4 * vfs operations that deal with files
6 * Copyright (C) International Business Machines Corp., 2002,2010
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 <linux/mount.h>
34 #include <linux/slab.h>
35 #include <asm/div64.h>
39 #include "cifsproto.h"
40 #include "cifs_unicode.h"
41 #include "cifs_debug.h"
42 #include "cifs_fs_sb.h"
45 static inline int cifs_convert_flags(unsigned int flags
)
47 if ((flags
& O_ACCMODE
) == O_RDONLY
)
49 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
51 else if ((flags
& O_ACCMODE
) == O_RDWR
) {
52 /* GENERIC_ALL is too much permission to request
53 can cause unnecessary access denied on create */
54 /* return GENERIC_ALL; */
55 return (GENERIC_READ
| GENERIC_WRITE
);
58 return (READ_CONTROL
| FILE_WRITE_ATTRIBUTES
| FILE_READ_ATTRIBUTES
|
59 FILE_WRITE_EA
| FILE_APPEND_DATA
| FILE_WRITE_DATA
|
63 static u32
cifs_posix_convert_flags(unsigned int flags
)
67 if ((flags
& O_ACCMODE
) == O_RDONLY
)
68 posix_flags
= SMB_O_RDONLY
;
69 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
70 posix_flags
= SMB_O_WRONLY
;
71 else if ((flags
& O_ACCMODE
) == O_RDWR
)
72 posix_flags
= SMB_O_RDWR
;
75 posix_flags
|= SMB_O_CREAT
;
77 posix_flags
|= SMB_O_EXCL
;
79 posix_flags
|= SMB_O_TRUNC
;
80 /* be safe and imply O_SYNC for O_DSYNC */
82 posix_flags
|= SMB_O_SYNC
;
83 if (flags
& O_DIRECTORY
)
84 posix_flags
|= SMB_O_DIRECTORY
;
85 if (flags
& O_NOFOLLOW
)
86 posix_flags
|= SMB_O_NOFOLLOW
;
88 posix_flags
|= SMB_O_DIRECT
;
93 static inline int cifs_get_disposition(unsigned int flags
)
95 if ((flags
& (O_CREAT
| O_EXCL
)) == (O_CREAT
| O_EXCL
))
97 else if ((flags
& (O_CREAT
| O_TRUNC
)) == (O_CREAT
| O_TRUNC
))
98 return FILE_OVERWRITE_IF
;
99 else if ((flags
& O_CREAT
) == O_CREAT
)
101 else if ((flags
& O_TRUNC
) == O_TRUNC
)
102 return FILE_OVERWRITE
;
107 static inline int cifs_open_inode_helper(struct inode
*inode
,
108 struct cifsTconInfo
*pTcon
, __u32 oplock
, FILE_ALL_INFO
*buf
,
109 char *full_path
, int xid
)
111 struct cifsInodeInfo
*pCifsInode
= CIFS_I(inode
);
112 struct timespec temp
;
115 if (pCifsInode
->clientCanCacheRead
) {
116 /* we have the inode open somewhere else
117 no need to discard cache data */
118 goto client_can_cache
;
121 /* BB need same check in cifs_create too? */
122 /* if not oplocked, invalidate inode pages if mtime or file
124 temp
= cifs_NTtimeToUnix(buf
->LastWriteTime
);
125 if (timespec_equal(&inode
->i_mtime
, &temp
) &&
127 (loff_t
)le64_to_cpu(buf
->EndOfFile
))) {
128 cFYI(1, "inode unchanged on server");
130 if (inode
->i_mapping
) {
131 /* BB no need to lock inode until after invalidate
132 since namei code should already have it locked? */
133 rc
= filemap_write_and_wait(inode
->i_mapping
);
135 pCifsInode
->write_behind_rc
= rc
;
137 cFYI(1, "invalidating remote inode since open detected it "
139 invalidate_remote_inode(inode
);
144 rc
= cifs_get_inode_info_unix(&inode
, full_path
, inode
->i_sb
,
147 rc
= cifs_get_inode_info(&inode
, full_path
, buf
, inode
->i_sb
,
150 if ((oplock
& 0xF) == OPLOCK_EXCLUSIVE
) {
151 pCifsInode
->clientCanCacheAll
= true;
152 pCifsInode
->clientCanCacheRead
= true;
153 cFYI(1, "Exclusive Oplock granted on inode %p", inode
);
154 } else if ((oplock
& 0xF) == OPLOCK_READ
)
155 pCifsInode
->clientCanCacheRead
= true;
160 int cifs_posix_open(char *full_path
, struct inode
**pinode
,
161 struct super_block
*sb
, int mode
, unsigned int f_flags
,
162 __u32
*poplock
, __u16
*pnetfid
, int xid
)
165 FILE_UNIX_BASIC_INFO
*presp_data
;
166 __u32 posix_flags
= 0;
167 struct cifs_sb_info
*cifs_sb
= CIFS_SB(sb
);
168 struct cifs_fattr fattr
;
169 struct tcon_link
*tlink
;
170 struct cifsTconInfo
*tcon
;
172 cFYI(1, "posix open %s", full_path
);
174 presp_data
= kzalloc(sizeof(FILE_UNIX_BASIC_INFO
), GFP_KERNEL
);
175 if (presp_data
== NULL
)
178 tlink
= cifs_sb_tlink(cifs_sb
);
184 tcon
= tlink_tcon(tlink
);
185 mode
&= ~current_umask();
187 posix_flags
= cifs_posix_convert_flags(f_flags
);
188 rc
= CIFSPOSIXCreate(xid
, tcon
, posix_flags
, mode
, pnetfid
, presp_data
,
189 poplock
, full_path
, cifs_sb
->local_nls
,
190 cifs_sb
->mnt_cifs_flags
&
191 CIFS_MOUNT_MAP_SPECIAL_CHR
);
192 cifs_put_tlink(tlink
);
197 if (presp_data
->Type
== cpu_to_le32(-1))
198 goto posix_open_ret
; /* open ok, caller does qpathinfo */
201 goto posix_open_ret
; /* caller does not need info */
203 cifs_unix_basic_to_fattr(&fattr
, presp_data
, cifs_sb
);
205 /* get new inode and set it up */
206 if (*pinode
== NULL
) {
207 cifs_fill_uniqueid(sb
, &fattr
);
208 *pinode
= cifs_iget(sb
, &fattr
);
214 cifs_fattr_to_inode(*pinode
, &fattr
);
222 int cifs_open(struct inode
*inode
, struct file
*file
)
227 struct cifs_sb_info
*cifs_sb
;
228 struct cifsTconInfo
*tcon
;
229 struct tcon_link
*tlink
;
230 struct cifsFileInfo
*pCifsFile
= NULL
;
231 struct cifsInodeInfo
*pCifsInode
;
232 char *full_path
= NULL
;
236 FILE_ALL_INFO
*buf
= NULL
;
240 cifs_sb
= CIFS_SB(inode
->i_sb
);
241 tlink
= cifs_sb_tlink(cifs_sb
);
244 return PTR_ERR(tlink
);
246 tcon
= tlink_tcon(tlink
);
248 pCifsInode
= CIFS_I(file
->f_path
.dentry
->d_inode
);
250 full_path
= build_path_from_dentry(file
->f_path
.dentry
);
251 if (full_path
== NULL
) {
256 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
257 inode
, file
->f_flags
, full_path
);
264 if (!tcon
->broken_posix_open
&& tcon
->unix_ext
&&
265 (tcon
->ses
->capabilities
& CAP_UNIX
) &&
266 (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
267 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
268 /* can not refresh inode info since size could be stale */
269 rc
= cifs_posix_open(full_path
, &inode
, inode
->i_sb
,
270 cifs_sb
->mnt_file_mode
/* ignored */,
271 file
->f_flags
, &oplock
, &netfid
, xid
);
273 cFYI(1, "posix open succeeded");
275 pCifsFile
= cifs_new_fileinfo(netfid
, file
, tlink
,
277 if (pCifsFile
== NULL
) {
278 CIFSSMBClose(xid
, tcon
, netfid
);
282 cifs_fscache_set_inode_cookie(inode
, file
);
285 } else if ((rc
== -EINVAL
) || (rc
== -EOPNOTSUPP
)) {
286 if (tcon
->ses
->serverNOS
)
287 cERROR(1, "server %s of type %s returned"
288 " unexpected error on SMB posix open"
289 ", disabling posix open support."
290 " Check if server update available.",
291 tcon
->ses
->serverName
,
292 tcon
->ses
->serverNOS
);
293 tcon
->broken_posix_open
= true;
294 } else if ((rc
!= -EIO
) && (rc
!= -EREMOTE
) &&
295 (rc
!= -EOPNOTSUPP
)) /* path not found or net err */
297 /* else fallthrough to retry open the old way on network i/o
301 desiredAccess
= cifs_convert_flags(file
->f_flags
);
303 /*********************************************************************
304 * open flag mapping table:
306 * POSIX Flag CIFS Disposition
307 * ---------- ----------------
308 * O_CREAT FILE_OPEN_IF
309 * O_CREAT | O_EXCL FILE_CREATE
310 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
311 * O_TRUNC FILE_OVERWRITE
312 * none of the above FILE_OPEN
314 * Note that there is not a direct match between disposition
315 * FILE_SUPERSEDE (ie create whether or not file exists although
316 * O_CREAT | O_TRUNC is similar but truncates the existing
317 * file rather than creating a new file as FILE_SUPERSEDE does
318 * (which uses the attributes / metadata passed in on open call)
320 *? O_SYNC is a reasonable match to CIFS writethrough flag
321 *? and the read write flags match reasonably. O_LARGEFILE
322 *? is irrelevant because largefile support is always used
323 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
324 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
325 *********************************************************************/
327 disposition
= cifs_get_disposition(file
->f_flags
);
329 /* BB pass O_SYNC flag through on file attributes .. BB */
331 /* Also refresh inode by passing in file_info buf returned by SMBOpen
332 and calling get_inode_info with returned buf (at least helps
333 non-Unix server case) */
335 /* BB we can not do this if this is the second open of a file
336 and the first handle has writebehind data, we might be
337 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
338 buf
= kmalloc(sizeof(FILE_ALL_INFO
), GFP_KERNEL
);
344 if (tcon
->ses
->capabilities
& CAP_NT_SMBS
)
345 rc
= CIFSSMBOpen(xid
, tcon
, full_path
, disposition
,
346 desiredAccess
, CREATE_NOT_DIR
, &netfid
, &oplock
, buf
,
347 cifs_sb
->local_nls
, cifs_sb
->mnt_cifs_flags
348 & CIFS_MOUNT_MAP_SPECIAL_CHR
);
350 rc
= -EIO
; /* no NT SMB support fall into legacy open below */
353 /* Old server, try legacy style OpenX */
354 rc
= SMBLegacyOpen(xid
, tcon
, full_path
, disposition
,
355 desiredAccess
, CREATE_NOT_DIR
, &netfid
, &oplock
, buf
,
356 cifs_sb
->local_nls
, cifs_sb
->mnt_cifs_flags
357 & CIFS_MOUNT_MAP_SPECIAL_CHR
);
360 cFYI(1, "cifs_open returned 0x%x", rc
);
364 rc
= cifs_open_inode_helper(inode
, tcon
, oplock
, buf
, full_path
, xid
);
368 pCifsFile
= cifs_new_fileinfo(netfid
, file
, tlink
, oplock
);
369 if (pCifsFile
== NULL
) {
374 cifs_fscache_set_inode_cookie(inode
, file
);
376 if (oplock
& CIFS_CREATE_ACTION
) {
377 /* time to set mode which we can not set earlier due to
378 problems creating new read-only files */
379 if (tcon
->unix_ext
) {
380 struct cifs_unix_set_info_args args
= {
381 .mode
= inode
->i_mode
,
384 .ctime
= NO_CHANGE_64
,
385 .atime
= NO_CHANGE_64
,
386 .mtime
= NO_CHANGE_64
,
389 CIFSSMBUnixSetPathInfo(xid
, tcon
, full_path
, &args
,
391 cifs_sb
->mnt_cifs_flags
&
392 CIFS_MOUNT_MAP_SPECIAL_CHR
);
400 cifs_put_tlink(tlink
);
404 /* Try to reacquire byte range locks that were released when session */
405 /* to server was lost */
406 static int cifs_relock_file(struct cifsFileInfo
*cifsFile
)
410 /* BB list all locks open on this file and relock */
415 static int cifs_reopen_file(struct cifsFileInfo
*pCifsFile
, bool can_flush
)
420 struct cifs_sb_info
*cifs_sb
;
421 struct cifsTconInfo
*tcon
;
422 struct cifsInodeInfo
*pCifsInode
;
424 char *full_path
= NULL
;
426 int disposition
= FILE_OPEN
;
430 mutex_lock(&pCifsFile
->fh_mutex
);
431 if (!pCifsFile
->invalidHandle
) {
432 mutex_unlock(&pCifsFile
->fh_mutex
);
438 inode
= pCifsFile
->dentry
->d_inode
;
439 cifs_sb
= CIFS_SB(inode
->i_sb
);
440 tcon
= tlink_tcon(pCifsFile
->tlink
);
442 /* can not grab rename sem here because various ops, including
443 those that already have the rename sem can end up causing writepage
444 to get called and if the server was down that means we end up here,
445 and we can never tell if the caller already has the rename_sem */
446 full_path
= build_path_from_dentry(pCifsFile
->dentry
);
447 if (full_path
== NULL
) {
449 mutex_unlock(&pCifsFile
->fh_mutex
);
454 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
455 inode
, pCifsFile
->f_flags
, full_path
);
462 if (tcon
->unix_ext
&& (tcon
->ses
->capabilities
& CAP_UNIX
) &&
463 (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
464 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
467 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
468 * original open. Must mask them off for a reopen.
470 unsigned int oflags
= pCifsFile
->f_flags
&
471 ~(O_CREAT
| O_EXCL
| O_TRUNC
);
473 rc
= cifs_posix_open(full_path
, NULL
, inode
->i_sb
,
474 cifs_sb
->mnt_file_mode
/* ignored */,
475 oflags
, &oplock
, &netfid
, xid
);
477 cFYI(1, "posix reopen succeeded");
480 /* fallthrough to retry open the old way on errors, especially
481 in the reconnect path it is important to retry hard */
484 desiredAccess
= cifs_convert_flags(pCifsFile
->f_flags
);
486 /* Can not refresh inode by passing in file_info buf to be returned
487 by SMBOpen and then calling get_inode_info with returned buf
488 since file might have write behind data that needs to be flushed
489 and server version of file size can be stale. If we knew for sure
490 that inode was not dirty locally we could do this */
492 rc
= CIFSSMBOpen(xid
, tcon
, full_path
, disposition
, desiredAccess
,
493 CREATE_NOT_DIR
, &netfid
, &oplock
, NULL
,
494 cifs_sb
->local_nls
, cifs_sb
->mnt_cifs_flags
&
495 CIFS_MOUNT_MAP_SPECIAL_CHR
);
497 mutex_unlock(&pCifsFile
->fh_mutex
);
498 cFYI(1, "cifs_open returned 0x%x", rc
);
499 cFYI(1, "oplock: %d", oplock
);
500 goto reopen_error_exit
;
504 pCifsFile
->netfid
= netfid
;
505 pCifsFile
->invalidHandle
= false;
506 mutex_unlock(&pCifsFile
->fh_mutex
);
507 pCifsInode
= CIFS_I(inode
);
510 rc
= filemap_write_and_wait(inode
->i_mapping
);
512 CIFS_I(inode
)->write_behind_rc
= rc
;
514 pCifsInode
->clientCanCacheAll
= false;
515 pCifsInode
->clientCanCacheRead
= false;
517 rc
= cifs_get_inode_info_unix(&inode
,
518 full_path
, inode
->i_sb
, xid
);
520 rc
= cifs_get_inode_info(&inode
,
521 full_path
, NULL
, inode
->i_sb
,
523 } /* else we are writing out data to server already
524 and could deadlock if we tried to flush data, and
525 since we do not know if we have data that would
526 invalidate the current end of file on the server
527 we can not go to the server to get the new inod
529 if ((oplock
& 0xF) == OPLOCK_EXCLUSIVE
) {
530 pCifsInode
->clientCanCacheAll
= true;
531 pCifsInode
->clientCanCacheRead
= true;
532 cFYI(1, "Exclusive Oplock granted on inode %p",
533 pCifsFile
->dentry
->d_inode
);
534 } else if ((oplock
& 0xF) == OPLOCK_READ
) {
535 pCifsInode
->clientCanCacheRead
= true;
536 pCifsInode
->clientCanCacheAll
= false;
538 pCifsInode
->clientCanCacheRead
= false;
539 pCifsInode
->clientCanCacheAll
= false;
541 cifs_relock_file(pCifsFile
);
549 int cifs_close(struct inode
*inode
, struct file
*file
)
553 struct cifs_sb_info
*cifs_sb
;
554 struct cifsTconInfo
*pTcon
;
555 struct cifsFileInfo
*pSMBFile
= file
->private_data
;
559 cifs_sb
= CIFS_SB(inode
->i_sb
);
560 pTcon
= tlink_tcon(pSMBFile
->tlink
);
562 struct cifsLockInfo
*li
, *tmp
;
563 write_lock(&GlobalSMBSeslock
);
564 pSMBFile
->closePend
= true;
566 /* no sense reconnecting to close a file that is
568 if (!pTcon
->need_reconnect
) {
569 write_unlock(&GlobalSMBSeslock
);
571 while ((atomic_read(&pSMBFile
->count
) != 1)
572 && (timeout
<= 2048)) {
573 /* Give write a better chance to get to
574 server ahead of the close. We do not
575 want to add a wait_q here as it would
576 increase the memory utilization as
577 the struct would be in each open file,
578 but this should give enough time to
580 cFYI(DBG2
, "close delay, write pending");
584 if (!pTcon
->need_reconnect
&&
585 !pSMBFile
->invalidHandle
)
586 rc
= CIFSSMBClose(xid
, pTcon
,
589 write_unlock(&GlobalSMBSeslock
);
591 write_unlock(&GlobalSMBSeslock
);
593 /* Delete any outstanding lock records.
594 We'll lose them when the file is closed anyway. */
595 mutex_lock(&pSMBFile
->lock_mutex
);
596 list_for_each_entry_safe(li
, tmp
, &pSMBFile
->llist
, llist
) {
597 list_del(&li
->llist
);
600 mutex_unlock(&pSMBFile
->lock_mutex
);
602 write_lock(&GlobalSMBSeslock
);
603 list_del(&pSMBFile
->flist
);
604 list_del(&pSMBFile
->tlist
);
605 write_unlock(&GlobalSMBSeslock
);
606 cifsFileInfo_put(file
->private_data
);
607 file
->private_data
= NULL
;
611 read_lock(&GlobalSMBSeslock
);
612 if (list_empty(&(CIFS_I(inode
)->openFileList
))) {
613 cFYI(1, "closing last open instance for inode %p", inode
);
614 /* if the file is not open we do not know if we can cache info
615 on this inode, much less write behind and read ahead */
616 CIFS_I(inode
)->clientCanCacheRead
= false;
617 CIFS_I(inode
)->clientCanCacheAll
= false;
619 read_unlock(&GlobalSMBSeslock
);
620 if ((rc
== 0) && CIFS_I(inode
)->write_behind_rc
)
621 rc
= CIFS_I(inode
)->write_behind_rc
;
626 int cifs_closedir(struct inode
*inode
, struct file
*file
)
630 struct cifsFileInfo
*pCFileStruct
= file
->private_data
;
633 cFYI(1, "Closedir inode = 0x%p", inode
);
638 struct cifsTconInfo
*pTcon
= tlink_tcon(pCFileStruct
->tlink
);
640 cFYI(1, "Freeing private data in close dir");
641 write_lock(&GlobalSMBSeslock
);
642 if (!pCFileStruct
->srch_inf
.endOfSearch
&&
643 !pCFileStruct
->invalidHandle
) {
644 pCFileStruct
->invalidHandle
= true;
645 write_unlock(&GlobalSMBSeslock
);
646 rc
= CIFSFindClose(xid
, pTcon
, pCFileStruct
->netfid
);
647 cFYI(1, "Closing uncompleted readdir with rc %d",
649 /* not much we can do if it fails anyway, ignore rc */
652 write_unlock(&GlobalSMBSeslock
);
653 ptmp
= pCFileStruct
->srch_inf
.ntwrk_buf_start
;
655 cFYI(1, "closedir free smb buf in srch struct");
656 pCFileStruct
->srch_inf
.ntwrk_buf_start
= NULL
;
657 if (pCFileStruct
->srch_inf
.smallBuf
)
658 cifs_small_buf_release(ptmp
);
660 cifs_buf_release(ptmp
);
662 cifs_put_tlink(pCFileStruct
->tlink
);
663 kfree(file
->private_data
);
664 file
->private_data
= NULL
;
666 /* BB can we lock the filestruct while this is going on? */
671 static int store_file_lock(struct cifsFileInfo
*fid
, __u64 len
,
672 __u64 offset
, __u8 lockType
)
674 struct cifsLockInfo
*li
=
675 kmalloc(sizeof(struct cifsLockInfo
), GFP_KERNEL
);
681 mutex_lock(&fid
->lock_mutex
);
682 list_add(&li
->llist
, &fid
->llist
);
683 mutex_unlock(&fid
->lock_mutex
);
687 int cifs_lock(struct file
*file
, int cmd
, struct file_lock
*pfLock
)
693 bool wait_flag
= false;
694 struct cifs_sb_info
*cifs_sb
;
695 struct cifsTconInfo
*tcon
;
697 __u8 lockType
= LOCKING_ANDX_LARGE_FILES
;
698 bool posix_locking
= 0;
700 length
= 1 + pfLock
->fl_end
- pfLock
->fl_start
;
704 cFYI(1, "Lock parm: 0x%x flockflags: "
705 "0x%x flocktype: 0x%x start: %lld end: %lld",
706 cmd
, pfLock
->fl_flags
, pfLock
->fl_type
, pfLock
->fl_start
,
709 if (pfLock
->fl_flags
& FL_POSIX
)
711 if (pfLock
->fl_flags
& FL_FLOCK
)
713 if (pfLock
->fl_flags
& FL_SLEEP
) {
714 cFYI(1, "Blocking lock");
717 if (pfLock
->fl_flags
& FL_ACCESS
)
718 cFYI(1, "Process suspended by mandatory locking - "
719 "not implemented yet");
720 if (pfLock
->fl_flags
& FL_LEASE
)
721 cFYI(1, "Lease on file - not implemented yet");
722 if (pfLock
->fl_flags
&
723 (~(FL_POSIX
| FL_FLOCK
| FL_SLEEP
| FL_ACCESS
| FL_LEASE
)))
724 cFYI(1, "Unknown lock flags 0x%x", pfLock
->fl_flags
);
726 if (pfLock
->fl_type
== F_WRLCK
) {
729 } else if (pfLock
->fl_type
== F_UNLCK
) {
732 /* Check if unlock includes more than
734 } else if (pfLock
->fl_type
== F_RDLCK
) {
736 lockType
|= LOCKING_ANDX_SHARED_LOCK
;
738 } else if (pfLock
->fl_type
== F_EXLCK
) {
741 } else if (pfLock
->fl_type
== F_SHLCK
) {
743 lockType
|= LOCKING_ANDX_SHARED_LOCK
;
746 cFYI(1, "Unknown type of lock");
748 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
749 tcon
= tlink_tcon(((struct cifsFileInfo
*)file
->private_data
)->tlink
);
751 if (file
->private_data
== NULL
) {
756 netfid
= ((struct cifsFileInfo
*)file
->private_data
)->netfid
;
758 if ((tcon
->ses
->capabilities
& CAP_UNIX
) &&
759 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
760 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
762 /* BB add code here to normalize offset and length to
763 account for negative length which we can not accept over the
768 if (lockType
& LOCKING_ANDX_SHARED_LOCK
)
769 posix_lock_type
= CIFS_RDLCK
;
771 posix_lock_type
= CIFS_WRLCK
;
772 rc
= CIFSSMBPosixLock(xid
, tcon
, netfid
, 1 /* get */,
774 posix_lock_type
, wait_flag
);
779 /* BB we could chain these into one lock request BB */
780 rc
= CIFSSMBLock(xid
, tcon
, netfid
, length
, pfLock
->fl_start
,
781 0, 1, lockType
, 0 /* wait flag */ );
783 rc
= CIFSSMBLock(xid
, tcon
, netfid
, length
,
784 pfLock
->fl_start
, 1 /* numUnlock */ ,
785 0 /* numLock */ , lockType
,
787 pfLock
->fl_type
= F_UNLCK
;
789 cERROR(1, "Error unlocking previously locked "
790 "range %d during test of lock", rc
);
794 /* if rc == ERR_SHARING_VIOLATION ? */
797 if (lockType
& LOCKING_ANDX_SHARED_LOCK
) {
798 pfLock
->fl_type
= F_WRLCK
;
800 rc
= CIFSSMBLock(xid
, tcon
, netfid
, length
,
801 pfLock
->fl_start
, 0, 1,
802 lockType
| LOCKING_ANDX_SHARED_LOCK
,
805 rc
= CIFSSMBLock(xid
, tcon
, netfid
,
806 length
, pfLock
->fl_start
, 1, 0,
808 LOCKING_ANDX_SHARED_LOCK
,
810 pfLock
->fl_type
= F_RDLCK
;
812 cERROR(1, "Error unlocking "
813 "previously locked range %d "
814 "during test of lock", rc
);
817 pfLock
->fl_type
= F_WRLCK
;
827 if (!numLock
&& !numUnlock
) {
828 /* if no lock or unlock then nothing
829 to do since we do not know what it is */
836 if (lockType
& LOCKING_ANDX_SHARED_LOCK
)
837 posix_lock_type
= CIFS_RDLCK
;
839 posix_lock_type
= CIFS_WRLCK
;
842 posix_lock_type
= CIFS_UNLCK
;
844 rc
= CIFSSMBPosixLock(xid
, tcon
, netfid
, 0 /* set */,
846 posix_lock_type
, wait_flag
);
848 struct cifsFileInfo
*fid
= file
->private_data
;
851 rc
= CIFSSMBLock(xid
, tcon
, netfid
, length
,
853 0, numLock
, lockType
, wait_flag
);
856 /* For Windows locks we must store them. */
857 rc
= store_file_lock(fid
, length
,
858 pfLock
->fl_start
, lockType
);
860 } else if (numUnlock
) {
861 /* For each stored lock that this unlock overlaps
862 completely, unlock it. */
864 struct cifsLockInfo
*li
, *tmp
;
867 mutex_lock(&fid
->lock_mutex
);
868 list_for_each_entry_safe(li
, tmp
, &fid
->llist
, llist
) {
869 if (pfLock
->fl_start
<= li
->offset
&&
870 (pfLock
->fl_start
+ length
) >=
871 (li
->offset
+ li
->length
)) {
872 stored_rc
= CIFSSMBLock(xid
, tcon
,
874 li
->length
, li
->offset
,
875 1, 0, li
->type
, false);
879 list_del(&li
->llist
);
884 mutex_unlock(&fid
->lock_mutex
);
888 if (pfLock
->fl_flags
& FL_POSIX
)
889 posix_lock_file_wait(file
, pfLock
);
895 * Set the timeout on write requests past EOF. For some servers (Windows)
896 * these calls can be very long.
898 * If we're writing >10M past the EOF we give a 180s timeout. Anything less
899 * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
900 * The 10M cutoff is totally arbitrary. A better scheme for this would be
901 * welcome if someone wants to suggest one.
903 * We may be able to do a better job with this if there were some way to
904 * declare that a file should be sparse.
907 cifs_write_timeout(struct cifsInodeInfo
*cifsi
, loff_t offset
)
909 if (offset
<= cifsi
->server_eof
)
911 else if (offset
> (cifsi
->server_eof
+ (10 * 1024 * 1024)))
912 return CIFS_VLONG_OP
;
917 /* update the file size (if needed) after a write */
919 cifs_update_eof(struct cifsInodeInfo
*cifsi
, loff_t offset
,
920 unsigned int bytes_written
)
922 loff_t end_of_write
= offset
+ bytes_written
;
924 if (end_of_write
> cifsi
->server_eof
)
925 cifsi
->server_eof
= end_of_write
;
928 ssize_t
cifs_user_write(struct file
*file
, const char __user
*write_data
,
929 size_t write_size
, loff_t
*poffset
)
932 unsigned int bytes_written
= 0;
933 unsigned int total_written
;
934 struct cifs_sb_info
*cifs_sb
;
935 struct cifsTconInfo
*pTcon
;
937 struct cifsFileInfo
*open_file
;
938 struct cifsInodeInfo
*cifsi
= CIFS_I(file
->f_path
.dentry
->d_inode
);
940 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
942 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
943 *poffset, file->f_path.dentry->d_name.name); */
945 if (file
->private_data
== NULL
)
948 open_file
= file
->private_data
;
949 pTcon
= tlink_tcon(open_file
->tlink
);
951 rc
= generic_write_checks(file
, poffset
, &write_size
, 0);
957 long_op
= cifs_write_timeout(cifsi
, *poffset
);
958 for (total_written
= 0; write_size
> total_written
;
959 total_written
+= bytes_written
) {
961 while (rc
== -EAGAIN
) {
962 if (file
->private_data
== NULL
) {
963 /* file has been closed on us */
965 /* if we have gotten here we have written some data
966 and blocked, and the file has been freed on us while
967 we blocked so return what we managed to write */
968 return total_written
;
970 if (open_file
->closePend
) {
973 return total_written
;
977 if (open_file
->invalidHandle
) {
978 /* we could deadlock if we called
979 filemap_fdatawait from here so tell
980 reopen_file not to flush data to server
982 rc
= cifs_reopen_file(open_file
, false);
987 rc
= CIFSSMBWrite(xid
, pTcon
,
989 min_t(const int, cifs_sb
->wsize
,
990 write_size
- total_written
),
991 *poffset
, &bytes_written
,
992 NULL
, write_data
+ total_written
, long_op
);
994 if (rc
|| (bytes_written
== 0)) {
1002 cifs_update_eof(cifsi
, *poffset
, bytes_written
);
1003 *poffset
+= bytes_written
;
1005 long_op
= CIFS_STD_OP
; /* subsequent writes fast -
1006 15 seconds is plenty */
1009 cifs_stats_bytes_written(pTcon
, total_written
);
1011 /* since the write may have blocked check these pointers again */
1012 if ((file
->f_path
.dentry
) && (file
->f_path
.dentry
->d_inode
)) {
1013 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1014 /* Do not update local mtime - server will set its actual value on write
1015 * inode->i_ctime = inode->i_mtime =
1016 * current_fs_time(inode->i_sb);*/
1017 if (total_written
> 0) {
1018 spin_lock(&inode
->i_lock
);
1019 if (*poffset
> file
->f_path
.dentry
->d_inode
->i_size
)
1020 i_size_write(file
->f_path
.dentry
->d_inode
,
1022 spin_unlock(&inode
->i_lock
);
1024 mark_inode_dirty_sync(file
->f_path
.dentry
->d_inode
);
1027 return total_written
;
1030 static ssize_t
cifs_write(struct cifsFileInfo
*open_file
,
1031 const char *write_data
, size_t write_size
,
1035 unsigned int bytes_written
= 0;
1036 unsigned int total_written
;
1037 struct cifs_sb_info
*cifs_sb
;
1038 struct cifsTconInfo
*pTcon
;
1040 struct dentry
*dentry
= open_file
->dentry
;
1041 struct cifsInodeInfo
*cifsi
= CIFS_I(dentry
->d_inode
);
1043 cifs_sb
= CIFS_SB(dentry
->d_sb
);
1045 cFYI(1, "write %zd bytes to offset %lld of %s", write_size
,
1046 *poffset
, dentry
->d_name
.name
);
1048 pTcon
= tlink_tcon(open_file
->tlink
);
1052 long_op
= cifs_write_timeout(cifsi
, *poffset
);
1053 for (total_written
= 0; write_size
> total_written
;
1054 total_written
+= bytes_written
) {
1056 while (rc
== -EAGAIN
) {
1057 if (open_file
->closePend
) {
1060 return total_written
;
1064 if (open_file
->invalidHandle
) {
1065 /* we could deadlock if we called
1066 filemap_fdatawait from here so tell
1067 reopen_file not to flush data to
1069 rc
= cifs_reopen_file(open_file
, false);
1073 if (experimEnabled
|| (pTcon
->ses
->server
&&
1074 ((pTcon
->ses
->server
->secMode
&
1075 (SECMODE_SIGN_REQUIRED
| SECMODE_SIGN_ENABLED
))
1080 len
= min((size_t)cifs_sb
->wsize
,
1081 write_size
- total_written
);
1082 /* iov[0] is reserved for smb header */
1083 iov
[1].iov_base
= (char *)write_data
+
1085 iov
[1].iov_len
= len
;
1086 rc
= CIFSSMBWrite2(xid
, pTcon
,
1087 open_file
->netfid
, len
,
1088 *poffset
, &bytes_written
,
1091 rc
= CIFSSMBWrite(xid
, pTcon
,
1093 min_t(const int, cifs_sb
->wsize
,
1094 write_size
- total_written
),
1095 *poffset
, &bytes_written
,
1096 write_data
+ total_written
,
1099 if (rc
|| (bytes_written
== 0)) {
1107 cifs_update_eof(cifsi
, *poffset
, bytes_written
);
1108 *poffset
+= bytes_written
;
1110 long_op
= CIFS_STD_OP
; /* subsequent writes fast -
1111 15 seconds is plenty */
1114 cifs_stats_bytes_written(pTcon
, total_written
);
1116 if (total_written
> 0) {
1117 spin_lock(&dentry
->d_inode
->i_lock
);
1118 if (*poffset
> dentry
->d_inode
->i_size
)
1119 i_size_write(dentry
->d_inode
, *poffset
);
1120 spin_unlock(&dentry
->d_inode
->i_lock
);
1122 mark_inode_dirty_sync(dentry
->d_inode
);
1124 return total_written
;
1127 #ifdef CONFIG_CIFS_EXPERIMENTAL
1128 struct cifsFileInfo
*find_readable_file(struct cifsInodeInfo
*cifs_inode
,
1131 struct cifsFileInfo
*open_file
= NULL
;
1132 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1134 /* only filter by fsuid on multiuser mounts */
1135 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1138 read_lock(&GlobalSMBSeslock
);
1139 /* we could simply get the first_list_entry since write-only entries
1140 are always at the end of the list but since the first entry might
1141 have a close pending, we go through the whole list */
1142 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1143 if (open_file
->closePend
)
1145 if (fsuid_only
&& open_file
->uid
!= current_fsuid())
1147 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_READ
) {
1148 if (!open_file
->invalidHandle
) {
1149 /* found a good file */
1150 /* lock it so it will not be closed on us */
1151 cifsFileInfo_get(open_file
);
1152 read_unlock(&GlobalSMBSeslock
);
1154 } /* else might as well continue, and look for
1155 another, or simply have the caller reopen it
1156 again rather than trying to fix this handle */
1157 } else /* write only file */
1158 break; /* write only files are last so must be done */
1160 read_unlock(&GlobalSMBSeslock
);
1165 struct cifsFileInfo
*find_writable_file(struct cifsInodeInfo
*cifs_inode
,
1168 struct cifsFileInfo
*open_file
;
1169 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1170 bool any_available
= false;
1173 /* Having a null inode here (because mapping->host was set to zero by
1174 the VFS or MM) should not happen but we had reports of on oops (due to
1175 it being zero) during stress testcases so we need to check for it */
1177 if (cifs_inode
== NULL
) {
1178 cERROR(1, "Null inode passed to cifs_writeable_file");
1183 /* only filter by fsuid on multiuser mounts */
1184 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1187 read_lock(&GlobalSMBSeslock
);
1189 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1190 if (open_file
->closePend
)
1192 if (!any_available
&& open_file
->pid
!= current
->tgid
)
1194 if (fsuid_only
&& open_file
->uid
!= current_fsuid())
1196 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
1197 cifsFileInfo_get(open_file
);
1199 if (!open_file
->invalidHandle
) {
1200 /* found a good writable file */
1201 read_unlock(&GlobalSMBSeslock
);
1205 read_unlock(&GlobalSMBSeslock
);
1206 /* Had to unlock since following call can block */
1207 rc
= cifs_reopen_file(open_file
, false);
1209 if (!open_file
->closePend
)
1211 else { /* start over in case this was deleted */
1212 /* since the list could be modified */
1213 read_lock(&GlobalSMBSeslock
);
1214 cifsFileInfo_put(open_file
);
1215 goto refind_writable
;
1219 /* if it fails, try another handle if possible -
1220 (we can not do this if closePending since
1221 loop could be modified - in which case we
1222 have to start at the beginning of the list
1223 again. Note that it would be bad
1224 to hold up writepages here (rather than
1225 in caller) with continuous retries */
1226 cFYI(1, "wp failed on reopen file");
1227 read_lock(&GlobalSMBSeslock
);
1228 /* can not use this handle, no write
1229 pending on this one after all */
1230 cifsFileInfo_put(open_file
);
1232 if (open_file
->closePend
) /* list could have changed */
1233 goto refind_writable
;
1234 /* else we simply continue to the next entry. Thus
1235 we do not loop on reopen errors. If we
1236 can not reopen the file, for example if we
1237 reconnected to a server with another client
1238 racing to delete or lock the file we would not
1239 make progress if we restarted before the beginning
1240 of the loop here. */
1243 /* couldn't find useable FH with same pid, try any available */
1244 if (!any_available
) {
1245 any_available
= true;
1246 goto refind_writable
;
1248 read_unlock(&GlobalSMBSeslock
);
1252 static int cifs_partialpagewrite(struct page
*page
, unsigned from
, unsigned to
)
1254 struct address_space
*mapping
= page
->mapping
;
1255 loff_t offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
1258 int bytes_written
= 0;
1259 struct cifs_sb_info
*cifs_sb
;
1260 struct inode
*inode
;
1261 struct cifsFileInfo
*open_file
;
1263 if (!mapping
|| !mapping
->host
)
1266 inode
= page
->mapping
->host
;
1267 cifs_sb
= CIFS_SB(inode
->i_sb
);
1269 offset
+= (loff_t
)from
;
1270 write_data
= kmap(page
);
1273 if ((to
> PAGE_CACHE_SIZE
) || (from
> to
)) {
1278 /* racing with truncate? */
1279 if (offset
> mapping
->host
->i_size
) {
1281 return 0; /* don't care */
1284 /* check to make sure that we are not extending the file */
1285 if (mapping
->host
->i_size
- offset
< (loff_t
)to
)
1286 to
= (unsigned)(mapping
->host
->i_size
- offset
);
1288 open_file
= find_writable_file(CIFS_I(mapping
->host
), false);
1290 bytes_written
= cifs_write(open_file
, write_data
,
1291 to
- from
, &offset
);
1292 cifsFileInfo_put(open_file
);
1293 /* Does mm or vfs already set times? */
1294 inode
->i_atime
= inode
->i_mtime
= current_fs_time(inode
->i_sb
);
1295 if ((bytes_written
> 0) && (offset
))
1297 else if (bytes_written
< 0)
1300 cFYI(1, "No writeable filehandles for inode");
1308 static int cifs_writepages(struct address_space
*mapping
,
1309 struct writeback_control
*wbc
)
1311 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
1312 unsigned int bytes_to_write
;
1313 unsigned int bytes_written
;
1314 struct cifs_sb_info
*cifs_sb
;
1318 int range_whole
= 0;
1325 struct cifsFileInfo
*open_file
;
1326 struct cifsTconInfo
*tcon
;
1327 struct cifsInodeInfo
*cifsi
= CIFS_I(mapping
->host
);
1329 struct pagevec pvec
;
1335 * BB: Is this meaningful for a non-block-device file system?
1336 * If it is, we should test it again after we do I/O
1338 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
1339 wbc
->encountered_congestion
= 1;
1343 cifs_sb
= CIFS_SB(mapping
->host
->i_sb
);
1346 * If wsize is smaller that the page cache size, default to writing
1347 * one page at a time via cifs_writepage
1349 if (cifs_sb
->wsize
< PAGE_CACHE_SIZE
)
1350 return generic_writepages(mapping
, wbc
);
1352 iov
= kmalloc(32 * sizeof(struct kvec
), GFP_KERNEL
);
1354 return generic_writepages(mapping
, wbc
);
1357 * if there's no open file, then this is likely to fail too,
1358 * but it'll at least handle the return. Maybe it should be
1361 open_file
= find_writable_file(CIFS_I(mapping
->host
), false);
1364 return generic_writepages(mapping
, wbc
);
1367 tcon
= tlink_tcon(open_file
->tlink
);
1368 if (!experimEnabled
&& tcon
->ses
->server
->secMode
&
1369 (SECMODE_SIGN_REQUIRED
| SECMODE_SIGN_ENABLED
)) {
1370 cifsFileInfo_put(open_file
);
1371 return generic_writepages(mapping
, wbc
);
1373 cifsFileInfo_put(open_file
);
1377 pagevec_init(&pvec
, 0);
1378 if (wbc
->range_cyclic
) {
1379 index
= mapping
->writeback_index
; /* Start from prev offset */
1382 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
1383 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
1384 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
1389 while (!done
&& (index
<= end
) &&
1390 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1391 PAGECACHE_TAG_DIRTY
,
1392 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1))) {
1401 for (i
= 0; i
< nr_pages
; i
++) {
1402 page
= pvec
.pages
[i
];
1404 * At this point we hold neither mapping->tree_lock nor
1405 * lock on the page itself: the page may be truncated or
1406 * invalidated (changing page->mapping to NULL), or even
1407 * swizzled back from swapper_space to tmpfs file
1413 else if (!trylock_page(page
))
1416 if (unlikely(page
->mapping
!= mapping
)) {
1421 if (!wbc
->range_cyclic
&& page
->index
> end
) {
1427 if (next
&& (page
->index
!= next
)) {
1428 /* Not next consecutive page */
1433 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
1434 wait_on_page_writeback(page
);
1436 if (PageWriteback(page
) ||
1437 !clear_page_dirty_for_io(page
)) {
1443 * This actually clears the dirty bit in the radix tree.
1444 * See cifs_writepage() for more commentary.
1446 set_page_writeback(page
);
1448 if (page_offset(page
) >= mapping
->host
->i_size
) {
1451 end_page_writeback(page
);
1456 * BB can we get rid of this? pages are held by pvec
1458 page_cache_get(page
);
1460 len
= min(mapping
->host
->i_size
- page_offset(page
),
1461 (loff_t
)PAGE_CACHE_SIZE
);
1463 /* reserve iov[0] for the smb header */
1465 iov
[n_iov
].iov_base
= kmap(page
);
1466 iov
[n_iov
].iov_len
= len
;
1467 bytes_to_write
+= len
;
1471 offset
= page_offset(page
);
1473 next
= page
->index
+ 1;
1474 if (bytes_to_write
+ PAGE_CACHE_SIZE
> cifs_sb
->wsize
)
1478 open_file
= find_writable_file(CIFS_I(mapping
->host
),
1481 cERROR(1, "No writable handles for inode");
1484 long_op
= cifs_write_timeout(cifsi
, offset
);
1485 rc
= CIFSSMBWrite2(xid
, tcon
, open_file
->netfid
,
1486 bytes_to_write
, offset
,
1487 &bytes_written
, iov
, n_iov
,
1489 cifsFileInfo_put(open_file
);
1490 cifs_update_eof(cifsi
, offset
, bytes_written
);
1493 if (rc
|| bytes_written
< bytes_to_write
) {
1494 cERROR(1, "Write2 ret %d, wrote %d",
1496 /* BB what if continued retry is
1497 requested via mount flags? */
1499 set_bit(AS_ENOSPC
, &mapping
->flags
);
1501 set_bit(AS_EIO
, &mapping
->flags
);
1503 cifs_stats_bytes_written(tcon
, bytes_written
);
1506 for (i
= 0; i
< n_iov
; i
++) {
1507 page
= pvec
.pages
[first
+ i
];
1508 /* Should we also set page error on
1509 success rc but too little data written? */
1510 /* BB investigate retry logic on temporary
1511 server crash cases and how recovery works
1512 when page marked as error */
1517 end_page_writeback(page
);
1518 page_cache_release(page
);
1520 if ((wbc
->nr_to_write
-= n_iov
) <= 0)
1524 /* Need to re-find the pages we skipped */
1525 index
= pvec
.pages
[0]->index
+ 1;
1527 pagevec_release(&pvec
);
1529 if (!scanned
&& !done
) {
1531 * We hit the last page and there is more work to be done: wrap
1532 * back to the start of the file
1538 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
1539 mapping
->writeback_index
= index
;
1546 static int cifs_writepage(struct page
*page
, struct writeback_control
*wbc
)
1552 /* BB add check for wbc flags */
1553 page_cache_get(page
);
1554 if (!PageUptodate(page
))
1555 cFYI(1, "ppw - page not up to date");
1558 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1560 * A writepage() implementation always needs to do either this,
1561 * or re-dirty the page with "redirty_page_for_writepage()" in
1562 * the case of a failure.
1564 * Just unlocking the page will cause the radix tree tag-bits
1565 * to fail to update with the state of the page correctly.
1567 set_page_writeback(page
);
1568 rc
= cifs_partialpagewrite(page
, 0, PAGE_CACHE_SIZE
);
1569 SetPageUptodate(page
); /* BB add check for error and Clearuptodate? */
1571 end_page_writeback(page
);
1572 page_cache_release(page
);
1577 static int cifs_write_end(struct file
*file
, struct address_space
*mapping
,
1578 loff_t pos
, unsigned len
, unsigned copied
,
1579 struct page
*page
, void *fsdata
)
1582 struct inode
*inode
= mapping
->host
;
1584 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1587 if (PageChecked(page
)) {
1589 SetPageUptodate(page
);
1590 ClearPageChecked(page
);
1591 } else if (!PageUptodate(page
) && copied
== PAGE_CACHE_SIZE
)
1592 SetPageUptodate(page
);
1594 if (!PageUptodate(page
)) {
1596 unsigned offset
= pos
& (PAGE_CACHE_SIZE
- 1);
1600 /* this is probably better than directly calling
1601 partialpage_write since in this function the file handle is
1602 known which we might as well leverage */
1603 /* BB check if anything else missing out of ppw
1604 such as updating last write time */
1605 page_data
= kmap(page
);
1606 rc
= cifs_write(file
->private_data
, page_data
+ offset
,
1608 /* if (rc < 0) should we set writebehind rc? */
1615 set_page_dirty(page
);
1619 spin_lock(&inode
->i_lock
);
1620 if (pos
> inode
->i_size
)
1621 i_size_write(inode
, pos
);
1622 spin_unlock(&inode
->i_lock
);
1626 page_cache_release(page
);
1631 int cifs_fsync(struct file
*file
, int datasync
)
1635 struct cifsTconInfo
*tcon
;
1636 struct cifsFileInfo
*smbfile
= file
->private_data
;
1637 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1641 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1642 file
->f_path
.dentry
->d_name
.name
, datasync
);
1644 rc
= filemap_write_and_wait(inode
->i_mapping
);
1646 rc
= CIFS_I(inode
)->write_behind_rc
;
1647 CIFS_I(inode
)->write_behind_rc
= 0;
1648 tcon
= tlink_tcon(smbfile
->tlink
);
1649 if (!rc
&& tcon
&& smbfile
&&
1650 !(CIFS_SB(inode
->i_sb
)->mnt_cifs_flags
& CIFS_MOUNT_NOSSYNC
))
1651 rc
= CIFSSMBFlush(xid
, tcon
, smbfile
->netfid
);
1658 /* static void cifs_sync_page(struct page *page)
1660 struct address_space *mapping;
1661 struct inode *inode;
1662 unsigned long index = page->index;
1663 unsigned int rpages = 0;
1666 cFYI(1, "sync page %p", page);
1667 mapping = page->mapping;
1670 inode = mapping->host;
1674 /* fill in rpages then
1675 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1677 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1687 * As file closes, flush all cached write data for this inode checking
1688 * for write behind errors.
1690 int cifs_flush(struct file
*file
, fl_owner_t id
)
1692 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1695 /* Rather than do the steps manually:
1696 lock the inode for writing
1697 loop through pages looking for write behind data (dirty pages)
1698 coalesce into contiguous 16K (or smaller) chunks to write to server
1699 send to server (prefer in parallel)
1700 deal with writebehind errors
1701 unlock inode for writing
1702 filemapfdatawrite appears easier for the time being */
1704 rc
= filemap_fdatawrite(inode
->i_mapping
);
1705 /* reset wb rc if we were able to write out dirty pages */
1707 rc
= CIFS_I(inode
)->write_behind_rc
;
1708 CIFS_I(inode
)->write_behind_rc
= 0;
1711 cFYI(1, "Flush inode %p file %p rc %d", inode
, file
, rc
);
1716 ssize_t
cifs_user_read(struct file
*file
, char __user
*read_data
,
1717 size_t read_size
, loff_t
*poffset
)
1720 unsigned int bytes_read
= 0;
1721 unsigned int total_read
= 0;
1722 unsigned int current_read_size
;
1723 struct cifs_sb_info
*cifs_sb
;
1724 struct cifsTconInfo
*pTcon
;
1726 struct cifsFileInfo
*open_file
;
1727 char *smb_read_data
;
1728 char __user
*current_offset
;
1729 struct smb_com_read_rsp
*pSMBr
;
1732 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
1734 if (file
->private_data
== NULL
) {
1739 open_file
= file
->private_data
;
1740 pTcon
= tlink_tcon(open_file
->tlink
);
1742 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
1743 cFYI(1, "attempting read on write only file instance");
1745 for (total_read
= 0, current_offset
= read_data
;
1746 read_size
> total_read
;
1747 total_read
+= bytes_read
, current_offset
+= bytes_read
) {
1748 current_read_size
= min_t(const int, read_size
- total_read
,
1751 smb_read_data
= NULL
;
1752 while (rc
== -EAGAIN
) {
1753 int buf_type
= CIFS_NO_BUFFER
;
1754 if ((open_file
->invalidHandle
) &&
1755 (!open_file
->closePend
)) {
1756 rc
= cifs_reopen_file(open_file
, true);
1760 rc
= CIFSSMBRead(xid
, pTcon
,
1762 current_read_size
, *poffset
,
1763 &bytes_read
, &smb_read_data
,
1765 pSMBr
= (struct smb_com_read_rsp
*)smb_read_data
;
1766 if (smb_read_data
) {
1767 if (copy_to_user(current_offset
,
1769 4 /* RFC1001 length field */ +
1770 le16_to_cpu(pSMBr
->DataOffset
),
1774 if (buf_type
== CIFS_SMALL_BUFFER
)
1775 cifs_small_buf_release(smb_read_data
);
1776 else if (buf_type
== CIFS_LARGE_BUFFER
)
1777 cifs_buf_release(smb_read_data
);
1778 smb_read_data
= NULL
;
1781 if (rc
|| (bytes_read
== 0)) {
1789 cifs_stats_bytes_read(pTcon
, bytes_read
);
1790 *poffset
+= bytes_read
;
1798 static ssize_t
cifs_read(struct file
*file
, char *read_data
, size_t read_size
,
1802 unsigned int bytes_read
= 0;
1803 unsigned int total_read
;
1804 unsigned int current_read_size
;
1805 struct cifs_sb_info
*cifs_sb
;
1806 struct cifsTconInfo
*pTcon
;
1808 char *current_offset
;
1809 struct cifsFileInfo
*open_file
;
1810 int buf_type
= CIFS_NO_BUFFER
;
1813 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
1815 if (file
->private_data
== NULL
) {
1820 open_file
= file
->private_data
;
1821 pTcon
= tlink_tcon(open_file
->tlink
);
1823 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
1824 cFYI(1, "attempting read on write only file instance");
1826 for (total_read
= 0, current_offset
= read_data
;
1827 read_size
> total_read
;
1828 total_read
+= bytes_read
, current_offset
+= bytes_read
) {
1829 current_read_size
= min_t(const int, read_size
- total_read
,
1831 /* For windows me and 9x we do not want to request more
1832 than it negotiated since it will refuse the read then */
1834 !(pTcon
->ses
->capabilities
& CAP_LARGE_FILES
)) {
1835 current_read_size
= min_t(const int, current_read_size
,
1836 pTcon
->ses
->server
->maxBuf
- 128);
1839 while (rc
== -EAGAIN
) {
1840 if ((open_file
->invalidHandle
) &&
1841 (!open_file
->closePend
)) {
1842 rc
= cifs_reopen_file(open_file
, true);
1846 rc
= CIFSSMBRead(xid
, pTcon
,
1848 current_read_size
, *poffset
,
1849 &bytes_read
, ¤t_offset
,
1852 if (rc
|| (bytes_read
== 0)) {
1860 cifs_stats_bytes_read(pTcon
, total_read
);
1861 *poffset
+= bytes_read
;
1868 int cifs_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1873 rc
= cifs_revalidate_file(file
);
1875 cFYI(1, "Validation prior to mmap failed, error=%d", rc
);
1879 rc
= generic_file_mmap(file
, vma
);
1885 static void cifs_copy_cache_pages(struct address_space
*mapping
,
1886 struct list_head
*pages
, int bytes_read
, char *data
)
1891 while (bytes_read
> 0) {
1892 if (list_empty(pages
))
1895 page
= list_entry(pages
->prev
, struct page
, lru
);
1896 list_del(&page
->lru
);
1898 if (add_to_page_cache_lru(page
, mapping
, page
->index
,
1900 page_cache_release(page
);
1901 cFYI(1, "Add page cache failed");
1902 data
+= PAGE_CACHE_SIZE
;
1903 bytes_read
-= PAGE_CACHE_SIZE
;
1906 page_cache_release(page
);
1908 target
= kmap_atomic(page
, KM_USER0
);
1910 if (PAGE_CACHE_SIZE
> bytes_read
) {
1911 memcpy(target
, data
, bytes_read
);
1912 /* zero the tail end of this partial page */
1913 memset(target
+ bytes_read
, 0,
1914 PAGE_CACHE_SIZE
- bytes_read
);
1917 memcpy(target
, data
, PAGE_CACHE_SIZE
);
1918 bytes_read
-= PAGE_CACHE_SIZE
;
1920 kunmap_atomic(target
, KM_USER0
);
1922 flush_dcache_page(page
);
1923 SetPageUptodate(page
);
1925 data
+= PAGE_CACHE_SIZE
;
1927 /* add page to FS-Cache */
1928 cifs_readpage_to_fscache(mapping
->host
, page
);
1933 static int cifs_readpages(struct file
*file
, struct address_space
*mapping
,
1934 struct list_head
*page_list
, unsigned num_pages
)
1940 struct cifs_sb_info
*cifs_sb
;
1941 struct cifsTconInfo
*pTcon
;
1942 unsigned int bytes_read
= 0;
1943 unsigned int read_size
, i
;
1944 char *smb_read_data
= NULL
;
1945 struct smb_com_read_rsp
*pSMBr
;
1946 struct cifsFileInfo
*open_file
;
1947 int buf_type
= CIFS_NO_BUFFER
;
1950 if (file
->private_data
== NULL
) {
1955 open_file
= file
->private_data
;
1956 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
1957 pTcon
= tlink_tcon(open_file
->tlink
);
1960 * Reads as many pages as possible from fscache. Returns -ENOBUFS
1961 * immediately if the cookie is negative
1963 rc
= cifs_readpages_from_fscache(mapping
->host
, mapping
, page_list
,
1968 cFYI(DBG2
, "rpages: num pages %d", num_pages
);
1969 for (i
= 0; i
< num_pages
; ) {
1970 unsigned contig_pages
;
1971 struct page
*tmp_page
;
1972 unsigned long expected_index
;
1974 if (list_empty(page_list
))
1977 page
= list_entry(page_list
->prev
, struct page
, lru
);
1978 offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
1980 /* count adjacent pages that we will read into */
1983 list_entry(page_list
->prev
, struct page
, lru
)->index
;
1984 list_for_each_entry_reverse(tmp_page
, page_list
, lru
) {
1985 if (tmp_page
->index
== expected_index
) {
1991 if (contig_pages
+ i
> num_pages
)
1992 contig_pages
= num_pages
- i
;
1994 /* for reads over a certain size could initiate async
1997 read_size
= contig_pages
* PAGE_CACHE_SIZE
;
1998 /* Read size needs to be in multiples of one page */
1999 read_size
= min_t(const unsigned int, read_size
,
2000 cifs_sb
->rsize
& PAGE_CACHE_MASK
);
2001 cFYI(DBG2
, "rpages: read size 0x%x contiguous pages %d",
2002 read_size
, contig_pages
);
2004 while (rc
== -EAGAIN
) {
2005 if ((open_file
->invalidHandle
) &&
2006 (!open_file
->closePend
)) {
2007 rc
= cifs_reopen_file(open_file
, true);
2012 rc
= CIFSSMBRead(xid
, pTcon
,
2015 &bytes_read
, &smb_read_data
,
2017 /* BB more RC checks ? */
2018 if (rc
== -EAGAIN
) {
2019 if (smb_read_data
) {
2020 if (buf_type
== CIFS_SMALL_BUFFER
)
2021 cifs_small_buf_release(smb_read_data
);
2022 else if (buf_type
== CIFS_LARGE_BUFFER
)
2023 cifs_buf_release(smb_read_data
);
2024 smb_read_data
= NULL
;
2028 if ((rc
< 0) || (smb_read_data
== NULL
)) {
2029 cFYI(1, "Read error in readpages: %d", rc
);
2031 } else if (bytes_read
> 0) {
2032 task_io_account_read(bytes_read
);
2033 pSMBr
= (struct smb_com_read_rsp
*)smb_read_data
;
2034 cifs_copy_cache_pages(mapping
, page_list
, bytes_read
,
2035 smb_read_data
+ 4 /* RFC1001 hdr */ +
2036 le16_to_cpu(pSMBr
->DataOffset
));
2038 i
+= bytes_read
>> PAGE_CACHE_SHIFT
;
2039 cifs_stats_bytes_read(pTcon
, bytes_read
);
2040 if ((bytes_read
& PAGE_CACHE_MASK
) != bytes_read
) {
2041 i
++; /* account for partial page */
2043 /* server copy of file can have smaller size
2045 /* BB do we need to verify this common case ?
2046 this case is ok - if we are at server EOF
2047 we will hit it on next read */
2052 cFYI(1, "No bytes read (%d) at offset %lld . "
2053 "Cleaning remaining pages from readahead list",
2054 bytes_read
, offset
);
2055 /* BB turn off caching and do new lookup on
2056 file size at server? */
2059 if (smb_read_data
) {
2060 if (buf_type
== CIFS_SMALL_BUFFER
)
2061 cifs_small_buf_release(smb_read_data
);
2062 else if (buf_type
== CIFS_LARGE_BUFFER
)
2063 cifs_buf_release(smb_read_data
);
2064 smb_read_data
= NULL
;
2069 /* need to free smb_read_data buf before exit */
2070 if (smb_read_data
) {
2071 if (buf_type
== CIFS_SMALL_BUFFER
)
2072 cifs_small_buf_release(smb_read_data
);
2073 else if (buf_type
== CIFS_LARGE_BUFFER
)
2074 cifs_buf_release(smb_read_data
);
2075 smb_read_data
= NULL
;
2083 static int cifs_readpage_worker(struct file
*file
, struct page
*page
,
2089 /* Is the page cached? */
2090 rc
= cifs_readpage_from_fscache(file
->f_path
.dentry
->d_inode
, page
);
2094 page_cache_get(page
);
2095 read_data
= kmap(page
);
2096 /* for reads over a certain size could initiate async read ahead */
2098 rc
= cifs_read(file
, read_data
, PAGE_CACHE_SIZE
, poffset
);
2103 cFYI(1, "Bytes read %d", rc
);
2105 file
->f_path
.dentry
->d_inode
->i_atime
=
2106 current_fs_time(file
->f_path
.dentry
->d_inode
->i_sb
);
2108 if (PAGE_CACHE_SIZE
> rc
)
2109 memset(read_data
+ rc
, 0, PAGE_CACHE_SIZE
- rc
);
2111 flush_dcache_page(page
);
2112 SetPageUptodate(page
);
2114 /* send this page to the cache */
2115 cifs_readpage_to_fscache(file
->f_path
.dentry
->d_inode
, page
);
2121 page_cache_release(page
);
2127 static int cifs_readpage(struct file
*file
, struct page
*page
)
2129 loff_t offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
2135 if (file
->private_data
== NULL
) {
2141 cFYI(1, "readpage %p at offset %d 0x%x\n",
2142 page
, (int)offset
, (int)offset
);
2144 rc
= cifs_readpage_worker(file
, page
, &offset
);
2152 static int is_inode_writable(struct cifsInodeInfo
*cifs_inode
)
2154 struct cifsFileInfo
*open_file
;
2156 read_lock(&GlobalSMBSeslock
);
2157 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
2158 if (open_file
->closePend
)
2160 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
2161 read_unlock(&GlobalSMBSeslock
);
2165 read_unlock(&GlobalSMBSeslock
);
2169 /* We do not want to update the file size from server for inodes
2170 open for write - to avoid races with writepage extending
2171 the file - in the future we could consider allowing
2172 refreshing the inode only on increases in the file size
2173 but this is tricky to do without racing with writebehind
2174 page caching in the current Linux kernel design */
2175 bool is_size_safe_to_change(struct cifsInodeInfo
*cifsInode
, __u64 end_of_file
)
2180 if (is_inode_writable(cifsInode
)) {
2181 /* This inode is open for write at least once */
2182 struct cifs_sb_info
*cifs_sb
;
2184 cifs_sb
= CIFS_SB(cifsInode
->vfs_inode
.i_sb
);
2185 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_DIRECT_IO
) {
2186 /* since no page cache to corrupt on directio
2187 we can change size safely */
2191 if (i_size_read(&cifsInode
->vfs_inode
) < end_of_file
)
2199 static int cifs_write_begin(struct file
*file
, struct address_space
*mapping
,
2200 loff_t pos
, unsigned len
, unsigned flags
,
2201 struct page
**pagep
, void **fsdata
)
2203 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
2204 loff_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
2205 loff_t page_start
= pos
& PAGE_MASK
;
2210 cFYI(1, "write_begin from %lld len %d", (long long)pos
, len
);
2212 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2218 if (PageUptodate(page
))
2222 * If we write a full page it will be up to date, no need to read from
2223 * the server. If the write is short, we'll end up doing a sync write
2226 if (len
== PAGE_CACHE_SIZE
)
2230 * optimize away the read when we have an oplock, and we're not
2231 * expecting to use any of the data we'd be reading in. That
2232 * is, when the page lies beyond the EOF, or straddles the EOF
2233 * and the write will cover all of the existing data.
2235 if (CIFS_I(mapping
->host
)->clientCanCacheRead
) {
2236 i_size
= i_size_read(mapping
->host
);
2237 if (page_start
>= i_size
||
2238 (offset
== 0 && (pos
+ len
) >= i_size
)) {
2239 zero_user_segments(page
, 0, offset
,
2243 * PageChecked means that the parts of the page
2244 * to which we're not writing are considered up
2245 * to date. Once the data is copied to the
2246 * page, it can be set uptodate.
2248 SetPageChecked(page
);
2253 if ((file
->f_flags
& O_ACCMODE
) != O_WRONLY
) {
2255 * might as well read a page, it is fast enough. If we get
2256 * an error, we don't need to return it. cifs_write_end will
2257 * do a sync write instead since PG_uptodate isn't set.
2259 cifs_readpage_worker(file
, page
, &page_start
);
2261 /* we could try using another file handle if there is one -
2262 but how would we lock it to prevent close of that handle
2263 racing with this read? In any case
2264 this will be written out by write_end so is fine */
2271 static int cifs_release_page(struct page
*page
, gfp_t gfp
)
2273 if (PagePrivate(page
))
2276 return cifs_fscache_release_page(page
, gfp
);
2279 static void cifs_invalidate_page(struct page
*page
, unsigned long offset
)
2281 struct cifsInodeInfo
*cifsi
= CIFS_I(page
->mapping
->host
);
2284 cifs_fscache_invalidate_page(page
, &cifsi
->vfs_inode
);
2287 void cifs_oplock_break(struct work_struct
*work
)
2289 struct cifsFileInfo
*cfile
= container_of(work
, struct cifsFileInfo
,
2291 struct inode
*inode
= cfile
->dentry
->d_inode
;
2292 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
2295 if (inode
&& S_ISREG(inode
->i_mode
)) {
2296 if (cinode
->clientCanCacheRead
)
2297 break_lease(inode
, O_RDONLY
);
2299 break_lease(inode
, O_WRONLY
);
2300 rc
= filemap_fdatawrite(inode
->i_mapping
);
2301 if (cinode
->clientCanCacheRead
== 0) {
2302 waitrc
= filemap_fdatawait(inode
->i_mapping
);
2303 invalidate_remote_inode(inode
);
2308 cinode
->write_behind_rc
= rc
;
2309 cFYI(1, "Oplock flush inode %p rc %d", inode
, rc
);
2313 * releasing stale oplock after recent reconnect of smb session using
2314 * a now incorrect file handle is not a data integrity issue but do
2315 * not bother sending an oplock release if session to server still is
2316 * disconnected since oplock already released by the server
2318 if (!cfile
->closePend
&& !cfile
->oplock_break_cancelled
) {
2319 rc
= CIFSSMBLock(0, tlink_tcon(cfile
->tlink
), cfile
->netfid
, 0,
2320 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE
, false);
2321 cFYI(1, "Oplock release rc = %d", rc
);
2325 * We might have kicked in before is_valid_oplock_break()
2326 * finished grabbing reference for us. Make sure it's done by
2327 * waiting for GlobalSMSSeslock.
2329 write_lock(&GlobalSMBSeslock
);
2330 write_unlock(&GlobalSMBSeslock
);
2332 cifs_oplock_break_put(cfile
);
2335 void cifs_oplock_break_get(struct cifsFileInfo
*cfile
)
2337 cifs_sb_active(cfile
->dentry
->d_sb
);
2338 cifsFileInfo_get(cfile
);
2341 void cifs_oplock_break_put(struct cifsFileInfo
*cfile
)
2343 cifsFileInfo_put(cfile
);
2344 cifs_sb_deactive(cfile
->dentry
->d_sb
);
2347 const struct address_space_operations cifs_addr_ops
= {
2348 .readpage
= cifs_readpage
,
2349 .readpages
= cifs_readpages
,
2350 .writepage
= cifs_writepage
,
2351 .writepages
= cifs_writepages
,
2352 .write_begin
= cifs_write_begin
,
2353 .write_end
= cifs_write_end
,
2354 .set_page_dirty
= __set_page_dirty_nobuffers
,
2355 .releasepage
= cifs_release_page
,
2356 .invalidatepage
= cifs_invalidate_page
,
2357 /* .sync_page = cifs_sync_page, */
2362 * cifs_readpages requires the server to support a buffer large enough to
2363 * contain the header plus one complete page of data. Otherwise, we need
2364 * to leave cifs_readpages out of the address space operations.
2366 const struct address_space_operations cifs_addr_ops_smallbuf
= {
2367 .readpage
= cifs_readpage
,
2368 .writepage
= cifs_writepage
,
2369 .writepages
= cifs_writepages
,
2370 .write_begin
= cifs_write_begin
,
2371 .write_end
= cifs_write_end
,
2372 .set_page_dirty
= __set_page_dirty_nobuffers
,
2373 .releasepage
= cifs_release_page
,
2374 .invalidatepage
= cifs_invalidate_page
,
2375 /* .sync_page = cifs_sync_page, */