]> git.proxmox.com Git - qemu.git/blob - block/raw-posix.c
Merge remote-tracking branch 'kwolf/for-anthony' into staging
[qemu.git] / block / raw-posix.c
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-char.h"
27 #include "qemu-log.h"
28 #include "block_int.h"
29 #include "module.h"
30 #include "block/raw-posix-aio.h"
31
32 #if defined(__APPLE__) && (__MACH__)
33 #include <paths.h>
34 #include <sys/param.h>
35 #include <IOKit/IOKitLib.h>
36 #include <IOKit/IOBSD.h>
37 #include <IOKit/storage/IOMediaBSDClient.h>
38 #include <IOKit/storage/IOMedia.h>
39 #include <IOKit/storage/IOCDMedia.h>
40 //#include <IOKit/storage/IOCDTypes.h>
41 #include <CoreFoundation/CoreFoundation.h>
42 #endif
43
44 #ifdef __sun__
45 #define _POSIX_PTHREAD_SEMANTICS 1
46 #include <sys/dkio.h>
47 #endif
48 #ifdef __linux__
49 #include <sys/types.h>
50 #include <sys/stat.h>
51 #include <sys/ioctl.h>
52 #include <sys/param.h>
53 #include <linux/cdrom.h>
54 #include <linux/fd.h>
55 #include <linux/fs.h>
56 #endif
57 #ifdef CONFIG_FIEMAP
58 #include <linux/fiemap.h>
59 #endif
60 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
61 #include <sys/disk.h>
62 #include <sys/cdio.h>
63 #endif
64
65 #ifdef __OpenBSD__
66 #include <sys/ioctl.h>
67 #include <sys/disklabel.h>
68 #include <sys/dkio.h>
69 #endif
70
71 #ifdef __NetBSD__
72 #include <sys/ioctl.h>
73 #include <sys/disklabel.h>
74 #include <sys/dkio.h>
75 #include <sys/disk.h>
76 #endif
77
78 #ifdef __DragonFly__
79 #include <sys/ioctl.h>
80 #include <sys/diskslice.h>
81 #endif
82
83 #ifdef CONFIG_XFS
84 #include <xfs/xfs.h>
85 #endif
86
87 //#define DEBUG_FLOPPY
88
89 //#define DEBUG_BLOCK
90 #if defined(DEBUG_BLOCK)
91 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
92 { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
93 #else
94 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
95 #endif
96
97 /* OS X does not have O_DSYNC */
98 #ifndef O_DSYNC
99 #ifdef O_SYNC
100 #define O_DSYNC O_SYNC
101 #elif defined(O_FSYNC)
102 #define O_DSYNC O_FSYNC
103 #endif
104 #endif
105
106 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
107 #ifndef O_DIRECT
108 #define O_DIRECT O_DSYNC
109 #endif
110
111 #define FTYPE_FILE 0
112 #define FTYPE_CD 1
113 #define FTYPE_FD 2
114
115 /* if the FD is not accessed during that time (in ns), we try to
116 reopen it to see if the disk has been changed */
117 #define FD_OPEN_TIMEOUT (1000000000)
118
119 #define MAX_BLOCKSIZE 4096
120
121 typedef struct BDRVRawState {
122 int fd;
123 int type;
124 int open_flags;
125 #if defined(__linux__)
126 /* linux floppy specific */
127 int64_t fd_open_time;
128 int64_t fd_error_time;
129 int fd_got_error;
130 int fd_media_changed;
131 #endif
132 #ifdef CONFIG_LINUX_AIO
133 int use_aio;
134 void *aio_ctx;
135 #endif
136 uint8_t *aligned_buf;
137 unsigned aligned_buf_size;
138 #ifdef CONFIG_XFS
139 bool is_xfs : 1;
140 #endif
141 } BDRVRawState;
142
143 static int fd_open(BlockDriverState *bs);
144 static int64_t raw_getlength(BlockDriverState *bs);
145
146 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
147 static int cdrom_reopen(BlockDriverState *bs);
148 #endif
149
150 #if defined(__NetBSD__)
151 static int raw_normalize_devicepath(const char **filename)
152 {
153 static char namebuf[PATH_MAX];
154 const char *dp, *fname;
155 struct stat sb;
156
157 fname = *filename;
158 dp = strrchr(fname, '/');
159 if (lstat(fname, &sb) < 0) {
160 fprintf(stderr, "%s: stat failed: %s\n",
161 fname, strerror(errno));
162 return -errno;
163 }
164
165 if (!S_ISBLK(sb.st_mode)) {
166 return 0;
167 }
168
169 if (dp == NULL) {
170 snprintf(namebuf, PATH_MAX, "r%s", fname);
171 } else {
172 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
173 (int)(dp - fname), fname, dp + 1);
174 }
175 fprintf(stderr, "%s is a block device", fname);
176 *filename = namebuf;
177 fprintf(stderr, ", using %s\n", *filename);
178
179 return 0;
180 }
181 #else
182 static int raw_normalize_devicepath(const char **filename)
183 {
184 return 0;
185 }
186 #endif
187
188 static int raw_open_common(BlockDriverState *bs, const char *filename,
189 int bdrv_flags, int open_flags)
190 {
191 BDRVRawState *s = bs->opaque;
192 int fd, ret;
193
194 ret = raw_normalize_devicepath(&filename);
195 if (ret != 0) {
196 return ret;
197 }
198
199 s->open_flags = open_flags | O_BINARY;
200 s->open_flags &= ~O_ACCMODE;
201 if (bdrv_flags & BDRV_O_RDWR) {
202 s->open_flags |= O_RDWR;
203 } else {
204 s->open_flags |= O_RDONLY;
205 }
206
207 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
208 * and O_DIRECT for no caching. */
209 if ((bdrv_flags & BDRV_O_NOCACHE))
210 s->open_flags |= O_DIRECT;
211 if (!(bdrv_flags & BDRV_O_CACHE_WB))
212 s->open_flags |= O_DSYNC;
213
214 s->fd = -1;
215 fd = qemu_open(filename, s->open_flags, 0644);
216 if (fd < 0) {
217 ret = -errno;
218 if (ret == -EROFS)
219 ret = -EACCES;
220 return ret;
221 }
222 s->fd = fd;
223 s->aligned_buf = NULL;
224
225 if ((bdrv_flags & BDRV_O_NOCACHE)) {
226 /*
227 * Allocate a buffer for read/modify/write cycles. Chose the size
228 * pessimistically as we don't know the block size yet.
229 */
230 s->aligned_buf_size = 32 * MAX_BLOCKSIZE;
231 s->aligned_buf = qemu_memalign(MAX_BLOCKSIZE, s->aligned_buf_size);
232 if (s->aligned_buf == NULL) {
233 goto out_close;
234 }
235 }
236
237 /* We're falling back to POSIX AIO in some cases so init always */
238 if (paio_init() < 0) {
239 goto out_free_buf;
240 }
241
242 #ifdef CONFIG_LINUX_AIO
243 /*
244 * Currently Linux do AIO only for files opened with O_DIRECT
245 * specified so check NOCACHE flag too
246 */
247 if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
248 (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
249
250 s->aio_ctx = laio_init();
251 if (!s->aio_ctx) {
252 goto out_free_buf;
253 }
254 s->use_aio = 1;
255 } else
256 #endif
257 {
258 #ifdef CONFIG_LINUX_AIO
259 s->use_aio = 0;
260 #endif
261 }
262
263 #ifdef CONFIG_XFS
264 if (platform_test_xfs_fd(s->fd)) {
265 s->is_xfs = 1;
266 }
267 #endif
268
269 return 0;
270
271 out_free_buf:
272 qemu_vfree(s->aligned_buf);
273 out_close:
274 qemu_close(fd);
275 return -errno;
276 }
277
278 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
279 {
280 BDRVRawState *s = bs->opaque;
281
282 s->type = FTYPE_FILE;
283 return raw_open_common(bs, filename, flags, 0);
284 }
285
286 /* XXX: use host sector size if necessary with:
287 #ifdef DIOCGSECTORSIZE
288 {
289 unsigned int sectorsize = 512;
290 if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
291 sectorsize > bufsize)
292 bufsize = sectorsize;
293 }
294 #endif
295 #ifdef CONFIG_COCOA
296 uint32_t blockSize = 512;
297 if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
298 bufsize = blockSize;
299 }
300 #endif
301 */
302
303 /*
304 * Check if all memory in this vector is sector aligned.
305 */
306 static int qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
307 {
308 int i;
309
310 for (i = 0; i < qiov->niov; i++) {
311 if ((uintptr_t) qiov->iov[i].iov_base % bs->buffer_alignment) {
312 return 0;
313 }
314 }
315
316 return 1;
317 }
318
319 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
320 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
321 BlockDriverCompletionFunc *cb, void *opaque, int type)
322 {
323 BDRVRawState *s = bs->opaque;
324
325 if (fd_open(bs) < 0)
326 return NULL;
327
328 /*
329 * If O_DIRECT is used the buffer needs to be aligned on a sector
330 * boundary. Check if this is the case or tell the low-level
331 * driver that it needs to copy the buffer.
332 */
333 if (s->aligned_buf) {
334 if (!qiov_is_aligned(bs, qiov)) {
335 type |= QEMU_AIO_MISALIGNED;
336 #ifdef CONFIG_LINUX_AIO
337 } else if (s->use_aio) {
338 return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
339 nb_sectors, cb, opaque, type);
340 #endif
341 }
342 }
343
344 return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
345 cb, opaque, type);
346 }
347
348 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
349 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
350 BlockDriverCompletionFunc *cb, void *opaque)
351 {
352 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
353 cb, opaque, QEMU_AIO_READ);
354 }
355
356 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
357 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
358 BlockDriverCompletionFunc *cb, void *opaque)
359 {
360 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
361 cb, opaque, QEMU_AIO_WRITE);
362 }
363
364 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
365 BlockDriverCompletionFunc *cb, void *opaque)
366 {
367 BDRVRawState *s = bs->opaque;
368
369 if (fd_open(bs) < 0)
370 return NULL;
371
372 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
373 }
374
375 static void raw_close(BlockDriverState *bs)
376 {
377 BDRVRawState *s = bs->opaque;
378 if (s->fd >= 0) {
379 qemu_close(s->fd);
380 s->fd = -1;
381 if (s->aligned_buf != NULL)
382 qemu_vfree(s->aligned_buf);
383 }
384 }
385
386 static int raw_truncate(BlockDriverState *bs, int64_t offset)
387 {
388 BDRVRawState *s = bs->opaque;
389 struct stat st;
390
391 if (fstat(s->fd, &st)) {
392 return -errno;
393 }
394
395 if (S_ISREG(st.st_mode)) {
396 if (ftruncate(s->fd, offset) < 0) {
397 return -errno;
398 }
399 } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
400 if (offset > raw_getlength(bs)) {
401 return -EINVAL;
402 }
403 } else {
404 return -ENOTSUP;
405 }
406
407 return 0;
408 }
409
410 #ifdef __OpenBSD__
411 static int64_t raw_getlength(BlockDriverState *bs)
412 {
413 BDRVRawState *s = bs->opaque;
414 int fd = s->fd;
415 struct stat st;
416
417 if (fstat(fd, &st))
418 return -1;
419 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
420 struct disklabel dl;
421
422 if (ioctl(fd, DIOCGDINFO, &dl))
423 return -1;
424 return (uint64_t)dl.d_secsize *
425 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
426 } else
427 return st.st_size;
428 }
429 #elif defined(__NetBSD__)
430 static int64_t raw_getlength(BlockDriverState *bs)
431 {
432 BDRVRawState *s = bs->opaque;
433 int fd = s->fd;
434 struct stat st;
435
436 if (fstat(fd, &st))
437 return -1;
438 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
439 struct dkwedge_info dkw;
440
441 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
442 return dkw.dkw_size * 512;
443 } else {
444 struct disklabel dl;
445
446 if (ioctl(fd, DIOCGDINFO, &dl))
447 return -1;
448 return (uint64_t)dl.d_secsize *
449 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
450 }
451 } else
452 return st.st_size;
453 }
454 #elif defined(__sun__)
455 static int64_t raw_getlength(BlockDriverState *bs)
456 {
457 BDRVRawState *s = bs->opaque;
458 struct dk_minfo minfo;
459 int ret;
460
461 ret = fd_open(bs);
462 if (ret < 0) {
463 return ret;
464 }
465
466 /*
467 * Use the DKIOCGMEDIAINFO ioctl to read the size.
468 */
469 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
470 if (ret != -1) {
471 return minfo.dki_lbsize * minfo.dki_capacity;
472 }
473
474 /*
475 * There are reports that lseek on some devices fails, but
476 * irc discussion said that contingency on contingency was overkill.
477 */
478 return lseek(s->fd, 0, SEEK_END);
479 }
480 #elif defined(CONFIG_BSD)
481 static int64_t raw_getlength(BlockDriverState *bs)
482 {
483 BDRVRawState *s = bs->opaque;
484 int fd = s->fd;
485 int64_t size;
486 struct stat sb;
487 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
488 int reopened = 0;
489 #endif
490 int ret;
491
492 ret = fd_open(bs);
493 if (ret < 0)
494 return ret;
495
496 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
497 again:
498 #endif
499 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
500 #ifdef DIOCGMEDIASIZE
501 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
502 #elif defined(DIOCGPART)
503 {
504 struct partinfo pi;
505 if (ioctl(fd, DIOCGPART, &pi) == 0)
506 size = pi.media_size;
507 else
508 size = 0;
509 }
510 if (size == 0)
511 #endif
512 #if defined(__APPLE__) && defined(__MACH__)
513 size = LONG_LONG_MAX;
514 #else
515 size = lseek(fd, 0LL, SEEK_END);
516 #endif
517 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
518 switch(s->type) {
519 case FTYPE_CD:
520 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
521 if (size == 2048LL * (unsigned)-1)
522 size = 0;
523 /* XXX no disc? maybe we need to reopen... */
524 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
525 reopened = 1;
526 goto again;
527 }
528 }
529 #endif
530 } else {
531 size = lseek(fd, 0, SEEK_END);
532 }
533 return size;
534 }
535 #else
536 static int64_t raw_getlength(BlockDriverState *bs)
537 {
538 BDRVRawState *s = bs->opaque;
539 int ret;
540
541 ret = fd_open(bs);
542 if (ret < 0) {
543 return ret;
544 }
545
546 return lseek(s->fd, 0, SEEK_END);
547 }
548 #endif
549
550 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
551 {
552 struct stat st;
553 BDRVRawState *s = bs->opaque;
554
555 if (fstat(s->fd, &st) < 0) {
556 return -errno;
557 }
558 return (int64_t)st.st_blocks * 512;
559 }
560
561 static int raw_create(const char *filename, QEMUOptionParameter *options)
562 {
563 int fd;
564 int result = 0;
565 int64_t total_size = 0;
566
567 /* Read out options */
568 while (options && options->name) {
569 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
570 total_size = options->value.n / BDRV_SECTOR_SIZE;
571 }
572 options++;
573 }
574
575 fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
576 0644);
577 if (fd < 0) {
578 result = -errno;
579 } else {
580 if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
581 result = -errno;
582 }
583 if (qemu_close(fd) != 0) {
584 result = -errno;
585 }
586 }
587 return result;
588 }
589
590 /*
591 * Returns true iff the specified sector is present in the disk image. Drivers
592 * not implementing the functionality are assumed to not support backing files,
593 * hence all their sectors are reported as allocated.
594 *
595 * If 'sector_num' is beyond the end of the disk image the return value is 0
596 * and 'pnum' is set to 0.
597 *
598 * 'pnum' is set to the number of sectors (including and immediately following
599 * the specified sector) that are known to be in the same
600 * allocated/unallocated state.
601 *
602 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
603 * beyond the end of the disk image it will be clamped.
604 */
605 static int coroutine_fn raw_co_is_allocated(BlockDriverState *bs,
606 int64_t sector_num,
607 int nb_sectors, int *pnum)
608 {
609 off_t start, data, hole;
610 int ret;
611
612 ret = fd_open(bs);
613 if (ret < 0) {
614 return ret;
615 }
616
617 start = sector_num * BDRV_SECTOR_SIZE;
618
619 #ifdef CONFIG_FIEMAP
620
621 BDRVRawState *s = bs->opaque;
622 struct {
623 struct fiemap fm;
624 struct fiemap_extent fe;
625 } f;
626
627 f.fm.fm_start = start;
628 f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE;
629 f.fm.fm_flags = 0;
630 f.fm.fm_extent_count = 1;
631 f.fm.fm_reserved = 0;
632 if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) {
633 /* Assume everything is allocated. */
634 *pnum = nb_sectors;
635 return 1;
636 }
637
638 if (f.fm.fm_mapped_extents == 0) {
639 /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
640 * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
641 */
642 off_t length = lseek(s->fd, 0, SEEK_END);
643 hole = f.fm.fm_start;
644 data = MIN(f.fm.fm_start + f.fm.fm_length, length);
645 } else {
646 data = f.fe.fe_logical;
647 hole = f.fe.fe_logical + f.fe.fe_length;
648 }
649
650 #elif defined SEEK_HOLE && defined SEEK_DATA
651
652 BDRVRawState *s = bs->opaque;
653
654 hole = lseek(s->fd, start, SEEK_HOLE);
655 if (hole == -1) {
656 /* -ENXIO indicates that sector_num was past the end of the file.
657 * There is a virtual hole there. */
658 assert(errno != -ENXIO);
659
660 /* Most likely EINVAL. Assume everything is allocated. */
661 *pnum = nb_sectors;
662 return 1;
663 }
664
665 if (hole > start) {
666 data = start;
667 } else {
668 /* On a hole. We need another syscall to find its end. */
669 data = lseek(s->fd, start, SEEK_DATA);
670 if (data == -1) {
671 data = lseek(s->fd, 0, SEEK_END);
672 }
673 }
674 #else
675 *pnum = nb_sectors;
676 return 1;
677 #endif
678
679 if (data <= start) {
680 /* On a data extent, compute sectors to the end of the extent. */
681 *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
682 return 1;
683 } else {
684 /* On a hole, compute sectors to the beginning of the next extent. */
685 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
686 return 0;
687 }
688 }
689
690 #ifdef CONFIG_XFS
691 static int xfs_discard(BDRVRawState *s, int64_t sector_num, int nb_sectors)
692 {
693 struct xfs_flock64 fl;
694
695 memset(&fl, 0, sizeof(fl));
696 fl.l_whence = SEEK_SET;
697 fl.l_start = sector_num << 9;
698 fl.l_len = (int64_t)nb_sectors << 9;
699
700 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
701 DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
702 return -errno;
703 }
704
705 return 0;
706 }
707 #endif
708
709 static coroutine_fn int raw_co_discard(BlockDriverState *bs,
710 int64_t sector_num, int nb_sectors)
711 {
712 #ifdef CONFIG_XFS
713 BDRVRawState *s = bs->opaque;
714
715 if (s->is_xfs) {
716 return xfs_discard(s, sector_num, nb_sectors);
717 }
718 #endif
719
720 return 0;
721 }
722
723 static QEMUOptionParameter raw_create_options[] = {
724 {
725 .name = BLOCK_OPT_SIZE,
726 .type = OPT_SIZE,
727 .help = "Virtual disk size"
728 },
729 { NULL }
730 };
731
732 static BlockDriver bdrv_file = {
733 .format_name = "file",
734 .protocol_name = "file",
735 .instance_size = sizeof(BDRVRawState),
736 .bdrv_probe = NULL, /* no probe for protocols */
737 .bdrv_file_open = raw_open,
738 .bdrv_close = raw_close,
739 .bdrv_create = raw_create,
740 .bdrv_co_discard = raw_co_discard,
741 .bdrv_co_is_allocated = raw_co_is_allocated,
742
743 .bdrv_aio_readv = raw_aio_readv,
744 .bdrv_aio_writev = raw_aio_writev,
745 .bdrv_aio_flush = raw_aio_flush,
746
747 .bdrv_truncate = raw_truncate,
748 .bdrv_getlength = raw_getlength,
749 .bdrv_get_allocated_file_size
750 = raw_get_allocated_file_size,
751
752 .create_options = raw_create_options,
753 };
754
755 /***********************************************/
756 /* host device */
757
758 #if defined(__APPLE__) && defined(__MACH__)
759 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
760 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
761
762 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
763 {
764 kern_return_t kernResult;
765 mach_port_t masterPort;
766 CFMutableDictionaryRef classesToMatch;
767
768 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
769 if ( KERN_SUCCESS != kernResult ) {
770 printf( "IOMasterPort returned %d\n", kernResult );
771 }
772
773 classesToMatch = IOServiceMatching( kIOCDMediaClass );
774 if ( classesToMatch == NULL ) {
775 printf( "IOServiceMatching returned a NULL dictionary.\n" );
776 } else {
777 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
778 }
779 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
780 if ( KERN_SUCCESS != kernResult )
781 {
782 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
783 }
784
785 return kernResult;
786 }
787
788 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
789 {
790 io_object_t nextMedia;
791 kern_return_t kernResult = KERN_FAILURE;
792 *bsdPath = '\0';
793 nextMedia = IOIteratorNext( mediaIterator );
794 if ( nextMedia )
795 {
796 CFTypeRef bsdPathAsCFString;
797 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
798 if ( bsdPathAsCFString ) {
799 size_t devPathLength;
800 strcpy( bsdPath, _PATH_DEV );
801 strcat( bsdPath, "r" );
802 devPathLength = strlen( bsdPath );
803 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
804 kernResult = KERN_SUCCESS;
805 }
806 CFRelease( bsdPathAsCFString );
807 }
808 IOObjectRelease( nextMedia );
809 }
810
811 return kernResult;
812 }
813
814 #endif
815
816 static int hdev_probe_device(const char *filename)
817 {
818 struct stat st;
819
820 /* allow a dedicated CD-ROM driver to match with a higher priority */
821 if (strstart(filename, "/dev/cdrom", NULL))
822 return 50;
823
824 if (stat(filename, &st) >= 0 &&
825 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
826 return 100;
827 }
828
829 return 0;
830 }
831
832 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
833 {
834 BDRVRawState *s = bs->opaque;
835
836 #if defined(__APPLE__) && defined(__MACH__)
837 if (strstart(filename, "/dev/cdrom", NULL)) {
838 kern_return_t kernResult;
839 io_iterator_t mediaIterator;
840 char bsdPath[ MAXPATHLEN ];
841 int fd;
842
843 kernResult = FindEjectableCDMedia( &mediaIterator );
844 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
845
846 if ( bsdPath[ 0 ] != '\0' ) {
847 strcat(bsdPath,"s0");
848 /* some CDs don't have a partition 0 */
849 fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
850 if (fd < 0) {
851 bsdPath[strlen(bsdPath)-1] = '1';
852 } else {
853 qemu_close(fd);
854 }
855 filename = bsdPath;
856 }
857
858 if ( mediaIterator )
859 IOObjectRelease( mediaIterator );
860 }
861 #endif
862
863 s->type = FTYPE_FILE;
864 #if defined(__linux__)
865 {
866 char resolved_path[ MAXPATHLEN ], *temp;
867
868 temp = realpath(filename, resolved_path);
869 if (temp && strstart(temp, "/dev/sg", NULL)) {
870 bs->sg = 1;
871 }
872 }
873 #endif
874
875 return raw_open_common(bs, filename, flags, 0);
876 }
877
878 #if defined(__linux__)
879 /* Note: we do not have a reliable method to detect if the floppy is
880 present. The current method is to try to open the floppy at every
881 I/O and to keep it opened during a few hundreds of ms. */
882 static int fd_open(BlockDriverState *bs)
883 {
884 BDRVRawState *s = bs->opaque;
885 int last_media_present;
886
887 if (s->type != FTYPE_FD)
888 return 0;
889 last_media_present = (s->fd >= 0);
890 if (s->fd >= 0 &&
891 (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
892 qemu_close(s->fd);
893 s->fd = -1;
894 #ifdef DEBUG_FLOPPY
895 printf("Floppy closed\n");
896 #endif
897 }
898 if (s->fd < 0) {
899 if (s->fd_got_error &&
900 (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
901 #ifdef DEBUG_FLOPPY
902 printf("No floppy (open delayed)\n");
903 #endif
904 return -EIO;
905 }
906 s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
907 if (s->fd < 0) {
908 s->fd_error_time = get_clock();
909 s->fd_got_error = 1;
910 if (last_media_present)
911 s->fd_media_changed = 1;
912 #ifdef DEBUG_FLOPPY
913 printf("No floppy\n");
914 #endif
915 return -EIO;
916 }
917 #ifdef DEBUG_FLOPPY
918 printf("Floppy opened\n");
919 #endif
920 }
921 if (!last_media_present)
922 s->fd_media_changed = 1;
923 s->fd_open_time = get_clock();
924 s->fd_got_error = 0;
925 return 0;
926 }
927
928 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
929 {
930 BDRVRawState *s = bs->opaque;
931
932 return ioctl(s->fd, req, buf);
933 }
934
935 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
936 unsigned long int req, void *buf,
937 BlockDriverCompletionFunc *cb, void *opaque)
938 {
939 BDRVRawState *s = bs->opaque;
940
941 if (fd_open(bs) < 0)
942 return NULL;
943 return paio_ioctl(bs, s->fd, req, buf, cb, opaque);
944 }
945
946 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
947 static int fd_open(BlockDriverState *bs)
948 {
949 BDRVRawState *s = bs->opaque;
950
951 /* this is just to ensure s->fd is sane (its called by io ops) */
952 if (s->fd >= 0)
953 return 0;
954 return -EIO;
955 }
956 #else /* !linux && !FreeBSD */
957
958 static int fd_open(BlockDriverState *bs)
959 {
960 return 0;
961 }
962
963 #endif /* !linux && !FreeBSD */
964
965 static int hdev_create(const char *filename, QEMUOptionParameter *options)
966 {
967 int fd;
968 int ret = 0;
969 struct stat stat_buf;
970 int64_t total_size = 0;
971
972 /* Read out options */
973 while (options && options->name) {
974 if (!strcmp(options->name, "size")) {
975 total_size = options->value.n / BDRV_SECTOR_SIZE;
976 }
977 options++;
978 }
979
980 fd = qemu_open(filename, O_WRONLY | O_BINARY);
981 if (fd < 0)
982 return -errno;
983
984 if (fstat(fd, &stat_buf) < 0)
985 ret = -errno;
986 else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode))
987 ret = -ENODEV;
988 else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE)
989 ret = -ENOSPC;
990
991 qemu_close(fd);
992 return ret;
993 }
994
995 static int hdev_has_zero_init(BlockDriverState *bs)
996 {
997 return 0;
998 }
999
1000 static BlockDriver bdrv_host_device = {
1001 .format_name = "host_device",
1002 .protocol_name = "host_device",
1003 .instance_size = sizeof(BDRVRawState),
1004 .bdrv_probe_device = hdev_probe_device,
1005 .bdrv_file_open = hdev_open,
1006 .bdrv_close = raw_close,
1007 .bdrv_create = hdev_create,
1008 .create_options = raw_create_options,
1009 .bdrv_has_zero_init = hdev_has_zero_init,
1010
1011 .bdrv_aio_readv = raw_aio_readv,
1012 .bdrv_aio_writev = raw_aio_writev,
1013 .bdrv_aio_flush = raw_aio_flush,
1014
1015 .bdrv_truncate = raw_truncate,
1016 .bdrv_getlength = raw_getlength,
1017 .bdrv_get_allocated_file_size
1018 = raw_get_allocated_file_size,
1019
1020 /* generic scsi device */
1021 #ifdef __linux__
1022 .bdrv_ioctl = hdev_ioctl,
1023 .bdrv_aio_ioctl = hdev_aio_ioctl,
1024 #endif
1025 };
1026
1027 #ifdef __linux__
1028 static int floppy_open(BlockDriverState *bs, const char *filename, int flags)
1029 {
1030 BDRVRawState *s = bs->opaque;
1031 int ret;
1032
1033 s->type = FTYPE_FD;
1034
1035 /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1036 ret = raw_open_common(bs, filename, flags, O_NONBLOCK);
1037 if (ret)
1038 return ret;
1039
1040 /* close fd so that we can reopen it as needed */
1041 qemu_close(s->fd);
1042 s->fd = -1;
1043 s->fd_media_changed = 1;
1044
1045 return 0;
1046 }
1047
1048 static int floppy_probe_device(const char *filename)
1049 {
1050 int fd, ret;
1051 int prio = 0;
1052 struct floppy_struct fdparam;
1053 struct stat st;
1054
1055 if (strstart(filename, "/dev/fd", NULL) &&
1056 !strstart(filename, "/dev/fdset/", NULL)) {
1057 prio = 50;
1058 }
1059
1060 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1061 if (fd < 0) {
1062 goto out;
1063 }
1064 ret = fstat(fd, &st);
1065 if (ret == -1 || !S_ISBLK(st.st_mode)) {
1066 goto outc;
1067 }
1068
1069 /* Attempt to detect via a floppy specific ioctl */
1070 ret = ioctl(fd, FDGETPRM, &fdparam);
1071 if (ret >= 0)
1072 prio = 100;
1073
1074 outc:
1075 qemu_close(fd);
1076 out:
1077 return prio;
1078 }
1079
1080
1081 static int floppy_is_inserted(BlockDriverState *bs)
1082 {
1083 return fd_open(bs) >= 0;
1084 }
1085
1086 static int floppy_media_changed(BlockDriverState *bs)
1087 {
1088 BDRVRawState *s = bs->opaque;
1089 int ret;
1090
1091 /*
1092 * XXX: we do not have a true media changed indication.
1093 * It does not work if the floppy is changed without trying to read it.
1094 */
1095 fd_open(bs);
1096 ret = s->fd_media_changed;
1097 s->fd_media_changed = 0;
1098 #ifdef DEBUG_FLOPPY
1099 printf("Floppy changed=%d\n", ret);
1100 #endif
1101 return ret;
1102 }
1103
1104 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
1105 {
1106 BDRVRawState *s = bs->opaque;
1107 int fd;
1108
1109 if (s->fd >= 0) {
1110 qemu_close(s->fd);
1111 s->fd = -1;
1112 }
1113 fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
1114 if (fd >= 0) {
1115 if (ioctl(fd, FDEJECT, 0) < 0)
1116 perror("FDEJECT");
1117 qemu_close(fd);
1118 }
1119 }
1120
1121 static BlockDriver bdrv_host_floppy = {
1122 .format_name = "host_floppy",
1123 .protocol_name = "host_floppy",
1124 .instance_size = sizeof(BDRVRawState),
1125 .bdrv_probe_device = floppy_probe_device,
1126 .bdrv_file_open = floppy_open,
1127 .bdrv_close = raw_close,
1128 .bdrv_create = hdev_create,
1129 .create_options = raw_create_options,
1130 .bdrv_has_zero_init = hdev_has_zero_init,
1131
1132 .bdrv_aio_readv = raw_aio_readv,
1133 .bdrv_aio_writev = raw_aio_writev,
1134 .bdrv_aio_flush = raw_aio_flush,
1135
1136 .bdrv_truncate = raw_truncate,
1137 .bdrv_getlength = raw_getlength,
1138 .bdrv_get_allocated_file_size
1139 = raw_get_allocated_file_size,
1140
1141 /* removable device support */
1142 .bdrv_is_inserted = floppy_is_inserted,
1143 .bdrv_media_changed = floppy_media_changed,
1144 .bdrv_eject = floppy_eject,
1145 };
1146
1147 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1148 {
1149 BDRVRawState *s = bs->opaque;
1150
1151 s->type = FTYPE_CD;
1152
1153 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
1154 return raw_open_common(bs, filename, flags, O_NONBLOCK);
1155 }
1156
1157 static int cdrom_probe_device(const char *filename)
1158 {
1159 int fd, ret;
1160 int prio = 0;
1161 struct stat st;
1162
1163 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1164 if (fd < 0) {
1165 goto out;
1166 }
1167 ret = fstat(fd, &st);
1168 if (ret == -1 || !S_ISBLK(st.st_mode)) {
1169 goto outc;
1170 }
1171
1172 /* Attempt to detect via a CDROM specific ioctl */
1173 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1174 if (ret >= 0)
1175 prio = 100;
1176
1177 outc:
1178 qemu_close(fd);
1179 out:
1180 return prio;
1181 }
1182
1183 static int cdrom_is_inserted(BlockDriverState *bs)
1184 {
1185 BDRVRawState *s = bs->opaque;
1186 int ret;
1187
1188 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1189 if (ret == CDS_DISC_OK)
1190 return 1;
1191 return 0;
1192 }
1193
1194 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1195 {
1196 BDRVRawState *s = bs->opaque;
1197
1198 if (eject_flag) {
1199 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
1200 perror("CDROMEJECT");
1201 } else {
1202 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
1203 perror("CDROMEJECT");
1204 }
1205 }
1206
1207 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1208 {
1209 BDRVRawState *s = bs->opaque;
1210
1211 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
1212 /*
1213 * Note: an error can happen if the distribution automatically
1214 * mounts the CD-ROM
1215 */
1216 /* perror("CDROM_LOCKDOOR"); */
1217 }
1218 }
1219
1220 static BlockDriver bdrv_host_cdrom = {
1221 .format_name = "host_cdrom",
1222 .protocol_name = "host_cdrom",
1223 .instance_size = sizeof(BDRVRawState),
1224 .bdrv_probe_device = cdrom_probe_device,
1225 .bdrv_file_open = cdrom_open,
1226 .bdrv_close = raw_close,
1227 .bdrv_create = hdev_create,
1228 .create_options = raw_create_options,
1229 .bdrv_has_zero_init = hdev_has_zero_init,
1230
1231 .bdrv_aio_readv = raw_aio_readv,
1232 .bdrv_aio_writev = raw_aio_writev,
1233 .bdrv_aio_flush = raw_aio_flush,
1234
1235 .bdrv_truncate = raw_truncate,
1236 .bdrv_getlength = raw_getlength,
1237 .bdrv_get_allocated_file_size
1238 = raw_get_allocated_file_size,
1239
1240 /* removable device support */
1241 .bdrv_is_inserted = cdrom_is_inserted,
1242 .bdrv_eject = cdrom_eject,
1243 .bdrv_lock_medium = cdrom_lock_medium,
1244
1245 /* generic scsi device */
1246 .bdrv_ioctl = hdev_ioctl,
1247 .bdrv_aio_ioctl = hdev_aio_ioctl,
1248 };
1249 #endif /* __linux__ */
1250
1251 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1252 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1253 {
1254 BDRVRawState *s = bs->opaque;
1255 int ret;
1256
1257 s->type = FTYPE_CD;
1258
1259 ret = raw_open_common(bs, filename, flags, 0);
1260 if (ret)
1261 return ret;
1262
1263 /* make sure the door isn't locked at this time */
1264 ioctl(s->fd, CDIOCALLOW);
1265 return 0;
1266 }
1267
1268 static int cdrom_probe_device(const char *filename)
1269 {
1270 if (strstart(filename, "/dev/cd", NULL) ||
1271 strstart(filename, "/dev/acd", NULL))
1272 return 100;
1273 return 0;
1274 }
1275
1276 static int cdrom_reopen(BlockDriverState *bs)
1277 {
1278 BDRVRawState *s = bs->opaque;
1279 int fd;
1280
1281 /*
1282 * Force reread of possibly changed/newly loaded disc,
1283 * FreeBSD seems to not notice sometimes...
1284 */
1285 if (s->fd >= 0)
1286 qemu_close(s->fd);
1287 fd = qemu_open(bs->filename, s->open_flags, 0644);
1288 if (fd < 0) {
1289 s->fd = -1;
1290 return -EIO;
1291 }
1292 s->fd = fd;
1293
1294 /* make sure the door isn't locked at this time */
1295 ioctl(s->fd, CDIOCALLOW);
1296 return 0;
1297 }
1298
1299 static int cdrom_is_inserted(BlockDriverState *bs)
1300 {
1301 return raw_getlength(bs) > 0;
1302 }
1303
1304 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1305 {
1306 BDRVRawState *s = bs->opaque;
1307
1308 if (s->fd < 0)
1309 return;
1310
1311 (void) ioctl(s->fd, CDIOCALLOW);
1312
1313 if (eject_flag) {
1314 if (ioctl(s->fd, CDIOCEJECT) < 0)
1315 perror("CDIOCEJECT");
1316 } else {
1317 if (ioctl(s->fd, CDIOCCLOSE) < 0)
1318 perror("CDIOCCLOSE");
1319 }
1320
1321 cdrom_reopen(bs);
1322 }
1323
1324 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1325 {
1326 BDRVRawState *s = bs->opaque;
1327
1328 if (s->fd < 0)
1329 return;
1330 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1331 /*
1332 * Note: an error can happen if the distribution automatically
1333 * mounts the CD-ROM
1334 */
1335 /* perror("CDROM_LOCKDOOR"); */
1336 }
1337 }
1338
1339 static BlockDriver bdrv_host_cdrom = {
1340 .format_name = "host_cdrom",
1341 .protocol_name = "host_cdrom",
1342 .instance_size = sizeof(BDRVRawState),
1343 .bdrv_probe_device = cdrom_probe_device,
1344 .bdrv_file_open = cdrom_open,
1345 .bdrv_close = raw_close,
1346 .bdrv_create = hdev_create,
1347 .create_options = raw_create_options,
1348 .bdrv_has_zero_init = hdev_has_zero_init,
1349
1350 .bdrv_aio_readv = raw_aio_readv,
1351 .bdrv_aio_writev = raw_aio_writev,
1352 .bdrv_aio_flush = raw_aio_flush,
1353
1354 .bdrv_truncate = raw_truncate,
1355 .bdrv_getlength = raw_getlength,
1356 .bdrv_get_allocated_file_size
1357 = raw_get_allocated_file_size,
1358
1359 /* removable device support */
1360 .bdrv_is_inserted = cdrom_is_inserted,
1361 .bdrv_eject = cdrom_eject,
1362 .bdrv_lock_medium = cdrom_lock_medium,
1363 };
1364 #endif /* __FreeBSD__ */
1365
1366 static void bdrv_file_init(void)
1367 {
1368 /*
1369 * Register all the drivers. Note that order is important, the driver
1370 * registered last will get probed first.
1371 */
1372 bdrv_register(&bdrv_file);
1373 bdrv_register(&bdrv_host_device);
1374 #ifdef __linux__
1375 bdrv_register(&bdrv_host_floppy);
1376 bdrv_register(&bdrv_host_cdrom);
1377 #endif
1378 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1379 bdrv_register(&bdrv_host_cdrom);
1380 #endif
1381 }
1382
1383 block_init(bdrv_file_init);