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