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
*oplock
,
173 struct cifs_fid
*fid
, unsigned int xid
)
178 int create_options
= CREATE_NOT_DIR
;
180 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
182 if (!server
->ops
->open
)
185 desired_access
= cifs_convert_flags(f_flags
);
187 /*********************************************************************
188 * open flag mapping table:
190 * POSIX Flag CIFS Disposition
191 * ---------- ----------------
192 * O_CREAT FILE_OPEN_IF
193 * O_CREAT | O_EXCL FILE_CREATE
194 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
195 * O_TRUNC FILE_OVERWRITE
196 * none of the above FILE_OPEN
198 * Note that there is not a direct match between disposition
199 * FILE_SUPERSEDE (ie create whether or not file exists although
200 * O_CREAT | O_TRUNC is similar but truncates the existing
201 * file rather than creating a new file as FILE_SUPERSEDE does
202 * (which uses the attributes / metadata passed in on open call)
204 *? O_SYNC is a reasonable match to CIFS writethrough flag
205 *? and the read write flags match reasonably. O_LARGEFILE
206 *? is irrelevant because largefile support is always used
207 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
208 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
209 *********************************************************************/
211 disposition
= cifs_get_disposition(f_flags
);
213 /* BB pass O_SYNC flag through on file attributes .. BB */
215 buf
= kmalloc(sizeof(FILE_ALL_INFO
), GFP_KERNEL
);
219 if (backup_cred(cifs_sb
))
220 create_options
|= CREATE_OPEN_BACKUP_INTENT
;
222 rc
= server
->ops
->open(xid
, tcon
, full_path
, disposition
,
223 desired_access
, create_options
, fid
, oplock
, buf
,
230 rc
= cifs_get_inode_info_unix(&inode
, full_path
, inode
->i_sb
,
233 rc
= cifs_get_inode_info(&inode
, full_path
, buf
, inode
->i_sb
,
241 struct cifsFileInfo
*
242 cifs_new_fileinfo(struct cifs_fid
*fid
, struct file
*file
,
243 struct tcon_link
*tlink
, __u32 oplock
)
245 struct dentry
*dentry
= file
->f_path
.dentry
;
246 struct inode
*inode
= dentry
->d_inode
;
247 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
248 struct cifsFileInfo
*cfile
;
249 struct cifs_fid_locks
*fdlocks
;
250 struct cifs_tcon
*tcon
= tlink_tcon(tlink
);
252 cfile
= kzalloc(sizeof(struct cifsFileInfo
), GFP_KERNEL
);
256 fdlocks
= kzalloc(sizeof(struct cifs_fid_locks
), GFP_KERNEL
);
262 INIT_LIST_HEAD(&fdlocks
->locks
);
263 fdlocks
->cfile
= cfile
;
264 cfile
->llist
= fdlocks
;
265 down_write(&cinode
->lock_sem
);
266 list_add(&fdlocks
->llist
, &cinode
->llist
);
267 up_write(&cinode
->lock_sem
);
270 cfile
->pid
= current
->tgid
;
271 cfile
->uid
= current_fsuid();
272 cfile
->dentry
= dget(dentry
);
273 cfile
->f_flags
= file
->f_flags
;
274 cfile
->invalidHandle
= false;
275 cfile
->tlink
= cifs_get_tlink(tlink
);
276 INIT_WORK(&cfile
->oplock_break
, cifs_oplock_break
);
277 mutex_init(&cfile
->fh_mutex
);
279 spin_lock(&cifs_file_list_lock
);
280 if (fid
->pending_open
->oplock
!= CIFS_OPLOCK_NO_CHANGE
)
281 oplock
= fid
->pending_open
->oplock
;
282 list_del(&fid
->pending_open
->olist
);
284 tlink_tcon(tlink
)->ses
->server
->ops
->set_fid(cfile
, fid
, oplock
);
286 list_add(&cfile
->tlist
, &tcon
->openFileList
);
287 /* if readable file instance put first in list*/
288 if (file
->f_mode
& FMODE_READ
)
289 list_add(&cfile
->flist
, &cinode
->openFileList
);
291 list_add_tail(&cfile
->flist
, &cinode
->openFileList
);
292 spin_unlock(&cifs_file_list_lock
);
294 file
->private_data
= cfile
;
298 struct cifsFileInfo
*
299 cifsFileInfo_get(struct cifsFileInfo
*cifs_file
)
301 spin_lock(&cifs_file_list_lock
);
302 cifsFileInfo_get_locked(cifs_file
);
303 spin_unlock(&cifs_file_list_lock
);
308 * Release a reference on the file private data. This may involve closing
309 * the filehandle out on the server. Must be called without holding
310 * cifs_file_list_lock.
312 void cifsFileInfo_put(struct cifsFileInfo
*cifs_file
)
314 struct inode
*inode
= cifs_file
->dentry
->d_inode
;
315 struct cifs_tcon
*tcon
= tlink_tcon(cifs_file
->tlink
);
316 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
317 struct cifsInodeInfo
*cifsi
= CIFS_I(inode
);
318 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
319 struct cifsLockInfo
*li
, *tmp
;
321 struct cifs_pending_open open
;
323 spin_lock(&cifs_file_list_lock
);
324 if (--cifs_file
->count
> 0) {
325 spin_unlock(&cifs_file_list_lock
);
329 if (server
->ops
->get_lease_key
)
330 server
->ops
->get_lease_key(inode
, &fid
);
332 /* store open in pending opens to make sure we don't miss lease break */
333 cifs_add_pending_open_locked(&fid
, cifs_file
->tlink
, &open
);
335 /* remove it from the lists */
336 list_del(&cifs_file
->flist
);
337 list_del(&cifs_file
->tlist
);
339 if (list_empty(&cifsi
->openFileList
)) {
340 cFYI(1, "closing last open instance for inode %p",
341 cifs_file
->dentry
->d_inode
);
343 * In strict cache mode we need invalidate mapping on the last
344 * close because it may cause a error when we open this file
345 * again and get at least level II oplock.
347 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_STRICT_IO
)
348 CIFS_I(inode
)->invalid_mapping
= true;
349 cifs_set_oplock_level(cifsi
, 0);
351 spin_unlock(&cifs_file_list_lock
);
353 cancel_work_sync(&cifs_file
->oplock_break
);
355 if (!tcon
->need_reconnect
&& !cifs_file
->invalidHandle
) {
356 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
361 if (server
->ops
->close
)
362 rc
= server
->ops
->close(xid
, tcon
, &cifs_file
->fid
);
366 cifs_del_pending_open(&open
);
369 * Delete any outstanding lock records. We'll lose them when the file
372 down_write(&cifsi
->lock_sem
);
373 list_for_each_entry_safe(li
, tmp
, &cifs_file
->llist
->locks
, llist
) {
374 list_del(&li
->llist
);
375 cifs_del_lock_waiters(li
);
378 list_del(&cifs_file
->llist
->llist
);
379 kfree(cifs_file
->llist
);
380 up_write(&cifsi
->lock_sem
);
382 cifs_put_tlink(cifs_file
->tlink
);
383 dput(cifs_file
->dentry
);
387 int cifs_open(struct inode
*inode
, struct file
*file
)
393 struct cifs_sb_info
*cifs_sb
;
394 struct TCP_Server_Info
*server
;
395 struct cifs_tcon
*tcon
;
396 struct tcon_link
*tlink
;
397 struct cifsFileInfo
*cfile
= NULL
;
398 char *full_path
= NULL
;
399 bool posix_open_ok
= false;
401 struct cifs_pending_open open
;
405 cifs_sb
= CIFS_SB(inode
->i_sb
);
406 tlink
= cifs_sb_tlink(cifs_sb
);
409 return PTR_ERR(tlink
);
411 tcon
= tlink_tcon(tlink
);
412 server
= tcon
->ses
->server
;
414 full_path
= build_path_from_dentry(file
->f_path
.dentry
);
415 if (full_path
== NULL
) {
420 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
421 inode
, file
->f_flags
, full_path
);
428 if (!tcon
->broken_posix_open
&& tcon
->unix_ext
&&
429 cap_unix(tcon
->ses
) && (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
430 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
431 /* can not refresh inode info since size could be stale */
432 rc
= cifs_posix_open(full_path
, &inode
, inode
->i_sb
,
433 cifs_sb
->mnt_file_mode
/* ignored */,
434 file
->f_flags
, &oplock
, &fid
.netfid
, xid
);
436 cFYI(1, "posix open succeeded");
437 posix_open_ok
= true;
438 } else if ((rc
== -EINVAL
) || (rc
== -EOPNOTSUPP
)) {
439 if (tcon
->ses
->serverNOS
)
440 cERROR(1, "server %s of type %s returned"
441 " unexpected error on SMB posix open"
442 ", disabling posix open support."
443 " Check if server update available.",
444 tcon
->ses
->serverName
,
445 tcon
->ses
->serverNOS
);
446 tcon
->broken_posix_open
= true;
447 } else if ((rc
!= -EIO
) && (rc
!= -EREMOTE
) &&
448 (rc
!= -EOPNOTSUPP
)) /* path not found or net err */
451 * Else fallthrough to retry open the old way on network i/o
456 if (server
->ops
->get_lease_key
)
457 server
->ops
->get_lease_key(inode
, &fid
);
459 cifs_add_pending_open(&fid
, tlink
, &open
);
461 if (!posix_open_ok
) {
462 if (server
->ops
->get_lease_key
)
463 server
->ops
->get_lease_key(inode
, &fid
);
465 rc
= cifs_nt_open(full_path
, inode
, cifs_sb
, tcon
,
466 file
->f_flags
, &oplock
, &fid
, xid
);
468 cifs_del_pending_open(&open
);
473 cfile
= cifs_new_fileinfo(&fid
, file
, tlink
, oplock
);
475 if (server
->ops
->close
)
476 server
->ops
->close(xid
, tcon
, &fid
);
477 cifs_del_pending_open(&open
);
482 cifs_fscache_set_inode_cookie(inode
, file
);
484 if ((oplock
& CIFS_CREATE_ACTION
) && !posix_open_ok
&& tcon
->unix_ext
) {
486 * Time to set mode which we can not set earlier due to
487 * problems creating new read-only files.
489 struct cifs_unix_set_info_args args
= {
490 .mode
= inode
->i_mode
,
493 .ctime
= NO_CHANGE_64
,
494 .atime
= NO_CHANGE_64
,
495 .mtime
= NO_CHANGE_64
,
498 CIFSSMBUnixSetFileInfo(xid
, tcon
, &args
, fid
.netfid
,
505 cifs_put_tlink(tlink
);
510 * Try to reacquire byte range locks that were released when session
513 static int cifs_relock_file(struct cifsFileInfo
*cifsFile
)
517 /* BB list all locks open on this file and relock */
523 cifs_reopen_file(struct cifsFileInfo
*cfile
, bool can_flush
)
528 struct cifs_sb_info
*cifs_sb
;
529 struct cifs_tcon
*tcon
;
530 struct TCP_Server_Info
*server
;
531 struct cifsInodeInfo
*cinode
;
533 char *full_path
= NULL
;
535 int disposition
= FILE_OPEN
;
536 int create_options
= CREATE_NOT_DIR
;
540 mutex_lock(&cfile
->fh_mutex
);
541 if (!cfile
->invalidHandle
) {
542 mutex_unlock(&cfile
->fh_mutex
);
548 inode
= cfile
->dentry
->d_inode
;
549 cifs_sb
= CIFS_SB(inode
->i_sb
);
550 tcon
= tlink_tcon(cfile
->tlink
);
551 server
= tcon
->ses
->server
;
554 * Can not grab rename sem here because various ops, including those
555 * that already have the rename sem can end up causing writepage to get
556 * called and if the server was down that means we end up here, and we
557 * can never tell if the caller already has the rename_sem.
559 full_path
= build_path_from_dentry(cfile
->dentry
);
560 if (full_path
== NULL
) {
562 mutex_unlock(&cfile
->fh_mutex
);
567 cFYI(1, "inode = 0x%p file flags 0x%x for %s", inode
, cfile
->f_flags
,
570 if (tcon
->ses
->server
->oplocks
)
575 if (tcon
->unix_ext
&& cap_unix(tcon
->ses
) &&
576 (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
577 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
579 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
580 * original open. Must mask them off for a reopen.
582 unsigned int oflags
= cfile
->f_flags
&
583 ~(O_CREAT
| O_EXCL
| O_TRUNC
);
585 rc
= cifs_posix_open(full_path
, NULL
, inode
->i_sb
,
586 cifs_sb
->mnt_file_mode
/* ignored */,
587 oflags
, &oplock
, &fid
.netfid
, xid
);
589 cFYI(1, "posix reopen succeeded");
593 * fallthrough to retry open the old way on errors, especially
594 * in the reconnect path it is important to retry hard
598 desired_access
= cifs_convert_flags(cfile
->f_flags
);
600 if (backup_cred(cifs_sb
))
601 create_options
|= CREATE_OPEN_BACKUP_INTENT
;
603 if (server
->ops
->get_lease_key
)
604 server
->ops
->get_lease_key(inode
, &fid
);
607 * Can not refresh inode by passing in file_info buf to be returned by
608 * CIFSSMBOpen and then calling get_inode_info with returned buf since
609 * file might have write behind data that needs to be flushed and server
610 * version of file size can be stale. If we knew for sure that inode was
611 * not dirty locally we could do this.
613 rc
= server
->ops
->open(xid
, tcon
, full_path
, disposition
,
614 desired_access
, create_options
, &fid
, &oplock
,
617 mutex_unlock(&cfile
->fh_mutex
);
618 cFYI(1, "cifs_reopen returned 0x%x", rc
);
619 cFYI(1, "oplock: %d", oplock
);
620 goto reopen_error_exit
;
624 cfile
->invalidHandle
= false;
625 mutex_unlock(&cfile
->fh_mutex
);
626 cinode
= CIFS_I(inode
);
629 rc
= filemap_write_and_wait(inode
->i_mapping
);
630 mapping_set_error(inode
->i_mapping
, rc
);
633 rc
= cifs_get_inode_info_unix(&inode
, full_path
,
636 rc
= cifs_get_inode_info(&inode
, full_path
, NULL
,
637 inode
->i_sb
, xid
, NULL
);
640 * Else we are writing out data to server already and could deadlock if
641 * we tried to flush data, and since we do not know if we have data that
642 * would invalidate the current end of file on the server we can not go
643 * to the server to get the new inode info.
646 server
->ops
->set_fid(cfile
, &fid
, oplock
);
647 cifs_relock_file(cfile
);
655 int cifs_close(struct inode
*inode
, struct file
*file
)
657 if (file
->private_data
!= NULL
) {
658 cifsFileInfo_put(file
->private_data
);
659 file
->private_data
= NULL
;
662 /* return code from the ->release op is always ignored */
666 int cifs_closedir(struct inode
*inode
, struct file
*file
)
670 struct cifsFileInfo
*cfile
= file
->private_data
;
671 struct cifs_tcon
*tcon
;
672 struct TCP_Server_Info
*server
;
675 cFYI(1, "Closedir inode = 0x%p", inode
);
681 tcon
= tlink_tcon(cfile
->tlink
);
682 server
= tcon
->ses
->server
;
684 cFYI(1, "Freeing private data in close dir");
685 spin_lock(&cifs_file_list_lock
);
686 if (!cfile
->srch_inf
.endOfSearch
&& !cfile
->invalidHandle
) {
687 cfile
->invalidHandle
= true;
688 spin_unlock(&cifs_file_list_lock
);
689 if (server
->ops
->close_dir
)
690 rc
= server
->ops
->close_dir(xid
, tcon
, &cfile
->fid
);
693 cFYI(1, "Closing uncompleted readdir with rc %d", rc
);
694 /* not much we can do if it fails anyway, ignore rc */
697 spin_unlock(&cifs_file_list_lock
);
699 buf
= cfile
->srch_inf
.ntwrk_buf_start
;
701 cFYI(1, "closedir free smb buf in srch struct");
702 cfile
->srch_inf
.ntwrk_buf_start
= NULL
;
703 if (cfile
->srch_inf
.smallBuf
)
704 cifs_small_buf_release(buf
);
706 cifs_buf_release(buf
);
709 cifs_put_tlink(cfile
->tlink
);
710 kfree(file
->private_data
);
711 file
->private_data
= NULL
;
712 /* BB can we lock the filestruct while this is going on? */
717 static struct cifsLockInfo
*
718 cifs_lock_init(__u64 offset
, __u64 length
, __u8 type
)
720 struct cifsLockInfo
*lock
=
721 kmalloc(sizeof(struct cifsLockInfo
), GFP_KERNEL
);
724 lock
->offset
= offset
;
725 lock
->length
= length
;
727 lock
->pid
= current
->tgid
;
728 INIT_LIST_HEAD(&lock
->blist
);
729 init_waitqueue_head(&lock
->block_q
);
734 cifs_del_lock_waiters(struct cifsLockInfo
*lock
)
736 struct cifsLockInfo
*li
, *tmp
;
737 list_for_each_entry_safe(li
, tmp
, &lock
->blist
, blist
) {
738 list_del_init(&li
->blist
);
739 wake_up(&li
->block_q
);
744 cifs_find_fid_lock_conflict(struct cifs_fid_locks
*fdlocks
, __u64 offset
,
745 __u64 length
, __u8 type
, struct cifsFileInfo
*cfile
,
746 struct cifsLockInfo
**conf_lock
, bool rw_check
)
748 struct cifsLockInfo
*li
;
749 struct cifsFileInfo
*cur_cfile
= fdlocks
->cfile
;
750 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
752 list_for_each_entry(li
, &fdlocks
->locks
, llist
) {
753 if (offset
+ length
<= li
->offset
||
754 offset
>= li
->offset
+ li
->length
)
756 if (rw_check
&& server
->ops
->compare_fids(cfile
, cur_cfile
) &&
757 current
->tgid
== li
->pid
)
759 if ((type
& server
->vals
->shared_lock_type
) &&
760 ((server
->ops
->compare_fids(cfile
, cur_cfile
) &&
761 current
->tgid
== li
->pid
) || type
== li
->type
))
771 cifs_find_lock_conflict(struct cifsFileInfo
*cfile
, __u64 offset
, __u64 length
,
772 __u8 type
, struct cifsLockInfo
**conf_lock
,
776 struct cifs_fid_locks
*cur
;
777 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
779 list_for_each_entry(cur
, &cinode
->llist
, llist
) {
780 rc
= cifs_find_fid_lock_conflict(cur
, offset
, length
, type
,
781 cfile
, conf_lock
, rw_check
);
790 * Check if there is another lock that prevents us to set the lock (mandatory
791 * style). If such a lock exists, update the flock structure with its
792 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
793 * or leave it the same if we can't. Returns 0 if we don't need to request to
794 * the server or 1 otherwise.
797 cifs_lock_test(struct cifsFileInfo
*cfile
, __u64 offset
, __u64 length
,
798 __u8 type
, struct file_lock
*flock
)
801 struct cifsLockInfo
*conf_lock
;
802 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
803 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
806 down_read(&cinode
->lock_sem
);
808 exist
= cifs_find_lock_conflict(cfile
, offset
, length
, type
,
811 flock
->fl_start
= conf_lock
->offset
;
812 flock
->fl_end
= conf_lock
->offset
+ conf_lock
->length
- 1;
813 flock
->fl_pid
= conf_lock
->pid
;
814 if (conf_lock
->type
& server
->vals
->shared_lock_type
)
815 flock
->fl_type
= F_RDLCK
;
817 flock
->fl_type
= F_WRLCK
;
818 } else if (!cinode
->can_cache_brlcks
)
821 flock
->fl_type
= F_UNLCK
;
823 up_read(&cinode
->lock_sem
);
828 cifs_lock_add(struct cifsFileInfo
*cfile
, struct cifsLockInfo
*lock
)
830 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
831 down_write(&cinode
->lock_sem
);
832 list_add_tail(&lock
->llist
, &cfile
->llist
->locks
);
833 up_write(&cinode
->lock_sem
);
837 * Set the byte-range lock (mandatory style). Returns:
838 * 1) 0, if we set the lock and don't need to request to the server;
839 * 2) 1, if no locks prevent us but we need to request to the server;
840 * 3) -EACCESS, if there is a lock that prevents us and wait is false.
843 cifs_lock_add_if(struct cifsFileInfo
*cfile
, struct cifsLockInfo
*lock
,
846 struct cifsLockInfo
*conf_lock
;
847 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
853 down_write(&cinode
->lock_sem
);
855 exist
= cifs_find_lock_conflict(cfile
, lock
->offset
, lock
->length
,
856 lock
->type
, &conf_lock
, false);
857 if (!exist
&& cinode
->can_cache_brlcks
) {
858 list_add_tail(&lock
->llist
, &cfile
->llist
->locks
);
859 up_write(&cinode
->lock_sem
);
868 list_add_tail(&lock
->blist
, &conf_lock
->blist
);
869 up_write(&cinode
->lock_sem
);
870 rc
= wait_event_interruptible(lock
->block_q
,
871 (lock
->blist
.prev
== &lock
->blist
) &&
872 (lock
->blist
.next
== &lock
->blist
));
875 down_write(&cinode
->lock_sem
);
876 list_del_init(&lock
->blist
);
879 up_write(&cinode
->lock_sem
);
884 * Check if there is another lock that prevents us to set the lock (posix
885 * style). If such a lock exists, update the flock structure with its
886 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
887 * or leave it the same if we can't. Returns 0 if we don't need to request to
888 * the server or 1 otherwise.
891 cifs_posix_lock_test(struct file
*file
, struct file_lock
*flock
)
894 struct cifsInodeInfo
*cinode
= CIFS_I(file
->f_path
.dentry
->d_inode
);
895 unsigned char saved_type
= flock
->fl_type
;
897 if ((flock
->fl_flags
& FL_POSIX
) == 0)
900 down_read(&cinode
->lock_sem
);
901 posix_test_lock(file
, flock
);
903 if (flock
->fl_type
== F_UNLCK
&& !cinode
->can_cache_brlcks
) {
904 flock
->fl_type
= saved_type
;
908 up_read(&cinode
->lock_sem
);
913 * Set the byte-range lock (posix style). Returns:
914 * 1) 0, if we set the lock and don't need to request to the server;
915 * 2) 1, if we need to request to the server;
916 * 3) <0, if the error occurs while setting the lock.
919 cifs_posix_lock_set(struct file
*file
, struct file_lock
*flock
)
921 struct cifsInodeInfo
*cinode
= CIFS_I(file
->f_path
.dentry
->d_inode
);
924 if ((flock
->fl_flags
& FL_POSIX
) == 0)
928 down_write(&cinode
->lock_sem
);
929 if (!cinode
->can_cache_brlcks
) {
930 up_write(&cinode
->lock_sem
);
934 rc
= posix_lock_file(file
, flock
, NULL
);
935 up_write(&cinode
->lock_sem
);
936 if (rc
== FILE_LOCK_DEFERRED
) {
937 rc
= wait_event_interruptible(flock
->fl_wait
, !flock
->fl_next
);
940 locks_delete_block(flock
);
946 cifs_push_mandatory_locks(struct cifsFileInfo
*cfile
)
949 int rc
= 0, stored_rc
;
950 struct cifsLockInfo
*li
, *tmp
;
951 struct cifs_tcon
*tcon
;
952 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
953 unsigned int num
, max_num
, max_buf
;
954 LOCKING_ANDX_RANGE
*buf
, *cur
;
955 int types
[] = {LOCKING_ANDX_LARGE_FILES
,
956 LOCKING_ANDX_SHARED_LOCK
| LOCKING_ANDX_LARGE_FILES
};
960 tcon
= tlink_tcon(cfile
->tlink
);
962 /* we are going to update can_cache_brlcks here - need a write access */
963 down_write(&cinode
->lock_sem
);
964 if (!cinode
->can_cache_brlcks
) {
965 up_write(&cinode
->lock_sem
);
971 * Accessing maxBuf is racy with cifs_reconnect - need to store value
972 * and check it for zero before using.
974 max_buf
= tcon
->ses
->server
->maxBuf
;
976 up_write(&cinode
->lock_sem
);
981 max_num
= (max_buf
- sizeof(struct smb_hdr
)) /
982 sizeof(LOCKING_ANDX_RANGE
);
983 buf
= kzalloc(max_num
* sizeof(LOCKING_ANDX_RANGE
), GFP_KERNEL
);
985 up_write(&cinode
->lock_sem
);
990 for (i
= 0; i
< 2; i
++) {
993 list_for_each_entry_safe(li
, tmp
, &cfile
->llist
->locks
, llist
) {
994 if (li
->type
!= types
[i
])
996 cur
->Pid
= cpu_to_le16(li
->pid
);
997 cur
->LengthLow
= cpu_to_le32((u32
)li
->length
);
998 cur
->LengthHigh
= cpu_to_le32((u32
)(li
->length
>>32));
999 cur
->OffsetLow
= cpu_to_le32((u32
)li
->offset
);
1000 cur
->OffsetHigh
= cpu_to_le32((u32
)(li
->offset
>>32));
1001 if (++num
== max_num
) {
1002 stored_rc
= cifs_lockv(xid
, tcon
,
1004 (__u8
)li
->type
, 0, num
,
1015 stored_rc
= cifs_lockv(xid
, tcon
, cfile
->fid
.netfid
,
1016 (__u8
)types
[i
], 0, num
, buf
);
1022 cinode
->can_cache_brlcks
= false;
1023 up_write(&cinode
->lock_sem
);
1030 /* copied from fs/locks.c with a name change */
1031 #define cifs_for_each_lock(inode, lockp) \
1032 for (lockp = &inode->i_flock; *lockp != NULL; \
1033 lockp = &(*lockp)->fl_next)
1035 struct lock_to_push
{
1036 struct list_head llist
;
1045 cifs_push_posix_locks(struct cifsFileInfo
*cfile
)
1047 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
1048 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1049 struct file_lock
*flock
, **before
;
1050 unsigned int count
= 0, i
= 0;
1051 int rc
= 0, xid
, type
;
1052 struct list_head locks_to_send
, *el
;
1053 struct lock_to_push
*lck
, *tmp
;
1058 /* we are going to update can_cache_brlcks here - need a write access */
1059 down_write(&cinode
->lock_sem
);
1060 if (!cinode
->can_cache_brlcks
) {
1061 up_write(&cinode
->lock_sem
);
1067 cifs_for_each_lock(cfile
->dentry
->d_inode
, before
) {
1068 if ((*before
)->fl_flags
& FL_POSIX
)
1073 INIT_LIST_HEAD(&locks_to_send
);
1076 * Allocating count locks is enough because no FL_POSIX locks can be
1077 * added to the list while we are holding cinode->lock_sem that
1078 * protects locking operations of this inode.
1080 for (; i
< count
; i
++) {
1081 lck
= kmalloc(sizeof(struct lock_to_push
), GFP_KERNEL
);
1086 list_add_tail(&lck
->llist
, &locks_to_send
);
1089 el
= locks_to_send
.next
;
1091 cifs_for_each_lock(cfile
->dentry
->d_inode
, before
) {
1093 if ((flock
->fl_flags
& FL_POSIX
) == 0)
1095 if (el
== &locks_to_send
) {
1097 * The list ended. We don't have enough allocated
1098 * structures - something is really wrong.
1100 cERROR(1, "Can't push all brlocks!");
1103 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1104 if (flock
->fl_type
== F_RDLCK
|| flock
->fl_type
== F_SHLCK
)
1108 lck
= list_entry(el
, struct lock_to_push
, llist
);
1109 lck
->pid
= flock
->fl_pid
;
1110 lck
->netfid
= cfile
->fid
.netfid
;
1111 lck
->length
= length
;
1113 lck
->offset
= flock
->fl_start
;
1118 list_for_each_entry_safe(lck
, tmp
, &locks_to_send
, llist
) {
1121 stored_rc
= CIFSSMBPosixLock(xid
, tcon
, lck
->netfid
, lck
->pid
,
1122 lck
->offset
, lck
->length
, NULL
,
1126 list_del(&lck
->llist
);
1131 cinode
->can_cache_brlcks
= false;
1132 up_write(&cinode
->lock_sem
);
1137 list_for_each_entry_safe(lck
, tmp
, &locks_to_send
, llist
) {
1138 list_del(&lck
->llist
);
1145 cifs_push_locks(struct cifsFileInfo
*cfile
)
1147 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cfile
->dentry
->d_sb
);
1148 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1150 if (cap_unix(tcon
->ses
) &&
1151 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
1152 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
1153 return cifs_push_posix_locks(cfile
);
1155 return tcon
->ses
->server
->ops
->push_mand_locks(cfile
);
1159 cifs_read_flock(struct file_lock
*flock
, __u32
*type
, int *lock
, int *unlock
,
1160 bool *wait_flag
, struct TCP_Server_Info
*server
)
1162 if (flock
->fl_flags
& FL_POSIX
)
1164 if (flock
->fl_flags
& FL_FLOCK
)
1166 if (flock
->fl_flags
& FL_SLEEP
) {
1167 cFYI(1, "Blocking lock");
1170 if (flock
->fl_flags
& FL_ACCESS
)
1171 cFYI(1, "Process suspended by mandatory locking - "
1172 "not implemented yet");
1173 if (flock
->fl_flags
& FL_LEASE
)
1174 cFYI(1, "Lease on file - not implemented yet");
1175 if (flock
->fl_flags
&
1176 (~(FL_POSIX
| FL_FLOCK
| FL_SLEEP
|
1177 FL_ACCESS
| FL_LEASE
| FL_CLOSE
)))
1178 cFYI(1, "Unknown lock flags 0x%x", flock
->fl_flags
);
1180 *type
= server
->vals
->large_lock_type
;
1181 if (flock
->fl_type
== F_WRLCK
) {
1182 cFYI(1, "F_WRLCK ");
1183 *type
|= server
->vals
->exclusive_lock_type
;
1185 } else if (flock
->fl_type
== F_UNLCK
) {
1187 *type
|= server
->vals
->unlock_lock_type
;
1189 /* Check if unlock includes more than one lock range */
1190 } else if (flock
->fl_type
== F_RDLCK
) {
1192 *type
|= server
->vals
->shared_lock_type
;
1194 } else if (flock
->fl_type
== F_EXLCK
) {
1196 *type
|= server
->vals
->exclusive_lock_type
;
1198 } else if (flock
->fl_type
== F_SHLCK
) {
1200 *type
|= server
->vals
->shared_lock_type
;
1203 cFYI(1, "Unknown type of lock");
1207 cifs_getlk(struct file
*file
, struct file_lock
*flock
, __u32 type
,
1208 bool wait_flag
, bool posix_lck
, unsigned int xid
)
1211 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1212 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
1213 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1214 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
1215 __u16 netfid
= cfile
->fid
.netfid
;
1218 int posix_lock_type
;
1220 rc
= cifs_posix_lock_test(file
, flock
);
1224 if (type
& server
->vals
->shared_lock_type
)
1225 posix_lock_type
= CIFS_RDLCK
;
1227 posix_lock_type
= CIFS_WRLCK
;
1228 rc
= CIFSSMBPosixLock(xid
, tcon
, netfid
, current
->tgid
,
1229 flock
->fl_start
, length
, flock
,
1230 posix_lock_type
, wait_flag
);
1234 rc
= cifs_lock_test(cfile
, flock
->fl_start
, length
, type
, flock
);
1238 /* BB we could chain these into one lock request BB */
1239 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
, type
,
1242 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1244 flock
->fl_type
= F_UNLCK
;
1246 cERROR(1, "Error unlocking previously locked "
1247 "range %d during test of lock", rc
);
1251 if (type
& server
->vals
->shared_lock_type
) {
1252 flock
->fl_type
= F_WRLCK
;
1256 type
&= ~server
->vals
->exclusive_lock_type
;
1258 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1259 type
| server
->vals
->shared_lock_type
,
1262 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1263 type
| server
->vals
->shared_lock_type
, 0, 1, false);
1264 flock
->fl_type
= F_RDLCK
;
1266 cERROR(1, "Error unlocking previously locked "
1267 "range %d during test of lock", rc
);
1269 flock
->fl_type
= F_WRLCK
;
1275 cifs_move_llist(struct list_head
*source
, struct list_head
*dest
)
1277 struct list_head
*li
, *tmp
;
1278 list_for_each_safe(li
, tmp
, source
)
1279 list_move(li
, dest
);
1283 cifs_free_llist(struct list_head
*llist
)
1285 struct cifsLockInfo
*li
, *tmp
;
1286 list_for_each_entry_safe(li
, tmp
, llist
, llist
) {
1287 cifs_del_lock_waiters(li
);
1288 list_del(&li
->llist
);
1294 cifs_unlock_range(struct cifsFileInfo
*cfile
, struct file_lock
*flock
,
1297 int rc
= 0, stored_rc
;
1298 int types
[] = {LOCKING_ANDX_LARGE_FILES
,
1299 LOCKING_ANDX_SHARED_LOCK
| LOCKING_ANDX_LARGE_FILES
};
1301 unsigned int max_num
, num
, max_buf
;
1302 LOCKING_ANDX_RANGE
*buf
, *cur
;
1303 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1304 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
1305 struct cifsLockInfo
*li
, *tmp
;
1306 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1307 struct list_head tmp_llist
;
1309 INIT_LIST_HEAD(&tmp_llist
);
1312 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1313 * and check it for zero before using.
1315 max_buf
= tcon
->ses
->server
->maxBuf
;
1319 max_num
= (max_buf
- sizeof(struct smb_hdr
)) /
1320 sizeof(LOCKING_ANDX_RANGE
);
1321 buf
= kzalloc(max_num
* sizeof(LOCKING_ANDX_RANGE
), GFP_KERNEL
);
1325 down_write(&cinode
->lock_sem
);
1326 for (i
= 0; i
< 2; i
++) {
1329 list_for_each_entry_safe(li
, tmp
, &cfile
->llist
->locks
, llist
) {
1330 if (flock
->fl_start
> li
->offset
||
1331 (flock
->fl_start
+ length
) <
1332 (li
->offset
+ li
->length
))
1334 if (current
->tgid
!= li
->pid
)
1336 if (types
[i
] != li
->type
)
1338 if (cinode
->can_cache_brlcks
) {
1340 * We can cache brlock requests - simply remove
1341 * a lock from the file's list.
1343 list_del(&li
->llist
);
1344 cifs_del_lock_waiters(li
);
1348 cur
->Pid
= cpu_to_le16(li
->pid
);
1349 cur
->LengthLow
= cpu_to_le32((u32
)li
->length
);
1350 cur
->LengthHigh
= cpu_to_le32((u32
)(li
->length
>>32));
1351 cur
->OffsetLow
= cpu_to_le32((u32
)li
->offset
);
1352 cur
->OffsetHigh
= cpu_to_le32((u32
)(li
->offset
>>32));
1354 * We need to save a lock here to let us add it again to
1355 * the file's list if the unlock range request fails on
1358 list_move(&li
->llist
, &tmp_llist
);
1359 if (++num
== max_num
) {
1360 stored_rc
= cifs_lockv(xid
, tcon
,
1362 li
->type
, num
, 0, buf
);
1365 * We failed on the unlock range
1366 * request - add all locks from the tmp
1367 * list to the head of the file's list.
1369 cifs_move_llist(&tmp_llist
,
1370 &cfile
->llist
->locks
);
1374 * The unlock range request succeed -
1375 * free the tmp list.
1377 cifs_free_llist(&tmp_llist
);
1384 stored_rc
= cifs_lockv(xid
, tcon
, cfile
->fid
.netfid
,
1385 types
[i
], num
, 0, buf
);
1387 cifs_move_llist(&tmp_llist
,
1388 &cfile
->llist
->locks
);
1391 cifs_free_llist(&tmp_llist
);
1395 up_write(&cinode
->lock_sem
);
1401 cifs_setlk(struct file
*file
, struct file_lock
*flock
, __u32 type
,
1402 bool wait_flag
, bool posix_lck
, int lock
, int unlock
,
1406 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1407 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
1408 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1409 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
1412 int posix_lock_type
;
1414 rc
= cifs_posix_lock_set(file
, flock
);
1418 if (type
& server
->vals
->shared_lock_type
)
1419 posix_lock_type
= CIFS_RDLCK
;
1421 posix_lock_type
= CIFS_WRLCK
;
1424 posix_lock_type
= CIFS_UNLCK
;
1426 rc
= CIFSSMBPosixLock(xid
, tcon
, cfile
->fid
.netfid
,
1427 current
->tgid
, flock
->fl_start
, length
,
1428 NULL
, posix_lock_type
, wait_flag
);
1433 struct cifsLockInfo
*lock
;
1435 lock
= cifs_lock_init(flock
->fl_start
, length
, type
);
1439 rc
= cifs_lock_add_if(cfile
, lock
, wait_flag
);
1445 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1446 type
, 1, 0, wait_flag
);
1452 cifs_lock_add(cfile
, lock
);
1454 rc
= server
->ops
->mand_unlock_range(cfile
, flock
, xid
);
1457 if (flock
->fl_flags
& FL_POSIX
)
1458 posix_lock_file_wait(file
, flock
);
1462 int cifs_lock(struct file
*file
, int cmd
, struct file_lock
*flock
)
1465 int lock
= 0, unlock
= 0;
1466 bool wait_flag
= false;
1467 bool posix_lck
= false;
1468 struct cifs_sb_info
*cifs_sb
;
1469 struct cifs_tcon
*tcon
;
1470 struct cifsInodeInfo
*cinode
;
1471 struct cifsFileInfo
*cfile
;
1478 cFYI(1, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld "
1479 "end: %lld", cmd
, flock
->fl_flags
, flock
->fl_type
,
1480 flock
->fl_start
, flock
->fl_end
);
1482 cfile
= (struct cifsFileInfo
*)file
->private_data
;
1483 tcon
= tlink_tcon(cfile
->tlink
);
1485 cifs_read_flock(flock
, &type
, &lock
, &unlock
, &wait_flag
,
1488 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
1489 netfid
= cfile
->fid
.netfid
;
1490 cinode
= CIFS_I(file
->f_path
.dentry
->d_inode
);
1492 if (cap_unix(tcon
->ses
) &&
1493 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
1494 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
1497 * BB add code here to normalize offset and length to account for
1498 * negative length which we can not accept over the wire.
1500 if (IS_GETLK(cmd
)) {
1501 rc
= cifs_getlk(file
, flock
, type
, wait_flag
, posix_lck
, xid
);
1506 if (!lock
&& !unlock
) {
1508 * if no lock or unlock then nothing to do since we do not
1515 rc
= cifs_setlk(file
, flock
, type
, wait_flag
, posix_lck
, lock
, unlock
,
1522 * update the file size (if needed) after a write. Should be called with
1523 * the inode->i_lock held
1526 cifs_update_eof(struct cifsInodeInfo
*cifsi
, loff_t offset
,
1527 unsigned int bytes_written
)
1529 loff_t end_of_write
= offset
+ bytes_written
;
1531 if (end_of_write
> cifsi
->server_eof
)
1532 cifsi
->server_eof
= end_of_write
;
1536 cifs_write(struct cifsFileInfo
*open_file
, __u32 pid
, const char *write_data
,
1537 size_t write_size
, loff_t
*offset
)
1540 unsigned int bytes_written
= 0;
1541 unsigned int total_written
;
1542 struct cifs_sb_info
*cifs_sb
;
1543 struct cifs_tcon
*tcon
;
1544 struct TCP_Server_Info
*server
;
1546 struct dentry
*dentry
= open_file
->dentry
;
1547 struct cifsInodeInfo
*cifsi
= CIFS_I(dentry
->d_inode
);
1548 struct cifs_io_parms io_parms
;
1550 cifs_sb
= CIFS_SB(dentry
->d_sb
);
1552 cFYI(1, "write %zd bytes to offset %lld of %s", write_size
,
1553 *offset
, dentry
->d_name
.name
);
1555 tcon
= tlink_tcon(open_file
->tlink
);
1556 server
= tcon
->ses
->server
;
1558 if (!server
->ops
->sync_write
)
1563 for (total_written
= 0; write_size
> total_written
;
1564 total_written
+= bytes_written
) {
1566 while (rc
== -EAGAIN
) {
1570 if (open_file
->invalidHandle
) {
1571 /* we could deadlock if we called
1572 filemap_fdatawait from here so tell
1573 reopen_file not to flush data to
1575 rc
= cifs_reopen_file(open_file
, false);
1580 len
= min((size_t)cifs_sb
->wsize
,
1581 write_size
- total_written
);
1582 /* iov[0] is reserved for smb header */
1583 iov
[1].iov_base
= (char *)write_data
+ total_written
;
1584 iov
[1].iov_len
= len
;
1586 io_parms
.tcon
= tcon
;
1587 io_parms
.offset
= *offset
;
1588 io_parms
.length
= len
;
1589 rc
= server
->ops
->sync_write(xid
, open_file
, &io_parms
,
1590 &bytes_written
, iov
, 1);
1592 if (rc
|| (bytes_written
== 0)) {
1600 spin_lock(&dentry
->d_inode
->i_lock
);
1601 cifs_update_eof(cifsi
, *offset
, bytes_written
);
1602 spin_unlock(&dentry
->d_inode
->i_lock
);
1603 *offset
+= bytes_written
;
1607 cifs_stats_bytes_written(tcon
, total_written
);
1609 if (total_written
> 0) {
1610 spin_lock(&dentry
->d_inode
->i_lock
);
1611 if (*offset
> dentry
->d_inode
->i_size
)
1612 i_size_write(dentry
->d_inode
, *offset
);
1613 spin_unlock(&dentry
->d_inode
->i_lock
);
1615 mark_inode_dirty_sync(dentry
->d_inode
);
1617 return total_written
;
1620 struct cifsFileInfo
*find_readable_file(struct cifsInodeInfo
*cifs_inode
,
1623 struct cifsFileInfo
*open_file
= NULL
;
1624 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1626 /* only filter by fsuid on multiuser mounts */
1627 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1630 spin_lock(&cifs_file_list_lock
);
1631 /* we could simply get the first_list_entry since write-only entries
1632 are always at the end of the list but since the first entry might
1633 have a close pending, we go through the whole list */
1634 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1635 if (fsuid_only
&& open_file
->uid
!= current_fsuid())
1637 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_READ
) {
1638 if (!open_file
->invalidHandle
) {
1639 /* found a good file */
1640 /* lock it so it will not be closed on us */
1641 cifsFileInfo_get_locked(open_file
);
1642 spin_unlock(&cifs_file_list_lock
);
1644 } /* else might as well continue, and look for
1645 another, or simply have the caller reopen it
1646 again rather than trying to fix this handle */
1647 } else /* write only file */
1648 break; /* write only files are last so must be done */
1650 spin_unlock(&cifs_file_list_lock
);
1654 struct cifsFileInfo
*find_writable_file(struct cifsInodeInfo
*cifs_inode
,
1657 struct cifsFileInfo
*open_file
, *inv_file
= NULL
;
1658 struct cifs_sb_info
*cifs_sb
;
1659 bool any_available
= false;
1661 unsigned int refind
= 0;
1663 /* Having a null inode here (because mapping->host was set to zero by
1664 the VFS or MM) should not happen but we had reports of on oops (due to
1665 it being zero) during stress testcases so we need to check for it */
1667 if (cifs_inode
== NULL
) {
1668 cERROR(1, "Null inode passed to cifs_writeable_file");
1673 cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1675 /* only filter by fsuid on multiuser mounts */
1676 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1679 spin_lock(&cifs_file_list_lock
);
1681 if (refind
> MAX_REOPEN_ATT
) {
1682 spin_unlock(&cifs_file_list_lock
);
1685 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1686 if (!any_available
&& open_file
->pid
!= current
->tgid
)
1688 if (fsuid_only
&& open_file
->uid
!= current_fsuid())
1690 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
1691 if (!open_file
->invalidHandle
) {
1692 /* found a good writable file */
1693 cifsFileInfo_get_locked(open_file
);
1694 spin_unlock(&cifs_file_list_lock
);
1698 inv_file
= open_file
;
1702 /* couldn't find useable FH with same pid, try any available */
1703 if (!any_available
) {
1704 any_available
= true;
1705 goto refind_writable
;
1709 any_available
= false;
1710 cifsFileInfo_get_locked(inv_file
);
1713 spin_unlock(&cifs_file_list_lock
);
1716 rc
= cifs_reopen_file(inv_file
, false);
1720 spin_lock(&cifs_file_list_lock
);
1721 list_move_tail(&inv_file
->flist
,
1722 &cifs_inode
->openFileList
);
1723 spin_unlock(&cifs_file_list_lock
);
1724 cifsFileInfo_put(inv_file
);
1725 spin_lock(&cifs_file_list_lock
);
1727 goto refind_writable
;
1734 static int cifs_partialpagewrite(struct page
*page
, unsigned from
, unsigned to
)
1736 struct address_space
*mapping
= page
->mapping
;
1737 loff_t offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
1740 int bytes_written
= 0;
1741 struct inode
*inode
;
1742 struct cifsFileInfo
*open_file
;
1744 if (!mapping
|| !mapping
->host
)
1747 inode
= page
->mapping
->host
;
1749 offset
+= (loff_t
)from
;
1750 write_data
= kmap(page
);
1753 if ((to
> PAGE_CACHE_SIZE
) || (from
> to
)) {
1758 /* racing with truncate? */
1759 if (offset
> mapping
->host
->i_size
) {
1761 return 0; /* don't care */
1764 /* check to make sure that we are not extending the file */
1765 if (mapping
->host
->i_size
- offset
< (loff_t
)to
)
1766 to
= (unsigned)(mapping
->host
->i_size
- offset
);
1768 open_file
= find_writable_file(CIFS_I(mapping
->host
), false);
1770 bytes_written
= cifs_write(open_file
, open_file
->pid
,
1771 write_data
, to
- from
, &offset
);
1772 cifsFileInfo_put(open_file
);
1773 /* Does mm or vfs already set times? */
1774 inode
->i_atime
= inode
->i_mtime
= current_fs_time(inode
->i_sb
);
1775 if ((bytes_written
> 0) && (offset
))
1777 else if (bytes_written
< 0)
1780 cFYI(1, "No writeable filehandles for inode");
1788 static int cifs_writepages(struct address_space
*mapping
,
1789 struct writeback_control
*wbc
)
1791 struct cifs_sb_info
*cifs_sb
= CIFS_SB(mapping
->host
->i_sb
);
1792 bool done
= false, scanned
= false, range_whole
= false;
1794 struct cifs_writedata
*wdata
;
1795 struct TCP_Server_Info
*server
;
1798 loff_t isize
= i_size_read(mapping
->host
);
1801 * If wsize is smaller than the page cache size, default to writing
1802 * one page at a time via cifs_writepage
1804 if (cifs_sb
->wsize
< PAGE_CACHE_SIZE
)
1805 return generic_writepages(mapping
, wbc
);
1807 if (wbc
->range_cyclic
) {
1808 index
= mapping
->writeback_index
; /* Start from prev offset */
1811 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
1812 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
1813 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
1818 while (!done
&& index
<= end
) {
1819 unsigned int i
, nr_pages
, found_pages
;
1820 pgoff_t next
= 0, tofind
;
1821 struct page
**pages
;
1823 tofind
= min((cifs_sb
->wsize
/ PAGE_CACHE_SIZE
) - 1,
1826 wdata
= cifs_writedata_alloc((unsigned int)tofind
,
1827 cifs_writev_complete
);
1834 * find_get_pages_tag seems to return a max of 256 on each
1835 * iteration, so we must call it several times in order to
1836 * fill the array or the wsize is effectively limited to
1837 * 256 * PAGE_CACHE_SIZE.
1840 pages
= wdata
->pages
;
1842 nr_pages
= find_get_pages_tag(mapping
, &index
,
1843 PAGECACHE_TAG_DIRTY
,
1845 found_pages
+= nr_pages
;
1848 } while (nr_pages
&& tofind
&& index
<= end
);
1850 if (found_pages
== 0) {
1851 kref_put(&wdata
->refcount
, cifs_writedata_release
);
1856 for (i
= 0; i
< found_pages
; i
++) {
1857 page
= wdata
->pages
[i
];
1859 * At this point we hold neither mapping->tree_lock nor
1860 * lock on the page itself: the page may be truncated or
1861 * invalidated (changing page->mapping to NULL), or even
1862 * swizzled back from swapper_space to tmpfs file
1868 else if (!trylock_page(page
))
1871 if (unlikely(page
->mapping
!= mapping
)) {
1876 if (!wbc
->range_cyclic
&& page
->index
> end
) {
1882 if (next
&& (page
->index
!= next
)) {
1883 /* Not next consecutive page */
1888 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
1889 wait_on_page_writeback(page
);
1891 if (PageWriteback(page
) ||
1892 !clear_page_dirty_for_io(page
)) {
1898 * This actually clears the dirty bit in the radix tree.
1899 * See cifs_writepage() for more commentary.
1901 set_page_writeback(page
);
1903 if (page_offset(page
) >= isize
) {
1906 end_page_writeback(page
);
1910 wdata
->pages
[i
] = page
;
1911 next
= page
->index
+ 1;
1915 /* reset index to refind any pages skipped */
1917 index
= wdata
->pages
[0]->index
+ 1;
1919 /* put any pages we aren't going to use */
1920 for (i
= nr_pages
; i
< found_pages
; i
++) {
1921 page_cache_release(wdata
->pages
[i
]);
1922 wdata
->pages
[i
] = NULL
;
1925 /* nothing to write? */
1926 if (nr_pages
== 0) {
1927 kref_put(&wdata
->refcount
, cifs_writedata_release
);
1931 wdata
->sync_mode
= wbc
->sync_mode
;
1932 wdata
->nr_pages
= nr_pages
;
1933 wdata
->offset
= page_offset(wdata
->pages
[0]);
1934 wdata
->pagesz
= PAGE_CACHE_SIZE
;
1936 min(isize
- page_offset(wdata
->pages
[nr_pages
- 1]),
1937 (loff_t
)PAGE_CACHE_SIZE
);
1938 wdata
->bytes
= ((nr_pages
- 1) * PAGE_CACHE_SIZE
) +
1942 if (wdata
->cfile
!= NULL
)
1943 cifsFileInfo_put(wdata
->cfile
);
1944 wdata
->cfile
= find_writable_file(CIFS_I(mapping
->host
),
1946 if (!wdata
->cfile
) {
1947 cERROR(1, "No writable handles for inode");
1951 wdata
->pid
= wdata
->cfile
->pid
;
1952 server
= tlink_tcon(wdata
->cfile
->tlink
)->ses
->server
;
1953 rc
= server
->ops
->async_writev(wdata
);
1954 } while (wbc
->sync_mode
== WB_SYNC_ALL
&& rc
== -EAGAIN
);
1956 for (i
= 0; i
< nr_pages
; ++i
)
1957 unlock_page(wdata
->pages
[i
]);
1959 /* send failure -- clean up the mess */
1961 for (i
= 0; i
< nr_pages
; ++i
) {
1963 redirty_page_for_writepage(wbc
,
1966 SetPageError(wdata
->pages
[i
]);
1967 end_page_writeback(wdata
->pages
[i
]);
1968 page_cache_release(wdata
->pages
[i
]);
1971 mapping_set_error(mapping
, rc
);
1973 kref_put(&wdata
->refcount
, cifs_writedata_release
);
1975 wbc
->nr_to_write
-= nr_pages
;
1976 if (wbc
->nr_to_write
<= 0)
1982 if (!scanned
&& !done
) {
1984 * We hit the last page and there is more work to be done: wrap
1985 * back to the start of the file
1992 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
1993 mapping
->writeback_index
= index
;
1999 cifs_writepage_locked(struct page
*page
, struct writeback_control
*wbc
)
2005 /* BB add check for wbc flags */
2006 page_cache_get(page
);
2007 if (!PageUptodate(page
))
2008 cFYI(1, "ppw - page not up to date");
2011 * Set the "writeback" flag, and clear "dirty" in the radix tree.
2013 * A writepage() implementation always needs to do either this,
2014 * or re-dirty the page with "redirty_page_for_writepage()" in
2015 * the case of a failure.
2017 * Just unlocking the page will cause the radix tree tag-bits
2018 * to fail to update with the state of the page correctly.
2020 set_page_writeback(page
);
2022 rc
= cifs_partialpagewrite(page
, 0, PAGE_CACHE_SIZE
);
2023 if (rc
== -EAGAIN
&& wbc
->sync_mode
== WB_SYNC_ALL
)
2025 else if (rc
== -EAGAIN
)
2026 redirty_page_for_writepage(wbc
, page
);
2030 SetPageUptodate(page
);
2031 end_page_writeback(page
);
2032 page_cache_release(page
);
2037 static int cifs_writepage(struct page
*page
, struct writeback_control
*wbc
)
2039 int rc
= cifs_writepage_locked(page
, wbc
);
2044 static int cifs_write_end(struct file
*file
, struct address_space
*mapping
,
2045 loff_t pos
, unsigned len
, unsigned copied
,
2046 struct page
*page
, void *fsdata
)
2049 struct inode
*inode
= mapping
->host
;
2050 struct cifsFileInfo
*cfile
= file
->private_data
;
2051 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cfile
->dentry
->d_sb
);
2054 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2057 pid
= current
->tgid
;
2059 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
2062 if (PageChecked(page
)) {
2064 SetPageUptodate(page
);
2065 ClearPageChecked(page
);
2066 } else if (!PageUptodate(page
) && copied
== PAGE_CACHE_SIZE
)
2067 SetPageUptodate(page
);
2069 if (!PageUptodate(page
)) {
2071 unsigned offset
= pos
& (PAGE_CACHE_SIZE
- 1);
2075 /* this is probably better than directly calling
2076 partialpage_write since in this function the file handle is
2077 known which we might as well leverage */
2078 /* BB check if anything else missing out of ppw
2079 such as updating last write time */
2080 page_data
= kmap(page
);
2081 rc
= cifs_write(cfile
, pid
, page_data
+ offset
, copied
, &pos
);
2082 /* if (rc < 0) should we set writebehind rc? */
2089 set_page_dirty(page
);
2093 spin_lock(&inode
->i_lock
);
2094 if (pos
> inode
->i_size
)
2095 i_size_write(inode
, pos
);
2096 spin_unlock(&inode
->i_lock
);
2100 page_cache_release(page
);
2105 int cifs_strict_fsync(struct file
*file
, loff_t start
, loff_t end
,
2110 struct cifs_tcon
*tcon
;
2111 struct TCP_Server_Info
*server
;
2112 struct cifsFileInfo
*smbfile
= file
->private_data
;
2113 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
2114 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
2116 rc
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
2119 mutex_lock(&inode
->i_mutex
);
2123 cFYI(1, "Sync file - name: %s datasync: 0x%x",
2124 file
->f_path
.dentry
->d_name
.name
, datasync
);
2126 if (!CIFS_I(inode
)->clientCanCacheRead
) {
2127 rc
= cifs_invalidate_mapping(inode
);
2129 cFYI(1, "rc: %d during invalidate phase", rc
);
2130 rc
= 0; /* don't care about it in fsync */
2134 tcon
= tlink_tcon(smbfile
->tlink
);
2135 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOSSYNC
)) {
2136 server
= tcon
->ses
->server
;
2137 if (server
->ops
->flush
)
2138 rc
= server
->ops
->flush(xid
, tcon
, &smbfile
->fid
);
2144 mutex_unlock(&inode
->i_mutex
);
2148 int cifs_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
2152 struct cifs_tcon
*tcon
;
2153 struct TCP_Server_Info
*server
;
2154 struct cifsFileInfo
*smbfile
= file
->private_data
;
2155 struct cifs_sb_info
*cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
2156 struct inode
*inode
= file
->f_mapping
->host
;
2158 rc
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
2161 mutex_lock(&inode
->i_mutex
);
2165 cFYI(1, "Sync file - name: %s datasync: 0x%x",
2166 file
->f_path
.dentry
->d_name
.name
, datasync
);
2168 tcon
= tlink_tcon(smbfile
->tlink
);
2169 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOSSYNC
)) {
2170 server
= tcon
->ses
->server
;
2171 if (server
->ops
->flush
)
2172 rc
= server
->ops
->flush(xid
, tcon
, &smbfile
->fid
);
2178 mutex_unlock(&inode
->i_mutex
);
2183 * As file closes, flush all cached write data for this inode checking
2184 * for write behind errors.
2186 int cifs_flush(struct file
*file
, fl_owner_t id
)
2188 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
2191 if (file
->f_mode
& FMODE_WRITE
)
2192 rc
= filemap_write_and_wait(inode
->i_mapping
);
2194 cFYI(1, "Flush inode %p file %p rc %d", inode
, file
, rc
);
2200 cifs_write_allocate_pages(struct page
**pages
, unsigned long num_pages
)
2205 for (i
= 0; i
< num_pages
; i
++) {
2206 pages
[i
] = alloc_page(GFP_KERNEL
|__GFP_HIGHMEM
);
2209 * save number of pages we have already allocated and
2210 * return with ENOMEM error
2219 for (i
= 0; i
< num_pages
; i
++)
2226 size_t get_numpages(const size_t wsize
, const size_t len
, size_t *cur_len
)
2231 clen
= min_t(const size_t, len
, wsize
);
2232 num_pages
= DIV_ROUND_UP(clen
, PAGE_SIZE
);
2241 cifs_uncached_writev_complete(struct work_struct
*work
)
2244 struct cifs_writedata
*wdata
= container_of(work
,
2245 struct cifs_writedata
, work
);
2246 struct inode
*inode
= wdata
->cfile
->dentry
->d_inode
;
2247 struct cifsInodeInfo
*cifsi
= CIFS_I(inode
);
2249 spin_lock(&inode
->i_lock
);
2250 cifs_update_eof(cifsi
, wdata
->offset
, wdata
->bytes
);
2251 if (cifsi
->server_eof
> inode
->i_size
)
2252 i_size_write(inode
, cifsi
->server_eof
);
2253 spin_unlock(&inode
->i_lock
);
2255 complete(&wdata
->done
);
2257 if (wdata
->result
!= -EAGAIN
) {
2258 for (i
= 0; i
< wdata
->nr_pages
; i
++)
2259 put_page(wdata
->pages
[i
]);
2262 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2265 /* attempt to send write to server, retry on any -EAGAIN errors */
2267 cifs_uncached_retry_writev(struct cifs_writedata
*wdata
)
2270 struct TCP_Server_Info
*server
;
2272 server
= tlink_tcon(wdata
->cfile
->tlink
)->ses
->server
;
2275 if (wdata
->cfile
->invalidHandle
) {
2276 rc
= cifs_reopen_file(wdata
->cfile
, false);
2280 rc
= server
->ops
->async_writev(wdata
);
2281 } while (rc
== -EAGAIN
);
2287 cifs_iovec_write(struct file
*file
, const struct iovec
*iov
,
2288 unsigned long nr_segs
, loff_t
*poffset
)
2290 unsigned long nr_pages
, i
;
2291 size_t copied
, len
, cur_len
;
2292 ssize_t total_written
= 0;
2295 struct cifsFileInfo
*open_file
;
2296 struct cifs_tcon
*tcon
;
2297 struct cifs_sb_info
*cifs_sb
;
2298 struct cifs_writedata
*wdata
, *tmp
;
2299 struct list_head wdata_list
;
2303 len
= iov_length(iov
, nr_segs
);
2307 rc
= generic_write_checks(file
, poffset
, &len
, 0);
2311 INIT_LIST_HEAD(&wdata_list
);
2312 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
2313 open_file
= file
->private_data
;
2314 tcon
= tlink_tcon(open_file
->tlink
);
2316 if (!tcon
->ses
->server
->ops
->async_writev
)
2321 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2322 pid
= open_file
->pid
;
2324 pid
= current
->tgid
;
2326 iov_iter_init(&it
, iov
, nr_segs
, len
, 0);
2330 nr_pages
= get_numpages(cifs_sb
->wsize
, len
, &cur_len
);
2331 wdata
= cifs_writedata_alloc(nr_pages
,
2332 cifs_uncached_writev_complete
);
2338 rc
= cifs_write_allocate_pages(wdata
->pages
, nr_pages
);
2345 for (i
= 0; i
< nr_pages
; i
++) {
2346 copied
= min_t(const size_t, cur_len
, PAGE_SIZE
);
2347 copied
= iov_iter_copy_from_user(wdata
->pages
[i
], &it
,
2350 iov_iter_advance(&it
, copied
);
2352 cur_len
= save_len
- cur_len
;
2354 wdata
->sync_mode
= WB_SYNC_ALL
;
2355 wdata
->nr_pages
= nr_pages
;
2356 wdata
->offset
= (__u64
)offset
;
2357 wdata
->cfile
= cifsFileInfo_get(open_file
);
2359 wdata
->bytes
= cur_len
;
2360 wdata
->pagesz
= PAGE_SIZE
;
2361 wdata
->tailsz
= cur_len
- ((nr_pages
- 1) * PAGE_SIZE
);
2362 rc
= cifs_uncached_retry_writev(wdata
);
2364 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2368 list_add_tail(&wdata
->list
, &wdata_list
);
2374 * If at least one write was successfully sent, then discard any rc
2375 * value from the later writes. If the other write succeeds, then
2376 * we'll end up returning whatever was written. If it fails, then
2377 * we'll get a new rc value from that.
2379 if (!list_empty(&wdata_list
))
2383 * Wait for and collect replies for any successful sends in order of
2384 * increasing offset. Once an error is hit or we get a fatal signal
2385 * while waiting, then return without waiting for any more replies.
2388 list_for_each_entry_safe(wdata
, tmp
, &wdata_list
, list
) {
2390 /* FIXME: freezable too? */
2391 rc
= wait_for_completion_killable(&wdata
->done
);
2394 else if (wdata
->result
)
2397 total_written
+= wdata
->bytes
;
2399 /* resend call if it's a retryable error */
2400 if (rc
== -EAGAIN
) {
2401 rc
= cifs_uncached_retry_writev(wdata
);
2405 list_del_init(&wdata
->list
);
2406 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2409 if (total_written
> 0)
2410 *poffset
+= total_written
;
2412 cifs_stats_bytes_written(tcon
, total_written
);
2413 return total_written
? total_written
: (ssize_t
)rc
;
2416 ssize_t
cifs_user_writev(struct kiocb
*iocb
, const struct iovec
*iov
,
2417 unsigned long nr_segs
, loff_t pos
)
2420 struct inode
*inode
;
2422 inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2425 * BB - optimize the way when signing is disabled. We can drop this
2426 * extra memory-to-memory copying and use iovec buffers for constructing
2430 written
= cifs_iovec_write(iocb
->ki_filp
, iov
, nr_segs
, &pos
);
2432 CIFS_I(inode
)->invalid_mapping
= true;
2440 cifs_writev(struct kiocb
*iocb
, const struct iovec
*iov
,
2441 unsigned long nr_segs
, loff_t pos
)
2443 struct file
*file
= iocb
->ki_filp
;
2444 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
2445 struct inode
*inode
= file
->f_mapping
->host
;
2446 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
2447 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
2448 ssize_t rc
= -EACCES
;
2450 BUG_ON(iocb
->ki_pos
!= pos
);
2452 sb_start_write(inode
->i_sb
);
2455 * We need to hold the sem to be sure nobody modifies lock list
2456 * with a brlock that prevents writing.
2458 down_read(&cinode
->lock_sem
);
2459 if (!cifs_find_lock_conflict(cfile
, pos
, iov_length(iov
, nr_segs
),
2460 server
->vals
->exclusive_lock_type
, NULL
,
2462 mutex_lock(&inode
->i_mutex
);
2463 rc
= __generic_file_aio_write(iocb
, iov
, nr_segs
,
2465 mutex_unlock(&inode
->i_mutex
);
2468 if (rc
> 0 || rc
== -EIOCBQUEUED
) {
2471 err
= generic_write_sync(file
, pos
, rc
);
2472 if (err
< 0 && rc
> 0)
2476 up_read(&cinode
->lock_sem
);
2477 sb_end_write(inode
->i_sb
);
2482 cifs_strict_writev(struct kiocb
*iocb
, const struct iovec
*iov
,
2483 unsigned long nr_segs
, loff_t pos
)
2485 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2486 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
2487 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
2488 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)
2489 iocb
->ki_filp
->private_data
;
2490 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
2492 #ifdef CONFIG_CIFS_SMB2
2494 * If we have an oplock for read and want to write a data to the file
2495 * we need to store it in the page cache and then push it to the server
2496 * to be sure the next read will get a valid data.
2498 if (!cinode
->clientCanCacheAll
&& cinode
->clientCanCacheRead
) {
2502 written
= generic_file_aio_write(iocb
, iov
, nr_segs
, pos
);
2503 rc
= filemap_fdatawrite(inode
->i_mapping
);
2512 * For non-oplocked files in strict cache mode we need to write the data
2513 * to the server exactly from the pos to pos+len-1 rather than flush all
2514 * affected pages because it may cause a error with mandatory locks on
2515 * these pages but not on the region from pos to ppos+len-1.
2518 if (!cinode
->clientCanCacheAll
)
2519 return cifs_user_writev(iocb
, iov
, nr_segs
, pos
);
2521 if (cap_unix(tcon
->ses
) &&
2522 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
2523 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
2524 return generic_file_aio_write(iocb
, iov
, nr_segs
, pos
);
2526 return cifs_writev(iocb
, iov
, nr_segs
, pos
);
2529 static struct cifs_readdata
*
2530 cifs_readdata_alloc(unsigned int nr_pages
, work_func_t complete
)
2532 struct cifs_readdata
*rdata
;
2534 rdata
= kzalloc(sizeof(*rdata
) + (sizeof(struct page
*) * nr_pages
),
2536 if (rdata
!= NULL
) {
2537 kref_init(&rdata
->refcount
);
2538 INIT_LIST_HEAD(&rdata
->list
);
2539 init_completion(&rdata
->done
);
2540 INIT_WORK(&rdata
->work
, complete
);
2547 cifs_readdata_release(struct kref
*refcount
)
2549 struct cifs_readdata
*rdata
= container_of(refcount
,
2550 struct cifs_readdata
, refcount
);
2553 cifsFileInfo_put(rdata
->cfile
);
2559 cifs_read_allocate_pages(struct cifs_readdata
*rdata
, unsigned int nr_pages
)
2565 for (i
= 0; i
< nr_pages
; i
++) {
2566 page
= alloc_page(GFP_KERNEL
|__GFP_HIGHMEM
);
2571 rdata
->pages
[i
] = page
;
2575 for (i
= 0; i
< nr_pages
; i
++) {
2576 put_page(rdata
->pages
[i
]);
2577 rdata
->pages
[i
] = NULL
;
2584 cifs_uncached_readdata_release(struct kref
*refcount
)
2586 struct cifs_readdata
*rdata
= container_of(refcount
,
2587 struct cifs_readdata
, refcount
);
2590 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
2591 put_page(rdata
->pages
[i
]);
2592 rdata
->pages
[i
] = NULL
;
2594 cifs_readdata_release(refcount
);
2598 cifs_retry_async_readv(struct cifs_readdata
*rdata
)
2601 struct TCP_Server_Info
*server
;
2603 server
= tlink_tcon(rdata
->cfile
->tlink
)->ses
->server
;
2606 if (rdata
->cfile
->invalidHandle
) {
2607 rc
= cifs_reopen_file(rdata
->cfile
, true);
2611 rc
= server
->ops
->async_readv(rdata
);
2612 } while (rc
== -EAGAIN
);
2618 * cifs_readdata_to_iov - copy data from pages in response to an iovec
2619 * @rdata: the readdata response with list of pages holding data
2620 * @iov: vector in which we should copy the data
2621 * @nr_segs: number of segments in vector
2622 * @offset: offset into file of the first iovec
2623 * @copied: used to return the amount of data copied to the iov
2625 * This function copies data from a list of pages in a readdata response into
2626 * an array of iovecs. It will first calculate where the data should go
2627 * based on the info in the readdata and then copy the data into that spot.
2630 cifs_readdata_to_iov(struct cifs_readdata
*rdata
, const struct iovec
*iov
,
2631 unsigned long nr_segs
, loff_t offset
, ssize_t
*copied
)
2635 size_t pos
= rdata
->offset
- offset
;
2636 ssize_t remaining
= rdata
->bytes
;
2637 unsigned char *pdata
;
2640 /* set up iov_iter and advance to the correct offset */
2641 iov_iter_init(&ii
, iov
, nr_segs
, iov_length(iov
, nr_segs
), 0);
2642 iov_iter_advance(&ii
, pos
);
2645 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
2647 struct page
*page
= rdata
->pages
[i
];
2649 /* copy a whole page or whatever's left */
2650 copy
= min_t(ssize_t
, remaining
, PAGE_SIZE
);
2652 /* ...but limit it to whatever space is left in the iov */
2653 copy
= min_t(ssize_t
, copy
, iov_iter_count(&ii
));
2655 /* go while there's data to be copied and no errors */
2658 rc
= memcpy_toiovecend(ii
.iov
, pdata
, ii
.iov_offset
,
2664 iov_iter_advance(&ii
, copy
);
2673 cifs_uncached_readv_complete(struct work_struct
*work
)
2675 struct cifs_readdata
*rdata
= container_of(work
,
2676 struct cifs_readdata
, work
);
2678 complete(&rdata
->done
);
2679 kref_put(&rdata
->refcount
, cifs_uncached_readdata_release
);
2683 cifs_uncached_read_into_pages(struct TCP_Server_Info
*server
,
2684 struct cifs_readdata
*rdata
, unsigned int len
)
2686 int total_read
= 0, result
= 0;
2688 unsigned int nr_pages
= rdata
->nr_pages
;
2691 rdata
->tailsz
= PAGE_SIZE
;
2692 for (i
= 0; i
< nr_pages
; i
++) {
2693 struct page
*page
= rdata
->pages
[i
];
2695 if (len
>= PAGE_SIZE
) {
2696 /* enough data to fill the page */
2697 iov
.iov_base
= kmap(page
);
2698 iov
.iov_len
= PAGE_SIZE
;
2699 cFYI(1, "%u: iov_base=%p iov_len=%zu",
2700 i
, iov
.iov_base
, iov
.iov_len
);
2702 } else if (len
> 0) {
2703 /* enough for partial page, fill and zero the rest */
2704 iov
.iov_base
= kmap(page
);
2706 cFYI(1, "%u: iov_base=%p iov_len=%zu",
2707 i
, iov
.iov_base
, iov
.iov_len
);
2708 memset(iov
.iov_base
+ len
, '\0', PAGE_SIZE
- len
);
2709 rdata
->tailsz
= len
;
2712 /* no need to hold page hostage */
2713 rdata
->pages
[i
] = NULL
;
2719 result
= cifs_readv_from_socket(server
, &iov
, 1, iov
.iov_len
);
2724 total_read
+= result
;
2727 return total_read
> 0 ? total_read
: result
;
2731 cifs_iovec_read(struct file
*file
, const struct iovec
*iov
,
2732 unsigned long nr_segs
, loff_t
*poffset
)
2735 size_t len
, cur_len
;
2736 ssize_t total_read
= 0;
2737 loff_t offset
= *poffset
;
2738 unsigned int npages
;
2739 struct cifs_sb_info
*cifs_sb
;
2740 struct cifs_tcon
*tcon
;
2741 struct cifsFileInfo
*open_file
;
2742 struct cifs_readdata
*rdata
, *tmp
;
2743 struct list_head rdata_list
;
2749 len
= iov_length(iov
, nr_segs
);
2753 INIT_LIST_HEAD(&rdata_list
);
2754 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
2755 open_file
= file
->private_data
;
2756 tcon
= tlink_tcon(open_file
->tlink
);
2758 if (!tcon
->ses
->server
->ops
->async_readv
)
2761 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2762 pid
= open_file
->pid
;
2764 pid
= current
->tgid
;
2766 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
2767 cFYI(1, "attempting read on write only file instance");
2770 cur_len
= min_t(const size_t, len
- total_read
, cifs_sb
->rsize
);
2771 npages
= DIV_ROUND_UP(cur_len
, PAGE_SIZE
);
2773 /* allocate a readdata struct */
2774 rdata
= cifs_readdata_alloc(npages
,
2775 cifs_uncached_readv_complete
);
2781 rc
= cifs_read_allocate_pages(rdata
, npages
);
2785 rdata
->cfile
= cifsFileInfo_get(open_file
);
2786 rdata
->nr_pages
= npages
;
2787 rdata
->offset
= offset
;
2788 rdata
->bytes
= cur_len
;
2790 rdata
->pagesz
= PAGE_SIZE
;
2791 rdata
->read_into_pages
= cifs_uncached_read_into_pages
;
2793 rc
= cifs_retry_async_readv(rdata
);
2796 kref_put(&rdata
->refcount
,
2797 cifs_uncached_readdata_release
);
2801 list_add_tail(&rdata
->list
, &rdata_list
);
2806 /* if at least one read request send succeeded, then reset rc */
2807 if (!list_empty(&rdata_list
))
2810 /* the loop below should proceed in the order of increasing offsets */
2812 list_for_each_entry_safe(rdata
, tmp
, &rdata_list
, list
) {
2816 /* FIXME: freezable sleep too? */
2817 rc
= wait_for_completion_killable(&rdata
->done
);
2820 else if (rdata
->result
)
2823 rc
= cifs_readdata_to_iov(rdata
, iov
,
2826 total_read
+= copied
;
2829 /* resend call if it's a retryable error */
2830 if (rc
== -EAGAIN
) {
2831 rc
= cifs_retry_async_readv(rdata
);
2835 list_del_init(&rdata
->list
);
2836 kref_put(&rdata
->refcount
, cifs_uncached_readdata_release
);
2839 cifs_stats_bytes_read(tcon
, total_read
);
2840 *poffset
+= total_read
;
2842 /* mask nodata case */
2846 return total_read
? total_read
: rc
;
2849 ssize_t
cifs_user_readv(struct kiocb
*iocb
, const struct iovec
*iov
,
2850 unsigned long nr_segs
, loff_t pos
)
2854 read
= cifs_iovec_read(iocb
->ki_filp
, iov
, nr_segs
, &pos
);
2862 cifs_strict_readv(struct kiocb
*iocb
, const struct iovec
*iov
,
2863 unsigned long nr_segs
, loff_t pos
)
2865 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2866 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
2867 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
2868 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)
2869 iocb
->ki_filp
->private_data
;
2870 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
2874 * In strict cache mode we need to read from the server all the time
2875 * if we don't have level II oplock because the server can delay mtime
2876 * change - so we can't make a decision about inode invalidating.
2877 * And we can also fail with pagereading if there are mandatory locks
2878 * on pages affected by this read but not on the region from pos to
2881 if (!cinode
->clientCanCacheRead
)
2882 return cifs_user_readv(iocb
, iov
, nr_segs
, pos
);
2884 if (cap_unix(tcon
->ses
) &&
2885 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
2886 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
2887 return generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
2890 * We need to hold the sem to be sure nobody modifies lock list
2891 * with a brlock that prevents reading.
2893 down_read(&cinode
->lock_sem
);
2894 if (!cifs_find_lock_conflict(cfile
, pos
, iov_length(iov
, nr_segs
),
2895 tcon
->ses
->server
->vals
->shared_lock_type
,
2897 rc
= generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
2898 up_read(&cinode
->lock_sem
);
2903 cifs_read(struct file
*file
, char *read_data
, size_t read_size
, loff_t
*offset
)
2906 unsigned int bytes_read
= 0;
2907 unsigned int total_read
;
2908 unsigned int current_read_size
;
2910 struct cifs_sb_info
*cifs_sb
;
2911 struct cifs_tcon
*tcon
;
2912 struct TCP_Server_Info
*server
;
2915 struct cifsFileInfo
*open_file
;
2916 struct cifs_io_parms io_parms
;
2917 int buf_type
= CIFS_NO_BUFFER
;
2921 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
2923 /* FIXME: set up handlers for larger reads and/or convert to async */
2924 rsize
= min_t(unsigned int, cifs_sb
->rsize
, CIFSMaxBufSize
);
2926 if (file
->private_data
== NULL
) {
2931 open_file
= file
->private_data
;
2932 tcon
= tlink_tcon(open_file
->tlink
);
2933 server
= tcon
->ses
->server
;
2935 if (!server
->ops
->sync_read
) {
2940 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2941 pid
= open_file
->pid
;
2943 pid
= current
->tgid
;
2945 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
2946 cFYI(1, "attempting read on write only file instance");
2948 for (total_read
= 0, cur_offset
= read_data
; read_size
> total_read
;
2949 total_read
+= bytes_read
, cur_offset
+= bytes_read
) {
2950 current_read_size
= min_t(uint
, read_size
- total_read
, rsize
);
2952 * For windows me and 9x we do not want to request more than it
2953 * negotiated since it will refuse the read then.
2955 if ((tcon
->ses
) && !(tcon
->ses
->capabilities
&
2956 tcon
->ses
->server
->vals
->cap_large_files
)) {
2957 current_read_size
= min_t(uint
, current_read_size
,
2961 while (rc
== -EAGAIN
) {
2962 if (open_file
->invalidHandle
) {
2963 rc
= cifs_reopen_file(open_file
, true);
2968 io_parms
.tcon
= tcon
;
2969 io_parms
.offset
= *offset
;
2970 io_parms
.length
= current_read_size
;
2971 rc
= server
->ops
->sync_read(xid
, open_file
, &io_parms
,
2972 &bytes_read
, &cur_offset
,
2975 if (rc
|| (bytes_read
== 0)) {
2983 cifs_stats_bytes_read(tcon
, total_read
);
2984 *offset
+= bytes_read
;
2992 * If the page is mmap'ed into a process' page tables, then we need to make
2993 * sure that it doesn't change while being written back.
2996 cifs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2998 struct page
*page
= vmf
->page
;
3001 return VM_FAULT_LOCKED
;
3004 static struct vm_operations_struct cifs_file_vm_ops
= {
3005 .fault
= filemap_fault
,
3006 .page_mkwrite
= cifs_page_mkwrite
,
3009 int cifs_file_strict_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3012 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
3016 if (!CIFS_I(inode
)->clientCanCacheRead
) {
3017 rc
= cifs_invalidate_mapping(inode
);
3022 rc
= generic_file_mmap(file
, vma
);
3024 vma
->vm_ops
= &cifs_file_vm_ops
;
3029 int cifs_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3034 rc
= cifs_revalidate_file(file
);
3036 cFYI(1, "Validation prior to mmap failed, error=%d", rc
);
3040 rc
= generic_file_mmap(file
, vma
);
3042 vma
->vm_ops
= &cifs_file_vm_ops
;
3048 cifs_readv_complete(struct work_struct
*work
)
3051 struct cifs_readdata
*rdata
= container_of(work
,
3052 struct cifs_readdata
, work
);
3054 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
3055 struct page
*page
= rdata
->pages
[i
];
3057 lru_cache_add_file(page
);
3059 if (rdata
->result
== 0) {
3060 flush_dcache_page(page
);
3061 SetPageUptodate(page
);
3066 if (rdata
->result
== 0)
3067 cifs_readpage_to_fscache(rdata
->mapping
->host
, page
);
3069 page_cache_release(page
);
3070 rdata
->pages
[i
] = NULL
;
3072 kref_put(&rdata
->refcount
, cifs_readdata_release
);
3076 cifs_readpages_read_into_pages(struct TCP_Server_Info
*server
,
3077 struct cifs_readdata
*rdata
, unsigned int len
)
3079 int total_read
= 0, result
= 0;
3083 unsigned int nr_pages
= rdata
->nr_pages
;
3086 /* determine the eof that the server (probably) has */
3087 eof
= CIFS_I(rdata
->mapping
->host
)->server_eof
;
3088 eof_index
= eof
? (eof
- 1) >> PAGE_CACHE_SHIFT
: 0;
3089 cFYI(1, "eof=%llu eof_index=%lu", eof
, eof_index
);
3091 rdata
->tailsz
= PAGE_CACHE_SIZE
;
3092 for (i
= 0; i
< nr_pages
; i
++) {
3093 struct page
*page
= rdata
->pages
[i
];
3095 if (len
>= PAGE_CACHE_SIZE
) {
3096 /* enough data to fill the page */
3097 iov
.iov_base
= kmap(page
);
3098 iov
.iov_len
= PAGE_CACHE_SIZE
;
3099 cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
3100 i
, page
->index
, iov
.iov_base
, iov
.iov_len
);
3101 len
-= PAGE_CACHE_SIZE
;
3102 } else if (len
> 0) {
3103 /* enough for partial page, fill and zero the rest */
3104 iov
.iov_base
= kmap(page
);
3106 cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
3107 i
, page
->index
, iov
.iov_base
, iov
.iov_len
);
3108 memset(iov
.iov_base
+ len
,
3109 '\0', PAGE_CACHE_SIZE
- len
);
3110 rdata
->tailsz
= len
;
3112 } else if (page
->index
> eof_index
) {
3114 * The VFS will not try to do readahead past the
3115 * i_size, but it's possible that we have outstanding
3116 * writes with gaps in the middle and the i_size hasn't
3117 * caught up yet. Populate those with zeroed out pages
3118 * to prevent the VFS from repeatedly attempting to
3119 * fill them until the writes are flushed.
3121 zero_user(page
, 0, PAGE_CACHE_SIZE
);
3122 lru_cache_add_file(page
);
3123 flush_dcache_page(page
);
3124 SetPageUptodate(page
);
3126 page_cache_release(page
);
3127 rdata
->pages
[i
] = NULL
;
3131 /* no need to hold page hostage */
3132 lru_cache_add_file(page
);
3134 page_cache_release(page
);
3135 rdata
->pages
[i
] = NULL
;
3140 result
= cifs_readv_from_socket(server
, &iov
, 1, iov
.iov_len
);
3145 total_read
+= result
;
3148 return total_read
> 0 ? total_read
: result
;
3151 static int cifs_readpages(struct file
*file
, struct address_space
*mapping
,
3152 struct list_head
*page_list
, unsigned num_pages
)
3155 struct list_head tmplist
;
3156 struct cifsFileInfo
*open_file
= file
->private_data
;
3157 struct cifs_sb_info
*cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
3158 unsigned int rsize
= cifs_sb
->rsize
;
3162 * Give up immediately if rsize is too small to read an entire page.
3163 * The VFS will fall back to readpage. We should never reach this
3164 * point however since we set ra_pages to 0 when the rsize is smaller
3165 * than a cache page.
3167 if (unlikely(rsize
< PAGE_CACHE_SIZE
))
3171 * Reads as many pages as possible from fscache. Returns -ENOBUFS
3172 * immediately if the cookie is negative
3174 rc
= cifs_readpages_from_fscache(mapping
->host
, mapping
, page_list
,
3179 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
3180 pid
= open_file
->pid
;
3182 pid
= current
->tgid
;
3185 INIT_LIST_HEAD(&tmplist
);
3187 cFYI(1, "%s: file=%p mapping=%p num_pages=%u", __func__
, file
,
3188 mapping
, num_pages
);
3191 * Start with the page at end of list and move it to private
3192 * list. Do the same with any following pages until we hit
3193 * the rsize limit, hit an index discontinuity, or run out of
3194 * pages. Issue the async read and then start the loop again
3195 * until the list is empty.
3197 * Note that list order is important. The page_list is in
3198 * the order of declining indexes. When we put the pages in
3199 * the rdata->pages, then we want them in increasing order.
3201 while (!list_empty(page_list
)) {
3203 unsigned int bytes
= PAGE_CACHE_SIZE
;
3204 unsigned int expected_index
;
3205 unsigned int nr_pages
= 1;
3207 struct page
*page
, *tpage
;
3208 struct cifs_readdata
*rdata
;
3210 page
= list_entry(page_list
->prev
, struct page
, lru
);
3213 * Lock the page and put it in the cache. Since no one else
3214 * should have access to this page, we're safe to simply set
3215 * PG_locked without checking it first.
3217 __set_page_locked(page
);
3218 rc
= add_to_page_cache_locked(page
, mapping
,
3219 page
->index
, GFP_KERNEL
);
3221 /* give up if we can't stick it in the cache */
3223 __clear_page_locked(page
);
3227 /* move first page to the tmplist */
3228 offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
3229 list_move_tail(&page
->lru
, &tmplist
);
3231 /* now try and add more pages onto the request */
3232 expected_index
= page
->index
+ 1;
3233 list_for_each_entry_safe_reverse(page
, tpage
, page_list
, lru
) {
3234 /* discontinuity ? */
3235 if (page
->index
!= expected_index
)
3238 /* would this page push the read over the rsize? */
3239 if (bytes
+ PAGE_CACHE_SIZE
> rsize
)
3242 __set_page_locked(page
);
3243 if (add_to_page_cache_locked(page
, mapping
,
3244 page
->index
, GFP_KERNEL
)) {
3245 __clear_page_locked(page
);
3248 list_move_tail(&page
->lru
, &tmplist
);
3249 bytes
+= PAGE_CACHE_SIZE
;
3254 rdata
= cifs_readdata_alloc(nr_pages
, cifs_readv_complete
);
3256 /* best to give up if we're out of mem */
3257 list_for_each_entry_safe(page
, tpage
, &tmplist
, lru
) {
3258 list_del(&page
->lru
);
3259 lru_cache_add_file(page
);
3261 page_cache_release(page
);
3267 rdata
->cfile
= cifsFileInfo_get(open_file
);
3268 rdata
->mapping
= mapping
;
3269 rdata
->offset
= offset
;
3270 rdata
->bytes
= bytes
;
3272 rdata
->pagesz
= PAGE_CACHE_SIZE
;
3273 rdata
->read_into_pages
= cifs_readpages_read_into_pages
;
3275 list_for_each_entry_safe(page
, tpage
, &tmplist
, lru
) {
3276 list_del(&page
->lru
);
3277 rdata
->pages
[rdata
->nr_pages
++] = page
;
3280 rc
= cifs_retry_async_readv(rdata
);
3282 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
3283 page
= rdata
->pages
[i
];
3284 lru_cache_add_file(page
);
3286 page_cache_release(page
);
3288 kref_put(&rdata
->refcount
, cifs_readdata_release
);
3292 kref_put(&rdata
->refcount
, cifs_readdata_release
);
3298 static int cifs_readpage_worker(struct file
*file
, struct page
*page
,
3304 /* Is the page cached? */
3305 rc
= cifs_readpage_from_fscache(file
->f_path
.dentry
->d_inode
, page
);
3309 page_cache_get(page
);
3310 read_data
= kmap(page
);
3311 /* for reads over a certain size could initiate async read ahead */
3313 rc
= cifs_read(file
, read_data
, PAGE_CACHE_SIZE
, poffset
);
3318 cFYI(1, "Bytes read %d", rc
);
3320 file
->f_path
.dentry
->d_inode
->i_atime
=
3321 current_fs_time(file
->f_path
.dentry
->d_inode
->i_sb
);
3323 if (PAGE_CACHE_SIZE
> rc
)
3324 memset(read_data
+ rc
, 0, PAGE_CACHE_SIZE
- rc
);
3326 flush_dcache_page(page
);
3327 SetPageUptodate(page
);
3329 /* send this page to the cache */
3330 cifs_readpage_to_fscache(file
->f_path
.dentry
->d_inode
, page
);
3336 page_cache_release(page
);
3342 static int cifs_readpage(struct file
*file
, struct page
*page
)
3344 loff_t offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
3350 if (file
->private_data
== NULL
) {
3356 cFYI(1, "readpage %p at offset %d 0x%x",
3357 page
, (int)offset
, (int)offset
);
3359 rc
= cifs_readpage_worker(file
, page
, &offset
);
3367 static int is_inode_writable(struct cifsInodeInfo
*cifs_inode
)
3369 struct cifsFileInfo
*open_file
;
3371 spin_lock(&cifs_file_list_lock
);
3372 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
3373 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
3374 spin_unlock(&cifs_file_list_lock
);
3378 spin_unlock(&cifs_file_list_lock
);
3382 /* We do not want to update the file size from server for inodes
3383 open for write - to avoid races with writepage extending
3384 the file - in the future we could consider allowing
3385 refreshing the inode only on increases in the file size
3386 but this is tricky to do without racing with writebehind
3387 page caching in the current Linux kernel design */
3388 bool is_size_safe_to_change(struct cifsInodeInfo
*cifsInode
, __u64 end_of_file
)
3393 if (is_inode_writable(cifsInode
)) {
3394 /* This inode is open for write at least once */
3395 struct cifs_sb_info
*cifs_sb
;
3397 cifs_sb
= CIFS_SB(cifsInode
->vfs_inode
.i_sb
);
3398 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_DIRECT_IO
) {
3399 /* since no page cache to corrupt on directio
3400 we can change size safely */
3404 if (i_size_read(&cifsInode
->vfs_inode
) < end_of_file
)
3412 static int cifs_write_begin(struct file
*file
, struct address_space
*mapping
,
3413 loff_t pos
, unsigned len
, unsigned flags
,
3414 struct page
**pagep
, void **fsdata
)
3416 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
3417 loff_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
3418 loff_t page_start
= pos
& PAGE_MASK
;
3423 cFYI(1, "write_begin from %lld len %d", (long long)pos
, len
);
3425 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3431 if (PageUptodate(page
))
3435 * If we write a full page it will be up to date, no need to read from
3436 * the server. If the write is short, we'll end up doing a sync write
3439 if (len
== PAGE_CACHE_SIZE
)
3443 * optimize away the read when we have an oplock, and we're not
3444 * expecting to use any of the data we'd be reading in. That
3445 * is, when the page lies beyond the EOF, or straddles the EOF
3446 * and the write will cover all of the existing data.
3448 if (CIFS_I(mapping
->host
)->clientCanCacheRead
) {
3449 i_size
= i_size_read(mapping
->host
);
3450 if (page_start
>= i_size
||
3451 (offset
== 0 && (pos
+ len
) >= i_size
)) {
3452 zero_user_segments(page
, 0, offset
,
3456 * PageChecked means that the parts of the page
3457 * to which we're not writing are considered up
3458 * to date. Once the data is copied to the
3459 * page, it can be set uptodate.
3461 SetPageChecked(page
);
3466 if ((file
->f_flags
& O_ACCMODE
) != O_WRONLY
) {
3468 * might as well read a page, it is fast enough. If we get
3469 * an error, we don't need to return it. cifs_write_end will
3470 * do a sync write instead since PG_uptodate isn't set.
3472 cifs_readpage_worker(file
, page
, &page_start
);
3474 /* we could try using another file handle if there is one -
3475 but how would we lock it to prevent close of that handle
3476 racing with this read? In any case
3477 this will be written out by write_end so is fine */
3484 static int cifs_release_page(struct page
*page
, gfp_t gfp
)
3486 if (PagePrivate(page
))
3489 return cifs_fscache_release_page(page
, gfp
);
3492 static void cifs_invalidate_page(struct page
*page
, unsigned long offset
)
3494 struct cifsInodeInfo
*cifsi
= CIFS_I(page
->mapping
->host
);
3497 cifs_fscache_invalidate_page(page
, &cifsi
->vfs_inode
);
3500 static int cifs_launder_page(struct page
*page
)
3503 loff_t range_start
= page_offset(page
);
3504 loff_t range_end
= range_start
+ (loff_t
)(PAGE_CACHE_SIZE
- 1);
3505 struct writeback_control wbc
= {
3506 .sync_mode
= WB_SYNC_ALL
,
3508 .range_start
= range_start
,
3509 .range_end
= range_end
,
3512 cFYI(1, "Launder page: %p", page
);
3514 if (clear_page_dirty_for_io(page
))
3515 rc
= cifs_writepage_locked(page
, &wbc
);
3517 cifs_fscache_invalidate_page(page
, page
->mapping
->host
);
3521 void cifs_oplock_break(struct work_struct
*work
)
3523 struct cifsFileInfo
*cfile
= container_of(work
, struct cifsFileInfo
,
3525 struct inode
*inode
= cfile
->dentry
->d_inode
;
3526 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
3527 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
3530 if (inode
&& S_ISREG(inode
->i_mode
)) {
3531 if (cinode
->clientCanCacheRead
)
3532 break_lease(inode
, O_RDONLY
);
3534 break_lease(inode
, O_WRONLY
);
3535 rc
= filemap_fdatawrite(inode
->i_mapping
);
3536 if (cinode
->clientCanCacheRead
== 0) {
3537 rc
= filemap_fdatawait(inode
->i_mapping
);
3538 mapping_set_error(inode
->i_mapping
, rc
);
3539 invalidate_remote_inode(inode
);
3541 cFYI(1, "Oplock flush inode %p rc %d", inode
, rc
);
3544 rc
= cifs_push_locks(cfile
);
3546 cERROR(1, "Push locks rc = %d", rc
);
3549 * releasing stale oplock after recent reconnect of smb session using
3550 * a now incorrect file handle is not a data integrity issue but do
3551 * not bother sending an oplock release if session to server still is
3552 * disconnected since oplock already released by the server
3554 if (!cfile
->oplock_break_cancelled
) {
3555 rc
= tcon
->ses
->server
->ops
->oplock_response(tcon
, &cfile
->fid
,
3557 cFYI(1, "Oplock release rc = %d", rc
);
3561 const struct address_space_operations cifs_addr_ops
= {
3562 .readpage
= cifs_readpage
,
3563 .readpages
= cifs_readpages
,
3564 .writepage
= cifs_writepage
,
3565 .writepages
= cifs_writepages
,
3566 .write_begin
= cifs_write_begin
,
3567 .write_end
= cifs_write_end
,
3568 .set_page_dirty
= __set_page_dirty_nobuffers
,
3569 .releasepage
= cifs_release_page
,
3570 .invalidatepage
= cifs_invalidate_page
,
3571 .launder_page
= cifs_launder_page
,
3575 * cifs_readpages requires the server to support a buffer large enough to
3576 * contain the header plus one complete page of data. Otherwise, we need
3577 * to leave cifs_readpages out of the address space operations.
3579 const struct address_space_operations cifs_addr_ops_smallbuf
= {
3580 .readpage
= cifs_readpage
,
3581 .writepage
= cifs_writepage
,
3582 .writepages
= cifs_writepages
,
3583 .write_begin
= cifs_write_begin
,
3584 .write_end
= cifs_write_end
,
3585 .set_page_dirty
= __set_page_dirty_nobuffers
,
3586 .releasepage
= cifs_release_page
,
3587 .invalidatepage
= cifs_invalidate_page
,
3588 .launder_page
= cifs_launder_page
,