block/blklogwrites.c

   1 /*
   2  * Write logging blk driver based on blkverify and blkdebug.
   3  *
   4  * Copyright (c) 2017 Tuomas Tynkkynen <tuomas@tuxera.com>
   5  * Copyright (c) 2018 Aapo Vienamo <aapo@tuxera.com>
   6  * Copyright (c) 2018 Ari Sundholm <ari@tuxera.com>
   7  *
   8  * This work is licensed under the terms of the GNU GPL, version 2 or later.
   9  * See the COPYING file in the top-level directory.
  10  */
  11
  12 #include "qemu/osdep.h"
  13 #include "qapi/error.h"
  14 #include "qemu/sockets.h" /* for EINPROGRESS on Windows */
  15 #include "block/block_int.h"
  16 #include "qapi/qmp/qdict.h"
  17 #include "qapi/qmp/qstring.h"
  18 #include "qemu/cutils.h"
  19 #include "qemu/module.h"
  20 #include "qemu/option.h"
  21
  22 /* Disk format stuff - taken from Linux drivers/md/dm-log-writes.c */
  23
  24 #define LOG_FLUSH_FLAG   (1 << 0)
  25 #define LOG_FUA_FLAG     (1 << 1)
  26 #define LOG_DISCARD_FLAG (1 << 2)
  27 #define LOG_MARK_FLAG    (1 << 3)
  28 #define LOG_FLAG_MASK    (LOG_FLUSH_FLAG \
  29                          | LOG_FUA_FLAG \
  30                          | LOG_DISCARD_FLAG \
  31                          | LOG_MARK_FLAG)
  32
  33 #define WRITE_LOG_VERSION 1ULL
  34 #define WRITE_LOG_MAGIC 0x6a736677736872ULL
  35
  36 /* All fields are little-endian. */
  37 struct log_write_super {
  38     uint64_t magic;
  39     uint64_t version;
  40     uint64_t nr_entries;
  41     uint32_t sectorsize;
  42 } QEMU_PACKED;
  43
  44 struct log_write_entry {
  45     uint64_t sector;
  46     uint64_t nr_sectors;
  47     uint64_t flags;
  48     uint64_t data_len;
  49 } QEMU_PACKED;
  50
  51 /* End of disk format structures. */
  52
  53 typedef struct {
  54     BdrvChild *log_file;
  55     uint32_t sectorsize;
  56     uint32_t sectorbits;
  57     uint64_t cur_log_sector;
  58     uint64_t nr_entries;
  59     uint64_t update_interval;
  60 } BDRVBlkLogWritesState;
  61
  62 static QemuOptsList runtime_opts = {
  63     .name = "blklogwrites",
  64     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
  65     .desc = {
  66         {
  67             .name = "log-append",
  68             .type = QEMU_OPT_BOOL,
  69             .help = "Append to an existing log",
  70         },
  71         {
  72             .name = "log-sector-size",
  73             .type = QEMU_OPT_SIZE,
  74             .help = "Log sector size",
  75         },
  76         {
  77             .name = "log-super-update-interval",
  78             .type = QEMU_OPT_NUMBER,
  79             .help = "Log superblock update interval (# of write requests)",
  80         },
  81         { /* end of list */ }
  82     },
  83 };
  84
  85 static inline uint32_t blk_log_writes_log2(uint32_t value)
  86 {
  87     assert(value > 0);
  88     return 31 - clz32(value);
  89 }
  90
  91 static inline bool blk_log_writes_sector_size_valid(uint32_t sector_size)
  92 {
  93     return is_power_of_2(sector_size) &&
  94         sector_size >= sizeof(struct log_write_super) &&
  95         sector_size >= sizeof(struct log_write_entry) &&
  96         sector_size < (1ull << 24);
  97 }
  98
  99 static uint64_t blk_log_writes_find_cur_log_sector(BdrvChild *log,
 100                                                    uint32_t sector_size,
 101                                                    uint64_t nr_entries,
 102                                                    Error **errp)
 103 {
 104     uint64_t cur_sector = 1;
 105     uint64_t cur_idx = 0;
 106     uint32_t sector_bits = blk_log_writes_log2(sector_size);
 107     struct log_write_entry cur_entry;
 108
 109     while (cur_idx < nr_entries) {
 110         int read_ret = bdrv_pread(log, cur_sector << sector_bits, &cur_entry,
 111                                   sizeof(cur_entry));
 112         if (read_ret < 0) {
 113             error_setg_errno(errp, -read_ret,
 114                              "Failed to read log entry %"PRIu64, cur_idx);
 115             return (uint64_t)-1ull;
 116         }
 117
 118         if (cur_entry.flags & ~cpu_to_le64(LOG_FLAG_MASK)) {
 119             error_setg(errp, "Invalid flags 0x%"PRIx64" in log entry %"PRIu64,
 120                        le64_to_cpu(cur_entry.flags), cur_idx);
 121             return (uint64_t)-1ull;
 122         }
 123
 124         /* Account for the sector of the entry itself */
 125         ++cur_sector;
 126
 127         /*
 128          * Account for the data of the write.
 129          * For discards, this data is not present.
 130          */
 131         if (!(cur_entry.flags & cpu_to_le64(LOG_DISCARD_FLAG))) {
 132             cur_sector += le64_to_cpu(cur_entry.nr_sectors);
 133         }
 134
 135         ++cur_idx;
 136     }
 137
 138     return cur_sector;
 139 }
 140
 141 static int blk_log_writes_open(BlockDriverState *bs, QDict *options, int flags,
 142                                Error **errp)
 143 {
 144     BDRVBlkLogWritesState *s = bs->opaque;
 145     QemuOpts *opts;
 146     Error *local_err = NULL;
 147     int ret;
 148     uint64_t log_sector_size;
 149     bool log_append;
 150
 151     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 152     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
 153         ret = -EINVAL;
 154         goto fail;
 155     }
 156
 157     /* Open the file */
 158     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
 159                                BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY, false,
 160                                errp);
 161     if (!bs->file) {
 162         ret = -EINVAL;
 163         goto fail;
 164     }
 165
 166     /* Open the log file */
 167     s->log_file = bdrv_open_child(NULL, options, "log", bs, &child_of_bds,
 168                                   BDRV_CHILD_METADATA, false, errp);
 169     if (!s->log_file) {
 170         ret = -EINVAL;
 171         goto fail;
 172     }
 173
 174     log_append = qemu_opt_get_bool(opts, "log-append", false);
 175
 176     if (log_append) {
 177         struct log_write_super log_sb = { 0, 0, 0, 0 };
 178
 179         if (qemu_opt_find(opts, "log-sector-size")) {
 180             ret = -EINVAL;
 181             error_setg(errp, "log-append and log-sector-size are mutually "
 182                        "exclusive");
 183             goto fail_log;
 184         }
 185
 186         /* Read log superblock or fake one for an empty log */
 187         if (!bdrv_getlength(s->log_file->bs)) {
 188             log_sb.magic      = cpu_to_le64(WRITE_LOG_MAGIC);
 189             log_sb.version    = cpu_to_le64(WRITE_LOG_VERSION);
 190             log_sb.nr_entries = cpu_to_le64(0);
 191             log_sb.sectorsize = cpu_to_le32(BDRV_SECTOR_SIZE);
 192         } else {
 193             ret = bdrv_pread(s->log_file, 0, &log_sb, sizeof(log_sb));
 194             if (ret < 0) {
 195                 error_setg_errno(errp, -ret, "Could not read log superblock");
 196                 goto fail_log;
 197             }
 198         }
 199
 200         if (log_sb.magic != cpu_to_le64(WRITE_LOG_MAGIC)) {
 201             ret = -EINVAL;
 202             error_setg(errp, "Invalid log superblock magic");
 203             goto fail_log;
 204         }
 205
 206         if (log_sb.version != cpu_to_le64(WRITE_LOG_VERSION)) {
 207             ret = -EINVAL;
 208             error_setg(errp, "Unsupported log version %"PRIu64,
 209                        le64_to_cpu(log_sb.version));
 210             goto fail_log;
 211         }
 212
 213         log_sector_size = le32_to_cpu(log_sb.sectorsize);
 214         s->cur_log_sector = 1;
 215         s->nr_entries = 0;
 216
 217         if (blk_log_writes_sector_size_valid(log_sector_size)) {
 218             s->cur_log_sector =
 219                 blk_log_writes_find_cur_log_sector(s->log_file, log_sector_size,
 220                                     le64_to_cpu(log_sb.nr_entries), &local_err);
 221             if (local_err) {
 222                 ret = -EINVAL;
 223                 error_propagate(errp, local_err);
 224                 goto fail_log;
 225             }
 226
 227             s->nr_entries = le64_to_cpu(log_sb.nr_entries);
 228         }
 229     } else {
 230         log_sector_size = qemu_opt_get_size(opts, "log-sector-size",
 231                                             BDRV_SECTOR_SIZE);
 232         s->cur_log_sector = 1;
 233         s->nr_entries = 0;
 234     }
 235
 236     if (!blk_log_writes_sector_size_valid(log_sector_size)) {
 237         ret = -EINVAL;
 238         error_setg(errp, "Invalid log sector size %"PRIu64, log_sector_size);
 239         goto fail_log;
 240     }
 241
 242     s->sectorsize = log_sector_size;
 243     s->sectorbits = blk_log_writes_log2(log_sector_size);
 244     s->update_interval = qemu_opt_get_number(opts, "log-super-update-interval",
 245                                              4096);
 246     if (!s->update_interval) {
 247         ret = -EINVAL;
 248         error_setg(errp, "Invalid log superblock update interval %"PRIu64,
 249                    s->update_interval);
 250         goto fail_log;
 251     }
 252
 253     ret = 0;
 254 fail_log:
 255     if (ret < 0) {
 256         bdrv_unref_child(bs, s->log_file);
 257         s->log_file = NULL;
 258     }
 259 fail:
 260     if (ret < 0) {
 261         bdrv_unref_child(bs, bs->file);
 262         bs->file = NULL;
 263     }
 264     qemu_opts_del(opts);
 265     return ret;
 266 }
 267
 268 static void blk_log_writes_close(BlockDriverState *bs)
 269 {
 270     BDRVBlkLogWritesState *s = bs->opaque;
 271
 272     bdrv_unref_child(bs, s->log_file);
 273     s->log_file = NULL;
 274 }
 275
 276 static int64_t blk_log_writes_getlength(BlockDriverState *bs)
 277 {
 278     return bdrv_getlength(bs->file->bs);
 279 }
 280
 281 static void blk_log_writes_child_perm(BlockDriverState *bs, BdrvChild *c,
 282                                       BdrvChildRole role,
 283                                       BlockReopenQueue *ro_q,
 284                                       uint64_t perm, uint64_t shrd,
 285                                       uint64_t *nperm, uint64_t *nshrd)
 286 {
 287     if (!c) {
 288         *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
 289         *nshrd = (shrd & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
 290         return;
 291     }
 292
 293     bdrv_default_perms(bs, c, role, ro_q, perm, shrd,
 294                        nperm, nshrd);
 295 }
 296
 297 static void blk_log_writes_refresh_limits(BlockDriverState *bs, Error **errp)
 298 {
 299     BDRVBlkLogWritesState *s = bs->opaque;
 300     bs->bl.request_alignment = s->sectorsize;
 301 }
 302
 303 static int coroutine_fn
 304 blk_log_writes_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
 305                          QEMUIOVector *qiov, BdrvRequestFlags flags)
 306 {
 307     return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
 308 }
 309
 310 typedef struct BlkLogWritesFileReq {
 311     BlockDriverState *bs;
 312     uint64_t offset;
 313     uint64_t bytes;
 314     int file_flags;
 315     QEMUIOVector *qiov;
 316     int (*func)(struct BlkLogWritesFileReq *r);
 317     int file_ret;
 318 } BlkLogWritesFileReq;
 319
 320 typedef struct {
 321     BlockDriverState *bs;
 322     QEMUIOVector *qiov;
 323     struct log_write_entry entry;
 324     uint64_t zero_size;
 325     int log_ret;
 326 } BlkLogWritesLogReq;
 327
 328 static void coroutine_fn blk_log_writes_co_do_log(BlkLogWritesLogReq *lr)
 329 {
 330     BDRVBlkLogWritesState *s = lr->bs->opaque;
 331     uint64_t cur_log_offset = s->cur_log_sector << s->sectorbits;
 332
 333     s->nr_entries++;
 334     s->cur_log_sector +=
 335             ROUND_UP(lr->qiov->size, s->sectorsize) >> s->sectorbits;
 336
 337     lr->log_ret = bdrv_co_pwritev(s->log_file, cur_log_offset, lr->qiov->size,
 338                                   lr->qiov, 0);
 339
 340     /* Logging for the "write zeroes" operation */
 341     if (lr->log_ret == 0 && lr->zero_size) {
 342         cur_log_offset = s->cur_log_sector << s->sectorbits;
 343         s->cur_log_sector +=
 344                 ROUND_UP(lr->zero_size, s->sectorsize) >> s->sectorbits;
 345
 346         lr->log_ret = bdrv_co_pwrite_zeroes(s->log_file, cur_log_offset,
 347                                             lr->zero_size, 0);
 348     }
 349
 350     /* Update super block on flush or every update interval */
 351     if (lr->log_ret == 0 && ((lr->entry.flags & LOG_FLUSH_FLAG)
 352         || (s->nr_entries % s->update_interval == 0)))
 353     {
 354         struct log_write_super super = {
 355             .magic      = cpu_to_le64(WRITE_LOG_MAGIC),
 356             .version    = cpu_to_le64(WRITE_LOG_VERSION),
 357             .nr_entries = cpu_to_le64(s->nr_entries),
 358             .sectorsize = cpu_to_le32(s->sectorsize),
 359         };
 360         void *zeroes = g_malloc0(s->sectorsize - sizeof(super));
 361         QEMUIOVector qiov;
 362
 363         qemu_iovec_init(&qiov, 2);
 364         qemu_iovec_add(&qiov, &super, sizeof(super));
 365         qemu_iovec_add(&qiov, zeroes, s->sectorsize - sizeof(super));
 366
 367         lr->log_ret =
 368             bdrv_co_pwritev(s->log_file, 0, s->sectorsize, &qiov, 0);
 369         if (lr->log_ret == 0) {
 370             lr->log_ret = bdrv_co_flush(s->log_file->bs);
 371         }
 372         qemu_iovec_destroy(&qiov);
 373         g_free(zeroes);
 374     }
 375 }
 376
 377 static void coroutine_fn blk_log_writes_co_do_file(BlkLogWritesFileReq *fr)
 378 {
 379     fr->file_ret = fr->func(fr);
 380 }
 381
 382 static int coroutine_fn
 383 blk_log_writes_co_log(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
 384                       QEMUIOVector *qiov, int flags,
 385                       int (*file_func)(BlkLogWritesFileReq *r),
 386                       uint64_t entry_flags, bool is_zero_write)
 387 {
 388     QEMUIOVector log_qiov;
 389     size_t niov = qiov ? qiov->niov : 0;
 390     BDRVBlkLogWritesState *s = bs->opaque;
 391     BlkLogWritesFileReq fr = {
 392         .bs         = bs,
 393         .offset     = offset,
 394         .bytes      = bytes,
 395         .file_flags = flags,
 396         .qiov       = qiov,
 397         .func       = file_func,
 398     };
 399     BlkLogWritesLogReq lr = {
 400         .bs             = bs,
 401         .qiov           = &log_qiov,
 402         .entry = {
 403             .sector     = cpu_to_le64(offset >> s->sectorbits),
 404             .nr_sectors = cpu_to_le64(bytes >> s->sectorbits),
 405             .flags      = cpu_to_le64(entry_flags),
 406             .data_len   = 0,
 407         },
 408         .zero_size = is_zero_write ? bytes : 0,
 409     };
 410     void *zeroes = g_malloc0(s->sectorsize - sizeof(lr.entry));
 411
 412     assert((1 << s->sectorbits) == s->sectorsize);
 413     assert(bs->bl.request_alignment == s->sectorsize);
 414     assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
 415     assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
 416
 417     qemu_iovec_init(&log_qiov, niov + 2);
 418     qemu_iovec_add(&log_qiov, &lr.entry, sizeof(lr.entry));
 419     qemu_iovec_add(&log_qiov, zeroes, s->sectorsize - sizeof(lr.entry));
 420     if (qiov) {
 421         qemu_iovec_concat(&log_qiov, qiov, 0, qiov->size);
 422     }
 423
 424     blk_log_writes_co_do_file(&fr);
 425     blk_log_writes_co_do_log(&lr);
 426
 427     qemu_iovec_destroy(&log_qiov);
 428     g_free(zeroes);
 429
 430     if (lr.log_ret < 0) {
 431         return lr.log_ret;
 432     }
 433
 434     return fr.file_ret;
 435 }
 436
 437 static int coroutine_fn
 438 blk_log_writes_co_do_file_pwritev(BlkLogWritesFileReq *fr)
 439 {
 440     return bdrv_co_pwritev(fr->bs->file, fr->offset, fr->bytes,
 441                            fr->qiov, fr->file_flags);
 442 }
 443
 444 static int coroutine_fn
 445 blk_log_writes_co_do_file_pwrite_zeroes(BlkLogWritesFileReq *fr)
 446 {
 447     return bdrv_co_pwrite_zeroes(fr->bs->file, fr->offset, fr->bytes,
 448                                  fr->file_flags);
 449 }
 450
 451 static int coroutine_fn blk_log_writes_co_do_file_flush(BlkLogWritesFileReq *fr)
 452 {
 453     return bdrv_co_flush(fr->bs->file->bs);
 454 }
 455
 456 static int coroutine_fn
 457 blk_log_writes_co_do_file_pdiscard(BlkLogWritesFileReq *fr)
 458 {
 459     return bdrv_co_pdiscard(fr->bs->file, fr->offset, fr->bytes);
 460 }
 461
 462 static int coroutine_fn
 463 blk_log_writes_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
 464                           QEMUIOVector *qiov, BdrvRequestFlags flags)
 465 {
 466     return blk_log_writes_co_log(bs, offset, bytes, qiov, flags,
 467                                  blk_log_writes_co_do_file_pwritev, 0, false);
 468 }
 469
 470 static int coroutine_fn
 471 blk_log_writes_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
 472                                 int64_t bytes, BdrvRequestFlags flags)
 473 {
 474     return blk_log_writes_co_log(bs, offset, bytes, NULL, flags,
 475                                  blk_log_writes_co_do_file_pwrite_zeroes, 0,
 476                                  true);
 477 }
 478
 479 static int coroutine_fn blk_log_writes_co_flush_to_disk(BlockDriverState *bs)
 480 {
 481     return blk_log_writes_co_log(bs, 0, 0, NULL, 0,
 482                                  blk_log_writes_co_do_file_flush,
 483                                  LOG_FLUSH_FLAG, false);
 484 }
 485
 486 static int coroutine_fn
 487 blk_log_writes_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
 488 {
 489     return blk_log_writes_co_log(bs, offset, bytes, NULL, 0,
 490                                  blk_log_writes_co_do_file_pdiscard,
 491                                  LOG_DISCARD_FLAG, false);
 492 }
 493
 494 static const char *const blk_log_writes_strong_runtime_opts[] = {
 495     "log-append",
 496     "log-sector-size",
 497
 498     NULL
 499 };
 500
 501 static BlockDriver bdrv_blk_log_writes = {
 502     .format_name            = "blklogwrites",
 503     .instance_size          = sizeof(BDRVBlkLogWritesState),
 504
 505     .bdrv_open              = blk_log_writes_open,
 506     .bdrv_close             = blk_log_writes_close,
 507     .bdrv_getlength         = blk_log_writes_getlength,
 508     .bdrv_child_perm        = blk_log_writes_child_perm,
 509     .bdrv_refresh_limits    = blk_log_writes_refresh_limits,
 510
 511     .bdrv_co_preadv         = blk_log_writes_co_preadv,
 512     .bdrv_co_pwritev        = blk_log_writes_co_pwritev,
 513     .bdrv_co_pwrite_zeroes  = blk_log_writes_co_pwrite_zeroes,
 514     .bdrv_co_flush_to_disk  = blk_log_writes_co_flush_to_disk,
 515     .bdrv_co_pdiscard       = blk_log_writes_co_pdiscard,
 516
 517     .is_filter              = true,
 518     .strong_runtime_opts    = blk_log_writes_strong_runtime_opts,
 519 };
 520
 521 static void bdrv_blk_log_writes_init(void)
 522 {
 523     bdrv_register(&bdrv_blk_log_writes);
 524 }
 525
 526 block_init(bdrv_blk_log_writes_init);