Samba 3: added Samba 3.0.24 sources
[tomato.git] / release / src / router / samba3 / source / smbd / fileio.c
blob375bfbe7cf3a381a14acdcbaf261e81694f5b6d3
1 /*
2 Unix SMB/Netbios implementation.
3 Version 1.9.
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 2 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, write to the Free Software
20 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #include "includes.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;
35 if(!wcp) {
36 return False;
39 if( n > wcp->data_size || pos < wcp->offset || pos + n > wcp->offset + wcp->data_size) {
40 return False;
43 memcpy(data, wcp->data + (pos - wcp->offset), n);
45 DO_PROFILE_INC(writecache_read_hits);
47 return True;
50 /****************************************************************************
51 Read from a file.
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) {
60 return -1;
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;
70 return n;
73 flush_write_cache(fsp, READ_FLUSH);
75 fsp->fh->pos = pos;
77 if (n > 0) {
78 #ifdef DMF_FIX
79 int numretries = 3;
80 tryagain:
81 readret = SMB_VFS_PREAD(fsp,fsp->fh->fd,data,n,pos);
83 if (readret == -1) {
84 if ((errno == EAGAIN) && numretries) {
85 DEBUG(3,("read_file EAGAIN retry in 10 seconds\n"));
86 (void)sleep(10);
87 --numretries;
88 goto tryagain;
90 return -1;
92 #else /* NO DMF fix. */
93 readret = SMB_VFS_PREAD(fsp,fsp->fh->fd,data,n,pos);
95 if (readret == -1) {
96 return -1;
98 #endif
99 if (readret > 0) {
100 ret += readret;
104 DEBUG(10,("read_file (%s): pos = %.0f, size = %lu, returned %lu\n",
105 fsp->fsp_name, (double)pos, (unsigned long)n, (long)ret ));
107 fsp->fh->pos += ret;
108 fsp->fh->position_information = fsp->fh->pos;
110 return(ret);
113 /* how many write cache buffers have been allocated */
114 static unsigned int allocated_write_caches;
116 /****************************************************************************
117 *Really* write to a file.
118 ****************************************************************************/
120 static ssize_t real_write_file(files_struct *fsp,const char *data, SMB_OFF_T pos, size_t n)
122 ssize_t ret;
124 if (pos == -1) {
125 ret = vfs_write_data(fsp, data, n);
126 } else {
127 fsp->fh->pos = pos;
128 if (pos && lp_strict_allocate(SNUM(fsp->conn))) {
129 if (vfs_fill_sparse(fsp, pos) == -1) {
130 return -1;
133 ret = vfs_pwrite_data(fsp, data, n, pos);
136 DEBUG(10,("real_write_file (%s): pos = %.0f, size = %lu, returned %ld\n",
137 fsp->fsp_name, (double)pos, (unsigned long)n, (long)ret ));
139 if (ret != -1) {
140 fsp->fh->pos += ret;
143 * It turns out that setting the last write time from a Windows
144 * client stops any subsequent writes from updating the write time.
145 * Doing this after the write gives a race condition here where
146 * a stat may see the changed write time before we reset it here,
147 * but it's cheaper than having to store the write time in shared
148 * memory and look it up using dev/inode across all running smbd's.
149 * The 99% solution will hopefully be good enough in this case. JRA.
152 if (fsp->pending_modtime) {
153 set_filetime(fsp->conn, fsp->fsp_name, fsp->pending_modtime);
155 /* If we didn't get the "set modtime" call ourselves, we must
156 store the last write time to restore on close. JRA. */
157 if (!fsp->pending_modtime_owner) {
158 fsp->last_write_time = time(NULL);
162 /* Yes - this is correct - writes don't update this. JRA. */
163 /* Found by Samba4 tests. */
164 #if 0
165 fsp->position_information = fsp->pos;
166 #endif
169 return ret;
172 /****************************************************************************
173 File size cache change.
174 Updates size on disk but doesn't flush the cache.
175 ****************************************************************************/
177 static int wcp_file_size_change(files_struct *fsp)
179 int ret;
180 write_cache *wcp = fsp->wcp;
182 wcp->file_size = wcp->offset + wcp->data_size;
183 ret = SMB_VFS_FTRUNCATE(fsp, fsp->fh->fd, wcp->file_size);
184 if (ret == -1) {
185 DEBUG(0,("wcp_file_size_change (%s): ftruncate of size %.0f error %s\n",
186 fsp->fsp_name, (double)wcp->file_size, strerror(errno) ));
188 return ret;
191 /****************************************************************************
192 Write to a file.
193 ****************************************************************************/
195 ssize_t write_file(files_struct *fsp, const char *data, SMB_OFF_T pos, size_t n)
197 write_cache *wcp = fsp->wcp;
198 ssize_t total_written = 0;
199 int write_path = -1;
201 if (fsp->print_file) {
202 fstring sharename;
203 uint32 jobid;
205 if (!rap_to_pjobid(fsp->rap_print_jobid, sharename, &jobid)) {
206 DEBUG(3,("write_file: Unable to map RAP jobid %u to jobid.\n",
207 (unsigned int)fsp->rap_print_jobid ));
208 errno = EBADF;
209 return -1;
212 return print_job_write(SNUM(fsp->conn), jobid, data, pos, n);
215 if (!fsp->can_write) {
216 errno = EPERM;
217 return(0);
220 if (!fsp->modified) {
221 SMB_STRUCT_STAT st;
222 fsp->modified = True;
224 if (SMB_VFS_FSTAT(fsp,fsp->fh->fd,&st) == 0) {
225 int dosmode = dos_mode(fsp->conn,fsp->fsp_name,&st);
226 if ((lp_store_dos_attributes(SNUM(fsp->conn)) || MAP_ARCHIVE(fsp->conn)) && !IS_DOS_ARCHIVE(dosmode)) {
227 file_set_dosmode(fsp->conn,fsp->fsp_name,dosmode | aARCH,&st, False);
231 * If this is the first write and we have an exclusive oplock then setup
232 * the write cache.
235 if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && !wcp) {
236 setup_write_cache(fsp, st.st_size);
237 wcp = fsp->wcp;
242 #ifdef WITH_PROFILE
243 DO_PROFILE_INC(writecache_total_writes);
244 if (!fsp->oplock_type) {
245 DO_PROFILE_INC(writecache_non_oplock_writes);
247 #endif
250 * If this file is level II oplocked then we need
251 * to grab the shared memory lock and inform all
252 * other files with a level II lock that they need
253 * to flush their read caches. We keep the lock over
254 * the shared memory area whilst doing this.
257 release_level_2_oplocks_on_change(fsp);
259 #ifdef WITH_PROFILE
260 if (profile_p && profile_p->writecache_total_writes % 500 == 0) {
261 DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u total=%u \
262 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
263 profile_p->writecache_init_writes,
264 profile_p->writecache_abutted_writes,
265 profile_p->writecache_total_writes,
266 profile_p->writecache_non_oplock_writes,
267 profile_p->writecache_allocated_write_caches,
268 profile_p->writecache_num_write_caches,
269 profile_p->writecache_direct_writes,
270 profile_p->writecache_num_perfect_writes,
271 profile_p->writecache_read_hits ));
273 DEBUG(3,("WRITECACHE: Flushes SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
274 profile_p->writecache_flushed_writes[SEEK_FLUSH],
275 profile_p->writecache_flushed_writes[READ_FLUSH],
276 profile_p->writecache_flushed_writes[WRITE_FLUSH],
277 profile_p->writecache_flushed_writes[READRAW_FLUSH],
278 profile_p->writecache_flushed_writes[OPLOCK_RELEASE_FLUSH],
279 profile_p->writecache_flushed_writes[CLOSE_FLUSH],
280 profile_p->writecache_flushed_writes[SYNC_FLUSH] ));
282 #endif
284 if(!wcp) {
285 DO_PROFILE_INC(writecache_direct_writes);
286 total_written = real_write_file(fsp, data, pos, n);
287 return total_written;
290 DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f wcp->data_size=%u\n",
291 fsp->fsp_name, fsp->fh->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size));
293 fsp->fh->pos = pos + n;
296 * If we have active cache and it isn't contiguous then we flush.
297 * NOTE: There is a small problem with running out of disk ....
300 if (wcp->data_size) {
301 BOOL cache_flush_needed = False;
303 if ((pos >= wcp->offset) && (pos <= wcp->offset + wcp->data_size)) {
305 /* ASCII art.... JRA.
307 +--------------+-----
308 | Cached data | Rest of allocated cache buffer....
309 +--------------+-----
311 +-------------------+
312 | Data to write |
313 +-------------------+
318 * Start of write overlaps or abutts the existing data.
321 size_t data_used = MIN((wcp->alloc_size - (pos - wcp->offset)), n);
323 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
326 * Update the current buffer size with the new data.
329 if(pos + data_used > wcp->offset + wcp->data_size) {
330 wcp->data_size = pos + data_used - wcp->offset;
334 * Update the file size if changed.
337 if (wcp->offset + wcp->data_size > wcp->file_size) {
338 if (wcp_file_size_change(fsp) == -1) {
339 return -1;
344 * If we used all the data then
345 * return here.
348 if(n == data_used) {
349 return n;
350 } else {
351 cache_flush_needed = True;
354 * Move the start of data forward by the amount used,
355 * cut down the amount left by the same amount.
358 data += data_used;
359 pos += data_used;
360 n -= data_used;
362 DO_PROFILE_INC(writecache_abutted_writes);
363 total_written = data_used;
365 write_path = 1;
367 } else if ((pos < wcp->offset) && (pos + n > wcp->offset) &&
368 (pos + n <= wcp->offset + wcp->alloc_size)) {
370 /* ASCII art.... JRA.
372 +---------------+
373 | Cache buffer |
374 +---------------+
376 +-------------------+
377 | Data to write |
378 +-------------------+
383 * End of write overlaps the existing data.
386 size_t data_used = pos + n - wcp->offset;
388 memcpy(wcp->data, data + n - data_used, data_used);
391 * Update the current buffer size with the new data.
394 if(pos + n > wcp->offset + wcp->data_size) {
395 wcp->data_size = pos + n - wcp->offset;
399 * Update the file size if changed.
402 if (wcp->offset + wcp->data_size > wcp->file_size) {
403 if (wcp_file_size_change(fsp) == -1) {
404 return -1;
409 * We don't need to move the start of data, but we
410 * cut down the amount left by the amount used.
413 n -= data_used;
416 * We cannot have used all the data here.
419 cache_flush_needed = True;
421 DO_PROFILE_INC(writecache_abutted_writes);
422 total_written = data_used;
424 write_path = 2;
426 } else if ( (pos >= wcp->file_size) &&
427 (wcp->offset + wcp->data_size == wcp->file_size) &&
428 (pos > wcp->offset + wcp->data_size) &&
429 (pos < wcp->offset + wcp->alloc_size) ) {
431 /* ASCII art.... JRA.
433 End of file ---->|
435 +---------------+---------------+
436 | Cached data | Cache buffer |
437 +---------------+---------------+
439 +-------------------+
440 | Data to write |
441 +-------------------+
446 * Non-contiguous write part of which fits within
447 * the cache buffer and is extending the file
448 * and the cache contents reflect the current
449 * data up to the current end of the file.
452 size_t data_used;
454 if(pos + n <= wcp->offset + wcp->alloc_size) {
455 data_used = n;
456 } else {
457 data_used = wcp->offset + wcp->alloc_size - pos;
461 * Fill in the non-continuous area with zeros.
464 memset(wcp->data + wcp->data_size, '\0',
465 pos - (wcp->offset + wcp->data_size) );
467 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
470 * Update the current buffer size with the new data.
473 if(pos + data_used > wcp->offset + wcp->data_size) {
474 wcp->data_size = pos + data_used - wcp->offset;
478 * Update the file size if changed.
481 if (wcp->offset + wcp->data_size > wcp->file_size) {
482 if (wcp_file_size_change(fsp) == -1) {
483 return -1;
488 * If we used all the data then
489 * return here.
492 if(n == data_used) {
493 return n;
494 } else {
495 cache_flush_needed = True;
499 * Move the start of data forward by the amount used,
500 * cut down the amount left by the same amount.
503 data += data_used;
504 pos += data_used;
505 n -= data_used;
507 DO_PROFILE_INC(writecache_abutted_writes);
508 total_written = data_used;
510 write_path = 3;
512 } else if ( (pos >= wcp->file_size) &&
513 (n == 1) &&
514 (pos < wcp->offset + 2*wcp->alloc_size) &&
515 (wcp->file_size == wcp->offset + wcp->data_size)) {
518 +---------------+
519 | Cached data |
520 +---------------+
522 +--------+
523 | 1 Byte |
524 +--------+
526 MS-Office seems to do this a lot to determine if there's enough
527 space on the filesystem to write a new file.
530 SMB_BIG_UINT new_start = wcp->offset + wcp->data_size;
532 flush_write_cache(fsp, WRITE_FLUSH);
533 wcp->offset = new_start;
534 wcp->data_size = pos - new_start + 1;
535 memset(wcp->data, '\0', wcp->data_size);
536 memcpy(wcp->data + wcp->data_size-1, data, 1);
539 * Update the file size if changed.
542 if (wcp->offset + wcp->data_size > wcp->file_size) {
543 if (wcp_file_size_change(fsp) == -1) {
544 return -1;
548 return n;
550 } else {
552 /* ASCII art..... JRA.
554 Case 1).
556 +---------------+---------------+
557 | Cached data | Cache buffer |
558 +---------------+---------------+
560 +-------------------+
561 | Data to write |
562 +-------------------+
564 Case 2).
566 +---------------+---------------+
567 | Cached data | Cache buffer |
568 +---------------+---------------+
570 +-------------------+
571 | Data to write |
572 +-------------------+
574 Case 3).
576 +---------------+---------------+
577 | Cached data | Cache buffer |
578 +---------------+---------------+
580 +-----------------------------------------------------+
581 | Data to write |
582 +-----------------------------------------------------+
587 * Write is bigger than buffer, or there is no overlap on the
588 * low or high ends.
591 DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
592 len = %u\n",fsp->fh->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size ));
595 * If write would fit in the cache, and is larger than
596 * the data already in the cache, flush the cache and
597 * preferentially copy the data new data into it. Otherwise
598 * just write the data directly.
601 if ( n <= wcp->alloc_size && n > wcp->data_size) {
602 cache_flush_needed = True;
603 } else {
604 ssize_t ret = real_write_file(fsp, data, pos, n);
607 * If the write overlaps the entire cache, then
608 * discard the current contents of the cache.
609 * Fix from Rasmus Borup Hansen rbh@math.ku.dk.
612 if ((pos <= wcp->offset) &&
613 (pos + n >= wcp->offset + wcp->data_size) ) {
614 DEBUG(9,("write_file: discarding overwritten write \
615 cache: fd = %d, off=%.0f, size=%u\n", fsp->fh->fd, (double)wcp->offset, (unsigned int)wcp->data_size ));
616 wcp->data_size = 0;
619 DO_PROFILE_INC(writecache_direct_writes);
620 if (ret == -1) {
621 return ret;
624 if (pos + ret > wcp->file_size) {
625 wcp->file_size = pos + ret;
628 return ret;
631 write_path = 4;
635 if (cache_flush_needed) {
636 DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
637 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
638 write_path, fsp->fh->fd, (double)wcp->file_size, (double)pos, (unsigned int)n,
639 (double)wcp->offset, (unsigned int)wcp->data_size ));
641 flush_write_cache(fsp, WRITE_FLUSH);
646 * If the write request is bigger than the cache
647 * size, write it all out.
650 if (n > wcp->alloc_size ) {
651 ssize_t ret = real_write_file(fsp, data, pos, n);
652 if (ret == -1) {
653 return -1;
656 if (pos + ret > wcp->file_size) {
657 wcp->file_size = pos + n;
660 DO_PROFILE_INC(writecache_direct_writes);
661 return total_written + n;
665 * If there's any data left, cache it.
668 if (n) {
669 #ifdef WITH_PROFILE
670 if (wcp->data_size) {
671 DO_PROFILE_INC(writecache_abutted_writes);
672 } else {
673 DO_PROFILE_INC(writecache_init_writes);
675 #endif
676 memcpy(wcp->data+wcp->data_size, data, n);
677 if (wcp->data_size == 0) {
678 wcp->offset = pos;
679 DO_PROFILE_INC(writecache_num_write_caches);
681 wcp->data_size += n;
684 * Update the file size if changed.
687 if (wcp->offset + wcp->data_size > wcp->file_size) {
688 if (wcp_file_size_change(fsp) == -1) {
689 return -1;
692 DEBUG(9,("wcp->offset = %.0f wcp->data_size = %u cache return %u\n",
693 (double)wcp->offset, (unsigned int)wcp->data_size, (unsigned int)n));
695 total_written += n;
696 return total_written; /* .... that's a write :) */
699 return total_written;
702 /****************************************************************************
703 Delete the write cache structure.
704 ****************************************************************************/
706 void delete_write_cache(files_struct *fsp)
708 write_cache *wcp;
710 if(!fsp) {
711 return;
714 if(!(wcp = fsp->wcp)) {
715 return;
718 DO_PROFILE_DEC(writecache_allocated_write_caches);
719 allocated_write_caches--;
721 SMB_ASSERT(wcp->data_size == 0);
723 SAFE_FREE(wcp->data);
724 SAFE_FREE(fsp->wcp);
726 DEBUG(10,("delete_write_cache: File %s deleted write cache\n", fsp->fsp_name ));
729 /****************************************************************************
730 Setup the write cache structure.
731 ****************************************************************************/
733 static BOOL setup_write_cache(files_struct *fsp, SMB_OFF_T file_size)
735 ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
736 write_cache *wcp;
738 if (allocated_write_caches >= MAX_WRITE_CACHES) {
739 return False;
742 if(alloc_size == 0 || fsp->wcp) {
743 return False;
746 if((wcp = SMB_MALLOC_P(write_cache)) == NULL) {
747 DEBUG(0,("setup_write_cache: malloc fail.\n"));
748 return False;
751 wcp->file_size = file_size;
752 wcp->offset = 0;
753 wcp->alloc_size = alloc_size;
754 wcp->data_size = 0;
755 if((wcp->data = (char *)SMB_MALLOC(wcp->alloc_size)) == NULL) {
756 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
757 (unsigned int)wcp->alloc_size ));
758 SAFE_FREE(wcp);
759 return False;
762 memset(wcp->data, '\0', wcp->alloc_size );
764 fsp->wcp = wcp;
765 DO_PROFILE_INC(writecache_allocated_write_caches);
766 allocated_write_caches++;
768 DEBUG(10,("setup_write_cache: File %s allocated write cache size %lu\n",
769 fsp->fsp_name, (unsigned long)wcp->alloc_size ));
771 return True;
774 /****************************************************************************
775 Cope with a size change.
776 ****************************************************************************/
778 void set_filelen_write_cache(files_struct *fsp, SMB_OFF_T file_size)
780 if(fsp->wcp) {
781 /* The cache *must* have been flushed before we do this. */
782 if (fsp->wcp->data_size != 0) {
783 pstring msg;
784 slprintf(msg, sizeof(msg)-1, "set_filelen_write_cache: size change \
785 on file %s with write cache size = %lu\n", fsp->fsp_name, (unsigned long)fsp->wcp->data_size );
786 smb_panic(msg);
788 fsp->wcp->file_size = file_size;
792 /*******************************************************************
793 Flush a write cache struct to disk.
794 ********************************************************************/
796 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
798 write_cache *wcp = fsp->wcp;
799 size_t data_size;
800 ssize_t ret;
802 if(!wcp || !wcp->data_size) {
803 return 0;
806 data_size = wcp->data_size;
807 wcp->data_size = 0;
809 DO_PROFILE_DEC_INC(writecache_num_write_caches,writecache_flushed_writes[reason]);
811 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
812 fsp->fh->fd, (double)wcp->offset, (unsigned int)data_size));
814 #ifdef WITH_PROFILE
815 if(data_size == wcp->alloc_size) {
816 DO_PROFILE_INC(writecache_num_perfect_writes);
818 #endif
820 ret = real_write_file(fsp, wcp->data, wcp->offset, data_size);
823 * Ensure file size if kept up to date if write extends file.
826 if ((ret != -1) && (wcp->offset + ret > wcp->file_size)) {
827 wcp->file_size = wcp->offset + ret;
830 return ret;
833 /*******************************************************************
834 sync a file
835 ********************************************************************/
837 void sync_file(connection_struct *conn, files_struct *fsp, BOOL write_through)
839 if (fsp->fh->fd == -1)
840 return;
842 if (lp_strict_sync(SNUM(conn)) &&
843 (lp_syncalways(SNUM(conn)) || write_through)) {
844 flush_write_cache(fsp, SYNC_FLUSH);
845 SMB_VFS_FSYNC(fsp,fsp->fh->fd);
849 /************************************************************
850 Perform a stat whether a valid fd or not.
851 ************************************************************/
853 int fsp_stat(files_struct *fsp, SMB_STRUCT_STAT *pst)
855 if (fsp->fh->fd == -1) {
856 return SMB_VFS_STAT(fsp->conn, fsp->fsp_name, pst);
857 } else {
858 return SMB_VFS_FSTAT(fsp,fsp->fh->fd, pst);