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