?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
hw/arm/virt: no ITS on older machine types
[qemu/kevin.git] / block / raw-posix.c
blob166e9d1ad5dd59211c12de4b04b140411fd3f173
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 use_linux_aio:1;
147 bool has_fallocate;
148 bool needs_alignment;
149 } BDRVRawState;
150
151 typedef struct BDRVRawReopenState {
152 int fd;
153 int open_flags;
154 } BDRVRawReopenState;
155
156 static int fd_open(BlockDriverState *bs);
157 static int64_t raw_getlength(BlockDriverState *bs);
158
159 typedef struct RawPosixAIOData {
160 BlockDriverState *bs;
161 int aio_fildes;
162 union {
163 struct iovec *aio_iov;
164 void *aio_ioctl_buf;
165 };
166 int aio_niov;
167 uint64_t aio_nbytes;
168 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
169 off_t aio_offset;
170 int aio_type;
171 } RawPosixAIOData;
172
173 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
174 static int cdrom_reopen(BlockDriverState *bs);
175 #endif
176
177 #if defined(__NetBSD__)
178 static int raw_normalize_devicepath(const char **filename)
179 {
180 static char namebuf[PATH_MAX];
181 const char *dp, *fname;
182 struct stat sb;
183
184 fname = *filename;
185 dp = strrchr(fname, '/');
186 if (lstat(fname, &sb) < 0) {
187 fprintf(stderr, "%s: stat failed: %s\n",
188 fname, strerror(errno));
189 return -errno;
190 }
191
192 if (!S_ISBLK(sb.st_mode)) {
193 return 0;
194 }
195
196 if (dp == NULL) {
197 snprintf(namebuf, PATH_MAX, "r%s", fname);
198 } else {
199 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
200 (int)(dp - fname), fname, dp + 1);
201 }
202 fprintf(stderr, "%s is a block device", fname);
203 *filename = namebuf;
204 fprintf(stderr, ", using %s\n", *filename);
205
206 return 0;
207 }
208 #else
209 static int raw_normalize_devicepath(const char **filename)
210 {
211 return 0;
212 }
213 #endif
214
215 /*
216 * Get logical block size via ioctl. On success store it in @sector_size_p.
217 */
218 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
219 {
220 unsigned int sector_size;
221 bool success = false;
222
223 errno = ENOTSUP;
224
225 /* Try a few ioctls to get the right size */
226 #ifdef BLKSSZGET
227 if (ioctl(fd, BLKSSZGET, &sector_size) >= 0) {
228 *sector_size_p = sector_size;
229 success = true;
230 }
231 #endif
232 #ifdef DKIOCGETBLOCKSIZE
233 if (ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) >= 0) {
234 *sector_size_p = sector_size;
235 success = true;
236 }
237 #endif
238 #ifdef DIOCGSECTORSIZE
239 if (ioctl(fd, DIOCGSECTORSIZE, &sector_size) >= 0) {
240 *sector_size_p = sector_size;
241 success = true;
242 }
243 #endif
244
245 return success ? 0 : -errno;
246 }
247
248 /**
249 * Get physical block size of @fd.
250 * On success, store it in @blk_size and return 0.
251 * On failure, return -errno.
252 */
253 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
254 {
255 #ifdef BLKPBSZGET
256 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
257 return -errno;
258 }
259 return 0;
260 #else
261 return -ENOTSUP;
262 #endif
263 }
264
265 /* Check if read is allowed with given memory buffer and length.
266 *
267 * This function is used to check O_DIRECT memory buffer and request alignment.
268 */
269 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
270 {
271 ssize_t ret = pread(fd, buf, len, 0);
272
273 if (ret >= 0) {
274 return true;
275 }
276
277 #ifdef __linux__
278 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
279 * other errors (e.g. real I/O error), which could happen on a failed
280 * drive, since we only care about probing alignment.
281 */
282 if (errno != EINVAL) {
283 return true;
284 }
285 #endif
286
287 return false;
288 }
289
290 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
291 {
292 BDRVRawState *s = bs->opaque;
293 char *buf;
294 size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize());
295
296 /* For SCSI generic devices the alignment is not really used.
297 With buffered I/O, we don't have any restrictions. */
298 if (bdrv_is_sg(bs) || !s->needs_alignment) {
299 bs->bl.request_alignment = 1;
300 s->buf_align = 1;
301 return;
302 }
303
304 bs->bl.request_alignment = 0;
305 s->buf_align = 0;
306 /* Let's try to use the logical blocksize for the alignment. */
307 if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) {
308 bs->bl.request_alignment = 0;
309 }
310 #ifdef CONFIG_XFS
311 if (s->is_xfs) {
312 struct dioattr da;
313 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
314 bs->bl.request_alignment = da.d_miniosz;
315 /* The kernel returns wrong information for d_mem */
316 /* s->buf_align = da.d_mem; */
317 }
318 }
319 #endif
320
321 /* If we could not get the sizes so far, we can only guess them */
322 if (!s->buf_align) {
323 size_t align;
324 buf = qemu_memalign(max_align, 2 * max_align);
325 for (align = 512; align <= max_align; align <<= 1) {
326 if (raw_is_io_aligned(fd, buf + align, max_align)) {
327 s->buf_align = align;
328 break;
329 }
330 }
331 qemu_vfree(buf);
332 }
333
334 if (!bs->bl.request_alignment) {
335 size_t align;
336 buf = qemu_memalign(s->buf_align, max_align);
337 for (align = 512; align <= max_align; align <<= 1) {
338 if (raw_is_io_aligned(fd, buf, align)) {
339 bs->bl.request_alignment = align;
340 break;
341 }
342 }
343 qemu_vfree(buf);
344 }
345
346 if (!s->buf_align || !bs->bl.request_alignment) {
347 error_setg(errp, "Could not find working O_DIRECT alignment");
348 error_append_hint(errp, "Try cache.direct=off\n");
349 }
350 }
351
352 static void raw_parse_flags(int bdrv_flags, int *open_flags)
353 {
354 assert(open_flags != NULL);
355
356 *open_flags |= O_BINARY;
357 *open_flags &= ~O_ACCMODE;
358 if (bdrv_flags & BDRV_O_RDWR) {
359 *open_flags |= O_RDWR;
360 } else {
361 *open_flags |= O_RDONLY;
362 }
363
364 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
365 * and O_DIRECT for no caching. */
366 if ((bdrv_flags & BDRV_O_NOCACHE)) {
367 *open_flags |= O_DIRECT;
368 }
369 }
370
371 static void raw_parse_filename(const char *filename, QDict *options,
372 Error **errp)
373 {
374 /* The filename does not have to be prefixed by the protocol name, since
375 * "file" is the default protocol; therefore, the return value of this
376 * function call can be ignored. */
377 strstart(filename, "file:", &filename);
378
379 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
380 }
381
382 static QemuOptsList raw_runtime_opts = {
383 .name = "raw",
384 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
385 .desc = {
386 {
387 .name = "filename",
388 .type = QEMU_OPT_STRING,
389 .help = "File name of the image",
390 },
391 {
392 .name = "aio",
393 .type = QEMU_OPT_STRING,
394 .help = "host AIO implementation (threads, native)",
395 },
396 { /* end of list */ }
397 },
398 };
399
400 static int raw_open_common(BlockDriverState *bs, QDict *options,
401 int bdrv_flags, int open_flags, Error **errp)
402 {
403 BDRVRawState *s = bs->opaque;
404 QemuOpts *opts;
405 Error *local_err = NULL;
406 const char *filename = NULL;
407 BlockdevAioOptions aio, aio_default;
408 int fd, ret;
409 struct stat st;
410
411 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
412 qemu_opts_absorb_qdict(opts, options, &local_err);
413 if (local_err) {
414 error_propagate(errp, local_err);
415 ret = -EINVAL;
416 goto fail;
417 }
418
419 filename = qemu_opt_get(opts, "filename");
420
421 ret = raw_normalize_devicepath(&filename);
422 if (ret != 0) {
423 error_setg_errno(errp, -ret, "Could not normalize device path");
424 goto fail;
425 }
426
427 aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO)
428 ? BLOCKDEV_AIO_OPTIONS_NATIVE
429 : BLOCKDEV_AIO_OPTIONS_THREADS;
430 aio = qapi_enum_parse(BlockdevAioOptions_lookup, qemu_opt_get(opts, "aio"),
431 BLOCKDEV_AIO_OPTIONS__MAX, aio_default, &local_err);
432 if (local_err) {
433 error_propagate(errp, local_err);
434 ret = -EINVAL;
435 goto fail;
436 }
437 s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE);
438
439 s->open_flags = open_flags;
440 raw_parse_flags(bdrv_flags, &s->open_flags);
441
442 s->fd = -1;
443 fd = qemu_open(filename, s->open_flags, 0644);
444 if (fd < 0) {
445 ret = -errno;
446 if (ret == -EROFS) {
447 ret = -EACCES;
448 }
449 goto fail;
450 }
451 s->fd = fd;
452
453 #ifdef CONFIG_LINUX_AIO
454 /* Currently Linux does AIO only for files opened with O_DIRECT */
455 if (s->use_linux_aio && !(s->open_flags & O_DIRECT)) {
456 error_setg(errp, "aio=native was specified, but it requires "
457 "cache.direct=on, which was not specified.");
458 ret = -EINVAL;
459 goto fail;
460 }
461 #else
462 if (s->use_linux_aio) {
463 error_setg(errp, "aio=native was specified, but is not supported "
464 "in this build.");
465 ret = -EINVAL;
466 goto fail;
467 }
468 #endif /* !defined(CONFIG_LINUX_AIO) */
469
470 s->has_discard = true;
471 s->has_write_zeroes = true;
472 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
473 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
474 s->needs_alignment = true;
475 }
476
477 if (fstat(s->fd, &st) < 0) {
478 ret = -errno;
479 error_setg_errno(errp, errno, "Could not stat file");
480 goto fail;
481 }
482 if (S_ISREG(st.st_mode)) {
483 s->discard_zeroes = true;
484 s->has_fallocate = true;
485 }
486 if (S_ISBLK(st.st_mode)) {
487 #ifdef BLKDISCARDZEROES
488 unsigned int arg;
489 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
490 s->discard_zeroes = true;
491 }
492 #endif
493 #ifdef __linux__
494 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
495 * not rely on the contents of discarded blocks unless using O_DIRECT.
496 * Same for BLKZEROOUT.
497 */
498 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
499 s->discard_zeroes = false;
500 s->has_write_zeroes = false;
501 }
502 #endif
503 }
504 #ifdef __FreeBSD__
505 if (S_ISCHR(st.st_mode)) {
506 /*
507 * The file is a char device (disk), which on FreeBSD isn't behind
508 * a pager, so force all requests to be aligned. This is needed
509 * so QEMU makes sure all IO operations on the device are aligned
510 * to sector size, or else FreeBSD will reject them with EINVAL.
511 */
512 s->needs_alignment = true;
513 }
514 #endif
515
516 #ifdef CONFIG_XFS
517 if (platform_test_xfs_fd(s->fd)) {
518 s->is_xfs = true;
519 }
520 #endif
521
522 ret = 0;
523 fail:
524 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
525 unlink(filename);
526 }
527 qemu_opts_del(opts);
528 return ret;
529 }
530
531 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
532 Error **errp)
533 {
534 BDRVRawState *s = bs->opaque;
535
536 s->type = FTYPE_FILE;
537 return raw_open_common(bs, options, flags, 0, errp);
538 }
539
540 static int raw_reopen_prepare(BDRVReopenState *state,
541 BlockReopenQueue *queue, Error **errp)
542 {
543 BDRVRawState *s;
544 BDRVRawReopenState *raw_s;
545 int ret = 0;
546 Error *local_err = NULL;
547
548 assert(state != NULL);
549 assert(state->bs != NULL);
550
551 s = state->bs->opaque;
552
553 state->opaque = g_new0(BDRVRawReopenState, 1);
554 raw_s = state->opaque;
555
556 if (s->type == FTYPE_CD) {
557 raw_s->open_flags |= O_NONBLOCK;
558 }
559
560 raw_parse_flags(state->flags, &raw_s->open_flags);
561
562 raw_s->fd = -1;
563
564 int fcntl_flags = O_APPEND | O_NONBLOCK;
565 #ifdef O_NOATIME
566 fcntl_flags |= O_NOATIME;
567 #endif
568
569 #ifdef O_ASYNC
570 /* Not all operating systems have O_ASYNC, and those that don't
571 * will not let us track the state into raw_s->open_flags (typically
572 * you achieve the same effect with an ioctl, for example I_SETSIG
573 * on Solaris). But we do not use O_ASYNC, so that's fine.
574 */
575 assert((s->open_flags & O_ASYNC) == 0);
576 #endif
577
578 if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
579 /* dup the original fd */
580 raw_s->fd = qemu_dup(s->fd);
581 if (raw_s->fd >= 0) {
582 ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
583 if (ret) {
584 qemu_close(raw_s->fd);
585 raw_s->fd = -1;
586 }
587 }
588 }
589
590 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
591 if (raw_s->fd == -1) {
592 const char *normalized_filename = state->bs->filename;
593 ret = raw_normalize_devicepath(&normalized_filename);
594 if (ret < 0) {
595 error_setg_errno(errp, -ret, "Could not normalize device path");
596 } else {
597 assert(!(raw_s->open_flags & O_CREAT));
598 raw_s->fd = qemu_open(normalized_filename, raw_s->open_flags);
599 if (raw_s->fd == -1) {
600 error_setg_errno(errp, errno, "Could not reopen file");
601 ret = -1;
602 }
603 }
604 }
605
606 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
607 * alignment with the new fd. */
608 if (raw_s->fd != -1) {
609 raw_probe_alignment(state->bs, raw_s->fd, &local_err);
610 if (local_err) {
611 qemu_close(raw_s->fd);
612 raw_s->fd = -1;
613 error_propagate(errp, local_err);
614 ret = -EINVAL;
615 }
616 }
617
618 return ret;
619 }
620
621 static void raw_reopen_commit(BDRVReopenState *state)
622 {
623 BDRVRawReopenState *raw_s = state->opaque;
624 BDRVRawState *s = state->bs->opaque;
625
626 s->open_flags = raw_s->open_flags;
627
628 qemu_close(s->fd);
629 s->fd = raw_s->fd;
630
631 g_free(state->opaque);
632 state->opaque = NULL;
633 }
634
635
636 static void raw_reopen_abort(BDRVReopenState *state)
637 {
638 BDRVRawReopenState *raw_s = state->opaque;
639
640 /* nothing to do if NULL, we didn't get far enough */
641 if (raw_s == NULL) {
642 return;
643 }
644
645 if (raw_s->fd >= 0) {
646 qemu_close(raw_s->fd);
647 raw_s->fd = -1;
648 }
649 g_free(state->opaque);
650 state->opaque = NULL;
651 }
652
653 static int hdev_get_max_transfer_length(int fd)
654 {
655 #ifdef BLKSECTGET
656 int max_sectors = 0;
657 if (ioctl(fd, BLKSECTGET, &max_sectors) == 0) {
658 return max_sectors;
659 } else {
660 return -errno;
661 }
662 #else
663 return -ENOSYS;
664 #endif
665 }
666
667 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
668 {
669 BDRVRawState *s = bs->opaque;
670 struct stat st;
671
672 if (!fstat(s->fd, &st)) {
673 if (S_ISBLK(st.st_mode)) {
674 int ret = hdev_get_max_transfer_length(s->fd);
675 if (ret > 0 && ret <= BDRV_REQUEST_MAX_SECTORS) {
676 bs->bl.max_transfer = pow2floor(ret << BDRV_SECTOR_BITS);
677 }
678 }
679 }
680
681 raw_probe_alignment(bs, s->fd, errp);
682 bs->bl.min_mem_alignment = s->buf_align;
683 bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
684 }
685
686 static int check_for_dasd(int fd)
687 {
688 #ifdef BIODASDINFO2
689 struct dasd_information2_t info = {0};
690
691 return ioctl(fd, BIODASDINFO2, &info);
692 #else
693 return -1;
694 #endif
695 }
696
697 /**
698 * Try to get @bs's logical and physical block size.
699 * On success, store them in @bsz and return zero.
700 * On failure, return negative errno.
701 */
702 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
703 {
704 BDRVRawState *s = bs->opaque;
705 int ret;
706
707 /* If DASD, get blocksizes */
708 if (check_for_dasd(s->fd) < 0) {
709 return -ENOTSUP;
710 }
711 ret = probe_logical_blocksize(s->fd, &bsz->log);
712 if (ret < 0) {
713 return ret;
714 }
715 return probe_physical_blocksize(s->fd, &bsz->phys);
716 }
717
718 /**
719 * Try to get @bs's geometry: cyls, heads, sectors.
720 * On success, store them in @geo and return 0.
721 * On failure return -errno.
722 * (Allows block driver to assign default geometry values that guest sees)
723 */
724 #ifdef __linux__
725 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
726 {
727 BDRVRawState *s = bs->opaque;
728 struct hd_geometry ioctl_geo = {0};
729
730 /* If DASD, get its geometry */
731 if (check_for_dasd(s->fd) < 0) {
732 return -ENOTSUP;
733 }
734 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
735 return -errno;
736 }
737 /* HDIO_GETGEO may return success even though geo contains zeros
738 (e.g. certain multipath setups) */
739 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
740 return -ENOTSUP;
741 }
742 /* Do not return a geometry for partition */
743 if (ioctl_geo.start != 0) {
744 return -ENOTSUP;
745 }
746 geo->heads = ioctl_geo.heads;
747 geo->sectors = ioctl_geo.sectors;
748 geo->cylinders = ioctl_geo.cylinders;
749
750 return 0;
751 }
752 #else /* __linux__ */
753 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
754 {
755 return -ENOTSUP;
756 }
757 #endif
758
759 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
760 {
761 int ret;
762
763 ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
764 if (ret == -1) {
765 return -errno;
766 }
767
768 return 0;
769 }
770
771 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
772 {
773 int ret;
774
775 ret = qemu_fdatasync(aiocb->aio_fildes);
776 if (ret == -1) {
777 return -errno;
778 }
779 return 0;
780 }
781
782 #ifdef CONFIG_PREADV
783
784 static bool preadv_present = true;
785
786 static ssize_t
787 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
788 {
789 return preadv(fd, iov, nr_iov, offset);
790 }
791
792 static ssize_t
793 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
794 {
795 return pwritev(fd, iov, nr_iov, offset);
796 }
797
798 #else
799
800 static bool preadv_present = false;
801
802 static ssize_t
803 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
804 {
805 return -ENOSYS;
806 }
807
808 static ssize_t
809 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
810 {
811 return -ENOSYS;
812 }
813
814 #endif
815
816 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
817 {
818 ssize_t len;
819
820 do {
821 if (aiocb->aio_type & QEMU_AIO_WRITE)
822 len = qemu_pwritev(aiocb->aio_fildes,
823 aiocb->aio_iov,
824 aiocb->aio_niov,
825 aiocb->aio_offset);
826 else
827 len = qemu_preadv(aiocb->aio_fildes,
828 aiocb->aio_iov,
829 aiocb->aio_niov,
830 aiocb->aio_offset);
831 } while (len == -1 && errno == EINTR);
832
833 if (len == -1) {
834 return -errno;
835 }
836 return len;
837 }
838
839 /*
840 * Read/writes the data to/from a given linear buffer.
841 *
842 * Returns the number of bytes handles or -errno in case of an error. Short
843 * reads are only returned if the end of the file is reached.
844 */
845 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
846 {
847 ssize_t offset = 0;
848 ssize_t len;
849
850 while (offset < aiocb->aio_nbytes) {
851 if (aiocb->aio_type & QEMU_AIO_WRITE) {
852 len = pwrite(aiocb->aio_fildes,
853 (const char *)buf + offset,
854 aiocb->aio_nbytes - offset,
855 aiocb->aio_offset + offset);
856 } else {
857 len = pread(aiocb->aio_fildes,
858 buf + offset,
859 aiocb->aio_nbytes - offset,
860 aiocb->aio_offset + offset);
861 }
862 if (len == -1 && errno == EINTR) {
863 continue;
864 } else if (len == -1 && errno == EINVAL &&
865 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
866 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
867 offset > 0) {
868 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
869 * after a short read. Assume that O_DIRECT short reads only occur
870 * at EOF. Therefore this is a short read, not an I/O error.
871 */
872 break;
873 } else if (len == -1) {
874 offset = -errno;
875 break;
876 } else if (len == 0) {
877 break;
878 }
879 offset += len;
880 }
881
882 return offset;
883 }
884
885 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
886 {
887 ssize_t nbytes;
888 char *buf;
889
890 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
891 /*
892 * If there is just a single buffer, and it is properly aligned
893 * we can just use plain pread/pwrite without any problems.
894 */
895 if (aiocb->aio_niov == 1) {
896 return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
897 }
898 /*
899 * We have more than one iovec, and all are properly aligned.
900 *
901 * Try preadv/pwritev first and fall back to linearizing the
902 * buffer if it's not supported.
903 */
904 if (preadv_present) {
905 nbytes = handle_aiocb_rw_vector(aiocb);
906 if (nbytes == aiocb->aio_nbytes ||
907 (nbytes < 0 && nbytes != -ENOSYS)) {
908 return nbytes;
909 }
910 preadv_present = false;
911 }
912
913 /*
914 * XXX(hch): short read/write. no easy way to handle the reminder
915 * using these interfaces. For now retry using plain
916 * pread/pwrite?
917 */
918 }
919
920 /*
921 * Ok, we have to do it the hard way, copy all segments into
922 * a single aligned buffer.
923 */
924 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
925 if (buf == NULL) {
926 return -ENOMEM;
927 }
928
929 if (aiocb->aio_type & QEMU_AIO_WRITE) {
930 char *p = buf;
931 int i;
932
933 for (i = 0; i < aiocb->aio_niov; ++i) {
934 memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
935 p += aiocb->aio_iov[i].iov_len;
936 }
937 assert(p - buf == aiocb->aio_nbytes);
938 }
939
940 nbytes = handle_aiocb_rw_linear(aiocb, buf);
941 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
942 char *p = buf;
943 size_t count = aiocb->aio_nbytes, copy;
944 int i;
945
946 for (i = 0; i < aiocb->aio_niov && count; ++i) {
947 copy = count;
948 if (copy > aiocb->aio_iov[i].iov_len) {
949 copy = aiocb->aio_iov[i].iov_len;
950 }
951 memcpy(aiocb->aio_iov[i].iov_base, p, copy);
952 assert(count >= copy);
953 p += copy;
954 count -= copy;
955 }
956 assert(count == 0);
957 }
958 qemu_vfree(buf);
959
960 return nbytes;
961 }
962
963 #ifdef CONFIG_XFS
964 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
965 {
966 struct xfs_flock64 fl;
967 int err;
968
969 memset(&fl, 0, sizeof(fl));
970 fl.l_whence = SEEK_SET;
971 fl.l_start = offset;
972 fl.l_len = bytes;
973
974 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
975 err = errno;
976 DPRINTF("cannot write zero range (%s)\n", strerror(errno));
977 return -err;
978 }
979
980 return 0;
981 }
982
983 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
984 {
985 struct xfs_flock64 fl;
986 int err;
987
988 memset(&fl, 0, sizeof(fl));
989 fl.l_whence = SEEK_SET;
990 fl.l_start = offset;
991 fl.l_len = bytes;
992
993 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
994 err = errno;
995 DPRINTF("cannot punch hole (%s)\n", strerror(errno));
996 return -err;
997 }
998
999 return 0;
1000 }
1001 #endif
1002
1003 static int translate_err(int err)
1004 {
1005 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1006 err == -ENOTTY) {
1007 err = -ENOTSUP;
1008 }
1009 return err;
1010 }
1011
1012 #ifdef CONFIG_FALLOCATE
1013 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1014 {
1015 do {
1016 if (fallocate(fd, mode, offset, len) == 0) {
1017 return 0;
1018 }
1019 } while (errno == EINTR);
1020 return translate_err(-errno);
1021 }
1022 #endif
1023
1024 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1025 {
1026 int ret = -ENOTSUP;
1027 BDRVRawState *s = aiocb->bs->opaque;
1028
1029 if (!s->has_write_zeroes) {
1030 return -ENOTSUP;
1031 }
1032
1033 #ifdef BLKZEROOUT
1034 do {
1035 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1036 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1037 return 0;
1038 }
1039 } while (errno == EINTR);
1040
1041 ret = translate_err(-errno);
1042 #endif
1043
1044 if (ret == -ENOTSUP) {
1045 s->has_write_zeroes = false;
1046 }
1047 return ret;
1048 }
1049
1050 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
1051 {
1052 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1053 BDRVRawState *s = aiocb->bs->opaque;
1054 #endif
1055
1056 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1057 return handle_aiocb_write_zeroes_block(aiocb);
1058 }
1059
1060 #ifdef CONFIG_XFS
1061 if (s->is_xfs) {
1062 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
1063 }
1064 #endif
1065
1066 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1067 if (s->has_write_zeroes) {
1068 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1069 aiocb->aio_offset, aiocb->aio_nbytes);
1070 if (ret == 0 || ret != -ENOTSUP) {
1071 return ret;
1072 }
1073 s->has_write_zeroes = false;
1074 }
1075 #endif
1076
1077 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1078 if (s->has_discard && s->has_fallocate) {
1079 int ret = do_fallocate(s->fd,
1080 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1081 aiocb->aio_offset, aiocb->aio_nbytes);
1082 if (ret == 0) {
1083 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1084 if (ret == 0 || ret != -ENOTSUP) {
1085 return ret;
1086 }
1087 s->has_fallocate = false;
1088 } else if (ret != -ENOTSUP) {
1089 return ret;
1090 } else {
1091 s->has_discard = false;
1092 }
1093 }
1094 #endif
1095
1096 #ifdef CONFIG_FALLOCATE
1097 if (s->has_fallocate && aiocb->aio_offset >= bdrv_getlength(aiocb->bs)) {
1098 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1099 if (ret == 0 || ret != -ENOTSUP) {
1100 return ret;
1101 }
1102 s->has_fallocate = false;
1103 }
1104 #endif
1105
1106 return -ENOTSUP;
1107 }
1108
1109 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
1110 {
1111 int ret = -EOPNOTSUPP;
1112 BDRVRawState *s = aiocb->bs->opaque;
1113
1114 if (!s->has_discard) {
1115 return -ENOTSUP;
1116 }
1117
1118 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1119 #ifdef BLKDISCARD
1120 do {
1121 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1122 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1123 return 0;
1124 }
1125 } while (errno == EINTR);
1126
1127 ret = -errno;
1128 #endif
1129 } else {
1130 #ifdef CONFIG_XFS
1131 if (s->is_xfs) {
1132 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1133 }
1134 #endif
1135
1136 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1137 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1138 aiocb->aio_offset, aiocb->aio_nbytes);
1139 #endif
1140 }
1141
1142 ret = translate_err(ret);
1143 if (ret == -ENOTSUP) {
1144 s->has_discard = false;
1145 }
1146 return ret;
1147 }
1148
1149 static int aio_worker(void *arg)
1150 {
1151 RawPosixAIOData *aiocb = arg;
1152 ssize_t ret = 0;
1153
1154 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
1155 case QEMU_AIO_READ:
1156 ret = handle_aiocb_rw(aiocb);
1157 if (ret >= 0 && ret < aiocb->aio_nbytes) {
1158 iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
1159 0, aiocb->aio_nbytes - ret);
1160
1161 ret = aiocb->aio_nbytes;
1162 }
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_WRITE:
1170 ret = handle_aiocb_rw(aiocb);
1171 if (ret == aiocb->aio_nbytes) {
1172 ret = 0;
1173 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1174 ret = -EINVAL;
1175 }
1176 break;
1177 case QEMU_AIO_FLUSH:
1178 ret = handle_aiocb_flush(aiocb);
1179 break;
1180 case QEMU_AIO_IOCTL:
1181 ret = handle_aiocb_ioctl(aiocb);
1182 break;
1183 case QEMU_AIO_DISCARD:
1184 ret = handle_aiocb_discard(aiocb);
1185 break;
1186 case QEMU_AIO_WRITE_ZEROES:
1187 ret = handle_aiocb_write_zeroes(aiocb);
1188 break;
1189 default:
1190 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1191 ret = -EINVAL;
1192 break;
1193 }
1194
1195 g_free(aiocb);
1196 return ret;
1197 }
1198
1199 static int paio_submit_co(BlockDriverState *bs, int fd,
1200 int64_t offset, QEMUIOVector *qiov,
1201 int count, int type)
1202 {
1203 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1204 ThreadPool *pool;
1205
1206 acb->bs = bs;
1207 acb->aio_type = type;
1208 acb->aio_fildes = fd;
1209
1210 acb->aio_nbytes = count;
1211 acb->aio_offset = offset;
1212
1213 if (qiov) {
1214 acb->aio_iov = qiov->iov;
1215 acb->aio_niov = qiov->niov;
1216 assert(qiov->size == count);
1217 }
1218
1219 trace_paio_submit_co(offset, count, type);
1220 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1221 return thread_pool_submit_co(pool, aio_worker, acb);
1222 }
1223
1224 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
1225 int64_t offset, QEMUIOVector *qiov, int count,
1226 BlockCompletionFunc *cb, void *opaque, int type)
1227 {
1228 RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1229 ThreadPool *pool;
1230
1231 acb->bs = bs;
1232 acb->aio_type = type;
1233 acb->aio_fildes = fd;
1234
1235 acb->aio_nbytes = count;
1236 acb->aio_offset = offset;
1237
1238 if (qiov) {
1239 acb->aio_iov = qiov->iov;
1240 acb->aio_niov = qiov->niov;
1241 assert(qiov->size == acb->aio_nbytes);
1242 }
1243
1244 trace_paio_submit(acb, opaque, offset, count, type);
1245 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1246 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1247 }
1248
1249 static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset,
1250 uint64_t bytes, QEMUIOVector *qiov, int type)
1251 {
1252 BDRVRawState *s = bs->opaque;
1253
1254 if (fd_open(bs) < 0)
1255 return -EIO;
1256
1257 /*
1258 * Check if the underlying device requires requests to be aligned,
1259 * and if the request we are trying to submit is aligned or not.
1260 * If this is the case tell the low-level driver that it needs
1261 * to copy the buffer.
1262 */
1263 if (s->needs_alignment) {
1264 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1265 type |= QEMU_AIO_MISALIGNED;
1266 #ifdef CONFIG_LINUX_AIO
1267 } else if (s->use_linux_aio) {
1268 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1269 assert(qiov->size == bytes);
1270 return laio_co_submit(bs, aio, s->fd, offset, qiov, type);
1271 #endif
1272 }
1273 }
1274
1275 return paio_submit_co(bs, s->fd, offset, qiov, bytes, type);
1276 }
1277
1278 static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset,
1279 uint64_t bytes, QEMUIOVector *qiov,
1280 int flags)
1281 {
1282 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ);
1283 }
1284
1285 static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
1286 uint64_t bytes, QEMUIOVector *qiov,
1287 int flags)
1288 {
1289 assert(flags == 0);
1290 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE);
1291 }
1292
1293 static void raw_aio_plug(BlockDriverState *bs)
1294 {
1295 #ifdef CONFIG_LINUX_AIO
1296 BDRVRawState *s = bs->opaque;
1297 if (s->use_linux_aio) {
1298 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1299 laio_io_plug(bs, aio);
1300 }
1301 #endif
1302 }
1303
1304 static void raw_aio_unplug(BlockDriverState *bs)
1305 {
1306 #ifdef CONFIG_LINUX_AIO
1307 BDRVRawState *s = bs->opaque;
1308 if (s->use_linux_aio) {
1309 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1310 laio_io_unplug(bs, aio);
1311 }
1312 #endif
1313 }
1314
1315 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
1316 BlockCompletionFunc *cb, void *opaque)
1317 {
1318 BDRVRawState *s = bs->opaque;
1319
1320 if (fd_open(bs) < 0)
1321 return NULL;
1322
1323 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1324 }
1325
1326 static void raw_close(BlockDriverState *bs)
1327 {
1328 BDRVRawState *s = bs->opaque;
1329
1330 if (s->fd >= 0) {
1331 qemu_close(s->fd);
1332 s->fd = -1;
1333 }
1334 }
1335
1336 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1337 {
1338 BDRVRawState *s = bs->opaque;
1339 struct stat st;
1340
1341 if (fstat(s->fd, &st)) {
1342 return -errno;
1343 }
1344
1345 if (S_ISREG(st.st_mode)) {
1346 if (ftruncate(s->fd, offset) < 0) {
1347 return -errno;
1348 }
1349 } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1350 if (offset > raw_getlength(bs)) {
1351 return -EINVAL;
1352 }
1353 } else {
1354 return -ENOTSUP;
1355 }
1356
1357 return 0;
1358 }
1359
1360 #ifdef __OpenBSD__
1361 static int64_t raw_getlength(BlockDriverState *bs)
1362 {
1363 BDRVRawState *s = bs->opaque;
1364 int fd = s->fd;
1365 struct stat st;
1366
1367 if (fstat(fd, &st))
1368 return -errno;
1369 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1370 struct disklabel dl;
1371
1372 if (ioctl(fd, DIOCGDINFO, &dl))
1373 return -errno;
1374 return (uint64_t)dl.d_secsize *
1375 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1376 } else
1377 return st.st_size;
1378 }
1379 #elif defined(__NetBSD__)
1380 static int64_t raw_getlength(BlockDriverState *bs)
1381 {
1382 BDRVRawState *s = bs->opaque;
1383 int fd = s->fd;
1384 struct stat st;
1385
1386 if (fstat(fd, &st))
1387 return -errno;
1388 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1389 struct dkwedge_info dkw;
1390
1391 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1392 return dkw.dkw_size * 512;
1393 } else {
1394 struct disklabel dl;
1395
1396 if (ioctl(fd, DIOCGDINFO, &dl))
1397 return -errno;
1398 return (uint64_t)dl.d_secsize *
1399 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1400 }
1401 } else
1402 return st.st_size;
1403 }
1404 #elif defined(__sun__)
1405 static int64_t raw_getlength(BlockDriverState *bs)
1406 {
1407 BDRVRawState *s = bs->opaque;
1408 struct dk_minfo minfo;
1409 int ret;
1410 int64_t size;
1411
1412 ret = fd_open(bs);
1413 if (ret < 0) {
1414 return ret;
1415 }
1416
1417 /*
1418 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1419 */
1420 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1421 if (ret != -1) {
1422 return minfo.dki_lbsize * minfo.dki_capacity;
1423 }
1424
1425 /*
1426 * There are reports that lseek on some devices fails, but
1427 * irc discussion said that contingency on contingency was overkill.
1428 */
1429 size = lseek(s->fd, 0, SEEK_END);
1430 if (size < 0) {
1431 return -errno;
1432 }
1433 return size;
1434 }
1435 #elif defined(CONFIG_BSD)
1436 static int64_t raw_getlength(BlockDriverState *bs)
1437 {
1438 BDRVRawState *s = bs->opaque;
1439 int fd = s->fd;
1440 int64_t size;
1441 struct stat sb;
1442 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1443 int reopened = 0;
1444 #endif
1445 int ret;
1446
1447 ret = fd_open(bs);
1448 if (ret < 0)
1449 return ret;
1450
1451 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1452 again:
1453 #endif
1454 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1455 #ifdef DIOCGMEDIASIZE
1456 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1457 #elif defined(DIOCGPART)
1458 {
1459 struct partinfo pi;
1460 if (ioctl(fd, DIOCGPART, &pi) == 0)
1461 size = pi.media_size;
1462 else
1463 size = 0;
1464 }
1465 if (size == 0)
1466 #endif
1467 #if defined(__APPLE__) && defined(__MACH__)
1468 {
1469 uint64_t sectors = 0;
1470 uint32_t sector_size = 0;
1471
1472 if (ioctl(fd, DKIOCGETBLOCKCOUNT, &sectors) == 0
1473 && ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) == 0) {
1474 size = sectors * sector_size;
1475 } else {
1476 size = lseek(fd, 0LL, SEEK_END);
1477 if (size < 0) {
1478 return -errno;
1479 }
1480 }
1481 }
1482 #else
1483 size = lseek(fd, 0LL, SEEK_END);
1484 if (size < 0) {
1485 return -errno;
1486 }
1487 #endif
1488 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1489 switch(s->type) {
1490 case FTYPE_CD:
1491 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1492 if (size == 2048LL * (unsigned)-1)
1493 size = 0;
1494 /* XXX no disc? maybe we need to reopen... */
1495 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1496 reopened = 1;
1497 goto again;
1498 }
1499 }
1500 #endif
1501 } else {
1502 size = lseek(fd, 0, SEEK_END);
1503 if (size < 0) {
1504 return -errno;
1505 }
1506 }
1507 return size;
1508 }
1509 #else
1510 static int64_t raw_getlength(BlockDriverState *bs)
1511 {
1512 BDRVRawState *s = bs->opaque;
1513 int ret;
1514 int64_t size;
1515
1516 ret = fd_open(bs);
1517 if (ret < 0) {
1518 return ret;
1519 }
1520
1521 size = lseek(s->fd, 0, SEEK_END);
1522 if (size < 0) {
1523 return -errno;
1524 }
1525 return size;
1526 }
1527 #endif
1528
1529 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1530 {
1531 struct stat st;
1532 BDRVRawState *s = bs->opaque;
1533
1534 if (fstat(s->fd, &st) < 0) {
1535 return -errno;
1536 }
1537 return (int64_t)st.st_blocks * 512;
1538 }
1539
1540 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
1541 {
1542 int fd;
1543 int result = 0;
1544 int64_t total_size = 0;
1545 bool nocow = false;
1546 PreallocMode prealloc;
1547 char *buf = NULL;
1548 Error *local_err = NULL;
1549
1550 strstart(filename, "file:", &filename);
1551
1552 /* Read out options */
1553 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1554 BDRV_SECTOR_SIZE);
1555 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
1556 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1557 prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
1558 PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
1559 &local_err);
1560 g_free(buf);
1561 if (local_err) {
1562 error_propagate(errp, local_err);
1563 result = -EINVAL;
1564 goto out;
1565 }
1566
1567 fd = qemu_open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY,
1568 0644);
1569 if (fd < 0) {
1570 result = -errno;
1571 error_setg_errno(errp, -result, "Could not create file");
1572 goto out;
1573 }
1574
1575 if (nocow) {
1576 #ifdef __linux__
1577 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1578 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1579 * will be ignored since any failure of this operation should not
1580 * block the left work.
1581 */
1582 int attr;
1583 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
1584 attr |= FS_NOCOW_FL;
1585 ioctl(fd, FS_IOC_SETFLAGS, &attr);
1586 }
1587 #endif
1588 }
1589
1590 if (ftruncate(fd, total_size) != 0) {
1591 result = -errno;
1592 error_setg_errno(errp, -result, "Could not resize file");
1593 goto out_close;
1594 }
1595
1596 switch (prealloc) {
1597 #ifdef CONFIG_POSIX_FALLOCATE
1598 case PREALLOC_MODE_FALLOC:
1599 /* posix_fallocate() doesn't set errno. */
1600 result = -posix_fallocate(fd, 0, total_size);
1601 if (result != 0) {
1602 error_setg_errno(errp, -result,
1603 "Could not preallocate data for the new file");
1604 }
1605 break;
1606 #endif
1607 case PREALLOC_MODE_FULL:
1608 {
1609 int64_t num = 0, left = total_size;
1610 buf = g_malloc0(65536);
1611
1612 while (left > 0) {
1613 num = MIN(left, 65536);
1614 result = write(fd, buf, num);
1615 if (result < 0) {
1616 result = -errno;
1617 error_setg_errno(errp, -result,
1618 "Could not write to the new file");
1619 break;
1620 }
1621 left -= result;
1622 }
1623 if (result >= 0) {
1624 result = fsync(fd);
1625 if (result < 0) {
1626 result = -errno;
1627 error_setg_errno(errp, -result,
1628 "Could not flush new file to disk");
1629 }
1630 }
1631 g_free(buf);
1632 break;
1633 }
1634 case PREALLOC_MODE_OFF:
1635 break;
1636 default:
1637 result = -EINVAL;
1638 error_setg(errp, "Unsupported preallocation mode: %s",
1639 PreallocMode_lookup[prealloc]);
1640 break;
1641 }
1642
1643 out_close:
1644 if (qemu_close(fd) != 0 && result == 0) {
1645 result = -errno;
1646 error_setg_errno(errp, -result, "Could not close the new file");
1647 }
1648 out:
1649 return result;
1650 }
1651
1652 /*
1653 * Find allocation range in @bs around offset @start.
1654 * May change underlying file descriptor's file offset.
1655 * If @start is not in a hole, store @start in @data, and the
1656 * beginning of the next hole in @hole, and return 0.
1657 * If @start is in a non-trailing hole, store @start in @hole and the
1658 * beginning of the next non-hole in @data, and return 0.
1659 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1660 * If we can't find out, return a negative errno other than -ENXIO.
1661 */
1662 static int find_allocation(BlockDriverState *bs, off_t start,
1663 off_t *data, off_t *hole)
1664 {
1665 #if defined SEEK_HOLE && defined SEEK_DATA
1666 BDRVRawState *s = bs->opaque;
1667 off_t offs;
1668
1669 /*
1670 * SEEK_DATA cases:
1671 * D1. offs == start: start is in data
1672 * D2. offs > start: start is in a hole, next data at offs
1673 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1674 * or start is beyond EOF
1675 * If the latter happens, the file has been truncated behind
1676 * our back since we opened it. All bets are off then.
1677 * Treating like a trailing hole is simplest.
1678 * D4. offs < 0, errno != ENXIO: we learned nothing
1679 */
1680 offs = lseek(s->fd, start, SEEK_DATA);
1681 if (offs < 0) {
1682 return -errno; /* D3 or D4 */
1683 }
1684 assert(offs >= start);
1685
1686 if (offs > start) {
1687 /* D2: in hole, next data at offs */
1688 *hole = start;
1689 *data = offs;
1690 return 0;
1691 }
1692
1693 /* D1: in data, end not yet known */
1694
1695 /*
1696 * SEEK_HOLE cases:
1697 * H1. offs == start: start is in a hole
1698 * If this happens here, a hole has been dug behind our back
1699 * since the previous lseek().
1700 * H2. offs > start: either start is in data, next hole at offs,
1701 * or start is in trailing hole, EOF at offs
1702 * Linux treats trailing holes like any other hole: offs ==
1703 * start. Solaris seeks to EOF instead: offs > start (blech).
1704 * If that happens here, a hole has been dug behind our back
1705 * since the previous lseek().
1706 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1707 * If this happens, the file has been truncated behind our
1708 * back since we opened it. Treat it like a trailing hole.
1709 * H4. offs < 0, errno != ENXIO: we learned nothing
1710 * Pretend we know nothing at all, i.e. "forget" about D1.
1711 */
1712 offs = lseek(s->fd, start, SEEK_HOLE);
1713 if (offs < 0) {
1714 return -errno; /* D1 and (H3 or H4) */
1715 }
1716 assert(offs >= start);
1717
1718 if (offs > start) {
1719 /*
1720 * D1 and H2: either in data, next hole at offs, or it was in
1721 * data but is now in a trailing hole. In the latter case,
1722 * all bets are off. Treating it as if it there was data all
1723 * the way to EOF is safe, so simply do that.
1724 */
1725 *data = start;
1726 *hole = offs;
1727 return 0;
1728 }
1729
1730 /* D1 and H1 */
1731 return -EBUSY;
1732 #else
1733 return -ENOTSUP;
1734 #endif
1735 }
1736
1737 /*
1738 * Returns the allocation status of the specified sectors.
1739 *
1740 * If 'sector_num' is beyond the end of the disk image the return value is 0
1741 * and 'pnum' is set to 0.
1742 *
1743 * 'pnum' is set to the number of sectors (including and immediately following
1744 * the specified sector) that are known to be in the same
1745 * allocated/unallocated state.
1746 *
1747 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1748 * beyond the end of the disk image it will be clamped.
1749 */
1750 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1751 int64_t sector_num,
1752 int nb_sectors, int *pnum,
1753 BlockDriverState **file)
1754 {
1755 off_t start, data = 0, hole = 0;
1756 int64_t total_size;
1757 int ret;
1758
1759 ret = fd_open(bs);
1760 if (ret < 0) {
1761 return ret;
1762 }
1763
1764 start = sector_num * BDRV_SECTOR_SIZE;
1765 total_size = bdrv_getlength(bs);
1766 if (total_size < 0) {
1767 return total_size;
1768 } else if (start >= total_size) {
1769 *pnum = 0;
1770 return 0;
1771 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1772 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1773 }
1774
1775 ret = find_allocation(bs, start, &data, &hole);
1776 if (ret == -ENXIO) {
1777 /* Trailing hole */
1778 *pnum = nb_sectors;
1779 ret = BDRV_BLOCK_ZERO;
1780 } else if (ret < 0) {
1781 /* No info available, so pretend there are no holes */
1782 *pnum = nb_sectors;
1783 ret = BDRV_BLOCK_DATA;
1784 } else if (data == start) {
1785 /* On a data extent, compute sectors to the end of the extent,
1786 * possibly including a partial sector at EOF. */
1787 *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1788 ret = BDRV_BLOCK_DATA;
1789 } else {
1790 /* On a hole, compute sectors to the beginning of the next extent. */
1791 assert(hole == start);
1792 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1793 ret = BDRV_BLOCK_ZERO;
1794 }
1795 *file = bs;
1796 return ret | BDRV_BLOCK_OFFSET_VALID | start;
1797 }
1798
1799 static coroutine_fn BlockAIOCB *raw_aio_pdiscard(BlockDriverState *bs,
1800 int64_t offset, int count,
1801 BlockCompletionFunc *cb, void *opaque)
1802 {
1803 BDRVRawState *s = bs->opaque;
1804
1805 return paio_submit(bs, s->fd, offset, NULL, count,
1806 cb, opaque, QEMU_AIO_DISCARD);
1807 }
1808
1809 static int coroutine_fn raw_co_pwrite_zeroes(
1810 BlockDriverState *bs, int64_t offset,
1811 int count, BdrvRequestFlags flags)
1812 {
1813 BDRVRawState *s = bs->opaque;
1814
1815 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1816 return paio_submit_co(bs, s->fd, offset, NULL, count,
1817 QEMU_AIO_WRITE_ZEROES);
1818 } else if (s->discard_zeroes) {
1819 return paio_submit_co(bs, s->fd, offset, NULL, count,
1820 QEMU_AIO_DISCARD);
1821 }
1822 return -ENOTSUP;
1823 }
1824
1825 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1826 {
1827 BDRVRawState *s = bs->opaque;
1828
1829 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1830 bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1831 return 0;
1832 }
1833
1834 static QemuOptsList raw_create_opts = {
1835 .name = "raw-create-opts",
1836 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
1837 .desc = {
1838 {
1839 .name = BLOCK_OPT_SIZE,
1840 .type = QEMU_OPT_SIZE,
1841 .help = "Virtual disk size"
1842 },
1843 {
1844 .name = BLOCK_OPT_NOCOW,
1845 .type = QEMU_OPT_BOOL,
1846 .help = "Turn off copy-on-write (valid only on btrfs)"
1847 },
1848 {
1849 .name = BLOCK_OPT_PREALLOC,
1850 .type = QEMU_OPT_STRING,
1851 .help = "Preallocation mode (allowed values: off, falloc, full)"
1852 },
1853 { /* end of list */ }
1854 }
1855 };
1856
1857 BlockDriver bdrv_file = {
1858 .format_name = "file",
1859 .protocol_name = "file",
1860 .instance_size = sizeof(BDRVRawState),
1861 .bdrv_needs_filename = true,
1862 .bdrv_probe = NULL, /* no probe for protocols */
1863 .bdrv_parse_filename = raw_parse_filename,
1864 .bdrv_file_open = raw_open,
1865 .bdrv_reopen_prepare = raw_reopen_prepare,
1866 .bdrv_reopen_commit = raw_reopen_commit,
1867 .bdrv_reopen_abort = raw_reopen_abort,
1868 .bdrv_close = raw_close,
1869 .bdrv_create = raw_create,
1870 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1871 .bdrv_co_get_block_status = raw_co_get_block_status,
1872 .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes,
1873
1874 .bdrv_co_preadv = raw_co_preadv,
1875 .bdrv_co_pwritev = raw_co_pwritev,
1876 .bdrv_aio_flush = raw_aio_flush,
1877 .bdrv_aio_pdiscard = raw_aio_pdiscard,
1878 .bdrv_refresh_limits = raw_refresh_limits,
1879 .bdrv_io_plug = raw_aio_plug,
1880 .bdrv_io_unplug = raw_aio_unplug,
1881
1882 .bdrv_truncate = raw_truncate,
1883 .bdrv_getlength = raw_getlength,
1884 .bdrv_get_info = raw_get_info,
1885 .bdrv_get_allocated_file_size
1886 = raw_get_allocated_file_size,
1887
1888 .create_opts = &raw_create_opts,
1889 };
1890
1891 /***********************************************/
1892 /* host device */
1893
1894 #if defined(__APPLE__) && defined(__MACH__)
1895 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
1896 CFIndex maxPathSize, int flags);
1897 static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
1898 {
1899 kern_return_t kernResult = KERN_FAILURE;
1900 mach_port_t masterPort;
1901 CFMutableDictionaryRef classesToMatch;
1902 const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
1903 char *mediaType = NULL;
1904
1905 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1906 if ( KERN_SUCCESS != kernResult ) {
1907 printf( "IOMasterPort returned %d\n", kernResult );
1908 }
1909
1910 int index;
1911 for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
1912 classesToMatch = IOServiceMatching(matching_array[index]);
1913 if (classesToMatch == NULL) {
1914 error_report("IOServiceMatching returned NULL for %s",
1915 matching_array[index]);
1916 continue;
1917 }
1918 CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
1919 kCFBooleanTrue);
1920 kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch,
1921 mediaIterator);
1922 if (kernResult != KERN_SUCCESS) {
1923 error_report("Note: IOServiceGetMatchingServices returned %d",
1924 kernResult);
1925 continue;
1926 }
1927
1928 /* If a match was found, leave the loop */
1929 if (*mediaIterator != 0) {
1930 DPRINTF("Matching using %s\n", matching_array[index]);
1931 mediaType = g_strdup(matching_array[index]);
1932 break;
1933 }
1934 }
1935 return mediaType;
1936 }
1937
1938 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
1939 CFIndex maxPathSize, int flags)
1940 {
1941 io_object_t nextMedia;
1942 kern_return_t kernResult = KERN_FAILURE;
1943 *bsdPath = '\0';
1944 nextMedia = IOIteratorNext( mediaIterator );
1945 if ( nextMedia )
1946 {
1947 CFTypeRef bsdPathAsCFString;
1948 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1949 if ( bsdPathAsCFString ) {
1950 size_t devPathLength;
1951 strcpy( bsdPath, _PATH_DEV );
1952 if (flags & BDRV_O_NOCACHE) {
1953 strcat(bsdPath, "r");
1954 }
1955 devPathLength = strlen( bsdPath );
1956 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1957 kernResult = KERN_SUCCESS;
1958 }
1959 CFRelease( bsdPathAsCFString );
1960 }
1961 IOObjectRelease( nextMedia );
1962 }
1963
1964 return kernResult;
1965 }
1966
1967 /* Sets up a real cdrom for use in QEMU */
1968 static bool setup_cdrom(char *bsd_path, Error **errp)
1969 {
1970 int index, num_of_test_partitions = 2, fd;
1971 char test_partition[MAXPATHLEN];
1972 bool partition_found = false;
1973
1974 /* look for a working partition */
1975 for (index = 0; index < num_of_test_partitions; index++) {
1976 snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
1977 index);
1978 fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE);
1979 if (fd >= 0) {
1980 partition_found = true;
1981 qemu_close(fd);
1982 break;
1983 }
1984 }
1985
1986 /* if a working partition on the device was not found */
1987 if (partition_found == false) {
1988 error_setg(errp, "Failed to find a working partition on disc");
1989 } else {
1990 DPRINTF("Using %s as optical disc\n", test_partition);
1991 pstrcpy(bsd_path, MAXPATHLEN, test_partition);
1992 }
1993 return partition_found;
1994 }
1995
1996 /* Prints directions on mounting and unmounting a device */
1997 static void print_unmounting_directions(const char *file_name)
1998 {
1999 error_report("If device %s is mounted on the desktop, unmount"
2000 " it first before using it in QEMU", file_name);
2001 error_report("Command to unmount device: diskutil unmountDisk %s",
2002 file_name);
2003 error_report("Command to mount device: diskutil mountDisk %s", file_name);
2004 }
2005
2006 #endif /* defined(__APPLE__) && defined(__MACH__) */
2007
2008 static int hdev_probe_device(const char *filename)
2009 {
2010 struct stat st;
2011
2012 /* allow a dedicated CD-ROM driver to match with a higher priority */
2013 if (strstart(filename, "/dev/cdrom", NULL))
2014 return 50;
2015
2016 if (stat(filename, &st) >= 0 &&
2017 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
2018 return 100;
2019 }
2020
2021 return 0;
2022 }
2023
2024 static int check_hdev_writable(BDRVRawState *s)
2025 {
2026 #if defined(BLKROGET)
2027 /* Linux block devices can be configured "read-only" using blockdev(8).
2028 * This is independent of device node permissions and therefore open(2)
2029 * with O_RDWR succeeds. Actual writes fail with EPERM.
2030 *
2031 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2032 * check for read-only block devices so that Linux block devices behave
2033 * properly.
2034 */
2035 struct stat st;
2036 int readonly = 0;
2037
2038 if (fstat(s->fd, &st)) {
2039 return -errno;
2040 }
2041
2042 if (!S_ISBLK(st.st_mode)) {
2043 return 0;
2044 }
2045
2046 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
2047 return -errno;
2048 }
2049
2050 if (readonly) {
2051 return -EACCES;
2052 }
2053 #endif /* defined(BLKROGET) */
2054 return 0;
2055 }
2056
2057 static void hdev_parse_filename(const char *filename, QDict *options,
2058 Error **errp)
2059 {
2060 /* The prefix is optional, just as for "file". */
2061 strstart(filename, "host_device:", &filename);
2062
2063 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2064 }
2065
2066 static bool hdev_is_sg(BlockDriverState *bs)
2067 {
2068
2069 #if defined(__linux__)
2070
2071 struct stat st;
2072 struct sg_scsi_id scsiid;
2073 int sg_version;
2074
2075 if (stat(bs->filename, &st) >= 0 && S_ISCHR(st.st_mode) &&
2076 !bdrv_ioctl(bs, SG_GET_VERSION_NUM, &sg_version) &&
2077 !bdrv_ioctl(bs, SG_GET_SCSI_ID, &scsiid)) {
2078 DPRINTF("SG device found: type=%d, version=%d\n",
2079 scsiid.scsi_type, sg_version);
2080 return true;
2081 }
2082
2083 #endif
2084
2085 return false;
2086 }
2087
2088 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
2089 Error **errp)
2090 {
2091 BDRVRawState *s = bs->opaque;
2092 Error *local_err = NULL;
2093 int ret;
2094
2095 #if defined(__APPLE__) && defined(__MACH__)
2096 const char *filename = qdict_get_str(options, "filename");
2097 char bsd_path[MAXPATHLEN] = "";
2098 bool error_occurred = false;
2099
2100 /* If using a real cdrom */
2101 if (strcmp(filename, "/dev/cdrom") == 0) {
2102 char *mediaType = NULL;
2103 kern_return_t ret_val;
2104 io_iterator_t mediaIterator = 0;
2105
2106 mediaType = FindEjectableOpticalMedia(&mediaIterator);
2107 if (mediaType == NULL) {
2108 error_setg(errp, "Please make sure your CD/DVD is in the optical"
2109 " drive");
2110 error_occurred = true;
2111 goto hdev_open_Mac_error;
2112 }
2113
2114 ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
2115 if (ret_val != KERN_SUCCESS) {
2116 error_setg(errp, "Could not get BSD path for optical drive");
2117 error_occurred = true;
2118 goto hdev_open_Mac_error;
2119 }
2120
2121 /* If a real optical drive was not found */
2122 if (bsd_path[0] == '\0') {
2123 error_setg(errp, "Failed to obtain bsd path for optical drive");
2124 error_occurred = true;
2125 goto hdev_open_Mac_error;
2126 }
2127
2128 /* If using a cdrom disc and finding a partition on the disc failed */
2129 if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
2130 setup_cdrom(bsd_path, errp) == false) {
2131 print_unmounting_directions(bsd_path);
2132 error_occurred = true;
2133 goto hdev_open_Mac_error;
2134 }
2135
2136 qdict_put(options, "filename", qstring_from_str(bsd_path));
2137
2138 hdev_open_Mac_error:
2139 g_free(mediaType);
2140 if (mediaIterator) {
2141 IOObjectRelease(mediaIterator);
2142 }
2143 if (error_occurred) {
2144 return -ENOENT;
2145 }
2146 }
2147 #endif /* defined(__APPLE__) && defined(__MACH__) */
2148
2149 s->type = FTYPE_FILE;
2150
2151 ret = raw_open_common(bs, options, flags, 0, &local_err);
2152 if (ret < 0) {
2153 error_propagate(errp, local_err);
2154 #if defined(__APPLE__) && defined(__MACH__)
2155 if (*bsd_path) {
2156 filename = bsd_path;
2157 }
2158 /* if a physical device experienced an error while being opened */
2159 if (strncmp(filename, "/dev/", 5) == 0) {
2160 print_unmounting_directions(filename);
2161 }
2162 #endif /* defined(__APPLE__) && defined(__MACH__) */
2163 return ret;
2164 }
2165
2166 /* Since this does ioctl the device must be already opened */
2167 bs->sg = hdev_is_sg(bs);
2168
2169 if (flags & BDRV_O_RDWR) {
2170 ret = check_hdev_writable(s);
2171 if (ret < 0) {
2172 raw_close(bs);
2173 error_setg_errno(errp, -ret, "The device is not writable");
2174 return ret;
2175 }
2176 }
2177
2178 return ret;
2179 }
2180
2181 #if defined(__linux__)
2182
2183 static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
2184 unsigned long int req, void *buf,
2185 BlockCompletionFunc *cb, void *opaque)
2186 {
2187 BDRVRawState *s = bs->opaque;
2188 RawPosixAIOData *acb;
2189 ThreadPool *pool;
2190
2191 if (fd_open(bs) < 0)
2192 return NULL;
2193
2194 acb = g_new(RawPosixAIOData, 1);
2195 acb->bs = bs;
2196 acb->aio_type = QEMU_AIO_IOCTL;
2197 acb->aio_fildes = s->fd;
2198 acb->aio_offset = 0;
2199 acb->aio_ioctl_buf = buf;
2200 acb->aio_ioctl_cmd = req;
2201 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
2202 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
2203 }
2204 #endif /* linux */
2205
2206 static int fd_open(BlockDriverState *bs)
2207 {
2208 BDRVRawState *s = bs->opaque;
2209
2210 /* this is just to ensure s->fd is sane (its called by io ops) */
2211 if (s->fd >= 0)
2212 return 0;
2213 return -EIO;
2214 }
2215
2216 static coroutine_fn BlockAIOCB *hdev_aio_pdiscard(BlockDriverState *bs,
2217 int64_t offset, int count,
2218 BlockCompletionFunc *cb, void *opaque)
2219 {
2220 BDRVRawState *s = bs->opaque;
2221
2222 if (fd_open(bs) < 0) {
2223 return NULL;
2224 }
2225 return paio_submit(bs, s->fd, offset, NULL, count,
2226 cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2227 }
2228
2229 static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs,
2230 int64_t offset, int count, BdrvRequestFlags flags)
2231 {
2232 BDRVRawState *s = bs->opaque;
2233 int rc;
2234
2235 rc = fd_open(bs);
2236 if (rc < 0) {
2237 return rc;
2238 }
2239 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2240 return paio_submit_co(bs, s->fd, offset, NULL, count,
2241 QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2242 } else if (s->discard_zeroes) {
2243 return paio_submit_co(bs, s->fd, offset, NULL, count,
2244 QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2245 }
2246 return -ENOTSUP;
2247 }
2248
2249 static int hdev_create(const char *filename, QemuOpts *opts,
2250 Error **errp)
2251 {
2252 int fd;
2253 int ret = 0;
2254 struct stat stat_buf;
2255 int64_t total_size = 0;
2256 bool has_prefix;
2257
2258 /* This function is used by both protocol block drivers and therefore either
2259 * of these prefixes may be given.
2260 * The return value has to be stored somewhere, otherwise this is an error
2261 * due to -Werror=unused-value. */
2262 has_prefix =
2263 strstart(filename, "host_device:", &filename) ||
2264 strstart(filename, "host_cdrom:" , &filename);
2265
2266 (void)has_prefix;
2267
2268 ret = raw_normalize_devicepath(&filename);
2269 if (ret < 0) {
2270 error_setg_errno(errp, -ret, "Could not normalize device path");
2271 return ret;
2272 }
2273
2274 /* Read out options */
2275 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2276 BDRV_SECTOR_SIZE);
2277
2278 fd = qemu_open(filename, O_WRONLY | O_BINARY);
2279 if (fd < 0) {
2280 ret = -errno;
2281 error_setg_errno(errp, -ret, "Could not open device");
2282 return ret;
2283 }
2284
2285 if (fstat(fd, &stat_buf) < 0) {
2286 ret = -errno;
2287 error_setg_errno(errp, -ret, "Could not stat device");
2288 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2289 error_setg(errp,
2290 "The given file is neither a block nor a character device");
2291 ret = -ENODEV;
2292 } else if (lseek(fd, 0, SEEK_END) < total_size) {
2293 error_setg(errp, "Device is too small");
2294 ret = -ENOSPC;
2295 }
2296
2297 qemu_close(fd);
2298 return ret;
2299 }
2300
2301 static BlockDriver bdrv_host_device = {
2302 .format_name = "host_device",
2303 .protocol_name = "host_device",
2304 .instance_size = sizeof(BDRVRawState),
2305 .bdrv_needs_filename = true,
2306 .bdrv_probe_device = hdev_probe_device,
2307 .bdrv_parse_filename = hdev_parse_filename,
2308 .bdrv_file_open = hdev_open,
2309 .bdrv_close = raw_close,
2310 .bdrv_reopen_prepare = raw_reopen_prepare,
2311 .bdrv_reopen_commit = raw_reopen_commit,
2312 .bdrv_reopen_abort = raw_reopen_abort,
2313 .bdrv_create = hdev_create,
2314 .create_opts = &raw_create_opts,
2315 .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes,
2316
2317 .bdrv_co_preadv = raw_co_preadv,
2318 .bdrv_co_pwritev = raw_co_pwritev,
2319 .bdrv_aio_flush = raw_aio_flush,
2320 .bdrv_aio_pdiscard = hdev_aio_pdiscard,
2321 .bdrv_refresh_limits = raw_refresh_limits,
2322 .bdrv_io_plug = raw_aio_plug,
2323 .bdrv_io_unplug = raw_aio_unplug,
2324
2325 .bdrv_truncate = raw_truncate,
2326 .bdrv_getlength = raw_getlength,
2327 .bdrv_get_info = raw_get_info,
2328 .bdrv_get_allocated_file_size
2329 = raw_get_allocated_file_size,
2330 .bdrv_probe_blocksizes = hdev_probe_blocksizes,
2331 .bdrv_probe_geometry = hdev_probe_geometry,
2332
2333 /* generic scsi device */
2334 #ifdef __linux__
2335 .bdrv_aio_ioctl = hdev_aio_ioctl,
2336 #endif
2337 };
2338
2339 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2340 static void cdrom_parse_filename(const char *filename, QDict *options,
2341 Error **errp)
2342 {
2343 /* The prefix is optional, just as for "file". */
2344 strstart(filename, "host_cdrom:", &filename);
2345
2346 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2347 }
2348 #endif
2349
2350 #ifdef __linux__
2351 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2352 Error **errp)
2353 {
2354 BDRVRawState *s = bs->opaque;
2355
2356 s->type = FTYPE_CD;
2357
2358 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2359 return raw_open_common(bs, options, flags, O_NONBLOCK, errp);
2360 }
2361
2362 static int cdrom_probe_device(const char *filename)
2363 {
2364 int fd, ret;
2365 int prio = 0;
2366 struct stat st;
2367
2368 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2369 if (fd < 0) {
2370 goto out;
2371 }
2372 ret = fstat(fd, &st);
2373 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2374 goto outc;
2375 }
2376
2377 /* Attempt to detect via a CDROM specific ioctl */
2378 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2379 if (ret >= 0)
2380 prio = 100;
2381
2382 outc:
2383 qemu_close(fd);
2384 out:
2385 return prio;
2386 }
2387
2388 static bool cdrom_is_inserted(BlockDriverState *bs)
2389 {
2390 BDRVRawState *s = bs->opaque;
2391 int ret;
2392
2393 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2394 return ret == CDS_DISC_OK;
2395 }
2396
2397 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2398 {
2399 BDRVRawState *s = bs->opaque;
2400
2401 if (eject_flag) {
2402 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2403 perror("CDROMEJECT");
2404 } else {
2405 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2406 perror("CDROMEJECT");
2407 }
2408 }
2409
2410 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2411 {
2412 BDRVRawState *s = bs->opaque;
2413
2414 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2415 /*
2416 * Note: an error can happen if the distribution automatically
2417 * mounts the CD-ROM
2418 */
2419 /* perror("CDROM_LOCKDOOR"); */
2420 }
2421 }
2422
2423 static BlockDriver bdrv_host_cdrom = {
2424 .format_name = "host_cdrom",
2425 .protocol_name = "host_cdrom",
2426 .instance_size = sizeof(BDRVRawState),
2427 .bdrv_needs_filename = true,
2428 .bdrv_probe_device = cdrom_probe_device,
2429 .bdrv_parse_filename = cdrom_parse_filename,
2430 .bdrv_file_open = cdrom_open,
2431 .bdrv_close = raw_close,
2432 .bdrv_reopen_prepare = raw_reopen_prepare,
2433 .bdrv_reopen_commit = raw_reopen_commit,
2434 .bdrv_reopen_abort = raw_reopen_abort,
2435 .bdrv_create = hdev_create,
2436 .create_opts = &raw_create_opts,
2437
2438
2439 .bdrv_co_preadv = raw_co_preadv,
2440 .bdrv_co_pwritev = raw_co_pwritev,
2441 .bdrv_aio_flush = raw_aio_flush,
2442 .bdrv_refresh_limits = raw_refresh_limits,
2443 .bdrv_io_plug = raw_aio_plug,
2444 .bdrv_io_unplug = raw_aio_unplug,
2445
2446 .bdrv_truncate = raw_truncate,
2447 .bdrv_getlength = raw_getlength,
2448 .has_variable_length = true,
2449 .bdrv_get_allocated_file_size
2450 = raw_get_allocated_file_size,
2451
2452 /* removable device support */
2453 .bdrv_is_inserted = cdrom_is_inserted,
2454 .bdrv_eject = cdrom_eject,
2455 .bdrv_lock_medium = cdrom_lock_medium,
2456
2457 /* generic scsi device */
2458 .bdrv_aio_ioctl = hdev_aio_ioctl,
2459 };
2460 #endif /* __linux__ */
2461
2462 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2463 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2464 Error **errp)
2465 {
2466 BDRVRawState *s = bs->opaque;
2467 Error *local_err = NULL;
2468 int ret;
2469
2470 s->type = FTYPE_CD;
2471
2472 ret = raw_open_common(bs, options, flags, 0, &local_err);
2473 if (ret) {
2474 error_propagate(errp, local_err);
2475 return ret;
2476 }
2477
2478 /* make sure the door isn't locked at this time */
2479 ioctl(s->fd, CDIOCALLOW);
2480 return 0;
2481 }
2482
2483 static int cdrom_probe_device(const char *filename)
2484 {
2485 if (strstart(filename, "/dev/cd", NULL) ||
2486 strstart(filename, "/dev/acd", NULL))
2487 return 100;
2488 return 0;
2489 }
2490
2491 static int cdrom_reopen(BlockDriverState *bs)
2492 {
2493 BDRVRawState *s = bs->opaque;
2494 int fd;
2495
2496 /*
2497 * Force reread of possibly changed/newly loaded disc,
2498 * FreeBSD seems to not notice sometimes...
2499 */
2500 if (s->fd >= 0)
2501 qemu_close(s->fd);
2502 fd = qemu_open(bs->filename, s->open_flags, 0644);
2503 if (fd < 0) {
2504 s->fd = -1;
2505 return -EIO;
2506 }
2507 s->fd = fd;
2508
2509 /* make sure the door isn't locked at this time */
2510 ioctl(s->fd, CDIOCALLOW);
2511 return 0;
2512 }
2513
2514 static bool cdrom_is_inserted(BlockDriverState *bs)
2515 {
2516 return raw_getlength(bs) > 0;
2517 }
2518
2519 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2520 {
2521 BDRVRawState *s = bs->opaque;
2522
2523 if (s->fd < 0)
2524 return;
2525
2526 (void) ioctl(s->fd, CDIOCALLOW);
2527
2528 if (eject_flag) {
2529 if (ioctl(s->fd, CDIOCEJECT) < 0)
2530 perror("CDIOCEJECT");
2531 } else {
2532 if (ioctl(s->fd, CDIOCCLOSE) < 0)
2533 perror("CDIOCCLOSE");
2534 }
2535
2536 cdrom_reopen(bs);
2537 }
2538
2539 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2540 {
2541 BDRVRawState *s = bs->opaque;
2542
2543 if (s->fd < 0)
2544 return;
2545 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2546 /*
2547 * Note: an error can happen if the distribution automatically
2548 * mounts the CD-ROM
2549 */
2550 /* perror("CDROM_LOCKDOOR"); */
2551 }
2552 }
2553
2554 static BlockDriver bdrv_host_cdrom = {
2555 .format_name = "host_cdrom",
2556 .protocol_name = "host_cdrom",
2557 .instance_size = sizeof(BDRVRawState),
2558 .bdrv_needs_filename = true,
2559 .bdrv_probe_device = cdrom_probe_device,
2560 .bdrv_parse_filename = cdrom_parse_filename,
2561 .bdrv_file_open = cdrom_open,
2562 .bdrv_close = raw_close,
2563 .bdrv_reopen_prepare = raw_reopen_prepare,
2564 .bdrv_reopen_commit = raw_reopen_commit,
2565 .bdrv_reopen_abort = raw_reopen_abort,
2566 .bdrv_create = hdev_create,
2567 .create_opts = &raw_create_opts,
2568
2569 .bdrv_co_preadv = raw_co_preadv,
2570 .bdrv_co_pwritev = raw_co_pwritev,
2571 .bdrv_aio_flush = raw_aio_flush,
2572 .bdrv_refresh_limits = raw_refresh_limits,
2573 .bdrv_io_plug = raw_aio_plug,
2574 .bdrv_io_unplug = raw_aio_unplug,
2575
2576 .bdrv_truncate = raw_truncate,
2577 .bdrv_getlength = raw_getlength,
2578 .has_variable_length = true,
2579 .bdrv_get_allocated_file_size
2580 = raw_get_allocated_file_size,
2581
2582 /* removable device support */
2583 .bdrv_is_inserted = cdrom_is_inserted,
2584 .bdrv_eject = cdrom_eject,
2585 .bdrv_lock_medium = cdrom_lock_medium,
2586 };
2587 #endif /* __FreeBSD__ */
2588
2589 static void bdrv_file_init(void)
2590 {
2591 /*
2592 * Register all the drivers. Note that order is important, the driver
2593 * registered last will get probed first.
2594 */
2595 bdrv_register(&bdrv_file);
2596 bdrv_register(&bdrv_host_device);
2597 #ifdef __linux__
2598 bdrv_register(&bdrv_host_cdrom);
2599 #endif
2600 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2601 bdrv_register(&bdrv_host_cdrom);
2602 #endif
2603 }
2604
2605 block_init(bdrv_file_init);
Random patches