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 (pfLock->fl_start + length) >=
771 (li->offset + li->length)) {
772 stored_rc = CIFSSMBLock(xid, pTcon,
774 li->length, li->offset,
775 1, 0, li->type, FALSE);
779 list_del(&li->llist);
783 mutex_unlock(&fid->lock_mutex);
787 if (pfLock->fl_flags & FL_POSIX)
788 posix_lock_file_wait(file, pfLock);
793 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
794 size_t write_size, loff_t *poffset)
797 unsigned int bytes_written = 0;
798 unsigned int total_written;
799 struct cifs_sb_info *cifs_sb;
800 struct cifsTconInfo *pTcon;
802 struct cifsFileInfo *open_file;
804 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
806 pTcon = cifs_sb->tcon;
809 (" write %d bytes to offset %lld of %s", write_size,
810 *poffset, file->f_path.dentry->d_name.name)); */
812 if (file->private_data == NULL)
814 open_file = (struct cifsFileInfo *) file->private_data;
818 if (*poffset > file->f_path.dentry->d_inode->i_size)
819 long_op = 2; /* writes past end of file can take a long time */
823 for (total_written = 0; write_size > total_written;
824 total_written += bytes_written) {
826 while (rc == -EAGAIN) {
827 if (file->private_data == NULL) {
828 /* file has been closed on us */
830 /* if we have gotten here we have written some data
831 and blocked, and the file has been freed on us while
832 we blocked so return what we managed to write */
833 return total_written;
835 if (open_file->closePend) {
838 return total_written;
842 if (open_file->invalidHandle) {
843 /* we could deadlock if we called
844 filemap_fdatawait from here so tell
845 reopen_file not to flush data to server
847 rc = cifs_reopen_file(file, FALSE);
852 rc = CIFSSMBWrite(xid, pTcon,
854 min_t(const int, cifs_sb->wsize,
855 write_size - total_written),
856 *poffset, &bytes_written,
857 NULL, write_data + total_written, long_op);
859 if (rc || (bytes_written == 0)) {
867 *poffset += bytes_written;
868 long_op = FALSE; /* subsequent writes fast -
869 15 seconds is plenty */
872 cifs_stats_bytes_written(pTcon, total_written);
874 /* since the write may have blocked check these pointers again */
875 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
876 struct inode *inode = file->f_path.dentry->d_inode;
877 /* Do not update local mtime - server will set its actual value on write
878 * inode->i_ctime = inode->i_mtime =
879 * current_fs_time(inode->i_sb);*/
880 if (total_written > 0) {
881 spin_lock(&inode->i_lock);
882 if (*poffset > file->f_path.dentry->d_inode->i_size)
883 i_size_write(file->f_path.dentry->d_inode,
885 spin_unlock(&inode->i_lock);
887 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
890 return total_written;
893 static ssize_t cifs_write(struct file *file, const char *write_data,
894 size_t write_size, loff_t *poffset)
897 unsigned int bytes_written = 0;
898 unsigned int total_written;
899 struct cifs_sb_info *cifs_sb;
900 struct cifsTconInfo *pTcon;
902 struct cifsFileInfo *open_file;
904 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
906 pTcon = cifs_sb->tcon;
908 cFYI(1, ("write %zd bytes to offset %lld of %s", write_size,
909 *poffset, file->f_path.dentry->d_name.name));
911 if (file->private_data == NULL)
913 open_file = (struct cifsFileInfo *)file->private_data;
917 if (*poffset > file->f_path.dentry->d_inode->i_size)
918 long_op = 2; /* writes past end of file can take a long time */
922 for (total_written = 0; write_size > total_written;
923 total_written += bytes_written) {
925 while (rc == -EAGAIN) {
926 if (file->private_data == NULL) {
927 /* file has been closed on us */
929 /* if we have gotten here we have written some data
930 and blocked, and the file has been freed on us
931 while we blocked so return what we managed to
933 return total_written;
935 if (open_file->closePend) {
938 return total_written;
942 if (open_file->invalidHandle) {
943 /* we could deadlock if we called
944 filemap_fdatawait from here so tell
945 reopen_file not to flush data to
947 rc = cifs_reopen_file(file, FALSE);
951 if (experimEnabled || (pTcon->ses->server &&
952 ((pTcon->ses->server->secMode &
953 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
958 len = min((size_t)cifs_sb->wsize,
959 write_size - total_written);
960 /* iov[0] is reserved for smb header */
961 iov[1].iov_base = (char *)write_data +
963 iov[1].iov_len = len;
964 rc = CIFSSMBWrite2(xid, pTcon,
965 open_file->netfid, len,
966 *poffset, &bytes_written,
969 rc = CIFSSMBWrite(xid, pTcon,
971 min_t(const int, cifs_sb->wsize,
972 write_size - total_written),
973 *poffset, &bytes_written,
974 write_data + total_written,
977 if (rc || (bytes_written == 0)) {
985 *poffset += bytes_written;
986 long_op = FALSE; /* subsequent writes fast -
987 15 seconds is plenty */
990 cifs_stats_bytes_written(pTcon, total_written);
992 /* since the write may have blocked check these pointers again */
993 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
994 /*BB We could make this contingent on superblock ATIME flag too */
995 /* file->f_path.dentry->d_inode->i_ctime =
996 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
997 if (total_written > 0) {
998 spin_lock(&file->f_path.dentry->d_inode->i_lock);
999 if (*poffset > file->f_path.dentry->d_inode->i_size)
1000 i_size_write(file->f_path.dentry->d_inode,
1002 spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1004 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1007 return total_written;
1010 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1012 struct cifsFileInfo *open_file;
1015 /* Having a null inode here (because mapping->host was set to zero by
1016 the VFS or MM) should not happen but we had reports of on oops (due to
1017 it being zero) during stress testcases so we need to check for it */
1019 if (cifs_inode == NULL) {
1020 cERROR(1, ("Null inode passed to cifs_writeable_file"));
1025 read_lock(&GlobalSMBSeslock);
1026 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1027 if (open_file->closePend)
1029 if (open_file->pfile &&
1030 ((open_file->pfile->f_flags & O_RDWR) ||
1031 (open_file->pfile->f_flags & O_WRONLY))) {
1032 atomic_inc(&open_file->wrtPending);
1033 read_unlock(&GlobalSMBSeslock);
1034 if ((open_file->invalidHandle) &&
1035 (!open_file->closePend) /* BB fixme -since the second clause can not be true remove it BB */) {
1036 rc = cifs_reopen_file(open_file->pfile, FALSE);
1037 /* if it fails, try another handle - might be */
1038 /* dangerous to hold up writepages with retry */
1041 ("failed on reopen file in wp"));
1042 read_lock(&GlobalSMBSeslock);
1043 /* can not use this handle, no write
1044 pending on this one after all */
1046 (&open_file->wrtPending);
1053 read_unlock(&GlobalSMBSeslock);
1057 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1059 struct address_space *mapping = page->mapping;
1060 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1063 int bytes_written = 0;
1064 struct cifs_sb_info *cifs_sb;
1065 struct cifsTconInfo *pTcon;
1066 struct inode *inode;
1067 struct cifsFileInfo *open_file;
1069 if (!mapping || !mapping->host)
1072 inode = page->mapping->host;
1073 cifs_sb = CIFS_SB(inode->i_sb);
1074 pTcon = cifs_sb->tcon;
1076 offset += (loff_t)from;
1077 write_data = kmap(page);
1080 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1085 /* racing with truncate? */
1086 if (offset > mapping->host->i_size) {
1088 return 0; /* don't care */
1091 /* check to make sure that we are not extending the file */
1092 if (mapping->host->i_size - offset < (loff_t)to)
1093 to = (unsigned)(mapping->host->i_size - offset);
1095 open_file = find_writable_file(CIFS_I(mapping->host));
1097 bytes_written = cifs_write(open_file->pfile, write_data,
1099 atomic_dec(&open_file->wrtPending);
1100 /* Does mm or vfs already set times? */
1101 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1102 if ((bytes_written > 0) && (offset)) {
1104 } else if (bytes_written < 0) {
1109 cFYI(1, ("No writeable filehandles for inode"));
1117 static int cifs_writepages(struct address_space *mapping,
1118 struct writeback_control *wbc)
1120 struct backing_dev_info *bdi = mapping->backing_dev_info;
1121 unsigned int bytes_to_write;
1122 unsigned int bytes_written;
1123 struct cifs_sb_info *cifs_sb;
1127 int range_whole = 0;
1134 struct cifsFileInfo *open_file;
1136 struct pagevec pvec;
1141 cifs_sb = CIFS_SB(mapping->host->i_sb);
1144 * If wsize is smaller that the page cache size, default to writing
1145 * one page at a time via cifs_writepage
1147 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1148 return generic_writepages(mapping, wbc);
1150 if ((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1151 if (cifs_sb->tcon->ses->server->secMode &
1152 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1153 if (!experimEnabled)
1154 return generic_writepages(mapping, wbc);
1156 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1158 return generic_writepages(mapping, wbc);
1162 * BB: Is this meaningful for a non-block-device file system?
1163 * If it is, we should test it again after we do I/O
1165 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1166 wbc->encountered_congestion = 1;
1173 pagevec_init(&pvec, 0);
1174 if (wbc->range_cyclic) {
1175 index = mapping->writeback_index; /* Start from prev offset */
1178 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1179 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1180 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1185 while (!done && (index <= end) &&
1186 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1187 PAGECACHE_TAG_DIRTY,
1188 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1197 for (i = 0; i < nr_pages; i++) {
1198 page = pvec.pages[i];
1200 * At this point we hold neither mapping->tree_lock nor
1201 * lock on the page itself: the page may be truncated or
1202 * invalidated (changing page->mapping to NULL), or even
1203 * swizzled back from swapper_space to tmpfs file
1209 else if (TestSetPageLocked(page))
1212 if (unlikely(page->mapping != mapping)) {
1217 if (!wbc->range_cyclic && page->index > end) {
1223 if (next && (page->index != next)) {
1224 /* Not next consecutive page */
1229 if (wbc->sync_mode != WB_SYNC_NONE)
1230 wait_on_page_writeback(page);
1232 if (PageWriteback(page) ||
1233 !clear_page_dirty_for_io(page)) {
1239 * This actually clears the dirty bit in the radix tree.
1240 * See cifs_writepage() for more commentary.
1242 set_page_writeback(page);
1244 if (page_offset(page) >= mapping->host->i_size) {
1247 end_page_writeback(page);
1252 * BB can we get rid of this? pages are held by pvec
1254 page_cache_get(page);
1256 len = min(mapping->host->i_size - page_offset(page),
1257 (loff_t)PAGE_CACHE_SIZE);
1259 /* reserve iov[0] for the smb header */
1261 iov[n_iov].iov_base = kmap(page);
1262 iov[n_iov].iov_len = len;
1263 bytes_to_write += len;
1267 offset = page_offset(page);
1269 next = page->index + 1;
1270 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1274 /* Search for a writable handle every time we call
1275 * CIFSSMBWrite2. We can't rely on the last handle
1276 * we used to still be valid
1278 open_file = find_writable_file(CIFS_I(mapping->host));
1280 cERROR(1, ("No writable handles for inode"));
1283 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1285 bytes_to_write, offset,
1286 &bytes_written, iov, n_iov,
1288 atomic_dec(&open_file->wrtPending);
1289 if (rc || bytes_written < bytes_to_write) {
1290 cERROR(1, ("Write2 ret %d, wrote %d",
1291 rc, bytes_written));
1292 /* BB what if continued retry is
1293 requested via mount flags? */
1294 set_bit(AS_EIO, &mapping->flags);
1296 cifs_stats_bytes_written(cifs_sb->tcon,
1300 for (i = 0; i < n_iov; i++) {
1301 page = pvec.pages[first + i];
1302 /* Should we also set page error on
1303 success rc but too little data written? */
1304 /* BB investigate retry logic on temporary
1305 server crash cases and how recovery works
1306 when page marked as error */
1311 end_page_writeback(page);
1312 page_cache_release(page);
1314 if ((wbc->nr_to_write -= n_iov) <= 0)
1318 pagevec_release(&pvec);
1320 if (!scanned && !done) {
1322 * We hit the last page and there is more work to be done: wrap
1323 * back to the start of the file
1329 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1330 mapping->writeback_index = index;
1337 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1343 /* BB add check for wbc flags */
1344 page_cache_get(page);
1345 if (!PageUptodate(page)) {
1346 cFYI(1, ("ppw - page not up to date"));
1350 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1352 * A writepage() implementation always needs to do either this,
1353 * or re-dirty the page with "redirty_page_for_writepage()" in
1354 * the case of a failure.
1356 * Just unlocking the page will cause the radix tree tag-bits
1357 * to fail to update with the state of the page correctly.
1359 set_page_writeback(page);
1360 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1361 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1363 end_page_writeback(page);
1364 page_cache_release(page);
1369 static int cifs_commit_write(struct file *file, struct page *page,
1370 unsigned offset, unsigned to)
1374 struct inode *inode = page->mapping->host;
1375 loff_t position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1379 cFYI(1, ("commit write for page %p up to position %lld for %d",
1380 page, position, to));
1381 spin_lock(&inode->i_lock);
1382 if (position > inode->i_size) {
1383 i_size_write(inode, position);
1385 spin_unlock(&inode->i_lock);
1386 if (!PageUptodate(page)) {
1387 position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + offset;
1388 /* can not rely on (or let) writepage write this data */
1390 cFYI(1, ("Illegal offsets, can not copy from %d to %d",
1395 /* this is probably better than directly calling
1396 partialpage_write since in this function the file handle is
1397 known which we might as well leverage */
1398 /* BB check if anything else missing out of ppw
1399 such as updating last write time */
1400 page_data = kmap(page);
1401 rc = cifs_write(file, page_data + offset, to-offset,
1405 /* else if (rc < 0) should we set writebehind rc? */
1408 set_page_dirty(page);
1415 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1419 struct inode *inode = file->f_path.dentry->d_inode;
1423 cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1424 dentry->d_name.name, datasync));
1426 rc = filemap_fdatawrite(inode->i_mapping);
1428 CIFS_I(inode)->write_behind_rc = 0;
1433 /* static void cifs_sync_page(struct page *page)
1435 struct address_space *mapping;
1436 struct inode *inode;
1437 unsigned long index = page->index;
1438 unsigned int rpages = 0;
1441 cFYI(1, ("sync page %p",page));
1442 mapping = page->mapping;
1445 inode = mapping->host;
1449 /* fill in rpages then
1450 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1452 /* cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1462 * As file closes, flush all cached write data for this inode checking
1463 * for write behind errors.
1465 int cifs_flush(struct file *file, fl_owner_t id)
1467 struct inode *inode = file->f_path.dentry->d_inode;
1470 /* Rather than do the steps manually:
1471 lock the inode for writing
1472 loop through pages looking for write behind data (dirty pages)
1473 coalesce into contiguous 16K (or smaller) chunks to write to server
1474 send to server (prefer in parallel)
1475 deal with writebehind errors
1476 unlock inode for writing
1477 filemapfdatawrite appears easier for the time being */
1479 rc = filemap_fdatawrite(inode->i_mapping);
1480 if (!rc) /* reset wb rc if we were able to write out dirty pages */
1481 CIFS_I(inode)->write_behind_rc = 0;
1483 cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
1488 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1489 size_t read_size, loff_t *poffset)
1492 unsigned int bytes_read = 0;
1493 unsigned int total_read = 0;
1494 unsigned int current_read_size;
1495 struct cifs_sb_info *cifs_sb;
1496 struct cifsTconInfo *pTcon;
1498 struct cifsFileInfo *open_file;
1499 char *smb_read_data;
1500 char __user *current_offset;
1501 struct smb_com_read_rsp *pSMBr;
1504 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1505 pTcon = cifs_sb->tcon;
1507 if (file->private_data == NULL) {
1511 open_file = (struct cifsFileInfo *)file->private_data;
1513 if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
1514 cFYI(1, ("attempting read on write only file instance"));
1516 for (total_read = 0, current_offset = read_data;
1517 read_size > total_read;
1518 total_read += bytes_read, current_offset += bytes_read) {
1519 current_read_size = min_t(const int, read_size - total_read,
1522 smb_read_data = NULL;
1523 while (rc == -EAGAIN) {
1524 int buf_type = CIFS_NO_BUFFER;
1525 if ((open_file->invalidHandle) &&
1526 (!open_file->closePend)) {
1527 rc = cifs_reopen_file(file, TRUE);
1531 rc = CIFSSMBRead(xid, pTcon,
1533 current_read_size, *poffset,
1534 &bytes_read, &smb_read_data,
1536 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1537 if (smb_read_data) {
1538 if (copy_to_user(current_offset,
1540 4 /* RFC1001 length field */ +
1541 le16_to_cpu(pSMBr->DataOffset),
1546 if (buf_type == CIFS_SMALL_BUFFER)
1547 cifs_small_buf_release(smb_read_data);
1548 else if (buf_type == CIFS_LARGE_BUFFER)
1549 cifs_buf_release(smb_read_data);
1550 smb_read_data = NULL;
1553 if (rc || (bytes_read == 0)) {
1561 cifs_stats_bytes_read(pTcon, bytes_read);
1562 *poffset += bytes_read;
1570 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1574 unsigned int bytes_read = 0;
1575 unsigned int total_read;
1576 unsigned int current_read_size;
1577 struct cifs_sb_info *cifs_sb;
1578 struct cifsTconInfo *pTcon;
1580 char *current_offset;
1581 struct cifsFileInfo *open_file;
1582 int buf_type = CIFS_NO_BUFFER;
1585 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1586 pTcon = cifs_sb->tcon;
1588 if (file->private_data == NULL) {
1592 open_file = (struct cifsFileInfo *)file->private_data;
1594 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1595 cFYI(1, ("attempting read on write only file instance"));
1597 for (total_read = 0, current_offset = read_data;
1598 read_size > total_read;
1599 total_read += bytes_read, current_offset += bytes_read) {
1600 current_read_size = min_t(const int, read_size - total_read,
1602 /* For windows me and 9x we do not want to request more
1603 than it negotiated since it will refuse the read then */
1605 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1606 current_read_size = min_t(const int, current_read_size,
1607 pTcon->ses->server->maxBuf - 128);
1610 while (rc == -EAGAIN) {
1611 if ((open_file->invalidHandle) &&
1612 (!open_file->closePend)) {
1613 rc = cifs_reopen_file(file, TRUE);
1617 rc = CIFSSMBRead(xid, pTcon,
1619 current_read_size, *poffset,
1620 &bytes_read, ¤t_offset,
1623 if (rc || (bytes_read == 0)) {
1631 cifs_stats_bytes_read(pTcon, total_read);
1632 *poffset += bytes_read;
1639 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1641 struct dentry *dentry = file->f_path.dentry;
1645 rc = cifs_revalidate(dentry);
1647 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1651 rc = generic_file_mmap(file, vma);
1657 static void cifs_copy_cache_pages(struct address_space *mapping,
1658 struct list_head *pages, int bytes_read, char *data,
1659 struct pagevec *plru_pvec)
1664 while (bytes_read > 0) {
1665 if (list_empty(pages))
1668 page = list_entry(pages->prev, struct page, lru);
1669 list_del(&page->lru);
1671 if (add_to_page_cache(page, mapping, page->index,
1673 page_cache_release(page);
1674 cFYI(1, ("Add page cache failed"));
1675 data += PAGE_CACHE_SIZE;
1676 bytes_read -= PAGE_CACHE_SIZE;
1680 target = kmap_atomic(page, KM_USER0);
1682 if (PAGE_CACHE_SIZE > bytes_read) {
1683 memcpy(target, data, bytes_read);
1684 /* zero the tail end of this partial page */
1685 memset(target + bytes_read, 0,
1686 PAGE_CACHE_SIZE - bytes_read);
1689 memcpy(target, data, PAGE_CACHE_SIZE);
1690 bytes_read -= PAGE_CACHE_SIZE;
1692 kunmap_atomic(target, KM_USER0);
1694 flush_dcache_page(page);
1695 SetPageUptodate(page);
1697 if (!pagevec_add(plru_pvec, page))
1698 __pagevec_lru_add(plru_pvec);
1699 data += PAGE_CACHE_SIZE;
1704 static int cifs_readpages(struct file *file, struct address_space *mapping,
1705 struct list_head *page_list, unsigned num_pages)
1711 struct cifs_sb_info *cifs_sb;
1712 struct cifsTconInfo *pTcon;
1714 unsigned int read_size, i;
1715 char *smb_read_data = NULL;
1716 struct smb_com_read_rsp *pSMBr;
1717 struct pagevec lru_pvec;
1718 struct cifsFileInfo *open_file;
1719 int buf_type = CIFS_NO_BUFFER;
1722 if (file->private_data == NULL) {
1726 open_file = (struct cifsFileInfo *)file->private_data;
1727 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1728 pTcon = cifs_sb->tcon;
1730 pagevec_init(&lru_pvec, 0);
1731 #ifdef CONFIG_CIFS_DEBUG2
1732 cFYI(1, ("rpages: num pages %d", num_pages));
1734 for (i = 0; i < num_pages; ) {
1735 unsigned contig_pages;
1736 struct page *tmp_page;
1737 unsigned long expected_index;
1739 if (list_empty(page_list))
1742 page = list_entry(page_list->prev, struct page, lru);
1743 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1745 /* count adjacent pages that we will read into */
1748 list_entry(page_list->prev, struct page, lru)->index;
1749 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1750 if (tmp_page->index == expected_index) {
1756 if (contig_pages + i > num_pages)
1757 contig_pages = num_pages - i;
1759 /* for reads over a certain size could initiate async
1762 read_size = contig_pages * PAGE_CACHE_SIZE;
1763 /* Read size needs to be in multiples of one page */
1764 read_size = min_t(const unsigned int, read_size,
1765 cifs_sb->rsize & PAGE_CACHE_MASK);
1766 #ifdef CONFIG_CIFS_DEBUG2
1767 cFYI(1, ("rpages: read size 0x%x contiguous pages %d",
1768 read_size, contig_pages));
1771 while (rc == -EAGAIN) {
1772 if ((open_file->invalidHandle) &&
1773 (!open_file->closePend)) {
1774 rc = cifs_reopen_file(file, TRUE);
1779 rc = CIFSSMBRead(xid, pTcon,
1782 &bytes_read, &smb_read_data,
1784 /* BB more RC checks ? */
1785 if (rc == -EAGAIN) {
1786 if (smb_read_data) {
1787 if (buf_type == CIFS_SMALL_BUFFER)
1788 cifs_small_buf_release(smb_read_data);
1789 else if (buf_type == CIFS_LARGE_BUFFER)
1790 cifs_buf_release(smb_read_data);
1791 smb_read_data = NULL;
1795 if ((rc < 0) || (smb_read_data == NULL)) {
1796 cFYI(1, ("Read error in readpages: %d", rc));
1798 } else if (bytes_read > 0) {
1799 task_io_account_read(bytes_read);
1800 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1801 cifs_copy_cache_pages(mapping, page_list, bytes_read,
1802 smb_read_data + 4 /* RFC1001 hdr */ +
1803 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
1805 i += bytes_read >> PAGE_CACHE_SHIFT;
1806 cifs_stats_bytes_read(pTcon, bytes_read);
1807 if ((int)(bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1808 i++; /* account for partial page */
1810 /* server copy of file can have smaller size
1812 /* BB do we need to verify this common case ?
1813 this case is ok - if we are at server EOF
1814 we will hit it on next read */
1819 cFYI(1, ("No bytes read (%d) at offset %lld . "
1820 "Cleaning remaining pages from readahead list",
1821 bytes_read, offset));
1822 /* BB turn off caching and do new lookup on
1823 file size at server? */
1826 if (smb_read_data) {
1827 if (buf_type == CIFS_SMALL_BUFFER)
1828 cifs_small_buf_release(smb_read_data);
1829 else if (buf_type == CIFS_LARGE_BUFFER)
1830 cifs_buf_release(smb_read_data);
1831 smb_read_data = NULL;
1836 pagevec_lru_add(&lru_pvec);
1838 /* need to free smb_read_data buf before exit */
1839 if (smb_read_data) {
1840 if (buf_type == CIFS_SMALL_BUFFER)
1841 cifs_small_buf_release(smb_read_data);
1842 else if (buf_type == CIFS_LARGE_BUFFER)
1843 cifs_buf_release(smb_read_data);
1844 smb_read_data = NULL;
1851 static int cifs_readpage_worker(struct file *file, struct page *page,
1857 page_cache_get(page);
1858 read_data = kmap(page);
1859 /* for reads over a certain size could initiate async read ahead */
1861 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1866 cFYI(1, ("Bytes read %d", rc));
1868 file->f_path.dentry->d_inode->i_atime =
1869 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1871 if (PAGE_CACHE_SIZE > rc)
1872 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1874 flush_dcache_page(page);
1875 SetPageUptodate(page);
1880 page_cache_release(page);
1884 static int cifs_readpage(struct file *file, struct page *page)
1886 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1892 if (file->private_data == NULL) {
1897 cFYI(1, ("readpage %p at offset %d 0x%x\n",
1898 page, (int)offset, (int)offset));
1900 rc = cifs_readpage_worker(file, page, &offset);
1908 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
1910 struct cifsFileInfo *open_file;
1912 read_lock(&GlobalSMBSeslock);
1913 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1914 if (open_file->closePend)
1916 if (open_file->pfile &&
1917 ((open_file->pfile->f_flags & O_RDWR) ||
1918 (open_file->pfile->f_flags & O_WRONLY))) {
1919 read_unlock(&GlobalSMBSeslock);
1923 read_unlock(&GlobalSMBSeslock);
1927 /* We do not want to update the file size from server for inodes
1928 open for write - to avoid races with writepage extending
1929 the file - in the future we could consider allowing
1930 refreshing the inode only on increases in the file size
1931 but this is tricky to do without racing with writebehind
1932 page caching in the current Linux kernel design */
1933 int is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
1938 if (is_inode_writable(cifsInode)) {
1939 /* This inode is open for write at least once */
1940 struct cifs_sb_info *cifs_sb;
1942 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
1943 if ( cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO ) {
1944 /* since no page cache to corrupt on directio
1945 we can change size safely */
1949 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
1957 static int cifs_prepare_write(struct file *file, struct page *page,
1958 unsigned from, unsigned to)
1964 cFYI(1, ("prepare write for page %p from %d to %d", page, from, to));
1965 if (PageUptodate(page))
1968 /* If we are writing a full page it will be up to date,
1969 no need to read from the server */
1970 if ((to == PAGE_CACHE_SIZE) && (from == 0)) {
1971 SetPageUptodate(page);
1975 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1976 i_size = i_size_read(page->mapping->host);
1978 if ((offset >= i_size) ||
1979 ((from == 0) && (offset + to) >= i_size)) {
1981 * We don't need to read data beyond the end of the file.
1982 * zero it, and set the page uptodate
1984 simple_prepare_write(file, page, from, to);
1985 SetPageUptodate(page);
1986 } else if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
1987 /* might as well read a page, it is fast enough */
1988 rc = cifs_readpage_worker(file, page, &offset);
1990 /* we could try using another file handle if there is one -
1991 but how would we lock it to prevent close of that handle
1992 racing with this read? In any case
1993 this will be written out by commit_write so is fine */
1996 /* we do not need to pass errors back
1997 e.g. if we do not have read access to the file
1998 because cifs_commit_write will do the right thing. -- shaggy */
2003 const struct address_space_operations cifs_addr_ops = {
2004 .readpage = cifs_readpage,
2005 .readpages = cifs_readpages,
2006 .writepage = cifs_writepage,
2007 .writepages = cifs_writepages,
2008 .prepare_write = cifs_prepare_write,
2009 .commit_write = cifs_commit_write,
2010 .set_page_dirty = __set_page_dirty_nobuffers,
2011 /* .sync_page = cifs_sync_page, */
2016 * cifs_readpages requires the server to support a buffer large enough to
2017 * contain the header plus one complete page of data. Otherwise, we need
2018 * to leave cifs_readpages out of the address space operations.
2020 const struct address_space_operations cifs_addr_ops_smallbuf = {
2021 .readpage = cifs_readpage,
2022 .writepage = cifs_writepage,
2023 .writepages = cifs_writepages,
2024 .prepare_write = cifs_prepare_write,
2025 .commit_write = cifs_commit_write,
2026 .set_page_dirty = __set_page_dirty_nobuffers,
2027 /* .sync_page = cifs_sync_page, */