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 <= 2048)) {
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);
526 /* We waited above to give the SMBWrite a chance to issue
527 on the wire (so we do not get SMBWrite returning EBADF
528 if writepages is racing with close. Note that writepages
529 does not specify a file handle, so it is possible for a file
530 to be opened twice, and the application close the "wrong"
531 file handle - in these cases we delay long enough to allow
532 the SMBWrite to get on the wire before the SMB Close.
533 We allow total wait here over 45 seconds, more than
534 oplock break time, and more than enough to allow any write
535 to complete on the server, or to time out on the client */
536 while ((atomic_read(&pSMBFile->wrtPending) != 0)
537 && (timeout <= 50000)) {
538 cERROR(1, ("writes pending, delay free of handle"));
542 kfree(pSMBFile->search_resume_name);
543 kfree(file->private_data);
544 file->private_data = NULL;
548 if (list_empty(&(CIFS_I(inode)->openFileList))) {
549 cFYI(1, ("closing last open instance for inode %p", inode));
550 /* if the file is not open we do not know if we can cache info
551 on this inode, much less write behind and read ahead */
552 CIFS_I(inode)->clientCanCacheRead = FALSE;
553 CIFS_I(inode)->clientCanCacheAll = FALSE;
555 if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
556 rc = CIFS_I(inode)->write_behind_rc;
561 int cifs_closedir(struct inode *inode, struct file *file)
565 struct cifsFileInfo *pCFileStruct =
566 (struct cifsFileInfo *)file->private_data;
569 cFYI(1, ("Closedir inode = 0x%p", inode));
574 struct cifsTconInfo *pTcon;
575 struct cifs_sb_info *cifs_sb =
576 CIFS_SB(file->f_path.dentry->d_sb);
578 pTcon = cifs_sb->tcon;
580 cFYI(1, ("Freeing private data in close dir"));
581 if ((pCFileStruct->srch_inf.endOfSearch == FALSE) &&
582 (pCFileStruct->invalidHandle == FALSE)) {
583 pCFileStruct->invalidHandle = TRUE;
584 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
585 cFYI(1, ("Closing uncompleted readdir with rc %d",
587 /* not much we can do if it fails anyway, ignore rc */
590 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
592 cFYI(1, ("closedir free smb buf in srch struct"));
593 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
594 if (pCFileStruct->srch_inf.smallBuf)
595 cifs_small_buf_release(ptmp);
597 cifs_buf_release(ptmp);
599 ptmp = pCFileStruct->search_resume_name;
601 cFYI(1, ("closedir free resume name"));
602 pCFileStruct->search_resume_name = NULL;
605 kfree(file->private_data);
606 file->private_data = NULL;
608 /* BB can we lock the filestruct while this is going on? */
613 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
614 __u64 offset, __u8 lockType)
616 struct cifsLockInfo *li =
617 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
623 mutex_lock(&fid->lock_mutex);
624 list_add(&li->llist, &fid->llist);
625 mutex_unlock(&fid->lock_mutex);
629 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
635 int wait_flag = FALSE;
636 struct cifs_sb_info *cifs_sb;
637 struct cifsTconInfo *pTcon;
639 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
642 length = 1 + pfLock->fl_end - pfLock->fl_start;
646 cFYI(1, ("Lock parm: 0x%x flockflags: "
647 "0x%x flocktype: 0x%x start: %lld end: %lld",
648 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
651 if (pfLock->fl_flags & FL_POSIX)
653 if (pfLock->fl_flags & FL_FLOCK)
655 if (pfLock->fl_flags & FL_SLEEP) {
656 cFYI(1, ("Blocking lock"));
659 if (pfLock->fl_flags & FL_ACCESS)
660 cFYI(1, ("Process suspended by mandatory locking - "
661 "not implemented yet"));
662 if (pfLock->fl_flags & FL_LEASE)
663 cFYI(1, ("Lease on file - not implemented yet"));
664 if (pfLock->fl_flags &
665 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
666 cFYI(1, ("Unknown lock flags 0x%x", pfLock->fl_flags));
668 if (pfLock->fl_type == F_WRLCK) {
669 cFYI(1, ("F_WRLCK "));
671 } else if (pfLock->fl_type == F_UNLCK) {
672 cFYI(1, ("F_UNLCK"));
674 /* Check if unlock includes more than
676 } else if (pfLock->fl_type == F_RDLCK) {
677 cFYI(1, ("F_RDLCK"));
678 lockType |= LOCKING_ANDX_SHARED_LOCK;
680 } else if (pfLock->fl_type == F_EXLCK) {
681 cFYI(1, ("F_EXLCK"));
683 } else if (pfLock->fl_type == F_SHLCK) {
684 cFYI(1, ("F_SHLCK"));
685 lockType |= LOCKING_ANDX_SHARED_LOCK;
688 cFYI(1, ("Unknown type of lock"));
690 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
691 pTcon = cifs_sb->tcon;
693 if (file->private_data == NULL) {
697 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
699 posix_locking = (cifs_sb->tcon->ses->capabilities & CAP_UNIX) &&
700 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(cifs_sb->tcon->fsUnixInfo.Capability));
702 /* BB add code here to normalize offset and length to
703 account for negative length which we can not accept over the
708 if (lockType & LOCKING_ANDX_SHARED_LOCK)
709 posix_lock_type = CIFS_RDLCK;
711 posix_lock_type = CIFS_WRLCK;
712 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 1 /* get */,
714 posix_lock_type, wait_flag);
719 /* BB we could chain these into one lock request BB */
720 rc = CIFSSMBLock(xid, pTcon, netfid, length, pfLock->fl_start,
721 0, 1, lockType, 0 /* wait flag */ );
723 rc = CIFSSMBLock(xid, pTcon, netfid, length,
724 pfLock->fl_start, 1 /* numUnlock */ ,
725 0 /* numLock */ , lockType,
727 pfLock->fl_type = F_UNLCK;
729 cERROR(1, ("Error unlocking previously locked "
730 "range %d during test of lock", rc));
734 /* if rc == ERR_SHARING_VIOLATION ? */
735 rc = 0; /* do not change lock type to unlock
736 since range in use */
743 if (!numLock && !numUnlock) {
744 /* if no lock or unlock then nothing
745 to do since we do not know what it is */
752 if (lockType & LOCKING_ANDX_SHARED_LOCK)
753 posix_lock_type = CIFS_RDLCK;
755 posix_lock_type = CIFS_WRLCK;
758 posix_lock_type = CIFS_UNLCK;
760 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 0 /* set */,
762 posix_lock_type, wait_flag);
764 struct cifsFileInfo *fid =
765 (struct cifsFileInfo *)file->private_data;
768 rc = CIFSSMBLock(xid, pTcon, netfid, length,
770 0, numLock, lockType, wait_flag);
773 /* For Windows locks we must store them. */
774 rc = store_file_lock(fid, length,
775 pfLock->fl_start, lockType);
777 } else if (numUnlock) {
778 /* For each stored lock that this unlock overlaps
779 completely, unlock it. */
781 struct cifsLockInfo *li, *tmp;
784 mutex_lock(&fid->lock_mutex);
785 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
786 if (pfLock->fl_start <= li->offset &&
787 (pfLock->fl_start + length) >=
788 (li->offset + li->length)) {
789 stored_rc = CIFSSMBLock(xid, pTcon,
791 li->length, li->offset,
792 1, 0, li->type, FALSE);
796 list_del(&li->llist);
800 mutex_unlock(&fid->lock_mutex);
804 if (pfLock->fl_flags & FL_POSIX)
805 posix_lock_file_wait(file, pfLock);
810 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
811 size_t write_size, loff_t *poffset)
814 unsigned int bytes_written = 0;
815 unsigned int total_written;
816 struct cifs_sb_info *cifs_sb;
817 struct cifsTconInfo *pTcon;
819 struct cifsFileInfo *open_file;
821 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
823 pTcon = cifs_sb->tcon;
826 (" write %d bytes to offset %lld of %s", write_size,
827 *poffset, file->f_path.dentry->d_name.name)); */
829 if (file->private_data == NULL)
831 open_file = (struct cifsFileInfo *) file->private_data;
835 if (*poffset > file->f_path.dentry->d_inode->i_size)
836 long_op = 2; /* writes past end of file can take a long time */
840 for (total_written = 0; write_size > total_written;
841 total_written += bytes_written) {
843 while (rc == -EAGAIN) {
844 if (file->private_data == NULL) {
845 /* file has been closed on us */
847 /* if we have gotten here we have written some data
848 and blocked, and the file has been freed on us while
849 we blocked so return what we managed to write */
850 return total_written;
852 if (open_file->closePend) {
855 return total_written;
859 if (open_file->invalidHandle) {
860 /* we could deadlock if we called
861 filemap_fdatawait from here so tell
862 reopen_file not to flush data to server
864 rc = cifs_reopen_file(file, FALSE);
869 rc = CIFSSMBWrite(xid, pTcon,
871 min_t(const int, cifs_sb->wsize,
872 write_size - total_written),
873 *poffset, &bytes_written,
874 NULL, write_data + total_written, long_op);
876 if (rc || (bytes_written == 0)) {
884 *poffset += bytes_written;
885 long_op = FALSE; /* subsequent writes fast -
886 15 seconds is plenty */
889 cifs_stats_bytes_written(pTcon, total_written);
891 /* since the write may have blocked check these pointers again */
892 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
893 struct inode *inode = file->f_path.dentry->d_inode;
894 /* Do not update local mtime - server will set its actual value on write
895 * inode->i_ctime = inode->i_mtime =
896 * current_fs_time(inode->i_sb);*/
897 if (total_written > 0) {
898 spin_lock(&inode->i_lock);
899 if (*poffset > file->f_path.dentry->d_inode->i_size)
900 i_size_write(file->f_path.dentry->d_inode,
902 spin_unlock(&inode->i_lock);
904 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
907 return total_written;
910 static ssize_t cifs_write(struct file *file, const char *write_data,
911 size_t write_size, loff_t *poffset)
914 unsigned int bytes_written = 0;
915 unsigned int total_written;
916 struct cifs_sb_info *cifs_sb;
917 struct cifsTconInfo *pTcon;
919 struct cifsFileInfo *open_file;
921 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
923 pTcon = cifs_sb->tcon;
925 cFYI(1, ("write %zd bytes to offset %lld of %s", write_size,
926 *poffset, file->f_path.dentry->d_name.name));
928 if (file->private_data == NULL)
930 open_file = (struct cifsFileInfo *)file->private_data;
934 if (*poffset > file->f_path.dentry->d_inode->i_size)
935 long_op = 2; /* writes past end of file can take a long time */
939 for (total_written = 0; write_size > total_written;
940 total_written += bytes_written) {
942 while (rc == -EAGAIN) {
943 if (file->private_data == NULL) {
944 /* file has been closed on us */
946 /* if we have gotten here we have written some data
947 and blocked, and the file has been freed on us
948 while we blocked so return what we managed to
950 return total_written;
952 if (open_file->closePend) {
955 return total_written;
959 if (open_file->invalidHandle) {
960 /* we could deadlock if we called
961 filemap_fdatawait from here so tell
962 reopen_file not to flush data to
964 rc = cifs_reopen_file(file, FALSE);
968 if (experimEnabled || (pTcon->ses->server &&
969 ((pTcon->ses->server->secMode &
970 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
975 len = min((size_t)cifs_sb->wsize,
976 write_size - total_written);
977 /* iov[0] is reserved for smb header */
978 iov[1].iov_base = (char *)write_data +
980 iov[1].iov_len = len;
981 rc = CIFSSMBWrite2(xid, pTcon,
982 open_file->netfid, len,
983 *poffset, &bytes_written,
986 rc = CIFSSMBWrite(xid, pTcon,
988 min_t(const int, cifs_sb->wsize,
989 write_size - total_written),
990 *poffset, &bytes_written,
991 write_data + total_written,
994 if (rc || (bytes_written == 0)) {
1002 *poffset += bytes_written;
1003 long_op = FALSE; /* subsequent writes fast -
1004 15 seconds is plenty */
1007 cifs_stats_bytes_written(pTcon, total_written);
1009 /* since the write may have blocked check these pointers again */
1010 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1011 /*BB We could make this contingent on superblock ATIME flag too */
1012 /* file->f_path.dentry->d_inode->i_ctime =
1013 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1014 if (total_written > 0) {
1015 spin_lock(&file->f_path.dentry->d_inode->i_lock);
1016 if (*poffset > file->f_path.dentry->d_inode->i_size)
1017 i_size_write(file->f_path.dentry->d_inode,
1019 spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1021 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1024 return total_written;
1027 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1029 struct cifsFileInfo *open_file;
1032 /* Having a null inode here (because mapping->host was set to zero by
1033 the VFS or MM) should not happen but we had reports of on oops (due to
1034 it being zero) during stress testcases so we need to check for it */
1036 if (cifs_inode == NULL) {
1037 cERROR(1, ("Null inode passed to cifs_writeable_file"));
1042 read_lock(&GlobalSMBSeslock);
1043 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1044 if (open_file->closePend)
1046 if (open_file->pfile &&
1047 ((open_file->pfile->f_flags & O_RDWR) ||
1048 (open_file->pfile->f_flags & O_WRONLY))) {
1049 atomic_inc(&open_file->wrtPending);
1050 read_unlock(&GlobalSMBSeslock);
1051 if (open_file->invalidHandle) {
1052 rc = cifs_reopen_file(open_file->pfile, FALSE);
1053 /* if it fails, try another handle - might be */
1054 /* dangerous to hold up writepages with retry */
1056 cFYI(1, ("wp failed on reopen file"));
1057 read_lock(&GlobalSMBSeslock);
1058 /* can not use this handle, no write
1059 pending on this one after all */
1060 atomic_dec(&open_file->wrtPending);
1064 if (open_file->closePend) {
1065 read_lock(&GlobalSMBSeslock);
1066 atomic_dec(&open_file->wrtPending);
1072 read_unlock(&GlobalSMBSeslock);
1076 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1078 struct address_space *mapping = page->mapping;
1079 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1082 int bytes_written = 0;
1083 struct cifs_sb_info *cifs_sb;
1084 struct cifsTconInfo *pTcon;
1085 struct inode *inode;
1086 struct cifsFileInfo *open_file;
1088 if (!mapping || !mapping->host)
1091 inode = page->mapping->host;
1092 cifs_sb = CIFS_SB(inode->i_sb);
1093 pTcon = cifs_sb->tcon;
1095 offset += (loff_t)from;
1096 write_data = kmap(page);
1099 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1104 /* racing with truncate? */
1105 if (offset > mapping->host->i_size) {
1107 return 0; /* don't care */
1110 /* check to make sure that we are not extending the file */
1111 if (mapping->host->i_size - offset < (loff_t)to)
1112 to = (unsigned)(mapping->host->i_size - offset);
1114 open_file = find_writable_file(CIFS_I(mapping->host));
1116 bytes_written = cifs_write(open_file->pfile, write_data,
1118 atomic_dec(&open_file->wrtPending);
1119 /* Does mm or vfs already set times? */
1120 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1121 if ((bytes_written > 0) && (offset)) {
1123 } else if (bytes_written < 0) {
1128 cFYI(1, ("No writeable filehandles for inode"));
1136 static int cifs_writepages(struct address_space *mapping,
1137 struct writeback_control *wbc)
1139 struct backing_dev_info *bdi = mapping->backing_dev_info;
1140 unsigned int bytes_to_write;
1141 unsigned int bytes_written;
1142 struct cifs_sb_info *cifs_sb;
1146 int range_whole = 0;
1153 struct cifsFileInfo *open_file;
1155 struct pagevec pvec;
1160 cifs_sb = CIFS_SB(mapping->host->i_sb);
1163 * If wsize is smaller that the page cache size, default to writing
1164 * one page at a time via cifs_writepage
1166 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1167 return generic_writepages(mapping, wbc);
1169 if ((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1170 if (cifs_sb->tcon->ses->server->secMode &
1171 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1172 if (!experimEnabled)
1173 return generic_writepages(mapping, wbc);
1175 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1177 return generic_writepages(mapping, wbc);
1181 * BB: Is this meaningful for a non-block-device file system?
1182 * If it is, we should test it again after we do I/O
1184 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1185 wbc->encountered_congestion = 1;
1192 pagevec_init(&pvec, 0);
1193 if (wbc->range_cyclic) {
1194 index = mapping->writeback_index; /* Start from prev offset */
1197 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1198 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1199 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1204 while (!done && (index <= end) &&
1205 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1206 PAGECACHE_TAG_DIRTY,
1207 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1216 for (i = 0; i < nr_pages; i++) {
1217 page = pvec.pages[i];
1219 * At this point we hold neither mapping->tree_lock nor
1220 * lock on the page itself: the page may be truncated or
1221 * invalidated (changing page->mapping to NULL), or even
1222 * swizzled back from swapper_space to tmpfs file
1228 else if (TestSetPageLocked(page))
1231 if (unlikely(page->mapping != mapping)) {
1236 if (!wbc->range_cyclic && page->index > end) {
1242 if (next && (page->index != next)) {
1243 /* Not next consecutive page */
1248 if (wbc->sync_mode != WB_SYNC_NONE)
1249 wait_on_page_writeback(page);
1251 if (PageWriteback(page) ||
1252 !clear_page_dirty_for_io(page)) {
1258 * This actually clears the dirty bit in the radix tree.
1259 * See cifs_writepage() for more commentary.
1261 set_page_writeback(page);
1263 if (page_offset(page) >= mapping->host->i_size) {
1266 end_page_writeback(page);
1271 * BB can we get rid of this? pages are held by pvec
1273 page_cache_get(page);
1275 len = min(mapping->host->i_size - page_offset(page),
1276 (loff_t)PAGE_CACHE_SIZE);
1278 /* reserve iov[0] for the smb header */
1280 iov[n_iov].iov_base = kmap(page);
1281 iov[n_iov].iov_len = len;
1282 bytes_to_write += len;
1286 offset = page_offset(page);
1288 next = page->index + 1;
1289 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1293 /* Search for a writable handle every time we call
1294 * CIFSSMBWrite2. We can't rely on the last handle
1295 * we used to still be valid
1297 open_file = find_writable_file(CIFS_I(mapping->host));
1299 cERROR(1, ("No writable handles for inode"));
1302 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1304 bytes_to_write, offset,
1305 &bytes_written, iov, n_iov,
1307 atomic_dec(&open_file->wrtPending);
1308 if (rc || bytes_written < bytes_to_write) {
1309 cERROR(1, ("Write2 ret %d, wrote %d",
1310 rc, bytes_written));
1311 /* BB what if continued retry is
1312 requested via mount flags? */
1313 set_bit(AS_EIO, &mapping->flags);
1315 cifs_stats_bytes_written(cifs_sb->tcon,
1319 for (i = 0; i < n_iov; i++) {
1320 page = pvec.pages[first + i];
1321 /* Should we also set page error on
1322 success rc but too little data written? */
1323 /* BB investigate retry logic on temporary
1324 server crash cases and how recovery works
1325 when page marked as error */
1330 end_page_writeback(page);
1331 page_cache_release(page);
1333 if ((wbc->nr_to_write -= n_iov) <= 0)
1337 pagevec_release(&pvec);
1339 if (!scanned && !done) {
1341 * We hit the last page and there is more work to be done: wrap
1342 * back to the start of the file
1348 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1349 mapping->writeback_index = index;
1356 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1362 /* BB add check for wbc flags */
1363 page_cache_get(page);
1364 if (!PageUptodate(page)) {
1365 cFYI(1, ("ppw - page not up to date"));
1369 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1371 * A writepage() implementation always needs to do either this,
1372 * or re-dirty the page with "redirty_page_for_writepage()" in
1373 * the case of a failure.
1375 * Just unlocking the page will cause the radix tree tag-bits
1376 * to fail to update with the state of the page correctly.
1378 set_page_writeback(page);
1379 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1380 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1382 end_page_writeback(page);
1383 page_cache_release(page);
1388 static int cifs_commit_write(struct file *file, struct page *page,
1389 unsigned offset, unsigned to)
1393 struct inode *inode = page->mapping->host;
1394 loff_t position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1398 cFYI(1, ("commit write for page %p up to position %lld for %d",
1399 page, position, to));
1400 spin_lock(&inode->i_lock);
1401 if (position > inode->i_size) {
1402 i_size_write(inode, position);
1404 spin_unlock(&inode->i_lock);
1405 if (!PageUptodate(page)) {
1406 position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + offset;
1407 /* can not rely on (or let) writepage write this data */
1409 cFYI(1, ("Illegal offsets, can not copy from %d to %d",
1414 /* this is probably better than directly calling
1415 partialpage_write since in this function the file handle is
1416 known which we might as well leverage */
1417 /* BB check if anything else missing out of ppw
1418 such as updating last write time */
1419 page_data = kmap(page);
1420 rc = cifs_write(file, page_data + offset, to-offset,
1424 /* else if (rc < 0) should we set writebehind rc? */
1427 set_page_dirty(page);
1434 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1438 struct inode *inode = file->f_path.dentry->d_inode;
1442 cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1443 dentry->d_name.name, datasync));
1445 rc = filemap_fdatawrite(inode->i_mapping);
1447 CIFS_I(inode)->write_behind_rc = 0;
1452 /* static void cifs_sync_page(struct page *page)
1454 struct address_space *mapping;
1455 struct inode *inode;
1456 unsigned long index = page->index;
1457 unsigned int rpages = 0;
1460 cFYI(1, ("sync page %p",page));
1461 mapping = page->mapping;
1464 inode = mapping->host;
1468 /* fill in rpages then
1469 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1471 /* cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1481 * As file closes, flush all cached write data for this inode checking
1482 * for write behind errors.
1484 int cifs_flush(struct file *file, fl_owner_t id)
1486 struct inode *inode = file->f_path.dentry->d_inode;
1489 /* Rather than do the steps manually:
1490 lock the inode for writing
1491 loop through pages looking for write behind data (dirty pages)
1492 coalesce into contiguous 16K (or smaller) chunks to write to server
1493 send to server (prefer in parallel)
1494 deal with writebehind errors
1495 unlock inode for writing
1496 filemapfdatawrite appears easier for the time being */
1498 rc = filemap_fdatawrite(inode->i_mapping);
1499 if (!rc) /* reset wb rc if we were able to write out dirty pages */
1500 CIFS_I(inode)->write_behind_rc = 0;
1502 cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
1507 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1508 size_t read_size, loff_t *poffset)
1511 unsigned int bytes_read = 0;
1512 unsigned int total_read = 0;
1513 unsigned int current_read_size;
1514 struct cifs_sb_info *cifs_sb;
1515 struct cifsTconInfo *pTcon;
1517 struct cifsFileInfo *open_file;
1518 char *smb_read_data;
1519 char __user *current_offset;
1520 struct smb_com_read_rsp *pSMBr;
1523 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1524 pTcon = cifs_sb->tcon;
1526 if (file->private_data == NULL) {
1530 open_file = (struct cifsFileInfo *)file->private_data;
1532 if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
1533 cFYI(1, ("attempting read on write only file instance"));
1535 for (total_read = 0, current_offset = read_data;
1536 read_size > total_read;
1537 total_read += bytes_read, current_offset += bytes_read) {
1538 current_read_size = min_t(const int, read_size - total_read,
1541 smb_read_data = NULL;
1542 while (rc == -EAGAIN) {
1543 int buf_type = CIFS_NO_BUFFER;
1544 if ((open_file->invalidHandle) &&
1545 (!open_file->closePend)) {
1546 rc = cifs_reopen_file(file, TRUE);
1550 rc = CIFSSMBRead(xid, pTcon,
1552 current_read_size, *poffset,
1553 &bytes_read, &smb_read_data,
1555 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1556 if (smb_read_data) {
1557 if (copy_to_user(current_offset,
1559 4 /* RFC1001 length field */ +
1560 le16_to_cpu(pSMBr->DataOffset),
1565 if (buf_type == CIFS_SMALL_BUFFER)
1566 cifs_small_buf_release(smb_read_data);
1567 else if (buf_type == CIFS_LARGE_BUFFER)
1568 cifs_buf_release(smb_read_data);
1569 smb_read_data = NULL;
1572 if (rc || (bytes_read == 0)) {
1580 cifs_stats_bytes_read(pTcon, bytes_read);
1581 *poffset += bytes_read;
1589 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1593 unsigned int bytes_read = 0;
1594 unsigned int total_read;
1595 unsigned int current_read_size;
1596 struct cifs_sb_info *cifs_sb;
1597 struct cifsTconInfo *pTcon;
1599 char *current_offset;
1600 struct cifsFileInfo *open_file;
1601 int buf_type = CIFS_NO_BUFFER;
1604 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1605 pTcon = cifs_sb->tcon;
1607 if (file->private_data == NULL) {
1611 open_file = (struct cifsFileInfo *)file->private_data;
1613 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1614 cFYI(1, ("attempting read on write only file instance"));
1616 for (total_read = 0, current_offset = read_data;
1617 read_size > total_read;
1618 total_read += bytes_read, current_offset += bytes_read) {
1619 current_read_size = min_t(const int, read_size - total_read,
1621 /* For windows me and 9x we do not want to request more
1622 than it negotiated since it will refuse the read then */
1624 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1625 current_read_size = min_t(const int, current_read_size,
1626 pTcon->ses->server->maxBuf - 128);
1629 while (rc == -EAGAIN) {
1630 if ((open_file->invalidHandle) &&
1631 (!open_file->closePend)) {
1632 rc = cifs_reopen_file(file, TRUE);
1636 rc = CIFSSMBRead(xid, pTcon,
1638 current_read_size, *poffset,
1639 &bytes_read, ¤t_offset,
1642 if (rc || (bytes_read == 0)) {
1650 cifs_stats_bytes_read(pTcon, total_read);
1651 *poffset += bytes_read;
1658 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1660 struct dentry *dentry = file->f_path.dentry;
1664 rc = cifs_revalidate(dentry);
1666 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1670 rc = generic_file_mmap(file, vma);
1676 static void cifs_copy_cache_pages(struct address_space *mapping,
1677 struct list_head *pages, int bytes_read, char *data,
1678 struct pagevec *plru_pvec)
1683 while (bytes_read > 0) {
1684 if (list_empty(pages))
1687 page = list_entry(pages->prev, struct page, lru);
1688 list_del(&page->lru);
1690 if (add_to_page_cache(page, mapping, page->index,
1692 page_cache_release(page);
1693 cFYI(1, ("Add page cache failed"));
1694 data += PAGE_CACHE_SIZE;
1695 bytes_read -= PAGE_CACHE_SIZE;
1699 target = kmap_atomic(page, KM_USER0);
1701 if (PAGE_CACHE_SIZE > bytes_read) {
1702 memcpy(target, data, bytes_read);
1703 /* zero the tail end of this partial page */
1704 memset(target + bytes_read, 0,
1705 PAGE_CACHE_SIZE - bytes_read);
1708 memcpy(target, data, PAGE_CACHE_SIZE);
1709 bytes_read -= PAGE_CACHE_SIZE;
1711 kunmap_atomic(target, KM_USER0);
1713 flush_dcache_page(page);
1714 SetPageUptodate(page);
1716 if (!pagevec_add(plru_pvec, page))
1717 __pagevec_lru_add(plru_pvec);
1718 data += PAGE_CACHE_SIZE;
1723 static int cifs_readpages(struct file *file, struct address_space *mapping,
1724 struct list_head *page_list, unsigned num_pages)
1730 struct cifs_sb_info *cifs_sb;
1731 struct cifsTconInfo *pTcon;
1733 unsigned int read_size, i;
1734 char *smb_read_data = NULL;
1735 struct smb_com_read_rsp *pSMBr;
1736 struct pagevec lru_pvec;
1737 struct cifsFileInfo *open_file;
1738 int buf_type = CIFS_NO_BUFFER;
1741 if (file->private_data == NULL) {
1745 open_file = (struct cifsFileInfo *)file->private_data;
1746 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1747 pTcon = cifs_sb->tcon;
1749 pagevec_init(&lru_pvec, 0);
1750 #ifdef CONFIG_CIFS_DEBUG2
1751 cFYI(1, ("rpages: num pages %d", num_pages));
1753 for (i = 0; i < num_pages; ) {
1754 unsigned contig_pages;
1755 struct page *tmp_page;
1756 unsigned long expected_index;
1758 if (list_empty(page_list))
1761 page = list_entry(page_list->prev, struct page, lru);
1762 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1764 /* count adjacent pages that we will read into */
1767 list_entry(page_list->prev, struct page, lru)->index;
1768 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1769 if (tmp_page->index == expected_index) {
1775 if (contig_pages + i > num_pages)
1776 contig_pages = num_pages - i;
1778 /* for reads over a certain size could initiate async
1781 read_size = contig_pages * PAGE_CACHE_SIZE;
1782 /* Read size needs to be in multiples of one page */
1783 read_size = min_t(const unsigned int, read_size,
1784 cifs_sb->rsize & PAGE_CACHE_MASK);
1785 #ifdef CONFIG_CIFS_DEBUG2
1786 cFYI(1, ("rpages: read size 0x%x contiguous pages %d",
1787 read_size, contig_pages));
1790 while (rc == -EAGAIN) {
1791 if ((open_file->invalidHandle) &&
1792 (!open_file->closePend)) {
1793 rc = cifs_reopen_file(file, TRUE);
1798 rc = CIFSSMBRead(xid, pTcon,
1801 &bytes_read, &smb_read_data,
1803 /* BB more RC checks ? */
1804 if (rc == -EAGAIN) {
1805 if (smb_read_data) {
1806 if (buf_type == CIFS_SMALL_BUFFER)
1807 cifs_small_buf_release(smb_read_data);
1808 else if (buf_type == CIFS_LARGE_BUFFER)
1809 cifs_buf_release(smb_read_data);
1810 smb_read_data = NULL;
1814 if ((rc < 0) || (smb_read_data == NULL)) {
1815 cFYI(1, ("Read error in readpages: %d", rc));
1817 } else if (bytes_read > 0) {
1818 task_io_account_read(bytes_read);
1819 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1820 cifs_copy_cache_pages(mapping, page_list, bytes_read,
1821 smb_read_data + 4 /* RFC1001 hdr */ +
1822 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
1824 i += bytes_read >> PAGE_CACHE_SHIFT;
1825 cifs_stats_bytes_read(pTcon, bytes_read);
1826 if ((int)(bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1827 i++; /* account for partial page */
1829 /* server copy of file can have smaller size
1831 /* BB do we need to verify this common case ?
1832 this case is ok - if we are at server EOF
1833 we will hit it on next read */
1838 cFYI(1, ("No bytes read (%d) at offset %lld . "
1839 "Cleaning remaining pages from readahead list",
1840 bytes_read, offset));
1841 /* BB turn off caching and do new lookup on
1842 file size at server? */
1845 if (smb_read_data) {
1846 if (buf_type == CIFS_SMALL_BUFFER)
1847 cifs_small_buf_release(smb_read_data);
1848 else if (buf_type == CIFS_LARGE_BUFFER)
1849 cifs_buf_release(smb_read_data);
1850 smb_read_data = NULL;
1855 pagevec_lru_add(&lru_pvec);
1857 /* need to free smb_read_data buf before exit */
1858 if (smb_read_data) {
1859 if (buf_type == CIFS_SMALL_BUFFER)
1860 cifs_small_buf_release(smb_read_data);
1861 else if (buf_type == CIFS_LARGE_BUFFER)
1862 cifs_buf_release(smb_read_data);
1863 smb_read_data = NULL;
1870 static int cifs_readpage_worker(struct file *file, struct page *page,
1876 page_cache_get(page);
1877 read_data = kmap(page);
1878 /* for reads over a certain size could initiate async read ahead */
1880 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1885 cFYI(1, ("Bytes read %d", rc));
1887 file->f_path.dentry->d_inode->i_atime =
1888 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1890 if (PAGE_CACHE_SIZE > rc)
1891 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1893 flush_dcache_page(page);
1894 SetPageUptodate(page);
1899 page_cache_release(page);
1903 static int cifs_readpage(struct file *file, struct page *page)
1905 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1911 if (file->private_data == NULL) {
1916 cFYI(1, ("readpage %p at offset %d 0x%x\n",
1917 page, (int)offset, (int)offset));
1919 rc = cifs_readpage_worker(file, page, &offset);
1927 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
1929 struct cifsFileInfo *open_file;
1931 read_lock(&GlobalSMBSeslock);
1932 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1933 if (open_file->closePend)
1935 if (open_file->pfile &&
1936 ((open_file->pfile->f_flags & O_RDWR) ||
1937 (open_file->pfile->f_flags & O_WRONLY))) {
1938 read_unlock(&GlobalSMBSeslock);
1942 read_unlock(&GlobalSMBSeslock);
1946 /* We do not want to update the file size from server for inodes
1947 open for write - to avoid races with writepage extending
1948 the file - in the future we could consider allowing
1949 refreshing the inode only on increases in the file size
1950 but this is tricky to do without racing with writebehind
1951 page caching in the current Linux kernel design */
1952 int is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
1957 if (is_inode_writable(cifsInode)) {
1958 /* This inode is open for write at least once */
1959 struct cifs_sb_info *cifs_sb;
1961 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
1962 if ( cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO ) {
1963 /* since no page cache to corrupt on directio
1964 we can change size safely */
1968 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
1976 static int cifs_prepare_write(struct file *file, struct page *page,
1977 unsigned from, unsigned to)
1983 cFYI(1, ("prepare write for page %p from %d to %d", page, from, to));
1984 if (PageUptodate(page))
1987 /* If we are writing a full page it will be up to date,
1988 no need to read from the server */
1989 if ((to == PAGE_CACHE_SIZE) && (from == 0)) {
1990 SetPageUptodate(page);
1994 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1995 i_size = i_size_read(page->mapping->host);
1997 if ((offset >= i_size) ||
1998 ((from == 0) && (offset + to) >= i_size)) {
2000 * We don't need to read data beyond the end of the file.
2001 * zero it, and set the page uptodate
2003 simple_prepare_write(file, page, from, to);
2004 SetPageUptodate(page);
2005 } else if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2006 /* might as well read a page, it is fast enough */
2007 rc = cifs_readpage_worker(file, page, &offset);
2009 /* we could try using another file handle if there is one -
2010 but how would we lock it to prevent close of that handle
2011 racing with this read? In any case
2012 this will be written out by commit_write so is fine */
2015 /* we do not need to pass errors back
2016 e.g. if we do not have read access to the file
2017 because cifs_commit_write will do the right thing. -- shaggy */
2022 const struct address_space_operations cifs_addr_ops = {
2023 .readpage = cifs_readpage,
2024 .readpages = cifs_readpages,
2025 .writepage = cifs_writepage,
2026 .writepages = cifs_writepages,
2027 .prepare_write = cifs_prepare_write,
2028 .commit_write = cifs_commit_write,
2029 .set_page_dirty = __set_page_dirty_nobuffers,
2030 /* .sync_page = cifs_sync_page, */
2035 * cifs_readpages requires the server to support a buffer large enough to
2036 * contain the header plus one complete page of data. Otherwise, we need
2037 * to leave cifs_readpages out of the address space operations.
2039 const struct address_space_operations cifs_addr_ops_smallbuf = {
2040 .readpage = cifs_readpage,
2041 .writepage = cifs_writepage,
2042 .writepages = cifs_writepages,
2043 .prepare_write = cifs_prepare_write,
2044 .commit_write = cifs_commit_write,
2045 .set_page_dirty = __set_page_dirty_nobuffers,
2046 /* .sync_page = cifs_sync_page, */