4 * vfs operations that deal with files
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
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>
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);
66 static inline int cifs_convert_flags(unsigned int flags)
68 if ((flags & O_ACCMODE) == O_RDONLY)
70 else if ((flags & O_ACCMODE) == O_WRONLY)
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);
82 static inline int cifs_get_disposition(unsigned int flags)
84 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
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)
90 else if ((flags & O_TRUNC) == O_TRUNC)
91 return FILE_OVERWRITE;
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;
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);
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
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"));
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 "
138 invalidate_remote_inode(file->f_path.dentry->d_inode);
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);
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;
160 int cifs_open(struct inode *inode, struct file *file)
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;
173 FILE_ALL_INFO *buf = NULL;
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,
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;
198 read_unlock(&GlobalSMBSeslock);
199 if (file->private_data != NULL) {
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) {
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);
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);
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);
272 rc = -EIO; /* no NT SMB support fall into legacy open below */
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);
282 cFYI(1, ("cifs_open returned 0x%x", rc));
286 kmalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
287 if (file->private_data == NULL) {
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);
297 rc = cifs_open_inode_helper(inode, file, pCifsInode,
299 &oplock, buf, full_path, xid);
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,
310 (__u64)-1, (__u64)-1, 0 /* dev */,
312 cifs_sb->mnt_cifs_flags &
313 CIFS_MOUNT_MAP_SPECIAL_CHR);
315 /* BB implement via Windows security descriptors eg
316 CIFSSMBWinSetPerms(xid, pTcon, full_path, mode,
318 in the meantime could set r/o dos attribute when
319 perms are eg: mode & 0222 == 0 */
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)
336 /* BB list all locks open on this file and relock */
341 static int cifs_reopen_file(struct inode *inode, struct file *file,
346 struct cifs_sb_info *cifs_sb;
347 struct cifsTconInfo *pTcon;
348 struct cifsFileInfo *pCifsFile;
349 struct cifsInodeInfo *pCifsInode;
350 char *full_path = NULL;
352 int disposition = FILE_OPEN;
357 if (file->private_data) {
358 pCifsFile = (struct cifsFileInfo *)file->private_data;
363 down(&pCifsFile->fh_sem);
364 if (pCifsFile->invalidHandle == FALSE) {
365 up(&pCifsFile->fh_sem);
370 if (file->f_path.dentry == NULL) {
371 up(&pCifsFile->fh_sem);
372 cFYI(1, ("failed file reopen, no valid name if dentry freed"));
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);
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);
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);
406 up(&pCifsFile->fh_sem);
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);
416 up(&pCifsFile->fh_sem);
417 cFYI(1, ("cifs_open returned 0x%x", rc));
418 cFYI(1, ("oplock: %d", oplock));
420 pCifsFile->netfid = netfid;
421 pCifsFile->invalidHandle = FALSE;
422 up(&pCifsFile->fh_sem);
423 pCifsInode = CIFS_I(inode);
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);
435 rc = cifs_get_inode_info(&inode,
436 full_path, NULL, inode->i_sb,
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
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;
453 pCifsInode->clientCanCacheRead = FALSE;
454 pCifsInode->clientCanCacheAll = FALSE;
456 cifs_relock_file(pCifsFile);
465 int cifs_close(struct inode *inode, struct file *file)
469 struct cifs_sb_info *cifs_sb;
470 struct cifsTconInfo *pTcon;
471 struct cifsFileInfo *pSMBFile =
472 (struct cifsFileInfo *)file->private_data;
476 cifs_sb = CIFS_SB(inode->i_sb);
477 pTcon = cifs_sb->tcon;
479 struct cifsLockInfo *li, *tmp;
481 pSMBFile->closePend = TRUE;
483 /* no sense reconnecting to close a file that is
485 if (pTcon->tidStatus != CifsNeedReconnect) {
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
496 #ifdef CONFIG_CIFS_DEBUG2
497 cFYI(1,("close delay, write pending"));
502 if(atomic_read(&pSMBFile->wrtPending))
503 cERROR(1,("close with pending writes"));
504 rc = CIFSSMBClose(xid, pTcon,
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);
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;
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;
541 int cifs_closedir(struct inode *inode, struct file *file)
545 struct cifsFileInfo *pCFileStruct =
546 (struct cifsFileInfo *)file->private_data;
549 cFYI(1, ("Closedir inode = 0x%p", inode));
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",
566 /* not much we can do if it fails anyway, ignore rc */
569 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
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);
576 cifs_buf_release(ptmp);
578 ptmp = pCFileStruct->search_resume_name;
580 cFYI(1, ("closedir free resume name"));
581 pCFileStruct->search_resume_name = NULL;
584 kfree(file->private_data);
585 file->private_data = NULL;
587 /* BB can we lock the filestruct while this is going on? */
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);
601 down(&fid->lock_sem);
602 list_add(&li->llist, &fid->llist);
607 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
613 int wait_flag = FALSE;
614 struct cifs_sb_info *cifs_sb;
615 struct cifsTconInfo *pTcon;
617 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
620 length = 1 + pfLock->fl_end - pfLock->fl_start;
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,
629 if (pfLock->fl_flags & FL_POSIX)
631 if (pfLock->fl_flags & FL_FLOCK)
633 if (pfLock->fl_flags & FL_SLEEP) {
634 cFYI(1, ("Blocking lock"));
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 "));
649 } else if (pfLock->fl_type == F_UNLCK) {
650 cFYI(1, ("F_UNLCK"));
652 /* Check if unlock includes more than
654 } else if (pfLock->fl_type == F_RDLCK) {
655 cFYI(1, ("F_RDLCK"));
656 lockType |= LOCKING_ANDX_SHARED_LOCK;
658 } else if (pfLock->fl_type == F_EXLCK) {
659 cFYI(1, ("F_EXLCK"));
661 } else if (pfLock->fl_type == F_SHLCK) {
662 cFYI(1, ("F_SHLCK"));
663 lockType |= LOCKING_ANDX_SHARED_LOCK;
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) {
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
686 if(lockType & LOCKING_ANDX_SHARED_LOCK)
687 posix_lock_type = CIFS_RDLCK;
689 posix_lock_type = CIFS_WRLCK;
690 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 1 /* get */,
692 posix_lock_type, wait_flag);
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 */ );
701 rc = CIFSSMBLock(xid, pTcon, netfid, length,
702 pfLock->fl_start, 1 /* numUnlock */ ,
703 0 /* numLock */ , lockType,
705 pfLock->fl_type = F_UNLCK;
707 cERROR(1, ("Error unlocking previously locked "
708 "range %d during test of lock", rc));
712 /* if rc == ERR_SHARING_VIOLATION ? */
713 rc = 0; /* do not change lock type to unlock
714 since range in use */
721 if (!numLock && !numUnlock) {
722 /* if no lock or unlock then nothing
723 to do since we do not know what it is */
730 if(lockType & LOCKING_ANDX_SHARED_LOCK)
731 posix_lock_type = CIFS_RDLCK;
733 posix_lock_type = CIFS_WRLCK;
736 posix_lock_type = CIFS_UNLCK;
738 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 0 /* set */,
740 posix_lock_type, wait_flag);
742 struct cifsFileInfo *fid = (struct cifsFileInfo *)file->private_data;
745 rc = CIFSSMBLock(xid, pTcon, netfid, length, pfLock->fl_start,
746 0, numLock, lockType, wait_flag);
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. */
757 struct cifsLockInfo *li, *tmp;
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);
770 list_del(&li->llist);
778 if (pfLock->fl_flags & FL_POSIX)
779 posix_lock_file_wait(file, pfLock);
784 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
785 size_t write_size, loff_t *poffset)
788 unsigned int bytes_written = 0;
789 unsigned int total_written;
790 struct cifs_sb_info *cifs_sb;
791 struct cifsTconInfo *pTcon;
793 struct cifsFileInfo *open_file;
795 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
797 pTcon = cifs_sb->tcon;
800 (" write %d bytes to offset %lld of %s", write_size,
801 *poffset, file->f_path.dentry->d_name.name)); */
803 if (file->private_data == NULL)
805 open_file = (struct cifsFileInfo *) file->private_data;
809 if (*poffset > file->f_path.dentry->d_inode->i_size)
810 long_op = 2; /* writes past end of file can take a long time */
814 for (total_written = 0; write_size > total_written;
815 total_written += bytes_written) {
817 while (rc == -EAGAIN) {
818 if (file->private_data == NULL) {
819 /* file has been closed on us */
821 /* if we have gotten here we have written some data
822 and blocked, and the file has been freed on us while
823 we blocked so return what we managed to write */
824 return total_written;
826 if (open_file->closePend) {
829 return total_written;
833 if (open_file->invalidHandle) {
834 if ((file->f_path.dentry == NULL) ||
835 (file->f_path.dentry->d_inode == NULL)) {
837 return total_written;
839 /* we could deadlock if we called
840 filemap_fdatawait from here so tell
841 reopen_file not to flush data to server
843 rc = cifs_reopen_file(file->f_path.dentry->d_inode,
849 rc = CIFSSMBWrite(xid, pTcon,
851 min_t(const int, cifs_sb->wsize,
852 write_size - total_written),
853 *poffset, &bytes_written,
854 NULL, write_data + total_written, long_op);
856 if (rc || (bytes_written == 0)) {
864 *poffset += bytes_written;
865 long_op = FALSE; /* subsequent writes fast -
866 15 seconds is plenty */
869 cifs_stats_bytes_written(pTcon, total_written);
871 /* since the write may have blocked check these pointers again */
872 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
873 struct inode *inode = file->f_path.dentry->d_inode;
874 /* Do not update local mtime - server will set its actual value on write
875 * inode->i_ctime = inode->i_mtime =
876 * current_fs_time(inode->i_sb);*/
877 if (total_written > 0) {
878 spin_lock(&inode->i_lock);
879 if (*poffset > file->f_path.dentry->d_inode->i_size)
880 i_size_write(file->f_path.dentry->d_inode,
882 spin_unlock(&inode->i_lock);
884 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
887 return total_written;
890 static ssize_t cifs_write(struct file *file, const char *write_data,
891 size_t write_size, loff_t *poffset)
894 unsigned int bytes_written = 0;
895 unsigned int total_written;
896 struct cifs_sb_info *cifs_sb;
897 struct cifsTconInfo *pTcon;
899 struct cifsFileInfo *open_file;
901 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
903 pTcon = cifs_sb->tcon;
905 cFYI(1,("write %zd bytes to offset %lld of %s", write_size,
906 *poffset, file->f_path.dentry->d_name.name));
908 if (file->private_data == NULL)
910 open_file = (struct cifsFileInfo *)file->private_data;
914 if (*poffset > file->f_path.dentry->d_inode->i_size)
915 long_op = 2; /* writes past end of file can take a long time */
919 for (total_written = 0; write_size > total_written;
920 total_written += bytes_written) {
922 while (rc == -EAGAIN) {
923 if (file->private_data == NULL) {
924 /* file has been closed on us */
926 /* if we have gotten here we have written some data
927 and blocked, and the file has been freed on us
928 while we blocked so return what we managed to
930 return total_written;
932 if (open_file->closePend) {
935 return total_written;
939 if (open_file->invalidHandle) {
940 /* we could deadlock if we called
941 filemap_fdatawait from here so tell
942 reopen_file not to flush data to
944 rc = cifs_reopen_file(file->f_path.dentry->d_inode,
949 if(experimEnabled || (pTcon->ses->server &&
950 ((pTcon->ses->server->secMode &
951 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
956 len = min((size_t)cifs_sb->wsize,
957 write_size - total_written);
958 /* iov[0] is reserved for smb header */
959 iov[1].iov_base = (char *)write_data +
961 iov[1].iov_len = len;
962 rc = CIFSSMBWrite2(xid, pTcon,
963 open_file->netfid, len,
964 *poffset, &bytes_written,
967 rc = CIFSSMBWrite(xid, pTcon,
969 min_t(const int, cifs_sb->wsize,
970 write_size - total_written),
971 *poffset, &bytes_written,
972 write_data + total_written,
975 if (rc || (bytes_written == 0)) {
983 *poffset += bytes_written;
984 long_op = FALSE; /* subsequent writes fast -
985 15 seconds is plenty */
988 cifs_stats_bytes_written(pTcon, total_written);
990 /* since the write may have blocked check these pointers again */
991 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
992 /*BB We could make this contingent on superblock ATIME flag too */
993 /* file->f_path.dentry->d_inode->i_ctime =
994 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
995 if (total_written > 0) {
996 spin_lock(&file->f_path.dentry->d_inode->i_lock);
997 if (*poffset > file->f_path.dentry->d_inode->i_size)
998 i_size_write(file->f_path.dentry->d_inode,
1000 spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1002 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1005 return total_written;
1008 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1010 struct cifsFileInfo *open_file;
1013 /* Having a null inode here (because mapping->host was set to zero by
1014 the VFS or MM) should not happen but we had reports of on oops (due to
1015 it being zero) during stress testcases so we need to check for it */
1017 if(cifs_inode == NULL) {
1018 cERROR(1,("Null inode passed to cifs_writeable_file"));
1023 read_lock(&GlobalSMBSeslock);
1024 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1025 if (open_file->closePend)
1027 if (open_file->pfile &&
1028 ((open_file->pfile->f_flags & O_RDWR) ||
1029 (open_file->pfile->f_flags & O_WRONLY))) {
1030 atomic_inc(&open_file->wrtPending);
1031 read_unlock(&GlobalSMBSeslock);
1032 if((open_file->invalidHandle) &&
1033 (!open_file->closePend) /* BB fixme -since the second clause can not be true remove it BB */) {
1034 rc = cifs_reopen_file(&cifs_inode->vfs_inode,
1035 open_file->pfile, FALSE);
1036 /* if it fails, try another handle - might be */
1037 /* dangerous to hold up writepages with retry */
1039 cFYI(1,("failed on reopen file in wp"));
1040 read_lock(&GlobalSMBSeslock);
1041 /* can not use this handle, no write
1042 pending on this one after all */
1044 (&open_file->wrtPending);
1051 read_unlock(&GlobalSMBSeslock);
1055 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1057 struct address_space *mapping = page->mapping;
1058 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1061 int bytes_written = 0;
1062 struct cifs_sb_info *cifs_sb;
1063 struct cifsTconInfo *pTcon;
1064 struct inode *inode;
1065 struct cifsFileInfo *open_file;
1067 if (!mapping || !mapping->host)
1070 inode = page->mapping->host;
1071 cifs_sb = CIFS_SB(inode->i_sb);
1072 pTcon = cifs_sb->tcon;
1074 offset += (loff_t)from;
1075 write_data = kmap(page);
1078 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1083 /* racing with truncate? */
1084 if (offset > mapping->host->i_size) {
1086 return 0; /* don't care */
1089 /* check to make sure that we are not extending the file */
1090 if (mapping->host->i_size - offset < (loff_t)to)
1091 to = (unsigned)(mapping->host->i_size - offset);
1093 open_file = find_writable_file(CIFS_I(mapping->host));
1095 bytes_written = cifs_write(open_file->pfile, write_data,
1097 atomic_dec(&open_file->wrtPending);
1098 /* Does mm or vfs already set times? */
1099 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1100 if ((bytes_written > 0) && (offset)) {
1102 } else if (bytes_written < 0) {
1107 cFYI(1, ("No writeable filehandles for inode"));
1115 static int cifs_writepages(struct address_space *mapping,
1116 struct writeback_control *wbc)
1118 struct backing_dev_info *bdi = mapping->backing_dev_info;
1119 unsigned int bytes_to_write;
1120 unsigned int bytes_written;
1121 struct cifs_sb_info *cifs_sb;
1125 int range_whole = 0;
1132 struct cifsFileInfo *open_file;
1134 struct pagevec pvec;
1139 cifs_sb = CIFS_SB(mapping->host->i_sb);
1142 * If wsize is smaller that the page cache size, default to writing
1143 * one page at a time via cifs_writepage
1145 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1146 return generic_writepages(mapping, wbc);
1148 if((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1149 if(cifs_sb->tcon->ses->server->secMode &
1150 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1152 return generic_writepages(mapping, wbc);
1154 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1156 return generic_writepages(mapping, wbc);
1160 * BB: Is this meaningful for a non-block-device file system?
1161 * If it is, we should test it again after we do I/O
1163 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1164 wbc->encountered_congestion = 1;
1171 pagevec_init(&pvec, 0);
1172 if (wbc->range_cyclic) {
1173 index = mapping->writeback_index; /* Start from prev offset */
1176 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1177 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1178 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1183 while (!done && (index <= end) &&
1184 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1185 PAGECACHE_TAG_DIRTY,
1186 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1195 for (i = 0; i < nr_pages; i++) {
1196 page = pvec.pages[i];
1198 * At this point we hold neither mapping->tree_lock nor
1199 * lock on the page itself: the page may be truncated or
1200 * invalidated (changing page->mapping to NULL), or even
1201 * swizzled back from swapper_space to tmpfs file
1207 else if (TestSetPageLocked(page))
1210 if (unlikely(page->mapping != mapping)) {
1215 if (!wbc->range_cyclic && page->index > end) {
1221 if (next && (page->index != next)) {
1222 /* Not next consecutive page */
1227 if (wbc->sync_mode != WB_SYNC_NONE)
1228 wait_on_page_writeback(page);
1230 if (PageWriteback(page) ||
1231 !clear_page_dirty_for_io(page)) {
1237 * This actually clears the dirty bit in the radix tree.
1238 * See cifs_writepage() for more commentary.
1240 set_page_writeback(page);
1242 if (page_offset(page) >= mapping->host->i_size) {
1245 end_page_writeback(page);
1250 * BB can we get rid of this? pages are held by pvec
1252 page_cache_get(page);
1254 len = min(mapping->host->i_size - page_offset(page),
1255 (loff_t)PAGE_CACHE_SIZE);
1257 /* reserve iov[0] for the smb header */
1259 iov[n_iov].iov_base = kmap(page);
1260 iov[n_iov].iov_len = len;
1261 bytes_to_write += len;
1265 offset = page_offset(page);
1267 next = page->index + 1;
1268 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1272 /* Search for a writable handle every time we call
1273 * CIFSSMBWrite2. We can't rely on the last handle
1274 * we used to still be valid
1276 open_file = find_writable_file(CIFS_I(mapping->host));
1278 cERROR(1, ("No writable handles for inode"));
1281 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1283 bytes_to_write, offset,
1284 &bytes_written, iov, n_iov,
1286 atomic_dec(&open_file->wrtPending);
1287 if (rc || bytes_written < bytes_to_write) {
1288 cERROR(1,("Write2 ret %d, written = %d",
1289 rc, bytes_written));
1290 /* BB what if continued retry is
1291 requested via mount flags? */
1292 set_bit(AS_EIO, &mapping->flags);
1294 cifs_stats_bytes_written(cifs_sb->tcon,
1298 for (i = 0; i < n_iov; i++) {
1299 page = pvec.pages[first + i];
1300 /* Should we also set page error on
1301 success rc but too little data written? */
1302 /* BB investigate retry logic on temporary
1303 server crash cases and how recovery works
1304 when page marked as error */
1309 end_page_writeback(page);
1310 page_cache_release(page);
1312 if ((wbc->nr_to_write -= n_iov) <= 0)
1316 pagevec_release(&pvec);
1318 if (!scanned && !done) {
1320 * We hit the last page and there is more work to be done: wrap
1321 * back to the start of the file
1327 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1328 mapping->writeback_index = index;
1335 static int cifs_writepage(struct page* page, struct writeback_control *wbc)
1341 /* BB add check for wbc flags */
1342 page_cache_get(page);
1343 if (!PageUptodate(page)) {
1344 cFYI(1, ("ppw - page not up to date"));
1348 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1350 * A writepage() implementation always needs to do either this,
1351 * or re-dirty the page with "redirty_page_for_writepage()" in
1352 * the case of a failure.
1354 * Just unlocking the page will cause the radix tree tag-bits
1355 * to fail to update with the state of the page correctly.
1357 set_page_writeback(page);
1358 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1359 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1361 end_page_writeback(page);
1362 page_cache_release(page);
1367 static int cifs_commit_write(struct file *file, struct page *page,
1368 unsigned offset, unsigned to)
1372 struct inode *inode = page->mapping->host;
1373 loff_t position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1377 cFYI(1, ("commit write for page %p up to position %lld for %d",
1378 page, position, to));
1379 spin_lock(&inode->i_lock);
1380 if (position > inode->i_size) {
1381 i_size_write(inode, position);
1382 /* if (file->private_data == NULL) {
1385 open_file = (struct cifsFileInfo *)file->private_data;
1386 cifs_sb = CIFS_SB(inode->i_sb);
1388 while (rc == -EAGAIN) {
1389 if ((open_file->invalidHandle) &&
1390 (!open_file->closePend)) {
1391 rc = cifs_reopen_file(
1392 file->f_path.dentry->d_inode, file);
1396 if (!open_file->closePend) {
1397 rc = CIFSSMBSetFileSize(xid,
1398 cifs_sb->tcon, position,
1400 open_file->pid, FALSE);
1406 cFYI(1, (" SetEOF (commit write) rc = %d", rc));
1409 spin_unlock(&inode->i_lock);
1410 if (!PageUptodate(page)) {
1411 position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + offset;
1412 /* can not rely on (or let) writepage write this data */
1414 cFYI(1, ("Illegal offsets, can not copy from %d to %d",
1419 /* this is probably better than directly calling
1420 partialpage_write since in this function the file handle is
1421 known which we might as well leverage */
1422 /* BB check if anything else missing out of ppw
1423 such as updating last write time */
1424 page_data = kmap(page);
1425 rc = cifs_write(file, page_data + offset, to-offset,
1429 /* else if (rc < 0) should we set writebehind rc? */
1432 set_page_dirty(page);
1439 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1443 struct inode *inode = file->f_path.dentry->d_inode;
1447 cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1448 dentry->d_name.name, datasync));
1450 rc = filemap_fdatawrite(inode->i_mapping);
1452 CIFS_I(inode)->write_behind_rc = 0;
1457 /* static void cifs_sync_page(struct page *page)
1459 struct address_space *mapping;
1460 struct inode *inode;
1461 unsigned long index = page->index;
1462 unsigned int rpages = 0;
1465 cFYI(1, ("sync page %p",page));
1466 mapping = page->mapping;
1469 inode = mapping->host;
1473 /* fill in rpages then
1474 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1476 /* cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1486 * As file closes, flush all cached write data for this inode checking
1487 * for write behind errors.
1489 int cifs_flush(struct file *file, fl_owner_t id)
1491 struct inode * inode = file->f_path.dentry->d_inode;
1494 /* Rather than do the steps manually:
1495 lock the inode for writing
1496 loop through pages looking for write behind data (dirty pages)
1497 coalesce into contiguous 16K (or smaller) chunks to write to server
1498 send to server (prefer in parallel)
1499 deal with writebehind errors
1500 unlock inode for writing
1501 filemapfdatawrite appears easier for the time being */
1503 rc = filemap_fdatawrite(inode->i_mapping);
1504 if (!rc) /* reset wb rc if we were able to write out dirty pages */
1505 CIFS_I(inode)->write_behind_rc = 0;
1507 cFYI(1, ("Flush inode %p file %p rc %d",inode,file,rc));
1512 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1513 size_t read_size, loff_t *poffset)
1516 unsigned int bytes_read = 0;
1517 unsigned int total_read = 0;
1518 unsigned int current_read_size;
1519 struct cifs_sb_info *cifs_sb;
1520 struct cifsTconInfo *pTcon;
1522 struct cifsFileInfo *open_file;
1523 char *smb_read_data;
1524 char __user *current_offset;
1525 struct smb_com_read_rsp *pSMBr;
1528 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1529 pTcon = cifs_sb->tcon;
1531 if (file->private_data == NULL) {
1535 open_file = (struct cifsFileInfo *)file->private_data;
1537 if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
1538 cFYI(1, ("attempting read on write only file instance"));
1540 for (total_read = 0, current_offset = read_data;
1541 read_size > total_read;
1542 total_read += bytes_read, current_offset += bytes_read) {
1543 current_read_size = min_t(const int, read_size - total_read,
1546 smb_read_data = NULL;
1547 while (rc == -EAGAIN) {
1548 int buf_type = CIFS_NO_BUFFER;
1549 if ((open_file->invalidHandle) &&
1550 (!open_file->closePend)) {
1551 rc = cifs_reopen_file(file->f_path.dentry->d_inode,
1556 rc = CIFSSMBRead(xid, pTcon,
1558 current_read_size, *poffset,
1559 &bytes_read, &smb_read_data,
1561 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1562 if (smb_read_data) {
1563 if (copy_to_user(current_offset,
1565 4 /* RFC1001 length field */ +
1566 le16_to_cpu(pSMBr->DataOffset),
1571 if(buf_type == CIFS_SMALL_BUFFER)
1572 cifs_small_buf_release(smb_read_data);
1573 else if(buf_type == CIFS_LARGE_BUFFER)
1574 cifs_buf_release(smb_read_data);
1575 smb_read_data = NULL;
1578 if (rc || (bytes_read == 0)) {
1586 cifs_stats_bytes_read(pTcon, bytes_read);
1587 *poffset += bytes_read;
1595 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1599 unsigned int bytes_read = 0;
1600 unsigned int total_read;
1601 unsigned int current_read_size;
1602 struct cifs_sb_info *cifs_sb;
1603 struct cifsTconInfo *pTcon;
1605 char *current_offset;
1606 struct cifsFileInfo *open_file;
1607 int buf_type = CIFS_NO_BUFFER;
1610 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1611 pTcon = cifs_sb->tcon;
1613 if (file->private_data == NULL) {
1617 open_file = (struct cifsFileInfo *)file->private_data;
1619 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1620 cFYI(1, ("attempting read on write only file instance"));
1622 for (total_read = 0, current_offset = read_data;
1623 read_size > total_read;
1624 total_read += bytes_read, current_offset += bytes_read) {
1625 current_read_size = min_t(const int, read_size - total_read,
1627 /* For windows me and 9x we do not want to request more
1628 than it negotiated since it will refuse the read then */
1630 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1631 current_read_size = min_t(const int, current_read_size,
1632 pTcon->ses->server->maxBuf - 128);
1635 while (rc == -EAGAIN) {
1636 if ((open_file->invalidHandle) &&
1637 (!open_file->closePend)) {
1638 rc = cifs_reopen_file(file->f_path.dentry->d_inode,
1643 rc = CIFSSMBRead(xid, pTcon,
1645 current_read_size, *poffset,
1646 &bytes_read, ¤t_offset,
1649 if (rc || (bytes_read == 0)) {
1657 cifs_stats_bytes_read(pTcon, total_read);
1658 *poffset += bytes_read;
1665 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1667 struct dentry *dentry = file->f_path.dentry;
1671 rc = cifs_revalidate(dentry);
1673 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1677 rc = generic_file_mmap(file, vma);
1683 static void cifs_copy_cache_pages(struct address_space *mapping,
1684 struct list_head *pages, int bytes_read, char *data,
1685 struct pagevec *plru_pvec)
1690 while (bytes_read > 0) {
1691 if (list_empty(pages))
1694 page = list_entry(pages->prev, struct page, lru);
1695 list_del(&page->lru);
1697 if (add_to_page_cache(page, mapping, page->index,
1699 page_cache_release(page);
1700 cFYI(1, ("Add page cache failed"));
1701 data += PAGE_CACHE_SIZE;
1702 bytes_read -= PAGE_CACHE_SIZE;
1706 target = kmap_atomic(page,KM_USER0);
1708 if (PAGE_CACHE_SIZE > bytes_read) {
1709 memcpy(target, data, bytes_read);
1710 /* zero the tail end of this partial page */
1711 memset(target + bytes_read, 0,
1712 PAGE_CACHE_SIZE - bytes_read);
1715 memcpy(target, data, PAGE_CACHE_SIZE);
1716 bytes_read -= PAGE_CACHE_SIZE;
1718 kunmap_atomic(target, KM_USER0);
1720 flush_dcache_page(page);
1721 SetPageUptodate(page);
1723 if (!pagevec_add(plru_pvec, page))
1724 __pagevec_lru_add(plru_pvec);
1725 data += PAGE_CACHE_SIZE;
1730 static int cifs_readpages(struct file *file, struct address_space *mapping,
1731 struct list_head *page_list, unsigned num_pages)
1737 struct cifs_sb_info *cifs_sb;
1738 struct cifsTconInfo *pTcon;
1740 unsigned int read_size,i;
1741 char *smb_read_data = NULL;
1742 struct smb_com_read_rsp *pSMBr;
1743 struct pagevec lru_pvec;
1744 struct cifsFileInfo *open_file;
1745 int buf_type = CIFS_NO_BUFFER;
1748 if (file->private_data == NULL) {
1752 open_file = (struct cifsFileInfo *)file->private_data;
1753 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1754 pTcon = cifs_sb->tcon;
1756 pagevec_init(&lru_pvec, 0);
1758 for (i = 0; i < num_pages; ) {
1759 unsigned contig_pages;
1760 struct page *tmp_page;
1761 unsigned long expected_index;
1763 if (list_empty(page_list))
1766 page = list_entry(page_list->prev, struct page, lru);
1767 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1769 /* count adjacent pages that we will read into */
1772 list_entry(page_list->prev, struct page, lru)->index;
1773 list_for_each_entry_reverse(tmp_page,page_list,lru) {
1774 if (tmp_page->index == expected_index) {
1780 if (contig_pages + i > num_pages)
1781 contig_pages = num_pages - i;
1783 /* for reads over a certain size could initiate async
1786 read_size = contig_pages * PAGE_CACHE_SIZE;
1787 /* Read size needs to be in multiples of one page */
1788 read_size = min_t(const unsigned int, read_size,
1789 cifs_sb->rsize & PAGE_CACHE_MASK);
1792 while (rc == -EAGAIN) {
1793 if ((open_file->invalidHandle) &&
1794 (!open_file->closePend)) {
1795 rc = cifs_reopen_file(file->f_path.dentry->d_inode,
1801 rc = CIFSSMBRead(xid, pTcon,
1804 &bytes_read, &smb_read_data,
1806 /* BB more RC checks ? */
1808 if (smb_read_data) {
1809 if(buf_type == CIFS_SMALL_BUFFER)
1810 cifs_small_buf_release(smb_read_data);
1811 else if(buf_type == CIFS_LARGE_BUFFER)
1812 cifs_buf_release(smb_read_data);
1813 smb_read_data = NULL;
1817 if ((rc < 0) || (smb_read_data == NULL)) {
1818 cFYI(1, ("Read error in readpages: %d", rc));
1820 } else if (bytes_read > 0) {
1821 task_io_account_read(bytes_read);
1822 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1823 cifs_copy_cache_pages(mapping, page_list, bytes_read,
1824 smb_read_data + 4 /* RFC1001 hdr */ +
1825 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
1827 i += bytes_read >> PAGE_CACHE_SHIFT;
1828 cifs_stats_bytes_read(pTcon, bytes_read);
1829 if ((int)(bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1830 i++; /* account for partial page */
1832 /* server copy of file can have smaller size
1834 /* BB do we need to verify this common case ?
1835 this case is ok - if we are at server EOF
1836 we will hit it on next read */
1841 cFYI(1, ("No bytes read (%d) at offset %lld . "
1842 "Cleaning remaining pages from readahead list",
1843 bytes_read, offset));
1844 /* BB turn off caching and do new lookup on
1845 file size at server? */
1848 if (smb_read_data) {
1849 if(buf_type == CIFS_SMALL_BUFFER)
1850 cifs_small_buf_release(smb_read_data);
1851 else if(buf_type == CIFS_LARGE_BUFFER)
1852 cifs_buf_release(smb_read_data);
1853 smb_read_data = NULL;
1858 pagevec_lru_add(&lru_pvec);
1860 /* need to free smb_read_data buf before exit */
1861 if (smb_read_data) {
1862 if(buf_type == CIFS_SMALL_BUFFER)
1863 cifs_small_buf_release(smb_read_data);
1864 else if(buf_type == CIFS_LARGE_BUFFER)
1865 cifs_buf_release(smb_read_data);
1866 smb_read_data = NULL;
1873 static int cifs_readpage_worker(struct file *file, struct page *page,
1879 page_cache_get(page);
1880 read_data = kmap(page);
1881 /* for reads over a certain size could initiate async read ahead */
1883 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1888 cFYI(1, ("Bytes read %d",rc));
1890 file->f_path.dentry->d_inode->i_atime =
1891 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1893 if (PAGE_CACHE_SIZE > rc)
1894 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1896 flush_dcache_page(page);
1897 SetPageUptodate(page);
1902 page_cache_release(page);
1906 static int cifs_readpage(struct file *file, struct page *page)
1908 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1914 if (file->private_data == NULL) {
1919 cFYI(1, ("readpage %p at offset %d 0x%x\n",
1920 page, (int)offset, (int)offset));
1922 rc = cifs_readpage_worker(file, page, &offset);
1930 /* We do not want to update the file size from server for inodes
1931 open for write - to avoid races with writepage extending
1932 the file - in the future we could consider allowing
1933 refreshing the inode only on increases in the file size
1934 but this is tricky to do without racing with writebehind
1935 page caching in the current Linux kernel design */
1936 int is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
1938 struct cifsFileInfo *open_file = NULL;
1941 open_file = find_writable_file(cifsInode);
1944 struct cifs_sb_info *cifs_sb;
1946 /* there is not actually a write pending so let
1947 this handle go free and allow it to
1948 be closable if needed */
1949 atomic_dec(&open_file->wrtPending);
1951 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
1952 if ( cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO ) {
1953 /* since no page cache to corrupt on directio
1954 we can change size safely */
1958 if(i_size_read(&cifsInode->vfs_inode) < end_of_file)
1966 static int cifs_prepare_write(struct file *file, struct page *page,
1967 unsigned from, unsigned to)
1973 cFYI(1, ("prepare write for page %p from %d to %d",page,from,to));
1974 if (PageUptodate(page))
1977 /* If we are writing a full page it will be up to date,
1978 no need to read from the server */
1979 if ((to == PAGE_CACHE_SIZE) && (from == 0)) {
1980 SetPageUptodate(page);
1984 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1985 i_size = i_size_read(page->mapping->host);
1987 if ((offset >= i_size) ||
1988 ((from == 0) && (offset + to) >= i_size)) {
1990 * We don't need to read data beyond the end of the file.
1991 * zero it, and set the page uptodate
1993 void *kaddr = kmap_atomic(page, KM_USER0);
1996 memset(kaddr, 0, from);
1997 if (to < PAGE_CACHE_SIZE)
1998 memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
1999 flush_dcache_page(page);
2000 kunmap_atomic(kaddr, KM_USER0);
2001 SetPageUptodate(page);
2002 } else if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2003 /* might as well read a page, it is fast enough */
2004 rc = cifs_readpage_worker(file, page, &offset);
2006 /* we could try using another file handle if there is one -
2007 but how would we lock it to prevent close of that handle
2008 racing with this read? In any case
2009 this will be written out by commit_write so is fine */
2012 /* we do not need to pass errors back
2013 e.g. if we do not have read access to the file
2014 because cifs_commit_write will do the right thing. -- shaggy */
2019 const struct address_space_operations cifs_addr_ops = {
2020 .readpage = cifs_readpage,
2021 .readpages = cifs_readpages,
2022 .writepage = cifs_writepage,
2023 .writepages = cifs_writepages,
2024 .prepare_write = cifs_prepare_write,
2025 .commit_write = cifs_commit_write,
2026 .set_page_dirty = __set_page_dirty_nobuffers,
2027 /* .sync_page = cifs_sync_page, */
2032 * cifs_readpages requires the server to support a buffer large enough to
2033 * contain the header plus one complete page of data. Otherwise, we need
2034 * to leave cifs_readpages out of the address space operations.
2036 const struct address_space_operations cifs_addr_ops_smallbuf = {
2037 .readpage = cifs_readpage,
2038 .writepage = cifs_writepage,
2039 .writepages = cifs_writepages,
2040 .prepare_write = cifs_prepare_write,
2041 .commit_write = cifs_commit_write,
2042 .set_page_dirty = __set_page_dirty_nobuffers,
2043 /* .sync_page = cifs_sync_page, */