1 /* 2 * Block driver for RAW files (posix) 3 * 4 * Copyright (c) 2006 Fabrice Bellard 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 25 #include "qemu/osdep.h" 26 #include "qapi/error.h" 27 #include "qemu/cutils.h" 28 #include "qemu/error-report.h" 29 #include "block/block-io.h" 30 #include "block/block_int.h" 31 #include "qemu/module.h" 32 #include "qemu/option.h" 33 #include "qemu/units.h" 34 #include "qemu/memalign.h" 35 #include "trace.h" 36 #include "block/thread-pool.h" 37 #include "qemu/iov.h" 38 #include "block/raw-aio.h" 39 #include "qapi/qmp/qdict.h" 40 #include "qapi/qmp/qstring.h" 41 42 #include "scsi/pr-manager.h" 43 #include "scsi/constants.h" 44 45 #if defined(__APPLE__) && (__MACH__) 46 #include <sys/ioctl.h> 47 #if defined(HAVE_HOST_BLOCK_DEVICE) 48 #include <paths.h> 49 #include <sys/param.h> 50 #include <sys/mount.h> 51 #include <IOKit/IOKitLib.h> 52 #include <IOKit/IOBSD.h> 53 #include <IOKit/storage/IOMediaBSDClient.h> 54 #include <IOKit/storage/IOMedia.h> 55 #include <IOKit/storage/IOCDMedia.h> 56 //#include <IOKit/storage/IOCDTypes.h> 57 #include <IOKit/storage/IODVDMedia.h> 58 #include <CoreFoundation/CoreFoundation.h> 59 #endif /* defined(HAVE_HOST_BLOCK_DEVICE) */ 60 #endif 61 62 #ifdef __sun__ 63 #define _POSIX_PTHREAD_SEMANTICS 1 64 #include <sys/dkio.h> 65 #endif 66 #ifdef __linux__ 67 #include <sys/ioctl.h> 68 #include <sys/param.h> 69 #include <sys/syscall.h> 70 #include <sys/vfs.h> 71 #if defined(CONFIG_BLKZONED) 72 #include <linux/blkzoned.h> 73 #endif 74 #include <linux/cdrom.h> 75 #include <linux/fd.h> 76 #include <linux/fs.h> 77 #include <linux/hdreg.h> 78 #include <linux/magic.h> 79 #include <scsi/sg.h> 80 #ifdef __s390__ 81 #include <asm/dasd.h> 82 #endif 83 #ifndef FS_NOCOW_FL 84 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */ 85 #endif 86 #endif 87 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE) 88 #include <linux/falloc.h> 89 #endif 90 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 91 #include <sys/disk.h> 92 #include <sys/cdio.h> 93 #endif 94 95 #ifdef __OpenBSD__ 96 #include <sys/ioctl.h> 97 #include <sys/disklabel.h> 98 #include <sys/dkio.h> 99 #endif 100 101 #ifdef __NetBSD__ 102 #include <sys/ioctl.h> 103 #include <sys/disklabel.h> 104 #include <sys/dkio.h> 105 #include <sys/disk.h> 106 #endif 107 108 #ifdef __DragonFly__ 109 #include <sys/ioctl.h> 110 #include <sys/diskslice.h> 111 #endif 112 113 /* OS X does not have O_DSYNC */ 114 #ifndef O_DSYNC 115 #ifdef O_SYNC 116 #define O_DSYNC O_SYNC 117 #elif defined(O_FSYNC) 118 #define O_DSYNC O_FSYNC 119 #endif 120 #endif 121 122 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ 123 #ifndef O_DIRECT 124 #define O_DIRECT O_DSYNC 125 #endif 126 127 #define FTYPE_FILE 0 128 #define FTYPE_CD 1 129 130 #define MAX_BLOCKSIZE 4096 131 132 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes, 133 * leaving a few more bytes for its future use. */ 134 #define RAW_LOCK_PERM_BASE 100 135 #define RAW_LOCK_SHARED_BASE 200 136 137 typedef struct BDRVRawState { 138 int fd; 139 bool use_lock; 140 int type; 141 int open_flags; 142 size_t buf_align; 143 144 /* The current permissions. */ 145 uint64_t perm; 146 uint64_t shared_perm; 147 148 /* The perms bits whose corresponding bytes are already locked in 149 * s->fd. */ 150 uint64_t locked_perm; 151 uint64_t locked_shared_perm; 152 153 uint64_t aio_max_batch; 154 155 int perm_change_fd; 156 int perm_change_flags; 157 BDRVReopenState *reopen_state; 158 159 bool has_discard:1; 160 bool has_write_zeroes:1; 161 bool use_linux_aio:1; 162 bool use_linux_io_uring:1; 163 int64_t *offset; /* offset of zone append operation */ 164 int page_cache_inconsistent; /* errno from fdatasync failure */ 165 bool has_fallocate; 166 bool needs_alignment; 167 bool force_alignment; 168 bool drop_cache; 169 bool check_cache_dropped; 170 struct { 171 uint64_t discard_nb_ok; 172 uint64_t discard_nb_failed; 173 uint64_t discard_bytes_ok; 174 } stats; 175 176 PRManager *pr_mgr; 177 } BDRVRawState; 178 179 typedef struct BDRVRawReopenState { 180 int open_flags; 181 bool drop_cache; 182 bool check_cache_dropped; 183 } BDRVRawReopenState; 184 185 static int fd_open(BlockDriverState *bs) 186 { 187 BDRVRawState *s = bs->opaque; 188 189 /* this is just to ensure s->fd is sane (its called by io ops) */ 190 if (s->fd >= 0) { 191 return 0; 192 } 193 return -EIO; 194 } 195 196 static int64_t raw_getlength(BlockDriverState *bs); 197 198 typedef struct RawPosixAIOData { 199 BlockDriverState *bs; 200 int aio_type; 201 int aio_fildes; 202 203 off_t aio_offset; 204 uint64_t aio_nbytes; 205 206 union { 207 struct { 208 struct iovec *iov; 209 int niov; 210 } io; 211 struct { 212 uint64_t cmd; 213 void *buf; 214 } ioctl; 215 struct { 216 int aio_fd2; 217 off_t aio_offset2; 218 } copy_range; 219 struct { 220 PreallocMode prealloc; 221 Error **errp; 222 } truncate; 223 struct { 224 unsigned int *nr_zones; 225 BlockZoneDescriptor *zones; 226 } zone_report; 227 struct { 228 unsigned long op; 229 } zone_mgmt; 230 }; 231 } RawPosixAIOData; 232 233 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 234 static int cdrom_reopen(BlockDriverState *bs); 235 #endif 236 237 /* 238 * Elide EAGAIN and EACCES details when failing to lock, as this 239 * indicates that the specified file region is already locked by 240 * another process, which is considered a common scenario. 241 */ 242 #define raw_lock_error_setg_errno(errp, err, fmt, ...) \ 243 do { \ 244 if ((err) == EAGAIN || (err) == EACCES) { \ 245 error_setg((errp), (fmt), ## __VA_ARGS__); \ 246 } else { \ 247 error_setg_errno((errp), (err), (fmt), ## __VA_ARGS__); \ 248 } \ 249 } while (0) 250 251 #if defined(__NetBSD__) 252 static int raw_normalize_devicepath(const char **filename, Error **errp) 253 { 254 static char namebuf[PATH_MAX]; 255 const char *dp, *fname; 256 struct stat sb; 257 258 fname = *filename; 259 dp = strrchr(fname, '/'); 260 if (lstat(fname, &sb) < 0) { 261 error_setg_file_open(errp, errno, fname); 262 return -errno; 263 } 264 265 if (!S_ISBLK(sb.st_mode)) { 266 return 0; 267 } 268 269 if (dp == NULL) { 270 snprintf(namebuf, PATH_MAX, "r%s", fname); 271 } else { 272 snprintf(namebuf, PATH_MAX, "%.*s/r%s", 273 (int)(dp - fname), fname, dp + 1); 274 } 275 *filename = namebuf; 276 warn_report("%s is a block device, using %s", fname, *filename); 277 278 return 0; 279 } 280 #else 281 static int raw_normalize_devicepath(const char **filename, Error **errp) 282 { 283 return 0; 284 } 285 #endif 286 287 /* 288 * Get logical block size via ioctl. On success store it in @sector_size_p. 289 */ 290 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) 291 { 292 unsigned int sector_size; 293 bool success = false; 294 int i; 295 296 errno = ENOTSUP; 297 static const unsigned long ioctl_list[] = { 298 #ifdef BLKSSZGET 299 BLKSSZGET, 300 #endif 301 #ifdef DKIOCGETBLOCKSIZE 302 DKIOCGETBLOCKSIZE, 303 #endif 304 #ifdef DIOCGSECTORSIZE 305 DIOCGSECTORSIZE, 306 #endif 307 }; 308 309 /* Try a few ioctls to get the right size */ 310 for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { 311 if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { 312 *sector_size_p = sector_size; 313 success = true; 314 } 315 } 316 317 return success ? 0 : -errno; 318 } 319 320 /** 321 * Get physical block size of @fd. 322 * On success, store it in @blk_size and return 0. 323 * On failure, return -errno. 324 */ 325 static int probe_physical_blocksize(int fd, unsigned int *blk_size) 326 { 327 #ifdef BLKPBSZGET 328 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { 329 return -errno; 330 } 331 return 0; 332 #else 333 return -ENOTSUP; 334 #endif 335 } 336 337 /* 338 * Returns true if no alignment restrictions are necessary even for files 339 * opened with O_DIRECT. 340 * 341 * raw_probe_alignment() probes the required alignment and assume that 1 means 342 * the probing failed, so it falls back to a safe default of 4k. This can be 343 * avoided if we know that byte alignment is okay for the file. 344 */ 345 static bool dio_byte_aligned(int fd) 346 { 347 #ifdef __linux__ 348 struct statfs buf; 349 int ret; 350 351 ret = fstatfs(fd, &buf); 352 if (ret == 0 && buf.f_type == NFS_SUPER_MAGIC) { 353 return true; 354 } 355 #endif 356 return false; 357 } 358 359 static bool raw_needs_alignment(BlockDriverState *bs) 360 { 361 BDRVRawState *s = bs->opaque; 362 363 if ((bs->open_flags & BDRV_O_NOCACHE) != 0 && !dio_byte_aligned(s->fd)) { 364 return true; 365 } 366 367 return s->force_alignment; 368 } 369 370 /* Check if read is allowed with given memory buffer and length. 371 * 372 * This function is used to check O_DIRECT memory buffer and request alignment. 373 */ 374 static bool raw_is_io_aligned(int fd, void *buf, size_t len) 375 { 376 ssize_t ret = pread(fd, buf, len, 0); 377 378 if (ret >= 0) { 379 return true; 380 } 381 382 #ifdef __linux__ 383 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore 384 * other errors (e.g. real I/O error), which could happen on a failed 385 * drive, since we only care about probing alignment. 386 */ 387 if (errno != EINVAL) { 388 return true; 389 } 390 #endif 391 392 return false; 393 } 394 395 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) 396 { 397 BDRVRawState *s = bs->opaque; 398 char *buf; 399 size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size()); 400 size_t alignments[] = {1, 512, 1024, 2048, 4096}; 401 402 /* For SCSI generic devices the alignment is not really used. 403 With buffered I/O, we don't have any restrictions. */ 404 if (bdrv_is_sg(bs) || !s->needs_alignment) { 405 bs->bl.request_alignment = 1; 406 s->buf_align = 1; 407 return; 408 } 409 410 bs->bl.request_alignment = 0; 411 s->buf_align = 0; 412 /* Let's try to use the logical blocksize for the alignment. */ 413 if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { 414 bs->bl.request_alignment = 0; 415 } 416 417 #ifdef __linux__ 418 /* 419 * The XFS ioctl definitions are shipped in extra packages that might 420 * not always be available. Since we just need the XFS_IOC_DIOINFO ioctl 421 * here, we simply use our own definition instead: 422 */ 423 struct xfs_dioattr { 424 uint32_t d_mem; 425 uint32_t d_miniosz; 426 uint32_t d_maxiosz; 427 } da; 428 if (ioctl(fd, _IOR('X', 30, struct xfs_dioattr), &da) >= 0) { 429 bs->bl.request_alignment = da.d_miniosz; 430 /* The kernel returns wrong information for d_mem */ 431 /* s->buf_align = da.d_mem; */ 432 } 433 #endif 434 435 /* 436 * If we could not get the sizes so far, we can only guess them. First try 437 * to detect request alignment, since it is more likely to succeed. Then 438 * try to detect buf_align, which cannot be detected in some cases (e.g. 439 * Gluster). If buf_align cannot be detected, we fallback to the value of 440 * request_alignment. 441 */ 442 443 if (!bs->bl.request_alignment) { 444 int i; 445 size_t align; 446 buf = qemu_memalign(max_align, max_align); 447 for (i = 0; i < ARRAY_SIZE(alignments); i++) { 448 align = alignments[i]; 449 if (raw_is_io_aligned(fd, buf, align)) { 450 /* Fallback to safe value. */ 451 bs->bl.request_alignment = (align != 1) ? align : max_align; 452 break; 453 } 454 } 455 qemu_vfree(buf); 456 } 457 458 if (!s->buf_align) { 459 int i; 460 size_t align; 461 buf = qemu_memalign(max_align, 2 * max_align); 462 for (i = 0; i < ARRAY_SIZE(alignments); i++) { 463 align = alignments[i]; 464 if (raw_is_io_aligned(fd, buf + align, max_align)) { 465 /* Fallback to request_alignment. */ 466 s->buf_align = (align != 1) ? align : bs->bl.request_alignment; 467 break; 468 } 469 } 470 qemu_vfree(buf); 471 } 472 473 if (!s->buf_align || !bs->bl.request_alignment) { 474 error_setg(errp, "Could not find working O_DIRECT alignment"); 475 error_append_hint(errp, "Try cache.direct=off\n"); 476 } 477 } 478 479 static int check_hdev_writable(int fd) 480 { 481 #if defined(BLKROGET) 482 /* Linux block devices can be configured "read-only" using blockdev(8). 483 * This is independent of device node permissions and therefore open(2) 484 * with O_RDWR succeeds. Actual writes fail with EPERM. 485 * 486 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly 487 * check for read-only block devices so that Linux block devices behave 488 * properly. 489 */ 490 struct stat st; 491 int readonly = 0; 492 493 if (fstat(fd, &st)) { 494 return -errno; 495 } 496 497 if (!S_ISBLK(st.st_mode)) { 498 return 0; 499 } 500 501 if (ioctl(fd, BLKROGET, &readonly) < 0) { 502 return -errno; 503 } 504 505 if (readonly) { 506 return -EACCES; 507 } 508 #endif /* defined(BLKROGET) */ 509 return 0; 510 } 511 512 static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) 513 { 514 bool read_write = false; 515 assert(open_flags != NULL); 516 517 *open_flags |= O_BINARY; 518 *open_flags &= ~O_ACCMODE; 519 520 if (bdrv_flags & BDRV_O_AUTO_RDONLY) { 521 read_write = has_writers; 522 } else if (bdrv_flags & BDRV_O_RDWR) { 523 read_write = true; 524 } 525 526 if (read_write) { 527 *open_flags |= O_RDWR; 528 } else { 529 *open_flags |= O_RDONLY; 530 } 531 532 /* Use O_DSYNC for write-through caching, no flags for write-back caching, 533 * and O_DIRECT for no caching. */ 534 if ((bdrv_flags & BDRV_O_NOCACHE)) { 535 *open_flags |= O_DIRECT; 536 } 537 } 538 539 static void raw_parse_filename(const char *filename, QDict *options, 540 Error **errp) 541 { 542 bdrv_parse_filename_strip_prefix(filename, "file:", options); 543 } 544 545 static QemuOptsList raw_runtime_opts = { 546 .name = "raw", 547 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), 548 .desc = { 549 { 550 .name = "filename", 551 .type = QEMU_OPT_STRING, 552 .help = "File name of the image", 553 }, 554 { 555 .name = "aio", 556 .type = QEMU_OPT_STRING, 557 .help = "host AIO implementation (threads, native, io_uring)", 558 }, 559 { 560 .name = "aio-max-batch", 561 .type = QEMU_OPT_NUMBER, 562 .help = "AIO max batch size (0 = auto handled by AIO backend, default: 0)", 563 }, 564 { 565 .name = "locking", 566 .type = QEMU_OPT_STRING, 567 .help = "file locking mode (on/off/auto, default: auto)", 568 }, 569 { 570 .name = "pr-manager", 571 .type = QEMU_OPT_STRING, 572 .help = "id of persistent reservation manager object (default: none)", 573 }, 574 #if defined(__linux__) 575 { 576 .name = "drop-cache", 577 .type = QEMU_OPT_BOOL, 578 .help = "invalidate page cache during live migration (default: on)", 579 }, 580 #endif 581 { 582 .name = "x-check-cache-dropped", 583 .type = QEMU_OPT_BOOL, 584 .help = "check that page cache was dropped on live migration (default: off)" 585 }, 586 { /* end of list */ } 587 }, 588 }; 589 590 static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; 591 592 static int raw_open_common(BlockDriverState *bs, QDict *options, 593 int bdrv_flags, int open_flags, 594 bool device, Error **errp) 595 { 596 BDRVRawState *s = bs->opaque; 597 QemuOpts *opts; 598 Error *local_err = NULL; 599 const char *filename = NULL; 600 const char *str; 601 BlockdevAioOptions aio, aio_default; 602 int fd, ret; 603 struct stat st; 604 OnOffAuto locking; 605 606 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 607 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 608 ret = -EINVAL; 609 goto fail; 610 } 611 612 filename = qemu_opt_get(opts, "filename"); 613 614 ret = raw_normalize_devicepath(&filename, errp); 615 if (ret != 0) { 616 goto fail; 617 } 618 619 if (bdrv_flags & BDRV_O_NATIVE_AIO) { 620 aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE; 621 #ifdef CONFIG_LINUX_IO_URING 622 } else if (bdrv_flags & BDRV_O_IO_URING) { 623 aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING; 624 #endif 625 } else { 626 aio_default = BLOCKDEV_AIO_OPTIONS_THREADS; 627 } 628 629 aio = qapi_enum_parse(&BlockdevAioOptions_lookup, 630 qemu_opt_get(opts, "aio"), 631 aio_default, &local_err); 632 if (local_err) { 633 error_propagate(errp, local_err); 634 ret = -EINVAL; 635 goto fail; 636 } 637 638 s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); 639 #ifdef CONFIG_LINUX_IO_URING 640 s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING); 641 #endif 642 643 s->aio_max_batch = qemu_opt_get_number(opts, "aio-max-batch", 0); 644 645 locking = qapi_enum_parse(&OnOffAuto_lookup, 646 qemu_opt_get(opts, "locking"), 647 ON_OFF_AUTO_AUTO, &local_err); 648 if (local_err) { 649 error_propagate(errp, local_err); 650 ret = -EINVAL; 651 goto fail; 652 } 653 switch (locking) { 654 case ON_OFF_AUTO_ON: 655 s->use_lock = true; 656 if (!qemu_has_ofd_lock()) { 657 warn_report("File lock requested but OFD locking syscall is " 658 "unavailable, falling back to POSIX file locks"); 659 error_printf("Due to the implementation, locks can be lost " 660 "unexpectedly.\n"); 661 } 662 break; 663 case ON_OFF_AUTO_OFF: 664 s->use_lock = false; 665 break; 666 case ON_OFF_AUTO_AUTO: 667 s->use_lock = qemu_has_ofd_lock(); 668 break; 669 default: 670 abort(); 671 } 672 673 str = qemu_opt_get(opts, "pr-manager"); 674 if (str) { 675 s->pr_mgr = pr_manager_lookup(str, &local_err); 676 if (local_err) { 677 error_propagate(errp, local_err); 678 ret = -EINVAL; 679 goto fail; 680 } 681 } 682 683 s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); 684 s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", 685 false); 686 687 s->open_flags = open_flags; 688 raw_parse_flags(bdrv_flags, &s->open_flags, false); 689 690 s->fd = -1; 691 fd = qemu_open(filename, s->open_flags, errp); 692 ret = fd < 0 ? -errno : 0; 693 694 if (ret < 0) { 695 if (ret == -EROFS) { 696 ret = -EACCES; 697 } 698 goto fail; 699 } 700 s->fd = fd; 701 702 /* Check s->open_flags rather than bdrv_flags due to auto-read-only */ 703 if (s->open_flags & O_RDWR) { 704 ret = check_hdev_writable(s->fd); 705 if (ret < 0) { 706 error_setg_errno(errp, -ret, "The device is not writable"); 707 goto fail; 708 } 709 } 710 711 s->perm = 0; 712 s->shared_perm = BLK_PERM_ALL; 713 714 #ifdef CONFIG_LINUX_AIO 715 /* Currently Linux does AIO only for files opened with O_DIRECT */ 716 if (s->use_linux_aio) { 717 if (!(s->open_flags & O_DIRECT)) { 718 error_setg(errp, "aio=native was specified, but it requires " 719 "cache.direct=on, which was not specified."); 720 ret = -EINVAL; 721 goto fail; 722 } 723 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { 724 error_prepend(errp, "Unable to use native AIO: "); 725 goto fail; 726 } 727 } 728 #else 729 if (s->use_linux_aio) { 730 error_setg(errp, "aio=native was specified, but is not supported " 731 "in this build."); 732 ret = -EINVAL; 733 goto fail; 734 } 735 #endif /* !defined(CONFIG_LINUX_AIO) */ 736 737 #ifdef CONFIG_LINUX_IO_URING 738 if (s->use_linux_io_uring) { 739 if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs), errp)) { 740 error_prepend(errp, "Unable to use io_uring: "); 741 goto fail; 742 } 743 } 744 #else 745 if (s->use_linux_io_uring) { 746 error_setg(errp, "aio=io_uring was specified, but is not supported " 747 "in this build."); 748 ret = -EINVAL; 749 goto fail; 750 } 751 #endif /* !defined(CONFIG_LINUX_IO_URING) */ 752 753 s->has_discard = true; 754 s->has_write_zeroes = true; 755 756 if (fstat(s->fd, &st) < 0) { 757 ret = -errno; 758 error_setg_errno(errp, errno, "Could not stat file"); 759 goto fail; 760 } 761 762 if (!device) { 763 if (!S_ISREG(st.st_mode)) { 764 error_setg(errp, "'%s' driver requires '%s' to be a regular file", 765 bs->drv->format_name, bs->filename); 766 ret = -EINVAL; 767 goto fail; 768 } else { 769 s->has_fallocate = true; 770 } 771 } else { 772 if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 773 error_setg(errp, "'%s' driver requires '%s' to be either " 774 "a character or block device", 775 bs->drv->format_name, bs->filename); 776 ret = -EINVAL; 777 goto fail; 778 } 779 } 780 #ifdef CONFIG_BLKZONED 781 /* 782 * The kernel page cache does not reliably work for writes to SWR zones 783 * of zoned block device because it can not guarantee the order of writes. 784 */ 785 if ((bs->bl.zoned != BLK_Z_NONE) && 786 (!(s->open_flags & O_DIRECT))) { 787 error_setg(errp, "The driver supports zoned devices, and it requires " 788 "cache.direct=on, which was not specified."); 789 return -EINVAL; /* No host kernel page cache */ 790 } 791 #endif 792 793 if (S_ISBLK(st.st_mode)) { 794 #ifdef __linux__ 795 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do 796 * not rely on the contents of discarded blocks unless using O_DIRECT. 797 * Same for BLKZEROOUT. 798 */ 799 if (!(bs->open_flags & BDRV_O_NOCACHE)) { 800 s->has_write_zeroes = false; 801 } 802 #endif 803 } 804 #ifdef __FreeBSD__ 805 if (S_ISCHR(st.st_mode)) { 806 /* 807 * The file is a char device (disk), which on FreeBSD isn't behind 808 * a pager, so force all requests to be aligned. This is needed 809 * so QEMU makes sure all IO operations on the device are aligned 810 * to sector size, or else FreeBSD will reject them with EINVAL. 811 */ 812 s->force_alignment = true; 813 } 814 #endif 815 s->needs_alignment = raw_needs_alignment(bs); 816 817 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; 818 if (S_ISREG(st.st_mode)) { 819 /* When extending regular files, we get zeros from the OS */ 820 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; 821 } 822 ret = 0; 823 fail: 824 if (ret < 0 && s->fd != -1) { 825 qemu_close(s->fd); 826 } 827 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { 828 unlink(filename); 829 } 830 qemu_opts_del(opts); 831 return ret; 832 } 833 834 static int raw_open(BlockDriverState *bs, QDict *options, int flags, 835 Error **errp) 836 { 837 BDRVRawState *s = bs->opaque; 838 839 s->type = FTYPE_FILE; 840 return raw_open_common(bs, options, flags, 0, false, errp); 841 } 842 843 typedef enum { 844 RAW_PL_PREPARE, 845 RAW_PL_COMMIT, 846 RAW_PL_ABORT, 847 } RawPermLockOp; 848 849 #define PERM_FOREACH(i) \ 850 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) 851 852 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the 853 * file; if @unlock == true, also unlock the unneeded bytes. 854 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. 855 */ 856 static int raw_apply_lock_bytes(BDRVRawState *s, int fd, 857 uint64_t perm_lock_bits, 858 uint64_t shared_perm_lock_bits, 859 bool unlock, Error **errp) 860 { 861 int ret; 862 int i; 863 uint64_t locked_perm, locked_shared_perm; 864 865 if (s) { 866 locked_perm = s->locked_perm; 867 locked_shared_perm = s->locked_shared_perm; 868 } else { 869 /* 870 * We don't have the previous bits, just lock/unlock for each of the 871 * requested bits. 872 */ 873 if (unlock) { 874 locked_perm = BLK_PERM_ALL; 875 locked_shared_perm = BLK_PERM_ALL; 876 } else { 877 locked_perm = 0; 878 locked_shared_perm = 0; 879 } 880 } 881 882 PERM_FOREACH(i) { 883 int off = RAW_LOCK_PERM_BASE + i; 884 uint64_t bit = (1ULL << i); 885 if ((perm_lock_bits & bit) && !(locked_perm & bit)) { 886 ret = qemu_lock_fd(fd, off, 1, false); 887 if (ret) { 888 raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", 889 off); 890 return ret; 891 } else if (s) { 892 s->locked_perm |= bit; 893 } 894 } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { 895 ret = qemu_unlock_fd(fd, off, 1); 896 if (ret) { 897 error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); 898 return ret; 899 } else if (s) { 900 s->locked_perm &= ~bit; 901 } 902 } 903 } 904 PERM_FOREACH(i) { 905 int off = RAW_LOCK_SHARED_BASE + i; 906 uint64_t bit = (1ULL << i); 907 if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { 908 ret = qemu_lock_fd(fd, off, 1, false); 909 if (ret) { 910 raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", 911 off); 912 return ret; 913 } else if (s) { 914 s->locked_shared_perm |= bit; 915 } 916 } else if (unlock && (locked_shared_perm & bit) && 917 !(shared_perm_lock_bits & bit)) { 918 ret = qemu_unlock_fd(fd, off, 1); 919 if (ret) { 920 error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); 921 return ret; 922 } else if (s) { 923 s->locked_shared_perm &= ~bit; 924 } 925 } 926 } 927 return 0; 928 } 929 930 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ 931 static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, 932 Error **errp) 933 { 934 int ret; 935 int i; 936 937 PERM_FOREACH(i) { 938 int off = RAW_LOCK_SHARED_BASE + i; 939 uint64_t p = 1ULL << i; 940 if (perm & p) { 941 ret = qemu_lock_fd_test(fd, off, 1, true); 942 if (ret) { 943 char *perm_name = bdrv_perm_names(p); 944 945 raw_lock_error_setg_errno(errp, -ret, 946 "Failed to get \"%s\" lock", 947 perm_name); 948 g_free(perm_name); 949 return ret; 950 } 951 } 952 } 953 PERM_FOREACH(i) { 954 int off = RAW_LOCK_PERM_BASE + i; 955 uint64_t p = 1ULL << i; 956 if (!(shared_perm & p)) { 957 ret = qemu_lock_fd_test(fd, off, 1, true); 958 if (ret) { 959 char *perm_name = bdrv_perm_names(p); 960 961 raw_lock_error_setg_errno(errp, -ret, 962 "Failed to get shared \"%s\" lock", 963 perm_name); 964 g_free(perm_name); 965 return ret; 966 } 967 } 968 } 969 return 0; 970 } 971 972 static int raw_handle_perm_lock(BlockDriverState *bs, 973 RawPermLockOp op, 974 uint64_t new_perm, uint64_t new_shared, 975 Error **errp) 976 { 977 BDRVRawState *s = bs->opaque; 978 int ret = 0; 979 Error *local_err = NULL; 980 981 if (!s->use_lock) { 982 return 0; 983 } 984 985 if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { 986 return 0; 987 } 988 989 switch (op) { 990 case RAW_PL_PREPARE: 991 if ((s->perm | new_perm) == s->perm && 992 (s->shared_perm & new_shared) == s->shared_perm) 993 { 994 /* 995 * We are going to unlock bytes, it should not fail. If it fail due 996 * to some fs-dependent permission-unrelated reasons (which occurs 997 * sometimes on NFS and leads to abort in bdrv_replace_child) we 998 * can't prevent such errors by any check here. And we ignore them 999 * anyway in ABORT and COMMIT. 1000 */ 1001 return 0; 1002 } 1003 ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, 1004 ~s->shared_perm | ~new_shared, 1005 false, errp); 1006 if (!ret) { 1007 ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); 1008 if (!ret) { 1009 return 0; 1010 } 1011 error_append_hint(errp, 1012 "Is another process using the image [%s]?\n", 1013 bs->filename); 1014 } 1015 /* fall through to unlock bytes. */ 1016 case RAW_PL_ABORT: 1017 raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, 1018 true, &local_err); 1019 if (local_err) { 1020 /* Theoretically the above call only unlocks bytes and it cannot 1021 * fail. Something weird happened, report it. 1022 */ 1023 warn_report_err(local_err); 1024 } 1025 break; 1026 case RAW_PL_COMMIT: 1027 raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, 1028 true, &local_err); 1029 if (local_err) { 1030 /* Theoretically the above call only unlocks bytes and it cannot 1031 * fail. Something weird happened, report it. 1032 */ 1033 warn_report_err(local_err); 1034 } 1035 break; 1036 } 1037 return ret; 1038 } 1039 1040 /* Sets a specific flag */ 1041 static int fcntl_setfl(int fd, int flag) 1042 { 1043 int flags; 1044 1045 flags = fcntl(fd, F_GETFL); 1046 if (flags == -1) { 1047 return -errno; 1048 } 1049 if (fcntl(fd, F_SETFL, flags | flag) == -1) { 1050 return -errno; 1051 } 1052 return 0; 1053 } 1054 1055 static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, 1056 int *open_flags, uint64_t perm, bool force_dup, 1057 Error **errp) 1058 { 1059 BDRVRawState *s = bs->opaque; 1060 int fd = -1; 1061 int ret; 1062 bool has_writers = perm & 1063 (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); 1064 int fcntl_flags = O_APPEND | O_NONBLOCK; 1065 #ifdef O_NOATIME 1066 fcntl_flags |= O_NOATIME; 1067 #endif 1068 1069 *open_flags = 0; 1070 if (s->type == FTYPE_CD) { 1071 *open_flags |= O_NONBLOCK; 1072 } 1073 1074 raw_parse_flags(flags, open_flags, has_writers); 1075 1076 #ifdef O_ASYNC 1077 /* Not all operating systems have O_ASYNC, and those that don't 1078 * will not let us track the state into rs->open_flags (typically 1079 * you achieve the same effect with an ioctl, for example I_SETSIG 1080 * on Solaris). But we do not use O_ASYNC, so that's fine. 1081 */ 1082 assert((s->open_flags & O_ASYNC) == 0); 1083 #endif 1084 1085 if (!force_dup && *open_flags == s->open_flags) { 1086 /* We're lucky, the existing fd is fine */ 1087 return s->fd; 1088 } 1089 1090 if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { 1091 /* dup the original fd */ 1092 fd = qemu_dup(s->fd); 1093 if (fd >= 0) { 1094 ret = fcntl_setfl(fd, *open_flags); 1095 if (ret) { 1096 qemu_close(fd); 1097 fd = -1; 1098 } 1099 } 1100 } 1101 1102 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ 1103 if (fd == -1) { 1104 const char *normalized_filename = bs->filename; 1105 ret = raw_normalize_devicepath(&normalized_filename, errp); 1106 if (ret >= 0) { 1107 fd = qemu_open(normalized_filename, *open_flags, errp); 1108 if (fd == -1) { 1109 return -1; 1110 } 1111 } 1112 } 1113 1114 if (fd != -1 && (*open_flags & O_RDWR)) { 1115 ret = check_hdev_writable(fd); 1116 if (ret < 0) { 1117 qemu_close(fd); 1118 error_setg_errno(errp, -ret, "The device is not writable"); 1119 return -1; 1120 } 1121 } 1122 1123 return fd; 1124 } 1125 1126 static int raw_reopen_prepare(BDRVReopenState *state, 1127 BlockReopenQueue *queue, Error **errp) 1128 { 1129 BDRVRawState *s; 1130 BDRVRawReopenState *rs; 1131 QemuOpts *opts; 1132 int ret; 1133 1134 assert(state != NULL); 1135 assert(state->bs != NULL); 1136 1137 s = state->bs->opaque; 1138 1139 state->opaque = g_new0(BDRVRawReopenState, 1); 1140 rs = state->opaque; 1141 1142 /* Handle options changes */ 1143 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 1144 if (!qemu_opts_absorb_qdict(opts, state->options, errp)) { 1145 ret = -EINVAL; 1146 goto out; 1147 } 1148 1149 rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); 1150 rs->check_cache_dropped = 1151 qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); 1152 1153 /* This driver's reopen function doesn't currently allow changing 1154 * other options, so let's put them back in the original QDict and 1155 * bdrv_reopen_prepare() will detect changes and complain. */ 1156 qemu_opts_to_qdict(opts, state->options); 1157 1158 /* 1159 * As part of reopen prepare we also want to create new fd by 1160 * raw_reconfigure_getfd(). But it wants updated "perm", when in 1161 * bdrv_reopen_multiple() .bdrv_reopen_prepare() callback called prior to 1162 * permission update. Happily, permission update is always a part 1163 * (a separate stage) of bdrv_reopen_multiple() so we can rely on this 1164 * fact and reconfigure fd in raw_check_perm(). 1165 */ 1166 1167 s->reopen_state = state; 1168 ret = 0; 1169 1170 out: 1171 qemu_opts_del(opts); 1172 return ret; 1173 } 1174 1175 static void raw_reopen_commit(BDRVReopenState *state) 1176 { 1177 BDRVRawReopenState *rs = state->opaque; 1178 BDRVRawState *s = state->bs->opaque; 1179 1180 s->drop_cache = rs->drop_cache; 1181 s->check_cache_dropped = rs->check_cache_dropped; 1182 s->open_flags = rs->open_flags; 1183 g_free(state->opaque); 1184 state->opaque = NULL; 1185 1186 assert(s->reopen_state == state); 1187 s->reopen_state = NULL; 1188 } 1189 1190 1191 static void raw_reopen_abort(BDRVReopenState *state) 1192 { 1193 BDRVRawReopenState *rs = state->opaque; 1194 BDRVRawState *s = state->bs->opaque; 1195 1196 /* nothing to do if NULL, we didn't get far enough */ 1197 if (rs == NULL) { 1198 return; 1199 } 1200 1201 g_free(state->opaque); 1202 state->opaque = NULL; 1203 1204 assert(s->reopen_state == state); 1205 s->reopen_state = NULL; 1206 } 1207 1208 static int hdev_get_max_hw_transfer(int fd, struct stat *st) 1209 { 1210 #ifdef BLKSECTGET 1211 if (S_ISBLK(st->st_mode)) { 1212 unsigned short max_sectors = 0; 1213 if (ioctl(fd, BLKSECTGET, &max_sectors) == 0) { 1214 return max_sectors * 512; 1215 } 1216 } else { 1217 int max_bytes = 0; 1218 if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) { 1219 return max_bytes; 1220 } 1221 } 1222 return -errno; 1223 #else 1224 return -ENOSYS; 1225 #endif 1226 } 1227 1228 /* 1229 * Get a sysfs attribute value as character string. 1230 */ 1231 #ifdef CONFIG_LINUX 1232 static int get_sysfs_str_val(struct stat *st, const char *attribute, 1233 char **val) { 1234 g_autofree char *sysfspath = NULL; 1235 size_t len; 1236 1237 if (!S_ISBLK(st->st_mode)) { 1238 return -ENOTSUP; 1239 } 1240 1241 sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/%s", 1242 major(st->st_rdev), minor(st->st_rdev), 1243 attribute); 1244 if (!g_file_get_contents(sysfspath, val, &len, NULL)) { 1245 return -ENOENT; 1246 } 1247 1248 /* The file is ended with '\n' */ 1249 char *p; 1250 p = *val; 1251 if (*(p + len - 1) == '\n') { 1252 *(p + len - 1) = '\0'; 1253 } 1254 return 0; 1255 } 1256 #endif 1257 1258 #if defined(CONFIG_BLKZONED) 1259 static int get_sysfs_zoned_model(struct stat *st, BlockZoneModel *zoned) 1260 { 1261 g_autofree char *val = NULL; 1262 int ret; 1263 1264 ret = get_sysfs_str_val(st, "zoned", &val); 1265 if (ret < 0) { 1266 return ret; 1267 } 1268 1269 if (strcmp(val, "host-managed") == 0) { 1270 *zoned = BLK_Z_HM; 1271 } else if (strcmp(val, "host-aware") == 0) { 1272 *zoned = BLK_Z_HA; 1273 } else if (strcmp(val, "none") == 0) { 1274 *zoned = BLK_Z_NONE; 1275 } else { 1276 return -ENOTSUP; 1277 } 1278 return 0; 1279 } 1280 #endif /* defined(CONFIG_BLKZONED) */ 1281 1282 /* 1283 * Get a sysfs attribute value as a long integer. 1284 */ 1285 #ifdef CONFIG_LINUX 1286 static long get_sysfs_long_val(struct stat *st, const char *attribute) 1287 { 1288 g_autofree char *str = NULL; 1289 const char *end; 1290 long val; 1291 int ret; 1292 1293 ret = get_sysfs_str_val(st, attribute, &str); 1294 if (ret < 0) { 1295 return ret; 1296 } 1297 1298 /* The file is ended with '\n', pass 'end' to accept that. */ 1299 ret = qemu_strtol(str, &end, 10, &val); 1300 if (ret == 0 && end && *end == '\0') { 1301 ret = val; 1302 } 1303 return ret; 1304 } 1305 #endif 1306 1307 static int hdev_get_max_segments(int fd, struct stat *st) 1308 { 1309 #ifdef CONFIG_LINUX 1310 int ret; 1311 1312 if (S_ISCHR(st->st_mode)) { 1313 if (ioctl(fd, SG_GET_SG_TABLESIZE, &ret) == 0) { 1314 return ret; 1315 } 1316 return -ENOTSUP; 1317 } 1318 return get_sysfs_long_val(st, "max_segments"); 1319 #else 1320 return -ENOTSUP; 1321 #endif 1322 } 1323 1324 #if defined(CONFIG_BLKZONED) 1325 /* 1326 * If the reset_all flag is true, then the wps of zone whose state is 1327 * not readonly or offline should be all reset to the start sector. 1328 * Else, take the real wp of the device. 1329 */ 1330 static int get_zones_wp(BlockDriverState *bs, int fd, int64_t offset, 1331 unsigned int nrz, bool reset_all) 1332 { 1333 struct blk_zone *blkz; 1334 size_t rep_size; 1335 uint64_t sector = offset >> BDRV_SECTOR_BITS; 1336 BlockZoneWps *wps = bs->wps; 1337 unsigned int j = offset / bs->bl.zone_size; 1338 unsigned int n = 0, i = 0; 1339 int ret; 1340 rep_size = sizeof(struct blk_zone_report) + nrz * sizeof(struct blk_zone); 1341 g_autofree struct blk_zone_report *rep = NULL; 1342 1343 rep = g_malloc(rep_size); 1344 blkz = (struct blk_zone *)(rep + 1); 1345 while (n < nrz) { 1346 memset(rep, 0, rep_size); 1347 rep->sector = sector; 1348 rep->nr_zones = nrz - n; 1349 1350 do { 1351 ret = ioctl(fd, BLKREPORTZONE, rep); 1352 } while (ret != 0 && errno == EINTR); 1353 if (ret != 0) { 1354 error_report("%d: ioctl BLKREPORTZONE at %" PRId64 " failed %d", 1355 fd, offset, errno); 1356 return -errno; 1357 } 1358 1359 if (!rep->nr_zones) { 1360 break; 1361 } 1362 1363 for (i = 0; i < rep->nr_zones; ++i, ++n, ++j) { 1364 /* 1365 * The wp tracking cares only about sequential writes required and 1366 * sequential write preferred zones so that the wp can advance to 1367 * the right location. 1368 * Use the most significant bit of the wp location to indicate the 1369 * zone type: 0 for SWR/SWP zones and 1 for conventional zones. 1370 */ 1371 if (blkz[i].type == BLK_ZONE_TYPE_CONVENTIONAL) { 1372 wps->wp[j] |= 1ULL << 63; 1373 } else { 1374 switch(blkz[i].cond) { 1375 case BLK_ZONE_COND_FULL: 1376 case BLK_ZONE_COND_READONLY: 1377 /* Zone not writable */ 1378 wps->wp[j] = (blkz[i].start + blkz[i].len) << BDRV_SECTOR_BITS; 1379 break; 1380 case BLK_ZONE_COND_OFFLINE: 1381 /* Zone not writable nor readable */ 1382 wps->wp[j] = (blkz[i].start) << BDRV_SECTOR_BITS; 1383 break; 1384 default: 1385 if (reset_all) { 1386 wps->wp[j] = blkz[i].start << BDRV_SECTOR_BITS; 1387 } else { 1388 wps->wp[j] = blkz[i].wp << BDRV_SECTOR_BITS; 1389 } 1390 break; 1391 } 1392 } 1393 } 1394 sector = blkz[i - 1].start + blkz[i - 1].len; 1395 } 1396 1397 return 0; 1398 } 1399 1400 static void update_zones_wp(BlockDriverState *bs, int fd, int64_t offset, 1401 unsigned int nrz) 1402 { 1403 if (get_zones_wp(bs, fd, offset, nrz, 0) < 0) { 1404 error_report("update zone wp failed"); 1405 } 1406 } 1407 1408 static void raw_refresh_zoned_limits(BlockDriverState *bs, struct stat *st, 1409 Error **errp) 1410 { 1411 BDRVRawState *s = bs->opaque; 1412 BlockZoneModel zoned; 1413 int ret; 1414 1415 bs->bl.zoned = BLK_Z_NONE; 1416 1417 ret = get_sysfs_zoned_model(st, &zoned); 1418 if (ret < 0 || zoned == BLK_Z_NONE) { 1419 return; 1420 } 1421 bs->bl.zoned = zoned; 1422 1423 ret = get_sysfs_long_val(st, "max_open_zones"); 1424 if (ret >= 0) { 1425 bs->bl.max_open_zones = ret; 1426 } 1427 1428 ret = get_sysfs_long_val(st, "max_active_zones"); 1429 if (ret >= 0) { 1430 bs->bl.max_active_zones = ret; 1431 } 1432 1433 /* 1434 * The zoned device must at least have zone size and nr_zones fields. 1435 */ 1436 ret = get_sysfs_long_val(st, "chunk_sectors"); 1437 if (ret < 0) { 1438 error_setg_errno(errp, -ret, "Unable to read chunk_sectors " 1439 "sysfs attribute"); 1440 return; 1441 } else if (!ret) { 1442 error_setg(errp, "Read 0 from chunk_sectors sysfs attribute"); 1443 return; 1444 } 1445 bs->bl.zone_size = ret << BDRV_SECTOR_BITS; 1446 1447 ret = get_sysfs_long_val(st, "nr_zones"); 1448 if (ret < 0) { 1449 error_setg_errno(errp, -ret, "Unable to read nr_zones " 1450 "sysfs attribute"); 1451 return; 1452 } else if (!ret) { 1453 error_setg(errp, "Read 0 from nr_zones sysfs attribute"); 1454 return; 1455 } 1456 bs->bl.nr_zones = ret; 1457 1458 ret = get_sysfs_long_val(st, "zone_append_max_bytes"); 1459 if (ret > 0) { 1460 bs->bl.max_append_sectors = ret >> BDRV_SECTOR_BITS; 1461 } 1462 1463 ret = get_sysfs_long_val(st, "physical_block_size"); 1464 if (ret >= 0) { 1465 bs->bl.write_granularity = ret; 1466 } 1467 1468 /* The refresh_limits() function can be called multiple times. */ 1469 g_free(bs->wps); 1470 bs->wps = g_malloc(sizeof(BlockZoneWps) + 1471 sizeof(int64_t) * bs->bl.nr_zones); 1472 ret = get_zones_wp(bs, s->fd, 0, bs->bl.nr_zones, 0); 1473 if (ret < 0) { 1474 error_setg_errno(errp, -ret, "report wps failed"); 1475 bs->wps = NULL; 1476 return; 1477 } 1478 qemu_co_mutex_init(&bs->wps->colock); 1479 } 1480 #else /* !defined(CONFIG_BLKZONED) */ 1481 static void raw_refresh_zoned_limits(BlockDriverState *bs, struct stat *st, 1482 Error **errp) 1483 { 1484 bs->bl.zoned = BLK_Z_NONE; 1485 } 1486 #endif /* !defined(CONFIG_BLKZONED) */ 1487 1488 static void raw_refresh_limits(BlockDriverState *bs, Error **errp) 1489 { 1490 BDRVRawState *s = bs->opaque; 1491 struct stat st; 1492 1493 s->needs_alignment = raw_needs_alignment(bs); 1494 raw_probe_alignment(bs, s->fd, errp); 1495 1496 bs->bl.min_mem_alignment = s->buf_align; 1497 bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size()); 1498 1499 /* 1500 * Maximum transfers are best effort, so it is okay to ignore any 1501 * errors. That said, based on the man page errors in fstat would be 1502 * very much unexpected; the only possible case seems to be ENOMEM. 1503 */ 1504 if (fstat(s->fd, &st)) { 1505 return; 1506 } 1507 1508 #if defined(__APPLE__) && (__MACH__) 1509 struct statfs buf; 1510 1511 if (!fstatfs(s->fd, &buf)) { 1512 bs->bl.opt_transfer = buf.f_iosize; 1513 bs->bl.pdiscard_alignment = buf.f_bsize; 1514 } 1515 #endif 1516 1517 if (bdrv_is_sg(bs) || S_ISBLK(st.st_mode)) { 1518 int ret = hdev_get_max_hw_transfer(s->fd, &st); 1519 1520 if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { 1521 bs->bl.max_hw_transfer = ret; 1522 } 1523 1524 ret = hdev_get_max_segments(s->fd, &st); 1525 if (ret > 0) { 1526 bs->bl.max_hw_iov = ret; 1527 } 1528 } 1529 1530 raw_refresh_zoned_limits(bs, &st, errp); 1531 } 1532 1533 static int check_for_dasd(int fd) 1534 { 1535 #ifdef BIODASDINFO2 1536 struct dasd_information2_t info = {0}; 1537 1538 return ioctl(fd, BIODASDINFO2, &info); 1539 #else 1540 return -1; 1541 #endif 1542 } 1543 1544 /** 1545 * Try to get @bs's logical and physical block size. 1546 * On success, store them in @bsz and return zero. 1547 * On failure, return negative errno. 1548 */ 1549 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) 1550 { 1551 BDRVRawState *s = bs->opaque; 1552 int ret; 1553 1554 /* If DASD or zoned devices, get blocksizes */ 1555 if (check_for_dasd(s->fd) < 0) { 1556 /* zoned devices are not DASD */ 1557 if (bs->bl.zoned == BLK_Z_NONE) { 1558 return -ENOTSUP; 1559 } 1560 } 1561 ret = probe_logical_blocksize(s->fd, &bsz->log); 1562 if (ret < 0) { 1563 return ret; 1564 } 1565 return probe_physical_blocksize(s->fd, &bsz->phys); 1566 } 1567 1568 /** 1569 * Try to get @bs's geometry: cyls, heads, sectors. 1570 * On success, store them in @geo and return 0. 1571 * On failure return -errno. 1572 * (Allows block driver to assign default geometry values that guest sees) 1573 */ 1574 #ifdef __linux__ 1575 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1576 { 1577 BDRVRawState *s = bs->opaque; 1578 struct hd_geometry ioctl_geo = {0}; 1579 1580 /* If DASD, get its geometry */ 1581 if (check_for_dasd(s->fd) < 0) { 1582 return -ENOTSUP; 1583 } 1584 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { 1585 return -errno; 1586 } 1587 /* HDIO_GETGEO may return success even though geo contains zeros 1588 (e.g. certain multipath setups) */ 1589 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { 1590 return -ENOTSUP; 1591 } 1592 /* Do not return a geometry for partition */ 1593 if (ioctl_geo.start != 0) { 1594 return -ENOTSUP; 1595 } 1596 geo->heads = ioctl_geo.heads; 1597 geo->sectors = ioctl_geo.sectors; 1598 geo->cylinders = ioctl_geo.cylinders; 1599 1600 return 0; 1601 } 1602 #else /* __linux__ */ 1603 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1604 { 1605 return -ENOTSUP; 1606 } 1607 #endif 1608 1609 #if defined(__linux__) 1610 static int handle_aiocb_ioctl(void *opaque) 1611 { 1612 RawPosixAIOData *aiocb = opaque; 1613 int ret; 1614 1615 ret = RETRY_ON_EINTR( 1616 ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf) 1617 ); 1618 if (ret == -1) { 1619 return -errno; 1620 } 1621 1622 return 0; 1623 } 1624 #endif /* linux */ 1625 1626 static int handle_aiocb_flush(void *opaque) 1627 { 1628 RawPosixAIOData *aiocb = opaque; 1629 BDRVRawState *s = aiocb->bs->opaque; 1630 int ret; 1631 1632 if (s->page_cache_inconsistent) { 1633 return -s->page_cache_inconsistent; 1634 } 1635 1636 ret = qemu_fdatasync(aiocb->aio_fildes); 1637 if (ret == -1) { 1638 trace_file_flush_fdatasync_failed(errno); 1639 1640 /* There is no clear definition of the semantics of a failing fsync(), 1641 * so we may have to assume the worst. The sad truth is that this 1642 * assumption is correct for Linux. Some pages are now probably marked 1643 * clean in the page cache even though they are inconsistent with the 1644 * on-disk contents. The next fdatasync() call would succeed, but no 1645 * further writeback attempt will be made. We can't get back to a state 1646 * in which we know what is on disk (we would have to rewrite 1647 * everything that was touched since the last fdatasync() at least), so 1648 * make bdrv_flush() fail permanently. Given that the behaviour isn't 1649 * really defined, I have little hope that other OSes are doing better. 1650 * 1651 * Obviously, this doesn't affect O_DIRECT, which bypasses the page 1652 * cache. */ 1653 if ((s->open_flags & O_DIRECT) == 0) { 1654 s->page_cache_inconsistent = errno; 1655 } 1656 return -errno; 1657 } 1658 return 0; 1659 } 1660 1661 #ifdef CONFIG_PREADV 1662 1663 static bool preadv_present = true; 1664 1665 static ssize_t 1666 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1667 { 1668 return preadv(fd, iov, nr_iov, offset); 1669 } 1670 1671 static ssize_t 1672 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1673 { 1674 return pwritev(fd, iov, nr_iov, offset); 1675 } 1676 1677 #else 1678 1679 static bool preadv_present = false; 1680 1681 static ssize_t 1682 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1683 { 1684 return -ENOSYS; 1685 } 1686 1687 static ssize_t 1688 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1689 { 1690 return -ENOSYS; 1691 } 1692 1693 #endif 1694 1695 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) 1696 { 1697 ssize_t len; 1698 1699 len = RETRY_ON_EINTR( 1700 (aiocb->aio_type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) ? 1701 qemu_pwritev(aiocb->aio_fildes, 1702 aiocb->io.iov, 1703 aiocb->io.niov, 1704 aiocb->aio_offset) : 1705 qemu_preadv(aiocb->aio_fildes, 1706 aiocb->io.iov, 1707 aiocb->io.niov, 1708 aiocb->aio_offset) 1709 ); 1710 1711 if (len == -1) { 1712 return -errno; 1713 } 1714 return len; 1715 } 1716 1717 /* 1718 * Read/writes the data to/from a given linear buffer. 1719 * 1720 * Returns the number of bytes handles or -errno in case of an error. Short 1721 * reads are only returned if the end of the file is reached. 1722 */ 1723 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) 1724 { 1725 ssize_t offset = 0; 1726 ssize_t len; 1727 1728 while (offset < aiocb->aio_nbytes) { 1729 if (aiocb->aio_type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) { 1730 len = pwrite(aiocb->aio_fildes, 1731 (const char *)buf + offset, 1732 aiocb->aio_nbytes - offset, 1733 aiocb->aio_offset + offset); 1734 } else { 1735 len = pread(aiocb->aio_fildes, 1736 buf + offset, 1737 aiocb->aio_nbytes - offset, 1738 aiocb->aio_offset + offset); 1739 } 1740 if (len == -1 && errno == EINTR) { 1741 continue; 1742 } else if (len == -1 && errno == EINVAL && 1743 (aiocb->bs->open_flags & BDRV_O_NOCACHE) && 1744 !(aiocb->aio_type & QEMU_AIO_WRITE) && 1745 offset > 0) { 1746 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned 1747 * after a short read. Assume that O_DIRECT short reads only occur 1748 * at EOF. Therefore this is a short read, not an I/O error. 1749 */ 1750 break; 1751 } else if (len == -1) { 1752 offset = -errno; 1753 break; 1754 } else if (len == 0) { 1755 break; 1756 } 1757 offset += len; 1758 } 1759 1760 return offset; 1761 } 1762 1763 static int handle_aiocb_rw(void *opaque) 1764 { 1765 RawPosixAIOData *aiocb = opaque; 1766 ssize_t nbytes; 1767 char *buf; 1768 1769 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { 1770 /* 1771 * If there is just a single buffer, and it is properly aligned 1772 * we can just use plain pread/pwrite without any problems. 1773 */ 1774 if (aiocb->io.niov == 1) { 1775 nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); 1776 goto out; 1777 } 1778 /* 1779 * We have more than one iovec, and all are properly aligned. 1780 * 1781 * Try preadv/pwritev first and fall back to linearizing the 1782 * buffer if it's not supported. 1783 */ 1784 if (preadv_present) { 1785 nbytes = handle_aiocb_rw_vector(aiocb); 1786 if (nbytes == aiocb->aio_nbytes || 1787 (nbytes < 0 && nbytes != -ENOSYS)) { 1788 goto out; 1789 } 1790 preadv_present = false; 1791 } 1792 1793 /* 1794 * XXX(hch): short read/write. no easy way to handle the reminder 1795 * using these interfaces. For now retry using plain 1796 * pread/pwrite? 1797 */ 1798 } 1799 1800 /* 1801 * Ok, we have to do it the hard way, copy all segments into 1802 * a single aligned buffer. 1803 */ 1804 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); 1805 if (buf == NULL) { 1806 nbytes = -ENOMEM; 1807 goto out; 1808 } 1809 1810 if (aiocb->aio_type & QEMU_AIO_WRITE) { 1811 char *p = buf; 1812 int i; 1813 1814 for (i = 0; i < aiocb->io.niov; ++i) { 1815 memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); 1816 p += aiocb->io.iov[i].iov_len; 1817 } 1818 assert(p - buf == aiocb->aio_nbytes); 1819 } 1820 1821 nbytes = handle_aiocb_rw_linear(aiocb, buf); 1822 if (!(aiocb->aio_type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND))) { 1823 char *p = buf; 1824 size_t count = aiocb->aio_nbytes, copy; 1825 int i; 1826 1827 for (i = 0; i < aiocb->io.niov && count; ++i) { 1828 copy = count; 1829 if (copy > aiocb->io.iov[i].iov_len) { 1830 copy = aiocb->io.iov[i].iov_len; 1831 } 1832 memcpy(aiocb->io.iov[i].iov_base, p, copy); 1833 assert(count >= copy); 1834 p += copy; 1835 count -= copy; 1836 } 1837 assert(count == 0); 1838 } 1839 qemu_vfree(buf); 1840 1841 out: 1842 if (nbytes == aiocb->aio_nbytes) { 1843 return 0; 1844 } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { 1845 if (aiocb->aio_type & QEMU_AIO_WRITE) { 1846 return -EINVAL; 1847 } else { 1848 iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, 1849 0, aiocb->aio_nbytes - nbytes); 1850 return 0; 1851 } 1852 } else { 1853 assert(nbytes < 0); 1854 return nbytes; 1855 } 1856 } 1857 1858 #if defined(CONFIG_FALLOCATE) || defined(BLKZEROOUT) || defined(BLKDISCARD) 1859 static int translate_err(int err) 1860 { 1861 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || 1862 err == -ENOTTY) { 1863 err = -ENOTSUP; 1864 } 1865 return err; 1866 } 1867 #endif 1868 1869 #ifdef CONFIG_FALLOCATE 1870 static int do_fallocate(int fd, int mode, off_t offset, off_t len) 1871 { 1872 do { 1873 if (fallocate(fd, mode, offset, len) == 0) { 1874 return 0; 1875 } 1876 } while (errno == EINTR); 1877 return translate_err(-errno); 1878 } 1879 #endif 1880 1881 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) 1882 { 1883 int ret = -ENOTSUP; 1884 BDRVRawState *s = aiocb->bs->opaque; 1885 1886 if (!s->has_write_zeroes) { 1887 return -ENOTSUP; 1888 } 1889 1890 #ifdef BLKZEROOUT 1891 /* The BLKZEROOUT implementation in the kernel doesn't set 1892 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow 1893 * fallbacks. */ 1894 if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { 1895 do { 1896 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1897 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { 1898 return 0; 1899 } 1900 } while (errno == EINTR); 1901 1902 ret = translate_err(-errno); 1903 if (ret == -ENOTSUP) { 1904 s->has_write_zeroes = false; 1905 } 1906 } 1907 #endif 1908 1909 return ret; 1910 } 1911 1912 static int handle_aiocb_write_zeroes(void *opaque) 1913 { 1914 RawPosixAIOData *aiocb = opaque; 1915 #ifdef CONFIG_FALLOCATE 1916 BDRVRawState *s = aiocb->bs->opaque; 1917 int64_t len; 1918 #endif 1919 1920 if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1921 return handle_aiocb_write_zeroes_block(aiocb); 1922 } 1923 1924 #ifdef CONFIG_FALLOCATE_ZERO_RANGE 1925 if (s->has_write_zeroes) { 1926 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, 1927 aiocb->aio_offset, aiocb->aio_nbytes); 1928 if (ret == -ENOTSUP) { 1929 s->has_write_zeroes = false; 1930 } else if (ret == 0 || ret != -EINVAL) { 1931 return ret; 1932 } 1933 /* 1934 * Note: Some file systems do not like unaligned byte ranges, and 1935 * return EINVAL in such a case, though they should not do it according 1936 * to the man-page of fallocate(). Thus we simply ignore this return 1937 * value and try the other fallbacks instead. 1938 */ 1939 } 1940 #endif 1941 1942 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1943 if (s->has_discard && s->has_fallocate) { 1944 int ret = do_fallocate(s->fd, 1945 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1946 aiocb->aio_offset, aiocb->aio_nbytes); 1947 if (ret == 0) { 1948 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1949 if (ret == 0 || ret != -ENOTSUP) { 1950 return ret; 1951 } 1952 s->has_fallocate = false; 1953 } else if (ret == -EINVAL) { 1954 /* 1955 * Some file systems like older versions of GPFS do not like un- 1956 * aligned byte ranges, and return EINVAL in such a case, though 1957 * they should not do it according to the man-page of fallocate(). 1958 * Warn about the bad filesystem and try the final fallback instead. 1959 */ 1960 warn_report_once("Your file system is misbehaving: " 1961 "fallocate(FALLOC_FL_PUNCH_HOLE) returned EINVAL. " 1962 "Please report this bug to your file system " 1963 "vendor."); 1964 } else if (ret != -ENOTSUP) { 1965 return ret; 1966 } else { 1967 s->has_discard = false; 1968 } 1969 } 1970 #endif 1971 1972 #ifdef CONFIG_FALLOCATE 1973 /* Last resort: we are trying to extend the file with zeroed data. This 1974 * can be done via fallocate(fd, 0) */ 1975 len = raw_getlength(aiocb->bs); 1976 if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { 1977 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1978 if (ret == 0 || ret != -ENOTSUP) { 1979 return ret; 1980 } 1981 s->has_fallocate = false; 1982 } 1983 #endif 1984 1985 return -ENOTSUP; 1986 } 1987 1988 static int handle_aiocb_write_zeroes_unmap(void *opaque) 1989 { 1990 RawPosixAIOData *aiocb = opaque; 1991 BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; 1992 1993 /* First try to write zeros and unmap at the same time */ 1994 1995 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1996 int ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1997 aiocb->aio_offset, aiocb->aio_nbytes); 1998 switch (ret) { 1999 case -ENOTSUP: 2000 case -EINVAL: 2001 case -EBUSY: 2002 break; 2003 default: 2004 return ret; 2005 } 2006 #endif 2007 2008 /* If we couldn't manage to unmap while guaranteed that the area reads as 2009 * all-zero afterwards, just write zeroes without unmapping */ 2010 return handle_aiocb_write_zeroes(aiocb); 2011 } 2012 2013 #ifndef HAVE_COPY_FILE_RANGE 2014 static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, 2015 off_t *out_off, size_t len, unsigned int flags) 2016 { 2017 #ifdef __NR_copy_file_range 2018 return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, 2019 out_off, len, flags); 2020 #else 2021 errno = ENOSYS; 2022 return -1; 2023 #endif 2024 } 2025 #endif 2026 2027 /* 2028 * parse_zone - Fill a zone descriptor 2029 */ 2030 #if defined(CONFIG_BLKZONED) 2031 static inline int parse_zone(struct BlockZoneDescriptor *zone, 2032 const struct blk_zone *blkz) { 2033 zone->start = blkz->start << BDRV_SECTOR_BITS; 2034 zone->length = blkz->len << BDRV_SECTOR_BITS; 2035 zone->wp = blkz->wp << BDRV_SECTOR_BITS; 2036 2037 #ifdef HAVE_BLK_ZONE_REP_CAPACITY 2038 zone->cap = blkz->capacity << BDRV_SECTOR_BITS; 2039 #else 2040 zone->cap = blkz->len << BDRV_SECTOR_BITS; 2041 #endif 2042 2043 switch (blkz->type) { 2044 case BLK_ZONE_TYPE_SEQWRITE_REQ: 2045 zone->type = BLK_ZT_SWR; 2046 break; 2047 case BLK_ZONE_TYPE_SEQWRITE_PREF: 2048 zone->type = BLK_ZT_SWP; 2049 break; 2050 case BLK_ZONE_TYPE_CONVENTIONAL: 2051 zone->type = BLK_ZT_CONV; 2052 break; 2053 default: 2054 error_report("Unsupported zone type: 0x%x", blkz->type); 2055 return -ENOTSUP; 2056 } 2057 2058 switch (blkz->cond) { 2059 case BLK_ZONE_COND_NOT_WP: 2060 zone->state = BLK_ZS_NOT_WP; 2061 break; 2062 case BLK_ZONE_COND_EMPTY: 2063 zone->state = BLK_ZS_EMPTY; 2064 break; 2065 case BLK_ZONE_COND_IMP_OPEN: 2066 zone->state = BLK_ZS_IOPEN; 2067 break; 2068 case BLK_ZONE_COND_EXP_OPEN: 2069 zone->state = BLK_ZS_EOPEN; 2070 break; 2071 case BLK_ZONE_COND_CLOSED: 2072 zone->state = BLK_ZS_CLOSED; 2073 break; 2074 case BLK_ZONE_COND_READONLY: 2075 zone->state = BLK_ZS_RDONLY; 2076 break; 2077 case BLK_ZONE_COND_FULL: 2078 zone->state = BLK_ZS_FULL; 2079 break; 2080 case BLK_ZONE_COND_OFFLINE: 2081 zone->state = BLK_ZS_OFFLINE; 2082 break; 2083 default: 2084 error_report("Unsupported zone state: 0x%x", blkz->cond); 2085 return -ENOTSUP; 2086 } 2087 return 0; 2088 } 2089 #endif 2090 2091 #if defined(CONFIG_BLKZONED) 2092 static int handle_aiocb_zone_report(void *opaque) 2093 { 2094 RawPosixAIOData *aiocb = opaque; 2095 int fd = aiocb->aio_fildes; 2096 unsigned int *nr_zones = aiocb->zone_report.nr_zones; 2097 BlockZoneDescriptor *zones = aiocb->zone_report.zones; 2098 /* zoned block devices use 512-byte sectors */ 2099 uint64_t sector = aiocb->aio_offset / 512; 2100 2101 struct blk_zone *blkz; 2102 size_t rep_size; 2103 unsigned int nrz; 2104 int ret; 2105 unsigned int n = 0, i = 0; 2106 2107 nrz = *nr_zones; 2108 rep_size = sizeof(struct blk_zone_report) + nrz * sizeof(struct blk_zone); 2109 g_autofree struct blk_zone_report *rep = NULL; 2110 rep = g_malloc(rep_size); 2111 2112 blkz = (struct blk_zone *)(rep + 1); 2113 while (n < nrz) { 2114 memset(rep, 0, rep_size); 2115 rep->sector = sector; 2116 rep->nr_zones = nrz - n; 2117 2118 do { 2119 ret = ioctl(fd, BLKREPORTZONE, rep); 2120 } while (ret != 0 && errno == EINTR); 2121 if (ret != 0) { 2122 error_report("%d: ioctl BLKREPORTZONE at %" PRId64 " failed %d", 2123 fd, sector, errno); 2124 return -errno; 2125 } 2126 2127 if (!rep->nr_zones) { 2128 break; 2129 } 2130 2131 for (i = 0; i < rep->nr_zones; i++, n++) { 2132 ret = parse_zone(&zones[n], &blkz[i]); 2133 if (ret != 0) { 2134 return ret; 2135 } 2136 2137 /* The next report should start after the last zone reported */ 2138 sector = blkz[i].start + blkz[i].len; 2139 } 2140 } 2141 2142 *nr_zones = n; 2143 return 0; 2144 } 2145 #endif 2146 2147 #if defined(CONFIG_BLKZONED) 2148 static int handle_aiocb_zone_mgmt(void *opaque) 2149 { 2150 RawPosixAIOData *aiocb = opaque; 2151 int fd = aiocb->aio_fildes; 2152 uint64_t sector = aiocb->aio_offset / 512; 2153 int64_t nr_sectors = aiocb->aio_nbytes / 512; 2154 struct blk_zone_range range; 2155 int ret; 2156 2157 /* Execute the operation */ 2158 range.sector = sector; 2159 range.nr_sectors = nr_sectors; 2160 do { 2161 ret = ioctl(fd, aiocb->zone_mgmt.op, &range); 2162 } while (ret != 0 && errno == EINTR); 2163 2164 return ret < 0 ? -errno : ret; 2165 } 2166 #endif 2167 2168 static int handle_aiocb_copy_range(void *opaque) 2169 { 2170 RawPosixAIOData *aiocb = opaque; 2171 uint64_t bytes = aiocb->aio_nbytes; 2172 off_t in_off = aiocb->aio_offset; 2173 off_t out_off = aiocb->copy_range.aio_offset2; 2174 2175 while (bytes) { 2176 ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, 2177 aiocb->copy_range.aio_fd2, &out_off, 2178 bytes, 0); 2179 trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, 2180 aiocb->copy_range.aio_fd2, out_off, bytes, 2181 0, ret); 2182 if (ret == 0) { 2183 /* No progress (e.g. when beyond EOF), let the caller fall back to 2184 * buffer I/O. */ 2185 return -ENOSPC; 2186 } 2187 if (ret < 0) { 2188 switch (errno) { 2189 case ENOSYS: 2190 return -ENOTSUP; 2191 case EINTR: 2192 continue; 2193 default: 2194 return -errno; 2195 } 2196 } 2197 bytes -= ret; 2198 } 2199 return 0; 2200 } 2201 2202 static int handle_aiocb_discard(void *opaque) 2203 { 2204 RawPosixAIOData *aiocb = opaque; 2205 int ret = -ENOTSUP; 2206 BDRVRawState *s = aiocb->bs->opaque; 2207 2208 if (!s->has_discard) { 2209 return -ENOTSUP; 2210 } 2211 2212 if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 2213 #ifdef BLKDISCARD 2214 do { 2215 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 2216 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { 2217 return 0; 2218 } 2219 } while (errno == EINTR); 2220 2221 ret = translate_err(-errno); 2222 #endif 2223 } else { 2224 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 2225 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 2226 aiocb->aio_offset, aiocb->aio_nbytes); 2227 ret = translate_err(ret); 2228 #elif defined(__APPLE__) && (__MACH__) 2229 fpunchhole_t fpunchhole; 2230 fpunchhole.fp_flags = 0; 2231 fpunchhole.reserved = 0; 2232 fpunchhole.fp_offset = aiocb->aio_offset; 2233 fpunchhole.fp_length = aiocb->aio_nbytes; 2234 if (fcntl(s->fd, F_PUNCHHOLE, &fpunchhole) == -1) { 2235 ret = errno == ENODEV ? -ENOTSUP : -errno; 2236 } else { 2237 ret = 0; 2238 } 2239 #endif 2240 } 2241 2242 if (ret == -ENOTSUP) { 2243 s->has_discard = false; 2244 } 2245 return ret; 2246 } 2247 2248 /* 2249 * Help alignment probing by allocating the first block. 2250 * 2251 * When reading with direct I/O from unallocated area on Gluster backed by XFS, 2252 * reading succeeds regardless of request length. In this case we fallback to 2253 * safe alignment which is not optimal. Allocating the first block avoids this 2254 * fallback. 2255 * 2256 * fd may be opened with O_DIRECT, but we don't know the buffer alignment or 2257 * request alignment, so we use safe values. 2258 * 2259 * Returns: 0 on success, -errno on failure. Since this is an optimization, 2260 * caller may ignore failures. 2261 */ 2262 static int allocate_first_block(int fd, size_t max_size) 2263 { 2264 size_t write_size = (max_size < MAX_BLOCKSIZE) 2265 ? BDRV_SECTOR_SIZE 2266 : MAX_BLOCKSIZE; 2267 size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size()); 2268 void *buf; 2269 ssize_t n; 2270 int ret; 2271 2272 buf = qemu_memalign(max_align, write_size); 2273 memset(buf, 0, write_size); 2274 2275 n = RETRY_ON_EINTR(pwrite(fd, buf, write_size, 0)); 2276 2277 ret = (n == -1) ? -errno : 0; 2278 2279 qemu_vfree(buf); 2280 return ret; 2281 } 2282 2283 static int handle_aiocb_truncate(void *opaque) 2284 { 2285 RawPosixAIOData *aiocb = opaque; 2286 int result = 0; 2287 int64_t current_length = 0; 2288 char *buf = NULL; 2289 struct stat st; 2290 int fd = aiocb->aio_fildes; 2291 int64_t offset = aiocb->aio_offset; 2292 PreallocMode prealloc = aiocb->truncate.prealloc; 2293 Error **errp = aiocb->truncate.errp; 2294 2295 if (fstat(fd, &st) < 0) { 2296 result = -errno; 2297 error_setg_errno(errp, -result, "Could not stat file"); 2298 return result; 2299 } 2300 2301 current_length = st.st_size; 2302 if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { 2303 error_setg(errp, "Cannot use preallocation for shrinking files"); 2304 return -ENOTSUP; 2305 } 2306 2307 switch (prealloc) { 2308 #ifdef CONFIG_POSIX_FALLOCATE 2309 case PREALLOC_MODE_FALLOC: 2310 /* 2311 * Truncating before posix_fallocate() makes it about twice slower on 2312 * file systems that do not support fallocate(), trying to check if a 2313 * block is allocated before allocating it, so don't do that here. 2314 */ 2315 if (offset != current_length) { 2316 result = -posix_fallocate(fd, current_length, 2317 offset - current_length); 2318 if (result != 0) { 2319 /* posix_fallocate() doesn't set errno. */ 2320 error_setg_errno(errp, -result, 2321 "Could not preallocate new data"); 2322 } else if (current_length == 0) { 2323 /* 2324 * posix_fallocate() uses fallocate() if the filesystem 2325 * supports it, or fallback to manually writing zeroes. If 2326 * fallocate() was used, unaligned reads from the fallocated 2327 * area in raw_probe_alignment() will succeed, hence we need to 2328 * allocate the first block. 2329 * 2330 * Optimize future alignment probing; ignore failures. 2331 */ 2332 allocate_first_block(fd, offset); 2333 } 2334 } else { 2335 result = 0; 2336 } 2337 goto out; 2338 #endif 2339 case PREALLOC_MODE_FULL: 2340 { 2341 int64_t num = 0, left = offset - current_length; 2342 off_t seek_result; 2343 2344 /* 2345 * Knowing the final size from the beginning could allow the file 2346 * system driver to do less allocations and possibly avoid 2347 * fragmentation of the file. 2348 */ 2349 if (ftruncate(fd, offset) != 0) { 2350 result = -errno; 2351 error_setg_errno(errp, -result, "Could not resize file"); 2352 goto out; 2353 } 2354 2355 buf = g_malloc0(65536); 2356 2357 seek_result = lseek(fd, current_length, SEEK_SET); 2358 if (seek_result < 0) { 2359 result = -errno; 2360 error_setg_errno(errp, -result, 2361 "Failed to seek to the old end of file"); 2362 goto out; 2363 } 2364 2365 while (left > 0) { 2366 num = MIN(left, 65536); 2367 result = write(fd, buf, num); 2368 if (result < 0) { 2369 if (errno == EINTR) { 2370 continue; 2371 } 2372 result = -errno; 2373 error_setg_errno(errp, -result, 2374 "Could not write zeros for preallocation"); 2375 goto out; 2376 } 2377 left -= result; 2378 } 2379 if (result >= 0) { 2380 result = fsync(fd); 2381 if (result < 0) { 2382 result = -errno; 2383 error_setg_errno(errp, -result, 2384 "Could not flush file to disk"); 2385 goto out; 2386 } 2387 } 2388 goto out; 2389 } 2390 case PREALLOC_MODE_OFF: 2391 if (ftruncate(fd, offset) != 0) { 2392 result = -errno; 2393 error_setg_errno(errp, -result, "Could not resize file"); 2394 } else if (current_length == 0 && offset > current_length) { 2395 /* Optimize future alignment probing; ignore failures. */ 2396 allocate_first_block(fd, offset); 2397 } 2398 return result; 2399 default: 2400 result = -ENOTSUP; 2401 error_setg(errp, "Unsupported preallocation mode: %s", 2402 PreallocMode_str(prealloc)); 2403 return result; 2404 } 2405 2406 out: 2407 if (result < 0) { 2408 if (ftruncate(fd, current_length) < 0) { 2409 error_report("Failed to restore old file length: %s", 2410 strerror(errno)); 2411 } 2412 } 2413 2414 g_free(buf); 2415 return result; 2416 } 2417 2418 static int coroutine_fn raw_thread_pool_submit(ThreadPoolFunc func, void *arg) 2419 { 2420 return thread_pool_submit_co(func, arg); 2421 } 2422 2423 /* 2424 * Check if all memory in this vector is sector aligned. 2425 */ 2426 static bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov) 2427 { 2428 int i; 2429 size_t alignment = bdrv_min_mem_align(bs); 2430 size_t len = bs->bl.request_alignment; 2431 IO_CODE(); 2432 2433 for (i = 0; i < qiov->niov; i++) { 2434 if ((uintptr_t) qiov->iov[i].iov_base % alignment) { 2435 return false; 2436 } 2437 if (qiov->iov[i].iov_len % len) { 2438 return false; 2439 } 2440 } 2441 2442 return true; 2443 } 2444 2445 static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, 2446 uint64_t bytes, QEMUIOVector *qiov, int type) 2447 { 2448 BDRVRawState *s = bs->opaque; 2449 RawPosixAIOData acb; 2450 int ret; 2451 2452 if (fd_open(bs) < 0) 2453 return -EIO; 2454 #if defined(CONFIG_BLKZONED) 2455 if ((type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) && bs->wps) { 2456 qemu_co_mutex_lock(&bs->wps->colock); 2457 if (type & QEMU_AIO_ZONE_APPEND && bs->bl.zone_size) { 2458 int index = offset / bs->bl.zone_size; 2459 offset = bs->wps->wp[index]; 2460 } 2461 } 2462 #endif 2463 2464 /* 2465 * When using O_DIRECT, the request must be aligned to be able to use 2466 * either libaio or io_uring interface. If not fail back to regular thread 2467 * pool read/write code which emulates this for us if we 2468 * set QEMU_AIO_MISALIGNED. 2469 */ 2470 if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) { 2471 type |= QEMU_AIO_MISALIGNED; 2472 #ifdef CONFIG_LINUX_IO_URING 2473 } else if (s->use_linux_io_uring) { 2474 assert(qiov->size == bytes); 2475 ret = luring_co_submit(bs, s->fd, offset, qiov, type); 2476 goto out; 2477 #endif 2478 #ifdef CONFIG_LINUX_AIO 2479 } else if (s->use_linux_aio) { 2480 assert(qiov->size == bytes); 2481 ret = laio_co_submit(s->fd, offset, qiov, type, 2482 s->aio_max_batch); 2483 goto out; 2484 #endif 2485 } 2486 2487 acb = (RawPosixAIOData) { 2488 .bs = bs, 2489 .aio_fildes = s->fd, 2490 .aio_type = type, 2491 .aio_offset = offset, 2492 .aio_nbytes = bytes, 2493 .io = { 2494 .iov = qiov->iov, 2495 .niov = qiov->niov, 2496 }, 2497 }; 2498 2499 assert(qiov->size == bytes); 2500 ret = raw_thread_pool_submit(handle_aiocb_rw, &acb); 2501 goto out; /* Avoid the compiler err of unused label */ 2502 2503 out: 2504 #if defined(CONFIG_BLKZONED) 2505 { 2506 BlockZoneWps *wps = bs->wps; 2507 if (ret == 0) { 2508 if ((type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) 2509 && wps && bs->bl.zone_size) { 2510 uint64_t *wp = &wps->wp[offset / bs->bl.zone_size]; 2511 if (!BDRV_ZT_IS_CONV(*wp)) { 2512 if (type & QEMU_AIO_ZONE_APPEND) { 2513 *s->offset = *wp; 2514 trace_zbd_zone_append_complete(bs, *s->offset 2515 >> BDRV_SECTOR_BITS); 2516 } 2517 /* Advance the wp if needed */ 2518 if (offset + bytes > *wp) { 2519 *wp = offset + bytes; 2520 } 2521 } 2522 } 2523 } else { 2524 if (type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) { 2525 update_zones_wp(bs, s->fd, 0, 1); 2526 } 2527 } 2528 2529 if ((type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) && wps) { 2530 qemu_co_mutex_unlock(&wps->colock); 2531 } 2532 } 2533 #endif 2534 return ret; 2535 } 2536 2537 static int coroutine_fn raw_co_preadv(BlockDriverState *bs, int64_t offset, 2538 int64_t bytes, QEMUIOVector *qiov, 2539 BdrvRequestFlags flags) 2540 { 2541 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); 2542 } 2543 2544 static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, int64_t offset, 2545 int64_t bytes, QEMUIOVector *qiov, 2546 BdrvRequestFlags flags) 2547 { 2548 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); 2549 } 2550 2551 static int coroutine_fn raw_co_flush_to_disk(BlockDriverState *bs) 2552 { 2553 BDRVRawState *s = bs->opaque; 2554 RawPosixAIOData acb; 2555 int ret; 2556 2557 ret = fd_open(bs); 2558 if (ret < 0) { 2559 return ret; 2560 } 2561 2562 acb = (RawPosixAIOData) { 2563 .bs = bs, 2564 .aio_fildes = s->fd, 2565 .aio_type = QEMU_AIO_FLUSH, 2566 }; 2567 2568 #ifdef CONFIG_LINUX_IO_URING 2569 if (s->use_linux_io_uring) { 2570 return luring_co_submit(bs, s->fd, 0, NULL, QEMU_AIO_FLUSH); 2571 } 2572 #endif 2573 return raw_thread_pool_submit(handle_aiocb_flush, &acb); 2574 } 2575 2576 static void raw_aio_attach_aio_context(BlockDriverState *bs, 2577 AioContext *new_context) 2578 { 2579 BDRVRawState __attribute__((unused)) *s = bs->opaque; 2580 #ifdef CONFIG_LINUX_AIO 2581 if (s->use_linux_aio) { 2582 Error *local_err = NULL; 2583 if (!aio_setup_linux_aio(new_context, &local_err)) { 2584 error_reportf_err(local_err, "Unable to use native AIO, " 2585 "falling back to thread pool: "); 2586 s->use_linux_aio = false; 2587 } 2588 } 2589 #endif 2590 #ifdef CONFIG_LINUX_IO_URING 2591 if (s->use_linux_io_uring) { 2592 Error *local_err = NULL; 2593 if (!aio_setup_linux_io_uring(new_context, &local_err)) { 2594 error_reportf_err(local_err, "Unable to use linux io_uring, " 2595 "falling back to thread pool: "); 2596 s->use_linux_io_uring = false; 2597 } 2598 } 2599 #endif 2600 } 2601 2602 static void raw_close(BlockDriverState *bs) 2603 { 2604 BDRVRawState *s = bs->opaque; 2605 2606 if (s->fd >= 0) { 2607 #if defined(CONFIG_BLKZONED) 2608 g_free(bs->wps); 2609 #endif 2610 qemu_close(s->fd); 2611 s->fd = -1; 2612 } 2613 } 2614 2615 /** 2616 * Truncates the given regular file @fd to @offset and, when growing, fills the 2617 * new space according to @prealloc. 2618 * 2619 * Returns: 0 on success, -errno on failure. 2620 */ 2621 static int coroutine_fn 2622 raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, 2623 PreallocMode prealloc, Error **errp) 2624 { 2625 RawPosixAIOData acb; 2626 2627 acb = (RawPosixAIOData) { 2628 .bs = bs, 2629 .aio_fildes = fd, 2630 .aio_type = QEMU_AIO_TRUNCATE, 2631 .aio_offset = offset, 2632 .truncate = { 2633 .prealloc = prealloc, 2634 .errp = errp, 2635 }, 2636 }; 2637 2638 return raw_thread_pool_submit(handle_aiocb_truncate, &acb); 2639 } 2640 2641 static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, 2642 bool exact, PreallocMode prealloc, 2643 BdrvRequestFlags flags, Error **errp) 2644 { 2645 BDRVRawState *s = bs->opaque; 2646 struct stat st; 2647 int ret; 2648 2649 if (fstat(s->fd, &st)) { 2650 ret = -errno; 2651 error_setg_errno(errp, -ret, "Failed to fstat() the file"); 2652 return ret; 2653 } 2654 2655 if (S_ISREG(st.st_mode)) { 2656 /* Always resizes to the exact @offset */ 2657 return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); 2658 } 2659 2660 if (prealloc != PREALLOC_MODE_OFF) { 2661 error_setg(errp, "Preallocation mode '%s' unsupported for this " 2662 "non-regular file", PreallocMode_str(prealloc)); 2663 return -ENOTSUP; 2664 } 2665 2666 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2667 int64_t cur_length = raw_getlength(bs); 2668 2669 if (offset != cur_length && exact) { 2670 error_setg(errp, "Cannot resize device files"); 2671 return -ENOTSUP; 2672 } else if (offset > cur_length) { 2673 error_setg(errp, "Cannot grow device files"); 2674 return -EINVAL; 2675 } 2676 } else { 2677 error_setg(errp, "Resizing this file is not supported"); 2678 return -ENOTSUP; 2679 } 2680 2681 return 0; 2682 } 2683 2684 #ifdef __OpenBSD__ 2685 static int64_t raw_getlength(BlockDriverState *bs) 2686 { 2687 BDRVRawState *s = bs->opaque; 2688 int fd = s->fd; 2689 struct stat st; 2690 2691 if (fstat(fd, &st)) 2692 return -errno; 2693 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2694 struct disklabel dl; 2695 2696 if (ioctl(fd, DIOCGDINFO, &dl)) 2697 return -errno; 2698 return (uint64_t)dl.d_secsize * 2699 dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2700 } else 2701 return st.st_size; 2702 } 2703 #elif defined(__NetBSD__) 2704 static int64_t raw_getlength(BlockDriverState *bs) 2705 { 2706 BDRVRawState *s = bs->opaque; 2707 int fd = s->fd; 2708 struct stat st; 2709 2710 if (fstat(fd, &st)) 2711 return -errno; 2712 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2713 struct dkwedge_info dkw; 2714 2715 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { 2716 return dkw.dkw_size * 512; 2717 } else { 2718 struct disklabel dl; 2719 2720 if (ioctl(fd, DIOCGDINFO, &dl)) 2721 return -errno; 2722 return (uint64_t)dl.d_secsize * 2723 dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2724 } 2725 } else 2726 return st.st_size; 2727 } 2728 #elif defined(__sun__) 2729 static int64_t raw_getlength(BlockDriverState *bs) 2730 { 2731 BDRVRawState *s = bs->opaque; 2732 struct dk_minfo minfo; 2733 int ret; 2734 int64_t size; 2735 2736 ret = fd_open(bs); 2737 if (ret < 0) { 2738 return ret; 2739 } 2740 2741 /* 2742 * Use the DKIOCGMEDIAINFO ioctl to read the size. 2743 */ 2744 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); 2745 if (ret != -1) { 2746 return minfo.dki_lbsize * minfo.dki_capacity; 2747 } 2748 2749 /* 2750 * There are reports that lseek on some devices fails, but 2751 * irc discussion said that contingency on contingency was overkill. 2752 */ 2753 size = lseek(s->fd, 0, SEEK_END); 2754 if (size < 0) { 2755 return -errno; 2756 } 2757 return size; 2758 } 2759 #elif defined(CONFIG_BSD) 2760 static int64_t raw_getlength(BlockDriverState *bs) 2761 { 2762 BDRVRawState *s = bs->opaque; 2763 int fd = s->fd; 2764 int64_t size; 2765 struct stat sb; 2766 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2767 int reopened = 0; 2768 #endif 2769 int ret; 2770 2771 ret = fd_open(bs); 2772 if (ret < 0) 2773 return ret; 2774 2775 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2776 again: 2777 #endif 2778 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { 2779 size = 0; 2780 #ifdef DIOCGMEDIASIZE 2781 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) { 2782 size = 0; 2783 } 2784 #endif 2785 #ifdef DIOCGPART 2786 if (size == 0) { 2787 struct partinfo pi; 2788 if (ioctl(fd, DIOCGPART, &pi) == 0) { 2789 size = pi.media_size; 2790 } 2791 } 2792 #endif 2793 #if defined(DKIOCGETBLOCKCOUNT) && defined(DKIOCGETBLOCKSIZE) 2794 if (size == 0) { 2795 uint64_t sectors = 0; 2796 uint32_t sector_size = 0; 2797 2798 if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 2799 && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { 2800 size = sectors * sector_size; 2801 } 2802 } 2803 #endif 2804 if (size == 0) { 2805 size = lseek(fd, 0LL, SEEK_END); 2806 } 2807 if (size < 0) { 2808 return -errno; 2809 } 2810 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 2811 switch(s->type) { 2812 case FTYPE_CD: 2813 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ 2814 if (size == 2048LL * (unsigned)-1) 2815 size = 0; 2816 /* XXX no disc? maybe we need to reopen... */ 2817 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { 2818 reopened = 1; 2819 goto again; 2820 } 2821 } 2822 #endif 2823 } else { 2824 size = lseek(fd, 0, SEEK_END); 2825 if (size < 0) { 2826 return -errno; 2827 } 2828 } 2829 return size; 2830 } 2831 #else 2832 static int64_t raw_getlength(BlockDriverState *bs) 2833 { 2834 BDRVRawState *s = bs->opaque; 2835 int ret; 2836 int64_t size; 2837 2838 ret = fd_open(bs); 2839 if (ret < 0) { 2840 return ret; 2841 } 2842 2843 size = lseek(s->fd, 0, SEEK_END); 2844 if (size < 0) { 2845 return -errno; 2846 } 2847 return size; 2848 } 2849 #endif 2850 2851 static int64_t coroutine_fn raw_co_getlength(BlockDriverState *bs) 2852 { 2853 return raw_getlength(bs); 2854 } 2855 2856 static int64_t coroutine_fn raw_co_get_allocated_file_size(BlockDriverState *bs) 2857 { 2858 struct stat st; 2859 BDRVRawState *s = bs->opaque; 2860 2861 if (fstat(s->fd, &st) < 0) { 2862 return -errno; 2863 } 2864 return (int64_t)st.st_blocks * 512; 2865 } 2866 2867 static int coroutine_fn 2868 raw_co_create(BlockdevCreateOptions *options, Error **errp) 2869 { 2870 BlockdevCreateOptionsFile *file_opts; 2871 Error *local_err = NULL; 2872 int fd; 2873 uint64_t perm, shared; 2874 int result = 0; 2875 2876 /* Validate options and set default values */ 2877 assert(options->driver == BLOCKDEV_DRIVER_FILE); 2878 file_opts = &options->u.file; 2879 2880 if (!file_opts->has_nocow) { 2881 file_opts->nocow = false; 2882 } 2883 if (!file_opts->has_preallocation) { 2884 file_opts->preallocation = PREALLOC_MODE_OFF; 2885 } 2886 if (!file_opts->has_extent_size_hint) { 2887 file_opts->extent_size_hint = 1 * MiB; 2888 } 2889 if (file_opts->extent_size_hint > UINT32_MAX) { 2890 result = -EINVAL; 2891 error_setg(errp, "Extent size hint is too large"); 2892 goto out; 2893 } 2894 2895 /* Create file */ 2896 fd = qemu_create(file_opts->filename, O_RDWR | O_BINARY, 0644, errp); 2897 if (fd < 0) { 2898 result = -errno; 2899 goto out; 2900 } 2901 2902 /* Take permissions: We want to discard everything, so we need 2903 * BLK_PERM_WRITE; and truncation to the desired size requires 2904 * BLK_PERM_RESIZE. 2905 * On the other hand, we cannot share the RESIZE permission 2906 * because we promise that after this function, the file has the 2907 * size given in the options. If someone else were to resize it 2908 * concurrently, we could not guarantee that. 2909 * Note that after this function, we can no longer guarantee that 2910 * the file is not touched by a third party, so it may be resized 2911 * then. */ 2912 perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; 2913 shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; 2914 2915 /* Step one: Take locks */ 2916 result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); 2917 if (result < 0) { 2918 goto out_close; 2919 } 2920 2921 /* Step two: Check that nobody else has taken conflicting locks */ 2922 result = raw_check_lock_bytes(fd, perm, shared, errp); 2923 if (result < 0) { 2924 error_append_hint(errp, 2925 "Is another process using the image [%s]?\n", 2926 file_opts->filename); 2927 goto out_unlock; 2928 } 2929 2930 /* Clear the file by truncating it to 0 */ 2931 result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); 2932 if (result < 0) { 2933 goto out_unlock; 2934 } 2935 2936 if (file_opts->nocow) { 2937 #ifdef __linux__ 2938 /* Set NOCOW flag to solve performance issue on fs like btrfs. 2939 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value 2940 * will be ignored since any failure of this operation should not 2941 * block the left work. 2942 */ 2943 int attr; 2944 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { 2945 attr |= FS_NOCOW_FL; 2946 ioctl(fd, FS_IOC_SETFLAGS, &attr); 2947 } 2948 #endif 2949 } 2950 #ifdef FS_IOC_FSSETXATTR 2951 /* 2952 * Try to set the extent size hint. Failure is not fatal, and a warning is 2953 * only printed if the option was explicitly specified. 2954 */ 2955 { 2956 struct fsxattr attr; 2957 result = ioctl(fd, FS_IOC_FSGETXATTR, &attr); 2958 if (result == 0) { 2959 attr.fsx_xflags |= FS_XFLAG_EXTSIZE; 2960 attr.fsx_extsize = file_opts->extent_size_hint; 2961 result = ioctl(fd, FS_IOC_FSSETXATTR, &attr); 2962 } 2963 if (result < 0 && file_opts->has_extent_size_hint && 2964 file_opts->extent_size_hint) 2965 { 2966 warn_report("Failed to set extent size hint: %s", 2967 strerror(errno)); 2968 } 2969 } 2970 #endif 2971 2972 /* Resize and potentially preallocate the file to the desired 2973 * final size */ 2974 result = raw_regular_truncate(NULL, fd, file_opts->size, 2975 file_opts->preallocation, errp); 2976 if (result < 0) { 2977 goto out_unlock; 2978 } 2979 2980 out_unlock: 2981 raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); 2982 if (local_err) { 2983 /* The above call should not fail, and if it does, that does 2984 * not mean the whole creation operation has failed. So 2985 * report it the user for their convenience, but do not report 2986 * it to the caller. */ 2987 warn_report_err(local_err); 2988 } 2989 2990 out_close: 2991 if (qemu_close(fd) != 0 && result == 0) { 2992 result = -errno; 2993 error_setg_errno(errp, -result, "Could not close the new file"); 2994 } 2995 out: 2996 return result; 2997 } 2998 2999 static int coroutine_fn GRAPH_RDLOCK 3000 raw_co_create_opts(BlockDriver *drv, const char *filename, 3001 QemuOpts *opts, Error **errp) 3002 { 3003 BlockdevCreateOptions options; 3004 int64_t total_size = 0; 3005 int64_t extent_size_hint = 0; 3006 bool has_extent_size_hint = false; 3007 bool nocow = false; 3008 PreallocMode prealloc; 3009 char *buf = NULL; 3010 Error *local_err = NULL; 3011 3012 /* Skip file: protocol prefix */ 3013 strstart(filename, "file:", &filename); 3014 3015 /* Read out options */ 3016 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 3017 BDRV_SECTOR_SIZE); 3018 if (qemu_opt_get(opts, BLOCK_OPT_EXTENT_SIZE_HINT)) { 3019 has_extent_size_hint = true; 3020 extent_size_hint = 3021 qemu_opt_get_size_del(opts, BLOCK_OPT_EXTENT_SIZE_HINT, -1); 3022 } 3023 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); 3024 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 3025 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, 3026 PREALLOC_MODE_OFF, &local_err); 3027 g_free(buf); 3028 if (local_err) { 3029 error_propagate(errp, local_err); 3030 return -EINVAL; 3031 } 3032 3033 options = (BlockdevCreateOptions) { 3034 .driver = BLOCKDEV_DRIVER_FILE, 3035 .u.file = { 3036 .filename = (char *) filename, 3037 .size = total_size, 3038 .has_preallocation = true, 3039 .preallocation = prealloc, 3040 .has_nocow = true, 3041 .nocow = nocow, 3042 .has_extent_size_hint = has_extent_size_hint, 3043 .extent_size_hint = extent_size_hint, 3044 }, 3045 }; 3046 return raw_co_create(&options, errp); 3047 } 3048 3049 static int coroutine_fn raw_co_delete_file(BlockDriverState *bs, 3050 Error **errp) 3051 { 3052 struct stat st; 3053 int ret; 3054 3055 if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) { 3056 error_setg_errno(errp, ENOENT, "%s is not a regular file", 3057 bs->filename); 3058 return -ENOENT; 3059 } 3060 3061 ret = unlink(bs->filename); 3062 if (ret < 0) { 3063 ret = -errno; 3064 error_setg_errno(errp, -ret, "Error when deleting file %s", 3065 bs->filename); 3066 } 3067 3068 return ret; 3069 } 3070 3071 /* 3072 * Find allocation range in @bs around offset @start. 3073 * May change underlying file descriptor's file offset. 3074 * If @start is not in a hole, store @start in @data, and the 3075 * beginning of the next hole in @hole, and return 0. 3076 * If @start is in a non-trailing hole, store @start in @hole and the 3077 * beginning of the next non-hole in @data, and return 0. 3078 * If @start is in a trailing hole or beyond EOF, return -ENXIO. 3079 * If we can't find out, return a negative errno other than -ENXIO. 3080 */ 3081 static int find_allocation(BlockDriverState *bs, off_t start, 3082 off_t *data, off_t *hole) 3083 { 3084 #if defined SEEK_HOLE && defined SEEK_DATA 3085 BDRVRawState *s = bs->opaque; 3086 off_t offs; 3087 3088 /* 3089 * SEEK_DATA cases: 3090 * D1. offs == start: start is in data 3091 * D2. offs > start: start is in a hole, next data at offs 3092 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 3093 * or start is beyond EOF 3094 * If the latter happens, the file has been truncated behind 3095 * our back since we opened it. All bets are off then. 3096 * Treating like a trailing hole is simplest. 3097 * D4. offs < 0, errno != ENXIO: we learned nothing 3098 */ 3099 offs = lseek(s->fd, start, SEEK_DATA); 3100 if (offs < 0) { 3101 return -errno; /* D3 or D4 */ 3102 } 3103 3104 if (offs < start) { 3105 /* This is not a valid return by lseek(). We are safe to just return 3106 * -EIO in this case, and we'll treat it like D4. */ 3107 return -EIO; 3108 } 3109 3110 if (offs > start) { 3111 /* D2: in hole, next data at offs */ 3112 *hole = start; 3113 *data = offs; 3114 return 0; 3115 } 3116 3117 /* D1: in data, end not yet known */ 3118 3119 /* 3120 * SEEK_HOLE cases: 3121 * H1. offs == start: start is in a hole 3122 * If this happens here, a hole has been dug behind our back 3123 * since the previous lseek(). 3124 * H2. offs > start: either start is in data, next hole at offs, 3125 * or start is in trailing hole, EOF at offs 3126 * Linux treats trailing holes like any other hole: offs == 3127 * start. Solaris seeks to EOF instead: offs > start (blech). 3128 * If that happens here, a hole has been dug behind our back 3129 * since the previous lseek(). 3130 * H3. offs < 0, errno = ENXIO: start is beyond EOF 3131 * If this happens, the file has been truncated behind our 3132 * back since we opened it. Treat it like a trailing hole. 3133 * H4. offs < 0, errno != ENXIO: we learned nothing 3134 * Pretend we know nothing at all, i.e. "forget" about D1. 3135 */ 3136 offs = lseek(s->fd, start, SEEK_HOLE); 3137 if (offs < 0) { 3138 return -errno; /* D1 and (H3 or H4) */ 3139 } 3140 3141 if (offs < start) { 3142 /* This is not a valid return by lseek(). We are safe to just return 3143 * -EIO in this case, and we'll treat it like H4. */ 3144 return -EIO; 3145 } 3146 3147 if (offs > start) { 3148 /* 3149 * D1 and H2: either in data, next hole at offs, or it was in 3150 * data but is now in a trailing hole. In the latter case, 3151 * all bets are off. Treating it as if it there was data all 3152 * the way to EOF is safe, so simply do that. 3153 */ 3154 *data = start; 3155 *hole = offs; 3156 return 0; 3157 } 3158 3159 /* D1 and H1 */ 3160 return -EBUSY; 3161 #else 3162 return -ENOTSUP; 3163 #endif 3164 } 3165 3166 /* 3167 * Returns the allocation status of the specified offset. 3168 * 3169 * The block layer guarantees 'offset' and 'bytes' are within bounds. 3170 * 3171 * 'pnum' is set to the number of bytes (including and immediately following 3172 * the specified offset) that are known to be in the same 3173 * allocated/unallocated state. 3174 * 3175 * 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may 3176 * well exceed it. 3177 */ 3178 static int coroutine_fn raw_co_block_status(BlockDriverState *bs, 3179 bool want_zero, 3180 int64_t offset, 3181 int64_t bytes, int64_t *pnum, 3182 int64_t *map, 3183 BlockDriverState **file) 3184 { 3185 off_t data = 0, hole = 0; 3186 int ret; 3187 3188 assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); 3189 3190 ret = fd_open(bs); 3191 if (ret < 0) { 3192 return ret; 3193 } 3194 3195 if (!want_zero) { 3196 *pnum = bytes; 3197 *map = offset; 3198 *file = bs; 3199 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 3200 } 3201 3202 ret = find_allocation(bs, offset, &data, &hole); 3203 if (ret == -ENXIO) { 3204 /* Trailing hole */ 3205 *pnum = bytes; 3206 ret = BDRV_BLOCK_ZERO; 3207 } else if (ret < 0) { 3208 /* No info available, so pretend there are no holes */ 3209 *pnum = bytes; 3210 ret = BDRV_BLOCK_DATA; 3211 } else if (data == offset) { 3212 /* On a data extent, compute bytes to the end of the extent, 3213 * possibly including a partial sector at EOF. */ 3214 *pnum = hole - offset; 3215 3216 /* 3217 * We are not allowed to return partial sectors, though, so 3218 * round up if necessary. 3219 */ 3220 if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { 3221 int64_t file_length = raw_getlength(bs); 3222 if (file_length > 0) { 3223 /* Ignore errors, this is just a safeguard */ 3224 assert(hole == file_length); 3225 } 3226 *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); 3227 } 3228 3229 ret = BDRV_BLOCK_DATA; 3230 } else { 3231 /* On a hole, compute bytes to the beginning of the next extent. */ 3232 assert(hole == offset); 3233 *pnum = data - offset; 3234 ret = BDRV_BLOCK_ZERO; 3235 } 3236 *map = offset; 3237 *file = bs; 3238 return ret | BDRV_BLOCK_OFFSET_VALID; 3239 } 3240 3241 #if defined(__linux__) 3242 /* Verify that the file is not in the page cache */ 3243 static void check_cache_dropped(BlockDriverState *bs, Error **errp) 3244 { 3245 const size_t window_size = 128 * 1024 * 1024; 3246 BDRVRawState *s = bs->opaque; 3247 void *window = NULL; 3248 size_t length = 0; 3249 unsigned char *vec; 3250 size_t page_size; 3251 off_t offset; 3252 off_t end; 3253 3254 /* mincore(2) page status information requires 1 byte per page */ 3255 page_size = sysconf(_SC_PAGESIZE); 3256 vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); 3257 3258 end = raw_getlength(bs); 3259 3260 for (offset = 0; offset < end; offset += window_size) { 3261 void *new_window; 3262 size_t new_length; 3263 size_t vec_end; 3264 size_t i; 3265 int ret; 3266 3267 /* Unmap previous window if size has changed */ 3268 new_length = MIN(end - offset, window_size); 3269 if (new_length != length) { 3270 munmap(window, length); 3271 window = NULL; 3272 length = 0; 3273 } 3274 3275 new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, 3276 s->fd, offset); 3277 if (new_window == MAP_FAILED) { 3278 error_setg_errno(errp, errno, "mmap failed"); 3279 break; 3280 } 3281 3282 window = new_window; 3283 length = new_length; 3284 3285 ret = mincore(window, length, vec); 3286 if (ret < 0) { 3287 error_setg_errno(errp, errno, "mincore failed"); 3288 break; 3289 } 3290 3291 vec_end = DIV_ROUND_UP(length, page_size); 3292 for (i = 0; i < vec_end; i++) { 3293 if (vec[i] & 0x1) { 3294 break; 3295 } 3296 } 3297 if (i < vec_end) { 3298 error_setg(errp, "page cache still in use!"); 3299 break; 3300 } 3301 } 3302 3303 if (window) { 3304 munmap(window, length); 3305 } 3306 3307 g_free(vec); 3308 } 3309 #endif /* __linux__ */ 3310 3311 static void coroutine_fn GRAPH_RDLOCK 3312 raw_co_invalidate_cache(BlockDriverState *bs, Error **errp) 3313 { 3314 BDRVRawState *s = bs->opaque; 3315 int ret; 3316 3317 ret = fd_open(bs); 3318 if (ret < 0) { 3319 error_setg_errno(errp, -ret, "The file descriptor is not open"); 3320 return; 3321 } 3322 3323 if (!s->drop_cache) { 3324 return; 3325 } 3326 3327 if (s->open_flags & O_DIRECT) { 3328 return; /* No host kernel page cache */ 3329 } 3330 3331 #if defined(__linux__) 3332 /* This sets the scene for the next syscall... */ 3333 ret = bdrv_co_flush(bs); 3334 if (ret < 0) { 3335 error_setg_errno(errp, -ret, "flush failed"); 3336 return; 3337 } 3338 3339 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a 3340 * process. These limitations are okay because we just fsynced the file, 3341 * we don't use mmap, and the file should not be in use by other processes. 3342 */ 3343 ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); 3344 if (ret != 0) { /* the return value is a positive errno */ 3345 error_setg_errno(errp, ret, "fadvise failed"); 3346 return; 3347 } 3348 3349 if (s->check_cache_dropped) { 3350 check_cache_dropped(bs, errp); 3351 } 3352 #else /* __linux__ */ 3353 /* Do nothing. Live migration to a remote host with cache.direct=off is 3354 * unsupported on other host operating systems. Cache consistency issues 3355 * may occur but no error is reported here, partly because that's the 3356 * historical behavior and partly because it's hard to differentiate valid 3357 * configurations that should not cause errors. 3358 */ 3359 #endif /* !__linux__ */ 3360 } 3361 3362 static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret) 3363 { 3364 if (ret) { 3365 s->stats.discard_nb_failed++; 3366 } else { 3367 s->stats.discard_nb_ok++; 3368 s->stats.discard_bytes_ok += nbytes; 3369 } 3370 } 3371 3372 /* 3373 * zone report - Get a zone block device's information in the form 3374 * of an array of zone descriptors. 3375 * zones is an array of zone descriptors to hold zone information on reply; 3376 * offset can be any byte within the entire size of the device; 3377 * nr_zones is the maximum number of sectors the command should operate on. 3378 */ 3379 #if defined(CONFIG_BLKZONED) 3380 static int coroutine_fn raw_co_zone_report(BlockDriverState *bs, int64_t offset, 3381 unsigned int *nr_zones, 3382 BlockZoneDescriptor *zones) { 3383 BDRVRawState *s = bs->opaque; 3384 RawPosixAIOData acb = (RawPosixAIOData) { 3385 .bs = bs, 3386 .aio_fildes = s->fd, 3387 .aio_type = QEMU_AIO_ZONE_REPORT, 3388 .aio_offset = offset, 3389 .zone_report = { 3390 .nr_zones = nr_zones, 3391 .zones = zones, 3392 }, 3393 }; 3394 3395 trace_zbd_zone_report(bs, *nr_zones, offset >> BDRV_SECTOR_BITS); 3396 return raw_thread_pool_submit(handle_aiocb_zone_report, &acb); 3397 } 3398 #endif 3399 3400 /* 3401 * zone management operations - Execute an operation on a zone 3402 */ 3403 #if defined(CONFIG_BLKZONED) 3404 static int coroutine_fn raw_co_zone_mgmt(BlockDriverState *bs, BlockZoneOp op, 3405 int64_t offset, int64_t len) { 3406 BDRVRawState *s = bs->opaque; 3407 RawPosixAIOData acb; 3408 int64_t zone_size, zone_size_mask; 3409 const char *op_name; 3410 unsigned long zo; 3411 int ret; 3412 BlockZoneWps *wps = bs->wps; 3413 int64_t capacity = bs->total_sectors << BDRV_SECTOR_BITS; 3414 3415 zone_size = bs->bl.zone_size; 3416 zone_size_mask = zone_size - 1; 3417 if (offset & zone_size_mask) { 3418 error_report("sector offset %" PRId64 " is not aligned to zone size " 3419 "%" PRId64 "", offset / 512, zone_size / 512); 3420 return -EINVAL; 3421 } 3422 3423 if (((offset + len) < capacity && len & zone_size_mask) || 3424 offset + len > capacity) { 3425 error_report("number of sectors %" PRId64 " is not aligned to zone size" 3426 " %" PRId64 "", len / 512, zone_size / 512); 3427 return -EINVAL; 3428 } 3429 3430 uint32_t i = offset / bs->bl.zone_size; 3431 uint32_t nrz = len / bs->bl.zone_size; 3432 uint64_t *wp = &wps->wp[i]; 3433 if (BDRV_ZT_IS_CONV(*wp) && len != capacity) { 3434 error_report("zone mgmt operations are not allowed for conventional zones"); 3435 return -EIO; 3436 } 3437 3438 switch (op) { 3439 case BLK_ZO_OPEN: 3440 op_name = "BLKOPENZONE"; 3441 zo = BLKOPENZONE; 3442 break; 3443 case BLK_ZO_CLOSE: 3444 op_name = "BLKCLOSEZONE"; 3445 zo = BLKCLOSEZONE; 3446 break; 3447 case BLK_ZO_FINISH: 3448 op_name = "BLKFINISHZONE"; 3449 zo = BLKFINISHZONE; 3450 break; 3451 case BLK_ZO_RESET: 3452 op_name = "BLKRESETZONE"; 3453 zo = BLKRESETZONE; 3454 break; 3455 default: 3456 error_report("Unsupported zone op: 0x%x", op); 3457 return -ENOTSUP; 3458 } 3459 3460 acb = (RawPosixAIOData) { 3461 .bs = bs, 3462 .aio_fildes = s->fd, 3463 .aio_type = QEMU_AIO_ZONE_MGMT, 3464 .aio_offset = offset, 3465 .aio_nbytes = len, 3466 .zone_mgmt = { 3467 .op = zo, 3468 }, 3469 }; 3470 3471 trace_zbd_zone_mgmt(bs, op_name, offset >> BDRV_SECTOR_BITS, 3472 len >> BDRV_SECTOR_BITS); 3473 ret = raw_thread_pool_submit(handle_aiocb_zone_mgmt, &acb); 3474 if (ret != 0) { 3475 update_zones_wp(bs, s->fd, offset, i); 3476 error_report("ioctl %s failed %d", op_name, ret); 3477 return ret; 3478 } 3479 3480 if (zo == BLKRESETZONE && len == capacity) { 3481 ret = get_zones_wp(bs, s->fd, 0, bs->bl.nr_zones, 1); 3482 if (ret < 0) { 3483 error_report("reporting single wp failed"); 3484 return ret; 3485 } 3486 } else if (zo == BLKRESETZONE) { 3487 for (unsigned int j = 0; j < nrz; ++j) { 3488 wp[j] = offset + j * zone_size; 3489 } 3490 } else if (zo == BLKFINISHZONE) { 3491 for (unsigned int j = 0; j < nrz; ++j) { 3492 /* The zoned device allows the last zone smaller that the 3493 * zone size. */ 3494 wp[j] = MIN(offset + (j + 1) * zone_size, offset + len); 3495 } 3496 } 3497 3498 return ret; 3499 } 3500 #endif 3501 3502 #if defined(CONFIG_BLKZONED) 3503 static int coroutine_fn raw_co_zone_append(BlockDriverState *bs, 3504 int64_t *offset, 3505 QEMUIOVector *qiov, 3506 BdrvRequestFlags flags) { 3507 assert(flags == 0); 3508 int64_t zone_size_mask = bs->bl.zone_size - 1; 3509 int64_t iov_len = 0; 3510 int64_t len = 0; 3511 BDRVRawState *s = bs->opaque; 3512 s->offset = offset; 3513 3514 if (*offset & zone_size_mask) { 3515 error_report("sector offset %" PRId64 " is not aligned to zone size " 3516 "%" PRId32 "", *offset / 512, bs->bl.zone_size / 512); 3517 return -EINVAL; 3518 } 3519 3520 int64_t wg = bs->bl.write_granularity; 3521 int64_t wg_mask = wg - 1; 3522 for (int i = 0; i < qiov->niov; i++) { 3523 iov_len = qiov->iov[i].iov_len; 3524 if (iov_len & wg_mask) { 3525 error_report("len of IOVector[%d] %" PRId64 " is not aligned to " 3526 "block size %" PRId64 "", i, iov_len, wg); 3527 return -EINVAL; 3528 } 3529 len += iov_len; 3530 } 3531 3532 trace_zbd_zone_append(bs, *offset >> BDRV_SECTOR_BITS); 3533 return raw_co_prw(bs, *offset, len, qiov, QEMU_AIO_ZONE_APPEND); 3534 } 3535 #endif 3536 3537 static coroutine_fn int 3538 raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes, 3539 bool blkdev) 3540 { 3541 BDRVRawState *s = bs->opaque; 3542 RawPosixAIOData acb; 3543 int ret; 3544 3545 acb = (RawPosixAIOData) { 3546 .bs = bs, 3547 .aio_fildes = s->fd, 3548 .aio_type = QEMU_AIO_DISCARD, 3549 .aio_offset = offset, 3550 .aio_nbytes = bytes, 3551 }; 3552 3553 if (blkdev) { 3554 acb.aio_type |= QEMU_AIO_BLKDEV; 3555 } 3556 3557 ret = raw_thread_pool_submit(handle_aiocb_discard, &acb); 3558 raw_account_discard(s, bytes, ret); 3559 return ret; 3560 } 3561 3562 static coroutine_fn int 3563 raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) 3564 { 3565 return raw_do_pdiscard(bs, offset, bytes, false); 3566 } 3567 3568 static int coroutine_fn 3569 raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, 3570 BdrvRequestFlags flags, bool blkdev) 3571 { 3572 BDRVRawState *s = bs->opaque; 3573 RawPosixAIOData acb; 3574 ThreadPoolFunc *handler; 3575 3576 #ifdef CONFIG_FALLOCATE 3577 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { 3578 BdrvTrackedRequest *req; 3579 3580 /* 3581 * This is a workaround for a bug in the Linux XFS driver, 3582 * where writes submitted through the AIO interface will be 3583 * discarded if they happen beyond a concurrently running 3584 * fallocate() that increases the file length (i.e., both the 3585 * write and the fallocate() happen beyond the EOF). 3586 * 3587 * To work around it, we extend the tracked request for this 3588 * zero write until INT64_MAX (effectively infinity), and mark 3589 * it as serializing. 3590 * 3591 * We have to enable this workaround for all filesystems and 3592 * AIO modes (not just XFS with aio=native), because for 3593 * remote filesystems we do not know the host configuration. 3594 */ 3595 3596 req = bdrv_co_get_self_request(bs); 3597 assert(req); 3598 assert(req->type == BDRV_TRACKED_WRITE); 3599 assert(req->offset <= offset); 3600 assert(req->offset + req->bytes >= offset + bytes); 3601 3602 req->bytes = BDRV_MAX_LENGTH - req->offset; 3603 3604 bdrv_check_request(req->offset, req->bytes, &error_abort); 3605 3606 bdrv_make_request_serialising(req, bs->bl.request_alignment); 3607 } 3608 #endif 3609 3610 acb = (RawPosixAIOData) { 3611 .bs = bs, 3612 .aio_fildes = s->fd, 3613 .aio_type = QEMU_AIO_WRITE_ZEROES, 3614 .aio_offset = offset, 3615 .aio_nbytes = bytes, 3616 }; 3617 3618 if (blkdev) { 3619 acb.aio_type |= QEMU_AIO_BLKDEV; 3620 } 3621 if (flags & BDRV_REQ_NO_FALLBACK) { 3622 acb.aio_type |= QEMU_AIO_NO_FALLBACK; 3623 } 3624 3625 if (flags & BDRV_REQ_MAY_UNMAP) { 3626 acb.aio_type |= QEMU_AIO_DISCARD; 3627 handler = handle_aiocb_write_zeroes_unmap; 3628 } else { 3629 handler = handle_aiocb_write_zeroes; 3630 } 3631 3632 return raw_thread_pool_submit(handler, &acb); 3633 } 3634 3635 static int coroutine_fn raw_co_pwrite_zeroes( 3636 BlockDriverState *bs, int64_t offset, 3637 int64_t bytes, BdrvRequestFlags flags) 3638 { 3639 return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); 3640 } 3641 3642 static int coroutine_fn 3643 raw_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 3644 { 3645 return 0; 3646 } 3647 3648 static ImageInfoSpecific *raw_get_specific_info(BlockDriverState *bs, 3649 Error **errp) 3650 { 3651 ImageInfoSpecificFile *file_info = g_new0(ImageInfoSpecificFile, 1); 3652 ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1); 3653 3654 *spec_info = (ImageInfoSpecific){ 3655 .type = IMAGE_INFO_SPECIFIC_KIND_FILE, 3656 .u.file.data = file_info, 3657 }; 3658 3659 #ifdef FS_IOC_FSGETXATTR 3660 { 3661 BDRVRawState *s = bs->opaque; 3662 struct fsxattr attr; 3663 int ret; 3664 3665 ret = ioctl(s->fd, FS_IOC_FSGETXATTR, &attr); 3666 if (!ret && attr.fsx_extsize != 0) { 3667 file_info->has_extent_size_hint = true; 3668 file_info->extent_size_hint = attr.fsx_extsize; 3669 } 3670 } 3671 #endif 3672 3673 return spec_info; 3674 } 3675 3676 static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs) 3677 { 3678 BDRVRawState *s = bs->opaque; 3679 return (BlockStatsSpecificFile) { 3680 .discard_nb_ok = s->stats.discard_nb_ok, 3681 .discard_nb_failed = s->stats.discard_nb_failed, 3682 .discard_bytes_ok = s->stats.discard_bytes_ok, 3683 }; 3684 } 3685 3686 static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs) 3687 { 3688 BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 3689 3690 stats->driver = BLOCKDEV_DRIVER_FILE; 3691 stats->u.file = get_blockstats_specific_file(bs); 3692 3693 return stats; 3694 } 3695 3696 #if defined(HAVE_HOST_BLOCK_DEVICE) 3697 static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs) 3698 { 3699 BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 3700 3701 stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE; 3702 stats->u.host_device = get_blockstats_specific_file(bs); 3703 3704 return stats; 3705 } 3706 #endif /* HAVE_HOST_BLOCK_DEVICE */ 3707 3708 static QemuOptsList raw_create_opts = { 3709 .name = "raw-create-opts", 3710 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), 3711 .desc = { 3712 { 3713 .name = BLOCK_OPT_SIZE, 3714 .type = QEMU_OPT_SIZE, 3715 .help = "Virtual disk size" 3716 }, 3717 { 3718 .name = BLOCK_OPT_NOCOW, 3719 .type = QEMU_OPT_BOOL, 3720 .help = "Turn off copy-on-write (valid only on btrfs)" 3721 }, 3722 { 3723 .name = BLOCK_OPT_PREALLOC, 3724 .type = QEMU_OPT_STRING, 3725 .help = "Preallocation mode (allowed values: off" 3726 #ifdef CONFIG_POSIX_FALLOCATE 3727 ", falloc" 3728 #endif 3729 ", full)" 3730 }, 3731 { 3732 .name = BLOCK_OPT_EXTENT_SIZE_HINT, 3733 .type = QEMU_OPT_SIZE, 3734 .help = "Extent size hint for the image file, 0 to disable" 3735 }, 3736 { /* end of list */ } 3737 } 3738 }; 3739 3740 static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, 3741 Error **errp) 3742 { 3743 BDRVRawState *s = bs->opaque; 3744 int input_flags = s->reopen_state ? s->reopen_state->flags : bs->open_flags; 3745 int open_flags; 3746 int ret; 3747 3748 /* We may need a new fd if auto-read-only switches the mode */ 3749 ret = raw_reconfigure_getfd(bs, input_flags, &open_flags, perm, 3750 false, errp); 3751 if (ret < 0) { 3752 return ret; 3753 } else if (ret != s->fd) { 3754 Error *local_err = NULL; 3755 3756 /* 3757 * Fail already check_perm() if we can't get a working O_DIRECT 3758 * alignment with the new fd. 3759 */ 3760 raw_probe_alignment(bs, ret, &local_err); 3761 if (local_err) { 3762 error_propagate(errp, local_err); 3763 return -EINVAL; 3764 } 3765 3766 s->perm_change_fd = ret; 3767 s->perm_change_flags = open_flags; 3768 } 3769 3770 /* Prepare permissions on old fd to avoid conflicts between old and new, 3771 * but keep everything locked that new will need. */ 3772 ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); 3773 if (ret < 0) { 3774 goto fail; 3775 } 3776 3777 /* Copy locks to the new fd */ 3778 if (s->perm_change_fd && s->use_lock) { 3779 ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, 3780 false, errp); 3781 if (ret < 0) { 3782 raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 3783 goto fail; 3784 } 3785 } 3786 return 0; 3787 3788 fail: 3789 if (s->perm_change_fd) { 3790 qemu_close(s->perm_change_fd); 3791 } 3792 s->perm_change_fd = 0; 3793 return ret; 3794 } 3795 3796 static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) 3797 { 3798 BDRVRawState *s = bs->opaque; 3799 3800 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is 3801 * called after .bdrv_reopen_commit) */ 3802 if (s->perm_change_fd && s->fd != s->perm_change_fd) { 3803 qemu_close(s->fd); 3804 s->fd = s->perm_change_fd; 3805 s->open_flags = s->perm_change_flags; 3806 } 3807 s->perm_change_fd = 0; 3808 3809 raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); 3810 s->perm = perm; 3811 s->shared_perm = shared; 3812 } 3813 3814 static void raw_abort_perm_update(BlockDriverState *bs) 3815 { 3816 BDRVRawState *s = bs->opaque; 3817 3818 /* For reopen, .bdrv_reopen_abort is called afterwards and will close 3819 * the file descriptor. */ 3820 if (s->perm_change_fd) { 3821 qemu_close(s->perm_change_fd); 3822 } 3823 s->perm_change_fd = 0; 3824 3825 raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 3826 } 3827 3828 static int coroutine_fn GRAPH_RDLOCK raw_co_copy_range_from( 3829 BlockDriverState *bs, BdrvChild *src, int64_t src_offset, 3830 BdrvChild *dst, int64_t dst_offset, int64_t bytes, 3831 BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) 3832 { 3833 return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, 3834 read_flags, write_flags); 3835 } 3836 3837 static int coroutine_fn GRAPH_RDLOCK 3838 raw_co_copy_range_to(BlockDriverState *bs, 3839 BdrvChild *src, int64_t src_offset, 3840 BdrvChild *dst, int64_t dst_offset, 3841 int64_t bytes, BdrvRequestFlags read_flags, 3842 BdrvRequestFlags write_flags) 3843 { 3844 RawPosixAIOData acb; 3845 BDRVRawState *s = bs->opaque; 3846 BDRVRawState *src_s; 3847 3848 assert(dst->bs == bs); 3849 if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { 3850 return -ENOTSUP; 3851 } 3852 3853 src_s = src->bs->opaque; 3854 if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { 3855 return -EIO; 3856 } 3857 3858 acb = (RawPosixAIOData) { 3859 .bs = bs, 3860 .aio_type = QEMU_AIO_COPY_RANGE, 3861 .aio_fildes = src_s->fd, 3862 .aio_offset = src_offset, 3863 .aio_nbytes = bytes, 3864 .copy_range = { 3865 .aio_fd2 = s->fd, 3866 .aio_offset2 = dst_offset, 3867 }, 3868 }; 3869 3870 return raw_thread_pool_submit(handle_aiocb_copy_range, &acb); 3871 } 3872 3873 BlockDriver bdrv_file = { 3874 .format_name = "file", 3875 .protocol_name = "file", 3876 .instance_size = sizeof(BDRVRawState), 3877 .bdrv_needs_filename = true, 3878 .bdrv_probe = NULL, /* no probe for protocols */ 3879 .bdrv_parse_filename = raw_parse_filename, 3880 .bdrv_file_open = raw_open, 3881 .bdrv_reopen_prepare = raw_reopen_prepare, 3882 .bdrv_reopen_commit = raw_reopen_commit, 3883 .bdrv_reopen_abort = raw_reopen_abort, 3884 .bdrv_close = raw_close, 3885 .bdrv_co_create = raw_co_create, 3886 .bdrv_co_create_opts = raw_co_create_opts, 3887 .bdrv_has_zero_init = bdrv_has_zero_init_1, 3888 .bdrv_co_block_status = raw_co_block_status, 3889 .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3890 .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, 3891 .bdrv_co_delete_file = raw_co_delete_file, 3892 3893 .bdrv_co_preadv = raw_co_preadv, 3894 .bdrv_co_pwritev = raw_co_pwritev, 3895 .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3896 .bdrv_co_pdiscard = raw_co_pdiscard, 3897 .bdrv_co_copy_range_from = raw_co_copy_range_from, 3898 .bdrv_co_copy_range_to = raw_co_copy_range_to, 3899 .bdrv_refresh_limits = raw_refresh_limits, 3900 .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3901 3902 .bdrv_co_truncate = raw_co_truncate, 3903 .bdrv_co_getlength = raw_co_getlength, 3904 .bdrv_co_get_info = raw_co_get_info, 3905 .bdrv_get_specific_info = raw_get_specific_info, 3906 .bdrv_co_get_allocated_file_size = raw_co_get_allocated_file_size, 3907 .bdrv_get_specific_stats = raw_get_specific_stats, 3908 .bdrv_check_perm = raw_check_perm, 3909 .bdrv_set_perm = raw_set_perm, 3910 .bdrv_abort_perm_update = raw_abort_perm_update, 3911 .create_opts = &raw_create_opts, 3912 .mutable_opts = mutable_opts, 3913 }; 3914 3915 /***********************************************/ 3916 /* host device */ 3917 3918 #if defined(HAVE_HOST_BLOCK_DEVICE) 3919 3920 #if defined(__APPLE__) && defined(__MACH__) 3921 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3922 CFIndex maxPathSize, int flags); 3923 3924 #if !defined(MAC_OS_VERSION_12_0) \ 3925 || (MAC_OS_X_VERSION_MIN_REQUIRED < MAC_OS_VERSION_12_0) 3926 #define IOMainPort IOMasterPort 3927 #endif 3928 3929 static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) 3930 { 3931 kern_return_t kernResult = KERN_FAILURE; 3932 mach_port_t mainPort; 3933 CFMutableDictionaryRef classesToMatch; 3934 const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; 3935 char *mediaType = NULL; 3936 3937 kernResult = IOMainPort(MACH_PORT_NULL, &mainPort); 3938 if ( KERN_SUCCESS != kernResult ) { 3939 printf("IOMainPort returned %d\n", kernResult); 3940 } 3941 3942 int index; 3943 for (index = 0; index < ARRAY_SIZE(matching_array); index++) { 3944 classesToMatch = IOServiceMatching(matching_array[index]); 3945 if (classesToMatch == NULL) { 3946 error_report("IOServiceMatching returned NULL for %s", 3947 matching_array[index]); 3948 continue; 3949 } 3950 CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), 3951 kCFBooleanTrue); 3952 kernResult = IOServiceGetMatchingServices(mainPort, classesToMatch, 3953 mediaIterator); 3954 if (kernResult != KERN_SUCCESS) { 3955 error_report("Note: IOServiceGetMatchingServices returned %d", 3956 kernResult); 3957 continue; 3958 } 3959 3960 /* If a match was found, leave the loop */ 3961 if (*mediaIterator != 0) { 3962 trace_file_FindEjectableOpticalMedia(matching_array[index]); 3963 mediaType = g_strdup(matching_array[index]); 3964 break; 3965 } 3966 } 3967 return mediaType; 3968 } 3969 3970 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3971 CFIndex maxPathSize, int flags) 3972 { 3973 io_object_t nextMedia; 3974 kern_return_t kernResult = KERN_FAILURE; 3975 *bsdPath = '\0'; 3976 nextMedia = IOIteratorNext( mediaIterator ); 3977 if ( nextMedia ) 3978 { 3979 CFTypeRef bsdPathAsCFString; 3980 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); 3981 if ( bsdPathAsCFString ) { 3982 size_t devPathLength; 3983 strcpy( bsdPath, _PATH_DEV ); 3984 if (flags & BDRV_O_NOCACHE) { 3985 strcat(bsdPath, "r"); 3986 } 3987 devPathLength = strlen( bsdPath ); 3988 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { 3989 kernResult = KERN_SUCCESS; 3990 } 3991 CFRelease( bsdPathAsCFString ); 3992 } 3993 IOObjectRelease( nextMedia ); 3994 } 3995 3996 return kernResult; 3997 } 3998 3999 /* Sets up a real cdrom for use in QEMU */ 4000 static bool setup_cdrom(char *bsd_path, Error **errp) 4001 { 4002 int index, num_of_test_partitions = 2, fd; 4003 char test_partition[MAXPATHLEN]; 4004 bool partition_found = false; 4005 4006 /* look for a working partition */ 4007 for (index = 0; index < num_of_test_partitions; index++) { 4008 snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, 4009 index); 4010 fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE, NULL); 4011 if (fd >= 0) { 4012 partition_found = true; 4013 qemu_close(fd); 4014 break; 4015 } 4016 } 4017 4018 /* if a working partition on the device was not found */ 4019 if (partition_found == false) { 4020 error_setg(errp, "Failed to find a working partition on disc"); 4021 } else { 4022 trace_file_setup_cdrom(test_partition); 4023 pstrcpy(bsd_path, MAXPATHLEN, test_partition); 4024 } 4025 return partition_found; 4026 } 4027 4028 /* Prints directions on mounting and unmounting a device */ 4029 static void print_unmounting_directions(const char *file_name) 4030 { 4031 error_report("If device %s is mounted on the desktop, unmount" 4032 " it first before using it in QEMU", file_name); 4033 error_report("Command to unmount device: diskutil unmountDisk %s", 4034 file_name); 4035 error_report("Command to mount device: diskutil mountDisk %s", file_name); 4036 } 4037 4038 #endif /* defined(__APPLE__) && defined(__MACH__) */ 4039 4040 static int hdev_probe_device(const char *filename) 4041 { 4042 struct stat st; 4043 4044 /* allow a dedicated CD-ROM driver to match with a higher priority */ 4045 if (strstart(filename, "/dev/cdrom", NULL)) 4046 return 50; 4047 4048 if (stat(filename, &st) >= 0 && 4049 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 4050 return 100; 4051 } 4052 4053 return 0; 4054 } 4055 4056 static void hdev_parse_filename(const char *filename, QDict *options, 4057 Error **errp) 4058 { 4059 bdrv_parse_filename_strip_prefix(filename, "host_device:", options); 4060 } 4061 4062 static bool hdev_is_sg(BlockDriverState *bs) 4063 { 4064 4065 #if defined(__linux__) 4066 4067 BDRVRawState *s = bs->opaque; 4068 struct stat st; 4069 struct sg_scsi_id scsiid; 4070 int sg_version; 4071 int ret; 4072 4073 if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { 4074 return false; 4075 } 4076 4077 ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); 4078 if (ret < 0) { 4079 return false; 4080 } 4081 4082 ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); 4083 if (ret >= 0) { 4084 trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); 4085 return true; 4086 } 4087 4088 #endif 4089 4090 return false; 4091 } 4092 4093 static int hdev_open(BlockDriverState *bs, QDict *options, int flags, 4094 Error **errp) 4095 { 4096 BDRVRawState *s = bs->opaque; 4097 int ret; 4098 4099 #if defined(__APPLE__) && defined(__MACH__) 4100 /* 4101 * Caution: while qdict_get_str() is fine, getting non-string types 4102 * would require more care. When @options come from -blockdev or 4103 * blockdev_add, its members are typed according to the QAPI 4104 * schema, but when they come from -drive, they're all QString. 4105 */ 4106 const char *filename = qdict_get_str(options, "filename"); 4107 char bsd_path[MAXPATHLEN] = ""; 4108 bool error_occurred = false; 4109 4110 /* If using a real cdrom */ 4111 if (strcmp(filename, "/dev/cdrom") == 0) { 4112 char *mediaType = NULL; 4113 kern_return_t ret_val; 4114 io_iterator_t mediaIterator = 0; 4115 4116 mediaType = FindEjectableOpticalMedia(&mediaIterator); 4117 if (mediaType == NULL) { 4118 error_setg(errp, "Please make sure your CD/DVD is in the optical" 4119 " drive"); 4120 error_occurred = true; 4121 goto hdev_open_Mac_error; 4122 } 4123 4124 ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); 4125 if (ret_val != KERN_SUCCESS) { 4126 error_setg(errp, "Could not get BSD path for optical drive"); 4127 error_occurred = true; 4128 goto hdev_open_Mac_error; 4129 } 4130 4131 /* If a real optical drive was not found */ 4132 if (bsd_path[0] == '\0') { 4133 error_setg(errp, "Failed to obtain bsd path for optical drive"); 4134 error_occurred = true; 4135 goto hdev_open_Mac_error; 4136 } 4137 4138 /* If using a cdrom disc and finding a partition on the disc failed */ 4139 if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && 4140 setup_cdrom(bsd_path, errp) == false) { 4141 print_unmounting_directions(bsd_path); 4142 error_occurred = true; 4143 goto hdev_open_Mac_error; 4144 } 4145 4146 qdict_put_str(options, "filename", bsd_path); 4147 4148 hdev_open_Mac_error: 4149 g_free(mediaType); 4150 if (mediaIterator) { 4151 IOObjectRelease(mediaIterator); 4152 } 4153 if (error_occurred) { 4154 return -ENOENT; 4155 } 4156 } 4157 #endif /* defined(__APPLE__) && defined(__MACH__) */ 4158 4159 s->type = FTYPE_FILE; 4160 4161 ret = raw_open_common(bs, options, flags, 0, true, errp); 4162 if (ret < 0) { 4163 #if defined(__APPLE__) && defined(__MACH__) 4164 if (*bsd_path) { 4165 filename = bsd_path; 4166 } 4167 /* if a physical device experienced an error while being opened */ 4168 if (strncmp(filename, "/dev/", 5) == 0) { 4169 print_unmounting_directions(filename); 4170 } 4171 #endif /* defined(__APPLE__) && defined(__MACH__) */ 4172 return ret; 4173 } 4174 4175 /* Since this does ioctl the device must be already opened */ 4176 bs->sg = hdev_is_sg(bs); 4177 4178 return ret; 4179 } 4180 4181 #if defined(__linux__) 4182 static int coroutine_fn 4183 hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) 4184 { 4185 BDRVRawState *s = bs->opaque; 4186 RawPosixAIOData acb; 4187 int ret; 4188 4189 ret = fd_open(bs); 4190 if (ret < 0) { 4191 return ret; 4192 } 4193 4194 if (req == SG_IO && s->pr_mgr) { 4195 struct sg_io_hdr *io_hdr = buf; 4196 if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || 4197 io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { 4198 return pr_manager_execute(s->pr_mgr, qemu_get_current_aio_context(), 4199 s->fd, io_hdr); 4200 } 4201 } 4202 4203 acb = (RawPosixAIOData) { 4204 .bs = bs, 4205 .aio_type = QEMU_AIO_IOCTL, 4206 .aio_fildes = s->fd, 4207 .aio_offset = 0, 4208 .ioctl = { 4209 .buf = buf, 4210 .cmd = req, 4211 }, 4212 }; 4213 4214 return raw_thread_pool_submit(handle_aiocb_ioctl, &acb); 4215 } 4216 #endif /* linux */ 4217 4218 static coroutine_fn int 4219 hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) 4220 { 4221 BDRVRawState *s = bs->opaque; 4222 int ret; 4223 4224 ret = fd_open(bs); 4225 if (ret < 0) { 4226 raw_account_discard(s, bytes, ret); 4227 return ret; 4228 } 4229 return raw_do_pdiscard(bs, offset, bytes, true); 4230 } 4231 4232 static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, 4233 int64_t offset, int64_t bytes, BdrvRequestFlags flags) 4234 { 4235 int rc; 4236 4237 rc = fd_open(bs); 4238 if (rc < 0) { 4239 return rc; 4240 } 4241 4242 return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); 4243 } 4244 4245 static BlockDriver bdrv_host_device = { 4246 .format_name = "host_device", 4247 .protocol_name = "host_device", 4248 .instance_size = sizeof(BDRVRawState), 4249 .bdrv_needs_filename = true, 4250 .bdrv_probe_device = hdev_probe_device, 4251 .bdrv_parse_filename = hdev_parse_filename, 4252 .bdrv_file_open = hdev_open, 4253 .bdrv_close = raw_close, 4254 .bdrv_reopen_prepare = raw_reopen_prepare, 4255 .bdrv_reopen_commit = raw_reopen_commit, 4256 .bdrv_reopen_abort = raw_reopen_abort, 4257 .bdrv_co_create_opts = bdrv_co_create_opts_simple, 4258 .create_opts = &bdrv_create_opts_simple, 4259 .mutable_opts = mutable_opts, 4260 .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 4261 .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, 4262 4263 .bdrv_co_preadv = raw_co_preadv, 4264 .bdrv_co_pwritev = raw_co_pwritev, 4265 .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 4266 .bdrv_co_pdiscard = hdev_co_pdiscard, 4267 .bdrv_co_copy_range_from = raw_co_copy_range_from, 4268 .bdrv_co_copy_range_to = raw_co_copy_range_to, 4269 .bdrv_refresh_limits = raw_refresh_limits, 4270 .bdrv_attach_aio_context = raw_aio_attach_aio_context, 4271 4272 .bdrv_co_truncate = raw_co_truncate, 4273 .bdrv_co_getlength = raw_co_getlength, 4274 .bdrv_co_get_info = raw_co_get_info, 4275 .bdrv_get_specific_info = raw_get_specific_info, 4276 .bdrv_co_get_allocated_file_size = raw_co_get_allocated_file_size, 4277 .bdrv_get_specific_stats = hdev_get_specific_stats, 4278 .bdrv_check_perm = raw_check_perm, 4279 .bdrv_set_perm = raw_set_perm, 4280 .bdrv_abort_perm_update = raw_abort_perm_update, 4281 .bdrv_probe_blocksizes = hdev_probe_blocksizes, 4282 .bdrv_probe_geometry = hdev_probe_geometry, 4283 4284 /* generic scsi device */ 4285 #ifdef __linux__ 4286 .bdrv_co_ioctl = hdev_co_ioctl, 4287 #endif 4288 4289 /* zoned device */ 4290 #if defined(CONFIG_BLKZONED) 4291 /* zone management operations */ 4292 .bdrv_co_zone_report = raw_co_zone_report, 4293 .bdrv_co_zone_mgmt = raw_co_zone_mgmt, 4294 .bdrv_co_zone_append = raw_co_zone_append, 4295 #endif 4296 }; 4297 4298 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 4299 static void cdrom_parse_filename(const char *filename, QDict *options, 4300 Error **errp) 4301 { 4302 bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); 4303 } 4304 4305 static void cdrom_refresh_limits(BlockDriverState *bs, Error **errp) 4306 { 4307 bs->bl.has_variable_length = true; 4308 raw_refresh_limits(bs, errp); 4309 } 4310 #endif 4311 4312 #ifdef __linux__ 4313 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 4314 Error **errp) 4315 { 4316 BDRVRawState *s = bs->opaque; 4317 4318 s->type = FTYPE_CD; 4319 4320 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ 4321 return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); 4322 } 4323 4324 static int cdrom_probe_device(const char *filename) 4325 { 4326 int fd, ret; 4327 int prio = 0; 4328 struct stat st; 4329 4330 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK, NULL); 4331 if (fd < 0) { 4332 goto out; 4333 } 4334 ret = fstat(fd, &st); 4335 if (ret == -1 || !S_ISBLK(st.st_mode)) { 4336 goto outc; 4337 } 4338 4339 /* Attempt to detect via a CDROM specific ioctl */ 4340 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 4341 if (ret >= 0) 4342 prio = 100; 4343 4344 outc: 4345 qemu_close(fd); 4346 out: 4347 return prio; 4348 } 4349 4350 static bool coroutine_fn cdrom_co_is_inserted(BlockDriverState *bs) 4351 { 4352 BDRVRawState *s = bs->opaque; 4353 int ret; 4354 4355 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 4356 return ret == CDS_DISC_OK; 4357 } 4358 4359 static void coroutine_fn cdrom_co_eject(BlockDriverState *bs, bool eject_flag) 4360 { 4361 BDRVRawState *s = bs->opaque; 4362 4363 if (eject_flag) { 4364 if (ioctl(s->fd, CDROMEJECT, NULL) < 0) 4365 perror("CDROMEJECT"); 4366 } else { 4367 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) 4368 perror("CDROMEJECT"); 4369 } 4370 } 4371 4372 static void coroutine_fn cdrom_co_lock_medium(BlockDriverState *bs, bool locked) 4373 { 4374 BDRVRawState *s = bs->opaque; 4375 4376 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { 4377 /* 4378 * Note: an error can happen if the distribution automatically 4379 * mounts the CD-ROM 4380 */ 4381 /* perror("CDROM_LOCKDOOR"); */ 4382 } 4383 } 4384 4385 static BlockDriver bdrv_host_cdrom = { 4386 .format_name = "host_cdrom", 4387 .protocol_name = "host_cdrom", 4388 .instance_size = sizeof(BDRVRawState), 4389 .bdrv_needs_filename = true, 4390 .bdrv_probe_device = cdrom_probe_device, 4391 .bdrv_parse_filename = cdrom_parse_filename, 4392 .bdrv_file_open = cdrom_open, 4393 .bdrv_close = raw_close, 4394 .bdrv_reopen_prepare = raw_reopen_prepare, 4395 .bdrv_reopen_commit = raw_reopen_commit, 4396 .bdrv_reopen_abort = raw_reopen_abort, 4397 .bdrv_co_create_opts = bdrv_co_create_opts_simple, 4398 .create_opts = &bdrv_create_opts_simple, 4399 .mutable_opts = mutable_opts, 4400 .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 4401 4402 .bdrv_co_preadv = raw_co_preadv, 4403 .bdrv_co_pwritev = raw_co_pwritev, 4404 .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 4405 .bdrv_refresh_limits = cdrom_refresh_limits, 4406 .bdrv_attach_aio_context = raw_aio_attach_aio_context, 4407 4408 .bdrv_co_truncate = raw_co_truncate, 4409 .bdrv_co_getlength = raw_co_getlength, 4410 .bdrv_co_get_allocated_file_size = raw_co_get_allocated_file_size, 4411 4412 /* removable device support */ 4413 .bdrv_co_is_inserted = cdrom_co_is_inserted, 4414 .bdrv_co_eject = cdrom_co_eject, 4415 .bdrv_co_lock_medium = cdrom_co_lock_medium, 4416 4417 /* generic scsi device */ 4418 .bdrv_co_ioctl = hdev_co_ioctl, 4419 }; 4420 #endif /* __linux__ */ 4421 4422 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 4423 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 4424 Error **errp) 4425 { 4426 BDRVRawState *s = bs->opaque; 4427 int ret; 4428 4429 s->type = FTYPE_CD; 4430 4431 ret = raw_open_common(bs, options, flags, 0, true, errp); 4432 if (ret) { 4433 return ret; 4434 } 4435 4436 /* make sure the door isn't locked at this time */ 4437 ioctl(s->fd, CDIOCALLOW); 4438 return 0; 4439 } 4440 4441 static int cdrom_probe_device(const char *filename) 4442 { 4443 if (strstart(filename, "/dev/cd", NULL) || 4444 strstart(filename, "/dev/acd", NULL)) 4445 return 100; 4446 return 0; 4447 } 4448 4449 static int cdrom_reopen(BlockDriverState *bs) 4450 { 4451 BDRVRawState *s = bs->opaque; 4452 int fd; 4453 4454 /* 4455 * Force reread of possibly changed/newly loaded disc, 4456 * FreeBSD seems to not notice sometimes... 4457 */ 4458 if (s->fd >= 0) 4459 qemu_close(s->fd); 4460 fd = qemu_open(bs->filename, s->open_flags, NULL); 4461 if (fd < 0) { 4462 s->fd = -1; 4463 return -EIO; 4464 } 4465 s->fd = fd; 4466 4467 /* make sure the door isn't locked at this time */ 4468 ioctl(s->fd, CDIOCALLOW); 4469 return 0; 4470 } 4471 4472 static bool coroutine_fn cdrom_co_is_inserted(BlockDriverState *bs) 4473 { 4474 return raw_getlength(bs) > 0; 4475 } 4476 4477 static void coroutine_fn cdrom_co_eject(BlockDriverState *bs, bool eject_flag) 4478 { 4479 BDRVRawState *s = bs->opaque; 4480 4481 if (s->fd < 0) 4482 return; 4483 4484 (void) ioctl(s->fd, CDIOCALLOW); 4485 4486 if (eject_flag) { 4487 if (ioctl(s->fd, CDIOCEJECT) < 0) 4488 perror("CDIOCEJECT"); 4489 } else { 4490 if (ioctl(s->fd, CDIOCCLOSE) < 0) 4491 perror("CDIOCCLOSE"); 4492 } 4493 4494 cdrom_reopen(bs); 4495 } 4496 4497 static void coroutine_fn cdrom_co_lock_medium(BlockDriverState *bs, bool locked) 4498 { 4499 BDRVRawState *s = bs->opaque; 4500 4501 if (s->fd < 0) 4502 return; 4503 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { 4504 /* 4505 * Note: an error can happen if the distribution automatically 4506 * mounts the CD-ROM 4507 */ 4508 /* perror("CDROM_LOCKDOOR"); */ 4509 } 4510 } 4511 4512 static BlockDriver bdrv_host_cdrom = { 4513 .format_name = "host_cdrom", 4514 .protocol_name = "host_cdrom", 4515 .instance_size = sizeof(BDRVRawState), 4516 .bdrv_needs_filename = true, 4517 .bdrv_probe_device = cdrom_probe_device, 4518 .bdrv_parse_filename = cdrom_parse_filename, 4519 .bdrv_file_open = cdrom_open, 4520 .bdrv_close = raw_close, 4521 .bdrv_reopen_prepare = raw_reopen_prepare, 4522 .bdrv_reopen_commit = raw_reopen_commit, 4523 .bdrv_reopen_abort = raw_reopen_abort, 4524 .bdrv_co_create_opts = bdrv_co_create_opts_simple, 4525 .create_opts = &bdrv_create_opts_simple, 4526 .mutable_opts = mutable_opts, 4527 4528 .bdrv_co_preadv = raw_co_preadv, 4529 .bdrv_co_pwritev = raw_co_pwritev, 4530 .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 4531 .bdrv_refresh_limits = cdrom_refresh_limits, 4532 .bdrv_attach_aio_context = raw_aio_attach_aio_context, 4533 4534 .bdrv_co_truncate = raw_co_truncate, 4535 .bdrv_co_getlength = raw_co_getlength, 4536 .bdrv_co_get_allocated_file_size = raw_co_get_allocated_file_size, 4537 4538 /* removable device support */ 4539 .bdrv_co_is_inserted = cdrom_co_is_inserted, 4540 .bdrv_co_eject = cdrom_co_eject, 4541 .bdrv_co_lock_medium = cdrom_co_lock_medium, 4542 }; 4543 #endif /* __FreeBSD__ */ 4544 4545 #endif /* HAVE_HOST_BLOCK_DEVICE */ 4546 4547 static void bdrv_file_init(void) 4548 { 4549 /* 4550 * Register all the drivers. Note that order is important, the driver 4551 * registered last will get probed first. 4552 */ 4553 bdrv_register(&bdrv_file); 4554 #if defined(HAVE_HOST_BLOCK_DEVICE) 4555 bdrv_register(&bdrv_host_device); 4556 #ifdef __linux__ 4557 bdrv_register(&bdrv_host_cdrom); 4558 #endif 4559 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 4560 bdrv_register(&bdrv_host_cdrom); 4561 #endif 4562 #endif /* HAVE_HOST_BLOCK_DEVICE */ 4563 } 4564 4565 block_init(bdrv_file_init); 4566