Revision 9ef91a67 block/raw-posix.c
| b/block/raw-posix.c | ||
|---|---|---|
| 27 | 27 |
#include "qemu-log.h" |
| 28 | 28 |
#include "block_int.h" |
| 29 | 29 |
#include "module.h" |
| 30 |
#include "posix-aio-compat.h"
|
|
| 30 |
#include "block/raw-posix-aio.h"
|
|
| 31 | 31 |
|
| 32 | 32 |
#ifdef CONFIG_COCOA |
| 33 | 33 |
#include <paths.h> |
| ... | ... | |
| 107 | 107 |
int type; |
| 108 | 108 |
unsigned int lseek_err_cnt; |
| 109 | 109 |
int open_flags; |
| 110 |
void *aio_ctx; |
|
| 110 | 111 |
#if defined(__linux__) |
| 111 | 112 |
/* linux floppy specific */ |
| 112 | 113 |
int64_t fd_open_time; |
| ... | ... | |
| 117 | 118 |
uint8_t* aligned_buf; |
| 118 | 119 |
} BDRVRawState; |
| 119 | 120 |
|
| 120 |
static int posix_aio_init(void); |
|
| 121 |
|
|
| 122 | 121 |
static int fd_open(BlockDriverState *bs); |
| 123 | 122 |
static int64_t raw_getlength(BlockDriverState *bs); |
| 124 | 123 |
|
| ... | ... | |
| 132 | 131 |
BDRVRawState *s = bs->opaque; |
| 133 | 132 |
int fd, ret; |
| 134 | 133 |
|
| 135 |
posix_aio_init(); |
|
| 136 |
|
|
| 137 | 134 |
s->lseek_err_cnt = 0; |
| 138 | 135 |
|
| 139 | 136 |
s->open_flags = open_flags | O_BINARY; |
| ... | ... | |
| 165 | 162 |
if ((bdrv_flags & BDRV_O_NOCACHE)) {
|
| 166 | 163 |
s->aligned_buf = qemu_blockalign(bs, ALIGNED_BUFFER_SIZE); |
| 167 | 164 |
if (s->aligned_buf == NULL) {
|
| 168 |
ret = -errno; |
|
| 169 |
close(fd); |
|
| 170 |
return ret; |
|
| 165 |
goto out_close; |
|
| 171 | 166 |
} |
| 172 | 167 |
} |
| 168 |
|
|
| 169 |
s->aio_ctx = paio_init(); |
|
| 170 |
if (!s->aio_ctx) {
|
|
| 171 |
goto out_free_buf; |
|
| 172 |
} |
|
| 173 |
|
|
| 173 | 174 |
return 0; |
| 175 |
|
|
| 176 |
out_free_buf: |
|
| 177 |
qemu_vfree(s->aligned_buf); |
|
| 178 |
out_close: |
|
| 179 |
close(fd); |
|
| 180 |
return -errno; |
|
| 174 | 181 |
} |
| 175 | 182 |
|
| 176 | 183 |
static int raw_open(BlockDriverState *bs, const char *filename, int flags) |
| ... | ... | |
| 487 | 494 |
return ret; |
| 488 | 495 |
} |
| 489 | 496 |
|
| 490 |
/***********************************************************/ |
|
| 491 |
/* Unix AIO using POSIX AIO */ |
|
| 492 |
|
|
| 493 |
typedef struct RawAIOCB {
|
|
| 494 |
BlockDriverAIOCB common; |
|
| 495 |
struct qemu_paiocb aiocb; |
|
| 496 |
struct RawAIOCB *next; |
|
| 497 |
int ret; |
|
| 498 |
} RawAIOCB; |
|
| 499 |
|
|
| 500 |
typedef struct PosixAioState |
|
| 501 |
{
|
|
| 502 |
int rfd, wfd; |
|
| 503 |
RawAIOCB *first_aio; |
|
| 504 |
} PosixAioState; |
|
| 505 |
|
|
| 506 |
static void posix_aio_read(void *opaque) |
|
| 507 |
{
|
|
| 508 |
PosixAioState *s = opaque; |
|
| 509 |
RawAIOCB *acb, **pacb; |
|
| 510 |
int ret; |
|
| 511 |
ssize_t len; |
|
| 512 |
|
|
| 513 |
/* read all bytes from signal pipe */ |
|
| 514 |
for (;;) {
|
|
| 515 |
char bytes[16]; |
|
| 516 |
|
|
| 517 |
len = read(s->rfd, bytes, sizeof(bytes)); |
|
| 518 |
if (len == -1 && errno == EINTR) |
|
| 519 |
continue; /* try again */ |
|
| 520 |
if (len == sizeof(bytes)) |
|
| 521 |
continue; /* more to read */ |
|
| 522 |
break; |
|
| 523 |
} |
|
| 524 |
|
|
| 525 |
for(;;) {
|
|
| 526 |
pacb = &s->first_aio; |
|
| 527 |
for(;;) {
|
|
| 528 |
acb = *pacb; |
|
| 529 |
if (!acb) |
|
| 530 |
goto the_end; |
|
| 531 |
ret = qemu_paio_error(&acb->aiocb); |
|
| 532 |
if (ret == ECANCELED) {
|
|
| 533 |
/* remove the request */ |
|
| 534 |
*pacb = acb->next; |
|
| 535 |
qemu_aio_release(acb); |
|
| 536 |
} else if (ret != EINPROGRESS) {
|
|
| 537 |
/* end of aio */ |
|
| 538 |
if (ret == 0) {
|
|
| 539 |
ret = qemu_paio_return(&acb->aiocb); |
|
| 540 |
if (ret == acb->aiocb.aio_nbytes) |
|
| 541 |
ret = 0; |
|
| 542 |
else |
|
| 543 |
ret = -EINVAL; |
|
| 544 |
} else {
|
|
| 545 |
ret = -ret; |
|
| 546 |
} |
|
| 547 |
/* remove the request */ |
|
| 548 |
*pacb = acb->next; |
|
| 549 |
/* call the callback */ |
|
| 550 |
acb->common.cb(acb->common.opaque, ret); |
|
| 551 |
qemu_aio_release(acb); |
|
| 552 |
break; |
|
| 553 |
} else {
|
|
| 554 |
pacb = &acb->next; |
|
| 555 |
} |
|
| 556 |
} |
|
| 557 |
} |
|
| 558 |
the_end: ; |
|
| 559 |
} |
|
| 560 |
|
|
| 561 |
static int posix_aio_flush(void *opaque) |
|
| 562 |
{
|
|
| 563 |
PosixAioState *s = opaque; |
|
| 564 |
return !!s->first_aio; |
|
| 565 |
} |
|
| 566 |
|
|
| 567 |
static PosixAioState *posix_aio_state; |
|
| 568 |
|
|
| 569 |
static void aio_signal_handler(int signum) |
|
| 570 |
{
|
|
| 571 |
if (posix_aio_state) {
|
|
| 572 |
char byte = 0; |
|
| 573 |
|
|
| 574 |
write(posix_aio_state->wfd, &byte, sizeof(byte)); |
|
| 575 |
} |
|
| 576 |
|
|
| 577 |
qemu_service_io(); |
|
| 578 |
} |
|
| 579 |
|
|
| 580 |
static int posix_aio_init(void) |
|
| 497 |
/* |
|
| 498 |
* Check if all memory in this vector is sector aligned. |
|
| 499 |
*/ |
|
| 500 |
static int qiov_is_aligned(QEMUIOVector *qiov) |
|
| 581 | 501 |
{
|
| 582 |
struct sigaction act; |
|
| 583 |
PosixAioState *s; |
|
| 584 |
int fds[2]; |
|
| 585 |
struct qemu_paioinit ai; |
|
| 586 |
|
|
| 587 |
if (posix_aio_state) |
|
| 588 |
return 0; |
|
| 589 |
|
|
| 590 |
s = qemu_malloc(sizeof(PosixAioState)); |
|
| 591 |
|
|
| 592 |
sigfillset(&act.sa_mask); |
|
| 593 |
act.sa_flags = 0; /* do not restart syscalls to interrupt select() */ |
|
| 594 |
act.sa_handler = aio_signal_handler; |
|
| 595 |
sigaction(SIGUSR2, &act, NULL); |
|
| 596 |
|
|
| 597 |
s->first_aio = NULL; |
|
| 598 |
if (pipe(fds) == -1) {
|
|
| 599 |
fprintf(stderr, "failed to create pipe\n"); |
|
| 600 |
return -errno; |
|
| 601 |
} |
|
| 602 |
|
|
| 603 |
s->rfd = fds[0]; |
|
| 604 |
s->wfd = fds[1]; |
|
| 605 |
|
|
| 606 |
fcntl(s->rfd, F_SETFL, O_NONBLOCK); |
|
| 607 |
fcntl(s->wfd, F_SETFL, O_NONBLOCK); |
|
| 608 |
|
|
| 609 |
qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s); |
|
| 610 |
|
|
| 611 |
memset(&ai, 0, sizeof(ai)); |
|
| 612 |
ai.aio_threads = 64; |
|
| 613 |
ai.aio_num = 64; |
|
| 614 |
qemu_paio_init(&ai); |
|
| 615 |
|
|
| 616 |
posix_aio_state = s; |
|
| 617 |
|
|
| 618 |
return 0; |
|
| 619 |
} |
|
| 502 |
int i; |
|
| 620 | 503 |
|
| 621 |
static void raw_aio_remove(RawAIOCB *acb) |
|
| 622 |
{
|
|
| 623 |
RawAIOCB **pacb; |
|
| 624 |
|
|
| 625 |
/* remove the callback from the queue */ |
|
| 626 |
pacb = &posix_aio_state->first_aio; |
|
| 627 |
for(;;) {
|
|
| 628 |
if (*pacb == NULL) {
|
|
| 629 |
fprintf(stderr, "raw_aio_remove: aio request not found!\n"); |
|
| 630 |
break; |
|
| 631 |
} else if (*pacb == acb) {
|
|
| 632 |
*pacb = acb->next; |
|
| 633 |
qemu_aio_release(acb); |
|
| 634 |
break; |
|
| 504 |
for (i = 0; i < qiov->niov; i++) {
|
|
| 505 |
if ((uintptr_t) qiov->iov[i].iov_base % 512) {
|
|
| 506 |
return 0; |
|
| 635 | 507 |
} |
| 636 |
pacb = &(*pacb)->next; |
|
| 637 | 508 |
} |
| 638 |
} |
|
| 639 |
|
|
| 640 |
static void raw_aio_cancel(BlockDriverAIOCB *blockacb) |
|
| 641 |
{
|
|
| 642 |
int ret; |
|
| 643 |
RawAIOCB *acb = (RawAIOCB *)blockacb; |
|
| 644 | 509 |
|
| 645 |
ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb); |
|
| 646 |
if (ret == QEMU_PAIO_NOTCANCELED) {
|
|
| 647 |
/* fail safe: if the aio could not be canceled, we wait for |
|
| 648 |
it */ |
|
| 649 |
while (qemu_paio_error(&acb->aiocb) == EINPROGRESS); |
|
| 650 |
} |
|
| 651 |
|
|
| 652 |
raw_aio_remove(acb); |
|
| 510 |
return 1; |
|
| 653 | 511 |
} |
| 654 | 512 |
|
| 655 |
static AIOPool raw_aio_pool = {
|
|
| 656 |
.aiocb_size = sizeof(RawAIOCB), |
|
| 657 |
.cancel = raw_aio_cancel, |
|
| 658 |
}; |
|
| 659 |
|
|
| 660 |
static RawAIOCB *raw_aio_setup(BlockDriverState *bs, int64_t sector_num, |
|
| 661 |
QEMUIOVector *qiov, int nb_sectors, |
|
| 662 |
BlockDriverCompletionFunc *cb, void *opaque) |
|
| 513 |
static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs, |
|
| 514 |
int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, |
|
| 515 |
BlockDriverCompletionFunc *cb, void *opaque, int type) |
|
| 663 | 516 |
{
|
| 664 | 517 |
BDRVRawState *s = bs->opaque; |
| 665 |
RawAIOCB *acb; |
|
| 666 | 518 |
|
| 667 | 519 |
if (fd_open(bs) < 0) |
| 668 | 520 |
return NULL; |
| 669 | 521 |
|
| 670 |
acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque); |
|
| 671 |
if (!acb) |
|
| 672 |
return NULL; |
|
| 673 |
acb->aiocb.aio_fildes = s->fd; |
|
| 674 |
acb->aiocb.ev_signo = SIGUSR2; |
|
| 675 |
acb->aiocb.aio_iov = qiov->iov; |
|
| 676 |
acb->aiocb.aio_niov = qiov->niov; |
|
| 677 |
acb->aiocb.aio_nbytes = nb_sectors * 512; |
|
| 678 |
acb->aiocb.aio_offset = sector_num * 512; |
|
| 679 |
acb->aiocb.aio_flags = 0; |
|
| 680 |
|
|
| 681 | 522 |
/* |
| 682 | 523 |
* If O_DIRECT is used the buffer needs to be aligned on a sector |
| 683 |
* boundary. Tell the low level code to ensure that in case it's
|
|
| 684 |
* not done yet.
|
|
| 524 |
* boundary. Check if this is the case or telll the low-level
|
|
| 525 |
* driver that it needs to copy the buffer.
|
|
| 685 | 526 |
*/ |
| 686 |
if (s->aligned_buf) |
|
| 687 |
acb->aiocb.aio_flags |= QEMU_AIO_SECTOR_ALIGNED; |
|
| 527 |
if (s->aligned_buf && !qiov_is_aligned(qiov)) {
|
|
| 528 |
type |= QEMU_AIO_MISALIGNED; |
|
| 529 |
} |
|
| 688 | 530 |
|
| 689 |
acb->next = posix_aio_state->first_aio; |
|
| 690 |
posix_aio_state->first_aio = acb; |
|
| 691 |
return acb; |
|
| 531 |
return paio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov, nb_sectors, |
|
| 532 |
cb, opaque, type); |
|
| 692 | 533 |
} |
| 693 | 534 |
|
| 694 | 535 |
static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs, |
| 695 | 536 |
int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, |
| 696 | 537 |
BlockDriverCompletionFunc *cb, void *opaque) |
| 697 | 538 |
{
|
| 698 |
RawAIOCB *acb; |
|
| 699 |
|
|
| 700 |
acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque); |
|
| 701 |
if (!acb) |
|
| 702 |
return NULL; |
|
| 703 |
if (qemu_paio_read(&acb->aiocb) < 0) {
|
|
| 704 |
raw_aio_remove(acb); |
|
| 705 |
return NULL; |
|
| 706 |
} |
|
| 707 |
return &acb->common; |
|
| 539 |
return raw_aio_submit(bs, sector_num, qiov, nb_sectors, |
|
| 540 |
cb, opaque, QEMU_AIO_READ); |
|
| 708 | 541 |
} |
| 709 | 542 |
|
| 710 | 543 |
static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs, |
| 711 | 544 |
int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, |
| 712 | 545 |
BlockDriverCompletionFunc *cb, void *opaque) |
| 713 | 546 |
{
|
| 714 |
RawAIOCB *acb; |
|
| 715 |
|
|
| 716 |
acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque); |
|
| 717 |
if (!acb) |
|
| 718 |
return NULL; |
|
| 719 |
if (qemu_paio_write(&acb->aiocb) < 0) {
|
|
| 720 |
raw_aio_remove(acb); |
|
| 721 |
return NULL; |
|
| 722 |
} |
|
| 723 |
return &acb->common; |
|
| 547 |
return raw_aio_submit(bs, sector_num, qiov, nb_sectors, |
|
| 548 |
cb, opaque, QEMU_AIO_WRITE); |
|
| 724 | 549 |
} |
| 725 | 550 |
|
| 726 | 551 |
static void raw_close(BlockDriverState *bs) |
| ... | ... | |
| 1085 | 910 |
BlockDriverCompletionFunc *cb, void *opaque) |
| 1086 | 911 |
{
|
| 1087 | 912 |
BDRVRawState *s = bs->opaque; |
| 1088 |
RawAIOCB *acb; |
|
| 1089 | 913 |
|
| 1090 | 914 |
if (fd_open(bs) < 0) |
| 1091 | 915 |
return NULL; |
| 1092 |
|
|
| 1093 |
acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque); |
|
| 1094 |
if (!acb) |
|
| 1095 |
return NULL; |
|
| 1096 |
acb->aiocb.aio_fildes = s->fd; |
|
| 1097 |
acb->aiocb.ev_signo = SIGUSR2; |
|
| 1098 |
acb->aiocb.aio_offset = 0; |
|
| 1099 |
acb->aiocb.aio_flags = 0; |
|
| 1100 |
|
|
| 1101 |
acb->next = posix_aio_state->first_aio; |
|
| 1102 |
posix_aio_state->first_aio = acb; |
|
| 1103 |
|
|
| 1104 |
acb->aiocb.aio_ioctl_buf = buf; |
|
| 1105 |
acb->aiocb.aio_ioctl_cmd = req; |
|
| 1106 |
if (qemu_paio_ioctl(&acb->aiocb) < 0) {
|
|
| 1107 |
raw_aio_remove(acb); |
|
| 1108 |
return NULL; |
|
| 1109 |
} |
|
| 1110 |
|
|
| 1111 |
return &acb->common; |
|
| 916 |
return paio_ioctl(bs, s->fd, req, buf, cb, opaque); |
|
| 1112 | 917 |
} |
| 1113 | 918 |
|
| 1114 | 919 |
#elif defined(__FreeBSD__) |
| ... | ... | |
| 1189 | 994 |
BDRVRawState *s = bs->opaque; |
| 1190 | 995 |
int ret; |
| 1191 | 996 |
|
| 1192 |
posix_aio_init(); |
|
| 1193 |
|
|
| 1194 | 997 |
s->type = FTYPE_FD; |
| 1195 | 998 |
|
| 1196 | 999 |
/* open will not fail even if no floppy is inserted, so add O_NONBLOCK */ |
Also available in: Unified diff