Implement cancellation method for dma async I/O (Avi Kivity)
[qemu-kvm/fedora.git] / block-raw-posix.c
blob1a1a1781bc51693d6e622b3ec12cff3832005432
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 #endif
60 #ifdef __OpenBSD__
61 #include <sys/ioctl.h>
62 #include <sys/disklabel.h>
63 #include <sys/dkio.h>
64 #endif
66 #ifdef __DragonFly__
67 #include <sys/ioctl.h>
68 #include <sys/diskslice.h>
69 #endif
71 //#define DEBUG_FLOPPY
73 //#define DEBUG_BLOCK
74 #if defined(DEBUG_BLOCK)
75 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (qemu_log_enabled()) \
76 { qemu_log(formatCstr, ##args); qemu_log_flush(); } } while (0)
77 #else
78 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
79 #endif
81 /* OS X does not have O_DSYNC */
82 #ifndef O_DSYNC
83 #define O_DSYNC O_SYNC
84 #endif
86 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
87 #ifndef O_DIRECT
88 #define O_DIRECT O_DSYNC
89 #endif
91 #define FTYPE_FILE 0
92 #define FTYPE_CD 1
93 #define FTYPE_FD 2
95 #define ALIGNED_BUFFER_SIZE (32 * 512)
97 /* if the FD is not accessed during that time (in ms), we try to
98 reopen it to see if the disk has been changed */
99 #define FD_OPEN_TIMEOUT 1000
101 typedef struct BDRVRawState {
102 int fd;
103 int type;
104 unsigned int lseek_err_cnt;
105 #if defined(__linux__)
106 /* linux floppy specific */
107 int fd_open_flags;
108 int64_t fd_open_time;
109 int64_t fd_error_time;
110 int fd_got_error;
111 int fd_media_changed;
112 #endif
113 uint8_t* aligned_buf;
114 } BDRVRawState;
116 static int posix_aio_init(void);
118 static int fd_open(BlockDriverState *bs);
120 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
122 BDRVRawState *s = bs->opaque;
123 int fd, open_flags, ret;
125 posix_aio_init();
127 s->lseek_err_cnt = 0;
129 open_flags = O_BINARY;
130 if ((flags & BDRV_O_ACCESS) == O_RDWR) {
131 open_flags |= O_RDWR;
132 } else {
133 open_flags |= O_RDONLY;
134 bs->read_only = 1;
136 if (flags & BDRV_O_CREAT)
137 open_flags |= O_CREAT | O_TRUNC;
139 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
140 * and O_DIRECT for no caching. */
141 if ((flags & BDRV_O_NOCACHE))
142 open_flags |= O_DIRECT;
143 else if (!(flags & BDRV_O_CACHE_WB))
144 open_flags |= O_DSYNC;
146 s->type = FTYPE_FILE;
148 fd = open(filename, open_flags, 0644);
149 if (fd < 0) {
150 ret = -errno;
151 if (ret == -EROFS)
152 ret = -EACCES;
153 return ret;
155 s->fd = fd;
156 s->aligned_buf = NULL;
157 if ((flags & BDRV_O_NOCACHE)) {
158 s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
159 if (s->aligned_buf == NULL) {
160 ret = -errno;
161 close(fd);
162 return ret;
165 return 0;
168 /* XXX: use host sector size if necessary with:
169 #ifdef DIOCGSECTORSIZE
171 unsigned int sectorsize = 512;
172 if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
173 sectorsize > bufsize)
174 bufsize = sectorsize;
176 #endif
177 #ifdef CONFIG_COCOA
178 u_int32_t blockSize = 512;
179 if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
180 bufsize = blockSize;
182 #endif
186 * offset and count are in bytes, but must be multiples of 512 for files
187 * opened with O_DIRECT. buf must be aligned to 512 bytes then.
189 * This function may be called without alignment if the caller ensures
190 * that O_DIRECT is not in effect.
192 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
193 uint8_t *buf, int count)
195 BDRVRawState *s = bs->opaque;
196 int ret;
198 ret = fd_open(bs);
199 if (ret < 0)
200 return ret;
202 if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
203 ++(s->lseek_err_cnt);
204 if(s->lseek_err_cnt <= 10) {
205 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
206 "] lseek failed : %d = %s\n",
207 s->fd, bs->filename, offset, buf, count,
208 bs->total_sectors, errno, strerror(errno));
210 return -1;
212 s->lseek_err_cnt=0;
214 ret = read(s->fd, buf, count);
215 if (ret == count)
216 goto label__raw_read__success;
218 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
219 "] read failed %d : %d = %s\n",
220 s->fd, bs->filename, offset, buf, count,
221 bs->total_sectors, ret, errno, strerror(errno));
223 /* Try harder for CDrom. */
224 if (bs->type == BDRV_TYPE_CDROM) {
225 lseek(s->fd, offset, SEEK_SET);
226 ret = read(s->fd, buf, count);
227 if (ret == count)
228 goto label__raw_read__success;
229 lseek(s->fd, offset, SEEK_SET);
230 ret = read(s->fd, buf, count);
231 if (ret == count)
232 goto label__raw_read__success;
234 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
235 "] retry read failed %d : %d = %s\n",
236 s->fd, bs->filename, offset, buf, count,
237 bs->total_sectors, ret, errno, strerror(errno));
240 label__raw_read__success:
242 return ret;
246 * offset and count are in bytes, but must be multiples of 512 for files
247 * opened with O_DIRECT. buf must be aligned to 512 bytes then.
249 * This function may be called without alignment if the caller ensures
250 * that O_DIRECT is not in effect.
252 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
253 const uint8_t *buf, int count)
255 BDRVRawState *s = bs->opaque;
256 int ret;
258 ret = fd_open(bs);
259 if (ret < 0)
260 return -errno;
262 if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
263 ++(s->lseek_err_cnt);
264 if(s->lseek_err_cnt) {
265 DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
266 PRId64 "] lseek failed : %d = %s\n",
267 s->fd, bs->filename, offset, buf, count,
268 bs->total_sectors, errno, strerror(errno));
270 return -EIO;
272 s->lseek_err_cnt = 0;
274 ret = write(s->fd, buf, count);
275 if (ret == count)
276 goto label__raw_write__success;
278 DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
279 "] write failed %d : %d = %s\n",
280 s->fd, bs->filename, offset, buf, count,
281 bs->total_sectors, ret, errno, strerror(errno));
283 label__raw_write__success:
285 return (ret < 0) ? -errno : ret;
290 * offset and count are in bytes and possibly not aligned. For files opened
291 * with O_DIRECT, necessary alignments are ensured before calling
292 * raw_pread_aligned to do the actual read.
294 static int raw_pread(BlockDriverState *bs, int64_t offset,
295 uint8_t *buf, int count)
297 BDRVRawState *s = bs->opaque;
298 int size, ret, shift, sum;
300 sum = 0;
302 if (s->aligned_buf != NULL) {
304 if (offset & 0x1ff) {
305 /* align offset on a 512 bytes boundary */
307 shift = offset & 0x1ff;
308 size = (shift + count + 0x1ff) & ~0x1ff;
309 if (size > ALIGNED_BUFFER_SIZE)
310 size = ALIGNED_BUFFER_SIZE;
311 ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
312 if (ret < 0)
313 return ret;
315 size = 512 - shift;
316 if (size > count)
317 size = count;
318 memcpy(buf, s->aligned_buf + shift, size);
320 buf += size;
321 offset += size;
322 count -= size;
323 sum += size;
325 if (count == 0)
326 return sum;
328 if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
330 /* read on aligned buffer */
332 while (count) {
334 size = (count + 0x1ff) & ~0x1ff;
335 if (size > ALIGNED_BUFFER_SIZE)
336 size = ALIGNED_BUFFER_SIZE;
338 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
339 if (ret < 0)
340 return ret;
342 size = ret;
343 if (size > count)
344 size = count;
346 memcpy(buf, s->aligned_buf, size);
348 buf += size;
349 offset += size;
350 count -= size;
351 sum += size;
354 return sum;
358 return raw_pread_aligned(bs, offset, buf, count) + sum;
361 static int raw_read(BlockDriverState *bs, int64_t sector_num,
362 uint8_t *buf, int nb_sectors)
364 int ret;
366 ret = raw_pread(bs, sector_num * 512, buf, nb_sectors * 512);
367 if (ret == (nb_sectors * 512))
368 ret = 0;
369 return ret;
373 * offset and count are in bytes and possibly not aligned. For files opened
374 * with O_DIRECT, necessary alignments are ensured before calling
375 * raw_pwrite_aligned to do the actual write.
377 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
378 const uint8_t *buf, int count)
380 BDRVRawState *s = bs->opaque;
381 int size, ret, shift, sum;
383 sum = 0;
385 if (s->aligned_buf != NULL) {
387 if (offset & 0x1ff) {
388 /* align offset on a 512 bytes boundary */
389 shift = offset & 0x1ff;
390 ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
391 if (ret < 0)
392 return ret;
394 size = 512 - shift;
395 if (size > count)
396 size = count;
397 memcpy(s->aligned_buf + shift, buf, size);
399 ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
400 if (ret < 0)
401 return ret;
403 buf += size;
404 offset += size;
405 count -= size;
406 sum += size;
408 if (count == 0)
409 return sum;
411 if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
413 while ((size = (count & ~0x1ff)) != 0) {
415 if (size > ALIGNED_BUFFER_SIZE)
416 size = ALIGNED_BUFFER_SIZE;
418 memcpy(s->aligned_buf, buf, size);
420 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
421 if (ret < 0)
422 return ret;
424 buf += ret;
425 offset += ret;
426 count -= ret;
427 sum += ret;
429 /* here, count < 512 because (count & ~0x1ff) == 0 */
430 if (count) {
431 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
432 if (ret < 0)
433 return ret;
434 memcpy(s->aligned_buf, buf, count);
436 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
437 if (ret < 0)
438 return ret;
439 if (count < ret)
440 ret = count;
442 sum += ret;
444 return sum;
447 return raw_pwrite_aligned(bs, offset, buf, count) + sum;
450 static int raw_write(BlockDriverState *bs, int64_t sector_num,
451 const uint8_t *buf, int nb_sectors)
453 int ret;
454 ret = raw_pwrite(bs, sector_num * 512, buf, nb_sectors * 512);
455 if (ret == (nb_sectors * 512))
456 ret = 0;
457 return ret;
460 #ifdef CONFIG_AIO
461 /***********************************************************/
462 /* Unix AIO using POSIX AIO */
464 typedef struct RawAIOCB {
465 BlockDriverAIOCB common;
466 struct qemu_paiocb aiocb;
467 struct RawAIOCB *next;
468 int ret;
469 } RawAIOCB;
471 typedef struct PosixAioState
473 int rfd, wfd;
474 RawAIOCB *first_aio;
475 } PosixAioState;
477 static void posix_aio_read(void *opaque)
479 PosixAioState *s = opaque;
480 RawAIOCB *acb, **pacb;
481 int ret;
482 ssize_t len;
484 /* read all bytes from signal pipe */
485 for (;;) {
486 char bytes[16];
488 len = read(s->rfd, bytes, sizeof(bytes));
489 if (len == -1 && errno == EINTR)
490 continue; /* try again */
491 if (len == sizeof(bytes))
492 continue; /* more to read */
493 break;
496 for(;;) {
497 pacb = &s->first_aio;
498 for(;;) {
499 acb = *pacb;
500 if (!acb)
501 goto the_end;
502 ret = qemu_paio_error(&acb->aiocb);
503 if (ret == ECANCELED) {
504 /* remove the request */
505 *pacb = acb->next;
506 qemu_aio_release(acb);
507 } else if (ret != EINPROGRESS) {
508 /* end of aio */
509 if (ret == 0) {
510 ret = qemu_paio_return(&acb->aiocb);
511 if (ret == acb->aiocb.aio_nbytes)
512 ret = 0;
513 else
514 ret = -EINVAL;
515 } else {
516 ret = -ret;
518 /* remove the request */
519 *pacb = acb->next;
520 /* call the callback */
521 acb->common.cb(acb->common.opaque, ret);
522 qemu_aio_release(acb);
523 break;
524 } else {
525 pacb = &acb->next;
529 the_end: ;
532 static int posix_aio_flush(void *opaque)
534 PosixAioState *s = opaque;
535 return !!s->first_aio;
538 static PosixAioState *posix_aio_state;
540 static void aio_signal_handler(int signum)
542 if (posix_aio_state) {
543 char byte = 0;
545 write(posix_aio_state->wfd, &byte, sizeof(byte));
548 qemu_service_io();
551 static int posix_aio_init(void)
553 struct sigaction act;
554 PosixAioState *s;
555 int fds[2];
556 struct qemu_paioinit ai;
558 if (posix_aio_state)
559 return 0;
561 s = qemu_malloc(sizeof(PosixAioState));
563 sigfillset(&act.sa_mask);
564 act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
565 act.sa_handler = aio_signal_handler;
566 sigaction(SIGUSR2, &act, NULL);
568 s->first_aio = NULL;
569 if (pipe(fds) == -1) {
570 fprintf(stderr, "failed to create pipe\n");
571 return -errno;
574 s->rfd = fds[0];
575 s->wfd = fds[1];
577 fcntl(s->rfd, F_SETFL, O_NONBLOCK);
578 fcntl(s->wfd, F_SETFL, O_NONBLOCK);
580 qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);
582 memset(&ai, 0, sizeof(ai));
583 ai.aio_threads = 64;
584 ai.aio_num = 64;
585 qemu_paio_init(&ai);
587 posix_aio_state = s;
589 return 0;
592 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
593 int64_t sector_num, uint8_t *buf, int nb_sectors,
594 BlockDriverCompletionFunc *cb, void *opaque)
596 BDRVRawState *s = bs->opaque;
597 RawAIOCB *acb;
599 if (fd_open(bs) < 0)
600 return NULL;
602 acb = qemu_aio_get(bs, cb, opaque);
603 if (!acb)
604 return NULL;
605 acb->aiocb.aio_fildes = s->fd;
606 acb->aiocb.ev_signo = SIGUSR2;
607 acb->aiocb.aio_buf = buf;
608 if (nb_sectors < 0)
609 acb->aiocb.aio_nbytes = -nb_sectors;
610 else
611 acb->aiocb.aio_nbytes = nb_sectors * 512;
612 acb->aiocb.aio_offset = sector_num * 512;
613 acb->next = posix_aio_state->first_aio;
614 posix_aio_state->first_aio = acb;
615 return acb;
618 static void raw_aio_em_cb(void* opaque)
620 RawAIOCB *acb = opaque;
621 acb->common.cb(acb->common.opaque, acb->ret);
622 qemu_aio_release(acb);
625 static void raw_aio_remove(RawAIOCB *acb)
627 RawAIOCB **pacb;
629 /* remove the callback from the queue */
630 pacb = &posix_aio_state->first_aio;
631 for(;;) {
632 if (*pacb == NULL) {
633 fprintf(stderr, "raw_aio_remove: aio request not found!\n");
634 break;
635 } else if (*pacb == acb) {
636 *pacb = acb->next;
637 qemu_aio_release(acb);
638 break;
640 pacb = &(*pacb)->next;
644 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
645 int64_t sector_num, uint8_t *buf, int nb_sectors,
646 BlockDriverCompletionFunc *cb, void *opaque)
648 RawAIOCB *acb;
651 * If O_DIRECT is used and the buffer is not aligned fall back
652 * to synchronous IO.
654 BDRVRawState *s = bs->opaque;
656 if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
657 QEMUBH *bh;
658 acb = qemu_aio_get(bs, cb, opaque);
659 acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
660 bh = qemu_bh_new(raw_aio_em_cb, acb);
661 qemu_bh_schedule(bh);
662 return &acb->common;
665 acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
666 if (!acb)
667 return NULL;
668 if (qemu_paio_read(&acb->aiocb) < 0) {
669 raw_aio_remove(acb);
670 return NULL;
672 return &acb->common;
675 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
676 int64_t sector_num, const uint8_t *buf, int nb_sectors,
677 BlockDriverCompletionFunc *cb, void *opaque)
679 RawAIOCB *acb;
682 * If O_DIRECT is used and the buffer is not aligned fall back
683 * to synchronous IO.
685 BDRVRawState *s = bs->opaque;
687 if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
688 QEMUBH *bh;
689 acb = qemu_aio_get(bs, cb, opaque);
690 acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
691 bh = qemu_bh_new(raw_aio_em_cb, acb);
692 qemu_bh_schedule(bh);
693 return &acb->common;
696 acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
697 if (!acb)
698 return NULL;
699 if (qemu_paio_write(&acb->aiocb) < 0) {
700 raw_aio_remove(acb);
701 return NULL;
703 return &acb->common;
706 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
708 int ret;
709 RawAIOCB *acb = (RawAIOCB *)blockacb;
711 ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
712 if (ret == QEMU_PAIO_NOTCANCELED) {
713 /* fail safe: if the aio could not be canceled, we wait for
714 it */
715 while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
718 raw_aio_remove(acb);
720 #else /* CONFIG_AIO */
721 static int posix_aio_init(void)
723 return 0;
725 #endif /* CONFIG_AIO */
728 static void raw_close(BlockDriverState *bs)
730 BDRVRawState *s = bs->opaque;
731 if (s->fd >= 0) {
732 close(s->fd);
733 s->fd = -1;
734 if (s->aligned_buf != NULL)
735 qemu_free(s->aligned_buf);
739 static int raw_truncate(BlockDriverState *bs, int64_t offset)
741 BDRVRawState *s = bs->opaque;
742 if (s->type != FTYPE_FILE)
743 return -ENOTSUP;
744 if (ftruncate(s->fd, offset) < 0)
745 return -errno;
746 return 0;
749 #ifdef __OpenBSD__
750 static int64_t raw_getlength(BlockDriverState *bs)
752 BDRVRawState *s = bs->opaque;
753 int fd = s->fd;
754 struct stat st;
756 if (fstat(fd, &st))
757 return -1;
758 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
759 struct disklabel dl;
761 if (ioctl(fd, DIOCGDINFO, &dl))
762 return -1;
763 return (uint64_t)dl.d_secsize *
764 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
765 } else
766 return st.st_size;
768 #else /* !__OpenBSD__ */
769 static int64_t raw_getlength(BlockDriverState *bs)
771 BDRVRawState *s = bs->opaque;
772 int fd = s->fd;
773 int64_t size;
774 #ifdef HOST_BSD
775 struct stat sb;
776 #endif
777 #ifdef __sun__
778 struct dk_minfo minfo;
779 int rv;
780 #endif
781 int ret;
783 ret = fd_open(bs);
784 if (ret < 0)
785 return ret;
787 #ifdef HOST_BSD
788 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
789 #ifdef DIOCGMEDIASIZE
790 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
791 #elif defined(DIOCGPART)
793 struct partinfo pi;
794 if (ioctl(fd, DIOCGPART, &pi) == 0)
795 size = pi.media_size;
796 else
797 size = 0;
799 if (size == 0)
800 #endif
801 #ifdef CONFIG_COCOA
802 size = LONG_LONG_MAX;
803 #else
804 size = lseek(fd, 0LL, SEEK_END);
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 "raw",
851 sizeof(BDRVRawState),
852 NULL, /* no probe for protocols */
853 raw_open,
854 NULL,
855 NULL,
856 raw_close,
857 raw_create,
858 raw_flush,
860 #ifdef CONFIG_AIO
861 .bdrv_aio_read = raw_aio_read,
862 .bdrv_aio_write = raw_aio_write,
863 .bdrv_aio_cancel = raw_aio_cancel,
864 .aiocb_size = sizeof(RawAIOCB),
865 #endif
867 .bdrv_read = raw_read,
868 .bdrv_write = raw_write,
869 .bdrv_truncate = raw_truncate,
870 .bdrv_getlength = raw_getlength,
873 /***********************************************/
874 /* host device */
876 #ifdef CONFIG_COCOA
877 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
878 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
880 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
882 kern_return_t kernResult;
883 mach_port_t masterPort;
884 CFMutableDictionaryRef classesToMatch;
886 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
887 if ( KERN_SUCCESS != kernResult ) {
888 printf( "IOMasterPort returned %d\n", kernResult );
891 classesToMatch = IOServiceMatching( kIOCDMediaClass );
892 if ( classesToMatch == NULL ) {
893 printf( "IOServiceMatching returned a NULL dictionary.\n" );
894 } else {
895 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
897 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
898 if ( KERN_SUCCESS != kernResult )
900 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
903 return kernResult;
906 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
908 io_object_t nextMedia;
909 kern_return_t kernResult = KERN_FAILURE;
910 *bsdPath = '\0';
911 nextMedia = IOIteratorNext( mediaIterator );
912 if ( nextMedia )
914 CFTypeRef bsdPathAsCFString;
915 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
916 if ( bsdPathAsCFString ) {
917 size_t devPathLength;
918 strcpy( bsdPath, _PATH_DEV );
919 strcat( bsdPath, "r" );
920 devPathLength = strlen( bsdPath );
921 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
922 kernResult = KERN_SUCCESS;
924 CFRelease( bsdPathAsCFString );
926 IOObjectRelease( nextMedia );
929 return kernResult;
932 #endif
934 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
936 BDRVRawState *s = bs->opaque;
937 int fd, open_flags, ret;
939 posix_aio_init();
941 #ifdef CONFIG_COCOA
942 if (strstart(filename, "/dev/cdrom", NULL)) {
943 kern_return_t kernResult;
944 io_iterator_t mediaIterator;
945 char bsdPath[ MAXPATHLEN ];
946 int fd;
948 kernResult = FindEjectableCDMedia( &mediaIterator );
949 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
951 if ( bsdPath[ 0 ] != '\0' ) {
952 strcat(bsdPath,"s0");
953 /* some CDs don't have a partition 0 */
954 fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
955 if (fd < 0) {
956 bsdPath[strlen(bsdPath)-1] = '1';
957 } else {
958 close(fd);
960 filename = bsdPath;
963 if ( mediaIterator )
964 IOObjectRelease( mediaIterator );
966 #endif
967 open_flags = O_BINARY;
968 if ((flags & BDRV_O_ACCESS) == O_RDWR) {
969 open_flags |= O_RDWR;
970 } else {
971 open_flags |= O_RDONLY;
972 bs->read_only = 1;
974 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
975 * and O_DIRECT for no caching. */
976 if ((flags & BDRV_O_NOCACHE))
977 open_flags |= O_DIRECT;
978 else if (!(flags & BDRV_O_CACHE_WB))
979 open_flags |= O_DSYNC;
981 s->type = FTYPE_FILE;
982 #if defined(__linux__)
983 if (strstart(filename, "/dev/cd", NULL)) {
984 /* open will not fail even if no CD is inserted */
985 open_flags |= O_NONBLOCK;
986 s->type = FTYPE_CD;
987 } else if (strstart(filename, "/dev/fd", NULL)) {
988 s->type = FTYPE_FD;
989 s->fd_open_flags = open_flags;
990 /* open will not fail even if no floppy is inserted */
991 open_flags |= O_NONBLOCK;
992 } else if (strstart(filename, "/dev/sg", NULL)) {
993 bs->sg = 1;
995 #endif
996 fd = open(filename, open_flags, 0644);
997 if (fd < 0) {
998 ret = -errno;
999 if (ret == -EROFS)
1000 ret = -EACCES;
1001 return ret;
1003 s->fd = fd;
1004 #if defined(__linux__)
1005 /* close fd so that we can reopen it as needed */
1006 if (s->type == FTYPE_FD) {
1007 close(s->fd);
1008 s->fd = -1;
1009 s->fd_media_changed = 1;
1011 #endif
1012 return 0;
1015 #if defined(__linux__)
1016 /* Note: we do not have a reliable method to detect if the floppy is
1017 present. The current method is to try to open the floppy at every
1018 I/O and to keep it opened during a few hundreds of ms. */
1019 static int fd_open(BlockDriverState *bs)
1021 BDRVRawState *s = bs->opaque;
1022 int last_media_present;
1024 if (s->type != FTYPE_FD)
1025 return 0;
1026 last_media_present = (s->fd >= 0);
1027 if (s->fd >= 0 &&
1028 (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1029 close(s->fd);
1030 s->fd = -1;
1031 #ifdef DEBUG_FLOPPY
1032 printf("Floppy closed\n");
1033 #endif
1035 if (s->fd < 0) {
1036 if (s->fd_got_error &&
1037 (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1038 #ifdef DEBUG_FLOPPY
1039 printf("No floppy (open delayed)\n");
1040 #endif
1041 return -EIO;
1043 s->fd = open(bs->filename, s->fd_open_flags);
1044 if (s->fd < 0) {
1045 s->fd_error_time = qemu_get_clock(rt_clock);
1046 s->fd_got_error = 1;
1047 if (last_media_present)
1048 s->fd_media_changed = 1;
1049 #ifdef DEBUG_FLOPPY
1050 printf("No floppy\n");
1051 #endif
1052 return -EIO;
1054 #ifdef DEBUG_FLOPPY
1055 printf("Floppy opened\n");
1056 #endif
1058 if (!last_media_present)
1059 s->fd_media_changed = 1;
1060 s->fd_open_time = qemu_get_clock(rt_clock);
1061 s->fd_got_error = 0;
1062 return 0;
1065 static int raw_is_inserted(BlockDriverState *bs)
1067 BDRVRawState *s = bs->opaque;
1068 int ret;
1070 switch(s->type) {
1071 case FTYPE_CD:
1072 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1073 if (ret == CDS_DISC_OK)
1074 return 1;
1075 else
1076 return 0;
1077 break;
1078 case FTYPE_FD:
1079 ret = fd_open(bs);
1080 return (ret >= 0);
1081 default:
1082 return 1;
1086 /* currently only used by fdc.c, but a CD version would be good too */
1087 static int raw_media_changed(BlockDriverState *bs)
1089 BDRVRawState *s = bs->opaque;
1091 switch(s->type) {
1092 case FTYPE_FD:
1094 int ret;
1095 /* XXX: we do not have a true media changed indication. It
1096 does not work if the floppy is changed without trying
1097 to read it */
1098 fd_open(bs);
1099 ret = s->fd_media_changed;
1100 s->fd_media_changed = 0;
1101 #ifdef DEBUG_FLOPPY
1102 printf("Floppy changed=%d\n", ret);
1103 #endif
1104 return ret;
1106 default:
1107 return -ENOTSUP;
1111 static int raw_eject(BlockDriverState *bs, int eject_flag)
1113 BDRVRawState *s = bs->opaque;
1115 switch(s->type) {
1116 case FTYPE_CD:
1117 if (eject_flag) {
1118 if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1119 perror("CDROMEJECT");
1120 } else {
1121 if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1122 perror("CDROMEJECT");
1124 break;
1125 case FTYPE_FD:
1127 int fd;
1128 if (s->fd >= 0) {
1129 close(s->fd);
1130 s->fd = -1;
1132 fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1133 if (fd >= 0) {
1134 if (ioctl(fd, FDEJECT, 0) < 0)
1135 perror("FDEJECT");
1136 close(fd);
1139 break;
1140 default:
1141 return -ENOTSUP;
1143 return 0;
1146 static int raw_set_locked(BlockDriverState *bs, int locked)
1148 BDRVRawState *s = bs->opaque;
1150 switch(s->type) {
1151 case FTYPE_CD:
1152 if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1153 /* Note: an error can happen if the distribution automatically
1154 mounts the CD-ROM */
1155 // perror("CDROM_LOCKDOOR");
1157 break;
1158 default:
1159 return -ENOTSUP;
1161 return 0;
1164 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1166 BDRVRawState *s = bs->opaque;
1168 return ioctl(s->fd, req, buf);
1170 #else
1172 static int fd_open(BlockDriverState *bs)
1174 return 0;
1177 static int raw_is_inserted(BlockDriverState *bs)
1179 return 1;
1182 static int raw_media_changed(BlockDriverState *bs)
1184 return -ENOTSUP;
1187 static int raw_eject(BlockDriverState *bs, int eject_flag)
1189 return -ENOTSUP;
1192 static int raw_set_locked(BlockDriverState *bs, int locked)
1194 return -ENOTSUP;
1197 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1199 return -ENOTSUP;
1201 #endif /* !linux */
1203 static int raw_sg_send_command(BlockDriverState *bs, void *buf, int count)
1205 return raw_pwrite(bs, -1, buf, count);
1208 static int raw_sg_recv_response(BlockDriverState *bs, void *buf, int count)
1210 return raw_pread(bs, -1, buf, count);
1213 static BlockDriverAIOCB *raw_sg_aio_read(BlockDriverState *bs,
1214 void *buf, int count,
1215 BlockDriverCompletionFunc *cb,
1216 void *opaque)
1218 return raw_aio_read(bs, 0, buf, -(int64_t)count, cb, opaque);
1221 static BlockDriverAIOCB *raw_sg_aio_write(BlockDriverState *bs,
1222 void *buf, int count,
1223 BlockDriverCompletionFunc *cb,
1224 void *opaque)
1226 return raw_aio_write(bs, 0, buf, -(int64_t)count, cb, opaque);
1229 BlockDriver bdrv_host_device = {
1230 .format_name = "host_device",
1231 .instance_size = sizeof(BDRVRawState),
1232 .bdrv_open = hdev_open,
1233 .bdrv_close = raw_close,
1234 .bdrv_flush = raw_flush,
1236 #ifdef CONFIG_AIO
1237 .bdrv_aio_read = raw_aio_read,
1238 .bdrv_aio_write = raw_aio_write,
1239 .bdrv_aio_cancel = raw_aio_cancel,
1240 .aiocb_size = sizeof(RawAIOCB),
1241 #endif
1243 .bdrv_read = raw_read,
1244 .bdrv_write = raw_write,
1245 .bdrv_getlength = raw_getlength,
1247 /* removable device support */
1248 .bdrv_is_inserted = raw_is_inserted,
1249 .bdrv_media_changed = raw_media_changed,
1250 .bdrv_eject = raw_eject,
1251 .bdrv_set_locked = raw_set_locked,
1252 /* generic scsi device */
1253 .bdrv_ioctl = raw_ioctl,
1254 .bdrv_sg_send_command = raw_sg_send_command,
1255 .bdrv_sg_recv_response = raw_sg_recv_response,
1256 .bdrv_sg_aio_read = raw_sg_aio_read,
1257 .bdrv_sg_aio_write = raw_sg_aio_write,