source3/smbd/fileio.c

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