2 * Simulate the Posix AIO using mmap/fork
4 * Copyright (C) Volker Lendecke 2008
5 * Copyright (C) Jeremy Allison 2010
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #include "system/filesys.h"
24 #include "system/shmem.h"
25 #include "smbd/smbd.h"
26 #include "smbd/globals.h"
27 #include "lib/async_req/async_sock.h"
28 #include "lib/util/tevent_unix.h"
29 #include "lib/util/sys_rw.h"
30 #include "lib/util/sys_rw_data.h"
31 #include "lib/util/msghdr.h"
32 #include "smbprofile.h"
34 #if !defined(HAVE_STRUCT_MSGHDR_MSG_CONTROL) && !defined(HAVE_STRUCT_MSGHDR_MSG_ACCRIGHTS)
35 # error Can not pass file descriptors
44 struct aio_fork_config
{
45 bool erratic_testing_mode
;
53 static int mmap_area_destructor(struct mmap_area
*area
)
55 munmap(discard_const(area
->ptr
), area
->size
);
59 static struct mmap_area
*mmap_area_init(TALLOC_CTX
*mem_ctx
, size_t size
)
61 struct mmap_area
*result
;
64 result
= talloc(mem_ctx
, struct mmap_area
);
66 DEBUG(0, ("talloc failed\n"));
70 fd
= open("/dev/zero", O_RDWR
);
72 DEBUG(3, ("open(\"/dev/zero\") failed: %s\n",
77 result
->ptr
= mmap(NULL
, size
, PROT_READ
|PROT_WRITE
,
78 MAP_SHARED
|MAP_FILE
, fd
, 0);
80 if (result
->ptr
== MAP_FAILED
) {
81 DEBUG(1, ("mmap failed: %s\n", strerror(errno
)));
86 talloc_set_destructor(result
, mmap_area_destructor
);
101 static const char *cmd_type_str(enum cmd_type cmd
)
116 result
= "<UNKNOWN>";
126 bool erratic_testing_mode
;
135 struct aio_child_list
;
138 struct aio_child
*prev
, *next
;
139 struct aio_child_list
*list
;
142 struct mmap_area
*map
;
143 bool dont_delete
; /* Marked as in use since last cleanup */
147 struct aio_child_list
{
148 struct aio_child
*children
;
149 struct tevent_timer
*cleanup_event
;
152 static void free_aio_children(void **p
)
157 static ssize_t
read_fd(int fd
, void *ptr
, size_t nbytes
, int *recvfd
)
160 struct msghdr msg
= { .msg_iov
= iov
, .msg_iovlen
= 1 };
162 size_t bufsize
= msghdr_prep_recv_fds(NULL
, NULL
, 0, 1);
163 uint8_t buf
[bufsize
];
165 msghdr_prep_recv_fds(&msg
, buf
, bufsize
, 1);
167 iov
[0].iov_base
= (void *)ptr
;
168 iov
[0].iov_len
= nbytes
;
171 n
= recvmsg(fd
, &msg
, 0);
172 } while ((n
== -1) && (errno
== EINTR
));
179 size_t num_fds
= msghdr_extract_fds(&msg
, NULL
, 0);
182 msghdr_extract_fds(&msg
, fds
, num_fds
);
187 for (i
=0; i
<num_fds
; i
++) {
201 static ssize_t
write_fd(int fd
, void *ptr
, size_t nbytes
, int sendfd
)
203 struct msghdr msg
= {0};
204 size_t bufsize
= msghdr_prep_fds(NULL
, NULL
, 0, &sendfd
, 1);
205 uint8_t buf
[bufsize
];
209 msghdr_prep_fds(&msg
, buf
, bufsize
, &sendfd
, 1);
211 iov
.iov_base
= (void *)ptr
;
212 iov
.iov_len
= nbytes
;
217 sent
= sendmsg(fd
, &msg
, 0);
218 } while ((sent
== -1) && (errno
== EINTR
));
223 static void aio_child_cleanup(struct tevent_context
*event_ctx
,
224 struct tevent_timer
*te
,
228 struct aio_child_list
*list
= talloc_get_type_abort(
229 private_data
, struct aio_child_list
);
230 struct aio_child
*child
, *next
;
232 TALLOC_FREE(list
->cleanup_event
);
234 for (child
= list
->children
; child
!= NULL
; child
= next
) {
238 DEBUG(10, ("child %d currently active\n",
243 if (child
->dont_delete
) {
244 DEBUG(10, ("Child %d was active since last cleanup\n",
246 child
->dont_delete
= false;
250 DEBUG(10, ("Child %d idle for more than 30 seconds, "
251 "deleting\n", (int)child
->pid
));
257 if (list
->children
!= NULL
) {
259 * Re-schedule the next cleanup round
261 list
->cleanup_event
= tevent_add_timer(server_event_context(), list
,
262 timeval_add(&now
, 30, 0),
263 aio_child_cleanup
, list
);
268 static struct aio_child_list
*init_aio_children(struct vfs_handle_struct
*handle
)
270 struct aio_child_list
*data
= NULL
;
272 if (SMB_VFS_HANDLE_TEST_DATA(handle
)) {
273 SMB_VFS_HANDLE_GET_DATA(handle
, data
, struct aio_child_list
,
278 data
= talloc_zero(NULL
, struct aio_child_list
);
285 * Regardless of whether the child_list had been around or not, make
286 * sure that we have a cleanup timed event. This timed event will
287 * delete itself when it finds that no children are around anymore.
290 if (data
->cleanup_event
== NULL
) {
291 data
->cleanup_event
= tevent_add_timer(server_event_context(), data
,
292 timeval_current_ofs(30, 0),
293 aio_child_cleanup
, data
);
294 if (data
->cleanup_event
== NULL
) {
300 if (!SMB_VFS_HANDLE_TEST_DATA(handle
)) {
301 SMB_VFS_HANDLE_SET_DATA(handle
, data
, free_aio_children
,
302 struct aio_child_list
, return False
);
308 static void aio_child_loop(int sockfd
, struct mmap_area
*map
)
313 struct rw_cmd cmd_struct
;
314 struct rw_ret ret_struct
;
315 struct timespec start
, end
;
317 ret
= read_fd(sockfd
, &cmd_struct
, sizeof(cmd_struct
), &fd
);
318 if (ret
!= sizeof(cmd_struct
)) {
319 DEBUG(10, ("read_fd returned %d: %s\n", (int)ret
,
324 DEBUG(10, ("aio_child_loop: %s %d bytes at %d from fd %d\n",
325 cmd_type_str(cmd_struct
.cmd
),
326 (int)cmd_struct
.n
, (int)cmd_struct
.offset
, fd
));
328 if (cmd_struct
.erratic_testing_mode
) {
330 * For developer testing, we want erratic behaviour for
336 * use generate_random_buffer, we just forked from a
337 * common parent state
339 generate_random_buffer(&randval
, sizeof(randval
));
340 msecs
= randval
+ 20;
341 DEBUG(10, ("delaying for %u msecs\n", msecs
));
345 ZERO_STRUCT(ret_struct
);
347 PROFILE_TIMESTAMP(&start
);
349 switch (cmd_struct
.cmd
) {
351 ret_struct
.size
= sys_pread(
352 fd
, discard_const(map
->ptr
), cmd_struct
.n
,
355 /* This breaks "make test" when run with aio_fork module. */
357 ret_struct
.size
= MAX(1, ret_struct
.size
* 0.9);
362 ret_struct
.size
= sys_pwrite(
363 fd
, discard_const(map
->ptr
), cmd_struct
.n
,
367 ret_struct
.size
= fsync(fd
);
370 ret_struct
.size
= -1;
374 PROFILE_TIMESTAMP(&end
);
375 ret_struct
.duration
= nsec_time_diff(&end
, &start
);
376 DEBUG(10, ("aio_child_loop: syscall returned %d\n",
377 (int)ret_struct
.size
));
379 if (ret_struct
.size
== -1) {
380 ret_struct
.ret_errno
= errno
;
384 * Close the fd before telling our parent we're done. The
385 * parent might close and re-open the file very quickly, and
386 * with system-level share modes (GPFS) we would get an
387 * unjustified SHARING_VIOLATION.
391 ret
= write_data(sockfd
, (char *)&ret_struct
,
393 if (ret
!= sizeof(ret_struct
)) {
394 DEBUG(10, ("could not write ret_struct: %s\n",
401 static int aio_child_destructor(struct aio_child
*child
)
405 SMB_ASSERT(!child
->busy
);
407 DEBUG(10, ("aio_child_destructor: removing child %d on fd %d\n",
408 (int)child
->pid
, child
->sockfd
));
411 * closing the sockfd makes the child not return from recvmsg() on RHEL
412 * 5.5 so instead force the child to exit by writing bad data to it
414 sys_write_v(child
->sockfd
, &c
, sizeof(c
));
415 close(child
->sockfd
);
416 DLIST_REMOVE(child
->list
->children
, child
);
421 * We have to close all fd's in open files, we might incorrectly hold a system
422 * level share mode on a file.
425 static struct files_struct
*close_fsp_fd(struct files_struct
*fsp
,
428 if ((fsp
->fh
!= NULL
) && (fsp
->fh
->fd
!= -1)) {
435 static int create_aio_child(struct smbd_server_connection
*sconn
,
436 struct aio_child_list
*children
,
438 struct aio_child
**presult
)
440 struct aio_child
*result
;
444 fdpair
[0] = fdpair
[1] = -1;
446 result
= talloc_zero(children
, struct aio_child
);
447 if (result
== NULL
) {
451 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, fdpair
) == -1) {
453 DEBUG(10, ("socketpair() failed: %s\n", strerror(errno
)));
457 DEBUG(10, ("fdpair = %d/%d\n", fdpair
[0], fdpair
[1]));
459 result
->map
= mmap_area_init(result
, map_size
);
460 if (result
->map
== NULL
) {
462 DEBUG(0, ("Could not create mmap area\n"));
466 result
->pid
= fork();
467 if (result
->pid
== -1) {
469 DEBUG(0, ("fork failed: %s\n", strerror(errno
)));
473 if (result
->pid
== 0) {
475 result
->sockfd
= fdpair
[1];
476 files_forall(sconn
, close_fsp_fd
, NULL
);
477 aio_child_loop(result
->sockfd
, result
->map
);
480 DEBUG(10, ("Child %d created with sockfd %d\n",
481 (int)result
->pid
, fdpair
[0]));
483 result
->sockfd
= fdpair
[0];
486 result
->list
= children
;
487 DLIST_ADD(children
->children
, result
);
489 talloc_set_destructor(result
, aio_child_destructor
);
496 if (fdpair
[0] != -1) close(fdpair
[0]);
497 if (fdpair
[1] != -1) close(fdpair
[1]);
503 static int get_idle_child(struct vfs_handle_struct
*handle
,
504 struct aio_child
**pchild
)
506 struct aio_child_list
*children
;
507 struct aio_child
*child
;
509 children
= init_aio_children(handle
);
510 if (children
== NULL
) {
514 for (child
= children
->children
; child
!= NULL
; child
= child
->next
) {
523 DEBUG(10, ("no idle child found, creating new one\n"));
525 ret
= create_aio_child(handle
->conn
->sconn
, children
,
528 DEBUG(10, ("create_aio_child failed: %s\n",
534 child
->dont_delete
= true;
541 struct aio_fork_pread_state
{
542 struct aio_child
*child
;
544 struct vfs_aio_state vfs_aio_state
;
547 static void aio_fork_pread_done(struct tevent_req
*subreq
);
549 static struct tevent_req
*aio_fork_pread_send(struct vfs_handle_struct
*handle
,
551 struct tevent_context
*ev
,
552 struct files_struct
*fsp
,
554 size_t n
, off_t offset
)
556 struct tevent_req
*req
, *subreq
;
557 struct aio_fork_pread_state
*state
;
561 struct aio_fork_config
*config
;
563 SMB_VFS_HANDLE_GET_DATA(handle
, config
,
564 struct aio_fork_config
,
567 req
= tevent_req_create(mem_ctx
, &state
, struct aio_fork_pread_state
);
573 /* TODO: support variable buffers */
574 tevent_req_error(req
, EINVAL
);
575 return tevent_req_post(req
, ev
);
578 err
= get_idle_child(handle
, &state
->child
);
580 tevent_req_error(req
, err
);
581 return tevent_req_post(req
, ev
);
588 cmd
.erratic_testing_mode
= config
->erratic_testing_mode
;
590 DEBUG(10, ("sending fd %d to child %d\n", fsp
->fh
->fd
,
591 (int)state
->child
->pid
));
594 * Not making this async. We're writing into an empty unix
595 * domain socket. This should never block.
597 written
= write_fd(state
->child
->sockfd
, &cmd
, sizeof(cmd
),
602 TALLOC_FREE(state
->child
);
604 DEBUG(10, ("write_fd failed: %s\n", strerror(err
)));
605 tevent_req_error(req
, err
);
606 return tevent_req_post(req
, ev
);
609 subreq
= read_packet_send(state
, ev
, state
->child
->sockfd
,
610 sizeof(struct rw_ret
), NULL
, NULL
);
611 if (tevent_req_nomem(subreq
, req
)) {
612 TALLOC_FREE(state
->child
); /* we sent sth down */
613 return tevent_req_post(req
, ev
);
615 tevent_req_set_callback(subreq
, aio_fork_pread_done
, req
);
619 static void aio_fork_pread_done(struct tevent_req
*subreq
)
621 struct tevent_req
*req
= tevent_req_callback_data(
622 subreq
, struct tevent_req
);
623 struct aio_fork_pread_state
*state
= tevent_req_data(
624 req
, struct aio_fork_pread_state
);
628 struct rw_ret
*retbuf
;
630 nread
= read_packet_recv(subreq
, talloc_tos(), &buf
, &err
);
633 TALLOC_FREE(state
->child
);
634 tevent_req_error(req
, err
);
638 state
->child
->busy
= false;
640 retbuf
= (struct rw_ret
*)buf
;
641 state
->ret
= retbuf
->size
;
642 state
->vfs_aio_state
.error
= retbuf
->ret_errno
;
643 state
->vfs_aio_state
.duration
= retbuf
->duration
;
644 tevent_req_done(req
);
647 static ssize_t
aio_fork_pread_recv(struct tevent_req
*req
,
648 struct vfs_aio_state
*vfs_aio_state
)
650 struct aio_fork_pread_state
*state
= tevent_req_data(
651 req
, struct aio_fork_pread_state
);
653 if (tevent_req_is_unix_error(req
, &vfs_aio_state
->error
)) {
656 *vfs_aio_state
= state
->vfs_aio_state
;
660 struct aio_fork_pwrite_state
{
661 struct aio_child
*child
;
663 struct vfs_aio_state vfs_aio_state
;
666 static void aio_fork_pwrite_done(struct tevent_req
*subreq
);
668 static struct tevent_req
*aio_fork_pwrite_send(
669 struct vfs_handle_struct
*handle
, TALLOC_CTX
*mem_ctx
,
670 struct tevent_context
*ev
, struct files_struct
*fsp
,
671 const void *data
, size_t n
, off_t offset
)
673 struct tevent_req
*req
, *subreq
;
674 struct aio_fork_pwrite_state
*state
;
678 struct aio_fork_config
*config
;
679 SMB_VFS_HANDLE_GET_DATA(handle
, config
,
680 struct aio_fork_config
,
683 req
= tevent_req_create(mem_ctx
, &state
, struct aio_fork_pwrite_state
);
689 /* TODO: support variable buffers */
690 tevent_req_error(req
, EINVAL
);
691 return tevent_req_post(req
, ev
);
694 err
= get_idle_child(handle
, &state
->child
);
696 tevent_req_error(req
, err
);
697 return tevent_req_post(req
, ev
);
704 cmd
.erratic_testing_mode
= config
->erratic_testing_mode
;
706 DEBUG(10, ("sending fd %d to child %d\n", fsp
->fh
->fd
,
707 (int)state
->child
->pid
));
710 * Not making this async. We're writing into an empty unix
711 * domain socket. This should never block.
713 written
= write_fd(state
->child
->sockfd
, &cmd
, sizeof(cmd
),
718 TALLOC_FREE(state
->child
);
720 DEBUG(10, ("write_fd failed: %s\n", strerror(err
)));
721 tevent_req_error(req
, err
);
722 return tevent_req_post(req
, ev
);
725 subreq
= read_packet_send(state
, ev
, state
->child
->sockfd
,
726 sizeof(struct rw_ret
), NULL
, NULL
);
727 if (tevent_req_nomem(subreq
, req
)) {
728 TALLOC_FREE(state
->child
); /* we sent sth down */
729 return tevent_req_post(req
, ev
);
731 tevent_req_set_callback(subreq
, aio_fork_pwrite_done
, req
);
735 static void aio_fork_pwrite_done(struct tevent_req
*subreq
)
737 struct tevent_req
*req
= tevent_req_callback_data(
738 subreq
, struct tevent_req
);
739 struct aio_fork_pwrite_state
*state
= tevent_req_data(
740 req
, struct aio_fork_pwrite_state
);
744 struct rw_ret
*retbuf
;
746 nread
= read_packet_recv(subreq
, talloc_tos(), &buf
, &err
);
749 TALLOC_FREE(state
->child
);
750 tevent_req_error(req
, err
);
754 state
->child
->busy
= false;
756 retbuf
= (struct rw_ret
*)buf
;
757 state
->ret
= retbuf
->size
;
758 state
->vfs_aio_state
.error
= retbuf
->ret_errno
;
759 state
->vfs_aio_state
.duration
= retbuf
->duration
;
760 tevent_req_done(req
);
763 static ssize_t
aio_fork_pwrite_recv(struct tevent_req
*req
,
764 struct vfs_aio_state
*vfs_aio_state
)
766 struct aio_fork_pwrite_state
*state
= tevent_req_data(
767 req
, struct aio_fork_pwrite_state
);
769 if (tevent_req_is_unix_error(req
, &vfs_aio_state
->error
)) {
772 *vfs_aio_state
= state
->vfs_aio_state
;
776 struct aio_fork_fsync_state
{
777 struct aio_child
*child
;
779 struct vfs_aio_state vfs_aio_state
;
782 static void aio_fork_fsync_done(struct tevent_req
*subreq
);
784 static struct tevent_req
*aio_fork_fsync_send(
785 struct vfs_handle_struct
*handle
, TALLOC_CTX
*mem_ctx
,
786 struct tevent_context
*ev
, struct files_struct
*fsp
)
788 struct tevent_req
*req
, *subreq
;
789 struct aio_fork_fsync_state
*state
;
793 struct aio_fork_config
*config
;
795 SMB_VFS_HANDLE_GET_DATA(handle
, config
,
796 struct aio_fork_config
,
799 req
= tevent_req_create(mem_ctx
, &state
, struct aio_fork_fsync_state
);
804 err
= get_idle_child(handle
, &state
->child
);
806 tevent_req_error(req
, err
);
807 return tevent_req_post(req
, ev
);
812 cmd
.erratic_testing_mode
= config
->erratic_testing_mode
;
814 DEBUG(10, ("sending fd %d to child %d\n", fsp
->fh
->fd
,
815 (int)state
->child
->pid
));
818 * Not making this async. We're writing into an empty unix
819 * domain socket. This should never block.
821 written
= write_fd(state
->child
->sockfd
, &cmd
, sizeof(cmd
),
826 TALLOC_FREE(state
->child
);
828 DEBUG(10, ("write_fd failed: %s\n", strerror(err
)));
829 tevent_req_error(req
, err
);
830 return tevent_req_post(req
, ev
);
833 subreq
= read_packet_send(state
, ev
, state
->child
->sockfd
,
834 sizeof(struct rw_ret
), NULL
, NULL
);
835 if (tevent_req_nomem(subreq
, req
)) {
836 TALLOC_FREE(state
->child
); /* we sent sth down */
837 return tevent_req_post(req
, ev
);
839 tevent_req_set_callback(subreq
, aio_fork_fsync_done
, req
);
843 static void aio_fork_fsync_done(struct tevent_req
*subreq
)
845 struct tevent_req
*req
= tevent_req_callback_data(
846 subreq
, struct tevent_req
);
847 struct aio_fork_fsync_state
*state
= tevent_req_data(
848 req
, struct aio_fork_fsync_state
);
852 struct rw_ret
*retbuf
;
854 nread
= read_packet_recv(subreq
, talloc_tos(), &buf
, &err
);
857 TALLOC_FREE(state
->child
);
858 tevent_req_error(req
, err
);
862 state
->child
->busy
= false;
864 retbuf
= (struct rw_ret
*)buf
;
865 state
->ret
= retbuf
->size
;
866 state
->vfs_aio_state
.error
= retbuf
->ret_errno
;
867 state
->vfs_aio_state
.duration
= retbuf
->duration
;
868 tevent_req_done(req
);
871 static int aio_fork_fsync_recv(struct tevent_req
*req
,
872 struct vfs_aio_state
*vfs_aio_state
)
874 struct aio_fork_fsync_state
*state
= tevent_req_data(
875 req
, struct aio_fork_fsync_state
);
877 if (tevent_req_is_unix_error(req
, &vfs_aio_state
->error
)) {
880 *vfs_aio_state
= state
->vfs_aio_state
;
884 static int aio_fork_connect(vfs_handle_struct
*handle
, const char *service
,
888 struct aio_fork_config
*config
;
889 ret
= SMB_VFS_NEXT_CONNECT(handle
, service
, user
);
895 config
= talloc_zero(handle
->conn
, struct aio_fork_config
);
897 SMB_VFS_NEXT_DISCONNECT(handle
);
898 DEBUG(0, ("talloc_zero() failed\n"));
902 config
->erratic_testing_mode
= lp_parm_bool(SNUM(handle
->conn
), "vfs_aio_fork",
903 "erratic_testing_mode", false);
905 SMB_VFS_HANDLE_SET_DATA(handle
, config
,
906 NULL
, struct aio_fork_config
,
912 static struct vfs_fn_pointers vfs_aio_fork_fns
= {
913 .connect_fn
= aio_fork_connect
,
914 .pread_send_fn
= aio_fork_pread_send
,
915 .pread_recv_fn
= aio_fork_pread_recv
,
916 .pwrite_send_fn
= aio_fork_pwrite_send
,
917 .pwrite_recv_fn
= aio_fork_pwrite_recv
,
918 .fsync_send_fn
= aio_fork_fsync_send
,
919 .fsync_recv_fn
= aio_fork_fsync_recv
,
922 NTSTATUS
vfs_aio_fork_init(TALLOC_CTX
*);
923 NTSTATUS
vfs_aio_fork_init(TALLOC_CTX
*ctx
)
925 return smb_register_vfs(SMB_VFS_INTERFACE_VERSION
,
926 "aio_fork", &vfs_aio_fork_fns
);