikely(new_pos != NOT_FOUND) ||
2067                     unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2068                         remove_range_unlocked(ic, &dio->range);
2069                         spin_unlock_irq(&ic->endio_wait.lock);
2070                         queue_work(ic->commit_wq, &ic->commit_work);
2071                         flush_workqueue(ic->commit_wq);
2072                         queue_work(ic->writer_wq, &ic->writer_work);
2073                         flush_workqueue(ic->writer_wq);
2074                         discard_retried = true;
2075                         goto lock_retry;
2076                 }
2077         }
2078         spin_unlock_irq(&ic->endio_wait.lock);
2079
2080         if (unlikely(journal_read_pos != NOT_FOUND)) {
2081                 journal_section = journal_read_pos / ic->journal_section_entries;
2082                 journal_entry = journal_read_pos % ic->journal_section_entries;
2083                 goto journal_read_write;
2084         }
2085
2086         if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2087                 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2088                                      dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2089                         struct bitmap_block_status *bbs;
2090
2091                         bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2092                         spin_lock(&bbs->bio_queue_lock);
2093                         bio_list_add(&bbs->bio_queue, bio);
2094                         spin_unlock(&bbs->bio_queue_lock);
2095                         queue_work(ic->writer_wq, &bbs->work);
2096                         return;
2097                 }
2098         }
2099
2100         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2101
2102         if (need_sync_io) {
2103                 init_completion(&read_comp);
2104                 dio->completion = &read_comp;
2105         } else
2106                 dio->completion = NULL;
2107
2108         dm_bio_record(&dio->bio_details, bio);
2109         bio_set_dev(bio, ic->dev->bdev);
2110         bio->bi_integrity = NULL;
2111         bio->bi_opf &= ~REQ_INTEGRITY;
2112         bio->bi_end_io = integrity_end_io;
2113         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2114
2115         if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2116                 integrity_metadata(&dio->work);
2117                 dm_integrity_flush_buffers(ic);
2118
2119                 dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2120                 dio->completion = NULL;
2121
2122                 generic_make_request(bio);
2123
2124                 return;
2125         }
2126
2127         generic_make_request(bio);
2128
2129         if (need_sync_io) {
2130                 wait_for_completion_io(&read_comp);
2131                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2132                     dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2133                         goto skip_check;
2134                 if (ic->mode == 'B') {
2135                         if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2136                                              dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2137                                 goto skip_check;
2138                 }
2139
2140                 if (likely(!bio->bi_status))
2141                         integrity_metadata(&dio->work);
2142                 else
2143 skip_check:
2144                         dec_in_flight(dio);
2145
2146         } else {
2147                 INIT_WORK(&dio->work, integrity_metadata);
2148                 queue_work(ic->metadata_wq, &dio->work);
2149         }
2150
2151         return;
2152
2153 journal_read_write:
2154         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2155                 goto lock_retry;
2156
2157         do_endio_flush(ic, dio);
2158 }
2159
2160
2161 static void integrity_bio_wait(struct work_struct *w)
2162 {
2163         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2164
2165         dm_integrity_map_continue(dio, false);
2166 }
2167
2168 static void pad_uncommitted(struct dm_integrity_c *ic)
2169 {
2170         if (ic->free_section_entry) {
2171                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2172                 ic->free_section_entry = 0;
2173                 ic->free_section++;
2174                 wraparound_section(ic, &ic->free_section);
2175                 ic->n_uncommitted_sections++;
2176         }
2177         if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2178                     (ic->n_uncommitted_sections + ic->n_committed_sections) *
2179                     ic->journal_section_entries + ic->free_sectors)) {
2180                 DMCRIT("journal_sections %u, journal_section_entries %u, "
2181                        "n_uncommitted_sections %u, n_committed_sections %u, "
2182                        "journal_section_entries %u, free_sectors %u",
2183                        ic->journal_sections, ic->journal_section_entries,
2184                        ic->n_uncommitted_sections, ic->n_committed_sections,
2185                        ic->journal_section_entries, ic->free_sectors);
2186         }
2187 }
2188
2189 static void integrity_commit(struct work_struct *w)
2190 {
2191         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2192         unsigned commit_start, commit_sections;
2193         unsigned i, j, n;
2194         struct bio *flushes;
2195
2196         del_timer(&ic->autocommit_timer);
2197
2198         spin_lock_irq(&ic->endio_wait.lock);
2199         flushes = bio_list_get(&ic->flush_bio_list);
2200         if (unlikely(ic->mode != 'J')) {
2201                 spin_unlock_irq(&ic->endio_wait.lock);
2202                 dm_integrity_flush_buffers(ic);
2203                 goto release_flush_bios;
2204         }
2205
2206         pad_uncommitted(ic);
2207         commit_start = ic->uncommitted_section;
2208         commit_sections = ic->n_uncommitted_sections;
2209         spin_unlock_irq(&ic->endio_wait.lock);
2210
2211         if (!commit_sections)
2212                 goto release_flush_bios;
2213
2214         i = commit_start;
2215         for (n = 0; n < commit_sections; n++) {
2216                 for (j = 0; j < ic->journal_section_entries; j++) {
2217                         struct journal_entry *je;
2218                         je = access_journal_entry(ic, i, j);
2219                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2220                 }
2221                 for (j = 0; j < ic->journal_section_sectors; j++) {
2222                         struct journal_sector *js;
2223                         js = access_journal(ic, i, j);
2224                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2225                 }
2226                 i++;
2227                 if (unlikely(i >= ic->journal_sections))
2228                         ic->commit_seq = next_commit_seq(ic->commit_seq);
2229                 wraparound_section(ic, &i);
2230         }
2231         smp_rmb();
2232
2233         write_journal(ic, commit_start, commit_sections);
2234
2235         spin_lock_irq(&ic->endio_wait.lock);
2236         ic->uncommitted_section += commit_sections;
2237         wraparound_section(ic, &ic->uncommitted_section);
2238         ic->n_uncommitted_sections -= commit_sections;
2239         ic->n_committed_sections += commit_sections;
2240         spin_unlock_irq(&ic->endio_wait.lock);
2241
2242         if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2243                 queue_work(ic->writer_wq, &ic->writer_work);
2244
2245 release_flush_bios:
2246         while (flushes) {
2247                 struct bio *next = flushes->bi_next;
2248                 flushes->bi_next = NULL;
2249                 do_endio(ic, flushes);
2250                 flushes = next;
2251         }
2252 }
2253
2254 static void complete_copy_from_journal(unsigned long error, void *context)
2255 {
2256         struct journal_io *io = context;
2257         struct journal_completion *comp = io->comp;
2258         struct dm_integrity_c *ic = comp->ic;
2259         remove_range(ic, &io->range);
2260         mempool_free(io, &ic->journal_io_mempool);
2261         if (unlikely(error != 0))
2262                 dm_integrity_io_error(ic, "copying from journal", -EIO);
2263         complete_journal_op(comp);
2264 }
2265
2266 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2267                                struct journal_entry *je)
2268 {
2269         unsigned s = 0;
2270         do {
2271                 js->commit_id = je->last_bytes[s];
2272                 js++;
2273         } while (++s < ic->sectors_per_block);
2274 }
2275
2276 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2277                              unsigned write_sections, bool from_replay)
2278 {
2279         unsigned i, j, n;
2280         struct journal_completion comp;
2281         struct blk_plug plug;
2282
2283         blk_start_plug(&plug);
2284
2285         comp.ic = ic;
2286         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2287         init_completion(&comp.comp);
2288
2289         i = write_start;
2290         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2291 #ifndef INTERNAL_VERIFY
2292                 if (unlikely(from_replay))
2293 #endif
2294                         rw_section_mac(ic, i, false);
2295                 for (j = 0; j < ic->journal_section_entries; j++) {
2296                         struct journal_entry *je = access_journal_entry(ic, i, j);
2297                         sector_t sec, area, offset;
2298                         unsigned k, l, next_loop;
2299                         sector_t metadata_block;
2300                         unsigned metadata_offset;
2301                         struct journal_io *io;
2302
2303                         if (journal_entry_is_unused(je))
2304                                 continue;
2305                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2306                         sec = journal_entry_get_sector(je);
2307                         if (unlikely(from_replay)) {
2308                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2309                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2310                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
2311                                 }
2312                         }
2313                         if (unlikely(sec >= ic->provided_data_sectors))
2314                                 continue;
2315                         get_area_and_offset(ic, sec, &area, &offset);
2316                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2317                         for (k = j + 1; k < ic->journal_section_entries; k++) {
2318                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2319                                 sector_t sec2, area2, offset2;
2320                                 if (journal_entry_is_unused(je2))
2321                                         break;
2322                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2323                                 sec2 = journal_entry_get_sector(je2);
2324                                 if (unlikely(sec2 >= ic->provided_data_sectors))
2325                                         break;
2326                                 get_area_and_offset(ic, sec2, &area2, &offset2);
2327                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2328                                         break;
2329                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2330                         }
2331                         next_loop = k - 1;
2332
2333                         io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2334                         io->comp = &comp;
2335                         io->range.logical_sector = sec;
2336                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2337
2338                         spin_lock_irq(&ic->endio_wait.lock);
2339                         add_new_range_and_wait(ic, &io->range);
2340
2341                         if (likely(!from_replay)) {
2342                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2343
2344                                 /* don't write if there is newer committed sector */
2345                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2346                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
2347
2348                                         journal_entry_set_unused(je2);
2349                                         remove_journal_node(ic, &section_node[j]);
2350                                         j++;
2351                                         sec += ic->sectors_per_block;
2352                                         offset += ic->sectors_per_block;
2353                                 }
2354                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2355                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2356
2357                                         journal_entry_set_unused(je2);
2358                                         remove_journal_node(ic, &section_node[k - 1]);
2359                                         k--;
2360                                 }
2361                                 if (j == k) {
2362                                         remove_range_unlocked(ic, &io->range);
2363                                         spin_unlock_irq(&ic->endio_wait.lock);
2364                                         mempool_free(io, &ic->journal_io_mempool);
2365                                         goto skip_io;
2366                                 }
2367                                 for (l = j; l < k; l++) {
2368                                         remove_journal_node(ic, &section_node[l]);
2369                                 }
2370                         }
2371                         spin_unlock_irq(&ic->endio_wait.lock);
2372
2373                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2374                         for (l = j; l < k; l++) {
2375                                 int r;
2376                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2377
2378                                 if (
2379 #ifndef INTERNAL_VERIFY
2380                                     unlikely(from_replay) &&
2381 #endif
2382                                     ic->internal_hash) {
2383                                         char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2384
2385                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2386                                                                   (char *)access_journal_data(ic, i, l), test_tag);
2387                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2388                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2389                                 }
2390
2391                                 journal_entry_set_unused(je2);
2392                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2393                                                         ic->tag_size, TAG_WRITE);
2394                                 if (unlikely(r)) {
2395                                         dm_integrity_io_error(ic, "reading tags", r);
2396                                 }
2397                         }
2398
2399                         atomic_inc(&comp.in_flight);
2400                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2401                                           (k - j) << ic->sb->log2_sectors_per_block,
2402                                           get_data_sector(ic, area, offset),
2403                                           complete_copy_from_journal, io);
2404 skip_io:
2405                         j = next_loop;
2406                 }
2407         }
2408
2409         dm_bufio_write_dirty_buffers_async(ic->bufio);
2410
2411         blk_finish_plug(&plug);
2412
2413         complete_journal_op(&comp);
2414         wait_for_completion_io(&comp.comp);
2415
2416         dm_integrity_flush_buffers(ic);
2417 }
2418
2419 static void integrity_writer(struct work_struct *w)
2420 {
2421         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2422         unsigned write_start, write_sections;
2423
2424         unsigned prev_free_sectors;
2425
2426         /* the following test is not needed, but it tests the replay code */
2427         if (unlikely(dm_suspended(ic->ti)) && !ic->meta_dev)
2428                 return;
2429
2430         spin_lock_irq(&ic->endio_wait.lock);
2431         write_start = ic->committed_section;
2432         write_sections = ic->n_committed_sections;
2433         spin_unlock_irq(&ic->endio_wait.lock);
2434
2435         if (!write_sections)
2436                 return;
2437
2438         do_journal_write(ic, write_start, write_sections, false);
2439
2440         spin_lock_irq(&ic->endio_wait.lock);
2441
2442         ic->committed_section += write_sections;
2443         wraparound_section(ic, &ic->committed_section);
2444         ic->n_committed_sections -= write_sections;
2445
2446         prev_free_sectors = ic->free_sectors;
2447         ic->free_sectors += write_sections * ic->journal_section_entries;
2448         if (unlikely(!prev_free_sectors))
2449                 wake_up_locked(&ic->endio_wait);
2450
2451         spin_unlock_irq(&ic->endio_wait.lock);
2452 }
2453
2454 static void recalc_write_super(struct dm_integrity_c *ic)
2455 {
2456         int r;
2457
2458         dm_integrity_flush_buffers(ic);
2459         if (dm_integrity_failed(ic))
2460                 return;
2461
2462         r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2463         if (unlikely(r))
2464                 dm_integrity_io_error(ic, "writing superblock", r);
2465 }
2466
2467 static void integrity_recalc(struct work_struct *w)
2468 {
2469         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2470         struct dm_integrity_range range;
2471         struct dm_io_request io_req;
2472         struct dm_io_region io_loc;
2473         sector_t area, offset;
2474         sector_t metadata_block;
2475         unsigned metadata_offset;
2476         sector_t logical_sector, n_sectors;
2477         __u8 *t;
2478         unsigned i;
2479         int r;
2480         unsigned super_counter = 0;
2481
2482         DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2483
2484         spin_lock_irq(&ic->endio_wait.lock);
2485
2486 next_chunk:
2487
2488         if (unlikely(dm_suspended(ic->ti)))
2489                 goto unlock_ret;
2490
2491         range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2492         if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2493                 if (ic->mode == 'B') {
2494                         DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2495                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2496                 }
2497                 goto unlock_ret;
2498         }
2499
2500         get_area_and_offset(ic, range.logical_sector, &area, &offset);
2501         range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2502         if (!ic->meta_dev)
2503                 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2504
2505         add_new_range_and_wait(ic, &range);
2506         spin_unlock_irq(&ic->endio_wait.lock);
2507         logical_sector = range.logical_sector;
2508         n_sectors = range.n_sectors;
2509
2510         if (ic->mode == 'B') {
2511                 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2512                         goto advance_and_next;
2513                 }
2514                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2515                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2516                         logical_sector += ic->sectors_per_block;
2517                         n_sectors -= ic->sectors_per_block;
2518                         cond_resched();
2519                 }
2520                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2521                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2522                         n_sectors -= ic->sectors_per_block;
2523                         cond_resched();
2524                 }
2525                 get_area_and_offset(ic, logical_sector, &area, &offset);
2526         }
2527
2528         DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2529
2530         if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2531                 recalc_write_super(ic);
2532                 if (ic->mode == 'B') {
2533                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2534                 }
2535                 super_counter = 0;
2536         }
2537
2538         if (unlikely(dm_integrity_failed(ic)))
2539                 goto err;
2540
2541         io_req.bi_op = REQ_OP_READ;
2542         io_req.bi_op_flags = 0;
2543         io_req.mem.type = DM_IO_VMA;
2544         io_req.mem.ptr.addr = ic->recalc_buffer;
2545         io_req.notify.fn = NULL;
2546         io_req.client = ic->io;
2547         io_loc.bdev = ic->dev->bdev;
2548         io_loc.sector = get_data_sector(ic, area, offset);
2549         io_loc.count = n_sectors;
2550
2551         r = dm_io(&io_req, 1, &io_loc, NULL);
2552         if (unlikely(r)) {
2553                 dm_integrity_io_error(ic, "reading data", r);
2554                 goto err;
2555         }
2556
2557         t = ic->recalc_tags;
2558         for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2559                 integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2560                 t += ic->tag_size;
2561         }
2562
2563         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2564
2565         r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2566         if (unlikely(r)) {
2567                 dm_integrity_io_error(ic, "writing tags", r);
2568                 goto err;
2569         }
2570
2571 advance_and_next:
2572         cond_resched();
2573
2574         spin_lock_irq(&ic->endio_wait.lock);
2575         remove_range_unlocked(ic, &range);
2576         ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2577         goto next_chunk;
2578
2579 err:
2580         remove_range(ic, &range);
2581         return;
2582
2583 unlock_ret:
2584         spin_unlock_irq(&ic->endio_wait.lock);
2585
2586         recalc_write_super(ic);
2587 }
2588
2589 static void bitmap_block_work(struct work_struct *w)
2590 {
2591         struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2592         struct dm_integrity_c *ic = bbs->ic;
2593         struct bio *bio;
2594         struct bio_list bio_queue;
2595         struct bio_list waiting;
2596
2597         bio_list_init(&waiting);
2598
2599         spin_lock(&bbs->bio_queue_lock);
2600         bio_queue = bbs->bio_queue;
2601         bio_list_init(&bbs->bio_queue);
2602         spin_unlock(&bbs->bio_queue_lock);
2603
2604         while ((bio = bio_list_pop(&bio_queue))) {
2605                 struct dm_integrity_io *dio;
2606
2607                 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2608
2609                 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2610                                     dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2611                         remove_range(ic, &dio->range);
2612                         INIT_WORK(&dio->work, integrity_bio_wait);
2613                         queue_work(ic->offload_wq, &dio->work);
2614                 } else {
2615                         block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2616                                         dio->range.n_sectors, BITMAP_OP_SET);
2617                         bio_list_add(&waiting, bio);
2618                 }
2619         }
2620
2621         if (bio_list_empty(&waiting))
2622                 return;
2623
2624         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC,
2625                            bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2626                            BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2627
2628         while ((bio = bio_list_pop(&waiting))) {
2629                 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2630
2631                 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2632                                 dio->range.n_sectors, BITMAP_OP_SET);
2633
2634                 remove_range(ic, &dio->range);
2635                 INIT_WORK(&dio->work, integrity_bio_wait);
2636                 queue_work(ic->offload_wq, &dio->work);
2637         }
2638
2639         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2640 }
2641
2642 static void bitmap_flush_work(struct work_struct *work)
2643 {
2644         struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2645         struct dm_integrity_range range;
2646         unsigned long limit;
2647         struct bio *bio;
2648
2649         dm_integrity_flush_buffers(ic);
2650
2651         range.logical_sector = 0;
2652         range.n_sectors = ic->provided_data_sectors;
2653
2654         spin_lock_irq(&ic->endio_wait.lock);
2655         add_new_range_and_wait(ic, &range);
2656         spin_unlock_irq(&ic->endio_wait.lock);
2657
2658         dm_integrity_flush_buffers(ic);
2659         if (ic->meta_dev)
2660                 blkdev_issue_flush(ic->dev->bdev, GFP_NOIO, NULL);
2661
2662         limit = ic->provided_data_sectors;
2663         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2664                 limit = le64_to_cpu(ic->sb->recalc_sector)
2665                         >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2666                         << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2667         }
2668         /*DEBUG_print("zeroing journal\n");*/
2669         block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2670         block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2671
2672         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2673                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2674
2675         spin_lock_irq(&ic->endio_wait.lock);
2676         remove_range_unlocked(ic, &range);
2677         while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2678                 bio_endio(bio);
2679                 spin_unlock_irq(&ic->endio_wait.lock);
2680                 spin_lock_irq(&ic->endio_wait.lock);
2681         }
2682         spin_unlock_irq(&ic->endio_wait.lock);
2683 }
2684
2685
2686 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2687                          unsigned n_sections, unsigned char commit_seq)
2688 {
2689         unsigned i, j, n;
2690
2691         if (!n_sections)
2692                 return;
2693
2694         for (n = 0; n < n_sections; n++) {
2695                 i = start_section + n;
2696                 wraparound_section(ic, &i);
2697                 for (j = 0; j < ic->journal_section_sectors; j++) {
2698                         struct journal_sector *js = access_journal(ic, i, j);
2699                         memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2700                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2701                 }
2702                 for (j = 0; j < ic->journal_section_entries; j++) {
2703                         struct journal_entry *je = access_journal_entry(ic, i, j);
2704                         journal_entry_set_unused(je);
2705                 }
2706         }
2707
2708         write_journal(ic, start_section, n_sections);
2709 }
2710
2711 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2712 {
2713         unsigned char k;
2714         for (k = 0; k < N_COMMIT_IDS; k++) {
2715                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2716                         return k;
2717         }
2718         dm_integrity_io_error(ic, "journal commit id", -EIO);
2719         return -EIO;
2720 }
2721
2722 static void replay_journal(struct dm_integrity_c *ic)
2723 {
2724         unsigned i, j;
2725         bool used_commit_ids[N_COMMIT_IDS];
2726         unsigned max_commit_id_sections[N_COMMIT_IDS];
2727         unsigned write_start, write_sections;
2728         unsigned continue_section;
2729         bool journal_empty;
2730         unsigned char unused, last_used, want_commit_seq;
2731
2732         if (ic->mode == 'R')
2733                 return;
2734
2735         if (ic->journal_uptodate)
2736                 return;
2737
2738         last_used = 0;
2739         write_start = 0;
2740
2741         if (!ic->just_formatted) {
2742                 DEBUG_print("reading journal\n");
2743                 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2744                 if (ic->journal_io)
2745                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2746                 if (ic->journal_io) {
2747                         struct journal_completion crypt_comp;
2748                         crypt_comp.ic = ic;
2749                         init_completion(&crypt_comp.comp);
2750                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2751                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2752                         wait_for_completion(&crypt_comp.comp);
2753                 }
2754                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2755         }
2756
2757         if (dm_integrity_failed(ic))
2758                 goto clear_journal;
2759
2760         journal_empty = true;
2761         memset(used_commit_ids, 0, sizeof used_commit_ids);
2762         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2763         for (i = 0; i < ic->journal_sections; i++) {
2764                 for (j = 0; j < ic->journal_section_sectors; j++) {
2765                         int k;
2766                         struct journal_sector *js = access_journal(ic, i, j);
2767                         k = find_commit_seq(ic, i, j, js->commit_id);
2768                         if (k < 0)
2769                                 goto clear_journal;
2770                         used_commit_ids[k] = true;
2771                         max_commit_id_sections[k] = i;
2772                 }
2773                 if (journal_empty) {
2774                         for (j = 0; j < ic->journal_section_entries; j++) {
2775                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2776                                 if (!journal_entry_is_unused(je)) {
2777                                         journal_empty = false;
2778                                         break;
2779                                 }
2780                         }
2781                 }
2782         }
2783
2784         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2785                 unused = N_COMMIT_IDS - 1;
2786                 while (unused && !used_commit_ids[unused - 1])
2787                         unused--;
2788         } else {
2789                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2790                         if (!used_commit_ids[unused])
2791                                 break;
2792                 if (unused == N_COMMIT_IDS) {
2793                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2794                         goto clear_journal;
2795                 }
2796         }
2797         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2798                     unused, used_commit_ids[0], used_commit_ids[1],
2799                     used_commit_ids[2], used_commit_ids[3]);
2800
2801         last_used = prev_commit_seq(unused);
2802         want_commit_seq = prev_commit_seq(last_used);
2803
2804         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2805                 journal_empty = true;
2806
2807         write_start = max_commit_id_sections[last_used] + 1;
2808         if (unlikely(write_start >= ic->journal_sections))
2809                 want_commit_seq = next_commit_seq(want_commit_seq);
2810         wraparound_section(ic, &write_start);
2811
2812         i = write_start;
2813         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2814                 for (j = 0; j < ic->journal_section_sectors; j++) {
2815                         struct journal_sector *js = access_journal(ic, i, j);
2816
2817                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2818                                 /*
2819                                  * This could be caused by crash during writing.
2820                                  * We won't replay the inconsistent part of the
2821                                  * journal.
2822                                  */
2823                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2824                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2825                                 goto brk;
2826                         }
2827                 }
2828                 i++;
2829                 if (unlikely(i >= ic->journal_sections))
2830                         want_commit_seq = next_commit_seq(want_commit_seq);
2831                 wraparound_section(ic, &i);
2832         }
2833 brk:
2834
2835         if (!journal_empty) {
2836                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2837                             write_sections, write_start, want_commit_seq);
2838                 do_journal_write(ic, write_start, write_sections, true);
2839         }
2840
2841         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2842                 continue_section = write_start;
2843                 ic->commit_seq = want_commit_seq;
2844                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2845         } else {
2846                 unsigned s;
2847                 unsigned char erase_seq;
2848 clear_journal:
2849                 DEBUG_print("clearing journal\n");
2850
2851                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2852                 s = write_start;
2853                 init_journal(ic, s, 1, erase_seq);
2854                 s++;
2855                 wraparound_section(ic, &s);
2856                 if (ic->journal_sections >= 2) {
2857                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2858                         s += ic->journal_sections - 2;
2859                         wraparound_section(ic, &s);
2860                         init_journal(ic, s, 1, erase_seq);
2861                 }
2862
2863                 continue_section = 0;
2864                 ic->commit_seq = next_commit_seq(erase_seq);
2865         }
2866
2867         ic->committed_section = continue_section;
2868         ic->n_committed_sections = 0;
2869
2870         ic->uncommitted_section = continue_section;
2871         ic->n_uncommitted_sections = 0;
2872
2873         ic->free_section = continue_section;
2874         ic->free_section_entry = 0;
2875         ic->free_sectors = ic->journal_entries;
2876
2877         ic->journal_tree_root = RB_ROOT;
2878         for (i = 0; i < ic->journal_entries; i++)
2879                 init_journal_node(&ic->journal_tree[i]);
2880 }
2881
2882 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
2883 {
2884         DEBUG_print("dm_integrity_enter_synchronous_mode\n");
2885
2886         if (ic->mode == 'B') {
2887                 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
2888                 ic->synchronous_mode = 1;
2889
2890                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2891                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2892                 flush_workqueue(ic->commit_wq);
2893         }
2894 }
2895
2896 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
2897 {
2898         struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
2899
2900         DEBUG_print("dm_integrity_reboot\n");
2901
2902         dm_integrity_enter_synchronous_mode(ic);
2903
2904         return NOTIFY_DONE;
2905 }
2906
2907 static void dm_integrity_postsuspend(struct dm_target *ti)
2908 {
2909         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2910         int r;
2911
2912         WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
2913
2914         del_timer_sync(&ic->autocommit_timer);
2915
2916         if (ic->recalc_wq)
2917                 drain_workqueue(ic->recalc_wq);
2918
2919         if (ic->mode == 'B')
2920                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2921
2922         queue_work(ic->commit_wq, &ic->commit_work);
2923         drain_workqueue(ic->commit_wq);
2924
2925         if (ic->mode == 'J') {
2926                 if (ic->meta_dev)
2927                         queue_work(ic->writer_wq, &ic->writer_work);
2928                 drain_workqueue(ic->writer_wq);
2929                 dm_integrity_flush_buffers(ic);
2930         }
2931
2932         if (ic->mode == 'B') {
2933                 dm_integrity_flush_buffers(ic);
2934 #if 1
2935                 /* set to 0 to test bitmap replay code */
2936                 init_journal(ic, 0, ic->journal_sections, 0);
2937                 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2938                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2939                 if (unlikely(r))
2940                         dm_integrity_io_error(ic, "writing superblock", r);
2941 #endif
2942         }
2943
2944         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2945
2946         ic->journal_uptodate = true;
2947 }
2948
2949 static void dm_integrity_resume(struct dm_target *ti)
2950 {
2951         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2952         __u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
2953         int r;
2954
2955         DEBUG_print("resume\n");
2956
2957         if (ic->provided_data_sectors != old_provided_data_sectors) {
2958                 if (ic->provided_data_sectors > old_provided_data_sectors &&
2959                     ic->mode == 'B' &&
2960                     ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
2961                         rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
2962                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2963                         block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
2964                                         ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
2965                         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2966                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2967                 }
2968
2969                 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
2970                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2971                 if (unlikely(r))
2972                         dm_integrity_io_error(ic, "writing superblock", r);
2973         }
2974
2975         if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
2976                 DEBUG_print("resume dirty_bitmap\n");
2977                 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
2978                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2979                 if (ic->mode == 'B') {
2980                         if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
2981                                 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
2982                                 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
2983                                 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
2984                                                      BITMAP_OP_TEST_ALL_CLEAR)) {
2985                                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2986                                         ic->sb->recalc_sector = cpu_to_le64(0);
2987                                 }
2988                         } else {
2989                                 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
2990                                             ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
2991                                 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
2992                                 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2993                                 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2994                                 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2995                                 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2996                                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2997                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2998                                 ic->sb->recalc_sector = cpu_to_le64(0);
2999                         }
3000                 } else {
3001                         if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3002                               block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR))) {
3003                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3004                                 ic->sb->recalc_sector = cpu_to_le64(0);
3005                         }
3006                         init_journal(ic, 0, ic->journal_sections, 0);
3007                         replay_journal(ic);
3008                         ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3009                 }
3010                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3011                 if (unlikely(r))
3012                         dm_integrity_io_error(ic, "writing superblock", r);
3013         } else {
3014                 replay_journal(ic);
3015                 if (ic->mode == 'B') {
3016                         ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3017                         ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3018                         r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3019                         if (unlikely(r))
3020                                 dm_integrity_io_error(ic, "writing superblock", r);
3021
3022                         block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3023                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3024                         block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3025                         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3026                             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3027                                 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3028                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3029                                 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3030                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3031                                 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3032                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3033                         }
3034                         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3035                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3036                 }
3037         }
3038
3039         DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3040         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3041                 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3042                 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3043                 if (recalc_pos < ic->provided_data_sectors) {
3044                         queue_work(ic->recalc_wq, &ic->recalc_work);
3045                 } else if (recalc_pos > ic->provided_data_sectors) {
3046                         ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3047                         recalc_write_super(ic);
3048                 }
3049         }
3050
3051         ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3052         ic->reboot_notifier.next = NULL;
3053         ic->reboot_notifier.priority = INT_MAX - 1;     /* be notified after md and before hardware drivers */
3054         WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3055
3056 #if 0
3057         /* set to 1 to stress test synchronous mode */
3058         dm_integrity_enter_synchronous_mode(ic);
3059 #endif
3060 }
3061
3062 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3063                                 unsigned status_flags, char *result, unsigned maxlen)
3064 {
3065         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3066         unsigned arg_count;
3067         size_t sz = 0;
3068
3069         switch (type) {
3070         case STATUSTYPE_INFO:
3071                 DMEMIT("%llu %llu",
3072                         atomic64_read(&ic->number_of_mismatches),
3073                         ic->provided_data_sectors);
3074                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3075                         DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3076                 else
3077                         DMEMIT(" -");
3078                 break;
3079
3080         case STATUSTYPE_TABLE: {
3081                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3082                 watermark_percentage += ic->journal_entries / 2;
3083                 do_div(watermark_percentage, ic->journal_entries);
3084                 arg_count = 3;
3085                 arg_count += !!ic->meta_dev;
3086                 arg_count += ic->sectors_per_block != 1;
3087                 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3088                 arg_count += ic->discard;
3089                 arg_count += ic->mode == 'J';
3090                 arg_count += ic->mode == 'J';
3091                 arg_count += ic->mode == 'B';
3092                 arg_count += ic->mode == 'B';
3093                 arg_count += !!ic->internal_hash_alg.alg_string;
3094                 arg_count += !!ic->journal_crypt_alg.alg_string;
3095                 arg_count += !!ic->journal_mac_alg.alg_string;
3096                 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3097                 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3098                        ic->tag_size, ic->mode, arg_count);
3099                 if (ic->meta_dev)
3100                         DMEMIT(" meta_device:%s", ic->meta_dev->name);
3101                 if (ic->sectors_per_block != 1)
3102                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3103                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3104                         DMEMIT(" recalculate");
3105                 if (ic->discard)
3106                         DMEMIT(" allow_discards");
3107                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3108                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3109                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3110                 if (ic->mode == 'J') {
3111                         DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
3112                         DMEMIT(" commit_time:%u", ic->autocommit_msec);
3113                 }
3114                 if (ic->mode == 'B') {
3115                         DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3116                         DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3117                 }
3118                 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3119                         DMEMIT(" fix_padding");
3120
3121 #define EMIT_ALG(a, n)                                                  \
3122                 do {                                                    \
3123                         if (ic->a.alg_string) {                         \
3124                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
3125                                 if (ic->a.key_string)                   \
3126                                         DMEMIT(":%s", ic->a.key_string);\
3127                         }                                               \
3128                 } while (0)
3129                 EMIT_ALG(internal_hash_alg, "internal_hash");
3130                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3131                 EMIT_ALG(journal_mac_alg, "journal_mac");
3132                 break;
3133         }
3134         }
3135 }
3136
3137 static int dm_integrity_iterate_devices(struct dm_target *ti,
3138                                         iterate_devices_callout_fn fn, void *data)
3139 {
3140         struct dm_integrity_c *ic = ti->private;
3141
3142         if (!ic->meta_dev)
3143                 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3144         else
3145                 return fn(ti, ic->dev, 0, ti->len, data);
3146 }
3147
3148 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3149 {
3150         struct dm_integrity_c *ic = ti->private;
3151
3152         if (ic->sectors_per_block > 1) {
3153                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3154                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3155                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3156         }
3157 }
3158
3159 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3160 {
3161         unsigned sector_space = JOURNAL_SECTOR_DATA;
3162
3163         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3164         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3165                                          JOURNAL_ENTRY_ROUNDUP);
3166
3167         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3168                 sector_space -= JOURNAL_MAC_PER_SECTOR;
3169         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3170         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3171         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3172         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3173 }
3174
3175 static int calculate_device_limits(struct dm_integrity_c *ic)
3176 {
3177         __u64 initial_sectors;
3178
3179         calculate_journal_section_size(ic);
3180         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3181         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3182                 return -EINVAL;
3183         ic->initial_sectors = initial_sectors;
3184
3185         if (!ic->meta_dev) {
3186                 sector_t last_sector, last_area, last_offset;
3187
3188                 /* we have to maintain excessive padding for compatibility with existing volumes */
3189                 __u64 metadata_run_padding =
3190                         ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3191                         (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3192                         (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3193
3194                 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3195                                             metadata_run_padding) >> SECTOR_SHIFT;
3196                 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3197                         ic->log2_metadata_run = __ffs(ic->metadata_run);
3198                 else
3199                         ic->log2_metadata_run = -1;
3200
3201                 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3202                 last_sector = get_data_sector(ic, last_area, last_offset);
3203                 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3204                         return -EINVAL;
3205         } else {
3206                 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3207                 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3208                                 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3209                 meta_size <<= ic->log2_buffer_sectors;
3210                 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3211                     ic->initial_sectors + meta_size > ic->meta_device_sectors)
3212                         return -EINVAL;
3213                 ic->metadata_run = 1;
3214                 ic->log2_metadata_run = 0;
3215         }
3216
3217         return 0;
3218 }
3219
3220 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3221 {
3222         if (!ic->meta_dev) {
3223                 int test_bit;
3224                 ic->provided_data_sectors = 0;
3225                 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3226                         __u64 prev_data_sectors = ic->provided_data_sectors;
3227
3228                         ic->provided_data_sectors |= (sector_t)1 << test_bit;
3229                         if (calculate_device_limits(ic))
3230                                 ic->provided_data_sectors = prev_data_sectors;
3231                 }
3232         } else {
3233                 ic->provided_data_sectors = ic->data_device_sectors;
3234                 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3235         }
3236 }
3237
3238 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3239 {
3240         unsigned journal_sections;
3241         int test_bit;
3242
3243         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3244         memcpy(ic->sb->magic, SB_MAGIC, 8);
3245         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3246         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3247         if (ic->journal_mac_alg.alg_string)
3248                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3249
3250         calculate_journal_section_size(ic);
3251         journal_sections = journal_sectors / ic->journal_section_sectors;
3252         if (!journal_sections)
3253                 journal_sections = 1;
3254
3255         if (!ic->meta_dev) {
3256                 if (ic->fix_padding)
3257                         ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3258                 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3259                 if (!interleave_sectors)
3260                         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3261                 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3262                 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3263                 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3264
3265                 get_provided_data_sectors(ic);
3266                 if (!ic->provided_data_sectors)
3267                         return -EINVAL;
3268         } else {
3269                 ic->sb->log2_interleave_sectors = 0;
3270
3271                 get_provided_data_sectors(ic);
3272                 if (!ic->provided_data_sectors)
3273                         return -EINVAL;
3274
3275 try_smaller_buffer:
3276                 ic->sb->journal_sections = cpu_to_le32(0);
3277                 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3278                         __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3279                         __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3280                         if (test_journal_sections > journal_sections)
3281                                 continue;
3282                         ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3283                         if (calculate_device_limits(ic))
3284                                 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3285
3286                 }
3287                 if (!le32_to_cpu(ic->sb->journal_sections)) {
3288                         if (ic->log2_buffer_sectors > 3) {
3289                                 ic->log2_buffer_sectors--;
3290                                 goto try_smaller_buffer;
3291                         }
3292                         return -EINVAL;
3293                 }
3294         }
3295
3296         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3297
3298         sb_set_version(ic);
3299
3300         return 0;
3301 }
3302
3303 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3304 {
3305         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3306         struct blk_integrity bi;
3307
3308         memset(&bi, 0, sizeof(bi));
3309         bi.profile = &dm_integrity_profile;
3310         bi.tuple_size = ic->tag_size;
3311         bi.tag_size = bi.tuple_size;
3312         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3313
3314         blk_integrity_register(disk, &bi);
3315         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3316 }
3317
3318 static void dm_integrity_free_page_list(struct page_list *pl)
3319 {
3320         unsigned i;
3321
3322         if (!pl)
3323                 return;
3324         for (i = 0; pl[i].page; i++)
3325                 __free_page(pl[i].page);
3326         kvfree(pl);
3327 }
3328
3329 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3330 {
3331         struct page_list *pl;
3332         unsigned i;
3333
3334         pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3335         if (!pl)
3336                 return NULL;
3337
3338         for (i = 0; i < n_pages; i++) {
3339                 pl[i].page = alloc_page(GFP_KERNEL);
3340                 if (!pl[i].page) {
3341                         dm_integrity_free_page_list(pl);
3342                         return NULL;
3343                 }
3344                 if (i)
3345                         pl[i - 1].next = &pl[i];
3346         }
3347         pl[i].page = NULL;
3348         pl[i].next = NULL;
3349
3350         return pl;
3351 }
3352
3353 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3354 {
3355         unsigned i;
3356         for (i = 0; i < ic->journal_sections; i++)
3357                 kvfree(sl[i]);
3358         kvfree(sl);
3359 }
3360
3361 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3362                                                                    struct page_list *pl)
3363 {
3364         struct scatterlist **sl;
3365         unsigned i;
3366
3367         sl = kvmalloc_array(ic->journal_sections,
3368                             sizeof(struct scatterlist *),
3369                             GFP_KERNEL | __GFP_ZERO);
3370         if (!sl)
3371                 return NULL;
3372
3373         for (i = 0; i < ic->journal_sections; i++) {
3374                 struct scatterlist *s;
3375                 unsigned start_index, start_offset;
3376                 unsigned end_index, end_offset;
3377                 unsigned n_pages;
3378                 unsigned idx;
3379
3380                 page_list_location(ic, i, 0, &start_index, &start_offset);
3381                 page_list_location(ic, i, ic->journal_section_sectors - 1,
3382                                    &end_index, &end_offset);
3383
3384                 n_pages = (end_index - start_index + 1);
3385
3386                 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3387                                    GFP_KERNEL);
3388                 if (!s) {
3389                         dm_integrity_free_journal_scatterlist(ic, sl);
3390                         return NULL;
3391                 }
3392
3393                 sg_init_table(s, n_pages);
3394                 for (idx = start_index; idx <= end_index; idx++) {
3395                         char *va = lowmem_page_address(pl[idx].page);
3396                         unsigned start = 0, end = PAGE_SIZE;
3397                         if (idx == start_index)
3398                                 start = start_offset;
3399                         if (idx == end_index)
3400                                 end = end_offset + (1 << SECTOR_SHIFT);
3401                         sg_set_buf(&s[idx - start_index], va + start, end - start);
3402                 }
3403
3404                 sl[i] = s;
3405         }
3406
3407         return sl;
3408 }
3409
3410 static void free_alg(struct alg_spec *a)
3411 {
3412         kzfree(a->alg_string);
3413         kzfree(a->key);
3414         memset(a, 0, sizeof *a);
3415 }
3416
3417 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3418 {
3419         char *k;
3420
3421         free_alg(a);
3422
3423         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3424         if (!a->alg_string)
3425                 goto nomem;
3426
3427         k = strchr(a->alg_string, ':');
3428         if (k) {
3429                 *k = 0;
3430                 a->key_string = k + 1;
3431                 if (strlen(a->key_string) & 1)
3432                         goto inval;
3433
3434                 a->key_size = strlen(a->key_string) / 2;
3435                 a->key = kmalloc(a->key_size, GFP_KERNEL);
3436                 if (!a->key)
3437                         goto nomem;
3438                 if (hex2bin(a->key, a->key_string, a->key_size))
3439                         goto inval;
3440         }
3441
3442         return 0;
3443 inval:
3444         *error = error_inval;
3445         return -EINVAL;
3446 nomem:
3447         *error = "Out of memory for an argument";
3448         return -ENOMEM;
3449 }
3450
3451 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3452                    char *error_alg, char *error_key)
3453 {
3454         int r;
3455
3456         if (a->alg_string) {
3457                 *hash = crypto_alloc_shash(a->alg_string, 0, 0);
3458                 if (IS_ERR(*hash)) {
3459                         *error = error_alg;
3460                         r = PTR_ERR(*hash);
3461                         *hash = NULL;
3462                         return r;
3463                 }
3464
3465                 if (a->key) {
3466                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
3467                         if (r) {
3468                                 *error = error_key;
3469                                 return r;
3470                         }
3471                 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3472                         *error = error_key;
3473                         return -ENOKEY;
3474                 }
3475         }
3476
3477         return 0;
3478 }
3479
3480 static int create_journal(struct dm_integrity_c *ic, char **error)
3481 {
3482         int r = 0;
3483         unsigned i;
3484         __u64 journal_pages, journal_desc_size, journal_tree_size;
3485         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3486         struct skcipher_request *req = NULL;
3487
3488         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3489         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3490         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3491         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3492
3493         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3494                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3495         journal_desc_size = journal_pages * sizeof(struct page_list);
3496         if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3497                 *error = "Journal doesn't fit into memory";
3498                 r = -ENOMEM;
3499                 goto bad;
3500         }
3501         ic->journal_pages = journal_pages;
3502
3503         ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3504         if (!ic->journal) {
3505                 *error = "Could not allocate memory for journal";
3506                 r = -ENOMEM;
3507                 goto bad;
3508         }
3509         if (ic->journal_crypt_alg.alg_string) {
3510                 unsigned ivsize, blocksize;
3511                 struct journal_completion comp;
3512
3513                 comp.ic = ic;
3514                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
3515                 if (IS_ERR(ic->journal_crypt)) {
3516                         *error = "Invalid journal cipher";
3517                         r = PTR_ERR(ic->journal_crypt);
3518                         ic->journal_crypt = NULL;
3519                         goto bad;
3520                 }
3521                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3522                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3523
3524                 if (ic->journal_crypt_alg.key) {
3525                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3526                                                    ic->journal_crypt_alg.key_size);
3527                         if (r) {
3528                                 *error = "Error setting encryption key";
3529                                 goto bad;
3530                         }
3531                 }
3532                 DEBUG_print("cipher %s, block size %u iv size %u\n",
3533                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3534
3535                 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3536                 if (!ic->journal_io) {
3537                         *error = "Could not allocate memory for journal io";
3538                         r = -ENOMEM;
3539                         goto bad;
3540                 }
3541
3542                 if (blocksize == 1) {
3543                         struct scatterlist *sg;
3544
3545                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3546                         if (!req) {
3547                                 *error = "Could not allocate crypt request";
3548                                 r = -ENOMEM;
3549                                 goto bad;
3550                         }
3551
3552                         crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3553                         if (!crypt_iv) {
3554                                 *error = "Could not allocate iv";
3555                                 r = -ENOMEM;
3556                                 goto bad;
3557                         }
3558
3559                         ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3560                         if (!ic->journal_xor) {
3561                                 *error = "Could not allocate memory for journal xor";
3562                                 r = -ENOMEM;
3563                                 goto bad;
3564                         }
3565
3566                         sg = kvmalloc_array(ic->journal_pages + 1,
3567                                             sizeof(struct scatterlist),
3568                                             GFP_KERNEL);
3569                         if (!sg) {
3570                                 *error = "Unable to allocate sg list";
3571                                 r = -ENOMEM;
3572                                 goto bad;
3573                         }
3574                         sg_init_table(sg, ic->journal_pages + 1);
3575                         for (i = 0; i < ic->journal_pages; i++) {
3576                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
3577                                 clear_page(va);
3578                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
3579                         }
3580                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3581
3582                         skcipher_request_set_crypt(req, sg, sg,
3583                                                    PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3584                         init_completion(&comp.comp);
3585                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3586                         if (do_crypt(true, req, &comp))
3587                                 wait_for_completion(&comp.comp);
3588                         kvfree(sg);
3589                         r = dm_integrity_failed(ic);
3590                         if (r) {
3591                                 *error = "Unable to encrypt journal";
3592                                 goto bad;
3593                         }
3594                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3595
3596                         crypto_free_skcipher(ic->journal_crypt);
3597                         ic->journal_crypt = NULL;
3598                 } else {
3599                         unsigned crypt_len = roundup(ivsize, blocksize);
3600
3601                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3602                         if (!req) {
3603                                 *error = "Could not allocate crypt request";
3604                                 r = -ENOMEM;
3605                                 goto bad;
3606                         }
3607
3608                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3609                         if (!crypt_iv) {
3610                                 *error = "Could not allocate iv";
3611                                 r = -ENOMEM;
3612                                 goto bad;
3613                         }
3614
3615                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3616                         if (!crypt_data) {
3617                                 *error = "Unable to allocate crypt data";
3618                                 r = -ENOMEM;
3619                                 goto bad;
3620                         }
3621
3622                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3623                         if (!ic->journal_scatterlist) {
3624                                 *error = "Unable to allocate sg list";
3625                                 r = -ENOMEM;
3626                                 goto bad;
3627                         }
3628                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3629                         if (!ic->journal_io_scatterlist) {
3630                                 *error = "Unable to allocate sg list";
3631                                 r = -ENOMEM;
3632                                 goto bad;
3633                         }
3634                         ic->sk_requests = kvmalloc_array(ic->journal_sections,
3635                                                          sizeof(struct skcipher_request *),
3636                                                          GFP_KERNEL | __GFP_ZERO);
3637                         if (!ic->sk_requests) {
3638                                 *error = "Unable to allocate sk requests";
3639                                 r = -ENOMEM;
3640                                 goto bad;
3641                         }
3642                         for (i = 0; i < ic->journal_sections; i++) {
3643                                 struct scatterlist sg;
3644                                 struct skcipher_request *section_req;
3645                                 __u32 section_le = cpu_to_le32(i);
3646
3647                                 memset(crypt_iv, 0x00, ivsize);
3648                                 memset(crypt_data, 0x00, crypt_len);
3649                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
3650
3651                                 sg_init_one(&sg, crypt_data, crypt_len);
3652                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3653                                 init_completion(&comp.comp);
3654                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3655                                 if (do_crypt(true, req, &comp))
3656                                         wait_for_completion(&comp.comp);
3657
3658                                 r = dm_integrity_failed(ic);
3659                                 if (r) {
3660                                         *error = "Unable to generate iv";
3661                                         goto bad;
3662                                 }
3663
3664                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3665                                 if (!section_req) {
3666                                         *error = "Unable to allocate crypt request";
3667                                         r = -ENOMEM;
3668                                         goto bad;
3669                                 }
3670                                 section_req->iv = kmalloc_array(ivsize, 2,
3671                                                                 GFP_KERNEL);
3672                                 if (!section_req->iv) {
3673                                         skcipher_request_free(section_req);
3674                                         *error = "Unable to allocate iv";
3675                                         r = -ENOMEM;
3676                                         goto bad;
3677                                 }
3678                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3679                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3680                                 ic->sk_requests[i] = section_req;
3681                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3682                         }
3683                 }
3684         }
3685
3686         for (i = 0; i < N_COMMIT_IDS; i++) {
3687                 unsigned j;
3688 retest_commit_id:
3689                 for (j = 0; j < i; j++) {
3690                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
3691                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3692                                 goto retest_commit_id;
3693                         }
3694                 }
3695                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3696         }
3697
3698         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3699         if (journal_tree_size > ULONG_MAX) {
3700                 *error = "Journal doesn't fit into memory";
3701                 r = -ENOMEM;
3702                 goto bad;
3703         }
3704         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3705         if (!ic->journal_tree) {
3706                 *error = "Could not allocate memory for journal tree";
3707                 r = -ENOMEM;
3708         }
3709 bad:
3710         kfree(crypt_data);
3711         kfree(crypt_iv);
3712         skcipher_request_free(req);
3713
3714         return r;
3715 }
3716
3717 /*
3718  * Construct a integrity mapping
3719  *
3720  * Arguments:
3721  *      device
3722  *      offset from the start of the device
3723  *      tag size
3724  *      D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3725  *      number of optional arguments
3726  *      optional arguments:
3727  *              journal_sectors
3728  *              interleave_sectors
3729  *              buffer_sectors
3730  *              journal_watermark
3731  *              commit_time
3732  *              meta_device
3733  *              block_size
3734  *              sectors_per_bit
3735  *              bitmap_flush_interval
3736  *              internal_hash
3737  *              journal_crypt
3738  *              journal_mac
3739  *              recalculate
3740  */
3741 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3742 {
3743         struct dm_integrity_c *ic;
3744         char dummy;
3745         int r;
3746         unsigned extra_args;
3747         struct dm_arg_set as;
3748         static const struct dm_arg _args[] = {
3749                 {0, 9, "Invalid number of feature args"},
3750         };
3751         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3752         bool should_write_sb;
3753         __u64 threshold;
3754         unsigned long long start;
3755         __s8 log2_sectors_per_bitmap_bit = -1;
3756         __s8 log2_blocks_per_bitmap_bit;
3757         __u64 bits_in_journal;
3758         __u64 n_bitmap_bits;
3759
3760 #define DIRECT_ARGUMENTS        4
3761
3762         if (argc <= DIRECT_ARGUMENTS) {
3763                 ti->error = "Invalid argument count";
3764                 return -EINVAL;
3765         }
3766
3767         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3768         if (!ic) {
3769                 ti->error = "Cannot allocate integrity context";
3770                 return -ENOMEM;
3771         }
3772         ti->private = ic;
3773         ti->per_io_data_size = sizeof(struct dm_integrity_io);
3774         ic->ti = ti;
3775
3776         ic->in_progress = RB_ROOT;
3777         INIT_LIST_HEAD(&ic->wait_list);
3778         init_waitqueue_head(&ic->endio_wait);
3779         bio_list_init(&ic->flush_bio_list);
3780         init_waitqueue_head(&ic->copy_to_journal_wait);
3781         init_completion(&ic->crypto_backoff);
3782         atomic64_set(&ic->number_of_mismatches, 0);
3783         ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
3784
3785         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3786         if (r) {
3787                 ti->error = "Device lookup failed";
3788                 goto bad;
3789         }
3790
3791         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3792                 ti->error = "Invalid starting offset";
3793                 r = -EINVAL;
3794                 goto bad;
3795         }
3796         ic->start = start;
3797
3798         if (strcmp(argv[2], "-")) {
3799                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3800                         ti->error = "Invalid tag size";
3801                         r = -EINVAL;
3802                         goto bad;
3803                 }
3804         }
3805
3806         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
3807             !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
3808                 ic->mode = argv[3][0];
3809         } else {
3810                 ti->error = "Invalid mode (expecting J, B, D, R)";
3811                 r = -EINVAL;
3812                 goto bad;
3813         }
3814
3815         journal_sectors = 0;
3816         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3817         buffer_sectors = DEFAULT_BUFFER_SECTORS;
3818         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3819         sync_msec = DEFAULT_SYNC_MSEC;
3820         ic->sectors_per_block = 1;
3821
3822         as.argc = argc - DIRECT_ARGUMENTS;
3823         as.argv = argv + DIRECT_ARGUMENTS;
3824         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3825         if (r)
3826                 goto bad;
3827
3828         while (extra_args--) {
3829                 const char *opt_string;
3830                 unsigned val;
3831                 unsigned long long llval;
3832                 opt_string = dm_shift_arg(&as);
3833                 if (!opt_string) {
3834                         r = -EINVAL;
3835                         ti->error = "Not enough feature arguments";
3836                         goto bad;
3837                 }
3838                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3839                         journal_sectors = val ? val : 1;
3840                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3841                         interleave_sectors = val;
3842                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3843                         buffer_sectors = val;
3844                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3845                         journal_watermark = val;
3846                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3847                         sync_msec = val;
3848                 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3849                         if (ic->meta_dev) {
3850                                 dm_put_device(ti, ic->meta_dev);
3851                                 ic->meta_dev = NULL;
3852                         }
3853                         r = dm_get_device(ti, strchr(opt_string, ':') + 1,
3854                                           dm_table_get_mode(ti->table), &ic->meta_dev);
3855                         if (r) {
3856                                 ti->error = "Device lookup failed";
3857                                 goto bad;
3858                         }
3859                 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3860                         if (val < 1 << SECTOR_SHIFT ||
3861                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3862                             (val & (val -1))) {
3863                                 r = -EINVAL;
3864                                 ti->error = "Invalid block_size argument";
3865                                 goto bad;
3866                         }
3867                         ic->sectors_per_block = val >> SECTOR_SHIFT;
3868                 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
3869                         log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
3870                 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
3871                         if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
3872                                 r = -EINVAL;
3873                                 ti->error = "Invalid bitmap_flush_interval argument";
3874                         }
3875                         ic->bitmap_flush_interval = msecs_to_jiffies(val);
3876                 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3877                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3878                                             "Invalid internal_hash argument");
3879                         if (r)
3880                                 goto bad;
3881                 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3882                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3883                                             "Invalid journal_crypt argument");
3884                         if (r)
3885                                 goto bad;
3886                 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3887                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
3888                                             "Invalid journal_mac argument");
3889                         if (r)
3890                                 goto bad;
3891                 } else if (!strcmp(opt_string, "recalculate")) {
3892                         ic->recalculate_flag = true;
3893                 } else if (!strcmp(opt_string, "allow_discards")) {
3894                         ic->discard = true;
3895                 } else if (!strcmp(opt_string, "fix_padding")) {
3896                         ic->fix_padding = true;
3897                 } else {
3898                         r = -EINVAL;
3899                         ti->error = "Invalid argument";
3900                         goto bad;
3901                 }
3902         }
3903
3904         ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3905         if (!ic->meta_dev)
3906                 ic->meta_device_sectors = ic->data_device_sectors;
3907         else
3908                 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3909
3910         if (!journal_sectors) {
3911                 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3912                                       ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3913         }
3914
3915         if (!buffer_sectors)
3916                 buffer_sectors = 1;
3917         ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3918
3919         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3920                     "Invalid internal hash", "Error setting internal hash key");
3921         if (r)
3922                 goto bad;
3923
3924         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3925                     "Invalid journal mac", "Error setting journal mac key");
3926         if (r)
3927                 goto bad;
3928
3929         if (!ic->tag_size) {
3930                 if (!ic->internal_hash) {
3931                         ti->error = "Unknown tag size";
3932                         r = -EINVAL;
3933                         goto bad;
3934                 }
3935                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
3936         }
3937         if (ic->tag_size > MAX_TAG_SIZE) {
3938                 ti->error = "Too big tag size";
3939                 r = -EINVAL;
3940                 goto bad;
3941         }
3942         if (!(ic->tag_size & (ic->tag_size - 1)))
3943                 ic->log2_tag_size = __ffs(ic->tag_size);
3944         else
3945                 ic->log2_tag_size = -1;
3946
3947         if (ic->mode == 'B' && !ic->internal_hash) {
3948                 r = -EINVAL;
3949                 ti->error = "Bitmap mode can be only used with internal hash";
3950                 goto bad;
3951         }
3952
3953         if (ic->discard && !ic->internal_hash) {
3954                 r = -EINVAL;
3955                 ti->error = "Discard can be only used with internal hash";
3956                 goto bad;
3957         }
3958
3959         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
3960         ic->autocommit_msec = sync_msec;
3961         timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
3962
3963         ic->io = dm_io_client_create();
3964         if (IS_ERR(ic->io)) {
3965                 r = PTR_ERR(ic->io);
3966                 ic->io = NULL;
3967                 ti->error = "Cannot allocate dm io";
3968                 goto bad;
3969         }
3970
3971         r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
3972         if (r) {
3973                 ti->error = "Cannot allocate mempool";
3974                 goto bad;
3975         }
3976
3977         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
3978                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
3979         if (!ic->metadata_wq) {
3980                 ti->error = "Cannot allocate workqueue";
3981                 r = -ENOMEM;
3982                 goto bad;
3983         }
3984
3985         /*
3986          * If this workqueue were percpu, it would cause bio reordering
3987          * and reduced performance.
3988          */
3989         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3990         if (!ic->wait_wq) {
3991                 ti->error = "Cannot allocate workqueue";
3992                 r = -ENOMEM;
3993                 goto bad;
3994         }
3995
3996         ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
3997                                           METADATA_WORKQUEUE_MAX_ACTIVE);
3998         if (!ic->offload_wq) {
3999                 ti->error = "Cannot allocate workqueue";
4000                 r = -ENOMEM;
4001                 goto bad;
4002         }
4003
4004         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4005         if (!ic->commit_wq) {
4006                 ti->error = "Cannot allocate workqueue";
4007                 r = -ENOMEM;
4008                 goto bad;
4009         }
4010         INIT_WORK(&ic->commit_work, integrity_commit);
4011
4012         if (ic->mode == 'J' || ic->mode == 'B') {
4013                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4014                 if (!ic->writer_wq) {
4015                         ti->error = "Cannot allocate workqueue";
4016                         r = -ENOMEM;
4017                         goto bad;
4018                 }
4019                 INIT_WORK(&ic->writer_work, integrity_writer);
4020         }
4021
4022         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4023         if (!ic->sb) {
4024                 r = -ENOMEM;
4025                 ti->error = "Cannot allocate superblock area";
4026                 goto bad;
4027         }
4028
4029         r = sync_rw_sb(ic, REQ_OP_READ, 0);
4030         if (r) {
4031                 ti->error = "Error reading superblock";
4032                 goto bad;
4033         }
4034         should_write_sb = false;
4035         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4036                 if (ic->mode != 'R') {
4037                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4038                                 r = -EINVAL;
4039                                 ti->error = "The device is not initialized";
4040                                 goto bad;
4041                         }
4042                 }
4043
4044                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4045                 if (r) {
4046                         ti->error = "Could not initialize superblock";
4047                         goto bad;
4048                 }
4049                 if (ic->mode != 'R')
4050                         should_write_sb = true;
4051         }
4052
4053         if (!ic->sb->version || ic->sb->version > SB_VERSION_4) {
4054                 r = -EINVAL;
4055                 ti->error = "Unknown version";
4056                 goto bad;
4057         }
4058         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4059                 r = -EINVAL;
4060                 ti->error = "Tag size doesn't match the information in superblock";
4061                 goto bad;
4062         }
4063         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4064                 r = -EINVAL;
4065                 ti->error = "Block size doesn't match the information in superblock";
4066                 goto bad;
4067         }
4068         if (!le32_to_cpu(ic->sb->journal_sections)) {
4069                 r = -EINVAL;
4070                 ti->error = "Corrupted superblock, journal_sections is 0";
4071                 goto bad;
4072         }
4073         /* make sure that ti->max_io_len doesn't overflow */
4074         if (!ic->meta_dev) {
4075                 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4076                     ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4077                         r = -EINVAL;
4078                         ti->error = "Invalid interleave_sectors in the superblock";
4079                         goto bad;
4080                 }
4081         } else {
4082                 if (ic->sb->log2_interleave_sectors) {
4083                         r = -EINVAL;
4084                         ti->error = "Invalid interleave_sectors in the superblock";
4085                         goto bad;
4086                 }
4087         }
4088         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4089                 r = -EINVAL;
4090                 ti->error = "Journal mac mismatch";
4091                 goto bad;
4092         }
4093
4094         get_provided_data_sectors(ic);
4095         if (!ic->provided_data_sectors) {
4096                 r = -EINVAL;
4097                 ti->error = "The device is too small";
4098                 goto bad;
4099         }
4100
4101 try_smaller_buffer:
4102         r = calculate_device_limits(ic);
4103         if (r) {
4104                 if (ic->meta_dev) {
4105                         if (ic->log2_buffer_sectors > 3) {
4106                                 ic->log2_buffer_sectors--;
4107                                 goto try_smaller_buffer;
4108                         }
4109                 }
4110                 ti->error = "The device is too small";
4111                 goto bad;
4112         }
4113
4114         if (log2_sectors_per_bitmap_bit < 0)
4115                 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4116         if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4117                 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4118
4119         bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4120         if (bits_in_journal > UINT_MAX)
4121                 bits_in_journal = UINT_MAX;
4122         while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4123                 log2_sectors_per_bitmap_bit++;
4124
4125         log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4126         ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4127         if (should_write_sb) {
4128                 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4129         }
4130         n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4131                                 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4132         ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4133
4134         if (!ic->meta_dev)
4135                 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4136
4137         if (ti->len > ic->provided_data_sectors) {
4138                 r = -EINVAL;
4139                 ti->error = "Not enough provided sectors for requested mapping size";
4140                 goto bad;
4141         }
4142
4143
4144         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4145         threshold += 50;
4146         do_div(threshold, 100);
4147         ic->free_sectors_threshold = threshold;
4148
4149         DEBUG_print("initialized:\n");
4150         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4151         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
4152         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4153         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
4154         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
4155         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
4156         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
4157         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4158         DEBUG_print("   data_device_sectors 0x%llx\n", i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT);
4159         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
4160         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
4161         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
4162         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4163         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4164         DEBUG_print("   bits_in_journal %llu\n", bits_in_journal);
4165
4166         if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4167                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4168                 ic->sb->recalc_sector = cpu_to_le64(0);
4169         }
4170
4171         if (ic->internal_hash) {
4172                 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4173                 if (!ic->recalc_wq ) {
4174                         ti->error = "Cannot allocate workqueue";
4175                         r = -ENOMEM;
4176                         goto bad;
4177                 }
4178                 INIT_WORK(&ic->recalc_work, integrity_recalc);
4179                 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4180                 if (!ic->recalc_buffer) {
4181                         ti->error = "Cannot allocate buffer for recalculating";
4182                         r = -ENOMEM;
4183                         goto bad;
4184                 }
4185                 ic->recalc_tags = kvmalloc_array(RECALC_SECTORS >> ic->sb->log2_sectors_per_block,
4186                                                  ic->tag_size, GFP_KERNEL);
4187                 if (!ic->recalc_tags) {
4188                         ti->error = "Cannot allocate tags for recalculating";
4189                         r = -ENOMEM;
4190                         goto bad;
4191                 }
4192         }
4193
4194         ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4195                         1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
4196         if (IS_ERR(ic->bufio)) {
4197                 r = PTR_ERR(ic->bufio);
4198                 ti->error = "Cannot initialize dm-bufio";
4199                 ic->bufio = NULL;
4200                 goto bad;
4201         }
4202         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4203
4204         if (ic->mode != 'R') {
4205                 r = create_journal(ic, &ti->error);
4206                 if (r)
4207                         goto bad;
4208
4209         }
4210
4211         if (ic->mode == 'B') {
4212                 unsigned i;
4213                 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4214
4215                 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4216                 if (!ic->recalc_bitmap) {
4217                         r = -ENOMEM;
4218                         goto bad;
4219                 }
4220                 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4221                 if (!ic->may_write_bitmap) {
4222                         r = -ENOMEM;
4223                         goto bad;
4224                 }
4225                 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4226                 if (!ic->bbs) {
4227                         r = -ENOMEM;
4228                         goto bad;
4229                 }
4230                 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4231                 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4232                         struct bitmap_block_status *bbs = &ic->bbs[i];
4233                         unsigned sector, pl_index, pl_offset;
4234
4235                         INIT_WORK(&bbs->work, bitmap_block_work);
4236                         bbs->ic = ic;
4237                         bbs->idx = i;
4238                         bio_list_init(&bbs->bio_queue);
4239                         spin_lock_init(&bbs->bio_queue_lock);
4240
4241                         sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4242                         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4243                         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4244
4245                         bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4246                 }
4247         }
4248
4249         if (should_write_sb) {
4250                 int r;
4251
4252                 init_journal(ic, 0, ic->journal_sections, 0);
4253                 r = dm_integrity_failed(ic);
4254                 if (unlikely(r)) {
4255                         ti->error = "Error initializing journal";
4256                         goto bad;
4257                 }
4258                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4259                 if (r) {
4260                         ti->error = "Error initializing superblock";
4261                         goto bad;
4262                 }
4263                 ic->just_formatted = true;
4264         }
4265
4266         if (!ic->meta_dev) {
4267                 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4268                 if (r)
4269                         goto bad;
4270         }
4271         if (ic->mode == 'B') {
4272                 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4273                 if (!max_io_len)
4274                         max_io_len = 1U << 31;
4275                 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4276                 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4277                         r = dm_set_target_max_io_len(ti, max_io_len);
4278                         if (r)
4279                                 goto bad;
4280                 }
4281         }
4282
4283         if (!ic->internal_hash)
4284                 dm_integrity_set(ti, ic);
4285
4286         ti->num_flush_bios = 1;
4287         ti->flush_supported = true;
4288         if (ic->discard)
4289                 ti->num_discard_bios = 1;
4290
4291         return 0;
4292
4293 bad:
4294         dm_integrity_dtr(ti);
4295         return r;
4296 }
4297
4298 static void dm_integrity_dtr(struct dm_target *ti)
4299 {
4300         struct dm_integrity_c *ic = ti->private;
4301
4302         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4303         BUG_ON(!list_empty(&ic->wait_list));
4304
4305         if (ic->metadata_wq)
4306                 destroy_workqueue(ic->metadata_wq);
4307         if (ic->wait_wq)
4308                 destroy_workqueue(ic->wait_wq);
4309         if (ic->offload_wq)
4310                 destroy_workqueue(ic->offload_wq);
4311         if (ic->commit_wq)
4312                 destroy_workqueue(ic->commit_wq);
4313         if (ic->writer_wq)
4314                 destroy_workqueue(ic->writer_wq);
4315         if (ic->recalc_wq)
4316                 destroy_workqueue(ic->recalc_wq);
4317         vfree(ic->recalc_buffer);
4318         kvfree(ic->recalc_tags);
4319         kvfree(ic->bbs);
4320         if (ic->bufio)
4321                 dm_bufio_client_destroy(ic->bufio);
4322         mempool_exit(&ic->journal_io_mempool);
4323         if (ic->io)
4324                 dm_io_client_destroy(ic->io);
4325         if (ic->dev)
4326                 dm_put_device(ti, ic->dev);
4327         if (ic->meta_dev)
4328                 dm_put_device(ti, ic->meta_dev);
4329         dm_integrity_free_page_list(ic->journal);
4330         dm_integrity_free_page_list(ic->journal_io);
4331         dm_integrity_free_page_list(ic->journal_xor);
4332         dm_integrity_free_page_list(ic->recalc_bitmap);
4333         dm_integrity_free_page_list(ic->may_write_bitmap);
4334         if (ic->journal_scatterlist)
4335                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4336         if (ic->journal_io_scatterlist)
4337                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4338         if (ic->sk_requests) {
4339                 unsigned i;
4340
4341                 for (i = 0; i < ic->journal_sections; i++) {
4342                         struct skcipher_request *req = ic->sk_requests[i];
4343                         if (req) {
4344                                 kzfree(req->iv);
4345                                 skcipher_request_free(req);
4346                         }
4347                 }
4348                 kvfree(ic->sk_requests);
4349         }
4350         kvfree(ic->journal_tree);
4351         if (ic->sb)
4352                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4353
4354         if (ic->internal_hash)
4355                 crypto_free_shash(ic->internal_hash);
4356         free_alg(&ic->internal_hash_alg);
4357
4358         if (ic->journal_crypt)
4359                 crypto_free_skcipher(ic->journal_crypt);
4360         free_alg(&ic->journal_crypt_alg);
4361
4362         if (ic->journal_mac)
4363                 crypto_free_shash(ic->journal_mac);
4364         free_alg(&ic->journal_mac_alg);
4365
4366         kfree(ic);
4367 }
4368
4369 static struct target_type integrity_target = {
4370         .name                   = "integrity",
4371         .version                = {1, 6, 0},
4372         .module                 = THIS_MODULE,
4373         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4374         .ctr                    = dm_integrity_ctr,
4375         .dtr                    = dm_integrity_dtr,
4376         .map                    = dm_integrity_map,
4377         .postsuspend            = dm_integrity_postsuspend,
4378         .resume                 = dm_integrity_resume,
4379         .status                 = dm_integrity_status,
4380         .iterate_devices        = dm_integrity_iterate_devices,
4381         .io_hints               = dm_integrity_io_hints,
4382 };
4383
4384 static int __init dm_integrity_init(void)
4385 {
4386         int r;
4387
4388         journal_io_cache = kmem_cache_create("integrity_journal_io",
4389                                              sizeof(struct journal_io), 0, 0, NULL);
4390         if (!journal_io_cache) {
4391                 DMERR("can't allocate journal io cache");
4392                 return -ENOMEM;
4393         }
4394
4395         r = dm_register_target(&integrity_target);
4396
4397         if (r < 0)
4398                 DMERR("register failed %d", r);
4399
4400         return r;
4401 }
4402
4403 static void __exit dm_integrity_exit(void)
4404 {
4405         dm_unregister_target(&integrity_target);
4406         kmem_cache_destroy(journal_io_cache);
4407 }
4408
4409 module_init(dm_integrity_init);
4410 module_exit(dm_integrity_exit);
4411
4412 MODULE_AUTHOR("Milan Broz");
4413 MODULE_AUTHOR("Mikulas Patocka");
4414 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4415 MODULE_LICENSE("GPL");