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);
78 return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
79 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
85 static inline int cifs_get_disposition(unsigned int flags)
87 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
89 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
90 return FILE_OVERWRITE_IF;
91 else if ((flags & O_CREAT) == O_CREAT)
93 else if ((flags & O_TRUNC) == O_TRUNC)
94 return FILE_OVERWRITE;
99 /* all arguments to this function must be checked for validity in caller */
100 static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
101 struct cifsInodeInfo *pCifsInode, struct cifsFileInfo *pCifsFile,
102 struct cifsTconInfo *pTcon, int *oplock, FILE_ALL_INFO *buf,
103 char *full_path, int xid)
105 struct timespec temp;
108 /* want handles we can use to read with first
109 in the list so we do not have to walk the
110 list to search for one in prepare_write */
111 if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
112 list_add_tail(&pCifsFile->flist,
113 &pCifsInode->openFileList);
115 list_add(&pCifsFile->flist,
116 &pCifsInode->openFileList);
118 write_unlock(&GlobalSMBSeslock);
119 if (pCifsInode->clientCanCacheRead) {
120 /* we have the inode open somewhere else
121 no need to discard cache data */
122 goto client_can_cache;
125 /* BB need same check in cifs_create too? */
126 /* if not oplocked, invalidate inode pages if mtime or file
128 temp = cifs_NTtimeToUnix(le64_to_cpu(buf->LastWriteTime));
129 if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
130 (file->f_path.dentry->d_inode->i_size ==
131 (loff_t)le64_to_cpu(buf->EndOfFile))) {
132 cFYI(1, ("inode unchanged on server"));
134 if (file->f_path.dentry->d_inode->i_mapping) {
135 /* BB no need to lock inode until after invalidate
136 since namei code should already have it locked? */
137 rc = filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
139 CIFS_I(file->f_path.dentry->d_inode)->write_behind_rc = rc;
141 cFYI(1, ("invalidating remote inode since open detected it "
143 invalidate_remote_inode(file->f_path.dentry->d_inode);
148 rc = cifs_get_inode_info_unix(&file->f_path.dentry->d_inode,
149 full_path, inode->i_sb, xid);
151 rc = cifs_get_inode_info(&file->f_path.dentry->d_inode,
152 full_path, buf, inode->i_sb, xid, NULL);
154 if ((*oplock & 0xF) == OPLOCK_EXCLUSIVE) {
155 pCifsInode->clientCanCacheAll = true;
156 pCifsInode->clientCanCacheRead = true;
157 cFYI(1, ("Exclusive Oplock granted on inode %p",
158 file->f_path.dentry->d_inode));
159 } else if ((*oplock & 0xF) == OPLOCK_READ)
160 pCifsInode->clientCanCacheRead = true;
165 int cifs_open(struct inode *inode, struct file *file)
169 struct cifs_sb_info *cifs_sb;
170 struct cifsTconInfo *pTcon;
171 struct cifsFileInfo *pCifsFile;
172 struct cifsInodeInfo *pCifsInode;
173 struct list_head *tmp;
174 char *full_path = NULL;
178 FILE_ALL_INFO *buf = NULL;
182 cifs_sb = CIFS_SB(inode->i_sb);
183 pTcon = cifs_sb->tcon;
185 if (file->f_flags & O_CREAT) {
186 /* search inode for this file and fill in file->private_data */
187 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
188 read_lock(&GlobalSMBSeslock);
189 list_for_each(tmp, &pCifsInode->openFileList) {
190 pCifsFile = list_entry(tmp, struct cifsFileInfo,
192 if ((pCifsFile->pfile == NULL) &&
193 (pCifsFile->pid == current->tgid)) {
194 /* mode set in cifs_create */
196 /* needed for writepage */
197 pCifsFile->pfile = file;
199 file->private_data = pCifsFile;
203 read_unlock(&GlobalSMBSeslock);
204 if (file->private_data != NULL) {
209 if (file->f_flags & O_EXCL)
210 cERROR(1, ("could not find file instance for "
211 "new file %p", file));
215 full_path = build_path_from_dentry(file->f_path.dentry);
216 if (full_path == NULL) {
221 cFYI(1, ("inode = 0x%p file flags are 0x%x for %s",
222 inode, file->f_flags, full_path));
223 desiredAccess = cifs_convert_flags(file->f_flags);
225 /*********************************************************************
226 * open flag mapping table:
228 * POSIX Flag CIFS Disposition
229 * ---------- ----------------
230 * O_CREAT FILE_OPEN_IF
231 * O_CREAT | O_EXCL FILE_CREATE
232 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
233 * O_TRUNC FILE_OVERWRITE
234 * none of the above FILE_OPEN
236 * Note that there is not a direct match between disposition
237 * FILE_SUPERSEDE (ie create whether or not file exists although
238 * O_CREAT | O_TRUNC is similar but truncates the existing
239 * file rather than creating a new file as FILE_SUPERSEDE does
240 * (which uses the attributes / metadata passed in on open call)
242 *? O_SYNC is a reasonable match to CIFS writethrough flag
243 *? and the read write flags match reasonably. O_LARGEFILE
244 *? is irrelevant because largefile support is always used
245 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
246 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
247 *********************************************************************/
249 disposition = cifs_get_disposition(file->f_flags);
256 /* BB pass O_SYNC flag through on file attributes .. BB */
258 /* Also refresh inode by passing in file_info buf returned by SMBOpen
259 and calling get_inode_info with returned buf (at least helps
260 non-Unix server case) */
262 /* BB we can not do this if this is the second open of a file
263 and the first handle has writebehind data, we might be
264 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
265 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
271 if (cifs_sb->tcon->ses->capabilities & CAP_NT_SMBS)
272 rc = CIFSSMBOpen(xid, pTcon, full_path, disposition,
273 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
274 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
275 & CIFS_MOUNT_MAP_SPECIAL_CHR);
277 rc = -EIO; /* no NT SMB support fall into legacy open below */
280 /* Old server, try legacy style OpenX */
281 rc = SMBLegacyOpen(xid, pTcon, full_path, disposition,
282 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
283 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
284 & CIFS_MOUNT_MAP_SPECIAL_CHR);
287 cFYI(1, ("cifs_open returned 0x%x", rc));
291 kmalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
292 if (file->private_data == NULL) {
296 pCifsFile = cifs_init_private(file->private_data, inode, file, netfid);
297 write_lock(&GlobalSMBSeslock);
298 list_add(&pCifsFile->tlist, &pTcon->openFileList);
300 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
302 rc = cifs_open_inode_helper(inode, file, pCifsInode,
304 &oplock, buf, full_path, xid);
306 write_unlock(&GlobalSMBSeslock);
309 if (oplock & CIFS_CREATE_ACTION) {
310 /* time to set mode which we can not set earlier due to
311 problems creating new read-only files */
312 if (pTcon->unix_ext) {
313 struct cifs_unix_set_info_args args = {
314 .mode = inode->i_mode,
317 .ctime = NO_CHANGE_64,
318 .atime = NO_CHANGE_64,
319 .mtime = NO_CHANGE_64,
322 CIFSSMBUnixSetInfo(xid, pTcon, full_path, &args,
324 cifs_sb->mnt_cifs_flags &
325 CIFS_MOUNT_MAP_SPECIAL_CHR);
336 /* Try to reacquire byte range locks that were released when session */
337 /* to server was lost */
338 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
342 /* BB list all locks open on this file and relock */
347 static int cifs_reopen_file(struct file *file, bool can_flush)
351 struct cifs_sb_info *cifs_sb;
352 struct cifsTconInfo *pTcon;
353 struct cifsFileInfo *pCifsFile;
354 struct cifsInodeInfo *pCifsInode;
356 char *full_path = NULL;
358 int disposition = FILE_OPEN;
361 if (file->private_data)
362 pCifsFile = (struct cifsFileInfo *)file->private_data;
367 down(&pCifsFile->fh_sem);
368 if (!pCifsFile->invalidHandle) {
369 up(&pCifsFile->fh_sem);
374 if (file->f_path.dentry == NULL) {
375 cERROR(1, ("no valid name if dentry freed"));
378 goto reopen_error_exit;
381 inode = file->f_path.dentry->d_inode;
383 cERROR(1, ("inode not valid"));
386 goto reopen_error_exit;
389 cifs_sb = CIFS_SB(inode->i_sb);
390 pTcon = cifs_sb->tcon;
392 /* can not grab rename sem here because various ops, including
393 those that already have the rename sem can end up causing writepage
394 to get called and if the server was down that means we end up here,
395 and we can never tell if the caller already has the rename_sem */
396 full_path = build_path_from_dentry(file->f_path.dentry);
397 if (full_path == NULL) {
400 up(&pCifsFile->fh_sem);
405 cFYI(1, ("inode = 0x%p file flags 0x%x for %s",
406 inode, file->f_flags, full_path));
407 desiredAccess = cifs_convert_flags(file->f_flags);
414 /* Can not refresh inode by passing in file_info buf to be returned
415 by SMBOpen and then calling get_inode_info with returned buf
416 since file might have write behind data that needs to be flushed
417 and server version of file size can be stale. If we knew for sure
418 that inode was not dirty locally we could do this */
420 rc = CIFSSMBOpen(xid, pTcon, full_path, disposition, desiredAccess,
421 CREATE_NOT_DIR, &netfid, &oplock, NULL,
422 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
423 CIFS_MOUNT_MAP_SPECIAL_CHR);
425 up(&pCifsFile->fh_sem);
426 cFYI(1, ("cifs_open returned 0x%x", rc));
427 cFYI(1, ("oplock: %d", oplock));
429 pCifsFile->netfid = netfid;
430 pCifsFile->invalidHandle = false;
431 up(&pCifsFile->fh_sem);
432 pCifsInode = CIFS_I(inode);
435 rc = filemap_write_and_wait(inode->i_mapping);
437 CIFS_I(inode)->write_behind_rc = rc;
438 /* temporarily disable caching while we
439 go to server to get inode info */
440 pCifsInode->clientCanCacheAll = false;
441 pCifsInode->clientCanCacheRead = false;
443 rc = cifs_get_inode_info_unix(&inode,
444 full_path, inode->i_sb, xid);
446 rc = cifs_get_inode_info(&inode,
447 full_path, NULL, inode->i_sb,
449 } /* else we are writing out data to server already
450 and could deadlock if we tried to flush data, and
451 since we do not know if we have data that would
452 invalidate the current end of file on the server
453 we can not go to the server to get the new inod
455 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
456 pCifsInode->clientCanCacheAll = true;
457 pCifsInode->clientCanCacheRead = true;
458 cFYI(1, ("Exclusive Oplock granted on inode %p",
459 file->f_path.dentry->d_inode));
460 } else if ((oplock & 0xF) == OPLOCK_READ) {
461 pCifsInode->clientCanCacheRead = true;
462 pCifsInode->clientCanCacheAll = false;
464 pCifsInode->clientCanCacheRead = false;
465 pCifsInode->clientCanCacheAll = false;
467 cifs_relock_file(pCifsFile);
476 int cifs_close(struct inode *inode, struct file *file)
480 struct cifs_sb_info *cifs_sb;
481 struct cifsTconInfo *pTcon;
482 struct cifsFileInfo *pSMBFile =
483 (struct cifsFileInfo *)file->private_data;
487 cifs_sb = CIFS_SB(inode->i_sb);
488 pTcon = cifs_sb->tcon;
490 struct cifsLockInfo *li, *tmp;
492 pSMBFile->closePend = true;
494 /* no sense reconnecting to close a file that is
496 if (pTcon->tidStatus != CifsNeedReconnect) {
498 while ((atomic_read(&pSMBFile->wrtPending) != 0)
499 && (timeout <= 2048)) {
500 /* Give write a better chance to get to
501 server ahead of the close. We do not
502 want to add a wait_q here as it would
503 increase the memory utilization as
504 the struct would be in each open file,
505 but this should give enough time to
508 ("close delay, write pending"));
512 if (atomic_read(&pSMBFile->wrtPending))
514 ("close with pending writes"));
515 rc = CIFSSMBClose(xid, pTcon,
520 /* Delete any outstanding lock records.
521 We'll lose them when the file is closed anyway. */
522 mutex_lock(&pSMBFile->lock_mutex);
523 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
524 list_del(&li->llist);
527 mutex_unlock(&pSMBFile->lock_mutex);
529 write_lock(&GlobalSMBSeslock);
530 list_del(&pSMBFile->flist);
531 list_del(&pSMBFile->tlist);
532 write_unlock(&GlobalSMBSeslock);
534 /* We waited above to give the SMBWrite a chance to issue
535 on the wire (so we do not get SMBWrite returning EBADF
536 if writepages is racing with close. Note that writepages
537 does not specify a file handle, so it is possible for a file
538 to be opened twice, and the application close the "wrong"
539 file handle - in these cases we delay long enough to allow
540 the SMBWrite to get on the wire before the SMB Close.
541 We allow total wait here over 45 seconds, more than
542 oplock break time, and more than enough to allow any write
543 to complete on the server, or to time out on the client */
544 while ((atomic_read(&pSMBFile->wrtPending) != 0)
545 && (timeout <= 50000)) {
546 cERROR(1, ("writes pending, delay free of handle"));
550 kfree(file->private_data);
551 file->private_data = NULL;
555 read_lock(&GlobalSMBSeslock);
556 if (list_empty(&(CIFS_I(inode)->openFileList))) {
557 cFYI(1, ("closing last open instance for inode %p", inode));
558 /* if the file is not open we do not know if we can cache info
559 on this inode, much less write behind and read ahead */
560 CIFS_I(inode)->clientCanCacheRead = false;
561 CIFS_I(inode)->clientCanCacheAll = false;
563 read_unlock(&GlobalSMBSeslock);
564 if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
565 rc = CIFS_I(inode)->write_behind_rc;
570 int cifs_closedir(struct inode *inode, struct file *file)
574 struct cifsFileInfo *pCFileStruct =
575 (struct cifsFileInfo *)file->private_data;
578 cFYI(1, ("Closedir inode = 0x%p", inode));
583 struct cifsTconInfo *pTcon;
584 struct cifs_sb_info *cifs_sb =
585 CIFS_SB(file->f_path.dentry->d_sb);
587 pTcon = cifs_sb->tcon;
589 cFYI(1, ("Freeing private data in close dir"));
590 if (!pCFileStruct->srch_inf.endOfSearch &&
591 !pCFileStruct->invalidHandle) {
592 pCFileStruct->invalidHandle = true;
593 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
594 cFYI(1, ("Closing uncompleted readdir with rc %d",
596 /* not much we can do if it fails anyway, ignore rc */
599 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
601 cFYI(1, ("closedir free smb buf in srch struct"));
602 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
603 if (pCFileStruct->srch_inf.smallBuf)
604 cifs_small_buf_release(ptmp);
606 cifs_buf_release(ptmp);
608 kfree(file->private_data);
609 file->private_data = NULL;
611 /* BB can we lock the filestruct while this is going on? */
616 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
617 __u64 offset, __u8 lockType)
619 struct cifsLockInfo *li =
620 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
626 mutex_lock(&fid->lock_mutex);
627 list_add(&li->llist, &fid->llist);
628 mutex_unlock(&fid->lock_mutex);
632 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
638 bool wait_flag = false;
639 struct cifs_sb_info *cifs_sb;
640 struct cifsTconInfo *pTcon;
642 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
645 length = 1 + pfLock->fl_end - pfLock->fl_start;
649 cFYI(1, ("Lock parm: 0x%x flockflags: "
650 "0x%x flocktype: 0x%x start: %lld end: %lld",
651 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
654 if (pfLock->fl_flags & FL_POSIX)
656 if (pfLock->fl_flags & FL_FLOCK)
658 if (pfLock->fl_flags & FL_SLEEP) {
659 cFYI(1, ("Blocking lock"));
662 if (pfLock->fl_flags & FL_ACCESS)
663 cFYI(1, ("Process suspended by mandatory locking - "
664 "not implemented yet"));
665 if (pfLock->fl_flags & FL_LEASE)
666 cFYI(1, ("Lease on file - not implemented yet"));
667 if (pfLock->fl_flags &
668 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
669 cFYI(1, ("Unknown lock flags 0x%x", pfLock->fl_flags));
671 if (pfLock->fl_type == F_WRLCK) {
672 cFYI(1, ("F_WRLCK "));
674 } else if (pfLock->fl_type == F_UNLCK) {
675 cFYI(1, ("F_UNLCK"));
677 /* Check if unlock includes more than
679 } else if (pfLock->fl_type == F_RDLCK) {
680 cFYI(1, ("F_RDLCK"));
681 lockType |= LOCKING_ANDX_SHARED_LOCK;
683 } else if (pfLock->fl_type == F_EXLCK) {
684 cFYI(1, ("F_EXLCK"));
686 } else if (pfLock->fl_type == F_SHLCK) {
687 cFYI(1, ("F_SHLCK"));
688 lockType |= LOCKING_ANDX_SHARED_LOCK;
691 cFYI(1, ("Unknown type of lock"));
693 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
694 pTcon = cifs_sb->tcon;
696 if (file->private_data == NULL) {
700 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
702 posix_locking = (cifs_sb->tcon->ses->capabilities & CAP_UNIX) &&
703 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(cifs_sb->tcon->fsUnixInfo.Capability));
705 /* BB add code here to normalize offset and length to
706 account for negative length which we can not accept over the
711 if (lockType & LOCKING_ANDX_SHARED_LOCK)
712 posix_lock_type = CIFS_RDLCK;
714 posix_lock_type = CIFS_WRLCK;
715 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 1 /* get */,
717 posix_lock_type, wait_flag);
722 /* BB we could chain these into one lock request BB */
723 rc = CIFSSMBLock(xid, pTcon, netfid, length, pfLock->fl_start,
724 0, 1, lockType, 0 /* wait flag */ );
726 rc = CIFSSMBLock(xid, pTcon, netfid, length,
727 pfLock->fl_start, 1 /* numUnlock */ ,
728 0 /* numLock */ , lockType,
730 pfLock->fl_type = F_UNLCK;
732 cERROR(1, ("Error unlocking previously locked "
733 "range %d during test of lock", rc));
737 /* if rc == ERR_SHARING_VIOLATION ? */
738 rc = 0; /* do not change lock type to unlock
739 since range in use */
746 if (!numLock && !numUnlock) {
747 /* if no lock or unlock then nothing
748 to do since we do not know what it is */
755 if (lockType & LOCKING_ANDX_SHARED_LOCK)
756 posix_lock_type = CIFS_RDLCK;
758 posix_lock_type = CIFS_WRLCK;
761 posix_lock_type = CIFS_UNLCK;
763 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 0 /* set */,
765 posix_lock_type, wait_flag);
767 struct cifsFileInfo *fid =
768 (struct cifsFileInfo *)file->private_data;
771 rc = CIFSSMBLock(xid, pTcon, netfid, length,
773 0, numLock, lockType, wait_flag);
776 /* For Windows locks we must store them. */
777 rc = store_file_lock(fid, length,
778 pfLock->fl_start, lockType);
780 } else if (numUnlock) {
781 /* For each stored lock that this unlock overlaps
782 completely, unlock it. */
784 struct cifsLockInfo *li, *tmp;
787 mutex_lock(&fid->lock_mutex);
788 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
789 if (pfLock->fl_start <= li->offset &&
790 (pfLock->fl_start + length) >=
791 (li->offset + li->length)) {
792 stored_rc = CIFSSMBLock(xid, pTcon,
794 li->length, li->offset,
795 1, 0, li->type, false);
799 list_del(&li->llist);
803 mutex_unlock(&fid->lock_mutex);
807 if (pfLock->fl_flags & FL_POSIX)
808 posix_lock_file_wait(file, pfLock);
813 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
814 size_t write_size, loff_t *poffset)
817 unsigned int bytes_written = 0;
818 unsigned int total_written;
819 struct cifs_sb_info *cifs_sb;
820 struct cifsTconInfo *pTcon;
822 struct cifsFileInfo *open_file;
824 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
826 pTcon = cifs_sb->tcon;
829 (" write %d bytes to offset %lld of %s", write_size,
830 *poffset, file->f_path.dentry->d_name.name)); */
832 if (file->private_data == NULL)
834 open_file = (struct cifsFileInfo *) file->private_data;
836 rc = generic_write_checks(file, poffset, &write_size, 0);
842 if (*poffset > file->f_path.dentry->d_inode->i_size)
843 long_op = CIFS_VLONG_OP; /* writes past EOF take long time */
845 long_op = CIFS_LONG_OP;
847 for (total_written = 0; write_size > total_written;
848 total_written += bytes_written) {
850 while (rc == -EAGAIN) {
851 if (file->private_data == NULL) {
852 /* file has been closed on us */
854 /* if we have gotten here we have written some data
855 and blocked, and the file has been freed on us while
856 we blocked so return what we managed to write */
857 return total_written;
859 if (open_file->closePend) {
862 return total_written;
866 if (open_file->invalidHandle) {
867 /* we could deadlock if we called
868 filemap_fdatawait from here so tell
869 reopen_file not to flush data to server
871 rc = cifs_reopen_file(file, false);
876 rc = CIFSSMBWrite(xid, pTcon,
878 min_t(const int, cifs_sb->wsize,
879 write_size - total_written),
880 *poffset, &bytes_written,
881 NULL, write_data + total_written, long_op);
883 if (rc || (bytes_written == 0)) {
891 *poffset += bytes_written;
892 long_op = CIFS_STD_OP; /* subsequent writes fast -
893 15 seconds is plenty */
896 cifs_stats_bytes_written(pTcon, total_written);
898 /* since the write may have blocked check these pointers again */
899 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
900 struct inode *inode = file->f_path.dentry->d_inode;
901 /* Do not update local mtime - server will set its actual value on write
902 * inode->i_ctime = inode->i_mtime =
903 * current_fs_time(inode->i_sb);*/
904 if (total_written > 0) {
905 spin_lock(&inode->i_lock);
906 if (*poffset > file->f_path.dentry->d_inode->i_size)
907 i_size_write(file->f_path.dentry->d_inode,
909 spin_unlock(&inode->i_lock);
911 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
914 return total_written;
917 static ssize_t cifs_write(struct file *file, const char *write_data,
918 size_t write_size, loff_t *poffset)
921 unsigned int bytes_written = 0;
922 unsigned int total_written;
923 struct cifs_sb_info *cifs_sb;
924 struct cifsTconInfo *pTcon;
926 struct cifsFileInfo *open_file;
928 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
930 pTcon = cifs_sb->tcon;
932 cFYI(1, ("write %zd bytes to offset %lld of %s", write_size,
933 *poffset, file->f_path.dentry->d_name.name));
935 if (file->private_data == NULL)
937 open_file = (struct cifsFileInfo *)file->private_data;
941 if (*poffset > file->f_path.dentry->d_inode->i_size)
942 long_op = CIFS_VLONG_OP; /* writes past EOF can be slow */
944 long_op = CIFS_LONG_OP;
946 for (total_written = 0; write_size > total_written;
947 total_written += bytes_written) {
949 while (rc == -EAGAIN) {
950 if (file->private_data == NULL) {
951 /* file has been closed on us */
953 /* if we have gotten here we have written some data
954 and blocked, and the file has been freed on us
955 while we blocked so return what we managed to
957 return total_written;
959 if (open_file->closePend) {
962 return total_written;
966 if (open_file->invalidHandle) {
967 /* we could deadlock if we called
968 filemap_fdatawait from here so tell
969 reopen_file not to flush data to
971 rc = cifs_reopen_file(file, false);
975 if (experimEnabled || (pTcon->ses->server &&
976 ((pTcon->ses->server->secMode &
977 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
982 len = min((size_t)cifs_sb->wsize,
983 write_size - total_written);
984 /* iov[0] is reserved for smb header */
985 iov[1].iov_base = (char *)write_data +
987 iov[1].iov_len = len;
988 rc = CIFSSMBWrite2(xid, pTcon,
989 open_file->netfid, len,
990 *poffset, &bytes_written,
993 rc = CIFSSMBWrite(xid, pTcon,
995 min_t(const int, cifs_sb->wsize,
996 write_size - total_written),
997 *poffset, &bytes_written,
998 write_data + total_written,
1001 if (rc || (bytes_written == 0)) {
1009 *poffset += bytes_written;
1010 long_op = CIFS_STD_OP; /* subsequent writes fast -
1011 15 seconds is plenty */
1014 cifs_stats_bytes_written(pTcon, total_written);
1016 /* since the write may have blocked check these pointers again */
1017 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1018 /*BB We could make this contingent on superblock ATIME flag too */
1019 /* file->f_path.dentry->d_inode->i_ctime =
1020 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1021 if (total_written > 0) {
1022 spin_lock(&file->f_path.dentry->d_inode->i_lock);
1023 if (*poffset > file->f_path.dentry->d_inode->i_size)
1024 i_size_write(file->f_path.dentry->d_inode,
1026 spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1028 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1031 return total_written;
1034 #ifdef CONFIG_CIFS_EXPERIMENTAL
1035 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode)
1037 struct cifsFileInfo *open_file = NULL;
1039 read_lock(&GlobalSMBSeslock);
1040 /* we could simply get the first_list_entry since write-only entries
1041 are always at the end of the list but since the first entry might
1042 have a close pending, we go through the whole list */
1043 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1044 if (open_file->closePend)
1046 if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
1047 (open_file->pfile->f_flags & O_RDONLY))) {
1048 if (!open_file->invalidHandle) {
1049 /* found a good file */
1050 /* lock it so it will not be closed on us */
1051 atomic_inc(&open_file->wrtPending);
1052 read_unlock(&GlobalSMBSeslock);
1054 } /* else might as well continue, and look for
1055 another, or simply have the caller reopen it
1056 again rather than trying to fix this handle */
1057 } else /* write only file */
1058 break; /* write only files are last so must be done */
1060 read_unlock(&GlobalSMBSeslock);
1065 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1067 struct cifsFileInfo *open_file;
1068 bool any_available = false;
1071 /* Having a null inode here (because mapping->host was set to zero by
1072 the VFS or MM) should not happen but we had reports of on oops (due to
1073 it being zero) during stress testcases so we need to check for it */
1075 if (cifs_inode == NULL) {
1076 cERROR(1, ("Null inode passed to cifs_writeable_file"));
1081 read_lock(&GlobalSMBSeslock);
1083 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1084 if (open_file->closePend ||
1085 (!any_available && open_file->pid != current->tgid))
1088 if (open_file->pfile &&
1089 ((open_file->pfile->f_flags & O_RDWR) ||
1090 (open_file->pfile->f_flags & O_WRONLY))) {
1091 atomic_inc(&open_file->wrtPending);
1093 if (!open_file->invalidHandle) {
1094 /* found a good writable file */
1095 read_unlock(&GlobalSMBSeslock);
1099 read_unlock(&GlobalSMBSeslock);
1100 /* Had to unlock since following call can block */
1101 rc = cifs_reopen_file(open_file->pfile, false);
1103 if (!open_file->closePend)
1105 else { /* start over in case this was deleted */
1106 /* since the list could be modified */
1107 read_lock(&GlobalSMBSeslock);
1108 atomic_dec(&open_file->wrtPending);
1109 goto refind_writable;
1113 /* if it fails, try another handle if possible -
1114 (we can not do this if closePending since
1115 loop could be modified - in which case we
1116 have to start at the beginning of the list
1117 again. Note that it would be bad
1118 to hold up writepages here (rather than
1119 in caller) with continuous retries */
1120 cFYI(1, ("wp failed on reopen file"));
1121 read_lock(&GlobalSMBSeslock);
1122 /* can not use this handle, no write
1123 pending on this one after all */
1124 atomic_dec(&open_file->wrtPending);
1126 if (open_file->closePend) /* list could have changed */
1127 goto refind_writable;
1128 /* else we simply continue to the next entry. Thus
1129 we do not loop on reopen errors. If we
1130 can not reopen the file, for example if we
1131 reconnected to a server with another client
1132 racing to delete or lock the file we would not
1133 make progress if we restarted before the beginning
1134 of the loop here. */
1137 /* couldn't find useable FH with same pid, try any available */
1138 if (!any_available) {
1139 any_available = true;
1140 goto refind_writable;
1142 read_unlock(&GlobalSMBSeslock);
1146 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1148 struct address_space *mapping = page->mapping;
1149 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1152 int bytes_written = 0;
1153 struct cifs_sb_info *cifs_sb;
1154 struct cifsTconInfo *pTcon;
1155 struct inode *inode;
1156 struct cifsFileInfo *open_file;
1158 if (!mapping || !mapping->host)
1161 inode = page->mapping->host;
1162 cifs_sb = CIFS_SB(inode->i_sb);
1163 pTcon = cifs_sb->tcon;
1165 offset += (loff_t)from;
1166 write_data = kmap(page);
1169 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1174 /* racing with truncate? */
1175 if (offset > mapping->host->i_size) {
1177 return 0; /* don't care */
1180 /* check to make sure that we are not extending the file */
1181 if (mapping->host->i_size - offset < (loff_t)to)
1182 to = (unsigned)(mapping->host->i_size - offset);
1184 open_file = find_writable_file(CIFS_I(mapping->host));
1186 bytes_written = cifs_write(open_file->pfile, write_data,
1188 atomic_dec(&open_file->wrtPending);
1189 /* Does mm or vfs already set times? */
1190 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1191 if ((bytes_written > 0) && (offset))
1193 else if (bytes_written < 0)
1196 cFYI(1, ("No writeable filehandles for inode"));
1204 static int cifs_writepages(struct address_space *mapping,
1205 struct writeback_control *wbc)
1207 struct backing_dev_info *bdi = mapping->backing_dev_info;
1208 unsigned int bytes_to_write;
1209 unsigned int bytes_written;
1210 struct cifs_sb_info *cifs_sb;
1214 int range_whole = 0;
1221 struct cifsFileInfo *open_file;
1223 struct pagevec pvec;
1228 cifs_sb = CIFS_SB(mapping->host->i_sb);
1231 * If wsize is smaller that the page cache size, default to writing
1232 * one page at a time via cifs_writepage
1234 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1235 return generic_writepages(mapping, wbc);
1237 if ((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1238 if (cifs_sb->tcon->ses->server->secMode &
1239 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1240 if (!experimEnabled)
1241 return generic_writepages(mapping, wbc);
1243 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1245 return generic_writepages(mapping, wbc);
1249 * BB: Is this meaningful for a non-block-device file system?
1250 * If it is, we should test it again after we do I/O
1252 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1253 wbc->encountered_congestion = 1;
1260 pagevec_init(&pvec, 0);
1261 if (wbc->range_cyclic) {
1262 index = mapping->writeback_index; /* Start from prev offset */
1265 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1266 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1267 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1272 while (!done && (index <= end) &&
1273 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1274 PAGECACHE_TAG_DIRTY,
1275 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1284 for (i = 0; i < nr_pages; i++) {
1285 page = pvec.pages[i];
1287 * At this point we hold neither mapping->tree_lock nor
1288 * lock on the page itself: the page may be truncated or
1289 * invalidated (changing page->mapping to NULL), or even
1290 * swizzled back from swapper_space to tmpfs file
1296 else if (!trylock_page(page))
1299 if (unlikely(page->mapping != mapping)) {
1304 if (!wbc->range_cyclic && page->index > end) {
1310 if (next && (page->index != next)) {
1311 /* Not next consecutive page */
1316 if (wbc->sync_mode != WB_SYNC_NONE)
1317 wait_on_page_writeback(page);
1319 if (PageWriteback(page) ||
1320 !clear_page_dirty_for_io(page)) {
1326 * This actually clears the dirty bit in the radix tree.
1327 * See cifs_writepage() for more commentary.
1329 set_page_writeback(page);
1331 if (page_offset(page) >= mapping->host->i_size) {
1334 end_page_writeback(page);
1339 * BB can we get rid of this? pages are held by pvec
1341 page_cache_get(page);
1343 len = min(mapping->host->i_size - page_offset(page),
1344 (loff_t)PAGE_CACHE_SIZE);
1346 /* reserve iov[0] for the smb header */
1348 iov[n_iov].iov_base = kmap(page);
1349 iov[n_iov].iov_len = len;
1350 bytes_to_write += len;
1354 offset = page_offset(page);
1356 next = page->index + 1;
1357 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1361 /* Search for a writable handle every time we call
1362 * CIFSSMBWrite2. We can't rely on the last handle
1363 * we used to still be valid
1365 open_file = find_writable_file(CIFS_I(mapping->host));
1367 cERROR(1, ("No writable handles for inode"));
1370 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1372 bytes_to_write, offset,
1373 &bytes_written, iov, n_iov,
1375 atomic_dec(&open_file->wrtPending);
1376 if (rc || bytes_written < bytes_to_write) {
1377 cERROR(1, ("Write2 ret %d, wrote %d",
1378 rc, bytes_written));
1379 /* BB what if continued retry is
1380 requested via mount flags? */
1382 set_bit(AS_ENOSPC, &mapping->flags);
1384 set_bit(AS_EIO, &mapping->flags);
1386 cifs_stats_bytes_written(cifs_sb->tcon,
1390 for (i = 0; i < n_iov; i++) {
1391 page = pvec.pages[first + i];
1392 /* Should we also set page error on
1393 success rc but too little data written? */
1394 /* BB investigate retry logic on temporary
1395 server crash cases and how recovery works
1396 when page marked as error */
1401 end_page_writeback(page);
1402 page_cache_release(page);
1404 if ((wbc->nr_to_write -= n_iov) <= 0)
1408 pagevec_release(&pvec);
1410 if (!scanned && !done) {
1412 * We hit the last page and there is more work to be done: wrap
1413 * back to the start of the file
1419 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1420 mapping->writeback_index = index;
1427 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1433 /* BB add check for wbc flags */
1434 page_cache_get(page);
1435 if (!PageUptodate(page))
1436 cFYI(1, ("ppw - page not up to date"));
1439 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1441 * A writepage() implementation always needs to do either this,
1442 * or re-dirty the page with "redirty_page_for_writepage()" in
1443 * the case of a failure.
1445 * Just unlocking the page will cause the radix tree tag-bits
1446 * to fail to update with the state of the page correctly.
1448 set_page_writeback(page);
1449 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1450 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1452 end_page_writeback(page);
1453 page_cache_release(page);
1458 static int cifs_commit_write(struct file *file, struct page *page,
1459 unsigned offset, unsigned to)
1463 struct inode *inode = page->mapping->host;
1464 loff_t position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1468 cFYI(1, ("commit write for page %p up to position %lld for %d",
1469 page, position, to));
1470 spin_lock(&inode->i_lock);
1471 if (position > inode->i_size)
1472 i_size_write(inode, position);
1474 spin_unlock(&inode->i_lock);
1475 if (!PageUptodate(page)) {
1476 position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + offset;
1477 /* can not rely on (or let) writepage write this data */
1479 cFYI(1, ("Illegal offsets, can not copy from %d to %d",
1484 /* this is probably better than directly calling
1485 partialpage_write since in this function the file handle is
1486 known which we might as well leverage */
1487 /* BB check if anything else missing out of ppw
1488 such as updating last write time */
1489 page_data = kmap(page);
1490 rc = cifs_write(file, page_data + offset, to-offset,
1494 /* else if (rc < 0) should we set writebehind rc? */
1497 set_page_dirty(page);
1504 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1508 struct inode *inode = file->f_path.dentry->d_inode;
1512 cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1513 dentry->d_name.name, datasync));
1515 rc = filemap_write_and_wait(inode->i_mapping);
1517 rc = CIFS_I(inode)->write_behind_rc;
1518 CIFS_I(inode)->write_behind_rc = 0;
1524 /* static void cifs_sync_page(struct page *page)
1526 struct address_space *mapping;
1527 struct inode *inode;
1528 unsigned long index = page->index;
1529 unsigned int rpages = 0;
1532 cFYI(1, ("sync page %p",page));
1533 mapping = page->mapping;
1536 inode = mapping->host;
1540 /* fill in rpages then
1541 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1543 /* cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1553 * As file closes, flush all cached write data for this inode checking
1554 * for write behind errors.
1556 int cifs_flush(struct file *file, fl_owner_t id)
1558 struct inode *inode = file->f_path.dentry->d_inode;
1561 /* Rather than do the steps manually:
1562 lock the inode for writing
1563 loop through pages looking for write behind data (dirty pages)
1564 coalesce into contiguous 16K (or smaller) chunks to write to server
1565 send to server (prefer in parallel)
1566 deal with writebehind errors
1567 unlock inode for writing
1568 filemapfdatawrite appears easier for the time being */
1570 rc = filemap_fdatawrite(inode->i_mapping);
1571 /* reset wb rc if we were able to write out dirty pages */
1573 rc = CIFS_I(inode)->write_behind_rc;
1574 CIFS_I(inode)->write_behind_rc = 0;
1577 cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
1582 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1583 size_t read_size, loff_t *poffset)
1586 unsigned int bytes_read = 0;
1587 unsigned int total_read = 0;
1588 unsigned int current_read_size;
1589 struct cifs_sb_info *cifs_sb;
1590 struct cifsTconInfo *pTcon;
1592 struct cifsFileInfo *open_file;
1593 char *smb_read_data;
1594 char __user *current_offset;
1595 struct smb_com_read_rsp *pSMBr;
1598 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1599 pTcon = cifs_sb->tcon;
1601 if (file->private_data == NULL) {
1605 open_file = (struct cifsFileInfo *)file->private_data;
1607 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1608 cFYI(1, ("attempting read on write only file instance"));
1610 for (total_read = 0, current_offset = read_data;
1611 read_size > total_read;
1612 total_read += bytes_read, current_offset += bytes_read) {
1613 current_read_size = min_t(const int, read_size - total_read,
1616 smb_read_data = NULL;
1617 while (rc == -EAGAIN) {
1618 int buf_type = CIFS_NO_BUFFER;
1619 if ((open_file->invalidHandle) &&
1620 (!open_file->closePend)) {
1621 rc = cifs_reopen_file(file, true);
1625 rc = CIFSSMBRead(xid, pTcon,
1627 current_read_size, *poffset,
1628 &bytes_read, &smb_read_data,
1630 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1631 if (smb_read_data) {
1632 if (copy_to_user(current_offset,
1634 4 /* RFC1001 length field */ +
1635 le16_to_cpu(pSMBr->DataOffset),
1639 if (buf_type == CIFS_SMALL_BUFFER)
1640 cifs_small_buf_release(smb_read_data);
1641 else if (buf_type == CIFS_LARGE_BUFFER)
1642 cifs_buf_release(smb_read_data);
1643 smb_read_data = NULL;
1646 if (rc || (bytes_read == 0)) {
1654 cifs_stats_bytes_read(pTcon, bytes_read);
1655 *poffset += bytes_read;
1663 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1667 unsigned int bytes_read = 0;
1668 unsigned int total_read;
1669 unsigned int current_read_size;
1670 struct cifs_sb_info *cifs_sb;
1671 struct cifsTconInfo *pTcon;
1673 char *current_offset;
1674 struct cifsFileInfo *open_file;
1675 int buf_type = CIFS_NO_BUFFER;
1678 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1679 pTcon = cifs_sb->tcon;
1681 if (file->private_data == NULL) {
1685 open_file = (struct cifsFileInfo *)file->private_data;
1687 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1688 cFYI(1, ("attempting read on write only file instance"));
1690 for (total_read = 0, current_offset = read_data;
1691 read_size > total_read;
1692 total_read += bytes_read, current_offset += bytes_read) {
1693 current_read_size = min_t(const int, read_size - total_read,
1695 /* For windows me and 9x we do not want to request more
1696 than it negotiated since it will refuse the read then */
1698 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1699 current_read_size = min_t(const int, current_read_size,
1700 pTcon->ses->server->maxBuf - 128);
1703 while (rc == -EAGAIN) {
1704 if ((open_file->invalidHandle) &&
1705 (!open_file->closePend)) {
1706 rc = cifs_reopen_file(file, true);
1710 rc = CIFSSMBRead(xid, pTcon,
1712 current_read_size, *poffset,
1713 &bytes_read, ¤t_offset,
1716 if (rc || (bytes_read == 0)) {
1724 cifs_stats_bytes_read(pTcon, total_read);
1725 *poffset += bytes_read;
1732 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1734 struct dentry *dentry = file->f_path.dentry;
1738 rc = cifs_revalidate(dentry);
1740 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1744 rc = generic_file_mmap(file, vma);
1750 static void cifs_copy_cache_pages(struct address_space *mapping,
1751 struct list_head *pages, int bytes_read, char *data,
1752 struct pagevec *plru_pvec)
1757 while (bytes_read > 0) {
1758 if (list_empty(pages))
1761 page = list_entry(pages->prev, struct page, lru);
1762 list_del(&page->lru);
1764 if (add_to_page_cache(page, mapping, page->index,
1766 page_cache_release(page);
1767 cFYI(1, ("Add page cache failed"));
1768 data += PAGE_CACHE_SIZE;
1769 bytes_read -= PAGE_CACHE_SIZE;
1773 target = kmap_atomic(page, KM_USER0);
1775 if (PAGE_CACHE_SIZE > bytes_read) {
1776 memcpy(target, data, bytes_read);
1777 /* zero the tail end of this partial page */
1778 memset(target + bytes_read, 0,
1779 PAGE_CACHE_SIZE - bytes_read);
1782 memcpy(target, data, PAGE_CACHE_SIZE);
1783 bytes_read -= PAGE_CACHE_SIZE;
1785 kunmap_atomic(target, KM_USER0);
1787 flush_dcache_page(page);
1788 SetPageUptodate(page);
1790 if (!pagevec_add(plru_pvec, page))
1791 __pagevec_lru_add(plru_pvec);
1792 data += PAGE_CACHE_SIZE;
1797 static int cifs_readpages(struct file *file, struct address_space *mapping,
1798 struct list_head *page_list, unsigned num_pages)
1804 struct cifs_sb_info *cifs_sb;
1805 struct cifsTconInfo *pTcon;
1806 unsigned int bytes_read = 0;
1807 unsigned int read_size, i;
1808 char *smb_read_data = NULL;
1809 struct smb_com_read_rsp *pSMBr;
1810 struct pagevec lru_pvec;
1811 struct cifsFileInfo *open_file;
1812 int buf_type = CIFS_NO_BUFFER;
1815 if (file->private_data == NULL) {
1819 open_file = (struct cifsFileInfo *)file->private_data;
1820 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1821 pTcon = cifs_sb->tcon;
1823 pagevec_init(&lru_pvec, 0);
1824 cFYI(DBG2, ("rpages: num pages %d", num_pages));
1825 for (i = 0; i < num_pages; ) {
1826 unsigned contig_pages;
1827 struct page *tmp_page;
1828 unsigned long expected_index;
1830 if (list_empty(page_list))
1833 page = list_entry(page_list->prev, struct page, lru);
1834 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1836 /* count adjacent pages that we will read into */
1839 list_entry(page_list->prev, struct page, lru)->index;
1840 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1841 if (tmp_page->index == expected_index) {
1847 if (contig_pages + i > num_pages)
1848 contig_pages = num_pages - i;
1850 /* for reads over a certain size could initiate async
1853 read_size = contig_pages * PAGE_CACHE_SIZE;
1854 /* Read size needs to be in multiples of one page */
1855 read_size = min_t(const unsigned int, read_size,
1856 cifs_sb->rsize & PAGE_CACHE_MASK);
1857 cFYI(DBG2, ("rpages: read size 0x%x contiguous pages %d",
1858 read_size, contig_pages));
1860 while (rc == -EAGAIN) {
1861 if ((open_file->invalidHandle) &&
1862 (!open_file->closePend)) {
1863 rc = cifs_reopen_file(file, true);
1868 rc = CIFSSMBRead(xid, pTcon,
1871 &bytes_read, &smb_read_data,
1873 /* BB more RC checks ? */
1874 if (rc == -EAGAIN) {
1875 if (smb_read_data) {
1876 if (buf_type == CIFS_SMALL_BUFFER)
1877 cifs_small_buf_release(smb_read_data);
1878 else if (buf_type == CIFS_LARGE_BUFFER)
1879 cifs_buf_release(smb_read_data);
1880 smb_read_data = NULL;
1884 if ((rc < 0) || (smb_read_data == NULL)) {
1885 cFYI(1, ("Read error in readpages: %d", rc));
1887 } else if (bytes_read > 0) {
1888 task_io_account_read(bytes_read);
1889 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1890 cifs_copy_cache_pages(mapping, page_list, bytes_read,
1891 smb_read_data + 4 /* RFC1001 hdr */ +
1892 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
1894 i += bytes_read >> PAGE_CACHE_SHIFT;
1895 cifs_stats_bytes_read(pTcon, bytes_read);
1896 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1897 i++; /* account for partial page */
1899 /* server copy of file can have smaller size
1901 /* BB do we need to verify this common case ?
1902 this case is ok - if we are at server EOF
1903 we will hit it on next read */
1908 cFYI(1, ("No bytes read (%d) at offset %lld . "
1909 "Cleaning remaining pages from readahead list",
1910 bytes_read, offset));
1911 /* BB turn off caching and do new lookup on
1912 file size at server? */
1915 if (smb_read_data) {
1916 if (buf_type == CIFS_SMALL_BUFFER)
1917 cifs_small_buf_release(smb_read_data);
1918 else if (buf_type == CIFS_LARGE_BUFFER)
1919 cifs_buf_release(smb_read_data);
1920 smb_read_data = NULL;
1925 pagevec_lru_add(&lru_pvec);
1927 /* need to free smb_read_data buf before exit */
1928 if (smb_read_data) {
1929 if (buf_type == CIFS_SMALL_BUFFER)
1930 cifs_small_buf_release(smb_read_data);
1931 else if (buf_type == CIFS_LARGE_BUFFER)
1932 cifs_buf_release(smb_read_data);
1933 smb_read_data = NULL;
1940 static int cifs_readpage_worker(struct file *file, struct page *page,
1946 page_cache_get(page);
1947 read_data = kmap(page);
1948 /* for reads over a certain size could initiate async read ahead */
1950 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1955 cFYI(1, ("Bytes read %d", rc));
1957 file->f_path.dentry->d_inode->i_atime =
1958 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1960 if (PAGE_CACHE_SIZE > rc)
1961 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1963 flush_dcache_page(page);
1964 SetPageUptodate(page);
1969 page_cache_release(page);
1973 static int cifs_readpage(struct file *file, struct page *page)
1975 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1981 if (file->private_data == NULL) {
1986 cFYI(1, ("readpage %p at offset %d 0x%x\n",
1987 page, (int)offset, (int)offset));
1989 rc = cifs_readpage_worker(file, page, &offset);
1997 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
1999 struct cifsFileInfo *open_file;
2001 read_lock(&GlobalSMBSeslock);
2002 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2003 if (open_file->closePend)
2005 if (open_file->pfile &&
2006 ((open_file->pfile->f_flags & O_RDWR) ||
2007 (open_file->pfile->f_flags & O_WRONLY))) {
2008 read_unlock(&GlobalSMBSeslock);
2012 read_unlock(&GlobalSMBSeslock);
2016 /* We do not want to update the file size from server for inodes
2017 open for write - to avoid races with writepage extending
2018 the file - in the future we could consider allowing
2019 refreshing the inode only on increases in the file size
2020 but this is tricky to do without racing with writebehind
2021 page caching in the current Linux kernel design */
2022 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2027 if (is_inode_writable(cifsInode)) {
2028 /* This inode is open for write at least once */
2029 struct cifs_sb_info *cifs_sb;
2031 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2032 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2033 /* since no page cache to corrupt on directio
2034 we can change size safely */
2038 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2046 static int cifs_prepare_write(struct file *file, struct page *page,
2047 unsigned from, unsigned to)
2053 cFYI(1, ("prepare write for page %p from %d to %d", page, from, to));
2054 if (PageUptodate(page))
2057 /* If we are writing a full page it will be up to date,
2058 no need to read from the server */
2059 if ((to == PAGE_CACHE_SIZE) && (from == 0)) {
2060 SetPageUptodate(page);
2064 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2065 i_size = i_size_read(page->mapping->host);
2067 if ((offset >= i_size) ||
2068 ((from == 0) && (offset + to) >= i_size)) {
2070 * We don't need to read data beyond the end of the file.
2071 * zero it, and set the page uptodate
2073 simple_prepare_write(file, page, from, to);
2074 SetPageUptodate(page);
2075 } else if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2076 /* might as well read a page, it is fast enough */
2077 rc = cifs_readpage_worker(file, page, &offset);
2079 /* we could try using another file handle if there is one -
2080 but how would we lock it to prevent close of that handle
2081 racing with this read? In any case
2082 this will be written out by commit_write so is fine */
2085 /* we do not need to pass errors back
2086 e.g. if we do not have read access to the file
2087 because cifs_commit_write will do the right thing. -- shaggy */
2092 const struct address_space_operations cifs_addr_ops = {
2093 .readpage = cifs_readpage,
2094 .readpages = cifs_readpages,
2095 .writepage = cifs_writepage,
2096 .writepages = cifs_writepages,
2097 .prepare_write = cifs_prepare_write,
2098 .commit_write = cifs_commit_write,
2099 .set_page_dirty = __set_page_dirty_nobuffers,
2100 /* .sync_page = cifs_sync_page, */
2105 * cifs_readpages requires the server to support a buffer large enough to
2106 * contain the header plus one complete page of data. Otherwise, we need
2107 * to leave cifs_readpages out of the address space operations.
2109 const struct address_space_operations cifs_addr_ops_smallbuf = {
2110 .readpage = cifs_readpage,
2111 .writepage = cifs_writepage,
2112 .writepages = cifs_writepages,
2113 .prepare_write = cifs_prepare_write,
2114 .commit_write = cifs_commit_write,
2115 .set_page_dirty = __set_page_dirty_nobuffers,
2116 /* .sync_page = cifs_sync_page, */