?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
qemu-img: Remove redundancy "ret = -1"
[qemu/kevin.git] / block / raw-posix.c
blob2bcc73dc37163d005961c78e207ca16c64bcc5cd
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
34 #if defined(__APPLE__) && (__MACH__)
35 #include <paths.h>
36 #include <sys/param.h>
37 #include <IOKit/IOKitLib.h>
38 #include <IOKit/IOBSD.h>
39 #include <IOKit/storage/IOMediaBSDClient.h>
40 #include <IOKit/storage/IOMedia.h>
41 #include <IOKit/storage/IOCDMedia.h>
42 //#include <IOKit/storage/IOCDTypes.h>
43 #include <CoreFoundation/CoreFoundation.h>
44 #endif
45
46 #ifdef __sun__
47 #define _POSIX_PTHREAD_SEMANTICS 1
48 #include <sys/dkio.h>
49 #endif
50 #ifdef __linux__
51 #include <sys/types.h>
52 #include <sys/stat.h>
53 #include <sys/ioctl.h>
54 #include <sys/param.h>
55 #include <linux/cdrom.h>
56 #include <linux/fd.h>
57 #include <linux/fs.h>
58 #ifndef FS_NOCOW_FL
59 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
60 #endif
61 #endif
62 #ifdef CONFIG_FIEMAP
63 #include <linux/fiemap.h>
64 #endif
65 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
66 #include <linux/falloc.h>
67 #endif
68 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
69 #include <sys/disk.h>
70 #include <sys/cdio.h>
71 #endif
72
73 #ifdef __OpenBSD__
74 #include <sys/ioctl.h>
75 #include <sys/disklabel.h>
76 #include <sys/dkio.h>
77 #endif
78
79 #ifdef __NetBSD__
80 #include <sys/ioctl.h>
81 #include <sys/disklabel.h>
82 #include <sys/dkio.h>
83 #include <sys/disk.h>
84 #endif
85
86 #ifdef __DragonFly__
87 #include <sys/ioctl.h>
88 #include <sys/diskslice.h>
89 #endif
90
91 #ifdef CONFIG_XFS
92 #include <xfs/xfs.h>
93 #endif
94
95 //#define DEBUG_FLOPPY
96
97 //#define DEBUG_BLOCK
98 #if defined(DEBUG_BLOCK)
99 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
100 { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
101 #else
102 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
103 #endif
104
105 /* OS X does not have O_DSYNC */
106 #ifndef O_DSYNC
107 #ifdef O_SYNC
108 #define O_DSYNC O_SYNC
109 #elif defined(O_FSYNC)
110 #define O_DSYNC O_FSYNC
111 #endif
112 #endif
113
114 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
115 #ifndef O_DIRECT
116 #define O_DIRECT O_DSYNC
117 #endif
118
119 #define FTYPE_FILE 0
120 #define FTYPE_CD 1
121 #define FTYPE_FD 2
122
123 /* if the FD is not accessed during that time (in ns), we try to
124 reopen it to see if the disk has been changed */
125 #define FD_OPEN_TIMEOUT (1000000000)
126
127 #define MAX_BLOCKSIZE 4096
128
129 typedef struct BDRVRawState {
130 int fd;
131 int type;
132 int open_flags;
133 size_t buf_align;
134
135 #if defined(__linux__)
136 /* linux floppy specific */
137 int64_t fd_open_time;
138 int64_t fd_error_time;
139 int fd_got_error;
140 int fd_media_changed;
141 #endif
142 #ifdef CONFIG_LINUX_AIO
143 int use_aio;
144 void *aio_ctx;
145 #endif
146 #ifdef CONFIG_XFS
147 bool is_xfs:1;
148 #endif
149 bool has_discard:1;
150 bool has_write_zeroes:1;
151 bool discard_zeroes:1;
152 #ifdef CONFIG_FIEMAP
153 bool skip_fiemap;
154 #endif
155 } BDRVRawState;
156
157 typedef struct BDRVRawReopenState {
158 int fd;
159 int open_flags;
160 #ifdef CONFIG_LINUX_AIO
161 int use_aio;
162 #endif
163 } BDRVRawReopenState;
164
165 static int fd_open(BlockDriverState *bs);
166 static int64_t raw_getlength(BlockDriverState *bs);
167
168 typedef struct RawPosixAIOData {
169 BlockDriverState *bs;
170 int aio_fildes;
171 union {
172 struct iovec *aio_iov;
173 void *aio_ioctl_buf;
174 };
175 int aio_niov;
176 uint64_t aio_nbytes;
177 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
178 off_t aio_offset;
179 int aio_type;
180 } RawPosixAIOData;
181
182 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
183 static int cdrom_reopen(BlockDriverState *bs);
184 #endif
185
186 #if defined(__NetBSD__)
187 static int raw_normalize_devicepath(const char **filename)
188 {
189 static char namebuf[PATH_MAX];
190 const char *dp, *fname;
191 struct stat sb;
192
193 fname = *filename;
194 dp = strrchr(fname, '/');
195 if (lstat(fname, &sb) < 0) {
196 fprintf(stderr, "%s: stat failed: %s\n",
197 fname, strerror(errno));
198 return -errno;
199 }
200
201 if (!S_ISBLK(sb.st_mode)) {
202 return 0;
203 }
204
205 if (dp == NULL) {
206 snprintf(namebuf, PATH_MAX, "r%s", fname);
207 } else {
208 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
209 (int)(dp - fname), fname, dp + 1);
210 }
211 fprintf(stderr, "%s is a block device", fname);
212 *filename = namebuf;
213 fprintf(stderr, ", using %s\n", *filename);
214
215 return 0;
216 }
217 #else
218 static int raw_normalize_devicepath(const char **filename)
219 {
220 return 0;
221 }
222 #endif
223
224 static void raw_probe_alignment(BlockDriverState *bs)
225 {
226 BDRVRawState *s = bs->opaque;
227 char *buf;
228 unsigned int sector_size;
229
230 /* For /dev/sg devices the alignment is not really used.
231 With buffered I/O, we don't have any restrictions. */
232 if (bs->sg || !(s->open_flags & O_DIRECT)) {
233 bs->request_alignment = 1;
234 s->buf_align = 1;
235 return;
236 }
237
238 /* Try a few ioctls to get the right size */
239 bs->request_alignment = 0;
240 s->buf_align = 0;
241
242 #ifdef BLKSSZGET
243 if (ioctl(s->fd, BLKSSZGET, &sector_size) >= 0) {
244 bs->request_alignment = sector_size;
245 }
246 #endif
247 #ifdef DKIOCGETBLOCKSIZE
248 if (ioctl(s->fd, DKIOCGETBLOCKSIZE, &sector_size) >= 0) {
249 bs->request_alignment = sector_size;
250 }
251 #endif
252 #ifdef DIOCGSECTORSIZE
253 if (ioctl(s->fd, DIOCGSECTORSIZE, &sector_size) >= 0) {
254 bs->request_alignment = sector_size;
255 }
256 #endif
257 #ifdef CONFIG_XFS
258 if (s->is_xfs) {
259 struct dioattr da;
260 if (xfsctl(NULL, s->fd, XFS_IOC_DIOINFO, &da) >= 0) {
261 bs->request_alignment = da.d_miniosz;
262 /* The kernel returns wrong information for d_mem */
263 /* s->buf_align = da.d_mem; */
264 }
265 }
266 #endif
267
268 /* If we could not get the sizes so far, we can only guess them */
269 if (!s->buf_align) {
270 size_t align;
271 buf = qemu_memalign(MAX_BLOCKSIZE, 2 * MAX_BLOCKSIZE);
272 for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
273 if (pread(s->fd, buf + align, MAX_BLOCKSIZE, 0) >= 0) {
274 s->buf_align = align;
275 break;
276 }
277 }
278 qemu_vfree(buf);
279 }
280
281 if (!bs->request_alignment) {
282 size_t align;
283 buf = qemu_memalign(s->buf_align, MAX_BLOCKSIZE);
284 for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
285 if (pread(s->fd, buf, align, 0) >= 0) {
286 bs->request_alignment = align;
287 break;
288 }
289 }
290 qemu_vfree(buf);
291 }
292 }
293
294 static void raw_parse_flags(int bdrv_flags, int *open_flags)
295 {
296 assert(open_flags != NULL);
297
298 *open_flags |= O_BINARY;
299 *open_flags &= ~O_ACCMODE;
300 if (bdrv_flags & BDRV_O_RDWR) {
301 *open_flags |= O_RDWR;
302 } else {
303 *open_flags |= O_RDONLY;
304 }
305
306 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
307 * and O_DIRECT for no caching. */
308 if ((bdrv_flags & BDRV_O_NOCACHE)) {
309 *open_flags |= O_DIRECT;
310 }
311 }
312
313 static void raw_detach_aio_context(BlockDriverState *bs)
314 {
315 #ifdef CONFIG_LINUX_AIO
316 BDRVRawState *s = bs->opaque;
317
318 if (s->use_aio) {
319 laio_detach_aio_context(s->aio_ctx, bdrv_get_aio_context(bs));
320 }
321 #endif
322 }
323
324 static void raw_attach_aio_context(BlockDriverState *bs,
325 AioContext *new_context)
326 {
327 #ifdef CONFIG_LINUX_AIO
328 BDRVRawState *s = bs->opaque;
329
330 if (s->use_aio) {
331 laio_attach_aio_context(s->aio_ctx, new_context);
332 }
333 #endif
334 }
335
336 #ifdef CONFIG_LINUX_AIO
337 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
338 {
339 int ret = -1;
340 assert(aio_ctx != NULL);
341 assert(use_aio != NULL);
342 /*
343 * Currently Linux do AIO only for files opened with O_DIRECT
344 * specified so check NOCACHE flag too
345 */
346 if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
347 (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
348
349 /* if non-NULL, laio_init() has already been run */
350 if (*aio_ctx == NULL) {
351 *aio_ctx = laio_init();
352 if (!*aio_ctx) {
353 goto error;
354 }
355 }
356 *use_aio = 1;
357 } else {
358 *use_aio = 0;
359 }
360
361 ret = 0;
362
363 error:
364 return ret;
365 }
366 #endif
367
368 static void raw_parse_filename(const char *filename, QDict *options,
369 Error **errp)
370 {
371 /* The filename does not have to be prefixed by the protocol name, since
372 * "file" is the default protocol; therefore, the return value of this
373 * function call can be ignored. */
374 strstart(filename, "file:", &filename);
375
376 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
377 }
378
379 static QemuOptsList raw_runtime_opts = {
380 .name = "raw",
381 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
382 .desc = {
383 {
384 .name = "filename",
385 .type = QEMU_OPT_STRING,
386 .help = "File name of the image",
387 },
388 { /* end of list */ }
389 },
390 };
391
392 static int raw_open_common(BlockDriverState *bs, QDict *options,
393 int bdrv_flags, int open_flags, Error **errp)
394 {
395 BDRVRawState *s = bs->opaque;
396 QemuOpts *opts;
397 Error *local_err = NULL;
398 const char *filename = NULL;
399 int fd, ret;
400 struct stat st;
401
402 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
403 qemu_opts_absorb_qdict(opts, options, &local_err);
404 if (local_err) {
405 error_propagate(errp, local_err);
406 ret = -EINVAL;
407 goto fail;
408 }
409
410 filename = qemu_opt_get(opts, "filename");
411
412 ret = raw_normalize_devicepath(&filename);
413 if (ret != 0) {
414 error_setg_errno(errp, -ret, "Could not normalize device path");
415 goto fail;
416 }
417
418 s->open_flags = open_flags;
419 raw_parse_flags(bdrv_flags, &s->open_flags);
420
421 s->fd = -1;
422 fd = qemu_open(filename, s->open_flags, 0644);
423 if (fd < 0) {
424 ret = -errno;
425 if (ret == -EROFS) {
426 ret = -EACCES;
427 }
428 goto fail;
429 }
430 s->fd = fd;
431
432 #ifdef CONFIG_LINUX_AIO
433 if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
434 qemu_close(fd);
435 ret = -errno;
436 error_setg_errno(errp, -ret, "Could not set AIO state");
437 goto fail;
438 }
439 #endif
440
441 s->has_discard = true;
442 s->has_write_zeroes = true;
443
444 if (fstat(s->fd, &st) < 0) {
445 error_setg_errno(errp, errno, "Could not stat file");
446 goto fail;
447 }
448 if (S_ISREG(st.st_mode)) {
449 s->discard_zeroes = true;
450 }
451 if (S_ISBLK(st.st_mode)) {
452 #ifdef BLKDISCARDZEROES
453 unsigned int arg;
454 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
455 s->discard_zeroes = true;
456 }
457 #endif
458 #ifdef __linux__
459 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
460 * not rely on the contents of discarded blocks unless using O_DIRECT.
461 * Same for BLKZEROOUT.
462 */
463 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
464 s->discard_zeroes = false;
465 s->has_write_zeroes = false;
466 }
467 #endif
468 }
469
470 #ifdef CONFIG_XFS
471 if (platform_test_xfs_fd(s->fd)) {
472 s->is_xfs = true;
473 }
474 #endif
475
476 raw_attach_aio_context(bs, bdrv_get_aio_context(bs));
477
478 ret = 0;
479 fail:
480 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
481 unlink(filename);
482 }
483 qemu_opts_del(opts);
484 return ret;
485 }
486
487 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
488 Error **errp)
489 {
490 BDRVRawState *s = bs->opaque;
491 Error *local_err = NULL;
492 int ret;
493
494 s->type = FTYPE_FILE;
495 ret = raw_open_common(bs, options, flags, 0, &local_err);
496 if (local_err) {
497 error_propagate(errp, local_err);
498 }
499 return ret;
500 }
501
502 static int raw_reopen_prepare(BDRVReopenState *state,
503 BlockReopenQueue *queue, Error **errp)
504 {
505 BDRVRawState *s;
506 BDRVRawReopenState *raw_s;
507 int ret = 0;
508
509 assert(state != NULL);
510 assert(state->bs != NULL);
511
512 s = state->bs->opaque;
513
514 state->opaque = g_malloc0(sizeof(BDRVRawReopenState));
515 raw_s = state->opaque;
516
517 #ifdef CONFIG_LINUX_AIO
518 raw_s->use_aio = s->use_aio;
519
520 /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
521 * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
522 * won't override aio_ctx if aio_ctx is non-NULL */
523 if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
524 error_setg(errp, "Could not set AIO state");
525 return -1;
526 }
527 #endif
528
529 if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
530 raw_s->open_flags |= O_NONBLOCK;
531 }
532
533 raw_parse_flags(state->flags, &raw_s->open_flags);
534
535 raw_s->fd = -1;
536
537 int fcntl_flags = O_APPEND | O_NONBLOCK;
538 #ifdef O_NOATIME
539 fcntl_flags |= O_NOATIME;
540 #endif
541
542 #ifdef O_ASYNC
543 /* Not all operating systems have O_ASYNC, and those that don't
544 * will not let us track the state into raw_s->open_flags (typically
545 * you achieve the same effect with an ioctl, for example I_SETSIG
546 * on Solaris). But we do not use O_ASYNC, so that's fine.
547 */
548 assert((s->open_flags & O_ASYNC) == 0);
549 #endif
550
551 if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
552 /* dup the original fd */
553 /* TODO: use qemu fcntl wrapper */
554 #ifdef F_DUPFD_CLOEXEC
555 raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
556 #else
557 raw_s->fd = dup(s->fd);
558 if (raw_s->fd != -1) {
559 qemu_set_cloexec(raw_s->fd);
560 }
561 #endif
562 if (raw_s->fd >= 0) {
563 ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
564 if (ret) {
565 qemu_close(raw_s->fd);
566 raw_s->fd = -1;
567 }
568 }
569 }
570
571 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
572 if (raw_s->fd == -1) {
573 assert(!(raw_s->open_flags & O_CREAT));
574 raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
575 if (raw_s->fd == -1) {
576 error_setg_errno(errp, errno, "Could not reopen file");
577 ret = -1;
578 }
579 }
580 return ret;
581 }
582
583 static void raw_reopen_commit(BDRVReopenState *state)
584 {
585 BDRVRawReopenState *raw_s = state->opaque;
586 BDRVRawState *s = state->bs->opaque;
587
588 s->open_flags = raw_s->open_flags;
589
590 qemu_close(s->fd);
591 s->fd = raw_s->fd;
592 #ifdef CONFIG_LINUX_AIO
593 s->use_aio = raw_s->use_aio;
594 #endif
595
596 g_free(state->opaque);
597 state->opaque = NULL;
598 }
599
600
601 static void raw_reopen_abort(BDRVReopenState *state)
602 {
603 BDRVRawReopenState *raw_s = state->opaque;
604
605 /* nothing to do if NULL, we didn't get far enough */
606 if (raw_s == NULL) {
607 return;
608 }
609
610 if (raw_s->fd >= 0) {
611 qemu_close(raw_s->fd);
612 raw_s->fd = -1;
613 }
614 g_free(state->opaque);
615 state->opaque = NULL;
616 }
617
618 static int raw_refresh_limits(BlockDriverState *bs)
619 {
620 BDRVRawState *s = bs->opaque;
621
622 raw_probe_alignment(bs);
623 bs->bl.opt_mem_alignment = s->buf_align;
624
625 return 0;
626 }
627
628 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
629 {
630 int ret;
631
632 ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
633 if (ret == -1) {
634 return -errno;
635 }
636
637 return 0;
638 }
639
640 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
641 {
642 int ret;
643
644 ret = qemu_fdatasync(aiocb->aio_fildes);
645 if (ret == -1) {
646 return -errno;
647 }
648 return 0;
649 }
650
651 #ifdef CONFIG_PREADV
652
653 static bool preadv_present = true;
654
655 static ssize_t
656 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
657 {
658 return preadv(fd, iov, nr_iov, offset);
659 }
660
661 static ssize_t
662 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
663 {
664 return pwritev(fd, iov, nr_iov, offset);
665 }
666
667 #else
668
669 static bool preadv_present = false;
670
671 static ssize_t
672 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
673 {
674 return -ENOSYS;
675 }
676
677 static ssize_t
678 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
679 {
680 return -ENOSYS;
681 }
682
683 #endif
684
685 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
686 {
687 ssize_t len;
688
689 do {
690 if (aiocb->aio_type & QEMU_AIO_WRITE)
691 len = qemu_pwritev(aiocb->aio_fildes,
692 aiocb->aio_iov,
693 aiocb->aio_niov,
694 aiocb->aio_offset);
695 else
696 len = qemu_preadv(aiocb->aio_fildes,
697 aiocb->aio_iov,
698 aiocb->aio_niov,
699 aiocb->aio_offset);
700 } while (len == -1 && errno == EINTR);
701
702 if (len == -1) {
703 return -errno;
704 }
705 return len;
706 }
707
708 /*
709 * Read/writes the data to/from a given linear buffer.
710 *
711 * Returns the number of bytes handles or -errno in case of an error. Short
712 * reads are only returned if the end of the file is reached.
713 */
714 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
715 {
716 ssize_t offset = 0;
717 ssize_t len;
718
719 while (offset < aiocb->aio_nbytes) {
720 if (aiocb->aio_type & QEMU_AIO_WRITE) {
721 len = pwrite(aiocb->aio_fildes,
722 (const char *)buf + offset,
723 aiocb->aio_nbytes - offset,
724 aiocb->aio_offset + offset);
725 } else {
726 len = pread(aiocb->aio_fildes,
727 buf + offset,
728 aiocb->aio_nbytes - offset,
729 aiocb->aio_offset + offset);
730 }
731 if (len == -1 && errno == EINTR) {
732 continue;
733 } else if (len == -1) {
734 offset = -errno;
735 break;
736 } else if (len == 0) {
737 break;
738 }
739 offset += len;
740 }
741
742 return offset;
743 }
744
745 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
746 {
747 ssize_t nbytes;
748 char *buf;
749
750 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
751 /*
752 * If there is just a single buffer, and it is properly aligned
753 * we can just use plain pread/pwrite without any problems.
754 */
755 if (aiocb->aio_niov == 1) {
756 return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
757 }
758 /*
759 * We have more than one iovec, and all are properly aligned.
760 *
761 * Try preadv/pwritev first and fall back to linearizing the
762 * buffer if it's not supported.
763 */
764 if (preadv_present) {
765 nbytes = handle_aiocb_rw_vector(aiocb);
766 if (nbytes == aiocb->aio_nbytes ||
767 (nbytes < 0 && nbytes != -ENOSYS)) {
768 return nbytes;
769 }
770 preadv_present = false;
771 }
772
773 /*
774 * XXX(hch): short read/write. no easy way to handle the reminder
775 * using these interfaces. For now retry using plain
776 * pread/pwrite?
777 */
778 }
779
780 /*
781 * Ok, we have to do it the hard way, copy all segments into
782 * a single aligned buffer.
783 */
784 buf = qemu_blockalign(aiocb->bs, aiocb->aio_nbytes);
785 if (aiocb->aio_type & QEMU_AIO_WRITE) {
786 char *p = buf;
787 int i;
788
789 for (i = 0; i < aiocb->aio_niov; ++i) {
790 memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
791 p += aiocb->aio_iov[i].iov_len;
792 }
793 assert(p - buf == aiocb->aio_nbytes);
794 }
795
796 nbytes = handle_aiocb_rw_linear(aiocb, buf);
797 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
798 char *p = buf;
799 size_t count = aiocb->aio_nbytes, copy;
800 int i;
801
802 for (i = 0; i < aiocb->aio_niov && count; ++i) {
803 copy = count;
804 if (copy > aiocb->aio_iov[i].iov_len) {
805 copy = aiocb->aio_iov[i].iov_len;
806 }
807 memcpy(aiocb->aio_iov[i].iov_base, p, copy);
808 assert(count >= copy);
809 p += copy;
810 count -= copy;
811 }
812 assert(count == 0);
813 }
814 qemu_vfree(buf);
815
816 return nbytes;
817 }
818
819 #ifdef CONFIG_XFS
820 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
821 {
822 struct xfs_flock64 fl;
823
824 memset(&fl, 0, sizeof(fl));
825 fl.l_whence = SEEK_SET;
826 fl.l_start = offset;
827 fl.l_len = bytes;
828
829 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
830 DEBUG_BLOCK_PRINT("cannot write zero range (%s)\n", strerror(errno));
831 return -errno;
832 }
833
834 return 0;
835 }
836
837 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
838 {
839 struct xfs_flock64 fl;
840
841 memset(&fl, 0, sizeof(fl));
842 fl.l_whence = SEEK_SET;
843 fl.l_start = offset;
844 fl.l_len = bytes;
845
846 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
847 DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
848 return -errno;
849 }
850
851 return 0;
852 }
853 #endif
854
855 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
856 {
857 int ret = -EOPNOTSUPP;
858 BDRVRawState *s = aiocb->bs->opaque;
859
860 if (s->has_write_zeroes == 0) {
861 return -ENOTSUP;
862 }
863
864 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
865 #ifdef BLKZEROOUT
866 do {
867 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
868 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
869 return 0;
870 }
871 } while (errno == EINTR);
872
873 ret = -errno;
874 #endif
875 } else {
876 #ifdef CONFIG_XFS
877 if (s->is_xfs) {
878 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
879 }
880 #endif
881 }
882
883 if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
884 ret == -ENOTTY) {
885 s->has_write_zeroes = false;
886 ret = -ENOTSUP;
887 }
888 return ret;
889 }
890
891 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
892 {
893 int ret = -EOPNOTSUPP;
894 BDRVRawState *s = aiocb->bs->opaque;
895
896 if (!s->has_discard) {
897 return -ENOTSUP;
898 }
899
900 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
901 #ifdef BLKDISCARD
902 do {
903 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
904 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
905 return 0;
906 }
907 } while (errno == EINTR);
908
909 ret = -errno;
910 #endif
911 } else {
912 #ifdef CONFIG_XFS
913 if (s->is_xfs) {
914 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
915 }
916 #endif
917
918 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
919 do {
920 if (fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
921 aiocb->aio_offset, aiocb->aio_nbytes) == 0) {
922 return 0;
923 }
924 } while (errno == EINTR);
925
926 ret = -errno;
927 #endif
928 }
929
930 if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
931 ret == -ENOTTY) {
932 s->has_discard = false;
933 ret = -ENOTSUP;
934 }
935 return ret;
936 }
937
938 static int aio_worker(void *arg)
939 {
940 RawPosixAIOData *aiocb = arg;
941 ssize_t ret = 0;
942
943 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
944 case QEMU_AIO_READ:
945 ret = handle_aiocb_rw(aiocb);
946 if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) {
947 iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
948 0, aiocb->aio_nbytes - ret);
949
950 ret = aiocb->aio_nbytes;
951 }
952 if (ret == aiocb->aio_nbytes) {
953 ret = 0;
954 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
955 ret = -EINVAL;
956 }
957 break;
958 case QEMU_AIO_WRITE:
959 ret = handle_aiocb_rw(aiocb);
960 if (ret == aiocb->aio_nbytes) {
961 ret = 0;
962 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
963 ret = -EINVAL;
964 }
965 break;
966 case QEMU_AIO_FLUSH:
967 ret = handle_aiocb_flush(aiocb);
968 break;
969 case QEMU_AIO_IOCTL:
970 ret = handle_aiocb_ioctl(aiocb);
971 break;
972 case QEMU_AIO_DISCARD:
973 ret = handle_aiocb_discard(aiocb);
974 break;
975 case QEMU_AIO_WRITE_ZEROES:
976 ret = handle_aiocb_write_zeroes(aiocb);
977 break;
978 default:
979 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
980 ret = -EINVAL;
981 break;
982 }
983
984 g_slice_free(RawPosixAIOData, aiocb);
985 return ret;
986 }
987
988 static int paio_submit_co(BlockDriverState *bs, int fd,
989 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
990 int type)
991 {
992 RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
993 ThreadPool *pool;
994
995 acb->bs = bs;
996 acb->aio_type = type;
997 acb->aio_fildes = fd;
998
999 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1000 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1001
1002 if (qiov) {
1003 acb->aio_iov = qiov->iov;
1004 acb->aio_niov = qiov->niov;
1005 assert(qiov->size == acb->aio_nbytes);
1006 }
1007
1008 trace_paio_submit_co(sector_num, nb_sectors, type);
1009 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1010 return thread_pool_submit_co(pool, aio_worker, acb);
1011 }
1012
1013 static BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd,
1014 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1015 BlockDriverCompletionFunc *cb, void *opaque, int type)
1016 {
1017 RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
1018 ThreadPool *pool;
1019
1020 acb->bs = bs;
1021 acb->aio_type = type;
1022 acb->aio_fildes = fd;
1023
1024 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1025 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1026
1027 if (qiov) {
1028 acb->aio_iov = qiov->iov;
1029 acb->aio_niov = qiov->niov;
1030 assert(qiov->size == acb->aio_nbytes);
1031 }
1032
1033 trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
1034 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1035 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1036 }
1037
1038 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
1039 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1040 BlockDriverCompletionFunc *cb, void *opaque, int type)
1041 {
1042 BDRVRawState *s = bs->opaque;
1043
1044 if (fd_open(bs) < 0)
1045 return NULL;
1046
1047 /*
1048 * If O_DIRECT is used the buffer needs to be aligned on a sector
1049 * boundary. Check if this is the case or tell the low-level
1050 * driver that it needs to copy the buffer.
1051 */
1052 if ((bs->open_flags & BDRV_O_NOCACHE)) {
1053 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1054 type |= QEMU_AIO_MISALIGNED;
1055 #ifdef CONFIG_LINUX_AIO
1056 } else if (s->use_aio) {
1057 return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
1058 nb_sectors, cb, opaque, type);
1059 #endif
1060 }
1061 }
1062
1063 return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
1064 cb, opaque, type);
1065 }
1066
1067 static void raw_aio_plug(BlockDriverState *bs)
1068 {
1069 #ifdef CONFIG_LINUX_AIO
1070 BDRVRawState *s = bs->opaque;
1071 if (s->use_aio) {
1072 laio_io_plug(bs, s->aio_ctx);
1073 }
1074 #endif
1075 }
1076
1077 static void raw_aio_unplug(BlockDriverState *bs)
1078 {
1079 #ifdef CONFIG_LINUX_AIO
1080 BDRVRawState *s = bs->opaque;
1081 if (s->use_aio) {
1082 laio_io_unplug(bs, s->aio_ctx, true);
1083 }
1084 #endif
1085 }
1086
1087 static void raw_aio_flush_io_queue(BlockDriverState *bs)
1088 {
1089 #ifdef CONFIG_LINUX_AIO
1090 BDRVRawState *s = bs->opaque;
1091 if (s->use_aio) {
1092 laio_io_unplug(bs, s->aio_ctx, false);
1093 }
1094 #endif
1095 }
1096
1097 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
1098 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1099 BlockDriverCompletionFunc *cb, void *opaque)
1100 {
1101 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1102 cb, opaque, QEMU_AIO_READ);
1103 }
1104
1105 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
1106 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1107 BlockDriverCompletionFunc *cb, void *opaque)
1108 {
1109 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1110 cb, opaque, QEMU_AIO_WRITE);
1111 }
1112
1113 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
1114 BlockDriverCompletionFunc *cb, void *opaque)
1115 {
1116 BDRVRawState *s = bs->opaque;
1117
1118 if (fd_open(bs) < 0)
1119 return NULL;
1120
1121 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1122 }
1123
1124 static void raw_close(BlockDriverState *bs)
1125 {
1126 BDRVRawState *s = bs->opaque;
1127
1128 raw_detach_aio_context(bs);
1129
1130 #ifdef CONFIG_LINUX_AIO
1131 if (s->use_aio) {
1132 laio_cleanup(s->aio_ctx);
1133 }
1134 #endif
1135 if (s->fd >= 0) {
1136 qemu_close(s->fd);
1137 s->fd = -1;
1138 }
1139 }
1140
1141 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1142 {
1143 BDRVRawState *s = bs->opaque;
1144 struct stat st;
1145
1146 if (fstat(s->fd, &st)) {
1147 return -errno;
1148 }
1149
1150 if (S_ISREG(st.st_mode)) {
1151 if (ftruncate(s->fd, offset) < 0) {
1152 return -errno;
1153 }
1154 } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1155 if (offset > raw_getlength(bs)) {
1156 return -EINVAL;
1157 }
1158 } else {
1159 return -ENOTSUP;
1160 }
1161
1162 return 0;
1163 }
1164
1165 #ifdef __OpenBSD__
1166 static int64_t raw_getlength(BlockDriverState *bs)
1167 {
1168 BDRVRawState *s = bs->opaque;
1169 int fd = s->fd;
1170 struct stat st;
1171
1172 if (fstat(fd, &st))
1173 return -errno;
1174 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1175 struct disklabel dl;
1176
1177 if (ioctl(fd, DIOCGDINFO, &dl))
1178 return -errno;
1179 return (uint64_t)dl.d_secsize *
1180 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1181 } else
1182 return st.st_size;
1183 }
1184 #elif defined(__NetBSD__)
1185 static int64_t raw_getlength(BlockDriverState *bs)
1186 {
1187 BDRVRawState *s = bs->opaque;
1188 int fd = s->fd;
1189 struct stat st;
1190
1191 if (fstat(fd, &st))
1192 return -errno;
1193 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1194 struct dkwedge_info dkw;
1195
1196 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1197 return dkw.dkw_size * 512;
1198 } else {
1199 struct disklabel dl;
1200
1201 if (ioctl(fd, DIOCGDINFO, &dl))
1202 return -errno;
1203 return (uint64_t)dl.d_secsize *
1204 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1205 }
1206 } else
1207 return st.st_size;
1208 }
1209 #elif defined(__sun__)
1210 static int64_t raw_getlength(BlockDriverState *bs)
1211 {
1212 BDRVRawState *s = bs->opaque;
1213 struct dk_minfo minfo;
1214 int ret;
1215 int64_t size;
1216
1217 ret = fd_open(bs);
1218 if (ret < 0) {
1219 return ret;
1220 }
1221
1222 /*
1223 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1224 */
1225 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1226 if (ret != -1) {
1227 return minfo.dki_lbsize * minfo.dki_capacity;
1228 }
1229
1230 /*
1231 * There are reports that lseek on some devices fails, but
1232 * irc discussion said that contingency on contingency was overkill.
1233 */
1234 size = lseek(s->fd, 0, SEEK_END);
1235 if (size < 0) {
1236 return -errno;
1237 }
1238 return size;
1239 }
1240 #elif defined(CONFIG_BSD)
1241 static int64_t raw_getlength(BlockDriverState *bs)
1242 {
1243 BDRVRawState *s = bs->opaque;
1244 int fd = s->fd;
1245 int64_t size;
1246 struct stat sb;
1247 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1248 int reopened = 0;
1249 #endif
1250 int ret;
1251
1252 ret = fd_open(bs);
1253 if (ret < 0)
1254 return ret;
1255
1256 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1257 again:
1258 #endif
1259 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1260 #ifdef DIOCGMEDIASIZE
1261 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1262 #elif defined(DIOCGPART)
1263 {
1264 struct partinfo pi;
1265 if (ioctl(fd, DIOCGPART, &pi) == 0)
1266 size = pi.media_size;
1267 else
1268 size = 0;
1269 }
1270 if (size == 0)
1271 #endif
1272 #if defined(__APPLE__) && defined(__MACH__)
1273 size = LLONG_MAX;
1274 #else
1275 size = lseek(fd, 0LL, SEEK_END);
1276 if (size < 0) {
1277 return -errno;
1278 }
1279 #endif
1280 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1281 switch(s->type) {
1282 case FTYPE_CD:
1283 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1284 if (size == 2048LL * (unsigned)-1)
1285 size = 0;
1286 /* XXX no disc? maybe we need to reopen... */
1287 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1288 reopened = 1;
1289 goto again;
1290 }
1291 }
1292 #endif
1293 } else {
1294 size = lseek(fd, 0, SEEK_END);
1295 if (size < 0) {
1296 return -errno;
1297 }
1298 }
1299 return size;
1300 }
1301 #else
1302 static int64_t raw_getlength(BlockDriverState *bs)
1303 {
1304 BDRVRawState *s = bs->opaque;
1305 int ret;
1306 int64_t size;
1307
1308 ret = fd_open(bs);
1309 if (ret < 0) {
1310 return ret;
1311 }
1312
1313 size = lseek(s->fd, 0, SEEK_END);
1314 if (size < 0) {
1315 return -errno;
1316 }
1317 return size;
1318 }
1319 #endif
1320
1321 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1322 {
1323 struct stat st;
1324 BDRVRawState *s = bs->opaque;
1325
1326 if (fstat(s->fd, &st) < 0) {
1327 return -errno;
1328 }
1329 return (int64_t)st.st_blocks * 512;
1330 }
1331
1332 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
1333 {
1334 int fd;
1335 int result = 0;
1336 int64_t total_size = 0;
1337 bool nocow = false;
1338
1339 strstart(filename, "file:", &filename);
1340
1341 /* Read out options */
1342 total_size =
1343 qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0) / BDRV_SECTOR_SIZE;
1344 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
1345
1346 fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
1347 0644);
1348 if (fd < 0) {
1349 result = -errno;
1350 error_setg_errno(errp, -result, "Could not create file");
1351 } else {
1352 if (nocow) {
1353 #ifdef __linux__
1354 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1355 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1356 * will be ignored since any failure of this operation should not
1357 * block the left work.
1358 */
1359 int attr;
1360 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
1361 attr |= FS_NOCOW_FL;
1362 ioctl(fd, FS_IOC_SETFLAGS, &attr);
1363 }
1364 #endif
1365 }
1366
1367 if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
1368 result = -errno;
1369 error_setg_errno(errp, -result, "Could not resize file");
1370 }
1371 if (qemu_close(fd) != 0) {
1372 result = -errno;
1373 error_setg_errno(errp, -result, "Could not close the new file");
1374 }
1375 }
1376 return result;
1377 }
1378
1379 static int64_t try_fiemap(BlockDriverState *bs, off_t start, off_t *data,
1380 off_t *hole, int nb_sectors, int *pnum)
1381 {
1382 #ifdef CONFIG_FIEMAP
1383 BDRVRawState *s = bs->opaque;
1384 int64_t ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1385 struct {
1386 struct fiemap fm;
1387 struct fiemap_extent fe;
1388 } f;
1389
1390 if (s->skip_fiemap) {
1391 return -ENOTSUP;
1392 }
1393
1394 f.fm.fm_start = start;
1395 f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE;
1396 f.fm.fm_flags = 0;
1397 f.fm.fm_extent_count = 1;
1398 f.fm.fm_reserved = 0;
1399 if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) {
1400 s->skip_fiemap = true;
1401 return -errno;
1402 }
1403
1404 if (f.fm.fm_mapped_extents == 0) {
1405 /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1406 * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1407 */
1408 off_t length = lseek(s->fd, 0, SEEK_END);
1409 *hole = f.fm.fm_start;
1410 *data = MIN(f.fm.fm_start + f.fm.fm_length, length);
1411 } else {
1412 *data = f.fe.fe_logical;
1413 *hole = f.fe.fe_logical + f.fe.fe_length;
1414 if (f.fe.fe_flags & FIEMAP_EXTENT_UNWRITTEN) {
1415 ret |= BDRV_BLOCK_ZERO;
1416 }
1417 }
1418
1419 return ret;
1420 #else
1421 return -ENOTSUP;
1422 #endif
1423 }
1424
1425 static int64_t try_seek_hole(BlockDriverState *bs, off_t start, off_t *data,
1426 off_t *hole, int *pnum)
1427 {
1428 #if defined SEEK_HOLE && defined SEEK_DATA
1429 BDRVRawState *s = bs->opaque;
1430
1431 *hole = lseek(s->fd, start, SEEK_HOLE);
1432 if (*hole == -1) {
1433 /* -ENXIO indicates that sector_num was past the end of the file.
1434 * There is a virtual hole there. */
1435 assert(errno != -ENXIO);
1436
1437 return -errno;
1438 }
1439
1440 if (*hole > start) {
1441 *data = start;
1442 } else {
1443 /* On a hole. We need another syscall to find its end. */
1444 *data = lseek(s->fd, start, SEEK_DATA);
1445 if (*data == -1) {
1446 *data = lseek(s->fd, 0, SEEK_END);
1447 }
1448 }
1449
1450 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1451 #else
1452 return -ENOTSUP;
1453 #endif
1454 }
1455
1456 /*
1457 * Returns true iff the specified sector is present in the disk image. Drivers
1458 * not implementing the functionality are assumed to not support backing files,
1459 * hence all their sectors are reported as allocated.
1460 *
1461 * If 'sector_num' is beyond the end of the disk image the return value is 0
1462 * and 'pnum' is set to 0.
1463 *
1464 * 'pnum' is set to the number of sectors (including and immediately following
1465 * the specified sector) that are known to be in the same
1466 * allocated/unallocated state.
1467 *
1468 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1469 * beyond the end of the disk image it will be clamped.
1470 */
1471 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1472 int64_t sector_num,
1473 int nb_sectors, int *pnum)
1474 {
1475 off_t start, data = 0, hole = 0;
1476 int64_t ret;
1477
1478 ret = fd_open(bs);
1479 if (ret < 0) {
1480 return ret;
1481 }
1482
1483 start = sector_num * BDRV_SECTOR_SIZE;
1484
1485 ret = try_fiemap(bs, start, &data, &hole, nb_sectors, pnum);
1486 if (ret < 0) {
1487 ret = try_seek_hole(bs, start, &data, &hole, pnum);
1488 if (ret < 0) {
1489 /* Assume everything is allocated. */
1490 data = 0;
1491 hole = start + nb_sectors * BDRV_SECTOR_SIZE;
1492 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1493 }
1494 }
1495
1496 if (data <= start) {
1497 /* On a data extent, compute sectors to the end of the extent. */
1498 *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1499 } else {
1500 /* On a hole, compute sectors to the beginning of the next extent. */
1501 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1502 ret &= ~BDRV_BLOCK_DATA;
1503 ret |= BDRV_BLOCK_ZERO;
1504 }
1505
1506 return ret;
1507 }
1508
1509 static coroutine_fn BlockDriverAIOCB *raw_aio_discard(BlockDriverState *bs,
1510 int64_t sector_num, int nb_sectors,
1511 BlockDriverCompletionFunc *cb, void *opaque)
1512 {
1513 BDRVRawState *s = bs->opaque;
1514
1515 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1516 cb, opaque, QEMU_AIO_DISCARD);
1517 }
1518
1519 static int coroutine_fn raw_co_write_zeroes(
1520 BlockDriverState *bs, int64_t sector_num,
1521 int nb_sectors, BdrvRequestFlags flags)
1522 {
1523 BDRVRawState *s = bs->opaque;
1524
1525 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1526 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1527 QEMU_AIO_WRITE_ZEROES);
1528 } else if (s->discard_zeroes) {
1529 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1530 QEMU_AIO_DISCARD);
1531 }
1532 return -ENOTSUP;
1533 }
1534
1535 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1536 {
1537 BDRVRawState *s = bs->opaque;
1538
1539 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1540 bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1541 return 0;
1542 }
1543
1544 static QemuOptsList raw_create_opts = {
1545 .name = "raw-create-opts",
1546 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
1547 .desc = {
1548 {
1549 .name = BLOCK_OPT_SIZE,
1550 .type = QEMU_OPT_SIZE,
1551 .help = "Virtual disk size"
1552 },
1553 {
1554 .name = BLOCK_OPT_NOCOW,
1555 .type = QEMU_OPT_BOOL,
1556 .help = "Turn off copy-on-write (valid only on btrfs)"
1557 },
1558 { /* end of list */ }
1559 }
1560 };
1561
1562 static BlockDriver bdrv_file = {
1563 .format_name = "file",
1564 .protocol_name = "file",
1565 .instance_size = sizeof(BDRVRawState),
1566 .bdrv_needs_filename = true,
1567 .bdrv_probe = NULL, /* no probe for protocols */
1568 .bdrv_parse_filename = raw_parse_filename,
1569 .bdrv_file_open = raw_open,
1570 .bdrv_reopen_prepare = raw_reopen_prepare,
1571 .bdrv_reopen_commit = raw_reopen_commit,
1572 .bdrv_reopen_abort = raw_reopen_abort,
1573 .bdrv_close = raw_close,
1574 .bdrv_create = raw_create,
1575 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1576 .bdrv_co_get_block_status = raw_co_get_block_status,
1577 .bdrv_co_write_zeroes = raw_co_write_zeroes,
1578
1579 .bdrv_aio_readv = raw_aio_readv,
1580 .bdrv_aio_writev = raw_aio_writev,
1581 .bdrv_aio_flush = raw_aio_flush,
1582 .bdrv_aio_discard = raw_aio_discard,
1583 .bdrv_refresh_limits = raw_refresh_limits,
1584 .bdrv_io_plug = raw_aio_plug,
1585 .bdrv_io_unplug = raw_aio_unplug,
1586 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
1587
1588 .bdrv_truncate = raw_truncate,
1589 .bdrv_getlength = raw_getlength,
1590 .bdrv_get_info = raw_get_info,
1591 .bdrv_get_allocated_file_size
1592 = raw_get_allocated_file_size,
1593
1594 .bdrv_detach_aio_context = raw_detach_aio_context,
1595 .bdrv_attach_aio_context = raw_attach_aio_context,
1596
1597 .create_opts = &raw_create_opts,
1598 };
1599
1600 /***********************************************/
1601 /* host device */
1602
1603 #if defined(__APPLE__) && defined(__MACH__)
1604 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1605 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1606
1607 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1608 {
1609 kern_return_t kernResult;
1610 mach_port_t masterPort;
1611 CFMutableDictionaryRef classesToMatch;
1612
1613 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1614 if ( KERN_SUCCESS != kernResult ) {
1615 printf( "IOMasterPort returned %d\n", kernResult );
1616 }
1617
1618 classesToMatch = IOServiceMatching( kIOCDMediaClass );
1619 if ( classesToMatch == NULL ) {
1620 printf( "IOServiceMatching returned a NULL dictionary.\n" );
1621 } else {
1622 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1623 }
1624 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1625 if ( KERN_SUCCESS != kernResult )
1626 {
1627 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1628 }
1629
1630 return kernResult;
1631 }
1632
1633 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
1634 {
1635 io_object_t nextMedia;
1636 kern_return_t kernResult = KERN_FAILURE;
1637 *bsdPath = '\0';
1638 nextMedia = IOIteratorNext( mediaIterator );
1639 if ( nextMedia )
1640 {
1641 CFTypeRef bsdPathAsCFString;
1642 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1643 if ( bsdPathAsCFString ) {
1644 size_t devPathLength;
1645 strcpy( bsdPath, _PATH_DEV );
1646 strcat( bsdPath, "r" );
1647 devPathLength = strlen( bsdPath );
1648 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1649 kernResult = KERN_SUCCESS;
1650 }
1651 CFRelease( bsdPathAsCFString );
1652 }
1653 IOObjectRelease( nextMedia );
1654 }
1655
1656 return kernResult;
1657 }
1658
1659 #endif
1660
1661 static int hdev_probe_device(const char *filename)
1662 {
1663 struct stat st;
1664
1665 /* allow a dedicated CD-ROM driver to match with a higher priority */
1666 if (strstart(filename, "/dev/cdrom", NULL))
1667 return 50;
1668
1669 if (stat(filename, &st) >= 0 &&
1670 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
1671 return 100;
1672 }
1673
1674 return 0;
1675 }
1676
1677 static int check_hdev_writable(BDRVRawState *s)
1678 {
1679 #if defined(BLKROGET)
1680 /* Linux block devices can be configured "read-only" using blockdev(8).
1681 * This is independent of device node permissions and therefore open(2)
1682 * with O_RDWR succeeds. Actual writes fail with EPERM.
1683 *
1684 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
1685 * check for read-only block devices so that Linux block devices behave
1686 * properly.
1687 */
1688 struct stat st;
1689 int readonly = 0;
1690
1691 if (fstat(s->fd, &st)) {
1692 return -errno;
1693 }
1694
1695 if (!S_ISBLK(st.st_mode)) {
1696 return 0;
1697 }
1698
1699 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
1700 return -errno;
1701 }
1702
1703 if (readonly) {
1704 return -EACCES;
1705 }
1706 #endif /* defined(BLKROGET) */
1707 return 0;
1708 }
1709
1710 static void hdev_parse_filename(const char *filename, QDict *options,
1711 Error **errp)
1712 {
1713 /* The prefix is optional, just as for "file". */
1714 strstart(filename, "host_device:", &filename);
1715
1716 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
1717 }
1718
1719 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
1720 Error **errp)
1721 {
1722 BDRVRawState *s = bs->opaque;
1723 Error *local_err = NULL;
1724 int ret;
1725 const char *filename = qdict_get_str(options, "filename");
1726
1727 #if defined(__APPLE__) && defined(__MACH__)
1728 if (strstart(filename, "/dev/cdrom", NULL)) {
1729 kern_return_t kernResult;
1730 io_iterator_t mediaIterator;
1731 char bsdPath[ MAXPATHLEN ];
1732 int fd;
1733
1734 kernResult = FindEjectableCDMedia( &mediaIterator );
1735 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1736
1737 if ( bsdPath[ 0 ] != '\0' ) {
1738 strcat(bsdPath,"s0");
1739 /* some CDs don't have a partition 0 */
1740 fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1741 if (fd < 0) {
1742 bsdPath[strlen(bsdPath)-1] = '1';
1743 } else {
1744 qemu_close(fd);
1745 }
1746 filename = bsdPath;
1747 qdict_put(options, "filename", qstring_from_str(filename));
1748 }
1749
1750 if ( mediaIterator )
1751 IOObjectRelease( mediaIterator );
1752 }
1753 #endif
1754
1755 s->type = FTYPE_FILE;
1756 #if defined(__linux__)
1757 {
1758 char resolved_path[ MAXPATHLEN ], *temp;
1759
1760 temp = realpath(filename, resolved_path);
1761 if (temp && strstart(temp, "/dev/sg", NULL)) {
1762 bs->sg = 1;
1763 }
1764 }
1765 #endif
1766
1767 ret = raw_open_common(bs, options, flags, 0, &local_err);
1768 if (ret < 0) {
1769 if (local_err) {
1770 error_propagate(errp, local_err);
1771 }
1772 return ret;
1773 }
1774
1775 if (flags & BDRV_O_RDWR) {
1776 ret = check_hdev_writable(s);
1777 if (ret < 0) {
1778 raw_close(bs);
1779 error_setg_errno(errp, -ret, "The device is not writable");
1780 return ret;
1781 }
1782 }
1783
1784 return ret;
1785 }
1786
1787 #if defined(__linux__)
1788 /* Note: we do not have a reliable method to detect if the floppy is
1789 present. The current method is to try to open the floppy at every
1790 I/O and to keep it opened during a few hundreds of ms. */
1791 static int fd_open(BlockDriverState *bs)
1792 {
1793 BDRVRawState *s = bs->opaque;
1794 int last_media_present;
1795
1796 if (s->type != FTYPE_FD)
1797 return 0;
1798 last_media_present = (s->fd >= 0);
1799 if (s->fd >= 0 &&
1800 (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1801 qemu_close(s->fd);
1802 s->fd = -1;
1803 #ifdef DEBUG_FLOPPY
1804 printf("Floppy closed\n");
1805 #endif
1806 }
1807 if (s->fd < 0) {
1808 if (s->fd_got_error &&
1809 (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1810 #ifdef DEBUG_FLOPPY
1811 printf("No floppy (open delayed)\n");
1812 #endif
1813 return -EIO;
1814 }
1815 s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
1816 if (s->fd < 0) {
1817 s->fd_error_time = get_clock();
1818 s->fd_got_error = 1;
1819 if (last_media_present)
1820 s->fd_media_changed = 1;
1821 #ifdef DEBUG_FLOPPY
1822 printf("No floppy\n");
1823 #endif
1824 return -EIO;
1825 }
1826 #ifdef DEBUG_FLOPPY
1827 printf("Floppy opened\n");
1828 #endif
1829 }
1830 if (!last_media_present)
1831 s->fd_media_changed = 1;
1832 s->fd_open_time = get_clock();
1833 s->fd_got_error = 0;
1834 return 0;
1835 }
1836
1837 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1838 {
1839 BDRVRawState *s = bs->opaque;
1840
1841 return ioctl(s->fd, req, buf);
1842 }
1843
1844 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1845 unsigned long int req, void *buf,
1846 BlockDriverCompletionFunc *cb, void *opaque)
1847 {
1848 BDRVRawState *s = bs->opaque;
1849 RawPosixAIOData *acb;
1850 ThreadPool *pool;
1851
1852 if (fd_open(bs) < 0)
1853 return NULL;
1854
1855 acb = g_slice_new(RawPosixAIOData);
1856 acb->bs = bs;
1857 acb->aio_type = QEMU_AIO_IOCTL;
1858 acb->aio_fildes = s->fd;
1859 acb->aio_offset = 0;
1860 acb->aio_ioctl_buf = buf;
1861 acb->aio_ioctl_cmd = req;
1862 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1863 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1864 }
1865
1866 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1867 static int fd_open(BlockDriverState *bs)
1868 {
1869 BDRVRawState *s = bs->opaque;
1870
1871 /* this is just to ensure s->fd is sane (its called by io ops) */
1872 if (s->fd >= 0)
1873 return 0;
1874 return -EIO;
1875 }
1876 #else /* !linux && !FreeBSD */
1877
1878 static int fd_open(BlockDriverState *bs)
1879 {
1880 return 0;
1881 }
1882
1883 #endif /* !linux && !FreeBSD */
1884
1885 static coroutine_fn BlockDriverAIOCB *hdev_aio_discard(BlockDriverState *bs,
1886 int64_t sector_num, int nb_sectors,
1887 BlockDriverCompletionFunc *cb, void *opaque)
1888 {
1889 BDRVRawState *s = bs->opaque;
1890
1891 if (fd_open(bs) < 0) {
1892 return NULL;
1893 }
1894 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1895 cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1896 }
1897
1898 static coroutine_fn int hdev_co_write_zeroes(BlockDriverState *bs,
1899 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
1900 {
1901 BDRVRawState *s = bs->opaque;
1902 int rc;
1903
1904 rc = fd_open(bs);
1905 if (rc < 0) {
1906 return rc;
1907 }
1908 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1909 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1910 QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
1911 } else if (s->discard_zeroes) {
1912 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1913 QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1914 }
1915 return -ENOTSUP;
1916 }
1917
1918 static int hdev_create(const char *filename, QemuOpts *opts,
1919 Error **errp)
1920 {
1921 int fd;
1922 int ret = 0;
1923 struct stat stat_buf;
1924 int64_t total_size = 0;
1925 bool has_prefix;
1926
1927 /* This function is used by all three protocol block drivers and therefore
1928 * any of these three prefixes may be given.
1929 * The return value has to be stored somewhere, otherwise this is an error
1930 * due to -Werror=unused-value. */
1931 has_prefix =
1932 strstart(filename, "host_device:", &filename) ||
1933 strstart(filename, "host_cdrom:" , &filename) ||
1934 strstart(filename, "host_floppy:", &filename);
1935
1936 (void)has_prefix;
1937
1938 /* Read out options */
1939 total_size =
1940 qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0) / BDRV_SECTOR_SIZE;
1941
1942 fd = qemu_open(filename, O_WRONLY | O_BINARY);
1943 if (fd < 0) {
1944 ret = -errno;
1945 error_setg_errno(errp, -ret, "Could not open device");
1946 return ret;
1947 }
1948
1949 if (fstat(fd, &stat_buf) < 0) {
1950 ret = -errno;
1951 error_setg_errno(errp, -ret, "Could not stat device");
1952 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
1953 error_setg(errp,
1954 "The given file is neither a block nor a character device");
1955 ret = -ENODEV;
1956 } else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE) {
1957 error_setg(errp, "Device is too small");
1958 ret = -ENOSPC;
1959 }
1960
1961 qemu_close(fd);
1962 return ret;
1963 }
1964
1965 static BlockDriver bdrv_host_device = {
1966 .format_name = "host_device",
1967 .protocol_name = "host_device",
1968 .instance_size = sizeof(BDRVRawState),
1969 .bdrv_needs_filename = true,
1970 .bdrv_probe_device = hdev_probe_device,
1971 .bdrv_parse_filename = hdev_parse_filename,
1972 .bdrv_file_open = hdev_open,
1973 .bdrv_close = raw_close,
1974 .bdrv_reopen_prepare = raw_reopen_prepare,
1975 .bdrv_reopen_commit = raw_reopen_commit,
1976 .bdrv_reopen_abort = raw_reopen_abort,
1977 .bdrv_create = hdev_create,
1978 .create_opts = &raw_create_opts,
1979 .bdrv_co_write_zeroes = hdev_co_write_zeroes,
1980
1981 .bdrv_aio_readv = raw_aio_readv,
1982 .bdrv_aio_writev = raw_aio_writev,
1983 .bdrv_aio_flush = raw_aio_flush,
1984 .bdrv_aio_discard = hdev_aio_discard,
1985 .bdrv_refresh_limits = raw_refresh_limits,
1986 .bdrv_io_plug = raw_aio_plug,
1987 .bdrv_io_unplug = raw_aio_unplug,
1988 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
1989
1990 .bdrv_truncate = raw_truncate,
1991 .bdrv_getlength = raw_getlength,
1992 .bdrv_get_info = raw_get_info,
1993 .bdrv_get_allocated_file_size
1994 = raw_get_allocated_file_size,
1995
1996 .bdrv_detach_aio_context = raw_detach_aio_context,
1997 .bdrv_attach_aio_context = raw_attach_aio_context,
1998
1999 /* generic scsi device */
2000 #ifdef __linux__
2001 .bdrv_ioctl = hdev_ioctl,
2002 .bdrv_aio_ioctl = hdev_aio_ioctl,
2003 #endif
2004 };
2005
2006 #ifdef __linux__
2007 static void floppy_parse_filename(const char *filename, QDict *options,
2008 Error **errp)
2009 {
2010 /* The prefix is optional, just as for "file". */
2011 strstart(filename, "host_floppy:", &filename);
2012
2013 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2014 }
2015
2016 static int floppy_open(BlockDriverState *bs, QDict *options, int flags,
2017 Error **errp)
2018 {
2019 BDRVRawState *s = bs->opaque;
2020 Error *local_err = NULL;
2021 int ret;
2022
2023 s->type = FTYPE_FD;
2024
2025 /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
2026 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2027 if (ret) {
2028 if (local_err) {
2029 error_propagate(errp, local_err);
2030 }
2031 return ret;
2032 }
2033
2034 /* close fd so that we can reopen it as needed */
2035 qemu_close(s->fd);
2036 s->fd = -1;
2037 s->fd_media_changed = 1;
2038
2039 return 0;
2040 }
2041
2042 static int floppy_probe_device(const char *filename)
2043 {
2044 int fd, ret;
2045 int prio = 0;
2046 struct floppy_struct fdparam;
2047 struct stat st;
2048
2049 if (strstart(filename, "/dev/fd", NULL) &&
2050 !strstart(filename, "/dev/fdset/", NULL)) {
2051 prio = 50;
2052 }
2053
2054 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2055 if (fd < 0) {
2056 goto out;
2057 }
2058 ret = fstat(fd, &st);
2059 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2060 goto outc;
2061 }
2062
2063 /* Attempt to detect via a floppy specific ioctl */
2064 ret = ioctl(fd, FDGETPRM, &fdparam);
2065 if (ret >= 0)
2066 prio = 100;
2067
2068 outc:
2069 qemu_close(fd);
2070 out:
2071 return prio;
2072 }
2073
2074
2075 static int floppy_is_inserted(BlockDriverState *bs)
2076 {
2077 return fd_open(bs) >= 0;
2078 }
2079
2080 static int floppy_media_changed(BlockDriverState *bs)
2081 {
2082 BDRVRawState *s = bs->opaque;
2083 int ret;
2084
2085 /*
2086 * XXX: we do not have a true media changed indication.
2087 * It does not work if the floppy is changed without trying to read it.
2088 */
2089 fd_open(bs);
2090 ret = s->fd_media_changed;
2091 s->fd_media_changed = 0;
2092 #ifdef DEBUG_FLOPPY
2093 printf("Floppy changed=%d\n", ret);
2094 #endif
2095 return ret;
2096 }
2097
2098 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
2099 {
2100 BDRVRawState *s = bs->opaque;
2101 int fd;
2102
2103 if (s->fd >= 0) {
2104 qemu_close(s->fd);
2105 s->fd = -1;
2106 }
2107 fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
2108 if (fd >= 0) {
2109 if (ioctl(fd, FDEJECT, 0) < 0)
2110 perror("FDEJECT");
2111 qemu_close(fd);
2112 }
2113 }
2114
2115 static BlockDriver bdrv_host_floppy = {
2116 .format_name = "host_floppy",
2117 .protocol_name = "host_floppy",
2118 .instance_size = sizeof(BDRVRawState),
2119 .bdrv_needs_filename = true,
2120 .bdrv_probe_device = floppy_probe_device,
2121 .bdrv_parse_filename = floppy_parse_filename,
2122 .bdrv_file_open = floppy_open,
2123 .bdrv_close = raw_close,
2124 .bdrv_reopen_prepare = raw_reopen_prepare,
2125 .bdrv_reopen_commit = raw_reopen_commit,
2126 .bdrv_reopen_abort = raw_reopen_abort,
2127 .bdrv_create = hdev_create,
2128 .create_opts = &raw_create_opts,
2129
2130 .bdrv_aio_readv = raw_aio_readv,
2131 .bdrv_aio_writev = raw_aio_writev,
2132 .bdrv_aio_flush = raw_aio_flush,
2133 .bdrv_refresh_limits = raw_refresh_limits,
2134 .bdrv_io_plug = raw_aio_plug,
2135 .bdrv_io_unplug = raw_aio_unplug,
2136 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2137
2138 .bdrv_truncate = raw_truncate,
2139 .bdrv_getlength = raw_getlength,
2140 .has_variable_length = true,
2141 .bdrv_get_allocated_file_size
2142 = raw_get_allocated_file_size,
2143
2144 .bdrv_detach_aio_context = raw_detach_aio_context,
2145 .bdrv_attach_aio_context = raw_attach_aio_context,
2146
2147 /* removable device support */
2148 .bdrv_is_inserted = floppy_is_inserted,
2149 .bdrv_media_changed = floppy_media_changed,
2150 .bdrv_eject = floppy_eject,
2151 };
2152 #endif
2153
2154 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2155 static void cdrom_parse_filename(const char *filename, QDict *options,
2156 Error **errp)
2157 {
2158 /* The prefix is optional, just as for "file". */
2159 strstart(filename, "host_cdrom:", &filename);
2160
2161 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2162 }
2163 #endif
2164
2165 #ifdef __linux__
2166 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2167 Error **errp)
2168 {
2169 BDRVRawState *s = bs->opaque;
2170 Error *local_err = NULL;
2171 int ret;
2172
2173 s->type = FTYPE_CD;
2174
2175 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2176 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2177 if (local_err) {
2178 error_propagate(errp, local_err);
2179 }
2180 return ret;
2181 }
2182
2183 static int cdrom_probe_device(const char *filename)
2184 {
2185 int fd, ret;
2186 int prio = 0;
2187 struct stat st;
2188
2189 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2190 if (fd < 0) {
2191 goto out;
2192 }
2193 ret = fstat(fd, &st);
2194 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2195 goto outc;
2196 }
2197
2198 /* Attempt to detect via a CDROM specific ioctl */
2199 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2200 if (ret >= 0)
2201 prio = 100;
2202
2203 outc:
2204 qemu_close(fd);
2205 out:
2206 return prio;
2207 }
2208
2209 static int cdrom_is_inserted(BlockDriverState *bs)
2210 {
2211 BDRVRawState *s = bs->opaque;
2212 int ret;
2213
2214 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2215 if (ret == CDS_DISC_OK)
2216 return 1;
2217 return 0;
2218 }
2219
2220 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2221 {
2222 BDRVRawState *s = bs->opaque;
2223
2224 if (eject_flag) {
2225 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2226 perror("CDROMEJECT");
2227 } else {
2228 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2229 perror("CDROMEJECT");
2230 }
2231 }
2232
2233 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2234 {
2235 BDRVRawState *s = bs->opaque;
2236
2237 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2238 /*
2239 * Note: an error can happen if the distribution automatically
2240 * mounts the CD-ROM
2241 */
2242 /* perror("CDROM_LOCKDOOR"); */
2243 }
2244 }
2245
2246 static BlockDriver bdrv_host_cdrom = {
2247 .format_name = "host_cdrom",
2248 .protocol_name = "host_cdrom",
2249 .instance_size = sizeof(BDRVRawState),
2250 .bdrv_needs_filename = true,
2251 .bdrv_probe_device = cdrom_probe_device,
2252 .bdrv_parse_filename = cdrom_parse_filename,
2253 .bdrv_file_open = cdrom_open,
2254 .bdrv_close = raw_close,
2255 .bdrv_reopen_prepare = raw_reopen_prepare,
2256 .bdrv_reopen_commit = raw_reopen_commit,
2257 .bdrv_reopen_abort = raw_reopen_abort,
2258 .bdrv_create = hdev_create,
2259 .create_opts = &raw_create_opts,
2260
2261 .bdrv_aio_readv = raw_aio_readv,
2262 .bdrv_aio_writev = raw_aio_writev,
2263 .bdrv_aio_flush = raw_aio_flush,
2264 .bdrv_refresh_limits = raw_refresh_limits,
2265 .bdrv_io_plug = raw_aio_plug,
2266 .bdrv_io_unplug = raw_aio_unplug,
2267 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2268
2269 .bdrv_truncate = raw_truncate,
2270 .bdrv_getlength = raw_getlength,
2271 .has_variable_length = true,
2272 .bdrv_get_allocated_file_size
2273 = raw_get_allocated_file_size,
2274
2275 .bdrv_detach_aio_context = raw_detach_aio_context,
2276 .bdrv_attach_aio_context = raw_attach_aio_context,
2277
2278 /* removable device support */
2279 .bdrv_is_inserted = cdrom_is_inserted,
2280 .bdrv_eject = cdrom_eject,
2281 .bdrv_lock_medium = cdrom_lock_medium,
2282
2283 /* generic scsi device */
2284 .bdrv_ioctl = hdev_ioctl,
2285 .bdrv_aio_ioctl = hdev_aio_ioctl,
2286 };
2287 #endif /* __linux__ */
2288
2289 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2290 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2291 Error **errp)
2292 {
2293 BDRVRawState *s = bs->opaque;
2294 Error *local_err = NULL;
2295 int ret;
2296
2297 s->type = FTYPE_CD;
2298
2299 ret = raw_open_common(bs, options, flags, 0, &local_err);
2300 if (ret) {
2301 if (local_err) {
2302 error_propagate(errp, local_err);
2303 }
2304 return ret;
2305 }
2306
2307 /* make sure the door isn't locked at this time */
2308 ioctl(s->fd, CDIOCALLOW);
2309 return 0;
2310 }
2311
2312 static int cdrom_probe_device(const char *filename)
2313 {
2314 if (strstart(filename, "/dev/cd", NULL) ||
2315 strstart(filename, "/dev/acd", NULL))
2316 return 100;
2317 return 0;
2318 }
2319
2320 static int cdrom_reopen(BlockDriverState *bs)
2321 {
2322 BDRVRawState *s = bs->opaque;
2323 int fd;
2324
2325 /*
2326 * Force reread of possibly changed/newly loaded disc,
2327 * FreeBSD seems to not notice sometimes...
2328 */
2329 if (s->fd >= 0)
2330 qemu_close(s->fd);
2331 fd = qemu_open(bs->filename, s->open_flags, 0644);
2332 if (fd < 0) {
2333 s->fd = -1;
2334 return -EIO;
2335 }
2336 s->fd = fd;
2337
2338 /* make sure the door isn't locked at this time */
2339 ioctl(s->fd, CDIOCALLOW);
2340 return 0;
2341 }
2342
2343 static int cdrom_is_inserted(BlockDriverState *bs)
2344 {
2345 return raw_getlength(bs) > 0;
2346 }
2347
2348 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2349 {
2350 BDRVRawState *s = bs->opaque;
2351
2352 if (s->fd < 0)
2353 return;
2354
2355 (void) ioctl(s->fd, CDIOCALLOW);
2356
2357 if (eject_flag) {
2358 if (ioctl(s->fd, CDIOCEJECT) < 0)
2359 perror("CDIOCEJECT");
2360 } else {
2361 if (ioctl(s->fd, CDIOCCLOSE) < 0)
2362 perror("CDIOCCLOSE");
2363 }
2364
2365 cdrom_reopen(bs);
2366 }
2367
2368 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2369 {
2370 BDRVRawState *s = bs->opaque;
2371
2372 if (s->fd < 0)
2373 return;
2374 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2375 /*
2376 * Note: an error can happen if the distribution automatically
2377 * mounts the CD-ROM
2378 */
2379 /* perror("CDROM_LOCKDOOR"); */
2380 }
2381 }
2382
2383 static BlockDriver bdrv_host_cdrom = {
2384 .format_name = "host_cdrom",
2385 .protocol_name = "host_cdrom",
2386 .instance_size = sizeof(BDRVRawState),
2387 .bdrv_needs_filename = true,
2388 .bdrv_probe_device = cdrom_probe_device,
2389 .bdrv_parse_filename = cdrom_parse_filename,
2390 .bdrv_file_open = cdrom_open,
2391 .bdrv_close = raw_close,
2392 .bdrv_reopen_prepare = raw_reopen_prepare,
2393 .bdrv_reopen_commit = raw_reopen_commit,
2394 .bdrv_reopen_abort = raw_reopen_abort,
2395 .bdrv_create = hdev_create,
2396 .create_opts = &raw_create_opts,
2397
2398 .bdrv_aio_readv = raw_aio_readv,
2399 .bdrv_aio_writev = raw_aio_writev,
2400 .bdrv_aio_flush = raw_aio_flush,
2401 .bdrv_refresh_limits = raw_refresh_limits,
2402 .bdrv_io_plug = raw_aio_plug,
2403 .bdrv_io_unplug = raw_aio_unplug,
2404 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2405
2406 .bdrv_truncate = raw_truncate,
2407 .bdrv_getlength = raw_getlength,
2408 .has_variable_length = true,
2409 .bdrv_get_allocated_file_size
2410 = raw_get_allocated_file_size,
2411
2412 .bdrv_detach_aio_context = raw_detach_aio_context,
2413 .bdrv_attach_aio_context = raw_attach_aio_context,
2414
2415 /* removable device support */
2416 .bdrv_is_inserted = cdrom_is_inserted,
2417 .bdrv_eject = cdrom_eject,
2418 .bdrv_lock_medium = cdrom_lock_medium,
2419 };
2420 #endif /* __FreeBSD__ */
2421
2422 static void bdrv_file_init(void)
2423 {
2424 /*
2425 * Register all the drivers. Note that order is important, the driver
2426 * registered last will get probed first.
2427 */
2428 bdrv_register(&bdrv_file);
2429 bdrv_register(&bdrv_host_device);
2430 #ifdef __linux__
2431 bdrv_register(&bdrv_host_floppy);
2432 bdrv_register(&bdrv_host_cdrom);
2433 #endif
2434 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2435 bdrv_register(&bdrv_host_cdrom);
2436 #endif
2437 }
2438
2439 block_init(bdrv_file_init);
Random patches