r5739: sync for 3.0.12rc1 (current with SAMBA_3_0 r5738)
[Samba/gbeck.git] / source / smbd / fileio.c
blob3048c27fa25ee967de23ac2efdb6a2c6168a5017
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.
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.
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.
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.
23 #include "includes.h"
25 static BOOL setup_write_cache(files_struct *, SMB_OFF_T);
27 /****************************************************************************
28 Read from write cache if we can.
29 ****************************************************************************/
32 static BOOL read_from_write_cache(files_struct *fsp,char *data,SMB_OFF_T pos,size_t n)
34 write_cache *wcp = fsp->wcp;
36 if(!wcp)
37 return False;
39 if(n > wcp->data_size || pos < wcp->offset || pos + n > wcp->offset + wcp->data_size)
40 return False;
42 memcpy(data, wcp->data + (pos - wcp->offset), n);
44 DO_PROFILE_INC(writecache_read_hits);
46 return True;
49 /****************************************************************************
50 Read from a file.
51 ****************************************************************************/
53 ssize_t read_file(files_struct *fsp,char *data,SMB_OFF_T pos,size_t n)
55 ssize_t ret=0,readret;
57 /* you can't read from print files */
58 if (fsp->print_file)
59 return -1;
62 * Serve from write cache if we can.
65 if(read_from_write_cache(fsp, data, pos, n)) {
66 fsp->pos = pos + n;
67 fsp->position_information = fsp->pos;
68 return n;
71 flush_write_cache(fsp, READ_FLUSH);
73 fsp->pos = pos;
75 if (n > 0) {
76 #ifdef DMF_FIX
77 int numretries = 3;
78 tryagain:
79 readret = SMB_VFS_PREAD(fsp,fsp->fd,data,n,pos);
81 if (readret == -1) {
82 if ((errno == EAGAIN) && numretries) {
83 DEBUG(3,("read_file EAGAIN retry in 10 seconds\n"));
84 (void)sleep(10);
85 --numretries;
86 goto tryagain;
88 return -1;
90 #else /* NO DMF fix. */
91 readret = SMB_VFS_PREAD(fsp,fsp->fd,data,n,pos);
93 if (readret == -1)
94 return -1;
95 #endif
96 if (readret > 0)
97 ret += readret;
100 DEBUG(10,("read_file (%s): pos = %.0f, size = %lu, returned %lu\n",
101 fsp->fsp_name, (double)pos, (unsigned long)n, (long)ret ));
103 fsp->pos += ret;
104 fsp->position_information = fsp->pos;
106 return(ret);
109 /* how many write cache buffers have been allocated */
110 static unsigned int allocated_write_caches;
112 /****************************************************************************
113 *Really* write to a file.
114 ****************************************************************************/
116 static ssize_t real_write_file(files_struct *fsp,char *data,SMB_OFF_T pos, size_t n)
118 ssize_t ret;
120 if (pos == -1)
121 ret = vfs_write_data(fsp, data, n);
122 else {
123 fsp->pos = pos;
124 ret = vfs_pwrite_data(fsp, data, n, pos);
127 DEBUG(10,("real_write_file (%s): pos = %.0f, size = %lu, returned %ld\n",
128 fsp->fsp_name, (double)pos, (unsigned long)n, (long)ret ));
130 if (ret != -1) {
131 fsp->pos += ret;
134 * It turns out that setting the last write time from a Windows
135 * client stops any subsequent writes from updating the write time.
136 * Doing this after the write gives a race condition here where
137 * a stat may see the changed write time before we reset it here,
138 * but it's cheaper than having to store the write time in shared
139 * memory and look it up using dev/inode across all running smbd's.
140 * The 99% solution will hopefully be good enough in this case. JRA.
143 if (fsp->pending_modtime) {
144 set_filetime(fsp->conn, fsp->fsp_name, fsp->pending_modtime);
146 /* If we didn't get the "set modtime" call ourselves, we must
147 store the last write time to restore on close. JRA. */
148 if (!fsp->pending_modtime_owner) {
149 fsp->last_write_time = time(NULL);
153 /* Yes - this is correct - writes don't update this. JRA. */
154 /* Found by Samba4 tests. */
155 #if 0
156 fsp->position_information = fsp->pos;
157 #endif
160 return ret;
163 /****************************************************************************
164 write to a file
165 ****************************************************************************/
167 ssize_t write_file(files_struct *fsp, char *data, SMB_OFF_T pos, size_t n)
169 write_cache *wcp = fsp->wcp;
170 ssize_t total_written = 0;
171 int write_path = -1;
173 if (fsp->print_file) {
174 fstring sharename;
175 uint32 jobid;
177 if (!rap_to_pjobid(fsp->rap_print_jobid, sharename, &jobid)) {
178 DEBUG(3,("write_file: Unable to map RAP jobid %u to jobid.\n",
179 (unsigned int)fsp->rap_print_jobid ));
180 errno = EBADF;
181 return -1;
184 return print_job_write(SNUM(fsp->conn), jobid, data, n);
187 if (!fsp->can_write) {
188 errno = EPERM;
189 return(0);
192 if (!fsp->modified) {
193 SMB_STRUCT_STAT st;
194 fsp->modified = True;
196 if (SMB_VFS_FSTAT(fsp,fsp->fd,&st) == 0) {
197 int dosmode = dos_mode(fsp->conn,fsp->fsp_name,&st);
198 fsp->size = (SMB_BIG_UINT)st.st_size;
199 if ((lp_store_dos_attributes(SNUM(fsp->conn)) || MAP_ARCHIVE(fsp->conn)) && !IS_DOS_ARCHIVE(dosmode)) {
200 file_set_dosmode(fsp->conn,fsp->fsp_name,dosmode | aARCH,&st, False);
204 * If this is the first write and we have an exclusive oplock then setup
205 * the write cache.
208 if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && !wcp) {
209 setup_write_cache(fsp, st.st_size);
210 wcp = fsp->wcp;
215 #ifdef WITH_PROFILE
216 DO_PROFILE_INC(writecache_total_writes);
217 if (!fsp->oplock_type) {
218 DO_PROFILE_INC(writecache_non_oplock_writes);
220 #endif
223 * If this file is level II oplocked then we need
224 * to grab the shared memory lock and inform all
225 * other files with a level II lock that they need
226 * to flush their read caches. We keep the lock over
227 * the shared memory area whilst doing this.
230 release_level_2_oplocks_on_change(fsp);
232 #ifdef WITH_PROFILE
233 if (profile_p && profile_p->writecache_total_writes % 500 == 0) {
234 DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u total=%u \
235 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
236 profile_p->writecache_init_writes,
237 profile_p->writecache_abutted_writes,
238 profile_p->writecache_total_writes,
239 profile_p->writecache_non_oplock_writes,
240 profile_p->writecache_allocated_write_caches,
241 profile_p->writecache_num_write_caches,
242 profile_p->writecache_direct_writes,
243 profile_p->writecache_num_perfect_writes,
244 profile_p->writecache_read_hits ));
246 DEBUG(3,("WRITECACHE: Flushes SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
247 profile_p->writecache_flushed_writes[SEEK_FLUSH],
248 profile_p->writecache_flushed_writes[READ_FLUSH],
249 profile_p->writecache_flushed_writes[WRITE_FLUSH],
250 profile_p->writecache_flushed_writes[READRAW_FLUSH],
251 profile_p->writecache_flushed_writes[OPLOCK_RELEASE_FLUSH],
252 profile_p->writecache_flushed_writes[CLOSE_FLUSH],
253 profile_p->writecache_flushed_writes[SYNC_FLUSH] ));
255 #endif
257 if(!wcp) {
258 DO_PROFILE_INC(writecache_direct_writes);
259 total_written = real_write_file(fsp, data, pos, n);
260 if ((total_written != -1) && (pos + total_written > (SMB_OFF_T)fsp->size))
261 fsp->size = (SMB_BIG_UINT)(pos + total_written);
262 return total_written;
265 DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f wcp->data_size=%u\n",
266 fsp->fsp_name, fsp->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size));
268 fsp->pos = pos + n;
271 * If we have active cache and it isn't contiguous then we flush.
272 * NOTE: There is a small problem with running out of disk ....
275 if (wcp->data_size) {
277 BOOL cache_flush_needed = False;
279 if ((pos >= wcp->offset) && (pos <= wcp->offset + wcp->data_size)) {
281 /* ASCII art.... JRA.
283 +--------------+-----
284 | Cached data | Rest of allocated cache buffer....
285 +--------------+-----
287 +-------------------+
288 | Data to write |
289 +-------------------+
294 * Start of write overlaps or abutts the existing data.
297 size_t data_used = MIN((wcp->alloc_size - (pos - wcp->offset)), n);
299 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
302 * Update the current buffer size with the new data.
305 if(pos + data_used > wcp->offset + wcp->data_size)
306 wcp->data_size = pos + data_used - wcp->offset;
309 * Update the file size if changed.
312 if (wcp->offset + wcp->data_size > wcp->file_size) {
313 wcp->file_size = wcp->offset + wcp->data_size;
314 fsp->size = (SMB_BIG_UINT)wcp->file_size;
318 * If we used all the data then
319 * return here.
322 if(n == data_used)
323 return n;
324 else
325 cache_flush_needed = True;
328 * Move the start of data forward by the amount used,
329 * cut down the amount left by the same amount.
332 data += data_used;
333 pos += data_used;
334 n -= data_used;
336 DO_PROFILE_INC(writecache_abutted_writes);
337 total_written = data_used;
339 write_path = 1;
341 } else if ((pos < wcp->offset) && (pos + n > wcp->offset) &&
342 (pos + n <= wcp->offset + wcp->alloc_size)) {
344 /* ASCII art.... JRA.
346 +---------------+
347 | Cache buffer |
348 +---------------+
350 +-------------------+
351 | Data to write |
352 +-------------------+
357 * End of write overlaps the existing data.
360 size_t data_used = pos + n - wcp->offset;
362 memcpy(wcp->data, data + n - data_used, data_used);
365 * Update the current buffer size with the new data.
368 if(pos + n > wcp->offset + wcp->data_size)
369 wcp->data_size = pos + n - wcp->offset;
372 * Update the file size if changed.
375 if (wcp->offset + wcp->data_size > wcp->file_size) {
376 wcp->file_size = wcp->offset + wcp->data_size;
377 fsp->size = (SMB_BIG_UINT)wcp->file_size;
381 * We don't need to move the start of data, but we
382 * cut down the amount left by the amount used.
385 n -= data_used;
388 * We cannot have used all the data here.
391 cache_flush_needed = True;
393 DO_PROFILE_INC(writecache_abutted_writes);
394 total_written = data_used;
396 write_path = 2;
398 } else if ( (pos >= wcp->file_size) &&
399 (wcp->offset + wcp->data_size == wcp->file_size) &&
400 (pos > wcp->offset + wcp->data_size) &&
401 (pos < wcp->offset + wcp->alloc_size) ) {
403 /* ASCII art.... JRA.
405 End of file ---->|
407 +---------------+---------------+
408 | Cached data | Cache buffer |
409 +---------------+---------------+
411 +-------------------+
412 | Data to write |
413 +-------------------+
418 * Non-contiguous write part of which fits within
419 * the cache buffer and is extending the file
420 * and the cache contents reflect the current
421 * data up to the current end of the file.
424 size_t data_used;
426 if(pos + n <= wcp->offset + wcp->alloc_size)
427 data_used = n;
428 else
429 data_used = wcp->offset + wcp->alloc_size - pos;
432 * Fill in the non-continuous area with zeros.
435 memset(wcp->data + wcp->data_size, '\0',
436 pos - (wcp->offset + wcp->data_size) );
438 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
441 * Update the current buffer size with the new data.
444 if(pos + data_used > wcp->offset + wcp->data_size)
445 wcp->data_size = pos + data_used - wcp->offset;
448 * Update the file size if changed.
451 if (wcp->offset + wcp->data_size > wcp->file_size) {
452 wcp->file_size = wcp->offset + wcp->data_size;
453 fsp->size = (SMB_BIG_UINT)wcp->file_size;
457 * If we used all the data then
458 * return here.
461 if(n == data_used)
462 return n;
463 else
464 cache_flush_needed = True;
467 * Move the start of data forward by the amount used,
468 * cut down the amount left by the same amount.
471 data += data_used;
472 pos += data_used;
473 n -= data_used;
475 DO_PROFILE_INC(writecache_abutted_writes);
476 total_written = data_used;
478 write_path = 3;
480 } else {
482 /* ASCII art..... JRA.
484 Case 1).
486 +---------------+---------------+
487 | Cached data | Cache buffer |
488 +---------------+---------------+
490 +-------------------+
491 | Data to write |
492 +-------------------+
494 Case 2).
496 +---------------+---------------+
497 | Cached data | Cache buffer |
498 +---------------+---------------+
500 +-------------------+
501 | Data to write |
502 +-------------------+
504 Case 3).
506 +---------------+---------------+
507 | Cached data | Cache buffer |
508 +---------------+---------------+
510 +-----------------------------------------------------+
511 | Data to write |
512 +-----------------------------------------------------+
517 * Write is bigger than buffer, or there is no overlap on the
518 * low or high ends.
521 DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
522 len = %u\n",fsp->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size ));
525 * Update the file size if needed.
528 if(pos + n > wcp->file_size) {
529 wcp->file_size = pos + n;
530 fsp->size = (SMB_BIG_UINT)wcp->file_size;
534 * If write would fit in the cache, and is larger than
535 * the data already in the cache, flush the cache and
536 * preferentially copy the data new data into it. Otherwise
537 * just write the data directly.
540 if ( n <= wcp->alloc_size && n > wcp->data_size) {
541 cache_flush_needed = True;
542 } else {
543 ssize_t ret = real_write_file(fsp, data, pos, n);
546 * If the write overlaps the entire cache, then
547 * discard the current contents of the cache.
548 * Fix from Rasmus Borup Hansen rbh@math.ku.dk.
551 if ((pos <= wcp->offset) &&
552 (pos + n >= wcp->offset + wcp->data_size) ) {
553 DEBUG(9,("write_file: discarding overwritten write \
554 cache: fd = %d, off=%.0f, size=%u\n", fsp->fd, (double)wcp->offset, (unsigned int)wcp->data_size ));
555 wcp->data_size = 0;
558 DO_PROFILE_INC(writecache_direct_writes);
559 if (ret == -1)
560 return ret;
562 if (pos + ret > wcp->file_size) {
563 wcp->file_size = pos + ret;
564 fsp->size = (SMB_BIG_UINT)wcp->file_size;
567 return ret;
570 write_path = 4;
574 if(wcp->data_size > wcp->file_size) {
575 wcp->file_size = wcp->data_size;
576 fsp->size = (SMB_BIG_UINT)wcp->file_size;
579 if (cache_flush_needed) {
580 DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
581 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
582 write_path, fsp->fd, (double)wcp->file_size, (double)pos, (unsigned int)n,
583 (double)wcp->offset, (unsigned int)wcp->data_size ));
585 flush_write_cache(fsp, WRITE_FLUSH);
590 * If the write request is bigger than the cache
591 * size, write it all out.
594 if (n > wcp->alloc_size ) {
595 ssize_t ret = real_write_file(fsp, data, pos, n);
596 if (ret == -1)
597 return -1;
599 if (pos + ret > wcp->file_size) {
600 wcp->file_size = pos + n;
601 fsp->size = (SMB_BIG_UINT)wcp->file_size;
604 DO_PROFILE_INC(writecache_direct_writes);
605 return total_written + n;
609 * If there's any data left, cache it.
612 if (n) {
613 #ifdef WITH_PROFILE
614 if (wcp->data_size) {
615 DO_PROFILE_INC(writecache_abutted_writes);
616 } else {
617 DO_PROFILE_INC(writecache_init_writes);
619 #endif
620 memcpy(wcp->data+wcp->data_size, data, n);
621 if (wcp->data_size == 0) {
622 wcp->offset = pos;
623 DO_PROFILE_INC(writecache_num_write_caches);
625 wcp->data_size += n;
628 * Update the file size if changed.
631 if (wcp->offset + wcp->data_size > wcp->file_size) {
632 wcp->file_size = wcp->offset + wcp->data_size;
633 fsp->size = (SMB_BIG_UINT)wcp->file_size;
635 DEBUG(9,("wcp->offset = %.0f wcp->data_size = %u cache return %u\n",
636 (double)wcp->offset, (unsigned int)wcp->data_size, (unsigned int)n));
638 total_written += n;
639 return total_written; /* .... that's a write :) */
642 return total_written;
645 /****************************************************************************
646 Delete the write cache structure.
647 ****************************************************************************/
649 void delete_write_cache(files_struct *fsp)
651 write_cache *wcp;
653 if(!fsp)
654 return;
656 if(!(wcp = fsp->wcp))
657 return;
659 DO_PROFILE_DEC(writecache_allocated_write_caches);
660 allocated_write_caches--;
662 SMB_ASSERT(wcp->data_size == 0);
664 SAFE_FREE(wcp->data);
665 SAFE_FREE(fsp->wcp);
667 DEBUG(10,("delete_write_cache: File %s deleted write cache\n", fsp->fsp_name ));
670 /****************************************************************************
671 Setup the write cache structure.
672 ****************************************************************************/
674 static BOOL setup_write_cache(files_struct *fsp, SMB_OFF_T file_size)
676 ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
677 write_cache *wcp;
679 if (allocated_write_caches >= MAX_WRITE_CACHES)
680 return False;
682 if(alloc_size == 0 || fsp->wcp)
683 return False;
685 if((wcp = SMB_MALLOC_P(write_cache)) == NULL) {
686 DEBUG(0,("setup_write_cache: malloc fail.\n"));
687 return False;
690 wcp->file_size = file_size;
691 wcp->offset = 0;
692 wcp->alloc_size = alloc_size;
693 wcp->data_size = 0;
694 if((wcp->data = SMB_MALLOC(wcp->alloc_size)) == NULL) {
695 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
696 (unsigned int)wcp->alloc_size ));
697 SAFE_FREE(wcp);
698 return False;
701 memset(wcp->data, '\0', wcp->alloc_size );
703 fsp->wcp = wcp;
704 DO_PROFILE_INC(writecache_allocated_write_caches);
705 allocated_write_caches++;
707 DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
708 fsp->fsp_name, (unsigned long)wcp->alloc_size ));
710 return True;
713 /****************************************************************************
714 Cope with a size change.
715 ****************************************************************************/
717 void set_filelen_write_cache(files_struct *fsp, SMB_OFF_T file_size)
719 fsp->size = (SMB_BIG_UINT)file_size;
720 if(fsp->wcp) {
721 /* The cache *must* have been flushed before we do this. */
722 if (fsp->wcp->data_size != 0) {
723 pstring msg;
724 slprintf(msg, sizeof(msg)-1, "set_filelen_write_cache: size change \
725 on file %s with write cache size = %lu\n", fsp->fsp_name, (unsigned long)fsp->wcp->data_size );
726 smb_panic(msg);
728 fsp->wcp->file_size = file_size;
732 /*******************************************************************
733 Flush a write cache struct to disk.
734 ********************************************************************/
736 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
738 write_cache *wcp = fsp->wcp;
739 size_t data_size;
740 ssize_t ret;
742 if(!wcp || !wcp->data_size)
743 return 0;
745 data_size = wcp->data_size;
746 wcp->data_size = 0;
748 DO_PROFILE_DEC_INC(writecache_num_write_caches,writecache_flushed_writes[reason]);
750 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
751 fsp->fd, (double)wcp->offset, (unsigned int)data_size));
753 #ifdef WITH_PROFILE
754 if(data_size == wcp->alloc_size)
755 DO_PROFILE_INC(writecache_num_perfect_writes);
756 #endif
758 ret = real_write_file(fsp, wcp->data, wcp->offset, data_size);
761 * Ensure file size if kept up to date if write extends file.
764 if ((ret != -1) && (wcp->offset + ret > wcp->file_size))
765 wcp->file_size = wcp->offset + ret;
767 return ret;
770 /*******************************************************************
771 sync a file
772 ********************************************************************/
774 void sync_file(connection_struct *conn, files_struct *fsp)
776 if(lp_strict_sync(SNUM(conn)) && fsp->fd != -1) {
777 flush_write_cache(fsp, SYNC_FLUSH);
778 SMB_VFS_FSYNC(fsp,fsp->fd);
783 /************************************************************
784 Perform a stat whether a valid fd or not.
785 ************************************************************/
787 int fsp_stat(files_struct *fsp, SMB_STRUCT_STAT *pst)
789 if (fsp->fd == -1)
790 return SMB_VFS_STAT(fsp->conn, fsp->fsp_name, pst);
791 else
792 return SMB_VFS_FSTAT(fsp,fsp->fd, pst);