?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
s3: Use SMB_VFS_PREAD_SEND in schedule_aio_read_and_X
[Samba/bb.git] / source3 / smbd / aio.c
blob5de561b370a13d76b25beb672dd2d2873620295a
1 /*
2 Unix SMB/Netbios implementation.
3 Version 3.0
4 async_io read handling using POSIX async io.
5 Copyright (C) Jeremy Allison 2005.
6
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 3 of the License, or
10 (at your option) any later version.
11
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.
16
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>.
19 */
20
21 #include "includes.h"
22 #include "smbd/smbd.h"
23 #include "smbd/globals.h"
24 #include "../lib/util/tevent_ntstatus.h"
25
26 #if defined(HAVE_AIO)
27
28 /* The signal we'll use to signify aio done. */
29 #ifndef RT_SIGNAL_AIO
30 #define RT_SIGNAL_AIO (SIGRTMIN+3)
31 #endif
32
33 #ifndef HAVE_STRUCT_SIGEVENT_SIGEV_VALUE_SIVAL_PTR
34 #ifdef HAVE_STRUCT_SIGEVENT_SIGEV_VALUE_SIGVAL_PTR
35 #define sival_int sigval_int
36 #define sival_ptr sigval_ptr
37 #endif
38 #endif
39
40 /****************************************************************************
41 The buffer we keep around whilst an aio request is in process.
42 *****************************************************************************/
43
44 struct aio_extra {
45 struct aio_extra *next, *prev;
46 SMB_STRUCT_AIOCB acb;
47 files_struct *fsp;
48 struct smb_request *smbreq;
49 DATA_BLOB outbuf;
50 struct lock_struct lock;
51 bool write_through;
52 int (*handle_completion)(struct aio_extra *ex, int errcode);
53 };
54
55 /****************************************************************************
56 Accessor function to return write_through state.
57 *****************************************************************************/
58
59 bool aio_write_through_requested(struct aio_extra *aio_ex)
60 {
61 return aio_ex->write_through;
62 }
63
64 /****************************************************************************
65 Initialize the signal handler for aio read/write.
66 *****************************************************************************/
67
68 static void smbd_aio_signal_handler(struct tevent_context *ev_ctx,
69 struct tevent_signal *se,
70 int signum, int count,
71 void *_info, void *private_data)
72 {
73 siginfo_t *info = (siginfo_t *)_info;
74 struct aio_extra *aio_ex = (struct aio_extra *)
75 info->si_value.sival_ptr;
76
77 smbd_aio_complete_aio_ex(aio_ex);
78 TALLOC_FREE(aio_ex);
79 }
80
81
82 bool initialize_async_io_handler(void)
83 {
84 static bool tried_signal_setup = false;
85
86 if (aio_signal_event) {
87 return true;
88 }
89 if (tried_signal_setup) {
90 return false;
91 }
92 tried_signal_setup = true;
93
94 aio_signal_event = tevent_add_signal(server_event_context(),
95 server_event_context(),
96 RT_SIGNAL_AIO, SA_SIGINFO,
97 smbd_aio_signal_handler,
98 NULL);
99 if (!aio_signal_event) {
100 DEBUG(10, ("Failed to setup RT_SIGNAL_AIO handler\n"));
101 return false;
102 }
103 return true;
104 }
105
106 static int handle_aio_read_complete(struct aio_extra *aio_ex, int errcode);
107 static int handle_aio_write_complete(struct aio_extra *aio_ex, int errcode);
108 static int handle_aio_smb2_read_complete(struct aio_extra *aio_ex, int errcode);
109 static int handle_aio_smb2_write_complete(struct aio_extra *aio_ex, int errcode);
110
111 static int aio_extra_destructor(struct aio_extra *aio_ex)
112 {
113 DLIST_REMOVE(aio_list_head, aio_ex);
114 outstanding_aio_calls--;
115 return 0;
116 }
117
118 /****************************************************************************
119 Create the extended aio struct we must keep around for the lifetime
120 of the aio call.
121 *****************************************************************************/
122
123 static struct aio_extra *create_aio_extra(TALLOC_CTX *mem_ctx,
124 files_struct *fsp,
125 size_t buflen)
126 {
127 struct aio_extra *aio_ex = talloc_zero(mem_ctx, struct aio_extra);
128
129 if (!aio_ex) {
130 return NULL;
131 }
132
133 /* The output buffer stored in the aio_ex is the start of
134 the smb return buffer. The buffer used in the acb
135 is the start of the reply data portion of that buffer. */
136
137 if (buflen) {
138 aio_ex->outbuf = data_blob_talloc(aio_ex, NULL, buflen);
139 if (!aio_ex->outbuf.data) {
140 TALLOC_FREE(aio_ex);
141 return NULL;
142 }
143 }
144 DLIST_ADD(aio_list_head, aio_ex);
145 talloc_set_destructor(aio_ex, aio_extra_destructor);
146 aio_ex->fsp = fsp;
147 outstanding_aio_calls++;
148 return aio_ex;
149 }
150
151 static void aio_pread_smb1_done(struct tevent_req *req);
152
153 /****************************************************************************
154 Set up an aio request from a SMBreadX call.
155 *****************************************************************************/
156
157 NTSTATUS schedule_aio_read_and_X(connection_struct *conn,
158 struct smb_request *smbreq,
159 files_struct *fsp, off_t startpos,
160 size_t smb_maxcnt)
161 {
162 struct aio_extra *aio_ex;
163 SMB_STRUCT_AIOCB *a;
164 size_t bufsize;
165 size_t min_aio_read_size = lp_aio_read_size(SNUM(conn));
166 struct tevent_req *req;
167
168 if (fsp->base_fsp != NULL) {
169 /* No AIO on streams yet */
170 DEBUG(10, ("AIO on streams not yet supported\n"));
171 return NT_STATUS_RETRY;
172 }
173
174 if ((!min_aio_read_size || (smb_maxcnt < min_aio_read_size))
175 && !SMB_VFS_AIO_FORCE(fsp)) {
176 /* Too small a read for aio request. */
177 DEBUG(10,("schedule_aio_read_and_X: read size (%u) too small "
178 "for minimum aio_read of %u\n",
179 (unsigned int)smb_maxcnt,
180 (unsigned int)min_aio_read_size ));
181 return NT_STATUS_RETRY;
182 }
183
184 /* Only do this on non-chained and non-chaining reads not using the
185 * write cache. */
186 if (req_is_in_chain(smbreq) || (lp_write_cache_size(SNUM(conn)) != 0)) {
187 return NT_STATUS_RETRY;
188 }
189
190 if (outstanding_aio_calls >= aio_pending_size) {
191 DEBUG(10,("schedule_aio_read_and_X: Already have %d aio "
192 "activities outstanding.\n",
193 outstanding_aio_calls ));
194 return NT_STATUS_RETRY;
195 }
196
197 /* The following is safe from integer wrap as we've already checked
198 smb_maxcnt is 128k or less. Wct is 12 for read replies */
199
200 bufsize = smb_size + 12 * 2 + smb_maxcnt;
201
202 if ((aio_ex = create_aio_extra(NULL, fsp, bufsize)) == NULL) {
203 DEBUG(10,("schedule_aio_read_and_X: malloc fail.\n"));
204 return NT_STATUS_NO_MEMORY;
205 }
206 aio_ex->handle_completion = handle_aio_read_complete;
207
208 construct_reply_common_req(smbreq, (char *)aio_ex->outbuf.data);
209 srv_set_message((char *)aio_ex->outbuf.data, 12, 0, True);
210 SCVAL(aio_ex->outbuf.data,smb_vwv0,0xFF); /* Never a chained reply. */
211
212 init_strict_lock_struct(fsp, (uint64_t)smbreq->smbpid,
213 (uint64_t)startpos, (uint64_t)smb_maxcnt, READ_LOCK,
214 &aio_ex->lock);
215
216 /* Take the lock until the AIO completes. */
217 if (!SMB_VFS_STRICT_LOCK(conn, fsp, &aio_ex->lock)) {
218 TALLOC_FREE(aio_ex);
219 return NT_STATUS_FILE_LOCK_CONFLICT;
220 }
221
222 a = &aio_ex->acb;
223
224 /* Now set up the aio record for the read call. */
225
226 a->aio_fildes = fsp->fh->fd;
227 a->aio_buf = smb_buf(aio_ex->outbuf.data);
228 a->aio_nbytes = smb_maxcnt;
229 a->aio_offset = startpos;
230 a->aio_sigevent.sigev_notify = SIGEV_SIGNAL;
231 a->aio_sigevent.sigev_signo = RT_SIGNAL_AIO;
232 a->aio_sigevent.sigev_value.sival_ptr = aio_ex;
233
234 req = SMB_VFS_PREAD_SEND(aio_ex, fsp->conn->sconn->ev_ctx,
235 fsp, smb_buf(aio_ex->outbuf.data),
236 smb_maxcnt, startpos);
237 if (req == NULL) {
238 DEBUG(0,("schedule_aio_read_and_X: aio_read failed. "
239 "Error %s\n", strerror(errno) ));
240 SMB_VFS_STRICT_UNLOCK(conn, fsp, &aio_ex->lock);
241 TALLOC_FREE(aio_ex);
242 return NT_STATUS_RETRY;
243 }
244 tevent_req_set_callback(req, aio_pread_smb1_done, aio_ex);
245
246 aio_ex->smbreq = talloc_move(aio_ex, &smbreq);
247
248 DEBUG(10,("schedule_aio_read_and_X: scheduled aio_read for file %s, "
249 "offset %.0f, len = %u (mid = %u)\n",
250 fsp_str_dbg(fsp), (double)startpos, (unsigned int)smb_maxcnt,
251 (unsigned int)aio_ex->smbreq->mid ));
252
253 return NT_STATUS_OK;
254 }
255
256 static void aio_pread_smb1_done(struct tevent_req *req)
257 {
258 struct aio_extra *aio_ex = tevent_req_callback_data(
259 req, struct aio_extra);
260 files_struct *fsp = aio_ex->fsp;
261 int outsize;
262 char *outbuf = (char *)aio_ex->outbuf.data;
263 char *data = smb_buf(outbuf);
264 ssize_t nread;
265 int err;
266
267 nread = SMB_VFS_PREAD_RECV(req, &err);
268 TALLOC_FREE(req);
269
270 DEBUG(10, ("pread_recv returned %d, err = %s\n", (int)nread,
271 (nread == -1) ? strerror(err) : "no error"));
272
273 if (fsp == NULL) {
274 DEBUG( 3, ("aio_pread_smb1_done: file closed whilst "
275 "aio outstanding (mid[%llu]).\n",
276 (unsigned long long)aio_ex->smbreq->mid));
277 TALLOC_FREE(aio_ex);
278 return;
279 }
280
281 /* Unlock now we're done. */
282 SMB_VFS_STRICT_UNLOCK(fsp->conn, fsp, &aio_ex->lock);
283
284 if (nread < 0) {
285 DEBUG( 3, ("handle_aio_read_complete: file %s nread == %d. "
286 "Error = %s\n", fsp_str_dbg(fsp), (int)nread,
287 strerror(err)));
288
289 ERROR_NT(map_nt_error_from_unix(err));
290 outsize = srv_set_message(outbuf,0,0,true);
291 } else {
292 outsize = srv_set_message(outbuf, 12, nread, False);
293 SSVAL(outbuf,smb_vwv2, 0xFFFF); /* Remaining - must be * -1. */
294 SSVAL(outbuf,smb_vwv5, nread);
295 SSVAL(outbuf,smb_vwv6, smb_offset(data,outbuf));
296 SSVAL(outbuf,smb_vwv7, ((nread >> 16) & 1));
297 SSVAL(smb_buf(outbuf), -2, nread);
298
299 aio_ex->fsp->fh->pos = aio_ex->acb.aio_offset + nread;
300 aio_ex->fsp->fh->position_information = aio_ex->fsp->fh->pos;
301
302 DEBUG( 3, ("handle_aio_read_complete file %s max=%d "
303 "nread=%d\n", fsp_str_dbg(fsp),
304 (int)aio_ex->acb.aio_nbytes, (int)nread ) );
305
306 }
307 smb_setlen(outbuf, outsize - 4);
308 show_msg(outbuf);
309 if (!srv_send_smb(aio_ex->smbreq->sconn, outbuf,
310 true, aio_ex->smbreq->seqnum+1,
311 IS_CONN_ENCRYPTED(fsp->conn), NULL)) {
312 exit_server_cleanly("handle_aio_read_complete: srv_send_smb "
313 "failed.");
314 }
315
316 DEBUG(10, ("handle_aio_read_complete: scheduled aio_read completed "
317 "for file %s, offset %.0f, len = %u\n",
318 fsp_str_dbg(fsp), (double)aio_ex->acb.aio_offset,
319 (unsigned int)nread));
320
321 TALLOC_FREE(aio_ex);
322 }
323
324 /****************************************************************************
325 Set up an aio request from a SMBwriteX call.
326 *****************************************************************************/
327
328 NTSTATUS schedule_aio_write_and_X(connection_struct *conn,
329 struct smb_request *smbreq,
330 files_struct *fsp, const char *data,
331 off_t startpos,
332 size_t numtowrite)
333 {
334 struct aio_extra *aio_ex;
335 SMB_STRUCT_AIOCB *a;
336 size_t bufsize;
337 size_t min_aio_write_size = lp_aio_write_size(SNUM(conn));
338 int ret;
339
340 if (fsp->base_fsp != NULL) {
341 /* No AIO on streams yet */
342 DEBUG(10, ("AIO on streams not yet supported\n"));
343 return NT_STATUS_RETRY;
344 }
345
346 if ((!min_aio_write_size || (numtowrite < min_aio_write_size))
347 && !SMB_VFS_AIO_FORCE(fsp)) {
348 /* Too small a write for aio request. */
349 DEBUG(10,("schedule_aio_write_and_X: write size (%u) too "
350 "small for minimum aio_write of %u\n",
351 (unsigned int)numtowrite,
352 (unsigned int)min_aio_write_size ));
353 return NT_STATUS_RETRY;
354 }
355
356 /* Only do this on non-chained and non-chaining writes not using the
357 * write cache. */
358 if (req_is_in_chain(smbreq) || (lp_write_cache_size(SNUM(conn)) != 0)) {
359 return NT_STATUS_RETRY;
360 }
361
362 if (outstanding_aio_calls >= aio_pending_size) {
363 DEBUG(3,("schedule_aio_write_and_X: Already have %d aio "
364 "activities outstanding.\n",
365 outstanding_aio_calls ));
366 DEBUG(10,("schedule_aio_write_and_X: failed to schedule "
367 "aio_write for file %s, offset %.0f, len = %u "
368 "(mid = %u)\n",
369 fsp_str_dbg(fsp), (double)startpos,
370 (unsigned int)numtowrite,
371 (unsigned int)smbreq->mid ));
372 return NT_STATUS_RETRY;
373 }
374
375 bufsize = smb_size + 6*2;
376
377 if (!(aio_ex = create_aio_extra(NULL, fsp, bufsize))) {
378 DEBUG(0,("schedule_aio_write_and_X: malloc fail.\n"));
379 return NT_STATUS_NO_MEMORY;
380 }
381 aio_ex->handle_completion = handle_aio_write_complete;
382 aio_ex->write_through = BITSETW(smbreq->vwv+7,0);
383
384 construct_reply_common_req(smbreq, (char *)aio_ex->outbuf.data);
385 srv_set_message((char *)aio_ex->outbuf.data, 6, 0, True);
386 SCVAL(aio_ex->outbuf.data,smb_vwv0,0xFF); /* Never a chained reply. */
387
388 init_strict_lock_struct(fsp, (uint64_t)smbreq->smbpid,
389 (uint64_t)startpos, (uint64_t)numtowrite, WRITE_LOCK,
390 &aio_ex->lock);
391
392 /* Take the lock until the AIO completes. */
393 if (!SMB_VFS_STRICT_LOCK(conn, fsp, &aio_ex->lock)) {
394 TALLOC_FREE(aio_ex);
395 return NT_STATUS_FILE_LOCK_CONFLICT;
396 }
397
398 a = &aio_ex->acb;
399
400 /* Now set up the aio record for the write call. */
401
402 a->aio_fildes = fsp->fh->fd;
403 a->aio_buf = discard_const_p(char, data);
404 a->aio_nbytes = numtowrite;
405 a->aio_offset = startpos;
406 a->aio_sigevent.sigev_notify = SIGEV_SIGNAL;
407 a->aio_sigevent.sigev_signo = RT_SIGNAL_AIO;
408 a->aio_sigevent.sigev_value.sival_ptr = aio_ex;
409
410 ret = SMB_VFS_AIO_WRITE(fsp, a);
411 if (ret == -1) {
412 DEBUG(3,("schedule_aio_wrote_and_X: aio_write failed. "
413 "Error %s\n", strerror(errno) ));
414 SMB_VFS_STRICT_UNLOCK(conn, fsp, &aio_ex->lock);
415 TALLOC_FREE(aio_ex);
416 return NT_STATUS_RETRY;
417 }
418
419 aio_ex->smbreq = talloc_move(aio_ex, &smbreq);
420
421 /* This should actually be improved to span the write. */
422 contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WRITE);
423 contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WRITE);
424
425 if (!aio_ex->write_through && !lp_syncalways(SNUM(fsp->conn))
426 && fsp->aio_write_behind) {
427 /* Lie to the client and immediately claim we finished the
428 * write. */
429 SSVAL(aio_ex->outbuf.data,smb_vwv2,numtowrite);
430 SSVAL(aio_ex->outbuf.data,smb_vwv4,(numtowrite>>16)&1);
431 show_msg((char *)aio_ex->outbuf.data);
432 if (!srv_send_smb(aio_ex->smbreq->sconn,
433 (char *)aio_ex->outbuf.data,
434 true, aio_ex->smbreq->seqnum+1,
435 IS_CONN_ENCRYPTED(fsp->conn),
436 &aio_ex->smbreq->pcd)) {
437 exit_server_cleanly("schedule_aio_write_and_X: "
438 "srv_send_smb failed.");
439 }
440 DEBUG(10,("schedule_aio_write_and_X: scheduled aio_write "
441 "behind for file %s\n", fsp_str_dbg(fsp)));
442 }
443
444 DEBUG(10,("schedule_aio_write_and_X: scheduled aio_write for file "
445 "%s, offset %.0f, len = %u (mid = %u) "
446 "outstanding_aio_calls = %d\n",
447 fsp_str_dbg(fsp), (double)startpos, (unsigned int)numtowrite,
448 (unsigned int)aio_ex->smbreq->mid, outstanding_aio_calls ));
449
450 return NT_STATUS_OK;
451 }
452
453 bool cancel_smb2_aio(struct smb_request *smbreq)
454 {
455 struct smbd_smb2_request *smb2req = smbreq->smb2req;
456 struct aio_extra *aio_ex = NULL;
457 int ret;
458
459 if (smb2req) {
460 aio_ex = talloc_get_type(smbreq->async_priv,
461 struct aio_extra);
462 }
463
464 if (aio_ex == NULL) {
465 return false;
466 }
467
468 if (aio_ex->fsp == NULL) {
469 return false;
470 }
471
472 ret = SMB_VFS_AIO_CANCEL(aio_ex->fsp, &aio_ex->acb);
473 if (ret != AIO_CANCELED) {
474 return false;
475 }
476
477 return true;
478 }
479
480 /****************************************************************************
481 Set up an aio request from a SMB2 read call.
482 *****************************************************************************/
483
484 NTSTATUS schedule_smb2_aio_read(connection_struct *conn,
485 struct smb_request *smbreq,
486 files_struct *fsp,
487 TALLOC_CTX *ctx,
488 DATA_BLOB *preadbuf,
489 off_t startpos,
490 size_t smb_maxcnt)
491 {
492 struct aio_extra *aio_ex;
493 SMB_STRUCT_AIOCB *a;
494 size_t min_aio_read_size = lp_aio_read_size(SNUM(conn));
495 int ret;
496
497 if (fsp->base_fsp != NULL) {
498 /* No AIO on streams yet */
499 DEBUG(10, ("AIO on streams not yet supported\n"));
500 return NT_STATUS_RETRY;
501 }
502
503 if ((!min_aio_read_size || (smb_maxcnt < min_aio_read_size))
504 && !SMB_VFS_AIO_FORCE(fsp)) {
505 /* Too small a read for aio request. */
506 DEBUG(10,("smb2: read size (%u) too small "
507 "for minimum aio_read of %u\n",
508 (unsigned int)smb_maxcnt,
509 (unsigned int)min_aio_read_size ));
510 return NT_STATUS_RETRY;
511 }
512
513 /* Only do this on reads not using the write cache. */
514 if (lp_write_cache_size(SNUM(conn)) != 0) {
515 return NT_STATUS_RETRY;
516 }
517
518 if (outstanding_aio_calls >= aio_pending_size) {
519 DEBUG(10,("smb2: Already have %d aio "
520 "activities outstanding.\n",
521 outstanding_aio_calls ));
522 return NT_STATUS_RETRY;
523 }
524
525 /* Create the out buffer. */
526 *preadbuf = data_blob_talloc(ctx, NULL, smb_maxcnt);
527 if (preadbuf->data == NULL) {
528 return NT_STATUS_NO_MEMORY;
529 }
530
531 if (!(aio_ex = create_aio_extra(smbreq->smb2req, fsp, 0))) {
532 return NT_STATUS_NO_MEMORY;
533 }
534 aio_ex->handle_completion = handle_aio_smb2_read_complete;
535
536 init_strict_lock_struct(fsp, (uint64_t)smbreq->smbpid,
537 (uint64_t)startpos, (uint64_t)smb_maxcnt, READ_LOCK,
538 &aio_ex->lock);
539
540 /* Take the lock until the AIO completes. */
541 if (!SMB_VFS_STRICT_LOCK(conn, fsp, &aio_ex->lock)) {
542 TALLOC_FREE(aio_ex);
543 return NT_STATUS_FILE_LOCK_CONFLICT;
544 }
545
546 a = &aio_ex->acb;
547
548 /* Now set up the aio record for the read call. */
549
550 a->aio_fildes = fsp->fh->fd;
551 a->aio_buf = preadbuf->data;
552 a->aio_nbytes = smb_maxcnt;
553 a->aio_offset = startpos;
554 a->aio_sigevent.sigev_notify = SIGEV_SIGNAL;
555 a->aio_sigevent.sigev_signo = RT_SIGNAL_AIO;
556 a->aio_sigevent.sigev_value.sival_ptr = aio_ex;
557
558 ret = SMB_VFS_AIO_READ(fsp, a);
559 if (ret == -1) {
560 DEBUG(0,("smb2: aio_read failed. "
561 "Error %s\n", strerror(errno) ));
562 SMB_VFS_STRICT_UNLOCK(conn, fsp, &aio_ex->lock);
563 TALLOC_FREE(aio_ex);
564 return NT_STATUS_RETRY;
565 }
566
567 /* We don't need talloc_move here as both aio_ex and
568 * smbreq are children of smbreq->smb2req. */
569 aio_ex->smbreq = smbreq;
570 smbreq->async_priv = aio_ex;
571
572 DEBUG(10,("smb2: scheduled aio_read for file %s, "
573 "offset %.0f, len = %u (mid = %u)\n",
574 fsp_str_dbg(fsp), (double)startpos, (unsigned int)smb_maxcnt,
575 (unsigned int)aio_ex->smbreq->mid ));
576
577 return NT_STATUS_OK;
578 }
579
580 /****************************************************************************
581 Set up an aio request from a SMB2write call.
582 *****************************************************************************/
583
584 NTSTATUS schedule_aio_smb2_write(connection_struct *conn,
585 struct smb_request *smbreq,
586 files_struct *fsp,
587 uint64_t in_offset,
588 DATA_BLOB in_data,
589 bool write_through)
590 {
591 struct aio_extra *aio_ex = NULL;
592 SMB_STRUCT_AIOCB *a = NULL;
593 size_t min_aio_write_size = lp_aio_write_size(SNUM(conn));
594 int ret;
595
596 if (fsp->base_fsp != NULL) {
597 /* No AIO on streams yet */
598 DEBUG(10, ("AIO on streams not yet supported\n"));
599 return NT_STATUS_RETRY;
600 }
601
602 if ((!min_aio_write_size || (in_data.length < min_aio_write_size))
603 && !SMB_VFS_AIO_FORCE(fsp)) {
604 /* Too small a write for aio request. */
605 DEBUG(10,("smb2: write size (%u) too "
606 "small for minimum aio_write of %u\n",
607 (unsigned int)in_data.length,
608 (unsigned int)min_aio_write_size ));
609 return NT_STATUS_RETRY;
610 }
611
612 /* Only do this on writes not using the write cache. */
613 if (lp_write_cache_size(SNUM(conn)) != 0) {
614 return NT_STATUS_RETRY;
615 }
616
617 if (outstanding_aio_calls >= aio_pending_size) {
618 DEBUG(3,("smb2: Already have %d aio "
619 "activities outstanding.\n",
620 outstanding_aio_calls ));
621 return NT_STATUS_RETRY;
622 }
623
624 if (!(aio_ex = create_aio_extra(smbreq->smb2req, fsp, 0))) {
625 return NT_STATUS_NO_MEMORY;
626 }
627
628 aio_ex->handle_completion = handle_aio_smb2_write_complete;
629 aio_ex->write_through = write_through;
630
631 init_strict_lock_struct(fsp, (uint64_t)smbreq->smbpid,
632 in_offset, (uint64_t)in_data.length, WRITE_LOCK,
633 &aio_ex->lock);
634
635 /* Take the lock until the AIO completes. */
636 if (!SMB_VFS_STRICT_LOCK(conn, fsp, &aio_ex->lock)) {
637 TALLOC_FREE(aio_ex);
638 return NT_STATUS_FILE_LOCK_CONFLICT;
639 }
640
641 a = &aio_ex->acb;
642
643 /* Now set up the aio record for the write call. */
644
645 a->aio_fildes = fsp->fh->fd;
646 a->aio_buf = in_data.data;
647 a->aio_nbytes = in_data.length;
648 a->aio_offset = in_offset;
649 a->aio_sigevent.sigev_notify = SIGEV_SIGNAL;
650 a->aio_sigevent.sigev_signo = RT_SIGNAL_AIO;
651 a->aio_sigevent.sigev_value.sival_ptr = aio_ex;
652
653 ret = SMB_VFS_AIO_WRITE(fsp, a);
654 if (ret == -1) {
655 DEBUG(3,("smb2: aio_write failed. "
656 "Error %s\n", strerror(errno) ));
657 SMB_VFS_STRICT_UNLOCK(conn, fsp, &aio_ex->lock);
658 TALLOC_FREE(aio_ex);
659 return NT_STATUS_RETRY;
660 }
661
662 /* We don't need talloc_move here as both aio_ex and
663 * smbreq are children of smbreq->smb2req. */
664 aio_ex->smbreq = smbreq;
665 smbreq->async_priv = aio_ex;
666
667 /* This should actually be improved to span the write. */
668 contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WRITE);
669 contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WRITE);
670
671 /*
672 * We don't want to do write behind due to ownership
673 * issues of the request structs. Maybe add it if I
674 * figure those out. JRA.
675 */
676
677 DEBUG(10,("smb2: scheduled aio_write for file "
678 "%s, offset %.0f, len = %u (mid = %u) "
679 "outstanding_aio_calls = %d\n",
680 fsp_str_dbg(fsp),
681 (double)in_offset,
682 (unsigned int)in_data.length,
683 (unsigned int)aio_ex->smbreq->mid,
684 outstanding_aio_calls ));
685
686 return NT_STATUS_OK;
687 }
688
689 /****************************************************************************
690 Complete the read and return the data or error back to the client.
691 Returns errno or zero if all ok.
692 *****************************************************************************/
693
694 static int handle_aio_read_complete(struct aio_extra *aio_ex, int errcode)
695 {
696 int outsize;
697 char *outbuf = (char *)aio_ex->outbuf.data;
698 char *data = smb_buf(outbuf);
699 ssize_t nread = SMB_VFS_AIO_RETURN(aio_ex->fsp,&aio_ex->acb);
700
701 if (nread < 0) {
702 /* We're relying here on the fact that if the fd is
703 closed then the aio will complete and aio_return
704 will return an error. Hopefully this is
705 true.... JRA. */
706
707 DEBUG( 3,( "handle_aio_read_complete: file %s nread == %d. "
708 "Error = %s\n",
709 fsp_str_dbg(aio_ex->fsp), (int)nread, strerror(errcode)));
710
711 ERROR_NT(map_nt_error_from_unix(errcode));
712 outsize = srv_set_message(outbuf,0,0,true);
713 } else {
714 outsize = srv_set_message(outbuf,12,nread,False);
715 SSVAL(outbuf,smb_vwv2,0xFFFF); /* Remaining - must be * -1. */
716 SSVAL(outbuf,smb_vwv5,nread);
717 SSVAL(outbuf,smb_vwv6,smb_offset(data,outbuf));
718 SSVAL(outbuf,smb_vwv7,((nread >> 16) & 1));
719 SSVAL(smb_buf(outbuf),-2,nread);
720
721 aio_ex->fsp->fh->pos = aio_ex->acb.aio_offset + nread;
722 aio_ex->fsp->fh->position_information = aio_ex->fsp->fh->pos;
723
724 DEBUG( 3, ( "handle_aio_read_complete file %s max=%d "
725 "nread=%d\n",
726 fsp_str_dbg(aio_ex->fsp),
727 (int)aio_ex->acb.aio_nbytes, (int)nread ) );
728
729 }
730 smb_setlen(outbuf,outsize - 4);
731 show_msg(outbuf);
732 if (!srv_send_smb(aio_ex->smbreq->sconn, outbuf,
733 true, aio_ex->smbreq->seqnum+1,
734 IS_CONN_ENCRYPTED(aio_ex->fsp->conn), NULL)) {
735 exit_server_cleanly("handle_aio_read_complete: srv_send_smb "
736 "failed.");
737 }
738
739 DEBUG(10,("handle_aio_read_complete: scheduled aio_read completed "
740 "for file %s, offset %.0f, len = %u\n",
741 fsp_str_dbg(aio_ex->fsp), (double)aio_ex->acb.aio_offset,
742 (unsigned int)nread ));
743
744 return errcode;
745 }
746
747 /****************************************************************************
748 Complete the write and return the data or error back to the client.
749 Returns error code or zero if all ok.
750 *****************************************************************************/
751
752 static int handle_aio_write_complete(struct aio_extra *aio_ex, int errcode)
753 {
754 files_struct *fsp = aio_ex->fsp;
755 char *outbuf = (char *)aio_ex->outbuf.data;
756 ssize_t numtowrite = aio_ex->acb.aio_nbytes;
757 ssize_t nwritten = SMB_VFS_AIO_RETURN(fsp,&aio_ex->acb);
758
759 if (fsp->aio_write_behind) {
760 if (nwritten != numtowrite) {
761 if (nwritten == -1) {
762 DEBUG(5,("handle_aio_write_complete: "
763 "aio_write_behind failed ! File %s "
764 "is corrupt ! Error %s\n",
765 fsp_str_dbg(fsp), strerror(errcode)));
766 } else {
767 DEBUG(0,("handle_aio_write_complete: "
768 "aio_write_behind failed ! File %s "
769 "is corrupt ! Wanted %u bytes but "
770 "only wrote %d\n", fsp_str_dbg(fsp),
771 (unsigned int)numtowrite,
772 (int)nwritten ));
773 errcode = EIO;
774 }
775 } else {
776 DEBUG(10,("handle_aio_write_complete: "
777 "aio_write_behind completed for file %s\n",
778 fsp_str_dbg(fsp)));
779 }
780 /* TODO: should no return 0 in case of an error !!! */
781 return 0;
782 }
783
784 /* We don't need outsize or set_message here as we've already set the
785 fixed size length when we set up the aio call. */
786
787 if(nwritten == -1) {
788 DEBUG( 3,( "handle_aio_write: file %s wanted %u bytes. "
789 "nwritten == %d. Error = %s\n",
790 fsp_str_dbg(fsp), (unsigned int)numtowrite,
791 (int)nwritten, strerror(errcode) ));
792
793 ERROR_NT(map_nt_error_from_unix(errcode));
794 srv_set_message(outbuf,0,0,true);
795 } else {
796 NTSTATUS status;
797
798 SSVAL(outbuf,smb_vwv2,nwritten);
799 SSVAL(outbuf,smb_vwv4,(nwritten>>16)&1);
800 if (nwritten < (ssize_t)numtowrite) {
801 SCVAL(outbuf,smb_rcls,ERRHRD);
802 SSVAL(outbuf,smb_err,ERRdiskfull);
803 }
804
805 DEBUG(3,("handle_aio_write: %s, num=%d wrote=%d\n",
806 fsp_fnum_dbg(fsp), (int)numtowrite, (int)nwritten));
807 status = sync_file(fsp->conn,fsp, aio_ex->write_through);
808 if (!NT_STATUS_IS_OK(status)) {
809 errcode = errno;
810 ERROR_BOTH(map_nt_error_from_unix(errcode),
811 ERRHRD, ERRdiskfull);
812 srv_set_message(outbuf,0,0,true);
813 DEBUG(5,("handle_aio_write: sync_file for %s returned %s\n",
814 fsp_str_dbg(fsp), nt_errstr(status)));
815 }
816
817 aio_ex->fsp->fh->pos = aio_ex->acb.aio_offset + nwritten;
818
819 mark_file_modified(aio_ex->fsp);
820 }
821
822 show_msg(outbuf);
823 if (!srv_send_smb(aio_ex->smbreq->sconn, outbuf,
824 true, aio_ex->smbreq->seqnum+1,
825 IS_CONN_ENCRYPTED(fsp->conn),
826 NULL)) {
827 exit_server_cleanly("handle_aio_write_complete: "
828 "srv_send_smb failed.");
829 }
830
831 DEBUG(10,("handle_aio_write_complete: scheduled aio_write completed "
832 "for file %s, offset %.0f, requested %u, written = %u\n",
833 fsp_str_dbg(fsp), (double)aio_ex->acb.aio_offset,
834 (unsigned int)numtowrite, (unsigned int)nwritten ));
835
836 return errcode;
837 }
838
839 /****************************************************************************
840 Complete the read and return the data or error back to the client.
841 Returns errno or zero if all ok.
842 *****************************************************************************/
843
844 static int handle_aio_smb2_read_complete(struct aio_extra *aio_ex, int errcode)
845 {
846 NTSTATUS status;
847 struct tevent_req *subreq = aio_ex->smbreq->smb2req->subreq;
848 ssize_t nread = SMB_VFS_AIO_RETURN(aio_ex->fsp,&aio_ex->acb);
849
850 /* Common error or success code processing for async or sync
851 read returns. */
852
853 status = smb2_read_complete(subreq, nread, errcode);
854
855 if (nread > 0) {
856 aio_ex->fsp->fh->pos = aio_ex->acb.aio_offset + nread;
857 aio_ex->fsp->fh->position_information = aio_ex->fsp->fh->pos;
858 }
859
860 DEBUG(10,("smb2: scheduled aio_read completed "
861 "for file %s, offset %.0f, len = %u "
862 "(errcode = %d, NTSTATUS = %s)\n",
863 fsp_str_dbg(aio_ex->fsp),
864 (double)aio_ex->acb.aio_offset,
865 (unsigned int)nread,
866 errcode,
867 nt_errstr(status) ));
868
869 if (!NT_STATUS_IS_OK(status)) {
870 tevent_req_nterror(subreq, status);
871 return errcode;
872 }
873
874 tevent_req_done(subreq);
875 return errcode;
876 }
877
878 /****************************************************************************
879 Complete the SMB2 write and return the data or error back to the client.
880 Returns error code or zero if all ok.
881 *****************************************************************************/
882
883 static int handle_aio_smb2_write_complete(struct aio_extra *aio_ex, int errcode)
884 {
885 files_struct *fsp = aio_ex->fsp;
886 ssize_t numtowrite = aio_ex->acb.aio_nbytes;
887 ssize_t nwritten = SMB_VFS_AIO_RETURN(fsp,&aio_ex->acb);
888 struct tevent_req *subreq = aio_ex->smbreq->smb2req->subreq;
889 NTSTATUS status;
890
891 status = smb2_write_complete(subreq, nwritten, errcode);
892
893 DEBUG(10,("smb2: scheduled aio_write completed "
894 "for file %s, offset %.0f, requested %u, "
895 "written = %u (errcode = %d, NTSTATUS = %s)\n",
896 fsp_str_dbg(fsp),
897 (double)aio_ex->acb.aio_offset,
898 (unsigned int)numtowrite,
899 (unsigned int)nwritten,
900 errcode,
901 nt_errstr(status) ));
902
903 if (!NT_STATUS_IS_OK(status)) {
904 tevent_req_nterror(subreq, status);
905 return errcode;
906 }
907
908 mark_file_modified(fsp);
909
910 tevent_req_done(subreq);
911 return errcode;
912 }
913
914 /****************************************************************************
915 Handle any aio completion. Returns True if finished (and sets *perr if err
916 was non-zero), False if not.
917 *****************************************************************************/
918
919 static bool handle_aio_completed(struct aio_extra *aio_ex, int *perr)
920 {
921 files_struct *fsp = NULL;
922 int err;
923
924 if(!aio_ex) {
925 DEBUG(3, ("handle_aio_completed: Non-existing aio_ex passed\n"));
926 return false;
927 }
928
929 if (!aio_ex->fsp) {
930 DEBUG(3, ("handle_aio_completed: aio_ex->fsp == NULL\n"));
931 return false;
932 }
933
934 fsp = aio_ex->fsp;
935
936 /* Ensure the operation has really completed. */
937 err = SMB_VFS_AIO_ERROR(fsp, &aio_ex->acb);
938 if (err == EINPROGRESS) {
939 DEBUG(10,( "handle_aio_completed: operation mid %llu still in "
940 "process for file %s\n",
941 (unsigned long long)aio_ex->smbreq->mid,
942 fsp_str_dbg(aio_ex->fsp)));
943 return False;
944 }
945
946 if (err == ECANCELED) {
947 DEBUG(10,( "handle_aio_completed: operation mid %llu canceled "
948 "for file %s\n",
949 (unsigned long long)aio_ex->smbreq->mid,
950 fsp_str_dbg(aio_ex->fsp)));
951 }
952
953 /* Unlock now we're done. */
954 SMB_VFS_STRICT_UNLOCK(fsp->conn, fsp, &aio_ex->lock);
955
956 err = aio_ex->handle_completion(aio_ex, err);
957 if (err) {
958 *perr = err; /* Only save non-zero errors. */
959 }
960
961 return True;
962 }
963
964 /****************************************************************************
965 Handle any aio completion inline.
966 *****************************************************************************/
967
968 void smbd_aio_complete_aio_ex(struct aio_extra *aio_ex)
969 {
970 files_struct *fsp = NULL;
971 int ret = 0;
972
973 DEBUG(10,("smbd_aio_complete_mid: mid[%llu]\n",
974 (unsigned long long)aio_ex->smbreq->mid));
975
976 fsp = aio_ex->fsp;
977 if (fsp == NULL) {
978 /* file was closed whilst I/O was outstanding. Just
979 * ignore. */
980 DEBUG( 3,( "smbd_aio_complete_mid: file closed whilst "
981 "aio outstanding (mid[%llu]).\n",
982 (unsigned long long)aio_ex->smbreq->mid));
983 return;
984 }
985
986 if (!handle_aio_completed(aio_ex, &ret)) {
987 return;
988 }
989 }
990
991 /****************************************************************************
992 We're doing write behind and the client closed the file. Wait up to 45
993 seconds (my arbitrary choice) for the aio to complete. Return 0 if all writes
994 completed, errno to return if not.
995 *****************************************************************************/
996
997 #define SMB_TIME_FOR_AIO_COMPLETE_WAIT 45
998
999 int wait_for_aio_completion(files_struct *fsp)
1000 {
1001 struct aio_extra *aio_ex;
1002 const SMB_STRUCT_AIOCB **aiocb_list;
1003 int aio_completion_count = 0;
1004 time_t start_time = time_mono(NULL);
1005 int seconds_left;
1006
1007 for (seconds_left = SMB_TIME_FOR_AIO_COMPLETE_WAIT;
1008 seconds_left >= 0;) {
1009 int err = 0;
1010 int i;
1011 struct timespec ts;
1012
1013 aio_completion_count = 0;
1014 for( aio_ex = aio_list_head; aio_ex; aio_ex = aio_ex->next) {
1015 if (aio_ex->fsp == fsp) {
1016 aio_completion_count++;
1017 }
1018 }
1019
1020 if (!aio_completion_count) {
1021 return 0;
1022 }
1023
1024 DEBUG(3,("wait_for_aio_completion: waiting for %d aio events "
1025 "to complete.\n", aio_completion_count ));
1026
1027 aiocb_list = SMB_MALLOC_ARRAY(const SMB_STRUCT_AIOCB *,
1028 aio_completion_count);
1029 if (!aiocb_list) {
1030 return ENOMEM;
1031 }
1032
1033 for( i = 0, aio_ex = aio_list_head;
1034 aio_ex;
1035 aio_ex = aio_ex->next) {
1036 if (aio_ex->fsp == fsp) {
1037 aiocb_list[i++] = &aio_ex->acb;
1038 }
1039 }
1040
1041 /* Now wait up to seconds_left for completion. */
1042 ts.tv_sec = seconds_left;
1043 ts.tv_nsec = 0;
1044
1045 DEBUG(10,("wait_for_aio_completion: %d events, doing a wait "
1046 "of %d seconds.\n",
1047 aio_completion_count, seconds_left ));
1048
1049 err = SMB_VFS_AIO_SUSPEND(fsp, aiocb_list,
1050 aio_completion_count, &ts);
1051
1052 DEBUG(10,("wait_for_aio_completion: returned err = %d, "
1053 "errno = %s\n", err, strerror(errno) ));
1054
1055 if (err == -1 && errno == EAGAIN) {
1056 DEBUG(0,("wait_for_aio_completion: aio_suspend timed "
1057 "out waiting for %d events after a wait of "
1058 "%d seconds\n", aio_completion_count,
1059 seconds_left));
1060 /* Timeout. */
1061 SAFE_FREE(aiocb_list);
1062 /* We're hosed here - IO may complete
1063 and trample over memory if we free
1064 the aio_ex struct, but if we don't
1065 we leak IO requests. I think smb_panic()
1066 if the right thing to do here. JRA.
1067 */
1068 smb_panic("AIO suspend timed out - cannot continue.");
1069 return EIO;
1070 }
1071
1072 /* One or more events might have completed - process them if
1073 * so. */
1074 for( i = 0; i < aio_completion_count; i++) {
1075 aio_ex = (struct aio_extra *)aiocb_list[i]->aio_sigevent.sigev_value.sival_ptr;
1076
1077 if (!handle_aio_completed(aio_ex, &err)) {
1078 continue;
1079 }
1080 TALLOC_FREE(aio_ex);
1081 }
1082
1083 SAFE_FREE(aiocb_list);
1084 seconds_left = SMB_TIME_FOR_AIO_COMPLETE_WAIT
1085 - (time_mono(NULL) - start_time);
1086 }
1087
1088 /* We timed out - we don't know why. Return ret if already an error,
1089 * else EIO. */
1090 DEBUG(10,("wait_for_aio_completion: aio_suspend timed out waiting "
1091 "for %d events\n",
1092 aio_completion_count));
1093
1094 return EIO;
1095 }
1096
1097 #else
1098
1099 bool initialize_async_io_handler(void)
1100 {
1101 return false;
1102 }
1103
1104 NTSTATUS schedule_aio_read_and_X(connection_struct *conn,
1105 struct smb_request *smbreq,
1106 files_struct *fsp, off_t startpos,
1107 size_t smb_maxcnt)
1108 {
1109 return NT_STATUS_RETRY;
1110 }
1111
1112 NTSTATUS schedule_aio_write_and_X(connection_struct *conn,
1113 struct smb_request *smbreq,
1114 files_struct *fsp, const char *data,
1115 off_t startpos,
1116 size_t numtowrite)
1117 {
1118 return NT_STATUS_RETRY;
1119 }
1120
1121 bool cancel_smb2_aio(struct smb_request *smbreq)
1122 {
1123 return false;
1124 }
1125
1126 NTSTATUS schedule_smb2_aio_read(connection_struct *conn,
1127 struct smb_request *smbreq,
1128 files_struct *fsp,
1129 TALLOC_CTX *ctx,
1130 DATA_BLOB *preadbuf,
1131 off_t startpos,
1132 size_t smb_maxcnt)
1133 {
1134 return NT_STATUS_RETRY;
1135 }
1136
1137 NTSTATUS schedule_aio_smb2_write(connection_struct *conn,
1138 struct smb_request *smbreq,
1139 files_struct *fsp,
1140 uint64_t in_offset,
1141 DATA_BLOB in_data,
1142 bool write_through)
1143 {
1144 return NT_STATUS_RETRY;
1145 }
1146
1147 int wait_for_aio_completion(files_struct *fsp)
1148 {
1149 return 0;
1150 }
1151
1152 void smbd_aio_complete_mid(uint64_t mid);
1153
1154 #endif