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 extern int DEBUGLEVEL;
26 static BOOL setup_write_cache(files_struct *, SMB_OFF_T);
28 /****************************************************************************
29 seek a file. Try to avoid the seek if possible
30 ****************************************************************************/
32 SMB_OFF_T seek_file(files_struct *fsp,SMB_OFF_T pos)
37 if (fsp->print_file && lp_postscript(fsp->conn->service))
40 seek_ret = fsp->conn->vfs_ops.lseek(fsp->fd,pos+offset,SEEK_SET);
43 * We want to maintain the fiction that we can seek
44 * on a fifo for file system purposes. This allows
45 * people to set up UNIX fifo's that feed data to Windows
49 if((seek_ret == -1) && (errno == ESPIPE)) {
50 seek_ret = pos+offset;
54 if((seek_ret == -1) || (seek_ret != pos+offset)) {
55 DEBUG(0,("seek_file: sys_lseek failed. Error was %s\n", strerror(errno) ));
60 fsp->pos = seek_ret - offset;
62 DEBUG(10,("seek_file: requested pos = %.0f, new pos = %.0f\n",
63 (double)(pos+offset), (double)fsp->pos ));
68 /****************************************************************************
69 Read from write cache if we can.
70 ****************************************************************************/
72 static unsigned int cache_read_hits;
74 BOOL read_from_write_cache(files_struct *fsp,char *data,SMB_OFF_T pos,size_t n)
76 write_cache *wcp = fsp->wcp;
81 if(n > wcp->data_size || pos < wcp->offset || pos + n > wcp->offset + wcp->data_size)
84 memcpy(data, wcp->data + (pos - wcp->offset), n);
91 /****************************************************************************
93 ****************************************************************************/
95 ssize_t read_file(files_struct *fsp,char *data,SMB_OFF_T pos,size_t n)
97 ssize_t ret=0,readret;
100 * Serve from write cache if we can.
102 if(read_from_write_cache(fsp, data, pos, n))
105 flush_write_cache(fsp, READ_FLUSH);
107 if (seek_file(fsp,pos) == -1) {
108 DEBUG(3,("read_file: Failed to seek to %.0f\n",(double)pos));
113 readret = fsp->conn->vfs_ops.read(fsp->fd,data,n);
114 if (readret > 0) ret += readret;
120 /* Write cache static counters. */
122 static unsigned int abutted_writes;
123 static unsigned int total_writes;
124 static unsigned int non_oplock_writes;
125 static unsigned int direct_writes;
126 static unsigned int init_writes;
127 static unsigned int flushed_writes;
128 static unsigned int num_perfect_writes;
129 static unsigned int flush_reasons[NUM_FLUSH_REASONS];
131 /* how many write cache buffers have been allocated */
132 static unsigned int allocated_write_caches;
133 static unsigned int num_write_caches;
135 /****************************************************************************
136 *Really* write to a file
137 ****************************************************************************/
139 static ssize_t real_write_file(files_struct *fsp,char *data,SMB_OFF_T pos, size_t n)
141 if ((pos != -1) && (seek_file(fsp,pos) == -1))
144 return write_data(fsp->fd,data,n);
147 /****************************************************************************
149 ****************************************************************************/
151 ssize_t write_file(files_struct *fsp, char *data, SMB_OFF_T pos, size_t n)
153 write_cache *wcp = fsp->wcp;
154 ssize_t total_written = 0;
157 if (!fsp->can_write) {
162 if (!fsp->modified) {
164 fsp->modified = True;
166 if (fsp->conn->vfs_ops.fstat(fsp->fd,&st) == 0) {
167 int dosmode = dos_mode(fsp->conn,fsp->fsp_name,&st);
168 if (MAP_ARCHIVE(fsp->conn) && !IS_DOS_ARCHIVE(dosmode)) {
169 file_chmod(fsp->conn,fsp->fsp_name,dosmode | aARCH,&st);
173 * If this is the first write and we have an exclusive oplock then setup
177 if ((fsp->oplock_type == EXCLUSIVE_OPLOCK) && !wcp) {
178 setup_write_cache(fsp, st.st_size);
185 if (!fsp->oplock_type) {
190 * If this file is level II oplocked then we need
191 * to grab the shared memory lock and inform all
192 * other files with a level II lock that they need
193 * to flush their read caches. We keep the lock over
194 * the shared memory area whilst doing this.
197 if (LEVEL_II_OPLOCK_TYPE(fsp->oplock_type)) {
198 share_mode_entry *share_list = NULL;
199 pid_t pid = getpid();
201 int num_share_modes = 0;
204 if (lock_share_entry_fsp(fsp) == False) {
205 DEBUG(0,("write_file: failed to lock share mode entry for file %s.\n", fsp->fsp_name ));
208 num_share_modes = get_share_modes(fsp->conn, fsp->dev, fsp->inode, &share_list);
210 for(i = 0; i < num_share_modes; i++) {
211 share_mode_entry *share_entry = &share_list[i];
214 * As there could have been multiple writes waiting at the lock_share_entry
215 * gate we may not be the first to enter. Hence the state of the op_types
216 * in the share mode entries may be partly NO_OPLOCK and partly LEVEL_II
217 * oplock. It will do no harm to re-send break messages to those smbd's
218 * that are still waiting their turn to remove their LEVEL_II state, and
219 * also no harm to ignore existing NO_OPLOCK states. JRA.
222 if (share_entry->op_type == NO_OPLOCK)
226 if (EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type)) {
227 DEBUG(0,("write_file: PANIC. share mode entry %d is an exlusive oplock !\n", i ));
232 * Check if this is a file we have open (including the
233 * file we've been called to do write_file on. If so
234 * then break it directly without releasing the lock.
237 if (pid == share_entry->pid) {
238 files_struct *new_fsp = file_find_dit(fsp->dev, fsp->inode, &share_entry->time);
240 /* Paranoia check... */
241 if(new_fsp == NULL) {
242 DEBUG(0,("write_file: PANIC. share mode entry %d is not a local file !\n", i ));
245 oplock_break_level2(new_fsp, True, token);
250 * This is a remote file and so we send an asynchronous
254 request_oplock_break(share_entry, fsp->dev, fsp->inode);
258 free((char *)share_list);
259 unlock_share_entry_fsp(fsp);
262 /* Paranoia check... */
263 if (LEVEL_II_OPLOCK_TYPE(fsp->oplock_type)) {
264 DEBUG(0,("write_file: PANIC. File %s still has a level II oplock.\n", fsp->fsp_name));
268 if (total_writes % 500 == 0) {
269 DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u flushes=%u total=%u \
270 nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
271 init_writes, abutted_writes, flushed_writes, total_writes, non_oplock_writes,
272 allocated_write_caches,
273 num_write_caches, direct_writes, num_perfect_writes, cache_read_hits ));
275 DEBUG(3,("WRITECACHE: SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
276 flush_reasons[SEEK_FLUSH],
277 flush_reasons[READ_FLUSH],
278 flush_reasons[WRITE_FLUSH],
279 flush_reasons[READRAW_FLUSH],
280 flush_reasons[OPLOCK_RELEASE_FLUSH],
281 flush_reasons[CLOSE_FLUSH],
282 flush_reasons[SYNC_FLUSH] ));
287 return real_write_file(fsp, data, pos, n);
290 DEBUG(9,("write_file(fd=%d pos=%d size=%d) wofs=%d wsize=%d\n",
291 fsp->fd, (int)pos, (int)n, (int)wcp->offset, (int)wcp->data_size));
294 * If we have active cache and it isn't contiguous then we flush.
295 * NOTE: There is a small problem with running out of disk ....
298 if (wcp->data_size) {
300 BOOL cache_flush_needed = False;
302 if ((pos >= wcp->offset) && (pos <= wcp->offset + wcp->data_size)) {
305 * Start of write overlaps or abutts the existing data.
308 size_t data_used = MIN((wcp->alloc_size - (pos - wcp->offset)), n);
310 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
313 * Update the current buffer size with the new data.
316 if(pos + data_used > wcp->offset + wcp->data_size)
317 wcp->data_size = pos + data_used - wcp->offset;
320 * If we used all the data then
327 cache_flush_needed = True;
330 * Move the start of data forward by the amount used,
331 * cut down the amount left by the same amount.
339 total_written = data_used;
343 } else if ((pos < wcp->offset) && (pos + n > wcp->offset) &&
344 (pos + n <= wcp->offset + wcp->alloc_size)) {
347 * End of write overlaps the existing data.
350 size_t data_used = pos + n - wcp->offset;
352 memcpy(wcp->data, data + n - data_used, data_used);
355 * Update the current buffer size with the new data.
358 if(pos + n > wcp->offset + wcp->data_size)
359 wcp->data_size = pos + n - wcp->offset;
362 * We don't need to move the start of data, but we
363 * cut down the amount left by the amount used.
369 * We cannot have used all the data here.
372 cache_flush_needed = True;
375 total_written = data_used;
379 } else if ( (pos >= wcp->file_size) &&
380 (pos > wcp->offset + wcp->data_size) &&
381 (pos < wcp->offset + wcp->alloc_size) ) {
384 * Non-contiguous write part of which fits within
385 * the cache buffer and is extending the file.
390 if(pos + n <= wcp->offset + wcp->alloc_size)
393 data_used = wcp->offset + wcp->alloc_size - pos;
396 * Fill in the non-continuous area with zeros.
399 memset(wcp->data + wcp->data_size, '\0',
400 pos - (wcp->offset + wcp->data_size) );
402 memcpy(wcp->data + (pos - wcp->offset), data, data_used);
405 * Update the current buffer size with the new data.
408 if(pos + data_used > wcp->offset + wcp->data_size)
409 wcp->data_size = pos + data_used - wcp->offset;
412 * Update the known file length.
415 wcp->file_size = wcp->offset + wcp->data_size;
418 if (set_filelen(fsp->fd, wcp->file_size) == -1) {
419 DEBUG(0,("write_file: error %s in setting file to length %.0f\n",
420 strerror(errno), (double)wcp->file_size ));
426 * If we used all the data then
433 cache_flush_needed = True;
436 * Move the start of data forward by the amount used,
437 * cut down the amount left by the same amount.
445 total_written = data_used;
452 * Write is bigger than buffer, or there is no overlap on the
456 DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
457 len = %u\n",fsp->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size ));
460 * Update the file size if needed.
463 if(pos + n > wcp->file_size)
464 wcp->file_size = pos + n;
467 * If write would fit in the cache, and is larger than
468 * the data already in the cache, flush the cache and
469 * preferentially copy the data new data into it. Otherwise
470 * just write the data directly.
473 if ( n <= wcp->alloc_size && n > wcp->data_size) {
474 cache_flush_needed = True;
477 return real_write_file(fsp, data, pos, n);
484 if(wcp->data_size > wcp->file_size)
485 wcp->file_size = wcp->data_size;
487 if (cache_flush_needed) {
490 DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
491 n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
492 write_path, fsp->fd, (double)wcp->file_size, (double)pos, (unsigned int)n,
493 (double)wcp->offset, (unsigned int)wcp->data_size ));
495 flush_write_cache(fsp, WRITE_FLUSH);
500 * If the write request is bigger than the cache
501 * size, write it all out.
504 if (n > wcp->alloc_size ) {
505 if(real_write_file(fsp, data, pos, n) == -1)
508 return total_written + n;
512 * If there's any data left, cache it.
516 if (wcp->data_size) {
518 DEBUG(9,("abutted write (%u)\n", abutted_writes));
522 memcpy(wcp->data+wcp->data_size, data, n);
523 if (wcp->data_size == 0) {
528 DEBUG(9,("cache return %u\n", (unsigned int)n));
530 return total_written; /* .... that's a write :) */
533 return total_written;
536 /****************************************************************************
537 Delete the write cache structure.
538 ****************************************************************************/
540 void delete_write_cache(files_struct *fsp)
547 if(!(wcp = fsp->wcp))
550 allocated_write_caches--;
552 SMB_ASSERT(wcp->data_size == 0);
560 /****************************************************************************
561 Setup the write cache structure.
562 ****************************************************************************/
564 static BOOL setup_write_cache(files_struct *fsp, SMB_OFF_T file_size)
566 ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
569 if (allocated_write_caches >= MAX_WRITE_CACHES) return False;
571 if(alloc_size == 0 || fsp->wcp)
574 if((wcp = (write_cache *)malloc(sizeof(write_cache))) == NULL) {
575 DEBUG(0,("setup_write_cache: malloc fail.\n"));
579 wcp->file_size = file_size;
581 wcp->alloc_size = alloc_size;
583 if((wcp->data = malloc(wcp->alloc_size)) == NULL) {
584 DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
585 (unsigned int)wcp->alloc_size ));
591 allocated_write_caches++;
596 /****************************************************************************
597 Cope with a size change.
598 ****************************************************************************/
600 void set_filelen_write_cache(files_struct *fsp, SMB_OFF_T file_size)
603 flush_write_cache(fsp, SIZECHANGE_FLUSH);
604 fsp->wcp->file_size = file_size;
608 /*******************************************************************
609 Flush a write cache struct to disk.
610 ********************************************************************/
612 ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
614 write_cache *wcp = fsp->wcp;
617 if(!wcp || !wcp->data_size)
620 data_size = wcp->data_size;
625 flush_reasons[(int)reason]++;
627 DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
628 fsp->fd, (double)wcp->offset, (unsigned int)data_size));
630 if(data_size == wcp->alloc_size)
631 num_perfect_writes++;
633 return real_write_file(fsp, wcp->data, wcp->offset, data_size);
636 /*******************************************************************
638 ********************************************************************/
640 void sys_fsync_file(connection_struct *conn, files_struct *fsp)
642 if(lp_strict_sync(SNUM(conn)) && fsp->fd != -1) {
643 flush_write_cache(fsp, SYNC_FLUSH);
644 conn->vfs_ops.fsync(fsp->fd);