2 Unix SMB/Netbios implementation.
4 read/write to a files_struct
5 Copyright (C) Andrew Tridgell 1992-1998
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 static BOOL setup_write_cache(files_struct *, SMB_OFF_T);
26 /****************************************************************************
27 seek a file. Try to avoid the seek if possible
28 ****************************************************************************/
30 SMB_OFF_T seek_file(files_struct *fsp,SMB_OFF_T pos)
35 if (fsp->print_file && lp_postscript(fsp->conn->service))
38 seek_ret = fsp->conn->vfs_ops.lseek(fsp,fsp->fd,pos+offset,SEEK_SET);
41 * We want to maintain the fiction that we can seek
42 * on a fifo for file system purposes. This allows
43 * people to set up UNIX fifo's that feed data to Windows
47 if((seek_ret == -1) && (errno == ESPIPE)) {
48 seek_ret = pos+offset;
52 if((seek_ret == -1) || (seek_ret != pos+offset)) {
53 DEBUG(0,("seek_file: sys_lseek failed. Error was %s\n", strerror(errno) ));
58 fsp->pos = seek_ret - offset;
60 DEBUG(10,("seek_file: requested pos = %.0f, new pos = %.0f\n",
61 (double)(pos+offset), (double)fsp->pos ));
66 /****************************************************************************
67 Read from write cache if we can.
68 ****************************************************************************/
71 BOOL read_from_write_cache(files_struct *fsp,char *data,SMB_OFF_T pos,size_t n)
73 write_cache *wcp = fsp->wcp;
78 if(n > wcp->data_size || pos < wcp->offset || pos + n > wcp->offset + wcp->data_size)
81 memcpy(data, wcp->data + (pos - wcp->offset), n);
83 DO_PROFILE_INC(writecache_read_hits);
88 /****************************************************************************
90 ****************************************************************************/
92 ssize_t read_file(files_struct *fsp,char *data,SMB_OFF_T pos,size_t n)
94 ssize_t ret=0,readret;
96 /* you can't read from print files */
101 * Serve from write cache if we can.
104 if(read_from_write_cache(fsp, data, pos, n))
107 flush_write_cache(fsp, READ_FLUSH);
109 if (seek_file(fsp,pos) == -1) {
110 DEBUG(3,("read_file: Failed to seek to %.0f\n",(double)pos));
118 readret = fsp->conn->vfs_ops.read(fsp,fsp->fd,data,n);
120 if ((errno == EAGAIN) && numretries) {
121 DEBUG(3,("read_file EAGAIN retry in 10 seconds\n"));
128 #else /* NO DMF fix. */
129 readret = fsp->conn->vfs_ops.read(fsp,fsp->fd,data,n);
140 /* how many write cache buffers have been allocated */
141 static unsigned int allocated_write_caches;
143 /****************************************************************************
144 *Really* write to a file.
145 ****************************************************************************/
147 static ssize_t real_write_file(files_struct *fsp,char *data,SMB_OFF_T pos, size_t n)
149 if ((pos != -1) && (seek_file(fsp,pos) == -1))
152 return vfs_write_data(fsp,data,n);
155 /****************************************************************************
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;
165 if (fsp->print_file) {
166 return print_job_write(fsp->print_jobid, data, n);
169 if (!fsp->can_write) {
174 if (!fsp->modified) {
176 fsp->modified = True;
178 if (fsp->conn->vfs_ops.fstat(fsp,fsp->fd,&st) == 0) {
179 int dosmode = dos_mode(fsp->conn,fsp->fsp_name,&st);
180 fsp->size = st.st_size;
181 if (MAP_ARCHIVE(fsp->conn) && !IS_DOS_ARCHIVE(dosmode)) {
182 file_chmod(fsp->conn,fsp->fsp_name,dosmode | aARCH,&st);
186 * If this is the first write and we have an exclusive oplock then setup
190 if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && !wcp) {
191 setup_write_cache(fsp, st.st_size);
198 DO_PROFILE_INC(writecache_total_writes);
199 if (!fsp->oplock_type) {
200 DO_PROFILE_INC(writecache_non_oplock_writes);
205 * If this file is level II oplocked then we need
206 * to grab the shared memory lock and inform all
207 * other files with a level II lock that they need
208 * to flush their read caches. We keep the lock over
209 * the shared memory area whilst doing this.
212 release_level_2_oplocks_on_change(fsp);
215 if (profile_p && profile_p->writecache_total_writes % 500 == 0) {
216 DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u total=%u \
217 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
218 profile_p->writecache_init_writes,
219 profile_p->writecache_abutted_writes,
220 profile_p->writecache_total_writes,
221 profile_p->writecache_non_oplock_writes,
222 profile_p->writecache_allocated_write_caches,
223 profile_p->writecache_num_write_caches,
224 profile_p->writecache_direct_writes,
225 profile_p->writecache_num_perfect_writes,
226 profile_p->writecache_read_hits ));
228 DEBUG(3,("WRITECACHE: Flushes SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
229 profile_p->writecache_flushed_writes[SEEK_FLUSH],
230 profile_p->writecache_flushed_writes[READ_FLUSH],
231 profile_p->writecache_flushed_writes[WRITE_FLUSH],
232 profile_p->writecache_flushed_writes[READRAW_FLUSH],
233 profile_p->writecache_flushed_writes[OPLOCK_RELEASE_FLUSH],
234 profile_p->writecache_flushed_writes[CLOSE_FLUSH],
235 profile_p->writecache_flushed_writes[SYNC_FLUSH] ));
240 DO_PROFILE_INC(writecache_direct_writes);
241 total_written = real_write_file(fsp, data, pos, n);
242 if ((total_written != -1) && (pos + total_written > fsp->size))
243 fsp->size = pos + total_written;
244 return total_written;
247 DEBUG(9,("write_file(fd=%d pos=%.0f size=%u) wcp->offset=%.0f wcp->data_size=%u\n",
248 fsp->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size));
251 * If we have active cache and it isn't contiguous then we flush.
252 * NOTE: There is a small problem with running out of disk ....
255 if (wcp->data_size) {
257 BOOL cache_flush_needed = False;
259 if ((pos >= wcp->offset) && (pos <= wcp->offset + wcp->data_size)) {
262 * Start of write overlaps or abutts the existing data.
265 size_t data_used = MIN((wcp->alloc_size - (pos - wcp->offset)), n);
267 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
270 * Update the current buffer size with the new data.
273 if(pos + data_used > wcp->offset + wcp->data_size)
274 wcp->data_size = pos + data_used - wcp->offset;
277 * Update the file size if changed.
280 if (wcp->offset + wcp->data_size > wcp->file_size)
281 fsp->size = wcp->file_size = wcp->offset + wcp->data_size;
284 * If we used all the data then
291 cache_flush_needed = True;
294 * Move the start of data forward by the amount used,
295 * cut down the amount left by the same amount.
302 DO_PROFILE_INC(writecache_abutted_writes);
303 total_written = data_used;
307 } else if ((pos < wcp->offset) && (pos + n > wcp->offset) &&
308 (pos + n <= wcp->offset + wcp->alloc_size)) {
311 * End of write overlaps the existing data.
314 size_t data_used = pos + n - wcp->offset;
316 memcpy(wcp->data, data + n - data_used, data_used);
319 * Update the current buffer size with the new data.
322 if(pos + n > wcp->offset + wcp->data_size)
323 wcp->data_size = pos + n - wcp->offset;
326 * Update the file size if changed.
329 if (wcp->offset + wcp->data_size > wcp->file_size)
330 fsp->size = wcp->file_size = wcp->offset + wcp->data_size;
333 * We don't need to move the start of data, but we
334 * cut down the amount left by the amount used.
340 * We cannot have used all the data here.
343 cache_flush_needed = True;
345 DO_PROFILE_INC(writecache_abutted_writes);
346 total_written = data_used;
350 } else if ( (pos >= wcp->file_size) &&
351 (wcp->offset + wcp->data_size == wcp->file_size) &&
352 (pos > wcp->offset + wcp->data_size) &&
353 (pos < wcp->offset + wcp->alloc_size) ) {
356 * Non-contiguous write part of which fits within
357 * the cache buffer and is extending the file
358 * and the cache contents reflect the current
359 * data up to the current end of the file.
364 if(pos + n <= wcp->offset + wcp->alloc_size)
367 data_used = wcp->offset + wcp->alloc_size - pos;
370 * Fill in the non-continuous area with zeros.
373 memset(wcp->data + wcp->data_size, '\0',
374 pos - (wcp->offset + wcp->data_size) );
376 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
379 * Update the current buffer size with the new data.
382 if(pos + data_used > wcp->offset + wcp->data_size)
383 wcp->data_size = pos + data_used - wcp->offset;
386 * Update the file size if changed.
389 if (wcp->offset + wcp->data_size > wcp->file_size)
390 fsp->size = wcp->file_size = wcp->offset + wcp->data_size;
393 * If we used all the data then
400 cache_flush_needed = True;
403 * Move the start of data forward by the amount used,
404 * cut down the amount left by the same amount.
411 DO_PROFILE_INC(writecache_abutted_writes);
412 total_written = data_used;
419 * Write is bigger than buffer, or there is no overlap on the
423 DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
424 len = %u\n",fsp->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size ));
427 * Update the file size if needed.
430 if(pos + n > wcp->file_size)
431 fsp->size = wcp->file_size = pos + n;
434 * If write would fit in the cache, and is larger than
435 * the data already in the cache, flush the cache and
436 * preferentially copy the data new data into it. Otherwise
437 * just write the data directly.
440 if ( n <= wcp->alloc_size && n > wcp->data_size) {
441 cache_flush_needed = True;
443 ssize_t ret = real_write_file(fsp, data, pos, n);
445 DO_PROFILE_INC(writecache_direct_writes);
449 if (pos + ret > wcp->file_size)
450 fsp->size = wcp->file_size = pos + ret;
459 if(wcp->data_size > wcp->file_size)
460 fsp->size = wcp->file_size = wcp->data_size;
462 if (cache_flush_needed) {
463 DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
464 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
465 write_path, fsp->fd, (double)wcp->file_size, (double)pos, (unsigned int)n,
466 (double)wcp->offset, (unsigned int)wcp->data_size ));
468 flush_write_cache(fsp, WRITE_FLUSH);
473 * If the write request is bigger than the cache
474 * size, write it all out.
477 if (n > wcp->alloc_size ) {
478 ssize_t ret = real_write_file(fsp, data, pos, n);
482 if (pos + ret > wcp->file_size)
483 fsp->size = wcp->file_size = pos + n;
485 DO_PROFILE_INC(writecache_direct_writes);
486 return total_written + n;
490 * If there's any data left, cache it.
495 if (wcp->data_size) {
496 DO_PROFILE_INC(writecache_abutted_writes);
498 DO_PROFILE_INC(writecache_init_writes);
501 memcpy(wcp->data+wcp->data_size, data, n);
502 if (wcp->data_size == 0) {
504 DO_PROFILE_INC(writecache_num_write_caches);
509 * Update the file size if changed.
512 if (wcp->offset + wcp->data_size > wcp->file_size)
513 fsp->size = wcp->file_size = wcp->offset + wcp->data_size;
514 DEBUG(9,("wcp->offset = %.0f wcp->data_size = %u cache return %u\n",
515 (double)wcp->offset, (unsigned int)wcp->data_size, (unsigned int)n));
518 return total_written; /* .... that's a write :) */
521 return total_written;
524 /****************************************************************************
525 Delete the write cache structure.
526 ****************************************************************************/
528 void delete_write_cache(files_struct *fsp)
535 if(!(wcp = fsp->wcp))
538 DO_PROFILE_DEC(writecache_allocated_write_caches);
539 allocated_write_caches--;
541 SMB_ASSERT(wcp->data_size == 0);
543 SAFE_FREE(wcp->data);
546 DEBUG(10,("delete_write_cache: File %s deleted write cache\n", fsp->fsp_name ));
550 /****************************************************************************
551 Setup the write cache structure.
552 ****************************************************************************/
554 static BOOL setup_write_cache(files_struct *fsp, SMB_OFF_T file_size)
556 ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
559 if (allocated_write_caches >= MAX_WRITE_CACHES)
562 if(alloc_size == 0 || fsp->wcp)
565 if((wcp = (write_cache *)malloc(sizeof(write_cache))) == NULL) {
566 DEBUG(0,("setup_write_cache: malloc fail.\n"));
570 wcp->file_size = file_size;
572 wcp->alloc_size = alloc_size;
574 if((wcp->data = malloc(wcp->alloc_size)) == NULL) {
575 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
576 (unsigned int)wcp->alloc_size ));
581 memset(wcp->data, '\0', wcp->alloc_size );
584 DO_PROFILE_INC(writecache_allocated_write_caches);
585 allocated_write_caches++;
587 DEBUG(10,("setup_write_cache: File %s allocated write cache size %u\n",
588 fsp->fsp_name, wcp->alloc_size ));
593 /****************************************************************************
594 Cope with a size change.
595 ****************************************************************************/
597 void set_filelen_write_cache(files_struct *fsp, SMB_OFF_T file_size)
599 fsp->size = file_size;
601 /* The cache *must* have been flushed before we do this. */
602 if (fsp->wcp->data_size != 0) {
604 slprintf(msg, sizeof(msg)-1, "set_filelen_write_cache: size change \
605 on file %s with write cache size = %u\n", fsp->fsp_name, fsp->wcp->data_size );
608 fsp->wcp->file_size = file_size;
612 /*******************************************************************
613 Flush a write cache struct to disk.
614 ********************************************************************/
616 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
618 write_cache *wcp = fsp->wcp;
622 if(!wcp || !wcp->data_size)
625 data_size = wcp->data_size;
628 DO_PROFILE_DEC_INC(writecache_num_write_caches,writecache_flushed_writes[reason]);
630 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
631 fsp->fd, (double)wcp->offset, (unsigned int)data_size));
634 if(data_size == wcp->alloc_size)
635 DO_PROFILE_INC(writecache_num_perfect_writes);
638 ret = real_write_file(fsp, wcp->data, wcp->offset, data_size);
641 * Ensure file size if kept up to date if write extends file.
644 if ((ret != -1) && (wcp->offset + ret > wcp->file_size))
645 wcp->file_size = wcp->offset + ret;
650 /*******************************************************************
652 ********************************************************************/
654 void sync_file(connection_struct *conn, files_struct *fsp)
656 if(lp_strict_sync(SNUM(conn)) && fsp->fd != -1) {
657 flush_write_cache(fsp, SYNC_FLUSH);
658 conn->vfs_ops.fsync(fsp,fsp->fd);