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(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 ));
146 * It turns out that setting the last write time from a Windows
147 * client stops any subsequent writes from updating the write time.
148 * Doing this after the write gives a race condition here where
149 * a stat may see the changed write time before we reset it here,
150 * but it's cheaper than having to store the write time in shared
151 * memory and look it up using dev/inode across all running smbd's.
152 * The 99% solution will hopefully be good enough in this case. JRA.
155 if (!null_timespec(fsp->pending_modtime)) {
156 set_filetime(fsp->conn, fsp->fsp_name, fsp->pending_modtime);
158 /* If we didn't get the "set modtime" call ourselves, we must
159 store the last write time to restore on close. JRA. */
160 if (!fsp->pending_modtime_owner) {
161 fsp->last_write_time = timespec_current();
165 /* Yes - this is correct - writes don't update this. JRA. */
166 /* Found by Samba4 tests. */
168 fsp->position_information = fsp->pos;
175 /****************************************************************************
176 File size cache change.
177 Updates size on disk but doesn't flush the cache.
178 ****************************************************************************/
180 static int wcp_file_size_change(files_struct *fsp)
183 write_cache *wcp = fsp->wcp;
185 wcp->file_size = wcp->offset + wcp->data_size;
186 ret = SMB_VFS_FTRUNCATE(fsp, fsp->fh->fd, wcp->file_size);
188 DEBUG(0,("wcp_file_size_change (%s): ftruncate of size %.0f error %s\n",
189 fsp->fsp_name, (double)wcp->file_size, strerror(errno) ));
194 /****************************************************************************
196 ****************************************************************************/
198 ssize_t write_file(struct smb_request *req,
204 write_cache *wcp = fsp->wcp;
205 ssize_t total_written = 0;
208 if (fsp->print_file) {
212 if (!rap_to_pjobid(fsp->rap_print_jobid, sharename, &jobid)) {
213 DEBUG(3,("write_file: Unable to map RAP jobid %u to jobid.\n",
214 (unsigned int)fsp->rap_print_jobid ));
219 return print_job_write(SNUM(fsp->conn), jobid, data, pos, n);
222 if (!fsp->can_write) {
227 if (!fsp->modified) {
229 fsp->modified = True;
231 if (SMB_VFS_FSTAT(fsp,fsp->fh->fd,&st) == 0) {
232 int dosmode = dos_mode(fsp->conn,fsp->fsp_name,&st);
233 if ((lp_store_dos_attributes(SNUM(fsp->conn)) || MAP_ARCHIVE(fsp->conn)) && !IS_DOS_ARCHIVE(dosmode)) {
234 file_set_dosmode(fsp->conn,fsp->fsp_name,dosmode | aARCH,&st, False);
238 * If this is the first write and we have an exclusive oplock then setup
242 if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && !wcp) {
243 setup_write_cache(fsp, st.st_size);
250 DO_PROFILE_INC(writecache_total_writes);
251 if (!fsp->oplock_type) {
252 DO_PROFILE_INC(writecache_non_oplock_writes);
257 * If this file is level II oplocked then we need
258 * to grab the shared memory lock and inform all
259 * other files with a level II lock that they need
260 * to flush their read caches. We keep the lock over
261 * the shared memory area whilst doing this.
264 release_level_2_oplocks_on_change(fsp);
267 if (profile_p && profile_p->writecache_total_writes % 500 == 0) {
268 DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u total=%u \
269 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
270 profile_p->writecache_init_writes,
271 profile_p->writecache_abutted_writes,
272 profile_p->writecache_total_writes,
273 profile_p->writecache_non_oplock_writes,
274 profile_p->writecache_allocated_write_caches,
275 profile_p->writecache_num_write_caches,
276 profile_p->writecache_direct_writes,
277 profile_p->writecache_num_perfect_writes,
278 profile_p->writecache_read_hits ));
280 DEBUG(3,("WRITECACHE: Flushes SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
281 profile_p->writecache_flushed_writes[SEEK_FLUSH],
282 profile_p->writecache_flushed_writes[READ_FLUSH],
283 profile_p->writecache_flushed_writes[WRITE_FLUSH],
284 profile_p->writecache_flushed_writes[READRAW_FLUSH],
285 profile_p->writecache_flushed_writes[OPLOCK_RELEASE_FLUSH],
286 profile_p->writecache_flushed_writes[CLOSE_FLUSH],
287 profile_p->writecache_flushed_writes[SYNC_FLUSH] ));
291 if (wcp && req->unread_bytes) {
292 /* If we're using receivefile don't
293 * deal with a write cache.
295 flush_write_cache(fsp, WRITE_FLUSH);
296 delete_write_cache(fsp);
301 DO_PROFILE_INC(writecache_direct_writes);
302 total_written = real_write_file(req, fsp, data, pos, n);
303 return total_written;
306 DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f wcp->data_size=%u\n",
307 fsp->fsp_name, fsp->fh->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size));
309 fsp->fh->pos = pos + n;
312 * If we have active cache and it isn't contiguous then we flush.
313 * NOTE: There is a small problem with running out of disk ....
316 if (wcp->data_size) {
317 bool cache_flush_needed = False;
319 if ((pos >= wcp->offset) && (pos <= wcp->offset + wcp->data_size)) {
321 /* ASCII art.... JRA.
323 +--------------+-----
324 | Cached data | Rest of allocated cache buffer....
325 +--------------+-----
327 +-------------------+
329 +-------------------+
334 * Start of write overlaps or abutts the existing data.
337 size_t data_used = MIN((wcp->alloc_size - (pos - wcp->offset)), n);
339 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
342 * Update the current buffer size with the new data.
345 if(pos + data_used > wcp->offset + wcp->data_size) {
346 wcp->data_size = pos + data_used - wcp->offset;
350 * Update the file size if changed.
353 if (wcp->offset + wcp->data_size > wcp->file_size) {
354 if (wcp_file_size_change(fsp) == -1) {
360 * If we used all the data then
367 cache_flush_needed = True;
370 * Move the start of data forward by the amount used,
371 * cut down the amount left by the same amount.
378 DO_PROFILE_INC(writecache_abutted_writes);
379 total_written = data_used;
383 } else if ((pos < wcp->offset) && (pos + n > wcp->offset) &&
384 (pos + n <= wcp->offset + wcp->alloc_size)) {
386 /* ASCII art.... JRA.
392 +-------------------+
394 +-------------------+
399 * End of write overlaps the existing data.
402 size_t data_used = pos + n - wcp->offset;
404 memcpy(wcp->data, data + n - data_used, data_used);
407 * Update the current buffer size with the new data.
410 if(pos + n > wcp->offset + wcp->data_size) {
411 wcp->data_size = pos + n - wcp->offset;
415 * Update the file size if changed.
418 if (wcp->offset + wcp->data_size > wcp->file_size) {
419 if (wcp_file_size_change(fsp) == -1) {
425 * We don't need to move the start of data, but we
426 * cut down the amount left by the amount used.
432 * We cannot have used all the data here.
435 cache_flush_needed = True;
437 DO_PROFILE_INC(writecache_abutted_writes);
438 total_written = data_used;
442 } else if ( (pos >= wcp->file_size) &&
443 (wcp->offset + wcp->data_size == wcp->file_size) &&
444 (pos > wcp->offset + wcp->data_size) &&
445 (pos < wcp->offset + wcp->alloc_size) ) {
447 /* ASCII art.... JRA.
451 +---------------+---------------+
452 | Cached data | Cache buffer |
453 +---------------+---------------+
455 +-------------------+
457 +-------------------+
462 * Non-contiguous write part of which fits within
463 * the cache buffer and is extending the file
464 * and the cache contents reflect the current
465 * data up to the current end of the file.
470 if(pos + n <= wcp->offset + wcp->alloc_size) {
473 data_used = wcp->offset + wcp->alloc_size - pos;
477 * Fill in the non-continuous area with zeros.
480 memset(wcp->data + wcp->data_size, '\0',
481 pos - (wcp->offset + wcp->data_size) );
483 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
486 * Update the current buffer size with the new data.
489 if(pos + data_used > wcp->offset + wcp->data_size) {
490 wcp->data_size = pos + data_used - wcp->offset;
494 * Update the file size if changed.
497 if (wcp->offset + wcp->data_size > wcp->file_size) {
498 if (wcp_file_size_change(fsp) == -1) {
504 * If we used all the data then
511 cache_flush_needed = True;
515 * Move the start of data forward by the amount used,
516 * cut down the amount left by the same amount.
523 DO_PROFILE_INC(writecache_abutted_writes);
524 total_written = data_used;
528 } else if ( (pos >= wcp->file_size) &&
530 (wcp->file_size == wcp->offset + wcp->data_size) &&
531 (pos < wcp->file_size + wcp->alloc_size)) {
537 +---------------+---------------+
538 | Cached data | Cache buffer |
539 +---------------+---------------+
541 |<------- allocated size ---------------->|
547 MS-Office seems to do this a lot to determine if there's enough
548 space on the filesystem to write a new file.
553 +-----------------------+--------+
554 | Zeroed Cached data | 1 Byte |
555 +-----------------------+--------+
558 flush_write_cache(fsp, WRITE_FLUSH);
559 wcp->offset = wcp->file_size;
560 wcp->data_size = pos - wcp->file_size + 1;
561 memset(wcp->data, '\0', wcp->data_size);
562 memcpy(wcp->data + wcp->data_size-1, data, 1);
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) {
578 /* ASCII art..... JRA.
582 +---------------+---------------+
583 | Cached data | Cache buffer |
584 +---------------+---------------+
586 +-------------------+
588 +-------------------+
592 +---------------+---------------+
593 | Cached data | Cache buffer |
594 +---------------+---------------+
596 +-------------------+
598 +-------------------+
602 +---------------+---------------+
603 | Cached data | Cache buffer |
604 +---------------+---------------+
606 +-----------------------------------------------------+
608 +-----------------------------------------------------+
613 * Write is bigger than buffer, or there is no overlap on the
617 DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
618 len = %u\n",fsp->fh->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size ));
621 * If write would fit in the cache, and is larger than
622 * the data already in the cache, flush the cache and
623 * preferentially copy the data new data into it. Otherwise
624 * just write the data directly.
627 if ( n <= wcp->alloc_size && n > wcp->data_size) {
628 cache_flush_needed = True;
630 ssize_t ret = real_write_file(NULL,fsp, data, pos, n);
633 * If the write overlaps the entire cache, then
634 * discard the current contents of the cache.
635 * Fix from Rasmus Borup Hansen rbh@math.ku.dk.
638 if ((pos <= wcp->offset) &&
639 (pos + n >= wcp->offset + wcp->data_size) ) {
640 DEBUG(9,("write_file: discarding overwritten write \
641 cache: fd = %d, off=%.0f, size=%u\n", fsp->fh->fd, (double)wcp->offset, (unsigned int)wcp->data_size ));
645 DO_PROFILE_INC(writecache_direct_writes);
650 if (pos + ret > wcp->file_size) {
651 wcp->file_size = pos + ret;
661 if (cache_flush_needed) {
662 DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
663 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
664 write_path, fsp->fh->fd, (double)wcp->file_size, (double)pos, (unsigned int)n,
665 (double)wcp->offset, (unsigned int)wcp->data_size ));
667 flush_write_cache(fsp, WRITE_FLUSH);
672 * If the write request is bigger than the cache
673 * size, write it all out.
676 if (n > wcp->alloc_size ) {
677 ssize_t ret = real_write_file(NULL,fsp, data, pos, n);
682 if (pos + ret > wcp->file_size) {
683 wcp->file_size = pos + n;
686 DO_PROFILE_INC(writecache_direct_writes);
687 return total_written + n;
691 * If there's any data left, cache it.
696 if (wcp->data_size) {
697 DO_PROFILE_INC(writecache_abutted_writes);
699 DO_PROFILE_INC(writecache_init_writes);
702 memcpy(wcp->data+wcp->data_size, data, n);
703 if (wcp->data_size == 0) {
705 DO_PROFILE_INC(writecache_num_write_caches);
710 * Update the file size if changed.
713 if (wcp->offset + wcp->data_size > wcp->file_size) {
714 if (wcp_file_size_change(fsp) == -1) {
718 DEBUG(9,("wcp->offset = %.0f wcp->data_size = %u cache return %u\n",
719 (double)wcp->offset, (unsigned int)wcp->data_size, (unsigned int)n));
722 return total_written; /* .... that's a write :) */
725 return total_written;
728 /****************************************************************************
729 Delete the write cache structure.
730 ****************************************************************************/
732 void delete_write_cache(files_struct *fsp)
740 if(!(wcp = fsp->wcp)) {
744 DO_PROFILE_DEC(writecache_allocated_write_caches);
745 allocated_write_caches--;
747 SMB_ASSERT(wcp->data_size == 0);
749 SAFE_FREE(wcp->data);
752 DEBUG(10,("delete_write_cache: File %s deleted write cache\n", fsp->fsp_name ));
755 /****************************************************************************
756 Setup the write cache structure.
757 ****************************************************************************/
759 static bool setup_write_cache(files_struct *fsp, SMB_OFF_T file_size)
761 ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
764 if (allocated_write_caches >= MAX_WRITE_CACHES) {
768 if(alloc_size == 0 || fsp->wcp) {
772 if((wcp = SMB_MALLOC_P(write_cache)) == NULL) {
773 DEBUG(0,("setup_write_cache: malloc fail.\n"));
777 wcp->file_size = file_size;
779 wcp->alloc_size = alloc_size;
781 if((wcp->data = (char *)SMB_MALLOC(wcp->alloc_size)) == NULL) {
782 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
783 (unsigned int)wcp->alloc_size ));
788 memset(wcp->data, '\0', wcp->alloc_size );
791 DO_PROFILE_INC(writecache_allocated_write_caches);
792 allocated_write_caches++;
794 DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
795 fsp->fsp_name, (unsigned long)wcp->alloc_size ));
800 /****************************************************************************
801 Cope with a size change.
802 ****************************************************************************/
804 void set_filelen_write_cache(files_struct *fsp, SMB_OFF_T file_size)
807 /* The cache *must* have been flushed before we do this. */
808 if (fsp->wcp->data_size != 0) {
810 asprintf(&msg, "set_filelen_write_cache: size change "
811 "on file %s with write cache size = %lu\n",
813 (unsigned long)fsp->wcp->data_size);
816 fsp->wcp->file_size = file_size;
820 /*******************************************************************
821 Flush a write cache struct to disk.
822 ********************************************************************/
824 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
826 write_cache *wcp = fsp->wcp;
830 if(!wcp || !wcp->data_size) {
834 data_size = wcp->data_size;
837 DO_PROFILE_DEC_INC(writecache_num_write_caches,writecache_flushed_writes[reason]);
839 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
840 fsp->fh->fd, (double)wcp->offset, (unsigned int)data_size));
843 if(data_size == wcp->alloc_size) {
844 DO_PROFILE_INC(writecache_num_perfect_writes);
848 ret = real_write_file(NULL, fsp, wcp->data, wcp->offset, data_size);
851 * Ensure file size if kept up to date if write extends file.
854 if ((ret != -1) && (wcp->offset + ret > wcp->file_size)) {
855 wcp->file_size = wcp->offset + ret;
861 /*******************************************************************
863 ********************************************************************/
865 NTSTATUS sync_file(connection_struct *conn, files_struct *fsp, bool write_through)
867 if (fsp->fh->fd == -1)
868 return NT_STATUS_INVALID_HANDLE;
870 if (lp_strict_sync(SNUM(conn)) &&
871 (lp_syncalways(SNUM(conn)) || write_through)) {
872 int ret = flush_write_cache(fsp, SYNC_FLUSH);
874 return map_nt_error_from_unix(errno);
876 ret = SMB_VFS_FSYNC(fsp,fsp->fh->fd);
878 return map_nt_error_from_unix(errno);
884 /************************************************************
885 Perform a stat whether a valid fd or not.
886 ************************************************************/
888 int fsp_stat(files_struct *fsp, SMB_STRUCT_STAT *pst)
890 if (fsp->fh->fd == -1) {
891 return SMB_VFS_STAT(fsp->conn, fsp->fsp_name, pst);
893 return SMB_VFS_FSTAT(fsp,fsp->fh->fd, pst);