Fix vectored aio bounce handling immediate errors (Avi Kivity)
[qemu-kvm/fedora.git] / block-raw-posix.c
blobbe6dd8333faa6137d9fe49c94ccbe92e38656a52
1 /*
2 * Block driver for RAW files (posix)
4 * Copyright (c) 2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
24 #include "qemu-common.h"
25 #include "qemu-timer.h"
26 #include "qemu-char.h"
27 #include "block_int.h"
28 #include <assert.h>
29 #ifdef CONFIG_AIO
30 #include "posix-aio-compat.h"
31 #endif
33 #ifdef CONFIG_COCOA
34 #include <paths.h>
35 #include <sys/param.h>
36 #include <IOKit/IOKitLib.h>
37 #include <IOKit/IOBSD.h>
38 #include <IOKit/storage/IOMediaBSDClient.h>
39 #include <IOKit/storage/IOMedia.h>
40 #include <IOKit/storage/IOCDMedia.h>
41 //#include <IOKit/storage/IOCDTypes.h>
42 #include <CoreFoundation/CoreFoundation.h>
43 #endif
45 #ifdef __sun__
46 #define _POSIX_PTHREAD_SEMANTICS 1
47 #include <signal.h>
48 #include <sys/dkio.h>
49 #endif
50 #ifdef __linux__
51 #include <sys/ioctl.h>
52 #include <linux/cdrom.h>
53 #include <linux/fd.h>
54 #endif
55 #ifdef __FreeBSD__
56 #include <signal.h>
57 #include <sys/disk.h>
58 #include <sys/cdio.h>
59 #endif
61 #ifdef __OpenBSD__
62 #include <sys/ioctl.h>
63 #include <sys/disklabel.h>
64 #include <sys/dkio.h>
65 #endif
67 #ifdef __DragonFly__
68 #include <sys/ioctl.h>
69 #include <sys/diskslice.h>
70 #endif
72 //#define DEBUG_FLOPPY
74 //#define DEBUG_BLOCK
75 #if defined(DEBUG_BLOCK)
76 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (qemu_log_enabled()) \
77 { qemu_log(formatCstr, ##args); qemu_log_flush(); } } while (0)
78 #else
79 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
80 #endif
82 /* OS X does not have O_DSYNC */
83 #ifndef O_DSYNC
84 #define O_DSYNC O_SYNC
85 #endif
87 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
88 #ifndef O_DIRECT
89 #define O_DIRECT O_DSYNC
90 #endif
92 #define FTYPE_FILE 0
93 #define FTYPE_CD 1
94 #define FTYPE_FD 2
96 #define ALIGNED_BUFFER_SIZE (32 * 512)
98 /* if the FD is not accessed during that time (in ms), we try to
99 reopen it to see if the disk has been changed */
100 #define FD_OPEN_TIMEOUT 1000
102 typedef struct BDRVRawState {
103 int fd;
104 int type;
105 unsigned int lseek_err_cnt;
106 #if defined(__linux__)
107 /* linux floppy specific */
108 int fd_open_flags;
109 int64_t fd_open_time;
110 int64_t fd_error_time;
111 int fd_got_error;
112 int fd_media_changed;
113 #endif
114 #if defined(__FreeBSD__)
115 int cd_open_flags;
116 #endif
117 uint8_t* aligned_buf;
118 } BDRVRawState;
120 static int posix_aio_init(void);
122 static int fd_open(BlockDriverState *bs);
124 #if defined(__FreeBSD__)
125 static int cd_open(BlockDriverState *bs);
126 #endif
128 static int raw_is_inserted(BlockDriverState *bs);
130 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
132 BDRVRawState *s = bs->opaque;
133 int fd, open_flags, ret;
135 posix_aio_init();
137 s->lseek_err_cnt = 0;
139 open_flags = O_BINARY;
140 if ((flags & BDRV_O_ACCESS) == O_RDWR) {
141 open_flags |= O_RDWR;
142 } else {
143 open_flags |= O_RDONLY;
144 bs->read_only = 1;
146 if (flags & BDRV_O_CREAT)
147 open_flags |= O_CREAT | O_TRUNC;
149 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
150 * and O_DIRECT for no caching. */
151 if ((flags & BDRV_O_NOCACHE))
152 open_flags |= O_DIRECT;
153 else if (!(flags & BDRV_O_CACHE_WB))
154 open_flags |= O_DSYNC;
156 s->type = FTYPE_FILE;
158 fd = open(filename, open_flags, 0644);
159 if (fd < 0) {
160 ret = -errno;
161 if (ret == -EROFS)
162 ret = -EACCES;
163 return ret;
165 s->fd = fd;
166 s->aligned_buf = NULL;
167 if ((flags & BDRV_O_NOCACHE)) {
168 s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
169 if (s->aligned_buf == NULL) {
170 ret = -errno;
171 close(fd);
172 return ret;
175 return 0;
178 /* XXX: use host sector size if necessary with:
179 #ifdef DIOCGSECTORSIZE
181 unsigned int sectorsize = 512;
182 if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
183 sectorsize > bufsize)
184 bufsize = sectorsize;
186 #endif
187 #ifdef CONFIG_COCOA
188 u_int32_t blockSize = 512;
189 if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
190 bufsize = blockSize;
192 #endif
196 * offset and count are in bytes, but must be multiples of 512 for files
197 * opened with O_DIRECT. buf must be aligned to 512 bytes then.
199 * This function may be called without alignment if the caller ensures
200 * that O_DIRECT is not in effect.
202 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
203 uint8_t *buf, int count)
205 BDRVRawState *s = bs->opaque;
206 int ret;
208 ret = fd_open(bs);
209 if (ret < 0)
210 return ret;
212 if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
213 ++(s->lseek_err_cnt);
214 if(s->lseek_err_cnt <= 10) {
215 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
216 "] lseek failed : %d = %s\n",
217 s->fd, bs->filename, offset, buf, count,
218 bs->total_sectors, errno, strerror(errno));
220 return -1;
222 s->lseek_err_cnt=0;
224 ret = read(s->fd, buf, count);
225 if (ret == count)
226 goto label__raw_read__success;
228 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
229 "] read failed %d : %d = %s\n",
230 s->fd, bs->filename, offset, buf, count,
231 bs->total_sectors, ret, errno, strerror(errno));
233 /* Try harder for CDrom. */
234 if (bs->type == BDRV_TYPE_CDROM) {
235 lseek(s->fd, offset, SEEK_SET);
236 ret = read(s->fd, buf, count);
237 if (ret == count)
238 goto label__raw_read__success;
239 lseek(s->fd, offset, SEEK_SET);
240 ret = read(s->fd, buf, count);
241 if (ret == count)
242 goto label__raw_read__success;
244 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
245 "] retry read failed %d : %d = %s\n",
246 s->fd, bs->filename, offset, buf, count,
247 bs->total_sectors, ret, errno, strerror(errno));
250 label__raw_read__success:
252 return ret;
256 * offset and count are in bytes, but must be multiples of 512 for files
257 * opened with O_DIRECT. buf must be aligned to 512 bytes then.
259 * This function may be called without alignment if the caller ensures
260 * that O_DIRECT is not in effect.
262 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
263 const uint8_t *buf, int count)
265 BDRVRawState *s = bs->opaque;
266 int ret;
268 ret = fd_open(bs);
269 if (ret < 0)
270 return -errno;
272 if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
273 ++(s->lseek_err_cnt);
274 if(s->lseek_err_cnt) {
275 DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
276 PRId64 "] lseek failed : %d = %s\n",
277 s->fd, bs->filename, offset, buf, count,
278 bs->total_sectors, errno, strerror(errno));
280 return -EIO;
282 s->lseek_err_cnt = 0;
284 ret = write(s->fd, buf, count);
285 if (ret == count)
286 goto label__raw_write__success;
288 DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
289 "] write failed %d : %d = %s\n",
290 s->fd, bs->filename, offset, buf, count,
291 bs->total_sectors, ret, errno, strerror(errno));
293 label__raw_write__success:
295 return (ret < 0) ? -errno : ret;
300 * offset and count are in bytes and possibly not aligned. For files opened
301 * with O_DIRECT, necessary alignments are ensured before calling
302 * raw_pread_aligned to do the actual read.
304 static int raw_pread(BlockDriverState *bs, int64_t offset,
305 uint8_t *buf, int count)
307 BDRVRawState *s = bs->opaque;
308 int size, ret, shift, sum;
310 sum = 0;
312 if (s->aligned_buf != NULL) {
314 if (offset & 0x1ff) {
315 /* align offset on a 512 bytes boundary */
317 shift = offset & 0x1ff;
318 size = (shift + count + 0x1ff) & ~0x1ff;
319 if (size > ALIGNED_BUFFER_SIZE)
320 size = ALIGNED_BUFFER_SIZE;
321 ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
322 if (ret < 0)
323 return ret;
325 size = 512 - shift;
326 if (size > count)
327 size = count;
328 memcpy(buf, s->aligned_buf + shift, size);
330 buf += size;
331 offset += size;
332 count -= size;
333 sum += size;
335 if (count == 0)
336 return sum;
338 if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
340 /* read on aligned buffer */
342 while (count) {
344 size = (count + 0x1ff) & ~0x1ff;
345 if (size > ALIGNED_BUFFER_SIZE)
346 size = ALIGNED_BUFFER_SIZE;
348 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
349 if (ret < 0)
350 return ret;
352 size = ret;
353 if (size > count)
354 size = count;
356 memcpy(buf, s->aligned_buf, size);
358 buf += size;
359 offset += size;
360 count -= size;
361 sum += size;
364 return sum;
368 return raw_pread_aligned(bs, offset, buf, count) + sum;
371 static int raw_read(BlockDriverState *bs, int64_t sector_num,
372 uint8_t *buf, int nb_sectors)
374 int ret;
376 ret = raw_pread(bs, sector_num * 512, buf, nb_sectors * 512);
377 if (ret == (nb_sectors * 512))
378 ret = 0;
379 return ret;
383 * offset and count are in bytes and possibly not aligned. For files opened
384 * with O_DIRECT, necessary alignments are ensured before calling
385 * raw_pwrite_aligned to do the actual write.
387 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
388 const uint8_t *buf, int count)
390 BDRVRawState *s = bs->opaque;
391 int size, ret, shift, sum;
393 sum = 0;
395 if (s->aligned_buf != NULL) {
397 if (offset & 0x1ff) {
398 /* align offset on a 512 bytes boundary */
399 shift = offset & 0x1ff;
400 ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
401 if (ret < 0)
402 return ret;
404 size = 512 - shift;
405 if (size > count)
406 size = count;
407 memcpy(s->aligned_buf + shift, buf, size);
409 ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
410 if (ret < 0)
411 return ret;
413 buf += size;
414 offset += size;
415 count -= size;
416 sum += size;
418 if (count == 0)
419 return sum;
421 if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
423 while ((size = (count & ~0x1ff)) != 0) {
425 if (size > ALIGNED_BUFFER_SIZE)
426 size = ALIGNED_BUFFER_SIZE;
428 memcpy(s->aligned_buf, buf, size);
430 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
431 if (ret < 0)
432 return ret;
434 buf += ret;
435 offset += ret;
436 count -= ret;
437 sum += ret;
439 /* here, count < 512 because (count & ~0x1ff) == 0 */
440 if (count) {
441 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
442 if (ret < 0)
443 return ret;
444 memcpy(s->aligned_buf, buf, count);
446 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
447 if (ret < 0)
448 return ret;
449 if (count < ret)
450 ret = count;
452 sum += ret;
454 return sum;
457 return raw_pwrite_aligned(bs, offset, buf, count) + sum;
460 static int raw_write(BlockDriverState *bs, int64_t sector_num,
461 const uint8_t *buf, int nb_sectors)
463 int ret;
464 ret = raw_pwrite(bs, sector_num * 512, buf, nb_sectors * 512);
465 if (ret == (nb_sectors * 512))
466 ret = 0;
467 return ret;
470 #ifdef CONFIG_AIO
471 /***********************************************************/
472 /* Unix AIO using POSIX AIO */
474 typedef struct RawAIOCB {
475 BlockDriverAIOCB common;
476 struct qemu_paiocb aiocb;
477 struct RawAIOCB *next;
478 int ret;
479 } RawAIOCB;
481 typedef struct PosixAioState
483 int rfd, wfd;
484 RawAIOCB *first_aio;
485 } PosixAioState;
487 static void posix_aio_read(void *opaque)
489 PosixAioState *s = opaque;
490 RawAIOCB *acb, **pacb;
491 int ret;
492 ssize_t len;
494 /* read all bytes from signal pipe */
495 for (;;) {
496 char bytes[16];
498 len = read(s->rfd, bytes, sizeof(bytes));
499 if (len == -1 && errno == EINTR)
500 continue; /* try again */
501 if (len == sizeof(bytes))
502 continue; /* more to read */
503 break;
506 for(;;) {
507 pacb = &s->first_aio;
508 for(;;) {
509 acb = *pacb;
510 if (!acb)
511 goto the_end;
512 ret = qemu_paio_error(&acb->aiocb);
513 if (ret == ECANCELED) {
514 /* remove the request */
515 *pacb = acb->next;
516 qemu_aio_release(acb);
517 } else if (ret != EINPROGRESS) {
518 /* end of aio */
519 if (ret == 0) {
520 ret = qemu_paio_return(&acb->aiocb);
521 if (ret == acb->aiocb.aio_nbytes)
522 ret = 0;
523 else
524 ret = -EINVAL;
525 } else {
526 ret = -ret;
528 /* remove the request */
529 *pacb = acb->next;
530 /* call the callback */
531 acb->common.cb(acb->common.opaque, ret);
532 qemu_aio_release(acb);
533 break;
534 } else {
535 pacb = &acb->next;
539 the_end: ;
542 static int posix_aio_flush(void *opaque)
544 PosixAioState *s = opaque;
545 return !!s->first_aio;
548 static PosixAioState *posix_aio_state;
550 static void aio_signal_handler(int signum)
552 if (posix_aio_state) {
553 char byte = 0;
555 write(posix_aio_state->wfd, &byte, sizeof(byte));
558 qemu_service_io();
561 static int posix_aio_init(void)
563 struct sigaction act;
564 PosixAioState *s;
565 int fds[2];
566 struct qemu_paioinit ai;
568 if (posix_aio_state)
569 return 0;
571 s = qemu_malloc(sizeof(PosixAioState));
573 sigfillset(&act.sa_mask);
574 act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
575 act.sa_handler = aio_signal_handler;
576 sigaction(SIGUSR2, &act, NULL);
578 s->first_aio = NULL;
579 if (pipe(fds) == -1) {
580 fprintf(stderr, "failed to create pipe\n");
581 return -errno;
584 s->rfd = fds[0];
585 s->wfd = fds[1];
587 fcntl(s->rfd, F_SETFL, O_NONBLOCK);
588 fcntl(s->wfd, F_SETFL, O_NONBLOCK);
590 qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);
592 memset(&ai, 0, sizeof(ai));
593 ai.aio_threads = 64;
594 ai.aio_num = 64;
595 qemu_paio_init(&ai);
597 posix_aio_state = s;
599 return 0;
602 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
603 int64_t sector_num, uint8_t *buf, int nb_sectors,
604 BlockDriverCompletionFunc *cb, void *opaque)
606 BDRVRawState *s = bs->opaque;
607 RawAIOCB *acb;
609 if (fd_open(bs) < 0)
610 return NULL;
612 acb = qemu_aio_get(bs, cb, opaque);
613 if (!acb)
614 return NULL;
615 acb->aiocb.aio_fildes = s->fd;
616 acb->aiocb.ev_signo = SIGUSR2;
617 acb->aiocb.aio_buf = buf;
618 if (nb_sectors < 0)
619 acb->aiocb.aio_nbytes = -nb_sectors;
620 else
621 acb->aiocb.aio_nbytes = nb_sectors * 512;
622 acb->aiocb.aio_offset = sector_num * 512;
623 acb->next = posix_aio_state->first_aio;
624 posix_aio_state->first_aio = acb;
625 return acb;
628 static void raw_aio_em_cb(void* opaque)
630 RawAIOCB *acb = opaque;
631 acb->common.cb(acb->common.opaque, acb->ret);
632 qemu_aio_release(acb);
635 static void raw_aio_remove(RawAIOCB *acb)
637 RawAIOCB **pacb;
639 /* remove the callback from the queue */
640 pacb = &posix_aio_state->first_aio;
641 for(;;) {
642 if (*pacb == NULL) {
643 fprintf(stderr, "raw_aio_remove: aio request not found!\n");
644 break;
645 } else if (*pacb == acb) {
646 *pacb = acb->next;
647 qemu_aio_release(acb);
648 break;
650 pacb = &(*pacb)->next;
654 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
655 int64_t sector_num, uint8_t *buf, int nb_sectors,
656 BlockDriverCompletionFunc *cb, void *opaque)
658 RawAIOCB *acb;
661 * If O_DIRECT is used and the buffer is not aligned fall back
662 * to synchronous IO.
664 BDRVRawState *s = bs->opaque;
666 if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
667 QEMUBH *bh;
668 acb = qemu_aio_get(bs, cb, opaque);
669 acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
670 bh = qemu_bh_new(raw_aio_em_cb, acb);
671 qemu_bh_schedule(bh);
672 return &acb->common;
675 acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
676 if (!acb)
677 return NULL;
678 if (qemu_paio_read(&acb->aiocb) < 0) {
679 raw_aio_remove(acb);
680 return NULL;
682 return &acb->common;
685 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
686 int64_t sector_num, const uint8_t *buf, int nb_sectors,
687 BlockDriverCompletionFunc *cb, void *opaque)
689 RawAIOCB *acb;
692 * If O_DIRECT is used and the buffer is not aligned fall back
693 * to synchronous IO.
695 BDRVRawState *s = bs->opaque;
697 if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
698 QEMUBH *bh;
699 acb = qemu_aio_get(bs, cb, opaque);
700 acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
701 bh = qemu_bh_new(raw_aio_em_cb, acb);
702 qemu_bh_schedule(bh);
703 return &acb->common;
706 acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
707 if (!acb)
708 return NULL;
709 if (qemu_paio_write(&acb->aiocb) < 0) {
710 raw_aio_remove(acb);
711 return NULL;
713 return &acb->common;
716 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
718 int ret;
719 RawAIOCB *acb = (RawAIOCB *)blockacb;
721 ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
722 if (ret == QEMU_PAIO_NOTCANCELED) {
723 /* fail safe: if the aio could not be canceled, we wait for
724 it */
725 while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
728 raw_aio_remove(acb);
730 #else /* CONFIG_AIO */
731 static int posix_aio_init(void)
733 return 0;
735 #endif /* CONFIG_AIO */
738 static void raw_close(BlockDriverState *bs)
740 BDRVRawState *s = bs->opaque;
741 if (s->fd >= 0) {
742 close(s->fd);
743 s->fd = -1;
744 if (s->aligned_buf != NULL)
745 qemu_free(s->aligned_buf);
749 static int raw_truncate(BlockDriverState *bs, int64_t offset)
751 BDRVRawState *s = bs->opaque;
752 if (s->type != FTYPE_FILE)
753 return -ENOTSUP;
754 if (ftruncate(s->fd, offset) < 0)
755 return -errno;
756 return 0;
759 #ifdef __OpenBSD__
760 static int64_t raw_getlength(BlockDriverState *bs)
762 BDRVRawState *s = bs->opaque;
763 int fd = s->fd;
764 struct stat st;
766 if (fstat(fd, &st))
767 return -1;
768 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
769 struct disklabel dl;
771 if (ioctl(fd, DIOCGDINFO, &dl))
772 return -1;
773 return (uint64_t)dl.d_secsize *
774 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
775 } else
776 return st.st_size;
778 #else /* !__OpenBSD__ */
779 static int64_t raw_getlength(BlockDriverState *bs)
781 BDRVRawState *s = bs->opaque;
782 int fd = s->fd;
783 int64_t size;
784 #ifdef HOST_BSD
785 struct stat sb;
786 #ifdef __FreeBSD__
787 int reopened = 0;
788 #endif
789 #endif
790 #ifdef __sun__
791 struct dk_minfo minfo;
792 int rv;
793 #endif
794 int ret;
796 ret = fd_open(bs);
797 if (ret < 0)
798 return ret;
800 #ifdef HOST_BSD
801 #ifdef __FreeBSD__
802 again:
803 #endif
804 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
805 #ifdef DIOCGMEDIASIZE
806 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
807 #elif defined(DIOCGPART)
809 struct partinfo pi;
810 if (ioctl(fd, DIOCGPART, &pi) == 0)
811 size = pi.media_size;
812 else
813 size = 0;
815 if (size == 0)
816 #endif
817 #ifdef CONFIG_COCOA
818 size = LONG_LONG_MAX;
819 #else
820 size = lseek(fd, 0LL, SEEK_END);
821 #endif
822 #ifdef __FreeBSD__
823 switch(s->type) {
824 case FTYPE_CD:
825 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
826 if (size == 2048LL * (unsigned)-1)
827 size = 0;
828 /* XXX no disc? maybe we need to reopen... */
829 if (size <= 0 && !reopened && cd_open(bs) >= 0) {
830 reopened = 1;
831 goto again;
834 #endif
835 } else
836 #endif
837 #ifdef __sun__
839 * use the DKIOCGMEDIAINFO ioctl to read the size.
841 rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
842 if ( rv != -1 ) {
843 size = minfo.dki_lbsize * minfo.dki_capacity;
844 } else /* there are reports that lseek on some devices
845 fails, but irc discussion said that contingency
846 on contingency was overkill */
847 #endif
849 size = lseek(fd, 0, SEEK_END);
851 return size;
853 #endif
855 static int raw_create(const char *filename, int64_t total_size,
856 const char *backing_file, int flags)
858 int fd;
860 if (flags || backing_file)
861 return -ENOTSUP;
863 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
864 0644);
865 if (fd < 0)
866 return -EIO;
867 ftruncate(fd, total_size * 512);
868 close(fd);
869 return 0;
872 static void raw_flush(BlockDriverState *bs)
874 BDRVRawState *s = bs->opaque;
875 fsync(s->fd);
878 BlockDriver bdrv_raw = {
879 "raw",
880 sizeof(BDRVRawState),
881 NULL, /* no probe for protocols */
882 raw_open,
883 NULL,
884 NULL,
885 raw_close,
886 raw_create,
887 raw_flush,
889 #ifdef CONFIG_AIO
890 .bdrv_aio_read = raw_aio_read,
891 .bdrv_aio_write = raw_aio_write,
892 .bdrv_aio_cancel = raw_aio_cancel,
893 .aiocb_size = sizeof(RawAIOCB),
894 #endif
896 .bdrv_read = raw_read,
897 .bdrv_write = raw_write,
898 .bdrv_truncate = raw_truncate,
899 .bdrv_getlength = raw_getlength,
902 /***********************************************/
903 /* host device */
905 #ifdef CONFIG_COCOA
906 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
907 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
909 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
911 kern_return_t kernResult;
912 mach_port_t masterPort;
913 CFMutableDictionaryRef classesToMatch;
915 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
916 if ( KERN_SUCCESS != kernResult ) {
917 printf( "IOMasterPort returned %d\n", kernResult );
920 classesToMatch = IOServiceMatching( kIOCDMediaClass );
921 if ( classesToMatch == NULL ) {
922 printf( "IOServiceMatching returned a NULL dictionary.\n" );
923 } else {
924 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
926 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
927 if ( KERN_SUCCESS != kernResult )
929 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
932 return kernResult;
935 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
937 io_object_t nextMedia;
938 kern_return_t kernResult = KERN_FAILURE;
939 *bsdPath = '\0';
940 nextMedia = IOIteratorNext( mediaIterator );
941 if ( nextMedia )
943 CFTypeRef bsdPathAsCFString;
944 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
945 if ( bsdPathAsCFString ) {
946 size_t devPathLength;
947 strcpy( bsdPath, _PATH_DEV );
948 strcat( bsdPath, "r" );
949 devPathLength = strlen( bsdPath );
950 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
951 kernResult = KERN_SUCCESS;
953 CFRelease( bsdPathAsCFString );
955 IOObjectRelease( nextMedia );
958 return kernResult;
961 #endif
963 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
965 BDRVRawState *s = bs->opaque;
966 int fd, open_flags, ret;
968 posix_aio_init();
970 #ifdef CONFIG_COCOA
971 if (strstart(filename, "/dev/cdrom", NULL)) {
972 kern_return_t kernResult;
973 io_iterator_t mediaIterator;
974 char bsdPath[ MAXPATHLEN ];
975 int fd;
977 kernResult = FindEjectableCDMedia( &mediaIterator );
978 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
980 if ( bsdPath[ 0 ] != '\0' ) {
981 strcat(bsdPath,"s0");
982 /* some CDs don't have a partition 0 */
983 fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
984 if (fd < 0) {
985 bsdPath[strlen(bsdPath)-1] = '1';
986 } else {
987 close(fd);
989 filename = bsdPath;
992 if ( mediaIterator )
993 IOObjectRelease( mediaIterator );
995 #endif
996 open_flags = O_BINARY;
997 if ((flags & BDRV_O_ACCESS) == O_RDWR) {
998 open_flags |= O_RDWR;
999 } else {
1000 open_flags |= O_RDONLY;
1001 bs->read_only = 1;
1003 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
1004 * and O_DIRECT for no caching. */
1005 if ((flags & BDRV_O_NOCACHE))
1006 open_flags |= O_DIRECT;
1007 else if (!(flags & BDRV_O_CACHE_WB))
1008 open_flags |= O_DSYNC;
1010 s->type = FTYPE_FILE;
1011 #if defined(__linux__)
1012 if (strstart(filename, "/dev/cd", NULL)) {
1013 /* open will not fail even if no CD is inserted */
1014 open_flags |= O_NONBLOCK;
1015 s->type = FTYPE_CD;
1016 } else if (strstart(filename, "/dev/fd", NULL)) {
1017 s->type = FTYPE_FD;
1018 s->fd_open_flags = open_flags;
1019 /* open will not fail even if no floppy is inserted */
1020 open_flags |= O_NONBLOCK;
1021 } else if (strstart(filename, "/dev/sg", NULL)) {
1022 bs->sg = 1;
1024 #endif
1025 #if defined(__FreeBSD__)
1026 if (strstart(filename, "/dev/cd", NULL) ||
1027 strstart(filename, "/dev/acd", NULL)) {
1028 s->type = FTYPE_CD;
1029 s->cd_open_flags = open_flags;
1031 #endif
1032 s->fd = -1;
1033 fd = open(filename, open_flags, 0644);
1034 if (fd < 0) {
1035 ret = -errno;
1036 if (ret == -EROFS)
1037 ret = -EACCES;
1038 return ret;
1040 s->fd = fd;
1041 #if defined(__FreeBSD__)
1042 /* make sure the door isnt locked at this time */
1043 if (s->type == FTYPE_CD)
1044 ioctl (s->fd, CDIOCALLOW);
1045 #endif
1046 #if defined(__linux__)
1047 /* close fd so that we can reopen it as needed */
1048 if (s->type == FTYPE_FD) {
1049 close(s->fd);
1050 s->fd = -1;
1051 s->fd_media_changed = 1;
1053 #endif
1054 return 0;
1057 #if defined(__linux__)
1058 /* Note: we do not have a reliable method to detect if the floppy is
1059 present. The current method is to try to open the floppy at every
1060 I/O and to keep it opened during a few hundreds of ms. */
1061 static int fd_open(BlockDriverState *bs)
1063 BDRVRawState *s = bs->opaque;
1064 int last_media_present;
1066 if (s->type != FTYPE_FD)
1067 return 0;
1068 last_media_present = (s->fd >= 0);
1069 if (s->fd >= 0 &&
1070 (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1071 close(s->fd);
1072 s->fd = -1;
1073 #ifdef DEBUG_FLOPPY
1074 printf("Floppy closed\n");
1075 #endif
1077 if (s->fd < 0) {
1078 if (s->fd_got_error &&
1079 (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1080 #ifdef DEBUG_FLOPPY
1081 printf("No floppy (open delayed)\n");
1082 #endif
1083 return -EIO;
1085 s->fd = open(bs->filename, s->fd_open_flags);
1086 if (s->fd < 0) {
1087 s->fd_error_time = qemu_get_clock(rt_clock);
1088 s->fd_got_error = 1;
1089 if (last_media_present)
1090 s->fd_media_changed = 1;
1091 #ifdef DEBUG_FLOPPY
1092 printf("No floppy\n");
1093 #endif
1094 return -EIO;
1096 #ifdef DEBUG_FLOPPY
1097 printf("Floppy opened\n");
1098 #endif
1100 if (!last_media_present)
1101 s->fd_media_changed = 1;
1102 s->fd_open_time = qemu_get_clock(rt_clock);
1103 s->fd_got_error = 0;
1104 return 0;
1107 static int raw_is_inserted(BlockDriverState *bs)
1109 BDRVRawState *s = bs->opaque;
1110 int ret;
1112 switch(s->type) {
1113 case FTYPE_CD:
1114 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1115 if (ret == CDS_DISC_OK)
1116 return 1;
1117 else
1118 return 0;
1119 break;
1120 case FTYPE_FD:
1121 ret = fd_open(bs);
1122 return (ret >= 0);
1123 default:
1124 return 1;
1128 /* currently only used by fdc.c, but a CD version would be good too */
1129 static int raw_media_changed(BlockDriverState *bs)
1131 BDRVRawState *s = bs->opaque;
1133 switch(s->type) {
1134 case FTYPE_FD:
1136 int ret;
1137 /* XXX: we do not have a true media changed indication. It
1138 does not work if the floppy is changed without trying
1139 to read it */
1140 fd_open(bs);
1141 ret = s->fd_media_changed;
1142 s->fd_media_changed = 0;
1143 #ifdef DEBUG_FLOPPY
1144 printf("Floppy changed=%d\n", ret);
1145 #endif
1146 return ret;
1148 default:
1149 return -ENOTSUP;
1153 static int raw_eject(BlockDriverState *bs, int eject_flag)
1155 BDRVRawState *s = bs->opaque;
1157 switch(s->type) {
1158 case FTYPE_CD:
1159 if (eject_flag) {
1160 if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1161 perror("CDROMEJECT");
1162 } else {
1163 if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1164 perror("CDROMEJECT");
1166 break;
1167 case FTYPE_FD:
1169 int fd;
1170 if (s->fd >= 0) {
1171 close(s->fd);
1172 s->fd = -1;
1174 fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1175 if (fd >= 0) {
1176 if (ioctl(fd, FDEJECT, 0) < 0)
1177 perror("FDEJECT");
1178 close(fd);
1181 break;
1182 default:
1183 return -ENOTSUP;
1185 return 0;
1188 static int raw_set_locked(BlockDriverState *bs, int locked)
1190 BDRVRawState *s = bs->opaque;
1192 switch(s->type) {
1193 case FTYPE_CD:
1194 if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1195 /* Note: an error can happen if the distribution automatically
1196 mounts the CD-ROM */
1197 // perror("CDROM_LOCKDOOR");
1199 break;
1200 default:
1201 return -ENOTSUP;
1203 return 0;
1206 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1208 BDRVRawState *s = bs->opaque;
1210 return ioctl(s->fd, req, buf);
1212 #elif defined(__FreeBSD__)
1214 static int fd_open(BlockDriverState *bs)
1216 BDRVRawState *s = bs->opaque;
1218 /* this is just to ensure s->fd is sane (its called by io ops) */
1219 if (s->fd >= 0)
1220 return 0;
1221 return -EIO;
1224 static int cd_open(BlockDriverState *bs)
1226 #if defined(__FreeBSD__)
1227 BDRVRawState *s = bs->opaque;
1228 int fd;
1230 switch(s->type) {
1231 case FTYPE_CD:
1232 /* XXX force reread of possibly changed/newly loaded disc,
1233 * FreeBSD seems to not notice sometimes... */
1234 if (s->fd >= 0)
1235 close (s->fd);
1236 fd = open(bs->filename, s->cd_open_flags, 0644);
1237 if (fd < 0) {
1238 s->fd = -1;
1239 return -EIO;
1241 s->fd = fd;
1242 /* make sure the door isnt locked at this time */
1243 ioctl (s->fd, CDIOCALLOW);
1245 #endif
1246 return 0;
1249 static int raw_is_inserted(BlockDriverState *bs)
1251 BDRVRawState *s = bs->opaque;
1253 switch(s->type) {
1254 case FTYPE_CD:
1255 return (raw_getlength(bs) > 0);
1256 case FTYPE_FD:
1257 /* XXX handle this */
1258 /* FALLTHRU */
1259 default:
1260 return 1;
1264 static int raw_media_changed(BlockDriverState *bs)
1266 return -ENOTSUP;
1269 static int raw_eject(BlockDriverState *bs, int eject_flag)
1271 BDRVRawState *s = bs->opaque;
1273 switch(s->type) {
1274 case FTYPE_CD:
1275 if (s->fd < 0)
1276 return -ENOTSUP;
1277 (void) ioctl (s->fd, CDIOCALLOW);
1278 if (eject_flag) {
1279 if (ioctl (s->fd, CDIOCEJECT) < 0)
1280 perror("CDIOCEJECT");
1281 } else {
1282 if (ioctl (s->fd, CDIOCCLOSE) < 0)
1283 perror("CDIOCCLOSE");
1285 if (cd_open(bs) < 0)
1286 return -ENOTSUP;
1287 break;
1288 case FTYPE_FD:
1289 /* XXX handle this */
1290 /* FALLTHRU */
1291 default:
1292 return -ENOTSUP;
1294 return 0;
1297 static int raw_set_locked(BlockDriverState *bs, int locked)
1299 BDRVRawState *s = bs->opaque;
1301 switch(s->type) {
1302 case FTYPE_CD:
1303 if (s->fd < 0)
1304 return -ENOTSUP;
1305 if (ioctl (s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1306 /* Note: an error can happen if the distribution automatically
1307 mounts the CD-ROM */
1308 // perror("CDROM_LOCKDOOR");
1310 break;
1311 default:
1312 return -ENOTSUP;
1314 return 0;
1317 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1319 return -ENOTSUP;
1321 #else /* !linux && !FreeBSD */
1323 static int fd_open(BlockDriverState *bs)
1325 return 0;
1328 static int raw_is_inserted(BlockDriverState *bs)
1330 return 1;
1333 static int raw_media_changed(BlockDriverState *bs)
1335 return -ENOTSUP;
1338 static int raw_eject(BlockDriverState *bs, int eject_flag)
1340 return -ENOTSUP;
1343 static int raw_set_locked(BlockDriverState *bs, int locked)
1345 return -ENOTSUP;
1348 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1350 return -ENOTSUP;
1352 #endif /* !linux && !FreeBSD */
1354 static int raw_sg_send_command(BlockDriverState *bs, void *buf, int count)
1356 return raw_pwrite(bs, -1, buf, count);
1359 static int raw_sg_recv_response(BlockDriverState *bs, void *buf, int count)
1361 return raw_pread(bs, -1, buf, count);
1364 static BlockDriverAIOCB *raw_sg_aio_read(BlockDriverState *bs,
1365 void *buf, int count,
1366 BlockDriverCompletionFunc *cb,
1367 void *opaque)
1369 return raw_aio_read(bs, 0, buf, -(int64_t)count, cb, opaque);
1372 static BlockDriverAIOCB *raw_sg_aio_write(BlockDriverState *bs,
1373 void *buf, int count,
1374 BlockDriverCompletionFunc *cb,
1375 void *opaque)
1377 return raw_aio_write(bs, 0, buf, -(int64_t)count, cb, opaque);
1380 BlockDriver bdrv_host_device = {
1381 .format_name = "host_device",
1382 .instance_size = sizeof(BDRVRawState),
1383 .bdrv_open = hdev_open,
1384 .bdrv_close = raw_close,
1385 .bdrv_flush = raw_flush,
1387 #ifdef CONFIG_AIO
1388 .bdrv_aio_read = raw_aio_read,
1389 .bdrv_aio_write = raw_aio_write,
1390 .bdrv_aio_cancel = raw_aio_cancel,
1391 .aiocb_size = sizeof(RawAIOCB),
1392 #endif
1394 .bdrv_read = raw_read,
1395 .bdrv_write = raw_write,
1396 .bdrv_getlength = raw_getlength,
1398 /* removable device support */
1399 .bdrv_is_inserted = raw_is_inserted,
1400 .bdrv_media_changed = raw_media_changed,
1401 .bdrv_eject = raw_eject,
1402 .bdrv_set_locked = raw_set_locked,
1403 /* generic scsi device */
1404 .bdrv_ioctl = raw_ioctl,
1405 .bdrv_sg_send_command = raw_sg_send_command,
1406 .bdrv_sg_recv_response = raw_sg_recv_response,
1407 .bdrv_sg_aio_read = raw_sg_aio_read,
1408 .bdrv_sg_aio_write = raw_sg_aio_write,