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