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 &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_of_bds, 169 BDRV_CHILD_METADATA, false, &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_child_perm(BlockDriverState *bs, BdrvChild *c, 284 BdrvChildRole role, 285 BlockReopenQueue *ro_q, 286 uint64_t perm, uint64_t shrd, 287 uint64_t *nperm, uint64_t *nshrd) 288 { 289 if (!c) { 290 *nperm = perm & DEFAULT_PERM_PASSTHROUGH; 291 *nshrd = (shrd & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED; 292 return; 293 } 294 295 bdrv_default_perms(bs, c, role, ro_q, perm, shrd, 296 nperm, nshrd); 297 } 298 299 static void blk_log_writes_refresh_limits(BlockDriverState *bs, Error **errp) 300 { 301 BDRVBlkLogWritesState *s = bs->opaque; 302 bs->bl.request_alignment = s->sectorsize; 303 } 304 305 static int coroutine_fn 306 blk_log_writes_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 307 QEMUIOVector *qiov, int flags) 308 { 309 return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags); 310 } 311 312 typedef struct BlkLogWritesFileReq { 313 BlockDriverState *bs; 314 uint64_t offset; 315 uint64_t bytes; 316 int file_flags; 317 QEMUIOVector *qiov; 318 int (*func)(struct BlkLogWritesFileReq *r); 319 int file_ret; 320 } BlkLogWritesFileReq; 321 322 typedef struct { 323 BlockDriverState *bs; 324 QEMUIOVector *qiov; 325 struct log_write_entry entry; 326 uint64_t zero_size; 327 int log_ret; 328 } BlkLogWritesLogReq; 329 330 static void coroutine_fn blk_log_writes_co_do_log(BlkLogWritesLogReq *lr) 331 { 332 BDRVBlkLogWritesState *s = lr->bs->opaque; 333 uint64_t cur_log_offset = s->cur_log_sector << s->sectorbits; 334 335 s->nr_entries++; 336 s->cur_log_sector += 337 ROUND_UP(lr->qiov->size, s->sectorsize) >> s->sectorbits; 338 339 lr->log_ret = bdrv_co_pwritev(s->log_file, cur_log_offset, lr->qiov->size, 340 lr->qiov, 0); 341 342 /* Logging for the "write zeroes" operation */ 343 if (lr->log_ret == 0 && lr->zero_size) { 344 cur_log_offset = s->cur_log_sector << s->sectorbits; 345 s->cur_log_sector += 346 ROUND_UP(lr->zero_size, s->sectorsize) >> s->sectorbits; 347 348 lr->log_ret = bdrv_co_pwrite_zeroes(s->log_file, cur_log_offset, 349 lr->zero_size, 0); 350 } 351 352 /* Update super block on flush or every update interval */ 353 if (lr->log_ret == 0 && ((lr->entry.flags & LOG_FLUSH_FLAG) 354 || (s->nr_entries % s->update_interval == 0))) 355 { 356 struct log_write_super super = { 357 .magic = cpu_to_le64(WRITE_LOG_MAGIC), 358 .version = cpu_to_le64(WRITE_LOG_VERSION), 359 .nr_entries = cpu_to_le64(s->nr_entries), 360 .sectorsize = cpu_to_le32(s->sectorsize), 361 }; 362 void *zeroes = g_malloc0(s->sectorsize - sizeof(super)); 363 QEMUIOVector qiov; 364 365 qemu_iovec_init(&qiov, 2); 366 qemu_iovec_add(&qiov, &super, sizeof(super)); 367 qemu_iovec_add(&qiov, zeroes, s->sectorsize - sizeof(super)); 368 369 lr->log_ret = 370 bdrv_co_pwritev(s->log_file, 0, s->sectorsize, &qiov, 0); 371 if (lr->log_ret == 0) { 372 lr->log_ret = bdrv_co_flush(s->log_file->bs); 373 } 374 qemu_iovec_destroy(&qiov); 375 g_free(zeroes); 376 } 377 } 378 379 static void coroutine_fn blk_log_writes_co_do_file(BlkLogWritesFileReq *fr) 380 { 381 fr->file_ret = fr->func(fr); 382 } 383 384 static int coroutine_fn 385 blk_log_writes_co_log(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 386 QEMUIOVector *qiov, int flags, 387 int (*file_func)(BlkLogWritesFileReq *r), 388 uint64_t entry_flags, bool is_zero_write) 389 { 390 QEMUIOVector log_qiov; 391 size_t niov = qiov ? qiov->niov : 0; 392 BDRVBlkLogWritesState *s = bs->opaque; 393 BlkLogWritesFileReq fr = { 394 .bs = bs, 395 .offset = offset, 396 .bytes = bytes, 397 .file_flags = flags, 398 .qiov = qiov, 399 .func = file_func, 400 }; 401 BlkLogWritesLogReq lr = { 402 .bs = bs, 403 .qiov = &log_qiov, 404 .entry = { 405 .sector = cpu_to_le64(offset >> s->sectorbits), 406 .nr_sectors = cpu_to_le64(bytes >> s->sectorbits), 407 .flags = cpu_to_le64(entry_flags), 408 .data_len = 0, 409 }, 410 .zero_size = is_zero_write ? bytes : 0, 411 }; 412 void *zeroes = g_malloc0(s->sectorsize - sizeof(lr.entry)); 413 414 assert((1 << s->sectorbits) == s->sectorsize); 415 assert(bs->bl.request_alignment == s->sectorsize); 416 assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment)); 417 assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment)); 418 419 qemu_iovec_init(&log_qiov, niov + 2); 420 qemu_iovec_add(&log_qiov, &lr.entry, sizeof(lr.entry)); 421 qemu_iovec_add(&log_qiov, zeroes, s->sectorsize - sizeof(lr.entry)); 422 if (qiov) { 423 qemu_iovec_concat(&log_qiov, qiov, 0, qiov->size); 424 } 425 426 blk_log_writes_co_do_file(&fr); 427 blk_log_writes_co_do_log(&lr); 428 429 qemu_iovec_destroy(&log_qiov); 430 g_free(zeroes); 431 432 if (lr.log_ret < 0) { 433 return lr.log_ret; 434 } 435 436 return fr.file_ret; 437 } 438 439 static int coroutine_fn 440 blk_log_writes_co_do_file_pwritev(BlkLogWritesFileReq *fr) 441 { 442 return bdrv_co_pwritev(fr->bs->file, fr->offset, fr->bytes, 443 fr->qiov, fr->file_flags); 444 } 445 446 static int coroutine_fn 447 blk_log_writes_co_do_file_pwrite_zeroes(BlkLogWritesFileReq *fr) 448 { 449 return bdrv_co_pwrite_zeroes(fr->bs->file, fr->offset, fr->bytes, 450 fr->file_flags); 451 } 452 453 static int coroutine_fn blk_log_writes_co_do_file_flush(BlkLogWritesFileReq *fr) 454 { 455 return bdrv_co_flush(fr->bs->file->bs); 456 } 457 458 static int coroutine_fn 459 blk_log_writes_co_do_file_pdiscard(BlkLogWritesFileReq *fr) 460 { 461 return bdrv_co_pdiscard(fr->bs->file, fr->offset, fr->bytes); 462 } 463 464 static int coroutine_fn 465 blk_log_writes_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 466 QEMUIOVector *qiov, int flags) 467 { 468 return blk_log_writes_co_log(bs, offset, bytes, qiov, flags, 469 blk_log_writes_co_do_file_pwritev, 0, false); 470 } 471 472 static int coroutine_fn 473 blk_log_writes_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, 474 BdrvRequestFlags flags) 475 { 476 return blk_log_writes_co_log(bs, offset, bytes, NULL, flags, 477 blk_log_writes_co_do_file_pwrite_zeroes, 0, 478 true); 479 } 480 481 static int coroutine_fn blk_log_writes_co_flush_to_disk(BlockDriverState *bs) 482 { 483 return blk_log_writes_co_log(bs, 0, 0, NULL, 0, 484 blk_log_writes_co_do_file_flush, 485 LOG_FLUSH_FLAG, false); 486 } 487 488 static int coroutine_fn 489 blk_log_writes_co_pdiscard(BlockDriverState *bs, int64_t offset, int count) 490 { 491 return blk_log_writes_co_log(bs, offset, count, NULL, 0, 492 blk_log_writes_co_do_file_pdiscard, 493 LOG_DISCARD_FLAG, false); 494 } 495 496 static const char *const blk_log_writes_strong_runtime_opts[] = { 497 "log-append", 498 "log-sector-size", 499 500 NULL 501 }; 502 503 static BlockDriver bdrv_blk_log_writes = { 504 .format_name = "blklogwrites", 505 .instance_size = sizeof(BDRVBlkLogWritesState), 506 507 .bdrv_open = blk_log_writes_open, 508 .bdrv_close = blk_log_writes_close, 509 .bdrv_getlength = blk_log_writes_getlength, 510 .bdrv_child_perm = blk_log_writes_child_perm, 511 .bdrv_refresh_limits = blk_log_writes_refresh_limits, 512 513 .bdrv_co_preadv = blk_log_writes_co_preadv, 514 .bdrv_co_pwritev = blk_log_writes_co_pwritev, 515 .bdrv_co_pwrite_zeroes = blk_log_writes_co_pwrite_zeroes, 516 .bdrv_co_flush_to_disk = blk_log_writes_co_flush_to_disk, 517 .bdrv_co_pdiscard = blk_log_writes_co_pdiscard, 518 519 .is_filter = true, 520 .strong_runtime_opts = blk_log_writes_strong_runtime_opts, 521 }; 522 523 static void bdrv_blk_log_writes_init(void) 524 { 525 bdrv_register(&bdrv_blk_log_writes); 526 } 527 528 block_init(bdrv_blk_log_writes_init); 529