?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
bsd-user: Implement strace support for print_ioctl syscall
[qemu/kevin.git] / block / raw-posix.c
blobc2b30be3d359555ec57ceabec60c89b128241b0e
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 #endif
59 #ifdef CONFIG_FIEMAP
60 #include <linux/fiemap.h>
61 #endif
62 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
63 #include <linux/falloc.h>
64 #endif
65 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
66 #include <sys/disk.h>
67 #include <sys/cdio.h>
68 #endif
69
70 #ifdef __OpenBSD__
71 #include <sys/ioctl.h>
72 #include <sys/disklabel.h>
73 #include <sys/dkio.h>
74 #endif
75
76 #ifdef __NetBSD__
77 #include <sys/ioctl.h>
78 #include <sys/disklabel.h>
79 #include <sys/dkio.h>
80 #include <sys/disk.h>
81 #endif
82
83 #ifdef __DragonFly__
84 #include <sys/ioctl.h>
85 #include <sys/diskslice.h>
86 #endif
87
88 #ifdef CONFIG_XFS
89 #include <xfs/xfs.h>
90 #endif
91
92 //#define DEBUG_FLOPPY
93
94 //#define DEBUG_BLOCK
95 #if defined(DEBUG_BLOCK)
96 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
97 { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
98 #else
99 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
100 #endif
101
102 /* OS X does not have O_DSYNC */
103 #ifndef O_DSYNC
104 #ifdef O_SYNC
105 #define O_DSYNC O_SYNC
106 #elif defined(O_FSYNC)
107 #define O_DSYNC O_FSYNC
108 #endif
109 #endif
110
111 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
112 #ifndef O_DIRECT
113 #define O_DIRECT O_DSYNC
114 #endif
115
116 #define FTYPE_FILE 0
117 #define FTYPE_CD 1
118 #define FTYPE_FD 2
119
120 /* if the FD is not accessed during that time (in ns), we try to
121 reopen it to see if the disk has been changed */
122 #define FD_OPEN_TIMEOUT (1000000000)
123
124 #define MAX_BLOCKSIZE 4096
125
126 typedef struct BDRVRawState {
127 int fd;
128 int type;
129 int open_flags;
130 size_t buf_align;
131
132 #if defined(__linux__)
133 /* linux floppy specific */
134 int64_t fd_open_time;
135 int64_t fd_error_time;
136 int fd_got_error;
137 int fd_media_changed;
138 #endif
139 #ifdef CONFIG_LINUX_AIO
140 int use_aio;
141 void *aio_ctx;
142 #endif
143 #ifdef CONFIG_XFS
144 bool is_xfs:1;
145 #endif
146 bool has_discard:1;
147 bool has_write_zeroes:1;
148 bool discard_zeroes:1;
149 #ifdef CONFIG_FIEMAP
150 bool skip_fiemap;
151 #endif
152 } BDRVRawState;
153
154 typedef struct BDRVRawReopenState {
155 int fd;
156 int open_flags;
157 #ifdef CONFIG_LINUX_AIO
158 int use_aio;
159 #endif
160 } BDRVRawReopenState;
161
162 static int fd_open(BlockDriverState *bs);
163 static int64_t raw_getlength(BlockDriverState *bs);
164
165 typedef struct RawPosixAIOData {
166 BlockDriverState *bs;
167 int aio_fildes;
168 union {
169 struct iovec *aio_iov;
170 void *aio_ioctl_buf;
171 };
172 int aio_niov;
173 uint64_t aio_nbytes;
174 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
175 off_t aio_offset;
176 int aio_type;
177 } RawPosixAIOData;
178
179 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
180 static int cdrom_reopen(BlockDriverState *bs);
181 #endif
182
183 #if defined(__NetBSD__)
184 static int raw_normalize_devicepath(const char **filename)
185 {
186 static char namebuf[PATH_MAX];
187 const char *dp, *fname;
188 struct stat sb;
189
190 fname = *filename;
191 dp = strrchr(fname, '/');
192 if (lstat(fname, &sb) < 0) {
193 fprintf(stderr, "%s: stat failed: %s\n",
194 fname, strerror(errno));
195 return -errno;
196 }
197
198 if (!S_ISBLK(sb.st_mode)) {
199 return 0;
200 }
201
202 if (dp == NULL) {
203 snprintf(namebuf, PATH_MAX, "r%s", fname);
204 } else {
205 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
206 (int)(dp - fname), fname, dp + 1);
207 }
208 fprintf(stderr, "%s is a block device", fname);
209 *filename = namebuf;
210 fprintf(stderr, ", using %s\n", *filename);
211
212 return 0;
213 }
214 #else
215 static int raw_normalize_devicepath(const char **filename)
216 {
217 return 0;
218 }
219 #endif
220
221 static void raw_probe_alignment(BlockDriverState *bs)
222 {
223 BDRVRawState *s = bs->opaque;
224 char *buf;
225 unsigned int sector_size;
226
227 /* For /dev/sg devices the alignment is not really used.
228 With buffered I/O, we don't have any restrictions. */
229 if (bs->sg || !(s->open_flags & O_DIRECT)) {
230 bs->request_alignment = 1;
231 s->buf_align = 1;
232 return;
233 }
234
235 /* Try a few ioctls to get the right size */
236 bs->request_alignment = 0;
237 s->buf_align = 0;
238
239 #ifdef BLKSSZGET
240 if (ioctl(s->fd, BLKSSZGET, &sector_size) >= 0) {
241 bs->request_alignment = sector_size;
242 }
243 #endif
244 #ifdef DKIOCGETBLOCKSIZE
245 if (ioctl(s->fd, DKIOCGETBLOCKSIZE, &sector_size) >= 0) {
246 bs->request_alignment = sector_size;
247 }
248 #endif
249 #ifdef DIOCGSECTORSIZE
250 if (ioctl(s->fd, DIOCGSECTORSIZE, &sector_size) >= 0) {
251 bs->request_alignment = sector_size;
252 }
253 #endif
254 #ifdef CONFIG_XFS
255 if (s->is_xfs) {
256 struct dioattr da;
257 if (xfsctl(NULL, s->fd, XFS_IOC_DIOINFO, &da) >= 0) {
258 bs->request_alignment = da.d_miniosz;
259 /* The kernel returns wrong information for d_mem */
260 /* s->buf_align = da.d_mem; */
261 }
262 }
263 #endif
264
265 /* If we could not get the sizes so far, we can only guess them */
266 if (!s->buf_align) {
267 size_t align;
268 buf = qemu_memalign(MAX_BLOCKSIZE, 2 * MAX_BLOCKSIZE);
269 for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
270 if (pread(s->fd, buf + align, MAX_BLOCKSIZE, 0) >= 0) {
271 s->buf_align = align;
272 break;
273 }
274 }
275 qemu_vfree(buf);
276 }
277
278 if (!bs->request_alignment) {
279 size_t align;
280 buf = qemu_memalign(s->buf_align, MAX_BLOCKSIZE);
281 for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
282 if (pread(s->fd, buf, align, 0) >= 0) {
283 bs->request_alignment = align;
284 break;
285 }
286 }
287 qemu_vfree(buf);
288 }
289 }
290
291 static void raw_parse_flags(int bdrv_flags, int *open_flags)
292 {
293 assert(open_flags != NULL);
294
295 *open_flags |= O_BINARY;
296 *open_flags &= ~O_ACCMODE;
297 if (bdrv_flags & BDRV_O_RDWR) {
298 *open_flags |= O_RDWR;
299 } else {
300 *open_flags |= O_RDONLY;
301 }
302
303 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
304 * and O_DIRECT for no caching. */
305 if ((bdrv_flags & BDRV_O_NOCACHE)) {
306 *open_flags |= O_DIRECT;
307 }
308 }
309
310 static void raw_detach_aio_context(BlockDriverState *bs)
311 {
312 #ifdef CONFIG_LINUX_AIO
313 BDRVRawState *s = bs->opaque;
314
315 if (s->use_aio) {
316 laio_detach_aio_context(s->aio_ctx, bdrv_get_aio_context(bs));
317 }
318 #endif
319 }
320
321 static void raw_attach_aio_context(BlockDriverState *bs,
322 AioContext *new_context)
323 {
324 #ifdef CONFIG_LINUX_AIO
325 BDRVRawState *s = bs->opaque;
326
327 if (s->use_aio) {
328 laio_attach_aio_context(s->aio_ctx, new_context);
329 }
330 #endif
331 }
332
333 #ifdef CONFIG_LINUX_AIO
334 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
335 {
336 int ret = -1;
337 assert(aio_ctx != NULL);
338 assert(use_aio != NULL);
339 /*
340 * Currently Linux do AIO only for files opened with O_DIRECT
341 * specified so check NOCACHE flag too
342 */
343 if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
344 (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
345
346 /* if non-NULL, laio_init() has already been run */
347 if (*aio_ctx == NULL) {
348 *aio_ctx = laio_init();
349 if (!*aio_ctx) {
350 goto error;
351 }
352 }
353 *use_aio = 1;
354 } else {
355 *use_aio = 0;
356 }
357
358 ret = 0;
359
360 error:
361 return ret;
362 }
363 #endif
364
365 static void raw_parse_filename(const char *filename, QDict *options,
366 Error **errp)
367 {
368 /* The filename does not have to be prefixed by the protocol name, since
369 * "file" is the default protocol; therefore, the return value of this
370 * function call can be ignored. */
371 strstart(filename, "file:", &filename);
372
373 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
374 }
375
376 static QemuOptsList raw_runtime_opts = {
377 .name = "raw",
378 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
379 .desc = {
380 {
381 .name = "filename",
382 .type = QEMU_OPT_STRING,
383 .help = "File name of the image",
384 },
385 { /* end of list */ }
386 },
387 };
388
389 static int raw_open_common(BlockDriverState *bs, QDict *options,
390 int bdrv_flags, int open_flags, Error **errp)
391 {
392 BDRVRawState *s = bs->opaque;
393 QemuOpts *opts;
394 Error *local_err = NULL;
395 const char *filename = NULL;
396 int fd, ret;
397 struct stat st;
398
399 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
400 qemu_opts_absorb_qdict(opts, options, &local_err);
401 if (local_err) {
402 error_propagate(errp, local_err);
403 ret = -EINVAL;
404 goto fail;
405 }
406
407 filename = qemu_opt_get(opts, "filename");
408
409 ret = raw_normalize_devicepath(&filename);
410 if (ret != 0) {
411 error_setg_errno(errp, -ret, "Could not normalize device path");
412 goto fail;
413 }
414
415 s->open_flags = open_flags;
416 raw_parse_flags(bdrv_flags, &s->open_flags);
417
418 s->fd = -1;
419 fd = qemu_open(filename, s->open_flags, 0644);
420 if (fd < 0) {
421 ret = -errno;
422 if (ret == -EROFS) {
423 ret = -EACCES;
424 }
425 goto fail;
426 }
427 s->fd = fd;
428
429 #ifdef CONFIG_LINUX_AIO
430 if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
431 qemu_close(fd);
432 ret = -errno;
433 error_setg_errno(errp, -ret, "Could not set AIO state");
434 goto fail;
435 }
436 #endif
437
438 s->has_discard = true;
439 s->has_write_zeroes = true;
440
441 if (fstat(s->fd, &st) < 0) {
442 error_setg_errno(errp, errno, "Could not stat file");
443 goto fail;
444 }
445 if (S_ISREG(st.st_mode)) {
446 s->discard_zeroes = true;
447 }
448 if (S_ISBLK(st.st_mode)) {
449 #ifdef BLKDISCARDZEROES
450 unsigned int arg;
451 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
452 s->discard_zeroes = true;
453 }
454 #endif
455 #ifdef __linux__
456 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
457 * not rely on the contents of discarded blocks unless using O_DIRECT.
458 * Same for BLKZEROOUT.
459 */
460 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
461 s->discard_zeroes = false;
462 s->has_write_zeroes = false;
463 }
464 #endif
465 }
466
467 #ifdef CONFIG_XFS
468 if (platform_test_xfs_fd(s->fd)) {
469 s->is_xfs = true;
470 }
471 #endif
472
473 raw_attach_aio_context(bs, bdrv_get_aio_context(bs));
474
475 ret = 0;
476 fail:
477 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
478 unlink(filename);
479 }
480 qemu_opts_del(opts);
481 return ret;
482 }
483
484 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
485 Error **errp)
486 {
487 BDRVRawState *s = bs->opaque;
488 Error *local_err = NULL;
489 int ret;
490
491 s->type = FTYPE_FILE;
492 ret = raw_open_common(bs, options, flags, 0, &local_err);
493 if (local_err) {
494 error_propagate(errp, local_err);
495 }
496 return ret;
497 }
498
499 static int raw_reopen_prepare(BDRVReopenState *state,
500 BlockReopenQueue *queue, Error **errp)
501 {
502 BDRVRawState *s;
503 BDRVRawReopenState *raw_s;
504 int ret = 0;
505
506 assert(state != NULL);
507 assert(state->bs != NULL);
508
509 s = state->bs->opaque;
510
511 state->opaque = g_malloc0(sizeof(BDRVRawReopenState));
512 raw_s = state->opaque;
513
514 #ifdef CONFIG_LINUX_AIO
515 raw_s->use_aio = s->use_aio;
516
517 /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
518 * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
519 * won't override aio_ctx if aio_ctx is non-NULL */
520 if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
521 error_setg(errp, "Could not set AIO state");
522 return -1;
523 }
524 #endif
525
526 if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
527 raw_s->open_flags |= O_NONBLOCK;
528 }
529
530 raw_parse_flags(state->flags, &raw_s->open_flags);
531
532 raw_s->fd = -1;
533
534 int fcntl_flags = O_APPEND | O_NONBLOCK;
535 #ifdef O_NOATIME
536 fcntl_flags |= O_NOATIME;
537 #endif
538
539 #ifdef O_ASYNC
540 /* Not all operating systems have O_ASYNC, and those that don't
541 * will not let us track the state into raw_s->open_flags (typically
542 * you achieve the same effect with an ioctl, for example I_SETSIG
543 * on Solaris). But we do not use O_ASYNC, so that's fine.
544 */
545 assert((s->open_flags & O_ASYNC) == 0);
546 #endif
547
548 if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
549 /* dup the original fd */
550 /* TODO: use qemu fcntl wrapper */
551 #ifdef F_DUPFD_CLOEXEC
552 raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
553 #else
554 raw_s->fd = dup(s->fd);
555 if (raw_s->fd != -1) {
556 qemu_set_cloexec(raw_s->fd);
557 }
558 #endif
559 if (raw_s->fd >= 0) {
560 ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
561 if (ret) {
562 qemu_close(raw_s->fd);
563 raw_s->fd = -1;
564 }
565 }
566 }
567
568 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
569 if (raw_s->fd == -1) {
570 assert(!(raw_s->open_flags & O_CREAT));
571 raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
572 if (raw_s->fd == -1) {
573 error_setg_errno(errp, errno, "Could not reopen file");
574 ret = -1;
575 }
576 }
577 return ret;
578 }
579
580 static void raw_reopen_commit(BDRVReopenState *state)
581 {
582 BDRVRawReopenState *raw_s = state->opaque;
583 BDRVRawState *s = state->bs->opaque;
584
585 s->open_flags = raw_s->open_flags;
586
587 qemu_close(s->fd);
588 s->fd = raw_s->fd;
589 #ifdef CONFIG_LINUX_AIO
590 s->use_aio = raw_s->use_aio;
591 #endif
592
593 g_free(state->opaque);
594 state->opaque = NULL;
595 }
596
597
598 static void raw_reopen_abort(BDRVReopenState *state)
599 {
600 BDRVRawReopenState *raw_s = state->opaque;
601
602 /* nothing to do if NULL, we didn't get far enough */
603 if (raw_s == NULL) {
604 return;
605 }
606
607 if (raw_s->fd >= 0) {
608 qemu_close(raw_s->fd);
609 raw_s->fd = -1;
610 }
611 g_free(state->opaque);
612 state->opaque = NULL;
613 }
614
615 static int raw_refresh_limits(BlockDriverState *bs)
616 {
617 BDRVRawState *s = bs->opaque;
618
619 raw_probe_alignment(bs);
620 bs->bl.opt_mem_alignment = s->buf_align;
621
622 return 0;
623 }
624
625 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
626 {
627 int ret;
628
629 ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
630 if (ret == -1) {
631 return -errno;
632 }
633
634 return 0;
635 }
636
637 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
638 {
639 int ret;
640
641 ret = qemu_fdatasync(aiocb->aio_fildes);
642 if (ret == -1) {
643 return -errno;
644 }
645 return 0;
646 }
647
648 #ifdef CONFIG_PREADV
649
650 static bool preadv_present = true;
651
652 static ssize_t
653 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
654 {
655 return preadv(fd, iov, nr_iov, offset);
656 }
657
658 static ssize_t
659 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
660 {
661 return pwritev(fd, iov, nr_iov, offset);
662 }
663
664 #else
665
666 static bool preadv_present = false;
667
668 static ssize_t
669 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
670 {
671 return -ENOSYS;
672 }
673
674 static ssize_t
675 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
676 {
677 return -ENOSYS;
678 }
679
680 #endif
681
682 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
683 {
684 ssize_t len;
685
686 do {
687 if (aiocb->aio_type & QEMU_AIO_WRITE)
688 len = qemu_pwritev(aiocb->aio_fildes,
689 aiocb->aio_iov,
690 aiocb->aio_niov,
691 aiocb->aio_offset);
692 else
693 len = qemu_preadv(aiocb->aio_fildes,
694 aiocb->aio_iov,
695 aiocb->aio_niov,
696 aiocb->aio_offset);
697 } while (len == -1 && errno == EINTR);
698
699 if (len == -1) {
700 return -errno;
701 }
702 return len;
703 }
704
705 /*
706 * Read/writes the data to/from a given linear buffer.
707 *
708 * Returns the number of bytes handles or -errno in case of an error. Short
709 * reads are only returned if the end of the file is reached.
710 */
711 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
712 {
713 ssize_t offset = 0;
714 ssize_t len;
715
716 while (offset < aiocb->aio_nbytes) {
717 if (aiocb->aio_type & QEMU_AIO_WRITE) {
718 len = pwrite(aiocb->aio_fildes,
719 (const char *)buf + offset,
720 aiocb->aio_nbytes - offset,
721 aiocb->aio_offset + offset);
722 } else {
723 len = pread(aiocb->aio_fildes,
724 buf + offset,
725 aiocb->aio_nbytes - offset,
726 aiocb->aio_offset + offset);
727 }
728 if (len == -1 && errno == EINTR) {
729 continue;
730 } else if (len == -1) {
731 offset = -errno;
732 break;
733 } else if (len == 0) {
734 break;
735 }
736 offset += len;
737 }
738
739 return offset;
740 }
741
742 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
743 {
744 ssize_t nbytes;
745 char *buf;
746
747 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
748 /*
749 * If there is just a single buffer, and it is properly aligned
750 * we can just use plain pread/pwrite without any problems.
751 */
752 if (aiocb->aio_niov == 1) {
753 return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
754 }
755 /*
756 * We have more than one iovec, and all are properly aligned.
757 *
758 * Try preadv/pwritev first and fall back to linearizing the
759 * buffer if it's not supported.
760 */
761 if (preadv_present) {
762 nbytes = handle_aiocb_rw_vector(aiocb);
763 if (nbytes == aiocb->aio_nbytes ||
764 (nbytes < 0 && nbytes != -ENOSYS)) {
765 return nbytes;
766 }
767 preadv_present = false;
768 }
769
770 /*
771 * XXX(hch): short read/write. no easy way to handle the reminder
772 * using these interfaces. For now retry using plain
773 * pread/pwrite?
774 */
775 }
776
777 /*
778 * Ok, we have to do it the hard way, copy all segments into
779 * a single aligned buffer.
780 */
781 buf = qemu_blockalign(aiocb->bs, aiocb->aio_nbytes);
782 if (aiocb->aio_type & QEMU_AIO_WRITE) {
783 char *p = buf;
784 int i;
785
786 for (i = 0; i < aiocb->aio_niov; ++i) {
787 memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
788 p += aiocb->aio_iov[i].iov_len;
789 }
790 }
791
792 nbytes = handle_aiocb_rw_linear(aiocb, buf);
793 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
794 char *p = buf;
795 size_t count = aiocb->aio_nbytes, copy;
796 int i;
797
798 for (i = 0; i < aiocb->aio_niov && count; ++i) {
799 copy = count;
800 if (copy > aiocb->aio_iov[i].iov_len) {
801 copy = aiocb->aio_iov[i].iov_len;
802 }
803 memcpy(aiocb->aio_iov[i].iov_base, p, copy);
804 p += copy;
805 count -= copy;
806 }
807 }
808 qemu_vfree(buf);
809
810 return nbytes;
811 }
812
813 #ifdef CONFIG_XFS
814 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
815 {
816 struct xfs_flock64 fl;
817
818 memset(&fl, 0, sizeof(fl));
819 fl.l_whence = SEEK_SET;
820 fl.l_start = offset;
821 fl.l_len = bytes;
822
823 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
824 DEBUG_BLOCK_PRINT("cannot write zero range (%s)\n", strerror(errno));
825 return -errno;
826 }
827
828 return 0;
829 }
830
831 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
832 {
833 struct xfs_flock64 fl;
834
835 memset(&fl, 0, sizeof(fl));
836 fl.l_whence = SEEK_SET;
837 fl.l_start = offset;
838 fl.l_len = bytes;
839
840 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
841 DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
842 return -errno;
843 }
844
845 return 0;
846 }
847 #endif
848
849 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
850 {
851 int ret = -EOPNOTSUPP;
852 BDRVRawState *s = aiocb->bs->opaque;
853
854 if (s->has_write_zeroes == 0) {
855 return -ENOTSUP;
856 }
857
858 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
859 #ifdef BLKZEROOUT
860 do {
861 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
862 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
863 return 0;
864 }
865 } while (errno == EINTR);
866
867 ret = -errno;
868 #endif
869 } else {
870 #ifdef CONFIG_XFS
871 if (s->is_xfs) {
872 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
873 }
874 #endif
875 }
876
877 if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
878 ret == -ENOTTY) {
879 s->has_write_zeroes = false;
880 ret = -ENOTSUP;
881 }
882 return ret;
883 }
884
885 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
886 {
887 int ret = -EOPNOTSUPP;
888 BDRVRawState *s = aiocb->bs->opaque;
889
890 if (!s->has_discard) {
891 return -ENOTSUP;
892 }
893
894 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
895 #ifdef BLKDISCARD
896 do {
897 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
898 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
899 return 0;
900 }
901 } while (errno == EINTR);
902
903 ret = -errno;
904 #endif
905 } else {
906 #ifdef CONFIG_XFS
907 if (s->is_xfs) {
908 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
909 }
910 #endif
911
912 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
913 do {
914 if (fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
915 aiocb->aio_offset, aiocb->aio_nbytes) == 0) {
916 return 0;
917 }
918 } while (errno == EINTR);
919
920 ret = -errno;
921 #endif
922 }
923
924 if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
925 ret == -ENOTTY) {
926 s->has_discard = false;
927 ret = -ENOTSUP;
928 }
929 return ret;
930 }
931
932 static int aio_worker(void *arg)
933 {
934 RawPosixAIOData *aiocb = arg;
935 ssize_t ret = 0;
936
937 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
938 case QEMU_AIO_READ:
939 ret = handle_aiocb_rw(aiocb);
940 if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) {
941 iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
942 0, aiocb->aio_nbytes - ret);
943
944 ret = aiocb->aio_nbytes;
945 }
946 if (ret == aiocb->aio_nbytes) {
947 ret = 0;
948 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
949 ret = -EINVAL;
950 }
951 break;
952 case QEMU_AIO_WRITE:
953 ret = handle_aiocb_rw(aiocb);
954 if (ret == aiocb->aio_nbytes) {
955 ret = 0;
956 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
957 ret = -EINVAL;
958 }
959 break;
960 case QEMU_AIO_FLUSH:
961 ret = handle_aiocb_flush(aiocb);
962 break;
963 case QEMU_AIO_IOCTL:
964 ret = handle_aiocb_ioctl(aiocb);
965 break;
966 case QEMU_AIO_DISCARD:
967 ret = handle_aiocb_discard(aiocb);
968 break;
969 case QEMU_AIO_WRITE_ZEROES:
970 ret = handle_aiocb_write_zeroes(aiocb);
971 break;
972 default:
973 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
974 ret = -EINVAL;
975 break;
976 }
977
978 g_slice_free(RawPosixAIOData, aiocb);
979 return ret;
980 }
981
982 static int paio_submit_co(BlockDriverState *bs, int fd,
983 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
984 int type)
985 {
986 RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
987 ThreadPool *pool;
988
989 acb->bs = bs;
990 acb->aio_type = type;
991 acb->aio_fildes = fd;
992
993 if (qiov) {
994 acb->aio_iov = qiov->iov;
995 acb->aio_niov = qiov->niov;
996 }
997 acb->aio_nbytes = nb_sectors * 512;
998 acb->aio_offset = sector_num * 512;
999
1000 trace_paio_submit_co(sector_num, nb_sectors, type);
1001 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1002 return thread_pool_submit_co(pool, aio_worker, acb);
1003 }
1004
1005 static BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd,
1006 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1007 BlockDriverCompletionFunc *cb, void *opaque, int type)
1008 {
1009 RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
1010 ThreadPool *pool;
1011
1012 acb->bs = bs;
1013 acb->aio_type = type;
1014 acb->aio_fildes = fd;
1015
1016 if (qiov) {
1017 acb->aio_iov = qiov->iov;
1018 acb->aio_niov = qiov->niov;
1019 }
1020 acb->aio_nbytes = nb_sectors * 512;
1021 acb->aio_offset = sector_num * 512;
1022
1023 trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
1024 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1025 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1026 }
1027
1028 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
1029 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1030 BlockDriverCompletionFunc *cb, void *opaque, int type)
1031 {
1032 BDRVRawState *s = bs->opaque;
1033
1034 if (fd_open(bs) < 0)
1035 return NULL;
1036
1037 /*
1038 * If O_DIRECT is used the buffer needs to be aligned on a sector
1039 * boundary. Check if this is the case or tell the low-level
1040 * driver that it needs to copy the buffer.
1041 */
1042 if ((bs->open_flags & BDRV_O_NOCACHE)) {
1043 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1044 type |= QEMU_AIO_MISALIGNED;
1045 #ifdef CONFIG_LINUX_AIO
1046 } else if (s->use_aio) {
1047 return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
1048 nb_sectors, cb, opaque, type);
1049 #endif
1050 }
1051 }
1052
1053 return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
1054 cb, opaque, type);
1055 }
1056
1057 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
1058 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1059 BlockDriverCompletionFunc *cb, void *opaque)
1060 {
1061 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1062 cb, opaque, QEMU_AIO_READ);
1063 }
1064
1065 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
1066 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1067 BlockDriverCompletionFunc *cb, void *opaque)
1068 {
1069 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1070 cb, opaque, QEMU_AIO_WRITE);
1071 }
1072
1073 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
1074 BlockDriverCompletionFunc *cb, void *opaque)
1075 {
1076 BDRVRawState *s = bs->opaque;
1077
1078 if (fd_open(bs) < 0)
1079 return NULL;
1080
1081 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1082 }
1083
1084 static void raw_close(BlockDriverState *bs)
1085 {
1086 BDRVRawState *s = bs->opaque;
1087
1088 raw_detach_aio_context(bs);
1089
1090 #ifdef CONFIG_LINUX_AIO
1091 if (s->use_aio) {
1092 laio_cleanup(s->aio_ctx);
1093 }
1094 #endif
1095 if (s->fd >= 0) {
1096 qemu_close(s->fd);
1097 s->fd = -1;
1098 }
1099 }
1100
1101 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1102 {
1103 BDRVRawState *s = bs->opaque;
1104 struct stat st;
1105
1106 if (fstat(s->fd, &st)) {
1107 return -errno;
1108 }
1109
1110 if (S_ISREG(st.st_mode)) {
1111 if (ftruncate(s->fd, offset) < 0) {
1112 return -errno;
1113 }
1114 } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1115 if (offset > raw_getlength(bs)) {
1116 return -EINVAL;
1117 }
1118 } else {
1119 return -ENOTSUP;
1120 }
1121
1122 return 0;
1123 }
1124
1125 #ifdef __OpenBSD__
1126 static int64_t raw_getlength(BlockDriverState *bs)
1127 {
1128 BDRVRawState *s = bs->opaque;
1129 int fd = s->fd;
1130 struct stat st;
1131
1132 if (fstat(fd, &st))
1133 return -1;
1134 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1135 struct disklabel dl;
1136
1137 if (ioctl(fd, DIOCGDINFO, &dl))
1138 return -1;
1139 return (uint64_t)dl.d_secsize *
1140 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1141 } else
1142 return st.st_size;
1143 }
1144 #elif defined(__NetBSD__)
1145 static int64_t raw_getlength(BlockDriverState *bs)
1146 {
1147 BDRVRawState *s = bs->opaque;
1148 int fd = s->fd;
1149 struct stat st;
1150
1151 if (fstat(fd, &st))
1152 return -1;
1153 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1154 struct dkwedge_info dkw;
1155
1156 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1157 return dkw.dkw_size * 512;
1158 } else {
1159 struct disklabel dl;
1160
1161 if (ioctl(fd, DIOCGDINFO, &dl))
1162 return -1;
1163 return (uint64_t)dl.d_secsize *
1164 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1165 }
1166 } else
1167 return st.st_size;
1168 }
1169 #elif defined(__sun__)
1170 static int64_t raw_getlength(BlockDriverState *bs)
1171 {
1172 BDRVRawState *s = bs->opaque;
1173 struct dk_minfo minfo;
1174 int ret;
1175
1176 ret = fd_open(bs);
1177 if (ret < 0) {
1178 return ret;
1179 }
1180
1181 /*
1182 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1183 */
1184 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1185 if (ret != -1) {
1186 return minfo.dki_lbsize * minfo.dki_capacity;
1187 }
1188
1189 /*
1190 * There are reports that lseek on some devices fails, but
1191 * irc discussion said that contingency on contingency was overkill.
1192 */
1193 return lseek(s->fd, 0, SEEK_END);
1194 }
1195 #elif defined(CONFIG_BSD)
1196 static int64_t raw_getlength(BlockDriverState *bs)
1197 {
1198 BDRVRawState *s = bs->opaque;
1199 int fd = s->fd;
1200 int64_t size;
1201 struct stat sb;
1202 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1203 int reopened = 0;
1204 #endif
1205 int ret;
1206
1207 ret = fd_open(bs);
1208 if (ret < 0)
1209 return ret;
1210
1211 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1212 again:
1213 #endif
1214 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1215 #ifdef DIOCGMEDIASIZE
1216 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1217 #elif defined(DIOCGPART)
1218 {
1219 struct partinfo pi;
1220 if (ioctl(fd, DIOCGPART, &pi) == 0)
1221 size = pi.media_size;
1222 else
1223 size = 0;
1224 }
1225 if (size == 0)
1226 #endif
1227 #if defined(__APPLE__) && defined(__MACH__)
1228 size = LLONG_MAX;
1229 #else
1230 size = lseek(fd, 0LL, SEEK_END);
1231 #endif
1232 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1233 switch(s->type) {
1234 case FTYPE_CD:
1235 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1236 if (size == 2048LL * (unsigned)-1)
1237 size = 0;
1238 /* XXX no disc? maybe we need to reopen... */
1239 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1240 reopened = 1;
1241 goto again;
1242 }
1243 }
1244 #endif
1245 } else {
1246 size = lseek(fd, 0, SEEK_END);
1247 }
1248 return size;
1249 }
1250 #else
1251 static int64_t raw_getlength(BlockDriverState *bs)
1252 {
1253 BDRVRawState *s = bs->opaque;
1254 int ret;
1255
1256 ret = fd_open(bs);
1257 if (ret < 0) {
1258 return ret;
1259 }
1260
1261 return lseek(s->fd, 0, SEEK_END);
1262 }
1263 #endif
1264
1265 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1266 {
1267 struct stat st;
1268 BDRVRawState *s = bs->opaque;
1269
1270 if (fstat(s->fd, &st) < 0) {
1271 return -errno;
1272 }
1273 return (int64_t)st.st_blocks * 512;
1274 }
1275
1276 static int raw_create(const char *filename, QEMUOptionParameter *options,
1277 Error **errp)
1278 {
1279 int fd;
1280 int result = 0;
1281 int64_t total_size = 0;
1282
1283 strstart(filename, "file:", &filename);
1284
1285 /* Read out options */
1286 while (options && options->name) {
1287 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1288 total_size = options->value.n / BDRV_SECTOR_SIZE;
1289 }
1290 options++;
1291 }
1292
1293 fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
1294 0644);
1295 if (fd < 0) {
1296 result = -errno;
1297 error_setg_errno(errp, -result, "Could not create file");
1298 } else {
1299 if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
1300 result = -errno;
1301 error_setg_errno(errp, -result, "Could not resize file");
1302 }
1303 if (qemu_close(fd) != 0) {
1304 result = -errno;
1305 error_setg_errno(errp, -result, "Could not close the new file");
1306 }
1307 }
1308 return result;
1309 }
1310
1311 static int64_t try_fiemap(BlockDriverState *bs, off_t start, off_t *data,
1312 off_t *hole, int nb_sectors, int *pnum)
1313 {
1314 #ifdef CONFIG_FIEMAP
1315 BDRVRawState *s = bs->opaque;
1316 int64_t ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1317 struct {
1318 struct fiemap fm;
1319 struct fiemap_extent fe;
1320 } f;
1321
1322 if (s->skip_fiemap) {
1323 return -ENOTSUP;
1324 }
1325
1326 f.fm.fm_start = start;
1327 f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE;
1328 f.fm.fm_flags = 0;
1329 f.fm.fm_extent_count = 1;
1330 f.fm.fm_reserved = 0;
1331 if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) {
1332 s->skip_fiemap = true;
1333 return -errno;
1334 }
1335
1336 if (f.fm.fm_mapped_extents == 0) {
1337 /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1338 * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1339 */
1340 off_t length = lseek(s->fd, 0, SEEK_END);
1341 *hole = f.fm.fm_start;
1342 *data = MIN(f.fm.fm_start + f.fm.fm_length, length);
1343 } else {
1344 *data = f.fe.fe_logical;
1345 *hole = f.fe.fe_logical + f.fe.fe_length;
1346 if (f.fe.fe_flags & FIEMAP_EXTENT_UNWRITTEN) {
1347 ret |= BDRV_BLOCK_ZERO;
1348 }
1349 }
1350
1351 return ret;
1352 #else
1353 return -ENOTSUP;
1354 #endif
1355 }
1356
1357 static int64_t try_seek_hole(BlockDriverState *bs, off_t start, off_t *data,
1358 off_t *hole, int *pnum)
1359 {
1360 #if defined SEEK_HOLE && defined SEEK_DATA
1361 BDRVRawState *s = bs->opaque;
1362
1363 *hole = lseek(s->fd, start, SEEK_HOLE);
1364 if (*hole == -1) {
1365 /* -ENXIO indicates that sector_num was past the end of the file.
1366 * There is a virtual hole there. */
1367 assert(errno != -ENXIO);
1368
1369 return -errno;
1370 }
1371
1372 if (*hole > start) {
1373 *data = start;
1374 } else {
1375 /* On a hole. We need another syscall to find its end. */
1376 *data = lseek(s->fd, start, SEEK_DATA);
1377 if (*data == -1) {
1378 *data = lseek(s->fd, 0, SEEK_END);
1379 }
1380 }
1381
1382 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1383 #else
1384 return -ENOTSUP;
1385 #endif
1386 }
1387
1388 /*
1389 * Returns true iff the specified sector is present in the disk image. Drivers
1390 * not implementing the functionality are assumed to not support backing files,
1391 * hence all their sectors are reported as allocated.
1392 *
1393 * If 'sector_num' is beyond the end of the disk image the return value is 0
1394 * and 'pnum' is set to 0.
1395 *
1396 * 'pnum' is set to the number of sectors (including and immediately following
1397 * the specified sector) that are known to be in the same
1398 * allocated/unallocated state.
1399 *
1400 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1401 * beyond the end of the disk image it will be clamped.
1402 */
1403 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1404 int64_t sector_num,
1405 int nb_sectors, int *pnum)
1406 {
1407 off_t start, data = 0, hole = 0;
1408 int64_t ret;
1409
1410 ret = fd_open(bs);
1411 if (ret < 0) {
1412 return ret;
1413 }
1414
1415 start = sector_num * BDRV_SECTOR_SIZE;
1416
1417 ret = try_fiemap(bs, start, &data, &hole, nb_sectors, pnum);
1418 if (ret < 0) {
1419 ret = try_seek_hole(bs, start, &data, &hole, pnum);
1420 if (ret < 0) {
1421 /* Assume everything is allocated. */
1422 data = 0;
1423 hole = start + nb_sectors * BDRV_SECTOR_SIZE;
1424 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1425 }
1426 }
1427
1428 if (data <= start) {
1429 /* On a data extent, compute sectors to the end of the extent. */
1430 *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1431 } else {
1432 /* On a hole, compute sectors to the beginning of the next extent. */
1433 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1434 ret &= ~BDRV_BLOCK_DATA;
1435 ret |= BDRV_BLOCK_ZERO;
1436 }
1437
1438 return ret;
1439 }
1440
1441 static coroutine_fn BlockDriverAIOCB *raw_aio_discard(BlockDriverState *bs,
1442 int64_t sector_num, int nb_sectors,
1443 BlockDriverCompletionFunc *cb, void *opaque)
1444 {
1445 BDRVRawState *s = bs->opaque;
1446
1447 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1448 cb, opaque, QEMU_AIO_DISCARD);
1449 }
1450
1451 static int coroutine_fn raw_co_write_zeroes(
1452 BlockDriverState *bs, int64_t sector_num,
1453 int nb_sectors, BdrvRequestFlags flags)
1454 {
1455 BDRVRawState *s = bs->opaque;
1456
1457 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1458 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1459 QEMU_AIO_WRITE_ZEROES);
1460 } else if (s->discard_zeroes) {
1461 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1462 QEMU_AIO_DISCARD);
1463 }
1464 return -ENOTSUP;
1465 }
1466
1467 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1468 {
1469 BDRVRawState *s = bs->opaque;
1470
1471 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1472 bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1473 return 0;
1474 }
1475
1476 static QEMUOptionParameter raw_create_options[] = {
1477 {
1478 .name = BLOCK_OPT_SIZE,
1479 .type = OPT_SIZE,
1480 .help = "Virtual disk size"
1481 },
1482 { NULL }
1483 };
1484
1485 static BlockDriver bdrv_file = {
1486 .format_name = "file",
1487 .protocol_name = "file",
1488 .instance_size = sizeof(BDRVRawState),
1489 .bdrv_needs_filename = true,
1490 .bdrv_probe = NULL, /* no probe for protocols */
1491 .bdrv_parse_filename = raw_parse_filename,
1492 .bdrv_file_open = raw_open,
1493 .bdrv_reopen_prepare = raw_reopen_prepare,
1494 .bdrv_reopen_commit = raw_reopen_commit,
1495 .bdrv_reopen_abort = raw_reopen_abort,
1496 .bdrv_close = raw_close,
1497 .bdrv_create = raw_create,
1498 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1499 .bdrv_co_get_block_status = raw_co_get_block_status,
1500 .bdrv_co_write_zeroes = raw_co_write_zeroes,
1501
1502 .bdrv_aio_readv = raw_aio_readv,
1503 .bdrv_aio_writev = raw_aio_writev,
1504 .bdrv_aio_flush = raw_aio_flush,
1505 .bdrv_aio_discard = raw_aio_discard,
1506 .bdrv_refresh_limits = raw_refresh_limits,
1507
1508 .bdrv_truncate = raw_truncate,
1509 .bdrv_getlength = raw_getlength,
1510 .bdrv_get_info = raw_get_info,
1511 .bdrv_get_allocated_file_size
1512 = raw_get_allocated_file_size,
1513
1514 .bdrv_detach_aio_context = raw_detach_aio_context,
1515 .bdrv_attach_aio_context = raw_attach_aio_context,
1516
1517 .create_options = raw_create_options,
1518 };
1519
1520 /***********************************************/
1521 /* host device */
1522
1523 #if defined(__APPLE__) && defined(__MACH__)
1524 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1525 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1526
1527 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1528 {
1529 kern_return_t kernResult;
1530 mach_port_t masterPort;
1531 CFMutableDictionaryRef classesToMatch;
1532
1533 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1534 if ( KERN_SUCCESS != kernResult ) {
1535 printf( "IOMasterPort returned %d\n", kernResult );
1536 }
1537
1538 classesToMatch = IOServiceMatching( kIOCDMediaClass );
1539 if ( classesToMatch == NULL ) {
1540 printf( "IOServiceMatching returned a NULL dictionary.\n" );
1541 } else {
1542 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1543 }
1544 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1545 if ( KERN_SUCCESS != kernResult )
1546 {
1547 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1548 }
1549
1550 return kernResult;
1551 }
1552
1553 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
1554 {
1555 io_object_t nextMedia;
1556 kern_return_t kernResult = KERN_FAILURE;
1557 *bsdPath = '\0';
1558 nextMedia = IOIteratorNext( mediaIterator );
1559 if ( nextMedia )
1560 {
1561 CFTypeRef bsdPathAsCFString;
1562 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1563 if ( bsdPathAsCFString ) {
1564 size_t devPathLength;
1565 strcpy( bsdPath, _PATH_DEV );
1566 strcat( bsdPath, "r" );
1567 devPathLength = strlen( bsdPath );
1568 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1569 kernResult = KERN_SUCCESS;
1570 }
1571 CFRelease( bsdPathAsCFString );
1572 }
1573 IOObjectRelease( nextMedia );
1574 }
1575
1576 return kernResult;
1577 }
1578
1579 #endif
1580
1581 static int hdev_probe_device(const char *filename)
1582 {
1583 struct stat st;
1584
1585 /* allow a dedicated CD-ROM driver to match with a higher priority */
1586 if (strstart(filename, "/dev/cdrom", NULL))
1587 return 50;
1588
1589 if (stat(filename, &st) >= 0 &&
1590 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
1591 return 100;
1592 }
1593
1594 return 0;
1595 }
1596
1597 static int check_hdev_writable(BDRVRawState *s)
1598 {
1599 #if defined(BLKROGET)
1600 /* Linux block devices can be configured "read-only" using blockdev(8).
1601 * This is independent of device node permissions and therefore open(2)
1602 * with O_RDWR succeeds. Actual writes fail with EPERM.
1603 *
1604 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
1605 * check for read-only block devices so that Linux block devices behave
1606 * properly.
1607 */
1608 struct stat st;
1609 int readonly = 0;
1610
1611 if (fstat(s->fd, &st)) {
1612 return -errno;
1613 }
1614
1615 if (!S_ISBLK(st.st_mode)) {
1616 return 0;
1617 }
1618
1619 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
1620 return -errno;
1621 }
1622
1623 if (readonly) {
1624 return -EACCES;
1625 }
1626 #endif /* defined(BLKROGET) */
1627 return 0;
1628 }
1629
1630 static void hdev_parse_filename(const char *filename, QDict *options,
1631 Error **errp)
1632 {
1633 /* The prefix is optional, just as for "file". */
1634 strstart(filename, "host_device:", &filename);
1635
1636 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
1637 }
1638
1639 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
1640 Error **errp)
1641 {
1642 BDRVRawState *s = bs->opaque;
1643 Error *local_err = NULL;
1644 int ret;
1645 const char *filename = qdict_get_str(options, "filename");
1646
1647 #if defined(__APPLE__) && defined(__MACH__)
1648 if (strstart(filename, "/dev/cdrom", NULL)) {
1649 kern_return_t kernResult;
1650 io_iterator_t mediaIterator;
1651 char bsdPath[ MAXPATHLEN ];
1652 int fd;
1653
1654 kernResult = FindEjectableCDMedia( &mediaIterator );
1655 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1656
1657 if ( bsdPath[ 0 ] != '\0' ) {
1658 strcat(bsdPath,"s0");
1659 /* some CDs don't have a partition 0 */
1660 fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1661 if (fd < 0) {
1662 bsdPath[strlen(bsdPath)-1] = '1';
1663 } else {
1664 qemu_close(fd);
1665 }
1666 filename = bsdPath;
1667 qdict_put(options, "filename", qstring_from_str(filename));
1668 }
1669
1670 if ( mediaIterator )
1671 IOObjectRelease( mediaIterator );
1672 }
1673 #endif
1674
1675 s->type = FTYPE_FILE;
1676 #if defined(__linux__)
1677 {
1678 char resolved_path[ MAXPATHLEN ], *temp;
1679
1680 temp = realpath(filename, resolved_path);
1681 if (temp && strstart(temp, "/dev/sg", NULL)) {
1682 bs->sg = 1;
1683 }
1684 }
1685 #endif
1686
1687 ret = raw_open_common(bs, options, flags, 0, &local_err);
1688 if (ret < 0) {
1689 if (local_err) {
1690 error_propagate(errp, local_err);
1691 }
1692 return ret;
1693 }
1694
1695 if (flags & BDRV_O_RDWR) {
1696 ret = check_hdev_writable(s);
1697 if (ret < 0) {
1698 raw_close(bs);
1699 error_setg_errno(errp, -ret, "The device is not writable");
1700 return ret;
1701 }
1702 }
1703
1704 return ret;
1705 }
1706
1707 #if defined(__linux__)
1708 /* Note: we do not have a reliable method to detect if the floppy is
1709 present. The current method is to try to open the floppy at every
1710 I/O and to keep it opened during a few hundreds of ms. */
1711 static int fd_open(BlockDriverState *bs)
1712 {
1713 BDRVRawState *s = bs->opaque;
1714 int last_media_present;
1715
1716 if (s->type != FTYPE_FD)
1717 return 0;
1718 last_media_present = (s->fd >= 0);
1719 if (s->fd >= 0 &&
1720 (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1721 qemu_close(s->fd);
1722 s->fd = -1;
1723 #ifdef DEBUG_FLOPPY
1724 printf("Floppy closed\n");
1725 #endif
1726 }
1727 if (s->fd < 0) {
1728 if (s->fd_got_error &&
1729 (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1730 #ifdef DEBUG_FLOPPY
1731 printf("No floppy (open delayed)\n");
1732 #endif
1733 return -EIO;
1734 }
1735 s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
1736 if (s->fd < 0) {
1737 s->fd_error_time = get_clock();
1738 s->fd_got_error = 1;
1739 if (last_media_present)
1740 s->fd_media_changed = 1;
1741 #ifdef DEBUG_FLOPPY
1742 printf("No floppy\n");
1743 #endif
1744 return -EIO;
1745 }
1746 #ifdef DEBUG_FLOPPY
1747 printf("Floppy opened\n");
1748 #endif
1749 }
1750 if (!last_media_present)
1751 s->fd_media_changed = 1;
1752 s->fd_open_time = get_clock();
1753 s->fd_got_error = 0;
1754 return 0;
1755 }
1756
1757 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1758 {
1759 BDRVRawState *s = bs->opaque;
1760
1761 return ioctl(s->fd, req, buf);
1762 }
1763
1764 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1765 unsigned long int req, void *buf,
1766 BlockDriverCompletionFunc *cb, void *opaque)
1767 {
1768 BDRVRawState *s = bs->opaque;
1769 RawPosixAIOData *acb;
1770 ThreadPool *pool;
1771
1772 if (fd_open(bs) < 0)
1773 return NULL;
1774
1775 acb = g_slice_new(RawPosixAIOData);
1776 acb->bs = bs;
1777 acb->aio_type = QEMU_AIO_IOCTL;
1778 acb->aio_fildes = s->fd;
1779 acb->aio_offset = 0;
1780 acb->aio_ioctl_buf = buf;
1781 acb->aio_ioctl_cmd = req;
1782 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1783 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1784 }
1785
1786 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1787 static int fd_open(BlockDriverState *bs)
1788 {
1789 BDRVRawState *s = bs->opaque;
1790
1791 /* this is just to ensure s->fd is sane (its called by io ops) */
1792 if (s->fd >= 0)
1793 return 0;
1794 return -EIO;
1795 }
1796 #else /* !linux && !FreeBSD */
1797
1798 static int fd_open(BlockDriverState *bs)
1799 {
1800 return 0;
1801 }
1802
1803 #endif /* !linux && !FreeBSD */
1804
1805 static coroutine_fn BlockDriverAIOCB *hdev_aio_discard(BlockDriverState *bs,
1806 int64_t sector_num, int nb_sectors,
1807 BlockDriverCompletionFunc *cb, void *opaque)
1808 {
1809 BDRVRawState *s = bs->opaque;
1810
1811 if (fd_open(bs) < 0) {
1812 return NULL;
1813 }
1814 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1815 cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1816 }
1817
1818 static coroutine_fn int hdev_co_write_zeroes(BlockDriverState *bs,
1819 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
1820 {
1821 BDRVRawState *s = bs->opaque;
1822 int rc;
1823
1824 rc = fd_open(bs);
1825 if (rc < 0) {
1826 return rc;
1827 }
1828 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1829 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1830 QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
1831 } else if (s->discard_zeroes) {
1832 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1833 QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1834 }
1835 return -ENOTSUP;
1836 }
1837
1838 static int hdev_create(const char *filename, QEMUOptionParameter *options,
1839 Error **errp)
1840 {
1841 int fd;
1842 int ret = 0;
1843 struct stat stat_buf;
1844 int64_t total_size = 0;
1845 bool has_prefix;
1846
1847 /* This function is used by all three protocol block drivers and therefore
1848 * any of these three prefixes may be given.
1849 * The return value has to be stored somewhere, otherwise this is an error
1850 * due to -Werror=unused-value. */
1851 has_prefix =
1852 strstart(filename, "host_device:", &filename) ||
1853 strstart(filename, "host_cdrom:" , &filename) ||
1854 strstart(filename, "host_floppy:", &filename);
1855
1856 (void)has_prefix;
1857
1858 /* Read out options */
1859 while (options && options->name) {
1860 if (!strcmp(options->name, "size")) {
1861 total_size = options->value.n / BDRV_SECTOR_SIZE;
1862 }
1863 options++;
1864 }
1865
1866 fd = qemu_open(filename, O_WRONLY | O_BINARY);
1867 if (fd < 0) {
1868 ret = -errno;
1869 error_setg_errno(errp, -ret, "Could not open device");
1870 return ret;
1871 }
1872
1873 if (fstat(fd, &stat_buf) < 0) {
1874 ret = -errno;
1875 error_setg_errno(errp, -ret, "Could not stat device");
1876 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
1877 error_setg(errp,
1878 "The given file is neither a block nor a character device");
1879 ret = -ENODEV;
1880 } else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE) {
1881 error_setg(errp, "Device is too small");
1882 ret = -ENOSPC;
1883 }
1884
1885 qemu_close(fd);
1886 return ret;
1887 }
1888
1889 static BlockDriver bdrv_host_device = {
1890 .format_name = "host_device",
1891 .protocol_name = "host_device",
1892 .instance_size = sizeof(BDRVRawState),
1893 .bdrv_needs_filename = true,
1894 .bdrv_probe_device = hdev_probe_device,
1895 .bdrv_parse_filename = hdev_parse_filename,
1896 .bdrv_file_open = hdev_open,
1897 .bdrv_close = raw_close,
1898 .bdrv_reopen_prepare = raw_reopen_prepare,
1899 .bdrv_reopen_commit = raw_reopen_commit,
1900 .bdrv_reopen_abort = raw_reopen_abort,
1901 .bdrv_create = hdev_create,
1902 .create_options = raw_create_options,
1903 .bdrv_co_write_zeroes = hdev_co_write_zeroes,
1904
1905 .bdrv_aio_readv = raw_aio_readv,
1906 .bdrv_aio_writev = raw_aio_writev,
1907 .bdrv_aio_flush = raw_aio_flush,
1908 .bdrv_aio_discard = hdev_aio_discard,
1909 .bdrv_refresh_limits = raw_refresh_limits,
1910
1911 .bdrv_truncate = raw_truncate,
1912 .bdrv_getlength = raw_getlength,
1913 .bdrv_get_info = raw_get_info,
1914 .bdrv_get_allocated_file_size
1915 = raw_get_allocated_file_size,
1916
1917 .bdrv_detach_aio_context = raw_detach_aio_context,
1918 .bdrv_attach_aio_context = raw_attach_aio_context,
1919
1920 /* generic scsi device */
1921 #ifdef __linux__
1922 .bdrv_ioctl = hdev_ioctl,
1923 .bdrv_aio_ioctl = hdev_aio_ioctl,
1924 #endif
1925 };
1926
1927 #ifdef __linux__
1928 static void floppy_parse_filename(const char *filename, QDict *options,
1929 Error **errp)
1930 {
1931 /* The prefix is optional, just as for "file". */
1932 strstart(filename, "host_floppy:", &filename);
1933
1934 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
1935 }
1936
1937 static int floppy_open(BlockDriverState *bs, QDict *options, int flags,
1938 Error **errp)
1939 {
1940 BDRVRawState *s = bs->opaque;
1941 Error *local_err = NULL;
1942 int ret;
1943
1944 s->type = FTYPE_FD;
1945
1946 /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1947 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
1948 if (ret) {
1949 if (local_err) {
1950 error_propagate(errp, local_err);
1951 }
1952 return ret;
1953 }
1954
1955 /* close fd so that we can reopen it as needed */
1956 qemu_close(s->fd);
1957 s->fd = -1;
1958 s->fd_media_changed = 1;
1959
1960 return 0;
1961 }
1962
1963 static int floppy_probe_device(const char *filename)
1964 {
1965 int fd, ret;
1966 int prio = 0;
1967 struct floppy_struct fdparam;
1968 struct stat st;
1969
1970 if (strstart(filename, "/dev/fd", NULL) &&
1971 !strstart(filename, "/dev/fdset/", NULL)) {
1972 prio = 50;
1973 }
1974
1975 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1976 if (fd < 0) {
1977 goto out;
1978 }
1979 ret = fstat(fd, &st);
1980 if (ret == -1 || !S_ISBLK(st.st_mode)) {
1981 goto outc;
1982 }
1983
1984 /* Attempt to detect via a floppy specific ioctl */
1985 ret = ioctl(fd, FDGETPRM, &fdparam);
1986 if (ret >= 0)
1987 prio = 100;
1988
1989 outc:
1990 qemu_close(fd);
1991 out:
1992 return prio;
1993 }
1994
1995
1996 static int floppy_is_inserted(BlockDriverState *bs)
1997 {
1998 return fd_open(bs) >= 0;
1999 }
2000
2001 static int floppy_media_changed(BlockDriverState *bs)
2002 {
2003 BDRVRawState *s = bs->opaque;
2004 int ret;
2005
2006 /*
2007 * XXX: we do not have a true media changed indication.
2008 * It does not work if the floppy is changed without trying to read it.
2009 */
2010 fd_open(bs);
2011 ret = s->fd_media_changed;
2012 s->fd_media_changed = 0;
2013 #ifdef DEBUG_FLOPPY
2014 printf("Floppy changed=%d\n", ret);
2015 #endif
2016 return ret;
2017 }
2018
2019 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
2020 {
2021 BDRVRawState *s = bs->opaque;
2022 int fd;
2023
2024 if (s->fd >= 0) {
2025 qemu_close(s->fd);
2026 s->fd = -1;
2027 }
2028 fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
2029 if (fd >= 0) {
2030 if (ioctl(fd, FDEJECT, 0) < 0)
2031 perror("FDEJECT");
2032 qemu_close(fd);
2033 }
2034 }
2035
2036 static BlockDriver bdrv_host_floppy = {
2037 .format_name = "host_floppy",
2038 .protocol_name = "host_floppy",
2039 .instance_size = sizeof(BDRVRawState),
2040 .bdrv_needs_filename = true,
2041 .bdrv_probe_device = floppy_probe_device,
2042 .bdrv_parse_filename = floppy_parse_filename,
2043 .bdrv_file_open = floppy_open,
2044 .bdrv_close = raw_close,
2045 .bdrv_reopen_prepare = raw_reopen_prepare,
2046 .bdrv_reopen_commit = raw_reopen_commit,
2047 .bdrv_reopen_abort = raw_reopen_abort,
2048 .bdrv_create = hdev_create,
2049 .create_options = raw_create_options,
2050
2051 .bdrv_aio_readv = raw_aio_readv,
2052 .bdrv_aio_writev = raw_aio_writev,
2053 .bdrv_aio_flush = raw_aio_flush,
2054 .bdrv_refresh_limits = raw_refresh_limits,
2055
2056 .bdrv_truncate = raw_truncate,
2057 .bdrv_getlength = raw_getlength,
2058 .has_variable_length = true,
2059 .bdrv_get_allocated_file_size
2060 = raw_get_allocated_file_size,
2061
2062 .bdrv_detach_aio_context = raw_detach_aio_context,
2063 .bdrv_attach_aio_context = raw_attach_aio_context,
2064
2065 /* removable device support */
2066 .bdrv_is_inserted = floppy_is_inserted,
2067 .bdrv_media_changed = floppy_media_changed,
2068 .bdrv_eject = floppy_eject,
2069 };
2070 #endif
2071
2072 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2073 static void cdrom_parse_filename(const char *filename, QDict *options,
2074 Error **errp)
2075 {
2076 /* The prefix is optional, just as for "file". */
2077 strstart(filename, "host_cdrom:", &filename);
2078
2079 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2080 }
2081 #endif
2082
2083 #ifdef __linux__
2084 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2085 Error **errp)
2086 {
2087 BDRVRawState *s = bs->opaque;
2088 Error *local_err = NULL;
2089 int ret;
2090
2091 s->type = FTYPE_CD;
2092
2093 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2094 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2095 if (local_err) {
2096 error_propagate(errp, local_err);
2097 }
2098 return ret;
2099 }
2100
2101 static int cdrom_probe_device(const char *filename)
2102 {
2103 int fd, ret;
2104 int prio = 0;
2105 struct stat st;
2106
2107 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2108 if (fd < 0) {
2109 goto out;
2110 }
2111 ret = fstat(fd, &st);
2112 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2113 goto outc;
2114 }
2115
2116 /* Attempt to detect via a CDROM specific ioctl */
2117 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2118 if (ret >= 0)
2119 prio = 100;
2120
2121 outc:
2122 qemu_close(fd);
2123 out:
2124 return prio;
2125 }
2126
2127 static int cdrom_is_inserted(BlockDriverState *bs)
2128 {
2129 BDRVRawState *s = bs->opaque;
2130 int ret;
2131
2132 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2133 if (ret == CDS_DISC_OK)
2134 return 1;
2135 return 0;
2136 }
2137
2138 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2139 {
2140 BDRVRawState *s = bs->opaque;
2141
2142 if (eject_flag) {
2143 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2144 perror("CDROMEJECT");
2145 } else {
2146 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2147 perror("CDROMEJECT");
2148 }
2149 }
2150
2151 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2152 {
2153 BDRVRawState *s = bs->opaque;
2154
2155 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2156 /*
2157 * Note: an error can happen if the distribution automatically
2158 * mounts the CD-ROM
2159 */
2160 /* perror("CDROM_LOCKDOOR"); */
2161 }
2162 }
2163
2164 static BlockDriver bdrv_host_cdrom = {
2165 .format_name = "host_cdrom",
2166 .protocol_name = "host_cdrom",
2167 .instance_size = sizeof(BDRVRawState),
2168 .bdrv_needs_filename = true,
2169 .bdrv_probe_device = cdrom_probe_device,
2170 .bdrv_parse_filename = cdrom_parse_filename,
2171 .bdrv_file_open = cdrom_open,
2172 .bdrv_close = raw_close,
2173 .bdrv_reopen_prepare = raw_reopen_prepare,
2174 .bdrv_reopen_commit = raw_reopen_commit,
2175 .bdrv_reopen_abort = raw_reopen_abort,
2176 .bdrv_create = hdev_create,
2177 .create_options = raw_create_options,
2178
2179 .bdrv_aio_readv = raw_aio_readv,
2180 .bdrv_aio_writev = raw_aio_writev,
2181 .bdrv_aio_flush = raw_aio_flush,
2182 .bdrv_refresh_limits = raw_refresh_limits,
2183
2184 .bdrv_truncate = raw_truncate,
2185 .bdrv_getlength = raw_getlength,
2186 .has_variable_length = true,
2187 .bdrv_get_allocated_file_size
2188 = raw_get_allocated_file_size,
2189
2190 .bdrv_detach_aio_context = raw_detach_aio_context,
2191 .bdrv_attach_aio_context = raw_attach_aio_context,
2192
2193 /* removable device support */
2194 .bdrv_is_inserted = cdrom_is_inserted,
2195 .bdrv_eject = cdrom_eject,
2196 .bdrv_lock_medium = cdrom_lock_medium,
2197
2198 /* generic scsi device */
2199 .bdrv_ioctl = hdev_ioctl,
2200 .bdrv_aio_ioctl = hdev_aio_ioctl,
2201 };
2202 #endif /* __linux__ */
2203
2204 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2205 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2206 Error **errp)
2207 {
2208 BDRVRawState *s = bs->opaque;
2209 Error *local_err = NULL;
2210 int ret;
2211
2212 s->type = FTYPE_CD;
2213
2214 ret = raw_open_common(bs, options, flags, 0, &local_err);
2215 if (ret) {
2216 if (local_err) {
2217 error_propagate(errp, local_err);
2218 }
2219 return ret;
2220 }
2221
2222 /* make sure the door isn't locked at this time */
2223 ioctl(s->fd, CDIOCALLOW);
2224 return 0;
2225 }
2226
2227 static int cdrom_probe_device(const char *filename)
2228 {
2229 if (strstart(filename, "/dev/cd", NULL) ||
2230 strstart(filename, "/dev/acd", NULL))
2231 return 100;
2232 return 0;
2233 }
2234
2235 static int cdrom_reopen(BlockDriverState *bs)
2236 {
2237 BDRVRawState *s = bs->opaque;
2238 int fd;
2239
2240 /*
2241 * Force reread of possibly changed/newly loaded disc,
2242 * FreeBSD seems to not notice sometimes...
2243 */
2244 if (s->fd >= 0)
2245 qemu_close(s->fd);
2246 fd = qemu_open(bs->filename, s->open_flags, 0644);
2247 if (fd < 0) {
2248 s->fd = -1;
2249 return -EIO;
2250 }
2251 s->fd = fd;
2252
2253 /* make sure the door isn't locked at this time */
2254 ioctl(s->fd, CDIOCALLOW);
2255 return 0;
2256 }
2257
2258 static int cdrom_is_inserted(BlockDriverState *bs)
2259 {
2260 return raw_getlength(bs) > 0;
2261 }
2262
2263 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2264 {
2265 BDRVRawState *s = bs->opaque;
2266
2267 if (s->fd < 0)
2268 return;
2269
2270 (void) ioctl(s->fd, CDIOCALLOW);
2271
2272 if (eject_flag) {
2273 if (ioctl(s->fd, CDIOCEJECT) < 0)
2274 perror("CDIOCEJECT");
2275 } else {
2276 if (ioctl(s->fd, CDIOCCLOSE) < 0)
2277 perror("CDIOCCLOSE");
2278 }
2279
2280 cdrom_reopen(bs);
2281 }
2282
2283 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2284 {
2285 BDRVRawState *s = bs->opaque;
2286
2287 if (s->fd < 0)
2288 return;
2289 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2290 /*
2291 * Note: an error can happen if the distribution automatically
2292 * mounts the CD-ROM
2293 */
2294 /* perror("CDROM_LOCKDOOR"); */
2295 }
2296 }
2297
2298 static BlockDriver bdrv_host_cdrom = {
2299 .format_name = "host_cdrom",
2300 .protocol_name = "host_cdrom",
2301 .instance_size = sizeof(BDRVRawState),
2302 .bdrv_needs_filename = true,
2303 .bdrv_probe_device = cdrom_probe_device,
2304 .bdrv_parse_filename = cdrom_parse_filename,
2305 .bdrv_file_open = cdrom_open,
2306 .bdrv_close = raw_close,
2307 .bdrv_reopen_prepare = raw_reopen_prepare,
2308 .bdrv_reopen_commit = raw_reopen_commit,
2309 .bdrv_reopen_abort = raw_reopen_abort,
2310 .bdrv_create = hdev_create,
2311 .create_options = raw_create_options,
2312
2313 .bdrv_aio_readv = raw_aio_readv,
2314 .bdrv_aio_writev = raw_aio_writev,
2315 .bdrv_aio_flush = raw_aio_flush,
2316 .bdrv_refresh_limits = raw_refresh_limits,
2317
2318 .bdrv_truncate = raw_truncate,
2319 .bdrv_getlength = raw_getlength,
2320 .has_variable_length = true,
2321 .bdrv_get_allocated_file_size
2322 = raw_get_allocated_file_size,
2323
2324 .bdrv_detach_aio_context = raw_detach_aio_context,
2325 .bdrv_attach_aio_context = raw_attach_aio_context,
2326
2327 /* removable device support */
2328 .bdrv_is_inserted = cdrom_is_inserted,
2329 .bdrv_eject = cdrom_eject,
2330 .bdrv_lock_medium = cdrom_lock_medium,
2331 };
2332 #endif /* __FreeBSD__ */
2333
2334 static void bdrv_file_init(void)
2335 {
2336 /*
2337 * Register all the drivers. Note that order is important, the driver
2338 * registered last will get probed first.
2339 */
2340 bdrv_register(&bdrv_file);
2341 bdrv_register(&bdrv_host_device);
2342 #ifdef __linux__
2343 bdrv_register(&bdrv_host_floppy);
2344 bdrv_register(&bdrv_host_cdrom);
2345 #endif
2346 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2347 bdrv_register(&bdrv_host_cdrom);
2348 #endif
2349 }
2350
2351 block_init(bdrv_file_init);
Random patches