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 mutex_init(&private_data->lock_mutex);
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 file *file, int can_flush)
345 struct cifs_sb_info *cifs_sb;
346 struct cifsTconInfo *pTcon;
347 struct cifsFileInfo *pCifsFile;
348 struct cifsInodeInfo *pCifsInode;
349 struct inode * inode;
350 char *full_path = NULL;
352 int disposition = FILE_OPEN;
355 if (file->private_data) {
356 pCifsFile = (struct cifsFileInfo *)file->private_data;
361 down(&pCifsFile->fh_sem);
362 if (pCifsFile->invalidHandle == FALSE) {
363 up(&pCifsFile->fh_sem);
368 if (file->f_path.dentry == NULL) {
369 cERROR(1, ("no valid name if dentry freed"));
372 goto reopen_error_exit;
375 inode = file->f_path.dentry->d_inode;
377 cERROR(1, ("inode not valid"));
380 goto reopen_error_exit;
383 cifs_sb = CIFS_SB(inode->i_sb);
384 pTcon = cifs_sb->tcon;
386 /* can not grab rename sem here because various ops, including
387 those that already have the rename sem can end up causing writepage
388 to get called and if the server was down that means we end up here,
389 and we can never tell if the caller already has the rename_sem */
390 full_path = build_path_from_dentry(file->f_path.dentry);
391 if (full_path == NULL) {
394 up(&pCifsFile->fh_sem);
399 cFYI(1, ("inode = 0x%p file flags 0x%x for %s",
400 inode, file->f_flags,full_path));
401 desiredAccess = cifs_convert_flags(file->f_flags);
408 /* Can not refresh inode by passing in file_info buf to be returned
409 by SMBOpen and then calling get_inode_info with returned buf
410 since file might have write behind data that needs to be flushed
411 and server version of file size can be stale. If we knew for sure
412 that inode was not dirty locally we could do this */
414 rc = CIFSSMBOpen(xid, pTcon, full_path, disposition, desiredAccess,
415 CREATE_NOT_DIR, &netfid, &oplock, NULL,
416 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
417 CIFS_MOUNT_MAP_SPECIAL_CHR);
419 up(&pCifsFile->fh_sem);
420 cFYI(1, ("cifs_open returned 0x%x", rc));
421 cFYI(1, ("oplock: %d", oplock));
423 pCifsFile->netfid = netfid;
424 pCifsFile->invalidHandle = FALSE;
425 up(&pCifsFile->fh_sem);
426 pCifsInode = CIFS_I(inode);
429 filemap_write_and_wait(inode->i_mapping);
430 /* temporarily disable caching while we
431 go to server to get inode info */
432 pCifsInode->clientCanCacheAll = FALSE;
433 pCifsInode->clientCanCacheRead = FALSE;
434 if (pTcon->ses->capabilities & CAP_UNIX)
435 rc = cifs_get_inode_info_unix(&inode,
436 full_path, inode->i_sb, xid);
438 rc = cifs_get_inode_info(&inode,
439 full_path, NULL, inode->i_sb,
441 } /* else we are writing out data to server already
442 and could deadlock if we tried to flush data, and
443 since we do not know if we have data that would
444 invalidate the current end of file on the server
445 we can not go to the server to get the new inod
447 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
448 pCifsInode->clientCanCacheAll = TRUE;
449 pCifsInode->clientCanCacheRead = TRUE;
450 cFYI(1, ("Exclusive Oplock granted on inode %p",
451 file->f_path.dentry->d_inode));
452 } else if ((oplock & 0xF) == OPLOCK_READ) {
453 pCifsInode->clientCanCacheRead = TRUE;
454 pCifsInode->clientCanCacheAll = FALSE;
456 pCifsInode->clientCanCacheRead = FALSE;
457 pCifsInode->clientCanCacheAll = FALSE;
459 cifs_relock_file(pCifsFile);
468 int cifs_close(struct inode *inode, struct file *file)
472 struct cifs_sb_info *cifs_sb;
473 struct cifsTconInfo *pTcon;
474 struct cifsFileInfo *pSMBFile =
475 (struct cifsFileInfo *)file->private_data;
479 cifs_sb = CIFS_SB(inode->i_sb);
480 pTcon = cifs_sb->tcon;
482 struct cifsLockInfo *li, *tmp;
484 pSMBFile->closePend = TRUE;
486 /* no sense reconnecting to close a file that is
488 if (pTcon->tidStatus != CifsNeedReconnect) {
490 while((atomic_read(&pSMBFile->wrtPending) != 0)
491 && (timeout < 1000) ) {
492 /* Give write a better chance to get to
493 server ahead of the close. We do not
494 want to add a wait_q here as it would
495 increase the memory utilization as
496 the struct would be in each open file,
497 but this should give enough time to
499 #ifdef CONFIG_CIFS_DEBUG2
500 cFYI(1,("close delay, write pending"));
505 if(atomic_read(&pSMBFile->wrtPending))
506 cERROR(1,("close with pending writes"));
507 rc = CIFSSMBClose(xid, pTcon,
512 /* Delete any outstanding lock records.
513 We'll lose them when the file is closed anyway. */
514 mutex_lock(&pSMBFile->lock_mutex);
515 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
516 list_del(&li->llist);
519 mutex_unlock(&pSMBFile->lock_mutex);
521 write_lock(&GlobalSMBSeslock);
522 list_del(&pSMBFile->flist);
523 list_del(&pSMBFile->tlist);
524 write_unlock(&GlobalSMBSeslock);
525 kfree(pSMBFile->search_resume_name);
526 kfree(file->private_data);
527 file->private_data = NULL;
531 if (list_empty(&(CIFS_I(inode)->openFileList))) {
532 cFYI(1, ("closing last open instance for inode %p", inode));
533 /* if the file is not open we do not know if we can cache info
534 on this inode, much less write behind and read ahead */
535 CIFS_I(inode)->clientCanCacheRead = FALSE;
536 CIFS_I(inode)->clientCanCacheAll = FALSE;
538 if ((rc ==0) && CIFS_I(inode)->write_behind_rc)
539 rc = CIFS_I(inode)->write_behind_rc;
544 int cifs_closedir(struct inode *inode, struct file *file)
548 struct cifsFileInfo *pCFileStruct =
549 (struct cifsFileInfo *)file->private_data;
552 cFYI(1, ("Closedir inode = 0x%p", inode));
557 struct cifsTconInfo *pTcon;
558 struct cifs_sb_info *cifs_sb = 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 = kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
604 mutex_lock(&fid->lock_mutex);
605 list_add(&li->llist, &fid->llist);
606 mutex_unlock(&fid->lock_mutex);
610 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
616 int wait_flag = FALSE;
617 struct cifs_sb_info *cifs_sb;
618 struct cifsTconInfo *pTcon;
620 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
623 length = 1 + pfLock->fl_end - pfLock->fl_start;
627 cFYI(1, ("Lock parm: 0x%x flockflags: "
628 "0x%x flocktype: 0x%x start: %lld end: %lld",
629 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
632 if (pfLock->fl_flags & FL_POSIX)
634 if (pfLock->fl_flags & FL_FLOCK)
636 if (pfLock->fl_flags & FL_SLEEP) {
637 cFYI(1, ("Blocking lock"));
640 if (pfLock->fl_flags & FL_ACCESS)
641 cFYI(1, ("Process suspended by mandatory locking - "
642 "not implemented yet"));
643 if (pfLock->fl_flags & FL_LEASE)
644 cFYI(1, ("Lease on file - not implemented yet"));
645 if (pfLock->fl_flags &
646 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
647 cFYI(1, ("Unknown lock flags 0x%x", pfLock->fl_flags));
649 if (pfLock->fl_type == F_WRLCK) {
650 cFYI(1, ("F_WRLCK "));
652 } else if (pfLock->fl_type == F_UNLCK) {
653 cFYI(1, ("F_UNLCK"));
655 /* Check if unlock includes more than
657 } else if (pfLock->fl_type == F_RDLCK) {
658 cFYI(1, ("F_RDLCK"));
659 lockType |= LOCKING_ANDX_SHARED_LOCK;
661 } else if (pfLock->fl_type == F_EXLCK) {
662 cFYI(1, ("F_EXLCK"));
664 } else if (pfLock->fl_type == F_SHLCK) {
665 cFYI(1, ("F_SHLCK"));
666 lockType |= LOCKING_ANDX_SHARED_LOCK;
669 cFYI(1, ("Unknown type of lock"));
671 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
672 pTcon = cifs_sb->tcon;
674 if (file->private_data == NULL) {
678 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
680 posix_locking = (cifs_sb->tcon->ses->capabilities & CAP_UNIX) &&
681 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(cifs_sb->tcon->fsUnixInfo.Capability));
683 /* BB add code here to normalize offset and length to
684 account for negative length which we can not accept over the
689 if(lockType & LOCKING_ANDX_SHARED_LOCK)
690 posix_lock_type = CIFS_RDLCK;
692 posix_lock_type = CIFS_WRLCK;
693 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 1 /* get */,
695 posix_lock_type, wait_flag);
700 /* BB we could chain these into one lock request BB */
701 rc = CIFSSMBLock(xid, pTcon, netfid, length, pfLock->fl_start,
702 0, 1, lockType, 0 /* wait flag */ );
704 rc = CIFSSMBLock(xid, pTcon, netfid, length,
705 pfLock->fl_start, 1 /* numUnlock */ ,
706 0 /* numLock */ , lockType,
708 pfLock->fl_type = F_UNLCK;
710 cERROR(1, ("Error unlocking previously locked "
711 "range %d during test of lock", rc));
715 /* if rc == ERR_SHARING_VIOLATION ? */
716 rc = 0; /* do not change lock type to unlock
717 since range in use */
724 if (!numLock && !numUnlock) {
725 /* if no lock or unlock then nothing
726 to do since we do not know what it is */
733 if(lockType & LOCKING_ANDX_SHARED_LOCK)
734 posix_lock_type = CIFS_RDLCK;
736 posix_lock_type = CIFS_WRLCK;
739 posix_lock_type = CIFS_UNLCK;
741 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 0 /* set */,
743 posix_lock_type, wait_flag);
745 struct cifsFileInfo *fid = (struct cifsFileInfo *)file->private_data;
748 rc = CIFSSMBLock(xid, pTcon, netfid, length, pfLock->fl_start,
749 0, numLock, lockType, wait_flag);
752 /* For Windows locks we must store them. */
753 rc = store_file_lock(fid, length,
754 pfLock->fl_start, lockType);
756 } else if (numUnlock) {
757 /* For each stored lock that this unlock overlaps
758 completely, unlock it. */
760 struct cifsLockInfo *li, *tmp;
763 mutex_lock(&fid->lock_mutex);
764 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
765 if (pfLock->fl_start <= li->offset &&
766 length >= li->length) {
767 stored_rc = CIFSSMBLock(xid, pTcon, netfid,
768 li->length, li->offset,
769 1, 0, li->type, FALSE);
773 list_del(&li->llist);
777 mutex_unlock(&fid->lock_mutex);
781 if (pfLock->fl_flags & FL_POSIX)
782 posix_lock_file_wait(file, pfLock);
787 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
788 size_t write_size, loff_t *poffset)
791 unsigned int bytes_written = 0;
792 unsigned int total_written;
793 struct cifs_sb_info *cifs_sb;
794 struct cifsTconInfo *pTcon;
796 struct cifsFileInfo *open_file;
798 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
800 pTcon = cifs_sb->tcon;
803 (" write %d bytes to offset %lld of %s", write_size,
804 *poffset, file->f_path.dentry->d_name.name)); */
806 if (file->private_data == NULL)
808 open_file = (struct cifsFileInfo *) file->private_data;
812 if (*poffset > file->f_path.dentry->d_inode->i_size)
813 long_op = 2; /* writes past end of file can take a long time */
817 for (total_written = 0; write_size > total_written;
818 total_written += bytes_written) {
820 while (rc == -EAGAIN) {
821 if (file->private_data == NULL) {
822 /* file has been closed on us */
824 /* if we have gotten here we have written some data
825 and blocked, and the file has been freed on us while
826 we blocked so return what we managed to write */
827 return total_written;
829 if (open_file->closePend) {
832 return total_written;
836 if (open_file->invalidHandle) {
837 /* we could deadlock if we called
838 filemap_fdatawait from here so tell
839 reopen_file not to flush data to server
841 rc = cifs_reopen_file(file, FALSE);
846 rc = CIFSSMBWrite(xid, pTcon,
848 min_t(const int, cifs_sb->wsize,
849 write_size - total_written),
850 *poffset, &bytes_written,
851 NULL, write_data + total_written, long_op);
853 if (rc || (bytes_written == 0)) {
861 *poffset += bytes_written;
862 long_op = FALSE; /* subsequent writes fast -
863 15 seconds is plenty */
866 cifs_stats_bytes_written(pTcon, total_written);
868 /* since the write may have blocked check these pointers again */
869 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
870 struct inode *inode = file->f_path.dentry->d_inode;
871 /* Do not update local mtime - server will set its actual value on write
872 * inode->i_ctime = inode->i_mtime =
873 * current_fs_time(inode->i_sb);*/
874 if (total_written > 0) {
875 spin_lock(&inode->i_lock);
876 if (*poffset > file->f_path.dentry->d_inode->i_size)
877 i_size_write(file->f_path.dentry->d_inode,
879 spin_unlock(&inode->i_lock);
881 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
884 return total_written;
887 static ssize_t cifs_write(struct file *file, const char *write_data,
888 size_t write_size, loff_t *poffset)
891 unsigned int bytes_written = 0;
892 unsigned int total_written;
893 struct cifs_sb_info *cifs_sb;
894 struct cifsTconInfo *pTcon;
896 struct cifsFileInfo *open_file;
898 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
900 pTcon = cifs_sb->tcon;
902 cFYI(1,("write %zd bytes to offset %lld of %s", write_size,
903 *poffset, file->f_path.dentry->d_name.name));
905 if (file->private_data == NULL)
907 open_file = (struct cifsFileInfo *)file->private_data;
911 if (*poffset > file->f_path.dentry->d_inode->i_size)
912 long_op = 2; /* writes past end of file can take a long time */
916 for (total_written = 0; write_size > total_written;
917 total_written += bytes_written) {
919 while (rc == -EAGAIN) {
920 if (file->private_data == NULL) {
921 /* file has been closed on us */
923 /* if we have gotten here we have written some data
924 and blocked, and the file has been freed on us
925 while we blocked so return what we managed to
927 return total_written;
929 if (open_file->closePend) {
932 return total_written;
936 if (open_file->invalidHandle) {
937 /* we could deadlock if we called
938 filemap_fdatawait from here so tell
939 reopen_file not to flush data to
941 rc = cifs_reopen_file(file, FALSE);
945 if(experimEnabled || (pTcon->ses->server &&
946 ((pTcon->ses->server->secMode &
947 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
952 len = min((size_t)cifs_sb->wsize,
953 write_size - total_written);
954 /* iov[0] is reserved for smb header */
955 iov[1].iov_base = (char *)write_data +
957 iov[1].iov_len = len;
958 rc = CIFSSMBWrite2(xid, pTcon,
959 open_file->netfid, len,
960 *poffset, &bytes_written,
963 rc = CIFSSMBWrite(xid, pTcon,
965 min_t(const int, cifs_sb->wsize,
966 write_size - total_written),
967 *poffset, &bytes_written,
968 write_data + total_written,
971 if (rc || (bytes_written == 0)) {
979 *poffset += bytes_written;
980 long_op = FALSE; /* subsequent writes fast -
981 15 seconds is plenty */
984 cifs_stats_bytes_written(pTcon, total_written);
986 /* since the write may have blocked check these pointers again */
987 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
988 /*BB We could make this contingent on superblock ATIME flag too */
989 /* file->f_path.dentry->d_inode->i_ctime =
990 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
991 if (total_written > 0) {
992 spin_lock(&file->f_path.dentry->d_inode->i_lock);
993 if (*poffset > file->f_path.dentry->d_inode->i_size)
994 i_size_write(file->f_path.dentry->d_inode,
996 spin_unlock(&file->f_path.dentry->d_inode->i_lock);
998 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1001 return total_written;
1004 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1006 struct cifsFileInfo *open_file;
1009 /* Having a null inode here (because mapping->host was set to zero by
1010 the VFS or MM) should not happen but we had reports of on oops (due to
1011 it being zero) during stress testcases so we need to check for it */
1013 if(cifs_inode == NULL) {
1014 cERROR(1,("Null inode passed to cifs_writeable_file"));
1019 read_lock(&GlobalSMBSeslock);
1020 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1021 if (open_file->closePend)
1023 if (open_file->pfile &&
1024 ((open_file->pfile->f_flags & O_RDWR) ||
1025 (open_file->pfile->f_flags & O_WRONLY))) {
1026 atomic_inc(&open_file->wrtPending);
1027 read_unlock(&GlobalSMBSeslock);
1028 if((open_file->invalidHandle) &&
1029 (!open_file->closePend) /* BB fixme -since the second clause can not be true remove it BB */) {
1030 rc = cifs_reopen_file(open_file->pfile, FALSE);
1031 /* if it fails, try another handle - might be */
1032 /* dangerous to hold up writepages with retry */
1034 cFYI(1,("failed on reopen file in wp"));
1035 read_lock(&GlobalSMBSeslock);
1036 /* can not use this handle, no write
1037 pending on this one after all */
1039 (&open_file->wrtPending);
1046 read_unlock(&GlobalSMBSeslock);
1050 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1052 struct address_space *mapping = page->mapping;
1053 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1056 int bytes_written = 0;
1057 struct cifs_sb_info *cifs_sb;
1058 struct cifsTconInfo *pTcon;
1059 struct inode *inode;
1060 struct cifsFileInfo *open_file;
1062 if (!mapping || !mapping->host)
1065 inode = page->mapping->host;
1066 cifs_sb = CIFS_SB(inode->i_sb);
1067 pTcon = cifs_sb->tcon;
1069 offset += (loff_t)from;
1070 write_data = kmap(page);
1073 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1078 /* racing with truncate? */
1079 if (offset > mapping->host->i_size) {
1081 return 0; /* don't care */
1084 /* check to make sure that we are not extending the file */
1085 if (mapping->host->i_size - offset < (loff_t)to)
1086 to = (unsigned)(mapping->host->i_size - offset);
1088 open_file = find_writable_file(CIFS_I(mapping->host));
1090 bytes_written = cifs_write(open_file->pfile, write_data,
1092 atomic_dec(&open_file->wrtPending);
1093 /* Does mm or vfs already set times? */
1094 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1095 if ((bytes_written > 0) && (offset)) {
1097 } else if (bytes_written < 0) {
1102 cFYI(1, ("No writeable filehandles for inode"));
1110 static int cifs_writepages(struct address_space *mapping,
1111 struct writeback_control *wbc)
1113 struct backing_dev_info *bdi = mapping->backing_dev_info;
1114 unsigned int bytes_to_write;
1115 unsigned int bytes_written;
1116 struct cifs_sb_info *cifs_sb;
1120 int range_whole = 0;
1127 struct cifsFileInfo *open_file;
1129 struct pagevec pvec;
1134 cifs_sb = CIFS_SB(mapping->host->i_sb);
1137 * If wsize is smaller that the page cache size, default to writing
1138 * one page at a time via cifs_writepage
1140 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1141 return generic_writepages(mapping, wbc);
1143 if((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1144 if(cifs_sb->tcon->ses->server->secMode &
1145 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1147 return generic_writepages(mapping, wbc);
1149 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1151 return generic_writepages(mapping, wbc);
1155 * BB: Is this meaningful for a non-block-device file system?
1156 * If it is, we should test it again after we do I/O
1158 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1159 wbc->encountered_congestion = 1;
1166 pagevec_init(&pvec, 0);
1167 if (wbc->range_cyclic) {
1168 index = mapping->writeback_index; /* Start from prev offset */
1171 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1172 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1173 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1178 while (!done && (index <= end) &&
1179 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1180 PAGECACHE_TAG_DIRTY,
1181 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1190 for (i = 0; i < nr_pages; i++) {
1191 page = pvec.pages[i];
1193 * At this point we hold neither mapping->tree_lock nor
1194 * lock on the page itself: the page may be truncated or
1195 * invalidated (changing page->mapping to NULL), or even
1196 * swizzled back from swapper_space to tmpfs file
1202 else if (TestSetPageLocked(page))
1205 if (unlikely(page->mapping != mapping)) {
1210 if (!wbc->range_cyclic && page->index > end) {
1216 if (next && (page->index != next)) {
1217 /* Not next consecutive page */
1222 if (wbc->sync_mode != WB_SYNC_NONE)
1223 wait_on_page_writeback(page);
1225 if (PageWriteback(page) ||
1226 !clear_page_dirty_for_io(page)) {
1232 * This actually clears the dirty bit in the radix tree.
1233 * See cifs_writepage() for more commentary.
1235 set_page_writeback(page);
1237 if (page_offset(page) >= mapping->host->i_size) {
1240 end_page_writeback(page);
1245 * BB can we get rid of this? pages are held by pvec
1247 page_cache_get(page);
1249 len = min(mapping->host->i_size - page_offset(page),
1250 (loff_t)PAGE_CACHE_SIZE);
1252 /* reserve iov[0] for the smb header */
1254 iov[n_iov].iov_base = kmap(page);
1255 iov[n_iov].iov_len = len;
1256 bytes_to_write += len;
1260 offset = page_offset(page);
1262 next = page->index + 1;
1263 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1267 /* Search for a writable handle every time we call
1268 * CIFSSMBWrite2. We can't rely on the last handle
1269 * we used to still be valid
1271 open_file = find_writable_file(CIFS_I(mapping->host));
1273 cERROR(1, ("No writable handles for inode"));
1276 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1278 bytes_to_write, offset,
1279 &bytes_written, iov, n_iov,
1281 atomic_dec(&open_file->wrtPending);
1282 if (rc || bytes_written < bytes_to_write) {
1283 cERROR(1,("Write2 ret %d, written = %d",
1284 rc, bytes_written));
1285 /* BB what if continued retry is
1286 requested via mount flags? */
1287 set_bit(AS_EIO, &mapping->flags);
1289 cifs_stats_bytes_written(cifs_sb->tcon,
1293 for (i = 0; i < n_iov; i++) {
1294 page = pvec.pages[first + i];
1295 /* Should we also set page error on
1296 success rc but too little data written? */
1297 /* BB investigate retry logic on temporary
1298 server crash cases and how recovery works
1299 when page marked as error */
1304 end_page_writeback(page);
1305 page_cache_release(page);
1307 if ((wbc->nr_to_write -= n_iov) <= 0)
1311 pagevec_release(&pvec);
1313 if (!scanned && !done) {
1315 * We hit the last page and there is more work to be done: wrap
1316 * back to the start of the file
1322 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1323 mapping->writeback_index = index;
1330 static int cifs_writepage(struct page* page, struct writeback_control *wbc)
1336 /* BB add check for wbc flags */
1337 page_cache_get(page);
1338 if (!PageUptodate(page)) {
1339 cFYI(1, ("ppw - page not up to date"));
1343 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1345 * A writepage() implementation always needs to do either this,
1346 * or re-dirty the page with "redirty_page_for_writepage()" in
1347 * the case of a failure.
1349 * Just unlocking the page will cause the radix tree tag-bits
1350 * to fail to update with the state of the page correctly.
1352 set_page_writeback(page);
1353 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1354 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1356 end_page_writeback(page);
1357 page_cache_release(page);
1362 static int cifs_commit_write(struct file *file, struct page *page,
1363 unsigned offset, unsigned to)
1367 struct inode *inode = page->mapping->host;
1368 loff_t position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1372 cFYI(1, ("commit write for page %p up to position %lld for %d",
1373 page, position, to));
1374 spin_lock(&inode->i_lock);
1375 if (position > inode->i_size) {
1376 i_size_write(inode, position);
1378 spin_unlock(&inode->i_lock);
1379 if (!PageUptodate(page)) {
1380 position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + offset;
1381 /* can not rely on (or let) writepage write this data */
1383 cFYI(1, ("Illegal offsets, can not copy from %d to %d",
1388 /* this is probably better than directly calling
1389 partialpage_write since in this function the file handle is
1390 known which we might as well leverage */
1391 /* BB check if anything else missing out of ppw
1392 such as updating last write time */
1393 page_data = kmap(page);
1394 rc = cifs_write(file, page_data + offset, to-offset,
1398 /* else if (rc < 0) should we set writebehind rc? */
1401 set_page_dirty(page);
1408 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1412 struct inode *inode = file->f_path.dentry->d_inode;
1416 cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1417 dentry->d_name.name, datasync));
1419 rc = filemap_fdatawrite(inode->i_mapping);
1421 CIFS_I(inode)->write_behind_rc = 0;
1426 /* static void cifs_sync_page(struct page *page)
1428 struct address_space *mapping;
1429 struct inode *inode;
1430 unsigned long index = page->index;
1431 unsigned int rpages = 0;
1434 cFYI(1, ("sync page %p",page));
1435 mapping = page->mapping;
1438 inode = mapping->host;
1442 /* fill in rpages then
1443 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1445 /* cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1455 * As file closes, flush all cached write data for this inode checking
1456 * for write behind errors.
1458 int cifs_flush(struct file *file, fl_owner_t id)
1460 struct inode * inode = file->f_path.dentry->d_inode;
1463 /* Rather than do the steps manually:
1464 lock the inode for writing
1465 loop through pages looking for write behind data (dirty pages)
1466 coalesce into contiguous 16K (or smaller) chunks to write to server
1467 send to server (prefer in parallel)
1468 deal with writebehind errors
1469 unlock inode for writing
1470 filemapfdatawrite appears easier for the time being */
1472 rc = filemap_fdatawrite(inode->i_mapping);
1473 if (!rc) /* reset wb rc if we were able to write out dirty pages */
1474 CIFS_I(inode)->write_behind_rc = 0;
1476 cFYI(1, ("Flush inode %p file %p rc %d",inode,file,rc));
1481 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1482 size_t read_size, loff_t *poffset)
1485 unsigned int bytes_read = 0;
1486 unsigned int total_read = 0;
1487 unsigned int current_read_size;
1488 struct cifs_sb_info *cifs_sb;
1489 struct cifsTconInfo *pTcon;
1491 struct cifsFileInfo *open_file;
1492 char *smb_read_data;
1493 char __user *current_offset;
1494 struct smb_com_read_rsp *pSMBr;
1497 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1498 pTcon = cifs_sb->tcon;
1500 if (file->private_data == NULL) {
1504 open_file = (struct cifsFileInfo *)file->private_data;
1506 if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
1507 cFYI(1, ("attempting read on write only file instance"));
1509 for (total_read = 0, current_offset = read_data;
1510 read_size > total_read;
1511 total_read += bytes_read, current_offset += bytes_read) {
1512 current_read_size = min_t(const int, read_size - total_read,
1515 smb_read_data = NULL;
1516 while (rc == -EAGAIN) {
1517 int buf_type = CIFS_NO_BUFFER;
1518 if ((open_file->invalidHandle) &&
1519 (!open_file->closePend)) {
1520 rc = cifs_reopen_file(file, TRUE);
1524 rc = CIFSSMBRead(xid, pTcon,
1526 current_read_size, *poffset,
1527 &bytes_read, &smb_read_data,
1529 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1530 if (smb_read_data) {
1531 if (copy_to_user(current_offset,
1533 4 /* RFC1001 length field */ +
1534 le16_to_cpu(pSMBr->DataOffset),
1539 if(buf_type == CIFS_SMALL_BUFFER)
1540 cifs_small_buf_release(smb_read_data);
1541 else if(buf_type == CIFS_LARGE_BUFFER)
1542 cifs_buf_release(smb_read_data);
1543 smb_read_data = NULL;
1546 if (rc || (bytes_read == 0)) {
1554 cifs_stats_bytes_read(pTcon, bytes_read);
1555 *poffset += bytes_read;
1563 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1567 unsigned int bytes_read = 0;
1568 unsigned int total_read;
1569 unsigned int current_read_size;
1570 struct cifs_sb_info *cifs_sb;
1571 struct cifsTconInfo *pTcon;
1573 char *current_offset;
1574 struct cifsFileInfo *open_file;
1575 int buf_type = CIFS_NO_BUFFER;
1578 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1579 pTcon = cifs_sb->tcon;
1581 if (file->private_data == NULL) {
1585 open_file = (struct cifsFileInfo *)file->private_data;
1587 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1588 cFYI(1, ("attempting read on write only file instance"));
1590 for (total_read = 0, current_offset = read_data;
1591 read_size > total_read;
1592 total_read += bytes_read, current_offset += bytes_read) {
1593 current_read_size = min_t(const int, read_size - total_read,
1595 /* For windows me and 9x we do not want to request more
1596 than it negotiated since it will refuse the read then */
1598 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1599 current_read_size = min_t(const int, current_read_size,
1600 pTcon->ses->server->maxBuf - 128);
1603 while (rc == -EAGAIN) {
1604 if ((open_file->invalidHandle) &&
1605 (!open_file->closePend)) {
1606 rc = cifs_reopen_file(file, TRUE);
1610 rc = CIFSSMBRead(xid, pTcon,
1612 current_read_size, *poffset,
1613 &bytes_read, ¤t_offset,
1616 if (rc || (bytes_read == 0)) {
1624 cifs_stats_bytes_read(pTcon, total_read);
1625 *poffset += bytes_read;
1632 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1634 struct dentry *dentry = file->f_path.dentry;
1638 rc = cifs_revalidate(dentry);
1640 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1644 rc = generic_file_mmap(file, vma);
1650 static void cifs_copy_cache_pages(struct address_space *mapping,
1651 struct list_head *pages, int bytes_read, char *data,
1652 struct pagevec *plru_pvec)
1657 while (bytes_read > 0) {
1658 if (list_empty(pages))
1661 page = list_entry(pages->prev, struct page, lru);
1662 list_del(&page->lru);
1664 if (add_to_page_cache(page, mapping, page->index,
1666 page_cache_release(page);
1667 cFYI(1, ("Add page cache failed"));
1668 data += PAGE_CACHE_SIZE;
1669 bytes_read -= PAGE_CACHE_SIZE;
1673 target = kmap_atomic(page,KM_USER0);
1675 if (PAGE_CACHE_SIZE > bytes_read) {
1676 memcpy(target, data, bytes_read);
1677 /* zero the tail end of this partial page */
1678 memset(target + bytes_read, 0,
1679 PAGE_CACHE_SIZE - bytes_read);
1682 memcpy(target, data, PAGE_CACHE_SIZE);
1683 bytes_read -= PAGE_CACHE_SIZE;
1685 kunmap_atomic(target, KM_USER0);
1687 flush_dcache_page(page);
1688 SetPageUptodate(page);
1690 if (!pagevec_add(plru_pvec, page))
1691 __pagevec_lru_add(plru_pvec);
1692 data += PAGE_CACHE_SIZE;
1697 static int cifs_readpages(struct file *file, struct address_space *mapping,
1698 struct list_head *page_list, unsigned num_pages)
1704 struct cifs_sb_info *cifs_sb;
1705 struct cifsTconInfo *pTcon;
1707 unsigned int read_size,i;
1708 char *smb_read_data = NULL;
1709 struct smb_com_read_rsp *pSMBr;
1710 struct pagevec lru_pvec;
1711 struct cifsFileInfo *open_file;
1712 int buf_type = CIFS_NO_BUFFER;
1715 if (file->private_data == NULL) {
1719 open_file = (struct cifsFileInfo *)file->private_data;
1720 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1721 pTcon = cifs_sb->tcon;
1723 pagevec_init(&lru_pvec, 0);
1725 for (i = 0; i < num_pages; ) {
1726 unsigned contig_pages;
1727 struct page *tmp_page;
1728 unsigned long expected_index;
1730 if (list_empty(page_list))
1733 page = list_entry(page_list->prev, struct page, lru);
1734 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1736 /* count adjacent pages that we will read into */
1739 list_entry(page_list->prev, struct page, lru)->index;
1740 list_for_each_entry_reverse(tmp_page,page_list,lru) {
1741 if (tmp_page->index == expected_index) {
1747 if (contig_pages + i > num_pages)
1748 contig_pages = num_pages - i;
1750 /* for reads over a certain size could initiate async
1753 read_size = contig_pages * PAGE_CACHE_SIZE;
1754 /* Read size needs to be in multiples of one page */
1755 read_size = min_t(const unsigned int, read_size,
1756 cifs_sb->rsize & PAGE_CACHE_MASK);
1759 while (rc == -EAGAIN) {
1760 if ((open_file->invalidHandle) &&
1761 (!open_file->closePend)) {
1762 rc = cifs_reopen_file(file, TRUE);
1767 rc = CIFSSMBRead(xid, pTcon,
1770 &bytes_read, &smb_read_data,
1772 /* BB more RC checks ? */
1774 if (smb_read_data) {
1775 if(buf_type == CIFS_SMALL_BUFFER)
1776 cifs_small_buf_release(smb_read_data);
1777 else if(buf_type == CIFS_LARGE_BUFFER)
1778 cifs_buf_release(smb_read_data);
1779 smb_read_data = NULL;
1783 if ((rc < 0) || (smb_read_data == NULL)) {
1784 cFYI(1, ("Read error in readpages: %d", rc));
1786 } else if (bytes_read > 0) {
1787 task_io_account_read(bytes_read);
1788 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1789 cifs_copy_cache_pages(mapping, page_list, bytes_read,
1790 smb_read_data + 4 /* RFC1001 hdr */ +
1791 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
1793 i += bytes_read >> PAGE_CACHE_SHIFT;
1794 cifs_stats_bytes_read(pTcon, bytes_read);
1795 if ((int)(bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1796 i++; /* account for partial page */
1798 /* server copy of file can have smaller size
1800 /* BB do we need to verify this common case ?
1801 this case is ok - if we are at server EOF
1802 we will hit it on next read */
1807 cFYI(1, ("No bytes read (%d) at offset %lld . "
1808 "Cleaning remaining pages from readahead list",
1809 bytes_read, offset));
1810 /* BB turn off caching and do new lookup on
1811 file size at server? */
1814 if (smb_read_data) {
1815 if(buf_type == CIFS_SMALL_BUFFER)
1816 cifs_small_buf_release(smb_read_data);
1817 else if(buf_type == CIFS_LARGE_BUFFER)
1818 cifs_buf_release(smb_read_data);
1819 smb_read_data = NULL;
1824 pagevec_lru_add(&lru_pvec);
1826 /* need to free smb_read_data buf before exit */
1827 if (smb_read_data) {
1828 if(buf_type == CIFS_SMALL_BUFFER)
1829 cifs_small_buf_release(smb_read_data);
1830 else if(buf_type == CIFS_LARGE_BUFFER)
1831 cifs_buf_release(smb_read_data);
1832 smb_read_data = NULL;
1839 static int cifs_readpage_worker(struct file *file, struct page *page,
1845 page_cache_get(page);
1846 read_data = kmap(page);
1847 /* for reads over a certain size could initiate async read ahead */
1849 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1854 cFYI(1, ("Bytes read %d",rc));
1856 file->f_path.dentry->d_inode->i_atime =
1857 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1859 if (PAGE_CACHE_SIZE > rc)
1860 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1862 flush_dcache_page(page);
1863 SetPageUptodate(page);
1868 page_cache_release(page);
1872 static int cifs_readpage(struct file *file, struct page *page)
1874 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1880 if (file->private_data == NULL) {
1885 cFYI(1, ("readpage %p at offset %d 0x%x\n",
1886 page, (int)offset, (int)offset));
1888 rc = cifs_readpage_worker(file, page, &offset);
1896 /* We do not want to update the file size from server for inodes
1897 open for write - to avoid races with writepage extending
1898 the file - in the future we could consider allowing
1899 refreshing the inode only on increases in the file size
1900 but this is tricky to do without racing with writebehind
1901 page caching in the current Linux kernel design */
1902 int is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
1904 struct cifsFileInfo *open_file = NULL;
1907 open_file = find_writable_file(cifsInode);
1910 struct cifs_sb_info *cifs_sb;
1912 /* there is not actually a write pending so let
1913 this handle go free and allow it to
1914 be closable if needed */
1915 atomic_dec(&open_file->wrtPending);
1917 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
1918 if ( cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO ) {
1919 /* since no page cache to corrupt on directio
1920 we can change size safely */
1924 if(i_size_read(&cifsInode->vfs_inode) < end_of_file)
1932 static int cifs_prepare_write(struct file *file, struct page *page,
1933 unsigned from, unsigned to)
1939 cFYI(1, ("prepare write for page %p from %d to %d",page,from,to));
1940 if (PageUptodate(page))
1943 /* If we are writing a full page it will be up to date,
1944 no need to read from the server */
1945 if ((to == PAGE_CACHE_SIZE) && (from == 0)) {
1946 SetPageUptodate(page);
1950 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1951 i_size = i_size_read(page->mapping->host);
1953 if ((offset >= i_size) ||
1954 ((from == 0) && (offset + to) >= i_size)) {
1956 * We don't need to read data beyond the end of the file.
1957 * zero it, and set the page uptodate
1959 void *kaddr = kmap_atomic(page, KM_USER0);
1962 memset(kaddr, 0, from);
1963 if (to < PAGE_CACHE_SIZE)
1964 memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
1965 flush_dcache_page(page);
1966 kunmap_atomic(kaddr, KM_USER0);
1967 SetPageUptodate(page);
1968 } else if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
1969 /* might as well read a page, it is fast enough */
1970 rc = cifs_readpage_worker(file, page, &offset);
1972 /* we could try using another file handle if there is one -
1973 but how would we lock it to prevent close of that handle
1974 racing with this read? In any case
1975 this will be written out by commit_write so is fine */
1978 /* we do not need to pass errors back
1979 e.g. if we do not have read access to the file
1980 because cifs_commit_write will do the right thing. -- shaggy */
1985 const struct address_space_operations cifs_addr_ops = {
1986 .readpage = cifs_readpage,
1987 .readpages = cifs_readpages,
1988 .writepage = cifs_writepage,
1989 .writepages = cifs_writepages,
1990 .prepare_write = cifs_prepare_write,
1991 .commit_write = cifs_commit_write,
1992 .set_page_dirty = __set_page_dirty_nobuffers,
1993 /* .sync_page = cifs_sync_page, */
1998 * cifs_readpages requires the server to support a buffer large enough to
1999 * contain the header plus one complete page of data. Otherwise, we need
2000 * to leave cifs_readpages out of the address space operations.
2002 const struct address_space_operations cifs_addr_ops_smallbuf = {
2003 .readpage = cifs_readpage,
2004 .writepage = cifs_writepage,
2005 .writepages = cifs_writepages,
2006 .prepare_write = cifs_prepare_write,
2007 .commit_write = cifs_commit_write,
2008 .set_page_dirty = __set_page_dirty_nobuffers,
2009 /* .sync_page = cifs_sync_page, */