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 ret = vfs_write_data(req, fsp, data, n);
122 if (pos && lp_strict_allocate(SNUM(fsp->conn) &&
124 if (vfs_fill_sparse(fsp, pos) == -1) {
128 ret = vfs_pwrite_data(req, fsp, data, n, pos);
131 DEBUG(10,("real_write_file (%s): pos = %.0f, size = %lu, returned %ld\n",
132 fsp_str_dbg(fsp), (double)pos, (unsigned long)n, (long)ret));
137 /* Yes - this is correct - writes don't update this. JRA. */
138 /* Found by Samba4 tests. */
140 fsp->position_information = fsp->pos;
147 /****************************************************************************
148 File size cache change.
149 Updates size on disk but doesn't flush the cache.
150 ****************************************************************************/
152 static int wcp_file_size_change(files_struct *fsp)
155 struct write_cache *wcp = fsp->wcp;
157 wcp->file_size = wcp->offset + wcp->data_size;
158 ret = SMB_VFS_FTRUNCATE(fsp, wcp->file_size);
160 DEBUG(0,("wcp_file_size_change (%s): ftruncate of size %.0f "
161 "error %s\n", fsp_str_dbg(fsp),
162 (double)wcp->file_size, strerror(errno)));
167 void update_write_time_handler(struct tevent_context *ctx,
168 struct tevent_timer *te,
172 files_struct *fsp = (files_struct *)private_data;
174 DEBUG(5, ("Update write time on %s\n", fsp_str_dbg(fsp)));
176 /* change the write time in the open file db. */
177 (void)set_write_time(fsp->file_id, timespec_current());
180 notify_fname(fsp->conn, NOTIFY_ACTION_MODIFIED,
181 FILE_NOTIFY_CHANGE_LAST_WRITE, fsp->fsp_name->base_name);
183 /* Remove the timed event handler. */
184 TALLOC_FREE(fsp->update_write_time_event);
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_flags & FSP_POSIX_FLAGS_OPEN) {
197 /* Don't use delayed writes on POSIX files. */
201 if (fsp->write_time_forced) {
202 /* No point - "sticky" write times
208 /* We need to remember someone did a write
209 * and update to current time on close. */
211 fsp->update_write_time_on_close = true;
213 if (fsp->update_write_time_triggered) {
215 * We only update the write time after 2 seconds
216 * on the first normal write. After that
217 * no other writes affect this until close.
221 fsp->update_write_time_triggered = true;
223 delay = lp_parm_int(SNUM(fsp->conn),
224 "smbd", "writetimeupdatedelay",
225 WRITE_TIME_UPDATE_USEC_DELAY);
227 DEBUG(5, ("Update write time %d usec later on %s\n",
228 delay, fsp_str_dbg(fsp)));
230 /* trigger the update 2 seconds later */
231 fsp->update_write_time_event =
232 tevent_add_timer(fsp->conn->sconn->ev_ctx, NULL,
233 timeval_current_ofs_usec(delay),
234 update_write_time_handler, fsp);
237 void trigger_write_time_update_immediate(struct files_struct *fsp)
239 struct smb_file_time ft;
241 if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
242 /* Don't use delayed writes on POSIX files. */
246 if (fsp->write_time_forced) {
248 * No point - "sticky" write times
254 TALLOC_FREE(fsp->update_write_time_event);
255 DEBUG(5, ("Update write time immediate on %s\n",
258 /* After an immediate update, reset the trigger. */
259 fsp->update_write_time_triggered = true;
260 fsp->update_write_time_on_close = false;
263 ft.mtime = timespec_current();
265 /* Update the time in the open file db. */
266 (void)set_write_time(fsp->file_id, ft.mtime);
268 /* Now set on disk - takes care of notify. */
269 (void)smb_set_file_time(fsp->conn, fsp, fsp->fsp_name, &ft, false);
272 void mark_file_modified(files_struct *fsp)
280 fsp->modified = true;
282 if (SMB_VFS_FSTAT(fsp, &fsp->fsp_name->st) != 0) {
285 trigger_write_time_update(fsp);
287 if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
290 if (!(lp_store_dos_attributes(SNUM(fsp->conn)) ||
291 MAP_ARCHIVE(fsp->conn))) {
295 dosmode = dos_mode(fsp->conn, fsp->fsp_name);
296 if (IS_DOS_ARCHIVE(dosmode)) {
299 file_set_dosmode(fsp->conn, fsp->fsp_name,
300 dosmode | FILE_ATTRIBUTE_ARCHIVE, NULL, false);
303 /****************************************************************************
305 ****************************************************************************/
307 ssize_t write_file(struct smb_request *req,
313 struct write_cache *wcp = fsp->wcp;
314 ssize_t total_written = 0;
317 if (fsp->print_file) {
321 ret = print_spool_write(fsp, data, n, pos, &t);
329 if (!fsp->can_write) {
335 * If this is the first write and we have an exclusive oplock
336 * then setup the write cache.
339 if (!fsp->modified &&
340 EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) &&
343 * Note: no write cache with leases!
344 * as the handles would have to share the write cache
345 * that's possible but an improvement for another day...
347 setup_write_cache(fsp, fsp->fsp_name->st.st_ex_size);
351 mark_file_modified(fsp);
353 DO_PROFILE_INC(writecache_total_writes);
354 if (!fsp->oplock_type) {
355 DO_PROFILE_INC(writecache_non_oplock_writes);
359 * If this file is level II oplocked then we need
360 * to grab the shared memory lock and inform all
361 * other files with a level II lock that they need
362 * to flush their read caches. We keep the lock over
363 * the shared memory area whilst doing this.
366 /* This should actually be improved to span the write. */
367 contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WRITE);
368 contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WRITE);
370 if (wcp && req->unread_bytes) {
371 /* If we're using receivefile don't
372 * deal with a write cache.
374 flush_write_cache(fsp, SAMBA_WRITE_FLUSH);
375 delete_write_cache(fsp);
380 DO_PROFILE_INC(writecache_direct_writes);
381 total_written = real_write_file(req, fsp, data, pos, n);
382 return total_written;
385 DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f "
386 "wcp->data_size=%u\n", fsp_str_dbg(fsp), fsp->fh->fd,
387 (double)pos, (unsigned int)n, (double)wcp->offset,
388 (unsigned int)wcp->data_size));
390 fsp->fh->pos = pos + n;
392 if ((n == 1) && (data[0] == '\0') && (pos > wcp->file_size)) {
396 * This is a 1-byte write of a 0 beyond the EOF and
397 * thus implicitly also beyond the current active
398 * write cache, the typical file-extending (and
399 * allocating, but we're using the write cache here)
400 * write done by Windows. We just have to ftruncate
401 * the file and rely on posix semantics to return
402 * zeros for non-written file data that is within the
405 * We can not use wcp_file_size_change here because we
406 * might have an existing write cache, and
407 * wcp_file_size_change assumes a change to just the
408 * end of the current write cache.
411 wcp->file_size = pos + 1;
412 ret = SMB_VFS_FTRUNCATE(fsp, wcp->file_size);
414 DEBUG(0, ("wcp_file_size_change (%s): ftruncate of "
415 "size %.0f error %s\n", fsp_str_dbg(fsp),
416 (double)wcp->file_size, strerror(errno)));
424 * If we have active cache and it isn't contiguous then we flush.
425 * NOTE: There is a small problem with running out of disk ....
428 if (wcp->data_size) {
429 bool cache_flush_needed = False;
431 if ((pos >= wcp->offset) &&
432 (pos <= wcp->offset + wcp->data_size)) {
434 /* ASCII art.... JRA.
436 +--------------+-----
437 | Cached data | Rest of allocated cache buffer....
438 +--------------+-----
440 +-------------------+
442 +-------------------+
447 * Start of write overlaps or abutts the existing data.
452 data_used = MIN((wcp->alloc_size - (pos - wcp->offset)),
455 memcpy(wcp->data + (pos - wcp->offset), data,
459 * Update the current buffer size with the new data.
462 if(pos + data_used > wcp->offset + wcp->data_size) {
463 wcp->data_size = pos + data_used - wcp->offset;
467 * Update the file size if changed.
470 if (wcp->offset + wcp->data_size > wcp->file_size) {
471 if (wcp_file_size_change(fsp) == -1) {
477 * If we used all the data then
484 cache_flush_needed = True;
487 * Move the start of data forward by the amount used,
488 * cut down the amount left by the same amount.
495 DO_PROFILE_INC(writecache_abutted_writes);
496 total_written = data_used;
500 } else if ((pos < wcp->offset) &&
501 (pos + n > wcp->offset) &&
502 (pos + n <= wcp->offset + wcp->alloc_size)) {
504 /* ASCII art.... JRA.
510 +-------------------+
512 +-------------------+
517 * End of write overlaps the existing data.
520 size_t data_used = pos + n - wcp->offset;
522 memcpy(wcp->data, data + n - data_used, data_used);
525 * Update the current buffer size with the new data.
528 if(pos + n > wcp->offset + wcp->data_size) {
529 wcp->data_size = pos + n - wcp->offset;
533 * Update the file size if changed.
536 if (wcp->offset + wcp->data_size > wcp->file_size) {
537 if (wcp_file_size_change(fsp) == -1) {
543 * We don't need to move the start of data, but we
544 * cut down the amount left by the amount used.
550 * We cannot have used all the data here.
553 cache_flush_needed = True;
555 DO_PROFILE_INC(writecache_abutted_writes);
556 total_written = data_used;
560 } else if ((pos >= wcp->file_size) &&
561 (wcp->offset + wcp->data_size == wcp->file_size) &&
562 (pos > wcp->offset + wcp->data_size) &&
563 (pos < wcp->offset + wcp->alloc_size) ) {
565 /* ASCII art.... JRA.
569 +---------------+---------------+
570 | Cached data | Cache buffer |
571 +---------------+---------------+
573 +-------------------+
575 +-------------------+
580 * Non-contiguous write part of which fits within
581 * the cache buffer and is extending the file
582 * and the cache contents reflect the current
583 * data up to the current end of the file.
588 if(pos + n <= wcp->offset + wcp->alloc_size) {
591 data_used = wcp->offset+wcp->alloc_size-pos;
595 * Fill in the non-continuous area with zeros.
598 memset(wcp->data + wcp->data_size, '\0',
599 pos - (wcp->offset + wcp->data_size) );
601 memcpy(wcp->data + (pos - wcp->offset), data,
605 * Update the current buffer size with the new data.
608 if(pos + data_used > wcp->offset + wcp->data_size) {
609 wcp->data_size = pos + data_used - wcp->offset;
613 * Update the file size if changed.
616 if (wcp->offset + wcp->data_size > wcp->file_size) {
617 if (wcp_file_size_change(fsp) == -1) {
623 * If we used all the data then
630 cache_flush_needed = True;
634 * Move the start of data forward by the amount used,
635 * cut down the amount left by the same amount.
642 DO_PROFILE_INC(writecache_abutted_writes);
643 total_written = data_used;
647 } else if ( (pos >= wcp->file_size) &&
649 (wcp->file_size == wcp->offset + wcp->data_size) &&
650 (pos < wcp->file_size + wcp->alloc_size)) {
656 +---------------+---------------+
657 | Cached data | Cache buffer |
658 +---------------+---------------+
660 |<------- allocated size ---------------->|
666 MS-Office seems to do this a lot to determine if
667 there's enough space on the filesystem to write a new
673 +-----------------------+--------+
674 | Zeroed Cached data | 1 Byte |
675 +-----------------------+--------+
678 flush_write_cache(fsp, SAMBA_WRITE_FLUSH);
679 wcp->offset = wcp->file_size;
680 wcp->data_size = pos - wcp->file_size + 1;
681 memset(wcp->data, '\0', wcp->data_size);
682 memcpy(wcp->data + wcp->data_size-1, data, 1);
685 * Update the file size if changed.
688 if (wcp->offset + wcp->data_size > wcp->file_size) {
689 if (wcp_file_size_change(fsp) == -1) {
698 /* ASCII art..... JRA.
702 +---------------+---------------+
703 | Cached data | Cache buffer |
704 +---------------+---------------+
712 +---------------+---------------+
713 | Cached data | Cache buffer |
714 +---------------+---------------+
716 +-------------------+
718 +-------------------+
722 +---------------+---------------+
723 | Cached data | Cache buffer |
724 +---------------+---------------+
726 +-----------------------------------------------------+
728 +-----------------------------------------------------+
733 * Write is bigger than buffer, or there is no
734 * overlap on the low or high ends.
737 DEBUG(9,("write_file: non cacheable write : fd = %d, "
738 "pos = %.0f, len = %u, "
739 "current cache pos = %.0f len = %u\n",
740 fsp->fh->fd, (double)pos, (unsigned int)n,
742 (unsigned int)wcp->data_size ));
745 * If write would fit in the cache, and is
746 * larger than the data already in the cache,
747 * flush the cache and preferentially copy the
748 * data new data into it. Otherwise just write
752 if ( n <= wcp->alloc_size && n > wcp->data_size) {
753 cache_flush_needed = True;
755 ssize_t ret = real_write_file(NULL, fsp, data,
759 * If the write overlaps the entire
760 * cache, then discard the current
761 * contents of the cache. Fix from
762 * Rasmus Borup Hansen rbh@math.ku.dk.
765 if ((pos <= wcp->offset) &&
766 (pos + n >= wcp->offset+wcp->data_size)) {
767 DEBUG(9,("write_file: discarding "
768 "overwritten write cache: "
769 "fd = %d, off=%.0f, "
770 "size=%u\n", fsp->fh->fd,
772 (unsigned)wcp->data_size));
776 DO_PROFILE_INC(writecache_direct_writes);
781 if (pos + ret > wcp->file_size) {
782 wcp->file_size = pos + ret;
792 if (cache_flush_needed) {
793 DEBUG(3, ("SAMBA_WRITE_FLUSH:%d: due to noncontinuous "
794 "write: fd = %d, size = %.0f, pos = %.0f, "
795 "n = %u, wcp->offset=%.0f, "
796 "wcp->data_size=%u\n",
797 write_path, fsp->fh->fd,
798 (double)wcp->file_size, (double)pos,
799 (unsigned int)n, (double)wcp->offset,
800 (unsigned int)wcp->data_size ));
802 flush_write_cache(fsp, SAMBA_WRITE_FLUSH);
807 * If the write request is bigger than the cache
808 * size, write it all out.
811 if (n > wcp->alloc_size ) {
812 ssize_t ret = real_write_file(NULL,fsp, data, pos, n);
817 if (pos + ret > wcp->file_size) {
818 wcp->file_size = pos + n;
821 DO_PROFILE_INC(writecache_direct_writes);
822 return total_written + n;
826 * If there's any data left, cache it.
830 DO_PROFILE_INC(writecache_cached_writes);
831 if (wcp->data_size) {
832 DO_PROFILE_INC(writecache_abutted_writes);
834 DO_PROFILE_INC(writecache_init_writes);
837 if ((wcp->data_size == 0)
838 && (pos > wcp->file_size)
839 && (pos + n <= wcp->file_size + wcp->alloc_size)) {
841 * This is a write completely beyond the
842 * current EOF, but within reach of the write
843 * cache. We expect fill-up writes pretty
844 * soon, so it does not make sense to start
845 * the write cache at the current
846 * offset. These fill-up writes would trigger
847 * separate pwrites or even unnecessary cache
848 * flushes because they overlap if this is a
849 * one-byte allocating write.
851 wcp->offset = wcp->file_size;
852 wcp->data_size = pos - wcp->file_size;
853 memset(wcp->data, 0, wcp->data_size);
856 memcpy(wcp->data+wcp->data_size, data, n);
857 if (wcp->data_size == 0) {
863 * Update the file size if changed.
866 if (wcp->offset + wcp->data_size > wcp->file_size) {
867 if (wcp_file_size_change(fsp) == -1) {
871 DEBUG(9, ("wcp->offset = %.0f wcp->data_size = %u cache "
873 (double)wcp->offset, (unsigned int)wcp->data_size,
877 return total_written; /* .... that's a write :) */
880 return total_written;
883 /****************************************************************************
884 Delete the write cache structure.
885 ****************************************************************************/
887 void delete_write_cache(files_struct *fsp)
889 struct write_cache *wcp;
895 if(!(wcp = fsp->wcp)) {
899 DO_PROFILE_INC(writecache_deallocations);
900 allocated_write_caches--;
902 SMB_ASSERT(wcp->data_size == 0);
904 SAFE_FREE(wcp->data);
907 DEBUG(10,("delete_write_cache: File %s deleted write cache\n",
911 /****************************************************************************
912 Setup the write cache structure.
913 ****************************************************************************/
915 static bool setup_write_cache(files_struct *fsp, off_t file_size)
917 ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
918 struct write_cache *wcp;
920 if (allocated_write_caches >= MAX_WRITE_CACHES) {
924 if(alloc_size == 0 || fsp->wcp) {
928 if((wcp = SMB_MALLOC_P(struct write_cache)) == NULL) {
929 DEBUG(0,("setup_write_cache: malloc fail.\n"));
933 wcp->file_size = file_size;
935 wcp->alloc_size = alloc_size;
937 if((wcp->data = (char *)SMB_MALLOC(wcp->alloc_size)) == NULL) {
938 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
939 (unsigned int)wcp->alloc_size ));
944 memset(wcp->data, '\0', wcp->alloc_size );
947 DO_PROFILE_INC(writecache_allocations);
948 allocated_write_caches++;
950 DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
951 fsp_str_dbg(fsp), (unsigned long)wcp->alloc_size));
956 /****************************************************************************
957 Cope with a size change.
958 ****************************************************************************/
960 void set_filelen_write_cache(files_struct *fsp, off_t file_size)
963 /* The cache *must* have been flushed before we do this. */
964 if (fsp->wcp->data_size != 0) {
966 if (asprintf(&msg, "set_filelen_write_cache: size change "
967 "on file %s with write cache size = %lu\n",
968 fsp->fsp_name->base_name,
969 (unsigned long)fsp->wcp->data_size) != -1) {
972 smb_panic("set_filelen_write_cache");
975 fsp->wcp->file_size = file_size;
979 /*******************************************************************
980 Flush a write cache struct to disk.
981 ********************************************************************/
983 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
985 struct write_cache *wcp = fsp->wcp;
989 if(!wcp || !wcp->data_size) {
993 data_size = wcp->data_size;
997 case SAMBA_SEEK_FLUSH:
998 DO_PROFILE_INC(writecache_flush_reason_seek);
1000 case SAMBA_READ_FLUSH:
1001 DO_PROFILE_INC(writecache_flush_reason_read);
1003 case SAMBA_WRITE_FLUSH:
1004 DO_PROFILE_INC(writecache_flush_reason_write);;
1006 case SAMBA_READRAW_FLUSH:
1007 DO_PROFILE_INC(writecache_flush_reason_readraw);
1009 case SAMBA_OPLOCK_RELEASE_FLUSH:
1010 DO_PROFILE_INC(writecache_flush_reason_oplock);
1012 case SAMBA_CLOSE_FLUSH:
1013 DO_PROFILE_INC(writecache_flush_reason_close);
1015 case SAMBA_SYNC_FLUSH:
1016 DO_PROFILE_INC(writecache_flush_reason_sync);
1018 case SAMBA_SIZECHANGE_FLUSH:
1019 DO_PROFILE_INC(writecache_flush_reason_sizechange);
1025 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
1026 fsp->fh->fd, (double)wcp->offset, (unsigned int)data_size));
1028 if(data_size == wcp->alloc_size) {
1029 DO_PROFILE_INC(writecache_perfect_writes);
1032 ret = real_write_file(NULL, fsp, wcp->data, wcp->offset, data_size);
1035 * Ensure file size if kept up to date if write extends file.
1038 if ((ret != -1) && (wcp->offset + ret > wcp->file_size)) {
1039 wcp->file_size = wcp->offset + ret;
1045 /*******************************************************************
1047 ********************************************************************/
1049 NTSTATUS sync_file(connection_struct *conn, files_struct *fsp, bool write_through)
1051 if (fsp->fh->fd == -1)
1052 return NT_STATUS_INVALID_HANDLE;
1054 if (lp_strict_sync(SNUM(conn)) &&
1055 (lp_sync_always(SNUM(conn)) || write_through)) {
1056 int ret = flush_write_cache(fsp, SAMBA_SYNC_FLUSH);
1058 return map_nt_error_from_unix(errno);
1060 ret = SMB_VFS_FSYNC(fsp);
1062 return map_nt_error_from_unix(errno);
1065 return NT_STATUS_OK;
1068 /************************************************************
1069 Perform a stat whether a valid fd or not.
1070 ************************************************************/
1072 int fsp_stat(files_struct *fsp)
1074 if (fsp->fh->fd == -1) {
1075 if (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) {
1076 return SMB_VFS_LSTAT(fsp->conn, fsp->fsp_name);
1078 return SMB_VFS_STAT(fsp->conn, fsp->fsp_name);
1081 return SMB_VFS_FSTAT(fsp, &fsp->fsp_name->st);