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