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 <linux/swap.h>
36 #include <asm/div64.h>
40 #include "cifsproto.h"
41 #include "cifs_unicode.h"
42 #include "cifs_debug.h"
43 #include "cifs_fs_sb.h"
46 static inline int cifs_convert_flags(unsigned int flags
)
48 if ((flags
& O_ACCMODE
) == O_RDONLY
)
50 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
52 else if ((flags
& O_ACCMODE
) == O_RDWR
) {
53 /* GENERIC_ALL is too much permission to request
54 can cause unnecessary access denied on create */
55 /* return GENERIC_ALL; */
56 return (GENERIC_READ
| GENERIC_WRITE
);
59 return (READ_CONTROL
| FILE_WRITE_ATTRIBUTES
| FILE_READ_ATTRIBUTES
|
60 FILE_WRITE_EA
| FILE_APPEND_DATA
| FILE_WRITE_DATA
|
64 static u32
cifs_posix_convert_flags(unsigned int flags
)
68 if ((flags
& O_ACCMODE
) == O_RDONLY
)
69 posix_flags
= SMB_O_RDONLY
;
70 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
71 posix_flags
= SMB_O_WRONLY
;
72 else if ((flags
& O_ACCMODE
) == O_RDWR
)
73 posix_flags
= SMB_O_RDWR
;
76 posix_flags
|= SMB_O_CREAT
;
78 posix_flags
|= SMB_O_EXCL
;
80 posix_flags
|= SMB_O_TRUNC
;
81 /* be safe and imply O_SYNC for O_DSYNC */
83 posix_flags
|= SMB_O_SYNC
;
84 if (flags
& O_DIRECTORY
)
85 posix_flags
|= SMB_O_DIRECTORY
;
86 if (flags
& O_NOFOLLOW
)
87 posix_flags
|= SMB_O_NOFOLLOW
;
89 posix_flags
|= SMB_O_DIRECT
;
94 static inline int cifs_get_disposition(unsigned int flags
)
96 if ((flags
& (O_CREAT
| O_EXCL
)) == (O_CREAT
| O_EXCL
))
98 else if ((flags
& (O_CREAT
| O_TRUNC
)) == (O_CREAT
| O_TRUNC
))
99 return FILE_OVERWRITE_IF
;
100 else if ((flags
& O_CREAT
) == O_CREAT
)
102 else if ((flags
& O_TRUNC
) == O_TRUNC
)
103 return FILE_OVERWRITE
;
108 int cifs_posix_open(char *full_path
, struct inode
**pinode
,
109 struct super_block
*sb
, int mode
, unsigned int f_flags
,
110 __u32
*poplock
, __u16
*pnetfid
, unsigned int xid
)
113 FILE_UNIX_BASIC_INFO
*presp_data
;
114 __u32 posix_flags
= 0;
115 struct cifs_sb_info
*cifs_sb
= CIFS_SB(sb
);
116 struct cifs_fattr fattr
;
117 struct tcon_link
*tlink
;
118 struct cifs_tcon
*tcon
;
120 cFYI(1, "posix open %s", full_path
);
122 presp_data
= kzalloc(sizeof(FILE_UNIX_BASIC_INFO
), GFP_KERNEL
);
123 if (presp_data
== NULL
)
126 tlink
= cifs_sb_tlink(cifs_sb
);
132 tcon
= tlink_tcon(tlink
);
133 mode
&= ~current_umask();
135 posix_flags
= cifs_posix_convert_flags(f_flags
);
136 rc
= CIFSPOSIXCreate(xid
, tcon
, posix_flags
, mode
, pnetfid
, presp_data
,
137 poplock
, full_path
, cifs_sb
->local_nls
,
138 cifs_sb
->mnt_cifs_flags
&
139 CIFS_MOUNT_MAP_SPECIAL_CHR
);
140 cifs_put_tlink(tlink
);
145 if (presp_data
->Type
== cpu_to_le32(-1))
146 goto posix_open_ret
; /* open ok, caller does qpathinfo */
149 goto posix_open_ret
; /* caller does not need info */
151 cifs_unix_basic_to_fattr(&fattr
, presp_data
, cifs_sb
);
153 /* get new inode and set it up */
154 if (*pinode
== NULL
) {
155 cifs_fill_uniqueid(sb
, &fattr
);
156 *pinode
= cifs_iget(sb
, &fattr
);
162 cifs_fattr_to_inode(*pinode
, &fattr
);
171 cifs_nt_open(char *full_path
, struct inode
*inode
, struct cifs_sb_info
*cifs_sb
,
172 struct cifs_tcon
*tcon
, unsigned int f_flags
, __u32
*poplock
,
173 __u16
*pnetfid
, unsigned int xid
)
178 int create_options
= CREATE_NOT_DIR
;
181 desiredAccess
= cifs_convert_flags(f_flags
);
183 /*********************************************************************
184 * open flag mapping table:
186 * POSIX Flag CIFS Disposition
187 * ---------- ----------------
188 * O_CREAT FILE_OPEN_IF
189 * O_CREAT | O_EXCL FILE_CREATE
190 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
191 * O_TRUNC FILE_OVERWRITE
192 * none of the above FILE_OPEN
194 * Note that there is not a direct match between disposition
195 * FILE_SUPERSEDE (ie create whether or not file exists although
196 * O_CREAT | O_TRUNC is similar but truncates the existing
197 * file rather than creating a new file as FILE_SUPERSEDE does
198 * (which uses the attributes / metadata passed in on open call)
200 *? O_SYNC is a reasonable match to CIFS writethrough flag
201 *? and the read write flags match reasonably. O_LARGEFILE
202 *? is irrelevant because largefile support is always used
203 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
204 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
205 *********************************************************************/
207 disposition
= cifs_get_disposition(f_flags
);
209 /* BB pass O_SYNC flag through on file attributes .. BB */
211 buf
= kmalloc(sizeof(FILE_ALL_INFO
), GFP_KERNEL
);
215 if (backup_cred(cifs_sb
))
216 create_options
|= CREATE_OPEN_BACKUP_INTENT
;
218 if (tcon
->ses
->capabilities
& CAP_NT_SMBS
)
219 rc
= CIFSSMBOpen(xid
, tcon
, full_path
, disposition
,
220 desiredAccess
, create_options
, pnetfid
, poplock
, buf
,
221 cifs_sb
->local_nls
, cifs_sb
->mnt_cifs_flags
222 & CIFS_MOUNT_MAP_SPECIAL_CHR
);
224 rc
= SMBLegacyOpen(xid
, tcon
, full_path
, disposition
,
225 desiredAccess
, CREATE_NOT_DIR
, pnetfid
, poplock
, buf
,
226 cifs_sb
->local_nls
, cifs_sb
->mnt_cifs_flags
227 & CIFS_MOUNT_MAP_SPECIAL_CHR
);
233 rc
= cifs_get_inode_info_unix(&inode
, full_path
, inode
->i_sb
,
236 rc
= cifs_get_inode_info(&inode
, full_path
, buf
, inode
->i_sb
,
244 struct cifsFileInfo
*
245 cifs_new_fileinfo(__u16 fileHandle
, struct file
*file
,
246 struct tcon_link
*tlink
, __u32 oplock
)
248 struct dentry
*dentry
= file
->f_path
.dentry
;
249 struct inode
*inode
= dentry
->d_inode
;
250 struct cifsInodeInfo
*pCifsInode
= CIFS_I(inode
);
251 struct cifsFileInfo
*pCifsFile
;
253 pCifsFile
= kzalloc(sizeof(struct cifsFileInfo
), GFP_KERNEL
);
254 if (pCifsFile
== NULL
)
257 pCifsFile
->count
= 1;
258 pCifsFile
->netfid
= fileHandle
;
259 pCifsFile
->pid
= current
->tgid
;
260 pCifsFile
->uid
= current_fsuid();
261 pCifsFile
->dentry
= dget(dentry
);
262 pCifsFile
->f_flags
= file
->f_flags
;
263 pCifsFile
->invalidHandle
= false;
264 pCifsFile
->tlink
= cifs_get_tlink(tlink
);
265 mutex_init(&pCifsFile
->fh_mutex
);
266 INIT_WORK(&pCifsFile
->oplock_break
, cifs_oplock_break
);
267 INIT_LIST_HEAD(&pCifsFile
->llist
);
269 spin_lock(&cifs_file_list_lock
);
270 list_add(&pCifsFile
->tlist
, &(tlink_tcon(tlink
)->openFileList
));
271 /* if readable file instance put first in list*/
272 if (file
->f_mode
& FMODE_READ
)
273 list_add(&pCifsFile
->flist
, &pCifsInode
->openFileList
);
275 list_add_tail(&pCifsFile
->flist
, &pCifsInode
->openFileList
);
276 spin_unlock(&cifs_file_list_lock
);
278 cifs_set_oplock_level(pCifsInode
, oplock
);
279 pCifsInode
->can_cache_brlcks
= pCifsInode
->clientCanCacheAll
;
281 file
->private_data
= pCifsFile
;
285 static void cifs_del_lock_waiters(struct cifsLockInfo
*lock
);
288 * Release a reference on the file private data. This may involve closing
289 * the filehandle out on the server. Must be called without holding
290 * cifs_file_list_lock.
292 void cifsFileInfo_put(struct cifsFileInfo
*cifs_file
)
294 struct inode
*inode
= cifs_file
->dentry
->d_inode
;
295 struct cifs_tcon
*tcon
= tlink_tcon(cifs_file
->tlink
);
296 struct cifsInodeInfo
*cifsi
= CIFS_I(inode
);
297 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
298 struct cifsLockInfo
*li
, *tmp
;
300 spin_lock(&cifs_file_list_lock
);
301 if (--cifs_file
->count
> 0) {
302 spin_unlock(&cifs_file_list_lock
);
306 /* remove it from the lists */
307 list_del(&cifs_file
->flist
);
308 list_del(&cifs_file
->tlist
);
310 if (list_empty(&cifsi
->openFileList
)) {
311 cFYI(1, "closing last open instance for inode %p",
312 cifs_file
->dentry
->d_inode
);
314 /* in strict cache mode we need invalidate mapping on the last
315 close because it may cause a error when we open this file
316 again and get at least level II oplock */
317 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_STRICT_IO
)
318 CIFS_I(inode
)->invalid_mapping
= true;
320 cifs_set_oplock_level(cifsi
, 0);
322 spin_unlock(&cifs_file_list_lock
);
324 cancel_work_sync(&cifs_file
->oplock_break
);
326 if (!tcon
->need_reconnect
&& !cifs_file
->invalidHandle
) {
330 rc
= CIFSSMBClose(xid
, tcon
, cifs_file
->netfid
);
334 /* Delete any outstanding lock records. We'll lose them when the file
337 mutex_lock(&cifsi
->lock_mutex
);
338 list_for_each_entry_safe(li
, tmp
, &cifs_file
->llist
, llist
) {
339 list_del(&li
->llist
);
340 cifs_del_lock_waiters(li
);
343 mutex_unlock(&cifsi
->lock_mutex
);
345 cifs_put_tlink(cifs_file
->tlink
);
346 dput(cifs_file
->dentry
);
350 int cifs_open(struct inode
*inode
, struct file
*file
)
355 struct cifs_sb_info
*cifs_sb
;
356 struct cifs_tcon
*tcon
;
357 struct tcon_link
*tlink
;
358 struct cifsFileInfo
*pCifsFile
= NULL
;
359 char *full_path
= NULL
;
360 bool posix_open_ok
= false;
365 cifs_sb
= CIFS_SB(inode
->i_sb
);
366 tlink
= cifs_sb_tlink(cifs_sb
);
369 return PTR_ERR(tlink
);
371 tcon
= tlink_tcon(tlink
);
373 full_path
= build_path_from_dentry(file
->f_path
.dentry
);
374 if (full_path
== NULL
) {
379 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
380 inode
, file
->f_flags
, full_path
);
382 if (tcon
->ses
->server
->oplocks
)
387 if (!tcon
->broken_posix_open
&& tcon
->unix_ext
&&
388 cap_unix(tcon
->ses
) && (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
389 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
390 /* can not refresh inode info since size could be stale */
391 rc
= cifs_posix_open(full_path
, &inode
, inode
->i_sb
,
392 cifs_sb
->mnt_file_mode
/* ignored */,
393 file
->f_flags
, &oplock
, &netfid
, xid
);
395 cFYI(1, "posix open succeeded");
396 posix_open_ok
= true;
397 } else if ((rc
== -EINVAL
) || (rc
== -EOPNOTSUPP
)) {
398 if (tcon
->ses
->serverNOS
)
399 cERROR(1, "server %s of type %s returned"
400 " unexpected error on SMB posix open"
401 ", disabling posix open support."
402 " Check if server update available.",
403 tcon
->ses
->serverName
,
404 tcon
->ses
->serverNOS
);
405 tcon
->broken_posix_open
= true;
406 } else if ((rc
!= -EIO
) && (rc
!= -EREMOTE
) &&
407 (rc
!= -EOPNOTSUPP
)) /* path not found or net err */
409 /* else fallthrough to retry open the old way on network i/o
413 if (!posix_open_ok
) {
414 rc
= cifs_nt_open(full_path
, inode
, cifs_sb
, tcon
,
415 file
->f_flags
, &oplock
, &netfid
, xid
);
420 pCifsFile
= cifs_new_fileinfo(netfid
, file
, tlink
, oplock
);
421 if (pCifsFile
== NULL
) {
422 CIFSSMBClose(xid
, tcon
, netfid
);
427 cifs_fscache_set_inode_cookie(inode
, file
);
429 if ((oplock
& CIFS_CREATE_ACTION
) && !posix_open_ok
&& tcon
->unix_ext
) {
430 /* time to set mode which we can not set earlier due to
431 problems creating new read-only files */
432 struct cifs_unix_set_info_args args
= {
433 .mode
= inode
->i_mode
,
436 .ctime
= NO_CHANGE_64
,
437 .atime
= NO_CHANGE_64
,
438 .mtime
= NO_CHANGE_64
,
441 CIFSSMBUnixSetFileInfo(xid
, tcon
, &args
, netfid
,
448 cifs_put_tlink(tlink
);
452 /* Try to reacquire byte range locks that were released when session */
453 /* to server was lost */
454 static int cifs_relock_file(struct cifsFileInfo
*cifsFile
)
458 /* BB list all locks open on this file and relock */
463 static int cifs_reopen_file(struct cifsFileInfo
*pCifsFile
, bool can_flush
)
468 struct cifs_sb_info
*cifs_sb
;
469 struct cifs_tcon
*tcon
;
470 struct cifsInodeInfo
*pCifsInode
;
472 char *full_path
= NULL
;
474 int disposition
= FILE_OPEN
;
475 int create_options
= CREATE_NOT_DIR
;
479 mutex_lock(&pCifsFile
->fh_mutex
);
480 if (!pCifsFile
->invalidHandle
) {
481 mutex_unlock(&pCifsFile
->fh_mutex
);
487 inode
= pCifsFile
->dentry
->d_inode
;
488 cifs_sb
= CIFS_SB(inode
->i_sb
);
489 tcon
= tlink_tcon(pCifsFile
->tlink
);
491 /* can not grab rename sem here because various ops, including
492 those that already have the rename sem can end up causing writepage
493 to get called and if the server was down that means we end up here,
494 and we can never tell if the caller already has the rename_sem */
495 full_path
= build_path_from_dentry(pCifsFile
->dentry
);
496 if (full_path
== NULL
) {
498 mutex_unlock(&pCifsFile
->fh_mutex
);
503 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
504 inode
, pCifsFile
->f_flags
, full_path
);
506 if (tcon
->ses
->server
->oplocks
)
511 if (tcon
->unix_ext
&& cap_unix(tcon
->ses
) &&
512 (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
513 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
515 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
516 * original open. Must mask them off for a reopen.
518 unsigned int oflags
= pCifsFile
->f_flags
&
519 ~(O_CREAT
| O_EXCL
| O_TRUNC
);
521 rc
= cifs_posix_open(full_path
, NULL
, inode
->i_sb
,
522 cifs_sb
->mnt_file_mode
/* ignored */,
523 oflags
, &oplock
, &netfid
, xid
);
525 cFYI(1, "posix reopen succeeded");
528 /* fallthrough to retry open the old way on errors, especially
529 in the reconnect path it is important to retry hard */
532 desiredAccess
= cifs_convert_flags(pCifsFile
->f_flags
);
534 if (backup_cred(cifs_sb
))
535 create_options
|= CREATE_OPEN_BACKUP_INTENT
;
537 /* Can not refresh inode by passing in file_info buf to be returned
538 by SMBOpen and then calling get_inode_info with returned buf
539 since file might have write behind data that needs to be flushed
540 and server version of file size can be stale. If we knew for sure
541 that inode was not dirty locally we could do this */
543 rc
= CIFSSMBOpen(xid
, tcon
, full_path
, disposition
, desiredAccess
,
544 create_options
, &netfid
, &oplock
, NULL
,
545 cifs_sb
->local_nls
, cifs_sb
->mnt_cifs_flags
&
546 CIFS_MOUNT_MAP_SPECIAL_CHR
);
548 mutex_unlock(&pCifsFile
->fh_mutex
);
549 cFYI(1, "cifs_open returned 0x%x", rc
);
550 cFYI(1, "oplock: %d", oplock
);
551 goto reopen_error_exit
;
555 pCifsFile
->netfid
= netfid
;
556 pCifsFile
->invalidHandle
= false;
557 mutex_unlock(&pCifsFile
->fh_mutex
);
558 pCifsInode
= CIFS_I(inode
);
561 rc
= filemap_write_and_wait(inode
->i_mapping
);
562 mapping_set_error(inode
->i_mapping
, rc
);
565 rc
= cifs_get_inode_info_unix(&inode
,
566 full_path
, inode
->i_sb
, xid
);
568 rc
= cifs_get_inode_info(&inode
,
569 full_path
, NULL
, inode
->i_sb
,
571 } /* else we are writing out data to server already
572 and could deadlock if we tried to flush data, and
573 since we do not know if we have data that would
574 invalidate the current end of file on the server
575 we can not go to the server to get the new inod
578 cifs_set_oplock_level(pCifsInode
, oplock
);
580 cifs_relock_file(pCifsFile
);
588 int cifs_close(struct inode
*inode
, struct file
*file
)
590 if (file
->private_data
!= NULL
) {
591 cifsFileInfo_put(file
->private_data
);
592 file
->private_data
= NULL
;
595 /* return code from the ->release op is always ignored */
599 int cifs_closedir(struct inode
*inode
, struct file
*file
)
603 struct cifsFileInfo
*pCFileStruct
= file
->private_data
;
606 cFYI(1, "Closedir inode = 0x%p", inode
);
611 struct cifs_tcon
*pTcon
= tlink_tcon(pCFileStruct
->tlink
);
613 cFYI(1, "Freeing private data in close dir");
614 spin_lock(&cifs_file_list_lock
);
615 if (!pCFileStruct
->srch_inf
.endOfSearch
&&
616 !pCFileStruct
->invalidHandle
) {
617 pCFileStruct
->invalidHandle
= true;
618 spin_unlock(&cifs_file_list_lock
);
619 rc
= CIFSFindClose(xid
, pTcon
, pCFileStruct
->netfid
);
620 cFYI(1, "Closing uncompleted readdir with rc %d",
622 /* not much we can do if it fails anyway, ignore rc */
625 spin_unlock(&cifs_file_list_lock
);
626 ptmp
= pCFileStruct
->srch_inf
.ntwrk_buf_start
;
628 cFYI(1, "closedir free smb buf in srch struct");
629 pCFileStruct
->srch_inf
.ntwrk_buf_start
= NULL
;
630 if (pCFileStruct
->srch_inf
.smallBuf
)
631 cifs_small_buf_release(ptmp
);
633 cifs_buf_release(ptmp
);
635 cifs_put_tlink(pCFileStruct
->tlink
);
636 kfree(file
->private_data
);
637 file
->private_data
= NULL
;
639 /* BB can we lock the filestruct while this is going on? */
644 static struct cifsLockInfo
*
645 cifs_lock_init(__u64 offset
, __u64 length
, __u8 type
)
647 struct cifsLockInfo
*lock
=
648 kmalloc(sizeof(struct cifsLockInfo
), GFP_KERNEL
);
651 lock
->offset
= offset
;
652 lock
->length
= length
;
654 lock
->pid
= current
->tgid
;
655 INIT_LIST_HEAD(&lock
->blist
);
656 init_waitqueue_head(&lock
->block_q
);
661 cifs_del_lock_waiters(struct cifsLockInfo
*lock
)
663 struct cifsLockInfo
*li
, *tmp
;
664 list_for_each_entry_safe(li
, tmp
, &lock
->blist
, blist
) {
665 list_del_init(&li
->blist
);
666 wake_up(&li
->block_q
);
671 cifs_find_fid_lock_conflict(struct cifsFileInfo
*cfile
, __u64 offset
,
672 __u64 length
, __u8 type
, struct cifsFileInfo
*cur
,
673 struct cifsLockInfo
**conf_lock
)
675 struct cifsLockInfo
*li
;
676 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
678 list_for_each_entry(li
, &cfile
->llist
, llist
) {
679 if (offset
+ length
<= li
->offset
||
680 offset
>= li
->offset
+ li
->length
)
682 else if ((type
& server
->vals
->shared_lock_type
) &&
683 ((server
->ops
->compare_fids(cur
, cfile
) &&
684 current
->tgid
== li
->pid
) || type
== li
->type
))
695 cifs_find_lock_conflict(struct cifsFileInfo
*cfile
, __u64 offset
, __u64 length
,
696 __u8 type
, struct cifsLockInfo
**conf_lock
)
699 struct cifsFileInfo
*fid
, *tmp
;
700 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
702 spin_lock(&cifs_file_list_lock
);
703 list_for_each_entry_safe(fid
, tmp
, &cinode
->openFileList
, flist
) {
704 rc
= cifs_find_fid_lock_conflict(fid
, offset
, length
, type
,
709 spin_unlock(&cifs_file_list_lock
);
715 * Check if there is another lock that prevents us to set the lock (mandatory
716 * style). If such a lock exists, update the flock structure with its
717 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
718 * or leave it the same if we can't. Returns 0 if we don't need to request to
719 * the server or 1 otherwise.
722 cifs_lock_test(struct cifsFileInfo
*cfile
, __u64 offset
, __u64 length
,
723 __u8 type
, struct file_lock
*flock
)
726 struct cifsLockInfo
*conf_lock
;
727 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
728 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
731 mutex_lock(&cinode
->lock_mutex
);
733 exist
= cifs_find_lock_conflict(cfile
, offset
, length
, type
,
736 flock
->fl_start
= conf_lock
->offset
;
737 flock
->fl_end
= conf_lock
->offset
+ conf_lock
->length
- 1;
738 flock
->fl_pid
= conf_lock
->pid
;
739 if (conf_lock
->type
& server
->vals
->shared_lock_type
)
740 flock
->fl_type
= F_RDLCK
;
742 flock
->fl_type
= F_WRLCK
;
743 } else if (!cinode
->can_cache_brlcks
)
746 flock
->fl_type
= F_UNLCK
;
748 mutex_unlock(&cinode
->lock_mutex
);
753 cifs_lock_add(struct cifsFileInfo
*cfile
, struct cifsLockInfo
*lock
)
755 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
756 mutex_lock(&cinode
->lock_mutex
);
757 list_add_tail(&lock
->llist
, &cfile
->llist
);
758 mutex_unlock(&cinode
->lock_mutex
);
762 * Set the byte-range lock (mandatory style). Returns:
763 * 1) 0, if we set the lock and don't need to request to the server;
764 * 2) 1, if no locks prevent us but we need to request to the server;
765 * 3) -EACCESS, if there is a lock that prevents us and wait is false.
768 cifs_lock_add_if(struct cifsFileInfo
*cfile
, struct cifsLockInfo
*lock
,
771 struct cifsLockInfo
*conf_lock
;
772 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
778 mutex_lock(&cinode
->lock_mutex
);
780 exist
= cifs_find_lock_conflict(cfile
, lock
->offset
, lock
->length
,
781 lock
->type
, &conf_lock
);
782 if (!exist
&& cinode
->can_cache_brlcks
) {
783 list_add_tail(&lock
->llist
, &cfile
->llist
);
784 mutex_unlock(&cinode
->lock_mutex
);
793 list_add_tail(&lock
->blist
, &conf_lock
->blist
);
794 mutex_unlock(&cinode
->lock_mutex
);
795 rc
= wait_event_interruptible(lock
->block_q
,
796 (lock
->blist
.prev
== &lock
->blist
) &&
797 (lock
->blist
.next
== &lock
->blist
));
800 mutex_lock(&cinode
->lock_mutex
);
801 list_del_init(&lock
->blist
);
804 mutex_unlock(&cinode
->lock_mutex
);
809 * Check if there is another lock that prevents us to set the lock (posix
810 * style). If such a lock exists, update the flock structure with its
811 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
812 * or leave it the same if we can't. Returns 0 if we don't need to request to
813 * the server or 1 otherwise.
816 cifs_posix_lock_test(struct file
*file
, struct file_lock
*flock
)
819 struct cifsInodeInfo
*cinode
= CIFS_I(file
->f_path
.dentry
->d_inode
);
820 unsigned char saved_type
= flock
->fl_type
;
822 if ((flock
->fl_flags
& FL_POSIX
) == 0)
825 mutex_lock(&cinode
->lock_mutex
);
826 posix_test_lock(file
, flock
);
828 if (flock
->fl_type
== F_UNLCK
&& !cinode
->can_cache_brlcks
) {
829 flock
->fl_type
= saved_type
;
833 mutex_unlock(&cinode
->lock_mutex
);
838 * Set the byte-range lock (posix style). Returns:
839 * 1) 0, if we set the lock and don't need to request to the server;
840 * 2) 1, if we need to request to the server;
841 * 3) <0, if the error occurs while setting the lock.
844 cifs_posix_lock_set(struct file
*file
, struct file_lock
*flock
)
846 struct cifsInodeInfo
*cinode
= CIFS_I(file
->f_path
.dentry
->d_inode
);
849 if ((flock
->fl_flags
& FL_POSIX
) == 0)
853 mutex_lock(&cinode
->lock_mutex
);
854 if (!cinode
->can_cache_brlcks
) {
855 mutex_unlock(&cinode
->lock_mutex
);
859 rc
= posix_lock_file(file
, flock
, NULL
);
860 mutex_unlock(&cinode
->lock_mutex
);
861 if (rc
== FILE_LOCK_DEFERRED
) {
862 rc
= wait_event_interruptible(flock
->fl_wait
, !flock
->fl_next
);
865 locks_delete_block(flock
);
871 cifs_push_mandatory_locks(struct cifsFileInfo
*cfile
)
874 int rc
= 0, stored_rc
;
875 struct cifsLockInfo
*li
, *tmp
;
876 struct cifs_tcon
*tcon
;
877 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
878 unsigned int num
, max_num
, max_buf
;
879 LOCKING_ANDX_RANGE
*buf
, *cur
;
880 int types
[] = {LOCKING_ANDX_LARGE_FILES
,
881 LOCKING_ANDX_SHARED_LOCK
| LOCKING_ANDX_LARGE_FILES
};
885 tcon
= tlink_tcon(cfile
->tlink
);
887 mutex_lock(&cinode
->lock_mutex
);
888 if (!cinode
->can_cache_brlcks
) {
889 mutex_unlock(&cinode
->lock_mutex
);
895 * Accessing maxBuf is racy with cifs_reconnect - need to store value
896 * and check it for zero before using.
898 max_buf
= tcon
->ses
->server
->maxBuf
;
900 mutex_unlock(&cinode
->lock_mutex
);
905 max_num
= (max_buf
- sizeof(struct smb_hdr
)) /
906 sizeof(LOCKING_ANDX_RANGE
);
907 buf
= kzalloc(max_num
* sizeof(LOCKING_ANDX_RANGE
), GFP_KERNEL
);
909 mutex_unlock(&cinode
->lock_mutex
);
914 for (i
= 0; i
< 2; i
++) {
917 list_for_each_entry_safe(li
, tmp
, &cfile
->llist
, llist
) {
918 if (li
->type
!= types
[i
])
920 cur
->Pid
= cpu_to_le16(li
->pid
);
921 cur
->LengthLow
= cpu_to_le32((u32
)li
->length
);
922 cur
->LengthHigh
= cpu_to_le32((u32
)(li
->length
>>32));
923 cur
->OffsetLow
= cpu_to_le32((u32
)li
->offset
);
924 cur
->OffsetHigh
= cpu_to_le32((u32
)(li
->offset
>>32));
925 if (++num
== max_num
) {
926 stored_rc
= cifs_lockv(xid
, tcon
, cfile
->netfid
,
927 (__u8
)li
->type
, 0, num
,
938 stored_rc
= cifs_lockv(xid
, tcon
, cfile
->netfid
,
939 (__u8
)types
[i
], 0, num
, buf
);
945 cinode
->can_cache_brlcks
= false;
946 mutex_unlock(&cinode
->lock_mutex
);
953 /* copied from fs/locks.c with a name change */
954 #define cifs_for_each_lock(inode, lockp) \
955 for (lockp = &inode->i_flock; *lockp != NULL; \
956 lockp = &(*lockp)->fl_next)
958 struct lock_to_push
{
959 struct list_head llist
;
968 cifs_push_posix_locks(struct cifsFileInfo
*cfile
)
970 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
971 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
972 struct file_lock
*flock
, **before
;
973 unsigned int count
= 0, i
= 0;
974 int rc
= 0, xid
, type
;
975 struct list_head locks_to_send
, *el
;
976 struct lock_to_push
*lck
, *tmp
;
981 mutex_lock(&cinode
->lock_mutex
);
982 if (!cinode
->can_cache_brlcks
) {
983 mutex_unlock(&cinode
->lock_mutex
);
989 cifs_for_each_lock(cfile
->dentry
->d_inode
, before
) {
990 if ((*before
)->fl_flags
& FL_POSIX
)
995 INIT_LIST_HEAD(&locks_to_send
);
998 * Allocating count locks is enough because no FL_POSIX locks can be
999 * added to the list while we are holding cinode->lock_mutex that
1000 * protects locking operations of this inode.
1002 for (; i
< count
; i
++) {
1003 lck
= kmalloc(sizeof(struct lock_to_push
), GFP_KERNEL
);
1008 list_add_tail(&lck
->llist
, &locks_to_send
);
1011 el
= locks_to_send
.next
;
1013 cifs_for_each_lock(cfile
->dentry
->d_inode
, before
) {
1015 if ((flock
->fl_flags
& FL_POSIX
) == 0)
1017 if (el
== &locks_to_send
) {
1019 * The list ended. We don't have enough allocated
1020 * structures - something is really wrong.
1022 cERROR(1, "Can't push all brlocks!");
1025 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1026 if (flock
->fl_type
== F_RDLCK
|| flock
->fl_type
== F_SHLCK
)
1030 lck
= list_entry(el
, struct lock_to_push
, llist
);
1031 lck
->pid
= flock
->fl_pid
;
1032 lck
->netfid
= cfile
->netfid
;
1033 lck
->length
= length
;
1035 lck
->offset
= flock
->fl_start
;
1040 list_for_each_entry_safe(lck
, tmp
, &locks_to_send
, llist
) {
1043 stored_rc
= CIFSSMBPosixLock(xid
, tcon
, lck
->netfid
, lck
->pid
,
1044 lck
->offset
, lck
->length
, NULL
,
1048 list_del(&lck
->llist
);
1053 cinode
->can_cache_brlcks
= false;
1054 mutex_unlock(&cinode
->lock_mutex
);
1059 list_for_each_entry_safe(lck
, tmp
, &locks_to_send
, llist
) {
1060 list_del(&lck
->llist
);
1067 cifs_push_locks(struct cifsFileInfo
*cfile
)
1069 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cfile
->dentry
->d_sb
);
1070 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1072 if (cap_unix(tcon
->ses
) &&
1073 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
1074 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
1075 return cifs_push_posix_locks(cfile
);
1077 return cifs_push_mandatory_locks(cfile
);
1081 cifs_read_flock(struct file_lock
*flock
, __u32
*type
, int *lock
, int *unlock
,
1082 bool *wait_flag
, struct TCP_Server_Info
*server
)
1084 if (flock
->fl_flags
& FL_POSIX
)
1086 if (flock
->fl_flags
& FL_FLOCK
)
1088 if (flock
->fl_flags
& FL_SLEEP
) {
1089 cFYI(1, "Blocking lock");
1092 if (flock
->fl_flags
& FL_ACCESS
)
1093 cFYI(1, "Process suspended by mandatory locking - "
1094 "not implemented yet");
1095 if (flock
->fl_flags
& FL_LEASE
)
1096 cFYI(1, "Lease on file - not implemented yet");
1097 if (flock
->fl_flags
&
1098 (~(FL_POSIX
| FL_FLOCK
| FL_SLEEP
| FL_ACCESS
| FL_LEASE
)))
1099 cFYI(1, "Unknown lock flags 0x%x", flock
->fl_flags
);
1101 *type
= server
->vals
->large_lock_type
;
1102 if (flock
->fl_type
== F_WRLCK
) {
1103 cFYI(1, "F_WRLCK ");
1104 *type
|= server
->vals
->exclusive_lock_type
;
1106 } else if (flock
->fl_type
== F_UNLCK
) {
1108 *type
|= server
->vals
->unlock_lock_type
;
1110 /* Check if unlock includes more than one lock range */
1111 } else if (flock
->fl_type
== F_RDLCK
) {
1113 *type
|= server
->vals
->shared_lock_type
;
1115 } else if (flock
->fl_type
== F_EXLCK
) {
1117 *type
|= server
->vals
->exclusive_lock_type
;
1119 } else if (flock
->fl_type
== F_SHLCK
) {
1121 *type
|= server
->vals
->shared_lock_type
;
1124 cFYI(1, "Unknown type of lock");
1128 cifs_mandatory_lock(unsigned int xid
, struct cifsFileInfo
*cfile
, __u64 offset
,
1129 __u64 length
, __u32 type
, int lock
, int unlock
, bool wait
)
1131 return CIFSSMBLock(xid
, tlink_tcon(cfile
->tlink
), cfile
->netfid
,
1132 current
->tgid
, length
, offset
, unlock
, lock
,
1133 (__u8
)type
, wait
, 0);
1137 cifs_getlk(struct file
*file
, struct file_lock
*flock
, __u32 type
,
1138 bool wait_flag
, bool posix_lck
, unsigned int xid
)
1141 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1142 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
1143 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1144 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
1145 __u16 netfid
= cfile
->netfid
;
1148 int posix_lock_type
;
1150 rc
= cifs_posix_lock_test(file
, flock
);
1154 if (type
& server
->vals
->shared_lock_type
)
1155 posix_lock_type
= CIFS_RDLCK
;
1157 posix_lock_type
= CIFS_WRLCK
;
1158 rc
= CIFSSMBPosixLock(xid
, tcon
, netfid
, current
->tgid
,
1159 flock
->fl_start
, length
, flock
,
1160 posix_lock_type
, wait_flag
);
1164 rc
= cifs_lock_test(cfile
, flock
->fl_start
, length
, type
, flock
);
1168 /* BB we could chain these into one lock request BB */
1169 rc
= cifs_mandatory_lock(xid
, cfile
, flock
->fl_start
, length
, type
,
1172 rc
= cifs_mandatory_lock(xid
, cfile
, flock
->fl_start
, length
,
1174 flock
->fl_type
= F_UNLCK
;
1176 cERROR(1, "Error unlocking previously locked "
1177 "range %d during test of lock", rc
);
1181 if (type
& server
->vals
->shared_lock_type
) {
1182 flock
->fl_type
= F_WRLCK
;
1186 rc
= cifs_mandatory_lock(xid
, cfile
, flock
->fl_start
, length
,
1187 type
| server
->vals
->shared_lock_type
, 1, 0,
1190 rc
= cifs_mandatory_lock(xid
, cfile
, flock
->fl_start
, length
,
1191 type
| server
->vals
->shared_lock_type
,
1193 flock
->fl_type
= F_RDLCK
;
1195 cERROR(1, "Error unlocking previously locked "
1196 "range %d during test of lock", rc
);
1198 flock
->fl_type
= F_WRLCK
;
1204 cifs_move_llist(struct list_head
*source
, struct list_head
*dest
)
1206 struct list_head
*li
, *tmp
;
1207 list_for_each_safe(li
, tmp
, source
)
1208 list_move(li
, dest
);
1212 cifs_free_llist(struct list_head
*llist
)
1214 struct cifsLockInfo
*li
, *tmp
;
1215 list_for_each_entry_safe(li
, tmp
, llist
, llist
) {
1216 cifs_del_lock_waiters(li
);
1217 list_del(&li
->llist
);
1223 cifs_unlock_range(struct cifsFileInfo
*cfile
, struct file_lock
*flock
,
1226 int rc
= 0, stored_rc
;
1227 int types
[] = {LOCKING_ANDX_LARGE_FILES
,
1228 LOCKING_ANDX_SHARED_LOCK
| LOCKING_ANDX_LARGE_FILES
};
1230 unsigned int max_num
, num
, max_buf
;
1231 LOCKING_ANDX_RANGE
*buf
, *cur
;
1232 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1233 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
1234 struct cifsLockInfo
*li
, *tmp
;
1235 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1236 struct list_head tmp_llist
;
1238 INIT_LIST_HEAD(&tmp_llist
);
1241 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1242 * and check it for zero before using.
1244 max_buf
= tcon
->ses
->server
->maxBuf
;
1248 max_num
= (max_buf
- sizeof(struct smb_hdr
)) /
1249 sizeof(LOCKING_ANDX_RANGE
);
1250 buf
= kzalloc(max_num
* sizeof(LOCKING_ANDX_RANGE
), GFP_KERNEL
);
1254 mutex_lock(&cinode
->lock_mutex
);
1255 for (i
= 0; i
< 2; i
++) {
1258 list_for_each_entry_safe(li
, tmp
, &cfile
->llist
, llist
) {
1259 if (flock
->fl_start
> li
->offset
||
1260 (flock
->fl_start
+ length
) <
1261 (li
->offset
+ li
->length
))
1263 if (current
->tgid
!= li
->pid
)
1265 if (types
[i
] != li
->type
)
1267 if (cinode
->can_cache_brlcks
) {
1269 * We can cache brlock requests - simply remove
1270 * a lock from the file's list.
1272 list_del(&li
->llist
);
1273 cifs_del_lock_waiters(li
);
1277 cur
->Pid
= cpu_to_le16(li
->pid
);
1278 cur
->LengthLow
= cpu_to_le32((u32
)li
->length
);
1279 cur
->LengthHigh
= cpu_to_le32((u32
)(li
->length
>>32));
1280 cur
->OffsetLow
= cpu_to_le32((u32
)li
->offset
);
1281 cur
->OffsetHigh
= cpu_to_le32((u32
)(li
->offset
>>32));
1283 * We need to save a lock here to let us add it again to
1284 * the file's list if the unlock range request fails on
1287 list_move(&li
->llist
, &tmp_llist
);
1288 if (++num
== max_num
) {
1289 stored_rc
= cifs_lockv(xid
, tcon
, cfile
->netfid
,
1290 li
->type
, num
, 0, buf
);
1293 * We failed on the unlock range
1294 * request - add all locks from the tmp
1295 * list to the head of the file's list.
1297 cifs_move_llist(&tmp_llist
,
1302 * The unlock range request succeed -
1303 * free the tmp list.
1305 cifs_free_llist(&tmp_llist
);
1312 stored_rc
= cifs_lockv(xid
, tcon
, cfile
->netfid
,
1313 types
[i
], num
, 0, buf
);
1315 cifs_move_llist(&tmp_llist
, &cfile
->llist
);
1318 cifs_free_llist(&tmp_llist
);
1322 mutex_unlock(&cinode
->lock_mutex
);
1328 cifs_setlk(struct file
*file
, struct file_lock
*flock
, __u32 type
,
1329 bool wait_flag
, bool posix_lck
, int lock
, int unlock
,
1333 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1334 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
1335 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1336 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
1337 __u16 netfid
= cfile
->netfid
;
1340 int posix_lock_type
;
1342 rc
= cifs_posix_lock_set(file
, flock
);
1346 if (type
& server
->vals
->shared_lock_type
)
1347 posix_lock_type
= CIFS_RDLCK
;
1349 posix_lock_type
= CIFS_WRLCK
;
1352 posix_lock_type
= CIFS_UNLCK
;
1354 rc
= CIFSSMBPosixLock(xid
, tcon
, netfid
, current
->tgid
,
1355 flock
->fl_start
, length
, NULL
,
1356 posix_lock_type
, wait_flag
);
1361 struct cifsLockInfo
*lock
;
1363 lock
= cifs_lock_init(flock
->fl_start
, length
, type
);
1367 rc
= cifs_lock_add_if(cfile
, lock
, wait_flag
);
1373 rc
= cifs_mandatory_lock(xid
, cfile
, flock
->fl_start
, length
,
1374 type
, 1, 0, wait_flag
);
1380 cifs_lock_add(cfile
, lock
);
1382 rc
= cifs_unlock_range(cfile
, flock
, xid
);
1385 if (flock
->fl_flags
& FL_POSIX
)
1386 posix_lock_file_wait(file
, flock
);
1390 int cifs_lock(struct file
*file
, int cmd
, struct file_lock
*flock
)
1393 int lock
= 0, unlock
= 0;
1394 bool wait_flag
= false;
1395 bool posix_lck
= false;
1396 struct cifs_sb_info
*cifs_sb
;
1397 struct cifs_tcon
*tcon
;
1398 struct cifsInodeInfo
*cinode
;
1399 struct cifsFileInfo
*cfile
;
1406 cFYI(1, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld "
1407 "end: %lld", cmd
, flock
->fl_flags
, flock
->fl_type
,
1408 flock
->fl_start
, flock
->fl_end
);
1410 cfile
= (struct cifsFileInfo
*)file
->private_data
;
1411 tcon
= tlink_tcon(cfile
->tlink
);
1413 cifs_read_flock(flock
, &type
, &lock
, &unlock
, &wait_flag
,
1416 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
1417 netfid
= cfile
->netfid
;
1418 cinode
= CIFS_I(file
->f_path
.dentry
->d_inode
);
1420 if (cap_unix(tcon
->ses
) &&
1421 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
1422 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
1425 * BB add code here to normalize offset and length to account for
1426 * negative length which we can not accept over the wire.
1428 if (IS_GETLK(cmd
)) {
1429 rc
= cifs_getlk(file
, flock
, type
, wait_flag
, posix_lck
, xid
);
1434 if (!lock
&& !unlock
) {
1436 * if no lock or unlock then nothing to do since we do not
1443 rc
= cifs_setlk(file
, flock
, type
, wait_flag
, posix_lck
, lock
, unlock
,
1450 * update the file size (if needed) after a write. Should be called with
1451 * the inode->i_lock held
1454 cifs_update_eof(struct cifsInodeInfo
*cifsi
, loff_t offset
,
1455 unsigned int bytes_written
)
1457 loff_t end_of_write
= offset
+ bytes_written
;
1459 if (end_of_write
> cifsi
->server_eof
)
1460 cifsi
->server_eof
= end_of_write
;
1463 static ssize_t
cifs_write(struct cifsFileInfo
*open_file
, __u32 pid
,
1464 const char *write_data
, size_t write_size
,
1468 unsigned int bytes_written
= 0;
1469 unsigned int total_written
;
1470 struct cifs_sb_info
*cifs_sb
;
1471 struct cifs_tcon
*pTcon
;
1473 struct dentry
*dentry
= open_file
->dentry
;
1474 struct cifsInodeInfo
*cifsi
= CIFS_I(dentry
->d_inode
);
1475 struct cifs_io_parms io_parms
;
1477 cifs_sb
= CIFS_SB(dentry
->d_sb
);
1479 cFYI(1, "write %zd bytes to offset %lld of %s", write_size
,
1480 *poffset
, dentry
->d_name
.name
);
1482 pTcon
= tlink_tcon(open_file
->tlink
);
1486 for (total_written
= 0; write_size
> total_written
;
1487 total_written
+= bytes_written
) {
1489 while (rc
== -EAGAIN
) {
1493 if (open_file
->invalidHandle
) {
1494 /* we could deadlock if we called
1495 filemap_fdatawait from here so tell
1496 reopen_file not to flush data to
1498 rc
= cifs_reopen_file(open_file
, false);
1503 len
= min((size_t)cifs_sb
->wsize
,
1504 write_size
- total_written
);
1505 /* iov[0] is reserved for smb header */
1506 iov
[1].iov_base
= (char *)write_data
+ total_written
;
1507 iov
[1].iov_len
= len
;
1508 io_parms
.netfid
= open_file
->netfid
;
1510 io_parms
.tcon
= pTcon
;
1511 io_parms
.offset
= *poffset
;
1512 io_parms
.length
= len
;
1513 rc
= CIFSSMBWrite2(xid
, &io_parms
, &bytes_written
, iov
,
1516 if (rc
|| (bytes_written
== 0)) {
1524 spin_lock(&dentry
->d_inode
->i_lock
);
1525 cifs_update_eof(cifsi
, *poffset
, bytes_written
);
1526 spin_unlock(&dentry
->d_inode
->i_lock
);
1527 *poffset
+= bytes_written
;
1531 cifs_stats_bytes_written(pTcon
, total_written
);
1533 if (total_written
> 0) {
1534 spin_lock(&dentry
->d_inode
->i_lock
);
1535 if (*poffset
> dentry
->d_inode
->i_size
)
1536 i_size_write(dentry
->d_inode
, *poffset
);
1537 spin_unlock(&dentry
->d_inode
->i_lock
);
1539 mark_inode_dirty_sync(dentry
->d_inode
);
1541 return total_written
;
1544 struct cifsFileInfo
*find_readable_file(struct cifsInodeInfo
*cifs_inode
,
1547 struct cifsFileInfo
*open_file
= NULL
;
1548 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1550 /* only filter by fsuid on multiuser mounts */
1551 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1554 spin_lock(&cifs_file_list_lock
);
1555 /* we could simply get the first_list_entry since write-only entries
1556 are always at the end of the list but since the first entry might
1557 have a close pending, we go through the whole list */
1558 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1559 if (fsuid_only
&& open_file
->uid
!= current_fsuid())
1561 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_READ
) {
1562 if (!open_file
->invalidHandle
) {
1563 /* found a good file */
1564 /* lock it so it will not be closed on us */
1565 cifsFileInfo_get(open_file
);
1566 spin_unlock(&cifs_file_list_lock
);
1568 } /* else might as well continue, and look for
1569 another, or simply have the caller reopen it
1570 again rather than trying to fix this handle */
1571 } else /* write only file */
1572 break; /* write only files are last so must be done */
1574 spin_unlock(&cifs_file_list_lock
);
1578 struct cifsFileInfo
*find_writable_file(struct cifsInodeInfo
*cifs_inode
,
1581 struct cifsFileInfo
*open_file
, *inv_file
= NULL
;
1582 struct cifs_sb_info
*cifs_sb
;
1583 bool any_available
= false;
1585 unsigned int refind
= 0;
1587 /* Having a null inode here (because mapping->host was set to zero by
1588 the VFS or MM) should not happen but we had reports of on oops (due to
1589 it being zero) during stress testcases so we need to check for it */
1591 if (cifs_inode
== NULL
) {
1592 cERROR(1, "Null inode passed to cifs_writeable_file");
1597 cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1599 /* only filter by fsuid on multiuser mounts */
1600 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1603 spin_lock(&cifs_file_list_lock
);
1605 if (refind
> MAX_REOPEN_ATT
) {
1606 spin_unlock(&cifs_file_list_lock
);
1609 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1610 if (!any_available
&& open_file
->pid
!= current
->tgid
)
1612 if (fsuid_only
&& open_file
->uid
!= current_fsuid())
1614 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
1615 if (!open_file
->invalidHandle
) {
1616 /* found a good writable file */
1617 cifsFileInfo_get(open_file
);
1618 spin_unlock(&cifs_file_list_lock
);
1622 inv_file
= open_file
;
1626 /* couldn't find useable FH with same pid, try any available */
1627 if (!any_available
) {
1628 any_available
= true;
1629 goto refind_writable
;
1633 any_available
= false;
1634 cifsFileInfo_get(inv_file
);
1637 spin_unlock(&cifs_file_list_lock
);
1640 rc
= cifs_reopen_file(inv_file
, false);
1644 spin_lock(&cifs_file_list_lock
);
1645 list_move_tail(&inv_file
->flist
,
1646 &cifs_inode
->openFileList
);
1647 spin_unlock(&cifs_file_list_lock
);
1648 cifsFileInfo_put(inv_file
);
1649 spin_lock(&cifs_file_list_lock
);
1651 goto refind_writable
;
1658 static int cifs_partialpagewrite(struct page
*page
, unsigned from
, unsigned to
)
1660 struct address_space
*mapping
= page
->mapping
;
1661 loff_t offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
1664 int bytes_written
= 0;
1665 struct inode
*inode
;
1666 struct cifsFileInfo
*open_file
;
1668 if (!mapping
|| !mapping
->host
)
1671 inode
= page
->mapping
->host
;
1673 offset
+= (loff_t
)from
;
1674 write_data
= kmap(page
);
1677 if ((to
> PAGE_CACHE_SIZE
) || (from
> to
)) {
1682 /* racing with truncate? */
1683 if (offset
> mapping
->host
->i_size
) {
1685 return 0; /* don't care */
1688 /* check to make sure that we are not extending the file */
1689 if (mapping
->host
->i_size
- offset
< (loff_t
)to
)
1690 to
= (unsigned)(mapping
->host
->i_size
- offset
);
1692 open_file
= find_writable_file(CIFS_I(mapping
->host
), false);
1694 bytes_written
= cifs_write(open_file
, open_file
->pid
,
1695 write_data
, to
- from
, &offset
);
1696 cifsFileInfo_put(open_file
);
1697 /* Does mm or vfs already set times? */
1698 inode
->i_atime
= inode
->i_mtime
= current_fs_time(inode
->i_sb
);
1699 if ((bytes_written
> 0) && (offset
))
1701 else if (bytes_written
< 0)
1704 cFYI(1, "No writeable filehandles for inode");
1713 * Marshal up the iov array, reserving the first one for the header. Also,
1717 cifs_writepages_marshal_iov(struct kvec
*iov
, struct cifs_writedata
*wdata
)
1720 struct inode
*inode
= wdata
->cfile
->dentry
->d_inode
;
1721 loff_t size
= i_size_read(inode
);
1723 /* marshal up the pages into iov array */
1725 for (i
= 0; i
< wdata
->nr_pages
; i
++) {
1726 iov
[i
+ 1].iov_len
= min(size
- page_offset(wdata
->pages
[i
]),
1727 (loff_t
)PAGE_CACHE_SIZE
);
1728 iov
[i
+ 1].iov_base
= kmap(wdata
->pages
[i
]);
1729 wdata
->bytes
+= iov
[i
+ 1].iov_len
;
1733 static int cifs_writepages(struct address_space
*mapping
,
1734 struct writeback_control
*wbc
)
1736 struct cifs_sb_info
*cifs_sb
= CIFS_SB(mapping
->host
->i_sb
);
1737 bool done
= false, scanned
= false, range_whole
= false;
1739 struct cifs_writedata
*wdata
;
1744 * If wsize is smaller than the page cache size, default to writing
1745 * one page at a time via cifs_writepage
1747 if (cifs_sb
->wsize
< PAGE_CACHE_SIZE
)
1748 return generic_writepages(mapping
, wbc
);
1750 if (wbc
->range_cyclic
) {
1751 index
= mapping
->writeback_index
; /* Start from prev offset */
1754 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
1755 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
1756 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
1761 while (!done
&& index
<= end
) {
1762 unsigned int i
, nr_pages
, found_pages
;
1763 pgoff_t next
= 0, tofind
;
1764 struct page
**pages
;
1766 tofind
= min((cifs_sb
->wsize
/ PAGE_CACHE_SIZE
) - 1,
1769 wdata
= cifs_writedata_alloc((unsigned int)tofind
,
1770 cifs_writev_complete
);
1777 * find_get_pages_tag seems to return a max of 256 on each
1778 * iteration, so we must call it several times in order to
1779 * fill the array or the wsize is effectively limited to
1780 * 256 * PAGE_CACHE_SIZE.
1783 pages
= wdata
->pages
;
1785 nr_pages
= find_get_pages_tag(mapping
, &index
,
1786 PAGECACHE_TAG_DIRTY
,
1788 found_pages
+= nr_pages
;
1791 } while (nr_pages
&& tofind
&& index
<= end
);
1793 if (found_pages
== 0) {
1794 kref_put(&wdata
->refcount
, cifs_writedata_release
);
1799 for (i
= 0; i
< found_pages
; i
++) {
1800 page
= wdata
->pages
[i
];
1802 * At this point we hold neither mapping->tree_lock nor
1803 * lock on the page itself: the page may be truncated or
1804 * invalidated (changing page->mapping to NULL), or even
1805 * swizzled back from swapper_space to tmpfs file
1811 else if (!trylock_page(page
))
1814 if (unlikely(page
->mapping
!= mapping
)) {
1819 if (!wbc
->range_cyclic
&& page
->index
> end
) {
1825 if (next
&& (page
->index
!= next
)) {
1826 /* Not next consecutive page */
1831 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
1832 wait_on_page_writeback(page
);
1834 if (PageWriteback(page
) ||
1835 !clear_page_dirty_for_io(page
)) {
1841 * This actually clears the dirty bit in the radix tree.
1842 * See cifs_writepage() for more commentary.
1844 set_page_writeback(page
);
1846 if (page_offset(page
) >= mapping
->host
->i_size
) {
1849 end_page_writeback(page
);
1853 wdata
->pages
[i
] = page
;
1854 next
= page
->index
+ 1;
1858 /* reset index to refind any pages skipped */
1860 index
= wdata
->pages
[0]->index
+ 1;
1862 /* put any pages we aren't going to use */
1863 for (i
= nr_pages
; i
< found_pages
; i
++) {
1864 page_cache_release(wdata
->pages
[i
]);
1865 wdata
->pages
[i
] = NULL
;
1868 /* nothing to write? */
1869 if (nr_pages
== 0) {
1870 kref_put(&wdata
->refcount
, cifs_writedata_release
);
1874 wdata
->sync_mode
= wbc
->sync_mode
;
1875 wdata
->nr_pages
= nr_pages
;
1876 wdata
->offset
= page_offset(wdata
->pages
[0]);
1877 wdata
->marshal_iov
= cifs_writepages_marshal_iov
;
1880 if (wdata
->cfile
!= NULL
)
1881 cifsFileInfo_put(wdata
->cfile
);
1882 wdata
->cfile
= find_writable_file(CIFS_I(mapping
->host
),
1884 if (!wdata
->cfile
) {
1885 cERROR(1, "No writable handles for inode");
1889 wdata
->pid
= wdata
->cfile
->pid
;
1890 rc
= cifs_async_writev(wdata
);
1891 } while (wbc
->sync_mode
== WB_SYNC_ALL
&& rc
== -EAGAIN
);
1893 for (i
= 0; i
< nr_pages
; ++i
)
1894 unlock_page(wdata
->pages
[i
]);
1896 /* send failure -- clean up the mess */
1898 for (i
= 0; i
< nr_pages
; ++i
) {
1900 redirty_page_for_writepage(wbc
,
1903 SetPageError(wdata
->pages
[i
]);
1904 end_page_writeback(wdata
->pages
[i
]);
1905 page_cache_release(wdata
->pages
[i
]);
1908 mapping_set_error(mapping
, rc
);
1910 kref_put(&wdata
->refcount
, cifs_writedata_release
);
1912 wbc
->nr_to_write
-= nr_pages
;
1913 if (wbc
->nr_to_write
<= 0)
1919 if (!scanned
&& !done
) {
1921 * We hit the last page and there is more work to be done: wrap
1922 * back to the start of the file
1929 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
1930 mapping
->writeback_index
= index
;
1936 cifs_writepage_locked(struct page
*page
, struct writeback_control
*wbc
)
1942 /* BB add check for wbc flags */
1943 page_cache_get(page
);
1944 if (!PageUptodate(page
))
1945 cFYI(1, "ppw - page not up to date");
1948 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1950 * A writepage() implementation always needs to do either this,
1951 * or re-dirty the page with "redirty_page_for_writepage()" in
1952 * the case of a failure.
1954 * Just unlocking the page will cause the radix tree tag-bits
1955 * to fail to update with the state of the page correctly.
1957 set_page_writeback(page
);
1959 rc
= cifs_partialpagewrite(page
, 0, PAGE_CACHE_SIZE
);
1960 if (rc
== -EAGAIN
&& wbc
->sync_mode
== WB_SYNC_ALL
)
1962 else if (rc
== -EAGAIN
)
1963 redirty_page_for_writepage(wbc
, page
);
1967 SetPageUptodate(page
);
1968 end_page_writeback(page
);
1969 page_cache_release(page
);
1974 static int cifs_writepage(struct page
*page
, struct writeback_control
*wbc
)
1976 int rc
= cifs_writepage_locked(page
, wbc
);
1981 static int cifs_write_end(struct file
*file
, struct address_space
*mapping
,
1982 loff_t pos
, unsigned len
, unsigned copied
,
1983 struct page
*page
, void *fsdata
)
1986 struct inode
*inode
= mapping
->host
;
1987 struct cifsFileInfo
*cfile
= file
->private_data
;
1988 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cfile
->dentry
->d_sb
);
1991 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
1994 pid
= current
->tgid
;
1996 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1999 if (PageChecked(page
)) {
2001 SetPageUptodate(page
);
2002 ClearPageChecked(page
);
2003 } else if (!PageUptodate(page
) && copied
== PAGE_CACHE_SIZE
)
2004 SetPageUptodate(page
);
2006 if (!PageUptodate(page
)) {
2008 unsigned offset
= pos
& (PAGE_CACHE_SIZE
- 1);
2012 /* this is probably better than directly calling
2013 partialpage_write since in this function the file handle is
2014 known which we might as well leverage */
2015 /* BB check if anything else missing out of ppw
2016 such as updating last write time */
2017 page_data
= kmap(page
);
2018 rc
= cifs_write(cfile
, pid
, page_data
+ offset
, copied
, &pos
);
2019 /* if (rc < 0) should we set writebehind rc? */
2026 set_page_dirty(page
);
2030 spin_lock(&inode
->i_lock
);
2031 if (pos
> inode
->i_size
)
2032 i_size_write(inode
, pos
);
2033 spin_unlock(&inode
->i_lock
);
2037 page_cache_release(page
);
2042 int cifs_strict_fsync(struct file
*file
, loff_t start
, loff_t end
,
2047 struct cifs_tcon
*tcon
;
2048 struct cifsFileInfo
*smbfile
= file
->private_data
;
2049 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
2050 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
2052 rc
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
2055 mutex_lock(&inode
->i_mutex
);
2059 cFYI(1, "Sync file - name: %s datasync: 0x%x",
2060 file
->f_path
.dentry
->d_name
.name
, datasync
);
2062 if (!CIFS_I(inode
)->clientCanCacheRead
) {
2063 rc
= cifs_invalidate_mapping(inode
);
2065 cFYI(1, "rc: %d during invalidate phase", rc
);
2066 rc
= 0; /* don't care about it in fsync */
2070 tcon
= tlink_tcon(smbfile
->tlink
);
2071 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOSSYNC
))
2072 rc
= CIFSSMBFlush(xid
, tcon
, smbfile
->netfid
);
2075 mutex_unlock(&inode
->i_mutex
);
2079 int cifs_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
2083 struct cifs_tcon
*tcon
;
2084 struct cifsFileInfo
*smbfile
= file
->private_data
;
2085 struct cifs_sb_info
*cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
2086 struct inode
*inode
= file
->f_mapping
->host
;
2088 rc
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
2091 mutex_lock(&inode
->i_mutex
);
2095 cFYI(1, "Sync file - name: %s datasync: 0x%x",
2096 file
->f_path
.dentry
->d_name
.name
, datasync
);
2098 tcon
= tlink_tcon(smbfile
->tlink
);
2099 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOSSYNC
))
2100 rc
= CIFSSMBFlush(xid
, tcon
, smbfile
->netfid
);
2103 mutex_unlock(&inode
->i_mutex
);
2108 * As file closes, flush all cached write data for this inode checking
2109 * for write behind errors.
2111 int cifs_flush(struct file
*file
, fl_owner_t id
)
2113 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
2116 if (file
->f_mode
& FMODE_WRITE
)
2117 rc
= filemap_write_and_wait(inode
->i_mapping
);
2119 cFYI(1, "Flush inode %p file %p rc %d", inode
, file
, rc
);
2125 cifs_write_allocate_pages(struct page
**pages
, unsigned long num_pages
)
2130 for (i
= 0; i
< num_pages
; i
++) {
2131 pages
[i
] = alloc_page(GFP_KERNEL
|__GFP_HIGHMEM
);
2134 * save number of pages we have already allocated and
2135 * return with ENOMEM error
2144 for (i
= 0; i
< num_pages
; i
++)
2151 size_t get_numpages(const size_t wsize
, const size_t len
, size_t *cur_len
)
2156 clen
= min_t(const size_t, len
, wsize
);
2157 num_pages
= DIV_ROUND_UP(clen
, PAGE_SIZE
);
2166 cifs_uncached_marshal_iov(struct kvec
*iov
, struct cifs_writedata
*wdata
)
2169 size_t bytes
= wdata
->bytes
;
2171 /* marshal up the pages into iov array */
2172 for (i
= 0; i
< wdata
->nr_pages
; i
++) {
2173 iov
[i
+ 1].iov_len
= min_t(size_t, bytes
, PAGE_SIZE
);
2174 iov
[i
+ 1].iov_base
= kmap(wdata
->pages
[i
]);
2175 bytes
-= iov
[i
+ 1].iov_len
;
2180 cifs_uncached_writev_complete(struct work_struct
*work
)
2183 struct cifs_writedata
*wdata
= container_of(work
,
2184 struct cifs_writedata
, work
);
2185 struct inode
*inode
= wdata
->cfile
->dentry
->d_inode
;
2186 struct cifsInodeInfo
*cifsi
= CIFS_I(inode
);
2188 spin_lock(&inode
->i_lock
);
2189 cifs_update_eof(cifsi
, wdata
->offset
, wdata
->bytes
);
2190 if (cifsi
->server_eof
> inode
->i_size
)
2191 i_size_write(inode
, cifsi
->server_eof
);
2192 spin_unlock(&inode
->i_lock
);
2194 complete(&wdata
->done
);
2196 if (wdata
->result
!= -EAGAIN
) {
2197 for (i
= 0; i
< wdata
->nr_pages
; i
++)
2198 put_page(wdata
->pages
[i
]);
2201 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2204 /* attempt to send write to server, retry on any -EAGAIN errors */
2206 cifs_uncached_retry_writev(struct cifs_writedata
*wdata
)
2211 if (wdata
->cfile
->invalidHandle
) {
2212 rc
= cifs_reopen_file(wdata
->cfile
, false);
2216 rc
= cifs_async_writev(wdata
);
2217 } while (rc
== -EAGAIN
);
2223 cifs_iovec_write(struct file
*file
, const struct iovec
*iov
,
2224 unsigned long nr_segs
, loff_t
*poffset
)
2226 unsigned long nr_pages
, i
;
2227 size_t copied
, len
, cur_len
;
2228 ssize_t total_written
= 0;
2231 struct cifsFileInfo
*open_file
;
2232 struct cifs_tcon
*tcon
;
2233 struct cifs_sb_info
*cifs_sb
;
2234 struct cifs_writedata
*wdata
, *tmp
;
2235 struct list_head wdata_list
;
2239 len
= iov_length(iov
, nr_segs
);
2243 rc
= generic_write_checks(file
, poffset
, &len
, 0);
2247 INIT_LIST_HEAD(&wdata_list
);
2248 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
2249 open_file
= file
->private_data
;
2250 tcon
= tlink_tcon(open_file
->tlink
);
2253 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2254 pid
= open_file
->pid
;
2256 pid
= current
->tgid
;
2258 iov_iter_init(&it
, iov
, nr_segs
, len
, 0);
2262 nr_pages
= get_numpages(cifs_sb
->wsize
, len
, &cur_len
);
2263 wdata
= cifs_writedata_alloc(nr_pages
,
2264 cifs_uncached_writev_complete
);
2270 rc
= cifs_write_allocate_pages(wdata
->pages
, nr_pages
);
2277 for (i
= 0; i
< nr_pages
; i
++) {
2278 copied
= min_t(const size_t, cur_len
, PAGE_SIZE
);
2279 copied
= iov_iter_copy_from_user(wdata
->pages
[i
], &it
,
2282 iov_iter_advance(&it
, copied
);
2284 cur_len
= save_len
- cur_len
;
2286 wdata
->sync_mode
= WB_SYNC_ALL
;
2287 wdata
->nr_pages
= nr_pages
;
2288 wdata
->offset
= (__u64
)offset
;
2289 wdata
->cfile
= cifsFileInfo_get(open_file
);
2291 wdata
->bytes
= cur_len
;
2292 wdata
->marshal_iov
= cifs_uncached_marshal_iov
;
2293 rc
= cifs_uncached_retry_writev(wdata
);
2295 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2299 list_add_tail(&wdata
->list
, &wdata_list
);
2305 * If at least one write was successfully sent, then discard any rc
2306 * value from the later writes. If the other write succeeds, then
2307 * we'll end up returning whatever was written. If it fails, then
2308 * we'll get a new rc value from that.
2310 if (!list_empty(&wdata_list
))
2314 * Wait for and collect replies for any successful sends in order of
2315 * increasing offset. Once an error is hit or we get a fatal signal
2316 * while waiting, then return without waiting for any more replies.
2319 list_for_each_entry_safe(wdata
, tmp
, &wdata_list
, list
) {
2321 /* FIXME: freezable too? */
2322 rc
= wait_for_completion_killable(&wdata
->done
);
2325 else if (wdata
->result
)
2328 total_written
+= wdata
->bytes
;
2330 /* resend call if it's a retryable error */
2331 if (rc
== -EAGAIN
) {
2332 rc
= cifs_uncached_retry_writev(wdata
);
2336 list_del_init(&wdata
->list
);
2337 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2340 if (total_written
> 0)
2341 *poffset
+= total_written
;
2343 cifs_stats_bytes_written(tcon
, total_written
);
2344 return total_written
? total_written
: (ssize_t
)rc
;
2347 ssize_t
cifs_user_writev(struct kiocb
*iocb
, const struct iovec
*iov
,
2348 unsigned long nr_segs
, loff_t pos
)
2351 struct inode
*inode
;
2353 inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2356 * BB - optimize the way when signing is disabled. We can drop this
2357 * extra memory-to-memory copying and use iovec buffers for constructing
2361 written
= cifs_iovec_write(iocb
->ki_filp
, iov
, nr_segs
, &pos
);
2363 CIFS_I(inode
)->invalid_mapping
= true;
2370 ssize_t
cifs_strict_writev(struct kiocb
*iocb
, const struct iovec
*iov
,
2371 unsigned long nr_segs
, loff_t pos
)
2373 struct inode
*inode
;
2375 inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2377 if (CIFS_I(inode
)->clientCanCacheAll
)
2378 return generic_file_aio_write(iocb
, iov
, nr_segs
, pos
);
2381 * In strict cache mode we need to write the data to the server exactly
2382 * from the pos to pos+len-1 rather than flush all affected pages
2383 * because it may cause a error with mandatory locks on these pages but
2384 * not on the region from pos to ppos+len-1.
2387 return cifs_user_writev(iocb
, iov
, nr_segs
, pos
);
2390 static struct cifs_readdata
*
2391 cifs_readdata_alloc(unsigned int nr_vecs
, work_func_t complete
)
2393 struct cifs_readdata
*rdata
;
2395 rdata
= kzalloc(sizeof(*rdata
) +
2396 sizeof(struct kvec
) * nr_vecs
, GFP_KERNEL
);
2397 if (rdata
!= NULL
) {
2398 kref_init(&rdata
->refcount
);
2399 INIT_LIST_HEAD(&rdata
->list
);
2400 init_completion(&rdata
->done
);
2401 INIT_WORK(&rdata
->work
, complete
);
2402 INIT_LIST_HEAD(&rdata
->pages
);
2408 cifs_readdata_release(struct kref
*refcount
)
2410 struct cifs_readdata
*rdata
= container_of(refcount
,
2411 struct cifs_readdata
, refcount
);
2414 cifsFileInfo_put(rdata
->cfile
);
2420 cifs_read_allocate_pages(struct list_head
*list
, unsigned int npages
)
2423 struct page
*page
, *tpage
;
2426 for (i
= 0; i
< npages
; i
++) {
2427 page
= alloc_page(GFP_KERNEL
|__GFP_HIGHMEM
);
2432 list_add(&page
->lru
, list
);
2436 list_for_each_entry_safe(page
, tpage
, list
, lru
) {
2437 list_del(&page
->lru
);
2445 cifs_uncached_readdata_release(struct kref
*refcount
)
2447 struct page
*page
, *tpage
;
2448 struct cifs_readdata
*rdata
= container_of(refcount
,
2449 struct cifs_readdata
, refcount
);
2451 list_for_each_entry_safe(page
, tpage
, &rdata
->pages
, lru
) {
2452 list_del(&page
->lru
);
2455 cifs_readdata_release(refcount
);
2459 cifs_retry_async_readv(struct cifs_readdata
*rdata
)
2464 if (rdata
->cfile
->invalidHandle
) {
2465 rc
= cifs_reopen_file(rdata
->cfile
, true);
2469 rc
= cifs_async_readv(rdata
);
2470 } while (rc
== -EAGAIN
);
2476 * cifs_readdata_to_iov - copy data from pages in response to an iovec
2477 * @rdata: the readdata response with list of pages holding data
2478 * @iov: vector in which we should copy the data
2479 * @nr_segs: number of segments in vector
2480 * @offset: offset into file of the first iovec
2481 * @copied: used to return the amount of data copied to the iov
2483 * This function copies data from a list of pages in a readdata response into
2484 * an array of iovecs. It will first calculate where the data should go
2485 * based on the info in the readdata and then copy the data into that spot.
2488 cifs_readdata_to_iov(struct cifs_readdata
*rdata
, const struct iovec
*iov
,
2489 unsigned long nr_segs
, loff_t offset
, ssize_t
*copied
)
2493 size_t pos
= rdata
->offset
- offset
;
2494 struct page
*page
, *tpage
;
2495 ssize_t remaining
= rdata
->bytes
;
2496 unsigned char *pdata
;
2498 /* set up iov_iter and advance to the correct offset */
2499 iov_iter_init(&ii
, iov
, nr_segs
, iov_length(iov
, nr_segs
), 0);
2500 iov_iter_advance(&ii
, pos
);
2503 list_for_each_entry_safe(page
, tpage
, &rdata
->pages
, lru
) {
2506 /* copy a whole page or whatever's left */
2507 copy
= min_t(ssize_t
, remaining
, PAGE_SIZE
);
2509 /* ...but limit it to whatever space is left in the iov */
2510 copy
= min_t(ssize_t
, copy
, iov_iter_count(&ii
));
2512 /* go while there's data to be copied and no errors */
2515 rc
= memcpy_toiovecend(ii
.iov
, pdata
, ii
.iov_offset
,
2521 iov_iter_advance(&ii
, copy
);
2525 list_del(&page
->lru
);
2533 cifs_uncached_readv_complete(struct work_struct
*work
)
2535 struct cifs_readdata
*rdata
= container_of(work
,
2536 struct cifs_readdata
, work
);
2538 /* if the result is non-zero then the pages weren't kmapped */
2539 if (rdata
->result
== 0) {
2542 list_for_each_entry(page
, &rdata
->pages
, lru
)
2546 complete(&rdata
->done
);
2547 kref_put(&rdata
->refcount
, cifs_uncached_readdata_release
);
2551 cifs_uncached_read_marshal_iov(struct cifs_readdata
*rdata
,
2552 unsigned int remaining
)
2555 struct page
*page
, *tpage
;
2558 list_for_each_entry_safe(page
, tpage
, &rdata
->pages
, lru
) {
2559 if (remaining
>= PAGE_SIZE
) {
2560 /* enough data to fill the page */
2561 rdata
->iov
[rdata
->nr_iov
].iov_base
= kmap(page
);
2562 rdata
->iov
[rdata
->nr_iov
].iov_len
= PAGE_SIZE
;
2563 cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
2564 rdata
->nr_iov
, page
->index
,
2565 rdata
->iov
[rdata
->nr_iov
].iov_base
,
2566 rdata
->iov
[rdata
->nr_iov
].iov_len
);
2569 remaining
-= PAGE_SIZE
;
2570 } else if (remaining
> 0) {
2571 /* enough for partial page, fill and zero the rest */
2572 rdata
->iov
[rdata
->nr_iov
].iov_base
= kmap(page
);
2573 rdata
->iov
[rdata
->nr_iov
].iov_len
= remaining
;
2574 cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
2575 rdata
->nr_iov
, page
->index
,
2576 rdata
->iov
[rdata
->nr_iov
].iov_base
,
2577 rdata
->iov
[rdata
->nr_iov
].iov_len
);
2578 memset(rdata
->iov
[rdata
->nr_iov
].iov_base
+ remaining
,
2579 '\0', PAGE_SIZE
- remaining
);
2584 /* no need to hold page hostage */
2585 list_del(&page
->lru
);
2594 cifs_iovec_read(struct file
*file
, const struct iovec
*iov
,
2595 unsigned long nr_segs
, loff_t
*poffset
)
2598 size_t len
, cur_len
;
2599 ssize_t total_read
= 0;
2600 loff_t offset
= *poffset
;
2601 unsigned int npages
;
2602 struct cifs_sb_info
*cifs_sb
;
2603 struct cifs_tcon
*tcon
;
2604 struct cifsFileInfo
*open_file
;
2605 struct cifs_readdata
*rdata
, *tmp
;
2606 struct list_head rdata_list
;
2612 len
= iov_length(iov
, nr_segs
);
2616 INIT_LIST_HEAD(&rdata_list
);
2617 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
2618 open_file
= file
->private_data
;
2619 tcon
= tlink_tcon(open_file
->tlink
);
2621 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2622 pid
= open_file
->pid
;
2624 pid
= current
->tgid
;
2626 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
2627 cFYI(1, "attempting read on write only file instance");
2630 cur_len
= min_t(const size_t, len
- total_read
, cifs_sb
->rsize
);
2631 npages
= DIV_ROUND_UP(cur_len
, PAGE_SIZE
);
2633 /* allocate a readdata struct */
2634 rdata
= cifs_readdata_alloc(npages
,
2635 cifs_uncached_readv_complete
);
2641 rc
= cifs_read_allocate_pages(&rdata
->pages
, npages
);
2645 rdata
->cfile
= cifsFileInfo_get(open_file
);
2646 rdata
->offset
= offset
;
2647 rdata
->bytes
= cur_len
;
2649 rdata
->marshal_iov
= cifs_uncached_read_marshal_iov
;
2651 rc
= cifs_retry_async_readv(rdata
);
2654 kref_put(&rdata
->refcount
,
2655 cifs_uncached_readdata_release
);
2659 list_add_tail(&rdata
->list
, &rdata_list
);
2664 /* if at least one read request send succeeded, then reset rc */
2665 if (!list_empty(&rdata_list
))
2668 /* the loop below should proceed in the order of increasing offsets */
2670 list_for_each_entry_safe(rdata
, tmp
, &rdata_list
, list
) {
2674 /* FIXME: freezable sleep too? */
2675 rc
= wait_for_completion_killable(&rdata
->done
);
2678 else if (rdata
->result
)
2681 rc
= cifs_readdata_to_iov(rdata
, iov
,
2684 total_read
+= copied
;
2687 /* resend call if it's a retryable error */
2688 if (rc
== -EAGAIN
) {
2689 rc
= cifs_retry_async_readv(rdata
);
2693 list_del_init(&rdata
->list
);
2694 kref_put(&rdata
->refcount
, cifs_uncached_readdata_release
);
2697 cifs_stats_bytes_read(tcon
, total_read
);
2698 *poffset
+= total_read
;
2700 return total_read
? total_read
: rc
;
2703 ssize_t
cifs_user_readv(struct kiocb
*iocb
, const struct iovec
*iov
,
2704 unsigned long nr_segs
, loff_t pos
)
2708 read
= cifs_iovec_read(iocb
->ki_filp
, iov
, nr_segs
, &pos
);
2715 ssize_t
cifs_strict_readv(struct kiocb
*iocb
, const struct iovec
*iov
,
2716 unsigned long nr_segs
, loff_t pos
)
2718 struct inode
*inode
;
2720 inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2722 if (CIFS_I(inode
)->clientCanCacheRead
)
2723 return generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
2726 * In strict cache mode we need to read from the server all the time
2727 * if we don't have level II oplock because the server can delay mtime
2728 * change - so we can't make a decision about inode invalidating.
2729 * And we can also fail with pagereading if there are mandatory locks
2730 * on pages affected by this read but not on the region from pos to
2734 return cifs_user_readv(iocb
, iov
, nr_segs
, pos
);
2737 static ssize_t
cifs_read(struct file
*file
, char *read_data
, size_t read_size
,
2741 unsigned int bytes_read
= 0;
2742 unsigned int total_read
;
2743 unsigned int current_read_size
;
2745 struct cifs_sb_info
*cifs_sb
;
2746 struct cifs_tcon
*tcon
;
2748 char *current_offset
;
2749 struct cifsFileInfo
*open_file
;
2750 struct cifs_io_parms io_parms
;
2751 int buf_type
= CIFS_NO_BUFFER
;
2755 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
2757 /* FIXME: set up handlers for larger reads and/or convert to async */
2758 rsize
= min_t(unsigned int, cifs_sb
->rsize
, CIFSMaxBufSize
);
2760 if (file
->private_data
== NULL
) {
2765 open_file
= file
->private_data
;
2766 tcon
= tlink_tcon(open_file
->tlink
);
2768 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2769 pid
= open_file
->pid
;
2771 pid
= current
->tgid
;
2773 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
2774 cFYI(1, "attempting read on write only file instance");
2776 for (total_read
= 0, current_offset
= read_data
;
2777 read_size
> total_read
;
2778 total_read
+= bytes_read
, current_offset
+= bytes_read
) {
2779 current_read_size
= min_t(uint
, read_size
- total_read
, rsize
);
2781 * For windows me and 9x we do not want to request more than it
2782 * negotiated since it will refuse the read then.
2784 if ((tcon
->ses
) && !(tcon
->ses
->capabilities
&
2785 tcon
->ses
->server
->vals
->cap_large_files
)) {
2786 current_read_size
= min_t(uint
, current_read_size
,
2790 while (rc
== -EAGAIN
) {
2791 if (open_file
->invalidHandle
) {
2792 rc
= cifs_reopen_file(open_file
, true);
2796 io_parms
.netfid
= open_file
->netfid
;
2798 io_parms
.tcon
= tcon
;
2799 io_parms
.offset
= *poffset
;
2800 io_parms
.length
= current_read_size
;
2801 rc
= CIFSSMBRead(xid
, &io_parms
, &bytes_read
,
2802 ¤t_offset
, &buf_type
);
2804 if (rc
|| (bytes_read
== 0)) {
2812 cifs_stats_bytes_read(tcon
, total_read
);
2813 *poffset
+= bytes_read
;
2821 * If the page is mmap'ed into a process' page tables, then we need to make
2822 * sure that it doesn't change while being written back.
2825 cifs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2827 struct page
*page
= vmf
->page
;
2830 return VM_FAULT_LOCKED
;
2833 static struct vm_operations_struct cifs_file_vm_ops
= {
2834 .fault
= filemap_fault
,
2835 .page_mkwrite
= cifs_page_mkwrite
,
2838 int cifs_file_strict_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2841 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
2845 if (!CIFS_I(inode
)->clientCanCacheRead
) {
2846 rc
= cifs_invalidate_mapping(inode
);
2851 rc
= generic_file_mmap(file
, vma
);
2853 vma
->vm_ops
= &cifs_file_vm_ops
;
2858 int cifs_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2863 rc
= cifs_revalidate_file(file
);
2865 cFYI(1, "Validation prior to mmap failed, error=%d", rc
);
2869 rc
= generic_file_mmap(file
, vma
);
2871 vma
->vm_ops
= &cifs_file_vm_ops
;
2877 cifs_readv_complete(struct work_struct
*work
)
2879 struct cifs_readdata
*rdata
= container_of(work
,
2880 struct cifs_readdata
, work
);
2881 struct page
*page
, *tpage
;
2883 list_for_each_entry_safe(page
, tpage
, &rdata
->pages
, lru
) {
2884 list_del(&page
->lru
);
2885 lru_cache_add_file(page
);
2887 if (rdata
->result
== 0) {
2889 flush_dcache_page(page
);
2890 SetPageUptodate(page
);
2895 if (rdata
->result
== 0)
2896 cifs_readpage_to_fscache(rdata
->mapping
->host
, page
);
2898 page_cache_release(page
);
2900 kref_put(&rdata
->refcount
, cifs_readdata_release
);
2904 cifs_readpages_marshal_iov(struct cifs_readdata
*rdata
, unsigned int remaining
)
2907 struct page
*page
, *tpage
;
2911 /* determine the eof that the server (probably) has */
2912 eof
= CIFS_I(rdata
->mapping
->host
)->server_eof
;
2913 eof_index
= eof
? (eof
- 1) >> PAGE_CACHE_SHIFT
: 0;
2914 cFYI(1, "eof=%llu eof_index=%lu", eof
, eof_index
);
2917 list_for_each_entry_safe(page
, tpage
, &rdata
->pages
, lru
) {
2918 if (remaining
>= PAGE_CACHE_SIZE
) {
2919 /* enough data to fill the page */
2920 rdata
->iov
[rdata
->nr_iov
].iov_base
= kmap(page
);
2921 rdata
->iov
[rdata
->nr_iov
].iov_len
= PAGE_CACHE_SIZE
;
2922 cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
2923 rdata
->nr_iov
, page
->index
,
2924 rdata
->iov
[rdata
->nr_iov
].iov_base
,
2925 rdata
->iov
[rdata
->nr_iov
].iov_len
);
2927 len
+= PAGE_CACHE_SIZE
;
2928 remaining
-= PAGE_CACHE_SIZE
;
2929 } else if (remaining
> 0) {
2930 /* enough for partial page, fill and zero the rest */
2931 rdata
->iov
[rdata
->nr_iov
].iov_base
= kmap(page
);
2932 rdata
->iov
[rdata
->nr_iov
].iov_len
= remaining
;
2933 cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
2934 rdata
->nr_iov
, page
->index
,
2935 rdata
->iov
[rdata
->nr_iov
].iov_base
,
2936 rdata
->iov
[rdata
->nr_iov
].iov_len
);
2937 memset(rdata
->iov
[rdata
->nr_iov
].iov_base
+ remaining
,
2938 '\0', PAGE_CACHE_SIZE
- remaining
);
2942 } else if (page
->index
> eof_index
) {
2944 * The VFS will not try to do readahead past the
2945 * i_size, but it's possible that we have outstanding
2946 * writes with gaps in the middle and the i_size hasn't
2947 * caught up yet. Populate those with zeroed out pages
2948 * to prevent the VFS from repeatedly attempting to
2949 * fill them until the writes are flushed.
2951 zero_user(page
, 0, PAGE_CACHE_SIZE
);
2952 list_del(&page
->lru
);
2953 lru_cache_add_file(page
);
2954 flush_dcache_page(page
);
2955 SetPageUptodate(page
);
2957 page_cache_release(page
);
2959 /* no need to hold page hostage */
2960 list_del(&page
->lru
);
2961 lru_cache_add_file(page
);
2963 page_cache_release(page
);
2970 static int cifs_readpages(struct file
*file
, struct address_space
*mapping
,
2971 struct list_head
*page_list
, unsigned num_pages
)
2974 struct list_head tmplist
;
2975 struct cifsFileInfo
*open_file
= file
->private_data
;
2976 struct cifs_sb_info
*cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
2977 unsigned int rsize
= cifs_sb
->rsize
;
2981 * Give up immediately if rsize is too small to read an entire page.
2982 * The VFS will fall back to readpage. We should never reach this
2983 * point however since we set ra_pages to 0 when the rsize is smaller
2984 * than a cache page.
2986 if (unlikely(rsize
< PAGE_CACHE_SIZE
))
2990 * Reads as many pages as possible from fscache. Returns -ENOBUFS
2991 * immediately if the cookie is negative
2993 rc
= cifs_readpages_from_fscache(mapping
->host
, mapping
, page_list
,
2998 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2999 pid
= open_file
->pid
;
3001 pid
= current
->tgid
;
3004 INIT_LIST_HEAD(&tmplist
);
3006 cFYI(1, "%s: file=%p mapping=%p num_pages=%u", __func__
, file
,
3007 mapping
, num_pages
);
3010 * Start with the page at end of list and move it to private
3011 * list. Do the same with any following pages until we hit
3012 * the rsize limit, hit an index discontinuity, or run out of
3013 * pages. Issue the async read and then start the loop again
3014 * until the list is empty.
3016 * Note that list order is important. The page_list is in
3017 * the order of declining indexes. When we put the pages in
3018 * the rdata->pages, then we want them in increasing order.
3020 while (!list_empty(page_list
)) {
3021 unsigned int bytes
= PAGE_CACHE_SIZE
;
3022 unsigned int expected_index
;
3023 unsigned int nr_pages
= 1;
3025 struct page
*page
, *tpage
;
3026 struct cifs_readdata
*rdata
;
3028 page
= list_entry(page_list
->prev
, struct page
, lru
);
3031 * Lock the page and put it in the cache. Since no one else
3032 * should have access to this page, we're safe to simply set
3033 * PG_locked without checking it first.
3035 __set_page_locked(page
);
3036 rc
= add_to_page_cache_locked(page
, mapping
,
3037 page
->index
, GFP_KERNEL
);
3039 /* give up if we can't stick it in the cache */
3041 __clear_page_locked(page
);
3045 /* move first page to the tmplist */
3046 offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
3047 list_move_tail(&page
->lru
, &tmplist
);
3049 /* now try and add more pages onto the request */
3050 expected_index
= page
->index
+ 1;
3051 list_for_each_entry_safe_reverse(page
, tpage
, page_list
, lru
) {
3052 /* discontinuity ? */
3053 if (page
->index
!= expected_index
)
3056 /* would this page push the read over the rsize? */
3057 if (bytes
+ PAGE_CACHE_SIZE
> rsize
)
3060 __set_page_locked(page
);
3061 if (add_to_page_cache_locked(page
, mapping
,
3062 page
->index
, GFP_KERNEL
)) {
3063 __clear_page_locked(page
);
3066 list_move_tail(&page
->lru
, &tmplist
);
3067 bytes
+= PAGE_CACHE_SIZE
;
3072 rdata
= cifs_readdata_alloc(nr_pages
, cifs_readv_complete
);
3074 /* best to give up if we're out of mem */
3075 list_for_each_entry_safe(page
, tpage
, &tmplist
, lru
) {
3076 list_del(&page
->lru
);
3077 lru_cache_add_file(page
);
3079 page_cache_release(page
);
3085 spin_lock(&cifs_file_list_lock
);
3086 spin_unlock(&cifs_file_list_lock
);
3087 rdata
->cfile
= cifsFileInfo_get(open_file
);
3088 rdata
->mapping
= mapping
;
3089 rdata
->offset
= offset
;
3090 rdata
->bytes
= bytes
;
3092 rdata
->marshal_iov
= cifs_readpages_marshal_iov
;
3093 list_splice_init(&tmplist
, &rdata
->pages
);
3095 rc
= cifs_retry_async_readv(rdata
);
3097 list_for_each_entry_safe(page
, tpage
, &rdata
->pages
,
3099 list_del(&page
->lru
);
3100 lru_cache_add_file(page
);
3102 page_cache_release(page
);
3104 kref_put(&rdata
->refcount
, cifs_readdata_release
);
3108 kref_put(&rdata
->refcount
, cifs_readdata_release
);
3114 static int cifs_readpage_worker(struct file
*file
, struct page
*page
,
3120 /* Is the page cached? */
3121 rc
= cifs_readpage_from_fscache(file
->f_path
.dentry
->d_inode
, page
);
3125 page_cache_get(page
);
3126 read_data
= kmap(page
);
3127 /* for reads over a certain size could initiate async read ahead */
3129 rc
= cifs_read(file
, read_data
, PAGE_CACHE_SIZE
, poffset
);
3134 cFYI(1, "Bytes read %d", rc
);
3136 file
->f_path
.dentry
->d_inode
->i_atime
=
3137 current_fs_time(file
->f_path
.dentry
->d_inode
->i_sb
);
3139 if (PAGE_CACHE_SIZE
> rc
)
3140 memset(read_data
+ rc
, 0, PAGE_CACHE_SIZE
- rc
);
3142 flush_dcache_page(page
);
3143 SetPageUptodate(page
);
3145 /* send this page to the cache */
3146 cifs_readpage_to_fscache(file
->f_path
.dentry
->d_inode
, page
);
3152 page_cache_release(page
);
3158 static int cifs_readpage(struct file
*file
, struct page
*page
)
3160 loff_t offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
3166 if (file
->private_data
== NULL
) {
3172 cFYI(1, "readpage %p at offset %d 0x%x",
3173 page
, (int)offset
, (int)offset
);
3175 rc
= cifs_readpage_worker(file
, page
, &offset
);
3183 static int is_inode_writable(struct cifsInodeInfo
*cifs_inode
)
3185 struct cifsFileInfo
*open_file
;
3187 spin_lock(&cifs_file_list_lock
);
3188 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
3189 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
3190 spin_unlock(&cifs_file_list_lock
);
3194 spin_unlock(&cifs_file_list_lock
);
3198 /* We do not want to update the file size from server for inodes
3199 open for write - to avoid races with writepage extending
3200 the file - in the future we could consider allowing
3201 refreshing the inode only on increases in the file size
3202 but this is tricky to do without racing with writebehind
3203 page caching in the current Linux kernel design */
3204 bool is_size_safe_to_change(struct cifsInodeInfo
*cifsInode
, __u64 end_of_file
)
3209 if (is_inode_writable(cifsInode
)) {
3210 /* This inode is open for write at least once */
3211 struct cifs_sb_info
*cifs_sb
;
3213 cifs_sb
= CIFS_SB(cifsInode
->vfs_inode
.i_sb
);
3214 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_DIRECT_IO
) {
3215 /* since no page cache to corrupt on directio
3216 we can change size safely */
3220 if (i_size_read(&cifsInode
->vfs_inode
) < end_of_file
)
3228 static int cifs_write_begin(struct file
*file
, struct address_space
*mapping
,
3229 loff_t pos
, unsigned len
, unsigned flags
,
3230 struct page
**pagep
, void **fsdata
)
3232 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
3233 loff_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
3234 loff_t page_start
= pos
& PAGE_MASK
;
3239 cFYI(1, "write_begin from %lld len %d", (long long)pos
, len
);
3241 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3247 if (PageUptodate(page
))
3251 * If we write a full page it will be up to date, no need to read from
3252 * the server. If the write is short, we'll end up doing a sync write
3255 if (len
== PAGE_CACHE_SIZE
)
3259 * optimize away the read when we have an oplock, and we're not
3260 * expecting to use any of the data we'd be reading in. That
3261 * is, when the page lies beyond the EOF, or straddles the EOF
3262 * and the write will cover all of the existing data.
3264 if (CIFS_I(mapping
->host
)->clientCanCacheRead
) {
3265 i_size
= i_size_read(mapping
->host
);
3266 if (page_start
>= i_size
||
3267 (offset
== 0 && (pos
+ len
) >= i_size
)) {
3268 zero_user_segments(page
, 0, offset
,
3272 * PageChecked means that the parts of the page
3273 * to which we're not writing are considered up
3274 * to date. Once the data is copied to the
3275 * page, it can be set uptodate.
3277 SetPageChecked(page
);
3282 if ((file
->f_flags
& O_ACCMODE
) != O_WRONLY
) {
3284 * might as well read a page, it is fast enough. If we get
3285 * an error, we don't need to return it. cifs_write_end will
3286 * do a sync write instead since PG_uptodate isn't set.
3288 cifs_readpage_worker(file
, page
, &page_start
);
3290 /* we could try using another file handle if there is one -
3291 but how would we lock it to prevent close of that handle
3292 racing with this read? In any case
3293 this will be written out by write_end so is fine */
3300 static int cifs_release_page(struct page
*page
, gfp_t gfp
)
3302 if (PagePrivate(page
))
3305 return cifs_fscache_release_page(page
, gfp
);
3308 static void cifs_invalidate_page(struct page
*page
, unsigned long offset
)
3310 struct cifsInodeInfo
*cifsi
= CIFS_I(page
->mapping
->host
);
3313 cifs_fscache_invalidate_page(page
, &cifsi
->vfs_inode
);
3316 static int cifs_launder_page(struct page
*page
)
3319 loff_t range_start
= page_offset(page
);
3320 loff_t range_end
= range_start
+ (loff_t
)(PAGE_CACHE_SIZE
- 1);
3321 struct writeback_control wbc
= {
3322 .sync_mode
= WB_SYNC_ALL
,
3324 .range_start
= range_start
,
3325 .range_end
= range_end
,
3328 cFYI(1, "Launder page: %p", page
);
3330 if (clear_page_dirty_for_io(page
))
3331 rc
= cifs_writepage_locked(page
, &wbc
);
3333 cifs_fscache_invalidate_page(page
, page
->mapping
->host
);
3337 void cifs_oplock_break(struct work_struct
*work
)
3339 struct cifsFileInfo
*cfile
= container_of(work
, struct cifsFileInfo
,
3341 struct inode
*inode
= cfile
->dentry
->d_inode
;
3342 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
3345 if (inode
&& S_ISREG(inode
->i_mode
)) {
3346 if (cinode
->clientCanCacheRead
)
3347 break_lease(inode
, O_RDONLY
);
3349 break_lease(inode
, O_WRONLY
);
3350 rc
= filemap_fdatawrite(inode
->i_mapping
);
3351 if (cinode
->clientCanCacheRead
== 0) {
3352 rc
= filemap_fdatawait(inode
->i_mapping
);
3353 mapping_set_error(inode
->i_mapping
, rc
);
3354 invalidate_remote_inode(inode
);
3356 cFYI(1, "Oplock flush inode %p rc %d", inode
, rc
);
3359 rc
= cifs_push_locks(cfile
);
3361 cERROR(1, "Push locks rc = %d", rc
);
3364 * releasing stale oplock after recent reconnect of smb session using
3365 * a now incorrect file handle is not a data integrity issue but do
3366 * not bother sending an oplock release if session to server still is
3367 * disconnected since oplock already released by the server
3369 if (!cfile
->oplock_break_cancelled
) {
3370 rc
= CIFSSMBLock(0, tlink_tcon(cfile
->tlink
), cfile
->netfid
,
3371 current
->tgid
, 0, 0, 0, 0,
3372 LOCKING_ANDX_OPLOCK_RELEASE
, false,
3373 cinode
->clientCanCacheRead
? 1 : 0);
3374 cFYI(1, "Oplock release rc = %d", rc
);
3378 const struct address_space_operations cifs_addr_ops
= {
3379 .readpage
= cifs_readpage
,
3380 .readpages
= cifs_readpages
,
3381 .writepage
= cifs_writepage
,
3382 .writepages
= cifs_writepages
,
3383 .write_begin
= cifs_write_begin
,
3384 .write_end
= cifs_write_end
,
3385 .set_page_dirty
= __set_page_dirty_nobuffers
,
3386 .releasepage
= cifs_release_page
,
3387 .invalidatepage
= cifs_invalidate_page
,
3388 .launder_page
= cifs_launder_page
,
3392 * cifs_readpages requires the server to support a buffer large enough to
3393 * contain the header plus one complete page of data. Otherwise, we need
3394 * to leave cifs_readpages out of the address space operations.
3396 const struct address_space_operations cifs_addr_ops_smallbuf
= {
3397 .readpage
= cifs_readpage
,
3398 .writepage
= cifs_writepage
,
3399 .writepages
= cifs_writepages
,
3400 .write_begin
= cifs_write_begin
,
3401 .write_end
= cifs_write_end
,
3402 .set_page_dirty
= __set_page_dirty_nobuffers
,
3403 .releasepage
= cifs_release_page
,
3404 .invalidatepage
= cifs_invalidate_page
,
3405 .launder_page
= cifs_launder_page
,