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
) {
65 * Serve from write cache if we can.
68 if(read_from_write_cache(fsp
, data
, pos
, n
)) {
69 fsp
->fh
->pos
= pos
+ n
;
70 fsp
->fh
->position_information
= fsp
->fh
->pos
;
74 flush_write_cache(fsp
, READ_FLUSH
);
82 readret
= SMB_VFS_PREAD(fsp
,data
,n
,pos
);
85 if ((errno
== EAGAIN
) && numretries
) {
86 DEBUG(3,("read_file EAGAIN retry in 10 seconds\n"));
93 #else /* NO DMF fix. */
94 readret
= SMB_VFS_PREAD(fsp
,data
,n
,pos
);
105 DEBUG(10,("read_file (%s): pos = %.0f, size = %lu, returned %lu\n",
106 fsp_str_dbg(fsp
), (double)pos
, (unsigned long)n
, (long)ret
));
109 fsp
->fh
->position_information
= fsp
->fh
->pos
;
114 /****************************************************************************
115 *Really* write to a file.
116 ****************************************************************************/
118 static ssize_t
real_write_file(struct smb_request
*req
,
127 ret
= vfs_write_data(req
, fsp
, data
, n
);
130 if (pos
&& lp_strict_allocate(SNUM(fsp
->conn
))) {
131 if (vfs_fill_sparse(fsp
, pos
) == -1) {
135 ret
= vfs_pwrite_data(req
, fsp
, data
, n
, pos
);
138 DEBUG(10,("real_write_file (%s): pos = %.0f, size = %lu, returned %ld\n",
139 fsp_str_dbg(fsp
), (double)pos
, (unsigned long)n
, (long)ret
));
144 /* Yes - this is correct - writes don't update this. JRA. */
145 /* Found by Samba4 tests. */
147 fsp
->position_information
= fsp
->pos
;
154 /****************************************************************************
155 File size cache change.
156 Updates size on disk but doesn't flush the cache.
157 ****************************************************************************/
159 static int wcp_file_size_change(files_struct
*fsp
)
162 write_cache
*wcp
= fsp
->wcp
;
164 wcp
->file_size
= wcp
->offset
+ wcp
->data_size
;
165 ret
= SMB_VFS_FTRUNCATE(fsp
, wcp
->file_size
);
167 DEBUG(0,("wcp_file_size_change (%s): ftruncate of size %.0f "
168 "error %s\n", fsp_str_dbg(fsp
),
169 (double)wcp
->file_size
, strerror(errno
)));
174 static void update_write_time_handler(struct event_context
*ctx
,
175 struct timed_event
*te
,
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_str_dbg(fsp
)));
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
->posix_open
) {
199 /* Don't use delayed writes on POSIX files. */
203 if (fsp
->write_time_forced
) {
204 /* No point - "sticky" write times
210 if (fsp
->update_write_time_triggered
) {
212 * We only update the write time
213 * on the first write. After that
214 * no other writes affect this.
218 fsp
->update_write_time_triggered
= true;
220 delay
= lp_parm_int(SNUM(fsp
->conn
),
221 "smbd", "writetimeupdatedelay",
222 WRITE_TIME_UPDATE_USEC_DELAY
);
224 /* trigger the update 2 seconds later */
225 fsp
->update_write_time_on_close
= true;
226 fsp
->update_write_time_event
=
227 event_add_timed(smbd_event_context(), NULL
,
228 timeval_current_ofs(0, delay
),
229 update_write_time_handler
, fsp
);
232 void trigger_write_time_update_immediate(struct files_struct
*fsp
)
234 if (fsp
->posix_open
) {
235 /* Don't use delayed writes on POSIX files. */
239 if (fsp
->write_time_forced
) {
241 * No point - "sticky" write times
247 TALLOC_FREE(fsp
->update_write_time_event
);
248 DEBUG(5, ("Update write time immediate on %s\n",
251 fsp
->update_write_time_triggered
= true;
253 fsp
->update_write_time_on_close
= false;
254 update_write_time(fsp
);
257 /****************************************************************************
259 ****************************************************************************/
261 ssize_t
write_file(struct smb_request
*req
,
267 write_cache
*wcp
= fsp
->wcp
;
268 ssize_t total_written
= 0;
271 if (fsp
->print_file
) {
274 if (!rap_to_pjobid(fsp
->rap_print_jobid
, NULL
, &jobid
)) {
275 DEBUG(3,("write_file: Unable to map RAP jobid %u to jobid.\n",
276 (unsigned int)fsp
->rap_print_jobid
));
281 return print_job_write(SNUM(fsp
->conn
), jobid
, data
, pos
, n
);
284 if (!fsp
->can_write
) {
289 if (!fsp
->modified
) {
290 fsp
->modified
= True
;
292 if (SMB_VFS_FSTAT(fsp
, &fsp
->fsp_name
->st
) == 0) {
294 trigger_write_time_update(fsp
);
295 dosmode
= dos_mode(fsp
->conn
, fsp
->fsp_name
);
296 if ((lp_store_dos_attributes(SNUM(fsp
->conn
)) ||
297 MAP_ARCHIVE(fsp
->conn
)) &&
298 !IS_DOS_ARCHIVE(dosmode
)) {
299 file_set_dosmode(fsp
->conn
, fsp
->fsp_name
,
300 dosmode
| aARCH
, NULL
, false);
304 * If this is the first write and we have an exclusive oplock then setup
308 if (EXCLUSIVE_OPLOCK_TYPE(fsp
->oplock_type
) && !wcp
) {
309 setup_write_cache(fsp
,
310 fsp
->fsp_name
->st
.st_ex_size
);
317 DO_PROFILE_INC(writecache_total_writes
);
318 if (!fsp
->oplock_type
) {
319 DO_PROFILE_INC(writecache_non_oplock_writes
);
324 * If this file is level II oplocked then we need
325 * to grab the shared memory lock and inform all
326 * other files with a level II lock that they need
327 * to flush their read caches. We keep the lock over
328 * the shared memory area whilst doing this.
331 /* This should actually be improved to span the write. */
332 contend_level2_oplocks_begin(fsp
, LEVEL2_CONTEND_WRITE
);
333 contend_level2_oplocks_end(fsp
, LEVEL2_CONTEND_WRITE
);
336 if (profile_p
&& profile_p
->writecache_total_writes
% 500 == 0) {
337 DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u total=%u \
338 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
339 profile_p
->writecache_init_writes
,
340 profile_p
->writecache_abutted_writes
,
341 profile_p
->writecache_total_writes
,
342 profile_p
->writecache_non_oplock_writes
,
343 profile_p
->writecache_allocated_write_caches
,
344 profile_p
->writecache_num_write_caches
,
345 profile_p
->writecache_direct_writes
,
346 profile_p
->writecache_num_perfect_writes
,
347 profile_p
->writecache_read_hits
));
349 DEBUG(3,("WRITECACHE: Flushes SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
350 profile_p
->writecache_flushed_writes
[SEEK_FLUSH
],
351 profile_p
->writecache_flushed_writes
[READ_FLUSH
],
352 profile_p
->writecache_flushed_writes
[WRITE_FLUSH
],
353 profile_p
->writecache_flushed_writes
[READRAW_FLUSH
],
354 profile_p
->writecache_flushed_writes
[OPLOCK_RELEASE_FLUSH
],
355 profile_p
->writecache_flushed_writes
[CLOSE_FLUSH
],
356 profile_p
->writecache_flushed_writes
[SYNC_FLUSH
] ));
360 if (wcp
&& req
->unread_bytes
) {
361 /* If we're using receivefile don't
362 * deal with a write cache.
364 flush_write_cache(fsp
, WRITE_FLUSH
);
365 delete_write_cache(fsp
);
370 DO_PROFILE_INC(writecache_direct_writes
);
371 total_written
= real_write_file(req
, fsp
, data
, pos
, n
);
372 return total_written
;
375 DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f "
376 "wcp->data_size=%u\n", fsp_str_dbg(fsp
), fsp
->fh
->fd
,
377 (double)pos
, (unsigned int)n
, (double)wcp
->offset
,
378 (unsigned int)wcp
->data_size
));
380 fsp
->fh
->pos
= pos
+ n
;
383 * If we have active cache and it isn't contiguous then we flush.
384 * NOTE: There is a small problem with running out of disk ....
387 if (wcp
->data_size
) {
388 bool cache_flush_needed
= False
;
390 if ((pos
>= wcp
->offset
) && (pos
<= wcp
->offset
+ wcp
->data_size
)) {
392 /* ASCII art.... JRA.
394 +--------------+-----
395 | Cached data | Rest of allocated cache buffer....
396 +--------------+-----
398 +-------------------+
400 +-------------------+
405 * Start of write overlaps or abutts the existing data.
408 size_t data_used
= MIN((wcp
->alloc_size
- (pos
- wcp
->offset
)), n
);
410 memcpy(wcp
->data
+ (pos
- wcp
->offset
), data
, data_used
);
413 * Update the current buffer size with the new data.
416 if(pos
+ data_used
> wcp
->offset
+ wcp
->data_size
) {
417 wcp
->data_size
= pos
+ data_used
- wcp
->offset
;
421 * Update the file size if changed.
424 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
425 if (wcp_file_size_change(fsp
) == -1) {
431 * If we used all the data then
438 cache_flush_needed
= True
;
441 * Move the start of data forward by the amount used,
442 * cut down the amount left by the same amount.
449 DO_PROFILE_INC(writecache_abutted_writes
);
450 total_written
= data_used
;
454 } else if ((pos
< wcp
->offset
) && (pos
+ n
> wcp
->offset
) &&
455 (pos
+ n
<= wcp
->offset
+ wcp
->alloc_size
)) {
457 /* ASCII art.... JRA.
463 +-------------------+
465 +-------------------+
470 * End of write overlaps the existing data.
473 size_t data_used
= pos
+ n
- wcp
->offset
;
475 memcpy(wcp
->data
, data
+ n
- data_used
, data_used
);
478 * Update the current buffer size with the new data.
481 if(pos
+ n
> wcp
->offset
+ wcp
->data_size
) {
482 wcp
->data_size
= pos
+ n
- wcp
->offset
;
486 * Update the file size if changed.
489 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
490 if (wcp_file_size_change(fsp
) == -1) {
496 * We don't need to move the start of data, but we
497 * cut down the amount left by the amount used.
503 * We cannot have used all the data here.
506 cache_flush_needed
= True
;
508 DO_PROFILE_INC(writecache_abutted_writes
);
509 total_written
= data_used
;
513 } else if ( (pos
>= wcp
->file_size
) &&
514 (wcp
->offset
+ wcp
->data_size
== wcp
->file_size
) &&
515 (pos
> wcp
->offset
+ wcp
->data_size
) &&
516 (pos
< wcp
->offset
+ wcp
->alloc_size
) ) {
518 /* ASCII art.... JRA.
522 +---------------+---------------+
523 | Cached data | Cache buffer |
524 +---------------+---------------+
526 +-------------------+
528 +-------------------+
533 * Non-contiguous write part of which fits within
534 * the cache buffer and is extending the file
535 * and the cache contents reflect the current
536 * data up to the current end of the file.
541 if(pos
+ n
<= wcp
->offset
+ wcp
->alloc_size
) {
544 data_used
= wcp
->offset
+ wcp
->alloc_size
- pos
;
548 * Fill in the non-continuous area with zeros.
551 memset(wcp
->data
+ wcp
->data_size
, '\0',
552 pos
- (wcp
->offset
+ wcp
->data_size
) );
554 memcpy(wcp
->data
+ (pos
- wcp
->offset
), data
, data_used
);
557 * Update the current buffer size with the new data.
560 if(pos
+ data_used
> wcp
->offset
+ wcp
->data_size
) {
561 wcp
->data_size
= pos
+ data_used
- wcp
->offset
;
565 * Update the file size if changed.
568 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
569 if (wcp_file_size_change(fsp
) == -1) {
575 * If we used all the data then
582 cache_flush_needed
= True
;
586 * Move the start of data forward by the amount used,
587 * cut down the amount left by the same amount.
594 DO_PROFILE_INC(writecache_abutted_writes
);
595 total_written
= data_used
;
599 } else if ( (pos
>= wcp
->file_size
) &&
601 (wcp
->file_size
== wcp
->offset
+ wcp
->data_size
) &&
602 (pos
< wcp
->file_size
+ wcp
->alloc_size
)) {
608 +---------------+---------------+
609 | Cached data | Cache buffer |
610 +---------------+---------------+
612 |<------- allocated size ---------------->|
618 MS-Office seems to do this a lot to determine if there's enough
619 space on the filesystem to write a new file.
624 +-----------------------+--------+
625 | Zeroed Cached data | 1 Byte |
626 +-----------------------+--------+
629 flush_write_cache(fsp
, WRITE_FLUSH
);
630 wcp
->offset
= wcp
->file_size
;
631 wcp
->data_size
= pos
- wcp
->file_size
+ 1;
632 memset(wcp
->data
, '\0', wcp
->data_size
);
633 memcpy(wcp
->data
+ wcp
->data_size
-1, data
, 1);
636 * Update the file size if changed.
639 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
640 if (wcp_file_size_change(fsp
) == -1) {
649 /* ASCII art..... JRA.
653 +---------------+---------------+
654 | Cached data | Cache buffer |
655 +---------------+---------------+
657 +-------------------+
659 +-------------------+
663 +---------------+---------------+
664 | Cached data | Cache buffer |
665 +---------------+---------------+
667 +-------------------+
669 +-------------------+
673 +---------------+---------------+
674 | Cached data | Cache buffer |
675 +---------------+---------------+
677 +-----------------------------------------------------+
679 +-----------------------------------------------------+
684 * Write is bigger than buffer, or there is no overlap on the
688 DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
689 len = %u\n",fsp
->fh
->fd
, (double)pos
, (unsigned int)n
, (double)wcp
->offset
, (unsigned int)wcp
->data_size
));
692 * If write would fit in the cache, and is larger than
693 * the data already in the cache, flush the cache and
694 * preferentially copy the data new data into it. Otherwise
695 * just write the data directly.
698 if ( n
<= wcp
->alloc_size
&& n
> wcp
->data_size
) {
699 cache_flush_needed
= True
;
701 ssize_t ret
= real_write_file(NULL
,fsp
, data
, pos
, n
);
704 * If the write overlaps the entire cache, then
705 * discard the current contents of the cache.
706 * Fix from Rasmus Borup Hansen rbh@math.ku.dk.
709 if ((pos
<= wcp
->offset
) &&
710 (pos
+ n
>= wcp
->offset
+ wcp
->data_size
) ) {
711 DEBUG(9,("write_file: discarding overwritten write \
712 cache: fd = %d, off=%.0f, size=%u\n", fsp
->fh
->fd
, (double)wcp
->offset
, (unsigned int)wcp
->data_size
));
716 DO_PROFILE_INC(writecache_direct_writes
);
721 if (pos
+ ret
> wcp
->file_size
) {
722 wcp
->file_size
= pos
+ ret
;
732 if (cache_flush_needed
) {
733 DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
734 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
735 write_path
, fsp
->fh
->fd
, (double)wcp
->file_size
, (double)pos
, (unsigned int)n
,
736 (double)wcp
->offset
, (unsigned int)wcp
->data_size
));
738 flush_write_cache(fsp
, WRITE_FLUSH
);
743 * If the write request is bigger than the cache
744 * size, write it all out.
747 if (n
> wcp
->alloc_size
) {
748 ssize_t ret
= real_write_file(NULL
,fsp
, data
, pos
, n
);
753 if (pos
+ ret
> wcp
->file_size
) {
754 wcp
->file_size
= pos
+ n
;
757 DO_PROFILE_INC(writecache_direct_writes
);
758 return total_written
+ n
;
762 * If there's any data left, cache it.
767 if (wcp
->data_size
) {
768 DO_PROFILE_INC(writecache_abutted_writes
);
770 DO_PROFILE_INC(writecache_init_writes
);
773 memcpy(wcp
->data
+wcp
->data_size
, data
, n
);
774 if (wcp
->data_size
== 0) {
776 DO_PROFILE_INC(writecache_num_write_caches
);
781 * Update the file size if changed.
784 if (wcp
->offset
+ wcp
->data_size
> wcp
->file_size
) {
785 if (wcp_file_size_change(fsp
) == -1) {
789 DEBUG(9,("wcp->offset = %.0f wcp->data_size = %u cache return %u\n",
790 (double)wcp
->offset
, (unsigned int)wcp
->data_size
, (unsigned int)n
));
793 return total_written
; /* .... that's a write :) */
796 return total_written
;
799 /****************************************************************************
800 Delete the write cache structure.
801 ****************************************************************************/
803 void delete_write_cache(files_struct
*fsp
)
811 if(!(wcp
= fsp
->wcp
)) {
815 DO_PROFILE_DEC(writecache_allocated_write_caches
);
816 allocated_write_caches
--;
818 SMB_ASSERT(wcp
->data_size
== 0);
820 SAFE_FREE(wcp
->data
);
823 DEBUG(10,("delete_write_cache: File %s deleted write cache\n",
827 /****************************************************************************
828 Setup the write cache structure.
829 ****************************************************************************/
831 static bool setup_write_cache(files_struct
*fsp
, SMB_OFF_T file_size
)
833 ssize_t alloc_size
= lp_write_cache_size(SNUM(fsp
->conn
));
836 if (allocated_write_caches
>= MAX_WRITE_CACHES
) {
840 if(alloc_size
== 0 || fsp
->wcp
) {
844 if((wcp
= SMB_MALLOC_P(write_cache
)) == NULL
) {
845 DEBUG(0,("setup_write_cache: malloc fail.\n"));
849 wcp
->file_size
= file_size
;
851 wcp
->alloc_size
= alloc_size
;
853 if((wcp
->data
= (char *)SMB_MALLOC(wcp
->alloc_size
)) == NULL
) {
854 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
855 (unsigned int)wcp
->alloc_size
));
860 memset(wcp
->data
, '\0', wcp
->alloc_size
);
863 DO_PROFILE_INC(writecache_allocated_write_caches
);
864 allocated_write_caches
++;
866 DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
867 fsp_str_dbg(fsp
), (unsigned long)wcp
->alloc_size
));
872 /****************************************************************************
873 Cope with a size change.
874 ****************************************************************************/
876 void set_filelen_write_cache(files_struct
*fsp
, SMB_OFF_T file_size
)
879 /* The cache *must* have been flushed before we do this. */
880 if (fsp
->wcp
->data_size
!= 0) {
882 if (asprintf(&msg
, "set_filelen_write_cache: size change "
883 "on file %s with write cache size = %lu\n",
884 fsp
->fsp_name
->base_name
,
885 (unsigned long)fsp
->wcp
->data_size
) != -1) {
888 smb_panic("set_filelen_write_cache");
891 fsp
->wcp
->file_size
= file_size
;
895 /*******************************************************************
896 Flush a write cache struct to disk.
897 ********************************************************************/
899 ssize_t
flush_write_cache(files_struct
*fsp
, enum flush_reason_enum reason
)
901 write_cache
*wcp
= fsp
->wcp
;
905 if(!wcp
|| !wcp
->data_size
) {
909 data_size
= wcp
->data_size
;
912 DO_PROFILE_DEC_INC(writecache_num_write_caches
,writecache_flushed_writes
[reason
]);
914 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
915 fsp
->fh
->fd
, (double)wcp
->offset
, (unsigned int)data_size
));
918 if(data_size
== wcp
->alloc_size
) {
919 DO_PROFILE_INC(writecache_num_perfect_writes
);
923 ret
= real_write_file(NULL
, fsp
, wcp
->data
, wcp
->offset
, data_size
);
926 * Ensure file size if kept up to date if write extends file.
929 if ((ret
!= -1) && (wcp
->offset
+ ret
> wcp
->file_size
)) {
930 wcp
->file_size
= wcp
->offset
+ ret
;
936 /*******************************************************************
938 ********************************************************************/
940 NTSTATUS
sync_file(connection_struct
*conn
, files_struct
*fsp
, bool write_through
)
942 if (fsp
->fh
->fd
== -1)
943 return NT_STATUS_INVALID_HANDLE
;
945 if (lp_strict_sync(SNUM(conn
)) &&
946 (lp_syncalways(SNUM(conn
)) || write_through
)) {
947 int ret
= flush_write_cache(fsp
, SYNC_FLUSH
);
949 return map_nt_error_from_unix(errno
);
951 ret
= SMB_VFS_FSYNC(fsp
);
953 return map_nt_error_from_unix(errno
);
959 /************************************************************
960 Perform a stat whether a valid fd or not.
961 ************************************************************/
963 int fsp_stat(files_struct
*fsp
, SMB_STRUCT_STAT
*pst
)
965 if (fsp
->fh
->fd
== -1) {
968 ret
= SMB_VFS_STAT(fsp
->conn
, fsp
->fsp_name
);
970 *pst
= fsp
->fsp_name
->st
;
974 return SMB_VFS_FSTAT(fsp
, pst
);