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_cached_reads);
  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, SAMBA_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         fsp->fh->pos = pos;
 119         if (pos && lp_strict_allocate(SNUM(fsp->conn) &&
 120                         !fsp->is_sparse)) {
 121                 if (vfs_fill_sparse(fsp, pos) == -1) {
 122                         return -1;
 123                 }
 124         }
 125         ret = vfs_pwrite_data(req, fsp, data, n, pos);
 126
 127         DEBUG(10,("real_write_file (%s): pos = %.0f, size = %lu, returned %ld\n",
 128                   fsp_str_dbg(fsp), (double)pos, (unsigned long)n, (long)ret));
 129
 130         if (ret != -1) {
 131                 fsp->fh->pos += ret;
 132
 133 /* Yes - this is correct - writes don't update this. JRA. */
 134 /* Found by Samba4 tests. */
 135 #if 0
 136                 fsp->position_information = fsp->pos;
 137 #endif
 138         }
 139
 140         return ret;
 141 }
 142
 143 /****************************************************************************
 144  File size cache change.
 145  Updates size on disk but doesn't flush the cache.
 146 ****************************************************************************/
 147
 148 static int wcp_file_size_change(files_struct *fsp)
 149 {
 150         int ret;
 151         struct write_cache *wcp = fsp->wcp;
 152
 153         wcp->file_size = wcp->offset + wcp->data_size;
 154         ret = SMB_VFS_FTRUNCATE(fsp, wcp->file_size);
 155         if (ret == -1) {
 156                 DEBUG(0,("wcp_file_size_change (%s): ftruncate of size %.0f "
 157                          "error %s\n", fsp_str_dbg(fsp),
 158                          (double)wcp->file_size, strerror(errno)));
 159         }
 160         return ret;
 161 }
 162
 163 void fsp_flush_write_time_update(struct files_struct *fsp)
 164 {
 165         /*
 166          * Note this won't expect any impersonation!
 167          * So don't call any SMB_VFS operations here!
 168          */
 169
 170         DEBUG(5, ("Update write time on %s\n", fsp_str_dbg(fsp)));
 171
 172         /* change the write time in the open file db. */
 173         (void)set_write_time(fsp->file_id, timespec_current());
 174
 175         /* And notify. */
 176         notify_fname(fsp->conn, NOTIFY_ACTION_MODIFIED,
 177                      FILE_NOTIFY_CHANGE_LAST_WRITE, fsp->fsp_name->base_name);
 178
 179         /* Remove the timed event handler. */
 180         TALLOC_FREE(fsp->update_write_time_event);
 181 }
 182
 183 static void update_write_time_handler(struct tevent_context *ctx,
 184                                       struct tevent_timer *te,
 185                                       struct timeval now,
 186                                       void *private_data)
 187 {
 188         files_struct *fsp = (files_struct *)private_data;
 189         fsp_flush_write_time_update(fsp);
 190 }
 191
 192 /*********************************************************
 193  Schedule a write time update for WRITE_TIME_UPDATE_USEC_DELAY
 194  in the future.
 195 *********************************************************/
 196
 197 void trigger_write_time_update(struct files_struct *fsp)
 198 {
 199         int delay;
 200
 201         if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
 202                 /* Don't use delayed writes on POSIX files. */
 203                 return;
 204         }
 205
 206         if (fsp->write_time_forced) {
 207                 /* No point - "sticky" write times
 208                  * in effect.
 209                  */
 210                 return;
 211         }
 212
 213         /* We need to remember someone did a write
 214          * and update to current time on close. */
 215
 216         fsp->update_write_time_on_close = true;
 217
 218         if (fsp->update_write_time_triggered) {
 219                 /*
 220                  * We only update the write time after 2 seconds
 221                  * on the first normal write. After that
 222                  * no other writes affect this until close.
 223                  */
 224                 return;
 225         }
 226         fsp->update_write_time_triggered = true;
 227
 228         delay = lp_parm_int(SNUM(fsp->conn),
 229                             "smbd", "writetimeupdatedelay",
 230                             WRITE_TIME_UPDATE_USEC_DELAY);
 231
 232         DEBUG(5, ("Update write time %d usec later on %s\n",
 233                   delay, fsp_str_dbg(fsp)));
 234
 235         /* trigger the update 2 seconds later */
 236         fsp->update_write_time_event =
 237                 tevent_add_timer(fsp->conn->sconn->ev_ctx, NULL,
 238                                  timeval_current_ofs_usec(delay),
 239                                  update_write_time_handler, fsp);
 240 }
 241
 242 void trigger_write_time_update_immediate(struct files_struct *fsp)
 243 {
 244         struct smb_file_time ft;
 245
 246         if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
 247                 /* Don't use delayed writes on POSIX files. */
 248                 return;
 249         }
 250
 251         if (fsp->write_time_forced) {
 252                 /*
 253                  * No point - "sticky" write times
 254                  * in effect.
 255                  */
 256                 return;
 257         }
 258
 259         TALLOC_FREE(fsp->update_write_time_event);
 260         DEBUG(5, ("Update write time immediate on %s\n",
 261                   fsp_str_dbg(fsp)));
 262
 263         /* After an immediate update, reset the trigger. */
 264         fsp->update_write_time_triggered = true;
 265         fsp->update_write_time_on_close = false;
 266
 267         ZERO_STRUCT(ft);
 268         ft.mtime = timespec_current();
 269
 270         /* Update the time in the open file db. */
 271         (void)set_write_time(fsp->file_id, ft.mtime);
 272
 273         /* Now set on disk - takes care of notify. */
 274         (void)smb_set_file_time(fsp->conn, fsp, fsp->fsp_name, &ft, false);
 275 }
 276
 277 void mark_file_modified(files_struct *fsp)
 278 {
 279         int dosmode;
 280
 281         if (fsp->modified) {
 282                 return;
 283         }
 284
 285         fsp->modified = true;
 286
 287         if (SMB_VFS_FSTAT(fsp, &fsp->fsp_name->st) != 0) {
 288                 return;
 289         }
 290         trigger_write_time_update(fsp);
 291
 292         if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
 293                 return;
 294         }
 295         if (!(lp_store_dos_attributes(SNUM(fsp->conn)) ||
 296               MAP_ARCHIVE(fsp->conn))) {
 297                 return;
 298         }
 299
 300         dosmode = dos_mode(fsp->conn, fsp->fsp_name);
 301         if (IS_DOS_ARCHIVE(dosmode)) {
 302                 return;
 303         }
 304         file_set_dosmode(fsp->conn, fsp->fsp_name,
 305                          dosmode | FILE_ATTRIBUTE_ARCHIVE, NULL, false);
 306 }
 307
 308 /****************************************************************************
 309  Write to a file.
 310 ****************************************************************************/
 311
 312 ssize_t write_file(struct smb_request *req,
 313                         files_struct *fsp,
 314                         const char *data,
 315                         off_t pos,
 316                         size_t n)
 317 {
 318         struct write_cache *wcp = fsp->wcp;
 319         ssize_t total_written = 0;
 320         int write_path = -1;
 321
 322         if (fsp->print_file) {
 323                 uint32_t t;
 324                 int ret;
 325
 326                 ret = print_spool_write(fsp, data, n, pos, &t);
 327                 if (ret) {
 328                         errno = ret;
 329                         return -1;
 330                 }
 331                 return t;
 332         }
 333
 334         if (!fsp->can_write) {
 335                 errno = EPERM;
 336                 return -1;
 337         }
 338
 339         /*
 340          * If this is the first write and we have an exclusive oplock
 341          * then setup the write cache.
 342          */
 343
 344         if (!fsp->modified &&
 345             EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) &&
 346             (wcp == NULL)) {
 347                 /*
 348                  * Note: no write cache with leases!
 349                  * as the handles would have to share the write cache
 350                  * that's possible but an improvement for another day...
 351                  */
 352                 setup_write_cache(fsp, fsp->fsp_name->st.st_ex_size);
 353                 wcp = fsp->wcp;
 354         }
 355
 356         mark_file_modified(fsp);
 357
 358         DO_PROFILE_INC(writecache_total_writes);
 359         if (!fsp->oplock_type) {
 360                 DO_PROFILE_INC(writecache_non_oplock_writes);
 361         }
 362
 363         /*
 364          * If this file is level II oplocked then we need
 365          * to grab the shared memory lock and inform all
 366          * other files with a level II lock that they need
 367          * to flush their read caches. We keep the lock over
 368          * the shared memory area whilst doing this.
 369          */
 370
 371         /* This should actually be improved to span the write. */
 372         contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WRITE);
 373         contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WRITE);
 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, SAMBA_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 "
 420                                   "size %.0f 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) &&
 437                     (pos <= wcp->offset + wcp->data_size)) {
 438
 439                         /* ASCII art.... JRA.
 440
 441       +--------------+-----
 442       | Cached data  | Rest of allocated cache buffer....
 443       +--------------+-----
 444
 445             +-------------------+
 446             | Data to write     |
 447             +-------------------+
 448
 449                         */
 450
 451                         /*
 452                          * Start of write overlaps or abutts the existing data.
 453                          */
 454
 455                         size_t data_used;
 456
 457                         data_used = MIN((wcp->alloc_size - (pos - wcp->offset)),
 458                                         n);
 459
 460                         memcpy(wcp->data + (pos - wcp->offset), data,
 461                                data_used);
 462
 463                         /*
 464                          * Update the current buffer size with the new data.
 465                          */
 466
 467                         if(pos + data_used > wcp->offset + wcp->data_size) {
 468                                 wcp->data_size = pos + data_used - wcp->offset;
 469                         }
 470
 471                         /*
 472                          * Update the file size if changed.
 473                          */
 474
 475                         if (wcp->offset + wcp->data_size > wcp->file_size) {
 476                                 if (wcp_file_size_change(fsp) == -1) {
 477                                         return -1;
 478                                 }
 479                         }
 480
 481                         /*
 482                          * If we used all the data then
 483                          * return here.
 484                          */
 485
 486                         if(n == data_used) {
 487                                 return n;
 488                         } else {
 489                                 cache_flush_needed = True;
 490                         }
 491                         /*
 492                          * Move the start of data forward by the amount used,
 493                          * cut down the amount left by the same amount.
 494                          */
 495
 496                         data += data_used;
 497                         pos += data_used;
 498                         n -= data_used;
 499
 500                         DO_PROFILE_INC(writecache_abutted_writes);
 501                         total_written = data_used;
 502
 503                         write_path = 1;
 504
 505                 } else if ((pos < wcp->offset) &&
 506                            (pos + n > wcp->offset) &&
 507                            (pos + n <= wcp->offset + wcp->alloc_size)) {
 508
 509                         /* ASCII art.... JRA.
 510
 511                         +---------------+
 512                         | Cache buffer  |
 513                         +---------------+
 514
 515             +-------------------+
 516             | Data to write     |
 517             +-------------------+
 518
 519                         */
 520
 521                         /*
 522                          * End of write overlaps the existing data.
 523                          */
 524
 525                         size_t data_used = pos + n - wcp->offset;
 526
 527                         memcpy(wcp->data, data + n - data_used, data_used);
 528
 529                         /*
 530                          * Update the current buffer size with the new data.
 531                          */
 532
 533                         if(pos + n > wcp->offset + wcp->data_size) {
 534                                 wcp->data_size = pos + n - wcp->offset;
 535                         }
 536
 537                         /*
 538                          * Update the file size if changed.
 539                          */
 540
 541                         if (wcp->offset + wcp->data_size > wcp->file_size) {
 542                                 if (wcp_file_size_change(fsp) == -1) {
 543                                         return -1;
 544                                 }
 545                         }
 546
 547                         /*
 548                          * We don't need to move the start of data, but we
 549                          * cut down the amount left by the amount used.
 550                          */
 551
 552                         n -= data_used;
 553
 554                         /*
 555                          * We cannot have used all the data here.
 556                          */
 557
 558                         cache_flush_needed = True;
 559
 560                         DO_PROFILE_INC(writecache_abutted_writes);
 561                         total_written = data_used;
 562
 563                         write_path = 2;
 564
 565                 } else if ((pos >= wcp->file_size) &&
 566                            (wcp->offset + wcp->data_size == wcp->file_size) &&
 567                            (pos > wcp->offset + wcp->data_size) &&
 568                            (pos < wcp->offset + wcp->alloc_size) ) {
 569
 570                         /* ASCII art.... JRA.
 571
 572                        End of file ---->|
 573
 574                         +---------------+---------------+
 575                         | Cached data   | Cache buffer  |
 576                         +---------------+---------------+
 577
 578                                               +-------------------+
 579                                               | Data to write     |
 580                                               +-------------------+
 581
 582                         */
 583
 584                         /*
 585                          * Non-contiguous write part of which fits within
 586                          * the cache buffer and is extending the file
 587                          * and the cache contents reflect the current
 588                          * data up to the current end of the file.
 589                          */
 590
 591                         size_t data_used;
 592
 593                         if(pos + n <= wcp->offset + wcp->alloc_size) {
 594                                 data_used = n;
 595                         } else {
 596                                 data_used = wcp->offset+wcp->alloc_size-pos;
 597                         }
 598
 599                         /*
 600                          * Fill in the non-continuous area with zeros.
 601                          */
 602
 603                         memset(wcp->data + wcp->data_size, '\0',
 604                                 pos - (wcp->offset + wcp->data_size) );
 605
 606                         memcpy(wcp->data + (pos - wcp->offset), data,
 607                                data_used);
 608
 609                         /*
 610                          * Update the current buffer size with the new data.
 611                          */
 612
 613                         if(pos + data_used > wcp->offset + wcp->data_size) {
 614                                 wcp->data_size = pos + data_used - wcp->offset;
 615                         }
 616
 617                         /*
 618                          * Update the file size if changed.
 619                          */
 620
 621                         if (wcp->offset + wcp->data_size > wcp->file_size) {
 622                                 if (wcp_file_size_change(fsp) == -1) {
 623                                         return -1;
 624                                 }
 625                         }
 626
 627                         /*
 628                          * If we used all the data then
 629                          * return here.
 630                          */
 631
 632                         if(n == data_used) {
 633                                 return n;
 634                         } else {
 635                                 cache_flush_needed = True;
 636                         }
 637
 638                         /*
 639                          * Move the start of data forward by the amount used,
 640                          * cut down the amount left by the same amount.
 641                          */
 642
 643                         data += data_used;
 644                         pos += data_used;
 645                         n -= data_used;
 646
 647                         DO_PROFILE_INC(writecache_abutted_writes);
 648                         total_written = data_used;
 649
 650                         write_path = 3;
 651
 652                 } else if ( (pos >= wcp->file_size) &&
 653                             (n == 1) &&
 654                             (wcp->file_size == wcp->offset + wcp->data_size) &&
 655                             (pos < wcp->file_size + wcp->alloc_size)) {
 656
 657                         /*
 658
 659                 End of file ---->|
 660
 661                  +---------------+---------------+
 662                  | Cached data   | Cache buffer  |
 663                  +---------------+---------------+
 664
 665                                  |<------- allocated size ---------------->|
 666
 667                                                          +--------+
 668                                                          | 1 Byte |
 669                                                          +--------+
 670
 671                         MS-Office seems to do this a lot to determine if
 672                         there's enough space on the filesystem to write a new
 673                         file.
 674
 675                         Change to :
 676
 677                 End of file ---->|
 678                                  +-----------------------+--------+
 679                                  | Zeroed Cached data    | 1 Byte |
 680                                  +-----------------------+--------+
 681                         */
 682
 683                         flush_write_cache(fsp, SAMBA_WRITE_FLUSH);
 684                         wcp->offset = wcp->file_size;
 685                         wcp->data_size = pos - wcp->file_size + 1;
 686                         memset(wcp->data, '\0', wcp->data_size);
 687                         memcpy(wcp->data + wcp->data_size-1, data, 1);
 688
 689                         /*
 690                          * Update the file size if changed.
 691                          */
 692
 693                         if (wcp->offset + wcp->data_size > wcp->file_size) {
 694                                 if (wcp_file_size_change(fsp) == -1) {
 695                                         return -1;
 696                                 }
 697                         }
 698
 699                         return n;
 700
 701                 } else {
 702
 703                         /* ASCII art..... JRA.
 704
 705    Case 1).
 706
 707                         +---------------+---------------+
 708                         | Cached data   | Cache buffer  |
 709                         +---------------+---------------+
 710
 711                                                               +---------------+
 712                                                               | Data to write |
 713                                                               +---------------+
 714
 715    Case 2).
 716
 717                            +---------------+---------------+
 718                            | Cached data   | Cache buffer  |
 719                            +---------------+---------------+
 720
 721    +-------------------+
 722    | Data to write     |
 723    +-------------------+
 724
 725     Case 3).
 726
 727                            +---------------+---------------+
 728                            | Cached data   | Cache buffer  |
 729                            +---------------+---------------+
 730
 731                   +-----------------------------------------------------+
 732                   | Data to write                                       |
 733                   +-----------------------------------------------------+
 734
 735                   */
 736
 737                         /*
 738                          * Write is bigger than buffer, or there is no
 739                          * overlap on the low or high ends.
 740                          */
 741
 742                         DEBUG(9,("write_file: non cacheable write : fd = %d, "
 743                                  "pos = %.0f, len = %u, "
 744                                  "current cache pos = %.0f len = %u\n",
 745                                  fsp->fh->fd, (double)pos, (unsigned int)n,
 746                                  (double)wcp->offset,
 747                                  (unsigned int)wcp->data_size ));
 748
 749                         /*
 750                          * If write would fit in the cache, and is
 751                          * larger than the data already in the cache,
 752                          * flush the cache and preferentially copy the
 753                          * data new data into it. Otherwise just write
 754                          * the data directly.
 755                          */
 756
 757                         if ( n <= wcp->alloc_size && n > wcp->data_size) {
 758                                 cache_flush_needed = True;
 759                         } else {
 760                                 ssize_t ret = real_write_file(NULL, fsp, data,
 761                                                               pos, n);
 762
 763                                 /*
 764                                  * If the write overlaps the entire
 765                                  * cache, then discard the current
 766                                  * contents of the cache.  Fix from
 767                                  * Rasmus Borup Hansen rbh@math.ku.dk.
 768                                  */
 769
 770                                 if ((pos <= wcp->offset) &&
 771                                     (pos + n >= wcp->offset+wcp->data_size)) {
 772                                         DEBUG(9,("write_file: discarding "
 773                                                  "overwritten write cache: "
 774                                                  "fd = %d, off=%.0f, "
 775                                                  "size=%u\n", fsp->fh->fd,
 776                                                  (double)wcp->offset,
 777                                                  (unsigned)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, ("SAMBA_WRITE_FLUSH:%d: due to noncontinuous "
 799                                   "write: fd = %d, size = %.0f, pos = %.0f, "
 800                                   "n = %u, wcp->offset=%.0f, "
 801                                   "wcp->data_size=%u\n",
 802                                   write_path, fsp->fh->fd,
 803                                   (double)wcp->file_size, (double)pos,
 804                                   (unsigned int)n, (double)wcp->offset,
 805                                   (unsigned int)wcp->data_size ));
 806
 807                         flush_write_cache(fsp, SAMBA_WRITE_FLUSH);
 808                 }
 809         }
 810
 811         /*
 812          * If the write request is bigger than the cache
 813          * size, write it all out.
 814          */
 815
 816         if (n > wcp->alloc_size ) {
 817                 ssize_t ret = real_write_file(NULL,fsp, data, pos, n);
 818                 if (ret == -1) {
 819                         return -1;
 820                 }
 821
 822                 if (pos + ret > wcp->file_size) {
 823                         wcp->file_size = pos + n;
 824                 }
 825
 826                 DO_PROFILE_INC(writecache_direct_writes);
 827                 return total_written + n;
 828         }
 829
 830         /*
 831          * If there's any data left, cache it.
 832          */
 833
 834         if (n) {
 835                 DO_PROFILE_INC(writecache_cached_writes);
 836                 if (wcp->data_size) {
 837                         DO_PROFILE_INC(writecache_abutted_writes);
 838                 } else {
 839                         DO_PROFILE_INC(writecache_init_writes);
 840                 }
 841
 842                 if ((wcp->data_size == 0)
 843                     && (pos > wcp->file_size)
 844                     && (pos + n <= wcp->file_size + wcp->alloc_size)) {
 845                         /*
 846                          * This is a write completely beyond the
 847                          * current EOF, but within reach of the write
 848                          * cache. We expect fill-up writes pretty
 849                          * soon, so it does not make sense to start
 850                          * the write cache at the current
 851                          * offset. These fill-up writes would trigger
 852                          * separate pwrites or even unnecessary cache
 853                          * flushes because they overlap if this is a
 854                          * one-byte allocating write.
 855                          */
 856                         wcp->offset = wcp->file_size;
 857                         wcp->data_size = pos - wcp->file_size;
 858                         memset(wcp->data, 0, wcp->data_size);
 859                 }
 860
 861                 memcpy(wcp->data+wcp->data_size, data, n);
 862                 if (wcp->data_size == 0) {
 863                         wcp->offset = pos;
 864                 }
 865                 wcp->data_size += n;
 866
 867                 /*
 868                  * Update the file size if changed.
 869                  */
 870
 871                 if (wcp->offset + wcp->data_size > wcp->file_size) {
 872                         if (wcp_file_size_change(fsp) == -1) {
 873                                 return -1;
 874                         }
 875                 }
 876                 DEBUG(9, ("wcp->offset = %.0f wcp->data_size = %u cache "
 877                           "return %u\n",
 878                           (double)wcp->offset, (unsigned int)wcp->data_size,
 879                           (unsigned int)n));
 880
 881                 total_written += n;
 882                 return total_written; /* .... that's a write :) */
 883         }
 884
 885         return total_written;
 886 }
 887
 888 /****************************************************************************
 889  Delete the write cache structure.
 890 ****************************************************************************/
 891
 892 void delete_write_cache(files_struct *fsp)
 893 {
 894         struct write_cache *wcp;
 895
 896         if(!fsp) {
 897                 return;
 898         }
 899
 900         if(!(wcp = fsp->wcp)) {
 901                 return;
 902         }
 903
 904         DO_PROFILE_INC(writecache_deallocations);
 905         allocated_write_caches--;
 906
 907         SMB_ASSERT(wcp->data_size == 0);
 908
 909         SAFE_FREE(wcp->data);
 910         SAFE_FREE(fsp->wcp);
 911
 912         DEBUG(10,("delete_write_cache: File %s deleted write cache\n",
 913                   fsp_str_dbg(fsp)));
 914 }
 915
 916 /****************************************************************************
 917  Setup the write cache structure.
 918 ****************************************************************************/
 919
 920 static bool setup_write_cache(files_struct *fsp, off_t file_size)
 921 {
 922         ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
 923         struct write_cache *wcp;
 924
 925         if (allocated_write_caches >= MAX_WRITE_CACHES) {
 926                 return False;
 927         }
 928
 929         if(alloc_size == 0 || fsp->wcp) {
 930                 return False;
 931         }
 932
 933         if((wcp = SMB_MALLOC_P(struct write_cache)) == NULL) {
 934                 DEBUG(0,("setup_write_cache: malloc fail.\n"));
 935                 return False;
 936         }
 937
 938         wcp->file_size = file_size;
 939         wcp->offset = 0;
 940         wcp->alloc_size = alloc_size;
 941         wcp->data_size = 0;
 942         if((wcp->data = (char *)SMB_MALLOC(wcp->alloc_size)) == NULL) {
 943                 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
 944                         (unsigned int)wcp->alloc_size ));
 945                 SAFE_FREE(wcp);
 946                 return False;
 947         }
 948
 949         memset(wcp->data, '\0', wcp->alloc_size );
 950
 951         fsp->wcp = wcp;
 952         DO_PROFILE_INC(writecache_allocations);
 953         allocated_write_caches++;
 954
 955         DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
 956                   fsp_str_dbg(fsp), (unsigned long)wcp->alloc_size));
 957
 958         return True;
 959 }
 960
 961 /****************************************************************************
 962  Cope with a size change.
 963 ****************************************************************************/
 964
 965 void set_filelen_write_cache(files_struct *fsp, off_t file_size)
 966 {
 967         if(fsp->wcp) {
 968                 /* The cache *must* have been flushed before we do this. */
 969                 if (fsp->wcp->data_size != 0) {
 970                         char *msg;
 971                         if (asprintf(&msg, "set_filelen_write_cache: size change "
 972                                  "on file %s with write cache size = %lu\n",
 973                                  fsp->fsp_name->base_name,
 974                                  (unsigned long)fsp->wcp->data_size) != -1) {
 975                                 smb_panic(msg);
 976                         } else {
 977                                 smb_panic("set_filelen_write_cache");
 978                         }
 979                 }
 980                 fsp->wcp->file_size = file_size;
 981         }
 982 }
 983
 984 /*******************************************************************
 985  Flush a write cache struct to disk.
 986 ********************************************************************/
 987
 988 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
 989 {
 990         struct write_cache *wcp = fsp->wcp;
 991         size_t data_size;
 992         ssize_t ret;
 993
 994         if(!wcp || !wcp->data_size) {
 995                 return 0;
 996         }
 997
 998         data_size = wcp->data_size;
 999         wcp->data_size = 0;
1000
1001         switch (reason) {
1002         case SAMBA_SEEK_FLUSH:
1003                 DO_PROFILE_INC(writecache_flush_reason_seek);
1004                 break;
1005         case SAMBA_READ_FLUSH:
1006                 DO_PROFILE_INC(writecache_flush_reason_read);
1007                 break;
1008         case SAMBA_WRITE_FLUSH:
1009                 DO_PROFILE_INC(writecache_flush_reason_write);;
1010                 break;
1011         case SAMBA_READRAW_FLUSH:
1012                 DO_PROFILE_INC(writecache_flush_reason_readraw);
1013                 break;
1014         case SAMBA_OPLOCK_RELEASE_FLUSH:
1015                 DO_PROFILE_INC(writecache_flush_reason_oplock);
1016                 break;
1017         case SAMBA_CLOSE_FLUSH:
1018                 DO_PROFILE_INC(writecache_flush_reason_close);
1019                 break;
1020         case SAMBA_SYNC_FLUSH:
1021                 DO_PROFILE_INC(writecache_flush_reason_sync);
1022                 break;
1023         case SAMBA_SIZECHANGE_FLUSH:
1024                 DO_PROFILE_INC(writecache_flush_reason_sizechange);
1025                 break;
1026         default:
1027                 break;
1028         }
1029
1030         DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
1031                 fsp->fh->fd, (double)wcp->offset, (unsigned int)data_size));
1032
1033         if(data_size == wcp->alloc_size) {
1034                 DO_PROFILE_INC(writecache_perfect_writes);
1035         }
1036
1037         ret = real_write_file(NULL, fsp, wcp->data, wcp->offset, data_size);
1038
1039         /*
1040          * Ensure file size if kept up to date if write extends file.
1041          */
1042
1043         if ((ret != -1) && (wcp->offset + ret > wcp->file_size)) {
1044                 wcp->file_size = wcp->offset + ret;
1045         }
1046
1047         return ret;
1048 }
1049
1050 /*******************************************************************
1051 sync a file
1052 ********************************************************************/
1053
1054 NTSTATUS sync_file(connection_struct *conn, files_struct *fsp, bool write_through)
1055 {
1056         if (fsp->fh->fd == -1)
1057                 return NT_STATUS_INVALID_HANDLE;
1058
1059         if (lp_strict_sync(SNUM(conn)) &&
1060             (lp_sync_always(SNUM(conn)) || write_through)) {
1061                 int ret = flush_write_cache(fsp, SAMBA_SYNC_FLUSH);
1062                 if (ret == -1) {
1063                         return map_nt_error_from_unix(errno);
1064                 }
1065                 ret = smb_vfs_fsync_sync(fsp);
1066                 if (ret == -1) {
1067                         return map_nt_error_from_unix(errno);
1068                 }
1069         }
1070         return NT_STATUS_OK;
1071 }
1072
1073 /************************************************************
1074  Perform a stat whether a valid fd or not.
1075 ************************************************************/
1076
1077 int fsp_stat(files_struct *fsp)
1078 {
1079         if (fsp->fh->fd == -1) {
1080                 if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
1081                         return SMB_VFS_LSTAT(fsp->conn, fsp->fsp_name);
1082                 } else {
1083                         return SMB_VFS_STAT(fsp->conn, fsp->fsp_name);
1084                 }
1085         } else {
1086                 return SMB_VFS_FSTAT(fsp, &fsp->fsp_name->st);
1087         }
1088 }