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/>.
23 #include "smbd/globals.h"
25 static bool setup_write_cache(files_struct
*, SMB_OFF_T
);
27 /****************************************************************************
28 Read from write cache if we can.
29 ****************************************************************************/
31 static bool read_from_write_cache(files_struct
*fsp
,char *data
,SMB_OFF_T pos
,size_t n
)
33 write_cache
*wcp
= fsp
->wcp
;
39 if( n
> wcp
->data_size
|| pos
< wcp
->offset
|| pos
+ n
> wcp
->offset
+ wcp
->data_size
) {
43 memcpy(data
, wcp
->data
+ (pos
- wcp
->offset
), n
);
45 DO_PROFILE_INC(writecache_read_hits
);
50 /****************************************************************************
52 ****************************************************************************/
54 ssize_t
read_file(files_struct
*fsp
,char *data
,SMB_OFF_T pos
,size_t n
)
56 ssize_t ret
=0,readret
;
58 /* you can't read from print files */
59 if (fsp
->print_file
) {
64 * Serve from write cache if we can.
67 if(read_from_write_cache(fsp
, data
, pos
, n
)) {
68 fsp
->fh
->pos
= pos
+ n
;
69 fsp
->fh
->position_information
= fsp
->fh
->pos
;
73 flush_write_cache(fsp
, READ_FLUSH
);
81 readret
= SMB_VFS_PREAD(fsp
,data
,n
,pos
);
84 if ((errno
== EAGAIN
) && numretries
) {
85 DEBUG(3,("read_file EAGAIN retry in 10 seconds\n"));
92 #else /* NO DMF fix. */
93 readret
= SMB_VFS_PREAD(fsp
,data
,n
,pos
);
104 DEBUG(10,("read_file (%s): pos = %.0f, size = %lu, returned %lu\n",
105 fsp
->fsp_name
, (double)pos
, (unsigned long)n
, (long)ret
));
108 fsp
->fh
->position_information
= fsp
->fh
->pos
;
113 /****************************************************************************
114 *Really* write to a file.
115 ****************************************************************************/
117 static ssize_t
real_write_file(struct smb_request
*req
,
126 ret
= vfs_write_data(req
, fsp
, data
, n
);
129 if (pos
&& lp_strict_allocate(SNUM(fsp
->conn
))) {
130 if (vfs_fill_sparse(fsp
, pos
) == -1) {
134 ret
= vfs_pwrite_data(req
, fsp
, data
, n
, pos
);
137 DEBUG(10,("real_write_file (%s): pos = %.0f, size = %lu, returned %ld\n",
138 fsp
->fsp_name
, (double)pos
, (unsigned long)n
, (long)ret
));
143 /* Yes - this is correct - writes don't update this. JRA. */
144 /* Found by Samba4 tests. */
146 fsp
->position_information
= fsp
->pos
;
153 /****************************************************************************
154 File size cache change.
155 Updates size on disk but doesn't flush the cache.
156 ****************************************************************************/
158 static int wcp_file_size_change(files_struct
*fsp
)
161 write_cache
*wcp
= fsp
->wcp
;
163 wcp
->file_size
= wcp
->offset
+ wcp
->data_size
;
164 ret
= SMB_VFS_FTRUNCATE(fsp
, wcp
->file_size
);
166 DEBUG(0,("wcp_file_size_change (%s): ftruncate of size %.0f error %s\n",
167 fsp
->fsp_name
, (double)wcp
->file_size
, strerror(errno
) ));
172 static void update_write_time_handler(struct event_context
*ctx
,
173 struct timed_event
*te
,
177 files_struct
*fsp
= (files_struct
*)private_data
;
179 /* Remove the timed event handler. */
180 TALLOC_FREE(fsp
->update_write_time_event
);
181 DEBUG(5, ("Update write time on %s\n", fsp
->fsp_name
));
183 /* change the write time if not already changed by someone else */
184 update_write_time(fsp
);
187 /*********************************************************
188 Schedule a write time update for WRITE_TIME_UPDATE_USEC_DELAY
190 *********************************************************/
192 void trigger_write_time_update(struct files_struct
*fsp
)
196 if (fsp
->write_time_forced
) {
197 /* No point - "sticky" write times
203 if (fsp
->update_write_time_triggered
) {
205 * We only update the write time
206 * on the first write. After that
207 * no other writes affect this.
211 fsp
->update_write_time_triggered
= true;
213 delay
= lp_parm_int(SNUM(fsp
->conn
),
214 "smbd", "writetimeupdatedelay",
215 WRITE_TIME_UPDATE_USEC_DELAY
);
217 /* trigger the update 2 seconds later */
218 fsp
->update_write_time_on_close
= true;
219 fsp
->update_write_time_event
=
220 event_add_timed(smbd_event_context(), NULL
,
221 timeval_current_ofs(0, delay
),
222 update_write_time_handler
, fsp
);
225 void trigger_write_time_update_immediate(struct files_struct
*fsp
)
227 if (fsp
->write_time_forced
) {
229 * No point - "sticky" write times
235 TALLOC_FREE(fsp
->update_write_time_event
);
236 DEBUG(5, ("Update write time immediate on %s\n", fsp
->fsp_name
));
238 fsp
->update_write_time_triggered
= true;
240 fsp
->update_write_time_on_close
= false;
241 update_write_time(fsp
);
244 /****************************************************************************
246 ****************************************************************************/
248 ssize_t
write_file(struct smb_request
*req
,
254 write_cache
*wcp
= fsp
->wcp
;
255 ssize_t total_written
= 0;
258 if (fsp
->print_file
) {
262 if (!rap_to_pjobid(fsp
->rap_print_jobid
, sharename
, &jobid
)) {
263 DEBUG(3,("write_file: Unable to map RAP jobid %u to jobid.\n",
264 (unsigned int)fsp
->rap_print_jobid
));
269 return print_job_write(SNUM(fsp
->conn
), jobid
, data
, pos
, n
);
272 if (!fsp
->can_write
) {
277 if (!fsp
->modified
) {
279 fsp
->modified
= True
;
281 if (SMB_VFS_FSTAT(fsp
, &st
) == 0) {
283 trigger_write_time_update(fsp
);
284 dosmode
= dos_mode(fsp
->conn
,fsp
->fsp_name
,&st
);
285 if ((lp_store_dos_attributes(SNUM(fsp
->conn
)) ||
286 MAP_ARCHIVE(fsp
->conn
)) &&
287 !IS_DOS_ARCHIVE(dosmode
)) {
288 file_set_dosmode(fsp
->conn
,fsp
->fsp_name
,
295 * If this is the first write and we have an exclusive oplock then setup
299 if (EXCLUSIVE_OPLOCK_TYPE(fsp
->oplock_type
) && !wcp
) {
300 setup_write_cache(fsp
, st
.st_size
);
307 DO_PROFILE_INC(writecache_total_writes
);
308 if (!fsp
->oplock_type
) {
309 DO_PROFILE_INC(writecache_non_oplock_writes
);
314 * If this file is level II oplocked then we need
315 * to grab the shared memory lock and inform all
316 * other files with a level II lock that they need
317 * to flush their read caches. We keep the lock over
318 * the shared memory area whilst doing this.
321 release_level_2_oplocks_on_change(fsp
);
324 if (profile_p
&& profile_p
->writecache_total_writes
% 500 == 0) {
325 DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u total=%u \
326 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
327 profile_p
->writecache_init_writes
,
328 profile_p
->writecache_abutted_writes
,
329 profile_p
->writecache_total_writes
,
330 profile_p
->writecache_non_oplock_writes
,
331 profile_p
->writecache_allocated_write_caches
,
332 profile_p
->writecache_num_write_caches
,
333 profile_p
->writecache_direct_writes
,
334 profile_p
->writecache_num_perfect_writes
,
335 profile_p
->writecache_read_hits
));
337 DEBUG(3,("WRITECACHE: Flushes SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
338 profile_p
->writecache_flushed_writes
[SEEK_FLUSH
],
339 profile_p
->writecache_flushed_writes
[READ_FLUSH
],
340 profile_p
->writecache_flushed_writes
[WRITE_FLUSH
],
341 profile_p
->writecache_flushed_writes
[READRAW_FLUSH
],
342 profile_p
->writecache_flushed_writes
[OPLOCK_RELEASE_FLUSH
],
343 profile_p
->writecache_flushed_writes
[CLOSE_FLUSH
],
344 profile_p
->writecache_flushed_writes
[SYNC_FLUSH
] ));
348 if (wcp
&& req
->unread_bytes
) {
349 /* If we're using receivefile don't
350 * deal with a write cache.
352 flush_write_cache(fsp
, WRITE_FLUSH
);
353 delete_write_cache(fsp
);
358 DO_PROFILE_INC(writecache_direct_writes
);
359 total_written
= real_write_file(req
, fsp
, data
, pos
, n
);
360 return total_written
;
363 DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f wcp->data_size=%u\n",
364 fsp
->fsp_name
, fsp
->fh
->fd
, (double)pos
, (unsigned int)n
, (double)wcp
->offset
, (unsigned int)wcp
->data_size
));
366 fsp
->fh
->pos
= pos
+ n
;
369 * If we have active cache and it isn't contiguous then we flush.
370 * NOTE: There is a small problem with running out of disk ....
373 if (wcp
->data_size
) {
374 bool cache_flush_needed
= False
;
376 if ((pos
>= wcp
->offset
) && (pos
<= wcp
->offset
+ wcp
->data_size
)) {
378 /* ASCII art.... JRA.
380 +--------------+-----
381 | Cached data | Rest of allocated cache buffer....
382 +--------------+-----
384 +-------------------+
386 +-------------------+
391 * Start of write overlaps or abutts the existing data.
394 size_t data_used
= MIN((wcp
->alloc_size
- (pos
- wcp
->offset
)), n
);
396 memcpy(wcp
->data
+ (pos
- wcp
->offset
), data
, data_used
);
399 * Update the current buffer size with the new data.
402 if(pos
+ data_used
> wcp
->offset
+ wcp
->data_size
) {
403 wcp
->data_size
= pos
+ data_used
- wcp
->offset
;
407 * Update the file size if changed.
410 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
411 if (wcp_file_size_change(fsp
) == -1) {
417 * If we used all the data then
424 cache_flush_needed
= True
;
427 * Move the start of data forward by the amount used,
428 * cut down the amount left by the same amount.
435 DO_PROFILE_INC(writecache_abutted_writes
);
436 total_written
= data_used
;
440 } else if ((pos
< wcp
->offset
) && (pos
+ n
> wcp
->offset
) &&
441 (pos
+ n
<= wcp
->offset
+ wcp
->alloc_size
)) {
443 /* ASCII art.... JRA.
449 +-------------------+
451 +-------------------+
456 * End of write overlaps the existing data.
459 size_t data_used
= pos
+ n
- wcp
->offset
;
461 memcpy(wcp
->data
, data
+ n
- data_used
, data_used
);
464 * Update the current buffer size with the new data.
467 if(pos
+ n
> wcp
->offset
+ wcp
->data_size
) {
468 wcp
->data_size
= pos
+ n
- wcp
->offset
;
472 * Update the file size if changed.
475 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
476 if (wcp_file_size_change(fsp
) == -1) {
482 * We don't need to move the start of data, but we
483 * cut down the amount left by the amount used.
489 * We cannot have used all the data here.
492 cache_flush_needed
= True
;
494 DO_PROFILE_INC(writecache_abutted_writes
);
495 total_written
= data_used
;
499 } else if ( (pos
>= wcp
->file_size
) &&
500 (wcp
->offset
+ wcp
->data_size
== wcp
->file_size
) &&
501 (pos
> wcp
->offset
+ wcp
->data_size
) &&
502 (pos
< wcp
->offset
+ wcp
->alloc_size
) ) {
504 /* ASCII art.... JRA.
508 +---------------+---------------+
509 | Cached data | Cache buffer |
510 +---------------+---------------+
512 +-------------------+
514 +-------------------+
519 * Non-contiguous write part of which fits within
520 * the cache buffer and is extending the file
521 * and the cache contents reflect the current
522 * data up to the current end of the file.
527 if(pos
+ n
<= wcp
->offset
+ wcp
->alloc_size
) {
530 data_used
= wcp
->offset
+ wcp
->alloc_size
- pos
;
534 * Fill in the non-continuous area with zeros.
537 memset(wcp
->data
+ wcp
->data_size
, '\0',
538 pos
- (wcp
->offset
+ wcp
->data_size
) );
540 memcpy(wcp
->data
+ (pos
- wcp
->offset
), data
, data_used
);
543 * Update the current buffer size with the new data.
546 if(pos
+ data_used
> wcp
->offset
+ wcp
->data_size
) {
547 wcp
->data_size
= pos
+ data_used
- wcp
->offset
;
551 * Update the file size if changed.
554 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
555 if (wcp_file_size_change(fsp
) == -1) {
561 * If we used all the data then
568 cache_flush_needed
= True
;
572 * Move the start of data forward by the amount used,
573 * cut down the amount left by the same amount.
580 DO_PROFILE_INC(writecache_abutted_writes
);
581 total_written
= data_used
;
585 } else if ( (pos
>= wcp
->file_size
) &&
587 (wcp
->file_size
== wcp
->offset
+ wcp
->data_size
) &&
588 (pos
< wcp
->file_size
+ wcp
->alloc_size
)) {
594 +---------------+---------------+
595 | Cached data | Cache buffer |
596 +---------------+---------------+
598 |<------- allocated size ---------------->|
604 MS-Office seems to do this a lot to determine if there's enough
605 space on the filesystem to write a new file.
610 +-----------------------+--------+
611 | Zeroed Cached data | 1 Byte |
612 +-----------------------+--------+
615 flush_write_cache(fsp
, WRITE_FLUSH
);
616 wcp
->offset
= wcp
->file_size
;
617 wcp
->data_size
= pos
- wcp
->file_size
+ 1;
618 memset(wcp
->data
, '\0', wcp
->data_size
);
619 memcpy(wcp
->data
+ wcp
->data_size
-1, data
, 1);
622 * Update the file size if changed.
625 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
626 if (wcp_file_size_change(fsp
) == -1) {
635 /* ASCII art..... JRA.
639 +---------------+---------------+
640 | Cached data | Cache buffer |
641 +---------------+---------------+
643 +-------------------+
645 +-------------------+
649 +---------------+---------------+
650 | Cached data | Cache buffer |
651 +---------------+---------------+
653 +-------------------+
655 +-------------------+
659 +---------------+---------------+
660 | Cached data | Cache buffer |
661 +---------------+---------------+
663 +-----------------------------------------------------+
665 +-----------------------------------------------------+
670 * Write is bigger than buffer, or there is no overlap on the
674 DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
675 len = %u\n",fsp
->fh
->fd
, (double)pos
, (unsigned int)n
, (double)wcp
->offset
, (unsigned int)wcp
->data_size
));
678 * If write would fit in the cache, and is larger than
679 * the data already in the cache, flush the cache and
680 * preferentially copy the data new data into it. Otherwise
681 * just write the data directly.
684 if ( n
<= wcp
->alloc_size
&& n
> wcp
->data_size
) {
685 cache_flush_needed
= True
;
687 ssize_t ret
= real_write_file(NULL
,fsp
, data
, pos
, n
);
690 * If the write overlaps the entire cache, then
691 * discard the current contents of the cache.
692 * Fix from Rasmus Borup Hansen rbh@math.ku.dk.
695 if ((pos
<= wcp
->offset
) &&
696 (pos
+ n
>= wcp
->offset
+ wcp
->data_size
) ) {
697 DEBUG(9,("write_file: discarding overwritten write \
698 cache: fd = %d, off=%.0f, size=%u\n", fsp
->fh
->fd
, (double)wcp
->offset
, (unsigned int)wcp
->data_size
));
702 DO_PROFILE_INC(writecache_direct_writes
);
707 if (pos
+ ret
> wcp
->file_size
) {
708 wcp
->file_size
= pos
+ ret
;
718 if (cache_flush_needed
) {
719 DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
720 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
721 write_path
, fsp
->fh
->fd
, (double)wcp
->file_size
, (double)pos
, (unsigned int)n
,
722 (double)wcp
->offset
, (unsigned int)wcp
->data_size
));
724 flush_write_cache(fsp
, WRITE_FLUSH
);
729 * If the write request is bigger than the cache
730 * size, write it all out.
733 if (n
> wcp
->alloc_size
) {
734 ssize_t ret
= real_write_file(NULL
,fsp
, data
, pos
, n
);
739 if (pos
+ ret
> wcp
->file_size
) {
740 wcp
->file_size
= pos
+ n
;
743 DO_PROFILE_INC(writecache_direct_writes
);
744 return total_written
+ n
;
748 * If there's any data left, cache it.
753 if (wcp
->data_size
) {
754 DO_PROFILE_INC(writecache_abutted_writes
);
756 DO_PROFILE_INC(writecache_init_writes
);
759 memcpy(wcp
->data
+wcp
->data_size
, data
, n
);
760 if (wcp
->data_size
== 0) {
762 DO_PROFILE_INC(writecache_num_write_caches
);
767 * Update the file size if changed.
770 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
771 if (wcp_file_size_change(fsp
) == -1) {
775 DEBUG(9,("wcp->offset = %.0f wcp->data_size = %u cache return %u\n",
776 (double)wcp
->offset
, (unsigned int)wcp
->data_size
, (unsigned int)n
));
779 return total_written
; /* .... that's a write :) */
782 return total_written
;
785 /****************************************************************************
786 Delete the write cache structure.
787 ****************************************************************************/
789 void delete_write_cache(files_struct
*fsp
)
797 if(!(wcp
= fsp
->wcp
)) {
801 DO_PROFILE_DEC(writecache_allocated_write_caches
);
802 allocated_write_caches
--;
804 SMB_ASSERT(wcp
->data_size
== 0);
806 SAFE_FREE(wcp
->data
);
809 DEBUG(10,("delete_write_cache: File %s deleted write cache\n", fsp
->fsp_name
));
812 /****************************************************************************
813 Setup the write cache structure.
814 ****************************************************************************/
816 static bool setup_write_cache(files_struct
*fsp
, SMB_OFF_T file_size
)
818 ssize_t alloc_size
= lp_write_cache_size(SNUM(fsp
->conn
));
821 if (allocated_write_caches
>= MAX_WRITE_CACHES
) {
825 if(alloc_size
== 0 || fsp
->wcp
) {
829 if((wcp
= SMB_MALLOC_P(write_cache
)) == NULL
) {
830 DEBUG(0,("setup_write_cache: malloc fail.\n"));
834 wcp
->file_size
= file_size
;
836 wcp
->alloc_size
= alloc_size
;
838 if((wcp
->data
= (char *)SMB_MALLOC(wcp
->alloc_size
)) == NULL
) {
839 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
840 (unsigned int)wcp
->alloc_size
));
845 memset(wcp
->data
, '\0', wcp
->alloc_size
);
848 DO_PROFILE_INC(writecache_allocated_write_caches
);
849 allocated_write_caches
++;
851 DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
852 fsp
->fsp_name
, (unsigned long)wcp
->alloc_size
));
857 /****************************************************************************
858 Cope with a size change.
859 ****************************************************************************/
861 void set_filelen_write_cache(files_struct
*fsp
, SMB_OFF_T file_size
)
864 /* The cache *must* have been flushed before we do this. */
865 if (fsp
->wcp
->data_size
!= 0) {
867 if (asprintf(&msg
, "set_filelen_write_cache: size change "
868 "on file %s with write cache size = %lu\n",
870 (unsigned long)fsp
->wcp
->data_size
) != -1) {
873 smb_panic("set_filelen_write_cache");
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
);