?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
generic-loader: Add a generic loader
[qemu/kevin.git] / block / raw-posix.c
blob6ed7547392704afbce2420f18a506792003db5da
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/osdep.h"
25 #include "qapi/error.h"
26 #include "qemu/cutils.h"
27 #include "qemu/error-report.h"
28 #include "qemu/timer.h"
29 #include "qemu/log.h"
30 #include "block/block_int.h"
31 #include "qemu/module.h"
32 #include "trace.h"
33 #include "block/thread-pool.h"
34 #include "qemu/iov.h"
35 #include "block/raw-aio.h"
36 #include "qapi/util.h"
37 #include "qapi/qmp/qstring.h"
38
39 #if defined(__APPLE__) && (__MACH__)
40 #include <paths.h>
41 #include <sys/param.h>
42 #include <IOKit/IOKitLib.h>
43 #include <IOKit/IOBSD.h>
44 #include <IOKit/storage/IOMediaBSDClient.h>
45 #include <IOKit/storage/IOMedia.h>
46 #include <IOKit/storage/IOCDMedia.h>
47 //#include <IOKit/storage/IOCDTypes.h>
48 #include <IOKit/storage/IODVDMedia.h>
49 #include <CoreFoundation/CoreFoundation.h>
50 #endif
51
52 #ifdef __sun__
53 #define _POSIX_PTHREAD_SEMANTICS 1
54 #include <sys/dkio.h>
55 #endif
56 #ifdef __linux__
57 #include <sys/ioctl.h>
58 #include <sys/param.h>
59 #include <linux/cdrom.h>
60 #include <linux/fd.h>
61 #include <linux/fs.h>
62 #include <linux/hdreg.h>
63 #include <scsi/sg.h>
64 #ifdef __s390__
65 #include <asm/dasd.h>
66 #endif
67 #ifndef FS_NOCOW_FL
68 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
69 #endif
70 #endif
71 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
72 #include <linux/falloc.h>
73 #endif
74 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
75 #include <sys/disk.h>
76 #include <sys/cdio.h>
77 #endif
78
79 #ifdef __OpenBSD__
80 #include <sys/ioctl.h>
81 #include <sys/disklabel.h>
82 #include <sys/dkio.h>
83 #endif
84
85 #ifdef __NetBSD__
86 #include <sys/ioctl.h>
87 #include <sys/disklabel.h>
88 #include <sys/dkio.h>
89 #include <sys/disk.h>
90 #endif
91
92 #ifdef __DragonFly__
93 #include <sys/ioctl.h>
94 #include <sys/diskslice.h>
95 #endif
96
97 #ifdef CONFIG_XFS
98 #include <xfs/xfs.h>
99 #endif
100
101 //#define DEBUG_BLOCK
102
103 #ifdef DEBUG_BLOCK
104 # define DEBUG_BLOCK_PRINT 1
105 #else
106 # define DEBUG_BLOCK_PRINT 0
107 #endif
108 #define DPRINTF(fmt, ...) \
109 do { \
110 if (DEBUG_BLOCK_PRINT) { \
111 printf(fmt, ## __VA_ARGS__); \
112 } \
113 } while (0)
114
115 /* OS X does not have O_DSYNC */
116 #ifndef O_DSYNC
117 #ifdef O_SYNC
118 #define O_DSYNC O_SYNC
119 #elif defined(O_FSYNC)
120 #define O_DSYNC O_FSYNC
121 #endif
122 #endif
123
124 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
125 #ifndef O_DIRECT
126 #define O_DIRECT O_DSYNC
127 #endif
128
129 #define FTYPE_FILE 0
130 #define FTYPE_CD 1
131
132 #define MAX_BLOCKSIZE 4096
133
134 typedef struct BDRVRawState {
135 int fd;
136 int type;
137 int open_flags;
138 size_t buf_align;
139
140 #ifdef CONFIG_XFS
141 bool is_xfs:1;
142 #endif
143 bool has_discard:1;
144 bool has_write_zeroes:1;
145 bool discard_zeroes:1;
146 bool has_fallocate;
147 bool needs_alignment;
148 } BDRVRawState;
149
150 typedef struct BDRVRawReopenState {
151 int fd;
152 int open_flags;
153 } BDRVRawReopenState;
154
155 static int fd_open(BlockDriverState *bs);
156 static int64_t raw_getlength(BlockDriverState *bs);
157
158 typedef struct RawPosixAIOData {
159 BlockDriverState *bs;
160 int aio_fildes;
161 union {
162 struct iovec *aio_iov;
163 void *aio_ioctl_buf;
164 };
165 int aio_niov;
166 uint64_t aio_nbytes;
167 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
168 off_t aio_offset;
169 int aio_type;
170 } RawPosixAIOData;
171
172 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
173 static int cdrom_reopen(BlockDriverState *bs);
174 #endif
175
176 #if defined(__NetBSD__)
177 static int raw_normalize_devicepath(const char **filename)
178 {
179 static char namebuf[PATH_MAX];
180 const char *dp, *fname;
181 struct stat sb;
182
183 fname = *filename;
184 dp = strrchr(fname, '/');
185 if (lstat(fname, &sb) < 0) {
186 fprintf(stderr, "%s: stat failed: %s\n",
187 fname, strerror(errno));
188 return -errno;
189 }
190
191 if (!S_ISBLK(sb.st_mode)) {
192 return 0;
193 }
194
195 if (dp == NULL) {
196 snprintf(namebuf, PATH_MAX, "r%s", fname);
197 } else {
198 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
199 (int)(dp - fname), fname, dp + 1);
200 }
201 fprintf(stderr, "%s is a block device", fname);
202 *filename = namebuf;
203 fprintf(stderr, ", using %s\n", *filename);
204
205 return 0;
206 }
207 #else
208 static int raw_normalize_devicepath(const char **filename)
209 {
210 return 0;
211 }
212 #endif
213
214 /*
215 * Get logical block size via ioctl. On success store it in @sector_size_p.
216 */
217 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
218 {
219 unsigned int sector_size;
220 bool success = false;
221
222 errno = ENOTSUP;
223
224 /* Try a few ioctls to get the right size */
225 #ifdef BLKSSZGET
226 if (ioctl(fd, BLKSSZGET, &sector_size) >= 0) {
227 *sector_size_p = sector_size;
228 success = true;
229 }
230 #endif
231 #ifdef DKIOCGETBLOCKSIZE
232 if (ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) >= 0) {
233 *sector_size_p = sector_size;
234 success = true;
235 }
236 #endif
237 #ifdef DIOCGSECTORSIZE
238 if (ioctl(fd, DIOCGSECTORSIZE, &sector_size) >= 0) {
239 *sector_size_p = sector_size;
240 success = true;
241 }
242 #endif
243
244 return success ? 0 : -errno;
245 }
246
247 /**
248 * Get physical block size of @fd.
249 * On success, store it in @blk_size and return 0.
250 * On failure, return -errno.
251 */
252 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
253 {
254 #ifdef BLKPBSZGET
255 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
256 return -errno;
257 }
258 return 0;
259 #else
260 return -ENOTSUP;
261 #endif
262 }
263
264 /* Check if read is allowed with given memory buffer and length.
265 *
266 * This function is used to check O_DIRECT memory buffer and request alignment.
267 */
268 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
269 {
270 ssize_t ret = pread(fd, buf, len, 0);
271
272 if (ret >= 0) {
273 return true;
274 }
275
276 #ifdef __linux__
277 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
278 * other errors (e.g. real I/O error), which could happen on a failed
279 * drive, since we only care about probing alignment.
280 */
281 if (errno != EINVAL) {
282 return true;
283 }
284 #endif
285
286 return false;
287 }
288
289 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
290 {
291 BDRVRawState *s = bs->opaque;
292 char *buf;
293 size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize());
294
295 /* For SCSI generic devices the alignment is not really used.
296 With buffered I/O, we don't have any restrictions. */
297 if (bdrv_is_sg(bs) || !s->needs_alignment) {
298 bs->bl.request_alignment = 1;
299 s->buf_align = 1;
300 return;
301 }
302
303 bs->bl.request_alignment = 0;
304 s->buf_align = 0;
305 /* Let's try to use the logical blocksize for the alignment. */
306 if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) {
307 bs->bl.request_alignment = 0;
308 }
309 #ifdef CONFIG_XFS
310 if (s->is_xfs) {
311 struct dioattr da;
312 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
313 bs->bl.request_alignment = da.d_miniosz;
314 /* The kernel returns wrong information for d_mem */
315 /* s->buf_align = da.d_mem; */
316 }
317 }
318 #endif
319
320 /* If we could not get the sizes so far, we can only guess them */
321 if (!s->buf_align) {
322 size_t align;
323 buf = qemu_memalign(max_align, 2 * max_align);
324 for (align = 512; align <= max_align; align <<= 1) {
325 if (raw_is_io_aligned(fd, buf + align, max_align)) {
326 s->buf_align = align;
327 break;
328 }
329 }
330 qemu_vfree(buf);
331 }
332
333 if (!bs->bl.request_alignment) {
334 size_t align;
335 buf = qemu_memalign(s->buf_align, max_align);
336 for (align = 512; align <= max_align; align <<= 1) {
337 if (raw_is_io_aligned(fd, buf, align)) {
338 bs->bl.request_alignment = align;
339 break;
340 }
341 }
342 qemu_vfree(buf);
343 }
344
345 if (!s->buf_align || !bs->bl.request_alignment) {
346 error_setg(errp, "Could not find working O_DIRECT alignment");
347 error_append_hint(errp, "Try cache.direct=off\n");
348 }
349 }
350
351 static void raw_parse_flags(int bdrv_flags, int *open_flags)
352 {
353 assert(open_flags != NULL);
354
355 *open_flags |= O_BINARY;
356 *open_flags &= ~O_ACCMODE;
357 if (bdrv_flags & BDRV_O_RDWR) {
358 *open_flags |= O_RDWR;
359 } else {
360 *open_flags |= O_RDONLY;
361 }
362
363 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
364 * and O_DIRECT for no caching. */
365 if ((bdrv_flags & BDRV_O_NOCACHE)) {
366 *open_flags |= O_DIRECT;
367 }
368 }
369
370 #ifdef CONFIG_LINUX_AIO
371 static bool raw_use_aio(int bdrv_flags)
372 {
373 /*
374 * Currently Linux do AIO only for files opened with O_DIRECT
375 * specified so check NOCACHE flag too
376 */
377 return (bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
378 (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO);
379 }
380 #endif
381
382 static void raw_parse_filename(const char *filename, QDict *options,
383 Error **errp)
384 {
385 /* The filename does not have to be prefixed by the protocol name, since
386 * "file" is the default protocol; therefore, the return value of this
387 * function call can be ignored. */
388 strstart(filename, "file:", &filename);
389
390 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
391 }
392
393 static QemuOptsList raw_runtime_opts = {
394 .name = "raw",
395 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
396 .desc = {
397 {
398 .name = "filename",
399 .type = QEMU_OPT_STRING,
400 .help = "File name of the image",
401 },
402 { /* end of list */ }
403 },
404 };
405
406 static int raw_open_common(BlockDriverState *bs, QDict *options,
407 int bdrv_flags, int open_flags, Error **errp)
408 {
409 BDRVRawState *s = bs->opaque;
410 QemuOpts *opts;
411 Error *local_err = NULL;
412 const char *filename = NULL;
413 int fd, ret;
414 struct stat st;
415
416 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
417 qemu_opts_absorb_qdict(opts, options, &local_err);
418 if (local_err) {
419 error_propagate(errp, local_err);
420 ret = -EINVAL;
421 goto fail;
422 }
423
424 filename = qemu_opt_get(opts, "filename");
425
426 ret = raw_normalize_devicepath(&filename);
427 if (ret != 0) {
428 error_setg_errno(errp, -ret, "Could not normalize device path");
429 goto fail;
430 }
431
432 s->open_flags = open_flags;
433 raw_parse_flags(bdrv_flags, &s->open_flags);
434
435 s->fd = -1;
436 fd = qemu_open(filename, s->open_flags, 0644);
437 if (fd < 0) {
438 ret = -errno;
439 if (ret == -EROFS) {
440 ret = -EACCES;
441 }
442 goto fail;
443 }
444 s->fd = fd;
445
446 #ifdef CONFIG_LINUX_AIO
447 if (!raw_use_aio(bdrv_flags) && (bdrv_flags & BDRV_O_NATIVE_AIO)) {
448 error_setg(errp, "aio=native was specified, but it requires "
449 "cache.direct=on, which was not specified.");
450 ret = -EINVAL;
451 goto fail;
452 }
453 #else
454 if (bdrv_flags & BDRV_O_NATIVE_AIO) {
455 error_setg(errp, "aio=native was specified, but is not supported "
456 "in this build.");
457 ret = -EINVAL;
458 goto fail;
459 }
460 #endif /* !defined(CONFIG_LINUX_AIO) */
461
462 s->has_discard = true;
463 s->has_write_zeroes = true;
464 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
465 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
466 s->needs_alignment = true;
467 }
468
469 if (fstat(s->fd, &st) < 0) {
470 ret = -errno;
471 error_setg_errno(errp, errno, "Could not stat file");
472 goto fail;
473 }
474 if (S_ISREG(st.st_mode)) {
475 s->discard_zeroes = true;
476 s->has_fallocate = true;
477 }
478 if (S_ISBLK(st.st_mode)) {
479 #ifdef BLKDISCARDZEROES
480 unsigned int arg;
481 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
482 s->discard_zeroes = true;
483 }
484 #endif
485 #ifdef __linux__
486 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
487 * not rely on the contents of discarded blocks unless using O_DIRECT.
488 * Same for BLKZEROOUT.
489 */
490 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
491 s->discard_zeroes = false;
492 s->has_write_zeroes = false;
493 }
494 #endif
495 }
496 #ifdef __FreeBSD__
497 if (S_ISCHR(st.st_mode)) {
498 /*
499 * The file is a char device (disk), which on FreeBSD isn't behind
500 * a pager, so force all requests to be aligned. This is needed
501 * so QEMU makes sure all IO operations on the device are aligned
502 * to sector size, or else FreeBSD will reject them with EINVAL.
503 */
504 s->needs_alignment = true;
505 }
506 #endif
507
508 #ifdef CONFIG_XFS
509 if (platform_test_xfs_fd(s->fd)) {
510 s->is_xfs = true;
511 }
512 #endif
513
514 ret = 0;
515 fail:
516 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
517 unlink(filename);
518 }
519 qemu_opts_del(opts);
520 return ret;
521 }
522
523 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
524 Error **errp)
525 {
526 BDRVRawState *s = bs->opaque;
527
528 s->type = FTYPE_FILE;
529 return raw_open_common(bs, options, flags, 0, errp);
530 }
531
532 static int raw_reopen_prepare(BDRVReopenState *state,
533 BlockReopenQueue *queue, Error **errp)
534 {
535 BDRVRawState *s;
536 BDRVRawReopenState *raw_s;
537 int ret = 0;
538 Error *local_err = NULL;
539
540 assert(state != NULL);
541 assert(state->bs != NULL);
542
543 s = state->bs->opaque;
544
545 state->opaque = g_new0(BDRVRawReopenState, 1);
546 raw_s = state->opaque;
547
548 if (s->type == FTYPE_CD) {
549 raw_s->open_flags |= O_NONBLOCK;
550 }
551
552 raw_parse_flags(state->flags, &raw_s->open_flags);
553
554 raw_s->fd = -1;
555
556 int fcntl_flags = O_APPEND | O_NONBLOCK;
557 #ifdef O_NOATIME
558 fcntl_flags |= O_NOATIME;
559 #endif
560
561 #ifdef O_ASYNC
562 /* Not all operating systems have O_ASYNC, and those that don't
563 * will not let us track the state into raw_s->open_flags (typically
564 * you achieve the same effect with an ioctl, for example I_SETSIG
565 * on Solaris). But we do not use O_ASYNC, so that's fine.
566 */
567 assert((s->open_flags & O_ASYNC) == 0);
568 #endif
569
570 if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
571 /* dup the original fd */
572 raw_s->fd = qemu_dup(s->fd);
573 if (raw_s->fd >= 0) {
574 ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
575 if (ret) {
576 qemu_close(raw_s->fd);
577 raw_s->fd = -1;
578 }
579 }
580 }
581
582 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
583 if (raw_s->fd == -1) {
584 const char *normalized_filename = state->bs->filename;
585 ret = raw_normalize_devicepath(&normalized_filename);
586 if (ret < 0) {
587 error_setg_errno(errp, -ret, "Could not normalize device path");
588 } else {
589 assert(!(raw_s->open_flags & O_CREAT));
590 raw_s->fd = qemu_open(normalized_filename, raw_s->open_flags);
591 if (raw_s->fd == -1) {
592 error_setg_errno(errp, errno, "Could not reopen file");
593 ret = -1;
594 }
595 }
596 }
597
598 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
599 * alignment with the new fd. */
600 if (raw_s->fd != -1) {
601 raw_probe_alignment(state->bs, raw_s->fd, &local_err);
602 if (local_err) {
603 qemu_close(raw_s->fd);
604 raw_s->fd = -1;
605 error_propagate(errp, local_err);
606 ret = -EINVAL;
607 }
608 }
609
610 return ret;
611 }
612
613 static void raw_reopen_commit(BDRVReopenState *state)
614 {
615 BDRVRawReopenState *raw_s = state->opaque;
616 BDRVRawState *s = state->bs->opaque;
617
618 s->open_flags = raw_s->open_flags;
619
620 qemu_close(s->fd);
621 s->fd = raw_s->fd;
622
623 g_free(state->opaque);
624 state->opaque = NULL;
625 }
626
627
628 static void raw_reopen_abort(BDRVReopenState *state)
629 {
630 BDRVRawReopenState *raw_s = state->opaque;
631
632 /* nothing to do if NULL, we didn't get far enough */
633 if (raw_s == NULL) {
634 return;
635 }
636
637 if (raw_s->fd >= 0) {
638 qemu_close(raw_s->fd);
639 raw_s->fd = -1;
640 }
641 g_free(state->opaque);
642 state->opaque = NULL;
643 }
644
645 static int hdev_get_max_transfer_length(int fd)
646 {
647 #ifdef BLKSECTGET
648 int max_sectors = 0;
649 if (ioctl(fd, BLKSECTGET, &max_sectors) == 0) {
650 return max_sectors;
651 } else {
652 return -errno;
653 }
654 #else
655 return -ENOSYS;
656 #endif
657 }
658
659 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
660 {
661 BDRVRawState *s = bs->opaque;
662 struct stat st;
663
664 if (!fstat(s->fd, &st)) {
665 if (S_ISBLK(st.st_mode)) {
666 int ret = hdev_get_max_transfer_length(s->fd);
667 if (ret > 0 && ret <= BDRV_REQUEST_MAX_SECTORS) {
668 bs->bl.max_transfer = pow2floor(ret << BDRV_SECTOR_BITS);
669 }
670 }
671 }
672
673 raw_probe_alignment(bs, s->fd, errp);
674 bs->bl.min_mem_alignment = s->buf_align;
675 bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
676 }
677
678 static int check_for_dasd(int fd)
679 {
680 #ifdef BIODASDINFO2
681 struct dasd_information2_t info = {0};
682
683 return ioctl(fd, BIODASDINFO2, &info);
684 #else
685 return -1;
686 #endif
687 }
688
689 /**
690 * Try to get @bs's logical and physical block size.
691 * On success, store them in @bsz and return zero.
692 * On failure, return negative errno.
693 */
694 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
695 {
696 BDRVRawState *s = bs->opaque;
697 int ret;
698
699 /* If DASD, get blocksizes */
700 if (check_for_dasd(s->fd) < 0) {
701 return -ENOTSUP;
702 }
703 ret = probe_logical_blocksize(s->fd, &bsz->log);
704 if (ret < 0) {
705 return ret;
706 }
707 return probe_physical_blocksize(s->fd, &bsz->phys);
708 }
709
710 /**
711 * Try to get @bs's geometry: cyls, heads, sectors.
712 * On success, store them in @geo and return 0.
713 * On failure return -errno.
714 * (Allows block driver to assign default geometry values that guest sees)
715 */
716 #ifdef __linux__
717 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
718 {
719 BDRVRawState *s = bs->opaque;
720 struct hd_geometry ioctl_geo = {0};
721
722 /* If DASD, get its geometry */
723 if (check_for_dasd(s->fd) < 0) {
724 return -ENOTSUP;
725 }
726 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
727 return -errno;
728 }
729 /* HDIO_GETGEO may return success even though geo contains zeros
730 (e.g. certain multipath setups) */
731 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
732 return -ENOTSUP;
733 }
734 /* Do not return a geometry for partition */
735 if (ioctl_geo.start != 0) {
736 return -ENOTSUP;
737 }
738 geo->heads = ioctl_geo.heads;
739 geo->sectors = ioctl_geo.sectors;
740 geo->cylinders = ioctl_geo.cylinders;
741
742 return 0;
743 }
744 #else /* __linux__ */
745 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
746 {
747 return -ENOTSUP;
748 }
749 #endif
750
751 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
752 {
753 int ret;
754
755 ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
756 if (ret == -1) {
757 return -errno;
758 }
759
760 return 0;
761 }
762
763 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
764 {
765 int ret;
766
767 ret = qemu_fdatasync(aiocb->aio_fildes);
768 if (ret == -1) {
769 return -errno;
770 }
771 return 0;
772 }
773
774 #ifdef CONFIG_PREADV
775
776 static bool preadv_present = true;
777
778 static ssize_t
779 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
780 {
781 return preadv(fd, iov, nr_iov, offset);
782 }
783
784 static ssize_t
785 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
786 {
787 return pwritev(fd, iov, nr_iov, offset);
788 }
789
790 #else
791
792 static bool preadv_present = false;
793
794 static ssize_t
795 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
796 {
797 return -ENOSYS;
798 }
799
800 static ssize_t
801 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
802 {
803 return -ENOSYS;
804 }
805
806 #endif
807
808 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
809 {
810 ssize_t len;
811
812 do {
813 if (aiocb->aio_type & QEMU_AIO_WRITE)
814 len = qemu_pwritev(aiocb->aio_fildes,
815 aiocb->aio_iov,
816 aiocb->aio_niov,
817 aiocb->aio_offset);
818 else
819 len = qemu_preadv(aiocb->aio_fildes,
820 aiocb->aio_iov,
821 aiocb->aio_niov,
822 aiocb->aio_offset);
823 } while (len == -1 && errno == EINTR);
824
825 if (len == -1) {
826 return -errno;
827 }
828 return len;
829 }
830
831 /*
832 * Read/writes the data to/from a given linear buffer.
833 *
834 * Returns the number of bytes handles or -errno in case of an error. Short
835 * reads are only returned if the end of the file is reached.
836 */
837 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
838 {
839 ssize_t offset = 0;
840 ssize_t len;
841
842 while (offset < aiocb->aio_nbytes) {
843 if (aiocb->aio_type & QEMU_AIO_WRITE) {
844 len = pwrite(aiocb->aio_fildes,
845 (const char *)buf + offset,
846 aiocb->aio_nbytes - offset,
847 aiocb->aio_offset + offset);
848 } else {
849 len = pread(aiocb->aio_fildes,
850 buf + offset,
851 aiocb->aio_nbytes - offset,
852 aiocb->aio_offset + offset);
853 }
854 if (len == -1 && errno == EINTR) {
855 continue;
856 } else if (len == -1 && errno == EINVAL &&
857 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
858 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
859 offset > 0) {
860 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
861 * after a short read. Assume that O_DIRECT short reads only occur
862 * at EOF. Therefore this is a short read, not an I/O error.
863 */
864 break;
865 } else if (len == -1) {
866 offset = -errno;
867 break;
868 } else if (len == 0) {
869 break;
870 }
871 offset += len;
872 }
873
874 return offset;
875 }
876
877 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
878 {
879 ssize_t nbytes;
880 char *buf;
881
882 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
883 /*
884 * If there is just a single buffer, and it is properly aligned
885 * we can just use plain pread/pwrite without any problems.
886 */
887 if (aiocb->aio_niov == 1) {
888 return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
889 }
890 /*
891 * We have more than one iovec, and all are properly aligned.
892 *
893 * Try preadv/pwritev first and fall back to linearizing the
894 * buffer if it's not supported.
895 */
896 if (preadv_present) {
897 nbytes = handle_aiocb_rw_vector(aiocb);
898 if (nbytes == aiocb->aio_nbytes ||
899 (nbytes < 0 && nbytes != -ENOSYS)) {
900 return nbytes;
901 }
902 preadv_present = false;
903 }
904
905 /*
906 * XXX(hch): short read/write. no easy way to handle the reminder
907 * using these interfaces. For now retry using plain
908 * pread/pwrite?
909 */
910 }
911
912 /*
913 * Ok, we have to do it the hard way, copy all segments into
914 * a single aligned buffer.
915 */
916 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
917 if (buf == NULL) {
918 return -ENOMEM;
919 }
920
921 if (aiocb->aio_type & QEMU_AIO_WRITE) {
922 char *p = buf;
923 int i;
924
925 for (i = 0; i < aiocb->aio_niov; ++i) {
926 memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
927 p += aiocb->aio_iov[i].iov_len;
928 }
929 assert(p - buf == aiocb->aio_nbytes);
930 }
931
932 nbytes = handle_aiocb_rw_linear(aiocb, buf);
933 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
934 char *p = buf;
935 size_t count = aiocb->aio_nbytes, copy;
936 int i;
937
938 for (i = 0; i < aiocb->aio_niov && count; ++i) {
939 copy = count;
940 if (copy > aiocb->aio_iov[i].iov_len) {
941 copy = aiocb->aio_iov[i].iov_len;
942 }
943 memcpy(aiocb->aio_iov[i].iov_base, p, copy);
944 assert(count >= copy);
945 p += copy;
946 count -= copy;
947 }
948 assert(count == 0);
949 }
950 qemu_vfree(buf);
951
952 return nbytes;
953 }
954
955 #ifdef CONFIG_XFS
956 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
957 {
958 struct xfs_flock64 fl;
959 int err;
960
961 memset(&fl, 0, sizeof(fl));
962 fl.l_whence = SEEK_SET;
963 fl.l_start = offset;
964 fl.l_len = bytes;
965
966 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
967 err = errno;
968 DPRINTF("cannot write zero range (%s)\n", strerror(errno));
969 return -err;
970 }
971
972 return 0;
973 }
974
975 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
976 {
977 struct xfs_flock64 fl;
978 int err;
979
980 memset(&fl, 0, sizeof(fl));
981 fl.l_whence = SEEK_SET;
982 fl.l_start = offset;
983 fl.l_len = bytes;
984
985 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
986 err = errno;
987 DPRINTF("cannot punch hole (%s)\n", strerror(errno));
988 return -err;
989 }
990
991 return 0;
992 }
993 #endif
994
995 static int translate_err(int err)
996 {
997 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
998 err == -ENOTTY) {
999 err = -ENOTSUP;
1000 }
1001 return err;
1002 }
1003
1004 #ifdef CONFIG_FALLOCATE
1005 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1006 {
1007 do {
1008 if (fallocate(fd, mode, offset, len) == 0) {
1009 return 0;
1010 }
1011 } while (errno == EINTR);
1012 return translate_err(-errno);
1013 }
1014 #endif
1015
1016 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1017 {
1018 int ret = -ENOTSUP;
1019 BDRVRawState *s = aiocb->bs->opaque;
1020
1021 if (!s->has_write_zeroes) {
1022 return -ENOTSUP;
1023 }
1024
1025 #ifdef BLKZEROOUT
1026 do {
1027 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1028 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1029 return 0;
1030 }
1031 } while (errno == EINTR);
1032
1033 ret = translate_err(-errno);
1034 #endif
1035
1036 if (ret == -ENOTSUP) {
1037 s->has_write_zeroes = false;
1038 }
1039 return ret;
1040 }
1041
1042 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
1043 {
1044 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1045 BDRVRawState *s = aiocb->bs->opaque;
1046 #endif
1047
1048 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1049 return handle_aiocb_write_zeroes_block(aiocb);
1050 }
1051
1052 #ifdef CONFIG_XFS
1053 if (s->is_xfs) {
1054 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
1055 }
1056 #endif
1057
1058 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1059 if (s->has_write_zeroes) {
1060 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1061 aiocb->aio_offset, aiocb->aio_nbytes);
1062 if (ret == 0 || ret != -ENOTSUP) {
1063 return ret;
1064 }
1065 s->has_write_zeroes = false;
1066 }
1067 #endif
1068
1069 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1070 if (s->has_discard && s->has_fallocate) {
1071 int ret = do_fallocate(s->fd,
1072 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1073 aiocb->aio_offset, aiocb->aio_nbytes);
1074 if (ret == 0) {
1075 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1076 if (ret == 0 || ret != -ENOTSUP) {
1077 return ret;
1078 }
1079 s->has_fallocate = false;
1080 } else if (ret != -ENOTSUP) {
1081 return ret;
1082 } else {
1083 s->has_discard = false;
1084 }
1085 }
1086 #endif
1087
1088 #ifdef CONFIG_FALLOCATE
1089 if (s->has_fallocate && aiocb->aio_offset >= bdrv_getlength(aiocb->bs)) {
1090 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1091 if (ret == 0 || ret != -ENOTSUP) {
1092 return ret;
1093 }
1094 s->has_fallocate = false;
1095 }
1096 #endif
1097
1098 return -ENOTSUP;
1099 }
1100
1101 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
1102 {
1103 int ret = -EOPNOTSUPP;
1104 BDRVRawState *s = aiocb->bs->opaque;
1105
1106 if (!s->has_discard) {
1107 return -ENOTSUP;
1108 }
1109
1110 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1111 #ifdef BLKDISCARD
1112 do {
1113 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1114 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1115 return 0;
1116 }
1117 } while (errno == EINTR);
1118
1119 ret = -errno;
1120 #endif
1121 } else {
1122 #ifdef CONFIG_XFS
1123 if (s->is_xfs) {
1124 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1125 }
1126 #endif
1127
1128 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1129 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1130 aiocb->aio_offset, aiocb->aio_nbytes);
1131 #endif
1132 }
1133
1134 ret = translate_err(ret);
1135 if (ret == -ENOTSUP) {
1136 s->has_discard = false;
1137 }
1138 return ret;
1139 }
1140
1141 static int aio_worker(void *arg)
1142 {
1143 RawPosixAIOData *aiocb = arg;
1144 ssize_t ret = 0;
1145
1146 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
1147 case QEMU_AIO_READ:
1148 ret = handle_aiocb_rw(aiocb);
1149 if (ret >= 0 && ret < aiocb->aio_nbytes) {
1150 iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
1151 0, aiocb->aio_nbytes - ret);
1152
1153 ret = aiocb->aio_nbytes;
1154 }
1155 if (ret == aiocb->aio_nbytes) {
1156 ret = 0;
1157 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1158 ret = -EINVAL;
1159 }
1160 break;
1161 case QEMU_AIO_WRITE:
1162 ret = handle_aiocb_rw(aiocb);
1163 if (ret == aiocb->aio_nbytes) {
1164 ret = 0;
1165 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1166 ret = -EINVAL;
1167 }
1168 break;
1169 case QEMU_AIO_FLUSH:
1170 ret = handle_aiocb_flush(aiocb);
1171 break;
1172 case QEMU_AIO_IOCTL:
1173 ret = handle_aiocb_ioctl(aiocb);
1174 break;
1175 case QEMU_AIO_DISCARD:
1176 ret = handle_aiocb_discard(aiocb);
1177 break;
1178 case QEMU_AIO_WRITE_ZEROES:
1179 ret = handle_aiocb_write_zeroes(aiocb);
1180 break;
1181 default:
1182 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1183 ret = -EINVAL;
1184 break;
1185 }
1186
1187 g_free(aiocb);
1188 return ret;
1189 }
1190
1191 static int paio_submit_co(BlockDriverState *bs, int fd,
1192 int64_t offset, QEMUIOVector *qiov,
1193 int count, int type)
1194 {
1195 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1196 ThreadPool *pool;
1197
1198 acb->bs = bs;
1199 acb->aio_type = type;
1200 acb->aio_fildes = fd;
1201
1202 acb->aio_nbytes = count;
1203 acb->aio_offset = offset;
1204
1205 if (qiov) {
1206 acb->aio_iov = qiov->iov;
1207 acb->aio_niov = qiov->niov;
1208 assert(qiov->size == count);
1209 }
1210
1211 trace_paio_submit_co(offset, count, type);
1212 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1213 return thread_pool_submit_co(pool, aio_worker, acb);
1214 }
1215
1216 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
1217 int64_t offset, QEMUIOVector *qiov, int count,
1218 BlockCompletionFunc *cb, void *opaque, int type)
1219 {
1220 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1221 ThreadPool *pool;
1222
1223 acb->bs = bs;
1224 acb->aio_type = type;
1225 acb->aio_fildes = fd;
1226
1227 acb->aio_nbytes = count;
1228 acb->aio_offset = offset;
1229
1230 if (qiov) {
1231 acb->aio_iov = qiov->iov;
1232 acb->aio_niov = qiov->niov;
1233 assert(qiov->size == acb->aio_nbytes);
1234 }
1235
1236 trace_paio_submit(acb, opaque, offset, count, type);
1237 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1238 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1239 }
1240
1241 static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset,
1242 uint64_t bytes, QEMUIOVector *qiov, int type)
1243 {
1244 BDRVRawState *s = bs->opaque;
1245
1246 if (fd_open(bs) < 0)
1247 return -EIO;
1248
1249 /*
1250 * Check if the underlying device requires requests to be aligned,
1251 * and if the request we are trying to submit is aligned or not.
1252 * If this is the case tell the low-level driver that it needs
1253 * to copy the buffer.
1254 */
1255 if (s->needs_alignment) {
1256 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1257 type |= QEMU_AIO_MISALIGNED;
1258 #ifdef CONFIG_LINUX_AIO
1259 } else if (bs->open_flags & BDRV_O_NATIVE_AIO) {
1260 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1261 assert(qiov->size == bytes);
1262 return laio_co_submit(bs, aio, s->fd, offset, qiov, type);
1263 #endif
1264 }
1265 }
1266
1267 return paio_submit_co(bs, s->fd, offset, qiov, bytes, type);
1268 }
1269
1270 static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset,
1271 uint64_t bytes, QEMUIOVector *qiov,
1272 int flags)
1273 {
1274 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ);
1275 }
1276
1277 static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
1278 uint64_t bytes, QEMUIOVector *qiov,
1279 int flags)
1280 {
1281 assert(flags == 0);
1282 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE);
1283 }
1284
1285 static void raw_aio_plug(BlockDriverState *bs)
1286 {
1287 #ifdef CONFIG_LINUX_AIO
1288 if (bs->open_flags & BDRV_O_NATIVE_AIO) {
1289 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1290 laio_io_plug(bs, aio);
1291 }
1292 #endif
1293 }
1294
1295 static void raw_aio_unplug(BlockDriverState *bs)
1296 {
1297 #ifdef CONFIG_LINUX_AIO
1298 if (bs->open_flags & BDRV_O_NATIVE_AIO) {
1299 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1300 laio_io_unplug(bs, aio);
1301 }
1302 #endif
1303 }
1304
1305 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
1306 BlockCompletionFunc *cb, void *opaque)
1307 {
1308 BDRVRawState *s = bs->opaque;
1309
1310 if (fd_open(bs) < 0)
1311 return NULL;
1312
1313 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1314 }
1315
1316 static void raw_close(BlockDriverState *bs)
1317 {
1318 BDRVRawState *s = bs->opaque;
1319
1320 if (s->fd >= 0) {
1321 qemu_close(s->fd);
1322 s->fd = -1;
1323 }
1324 }
1325
1326 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1327 {
1328 BDRVRawState *s = bs->opaque;
1329 struct stat st;
1330
1331 if (fstat(s->fd, &st)) {
1332 return -errno;
1333 }
1334
1335 if (S_ISREG(st.st_mode)) {
1336 if (ftruncate(s->fd, offset) < 0) {
1337 return -errno;
1338 }
1339 } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1340 if (offset > raw_getlength(bs)) {
1341 return -EINVAL;
1342 }
1343 } else {
1344 return -ENOTSUP;
1345 }
1346
1347 return 0;
1348 }
1349
1350 #ifdef __OpenBSD__
1351 static int64_t raw_getlength(BlockDriverState *bs)
1352 {
1353 BDRVRawState *s = bs->opaque;
1354 int fd = s->fd;
1355 struct stat st;
1356
1357 if (fstat(fd, &st))
1358 return -errno;
1359 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1360 struct disklabel dl;
1361
1362 if (ioctl(fd, DIOCGDINFO, &dl))
1363 return -errno;
1364 return (uint64_t)dl.d_secsize *
1365 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1366 } else
1367 return st.st_size;
1368 }
1369 #elif defined(__NetBSD__)
1370 static int64_t raw_getlength(BlockDriverState *bs)
1371 {
1372 BDRVRawState *s = bs->opaque;
1373 int fd = s->fd;
1374 struct stat st;
1375
1376 if (fstat(fd, &st))
1377 return -errno;
1378 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1379 struct dkwedge_info dkw;
1380
1381 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1382 return dkw.dkw_size * 512;
1383 } else {
1384 struct disklabel dl;
1385
1386 if (ioctl(fd, DIOCGDINFO, &dl))
1387 return -errno;
1388 return (uint64_t)dl.d_secsize *
1389 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1390 }
1391 } else
1392 return st.st_size;
1393 }
1394 #elif defined(__sun__)
1395 static int64_t raw_getlength(BlockDriverState *bs)
1396 {
1397 BDRVRawState *s = bs->opaque;
1398 struct dk_minfo minfo;
1399 int ret;
1400 int64_t size;
1401
1402 ret = fd_open(bs);
1403 if (ret < 0) {
1404 return ret;
1405 }
1406
1407 /*
1408 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1409 */
1410 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1411 if (ret != -1) {
1412 return minfo.dki_lbsize * minfo.dki_capacity;
1413 }
1414
1415 /*
1416 * There are reports that lseek on some devices fails, but
1417 * irc discussion said that contingency on contingency was overkill.
1418 */
1419 size = lseek(s->fd, 0, SEEK_END);
1420 if (size < 0) {
1421 return -errno;
1422 }
1423 return size;
1424 }
1425 #elif defined(CONFIG_BSD)
1426 static int64_t raw_getlength(BlockDriverState *bs)
1427 {
1428 BDRVRawState *s = bs->opaque;
1429 int fd = s->fd;
1430 int64_t size;
1431 struct stat sb;
1432 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1433 int reopened = 0;
1434 #endif
1435 int ret;
1436
1437 ret = fd_open(bs);
1438 if (ret < 0)
1439 return ret;
1440
1441 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1442 again:
1443 #endif
1444 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1445 #ifdef DIOCGMEDIASIZE
1446 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1447 #elif defined(DIOCGPART)
1448 {
1449 struct partinfo pi;
1450 if (ioctl(fd, DIOCGPART, &pi) == 0)
1451 size = pi.media_size;
1452 else
1453 size = 0;
1454 }
1455 if (size == 0)
1456 #endif
1457 #if defined(__APPLE__) && defined(__MACH__)
1458 {
1459 uint64_t sectors = 0;
1460 uint32_t sector_size = 0;
1461
1462 if (ioctl(fd, DKIOCGETBLOCKCOUNT, &sectors) == 0
1463 && ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) == 0) {
1464 size = sectors * sector_size;
1465 } else {
1466 size = lseek(fd, 0LL, SEEK_END);
1467 if (size < 0) {
1468 return -errno;
1469 }
1470 }
1471 }
1472 #else
1473 size = lseek(fd, 0LL, SEEK_END);
1474 if (size < 0) {
1475 return -errno;
1476 }
1477 #endif
1478 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1479 switch(s->type) {
1480 case FTYPE_CD:
1481 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1482 if (size == 2048LL * (unsigned)-1)
1483 size = 0;
1484 /* XXX no disc? maybe we need to reopen... */
1485 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1486 reopened = 1;
1487 goto again;
1488 }
1489 }
1490 #endif
1491 } else {
1492 size = lseek(fd, 0, SEEK_END);
1493 if (size < 0) {
1494 return -errno;
1495 }
1496 }
1497 return size;
1498 }
1499 #else
1500 static int64_t raw_getlength(BlockDriverState *bs)
1501 {
1502 BDRVRawState *s = bs->opaque;
1503 int ret;
1504 int64_t size;
1505
1506 ret = fd_open(bs);
1507 if (ret < 0) {
1508 return ret;
1509 }
1510
1511 size = lseek(s->fd, 0, SEEK_END);
1512 if (size < 0) {
1513 return -errno;
1514 }
1515 return size;
1516 }
1517 #endif
1518
1519 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1520 {
1521 struct stat st;
1522 BDRVRawState *s = bs->opaque;
1523
1524 if (fstat(s->fd, &st) < 0) {
1525 return -errno;
1526 }
1527 return (int64_t)st.st_blocks * 512;
1528 }
1529
1530 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
1531 {
1532 int fd;
1533 int result = 0;
1534 int64_t total_size = 0;
1535 bool nocow = false;
1536 PreallocMode prealloc;
1537 char *buf = NULL;
1538 Error *local_err = NULL;
1539
1540 strstart(filename, "file:", &filename);
1541
1542 /* Read out options */
1543 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1544 BDRV_SECTOR_SIZE);
1545 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
1546 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1547 prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
1548 PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
1549 &local_err);
1550 g_free(buf);
1551 if (local_err) {
1552 error_propagate(errp, local_err);
1553 result = -EINVAL;
1554 goto out;
1555 }
1556
1557 fd = qemu_open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY,
1558 0644);
1559 if (fd < 0) {
1560 result = -errno;
1561 error_setg_errno(errp, -result, "Could not create file");
1562 goto out;
1563 }
1564
1565 if (nocow) {
1566 #ifdef __linux__
1567 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1568 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1569 * will be ignored since any failure of this operation should not
1570 * block the left work.
1571 */
1572 int attr;
1573 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
1574 attr |= FS_NOCOW_FL;
1575 ioctl(fd, FS_IOC_SETFLAGS, &attr);
1576 }
1577 #endif
1578 }
1579
1580 if (ftruncate(fd, total_size) != 0) {
1581 result = -errno;
1582 error_setg_errno(errp, -result, "Could not resize file");
1583 goto out_close;
1584 }
1585
1586 switch (prealloc) {
1587 #ifdef CONFIG_POSIX_FALLOCATE
1588 case PREALLOC_MODE_FALLOC:
1589 /* posix_fallocate() doesn't set errno. */
1590 result = -posix_fallocate(fd, 0, total_size);
1591 if (result != 0) {
1592 error_setg_errno(errp, -result,
1593 "Could not preallocate data for the new file");
1594 }
1595 break;
1596 #endif
1597 case PREALLOC_MODE_FULL:
1598 {
1599 int64_t num = 0, left = total_size;
1600 buf = g_malloc0(65536);
1601
1602 while (left > 0) {
1603 num = MIN(left, 65536);
1604 result = write(fd, buf, num);
1605 if (result < 0) {
1606 result = -errno;
1607 error_setg_errno(errp, -result,
1608 "Could not write to the new file");
1609 break;
1610 }
1611 left -= result;
1612 }
1613 if (result >= 0) {
1614 result = fsync(fd);
1615 if (result < 0) {
1616 result = -errno;
1617 error_setg_errno(errp, -result,
1618 "Could not flush new file to disk");
1619 }
1620 }
1621 g_free(buf);
1622 break;
1623 }
1624 case PREALLOC_MODE_OFF:
1625 break;
1626 default:
1627 result = -EINVAL;
1628 error_setg(errp, "Unsupported preallocation mode: %s",
1629 PreallocMode_lookup[prealloc]);
1630 break;
1631 }
1632
1633 out_close:
1634 if (qemu_close(fd) != 0 && result == 0) {
1635 result = -errno;
1636 error_setg_errno(errp, -result, "Could not close the new file");
1637 }
1638 out:
1639 return result;
1640 }
1641
1642 /*
1643 * Find allocation range in @bs around offset @start.
1644 * May change underlying file descriptor's file offset.
1645 * If @start is not in a hole, store @start in @data, and the
1646 * beginning of the next hole in @hole, and return 0.
1647 * If @start is in a non-trailing hole, store @start in @hole and the
1648 * beginning of the next non-hole in @data, and return 0.
1649 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1650 * If we can't find out, return a negative errno other than -ENXIO.
1651 */
1652 static int find_allocation(BlockDriverState *bs, off_t start,
1653 off_t *data, off_t *hole)
1654 {
1655 #if defined SEEK_HOLE && defined SEEK_DATA
1656 BDRVRawState *s = bs->opaque;
1657 off_t offs;
1658
1659 /*
1660 * SEEK_DATA cases:
1661 * D1. offs == start: start is in data
1662 * D2. offs > start: start is in a hole, next data at offs
1663 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1664 * or start is beyond EOF
1665 * If the latter happens, the file has been truncated behind
1666 * our back since we opened it. All bets are off then.
1667 * Treating like a trailing hole is simplest.
1668 * D4. offs < 0, errno != ENXIO: we learned nothing
1669 */
1670 offs = lseek(s->fd, start, SEEK_DATA);
1671 if (offs < 0) {
1672 return -errno; /* D3 or D4 */
1673 }
1674 assert(offs >= start);
1675
1676 if (offs > start) {
1677 /* D2: in hole, next data at offs */
1678 *hole = start;
1679 *data = offs;
1680 return 0;
1681 }
1682
1683 /* D1: in data, end not yet known */
1684
1685 /*
1686 * SEEK_HOLE cases:
1687 * H1. offs == start: start is in a hole
1688 * If this happens here, a hole has been dug behind our back
1689 * since the previous lseek().
1690 * H2. offs > start: either start is in data, next hole at offs,
1691 * or start is in trailing hole, EOF at offs
1692 * Linux treats trailing holes like any other hole: offs ==
1693 * start. Solaris seeks to EOF instead: offs > start (blech).
1694 * If that happens here, a hole has been dug behind our back
1695 * since the previous lseek().
1696 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1697 * If this happens, the file has been truncated behind our
1698 * back since we opened it. Treat it like a trailing hole.
1699 * H4. offs < 0, errno != ENXIO: we learned nothing
1700 * Pretend we know nothing at all, i.e. "forget" about D1.
1701 */
1702 offs = lseek(s->fd, start, SEEK_HOLE);
1703 if (offs < 0) {
1704 return -errno; /* D1 and (H3 or H4) */
1705 }
1706 assert(offs >= start);
1707
1708 if (offs > start) {
1709 /*
1710 * D1 and H2: either in data, next hole at offs, or it was in
1711 * data but is now in a trailing hole. In the latter case,
1712 * all bets are off. Treating it as if it there was data all
1713 * the way to EOF is safe, so simply do that.
1714 */
1715 *data = start;
1716 *hole = offs;
1717 return 0;
1718 }
1719
1720 /* D1 and H1 */
1721 return -EBUSY;
1722 #else
1723 return -ENOTSUP;
1724 #endif
1725 }
1726
1727 /*
1728 * Returns the allocation status of the specified sectors.
1729 *
1730 * If 'sector_num' is beyond the end of the disk image the return value is 0
1731 * and 'pnum' is set to 0.
1732 *
1733 * 'pnum' is set to the number of sectors (including and immediately following
1734 * the specified sector) that are known to be in the same
1735 * allocated/unallocated state.
1736 *
1737 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1738 * beyond the end of the disk image it will be clamped.
1739 */
1740 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1741 int64_t sector_num,
1742 int nb_sectors, int *pnum,
1743 BlockDriverState **file)
1744 {
1745 off_t start, data = 0, hole = 0;
1746 int64_t total_size;
1747 int ret;
1748
1749 ret = fd_open(bs);
1750 if (ret < 0) {
1751 return ret;
1752 }
1753
1754 start = sector_num * BDRV_SECTOR_SIZE;
1755 total_size = bdrv_getlength(bs);
1756 if (total_size < 0) {
1757 return total_size;
1758 } else if (start >= total_size) {
1759 *pnum = 0;
1760 return 0;
1761 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1762 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1763 }
1764
1765 ret = find_allocation(bs, start, &data, &hole);
1766 if (ret == -ENXIO) {
1767 /* Trailing hole */
1768 *pnum = nb_sectors;
1769 ret = BDRV_BLOCK_ZERO;
1770 } else if (ret < 0) {
1771 /* No info available, so pretend there are no holes */
1772 *pnum = nb_sectors;
1773 ret = BDRV_BLOCK_DATA;
1774 } else if (data == start) {
1775 /* On a data extent, compute sectors to the end of the extent,
1776 * possibly including a partial sector at EOF. */
1777 *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1778 ret = BDRV_BLOCK_DATA;
1779 } else {
1780 /* On a hole, compute sectors to the beginning of the next extent. */
1781 assert(hole == start);
1782 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1783 ret = BDRV_BLOCK_ZERO;
1784 }
1785 *file = bs;
1786 return ret | BDRV_BLOCK_OFFSET_VALID | start;
1787 }
1788
1789 static coroutine_fn BlockAIOCB *raw_aio_pdiscard(BlockDriverState *bs,
1790 int64_t offset, int count,
1791 BlockCompletionFunc *cb, void *opaque)
1792 {
1793 BDRVRawState *s = bs->opaque;
1794
1795 return paio_submit(bs, s->fd, offset, NULL, count,
1796 cb, opaque, QEMU_AIO_DISCARD);
1797 }
1798
1799 static int coroutine_fn raw_co_pwrite_zeroes(
1800 BlockDriverState *bs, int64_t offset,
1801 int count, BdrvRequestFlags flags)
1802 {
1803 BDRVRawState *s = bs->opaque;
1804
1805 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1806 return paio_submit_co(bs, s->fd, offset, NULL, count,
1807 QEMU_AIO_WRITE_ZEROES);
1808 } else if (s->discard_zeroes) {
1809 return paio_submit_co(bs, s->fd, offset, NULL, count,
1810 QEMU_AIO_DISCARD);
1811 }
1812 return -ENOTSUP;
1813 }
1814
1815 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1816 {
1817 BDRVRawState *s = bs->opaque;
1818
1819 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1820 bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1821 return 0;
1822 }
1823
1824 static QemuOptsList raw_create_opts = {
1825 .name = "raw-create-opts",
1826 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
1827 .desc = {
1828 {
1829 .name = BLOCK_OPT_SIZE,
1830 .type = QEMU_OPT_SIZE,
1831 .help = "Virtual disk size"
1832 },
1833 {
1834 .name = BLOCK_OPT_NOCOW,
1835 .type = QEMU_OPT_BOOL,
1836 .help = "Turn off copy-on-write (valid only on btrfs)"
1837 },
1838 {
1839 .name = BLOCK_OPT_PREALLOC,
1840 .type = QEMU_OPT_STRING,
1841 .help = "Preallocation mode (allowed values: off, falloc, full)"
1842 },
1843 { /* end of list */ }
1844 }
1845 };
1846
1847 BlockDriver bdrv_file = {
1848 .format_name = "file",
1849 .protocol_name = "file",
1850 .instance_size = sizeof(BDRVRawState),
1851 .bdrv_needs_filename = true,
1852 .bdrv_probe = NULL, /* no probe for protocols */
1853 .bdrv_parse_filename = raw_parse_filename,
1854 .bdrv_file_open = raw_open,
1855 .bdrv_reopen_prepare = raw_reopen_prepare,
1856 .bdrv_reopen_commit = raw_reopen_commit,
1857 .bdrv_reopen_abort = raw_reopen_abort,
1858 .bdrv_close = raw_close,
1859 .bdrv_create = raw_create,
1860 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1861 .bdrv_co_get_block_status = raw_co_get_block_status,
1862 .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes,
1863
1864 .bdrv_co_preadv = raw_co_preadv,
1865 .bdrv_co_pwritev = raw_co_pwritev,
1866 .bdrv_aio_flush = raw_aio_flush,
1867 .bdrv_aio_pdiscard = raw_aio_pdiscard,
1868 .bdrv_refresh_limits = raw_refresh_limits,
1869 .bdrv_io_plug = raw_aio_plug,
1870 .bdrv_io_unplug = raw_aio_unplug,
1871
1872 .bdrv_truncate = raw_truncate,
1873 .bdrv_getlength = raw_getlength,
1874 .bdrv_get_info = raw_get_info,
1875 .bdrv_get_allocated_file_size
1876 = raw_get_allocated_file_size,
1877
1878 .create_opts = &raw_create_opts,
1879 };
1880
1881 /***********************************************/
1882 /* host device */
1883
1884 #if defined(__APPLE__) && defined(__MACH__)
1885 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
1886 CFIndex maxPathSize, int flags);
1887 static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
1888 {
1889 kern_return_t kernResult = KERN_FAILURE;
1890 mach_port_t masterPort;
1891 CFMutableDictionaryRef classesToMatch;
1892 const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
1893 char *mediaType = NULL;
1894
1895 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1896 if ( KERN_SUCCESS != kernResult ) {
1897 printf( "IOMasterPort returned %d\n", kernResult );
1898 }
1899
1900 int index;
1901 for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
1902 classesToMatch = IOServiceMatching(matching_array[index]);
1903 if (classesToMatch == NULL) {
1904 error_report("IOServiceMatching returned NULL for %s",
1905 matching_array[index]);
1906 continue;
1907 }
1908 CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
1909 kCFBooleanTrue);
1910 kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch,
1911 mediaIterator);
1912 if (kernResult != KERN_SUCCESS) {
1913 error_report("Note: IOServiceGetMatchingServices returned %d",
1914 kernResult);
1915 continue;
1916 }
1917
1918 /* If a match was found, leave the loop */
1919 if (*mediaIterator != 0) {
1920 DPRINTF("Matching using %s\n", matching_array[index]);
1921 mediaType = g_strdup(matching_array[index]);
1922 break;
1923 }
1924 }
1925 return mediaType;
1926 }
1927
1928 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
1929 CFIndex maxPathSize, int flags)
1930 {
1931 io_object_t nextMedia;
1932 kern_return_t kernResult = KERN_FAILURE;
1933 *bsdPath = '\0';
1934 nextMedia = IOIteratorNext( mediaIterator );
1935 if ( nextMedia )
1936 {
1937 CFTypeRef bsdPathAsCFString;
1938 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1939 if ( bsdPathAsCFString ) {
1940 size_t devPathLength;
1941 strcpy( bsdPath, _PATH_DEV );
1942 if (flags & BDRV_O_NOCACHE) {
1943 strcat(bsdPath, "r");
1944 }
1945 devPathLength = strlen( bsdPath );
1946 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1947 kernResult = KERN_SUCCESS;
1948 }
1949 CFRelease( bsdPathAsCFString );
1950 }
1951 IOObjectRelease( nextMedia );
1952 }
1953
1954 return kernResult;
1955 }
1956
1957 /* Sets up a real cdrom for use in QEMU */
1958 static bool setup_cdrom(char *bsd_path, Error **errp)
1959 {
1960 int index, num_of_test_partitions = 2, fd;
1961 char test_partition[MAXPATHLEN];
1962 bool partition_found = false;
1963
1964 /* look for a working partition */
1965 for (index = 0; index < num_of_test_partitions; index++) {
1966 snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
1967 index);
1968 fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE);
1969 if (fd >= 0) {
1970 partition_found = true;
1971 qemu_close(fd);
1972 break;
1973 }
1974 }
1975
1976 /* if a working partition on the device was not found */
1977 if (partition_found == false) {
1978 error_setg(errp, "Failed to find a working partition on disc");
1979 } else {
1980 DPRINTF("Using %s as optical disc\n", test_partition);
1981 pstrcpy(bsd_path, MAXPATHLEN, test_partition);
1982 }
1983 return partition_found;
1984 }
1985
1986 /* Prints directions on mounting and unmounting a device */
1987 static void print_unmounting_directions(const char *file_name)
1988 {
1989 error_report("If device %s is mounted on the desktop, unmount"
1990 " it first before using it in QEMU", file_name);
1991 error_report("Command to unmount device: diskutil unmountDisk %s",
1992 file_name);
1993 error_report("Command to mount device: diskutil mountDisk %s", file_name);
1994 }
1995
1996 #endif /* defined(__APPLE__) && defined(__MACH__) */
1997
1998 static int hdev_probe_device(const char *filename)
1999 {
2000 struct stat st;
2001
2002 /* allow a dedicated CD-ROM driver to match with a higher priority */
2003 if (strstart(filename, "/dev/cdrom", NULL))
2004 return 50;
2005
2006 if (stat(filename, &st) >= 0 &&
2007 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
2008 return 100;
2009 }
2010
2011 return 0;
2012 }
2013
2014 static int check_hdev_writable(BDRVRawState *s)
2015 {
2016 #if defined(BLKROGET)
2017 /* Linux block devices can be configured "read-only" using blockdev(8).
2018 * This is independent of device node permissions and therefore open(2)
2019 * with O_RDWR succeeds. Actual writes fail with EPERM.
2020 *
2021 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2022 * check for read-only block devices so that Linux block devices behave
2023 * properly.
2024 */
2025 struct stat st;
2026 int readonly = 0;
2027
2028 if (fstat(s->fd, &st)) {
2029 return -errno;
2030 }
2031
2032 if (!S_ISBLK(st.st_mode)) {
2033 return 0;
2034 }
2035
2036 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
2037 return -errno;
2038 }
2039
2040 if (readonly) {
2041 return -EACCES;
2042 }
2043 #endif /* defined(BLKROGET) */
2044 return 0;
2045 }
2046
2047 static void hdev_parse_filename(const char *filename, QDict *options,
2048 Error **errp)
2049 {
2050 /* The prefix is optional, just as for "file". */
2051 strstart(filename, "host_device:", &filename);
2052
2053 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2054 }
2055
2056 static bool hdev_is_sg(BlockDriverState *bs)
2057 {
2058
2059 #if defined(__linux__)
2060
2061 struct stat st;
2062 struct sg_scsi_id scsiid;
2063 int sg_version;
2064
2065 if (stat(bs->filename, &st) >= 0 && S_ISCHR(st.st_mode) &&
2066 !bdrv_ioctl(bs, SG_GET_VERSION_NUM, &sg_version) &&
2067 !bdrv_ioctl(bs, SG_GET_SCSI_ID, &scsiid)) {
2068 DPRINTF("SG device found: type=%d, version=%d\n",
2069 scsiid.scsi_type, sg_version);
2070 return true;
2071 }
2072
2073 #endif
2074
2075 return false;
2076 }
2077
2078 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
2079 Error **errp)
2080 {
2081 BDRVRawState *s = bs->opaque;
2082 Error *local_err = NULL;
2083 int ret;
2084
2085 #if defined(__APPLE__) && defined(__MACH__)
2086 const char *filename = qdict_get_str(options, "filename");
2087 char bsd_path[MAXPATHLEN] = "";
2088 bool error_occurred = false;
2089
2090 /* If using a real cdrom */
2091 if (strcmp(filename, "/dev/cdrom") == 0) {
2092 char *mediaType = NULL;
2093 kern_return_t ret_val;
2094 io_iterator_t mediaIterator = 0;
2095
2096 mediaType = FindEjectableOpticalMedia(&mediaIterator);
2097 if (mediaType == NULL) {
2098 error_setg(errp, "Please make sure your CD/DVD is in the optical"
2099 " drive");
2100 error_occurred = true;
2101 goto hdev_open_Mac_error;
2102 }
2103
2104 ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
2105 if (ret_val != KERN_SUCCESS) {
2106 error_setg(errp, "Could not get BSD path for optical drive");
2107 error_occurred = true;
2108 goto hdev_open_Mac_error;
2109 }
2110
2111 /* If a real optical drive was not found */
2112 if (bsd_path[0] == '\0') {
2113 error_setg(errp, "Failed to obtain bsd path for optical drive");
2114 error_occurred = true;
2115 goto hdev_open_Mac_error;
2116 }
2117
2118 /* If using a cdrom disc and finding a partition on the disc failed */
2119 if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
2120 setup_cdrom(bsd_path, errp) == false) {
2121 print_unmounting_directions(bsd_path);
2122 error_occurred = true;
2123 goto hdev_open_Mac_error;
2124 }
2125
2126 qdict_put(options, "filename", qstring_from_str(bsd_path));
2127
2128 hdev_open_Mac_error:
2129 g_free(mediaType);
2130 if (mediaIterator) {
2131 IOObjectRelease(mediaIterator);
2132 }
2133 if (error_occurred) {
2134 return -ENOENT;
2135 }
2136 }
2137 #endif /* defined(__APPLE__) && defined(__MACH__) */
2138
2139 s->type = FTYPE_FILE;
2140
2141 ret = raw_open_common(bs, options, flags, 0, &local_err);
2142 if (ret < 0) {
2143 error_propagate(errp, local_err);
2144 #if defined(__APPLE__) && defined(__MACH__)
2145 if (*bsd_path) {
2146 filename = bsd_path;
2147 }
2148 /* if a physical device experienced an error while being opened */
2149 if (strncmp(filename, "/dev/", 5) == 0) {
2150 print_unmounting_directions(filename);
2151 }
2152 #endif /* defined(__APPLE__) && defined(__MACH__) */
2153 return ret;
2154 }
2155
2156 /* Since this does ioctl the device must be already opened */
2157 bs->sg = hdev_is_sg(bs);
2158
2159 if (flags & BDRV_O_RDWR) {
2160 ret = check_hdev_writable(s);
2161 if (ret < 0) {
2162 raw_close(bs);
2163 error_setg_errno(errp, -ret, "The device is not writable");
2164 return ret;
2165 }
2166 }
2167
2168 return ret;
2169 }
2170
2171 #if defined(__linux__)
2172
2173 static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
2174 unsigned long int req, void *buf,
2175 BlockCompletionFunc *cb, void *opaque)
2176 {
2177 BDRVRawState *s = bs->opaque;
2178 RawPosixAIOData *acb;
2179 ThreadPool *pool;
2180
2181 if (fd_open(bs) < 0)
2182 return NULL;
2183
2184 acb = g_new(RawPosixAIOData, 1);
2185 acb->bs = bs;
2186 acb->aio_type = QEMU_AIO_IOCTL;
2187 acb->aio_fildes = s->fd;
2188 acb->aio_offset = 0;
2189 acb->aio_ioctl_buf = buf;
2190 acb->aio_ioctl_cmd = req;
2191 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
2192 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
2193 }
2194 #endif /* linux */
2195
2196 static int fd_open(BlockDriverState *bs)
2197 {
2198 BDRVRawState *s = bs->opaque;
2199
2200 /* this is just to ensure s->fd is sane (its called by io ops) */
2201 if (s->fd >= 0)
2202 return 0;
2203 return -EIO;
2204 }
2205
2206 static coroutine_fn BlockAIOCB *hdev_aio_pdiscard(BlockDriverState *bs,
2207 int64_t offset, int count,
2208 BlockCompletionFunc *cb, void *opaque)
2209 {
2210 BDRVRawState *s = bs->opaque;
2211
2212 if (fd_open(bs) < 0) {
2213 return NULL;
2214 }
2215 return paio_submit(bs, s->fd, offset, NULL, count,
2216 cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2217 }
2218
2219 static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs,
2220 int64_t offset, int count, BdrvRequestFlags flags)
2221 {
2222 BDRVRawState *s = bs->opaque;
2223 int rc;
2224
2225 rc = fd_open(bs);
2226 if (rc < 0) {
2227 return rc;
2228 }
2229 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2230 return paio_submit_co(bs, s->fd, offset, NULL, count,
2231 QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2232 } else if (s->discard_zeroes) {
2233 return paio_submit_co(bs, s->fd, offset, NULL, count,
2234 QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2235 }
2236 return -ENOTSUP;
2237 }
2238
2239 static int hdev_create(const char *filename, QemuOpts *opts,
2240 Error **errp)
2241 {
2242 int fd;
2243 int ret = 0;
2244 struct stat stat_buf;
2245 int64_t total_size = 0;
2246 bool has_prefix;
2247
2248 /* This function is used by both protocol block drivers and therefore either
2249 * of these prefixes may be given.
2250 * The return value has to be stored somewhere, otherwise this is an error
2251 * due to -Werror=unused-value. */
2252 has_prefix =
2253 strstart(filename, "host_device:", &filename) ||
2254 strstart(filename, "host_cdrom:" , &filename);
2255
2256 (void)has_prefix;
2257
2258 ret = raw_normalize_devicepath(&filename);
2259 if (ret < 0) {
2260 error_setg_errno(errp, -ret, "Could not normalize device path");
2261 return ret;
2262 }
2263
2264 /* Read out options */
2265 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2266 BDRV_SECTOR_SIZE);
2267
2268 fd = qemu_open(filename, O_WRONLY | O_BINARY);
2269 if (fd < 0) {
2270 ret = -errno;
2271 error_setg_errno(errp, -ret, "Could not open device");
2272 return ret;
2273 }
2274
2275 if (fstat(fd, &stat_buf) < 0) {
2276 ret = -errno;
2277 error_setg_errno(errp, -ret, "Could not stat device");
2278 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2279 error_setg(errp,
2280 "The given file is neither a block nor a character device");
2281 ret = -ENODEV;
2282 } else if (lseek(fd, 0, SEEK_END) < total_size) {
2283 error_setg(errp, "Device is too small");
2284 ret = -ENOSPC;
2285 }
2286
2287 qemu_close(fd);
2288 return ret;
2289 }
2290
2291 static BlockDriver bdrv_host_device = {
2292 .format_name = "host_device",
2293 .protocol_name = "host_device",
2294 .instance_size = sizeof(BDRVRawState),
2295 .bdrv_needs_filename = true,
2296 .bdrv_probe_device = hdev_probe_device,
2297 .bdrv_parse_filename = hdev_parse_filename,
2298 .bdrv_file_open = hdev_open,
2299 .bdrv_close = raw_close,
2300 .bdrv_reopen_prepare = raw_reopen_prepare,
2301 .bdrv_reopen_commit = raw_reopen_commit,
2302 .bdrv_reopen_abort = raw_reopen_abort,
2303 .bdrv_create = hdev_create,
2304 .create_opts = &raw_create_opts,
2305 .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes,
2306
2307 .bdrv_co_preadv = raw_co_preadv,
2308 .bdrv_co_pwritev = raw_co_pwritev,
2309 .bdrv_aio_flush = raw_aio_flush,
2310 .bdrv_aio_pdiscard = hdev_aio_pdiscard,
2311 .bdrv_refresh_limits = raw_refresh_limits,
2312 .bdrv_io_plug = raw_aio_plug,
2313 .bdrv_io_unplug = raw_aio_unplug,
2314
2315 .bdrv_truncate = raw_truncate,
2316 .bdrv_getlength = raw_getlength,
2317 .bdrv_get_info = raw_get_info,
2318 .bdrv_get_allocated_file_size
2319 = raw_get_allocated_file_size,
2320 .bdrv_probe_blocksizes = hdev_probe_blocksizes,
2321 .bdrv_probe_geometry = hdev_probe_geometry,
2322
2323 /* generic scsi device */
2324 #ifdef __linux__
2325 .bdrv_aio_ioctl = hdev_aio_ioctl,
2326 #endif
2327 };
2328
2329 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2330 static void cdrom_parse_filename(const char *filename, QDict *options,
2331 Error **errp)
2332 {
2333 /* The prefix is optional, just as for "file". */
2334 strstart(filename, "host_cdrom:", &filename);
2335
2336 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2337 }
2338 #endif
2339
2340 #ifdef __linux__
2341 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2342 Error **errp)
2343 {
2344 BDRVRawState *s = bs->opaque;
2345
2346 s->type = FTYPE_CD;
2347
2348 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2349 return raw_open_common(bs, options, flags, O_NONBLOCK, errp);
2350 }
2351
2352 static int cdrom_probe_device(const char *filename)
2353 {
2354 int fd, ret;
2355 int prio = 0;
2356 struct stat st;
2357
2358 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2359 if (fd < 0) {
2360 goto out;
2361 }
2362 ret = fstat(fd, &st);
2363 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2364 goto outc;
2365 }
2366
2367 /* Attempt to detect via a CDROM specific ioctl */
2368 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2369 if (ret >= 0)
2370 prio = 100;
2371
2372 outc:
2373 qemu_close(fd);
2374 out:
2375 return prio;
2376 }
2377
2378 static bool cdrom_is_inserted(BlockDriverState *bs)
2379 {
2380 BDRVRawState *s = bs->opaque;
2381 int ret;
2382
2383 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2384 return ret == CDS_DISC_OK;
2385 }
2386
2387 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2388 {
2389 BDRVRawState *s = bs->opaque;
2390
2391 if (eject_flag) {
2392 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2393 perror("CDROMEJECT");
2394 } else {
2395 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2396 perror("CDROMEJECT");
2397 }
2398 }
2399
2400 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2401 {
2402 BDRVRawState *s = bs->opaque;
2403
2404 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2405 /*
2406 * Note: an error can happen if the distribution automatically
2407 * mounts the CD-ROM
2408 */
2409 /* perror("CDROM_LOCKDOOR"); */
2410 }
2411 }
2412
2413 static BlockDriver bdrv_host_cdrom = {
2414 .format_name = "host_cdrom",
2415 .protocol_name = "host_cdrom",
2416 .instance_size = sizeof(BDRVRawState),
2417 .bdrv_needs_filename = true,
2418 .bdrv_probe_device = cdrom_probe_device,
2419 .bdrv_parse_filename = cdrom_parse_filename,
2420 .bdrv_file_open = cdrom_open,
2421 .bdrv_close = raw_close,
2422 .bdrv_reopen_prepare = raw_reopen_prepare,
2423 .bdrv_reopen_commit = raw_reopen_commit,
2424 .bdrv_reopen_abort = raw_reopen_abort,
2425 .bdrv_create = hdev_create,
2426 .create_opts = &raw_create_opts,
2427
2428
2429 .bdrv_co_preadv = raw_co_preadv,
2430 .bdrv_co_pwritev = raw_co_pwritev,
2431 .bdrv_aio_flush = raw_aio_flush,
2432 .bdrv_refresh_limits = raw_refresh_limits,
2433 .bdrv_io_plug = raw_aio_plug,
2434 .bdrv_io_unplug = raw_aio_unplug,
2435
2436 .bdrv_truncate = raw_truncate,
2437 .bdrv_getlength = raw_getlength,
2438 .has_variable_length = true,
2439 .bdrv_get_allocated_file_size
2440 = raw_get_allocated_file_size,
2441
2442 /* removable device support */
2443 .bdrv_is_inserted = cdrom_is_inserted,
2444 .bdrv_eject = cdrom_eject,
2445 .bdrv_lock_medium = cdrom_lock_medium,
2446
2447 /* generic scsi device */
2448 .bdrv_aio_ioctl = hdev_aio_ioctl,
2449 };
2450 #endif /* __linux__ */
2451
2452 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2453 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2454 Error **errp)
2455 {
2456 BDRVRawState *s = bs->opaque;
2457 Error *local_err = NULL;
2458 int ret;
2459
2460 s->type = FTYPE_CD;
2461
2462 ret = raw_open_common(bs, options, flags, 0, &local_err);
2463 if (ret) {
2464 error_propagate(errp, local_err);
2465 return ret;
2466 }
2467
2468 /* make sure the door isn't locked at this time */
2469 ioctl(s->fd, CDIOCALLOW);
2470 return 0;
2471 }
2472
2473 static int cdrom_probe_device(const char *filename)
2474 {
2475 if (strstart(filename, "/dev/cd", NULL) ||
2476 strstart(filename, "/dev/acd", NULL))
2477 return 100;
2478 return 0;
2479 }
2480
2481 static int cdrom_reopen(BlockDriverState *bs)
2482 {
2483 BDRVRawState *s = bs->opaque;
2484 int fd;
2485
2486 /*
2487 * Force reread of possibly changed/newly loaded disc,
2488 * FreeBSD seems to not notice sometimes...
2489 */
2490 if (s->fd >= 0)
2491 qemu_close(s->fd);
2492 fd = qemu_open(bs->filename, s->open_flags, 0644);
2493 if (fd < 0) {
2494 s->fd = -1;
2495 return -EIO;
2496 }
2497 s->fd = fd;
2498
2499 /* make sure the door isn't locked at this time */
2500 ioctl(s->fd, CDIOCALLOW);
2501 return 0;
2502 }
2503
2504 static bool cdrom_is_inserted(BlockDriverState *bs)
2505 {
2506 return raw_getlength(bs) > 0;
2507 }
2508
2509 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2510 {
2511 BDRVRawState *s = bs->opaque;
2512
2513 if (s->fd < 0)
2514 return;
2515
2516 (void) ioctl(s->fd, CDIOCALLOW);
2517
2518 if (eject_flag) {
2519 if (ioctl(s->fd, CDIOCEJECT) < 0)
2520 perror("CDIOCEJECT");
2521 } else {
2522 if (ioctl(s->fd, CDIOCCLOSE) < 0)
2523 perror("CDIOCCLOSE");
2524 }
2525
2526 cdrom_reopen(bs);
2527 }
2528
2529 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2530 {
2531 BDRVRawState *s = bs->opaque;
2532
2533 if (s->fd < 0)
2534 return;
2535 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2536 /*
2537 * Note: an error can happen if the distribution automatically
2538 * mounts the CD-ROM
2539 */
2540 /* perror("CDROM_LOCKDOOR"); */
2541 }
2542 }
2543
2544 static BlockDriver bdrv_host_cdrom = {
2545 .format_name = "host_cdrom",
2546 .protocol_name = "host_cdrom",
2547 .instance_size = sizeof(BDRVRawState),
2548 .bdrv_needs_filename = true,
2549 .bdrv_probe_device = cdrom_probe_device,
2550 .bdrv_parse_filename = cdrom_parse_filename,
2551 .bdrv_file_open = cdrom_open,
2552 .bdrv_close = raw_close,
2553 .bdrv_reopen_prepare = raw_reopen_prepare,
2554 .bdrv_reopen_commit = raw_reopen_commit,
2555 .bdrv_reopen_abort = raw_reopen_abort,
2556 .bdrv_create = hdev_create,
2557 .create_opts = &raw_create_opts,
2558
2559 .bdrv_co_preadv = raw_co_preadv,
2560 .bdrv_co_pwritev = raw_co_pwritev,
2561 .bdrv_aio_flush = raw_aio_flush,
2562 .bdrv_refresh_limits = raw_refresh_limits,
2563 .bdrv_io_plug = raw_aio_plug,
2564 .bdrv_io_unplug = raw_aio_unplug,
2565
2566 .bdrv_truncate = raw_truncate,
2567 .bdrv_getlength = raw_getlength,
2568 .has_variable_length = true,
2569 .bdrv_get_allocated_file_size
2570 = raw_get_allocated_file_size,
2571
2572 /* removable device support */
2573 .bdrv_is_inserted = cdrom_is_inserted,
2574 .bdrv_eject = cdrom_eject,
2575 .bdrv_lock_medium = cdrom_lock_medium,
2576 };
2577 #endif /* __FreeBSD__ */
2578
2579 static void bdrv_file_init(void)
2580 {
2581 /*
2582 * Register all the drivers. Note that order is important, the driver
2583 * registered last will get probed first.
2584 */
2585 bdrv_register(&bdrv_file);
2586 bdrv_register(&bdrv_host_device);
2587 #ifdef __linux__
2588 bdrv_register(&bdrv_host_cdrom);
2589 #endif
2590 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2591 bdrv_register(&bdrv_host_cdrom);
2592 #endif
2593 }
2594
2595 block_init(bdrv_file_init);
Random patches