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->posix_open) {
197 /* Don't use delayed writes on POSIX files. */
201 if (fsp->write_time_forced) {
202 /* No point - "sticky" write times
208 if (fsp->update_write_time_triggered) {
210 * We only update the write time
211 * on the first write. After that
212 * no other writes affect this.
216 fsp->update_write_time_triggered = true;
218 delay = lp_parm_int(SNUM(fsp->conn),
219 "smbd", "writetimeupdatedelay",
220 WRITE_TIME_UPDATE_USEC_DELAY);
222 /* trigger the update 2 seconds later */
223 fsp->update_write_time_on_close = true;
224 fsp->update_write_time_event =
225 event_add_timed(smbd_event_context(), NULL,
226 timeval_current_ofs(0, delay),
227 update_write_time_handler, fsp);
230 void trigger_write_time_update_immediate(struct files_struct *fsp)
232 if (fsp->posix_open) {
233 /* Don't use delayed writes on POSIX files. */
237 if (fsp->write_time_forced) {
239 * No point - "sticky" write times
245 TALLOC_FREE(fsp->update_write_time_event);
246 DEBUG(5, ("Update write time immediate on %s\n", fsp->fsp_name));
248 fsp->update_write_time_triggered = true;
250 fsp->update_write_time_on_close = false;
251 update_write_time(fsp);
254 /****************************************************************************
256 ****************************************************************************/
258 ssize_t write_file(struct smb_request *req,
264 write_cache *wcp = fsp->wcp;
265 ssize_t total_written = 0;
268 if (fsp->print_file) {
271 if (!rap_to_pjobid(fsp->rap_print_jobid, NULL, &jobid)) {
272 DEBUG(3,("write_file: Unable to map RAP jobid %u to jobid.\n",
273 (unsigned int)fsp->rap_print_jobid ));
278 return print_job_write(SNUM(fsp->conn), jobid, data, pos, n);
281 if (!fsp->can_write) {
286 if (!fsp->modified) {
288 fsp->modified = True;
290 if (SMB_VFS_FSTAT(fsp, &st) == 0) {
292 trigger_write_time_update(fsp);
293 dosmode = dos_mode(fsp->conn,fsp->fsp_name,&st);
294 if ((lp_store_dos_attributes(SNUM(fsp->conn)) ||
295 MAP_ARCHIVE(fsp->conn)) &&
296 !IS_DOS_ARCHIVE(dosmode)) {
297 file_set_dosmode(fsp->conn,fsp->fsp_name,
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, st.st_ex_size);
316 DO_PROFILE_INC(writecache_total_writes);
317 if (!fsp->oplock_type) {
318 DO_PROFILE_INC(writecache_non_oplock_writes);
323 * If this file is level II oplocked then we need
324 * to grab the shared memory lock and inform all
325 * other files with a level II lock that they need
326 * to flush their read caches. We keep the lock over
327 * the shared memory area whilst doing this.
330 /* This should actually be improved to span the write. */
331 contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WRITE);
332 contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WRITE);
335 if (profile_p && profile_p->writecache_total_writes % 500 == 0) {
336 DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u total=%u \
337 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
338 profile_p->writecache_init_writes,
339 profile_p->writecache_abutted_writes,
340 profile_p->writecache_total_writes,
341 profile_p->writecache_non_oplock_writes,
342 profile_p->writecache_allocated_write_caches,
343 profile_p->writecache_num_write_caches,
344 profile_p->writecache_direct_writes,
345 profile_p->writecache_num_perfect_writes,
346 profile_p->writecache_read_hits ));
348 DEBUG(3,("WRITECACHE: Flushes SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
349 profile_p->writecache_flushed_writes[SEEK_FLUSH],
350 profile_p->writecache_flushed_writes[READ_FLUSH],
351 profile_p->writecache_flushed_writes[WRITE_FLUSH],
352 profile_p->writecache_flushed_writes[READRAW_FLUSH],
353 profile_p->writecache_flushed_writes[OPLOCK_RELEASE_FLUSH],
354 profile_p->writecache_flushed_writes[CLOSE_FLUSH],
355 profile_p->writecache_flushed_writes[SYNC_FLUSH] ));
359 if (wcp && req->unread_bytes) {
360 /* If we're using receivefile don't
361 * deal with a write cache.
363 flush_write_cache(fsp, WRITE_FLUSH);
364 delete_write_cache(fsp);
369 DO_PROFILE_INC(writecache_direct_writes);
370 total_written = real_write_file(req, fsp, data, pos, n);
371 return total_written;
374 DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f wcp->data_size=%u\n",
375 fsp->fsp_name, fsp->fh->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size));
377 fsp->fh->pos = pos + n;
380 * If we have active cache and it isn't contiguous then we flush.
381 * NOTE: There is a small problem with running out of disk ....
384 if (wcp->data_size) {
385 bool cache_flush_needed = False;
387 if ((pos >= wcp->offset) && (pos <= wcp->offset + wcp->data_size)) {
389 /* ASCII art.... JRA.
391 +--------------+-----
392 | Cached data | Rest of allocated cache buffer....
393 +--------------+-----
395 +-------------------+
397 +-------------------+
402 * Start of write overlaps or abutts the existing data.
405 size_t data_used = MIN((wcp->alloc_size - (pos - wcp->offset)), n);
407 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
410 * Update the current buffer size with the new data.
413 if(pos + data_used > wcp->offset + wcp->data_size) {
414 wcp->data_size = pos + data_used - wcp->offset;
418 * Update the file size if changed.
421 if (wcp->offset + wcp->data_size > wcp->file_size) {
422 if (wcp_file_size_change(fsp) == -1) {
428 * If we used all the data then
435 cache_flush_needed = True;
438 * Move the start of data forward by the amount used,
439 * cut down the amount left by the same amount.
446 DO_PROFILE_INC(writecache_abutted_writes);
447 total_written = data_used;
451 } else if ((pos < wcp->offset) && (pos + n > wcp->offset) &&
452 (pos + n <= wcp->offset + wcp->alloc_size)) {
454 /* ASCII art.... JRA.
460 +-------------------+
462 +-------------------+
467 * End of write overlaps the existing data.
470 size_t data_used = pos + n - wcp->offset;
472 memcpy(wcp->data, data + n - data_used, data_used);
475 * Update the current buffer size with the new data.
478 if(pos + n > wcp->offset + wcp->data_size) {
479 wcp->data_size = pos + n - wcp->offset;
483 * Update the file size if changed.
486 if (wcp->offset + wcp->data_size > wcp->file_size) {
487 if (wcp_file_size_change(fsp) == -1) {
493 * We don't need to move the start of data, but we
494 * cut down the amount left by the amount used.
500 * We cannot have used all the data here.
503 cache_flush_needed = True;
505 DO_PROFILE_INC(writecache_abutted_writes);
506 total_written = data_used;
510 } else if ( (pos >= wcp->file_size) &&
511 (wcp->offset + wcp->data_size == wcp->file_size) &&
512 (pos > wcp->offset + wcp->data_size) &&
513 (pos < wcp->offset + wcp->alloc_size) ) {
515 /* ASCII art.... JRA.
519 +---------------+---------------+
520 | Cached data | Cache buffer |
521 +---------------+---------------+
523 +-------------------+
525 +-------------------+
530 * Non-contiguous write part of which fits within
531 * the cache buffer and is extending the file
532 * and the cache contents reflect the current
533 * data up to the current end of the file.
538 if(pos + n <= wcp->offset + wcp->alloc_size) {
541 data_used = wcp->offset + wcp->alloc_size - pos;
545 * Fill in the non-continuous area with zeros.
548 memset(wcp->data + wcp->data_size, '\0',
549 pos - (wcp->offset + wcp->data_size) );
551 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
554 * Update the current buffer size with the new data.
557 if(pos + data_used > wcp->offset + wcp->data_size) {
558 wcp->data_size = pos + data_used - wcp->offset;
562 * Update the file size if changed.
565 if (wcp->offset + wcp->data_size > wcp->file_size) {
566 if (wcp_file_size_change(fsp) == -1) {
572 * If we used all the data then
579 cache_flush_needed = True;
583 * Move the start of data forward by the amount used,
584 * cut down the amount left by the same amount.
591 DO_PROFILE_INC(writecache_abutted_writes);
592 total_written = data_used;
596 } else if ( (pos >= wcp->file_size) &&
598 (wcp->file_size == wcp->offset + wcp->data_size) &&
599 (pos < wcp->file_size + wcp->alloc_size)) {
605 +---------------+---------------+
606 | Cached data | Cache buffer |
607 +---------------+---------------+
609 |<------- allocated size ---------------->|
615 MS-Office seems to do this a lot to determine if there's enough
616 space on the filesystem to write a new file.
621 +-----------------------+--------+
622 | Zeroed Cached data | 1 Byte |
623 +-----------------------+--------+
626 flush_write_cache(fsp, WRITE_FLUSH);
627 wcp->offset = wcp->file_size;
628 wcp->data_size = pos - wcp->file_size + 1;
629 memset(wcp->data, '\0', wcp->data_size);
630 memcpy(wcp->data + wcp->data_size-1, data, 1);
633 * Update the file size if changed.
636 if (wcp->offset + wcp->data_size > wcp->file_size) {
637 if (wcp_file_size_change(fsp) == -1) {
646 /* ASCII art..... JRA.
650 +---------------+---------------+
651 | Cached data | Cache buffer |
652 +---------------+---------------+
654 +-------------------+
656 +-------------------+
660 +---------------+---------------+
661 | Cached data | Cache buffer |
662 +---------------+---------------+
664 +-------------------+
666 +-------------------+
670 +---------------+---------------+
671 | Cached data | Cache buffer |
672 +---------------+---------------+
674 +-----------------------------------------------------+
676 +-----------------------------------------------------+
681 * Write is bigger than buffer, or there is no overlap on the
685 DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
686 len = %u\n",fsp->fh->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size ));
689 * If write would fit in the cache, and is larger than
690 * the data already in the cache, flush the cache and
691 * preferentially copy the data new data into it. Otherwise
692 * just write the data directly.
695 if ( n <= wcp->alloc_size && n > wcp->data_size) {
696 cache_flush_needed = True;
698 ssize_t ret = real_write_file(NULL,fsp, data, pos, n);
701 * If the write overlaps the entire cache, then
702 * discard the current contents of the cache.
703 * Fix from Rasmus Borup Hansen rbh@math.ku.dk.
706 if ((pos <= wcp->offset) &&
707 (pos + n >= wcp->offset + wcp->data_size) ) {
708 DEBUG(9,("write_file: discarding overwritten write \
709 cache: fd = %d, off=%.0f, size=%u\n", fsp->fh->fd, (double)wcp->offset, (unsigned int)wcp->data_size ));
713 DO_PROFILE_INC(writecache_direct_writes);
718 if (pos + ret > wcp->file_size) {
719 wcp->file_size = pos + ret;
729 if (cache_flush_needed) {
730 DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
731 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
732 write_path, fsp->fh->fd, (double)wcp->file_size, (double)pos, (unsigned int)n,
733 (double)wcp->offset, (unsigned int)wcp->data_size ));
735 flush_write_cache(fsp, WRITE_FLUSH);
740 * If the write request is bigger than the cache
741 * size, write it all out.
744 if (n > wcp->alloc_size ) {
745 ssize_t ret = real_write_file(NULL,fsp, data, pos, n);
750 if (pos + ret > wcp->file_size) {
751 wcp->file_size = pos + n;
754 DO_PROFILE_INC(writecache_direct_writes);
755 return total_written + n;
759 * If there's any data left, cache it.
764 if (wcp->data_size) {
765 DO_PROFILE_INC(writecache_abutted_writes);
767 DO_PROFILE_INC(writecache_init_writes);
770 memcpy(wcp->data+wcp->data_size, data, n);
771 if (wcp->data_size == 0) {
773 DO_PROFILE_INC(writecache_num_write_caches);
778 * Update the file size if changed.
781 if (wcp->offset + wcp->data_size > wcp->file_size) {
782 if (wcp_file_size_change(fsp) == -1) {
786 DEBUG(9,("wcp->offset = %.0f wcp->data_size = %u cache return %u\n",
787 (double)wcp->offset, (unsigned int)wcp->data_size, (unsigned int)n));
790 return total_written; /* .... that's a write :) */
793 return total_written;
796 /****************************************************************************
797 Delete the write cache structure.
798 ****************************************************************************/
800 void delete_write_cache(files_struct *fsp)
808 if(!(wcp = fsp->wcp)) {
812 DO_PROFILE_DEC(writecache_allocated_write_caches);
813 allocated_write_caches--;
815 SMB_ASSERT(wcp->data_size == 0);
817 SAFE_FREE(wcp->data);
820 DEBUG(10,("delete_write_cache: File %s deleted write cache\n", fsp->fsp_name ));
823 /****************************************************************************
824 Setup the write cache structure.
825 ****************************************************************************/
827 static bool setup_write_cache(files_struct *fsp, SMB_OFF_T file_size)
829 ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
832 if (allocated_write_caches >= MAX_WRITE_CACHES) {
836 if(alloc_size == 0 || fsp->wcp) {
840 if((wcp = SMB_MALLOC_P(write_cache)) == NULL) {
841 DEBUG(0,("setup_write_cache: malloc fail.\n"));
845 wcp->file_size = file_size;
847 wcp->alloc_size = alloc_size;
849 if((wcp->data = (char *)SMB_MALLOC(wcp->alloc_size)) == NULL) {
850 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
851 (unsigned int)wcp->alloc_size ));
856 memset(wcp->data, '\0', wcp->alloc_size );
859 DO_PROFILE_INC(writecache_allocated_write_caches);
860 allocated_write_caches++;
862 DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
863 fsp->fsp_name, (unsigned long)wcp->alloc_size ));
868 /****************************************************************************
869 Cope with a size change.
870 ****************************************************************************/
872 void set_filelen_write_cache(files_struct *fsp, SMB_OFF_T file_size)
875 /* The cache *must* have been flushed before we do this. */
876 if (fsp->wcp->data_size != 0) {
878 if (asprintf(&msg, "set_filelen_write_cache: size change "
879 "on file %s with write cache size = %lu\n",
881 (unsigned long)fsp->wcp->data_size) != -1) {
884 smb_panic("set_filelen_write_cache");
887 fsp->wcp->file_size = file_size;
891 /*******************************************************************
892 Flush a write cache struct to disk.
893 ********************************************************************/
895 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
897 write_cache *wcp = fsp->wcp;
901 if(!wcp || !wcp->data_size) {
905 data_size = wcp->data_size;
908 DO_PROFILE_DEC_INC(writecache_num_write_caches,writecache_flushed_writes[reason]);
910 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
911 fsp->fh->fd, (double)wcp->offset, (unsigned int)data_size));
914 if(data_size == wcp->alloc_size) {
915 DO_PROFILE_INC(writecache_num_perfect_writes);
919 ret = real_write_file(NULL, fsp, wcp->data, wcp->offset, data_size);
922 * Ensure file size if kept up to date if write extends file.
925 if ((ret != -1) && (wcp->offset + ret > wcp->file_size)) {
926 wcp->file_size = wcp->offset + ret;
932 /*******************************************************************
934 ********************************************************************/
936 NTSTATUS sync_file(connection_struct *conn, files_struct *fsp, bool write_through)
938 if (fsp->fh->fd == -1)
939 return NT_STATUS_INVALID_HANDLE;
941 if (lp_strict_sync(SNUM(conn)) &&
942 (lp_syncalways(SNUM(conn)) || write_through)) {
943 int ret = flush_write_cache(fsp, SYNC_FLUSH);
945 return map_nt_error_from_unix(errno);
947 ret = SMB_VFS_FSYNC(fsp);
949 return map_nt_error_from_unix(errno);
955 /************************************************************
956 Perform a stat whether a valid fd or not.
957 ************************************************************/
959 int fsp_stat(files_struct *fsp, SMB_STRUCT_STAT *pst)
961 if (fsp->fh->fd == -1) {
962 return vfs_stat_smb_fname(fsp->conn, fsp->fsp_name, pst);
964 return SMB_VFS_FSTAT(fsp, pst);