source3/smbd/aio.c

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