Guest debugging support for KVM (Jan Kiszka)
[sniper_test.git] / block-raw-posix.c
blob696128b80b66177908b19bb1c00cca1a85823ba8
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 return raw_pread(bs, sector_num * 512, buf, (uint64_t)nb_sectors * 512);
368 * offset and count are in bytes and possibly not aligned. For files opened
369 * with O_DIRECT, necessary alignments are ensured before calling
370 * raw_pwrite_aligned to do the actual write.
372 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
373 const uint8_t *buf, int count)
375 BDRVRawState *s = bs->opaque;
376 int size, ret, shift, sum;
378 sum = 0;
380 if (s->aligned_buf != NULL) {
382 if (offset & 0x1ff) {
383 /* align offset on a 512 bytes boundary */
384 shift = offset & 0x1ff;
385 ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
386 if (ret < 0)
387 return ret;
389 size = 512 - shift;
390 if (size > count)
391 size = count;
392 memcpy(s->aligned_buf + shift, buf, size);
394 ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
395 if (ret < 0)
396 return ret;
398 buf += size;
399 offset += size;
400 count -= size;
401 sum += size;
403 if (count == 0)
404 return sum;
406 if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
408 while ((size = (count & ~0x1ff)) != 0) {
410 if (size > ALIGNED_BUFFER_SIZE)
411 size = ALIGNED_BUFFER_SIZE;
413 memcpy(s->aligned_buf, buf, size);
415 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
416 if (ret < 0)
417 return ret;
419 buf += ret;
420 offset += ret;
421 count -= ret;
422 sum += ret;
424 /* here, count < 512 because (count & ~0x1ff) == 0 */
425 if (count) {
426 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
427 if (ret < 0)
428 return ret;
429 memcpy(s->aligned_buf, buf, count);
431 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
432 if (ret < 0)
433 return ret;
434 if (count < ret)
435 ret = count;
437 sum += ret;
439 return sum;
442 return raw_pwrite_aligned(bs, offset, buf, count) + sum;
445 static int raw_write(BlockDriverState *bs, int64_t sector_num,
446 const uint8_t *buf, int nb_sectors)
448 return raw_pwrite(bs, sector_num * 512, buf, (uint64_t)nb_sectors * 512);
451 #ifdef CONFIG_AIO
452 /***********************************************************/
453 /* Unix AIO using POSIX AIO */
455 typedef struct RawAIOCB {
456 BlockDriverAIOCB common;
457 struct qemu_paiocb aiocb;
458 struct RawAIOCB *next;
459 int ret;
460 } RawAIOCB;
462 typedef struct PosixAioState
464 int rfd, wfd;
465 RawAIOCB *first_aio;
466 } PosixAioState;
468 static void posix_aio_read(void *opaque)
470 PosixAioState *s = opaque;
471 RawAIOCB *acb, **pacb;
472 int ret;
473 ssize_t len;
475 /* read all bytes from signal pipe */
476 for (;;) {
477 char bytes[16];
479 len = read(s->rfd, bytes, sizeof(bytes));
480 if (len == -1 && errno == EINTR)
481 continue; /* try again */
482 if (len == sizeof(bytes))
483 continue; /* more to read */
484 break;
487 for(;;) {
488 pacb = &s->first_aio;
489 for(;;) {
490 acb = *pacb;
491 if (!acb)
492 goto the_end;
493 ret = qemu_paio_error(&acb->aiocb);
494 if (ret == ECANCELED) {
495 /* remove the request */
496 *pacb = acb->next;
497 qemu_aio_release(acb);
498 } else if (ret != EINPROGRESS) {
499 /* end of aio */
500 if (ret == 0) {
501 ret = qemu_paio_return(&acb->aiocb);
502 if (ret == acb->aiocb.aio_nbytes)
503 ret = 0;
504 else
505 ret = -EINVAL;
506 } else {
507 ret = -ret;
509 /* remove the request */
510 *pacb = acb->next;
511 /* call the callback */
512 acb->common.cb(acb->common.opaque, ret);
513 qemu_aio_release(acb);
514 break;
515 } else {
516 pacb = &acb->next;
520 the_end: ;
523 static int posix_aio_flush(void *opaque)
525 PosixAioState *s = opaque;
526 return !!s->first_aio;
529 static PosixAioState *posix_aio_state;
531 static void aio_signal_handler(int signum)
533 if (posix_aio_state) {
534 char byte = 0;
536 write(posix_aio_state->wfd, &byte, sizeof(byte));
539 qemu_service_io();
542 static int posix_aio_init(void)
544 struct sigaction act;
545 PosixAioState *s;
546 int fds[2];
547 struct qemu_paioinit ai;
549 if (posix_aio_state)
550 return 0;
552 s = qemu_malloc(sizeof(PosixAioState));
554 sigfillset(&act.sa_mask);
555 act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
556 act.sa_handler = aio_signal_handler;
557 sigaction(SIGUSR2, &act, NULL);
559 s->first_aio = NULL;
560 if (pipe(fds) == -1) {
561 fprintf(stderr, "failed to create pipe\n");
562 return -errno;
565 s->rfd = fds[0];
566 s->wfd = fds[1];
568 fcntl(s->rfd, F_SETFL, O_NONBLOCK);
569 fcntl(s->wfd, F_SETFL, O_NONBLOCK);
571 qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);
573 memset(&ai, 0, sizeof(ai));
574 ai.aio_threads = 64;
575 ai.aio_num = 64;
576 qemu_paio_init(&ai);
578 posix_aio_state = s;
580 return 0;
583 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
584 int64_t sector_num, uint8_t *buf, int nb_sectors,
585 BlockDriverCompletionFunc *cb, void *opaque)
587 BDRVRawState *s = bs->opaque;
588 RawAIOCB *acb;
590 if (fd_open(bs) < 0)
591 return NULL;
593 acb = qemu_aio_get(bs, cb, opaque);
594 if (!acb)
595 return NULL;
596 acb->aiocb.aio_fildes = s->fd;
597 acb->aiocb.ev_signo = SIGUSR2;
598 acb->aiocb.aio_buf = buf;
599 if (nb_sectors < 0)
600 acb->aiocb.aio_nbytes = -nb_sectors;
601 else
602 acb->aiocb.aio_nbytes = nb_sectors * 512;
603 acb->aiocb.aio_offset = sector_num * 512;
604 acb->next = posix_aio_state->first_aio;
605 posix_aio_state->first_aio = acb;
606 return acb;
609 static void raw_aio_em_cb(void* opaque)
611 RawAIOCB *acb = opaque;
612 acb->common.cb(acb->common.opaque, acb->ret);
613 qemu_aio_release(acb);
616 static void raw_aio_remove(RawAIOCB *acb)
618 RawAIOCB **pacb;
620 /* remove the callback from the queue */
621 pacb = &posix_aio_state->first_aio;
622 for(;;) {
623 if (*pacb == NULL) {
624 fprintf(stderr, "raw_aio_remove: aio request not found!\n");
625 break;
626 } else if (*pacb == acb) {
627 *pacb = acb->next;
628 qemu_aio_release(acb);
629 break;
631 pacb = &(*pacb)->next;
635 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
636 int64_t sector_num, uint8_t *buf, int nb_sectors,
637 BlockDriverCompletionFunc *cb, void *opaque)
639 RawAIOCB *acb;
642 * If O_DIRECT is used and the buffer is not aligned fall back
643 * to synchronous IO.
645 BDRVRawState *s = bs->opaque;
647 if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
648 QEMUBH *bh;
649 acb = qemu_aio_get(bs, cb, opaque);
650 acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
651 bh = qemu_bh_new(raw_aio_em_cb, acb);
652 qemu_bh_schedule(bh);
653 return &acb->common;
656 acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
657 if (!acb)
658 return NULL;
659 if (qemu_paio_read(&acb->aiocb) < 0) {
660 raw_aio_remove(acb);
661 return NULL;
663 return &acb->common;
666 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
667 int64_t sector_num, const uint8_t *buf, int nb_sectors,
668 BlockDriverCompletionFunc *cb, void *opaque)
670 RawAIOCB *acb;
673 * If O_DIRECT is used and the buffer is not aligned fall back
674 * to synchronous IO.
676 BDRVRawState *s = bs->opaque;
678 if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
679 QEMUBH *bh;
680 acb = qemu_aio_get(bs, cb, opaque);
681 acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
682 bh = qemu_bh_new(raw_aio_em_cb, acb);
683 qemu_bh_schedule(bh);
684 return &acb->common;
687 acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
688 if (!acb)
689 return NULL;
690 if (qemu_paio_write(&acb->aiocb) < 0) {
691 raw_aio_remove(acb);
692 return NULL;
694 return &acb->common;
697 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
699 int ret;
700 RawAIOCB *acb = (RawAIOCB *)blockacb;
702 ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
703 if (ret == QEMU_PAIO_NOTCANCELED) {
704 /* fail safe: if the aio could not be canceled, we wait for
705 it */
706 while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
709 raw_aio_remove(acb);
711 #else /* CONFIG_AIO */
712 static int posix_aio_init(void)
714 return 0;
716 #endif /* CONFIG_AIO */
719 static void raw_close(BlockDriverState *bs)
721 BDRVRawState *s = bs->opaque;
722 if (s->fd >= 0) {
723 close(s->fd);
724 s->fd = -1;
725 if (s->aligned_buf != NULL)
726 qemu_free(s->aligned_buf);
730 static int raw_truncate(BlockDriverState *bs, int64_t offset)
732 BDRVRawState *s = bs->opaque;
733 if (s->type != FTYPE_FILE)
734 return -ENOTSUP;
735 if (ftruncate(s->fd, offset) < 0)
736 return -errno;
737 return 0;
740 #ifdef __OpenBSD__
741 static int64_t raw_getlength(BlockDriverState *bs)
743 BDRVRawState *s = bs->opaque;
744 int fd = s->fd;
745 struct stat st;
747 if (fstat(fd, &st))
748 return -1;
749 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
750 struct disklabel dl;
752 if (ioctl(fd, DIOCGDINFO, &dl))
753 return -1;
754 return (uint64_t)dl.d_secsize *
755 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
756 } else
757 return st.st_size;
759 #else /* !__OpenBSD__ */
760 static int64_t raw_getlength(BlockDriverState *bs)
762 BDRVRawState *s = bs->opaque;
763 int fd = s->fd;
764 int64_t size;
765 #ifdef HOST_BSD
766 struct stat sb;
767 #endif
768 #ifdef __sun__
769 struct dk_minfo minfo;
770 int rv;
771 #endif
772 int ret;
774 ret = fd_open(bs);
775 if (ret < 0)
776 return ret;
778 #ifdef HOST_BSD
779 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
780 #ifdef DIOCGMEDIASIZE
781 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
782 #elif defined(DIOCGPART)
784 struct partinfo pi;
785 if (ioctl(fd, DIOCGPART, &pi) == 0)
786 size = pi.media_size;
787 else
788 size = 0;
790 if (size == 0)
791 #endif
792 #ifdef CONFIG_COCOA
793 size = LONG_LONG_MAX;
794 #else
795 size = lseek(fd, 0LL, SEEK_END);
796 #endif
797 } else
798 #endif
799 #ifdef __sun__
801 * use the DKIOCGMEDIAINFO ioctl to read the size.
803 rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
804 if ( rv != -1 ) {
805 size = minfo.dki_lbsize * minfo.dki_capacity;
806 } else /* there are reports that lseek on some devices
807 fails, but irc discussion said that contingency
808 on contingency was overkill */
809 #endif
811 size = lseek(fd, 0, SEEK_END);
813 return size;
815 #endif
817 static int raw_create(const char *filename, int64_t total_size,
818 const char *backing_file, int flags)
820 int fd;
822 if (flags || backing_file)
823 return -ENOTSUP;
825 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
826 0644);
827 if (fd < 0)
828 return -EIO;
829 ftruncate(fd, total_size * 512);
830 close(fd);
831 return 0;
834 static void raw_flush(BlockDriverState *bs)
836 BDRVRawState *s = bs->opaque;
837 fsync(s->fd);
840 BlockDriver bdrv_raw = {
841 "raw",
842 sizeof(BDRVRawState),
843 NULL, /* no probe for protocols */
844 raw_open,
845 NULL,
846 NULL,
847 raw_close,
848 raw_create,
849 raw_flush,
851 #ifdef CONFIG_AIO
852 .bdrv_aio_read = raw_aio_read,
853 .bdrv_aio_write = raw_aio_write,
854 .bdrv_aio_cancel = raw_aio_cancel,
855 .aiocb_size = sizeof(RawAIOCB),
856 #endif
858 .bdrv_read = raw_read,
859 .bdrv_write = raw_write,
860 .bdrv_truncate = raw_truncate,
861 .bdrv_getlength = raw_getlength,
864 /***********************************************/
865 /* host device */
867 #ifdef CONFIG_COCOA
868 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
869 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
871 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
873 kern_return_t kernResult;
874 mach_port_t masterPort;
875 CFMutableDictionaryRef classesToMatch;
877 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
878 if ( KERN_SUCCESS != kernResult ) {
879 printf( "IOMasterPort returned %d\n", kernResult );
882 classesToMatch = IOServiceMatching( kIOCDMediaClass );
883 if ( classesToMatch == NULL ) {
884 printf( "IOServiceMatching returned a NULL dictionary.\n" );
885 } else {
886 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
888 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
889 if ( KERN_SUCCESS != kernResult )
891 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
894 return kernResult;
897 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
899 io_object_t nextMedia;
900 kern_return_t kernResult = KERN_FAILURE;
901 *bsdPath = '\0';
902 nextMedia = IOIteratorNext( mediaIterator );
903 if ( nextMedia )
905 CFTypeRef bsdPathAsCFString;
906 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
907 if ( bsdPathAsCFString ) {
908 size_t devPathLength;
909 strcpy( bsdPath, _PATH_DEV );
910 strcat( bsdPath, "r" );
911 devPathLength = strlen( bsdPath );
912 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
913 kernResult = KERN_SUCCESS;
915 CFRelease( bsdPathAsCFString );
917 IOObjectRelease( nextMedia );
920 return kernResult;
923 #endif
925 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
927 BDRVRawState *s = bs->opaque;
928 int fd, open_flags, ret;
930 posix_aio_init();
932 #ifdef CONFIG_COCOA
933 if (strstart(filename, "/dev/cdrom", NULL)) {
934 kern_return_t kernResult;
935 io_iterator_t mediaIterator;
936 char bsdPath[ MAXPATHLEN ];
937 int fd;
939 kernResult = FindEjectableCDMedia( &mediaIterator );
940 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
942 if ( bsdPath[ 0 ] != '\0' ) {
943 strcat(bsdPath,"s0");
944 /* some CDs don't have a partition 0 */
945 fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
946 if (fd < 0) {
947 bsdPath[strlen(bsdPath)-1] = '1';
948 } else {
949 close(fd);
951 filename = bsdPath;
954 if ( mediaIterator )
955 IOObjectRelease( mediaIterator );
957 #endif
958 open_flags = O_BINARY;
959 if ((flags & BDRV_O_ACCESS) == O_RDWR) {
960 open_flags |= O_RDWR;
961 } else {
962 open_flags |= O_RDONLY;
963 bs->read_only = 1;
965 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
966 * and O_DIRECT for no caching. */
967 if ((flags & BDRV_O_NOCACHE))
968 open_flags |= O_DIRECT;
969 else if (!(flags & BDRV_O_CACHE_WB))
970 open_flags |= O_DSYNC;
972 s->type = FTYPE_FILE;
973 #if defined(__linux__)
974 if (strstart(filename, "/dev/cd", NULL)) {
975 /* open will not fail even if no CD is inserted */
976 open_flags |= O_NONBLOCK;
977 s->type = FTYPE_CD;
978 } else if (strstart(filename, "/dev/fd", NULL)) {
979 s->type = FTYPE_FD;
980 s->fd_open_flags = open_flags;
981 /* open will not fail even if no floppy is inserted */
982 open_flags |= O_NONBLOCK;
983 } else if (strstart(filename, "/dev/sg", NULL)) {
984 bs->sg = 1;
986 #endif
987 fd = open(filename, open_flags, 0644);
988 if (fd < 0) {
989 ret = -errno;
990 if (ret == -EROFS)
991 ret = -EACCES;
992 return ret;
994 s->fd = fd;
995 #if defined(__linux__)
996 /* close fd so that we can reopen it as needed */
997 if (s->type == FTYPE_FD) {
998 close(s->fd);
999 s->fd = -1;
1000 s->fd_media_changed = 1;
1002 #endif
1003 return 0;
1006 #if defined(__linux__)
1007 /* Note: we do not have a reliable method to detect if the floppy is
1008 present. The current method is to try to open the floppy at every
1009 I/O and to keep it opened during a few hundreds of ms. */
1010 static int fd_open(BlockDriverState *bs)
1012 BDRVRawState *s = bs->opaque;
1013 int last_media_present;
1015 if (s->type != FTYPE_FD)
1016 return 0;
1017 last_media_present = (s->fd >= 0);
1018 if (s->fd >= 0 &&
1019 (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1020 close(s->fd);
1021 s->fd = -1;
1022 #ifdef DEBUG_FLOPPY
1023 printf("Floppy closed\n");
1024 #endif
1026 if (s->fd < 0) {
1027 if (s->fd_got_error &&
1028 (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1029 #ifdef DEBUG_FLOPPY
1030 printf("No floppy (open delayed)\n");
1031 #endif
1032 return -EIO;
1034 s->fd = open(bs->filename, s->fd_open_flags);
1035 if (s->fd < 0) {
1036 s->fd_error_time = qemu_get_clock(rt_clock);
1037 s->fd_got_error = 1;
1038 if (last_media_present)
1039 s->fd_media_changed = 1;
1040 #ifdef DEBUG_FLOPPY
1041 printf("No floppy\n");
1042 #endif
1043 return -EIO;
1045 #ifdef DEBUG_FLOPPY
1046 printf("Floppy opened\n");
1047 #endif
1049 if (!last_media_present)
1050 s->fd_media_changed = 1;
1051 s->fd_open_time = qemu_get_clock(rt_clock);
1052 s->fd_got_error = 0;
1053 return 0;
1056 static int raw_is_inserted(BlockDriverState *bs)
1058 BDRVRawState *s = bs->opaque;
1059 int ret;
1061 switch(s->type) {
1062 case FTYPE_CD:
1063 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1064 if (ret == CDS_DISC_OK)
1065 return 1;
1066 else
1067 return 0;
1068 break;
1069 case FTYPE_FD:
1070 ret = fd_open(bs);
1071 return (ret >= 0);
1072 default:
1073 return 1;
1077 /* currently only used by fdc.c, but a CD version would be good too */
1078 static int raw_media_changed(BlockDriverState *bs)
1080 BDRVRawState *s = bs->opaque;
1082 switch(s->type) {
1083 case FTYPE_FD:
1085 int ret;
1086 /* XXX: we do not have a true media changed indication. It
1087 does not work if the floppy is changed without trying
1088 to read it */
1089 fd_open(bs);
1090 ret = s->fd_media_changed;
1091 s->fd_media_changed = 0;
1092 #ifdef DEBUG_FLOPPY
1093 printf("Floppy changed=%d\n", ret);
1094 #endif
1095 return ret;
1097 default:
1098 return -ENOTSUP;
1102 static int raw_eject(BlockDriverState *bs, int eject_flag)
1104 BDRVRawState *s = bs->opaque;
1106 switch(s->type) {
1107 case FTYPE_CD:
1108 if (eject_flag) {
1109 if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1110 perror("CDROMEJECT");
1111 } else {
1112 if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1113 perror("CDROMEJECT");
1115 break;
1116 case FTYPE_FD:
1118 int fd;
1119 if (s->fd >= 0) {
1120 close(s->fd);
1121 s->fd = -1;
1123 fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1124 if (fd >= 0) {
1125 if (ioctl(fd, FDEJECT, 0) < 0)
1126 perror("FDEJECT");
1127 close(fd);
1130 break;
1131 default:
1132 return -ENOTSUP;
1134 return 0;
1137 static int raw_set_locked(BlockDriverState *bs, int locked)
1139 BDRVRawState *s = bs->opaque;
1141 switch(s->type) {
1142 case FTYPE_CD:
1143 if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1144 /* Note: an error can happen if the distribution automatically
1145 mounts the CD-ROM */
1146 // perror("CDROM_LOCKDOOR");
1148 break;
1149 default:
1150 return -ENOTSUP;
1152 return 0;
1155 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1157 BDRVRawState *s = bs->opaque;
1159 return ioctl(s->fd, req, buf);
1161 #else
1163 static int fd_open(BlockDriverState *bs)
1165 return 0;
1168 static int raw_is_inserted(BlockDriverState *bs)
1170 return 1;
1173 static int raw_media_changed(BlockDriverState *bs)
1175 return -ENOTSUP;
1178 static int raw_eject(BlockDriverState *bs, int eject_flag)
1180 return -ENOTSUP;
1183 static int raw_set_locked(BlockDriverState *bs, int locked)
1185 return -ENOTSUP;
1188 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1190 return -ENOTSUP;
1192 #endif /* !linux */
1194 static int raw_sg_send_command(BlockDriverState *bs, void *buf, int count)
1196 return raw_pwrite(bs, -1, buf, count);
1199 static int raw_sg_recv_response(BlockDriverState *bs, void *buf, int count)
1201 return raw_pread(bs, -1, buf, count);
1204 static BlockDriverAIOCB *raw_sg_aio_read(BlockDriverState *bs,
1205 void *buf, int count,
1206 BlockDriverCompletionFunc *cb,
1207 void *opaque)
1209 return raw_aio_read(bs, 0, buf, -(int64_t)count, cb, opaque);
1212 static BlockDriverAIOCB *raw_sg_aio_write(BlockDriverState *bs,
1213 void *buf, int count,
1214 BlockDriverCompletionFunc *cb,
1215 void *opaque)
1217 return raw_aio_write(bs, 0, buf, -(int64_t)count, cb, opaque);
1220 BlockDriver bdrv_host_device = {
1221 .format_name = "host_device",
1222 .instance_size = sizeof(BDRVRawState),
1223 .bdrv_open = hdev_open,
1224 .bdrv_close = raw_close,
1225 .bdrv_flush = raw_flush,
1227 #ifdef CONFIG_AIO
1228 .bdrv_aio_read = raw_aio_read,
1229 .bdrv_aio_write = raw_aio_write,
1230 .bdrv_aio_cancel = raw_aio_cancel,
1231 .aiocb_size = sizeof(RawAIOCB),
1232 #endif
1234 .bdrv_read = raw_read,
1235 .bdrv_write = raw_write,
1236 .bdrv_getlength = raw_getlength,
1238 /* removable device support */
1239 .bdrv_is_inserted = raw_is_inserted,
1240 .bdrv_media_changed = raw_media_changed,
1241 .bdrv_eject = raw_eject,
1242 .bdrv_set_locked = raw_set_locked,
1243 /* generic scsi device */
1244 .bdrv_ioctl = raw_ioctl,
1245 .bdrv_sg_send_command = raw_sg_send_command,
1246 .bdrv_sg_recv_response = raw_sg_recv_response,
1247 .bdrv_sg_aio_read = raw_sg_aio_read,
1248 .bdrv_sg_aio_write = raw_sg_aio_write,