virtio-blk: add SGI_IO passthru support
[qemu-kvm/fedora.git] / block-raw-posix.c
blob0663c0695b4054c956d5a129050614b04712a743
1 /*
2 * Block driver for RAW files (posix)
4 * Copyright (c) 2006 Fabrice Bellard
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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.
24 #include "qemu-common.h"
25 #include "qemu-timer.h"
26 #include "qemu-char.h"
27 #include "block_int.h"
28 #include <assert.h>
29 #ifdef CONFIG_AIO
30 #include "posix-aio-compat.h"
31 #endif
33 #ifdef CONFIG_COCOA
34 #include <paths.h>
35 #include <sys/param.h>
36 #include <IOKit/IOKitLib.h>
37 #include <IOKit/IOBSD.h>
38 #include <IOKit/storage/IOMediaBSDClient.h>
39 #include <IOKit/storage/IOMedia.h>
40 #include <IOKit/storage/IOCDMedia.h>
41 //#include <IOKit/storage/IOCDTypes.h>
42 #include <CoreFoundation/CoreFoundation.h>
43 #endif
45 #ifdef __sun__
46 #define _POSIX_PTHREAD_SEMANTICS 1
47 #include <signal.h>
48 #include <sys/dkio.h>
49 #endif
50 #ifdef __linux__
51 #include <sys/ioctl.h>
52 #include <linux/cdrom.h>
53 #include <linux/fd.h>
54 #endif
55 #ifdef __FreeBSD__
56 #include <signal.h>
57 #include <sys/disk.h>
58 #include <sys/cdio.h>
59 #endif
61 #ifdef __OpenBSD__
62 #include <sys/ioctl.h>
63 #include <sys/disklabel.h>
64 #include <sys/dkio.h>
65 #endif
67 #ifdef __DragonFly__
68 #include <sys/ioctl.h>
69 #include <sys/diskslice.h>
70 #endif
72 //#define DEBUG_FLOPPY
74 //#define DEBUG_BLOCK
75 #if defined(DEBUG_BLOCK)
76 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (qemu_log_enabled()) \
77 { qemu_log(formatCstr, ##args); qemu_log_flush(); } } while (0)
78 #else
79 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
80 #endif
82 /* OS X does not have O_DSYNC */
83 #ifndef O_DSYNC
84 #define O_DSYNC O_SYNC
85 #endif
87 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
88 #ifndef O_DIRECT
89 #define O_DIRECT O_DSYNC
90 #endif
92 #define FTYPE_FILE 0
93 #define FTYPE_CD 1
94 #define FTYPE_FD 2
96 #define ALIGNED_BUFFER_SIZE (32 * 512)
98 /* if the FD is not accessed during that time (in ms), we try to
99 reopen it to see if the disk has been changed */
100 #define FD_OPEN_TIMEOUT 1000
102 typedef struct BDRVRawState {
103 int fd;
104 int type;
105 unsigned int lseek_err_cnt;
106 #if defined(__linux__)
107 /* linux floppy specific */
108 int fd_open_flags;
109 int64_t fd_open_time;
110 int64_t fd_error_time;
111 int fd_got_error;
112 int fd_media_changed;
113 #endif
114 #if defined(__FreeBSD__)
115 int cd_open_flags;
116 #endif
117 uint8_t* aligned_buf;
118 } BDRVRawState;
120 static int posix_aio_init(void);
122 static int fd_open(BlockDriverState *bs);
124 #if defined(__FreeBSD__)
125 static int cd_open(BlockDriverState *bs);
126 #endif
128 static int raw_is_inserted(BlockDriverState *bs);
130 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
132 BDRVRawState *s = bs->opaque;
133 int fd, open_flags, ret;
135 posix_aio_init();
137 s->lseek_err_cnt = 0;
139 open_flags = O_BINARY;
140 if ((flags & BDRV_O_ACCESS) == O_RDWR) {
141 open_flags |= O_RDWR;
142 } else {
143 open_flags |= O_RDONLY;
144 bs->read_only = 1;
146 if (flags & BDRV_O_CREAT)
147 open_flags |= O_CREAT | O_TRUNC;
149 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
150 * and O_DIRECT for no caching. */
151 if ((flags & BDRV_O_NOCACHE))
152 open_flags |= O_DIRECT;
153 else if (!(flags & BDRV_O_CACHE_WB))
154 open_flags |= O_DSYNC;
156 s->type = FTYPE_FILE;
158 fd = open(filename, open_flags, 0644);
159 if (fd < 0) {
160 ret = -errno;
161 if (ret == -EROFS)
162 ret = -EACCES;
163 return ret;
165 s->fd = fd;
166 s->aligned_buf = NULL;
167 if ((flags & BDRV_O_NOCACHE)) {
168 s->aligned_buf = qemu_blockalign(bs, ALIGNED_BUFFER_SIZE);
169 if (s->aligned_buf == NULL) {
170 ret = -errno;
171 close(fd);
172 return ret;
175 return 0;
178 /* XXX: use host sector size if necessary with:
179 #ifdef DIOCGSECTORSIZE
181 unsigned int sectorsize = 512;
182 if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
183 sectorsize > bufsize)
184 bufsize = sectorsize;
186 #endif
187 #ifdef CONFIG_COCOA
188 u_int32_t blockSize = 512;
189 if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
190 bufsize = blockSize;
192 #endif
196 * offset and count are in bytes, but must be multiples of 512 for files
197 * opened with O_DIRECT. buf must be aligned to 512 bytes then.
199 * This function may be called without alignment if the caller ensures
200 * that O_DIRECT is not in effect.
202 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
203 uint8_t *buf, int count)
205 BDRVRawState *s = bs->opaque;
206 int ret;
208 ret = fd_open(bs);
209 if (ret < 0)
210 return ret;
212 if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
213 ++(s->lseek_err_cnt);
214 if(s->lseek_err_cnt <= 10) {
215 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
216 "] lseek failed : %d = %s\n",
217 s->fd, bs->filename, offset, buf, count,
218 bs->total_sectors, errno, strerror(errno));
220 return -1;
222 s->lseek_err_cnt=0;
224 ret = read(s->fd, buf, count);
225 if (ret == count)
226 goto label__raw_read__success;
228 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
229 "] read failed %d : %d = %s\n",
230 s->fd, bs->filename, offset, buf, count,
231 bs->total_sectors, ret, errno, strerror(errno));
233 /* Try harder for CDrom. */
234 if (bs->type == BDRV_TYPE_CDROM) {
235 lseek(s->fd, offset, SEEK_SET);
236 ret = read(s->fd, buf, count);
237 if (ret == count)
238 goto label__raw_read__success;
239 lseek(s->fd, offset, SEEK_SET);
240 ret = read(s->fd, buf, count);
241 if (ret == count)
242 goto label__raw_read__success;
244 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
245 "] retry read failed %d : %d = %s\n",
246 s->fd, bs->filename, offset, buf, count,
247 bs->total_sectors, ret, errno, strerror(errno));
250 label__raw_read__success:
252 return ret;
256 * offset and count are in bytes, but must be multiples of 512 for files
257 * opened with O_DIRECT. buf must be aligned to 512 bytes then.
259 * This function may be called without alignment if the caller ensures
260 * that O_DIRECT is not in effect.
262 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
263 const uint8_t *buf, int count)
265 BDRVRawState *s = bs->opaque;
266 int ret;
268 ret = fd_open(bs);
269 if (ret < 0)
270 return -errno;
272 if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
273 ++(s->lseek_err_cnt);
274 if(s->lseek_err_cnt) {
275 DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
276 PRId64 "] lseek failed : %d = %s\n",
277 s->fd, bs->filename, offset, buf, count,
278 bs->total_sectors, errno, strerror(errno));
280 return -EIO;
282 s->lseek_err_cnt = 0;
284 ret = write(s->fd, buf, count);
285 if (ret == count)
286 goto label__raw_write__success;
288 DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
289 "] write failed %d : %d = %s\n",
290 s->fd, bs->filename, offset, buf, count,
291 bs->total_sectors, ret, errno, strerror(errno));
293 label__raw_write__success:
295 return (ret < 0) ? -errno : ret;
300 * offset and count are in bytes and possibly not aligned. For files opened
301 * with O_DIRECT, necessary alignments are ensured before calling
302 * raw_pread_aligned to do the actual read.
304 static int raw_pread(BlockDriverState *bs, int64_t offset,
305 uint8_t *buf, int count)
307 BDRVRawState *s = bs->opaque;
308 int size, ret, shift, sum;
310 sum = 0;
312 if (s->aligned_buf != NULL) {
314 if (offset & 0x1ff) {
315 /* align offset on a 512 bytes boundary */
317 shift = offset & 0x1ff;
318 size = (shift + count + 0x1ff) & ~0x1ff;
319 if (size > ALIGNED_BUFFER_SIZE)
320 size = ALIGNED_BUFFER_SIZE;
321 ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
322 if (ret < 0)
323 return ret;
325 size = 512 - shift;
326 if (size > count)
327 size = count;
328 memcpy(buf, s->aligned_buf + shift, size);
330 buf += size;
331 offset += size;
332 count -= size;
333 sum += size;
335 if (count == 0)
336 return sum;
338 if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
340 /* read on aligned buffer */
342 while (count) {
344 size = (count + 0x1ff) & ~0x1ff;
345 if (size > ALIGNED_BUFFER_SIZE)
346 size = ALIGNED_BUFFER_SIZE;
348 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
349 if (ret < 0)
350 return ret;
352 size = ret;
353 if (size > count)
354 size = count;
356 memcpy(buf, s->aligned_buf, size);
358 buf += size;
359 offset += size;
360 count -= size;
361 sum += size;
364 return sum;
368 return raw_pread_aligned(bs, offset, buf, count) + sum;
371 static int raw_read(BlockDriverState *bs, int64_t sector_num,
372 uint8_t *buf, int nb_sectors)
374 int ret;
376 ret = raw_pread(bs, sector_num * 512, buf, nb_sectors * 512);
377 if (ret == (nb_sectors * 512))
378 ret = 0;
379 return ret;
383 * offset and count are in bytes and possibly not aligned. For files opened
384 * with O_DIRECT, necessary alignments are ensured before calling
385 * raw_pwrite_aligned to do the actual write.
387 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
388 const uint8_t *buf, int count)
390 BDRVRawState *s = bs->opaque;
391 int size, ret, shift, sum;
393 sum = 0;
395 if (s->aligned_buf != NULL) {
397 if (offset & 0x1ff) {
398 /* align offset on a 512 bytes boundary */
399 shift = offset & 0x1ff;
400 ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
401 if (ret < 0)
402 return ret;
404 size = 512 - shift;
405 if (size > count)
406 size = count;
407 memcpy(s->aligned_buf + shift, buf, size);
409 ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
410 if (ret < 0)
411 return ret;
413 buf += size;
414 offset += size;
415 count -= size;
416 sum += size;
418 if (count == 0)
419 return sum;
421 if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
423 while ((size = (count & ~0x1ff)) != 0) {
425 if (size > ALIGNED_BUFFER_SIZE)
426 size = ALIGNED_BUFFER_SIZE;
428 memcpy(s->aligned_buf, buf, size);
430 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
431 if (ret < 0)
432 return ret;
434 buf += ret;
435 offset += ret;
436 count -= ret;
437 sum += ret;
439 /* here, count < 512 because (count & ~0x1ff) == 0 */
440 if (count) {
441 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
442 if (ret < 0)
443 return ret;
444 memcpy(s->aligned_buf, buf, count);
446 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
447 if (ret < 0)
448 return ret;
449 if (count < ret)
450 ret = count;
452 sum += ret;
454 return sum;
457 return raw_pwrite_aligned(bs, offset, buf, count) + sum;
460 static int raw_write(BlockDriverState *bs, int64_t sector_num,
461 const uint8_t *buf, int nb_sectors)
463 int ret;
464 ret = raw_pwrite(bs, sector_num * 512, buf, nb_sectors * 512);
465 if (ret == (nb_sectors * 512))
466 ret = 0;
467 return ret;
470 #ifdef CONFIG_AIO
471 /***********************************************************/
472 /* Unix AIO using POSIX AIO */
474 typedef struct RawAIOCB {
475 BlockDriverAIOCB common;
476 struct qemu_paiocb aiocb;
477 struct RawAIOCB *next;
478 int ret;
479 } RawAIOCB;
481 typedef struct PosixAioState
483 int rfd, wfd;
484 RawAIOCB *first_aio;
485 } PosixAioState;
487 static void posix_aio_read(void *opaque)
489 PosixAioState *s = opaque;
490 RawAIOCB *acb, **pacb;
491 int ret;
492 ssize_t len;
494 /* read all bytes from signal pipe */
495 for (;;) {
496 char bytes[16];
498 len = read(s->rfd, bytes, sizeof(bytes));
499 if (len == -1 && errno == EINTR)
500 continue; /* try again */
501 if (len == sizeof(bytes))
502 continue; /* more to read */
503 break;
506 for(;;) {
507 pacb = &s->first_aio;
508 for(;;) {
509 acb = *pacb;
510 if (!acb)
511 goto the_end;
512 ret = qemu_paio_error(&acb->aiocb);
513 if (ret == ECANCELED) {
514 /* remove the request */
515 *pacb = acb->next;
516 qemu_aio_release(acb);
517 } else if (ret != EINPROGRESS) {
518 /* end of aio */
519 if (ret == 0) {
520 ret = qemu_paio_return(&acb->aiocb);
521 if (ret == acb->aiocb.aio_nbytes)
522 ret = 0;
523 else
524 ret = -EINVAL;
525 } else {
526 ret = -ret;
528 /* remove the request */
529 *pacb = acb->next;
530 /* call the callback */
531 acb->common.cb(acb->common.opaque, ret);
532 qemu_aio_release(acb);
533 break;
534 } else {
535 pacb = &acb->next;
539 the_end: ;
542 static int posix_aio_flush(void *opaque)
544 PosixAioState *s = opaque;
545 return !!s->first_aio;
548 static PosixAioState *posix_aio_state;
550 static void aio_signal_handler(int signum)
552 if (posix_aio_state) {
553 char byte = 0;
555 write(posix_aio_state->wfd, &byte, sizeof(byte));
558 qemu_service_io();
561 static int posix_aio_init(void)
563 struct sigaction act;
564 PosixAioState *s;
565 int fds[2];
566 struct qemu_paioinit ai;
568 if (posix_aio_state)
569 return 0;
571 s = qemu_malloc(sizeof(PosixAioState));
573 sigfillset(&act.sa_mask);
574 act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
575 act.sa_handler = aio_signal_handler;
576 sigaction(SIGUSR2, &act, NULL);
578 s->first_aio = NULL;
579 if (pipe(fds) == -1) {
580 fprintf(stderr, "failed to create pipe\n");
581 return -errno;
584 s->rfd = fds[0];
585 s->wfd = fds[1];
587 fcntl(s->rfd, F_SETFL, O_NONBLOCK);
588 fcntl(s->wfd, F_SETFL, O_NONBLOCK);
590 qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);
592 memset(&ai, 0, sizeof(ai));
593 ai.aio_threads = 64;
594 ai.aio_num = 64;
595 qemu_paio_init(&ai);
597 posix_aio_state = s;
599 return 0;
602 static RawAIOCB *raw_aio_setup(BlockDriverState *bs, int64_t sector_num,
603 QEMUIOVector *qiov, int nb_sectors,
604 BlockDriverCompletionFunc *cb, void *opaque)
606 BDRVRawState *s = bs->opaque;
607 RawAIOCB *acb;
609 if (fd_open(bs) < 0)
610 return NULL;
612 acb = qemu_aio_get(bs, cb, opaque);
613 if (!acb)
614 return NULL;
615 acb->aiocb.aio_fildes = s->fd;
616 acb->aiocb.ev_signo = SIGUSR2;
617 acb->aiocb.aio_iov = qiov->iov;
618 acb->aiocb.aio_niov = qiov->niov;
619 acb->aiocb.aio_nbytes = nb_sectors * 512;
620 acb->aiocb.aio_offset = sector_num * 512;
621 acb->aiocb.aio_flags = 0;
624 * If O_DIRECT is used the buffer needs to be aligned on a sector
625 * boundary. Tell the low level code to ensure that in case it's
626 * not done yet.
628 if (s->aligned_buf)
629 acb->aiocb.aio_flags |= QEMU_AIO_SECTOR_ALIGNED;
631 acb->next = posix_aio_state->first_aio;
632 posix_aio_state->first_aio = acb;
633 return acb;
636 static void raw_aio_remove(RawAIOCB *acb)
638 RawAIOCB **pacb;
640 /* remove the callback from the queue */
641 pacb = &posix_aio_state->first_aio;
642 for(;;) {
643 if (*pacb == NULL) {
644 fprintf(stderr, "raw_aio_remove: aio request not found!\n");
645 break;
646 } else if (*pacb == acb) {
647 *pacb = acb->next;
648 qemu_aio_release(acb);
649 break;
651 pacb = &(*pacb)->next;
655 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
656 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
657 BlockDriverCompletionFunc *cb, void *opaque)
659 RawAIOCB *acb;
661 acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque);
662 if (!acb)
663 return NULL;
664 if (qemu_paio_read(&acb->aiocb) < 0) {
665 raw_aio_remove(acb);
666 return NULL;
668 return &acb->common;
671 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
672 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
673 BlockDriverCompletionFunc *cb, void *opaque)
675 RawAIOCB *acb;
677 acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque);
678 if (!acb)
679 return NULL;
680 if (qemu_paio_write(&acb->aiocb) < 0) {
681 raw_aio_remove(acb);
682 return NULL;
684 return &acb->common;
687 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
689 int ret;
690 RawAIOCB *acb = (RawAIOCB *)blockacb;
692 ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
693 if (ret == QEMU_PAIO_NOTCANCELED) {
694 /* fail safe: if the aio could not be canceled, we wait for
695 it */
696 while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
699 raw_aio_remove(acb);
701 #else /* CONFIG_AIO */
702 static int posix_aio_init(void)
704 return 0;
706 #endif /* CONFIG_AIO */
709 static void raw_close(BlockDriverState *bs)
711 BDRVRawState *s = bs->opaque;
712 if (s->fd >= 0) {
713 close(s->fd);
714 s->fd = -1;
715 if (s->aligned_buf != NULL)
716 qemu_free(s->aligned_buf);
720 static int raw_truncate(BlockDriverState *bs, int64_t offset)
722 BDRVRawState *s = bs->opaque;
723 if (s->type != FTYPE_FILE)
724 return -ENOTSUP;
725 if (ftruncate(s->fd, offset) < 0)
726 return -errno;
727 return 0;
730 #ifdef __OpenBSD__
731 static int64_t raw_getlength(BlockDriverState *bs)
733 BDRVRawState *s = bs->opaque;
734 int fd = s->fd;
735 struct stat st;
737 if (fstat(fd, &st))
738 return -1;
739 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
740 struct disklabel dl;
742 if (ioctl(fd, DIOCGDINFO, &dl))
743 return -1;
744 return (uint64_t)dl.d_secsize *
745 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
746 } else
747 return st.st_size;
749 #else /* !__OpenBSD__ */
750 static int64_t raw_getlength(BlockDriverState *bs)
752 BDRVRawState *s = bs->opaque;
753 int fd = s->fd;
754 int64_t size;
755 #ifdef HOST_BSD
756 struct stat sb;
757 #ifdef __FreeBSD__
758 int reopened = 0;
759 #endif
760 #endif
761 #ifdef __sun__
762 struct dk_minfo minfo;
763 int rv;
764 #endif
765 int ret;
767 ret = fd_open(bs);
768 if (ret < 0)
769 return ret;
771 #ifdef HOST_BSD
772 #ifdef __FreeBSD__
773 again:
774 #endif
775 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
776 #ifdef DIOCGMEDIASIZE
777 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
778 #elif defined(DIOCGPART)
780 struct partinfo pi;
781 if (ioctl(fd, DIOCGPART, &pi) == 0)
782 size = pi.media_size;
783 else
784 size = 0;
786 if (size == 0)
787 #endif
788 #ifdef CONFIG_COCOA
789 size = LONG_LONG_MAX;
790 #else
791 size = lseek(fd, 0LL, SEEK_END);
792 #endif
793 #ifdef __FreeBSD__
794 switch(s->type) {
795 case FTYPE_CD:
796 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
797 if (size == 2048LL * (unsigned)-1)
798 size = 0;
799 /* XXX no disc? maybe we need to reopen... */
800 if (size <= 0 && !reopened && cd_open(bs) >= 0) {
801 reopened = 1;
802 goto again;
805 #endif
806 } else
807 #endif
808 #ifdef __sun__
810 * use the DKIOCGMEDIAINFO ioctl to read the size.
812 rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
813 if ( rv != -1 ) {
814 size = minfo.dki_lbsize * minfo.dki_capacity;
815 } else /* there are reports that lseek on some devices
816 fails, but irc discussion said that contingency
817 on contingency was overkill */
818 #endif
820 size = lseek(fd, 0, SEEK_END);
822 return size;
824 #endif
826 static int raw_create(const char *filename, int64_t total_size,
827 const char *backing_file, int flags)
829 int fd;
831 if (flags || backing_file)
832 return -ENOTSUP;
834 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
835 0644);
836 if (fd < 0)
837 return -EIO;
838 ftruncate(fd, total_size * 512);
839 close(fd);
840 return 0;
843 static void raw_flush(BlockDriverState *bs)
845 BDRVRawState *s = bs->opaque;
846 fsync(s->fd);
849 BlockDriver bdrv_raw = {
850 .format_name = "raw",
851 .instance_size = sizeof(BDRVRawState),
852 .bdrv_probe = NULL, /* no probe for protocols */
853 .bdrv_open = raw_open,
854 .bdrv_read = raw_read,
855 .bdrv_write = raw_write,
856 .bdrv_close = raw_close,
857 .bdrv_create = raw_create,
858 .bdrv_flush = raw_flush,
860 #ifdef CONFIG_AIO
861 .bdrv_aio_readv = raw_aio_readv,
862 .bdrv_aio_writev = raw_aio_writev,
863 .bdrv_aio_cancel = raw_aio_cancel,
864 .aiocb_size = sizeof(RawAIOCB),
865 #endif
867 .bdrv_truncate = raw_truncate,
868 .bdrv_getlength = raw_getlength,
871 /***********************************************/
872 /* host device */
874 #ifdef CONFIG_COCOA
875 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
876 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
878 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
880 kern_return_t kernResult;
881 mach_port_t masterPort;
882 CFMutableDictionaryRef classesToMatch;
884 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
885 if ( KERN_SUCCESS != kernResult ) {
886 printf( "IOMasterPort returned %d\n", kernResult );
889 classesToMatch = IOServiceMatching( kIOCDMediaClass );
890 if ( classesToMatch == NULL ) {
891 printf( "IOServiceMatching returned a NULL dictionary.\n" );
892 } else {
893 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
895 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
896 if ( KERN_SUCCESS != kernResult )
898 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
901 return kernResult;
904 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
906 io_object_t nextMedia;
907 kern_return_t kernResult = KERN_FAILURE;
908 *bsdPath = '\0';
909 nextMedia = IOIteratorNext( mediaIterator );
910 if ( nextMedia )
912 CFTypeRef bsdPathAsCFString;
913 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
914 if ( bsdPathAsCFString ) {
915 size_t devPathLength;
916 strcpy( bsdPath, _PATH_DEV );
917 strcat( bsdPath, "r" );
918 devPathLength = strlen( bsdPath );
919 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
920 kernResult = KERN_SUCCESS;
922 CFRelease( bsdPathAsCFString );
924 IOObjectRelease( nextMedia );
927 return kernResult;
930 #endif
932 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
934 BDRVRawState *s = bs->opaque;
935 int fd, open_flags, ret;
937 posix_aio_init();
939 #ifdef CONFIG_COCOA
940 if (strstart(filename, "/dev/cdrom", NULL)) {
941 kern_return_t kernResult;
942 io_iterator_t mediaIterator;
943 char bsdPath[ MAXPATHLEN ];
944 int fd;
946 kernResult = FindEjectableCDMedia( &mediaIterator );
947 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
949 if ( bsdPath[ 0 ] != '\0' ) {
950 strcat(bsdPath,"s0");
951 /* some CDs don't have a partition 0 */
952 fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
953 if (fd < 0) {
954 bsdPath[strlen(bsdPath)-1] = '1';
955 } else {
956 close(fd);
958 filename = bsdPath;
961 if ( mediaIterator )
962 IOObjectRelease( mediaIterator );
964 #endif
965 open_flags = O_BINARY;
966 if ((flags & BDRV_O_ACCESS) == O_RDWR) {
967 open_flags |= O_RDWR;
968 } else {
969 open_flags |= O_RDONLY;
970 bs->read_only = 1;
972 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
973 * and O_DIRECT for no caching. */
974 if ((flags & BDRV_O_NOCACHE))
975 open_flags |= O_DIRECT;
976 else if (!(flags & BDRV_O_CACHE_WB))
977 open_flags |= O_DSYNC;
979 s->type = FTYPE_FILE;
980 #if defined(__linux__)
981 if (strstart(filename, "/dev/cd", NULL)) {
982 /* open will not fail even if no CD is inserted */
983 open_flags |= O_NONBLOCK;
984 s->type = FTYPE_CD;
985 } else if (strstart(filename, "/dev/fd", NULL)) {
986 s->type = FTYPE_FD;
987 s->fd_open_flags = open_flags;
988 /* open will not fail even if no floppy is inserted */
989 open_flags |= O_NONBLOCK;
990 #ifdef CONFIG_AIO
991 } else if (strstart(filename, "/dev/sg", NULL)) {
992 bs->sg = 1;
993 #endif
995 #endif
996 #if defined(__FreeBSD__)
997 if (strstart(filename, "/dev/cd", NULL) ||
998 strstart(filename, "/dev/acd", NULL)) {
999 s->type = FTYPE_CD;
1000 s->cd_open_flags = open_flags;
1002 #endif
1003 s->fd = -1;
1004 fd = open(filename, open_flags, 0644);
1005 if (fd < 0) {
1006 ret = -errno;
1007 if (ret == -EROFS)
1008 ret = -EACCES;
1009 return ret;
1011 s->fd = fd;
1012 #if defined(__FreeBSD__)
1013 /* make sure the door isnt locked at this time */
1014 if (s->type == FTYPE_CD)
1015 ioctl (s->fd, CDIOCALLOW);
1016 #endif
1017 #if defined(__linux__)
1018 /* close fd so that we can reopen it as needed */
1019 if (s->type == FTYPE_FD) {
1020 close(s->fd);
1021 s->fd = -1;
1022 s->fd_media_changed = 1;
1024 #endif
1025 return 0;
1028 #if defined(__linux__)
1029 /* Note: we do not have a reliable method to detect if the floppy is
1030 present. The current method is to try to open the floppy at every
1031 I/O and to keep it opened during a few hundreds of ms. */
1032 static int fd_open(BlockDriverState *bs)
1034 BDRVRawState *s = bs->opaque;
1035 int last_media_present;
1037 if (s->type != FTYPE_FD)
1038 return 0;
1039 last_media_present = (s->fd >= 0);
1040 if (s->fd >= 0 &&
1041 (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1042 close(s->fd);
1043 s->fd = -1;
1044 #ifdef DEBUG_FLOPPY
1045 printf("Floppy closed\n");
1046 #endif
1048 if (s->fd < 0) {
1049 if (s->fd_got_error &&
1050 (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1051 #ifdef DEBUG_FLOPPY
1052 printf("No floppy (open delayed)\n");
1053 #endif
1054 return -EIO;
1056 s->fd = open(bs->filename, s->fd_open_flags);
1057 if (s->fd < 0) {
1058 s->fd_error_time = qemu_get_clock(rt_clock);
1059 s->fd_got_error = 1;
1060 if (last_media_present)
1061 s->fd_media_changed = 1;
1062 #ifdef DEBUG_FLOPPY
1063 printf("No floppy\n");
1064 #endif
1065 return -EIO;
1067 #ifdef DEBUG_FLOPPY
1068 printf("Floppy opened\n");
1069 #endif
1071 if (!last_media_present)
1072 s->fd_media_changed = 1;
1073 s->fd_open_time = qemu_get_clock(rt_clock);
1074 s->fd_got_error = 0;
1075 return 0;
1078 static int raw_is_inserted(BlockDriverState *bs)
1080 BDRVRawState *s = bs->opaque;
1081 int ret;
1083 switch(s->type) {
1084 case FTYPE_CD:
1085 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1086 if (ret == CDS_DISC_OK)
1087 return 1;
1088 else
1089 return 0;
1090 break;
1091 case FTYPE_FD:
1092 ret = fd_open(bs);
1093 return (ret >= 0);
1094 default:
1095 return 1;
1099 /* currently only used by fdc.c, but a CD version would be good too */
1100 static int raw_media_changed(BlockDriverState *bs)
1102 BDRVRawState *s = bs->opaque;
1104 switch(s->type) {
1105 case FTYPE_FD:
1107 int ret;
1108 /* XXX: we do not have a true media changed indication. It
1109 does not work if the floppy is changed without trying
1110 to read it */
1111 fd_open(bs);
1112 ret = s->fd_media_changed;
1113 s->fd_media_changed = 0;
1114 #ifdef DEBUG_FLOPPY
1115 printf("Floppy changed=%d\n", ret);
1116 #endif
1117 return ret;
1119 default:
1120 return -ENOTSUP;
1124 static int raw_eject(BlockDriverState *bs, int eject_flag)
1126 BDRVRawState *s = bs->opaque;
1128 switch(s->type) {
1129 case FTYPE_CD:
1130 if (eject_flag) {
1131 if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1132 perror("CDROMEJECT");
1133 } else {
1134 if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1135 perror("CDROMEJECT");
1137 break;
1138 case FTYPE_FD:
1140 int fd;
1141 if (s->fd >= 0) {
1142 close(s->fd);
1143 s->fd = -1;
1145 fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1146 if (fd >= 0) {
1147 if (ioctl(fd, FDEJECT, 0) < 0)
1148 perror("FDEJECT");
1149 close(fd);
1152 break;
1153 default:
1154 return -ENOTSUP;
1156 return 0;
1159 static int raw_set_locked(BlockDriverState *bs, int locked)
1161 BDRVRawState *s = bs->opaque;
1163 switch(s->type) {
1164 case FTYPE_CD:
1165 if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1166 /* Note: an error can happen if the distribution automatically
1167 mounts the CD-ROM */
1168 // perror("CDROM_LOCKDOOR");
1170 break;
1171 default:
1172 return -ENOTSUP;
1174 return 0;
1177 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1179 BDRVRawState *s = bs->opaque;
1181 return ioctl(s->fd, req, buf);
1184 #ifdef CONFIG_AIO
1185 static BlockDriverAIOCB *raw_aio_ioctl(BlockDriverState *bs,
1186 unsigned long int req, void *buf,
1187 BlockDriverCompletionFunc *cb, void *opaque)
1189 BDRVRawState *s = bs->opaque;
1190 RawAIOCB *acb;
1192 if (fd_open(bs) < 0)
1193 return NULL;
1195 acb = qemu_aio_get(bs, cb, opaque);
1196 if (!acb)
1197 return NULL;
1198 acb->aiocb.aio_fildes = s->fd;
1199 acb->aiocb.ev_signo = SIGUSR2;
1200 acb->aiocb.aio_offset = 0;
1201 acb->aiocb.aio_flags = 0;
1203 acb->next = posix_aio_state->first_aio;
1204 posix_aio_state->first_aio = acb;
1206 acb->aiocb.aio_ioctl_buf = buf;
1207 acb->aiocb.aio_ioctl_cmd = req;
1208 if (qemu_paio_ioctl(&acb->aiocb) < 0) {
1209 raw_aio_remove(acb);
1210 return NULL;
1213 return &acb->common;
1215 #endif
1217 #elif defined(__FreeBSD__)
1219 static int fd_open(BlockDriverState *bs)
1221 BDRVRawState *s = bs->opaque;
1223 /* this is just to ensure s->fd is sane (its called by io ops) */
1224 if (s->fd >= 0)
1225 return 0;
1226 return -EIO;
1229 static int cd_open(BlockDriverState *bs)
1231 #if defined(__FreeBSD__)
1232 BDRVRawState *s = bs->opaque;
1233 int fd;
1235 switch(s->type) {
1236 case FTYPE_CD:
1237 /* XXX force reread of possibly changed/newly loaded disc,
1238 * FreeBSD seems to not notice sometimes... */
1239 if (s->fd >= 0)
1240 close (s->fd);
1241 fd = open(bs->filename, s->cd_open_flags, 0644);
1242 if (fd < 0) {
1243 s->fd = -1;
1244 return -EIO;
1246 s->fd = fd;
1247 /* make sure the door isnt locked at this time */
1248 ioctl (s->fd, CDIOCALLOW);
1250 #endif
1251 return 0;
1254 static int raw_is_inserted(BlockDriverState *bs)
1256 BDRVRawState *s = bs->opaque;
1258 switch(s->type) {
1259 case FTYPE_CD:
1260 return (raw_getlength(bs) > 0);
1261 case FTYPE_FD:
1262 /* XXX handle this */
1263 /* FALLTHRU */
1264 default:
1265 return 1;
1269 static int raw_media_changed(BlockDriverState *bs)
1271 return -ENOTSUP;
1274 static int raw_eject(BlockDriverState *bs, int eject_flag)
1276 BDRVRawState *s = bs->opaque;
1278 switch(s->type) {
1279 case FTYPE_CD:
1280 if (s->fd < 0)
1281 return -ENOTSUP;
1282 (void) ioctl (s->fd, CDIOCALLOW);
1283 if (eject_flag) {
1284 if (ioctl (s->fd, CDIOCEJECT) < 0)
1285 perror("CDIOCEJECT");
1286 } else {
1287 if (ioctl (s->fd, CDIOCCLOSE) < 0)
1288 perror("CDIOCCLOSE");
1290 if (cd_open(bs) < 0)
1291 return -ENOTSUP;
1292 break;
1293 case FTYPE_FD:
1294 /* XXX handle this */
1295 /* FALLTHRU */
1296 default:
1297 return -ENOTSUP;
1299 return 0;
1302 static int raw_set_locked(BlockDriverState *bs, int locked)
1304 BDRVRawState *s = bs->opaque;
1306 switch(s->type) {
1307 case FTYPE_CD:
1308 if (s->fd < 0)
1309 return -ENOTSUP;
1310 if (ioctl (s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1311 /* Note: an error can happen if the distribution automatically
1312 mounts the CD-ROM */
1313 // perror("CDROM_LOCKDOOR");
1315 break;
1316 default:
1317 return -ENOTSUP;
1319 return 0;
1322 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1324 return -ENOTSUP;
1326 #else /* !linux && !FreeBSD */
1328 static int fd_open(BlockDriverState *bs)
1330 return 0;
1333 static int raw_is_inserted(BlockDriverState *bs)
1335 return 1;
1338 static int raw_media_changed(BlockDriverState *bs)
1340 return -ENOTSUP;
1343 static int raw_eject(BlockDriverState *bs, int eject_flag)
1345 return -ENOTSUP;
1348 static int raw_set_locked(BlockDriverState *bs, int locked)
1350 return -ENOTSUP;
1353 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1355 return -ENOTSUP;
1358 static BlockDriverAIOCB *raw_aio_ioctl(BlockDriverState *bs,
1359 unsigned long int req, void *buf,
1360 BlockDriverCompletionFunc *cb, void *opaque)
1362 return NULL;
1364 #endif /* !linux && !FreeBSD */
1366 #if defined(__linux__) || defined(__FreeBSD__)
1367 static int hdev_create(const char *filename, int64_t total_size,
1368 const char *backing_file, int flags)
1370 int fd;
1371 int ret = 0;
1372 struct stat stat_buf;
1374 if (flags || backing_file)
1375 return -ENOTSUP;
1377 fd = open(filename, O_WRONLY | O_BINARY);
1378 if (fd < 0)
1379 return -EIO;
1381 if (fstat(fd, &stat_buf) < 0)
1382 ret = -EIO;
1383 else if (!S_ISBLK(stat_buf.st_mode))
1384 ret = -EIO;
1385 else if (lseek(fd, 0, SEEK_END) < total_size * 512)
1386 ret = -ENOSPC;
1388 close(fd);
1389 return ret;
1392 #else /* !(linux || freebsd) */
1394 static int hdev_create(const char *filename, int64_t total_size,
1395 const char *backing_file, int flags)
1397 return -ENOTSUP;
1399 #endif
1401 BlockDriver bdrv_host_device = {
1402 .format_name = "host_device",
1403 .instance_size = sizeof(BDRVRawState),
1404 .bdrv_open = hdev_open,
1405 .bdrv_close = raw_close,
1406 .bdrv_create = hdev_create,
1407 .bdrv_flush = raw_flush,
1409 #ifdef CONFIG_AIO
1410 .bdrv_aio_readv = raw_aio_readv,
1411 .bdrv_aio_writev = raw_aio_writev,
1412 .bdrv_aio_cancel = raw_aio_cancel,
1413 .aiocb_size = sizeof(RawAIOCB),
1414 #endif
1416 .bdrv_read = raw_read,
1417 .bdrv_write = raw_write,
1418 .bdrv_getlength = raw_getlength,
1420 /* removable device support */
1421 .bdrv_is_inserted = raw_is_inserted,
1422 .bdrv_media_changed = raw_media_changed,
1423 .bdrv_eject = raw_eject,
1424 .bdrv_set_locked = raw_set_locked,
1425 /* generic scsi device */
1426 .bdrv_ioctl = raw_ioctl,
1427 #ifdef CONFIG_AIO
1428 .bdrv_aio_ioctl = raw_aio_ioctl,
1429 #endif