source3/smbd/fileio.c

   1 #define OLD_NTDOMAIN 1
   2 /*
   3    Unix SMB/Netbios implementation.
   4    Version 1.9.
   5    read/write to a files_struct
   6    Copyright (C) Andrew Tridgell 1992-1998
   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 2 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, write to the Free Software
  20    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  21 */
  22
  23 #include "includes.h"
  24
  25 extern int DEBUGLEVEL;
  26
  27 static BOOL setup_write_cache(files_struct *, SMB_OFF_T);
  28
  29 /****************************************************************************
  30 seek a file. Try to avoid the seek if possible
  31 ****************************************************************************/
  32
  33 SMB_OFF_T seek_file(files_struct *fsp,SMB_OFF_T pos)
  34 {
  35   SMB_OFF_T offset = 0;
  36   SMB_OFF_T seek_ret;
  37
  38   if (fsp->print_file && lp_postscript(fsp->conn->service))
  39     offset = 3;
  40
  41   seek_ret = fsp->conn->vfs_ops.lseek(fsp->fd,pos+offset,SEEK_SET);
  42
  43   /*
  44    * We want to maintain the fiction that we can seek
  45    * on a fifo for file system purposes. This allows
  46    * people to set up UNIX fifo's that feed data to Windows
  47    * applications. JRA.
  48    */
  49
  50   if((seek_ret == -1) && (errno == ESPIPE)) {
  51     seek_ret = pos+offset;
  52     errno = 0;
  53   }
  54
  55   if((seek_ret == -1) || (seek_ret != pos+offset)) {
  56     DEBUG(0,("seek_file: sys_lseek failed. Error was %s\n", strerror(errno) ));
  57     fsp->pos = -1;
  58     return -1;
  59   }
  60
  61   fsp->pos = seek_ret - offset;
  62
  63   DEBUG(10,("seek_file: requested pos = %.0f, new pos = %.0f\n",
  64         (double)(pos+offset), (double)fsp->pos ));
  65
  66   return(fsp->pos);
  67 }
  68
  69 /****************************************************************************
  70  Read from write cache if we can.
  71 ****************************************************************************/
  72
  73 static unsigned int cache_read_hits;
  74
  75 BOOL read_from_write_cache(files_struct *fsp,char *data,SMB_OFF_T pos,size_t n)
  76 {
  77   write_cache *wcp = fsp->wcp;
  78
  79   if(!wcp)
  80     return False;
  81
  82   if(n > wcp->data_size || pos < wcp->offset || pos + n > wcp->offset + wcp->data_size)
  83     return False;
  84
  85   memcpy(data, wcp->data + (pos - wcp->offset), n);
  86
  87   cache_read_hits++;
  88
  89   return True;
  90 }
  91
  92 /****************************************************************************
  93 read from a file
  94 ****************************************************************************/
  95
  96 ssize_t read_file(files_struct *fsp,char *data,SMB_OFF_T pos,size_t n)
  97 {
  98   ssize_t ret=0,readret;
  99
 100   /* you can't read from print files */
 101   if (fsp->print_file) {
 102           return -1;
 103   }
 104
 105   /*
 106    * Serve from write cache if we can.
 107    */
 108   if(read_from_write_cache(fsp, data, pos, n))
 109     return n;
 110
 111   flush_write_cache(fsp, READ_FLUSH);
 112
 113   if (seek_file(fsp,pos) == -1) {
 114     DEBUG(3,("read_file: Failed to seek to %.0f\n",(double)pos));
 115     return(ret);
 116   }
 117
 118   if (n > 0) {
 119     readret = fsp->conn->vfs_ops.read(fsp->fd,data,n);
 120     if (readret == -1)
 121       return -1;
 122     if (readret > 0) ret += readret;
 123   }
 124
 125   return(ret);
 126 }
 127
 128 /* Write cache static counters. */
 129
 130 static unsigned int abutted_writes;
 131 static unsigned int total_writes;
 132 static unsigned int non_oplock_writes;
 133 static unsigned int direct_writes;
 134 static unsigned int init_writes;
 135 static unsigned int flushed_writes;
 136 static unsigned int num_perfect_writes;
 137 static unsigned int flush_reasons[NUM_FLUSH_REASONS];
 138
 139 /* how many write cache buffers have been allocated */
 140 static unsigned int allocated_write_caches;
 141 static unsigned int num_write_caches;
 142
 143 /****************************************************************************
 144  *Really* write to a file
 145 ****************************************************************************/
 146
 147 static ssize_t real_write_file(files_struct *fsp,char *data,SMB_OFF_T pos, size_t n)
 148 {
 149   if ((pos != -1) && (seek_file(fsp,pos) == -1))
 150     return -1;
 151
 152   return write_data(fsp->fd,data,n);
 153 }
 154
 155 /****************************************************************************
 156 write to a file
 157 ****************************************************************************/
 158
 159 ssize_t write_file(files_struct *fsp, char *data, SMB_OFF_T pos, size_t n)
 160 {
 161   write_cache *wcp = fsp->wcp;
 162   ssize_t total_written = 0;
 163   int write_path = -1;
 164
 165   if (fsp->print_file) {
 166           return print_job_write(fsp->print_jobid, data, n);
 167   }
 168
 169   if (!fsp->can_write) {
 170     errno = EPERM;
 171     return(0);
 172   }
 173
 174   if (!fsp->modified) {
 175     SMB_STRUCT_STAT st;
 176     fsp->modified = True;
 177
 178     if (fsp->conn->vfs_ops.fstat(fsp->fd,&st) == 0) {
 179       int dosmode = dos_mode(fsp->conn,fsp->fsp_name,&st);
 180       if (MAP_ARCHIVE(fsp->conn) && !IS_DOS_ARCHIVE(dosmode)) {
 181         file_chmod(fsp->conn,fsp->fsp_name,dosmode | aARCH,&st);
 182       }
 183
 184       /*
 185        * If this is the first write and we have an exclusive oplock then setup
 186        * the write cache.
 187        */
 188
 189       if ((fsp->oplock_type == EXCLUSIVE_OPLOCK) && !wcp) {
 190         setup_write_cache(fsp, st.st_size);
 191         wcp = fsp->wcp;
 192       }
 193     }
 194   }
 195
 196   total_writes++;
 197   if (!fsp->oplock_type) {
 198     non_oplock_writes++;
 199   }
 200
 201   /*
 202    * If this file is level II oplocked then we need
 203    * to grab the shared memory lock and inform all
 204    * other files with a level II lock that they need
 205    * to flush their read caches. We keep the lock over
 206    * the shared memory area whilst doing this.
 207    */
 208
 209   if (LEVEL_II_OPLOCK_TYPE(fsp->oplock_type)) {
 210     share_mode_entry *share_list = NULL;
 211     pid_t pid = sys_getpid();
 212     int token = -1;
 213     int num_share_modes = 0;
 214     int i;
 215
 216     if (lock_share_entry_fsp(fsp) == False) {
 217       DEBUG(0,("write_file: failed to lock share mode entry for file %s.\n", fsp->fsp_name ));
 218     }
 219
 220     num_share_modes = get_share_modes(fsp->conn, fsp->dev, fsp->inode, &share_list);
 221
 222     for(i = 0; i < num_share_modes; i++) {
 223       share_mode_entry *share_entry = &share_list[i];
 224
 225       /*
 226        * As there could have been multiple writes waiting at the lock_share_entry
 227        * gate we may not be the first to enter. Hence the state of the op_types
 228        * in the share mode entries may be partly NO_OPLOCK and partly LEVEL_II
 229        * oplock. It will do no harm to re-send break messages to those smbd's
 230        * that are still waiting their turn to remove their LEVEL_II state, and
 231        * also no harm to ignore existing NO_OPLOCK states. JRA.
 232        */
 233
 234       if (share_entry->op_type == NO_OPLOCK)
 235         continue;
 236
 237       /* Paranoia .... */
 238       if (EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type)) {
 239         DEBUG(0,("write_file: PANIC. share mode entry %d is an exlusive oplock !\n", i ));
 240         unlock_share_entry(fsp->conn, fsp->dev, fsp->inode);
 241         abort();
 242       }
 243
 244       /*
 245        * Check if this is a file we have open (including the
 246        * file we've been called to do write_file on. If so
 247        * then break it directly without releasing the lock.
 248        */
 249
 250       if (pid == share_entry->pid) {
 251         files_struct *new_fsp = file_find_dit(fsp->dev, fsp->inode, &share_entry->time);
 252
 253         /* Paranoia check... */
 254         if(new_fsp == NULL) {
 255           DEBUG(0,("write_file: PANIC. share mode entry %d is not a local file !\n", i ));
 256           unlock_share_entry(fsp->conn, fsp->dev, fsp->inode);
 257           abort();
 258         }
 259         oplock_break_level2(new_fsp, True, token);
 260
 261       } else {
 262
 263         /*
 264          * This is a remote file and so we send an asynchronous
 265          * message.
 266          */
 267
 268         request_oplock_break(share_entry, fsp->dev, fsp->inode);
 269       }
 270     }
 271
 272     free((char *)share_list);
 273     unlock_share_entry_fsp(fsp);
 274   }
 275
 276   /* Paranoia check... */
 277   if (LEVEL_II_OPLOCK_TYPE(fsp->oplock_type)) {
 278     DEBUG(0,("write_file: PANIC. File %s still has a level II oplock.\n", fsp->fsp_name));
 279     abort();
 280   }
 281
 282   if (total_writes % 500 == 0) {
 283     DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u flushes=%u total=%u \
 284 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
 285          init_writes, abutted_writes, flushed_writes, total_writes, non_oplock_writes,
 286          allocated_write_caches,
 287          num_write_caches, direct_writes, num_perfect_writes, cache_read_hits ));
 288
 289     DEBUG(3,("WRITECACHE: SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
 290     flush_reasons[SEEK_FLUSH],
 291     flush_reasons[READ_FLUSH],
 292     flush_reasons[WRITE_FLUSH],
 293     flush_reasons[READRAW_FLUSH],
 294     flush_reasons[OPLOCK_RELEASE_FLUSH],
 295     flush_reasons[CLOSE_FLUSH],
 296     flush_reasons[SYNC_FLUSH] ));
 297   }
 298
 299   if(!wcp) {
 300     direct_writes++;
 301     return real_write_file(fsp, data, pos, n);
 302   }
 303
 304   DEBUG(9,("write_file(fd=%d pos=%d size=%d) wofs=%d wsize=%d\n",
 305            fsp->fd, (int)pos, (int)n, (int)wcp->offset, (int)wcp->data_size));
 306
 307   /*
 308    * If we have active cache and it isn't contiguous then we flush.
 309    * NOTE: There is a small problem with running out of disk ....
 310    */
 311
 312   if (wcp->data_size) {
 313
 314     BOOL cache_flush_needed = False;
 315
 316     if ((pos >= wcp->offset) && (pos <= wcp->offset + wcp->data_size)) {
 317
 318       /*
 319        * Start of write overlaps or abutts the existing data.
 320        */
 321
 322       size_t data_used = MIN((wcp->alloc_size - (pos - wcp->offset)), n);
 323
 324       memcpy(wcp->data + (pos - wcp->offset), data, data_used);
 325
 326       /*
 327        * Update the current buffer size with the new data.
 328        */
 329
 330       if(pos + data_used > wcp->offset + wcp->data_size)
 331         wcp->data_size = pos + data_used - wcp->offset;
 332
 333       /*
 334        * If we used all the data then
 335        * return here.
 336        */
 337
 338       if(n == data_used)
 339         return n;
 340       else
 341         cache_flush_needed = True;
 342
 343       /*
 344        * Move the start of data forward by the amount used,
 345        * cut down the amount left by the same amount.
 346        */
 347
 348       data += data_used;
 349       pos += data_used;
 350       n -= data_used;
 351
 352       abutted_writes++;
 353       total_written = data_used;
 354
 355       write_path = 1;
 356
 357     } else if ((pos < wcp->offset) && (pos + n > wcp->offset) &&
 358                (pos + n <= wcp->offset + wcp->alloc_size)) {
 359
 360       /*
 361        * End of write overlaps the existing data.
 362        */
 363
 364       size_t data_used = pos + n - wcp->offset;
 365
 366       memcpy(wcp->data, data + n - data_used, data_used);
 367
 368       /*
 369        * Update the current buffer size with the new data.
 370        */
 371
 372       if(pos + n > wcp->offset + wcp->data_size)
 373         wcp->data_size = pos + n - wcp->offset;
 374
 375       /*
 376        * We don't need to move the start of data, but we
 377        * cut down the amount left by the amount used.
 378        */
 379
 380       n -= data_used;
 381
 382       /*
 383        * We cannot have used all the data here.
 384        */
 385
 386       cache_flush_needed = True;
 387
 388       abutted_writes++;
 389       total_written = data_used;
 390
 391       write_path = 2;
 392
 393     } else if ( (pos >= wcp->file_size) &&
 394                 (pos > wcp->offset + wcp->data_size) &&
 395                 (pos < wcp->offset + wcp->alloc_size) ) {
 396
 397       /*
 398        * Non-contiguous write part of which fits within
 399        * the cache buffer and is extending the file.
 400        */
 401
 402       size_t data_used;
 403
 404       if(pos + n <= wcp->offset + wcp->alloc_size)
 405         data_used = n;
 406       else
 407         data_used = wcp->offset + wcp->alloc_size - pos;
 408
 409       /*
 410        * Fill in the non-continuous area with zeros.
 411        */
 412
 413       memset(wcp->data + wcp->data_size, '\0',
 414              pos - (wcp->offset + wcp->data_size) );
 415
 416       memcpy(wcp->data + (pos - wcp->offset), data, data_used);
 417
 418       /*
 419        * Update the current buffer size with the new data.
 420        */
 421
 422       if(pos + data_used > wcp->offset + wcp->data_size)
 423         wcp->data_size = pos + data_used - wcp->offset;
 424
 425       /*
 426        * Update the known file length.
 427        */
 428
 429       wcp->file_size = wcp->offset + wcp->data_size;
 430
 431 #if 0
 432       if (set_filelen(fsp->fd, wcp->file_size) == -1) {
 433         DEBUG(0,("write_file: error %s in setting file to length %.0f\n",
 434           strerror(errno), (double)wcp->file_size ));
 435         return -1;
 436       }
 437 #endif
 438
 439       /*
 440        * If we used all the data then
 441        * return here.
 442        */
 443
 444       if(n == data_used)
 445         return n;
 446       else
 447         cache_flush_needed = True;
 448
 449       /*
 450        * Move the start of data forward by the amount used,
 451        * cut down the amount left by the same amount.
 452        */
 453
 454       data += data_used;
 455       pos += data_used;
 456       n -= data_used;
 457
 458       abutted_writes++;
 459       total_written = data_used;
 460
 461       write_path = 3;
 462
 463     } else {
 464
 465       /*
 466        * Write is bigger than buffer, or there is no overlap on the
 467        * low or high ends.
 468        */
 469
 470       DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
 471 len = %u\n",fsp->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size ));
 472
 473       /*
 474        * Update the file size if needed.
 475        */
 476
 477       if(pos + n > wcp->file_size)
 478         wcp->file_size = pos + n;
 479
 480       /*
 481        * If write would fit in the cache, and is larger than
 482        * the data already in the cache, flush the cache and
 483        * preferentially copy the data new data into it. Otherwise
 484        * just write the data directly.
 485        */
 486
 487       if ( n <= wcp->alloc_size && n > wcp->data_size) {
 488         cache_flush_needed = True;
 489       } else {
 490         direct_writes++;
 491         return real_write_file(fsp, data, pos, n);
 492       }
 493
 494       write_path = 4;
 495
 496     }
 497
 498     if(wcp->data_size > wcp->file_size)
 499       wcp->file_size = wcp->data_size;
 500
 501     if (cache_flush_needed) {
 502       flushed_writes++;
 503
 504       DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
 505 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
 506              write_path, fsp->fd, (double)wcp->file_size, (double)pos, (unsigned int)n,
 507              (double)wcp->offset, (unsigned int)wcp->data_size ));
 508
 509       flush_write_cache(fsp, WRITE_FLUSH);
 510     }
 511   }
 512
 513   /*
 514    * If the write request is bigger than the cache
 515    * size, write it all out.
 516    */
 517
 518   if (n > wcp->alloc_size ) {
 519     if(real_write_file(fsp, data, pos, n) == -1)
 520       return -1;
 521     direct_writes++;
 522     return total_written + n;
 523   }
 524
 525   /*
 526    * If there's any data left, cache it.
 527    */
 528
 529   if (n) {
 530     if (wcp->data_size) {
 531       abutted_writes++;
 532       DEBUG(9,("abutted write (%u)\n", abutted_writes));
 533     } else {
 534       init_writes++;
 535     }
 536     memcpy(wcp->data+wcp->data_size, data, n);
 537     if (wcp->data_size == 0) {
 538       wcp->offset = pos;
 539       num_write_caches++;
 540     }
 541     wcp->data_size += n;
 542     DEBUG(9,("cache return %u\n", (unsigned int)n));
 543     total_written += n;
 544     return total_written; /* .... that's a write :) */
 545   }
 546
 547   return total_written;
 548 }
 549
 550 /****************************************************************************
 551  Delete the write cache structure.
 552 ****************************************************************************/
 553
 554 void delete_write_cache(files_struct *fsp)
 555 {
 556   write_cache *wcp;
 557
 558   if(!fsp)
 559     return;
 560
 561   if(!(wcp = fsp->wcp))
 562     return;
 563
 564   allocated_write_caches--;
 565
 566   SMB_ASSERT(wcp->data_size == 0);
 567
 568   free(wcp->data);
 569   free(wcp);
 570
 571   fsp->wcp = NULL;
 572 }
 573
 574 /****************************************************************************
 575  Setup the write cache structure.
 576 ****************************************************************************/
 577
 578 static BOOL setup_write_cache(files_struct *fsp, SMB_OFF_T file_size)
 579 {
 580   ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
 581   write_cache *wcp;
 582
 583   if (allocated_write_caches >= MAX_WRITE_CACHES) return False;
 584
 585   if(alloc_size == 0 || fsp->wcp)
 586     return False;
 587
 588   if((wcp = (write_cache *)malloc(sizeof(write_cache))) == NULL) {
 589     DEBUG(0,("setup_write_cache: malloc fail.\n"));
 590     return False;
 591   }
 592
 593   wcp->file_size = file_size;
 594   wcp->offset = 0;
 595   wcp->alloc_size = alloc_size;
 596   wcp->data_size = 0;
 597   if((wcp->data = malloc(wcp->alloc_size)) == NULL) {
 598     DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
 599           (unsigned int)wcp->alloc_size ));
 600     free(wcp);
 601     return False;
 602   }
 603
 604   fsp->wcp = wcp;
 605   allocated_write_caches++;
 606
 607   return True;
 608 }
 609
 610 /****************************************************************************
 611  Cope with a size change.
 612 ****************************************************************************/
 613
 614 void set_filelen_write_cache(files_struct *fsp, SMB_OFF_T file_size)
 615 {
 616   if(fsp->wcp) {
 617     flush_write_cache(fsp, SIZECHANGE_FLUSH);
 618     fsp->wcp->file_size = file_size;
 619   }
 620 }
 621
 622 /*******************************************************************
 623  Flush a write cache struct to disk.
 624 ********************************************************************/
 625
 626 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
 627 {
 628   write_cache *wcp = fsp->wcp;
 629   size_t data_size;
 630
 631   if(!wcp || !wcp->data_size)
 632     return 0;
 633
 634   data_size = wcp->data_size;
 635   wcp->data_size = 0;
 636
 637   num_write_caches--;
 638
 639   flush_reasons[(int)reason]++;
 640
 641   DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
 642            fsp->fd, (double)wcp->offset, (unsigned int)data_size));
 643
 644   if(data_size == wcp->alloc_size)
 645     num_perfect_writes++;
 646
 647   return real_write_file(fsp, wcp->data, wcp->offset, data_size);
 648 }
 649
 650 /*******************************************************************
 651 sync a file
 652 ********************************************************************/
 653
 654 void sync_file(connection_struct *conn, files_struct *fsp)
 655 {
 656     if(lp_strict_sync(SNUM(conn)) && fsp->fd != -1) {
 657       flush_write_cache(fsp, SYNC_FLUSH);
 658       conn->vfs_ops.fsync(fsp->fd);
 659     }
 660 }
 661 #undef OLD_NTDOMAIN