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
);
134 mapping_set_error(inode
->i_mapping
, rc
);
136 cFYI(1, "invalidating remote inode since open detected it "
138 invalidate_remote_inode(inode
);
143 rc
= cifs_get_inode_info_unix(&inode
, full_path
, inode
->i_sb
,
146 rc
= cifs_get_inode_info(&inode
, full_path
, buf
, inode
->i_sb
,
149 if ((oplock
& 0xF) == OPLOCK_EXCLUSIVE
) {
150 pCifsInode
->clientCanCacheAll
= true;
151 pCifsInode
->clientCanCacheRead
= true;
152 cFYI(1, "Exclusive Oplock granted on inode %p", inode
);
153 } else if ((oplock
& 0xF) == OPLOCK_READ
)
154 pCifsInode
->clientCanCacheRead
= true;
159 int cifs_posix_open(char *full_path
, struct inode
**pinode
,
160 struct super_block
*sb
, int mode
, unsigned int f_flags
,
161 __u32
*poplock
, __u16
*pnetfid
, int xid
)
164 FILE_UNIX_BASIC_INFO
*presp_data
;
165 __u32 posix_flags
= 0;
166 struct cifs_sb_info
*cifs_sb
= CIFS_SB(sb
);
167 struct cifs_fattr fattr
;
168 struct tcon_link
*tlink
;
169 struct cifsTconInfo
*tcon
;
171 cFYI(1, "posix open %s", full_path
);
173 presp_data
= kzalloc(sizeof(FILE_UNIX_BASIC_INFO
), GFP_KERNEL
);
174 if (presp_data
== NULL
)
177 tlink
= cifs_sb_tlink(cifs_sb
);
183 tcon
= tlink_tcon(tlink
);
184 mode
&= ~current_umask();
186 posix_flags
= cifs_posix_convert_flags(f_flags
);
187 rc
= CIFSPOSIXCreate(xid
, tcon
, posix_flags
, mode
, pnetfid
, presp_data
,
188 poplock
, full_path
, cifs_sb
->local_nls
,
189 cifs_sb
->mnt_cifs_flags
&
190 CIFS_MOUNT_MAP_SPECIAL_CHR
);
191 cifs_put_tlink(tlink
);
196 if (presp_data
->Type
== cpu_to_le32(-1))
197 goto posix_open_ret
; /* open ok, caller does qpathinfo */
200 goto posix_open_ret
; /* caller does not need info */
202 cifs_unix_basic_to_fattr(&fattr
, presp_data
, cifs_sb
);
204 /* get new inode and set it up */
205 if (*pinode
== NULL
) {
206 cifs_fill_uniqueid(sb
, &fattr
);
207 *pinode
= cifs_iget(sb
, &fattr
);
213 cifs_fattr_to_inode(*pinode
, &fattr
);
221 struct cifsFileInfo
*
222 cifs_new_fileinfo(__u16 fileHandle
, struct file
*file
,
223 struct tcon_link
*tlink
, __u32 oplock
)
225 struct dentry
*dentry
= file
->f_path
.dentry
;
226 struct inode
*inode
= dentry
->d_inode
;
227 struct cifsInodeInfo
*pCifsInode
= CIFS_I(inode
);
228 struct cifsFileInfo
*pCifsFile
;
230 pCifsFile
= kzalloc(sizeof(struct cifsFileInfo
), GFP_KERNEL
);
231 if (pCifsFile
== NULL
)
234 pCifsFile
->count
= 1;
235 pCifsFile
->netfid
= fileHandle
;
236 pCifsFile
->pid
= current
->tgid
;
237 pCifsFile
->uid
= current_fsuid();
238 pCifsFile
->dentry
= dget(dentry
);
239 pCifsFile
->f_flags
= file
->f_flags
;
240 pCifsFile
->invalidHandle
= false;
241 pCifsFile
->tlink
= cifs_get_tlink(tlink
);
242 mutex_init(&pCifsFile
->fh_mutex
);
243 mutex_init(&pCifsFile
->lock_mutex
);
244 INIT_LIST_HEAD(&pCifsFile
->llist
);
245 INIT_WORK(&pCifsFile
->oplock_break
, cifs_oplock_break
);
247 spin_lock(&cifs_file_list_lock
);
248 list_add(&pCifsFile
->tlist
, &(tlink_tcon(tlink
)->openFileList
));
249 /* if readable file instance put first in list*/
250 if (file
->f_mode
& FMODE_READ
)
251 list_add(&pCifsFile
->flist
, &pCifsInode
->openFileList
);
253 list_add_tail(&pCifsFile
->flist
, &pCifsInode
->openFileList
);
254 spin_unlock(&cifs_file_list_lock
);
256 if ((oplock
& 0xF) == OPLOCK_EXCLUSIVE
) {
257 pCifsInode
->clientCanCacheAll
= true;
258 pCifsInode
->clientCanCacheRead
= true;
259 cFYI(1, "Exclusive Oplock inode %p", inode
);
260 } else if ((oplock
& 0xF) == OPLOCK_READ
)
261 pCifsInode
->clientCanCacheRead
= true;
263 file
->private_data
= pCifsFile
;
268 * Release a reference on the file private data. This may involve closing
269 * the filehandle out on the server. Must be called without holding
270 * cifs_file_list_lock.
272 void cifsFileInfo_put(struct cifsFileInfo
*cifs_file
)
274 struct cifsTconInfo
*tcon
= tlink_tcon(cifs_file
->tlink
);
275 struct cifsInodeInfo
*cifsi
= CIFS_I(cifs_file
->dentry
->d_inode
);
276 struct cifsLockInfo
*li
, *tmp
;
278 spin_lock(&cifs_file_list_lock
);
279 if (--cifs_file
->count
> 0) {
280 spin_unlock(&cifs_file_list_lock
);
284 /* remove it from the lists */
285 list_del(&cifs_file
->flist
);
286 list_del(&cifs_file
->tlist
);
288 if (list_empty(&cifsi
->openFileList
)) {
289 cFYI(1, "closing last open instance for inode %p",
290 cifs_file
->dentry
->d_inode
);
291 cifsi
->clientCanCacheRead
= false;
292 cifsi
->clientCanCacheAll
= false;
294 spin_unlock(&cifs_file_list_lock
);
296 if (!tcon
->need_reconnect
&& !cifs_file
->invalidHandle
) {
300 rc
= CIFSSMBClose(xid
, tcon
, cifs_file
->netfid
);
304 /* Delete any outstanding lock records. We'll lose them when the file
307 mutex_lock(&cifs_file
->lock_mutex
);
308 list_for_each_entry_safe(li
, tmp
, &cifs_file
->llist
, llist
) {
309 list_del(&li
->llist
);
312 mutex_unlock(&cifs_file
->lock_mutex
);
314 cifs_put_tlink(cifs_file
->tlink
);
315 dput(cifs_file
->dentry
);
319 int cifs_open(struct inode
*inode
, struct file
*file
)
324 struct cifs_sb_info
*cifs_sb
;
325 struct cifsTconInfo
*tcon
;
326 struct tcon_link
*tlink
;
327 struct cifsFileInfo
*pCifsFile
= NULL
;
328 struct cifsInodeInfo
*pCifsInode
;
329 char *full_path
= NULL
;
333 FILE_ALL_INFO
*buf
= NULL
;
337 cifs_sb
= CIFS_SB(inode
->i_sb
);
338 tlink
= cifs_sb_tlink(cifs_sb
);
341 return PTR_ERR(tlink
);
343 tcon
= tlink_tcon(tlink
);
345 pCifsInode
= CIFS_I(file
->f_path
.dentry
->d_inode
);
347 full_path
= build_path_from_dentry(file
->f_path
.dentry
);
348 if (full_path
== NULL
) {
353 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
354 inode
, file
->f_flags
, full_path
);
361 if (!tcon
->broken_posix_open
&& tcon
->unix_ext
&&
362 (tcon
->ses
->capabilities
& CAP_UNIX
) &&
363 (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
364 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
365 /* can not refresh inode info since size could be stale */
366 rc
= cifs_posix_open(full_path
, &inode
, inode
->i_sb
,
367 cifs_sb
->mnt_file_mode
/* ignored */,
368 file
->f_flags
, &oplock
, &netfid
, xid
);
370 cFYI(1, "posix open succeeded");
372 pCifsFile
= cifs_new_fileinfo(netfid
, file
, tlink
,
374 if (pCifsFile
== NULL
) {
375 CIFSSMBClose(xid
, tcon
, netfid
);
379 cifs_fscache_set_inode_cookie(inode
, file
);
382 } else if ((rc
== -EINVAL
) || (rc
== -EOPNOTSUPP
)) {
383 if (tcon
->ses
->serverNOS
)
384 cERROR(1, "server %s of type %s returned"
385 " unexpected error on SMB posix open"
386 ", disabling posix open support."
387 " Check if server update available.",
388 tcon
->ses
->serverName
,
389 tcon
->ses
->serverNOS
);
390 tcon
->broken_posix_open
= true;
391 } else if ((rc
!= -EIO
) && (rc
!= -EREMOTE
) &&
392 (rc
!= -EOPNOTSUPP
)) /* path not found or net err */
394 /* else fallthrough to retry open the old way on network i/o
398 desiredAccess
= cifs_convert_flags(file
->f_flags
);
400 /*********************************************************************
401 * open flag mapping table:
403 * POSIX Flag CIFS Disposition
404 * ---------- ----------------
405 * O_CREAT FILE_OPEN_IF
406 * O_CREAT | O_EXCL FILE_CREATE
407 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
408 * O_TRUNC FILE_OVERWRITE
409 * none of the above FILE_OPEN
411 * Note that there is not a direct match between disposition
412 * FILE_SUPERSEDE (ie create whether or not file exists although
413 * O_CREAT | O_TRUNC is similar but truncates the existing
414 * file rather than creating a new file as FILE_SUPERSEDE does
415 * (which uses the attributes / metadata passed in on open call)
417 *? O_SYNC is a reasonable match to CIFS writethrough flag
418 *? and the read write flags match reasonably. O_LARGEFILE
419 *? is irrelevant because largefile support is always used
420 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
421 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
422 *********************************************************************/
424 disposition
= cifs_get_disposition(file
->f_flags
);
426 /* BB pass O_SYNC flag through on file attributes .. BB */
428 /* Also refresh inode by passing in file_info buf returned by SMBOpen
429 and calling get_inode_info with returned buf (at least helps
430 non-Unix server case) */
432 /* BB we can not do this if this is the second open of a file
433 and the first handle has writebehind data, we might be
434 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
435 buf
= kmalloc(sizeof(FILE_ALL_INFO
), GFP_KERNEL
);
441 if (tcon
->ses
->capabilities
& CAP_NT_SMBS
)
442 rc
= CIFSSMBOpen(xid
, tcon
, full_path
, disposition
,
443 desiredAccess
, CREATE_NOT_DIR
, &netfid
, &oplock
, buf
,
444 cifs_sb
->local_nls
, cifs_sb
->mnt_cifs_flags
445 & CIFS_MOUNT_MAP_SPECIAL_CHR
);
447 rc
= -EIO
; /* no NT SMB support fall into legacy open below */
450 /* Old server, try legacy style OpenX */
451 rc
= SMBLegacyOpen(xid
, tcon
, full_path
, disposition
,
452 desiredAccess
, CREATE_NOT_DIR
, &netfid
, &oplock
, buf
,
453 cifs_sb
->local_nls
, cifs_sb
->mnt_cifs_flags
454 & CIFS_MOUNT_MAP_SPECIAL_CHR
);
457 cFYI(1, "cifs_open returned 0x%x", rc
);
461 rc
= cifs_open_inode_helper(inode
, tcon
, oplock
, buf
, full_path
, xid
);
465 pCifsFile
= cifs_new_fileinfo(netfid
, file
, tlink
, oplock
);
466 if (pCifsFile
== NULL
) {
471 cifs_fscache_set_inode_cookie(inode
, file
);
473 if (oplock
& CIFS_CREATE_ACTION
) {
474 /* time to set mode which we can not set earlier due to
475 problems creating new read-only files */
476 if (tcon
->unix_ext
) {
477 struct cifs_unix_set_info_args args
= {
478 .mode
= inode
->i_mode
,
481 .ctime
= NO_CHANGE_64
,
482 .atime
= NO_CHANGE_64
,
483 .mtime
= NO_CHANGE_64
,
486 CIFSSMBUnixSetPathInfo(xid
, tcon
, full_path
, &args
,
488 cifs_sb
->mnt_cifs_flags
&
489 CIFS_MOUNT_MAP_SPECIAL_CHR
);
497 cifs_put_tlink(tlink
);
501 /* Try to reacquire byte range locks that were released when session */
502 /* to server was lost */
503 static int cifs_relock_file(struct cifsFileInfo
*cifsFile
)
507 /* BB list all locks open on this file and relock */
512 static int cifs_reopen_file(struct cifsFileInfo
*pCifsFile
, bool can_flush
)
517 struct cifs_sb_info
*cifs_sb
;
518 struct cifsTconInfo
*tcon
;
519 struct cifsInodeInfo
*pCifsInode
;
521 char *full_path
= NULL
;
523 int disposition
= FILE_OPEN
;
527 mutex_lock(&pCifsFile
->fh_mutex
);
528 if (!pCifsFile
->invalidHandle
) {
529 mutex_unlock(&pCifsFile
->fh_mutex
);
535 inode
= pCifsFile
->dentry
->d_inode
;
536 cifs_sb
= CIFS_SB(inode
->i_sb
);
537 tcon
= tlink_tcon(pCifsFile
->tlink
);
539 /* can not grab rename sem here because various ops, including
540 those that already have the rename sem can end up causing writepage
541 to get called and if the server was down that means we end up here,
542 and we can never tell if the caller already has the rename_sem */
543 full_path
= build_path_from_dentry(pCifsFile
->dentry
);
544 if (full_path
== NULL
) {
546 mutex_unlock(&pCifsFile
->fh_mutex
);
551 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
552 inode
, pCifsFile
->f_flags
, full_path
);
559 if (tcon
->unix_ext
&& (tcon
->ses
->capabilities
& CAP_UNIX
) &&
560 (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
561 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
564 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
565 * original open. Must mask them off for a reopen.
567 unsigned int oflags
= pCifsFile
->f_flags
&
568 ~(O_CREAT
| O_EXCL
| O_TRUNC
);
570 rc
= cifs_posix_open(full_path
, NULL
, inode
->i_sb
,
571 cifs_sb
->mnt_file_mode
/* ignored */,
572 oflags
, &oplock
, &netfid
, xid
);
574 cFYI(1, "posix reopen succeeded");
577 /* fallthrough to retry open the old way on errors, especially
578 in the reconnect path it is important to retry hard */
581 desiredAccess
= cifs_convert_flags(pCifsFile
->f_flags
);
583 /* Can not refresh inode by passing in file_info buf to be returned
584 by SMBOpen and then calling get_inode_info with returned buf
585 since file might have write behind data that needs to be flushed
586 and server version of file size can be stale. If we knew for sure
587 that inode was not dirty locally we could do this */
589 rc
= CIFSSMBOpen(xid
, tcon
, full_path
, disposition
, desiredAccess
,
590 CREATE_NOT_DIR
, &netfid
, &oplock
, NULL
,
591 cifs_sb
->local_nls
, cifs_sb
->mnt_cifs_flags
&
592 CIFS_MOUNT_MAP_SPECIAL_CHR
);
594 mutex_unlock(&pCifsFile
->fh_mutex
);
595 cFYI(1, "cifs_open returned 0x%x", rc
);
596 cFYI(1, "oplock: %d", oplock
);
597 goto reopen_error_exit
;
601 pCifsFile
->netfid
= netfid
;
602 pCifsFile
->invalidHandle
= false;
603 mutex_unlock(&pCifsFile
->fh_mutex
);
604 pCifsInode
= CIFS_I(inode
);
607 rc
= filemap_write_and_wait(inode
->i_mapping
);
608 mapping_set_error(inode
->i_mapping
, rc
);
610 pCifsInode
->clientCanCacheAll
= false;
611 pCifsInode
->clientCanCacheRead
= false;
613 rc
= cifs_get_inode_info_unix(&inode
,
614 full_path
, inode
->i_sb
, xid
);
616 rc
= cifs_get_inode_info(&inode
,
617 full_path
, NULL
, inode
->i_sb
,
619 } /* else we are writing out data to server already
620 and could deadlock if we tried to flush data, and
621 since we do not know if we have data that would
622 invalidate the current end of file on the server
623 we can not go to the server to get the new inod
625 if ((oplock
& 0xF) == OPLOCK_EXCLUSIVE
) {
626 pCifsInode
->clientCanCacheAll
= true;
627 pCifsInode
->clientCanCacheRead
= true;
628 cFYI(1, "Exclusive Oplock granted on inode %p",
629 pCifsFile
->dentry
->d_inode
);
630 } else if ((oplock
& 0xF) == OPLOCK_READ
) {
631 pCifsInode
->clientCanCacheRead
= true;
632 pCifsInode
->clientCanCacheAll
= false;
634 pCifsInode
->clientCanCacheRead
= false;
635 pCifsInode
->clientCanCacheAll
= false;
637 cifs_relock_file(pCifsFile
);
645 int cifs_close(struct inode
*inode
, struct file
*file
)
647 cifsFileInfo_put(file
->private_data
);
648 file
->private_data
= NULL
;
650 /* return code from the ->release op is always ignored */
654 int cifs_closedir(struct inode
*inode
, struct file
*file
)
658 struct cifsFileInfo
*pCFileStruct
= file
->private_data
;
661 cFYI(1, "Closedir inode = 0x%p", inode
);
666 struct cifsTconInfo
*pTcon
= tlink_tcon(pCFileStruct
->tlink
);
668 cFYI(1, "Freeing private data in close dir");
669 spin_lock(&cifs_file_list_lock
);
670 if (!pCFileStruct
->srch_inf
.endOfSearch
&&
671 !pCFileStruct
->invalidHandle
) {
672 pCFileStruct
->invalidHandle
= true;
673 spin_unlock(&cifs_file_list_lock
);
674 rc
= CIFSFindClose(xid
, pTcon
, pCFileStruct
->netfid
);
675 cFYI(1, "Closing uncompleted readdir with rc %d",
677 /* not much we can do if it fails anyway, ignore rc */
680 spin_unlock(&cifs_file_list_lock
);
681 ptmp
= pCFileStruct
->srch_inf
.ntwrk_buf_start
;
683 cFYI(1, "closedir free smb buf in srch struct");
684 pCFileStruct
->srch_inf
.ntwrk_buf_start
= NULL
;
685 if (pCFileStruct
->srch_inf
.smallBuf
)
686 cifs_small_buf_release(ptmp
);
688 cifs_buf_release(ptmp
);
690 cifs_put_tlink(pCFileStruct
->tlink
);
691 kfree(file
->private_data
);
692 file
->private_data
= NULL
;
694 /* BB can we lock the filestruct while this is going on? */
699 static int store_file_lock(struct cifsFileInfo
*fid
, __u64 len
,
700 __u64 offset
, __u8 lockType
)
702 struct cifsLockInfo
*li
=
703 kmalloc(sizeof(struct cifsLockInfo
), GFP_KERNEL
);
709 mutex_lock(&fid
->lock_mutex
);
710 list_add(&li
->llist
, &fid
->llist
);
711 mutex_unlock(&fid
->lock_mutex
);
715 int cifs_lock(struct file
*file
, int cmd
, struct file_lock
*pfLock
)
721 bool wait_flag
= false;
722 struct cifs_sb_info
*cifs_sb
;
723 struct cifsTconInfo
*tcon
;
725 __u8 lockType
= LOCKING_ANDX_LARGE_FILES
;
726 bool posix_locking
= 0;
728 length
= 1 + pfLock
->fl_end
- pfLock
->fl_start
;
732 cFYI(1, "Lock parm: 0x%x flockflags: "
733 "0x%x flocktype: 0x%x start: %lld end: %lld",
734 cmd
, pfLock
->fl_flags
, pfLock
->fl_type
, pfLock
->fl_start
,
737 if (pfLock
->fl_flags
& FL_POSIX
)
739 if (pfLock
->fl_flags
& FL_FLOCK
)
741 if (pfLock
->fl_flags
& FL_SLEEP
) {
742 cFYI(1, "Blocking lock");
745 if (pfLock
->fl_flags
& FL_ACCESS
)
746 cFYI(1, "Process suspended by mandatory locking - "
747 "not implemented yet");
748 if (pfLock
->fl_flags
& FL_LEASE
)
749 cFYI(1, "Lease on file - not implemented yet");
750 if (pfLock
->fl_flags
&
751 (~(FL_POSIX
| FL_FLOCK
| FL_SLEEP
| FL_ACCESS
| FL_LEASE
)))
752 cFYI(1, "Unknown lock flags 0x%x", pfLock
->fl_flags
);
754 if (pfLock
->fl_type
== F_WRLCK
) {
757 } else if (pfLock
->fl_type
== F_UNLCK
) {
760 /* Check if unlock includes more than
762 } else if (pfLock
->fl_type
== F_RDLCK
) {
764 lockType
|= LOCKING_ANDX_SHARED_LOCK
;
766 } else if (pfLock
->fl_type
== F_EXLCK
) {
769 } else if (pfLock
->fl_type
== F_SHLCK
) {
771 lockType
|= LOCKING_ANDX_SHARED_LOCK
;
774 cFYI(1, "Unknown type of lock");
776 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
777 tcon
= tlink_tcon(((struct cifsFileInfo
*)file
->private_data
)->tlink
);
779 if (file
->private_data
== NULL
) {
784 netfid
= ((struct cifsFileInfo
*)file
->private_data
)->netfid
;
786 if ((tcon
->ses
->capabilities
& CAP_UNIX
) &&
787 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
788 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
790 /* BB add code here to normalize offset and length to
791 account for negative length which we can not accept over the
796 if (lockType
& LOCKING_ANDX_SHARED_LOCK
)
797 posix_lock_type
= CIFS_RDLCK
;
799 posix_lock_type
= CIFS_WRLCK
;
800 rc
= CIFSSMBPosixLock(xid
, tcon
, netfid
, 1 /* get */,
802 posix_lock_type
, wait_flag
);
807 /* BB we could chain these into one lock request BB */
808 rc
= CIFSSMBLock(xid
, tcon
, netfid
, length
, pfLock
->fl_start
,
809 0, 1, lockType
, 0 /* wait flag */ );
811 rc
= CIFSSMBLock(xid
, tcon
, netfid
, length
,
812 pfLock
->fl_start
, 1 /* numUnlock */ ,
813 0 /* numLock */ , lockType
,
815 pfLock
->fl_type
= F_UNLCK
;
817 cERROR(1, "Error unlocking previously locked "
818 "range %d during test of lock", rc
);
822 /* if rc == ERR_SHARING_VIOLATION ? */
825 if (lockType
& LOCKING_ANDX_SHARED_LOCK
) {
826 pfLock
->fl_type
= F_WRLCK
;
828 rc
= CIFSSMBLock(xid
, tcon
, netfid
, length
,
829 pfLock
->fl_start
, 0, 1,
830 lockType
| LOCKING_ANDX_SHARED_LOCK
,
833 rc
= CIFSSMBLock(xid
, tcon
, netfid
,
834 length
, pfLock
->fl_start
, 1, 0,
836 LOCKING_ANDX_SHARED_LOCK
,
838 pfLock
->fl_type
= F_RDLCK
;
840 cERROR(1, "Error unlocking "
841 "previously locked range %d "
842 "during test of lock", rc
);
845 pfLock
->fl_type
= F_WRLCK
;
855 if (!numLock
&& !numUnlock
) {
856 /* if no lock or unlock then nothing
857 to do since we do not know what it is */
864 if (lockType
& LOCKING_ANDX_SHARED_LOCK
)
865 posix_lock_type
= CIFS_RDLCK
;
867 posix_lock_type
= CIFS_WRLCK
;
870 posix_lock_type
= CIFS_UNLCK
;
872 rc
= CIFSSMBPosixLock(xid
, tcon
, netfid
, 0 /* set */,
874 posix_lock_type
, wait_flag
);
876 struct cifsFileInfo
*fid
= file
->private_data
;
879 rc
= CIFSSMBLock(xid
, tcon
, netfid
, length
,
881 0, numLock
, lockType
, wait_flag
);
884 /* For Windows locks we must store them. */
885 rc
= store_file_lock(fid
, length
,
886 pfLock
->fl_start
, lockType
);
888 } else if (numUnlock
) {
889 /* For each stored lock that this unlock overlaps
890 completely, unlock it. */
892 struct cifsLockInfo
*li
, *tmp
;
895 mutex_lock(&fid
->lock_mutex
);
896 list_for_each_entry_safe(li
, tmp
, &fid
->llist
, llist
) {
897 if (pfLock
->fl_start
<= li
->offset
&&
898 (pfLock
->fl_start
+ length
) >=
899 (li
->offset
+ li
->length
)) {
900 stored_rc
= CIFSSMBLock(xid
, tcon
,
902 li
->length
, li
->offset
,
903 1, 0, li
->type
, false);
907 list_del(&li
->llist
);
912 mutex_unlock(&fid
->lock_mutex
);
916 if (pfLock
->fl_flags
& FL_POSIX
)
917 posix_lock_file_wait(file
, pfLock
);
923 * Set the timeout on write requests past EOF. For some servers (Windows)
924 * these calls can be very long.
926 * If we're writing >10M past the EOF we give a 180s timeout. Anything less
927 * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
928 * The 10M cutoff is totally arbitrary. A better scheme for this would be
929 * welcome if someone wants to suggest one.
931 * We may be able to do a better job with this if there were some way to
932 * declare that a file should be sparse.
935 cifs_write_timeout(struct cifsInodeInfo
*cifsi
, loff_t offset
)
937 if (offset
<= cifsi
->server_eof
)
939 else if (offset
> (cifsi
->server_eof
+ (10 * 1024 * 1024)))
940 return CIFS_VLONG_OP
;
945 /* update the file size (if needed) after a write */
947 cifs_update_eof(struct cifsInodeInfo
*cifsi
, loff_t offset
,
948 unsigned int bytes_written
)
950 loff_t end_of_write
= offset
+ bytes_written
;
952 if (end_of_write
> cifsi
->server_eof
)
953 cifsi
->server_eof
= end_of_write
;
956 ssize_t
cifs_user_write(struct file
*file
, const char __user
*write_data
,
957 size_t write_size
, loff_t
*poffset
)
960 unsigned int bytes_written
= 0;
961 unsigned int total_written
;
962 struct cifs_sb_info
*cifs_sb
;
963 struct cifsTconInfo
*pTcon
;
965 struct cifsFileInfo
*open_file
;
966 struct cifsInodeInfo
*cifsi
= CIFS_I(file
->f_path
.dentry
->d_inode
);
968 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
970 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
971 *poffset, file->f_path.dentry->d_name.name); */
973 if (file
->private_data
== NULL
)
976 open_file
= file
->private_data
;
977 pTcon
= tlink_tcon(open_file
->tlink
);
979 rc
= generic_write_checks(file
, poffset
, &write_size
, 0);
985 long_op
= cifs_write_timeout(cifsi
, *poffset
);
986 for (total_written
= 0; write_size
> total_written
;
987 total_written
+= bytes_written
) {
989 while (rc
== -EAGAIN
) {
990 if (file
->private_data
== NULL
) {
991 /* file has been closed on us */
993 /* if we have gotten here we have written some data
994 and blocked, and the file has been freed on us while
995 we blocked so return what we managed to write */
996 return total_written
;
998 if (open_file
->invalidHandle
) {
999 /* we could deadlock if we called
1000 filemap_fdatawait from here so tell
1001 reopen_file not to flush data to server
1003 rc
= cifs_reopen_file(open_file
, false);
1008 rc
= CIFSSMBWrite(xid
, pTcon
,
1010 min_t(const int, cifs_sb
->wsize
,
1011 write_size
- total_written
),
1012 *poffset
, &bytes_written
,
1013 NULL
, write_data
+ total_written
, long_op
);
1015 if (rc
|| (bytes_written
== 0)) {
1023 cifs_update_eof(cifsi
, *poffset
, bytes_written
);
1024 *poffset
+= bytes_written
;
1026 long_op
= CIFS_STD_OP
; /* subsequent writes fast -
1027 15 seconds is plenty */
1030 cifs_stats_bytes_written(pTcon
, total_written
);
1032 /* since the write may have blocked check these pointers again */
1033 if ((file
->f_path
.dentry
) && (file
->f_path
.dentry
->d_inode
)) {
1034 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1035 /* Do not update local mtime - server will set its actual value on write
1036 * inode->i_ctime = inode->i_mtime =
1037 * current_fs_time(inode->i_sb);*/
1038 if (total_written
> 0) {
1039 spin_lock(&inode
->i_lock
);
1040 if (*poffset
> file
->f_path
.dentry
->d_inode
->i_size
)
1041 i_size_write(file
->f_path
.dentry
->d_inode
,
1043 spin_unlock(&inode
->i_lock
);
1045 mark_inode_dirty_sync(file
->f_path
.dentry
->d_inode
);
1048 return total_written
;
1051 static ssize_t
cifs_write(struct cifsFileInfo
*open_file
,
1052 const char *write_data
, size_t write_size
,
1056 unsigned int bytes_written
= 0;
1057 unsigned int total_written
;
1058 struct cifs_sb_info
*cifs_sb
;
1059 struct cifsTconInfo
*pTcon
;
1061 struct dentry
*dentry
= open_file
->dentry
;
1062 struct cifsInodeInfo
*cifsi
= CIFS_I(dentry
->d_inode
);
1064 cifs_sb
= CIFS_SB(dentry
->d_sb
);
1066 cFYI(1, "write %zd bytes to offset %lld of %s", write_size
,
1067 *poffset
, dentry
->d_name
.name
);
1069 pTcon
= tlink_tcon(open_file
->tlink
);
1073 long_op
= cifs_write_timeout(cifsi
, *poffset
);
1074 for (total_written
= 0; write_size
> total_written
;
1075 total_written
+= bytes_written
) {
1077 while (rc
== -EAGAIN
) {
1078 if (open_file
->invalidHandle
) {
1079 /* we could deadlock if we called
1080 filemap_fdatawait from here so tell
1081 reopen_file not to flush data to
1083 rc
= cifs_reopen_file(open_file
, false);
1087 if (experimEnabled
|| (pTcon
->ses
->server
&&
1088 ((pTcon
->ses
->server
->secMode
&
1089 (SECMODE_SIGN_REQUIRED
| SECMODE_SIGN_ENABLED
))
1094 len
= min((size_t)cifs_sb
->wsize
,
1095 write_size
- total_written
);
1096 /* iov[0] is reserved for smb header */
1097 iov
[1].iov_base
= (char *)write_data
+
1099 iov
[1].iov_len
= len
;
1100 rc
= CIFSSMBWrite2(xid
, pTcon
,
1101 open_file
->netfid
, len
,
1102 *poffset
, &bytes_written
,
1105 rc
= CIFSSMBWrite(xid
, pTcon
,
1107 min_t(const int, cifs_sb
->wsize
,
1108 write_size
- total_written
),
1109 *poffset
, &bytes_written
,
1110 write_data
+ total_written
,
1113 if (rc
|| (bytes_written
== 0)) {
1121 cifs_update_eof(cifsi
, *poffset
, bytes_written
);
1122 *poffset
+= bytes_written
;
1124 long_op
= CIFS_STD_OP
; /* subsequent writes fast -
1125 15 seconds is plenty */
1128 cifs_stats_bytes_written(pTcon
, total_written
);
1130 if (total_written
> 0) {
1131 spin_lock(&dentry
->d_inode
->i_lock
);
1132 if (*poffset
> dentry
->d_inode
->i_size
)
1133 i_size_write(dentry
->d_inode
, *poffset
);
1134 spin_unlock(&dentry
->d_inode
->i_lock
);
1136 mark_inode_dirty_sync(dentry
->d_inode
);
1138 return total_written
;
1141 #ifdef CONFIG_CIFS_EXPERIMENTAL
1142 struct cifsFileInfo
*find_readable_file(struct cifsInodeInfo
*cifs_inode
,
1145 struct cifsFileInfo
*open_file
= NULL
;
1146 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1148 /* only filter by fsuid on multiuser mounts */
1149 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1152 spin_lock(&cifs_file_list_lock
);
1153 /* we could simply get the first_list_entry since write-only entries
1154 are always at the end of the list but since the first entry might
1155 have a close pending, we go through the whole list */
1156 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1157 if (fsuid_only
&& open_file
->uid
!= current_fsuid())
1159 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_READ
) {
1160 if (!open_file
->invalidHandle
) {
1161 /* found a good file */
1162 /* lock it so it will not be closed on us */
1163 cifsFileInfo_get(open_file
);
1164 spin_unlock(&cifs_file_list_lock
);
1166 } /* else might as well continue, and look for
1167 another, or simply have the caller reopen it
1168 again rather than trying to fix this handle */
1169 } else /* write only file */
1170 break; /* write only files are last so must be done */
1172 spin_unlock(&cifs_file_list_lock
);
1177 struct cifsFileInfo
*find_writable_file(struct cifsInodeInfo
*cifs_inode
,
1180 struct cifsFileInfo
*open_file
;
1181 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1182 bool any_available
= false;
1185 /* Having a null inode here (because mapping->host was set to zero by
1186 the VFS or MM) should not happen but we had reports of on oops (due to
1187 it being zero) during stress testcases so we need to check for it */
1189 if (cifs_inode
== NULL
) {
1190 cERROR(1, "Null inode passed to cifs_writeable_file");
1195 /* only filter by fsuid on multiuser mounts */
1196 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1199 spin_lock(&cifs_file_list_lock
);
1201 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1202 if (!any_available
&& open_file
->pid
!= current
->tgid
)
1204 if (fsuid_only
&& open_file
->uid
!= current_fsuid())
1206 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
1207 cifsFileInfo_get(open_file
);
1209 if (!open_file
->invalidHandle
) {
1210 /* found a good writable file */
1211 spin_unlock(&cifs_file_list_lock
);
1215 spin_unlock(&cifs_file_list_lock
);
1217 /* Had to unlock since following call can block */
1218 rc
= cifs_reopen_file(open_file
, false);
1222 /* if it fails, try another handle if possible */
1223 cFYI(1, "wp failed on reopen file");
1224 cifsFileInfo_put(open_file
);
1226 spin_lock(&cifs_file_list_lock
);
1228 /* else we simply continue to the next entry. Thus
1229 we do not loop on reopen errors. If we
1230 can not reopen the file, for example if we
1231 reconnected to a server with another client
1232 racing to delete or lock the file we would not
1233 make progress if we restarted before the beginning
1234 of the loop here. */
1237 /* couldn't find useable FH with same pid, try any available */
1238 if (!any_available
) {
1239 any_available
= true;
1240 goto refind_writable
;
1242 spin_unlock(&cifs_file_list_lock
);
1246 static int cifs_partialpagewrite(struct page
*page
, unsigned from
, unsigned to
)
1248 struct address_space
*mapping
= page
->mapping
;
1249 loff_t offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
1252 int bytes_written
= 0;
1253 struct cifs_sb_info
*cifs_sb
;
1254 struct inode
*inode
;
1255 struct cifsFileInfo
*open_file
;
1257 if (!mapping
|| !mapping
->host
)
1260 inode
= page
->mapping
->host
;
1261 cifs_sb
= CIFS_SB(inode
->i_sb
);
1263 offset
+= (loff_t
)from
;
1264 write_data
= kmap(page
);
1267 if ((to
> PAGE_CACHE_SIZE
) || (from
> to
)) {
1272 /* racing with truncate? */
1273 if (offset
> mapping
->host
->i_size
) {
1275 return 0; /* don't care */
1278 /* check to make sure that we are not extending the file */
1279 if (mapping
->host
->i_size
- offset
< (loff_t
)to
)
1280 to
= (unsigned)(mapping
->host
->i_size
- offset
);
1282 open_file
= find_writable_file(CIFS_I(mapping
->host
), false);
1284 bytes_written
= cifs_write(open_file
, write_data
,
1285 to
- from
, &offset
);
1286 cifsFileInfo_put(open_file
);
1287 /* Does mm or vfs already set times? */
1288 inode
->i_atime
= inode
->i_mtime
= current_fs_time(inode
->i_sb
);
1289 if ((bytes_written
> 0) && (offset
))
1291 else if (bytes_written
< 0)
1294 cFYI(1, "No writeable filehandles for inode");
1302 static int cifs_writepages(struct address_space
*mapping
,
1303 struct writeback_control
*wbc
)
1305 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
1306 unsigned int bytes_to_write
;
1307 unsigned int bytes_written
;
1308 struct cifs_sb_info
*cifs_sb
;
1312 int range_whole
= 0;
1319 struct cifsFileInfo
*open_file
;
1320 struct cifsTconInfo
*tcon
;
1321 struct cifsInodeInfo
*cifsi
= CIFS_I(mapping
->host
);
1323 struct pagevec pvec
;
1329 * BB: Is this meaningful for a non-block-device file system?
1330 * If it is, we should test it again after we do I/O
1332 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
1333 wbc
->encountered_congestion
= 1;
1337 cifs_sb
= CIFS_SB(mapping
->host
->i_sb
);
1340 * If wsize is smaller that the page cache size, default to writing
1341 * one page at a time via cifs_writepage
1343 if (cifs_sb
->wsize
< PAGE_CACHE_SIZE
)
1344 return generic_writepages(mapping
, wbc
);
1346 iov
= kmalloc(32 * sizeof(struct kvec
), GFP_KERNEL
);
1348 return generic_writepages(mapping
, wbc
);
1351 * if there's no open file, then this is likely to fail too,
1352 * but it'll at least handle the return. Maybe it should be
1355 open_file
= find_writable_file(CIFS_I(mapping
->host
), false);
1358 return generic_writepages(mapping
, wbc
);
1361 tcon
= tlink_tcon(open_file
->tlink
);
1362 if (!experimEnabled
&& tcon
->ses
->server
->secMode
&
1363 (SECMODE_SIGN_REQUIRED
| SECMODE_SIGN_ENABLED
)) {
1364 cifsFileInfo_put(open_file
);
1366 return generic_writepages(mapping
, wbc
);
1368 cifsFileInfo_put(open_file
);
1372 pagevec_init(&pvec
, 0);
1373 if (wbc
->range_cyclic
) {
1374 index
= mapping
->writeback_index
; /* Start from prev offset */
1377 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
1378 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
1379 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
1384 while (!done
&& (index
<= end
) &&
1385 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1386 PAGECACHE_TAG_DIRTY
,
1387 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1))) {
1396 for (i
= 0; i
< nr_pages
; i
++) {
1397 page
= pvec
.pages
[i
];
1399 * At this point we hold neither mapping->tree_lock nor
1400 * lock on the page itself: the page may be truncated or
1401 * invalidated (changing page->mapping to NULL), or even
1402 * swizzled back from swapper_space to tmpfs file
1408 else if (!trylock_page(page
))
1411 if (unlikely(page
->mapping
!= mapping
)) {
1416 if (!wbc
->range_cyclic
&& page
->index
> end
) {
1422 if (next
&& (page
->index
!= next
)) {
1423 /* Not next consecutive page */
1428 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
1429 wait_on_page_writeback(page
);
1431 if (PageWriteback(page
) ||
1432 !clear_page_dirty_for_io(page
)) {
1438 * This actually clears the dirty bit in the radix tree.
1439 * See cifs_writepage() for more commentary.
1441 set_page_writeback(page
);
1443 if (page_offset(page
) >= mapping
->host
->i_size
) {
1446 end_page_writeback(page
);
1451 * BB can we get rid of this? pages are held by pvec
1453 page_cache_get(page
);
1455 len
= min(mapping
->host
->i_size
- page_offset(page
),
1456 (loff_t
)PAGE_CACHE_SIZE
);
1458 /* reserve iov[0] for the smb header */
1460 iov
[n_iov
].iov_base
= kmap(page
);
1461 iov
[n_iov
].iov_len
= len
;
1462 bytes_to_write
+= len
;
1466 offset
= page_offset(page
);
1468 next
= page
->index
+ 1;
1469 if (bytes_to_write
+ PAGE_CACHE_SIZE
> cifs_sb
->wsize
)
1473 open_file
= find_writable_file(CIFS_I(mapping
->host
),
1476 cERROR(1, "No writable handles for inode");
1479 long_op
= cifs_write_timeout(cifsi
, offset
);
1480 rc
= CIFSSMBWrite2(xid
, tcon
, open_file
->netfid
,
1481 bytes_to_write
, offset
,
1482 &bytes_written
, iov
, n_iov
,
1484 cifsFileInfo_put(open_file
);
1485 cifs_update_eof(cifsi
, offset
, bytes_written
);
1488 if (rc
|| bytes_written
< bytes_to_write
) {
1489 cERROR(1, "Write2 ret %d, wrote %d",
1491 mapping_set_error(mapping
, rc
);
1493 cifs_stats_bytes_written(tcon
, bytes_written
);
1496 for (i
= 0; i
< n_iov
; i
++) {
1497 page
= pvec
.pages
[first
+ i
];
1498 /* Should we also set page error on
1499 success rc but too little data written? */
1500 /* BB investigate retry logic on temporary
1501 server crash cases and how recovery works
1502 when page marked as error */
1507 end_page_writeback(page
);
1508 page_cache_release(page
);
1510 if ((wbc
->nr_to_write
-= n_iov
) <= 0)
1514 /* Need to re-find the pages we skipped */
1515 index
= pvec
.pages
[0]->index
+ 1;
1517 pagevec_release(&pvec
);
1519 if (!scanned
&& !done
) {
1521 * We hit the last page and there is more work to be done: wrap
1522 * back to the start of the file
1528 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
1529 mapping
->writeback_index
= index
;
1536 static int cifs_writepage(struct page
*page
, struct writeback_control
*wbc
)
1542 /* BB add check for wbc flags */
1543 page_cache_get(page
);
1544 if (!PageUptodate(page
))
1545 cFYI(1, "ppw - page not up to date");
1548 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1550 * A writepage() implementation always needs to do either this,
1551 * or re-dirty the page with "redirty_page_for_writepage()" in
1552 * the case of a failure.
1554 * Just unlocking the page will cause the radix tree tag-bits
1555 * to fail to update with the state of the page correctly.
1557 set_page_writeback(page
);
1558 rc
= cifs_partialpagewrite(page
, 0, PAGE_CACHE_SIZE
);
1559 SetPageUptodate(page
); /* BB add check for error and Clearuptodate? */
1561 end_page_writeback(page
);
1562 page_cache_release(page
);
1567 static int cifs_write_end(struct file
*file
, struct address_space
*mapping
,
1568 loff_t pos
, unsigned len
, unsigned copied
,
1569 struct page
*page
, void *fsdata
)
1572 struct inode
*inode
= mapping
->host
;
1574 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1577 if (PageChecked(page
)) {
1579 SetPageUptodate(page
);
1580 ClearPageChecked(page
);
1581 } else if (!PageUptodate(page
) && copied
== PAGE_CACHE_SIZE
)
1582 SetPageUptodate(page
);
1584 if (!PageUptodate(page
)) {
1586 unsigned offset
= pos
& (PAGE_CACHE_SIZE
- 1);
1590 /* this is probably better than directly calling
1591 partialpage_write since in this function the file handle is
1592 known which we might as well leverage */
1593 /* BB check if anything else missing out of ppw
1594 such as updating last write time */
1595 page_data
= kmap(page
);
1596 rc
= cifs_write(file
->private_data
, page_data
+ offset
,
1598 /* if (rc < 0) should we set writebehind rc? */
1605 set_page_dirty(page
);
1609 spin_lock(&inode
->i_lock
);
1610 if (pos
> inode
->i_size
)
1611 i_size_write(inode
, pos
);
1612 spin_unlock(&inode
->i_lock
);
1616 page_cache_release(page
);
1621 int cifs_fsync(struct file
*file
, int datasync
)
1625 struct cifsTconInfo
*tcon
;
1626 struct cifsFileInfo
*smbfile
= file
->private_data
;
1627 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1631 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1632 file
->f_path
.dentry
->d_name
.name
, datasync
);
1634 rc
= filemap_write_and_wait(inode
->i_mapping
);
1636 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
1638 tcon
= tlink_tcon(smbfile
->tlink
);
1639 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOSSYNC
))
1640 rc
= CIFSSMBFlush(xid
, tcon
, smbfile
->netfid
);
1647 /* static void cifs_sync_page(struct page *page)
1649 struct address_space *mapping;
1650 struct inode *inode;
1651 unsigned long index = page->index;
1652 unsigned int rpages = 0;
1655 cFYI(1, "sync page %p", page);
1656 mapping = page->mapping;
1659 inode = mapping->host;
1663 /* fill in rpages then
1664 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1666 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1676 * As file closes, flush all cached write data for this inode checking
1677 * for write behind errors.
1679 int cifs_flush(struct file
*file
, fl_owner_t id
)
1681 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1684 if (file
->f_mode
& FMODE_WRITE
)
1685 rc
= filemap_write_and_wait(inode
->i_mapping
);
1687 cFYI(1, "Flush inode %p file %p rc %d", inode
, file
, rc
);
1692 ssize_t
cifs_user_read(struct file
*file
, char __user
*read_data
,
1693 size_t read_size
, loff_t
*poffset
)
1696 unsigned int bytes_read
= 0;
1697 unsigned int total_read
= 0;
1698 unsigned int current_read_size
;
1699 struct cifs_sb_info
*cifs_sb
;
1700 struct cifsTconInfo
*pTcon
;
1702 struct cifsFileInfo
*open_file
;
1703 char *smb_read_data
;
1704 char __user
*current_offset
;
1705 struct smb_com_read_rsp
*pSMBr
;
1708 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
1710 if (file
->private_data
== NULL
) {
1715 open_file
= file
->private_data
;
1716 pTcon
= tlink_tcon(open_file
->tlink
);
1718 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
1719 cFYI(1, "attempting read on write only file instance");
1721 for (total_read
= 0, current_offset
= read_data
;
1722 read_size
> total_read
;
1723 total_read
+= bytes_read
, current_offset
+= bytes_read
) {
1724 current_read_size
= min_t(const int, read_size
- total_read
,
1727 smb_read_data
= NULL
;
1728 while (rc
== -EAGAIN
) {
1729 int buf_type
= CIFS_NO_BUFFER
;
1730 if (open_file
->invalidHandle
) {
1731 rc
= cifs_reopen_file(open_file
, true);
1735 rc
= CIFSSMBRead(xid
, pTcon
,
1737 current_read_size
, *poffset
,
1738 &bytes_read
, &smb_read_data
,
1740 pSMBr
= (struct smb_com_read_rsp
*)smb_read_data
;
1741 if (smb_read_data
) {
1742 if (copy_to_user(current_offset
,
1744 4 /* RFC1001 length field */ +
1745 le16_to_cpu(pSMBr
->DataOffset
),
1749 if (buf_type
== CIFS_SMALL_BUFFER
)
1750 cifs_small_buf_release(smb_read_data
);
1751 else if (buf_type
== CIFS_LARGE_BUFFER
)
1752 cifs_buf_release(smb_read_data
);
1753 smb_read_data
= NULL
;
1756 if (rc
|| (bytes_read
== 0)) {
1764 cifs_stats_bytes_read(pTcon
, bytes_read
);
1765 *poffset
+= bytes_read
;
1773 static ssize_t
cifs_read(struct file
*file
, char *read_data
, size_t read_size
,
1777 unsigned int bytes_read
= 0;
1778 unsigned int total_read
;
1779 unsigned int current_read_size
;
1780 struct cifs_sb_info
*cifs_sb
;
1781 struct cifsTconInfo
*pTcon
;
1783 char *current_offset
;
1784 struct cifsFileInfo
*open_file
;
1785 int buf_type
= CIFS_NO_BUFFER
;
1788 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
1790 if (file
->private_data
== NULL
) {
1795 open_file
= file
->private_data
;
1796 pTcon
= tlink_tcon(open_file
->tlink
);
1798 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
1799 cFYI(1, "attempting read on write only file instance");
1801 for (total_read
= 0, current_offset
= read_data
;
1802 read_size
> total_read
;
1803 total_read
+= bytes_read
, current_offset
+= bytes_read
) {
1804 current_read_size
= min_t(const int, read_size
- total_read
,
1806 /* For windows me and 9x we do not want to request more
1807 than it negotiated since it will refuse the read then */
1809 !(pTcon
->ses
->capabilities
& CAP_LARGE_FILES
)) {
1810 current_read_size
= min_t(const int, current_read_size
,
1811 pTcon
->ses
->server
->maxBuf
- 128);
1814 while (rc
== -EAGAIN
) {
1815 if (open_file
->invalidHandle
) {
1816 rc
= cifs_reopen_file(open_file
, true);
1820 rc
= CIFSSMBRead(xid
, pTcon
,
1822 current_read_size
, *poffset
,
1823 &bytes_read
, ¤t_offset
,
1826 if (rc
|| (bytes_read
== 0)) {
1834 cifs_stats_bytes_read(pTcon
, total_read
);
1835 *poffset
+= bytes_read
;
1842 int cifs_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1847 rc
= cifs_revalidate_file(file
);
1849 cFYI(1, "Validation prior to mmap failed, error=%d", rc
);
1853 rc
= generic_file_mmap(file
, vma
);
1859 static void cifs_copy_cache_pages(struct address_space
*mapping
,
1860 struct list_head
*pages
, int bytes_read
, char *data
)
1865 while (bytes_read
> 0) {
1866 if (list_empty(pages
))
1869 page
= list_entry(pages
->prev
, struct page
, lru
);
1870 list_del(&page
->lru
);
1872 if (add_to_page_cache_lru(page
, mapping
, page
->index
,
1874 page_cache_release(page
);
1875 cFYI(1, "Add page cache failed");
1876 data
+= PAGE_CACHE_SIZE
;
1877 bytes_read
-= PAGE_CACHE_SIZE
;
1880 page_cache_release(page
);
1882 target
= kmap_atomic(page
, KM_USER0
);
1884 if (PAGE_CACHE_SIZE
> bytes_read
) {
1885 memcpy(target
, data
, bytes_read
);
1886 /* zero the tail end of this partial page */
1887 memset(target
+ bytes_read
, 0,
1888 PAGE_CACHE_SIZE
- bytes_read
);
1891 memcpy(target
, data
, PAGE_CACHE_SIZE
);
1892 bytes_read
-= PAGE_CACHE_SIZE
;
1894 kunmap_atomic(target
, KM_USER0
);
1896 flush_dcache_page(page
);
1897 SetPageUptodate(page
);
1899 data
+= PAGE_CACHE_SIZE
;
1901 /* add page to FS-Cache */
1902 cifs_readpage_to_fscache(mapping
->host
, page
);
1907 static int cifs_readpages(struct file
*file
, struct address_space
*mapping
,
1908 struct list_head
*page_list
, unsigned num_pages
)
1914 struct cifs_sb_info
*cifs_sb
;
1915 struct cifsTconInfo
*pTcon
;
1916 unsigned int bytes_read
= 0;
1917 unsigned int read_size
, i
;
1918 char *smb_read_data
= NULL
;
1919 struct smb_com_read_rsp
*pSMBr
;
1920 struct cifsFileInfo
*open_file
;
1921 int buf_type
= CIFS_NO_BUFFER
;
1924 if (file
->private_data
== NULL
) {
1929 open_file
= file
->private_data
;
1930 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
1931 pTcon
= tlink_tcon(open_file
->tlink
);
1934 * Reads as many pages as possible from fscache. Returns -ENOBUFS
1935 * immediately if the cookie is negative
1937 rc
= cifs_readpages_from_fscache(mapping
->host
, mapping
, page_list
,
1942 cFYI(DBG2
, "rpages: num pages %d", num_pages
);
1943 for (i
= 0; i
< num_pages
; ) {
1944 unsigned contig_pages
;
1945 struct page
*tmp_page
;
1946 unsigned long expected_index
;
1948 if (list_empty(page_list
))
1951 page
= list_entry(page_list
->prev
, struct page
, lru
);
1952 offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
1954 /* count adjacent pages that we will read into */
1957 list_entry(page_list
->prev
, struct page
, lru
)->index
;
1958 list_for_each_entry_reverse(tmp_page
, page_list
, lru
) {
1959 if (tmp_page
->index
== expected_index
) {
1965 if (contig_pages
+ i
> num_pages
)
1966 contig_pages
= num_pages
- i
;
1968 /* for reads over a certain size could initiate async
1971 read_size
= contig_pages
* PAGE_CACHE_SIZE
;
1972 /* Read size needs to be in multiples of one page */
1973 read_size
= min_t(const unsigned int, read_size
,
1974 cifs_sb
->rsize
& PAGE_CACHE_MASK
);
1975 cFYI(DBG2
, "rpages: read size 0x%x contiguous pages %d",
1976 read_size
, contig_pages
);
1978 while (rc
== -EAGAIN
) {
1979 if (open_file
->invalidHandle
) {
1980 rc
= cifs_reopen_file(open_file
, true);
1985 rc
= CIFSSMBRead(xid
, pTcon
,
1988 &bytes_read
, &smb_read_data
,
1990 /* BB more RC checks ? */
1991 if (rc
== -EAGAIN
) {
1992 if (smb_read_data
) {
1993 if (buf_type
== CIFS_SMALL_BUFFER
)
1994 cifs_small_buf_release(smb_read_data
);
1995 else if (buf_type
== CIFS_LARGE_BUFFER
)
1996 cifs_buf_release(smb_read_data
);
1997 smb_read_data
= NULL
;
2001 if ((rc
< 0) || (smb_read_data
== NULL
)) {
2002 cFYI(1, "Read error in readpages: %d", rc
);
2004 } else if (bytes_read
> 0) {
2005 task_io_account_read(bytes_read
);
2006 pSMBr
= (struct smb_com_read_rsp
*)smb_read_data
;
2007 cifs_copy_cache_pages(mapping
, page_list
, bytes_read
,
2008 smb_read_data
+ 4 /* RFC1001 hdr */ +
2009 le16_to_cpu(pSMBr
->DataOffset
));
2011 i
+= bytes_read
>> PAGE_CACHE_SHIFT
;
2012 cifs_stats_bytes_read(pTcon
, bytes_read
);
2013 if ((bytes_read
& PAGE_CACHE_MASK
) != bytes_read
) {
2014 i
++; /* account for partial page */
2016 /* server copy of file can have smaller size
2018 /* BB do we need to verify this common case ?
2019 this case is ok - if we are at server EOF
2020 we will hit it on next read */
2025 cFYI(1, "No bytes read (%d) at offset %lld . "
2026 "Cleaning remaining pages from readahead list",
2027 bytes_read
, offset
);
2028 /* BB turn off caching and do new lookup on
2029 file size at server? */
2032 if (smb_read_data
) {
2033 if (buf_type
== CIFS_SMALL_BUFFER
)
2034 cifs_small_buf_release(smb_read_data
);
2035 else if (buf_type
== CIFS_LARGE_BUFFER
)
2036 cifs_buf_release(smb_read_data
);
2037 smb_read_data
= NULL
;
2042 /* need to free smb_read_data buf before exit */
2043 if (smb_read_data
) {
2044 if (buf_type
== CIFS_SMALL_BUFFER
)
2045 cifs_small_buf_release(smb_read_data
);
2046 else if (buf_type
== CIFS_LARGE_BUFFER
)
2047 cifs_buf_release(smb_read_data
);
2048 smb_read_data
= NULL
;
2056 static int cifs_readpage_worker(struct file
*file
, struct page
*page
,
2062 /* Is the page cached? */
2063 rc
= cifs_readpage_from_fscache(file
->f_path
.dentry
->d_inode
, page
);
2067 page_cache_get(page
);
2068 read_data
= kmap(page
);
2069 /* for reads over a certain size could initiate async read ahead */
2071 rc
= cifs_read(file
, read_data
, PAGE_CACHE_SIZE
, poffset
);
2076 cFYI(1, "Bytes read %d", rc
);
2078 file
->f_path
.dentry
->d_inode
->i_atime
=
2079 current_fs_time(file
->f_path
.dentry
->d_inode
->i_sb
);
2081 if (PAGE_CACHE_SIZE
> rc
)
2082 memset(read_data
+ rc
, 0, PAGE_CACHE_SIZE
- rc
);
2084 flush_dcache_page(page
);
2085 SetPageUptodate(page
);
2087 /* send this page to the cache */
2088 cifs_readpage_to_fscache(file
->f_path
.dentry
->d_inode
, page
);
2094 page_cache_release(page
);
2100 static int cifs_readpage(struct file
*file
, struct page
*page
)
2102 loff_t offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
2108 if (file
->private_data
== NULL
) {
2114 cFYI(1, "readpage %p at offset %d 0x%x\n",
2115 page
, (int)offset
, (int)offset
);
2117 rc
= cifs_readpage_worker(file
, page
, &offset
);
2125 static int is_inode_writable(struct cifsInodeInfo
*cifs_inode
)
2127 struct cifsFileInfo
*open_file
;
2129 spin_lock(&cifs_file_list_lock
);
2130 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
2131 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
2132 spin_unlock(&cifs_file_list_lock
);
2136 spin_unlock(&cifs_file_list_lock
);
2140 /* We do not want to update the file size from server for inodes
2141 open for write - to avoid races with writepage extending
2142 the file - in the future we could consider allowing
2143 refreshing the inode only on increases in the file size
2144 but this is tricky to do without racing with writebehind
2145 page caching in the current Linux kernel design */
2146 bool is_size_safe_to_change(struct cifsInodeInfo
*cifsInode
, __u64 end_of_file
)
2151 if (is_inode_writable(cifsInode
)) {
2152 /* This inode is open for write at least once */
2153 struct cifs_sb_info
*cifs_sb
;
2155 cifs_sb
= CIFS_SB(cifsInode
->vfs_inode
.i_sb
);
2156 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_DIRECT_IO
) {
2157 /* since no page cache to corrupt on directio
2158 we can change size safely */
2162 if (i_size_read(&cifsInode
->vfs_inode
) < end_of_file
)
2170 static int cifs_write_begin(struct file
*file
, struct address_space
*mapping
,
2171 loff_t pos
, unsigned len
, unsigned flags
,
2172 struct page
**pagep
, void **fsdata
)
2174 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
2175 loff_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
2176 loff_t page_start
= pos
& PAGE_MASK
;
2181 cFYI(1, "write_begin from %lld len %d", (long long)pos
, len
);
2183 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2189 if (PageUptodate(page
))
2193 * If we write a full page it will be up to date, no need to read from
2194 * the server. If the write is short, we'll end up doing a sync write
2197 if (len
== PAGE_CACHE_SIZE
)
2201 * optimize away the read when we have an oplock, and we're not
2202 * expecting to use any of the data we'd be reading in. That
2203 * is, when the page lies beyond the EOF, or straddles the EOF
2204 * and the write will cover all of the existing data.
2206 if (CIFS_I(mapping
->host
)->clientCanCacheRead
) {
2207 i_size
= i_size_read(mapping
->host
);
2208 if (page_start
>= i_size
||
2209 (offset
== 0 && (pos
+ len
) >= i_size
)) {
2210 zero_user_segments(page
, 0, offset
,
2214 * PageChecked means that the parts of the page
2215 * to which we're not writing are considered up
2216 * to date. Once the data is copied to the
2217 * page, it can be set uptodate.
2219 SetPageChecked(page
);
2224 if ((file
->f_flags
& O_ACCMODE
) != O_WRONLY
) {
2226 * might as well read a page, it is fast enough. If we get
2227 * an error, we don't need to return it. cifs_write_end will
2228 * do a sync write instead since PG_uptodate isn't set.
2230 cifs_readpage_worker(file
, page
, &page_start
);
2232 /* we could try using another file handle if there is one -
2233 but how would we lock it to prevent close of that handle
2234 racing with this read? In any case
2235 this will be written out by write_end so is fine */
2242 static int cifs_release_page(struct page
*page
, gfp_t gfp
)
2244 if (PagePrivate(page
))
2247 return cifs_fscache_release_page(page
, gfp
);
2250 static void cifs_invalidate_page(struct page
*page
, unsigned long offset
)
2252 struct cifsInodeInfo
*cifsi
= CIFS_I(page
->mapping
->host
);
2255 cifs_fscache_invalidate_page(page
, &cifsi
->vfs_inode
);
2258 void cifs_oplock_break(struct work_struct
*work
)
2260 struct cifsFileInfo
*cfile
= container_of(work
, struct cifsFileInfo
,
2262 struct inode
*inode
= cfile
->dentry
->d_inode
;
2263 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
2266 if (inode
&& S_ISREG(inode
->i_mode
)) {
2267 if (cinode
->clientCanCacheRead
)
2268 break_lease(inode
, O_RDONLY
);
2270 break_lease(inode
, O_WRONLY
);
2271 rc
= filemap_fdatawrite(inode
->i_mapping
);
2272 if (cinode
->clientCanCacheRead
== 0) {
2273 rc
= filemap_fdatawait(inode
->i_mapping
);
2274 mapping_set_error(inode
->i_mapping
, rc
);
2275 invalidate_remote_inode(inode
);
2277 cFYI(1, "Oplock flush inode %p rc %d", inode
, rc
);
2281 * releasing stale oplock after recent reconnect of smb session using
2282 * a now incorrect file handle is not a data integrity issue but do
2283 * not bother sending an oplock release if session to server still is
2284 * disconnected since oplock already released by the server
2286 if (!cfile
->oplock_break_cancelled
) {
2287 rc
= CIFSSMBLock(0, tlink_tcon(cfile
->tlink
), cfile
->netfid
, 0,
2288 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE
, false);
2289 cFYI(1, "Oplock release rc = %d", rc
);
2293 * We might have kicked in before is_valid_oplock_break()
2294 * finished grabbing reference for us. Make sure it's done by
2295 * waiting for cifs_file_list_lock.
2297 spin_lock(&cifs_file_list_lock
);
2298 spin_unlock(&cifs_file_list_lock
);
2300 cifs_oplock_break_put(cfile
);
2303 /* must be called while holding cifs_file_list_lock */
2304 void cifs_oplock_break_get(struct cifsFileInfo
*cfile
)
2306 cifs_sb_active(cfile
->dentry
->d_sb
);
2307 cifsFileInfo_get(cfile
);
2310 void cifs_oplock_break_put(struct cifsFileInfo
*cfile
)
2312 cifsFileInfo_put(cfile
);
2313 cifs_sb_deactive(cfile
->dentry
->d_sb
);
2316 const struct address_space_operations cifs_addr_ops
= {
2317 .readpage
= cifs_readpage
,
2318 .readpages
= cifs_readpages
,
2319 .writepage
= cifs_writepage
,
2320 .writepages
= cifs_writepages
,
2321 .write_begin
= cifs_write_begin
,
2322 .write_end
= cifs_write_end
,
2323 .set_page_dirty
= __set_page_dirty_nobuffers
,
2324 .releasepage
= cifs_release_page
,
2325 .invalidatepage
= cifs_invalidate_page
,
2326 /* .sync_page = cifs_sync_page, */
2331 * cifs_readpages requires the server to support a buffer large enough to
2332 * contain the header plus one complete page of data. Otherwise, we need
2333 * to leave cifs_readpages out of the address space operations.
2335 const struct address_space_operations cifs_addr_ops_smallbuf
= {
2336 .readpage
= cifs_readpage
,
2337 .writepage
= cifs_writepage
,
2338 .writepages
= cifs_writepages
,
2339 .write_begin
= cifs_write_begin
,
2340 .write_end
= cifs_write_end
,
2341 .set_page_dirty
= __set_page_dirty_nobuffers
,
2342 .releasepage
= cifs_release_page
,
2343 .invalidatepage
= cifs_invalidate_page
,
2344 /* .sync_page = cifs_sync_page, */