add WITH_SENDFILE profiling data (from Pierre Belanger)
[Samba.git] / source / smbd / fileio.c
blob06bce01e78f5b65d810897e66ed1a5a527a5021e
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 Seek a file. Try to avoid the seek if possible.
29 ****************************************************************************/
31 SMB_OFF_T seek_file(files_struct *fsp,SMB_OFF_T pos)
33 SMB_OFF_T offset = 0;
34 SMB_OFF_T seek_ret;
36 if (fsp->print_file && lp_postscript(fsp->conn->service))
37 offset = 3;
39 seek_ret = fsp->conn->vfs_ops.lseek(fsp,fsp->fd,pos+offset,SEEK_SET);
41 if(seek_ret == -1) {
42 DEBUG(0,("seek_file: (%s) sys_lseek failed. Error was %s\n",
43 fsp->fsp_name, strerror(errno) ));
44 fsp->pos = -1;
45 return -1;
48 fsp->pos = seek_ret - offset;
50 DEBUG(10,("seek_file (%s): requested pos = %.0f, new pos = %.0f\n",
51 fsp->fsp_name, (double)(pos+offset), (double)fsp->pos ));
53 return(fsp->pos);
56 /****************************************************************************
57 Read from write cache if we can.
58 ****************************************************************************/
61 BOOL read_from_write_cache(files_struct *fsp,char *data,SMB_OFF_T pos,size_t n)
63 write_cache *wcp = fsp->wcp;
65 if(!wcp)
66 return False;
68 if(n > wcp->data_size || pos < wcp->offset || pos + n > wcp->offset + wcp->data_size)
69 return False;
71 memcpy(data, wcp->data + (pos - wcp->offset), n);
73 DO_PROFILE_INC(writecache_read_hits);
75 return True;
78 /****************************************************************************
79 Read from a file.
80 ****************************************************************************/
82 ssize_t read_file(files_struct *fsp,char *data,SMB_OFF_T pos,size_t n)
84 ssize_t ret=0,readret;
86 /* you can't read from print files */
87 if (fsp->print_file)
88 return -1;
91 * Serve from write cache if we can.
94 if(read_from_write_cache(fsp, data, pos, n))
95 return n;
97 flush_write_cache(fsp, READ_FLUSH);
99 if (seek_file(fsp,pos) == -1) {
100 DEBUG(3,("read_file: Failed to seek to %.0f\n",(double)pos));
101 return(ret);
104 if (n > 0) {
105 #ifdef DMF_FIX
106 int numretries = 3;
107 tryagain:
108 readret = fsp->conn->vfs_ops.read(fsp,fsp->fd,data,n);
109 if (readret == -1) {
110 if ((errno == EAGAIN) && numretries) {
111 DEBUG(3,("read_file EAGAIN retry in 10 seconds\n"));
112 (void)sleep(10);
113 --numretries;
114 goto tryagain;
116 return -1;
118 #else /* NO DMF fix. */
119 readret = fsp->conn->vfs_ops.read(fsp,fsp->fd,data,n);
120 if (readret == -1)
121 return -1;
122 #endif
123 if (readret > 0)
124 ret += readret;
127 DEBUG(10,("read_file (%s): pos = %.0f, size = %lu, returned %lu\n",
128 fsp->fsp_name, (double)pos, (unsigned long)n, (long)ret ));
130 return(ret);
133 /* how many write cache buffers have been allocated */
134 static unsigned int allocated_write_caches;
136 /****************************************************************************
137 *Really* write to a file.
138 ****************************************************************************/
140 static ssize_t real_write_file(files_struct *fsp,char *data,SMB_OFF_T pos, size_t n)
142 ssize_t ret;
144 if ((pos != -1) && (seek_file(fsp,pos) == -1))
145 return -1;
147 ret = vfs_write_data(fsp,data,n);
149 DEBUG(10,("real_write_file (%s): pos = %.0f, size = %lu, returned %ld\n",
150 fsp->fsp_name, (double)pos, (unsigned long)n, (long)ret ));
152 return ret;
155 /****************************************************************************
156 write to a file
157 ****************************************************************************/
159 ssize_t write_file(files_struct *fsp, char *data, SMB_OFF_T pos, size_t n)
161 write_cache *wcp = fsp->wcp;
162 ssize_t total_written = 0;
163 int write_path = -1;
165 if (fsp->print_file)
166 return print_job_write(fsp->print_jobid, data, n);
168 if (!fsp->can_write) {
169 errno = EPERM;
170 return(0);
173 if (!fsp->modified) {
174 SMB_STRUCT_STAT st;
175 fsp->modified = True;
177 if (fsp->conn->vfs_ops.fstat(fsp,fsp->fd,&st) == 0) {
178 int dosmode = dos_mode(fsp->conn,fsp->fsp_name,&st);
179 fsp->size = st.st_size;
180 if (MAP_ARCHIVE(fsp->conn) && !IS_DOS_ARCHIVE(dosmode))
181 file_chmod(fsp->conn,fsp->fsp_name,dosmode | aARCH,&st);
184 * If this is the first write and we have an exclusive oplock then setup
185 * the write cache.
188 if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && !wcp) {
189 setup_write_cache(fsp, st.st_size);
190 wcp = fsp->wcp;
195 #ifdef WITH_PROFILE
196 DO_PROFILE_INC(writecache_total_writes);
197 if (!fsp->oplock_type) {
198 DO_PROFILE_INC(writecache_non_oplock_writes);
200 #endif
203 * If this file is level II oplocked then we need
204 * to grab the shared memory lock and inform all
205 * other files with a level II lock that they need
206 * to flush their read caches. We keep the lock over
207 * the shared memory area whilst doing this.
210 release_level_2_oplocks_on_change(fsp);
212 #ifdef WITH_PROFILE
213 if (profile_p && profile_p->writecache_total_writes % 500 == 0) {
214 DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u total=%u \
215 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
216 profile_p->writecache_init_writes,
217 profile_p->writecache_abutted_writes,
218 profile_p->writecache_total_writes,
219 profile_p->writecache_non_oplock_writes,
220 profile_p->writecache_allocated_write_caches,
221 profile_p->writecache_num_write_caches,
222 profile_p->writecache_direct_writes,
223 profile_p->writecache_num_perfect_writes,
224 profile_p->writecache_read_hits ));
226 DEBUG(3,("WRITECACHE: Flushes SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
227 profile_p->writecache_flushed_writes[SEEK_FLUSH],
228 profile_p->writecache_flushed_writes[READ_FLUSH],
229 profile_p->writecache_flushed_writes[WRITE_FLUSH],
230 profile_p->writecache_flushed_writes[READRAW_FLUSH],
231 profile_p->writecache_flushed_writes[OPLOCK_RELEASE_FLUSH],
232 profile_p->writecache_flushed_writes[CLOSE_FLUSH],
233 profile_p->writecache_flushed_writes[SYNC_FLUSH] ));
235 #endif
237 if(!wcp) {
238 DO_PROFILE_INC(writecache_direct_writes);
239 total_written = real_write_file(fsp, data, pos, n);
240 if ((total_written != -1) && (pos + total_written > fsp->size))
241 fsp->size = pos + total_written;
242 return total_written;
245 DEBUG(9,("write_file (%s)(fd=%d pos=%.0f size=%u) wcp->offset=%.0f wcp->data_size=%u\n",
246 fsp->fsp_name, fsp->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size));
249 * If we have active cache and it isn't contiguous then we flush.
250 * NOTE: There is a small problem with running out of disk ....
253 if (wcp->data_size) {
255 BOOL cache_flush_needed = False;
257 if ((pos >= wcp->offset) && (pos <= wcp->offset + wcp->data_size)) {
259 /* ASCII art.... JRA.
261 +--------------+-----
262 | Cached data | Rest of allocated cache buffer....
263 +--------------+-----
265 +-------------------+
266 | Data to write |
267 +-------------------+
272 * Start of write overlaps or abutts the existing data.
275 size_t data_used = MIN((wcp->alloc_size - (pos - wcp->offset)), n);
277 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
280 * Update the current buffer size with the new data.
283 if(pos + data_used > wcp->offset + wcp->data_size)
284 wcp->data_size = pos + data_used - wcp->offset;
287 * Update the file size if changed.
290 if (wcp->offset + wcp->data_size > wcp->file_size)
291 fsp->size = wcp->file_size = wcp->offset + wcp->data_size;
294 * If we used all the data then
295 * return here.
298 if(n == data_used)
299 return n;
300 else
301 cache_flush_needed = True;
304 * Move the start of data forward by the amount used,
305 * cut down the amount left by the same amount.
308 data += data_used;
309 pos += data_used;
310 n -= data_used;
312 DO_PROFILE_INC(writecache_abutted_writes);
313 total_written = data_used;
315 write_path = 1;
317 } else if ((pos < wcp->offset) && (pos + n > wcp->offset) &&
318 (pos + n <= wcp->offset + wcp->alloc_size)) {
320 /* ASCII art.... JRA.
322 +---------------+
323 | Cache buffer |
324 +---------------+
326 +-------------------+
327 | Data to write |
328 +-------------------+
333 * End of write overlaps the existing data.
336 size_t data_used = pos + n - wcp->offset;
338 memcpy(wcp->data, data + n - data_used, data_used);
341 * Update the current buffer size with the new data.
344 if(pos + n > wcp->offset + wcp->data_size)
345 wcp->data_size = pos + n - wcp->offset;
348 * Update the file size if changed.
351 if (wcp->offset + wcp->data_size > wcp->file_size)
352 fsp->size = wcp->file_size = wcp->offset + wcp->data_size;
355 * We don't need to move the start of data, but we
356 * cut down the amount left by the amount used.
359 n -= data_used;
362 * We cannot have used all the data here.
365 cache_flush_needed = True;
367 DO_PROFILE_INC(writecache_abutted_writes);
368 total_written = data_used;
370 write_path = 2;
372 } else if ( (pos >= wcp->file_size) &&
373 (wcp->offset + wcp->data_size == wcp->file_size) &&
374 (pos > wcp->offset + wcp->data_size) &&
375 (pos < wcp->offset + wcp->alloc_size) ) {
377 /* ASCII art.... JRA.
379 End of file ---->|
381 +---------------+---------------+
382 | Cached data | Cache buffer |
383 +---------------+---------------+
385 +-------------------+
386 | Data to write |
387 +-------------------+
392 * Non-contiguous write part of which fits within
393 * the cache buffer and is extending the file
394 * and the cache contents reflect the current
395 * data up to the current end of the file.
398 size_t data_used;
400 if(pos + n <= wcp->offset + wcp->alloc_size)
401 data_used = n;
402 else
403 data_used = wcp->offset + wcp->alloc_size - pos;
406 * Fill in the non-continuous area with zeros.
409 memset(wcp->data + wcp->data_size, '\0',
410 pos - (wcp->offset + wcp->data_size) );
412 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
415 * Update the current buffer size with the new data.
418 if(pos + data_used > wcp->offset + wcp->data_size)
419 wcp->data_size = pos + data_used - wcp->offset;
422 * Update the file size if changed.
425 if (wcp->offset + wcp->data_size > wcp->file_size)
426 fsp->size = wcp->file_size = wcp->offset + wcp->data_size;
429 * If we used all the data then
430 * return here.
433 if(n == data_used)
434 return n;
435 else
436 cache_flush_needed = True;
439 * Move the start of data forward by the amount used,
440 * cut down the amount left by the same amount.
443 data += data_used;
444 pos += data_used;
445 n -= data_used;
447 DO_PROFILE_INC(writecache_abutted_writes);
448 total_written = data_used;
450 write_path = 3;
452 } else {
454 /* ASCII art..... JRA.
456 Case 1).
458 +---------------+---------------+
459 | Cached data | Cache buffer |
460 +---------------+---------------+
462 +-------------------+
463 | Data to write |
464 +-------------------+
466 Case 2).
468 +---------------+---------------+
469 | Cached data | Cache buffer |
470 +---------------+---------------+
472 +-------------------+
473 | Data to write |
474 +-------------------+
476 Case 3).
478 +---------------+---------------+
479 | Cached data | Cache buffer |
480 +---------------+---------------+
482 +-----------------------------------------------------+
483 | Data to write |
484 +-----------------------------------------------------+
489 * Write is bigger than buffer, or there is no overlap on the
490 * low or high ends.
493 DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
494 len = %u\n",fsp->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size ));
497 * Update the file size if needed.
500 if(pos + n > wcp->file_size)
501 fsp->size = wcp->file_size = pos + n;
504 * If write would fit in the cache, and is larger than
505 * the data already in the cache, flush the cache and
506 * preferentially copy the data new data into it. Otherwise
507 * just write the data directly.
510 if ( n <= wcp->alloc_size && n > wcp->data_size) {
511 cache_flush_needed = True;
512 } else {
513 ssize_t ret = real_write_file(fsp, data, pos, n);
516 * If the write overlaps the entire cache, then
517 * discard the current contents of the cache.
518 * Fix from Rasmus Borup Hansen rbh@math.ku.dk.
521 if ((pos <= wcp->offset) &&
522 (pos + n >= wcp->offset + wcp->data_size) ) {
523 DEBUG(9,("write_file: discarding overwritten write \
524 cache: fd = %d, off=%.0f, size=%u\n", fsp->fd, (double)wcp->offset, (unsigned int)wcp->data_size ));
525 wcp->data_size = 0;
528 DO_PROFILE_INC(writecache_direct_writes);
529 if (ret == -1)
530 return ret;
532 if (pos + ret > wcp->file_size)
533 fsp->size = wcp->file_size = pos + ret;
535 return ret;
538 write_path = 4;
542 if(wcp->data_size > wcp->file_size)
543 fsp->size = wcp->file_size = wcp->data_size;
545 if (cache_flush_needed) {
546 DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
547 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
548 write_path, fsp->fd, (double)wcp->file_size, (double)pos, (unsigned int)n,
549 (double)wcp->offset, (unsigned int)wcp->data_size ));
551 flush_write_cache(fsp, WRITE_FLUSH);
556 * If the write request is bigger than the cache
557 * size, write it all out.
560 if (n > wcp->alloc_size ) {
561 ssize_t ret = real_write_file(fsp, data, pos, n);
562 if (ret == -1)
563 return -1;
565 if (pos + ret > wcp->file_size)
566 fsp->size = wcp->file_size = pos + n;
568 DO_PROFILE_INC(writecache_direct_writes);
569 return total_written + n;
573 * If there's any data left, cache it.
576 if (n) {
577 #ifdef WITH_PROFILE
578 if (wcp->data_size) {
579 DO_PROFILE_INC(writecache_abutted_writes);
580 } else {
581 DO_PROFILE_INC(writecache_init_writes);
583 #endif
584 memcpy(wcp->data+wcp->data_size, data, n);
585 if (wcp->data_size == 0) {
586 wcp->offset = pos;
587 DO_PROFILE_INC(writecache_num_write_caches);
589 wcp->data_size += n;
592 * Update the file size if changed.
595 if (wcp->offset + wcp->data_size > wcp->file_size)
596 fsp->size = wcp->file_size = wcp->offset + wcp->data_size;
597 DEBUG(9,("wcp->offset = %.0f wcp->data_size = %u cache return %u\n",
598 (double)wcp->offset, (unsigned int)wcp->data_size, (unsigned int)n));
600 total_written += n;
601 return total_written; /* .... that's a write :) */
604 return total_written;
607 /****************************************************************************
608 Delete the write cache structure.
609 ****************************************************************************/
611 void delete_write_cache(files_struct *fsp)
613 write_cache *wcp;
615 if(!fsp)
616 return;
618 if(!(wcp = fsp->wcp))
619 return;
621 DO_PROFILE_DEC(writecache_allocated_write_caches);
622 allocated_write_caches--;
624 SMB_ASSERT(wcp->data_size == 0);
626 SAFE_FREE(wcp->data);
627 SAFE_FREE(fsp->wcp);
629 DEBUG(10,("delete_write_cache: File %s deleted write cache\n", fsp->fsp_name ));
632 /****************************************************************************
633 Setup the write cache structure.
634 ****************************************************************************/
636 static BOOL setup_write_cache(files_struct *fsp, SMB_OFF_T file_size)
638 ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
639 write_cache *wcp;
641 if (allocated_write_caches >= MAX_WRITE_CACHES)
642 return False;
644 if(alloc_size == 0 || fsp->wcp)
645 return False;
647 if((wcp = (write_cache *)malloc(sizeof(write_cache))) == NULL) {
648 DEBUG(0,("setup_write_cache: malloc fail.\n"));
649 return False;
652 wcp->file_size = file_size;
653 wcp->offset = 0;
654 wcp->alloc_size = alloc_size;
655 wcp->data_size = 0;
656 if((wcp->data = malloc(wcp->alloc_size)) == NULL) {
657 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
658 (unsigned int)wcp->alloc_size ));
659 SAFE_FREE(wcp);
660 return False;
663 memset(wcp->data, '\0', wcp->alloc_size );
665 fsp->wcp = wcp;
666 DO_PROFILE_INC(writecache_allocated_write_caches);
667 allocated_write_caches++;
669 DEBUG(10,("setup_write_cache: File %s allocated write cache size %u\n",
670 fsp->fsp_name, wcp->alloc_size ));
672 return True;
675 /****************************************************************************
676 Cope with a size change.
677 ****************************************************************************/
679 void set_filelen_write_cache(files_struct *fsp, SMB_OFF_T file_size)
681 fsp->size = file_size;
682 if(fsp->wcp) {
683 /* The cache *must* have been flushed before we do this. */
684 if (fsp->wcp->data_size != 0) {
685 pstring msg;
686 slprintf(msg, sizeof(msg)-1, "set_filelen_write_cache: size change \
687 on file %s with write cache size = %u\n", fsp->fsp_name, fsp->wcp->data_size );
688 smb_panic(msg);
690 fsp->wcp->file_size = file_size;
694 /*******************************************************************
695 Flush a write cache struct to disk.
696 ********************************************************************/
698 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
700 write_cache *wcp = fsp->wcp;
701 size_t data_size;
702 ssize_t ret;
704 if(!wcp || !wcp->data_size)
705 return 0;
707 data_size = wcp->data_size;
708 wcp->data_size = 0;
710 DO_PROFILE_DEC_INC(writecache_num_write_caches,writecache_flushed_writes[reason]);
712 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
713 fsp->fd, (double)wcp->offset, (unsigned int)data_size));
715 #ifdef WITH_PROFILE
716 if(data_size == wcp->alloc_size)
717 DO_PROFILE_INC(writecache_num_perfect_writes);
718 #endif
720 ret = real_write_file(fsp, wcp->data, wcp->offset, data_size);
723 * Ensure file size if kept up to date if write extends file.
726 if ((ret != -1) && (wcp->offset + ret > wcp->file_size))
727 wcp->file_size = wcp->offset + ret;
729 return ret;
732 /*******************************************************************
733 sync a file
734 ********************************************************************/
736 void sync_file(connection_struct *conn, files_struct *fsp)
738 if(lp_strict_sync(SNUM(conn)) && fsp->fd != -1) {
739 flush_write_cache(fsp, SYNC_FLUSH);
740 conn->vfs_ops.fsync(fsp,fsp->fd);
745 /************************************************************
746 Perform a stat whether a valid fd or not.
747 ************************************************************/
749 int fsp_stat(files_struct *fsp, SMB_STRUCT_STAT *pst)
751 if (fsp->fd == -1)
752 return vfs_stat(fsp->conn, fsp->fsp_name, pst);
753 else
754 return vfs_fstat(fsp,fsp->fd, pst);