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 #include "smbd/smbd.h"
25 #include "smbd/globals.h"
26 #include "smbprofile.h"
36 static bool setup_write_cache(files_struct *, off_t);
38 /****************************************************************************
39 Read from write cache if we can.
40 ****************************************************************************/
42 static bool read_from_write_cache(files_struct *fsp,char *data,off_t pos,size_t n)
44 struct write_cache *wcp = fsp->wcp;
50 if( n > wcp->data_size || pos < wcp->offset || pos + n > wcp->offset + wcp->data_size) {
54 memcpy(data, wcp->data + (pos - wcp->offset), n);
56 DO_PROFILE_INC(writecache_cached_reads);
61 /****************************************************************************
63 ****************************************************************************/
65 ssize_t read_file(files_struct *fsp,char *data,off_t pos,size_t n)
69 /* you can't read from print files */
70 if (fsp->print_file) {
76 * Serve from write cache if we can.
79 if(read_from_write_cache(fsp, data, pos, n)) {
80 fsp->fh->pos = pos + n;
81 fsp->fh->position_information = fsp->fh->pos;
85 flush_write_cache(fsp, SAMBA_READ_FLUSH);
90 ret = SMB_VFS_PREAD(fsp,data,n,pos);
97 DEBUG(10,("read_file (%s): pos = %.0f, size = %lu, returned %lu\n",
98 fsp_str_dbg(fsp), (double)pos, (unsigned long)n, (long)ret));
101 fsp->fh->position_information = fsp->fh->pos;
106 /****************************************************************************
107 *Really* write to a file.
108 ****************************************************************************/
110 static ssize_t real_write_file(struct smb_request *req,
119 if (pos && lp_strict_allocate(SNUM(fsp->conn) &&
121 if (vfs_fill_sparse(fsp, pos) == -1) {
125 ret = vfs_pwrite_data(req, fsp, data, n, pos);
127 DEBUG(10,("real_write_file (%s): pos = %.0f, size = %lu, returned %ld\n",
128 fsp_str_dbg(fsp), (double)pos, (unsigned long)n, (long)ret));
133 /* Yes - this is correct - writes don't update this. JRA. */
134 /* Found by Samba4 tests. */
136 fsp->position_information = fsp->pos;
143 /****************************************************************************
144 File size cache change.
145 Updates size on disk but doesn't flush the cache.
146 ****************************************************************************/
148 static int wcp_file_size_change(files_struct *fsp)
151 struct write_cache *wcp = fsp->wcp;
153 wcp->file_size = wcp->offset + wcp->data_size;
154 ret = SMB_VFS_FTRUNCATE(fsp, wcp->file_size);
156 DEBUG(0,("wcp_file_size_change (%s): ftruncate of size %.0f "
157 "error %s\n", fsp_str_dbg(fsp),
158 (double)wcp->file_size, strerror(errno)));
163 void fsp_flush_write_time_update(struct files_struct *fsp)
166 * Note this won't expect any impersonation!
167 * So don't call any SMB_VFS operations here!
170 DEBUG(5, ("Update write time on %s\n", fsp_str_dbg(fsp)));
172 /* change the write time in the open file db. */
173 (void)set_write_time(fsp->file_id, timespec_current());
176 notify_fname(fsp->conn, NOTIFY_ACTION_MODIFIED,
177 FILE_NOTIFY_CHANGE_LAST_WRITE, fsp->fsp_name->base_name);
179 /* Remove the timed event handler. */
180 TALLOC_FREE(fsp->update_write_time_event);
183 static void update_write_time_handler(struct tevent_context *ctx,
184 struct tevent_timer *te,
188 files_struct *fsp = (files_struct *)private_data;
189 fsp_flush_write_time_update(fsp);
192 /*********************************************************
193 Schedule a write time update for WRITE_TIME_UPDATE_USEC_DELAY
195 *********************************************************/
197 void trigger_write_time_update(struct files_struct *fsp)
201 if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
202 /* Don't use delayed writes on POSIX files. */
206 if (fsp->write_time_forced) {
207 /* No point - "sticky" write times
213 /* We need to remember someone did a write
214 * and update to current time on close. */
216 fsp->update_write_time_on_close = true;
218 if (fsp->update_write_time_triggered) {
220 * We only update the write time after 2 seconds
221 * on the first normal write. After that
222 * no other writes affect this until close.
226 fsp->update_write_time_triggered = true;
228 delay = lp_parm_int(SNUM(fsp->conn),
229 "smbd", "writetimeupdatedelay",
230 WRITE_TIME_UPDATE_USEC_DELAY);
232 DEBUG(5, ("Update write time %d usec later on %s\n",
233 delay, fsp_str_dbg(fsp)));
236 * trigger the update 2 seconds later
238 * Note that update_write_time_handler()
239 * => fsp_flush_write_time_update()
240 * won't do any SMB_VFS calls and don't
241 * need impersonation. So we use the
242 * raw event context for this.
244 fsp->update_write_time_event =
245 tevent_add_timer(fsp->conn->sconn->raw_ev_ctx, NULL,
246 timeval_current_ofs_usec(delay),
247 update_write_time_handler, fsp);
250 void trigger_write_time_update_immediate(struct files_struct *fsp)
252 struct smb_file_time ft;
254 if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
255 /* Don't use delayed writes on POSIX files. */
259 if (fsp->write_time_forced) {
261 * No point - "sticky" write times
267 TALLOC_FREE(fsp->update_write_time_event);
268 DEBUG(5, ("Update write time immediate on %s\n",
271 /* After an immediate update, reset the trigger. */
272 fsp->update_write_time_triggered = true;
273 fsp->update_write_time_on_close = false;
276 ft.mtime = timespec_current();
278 /* Update the time in the open file db. */
279 (void)set_write_time(fsp->file_id, ft.mtime);
281 /* Now set on disk - takes care of notify. */
282 (void)smb_set_file_time(fsp->conn, fsp, fsp->fsp_name, &ft, false);
285 void mark_file_modified(files_struct *fsp)
293 fsp->modified = true;
295 if (SMB_VFS_FSTAT(fsp, &fsp->fsp_name->st) != 0) {
298 trigger_write_time_update(fsp);
300 if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
303 if (!(lp_store_dos_attributes(SNUM(fsp->conn)) ||
304 MAP_ARCHIVE(fsp->conn))) {
308 dosmode = dos_mode(fsp->conn, fsp->fsp_name);
309 if (IS_DOS_ARCHIVE(dosmode)) {
312 file_set_dosmode(fsp->conn, fsp->fsp_name,
313 dosmode | FILE_ATTRIBUTE_ARCHIVE, NULL, false);
316 /****************************************************************************
318 ****************************************************************************/
320 ssize_t write_file(struct smb_request *req,
326 struct write_cache *wcp = fsp->wcp;
327 ssize_t total_written = 0;
330 if (fsp->print_file) {
334 ret = print_spool_write(fsp, data, n, pos, &t);
342 if (!fsp->can_write) {
348 * If this is the first write and we have an exclusive oplock
349 * then setup the write cache.
352 if (!fsp->modified &&
353 EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) &&
356 * Note: no write cache with leases!
357 * as the handles would have to share the write cache
358 * that's possible but an improvement for another day...
360 setup_write_cache(fsp, fsp->fsp_name->st.st_ex_size);
364 mark_file_modified(fsp);
366 DO_PROFILE_INC(writecache_total_writes);
367 if (!fsp->oplock_type) {
368 DO_PROFILE_INC(writecache_non_oplock_writes);
372 * If this file is level II oplocked then we need
373 * to grab the shared memory lock and inform all
374 * other files with a level II lock that they need
375 * to flush their read caches. We keep the lock over
376 * the shared memory area whilst doing this.
379 /* This should actually be improved to span the write. */
380 contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WRITE);
381 contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WRITE);
383 if (wcp && req->unread_bytes) {
384 /* If we're using receivefile don't
385 * deal with a write cache.
387 flush_write_cache(fsp, SAMBA_WRITE_FLUSH);
388 delete_write_cache(fsp);
393 DO_PROFILE_INC(writecache_direct_writes);
394 total_written = real_write_file(req, fsp, data, pos, n);
395 return total_written;
398 DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f "
399 "wcp->data_size=%u\n", fsp_str_dbg(fsp), fsp->fh->fd,
400 (double)pos, (unsigned int)n, (double)wcp->offset,
401 (unsigned int)wcp->data_size));
403 fsp->fh->pos = pos + n;
405 if ((n == 1) && (data[0] == '\0') && (pos > wcp->file_size)) {
409 * This is a 1-byte write of a 0 beyond the EOF and
410 * thus implicitly also beyond the current active
411 * write cache, the typical file-extending (and
412 * allocating, but we're using the write cache here)
413 * write done by Windows. We just have to ftruncate
414 * the file and rely on posix semantics to return
415 * zeros for non-written file data that is within the
418 * We can not use wcp_file_size_change here because we
419 * might have an existing write cache, and
420 * wcp_file_size_change assumes a change to just the
421 * end of the current write cache.
424 wcp->file_size = pos + 1;
425 ret = SMB_VFS_FTRUNCATE(fsp, wcp->file_size);
427 DEBUG(0, ("wcp_file_size_change (%s): ftruncate of "
428 "size %.0f error %s\n", fsp_str_dbg(fsp),
429 (double)wcp->file_size, strerror(errno)));
437 * If we have active cache and it isn't contiguous then we flush.
438 * NOTE: There is a small problem with running out of disk ....
441 if (wcp->data_size) {
442 bool cache_flush_needed = False;
444 if ((pos >= wcp->offset) &&
445 (pos <= wcp->offset + wcp->data_size)) {
447 /* ASCII art.... JRA.
449 +--------------+-----
450 | Cached data | Rest of allocated cache buffer....
451 +--------------+-----
453 +-------------------+
455 +-------------------+
460 * Start of write overlaps or abutts the existing data.
465 data_used = MIN((wcp->alloc_size - (pos - wcp->offset)),
468 memcpy(wcp->data + (pos - wcp->offset), data,
472 * Update the current buffer size with the new data.
475 if(pos + data_used > wcp->offset + wcp->data_size) {
476 wcp->data_size = pos + data_used - wcp->offset;
480 * Update the file size if changed.
483 if (wcp->offset + wcp->data_size > wcp->file_size) {
484 if (wcp_file_size_change(fsp) == -1) {
490 * If we used all the data then
497 cache_flush_needed = True;
500 * Move the start of data forward by the amount used,
501 * cut down the amount left by the same amount.
508 DO_PROFILE_INC(writecache_abutted_writes);
509 total_written = data_used;
513 } else if ((pos < wcp->offset) &&
514 (pos + n > wcp->offset) &&
515 (pos + n <= wcp->offset + wcp->alloc_size)) {
517 /* ASCII art.... JRA.
523 +-------------------+
525 +-------------------+
530 * End of write overlaps the existing data.
533 size_t data_used = pos + n - wcp->offset;
535 memcpy(wcp->data, data + n - data_used, data_used);
538 * Update the current buffer size with the new data.
541 if(pos + n > wcp->offset + wcp->data_size) {
542 wcp->data_size = pos + n - wcp->offset;
546 * Update the file size if changed.
549 if (wcp->offset + wcp->data_size > wcp->file_size) {
550 if (wcp_file_size_change(fsp) == -1) {
556 * We don't need to move the start of data, but we
557 * cut down the amount left by the amount used.
563 * We cannot have used all the data here.
566 cache_flush_needed = True;
568 DO_PROFILE_INC(writecache_abutted_writes);
569 total_written = data_used;
573 } else if ((pos >= wcp->file_size) &&
574 (wcp->offset + wcp->data_size == wcp->file_size) &&
575 (pos > wcp->offset + wcp->data_size) &&
576 (pos < wcp->offset + wcp->alloc_size) ) {
578 /* ASCII art.... JRA.
582 +---------------+---------------+
583 | Cached data | Cache buffer |
584 +---------------+---------------+
586 +-------------------+
588 +-------------------+
593 * Non-contiguous write part of which fits within
594 * the cache buffer and is extending the file
595 * and the cache contents reflect the current
596 * data up to the current end of the file.
601 if(pos + n <= wcp->offset + wcp->alloc_size) {
604 data_used = wcp->offset+wcp->alloc_size-pos;
608 * Fill in the non-continuous area with zeros.
611 memset(wcp->data + wcp->data_size, '\0',
612 pos - (wcp->offset + wcp->data_size) );
614 memcpy(wcp->data + (pos - wcp->offset), data,
618 * Update the current buffer size with the new data.
621 if(pos + data_used > wcp->offset + wcp->data_size) {
622 wcp->data_size = pos + data_used - wcp->offset;
626 * Update the file size if changed.
629 if (wcp->offset + wcp->data_size > wcp->file_size) {
630 if (wcp_file_size_change(fsp) == -1) {
636 * If we used all the data then
643 cache_flush_needed = True;
647 * Move the start of data forward by the amount used,
648 * cut down the amount left by the same amount.
655 DO_PROFILE_INC(writecache_abutted_writes);
656 total_written = data_used;
660 } else if ( (pos >= wcp->file_size) &&
662 (wcp->file_size == wcp->offset + wcp->data_size) &&
663 (pos < wcp->file_size + wcp->alloc_size)) {
669 +---------------+---------------+
670 | Cached data | Cache buffer |
671 +---------------+---------------+
673 |<------- allocated size ---------------->|
679 MS-Office seems to do this a lot to determine if
680 there's enough space on the filesystem to write a new
686 +-----------------------+--------+
687 | Zeroed Cached data | 1 Byte |
688 +-----------------------+--------+
691 flush_write_cache(fsp, SAMBA_WRITE_FLUSH);
692 wcp->offset = wcp->file_size;
693 wcp->data_size = pos - wcp->file_size + 1;
694 memset(wcp->data, '\0', wcp->data_size);
695 memcpy(wcp->data + wcp->data_size-1, data, 1);
698 * Update the file size if changed.
701 if (wcp->offset + wcp->data_size > wcp->file_size) {
702 if (wcp_file_size_change(fsp) == -1) {
711 /* ASCII art..... JRA.
715 +---------------+---------------+
716 | Cached data | Cache buffer |
717 +---------------+---------------+
725 +---------------+---------------+
726 | Cached data | Cache buffer |
727 +---------------+---------------+
729 +-------------------+
731 +-------------------+
735 +---------------+---------------+
736 | Cached data | Cache buffer |
737 +---------------+---------------+
739 +-----------------------------------------------------+
741 +-----------------------------------------------------+
746 * Write is bigger than buffer, or there is no
747 * overlap on the low or high ends.
750 DEBUG(9,("write_file: non cacheable write : fd = %d, "
751 "pos = %.0f, len = %u, "
752 "current cache pos = %.0f len = %u\n",
753 fsp->fh->fd, (double)pos, (unsigned int)n,
755 (unsigned int)wcp->data_size ));
758 * If write would fit in the cache, and is
759 * larger than the data already in the cache,
760 * flush the cache and preferentially copy the
761 * data new data into it. Otherwise just write
765 if ( n <= wcp->alloc_size && n > wcp->data_size) {
766 cache_flush_needed = True;
768 ssize_t ret = real_write_file(NULL, fsp, data,
772 * If the write overlaps the entire
773 * cache, then discard the current
774 * contents of the cache. Fix from
775 * Rasmus Borup Hansen rbh@math.ku.dk.
778 if ((pos <= wcp->offset) &&
779 (pos + n >= wcp->offset+wcp->data_size)) {
780 DEBUG(9,("write_file: discarding "
781 "overwritten write cache: "
782 "fd = %d, off=%.0f, "
783 "size=%u\n", fsp->fh->fd,
785 (unsigned)wcp->data_size));
789 DO_PROFILE_INC(writecache_direct_writes);
794 if (pos + ret > wcp->file_size) {
795 wcp->file_size = pos + ret;
805 if (cache_flush_needed) {
806 DEBUG(3, ("SAMBA_WRITE_FLUSH:%d: due to noncontinuous "
807 "write: fd = %d, size = %.0f, pos = %.0f, "
808 "n = %u, wcp->offset=%.0f, "
809 "wcp->data_size=%u\n",
810 write_path, fsp->fh->fd,
811 (double)wcp->file_size, (double)pos,
812 (unsigned int)n, (double)wcp->offset,
813 (unsigned int)wcp->data_size ));
815 flush_write_cache(fsp, SAMBA_WRITE_FLUSH);
820 * If the write request is bigger than the cache
821 * size, write it all out.
824 if (n > wcp->alloc_size ) {
825 ssize_t ret = real_write_file(NULL,fsp, data, pos, n);
830 if (pos + ret > wcp->file_size) {
831 wcp->file_size = pos + n;
834 DO_PROFILE_INC(writecache_direct_writes);
835 return total_written + n;
839 * If there's any data left, cache it.
843 DO_PROFILE_INC(writecache_cached_writes);
844 if (wcp->data_size) {
845 DO_PROFILE_INC(writecache_abutted_writes);
847 DO_PROFILE_INC(writecache_init_writes);
850 if ((wcp->data_size == 0)
851 && (pos > wcp->file_size)
852 && (pos + n <= wcp->file_size + wcp->alloc_size)) {
854 * This is a write completely beyond the
855 * current EOF, but within reach of the write
856 * cache. We expect fill-up writes pretty
857 * soon, so it does not make sense to start
858 * the write cache at the current
859 * offset. These fill-up writes would trigger
860 * separate pwrites or even unnecessary cache
861 * flushes because they overlap if this is a
862 * one-byte allocating write.
864 wcp->offset = wcp->file_size;
865 wcp->data_size = pos - wcp->file_size;
866 memset(wcp->data, 0, wcp->data_size);
869 memcpy(wcp->data+wcp->data_size, data, n);
870 if (wcp->data_size == 0) {
876 * Update the file size if changed.
879 if (wcp->offset + wcp->data_size > wcp->file_size) {
880 if (wcp_file_size_change(fsp) == -1) {
884 DEBUG(9, ("wcp->offset = %.0f wcp->data_size = %u cache "
886 (double)wcp->offset, (unsigned int)wcp->data_size,
890 return total_written; /* .... that's a write :) */
893 return total_written;
896 /****************************************************************************
897 Delete the write cache structure.
898 ****************************************************************************/
900 void delete_write_cache(files_struct *fsp)
902 struct write_cache *wcp;
908 if(!(wcp = fsp->wcp)) {
912 DO_PROFILE_INC(writecache_deallocations);
913 allocated_write_caches--;
915 SMB_ASSERT(wcp->data_size == 0);
917 SAFE_FREE(wcp->data);
920 DEBUG(10,("delete_write_cache: File %s deleted write cache\n",
924 /****************************************************************************
925 Setup the write cache structure.
926 ****************************************************************************/
928 static bool setup_write_cache(files_struct *fsp, off_t file_size)
930 ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
931 struct write_cache *wcp;
933 if (allocated_write_caches >= MAX_WRITE_CACHES) {
937 if(alloc_size == 0 || fsp->wcp) {
941 if((wcp = SMB_MALLOC_P(struct write_cache)) == NULL) {
942 DEBUG(0,("setup_write_cache: malloc fail.\n"));
946 wcp->file_size = file_size;
948 wcp->alloc_size = alloc_size;
950 if((wcp->data = (char *)SMB_MALLOC(wcp->alloc_size)) == NULL) {
951 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
952 (unsigned int)wcp->alloc_size ));
957 memset(wcp->data, '\0', wcp->alloc_size );
960 DO_PROFILE_INC(writecache_allocations);
961 allocated_write_caches++;
963 DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
964 fsp_str_dbg(fsp), (unsigned long)wcp->alloc_size));
969 /****************************************************************************
970 Cope with a size change.
971 ****************************************************************************/
973 void set_filelen_write_cache(files_struct *fsp, off_t file_size)
976 /* The cache *must* have been flushed before we do this. */
977 if (fsp->wcp->data_size != 0) {
979 if (asprintf(&msg, "set_filelen_write_cache: size change "
980 "on file %s with write cache size = %lu\n",
981 fsp->fsp_name->base_name,
982 (unsigned long)fsp->wcp->data_size) != -1) {
985 smb_panic("set_filelen_write_cache");
988 fsp->wcp->file_size = file_size;
992 /*******************************************************************
993 Flush a write cache struct to disk.
994 ********************************************************************/
996 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
998 struct write_cache *wcp = fsp->wcp;
1002 if(!wcp || !wcp->data_size) {
1006 data_size = wcp->data_size;
1010 case SAMBA_SEEK_FLUSH:
1011 DO_PROFILE_INC(writecache_flush_reason_seek);
1013 case SAMBA_READ_FLUSH:
1014 DO_PROFILE_INC(writecache_flush_reason_read);
1016 case SAMBA_WRITE_FLUSH:
1017 DO_PROFILE_INC(writecache_flush_reason_write);;
1019 case SAMBA_READRAW_FLUSH:
1020 DO_PROFILE_INC(writecache_flush_reason_readraw);
1022 case SAMBA_OPLOCK_RELEASE_FLUSH:
1023 DO_PROFILE_INC(writecache_flush_reason_oplock);
1025 case SAMBA_CLOSE_FLUSH:
1026 DO_PROFILE_INC(writecache_flush_reason_close);
1028 case SAMBA_SYNC_FLUSH:
1029 DO_PROFILE_INC(writecache_flush_reason_sync);
1031 case SAMBA_SIZECHANGE_FLUSH:
1032 DO_PROFILE_INC(writecache_flush_reason_sizechange);
1038 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
1039 fsp->fh->fd, (double)wcp->offset, (unsigned int)data_size));
1041 if(data_size == wcp->alloc_size) {
1042 DO_PROFILE_INC(writecache_perfect_writes);
1045 ret = real_write_file(NULL, fsp, wcp->data, wcp->offset, data_size);
1048 * Ensure file size if kept up to date if write extends file.
1051 if ((ret != -1) && (wcp->offset + ret > wcp->file_size)) {
1052 wcp->file_size = wcp->offset + ret;
1058 /*******************************************************************
1060 ********************************************************************/
1062 NTSTATUS sync_file(connection_struct *conn, files_struct *fsp, bool write_through)
1064 if (fsp->fh->fd == -1)
1065 return NT_STATUS_INVALID_HANDLE;
1067 if (lp_strict_sync(SNUM(conn)) &&
1068 (lp_sync_always(SNUM(conn)) || write_through)) {
1069 int ret = flush_write_cache(fsp, SAMBA_SYNC_FLUSH);
1071 return map_nt_error_from_unix(errno);
1073 ret = smb_vfs_fsync_sync(fsp);
1075 return map_nt_error_from_unix(errno);
1078 return NT_STATUS_OK;
1081 /************************************************************
1082 Perform a stat whether a valid fd or not.
1083 ************************************************************/
1085 int fsp_stat(files_struct *fsp)
1087 if (fsp->fh->fd == -1) {
1088 if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
1089 return SMB_VFS_LSTAT(fsp->conn, fsp->fsp_name);
1091 return SMB_VFS_STAT(fsp->conn, fsp->fsp_name);
1094 return SMB_VFS_FSTAT(fsp, &fsp->fsp_name->st);