fully split aio_pool from BlockDriver
[qemu.git] / block / raw-posix.c
blob38c4aa71af4d63599cbf26e1b63a3fd8029a8b08
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 "module.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, ...) do { if (qemu_log_enabled()) \
77 { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
78 #else
79 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
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 < 0) ? -errno : 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 void raw_aio_remove(RawAIOCB *acb)
604 RawAIOCB **pacb;
606 /* remove the callback from the queue */
607 pacb = &posix_aio_state->first_aio;
608 for(;;) {
609 if (*pacb == NULL) {
610 fprintf(stderr, "raw_aio_remove: aio request not found!\n");
611 break;
612 } else if (*pacb == acb) {
613 *pacb = acb->next;
614 qemu_aio_release(acb);
615 break;
617 pacb = &(*pacb)->next;
621 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
623 int ret;
624 RawAIOCB *acb = (RawAIOCB *)blockacb;
626 ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
627 if (ret == QEMU_PAIO_NOTCANCELED) {
628 /* fail safe: if the aio could not be canceled, we wait for
629 it */
630 while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
633 raw_aio_remove(acb);
636 static AIOPool raw_aio_pool = {
637 .aiocb_size = sizeof(RawAIOCB),
638 .cancel = raw_aio_cancel,
641 static RawAIOCB *raw_aio_setup(BlockDriverState *bs, int64_t sector_num,
642 QEMUIOVector *qiov, int nb_sectors,
643 BlockDriverCompletionFunc *cb, void *opaque)
645 BDRVRawState *s = bs->opaque;
646 RawAIOCB *acb;
648 if (fd_open(bs) < 0)
649 return NULL;
651 acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque);
652 if (!acb)
653 return NULL;
654 acb->aiocb.aio_fildes = s->fd;
655 acb->aiocb.ev_signo = SIGUSR2;
656 acb->aiocb.aio_iov = qiov->iov;
657 acb->aiocb.aio_niov = qiov->niov;
658 acb->aiocb.aio_nbytes = nb_sectors * 512;
659 acb->aiocb.aio_offset = sector_num * 512;
660 acb->aiocb.aio_flags = 0;
663 * If O_DIRECT is used the buffer needs to be aligned on a sector
664 * boundary. Tell the low level code to ensure that in case it's
665 * not done yet.
667 if (s->aligned_buf)
668 acb->aiocb.aio_flags |= QEMU_AIO_SECTOR_ALIGNED;
670 acb->next = posix_aio_state->first_aio;
671 posix_aio_state->first_aio = acb;
672 return acb;
675 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
676 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
677 BlockDriverCompletionFunc *cb, void *opaque)
679 RawAIOCB *acb;
681 acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque);
682 if (!acb)
683 return NULL;
684 if (qemu_paio_read(&acb->aiocb) < 0) {
685 raw_aio_remove(acb);
686 return NULL;
688 return &acb->common;
691 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
692 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
693 BlockDriverCompletionFunc *cb, void *opaque)
695 RawAIOCB *acb;
697 acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque);
698 if (!acb)
699 return NULL;
700 if (qemu_paio_write(&acb->aiocb) < 0) {
701 raw_aio_remove(acb);
702 return NULL;
704 return &acb->common;
706 #else /* CONFIG_AIO */
707 static int posix_aio_init(void)
709 return 0;
711 #endif /* CONFIG_AIO */
714 static void raw_close(BlockDriverState *bs)
716 BDRVRawState *s = bs->opaque;
717 if (s->fd >= 0) {
718 close(s->fd);
719 s->fd = -1;
720 if (s->aligned_buf != NULL)
721 qemu_free(s->aligned_buf);
725 static int raw_truncate(BlockDriverState *bs, int64_t offset)
727 BDRVRawState *s = bs->opaque;
728 if (s->type != FTYPE_FILE)
729 return -ENOTSUP;
730 if (ftruncate(s->fd, offset) < 0)
731 return -errno;
732 return 0;
735 #ifdef __OpenBSD__
736 static int64_t raw_getlength(BlockDriverState *bs)
738 BDRVRawState *s = bs->opaque;
739 int fd = s->fd;
740 struct stat st;
742 if (fstat(fd, &st))
743 return -1;
744 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
745 struct disklabel dl;
747 if (ioctl(fd, DIOCGDINFO, &dl))
748 return -1;
749 return (uint64_t)dl.d_secsize *
750 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
751 } else
752 return st.st_size;
754 #else /* !__OpenBSD__ */
755 static int64_t raw_getlength(BlockDriverState *bs)
757 BDRVRawState *s = bs->opaque;
758 int fd = s->fd;
759 int64_t size;
760 #ifdef HOST_BSD
761 struct stat sb;
762 #ifdef __FreeBSD__
763 int reopened = 0;
764 #endif
765 #endif
766 #ifdef __sun__
767 struct dk_minfo minfo;
768 int rv;
769 #endif
770 int ret;
772 ret = fd_open(bs);
773 if (ret < 0)
774 return ret;
776 #ifdef HOST_BSD
777 #ifdef __FreeBSD__
778 again:
779 #endif
780 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
781 #ifdef DIOCGMEDIASIZE
782 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
783 #elif defined(DIOCGPART)
785 struct partinfo pi;
786 if (ioctl(fd, DIOCGPART, &pi) == 0)
787 size = pi.media_size;
788 else
789 size = 0;
791 if (size == 0)
792 #endif
793 #ifdef CONFIG_COCOA
794 size = LONG_LONG_MAX;
795 #else
796 size = lseek(fd, 0LL, SEEK_END);
797 #endif
798 #ifdef __FreeBSD__
799 switch(s->type) {
800 case FTYPE_CD:
801 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
802 if (size == 2048LL * (unsigned)-1)
803 size = 0;
804 /* XXX no disc? maybe we need to reopen... */
805 if (size <= 0 && !reopened && cd_open(bs) >= 0) {
806 reopened = 1;
807 goto again;
810 #endif
811 } else
812 #endif
813 #ifdef __sun__
815 * use the DKIOCGMEDIAINFO ioctl to read the size.
817 rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
818 if ( rv != -1 ) {
819 size = minfo.dki_lbsize * minfo.dki_capacity;
820 } else /* there are reports that lseek on some devices
821 fails, but irc discussion said that contingency
822 on contingency was overkill */
823 #endif
825 size = lseek(fd, 0, SEEK_END);
827 return size;
829 #endif
831 static int raw_create(const char *filename, QEMUOptionParameter *options)
833 int fd;
834 int64_t total_size = 0;
836 /* Read out options */
837 while (options && options->name) {
838 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
839 total_size = options->value.n / 512;
841 options++;
844 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
845 0644);
846 if (fd < 0)
847 return -EIO;
848 ftruncate(fd, total_size * 512);
849 close(fd);
850 return 0;
853 static void raw_flush(BlockDriverState *bs)
855 BDRVRawState *s = bs->opaque;
856 fsync(s->fd);
860 static QEMUOptionParameter raw_create_options[] = {
861 { BLOCK_OPT_SIZE, OPT_SIZE },
862 { NULL }
865 static BlockDriver bdrv_raw = {
866 .format_name = "raw",
867 .instance_size = sizeof(BDRVRawState),
868 .bdrv_probe = NULL, /* no probe for protocols */
869 .bdrv_open = raw_open,
870 .bdrv_read = raw_read,
871 .bdrv_write = raw_write,
872 .bdrv_close = raw_close,
873 .bdrv_create = raw_create,
874 .bdrv_flush = raw_flush,
876 #ifdef CONFIG_AIO
877 .bdrv_aio_readv = raw_aio_readv,
878 .bdrv_aio_writev = raw_aio_writev,
879 #endif
881 .bdrv_truncate = raw_truncate,
882 .bdrv_getlength = raw_getlength,
884 .create_options = raw_create_options,
887 /***********************************************/
888 /* host device */
890 #ifdef CONFIG_COCOA
891 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
892 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
894 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
896 kern_return_t kernResult;
897 mach_port_t masterPort;
898 CFMutableDictionaryRef classesToMatch;
900 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
901 if ( KERN_SUCCESS != kernResult ) {
902 printf( "IOMasterPort returned %d\n", kernResult );
905 classesToMatch = IOServiceMatching( kIOCDMediaClass );
906 if ( classesToMatch == NULL ) {
907 printf( "IOServiceMatching returned a NULL dictionary.\n" );
908 } else {
909 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
911 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
912 if ( KERN_SUCCESS != kernResult )
914 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
917 return kernResult;
920 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
922 io_object_t nextMedia;
923 kern_return_t kernResult = KERN_FAILURE;
924 *bsdPath = '\0';
925 nextMedia = IOIteratorNext( mediaIterator );
926 if ( nextMedia )
928 CFTypeRef bsdPathAsCFString;
929 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
930 if ( bsdPathAsCFString ) {
931 size_t devPathLength;
932 strcpy( bsdPath, _PATH_DEV );
933 strcat( bsdPath, "r" );
934 devPathLength = strlen( bsdPath );
935 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
936 kernResult = KERN_SUCCESS;
938 CFRelease( bsdPathAsCFString );
940 IOObjectRelease( nextMedia );
943 return kernResult;
946 #endif
948 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
950 BDRVRawState *s = bs->opaque;
951 int fd, open_flags, ret;
953 posix_aio_init();
955 #ifdef CONFIG_COCOA
956 if (strstart(filename, "/dev/cdrom", NULL)) {
957 kern_return_t kernResult;
958 io_iterator_t mediaIterator;
959 char bsdPath[ MAXPATHLEN ];
960 int fd;
962 kernResult = FindEjectableCDMedia( &mediaIterator );
963 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
965 if ( bsdPath[ 0 ] != '\0' ) {
966 strcat(bsdPath,"s0");
967 /* some CDs don't have a partition 0 */
968 fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
969 if (fd < 0) {
970 bsdPath[strlen(bsdPath)-1] = '1';
971 } else {
972 close(fd);
974 filename = bsdPath;
977 if ( mediaIterator )
978 IOObjectRelease( mediaIterator );
980 #endif
981 open_flags = O_BINARY;
982 if ((flags & BDRV_O_ACCESS) == O_RDWR) {
983 open_flags |= O_RDWR;
984 } else {
985 open_flags |= O_RDONLY;
986 bs->read_only = 1;
988 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
989 * and O_DIRECT for no caching. */
990 if ((flags & BDRV_O_NOCACHE))
991 open_flags |= O_DIRECT;
992 else if (!(flags & BDRV_O_CACHE_WB))
993 open_flags |= O_DSYNC;
995 s->type = FTYPE_FILE;
996 #if defined(__linux__)
997 if (strstart(filename, "/dev/cd", NULL)) {
998 /* open will not fail even if no CD is inserted */
999 open_flags |= O_NONBLOCK;
1000 s->type = FTYPE_CD;
1001 } else if (strstart(filename, "/dev/fd", NULL)) {
1002 s->type = FTYPE_FD;
1003 s->fd_open_flags = open_flags;
1004 /* open will not fail even if no floppy is inserted */
1005 open_flags |= O_NONBLOCK;
1006 #ifdef CONFIG_AIO
1007 } else if (strstart(filename, "/dev/sg", NULL)) {
1008 bs->sg = 1;
1009 #endif
1011 #endif
1012 #if defined(__FreeBSD__)
1013 if (strstart(filename, "/dev/cd", NULL) ||
1014 strstart(filename, "/dev/acd", NULL)) {
1015 s->type = FTYPE_CD;
1016 s->cd_open_flags = open_flags;
1018 #endif
1019 s->fd = -1;
1020 fd = open(filename, open_flags, 0644);
1021 if (fd < 0) {
1022 ret = -errno;
1023 if (ret == -EROFS)
1024 ret = -EACCES;
1025 return ret;
1027 s->fd = fd;
1028 #if defined(__FreeBSD__)
1029 /* make sure the door isnt locked at this time */
1030 if (s->type == FTYPE_CD)
1031 ioctl (s->fd, CDIOCALLOW);
1032 #endif
1033 #if defined(__linux__)
1034 /* close fd so that we can reopen it as needed */
1035 if (s->type == FTYPE_FD) {
1036 close(s->fd);
1037 s->fd = -1;
1038 s->fd_media_changed = 1;
1040 #endif
1041 return 0;
1044 #if defined(__linux__)
1045 /* Note: we do not have a reliable method to detect if the floppy is
1046 present. The current method is to try to open the floppy at every
1047 I/O and to keep it opened during a few hundreds of ms. */
1048 static int fd_open(BlockDriverState *bs)
1050 BDRVRawState *s = bs->opaque;
1051 int last_media_present;
1053 if (s->type != FTYPE_FD)
1054 return 0;
1055 last_media_present = (s->fd >= 0);
1056 if (s->fd >= 0 &&
1057 (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1058 close(s->fd);
1059 s->fd = -1;
1060 #ifdef DEBUG_FLOPPY
1061 printf("Floppy closed\n");
1062 #endif
1064 if (s->fd < 0) {
1065 if (s->fd_got_error &&
1066 (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1067 #ifdef DEBUG_FLOPPY
1068 printf("No floppy (open delayed)\n");
1069 #endif
1070 return -EIO;
1072 s->fd = open(bs->filename, s->fd_open_flags);
1073 if (s->fd < 0) {
1074 s->fd_error_time = qemu_get_clock(rt_clock);
1075 s->fd_got_error = 1;
1076 if (last_media_present)
1077 s->fd_media_changed = 1;
1078 #ifdef DEBUG_FLOPPY
1079 printf("No floppy\n");
1080 #endif
1081 return -EIO;
1083 #ifdef DEBUG_FLOPPY
1084 printf("Floppy opened\n");
1085 #endif
1087 if (!last_media_present)
1088 s->fd_media_changed = 1;
1089 s->fd_open_time = qemu_get_clock(rt_clock);
1090 s->fd_got_error = 0;
1091 return 0;
1094 static int raw_is_inserted(BlockDriverState *bs)
1096 BDRVRawState *s = bs->opaque;
1097 int ret;
1099 switch(s->type) {
1100 case FTYPE_CD:
1101 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1102 if (ret == CDS_DISC_OK)
1103 return 1;
1104 else
1105 return 0;
1106 break;
1107 case FTYPE_FD:
1108 ret = fd_open(bs);
1109 return (ret >= 0);
1110 default:
1111 return 1;
1115 /* currently only used by fdc.c, but a CD version would be good too */
1116 static int raw_media_changed(BlockDriverState *bs)
1118 BDRVRawState *s = bs->opaque;
1120 switch(s->type) {
1121 case FTYPE_FD:
1123 int ret;
1124 /* XXX: we do not have a true media changed indication. It
1125 does not work if the floppy is changed without trying
1126 to read it */
1127 fd_open(bs);
1128 ret = s->fd_media_changed;
1129 s->fd_media_changed = 0;
1130 #ifdef DEBUG_FLOPPY
1131 printf("Floppy changed=%d\n", ret);
1132 #endif
1133 return ret;
1135 default:
1136 return -ENOTSUP;
1140 static int raw_eject(BlockDriverState *bs, int eject_flag)
1142 BDRVRawState *s = bs->opaque;
1144 switch(s->type) {
1145 case FTYPE_CD:
1146 if (eject_flag) {
1147 if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1148 perror("CDROMEJECT");
1149 } else {
1150 if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1151 perror("CDROMEJECT");
1153 break;
1154 case FTYPE_FD:
1156 int fd;
1157 if (s->fd >= 0) {
1158 close(s->fd);
1159 s->fd = -1;
1161 fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1162 if (fd >= 0) {
1163 if (ioctl(fd, FDEJECT, 0) < 0)
1164 perror("FDEJECT");
1165 close(fd);
1168 break;
1169 default:
1170 return -ENOTSUP;
1172 return 0;
1175 static int raw_set_locked(BlockDriverState *bs, int locked)
1177 BDRVRawState *s = bs->opaque;
1179 switch(s->type) {
1180 case FTYPE_CD:
1181 if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1182 /* Note: an error can happen if the distribution automatically
1183 mounts the CD-ROM */
1184 // perror("CDROM_LOCKDOOR");
1186 break;
1187 default:
1188 return -ENOTSUP;
1190 return 0;
1193 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1195 BDRVRawState *s = bs->opaque;
1197 return ioctl(s->fd, req, buf);
1200 #ifdef CONFIG_AIO
1201 static BlockDriverAIOCB *raw_aio_ioctl(BlockDriverState *bs,
1202 unsigned long int req, void *buf,
1203 BlockDriverCompletionFunc *cb, void *opaque)
1205 BDRVRawState *s = bs->opaque;
1206 RawAIOCB *acb;
1208 if (fd_open(bs) < 0)
1209 return NULL;
1211 acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque);
1212 if (!acb)
1213 return NULL;
1214 acb->aiocb.aio_fildes = s->fd;
1215 acb->aiocb.ev_signo = SIGUSR2;
1216 acb->aiocb.aio_offset = 0;
1217 acb->aiocb.aio_flags = 0;
1219 acb->next = posix_aio_state->first_aio;
1220 posix_aio_state->first_aio = acb;
1222 acb->aiocb.aio_ioctl_buf = buf;
1223 acb->aiocb.aio_ioctl_cmd = req;
1224 if (qemu_paio_ioctl(&acb->aiocb) < 0) {
1225 raw_aio_remove(acb);
1226 return NULL;
1229 return &acb->common;
1231 #endif
1233 #elif defined(__FreeBSD__)
1235 static int fd_open(BlockDriverState *bs)
1237 BDRVRawState *s = bs->opaque;
1239 /* this is just to ensure s->fd is sane (its called by io ops) */
1240 if (s->fd >= 0)
1241 return 0;
1242 return -EIO;
1245 static int cd_open(BlockDriverState *bs)
1247 #if defined(__FreeBSD__)
1248 BDRVRawState *s = bs->opaque;
1249 int fd;
1251 switch(s->type) {
1252 case FTYPE_CD:
1253 /* XXX force reread of possibly changed/newly loaded disc,
1254 * FreeBSD seems to not notice sometimes... */
1255 if (s->fd >= 0)
1256 close (s->fd);
1257 fd = open(bs->filename, s->cd_open_flags, 0644);
1258 if (fd < 0) {
1259 s->fd = -1;
1260 return -EIO;
1262 s->fd = fd;
1263 /* make sure the door isnt locked at this time */
1264 ioctl (s->fd, CDIOCALLOW);
1266 #endif
1267 return 0;
1270 static int raw_is_inserted(BlockDriverState *bs)
1272 BDRVRawState *s = bs->opaque;
1274 switch(s->type) {
1275 case FTYPE_CD:
1276 return (raw_getlength(bs) > 0);
1277 case FTYPE_FD:
1278 /* XXX handle this */
1279 /* FALLTHRU */
1280 default:
1281 return 1;
1285 static int raw_media_changed(BlockDriverState *bs)
1287 return -ENOTSUP;
1290 static int raw_eject(BlockDriverState *bs, int eject_flag)
1292 BDRVRawState *s = bs->opaque;
1294 switch(s->type) {
1295 case FTYPE_CD:
1296 if (s->fd < 0)
1297 return -ENOTSUP;
1298 (void) ioctl (s->fd, CDIOCALLOW);
1299 if (eject_flag) {
1300 if (ioctl (s->fd, CDIOCEJECT) < 0)
1301 perror("CDIOCEJECT");
1302 } else {
1303 if (ioctl (s->fd, CDIOCCLOSE) < 0)
1304 perror("CDIOCCLOSE");
1306 if (cd_open(bs) < 0)
1307 return -ENOTSUP;
1308 break;
1309 case FTYPE_FD:
1310 /* XXX handle this */
1311 /* FALLTHRU */
1312 default:
1313 return -ENOTSUP;
1315 return 0;
1318 static int raw_set_locked(BlockDriverState *bs, int locked)
1320 BDRVRawState *s = bs->opaque;
1322 switch(s->type) {
1323 case FTYPE_CD:
1324 if (s->fd < 0)
1325 return -ENOTSUP;
1326 if (ioctl (s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1327 /* Note: an error can happen if the distribution automatically
1328 mounts the CD-ROM */
1329 // perror("CDROM_LOCKDOOR");
1331 break;
1332 default:
1333 return -ENOTSUP;
1335 return 0;
1338 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1340 return -ENOTSUP;
1342 #else /* !linux && !FreeBSD */
1344 static int fd_open(BlockDriverState *bs)
1346 return 0;
1349 static int raw_is_inserted(BlockDriverState *bs)
1351 return 1;
1354 static int raw_media_changed(BlockDriverState *bs)
1356 return -ENOTSUP;
1359 static int raw_eject(BlockDriverState *bs, int eject_flag)
1361 return -ENOTSUP;
1364 static int raw_set_locked(BlockDriverState *bs, int locked)
1366 return -ENOTSUP;
1369 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1371 return -ENOTSUP;
1374 static BlockDriverAIOCB *raw_aio_ioctl(BlockDriverState *bs,
1375 unsigned long int req, void *buf,
1376 BlockDriverCompletionFunc *cb, void *opaque)
1378 return NULL;
1380 #endif /* !linux && !FreeBSD */
1382 static int hdev_create(const char *filename, QEMUOptionParameter *options)
1384 int fd;
1385 int ret = 0;
1386 struct stat stat_buf;
1387 int64_t total_size = 0;
1389 /* Read out options */
1390 while (options && options->name) {
1391 if (!strcmp(options->name, "size")) {
1392 total_size = options->value.n / 512;
1394 options++;
1397 fd = open(filename, O_WRONLY | O_BINARY);
1398 if (fd < 0)
1399 return -EIO;
1401 if (fstat(fd, &stat_buf) < 0)
1402 ret = -EIO;
1403 else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode))
1404 ret = -EIO;
1405 else if (lseek(fd, 0, SEEK_END) < total_size * 512)
1406 ret = -ENOSPC;
1408 close(fd);
1409 return ret;
1412 static BlockDriver bdrv_host_device = {
1413 .format_name = "host_device",
1414 .instance_size = sizeof(BDRVRawState),
1415 .bdrv_open = hdev_open,
1416 .bdrv_close = raw_close,
1417 .bdrv_create = hdev_create,
1418 .bdrv_flush = raw_flush,
1420 #ifdef CONFIG_AIO
1421 .bdrv_aio_readv = raw_aio_readv,
1422 .bdrv_aio_writev = raw_aio_writev,
1423 #endif
1425 .bdrv_read = raw_read,
1426 .bdrv_write = raw_write,
1427 .bdrv_getlength = raw_getlength,
1429 /* removable device support */
1430 .bdrv_is_inserted = raw_is_inserted,
1431 .bdrv_media_changed = raw_media_changed,
1432 .bdrv_eject = raw_eject,
1433 .bdrv_set_locked = raw_set_locked,
1434 /* generic scsi device */
1435 .bdrv_ioctl = raw_ioctl,
1436 #ifdef CONFIG_AIO
1437 .bdrv_aio_ioctl = raw_aio_ioctl,
1438 #endif
1441 static void bdrv_raw_init(void)
1443 bdrv_register(&bdrv_raw);
1444 bdrv_register(&bdrv_host_device);
1447 block_init(bdrv_raw_init);