source3/smbd/fileio.c

   1 /*
   2    Unix SMB/Netbios implementation.
   3    Version 1.9.
   4    read/write to a files_struct
   5    Copyright (C) Andrew Tridgell 1992-1998
   6    Copyright (C) Jeremy Allison 2000-2002. - write cache.
   7
   8    This program is free software; you can redistribute it and/or modify
   9    it under the terms of the GNU General Public License as published by
  10    the Free Software Foundation; either version 3 of the License, or
  11    (at your option) any later version.
  12
  13    This program is distributed in the hope that it will be useful,
  14    but WITHOUT ANY WARRANTY; without even the implied warranty of
  15    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16    GNU General Public License for more details.
  17
  18    You should have received a copy of the GNU General Public License
  19    along with this program.  If not, see <http://www.gnu.org/licenses/>.
  20 */
  21
  22 #include "includes.h"
  23 #include "printing.h"
  24 #include "smbd/smbd.h"
  25 #include "smbd/globals.h"
  26 #include "smbprofile.h"
  27
  28 struct write_cache {
  29         off_t file_size;
  30         off_t offset;
  31         size_t alloc_size;
  32         size_t data_size;
  33         char *data;
  34 };
  35
  36 static bool setup_write_cache(files_struct *, off_t);
  37
  38 /****************************************************************************
  39  Read from write cache if we can.
  40 ****************************************************************************/
  41
  42 static bool read_from_write_cache(files_struct *fsp,char *data,off_t pos,size_t n)
  43 {
  44         struct write_cache *wcp = fsp->wcp;
  45
  46         if(!wcp) {
  47                 return False;
  48         }
  49
  50         if( n > wcp->data_size || pos < wcp->offset || pos + n > wcp->offset + wcp->data_size) {
  51                 return False;
  52         }
  53
  54         memcpy(data, wcp->data + (pos - wcp->offset), n);
  55
  56         DO_PROFILE_INC(writecache_cached_reads);
  57
  58         return True;
  59 }
  60
  61 /****************************************************************************
  62  Read from a file.
  63 ****************************************************************************/
  64
  65 ssize_t read_file(files_struct *fsp,char *data,off_t pos,size_t n)
  66 {
  67         ssize_t ret = 0;
  68
  69         /* you can't read from print files */
  70         if (fsp->print_file) {
  71                 errno = EBADF;
  72                 return -1;
  73         }
  74
  75         /*
  76          * Serve from write cache if we can.
  77          */
  78
  79         if(read_from_write_cache(fsp, data, pos, n)) {
  80                 fsp->fh->pos = pos + n;
  81                 fsp->fh->position_information = fsp->fh->pos;
  82                 return n;
  83         }
  84
  85         flush_write_cache(fsp, SAMBA_READ_FLUSH);
  86
  87         fsp->fh->pos = pos;
  88
  89         if (n > 0) {
  90                 ret = SMB_VFS_PREAD(fsp,data,n,pos);
  91
  92                 if (ret == -1) {
  93                         return -1;
  94                 }
  95         }
  96
  97         DEBUG(10,("read_file (%s): pos = %.0f, size = %lu, returned %lu\n",
  98                   fsp_str_dbg(fsp), (double)pos, (unsigned long)n, (long)ret));
  99
 100         fsp->fh->pos += ret;
 101         fsp->fh->position_information = fsp->fh->pos;
 102
 103         return(ret);
 104 }
 105
 106 /****************************************************************************
 107  *Really* write to a file.
 108 ****************************************************************************/
 109
 110 static ssize_t real_write_file(struct smb_request *req,
 111                                 files_struct *fsp,
 112                                 const char *data,
 113                                 off_t pos,
 114                                 size_t n)
 115 {
 116         ssize_t ret;
 117
 118         if (pos == -1) {
 119                 ret = vfs_write_data(req, fsp, data, n);
 120         } else {
 121                 fsp->fh->pos = pos;
 122                 if (pos && lp_strict_allocate(SNUM(fsp->conn) &&
 123                                 !fsp->is_sparse)) {
 124                         if (vfs_fill_sparse(fsp, pos) == -1) {
 125                                 return -1;
 126                         }
 127                 }
 128                 ret = vfs_pwrite_data(req, fsp, data, n, pos);
 129         }
 130
 131         DEBUG(10,("real_write_file (%s): pos = %.0f, size = %lu, returned %ld\n",
 132                   fsp_str_dbg(fsp), (double)pos, (unsigned long)n, (long)ret));
 133
 134         if (ret != -1) {
 135                 fsp->fh->pos += ret;
 136
 137 /* Yes - this is correct - writes don't update this. JRA. */
 138 /* Found by Samba4 tests. */
 139 #if 0
 140                 fsp->position_information = fsp->pos;
 141 #endif
 142         }
 143
 144         return ret;
 145 }
 146
 147 /****************************************************************************
 148  File size cache change.
 149  Updates size on disk but doesn't flush the cache.
 150 ****************************************************************************/
 151
 152 static int wcp_file_size_change(files_struct *fsp)
 153 {
 154         int ret;
 155         struct write_cache *wcp = fsp->wcp;
 156
 157         wcp->file_size = wcp->offset + wcp->data_size;
 158         ret = SMB_VFS_FTRUNCATE(fsp, wcp->file_size);
 159         if (ret == -1) {
 160                 DEBUG(0,("wcp_file_size_change (%s): ftruncate of size %.0f "
 161                          "error %s\n", fsp_str_dbg(fsp),
 162                          (double)wcp->file_size, strerror(errno)));
 163         }
 164         return ret;
 165 }
 166
 167 void update_write_time_handler(struct tevent_context *ctx,
 168                                       struct tevent_timer *te,
 169                                       struct timeval now,
 170                                       void *private_data)
 171 {
 172         files_struct *fsp = (files_struct *)private_data;
 173
 174         DEBUG(5, ("Update write time on %s\n", fsp_str_dbg(fsp)));
 175
 176         /* change the write time in the open file db. */
 177         (void)set_write_time(fsp->file_id, timespec_current());
 178
 179         /* And notify. */
 180         notify_fname(fsp->conn, NOTIFY_ACTION_MODIFIED,
 181                      FILE_NOTIFY_CHANGE_LAST_WRITE, fsp->fsp_name->base_name);
 182
 183         /* Remove the timed event handler. */
 184         TALLOC_FREE(fsp->update_write_time_event);
 185 }
 186
 187 /*********************************************************
 188  Schedule a write time update for WRITE_TIME_UPDATE_USEC_DELAY
 189  in the future.
 190 *********************************************************/
 191
 192 void trigger_write_time_update(struct files_struct *fsp)
 193 {
 194         int delay;
 195
 196         if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
 197                 /* Don't use delayed writes on POSIX files. */
 198                 return;
 199         }
 200
 201         if (fsp->write_time_forced) {
 202                 /* No point - "sticky" write times
 203                  * in effect.
 204                  */
 205                 return;
 206         }
 207
 208         /* We need to remember someone did a write
 209          * and update to current time on close. */
 210
 211         fsp->update_write_time_on_close = true;
 212
 213         if (fsp->update_write_time_triggered) {
 214                 /*
 215                  * We only update the write time after 2 seconds
 216                  * on the first normal write. After that
 217                  * no other writes affect this until close.
 218                  */
 219                 return;
 220         }
 221         fsp->update_write_time_triggered = true;
 222
 223         delay = lp_parm_int(SNUM(fsp->conn),
 224                             "smbd", "writetimeupdatedelay",
 225                             WRITE_TIME_UPDATE_USEC_DELAY);
 226
 227         DEBUG(5, ("Update write time %d usec later on %s\n",
 228                   delay, fsp_str_dbg(fsp)));
 229
 230         /* trigger the update 2 seconds later */
 231         fsp->update_write_time_event =
 232                 tevent_add_timer(fsp->conn->sconn->ev_ctx, NULL,
 233                                  timeval_current_ofs_usec(delay),
 234                                  update_write_time_handler, fsp);
 235 }
 236
 237 void trigger_write_time_update_immediate(struct files_struct *fsp)
 238 {
 239         struct smb_file_time ft;
 240
 241         if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
 242                 /* Don't use delayed writes on POSIX files. */
 243                 return;
 244         }
 245
 246         if (fsp->write_time_forced) {
 247                 /*
 248                  * No point - "sticky" write times
 249                  * in effect.
 250                  */
 251                 return;
 252         }
 253
 254         TALLOC_FREE(fsp->update_write_time_event);
 255         DEBUG(5, ("Update write time immediate on %s\n",
 256                   fsp_str_dbg(fsp)));
 257
 258         /* After an immediate update, reset the trigger. */
 259         fsp->update_write_time_triggered = true;
 260         fsp->update_write_time_on_close = false;
 261
 262         ZERO_STRUCT(ft);
 263         ft.mtime = timespec_current();
 264
 265         /* Update the time in the open file db. */
 266         (void)set_write_time(fsp->file_id, ft.mtime);
 267
 268         /* Now set on disk - takes care of notify. */
 269         (void)smb_set_file_time(fsp->conn, fsp, fsp->fsp_name, &ft, false);
 270 }
 271
 272 void mark_file_modified(files_struct *fsp)
 273 {
 274         int dosmode;
 275
 276         if (fsp->modified) {
 277                 return;
 278         }
 279
 280         fsp->modified = true;
 281
 282         if (SMB_VFS_FSTAT(fsp, &fsp->fsp_name->st) != 0) {
 283                 return;
 284         }
 285         trigger_write_time_update(fsp);
 286
 287         if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
 288                 return;
 289         }
 290         if (!(lp_store_dos_attributes(SNUM(fsp->conn)) ||
 291               MAP_ARCHIVE(fsp->conn))) {
 292                 return;
 293         }
 294
 295         dosmode = dos_mode(fsp->conn, fsp->fsp_name);
 296         if (IS_DOS_ARCHIVE(dosmode)) {
 297                 return;
 298         }
 299         file_set_dosmode(fsp->conn, fsp->fsp_name,
 300                          dosmode | FILE_ATTRIBUTE_ARCHIVE, NULL, false);
 301 }
 302
 303 /****************************************************************************
 304  Write to a file.
 305 ****************************************************************************/
 306
 307 ssize_t write_file(struct smb_request *req,
 308                         files_struct *fsp,
 309                         const char *data,
 310                         off_t pos,
 311                         size_t n)
 312 {
 313         struct write_cache *wcp = fsp->wcp;
 314         ssize_t total_written = 0;
 315         int write_path = -1;
 316
 317         if (fsp->print_file) {
 318                 uint32_t t;
 319                 int ret;
 320
 321                 ret = print_spool_write(fsp, data, n, pos, &t);
 322                 if (ret) {
 323                         errno = ret;
 324                         return -1;
 325                 }
 326                 return t;
 327         }
 328
 329         if (!fsp->can_write) {
 330                 errno = EPERM;
 331                 return -1;
 332         }
 333
 334         /*
 335          * If this is the first write and we have an exclusive oplock
 336          * then setup the write cache.
 337          */
 338
 339         if (!fsp->modified &&
 340             EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) &&
 341             (wcp == NULL)) {
 342                 /*
 343                  * Note: no write cache with leases!
 344                  * as the handles would have to share the write cache
 345                  * that's possible but an improvement for another day...
 346                  */
 347                 setup_write_cache(fsp, fsp->fsp_name->st.st_ex_size);
 348                 wcp = fsp->wcp;
 349         }
 350
 351         mark_file_modified(fsp);
 352
 353         DO_PROFILE_INC(writecache_total_writes);
 354         if (!fsp->oplock_type) {
 355                 DO_PROFILE_INC(writecache_non_oplock_writes);
 356         }
 357
 358         /*
 359          * If this file is level II oplocked then we need
 360          * to grab the shared memory lock and inform all
 361          * other files with a level II lock that they need
 362          * to flush their read caches. We keep the lock over
 363          * the shared memory area whilst doing this.
 364          */
 365
 366         /* This should actually be improved to span the write. */
 367         contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WRITE);
 368         contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WRITE);
 369
 370         if (wcp && req->unread_bytes) {
 371                 /* If we're using receivefile don't
 372                  * deal with a write cache.
 373                  */
 374                 flush_write_cache(fsp, SAMBA_WRITE_FLUSH);
 375                 delete_write_cache(fsp);
 376                 wcp = NULL;
 377         }
 378
 379         if(!wcp) {
 380                 DO_PROFILE_INC(writecache_direct_writes);
 381                 total_written = real_write_file(req, fsp, data, pos, n);
 382                 return total_written;
 383         }
 384
 385         DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f "
 386                  "wcp->data_size=%u\n", fsp_str_dbg(fsp), fsp->fh->fd,
 387                  (double)pos, (unsigned int)n, (double)wcp->offset,
 388                  (unsigned int)wcp->data_size));
 389
 390         fsp->fh->pos = pos + n;
 391
 392         if ((n == 1) && (data[0] == '\0') && (pos > wcp->file_size)) {
 393                 int ret;
 394
 395                 /*
 396                  * This is a 1-byte write of a 0 beyond the EOF and
 397                  * thus implicitly also beyond the current active
 398                  * write cache, the typical file-extending (and
 399                  * allocating, but we're using the write cache here)
 400                  * write done by Windows. We just have to ftruncate
 401                  * the file and rely on posix semantics to return
 402                  * zeros for non-written file data that is within the
 403                  * file length.
 404                  *
 405                  * We can not use wcp_file_size_change here because we
 406                  * might have an existing write cache, and
 407                  * wcp_file_size_change assumes a change to just the
 408                  * end of the current write cache.
 409                  */
 410
 411                 wcp->file_size = pos + 1;
 412                 ret = SMB_VFS_FTRUNCATE(fsp, wcp->file_size);
 413                 if (ret == -1) {
 414                         DEBUG(0, ("wcp_file_size_change (%s): ftruncate of "
 415                                   "size %.0f error %s\n", fsp_str_dbg(fsp),
 416                                   (double)wcp->file_size, strerror(errno)));
 417                         return -1;
 418                 }
 419                 return 1;
 420         }
 421
 422
 423         /*
 424          * If we have active cache and it isn't contiguous then we flush.
 425          * NOTE: There is a small problem with running out of disk ....
 426          */
 427
 428         if (wcp->data_size) {
 429                 bool cache_flush_needed = False;
 430
 431                 if ((pos >= wcp->offset) &&
 432                     (pos <= wcp->offset + wcp->data_size)) {
 433
 434                         /* ASCII art.... JRA.
 435
 436       +--------------+-----
 437       | Cached data  | Rest of allocated cache buffer....
 438       +--------------+-----
 439
 440             +-------------------+
 441             | Data to write     |
 442             +-------------------+
 443
 444                         */
 445
 446                         /*
 447                          * Start of write overlaps or abutts the existing data.
 448                          */
 449
 450                         size_t data_used;
 451
 452                         data_used = MIN((wcp->alloc_size - (pos - wcp->offset)),
 453                                         n);
 454
 455                         memcpy(wcp->data + (pos - wcp->offset), data,
 456                                data_used);
 457
 458                         /*
 459                          * Update the current buffer size with the new data.
 460                          */
 461
 462                         if(pos + data_used > wcp->offset + wcp->data_size) {
 463                                 wcp->data_size = pos + data_used - wcp->offset;
 464                         }
 465
 466                         /*
 467                          * Update the file size if changed.
 468                          */
 469
 470                         if (wcp->offset + wcp->data_size > wcp->file_size) {
 471                                 if (wcp_file_size_change(fsp) == -1) {
 472                                         return -1;
 473                                 }
 474                         }
 475
 476                         /*
 477                          * If we used all the data then
 478                          * return here.
 479                          */
 480
 481                         if(n == data_used) {
 482                                 return n;
 483                         } else {
 484                                 cache_flush_needed = True;
 485                         }
 486                         /*
 487                          * Move the start of data forward by the amount used,
 488                          * cut down the amount left by the same amount.
 489                          */
 490
 491                         data += data_used;
 492                         pos += data_used;
 493                         n -= data_used;
 494
 495                         DO_PROFILE_INC(writecache_abutted_writes);
 496                         total_written = data_used;
 497
 498                         write_path = 1;
 499
 500                 } else if ((pos < wcp->offset) &&
 501                            (pos + n > wcp->offset) &&
 502                            (pos + n <= wcp->offset + wcp->alloc_size)) {
 503
 504                         /* ASCII art.... JRA.
 505
 506                         +---------------+
 507                         | Cache buffer  |
 508                         +---------------+
 509
 510             +-------------------+
 511             | Data to write     |
 512             +-------------------+
 513
 514                         */
 515
 516                         /*
 517                          * End of write overlaps the existing data.
 518                          */
 519
 520                         size_t data_used = pos + n - wcp->offset;
 521
 522                         memcpy(wcp->data, data + n - data_used, data_used);
 523
 524                         /*
 525                          * Update the current buffer size with the new data.
 526                          */
 527
 528                         if(pos + n > wcp->offset + wcp->data_size) {
 529                                 wcp->data_size = pos + n - wcp->offset;
 530                         }
 531
 532                         /*
 533                          * Update the file size if changed.
 534                          */
 535
 536                         if (wcp->offset + wcp->data_size > wcp->file_size) {
 537                                 if (wcp_file_size_change(fsp) == -1) {
 538                                         return -1;
 539                                 }
 540                         }
 541
 542                         /*
 543                          * We don't need to move the start of data, but we
 544                          * cut down the amount left by the amount used.
 545                          */
 546
 547                         n -= data_used;
 548
 549                         /*
 550                          * We cannot have used all the data here.
 551                          */
 552
 553                         cache_flush_needed = True;
 554
 555                         DO_PROFILE_INC(writecache_abutted_writes);
 556                         total_written = data_used;
 557
 558                         write_path = 2;
 559
 560                 } else if ((pos >= wcp->file_size) &&
 561                            (wcp->offset + wcp->data_size == wcp->file_size) &&
 562                            (pos > wcp->offset + wcp->data_size) &&
 563                            (pos < wcp->offset + wcp->alloc_size) ) {
 564
 565                         /* ASCII art.... JRA.
 566
 567                        End of file ---->|
 568
 569                         +---------------+---------------+
 570                         | Cached data   | Cache buffer  |
 571                         +---------------+---------------+
 572
 573                                               +-------------------+
 574                                               | Data to write     |
 575                                               +-------------------+
 576
 577                         */
 578
 579                         /*
 580                          * Non-contiguous write part of which fits within
 581                          * the cache buffer and is extending the file
 582                          * and the cache contents reflect the current
 583                          * data up to the current end of the file.
 584                          */
 585
 586                         size_t data_used;
 587
 588                         if(pos + n <= wcp->offset + wcp->alloc_size) {
 589                                 data_used = n;
 590                         } else {
 591                                 data_used = wcp->offset+wcp->alloc_size-pos;
 592                         }
 593
 594                         /*
 595                          * Fill in the non-continuous area with zeros.
 596                          */
 597
 598                         memset(wcp->data + wcp->data_size, '\0',
 599                                 pos - (wcp->offset + wcp->data_size) );
 600
 601                         memcpy(wcp->data + (pos - wcp->offset), data,
 602                                data_used);
 603
 604                         /*
 605                          * Update the current buffer size with the new data.
 606                          */
 607
 608                         if(pos + data_used > wcp->offset + wcp->data_size) {
 609                                 wcp->data_size = pos + data_used - wcp->offset;
 610                         }
 611
 612                         /*
 613                          * Update the file size if changed.
 614                          */
 615
 616                         if (wcp->offset + wcp->data_size > wcp->file_size) {
 617                                 if (wcp_file_size_change(fsp) == -1) {
 618                                         return -1;
 619                                 }
 620                         }
 621
 622                         /*
 623                          * If we used all the data then
 624                          * return here.
 625                          */
 626
 627                         if(n == data_used) {
 628                                 return n;
 629                         } else {
 630                                 cache_flush_needed = True;
 631                         }
 632
 633                         /*
 634                          * Move the start of data forward by the amount used,
 635                          * cut down the amount left by the same amount.
 636                          */
 637
 638                         data += data_used;
 639                         pos += data_used;
 640                         n -= data_used;
 641
 642                         DO_PROFILE_INC(writecache_abutted_writes);
 643                         total_written = data_used;
 644
 645                         write_path = 3;
 646
 647                 } else if ( (pos >= wcp->file_size) &&
 648                             (n == 1) &&
 649                             (wcp->file_size == wcp->offset + wcp->data_size) &&
 650                             (pos < wcp->file_size + wcp->alloc_size)) {
 651
 652                         /*
 653
 654                 End of file ---->|
 655
 656                  +---------------+---------------+
 657                  | Cached data   | Cache buffer  |
 658                  +---------------+---------------+
 659
 660                                  |<------- allocated size ---------------->|
 661
 662                                                          +--------+
 663                                                          | 1 Byte |
 664                                                          +--------+
 665
 666                         MS-Office seems to do this a lot to determine if
 667                         there's enough space on the filesystem to write a new
 668                         file.
 669
 670                         Change to :
 671
 672                 End of file ---->|
 673                                  +-----------------------+--------+
 674                                  | Zeroed Cached data    | 1 Byte |
 675                                  +-----------------------+--------+
 676                         */
 677
 678                         flush_write_cache(fsp, SAMBA_WRITE_FLUSH);
 679                         wcp->offset = wcp->file_size;
 680                         wcp->data_size = pos - wcp->file_size + 1;
 681                         memset(wcp->data, '\0', wcp->data_size);
 682                         memcpy(wcp->data + wcp->data_size-1, data, 1);
 683
 684                         /*
 685                          * Update the file size if changed.
 686                          */
 687
 688                         if (wcp->offset + wcp->data_size > wcp->file_size) {
 689                                 if (wcp_file_size_change(fsp) == -1) {
 690                                         return -1;
 691                                 }
 692                         }
 693
 694                         return n;
 695
 696                 } else {
 697
 698                         /* ASCII art..... JRA.
 699
 700    Case 1).
 701
 702                         +---------------+---------------+
 703                         | Cached data   | Cache buffer  |
 704                         +---------------+---------------+
 705
 706                                                               +---------------+
 707                                                               | Data to write |
 708                                                               +---------------+
 709
 710    Case 2).
 711
 712                            +---------------+---------------+
 713                            | Cached data   | Cache buffer  |
 714                            +---------------+---------------+
 715
 716    +-------------------+
 717    | Data to write     |
 718    +-------------------+
 719
 720     Case 3).
 721
 722                            +---------------+---------------+
 723                            | Cached data   | Cache buffer  |
 724                            +---------------+---------------+
 725
 726                   +-----------------------------------------------------+
 727                   | Data to write                                       |
 728                   +-----------------------------------------------------+
 729
 730                   */
 731
 732                         /*
 733                          * Write is bigger than buffer, or there is no
 734                          * overlap on the low or high ends.
 735                          */
 736
 737                         DEBUG(9,("write_file: non cacheable write : fd = %d, "
 738                                  "pos = %.0f, len = %u, "
 739                                  "current cache pos = %.0f len = %u\n",
 740                                  fsp->fh->fd, (double)pos, (unsigned int)n,
 741                                  (double)wcp->offset,
 742                                  (unsigned int)wcp->data_size ));
 743
 744                         /*
 745                          * If write would fit in the cache, and is
 746                          * larger than the data already in the cache,
 747                          * flush the cache and preferentially copy the
 748                          * data new data into it. Otherwise just write
 749                          * the data directly.
 750                          */
 751
 752                         if ( n <= wcp->alloc_size && n > wcp->data_size) {
 753                                 cache_flush_needed = True;
 754                         } else {
 755                                 ssize_t ret = real_write_file(NULL, fsp, data,
 756                                                               pos, n);
 757
 758                                 /*
 759                                  * If the write overlaps the entire
 760                                  * cache, then discard the current
 761                                  * contents of the cache.  Fix from
 762                                  * Rasmus Borup Hansen rbh@math.ku.dk.
 763                                  */
 764
 765                                 if ((pos <= wcp->offset) &&
 766                                     (pos + n >= wcp->offset+wcp->data_size)) {
 767                                         DEBUG(9,("write_file: discarding "
 768                                                  "overwritten write cache: "
 769                                                  "fd = %d, off=%.0f, "
 770                                                  "size=%u\n", fsp->fh->fd,
 771                                                  (double)wcp->offset,
 772                                                  (unsigned)wcp->data_size));
 773                                         wcp->data_size = 0;
 774                                 }
 775
 776                                 DO_PROFILE_INC(writecache_direct_writes);
 777                                 if (ret == -1) {
 778                                         return ret;
 779                                 }
 780
 781                                 if (pos + ret > wcp->file_size) {
 782                                         wcp->file_size = pos + ret;
 783                                 }
 784
 785                                 return ret;
 786                         }
 787
 788                         write_path = 4;
 789
 790                 }
 791
 792                 if (cache_flush_needed) {
 793                         DEBUG(3, ("SAMBA_WRITE_FLUSH:%d: due to noncontinuous "
 794                                   "write: fd = %d, size = %.0f, pos = %.0f, "
 795                                   "n = %u, wcp->offset=%.0f, "
 796                                   "wcp->data_size=%u\n",
 797                                   write_path, fsp->fh->fd,
 798                                   (double)wcp->file_size, (double)pos,
 799                                   (unsigned int)n, (double)wcp->offset,
 800                                   (unsigned int)wcp->data_size ));
 801
 802                         flush_write_cache(fsp, SAMBA_WRITE_FLUSH);
 803                 }
 804         }
 805
 806         /*
 807          * If the write request is bigger than the cache
 808          * size, write it all out.
 809          */
 810
 811         if (n > wcp->alloc_size ) {
 812                 ssize_t ret = real_write_file(NULL,fsp, data, pos, n);
 813                 if (ret == -1) {
 814                         return -1;
 815                 }
 816
 817                 if (pos + ret > wcp->file_size) {
 818                         wcp->file_size = pos + n;
 819                 }
 820
 821                 DO_PROFILE_INC(writecache_direct_writes);
 822                 return total_written + n;
 823         }
 824
 825         /*
 826          * If there's any data left, cache it.
 827          */
 828
 829         if (n) {
 830                 DO_PROFILE_INC(writecache_cached_writes);
 831                 if (wcp->data_size) {
 832                         DO_PROFILE_INC(writecache_abutted_writes);
 833                 } else {
 834                         DO_PROFILE_INC(writecache_init_writes);
 835                 }
 836
 837                 if ((wcp->data_size == 0)
 838                     && (pos > wcp->file_size)
 839                     && (pos + n <= wcp->file_size + wcp->alloc_size)) {
 840                         /*
 841                          * This is a write completely beyond the
 842                          * current EOF, but within reach of the write
 843                          * cache. We expect fill-up writes pretty
 844                          * soon, so it does not make sense to start
 845                          * the write cache at the current
 846                          * offset. These fill-up writes would trigger
 847                          * separate pwrites or even unnecessary cache
 848                          * flushes because they overlap if this is a
 849                          * one-byte allocating write.
 850                          */
 851                         wcp->offset = wcp->file_size;
 852                         wcp->data_size = pos - wcp->file_size;
 853                         memset(wcp->data, 0, wcp->data_size);
 854                 }
 855
 856                 memcpy(wcp->data+wcp->data_size, data, n);
 857                 if (wcp->data_size == 0) {
 858                         wcp->offset = pos;
 859                 }
 860                 wcp->data_size += n;
 861
 862                 /*
 863                  * Update the file size if changed.
 864                  */
 865
 866                 if (wcp->offset + wcp->data_size > wcp->file_size) {
 867                         if (wcp_file_size_change(fsp) == -1) {
 868                                 return -1;
 869                         }
 870                 }
 871                 DEBUG(9, ("wcp->offset = %.0f wcp->data_size = %u cache "
 872                           "return %u\n",
 873                           (double)wcp->offset, (unsigned int)wcp->data_size,
 874                           (unsigned int)n));
 875
 876                 total_written += n;
 877                 return total_written; /* .... that's a write :) */
 878         }
 879
 880         return total_written;
 881 }
 882
 883 /****************************************************************************
 884  Delete the write cache structure.
 885 ****************************************************************************/
 886
 887 void delete_write_cache(files_struct *fsp)
 888 {
 889         struct write_cache *wcp;
 890
 891         if(!fsp) {
 892                 return;
 893         }
 894
 895         if(!(wcp = fsp->wcp)) {
 896                 return;
 897         }
 898
 899         DO_PROFILE_INC(writecache_deallocations);
 900         allocated_write_caches--;
 901
 902         SMB_ASSERT(wcp->data_size == 0);
 903
 904         SAFE_FREE(wcp->data);
 905         SAFE_FREE(fsp->wcp);
 906
 907         DEBUG(10,("delete_write_cache: File %s deleted write cache\n",
 908                   fsp_str_dbg(fsp)));
 909 }
 910
 911 /****************************************************************************
 912  Setup the write cache structure.
 913 ****************************************************************************/
 914
 915 static bool setup_write_cache(files_struct *fsp, off_t file_size)
 916 {
 917         ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
 918         struct write_cache *wcp;
 919
 920         if (allocated_write_caches >= MAX_WRITE_CACHES) {
 921                 return False;
 922         }
 923
 924         if(alloc_size == 0 || fsp->wcp) {
 925                 return False;
 926         }
 927
 928         if((wcp = SMB_MALLOC_P(struct write_cache)) == NULL) {
 929                 DEBUG(0,("setup_write_cache: malloc fail.\n"));
 930                 return False;
 931         }
 932
 933         wcp->file_size = file_size;
 934         wcp->offset = 0;
 935         wcp->alloc_size = alloc_size;
 936         wcp->data_size = 0;
 937         if((wcp->data = (char *)SMB_MALLOC(wcp->alloc_size)) == NULL) {
 938                 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
 939                         (unsigned int)wcp->alloc_size ));
 940                 SAFE_FREE(wcp);
 941                 return False;
 942         }
 943
 944         memset(wcp->data, '\0', wcp->alloc_size );
 945
 946         fsp->wcp = wcp;
 947         DO_PROFILE_INC(writecache_allocations);
 948         allocated_write_caches++;
 949
 950         DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
 951                   fsp_str_dbg(fsp), (unsigned long)wcp->alloc_size));
 952
 953         return True;
 954 }
 955
 956 /****************************************************************************
 957  Cope with a size change.
 958 ****************************************************************************/
 959
 960 void set_filelen_write_cache(files_struct *fsp, off_t file_size)
 961 {
 962         if(fsp->wcp) {
 963                 /* The cache *must* have been flushed before we do this. */
 964                 if (fsp->wcp->data_size != 0) {
 965                         char *msg;
 966                         if (asprintf(&msg, "set_filelen_write_cache: size change "
 967                                  "on file %s with write cache size = %lu\n",
 968                                  fsp->fsp_name->base_name,
 969                                  (unsigned long)fsp->wcp->data_size) != -1) {
 970                                 smb_panic(msg);
 971                         } else {
 972                                 smb_panic("set_filelen_write_cache");
 973                         }
 974                 }
 975                 fsp->wcp->file_size = file_size;
 976         }
 977 }
 978
 979 /*******************************************************************
 980  Flush a write cache struct to disk.
 981 ********************************************************************/
 982
 983 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
 984 {
 985         struct write_cache *wcp = fsp->wcp;
 986         size_t data_size;
 987         ssize_t ret;
 988
 989         if(!wcp || !wcp->data_size) {
 990                 return 0;
 991         }
 992
 993         data_size = wcp->data_size;
 994         wcp->data_size = 0;
 995
 996         switch (reason) {
 997         case SAMBA_SEEK_FLUSH:
 998                 DO_PROFILE_INC(writecache_flush_reason_seek);
 999                 break;
1000         case SAMBA_READ_FLUSH:
1001                 DO_PROFILE_INC(writecache_flush_reason_read);
1002                 break;
1003         case SAMBA_WRITE_FLUSH:
1004                 DO_PROFILE_INC(writecache_flush_reason_write);;
1005                 break;
1006         case SAMBA_READRAW_FLUSH:
1007                 DO_PROFILE_INC(writecache_flush_reason_readraw);
1008                 break;
1009         case SAMBA_OPLOCK_RELEASE_FLUSH:
1010                 DO_PROFILE_INC(writecache_flush_reason_oplock);
1011                 break;
1012         case SAMBA_CLOSE_FLUSH:
1013                 DO_PROFILE_INC(writecache_flush_reason_close);
1014                 break;
1015         case SAMBA_SYNC_FLUSH:
1016                 DO_PROFILE_INC(writecache_flush_reason_sync);
1017                 break;
1018         case SAMBA_SIZECHANGE_FLUSH:
1019                 DO_PROFILE_INC(writecache_flush_reason_sizechange);
1020                 break;
1021         default:
1022                 break;
1023         }
1024
1025         DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
1026                 fsp->fh->fd, (double)wcp->offset, (unsigned int)data_size));
1027
1028         if(data_size == wcp->alloc_size) {
1029                 DO_PROFILE_INC(writecache_perfect_writes);
1030         }
1031
1032         ret = real_write_file(NULL, fsp, wcp->data, wcp->offset, data_size);
1033
1034         /*
1035          * Ensure file size if kept up to date if write extends file.
1036          */
1037
1038         if ((ret != -1) && (wcp->offset + ret > wcp->file_size)) {
1039                 wcp->file_size = wcp->offset + ret;
1040         }
1041
1042         return ret;
1043 }
1044
1045 /*******************************************************************
1046 sync a file
1047 ********************************************************************/
1048
1049 NTSTATUS sync_file(connection_struct *conn, files_struct *fsp, bool write_through)
1050 {
1051         if (fsp->fh->fd == -1)
1052                 return NT_STATUS_INVALID_HANDLE;
1053
1054         if (lp_strict_sync(SNUM(conn)) &&
1055             (lp_sync_always(SNUM(conn)) || write_through)) {
1056                 int ret = flush_write_cache(fsp, SAMBA_SYNC_FLUSH);
1057                 if (ret == -1) {
1058                         return map_nt_error_from_unix(errno);
1059                 }
1060                 ret = SMB_VFS_FSYNC(fsp);
1061                 if (ret == -1) {
1062                         return map_nt_error_from_unix(errno);
1063                 }
1064         }
1065         return NT_STATUS_OK;
1066 }
1067
1068 /************************************************************
1069  Perform a stat whether a valid fd or not.
1070 ************************************************************/
1071
1072 int fsp_stat(files_struct *fsp)
1073 {
1074         if (fsp->fh->fd == -1) {
1075                 if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
1076                         return SMB_VFS_LSTAT(fsp->conn, fsp->fsp_name);
1077                 } else {
1078                         return SMB_VFS_STAT(fsp->conn, fsp->fsp_name);
1079                 }
1080         } else {
1081                 return SMB_VFS_FSTAT(fsp, &fsp->fsp_name->st);
1082         }
1083 }