2 * Simulate the Posix AIO using mmap/fork
4 * Copyright (C) Volker Lendecke 2008
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
28 static int mmap_area_destructor(struct mmap_area
*area
)
30 munmap((void *)area
->ptr
, area
->size
);
34 static struct mmap_area
*mmap_area_init(TALLOC_CTX
*mem_ctx
, size_t size
)
36 struct mmap_area
*result
;
39 result
= talloc(mem_ctx
, struct mmap_area
);
41 DEBUG(0, ("talloc failed\n"));
45 fd
= open("/dev/zero", O_RDWR
);
47 DEBUG(3, ("open(\"/dev/zero\") failed: %s\n",
52 result
->ptr
= mmap(NULL
, size
, PROT_READ
|PROT_WRITE
,
53 MAP_SHARED
|MAP_FILE
, fd
, 0);
54 if (result
->ptr
== MAP_FAILED
) {
55 DEBUG(1, ("mmap failed: %s\n", strerror(errno
)));
60 talloc_set_destructor(result
, mmap_area_destructor
);
80 struct aio_child_list
;
83 struct aio_child
*prev
, *next
;
84 struct aio_child_list
*list
;
85 SMB_STRUCT_AIOCB
*aiocb
;
88 struct fd_event
*sock_event
;
90 struct mmap_area
*map
; /* ==NULL means write request */
91 bool dont_delete
; /* Marked as in use since last cleanup */
96 struct aio_child_list
{
97 struct aio_child
*children
;
98 struct timed_event
*cleanup_event
;
101 static void free_aio_children(void **p
)
106 static ssize_t
read_fd(int fd
, void *ptr
, size_t nbytes
, int *recvfd
)
111 #ifndef HAVE_MSGHDR_MSG_CONTROL
115 #ifdef HAVE_MSGHDR_MSG_CONTROL
118 char control
[CMSG_SPACE(sizeof(int))];
120 struct cmsghdr
*cmptr
;
122 msg
.msg_control
= control_un
.control
;
123 msg
.msg_controllen
= sizeof(control_un
.control
);
125 #if HAVE_MSGHDR_MSG_ACCTRIGHTS
126 msg
.msg_accrights
= (caddr_t
) &newfd
;
127 msg
.msg_accrightslen
= sizeof(int);
129 #error Can not pass file descriptors
136 iov
[0].iov_base
= ptr
;
137 iov
[0].iov_len
= nbytes
;
141 if ( (n
= recvmsg(fd
, &msg
, 0)) <= 0) {
145 #ifdef HAVE_MSGHDR_MSG_CONTROL
146 if ((cmptr
= CMSG_FIRSTHDR(&msg
)) != NULL
147 && cmptr
->cmsg_len
== CMSG_LEN(sizeof(int))) {
148 if (cmptr
->cmsg_level
!= SOL_SOCKET
) {
149 DEBUG(10, ("control level != SOL_SOCKET"));
153 if (cmptr
->cmsg_type
!= SCM_RIGHTS
) {
154 DEBUG(10, ("control type != SCM_RIGHTS"));
158 *recvfd
= *((int *) CMSG_DATA(cmptr
));
160 *recvfd
= -1; /* descriptor was not passed */
163 if (msg
.msg_accrightslen
== sizeof(int)) {
167 *recvfd
= -1; /* descriptor was not passed */
174 static ssize_t
write_fd(int fd
, void *ptr
, size_t nbytes
, int sendfd
)
179 #ifdef HAVE_MSGHDR_MSG_CONTROL
182 char control
[CMSG_SPACE(sizeof(int))];
184 struct cmsghdr
*cmptr
;
187 ZERO_STRUCT(control_un
);
189 msg
.msg_control
= control_un
.control
;
190 msg
.msg_controllen
= sizeof(control_un
.control
);
192 cmptr
= CMSG_FIRSTHDR(&msg
);
193 cmptr
->cmsg_len
= CMSG_LEN(sizeof(int));
194 cmptr
->cmsg_level
= SOL_SOCKET
;
195 cmptr
->cmsg_type
= SCM_RIGHTS
;
196 *((int *) CMSG_DATA(cmptr
)) = sendfd
;
199 msg
.msg_accrights
= (caddr_t
) &sendfd
;
200 msg
.msg_accrightslen
= sizeof(int);
207 iov
[0].iov_base
= ptr
;
208 iov
[0].iov_len
= nbytes
;
212 return (sendmsg(fd
, &msg
, 0));
215 static void aio_child_cleanup(struct event_context
*event_ctx
,
216 struct timed_event
*te
,
217 const struct timeval
*now
,
220 struct aio_child_list
*list
= talloc_get_type_abort(
221 private_data
, struct aio_child_list
);
222 struct aio_child
*child
, *next
;
224 TALLOC_FREE(list
->cleanup_event
);
226 for (child
= list
->children
; child
!= NULL
; child
= next
) {
229 if (child
->aiocb
!= NULL
) {
230 DEBUG(10, ("child %d currently active\n",
235 if (child
->dont_delete
) {
236 DEBUG(10, ("Child %d was active since last cleanup\n",
238 child
->dont_delete
= false;
242 DEBUG(10, ("Child %d idle for more than 30 seconds, "
243 "deleting\n", (int)child
->pid
));
248 if (list
->children
!= NULL
) {
250 * Re-schedule the next cleanup round
252 list
->cleanup_event
= event_add_timed(smbd_event_context(), list
,
253 timeval_add(now
, 30, 0),
255 aio_child_cleanup
, list
);
260 static struct aio_child_list
*init_aio_children(struct vfs_handle_struct
*handle
)
262 struct aio_child_list
*data
= NULL
;
264 if (SMB_VFS_HANDLE_TEST_DATA(handle
)) {
265 SMB_VFS_HANDLE_GET_DATA(handle
, data
, struct aio_child_list
,
270 data
= TALLOC_ZERO_P(NULL
, struct aio_child_list
);
277 * Regardless of whether the child_list had been around or not, make
278 * sure that we have a cleanup timed event. This timed event will
279 * delete itself when it finds that no children are around anymore.
282 if (data
->cleanup_event
== NULL
) {
283 data
->cleanup_event
= event_add_timed(smbd_event_context(), data
,
284 timeval_current_ofs(30, 0),
286 aio_child_cleanup
, data
);
287 if (data
->cleanup_event
== NULL
) {
293 if (!SMB_VFS_HANDLE_TEST_DATA(handle
)) {
294 SMB_VFS_HANDLE_SET_DATA(handle
, data
, free_aio_children
,
295 struct aio_child_list
, return False
);
301 static void aio_child_loop(int sockfd
, struct mmap_area
*map
)
306 struct rw_cmd cmd_struct
;
307 struct rw_ret ret_struct
;
309 ret
= read_fd(sockfd
, &cmd_struct
, sizeof(cmd_struct
), &fd
);
310 if (ret
!= sizeof(cmd_struct
)) {
311 DEBUG(10, ("read_fd returned %d: %s\n", (int)ret
,
316 DEBUG(10, ("aio_child_loop: %s %d bytes at %d from fd %d\n",
317 cmd_struct
.read_cmd
? "read" : "write",
318 (int)cmd_struct
.n
, (int)cmd_struct
.offset
, fd
));
320 #ifdef ENABLE_BUILD_FARM_HACKS
323 * In the build farm, we want erratic behaviour for
329 * use generate_random_buffer, we just forked from a
330 * common parent state
332 generate_random_buffer(&randval
, sizeof(randval
));
333 msecs
= randval
+ 20;
334 DEBUG(10, ("delaying for %u msecs\n", msecs
));
340 ZERO_STRUCT(ret_struct
);
342 if (cmd_struct
.read_cmd
) {
343 ret_struct
.size
= sys_pread(
344 fd
, (void *)map
->ptr
, cmd_struct
.n
,
348 ret_struct
.size
= sys_pwrite(
349 fd
, (void *)map
->ptr
, cmd_struct
.n
,
353 DEBUG(10, ("aio_child_loop: syscall returned %d\n",
354 (int)ret_struct
.size
));
356 if (ret_struct
.size
== -1) {
357 ret_struct
.ret_errno
= errno
;
360 ret
= write_data(sockfd
, (char *)&ret_struct
,
362 if (ret
!= sizeof(ret_struct
)) {
363 DEBUG(10, ("could not write ret_struct: %s\n",
372 static void handle_aio_completion(struct event_context
*event_ctx
,
373 struct fd_event
*event
, uint16 flags
,
376 struct aio_child
*child
= (struct aio_child
*)p
;
379 DEBUG(10, ("handle_aio_completion called with flags=%d\n", flags
));
381 if ((flags
& EVENT_FD_READ
) == 0) {
385 if (!NT_STATUS_IS_OK(read_data(child
->sockfd
,
386 (char *)&child
->retval
,
387 sizeof(child
->retval
)))) {
388 DEBUG(0, ("aio child %d died\n", (int)child
->pid
));
389 child
->retval
.size
= -1;
390 child
->retval
.ret_errno
= EIO
;
393 if (child
->cancelled
) {
395 child
->cancelled
= false;
399 if (child
->read_cmd
&& (child
->retval
.size
> 0)) {
400 SMB_ASSERT(child
->retval
.size
<= child
->aiocb
->aio_nbytes
);
401 memcpy((void *)child
->aiocb
->aio_buf
, (void *)child
->map
->ptr
,
405 mid
= child
->aiocb
->aio_sigevent
.sigev_value
.sival_int
;
407 DEBUG(10, ("mid %d finished\n", (int)mid
));
409 aio_request_done(mid
);
413 static int aio_child_destructor(struct aio_child
*child
)
415 SMB_ASSERT((child
->aiocb
== NULL
) || child
->cancelled
);
416 close(child
->sockfd
);
417 DLIST_REMOVE(child
->list
->children
, child
);
421 static NTSTATUS
create_aio_child(struct aio_child_list
*children
,
423 struct aio_child
**presult
)
425 struct aio_child
*result
;
429 fdpair
[0] = fdpair
[1] = -1;
431 result
= TALLOC_ZERO_P(children
, struct aio_child
);
432 NT_STATUS_HAVE_NO_MEMORY(result
);
434 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, fdpair
) == -1) {
435 status
= map_nt_error_from_unix(errno
);
436 DEBUG(10, ("socketpair() failed: %s\n", strerror(errno
)));
441 DEBUG(10, ("fdpair = %d/%d\n", fdpair
[0], fdpair
[1]));
443 result
->map
= mmap_area_init(result
, map_size
);
444 if (result
->map
== NULL
) {
445 status
= map_nt_error_from_unix(errno
);
446 DEBUG(0, ("Could not create mmap area\n"));
450 result
->pid
= sys_fork();
451 if (result
->pid
== -1) {
452 status
= map_nt_error_from_unix(errno
);
453 DEBUG(0, ("fork failed: %s\n", strerror(errno
)));
457 if (result
->pid
== 0) {
459 result
->sockfd
= fdpair
[1];
460 aio_child_loop(result
->sockfd
, result
->map
);
463 DEBUG(10, ("Child %d created\n", result
->pid
));
465 result
->sockfd
= fdpair
[0];
468 result
->sock_event
= event_add_fd(smbd_event_context(), result
,
469 result
->sockfd
, EVENT_FD_READ
,
470 handle_aio_completion
,
472 if (result
->sock_event
== NULL
) {
473 status
= NT_STATUS_NO_MEMORY
;
474 DEBUG(0, ("event_add_fd failed\n"));
478 result
->list
= children
;
479 DLIST_ADD(children
->children
, result
);
481 talloc_set_destructor(result
, aio_child_destructor
);
488 if (fdpair
[0] != -1) close(fdpair
[0]);
489 if (fdpair
[1] != -1) close(fdpair
[1]);
495 static NTSTATUS
get_idle_child(struct vfs_handle_struct
*handle
,
496 struct aio_child
**pchild
)
498 struct aio_child_list
*children
;
499 struct aio_child
*child
;
502 children
= init_aio_children(handle
);
503 if (children
== NULL
) {
504 return NT_STATUS_NO_MEMORY
;
507 for (child
= children
->children
; child
!= NULL
; child
= child
->next
) {
508 if (child
->aiocb
== NULL
) {
515 DEBUG(10, ("no idle child found, creating new one\n"));
517 status
= create_aio_child(children
, 128*1024, &child
);
518 if (!NT_STATUS_IS_OK(status
)) {
519 DEBUG(10, ("create_aio_child failed: %s\n",
525 child
->dont_delete
= true;
531 static int aio_fork_read(struct vfs_handle_struct
*handle
,
532 struct files_struct
*fsp
, SMB_STRUCT_AIOCB
*aiocb
)
534 struct aio_child
*child
;
539 if (aiocb
->aio_nbytes
> 128*1024) {
540 /* TODO: support variable buffers */
545 status
= get_idle_child(handle
, &child
);
546 if (!NT_STATUS_IS_OK(status
)) {
547 DEBUG(10, ("Could not get an idle child\n"));
551 child
->read_cmd
= true;
552 child
->aiocb
= aiocb
;
553 child
->retval
.ret_errno
= EINPROGRESS
;
556 cmd
.n
= aiocb
->aio_nbytes
;
557 cmd
.offset
= aiocb
->aio_offset
;
558 cmd
.read_cmd
= child
->read_cmd
;
560 DEBUG(10, ("sending fd %d to child %d\n", fsp
->fh
->fd
,
563 ret
= write_fd(child
->sockfd
, &cmd
, sizeof(cmd
), fsp
->fh
->fd
);
565 DEBUG(10, ("write_fd failed: %s\n", strerror(errno
)));
572 static int aio_fork_write(struct vfs_handle_struct
*handle
,
573 struct files_struct
*fsp
, SMB_STRUCT_AIOCB
*aiocb
)
575 struct aio_child
*child
;
580 if (aiocb
->aio_nbytes
> 128*1024) {
581 /* TODO: support variable buffers */
586 status
= get_idle_child(handle
, &child
);
587 if (!NT_STATUS_IS_OK(status
)) {
588 DEBUG(10, ("Could not get an idle child\n"));
592 child
->read_cmd
= false;
593 child
->aiocb
= aiocb
;
594 child
->retval
.ret_errno
= EINPROGRESS
;
596 memcpy((void *)child
->map
->ptr
, (void *)aiocb
->aio_buf
,
600 cmd
.n
= aiocb
->aio_nbytes
;
601 cmd
.offset
= aiocb
->aio_offset
;
602 cmd
.read_cmd
= child
->read_cmd
;
604 DEBUG(10, ("sending fd %d to child %d\n", fsp
->fh
->fd
,
607 ret
= write_fd(child
->sockfd
, &cmd
, sizeof(cmd
), fsp
->fh
->fd
);
609 DEBUG(10, ("write_fd failed: %s\n", strerror(errno
)));
616 static struct aio_child
*aio_fork_find_child(struct vfs_handle_struct
*handle
,
617 SMB_STRUCT_AIOCB
*aiocb
)
619 struct aio_child_list
*children
;
620 struct aio_child
*child
;
622 children
= init_aio_children(handle
);
623 if (children
== NULL
) {
627 for (child
= children
->children
; child
!= NULL
; child
= child
->next
) {
628 if (child
->aiocb
== aiocb
) {
636 static ssize_t
aio_fork_return_fn(struct vfs_handle_struct
*handle
,
637 struct files_struct
*fsp
,
638 SMB_STRUCT_AIOCB
*aiocb
)
640 struct aio_child
*child
= aio_fork_find_child(handle
, aiocb
);
644 DEBUG(0, ("returning EINVAL\n"));
650 if (child
->retval
.size
== -1) {
651 errno
= child
->retval
.ret_errno
;
654 return child
->retval
.size
;
657 static int aio_fork_cancel(struct vfs_handle_struct
*handle
,
658 struct files_struct
*fsp
,
659 SMB_STRUCT_AIOCB
*aiocb
)
661 struct aio_child_list
*children
;
662 struct aio_child
*child
;
664 children
= init_aio_children(handle
);
665 if (children
== NULL
) {
670 for (child
= children
->children
; child
!= NULL
; child
= child
->next
) {
671 if (child
->aiocb
== NULL
) {
674 if (child
->aiocb
->aio_fildes
!= fsp
->fh
->fd
) {
677 if ((aiocb
!= NULL
) && (child
->aiocb
!= aiocb
)) {
682 * We let the child do its job, but we discard the result when
686 child
->cancelled
= true;
692 static int aio_fork_error_fn(struct vfs_handle_struct
*handle
,
693 struct files_struct
*fsp
,
694 SMB_STRUCT_AIOCB
*aiocb
)
696 struct aio_child
*child
= aio_fork_find_child(handle
, aiocb
);
703 return child
->retval
.ret_errno
;
706 /* VFS operations structure */
708 static vfs_op_tuple aio_fork_ops
[] = {
709 {SMB_VFS_OP(aio_fork_read
), SMB_VFS_OP_AIO_READ
,
710 SMB_VFS_LAYER_TRANSPARENT
},
711 {SMB_VFS_OP(aio_fork_write
), SMB_VFS_OP_AIO_WRITE
,
712 SMB_VFS_LAYER_TRANSPARENT
},
713 {SMB_VFS_OP(aio_fork_return_fn
), SMB_VFS_OP_AIO_RETURN
,
714 SMB_VFS_LAYER_TRANSPARENT
},
715 {SMB_VFS_OP(aio_fork_cancel
), SMB_VFS_OP_AIO_CANCEL
,
716 SMB_VFS_LAYER_TRANSPARENT
},
717 {SMB_VFS_OP(aio_fork_error_fn
), SMB_VFS_OP_AIO_ERROR
,
718 SMB_VFS_LAYER_TRANSPARENT
},
719 {SMB_VFS_OP(NULL
), SMB_VFS_OP_NOOP
,
723 NTSTATUS
vfs_aio_fork_init(void);
724 NTSTATUS
vfs_aio_fork_init(void)
726 return smb_register_vfs(SMB_VFS_INTERFACE_VERSION
,
727 "aio_fork", aio_fork_ops
);