source3/modules/vfs_aio_fork.c

   1 /*
   2  * Simulate the Posix AIO using mmap/fork
   3  *
   4  * Copyright (C) Volker Lendecke 2008
   5  * Copyright (C) Jeremy Allison 2010
   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 2 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, write to the Free Software
  19  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #include "includes.h"
  23
  24 struct mmap_area {
  25         size_t size;
  26         volatile void *ptr;
  27 };
  28
  29 static int mmap_area_destructor(struct mmap_area *area)
  30 {
  31         munmap((void *)area->ptr, area->size);
  32         return 0;
  33 }
  34
  35 static struct mmap_area *mmap_area_init(TALLOC_CTX *mem_ctx, size_t size)
  36 {
  37         struct mmap_area *result;
  38         int fd;
  39
  40         result = talloc(mem_ctx, struct mmap_area);
  41         if (result == NULL) {
  42                 DEBUG(0, ("talloc failed\n"));
  43                 goto fail;
  44         }
  45
  46         fd = open("/dev/zero", O_RDWR);
  47         if (fd == -1) {
  48                 DEBUG(3, ("open(\"/dev/zero\") failed: %s\n",
  49                           strerror(errno)));
  50                 goto fail;
  51         }
  52
  53         result->ptr = mmap(NULL, size, PROT_READ|PROT_WRITE,
  54                            MAP_SHARED|MAP_FILE, fd, 0);
  55         if (result->ptr == MAP_FAILED) {
  56                 DEBUG(1, ("mmap failed: %s\n", strerror(errno)));
  57                 goto fail;
  58         }
  59
  60         close(fd);
  61
  62         result->size = size;
  63         talloc_set_destructor(result, mmap_area_destructor);
  64
  65         return result;
  66
  67 fail:
  68         TALLOC_FREE(result);
  69         return NULL;
  70 }
  71
  72 struct rw_cmd {
  73         size_t n;
  74         SMB_OFF_T offset;
  75         bool read_cmd;
  76 };
  77
  78 struct rw_ret {
  79         ssize_t size;
  80         int ret_errno;
  81 };
  82
  83 struct aio_child_list;
  84
  85 struct aio_child {
  86         struct aio_child *prev, *next;
  87         struct aio_child_list *list;
  88         SMB_STRUCT_AIOCB *aiocb;
  89         pid_t pid;
  90         int sockfd;
  91         struct fd_event *sock_event;
  92         struct rw_ret retval;
  93         struct mmap_area *map;  /* ==NULL means write request */
  94         bool dont_delete;       /* Marked as in use since last cleanup */
  95         bool cancelled;
  96         bool read_cmd;
  97 };
  98
  99 struct aio_child_list {
 100         struct aio_child *children;
 101         struct timed_event *cleanup_event;
 102 };
 103
 104 static void free_aio_children(void **p)
 105 {
 106         TALLOC_FREE(*p);
 107 }
 108
 109 static ssize_t read_fd(int fd, void *ptr, size_t nbytes, int *recvfd)
 110 {
 111         struct msghdr msg;
 112         struct iovec iov[1];
 113         ssize_t n;
 114 #ifndef HAVE_MSGHDR_MSG_CONTROL
 115         int newfd;
 116 #endif
 117
 118 #ifdef  HAVE_MSGHDR_MSG_CONTROL
 119         union {
 120           struct cmsghdr        cm;
 121           char                          control[CMSG_SPACE(sizeof(int))];
 122         } control_un;
 123         struct cmsghdr  *cmptr;
 124
 125         msg.msg_control = control_un.control;
 126         msg.msg_controllen = sizeof(control_un.control);
 127 #else
 128 #if HAVE_MSGHDR_MSG_ACCTRIGHTS
 129         msg.msg_accrights = (caddr_t) &newfd;
 130         msg.msg_accrightslen = sizeof(int);
 131 #else
 132 #error Can not pass file descriptors
 133 #endif
 134 #endif
 135
 136         msg.msg_name = NULL;
 137         msg.msg_namelen = 0;
 138
 139         iov[0].iov_base = (void *)ptr;
 140         iov[0].iov_len = nbytes;
 141         msg.msg_iov = iov;
 142         msg.msg_iovlen = 1;
 143
 144         if ( (n = recvmsg(fd, &msg, 0)) <= 0) {
 145                 return(n);
 146         }
 147
 148 #ifdef  HAVE_MSGHDR_MSG_CONTROL
 149         if ((cmptr = CMSG_FIRSTHDR(&msg)) != NULL
 150             && cmptr->cmsg_len == CMSG_LEN(sizeof(int))) {
 151                 if (cmptr->cmsg_level != SOL_SOCKET) {
 152                         DEBUG(10, ("control level != SOL_SOCKET"));
 153                         errno = EINVAL;
 154                         return -1;
 155                 }
 156                 if (cmptr->cmsg_type != SCM_RIGHTS) {
 157                         DEBUG(10, ("control type != SCM_RIGHTS"));
 158                         errno = EINVAL;
 159                         return -1;
 160                 }
 161                 *recvfd = *((int *) CMSG_DATA(cmptr));
 162         } else {
 163                 *recvfd = -1;           /* descriptor was not passed */
 164         }
 165 #else
 166         if (msg.msg_accrightslen == sizeof(int)) {
 167                 *recvfd = newfd;
 168         }
 169         else {
 170                 *recvfd = -1;           /* descriptor was not passed */
 171         }
 172 #endif
 173
 174         return(n);
 175 }
 176
 177 static ssize_t write_fd(int fd, void *ptr, size_t nbytes, int sendfd)
 178 {
 179         struct msghdr   msg;
 180         struct iovec    iov[1];
 181
 182 #ifdef  HAVE_MSGHDR_MSG_CONTROL
 183         union {
 184                 struct cmsghdr  cm;
 185                 char control[CMSG_SPACE(sizeof(int))];
 186         } control_un;
 187         struct cmsghdr  *cmptr;
 188
 189         ZERO_STRUCT(msg);
 190         ZERO_STRUCT(control_un);
 191
 192         msg.msg_control = control_un.control;
 193         msg.msg_controllen = sizeof(control_un.control);
 194
 195         cmptr = CMSG_FIRSTHDR(&msg);
 196         cmptr->cmsg_len = CMSG_LEN(sizeof(int));
 197         cmptr->cmsg_level = SOL_SOCKET;
 198         cmptr->cmsg_type = SCM_RIGHTS;
 199         *((int *) CMSG_DATA(cmptr)) = sendfd;
 200 #else
 201         ZERO_STRUCT(msg);
 202         msg.msg_accrights = (caddr_t) &sendfd;
 203         msg.msg_accrightslen = sizeof(int);
 204 #endif
 205
 206         msg.msg_name = NULL;
 207         msg.msg_namelen = 0;
 208
 209         ZERO_STRUCT(iov);
 210         iov[0].iov_base = (void *)ptr;
 211         iov[0].iov_len = nbytes;
 212         msg.msg_iov = iov;
 213         msg.msg_iovlen = 1;
 214
 215         return (sendmsg(fd, &msg, 0));
 216 }
 217
 218 static void aio_child_cleanup(struct event_context *event_ctx,
 219                               struct timed_event *te,
 220                               struct timeval now,
 221                               void *private_data)
 222 {
 223         struct aio_child_list *list = talloc_get_type_abort(
 224                 private_data, struct aio_child_list);
 225         struct aio_child *child, *next;
 226
 227         TALLOC_FREE(list->cleanup_event);
 228
 229         for (child = list->children; child != NULL; child = next) {
 230                 next = child->next;
 231
 232                 if (child->aiocb != NULL) {
 233                         DEBUG(10, ("child %d currently active\n",
 234                                    (int)child->pid));
 235                         continue;
 236                 }
 237
 238                 if (child->dont_delete) {
 239                         DEBUG(10, ("Child %d was active since last cleanup\n",
 240                                    (int)child->pid));
 241                         child->dont_delete = false;
 242                         continue;
 243                 }
 244
 245                 DEBUG(10, ("Child %d idle for more than 30 seconds, "
 246                            "deleting\n", (int)child->pid));
 247
 248                 TALLOC_FREE(child);
 249         }
 250
 251         if (list->children != NULL) {
 252                 /*
 253                  * Re-schedule the next cleanup round
 254                  */
 255                 list->cleanup_event = event_add_timed(smbd_event_context(), list,
 256                                                       timeval_add(&now, 30, 0),
 257                                                       aio_child_cleanup, list);
 258
 259         }
 260 }
 261
 262 static struct aio_child_list *init_aio_children(struct vfs_handle_struct *handle)
 263 {
 264         struct aio_child_list *data = NULL;
 265
 266         if (SMB_VFS_HANDLE_TEST_DATA(handle)) {
 267                 SMB_VFS_HANDLE_GET_DATA(handle, data, struct aio_child_list,
 268                                         return NULL);
 269         }
 270
 271         if (data == NULL) {
 272                 data = TALLOC_ZERO_P(NULL, struct aio_child_list);
 273                 if (data == NULL) {
 274                         return NULL;
 275                 }
 276         }
 277
 278         /*
 279          * Regardless of whether the child_list had been around or not, make
 280          * sure that we have a cleanup timed event. This timed event will
 281          * delete itself when it finds that no children are around anymore.
 282          */
 283
 284         if (data->cleanup_event == NULL) {
 285                 data->cleanup_event = event_add_timed(smbd_event_context(), data,
 286                                                       timeval_current_ofs(30, 0),
 287                                                       aio_child_cleanup, data);
 288                 if (data->cleanup_event == NULL) {
 289                         TALLOC_FREE(data);
 290                         return NULL;
 291                 }
 292         }
 293
 294         if (!SMB_VFS_HANDLE_TEST_DATA(handle)) {
 295                 SMB_VFS_HANDLE_SET_DATA(handle, data, free_aio_children,
 296                                         struct aio_child_list, return False);
 297         }
 298
 299         return data;
 300 }
 301
 302 static void aio_child_loop(int sockfd, struct mmap_area *map)
 303 {
 304         while (true) {
 305                 int fd = -1;
 306                 ssize_t ret;
 307                 struct rw_cmd cmd_struct;
 308                 struct rw_ret ret_struct;
 309
 310                 ret = read_fd(sockfd, &cmd_struct, sizeof(cmd_struct), &fd);
 311                 if (ret != sizeof(cmd_struct)) {
 312                         DEBUG(10, ("read_fd returned %d: %s\n", (int)ret,
 313                                    strerror(errno)));
 314                         exit(1);
 315                 }
 316
 317                 DEBUG(10, ("aio_child_loop: %s %d bytes at %d from fd %d\n",
 318                            cmd_struct.read_cmd ? "read" : "write",
 319                            (int)cmd_struct.n, (int)cmd_struct.offset, fd));
 320
 321 #ifdef ENABLE_BUILD_FARM_HACKS
 322                 {
 323                         /*
 324                          * In the build farm, we want erratic behaviour for
 325                          * async I/O times
 326                          */
 327                         uint8_t randval;
 328                         unsigned msecs;
 329                         /*
 330                          * use generate_random_buffer, we just forked from a
 331                          * common parent state
 332                          */
 333                         generate_random_buffer(&randval, sizeof(randval));
 334                         msecs = randval + 20;
 335                         DEBUG(10, ("delaying for %u msecs\n", msecs));
 336                         smb_msleep(msecs);
 337                 }
 338 #endif
 339
 340
 341                 ZERO_STRUCT(ret_struct);
 342
 343                 if (cmd_struct.read_cmd) {
 344                         ret_struct.size = sys_pread(
 345                                 fd, (void *)map->ptr, cmd_struct.n,
 346                                 cmd_struct.offset);
 347 #if 0
 348 /* This breaks "make test" when run with aio_fork module. */
 349 #ifdef ENABLE_BUILD_FARM_HACKS
 350                         ret_struct.size = MAX(1, ret_struct.size * 0.9);
 351 #endif
 352 #endif
 353                 }
 354                 else {
 355                         ret_struct.size = sys_pwrite(
 356                                 fd, (void *)map->ptr, cmd_struct.n,
 357                                 cmd_struct.offset);
 358                 }
 359
 360                 DEBUG(10, ("aio_child_loop: syscall returned %d\n",
 361                            (int)ret_struct.size));
 362
 363                 if (ret_struct.size == -1) {
 364                         ret_struct.ret_errno = errno;
 365                 }
 366
 367                 /*
 368                  * Close the fd before telling our parent we're done. The
 369                  * parent might close and re-open the file very quickly, and
 370                  * with system-level share modes (GPFS) we would get an
 371                  * unjustified SHARING_VIOLATION.
 372                  */
 373                 close(fd);
 374
 375                 ret = write_data(sockfd, (char *)&ret_struct,
 376                                  sizeof(ret_struct));
 377                 if (ret != sizeof(ret_struct)) {
 378                         DEBUG(10, ("could not write ret_struct: %s\n",
 379                                    strerror(errno)));
 380                         exit(2);
 381                 }
 382         }
 383 }
 384
 385 static void handle_aio_completion(struct event_context *event_ctx,
 386                                   struct fd_event *event, uint16 flags,
 387                                   void *p)
 388 {
 389         struct aio_extra *aio_ex = NULL;
 390         struct aio_child *child = (struct aio_child *)p;
 391
 392         DEBUG(10, ("handle_aio_completion called with flags=%d\n", flags));
 393
 394         if ((flags & EVENT_FD_READ) == 0) {
 395                 return;
 396         }
 397
 398         if (!NT_STATUS_IS_OK(read_data(child->sockfd,
 399                                        (char *)&child->retval,
 400                                        sizeof(child->retval)))) {
 401                 DEBUG(0, ("aio child %d died\n", (int)child->pid));
 402                 child->retval.size = -1;
 403                 child->retval.ret_errno = EIO;
 404         }
 405
 406         if (child->cancelled) {
 407                 child->aiocb = NULL;
 408                 child->cancelled = false;
 409                 return;
 410         }
 411
 412         if (child->read_cmd && (child->retval.size > 0)) {
 413                 SMB_ASSERT(child->retval.size <= child->aiocb->aio_nbytes);
 414                 memcpy((void *)child->aiocb->aio_buf, (void *)child->map->ptr,
 415                        child->retval.size);
 416         }
 417
 418         aio_ex = (struct aio_extra *)child->aiocb->aio_sigevent.sigev_value.sival_ptr;
 419         smbd_aio_complete_aio_ex(aio_ex);
 420 }
 421
 422 static int aio_child_destructor(struct aio_child *child)
 423 {
 424         SMB_ASSERT((child->aiocb == NULL) || child->cancelled);
 425         close(child->sockfd);
 426         DLIST_REMOVE(child->list->children, child);
 427         return 0;
 428 }
 429
 430 /*
 431  * We have to close all fd's in open files, we might incorrectly hold a system
 432  * level share mode on a file.
 433  */
 434
 435 static struct files_struct *close_fsp_fd(struct files_struct *fsp,
 436                                          void *private_data)
 437 {
 438         if ((fsp->fh != NULL) && (fsp->fh->fd != -1)) {
 439                 close(fsp->fh->fd);
 440                 fsp->fh->fd = -1;
 441         }
 442         return NULL;
 443 }
 444
 445 static NTSTATUS create_aio_child(struct smbd_server_connection *sconn,
 446                                  struct aio_child_list *children,
 447                                  size_t map_size,
 448                                  struct aio_child **presult)
 449 {
 450         struct aio_child *result;
 451         int fdpair[2];
 452         NTSTATUS status;
 453
 454         fdpair[0] = fdpair[1] = -1;
 455
 456         result = TALLOC_ZERO_P(children, struct aio_child);
 457         NT_STATUS_HAVE_NO_MEMORY(result);
 458
 459         if (socketpair(AF_UNIX, SOCK_STREAM, 0, fdpair) == -1) {
 460                 status = map_nt_error_from_unix(errno);
 461                 DEBUG(10, ("socketpair() failed: %s\n", strerror(errno)));
 462                 goto fail;
 463         }
 464
 465         DEBUG(10, ("fdpair = %d/%d\n", fdpair[0], fdpair[1]));
 466
 467         result->map = mmap_area_init(result, map_size);
 468         if (result->map == NULL) {
 469                 status = map_nt_error_from_unix(errno);
 470                 DEBUG(0, ("Could not create mmap area\n"));
 471                 goto fail;
 472         }
 473
 474         result->pid = sys_fork();
 475         if (result->pid == -1) {
 476                 status = map_nt_error_from_unix(errno);
 477                 DEBUG(0, ("fork failed: %s\n", strerror(errno)));
 478                 goto fail;
 479         }
 480
 481         if (result->pid == 0) {
 482                 close(fdpair[0]);
 483                 result->sockfd = fdpair[1];
 484                 files_forall(sconn, close_fsp_fd, NULL);
 485                 aio_child_loop(result->sockfd, result->map);
 486         }
 487
 488         DEBUG(10, ("Child %d created\n", result->pid));
 489
 490         result->sockfd = fdpair[0];
 491         close(fdpair[1]);
 492
 493         result->sock_event = event_add_fd(smbd_event_context(), result,
 494                                           result->sockfd, EVENT_FD_READ,
 495                                           handle_aio_completion,
 496                                           result);
 497         if (result->sock_event == NULL) {
 498                 status = NT_STATUS_NO_MEMORY;
 499                 DEBUG(0, ("event_add_fd failed\n"));
 500                 goto fail;
 501         }
 502
 503         result->list = children;
 504         DLIST_ADD(children->children, result);
 505
 506         talloc_set_destructor(result, aio_child_destructor);
 507
 508         *presult = result;
 509
 510         return NT_STATUS_OK;
 511
 512  fail:
 513         if (fdpair[0] != -1) close(fdpair[0]);
 514         if (fdpair[1] != -1) close(fdpair[1]);
 515         TALLOC_FREE(result);
 516
 517         return status;
 518 }
 519
 520 static NTSTATUS get_idle_child(struct vfs_handle_struct *handle,
 521                                struct aio_child **pchild)
 522 {
 523         struct aio_child_list *children;
 524         struct aio_child *child;
 525         NTSTATUS status;
 526
 527         children = init_aio_children(handle);
 528         if (children == NULL) {
 529                 return NT_STATUS_NO_MEMORY;
 530         }
 531
 532         for (child = children->children; child != NULL; child = child->next) {
 533                 if (child->aiocb == NULL) {
 534                         /* idle */
 535                         break;
 536                 }
 537         }
 538
 539         if (child == NULL) {
 540                 DEBUG(10, ("no idle child found, creating new one\n"));
 541
 542                 status = create_aio_child(handle->conn->sconn, children,
 543                                           128*1024, &child);
 544                 if (!NT_STATUS_IS_OK(status)) {
 545                         DEBUG(10, ("create_aio_child failed: %s\n",
 546                                    nt_errstr(status)));
 547                         return status;
 548                 }
 549         }
 550
 551         child->dont_delete = true;
 552
 553         *pchild = child;
 554         return NT_STATUS_OK;
 555 }
 556
 557 static int aio_fork_read(struct vfs_handle_struct *handle,
 558                          struct files_struct *fsp, SMB_STRUCT_AIOCB *aiocb)
 559 {
 560         struct aio_child *child;
 561         struct rw_cmd cmd;
 562         ssize_t ret;
 563         NTSTATUS status;
 564
 565         if (aiocb->aio_nbytes > 128*1024) {
 566                 /* TODO: support variable buffers */
 567                 errno = EINVAL;
 568                 return -1;
 569         }
 570
 571         status = get_idle_child(handle, &child);
 572         if (!NT_STATUS_IS_OK(status)) {
 573                 DEBUG(10, ("Could not get an idle child\n"));
 574                 return -1;
 575         }
 576
 577         child->read_cmd = true;
 578         child->aiocb = aiocb;
 579         child->retval.ret_errno = EINPROGRESS;
 580
 581         ZERO_STRUCT(cmd);
 582         cmd.n = aiocb->aio_nbytes;
 583         cmd.offset = aiocb->aio_offset;
 584         cmd.read_cmd = child->read_cmd;
 585
 586         DEBUG(10, ("sending fd %d to child %d\n", fsp->fh->fd,
 587                    (int)child->pid));
 588
 589         ret = write_fd(child->sockfd, &cmd, sizeof(cmd), fsp->fh->fd);
 590         if (ret == -1) {
 591                 DEBUG(10, ("write_fd failed: %s\n", strerror(errno)));
 592                 return -1;
 593         }
 594
 595         return 0;
 596 }
 597
 598 static int aio_fork_write(struct vfs_handle_struct *handle,
 599                           struct files_struct *fsp, SMB_STRUCT_AIOCB *aiocb)
 600 {
 601         struct aio_child *child;
 602         struct rw_cmd cmd;
 603         ssize_t ret;
 604         NTSTATUS status;
 605
 606         if (aiocb->aio_nbytes > 128*1024) {
 607                 /* TODO: support variable buffers */
 608                 errno = EINVAL;
 609                 return -1;
 610         }
 611
 612         status = get_idle_child(handle, &child);
 613         if (!NT_STATUS_IS_OK(status)) {
 614                 DEBUG(10, ("Could not get an idle child\n"));
 615                 return -1;
 616         }
 617
 618         child->read_cmd = false;
 619         child->aiocb = aiocb;
 620         child->retval.ret_errno = EINPROGRESS;
 621
 622         memcpy((void *)child->map->ptr, (void *)aiocb->aio_buf,
 623                aiocb->aio_nbytes);
 624
 625         ZERO_STRUCT(cmd);
 626         cmd.n = aiocb->aio_nbytes;
 627         cmd.offset = aiocb->aio_offset;
 628         cmd.read_cmd = child->read_cmd;
 629
 630         DEBUG(10, ("sending fd %d to child %d\n", fsp->fh->fd,
 631                    (int)child->pid));
 632
 633         ret = write_fd(child->sockfd, &cmd, sizeof(cmd), fsp->fh->fd);
 634         if (ret == -1) {
 635                 DEBUG(10, ("write_fd failed: %s\n", strerror(errno)));
 636                 return -1;
 637         }
 638
 639         return 0;
 640 }
 641
 642 static struct aio_child *aio_fork_find_child(struct vfs_handle_struct *handle,
 643                                              SMB_STRUCT_AIOCB *aiocb)
 644 {
 645         struct aio_child_list *children;
 646         struct aio_child *child;
 647
 648         children = init_aio_children(handle);
 649         if (children == NULL) {
 650                 return NULL;
 651         }
 652
 653         for (child = children->children; child != NULL; child = child->next) {
 654                 if (child->aiocb == aiocb) {
 655                         return child;
 656                 }
 657         }
 658
 659         return NULL;
 660 }
 661
 662 static ssize_t aio_fork_return_fn(struct vfs_handle_struct *handle,
 663                                   struct files_struct *fsp,
 664                                   SMB_STRUCT_AIOCB *aiocb)
 665 {
 666         struct aio_child *child = aio_fork_find_child(handle, aiocb);
 667
 668         if (child == NULL) {
 669                 errno = EINVAL;
 670                 DEBUG(0, ("returning EINVAL\n"));
 671                 return -1;
 672         }
 673
 674         child->aiocb = NULL;
 675
 676         if (child->retval.size == -1) {
 677                 errno = child->retval.ret_errno;
 678         }
 679
 680         return child->retval.size;
 681 }
 682
 683 static int aio_fork_cancel(struct vfs_handle_struct *handle,
 684                            struct files_struct *fsp,
 685                            SMB_STRUCT_AIOCB *aiocb)
 686 {
 687         struct aio_child_list *children;
 688         struct aio_child *child;
 689
 690         children = init_aio_children(handle);
 691         if (children == NULL) {
 692                 errno = EINVAL;
 693                 return -1;
 694         }
 695
 696         for (child = children->children; child != NULL; child = child->next) {
 697                 if (child->aiocb == NULL) {
 698                         continue;
 699                 }
 700                 if (child->aiocb->aio_fildes != fsp->fh->fd) {
 701                         continue;
 702                 }
 703                 if ((aiocb != NULL) && (child->aiocb != aiocb)) {
 704                         continue;
 705                 }
 706
 707                 /*
 708                  * We let the child do its job, but we discard the result when
 709                  * it's finished.
 710                  */
 711
 712                 child->cancelled = true;
 713         }
 714
 715         return AIO_CANCELED;
 716 }
 717
 718 static int aio_fork_error_fn(struct vfs_handle_struct *handle,
 719                              struct files_struct *fsp,
 720                              SMB_STRUCT_AIOCB *aiocb)
 721 {
 722         struct aio_child *child = aio_fork_find_child(handle, aiocb);
 723
 724         if (child == NULL) {
 725                 errno = EINVAL;
 726                 return -1;
 727         }
 728
 729         return child->retval.ret_errno;
 730 }
 731
 732 static void aio_fork_suspend_timed_out(struct tevent_context *event_ctx,
 733                                         struct tevent_timer *te,
 734                                         struct timeval now,
 735                                         void *private_data)
 736 {
 737         bool *timed_out = (bool *)private_data;
 738         /* Remove this timed event handler. */
 739         TALLOC_FREE(te);
 740         *timed_out = true;
 741 }
 742
 743 static int aio_fork_suspend(struct vfs_handle_struct *handle,
 744                         struct files_struct *fsp,
 745                         const SMB_STRUCT_AIOCB * const aiocb_array[],
 746                         int n,
 747                         const struct timespec *timeout)
 748 {
 749         struct aio_child_list *children = NULL;
 750         TALLOC_CTX *frame = talloc_stackframe();
 751         struct event_context *ev = NULL;
 752         int i;
 753         int ret = -1;
 754         bool timed_out = false;
 755
 756         children = init_aio_children(handle);
 757         if (children == NULL) {
 758                 errno = EINVAL;
 759                 goto out;
 760         }
 761
 762         /* This is a blocking call, and has to use a sub-event loop. */
 763         ev = event_context_init(frame);
 764         if (ev == NULL) {
 765                 errno = ENOMEM;
 766                 goto out;
 767         }
 768
 769         if (timeout) {
 770                 struct timeval tv = convert_timespec_to_timeval(*timeout);
 771                 struct tevent_timer *te = tevent_add_timer(ev,
 772                                                 frame,
 773                                                 timeval_current_ofs(tv.tv_sec,
 774                                                                     tv.tv_usec),
 775                                                 aio_fork_suspend_timed_out,
 776                                                 &timed_out);
 777                 if (!te) {
 778                         errno = ENOMEM;
 779                         goto out;
 780                 }
 781         }
 782
 783         for (i = 0; i < n; i++) {
 784                 struct aio_child *child = NULL;
 785                 const SMB_STRUCT_AIOCB *aiocb = aiocb_array[i];
 786
 787                 if (!aiocb) {
 788                         continue;
 789                 }
 790
 791                 /*
 792                  * We're going to cheat here. We know that smbd/aio.c
 793                  * only calls this when it's waiting for every single
 794                  * outstanding call to finish on a close, so just wait
 795                  * individually for each IO to complete. We don't care
 796                  * what order they finish - only that they all do. JRA.
 797                  */
 798
 799                 for (child = children->children; child != NULL; child = child->next) {
 800                         if (child->aiocb == NULL) {
 801                                 continue;
 802                         }
 803                         if (child->aiocb->aio_fildes != fsp->fh->fd) {
 804                                 continue;
 805                         }
 806                         if (child->aiocb != aiocb) {
 807                                 continue;
 808                         }
 809
 810                         if (child->aiocb->aio_sigevent.sigev_value.sival_ptr == NULL) {
 811                                 continue;
 812                         }
 813
 814                         /* We're never using this event on the
 815                          * main event context again... */
 816                         TALLOC_FREE(child->sock_event);
 817
 818                         child->sock_event = event_add_fd(ev,
 819                                                 child,
 820                                                 child->sockfd,
 821                                                 EVENT_FD_READ,
 822                                                 handle_aio_completion,
 823                                                 child);
 824
 825                         while (1) {
 826                                 if (tevent_loop_once(ev) == -1) {
 827                                         goto out;
 828                                 }
 829
 830                                 if (timed_out) {
 831                                         errno = EAGAIN;
 832                                         goto out;
 833                                 }
 834
 835                                 /* We set child->aiocb to NULL in our hooked
 836                                  * AIO_RETURN(). */
 837                                 if (child->aiocb == NULL) {
 838                                         break;
 839                                 }
 840                         }
 841                 }
 842         }
 843
 844         ret = 0;
 845
 846   out:
 847
 848         TALLOC_FREE(frame);
 849         return ret;
 850 }
 851
 852 static struct vfs_fn_pointers vfs_aio_fork_fns = {
 853         .aio_read = aio_fork_read,
 854         .aio_write = aio_fork_write,
 855         .aio_return_fn = aio_fork_return_fn,
 856         .aio_cancel = aio_fork_cancel,
 857         .aio_error_fn = aio_fork_error_fn,
 858         .aio_suspend = aio_fork_suspend,
 859 };
 860
 861 NTSTATUS vfs_aio_fork_init(void);
 862 NTSTATUS vfs_aio_fork_init(void)
 863 {
 864         return smb_register_vfs(SMB_VFS_INTERFACE_VERSION,
 865                                 "aio_fork", &vfs_aio_fork_fns);
 866 }