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,fsp->fh->fd,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,fsp->fh->fd,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(files_struct *fsp,const char *data, SMB_OFF_T pos, size_t n)
124 ret = vfs_write_data(fsp, data, n);
127 if (pos && lp_strict_allocate(SNUM(fsp->conn))) {
128 if (vfs_fill_sparse(fsp, pos) == -1) {
132 ret = vfs_pwrite_data(fsp, data, n, pos);
135 DEBUG(10,("real_write_file (%s): pos = %.0f, size = %lu, returned %ld\n",
136 fsp->fsp_name, (double)pos, (unsigned long)n, (long)ret ));
142 * It turns out that setting the last write time from a Windows
143 * client stops any subsequent writes from updating the write time.
144 * Doing this after the write gives a race condition here where
145 * a stat may see the changed write time before we reset it here,
146 * but it's cheaper than having to store the write time in shared
147 * memory and look it up using dev/inode across all running smbd's.
148 * The 99% solution will hopefully be good enough in this case. JRA.
151 if (!null_timespec(fsp->pending_modtime)) {
152 set_filetime(fsp->conn, fsp->fsp_name, fsp->pending_modtime);
154 /* If we didn't get the "set modtime" call ourselves, we must
155 store the last write time to restore on close. JRA. */
156 if (!fsp->pending_modtime_owner) {
157 fsp->last_write_time = timespec_current();
161 /* Yes - this is correct - writes don't update this. JRA. */
162 /* Found by Samba4 tests. */
164 fsp->position_information = fsp->pos;
171 /****************************************************************************
172 File size cache change.
173 Updates size on disk but doesn't flush the cache.
174 ****************************************************************************/
176 static int wcp_file_size_change(files_struct *fsp)
179 write_cache *wcp = fsp->wcp;
181 wcp->file_size = wcp->offset + wcp->data_size;
182 ret = SMB_VFS_FTRUNCATE(fsp, fsp->fh->fd, wcp->file_size);
184 DEBUG(0,("wcp_file_size_change (%s): ftruncate of size %.0f error %s\n",
185 fsp->fsp_name, (double)wcp->file_size, strerror(errno) ));
190 /****************************************************************************
192 ****************************************************************************/
194 ssize_t write_file(files_struct *fsp, const char *data, SMB_OFF_T pos, size_t n)
196 write_cache *wcp = fsp->wcp;
197 ssize_t total_written = 0;
200 if (fsp->print_file) {
204 if (!rap_to_pjobid(fsp->rap_print_jobid, sharename, &jobid)) {
205 DEBUG(3,("write_file: Unable to map RAP jobid %u to jobid.\n",
206 (unsigned int)fsp->rap_print_jobid ));
211 return print_job_write(SNUM(fsp->conn), jobid, data, pos, n);
214 if (!fsp->can_write) {
219 if (!fsp->modified) {
221 fsp->modified = True;
223 if (SMB_VFS_FSTAT(fsp,fsp->fh->fd,&st) == 0) {
224 int dosmode = dos_mode(fsp->conn,fsp->fsp_name,&st);
225 if ((lp_store_dos_attributes(SNUM(fsp->conn)) || MAP_ARCHIVE(fsp->conn)) && !IS_DOS_ARCHIVE(dosmode)) {
226 file_set_dosmode(fsp->conn,fsp->fsp_name,dosmode | aARCH,&st, False);
230 * If this is the first write and we have an exclusive oplock then setup
234 if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && !wcp) {
235 setup_write_cache(fsp, st.st_size);
242 DO_PROFILE_INC(writecache_total_writes);
243 if (!fsp->oplock_type) {
244 DO_PROFILE_INC(writecache_non_oplock_writes);
249 * If this file is level II oplocked then we need
250 * to grab the shared memory lock and inform all
251 * other files with a level II lock that they need
252 * to flush their read caches. We keep the lock over
253 * the shared memory area whilst doing this.
256 release_level_2_oplocks_on_change(fsp);
259 if (profile_p && profile_p->writecache_total_writes % 500 == 0) {
260 DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u total=%u \
261 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
262 profile_p->writecache_init_writes,
263 profile_p->writecache_abutted_writes,
264 profile_p->writecache_total_writes,
265 profile_p->writecache_non_oplock_writes,
266 profile_p->writecache_allocated_write_caches,
267 profile_p->writecache_num_write_caches,
268 profile_p->writecache_direct_writes,
269 profile_p->writecache_num_perfect_writes,
270 profile_p->writecache_read_hits ));
272 DEBUG(3,("WRITECACHE: Flushes SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
273 profile_p->writecache_flushed_writes[SEEK_FLUSH],
274 profile_p->writecache_flushed_writes[READ_FLUSH],
275 profile_p->writecache_flushed_writes[WRITE_FLUSH],
276 profile_p->writecache_flushed_writes[READRAW_FLUSH],
277 profile_p->writecache_flushed_writes[OPLOCK_RELEASE_FLUSH],
278 profile_p->writecache_flushed_writes[CLOSE_FLUSH],
279 profile_p->writecache_flushed_writes[SYNC_FLUSH] ));
284 DO_PROFILE_INC(writecache_direct_writes);
285 total_written = real_write_file(fsp, data, pos, n);
286 return total_written;
289 DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f wcp->data_size=%u\n",
290 fsp->fsp_name, fsp->fh->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size));
292 fsp->fh->pos = pos + n;
295 * If we have active cache and it isn't contiguous then we flush.
296 * NOTE: There is a small problem with running out of disk ....
299 if (wcp->data_size) {
300 BOOL cache_flush_needed = False;
302 if ((pos >= wcp->offset) && (pos <= wcp->offset + wcp->data_size)) {
304 /* ASCII art.... JRA.
306 +--------------+-----
307 | Cached data | Rest of allocated cache buffer....
308 +--------------+-----
310 +-------------------+
312 +-------------------+
317 * Start of write overlaps or abutts the existing data.
320 size_t data_used = MIN((wcp->alloc_size - (pos - wcp->offset)), n);
322 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
325 * Update the current buffer size with the new data.
328 if(pos + data_used > wcp->offset + wcp->data_size) {
329 wcp->data_size = pos + data_used - wcp->offset;
333 * Update the file size if changed.
336 if (wcp->offset + wcp->data_size > wcp->file_size) {
337 if (wcp_file_size_change(fsp) == -1) {
343 * If we used all the data then
350 cache_flush_needed = True;
353 * Move the start of data forward by the amount used,
354 * cut down the amount left by the same amount.
361 DO_PROFILE_INC(writecache_abutted_writes);
362 total_written = data_used;
366 } else if ((pos < wcp->offset) && (pos + n > wcp->offset) &&
367 (pos + n <= wcp->offset + wcp->alloc_size)) {
369 /* ASCII art.... JRA.
375 +-------------------+
377 +-------------------+
382 * End of write overlaps the existing data.
385 size_t data_used = pos + n - wcp->offset;
387 memcpy(wcp->data, data + n - data_used, data_used);
390 * Update the current buffer size with the new data.
393 if(pos + n > wcp->offset + wcp->data_size) {
394 wcp->data_size = pos + n - wcp->offset;
398 * Update the file size if changed.
401 if (wcp->offset + wcp->data_size > wcp->file_size) {
402 if (wcp_file_size_change(fsp) == -1) {
408 * We don't need to move the start of data, but we
409 * cut down the amount left by the amount used.
415 * We cannot have used all the data here.
418 cache_flush_needed = True;
420 DO_PROFILE_INC(writecache_abutted_writes);
421 total_written = data_used;
425 } else if ( (pos >= wcp->file_size) &&
426 (wcp->offset + wcp->data_size == wcp->file_size) &&
427 (pos > wcp->offset + wcp->data_size) &&
428 (pos < wcp->offset + wcp->alloc_size) ) {
430 /* ASCII art.... JRA.
434 +---------------+---------------+
435 | Cached data | Cache buffer |
436 +---------------+---------------+
438 +-------------------+
440 +-------------------+
445 * Non-contiguous write part of which fits within
446 * the cache buffer and is extending the file
447 * and the cache contents reflect the current
448 * data up to the current end of the file.
453 if(pos + n <= wcp->offset + wcp->alloc_size) {
456 data_used = wcp->offset + wcp->alloc_size - pos;
460 * Fill in the non-continuous area with zeros.
463 memset(wcp->data + wcp->data_size, '\0',
464 pos - (wcp->offset + wcp->data_size) );
466 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
469 * Update the current buffer size with the new data.
472 if(pos + data_used > wcp->offset + wcp->data_size) {
473 wcp->data_size = pos + data_used - wcp->offset;
477 * Update the file size if changed.
480 if (wcp->offset + wcp->data_size > wcp->file_size) {
481 if (wcp_file_size_change(fsp) == -1) {
487 * If we used all the data then
494 cache_flush_needed = True;
498 * Move the start of data forward by the amount used,
499 * cut down the amount left by the same amount.
506 DO_PROFILE_INC(writecache_abutted_writes);
507 total_written = data_used;
511 } else if ( (pos >= wcp->file_size) &&
513 (wcp->file_size == wcp->offset + wcp->data_size) &&
514 (pos < wcp->file_size + wcp->alloc_size)) {
520 +---------------+---------------+
521 | Cached data | Cache buffer |
522 +---------------+---------------+
524 |<------- allocated size ---------------->|
530 MS-Office seems to do this a lot to determine if there's enough
531 space on the filesystem to write a new file.
536 +-----------------------+--------+
537 | Zeroed Cached data | 1 Byte |
538 +-----------------------+--------+
541 flush_write_cache(fsp, WRITE_FLUSH);
542 wcp->offset = wcp->file_size;
543 wcp->data_size = pos - wcp->file_size + 1;
544 memset(wcp->data, '\0', wcp->data_size);
545 memcpy(wcp->data + wcp->data_size-1, data, 1);
548 * Update the file size if changed.
551 if (wcp->offset + wcp->data_size > wcp->file_size) {
552 if (wcp_file_size_change(fsp) == -1) {
561 /* ASCII art..... JRA.
565 +---------------+---------------+
566 | Cached data | Cache buffer |
567 +---------------+---------------+
569 +-------------------+
571 +-------------------+
575 +---------------+---------------+
576 | Cached data | Cache buffer |
577 +---------------+---------------+
579 +-------------------+
581 +-------------------+
585 +---------------+---------------+
586 | Cached data | Cache buffer |
587 +---------------+---------------+
589 +-----------------------------------------------------+
591 +-----------------------------------------------------+
596 * Write is bigger than buffer, or there is no overlap on the
600 DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
601 len = %u\n",fsp->fh->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size ));
604 * If write would fit in the cache, and is larger than
605 * the data already in the cache, flush the cache and
606 * preferentially copy the data new data into it. Otherwise
607 * just write the data directly.
610 if ( n <= wcp->alloc_size && n > wcp->data_size) {
611 cache_flush_needed = True;
613 ssize_t ret = real_write_file(fsp, data, pos, n);
616 * If the write overlaps the entire cache, then
617 * discard the current contents of the cache.
618 * Fix from Rasmus Borup Hansen rbh@math.ku.dk.
621 if ((pos <= wcp->offset) &&
622 (pos + n >= wcp->offset + wcp->data_size) ) {
623 DEBUG(9,("write_file: discarding overwritten write \
624 cache: fd = %d, off=%.0f, size=%u\n", fsp->fh->fd, (double)wcp->offset, (unsigned int)wcp->data_size ));
628 DO_PROFILE_INC(writecache_direct_writes);
633 if (pos + ret > wcp->file_size) {
634 wcp->file_size = pos + ret;
644 if (cache_flush_needed) {
645 DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
646 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
647 write_path, fsp->fh->fd, (double)wcp->file_size, (double)pos, (unsigned int)n,
648 (double)wcp->offset, (unsigned int)wcp->data_size ));
650 flush_write_cache(fsp, WRITE_FLUSH);
655 * If the write request is bigger than the cache
656 * size, write it all out.
659 if (n > wcp->alloc_size ) {
660 ssize_t ret = real_write_file(fsp, data, pos, n);
665 if (pos + ret > wcp->file_size) {
666 wcp->file_size = pos + n;
669 DO_PROFILE_INC(writecache_direct_writes);
670 return total_written + n;
674 * If there's any data left, cache it.
679 if (wcp->data_size) {
680 DO_PROFILE_INC(writecache_abutted_writes);
682 DO_PROFILE_INC(writecache_init_writes);
685 memcpy(wcp->data+wcp->data_size, data, n);
686 if (wcp->data_size == 0) {
688 DO_PROFILE_INC(writecache_num_write_caches);
693 * Update the file size if changed.
696 if (wcp->offset + wcp->data_size > wcp->file_size) {
697 if (wcp_file_size_change(fsp) == -1) {
701 DEBUG(9,("wcp->offset = %.0f wcp->data_size = %u cache return %u\n",
702 (double)wcp->offset, (unsigned int)wcp->data_size, (unsigned int)n));
705 return total_written; /* .... that's a write :) */
708 return total_written;
711 /****************************************************************************
712 Delete the write cache structure.
713 ****************************************************************************/
715 void delete_write_cache(files_struct *fsp)
723 if(!(wcp = fsp->wcp)) {
727 DO_PROFILE_DEC(writecache_allocated_write_caches);
728 allocated_write_caches--;
730 SMB_ASSERT(wcp->data_size == 0);
732 SAFE_FREE(wcp->data);
735 DEBUG(10,("delete_write_cache: File %s deleted write cache\n", fsp->fsp_name ));
738 /****************************************************************************
739 Setup the write cache structure.
740 ****************************************************************************/
742 static BOOL setup_write_cache(files_struct *fsp, SMB_OFF_T file_size)
744 ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
747 if (allocated_write_caches >= MAX_WRITE_CACHES) {
751 if(alloc_size == 0 || fsp->wcp) {
755 if((wcp = SMB_MALLOC_P(write_cache)) == NULL) {
756 DEBUG(0,("setup_write_cache: malloc fail.\n"));
760 wcp->file_size = file_size;
762 wcp->alloc_size = alloc_size;
764 if((wcp->data = (char *)SMB_MALLOC(wcp->alloc_size)) == NULL) {
765 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
766 (unsigned int)wcp->alloc_size ));
771 memset(wcp->data, '\0', wcp->alloc_size );
774 DO_PROFILE_INC(writecache_allocated_write_caches);
775 allocated_write_caches++;
777 DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
778 fsp->fsp_name, (unsigned long)wcp->alloc_size ));
783 /****************************************************************************
784 Cope with a size change.
785 ****************************************************************************/
787 void set_filelen_write_cache(files_struct *fsp, SMB_OFF_T file_size)
790 /* The cache *must* have been flushed before we do this. */
791 if (fsp->wcp->data_size != 0) {
793 asprintf(&msg, "set_filelen_write_cache: size change "
794 "on file %s with write cache size = %lu\n",
796 (unsigned long)fsp->wcp->data_size);
799 fsp->wcp->file_size = file_size;
803 /*******************************************************************
804 Flush a write cache struct to disk.
805 ********************************************************************/
807 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
809 write_cache *wcp = fsp->wcp;
813 if(!wcp || !wcp->data_size) {
817 data_size = wcp->data_size;
820 DO_PROFILE_DEC_INC(writecache_num_write_caches,writecache_flushed_writes[reason]);
822 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
823 fsp->fh->fd, (double)wcp->offset, (unsigned int)data_size));
826 if(data_size == wcp->alloc_size) {
827 DO_PROFILE_INC(writecache_num_perfect_writes);
831 ret = real_write_file(fsp, wcp->data, wcp->offset, data_size);
834 * Ensure file size if kept up to date if write extends file.
837 if ((ret != -1) && (wcp->offset + ret > wcp->file_size)) {
838 wcp->file_size = wcp->offset + ret;
844 /*******************************************************************
846 ********************************************************************/
848 NTSTATUS sync_file(connection_struct *conn, files_struct *fsp, BOOL write_through)
850 if (fsp->fh->fd == -1)
851 return NT_STATUS_INVALID_HANDLE;
853 if (lp_strict_sync(SNUM(conn)) &&
854 (lp_syncalways(SNUM(conn)) || write_through)) {
855 int ret = flush_write_cache(fsp, SYNC_FLUSH);
857 return map_nt_error_from_unix(errno);
859 ret = SMB_VFS_FSYNC(fsp,fsp->fh->fd);
861 return map_nt_error_from_unix(errno);
867 /************************************************************
868 Perform a stat whether a valid fd or not.
869 ************************************************************/
871 int fsp_stat(files_struct *fsp, SMB_STRUCT_STAT *pst)
873 if (fsp->fh->fd == -1) {
874 return SMB_VFS_STAT(fsp->conn, fsp->fsp_name, pst);
876 return SMB_VFS_FSTAT(fsp,fsp->fh->fd, pst);