2 Unix SMB/Netbios implementation.
4 async_io read handling using POSIX async io.
5 Copyright (C) Jeremy Allison 2005.
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.
26 /* The signal we'll use to signify aio done. */
28 #define RT_SIGNAL_AIO (SIGRTMIN+3)
31 /****************************************************************************
32 The buffer we keep around whilst an aio request is in process.
33 *****************************************************************************/
36 struct aio_extra *next, *prev;
45 static struct aio_extra *aio_list_head;
47 /****************************************************************************
48 Create the extended aio struct we must keep around for the lifetime
50 *****************************************************************************/
52 static struct aio_extra *create_aio_ex_read(files_struct *fsp, size_t buflen,
55 struct aio_extra *aio_ex = SMB_MALLOC_P(struct aio_extra);
61 /* The output buffer stored in the aio_ex is the start of
62 the smb return buffer. The buffer used in the acb
63 is the start of the reply data portion of that buffer. */
64 aio_ex->outbuf = SMB_MALLOC_ARRAY(char, buflen);
65 if (!aio_ex->outbuf) {
69 DLIST_ADD(aio_list_head, aio_ex);
71 aio_ex->read_req = True;
76 /****************************************************************************
77 Create the extended aio struct we must keep around for the lifetime
78 of the aio_write call.
79 *****************************************************************************/
81 static struct aio_extra *create_aio_ex_write(files_struct *fsp,
86 struct aio_extra *aio_ex = SMB_MALLOC_P(struct aio_extra);
93 /* We need space for an output reply of outbuflen bytes. */
94 aio_ex->outbuf = SMB_MALLOC_ARRAY(char, outbuflen);
95 if (!aio_ex->outbuf) {
100 if (!(aio_ex->inbuf = SMB_MALLOC_ARRAY(char, inbuflen))) {
101 SAFE_FREE(aio_ex->outbuf);
106 DLIST_ADD(aio_list_head, aio_ex);
108 aio_ex->read_req = False;
113 /****************************************************************************
114 Delete the extended aio struct.
115 *****************************************************************************/
117 static void delete_aio_ex(struct aio_extra *aio_ex)
119 DLIST_REMOVE(aio_list_head, aio_ex);
120 SAFE_FREE(aio_ex->inbuf);
121 SAFE_FREE(aio_ex->outbuf);
125 /****************************************************************************
126 Given the aiocb struct find the extended aio struct containing it.
127 *****************************************************************************/
129 static struct aio_extra *find_aio_ex(uint16 mid)
133 for( p = aio_list_head; p; p = p->next) {
141 /****************************************************************************
142 We can have these many aio buffers in flight.
143 *****************************************************************************/
145 #define AIO_PENDING_SIZE 10
146 static sig_atomic_t signals_received;
147 static int outstanding_aio_calls;
148 static uint16 aio_pending_array[AIO_PENDING_SIZE];
150 /****************************************************************************
151 Signal handler when an aio request completes.
152 *****************************************************************************/
154 static void signal_handler(int sig, siginfo_t *info, void *unused)
156 if (signals_received < AIO_PENDING_SIZE) {
157 aio_pending_array[signals_received] = info->si_value.sival_int;
159 } /* Else signal is lost. */
160 sys_select_signal(RT_SIGNAL_AIO);
163 /****************************************************************************
164 Is there a signal waiting ?
165 *****************************************************************************/
167 BOOL aio_finished(void)
169 return (signals_received != 0);
172 /****************************************************************************
173 Initialize the signal handler for aio read/write.
174 *****************************************************************************/
176 void initialize_async_io_handler(void)
178 struct sigaction act;
181 act.sa_sigaction = signal_handler;
182 act.sa_flags = SA_SIGINFO;
183 sigemptyset( &act.sa_mask );
184 if (sigaction(RT_SIGNAL_AIO, &act, NULL) != 0) {
185 DEBUG(0,("Failed to setup RT_SIGNAL_AIO handler\n"));
188 /* the signal can start off blocked due to a bug in bash */
189 BlockSignals(False, RT_SIGNAL_AIO);
192 /****************************************************************************
193 Set up an aio request from a SMBreadX call.
194 *****************************************************************************/
196 BOOL schedule_aio_read_and_X(connection_struct *conn,
197 char *inbuf, char *outbuf,
198 int length, int len_outbuf,
199 files_struct *fsp, SMB_OFF_T startpos,
202 struct aio_extra *aio_ex;
205 size_t min_aio_read_size = lp_aio_read_size(SNUM(conn));
207 if (!min_aio_read_size || (smb_maxcnt < min_aio_read_size)) {
208 /* Too small a read for aio request. */
209 DEBUG(10,("schedule_aio_read_and_X: read size (%u) too small "
210 "for minimum aio_read of %u\n",
211 (unsigned int)smb_maxcnt,
212 (unsigned int)min_aio_read_size ));
216 /* Only do this on non-chained and non-chaining reads not using the
218 if (chain_size !=0 || (CVAL(inbuf,smb_vwv0) != 0xFF)
219 || (lp_write_cache_size(SNUM(conn)) != 0) ) {
223 if (outstanding_aio_calls >= AIO_PENDING_SIZE) {
224 DEBUG(10,("schedule_aio_read_and_X: Already have %d aio "
225 "activities outstanding.\n",
226 outstanding_aio_calls ));
230 /* The following is safe from integer wrap as we've already
231 checked smb_maxcnt is 128k or less. */
232 bufsize = PTR_DIFF(smb_buf(outbuf),outbuf) + smb_maxcnt;
234 if ((aio_ex = create_aio_ex_read(fsp, bufsize,
235 SVAL(inbuf,smb_mid))) == NULL) {
236 DEBUG(10,("schedule_aio_read_and_X: malloc fail.\n"));
240 /* Copy the SMB header already setup in outbuf. */
241 memcpy(aio_ex->outbuf, outbuf, smb_buf(outbuf) - outbuf);
242 SCVAL(aio_ex->outbuf,smb_vwv0,0xFF); /* Never a chained reply. */
246 /* Now set up the aio record for the read call. */
248 a->aio_fildes = fsp->fh->fd;
249 a->aio_buf = smb_buf(aio_ex->outbuf);
250 a->aio_nbytes = smb_maxcnt;
251 a->aio_offset = startpos;
252 a->aio_sigevent.sigev_notify = SIGEV_SIGNAL;
253 a->aio_sigevent.sigev_signo = RT_SIGNAL_AIO;
254 a->aio_sigevent.sigev_value.sival_int = aio_ex->mid;
257 if (SMB_VFS_AIO_READ(fsp,a) == -1) {
258 DEBUG(0,("schedule_aio_read_and_X: aio_read failed. "
259 "Error %s\n", strerror(errno) ));
260 delete_aio_ex(aio_ex);
266 DEBUG(10,("schedule_aio_read_and_X: scheduled aio_read for file %s, "
267 "offset %.0f, len = %u (mid = %u)\n",
268 fsp->fsp_name, (double)startpos, (unsigned int)smb_maxcnt,
269 (unsigned int)aio_ex->mid ));
271 srv_defer_sign_response(aio_ex->mid);
272 outstanding_aio_calls++;
276 /****************************************************************************
277 Set up an aio request from a SMBwriteX call.
278 *****************************************************************************/
280 BOOL schedule_aio_write_and_X(connection_struct *conn,
281 char *inbuf, char *outbuf,
282 int length, int len_outbuf,
283 files_struct *fsp, char *data,
287 struct aio_extra *aio_ex;
289 size_t inbufsize, outbufsize;
290 BOOL write_through = BITSETW(inbuf+smb_vwv7,0);
291 size_t min_aio_write_size = lp_aio_write_size(SNUM(conn));
293 if (!min_aio_write_size || (numtowrite < min_aio_write_size)) {
294 /* Too small a write for aio request. */
295 DEBUG(10,("schedule_aio_write_and_X: write size (%u) too "
296 "small for minimum aio_write of %u\n",
297 (unsigned int)numtowrite,
298 (unsigned int)min_aio_write_size ));
302 /* Only do this on non-chained and non-chaining reads not using the
304 if (chain_size !=0 || (CVAL(inbuf,smb_vwv0) != 0xFF)
305 || (lp_write_cache_size(SNUM(conn)) != 0) ) {
309 if (outstanding_aio_calls >= AIO_PENDING_SIZE) {
310 DEBUG(3,("schedule_aio_write_and_X: Already have %d aio "
311 "activities outstanding.\n",
312 outstanding_aio_calls ));
313 DEBUG(10,("schedule_aio_write_and_X: failed to schedule "
314 "aio_write for file %s, offset %.0f, len = %u "
316 fsp->fsp_name, (double)startpos,
317 (unsigned int)numtowrite,
318 (unsigned int)SVAL(inbuf,smb_mid) ));
322 inbufsize = smb_len(inbuf) + 4;
323 outbufsize = smb_len(outbuf) + 4;
324 if (!(aio_ex = create_aio_ex_write(fsp, inbufsize, outbufsize,
325 SVAL(inbuf,smb_mid)))) {
326 DEBUG(0,("schedule_aio_write_and_X: malloc fail.\n"));
330 /* Copy the SMB header already setup in outbuf. */
331 memcpy(aio_ex->inbuf, inbuf, inbufsize);
333 /* Copy the SMB header already setup in outbuf. */
334 memcpy(aio_ex->outbuf, outbuf, outbufsize);
335 SCVAL(aio_ex->outbuf,smb_vwv0,0xFF); /* Never a chained reply. */
339 /* Now set up the aio record for the write call. */
341 a->aio_fildes = fsp->fh->fd;
342 a->aio_buf = aio_ex->inbuf + (PTR_DIFF(data, inbuf));
343 a->aio_nbytes = numtowrite;
344 a->aio_offset = startpos;
345 a->aio_sigevent.sigev_notify = SIGEV_SIGNAL;
346 a->aio_sigevent.sigev_signo = RT_SIGNAL_AIO;
347 a->aio_sigevent.sigev_value.sival_int = aio_ex->mid;
350 if (SMB_VFS_AIO_WRITE(fsp,a) == -1) {
351 DEBUG(3,("schedule_aio_wrote_and_X: aio_write failed. "
352 "Error %s\n", strerror(errno) ));
353 delete_aio_ex(aio_ex);
359 if (!write_through && !lp_syncalways(SNUM(fsp->conn))
360 && fsp->aio_write_behind) {
361 /* Lie to the client and immediately claim we finished the
363 SSVAL(aio_ex->outbuf,smb_vwv2,numtowrite);
364 SSVAL(aio_ex->outbuf,smb_vwv4,(numtowrite>>16)&1);
365 show_msg(aio_ex->outbuf);
366 if (!send_smb(smbd_server_fd(),aio_ex->outbuf)) {
367 exit_server_cleanly("handle_aio_write: send_smb "
370 DEBUG(10,("schedule_aio_write_and_X: scheduled aio_write "
371 "behind for file %s\n", fsp->fsp_name ));
373 srv_defer_sign_response(aio_ex->mid);
375 outstanding_aio_calls++;
377 DEBUG(10,("schedule_aio_write_and_X: scheduled aio_write for file "
378 "%s, offset %.0f, len = %u (mid = %u) "
379 "outstanding_aio_calls = %d\n",
380 fsp->fsp_name, (double)startpos, (unsigned int)numtowrite,
381 (unsigned int)aio_ex->mid, outstanding_aio_calls ));
387 /****************************************************************************
388 Complete the read and return the data or error back to the client.
389 Returns errno or zero if all ok.
390 *****************************************************************************/
392 static int handle_aio_read_complete(struct aio_extra *aio_ex)
396 char *outbuf = aio_ex->outbuf;
397 char *data = smb_buf(outbuf);
398 ssize_t nread = SMB_VFS_AIO_RETURN(aio_ex->fsp,&aio_ex->acb);
401 /* We're relying here on the fact that if the fd is
402 closed then the aio will complete and aio_return
403 will return an error. Hopefully this is
406 /* If errno is ECANCELED then don't return anything to the
408 if (errno == ECANCELED) {
409 srv_cancel_sign_response(aio_ex->mid);
413 DEBUG( 3,( "handle_aio_read_complete: file %s nread == -1. "
415 aio_ex->fsp->fsp_name, strerror(errno) ));
417 outsize = (UNIXERROR(ERRDOS,ERRnoaccess));
420 outsize = set_message(outbuf,12,nread,False);
421 SSVAL(outbuf,smb_vwv2,0xFFFF); /* Remaining - must be * -1. */
422 SSVAL(outbuf,smb_vwv5,nread);
423 SSVAL(outbuf,smb_vwv6,smb_offset(data,outbuf));
424 SSVAL(outbuf,smb_vwv7,((nread >> 16) & 1));
425 SSVAL(smb_buf(outbuf),-2,nread);
427 DEBUG( 3, ( "handle_aio_read_complete file %s max=%d "
429 aio_ex->fsp->fsp_name,
430 aio_ex->acb.aio_nbytes, (int)nread ) );
433 smb_setlen(outbuf,outsize - 4);
435 if (!send_smb(smbd_server_fd(),outbuf)) {
436 exit_server_cleanly("handle_aio_read_complete: send_smb "
440 DEBUG(10,("handle_aio_read_complete: scheduled aio_read completed "
441 "for file %s, offset %.0f, len = %u\n",
442 aio_ex->fsp->fsp_name, (double)aio_ex->acb.aio_offset,
443 (unsigned int)nread ));
448 /****************************************************************************
449 Complete the write and return the data or error back to the client.
450 Returns errno or zero if all ok.
451 *****************************************************************************/
453 static int handle_aio_write_complete(struct aio_extra *aio_ex)
456 files_struct *fsp = aio_ex->fsp;
457 char *outbuf = aio_ex->outbuf;
458 ssize_t numtowrite = aio_ex->acb.aio_nbytes;
459 ssize_t nwritten = SMB_VFS_AIO_RETURN(fsp,&aio_ex->acb);
461 if (fsp->aio_write_behind) {
462 if (nwritten != numtowrite) {
463 if (nwritten == -1) {
464 DEBUG(5,("handle_aio_write_complete: "
465 "aio_write_behind failed ! File %s "
466 "is corrupt ! Error %s\n",
467 fsp->fsp_name, strerror(errno) ));
470 DEBUG(0,("handle_aio_write_complete: "
471 "aio_write_behind failed ! File %s "
472 "is corrupt ! Wanted %u bytes but "
473 "only wrote %d\n", fsp->fsp_name,
474 (unsigned int)numtowrite,
479 DEBUG(10,("handle_aio_write_complete: "
480 "aio_write_behind completed for file %s\n",
486 /* We don't need outsize or set_message here as we've already set the
487 fixed size length when we set up the aio call. */
490 DEBUG( 3,( "handle_aio_write: file %s wanted %u bytes. "
491 "nwritten == %d. Error = %s\n",
492 fsp->fsp_name, (unsigned int)numtowrite,
493 (int)nwritten, strerror(errno) ));
495 /* If errno is ECANCELED then don't return anything to the
497 if (errno == ECANCELED) {
498 srv_cancel_sign_response(aio_ex->mid);
502 UNIXERROR(ERRHRD,ERRdiskfull);
505 BOOL write_through = BITSETW(aio_ex->inbuf+smb_vwv7,0);
508 SSVAL(outbuf,smb_vwv2,nwritten);
509 SSVAL(outbuf,smb_vwv4,(nwritten>>16)&1);
510 if (nwritten < (ssize_t)numtowrite) {
511 SCVAL(outbuf,smb_rcls,ERRHRD);
512 SSVAL(outbuf,smb_err,ERRdiskfull);
515 DEBUG(3,("handle_aio_write: fnum=%d num=%d wrote=%d\n",
516 fsp->fnum, (int)numtowrite, (int)nwritten));
517 status = sync_file(fsp->conn,fsp, write_through);
518 if (!NT_STATUS_IS_OK(status)) {
519 UNIXERROR(ERRHRD,ERRdiskfull);
521 DEBUG(5,("handle_aio_write: sync_file for %s returned %s\n",
522 fsp->fsp_name, nt_errstr(status) ));
527 if (!send_smb(smbd_server_fd(),outbuf)) {
528 exit_server_cleanly("handle_aio_write: send_smb failed.");
531 DEBUG(10,("handle_aio_write_complete: scheduled aio_write completed "
532 "for file %s, offset %.0f, requested %u, written = %u\n",
533 fsp->fsp_name, (double)aio_ex->acb.aio_offset,
534 (unsigned int)numtowrite, (unsigned int)nwritten ));
539 /****************************************************************************
540 Handle any aio completion. Returns True if finished (and sets *perr if err
541 was non-zero), False if not.
542 *****************************************************************************/
544 static BOOL handle_aio_completed(struct aio_extra *aio_ex, int *perr)
548 /* Ensure the operation has really completed. */
549 if (SMB_VFS_AIO_ERROR(aio_ex->fsp, &aio_ex->acb) == EINPROGRESS) {
550 DEBUG(10,( "handle_aio_completed: operation mid %u still in "
551 "process for file %s\n",
552 aio_ex->mid, aio_ex->fsp->fsp_name ));
556 if (aio_ex->read_req) {
557 err = handle_aio_read_complete(aio_ex);
559 err = handle_aio_write_complete(aio_ex);
563 *perr = err; /* Only save non-zero errors. */
569 /****************************************************************************
570 Handle any aio completion inline.
571 Returns non-zero errno if fail or zero if all ok.
572 *****************************************************************************/
574 int process_aio_queue(void)
579 BlockSignals(True, RT_SIGNAL_AIO);
581 DEBUG(10,("process_aio_queue: signals_received = %d\n",
582 (int)signals_received));
583 DEBUG(10,("process_aio_queue: outstanding_aio_calls = %d\n",
584 outstanding_aio_calls));
586 if (!signals_received) {
587 BlockSignals(False, RT_SIGNAL_AIO);
591 /* Drain all the complete aio_reads. */
592 for (i = 0; i < signals_received; i++) {
593 uint16 mid = aio_pending_array[i];
594 files_struct *fsp = NULL;
595 struct aio_extra *aio_ex = find_aio_ex(mid);
598 DEBUG(3,("process_aio_queue: Can't find record to "
599 "match mid %u.\n", (unsigned int)mid));
600 srv_cancel_sign_response(mid);
606 /* file was closed whilst I/O was outstanding. Just
608 DEBUG( 3,( "process_aio_queue: file closed whilst "
609 "aio outstanding.\n"));
610 srv_cancel_sign_response(mid);
614 if (!handle_aio_completed(aio_ex, &ret)) {
618 delete_aio_ex(aio_ex);
621 outstanding_aio_calls -= signals_received;
622 signals_received = 0;
623 BlockSignals(False, RT_SIGNAL_AIO);
627 /****************************************************************************
628 We're doing write behind and the client closed the file. Wait up to 30
629 seconds (my arbitrary choice) for the aio to complete. Return 0 if all writes
630 completed, errno to return if not.
631 *****************************************************************************/
633 #define SMB_TIME_FOR_AIO_COMPLETE_WAIT 29
635 int wait_for_aio_completion(files_struct *fsp)
637 struct aio_extra *aio_ex;
638 const SMB_STRUCT_AIOCB **aiocb_list;
639 int aio_completion_count = 0;
640 time_t start_time = time(NULL);
643 for (seconds_left = SMB_TIME_FOR_AIO_COMPLETE_WAIT;
644 seconds_left >= 0;) {
649 aio_completion_count = 0;
650 for( aio_ex = aio_list_head; aio_ex; aio_ex = aio_ex->next) {
651 if (aio_ex->fsp == fsp) {
652 aio_completion_count++;
656 if (!aio_completion_count) {
660 DEBUG(3,("wait_for_aio_completion: waiting for %d aio events "
661 "to complete.\n", aio_completion_count ));
663 aiocb_list = SMB_MALLOC_ARRAY(const SMB_STRUCT_AIOCB *,
664 aio_completion_count);
669 for( i = 0, aio_ex = aio_list_head;
671 aio_ex = aio_ex->next) {
672 if (aio_ex->fsp == fsp) {
673 aiocb_list[i++] = &aio_ex->acb;
677 /* Now wait up to seconds_left for completion. */
678 ts.tv_sec = seconds_left;
681 DEBUG(10,("wait_for_aio_completion: %d events, doing a wait "
683 aio_completion_count, seconds_left ));
685 err = SMB_VFS_AIO_SUSPEND(fsp, aiocb_list,
686 aio_completion_count, &ts);
688 DEBUG(10,("wait_for_aio_completion: returned err = %d, "
689 "errno = %s\n", err, strerror(errno) ));
691 if (err == -1 && errno == EAGAIN) {
692 DEBUG(0,("wait_for_aio_completion: aio_suspend timed "
693 "out waiting for %d events after a wait of "
694 "%d seconds\n", aio_completion_count,
697 cancel_aio_by_fsp(fsp);
698 SAFE_FREE(aiocb_list);
702 /* One or more events might have completed - process them if
704 for( i = 0; i < aio_completion_count; i++) {
705 uint16 mid = aiocb_list[i]->aio_sigevent.sigev_value.sival_int;
707 aio_ex = find_aio_ex(mid);
710 DEBUG(0, ("wait_for_aio_completion: mid %u "
711 "doesn't match an aio record\n",
712 (unsigned int)mid ));
716 if (!handle_aio_completed(aio_ex, &err)) {
719 delete_aio_ex(aio_ex);
722 SAFE_FREE(aiocb_list);
723 seconds_left = SMB_TIME_FOR_AIO_COMPLETE_WAIT
724 - (time(NULL) - start_time);
727 /* We timed out - we don't know why. Return ret if already an error,
729 DEBUG(10,("wait_for_aio_completion: aio_suspend timed out waiting "
731 aio_completion_count));
736 /****************************************************************************
737 Cancel any outstanding aio requests. The client doesn't care about the reply.
738 *****************************************************************************/
740 void cancel_aio_by_fsp(files_struct *fsp)
742 struct aio_extra *aio_ex;
744 for( aio_ex = aio_list_head; aio_ex; aio_ex = aio_ex->next) {
745 if (aio_ex->fsp == fsp) {
746 /* Don't delete the aio_extra record as we may have
747 completed and don't yet know it. Just do the
748 aio_cancel call and return. */
749 SMB_VFS_AIO_CANCEL(fsp,fsp->fh->fd, &aio_ex->acb);
750 aio_ex->fsp = NULL; /* fsp will be closed when we
757 BOOL aio_finished(void)
762 void initialize_async_io_handler(void)
766 int process_aio_queue(void)
771 BOOL schedule_aio_read_and_X(connection_struct *conn,
772 char *inbuf, char *outbuf,
773 int length, int len_outbuf,
774 files_struct *fsp, SMB_OFF_T startpos,
780 BOOL schedule_aio_write_and_X(connection_struct *conn,
781 char *inbuf, char *outbuf,
782 int length, int len_outbuf,
783 files_struct *fsp, char *data,
790 void cancel_aio_by_fsp(files_struct *fsp)
794 BOOL wait_for_aio_completion(files_struct *fsp)