hw/misc/allwinner-h3-dramc: enforce 64-bit multiply when calculating row mirror address
[qemu/ar7.git] / block / file-posix.c
blob7e19bbff5f67a7596795b726ac044a040b99c45e
1 /*
2 * Block driver for RAW files (posix)
4 * Copyright (c) 2006 Fabrice Bellard
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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.
25 #include "qemu/osdep.h"
26 #include "qemu-common.h"
27 #include "qapi/error.h"
28 #include "qemu/cutils.h"
29 #include "qemu/error-report.h"
30 #include "block/block_int.h"
31 #include "qemu/module.h"
32 #include "qemu/option.h"
33 #include "trace.h"
34 #include "block/thread-pool.h"
35 #include "qemu/iov.h"
36 #include "block/raw-aio.h"
37 #include "qapi/qmp/qdict.h"
38 #include "qapi/qmp/qstring.h"
40 #include "scsi/pr-manager.h"
41 #include "scsi/constants.h"
43 #if defined(__APPLE__) && (__MACH__)
44 #include <paths.h>
45 #include <sys/param.h>
46 #include <IOKit/IOKitLib.h>
47 #include <IOKit/IOBSD.h>
48 #include <IOKit/storage/IOMediaBSDClient.h>
49 #include <IOKit/storage/IOMedia.h>
50 #include <IOKit/storage/IOCDMedia.h>
51 //#include <IOKit/storage/IOCDTypes.h>
52 #include <IOKit/storage/IODVDMedia.h>
53 #include <CoreFoundation/CoreFoundation.h>
54 #endif
56 #ifdef __sun__
57 #define _POSIX_PTHREAD_SEMANTICS 1
58 #include <sys/dkio.h>
59 #endif
60 #ifdef __linux__
61 #include <sys/ioctl.h>
62 #include <sys/param.h>
63 #include <sys/syscall.h>
64 #include <linux/cdrom.h>
65 #include <linux/fd.h>
66 #include <linux/fs.h>
67 #include <linux/hdreg.h>
68 #include <scsi/sg.h>
69 #ifdef __s390__
70 #include <asm/dasd.h>
71 #endif
72 #ifndef FS_NOCOW_FL
73 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
74 #endif
75 #endif
76 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
77 #include <linux/falloc.h>
78 #endif
79 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
80 #include <sys/disk.h>
81 #include <sys/cdio.h>
82 #endif
84 #ifdef __OpenBSD__
85 #include <sys/ioctl.h>
86 #include <sys/disklabel.h>
87 #include <sys/dkio.h>
88 #endif
90 #ifdef __NetBSD__
91 #include <sys/ioctl.h>
92 #include <sys/disklabel.h>
93 #include <sys/dkio.h>
94 #include <sys/disk.h>
95 #endif
97 #ifdef __DragonFly__
98 #include <sys/ioctl.h>
99 #include <sys/diskslice.h>
100 #endif
102 #ifdef CONFIG_XFS
103 #include <xfs/xfs.h>
104 #endif
106 #include "trace.h"
108 /* OS X does not have O_DSYNC */
109 #ifndef O_DSYNC
110 #ifdef O_SYNC
111 #define O_DSYNC O_SYNC
112 #elif defined(O_FSYNC)
113 #define O_DSYNC O_FSYNC
114 #endif
115 #endif
117 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
118 #ifndef O_DIRECT
119 #define O_DIRECT O_DSYNC
120 #endif
122 #define FTYPE_FILE 0
123 #define FTYPE_CD 1
125 #define MAX_BLOCKSIZE 4096
127 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
128 * leaving a few more bytes for its future use. */
129 #define RAW_LOCK_PERM_BASE 100
130 #define RAW_LOCK_SHARED_BASE 200
132 typedef struct BDRVRawState {
133 int fd;
134 bool use_lock;
135 int type;
136 int open_flags;
137 size_t buf_align;
139 /* The current permissions. */
140 uint64_t perm;
141 uint64_t shared_perm;
143 /* The perms bits whose corresponding bytes are already locked in
144 * s->fd. */
145 uint64_t locked_perm;
146 uint64_t locked_shared_perm;
148 int perm_change_fd;
149 int perm_change_flags;
150 BDRVReopenState *reopen_state;
152 #ifdef CONFIG_XFS
153 bool is_xfs:1;
154 #endif
155 bool has_discard:1;
156 bool has_write_zeroes:1;
157 bool discard_zeroes:1;
158 bool use_linux_aio:1;
159 bool use_linux_io_uring:1;
160 bool page_cache_inconsistent:1;
161 bool has_fallocate;
162 bool needs_alignment;
163 bool drop_cache;
164 bool check_cache_dropped;
165 struct {
166 uint64_t discard_nb_ok;
167 uint64_t discard_nb_failed;
168 uint64_t discard_bytes_ok;
169 } stats;
171 PRManager *pr_mgr;
172 } BDRVRawState;
174 typedef struct BDRVRawReopenState {
175 int fd;
176 int open_flags;
177 bool drop_cache;
178 bool check_cache_dropped;
179 } BDRVRawReopenState;
181 static int fd_open(BlockDriverState *bs);
182 static int64_t raw_getlength(BlockDriverState *bs);
184 typedef struct RawPosixAIOData {
185 BlockDriverState *bs;
186 int aio_type;
187 int aio_fildes;
189 off_t aio_offset;
190 uint64_t aio_nbytes;
192 union {
193 struct {
194 struct iovec *iov;
195 int niov;
196 } io;
197 struct {
198 uint64_t cmd;
199 void *buf;
200 } ioctl;
201 struct {
202 int aio_fd2;
203 off_t aio_offset2;
204 } copy_range;
205 struct {
206 PreallocMode prealloc;
207 Error **errp;
208 } truncate;
210 } RawPosixAIOData;
212 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
213 static int cdrom_reopen(BlockDriverState *bs);
214 #endif
216 #if defined(__NetBSD__)
217 static int raw_normalize_devicepath(const char **filename, Error **errp)
219 static char namebuf[PATH_MAX];
220 const char *dp, *fname;
221 struct stat sb;
223 fname = *filename;
224 dp = strrchr(fname, '/');
225 if (lstat(fname, &sb) < 0) {
226 error_setg_file_open(errp, errno, fname);
227 return -errno;
230 if (!S_ISBLK(sb.st_mode)) {
231 return 0;
234 if (dp == NULL) {
235 snprintf(namebuf, PATH_MAX, "r%s", fname);
236 } else {
237 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
238 (int)(dp - fname), fname, dp + 1);
240 *filename = namebuf;
241 warn_report("%s is a block device, using %s", fname, *filename);
243 return 0;
245 #else
246 static int raw_normalize_devicepath(const char **filename, Error **errp)
248 return 0;
250 #endif
253 * Get logical block size via ioctl. On success store it in @sector_size_p.
255 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
257 unsigned int sector_size;
258 bool success = false;
259 int i;
261 errno = ENOTSUP;
262 static const unsigned long ioctl_list[] = {
263 #ifdef BLKSSZGET
264 BLKSSZGET,
265 #endif
266 #ifdef DKIOCGETBLOCKSIZE
267 DKIOCGETBLOCKSIZE,
268 #endif
269 #ifdef DIOCGSECTORSIZE
270 DIOCGSECTORSIZE,
271 #endif
274 /* Try a few ioctls to get the right size */
275 for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) {
276 if (ioctl(fd, ioctl_list[i], &sector_size) >= 0) {
277 *sector_size_p = sector_size;
278 success = true;
282 return success ? 0 : -errno;
286 * Get physical block size of @fd.
287 * On success, store it in @blk_size and return 0.
288 * On failure, return -errno.
290 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
292 #ifdef BLKPBSZGET
293 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
294 return -errno;
296 return 0;
297 #else
298 return -ENOTSUP;
299 #endif
302 /* Check if read is allowed with given memory buffer and length.
304 * This function is used to check O_DIRECT memory buffer and request alignment.
306 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
308 ssize_t ret = pread(fd, buf, len, 0);
310 if (ret >= 0) {
311 return true;
314 #ifdef __linux__
315 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
316 * other errors (e.g. real I/O error), which could happen on a failed
317 * drive, since we only care about probing alignment.
319 if (errno != EINVAL) {
320 return true;
322 #endif
324 return false;
327 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
329 BDRVRawState *s = bs->opaque;
330 char *buf;
331 size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size);
332 size_t alignments[] = {1, 512, 1024, 2048, 4096};
334 /* For SCSI generic devices the alignment is not really used.
335 With buffered I/O, we don't have any restrictions. */
336 if (bdrv_is_sg(bs) || !s->needs_alignment) {
337 bs->bl.request_alignment = 1;
338 s->buf_align = 1;
339 return;
342 bs->bl.request_alignment = 0;
343 s->buf_align = 0;
344 /* Let's try to use the logical blocksize for the alignment. */
345 if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) {
346 bs->bl.request_alignment = 0;
348 #ifdef CONFIG_XFS
349 if (s->is_xfs) {
350 struct dioattr da;
351 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
352 bs->bl.request_alignment = da.d_miniosz;
353 /* The kernel returns wrong information for d_mem */
354 /* s->buf_align = da.d_mem; */
357 #endif
360 * If we could not get the sizes so far, we can only guess them. First try
361 * to detect request alignment, since it is more likely to succeed. Then
362 * try to detect buf_align, which cannot be detected in some cases (e.g.
363 * Gluster). If buf_align cannot be detected, we fallback to the value of
364 * request_alignment.
367 if (!bs->bl.request_alignment) {
368 int i;
369 size_t align;
370 buf = qemu_memalign(max_align, max_align);
371 for (i = 0; i < ARRAY_SIZE(alignments); i++) {
372 align = alignments[i];
373 if (raw_is_io_aligned(fd, buf, align)) {
374 /* Fallback to safe value. */
375 bs->bl.request_alignment = (align != 1) ? align : max_align;
376 break;
379 qemu_vfree(buf);
382 if (!s->buf_align) {
383 int i;
384 size_t align;
385 buf = qemu_memalign(max_align, 2 * max_align);
386 for (i = 0; i < ARRAY_SIZE(alignments); i++) {
387 align = alignments[i];
388 if (raw_is_io_aligned(fd, buf + align, max_align)) {
389 /* Fallback to request_alignment. */
390 s->buf_align = (align != 1) ? align : bs->bl.request_alignment;
391 break;
394 qemu_vfree(buf);
397 if (!s->buf_align || !bs->bl.request_alignment) {
398 error_setg(errp, "Could not find working O_DIRECT alignment");
399 error_append_hint(errp, "Try cache.direct=off\n");
403 static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers)
405 bool read_write = false;
406 assert(open_flags != NULL);
408 *open_flags |= O_BINARY;
409 *open_flags &= ~O_ACCMODE;
411 if (bdrv_flags & BDRV_O_AUTO_RDONLY) {
412 read_write = has_writers;
413 } else if (bdrv_flags & BDRV_O_RDWR) {
414 read_write = true;
417 if (read_write) {
418 *open_flags |= O_RDWR;
419 } else {
420 *open_flags |= O_RDONLY;
423 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
424 * and O_DIRECT for no caching. */
425 if ((bdrv_flags & BDRV_O_NOCACHE)) {
426 *open_flags |= O_DIRECT;
430 static void raw_parse_filename(const char *filename, QDict *options,
431 Error **errp)
433 bdrv_parse_filename_strip_prefix(filename, "file:", options);
436 static QemuOptsList raw_runtime_opts = {
437 .name = "raw",
438 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
439 .desc = {
441 .name = "filename",
442 .type = QEMU_OPT_STRING,
443 .help = "File name of the image",
446 .name = "aio",
447 .type = QEMU_OPT_STRING,
448 .help = "host AIO implementation (threads, native, io_uring)",
451 .name = "locking",
452 .type = QEMU_OPT_STRING,
453 .help = "file locking mode (on/off/auto, default: auto)",
456 .name = "pr-manager",
457 .type = QEMU_OPT_STRING,
458 .help = "id of persistent reservation manager object (default: none)",
460 #if defined(__linux__)
462 .name = "drop-cache",
463 .type = QEMU_OPT_BOOL,
464 .help = "invalidate page cache during live migration (default: on)",
466 #endif
468 .name = "x-check-cache-dropped",
469 .type = QEMU_OPT_BOOL,
470 .help = "check that page cache was dropped on live migration (default: off)"
472 { /* end of list */ }
476 static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL };
478 static int raw_open_common(BlockDriverState *bs, QDict *options,
479 int bdrv_flags, int open_flags,
480 bool device, Error **errp)
482 BDRVRawState *s = bs->opaque;
483 QemuOpts *opts;
484 Error *local_err = NULL;
485 const char *filename = NULL;
486 const char *str;
487 BlockdevAioOptions aio, aio_default;
488 int fd, ret;
489 struct stat st;
490 OnOffAuto locking;
492 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
493 qemu_opts_absorb_qdict(opts, options, &local_err);
494 if (local_err) {
495 error_propagate(errp, local_err);
496 ret = -EINVAL;
497 goto fail;
500 filename = qemu_opt_get(opts, "filename");
502 ret = raw_normalize_devicepath(&filename, errp);
503 if (ret != 0) {
504 goto fail;
507 if (bdrv_flags & BDRV_O_NATIVE_AIO) {
508 aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE;
509 #ifdef CONFIG_LINUX_IO_URING
510 } else if (bdrv_flags & BDRV_O_IO_URING) {
511 aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING;
512 #endif
513 } else {
514 aio_default = BLOCKDEV_AIO_OPTIONS_THREADS;
517 aio = qapi_enum_parse(&BlockdevAioOptions_lookup,
518 qemu_opt_get(opts, "aio"),
519 aio_default, &local_err);
520 if (local_err) {
521 error_propagate(errp, local_err);
522 ret = -EINVAL;
523 goto fail;
526 s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE);
527 #ifdef CONFIG_LINUX_IO_URING
528 s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING);
529 #endif
531 locking = qapi_enum_parse(&OnOffAuto_lookup,
532 qemu_opt_get(opts, "locking"),
533 ON_OFF_AUTO_AUTO, &local_err);
534 if (local_err) {
535 error_propagate(errp, local_err);
536 ret = -EINVAL;
537 goto fail;
539 switch (locking) {
540 case ON_OFF_AUTO_ON:
541 s->use_lock = true;
542 if (!qemu_has_ofd_lock()) {
543 warn_report("File lock requested but OFD locking syscall is "
544 "unavailable, falling back to POSIX file locks");
545 error_printf("Due to the implementation, locks can be lost "
546 "unexpectedly.\n");
548 break;
549 case ON_OFF_AUTO_OFF:
550 s->use_lock = false;
551 break;
552 case ON_OFF_AUTO_AUTO:
553 s->use_lock = qemu_has_ofd_lock();
554 break;
555 default:
556 abort();
559 str = qemu_opt_get(opts, "pr-manager");
560 if (str) {
561 s->pr_mgr = pr_manager_lookup(str, &local_err);
562 if (local_err) {
563 error_propagate(errp, local_err);
564 ret = -EINVAL;
565 goto fail;
569 s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true);
570 s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped",
571 false);
573 s->open_flags = open_flags;
574 raw_parse_flags(bdrv_flags, &s->open_flags, false);
576 s->fd = -1;
577 fd = qemu_open(filename, s->open_flags, 0644);
578 ret = fd < 0 ? -errno : 0;
580 if (ret < 0) {
581 error_setg_file_open(errp, -ret, filename);
582 if (ret == -EROFS) {
583 ret = -EACCES;
585 goto fail;
587 s->fd = fd;
589 s->perm = 0;
590 s->shared_perm = BLK_PERM_ALL;
592 #ifdef CONFIG_LINUX_AIO
593 /* Currently Linux does AIO only for files opened with O_DIRECT */
594 if (s->use_linux_aio) {
595 if (!(s->open_flags & O_DIRECT)) {
596 error_setg(errp, "aio=native was specified, but it requires "
597 "cache.direct=on, which was not specified.");
598 ret = -EINVAL;
599 goto fail;
601 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) {
602 error_prepend(errp, "Unable to use native AIO: ");
603 goto fail;
606 #else
607 if (s->use_linux_aio) {
608 error_setg(errp, "aio=native was specified, but is not supported "
609 "in this build.");
610 ret = -EINVAL;
611 goto fail;
613 #endif /* !defined(CONFIG_LINUX_AIO) */
615 #ifdef CONFIG_LINUX_IO_URING
616 if (s->use_linux_io_uring) {
617 if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs), errp)) {
618 error_prepend(errp, "Unable to use io_uring: ");
619 goto fail;
622 #else
623 if (s->use_linux_io_uring) {
624 error_setg(errp, "aio=io_uring was specified, but is not supported "
625 "in this build.");
626 ret = -EINVAL;
627 goto fail;
629 #endif /* !defined(CONFIG_LINUX_IO_URING) */
631 s->has_discard = true;
632 s->has_write_zeroes = true;
633 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
634 s->needs_alignment = true;
637 if (fstat(s->fd, &st) < 0) {
638 ret = -errno;
639 error_setg_errno(errp, errno, "Could not stat file");
640 goto fail;
643 if (!device) {
644 if (S_ISBLK(st.st_mode)) {
645 warn_report("Opening a block device as a file using the '%s' "
646 "driver is deprecated", bs->drv->format_name);
647 } else if (S_ISCHR(st.st_mode)) {
648 warn_report("Opening a character device as a file using the '%s' "
649 "driver is deprecated", bs->drv->format_name);
650 } else if (!S_ISREG(st.st_mode)) {
651 error_setg(errp, "A regular file was expected by the '%s' driver, "
652 "but something else was given", bs->drv->format_name);
653 ret = -EINVAL;
654 goto fail;
655 } else {
656 s->discard_zeroes = true;
657 s->has_fallocate = true;
659 } else {
660 if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
661 error_setg(errp, "'%s' driver expects either "
662 "a character or block device", bs->drv->format_name);
663 ret = -EINVAL;
664 goto fail;
668 if (S_ISBLK(st.st_mode)) {
669 #ifdef BLKDISCARDZEROES
670 unsigned int arg;
671 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
672 s->discard_zeroes = true;
674 #endif
675 #ifdef __linux__
676 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
677 * not rely on the contents of discarded blocks unless using O_DIRECT.
678 * Same for BLKZEROOUT.
680 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
681 s->discard_zeroes = false;
682 s->has_write_zeroes = false;
684 #endif
686 #ifdef __FreeBSD__
687 if (S_ISCHR(st.st_mode)) {
689 * The file is a char device (disk), which on FreeBSD isn't behind
690 * a pager, so force all requests to be aligned. This is needed
691 * so QEMU makes sure all IO operations on the device are aligned
692 * to sector size, or else FreeBSD will reject them with EINVAL.
694 s->needs_alignment = true;
696 #endif
698 #ifdef CONFIG_XFS
699 if (platform_test_xfs_fd(s->fd)) {
700 s->is_xfs = true;
702 #endif
704 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK;
705 ret = 0;
706 fail:
707 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
708 unlink(filename);
710 qemu_opts_del(opts);
711 return ret;
714 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
715 Error **errp)
717 BDRVRawState *s = bs->opaque;
719 s->type = FTYPE_FILE;
720 return raw_open_common(bs, options, flags, 0, false, errp);
723 typedef enum {
724 RAW_PL_PREPARE,
725 RAW_PL_COMMIT,
726 RAW_PL_ABORT,
727 } RawPermLockOp;
729 #define PERM_FOREACH(i) \
730 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
732 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
733 * file; if @unlock == true, also unlock the unneeded bytes.
734 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
736 static int raw_apply_lock_bytes(BDRVRawState *s, int fd,
737 uint64_t perm_lock_bits,
738 uint64_t shared_perm_lock_bits,
739 bool unlock, Error **errp)
741 int ret;
742 int i;
743 uint64_t locked_perm, locked_shared_perm;
745 if (s) {
746 locked_perm = s->locked_perm;
747 locked_shared_perm = s->locked_shared_perm;
748 } else {
750 * We don't have the previous bits, just lock/unlock for each of the
751 * requested bits.
753 if (unlock) {
754 locked_perm = BLK_PERM_ALL;
755 locked_shared_perm = BLK_PERM_ALL;
756 } else {
757 locked_perm = 0;
758 locked_shared_perm = 0;
762 PERM_FOREACH(i) {
763 int off = RAW_LOCK_PERM_BASE + i;
764 uint64_t bit = (1ULL << i);
765 if ((perm_lock_bits & bit) && !(locked_perm & bit)) {
766 ret = qemu_lock_fd(fd, off, 1, false);
767 if (ret) {
768 error_setg(errp, "Failed to lock byte %d", off);
769 return ret;
770 } else if (s) {
771 s->locked_perm |= bit;
773 } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) {
774 ret = qemu_unlock_fd(fd, off, 1);
775 if (ret) {
776 error_setg(errp, "Failed to unlock byte %d", off);
777 return ret;
778 } else if (s) {
779 s->locked_perm &= ~bit;
783 PERM_FOREACH(i) {
784 int off = RAW_LOCK_SHARED_BASE + i;
785 uint64_t bit = (1ULL << i);
786 if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) {
787 ret = qemu_lock_fd(fd, off, 1, false);
788 if (ret) {
789 error_setg(errp, "Failed to lock byte %d", off);
790 return ret;
791 } else if (s) {
792 s->locked_shared_perm |= bit;
794 } else if (unlock && (locked_shared_perm & bit) &&
795 !(shared_perm_lock_bits & bit)) {
796 ret = qemu_unlock_fd(fd, off, 1);
797 if (ret) {
798 error_setg(errp, "Failed to unlock byte %d", off);
799 return ret;
800 } else if (s) {
801 s->locked_shared_perm &= ~bit;
805 return 0;
808 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
809 static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm,
810 Error **errp)
812 int ret;
813 int i;
815 PERM_FOREACH(i) {
816 int off = RAW_LOCK_SHARED_BASE + i;
817 uint64_t p = 1ULL << i;
818 if (perm & p) {
819 ret = qemu_lock_fd_test(fd, off, 1, true);
820 if (ret) {
821 char *perm_name = bdrv_perm_names(p);
822 error_setg(errp,
823 "Failed to get \"%s\" lock",
824 perm_name);
825 g_free(perm_name);
826 return ret;
830 PERM_FOREACH(i) {
831 int off = RAW_LOCK_PERM_BASE + i;
832 uint64_t p = 1ULL << i;
833 if (!(shared_perm & p)) {
834 ret = qemu_lock_fd_test(fd, off, 1, true);
835 if (ret) {
836 char *perm_name = bdrv_perm_names(p);
837 error_setg(errp,
838 "Failed to get shared \"%s\" lock",
839 perm_name);
840 g_free(perm_name);
841 return ret;
845 return 0;
848 static int raw_handle_perm_lock(BlockDriverState *bs,
849 RawPermLockOp op,
850 uint64_t new_perm, uint64_t new_shared,
851 Error **errp)
853 BDRVRawState *s = bs->opaque;
854 int ret = 0;
855 Error *local_err = NULL;
857 if (!s->use_lock) {
858 return 0;
861 if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) {
862 return 0;
865 switch (op) {
866 case RAW_PL_PREPARE:
867 if ((s->perm | new_perm) == s->perm &&
868 (s->shared_perm & new_shared) == s->shared_perm)
871 * We are going to unlock bytes, it should not fail. If it fail due
872 * to some fs-dependent permission-unrelated reasons (which occurs
873 * sometimes on NFS and leads to abort in bdrv_replace_child) we
874 * can't prevent such errors by any check here. And we ignore them
875 * anyway in ABORT and COMMIT.
877 return 0;
879 ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm,
880 ~s->shared_perm | ~new_shared,
881 false, errp);
882 if (!ret) {
883 ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp);
884 if (!ret) {
885 return 0;
887 error_append_hint(errp,
888 "Is another process using the image [%s]?\n",
889 bs->filename);
891 /* fall through to unlock bytes. */
892 case RAW_PL_ABORT:
893 raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm,
894 true, &local_err);
895 if (local_err) {
896 /* Theoretically the above call only unlocks bytes and it cannot
897 * fail. Something weird happened, report it.
899 warn_report_err(local_err);
901 break;
902 case RAW_PL_COMMIT:
903 raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared,
904 true, &local_err);
905 if (local_err) {
906 /* Theoretically the above call only unlocks bytes and it cannot
907 * fail. Something weird happened, report it.
909 warn_report_err(local_err);
911 break;
913 return ret;
916 static int raw_reconfigure_getfd(BlockDriverState *bs, int flags,
917 int *open_flags, uint64_t perm, bool force_dup,
918 Error **errp)
920 BDRVRawState *s = bs->opaque;
921 int fd = -1;
922 int ret;
923 bool has_writers = perm &
924 (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE);
925 int fcntl_flags = O_APPEND | O_NONBLOCK;
926 #ifdef O_NOATIME
927 fcntl_flags |= O_NOATIME;
928 #endif
930 *open_flags = 0;
931 if (s->type == FTYPE_CD) {
932 *open_flags |= O_NONBLOCK;
935 raw_parse_flags(flags, open_flags, has_writers);
937 #ifdef O_ASYNC
938 /* Not all operating systems have O_ASYNC, and those that don't
939 * will not let us track the state into rs->open_flags (typically
940 * you achieve the same effect with an ioctl, for example I_SETSIG
941 * on Solaris). But we do not use O_ASYNC, so that's fine.
943 assert((s->open_flags & O_ASYNC) == 0);
944 #endif
946 if (!force_dup && *open_flags == s->open_flags) {
947 /* We're lucky, the existing fd is fine */
948 return s->fd;
951 if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
952 /* dup the original fd */
953 fd = qemu_dup(s->fd);
954 if (fd >= 0) {
955 ret = fcntl_setfl(fd, *open_flags);
956 if (ret) {
957 qemu_close(fd);
958 fd = -1;
963 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
964 if (fd == -1) {
965 const char *normalized_filename = bs->filename;
966 ret = raw_normalize_devicepath(&normalized_filename, errp);
967 if (ret >= 0) {
968 assert(!(*open_flags & O_CREAT));
969 fd = qemu_open(normalized_filename, *open_flags);
970 if (fd == -1) {
971 error_setg_errno(errp, errno, "Could not reopen file");
972 return -1;
977 return fd;
980 static int raw_reopen_prepare(BDRVReopenState *state,
981 BlockReopenQueue *queue, Error **errp)
983 BDRVRawState *s;
984 BDRVRawReopenState *rs;
985 QemuOpts *opts;
986 int ret;
987 Error *local_err = NULL;
989 assert(state != NULL);
990 assert(state->bs != NULL);
992 s = state->bs->opaque;
994 state->opaque = g_new0(BDRVRawReopenState, 1);
995 rs = state->opaque;
997 /* Handle options changes */
998 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
999 qemu_opts_absorb_qdict(opts, state->options, &local_err);
1000 if (local_err) {
1001 error_propagate(errp, local_err);
1002 ret = -EINVAL;
1003 goto out;
1006 rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true);
1007 rs->check_cache_dropped =
1008 qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false);
1010 /* This driver's reopen function doesn't currently allow changing
1011 * other options, so let's put them back in the original QDict and
1012 * bdrv_reopen_prepare() will detect changes and complain. */
1013 qemu_opts_to_qdict(opts, state->options);
1015 rs->fd = raw_reconfigure_getfd(state->bs, state->flags, &rs->open_flags,
1016 state->perm, true, &local_err);
1017 if (local_err) {
1018 error_propagate(errp, local_err);
1019 ret = -1;
1020 goto out;
1023 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
1024 * alignment with the new fd. */
1025 if (rs->fd != -1) {
1026 raw_probe_alignment(state->bs, rs->fd, &local_err);
1027 if (local_err) {
1028 error_propagate(errp, local_err);
1029 ret = -EINVAL;
1030 goto out_fd;
1034 s->reopen_state = state;
1035 ret = 0;
1036 out_fd:
1037 if (ret < 0) {
1038 qemu_close(rs->fd);
1039 rs->fd = -1;
1041 out:
1042 qemu_opts_del(opts);
1043 return ret;
1046 static void raw_reopen_commit(BDRVReopenState *state)
1048 BDRVRawReopenState *rs = state->opaque;
1049 BDRVRawState *s = state->bs->opaque;
1051 s->drop_cache = rs->drop_cache;
1052 s->check_cache_dropped = rs->check_cache_dropped;
1053 s->open_flags = rs->open_flags;
1055 qemu_close(s->fd);
1056 s->fd = rs->fd;
1058 g_free(state->opaque);
1059 state->opaque = NULL;
1061 assert(s->reopen_state == state);
1062 s->reopen_state = NULL;
1066 static void raw_reopen_abort(BDRVReopenState *state)
1068 BDRVRawReopenState *rs = state->opaque;
1069 BDRVRawState *s = state->bs->opaque;
1071 /* nothing to do if NULL, we didn't get far enough */
1072 if (rs == NULL) {
1073 return;
1076 if (rs->fd >= 0) {
1077 qemu_close(rs->fd);
1078 rs->fd = -1;
1080 g_free(state->opaque);
1081 state->opaque = NULL;
1083 assert(s->reopen_state == state);
1084 s->reopen_state = NULL;
1087 static int sg_get_max_transfer_length(int fd)
1089 #ifdef BLKSECTGET
1090 int max_bytes = 0;
1092 if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) {
1093 return max_bytes;
1094 } else {
1095 return -errno;
1097 #else
1098 return -ENOSYS;
1099 #endif
1102 static int sg_get_max_segments(int fd)
1104 #ifdef CONFIG_LINUX
1105 char buf[32];
1106 const char *end;
1107 char *sysfspath = NULL;
1108 int ret;
1109 int sysfd = -1;
1110 long max_segments;
1111 struct stat st;
1113 if (fstat(fd, &st)) {
1114 ret = -errno;
1115 goto out;
1118 sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1119 major(st.st_rdev), minor(st.st_rdev));
1120 sysfd = open(sysfspath, O_RDONLY);
1121 if (sysfd == -1) {
1122 ret = -errno;
1123 goto out;
1125 do {
1126 ret = read(sysfd, buf, sizeof(buf) - 1);
1127 } while (ret == -1 && errno == EINTR);
1128 if (ret < 0) {
1129 ret = -errno;
1130 goto out;
1131 } else if (ret == 0) {
1132 ret = -EIO;
1133 goto out;
1135 buf[ret] = 0;
1136 /* The file is ended with '\n', pass 'end' to accept that. */
1137 ret = qemu_strtol(buf, &end, 10, &max_segments);
1138 if (ret == 0 && end && *end == '\n') {
1139 ret = max_segments;
1142 out:
1143 if (sysfd != -1) {
1144 close(sysfd);
1146 g_free(sysfspath);
1147 return ret;
1148 #else
1149 return -ENOTSUP;
1150 #endif
1153 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
1155 BDRVRawState *s = bs->opaque;
1157 if (bs->sg) {
1158 int ret = sg_get_max_transfer_length(s->fd);
1160 if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) {
1161 bs->bl.max_transfer = pow2floor(ret);
1164 ret = sg_get_max_segments(s->fd);
1165 if (ret > 0) {
1166 bs->bl.max_transfer = MIN(bs->bl.max_transfer,
1167 ret * qemu_real_host_page_size);
1171 raw_probe_alignment(bs, s->fd, errp);
1172 bs->bl.min_mem_alignment = s->buf_align;
1173 bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size);
1176 static int check_for_dasd(int fd)
1178 #ifdef BIODASDINFO2
1179 struct dasd_information2_t info = {0};
1181 return ioctl(fd, BIODASDINFO2, &info);
1182 #else
1183 return -1;
1184 #endif
1188 * Try to get @bs's logical and physical block size.
1189 * On success, store them in @bsz and return zero.
1190 * On failure, return negative errno.
1192 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
1194 BDRVRawState *s = bs->opaque;
1195 int ret;
1197 /* If DASD, get blocksizes */
1198 if (check_for_dasd(s->fd) < 0) {
1199 return -ENOTSUP;
1201 ret = probe_logical_blocksize(s->fd, &bsz->log);
1202 if (ret < 0) {
1203 return ret;
1205 return probe_physical_blocksize(s->fd, &bsz->phys);
1209 * Try to get @bs's geometry: cyls, heads, sectors.
1210 * On success, store them in @geo and return 0.
1211 * On failure return -errno.
1212 * (Allows block driver to assign default geometry values that guest sees)
1214 #ifdef __linux__
1215 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1217 BDRVRawState *s = bs->opaque;
1218 struct hd_geometry ioctl_geo = {0};
1220 /* If DASD, get its geometry */
1221 if (check_for_dasd(s->fd) < 0) {
1222 return -ENOTSUP;
1224 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
1225 return -errno;
1227 /* HDIO_GETGEO may return success even though geo contains zeros
1228 (e.g. certain multipath setups) */
1229 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
1230 return -ENOTSUP;
1232 /* Do not return a geometry for partition */
1233 if (ioctl_geo.start != 0) {
1234 return -ENOTSUP;
1236 geo->heads = ioctl_geo.heads;
1237 geo->sectors = ioctl_geo.sectors;
1238 geo->cylinders = ioctl_geo.cylinders;
1240 return 0;
1242 #else /* __linux__ */
1243 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1245 return -ENOTSUP;
1247 #endif
1249 #if defined(__linux__)
1250 static int handle_aiocb_ioctl(void *opaque)
1252 RawPosixAIOData *aiocb = opaque;
1253 int ret;
1255 ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf);
1256 if (ret == -1) {
1257 return -errno;
1260 return 0;
1262 #endif /* linux */
1264 static int handle_aiocb_flush(void *opaque)
1266 RawPosixAIOData *aiocb = opaque;
1267 BDRVRawState *s = aiocb->bs->opaque;
1268 int ret;
1270 if (s->page_cache_inconsistent) {
1271 return -EIO;
1274 ret = qemu_fdatasync(aiocb->aio_fildes);
1275 if (ret == -1) {
1276 /* There is no clear definition of the semantics of a failing fsync(),
1277 * so we may have to assume the worst. The sad truth is that this
1278 * assumption is correct for Linux. Some pages are now probably marked
1279 * clean in the page cache even though they are inconsistent with the
1280 * on-disk contents. The next fdatasync() call would succeed, but no
1281 * further writeback attempt will be made. We can't get back to a state
1282 * in which we know what is on disk (we would have to rewrite
1283 * everything that was touched since the last fdatasync() at least), so
1284 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1285 * really defined, I have little hope that other OSes are doing better.
1287 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1288 * cache. */
1289 if ((s->open_flags & O_DIRECT) == 0) {
1290 s->page_cache_inconsistent = true;
1292 return -errno;
1294 return 0;
1297 #ifdef CONFIG_PREADV
1299 static bool preadv_present = true;
1301 static ssize_t
1302 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1304 return preadv(fd, iov, nr_iov, offset);
1307 static ssize_t
1308 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1310 return pwritev(fd, iov, nr_iov, offset);
1313 #else
1315 static bool preadv_present = false;
1317 static ssize_t
1318 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1320 return -ENOSYS;
1323 static ssize_t
1324 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1326 return -ENOSYS;
1329 #endif
1331 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
1333 ssize_t len;
1335 do {
1336 if (aiocb->aio_type & QEMU_AIO_WRITE)
1337 len = qemu_pwritev(aiocb->aio_fildes,
1338 aiocb->io.iov,
1339 aiocb->io.niov,
1340 aiocb->aio_offset);
1341 else
1342 len = qemu_preadv(aiocb->aio_fildes,
1343 aiocb->io.iov,
1344 aiocb->io.niov,
1345 aiocb->aio_offset);
1346 } while (len == -1 && errno == EINTR);
1348 if (len == -1) {
1349 return -errno;
1351 return len;
1355 * Read/writes the data to/from a given linear buffer.
1357 * Returns the number of bytes handles or -errno in case of an error. Short
1358 * reads are only returned if the end of the file is reached.
1360 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
1362 ssize_t offset = 0;
1363 ssize_t len;
1365 while (offset < aiocb->aio_nbytes) {
1366 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1367 len = pwrite(aiocb->aio_fildes,
1368 (const char *)buf + offset,
1369 aiocb->aio_nbytes - offset,
1370 aiocb->aio_offset + offset);
1371 } else {
1372 len = pread(aiocb->aio_fildes,
1373 buf + offset,
1374 aiocb->aio_nbytes - offset,
1375 aiocb->aio_offset + offset);
1377 if (len == -1 && errno == EINTR) {
1378 continue;
1379 } else if (len == -1 && errno == EINVAL &&
1380 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
1381 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
1382 offset > 0) {
1383 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1384 * after a short read. Assume that O_DIRECT short reads only occur
1385 * at EOF. Therefore this is a short read, not an I/O error.
1387 break;
1388 } else if (len == -1) {
1389 offset = -errno;
1390 break;
1391 } else if (len == 0) {
1392 break;
1394 offset += len;
1397 return offset;
1400 static int handle_aiocb_rw(void *opaque)
1402 RawPosixAIOData *aiocb = opaque;
1403 ssize_t nbytes;
1404 char *buf;
1406 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
1408 * If there is just a single buffer, and it is properly aligned
1409 * we can just use plain pread/pwrite without any problems.
1411 if (aiocb->io.niov == 1) {
1412 nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base);
1413 goto out;
1416 * We have more than one iovec, and all are properly aligned.
1418 * Try preadv/pwritev first and fall back to linearizing the
1419 * buffer if it's not supported.
1421 if (preadv_present) {
1422 nbytes = handle_aiocb_rw_vector(aiocb);
1423 if (nbytes == aiocb->aio_nbytes ||
1424 (nbytes < 0 && nbytes != -ENOSYS)) {
1425 goto out;
1427 preadv_present = false;
1431 * XXX(hch): short read/write. no easy way to handle the reminder
1432 * using these interfaces. For now retry using plain
1433 * pread/pwrite?
1438 * Ok, we have to do it the hard way, copy all segments into
1439 * a single aligned buffer.
1441 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
1442 if (buf == NULL) {
1443 nbytes = -ENOMEM;
1444 goto out;
1447 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1448 char *p = buf;
1449 int i;
1451 for (i = 0; i < aiocb->io.niov; ++i) {
1452 memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len);
1453 p += aiocb->io.iov[i].iov_len;
1455 assert(p - buf == aiocb->aio_nbytes);
1458 nbytes = handle_aiocb_rw_linear(aiocb, buf);
1459 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
1460 char *p = buf;
1461 size_t count = aiocb->aio_nbytes, copy;
1462 int i;
1464 for (i = 0; i < aiocb->io.niov && count; ++i) {
1465 copy = count;
1466 if (copy > aiocb->io.iov[i].iov_len) {
1467 copy = aiocb->io.iov[i].iov_len;
1469 memcpy(aiocb->io.iov[i].iov_base, p, copy);
1470 assert(count >= copy);
1471 p += copy;
1472 count -= copy;
1474 assert(count == 0);
1476 qemu_vfree(buf);
1478 out:
1479 if (nbytes == aiocb->aio_nbytes) {
1480 return 0;
1481 } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) {
1482 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1483 return -EINVAL;
1484 } else {
1485 iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes,
1486 0, aiocb->aio_nbytes - nbytes);
1487 return 0;
1489 } else {
1490 assert(nbytes < 0);
1491 return nbytes;
1495 static int translate_err(int err)
1497 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1498 err == -ENOTTY) {
1499 err = -ENOTSUP;
1501 return err;
1504 #ifdef CONFIG_FALLOCATE
1505 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1507 do {
1508 if (fallocate(fd, mode, offset, len) == 0) {
1509 return 0;
1511 } while (errno == EINTR);
1512 return translate_err(-errno);
1514 #endif
1516 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1518 int ret = -ENOTSUP;
1519 BDRVRawState *s = aiocb->bs->opaque;
1521 if (!s->has_write_zeroes) {
1522 return -ENOTSUP;
1525 #ifdef BLKZEROOUT
1526 /* The BLKZEROOUT implementation in the kernel doesn't set
1527 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1528 * fallbacks. */
1529 if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) {
1530 do {
1531 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1532 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1533 return 0;
1535 } while (errno == EINTR);
1537 ret = translate_err(-errno);
1538 if (ret == -ENOTSUP) {
1539 s->has_write_zeroes = false;
1542 #endif
1544 return ret;
1547 static int handle_aiocb_write_zeroes(void *opaque)
1549 RawPosixAIOData *aiocb = opaque;
1550 #ifdef CONFIG_FALLOCATE
1551 BDRVRawState *s = aiocb->bs->opaque;
1552 int64_t len;
1553 #endif
1555 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1556 return handle_aiocb_write_zeroes_block(aiocb);
1559 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1560 if (s->has_write_zeroes) {
1561 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1562 aiocb->aio_offset, aiocb->aio_nbytes);
1563 if (ret == -EINVAL) {
1565 * Allow falling back to pwrite for file systems that
1566 * do not support fallocate() for an unaligned byte range.
1568 return -ENOTSUP;
1570 if (ret == 0 || ret != -ENOTSUP) {
1571 return ret;
1573 s->has_write_zeroes = false;
1575 #endif
1577 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1578 if (s->has_discard && s->has_fallocate) {
1579 int ret = do_fallocate(s->fd,
1580 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1581 aiocb->aio_offset, aiocb->aio_nbytes);
1582 if (ret == 0) {
1583 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1584 if (ret == 0 || ret != -ENOTSUP) {
1585 return ret;
1587 s->has_fallocate = false;
1588 } else if (ret != -ENOTSUP) {
1589 return ret;
1590 } else {
1591 s->has_discard = false;
1594 #endif
1596 #ifdef CONFIG_FALLOCATE
1597 /* Last resort: we are trying to extend the file with zeroed data. This
1598 * can be done via fallocate(fd, 0) */
1599 len = bdrv_getlength(aiocb->bs);
1600 if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) {
1601 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1602 if (ret == 0 || ret != -ENOTSUP) {
1603 return ret;
1605 s->has_fallocate = false;
1607 #endif
1609 return -ENOTSUP;
1612 static int handle_aiocb_write_zeroes_unmap(void *opaque)
1614 RawPosixAIOData *aiocb = opaque;
1615 BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque;
1616 int ret;
1618 /* First try to write zeros and unmap at the same time */
1620 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1621 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1622 aiocb->aio_offset, aiocb->aio_nbytes);
1623 if (ret != -ENOTSUP) {
1624 return ret;
1626 #endif
1628 /* If we couldn't manage to unmap while guaranteed that the area reads as
1629 * all-zero afterwards, just write zeroes without unmapping */
1630 ret = handle_aiocb_write_zeroes(aiocb);
1631 return ret;
1634 #ifndef HAVE_COPY_FILE_RANGE
1635 static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd,
1636 off_t *out_off, size_t len, unsigned int flags)
1638 #ifdef __NR_copy_file_range
1639 return syscall(__NR_copy_file_range, in_fd, in_off, out_fd,
1640 out_off, len, flags);
1641 #else
1642 errno = ENOSYS;
1643 return -1;
1644 #endif
1646 #endif
1648 static int handle_aiocb_copy_range(void *opaque)
1650 RawPosixAIOData *aiocb = opaque;
1651 uint64_t bytes = aiocb->aio_nbytes;
1652 off_t in_off = aiocb->aio_offset;
1653 off_t out_off = aiocb->copy_range.aio_offset2;
1655 while (bytes) {
1656 ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off,
1657 aiocb->copy_range.aio_fd2, &out_off,
1658 bytes, 0);
1659 trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off,
1660 aiocb->copy_range.aio_fd2, out_off, bytes,
1661 0, ret);
1662 if (ret == 0) {
1663 /* No progress (e.g. when beyond EOF), let the caller fall back to
1664 * buffer I/O. */
1665 return -ENOSPC;
1667 if (ret < 0) {
1668 switch (errno) {
1669 case ENOSYS:
1670 return -ENOTSUP;
1671 case EINTR:
1672 continue;
1673 default:
1674 return -errno;
1677 bytes -= ret;
1679 return 0;
1682 static int handle_aiocb_discard(void *opaque)
1684 RawPosixAIOData *aiocb = opaque;
1685 int ret = -EOPNOTSUPP;
1686 BDRVRawState *s = aiocb->bs->opaque;
1688 if (!s->has_discard) {
1689 return -ENOTSUP;
1692 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1693 #ifdef BLKDISCARD
1694 do {
1695 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1696 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1697 return 0;
1699 } while (errno == EINTR);
1701 ret = -errno;
1702 #endif
1703 } else {
1704 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1705 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1706 aiocb->aio_offset, aiocb->aio_nbytes);
1707 #endif
1710 ret = translate_err(ret);
1711 if (ret == -ENOTSUP) {
1712 s->has_discard = false;
1714 return ret;
1718 * Help alignment probing by allocating the first block.
1720 * When reading with direct I/O from unallocated area on Gluster backed by XFS,
1721 * reading succeeds regardless of request length. In this case we fallback to
1722 * safe alignment which is not optimal. Allocating the first block avoids this
1723 * fallback.
1725 * fd may be opened with O_DIRECT, but we don't know the buffer alignment or
1726 * request alignment, so we use safe values.
1728 * Returns: 0 on success, -errno on failure. Since this is an optimization,
1729 * caller may ignore failures.
1731 static int allocate_first_block(int fd, size_t max_size)
1733 size_t write_size = (max_size < MAX_BLOCKSIZE)
1734 ? BDRV_SECTOR_SIZE
1735 : MAX_BLOCKSIZE;
1736 size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size);
1737 void *buf;
1738 ssize_t n;
1739 int ret;
1741 buf = qemu_memalign(max_align, write_size);
1742 memset(buf, 0, write_size);
1744 do {
1745 n = pwrite(fd, buf, write_size, 0);
1746 } while (n == -1 && errno == EINTR);
1748 ret = (n == -1) ? -errno : 0;
1750 qemu_vfree(buf);
1751 return ret;
1754 static int handle_aiocb_truncate(void *opaque)
1756 RawPosixAIOData *aiocb = opaque;
1757 int result = 0;
1758 int64_t current_length = 0;
1759 char *buf = NULL;
1760 struct stat st;
1761 int fd = aiocb->aio_fildes;
1762 int64_t offset = aiocb->aio_offset;
1763 PreallocMode prealloc = aiocb->truncate.prealloc;
1764 Error **errp = aiocb->truncate.errp;
1766 if (fstat(fd, &st) < 0) {
1767 result = -errno;
1768 error_setg_errno(errp, -result, "Could not stat file");
1769 return result;
1772 current_length = st.st_size;
1773 if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1774 error_setg(errp, "Cannot use preallocation for shrinking files");
1775 return -ENOTSUP;
1778 switch (prealloc) {
1779 #ifdef CONFIG_POSIX_FALLOCATE
1780 case PREALLOC_MODE_FALLOC:
1782 * Truncating before posix_fallocate() makes it about twice slower on
1783 * file systems that do not support fallocate(), trying to check if a
1784 * block is allocated before allocating it, so don't do that here.
1786 if (offset != current_length) {
1787 result = -posix_fallocate(fd, current_length,
1788 offset - current_length);
1789 if (result != 0) {
1790 /* posix_fallocate() doesn't set errno. */
1791 error_setg_errno(errp, -result,
1792 "Could not preallocate new data");
1793 } else if (current_length == 0) {
1795 * posix_fallocate() uses fallocate() if the filesystem
1796 * supports it, or fallback to manually writing zeroes. If
1797 * fallocate() was used, unaligned reads from the fallocated
1798 * area in raw_probe_alignment() will succeed, hence we need to
1799 * allocate the first block.
1801 * Optimize future alignment probing; ignore failures.
1803 allocate_first_block(fd, offset);
1805 } else {
1806 result = 0;
1808 goto out;
1809 #endif
1810 case PREALLOC_MODE_FULL:
1812 int64_t num = 0, left = offset - current_length;
1813 off_t seek_result;
1816 * Knowing the final size from the beginning could allow the file
1817 * system driver to do less allocations and possibly avoid
1818 * fragmentation of the file.
1820 if (ftruncate(fd, offset) != 0) {
1821 result = -errno;
1822 error_setg_errno(errp, -result, "Could not resize file");
1823 goto out;
1826 buf = g_malloc0(65536);
1828 seek_result = lseek(fd, current_length, SEEK_SET);
1829 if (seek_result < 0) {
1830 result = -errno;
1831 error_setg_errno(errp, -result,
1832 "Failed to seek to the old end of file");
1833 goto out;
1836 while (left > 0) {
1837 num = MIN(left, 65536);
1838 result = write(fd, buf, num);
1839 if (result < 0) {
1840 if (errno == EINTR) {
1841 continue;
1843 result = -errno;
1844 error_setg_errno(errp, -result,
1845 "Could not write zeros for preallocation");
1846 goto out;
1848 left -= result;
1850 if (result >= 0) {
1851 result = fsync(fd);
1852 if (result < 0) {
1853 result = -errno;
1854 error_setg_errno(errp, -result,
1855 "Could not flush file to disk");
1856 goto out;
1859 goto out;
1861 case PREALLOC_MODE_OFF:
1862 if (ftruncate(fd, offset) != 0) {
1863 result = -errno;
1864 error_setg_errno(errp, -result, "Could not resize file");
1865 } else if (current_length == 0 && offset > current_length) {
1866 /* Optimize future alignment probing; ignore failures. */
1867 allocate_first_block(fd, offset);
1869 return result;
1870 default:
1871 result = -ENOTSUP;
1872 error_setg(errp, "Unsupported preallocation mode: %s",
1873 PreallocMode_str(prealloc));
1874 return result;
1877 out:
1878 if (result < 0) {
1879 if (ftruncate(fd, current_length) < 0) {
1880 error_report("Failed to restore old file length: %s",
1881 strerror(errno));
1885 g_free(buf);
1886 return result;
1889 static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs,
1890 ThreadPoolFunc func, void *arg)
1892 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1893 ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1894 return thread_pool_submit_co(pool, func, arg);
1897 static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset,
1898 uint64_t bytes, QEMUIOVector *qiov, int type)
1900 BDRVRawState *s = bs->opaque;
1901 RawPosixAIOData acb;
1903 if (fd_open(bs) < 0)
1904 return -EIO;
1907 * When using O_DIRECT, the request must be aligned to be able to use
1908 * either libaio or io_uring interface. If not fail back to regular thread
1909 * pool read/write code which emulates this for us if we
1910 * set QEMU_AIO_MISALIGNED.
1912 if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) {
1913 type |= QEMU_AIO_MISALIGNED;
1914 #ifdef CONFIG_LINUX_IO_URING
1915 } else if (s->use_linux_io_uring) {
1916 LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs));
1917 assert(qiov->size == bytes);
1918 return luring_co_submit(bs, aio, s->fd, offset, qiov, type);
1919 #endif
1920 #ifdef CONFIG_LINUX_AIO
1921 } else if (s->use_linux_aio) {
1922 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1923 assert(qiov->size == bytes);
1924 return laio_co_submit(bs, aio, s->fd, offset, qiov, type);
1925 #endif
1928 acb = (RawPosixAIOData) {
1929 .bs = bs,
1930 .aio_fildes = s->fd,
1931 .aio_type = type,
1932 .aio_offset = offset,
1933 .aio_nbytes = bytes,
1934 .io = {
1935 .iov = qiov->iov,
1936 .niov = qiov->niov,
1940 assert(qiov->size == bytes);
1941 return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb);
1944 static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset,
1945 uint64_t bytes, QEMUIOVector *qiov,
1946 int flags)
1948 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ);
1951 static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
1952 uint64_t bytes, QEMUIOVector *qiov,
1953 int flags)
1955 assert(flags == 0);
1956 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE);
1959 static void raw_aio_plug(BlockDriverState *bs)
1961 BDRVRawState __attribute__((unused)) *s = bs->opaque;
1962 #ifdef CONFIG_LINUX_AIO
1963 if (s->use_linux_aio) {
1964 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1965 laio_io_plug(bs, aio);
1967 #endif
1968 #ifdef CONFIG_LINUX_IO_URING
1969 if (s->use_linux_io_uring) {
1970 LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs));
1971 luring_io_plug(bs, aio);
1973 #endif
1976 static void raw_aio_unplug(BlockDriverState *bs)
1978 BDRVRawState __attribute__((unused)) *s = bs->opaque;
1979 #ifdef CONFIG_LINUX_AIO
1980 if (s->use_linux_aio) {
1981 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1982 laio_io_unplug(bs, aio);
1984 #endif
1985 #ifdef CONFIG_LINUX_IO_URING
1986 if (s->use_linux_io_uring) {
1987 LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs));
1988 luring_io_unplug(bs, aio);
1990 #endif
1993 static int raw_co_flush_to_disk(BlockDriverState *bs)
1995 BDRVRawState *s = bs->opaque;
1996 RawPosixAIOData acb;
1997 int ret;
1999 ret = fd_open(bs);
2000 if (ret < 0) {
2001 return ret;
2004 acb = (RawPosixAIOData) {
2005 .bs = bs,
2006 .aio_fildes = s->fd,
2007 .aio_type = QEMU_AIO_FLUSH,
2010 #ifdef CONFIG_LINUX_IO_URING
2011 if (s->use_linux_io_uring) {
2012 LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs));
2013 return luring_co_submit(bs, aio, s->fd, 0, NULL, QEMU_AIO_FLUSH);
2015 #endif
2016 return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb);
2019 static void raw_aio_attach_aio_context(BlockDriverState *bs,
2020 AioContext *new_context)
2022 BDRVRawState __attribute__((unused)) *s = bs->opaque;
2023 #ifdef CONFIG_LINUX_AIO
2024 if (s->use_linux_aio) {
2025 Error *local_err = NULL;
2026 if (!aio_setup_linux_aio(new_context, &local_err)) {
2027 error_reportf_err(local_err, "Unable to use native AIO, "
2028 "falling back to thread pool: ");
2029 s->use_linux_aio = false;
2032 #endif
2033 #ifdef CONFIG_LINUX_IO_URING
2034 if (s->use_linux_io_uring) {
2035 Error *local_err;
2036 if (!aio_setup_linux_io_uring(new_context, &local_err)) {
2037 error_reportf_err(local_err, "Unable to use linux io_uring, "
2038 "falling back to thread pool: ");
2039 s->use_linux_io_uring = false;
2042 #endif
2045 static void raw_close(BlockDriverState *bs)
2047 BDRVRawState *s = bs->opaque;
2049 if (s->fd >= 0) {
2050 qemu_close(s->fd);
2051 s->fd = -1;
2056 * Truncates the given regular file @fd to @offset and, when growing, fills the
2057 * new space according to @prealloc.
2059 * Returns: 0 on success, -errno on failure.
2061 static int coroutine_fn
2062 raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset,
2063 PreallocMode prealloc, Error **errp)
2065 RawPosixAIOData acb;
2067 acb = (RawPosixAIOData) {
2068 .bs = bs,
2069 .aio_fildes = fd,
2070 .aio_type = QEMU_AIO_TRUNCATE,
2071 .aio_offset = offset,
2072 .truncate = {
2073 .prealloc = prealloc,
2074 .errp = errp,
2078 return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb);
2081 static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset,
2082 bool exact, PreallocMode prealloc,
2083 Error **errp)
2085 BDRVRawState *s = bs->opaque;
2086 struct stat st;
2087 int ret;
2089 if (fstat(s->fd, &st)) {
2090 ret = -errno;
2091 error_setg_errno(errp, -ret, "Failed to fstat() the file");
2092 return ret;
2095 if (S_ISREG(st.st_mode)) {
2096 /* Always resizes to the exact @offset */
2097 return raw_regular_truncate(bs, s->fd, offset, prealloc, errp);
2100 if (prealloc != PREALLOC_MODE_OFF) {
2101 error_setg(errp, "Preallocation mode '%s' unsupported for this "
2102 "non-regular file", PreallocMode_str(prealloc));
2103 return -ENOTSUP;
2106 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2107 int64_t cur_length = raw_getlength(bs);
2109 if (offset != cur_length && exact) {
2110 error_setg(errp, "Cannot resize device files");
2111 return -ENOTSUP;
2112 } else if (offset > cur_length) {
2113 error_setg(errp, "Cannot grow device files");
2114 return -EINVAL;
2116 } else {
2117 error_setg(errp, "Resizing this file is not supported");
2118 return -ENOTSUP;
2121 return 0;
2124 #ifdef __OpenBSD__
2125 static int64_t raw_getlength(BlockDriverState *bs)
2127 BDRVRawState *s = bs->opaque;
2128 int fd = s->fd;
2129 struct stat st;
2131 if (fstat(fd, &st))
2132 return -errno;
2133 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2134 struct disklabel dl;
2136 if (ioctl(fd, DIOCGDINFO, &dl))
2137 return -errno;
2138 return (uint64_t)dl.d_secsize *
2139 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
2140 } else
2141 return st.st_size;
2143 #elif defined(__NetBSD__)
2144 static int64_t raw_getlength(BlockDriverState *bs)
2146 BDRVRawState *s = bs->opaque;
2147 int fd = s->fd;
2148 struct stat st;
2150 if (fstat(fd, &st))
2151 return -errno;
2152 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2153 struct dkwedge_info dkw;
2155 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
2156 return dkw.dkw_size * 512;
2157 } else {
2158 struct disklabel dl;
2160 if (ioctl(fd, DIOCGDINFO, &dl))
2161 return -errno;
2162 return (uint64_t)dl.d_secsize *
2163 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
2165 } else
2166 return st.st_size;
2168 #elif defined(__sun__)
2169 static int64_t raw_getlength(BlockDriverState *bs)
2171 BDRVRawState *s = bs->opaque;
2172 struct dk_minfo minfo;
2173 int ret;
2174 int64_t size;
2176 ret = fd_open(bs);
2177 if (ret < 0) {
2178 return ret;
2182 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2184 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
2185 if (ret != -1) {
2186 return minfo.dki_lbsize * minfo.dki_capacity;
2190 * There are reports that lseek on some devices fails, but
2191 * irc discussion said that contingency on contingency was overkill.
2193 size = lseek(s->fd, 0, SEEK_END);
2194 if (size < 0) {
2195 return -errno;
2197 return size;
2199 #elif defined(CONFIG_BSD)
2200 static int64_t raw_getlength(BlockDriverState *bs)
2202 BDRVRawState *s = bs->opaque;
2203 int fd = s->fd;
2204 int64_t size;
2205 struct stat sb;
2206 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2207 int reopened = 0;
2208 #endif
2209 int ret;
2211 ret = fd_open(bs);
2212 if (ret < 0)
2213 return ret;
2215 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2216 again:
2217 #endif
2218 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
2219 #ifdef DIOCGMEDIASIZE
2220 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
2221 #elif defined(DIOCGPART)
2223 struct partinfo pi;
2224 if (ioctl(fd, DIOCGPART, &pi) == 0)
2225 size = pi.media_size;
2226 else
2227 size = 0;
2229 if (size == 0)
2230 #endif
2231 #if defined(__APPLE__) && defined(__MACH__)
2233 uint64_t sectors = 0;
2234 uint32_t sector_size = 0;
2236 if (ioctl(fd, DKIOCGETBLOCKCOUNT, &sectors) == 0
2237 && ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) == 0) {
2238 size = sectors * sector_size;
2239 } else {
2240 size = lseek(fd, 0LL, SEEK_END);
2241 if (size < 0) {
2242 return -errno;
2246 #else
2247 size = lseek(fd, 0LL, SEEK_END);
2248 if (size < 0) {
2249 return -errno;
2251 #endif
2252 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2253 switch(s->type) {
2254 case FTYPE_CD:
2255 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2256 if (size == 2048LL * (unsigned)-1)
2257 size = 0;
2258 /* XXX no disc? maybe we need to reopen... */
2259 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
2260 reopened = 1;
2261 goto again;
2264 #endif
2265 } else {
2266 size = lseek(fd, 0, SEEK_END);
2267 if (size < 0) {
2268 return -errno;
2271 return size;
2273 #else
2274 static int64_t raw_getlength(BlockDriverState *bs)
2276 BDRVRawState *s = bs->opaque;
2277 int ret;
2278 int64_t size;
2280 ret = fd_open(bs);
2281 if (ret < 0) {
2282 return ret;
2285 size = lseek(s->fd, 0, SEEK_END);
2286 if (size < 0) {
2287 return -errno;
2289 return size;
2291 #endif
2293 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
2295 struct stat st;
2296 BDRVRawState *s = bs->opaque;
2298 if (fstat(s->fd, &st) < 0) {
2299 return -errno;
2301 return (int64_t)st.st_blocks * 512;
2304 static int coroutine_fn
2305 raw_co_create(BlockdevCreateOptions *options, Error **errp)
2307 BlockdevCreateOptionsFile *file_opts;
2308 Error *local_err = NULL;
2309 int fd;
2310 uint64_t perm, shared;
2311 int result = 0;
2313 /* Validate options and set default values */
2314 assert(options->driver == BLOCKDEV_DRIVER_FILE);
2315 file_opts = &options->u.file;
2317 if (!file_opts->has_nocow) {
2318 file_opts->nocow = false;
2320 if (!file_opts->has_preallocation) {
2321 file_opts->preallocation = PREALLOC_MODE_OFF;
2324 /* Create file */
2325 fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_BINARY, 0644);
2326 if (fd < 0) {
2327 result = -errno;
2328 error_setg_errno(errp, -result, "Could not create file");
2329 goto out;
2332 /* Take permissions: We want to discard everything, so we need
2333 * BLK_PERM_WRITE; and truncation to the desired size requires
2334 * BLK_PERM_RESIZE.
2335 * On the other hand, we cannot share the RESIZE permission
2336 * because we promise that after this function, the file has the
2337 * size given in the options. If someone else were to resize it
2338 * concurrently, we could not guarantee that.
2339 * Note that after this function, we can no longer guarantee that
2340 * the file is not touched by a third party, so it may be resized
2341 * then. */
2342 perm = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2343 shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE;
2345 /* Step one: Take locks */
2346 result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp);
2347 if (result < 0) {
2348 goto out_close;
2351 /* Step two: Check that nobody else has taken conflicting locks */
2352 result = raw_check_lock_bytes(fd, perm, shared, errp);
2353 if (result < 0) {
2354 error_append_hint(errp,
2355 "Is another process using the image [%s]?\n",
2356 file_opts->filename);
2357 goto out_unlock;
2360 /* Clear the file by truncating it to 0 */
2361 result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp);
2362 if (result < 0) {
2363 goto out_unlock;
2366 if (file_opts->nocow) {
2367 #ifdef __linux__
2368 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2369 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2370 * will be ignored since any failure of this operation should not
2371 * block the left work.
2373 int attr;
2374 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
2375 attr |= FS_NOCOW_FL;
2376 ioctl(fd, FS_IOC_SETFLAGS, &attr);
2378 #endif
2381 /* Resize and potentially preallocate the file to the desired
2382 * final size */
2383 result = raw_regular_truncate(NULL, fd, file_opts->size,
2384 file_opts->preallocation, errp);
2385 if (result < 0) {
2386 goto out_unlock;
2389 out_unlock:
2390 raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err);
2391 if (local_err) {
2392 /* The above call should not fail, and if it does, that does
2393 * not mean the whole creation operation has failed. So
2394 * report it the user for their convenience, but do not report
2395 * it to the caller. */
2396 warn_report_err(local_err);
2399 out_close:
2400 if (qemu_close(fd) != 0 && result == 0) {
2401 result = -errno;
2402 error_setg_errno(errp, -result, "Could not close the new file");
2404 out:
2405 return result;
2408 static int coroutine_fn raw_co_create_opts(BlockDriver *drv,
2409 const char *filename,
2410 QemuOpts *opts,
2411 Error **errp)
2413 BlockdevCreateOptions options;
2414 int64_t total_size = 0;
2415 bool nocow = false;
2416 PreallocMode prealloc;
2417 char *buf = NULL;
2418 Error *local_err = NULL;
2420 /* Skip file: protocol prefix */
2421 strstart(filename, "file:", &filename);
2423 /* Read out options */
2424 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2425 BDRV_SECTOR_SIZE);
2426 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
2427 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2428 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
2429 PREALLOC_MODE_OFF, &local_err);
2430 g_free(buf);
2431 if (local_err) {
2432 error_propagate(errp, local_err);
2433 return -EINVAL;
2436 options = (BlockdevCreateOptions) {
2437 .driver = BLOCKDEV_DRIVER_FILE,
2438 .u.file = {
2439 .filename = (char *) filename,
2440 .size = total_size,
2441 .has_preallocation = true,
2442 .preallocation = prealloc,
2443 .has_nocow = true,
2444 .nocow = nocow,
2447 return raw_co_create(&options, errp);
2450 static int coroutine_fn raw_co_delete_file(BlockDriverState *bs,
2451 Error **errp)
2453 struct stat st;
2454 int ret;
2456 if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) {
2457 error_setg_errno(errp, ENOENT, "%s is not a regular file",
2458 bs->filename);
2459 return -ENOENT;
2462 ret = unlink(bs->filename);
2463 if (ret < 0) {
2464 ret = -errno;
2465 error_setg_errno(errp, -ret, "Error when deleting file %s",
2466 bs->filename);
2469 return ret;
2473 * Find allocation range in @bs around offset @start.
2474 * May change underlying file descriptor's file offset.
2475 * If @start is not in a hole, store @start in @data, and the
2476 * beginning of the next hole in @hole, and return 0.
2477 * If @start is in a non-trailing hole, store @start in @hole and the
2478 * beginning of the next non-hole in @data, and return 0.
2479 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2480 * If we can't find out, return a negative errno other than -ENXIO.
2482 static int find_allocation(BlockDriverState *bs, off_t start,
2483 off_t *data, off_t *hole)
2485 #if defined SEEK_HOLE && defined SEEK_DATA
2486 BDRVRawState *s = bs->opaque;
2487 off_t offs;
2490 * SEEK_DATA cases:
2491 * D1. offs == start: start is in data
2492 * D2. offs > start: start is in a hole, next data at offs
2493 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2494 * or start is beyond EOF
2495 * If the latter happens, the file has been truncated behind
2496 * our back since we opened it. All bets are off then.
2497 * Treating like a trailing hole is simplest.
2498 * D4. offs < 0, errno != ENXIO: we learned nothing
2500 offs = lseek(s->fd, start, SEEK_DATA);
2501 if (offs < 0) {
2502 return -errno; /* D3 or D4 */
2505 if (offs < start) {
2506 /* This is not a valid return by lseek(). We are safe to just return
2507 * -EIO in this case, and we'll treat it like D4. */
2508 return -EIO;
2511 if (offs > start) {
2512 /* D2: in hole, next data at offs */
2513 *hole = start;
2514 *data = offs;
2515 return 0;
2518 /* D1: in data, end not yet known */
2521 * SEEK_HOLE cases:
2522 * H1. offs == start: start is in a hole
2523 * If this happens here, a hole has been dug behind our back
2524 * since the previous lseek().
2525 * H2. offs > start: either start is in data, next hole at offs,
2526 * or start is in trailing hole, EOF at offs
2527 * Linux treats trailing holes like any other hole: offs ==
2528 * start. Solaris seeks to EOF instead: offs > start (blech).
2529 * If that happens here, a hole has been dug behind our back
2530 * since the previous lseek().
2531 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2532 * If this happens, the file has been truncated behind our
2533 * back since we opened it. Treat it like a trailing hole.
2534 * H4. offs < 0, errno != ENXIO: we learned nothing
2535 * Pretend we know nothing at all, i.e. "forget" about D1.
2537 offs = lseek(s->fd, start, SEEK_HOLE);
2538 if (offs < 0) {
2539 return -errno; /* D1 and (H3 or H4) */
2542 if (offs < start) {
2543 /* This is not a valid return by lseek(). We are safe to just return
2544 * -EIO in this case, and we'll treat it like H4. */
2545 return -EIO;
2548 if (offs > start) {
2550 * D1 and H2: either in data, next hole at offs, or it was in
2551 * data but is now in a trailing hole. In the latter case,
2552 * all bets are off. Treating it as if it there was data all
2553 * the way to EOF is safe, so simply do that.
2555 *data = start;
2556 *hole = offs;
2557 return 0;
2560 /* D1 and H1 */
2561 return -EBUSY;
2562 #else
2563 return -ENOTSUP;
2564 #endif
2568 * Returns the allocation status of the specified offset.
2570 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2572 * 'pnum' is set to the number of bytes (including and immediately following
2573 * the specified offset) that are known to be in the same
2574 * allocated/unallocated state.
2576 * 'bytes' is the max value 'pnum' should be set to.
2578 static int coroutine_fn raw_co_block_status(BlockDriverState *bs,
2579 bool want_zero,
2580 int64_t offset,
2581 int64_t bytes, int64_t *pnum,
2582 int64_t *map,
2583 BlockDriverState **file)
2585 off_t data = 0, hole = 0;
2586 int ret;
2588 assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment));
2590 ret = fd_open(bs);
2591 if (ret < 0) {
2592 return ret;
2595 if (!want_zero) {
2596 *pnum = bytes;
2597 *map = offset;
2598 *file = bs;
2599 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
2602 ret = find_allocation(bs, offset, &data, &hole);
2603 if (ret == -ENXIO) {
2604 /* Trailing hole */
2605 *pnum = bytes;
2606 ret = BDRV_BLOCK_ZERO;
2607 } else if (ret < 0) {
2608 /* No info available, so pretend there are no holes */
2609 *pnum = bytes;
2610 ret = BDRV_BLOCK_DATA;
2611 } else if (data == offset) {
2612 /* On a data extent, compute bytes to the end of the extent,
2613 * possibly including a partial sector at EOF. */
2614 *pnum = MIN(bytes, hole - offset);
2617 * We are not allowed to return partial sectors, though, so
2618 * round up if necessary.
2620 if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) {
2621 int64_t file_length = raw_getlength(bs);
2622 if (file_length > 0) {
2623 /* Ignore errors, this is just a safeguard */
2624 assert(hole == file_length);
2626 *pnum = ROUND_UP(*pnum, bs->bl.request_alignment);
2629 ret = BDRV_BLOCK_DATA;
2630 } else {
2631 /* On a hole, compute bytes to the beginning of the next extent. */
2632 assert(hole == offset);
2633 *pnum = MIN(bytes, data - offset);
2634 ret = BDRV_BLOCK_ZERO;
2636 *map = offset;
2637 *file = bs;
2638 return ret | BDRV_BLOCK_OFFSET_VALID;
2641 #if defined(__linux__)
2642 /* Verify that the file is not in the page cache */
2643 static void check_cache_dropped(BlockDriverState *bs, Error **errp)
2645 const size_t window_size = 128 * 1024 * 1024;
2646 BDRVRawState *s = bs->opaque;
2647 void *window = NULL;
2648 size_t length = 0;
2649 unsigned char *vec;
2650 size_t page_size;
2651 off_t offset;
2652 off_t end;
2654 /* mincore(2) page status information requires 1 byte per page */
2655 page_size = sysconf(_SC_PAGESIZE);
2656 vec = g_malloc(DIV_ROUND_UP(window_size, page_size));
2658 end = raw_getlength(bs);
2660 for (offset = 0; offset < end; offset += window_size) {
2661 void *new_window;
2662 size_t new_length;
2663 size_t vec_end;
2664 size_t i;
2665 int ret;
2667 /* Unmap previous window if size has changed */
2668 new_length = MIN(end - offset, window_size);
2669 if (new_length != length) {
2670 munmap(window, length);
2671 window = NULL;
2672 length = 0;
2675 new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE,
2676 s->fd, offset);
2677 if (new_window == MAP_FAILED) {
2678 error_setg_errno(errp, errno, "mmap failed");
2679 break;
2682 window = new_window;
2683 length = new_length;
2685 ret = mincore(window, length, vec);
2686 if (ret < 0) {
2687 error_setg_errno(errp, errno, "mincore failed");
2688 break;
2691 vec_end = DIV_ROUND_UP(length, page_size);
2692 for (i = 0; i < vec_end; i++) {
2693 if (vec[i] & 0x1) {
2694 error_setg(errp, "page cache still in use!");
2695 break;
2700 if (window) {
2701 munmap(window, length);
2704 g_free(vec);
2706 #endif /* __linux__ */
2708 static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs,
2709 Error **errp)
2711 BDRVRawState *s = bs->opaque;
2712 int ret;
2714 ret = fd_open(bs);
2715 if (ret < 0) {
2716 error_setg_errno(errp, -ret, "The file descriptor is not open");
2717 return;
2720 if (!s->drop_cache) {
2721 return;
2724 if (s->open_flags & O_DIRECT) {
2725 return; /* No host kernel page cache */
2728 #if defined(__linux__)
2729 /* This sets the scene for the next syscall... */
2730 ret = bdrv_co_flush(bs);
2731 if (ret < 0) {
2732 error_setg_errno(errp, -ret, "flush failed");
2733 return;
2736 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2737 * process. These limitations are okay because we just fsynced the file,
2738 * we don't use mmap, and the file should not be in use by other processes.
2740 ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED);
2741 if (ret != 0) { /* the return value is a positive errno */
2742 error_setg_errno(errp, ret, "fadvise failed");
2743 return;
2746 if (s->check_cache_dropped) {
2747 check_cache_dropped(bs, errp);
2749 #else /* __linux__ */
2750 /* Do nothing. Live migration to a remote host with cache.direct=off is
2751 * unsupported on other host operating systems. Cache consistency issues
2752 * may occur but no error is reported here, partly because that's the
2753 * historical behavior and partly because it's hard to differentiate valid
2754 * configurations that should not cause errors.
2756 #endif /* !__linux__ */
2759 static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret)
2761 if (ret) {
2762 s->stats.discard_nb_failed++;
2763 } else {
2764 s->stats.discard_nb_ok++;
2765 s->stats.discard_bytes_ok += nbytes;
2769 static coroutine_fn int
2770 raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev)
2772 BDRVRawState *s = bs->opaque;
2773 RawPosixAIOData acb;
2774 int ret;
2776 acb = (RawPosixAIOData) {
2777 .bs = bs,
2778 .aio_fildes = s->fd,
2779 .aio_type = QEMU_AIO_DISCARD,
2780 .aio_offset = offset,
2781 .aio_nbytes = bytes,
2784 if (blkdev) {
2785 acb.aio_type |= QEMU_AIO_BLKDEV;
2788 ret = raw_thread_pool_submit(bs, handle_aiocb_discard, &acb);
2789 raw_account_discard(s, bytes, ret);
2790 return ret;
2793 static coroutine_fn int
2794 raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
2796 return raw_do_pdiscard(bs, offset, bytes, false);
2799 static int coroutine_fn
2800 raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes,
2801 BdrvRequestFlags flags, bool blkdev)
2803 BDRVRawState *s = bs->opaque;
2804 RawPosixAIOData acb;
2805 ThreadPoolFunc *handler;
2807 #ifdef CONFIG_FALLOCATE
2808 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
2809 BdrvTrackedRequest *req;
2810 uint64_t end;
2813 * This is a workaround for a bug in the Linux XFS driver,
2814 * where writes submitted through the AIO interface will be
2815 * discarded if they happen beyond a concurrently running
2816 * fallocate() that increases the file length (i.e., both the
2817 * write and the fallocate() happen beyond the EOF).
2819 * To work around it, we extend the tracked request for this
2820 * zero write until INT64_MAX (effectively infinity), and mark
2821 * it as serializing.
2823 * We have to enable this workaround for all filesystems and
2824 * AIO modes (not just XFS with aio=native), because for
2825 * remote filesystems we do not know the host configuration.
2828 req = bdrv_co_get_self_request(bs);
2829 assert(req);
2830 assert(req->type == BDRV_TRACKED_WRITE);
2831 assert(req->offset <= offset);
2832 assert(req->offset + req->bytes >= offset + bytes);
2834 end = INT64_MAX & -(uint64_t)bs->bl.request_alignment;
2835 req->bytes = end - req->offset;
2836 req->overlap_bytes = req->bytes;
2838 bdrv_mark_request_serialising(req, bs->bl.request_alignment);
2840 #endif
2842 acb = (RawPosixAIOData) {
2843 .bs = bs,
2844 .aio_fildes = s->fd,
2845 .aio_type = QEMU_AIO_WRITE_ZEROES,
2846 .aio_offset = offset,
2847 .aio_nbytes = bytes,
2850 if (blkdev) {
2851 acb.aio_type |= QEMU_AIO_BLKDEV;
2853 if (flags & BDRV_REQ_NO_FALLBACK) {
2854 acb.aio_type |= QEMU_AIO_NO_FALLBACK;
2857 if (flags & BDRV_REQ_MAY_UNMAP) {
2858 acb.aio_type |= QEMU_AIO_DISCARD;
2859 handler = handle_aiocb_write_zeroes_unmap;
2860 } else {
2861 handler = handle_aiocb_write_zeroes;
2864 return raw_thread_pool_submit(bs, handler, &acb);
2867 static int coroutine_fn raw_co_pwrite_zeroes(
2868 BlockDriverState *bs, int64_t offset,
2869 int bytes, BdrvRequestFlags flags)
2871 return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false);
2874 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2876 BDRVRawState *s = bs->opaque;
2878 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
2879 return 0;
2882 static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs)
2884 BDRVRawState *s = bs->opaque;
2885 return (BlockStatsSpecificFile) {
2886 .discard_nb_ok = s->stats.discard_nb_ok,
2887 .discard_nb_failed = s->stats.discard_nb_failed,
2888 .discard_bytes_ok = s->stats.discard_bytes_ok,
2892 static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs)
2894 BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1);
2896 stats->driver = BLOCKDEV_DRIVER_FILE;
2897 stats->u.file = get_blockstats_specific_file(bs);
2899 return stats;
2902 static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs)
2904 BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1);
2906 stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE;
2907 stats->u.host_device = get_blockstats_specific_file(bs);
2909 return stats;
2912 static QemuOptsList raw_create_opts = {
2913 .name = "raw-create-opts",
2914 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
2915 .desc = {
2917 .name = BLOCK_OPT_SIZE,
2918 .type = QEMU_OPT_SIZE,
2919 .help = "Virtual disk size"
2922 .name = BLOCK_OPT_NOCOW,
2923 .type = QEMU_OPT_BOOL,
2924 .help = "Turn off copy-on-write (valid only on btrfs)"
2927 .name = BLOCK_OPT_PREALLOC,
2928 .type = QEMU_OPT_STRING,
2929 .help = "Preallocation mode (allowed values: off"
2930 #ifdef CONFIG_POSIX_FALLOCATE
2931 ", falloc"
2932 #endif
2933 ", full)"
2935 { /* end of list */ }
2939 static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared,
2940 Error **errp)
2942 BDRVRawState *s = bs->opaque;
2943 BDRVRawReopenState *rs = NULL;
2944 int open_flags;
2945 int ret;
2947 if (s->perm_change_fd) {
2949 * In the context of reopen, this function may be called several times
2950 * (directly and recursively while change permissions of the parent).
2951 * This is even true for children that don't inherit from the original
2952 * reopen node, so s->reopen_state is not set.
2954 * Ignore all but the first call.
2956 return 0;
2959 if (s->reopen_state) {
2960 /* We already have a new file descriptor to set permissions for */
2961 assert(s->reopen_state->perm == perm);
2962 assert(s->reopen_state->shared_perm == shared);
2963 rs = s->reopen_state->opaque;
2964 s->perm_change_fd = rs->fd;
2965 s->perm_change_flags = rs->open_flags;
2966 } else {
2967 /* We may need a new fd if auto-read-only switches the mode */
2968 ret = raw_reconfigure_getfd(bs, bs->open_flags, &open_flags, perm,
2969 false, errp);
2970 if (ret < 0) {
2971 return ret;
2972 } else if (ret != s->fd) {
2973 s->perm_change_fd = ret;
2974 s->perm_change_flags = open_flags;
2978 /* Prepare permissions on old fd to avoid conflicts between old and new,
2979 * but keep everything locked that new will need. */
2980 ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp);
2981 if (ret < 0) {
2982 goto fail;
2985 /* Copy locks to the new fd */
2986 if (s->perm_change_fd) {
2987 ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared,
2988 false, errp);
2989 if (ret < 0) {
2990 raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
2991 goto fail;
2994 return 0;
2996 fail:
2997 if (s->perm_change_fd && !s->reopen_state) {
2998 qemu_close(s->perm_change_fd);
3000 s->perm_change_fd = 0;
3001 return ret;
3004 static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared)
3006 BDRVRawState *s = bs->opaque;
3008 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
3009 * called after .bdrv_reopen_commit) */
3010 if (s->perm_change_fd && s->fd != s->perm_change_fd) {
3011 qemu_close(s->fd);
3012 s->fd = s->perm_change_fd;
3013 s->open_flags = s->perm_change_flags;
3015 s->perm_change_fd = 0;
3017 raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL);
3018 s->perm = perm;
3019 s->shared_perm = shared;
3022 static void raw_abort_perm_update(BlockDriverState *bs)
3024 BDRVRawState *s = bs->opaque;
3026 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
3027 * the file descriptor. */
3028 if (s->perm_change_fd && !s->reopen_state) {
3029 qemu_close(s->perm_change_fd);
3031 s->perm_change_fd = 0;
3033 raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
3036 static int coroutine_fn raw_co_copy_range_from(
3037 BlockDriverState *bs, BdrvChild *src, uint64_t src_offset,
3038 BdrvChild *dst, uint64_t dst_offset, uint64_t bytes,
3039 BdrvRequestFlags read_flags, BdrvRequestFlags write_flags)
3041 return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes,
3042 read_flags, write_flags);
3045 static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs,
3046 BdrvChild *src,
3047 uint64_t src_offset,
3048 BdrvChild *dst,
3049 uint64_t dst_offset,
3050 uint64_t bytes,
3051 BdrvRequestFlags read_flags,
3052 BdrvRequestFlags write_flags)
3054 RawPosixAIOData acb;
3055 BDRVRawState *s = bs->opaque;
3056 BDRVRawState *src_s;
3058 assert(dst->bs == bs);
3059 if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) {
3060 return -ENOTSUP;
3063 src_s = src->bs->opaque;
3064 if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) {
3065 return -EIO;
3068 acb = (RawPosixAIOData) {
3069 .bs = bs,
3070 .aio_type = QEMU_AIO_COPY_RANGE,
3071 .aio_fildes = src_s->fd,
3072 .aio_offset = src_offset,
3073 .aio_nbytes = bytes,
3074 .copy_range = {
3075 .aio_fd2 = s->fd,
3076 .aio_offset2 = dst_offset,
3080 return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb);
3083 BlockDriver bdrv_file = {
3084 .format_name = "file",
3085 .protocol_name = "file",
3086 .instance_size = sizeof(BDRVRawState),
3087 .bdrv_needs_filename = true,
3088 .bdrv_probe = NULL, /* no probe for protocols */
3089 .bdrv_parse_filename = raw_parse_filename,
3090 .bdrv_file_open = raw_open,
3091 .bdrv_reopen_prepare = raw_reopen_prepare,
3092 .bdrv_reopen_commit = raw_reopen_commit,
3093 .bdrv_reopen_abort = raw_reopen_abort,
3094 .bdrv_close = raw_close,
3095 .bdrv_co_create = raw_co_create,
3096 .bdrv_co_create_opts = raw_co_create_opts,
3097 .bdrv_has_zero_init = bdrv_has_zero_init_1,
3098 .bdrv_has_zero_init_truncate = bdrv_has_zero_init_1,
3099 .bdrv_co_block_status = raw_co_block_status,
3100 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
3101 .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes,
3102 .bdrv_co_delete_file = raw_co_delete_file,
3104 .bdrv_co_preadv = raw_co_preadv,
3105 .bdrv_co_pwritev = raw_co_pwritev,
3106 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3107 .bdrv_co_pdiscard = raw_co_pdiscard,
3108 .bdrv_co_copy_range_from = raw_co_copy_range_from,
3109 .bdrv_co_copy_range_to = raw_co_copy_range_to,
3110 .bdrv_refresh_limits = raw_refresh_limits,
3111 .bdrv_io_plug = raw_aio_plug,
3112 .bdrv_io_unplug = raw_aio_unplug,
3113 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
3115 .bdrv_co_truncate = raw_co_truncate,
3116 .bdrv_getlength = raw_getlength,
3117 .bdrv_get_info = raw_get_info,
3118 .bdrv_get_allocated_file_size
3119 = raw_get_allocated_file_size,
3120 .bdrv_get_specific_stats = raw_get_specific_stats,
3121 .bdrv_check_perm = raw_check_perm,
3122 .bdrv_set_perm = raw_set_perm,
3123 .bdrv_abort_perm_update = raw_abort_perm_update,
3124 .create_opts = &raw_create_opts,
3125 .mutable_opts = mutable_opts,
3128 /***********************************************/
3129 /* host device */
3131 #if defined(__APPLE__) && defined(__MACH__)
3132 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
3133 CFIndex maxPathSize, int flags);
3134 static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
3136 kern_return_t kernResult = KERN_FAILURE;
3137 mach_port_t masterPort;
3138 CFMutableDictionaryRef classesToMatch;
3139 const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
3140 char *mediaType = NULL;
3142 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
3143 if ( KERN_SUCCESS != kernResult ) {
3144 printf( "IOMasterPort returned %d\n", kernResult );
3147 int index;
3148 for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
3149 classesToMatch = IOServiceMatching(matching_array[index]);
3150 if (classesToMatch == NULL) {
3151 error_report("IOServiceMatching returned NULL for %s",
3152 matching_array[index]);
3153 continue;
3155 CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
3156 kCFBooleanTrue);
3157 kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch,
3158 mediaIterator);
3159 if (kernResult != KERN_SUCCESS) {
3160 error_report("Note: IOServiceGetMatchingServices returned %d",
3161 kernResult);
3162 continue;
3165 /* If a match was found, leave the loop */
3166 if (*mediaIterator != 0) {
3167 trace_file_FindEjectableOpticalMedia(matching_array[index]);
3168 mediaType = g_strdup(matching_array[index]);
3169 break;
3172 return mediaType;
3175 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
3176 CFIndex maxPathSize, int flags)
3178 io_object_t nextMedia;
3179 kern_return_t kernResult = KERN_FAILURE;
3180 *bsdPath = '\0';
3181 nextMedia = IOIteratorNext( mediaIterator );
3182 if ( nextMedia )
3184 CFTypeRef bsdPathAsCFString;
3185 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
3186 if ( bsdPathAsCFString ) {
3187 size_t devPathLength;
3188 strcpy( bsdPath, _PATH_DEV );
3189 if (flags & BDRV_O_NOCACHE) {
3190 strcat(bsdPath, "r");
3192 devPathLength = strlen( bsdPath );
3193 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
3194 kernResult = KERN_SUCCESS;
3196 CFRelease( bsdPathAsCFString );
3198 IOObjectRelease( nextMedia );
3201 return kernResult;
3204 /* Sets up a real cdrom for use in QEMU */
3205 static bool setup_cdrom(char *bsd_path, Error **errp)
3207 int index, num_of_test_partitions = 2, fd;
3208 char test_partition[MAXPATHLEN];
3209 bool partition_found = false;
3211 /* look for a working partition */
3212 for (index = 0; index < num_of_test_partitions; index++) {
3213 snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
3214 index);
3215 fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE);
3216 if (fd >= 0) {
3217 partition_found = true;
3218 qemu_close(fd);
3219 break;
3223 /* if a working partition on the device was not found */
3224 if (partition_found == false) {
3225 error_setg(errp, "Failed to find a working partition on disc");
3226 } else {
3227 trace_file_setup_cdrom(test_partition);
3228 pstrcpy(bsd_path, MAXPATHLEN, test_partition);
3230 return partition_found;
3233 /* Prints directions on mounting and unmounting a device */
3234 static void print_unmounting_directions(const char *file_name)
3236 error_report("If device %s is mounted on the desktop, unmount"
3237 " it first before using it in QEMU", file_name);
3238 error_report("Command to unmount device: diskutil unmountDisk %s",
3239 file_name);
3240 error_report("Command to mount device: diskutil mountDisk %s", file_name);
3243 #endif /* defined(__APPLE__) && defined(__MACH__) */
3245 static int hdev_probe_device(const char *filename)
3247 struct stat st;
3249 /* allow a dedicated CD-ROM driver to match with a higher priority */
3250 if (strstart(filename, "/dev/cdrom", NULL))
3251 return 50;
3253 if (stat(filename, &st) >= 0 &&
3254 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
3255 return 100;
3258 return 0;
3261 static int check_hdev_writable(BDRVRawState *s)
3263 #if defined(BLKROGET)
3264 /* Linux block devices can be configured "read-only" using blockdev(8).
3265 * This is independent of device node permissions and therefore open(2)
3266 * with O_RDWR succeeds. Actual writes fail with EPERM.
3268 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
3269 * check for read-only block devices so that Linux block devices behave
3270 * properly.
3272 struct stat st;
3273 int readonly = 0;
3275 if (fstat(s->fd, &st)) {
3276 return -errno;
3279 if (!S_ISBLK(st.st_mode)) {
3280 return 0;
3283 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
3284 return -errno;
3287 if (readonly) {
3288 return -EACCES;
3290 #endif /* defined(BLKROGET) */
3291 return 0;
3294 static void hdev_parse_filename(const char *filename, QDict *options,
3295 Error **errp)
3297 bdrv_parse_filename_strip_prefix(filename, "host_device:", options);
3300 static bool hdev_is_sg(BlockDriverState *bs)
3303 #if defined(__linux__)
3305 BDRVRawState *s = bs->opaque;
3306 struct stat st;
3307 struct sg_scsi_id scsiid;
3308 int sg_version;
3309 int ret;
3311 if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) {
3312 return false;
3315 ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version);
3316 if (ret < 0) {
3317 return false;
3320 ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid);
3321 if (ret >= 0) {
3322 trace_file_hdev_is_sg(scsiid.scsi_type, sg_version);
3323 return true;
3326 #endif
3328 return false;
3331 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
3332 Error **errp)
3334 BDRVRawState *s = bs->opaque;
3335 Error *local_err = NULL;
3336 int ret;
3338 #if defined(__APPLE__) && defined(__MACH__)
3340 * Caution: while qdict_get_str() is fine, getting non-string types
3341 * would require more care. When @options come from -blockdev or
3342 * blockdev_add, its members are typed according to the QAPI
3343 * schema, but when they come from -drive, they're all QString.
3345 const char *filename = qdict_get_str(options, "filename");
3346 char bsd_path[MAXPATHLEN] = "";
3347 bool error_occurred = false;
3349 /* If using a real cdrom */
3350 if (strcmp(filename, "/dev/cdrom") == 0) {
3351 char *mediaType = NULL;
3352 kern_return_t ret_val;
3353 io_iterator_t mediaIterator = 0;
3355 mediaType = FindEjectableOpticalMedia(&mediaIterator);
3356 if (mediaType == NULL) {
3357 error_setg(errp, "Please make sure your CD/DVD is in the optical"
3358 " drive");
3359 error_occurred = true;
3360 goto hdev_open_Mac_error;
3363 ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
3364 if (ret_val != KERN_SUCCESS) {
3365 error_setg(errp, "Could not get BSD path for optical drive");
3366 error_occurred = true;
3367 goto hdev_open_Mac_error;
3370 /* If a real optical drive was not found */
3371 if (bsd_path[0] == '\0') {
3372 error_setg(errp, "Failed to obtain bsd path for optical drive");
3373 error_occurred = true;
3374 goto hdev_open_Mac_error;
3377 /* If using a cdrom disc and finding a partition on the disc failed */
3378 if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
3379 setup_cdrom(bsd_path, errp) == false) {
3380 print_unmounting_directions(bsd_path);
3381 error_occurred = true;
3382 goto hdev_open_Mac_error;
3385 qdict_put_str(options, "filename", bsd_path);
3387 hdev_open_Mac_error:
3388 g_free(mediaType);
3389 if (mediaIterator) {
3390 IOObjectRelease(mediaIterator);
3392 if (error_occurred) {
3393 return -ENOENT;
3396 #endif /* defined(__APPLE__) && defined(__MACH__) */
3398 s->type = FTYPE_FILE;
3400 ret = raw_open_common(bs, options, flags, 0, true, &local_err);
3401 if (ret < 0) {
3402 error_propagate(errp, local_err);
3403 #if defined(__APPLE__) && defined(__MACH__)
3404 if (*bsd_path) {
3405 filename = bsd_path;
3407 /* if a physical device experienced an error while being opened */
3408 if (strncmp(filename, "/dev/", 5) == 0) {
3409 print_unmounting_directions(filename);
3411 #endif /* defined(__APPLE__) && defined(__MACH__) */
3412 return ret;
3415 /* Since this does ioctl the device must be already opened */
3416 bs->sg = hdev_is_sg(bs);
3418 if (flags & BDRV_O_RDWR) {
3419 ret = check_hdev_writable(s);
3420 if (ret < 0) {
3421 raw_close(bs);
3422 error_setg_errno(errp, -ret, "The device is not writable");
3423 return ret;
3427 return ret;
3430 #if defined(__linux__)
3431 static int coroutine_fn
3432 hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3434 BDRVRawState *s = bs->opaque;
3435 RawPosixAIOData acb;
3436 int ret;
3438 ret = fd_open(bs);
3439 if (ret < 0) {
3440 return ret;
3443 if (req == SG_IO && s->pr_mgr) {
3444 struct sg_io_hdr *io_hdr = buf;
3445 if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT ||
3446 io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) {
3447 return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs),
3448 s->fd, io_hdr);
3452 acb = (RawPosixAIOData) {
3453 .bs = bs,
3454 .aio_type = QEMU_AIO_IOCTL,
3455 .aio_fildes = s->fd,
3456 .aio_offset = 0,
3457 .ioctl = {
3458 .buf = buf,
3459 .cmd = req,
3463 return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb);
3465 #endif /* linux */
3467 static int fd_open(BlockDriverState *bs)
3469 BDRVRawState *s = bs->opaque;
3471 /* this is just to ensure s->fd is sane (its called by io ops) */
3472 if (s->fd >= 0)
3473 return 0;
3474 return -EIO;
3477 static coroutine_fn int
3478 hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
3480 BDRVRawState *s = bs->opaque;
3481 int ret;
3483 ret = fd_open(bs);
3484 if (ret < 0) {
3485 raw_account_discard(s, bytes, ret);
3486 return ret;
3488 return raw_do_pdiscard(bs, offset, bytes, true);
3491 static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs,
3492 int64_t offset, int bytes, BdrvRequestFlags flags)
3494 int rc;
3496 rc = fd_open(bs);
3497 if (rc < 0) {
3498 return rc;
3501 return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true);
3504 static BlockDriver bdrv_host_device = {
3505 .format_name = "host_device",
3506 .protocol_name = "host_device",
3507 .instance_size = sizeof(BDRVRawState),
3508 .bdrv_needs_filename = true,
3509 .bdrv_probe_device = hdev_probe_device,
3510 .bdrv_parse_filename = hdev_parse_filename,
3511 .bdrv_file_open = hdev_open,
3512 .bdrv_close = raw_close,
3513 .bdrv_reopen_prepare = raw_reopen_prepare,
3514 .bdrv_reopen_commit = raw_reopen_commit,
3515 .bdrv_reopen_abort = raw_reopen_abort,
3516 .bdrv_co_create_opts = bdrv_co_create_opts_simple,
3517 .create_opts = &bdrv_create_opts_simple,
3518 .mutable_opts = mutable_opts,
3519 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
3520 .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes,
3522 .bdrv_co_preadv = raw_co_preadv,
3523 .bdrv_co_pwritev = raw_co_pwritev,
3524 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3525 .bdrv_co_pdiscard = hdev_co_pdiscard,
3526 .bdrv_co_copy_range_from = raw_co_copy_range_from,
3527 .bdrv_co_copy_range_to = raw_co_copy_range_to,
3528 .bdrv_refresh_limits = raw_refresh_limits,
3529 .bdrv_io_plug = raw_aio_plug,
3530 .bdrv_io_unplug = raw_aio_unplug,
3531 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
3533 .bdrv_co_truncate = raw_co_truncate,
3534 .bdrv_getlength = raw_getlength,
3535 .bdrv_get_info = raw_get_info,
3536 .bdrv_get_allocated_file_size
3537 = raw_get_allocated_file_size,
3538 .bdrv_get_specific_stats = hdev_get_specific_stats,
3539 .bdrv_check_perm = raw_check_perm,
3540 .bdrv_set_perm = raw_set_perm,
3541 .bdrv_abort_perm_update = raw_abort_perm_update,
3542 .bdrv_probe_blocksizes = hdev_probe_blocksizes,
3543 .bdrv_probe_geometry = hdev_probe_geometry,
3545 /* generic scsi device */
3546 #ifdef __linux__
3547 .bdrv_co_ioctl = hdev_co_ioctl,
3548 #endif
3551 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3552 static void cdrom_parse_filename(const char *filename, QDict *options,
3553 Error **errp)
3555 bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options);
3557 #endif
3559 #ifdef __linux__
3560 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
3561 Error **errp)
3563 BDRVRawState *s = bs->opaque;
3565 s->type = FTYPE_CD;
3567 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3568 return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp);
3571 static int cdrom_probe_device(const char *filename)
3573 int fd, ret;
3574 int prio = 0;
3575 struct stat st;
3577 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
3578 if (fd < 0) {
3579 goto out;
3581 ret = fstat(fd, &st);
3582 if (ret == -1 || !S_ISBLK(st.st_mode)) {
3583 goto outc;
3586 /* Attempt to detect via a CDROM specific ioctl */
3587 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
3588 if (ret >= 0)
3589 prio = 100;
3591 outc:
3592 qemu_close(fd);
3593 out:
3594 return prio;
3597 static bool cdrom_is_inserted(BlockDriverState *bs)
3599 BDRVRawState *s = bs->opaque;
3600 int ret;
3602 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
3603 return ret == CDS_DISC_OK;
3606 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
3608 BDRVRawState *s = bs->opaque;
3610 if (eject_flag) {
3611 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
3612 perror("CDROMEJECT");
3613 } else {
3614 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
3615 perror("CDROMEJECT");
3619 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
3621 BDRVRawState *s = bs->opaque;
3623 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
3625 * Note: an error can happen if the distribution automatically
3626 * mounts the CD-ROM
3628 /* perror("CDROM_LOCKDOOR"); */
3632 static BlockDriver bdrv_host_cdrom = {
3633 .format_name = "host_cdrom",
3634 .protocol_name = "host_cdrom",
3635 .instance_size = sizeof(BDRVRawState),
3636 .bdrv_needs_filename = true,
3637 .bdrv_probe_device = cdrom_probe_device,
3638 .bdrv_parse_filename = cdrom_parse_filename,
3639 .bdrv_file_open = cdrom_open,
3640 .bdrv_close = raw_close,
3641 .bdrv_reopen_prepare = raw_reopen_prepare,
3642 .bdrv_reopen_commit = raw_reopen_commit,
3643 .bdrv_reopen_abort = raw_reopen_abort,
3644 .bdrv_co_create_opts = bdrv_co_create_opts_simple,
3645 .create_opts = &bdrv_create_opts_simple,
3646 .mutable_opts = mutable_opts,
3647 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
3649 .bdrv_co_preadv = raw_co_preadv,
3650 .bdrv_co_pwritev = raw_co_pwritev,
3651 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3652 .bdrv_refresh_limits = raw_refresh_limits,
3653 .bdrv_io_plug = raw_aio_plug,
3654 .bdrv_io_unplug = raw_aio_unplug,
3655 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
3657 .bdrv_co_truncate = raw_co_truncate,
3658 .bdrv_getlength = raw_getlength,
3659 .has_variable_length = true,
3660 .bdrv_get_allocated_file_size
3661 = raw_get_allocated_file_size,
3663 /* removable device support */
3664 .bdrv_is_inserted = cdrom_is_inserted,
3665 .bdrv_eject = cdrom_eject,
3666 .bdrv_lock_medium = cdrom_lock_medium,
3668 /* generic scsi device */
3669 .bdrv_co_ioctl = hdev_co_ioctl,
3671 #endif /* __linux__ */
3673 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3674 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
3675 Error **errp)
3677 BDRVRawState *s = bs->opaque;
3678 Error *local_err = NULL;
3679 int ret;
3681 s->type = FTYPE_CD;
3683 ret = raw_open_common(bs, options, flags, 0, true, &local_err);
3684 if (ret) {
3685 error_propagate(errp, local_err);
3686 return ret;
3689 /* make sure the door isn't locked at this time */
3690 ioctl(s->fd, CDIOCALLOW);
3691 return 0;
3694 static int cdrom_probe_device(const char *filename)
3696 if (strstart(filename, "/dev/cd", NULL) ||
3697 strstart(filename, "/dev/acd", NULL))
3698 return 100;
3699 return 0;
3702 static int cdrom_reopen(BlockDriverState *bs)
3704 BDRVRawState *s = bs->opaque;
3705 int fd;
3708 * Force reread of possibly changed/newly loaded disc,
3709 * FreeBSD seems to not notice sometimes...
3711 if (s->fd >= 0)
3712 qemu_close(s->fd);
3713 fd = qemu_open(bs->filename, s->open_flags, 0644);
3714 if (fd < 0) {
3715 s->fd = -1;
3716 return -EIO;
3718 s->fd = fd;
3720 /* make sure the door isn't locked at this time */
3721 ioctl(s->fd, CDIOCALLOW);
3722 return 0;
3725 static bool cdrom_is_inserted(BlockDriverState *bs)
3727 return raw_getlength(bs) > 0;
3730 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
3732 BDRVRawState *s = bs->opaque;
3734 if (s->fd < 0)
3735 return;
3737 (void) ioctl(s->fd, CDIOCALLOW);
3739 if (eject_flag) {
3740 if (ioctl(s->fd, CDIOCEJECT) < 0)
3741 perror("CDIOCEJECT");
3742 } else {
3743 if (ioctl(s->fd, CDIOCCLOSE) < 0)
3744 perror("CDIOCCLOSE");
3747 cdrom_reopen(bs);
3750 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
3752 BDRVRawState *s = bs->opaque;
3754 if (s->fd < 0)
3755 return;
3756 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
3758 * Note: an error can happen if the distribution automatically
3759 * mounts the CD-ROM
3761 /* perror("CDROM_LOCKDOOR"); */
3765 static BlockDriver bdrv_host_cdrom = {
3766 .format_name = "host_cdrom",
3767 .protocol_name = "host_cdrom",
3768 .instance_size = sizeof(BDRVRawState),
3769 .bdrv_needs_filename = true,
3770 .bdrv_probe_device = cdrom_probe_device,
3771 .bdrv_parse_filename = cdrom_parse_filename,
3772 .bdrv_file_open = cdrom_open,
3773 .bdrv_close = raw_close,
3774 .bdrv_reopen_prepare = raw_reopen_prepare,
3775 .bdrv_reopen_commit = raw_reopen_commit,
3776 .bdrv_reopen_abort = raw_reopen_abort,
3777 .bdrv_co_create_opts = bdrv_co_create_opts_simple,
3778 .create_opts = &bdrv_create_opts_simple,
3779 .mutable_opts = mutable_opts,
3781 .bdrv_co_preadv = raw_co_preadv,
3782 .bdrv_co_pwritev = raw_co_pwritev,
3783 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3784 .bdrv_refresh_limits = raw_refresh_limits,
3785 .bdrv_io_plug = raw_aio_plug,
3786 .bdrv_io_unplug = raw_aio_unplug,
3787 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
3789 .bdrv_co_truncate = raw_co_truncate,
3790 .bdrv_getlength = raw_getlength,
3791 .has_variable_length = true,
3792 .bdrv_get_allocated_file_size
3793 = raw_get_allocated_file_size,
3795 /* removable device support */
3796 .bdrv_is_inserted = cdrom_is_inserted,
3797 .bdrv_eject = cdrom_eject,
3798 .bdrv_lock_medium = cdrom_lock_medium,
3800 #endif /* __FreeBSD__ */
3802 static void bdrv_file_init(void)
3805 * Register all the drivers. Note that order is important, the driver
3806 * registered last will get probed first.
3808 bdrv_register(&bdrv_file);
3809 bdrv_register(&bdrv_host_device);
3810 #ifdef __linux__
3811 bdrv_register(&bdrv_host_cdrom);
3812 #endif
3813 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3814 bdrv_register(&bdrv_host_cdrom);
3815 #endif
3818 block_init(bdrv_file_init);