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,
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, fsp);
227 void trigger_write_time_update_immediate(struct files_struct *fsp)
229 if (fsp->write_time_forced) {
231 * No point - "sticky" write times
237 TALLOC_FREE(fsp->update_write_time_event);
238 DEBUG(5, ("Update write time immediate on %s\n", fsp->fsp_name));
240 fsp->update_write_time_triggered = true;
242 fsp->update_write_time_on_close = false;
243 update_write_time(fsp);
246 /****************************************************************************
248 ****************************************************************************/
250 ssize_t write_file(struct smb_request *req,
256 write_cache *wcp = fsp->wcp;
257 ssize_t total_written = 0;
260 if (fsp->print_file) {
264 if (!rap_to_pjobid(fsp->rap_print_jobid, sharename, &jobid)) {
265 DEBUG(3,("write_file: Unable to map RAP jobid %u to jobid.\n",
266 (unsigned int)fsp->rap_print_jobid ));
271 return print_job_write(SNUM(fsp->conn), jobid, data, pos, n);
274 if (!fsp->can_write) {
279 if (!fsp->modified) {
281 fsp->modified = True;
283 if (SMB_VFS_FSTAT(fsp, &st) == 0) {
285 trigger_write_time_update(fsp);
286 dosmode = dos_mode(fsp->conn,fsp->fsp_name,&st);
287 if ((lp_store_dos_attributes(SNUM(fsp->conn)) ||
288 MAP_ARCHIVE(fsp->conn)) &&
289 !IS_DOS_ARCHIVE(dosmode)) {
290 file_set_dosmode(fsp->conn,fsp->fsp_name,
297 * If this is the first write and we have an exclusive oplock then setup
301 if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && !wcp) {
302 setup_write_cache(fsp, st.st_size);
309 DO_PROFILE_INC(writecache_total_writes);
310 if (!fsp->oplock_type) {
311 DO_PROFILE_INC(writecache_non_oplock_writes);
316 * If this file is level II oplocked then we need
317 * to grab the shared memory lock and inform all
318 * other files with a level II lock that they need
319 * to flush their read caches. We keep the lock over
320 * the shared memory area whilst doing this.
323 release_level_2_oplocks_on_change(fsp);
326 if (profile_p && profile_p->writecache_total_writes % 500 == 0) {
327 DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u total=%u \
328 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
329 profile_p->writecache_init_writes,
330 profile_p->writecache_abutted_writes,
331 profile_p->writecache_total_writes,
332 profile_p->writecache_non_oplock_writes,
333 profile_p->writecache_allocated_write_caches,
334 profile_p->writecache_num_write_caches,
335 profile_p->writecache_direct_writes,
336 profile_p->writecache_num_perfect_writes,
337 profile_p->writecache_read_hits ));
339 DEBUG(3,("WRITECACHE: Flushes SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
340 profile_p->writecache_flushed_writes[SEEK_FLUSH],
341 profile_p->writecache_flushed_writes[READ_FLUSH],
342 profile_p->writecache_flushed_writes[WRITE_FLUSH],
343 profile_p->writecache_flushed_writes[READRAW_FLUSH],
344 profile_p->writecache_flushed_writes[OPLOCK_RELEASE_FLUSH],
345 profile_p->writecache_flushed_writes[CLOSE_FLUSH],
346 profile_p->writecache_flushed_writes[SYNC_FLUSH] ));
350 if (wcp && req->unread_bytes) {
351 /* If we're using receivefile don't
352 * deal with a write cache.
354 flush_write_cache(fsp, WRITE_FLUSH);
355 delete_write_cache(fsp);
360 DO_PROFILE_INC(writecache_direct_writes);
361 total_written = real_write_file(req, fsp, data, pos, n);
362 return total_written;
365 DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f wcp->data_size=%u\n",
366 fsp->fsp_name, fsp->fh->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size));
368 fsp->fh->pos = pos + n;
371 * If we have active cache and it isn't contiguous then we flush.
372 * NOTE: There is a small problem with running out of disk ....
375 if (wcp->data_size) {
376 bool cache_flush_needed = False;
378 if ((pos >= wcp->offset) && (pos <= wcp->offset + wcp->data_size)) {
380 /* ASCII art.... JRA.
382 +--------------+-----
383 | Cached data | Rest of allocated cache buffer....
384 +--------------+-----
386 +-------------------+
388 +-------------------+
393 * Start of write overlaps or abutts the existing data.
396 size_t data_used = MIN((wcp->alloc_size - (pos - wcp->offset)), n);
398 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
401 * Update the current buffer size with the new data.
404 if(pos + data_used > wcp->offset + wcp->data_size) {
405 wcp->data_size = pos + data_used - wcp->offset;
409 * Update the file size if changed.
412 if (wcp->offset + wcp->data_size > wcp->file_size) {
413 if (wcp_file_size_change(fsp) == -1) {
419 * If we used all the data then
426 cache_flush_needed = True;
429 * Move the start of data forward by the amount used,
430 * cut down the amount left by the same amount.
437 DO_PROFILE_INC(writecache_abutted_writes);
438 total_written = data_used;
442 } else if ((pos < wcp->offset) && (pos + n > wcp->offset) &&
443 (pos + n <= wcp->offset + wcp->alloc_size)) {
445 /* ASCII art.... JRA.
451 +-------------------+
453 +-------------------+
458 * End of write overlaps the existing data.
461 size_t data_used = pos + n - wcp->offset;
463 memcpy(wcp->data, data + n - data_used, data_used);
466 * Update the current buffer size with the new data.
469 if(pos + n > wcp->offset + wcp->data_size) {
470 wcp->data_size = pos + n - wcp->offset;
474 * Update the file size if changed.
477 if (wcp->offset + wcp->data_size > wcp->file_size) {
478 if (wcp_file_size_change(fsp) == -1) {
484 * We don't need to move the start of data, but we
485 * cut down the amount left by the amount used.
491 * We cannot have used all the data here.
494 cache_flush_needed = True;
496 DO_PROFILE_INC(writecache_abutted_writes);
497 total_written = data_used;
501 } else if ( (pos >= wcp->file_size) &&
502 (wcp->offset + wcp->data_size == wcp->file_size) &&
503 (pos > wcp->offset + wcp->data_size) &&
504 (pos < wcp->offset + wcp->alloc_size) ) {
506 /* ASCII art.... JRA.
510 +---------------+---------------+
511 | Cached data | Cache buffer |
512 +---------------+---------------+
514 +-------------------+
516 +-------------------+
521 * Non-contiguous write part of which fits within
522 * the cache buffer and is extending the file
523 * and the cache contents reflect the current
524 * data up to the current end of the file.
529 if(pos + n <= wcp->offset + wcp->alloc_size) {
532 data_used = wcp->offset + wcp->alloc_size - pos;
536 * Fill in the non-continuous area with zeros.
539 memset(wcp->data + wcp->data_size, '\0',
540 pos - (wcp->offset + wcp->data_size) );
542 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
545 * Update the current buffer size with the new data.
548 if(pos + data_used > wcp->offset + wcp->data_size) {
549 wcp->data_size = pos + data_used - wcp->offset;
553 * Update the file size if changed.
556 if (wcp->offset + wcp->data_size > wcp->file_size) {
557 if (wcp_file_size_change(fsp) == -1) {
563 * If we used all the data then
570 cache_flush_needed = True;
574 * Move the start of data forward by the amount used,
575 * cut down the amount left by the same amount.
582 DO_PROFILE_INC(writecache_abutted_writes);
583 total_written = data_used;
587 } else if ( (pos >= wcp->file_size) &&
589 (wcp->file_size == wcp->offset + wcp->data_size) &&
590 (pos < wcp->file_size + wcp->alloc_size)) {
596 +---------------+---------------+
597 | Cached data | Cache buffer |
598 +---------------+---------------+
600 |<------- allocated size ---------------->|
606 MS-Office seems to do this a lot to determine if there's enough
607 space on the filesystem to write a new file.
612 +-----------------------+--------+
613 | Zeroed Cached data | 1 Byte |
614 +-----------------------+--------+
617 flush_write_cache(fsp, WRITE_FLUSH);
618 wcp->offset = wcp->file_size;
619 wcp->data_size = pos - wcp->file_size + 1;
620 memset(wcp->data, '\0', wcp->data_size);
621 memcpy(wcp->data + wcp->data_size-1, data, 1);
624 * Update the file size if changed.
627 if (wcp->offset + wcp->data_size > wcp->file_size) {
628 if (wcp_file_size_change(fsp) == -1) {
637 /* ASCII art..... JRA.
641 +---------------+---------------+
642 | Cached data | Cache buffer |
643 +---------------+---------------+
645 +-------------------+
647 +-------------------+
651 +---------------+---------------+
652 | Cached data | Cache buffer |
653 +---------------+---------------+
655 +-------------------+
657 +-------------------+
661 +---------------+---------------+
662 | Cached data | Cache buffer |
663 +---------------+---------------+
665 +-----------------------------------------------------+
667 +-----------------------------------------------------+
672 * Write is bigger than buffer, or there is no overlap on the
676 DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
677 len = %u\n",fsp->fh->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size ));
680 * If write would fit in the cache, and is larger than
681 * the data already in the cache, flush the cache and
682 * preferentially copy the data new data into it. Otherwise
683 * just write the data directly.
686 if ( n <= wcp->alloc_size && n > wcp->data_size) {
687 cache_flush_needed = True;
689 ssize_t ret = real_write_file(NULL,fsp, data, pos, n);
692 * If the write overlaps the entire cache, then
693 * discard the current contents of the cache.
694 * Fix from Rasmus Borup Hansen rbh@math.ku.dk.
697 if ((pos <= wcp->offset) &&
698 (pos + n >= wcp->offset + wcp->data_size) ) {
699 DEBUG(9,("write_file: discarding overwritten write \
700 cache: fd = %d, off=%.0f, size=%u\n", fsp->fh->fd, (double)wcp->offset, (unsigned int)wcp->data_size ));
704 DO_PROFILE_INC(writecache_direct_writes);
709 if (pos + ret > wcp->file_size) {
710 wcp->file_size = pos + ret;
720 if (cache_flush_needed) {
721 DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
722 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
723 write_path, fsp->fh->fd, (double)wcp->file_size, (double)pos, (unsigned int)n,
724 (double)wcp->offset, (unsigned int)wcp->data_size ));
726 flush_write_cache(fsp, WRITE_FLUSH);
731 * If the write request is bigger than the cache
732 * size, write it all out.
735 if (n > wcp->alloc_size ) {
736 ssize_t ret = real_write_file(NULL,fsp, data, pos, n);
741 if (pos + ret > wcp->file_size) {
742 wcp->file_size = pos + n;
745 DO_PROFILE_INC(writecache_direct_writes);
746 return total_written + n;
750 * If there's any data left, cache it.
755 if (wcp->data_size) {
756 DO_PROFILE_INC(writecache_abutted_writes);
758 DO_PROFILE_INC(writecache_init_writes);
761 memcpy(wcp->data+wcp->data_size, data, n);
762 if (wcp->data_size == 0) {
764 DO_PROFILE_INC(writecache_num_write_caches);
769 * Update the file size if changed.
772 if (wcp->offset + wcp->data_size > wcp->file_size) {
773 if (wcp_file_size_change(fsp) == -1) {
777 DEBUG(9,("wcp->offset = %.0f wcp->data_size = %u cache return %u\n",
778 (double)wcp->offset, (unsigned int)wcp->data_size, (unsigned int)n));
781 return total_written; /* .... that's a write :) */
784 return total_written;
787 /****************************************************************************
788 Delete the write cache structure.
789 ****************************************************************************/
791 void delete_write_cache(files_struct *fsp)
799 if(!(wcp = fsp->wcp)) {
803 DO_PROFILE_DEC(writecache_allocated_write_caches);
804 allocated_write_caches--;
806 SMB_ASSERT(wcp->data_size == 0);
808 SAFE_FREE(wcp->data);
811 DEBUG(10,("delete_write_cache: File %s deleted write cache\n", fsp->fsp_name ));
814 /****************************************************************************
815 Setup the write cache structure.
816 ****************************************************************************/
818 static bool setup_write_cache(files_struct *fsp, SMB_OFF_T file_size)
820 ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
823 if (allocated_write_caches >= MAX_WRITE_CACHES) {
827 if(alloc_size == 0 || fsp->wcp) {
831 if((wcp = SMB_MALLOC_P(write_cache)) == NULL) {
832 DEBUG(0,("setup_write_cache: malloc fail.\n"));
836 wcp->file_size = file_size;
838 wcp->alloc_size = alloc_size;
840 if((wcp->data = (char *)SMB_MALLOC(wcp->alloc_size)) == NULL) {
841 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
842 (unsigned int)wcp->alloc_size ));
847 memset(wcp->data, '\0', wcp->alloc_size );
850 DO_PROFILE_INC(writecache_allocated_write_caches);
851 allocated_write_caches++;
853 DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
854 fsp->fsp_name, (unsigned long)wcp->alloc_size ));
859 /****************************************************************************
860 Cope with a size change.
861 ****************************************************************************/
863 void set_filelen_write_cache(files_struct *fsp, SMB_OFF_T file_size)
866 /* The cache *must* have been flushed before we do this. */
867 if (fsp->wcp->data_size != 0) {
869 if (asprintf(&msg, "set_filelen_write_cache: size change "
870 "on file %s with write cache size = %lu\n",
872 (unsigned long)fsp->wcp->data_size) != -1) {
875 smb_panic("set_filelen_write_cache");
878 fsp->wcp->file_size = file_size;
882 /*******************************************************************
883 Flush a write cache struct to disk.
884 ********************************************************************/
886 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
888 write_cache *wcp = fsp->wcp;
892 if(!wcp || !wcp->data_size) {
896 data_size = wcp->data_size;
899 DO_PROFILE_DEC_INC(writecache_num_write_caches,writecache_flushed_writes[reason]);
901 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
902 fsp->fh->fd, (double)wcp->offset, (unsigned int)data_size));
905 if(data_size == wcp->alloc_size) {
906 DO_PROFILE_INC(writecache_num_perfect_writes);
910 ret = real_write_file(NULL, fsp, wcp->data, wcp->offset, data_size);
913 * Ensure file size if kept up to date if write extends file.
916 if ((ret != -1) && (wcp->offset + ret > wcp->file_size)) {
917 wcp->file_size = wcp->offset + ret;
923 /*******************************************************************
925 ********************************************************************/
927 NTSTATUS sync_file(connection_struct *conn, files_struct *fsp, bool write_through)
929 if (fsp->fh->fd == -1)
930 return NT_STATUS_INVALID_HANDLE;
932 if (lp_strict_sync(SNUM(conn)) &&
933 (lp_syncalways(SNUM(conn)) || write_through)) {
934 int ret = flush_write_cache(fsp, SYNC_FLUSH);
936 return map_nt_error_from_unix(errno);
938 ret = SMB_VFS_FSYNC(fsp);
940 return map_nt_error_from_unix(errno);
946 /************************************************************
947 Perform a stat whether a valid fd or not.
948 ************************************************************/
950 int fsp_stat(files_struct *fsp, SMB_STRUCT_STAT *pst)
952 if (fsp->fh->fd == -1) {
953 return SMB_VFS_STAT(fsp->conn, fsp->fsp_name, pst);
955 return SMB_VFS_FSTAT(fsp, pst);