4 * vfs operations that deal with files
6 * Copyright (C) International Business Machines Corp., 2002,2007
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/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/delay.h>
33 #include <asm/div64.h>
37 #include "cifsproto.h"
38 #include "cifs_unicode.h"
39 #include "cifs_debug.h"
40 #include "cifs_fs_sb.h"
42 static inline struct cifsFileInfo *cifs_init_private(
43 struct cifsFileInfo *private_data, struct inode *inode,
44 struct file *file, __u16 netfid)
46 memset(private_data, 0, sizeof(struct cifsFileInfo));
47 private_data->netfid = netfid;
48 private_data->pid = current->tgid;
49 init_MUTEX(&private_data->fh_sem);
50 mutex_init(&private_data->lock_mutex);
51 INIT_LIST_HEAD(&private_data->llist);
52 private_data->pfile = file; /* needed for writepage */
53 private_data->pInode = inode;
54 private_data->invalidHandle = FALSE;
55 private_data->closePend = FALSE;
56 /* we have to track num writers to the inode, since writepages
57 does not tell us which handle the write is for so there can
58 be a close (overlapping with write) of the filehandle that
59 cifs_writepages chose to use */
60 atomic_set(&private_data->wrtPending, 0);
65 static inline int cifs_convert_flags(unsigned int flags)
67 if ((flags & O_ACCMODE) == O_RDONLY)
69 else if ((flags & O_ACCMODE) == O_WRONLY)
71 else if ((flags & O_ACCMODE) == O_RDWR) {
72 /* GENERIC_ALL is too much permission to request
73 can cause unnecessary access denied on create */
74 /* return GENERIC_ALL; */
75 return (GENERIC_READ | GENERIC_WRITE);
81 static inline int cifs_get_disposition(unsigned int flags)
83 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
85 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
86 return FILE_OVERWRITE_IF;
87 else if ((flags & O_CREAT) == O_CREAT)
89 else if ((flags & O_TRUNC) == O_TRUNC)
90 return FILE_OVERWRITE;
95 /* all arguments to this function must be checked for validity in caller */
96 static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
97 struct cifsInodeInfo *pCifsInode, struct cifsFileInfo *pCifsFile,
98 struct cifsTconInfo *pTcon, int *oplock, FILE_ALL_INFO *buf,
99 char *full_path, int xid)
101 struct timespec temp;
104 /* want handles we can use to read with first
105 in the list so we do not have to walk the
106 list to search for one in prepare_write */
107 if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
108 list_add_tail(&pCifsFile->flist,
109 &pCifsInode->openFileList);
111 list_add(&pCifsFile->flist,
112 &pCifsInode->openFileList);
114 write_unlock(&GlobalSMBSeslock);
115 if (pCifsInode->clientCanCacheRead) {
116 /* we have the inode open somewhere else
117 no need to discard cache data */
118 goto client_can_cache;
121 /* BB need same check in cifs_create too? */
122 /* if not oplocked, invalidate inode pages if mtime or file
124 temp = cifs_NTtimeToUnix(le64_to_cpu(buf->LastWriteTime));
125 if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
126 (file->f_path.dentry->d_inode->i_size ==
127 (loff_t)le64_to_cpu(buf->EndOfFile))) {
128 cFYI(1, ("inode unchanged on server"));
130 if (file->f_path.dentry->d_inode->i_mapping) {
131 /* BB no need to lock inode until after invalidate
132 since namei code should already have it locked? */
133 filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
135 cFYI(1, ("invalidating remote inode since open detected it "
137 invalidate_remote_inode(file->f_path.dentry->d_inode);
141 if (pTcon->ses->capabilities & CAP_UNIX)
142 rc = cifs_get_inode_info_unix(&file->f_path.dentry->d_inode,
143 full_path, inode->i_sb, xid);
145 rc = cifs_get_inode_info(&file->f_path.dentry->d_inode,
146 full_path, buf, inode->i_sb, xid);
148 if ((*oplock & 0xF) == OPLOCK_EXCLUSIVE) {
149 pCifsInode->clientCanCacheAll = TRUE;
150 pCifsInode->clientCanCacheRead = TRUE;
151 cFYI(1, ("Exclusive Oplock granted on inode %p",
152 file->f_path.dentry->d_inode));
153 } else if ((*oplock & 0xF) == OPLOCK_READ)
154 pCifsInode->clientCanCacheRead = TRUE;
159 int cifs_open(struct inode *inode, struct file *file)
163 struct cifs_sb_info *cifs_sb;
164 struct cifsTconInfo *pTcon;
165 struct cifsFileInfo *pCifsFile;
166 struct cifsInodeInfo *pCifsInode;
167 struct list_head *tmp;
168 char *full_path = NULL;
172 FILE_ALL_INFO *buf = NULL;
176 cifs_sb = CIFS_SB(inode->i_sb);
177 pTcon = cifs_sb->tcon;
179 if (file->f_flags & O_CREAT) {
180 /* search inode for this file and fill in file->private_data */
181 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
182 read_lock(&GlobalSMBSeslock);
183 list_for_each(tmp, &pCifsInode->openFileList) {
184 pCifsFile = list_entry(tmp, struct cifsFileInfo,
186 if ((pCifsFile->pfile == NULL) &&
187 (pCifsFile->pid == current->tgid)) {
188 /* mode set in cifs_create */
190 /* needed for writepage */
191 pCifsFile->pfile = file;
193 file->private_data = pCifsFile;
197 read_unlock(&GlobalSMBSeslock);
198 if (file->private_data != NULL) {
203 if (file->f_flags & O_EXCL)
204 cERROR(1, ("could not find file instance for "
205 "new file %p", file));
209 full_path = build_path_from_dentry(file->f_path.dentry);
210 if (full_path == NULL) {
215 cFYI(1, ("inode = 0x%p file flags are 0x%x for %s",
216 inode, file->f_flags, full_path));
217 desiredAccess = cifs_convert_flags(file->f_flags);
219 /*********************************************************************
220 * open flag mapping table:
222 * POSIX Flag CIFS Disposition
223 * ---------- ----------------
224 * O_CREAT FILE_OPEN_IF
225 * O_CREAT | O_EXCL FILE_CREATE
226 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
227 * O_TRUNC FILE_OVERWRITE
228 * none of the above FILE_OPEN
230 * Note that there is not a direct match between disposition
231 * FILE_SUPERSEDE (ie create whether or not file exists although
232 * O_CREAT | O_TRUNC is similar but truncates the existing
233 * file rather than creating a new file as FILE_SUPERSEDE does
234 * (which uses the attributes / metadata passed in on open call)
236 *? O_SYNC is a reasonable match to CIFS writethrough flag
237 *? and the read write flags match reasonably. O_LARGEFILE
238 *? is irrelevant because largefile support is always used
239 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
240 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
241 *********************************************************************/
243 disposition = cifs_get_disposition(file->f_flags);
250 /* BB pass O_SYNC flag through on file attributes .. BB */
252 /* Also refresh inode by passing in file_info buf returned by SMBOpen
253 and calling get_inode_info with returned buf (at least helps
254 non-Unix server case) */
256 /* BB we can not do this if this is the second open of a file
257 and the first handle has writebehind data, we might be
258 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
259 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
265 if (cifs_sb->tcon->ses->capabilities & CAP_NT_SMBS)
266 rc = CIFSSMBOpen(xid, pTcon, full_path, disposition,
267 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
268 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
269 & CIFS_MOUNT_MAP_SPECIAL_CHR);
271 rc = -EIO; /* no NT SMB support fall into legacy open below */
274 /* Old server, try legacy style OpenX */
275 rc = SMBLegacyOpen(xid, pTcon, full_path, disposition,
276 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
277 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
278 & CIFS_MOUNT_MAP_SPECIAL_CHR);
281 cFYI(1, ("cifs_open returned 0x%x", rc));
285 kmalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
286 if (file->private_data == NULL) {
290 pCifsFile = cifs_init_private(file->private_data, inode, file, netfid);
291 write_lock(&GlobalSMBSeslock);
292 list_add(&pCifsFile->tlist, &pTcon->openFileList);
294 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
296 rc = cifs_open_inode_helper(inode, file, pCifsInode,
298 &oplock, buf, full_path, xid);
300 write_unlock(&GlobalSMBSeslock);
303 if (oplock & CIFS_CREATE_ACTION) {
304 /* time to set mode which we can not set earlier due to
305 problems creating new read-only files */
306 if (cifs_sb->tcon->ses->capabilities & CAP_UNIX) {
307 CIFSSMBUnixSetPerms(xid, pTcon, full_path,
309 (__u64)-1, (__u64)-1, 0 /* dev */,
311 cifs_sb->mnt_cifs_flags &
312 CIFS_MOUNT_MAP_SPECIAL_CHR);
314 /* BB implement via Windows security descriptors eg
315 CIFSSMBWinSetPerms(xid, pTcon, full_path, mode,
317 in the meantime could set r/o dos attribute when
318 perms are eg: mode & 0222 == 0 */
329 /* Try to reacquire byte range locks that were released when session */
330 /* to server was lost */
331 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
335 /* BB list all locks open on this file and relock */
340 static int cifs_reopen_file(struct file *file, int can_flush)
344 struct cifs_sb_info *cifs_sb;
345 struct cifsTconInfo *pTcon;
346 struct cifsFileInfo *pCifsFile;
347 struct cifsInodeInfo *pCifsInode;
349 char *full_path = NULL;
351 int disposition = FILE_OPEN;
354 if (file->private_data) {
355 pCifsFile = (struct cifsFileInfo *)file->private_data;
360 down(&pCifsFile->fh_sem);
361 if (pCifsFile->invalidHandle == FALSE) {
362 up(&pCifsFile->fh_sem);
367 if (file->f_path.dentry == NULL) {
368 cERROR(1, ("no valid name if dentry freed"));
371 goto reopen_error_exit;
374 inode = file->f_path.dentry->d_inode;
376 cERROR(1, ("inode not valid"));
379 goto reopen_error_exit;
382 cifs_sb = CIFS_SB(inode->i_sb);
383 pTcon = cifs_sb->tcon;
385 /* can not grab rename sem here because various ops, including
386 those that already have the rename sem can end up causing writepage
387 to get called and if the server was down that means we end up here,
388 and we can never tell if the caller already has the rename_sem */
389 full_path = build_path_from_dentry(file->f_path.dentry);
390 if (full_path == NULL) {
393 up(&pCifsFile->fh_sem);
398 cFYI(1, ("inode = 0x%p file flags 0x%x for %s",
399 inode, file->f_flags, full_path));
400 desiredAccess = cifs_convert_flags(file->f_flags);
407 /* Can not refresh inode by passing in file_info buf to be returned
408 by SMBOpen and then calling get_inode_info with returned buf
409 since file might have write behind data that needs to be flushed
410 and server version of file size can be stale. If we knew for sure
411 that inode was not dirty locally we could do this */
413 rc = CIFSSMBOpen(xid, pTcon, full_path, disposition, desiredAccess,
414 CREATE_NOT_DIR, &netfid, &oplock, NULL,
415 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
416 CIFS_MOUNT_MAP_SPECIAL_CHR);
418 up(&pCifsFile->fh_sem);
419 cFYI(1, ("cifs_open returned 0x%x", rc));
420 cFYI(1, ("oplock: %d", oplock));
422 pCifsFile->netfid = netfid;
423 pCifsFile->invalidHandle = FALSE;
424 up(&pCifsFile->fh_sem);
425 pCifsInode = CIFS_I(inode);
428 filemap_write_and_wait(inode->i_mapping);
429 /* temporarily disable caching while we
430 go to server to get inode info */
431 pCifsInode->clientCanCacheAll = FALSE;
432 pCifsInode->clientCanCacheRead = FALSE;
433 if (pTcon->ses->capabilities & CAP_UNIX)
434 rc = cifs_get_inode_info_unix(&inode,
435 full_path, inode->i_sb, xid);
437 rc = cifs_get_inode_info(&inode,
438 full_path, NULL, inode->i_sb,
440 } /* else we are writing out data to server already
441 and could deadlock if we tried to flush data, and
442 since we do not know if we have data that would
443 invalidate the current end of file on the server
444 we can not go to the server to get the new inod
446 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
447 pCifsInode->clientCanCacheAll = TRUE;
448 pCifsInode->clientCanCacheRead = TRUE;
449 cFYI(1, ("Exclusive Oplock granted on inode %p",
450 file->f_path.dentry->d_inode));
451 } else if ((oplock & 0xF) == OPLOCK_READ) {
452 pCifsInode->clientCanCacheRead = TRUE;
453 pCifsInode->clientCanCacheAll = FALSE;
455 pCifsInode->clientCanCacheRead = FALSE;
456 pCifsInode->clientCanCacheAll = FALSE;
458 cifs_relock_file(pCifsFile);
467 int cifs_close(struct inode *inode, struct file *file)
471 struct cifs_sb_info *cifs_sb;
472 struct cifsTconInfo *pTcon;
473 struct cifsFileInfo *pSMBFile =
474 (struct cifsFileInfo *)file->private_data;
478 cifs_sb = CIFS_SB(inode->i_sb);
479 pTcon = cifs_sb->tcon;
481 struct cifsLockInfo *li, *tmp;
483 pSMBFile->closePend = TRUE;
485 /* no sense reconnecting to close a file that is
487 if (pTcon->tidStatus != CifsNeedReconnect) {
489 while ((atomic_read(&pSMBFile->wrtPending) != 0)
490 && (timeout < 1000) ) {
491 /* Give write a better chance to get to
492 server ahead of the close. We do not
493 want to add a wait_q here as it would
494 increase the memory utilization as
495 the struct would be in each open file,
496 but this should give enough time to
498 #ifdef CONFIG_CIFS_DEBUG2
499 cFYI(1, ("close delay, write pending"));
504 if (atomic_read(&pSMBFile->wrtPending))
505 cERROR(1,("close with pending writes"));
506 rc = CIFSSMBClose(xid, pTcon,
511 /* Delete any outstanding lock records.
512 We'll lose them when the file is closed anyway. */
513 mutex_lock(&pSMBFile->lock_mutex);
514 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
515 list_del(&li->llist);
518 mutex_unlock(&pSMBFile->lock_mutex);
520 write_lock(&GlobalSMBSeslock);
521 list_del(&pSMBFile->flist);
522 list_del(&pSMBFile->tlist);
523 write_unlock(&GlobalSMBSeslock);
524 kfree(pSMBFile->search_resume_name);
525 kfree(file->private_data);
526 file->private_data = NULL;
530 if (list_empty(&(CIFS_I(inode)->openFileList))) {
531 cFYI(1, ("closing last open instance for inode %p", inode));
532 /* if the file is not open we do not know if we can cache info
533 on this inode, much less write behind and read ahead */
534 CIFS_I(inode)->clientCanCacheRead = FALSE;
535 CIFS_I(inode)->clientCanCacheAll = FALSE;
537 if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
538 rc = CIFS_I(inode)->write_behind_rc;
543 int cifs_closedir(struct inode *inode, struct file *file)
547 struct cifsFileInfo *pCFileStruct =
548 (struct cifsFileInfo *)file->private_data;
551 cFYI(1, ("Closedir inode = 0x%p", inode));
556 struct cifsTconInfo *pTcon;
557 struct cifs_sb_info *cifs_sb =
558 CIFS_SB(file->f_path.dentry->d_sb);
560 pTcon = cifs_sb->tcon;
562 cFYI(1, ("Freeing private data in close dir"));
563 if ((pCFileStruct->srch_inf.endOfSearch == FALSE) &&
564 (pCFileStruct->invalidHandle == FALSE)) {
565 pCFileStruct->invalidHandle = TRUE;
566 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
567 cFYI(1, ("Closing uncompleted readdir with rc %d",
569 /* not much we can do if it fails anyway, ignore rc */
572 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
574 cFYI(1, ("closedir free smb buf in srch struct"));
575 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
576 if (pCFileStruct->srch_inf.smallBuf)
577 cifs_small_buf_release(ptmp);
579 cifs_buf_release(ptmp);
581 ptmp = pCFileStruct->search_resume_name;
583 cFYI(1, ("closedir free resume name"));
584 pCFileStruct->search_resume_name = NULL;
587 kfree(file->private_data);
588 file->private_data = NULL;
590 /* BB can we lock the filestruct while this is going on? */
595 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
596 __u64 offset, __u8 lockType)
598 struct cifsLockInfo *li =
599 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
605 mutex_lock(&fid->lock_mutex);
606 list_add(&li->llist, &fid->llist);
607 mutex_unlock(&fid->lock_mutex);
611 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
617 int wait_flag = FALSE;
618 struct cifs_sb_info *cifs_sb;
619 struct cifsTconInfo *pTcon;
621 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
624 length = 1 + pfLock->fl_end - pfLock->fl_start;
628 cFYI(1, ("Lock parm: 0x%x flockflags: "
629 "0x%x flocktype: 0x%x start: %lld end: %lld",
630 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
633 if (pfLock->fl_flags & FL_POSIX)
635 if (pfLock->fl_flags & FL_FLOCK)
637 if (pfLock->fl_flags & FL_SLEEP) {
638 cFYI(1, ("Blocking lock"));
641 if (pfLock->fl_flags & FL_ACCESS)
642 cFYI(1, ("Process suspended by mandatory locking - "
643 "not implemented yet"));
644 if (pfLock->fl_flags & FL_LEASE)
645 cFYI(1, ("Lease on file - not implemented yet"));
646 if (pfLock->fl_flags &
647 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
648 cFYI(1, ("Unknown lock flags 0x%x", pfLock->fl_flags));
650 if (pfLock->fl_type == F_WRLCK) {
651 cFYI(1, ("F_WRLCK "));
653 } else if (pfLock->fl_type == F_UNLCK) {
654 cFYI(1, ("F_UNLCK"));
656 /* Check if unlock includes more than
658 } else if (pfLock->fl_type == F_RDLCK) {
659 cFYI(1, ("F_RDLCK"));
660 lockType |= LOCKING_ANDX_SHARED_LOCK;
662 } else if (pfLock->fl_type == F_EXLCK) {
663 cFYI(1, ("F_EXLCK"));
665 } else if (pfLock->fl_type == F_SHLCK) {
666 cFYI(1, ("F_SHLCK"));
667 lockType |= LOCKING_ANDX_SHARED_LOCK;
670 cFYI(1, ("Unknown type of lock"));
672 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
673 pTcon = cifs_sb->tcon;
675 if (file->private_data == NULL) {
679 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
681 posix_locking = (cifs_sb->tcon->ses->capabilities & CAP_UNIX) &&
682 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(cifs_sb->tcon->fsUnixInfo.Capability));
684 /* BB add code here to normalize offset and length to
685 account for negative length which we can not accept over the
690 if (lockType & LOCKING_ANDX_SHARED_LOCK)
691 posix_lock_type = CIFS_RDLCK;
693 posix_lock_type = CIFS_WRLCK;
694 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 1 /* get */,
696 posix_lock_type, wait_flag);
701 /* BB we could chain these into one lock request BB */
702 rc = CIFSSMBLock(xid, pTcon, netfid, length, pfLock->fl_start,
703 0, 1, lockType, 0 /* wait flag */ );
705 rc = CIFSSMBLock(xid, pTcon, netfid, length,
706 pfLock->fl_start, 1 /* numUnlock */ ,
707 0 /* numLock */ , lockType,
709 pfLock->fl_type = F_UNLCK;
711 cERROR(1, ("Error unlocking previously locked "
712 "range %d during test of lock", rc));
716 /* if rc == ERR_SHARING_VIOLATION ? */
717 rc = 0; /* do not change lock type to unlock
718 since range in use */
725 if (!numLock && !numUnlock) {
726 /* if no lock or unlock then nothing
727 to do since we do not know what it is */
734 if (lockType & LOCKING_ANDX_SHARED_LOCK)
735 posix_lock_type = CIFS_RDLCK;
737 posix_lock_type = CIFS_WRLCK;
740 posix_lock_type = CIFS_UNLCK;
742 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 0 /* set */,
744 posix_lock_type, wait_flag);
746 struct cifsFileInfo *fid =
747 (struct cifsFileInfo *)file->private_data;
750 rc = CIFSSMBLock(xid, pTcon, netfid, length,
752 0, numLock, lockType, wait_flag);
755 /* For Windows locks we must store them. */
756 rc = store_file_lock(fid, length,
757 pfLock->fl_start, lockType);
759 } else if (numUnlock) {
760 /* For each stored lock that this unlock overlaps
761 completely, unlock it. */
763 struct cifsLockInfo *li, *tmp;
766 mutex_lock(&fid->lock_mutex);
767 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
768 if (pfLock->fl_start <= li->offset &&
769 length >= li->length) {
770 stored_rc = CIFSSMBLock(xid, pTcon,
772 li->length, li->offset,
773 1, 0, li->type, FALSE);
777 list_del(&li->llist);
781 mutex_unlock(&fid->lock_mutex);
785 if (pfLock->fl_flags & FL_POSIX)
786 posix_lock_file_wait(file, pfLock);
791 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
792 size_t write_size, loff_t *poffset)
795 unsigned int bytes_written = 0;
796 unsigned int total_written;
797 struct cifs_sb_info *cifs_sb;
798 struct cifsTconInfo *pTcon;
800 struct cifsFileInfo *open_file;
802 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
804 pTcon = cifs_sb->tcon;
807 (" write %d bytes to offset %lld of %s", write_size,
808 *poffset, file->f_path.dentry->d_name.name)); */
810 if (file->private_data == NULL)
812 open_file = (struct cifsFileInfo *) file->private_data;
816 if (*poffset > file->f_path.dentry->d_inode->i_size)
817 long_op = 2; /* writes past end of file can take a long time */
821 for (total_written = 0; write_size > total_written;
822 total_written += bytes_written) {
824 while (rc == -EAGAIN) {
825 if (file->private_data == NULL) {
826 /* file has been closed on us */
828 /* if we have gotten here we have written some data
829 and blocked, and the file has been freed on us while
830 we blocked so return what we managed to write */
831 return total_written;
833 if (open_file->closePend) {
836 return total_written;
840 if (open_file->invalidHandle) {
841 /* we could deadlock if we called
842 filemap_fdatawait from here so tell
843 reopen_file not to flush data to server
845 rc = cifs_reopen_file(file, FALSE);
850 rc = CIFSSMBWrite(xid, pTcon,
852 min_t(const int, cifs_sb->wsize,
853 write_size - total_written),
854 *poffset, &bytes_written,
855 NULL, write_data + total_written, long_op);
857 if (rc || (bytes_written == 0)) {
865 *poffset += bytes_written;
866 long_op = FALSE; /* subsequent writes fast -
867 15 seconds is plenty */
870 cifs_stats_bytes_written(pTcon, total_written);
872 /* since the write may have blocked check these pointers again */
873 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
874 struct inode *inode = file->f_path.dentry->d_inode;
875 /* Do not update local mtime - server will set its actual value on write
876 * inode->i_ctime = inode->i_mtime =
877 * current_fs_time(inode->i_sb);*/
878 if (total_written > 0) {
879 spin_lock(&inode->i_lock);
880 if (*poffset > file->f_path.dentry->d_inode->i_size)
881 i_size_write(file->f_path.dentry->d_inode,
883 spin_unlock(&inode->i_lock);
885 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
888 return total_written;
891 static ssize_t cifs_write(struct file *file, const char *write_data,
892 size_t write_size, loff_t *poffset)
895 unsigned int bytes_written = 0;
896 unsigned int total_written;
897 struct cifs_sb_info *cifs_sb;
898 struct cifsTconInfo *pTcon;
900 struct cifsFileInfo *open_file;
902 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
904 pTcon = cifs_sb->tcon;
906 cFYI(1, ("write %zd bytes to offset %lld of %s", write_size,
907 *poffset, file->f_path.dentry->d_name.name));
909 if (file->private_data == NULL)
911 open_file = (struct cifsFileInfo *)file->private_data;
915 if (*poffset > file->f_path.dentry->d_inode->i_size)
916 long_op = 2; /* writes past end of file can take a long time */
920 for (total_written = 0; write_size > total_written;
921 total_written += bytes_written) {
923 while (rc == -EAGAIN) {
924 if (file->private_data == NULL) {
925 /* file has been closed on us */
927 /* if we have gotten here we have written some data
928 and blocked, and the file has been freed on us
929 while we blocked so return what we managed to
931 return total_written;
933 if (open_file->closePend) {
936 return total_written;
940 if (open_file->invalidHandle) {
941 /* we could deadlock if we called
942 filemap_fdatawait from here so tell
943 reopen_file not to flush data to
945 rc = cifs_reopen_file(file, FALSE);
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(open_file->pfile, FALSE);
1035 /* if it fails, try another handle - might be */
1036 /* dangerous to hold up writepages with retry */
1039 ("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))
1151 if (!experimEnabled)
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);
1383 spin_unlock(&inode->i_lock);
1384 if (!PageUptodate(page)) {
1385 position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + offset;
1386 /* can not rely on (or let) writepage write this data */
1388 cFYI(1, ("Illegal offsets, can not copy from %d to %d",
1393 /* this is probably better than directly calling
1394 partialpage_write since in this function the file handle is
1395 known which we might as well leverage */
1396 /* BB check if anything else missing out of ppw
1397 such as updating last write time */
1398 page_data = kmap(page);
1399 rc = cifs_write(file, page_data + offset, to-offset,
1403 /* else if (rc < 0) should we set writebehind rc? */
1406 set_page_dirty(page);
1413 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1417 struct inode *inode = file->f_path.dentry->d_inode;
1421 cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1422 dentry->d_name.name, datasync));
1424 rc = filemap_fdatawrite(inode->i_mapping);
1426 CIFS_I(inode)->write_behind_rc = 0;
1431 /* static void cifs_sync_page(struct page *page)
1433 struct address_space *mapping;
1434 struct inode *inode;
1435 unsigned long index = page->index;
1436 unsigned int rpages = 0;
1439 cFYI(1, ("sync page %p",page));
1440 mapping = page->mapping;
1443 inode = mapping->host;
1447 /* fill in rpages then
1448 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1450 /* cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1460 * As file closes, flush all cached write data for this inode checking
1461 * for write behind errors.
1463 int cifs_flush(struct file *file, fl_owner_t id)
1465 struct inode *inode = file->f_path.dentry->d_inode;
1468 /* Rather than do the steps manually:
1469 lock the inode for writing
1470 loop through pages looking for write behind data (dirty pages)
1471 coalesce into contiguous 16K (or smaller) chunks to write to server
1472 send to server (prefer in parallel)
1473 deal with writebehind errors
1474 unlock inode for writing
1475 filemapfdatawrite appears easier for the time being */
1477 rc = filemap_fdatawrite(inode->i_mapping);
1478 if (!rc) /* reset wb rc if we were able to write out dirty pages */
1479 CIFS_I(inode)->write_behind_rc = 0;
1481 cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
1486 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1487 size_t read_size, loff_t *poffset)
1490 unsigned int bytes_read = 0;
1491 unsigned int total_read = 0;
1492 unsigned int current_read_size;
1493 struct cifs_sb_info *cifs_sb;
1494 struct cifsTconInfo *pTcon;
1496 struct cifsFileInfo *open_file;
1497 char *smb_read_data;
1498 char __user *current_offset;
1499 struct smb_com_read_rsp *pSMBr;
1502 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1503 pTcon = cifs_sb->tcon;
1505 if (file->private_data == NULL) {
1509 open_file = (struct cifsFileInfo *)file->private_data;
1511 if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
1512 cFYI(1, ("attempting read on write only file instance"));
1514 for (total_read = 0, current_offset = read_data;
1515 read_size > total_read;
1516 total_read += bytes_read, current_offset += bytes_read) {
1517 current_read_size = min_t(const int, read_size - total_read,
1520 smb_read_data = NULL;
1521 while (rc == -EAGAIN) {
1522 int buf_type = CIFS_NO_BUFFER;
1523 if ((open_file->invalidHandle) &&
1524 (!open_file->closePend)) {
1525 rc = cifs_reopen_file(file, TRUE);
1529 rc = CIFSSMBRead(xid, pTcon,
1531 current_read_size, *poffset,
1532 &bytes_read, &smb_read_data,
1534 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1535 if (smb_read_data) {
1536 if (copy_to_user(current_offset,
1538 4 /* RFC1001 length field */ +
1539 le16_to_cpu(pSMBr->DataOffset),
1544 if (buf_type == CIFS_SMALL_BUFFER)
1545 cifs_small_buf_release(smb_read_data);
1546 else if (buf_type == CIFS_LARGE_BUFFER)
1547 cifs_buf_release(smb_read_data);
1548 smb_read_data = NULL;
1551 if (rc || (bytes_read == 0)) {
1559 cifs_stats_bytes_read(pTcon, bytes_read);
1560 *poffset += bytes_read;
1568 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1572 unsigned int bytes_read = 0;
1573 unsigned int total_read;
1574 unsigned int current_read_size;
1575 struct cifs_sb_info *cifs_sb;
1576 struct cifsTconInfo *pTcon;
1578 char *current_offset;
1579 struct cifsFileInfo *open_file;
1580 int buf_type = CIFS_NO_BUFFER;
1583 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1584 pTcon = cifs_sb->tcon;
1586 if (file->private_data == NULL) {
1590 open_file = (struct cifsFileInfo *)file->private_data;
1592 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1593 cFYI(1, ("attempting read on write only file instance"));
1595 for (total_read = 0, current_offset = read_data;
1596 read_size > total_read;
1597 total_read += bytes_read, current_offset += bytes_read) {
1598 current_read_size = min_t(const int, read_size - total_read,
1600 /* For windows me and 9x we do not want to request more
1601 than it negotiated since it will refuse the read then */
1603 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1604 current_read_size = min_t(const int, current_read_size,
1605 pTcon->ses->server->maxBuf - 128);
1608 while (rc == -EAGAIN) {
1609 if ((open_file->invalidHandle) &&
1610 (!open_file->closePend)) {
1611 rc = cifs_reopen_file(file, TRUE);
1615 rc = CIFSSMBRead(xid, pTcon,
1617 current_read_size, *poffset,
1618 &bytes_read, ¤t_offset,
1621 if (rc || (bytes_read == 0)) {
1629 cifs_stats_bytes_read(pTcon, total_read);
1630 *poffset += bytes_read;
1637 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1639 struct dentry *dentry = file->f_path.dentry;
1643 rc = cifs_revalidate(dentry);
1645 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1649 rc = generic_file_mmap(file, vma);
1655 static void cifs_copy_cache_pages(struct address_space *mapping,
1656 struct list_head *pages, int bytes_read, char *data,
1657 struct pagevec *plru_pvec)
1662 while (bytes_read > 0) {
1663 if (list_empty(pages))
1666 page = list_entry(pages->prev, struct page, lru);
1667 list_del(&page->lru);
1669 if (add_to_page_cache(page, mapping, page->index,
1671 page_cache_release(page);
1672 cFYI(1, ("Add page cache failed"));
1673 data += PAGE_CACHE_SIZE;
1674 bytes_read -= PAGE_CACHE_SIZE;
1678 target = kmap_atomic(page, KM_USER0);
1680 if (PAGE_CACHE_SIZE > bytes_read) {
1681 memcpy(target, data, bytes_read);
1682 /* zero the tail end of this partial page */
1683 memset(target + bytes_read, 0,
1684 PAGE_CACHE_SIZE - bytes_read);
1687 memcpy(target, data, PAGE_CACHE_SIZE);
1688 bytes_read -= PAGE_CACHE_SIZE;
1690 kunmap_atomic(target, KM_USER0);
1692 flush_dcache_page(page);
1693 SetPageUptodate(page);
1695 if (!pagevec_add(plru_pvec, page))
1696 __pagevec_lru_add(plru_pvec);
1697 data += PAGE_CACHE_SIZE;
1702 static int cifs_readpages(struct file *file, struct address_space *mapping,
1703 struct list_head *page_list, unsigned num_pages)
1709 struct cifs_sb_info *cifs_sb;
1710 struct cifsTconInfo *pTcon;
1712 unsigned int read_size, i;
1713 char *smb_read_data = NULL;
1714 struct smb_com_read_rsp *pSMBr;
1715 struct pagevec lru_pvec;
1716 struct cifsFileInfo *open_file;
1717 int buf_type = CIFS_NO_BUFFER;
1720 if (file->private_data == NULL) {
1724 open_file = (struct cifsFileInfo *)file->private_data;
1725 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1726 pTcon = cifs_sb->tcon;
1728 pagevec_init(&lru_pvec, 0);
1729 #ifdef CONFIG_CIFS_DEBUG2
1730 cFYI(1, ("rpages: num pages %d", num_pages));
1732 for (i = 0; i < num_pages; ) {
1733 unsigned contig_pages;
1734 struct page *tmp_page;
1735 unsigned long expected_index;
1737 if (list_empty(page_list))
1740 page = list_entry(page_list->prev, struct page, lru);
1741 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1743 /* count adjacent pages that we will read into */
1746 list_entry(page_list->prev, struct page, lru)->index;
1747 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1748 if (tmp_page->index == expected_index) {
1754 if (contig_pages + i > num_pages)
1755 contig_pages = num_pages - i;
1757 /* for reads over a certain size could initiate async
1760 read_size = contig_pages * PAGE_CACHE_SIZE;
1761 /* Read size needs to be in multiples of one page */
1762 read_size = min_t(const unsigned int, read_size,
1763 cifs_sb->rsize & PAGE_CACHE_MASK);
1764 #ifdef CONFIG_CIFS_DEBUG2
1765 cFYI(1, ("rpages: read size 0x%x contiguous pages %d",
1766 read_size, contig_pages));
1769 while (rc == -EAGAIN) {
1770 if ((open_file->invalidHandle) &&
1771 (!open_file->closePend)) {
1772 rc = cifs_reopen_file(file, TRUE);
1777 rc = CIFSSMBRead(xid, pTcon,
1780 &bytes_read, &smb_read_data,
1782 /* BB more RC checks ? */
1783 if (rc == -EAGAIN) {
1784 if (smb_read_data) {
1785 if (buf_type == CIFS_SMALL_BUFFER)
1786 cifs_small_buf_release(smb_read_data);
1787 else if (buf_type == CIFS_LARGE_BUFFER)
1788 cifs_buf_release(smb_read_data);
1789 smb_read_data = NULL;
1793 if ((rc < 0) || (smb_read_data == NULL)) {
1794 cFYI(1, ("Read error in readpages: %d", rc));
1796 } else if (bytes_read > 0) {
1797 task_io_account_read(bytes_read);
1798 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1799 cifs_copy_cache_pages(mapping, page_list, bytes_read,
1800 smb_read_data + 4 /* RFC1001 hdr */ +
1801 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
1803 i += bytes_read >> PAGE_CACHE_SHIFT;
1804 cifs_stats_bytes_read(pTcon, bytes_read);
1805 if ((int)(bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1806 i++; /* account for partial page */
1808 /* server copy of file can have smaller size
1810 /* BB do we need to verify this common case ?
1811 this case is ok - if we are at server EOF
1812 we will hit it on next read */
1817 cFYI(1, ("No bytes read (%d) at offset %lld . "
1818 "Cleaning remaining pages from readahead list",
1819 bytes_read, offset));
1820 /* BB turn off caching and do new lookup on
1821 file size at server? */
1824 if (smb_read_data) {
1825 if (buf_type == CIFS_SMALL_BUFFER)
1826 cifs_small_buf_release(smb_read_data);
1827 else if (buf_type == CIFS_LARGE_BUFFER)
1828 cifs_buf_release(smb_read_data);
1829 smb_read_data = NULL;
1834 pagevec_lru_add(&lru_pvec);
1836 /* need to free smb_read_data buf before exit */
1837 if (smb_read_data) {
1838 if (buf_type == CIFS_SMALL_BUFFER)
1839 cifs_small_buf_release(smb_read_data);
1840 else if (buf_type == CIFS_LARGE_BUFFER)
1841 cifs_buf_release(smb_read_data);
1842 smb_read_data = NULL;
1849 static int cifs_readpage_worker(struct file *file, struct page *page,
1855 page_cache_get(page);
1856 read_data = kmap(page);
1857 /* for reads over a certain size could initiate async read ahead */
1859 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1864 cFYI(1, ("Bytes read %d", rc));
1866 file->f_path.dentry->d_inode->i_atime =
1867 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1869 if (PAGE_CACHE_SIZE > rc)
1870 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1872 flush_dcache_page(page);
1873 SetPageUptodate(page);
1878 page_cache_release(page);
1882 static int cifs_readpage(struct file *file, struct page *page)
1884 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1890 if (file->private_data == NULL) {
1895 cFYI(1, ("readpage %p at offset %d 0x%x\n",
1896 page, (int)offset, (int)offset));
1898 rc = cifs_readpage_worker(file, page, &offset);
1906 /* We do not want to update the file size from server for inodes
1907 open for write - to avoid races with writepage extending
1908 the file - in the future we could consider allowing
1909 refreshing the inode only on increases in the file size
1910 but this is tricky to do without racing with writebehind
1911 page caching in the current Linux kernel design */
1912 int is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
1914 struct cifsFileInfo *open_file = NULL;
1917 open_file = find_writable_file(cifsInode);
1920 struct cifs_sb_info *cifs_sb;
1922 /* there is not actually a write pending so let
1923 this handle go free and allow it to
1924 be closable if needed */
1925 atomic_dec(&open_file->wrtPending);
1927 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
1928 if ( cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO ) {
1929 /* since no page cache to corrupt on directio
1930 we can change size safely */
1934 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
1942 static int cifs_prepare_write(struct file *file, struct page *page,
1943 unsigned from, unsigned to)
1949 cFYI(1, ("prepare write for page %p from %d to %d", page, from, to));
1950 if (PageUptodate(page))
1953 /* If we are writing a full page it will be up to date,
1954 no need to read from the server */
1955 if ((to == PAGE_CACHE_SIZE) && (from == 0)) {
1956 SetPageUptodate(page);
1960 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1961 i_size = i_size_read(page->mapping->host);
1963 if ((offset >= i_size) ||
1964 ((from == 0) && (offset + to) >= i_size)) {
1966 * We don't need to read data beyond the end of the file.
1967 * zero it, and set the page uptodate
1969 void *kaddr = kmap_atomic(page, KM_USER0);
1972 memset(kaddr, 0, from);
1973 if (to < PAGE_CACHE_SIZE)
1974 memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
1975 flush_dcache_page(page);
1976 kunmap_atomic(kaddr, KM_USER0);
1977 SetPageUptodate(page);
1978 } else if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
1979 /* might as well read a page, it is fast enough */
1980 rc = cifs_readpage_worker(file, page, &offset);
1982 /* we could try using another file handle if there is one -
1983 but how would we lock it to prevent close of that handle
1984 racing with this read? In any case
1985 this will be written out by commit_write so is fine */
1988 /* we do not need to pass errors back
1989 e.g. if we do not have read access to the file
1990 because cifs_commit_write will do the right thing. -- shaggy */
1995 const struct address_space_operations cifs_addr_ops = {
1996 .readpage = cifs_readpage,
1997 .readpages = cifs_readpages,
1998 .writepage = cifs_writepage,
1999 .writepages = cifs_writepages,
2000 .prepare_write = cifs_prepare_write,
2001 .commit_write = cifs_commit_write,
2002 .set_page_dirty = __set_page_dirty_nobuffers,
2003 /* .sync_page = cifs_sync_page, */
2008 * cifs_readpages requires the server to support a buffer large enough to
2009 * contain the header plus one complete page of data. Otherwise, we need
2010 * to leave cifs_readpages out of the address space operations.
2012 const struct address_space_operations cifs_addr_ops_smallbuf = {
2013 .readpage = cifs_readpage,
2014 .writepage = cifs_writepage,
2015 .writepages = cifs_writepages,
2016 .prepare_write = cifs_prepare_write,
2017 .commit_write = cifs_commit_write,
2018 .set_page_dirty = __set_page_dirty_nobuffers,
2019 /* .sync_page = cifs_sync_page, */