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