2 Unix SMB/Netbios implementation.
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 3 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, see <http://www.gnu.org/licenses/>.
24 static bool setup_write_cache(files_struct
*, SMB_OFF_T
);
26 /****************************************************************************
27 Read from write cache if we can.
28 ****************************************************************************/
30 static bool read_from_write_cache(files_struct
*fsp
,char *data
,SMB_OFF_T pos
,size_t n
)
32 write_cache
*wcp
= fsp
->wcp
;
38 if( n
> wcp
->data_size
|| pos
< wcp
->offset
|| pos
+ n
> wcp
->offset
+ wcp
->data_size
) {
42 memcpy(data
, wcp
->data
+ (pos
- wcp
->offset
), n
);
44 DO_PROFILE_INC(writecache_read_hits
);
49 /****************************************************************************
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
) {
63 * Serve from write cache if we can.
66 if(read_from_write_cache(fsp
, data
, pos
, n
)) {
67 fsp
->fh
->pos
= pos
+ n
;
68 fsp
->fh
->position_information
= fsp
->fh
->pos
;
72 flush_write_cache(fsp
, READ_FLUSH
);
80 readret
= SMB_VFS_PREAD(fsp
,data
,n
,pos
);
83 if ((errno
== EAGAIN
) && numretries
) {
84 DEBUG(3,("read_file EAGAIN retry in 10 seconds\n"));
91 #else /* NO DMF fix. */
92 readret
= SMB_VFS_PREAD(fsp
,data
,n
,pos
);
103 DEBUG(10,("read_file (%s): pos = %.0f, size = %lu, returned %lu\n",
104 fsp
->fsp_name
, (double)pos
, (unsigned long)n
, (long)ret
));
107 fsp
->fh
->position_information
= fsp
->fh
->pos
;
112 /* how many write cache buffers have been allocated */
113 static unsigned int allocated_write_caches
;
115 /****************************************************************************
116 *Really* write to a file.
117 ****************************************************************************/
119 static ssize_t
real_write_file(struct smb_request
*req
,
128 ret
= vfs_write_data(req
, fsp
, data
, n
);
131 if (pos
&& lp_strict_allocate(SNUM(fsp
->conn
))) {
132 if (vfs_fill_sparse(fsp
, pos
) == -1) {
136 ret
= vfs_pwrite_data(req
, fsp
, data
, n
, pos
);
139 DEBUG(10,("real_write_file (%s): pos = %.0f, size = %lu, returned %ld\n",
140 fsp
->fsp_name
, (double)pos
, (unsigned long)n
, (long)ret
));
145 /* Yes - this is correct - writes don't update this. JRA. */
146 /* Found by Samba4 tests. */
148 fsp
->position_information
= fsp
->pos
;
155 /****************************************************************************
156 File size cache change.
157 Updates size on disk but doesn't flush the cache.
158 ****************************************************************************/
160 static int wcp_file_size_change(files_struct
*fsp
)
163 write_cache
*wcp
= fsp
->wcp
;
165 wcp
->file_size
= wcp
->offset
+ wcp
->data_size
;
166 ret
= SMB_VFS_FTRUNCATE(fsp
, wcp
->file_size
);
168 DEBUG(0,("wcp_file_size_change (%s): ftruncate of size %.0f error %s\n",
169 fsp
->fsp_name
, (double)wcp
->file_size
, strerror(errno
) ));
174 static void update_write_time_handler(struct event_context
*ctx
,
175 struct timed_event
*te
,
176 const struct timeval
*now
,
179 files_struct
*fsp
= (files_struct
*)private_data
;
181 /* Remove the timed event handler. */
182 TALLOC_FREE(fsp
->update_write_time_event
);
183 DEBUG(5, ("Update write time on %s\n", fsp
->fsp_name
));
185 /* change the write time if not already changed by someone else */
186 update_write_time(fsp
);
189 /*********************************************************
190 Schedule a write time update for WRITE_TIME_UPDATE_USEC_DELAY
192 *********************************************************/
194 void trigger_write_time_update(struct files_struct
*fsp
)
198 if (fsp
->write_time_forced
) {
199 /* No point - "sticky" write times
205 if (fsp
->update_write_time_triggered
) {
207 * We only update the write time
208 * on the first write. After that
209 * no other writes affect this.
213 fsp
->update_write_time_triggered
= true;
215 delay
= lp_parm_int(SNUM(fsp
->conn
),
216 "smbd", "writetimeupdatedelay",
217 WRITE_TIME_UPDATE_USEC_DELAY
);
219 /* trigger the update 2 seconds later */
220 fsp
->update_write_time_on_close
= true;
221 fsp
->update_write_time_event
=
222 event_add_timed(smbd_event_context(), NULL
,
223 timeval_current_ofs(0, delay
),
224 "update_write_time_handler",
225 update_write_time_handler
, fsp
);
228 void trigger_write_time_update_immediate(struct files_struct
*fsp
)
230 if (fsp
->write_time_forced
) {
232 * No point - "sticky" write times
238 TALLOC_FREE(fsp
->update_write_time_event
);
239 DEBUG(5, ("Update write time immediate on %s\n", fsp
->fsp_name
));
241 fsp
->update_write_time_triggered
= true;
243 fsp
->update_write_time_on_close
= false;
244 update_write_time(fsp
);
247 /****************************************************************************
249 ****************************************************************************/
251 ssize_t
write_file(struct smb_request
*req
,
257 write_cache
*wcp
= fsp
->wcp
;
258 ssize_t total_written
= 0;
261 if (fsp
->print_file
) {
265 if (!rap_to_pjobid(fsp
->rap_print_jobid
, sharename
, &jobid
)) {
266 DEBUG(3,("write_file: Unable to map RAP jobid %u to jobid.\n",
267 (unsigned int)fsp
->rap_print_jobid
));
272 return print_job_write(SNUM(fsp
->conn
), jobid
, data
, pos
, n
);
275 if (!fsp
->can_write
) {
280 if (!fsp
->modified
) {
282 fsp
->modified
= True
;
284 if (SMB_VFS_FSTAT(fsp
, &st
) == 0) {
286 trigger_write_time_update(fsp
);
287 dosmode
= dos_mode(fsp
->conn
,fsp
->fsp_name
,&st
);
288 if ((lp_store_dos_attributes(SNUM(fsp
->conn
)) ||
289 MAP_ARCHIVE(fsp
->conn
)) &&
290 !IS_DOS_ARCHIVE(dosmode
)) {
291 file_set_dosmode(fsp
->conn
,fsp
->fsp_name
,
298 * If this is the first write and we have an exclusive oplock then setup
302 if (EXCLUSIVE_OPLOCK_TYPE(fsp
->oplock_type
) && !wcp
) {
303 setup_write_cache(fsp
, st
.st_size
);
310 DO_PROFILE_INC(writecache_total_writes
);
311 if (!fsp
->oplock_type
) {
312 DO_PROFILE_INC(writecache_non_oplock_writes
);
317 * If this file is level II oplocked then we need
318 * to grab the shared memory lock and inform all
319 * other files with a level II lock that they need
320 * to flush their read caches. We keep the lock over
321 * the shared memory area whilst doing this.
324 release_level_2_oplocks_on_change(fsp
);
327 if (profile_p
&& profile_p
->writecache_total_writes
% 500 == 0) {
328 DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u total=%u \
329 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
330 profile_p
->writecache_init_writes
,
331 profile_p
->writecache_abutted_writes
,
332 profile_p
->writecache_total_writes
,
333 profile_p
->writecache_non_oplock_writes
,
334 profile_p
->writecache_allocated_write_caches
,
335 profile_p
->writecache_num_write_caches
,
336 profile_p
->writecache_direct_writes
,
337 profile_p
->writecache_num_perfect_writes
,
338 profile_p
->writecache_read_hits
));
340 DEBUG(3,("WRITECACHE: Flushes SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
341 profile_p
->writecache_flushed_writes
[SEEK_FLUSH
],
342 profile_p
->writecache_flushed_writes
[READ_FLUSH
],
343 profile_p
->writecache_flushed_writes
[WRITE_FLUSH
],
344 profile_p
->writecache_flushed_writes
[READRAW_FLUSH
],
345 profile_p
->writecache_flushed_writes
[OPLOCK_RELEASE_FLUSH
],
346 profile_p
->writecache_flushed_writes
[CLOSE_FLUSH
],
347 profile_p
->writecache_flushed_writes
[SYNC_FLUSH
] ));
351 if (wcp
&& req
->unread_bytes
) {
352 /* If we're using receivefile don't
353 * deal with a write cache.
355 flush_write_cache(fsp
, WRITE_FLUSH
);
356 delete_write_cache(fsp
);
361 DO_PROFILE_INC(writecache_direct_writes
);
362 total_written
= real_write_file(req
, fsp
, data
, pos
, n
);
363 return total_written
;
366 DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f wcp->data_size=%u\n",
367 fsp
->fsp_name
, fsp
->fh
->fd
, (double)pos
, (unsigned int)n
, (double)wcp
->offset
, (unsigned int)wcp
->data_size
));
369 fsp
->fh
->pos
= pos
+ n
;
372 * If we have active cache and it isn't contiguous then we flush.
373 * NOTE: There is a small problem with running out of disk ....
376 if (wcp
->data_size
) {
377 bool cache_flush_needed
= False
;
379 if ((pos
>= wcp
->offset
) && (pos
<= wcp
->offset
+ wcp
->data_size
)) {
381 /* ASCII art.... JRA.
383 +--------------+-----
384 | Cached data | Rest of allocated cache buffer....
385 +--------------+-----
387 +-------------------+
389 +-------------------+
394 * Start of write overlaps or abutts the existing data.
397 size_t data_used
= MIN((wcp
->alloc_size
- (pos
- wcp
->offset
)), n
);
399 memcpy(wcp
->data
+ (pos
- wcp
->offset
), data
, data_used
);
402 * Update the current buffer size with the new data.
405 if(pos
+ data_used
> wcp
->offset
+ wcp
->data_size
) {
406 wcp
->data_size
= pos
+ data_used
- wcp
->offset
;
410 * Update the file size if changed.
413 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
414 if (wcp_file_size_change(fsp
) == -1) {
420 * If we used all the data then
427 cache_flush_needed
= True
;
430 * Move the start of data forward by the amount used,
431 * cut down the amount left by the same amount.
438 DO_PROFILE_INC(writecache_abutted_writes
);
439 total_written
= data_used
;
443 } else if ((pos
< wcp
->offset
) && (pos
+ n
> wcp
->offset
) &&
444 (pos
+ n
<= wcp
->offset
+ wcp
->alloc_size
)) {
446 /* ASCII art.... JRA.
452 +-------------------+
454 +-------------------+
459 * End of write overlaps the existing data.
462 size_t data_used
= pos
+ n
- wcp
->offset
;
464 memcpy(wcp
->data
, data
+ n
- data_used
, data_used
);
467 * Update the current buffer size with the new data.
470 if(pos
+ n
> wcp
->offset
+ wcp
->data_size
) {
471 wcp
->data_size
= pos
+ n
- wcp
->offset
;
475 * Update the file size if changed.
478 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
479 if (wcp_file_size_change(fsp
) == -1) {
485 * We don't need to move the start of data, but we
486 * cut down the amount left by the amount used.
492 * We cannot have used all the data here.
495 cache_flush_needed
= True
;
497 DO_PROFILE_INC(writecache_abutted_writes
);
498 total_written
= data_used
;
502 } else if ( (pos
>= wcp
->file_size
) &&
503 (wcp
->offset
+ wcp
->data_size
== wcp
->file_size
) &&
504 (pos
> wcp
->offset
+ wcp
->data_size
) &&
505 (pos
< wcp
->offset
+ wcp
->alloc_size
) ) {
507 /* ASCII art.... JRA.
511 +---------------+---------------+
512 | Cached data | Cache buffer |
513 +---------------+---------------+
515 +-------------------+
517 +-------------------+
522 * Non-contiguous write part of which fits within
523 * the cache buffer and is extending the file
524 * and the cache contents reflect the current
525 * data up to the current end of the file.
530 if(pos
+ n
<= wcp
->offset
+ wcp
->alloc_size
) {
533 data_used
= wcp
->offset
+ wcp
->alloc_size
- pos
;
537 * Fill in the non-continuous area with zeros.
540 memset(wcp
->data
+ wcp
->data_size
, '\0',
541 pos
- (wcp
->offset
+ wcp
->data_size
) );
543 memcpy(wcp
->data
+ (pos
- wcp
->offset
), data
, data_used
);
546 * Update the current buffer size with the new data.
549 if(pos
+ data_used
> wcp
->offset
+ wcp
->data_size
) {
550 wcp
->data_size
= pos
+ data_used
- wcp
->offset
;
554 * Update the file size if changed.
557 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
558 if (wcp_file_size_change(fsp
) == -1) {
564 * If we used all the data then
571 cache_flush_needed
= True
;
575 * Move the start of data forward by the amount used,
576 * cut down the amount left by the same amount.
583 DO_PROFILE_INC(writecache_abutted_writes
);
584 total_written
= data_used
;
588 } else if ( (pos
>= wcp
->file_size
) &&
590 (wcp
->file_size
== wcp
->offset
+ wcp
->data_size
) &&
591 (pos
< wcp
->file_size
+ wcp
->alloc_size
)) {
597 +---------------+---------------+
598 | Cached data | Cache buffer |
599 +---------------+---------------+
601 |<------- allocated size ---------------->|
607 MS-Office seems to do this a lot to determine if there's enough
608 space on the filesystem to write a new file.
613 +-----------------------+--------+
614 | Zeroed Cached data | 1 Byte |
615 +-----------------------+--------+
618 flush_write_cache(fsp
, WRITE_FLUSH
);
619 wcp
->offset
= wcp
->file_size
;
620 wcp
->data_size
= pos
- wcp
->file_size
+ 1;
621 memset(wcp
->data
, '\0', wcp
->data_size
);
622 memcpy(wcp
->data
+ wcp
->data_size
-1, data
, 1);
625 * Update the file size if changed.
628 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
629 if (wcp_file_size_change(fsp
) == -1) {
638 /* ASCII art..... JRA.
642 +---------------+---------------+
643 | Cached data | Cache buffer |
644 +---------------+---------------+
646 +-------------------+
648 +-------------------+
652 +---------------+---------------+
653 | Cached data | Cache buffer |
654 +---------------+---------------+
656 +-------------------+
658 +-------------------+
662 +---------------+---------------+
663 | Cached data | Cache buffer |
664 +---------------+---------------+
666 +-----------------------------------------------------+
668 +-----------------------------------------------------+
673 * Write is bigger than buffer, or there is no overlap on the
677 DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
678 len = %u\n",fsp
->fh
->fd
, (double)pos
, (unsigned int)n
, (double)wcp
->offset
, (unsigned int)wcp
->data_size
));
681 * If write would fit in the cache, and is larger than
682 * the data already in the cache, flush the cache and
683 * preferentially copy the data new data into it. Otherwise
684 * just write the data directly.
687 if ( n
<= wcp
->alloc_size
&& n
> wcp
->data_size
) {
688 cache_flush_needed
= True
;
690 ssize_t ret
= real_write_file(NULL
,fsp
, data
, pos
, n
);
693 * If the write overlaps the entire cache, then
694 * discard the current contents of the cache.
695 * Fix from Rasmus Borup Hansen rbh@math.ku.dk.
698 if ((pos
<= wcp
->offset
) &&
699 (pos
+ n
>= wcp
->offset
+ wcp
->data_size
) ) {
700 DEBUG(9,("write_file: discarding overwritten write \
701 cache: fd = %d, off=%.0f, size=%u\n", fsp
->fh
->fd
, (double)wcp
->offset
, (unsigned int)wcp
->data_size
));
705 DO_PROFILE_INC(writecache_direct_writes
);
710 if (pos
+ ret
> wcp
->file_size
) {
711 wcp
->file_size
= pos
+ ret
;
721 if (cache_flush_needed
) {
722 DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
723 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
724 write_path
, fsp
->fh
->fd
, (double)wcp
->file_size
, (double)pos
, (unsigned int)n
,
725 (double)wcp
->offset
, (unsigned int)wcp
->data_size
));
727 flush_write_cache(fsp
, WRITE_FLUSH
);
732 * If the write request is bigger than the cache
733 * size, write it all out.
736 if (n
> wcp
->alloc_size
) {
737 ssize_t ret
= real_write_file(NULL
,fsp
, data
, pos
, n
);
742 if (pos
+ ret
> wcp
->file_size
) {
743 wcp
->file_size
= pos
+ n
;
746 DO_PROFILE_INC(writecache_direct_writes
);
747 return total_written
+ n
;
751 * If there's any data left, cache it.
756 if (wcp
->data_size
) {
757 DO_PROFILE_INC(writecache_abutted_writes
);
759 DO_PROFILE_INC(writecache_init_writes
);
762 memcpy(wcp
->data
+wcp
->data_size
, data
, n
);
763 if (wcp
->data_size
== 0) {
765 DO_PROFILE_INC(writecache_num_write_caches
);
770 * Update the file size if changed.
773 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
774 if (wcp_file_size_change(fsp
) == -1) {
778 DEBUG(9,("wcp->offset = %.0f wcp->data_size = %u cache return %u\n",
779 (double)wcp
->offset
, (unsigned int)wcp
->data_size
, (unsigned int)n
));
782 return total_written
; /* .... that's a write :) */
785 return total_written
;
788 /****************************************************************************
789 Delete the write cache structure.
790 ****************************************************************************/
792 void delete_write_cache(files_struct
*fsp
)
800 if(!(wcp
= fsp
->wcp
)) {
804 DO_PROFILE_DEC(writecache_allocated_write_caches
);
805 allocated_write_caches
--;
807 SMB_ASSERT(wcp
->data_size
== 0);
809 SAFE_FREE(wcp
->data
);
812 DEBUG(10,("delete_write_cache: File %s deleted write cache\n", fsp
->fsp_name
));
815 /****************************************************************************
816 Setup the write cache structure.
817 ****************************************************************************/
819 static bool setup_write_cache(files_struct
*fsp
, SMB_OFF_T file_size
)
821 ssize_t alloc_size
= lp_write_cache_size(SNUM(fsp
->conn
));
824 if (allocated_write_caches
>= MAX_WRITE_CACHES
) {
828 if(alloc_size
== 0 || fsp
->wcp
) {
832 if((wcp
= SMB_MALLOC_P(write_cache
)) == NULL
) {
833 DEBUG(0,("setup_write_cache: malloc fail.\n"));
837 wcp
->file_size
= file_size
;
839 wcp
->alloc_size
= alloc_size
;
841 if((wcp
->data
= (char *)SMB_MALLOC(wcp
->alloc_size
)) == NULL
) {
842 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
843 (unsigned int)wcp
->alloc_size
));
848 memset(wcp
->data
, '\0', wcp
->alloc_size
);
851 DO_PROFILE_INC(writecache_allocated_write_caches
);
852 allocated_write_caches
++;
854 DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
855 fsp
->fsp_name
, (unsigned long)wcp
->alloc_size
));
860 /****************************************************************************
861 Cope with a size change.
862 ****************************************************************************/
864 void set_filelen_write_cache(files_struct
*fsp
, SMB_OFF_T file_size
)
867 /* The cache *must* have been flushed before we do this. */
868 if (fsp
->wcp
->data_size
!= 0) {
870 asprintf(&msg
, "set_filelen_write_cache: size change "
871 "on file %s with write cache size = %lu\n",
873 (unsigned long)fsp
->wcp
->data_size
);
876 fsp
->wcp
->file_size
= file_size
;
880 /*******************************************************************
881 Flush a write cache struct to disk.
882 ********************************************************************/
884 ssize_t
flush_write_cache(files_struct
*fsp
, enum flush_reason_enum reason
)
886 write_cache
*wcp
= fsp
->wcp
;
890 if(!wcp
|| !wcp
->data_size
) {
894 data_size
= wcp
->data_size
;
897 DO_PROFILE_DEC_INC(writecache_num_write_caches
,writecache_flushed_writes
[reason
]);
899 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
900 fsp
->fh
->fd
, (double)wcp
->offset
, (unsigned int)data_size
));
903 if(data_size
== wcp
->alloc_size
) {
904 DO_PROFILE_INC(writecache_num_perfect_writes
);
908 ret
= real_write_file(NULL
, fsp
, wcp
->data
, wcp
->offset
, data_size
);
911 * Ensure file size if kept up to date if write extends file.
914 if ((ret
!= -1) && (wcp
->offset
+ ret
> wcp
->file_size
)) {
915 wcp
->file_size
= wcp
->offset
+ ret
;
921 /*******************************************************************
923 ********************************************************************/
925 NTSTATUS
sync_file(connection_struct
*conn
, files_struct
*fsp
, bool write_through
)
927 if (fsp
->fh
->fd
== -1)
928 return NT_STATUS_INVALID_HANDLE
;
930 if (lp_strict_sync(SNUM(conn
)) &&
931 (lp_syncalways(SNUM(conn
)) || write_through
)) {
932 int ret
= flush_write_cache(fsp
, SYNC_FLUSH
);
934 return map_nt_error_from_unix(errno
);
936 ret
= SMB_VFS_FSYNC(fsp
);
938 return map_nt_error_from_unix(errno
);
944 /************************************************************
945 Perform a stat whether a valid fd or not.
946 ************************************************************/
948 int fsp_stat(files_struct
*fsp
, SMB_STRUCT_STAT
*pst
)
950 if (fsp
->fh
->fd
== -1) {
951 return SMB_VFS_STAT(fsp
->conn
, fsp
->fsp_name
, pst
);
953 return SMB_VFS_FSTAT(fsp
, pst
);