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->write_time_forced) {
197 /* No point - "sticky" write times
203 if (fsp->update_write_time_triggered) {
205 * We only update the write time
206 * on the first write. After that
207 * no other writes affect this.
211 fsp->update_write_time_triggered = true;
213 delay = lp_parm_int(SNUM(fsp->conn),
214 "smbd", "writetimeupdatedelay",
215 WRITE_TIME_UPDATE_USEC_DELAY);
217 /* trigger the update 2 seconds later */
218 fsp->update_write_time_on_close = true;
219 fsp->update_write_time_event =
220 event_add_timed(smbd_event_context(), NULL,
221 timeval_current_ofs(0, delay),
222 update_write_time_handler, fsp);
225 void trigger_write_time_update_immediate(struct files_struct *fsp)
227 if (fsp->write_time_forced) {
229 * No point - "sticky" write times
235 TALLOC_FREE(fsp->update_write_time_event);
236 DEBUG(5, ("Update write time immediate on %s\n", fsp->fsp_name));
238 fsp->update_write_time_triggered = true;
240 fsp->update_write_time_on_close = false;
241 update_write_time(fsp);
244 /****************************************************************************
246 ****************************************************************************/
248 ssize_t write_file(struct smb_request *req,
254 write_cache *wcp = fsp->wcp;
255 ssize_t total_written = 0;
258 if (fsp->print_file) {
261 if (!rap_to_pjobid(fsp->rap_print_jobid, NULL, &jobid)) {
262 DEBUG(3,("write_file: Unable to map RAP jobid %u to jobid.\n",
263 (unsigned int)fsp->rap_print_jobid ));
268 return print_job_write(SNUM(fsp->conn), jobid, data, pos, n);
271 if (!fsp->can_write) {
276 if (!fsp->modified) {
278 fsp->modified = True;
280 if (SMB_VFS_FSTAT(fsp, &st) == 0) {
282 trigger_write_time_update(fsp);
283 dosmode = dos_mode(fsp->conn,fsp->fsp_name,&st);
284 if ((lp_store_dos_attributes(SNUM(fsp->conn)) ||
285 MAP_ARCHIVE(fsp->conn)) &&
286 !IS_DOS_ARCHIVE(dosmode)) {
287 file_set_dosmode(fsp->conn,fsp->fsp_name,
294 * If this is the first write and we have an exclusive oplock then setup
298 if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && !wcp) {
299 setup_write_cache(fsp, st.st_size);
306 DO_PROFILE_INC(writecache_total_writes);
307 if (!fsp->oplock_type) {
308 DO_PROFILE_INC(writecache_non_oplock_writes);
313 * If this file is level II oplocked then we need
314 * to grab the shared memory lock and inform all
315 * other files with a level II lock that they need
316 * to flush their read caches. We keep the lock over
317 * the shared memory area whilst doing this.
320 /* This should actually be improved to span the write. */
321 contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WRITE);
322 contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WRITE);
325 if (profile_p && profile_p->writecache_total_writes % 500 == 0) {
326 DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u total=%u \
327 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
328 profile_p->writecache_init_writes,
329 profile_p->writecache_abutted_writes,
330 profile_p->writecache_total_writes,
331 profile_p->writecache_non_oplock_writes,
332 profile_p->writecache_allocated_write_caches,
333 profile_p->writecache_num_write_caches,
334 profile_p->writecache_direct_writes,
335 profile_p->writecache_num_perfect_writes,
336 profile_p->writecache_read_hits ));
338 DEBUG(3,("WRITECACHE: Flushes SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
339 profile_p->writecache_flushed_writes[SEEK_FLUSH],
340 profile_p->writecache_flushed_writes[READ_FLUSH],
341 profile_p->writecache_flushed_writes[WRITE_FLUSH],
342 profile_p->writecache_flushed_writes[READRAW_FLUSH],
343 profile_p->writecache_flushed_writes[OPLOCK_RELEASE_FLUSH],
344 profile_p->writecache_flushed_writes[CLOSE_FLUSH],
345 profile_p->writecache_flushed_writes[SYNC_FLUSH] ));
349 if (wcp && req->unread_bytes) {
350 /* If we're using receivefile don't
351 * deal with a write cache.
353 flush_write_cache(fsp, WRITE_FLUSH);
354 delete_write_cache(fsp);
359 DO_PROFILE_INC(writecache_direct_writes);
360 total_written = real_write_file(req, fsp, data, pos, n);
361 return total_written;
364 DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f wcp->data_size=%u\n",
365 fsp->fsp_name, fsp->fh->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size));
367 fsp->fh->pos = pos + n;
370 * If we have active cache and it isn't contiguous then we flush.
371 * NOTE: There is a small problem with running out of disk ....
374 if (wcp->data_size) {
375 bool cache_flush_needed = False;
377 if ((pos >= wcp->offset) && (pos <= wcp->offset + wcp->data_size)) {
379 /* ASCII art.... JRA.
381 +--------------+-----
382 | Cached data | Rest of allocated cache buffer....
383 +--------------+-----
385 +-------------------+
387 +-------------------+
392 * Start of write overlaps or abutts the existing data.
395 size_t data_used = MIN((wcp->alloc_size - (pos - wcp->offset)), n);
397 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
400 * Update the current buffer size with the new data.
403 if(pos + data_used > wcp->offset + wcp->data_size) {
404 wcp->data_size = pos + data_used - wcp->offset;
408 * Update the file size if changed.
411 if (wcp->offset + wcp->data_size > wcp->file_size) {
412 if (wcp_file_size_change(fsp) == -1) {
418 * If we used all the data then
425 cache_flush_needed = True;
428 * Move the start of data forward by the amount used,
429 * cut down the amount left by the same amount.
436 DO_PROFILE_INC(writecache_abutted_writes);
437 total_written = data_used;
441 } else if ((pos < wcp->offset) && (pos + n > wcp->offset) &&
442 (pos + n <= wcp->offset + wcp->alloc_size)) {
444 /* ASCII art.... JRA.
450 +-------------------+
452 +-------------------+
457 * End of write overlaps the existing data.
460 size_t data_used = pos + n - wcp->offset;
462 memcpy(wcp->data, data + n - data_used, data_used);
465 * Update the current buffer size with the new data.
468 if(pos + n > wcp->offset + wcp->data_size) {
469 wcp->data_size = pos + n - wcp->offset;
473 * Update the file size if changed.
476 if (wcp->offset + wcp->data_size > wcp->file_size) {
477 if (wcp_file_size_change(fsp) == -1) {
483 * We don't need to move the start of data, but we
484 * cut down the amount left by the amount used.
490 * We cannot have used all the data here.
493 cache_flush_needed = True;
495 DO_PROFILE_INC(writecache_abutted_writes);
496 total_written = data_used;
500 } else if ( (pos >= wcp->file_size) &&
501 (wcp->offset + wcp->data_size == wcp->file_size) &&
502 (pos > wcp->offset + wcp->data_size) &&
503 (pos < wcp->offset + wcp->alloc_size) ) {
505 /* ASCII art.... JRA.
509 +---------------+---------------+
510 | Cached data | Cache buffer |
511 +---------------+---------------+
513 +-------------------+
515 +-------------------+
520 * Non-contiguous write part of which fits within
521 * the cache buffer and is extending the file
522 * and the cache contents reflect the current
523 * data up to the current end of the file.
528 if(pos + n <= wcp->offset + wcp->alloc_size) {
531 data_used = wcp->offset + wcp->alloc_size - pos;
535 * Fill in the non-continuous area with zeros.
538 memset(wcp->data + wcp->data_size, '\0',
539 pos - (wcp->offset + wcp->data_size) );
541 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
544 * Update the current buffer size with the new data.
547 if(pos + data_used > wcp->offset + wcp->data_size) {
548 wcp->data_size = pos + data_used - wcp->offset;
552 * Update the file size if changed.
555 if (wcp->offset + wcp->data_size > wcp->file_size) {
556 if (wcp_file_size_change(fsp) == -1) {
562 * If we used all the data then
569 cache_flush_needed = True;
573 * Move the start of data forward by the amount used,
574 * cut down the amount left by the same amount.
581 DO_PROFILE_INC(writecache_abutted_writes);
582 total_written = data_used;
586 } else if ( (pos >= wcp->file_size) &&
588 (wcp->file_size == wcp->offset + wcp->data_size) &&
589 (pos < wcp->file_size + wcp->alloc_size)) {
595 +---------------+---------------+
596 | Cached data | Cache buffer |
597 +---------------+---------------+
599 |<------- allocated size ---------------->|
605 MS-Office seems to do this a lot to determine if there's enough
606 space on the filesystem to write a new file.
611 +-----------------------+--------+
612 | Zeroed Cached data | 1 Byte |
613 +-----------------------+--------+
616 flush_write_cache(fsp, WRITE_FLUSH);
617 wcp->offset = wcp->file_size;
618 wcp->data_size = pos - wcp->file_size + 1;
619 memset(wcp->data, '\0', wcp->data_size);
620 memcpy(wcp->data + wcp->data_size-1, data, 1);
623 * Update the file size if changed.
626 if (wcp->offset + wcp->data_size > wcp->file_size) {
627 if (wcp_file_size_change(fsp) == -1) {
636 /* ASCII art..... JRA.
640 +---------------+---------------+
641 | Cached data | Cache buffer |
642 +---------------+---------------+
644 +-------------------+
646 +-------------------+
650 +---------------+---------------+
651 | Cached data | Cache buffer |
652 +---------------+---------------+
654 +-------------------+
656 +-------------------+
660 +---------------+---------------+
661 | Cached data | Cache buffer |
662 +---------------+---------------+
664 +-----------------------------------------------------+
666 +-----------------------------------------------------+
671 * Write is bigger than buffer, or there is no overlap on the
675 DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
676 len = %u\n",fsp->fh->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size ));
679 * If write would fit in the cache, and is larger than
680 * the data already in the cache, flush the cache and
681 * preferentially copy the data new data into it. Otherwise
682 * just write the data directly.
685 if ( n <= wcp->alloc_size && n > wcp->data_size) {
686 cache_flush_needed = True;
688 ssize_t ret = real_write_file(NULL,fsp, data, pos, n);
691 * If the write overlaps the entire cache, then
692 * discard the current contents of the cache.
693 * Fix from Rasmus Borup Hansen rbh@math.ku.dk.
696 if ((pos <= wcp->offset) &&
697 (pos + n >= wcp->offset + wcp->data_size) ) {
698 DEBUG(9,("write_file: discarding overwritten write \
699 cache: fd = %d, off=%.0f, size=%u\n", fsp->fh->fd, (double)wcp->offset, (unsigned int)wcp->data_size ));
703 DO_PROFILE_INC(writecache_direct_writes);
708 if (pos + ret > wcp->file_size) {
709 wcp->file_size = pos + ret;
719 if (cache_flush_needed) {
720 DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
721 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
722 write_path, fsp->fh->fd, (double)wcp->file_size, (double)pos, (unsigned int)n,
723 (double)wcp->offset, (unsigned int)wcp->data_size ));
725 flush_write_cache(fsp, WRITE_FLUSH);
730 * If the write request is bigger than the cache
731 * size, write it all out.
734 if (n > wcp->alloc_size ) {
735 ssize_t ret = real_write_file(NULL,fsp, data, pos, n);
740 if (pos + ret > wcp->file_size) {
741 wcp->file_size = pos + n;
744 DO_PROFILE_INC(writecache_direct_writes);
745 return total_written + n;
749 * If there's any data left, cache it.
754 if (wcp->data_size) {
755 DO_PROFILE_INC(writecache_abutted_writes);
757 DO_PROFILE_INC(writecache_init_writes);
760 memcpy(wcp->data+wcp->data_size, data, n);
761 if (wcp->data_size == 0) {
763 DO_PROFILE_INC(writecache_num_write_caches);
768 * Update the file size if changed.
771 if (wcp->offset + wcp->data_size > wcp->file_size) {
772 if (wcp_file_size_change(fsp) == -1) {
776 DEBUG(9,("wcp->offset = %.0f wcp->data_size = %u cache return %u\n",
777 (double)wcp->offset, (unsigned int)wcp->data_size, (unsigned int)n));
780 return total_written; /* .... that's a write :) */
783 return total_written;
786 /****************************************************************************
787 Delete the write cache structure.
788 ****************************************************************************/
790 void delete_write_cache(files_struct *fsp)
798 if(!(wcp = fsp->wcp)) {
802 DO_PROFILE_DEC(writecache_allocated_write_caches);
803 allocated_write_caches--;
805 SMB_ASSERT(wcp->data_size == 0);
807 SAFE_FREE(wcp->data);
810 DEBUG(10,("delete_write_cache: File %s deleted write cache\n", fsp->fsp_name ));
813 /****************************************************************************
814 Setup the write cache structure.
815 ****************************************************************************/
817 static bool setup_write_cache(files_struct *fsp, SMB_OFF_T file_size)
819 ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
822 if (allocated_write_caches >= MAX_WRITE_CACHES) {
826 if(alloc_size == 0 || fsp->wcp) {
830 if((wcp = SMB_MALLOC_P(write_cache)) == NULL) {
831 DEBUG(0,("setup_write_cache: malloc fail.\n"));
835 wcp->file_size = file_size;
837 wcp->alloc_size = alloc_size;
839 if((wcp->data = (char *)SMB_MALLOC(wcp->alloc_size)) == NULL) {
840 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
841 (unsigned int)wcp->alloc_size ));
846 memset(wcp->data, '\0', wcp->alloc_size );
849 DO_PROFILE_INC(writecache_allocated_write_caches);
850 allocated_write_caches++;
852 DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
853 fsp->fsp_name, (unsigned long)wcp->alloc_size ));
858 /****************************************************************************
859 Cope with a size change.
860 ****************************************************************************/
862 void set_filelen_write_cache(files_struct *fsp, SMB_OFF_T file_size)
865 /* The cache *must* have been flushed before we do this. */
866 if (fsp->wcp->data_size != 0) {
868 if (asprintf(&msg, "set_filelen_write_cache: size change "
869 "on file %s with write cache size = %lu\n",
871 (unsigned long)fsp->wcp->data_size) != -1) {
874 smb_panic("set_filelen_write_cache");
877 fsp->wcp->file_size = file_size;
881 /*******************************************************************
882 Flush a write cache struct to disk.
883 ********************************************************************/
885 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
887 write_cache *wcp = fsp->wcp;
891 if(!wcp || !wcp->data_size) {
895 data_size = wcp->data_size;
898 DO_PROFILE_DEC_INC(writecache_num_write_caches,writecache_flushed_writes[reason]);
900 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
901 fsp->fh->fd, (double)wcp->offset, (unsigned int)data_size));
904 if(data_size == wcp->alloc_size) {
905 DO_PROFILE_INC(writecache_num_perfect_writes);
909 ret = real_write_file(NULL, fsp, wcp->data, wcp->offset, data_size);
912 * Ensure file size if kept up to date if write extends file.
915 if ((ret != -1) && (wcp->offset + ret > wcp->file_size)) {
916 wcp->file_size = wcp->offset + ret;
922 /*******************************************************************
924 ********************************************************************/
926 NTSTATUS sync_file(connection_struct *conn, files_struct *fsp, bool write_through)
928 if (fsp->fh->fd == -1)
929 return NT_STATUS_INVALID_HANDLE;
931 if (lp_strict_sync(SNUM(conn)) &&
932 (lp_syncalways(SNUM(conn)) || write_through)) {
933 int ret = flush_write_cache(fsp, SYNC_FLUSH);
935 return map_nt_error_from_unix(errno);
937 ret = SMB_VFS_FSYNC(fsp);
939 return map_nt_error_from_unix(errno);
945 /************************************************************
946 Perform a stat whether a valid fd or not.
947 ************************************************************/
949 int fsp_stat(files_struct *fsp, SMB_STRUCT_STAT *pst)
951 if (fsp->fh->fd == -1) {
952 return SMB_VFS_STAT(fsp->conn, fsp->fsp_name, pst);
954 return SMB_VFS_FSTAT(fsp, pst);