?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
blockdev: check for BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE
[qemu/qmp-unstable.git] / block / raw-posix.c
blobb1af77e47fc2e1aad55127b6383fc57e3540b6d3
1 /*
2 * Block driver for RAW files (posix)
3 *
4 * Copyright (c) 2006 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24 #include "qemu-common.h"
25 #include "qemu/timer.h"
26 #include "qemu/log.h"
27 #include "block/block_int.h"
28 #include "qemu/module.h"
29 #include "trace.h"
30 #include "block/thread-pool.h"
31 #include "qemu/iov.h"
32 #include "raw-aio.h"
33 #include "qapi/util.h"
34
35 #if defined(__APPLE__) && (__MACH__)
36 #include <paths.h>
37 #include <sys/param.h>
38 #include <IOKit/IOKitLib.h>
39 #include <IOKit/IOBSD.h>
40 #include <IOKit/storage/IOMediaBSDClient.h>
41 #include <IOKit/storage/IOMedia.h>
42 #include <IOKit/storage/IOCDMedia.h>
43 //#include <IOKit/storage/IOCDTypes.h>
44 #include <CoreFoundation/CoreFoundation.h>
45 #endif
46
47 #ifdef __sun__
48 #define _POSIX_PTHREAD_SEMANTICS 1
49 #include <sys/dkio.h>
50 #endif
51 #ifdef __linux__
52 #include <sys/types.h>
53 #include <sys/stat.h>
54 #include <sys/ioctl.h>
55 #include <sys/param.h>
56 #include <linux/cdrom.h>
57 #include <linux/fd.h>
58 #include <linux/fs.h>
59 #ifndef FS_NOCOW_FL
60 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
61 #endif
62 #endif
63 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
64 #include <linux/falloc.h>
65 #endif
66 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
67 #include <sys/disk.h>
68 #include <sys/cdio.h>
69 #endif
70
71 #ifdef __OpenBSD__
72 #include <sys/ioctl.h>
73 #include <sys/disklabel.h>
74 #include <sys/dkio.h>
75 #endif
76
77 #ifdef __NetBSD__
78 #include <sys/ioctl.h>
79 #include <sys/disklabel.h>
80 #include <sys/dkio.h>
81 #include <sys/disk.h>
82 #endif
83
84 #ifdef __DragonFly__
85 #include <sys/ioctl.h>
86 #include <sys/diskslice.h>
87 #endif
88
89 #ifdef CONFIG_XFS
90 #include <xfs/xfs.h>
91 #endif
92
93 //#define DEBUG_FLOPPY
94
95 //#define DEBUG_BLOCK
96 #if defined(DEBUG_BLOCK)
97 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
98 { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
99 #else
100 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
101 #endif
102
103 /* OS X does not have O_DSYNC */
104 #ifndef O_DSYNC
105 #ifdef O_SYNC
106 #define O_DSYNC O_SYNC
107 #elif defined(O_FSYNC)
108 #define O_DSYNC O_FSYNC
109 #endif
110 #endif
111
112 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
113 #ifndef O_DIRECT
114 #define O_DIRECT O_DSYNC
115 #endif
116
117 #define FTYPE_FILE 0
118 #define FTYPE_CD 1
119 #define FTYPE_FD 2
120
121 /* if the FD is not accessed during that time (in ns), we try to
122 reopen it to see if the disk has been changed */
123 #define FD_OPEN_TIMEOUT (1000000000)
124
125 #define MAX_BLOCKSIZE 4096
126
127 typedef struct BDRVRawState {
128 int fd;
129 int type;
130 int open_flags;
131 size_t buf_align;
132
133 #if defined(__linux__)
134 /* linux floppy specific */
135 int64_t fd_open_time;
136 int64_t fd_error_time;
137 int fd_got_error;
138 int fd_media_changed;
139 #endif
140 #ifdef CONFIG_LINUX_AIO
141 int use_aio;
142 void *aio_ctx;
143 #endif
144 #ifdef CONFIG_XFS
145 bool is_xfs:1;
146 #endif
147 bool has_discard:1;
148 bool has_write_zeroes:1;
149 bool discard_zeroes:1;
150 bool needs_alignment;
151 } BDRVRawState;
152
153 typedef struct BDRVRawReopenState {
154 int fd;
155 int open_flags;
156 #ifdef CONFIG_LINUX_AIO
157 int use_aio;
158 #endif
159 } BDRVRawReopenState;
160
161 static int fd_open(BlockDriverState *bs);
162 static int64_t raw_getlength(BlockDriverState *bs);
163
164 typedef struct RawPosixAIOData {
165 BlockDriverState *bs;
166 int aio_fildes;
167 union {
168 struct iovec *aio_iov;
169 void *aio_ioctl_buf;
170 };
171 int aio_niov;
172 uint64_t aio_nbytes;
173 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
174 off_t aio_offset;
175 int aio_type;
176 } RawPosixAIOData;
177
178 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
179 static int cdrom_reopen(BlockDriverState *bs);
180 #endif
181
182 #if defined(__NetBSD__)
183 static int raw_normalize_devicepath(const char **filename)
184 {
185 static char namebuf[PATH_MAX];
186 const char *dp, *fname;
187 struct stat sb;
188
189 fname = *filename;
190 dp = strrchr(fname, '/');
191 if (lstat(fname, &sb) < 0) {
192 fprintf(stderr, "%s: stat failed: %s\n",
193 fname, strerror(errno));
194 return -errno;
195 }
196
197 if (!S_ISBLK(sb.st_mode)) {
198 return 0;
199 }
200
201 if (dp == NULL) {
202 snprintf(namebuf, PATH_MAX, "r%s", fname);
203 } else {
204 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
205 (int)(dp - fname), fname, dp + 1);
206 }
207 fprintf(stderr, "%s is a block device", fname);
208 *filename = namebuf;
209 fprintf(stderr, ", using %s\n", *filename);
210
211 return 0;
212 }
213 #else
214 static int raw_normalize_devicepath(const char **filename)
215 {
216 return 0;
217 }
218 #endif
219
220 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
221 {
222 BDRVRawState *s = bs->opaque;
223 char *buf;
224 unsigned int sector_size;
225
226 /* For /dev/sg devices the alignment is not really used.
227 With buffered I/O, we don't have any restrictions. */
228 if (bs->sg || !s->needs_alignment) {
229 bs->request_alignment = 1;
230 s->buf_align = 1;
231 return;
232 }
233
234 /* Try a few ioctls to get the right size */
235 bs->request_alignment = 0;
236 s->buf_align = 0;
237
238 #ifdef BLKSSZGET
239 if (ioctl(fd, BLKSSZGET, &sector_size) >= 0) {
240 bs->request_alignment = sector_size;
241 }
242 #endif
243 #ifdef DKIOCGETBLOCKSIZE
244 if (ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) >= 0) {
245 bs->request_alignment = sector_size;
246 }
247 #endif
248 #ifdef DIOCGSECTORSIZE
249 if (ioctl(fd, DIOCGSECTORSIZE, &sector_size) >= 0) {
250 bs->request_alignment = sector_size;
251 }
252 #endif
253 #ifdef CONFIG_XFS
254 if (s->is_xfs) {
255 struct dioattr da;
256 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
257 bs->request_alignment = da.d_miniosz;
258 /* The kernel returns wrong information for d_mem */
259 /* s->buf_align = da.d_mem; */
260 }
261 }
262 #endif
263
264 /* If we could not get the sizes so far, we can only guess them */
265 if (!s->buf_align) {
266 size_t align;
267 buf = qemu_memalign(MAX_BLOCKSIZE, 2 * MAX_BLOCKSIZE);
268 for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
269 if (pread(fd, buf + align, MAX_BLOCKSIZE, 0) >= 0) {
270 s->buf_align = align;
271 break;
272 }
273 }
274 qemu_vfree(buf);
275 }
276
277 if (!bs->request_alignment) {
278 size_t align;
279 buf = qemu_memalign(s->buf_align, MAX_BLOCKSIZE);
280 for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
281 if (pread(fd, buf, align, 0) >= 0) {
282 bs->request_alignment = align;
283 break;
284 }
285 }
286 qemu_vfree(buf);
287 }
288
289 if (!s->buf_align || !bs->request_alignment) {
290 error_setg(errp, "Could not find working O_DIRECT alignment. "
291 "Try cache.direct=off.");
292 }
293 }
294
295 static void raw_parse_flags(int bdrv_flags, int *open_flags)
296 {
297 assert(open_flags != NULL);
298
299 *open_flags |= O_BINARY;
300 *open_flags &= ~O_ACCMODE;
301 if (bdrv_flags & BDRV_O_RDWR) {
302 *open_flags |= O_RDWR;
303 } else {
304 *open_flags |= O_RDONLY;
305 }
306
307 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
308 * and O_DIRECT for no caching. */
309 if ((bdrv_flags & BDRV_O_NOCACHE)) {
310 *open_flags |= O_DIRECT;
311 }
312 }
313
314 static void raw_detach_aio_context(BlockDriverState *bs)
315 {
316 #ifdef CONFIG_LINUX_AIO
317 BDRVRawState *s = bs->opaque;
318
319 if (s->use_aio) {
320 laio_detach_aio_context(s->aio_ctx, bdrv_get_aio_context(bs));
321 }
322 #endif
323 }
324
325 static void raw_attach_aio_context(BlockDriverState *bs,
326 AioContext *new_context)
327 {
328 #ifdef CONFIG_LINUX_AIO
329 BDRVRawState *s = bs->opaque;
330
331 if (s->use_aio) {
332 laio_attach_aio_context(s->aio_ctx, new_context);
333 }
334 #endif
335 }
336
337 #ifdef CONFIG_LINUX_AIO
338 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
339 {
340 int ret = -1;
341 assert(aio_ctx != NULL);
342 assert(use_aio != NULL);
343 /*
344 * Currently Linux do AIO only for files opened with O_DIRECT
345 * specified so check NOCACHE flag too
346 */
347 if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
348 (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
349
350 /* if non-NULL, laio_init() has already been run */
351 if (*aio_ctx == NULL) {
352 *aio_ctx = laio_init();
353 if (!*aio_ctx) {
354 goto error;
355 }
356 }
357 *use_aio = 1;
358 } else {
359 *use_aio = 0;
360 }
361
362 ret = 0;
363
364 error:
365 return ret;
366 }
367 #endif
368
369 static void raw_parse_filename(const char *filename, QDict *options,
370 Error **errp)
371 {
372 /* The filename does not have to be prefixed by the protocol name, since
373 * "file" is the default protocol; therefore, the return value of this
374 * function call can be ignored. */
375 strstart(filename, "file:", &filename);
376
377 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
378 }
379
380 static QemuOptsList raw_runtime_opts = {
381 .name = "raw",
382 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
383 .desc = {
384 {
385 .name = "filename",
386 .type = QEMU_OPT_STRING,
387 .help = "File name of the image",
388 },
389 { /* end of list */ }
390 },
391 };
392
393 static int raw_open_common(BlockDriverState *bs, QDict *options,
394 int bdrv_flags, int open_flags, Error **errp)
395 {
396 BDRVRawState *s = bs->opaque;
397 QemuOpts *opts;
398 Error *local_err = NULL;
399 const char *filename = NULL;
400 int fd, ret;
401 struct stat st;
402
403 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
404 qemu_opts_absorb_qdict(opts, options, &local_err);
405 if (local_err) {
406 error_propagate(errp, local_err);
407 ret = -EINVAL;
408 goto fail;
409 }
410
411 filename = qemu_opt_get(opts, "filename");
412
413 ret = raw_normalize_devicepath(&filename);
414 if (ret != 0) {
415 error_setg_errno(errp, -ret, "Could not normalize device path");
416 goto fail;
417 }
418
419 s->open_flags = open_flags;
420 raw_parse_flags(bdrv_flags, &s->open_flags);
421
422 s->fd = -1;
423 fd = qemu_open(filename, s->open_flags, 0644);
424 if (fd < 0) {
425 ret = -errno;
426 if (ret == -EROFS) {
427 ret = -EACCES;
428 }
429 goto fail;
430 }
431 s->fd = fd;
432
433 #ifdef CONFIG_LINUX_AIO
434 if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
435 qemu_close(fd);
436 ret = -errno;
437 error_setg_errno(errp, -ret, "Could not set AIO state");
438 goto fail;
439 }
440 #endif
441
442 s->has_discard = true;
443 s->has_write_zeroes = true;
444 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
445 s->needs_alignment = true;
446 }
447
448 if (fstat(s->fd, &st) < 0) {
449 error_setg_errno(errp, errno, "Could not stat file");
450 goto fail;
451 }
452 if (S_ISREG(st.st_mode)) {
453 s->discard_zeroes = true;
454 }
455 if (S_ISBLK(st.st_mode)) {
456 #ifdef BLKDISCARDZEROES
457 unsigned int arg;
458 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
459 s->discard_zeroes = true;
460 }
461 #endif
462 #ifdef __linux__
463 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
464 * not rely on the contents of discarded blocks unless using O_DIRECT.
465 * Same for BLKZEROOUT.
466 */
467 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
468 s->discard_zeroes = false;
469 s->has_write_zeroes = false;
470 }
471 #endif
472 }
473 #ifdef __FreeBSD__
474 if (S_ISCHR(st.st_mode)) {
475 /*
476 * The file is a char device (disk), which on FreeBSD isn't behind
477 * a pager, so force all requests to be aligned. This is needed
478 * so QEMU makes sure all IO operations on the device are aligned
479 * to sector size, or else FreeBSD will reject them with EINVAL.
480 */
481 s->needs_alignment = true;
482 }
483 #endif
484
485 #ifdef CONFIG_XFS
486 if (platform_test_xfs_fd(s->fd)) {
487 s->is_xfs = true;
488 }
489 #endif
490
491 raw_attach_aio_context(bs, bdrv_get_aio_context(bs));
492
493 ret = 0;
494 fail:
495 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
496 unlink(filename);
497 }
498 qemu_opts_del(opts);
499 return ret;
500 }
501
502 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
503 Error **errp)
504 {
505 BDRVRawState *s = bs->opaque;
506 Error *local_err = NULL;
507 int ret;
508
509 s->type = FTYPE_FILE;
510 ret = raw_open_common(bs, options, flags, 0, &local_err);
511 if (local_err) {
512 error_propagate(errp, local_err);
513 }
514 return ret;
515 }
516
517 static int raw_reopen_prepare(BDRVReopenState *state,
518 BlockReopenQueue *queue, Error **errp)
519 {
520 BDRVRawState *s;
521 BDRVRawReopenState *raw_s;
522 int ret = 0;
523 Error *local_err = NULL;
524
525 assert(state != NULL);
526 assert(state->bs != NULL);
527
528 s = state->bs->opaque;
529
530 state->opaque = g_new0(BDRVRawReopenState, 1);
531 raw_s = state->opaque;
532
533 #ifdef CONFIG_LINUX_AIO
534 raw_s->use_aio = s->use_aio;
535
536 /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
537 * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
538 * won't override aio_ctx if aio_ctx is non-NULL */
539 if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
540 error_setg(errp, "Could not set AIO state");
541 return -1;
542 }
543 #endif
544
545 if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
546 raw_s->open_flags |= O_NONBLOCK;
547 }
548
549 raw_parse_flags(state->flags, &raw_s->open_flags);
550
551 raw_s->fd = -1;
552
553 int fcntl_flags = O_APPEND | O_NONBLOCK;
554 #ifdef O_NOATIME
555 fcntl_flags |= O_NOATIME;
556 #endif
557
558 #ifdef O_ASYNC
559 /* Not all operating systems have O_ASYNC, and those that don't
560 * will not let us track the state into raw_s->open_flags (typically
561 * you achieve the same effect with an ioctl, for example I_SETSIG
562 * on Solaris). But we do not use O_ASYNC, so that's fine.
563 */
564 assert((s->open_flags & O_ASYNC) == 0);
565 #endif
566
567 if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
568 /* dup the original fd */
569 /* TODO: use qemu fcntl wrapper */
570 #ifdef F_DUPFD_CLOEXEC
571 raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
572 #else
573 raw_s->fd = dup(s->fd);
574 if (raw_s->fd != -1) {
575 qemu_set_cloexec(raw_s->fd);
576 }
577 #endif
578 if (raw_s->fd >= 0) {
579 ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
580 if (ret) {
581 qemu_close(raw_s->fd);
582 raw_s->fd = -1;
583 }
584 }
585 }
586
587 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
588 if (raw_s->fd == -1) {
589 assert(!(raw_s->open_flags & O_CREAT));
590 raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
591 if (raw_s->fd == -1) {
592 error_setg_errno(errp, errno, "Could not reopen file");
593 ret = -1;
594 }
595 }
596
597 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
598 * alignment with the new fd. */
599 if (raw_s->fd != -1) {
600 raw_probe_alignment(state->bs, raw_s->fd, &local_err);
601 if (local_err) {
602 qemu_close(raw_s->fd);
603 raw_s->fd = -1;
604 error_propagate(errp, local_err);
605 ret = -EINVAL;
606 }
607 }
608
609 return ret;
610 }
611
612 static void raw_reopen_commit(BDRVReopenState *state)
613 {
614 BDRVRawReopenState *raw_s = state->opaque;
615 BDRVRawState *s = state->bs->opaque;
616
617 s->open_flags = raw_s->open_flags;
618
619 qemu_close(s->fd);
620 s->fd = raw_s->fd;
621 #ifdef CONFIG_LINUX_AIO
622 s->use_aio = raw_s->use_aio;
623 #endif
624
625 g_free(state->opaque);
626 state->opaque = NULL;
627 }
628
629
630 static void raw_reopen_abort(BDRVReopenState *state)
631 {
632 BDRVRawReopenState *raw_s = state->opaque;
633
634 /* nothing to do if NULL, we didn't get far enough */
635 if (raw_s == NULL) {
636 return;
637 }
638
639 if (raw_s->fd >= 0) {
640 qemu_close(raw_s->fd);
641 raw_s->fd = -1;
642 }
643 g_free(state->opaque);
644 state->opaque = NULL;
645 }
646
647 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
648 {
649 BDRVRawState *s = bs->opaque;
650
651 raw_probe_alignment(bs, s->fd, errp);
652 bs->bl.opt_mem_alignment = s->buf_align;
653 }
654
655 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
656 {
657 int ret;
658
659 ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
660 if (ret == -1) {
661 return -errno;
662 }
663
664 return 0;
665 }
666
667 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
668 {
669 int ret;
670
671 ret = qemu_fdatasync(aiocb->aio_fildes);
672 if (ret == -1) {
673 return -errno;
674 }
675 return 0;
676 }
677
678 #ifdef CONFIG_PREADV
679
680 static bool preadv_present = true;
681
682 static ssize_t
683 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
684 {
685 return preadv(fd, iov, nr_iov, offset);
686 }
687
688 static ssize_t
689 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
690 {
691 return pwritev(fd, iov, nr_iov, offset);
692 }
693
694 #else
695
696 static bool preadv_present = false;
697
698 static ssize_t
699 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
700 {
701 return -ENOSYS;
702 }
703
704 static ssize_t
705 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
706 {
707 return -ENOSYS;
708 }
709
710 #endif
711
712 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
713 {
714 ssize_t len;
715
716 do {
717 if (aiocb->aio_type & QEMU_AIO_WRITE)
718 len = qemu_pwritev(aiocb->aio_fildes,
719 aiocb->aio_iov,
720 aiocb->aio_niov,
721 aiocb->aio_offset);
722 else
723 len = qemu_preadv(aiocb->aio_fildes,
724 aiocb->aio_iov,
725 aiocb->aio_niov,
726 aiocb->aio_offset);
727 } while (len == -1 && errno == EINTR);
728
729 if (len == -1) {
730 return -errno;
731 }
732 return len;
733 }
734
735 /*
736 * Read/writes the data to/from a given linear buffer.
737 *
738 * Returns the number of bytes handles or -errno in case of an error. Short
739 * reads are only returned if the end of the file is reached.
740 */
741 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
742 {
743 ssize_t offset = 0;
744 ssize_t len;
745
746 while (offset < aiocb->aio_nbytes) {
747 if (aiocb->aio_type & QEMU_AIO_WRITE) {
748 len = pwrite(aiocb->aio_fildes,
749 (const char *)buf + offset,
750 aiocb->aio_nbytes - offset,
751 aiocb->aio_offset + offset);
752 } else {
753 len = pread(aiocb->aio_fildes,
754 buf + offset,
755 aiocb->aio_nbytes - offset,
756 aiocb->aio_offset + offset);
757 }
758 if (len == -1 && errno == EINTR) {
759 continue;
760 } else if (len == -1 && errno == EINVAL &&
761 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
762 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
763 offset > 0) {
764 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
765 * after a short read. Assume that O_DIRECT short reads only occur
766 * at EOF. Therefore this is a short read, not an I/O error.
767 */
768 break;
769 } else if (len == -1) {
770 offset = -errno;
771 break;
772 } else if (len == 0) {
773 break;
774 }
775 offset += len;
776 }
777
778 return offset;
779 }
780
781 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
782 {
783 ssize_t nbytes;
784 char *buf;
785
786 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
787 /*
788 * If there is just a single buffer, and it is properly aligned
789 * we can just use plain pread/pwrite without any problems.
790 */
791 if (aiocb->aio_niov == 1) {
792 return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
793 }
794 /*
795 * We have more than one iovec, and all are properly aligned.
796 *
797 * Try preadv/pwritev first and fall back to linearizing the
798 * buffer if it's not supported.
799 */
800 if (preadv_present) {
801 nbytes = handle_aiocb_rw_vector(aiocb);
802 if (nbytes == aiocb->aio_nbytes ||
803 (nbytes < 0 && nbytes != -ENOSYS)) {
804 return nbytes;
805 }
806 preadv_present = false;
807 }
808
809 /*
810 * XXX(hch): short read/write. no easy way to handle the reminder
811 * using these interfaces. For now retry using plain
812 * pread/pwrite?
813 */
814 }
815
816 /*
817 * Ok, we have to do it the hard way, copy all segments into
818 * a single aligned buffer.
819 */
820 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
821 if (buf == NULL) {
822 return -ENOMEM;
823 }
824
825 if (aiocb->aio_type & QEMU_AIO_WRITE) {
826 char *p = buf;
827 int i;
828
829 for (i = 0; i < aiocb->aio_niov; ++i) {
830 memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
831 p += aiocb->aio_iov[i].iov_len;
832 }
833 assert(p - buf == aiocb->aio_nbytes);
834 }
835
836 nbytes = handle_aiocb_rw_linear(aiocb, buf);
837 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
838 char *p = buf;
839 size_t count = aiocb->aio_nbytes, copy;
840 int i;
841
842 for (i = 0; i < aiocb->aio_niov && count; ++i) {
843 copy = count;
844 if (copy > aiocb->aio_iov[i].iov_len) {
845 copy = aiocb->aio_iov[i].iov_len;
846 }
847 memcpy(aiocb->aio_iov[i].iov_base, p, copy);
848 assert(count >= copy);
849 p += copy;
850 count -= copy;
851 }
852 assert(count == 0);
853 }
854 qemu_vfree(buf);
855
856 return nbytes;
857 }
858
859 #ifdef CONFIG_XFS
860 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
861 {
862 struct xfs_flock64 fl;
863
864 memset(&fl, 0, sizeof(fl));
865 fl.l_whence = SEEK_SET;
866 fl.l_start = offset;
867 fl.l_len = bytes;
868
869 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
870 DEBUG_BLOCK_PRINT("cannot write zero range (%s)\n", strerror(errno));
871 return -errno;
872 }
873
874 return 0;
875 }
876
877 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
878 {
879 struct xfs_flock64 fl;
880
881 memset(&fl, 0, sizeof(fl));
882 fl.l_whence = SEEK_SET;
883 fl.l_start = offset;
884 fl.l_len = bytes;
885
886 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
887 DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
888 return -errno;
889 }
890
891 return 0;
892 }
893 #endif
894
895 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
896 {
897 int ret = -EOPNOTSUPP;
898 BDRVRawState *s = aiocb->bs->opaque;
899
900 if (s->has_write_zeroes == 0) {
901 return -ENOTSUP;
902 }
903
904 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
905 #ifdef BLKZEROOUT
906 do {
907 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
908 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
909 return 0;
910 }
911 } while (errno == EINTR);
912
913 ret = -errno;
914 #endif
915 } else {
916 #ifdef CONFIG_XFS
917 if (s->is_xfs) {
918 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
919 }
920 #endif
921 }
922
923 if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
924 ret == -ENOTTY) {
925 s->has_write_zeroes = false;
926 ret = -ENOTSUP;
927 }
928 return ret;
929 }
930
931 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
932 {
933 int ret = -EOPNOTSUPP;
934 BDRVRawState *s = aiocb->bs->opaque;
935
936 if (!s->has_discard) {
937 return -ENOTSUP;
938 }
939
940 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
941 #ifdef BLKDISCARD
942 do {
943 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
944 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
945 return 0;
946 }
947 } while (errno == EINTR);
948
949 ret = -errno;
950 #endif
951 } else {
952 #ifdef CONFIG_XFS
953 if (s->is_xfs) {
954 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
955 }
956 #endif
957
958 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
959 do {
960 if (fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
961 aiocb->aio_offset, aiocb->aio_nbytes) == 0) {
962 return 0;
963 }
964 } while (errno == EINTR);
965
966 ret = -errno;
967 #endif
968 }
969
970 if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
971 ret == -ENOTTY) {
972 s->has_discard = false;
973 ret = -ENOTSUP;
974 }
975 return ret;
976 }
977
978 static int aio_worker(void *arg)
979 {
980 RawPosixAIOData *aiocb = arg;
981 ssize_t ret = 0;
982
983 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
984 case QEMU_AIO_READ:
985 ret = handle_aiocb_rw(aiocb);
986 if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) {
987 iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
988 0, aiocb->aio_nbytes - ret);
989
990 ret = aiocb->aio_nbytes;
991 }
992 if (ret == aiocb->aio_nbytes) {
993 ret = 0;
994 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
995 ret = -EINVAL;
996 }
997 break;
998 case QEMU_AIO_WRITE:
999 ret = handle_aiocb_rw(aiocb);
1000 if (ret == aiocb->aio_nbytes) {
1001 ret = 0;
1002 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1003 ret = -EINVAL;
1004 }
1005 break;
1006 case QEMU_AIO_FLUSH:
1007 ret = handle_aiocb_flush(aiocb);
1008 break;
1009 case QEMU_AIO_IOCTL:
1010 ret = handle_aiocb_ioctl(aiocb);
1011 break;
1012 case QEMU_AIO_DISCARD:
1013 ret = handle_aiocb_discard(aiocb);
1014 break;
1015 case QEMU_AIO_WRITE_ZEROES:
1016 ret = handle_aiocb_write_zeroes(aiocb);
1017 break;
1018 default:
1019 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1020 ret = -EINVAL;
1021 break;
1022 }
1023
1024 g_slice_free(RawPosixAIOData, aiocb);
1025 return ret;
1026 }
1027
1028 static int paio_submit_co(BlockDriverState *bs, int fd,
1029 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1030 int type)
1031 {
1032 RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
1033 ThreadPool *pool;
1034
1035 acb->bs = bs;
1036 acb->aio_type = type;
1037 acb->aio_fildes = fd;
1038
1039 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1040 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1041
1042 if (qiov) {
1043 acb->aio_iov = qiov->iov;
1044 acb->aio_niov = qiov->niov;
1045 assert(qiov->size == acb->aio_nbytes);
1046 }
1047
1048 trace_paio_submit_co(sector_num, nb_sectors, type);
1049 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1050 return thread_pool_submit_co(pool, aio_worker, acb);
1051 }
1052
1053 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
1054 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1055 BlockCompletionFunc *cb, void *opaque, int type)
1056 {
1057 RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
1058 ThreadPool *pool;
1059
1060 acb->bs = bs;
1061 acb->aio_type = type;
1062 acb->aio_fildes = fd;
1063
1064 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1065 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1066
1067 if (qiov) {
1068 acb->aio_iov = qiov->iov;
1069 acb->aio_niov = qiov->niov;
1070 assert(qiov->size == acb->aio_nbytes);
1071 }
1072
1073 trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
1074 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1075 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1076 }
1077
1078 static BlockAIOCB *raw_aio_submit(BlockDriverState *bs,
1079 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1080 BlockCompletionFunc *cb, void *opaque, int type)
1081 {
1082 BDRVRawState *s = bs->opaque;
1083
1084 if (fd_open(bs) < 0)
1085 return NULL;
1086
1087 /*
1088 * Check if the underlying device requires requests to be aligned,
1089 * and if the request we are trying to submit is aligned or not.
1090 * If this is the case tell the low-level driver that it needs
1091 * to copy the buffer.
1092 */
1093 if (s->needs_alignment) {
1094 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1095 type |= QEMU_AIO_MISALIGNED;
1096 #ifdef CONFIG_LINUX_AIO
1097 } else if (s->use_aio) {
1098 return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
1099 nb_sectors, cb, opaque, type);
1100 #endif
1101 }
1102 }
1103
1104 return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
1105 cb, opaque, type);
1106 }
1107
1108 static void raw_aio_plug(BlockDriverState *bs)
1109 {
1110 #ifdef CONFIG_LINUX_AIO
1111 BDRVRawState *s = bs->opaque;
1112 if (s->use_aio) {
1113 laio_io_plug(bs, s->aio_ctx);
1114 }
1115 #endif
1116 }
1117
1118 static void raw_aio_unplug(BlockDriverState *bs)
1119 {
1120 #ifdef CONFIG_LINUX_AIO
1121 BDRVRawState *s = bs->opaque;
1122 if (s->use_aio) {
1123 laio_io_unplug(bs, s->aio_ctx, true);
1124 }
1125 #endif
1126 }
1127
1128 static void raw_aio_flush_io_queue(BlockDriverState *bs)
1129 {
1130 #ifdef CONFIG_LINUX_AIO
1131 BDRVRawState *s = bs->opaque;
1132 if (s->use_aio) {
1133 laio_io_unplug(bs, s->aio_ctx, false);
1134 }
1135 #endif
1136 }
1137
1138 static BlockAIOCB *raw_aio_readv(BlockDriverState *bs,
1139 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1140 BlockCompletionFunc *cb, void *opaque)
1141 {
1142 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1143 cb, opaque, QEMU_AIO_READ);
1144 }
1145
1146 static BlockAIOCB *raw_aio_writev(BlockDriverState *bs,
1147 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1148 BlockCompletionFunc *cb, void *opaque)
1149 {
1150 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1151 cb, opaque, QEMU_AIO_WRITE);
1152 }
1153
1154 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
1155 BlockCompletionFunc *cb, void *opaque)
1156 {
1157 BDRVRawState *s = bs->opaque;
1158
1159 if (fd_open(bs) < 0)
1160 return NULL;
1161
1162 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1163 }
1164
1165 static void raw_close(BlockDriverState *bs)
1166 {
1167 BDRVRawState *s = bs->opaque;
1168
1169 raw_detach_aio_context(bs);
1170
1171 #ifdef CONFIG_LINUX_AIO
1172 if (s->use_aio) {
1173 laio_cleanup(s->aio_ctx);
1174 }
1175 #endif
1176 if (s->fd >= 0) {
1177 qemu_close(s->fd);
1178 s->fd = -1;
1179 }
1180 }
1181
1182 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1183 {
1184 BDRVRawState *s = bs->opaque;
1185 struct stat st;
1186
1187 if (fstat(s->fd, &st)) {
1188 return -errno;
1189 }
1190
1191 if (S_ISREG(st.st_mode)) {
1192 if (ftruncate(s->fd, offset) < 0) {
1193 return -errno;
1194 }
1195 } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1196 if (offset > raw_getlength(bs)) {
1197 return -EINVAL;
1198 }
1199 } else {
1200 return -ENOTSUP;
1201 }
1202
1203 return 0;
1204 }
1205
1206 #ifdef __OpenBSD__
1207 static int64_t raw_getlength(BlockDriverState *bs)
1208 {
1209 BDRVRawState *s = bs->opaque;
1210 int fd = s->fd;
1211 struct stat st;
1212
1213 if (fstat(fd, &st))
1214 return -errno;
1215 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1216 struct disklabel dl;
1217
1218 if (ioctl(fd, DIOCGDINFO, &dl))
1219 return -errno;
1220 return (uint64_t)dl.d_secsize *
1221 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1222 } else
1223 return st.st_size;
1224 }
1225 #elif defined(__NetBSD__)
1226 static int64_t raw_getlength(BlockDriverState *bs)
1227 {
1228 BDRVRawState *s = bs->opaque;
1229 int fd = s->fd;
1230 struct stat st;
1231
1232 if (fstat(fd, &st))
1233 return -errno;
1234 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1235 struct dkwedge_info dkw;
1236
1237 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1238 return dkw.dkw_size * 512;
1239 } else {
1240 struct disklabel dl;
1241
1242 if (ioctl(fd, DIOCGDINFO, &dl))
1243 return -errno;
1244 return (uint64_t)dl.d_secsize *
1245 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1246 }
1247 } else
1248 return st.st_size;
1249 }
1250 #elif defined(__sun__)
1251 static int64_t raw_getlength(BlockDriverState *bs)
1252 {
1253 BDRVRawState *s = bs->opaque;
1254 struct dk_minfo minfo;
1255 int ret;
1256 int64_t size;
1257
1258 ret = fd_open(bs);
1259 if (ret < 0) {
1260 return ret;
1261 }
1262
1263 /*
1264 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1265 */
1266 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1267 if (ret != -1) {
1268 return minfo.dki_lbsize * minfo.dki_capacity;
1269 }
1270
1271 /*
1272 * There are reports that lseek on some devices fails, but
1273 * irc discussion said that contingency on contingency was overkill.
1274 */
1275 size = lseek(s->fd, 0, SEEK_END);
1276 if (size < 0) {
1277 return -errno;
1278 }
1279 return size;
1280 }
1281 #elif defined(CONFIG_BSD)
1282 static int64_t raw_getlength(BlockDriverState *bs)
1283 {
1284 BDRVRawState *s = bs->opaque;
1285 int fd = s->fd;
1286 int64_t size;
1287 struct stat sb;
1288 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1289 int reopened = 0;
1290 #endif
1291 int ret;
1292
1293 ret = fd_open(bs);
1294 if (ret < 0)
1295 return ret;
1296
1297 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1298 again:
1299 #endif
1300 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1301 #ifdef DIOCGMEDIASIZE
1302 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1303 #elif defined(DIOCGPART)
1304 {
1305 struct partinfo pi;
1306 if (ioctl(fd, DIOCGPART, &pi) == 0)
1307 size = pi.media_size;
1308 else
1309 size = 0;
1310 }
1311 if (size == 0)
1312 #endif
1313 #if defined(__APPLE__) && defined(__MACH__)
1314 size = LLONG_MAX;
1315 #else
1316 size = lseek(fd, 0LL, SEEK_END);
1317 if (size < 0) {
1318 return -errno;
1319 }
1320 #endif
1321 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1322 switch(s->type) {
1323 case FTYPE_CD:
1324 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1325 if (size == 2048LL * (unsigned)-1)
1326 size = 0;
1327 /* XXX no disc? maybe we need to reopen... */
1328 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1329 reopened = 1;
1330 goto again;
1331 }
1332 }
1333 #endif
1334 } else {
1335 size = lseek(fd, 0, SEEK_END);
1336 if (size < 0) {
1337 return -errno;
1338 }
1339 }
1340 return size;
1341 }
1342 #else
1343 static int64_t raw_getlength(BlockDriverState *bs)
1344 {
1345 BDRVRawState *s = bs->opaque;
1346 int ret;
1347 int64_t size;
1348
1349 ret = fd_open(bs);
1350 if (ret < 0) {
1351 return ret;
1352 }
1353
1354 size = lseek(s->fd, 0, SEEK_END);
1355 if (size < 0) {
1356 return -errno;
1357 }
1358 return size;
1359 }
1360 #endif
1361
1362 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1363 {
1364 struct stat st;
1365 BDRVRawState *s = bs->opaque;
1366
1367 if (fstat(s->fd, &st) < 0) {
1368 return -errno;
1369 }
1370 return (int64_t)st.st_blocks * 512;
1371 }
1372
1373 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
1374 {
1375 int fd;
1376 int result = 0;
1377 int64_t total_size = 0;
1378 bool nocow = false;
1379 PreallocMode prealloc;
1380 char *buf = NULL;
1381 Error *local_err = NULL;
1382
1383 strstart(filename, "file:", &filename);
1384
1385 /* Read out options */
1386 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1387 BDRV_SECTOR_SIZE);
1388 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
1389 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1390 prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
1391 PREALLOC_MODE_MAX, PREALLOC_MODE_OFF,
1392 &local_err);
1393 g_free(buf);
1394 if (local_err) {
1395 error_propagate(errp, local_err);
1396 result = -EINVAL;
1397 goto out;
1398 }
1399
1400 fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
1401 0644);
1402 if (fd < 0) {
1403 result = -errno;
1404 error_setg_errno(errp, -result, "Could not create file");
1405 goto out;
1406 }
1407
1408 if (nocow) {
1409 #ifdef __linux__
1410 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1411 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1412 * will be ignored since any failure of this operation should not
1413 * block the left work.
1414 */
1415 int attr;
1416 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
1417 attr |= FS_NOCOW_FL;
1418 ioctl(fd, FS_IOC_SETFLAGS, &attr);
1419 }
1420 #endif
1421 }
1422
1423 if (ftruncate(fd, total_size) != 0) {
1424 result = -errno;
1425 error_setg_errno(errp, -result, "Could not resize file");
1426 goto out_close;
1427 }
1428
1429 switch (prealloc) {
1430 #ifdef CONFIG_POSIX_FALLOCATE
1431 case PREALLOC_MODE_FALLOC:
1432 /* posix_fallocate() doesn't set errno. */
1433 result = -posix_fallocate(fd, 0, total_size);
1434 if (result != 0) {
1435 error_setg_errno(errp, -result,
1436 "Could not preallocate data for the new file");
1437 }
1438 break;
1439 #endif
1440 case PREALLOC_MODE_FULL:
1441 {
1442 int64_t num = 0, left = total_size;
1443 buf = g_malloc0(65536);
1444
1445 while (left > 0) {
1446 num = MIN(left, 65536);
1447 result = write(fd, buf, num);
1448 if (result < 0) {
1449 result = -errno;
1450 error_setg_errno(errp, -result,
1451 "Could not write to the new file");
1452 break;
1453 }
1454 left -= result;
1455 }
1456 if (result >= 0) {
1457 result = fsync(fd);
1458 if (result < 0) {
1459 result = -errno;
1460 error_setg_errno(errp, -result,
1461 "Could not flush new file to disk");
1462 }
1463 }
1464 g_free(buf);
1465 break;
1466 }
1467 case PREALLOC_MODE_OFF:
1468 break;
1469 default:
1470 result = -EINVAL;
1471 error_setg(errp, "Unsupported preallocation mode: %s",
1472 PreallocMode_lookup[prealloc]);
1473 break;
1474 }
1475
1476 out_close:
1477 if (qemu_close(fd) != 0 && result == 0) {
1478 result = -errno;
1479 error_setg_errno(errp, -result, "Could not close the new file");
1480 }
1481 out:
1482 return result;
1483 }
1484
1485 /*
1486 * Find allocation range in @bs around offset @start.
1487 * May change underlying file descriptor's file offset.
1488 * If @start is not in a hole, store @start in @data, and the
1489 * beginning of the next hole in @hole, and return 0.
1490 * If @start is in a non-trailing hole, store @start in @hole and the
1491 * beginning of the next non-hole in @data, and return 0.
1492 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1493 * If we can't find out, return a negative errno other than -ENXIO.
1494 */
1495 static int find_allocation(BlockDriverState *bs, off_t start,
1496 off_t *data, off_t *hole)
1497 {
1498 #if defined SEEK_HOLE && defined SEEK_DATA
1499 BDRVRawState *s = bs->opaque;
1500 off_t offs;
1501
1502 /*
1503 * SEEK_DATA cases:
1504 * D1. offs == start: start is in data
1505 * D2. offs > start: start is in a hole, next data at offs
1506 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1507 * or start is beyond EOF
1508 * If the latter happens, the file has been truncated behind
1509 * our back since we opened it. All bets are off then.
1510 * Treating like a trailing hole is simplest.
1511 * D4. offs < 0, errno != ENXIO: we learned nothing
1512 */
1513 offs = lseek(s->fd, start, SEEK_DATA);
1514 if (offs < 0) {
1515 return -errno; /* D3 or D4 */
1516 }
1517 assert(offs >= start);
1518
1519 if (offs > start) {
1520 /* D2: in hole, next data at offs */
1521 *hole = start;
1522 *data = offs;
1523 return 0;
1524 }
1525
1526 /* D1: in data, end not yet known */
1527
1528 /*
1529 * SEEK_HOLE cases:
1530 * H1. offs == start: start is in a hole
1531 * If this happens here, a hole has been dug behind our back
1532 * since the previous lseek().
1533 * H2. offs > start: either start is in data, next hole at offs,
1534 * or start is in trailing hole, EOF at offs
1535 * Linux treats trailing holes like any other hole: offs ==
1536 * start. Solaris seeks to EOF instead: offs > start (blech).
1537 * If that happens here, a hole has been dug behind our back
1538 * since the previous lseek().
1539 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1540 * If this happens, the file has been truncated behind our
1541 * back since we opened it. Treat it like a trailing hole.
1542 * H4. offs < 0, errno != ENXIO: we learned nothing
1543 * Pretend we know nothing at all, i.e. "forget" about D1.
1544 */
1545 offs = lseek(s->fd, start, SEEK_HOLE);
1546 if (offs < 0) {
1547 return -errno; /* D1 and (H3 or H4) */
1548 }
1549 assert(offs >= start);
1550
1551 if (offs > start) {
1552 /*
1553 * D1 and H2: either in data, next hole at offs, or it was in
1554 * data but is now in a trailing hole. In the latter case,
1555 * all bets are off. Treating it as if it there was data all
1556 * the way to EOF is safe, so simply do that.
1557 */
1558 *data = start;
1559 *hole = offs;
1560 return 0;
1561 }
1562
1563 /* D1 and H1 */
1564 return -EBUSY;
1565 #else
1566 return -ENOTSUP;
1567 #endif
1568 }
1569
1570 /*
1571 * Returns the allocation status of the specified sectors.
1572 *
1573 * If 'sector_num' is beyond the end of the disk image the return value is 0
1574 * and 'pnum' is set to 0.
1575 *
1576 * 'pnum' is set to the number of sectors (including and immediately following
1577 * the specified sector) that are known to be in the same
1578 * allocated/unallocated state.
1579 *
1580 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1581 * beyond the end of the disk image it will be clamped.
1582 */
1583 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1584 int64_t sector_num,
1585 int nb_sectors, int *pnum)
1586 {
1587 off_t start, data = 0, hole = 0;
1588 int64_t total_size;
1589 int ret;
1590
1591 ret = fd_open(bs);
1592 if (ret < 0) {
1593 return ret;
1594 }
1595
1596 start = sector_num * BDRV_SECTOR_SIZE;
1597 total_size = bdrv_getlength(bs);
1598 if (total_size < 0) {
1599 return total_size;
1600 } else if (start >= total_size) {
1601 *pnum = 0;
1602 return 0;
1603 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1604 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1605 }
1606
1607 ret = find_allocation(bs, start, &data, &hole);
1608 if (ret == -ENXIO) {
1609 /* Trailing hole */
1610 *pnum = nb_sectors;
1611 ret = BDRV_BLOCK_ZERO;
1612 } else if (ret < 0) {
1613 /* No info available, so pretend there are no holes */
1614 *pnum = nb_sectors;
1615 ret = BDRV_BLOCK_DATA;
1616 } else if (data == start) {
1617 /* On a data extent, compute sectors to the end of the extent. */
1618 *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1619 ret = BDRV_BLOCK_DATA;
1620 } else {
1621 /* On a hole, compute sectors to the beginning of the next extent. */
1622 assert(hole == start);
1623 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1624 ret = BDRV_BLOCK_ZERO;
1625 }
1626 return ret | BDRV_BLOCK_OFFSET_VALID | start;
1627 }
1628
1629 static coroutine_fn BlockAIOCB *raw_aio_discard(BlockDriverState *bs,
1630 int64_t sector_num, int nb_sectors,
1631 BlockCompletionFunc *cb, void *opaque)
1632 {
1633 BDRVRawState *s = bs->opaque;
1634
1635 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1636 cb, opaque, QEMU_AIO_DISCARD);
1637 }
1638
1639 static int coroutine_fn raw_co_write_zeroes(
1640 BlockDriverState *bs, int64_t sector_num,
1641 int nb_sectors, BdrvRequestFlags flags)
1642 {
1643 BDRVRawState *s = bs->opaque;
1644
1645 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1646 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1647 QEMU_AIO_WRITE_ZEROES);
1648 } else if (s->discard_zeroes) {
1649 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1650 QEMU_AIO_DISCARD);
1651 }
1652 return -ENOTSUP;
1653 }
1654
1655 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1656 {
1657 BDRVRawState *s = bs->opaque;
1658
1659 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1660 bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1661 return 0;
1662 }
1663
1664 static QemuOptsList raw_create_opts = {
1665 .name = "raw-create-opts",
1666 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
1667 .desc = {
1668 {
1669 .name = BLOCK_OPT_SIZE,
1670 .type = QEMU_OPT_SIZE,
1671 .help = "Virtual disk size"
1672 },
1673 {
1674 .name = BLOCK_OPT_NOCOW,
1675 .type = QEMU_OPT_BOOL,
1676 .help = "Turn off copy-on-write (valid only on btrfs)"
1677 },
1678 {
1679 .name = BLOCK_OPT_PREALLOC,
1680 .type = QEMU_OPT_STRING,
1681 .help = "Preallocation mode (allowed values: off, falloc, full)"
1682 },
1683 { /* end of list */ }
1684 }
1685 };
1686
1687 static BlockDriver bdrv_file = {
1688 .format_name = "file",
1689 .protocol_name = "file",
1690 .instance_size = sizeof(BDRVRawState),
1691 .bdrv_needs_filename = true,
1692 .bdrv_probe = NULL, /* no probe for protocols */
1693 .bdrv_parse_filename = raw_parse_filename,
1694 .bdrv_file_open = raw_open,
1695 .bdrv_reopen_prepare = raw_reopen_prepare,
1696 .bdrv_reopen_commit = raw_reopen_commit,
1697 .bdrv_reopen_abort = raw_reopen_abort,
1698 .bdrv_close = raw_close,
1699 .bdrv_create = raw_create,
1700 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1701 .bdrv_co_get_block_status = raw_co_get_block_status,
1702 .bdrv_co_write_zeroes = raw_co_write_zeroes,
1703
1704 .bdrv_aio_readv = raw_aio_readv,
1705 .bdrv_aio_writev = raw_aio_writev,
1706 .bdrv_aio_flush = raw_aio_flush,
1707 .bdrv_aio_discard = raw_aio_discard,
1708 .bdrv_refresh_limits = raw_refresh_limits,
1709 .bdrv_io_plug = raw_aio_plug,
1710 .bdrv_io_unplug = raw_aio_unplug,
1711 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
1712
1713 .bdrv_truncate = raw_truncate,
1714 .bdrv_getlength = raw_getlength,
1715 .bdrv_get_info = raw_get_info,
1716 .bdrv_get_allocated_file_size
1717 = raw_get_allocated_file_size,
1718
1719 .bdrv_detach_aio_context = raw_detach_aio_context,
1720 .bdrv_attach_aio_context = raw_attach_aio_context,
1721
1722 .create_opts = &raw_create_opts,
1723 };
1724
1725 /***********************************************/
1726 /* host device */
1727
1728 #if defined(__APPLE__) && defined(__MACH__)
1729 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1730 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1731
1732 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1733 {
1734 kern_return_t kernResult;
1735 mach_port_t masterPort;
1736 CFMutableDictionaryRef classesToMatch;
1737
1738 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1739 if ( KERN_SUCCESS != kernResult ) {
1740 printf( "IOMasterPort returned %d\n", kernResult );
1741 }
1742
1743 classesToMatch = IOServiceMatching( kIOCDMediaClass );
1744 if ( classesToMatch == NULL ) {
1745 printf( "IOServiceMatching returned a NULL dictionary.\n" );
1746 } else {
1747 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1748 }
1749 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1750 if ( KERN_SUCCESS != kernResult )
1751 {
1752 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1753 }
1754
1755 return kernResult;
1756 }
1757
1758 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
1759 {
1760 io_object_t nextMedia;
1761 kern_return_t kernResult = KERN_FAILURE;
1762 *bsdPath = '\0';
1763 nextMedia = IOIteratorNext( mediaIterator );
1764 if ( nextMedia )
1765 {
1766 CFTypeRef bsdPathAsCFString;
1767 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1768 if ( bsdPathAsCFString ) {
1769 size_t devPathLength;
1770 strcpy( bsdPath, _PATH_DEV );
1771 strcat( bsdPath, "r" );
1772 devPathLength = strlen( bsdPath );
1773 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1774 kernResult = KERN_SUCCESS;
1775 }
1776 CFRelease( bsdPathAsCFString );
1777 }
1778 IOObjectRelease( nextMedia );
1779 }
1780
1781 return kernResult;
1782 }
1783
1784 #endif
1785
1786 static int hdev_probe_device(const char *filename)
1787 {
1788 struct stat st;
1789
1790 /* allow a dedicated CD-ROM driver to match with a higher priority */
1791 if (strstart(filename, "/dev/cdrom", NULL))
1792 return 50;
1793
1794 if (stat(filename, &st) >= 0 &&
1795 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
1796 return 100;
1797 }
1798
1799 return 0;
1800 }
1801
1802 static int check_hdev_writable(BDRVRawState *s)
1803 {
1804 #if defined(BLKROGET)
1805 /* Linux block devices can be configured "read-only" using blockdev(8).
1806 * This is independent of device node permissions and therefore open(2)
1807 * with O_RDWR succeeds. Actual writes fail with EPERM.
1808 *
1809 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
1810 * check for read-only block devices so that Linux block devices behave
1811 * properly.
1812 */
1813 struct stat st;
1814 int readonly = 0;
1815
1816 if (fstat(s->fd, &st)) {
1817 return -errno;
1818 }
1819
1820 if (!S_ISBLK(st.st_mode)) {
1821 return 0;
1822 }
1823
1824 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
1825 return -errno;
1826 }
1827
1828 if (readonly) {
1829 return -EACCES;
1830 }
1831 #endif /* defined(BLKROGET) */
1832 return 0;
1833 }
1834
1835 static void hdev_parse_filename(const char *filename, QDict *options,
1836 Error **errp)
1837 {
1838 /* The prefix is optional, just as for "file". */
1839 strstart(filename, "host_device:", &filename);
1840
1841 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
1842 }
1843
1844 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
1845 Error **errp)
1846 {
1847 BDRVRawState *s = bs->opaque;
1848 Error *local_err = NULL;
1849 int ret;
1850 const char *filename = qdict_get_str(options, "filename");
1851
1852 #if defined(__APPLE__) && defined(__MACH__)
1853 if (strstart(filename, "/dev/cdrom", NULL)) {
1854 kern_return_t kernResult;
1855 io_iterator_t mediaIterator;
1856 char bsdPath[ MAXPATHLEN ];
1857 int fd;
1858
1859 kernResult = FindEjectableCDMedia( &mediaIterator );
1860 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1861
1862 if ( bsdPath[ 0 ] != '\0' ) {
1863 strcat(bsdPath,"s0");
1864 /* some CDs don't have a partition 0 */
1865 fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1866 if (fd < 0) {
1867 bsdPath[strlen(bsdPath)-1] = '1';
1868 } else {
1869 qemu_close(fd);
1870 }
1871 filename = bsdPath;
1872 qdict_put(options, "filename", qstring_from_str(filename));
1873 }
1874
1875 if ( mediaIterator )
1876 IOObjectRelease( mediaIterator );
1877 }
1878 #endif
1879
1880 s->type = FTYPE_FILE;
1881 #if defined(__linux__)
1882 {
1883 char resolved_path[ MAXPATHLEN ], *temp;
1884
1885 temp = realpath(filename, resolved_path);
1886 if (temp && strstart(temp, "/dev/sg", NULL)) {
1887 bs->sg = 1;
1888 }
1889 }
1890 #endif
1891
1892 ret = raw_open_common(bs, options, flags, 0, &local_err);
1893 if (ret < 0) {
1894 if (local_err) {
1895 error_propagate(errp, local_err);
1896 }
1897 return ret;
1898 }
1899
1900 if (flags & BDRV_O_RDWR) {
1901 ret = check_hdev_writable(s);
1902 if (ret < 0) {
1903 raw_close(bs);
1904 error_setg_errno(errp, -ret, "The device is not writable");
1905 return ret;
1906 }
1907 }
1908
1909 return ret;
1910 }
1911
1912 #if defined(__linux__)
1913 /* Note: we do not have a reliable method to detect if the floppy is
1914 present. The current method is to try to open the floppy at every
1915 I/O and to keep it opened during a few hundreds of ms. */
1916 static int fd_open(BlockDriverState *bs)
1917 {
1918 BDRVRawState *s = bs->opaque;
1919 int last_media_present;
1920
1921 if (s->type != FTYPE_FD)
1922 return 0;
1923 last_media_present = (s->fd >= 0);
1924 if (s->fd >= 0 &&
1925 (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1926 qemu_close(s->fd);
1927 s->fd = -1;
1928 #ifdef DEBUG_FLOPPY
1929 printf("Floppy closed\n");
1930 #endif
1931 }
1932 if (s->fd < 0) {
1933 if (s->fd_got_error &&
1934 (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1935 #ifdef DEBUG_FLOPPY
1936 printf("No floppy (open delayed)\n");
1937 #endif
1938 return -EIO;
1939 }
1940 s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
1941 if (s->fd < 0) {
1942 s->fd_error_time = get_clock();
1943 s->fd_got_error = 1;
1944 if (last_media_present)
1945 s->fd_media_changed = 1;
1946 #ifdef DEBUG_FLOPPY
1947 printf("No floppy\n");
1948 #endif
1949 return -EIO;
1950 }
1951 #ifdef DEBUG_FLOPPY
1952 printf("Floppy opened\n");
1953 #endif
1954 }
1955 if (!last_media_present)
1956 s->fd_media_changed = 1;
1957 s->fd_open_time = get_clock();
1958 s->fd_got_error = 0;
1959 return 0;
1960 }
1961
1962 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1963 {
1964 BDRVRawState *s = bs->opaque;
1965
1966 return ioctl(s->fd, req, buf);
1967 }
1968
1969 static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1970 unsigned long int req, void *buf,
1971 BlockCompletionFunc *cb, void *opaque)
1972 {
1973 BDRVRawState *s = bs->opaque;
1974 RawPosixAIOData *acb;
1975 ThreadPool *pool;
1976
1977 if (fd_open(bs) < 0)
1978 return NULL;
1979
1980 acb = g_slice_new(RawPosixAIOData);
1981 acb->bs = bs;
1982 acb->aio_type = QEMU_AIO_IOCTL;
1983 acb->aio_fildes = s->fd;
1984 acb->aio_offset = 0;
1985 acb->aio_ioctl_buf = buf;
1986 acb->aio_ioctl_cmd = req;
1987 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1988 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1989 }
1990
1991 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1992 static int fd_open(BlockDriverState *bs)
1993 {
1994 BDRVRawState *s = bs->opaque;
1995
1996 /* this is just to ensure s->fd is sane (its called by io ops) */
1997 if (s->fd >= 0)
1998 return 0;
1999 return -EIO;
2000 }
2001 #else /* !linux && !FreeBSD */
2002
2003 static int fd_open(BlockDriverState *bs)
2004 {
2005 return 0;
2006 }
2007
2008 #endif /* !linux && !FreeBSD */
2009
2010 static coroutine_fn BlockAIOCB *hdev_aio_discard(BlockDriverState *bs,
2011 int64_t sector_num, int nb_sectors,
2012 BlockCompletionFunc *cb, void *opaque)
2013 {
2014 BDRVRawState *s = bs->opaque;
2015
2016 if (fd_open(bs) < 0) {
2017 return NULL;
2018 }
2019 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
2020 cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2021 }
2022
2023 static coroutine_fn int hdev_co_write_zeroes(BlockDriverState *bs,
2024 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2025 {
2026 BDRVRawState *s = bs->opaque;
2027 int rc;
2028
2029 rc = fd_open(bs);
2030 if (rc < 0) {
2031 return rc;
2032 }
2033 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2034 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2035 QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2036 } else if (s->discard_zeroes) {
2037 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2038 QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2039 }
2040 return -ENOTSUP;
2041 }
2042
2043 static int hdev_create(const char *filename, QemuOpts *opts,
2044 Error **errp)
2045 {
2046 int fd;
2047 int ret = 0;
2048 struct stat stat_buf;
2049 int64_t total_size = 0;
2050 bool has_prefix;
2051
2052 /* This function is used by all three protocol block drivers and therefore
2053 * any of these three prefixes may be given.
2054 * The return value has to be stored somewhere, otherwise this is an error
2055 * due to -Werror=unused-value. */
2056 has_prefix =
2057 strstart(filename, "host_device:", &filename) ||
2058 strstart(filename, "host_cdrom:" , &filename) ||
2059 strstart(filename, "host_floppy:", &filename);
2060
2061 (void)has_prefix;
2062
2063 /* Read out options */
2064 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2065 BDRV_SECTOR_SIZE);
2066
2067 fd = qemu_open(filename, O_WRONLY | O_BINARY);
2068 if (fd < 0) {
2069 ret = -errno;
2070 error_setg_errno(errp, -ret, "Could not open device");
2071 return ret;
2072 }
2073
2074 if (fstat(fd, &stat_buf) < 0) {
2075 ret = -errno;
2076 error_setg_errno(errp, -ret, "Could not stat device");
2077 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2078 error_setg(errp,
2079 "The given file is neither a block nor a character device");
2080 ret = -ENODEV;
2081 } else if (lseek(fd, 0, SEEK_END) < total_size) {
2082 error_setg(errp, "Device is too small");
2083 ret = -ENOSPC;
2084 }
2085
2086 qemu_close(fd);
2087 return ret;
2088 }
2089
2090 static BlockDriver bdrv_host_device = {
2091 .format_name = "host_device",
2092 .protocol_name = "host_device",
2093 .instance_size = sizeof(BDRVRawState),
2094 .bdrv_needs_filename = true,
2095 .bdrv_probe_device = hdev_probe_device,
2096 .bdrv_parse_filename = hdev_parse_filename,
2097 .bdrv_file_open = hdev_open,
2098 .bdrv_close = raw_close,
2099 .bdrv_reopen_prepare = raw_reopen_prepare,
2100 .bdrv_reopen_commit = raw_reopen_commit,
2101 .bdrv_reopen_abort = raw_reopen_abort,
2102 .bdrv_create = hdev_create,
2103 .create_opts = &raw_create_opts,
2104 .bdrv_co_write_zeroes = hdev_co_write_zeroes,
2105
2106 .bdrv_aio_readv = raw_aio_readv,
2107 .bdrv_aio_writev = raw_aio_writev,
2108 .bdrv_aio_flush = raw_aio_flush,
2109 .bdrv_aio_discard = hdev_aio_discard,
2110 .bdrv_refresh_limits = raw_refresh_limits,
2111 .bdrv_io_plug = raw_aio_plug,
2112 .bdrv_io_unplug = raw_aio_unplug,
2113 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2114
2115 .bdrv_truncate = raw_truncate,
2116 .bdrv_getlength = raw_getlength,
2117 .bdrv_get_info = raw_get_info,
2118 .bdrv_get_allocated_file_size
2119 = raw_get_allocated_file_size,
2120
2121 .bdrv_detach_aio_context = raw_detach_aio_context,
2122 .bdrv_attach_aio_context = raw_attach_aio_context,
2123
2124 /* generic scsi device */
2125 #ifdef __linux__
2126 .bdrv_ioctl = hdev_ioctl,
2127 .bdrv_aio_ioctl = hdev_aio_ioctl,
2128 #endif
2129 };
2130
2131 #ifdef __linux__
2132 static void floppy_parse_filename(const char *filename, QDict *options,
2133 Error **errp)
2134 {
2135 /* The prefix is optional, just as for "file". */
2136 strstart(filename, "host_floppy:", &filename);
2137
2138 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2139 }
2140
2141 static int floppy_open(BlockDriverState *bs, QDict *options, int flags,
2142 Error **errp)
2143 {
2144 BDRVRawState *s = bs->opaque;
2145 Error *local_err = NULL;
2146 int ret;
2147
2148 s->type = FTYPE_FD;
2149
2150 /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
2151 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2152 if (ret) {
2153 if (local_err) {
2154 error_propagate(errp, local_err);
2155 }
2156 return ret;
2157 }
2158
2159 /* close fd so that we can reopen it as needed */
2160 qemu_close(s->fd);
2161 s->fd = -1;
2162 s->fd_media_changed = 1;
2163
2164 return 0;
2165 }
2166
2167 static int floppy_probe_device(const char *filename)
2168 {
2169 int fd, ret;
2170 int prio = 0;
2171 struct floppy_struct fdparam;
2172 struct stat st;
2173
2174 if (strstart(filename, "/dev/fd", NULL) &&
2175 !strstart(filename, "/dev/fdset/", NULL)) {
2176 prio = 50;
2177 }
2178
2179 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2180 if (fd < 0) {
2181 goto out;
2182 }
2183 ret = fstat(fd, &st);
2184 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2185 goto outc;
2186 }
2187
2188 /* Attempt to detect via a floppy specific ioctl */
2189 ret = ioctl(fd, FDGETPRM, &fdparam);
2190 if (ret >= 0)
2191 prio = 100;
2192
2193 outc:
2194 qemu_close(fd);
2195 out:
2196 return prio;
2197 }
2198
2199
2200 static int floppy_is_inserted(BlockDriverState *bs)
2201 {
2202 return fd_open(bs) >= 0;
2203 }
2204
2205 static int floppy_media_changed(BlockDriverState *bs)
2206 {
2207 BDRVRawState *s = bs->opaque;
2208 int ret;
2209
2210 /*
2211 * XXX: we do not have a true media changed indication.
2212 * It does not work if the floppy is changed without trying to read it.
2213 */
2214 fd_open(bs);
2215 ret = s->fd_media_changed;
2216 s->fd_media_changed = 0;
2217 #ifdef DEBUG_FLOPPY
2218 printf("Floppy changed=%d\n", ret);
2219 #endif
2220 return ret;
2221 }
2222
2223 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
2224 {
2225 BDRVRawState *s = bs->opaque;
2226 int fd;
2227
2228 if (s->fd >= 0) {
2229 qemu_close(s->fd);
2230 s->fd = -1;
2231 }
2232 fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
2233 if (fd >= 0) {
2234 if (ioctl(fd, FDEJECT, 0) < 0)
2235 perror("FDEJECT");
2236 qemu_close(fd);
2237 }
2238 }
2239
2240 static BlockDriver bdrv_host_floppy = {
2241 .format_name = "host_floppy",
2242 .protocol_name = "host_floppy",
2243 .instance_size = sizeof(BDRVRawState),
2244 .bdrv_needs_filename = true,
2245 .bdrv_probe_device = floppy_probe_device,
2246 .bdrv_parse_filename = floppy_parse_filename,
2247 .bdrv_file_open = floppy_open,
2248 .bdrv_close = raw_close,
2249 .bdrv_reopen_prepare = raw_reopen_prepare,
2250 .bdrv_reopen_commit = raw_reopen_commit,
2251 .bdrv_reopen_abort = raw_reopen_abort,
2252 .bdrv_create = hdev_create,
2253 .create_opts = &raw_create_opts,
2254
2255 .bdrv_aio_readv = raw_aio_readv,
2256 .bdrv_aio_writev = raw_aio_writev,
2257 .bdrv_aio_flush = raw_aio_flush,
2258 .bdrv_refresh_limits = raw_refresh_limits,
2259 .bdrv_io_plug = raw_aio_plug,
2260 .bdrv_io_unplug = raw_aio_unplug,
2261 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2262
2263 .bdrv_truncate = raw_truncate,
2264 .bdrv_getlength = raw_getlength,
2265 .has_variable_length = true,
2266 .bdrv_get_allocated_file_size
2267 = raw_get_allocated_file_size,
2268
2269 .bdrv_detach_aio_context = raw_detach_aio_context,
2270 .bdrv_attach_aio_context = raw_attach_aio_context,
2271
2272 /* removable device support */
2273 .bdrv_is_inserted = floppy_is_inserted,
2274 .bdrv_media_changed = floppy_media_changed,
2275 .bdrv_eject = floppy_eject,
2276 };
2277 #endif
2278
2279 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2280 static void cdrom_parse_filename(const char *filename, QDict *options,
2281 Error **errp)
2282 {
2283 /* The prefix is optional, just as for "file". */
2284 strstart(filename, "host_cdrom:", &filename);
2285
2286 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2287 }
2288 #endif
2289
2290 #ifdef __linux__
2291 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2292 Error **errp)
2293 {
2294 BDRVRawState *s = bs->opaque;
2295 Error *local_err = NULL;
2296 int ret;
2297
2298 s->type = FTYPE_CD;
2299
2300 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2301 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2302 if (local_err) {
2303 error_propagate(errp, local_err);
2304 }
2305 return ret;
2306 }
2307
2308 static int cdrom_probe_device(const char *filename)
2309 {
2310 int fd, ret;
2311 int prio = 0;
2312 struct stat st;
2313
2314 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2315 if (fd < 0) {
2316 goto out;
2317 }
2318 ret = fstat(fd, &st);
2319 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2320 goto outc;
2321 }
2322
2323 /* Attempt to detect via a CDROM specific ioctl */
2324 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2325 if (ret >= 0)
2326 prio = 100;
2327
2328 outc:
2329 qemu_close(fd);
2330 out:
2331 return prio;
2332 }
2333
2334 static int cdrom_is_inserted(BlockDriverState *bs)
2335 {
2336 BDRVRawState *s = bs->opaque;
2337 int ret;
2338
2339 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2340 if (ret == CDS_DISC_OK)
2341 return 1;
2342 return 0;
2343 }
2344
2345 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2346 {
2347 BDRVRawState *s = bs->opaque;
2348
2349 if (eject_flag) {
2350 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2351 perror("CDROMEJECT");
2352 } else {
2353 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2354 perror("CDROMEJECT");
2355 }
2356 }
2357
2358 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2359 {
2360 BDRVRawState *s = bs->opaque;
2361
2362 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2363 /*
2364 * Note: an error can happen if the distribution automatically
2365 * mounts the CD-ROM
2366 */
2367 /* perror("CDROM_LOCKDOOR"); */
2368 }
2369 }
2370
2371 static BlockDriver bdrv_host_cdrom = {
2372 .format_name = "host_cdrom",
2373 .protocol_name = "host_cdrom",
2374 .instance_size = sizeof(BDRVRawState),
2375 .bdrv_needs_filename = true,
2376 .bdrv_probe_device = cdrom_probe_device,
2377 .bdrv_parse_filename = cdrom_parse_filename,
2378 .bdrv_file_open = cdrom_open,
2379 .bdrv_close = raw_close,
2380 .bdrv_reopen_prepare = raw_reopen_prepare,
2381 .bdrv_reopen_commit = raw_reopen_commit,
2382 .bdrv_reopen_abort = raw_reopen_abort,
2383 .bdrv_create = hdev_create,
2384 .create_opts = &raw_create_opts,
2385
2386 .bdrv_aio_readv = raw_aio_readv,
2387 .bdrv_aio_writev = raw_aio_writev,
2388 .bdrv_aio_flush = raw_aio_flush,
2389 .bdrv_refresh_limits = raw_refresh_limits,
2390 .bdrv_io_plug = raw_aio_plug,
2391 .bdrv_io_unplug = raw_aio_unplug,
2392 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2393
2394 .bdrv_truncate = raw_truncate,
2395 .bdrv_getlength = raw_getlength,
2396 .has_variable_length = true,
2397 .bdrv_get_allocated_file_size
2398 = raw_get_allocated_file_size,
2399
2400 .bdrv_detach_aio_context = raw_detach_aio_context,
2401 .bdrv_attach_aio_context = raw_attach_aio_context,
2402
2403 /* removable device support */
2404 .bdrv_is_inserted = cdrom_is_inserted,
2405 .bdrv_eject = cdrom_eject,
2406 .bdrv_lock_medium = cdrom_lock_medium,
2407
2408 /* generic scsi device */
2409 .bdrv_ioctl = hdev_ioctl,
2410 .bdrv_aio_ioctl = hdev_aio_ioctl,
2411 };
2412 #endif /* __linux__ */
2413
2414 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2415 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2416 Error **errp)
2417 {
2418 BDRVRawState *s = bs->opaque;
2419 Error *local_err = NULL;
2420 int ret;
2421
2422 s->type = FTYPE_CD;
2423
2424 ret = raw_open_common(bs, options, flags, 0, &local_err);
2425 if (ret) {
2426 if (local_err) {
2427 error_propagate(errp, local_err);
2428 }
2429 return ret;
2430 }
2431
2432 /* make sure the door isn't locked at this time */
2433 ioctl(s->fd, CDIOCALLOW);
2434 return 0;
2435 }
2436
2437 static int cdrom_probe_device(const char *filename)
2438 {
2439 if (strstart(filename, "/dev/cd", NULL) ||
2440 strstart(filename, "/dev/acd", NULL))
2441 return 100;
2442 return 0;
2443 }
2444
2445 static int cdrom_reopen(BlockDriverState *bs)
2446 {
2447 BDRVRawState *s = bs->opaque;
2448 int fd;
2449
2450 /*
2451 * Force reread of possibly changed/newly loaded disc,
2452 * FreeBSD seems to not notice sometimes...
2453 */
2454 if (s->fd >= 0)
2455 qemu_close(s->fd);
2456 fd = qemu_open(bs->filename, s->open_flags, 0644);
2457 if (fd < 0) {
2458 s->fd = -1;
2459 return -EIO;
2460 }
2461 s->fd = fd;
2462
2463 /* make sure the door isn't locked at this time */
2464 ioctl(s->fd, CDIOCALLOW);
2465 return 0;
2466 }
2467
2468 static int cdrom_is_inserted(BlockDriverState *bs)
2469 {
2470 return raw_getlength(bs) > 0;
2471 }
2472
2473 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2474 {
2475 BDRVRawState *s = bs->opaque;
2476
2477 if (s->fd < 0)
2478 return;
2479
2480 (void) ioctl(s->fd, CDIOCALLOW);
2481
2482 if (eject_flag) {
2483 if (ioctl(s->fd, CDIOCEJECT) < 0)
2484 perror("CDIOCEJECT");
2485 } else {
2486 if (ioctl(s->fd, CDIOCCLOSE) < 0)
2487 perror("CDIOCCLOSE");
2488 }
2489
2490 cdrom_reopen(bs);
2491 }
2492
2493 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2494 {
2495 BDRVRawState *s = bs->opaque;
2496
2497 if (s->fd < 0)
2498 return;
2499 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2500 /*
2501 * Note: an error can happen if the distribution automatically
2502 * mounts the CD-ROM
2503 */
2504 /* perror("CDROM_LOCKDOOR"); */
2505 }
2506 }
2507
2508 static BlockDriver bdrv_host_cdrom = {
2509 .format_name = "host_cdrom",
2510 .protocol_name = "host_cdrom",
2511 .instance_size = sizeof(BDRVRawState),
2512 .bdrv_needs_filename = true,
2513 .bdrv_probe_device = cdrom_probe_device,
2514 .bdrv_parse_filename = cdrom_parse_filename,
2515 .bdrv_file_open = cdrom_open,
2516 .bdrv_close = raw_close,
2517 .bdrv_reopen_prepare = raw_reopen_prepare,
2518 .bdrv_reopen_commit = raw_reopen_commit,
2519 .bdrv_reopen_abort = raw_reopen_abort,
2520 .bdrv_create = hdev_create,
2521 .create_opts = &raw_create_opts,
2522
2523 .bdrv_aio_readv = raw_aio_readv,
2524 .bdrv_aio_writev = raw_aio_writev,
2525 .bdrv_aio_flush = raw_aio_flush,
2526 .bdrv_refresh_limits = raw_refresh_limits,
2527 .bdrv_io_plug = raw_aio_plug,
2528 .bdrv_io_unplug = raw_aio_unplug,
2529 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2530
2531 .bdrv_truncate = raw_truncate,
2532 .bdrv_getlength = raw_getlength,
2533 .has_variable_length = true,
2534 .bdrv_get_allocated_file_size
2535 = raw_get_allocated_file_size,
2536
2537 .bdrv_detach_aio_context = raw_detach_aio_context,
2538 .bdrv_attach_aio_context = raw_attach_aio_context,
2539
2540 /* removable device support */
2541 .bdrv_is_inserted = cdrom_is_inserted,
2542 .bdrv_eject = cdrom_eject,
2543 .bdrv_lock_medium = cdrom_lock_medium,
2544 };
2545 #endif /* __FreeBSD__ */
2546
2547 static void bdrv_file_init(void)
2548 {
2549 /*
2550 * Register all the drivers. Note that order is important, the driver
2551 * registered last will get probed first.
2552 */
2553 bdrv_register(&bdrv_file);
2554 bdrv_register(&bdrv_host_device);
2555 #ifdef __linux__
2556 bdrv_register(&bdrv_host_floppy);
2557 bdrv_register(&bdrv_host_cdrom);
2558 #endif
2559 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2560 bdrv_register(&bdrv_host_cdrom);
2561 #endif
2562 }
2563
2564 block_init(bdrv_file_init);
QEMU Monitor Protocol