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 write_begin */
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;
491 write_lock(&GlobalSMBSeslock);
492 pSMBFile->closePend = true;
494 /* no sense reconnecting to close a file that is
496 if (!pTcon->need_reconnect) {
497 write_unlock(&GlobalSMBSeslock);
499 while ((atomic_read(&pSMBFile->wrtPending) != 0)
500 && (timeout <= 2048)) {
501 /* Give write a better chance to get to
502 server ahead of the close. We do not
503 want to add a wait_q here as it would
504 increase the memory utilization as
505 the struct would be in each open file,
506 but this should give enough time to
509 ("close delay, write pending"));
513 if (atomic_read(&pSMBFile->wrtPending))
514 cERROR(1, ("close with pending write"));
515 if (!pTcon->need_reconnect &&
516 !pSMBFile->invalidHandle)
517 rc = CIFSSMBClose(xid, pTcon,
520 write_unlock(&GlobalSMBSeslock);
522 write_unlock(&GlobalSMBSeslock);
524 /* Delete any outstanding lock records.
525 We'll lose them when the file is closed anyway. */
526 mutex_lock(&pSMBFile->lock_mutex);
527 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
528 list_del(&li->llist);
531 mutex_unlock(&pSMBFile->lock_mutex);
533 write_lock(&GlobalSMBSeslock);
534 list_del(&pSMBFile->flist);
535 list_del(&pSMBFile->tlist);
536 write_unlock(&GlobalSMBSeslock);
538 /* We waited above to give the SMBWrite a chance to issue
539 on the wire (so we do not get SMBWrite returning EBADF
540 if writepages is racing with close. Note that writepages
541 does not specify a file handle, so it is possible for a file
542 to be opened twice, and the application close the "wrong"
543 file handle - in these cases we delay long enough to allow
544 the SMBWrite to get on the wire before the SMB Close.
545 We allow total wait here over 45 seconds, more than
546 oplock break time, and more than enough to allow any write
547 to complete on the server, or to time out on the client */
548 while ((atomic_read(&pSMBFile->wrtPending) != 0)
549 && (timeout <= 50000)) {
550 cERROR(1, ("writes pending, delay free of handle"));
554 kfree(file->private_data);
555 file->private_data = NULL;
559 read_lock(&GlobalSMBSeslock);
560 if (list_empty(&(CIFS_I(inode)->openFileList))) {
561 cFYI(1, ("closing last open instance for inode %p", inode));
562 /* if the file is not open we do not know if we can cache info
563 on this inode, much less write behind and read ahead */
564 CIFS_I(inode)->clientCanCacheRead = false;
565 CIFS_I(inode)->clientCanCacheAll = false;
567 read_unlock(&GlobalSMBSeslock);
568 if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
569 rc = CIFS_I(inode)->write_behind_rc;
574 int cifs_closedir(struct inode *inode, struct file *file)
578 struct cifsFileInfo *pCFileStruct =
579 (struct cifsFileInfo *)file->private_data;
582 cFYI(1, ("Closedir inode = 0x%p", inode));
587 struct cifsTconInfo *pTcon;
588 struct cifs_sb_info *cifs_sb =
589 CIFS_SB(file->f_path.dentry->d_sb);
591 pTcon = cifs_sb->tcon;
593 cFYI(1, ("Freeing private data in close dir"));
594 write_lock(&GlobalSMBSeslock);
595 if (!pCFileStruct->srch_inf.endOfSearch &&
596 !pCFileStruct->invalidHandle) {
597 pCFileStruct->invalidHandle = true;
598 write_unlock(&GlobalSMBSeslock);
599 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
600 cFYI(1, ("Closing uncompleted readdir with rc %d",
602 /* not much we can do if it fails anyway, ignore rc */
605 write_unlock(&GlobalSMBSeslock);
606 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
608 cFYI(1, ("closedir free smb buf in srch struct"));
609 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
610 if (pCFileStruct->srch_inf.smallBuf)
611 cifs_small_buf_release(ptmp);
613 cifs_buf_release(ptmp);
615 kfree(file->private_data);
616 file->private_data = NULL;
618 /* BB can we lock the filestruct while this is going on? */
623 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
624 __u64 offset, __u8 lockType)
626 struct cifsLockInfo *li =
627 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
633 mutex_lock(&fid->lock_mutex);
634 list_add(&li->llist, &fid->llist);
635 mutex_unlock(&fid->lock_mutex);
639 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
645 bool wait_flag = false;
646 struct cifs_sb_info *cifs_sb;
647 struct cifsTconInfo *tcon;
649 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
650 bool posix_locking = 0;
652 length = 1 + pfLock->fl_end - pfLock->fl_start;
656 cFYI(1, ("Lock parm: 0x%x flockflags: "
657 "0x%x flocktype: 0x%x start: %lld end: %lld",
658 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
661 if (pfLock->fl_flags & FL_POSIX)
663 if (pfLock->fl_flags & FL_FLOCK)
665 if (pfLock->fl_flags & FL_SLEEP) {
666 cFYI(1, ("Blocking lock"));
669 if (pfLock->fl_flags & FL_ACCESS)
670 cFYI(1, ("Process suspended by mandatory locking - "
671 "not implemented yet"));
672 if (pfLock->fl_flags & FL_LEASE)
673 cFYI(1, ("Lease on file - not implemented yet"));
674 if (pfLock->fl_flags &
675 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
676 cFYI(1, ("Unknown lock flags 0x%x", pfLock->fl_flags));
678 if (pfLock->fl_type == F_WRLCK) {
679 cFYI(1, ("F_WRLCK "));
681 } else if (pfLock->fl_type == F_UNLCK) {
682 cFYI(1, ("F_UNLCK"));
684 /* Check if unlock includes more than
686 } else if (pfLock->fl_type == F_RDLCK) {
687 cFYI(1, ("F_RDLCK"));
688 lockType |= LOCKING_ANDX_SHARED_LOCK;
690 } else if (pfLock->fl_type == F_EXLCK) {
691 cFYI(1, ("F_EXLCK"));
693 } else if (pfLock->fl_type == F_SHLCK) {
694 cFYI(1, ("F_SHLCK"));
695 lockType |= LOCKING_ANDX_SHARED_LOCK;
698 cFYI(1, ("Unknown type of lock"));
700 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
701 tcon = cifs_sb->tcon;
703 if (file->private_data == NULL) {
707 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
709 if ((tcon->ses->capabilities & CAP_UNIX) &&
710 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
711 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
713 /* BB add code here to normalize offset and length to
714 account for negative length which we can not accept over the
719 if (lockType & LOCKING_ANDX_SHARED_LOCK)
720 posix_lock_type = CIFS_RDLCK;
722 posix_lock_type = CIFS_WRLCK;
723 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
725 posix_lock_type, wait_flag);
730 /* BB we could chain these into one lock request BB */
731 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
732 0, 1, lockType, 0 /* wait flag */ );
734 rc = CIFSSMBLock(xid, tcon, netfid, length,
735 pfLock->fl_start, 1 /* numUnlock */ ,
736 0 /* numLock */ , lockType,
738 pfLock->fl_type = F_UNLCK;
740 cERROR(1, ("Error unlocking previously locked "
741 "range %d during test of lock", rc));
745 /* if rc == ERR_SHARING_VIOLATION ? */
746 rc = 0; /* do not change lock type to unlock
747 since range in use */
754 if (!numLock && !numUnlock) {
755 /* if no lock or unlock then nothing
756 to do since we do not know what it is */
763 if (lockType & LOCKING_ANDX_SHARED_LOCK)
764 posix_lock_type = CIFS_RDLCK;
766 posix_lock_type = CIFS_WRLCK;
769 posix_lock_type = CIFS_UNLCK;
771 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
773 posix_lock_type, wait_flag);
775 struct cifsFileInfo *fid =
776 (struct cifsFileInfo *)file->private_data;
779 rc = CIFSSMBLock(xid, tcon, netfid, length,
781 0, numLock, lockType, wait_flag);
784 /* For Windows locks we must store them. */
785 rc = store_file_lock(fid, length,
786 pfLock->fl_start, lockType);
788 } else if (numUnlock) {
789 /* For each stored lock that this unlock overlaps
790 completely, unlock it. */
792 struct cifsLockInfo *li, *tmp;
795 mutex_lock(&fid->lock_mutex);
796 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
797 if (pfLock->fl_start <= li->offset &&
798 (pfLock->fl_start + length) >=
799 (li->offset + li->length)) {
800 stored_rc = CIFSSMBLock(xid, tcon,
802 li->length, li->offset,
803 1, 0, li->type, false);
807 list_del(&li->llist);
811 mutex_unlock(&fid->lock_mutex);
815 if (pfLock->fl_flags & FL_POSIX)
816 posix_lock_file_wait(file, pfLock);
821 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
822 size_t write_size, loff_t *poffset)
825 unsigned int bytes_written = 0;
826 unsigned int total_written;
827 struct cifs_sb_info *cifs_sb;
828 struct cifsTconInfo *pTcon;
830 struct cifsFileInfo *open_file;
832 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
834 pTcon = cifs_sb->tcon;
837 (" write %d bytes to offset %lld of %s", write_size,
838 *poffset, file->f_path.dentry->d_name.name)); */
840 if (file->private_data == NULL)
842 open_file = (struct cifsFileInfo *) file->private_data;
844 rc = generic_write_checks(file, poffset, &write_size, 0);
850 if (*poffset > file->f_path.dentry->d_inode->i_size)
851 long_op = CIFS_VLONG_OP; /* writes past EOF take long time */
853 long_op = CIFS_LONG_OP;
855 for (total_written = 0; write_size > total_written;
856 total_written += bytes_written) {
858 while (rc == -EAGAIN) {
859 if (file->private_data == NULL) {
860 /* file has been closed on us */
862 /* if we have gotten here we have written some data
863 and blocked, and the file has been freed on us while
864 we blocked so return what we managed to write */
865 return total_written;
867 if (open_file->closePend) {
870 return total_written;
874 if (open_file->invalidHandle) {
875 /* we could deadlock if we called
876 filemap_fdatawait from here so tell
877 reopen_file not to flush data to server
879 rc = cifs_reopen_file(file, false);
884 rc = CIFSSMBWrite(xid, pTcon,
886 min_t(const int, cifs_sb->wsize,
887 write_size - total_written),
888 *poffset, &bytes_written,
889 NULL, write_data + total_written, long_op);
891 if (rc || (bytes_written == 0)) {
899 *poffset += bytes_written;
900 long_op = CIFS_STD_OP; /* subsequent writes fast -
901 15 seconds is plenty */
904 cifs_stats_bytes_written(pTcon, total_written);
906 /* since the write may have blocked check these pointers again */
907 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
908 struct inode *inode = file->f_path.dentry->d_inode;
909 /* Do not update local mtime - server will set its actual value on write
910 * inode->i_ctime = inode->i_mtime =
911 * current_fs_time(inode->i_sb);*/
912 if (total_written > 0) {
913 spin_lock(&inode->i_lock);
914 if (*poffset > file->f_path.dentry->d_inode->i_size)
915 i_size_write(file->f_path.dentry->d_inode,
917 spin_unlock(&inode->i_lock);
919 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
922 return total_written;
925 static ssize_t cifs_write(struct file *file, const char *write_data,
926 size_t write_size, loff_t *poffset)
929 unsigned int bytes_written = 0;
930 unsigned int total_written;
931 struct cifs_sb_info *cifs_sb;
932 struct cifsTconInfo *pTcon;
934 struct cifsFileInfo *open_file;
936 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
938 pTcon = cifs_sb->tcon;
940 cFYI(1, ("write %zd bytes to offset %lld of %s", write_size,
941 *poffset, file->f_path.dentry->d_name.name));
943 if (file->private_data == NULL)
945 open_file = (struct cifsFileInfo *)file->private_data;
949 if (*poffset > file->f_path.dentry->d_inode->i_size)
950 long_op = CIFS_VLONG_OP; /* writes past EOF can be slow */
952 long_op = CIFS_LONG_OP;
954 for (total_written = 0; write_size > total_written;
955 total_written += bytes_written) {
957 while (rc == -EAGAIN) {
958 if (file->private_data == NULL) {
959 /* file has been closed on us */
961 /* if we have gotten here we have written some data
962 and blocked, and the file has been freed on us
963 while we blocked so return what we managed to
965 return total_written;
967 if (open_file->closePend) {
970 return total_written;
974 if (open_file->invalidHandle) {
975 /* we could deadlock if we called
976 filemap_fdatawait from here so tell
977 reopen_file not to flush data to
979 rc = cifs_reopen_file(file, false);
983 if (experimEnabled || (pTcon->ses->server &&
984 ((pTcon->ses->server->secMode &
985 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
990 len = min((size_t)cifs_sb->wsize,
991 write_size - total_written);
992 /* iov[0] is reserved for smb header */
993 iov[1].iov_base = (char *)write_data +
995 iov[1].iov_len = len;
996 rc = CIFSSMBWrite2(xid, pTcon,
997 open_file->netfid, len,
998 *poffset, &bytes_written,
1001 rc = CIFSSMBWrite(xid, pTcon,
1003 min_t(const int, cifs_sb->wsize,
1004 write_size - total_written),
1005 *poffset, &bytes_written,
1006 write_data + total_written,
1009 if (rc || (bytes_written == 0)) {
1017 *poffset += bytes_written;
1018 long_op = CIFS_STD_OP; /* subsequent writes fast -
1019 15 seconds is plenty */
1022 cifs_stats_bytes_written(pTcon, total_written);
1024 /* since the write may have blocked check these pointers again */
1025 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1026 /*BB We could make this contingent on superblock ATIME flag too */
1027 /* file->f_path.dentry->d_inode->i_ctime =
1028 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1029 if (total_written > 0) {
1030 spin_lock(&file->f_path.dentry->d_inode->i_lock);
1031 if (*poffset > file->f_path.dentry->d_inode->i_size)
1032 i_size_write(file->f_path.dentry->d_inode,
1034 spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1036 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1039 return total_written;
1042 #ifdef CONFIG_CIFS_EXPERIMENTAL
1043 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode)
1045 struct cifsFileInfo *open_file = NULL;
1047 read_lock(&GlobalSMBSeslock);
1048 /* we could simply get the first_list_entry since write-only entries
1049 are always at the end of the list but since the first entry might
1050 have a close pending, we go through the whole list */
1051 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1052 if (open_file->closePend)
1054 if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
1055 (open_file->pfile->f_flags & O_RDONLY))) {
1056 if (!open_file->invalidHandle) {
1057 /* found a good file */
1058 /* lock it so it will not be closed on us */
1059 atomic_inc(&open_file->wrtPending);
1060 read_unlock(&GlobalSMBSeslock);
1062 } /* else might as well continue, and look for
1063 another, or simply have the caller reopen it
1064 again rather than trying to fix this handle */
1065 } else /* write only file */
1066 break; /* write only files are last so must be done */
1068 read_unlock(&GlobalSMBSeslock);
1073 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1075 struct cifsFileInfo *open_file;
1076 bool any_available = false;
1079 /* Having a null inode here (because mapping->host was set to zero by
1080 the VFS or MM) should not happen but we had reports of on oops (due to
1081 it being zero) during stress testcases so we need to check for it */
1083 if (cifs_inode == NULL) {
1084 cERROR(1, ("Null inode passed to cifs_writeable_file"));
1089 read_lock(&GlobalSMBSeslock);
1091 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1092 if (open_file->closePend ||
1093 (!any_available && open_file->pid != current->tgid))
1096 if (open_file->pfile &&
1097 ((open_file->pfile->f_flags & O_RDWR) ||
1098 (open_file->pfile->f_flags & O_WRONLY))) {
1099 atomic_inc(&open_file->wrtPending);
1101 if (!open_file->invalidHandle) {
1102 /* found a good writable file */
1103 read_unlock(&GlobalSMBSeslock);
1107 read_unlock(&GlobalSMBSeslock);
1108 /* Had to unlock since following call can block */
1109 rc = cifs_reopen_file(open_file->pfile, false);
1111 if (!open_file->closePend)
1113 else { /* start over in case this was deleted */
1114 /* since the list could be modified */
1115 read_lock(&GlobalSMBSeslock);
1116 atomic_dec(&open_file->wrtPending);
1117 goto refind_writable;
1121 /* if it fails, try another handle if possible -
1122 (we can not do this if closePending since
1123 loop could be modified - in which case we
1124 have to start at the beginning of the list
1125 again. Note that it would be bad
1126 to hold up writepages here (rather than
1127 in caller) with continuous retries */
1128 cFYI(1, ("wp failed on reopen file"));
1129 read_lock(&GlobalSMBSeslock);
1130 /* can not use this handle, no write
1131 pending on this one after all */
1132 atomic_dec(&open_file->wrtPending);
1134 if (open_file->closePend) /* list could have changed */
1135 goto refind_writable;
1136 /* else we simply continue to the next entry. Thus
1137 we do not loop on reopen errors. If we
1138 can not reopen the file, for example if we
1139 reconnected to a server with another client
1140 racing to delete or lock the file we would not
1141 make progress if we restarted before the beginning
1142 of the loop here. */
1145 /* couldn't find useable FH with same pid, try any available */
1146 if (!any_available) {
1147 any_available = true;
1148 goto refind_writable;
1150 read_unlock(&GlobalSMBSeslock);
1154 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1156 struct address_space *mapping = page->mapping;
1157 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1160 int bytes_written = 0;
1161 struct cifs_sb_info *cifs_sb;
1162 struct cifsTconInfo *pTcon;
1163 struct inode *inode;
1164 struct cifsFileInfo *open_file;
1166 if (!mapping || !mapping->host)
1169 inode = page->mapping->host;
1170 cifs_sb = CIFS_SB(inode->i_sb);
1171 pTcon = cifs_sb->tcon;
1173 offset += (loff_t)from;
1174 write_data = kmap(page);
1177 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1182 /* racing with truncate? */
1183 if (offset > mapping->host->i_size) {
1185 return 0; /* don't care */
1188 /* check to make sure that we are not extending the file */
1189 if (mapping->host->i_size - offset < (loff_t)to)
1190 to = (unsigned)(mapping->host->i_size - offset);
1192 open_file = find_writable_file(CIFS_I(mapping->host));
1194 bytes_written = cifs_write(open_file->pfile, write_data,
1196 atomic_dec(&open_file->wrtPending);
1197 /* Does mm or vfs already set times? */
1198 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1199 if ((bytes_written > 0) && (offset))
1201 else if (bytes_written < 0)
1204 cFYI(1, ("No writeable filehandles for inode"));
1212 static int cifs_writepages(struct address_space *mapping,
1213 struct writeback_control *wbc)
1215 struct backing_dev_info *bdi = mapping->backing_dev_info;
1216 unsigned int bytes_to_write;
1217 unsigned int bytes_written;
1218 struct cifs_sb_info *cifs_sb;
1222 int range_whole = 0;
1229 struct cifsFileInfo *open_file;
1231 struct pagevec pvec;
1236 cifs_sb = CIFS_SB(mapping->host->i_sb);
1239 * If wsize is smaller that the page cache size, default to writing
1240 * one page at a time via cifs_writepage
1242 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1243 return generic_writepages(mapping, wbc);
1245 if ((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1246 if (cifs_sb->tcon->ses->server->secMode &
1247 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1248 if (!experimEnabled)
1249 return generic_writepages(mapping, wbc);
1251 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1253 return generic_writepages(mapping, wbc);
1257 * BB: Is this meaningful for a non-block-device file system?
1258 * If it is, we should test it again after we do I/O
1260 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1261 wbc->encountered_congestion = 1;
1268 pagevec_init(&pvec, 0);
1269 if (wbc->range_cyclic) {
1270 index = mapping->writeback_index; /* Start from prev offset */
1273 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1274 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1275 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1280 while (!done && (index <= end) &&
1281 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1282 PAGECACHE_TAG_DIRTY,
1283 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1292 for (i = 0; i < nr_pages; i++) {
1293 page = pvec.pages[i];
1295 * At this point we hold neither mapping->tree_lock nor
1296 * lock on the page itself: the page may be truncated or
1297 * invalidated (changing page->mapping to NULL), or even
1298 * swizzled back from swapper_space to tmpfs file
1304 else if (!trylock_page(page))
1307 if (unlikely(page->mapping != mapping)) {
1312 if (!wbc->range_cyclic && page->index > end) {
1318 if (next && (page->index != next)) {
1319 /* Not next consecutive page */
1324 if (wbc->sync_mode != WB_SYNC_NONE)
1325 wait_on_page_writeback(page);
1327 if (PageWriteback(page) ||
1328 !clear_page_dirty_for_io(page)) {
1334 * This actually clears the dirty bit in the radix tree.
1335 * See cifs_writepage() for more commentary.
1337 set_page_writeback(page);
1339 if (page_offset(page) >= mapping->host->i_size) {
1342 end_page_writeback(page);
1347 * BB can we get rid of this? pages are held by pvec
1349 page_cache_get(page);
1351 len = min(mapping->host->i_size - page_offset(page),
1352 (loff_t)PAGE_CACHE_SIZE);
1354 /* reserve iov[0] for the smb header */
1356 iov[n_iov].iov_base = kmap(page);
1357 iov[n_iov].iov_len = len;
1358 bytes_to_write += len;
1362 offset = page_offset(page);
1364 next = page->index + 1;
1365 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1369 /* Search for a writable handle every time we call
1370 * CIFSSMBWrite2. We can't rely on the last handle
1371 * we used to still be valid
1373 open_file = find_writable_file(CIFS_I(mapping->host));
1375 cERROR(1, ("No writable handles for inode"));
1378 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1380 bytes_to_write, offset,
1381 &bytes_written, iov, n_iov,
1383 atomic_dec(&open_file->wrtPending);
1384 if (rc || bytes_written < bytes_to_write) {
1385 cERROR(1, ("Write2 ret %d, wrote %d",
1386 rc, bytes_written));
1387 /* BB what if continued retry is
1388 requested via mount flags? */
1390 set_bit(AS_ENOSPC, &mapping->flags);
1392 set_bit(AS_EIO, &mapping->flags);
1394 cifs_stats_bytes_written(cifs_sb->tcon,
1398 for (i = 0; i < n_iov; i++) {
1399 page = pvec.pages[first + i];
1400 /* Should we also set page error on
1401 success rc but too little data written? */
1402 /* BB investigate retry logic on temporary
1403 server crash cases and how recovery works
1404 when page marked as error */
1409 end_page_writeback(page);
1410 page_cache_release(page);
1412 if ((wbc->nr_to_write -= n_iov) <= 0)
1416 /* Need to re-find the pages we skipped */
1417 index = pvec.pages[0]->index + 1;
1419 pagevec_release(&pvec);
1421 if (!scanned && !done) {
1423 * We hit the last page and there is more work to be done: wrap
1424 * back to the start of the file
1430 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1431 mapping->writeback_index = index;
1438 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1444 /* BB add check for wbc flags */
1445 page_cache_get(page);
1446 if (!PageUptodate(page))
1447 cFYI(1, ("ppw - page not up to date"));
1450 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1452 * A writepage() implementation always needs to do either this,
1453 * or re-dirty the page with "redirty_page_for_writepage()" in
1454 * the case of a failure.
1456 * Just unlocking the page will cause the radix tree tag-bits
1457 * to fail to update with the state of the page correctly.
1459 set_page_writeback(page);
1460 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1461 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1463 end_page_writeback(page);
1464 page_cache_release(page);
1469 static int cifs_write_end(struct file *file, struct address_space *mapping,
1470 loff_t pos, unsigned len, unsigned copied,
1471 struct page *page, void *fsdata)
1474 struct inode *inode = mapping->host;
1476 cFYI(1, ("write_end for page %p from pos %lld with %d bytes",
1477 page, pos, copied));
1479 if (PageChecked(page)) {
1481 SetPageUptodate(page);
1482 ClearPageChecked(page);
1483 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1484 SetPageUptodate(page);
1486 if (!PageUptodate(page)) {
1488 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1492 /* this is probably better than directly calling
1493 partialpage_write since in this function the file handle is
1494 known which we might as well leverage */
1495 /* BB check if anything else missing out of ppw
1496 such as updating last write time */
1497 page_data = kmap(page);
1498 rc = cifs_write(file, page_data + offset, copied, &pos);
1499 /* if (rc < 0) should we set writebehind rc? */
1506 set_page_dirty(page);
1510 spin_lock(&inode->i_lock);
1511 if (pos > inode->i_size)
1512 i_size_write(inode, pos);
1513 spin_unlock(&inode->i_lock);
1517 page_cache_release(page);
1522 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1526 struct cifsTconInfo *tcon;
1527 struct cifsFileInfo *smbfile =
1528 (struct cifsFileInfo *)file->private_data;
1529 struct inode *inode = file->f_path.dentry->d_inode;
1533 cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1534 dentry->d_name.name, datasync));
1536 rc = filemap_write_and_wait(inode->i_mapping);
1538 rc = CIFS_I(inode)->write_behind_rc;
1539 CIFS_I(inode)->write_behind_rc = 0;
1540 tcon = CIFS_SB(inode->i_sb)->tcon;
1541 if (!rc && tcon && smbfile)
1542 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1549 /* static void cifs_sync_page(struct page *page)
1551 struct address_space *mapping;
1552 struct inode *inode;
1553 unsigned long index = page->index;
1554 unsigned int rpages = 0;
1557 cFYI(1, ("sync page %p",page));
1558 mapping = page->mapping;
1561 inode = mapping->host;
1565 /* fill in rpages then
1566 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1568 /* cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1578 * As file closes, flush all cached write data for this inode checking
1579 * for write behind errors.
1581 int cifs_flush(struct file *file, fl_owner_t id)
1583 struct inode *inode = file->f_path.dentry->d_inode;
1586 /* Rather than do the steps manually:
1587 lock the inode for writing
1588 loop through pages looking for write behind data (dirty pages)
1589 coalesce into contiguous 16K (or smaller) chunks to write to server
1590 send to server (prefer in parallel)
1591 deal with writebehind errors
1592 unlock inode for writing
1593 filemapfdatawrite appears easier for the time being */
1595 rc = filemap_fdatawrite(inode->i_mapping);
1596 /* reset wb rc if we were able to write out dirty pages */
1598 rc = CIFS_I(inode)->write_behind_rc;
1599 CIFS_I(inode)->write_behind_rc = 0;
1602 cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
1607 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1608 size_t read_size, loff_t *poffset)
1611 unsigned int bytes_read = 0;
1612 unsigned int total_read = 0;
1613 unsigned int current_read_size;
1614 struct cifs_sb_info *cifs_sb;
1615 struct cifsTconInfo *pTcon;
1617 struct cifsFileInfo *open_file;
1618 char *smb_read_data;
1619 char __user *current_offset;
1620 struct smb_com_read_rsp *pSMBr;
1623 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1624 pTcon = cifs_sb->tcon;
1626 if (file->private_data == NULL) {
1630 open_file = (struct cifsFileInfo *)file->private_data;
1632 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1633 cFYI(1, ("attempting read on write only file instance"));
1635 for (total_read = 0, current_offset = read_data;
1636 read_size > total_read;
1637 total_read += bytes_read, current_offset += bytes_read) {
1638 current_read_size = min_t(const int, read_size - total_read,
1641 smb_read_data = NULL;
1642 while (rc == -EAGAIN) {
1643 int buf_type = CIFS_NO_BUFFER;
1644 if ((open_file->invalidHandle) &&
1645 (!open_file->closePend)) {
1646 rc = cifs_reopen_file(file, true);
1650 rc = CIFSSMBRead(xid, pTcon,
1652 current_read_size, *poffset,
1653 &bytes_read, &smb_read_data,
1655 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1656 if (smb_read_data) {
1657 if (copy_to_user(current_offset,
1659 4 /* RFC1001 length field */ +
1660 le16_to_cpu(pSMBr->DataOffset),
1664 if (buf_type == CIFS_SMALL_BUFFER)
1665 cifs_small_buf_release(smb_read_data);
1666 else if (buf_type == CIFS_LARGE_BUFFER)
1667 cifs_buf_release(smb_read_data);
1668 smb_read_data = NULL;
1671 if (rc || (bytes_read == 0)) {
1679 cifs_stats_bytes_read(pTcon, bytes_read);
1680 *poffset += bytes_read;
1688 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1692 unsigned int bytes_read = 0;
1693 unsigned int total_read;
1694 unsigned int current_read_size;
1695 struct cifs_sb_info *cifs_sb;
1696 struct cifsTconInfo *pTcon;
1698 char *current_offset;
1699 struct cifsFileInfo *open_file;
1700 int buf_type = CIFS_NO_BUFFER;
1703 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1704 pTcon = cifs_sb->tcon;
1706 if (file->private_data == NULL) {
1710 open_file = (struct cifsFileInfo *)file->private_data;
1712 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1713 cFYI(1, ("attempting read on write only file instance"));
1715 for (total_read = 0, current_offset = read_data;
1716 read_size > total_read;
1717 total_read += bytes_read, current_offset += bytes_read) {
1718 current_read_size = min_t(const int, read_size - total_read,
1720 /* For windows me and 9x we do not want to request more
1721 than it negotiated since it will refuse the read then */
1723 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1724 current_read_size = min_t(const int, current_read_size,
1725 pTcon->ses->server->maxBuf - 128);
1728 while (rc == -EAGAIN) {
1729 if ((open_file->invalidHandle) &&
1730 (!open_file->closePend)) {
1731 rc = cifs_reopen_file(file, true);
1735 rc = CIFSSMBRead(xid, pTcon,
1737 current_read_size, *poffset,
1738 &bytes_read, ¤t_offset,
1741 if (rc || (bytes_read == 0)) {
1749 cifs_stats_bytes_read(pTcon, total_read);
1750 *poffset += bytes_read;
1757 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1759 struct dentry *dentry = file->f_path.dentry;
1763 rc = cifs_revalidate(dentry);
1765 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1769 rc = generic_file_mmap(file, vma);
1775 static void cifs_copy_cache_pages(struct address_space *mapping,
1776 struct list_head *pages, int bytes_read, char *data,
1777 struct pagevec *plru_pvec)
1782 while (bytes_read > 0) {
1783 if (list_empty(pages))
1786 page = list_entry(pages->prev, struct page, lru);
1787 list_del(&page->lru);
1789 if (add_to_page_cache(page, mapping, page->index,
1791 page_cache_release(page);
1792 cFYI(1, ("Add page cache failed"));
1793 data += PAGE_CACHE_SIZE;
1794 bytes_read -= PAGE_CACHE_SIZE;
1798 target = kmap_atomic(page, KM_USER0);
1800 if (PAGE_CACHE_SIZE > bytes_read) {
1801 memcpy(target, data, bytes_read);
1802 /* zero the tail end of this partial page */
1803 memset(target + bytes_read, 0,
1804 PAGE_CACHE_SIZE - bytes_read);
1807 memcpy(target, data, PAGE_CACHE_SIZE);
1808 bytes_read -= PAGE_CACHE_SIZE;
1810 kunmap_atomic(target, KM_USER0);
1812 flush_dcache_page(page);
1813 SetPageUptodate(page);
1815 if (!pagevec_add(plru_pvec, page))
1816 __pagevec_lru_add_file(plru_pvec);
1817 data += PAGE_CACHE_SIZE;
1822 static int cifs_readpages(struct file *file, struct address_space *mapping,
1823 struct list_head *page_list, unsigned num_pages)
1829 struct cifs_sb_info *cifs_sb;
1830 struct cifsTconInfo *pTcon;
1831 unsigned int bytes_read = 0;
1832 unsigned int read_size, i;
1833 char *smb_read_data = NULL;
1834 struct smb_com_read_rsp *pSMBr;
1835 struct pagevec lru_pvec;
1836 struct cifsFileInfo *open_file;
1837 int buf_type = CIFS_NO_BUFFER;
1840 if (file->private_data == NULL) {
1844 open_file = (struct cifsFileInfo *)file->private_data;
1845 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1846 pTcon = cifs_sb->tcon;
1848 pagevec_init(&lru_pvec, 0);
1849 cFYI(DBG2, ("rpages: num pages %d", num_pages));
1850 for (i = 0; i < num_pages; ) {
1851 unsigned contig_pages;
1852 struct page *tmp_page;
1853 unsigned long expected_index;
1855 if (list_empty(page_list))
1858 page = list_entry(page_list->prev, struct page, lru);
1859 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1861 /* count adjacent pages that we will read into */
1864 list_entry(page_list->prev, struct page, lru)->index;
1865 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1866 if (tmp_page->index == expected_index) {
1872 if (contig_pages + i > num_pages)
1873 contig_pages = num_pages - i;
1875 /* for reads over a certain size could initiate async
1878 read_size = contig_pages * PAGE_CACHE_SIZE;
1879 /* Read size needs to be in multiples of one page */
1880 read_size = min_t(const unsigned int, read_size,
1881 cifs_sb->rsize & PAGE_CACHE_MASK);
1882 cFYI(DBG2, ("rpages: read size 0x%x contiguous pages %d",
1883 read_size, contig_pages));
1885 while (rc == -EAGAIN) {
1886 if ((open_file->invalidHandle) &&
1887 (!open_file->closePend)) {
1888 rc = cifs_reopen_file(file, true);
1893 rc = CIFSSMBRead(xid, pTcon,
1896 &bytes_read, &smb_read_data,
1898 /* BB more RC checks ? */
1899 if (rc == -EAGAIN) {
1900 if (smb_read_data) {
1901 if (buf_type == CIFS_SMALL_BUFFER)
1902 cifs_small_buf_release(smb_read_data);
1903 else if (buf_type == CIFS_LARGE_BUFFER)
1904 cifs_buf_release(smb_read_data);
1905 smb_read_data = NULL;
1909 if ((rc < 0) || (smb_read_data == NULL)) {
1910 cFYI(1, ("Read error in readpages: %d", rc));
1912 } else if (bytes_read > 0) {
1913 task_io_account_read(bytes_read);
1914 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1915 cifs_copy_cache_pages(mapping, page_list, bytes_read,
1916 smb_read_data + 4 /* RFC1001 hdr */ +
1917 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
1919 i += bytes_read >> PAGE_CACHE_SHIFT;
1920 cifs_stats_bytes_read(pTcon, bytes_read);
1921 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1922 i++; /* account for partial page */
1924 /* server copy of file can have smaller size
1926 /* BB do we need to verify this common case ?
1927 this case is ok - if we are at server EOF
1928 we will hit it on next read */
1933 cFYI(1, ("No bytes read (%d) at offset %lld . "
1934 "Cleaning remaining pages from readahead list",
1935 bytes_read, offset));
1936 /* BB turn off caching and do new lookup on
1937 file size at server? */
1940 if (smb_read_data) {
1941 if (buf_type == CIFS_SMALL_BUFFER)
1942 cifs_small_buf_release(smb_read_data);
1943 else if (buf_type == CIFS_LARGE_BUFFER)
1944 cifs_buf_release(smb_read_data);
1945 smb_read_data = NULL;
1950 pagevec_lru_add_file(&lru_pvec);
1952 /* need to free smb_read_data buf before exit */
1953 if (smb_read_data) {
1954 if (buf_type == CIFS_SMALL_BUFFER)
1955 cifs_small_buf_release(smb_read_data);
1956 else if (buf_type == CIFS_LARGE_BUFFER)
1957 cifs_buf_release(smb_read_data);
1958 smb_read_data = NULL;
1965 static int cifs_readpage_worker(struct file *file, struct page *page,
1971 page_cache_get(page);
1972 read_data = kmap(page);
1973 /* for reads over a certain size could initiate async read ahead */
1975 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1980 cFYI(1, ("Bytes read %d", rc));
1982 file->f_path.dentry->d_inode->i_atime =
1983 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1985 if (PAGE_CACHE_SIZE > rc)
1986 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1988 flush_dcache_page(page);
1989 SetPageUptodate(page);
1994 page_cache_release(page);
1998 static int cifs_readpage(struct file *file, struct page *page)
2000 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2006 if (file->private_data == NULL) {
2011 cFYI(1, ("readpage %p at offset %d 0x%x\n",
2012 page, (int)offset, (int)offset));
2014 rc = cifs_readpage_worker(file, page, &offset);
2022 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2024 struct cifsFileInfo *open_file;
2026 read_lock(&GlobalSMBSeslock);
2027 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2028 if (open_file->closePend)
2030 if (open_file->pfile &&
2031 ((open_file->pfile->f_flags & O_RDWR) ||
2032 (open_file->pfile->f_flags & O_WRONLY))) {
2033 read_unlock(&GlobalSMBSeslock);
2037 read_unlock(&GlobalSMBSeslock);
2041 /* We do not want to update the file size from server for inodes
2042 open for write - to avoid races with writepage extending
2043 the file - in the future we could consider allowing
2044 refreshing the inode only on increases in the file size
2045 but this is tricky to do without racing with writebehind
2046 page caching in the current Linux kernel design */
2047 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2052 if (is_inode_writable(cifsInode)) {
2053 /* This inode is open for write at least once */
2054 struct cifs_sb_info *cifs_sb;
2056 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2057 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2058 /* since no page cache to corrupt on directio
2059 we can change size safely */
2063 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2071 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2072 loff_t pos, unsigned len, unsigned flags,
2073 struct page **pagep, void **fsdata)
2075 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2076 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2077 loff_t page_start = pos & PAGE_MASK;
2082 cFYI(1, ("write_begin from %lld len %d", (long long)pos, len));
2084 page = grab_cache_page_write_begin(mapping, index, flags);
2090 if (PageUptodate(page))
2094 * If we write a full page it will be up to date, no need to read from
2095 * the server. If the write is short, we'll end up doing a sync write
2098 if (len == PAGE_CACHE_SIZE)
2102 * optimize away the read when we have an oplock, and we're not
2103 * expecting to use any of the data we'd be reading in. That
2104 * is, when the page lies beyond the EOF, or straddles the EOF
2105 * and the write will cover all of the existing data.
2107 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2108 i_size = i_size_read(mapping->host);
2109 if (page_start >= i_size ||
2110 (offset == 0 && (pos + len) >= i_size)) {
2111 zero_user_segments(page, 0, offset,
2115 * PageChecked means that the parts of the page
2116 * to which we're not writing are considered up
2117 * to date. Once the data is copied to the
2118 * page, it can be set uptodate.
2120 SetPageChecked(page);
2125 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2127 * might as well read a page, it is fast enough. If we get
2128 * an error, we don't need to return it. cifs_write_end will
2129 * do a sync write instead since PG_uptodate isn't set.
2131 cifs_readpage_worker(file, page, &page_start);
2133 /* we could try using another file handle if there is one -
2134 but how would we lock it to prevent close of that handle
2135 racing with this read? In any case
2136 this will be written out by write_end so is fine */
2143 const struct address_space_operations cifs_addr_ops = {
2144 .readpage = cifs_readpage,
2145 .readpages = cifs_readpages,
2146 .writepage = cifs_writepage,
2147 .writepages = cifs_writepages,
2148 .write_begin = cifs_write_begin,
2149 .write_end = cifs_write_end,
2150 .set_page_dirty = __set_page_dirty_nobuffers,
2151 /* .sync_page = cifs_sync_page, */
2156 * cifs_readpages requires the server to support a buffer large enough to
2157 * contain the header plus one complete page of data. Otherwise, we need
2158 * to leave cifs_readpages out of the address space operations.
2160 const struct address_space_operations cifs_addr_ops_smallbuf = {
2161 .readpage = cifs_readpage,
2162 .writepage = cifs_writepage,
2163 .writepages = cifs_writepages,
2164 .write_begin = cifs_write_begin,
2165 .write_end = cifs_write_end,
2166 .set_page_dirty = __set_page_dirty_nobuffers,
2167 /* .sync_page = cifs_sync_page, */