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