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