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