arm: Update code for several boards to match latest QEMU API
[qemu/ar7.git] / block / file-posix.c
blob36041ab5cf120453b9f2a3fb4a802e7c72e190c5
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 "qapi/error.h"
27 #include "qemu/cutils.h"
28 #include "qemu/error-report.h"
29 #include "block/block_int.h"
30 #include "qemu/module.h"
31 #include "qemu/option.h"
32 #include "trace.h"
33 #include "block/thread-pool.h"
34 #include "qemu/iov.h"
35 #include "block/raw-aio.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qstring.h"
39 #include "scsi/pr-manager.h"
40 #include "scsi/constants.h"
42 #if defined(__APPLE__) && (__MACH__)
43 #include <paths.h>
44 #include <sys/param.h>
45 #include <IOKit/IOKitLib.h>
46 #include <IOKit/IOBSD.h>
47 #include <IOKit/storage/IOMediaBSDClient.h>
48 #include <IOKit/storage/IOMedia.h>
49 #include <IOKit/storage/IOCDMedia.h>
50 //#include <IOKit/storage/IOCDTypes.h>
51 #include <IOKit/storage/IODVDMedia.h>
52 #include <CoreFoundation/CoreFoundation.h>
53 #endif
55 #ifdef __sun__
56 #define _POSIX_PTHREAD_SEMANTICS 1
57 #include <sys/dkio.h>
58 #endif
59 #ifdef __linux__
60 #include <sys/ioctl.h>
61 #include <sys/param.h>
62 #include <sys/syscall.h>
63 #include <linux/cdrom.h>
64 #include <linux/fd.h>
65 #include <linux/fs.h>
66 #include <linux/hdreg.h>
67 #include <scsi/sg.h>
68 #ifdef __s390__
69 #include <asm/dasd.h>
70 #endif
71 #ifndef FS_NOCOW_FL
72 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
73 #endif
74 #endif
75 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
76 #include <linux/falloc.h>
77 #endif
78 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
79 #include <sys/disk.h>
80 #include <sys/cdio.h>
81 #endif
83 #ifdef __OpenBSD__
84 #include <sys/ioctl.h>
85 #include <sys/disklabel.h>
86 #include <sys/dkio.h>
87 #endif
89 #ifdef __NetBSD__
90 #include <sys/ioctl.h>
91 #include <sys/disklabel.h>
92 #include <sys/dkio.h>
93 #include <sys/disk.h>
94 #endif
96 #ifdef __DragonFly__
97 #include <sys/ioctl.h>
98 #include <sys/diskslice.h>
99 #endif
101 #ifdef CONFIG_XFS
102 #include <xfs/xfs.h>
103 #endif
105 #include "trace.h"
107 /* OS X does not have O_DSYNC */
108 #ifndef O_DSYNC
109 #ifdef O_SYNC
110 #define O_DSYNC O_SYNC
111 #elif defined(O_FSYNC)
112 #define O_DSYNC O_FSYNC
113 #endif
114 #endif
116 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
117 #ifndef O_DIRECT
118 #define O_DIRECT O_DSYNC
119 #endif
121 #define FTYPE_FILE 0
122 #define FTYPE_CD 1
124 #define MAX_BLOCKSIZE 4096
126 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
127 * leaving a few more bytes for its future use. */
128 #define RAW_LOCK_PERM_BASE 100
129 #define RAW_LOCK_SHARED_BASE 200
131 typedef struct BDRVRawState {
132 int fd;
133 bool use_lock;
134 int type;
135 int open_flags;
136 size_t buf_align;
138 /* The current permissions. */
139 uint64_t perm;
140 uint64_t shared_perm;
142 /* The perms bits whose corresponding bytes are already locked in
143 * s->fd. */
144 uint64_t locked_perm;
145 uint64_t locked_shared_perm;
147 int perm_change_fd;
148 BDRVReopenState *reopen_state;
150 #ifdef CONFIG_XFS
151 bool is_xfs:1;
152 #endif
153 bool has_discard:1;
154 bool has_write_zeroes:1;
155 bool discard_zeroes:1;
156 bool use_linux_aio:1;
157 bool page_cache_inconsistent:1;
158 bool has_fallocate;
159 bool needs_alignment;
160 bool drop_cache;
161 bool check_cache_dropped;
163 PRManager *pr_mgr;
164 } BDRVRawState;
166 typedef struct BDRVRawReopenState {
167 int fd;
168 int open_flags;
169 bool drop_cache;
170 bool check_cache_dropped;
171 } BDRVRawReopenState;
173 static int fd_open(BlockDriverState *bs);
174 static int64_t raw_getlength(BlockDriverState *bs);
176 typedef struct RawPosixAIOData {
177 BlockDriverState *bs;
178 int aio_type;
179 int aio_fildes;
181 off_t aio_offset;
182 uint64_t aio_nbytes;
184 union {
185 struct {
186 struct iovec *iov;
187 int niov;
188 } io;
189 struct {
190 uint64_t cmd;
191 void *buf;
192 } ioctl;
193 struct {
194 int aio_fd2;
195 off_t aio_offset2;
196 } copy_range;
197 struct {
198 PreallocMode prealloc;
199 Error **errp;
200 } truncate;
202 } RawPosixAIOData;
204 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
205 static int cdrom_reopen(BlockDriverState *bs);
206 #endif
208 #if defined(__NetBSD__)
209 static int raw_normalize_devicepath(const char **filename, Error **errp)
211 static char namebuf[PATH_MAX];
212 const char *dp, *fname;
213 struct stat sb;
215 fname = *filename;
216 dp = strrchr(fname, '/');
217 if (lstat(fname, &sb) < 0) {
218 error_setg_errno(errp, errno, "%s: stat failed", fname);
219 return -errno;
222 if (!S_ISBLK(sb.st_mode)) {
223 return 0;
226 if (dp == NULL) {
227 snprintf(namebuf, PATH_MAX, "r%s", fname);
228 } else {
229 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
230 (int)(dp - fname), fname, dp + 1);
232 *filename = namebuf;
233 warn_report("%s is a block device, using %s", fname, *filename);
235 return 0;
237 #else
238 static int raw_normalize_devicepath(const char **filename, Error **errp)
240 return 0;
242 #endif
245 * Get logical block size via ioctl. On success store it in @sector_size_p.
247 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
249 unsigned int sector_size;
250 bool success = false;
251 int i;
253 errno = ENOTSUP;
254 static const unsigned long ioctl_list[] = {
255 #ifdef BLKSSZGET
256 BLKSSZGET,
257 #endif
258 #ifdef DKIOCGETBLOCKSIZE
259 DKIOCGETBLOCKSIZE,
260 #endif
261 #ifdef DIOCGSECTORSIZE
262 DIOCGSECTORSIZE,
263 #endif
266 /* Try a few ioctls to get the right size */
267 for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) {
268 if (ioctl(fd, ioctl_list[i], &sector_size) >= 0) {
269 *sector_size_p = sector_size;
270 success = true;
274 return success ? 0 : -errno;
278 * Get physical block size of @fd.
279 * On success, store it in @blk_size and return 0.
280 * On failure, return -errno.
282 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
284 #ifdef BLKPBSZGET
285 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
286 return -errno;
288 return 0;
289 #else
290 return -ENOTSUP;
291 #endif
294 /* Check if read is allowed with given memory buffer and length.
296 * This function is used to check O_DIRECT memory buffer and request alignment.
298 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
300 ssize_t ret = pread(fd, buf, len, 0);
302 if (ret >= 0) {
303 return true;
306 #ifdef __linux__
307 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
308 * other errors (e.g. real I/O error), which could happen on a failed
309 * drive, since we only care about probing alignment.
311 if (errno != EINVAL) {
312 return true;
314 #endif
316 return false;
319 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
321 BDRVRawState *s = bs->opaque;
322 char *buf;
323 size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize());
325 /* For SCSI generic devices the alignment is not really used.
326 With buffered I/O, we don't have any restrictions. */
327 if (bdrv_is_sg(bs) || !s->needs_alignment) {
328 bs->bl.request_alignment = 1;
329 s->buf_align = 1;
330 return;
333 bs->bl.request_alignment = 0;
334 s->buf_align = 0;
335 /* Let's try to use the logical blocksize for the alignment. */
336 if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) {
337 bs->bl.request_alignment = 0;
339 #ifdef CONFIG_XFS
340 if (s->is_xfs) {
341 struct dioattr da;
342 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
343 bs->bl.request_alignment = da.d_miniosz;
344 /* The kernel returns wrong information for d_mem */
345 /* s->buf_align = da.d_mem; */
348 #endif
350 /* If we could not get the sizes so far, we can only guess them */
351 if (!s->buf_align) {
352 size_t align;
353 buf = qemu_memalign(max_align, 2 * max_align);
354 for (align = 512; align <= max_align; align <<= 1) {
355 if (raw_is_io_aligned(fd, buf + align, max_align)) {
356 s->buf_align = align;
357 break;
360 qemu_vfree(buf);
363 if (!bs->bl.request_alignment) {
364 size_t align;
365 buf = qemu_memalign(s->buf_align, max_align);
366 for (align = 512; align <= max_align; align <<= 1) {
367 if (raw_is_io_aligned(fd, buf, align)) {
368 bs->bl.request_alignment = align;
369 break;
372 qemu_vfree(buf);
375 if (!s->buf_align || !bs->bl.request_alignment) {
376 error_setg(errp, "Could not find working O_DIRECT alignment");
377 error_append_hint(errp, "Try cache.direct=off\n");
381 static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers)
383 bool read_write = false;
384 assert(open_flags != NULL);
386 *open_flags |= O_BINARY;
387 *open_flags &= ~O_ACCMODE;
389 if (bdrv_flags & BDRV_O_AUTO_RDONLY) {
390 read_write = has_writers;
391 } else if (bdrv_flags & BDRV_O_RDWR) {
392 read_write = true;
395 if (read_write) {
396 *open_flags |= O_RDWR;
397 } else {
398 *open_flags |= O_RDONLY;
401 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
402 * and O_DIRECT for no caching. */
403 if ((bdrv_flags & BDRV_O_NOCACHE)) {
404 *open_flags |= O_DIRECT;
408 static void raw_parse_filename(const char *filename, QDict *options,
409 Error **errp)
411 bdrv_parse_filename_strip_prefix(filename, "file:", options);
414 static QemuOptsList raw_runtime_opts = {
415 .name = "raw",
416 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
417 .desc = {
419 .name = "filename",
420 .type = QEMU_OPT_STRING,
421 .help = "File name of the image",
424 .name = "aio",
425 .type = QEMU_OPT_STRING,
426 .help = "host AIO implementation (threads, native)",
429 .name = "locking",
430 .type = QEMU_OPT_STRING,
431 .help = "file locking mode (on/off/auto, default: auto)",
434 .name = "pr-manager",
435 .type = QEMU_OPT_STRING,
436 .help = "id of persistent reservation manager object (default: none)",
438 #if defined(__linux__)
440 .name = "drop-cache",
441 .type = QEMU_OPT_BOOL,
442 .help = "invalidate page cache during live migration (default: on)",
444 #endif
446 .name = "x-check-cache-dropped",
447 .type = QEMU_OPT_BOOL,
448 .help = "check that page cache was dropped on live migration (default: off)"
450 { /* end of list */ }
454 static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL };
456 static int raw_open_common(BlockDriverState *bs, QDict *options,
457 int bdrv_flags, int open_flags,
458 bool device, Error **errp)
460 BDRVRawState *s = bs->opaque;
461 QemuOpts *opts;
462 Error *local_err = NULL;
463 const char *filename = NULL;
464 const char *str;
465 BlockdevAioOptions aio, aio_default;
466 int fd, ret;
467 struct stat st;
468 OnOffAuto locking;
470 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
471 qemu_opts_absorb_qdict(opts, options, &local_err);
472 if (local_err) {
473 error_propagate(errp, local_err);
474 ret = -EINVAL;
475 goto fail;
478 filename = qemu_opt_get(opts, "filename");
480 ret = raw_normalize_devicepath(&filename, errp);
481 if (ret != 0) {
482 goto fail;
485 aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO)
486 ? BLOCKDEV_AIO_OPTIONS_NATIVE
487 : BLOCKDEV_AIO_OPTIONS_THREADS;
488 aio = qapi_enum_parse(&BlockdevAioOptions_lookup,
489 qemu_opt_get(opts, "aio"),
490 aio_default, &local_err);
491 if (local_err) {
492 error_propagate(errp, local_err);
493 ret = -EINVAL;
494 goto fail;
496 s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE);
498 locking = qapi_enum_parse(&OnOffAuto_lookup,
499 qemu_opt_get(opts, "locking"),
500 ON_OFF_AUTO_AUTO, &local_err);
501 if (local_err) {
502 error_propagate(errp, local_err);
503 ret = -EINVAL;
504 goto fail;
506 switch (locking) {
507 case ON_OFF_AUTO_ON:
508 s->use_lock = true;
509 if (!qemu_has_ofd_lock()) {
510 warn_report("File lock requested but OFD locking syscall is "
511 "unavailable, falling back to POSIX file locks");
512 error_printf("Due to the implementation, locks can be lost "
513 "unexpectedly.\n");
515 break;
516 case ON_OFF_AUTO_OFF:
517 s->use_lock = false;
518 break;
519 case ON_OFF_AUTO_AUTO:
520 s->use_lock = qemu_has_ofd_lock();
521 break;
522 default:
523 abort();
526 str = qemu_opt_get(opts, "pr-manager");
527 if (str) {
528 s->pr_mgr = pr_manager_lookup(str, &local_err);
529 if (local_err) {
530 error_propagate(errp, local_err);
531 ret = -EINVAL;
532 goto fail;
536 s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true);
537 s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped",
538 false);
540 s->open_flags = open_flags;
541 raw_parse_flags(bdrv_flags, &s->open_flags, false);
543 s->fd = -1;
544 fd = qemu_open(filename, s->open_flags, 0644);
545 ret = fd < 0 ? -errno : 0;
547 if (ret < 0) {
548 error_setg_errno(errp, -ret, "Could not open '%s'", filename);
549 if (ret == -EROFS) {
550 ret = -EACCES;
552 goto fail;
554 s->fd = fd;
556 s->perm = 0;
557 s->shared_perm = BLK_PERM_ALL;
559 #ifdef CONFIG_LINUX_AIO
560 /* Currently Linux does AIO only for files opened with O_DIRECT */
561 if (s->use_linux_aio) {
562 if (!(s->open_flags & O_DIRECT)) {
563 error_setg(errp, "aio=native was specified, but it requires "
564 "cache.direct=on, which was not specified.");
565 ret = -EINVAL;
566 goto fail;
568 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) {
569 error_prepend(errp, "Unable to use native AIO: ");
570 goto fail;
573 #else
574 if (s->use_linux_aio) {
575 error_setg(errp, "aio=native was specified, but is not supported "
576 "in this build.");
577 ret = -EINVAL;
578 goto fail;
580 #endif /* !defined(CONFIG_LINUX_AIO) */
582 s->has_discard = true;
583 s->has_write_zeroes = true;
584 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
585 s->needs_alignment = true;
588 if (fstat(s->fd, &st) < 0) {
589 ret = -errno;
590 error_setg_errno(errp, errno, "Could not stat file");
591 goto fail;
594 if (!device) {
595 if (S_ISBLK(st.st_mode)) {
596 warn_report("Opening a block device as a file using the '%s' "
597 "driver is deprecated", bs->drv->format_name);
598 } else if (S_ISCHR(st.st_mode)) {
599 warn_report("Opening a character device as a file using the '%s' "
600 "driver is deprecated", bs->drv->format_name);
601 } else if (!S_ISREG(st.st_mode)) {
602 error_setg(errp, "A regular file was expected by the '%s' driver, "
603 "but something else was given", bs->drv->format_name);
604 ret = -EINVAL;
605 goto fail;
606 } else {
607 s->discard_zeroes = true;
608 s->has_fallocate = true;
610 } else {
611 if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
612 error_setg(errp, "'%s' driver expects either "
613 "a character or block device", bs->drv->format_name);
614 ret = -EINVAL;
615 goto fail;
619 if (S_ISBLK(st.st_mode)) {
620 #ifdef BLKDISCARDZEROES
621 unsigned int arg;
622 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
623 s->discard_zeroes = true;
625 #endif
626 #ifdef __linux__
627 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
628 * not rely on the contents of discarded blocks unless using O_DIRECT.
629 * Same for BLKZEROOUT.
631 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
632 s->discard_zeroes = false;
633 s->has_write_zeroes = false;
635 #endif
637 #ifdef __FreeBSD__
638 if (S_ISCHR(st.st_mode)) {
640 * The file is a char device (disk), which on FreeBSD isn't behind
641 * a pager, so force all requests to be aligned. This is needed
642 * so QEMU makes sure all IO operations on the device are aligned
643 * to sector size, or else FreeBSD will reject them with EINVAL.
645 s->needs_alignment = true;
647 #endif
649 #ifdef CONFIG_XFS
650 if (platform_test_xfs_fd(s->fd)) {
651 s->is_xfs = true;
653 #endif
655 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
656 ret = 0;
657 fail:
658 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
659 unlink(filename);
661 qemu_opts_del(opts);
662 return ret;
665 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
666 Error **errp)
668 BDRVRawState *s = bs->opaque;
670 s->type = FTYPE_FILE;
671 return raw_open_common(bs, options, flags, 0, false, errp);
674 typedef enum {
675 RAW_PL_PREPARE,
676 RAW_PL_COMMIT,
677 RAW_PL_ABORT,
678 } RawPermLockOp;
680 #define PERM_FOREACH(i) \
681 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
683 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
684 * file; if @unlock == true, also unlock the unneeded bytes.
685 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
687 static int raw_apply_lock_bytes(BDRVRawState *s, int fd,
688 uint64_t perm_lock_bits,
689 uint64_t shared_perm_lock_bits,
690 bool unlock, Error **errp)
692 int ret;
693 int i;
694 uint64_t locked_perm, locked_shared_perm;
696 if (s) {
697 locked_perm = s->locked_perm;
698 locked_shared_perm = s->locked_shared_perm;
699 } else {
701 * We don't have the previous bits, just lock/unlock for each of the
702 * requested bits.
704 if (unlock) {
705 locked_perm = BLK_PERM_ALL;
706 locked_shared_perm = BLK_PERM_ALL;
707 } else {
708 locked_perm = 0;
709 locked_shared_perm = 0;
713 PERM_FOREACH(i) {
714 int off = RAW_LOCK_PERM_BASE + i;
715 uint64_t bit = (1ULL << i);
716 if ((perm_lock_bits & bit) && !(locked_perm & bit)) {
717 ret = qemu_lock_fd(fd, off, 1, false);
718 if (ret) {
719 error_setg(errp, "Failed to lock byte %d", off);
720 return ret;
721 } else if (s) {
722 s->locked_perm |= bit;
724 } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) {
725 ret = qemu_unlock_fd(fd, off, 1);
726 if (ret) {
727 error_setg(errp, "Failed to unlock byte %d", off);
728 return ret;
729 } else if (s) {
730 s->locked_perm &= ~bit;
734 PERM_FOREACH(i) {
735 int off = RAW_LOCK_SHARED_BASE + i;
736 uint64_t bit = (1ULL << i);
737 if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) {
738 ret = qemu_lock_fd(fd, off, 1, false);
739 if (ret) {
740 error_setg(errp, "Failed to lock byte %d", off);
741 return ret;
742 } else if (s) {
743 s->locked_shared_perm |= bit;
745 } else if (unlock && (locked_shared_perm & bit) &&
746 !(shared_perm_lock_bits & bit)) {
747 ret = qemu_unlock_fd(fd, off, 1);
748 if (ret) {
749 error_setg(errp, "Failed to unlock byte %d", off);
750 return ret;
751 } else if (s) {
752 s->locked_shared_perm &= ~bit;
756 return 0;
759 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
760 static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm,
761 Error **errp)
763 int ret;
764 int i;
766 PERM_FOREACH(i) {
767 int off = RAW_LOCK_SHARED_BASE + i;
768 uint64_t p = 1ULL << i;
769 if (perm & p) {
770 ret = qemu_lock_fd_test(fd, off, 1, true);
771 if (ret) {
772 char *perm_name = bdrv_perm_names(p);
773 error_setg(errp,
774 "Failed to get \"%s\" lock",
775 perm_name);
776 g_free(perm_name);
777 return ret;
781 PERM_FOREACH(i) {
782 int off = RAW_LOCK_PERM_BASE + i;
783 uint64_t p = 1ULL << i;
784 if (!(shared_perm & p)) {
785 ret = qemu_lock_fd_test(fd, off, 1, true);
786 if (ret) {
787 char *perm_name = bdrv_perm_names(p);
788 error_setg(errp,
789 "Failed to get shared \"%s\" lock",
790 perm_name);
791 g_free(perm_name);
792 return ret;
796 return 0;
799 static int raw_handle_perm_lock(BlockDriverState *bs,
800 RawPermLockOp op,
801 uint64_t new_perm, uint64_t new_shared,
802 Error **errp)
804 BDRVRawState *s = bs->opaque;
805 int ret = 0;
806 Error *local_err = NULL;
808 if (!s->use_lock) {
809 return 0;
812 if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) {
813 return 0;
816 switch (op) {
817 case RAW_PL_PREPARE:
818 ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm,
819 ~s->shared_perm | ~new_shared,
820 false, errp);
821 if (!ret) {
822 ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp);
823 if (!ret) {
824 return 0;
826 error_append_hint(errp,
827 "Is another process using the image [%s]?\n",
828 bs->filename);
830 op = RAW_PL_ABORT;
831 /* fall through to unlock bytes. */
832 case RAW_PL_ABORT:
833 raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm,
834 true, &local_err);
835 if (local_err) {
836 /* Theoretically the above call only unlocks bytes and it cannot
837 * fail. Something weird happened, report it.
839 warn_report_err(local_err);
841 break;
842 case RAW_PL_COMMIT:
843 raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared,
844 true, &local_err);
845 if (local_err) {
846 /* Theoretically the above call only unlocks bytes and it cannot
847 * fail. Something weird happened, report it.
849 warn_report_err(local_err);
851 break;
853 return ret;
856 static int raw_reconfigure_getfd(BlockDriverState *bs, int flags,
857 int *open_flags, uint64_t perm, bool force_dup,
858 Error **errp)
860 BDRVRawState *s = bs->opaque;
861 int fd = -1;
862 int ret;
863 bool has_writers = perm &
864 (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE);
865 int fcntl_flags = O_APPEND | O_NONBLOCK;
866 #ifdef O_NOATIME
867 fcntl_flags |= O_NOATIME;
868 #endif
870 *open_flags = 0;
871 if (s->type == FTYPE_CD) {
872 *open_flags |= O_NONBLOCK;
875 raw_parse_flags(flags, open_flags, has_writers);
877 #ifdef O_ASYNC
878 /* Not all operating systems have O_ASYNC, and those that don't
879 * will not let us track the state into rs->open_flags (typically
880 * you achieve the same effect with an ioctl, for example I_SETSIG
881 * on Solaris). But we do not use O_ASYNC, so that's fine.
883 assert((s->open_flags & O_ASYNC) == 0);
884 #endif
886 if (!force_dup && *open_flags == s->open_flags) {
887 /* We're lucky, the existing fd is fine */
888 return s->fd;
891 if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
892 /* dup the original fd */
893 fd = qemu_dup(s->fd);
894 if (fd >= 0) {
895 ret = fcntl_setfl(fd, *open_flags);
896 if (ret) {
897 qemu_close(fd);
898 fd = -1;
903 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
904 if (fd == -1) {
905 const char *normalized_filename = bs->filename;
906 ret = raw_normalize_devicepath(&normalized_filename, errp);
907 if (ret >= 0) {
908 assert(!(*open_flags & O_CREAT));
909 fd = qemu_open(normalized_filename, *open_flags);
910 if (fd == -1) {
911 error_setg_errno(errp, errno, "Could not reopen file");
912 return -1;
917 return fd;
920 static int raw_reopen_prepare(BDRVReopenState *state,
921 BlockReopenQueue *queue, Error **errp)
923 BDRVRawState *s;
924 BDRVRawReopenState *rs;
925 QemuOpts *opts;
926 int ret;
927 Error *local_err = NULL;
929 assert(state != NULL);
930 assert(state->bs != NULL);
932 s = state->bs->opaque;
934 state->opaque = g_new0(BDRVRawReopenState, 1);
935 rs = state->opaque;
937 /* Handle options changes */
938 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
939 qemu_opts_absorb_qdict(opts, state->options, &local_err);
940 if (local_err) {
941 error_propagate(errp, local_err);
942 ret = -EINVAL;
943 goto out;
946 rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true);
947 rs->check_cache_dropped =
948 qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false);
950 /* This driver's reopen function doesn't currently allow changing
951 * other options, so let's put them back in the original QDict and
952 * bdrv_reopen_prepare() will detect changes and complain. */
953 qemu_opts_to_qdict(opts, state->options);
955 rs->fd = raw_reconfigure_getfd(state->bs, state->flags, &rs->open_flags,
956 state->perm, true, &local_err);
957 if (local_err) {
958 error_propagate(errp, local_err);
959 ret = -1;
960 goto out;
963 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
964 * alignment with the new fd. */
965 if (rs->fd != -1) {
966 raw_probe_alignment(state->bs, rs->fd, &local_err);
967 if (local_err) {
968 error_propagate(errp, local_err);
969 ret = -EINVAL;
970 goto out_fd;
974 s->reopen_state = state;
975 ret = 0;
976 out_fd:
977 if (ret < 0) {
978 qemu_close(rs->fd);
979 rs->fd = -1;
981 out:
982 qemu_opts_del(opts);
983 return ret;
986 static void raw_reopen_commit(BDRVReopenState *state)
988 BDRVRawReopenState *rs = state->opaque;
989 BDRVRawState *s = state->bs->opaque;
991 s->drop_cache = rs->drop_cache;
992 s->check_cache_dropped = rs->check_cache_dropped;
993 s->open_flags = rs->open_flags;
995 qemu_close(s->fd);
996 s->fd = rs->fd;
998 g_free(state->opaque);
999 state->opaque = NULL;
1001 assert(s->reopen_state == state);
1002 s->reopen_state = NULL;
1006 static void raw_reopen_abort(BDRVReopenState *state)
1008 BDRVRawReopenState *rs = state->opaque;
1009 BDRVRawState *s = state->bs->opaque;
1011 /* nothing to do if NULL, we didn't get far enough */
1012 if (rs == NULL) {
1013 return;
1016 if (rs->fd >= 0) {
1017 qemu_close(rs->fd);
1018 rs->fd = -1;
1020 g_free(state->opaque);
1021 state->opaque = NULL;
1023 assert(s->reopen_state == state);
1024 s->reopen_state = NULL;
1027 static int hdev_get_max_transfer_length(BlockDriverState *bs, int fd)
1029 #ifdef BLKSECTGET
1030 int max_bytes = 0;
1031 short max_sectors = 0;
1032 if (bs->sg && ioctl(fd, BLKSECTGET, &max_bytes) == 0) {
1033 return max_bytes;
1034 } else if (!bs->sg && ioctl(fd, BLKSECTGET, &max_sectors) == 0) {
1035 return max_sectors << BDRV_SECTOR_BITS;
1036 } else {
1037 return -errno;
1039 #else
1040 return -ENOSYS;
1041 #endif
1044 static int hdev_get_max_segments(const struct stat *st)
1046 #ifdef CONFIG_LINUX
1047 char buf[32];
1048 const char *end;
1049 char *sysfspath;
1050 int ret;
1051 int fd = -1;
1052 long max_segments;
1054 sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1055 major(st->st_rdev), minor(st->st_rdev));
1056 fd = open(sysfspath, O_RDONLY);
1057 if (fd == -1) {
1058 ret = -errno;
1059 goto out;
1061 do {
1062 ret = read(fd, buf, sizeof(buf) - 1);
1063 } while (ret == -1 && errno == EINTR);
1064 if (ret < 0) {
1065 ret = -errno;
1066 goto out;
1067 } else if (ret == 0) {
1068 ret = -EIO;
1069 goto out;
1071 buf[ret] = 0;
1072 /* The file is ended with '\n', pass 'end' to accept that. */
1073 ret = qemu_strtol(buf, &end, 10, &max_segments);
1074 if (ret == 0 && end && *end == '\n') {
1075 ret = max_segments;
1078 out:
1079 if (fd != -1) {
1080 close(fd);
1082 g_free(sysfspath);
1083 return ret;
1084 #else
1085 return -ENOTSUP;
1086 #endif
1089 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
1091 BDRVRawState *s = bs->opaque;
1092 struct stat st;
1094 if (!fstat(s->fd, &st)) {
1095 if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode)) {
1096 int ret = hdev_get_max_transfer_length(bs, s->fd);
1097 if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) {
1098 bs->bl.max_transfer = pow2floor(ret);
1100 ret = hdev_get_max_segments(&st);
1101 if (ret > 0) {
1102 bs->bl.max_transfer = MIN(bs->bl.max_transfer,
1103 ret * getpagesize());
1108 raw_probe_alignment(bs, s->fd, errp);
1109 bs->bl.min_mem_alignment = s->buf_align;
1110 bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
1113 static int check_for_dasd(int fd)
1115 #ifdef BIODASDINFO2
1116 struct dasd_information2_t info = {0};
1118 return ioctl(fd, BIODASDINFO2, &info);
1119 #else
1120 return -1;
1121 #endif
1125 * Try to get @bs's logical and physical block size.
1126 * On success, store them in @bsz and return zero.
1127 * On failure, return negative errno.
1129 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
1131 BDRVRawState *s = bs->opaque;
1132 int ret;
1134 /* If DASD, get blocksizes */
1135 if (check_for_dasd(s->fd) < 0) {
1136 return -ENOTSUP;
1138 ret = probe_logical_blocksize(s->fd, &bsz->log);
1139 if (ret < 0) {
1140 return ret;
1142 return probe_physical_blocksize(s->fd, &bsz->phys);
1146 * Try to get @bs's geometry: cyls, heads, sectors.
1147 * On success, store them in @geo and return 0.
1148 * On failure return -errno.
1149 * (Allows block driver to assign default geometry values that guest sees)
1151 #ifdef __linux__
1152 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1154 BDRVRawState *s = bs->opaque;
1155 struct hd_geometry ioctl_geo = {0};
1157 /* If DASD, get its geometry */
1158 if (check_for_dasd(s->fd) < 0) {
1159 return -ENOTSUP;
1161 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
1162 return -errno;
1164 /* HDIO_GETGEO may return success even though geo contains zeros
1165 (e.g. certain multipath setups) */
1166 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
1167 return -ENOTSUP;
1169 /* Do not return a geometry for partition */
1170 if (ioctl_geo.start != 0) {
1171 return -ENOTSUP;
1173 geo->heads = ioctl_geo.heads;
1174 geo->sectors = ioctl_geo.sectors;
1175 geo->cylinders = ioctl_geo.cylinders;
1177 return 0;
1179 #else /* __linux__ */
1180 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1182 return -ENOTSUP;
1184 #endif
1186 #if defined(__linux__)
1187 static int handle_aiocb_ioctl(void *opaque)
1189 RawPosixAIOData *aiocb = opaque;
1190 int ret;
1192 ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf);
1193 if (ret == -1) {
1194 return -errno;
1197 return 0;
1199 #endif /* linux */
1201 static int handle_aiocb_flush(void *opaque)
1203 RawPosixAIOData *aiocb = opaque;
1204 BDRVRawState *s = aiocb->bs->opaque;
1205 int ret;
1207 if (s->page_cache_inconsistent) {
1208 return -EIO;
1211 ret = qemu_fdatasync(aiocb->aio_fildes);
1212 if (ret == -1) {
1213 /* There is no clear definition of the semantics of a failing fsync(),
1214 * so we may have to assume the worst. The sad truth is that this
1215 * assumption is correct for Linux. Some pages are now probably marked
1216 * clean in the page cache even though they are inconsistent with the
1217 * on-disk contents. The next fdatasync() call would succeed, but no
1218 * further writeback attempt will be made. We can't get back to a state
1219 * in which we know what is on disk (we would have to rewrite
1220 * everything that was touched since the last fdatasync() at least), so
1221 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1222 * really defined, I have little hope that other OSes are doing better.
1224 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1225 * cache. */
1226 if ((s->open_flags & O_DIRECT) == 0) {
1227 s->page_cache_inconsistent = true;
1229 return -errno;
1231 return 0;
1234 #ifdef CONFIG_PREADV
1236 static bool preadv_present = true;
1238 static ssize_t
1239 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1241 return preadv(fd, iov, nr_iov, offset);
1244 static ssize_t
1245 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1247 return pwritev(fd, iov, nr_iov, offset);
1250 #else
1252 static bool preadv_present = false;
1254 static ssize_t
1255 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1257 return -ENOSYS;
1260 static ssize_t
1261 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1263 return -ENOSYS;
1266 #endif
1268 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
1270 ssize_t len;
1272 do {
1273 if (aiocb->aio_type & QEMU_AIO_WRITE)
1274 len = qemu_pwritev(aiocb->aio_fildes,
1275 aiocb->io.iov,
1276 aiocb->io.niov,
1277 aiocb->aio_offset);
1278 else
1279 len = qemu_preadv(aiocb->aio_fildes,
1280 aiocb->io.iov,
1281 aiocb->io.niov,
1282 aiocb->aio_offset);
1283 } while (len == -1 && errno == EINTR);
1285 if (len == -1) {
1286 return -errno;
1288 return len;
1292 * Read/writes the data to/from a given linear buffer.
1294 * Returns the number of bytes handles or -errno in case of an error. Short
1295 * reads are only returned if the end of the file is reached.
1297 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
1299 ssize_t offset = 0;
1300 ssize_t len;
1302 while (offset < aiocb->aio_nbytes) {
1303 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1304 len = pwrite(aiocb->aio_fildes,
1305 (const char *)buf + offset,
1306 aiocb->aio_nbytes - offset,
1307 aiocb->aio_offset + offset);
1308 } else {
1309 len = pread(aiocb->aio_fildes,
1310 buf + offset,
1311 aiocb->aio_nbytes - offset,
1312 aiocb->aio_offset + offset);
1314 if (len == -1 && errno == EINTR) {
1315 continue;
1316 } else if (len == -1 && errno == EINVAL &&
1317 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
1318 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
1319 offset > 0) {
1320 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1321 * after a short read. Assume that O_DIRECT short reads only occur
1322 * at EOF. Therefore this is a short read, not an I/O error.
1324 break;
1325 } else if (len == -1) {
1326 offset = -errno;
1327 break;
1328 } else if (len == 0) {
1329 break;
1331 offset += len;
1334 return offset;
1337 static int handle_aiocb_rw(void *opaque)
1339 RawPosixAIOData *aiocb = opaque;
1340 ssize_t nbytes;
1341 char *buf;
1343 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
1345 * If there is just a single buffer, and it is properly aligned
1346 * we can just use plain pread/pwrite without any problems.
1348 if (aiocb->io.niov == 1) {
1349 nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base);
1350 goto out;
1353 * We have more than one iovec, and all are properly aligned.
1355 * Try preadv/pwritev first and fall back to linearizing the
1356 * buffer if it's not supported.
1358 if (preadv_present) {
1359 nbytes = handle_aiocb_rw_vector(aiocb);
1360 if (nbytes == aiocb->aio_nbytes ||
1361 (nbytes < 0 && nbytes != -ENOSYS)) {
1362 goto out;
1364 preadv_present = false;
1368 * XXX(hch): short read/write. no easy way to handle the reminder
1369 * using these interfaces. For now retry using plain
1370 * pread/pwrite?
1375 * Ok, we have to do it the hard way, copy all segments into
1376 * a single aligned buffer.
1378 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
1379 if (buf == NULL) {
1380 nbytes = -ENOMEM;
1381 goto out;
1384 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1385 char *p = buf;
1386 int i;
1388 for (i = 0; i < aiocb->io.niov; ++i) {
1389 memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len);
1390 p += aiocb->io.iov[i].iov_len;
1392 assert(p - buf == aiocb->aio_nbytes);
1395 nbytes = handle_aiocb_rw_linear(aiocb, buf);
1396 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
1397 char *p = buf;
1398 size_t count = aiocb->aio_nbytes, copy;
1399 int i;
1401 for (i = 0; i < aiocb->io.niov && count; ++i) {
1402 copy = count;
1403 if (copy > aiocb->io.iov[i].iov_len) {
1404 copy = aiocb->io.iov[i].iov_len;
1406 memcpy(aiocb->io.iov[i].iov_base, p, copy);
1407 assert(count >= copy);
1408 p += copy;
1409 count -= copy;
1411 assert(count == 0);
1413 qemu_vfree(buf);
1415 out:
1416 if (nbytes == aiocb->aio_nbytes) {
1417 return 0;
1418 } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) {
1419 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1420 return -EINVAL;
1421 } else {
1422 iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes,
1423 0, aiocb->aio_nbytes - nbytes);
1424 return 0;
1426 } else {
1427 assert(nbytes < 0);
1428 return nbytes;
1432 #ifdef CONFIG_XFS
1433 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
1435 struct xfs_flock64 fl;
1436 int err;
1438 memset(&fl, 0, sizeof(fl));
1439 fl.l_whence = SEEK_SET;
1440 fl.l_start = offset;
1441 fl.l_len = bytes;
1443 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
1444 err = errno;
1445 trace_file_xfs_write_zeroes(strerror(errno));
1446 return -err;
1449 return 0;
1452 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
1454 struct xfs_flock64 fl;
1455 int err;
1457 memset(&fl, 0, sizeof(fl));
1458 fl.l_whence = SEEK_SET;
1459 fl.l_start = offset;
1460 fl.l_len = bytes;
1462 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
1463 err = errno;
1464 trace_file_xfs_discard(strerror(errno));
1465 return -err;
1468 return 0;
1470 #endif
1472 static int translate_err(int err)
1474 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1475 err == -ENOTTY) {
1476 err = -ENOTSUP;
1478 return err;
1481 #ifdef CONFIG_FALLOCATE
1482 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1484 do {
1485 if (fallocate(fd, mode, offset, len) == 0) {
1486 return 0;
1488 } while (errno == EINTR);
1489 return translate_err(-errno);
1491 #endif
1493 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1495 int ret = -ENOTSUP;
1496 BDRVRawState *s = aiocb->bs->opaque;
1498 if (!s->has_write_zeroes) {
1499 return -ENOTSUP;
1502 #ifdef BLKZEROOUT
1503 do {
1504 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1505 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1506 return 0;
1508 } while (errno == EINTR);
1510 ret = translate_err(-errno);
1511 #endif
1513 if (ret == -ENOTSUP) {
1514 s->has_write_zeroes = false;
1516 return ret;
1519 static int handle_aiocb_write_zeroes(void *opaque)
1521 RawPosixAIOData *aiocb = opaque;
1522 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1523 BDRVRawState *s = aiocb->bs->opaque;
1524 #endif
1525 #ifdef CONFIG_FALLOCATE
1526 int64_t len;
1527 #endif
1529 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1530 return handle_aiocb_write_zeroes_block(aiocb);
1533 #ifdef CONFIG_XFS
1534 if (s->is_xfs) {
1535 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
1537 #endif
1539 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1540 if (s->has_write_zeroes) {
1541 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1542 aiocb->aio_offset, aiocb->aio_nbytes);
1543 if (ret == 0 || ret != -ENOTSUP) {
1544 return ret;
1546 s->has_write_zeroes = false;
1548 #endif
1550 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1551 if (s->has_discard && s->has_fallocate) {
1552 int ret = do_fallocate(s->fd,
1553 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1554 aiocb->aio_offset, aiocb->aio_nbytes);
1555 if (ret == 0) {
1556 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1557 if (ret == 0 || ret != -ENOTSUP) {
1558 return ret;
1560 s->has_fallocate = false;
1561 } else if (ret != -ENOTSUP) {
1562 return ret;
1563 } else {
1564 s->has_discard = false;
1567 #endif
1569 #ifdef CONFIG_FALLOCATE
1570 /* Last resort: we are trying to extend the file with zeroed data. This
1571 * can be done via fallocate(fd, 0) */
1572 len = bdrv_getlength(aiocb->bs);
1573 if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) {
1574 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1575 if (ret == 0 || ret != -ENOTSUP) {
1576 return ret;
1578 s->has_fallocate = false;
1580 #endif
1582 return -ENOTSUP;
1585 static int handle_aiocb_write_zeroes_unmap(void *opaque)
1587 RawPosixAIOData *aiocb = opaque;
1588 BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque;
1589 int ret;
1591 /* First try to write zeros and unmap at the same time */
1593 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1594 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1595 aiocb->aio_offset, aiocb->aio_nbytes);
1596 if (ret != -ENOTSUP) {
1597 return ret;
1599 #endif
1601 #ifdef CONFIG_XFS
1602 if (s->is_xfs) {
1603 /* xfs_discard() guarantees that the discarded area reads as all-zero
1604 * afterwards, so we can use it here. */
1605 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1607 #endif
1609 /* If we couldn't manage to unmap while guaranteed that the area reads as
1610 * all-zero afterwards, just write zeroes without unmapping */
1611 ret = handle_aiocb_write_zeroes(aiocb);
1612 return ret;
1615 #ifndef HAVE_COPY_FILE_RANGE
1617 /* Avoid conflict with system header file. */
1618 #define copy_file_range(a, b, c, d, e, f) local_copy_file_range(a, b, c, d, e, f)
1620 static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd,
1621 off_t *out_off, size_t len, unsigned int flags)
1623 #ifdef __NR_copy_file_range
1624 return syscall(__NR_copy_file_range, in_fd, in_off, out_fd,
1625 out_off, len, flags);
1626 #else
1627 errno = ENOSYS;
1628 return -1;
1629 #endif
1631 #endif
1633 static int handle_aiocb_copy_range(void *opaque)
1635 RawPosixAIOData *aiocb = opaque;
1636 uint64_t bytes = aiocb->aio_nbytes;
1637 off_t in_off = aiocb->aio_offset;
1638 off_t out_off = aiocb->copy_range.aio_offset2;
1640 while (bytes) {
1641 ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off,
1642 aiocb->copy_range.aio_fd2, &out_off,
1643 bytes, 0);
1644 trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off,
1645 aiocb->copy_range.aio_fd2, out_off, bytes,
1646 0, ret);
1647 if (ret == 0) {
1648 /* No progress (e.g. when beyond EOF), let the caller fall back to
1649 * buffer I/O. */
1650 return -ENOSPC;
1652 if (ret < 0) {
1653 switch (errno) {
1654 case ENOSYS:
1655 return -ENOTSUP;
1656 case EINTR:
1657 continue;
1658 default:
1659 return -errno;
1662 bytes -= ret;
1664 return 0;
1667 static int handle_aiocb_discard(void *opaque)
1669 RawPosixAIOData *aiocb = opaque;
1670 int ret = -EOPNOTSUPP;
1671 BDRVRawState *s = aiocb->bs->opaque;
1673 if (!s->has_discard) {
1674 return -ENOTSUP;
1677 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1678 #ifdef BLKDISCARD
1679 do {
1680 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1681 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1682 return 0;
1684 } while (errno == EINTR);
1686 ret = -errno;
1687 #endif
1688 } else {
1689 #ifdef CONFIG_XFS
1690 if (s->is_xfs) {
1691 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1693 #endif
1695 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1696 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1697 aiocb->aio_offset, aiocb->aio_nbytes);
1698 #endif
1701 ret = translate_err(ret);
1702 if (ret == -ENOTSUP) {
1703 s->has_discard = false;
1705 return ret;
1708 static int handle_aiocb_truncate(void *opaque)
1710 RawPosixAIOData *aiocb = opaque;
1711 int result = 0;
1712 int64_t current_length = 0;
1713 char *buf = NULL;
1714 struct stat st;
1715 int fd = aiocb->aio_fildes;
1716 int64_t offset = aiocb->aio_offset;
1717 PreallocMode prealloc = aiocb->truncate.prealloc;
1718 Error **errp = aiocb->truncate.errp;
1720 if (fstat(fd, &st) < 0) {
1721 result = -errno;
1722 error_setg_errno(errp, -result, "Could not stat file");
1723 return result;
1726 current_length = st.st_size;
1727 if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1728 error_setg(errp, "Cannot use preallocation for shrinking files");
1729 return -ENOTSUP;
1732 switch (prealloc) {
1733 #ifdef CONFIG_POSIX_FALLOCATE
1734 case PREALLOC_MODE_FALLOC:
1736 * Truncating before posix_fallocate() makes it about twice slower on
1737 * file systems that do not support fallocate(), trying to check if a
1738 * block is allocated before allocating it, so don't do that here.
1740 if (offset != current_length) {
1741 result = -posix_fallocate(fd, current_length,
1742 offset - current_length);
1743 if (result != 0) {
1744 /* posix_fallocate() doesn't set errno. */
1745 error_setg_errno(errp, -result,
1746 "Could not preallocate new data");
1748 } else {
1749 result = 0;
1751 goto out;
1752 #endif
1753 case PREALLOC_MODE_FULL:
1755 int64_t num = 0, left = offset - current_length;
1756 off_t seek_result;
1759 * Knowing the final size from the beginning could allow the file
1760 * system driver to do less allocations and possibly avoid
1761 * fragmentation of the file.
1763 if (ftruncate(fd, offset) != 0) {
1764 result = -errno;
1765 error_setg_errno(errp, -result, "Could not resize file");
1766 goto out;
1769 buf = g_malloc0(65536);
1771 seek_result = lseek(fd, current_length, SEEK_SET);
1772 if (seek_result < 0) {
1773 result = -errno;
1774 error_setg_errno(errp, -result,
1775 "Failed to seek to the old end of file");
1776 goto out;
1779 while (left > 0) {
1780 num = MIN(left, 65536);
1781 result = write(fd, buf, num);
1782 if (result < 0) {
1783 if (errno == EINTR) {
1784 continue;
1786 result = -errno;
1787 error_setg_errno(errp, -result,
1788 "Could not write zeros for preallocation");
1789 goto out;
1791 left -= result;
1793 if (result >= 0) {
1794 result = fsync(fd);
1795 if (result < 0) {
1796 result = -errno;
1797 error_setg_errno(errp, -result,
1798 "Could not flush file to disk");
1799 goto out;
1802 goto out;
1804 case PREALLOC_MODE_OFF:
1805 if (ftruncate(fd, offset) != 0) {
1806 result = -errno;
1807 error_setg_errno(errp, -result, "Could not resize file");
1809 return result;
1810 default:
1811 result = -ENOTSUP;
1812 error_setg(errp, "Unsupported preallocation mode: %s",
1813 PreallocMode_str(prealloc));
1814 return result;
1817 out:
1818 if (result < 0) {
1819 if (ftruncate(fd, current_length) < 0) {
1820 error_report("Failed to restore old file length: %s",
1821 strerror(errno));
1825 g_free(buf);
1826 return result;
1829 static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs,
1830 ThreadPoolFunc func, void *arg)
1832 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1833 ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1834 return thread_pool_submit_co(pool, func, arg);
1837 static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset,
1838 uint64_t bytes, QEMUIOVector *qiov, int type)
1840 BDRVRawState *s = bs->opaque;
1841 RawPosixAIOData acb;
1843 if (fd_open(bs) < 0)
1844 return -EIO;
1847 * Check if the underlying device requires requests to be aligned,
1848 * and if the request we are trying to submit is aligned or not.
1849 * If this is the case tell the low-level driver that it needs
1850 * to copy the buffer.
1852 if (s->needs_alignment) {
1853 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1854 type |= QEMU_AIO_MISALIGNED;
1855 #ifdef CONFIG_LINUX_AIO
1856 } else if (s->use_linux_aio) {
1857 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1858 assert(qiov->size == bytes);
1859 return laio_co_submit(bs, aio, s->fd, offset, qiov, type);
1860 #endif
1864 acb = (RawPosixAIOData) {
1865 .bs = bs,
1866 .aio_fildes = s->fd,
1867 .aio_type = type,
1868 .aio_offset = offset,
1869 .aio_nbytes = bytes,
1870 .io = {
1871 .iov = qiov->iov,
1872 .niov = qiov->niov,
1876 assert(qiov->size == bytes);
1877 return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb);
1880 static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset,
1881 uint64_t bytes, QEMUIOVector *qiov,
1882 int flags)
1884 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ);
1887 static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
1888 uint64_t bytes, QEMUIOVector *qiov,
1889 int flags)
1891 assert(flags == 0);
1892 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE);
1895 static void raw_aio_plug(BlockDriverState *bs)
1897 #ifdef CONFIG_LINUX_AIO
1898 BDRVRawState *s = bs->opaque;
1899 if (s->use_linux_aio) {
1900 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1901 laio_io_plug(bs, aio);
1903 #endif
1906 static void raw_aio_unplug(BlockDriverState *bs)
1908 #ifdef CONFIG_LINUX_AIO
1909 BDRVRawState *s = bs->opaque;
1910 if (s->use_linux_aio) {
1911 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1912 laio_io_unplug(bs, aio);
1914 #endif
1917 static int raw_co_flush_to_disk(BlockDriverState *bs)
1919 BDRVRawState *s = bs->opaque;
1920 RawPosixAIOData acb;
1921 int ret;
1923 ret = fd_open(bs);
1924 if (ret < 0) {
1925 return ret;
1928 acb = (RawPosixAIOData) {
1929 .bs = bs,
1930 .aio_fildes = s->fd,
1931 .aio_type = QEMU_AIO_FLUSH,
1934 return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb);
1937 static void raw_aio_attach_aio_context(BlockDriverState *bs,
1938 AioContext *new_context)
1940 #ifdef CONFIG_LINUX_AIO
1941 BDRVRawState *s = bs->opaque;
1942 if (s->use_linux_aio) {
1943 Error *local_err;
1944 if (!aio_setup_linux_aio(new_context, &local_err)) {
1945 error_reportf_err(local_err, "Unable to use native AIO, "
1946 "falling back to thread pool: ");
1947 s->use_linux_aio = false;
1950 #endif
1953 static void raw_close(BlockDriverState *bs)
1955 BDRVRawState *s = bs->opaque;
1957 if (s->fd >= 0) {
1958 qemu_close(s->fd);
1959 s->fd = -1;
1964 * Truncates the given regular file @fd to @offset and, when growing, fills the
1965 * new space according to @prealloc.
1967 * Returns: 0 on success, -errno on failure.
1969 static int coroutine_fn
1970 raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset,
1971 PreallocMode prealloc, Error **errp)
1973 RawPosixAIOData acb;
1975 acb = (RawPosixAIOData) {
1976 .bs = bs,
1977 .aio_fildes = fd,
1978 .aio_type = QEMU_AIO_TRUNCATE,
1979 .aio_offset = offset,
1980 .truncate = {
1981 .prealloc = prealloc,
1982 .errp = errp,
1986 return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb);
1989 static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset,
1990 PreallocMode prealloc, Error **errp)
1992 BDRVRawState *s = bs->opaque;
1993 struct stat st;
1994 int ret;
1996 if (fstat(s->fd, &st)) {
1997 ret = -errno;
1998 error_setg_errno(errp, -ret, "Failed to fstat() the file");
1999 return ret;
2002 if (S_ISREG(st.st_mode)) {
2003 return raw_regular_truncate(bs, s->fd, offset, prealloc, errp);
2006 if (prealloc != PREALLOC_MODE_OFF) {
2007 error_setg(errp, "Preallocation mode '%s' unsupported for this "
2008 "non-regular file", PreallocMode_str(prealloc));
2009 return -ENOTSUP;
2012 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2013 if (offset > raw_getlength(bs)) {
2014 error_setg(errp, "Cannot grow device files");
2015 return -EINVAL;
2017 } else {
2018 error_setg(errp, "Resizing this file is not supported");
2019 return -ENOTSUP;
2022 return 0;
2025 #ifdef __OpenBSD__
2026 static int64_t raw_getlength(BlockDriverState *bs)
2028 BDRVRawState *s = bs->opaque;
2029 int fd = s->fd;
2030 struct stat st;
2032 if (fstat(fd, &st))
2033 return -errno;
2034 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2035 struct disklabel dl;
2037 if (ioctl(fd, DIOCGDINFO, &dl))
2038 return -errno;
2039 return (uint64_t)dl.d_secsize *
2040 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
2041 } else
2042 return st.st_size;
2044 #elif defined(__NetBSD__)
2045 static int64_t raw_getlength(BlockDriverState *bs)
2047 BDRVRawState *s = bs->opaque;
2048 int fd = s->fd;
2049 struct stat st;
2051 if (fstat(fd, &st))
2052 return -errno;
2053 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2054 struct dkwedge_info dkw;
2056 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
2057 return dkw.dkw_size * 512;
2058 } else {
2059 struct disklabel dl;
2061 if (ioctl(fd, DIOCGDINFO, &dl))
2062 return -errno;
2063 return (uint64_t)dl.d_secsize *
2064 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
2066 } else
2067 return st.st_size;
2069 #elif defined(__sun__)
2070 static int64_t raw_getlength(BlockDriverState *bs)
2072 BDRVRawState *s = bs->opaque;
2073 struct dk_minfo minfo;
2074 int ret;
2075 int64_t size;
2077 ret = fd_open(bs);
2078 if (ret < 0) {
2079 return ret;
2083 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2085 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
2086 if (ret != -1) {
2087 return minfo.dki_lbsize * minfo.dki_capacity;
2091 * There are reports that lseek on some devices fails, but
2092 * irc discussion said that contingency on contingency was overkill.
2094 size = lseek(s->fd, 0, SEEK_END);
2095 if (size < 0) {
2096 return -errno;
2098 return size;
2100 #elif defined(CONFIG_BSD)
2101 static int64_t raw_getlength(BlockDriverState *bs)
2103 BDRVRawState *s = bs->opaque;
2104 int fd = s->fd;
2105 int64_t size;
2106 struct stat sb;
2107 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2108 int reopened = 0;
2109 #endif
2110 int ret;
2112 ret = fd_open(bs);
2113 if (ret < 0)
2114 return ret;
2116 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2117 again:
2118 #endif
2119 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
2120 #ifdef DIOCGMEDIASIZE
2121 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
2122 #elif defined(DIOCGPART)
2124 struct partinfo pi;
2125 if (ioctl(fd, DIOCGPART, &pi) == 0)
2126 size = pi.media_size;
2127 else
2128 size = 0;
2130 if (size == 0)
2131 #endif
2132 #if defined(__APPLE__) && defined(__MACH__)
2134 uint64_t sectors = 0;
2135 uint32_t sector_size = 0;
2137 if (ioctl(fd, DKIOCGETBLOCKCOUNT, &sectors) == 0
2138 && ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) == 0) {
2139 size = sectors * sector_size;
2140 } else {
2141 size = lseek(fd, 0LL, SEEK_END);
2142 if (size < 0) {
2143 return -errno;
2147 #else
2148 size = lseek(fd, 0LL, SEEK_END);
2149 if (size < 0) {
2150 return -errno;
2152 #endif
2153 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2154 switch(s->type) {
2155 case FTYPE_CD:
2156 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2157 if (size == 2048LL * (unsigned)-1)
2158 size = 0;
2159 /* XXX no disc? maybe we need to reopen... */
2160 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
2161 reopened = 1;
2162 goto again;
2165 #endif
2166 } else {
2167 size = lseek(fd, 0, SEEK_END);
2168 if (size < 0) {
2169 return -errno;
2172 return size;
2174 #else
2175 static int64_t raw_getlength(BlockDriverState *bs)
2177 BDRVRawState *s = bs->opaque;
2178 int ret;
2179 int64_t size;
2181 ret = fd_open(bs);
2182 if (ret < 0) {
2183 return ret;
2186 size = lseek(s->fd, 0, SEEK_END);
2187 if (size < 0) {
2188 return -errno;
2190 return size;
2192 #endif
2194 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
2196 struct stat st;
2197 BDRVRawState *s = bs->opaque;
2199 if (fstat(s->fd, &st) < 0) {
2200 return -errno;
2202 return (int64_t)st.st_blocks * 512;
2205 static int coroutine_fn
2206 raw_co_create(BlockdevCreateOptions *options, Error **errp)
2208 BlockdevCreateOptionsFile *file_opts;
2209 Error *local_err = NULL;
2210 int fd;
2211 uint64_t perm, shared;
2212 int result = 0;
2214 /* Validate options and set default values */
2215 assert(options->driver == BLOCKDEV_DRIVER_FILE);
2216 file_opts = &options->u.file;
2218 if (!file_opts->has_nocow) {
2219 file_opts->nocow = false;
2221 if (!file_opts->has_preallocation) {
2222 file_opts->preallocation = PREALLOC_MODE_OFF;
2225 /* Create file */
2226 fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_BINARY, 0644);
2227 if (fd < 0) {
2228 result = -errno;
2229 error_setg_errno(errp, -result, "Could not create file");
2230 goto out;
2233 /* Take permissions: We want to discard everything, so we need
2234 * BLK_PERM_WRITE; and truncation to the desired size requires
2235 * BLK_PERM_RESIZE.
2236 * On the other hand, we cannot share the RESIZE permission
2237 * because we promise that after this function, the file has the
2238 * size given in the options. If someone else were to resize it
2239 * concurrently, we could not guarantee that.
2240 * Note that after this function, we can no longer guarantee that
2241 * the file is not touched by a third party, so it may be resized
2242 * then. */
2243 perm = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2244 shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE;
2246 /* Step one: Take locks */
2247 result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp);
2248 if (result < 0) {
2249 goto out_close;
2252 /* Step two: Check that nobody else has taken conflicting locks */
2253 result = raw_check_lock_bytes(fd, perm, shared, errp);
2254 if (result < 0) {
2255 error_append_hint(errp,
2256 "Is another process using the image [%s]?\n",
2257 file_opts->filename);
2258 goto out_unlock;
2261 /* Clear the file by truncating it to 0 */
2262 result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp);
2263 if (result < 0) {
2264 goto out_unlock;
2267 if (file_opts->nocow) {
2268 #ifdef __linux__
2269 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2270 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2271 * will be ignored since any failure of this operation should not
2272 * block the left work.
2274 int attr;
2275 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
2276 attr |= FS_NOCOW_FL;
2277 ioctl(fd, FS_IOC_SETFLAGS, &attr);
2279 #endif
2282 /* Resize and potentially preallocate the file to the desired
2283 * final size */
2284 result = raw_regular_truncate(NULL, fd, file_opts->size,
2285 file_opts->preallocation, errp);
2286 if (result < 0) {
2287 goto out_unlock;
2290 out_unlock:
2291 raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err);
2292 if (local_err) {
2293 /* The above call should not fail, and if it does, that does
2294 * not mean the whole creation operation has failed. So
2295 * report it the user for their convenience, but do not report
2296 * it to the caller. */
2297 warn_report_err(local_err);
2300 out_close:
2301 if (qemu_close(fd) != 0 && result == 0) {
2302 result = -errno;
2303 error_setg_errno(errp, -result, "Could not close the new file");
2305 out:
2306 return result;
2309 static int coroutine_fn raw_co_create_opts(const char *filename, QemuOpts *opts,
2310 Error **errp)
2312 BlockdevCreateOptions options;
2313 int64_t total_size = 0;
2314 bool nocow = false;
2315 PreallocMode prealloc;
2316 char *buf = NULL;
2317 Error *local_err = NULL;
2319 /* Skip file: protocol prefix */
2320 strstart(filename, "file:", &filename);
2322 /* Read out options */
2323 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2324 BDRV_SECTOR_SIZE);
2325 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
2326 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2327 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
2328 PREALLOC_MODE_OFF, &local_err);
2329 g_free(buf);
2330 if (local_err) {
2331 error_propagate(errp, local_err);
2332 return -EINVAL;
2335 options = (BlockdevCreateOptions) {
2336 .driver = BLOCKDEV_DRIVER_FILE,
2337 .u.file = {
2338 .filename = (char *) filename,
2339 .size = total_size,
2340 .has_preallocation = true,
2341 .preallocation = prealloc,
2342 .has_nocow = true,
2343 .nocow = nocow,
2346 return raw_co_create(&options, errp);
2350 * Find allocation range in @bs around offset @start.
2351 * May change underlying file descriptor's file offset.
2352 * If @start is not in a hole, store @start in @data, and the
2353 * beginning of the next hole in @hole, and return 0.
2354 * If @start is in a non-trailing hole, store @start in @hole and the
2355 * beginning of the next non-hole in @data, and return 0.
2356 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2357 * If we can't find out, return a negative errno other than -ENXIO.
2359 static int find_allocation(BlockDriverState *bs, off_t start,
2360 off_t *data, off_t *hole)
2362 #if defined SEEK_HOLE && defined SEEK_DATA
2363 BDRVRawState *s = bs->opaque;
2364 off_t offs;
2367 * SEEK_DATA cases:
2368 * D1. offs == start: start is in data
2369 * D2. offs > start: start is in a hole, next data at offs
2370 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2371 * or start is beyond EOF
2372 * If the latter happens, the file has been truncated behind
2373 * our back since we opened it. All bets are off then.
2374 * Treating like a trailing hole is simplest.
2375 * D4. offs < 0, errno != ENXIO: we learned nothing
2377 offs = lseek(s->fd, start, SEEK_DATA);
2378 if (offs < 0) {
2379 return -errno; /* D3 or D4 */
2382 if (offs < start) {
2383 /* This is not a valid return by lseek(). We are safe to just return
2384 * -EIO in this case, and we'll treat it like D4. */
2385 return -EIO;
2388 if (offs > start) {
2389 /* D2: in hole, next data at offs */
2390 *hole = start;
2391 *data = offs;
2392 return 0;
2395 /* D1: in data, end not yet known */
2398 * SEEK_HOLE cases:
2399 * H1. offs == start: start is in a hole
2400 * If this happens here, a hole has been dug behind our back
2401 * since the previous lseek().
2402 * H2. offs > start: either start is in data, next hole at offs,
2403 * or start is in trailing hole, EOF at offs
2404 * Linux treats trailing holes like any other hole: offs ==
2405 * start. Solaris seeks to EOF instead: offs > start (blech).
2406 * If that happens here, a hole has been dug behind our back
2407 * since the previous lseek().
2408 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2409 * If this happens, the file has been truncated behind our
2410 * back since we opened it. Treat it like a trailing hole.
2411 * H4. offs < 0, errno != ENXIO: we learned nothing
2412 * Pretend we know nothing at all, i.e. "forget" about D1.
2414 offs = lseek(s->fd, start, SEEK_HOLE);
2415 if (offs < 0) {
2416 return -errno; /* D1 and (H3 or H4) */
2419 if (offs < start) {
2420 /* This is not a valid return by lseek(). We are safe to just return
2421 * -EIO in this case, and we'll treat it like H4. */
2422 return -EIO;
2425 if (offs > start) {
2427 * D1 and H2: either in data, next hole at offs, or it was in
2428 * data but is now in a trailing hole. In the latter case,
2429 * all bets are off. Treating it as if it there was data all
2430 * the way to EOF is safe, so simply do that.
2432 *data = start;
2433 *hole = offs;
2434 return 0;
2437 /* D1 and H1 */
2438 return -EBUSY;
2439 #else
2440 return -ENOTSUP;
2441 #endif
2445 * Returns the allocation status of the specified offset.
2447 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2449 * 'pnum' is set to the number of bytes (including and immediately following
2450 * the specified offset) that are known to be in the same
2451 * allocated/unallocated state.
2453 * 'bytes' is the max value 'pnum' should be set to.
2455 static int coroutine_fn raw_co_block_status(BlockDriverState *bs,
2456 bool want_zero,
2457 int64_t offset,
2458 int64_t bytes, int64_t *pnum,
2459 int64_t *map,
2460 BlockDriverState **file)
2462 off_t data = 0, hole = 0;
2463 int ret;
2465 ret = fd_open(bs);
2466 if (ret < 0) {
2467 return ret;
2470 if (!want_zero) {
2471 *pnum = bytes;
2472 *map = offset;
2473 *file = bs;
2474 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
2477 ret = find_allocation(bs, offset, &data, &hole);
2478 if (ret == -ENXIO) {
2479 /* Trailing hole */
2480 *pnum = bytes;
2481 ret = BDRV_BLOCK_ZERO;
2482 } else if (ret < 0) {
2483 /* No info available, so pretend there are no holes */
2484 *pnum = bytes;
2485 ret = BDRV_BLOCK_DATA;
2486 } else if (data == offset) {
2487 /* On a data extent, compute bytes to the end of the extent,
2488 * possibly including a partial sector at EOF. */
2489 *pnum = MIN(bytes, hole - offset);
2490 ret = BDRV_BLOCK_DATA;
2491 } else {
2492 /* On a hole, compute bytes to the beginning of the next extent. */
2493 assert(hole == offset);
2494 *pnum = MIN(bytes, data - offset);
2495 ret = BDRV_BLOCK_ZERO;
2497 *map = offset;
2498 *file = bs;
2499 return ret | BDRV_BLOCK_OFFSET_VALID;
2502 #if defined(__linux__)
2503 /* Verify that the file is not in the page cache */
2504 static void check_cache_dropped(BlockDriverState *bs, Error **errp)
2506 const size_t window_size = 128 * 1024 * 1024;
2507 BDRVRawState *s = bs->opaque;
2508 void *window = NULL;
2509 size_t length = 0;
2510 unsigned char *vec;
2511 size_t page_size;
2512 off_t offset;
2513 off_t end;
2515 /* mincore(2) page status information requires 1 byte per page */
2516 page_size = sysconf(_SC_PAGESIZE);
2517 vec = g_malloc(DIV_ROUND_UP(window_size, page_size));
2519 end = raw_getlength(bs);
2521 for (offset = 0; offset < end; offset += window_size) {
2522 void *new_window;
2523 size_t new_length;
2524 size_t vec_end;
2525 size_t i;
2526 int ret;
2528 /* Unmap previous window if size has changed */
2529 new_length = MIN(end - offset, window_size);
2530 if (new_length != length) {
2531 munmap(window, length);
2532 window = NULL;
2533 length = 0;
2536 new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE,
2537 s->fd, offset);
2538 if (new_window == MAP_FAILED) {
2539 error_setg_errno(errp, errno, "mmap failed");
2540 break;
2543 window = new_window;
2544 length = new_length;
2546 ret = mincore(window, length, vec);
2547 if (ret < 0) {
2548 error_setg_errno(errp, errno, "mincore failed");
2549 break;
2552 vec_end = DIV_ROUND_UP(length, page_size);
2553 for (i = 0; i < vec_end; i++) {
2554 if (vec[i] & 0x1) {
2555 error_setg(errp, "page cache still in use!");
2556 break;
2561 if (window) {
2562 munmap(window, length);
2565 g_free(vec);
2567 #endif /* __linux__ */
2569 static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs,
2570 Error **errp)
2572 BDRVRawState *s = bs->opaque;
2573 int ret;
2575 ret = fd_open(bs);
2576 if (ret < 0) {
2577 error_setg_errno(errp, -ret, "The file descriptor is not open");
2578 return;
2581 if (!s->drop_cache) {
2582 return;
2585 if (s->open_flags & O_DIRECT) {
2586 return; /* No host kernel page cache */
2589 #if defined(__linux__)
2590 /* This sets the scene for the next syscall... */
2591 ret = bdrv_co_flush(bs);
2592 if (ret < 0) {
2593 error_setg_errno(errp, -ret, "flush failed");
2594 return;
2597 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2598 * process. These limitations are okay because we just fsynced the file,
2599 * we don't use mmap, and the file should not be in use by other processes.
2601 ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED);
2602 if (ret != 0) { /* the return value is a positive errno */
2603 error_setg_errno(errp, ret, "fadvise failed");
2604 return;
2607 if (s->check_cache_dropped) {
2608 check_cache_dropped(bs, errp);
2610 #else /* __linux__ */
2611 /* Do nothing. Live migration to a remote host with cache.direct=off is
2612 * unsupported on other host operating systems. Cache consistency issues
2613 * may occur but no error is reported here, partly because that's the
2614 * historical behavior and partly because it's hard to differentiate valid
2615 * configurations that should not cause errors.
2617 #endif /* !__linux__ */
2620 static coroutine_fn int
2621 raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev)
2623 BDRVRawState *s = bs->opaque;
2624 RawPosixAIOData acb;
2626 acb = (RawPosixAIOData) {
2627 .bs = bs,
2628 .aio_fildes = s->fd,
2629 .aio_type = QEMU_AIO_DISCARD,
2630 .aio_offset = offset,
2631 .aio_nbytes = bytes,
2634 if (blkdev) {
2635 acb.aio_type |= QEMU_AIO_BLKDEV;
2638 return raw_thread_pool_submit(bs, handle_aiocb_discard, &acb);
2641 static coroutine_fn int
2642 raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
2644 return raw_do_pdiscard(bs, offset, bytes, false);
2647 static int coroutine_fn
2648 raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes,
2649 BdrvRequestFlags flags, bool blkdev)
2651 BDRVRawState *s = bs->opaque;
2652 RawPosixAIOData acb;
2653 ThreadPoolFunc *handler;
2655 acb = (RawPosixAIOData) {
2656 .bs = bs,
2657 .aio_fildes = s->fd,
2658 .aio_type = QEMU_AIO_WRITE_ZEROES,
2659 .aio_offset = offset,
2660 .aio_nbytes = bytes,
2663 if (blkdev) {
2664 acb.aio_type |= QEMU_AIO_BLKDEV;
2667 if (flags & BDRV_REQ_MAY_UNMAP) {
2668 acb.aio_type |= QEMU_AIO_DISCARD;
2669 handler = handle_aiocb_write_zeroes_unmap;
2670 } else {
2671 handler = handle_aiocb_write_zeroes;
2674 return raw_thread_pool_submit(bs, handler, &acb);
2677 static int coroutine_fn raw_co_pwrite_zeroes(
2678 BlockDriverState *bs, int64_t offset,
2679 int bytes, BdrvRequestFlags flags)
2681 return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false);
2684 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2686 BDRVRawState *s = bs->opaque;
2688 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
2689 return 0;
2692 static QemuOptsList raw_create_opts = {
2693 .name = "raw-create-opts",
2694 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
2695 .desc = {
2697 .name = BLOCK_OPT_SIZE,
2698 .type = QEMU_OPT_SIZE,
2699 .help = "Virtual disk size"
2702 .name = BLOCK_OPT_NOCOW,
2703 .type = QEMU_OPT_BOOL,
2704 .help = "Turn off copy-on-write (valid only on btrfs)"
2707 .name = BLOCK_OPT_PREALLOC,
2708 .type = QEMU_OPT_STRING,
2709 .help = "Preallocation mode (allowed values: off, falloc, full)"
2711 { /* end of list */ }
2715 static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared,
2716 Error **errp)
2718 BDRVRawState *s = bs->opaque;
2719 BDRVRawReopenState *rs = NULL;
2720 int open_flags;
2721 int ret;
2723 if (s->perm_change_fd) {
2725 * In the context of reopen, this function may be called several times
2726 * (directly and recursively while change permissions of the parent).
2727 * This is even true for children that don't inherit from the original
2728 * reopen node, so s->reopen_state is not set.
2730 * Ignore all but the first call.
2732 return 0;
2735 if (s->reopen_state) {
2736 /* We already have a new file descriptor to set permissions for */
2737 assert(s->reopen_state->perm == perm);
2738 assert(s->reopen_state->shared_perm == shared);
2739 rs = s->reopen_state->opaque;
2740 s->perm_change_fd = rs->fd;
2741 } else {
2742 /* We may need a new fd if auto-read-only switches the mode */
2743 ret = raw_reconfigure_getfd(bs, bs->open_flags, &open_flags, perm,
2744 false, errp);
2745 if (ret < 0) {
2746 return ret;
2747 } else if (ret != s->fd) {
2748 s->perm_change_fd = ret;
2752 /* Prepare permissions on old fd to avoid conflicts between old and new,
2753 * but keep everything locked that new will need. */
2754 ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp);
2755 if (ret < 0) {
2756 goto fail;
2759 /* Copy locks to the new fd */
2760 if (s->perm_change_fd) {
2761 ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared,
2762 false, errp);
2763 if (ret < 0) {
2764 raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
2765 goto fail;
2768 return 0;
2770 fail:
2771 if (s->perm_change_fd && !s->reopen_state) {
2772 qemu_close(s->perm_change_fd);
2774 s->perm_change_fd = 0;
2775 return ret;
2778 static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared)
2780 BDRVRawState *s = bs->opaque;
2782 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
2783 * called after .bdrv_reopen_commit) */
2784 if (s->perm_change_fd && s->fd != s->perm_change_fd) {
2785 qemu_close(s->fd);
2786 s->fd = s->perm_change_fd;
2788 s->perm_change_fd = 0;
2790 raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL);
2791 s->perm = perm;
2792 s->shared_perm = shared;
2795 static void raw_abort_perm_update(BlockDriverState *bs)
2797 BDRVRawState *s = bs->opaque;
2799 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
2800 * the file descriptor. */
2801 if (s->perm_change_fd && !s->reopen_state) {
2802 qemu_close(s->perm_change_fd);
2804 s->perm_change_fd = 0;
2806 raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
2809 static int coroutine_fn raw_co_copy_range_from(
2810 BlockDriverState *bs, BdrvChild *src, uint64_t src_offset,
2811 BdrvChild *dst, uint64_t dst_offset, uint64_t bytes,
2812 BdrvRequestFlags read_flags, BdrvRequestFlags write_flags)
2814 return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes,
2815 read_flags, write_flags);
2818 static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs,
2819 BdrvChild *src,
2820 uint64_t src_offset,
2821 BdrvChild *dst,
2822 uint64_t dst_offset,
2823 uint64_t bytes,
2824 BdrvRequestFlags read_flags,
2825 BdrvRequestFlags write_flags)
2827 RawPosixAIOData acb;
2828 BDRVRawState *s = bs->opaque;
2829 BDRVRawState *src_s;
2831 assert(dst->bs == bs);
2832 if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) {
2833 return -ENOTSUP;
2836 src_s = src->bs->opaque;
2837 if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) {
2838 return -EIO;
2841 acb = (RawPosixAIOData) {
2842 .bs = bs,
2843 .aio_type = QEMU_AIO_COPY_RANGE,
2844 .aio_fildes = src_s->fd,
2845 .aio_offset = src_offset,
2846 .aio_nbytes = bytes,
2847 .copy_range = {
2848 .aio_fd2 = s->fd,
2849 .aio_offset2 = dst_offset,
2853 return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb);
2856 BlockDriver bdrv_file = {
2857 .format_name = "file",
2858 .protocol_name = "file",
2859 .instance_size = sizeof(BDRVRawState),
2860 .bdrv_needs_filename = true,
2861 .bdrv_probe = NULL, /* no probe for protocols */
2862 .bdrv_parse_filename = raw_parse_filename,
2863 .bdrv_file_open = raw_open,
2864 .bdrv_reopen_prepare = raw_reopen_prepare,
2865 .bdrv_reopen_commit = raw_reopen_commit,
2866 .bdrv_reopen_abort = raw_reopen_abort,
2867 .bdrv_close = raw_close,
2868 .bdrv_co_create = raw_co_create,
2869 .bdrv_co_create_opts = raw_co_create_opts,
2870 .bdrv_has_zero_init = bdrv_has_zero_init_1,
2871 .bdrv_co_block_status = raw_co_block_status,
2872 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
2873 .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes,
2875 .bdrv_co_preadv = raw_co_preadv,
2876 .bdrv_co_pwritev = raw_co_pwritev,
2877 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
2878 .bdrv_co_pdiscard = raw_co_pdiscard,
2879 .bdrv_co_copy_range_from = raw_co_copy_range_from,
2880 .bdrv_co_copy_range_to = raw_co_copy_range_to,
2881 .bdrv_refresh_limits = raw_refresh_limits,
2882 .bdrv_io_plug = raw_aio_plug,
2883 .bdrv_io_unplug = raw_aio_unplug,
2884 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
2886 .bdrv_co_truncate = raw_co_truncate,
2887 .bdrv_getlength = raw_getlength,
2888 .bdrv_get_info = raw_get_info,
2889 .bdrv_get_allocated_file_size
2890 = raw_get_allocated_file_size,
2891 .bdrv_check_perm = raw_check_perm,
2892 .bdrv_set_perm = raw_set_perm,
2893 .bdrv_abort_perm_update = raw_abort_perm_update,
2894 .create_opts = &raw_create_opts,
2895 .mutable_opts = mutable_opts,
2898 /***********************************************/
2899 /* host device */
2901 #if defined(__APPLE__) && defined(__MACH__)
2902 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
2903 CFIndex maxPathSize, int flags);
2904 static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
2906 kern_return_t kernResult = KERN_FAILURE;
2907 mach_port_t masterPort;
2908 CFMutableDictionaryRef classesToMatch;
2909 const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
2910 char *mediaType = NULL;
2912 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
2913 if ( KERN_SUCCESS != kernResult ) {
2914 printf( "IOMasterPort returned %d\n", kernResult );
2917 int index;
2918 for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
2919 classesToMatch = IOServiceMatching(matching_array[index]);
2920 if (classesToMatch == NULL) {
2921 error_report("IOServiceMatching returned NULL for %s",
2922 matching_array[index]);
2923 continue;
2925 CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
2926 kCFBooleanTrue);
2927 kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch,
2928 mediaIterator);
2929 if (kernResult != KERN_SUCCESS) {
2930 error_report("Note: IOServiceGetMatchingServices returned %d",
2931 kernResult);
2932 continue;
2935 /* If a match was found, leave the loop */
2936 if (*mediaIterator != 0) {
2937 trace_file_FindEjectableOpticalMedia(matching_array[index]);
2938 mediaType = g_strdup(matching_array[index]);
2939 break;
2942 return mediaType;
2945 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
2946 CFIndex maxPathSize, int flags)
2948 io_object_t nextMedia;
2949 kern_return_t kernResult = KERN_FAILURE;
2950 *bsdPath = '\0';
2951 nextMedia = IOIteratorNext( mediaIterator );
2952 if ( nextMedia )
2954 CFTypeRef bsdPathAsCFString;
2955 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
2956 if ( bsdPathAsCFString ) {
2957 size_t devPathLength;
2958 strcpy( bsdPath, _PATH_DEV );
2959 if (flags & BDRV_O_NOCACHE) {
2960 strcat(bsdPath, "r");
2962 devPathLength = strlen( bsdPath );
2963 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
2964 kernResult = KERN_SUCCESS;
2966 CFRelease( bsdPathAsCFString );
2968 IOObjectRelease( nextMedia );
2971 return kernResult;
2974 /* Sets up a real cdrom for use in QEMU */
2975 static bool setup_cdrom(char *bsd_path, Error **errp)
2977 int index, num_of_test_partitions = 2, fd;
2978 char test_partition[MAXPATHLEN];
2979 bool partition_found = false;
2981 /* look for a working partition */
2982 for (index = 0; index < num_of_test_partitions; index++) {
2983 snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
2984 index);
2985 fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE);
2986 if (fd >= 0) {
2987 partition_found = true;
2988 qemu_close(fd);
2989 break;
2993 /* if a working partition on the device was not found */
2994 if (partition_found == false) {
2995 error_setg(errp, "Failed to find a working partition on disc");
2996 } else {
2997 trace_file_setup_cdrom(test_partition);
2998 pstrcpy(bsd_path, MAXPATHLEN, test_partition);
3000 return partition_found;
3003 /* Prints directions on mounting and unmounting a device */
3004 static void print_unmounting_directions(const char *file_name)
3006 error_report("If device %s is mounted on the desktop, unmount"
3007 " it first before using it in QEMU", file_name);
3008 error_report("Command to unmount device: diskutil unmountDisk %s",
3009 file_name);
3010 error_report("Command to mount device: diskutil mountDisk %s", file_name);
3013 #endif /* defined(__APPLE__) && defined(__MACH__) */
3015 static int hdev_probe_device(const char *filename)
3017 struct stat st;
3019 /* allow a dedicated CD-ROM driver to match with a higher priority */
3020 if (strstart(filename, "/dev/cdrom", NULL))
3021 return 50;
3023 if (stat(filename, &st) >= 0 &&
3024 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
3025 return 100;
3028 return 0;
3031 static int check_hdev_writable(BDRVRawState *s)
3033 #if defined(BLKROGET)
3034 /* Linux block devices can be configured "read-only" using blockdev(8).
3035 * This is independent of device node permissions and therefore open(2)
3036 * with O_RDWR succeeds. Actual writes fail with EPERM.
3038 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
3039 * check for read-only block devices so that Linux block devices behave
3040 * properly.
3042 struct stat st;
3043 int readonly = 0;
3045 if (fstat(s->fd, &st)) {
3046 return -errno;
3049 if (!S_ISBLK(st.st_mode)) {
3050 return 0;
3053 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
3054 return -errno;
3057 if (readonly) {
3058 return -EACCES;
3060 #endif /* defined(BLKROGET) */
3061 return 0;
3064 static void hdev_parse_filename(const char *filename, QDict *options,
3065 Error **errp)
3067 bdrv_parse_filename_strip_prefix(filename, "host_device:", options);
3070 static bool hdev_is_sg(BlockDriverState *bs)
3073 #if defined(__linux__)
3075 BDRVRawState *s = bs->opaque;
3076 struct stat st;
3077 struct sg_scsi_id scsiid;
3078 int sg_version;
3079 int ret;
3081 if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) {
3082 return false;
3085 ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version);
3086 if (ret < 0) {
3087 return false;
3090 ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid);
3091 if (ret >= 0) {
3092 trace_file_hdev_is_sg(scsiid.scsi_type, sg_version);
3093 return true;
3096 #endif
3098 return false;
3101 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
3102 Error **errp)
3104 BDRVRawState *s = bs->opaque;
3105 Error *local_err = NULL;
3106 int ret;
3108 #if defined(__APPLE__) && defined(__MACH__)
3110 * Caution: while qdict_get_str() is fine, getting non-string types
3111 * would require more care. When @options come from -blockdev or
3112 * blockdev_add, its members are typed according to the QAPI
3113 * schema, but when they come from -drive, they're all QString.
3115 const char *filename = qdict_get_str(options, "filename");
3116 char bsd_path[MAXPATHLEN] = "";
3117 bool error_occurred = false;
3119 /* If using a real cdrom */
3120 if (strcmp(filename, "/dev/cdrom") == 0) {
3121 char *mediaType = NULL;
3122 kern_return_t ret_val;
3123 io_iterator_t mediaIterator = 0;
3125 mediaType = FindEjectableOpticalMedia(&mediaIterator);
3126 if (mediaType == NULL) {
3127 error_setg(errp, "Please make sure your CD/DVD is in the optical"
3128 " drive");
3129 error_occurred = true;
3130 goto hdev_open_Mac_error;
3133 ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
3134 if (ret_val != KERN_SUCCESS) {
3135 error_setg(errp, "Could not get BSD path for optical drive");
3136 error_occurred = true;
3137 goto hdev_open_Mac_error;
3140 /* If a real optical drive was not found */
3141 if (bsd_path[0] == '\0') {
3142 error_setg(errp, "Failed to obtain bsd path for optical drive");
3143 error_occurred = true;
3144 goto hdev_open_Mac_error;
3147 /* If using a cdrom disc and finding a partition on the disc failed */
3148 if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
3149 setup_cdrom(bsd_path, errp) == false) {
3150 print_unmounting_directions(bsd_path);
3151 error_occurred = true;
3152 goto hdev_open_Mac_error;
3155 qdict_put_str(options, "filename", bsd_path);
3157 hdev_open_Mac_error:
3158 g_free(mediaType);
3159 if (mediaIterator) {
3160 IOObjectRelease(mediaIterator);
3162 if (error_occurred) {
3163 return -ENOENT;
3166 #endif /* defined(__APPLE__) && defined(__MACH__) */
3168 s->type = FTYPE_FILE;
3170 ret = raw_open_common(bs, options, flags, 0, true, &local_err);
3171 if (ret < 0) {
3172 error_propagate(errp, local_err);
3173 #if defined(__APPLE__) && defined(__MACH__)
3174 if (*bsd_path) {
3175 filename = bsd_path;
3177 /* if a physical device experienced an error while being opened */
3178 if (strncmp(filename, "/dev/", 5) == 0) {
3179 print_unmounting_directions(filename);
3181 #endif /* defined(__APPLE__) && defined(__MACH__) */
3182 return ret;
3185 /* Since this does ioctl the device must be already opened */
3186 bs->sg = hdev_is_sg(bs);
3188 if (flags & BDRV_O_RDWR) {
3189 ret = check_hdev_writable(s);
3190 if (ret < 0) {
3191 raw_close(bs);
3192 error_setg_errno(errp, -ret, "The device is not writable");
3193 return ret;
3197 return ret;
3200 #if defined(__linux__)
3201 static int coroutine_fn
3202 hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3204 BDRVRawState *s = bs->opaque;
3205 RawPosixAIOData acb;
3206 int ret;
3208 ret = fd_open(bs);
3209 if (ret < 0) {
3210 return ret;
3213 if (req == SG_IO && s->pr_mgr) {
3214 struct sg_io_hdr *io_hdr = buf;
3215 if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT ||
3216 io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) {
3217 return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs),
3218 s->fd, io_hdr);
3222 acb = (RawPosixAIOData) {
3223 .bs = bs,
3224 .aio_type = QEMU_AIO_IOCTL,
3225 .aio_fildes = s->fd,
3226 .aio_offset = 0,
3227 .ioctl = {
3228 .buf = buf,
3229 .cmd = req,
3233 return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb);
3235 #endif /* linux */
3237 static int fd_open(BlockDriverState *bs)
3239 BDRVRawState *s = bs->opaque;
3241 /* this is just to ensure s->fd is sane (its called by io ops) */
3242 if (s->fd >= 0)
3243 return 0;
3244 return -EIO;
3247 static coroutine_fn int
3248 hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
3250 int ret;
3252 ret = fd_open(bs);
3253 if (ret < 0) {
3254 return ret;
3256 return raw_do_pdiscard(bs, offset, bytes, true);
3259 static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs,
3260 int64_t offset, int bytes, BdrvRequestFlags flags)
3262 int rc;
3264 rc = fd_open(bs);
3265 if (rc < 0) {
3266 return rc;
3269 return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true);
3272 static int coroutine_fn hdev_co_create_opts(const char *filename, QemuOpts *opts,
3273 Error **errp)
3275 int fd;
3276 int ret = 0;
3277 struct stat stat_buf;
3278 int64_t total_size = 0;
3279 bool has_prefix;
3281 /* This function is used by both protocol block drivers and therefore either
3282 * of these prefixes may be given.
3283 * The return value has to be stored somewhere, otherwise this is an error
3284 * due to -Werror=unused-value. */
3285 has_prefix =
3286 strstart(filename, "host_device:", &filename) ||
3287 strstart(filename, "host_cdrom:" , &filename);
3289 (void)has_prefix;
3291 ret = raw_normalize_devicepath(&filename, errp);
3292 if (ret < 0) {
3293 return ret;
3296 /* Read out options */
3297 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
3298 BDRV_SECTOR_SIZE);
3300 fd = qemu_open(filename, O_WRONLY | O_BINARY);
3301 if (fd < 0) {
3302 ret = -errno;
3303 error_setg_errno(errp, -ret, "Could not open device");
3304 return ret;
3307 if (fstat(fd, &stat_buf) < 0) {
3308 ret = -errno;
3309 error_setg_errno(errp, -ret, "Could not stat device");
3310 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
3311 error_setg(errp,
3312 "The given file is neither a block nor a character device");
3313 ret = -ENODEV;
3314 } else if (lseek(fd, 0, SEEK_END) < total_size) {
3315 error_setg(errp, "Device is too small");
3316 ret = -ENOSPC;
3319 if (!ret && total_size) {
3320 uint8_t buf[BDRV_SECTOR_SIZE] = { 0 };
3321 int64_t zero_size = MIN(BDRV_SECTOR_SIZE, total_size);
3322 if (lseek(fd, 0, SEEK_SET) == -1) {
3323 ret = -errno;
3324 } else {
3325 ret = qemu_write_full(fd, buf, zero_size);
3326 ret = ret == zero_size ? 0 : -errno;
3329 qemu_close(fd);
3330 return ret;
3333 static BlockDriver bdrv_host_device = {
3334 .format_name = "host_device",
3335 .protocol_name = "host_device",
3336 .instance_size = sizeof(BDRVRawState),
3337 .bdrv_needs_filename = true,
3338 .bdrv_probe_device = hdev_probe_device,
3339 .bdrv_parse_filename = hdev_parse_filename,
3340 .bdrv_file_open = hdev_open,
3341 .bdrv_close = raw_close,
3342 .bdrv_reopen_prepare = raw_reopen_prepare,
3343 .bdrv_reopen_commit = raw_reopen_commit,
3344 .bdrv_reopen_abort = raw_reopen_abort,
3345 .bdrv_co_create_opts = hdev_co_create_opts,
3346 .create_opts = &raw_create_opts,
3347 .mutable_opts = mutable_opts,
3348 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
3349 .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes,
3351 .bdrv_co_preadv = raw_co_preadv,
3352 .bdrv_co_pwritev = raw_co_pwritev,
3353 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3354 .bdrv_co_pdiscard = hdev_co_pdiscard,
3355 .bdrv_co_copy_range_from = raw_co_copy_range_from,
3356 .bdrv_co_copy_range_to = raw_co_copy_range_to,
3357 .bdrv_refresh_limits = raw_refresh_limits,
3358 .bdrv_io_plug = raw_aio_plug,
3359 .bdrv_io_unplug = raw_aio_unplug,
3360 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
3362 .bdrv_co_truncate = raw_co_truncate,
3363 .bdrv_getlength = raw_getlength,
3364 .bdrv_get_info = raw_get_info,
3365 .bdrv_get_allocated_file_size
3366 = raw_get_allocated_file_size,
3367 .bdrv_check_perm = raw_check_perm,
3368 .bdrv_set_perm = raw_set_perm,
3369 .bdrv_abort_perm_update = raw_abort_perm_update,
3370 .bdrv_probe_blocksizes = hdev_probe_blocksizes,
3371 .bdrv_probe_geometry = hdev_probe_geometry,
3373 /* generic scsi device */
3374 #ifdef __linux__
3375 .bdrv_co_ioctl = hdev_co_ioctl,
3376 #endif
3379 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3380 static void cdrom_parse_filename(const char *filename, QDict *options,
3381 Error **errp)
3383 bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options);
3385 #endif
3387 #ifdef __linux__
3388 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
3389 Error **errp)
3391 BDRVRawState *s = bs->opaque;
3393 s->type = FTYPE_CD;
3395 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3396 return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp);
3399 static int cdrom_probe_device(const char *filename)
3401 int fd, ret;
3402 int prio = 0;
3403 struct stat st;
3405 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
3406 if (fd < 0) {
3407 goto out;
3409 ret = fstat(fd, &st);
3410 if (ret == -1 || !S_ISBLK(st.st_mode)) {
3411 goto outc;
3414 /* Attempt to detect via a CDROM specific ioctl */
3415 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
3416 if (ret >= 0)
3417 prio = 100;
3419 outc:
3420 qemu_close(fd);
3421 out:
3422 return prio;
3425 static bool cdrom_is_inserted(BlockDriverState *bs)
3427 BDRVRawState *s = bs->opaque;
3428 int ret;
3430 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
3431 return ret == CDS_DISC_OK;
3434 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
3436 BDRVRawState *s = bs->opaque;
3438 if (eject_flag) {
3439 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
3440 perror("CDROMEJECT");
3441 } else {
3442 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
3443 perror("CDROMEJECT");
3447 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
3449 BDRVRawState *s = bs->opaque;
3451 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
3453 * Note: an error can happen if the distribution automatically
3454 * mounts the CD-ROM
3456 /* perror("CDROM_LOCKDOOR"); */
3460 static BlockDriver bdrv_host_cdrom = {
3461 .format_name = "host_cdrom",
3462 .protocol_name = "host_cdrom",
3463 .instance_size = sizeof(BDRVRawState),
3464 .bdrv_needs_filename = true,
3465 .bdrv_probe_device = cdrom_probe_device,
3466 .bdrv_parse_filename = cdrom_parse_filename,
3467 .bdrv_file_open = cdrom_open,
3468 .bdrv_close = raw_close,
3469 .bdrv_reopen_prepare = raw_reopen_prepare,
3470 .bdrv_reopen_commit = raw_reopen_commit,
3471 .bdrv_reopen_abort = raw_reopen_abort,
3472 .bdrv_co_create_opts = hdev_co_create_opts,
3473 .create_opts = &raw_create_opts,
3474 .mutable_opts = mutable_opts,
3475 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
3478 .bdrv_co_preadv = raw_co_preadv,
3479 .bdrv_co_pwritev = raw_co_pwritev,
3480 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3481 .bdrv_refresh_limits = raw_refresh_limits,
3482 .bdrv_io_plug = raw_aio_plug,
3483 .bdrv_io_unplug = raw_aio_unplug,
3484 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
3486 .bdrv_co_truncate = raw_co_truncate,
3487 .bdrv_getlength = raw_getlength,
3488 .has_variable_length = true,
3489 .bdrv_get_allocated_file_size
3490 = raw_get_allocated_file_size,
3492 /* removable device support */
3493 .bdrv_is_inserted = cdrom_is_inserted,
3494 .bdrv_eject = cdrom_eject,
3495 .bdrv_lock_medium = cdrom_lock_medium,
3497 /* generic scsi device */
3498 .bdrv_co_ioctl = hdev_co_ioctl,
3500 #endif /* __linux__ */
3502 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3503 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
3504 Error **errp)
3506 BDRVRawState *s = bs->opaque;
3507 Error *local_err = NULL;
3508 int ret;
3510 s->type = FTYPE_CD;
3512 ret = raw_open_common(bs, options, flags, 0, true, &local_err);
3513 if (ret) {
3514 error_propagate(errp, local_err);
3515 return ret;
3518 /* make sure the door isn't locked at this time */
3519 ioctl(s->fd, CDIOCALLOW);
3520 return 0;
3523 static int cdrom_probe_device(const char *filename)
3525 if (strstart(filename, "/dev/cd", NULL) ||
3526 strstart(filename, "/dev/acd", NULL))
3527 return 100;
3528 return 0;
3531 static int cdrom_reopen(BlockDriverState *bs)
3533 BDRVRawState *s = bs->opaque;
3534 int fd;
3537 * Force reread of possibly changed/newly loaded disc,
3538 * FreeBSD seems to not notice sometimes...
3540 if (s->fd >= 0)
3541 qemu_close(s->fd);
3542 fd = qemu_open(bs->filename, s->open_flags, 0644);
3543 if (fd < 0) {
3544 s->fd = -1;
3545 return -EIO;
3547 s->fd = fd;
3549 /* make sure the door isn't locked at this time */
3550 ioctl(s->fd, CDIOCALLOW);
3551 return 0;
3554 static bool cdrom_is_inserted(BlockDriverState *bs)
3556 return raw_getlength(bs) > 0;
3559 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
3561 BDRVRawState *s = bs->opaque;
3563 if (s->fd < 0)
3564 return;
3566 (void) ioctl(s->fd, CDIOCALLOW);
3568 if (eject_flag) {
3569 if (ioctl(s->fd, CDIOCEJECT) < 0)
3570 perror("CDIOCEJECT");
3571 } else {
3572 if (ioctl(s->fd, CDIOCCLOSE) < 0)
3573 perror("CDIOCCLOSE");
3576 cdrom_reopen(bs);
3579 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
3581 BDRVRawState *s = bs->opaque;
3583 if (s->fd < 0)
3584 return;
3585 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
3587 * Note: an error can happen if the distribution automatically
3588 * mounts the CD-ROM
3590 /* perror("CDROM_LOCKDOOR"); */
3594 static BlockDriver bdrv_host_cdrom = {
3595 .format_name = "host_cdrom",
3596 .protocol_name = "host_cdrom",
3597 .instance_size = sizeof(BDRVRawState),
3598 .bdrv_needs_filename = true,
3599 .bdrv_probe_device = cdrom_probe_device,
3600 .bdrv_parse_filename = cdrom_parse_filename,
3601 .bdrv_file_open = cdrom_open,
3602 .bdrv_close = raw_close,
3603 .bdrv_reopen_prepare = raw_reopen_prepare,
3604 .bdrv_reopen_commit = raw_reopen_commit,
3605 .bdrv_reopen_abort = raw_reopen_abort,
3606 .bdrv_co_create_opts = hdev_co_create_opts,
3607 .create_opts = &raw_create_opts,
3608 .mutable_opts = mutable_opts,
3610 .bdrv_co_preadv = raw_co_preadv,
3611 .bdrv_co_pwritev = raw_co_pwritev,
3612 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3613 .bdrv_refresh_limits = raw_refresh_limits,
3614 .bdrv_io_plug = raw_aio_plug,
3615 .bdrv_io_unplug = raw_aio_unplug,
3616 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
3618 .bdrv_co_truncate = raw_co_truncate,
3619 .bdrv_getlength = raw_getlength,
3620 .has_variable_length = true,
3621 .bdrv_get_allocated_file_size
3622 = raw_get_allocated_file_size,
3624 /* removable device support */
3625 .bdrv_is_inserted = cdrom_is_inserted,
3626 .bdrv_eject = cdrom_eject,
3627 .bdrv_lock_medium = cdrom_lock_medium,
3629 #endif /* __FreeBSD__ */
3631 static void bdrv_file_init(void)
3634 * Register all the drivers. Note that order is important, the driver
3635 * registered last will get probed first.
3637 bdrv_register(&bdrv_file);
3638 bdrv_register(&bdrv_host_device);
3639 #ifdef __linux__
3640 bdrv_register(&bdrv_host_cdrom);
3641 #endif
3642 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3643 bdrv_register(&bdrv_host_cdrom);
3644 #endif
3647 block_init(bdrv_file_init);