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