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/option.h" 20 21 /* Disk format stuff - taken from Linux drivers/md/dm-log-writes.c */ 22 23 #define LOG_FLUSH_FLAG (1 << 0) 24 #define LOG_FUA_FLAG (1 << 1) 25 #define LOG_DISCARD_FLAG (1 << 2) 26 #define LOG_MARK_FLAG (1 << 3) 27 #define LOG_FLAG_MASK (LOG_FLUSH_FLAG \ 28 | LOG_FUA_FLAG \ 29 | LOG_DISCARD_FLAG \ 30 | LOG_MARK_FLAG) 31 32 #define WRITE_LOG_VERSION 1ULL 33 #define WRITE_LOG_MAGIC 0x6a736677736872ULL 34 35 /* All fields are little-endian. */ 36 struct log_write_super { 37 uint64_t magic; 38 uint64_t version; 39 uint64_t nr_entries; 40 uint32_t sectorsize; 41 } QEMU_PACKED; 42 43 struct log_write_entry { 44 uint64_t sector; 45 uint64_t nr_sectors; 46 uint64_t flags; 47 uint64_t data_len; 48 } QEMU_PACKED; 49 50 /* End of disk format structures. */ 51 52 typedef struct { 53 BdrvChild *log_file; 54 uint32_t sectorsize; 55 uint32_t sectorbits; 56 uint64_t cur_log_sector; 57 uint64_t nr_entries; 58 uint64_t update_interval; 59 } BDRVBlkLogWritesState; 60 61 static QemuOptsList runtime_opts = { 62 .name = "blklogwrites", 63 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), 64 .desc = { 65 { 66 .name = "log-append", 67 .type = QEMU_OPT_BOOL, 68 .help = "Append to an existing log", 69 }, 70 { 71 .name = "log-sector-size", 72 .type = QEMU_OPT_SIZE, 73 .help = "Log sector size", 74 }, 75 { 76 .name = "log-super-update-interval", 77 .type = QEMU_OPT_NUMBER, 78 .help = "Log superblock update interval (# of write requests)", 79 }, 80 { /* end of list */ } 81 }, 82 }; 83 84 static inline uint32_t blk_log_writes_log2(uint32_t value) 85 { 86 assert(value > 0); 87 return 31 - clz32(value); 88 } 89 90 static inline bool blk_log_writes_sector_size_valid(uint32_t sector_size) 91 { 92 return is_power_of_2(sector_size) && 93 sector_size >= sizeof(struct log_write_super) && 94 sector_size >= sizeof(struct log_write_entry) && 95 sector_size < (1ull << 24); 96 } 97 98 static uint64_t blk_log_writes_find_cur_log_sector(BdrvChild *log, 99 uint32_t sector_size, 100 uint64_t nr_entries, 101 Error **errp) 102 { 103 uint64_t cur_sector = 1; 104 uint64_t cur_idx = 0; 105 uint32_t sector_bits = blk_log_writes_log2(sector_size); 106 struct log_write_entry cur_entry; 107 108 while (cur_idx < nr_entries) { 109 int read_ret = bdrv_pread(log, cur_sector << sector_bits, &cur_entry, 110 sizeof(cur_entry)); 111 if (read_ret < 0) { 112 error_setg_errno(errp, -read_ret, 113 "Failed to read log entry %"PRIu64, cur_idx); 114 return (uint64_t)-1ull; 115 } 116 117 if (cur_entry.flags & ~cpu_to_le64(LOG_FLAG_MASK)) { 118 error_setg(errp, "Invalid flags 0x%"PRIx64" in log entry %"PRIu64, 119 le64_to_cpu(cur_entry.flags), cur_idx); 120 return (uint64_t)-1ull; 121 } 122 123 /* Account for the sector of the entry itself */ 124 ++cur_sector; 125 126 /* 127 * Account for the data of the write. 128 * For discards, this data is not present. 129 */ 130 if (!(cur_entry.flags & cpu_to_le64(LOG_DISCARD_FLAG))) { 131 cur_sector += le64_to_cpu(cur_entry.nr_sectors); 132 } 133 134 ++cur_idx; 135 } 136 137 return cur_sector; 138 } 139 140 static int blk_log_writes_open(BlockDriverState *bs, QDict *options, int flags, 141 Error **errp) 142 { 143 BDRVBlkLogWritesState *s = bs->opaque; 144 QemuOpts *opts; 145 Error *local_err = NULL; 146 int ret; 147 uint64_t log_sector_size; 148 bool log_append; 149 150 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); 151 qemu_opts_absorb_qdict(opts, options, &local_err); 152 if (local_err) { 153 ret = -EINVAL; 154 error_propagate(errp, local_err); 155 goto fail; 156 } 157 158 /* Open the file */ 159 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, false, 160 &local_err); 161 if (local_err) { 162 ret = -EINVAL; 163 error_propagate(errp, local_err); 164 goto fail; 165 } 166 167 /* Open the log file */ 168 s->log_file = bdrv_open_child(NULL, options, "log", bs, &child_file, false, 169 &local_err); 170 if (local_err) { 171 ret = -EINVAL; 172 error_propagate(errp, local_err); 173 goto fail; 174 } 175 176 log_append = qemu_opt_get_bool(opts, "log-append", false); 177 178 if (log_append) { 179 struct log_write_super log_sb = { 0, 0, 0, 0 }; 180 181 if (qemu_opt_find(opts, "log-sector-size")) { 182 ret = -EINVAL; 183 error_setg(errp, "log-append and log-sector-size are mutually " 184 "exclusive"); 185 goto fail_log; 186 } 187 188 /* Read log superblock or fake one for an empty log */ 189 if (!bdrv_getlength(s->log_file->bs)) { 190 log_sb.magic = cpu_to_le64(WRITE_LOG_MAGIC); 191 log_sb.version = cpu_to_le64(WRITE_LOG_VERSION); 192 log_sb.nr_entries = cpu_to_le64(0); 193 log_sb.sectorsize = cpu_to_le32(BDRV_SECTOR_SIZE); 194 } else { 195 ret = bdrv_pread(s->log_file, 0, &log_sb, sizeof(log_sb)); 196 if (ret < 0) { 197 error_setg_errno(errp, -ret, "Could not read log superblock"); 198 goto fail_log; 199 } 200 } 201 202 if (log_sb.magic != cpu_to_le64(WRITE_LOG_MAGIC)) { 203 ret = -EINVAL; 204 error_setg(errp, "Invalid log superblock magic"); 205 goto fail_log; 206 } 207 208 if (log_sb.version != cpu_to_le64(WRITE_LOG_VERSION)) { 209 ret = -EINVAL; 210 error_setg(errp, "Unsupported log version %"PRIu64, 211 le64_to_cpu(log_sb.version)); 212 goto fail_log; 213 } 214 215 log_sector_size = le32_to_cpu(log_sb.sectorsize); 216 s->cur_log_sector = 1; 217 s->nr_entries = 0; 218 219 if (blk_log_writes_sector_size_valid(log_sector_size)) { 220 s->cur_log_sector = 221 blk_log_writes_find_cur_log_sector(s->log_file, log_sector_size, 222 le64_to_cpu(log_sb.nr_entries), &local_err); 223 if (local_err) { 224 ret = -EINVAL; 225 error_propagate(errp, local_err); 226 goto fail_log; 227 } 228 229 s->nr_entries = le64_to_cpu(log_sb.nr_entries); 230 } 231 } else { 232 log_sector_size = qemu_opt_get_size(opts, "log-sector-size", 233 BDRV_SECTOR_SIZE); 234 s->cur_log_sector = 1; 235 s->nr_entries = 0; 236 } 237 238 if (!blk_log_writes_sector_size_valid(log_sector_size)) { 239 ret = -EINVAL; 240 error_setg(errp, "Invalid log sector size %"PRIu64, log_sector_size); 241 goto fail_log; 242 } 243 244 s->sectorsize = log_sector_size; 245 s->sectorbits = blk_log_writes_log2(log_sector_size); 246 s->update_interval = qemu_opt_get_number(opts, "log-super-update-interval", 247 4096); 248 if (!s->update_interval) { 249 ret = -EINVAL; 250 error_setg(errp, "Invalid log superblock update interval %"PRIu64, 251 s->update_interval); 252 goto fail_log; 253 } 254 255 ret = 0; 256 fail_log: 257 if (ret < 0) { 258 bdrv_unref_child(bs, s->log_file); 259 s->log_file = NULL; 260 } 261 fail: 262 if (ret < 0) { 263 bdrv_unref_child(bs, bs->file); 264 bs->file = NULL; 265 } 266 qemu_opts_del(opts); 267 return ret; 268 } 269 270 static void blk_log_writes_close(BlockDriverState *bs) 271 { 272 BDRVBlkLogWritesState *s = bs->opaque; 273 274 bdrv_unref_child(bs, s->log_file); 275 s->log_file = NULL; 276 } 277 278 static int64_t blk_log_writes_getlength(BlockDriverState *bs) 279 { 280 return bdrv_getlength(bs->file->bs); 281 } 282 283 static void blk_log_writes_refresh_filename(BlockDriverState *bs, 284 QDict *options) 285 { 286 BDRVBlkLogWritesState *s = bs->opaque; 287 288 /* bs->file->bs has already been refreshed */ 289 bdrv_refresh_filename(s->log_file->bs); 290 291 if (bs->file->bs->full_open_options 292 && s->log_file->bs->full_open_options) 293 { 294 QDict *opts = qdict_new(); 295 qdict_put_str(opts, "driver", "blklogwrites"); 296 297 qobject_ref(bs->file->bs->full_open_options); 298 qdict_put(opts, "file", bs->file->bs->full_open_options); 299 qobject_ref(s->log_file->bs->full_open_options); 300 qdict_put(opts, "log", s->log_file->bs->full_open_options); 301 qdict_put_int(opts, "log-sector-size", s->sectorsize); 302 303 bs->full_open_options = opts; 304 } 305 } 306 307 static void blk_log_writes_child_perm(BlockDriverState *bs, BdrvChild *c, 308 const BdrvChildRole *role, 309 BlockReopenQueue *ro_q, 310 uint64_t perm, uint64_t shrd, 311 uint64_t *nperm, uint64_t *nshrd) 312 { 313 if (!c) { 314 *nperm = perm & DEFAULT_PERM_PASSTHROUGH; 315 *nshrd = (shrd & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED; 316 return; 317 } 318 319 if (!strcmp(c->name, "log")) { 320 bdrv_format_default_perms(bs, c, role, ro_q, perm, shrd, nperm, nshrd); 321 } else { 322 bdrv_filter_default_perms(bs, c, role, ro_q, perm, shrd, nperm, nshrd); 323 } 324 } 325 326 static void blk_log_writes_refresh_limits(BlockDriverState *bs, Error **errp) 327 { 328 BDRVBlkLogWritesState *s = bs->opaque; 329 bs->bl.request_alignment = s->sectorsize; 330 } 331 332 static int coroutine_fn 333 blk_log_writes_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 334 QEMUIOVector *qiov, int flags) 335 { 336 return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags); 337 } 338 339 typedef struct BlkLogWritesFileReq { 340 BlockDriverState *bs; 341 uint64_t offset; 342 uint64_t bytes; 343 int file_flags; 344 QEMUIOVector *qiov; 345 int (*func)(struct BlkLogWritesFileReq *r); 346 int file_ret; 347 } BlkLogWritesFileReq; 348 349 typedef struct { 350 BlockDriverState *bs; 351 QEMUIOVector *qiov; 352 struct log_write_entry entry; 353 uint64_t zero_size; 354 int log_ret; 355 } BlkLogWritesLogReq; 356 357 static void coroutine_fn blk_log_writes_co_do_log(BlkLogWritesLogReq *lr) 358 { 359 BDRVBlkLogWritesState *s = lr->bs->opaque; 360 uint64_t cur_log_offset = s->cur_log_sector << s->sectorbits; 361 362 s->nr_entries++; 363 s->cur_log_sector += 364 ROUND_UP(lr->qiov->size, s->sectorsize) >> s->sectorbits; 365 366 lr->log_ret = bdrv_co_pwritev(s->log_file, cur_log_offset, lr->qiov->size, 367 lr->qiov, 0); 368 369 /* Logging for the "write zeroes" operation */ 370 if (lr->log_ret == 0 && lr->zero_size) { 371 cur_log_offset = s->cur_log_sector << s->sectorbits; 372 s->cur_log_sector += 373 ROUND_UP(lr->zero_size, s->sectorsize) >> s->sectorbits; 374 375 lr->log_ret = bdrv_co_pwrite_zeroes(s->log_file, cur_log_offset, 376 lr->zero_size, 0); 377 } 378 379 /* Update super block on flush or every update interval */ 380 if (lr->log_ret == 0 && ((lr->entry.flags & LOG_FLUSH_FLAG) 381 || (s->nr_entries % s->update_interval == 0))) 382 { 383 struct log_write_super super = { 384 .magic = cpu_to_le64(WRITE_LOG_MAGIC), 385 .version = cpu_to_le64(WRITE_LOG_VERSION), 386 .nr_entries = cpu_to_le64(s->nr_entries), 387 .sectorsize = cpu_to_le32(s->sectorsize), 388 }; 389 void *zeroes = g_malloc0(s->sectorsize - sizeof(super)); 390 QEMUIOVector qiov; 391 392 qemu_iovec_init(&qiov, 2); 393 qemu_iovec_add(&qiov, &super, sizeof(super)); 394 qemu_iovec_add(&qiov, zeroes, s->sectorsize - sizeof(super)); 395 396 lr->log_ret = 397 bdrv_co_pwritev(s->log_file, 0, s->sectorsize, &qiov, 0); 398 if (lr->log_ret == 0) { 399 lr->log_ret = bdrv_co_flush(s->log_file->bs); 400 } 401 qemu_iovec_destroy(&qiov); 402 g_free(zeroes); 403 } 404 } 405 406 static void coroutine_fn blk_log_writes_co_do_file(BlkLogWritesFileReq *fr) 407 { 408 fr->file_ret = fr->func(fr); 409 } 410 411 static int coroutine_fn 412 blk_log_writes_co_log(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 413 QEMUIOVector *qiov, int flags, 414 int (*file_func)(BlkLogWritesFileReq *r), 415 uint64_t entry_flags, bool is_zero_write) 416 { 417 QEMUIOVector log_qiov; 418 size_t niov = qiov ? qiov->niov : 0; 419 BDRVBlkLogWritesState *s = bs->opaque; 420 BlkLogWritesFileReq fr = { 421 .bs = bs, 422 .offset = offset, 423 .bytes = bytes, 424 .file_flags = flags, 425 .qiov = qiov, 426 .func = file_func, 427 }; 428 BlkLogWritesLogReq lr = { 429 .bs = bs, 430 .qiov = &log_qiov, 431 .entry = { 432 .sector = cpu_to_le64(offset >> s->sectorbits), 433 .nr_sectors = cpu_to_le64(bytes >> s->sectorbits), 434 .flags = cpu_to_le64(entry_flags), 435 .data_len = 0, 436 }, 437 .zero_size = is_zero_write ? bytes : 0, 438 }; 439 void *zeroes = g_malloc0(s->sectorsize - sizeof(lr.entry)); 440 441 assert((1 << s->sectorbits) == s->sectorsize); 442 assert(bs->bl.request_alignment == s->sectorsize); 443 assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment)); 444 assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment)); 445 446 qemu_iovec_init(&log_qiov, niov + 2); 447 qemu_iovec_add(&log_qiov, &lr.entry, sizeof(lr.entry)); 448 qemu_iovec_add(&log_qiov, zeroes, s->sectorsize - sizeof(lr.entry)); 449 if (qiov) { 450 qemu_iovec_concat(&log_qiov, qiov, 0, qiov->size); 451 } 452 453 blk_log_writes_co_do_file(&fr); 454 blk_log_writes_co_do_log(&lr); 455 456 qemu_iovec_destroy(&log_qiov); 457 g_free(zeroes); 458 459 if (lr.log_ret < 0) { 460 return lr.log_ret; 461 } 462 463 return fr.file_ret; 464 } 465 466 static int coroutine_fn 467 blk_log_writes_co_do_file_pwritev(BlkLogWritesFileReq *fr) 468 { 469 return bdrv_co_pwritev(fr->bs->file, fr->offset, fr->bytes, 470 fr->qiov, fr->file_flags); 471 } 472 473 static int coroutine_fn 474 blk_log_writes_co_do_file_pwrite_zeroes(BlkLogWritesFileReq *fr) 475 { 476 return bdrv_co_pwrite_zeroes(fr->bs->file, fr->offset, fr->bytes, 477 fr->file_flags); 478 } 479 480 static int coroutine_fn blk_log_writes_co_do_file_flush(BlkLogWritesFileReq *fr) 481 { 482 return bdrv_co_flush(fr->bs->file->bs); 483 } 484 485 static int coroutine_fn 486 blk_log_writes_co_do_file_pdiscard(BlkLogWritesFileReq *fr) 487 { 488 return bdrv_co_pdiscard(fr->bs->file, fr->offset, fr->bytes); 489 } 490 491 static int coroutine_fn 492 blk_log_writes_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 493 QEMUIOVector *qiov, int flags) 494 { 495 return blk_log_writes_co_log(bs, offset, bytes, qiov, flags, 496 blk_log_writes_co_do_file_pwritev, 0, false); 497 } 498 499 static int coroutine_fn 500 blk_log_writes_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, 501 BdrvRequestFlags flags) 502 { 503 return blk_log_writes_co_log(bs, offset, bytes, NULL, flags, 504 blk_log_writes_co_do_file_pwrite_zeroes, 0, 505 true); 506 } 507 508 static int coroutine_fn blk_log_writes_co_flush_to_disk(BlockDriverState *bs) 509 { 510 return blk_log_writes_co_log(bs, 0, 0, NULL, 0, 511 blk_log_writes_co_do_file_flush, 512 LOG_FLUSH_FLAG, false); 513 } 514 515 static int coroutine_fn 516 blk_log_writes_co_pdiscard(BlockDriverState *bs, int64_t offset, int count) 517 { 518 return blk_log_writes_co_log(bs, offset, count, NULL, 0, 519 blk_log_writes_co_do_file_pdiscard, 520 LOG_DISCARD_FLAG, false); 521 } 522 523 static BlockDriver bdrv_blk_log_writes = { 524 .format_name = "blklogwrites", 525 .instance_size = sizeof(BDRVBlkLogWritesState), 526 527 .bdrv_open = blk_log_writes_open, 528 .bdrv_close = blk_log_writes_close, 529 .bdrv_getlength = blk_log_writes_getlength, 530 .bdrv_refresh_filename = blk_log_writes_refresh_filename, 531 .bdrv_child_perm = blk_log_writes_child_perm, 532 .bdrv_refresh_limits = blk_log_writes_refresh_limits, 533 534 .bdrv_co_preadv = blk_log_writes_co_preadv, 535 .bdrv_co_pwritev = blk_log_writes_co_pwritev, 536 .bdrv_co_pwrite_zeroes = blk_log_writes_co_pwrite_zeroes, 537 .bdrv_co_flush_to_disk = blk_log_writes_co_flush_to_disk, 538 .bdrv_co_pdiscard = blk_log_writes_co_pdiscard, 539 .bdrv_co_block_status = bdrv_co_block_status_from_file, 540 541 .is_filter = true, 542 }; 543 544 static void bdrv_blk_log_writes_init(void) 545 { 546 bdrv_register(&bdrv_blk_log_writes); 547 } 548 549 block_init(bdrv_blk_log_writes_init); 550