libcli/named_pipe_auth: we need to hide length of the message mode header from the...
[Samba/gebeck_regimport.git] / source3 / modules / vfs_aio_fork.c
blob8fd89362bc48c9396bcd77946cd9e98e595d409c
1 /*
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.
21 #include "includes.h"
23 struct mmap_area {
24 size_t size;
25 volatile void *ptr;
28 static int mmap_area_destructor(struct mmap_area *area)
30 munmap((void *)area->ptr, area->size);
31 return 0;
34 static struct mmap_area *mmap_area_init(TALLOC_CTX *mem_ctx, size_t size)
36 struct mmap_area *result;
37 int fd;
39 result = talloc(mem_ctx, struct mmap_area);
40 if (result == NULL) {
41 DEBUG(0, ("talloc failed\n"));
42 goto fail;
45 fd = open("/dev/zero", O_RDWR);
46 if (fd == -1) {
47 DEBUG(3, ("open(\"/dev/zero\") failed: %s\n",
48 strerror(errno)));
49 goto fail;
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)));
56 goto fail;
59 close(fd);
61 result->size = size;
62 talloc_set_destructor(result, mmap_area_destructor);
64 return result;
66 fail:
67 TALLOC_FREE(result);
68 return NULL;
71 struct rw_cmd {
72 size_t n;
73 SMB_OFF_T offset;
74 bool read_cmd;
77 struct rw_ret {
78 ssize_t size;
79 int ret_errno;
82 struct aio_child_list;
84 struct aio_child {
85 struct aio_child *prev, *next;
86 struct aio_child_list *list;
87 SMB_STRUCT_AIOCB *aiocb;
88 pid_t pid;
89 int sockfd;
90 struct fd_event *sock_event;
91 struct rw_ret retval;
92 struct mmap_area *map; /* ==NULL means write request */
93 bool dont_delete; /* Marked as in use since last cleanup */
94 bool cancelled;
95 bool read_cmd;
98 struct aio_child_list {
99 struct aio_child *children;
100 struct timed_event *cleanup_event;
103 static void free_aio_children(void **p)
105 TALLOC_FREE(*p);
108 static ssize_t read_fd(int fd, void *ptr, size_t nbytes, int *recvfd)
110 struct msghdr msg;
111 struct iovec iov[1];
112 ssize_t n;
113 #ifndef HAVE_MSGHDR_MSG_CONTROL
114 int newfd;
115 #endif
117 #ifdef HAVE_MSGHDR_MSG_CONTROL
118 union {
119 struct cmsghdr cm;
120 char control[CMSG_SPACE(sizeof(int))];
121 } control_un;
122 struct cmsghdr *cmptr;
124 msg.msg_control = control_un.control;
125 msg.msg_controllen = sizeof(control_un.control);
126 #else
127 #if HAVE_MSGHDR_MSG_ACCTRIGHTS
128 msg.msg_accrights = (caddr_t) &newfd;
129 msg.msg_accrightslen = sizeof(int);
130 #else
131 #error Can not pass file descriptors
132 #endif
133 #endif
135 msg.msg_name = NULL;
136 msg.msg_namelen = 0;
138 iov[0].iov_base = (void *)ptr;
139 iov[0].iov_len = nbytes;
140 msg.msg_iov = iov;
141 msg.msg_iovlen = 1;
143 if ( (n = recvmsg(fd, &msg, 0)) <= 0) {
144 return(n);
147 #ifdef HAVE_MSGHDR_MSG_CONTROL
148 if ((cmptr = CMSG_FIRSTHDR(&msg)) != NULL
149 && cmptr->cmsg_len == CMSG_LEN(sizeof(int))) {
150 if (cmptr->cmsg_level != SOL_SOCKET) {
151 DEBUG(10, ("control level != SOL_SOCKET"));
152 errno = EINVAL;
153 return -1;
155 if (cmptr->cmsg_type != SCM_RIGHTS) {
156 DEBUG(10, ("control type != SCM_RIGHTS"));
157 errno = EINVAL;
158 return -1;
160 *recvfd = *((int *) CMSG_DATA(cmptr));
161 } else {
162 *recvfd = -1; /* descriptor was not passed */
164 #else
165 if (msg.msg_accrightslen == sizeof(int)) {
166 *recvfd = newfd;
168 else {
169 *recvfd = -1; /* descriptor was not passed */
171 #endif
173 return(n);
176 static ssize_t write_fd(int fd, void *ptr, size_t nbytes, int sendfd)
178 struct msghdr msg;
179 struct iovec iov[1];
181 #ifdef HAVE_MSGHDR_MSG_CONTROL
182 union {
183 struct cmsghdr cm;
184 char control[CMSG_SPACE(sizeof(int))];
185 } control_un;
186 struct cmsghdr *cmptr;
188 ZERO_STRUCT(msg);
189 ZERO_STRUCT(control_un);
191 msg.msg_control = control_un.control;
192 msg.msg_controllen = sizeof(control_un.control);
194 cmptr = CMSG_FIRSTHDR(&msg);
195 cmptr->cmsg_len = CMSG_LEN(sizeof(int));
196 cmptr->cmsg_level = SOL_SOCKET;
197 cmptr->cmsg_type = SCM_RIGHTS;
198 *((int *) CMSG_DATA(cmptr)) = sendfd;
199 #else
200 ZERO_STRUCT(msg);
201 msg.msg_accrights = (caddr_t) &sendfd;
202 msg.msg_accrightslen = sizeof(int);
203 #endif
205 msg.msg_name = NULL;
206 msg.msg_namelen = 0;
208 ZERO_STRUCT(iov);
209 iov[0].iov_base = (void *)ptr;
210 iov[0].iov_len = nbytes;
211 msg.msg_iov = iov;
212 msg.msg_iovlen = 1;
214 return (sendmsg(fd, &msg, 0));
217 static void aio_child_cleanup(struct event_context *event_ctx,
218 struct timed_event *te,
219 struct timeval now,
220 void *private_data)
222 struct aio_child_list *list = talloc_get_type_abort(
223 private_data, struct aio_child_list);
224 struct aio_child *child, *next;
226 TALLOC_FREE(list->cleanup_event);
228 for (child = list->children; child != NULL; child = next) {
229 next = child->next;
231 if (child->aiocb != NULL) {
232 DEBUG(10, ("child %d currently active\n",
233 (int)child->pid));
234 continue;
237 if (child->dont_delete) {
238 DEBUG(10, ("Child %d was active since last cleanup\n",
239 (int)child->pid));
240 child->dont_delete = false;
241 continue;
244 DEBUG(10, ("Child %d idle for more than 30 seconds, "
245 "deleting\n", (int)child->pid));
247 TALLOC_FREE(child);
250 if (list->children != NULL) {
252 * Re-schedule the next cleanup round
254 list->cleanup_event = event_add_timed(smbd_event_context(), list,
255 timeval_add(&now, 30, 0),
256 aio_child_cleanup, list);
261 static struct aio_child_list *init_aio_children(struct vfs_handle_struct *handle)
263 struct aio_child_list *data = NULL;
265 if (SMB_VFS_HANDLE_TEST_DATA(handle)) {
266 SMB_VFS_HANDLE_GET_DATA(handle, data, struct aio_child_list,
267 return NULL);
270 if (data == NULL) {
271 data = TALLOC_ZERO_P(NULL, struct aio_child_list);
272 if (data == NULL) {
273 return NULL;
278 * Regardless of whether the child_list had been around or not, make
279 * sure that we have a cleanup timed event. This timed event will
280 * delete itself when it finds that no children are around anymore.
283 if (data->cleanup_event == NULL) {
284 data->cleanup_event = event_add_timed(smbd_event_context(), data,
285 timeval_current_ofs(30, 0),
286 aio_child_cleanup, data);
287 if (data->cleanup_event == NULL) {
288 TALLOC_FREE(data);
289 return 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);
298 return data;
301 static void aio_child_loop(int sockfd, struct mmap_area *map)
303 while (true) {
304 int fd = -1;
305 ssize_t ret;
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,
312 strerror(errno)));
313 exit(1);
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
324 * async I/O times
326 uint8_t randval;
327 unsigned msecs;
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));
335 smb_msleep(msecs);
337 #endif
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,
345 cmd_struct.offset);
346 #ifdef ENABLE_BUILD_FARM_HACKS
347 ret_struct.size = MAX(1, ret_struct.size * 0.9);
348 #endif
350 else {
351 ret_struct.size = sys_pwrite(
352 fd, (void *)map->ptr, cmd_struct.n,
353 cmd_struct.offset);
356 DEBUG(10, ("aio_child_loop: syscall returned %d\n",
357 (int)ret_struct.size));
359 if (ret_struct.size == -1) {
360 ret_struct.ret_errno = errno;
364 * Close the fd before telling our parent we're done. The
365 * parent might close and re-open the file very quickly, and
366 * with system-level share modes (GPFS) we would get an
367 * unjustified SHARING_VIOLATION.
369 close(fd);
371 ret = write_data(sockfd, (char *)&ret_struct,
372 sizeof(ret_struct));
373 if (ret != sizeof(ret_struct)) {
374 DEBUG(10, ("could not write ret_struct: %s\n",
375 strerror(errno)));
376 exit(2);
381 static void handle_aio_completion(struct event_context *event_ctx,
382 struct fd_event *event, uint16 flags,
383 void *p)
385 struct aio_child *child = (struct aio_child *)p;
386 uint64_t mid;
388 DEBUG(10, ("handle_aio_completion called with flags=%d\n", flags));
390 if ((flags & EVENT_FD_READ) == 0) {
391 return;
394 if (!NT_STATUS_IS_OK(read_data(child->sockfd,
395 (char *)&child->retval,
396 sizeof(child->retval)))) {
397 DEBUG(0, ("aio child %d died\n", (int)child->pid));
398 child->retval.size = -1;
399 child->retval.ret_errno = EIO;
402 if (child->cancelled) {
403 child->aiocb = NULL;
404 child->cancelled = false;
405 return;
408 if (child->read_cmd && (child->retval.size > 0)) {
409 SMB_ASSERT(child->retval.size <= child->aiocb->aio_nbytes);
410 memcpy((void *)child->aiocb->aio_buf, (void *)child->map->ptr,
411 child->retval.size);
414 /* FIXME - this won't work for SMB2. */
415 mid = (uint64_t)child->aiocb->aio_sigevent.sigev_value.sival_int;
417 DEBUG(10, ("mid %d finished\n", (int)mid));
419 smbd_aio_complete_mid(mid);
422 static int aio_child_destructor(struct aio_child *child)
424 SMB_ASSERT((child->aiocb == NULL) || child->cancelled);
425 close(child->sockfd);
426 DLIST_REMOVE(child->list->children, child);
427 return 0;
431 * We have to close all fd's in open files, we might incorrectly hold a system
432 * level share mode on a file.
435 static struct files_struct *close_fsp_fd(struct files_struct *fsp,
436 void *private_data)
438 if ((fsp->fh != NULL) && (fsp->fh->fd != -1)) {
439 close(fsp->fh->fd);
440 fsp->fh->fd = -1;
442 return NULL;
445 static NTSTATUS create_aio_child(struct aio_child_list *children,
446 size_t map_size,
447 struct aio_child **presult)
449 struct aio_child *result;
450 int fdpair[2];
451 NTSTATUS status;
453 fdpair[0] = fdpair[1] = -1;
455 result = TALLOC_ZERO_P(children, struct aio_child);
456 NT_STATUS_HAVE_NO_MEMORY(result);
458 if (socketpair(AF_UNIX, SOCK_STREAM, 0, fdpair) == -1) {
459 status = map_nt_error_from_unix(errno);
460 DEBUG(10, ("socketpair() failed: %s\n", strerror(errno)));
461 goto fail;
464 DEBUG(10, ("fdpair = %d/%d\n", fdpair[0], fdpair[1]));
466 result->map = mmap_area_init(result, map_size);
467 if (result->map == NULL) {
468 status = map_nt_error_from_unix(errno);
469 DEBUG(0, ("Could not create mmap area\n"));
470 goto fail;
473 result->pid = sys_fork();
474 if (result->pid == -1) {
475 status = map_nt_error_from_unix(errno);
476 DEBUG(0, ("fork failed: %s\n", strerror(errno)));
477 goto fail;
480 if (result->pid == 0) {
481 close(fdpair[0]);
482 result->sockfd = fdpair[1];
483 files_forall(close_fsp_fd, NULL);
484 aio_child_loop(result->sockfd, result->map);
487 DEBUG(10, ("Child %d created\n", result->pid));
489 result->sockfd = fdpair[0];
490 close(fdpair[1]);
492 result->sock_event = event_add_fd(smbd_event_context(), result,
493 result->sockfd, EVENT_FD_READ,
494 handle_aio_completion,
495 result);
496 if (result->sock_event == NULL) {
497 status = NT_STATUS_NO_MEMORY;
498 DEBUG(0, ("event_add_fd failed\n"));
499 goto fail;
502 result->list = children;
503 DLIST_ADD(children->children, result);
505 talloc_set_destructor(result, aio_child_destructor);
507 *presult = result;
509 return NT_STATUS_OK;
511 fail:
512 if (fdpair[0] != -1) close(fdpair[0]);
513 if (fdpair[1] != -1) close(fdpair[1]);
514 TALLOC_FREE(result);
516 return status;
519 static NTSTATUS get_idle_child(struct vfs_handle_struct *handle,
520 struct aio_child **pchild)
522 struct aio_child_list *children;
523 struct aio_child *child;
524 NTSTATUS status;
526 children = init_aio_children(handle);
527 if (children == NULL) {
528 return NT_STATUS_NO_MEMORY;
531 for (child = children->children; child != NULL; child = child->next) {
532 if (child->aiocb == NULL) {
533 /* idle */
534 break;
538 if (child == NULL) {
539 DEBUG(10, ("no idle child found, creating new one\n"));
541 status = create_aio_child(children, 128*1024, &child);
542 if (!NT_STATUS_IS_OK(status)) {
543 DEBUG(10, ("create_aio_child failed: %s\n",
544 nt_errstr(status)));
545 return status;
549 child->dont_delete = true;
551 *pchild = child;
552 return NT_STATUS_OK;
555 static int aio_fork_read(struct vfs_handle_struct *handle,
556 struct files_struct *fsp, SMB_STRUCT_AIOCB *aiocb)
558 struct aio_child *child;
559 struct rw_cmd cmd;
560 ssize_t ret;
561 NTSTATUS status;
563 if (aiocb->aio_nbytes > 128*1024) {
564 /* TODO: support variable buffers */
565 errno = EINVAL;
566 return -1;
569 status = get_idle_child(handle, &child);
570 if (!NT_STATUS_IS_OK(status)) {
571 DEBUG(10, ("Could not get an idle child\n"));
572 return -1;
575 child->read_cmd = true;
576 child->aiocb = aiocb;
577 child->retval.ret_errno = EINPROGRESS;
579 ZERO_STRUCT(cmd);
580 cmd.n = aiocb->aio_nbytes;
581 cmd.offset = aiocb->aio_offset;
582 cmd.read_cmd = child->read_cmd;
584 DEBUG(10, ("sending fd %d to child %d\n", fsp->fh->fd,
585 (int)child->pid));
587 ret = write_fd(child->sockfd, &cmd, sizeof(cmd), fsp->fh->fd);
588 if (ret == -1) {
589 DEBUG(10, ("write_fd failed: %s\n", strerror(errno)));
590 return -1;
593 return 0;
596 static int aio_fork_write(struct vfs_handle_struct *handle,
597 struct files_struct *fsp, SMB_STRUCT_AIOCB *aiocb)
599 struct aio_child *child;
600 struct rw_cmd cmd;
601 ssize_t ret;
602 NTSTATUS status;
604 if (aiocb->aio_nbytes > 128*1024) {
605 /* TODO: support variable buffers */
606 errno = EINVAL;
607 return -1;
610 status = get_idle_child(handle, &child);
611 if (!NT_STATUS_IS_OK(status)) {
612 DEBUG(10, ("Could not get an idle child\n"));
613 return -1;
616 child->read_cmd = false;
617 child->aiocb = aiocb;
618 child->retval.ret_errno = EINPROGRESS;
620 memcpy((void *)child->map->ptr, (void *)aiocb->aio_buf,
621 aiocb->aio_nbytes);
623 ZERO_STRUCT(cmd);
624 cmd.n = aiocb->aio_nbytes;
625 cmd.offset = aiocb->aio_offset;
626 cmd.read_cmd = child->read_cmd;
628 DEBUG(10, ("sending fd %d to child %d\n", fsp->fh->fd,
629 (int)child->pid));
631 ret = write_fd(child->sockfd, &cmd, sizeof(cmd), fsp->fh->fd);
632 if (ret == -1) {
633 DEBUG(10, ("write_fd failed: %s\n", strerror(errno)));
634 return -1;
637 return 0;
640 static struct aio_child *aio_fork_find_child(struct vfs_handle_struct *handle,
641 SMB_STRUCT_AIOCB *aiocb)
643 struct aio_child_list *children;
644 struct aio_child *child;
646 children = init_aio_children(handle);
647 if (children == NULL) {
648 return NULL;
651 for (child = children->children; child != NULL; child = child->next) {
652 if (child->aiocb == aiocb) {
653 return child;
657 return NULL;
660 static ssize_t aio_fork_return_fn(struct vfs_handle_struct *handle,
661 struct files_struct *fsp,
662 SMB_STRUCT_AIOCB *aiocb)
664 struct aio_child *child = aio_fork_find_child(handle, aiocb);
666 if (child == NULL) {
667 errno = EINVAL;
668 DEBUG(0, ("returning EINVAL\n"));
669 return -1;
672 child->aiocb = NULL;
674 if (child->retval.size == -1) {
675 errno = child->retval.ret_errno;
678 return child->retval.size;
681 static int aio_fork_cancel(struct vfs_handle_struct *handle,
682 struct files_struct *fsp,
683 SMB_STRUCT_AIOCB *aiocb)
685 struct aio_child_list *children;
686 struct aio_child *child;
688 children = init_aio_children(handle);
689 if (children == NULL) {
690 errno = EINVAL;
691 return -1;
694 for (child = children->children; child != NULL; child = child->next) {
695 if (child->aiocb == NULL) {
696 continue;
698 if (child->aiocb->aio_fildes != fsp->fh->fd) {
699 continue;
701 if ((aiocb != NULL) && (child->aiocb != aiocb)) {
702 continue;
706 * We let the child do its job, but we discard the result when
707 * it's finished.
710 child->cancelled = true;
713 return AIO_CANCELED;
716 static int aio_fork_error_fn(struct vfs_handle_struct *handle,
717 struct files_struct *fsp,
718 SMB_STRUCT_AIOCB *aiocb)
720 struct aio_child *child = aio_fork_find_child(handle, aiocb);
722 if (child == NULL) {
723 errno = EINVAL;
724 return -1;
727 return child->retval.ret_errno;
730 static struct vfs_fn_pointers vfs_aio_fork_fns = {
731 .aio_read = aio_fork_read,
732 .aio_write = aio_fork_write,
733 .aio_return_fn = aio_fork_return_fn,
734 .aio_cancel = aio_fork_cancel,
735 .aio_error_fn = aio_fork_error_fn,
738 NTSTATUS vfs_aio_fork_init(void);
739 NTSTATUS vfs_aio_fork_init(void)
741 return smb_register_vfs(SMB_VFS_INTERFACE_VERSION,
742 "aio_fork", &vfs_aio_fork_fns);