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);
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 (pTcon->unix_ext) {
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;
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))
506 ("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 =
559 CIFS_SB(file->f_path.dentry->d_sb);
561 pTcon = cifs_sb->tcon;
563 cFYI(1, ("Freeing private data in close dir"));
564 if ((pCFileStruct->srch_inf.endOfSearch == FALSE) &&
565 (pCFileStruct->invalidHandle == FALSE)) {
566 pCFileStruct->invalidHandle = TRUE;
567 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
568 cFYI(1, ("Closing uncompleted readdir with rc %d",
570 /* not much we can do if it fails anyway, ignore rc */
573 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
575 cFYI(1, ("closedir free smb buf in srch struct"));
576 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
577 if (pCFileStruct->srch_inf.smallBuf)
578 cifs_small_buf_release(ptmp);
580 cifs_buf_release(ptmp);
582 ptmp = pCFileStruct->search_resume_name;
584 cFYI(1, ("closedir free resume name"));
585 pCFileStruct->search_resume_name = NULL;
588 kfree(file->private_data);
589 file->private_data = NULL;
591 /* BB can we lock the filestruct while this is going on? */
596 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
597 __u64 offset, __u8 lockType)
599 struct cifsLockInfo *li =
600 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
606 mutex_lock(&fid->lock_mutex);
607 list_add(&li->llist, &fid->llist);
608 mutex_unlock(&fid->lock_mutex);
612 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
618 int wait_flag = FALSE;
619 struct cifs_sb_info *cifs_sb;
620 struct cifsTconInfo *pTcon;
622 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
625 length = 1 + pfLock->fl_end - pfLock->fl_start;
629 cFYI(1, ("Lock parm: 0x%x flockflags: "
630 "0x%x flocktype: 0x%x start: %lld end: %lld",
631 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
634 if (pfLock->fl_flags & FL_POSIX)
636 if (pfLock->fl_flags & FL_FLOCK)
638 if (pfLock->fl_flags & FL_SLEEP) {
639 cFYI(1, ("Blocking lock"));
642 if (pfLock->fl_flags & FL_ACCESS)
643 cFYI(1, ("Process suspended by mandatory locking - "
644 "not implemented yet"));
645 if (pfLock->fl_flags & FL_LEASE)
646 cFYI(1, ("Lease on file - not implemented yet"));
647 if (pfLock->fl_flags &
648 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
649 cFYI(1, ("Unknown lock flags 0x%x", pfLock->fl_flags));
651 if (pfLock->fl_type == F_WRLCK) {
652 cFYI(1, ("F_WRLCK "));
654 } else if (pfLock->fl_type == F_UNLCK) {
655 cFYI(1, ("F_UNLCK"));
657 /* Check if unlock includes more than
659 } else if (pfLock->fl_type == F_RDLCK) {
660 cFYI(1, ("F_RDLCK"));
661 lockType |= LOCKING_ANDX_SHARED_LOCK;
663 } else if (pfLock->fl_type == F_EXLCK) {
664 cFYI(1, ("F_EXLCK"));
666 } else if (pfLock->fl_type == F_SHLCK) {
667 cFYI(1, ("F_SHLCK"));
668 lockType |= LOCKING_ANDX_SHARED_LOCK;
671 cFYI(1, ("Unknown type of lock"));
673 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
674 pTcon = cifs_sb->tcon;
676 if (file->private_data == NULL) {
680 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
682 posix_locking = (cifs_sb->tcon->ses->capabilities & CAP_UNIX) &&
683 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(cifs_sb->tcon->fsUnixInfo.Capability));
685 /* BB add code here to normalize offset and length to
686 account for negative length which we can not accept over the
691 if (lockType & LOCKING_ANDX_SHARED_LOCK)
692 posix_lock_type = CIFS_RDLCK;
694 posix_lock_type = CIFS_WRLCK;
695 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 1 /* get */,
697 posix_lock_type, wait_flag);
702 /* BB we could chain these into one lock request BB */
703 rc = CIFSSMBLock(xid, pTcon, netfid, length, pfLock->fl_start,
704 0, 1, lockType, 0 /* wait flag */ );
706 rc = CIFSSMBLock(xid, pTcon, netfid, length,
707 pfLock->fl_start, 1 /* numUnlock */ ,
708 0 /* numLock */ , lockType,
710 pfLock->fl_type = F_UNLCK;
712 cERROR(1, ("Error unlocking previously locked "
713 "range %d during test of lock", rc));
717 /* if rc == ERR_SHARING_VIOLATION ? */
718 rc = 0; /* do not change lock type to unlock
719 since range in use */
726 if (!numLock && !numUnlock) {
727 /* if no lock or unlock then nothing
728 to do since we do not know what it is */
735 if (lockType & LOCKING_ANDX_SHARED_LOCK)
736 posix_lock_type = CIFS_RDLCK;
738 posix_lock_type = CIFS_WRLCK;
741 posix_lock_type = CIFS_UNLCK;
743 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 0 /* set */,
745 posix_lock_type, wait_flag);
747 struct cifsFileInfo *fid =
748 (struct cifsFileInfo *)file->private_data;
751 rc = CIFSSMBLock(xid, pTcon, netfid, length,
753 0, numLock, lockType, wait_flag);
756 /* For Windows locks we must store them. */
757 rc = store_file_lock(fid, length,
758 pfLock->fl_start, lockType);
760 } else if (numUnlock) {
761 /* For each stored lock that this unlock overlaps
762 completely, unlock it. */
764 struct cifsLockInfo *li, *tmp;
767 mutex_lock(&fid->lock_mutex);
768 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
769 if (pfLock->fl_start <= li->offset &&
770 length >= li->length) {
771 stored_rc = CIFSSMBLock(xid, pTcon,
773 li->length, li->offset,
774 1, 0, li->type, FALSE);
778 list_del(&li->llist);
782 mutex_unlock(&fid->lock_mutex);
786 if (pfLock->fl_flags & FL_POSIX)
787 posix_lock_file_wait(file, pfLock);
792 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
793 size_t write_size, loff_t *poffset)
796 unsigned int bytes_written = 0;
797 unsigned int total_written;
798 struct cifs_sb_info *cifs_sb;
799 struct cifsTconInfo *pTcon;
801 struct cifsFileInfo *open_file;
803 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
805 pTcon = cifs_sb->tcon;
808 (" write %d bytes to offset %lld of %s", write_size,
809 *poffset, file->f_path.dentry->d_name.name)); */
811 if (file->private_data == NULL)
813 open_file = (struct cifsFileInfo *) file->private_data;
817 if (*poffset > file->f_path.dentry->d_inode->i_size)
818 long_op = 2; /* writes past end of file can take a long time */
822 for (total_written = 0; write_size > total_written;
823 total_written += bytes_written) {
825 while (rc == -EAGAIN) {
826 if (file->private_data == NULL) {
827 /* file has been closed on us */
829 /* if we have gotten here we have written some data
830 and blocked, and the file has been freed on us while
831 we blocked so return what we managed to write */
832 return total_written;
834 if (open_file->closePend) {
837 return total_written;
841 if (open_file->invalidHandle) {
842 /* we could deadlock if we called
843 filemap_fdatawait from here so tell
844 reopen_file not to flush data to server
846 rc = cifs_reopen_file(file, FALSE);
851 rc = CIFSSMBWrite(xid, pTcon,
853 min_t(const int, cifs_sb->wsize,
854 write_size - total_written),
855 *poffset, &bytes_written,
856 NULL, write_data + total_written, long_op);
858 if (rc || (bytes_written == 0)) {
866 *poffset += bytes_written;
867 long_op = FALSE; /* subsequent writes fast -
868 15 seconds is plenty */
871 cifs_stats_bytes_written(pTcon, total_written);
873 /* since the write may have blocked check these pointers again */
874 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
875 struct inode *inode = file->f_path.dentry->d_inode;
876 /* Do not update local mtime - server will set its actual value on write
877 * inode->i_ctime = inode->i_mtime =
878 * current_fs_time(inode->i_sb);*/
879 if (total_written > 0) {
880 spin_lock(&inode->i_lock);
881 if (*poffset > file->f_path.dentry->d_inode->i_size)
882 i_size_write(file->f_path.dentry->d_inode,
884 spin_unlock(&inode->i_lock);
886 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
889 return total_written;
892 static ssize_t cifs_write(struct file *file, const char *write_data,
893 size_t write_size, loff_t *poffset)
896 unsigned int bytes_written = 0;
897 unsigned int total_written;
898 struct cifs_sb_info *cifs_sb;
899 struct cifsTconInfo *pTcon;
901 struct cifsFileInfo *open_file;
903 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
905 pTcon = cifs_sb->tcon;
907 cFYI(1, ("write %zd bytes to offset %lld of %s", write_size,
908 *poffset, file->f_path.dentry->d_name.name));
910 if (file->private_data == NULL)
912 open_file = (struct cifsFileInfo *)file->private_data;
916 if (*poffset > file->f_path.dentry->d_inode->i_size)
917 long_op = 2; /* writes past end of file can take a long time */
921 for (total_written = 0; write_size > total_written;
922 total_written += bytes_written) {
924 while (rc == -EAGAIN) {
925 if (file->private_data == NULL) {
926 /* file has been closed on us */
928 /* if we have gotten here we have written some data
929 and blocked, and the file has been freed on us
930 while we blocked so return what we managed to
932 return total_written;
934 if (open_file->closePend) {
937 return total_written;
941 if (open_file->invalidHandle) {
942 /* we could deadlock if we called
943 filemap_fdatawait from here so tell
944 reopen_file not to flush data to
946 rc = cifs_reopen_file(file, FALSE);
950 if (experimEnabled || (pTcon->ses->server &&
951 ((pTcon->ses->server->secMode &
952 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
957 len = min((size_t)cifs_sb->wsize,
958 write_size - total_written);
959 /* iov[0] is reserved for smb header */
960 iov[1].iov_base = (char *)write_data +
962 iov[1].iov_len = len;
963 rc = CIFSSMBWrite2(xid, pTcon,
964 open_file->netfid, len,
965 *poffset, &bytes_written,
968 rc = CIFSSMBWrite(xid, pTcon,
970 min_t(const int, cifs_sb->wsize,
971 write_size - total_written),
972 *poffset, &bytes_written,
973 write_data + total_written,
976 if (rc || (bytes_written == 0)) {
984 *poffset += bytes_written;
985 long_op = FALSE; /* subsequent writes fast -
986 15 seconds is plenty */
989 cifs_stats_bytes_written(pTcon, total_written);
991 /* since the write may have blocked check these pointers again */
992 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
993 /*BB We could make this contingent on superblock ATIME flag too */
994 /* file->f_path.dentry->d_inode->i_ctime =
995 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
996 if (total_written > 0) {
997 spin_lock(&file->f_path.dentry->d_inode->i_lock);
998 if (*poffset > file->f_path.dentry->d_inode->i_size)
999 i_size_write(file->f_path.dentry->d_inode,
1001 spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1003 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1006 return total_written;
1009 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1011 struct cifsFileInfo *open_file;
1014 /* Having a null inode here (because mapping->host was set to zero by
1015 the VFS or MM) should not happen but we had reports of on oops (due to
1016 it being zero) during stress testcases so we need to check for it */
1018 if (cifs_inode == NULL) {
1019 cERROR(1, ("Null inode passed to cifs_writeable_file"));
1024 read_lock(&GlobalSMBSeslock);
1025 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1026 if (open_file->closePend)
1028 if (open_file->pfile &&
1029 ((open_file->pfile->f_flags & O_RDWR) ||
1030 (open_file->pfile->f_flags & O_WRONLY))) {
1031 atomic_inc(&open_file->wrtPending);
1032 read_unlock(&GlobalSMBSeslock);
1033 if ((open_file->invalidHandle) &&
1034 (!open_file->closePend) /* BB fixme -since the second clause can not be true remove it BB */) {
1035 rc = cifs_reopen_file(open_file->pfile, FALSE);
1036 /* if it fails, try another handle - might be */
1037 /* dangerous to hold up writepages with retry */
1040 ("failed on reopen file in wp"));
1041 read_lock(&GlobalSMBSeslock);
1042 /* can not use this handle, no write
1043 pending on this one after all */
1045 (&open_file->wrtPending);
1052 read_unlock(&GlobalSMBSeslock);
1056 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1058 struct address_space *mapping = page->mapping;
1059 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1062 int bytes_written = 0;
1063 struct cifs_sb_info *cifs_sb;
1064 struct cifsTconInfo *pTcon;
1065 struct inode *inode;
1066 struct cifsFileInfo *open_file;
1068 if (!mapping || !mapping->host)
1071 inode = page->mapping->host;
1072 cifs_sb = CIFS_SB(inode->i_sb);
1073 pTcon = cifs_sb->tcon;
1075 offset += (loff_t)from;
1076 write_data = kmap(page);
1079 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1084 /* racing with truncate? */
1085 if (offset > mapping->host->i_size) {
1087 return 0; /* don't care */
1090 /* check to make sure that we are not extending the file */
1091 if (mapping->host->i_size - offset < (loff_t)to)
1092 to = (unsigned)(mapping->host->i_size - offset);
1094 open_file = find_writable_file(CIFS_I(mapping->host));
1096 bytes_written = cifs_write(open_file->pfile, write_data,
1098 atomic_dec(&open_file->wrtPending);
1099 /* Does mm or vfs already set times? */
1100 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1101 if ((bytes_written > 0) && (offset)) {
1103 } else if (bytes_written < 0) {
1108 cFYI(1, ("No writeable filehandles for inode"));
1116 static int cifs_writepages(struct address_space *mapping,
1117 struct writeback_control *wbc)
1119 struct backing_dev_info *bdi = mapping->backing_dev_info;
1120 unsigned int bytes_to_write;
1121 unsigned int bytes_written;
1122 struct cifs_sb_info *cifs_sb;
1126 int range_whole = 0;
1133 struct cifsFileInfo *open_file;
1135 struct pagevec pvec;
1140 cifs_sb = CIFS_SB(mapping->host->i_sb);
1143 * If wsize is smaller that the page cache size, default to writing
1144 * one page at a time via cifs_writepage
1146 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1147 return generic_writepages(mapping, wbc);
1149 if ((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1150 if (cifs_sb->tcon->ses->server->secMode &
1151 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1152 if (!experimEnabled)
1153 return generic_writepages(mapping, wbc);
1155 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1157 return generic_writepages(mapping, wbc);
1161 * BB: Is this meaningful for a non-block-device file system?
1162 * If it is, we should test it again after we do I/O
1164 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1165 wbc->encountered_congestion = 1;
1172 pagevec_init(&pvec, 0);
1173 if (wbc->range_cyclic) {
1174 index = mapping->writeback_index; /* Start from prev offset */
1177 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1178 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1179 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1184 while (!done && (index <= end) &&
1185 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1186 PAGECACHE_TAG_DIRTY,
1187 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1196 for (i = 0; i < nr_pages; i++) {
1197 page = pvec.pages[i];
1199 * At this point we hold neither mapping->tree_lock nor
1200 * lock on the page itself: the page may be truncated or
1201 * invalidated (changing page->mapping to NULL), or even
1202 * swizzled back from swapper_space to tmpfs file
1208 else if (TestSetPageLocked(page))
1211 if (unlikely(page->mapping != mapping)) {
1216 if (!wbc->range_cyclic && page->index > end) {
1222 if (next && (page->index != next)) {
1223 /* Not next consecutive page */
1228 if (wbc->sync_mode != WB_SYNC_NONE)
1229 wait_on_page_writeback(page);
1231 if (PageWriteback(page) ||
1232 !clear_page_dirty_for_io(page)) {
1238 * This actually clears the dirty bit in the radix tree.
1239 * See cifs_writepage() for more commentary.
1241 set_page_writeback(page);
1243 if (page_offset(page) >= mapping->host->i_size) {
1246 end_page_writeback(page);
1251 * BB can we get rid of this? pages are held by pvec
1253 page_cache_get(page);
1255 len = min(mapping->host->i_size - page_offset(page),
1256 (loff_t)PAGE_CACHE_SIZE);
1258 /* reserve iov[0] for the smb header */
1260 iov[n_iov].iov_base = kmap(page);
1261 iov[n_iov].iov_len = len;
1262 bytes_to_write += len;
1266 offset = page_offset(page);
1268 next = page->index + 1;
1269 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1273 /* Search for a writable handle every time we call
1274 * CIFSSMBWrite2. We can't rely on the last handle
1275 * we used to still be valid
1277 open_file = find_writable_file(CIFS_I(mapping->host));
1279 cERROR(1, ("No writable handles for inode"));
1282 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1284 bytes_to_write, offset,
1285 &bytes_written, iov, n_iov,
1287 atomic_dec(&open_file->wrtPending);
1288 if (rc || bytes_written < bytes_to_write) {
1289 cERROR(1, ("Write2 ret %d, wrote %d",
1290 rc, bytes_written));
1291 /* BB what if continued retry is
1292 requested via mount flags? */
1293 set_bit(AS_EIO, &mapping->flags);
1295 cifs_stats_bytes_written(cifs_sb->tcon,
1299 for (i = 0; i < n_iov; i++) {
1300 page = pvec.pages[first + i];
1301 /* Should we also set page error on
1302 success rc but too little data written? */
1303 /* BB investigate retry logic on temporary
1304 server crash cases and how recovery works
1305 when page marked as error */
1310 end_page_writeback(page);
1311 page_cache_release(page);
1313 if ((wbc->nr_to_write -= n_iov) <= 0)
1317 pagevec_release(&pvec);
1319 if (!scanned && !done) {
1321 * We hit the last page and there is more work to be done: wrap
1322 * back to the start of the file
1328 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1329 mapping->writeback_index = index;
1336 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1342 /* BB add check for wbc flags */
1343 page_cache_get(page);
1344 if (!PageUptodate(page)) {
1345 cFYI(1, ("ppw - page not up to date"));
1349 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1351 * A writepage() implementation always needs to do either this,
1352 * or re-dirty the page with "redirty_page_for_writepage()" in
1353 * the case of a failure.
1355 * Just unlocking the page will cause the radix tree tag-bits
1356 * to fail to update with the state of the page correctly.
1358 set_page_writeback(page);
1359 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1360 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1362 end_page_writeback(page);
1363 page_cache_release(page);
1368 static int cifs_commit_write(struct file *file, struct page *page,
1369 unsigned offset, unsigned to)
1373 struct inode *inode = page->mapping->host;
1374 loff_t position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1378 cFYI(1, ("commit write for page %p up to position %lld for %d",
1379 page, position, to));
1380 spin_lock(&inode->i_lock);
1381 if (position > inode->i_size) {
1382 i_size_write(inode, position);
1384 spin_unlock(&inode->i_lock);
1385 if (!PageUptodate(page)) {
1386 position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + offset;
1387 /* can not rely on (or let) writepage write this data */
1389 cFYI(1, ("Illegal offsets, can not copy from %d to %d",
1394 /* this is probably better than directly calling
1395 partialpage_write since in this function the file handle is
1396 known which we might as well leverage */
1397 /* BB check if anything else missing out of ppw
1398 such as updating last write time */
1399 page_data = kmap(page);
1400 rc = cifs_write(file, page_data + offset, to-offset,
1404 /* else if (rc < 0) should we set writebehind rc? */
1407 set_page_dirty(page);
1414 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1418 struct inode *inode = file->f_path.dentry->d_inode;
1422 cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1423 dentry->d_name.name, datasync));
1425 rc = filemap_fdatawrite(inode->i_mapping);
1427 CIFS_I(inode)->write_behind_rc = 0;
1432 /* static void cifs_sync_page(struct page *page)
1434 struct address_space *mapping;
1435 struct inode *inode;
1436 unsigned long index = page->index;
1437 unsigned int rpages = 0;
1440 cFYI(1, ("sync page %p",page));
1441 mapping = page->mapping;
1444 inode = mapping->host;
1448 /* fill in rpages then
1449 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1451 /* cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1461 * As file closes, flush all cached write data for this inode checking
1462 * for write behind errors.
1464 int cifs_flush(struct file *file, fl_owner_t id)
1466 struct inode *inode = file->f_path.dentry->d_inode;
1469 /* Rather than do the steps manually:
1470 lock the inode for writing
1471 loop through pages looking for write behind data (dirty pages)
1472 coalesce into contiguous 16K (or smaller) chunks to write to server
1473 send to server (prefer in parallel)
1474 deal with writebehind errors
1475 unlock inode for writing
1476 filemapfdatawrite appears easier for the time being */
1478 rc = filemap_fdatawrite(inode->i_mapping);
1479 if (!rc) /* reset wb rc if we were able to write out dirty pages */
1480 CIFS_I(inode)->write_behind_rc = 0;
1482 cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
1487 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1488 size_t read_size, loff_t *poffset)
1491 unsigned int bytes_read = 0;
1492 unsigned int total_read = 0;
1493 unsigned int current_read_size;
1494 struct cifs_sb_info *cifs_sb;
1495 struct cifsTconInfo *pTcon;
1497 struct cifsFileInfo *open_file;
1498 char *smb_read_data;
1499 char __user *current_offset;
1500 struct smb_com_read_rsp *pSMBr;
1503 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1504 pTcon = cifs_sb->tcon;
1506 if (file->private_data == NULL) {
1510 open_file = (struct cifsFileInfo *)file->private_data;
1512 if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
1513 cFYI(1, ("attempting read on write only file instance"));
1515 for (total_read = 0, current_offset = read_data;
1516 read_size > total_read;
1517 total_read += bytes_read, current_offset += bytes_read) {
1518 current_read_size = min_t(const int, read_size - total_read,
1521 smb_read_data = NULL;
1522 while (rc == -EAGAIN) {
1523 int buf_type = CIFS_NO_BUFFER;
1524 if ((open_file->invalidHandle) &&
1525 (!open_file->closePend)) {
1526 rc = cifs_reopen_file(file, TRUE);
1530 rc = CIFSSMBRead(xid, pTcon,
1532 current_read_size, *poffset,
1533 &bytes_read, &smb_read_data,
1535 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1536 if (smb_read_data) {
1537 if (copy_to_user(current_offset,
1539 4 /* RFC1001 length field */ +
1540 le16_to_cpu(pSMBr->DataOffset),
1545 if (buf_type == CIFS_SMALL_BUFFER)
1546 cifs_small_buf_release(smb_read_data);
1547 else if (buf_type == CIFS_LARGE_BUFFER)
1548 cifs_buf_release(smb_read_data);
1549 smb_read_data = NULL;
1552 if (rc || (bytes_read == 0)) {
1560 cifs_stats_bytes_read(pTcon, bytes_read);
1561 *poffset += bytes_read;
1569 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1573 unsigned int bytes_read = 0;
1574 unsigned int total_read;
1575 unsigned int current_read_size;
1576 struct cifs_sb_info *cifs_sb;
1577 struct cifsTconInfo *pTcon;
1579 char *current_offset;
1580 struct cifsFileInfo *open_file;
1581 int buf_type = CIFS_NO_BUFFER;
1584 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1585 pTcon = cifs_sb->tcon;
1587 if (file->private_data == NULL) {
1591 open_file = (struct cifsFileInfo *)file->private_data;
1593 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1594 cFYI(1, ("attempting read on write only file instance"));
1596 for (total_read = 0, current_offset = read_data;
1597 read_size > total_read;
1598 total_read += bytes_read, current_offset += bytes_read) {
1599 current_read_size = min_t(const int, read_size - total_read,
1601 /* For windows me and 9x we do not want to request more
1602 than it negotiated since it will refuse the read then */
1604 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1605 current_read_size = min_t(const int, current_read_size,
1606 pTcon->ses->server->maxBuf - 128);
1609 while (rc == -EAGAIN) {
1610 if ((open_file->invalidHandle) &&
1611 (!open_file->closePend)) {
1612 rc = cifs_reopen_file(file, TRUE);
1616 rc = CIFSSMBRead(xid, pTcon,
1618 current_read_size, *poffset,
1619 &bytes_read, ¤t_offset,
1622 if (rc || (bytes_read == 0)) {
1630 cifs_stats_bytes_read(pTcon, total_read);
1631 *poffset += bytes_read;
1638 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1640 struct dentry *dentry = file->f_path.dentry;
1644 rc = cifs_revalidate(dentry);
1646 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1650 rc = generic_file_mmap(file, vma);
1656 static void cifs_copy_cache_pages(struct address_space *mapping,
1657 struct list_head *pages, int bytes_read, char *data,
1658 struct pagevec *plru_pvec)
1663 while (bytes_read > 0) {
1664 if (list_empty(pages))
1667 page = list_entry(pages->prev, struct page, lru);
1668 list_del(&page->lru);
1670 if (add_to_page_cache(page, mapping, page->index,
1672 page_cache_release(page);
1673 cFYI(1, ("Add page cache failed"));
1674 data += PAGE_CACHE_SIZE;
1675 bytes_read -= PAGE_CACHE_SIZE;
1679 target = kmap_atomic(page, KM_USER0);
1681 if (PAGE_CACHE_SIZE > bytes_read) {
1682 memcpy(target, data, bytes_read);
1683 /* zero the tail end of this partial page */
1684 memset(target + bytes_read, 0,
1685 PAGE_CACHE_SIZE - bytes_read);
1688 memcpy(target, data, PAGE_CACHE_SIZE);
1689 bytes_read -= PAGE_CACHE_SIZE;
1691 kunmap_atomic(target, KM_USER0);
1693 flush_dcache_page(page);
1694 SetPageUptodate(page);
1696 if (!pagevec_add(plru_pvec, page))
1697 __pagevec_lru_add(plru_pvec);
1698 data += PAGE_CACHE_SIZE;
1703 static int cifs_readpages(struct file *file, struct address_space *mapping,
1704 struct list_head *page_list, unsigned num_pages)
1710 struct cifs_sb_info *cifs_sb;
1711 struct cifsTconInfo *pTcon;
1713 unsigned int read_size, i;
1714 char *smb_read_data = NULL;
1715 struct smb_com_read_rsp *pSMBr;
1716 struct pagevec lru_pvec;
1717 struct cifsFileInfo *open_file;
1718 int buf_type = CIFS_NO_BUFFER;
1721 if (file->private_data == NULL) {
1725 open_file = (struct cifsFileInfo *)file->private_data;
1726 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1727 pTcon = cifs_sb->tcon;
1729 pagevec_init(&lru_pvec, 0);
1730 #ifdef CONFIG_CIFS_DEBUG2
1731 cFYI(1, ("rpages: num pages %d", num_pages));
1733 for (i = 0; i < num_pages; ) {
1734 unsigned contig_pages;
1735 struct page *tmp_page;
1736 unsigned long expected_index;
1738 if (list_empty(page_list))
1741 page = list_entry(page_list->prev, struct page, lru);
1742 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1744 /* count adjacent pages that we will read into */
1747 list_entry(page_list->prev, struct page, lru)->index;
1748 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1749 if (tmp_page->index == expected_index) {
1755 if (contig_pages + i > num_pages)
1756 contig_pages = num_pages - i;
1758 /* for reads over a certain size could initiate async
1761 read_size = contig_pages * PAGE_CACHE_SIZE;
1762 /* Read size needs to be in multiples of one page */
1763 read_size = min_t(const unsigned int, read_size,
1764 cifs_sb->rsize & PAGE_CACHE_MASK);
1765 #ifdef CONFIG_CIFS_DEBUG2
1766 cFYI(1, ("rpages: read size 0x%x contiguous pages %d",
1767 read_size, contig_pages));
1770 while (rc == -EAGAIN) {
1771 if ((open_file->invalidHandle) &&
1772 (!open_file->closePend)) {
1773 rc = cifs_reopen_file(file, TRUE);
1778 rc = CIFSSMBRead(xid, pTcon,
1781 &bytes_read, &smb_read_data,
1783 /* BB more RC checks ? */
1784 if (rc == -EAGAIN) {
1785 if (smb_read_data) {
1786 if (buf_type == CIFS_SMALL_BUFFER)
1787 cifs_small_buf_release(smb_read_data);
1788 else if (buf_type == CIFS_LARGE_BUFFER)
1789 cifs_buf_release(smb_read_data);
1790 smb_read_data = NULL;
1794 if ((rc < 0) || (smb_read_data == NULL)) {
1795 cFYI(1, ("Read error in readpages: %d", rc));
1797 } else if (bytes_read > 0) {
1798 task_io_account_read(bytes_read);
1799 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1800 cifs_copy_cache_pages(mapping, page_list, bytes_read,
1801 smb_read_data + 4 /* RFC1001 hdr */ +
1802 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
1804 i += bytes_read >> PAGE_CACHE_SHIFT;
1805 cifs_stats_bytes_read(pTcon, bytes_read);
1806 if ((int)(bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1807 i++; /* account for partial page */
1809 /* server copy of file can have smaller size
1811 /* BB do we need to verify this common case ?
1812 this case is ok - if we are at server EOF
1813 we will hit it on next read */
1818 cFYI(1, ("No bytes read (%d) at offset %lld . "
1819 "Cleaning remaining pages from readahead list",
1820 bytes_read, offset));
1821 /* BB turn off caching and do new lookup on
1822 file size at server? */
1825 if (smb_read_data) {
1826 if (buf_type == CIFS_SMALL_BUFFER)
1827 cifs_small_buf_release(smb_read_data);
1828 else if (buf_type == CIFS_LARGE_BUFFER)
1829 cifs_buf_release(smb_read_data);
1830 smb_read_data = NULL;
1835 pagevec_lru_add(&lru_pvec);
1837 /* need to free smb_read_data buf before exit */
1838 if (smb_read_data) {
1839 if (buf_type == CIFS_SMALL_BUFFER)
1840 cifs_small_buf_release(smb_read_data);
1841 else if (buf_type == CIFS_LARGE_BUFFER)
1842 cifs_buf_release(smb_read_data);
1843 smb_read_data = NULL;
1850 static int cifs_readpage_worker(struct file *file, struct page *page,
1856 page_cache_get(page);
1857 read_data = kmap(page);
1858 /* for reads over a certain size could initiate async read ahead */
1860 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1865 cFYI(1, ("Bytes read %d", rc));
1867 file->f_path.dentry->d_inode->i_atime =
1868 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1870 if (PAGE_CACHE_SIZE > rc)
1871 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1873 flush_dcache_page(page);
1874 SetPageUptodate(page);
1879 page_cache_release(page);
1883 static int cifs_readpage(struct file *file, struct page *page)
1885 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1891 if (file->private_data == NULL) {
1896 cFYI(1, ("readpage %p at offset %d 0x%x\n",
1897 page, (int)offset, (int)offset));
1899 rc = cifs_readpage_worker(file, page, &offset);
1907 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
1909 struct cifsFileInfo *open_file;
1911 read_lock(&GlobalSMBSeslock);
1912 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1913 if (open_file->closePend)
1915 if (open_file->pfile &&
1916 ((open_file->pfile->f_flags & O_RDWR) ||
1917 (open_file->pfile->f_flags & O_WRONLY))) {
1918 read_unlock(&GlobalSMBSeslock);
1922 read_unlock(&GlobalSMBSeslock);
1926 /* We do not want to update the file size from server for inodes
1927 open for write - to avoid races with writepage extending
1928 the file - in the future we could consider allowing
1929 refreshing the inode only on increases in the file size
1930 but this is tricky to do without racing with writebehind
1931 page caching in the current Linux kernel design */
1932 int is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
1937 if (is_inode_writable(cifsInode)) {
1938 /* This inode is open for write at least once */
1939 struct cifs_sb_info *cifs_sb;
1941 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
1942 if ( cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO ) {
1943 /* since no page cache to corrupt on directio
1944 we can change size safely */
1948 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
1956 static int cifs_prepare_write(struct file *file, struct page *page,
1957 unsigned from, unsigned to)
1963 cFYI(1, ("prepare write for page %p from %d to %d", page, from, to));
1964 if (PageUptodate(page))
1967 /* If we are writing a full page it will be up to date,
1968 no need to read from the server */
1969 if ((to == PAGE_CACHE_SIZE) && (from == 0)) {
1970 SetPageUptodate(page);
1974 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1975 i_size = i_size_read(page->mapping->host);
1977 if ((offset >= i_size) ||
1978 ((from == 0) && (offset + to) >= i_size)) {
1980 * We don't need to read data beyond the end of the file.
1981 * zero it, and set the page uptodate
1983 simple_prepare_write(file, page, from, to);
1984 SetPageUptodate(page);
1985 } else if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
1986 /* might as well read a page, it is fast enough */
1987 rc = cifs_readpage_worker(file, page, &offset);
1989 /* we could try using another file handle if there is one -
1990 but how would we lock it to prevent close of that handle
1991 racing with this read? In any case
1992 this will be written out by commit_write so is fine */
1995 /* we do not need to pass errors back
1996 e.g. if we do not have read access to the file
1997 because cifs_commit_write will do the right thing. -- shaggy */
2002 const struct address_space_operations cifs_addr_ops = {
2003 .readpage = cifs_readpage,
2004 .readpages = cifs_readpages,
2005 .writepage = cifs_writepage,
2006 .writepages = cifs_writepages,
2007 .prepare_write = cifs_prepare_write,
2008 .commit_write = cifs_commit_write,
2009 .set_page_dirty = __set_page_dirty_nobuffers,
2010 /* .sync_page = cifs_sync_page, */
2015 * cifs_readpages requires the server to support a buffer large enough to
2016 * contain the header plus one complete page of data. Otherwise, we need
2017 * to leave cifs_readpages out of the address space operations.
2019 const struct address_space_operations cifs_addr_ops_smallbuf = {
2020 .readpage = cifs_readpage,
2021 .writepage = cifs_writepage,
2022 .writepages = cifs_writepages,
2023 .prepare_write = cifs_prepare_write,
2024 .commit_write = cifs_commit_write,
2025 .set_page_dirty = __set_page_dirty_nobuffers,
2026 /* .sync_page = cifs_sync_page, */