PCI: pci.txt fix __devexit() usage
[linux-2.6.22.y-op.git] / fs / cifs / file.c
blob07ff9351e9ee71d5586f47827f92f1f9c1f7b4ff
1 /*
2 * fs/cifs/file.c
4 * vfs operations that deal with files
5 *
6 * Copyright (C) International Business Machines Corp., 2002,2003
7 * Author(s): Steve French (sfrench@us.ibm.com)
8 * Jeremy Allison (jra@samba.org)
10 * This library is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU Lesser General Public License as published
12 * by the Free Software Foundation; either version 2.1 of the License, or
13 * (at your option) any later version.
15 * This library is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
18 * the GNU Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public License
21 * along with this library; if not, write to the Free Software
22 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
24 #include <linux/fs.h>
25 #include <linux/backing-dev.h>
26 #include <linux/stat.h>
27 #include <linux/fcntl.h>
28 #include <linux/pagemap.h>
29 #include <linux/pagevec.h>
30 #include <linux/smp_lock.h>
31 #include <linux/writeback.h>
32 #include <linux/task_io_accounting_ops.h>
33 #include <linux/delay.h>
34 #include <asm/div64.h>
35 #include "cifsfs.h"
36 #include "cifspdu.h"
37 #include "cifsglob.h"
38 #include "cifsproto.h"
39 #include "cifs_unicode.h"
40 #include "cifs_debug.h"
41 #include "cifs_fs_sb.h"
43 static inline struct cifsFileInfo *cifs_init_private(
44 struct cifsFileInfo *private_data, struct inode *inode,
45 struct file *file, __u16 netfid)
47 memset(private_data, 0, sizeof(struct cifsFileInfo));
48 private_data->netfid = netfid;
49 private_data->pid = current->tgid;
50 init_MUTEX(&private_data->fh_sem);
51 init_MUTEX(&private_data->lock_sem);
52 INIT_LIST_HEAD(&private_data->llist);
53 private_data->pfile = file; /* needed for writepage */
54 private_data->pInode = inode;
55 private_data->invalidHandle = FALSE;
56 private_data->closePend = FALSE;
57 /* we have to track num writers to the inode, since writepages
58 does not tell us which handle the write is for so there can
59 be a close (overlapping with write) of the filehandle that
60 cifs_writepages chose to use */
61 atomic_set(&private_data->wrtPending,0);
63 return private_data;
66 static inline int cifs_convert_flags(unsigned int flags)
68 if ((flags & O_ACCMODE) == O_RDONLY)
69 return GENERIC_READ;
70 else if ((flags & O_ACCMODE) == O_WRONLY)
71 return GENERIC_WRITE;
72 else if ((flags & O_ACCMODE) == O_RDWR) {
73 /* GENERIC_ALL is too much permission to request
74 can cause unnecessary access denied on create */
75 /* return GENERIC_ALL; */
76 return (GENERIC_READ | GENERIC_WRITE);
79 return 0x20197;
82 static inline int cifs_get_disposition(unsigned int flags)
84 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
85 return FILE_CREATE;
86 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
87 return FILE_OVERWRITE_IF;
88 else if ((flags & O_CREAT) == O_CREAT)
89 return FILE_OPEN_IF;
90 else if ((flags & O_TRUNC) == O_TRUNC)
91 return FILE_OVERWRITE;
92 else
93 return FILE_OPEN;
96 /* all arguments to this function must be checked for validity in caller */
97 static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
98 struct cifsInodeInfo *pCifsInode, struct cifsFileInfo *pCifsFile,
99 struct cifsTconInfo *pTcon, int *oplock, FILE_ALL_INFO *buf,
100 char *full_path, int xid)
102 struct timespec temp;
103 int rc;
105 /* want handles we can use to read with first
106 in the list so we do not have to walk the
107 list to search for one in prepare_write */
108 if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
109 list_add_tail(&pCifsFile->flist,
110 &pCifsInode->openFileList);
111 } else {
112 list_add(&pCifsFile->flist,
113 &pCifsInode->openFileList);
115 write_unlock(&GlobalSMBSeslock);
116 if (pCifsInode->clientCanCacheRead) {
117 /* we have the inode open somewhere else
118 no need to discard cache data */
119 goto client_can_cache;
122 /* BB need same check in cifs_create too? */
123 /* if not oplocked, invalidate inode pages if mtime or file
124 size changed */
125 temp = cifs_NTtimeToUnix(le64_to_cpu(buf->LastWriteTime));
126 if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
127 (file->f_path.dentry->d_inode->i_size ==
128 (loff_t)le64_to_cpu(buf->EndOfFile))) {
129 cFYI(1, ("inode unchanged on server"));
130 } else {
131 if (file->f_path.dentry->d_inode->i_mapping) {
132 /* BB no need to lock inode until after invalidate
133 since namei code should already have it locked? */
134 filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
136 cFYI(1, ("invalidating remote inode since open detected it "
137 "changed"));
138 invalidate_remote_inode(file->f_path.dentry->d_inode);
141 client_can_cache:
142 if (pTcon->ses->capabilities & CAP_UNIX)
143 rc = cifs_get_inode_info_unix(&file->f_path.dentry->d_inode,
144 full_path, inode->i_sb, xid);
145 else
146 rc = cifs_get_inode_info(&file->f_path.dentry->d_inode,
147 full_path, buf, inode->i_sb, xid);
149 if ((*oplock & 0xF) == OPLOCK_EXCLUSIVE) {
150 pCifsInode->clientCanCacheAll = TRUE;
151 pCifsInode->clientCanCacheRead = TRUE;
152 cFYI(1, ("Exclusive Oplock granted on inode %p",
153 file->f_path.dentry->d_inode));
154 } else if ((*oplock & 0xF) == OPLOCK_READ)
155 pCifsInode->clientCanCacheRead = TRUE;
157 return rc;
160 int cifs_open(struct inode *inode, struct file *file)
162 int rc = -EACCES;
163 int xid, oplock;
164 struct cifs_sb_info *cifs_sb;
165 struct cifsTconInfo *pTcon;
166 struct cifsFileInfo *pCifsFile;
167 struct cifsInodeInfo *pCifsInode;
168 struct list_head *tmp;
169 char *full_path = NULL;
170 int desiredAccess;
171 int disposition;
172 __u16 netfid;
173 FILE_ALL_INFO *buf = NULL;
175 xid = GetXid();
177 cifs_sb = CIFS_SB(inode->i_sb);
178 pTcon = cifs_sb->tcon;
180 if (file->f_flags & O_CREAT) {
181 /* search inode for this file and fill in file->private_data */
182 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
183 read_lock(&GlobalSMBSeslock);
184 list_for_each(tmp, &pCifsInode->openFileList) {
185 pCifsFile = list_entry(tmp, struct cifsFileInfo,
186 flist);
187 if ((pCifsFile->pfile == NULL) &&
188 (pCifsFile->pid == current->tgid)) {
189 /* mode set in cifs_create */
191 /* needed for writepage */
192 pCifsFile->pfile = file;
194 file->private_data = pCifsFile;
195 break;
198 read_unlock(&GlobalSMBSeslock);
199 if (file->private_data != NULL) {
200 rc = 0;
201 FreeXid(xid);
202 return rc;
203 } else {
204 if (file->f_flags & O_EXCL)
205 cERROR(1, ("could not find file instance for "
206 "new file %p", file));
210 full_path = build_path_from_dentry(file->f_path.dentry);
211 if (full_path == NULL) {
212 FreeXid(xid);
213 return -ENOMEM;
216 cFYI(1, (" inode = 0x%p file flags are 0x%x for %s",
217 inode, file->f_flags, full_path));
218 desiredAccess = cifs_convert_flags(file->f_flags);
220 /*********************************************************************
221 * open flag mapping table:
223 * POSIX Flag CIFS Disposition
224 * ---------- ----------------
225 * O_CREAT FILE_OPEN_IF
226 * O_CREAT | O_EXCL FILE_CREATE
227 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
228 * O_TRUNC FILE_OVERWRITE
229 * none of the above FILE_OPEN
231 * Note that there is not a direct match between disposition
232 * FILE_SUPERSEDE (ie create whether or not file exists although
233 * O_CREAT | O_TRUNC is similar but truncates the existing
234 * file rather than creating a new file as FILE_SUPERSEDE does
235 * (which uses the attributes / metadata passed in on open call)
237 *? O_SYNC is a reasonable match to CIFS writethrough flag
238 *? and the read write flags match reasonably. O_LARGEFILE
239 *? is irrelevant because largefile support is always used
240 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
241 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
242 *********************************************************************/
244 disposition = cifs_get_disposition(file->f_flags);
246 if (oplockEnabled)
247 oplock = REQ_OPLOCK;
248 else
249 oplock = FALSE;
251 /* BB pass O_SYNC flag through on file attributes .. BB */
253 /* Also refresh inode by passing in file_info buf returned by SMBOpen
254 and calling get_inode_info with returned buf (at least helps
255 non-Unix server case) */
257 /* BB we can not do this if this is the second open of a file
258 and the first handle has writebehind data, we might be
259 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
260 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
261 if (!buf) {
262 rc = -ENOMEM;
263 goto out;
266 if (cifs_sb->tcon->ses->capabilities & CAP_NT_SMBS)
267 rc = CIFSSMBOpen(xid, pTcon, full_path, disposition,
268 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
269 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
270 & CIFS_MOUNT_MAP_SPECIAL_CHR);
271 else
272 rc = -EIO; /* no NT SMB support fall into legacy open below */
274 if (rc == -EIO) {
275 /* Old server, try legacy style OpenX */
276 rc = SMBLegacyOpen(xid, pTcon, full_path, disposition,
277 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
278 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
279 & CIFS_MOUNT_MAP_SPECIAL_CHR);
281 if (rc) {
282 cFYI(1, ("cifs_open returned 0x%x", rc));
283 goto out;
285 file->private_data =
286 kmalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
287 if (file->private_data == NULL) {
288 rc = -ENOMEM;
289 goto out;
291 pCifsFile = cifs_init_private(file->private_data, inode, file, netfid);
292 write_lock(&GlobalSMBSeslock);
293 list_add(&pCifsFile->tlist, &pTcon->openFileList);
295 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
296 if (pCifsInode) {
297 rc = cifs_open_inode_helper(inode, file, pCifsInode,
298 pCifsFile, pTcon,
299 &oplock, buf, full_path, xid);
300 } else {
301 write_unlock(&GlobalSMBSeslock);
304 if (oplock & CIFS_CREATE_ACTION) {
305 /* time to set mode which we can not set earlier due to
306 problems creating new read-only files */
307 if (cifs_sb->tcon->ses->capabilities & CAP_UNIX) {
308 CIFSSMBUnixSetPerms(xid, pTcon, full_path,
309 inode->i_mode,
310 (__u64)-1, (__u64)-1, 0 /* dev */,
311 cifs_sb->local_nls,
312 cifs_sb->mnt_cifs_flags &
313 CIFS_MOUNT_MAP_SPECIAL_CHR);
314 } else {
315 /* BB implement via Windows security descriptors eg
316 CIFSSMBWinSetPerms(xid, pTcon, full_path, mode,
317 -1, -1, local_nls);
318 in the meantime could set r/o dos attribute when
319 perms are eg: mode & 0222 == 0 */
323 out:
324 kfree(buf);
325 kfree(full_path);
326 FreeXid(xid);
327 return rc;
330 /* Try to reacquire byte range locks that were released when session */
331 /* to server was lost */
332 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
334 int rc = 0;
336 /* BB list all locks open on this file and relock */
338 return rc;
341 static int cifs_reopen_file(struct inode *inode, struct file *file,
342 int can_flush)
344 int rc = -EACCES;
345 int xid, oplock;
346 struct cifs_sb_info *cifs_sb;
347 struct cifsTconInfo *pTcon;
348 struct cifsFileInfo *pCifsFile;
349 struct cifsInodeInfo *pCifsInode;
350 char *full_path = NULL;
351 int desiredAccess;
352 int disposition = FILE_OPEN;
353 __u16 netfid;
355 if (inode == NULL)
356 return -EBADF;
357 if (file->private_data) {
358 pCifsFile = (struct cifsFileInfo *)file->private_data;
359 } else
360 return -EBADF;
362 xid = GetXid();
363 down(&pCifsFile->fh_sem);
364 if (pCifsFile->invalidHandle == FALSE) {
365 up(&pCifsFile->fh_sem);
366 FreeXid(xid);
367 return 0;
370 if (file->f_path.dentry == NULL) {
371 up(&pCifsFile->fh_sem);
372 cFYI(1, ("failed file reopen, no valid name if dentry freed"));
373 FreeXid(xid);
374 return -EBADF;
376 cifs_sb = CIFS_SB(inode->i_sb);
377 pTcon = cifs_sb->tcon;
378 /* can not grab rename sem here because various ops, including
379 those that already have the rename sem can end up causing writepage
380 to get called and if the server was down that means we end up here,
381 and we can never tell if the caller already has the rename_sem */
382 full_path = build_path_from_dentry(file->f_path.dentry);
383 if (full_path == NULL) {
384 up(&pCifsFile->fh_sem);
385 FreeXid(xid);
386 return -ENOMEM;
389 cFYI(1, (" inode = 0x%p file flags are 0x%x for %s",
390 inode, file->f_flags,full_path));
391 desiredAccess = cifs_convert_flags(file->f_flags);
393 if (oplockEnabled)
394 oplock = REQ_OPLOCK;
395 else
396 oplock = FALSE;
398 /* Can not refresh inode by passing in file_info buf to be returned
399 by SMBOpen and then calling get_inode_info with returned buf
400 since file might have write behind data that needs to be flushed
401 and server version of file size can be stale. If we knew for sure
402 that inode was not dirty locally we could do this */
404 /* buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
405 if (buf == 0) {
406 up(&pCifsFile->fh_sem);
407 kfree(full_path);
408 FreeXid(xid);
409 return -ENOMEM;
410 } */
411 rc = CIFSSMBOpen(xid, pTcon, full_path, disposition, desiredAccess,
412 CREATE_NOT_DIR, &netfid, &oplock, NULL,
413 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
414 CIFS_MOUNT_MAP_SPECIAL_CHR);
415 if (rc) {
416 up(&pCifsFile->fh_sem);
417 cFYI(1, ("cifs_open returned 0x%x", rc));
418 cFYI(1, ("oplock: %d", oplock));
419 } else {
420 pCifsFile->netfid = netfid;
421 pCifsFile->invalidHandle = FALSE;
422 up(&pCifsFile->fh_sem);
423 pCifsInode = CIFS_I(inode);
424 if (pCifsInode) {
425 if (can_flush) {
426 filemap_write_and_wait(inode->i_mapping);
427 /* temporarily disable caching while we
428 go to server to get inode info */
429 pCifsInode->clientCanCacheAll = FALSE;
430 pCifsInode->clientCanCacheRead = FALSE;
431 if (pTcon->ses->capabilities & CAP_UNIX)
432 rc = cifs_get_inode_info_unix(&inode,
433 full_path, inode->i_sb, xid);
434 else
435 rc = cifs_get_inode_info(&inode,
436 full_path, NULL, inode->i_sb,
437 xid);
438 } /* else we are writing out data to server already
439 and could deadlock if we tried to flush data, and
440 since we do not know if we have data that would
441 invalidate the current end of file on the server
442 we can not go to the server to get the new inod
443 info */
444 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
445 pCifsInode->clientCanCacheAll = TRUE;
446 pCifsInode->clientCanCacheRead = TRUE;
447 cFYI(1, ("Exclusive Oplock granted on inode %p",
448 file->f_path.dentry->d_inode));
449 } else if ((oplock & 0xF) == OPLOCK_READ) {
450 pCifsInode->clientCanCacheRead = TRUE;
451 pCifsInode->clientCanCacheAll = FALSE;
452 } else {
453 pCifsInode->clientCanCacheRead = FALSE;
454 pCifsInode->clientCanCacheAll = FALSE;
456 cifs_relock_file(pCifsFile);
460 kfree(full_path);
461 FreeXid(xid);
462 return rc;
465 int cifs_close(struct inode *inode, struct file *file)
467 int rc = 0;
468 int xid;
469 struct cifs_sb_info *cifs_sb;
470 struct cifsTconInfo *pTcon;
471 struct cifsFileInfo *pSMBFile =
472 (struct cifsFileInfo *)file->private_data;
474 xid = GetXid();
476 cifs_sb = CIFS_SB(inode->i_sb);
477 pTcon = cifs_sb->tcon;
478 if (pSMBFile) {
479 struct cifsLockInfo *li, *tmp;
481 pSMBFile->closePend = TRUE;
482 if (pTcon) {
483 /* no sense reconnecting to close a file that is
484 already closed */
485 if (pTcon->tidStatus != CifsNeedReconnect) {
486 int timeout = 2;
487 while((atomic_read(&pSMBFile->wrtPending) != 0)
488 && (timeout < 1000) ) {
489 /* Give write a better chance to get to
490 server ahead of the close. We do not
491 want to add a wait_q here as it would
492 increase the memory utilization as
493 the struct would be in each open file,
494 but this should give enough time to
495 clear the socket */
496 #ifdef CONFIG_CIFS_DEBUG2
497 cFYI(1,("close delay, write pending"));
498 #endif /* DEBUG2 */
499 msleep(timeout);
500 timeout *= 4;
502 if(atomic_read(&pSMBFile->wrtPending))
503 cERROR(1,("close with pending writes"));
504 rc = CIFSSMBClose(xid, pTcon,
505 pSMBFile->netfid);
509 /* Delete any outstanding lock records.
510 We'll lose them when the file is closed anyway. */
511 down(&pSMBFile->lock_sem);
512 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
513 list_del(&li->llist);
514 kfree(li);
516 up(&pSMBFile->lock_sem);
518 write_lock(&GlobalSMBSeslock);
519 list_del(&pSMBFile->flist);
520 list_del(&pSMBFile->tlist);
521 write_unlock(&GlobalSMBSeslock);
522 kfree(pSMBFile->search_resume_name);
523 kfree(file->private_data);
524 file->private_data = NULL;
525 } else
526 rc = -EBADF;
528 if (list_empty(&(CIFS_I(inode)->openFileList))) {
529 cFYI(1, ("closing last open instance for inode %p", inode));
530 /* if the file is not open we do not know if we can cache info
531 on this inode, much less write behind and read ahead */
532 CIFS_I(inode)->clientCanCacheRead = FALSE;
533 CIFS_I(inode)->clientCanCacheAll = FALSE;
535 if ((rc ==0) && CIFS_I(inode)->write_behind_rc)
536 rc = CIFS_I(inode)->write_behind_rc;
537 FreeXid(xid);
538 return rc;
541 int cifs_closedir(struct inode *inode, struct file *file)
543 int rc = 0;
544 int xid;
545 struct cifsFileInfo *pCFileStruct =
546 (struct cifsFileInfo *)file->private_data;
547 char *ptmp;
549 cFYI(1, ("Closedir inode = 0x%p", inode));
551 xid = GetXid();
553 if (pCFileStruct) {
554 struct cifsTconInfo *pTcon;
555 struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
557 pTcon = cifs_sb->tcon;
559 cFYI(1, ("Freeing private data in close dir"));
560 if ((pCFileStruct->srch_inf.endOfSearch == FALSE) &&
561 (pCFileStruct->invalidHandle == FALSE)) {
562 pCFileStruct->invalidHandle = TRUE;
563 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
564 cFYI(1, ("Closing uncompleted readdir with rc %d",
565 rc));
566 /* not much we can do if it fails anyway, ignore rc */
567 rc = 0;
569 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
570 if (ptmp) {
571 cFYI(1, ("closedir free smb buf in srch struct"));
572 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
573 if(pCFileStruct->srch_inf.smallBuf)
574 cifs_small_buf_release(ptmp);
575 else
576 cifs_buf_release(ptmp);
578 ptmp = pCFileStruct->search_resume_name;
579 if (ptmp) {
580 cFYI(1, ("closedir free resume name"));
581 pCFileStruct->search_resume_name = NULL;
582 kfree(ptmp);
584 kfree(file->private_data);
585 file->private_data = NULL;
587 /* BB can we lock the filestruct while this is going on? */
588 FreeXid(xid);
589 return rc;
592 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
593 __u64 offset, __u8 lockType)
595 struct cifsLockInfo *li = kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
596 if (li == NULL)
597 return -ENOMEM;
598 li->offset = offset;
599 li->length = len;
600 li->type = lockType;
601 down(&fid->lock_sem);
602 list_add(&li->llist, &fid->llist);
603 up(&fid->lock_sem);
604 return 0;
607 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
609 int rc, xid;
610 __u32 numLock = 0;
611 __u32 numUnlock = 0;
612 __u64 length;
613 int wait_flag = FALSE;
614 struct cifs_sb_info *cifs_sb;
615 struct cifsTconInfo *pTcon;
616 __u16 netfid;
617 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
618 int posix_locking;
620 length = 1 + pfLock->fl_end - pfLock->fl_start;
621 rc = -EACCES;
622 xid = GetXid();
624 cFYI(1, ("Lock parm: 0x%x flockflags: "
625 "0x%x flocktype: 0x%x start: %lld end: %lld",
626 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
627 pfLock->fl_end));
629 if (pfLock->fl_flags & FL_POSIX)
630 cFYI(1, ("Posix"));
631 if (pfLock->fl_flags & FL_FLOCK)
632 cFYI(1, ("Flock"));
633 if (pfLock->fl_flags & FL_SLEEP) {
634 cFYI(1, ("Blocking lock"));
635 wait_flag = TRUE;
637 if (pfLock->fl_flags & FL_ACCESS)
638 cFYI(1, ("Process suspended by mandatory locking - "
639 "not implemented yet"));
640 if (pfLock->fl_flags & FL_LEASE)
641 cFYI(1, ("Lease on file - not implemented yet"));
642 if (pfLock->fl_flags &
643 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
644 cFYI(1, ("Unknown lock flags 0x%x", pfLock->fl_flags));
646 if (pfLock->fl_type == F_WRLCK) {
647 cFYI(1, ("F_WRLCK "));
648 numLock = 1;
649 } else if (pfLock->fl_type == F_UNLCK) {
650 cFYI(1, ("F_UNLCK"));
651 numUnlock = 1;
652 /* Check if unlock includes more than
653 one lock range */
654 } else if (pfLock->fl_type == F_RDLCK) {
655 cFYI(1, ("F_RDLCK"));
656 lockType |= LOCKING_ANDX_SHARED_LOCK;
657 numLock = 1;
658 } else if (pfLock->fl_type == F_EXLCK) {
659 cFYI(1, ("F_EXLCK"));
660 numLock = 1;
661 } else if (pfLock->fl_type == F_SHLCK) {
662 cFYI(1, ("F_SHLCK"));
663 lockType |= LOCKING_ANDX_SHARED_LOCK;
664 numLock = 1;
665 } else
666 cFYI(1, ("Unknown type of lock"));
668 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
669 pTcon = cifs_sb->tcon;
671 if (file->private_data == NULL) {
672 FreeXid(xid);
673 return -EBADF;
675 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
677 posix_locking = (cifs_sb->tcon->ses->capabilities & CAP_UNIX) &&
678 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(cifs_sb->tcon->fsUnixInfo.Capability));
680 /* BB add code here to normalize offset and length to
681 account for negative length which we can not accept over the
682 wire */
683 if (IS_GETLK(cmd)) {
684 if(posix_locking) {
685 int posix_lock_type;
686 if(lockType & LOCKING_ANDX_SHARED_LOCK)
687 posix_lock_type = CIFS_RDLCK;
688 else
689 posix_lock_type = CIFS_WRLCK;
690 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 1 /* get */,
691 length, pfLock,
692 posix_lock_type, wait_flag);
693 FreeXid(xid);
694 return rc;
697 /* BB we could chain these into one lock request BB */
698 rc = CIFSSMBLock(xid, pTcon, netfid, length, pfLock->fl_start,
699 0, 1, lockType, 0 /* wait flag */ );
700 if (rc == 0) {
701 rc = CIFSSMBLock(xid, pTcon, netfid, length,
702 pfLock->fl_start, 1 /* numUnlock */ ,
703 0 /* numLock */ , lockType,
704 0 /* wait flag */ );
705 pfLock->fl_type = F_UNLCK;
706 if (rc != 0)
707 cERROR(1, ("Error unlocking previously locked "
708 "range %d during test of lock", rc));
709 rc = 0;
711 } else {
712 /* if rc == ERR_SHARING_VIOLATION ? */
713 rc = 0; /* do not change lock type to unlock
714 since range in use */
717 FreeXid(xid);
718 return rc;
721 if (!numLock && !numUnlock) {
722 /* if no lock or unlock then nothing
723 to do since we do not know what it is */
724 FreeXid(xid);
725 return -EOPNOTSUPP;
728 if (posix_locking) {
729 int posix_lock_type;
730 if(lockType & LOCKING_ANDX_SHARED_LOCK)
731 posix_lock_type = CIFS_RDLCK;
732 else
733 posix_lock_type = CIFS_WRLCK;
735 if(numUnlock == 1)
736 posix_lock_type = CIFS_UNLCK;
738 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 0 /* set */,
739 length, pfLock,
740 posix_lock_type, wait_flag);
741 } else {
742 struct cifsFileInfo *fid = (struct cifsFileInfo *)file->private_data;
744 if (numLock) {
745 rc = CIFSSMBLock(xid, pTcon, netfid, length, pfLock->fl_start,
746 0, numLock, lockType, wait_flag);
748 if (rc == 0) {
749 /* For Windows locks we must store them. */
750 rc = store_file_lock(fid, length,
751 pfLock->fl_start, lockType);
753 } else if (numUnlock) {
754 /* For each stored lock that this unlock overlaps
755 completely, unlock it. */
756 int stored_rc = 0;
757 struct cifsLockInfo *li, *tmp;
759 rc = 0;
760 down(&fid->lock_sem);
761 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
762 if (pfLock->fl_start <= li->offset &&
763 length >= li->length) {
764 stored_rc = CIFSSMBLock(xid, pTcon, netfid,
765 li->length, li->offset,
766 1, 0, li->type, FALSE);
767 if (stored_rc)
768 rc = stored_rc;
770 list_del(&li->llist);
771 kfree(li);
774 up(&fid->lock_sem);
778 if (pfLock->fl_flags & FL_POSIX)
779 posix_lock_file_wait(file, pfLock);
780 FreeXid(xid);
781 return rc;
784 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
785 size_t write_size, loff_t *poffset)
787 int rc = 0;
788 unsigned int bytes_written = 0;
789 unsigned int total_written;
790 struct cifs_sb_info *cifs_sb;
791 struct cifsTconInfo *pTcon;
792 int xid, long_op;
793 struct cifsFileInfo *open_file;
795 if (file->f_path.dentry == NULL)
796 return -EBADF;
798 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
799 if (cifs_sb == NULL)
800 return -EBADF;
802 pTcon = cifs_sb->tcon;
804 /* cFYI(1,
805 (" write %d bytes to offset %lld of %s", write_size,
806 *poffset, file->f_path.dentry->d_name.name)); */
808 if (file->private_data == NULL)
809 return -EBADF;
810 else
811 open_file = (struct cifsFileInfo *) file->private_data;
813 xid = GetXid();
814 if (file->f_path.dentry->d_inode == NULL) {
815 FreeXid(xid);
816 return -EBADF;
819 if (*poffset > file->f_path.dentry->d_inode->i_size)
820 long_op = 2; /* writes past end of file can take a long time */
821 else
822 long_op = 1;
824 for (total_written = 0; write_size > total_written;
825 total_written += bytes_written) {
826 rc = -EAGAIN;
827 while (rc == -EAGAIN) {
828 if (file->private_data == NULL) {
829 /* file has been closed on us */
830 FreeXid(xid);
831 /* if we have gotten here we have written some data
832 and blocked, and the file has been freed on us while
833 we blocked so return what we managed to write */
834 return total_written;
836 if (open_file->closePend) {
837 FreeXid(xid);
838 if (total_written)
839 return total_written;
840 else
841 return -EBADF;
843 if (open_file->invalidHandle) {
844 if ((file->f_path.dentry == NULL) ||
845 (file->f_path.dentry->d_inode == NULL)) {
846 FreeXid(xid);
847 return total_written;
849 /* we could deadlock if we called
850 filemap_fdatawait from here so tell
851 reopen_file not to flush data to server
852 now */
853 rc = cifs_reopen_file(file->f_path.dentry->d_inode,
854 file, FALSE);
855 if (rc != 0)
856 break;
859 rc = CIFSSMBWrite(xid, pTcon,
860 open_file->netfid,
861 min_t(const int, cifs_sb->wsize,
862 write_size - total_written),
863 *poffset, &bytes_written,
864 NULL, write_data + total_written, long_op);
866 if (rc || (bytes_written == 0)) {
867 if (total_written)
868 break;
869 else {
870 FreeXid(xid);
871 return rc;
873 } else
874 *poffset += bytes_written;
875 long_op = FALSE; /* subsequent writes fast -
876 15 seconds is plenty */
879 cifs_stats_bytes_written(pTcon, total_written);
881 /* since the write may have blocked check these pointers again */
882 if (file->f_path.dentry) {
883 if (file->f_path.dentry->d_inode) {
884 struct inode *inode = file->f_path.dentry->d_inode;
885 inode->i_ctime = inode->i_mtime =
886 current_fs_time(inode->i_sb);
887 if (total_written > 0) {
888 if (*poffset > file->f_path.dentry->d_inode->i_size)
889 i_size_write(file->f_path.dentry->d_inode,
890 *poffset);
892 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
895 FreeXid(xid);
896 return total_written;
899 static ssize_t cifs_write(struct file *file, const char *write_data,
900 size_t write_size, loff_t *poffset)
902 int rc = 0;
903 unsigned int bytes_written = 0;
904 unsigned int total_written;
905 struct cifs_sb_info *cifs_sb;
906 struct cifsTconInfo *pTcon;
907 int xid, long_op;
908 struct cifsFileInfo *open_file;
910 if (file->f_path.dentry == NULL)
911 return -EBADF;
913 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
914 if (cifs_sb == NULL)
915 return -EBADF;
917 pTcon = cifs_sb->tcon;
919 cFYI(1,("write %zd bytes to offset %lld of %s", write_size,
920 *poffset, file->f_path.dentry->d_name.name));
922 if (file->private_data == NULL)
923 return -EBADF;
924 else
925 open_file = (struct cifsFileInfo *)file->private_data;
927 xid = GetXid();
928 if (file->f_path.dentry->d_inode == NULL) {
929 FreeXid(xid);
930 return -EBADF;
933 if (*poffset > file->f_path.dentry->d_inode->i_size)
934 long_op = 2; /* writes past end of file can take a long time */
935 else
936 long_op = 1;
938 for (total_written = 0; write_size > total_written;
939 total_written += bytes_written) {
940 rc = -EAGAIN;
941 while (rc == -EAGAIN) {
942 if (file->private_data == NULL) {
943 /* file has been closed on us */
944 FreeXid(xid);
945 /* if we have gotten here we have written some data
946 and blocked, and the file has been freed on us
947 while we blocked so return what we managed to
948 write */
949 return total_written;
951 if (open_file->closePend) {
952 FreeXid(xid);
953 if (total_written)
954 return total_written;
955 else
956 return -EBADF;
958 if (open_file->invalidHandle) {
959 if ((file->f_path.dentry == NULL) ||
960 (file->f_path.dentry->d_inode == NULL)) {
961 FreeXid(xid);
962 return total_written;
964 /* we could deadlock if we called
965 filemap_fdatawait from here so tell
966 reopen_file not to flush data to
967 server now */
968 rc = cifs_reopen_file(file->f_path.dentry->d_inode,
969 file, FALSE);
970 if (rc != 0)
971 break;
973 if(experimEnabled || (pTcon->ses->server &&
974 ((pTcon->ses->server->secMode &
975 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
976 == 0))) {
977 struct kvec iov[2];
978 unsigned int len;
980 len = min((size_t)cifs_sb->wsize,
981 write_size - total_written);
982 /* iov[0] is reserved for smb header */
983 iov[1].iov_base = (char *)write_data +
984 total_written;
985 iov[1].iov_len = len;
986 rc = CIFSSMBWrite2(xid, pTcon,
987 open_file->netfid, len,
988 *poffset, &bytes_written,
989 iov, 1, long_op);
990 } else
991 rc = CIFSSMBWrite(xid, pTcon,
992 open_file->netfid,
993 min_t(const int, cifs_sb->wsize,
994 write_size - total_written),
995 *poffset, &bytes_written,
996 write_data + total_written,
997 NULL, long_op);
999 if (rc || (bytes_written == 0)) {
1000 if (total_written)
1001 break;
1002 else {
1003 FreeXid(xid);
1004 return rc;
1006 } else
1007 *poffset += bytes_written;
1008 long_op = FALSE; /* subsequent writes fast -
1009 15 seconds is plenty */
1012 cifs_stats_bytes_written(pTcon, total_written);
1014 /* since the write may have blocked check these pointers again */
1015 if (file->f_path.dentry) {
1016 if (file->f_path.dentry->d_inode) {
1017 file->f_path.dentry->d_inode->i_ctime =
1018 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;
1019 if (total_written > 0) {
1020 if (*poffset > file->f_path.dentry->d_inode->i_size)
1021 i_size_write(file->f_path.dentry->d_inode,
1022 *poffset);
1024 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1027 FreeXid(xid);
1028 return total_written;
1031 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1033 struct cifsFileInfo *open_file;
1034 int rc;
1036 /* Having a null inode here (because mapping->host was set to zero by
1037 the VFS or MM) should not happen but we had reports of on oops (due to
1038 it being zero) during stress testcases so we need to check for it */
1040 if(cifs_inode == NULL) {
1041 cERROR(1,("Null inode passed to cifs_writeable_file"));
1042 dump_stack();
1043 return NULL;
1046 read_lock(&GlobalSMBSeslock);
1047 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1048 if (open_file->closePend)
1049 continue;
1050 if (open_file->pfile &&
1051 ((open_file->pfile->f_flags & O_RDWR) ||
1052 (open_file->pfile->f_flags & O_WRONLY))) {
1053 atomic_inc(&open_file->wrtPending);
1054 read_unlock(&GlobalSMBSeslock);
1055 if((open_file->invalidHandle) &&
1056 (!open_file->closePend) /* BB fixme -since the second clause can not be true remove it BB */) {
1057 rc = cifs_reopen_file(&cifs_inode->vfs_inode,
1058 open_file->pfile, FALSE);
1059 /* if it fails, try another handle - might be */
1060 /* dangerous to hold up writepages with retry */
1061 if(rc) {
1062 cFYI(1,("failed on reopen file in wp"));
1063 read_lock(&GlobalSMBSeslock);
1064 /* can not use this handle, no write
1065 pending on this one after all */
1066 atomic_dec
1067 (&open_file->wrtPending);
1068 continue;
1071 return open_file;
1074 read_unlock(&GlobalSMBSeslock);
1075 return NULL;
1078 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1080 struct address_space *mapping = page->mapping;
1081 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1082 char *write_data;
1083 int rc = -EFAULT;
1084 int bytes_written = 0;
1085 struct cifs_sb_info *cifs_sb;
1086 struct cifsTconInfo *pTcon;
1087 struct inode *inode;
1088 struct cifsFileInfo *open_file;
1090 if (!mapping || !mapping->host)
1091 return -EFAULT;
1093 inode = page->mapping->host;
1094 cifs_sb = CIFS_SB(inode->i_sb);
1095 pTcon = cifs_sb->tcon;
1097 offset += (loff_t)from;
1098 write_data = kmap(page);
1099 write_data += from;
1101 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1102 kunmap(page);
1103 return -EIO;
1106 /* racing with truncate? */
1107 if (offset > mapping->host->i_size) {
1108 kunmap(page);
1109 return 0; /* don't care */
1112 /* check to make sure that we are not extending the file */
1113 if (mapping->host->i_size - offset < (loff_t)to)
1114 to = (unsigned)(mapping->host->i_size - offset);
1116 open_file = find_writable_file(CIFS_I(mapping->host));
1117 if (open_file) {
1118 bytes_written = cifs_write(open_file->pfile, write_data,
1119 to-from, &offset);
1120 atomic_dec(&open_file->wrtPending);
1121 /* Does mm or vfs already set times? */
1122 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1123 if ((bytes_written > 0) && (offset)) {
1124 rc = 0;
1125 } else if (bytes_written < 0) {
1126 if (rc != -EBADF)
1127 rc = bytes_written;
1129 } else {
1130 cFYI(1, ("No writeable filehandles for inode"));
1131 rc = -EIO;
1134 kunmap(page);
1135 return rc;
1138 static int cifs_writepages(struct address_space *mapping,
1139 struct writeback_control *wbc)
1141 struct backing_dev_info *bdi = mapping->backing_dev_info;
1142 unsigned int bytes_to_write;
1143 unsigned int bytes_written;
1144 struct cifs_sb_info *cifs_sb;
1145 int done = 0;
1146 pgoff_t end;
1147 pgoff_t index;
1148 int range_whole = 0;
1149 struct kvec * iov;
1150 int len;
1151 int n_iov = 0;
1152 pgoff_t next;
1153 int nr_pages;
1154 __u64 offset = 0;
1155 struct cifsFileInfo *open_file;
1156 struct page *page;
1157 struct pagevec pvec;
1158 int rc = 0;
1159 int scanned = 0;
1160 int xid;
1162 cifs_sb = CIFS_SB(mapping->host->i_sb);
1165 * If wsize is smaller that the page cache size, default to writing
1166 * one page at a time via cifs_writepage
1168 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1169 return generic_writepages(mapping, wbc);
1171 if((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1172 if(cifs_sb->tcon->ses->server->secMode &
1173 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1174 if(!experimEnabled)
1175 return generic_writepages(mapping, wbc);
1177 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1178 if(iov == NULL)
1179 return generic_writepages(mapping, wbc);
1183 * BB: Is this meaningful for a non-block-device file system?
1184 * If it is, we should test it again after we do I/O
1186 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1187 wbc->encountered_congestion = 1;
1188 kfree(iov);
1189 return 0;
1192 xid = GetXid();
1194 pagevec_init(&pvec, 0);
1195 if (wbc->range_cyclic) {
1196 index = mapping->writeback_index; /* Start from prev offset */
1197 end = -1;
1198 } else {
1199 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1200 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1201 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1202 range_whole = 1;
1203 scanned = 1;
1205 retry:
1206 while (!done && (index <= end) &&
1207 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1208 PAGECACHE_TAG_DIRTY,
1209 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1210 int first;
1211 unsigned int i;
1213 first = -1;
1214 next = 0;
1215 n_iov = 0;
1216 bytes_to_write = 0;
1218 for (i = 0; i < nr_pages; i++) {
1219 page = pvec.pages[i];
1221 * At this point we hold neither mapping->tree_lock nor
1222 * lock on the page itself: the page may be truncated or
1223 * invalidated (changing page->mapping to NULL), or even
1224 * swizzled back from swapper_space to tmpfs file
1225 * mapping
1228 if (first < 0)
1229 lock_page(page);
1230 else if (TestSetPageLocked(page))
1231 break;
1233 if (unlikely(page->mapping != mapping)) {
1234 unlock_page(page);
1235 break;
1238 if (!wbc->range_cyclic && page->index > end) {
1239 done = 1;
1240 unlock_page(page);
1241 break;
1244 if (next && (page->index != next)) {
1245 /* Not next consecutive page */
1246 unlock_page(page);
1247 break;
1250 if (wbc->sync_mode != WB_SYNC_NONE)
1251 wait_on_page_writeback(page);
1253 if (PageWriteback(page) ||
1254 !clear_page_dirty_for_io(page)) {
1255 unlock_page(page);
1256 break;
1260 * This actually clears the dirty bit in the radix tree.
1261 * See cifs_writepage() for more commentary.
1263 set_page_writeback(page);
1265 if (page_offset(page) >= mapping->host->i_size) {
1266 done = 1;
1267 unlock_page(page);
1268 end_page_writeback(page);
1269 break;
1273 * BB can we get rid of this? pages are held by pvec
1275 page_cache_get(page);
1277 len = min(mapping->host->i_size - page_offset(page),
1278 (loff_t)PAGE_CACHE_SIZE);
1280 /* reserve iov[0] for the smb header */
1281 n_iov++;
1282 iov[n_iov].iov_base = kmap(page);
1283 iov[n_iov].iov_len = len;
1284 bytes_to_write += len;
1286 if (first < 0) {
1287 first = i;
1288 offset = page_offset(page);
1290 next = page->index + 1;
1291 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1292 break;
1294 if (n_iov) {
1295 /* Search for a writable handle every time we call
1296 * CIFSSMBWrite2. We can't rely on the last handle
1297 * we used to still be valid
1299 open_file = find_writable_file(CIFS_I(mapping->host));
1300 if (!open_file) {
1301 cERROR(1, ("No writable handles for inode"));
1302 rc = -EBADF;
1303 } else {
1304 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1305 open_file->netfid,
1306 bytes_to_write, offset,
1307 &bytes_written, iov, n_iov,
1309 atomic_dec(&open_file->wrtPending);
1310 if (rc || bytes_written < bytes_to_write) {
1311 cERROR(1,("Write2 ret %d, written = %d",
1312 rc, bytes_written));
1313 /* BB what if continued retry is
1314 requested via mount flags? */
1315 set_bit(AS_EIO, &mapping->flags);
1316 } else {
1317 cifs_stats_bytes_written(cifs_sb->tcon,
1318 bytes_written);
1321 for (i = 0; i < n_iov; i++) {
1322 page = pvec.pages[first + i];
1323 /* Should we also set page error on
1324 success rc but too little data written? */
1325 /* BB investigate retry logic on temporary
1326 server crash cases and how recovery works
1327 when page marked as error */
1328 if(rc)
1329 SetPageError(page);
1330 kunmap(page);
1331 unlock_page(page);
1332 end_page_writeback(page);
1333 page_cache_release(page);
1335 if ((wbc->nr_to_write -= n_iov) <= 0)
1336 done = 1;
1337 index = next;
1339 pagevec_release(&pvec);
1341 if (!scanned && !done) {
1343 * We hit the last page and there is more work to be done: wrap
1344 * back to the start of the file
1346 scanned = 1;
1347 index = 0;
1348 goto retry;
1350 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1351 mapping->writeback_index = index;
1353 FreeXid(xid);
1354 kfree(iov);
1355 return rc;
1358 static int cifs_writepage(struct page* page, struct writeback_control *wbc)
1360 int rc = -EFAULT;
1361 int xid;
1363 xid = GetXid();
1364 /* BB add check for wbc flags */
1365 page_cache_get(page);
1366 if (!PageUptodate(page)) {
1367 cFYI(1, ("ppw - page not up to date"));
1371 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1373 * A writepage() implementation always needs to do either this,
1374 * or re-dirty the page with "redirty_page_for_writepage()" in
1375 * the case of a failure.
1377 * Just unlocking the page will cause the radix tree tag-bits
1378 * to fail to update with the state of the page correctly.
1380 set_page_writeback(page);
1381 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1382 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1383 unlock_page(page);
1384 end_page_writeback(page);
1385 page_cache_release(page);
1386 FreeXid(xid);
1387 return rc;
1390 static int cifs_commit_write(struct file *file, struct page *page,
1391 unsigned offset, unsigned to)
1393 int xid;
1394 int rc = 0;
1395 struct inode *inode = page->mapping->host;
1396 loff_t position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1397 char *page_data;
1399 xid = GetXid();
1400 cFYI(1, ("commit write for page %p up to position %lld for %d",
1401 page, position, to));
1402 if (position > inode->i_size) {
1403 i_size_write(inode, position);
1404 /* if (file->private_data == NULL) {
1405 rc = -EBADF;
1406 } else {
1407 open_file = (struct cifsFileInfo *)file->private_data;
1408 cifs_sb = CIFS_SB(inode->i_sb);
1409 rc = -EAGAIN;
1410 while (rc == -EAGAIN) {
1411 if ((open_file->invalidHandle) &&
1412 (!open_file->closePend)) {
1413 rc = cifs_reopen_file(
1414 file->f_path.dentry->d_inode, file);
1415 if (rc != 0)
1416 break;
1418 if (!open_file->closePend) {
1419 rc = CIFSSMBSetFileSize(xid,
1420 cifs_sb->tcon, position,
1421 open_file->netfid,
1422 open_file->pid, FALSE);
1423 } else {
1424 rc = -EBADF;
1425 break;
1428 cFYI(1, (" SetEOF (commit write) rc = %d", rc));
1429 } */
1431 if (!PageUptodate(page)) {
1432 position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + offset;
1433 /* can not rely on (or let) writepage write this data */
1434 if (to < offset) {
1435 cFYI(1, ("Illegal offsets, can not copy from %d to %d",
1436 offset, to));
1437 FreeXid(xid);
1438 return rc;
1440 /* this is probably better than directly calling
1441 partialpage_write since in this function the file handle is
1442 known which we might as well leverage */
1443 /* BB check if anything else missing out of ppw
1444 such as updating last write time */
1445 page_data = kmap(page);
1446 rc = cifs_write(file, page_data + offset, to-offset,
1447 &position);
1448 if (rc > 0)
1449 rc = 0;
1450 /* else if (rc < 0) should we set writebehind rc? */
1451 kunmap(page);
1452 } else {
1453 set_page_dirty(page);
1456 FreeXid(xid);
1457 return rc;
1460 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1462 int xid;
1463 int rc = 0;
1464 struct inode *inode = file->f_path.dentry->d_inode;
1466 xid = GetXid();
1468 cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1469 dentry->d_name.name, datasync));
1471 rc = filemap_fdatawrite(inode->i_mapping);
1472 if (rc == 0)
1473 CIFS_I(inode)->write_behind_rc = 0;
1474 FreeXid(xid);
1475 return rc;
1478 /* static void cifs_sync_page(struct page *page)
1480 struct address_space *mapping;
1481 struct inode *inode;
1482 unsigned long index = page->index;
1483 unsigned int rpages = 0;
1484 int rc = 0;
1486 cFYI(1, ("sync page %p",page));
1487 mapping = page->mapping;
1488 if (!mapping)
1489 return 0;
1490 inode = mapping->host;
1491 if (!inode)
1492 return; */
1494 /* fill in rpages then
1495 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1497 /* cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1499 #if 0
1500 if (rc < 0)
1501 return rc;
1502 return 0;
1503 #endif
1504 } */
1507 * As file closes, flush all cached write data for this inode checking
1508 * for write behind errors.
1510 int cifs_flush(struct file *file, fl_owner_t id)
1512 struct inode * inode = file->f_path.dentry->d_inode;
1513 int rc = 0;
1515 /* Rather than do the steps manually:
1516 lock the inode for writing
1517 loop through pages looking for write behind data (dirty pages)
1518 coalesce into contiguous 16K (or smaller) chunks to write to server
1519 send to server (prefer in parallel)
1520 deal with writebehind errors
1521 unlock inode for writing
1522 filemapfdatawrite appears easier for the time being */
1524 rc = filemap_fdatawrite(inode->i_mapping);
1525 if (!rc) /* reset wb rc if we were able to write out dirty pages */
1526 CIFS_I(inode)->write_behind_rc = 0;
1528 cFYI(1, ("Flush inode %p file %p rc %d",inode,file,rc));
1530 return rc;
1533 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1534 size_t read_size, loff_t *poffset)
1536 int rc = -EACCES;
1537 unsigned int bytes_read = 0;
1538 unsigned int total_read = 0;
1539 unsigned int current_read_size;
1540 struct cifs_sb_info *cifs_sb;
1541 struct cifsTconInfo *pTcon;
1542 int xid;
1543 struct cifsFileInfo *open_file;
1544 char *smb_read_data;
1545 char __user *current_offset;
1546 struct smb_com_read_rsp *pSMBr;
1548 xid = GetXid();
1549 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1550 pTcon = cifs_sb->tcon;
1552 if (file->private_data == NULL) {
1553 FreeXid(xid);
1554 return -EBADF;
1556 open_file = (struct cifsFileInfo *)file->private_data;
1558 if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
1559 cFYI(1, ("attempting read on write only file instance"));
1561 for (total_read = 0, current_offset = read_data;
1562 read_size > total_read;
1563 total_read += bytes_read, current_offset += bytes_read) {
1564 current_read_size = min_t(const int, read_size - total_read,
1565 cifs_sb->rsize);
1566 rc = -EAGAIN;
1567 smb_read_data = NULL;
1568 while (rc == -EAGAIN) {
1569 int buf_type = CIFS_NO_BUFFER;
1570 if ((open_file->invalidHandle) &&
1571 (!open_file->closePend)) {
1572 rc = cifs_reopen_file(file->f_path.dentry->d_inode,
1573 file, TRUE);
1574 if (rc != 0)
1575 break;
1577 rc = CIFSSMBRead(xid, pTcon,
1578 open_file->netfid,
1579 current_read_size, *poffset,
1580 &bytes_read, &smb_read_data,
1581 &buf_type);
1582 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1583 if (smb_read_data) {
1584 if (copy_to_user(current_offset,
1585 smb_read_data +
1586 4 /* RFC1001 length field */ +
1587 le16_to_cpu(pSMBr->DataOffset),
1588 bytes_read)) {
1589 rc = -EFAULT;
1592 if(buf_type == CIFS_SMALL_BUFFER)
1593 cifs_small_buf_release(smb_read_data);
1594 else if(buf_type == CIFS_LARGE_BUFFER)
1595 cifs_buf_release(smb_read_data);
1596 smb_read_data = NULL;
1599 if (rc || (bytes_read == 0)) {
1600 if (total_read) {
1601 break;
1602 } else {
1603 FreeXid(xid);
1604 return rc;
1606 } else {
1607 cifs_stats_bytes_read(pTcon, bytes_read);
1608 *poffset += bytes_read;
1611 FreeXid(xid);
1612 return total_read;
1616 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1617 loff_t *poffset)
1619 int rc = -EACCES;
1620 unsigned int bytes_read = 0;
1621 unsigned int total_read;
1622 unsigned int current_read_size;
1623 struct cifs_sb_info *cifs_sb;
1624 struct cifsTconInfo *pTcon;
1625 int xid;
1626 char *current_offset;
1627 struct cifsFileInfo *open_file;
1628 int buf_type = CIFS_NO_BUFFER;
1630 xid = GetXid();
1631 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1632 pTcon = cifs_sb->tcon;
1634 if (file->private_data == NULL) {
1635 FreeXid(xid);
1636 return -EBADF;
1638 open_file = (struct cifsFileInfo *)file->private_data;
1640 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1641 cFYI(1, ("attempting read on write only file instance"));
1643 for (total_read = 0, current_offset = read_data;
1644 read_size > total_read;
1645 total_read += bytes_read, current_offset += bytes_read) {
1646 current_read_size = min_t(const int, read_size - total_read,
1647 cifs_sb->rsize);
1648 /* For windows me and 9x we do not want to request more
1649 than it negotiated since it will refuse the read then */
1650 if((pTcon->ses) &&
1651 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1652 current_read_size = min_t(const int, current_read_size,
1653 pTcon->ses->server->maxBuf - 128);
1655 rc = -EAGAIN;
1656 while (rc == -EAGAIN) {
1657 if ((open_file->invalidHandle) &&
1658 (!open_file->closePend)) {
1659 rc = cifs_reopen_file(file->f_path.dentry->d_inode,
1660 file, TRUE);
1661 if (rc != 0)
1662 break;
1664 rc = CIFSSMBRead(xid, pTcon,
1665 open_file->netfid,
1666 current_read_size, *poffset,
1667 &bytes_read, &current_offset,
1668 &buf_type);
1670 if (rc || (bytes_read == 0)) {
1671 if (total_read) {
1672 break;
1673 } else {
1674 FreeXid(xid);
1675 return rc;
1677 } else {
1678 cifs_stats_bytes_read(pTcon, total_read);
1679 *poffset += bytes_read;
1682 FreeXid(xid);
1683 return total_read;
1686 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1688 struct dentry *dentry = file->f_path.dentry;
1689 int rc, xid;
1691 xid = GetXid();
1692 rc = cifs_revalidate(dentry);
1693 if (rc) {
1694 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1695 FreeXid(xid);
1696 return rc;
1698 rc = generic_file_mmap(file, vma);
1699 FreeXid(xid);
1700 return rc;
1704 static void cifs_copy_cache_pages(struct address_space *mapping,
1705 struct list_head *pages, int bytes_read, char *data,
1706 struct pagevec *plru_pvec)
1708 struct page *page;
1709 char *target;
1711 while (bytes_read > 0) {
1712 if (list_empty(pages))
1713 break;
1715 page = list_entry(pages->prev, struct page, lru);
1716 list_del(&page->lru);
1718 if (add_to_page_cache(page, mapping, page->index,
1719 GFP_KERNEL)) {
1720 page_cache_release(page);
1721 cFYI(1, ("Add page cache failed"));
1722 data += PAGE_CACHE_SIZE;
1723 bytes_read -= PAGE_CACHE_SIZE;
1724 continue;
1727 target = kmap_atomic(page,KM_USER0);
1729 if (PAGE_CACHE_SIZE > bytes_read) {
1730 memcpy(target, data, bytes_read);
1731 /* zero the tail end of this partial page */
1732 memset(target + bytes_read, 0,
1733 PAGE_CACHE_SIZE - bytes_read);
1734 bytes_read = 0;
1735 } else {
1736 memcpy(target, data, PAGE_CACHE_SIZE);
1737 bytes_read -= PAGE_CACHE_SIZE;
1739 kunmap_atomic(target, KM_USER0);
1741 flush_dcache_page(page);
1742 SetPageUptodate(page);
1743 unlock_page(page);
1744 if (!pagevec_add(plru_pvec, page))
1745 __pagevec_lru_add(plru_pvec);
1746 data += PAGE_CACHE_SIZE;
1748 return;
1751 static int cifs_readpages(struct file *file, struct address_space *mapping,
1752 struct list_head *page_list, unsigned num_pages)
1754 int rc = -EACCES;
1755 int xid;
1756 loff_t offset;
1757 struct page *page;
1758 struct cifs_sb_info *cifs_sb;
1759 struct cifsTconInfo *pTcon;
1760 int bytes_read = 0;
1761 unsigned int read_size,i;
1762 char *smb_read_data = NULL;
1763 struct smb_com_read_rsp *pSMBr;
1764 struct pagevec lru_pvec;
1765 struct cifsFileInfo *open_file;
1766 int buf_type = CIFS_NO_BUFFER;
1768 xid = GetXid();
1769 if (file->private_data == NULL) {
1770 FreeXid(xid);
1771 return -EBADF;
1773 open_file = (struct cifsFileInfo *)file->private_data;
1774 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1775 pTcon = cifs_sb->tcon;
1777 pagevec_init(&lru_pvec, 0);
1779 for (i = 0; i < num_pages; ) {
1780 unsigned contig_pages;
1781 struct page *tmp_page;
1782 unsigned long expected_index;
1784 if (list_empty(page_list))
1785 break;
1787 page = list_entry(page_list->prev, struct page, lru);
1788 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1790 /* count adjacent pages that we will read into */
1791 contig_pages = 0;
1792 expected_index =
1793 list_entry(page_list->prev, struct page, lru)->index;
1794 list_for_each_entry_reverse(tmp_page,page_list,lru) {
1795 if (tmp_page->index == expected_index) {
1796 contig_pages++;
1797 expected_index++;
1798 } else
1799 break;
1801 if (contig_pages + i > num_pages)
1802 contig_pages = num_pages - i;
1804 /* for reads over a certain size could initiate async
1805 read ahead */
1807 read_size = contig_pages * PAGE_CACHE_SIZE;
1808 /* Read size needs to be in multiples of one page */
1809 read_size = min_t(const unsigned int, read_size,
1810 cifs_sb->rsize & PAGE_CACHE_MASK);
1812 rc = -EAGAIN;
1813 while (rc == -EAGAIN) {
1814 if ((open_file->invalidHandle) &&
1815 (!open_file->closePend)) {
1816 rc = cifs_reopen_file(file->f_path.dentry->d_inode,
1817 file, TRUE);
1818 if (rc != 0)
1819 break;
1822 rc = CIFSSMBRead(xid, pTcon,
1823 open_file->netfid,
1824 read_size, offset,
1825 &bytes_read, &smb_read_data,
1826 &buf_type);
1827 /* BB more RC checks ? */
1828 if (rc== -EAGAIN) {
1829 if (smb_read_data) {
1830 if(buf_type == CIFS_SMALL_BUFFER)
1831 cifs_small_buf_release(smb_read_data);
1832 else if(buf_type == CIFS_LARGE_BUFFER)
1833 cifs_buf_release(smb_read_data);
1834 smb_read_data = NULL;
1838 if ((rc < 0) || (smb_read_data == NULL)) {
1839 cFYI(1, ("Read error in readpages: %d", rc));
1840 break;
1841 } else if (bytes_read > 0) {
1842 task_io_account_read(bytes_read);
1843 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1844 cifs_copy_cache_pages(mapping, page_list, bytes_read,
1845 smb_read_data + 4 /* RFC1001 hdr */ +
1846 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
1848 i += bytes_read >> PAGE_CACHE_SHIFT;
1849 cifs_stats_bytes_read(pTcon, bytes_read);
1850 if ((int)(bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1851 i++; /* account for partial page */
1853 /* server copy of file can have smaller size
1854 than client */
1855 /* BB do we need to verify this common case ?
1856 this case is ok - if we are at server EOF
1857 we will hit it on next read */
1859 /* break; */
1861 } else {
1862 cFYI(1, ("No bytes read (%d) at offset %lld . "
1863 "Cleaning remaining pages from readahead list",
1864 bytes_read, offset));
1865 /* BB turn off caching and do new lookup on
1866 file size at server? */
1867 break;
1869 if (smb_read_data) {
1870 if(buf_type == CIFS_SMALL_BUFFER)
1871 cifs_small_buf_release(smb_read_data);
1872 else if(buf_type == CIFS_LARGE_BUFFER)
1873 cifs_buf_release(smb_read_data);
1874 smb_read_data = NULL;
1876 bytes_read = 0;
1879 pagevec_lru_add(&lru_pvec);
1881 /* need to free smb_read_data buf before exit */
1882 if (smb_read_data) {
1883 if(buf_type == CIFS_SMALL_BUFFER)
1884 cifs_small_buf_release(smb_read_data);
1885 else if(buf_type == CIFS_LARGE_BUFFER)
1886 cifs_buf_release(smb_read_data);
1887 smb_read_data = NULL;
1890 FreeXid(xid);
1891 return rc;
1894 static int cifs_readpage_worker(struct file *file, struct page *page,
1895 loff_t *poffset)
1897 char *read_data;
1898 int rc;
1900 page_cache_get(page);
1901 read_data = kmap(page);
1902 /* for reads over a certain size could initiate async read ahead */
1904 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1906 if (rc < 0)
1907 goto io_error;
1908 else
1909 cFYI(1, ("Bytes read %d",rc));
1911 file->f_path.dentry->d_inode->i_atime =
1912 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1914 if (PAGE_CACHE_SIZE > rc)
1915 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1917 flush_dcache_page(page);
1918 SetPageUptodate(page);
1919 rc = 0;
1921 io_error:
1922 kunmap(page);
1923 page_cache_release(page);
1924 return rc;
1927 static int cifs_readpage(struct file *file, struct page *page)
1929 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1930 int rc = -EACCES;
1931 int xid;
1933 xid = GetXid();
1935 if (file->private_data == NULL) {
1936 FreeXid(xid);
1937 return -EBADF;
1940 cFYI(1, ("readpage %p at offset %d 0x%x\n",
1941 page, (int)offset, (int)offset));
1943 rc = cifs_readpage_worker(file, page, &offset);
1945 unlock_page(page);
1947 FreeXid(xid);
1948 return rc;
1951 /* We do not want to update the file size from server for inodes
1952 open for write - to avoid races with writepage extending
1953 the file - in the future we could consider allowing
1954 refreshing the inode only on increases in the file size
1955 but this is tricky to do without racing with writebehind
1956 page caching in the current Linux kernel design */
1957 int is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
1959 struct cifsFileInfo *open_file = NULL;
1961 if (cifsInode)
1962 open_file = find_writable_file(cifsInode);
1964 if(open_file) {
1965 struct cifs_sb_info *cifs_sb;
1967 /* there is not actually a write pending so let
1968 this handle go free and allow it to
1969 be closable if needed */
1970 atomic_dec(&open_file->wrtPending);
1972 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
1973 if ( cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO ) {
1974 /* since no page cache to corrupt on directio
1975 we can change size safely */
1976 return 1;
1979 if(i_size_read(&cifsInode->vfs_inode) < end_of_file)
1980 return 1;
1982 return 0;
1983 } else
1984 return 1;
1987 static int cifs_prepare_write(struct file *file, struct page *page,
1988 unsigned from, unsigned to)
1990 int rc = 0;
1991 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1992 cFYI(1, ("prepare write for page %p from %d to %d",page,from,to));
1993 if (!PageUptodate(page)) {
1994 /* if (to - from != PAGE_CACHE_SIZE) {
1995 void *kaddr = kmap_atomic(page, KM_USER0);
1996 memset(kaddr, 0, from);
1997 memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
1998 flush_dcache_page(page);
1999 kunmap_atomic(kaddr, KM_USER0);
2000 } */
2001 /* If we are writing a full page it will be up to date,
2002 no need to read from the server */
2003 if ((to == PAGE_CACHE_SIZE) && (from == 0))
2004 SetPageUptodate(page);
2006 /* might as well read a page, it is fast enough */
2007 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2008 rc = cifs_readpage_worker(file, page, &offset);
2009 } else {
2010 /* should we try using another file handle if there is one -
2011 how would we lock it to prevent close of that handle
2012 racing with this read?
2013 In any case this will be written out by commit_write */
2017 /* BB should we pass any errors back?
2018 e.g. if we do not have read access to the file */
2019 return 0;
2022 const struct address_space_operations cifs_addr_ops = {
2023 .readpage = cifs_readpage,
2024 .readpages = cifs_readpages,
2025 .writepage = cifs_writepage,
2026 .writepages = cifs_writepages,
2027 .prepare_write = cifs_prepare_write,
2028 .commit_write = cifs_commit_write,
2029 .set_page_dirty = __set_page_dirty_nobuffers,
2030 /* .sync_page = cifs_sync_page, */
2031 /* .direct_IO = */
2035 * cifs_readpages requires the server to support a buffer large enough to
2036 * contain the header plus one complete page of data. Otherwise, we need
2037 * to leave cifs_readpages out of the address space operations.
2039 const struct address_space_operations cifs_addr_ops_smallbuf = {
2040 .readpage = cifs_readpage,
2041 .writepage = cifs_writepage,
2042 .writepages = cifs_writepages,
2043 .prepare_write = cifs_prepare_write,
2044 .commit_write = cifs_commit_write,
2045 .set_page_dirty = __set_page_dirty_nobuffers,
2046 /* .sync_page = cifs_sync_page, */
2047 /* .direct_IO = */