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
;
360 if (server
->ops
->close
)
361 server
->ops
->close(xid
, tcon
, &cifs_file
->fid
);
365 cifs_del_pending_open(&open
);
368 * Delete any outstanding lock records. We'll lose them when the file
371 down_write(&cifsi
->lock_sem
);
372 list_for_each_entry_safe(li
, tmp
, &cifs_file
->llist
->locks
, llist
) {
373 list_del(&li
->llist
);
374 cifs_del_lock_waiters(li
);
377 list_del(&cifs_file
->llist
->llist
);
378 kfree(cifs_file
->llist
);
379 up_write(&cifsi
->lock_sem
);
381 cifs_put_tlink(cifs_file
->tlink
);
382 dput(cifs_file
->dentry
);
386 int cifs_open(struct inode
*inode
, struct file
*file
)
392 struct cifs_sb_info
*cifs_sb
;
393 struct TCP_Server_Info
*server
;
394 struct cifs_tcon
*tcon
;
395 struct tcon_link
*tlink
;
396 struct cifsFileInfo
*cfile
= NULL
;
397 char *full_path
= NULL
;
398 bool posix_open_ok
= false;
400 struct cifs_pending_open open
;
404 cifs_sb
= CIFS_SB(inode
->i_sb
);
405 tlink
= cifs_sb_tlink(cifs_sb
);
408 return PTR_ERR(tlink
);
410 tcon
= tlink_tcon(tlink
);
411 server
= tcon
->ses
->server
;
413 full_path
= build_path_from_dentry(file
->f_path
.dentry
);
414 if (full_path
== NULL
) {
419 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
420 inode
, file
->f_flags
, full_path
);
427 if (!tcon
->broken_posix_open
&& tcon
->unix_ext
&&
428 cap_unix(tcon
->ses
) && (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
429 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
430 /* can not refresh inode info since size could be stale */
431 rc
= cifs_posix_open(full_path
, &inode
, inode
->i_sb
,
432 cifs_sb
->mnt_file_mode
/* ignored */,
433 file
->f_flags
, &oplock
, &fid
.netfid
, xid
);
435 cFYI(1, "posix open succeeded");
436 posix_open_ok
= true;
437 } else if ((rc
== -EINVAL
) || (rc
== -EOPNOTSUPP
)) {
438 if (tcon
->ses
->serverNOS
)
439 cERROR(1, "server %s of type %s returned"
440 " unexpected error on SMB posix open"
441 ", disabling posix open support."
442 " Check if server update available.",
443 tcon
->ses
->serverName
,
444 tcon
->ses
->serverNOS
);
445 tcon
->broken_posix_open
= true;
446 } else if ((rc
!= -EIO
) && (rc
!= -EREMOTE
) &&
447 (rc
!= -EOPNOTSUPP
)) /* path not found or net err */
450 * Else fallthrough to retry open the old way on network i/o
455 if (server
->ops
->get_lease_key
)
456 server
->ops
->get_lease_key(inode
, &fid
);
458 cifs_add_pending_open(&fid
, tlink
, &open
);
460 if (!posix_open_ok
) {
461 if (server
->ops
->get_lease_key
)
462 server
->ops
->get_lease_key(inode
, &fid
);
464 rc
= cifs_nt_open(full_path
, inode
, cifs_sb
, tcon
,
465 file
->f_flags
, &oplock
, &fid
, xid
);
467 cifs_del_pending_open(&open
);
472 cfile
= cifs_new_fileinfo(&fid
, file
, tlink
, oplock
);
474 if (server
->ops
->close
)
475 server
->ops
->close(xid
, tcon
, &fid
);
476 cifs_del_pending_open(&open
);
481 cifs_fscache_set_inode_cookie(inode
, file
);
483 if ((oplock
& CIFS_CREATE_ACTION
) && !posix_open_ok
&& tcon
->unix_ext
) {
485 * Time to set mode which we can not set earlier due to
486 * problems creating new read-only files.
488 struct cifs_unix_set_info_args args
= {
489 .mode
= inode
->i_mode
,
492 .ctime
= NO_CHANGE_64
,
493 .atime
= NO_CHANGE_64
,
494 .mtime
= NO_CHANGE_64
,
497 CIFSSMBUnixSetFileInfo(xid
, tcon
, &args
, fid
.netfid
,
504 cifs_put_tlink(tlink
);
508 static int cifs_push_posix_locks(struct cifsFileInfo
*cfile
);
511 * Try to reacquire byte range locks that were released when session
512 * to server was lost.
515 cifs_relock_file(struct cifsFileInfo
*cfile
)
517 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cfile
->dentry
->d_sb
);
518 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
519 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
522 /* we are going to update can_cache_brlcks here - need a write access */
523 down_write(&cinode
->lock_sem
);
524 if (cinode
->can_cache_brlcks
) {
525 /* can cache locks - no need to push them */
526 up_write(&cinode
->lock_sem
);
530 if (cap_unix(tcon
->ses
) &&
531 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
532 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
533 rc
= cifs_push_posix_locks(cfile
);
535 rc
= tcon
->ses
->server
->ops
->push_mand_locks(cfile
);
537 up_write(&cinode
->lock_sem
);
542 cifs_reopen_file(struct cifsFileInfo
*cfile
, bool can_flush
)
547 struct cifs_sb_info
*cifs_sb
;
548 struct cifs_tcon
*tcon
;
549 struct TCP_Server_Info
*server
;
550 struct cifsInodeInfo
*cinode
;
552 char *full_path
= NULL
;
554 int disposition
= FILE_OPEN
;
555 int create_options
= CREATE_NOT_DIR
;
559 mutex_lock(&cfile
->fh_mutex
);
560 if (!cfile
->invalidHandle
) {
561 mutex_unlock(&cfile
->fh_mutex
);
567 inode
= cfile
->dentry
->d_inode
;
568 cifs_sb
= CIFS_SB(inode
->i_sb
);
569 tcon
= tlink_tcon(cfile
->tlink
);
570 server
= tcon
->ses
->server
;
573 * Can not grab rename sem here because various ops, including those
574 * that already have the rename sem can end up causing writepage to get
575 * called and if the server was down that means we end up here, and we
576 * can never tell if the caller already has the rename_sem.
578 full_path
= build_path_from_dentry(cfile
->dentry
);
579 if (full_path
== NULL
) {
581 mutex_unlock(&cfile
->fh_mutex
);
586 cFYI(1, "inode = 0x%p file flags 0x%x for %s", inode
, cfile
->f_flags
,
589 if (tcon
->ses
->server
->oplocks
)
594 if (tcon
->unix_ext
&& cap_unix(tcon
->ses
) &&
595 (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
596 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
598 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
599 * original open. Must mask them off for a reopen.
601 unsigned int oflags
= cfile
->f_flags
&
602 ~(O_CREAT
| O_EXCL
| O_TRUNC
);
604 rc
= cifs_posix_open(full_path
, NULL
, inode
->i_sb
,
605 cifs_sb
->mnt_file_mode
/* ignored */,
606 oflags
, &oplock
, &fid
.netfid
, xid
);
608 cFYI(1, "posix reopen succeeded");
612 * fallthrough to retry open the old way on errors, especially
613 * in the reconnect path it is important to retry hard
617 desired_access
= cifs_convert_flags(cfile
->f_flags
);
619 if (backup_cred(cifs_sb
))
620 create_options
|= CREATE_OPEN_BACKUP_INTENT
;
622 if (server
->ops
->get_lease_key
)
623 server
->ops
->get_lease_key(inode
, &fid
);
626 * Can not refresh inode by passing in file_info buf to be returned by
627 * CIFSSMBOpen and then calling get_inode_info with returned buf since
628 * file might have write behind data that needs to be flushed and server
629 * version of file size can be stale. If we knew for sure that inode was
630 * not dirty locally we could do this.
632 rc
= server
->ops
->open(xid
, tcon
, full_path
, disposition
,
633 desired_access
, create_options
, &fid
, &oplock
,
636 mutex_unlock(&cfile
->fh_mutex
);
637 cFYI(1, "cifs_reopen returned 0x%x", rc
);
638 cFYI(1, "oplock: %d", oplock
);
639 goto reopen_error_exit
;
643 cfile
->invalidHandle
= false;
644 mutex_unlock(&cfile
->fh_mutex
);
645 cinode
= CIFS_I(inode
);
648 rc
= filemap_write_and_wait(inode
->i_mapping
);
649 mapping_set_error(inode
->i_mapping
, rc
);
652 rc
= cifs_get_inode_info_unix(&inode
, full_path
,
655 rc
= cifs_get_inode_info(&inode
, full_path
, NULL
,
656 inode
->i_sb
, xid
, NULL
);
659 * Else we are writing out data to server already and could deadlock if
660 * we tried to flush data, and since we do not know if we have data that
661 * would invalidate the current end of file on the server we can not go
662 * to the server to get the new inode info.
665 server
->ops
->set_fid(cfile
, &fid
, oplock
);
666 cifs_relock_file(cfile
);
674 int cifs_close(struct inode
*inode
, struct file
*file
)
676 if (file
->private_data
!= NULL
) {
677 cifsFileInfo_put(file
->private_data
);
678 file
->private_data
= NULL
;
681 /* return code from the ->release op is always ignored */
685 int cifs_closedir(struct inode
*inode
, struct file
*file
)
689 struct cifsFileInfo
*cfile
= file
->private_data
;
690 struct cifs_tcon
*tcon
;
691 struct TCP_Server_Info
*server
;
694 cFYI(1, "Closedir inode = 0x%p", inode
);
700 tcon
= tlink_tcon(cfile
->tlink
);
701 server
= tcon
->ses
->server
;
703 cFYI(1, "Freeing private data in close dir");
704 spin_lock(&cifs_file_list_lock
);
705 if (!cfile
->srch_inf
.endOfSearch
&& !cfile
->invalidHandle
) {
706 cfile
->invalidHandle
= true;
707 spin_unlock(&cifs_file_list_lock
);
708 if (server
->ops
->close_dir
)
709 rc
= server
->ops
->close_dir(xid
, tcon
, &cfile
->fid
);
712 cFYI(1, "Closing uncompleted readdir with rc %d", rc
);
713 /* not much we can do if it fails anyway, ignore rc */
716 spin_unlock(&cifs_file_list_lock
);
718 buf
= cfile
->srch_inf
.ntwrk_buf_start
;
720 cFYI(1, "closedir free smb buf in srch struct");
721 cfile
->srch_inf
.ntwrk_buf_start
= NULL
;
722 if (cfile
->srch_inf
.smallBuf
)
723 cifs_small_buf_release(buf
);
725 cifs_buf_release(buf
);
728 cifs_put_tlink(cfile
->tlink
);
729 kfree(file
->private_data
);
730 file
->private_data
= NULL
;
731 /* BB can we lock the filestruct while this is going on? */
736 static struct cifsLockInfo
*
737 cifs_lock_init(__u64 offset
, __u64 length
, __u8 type
)
739 struct cifsLockInfo
*lock
=
740 kmalloc(sizeof(struct cifsLockInfo
), GFP_KERNEL
);
743 lock
->offset
= offset
;
744 lock
->length
= length
;
746 lock
->pid
= current
->tgid
;
747 INIT_LIST_HEAD(&lock
->blist
);
748 init_waitqueue_head(&lock
->block_q
);
753 cifs_del_lock_waiters(struct cifsLockInfo
*lock
)
755 struct cifsLockInfo
*li
, *tmp
;
756 list_for_each_entry_safe(li
, tmp
, &lock
->blist
, blist
) {
757 list_del_init(&li
->blist
);
758 wake_up(&li
->block_q
);
762 #define CIFS_LOCK_OP 0
763 #define CIFS_READ_OP 1
764 #define CIFS_WRITE_OP 2
766 /* @rw_check : 0 - no op, 1 - read, 2 - write */
768 cifs_find_fid_lock_conflict(struct cifs_fid_locks
*fdlocks
, __u64 offset
,
769 __u64 length
, __u8 type
, struct cifsFileInfo
*cfile
,
770 struct cifsLockInfo
**conf_lock
, int rw_check
)
772 struct cifsLockInfo
*li
;
773 struct cifsFileInfo
*cur_cfile
= fdlocks
->cfile
;
774 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
776 list_for_each_entry(li
, &fdlocks
->locks
, llist
) {
777 if (offset
+ length
<= li
->offset
||
778 offset
>= li
->offset
+ li
->length
)
780 if (rw_check
!= CIFS_LOCK_OP
&& current
->tgid
== li
->pid
&&
781 server
->ops
->compare_fids(cfile
, cur_cfile
)) {
782 /* shared lock prevents write op through the same fid */
783 if (!(li
->type
& server
->vals
->shared_lock_type
) ||
784 rw_check
!= CIFS_WRITE_OP
)
787 if ((type
& server
->vals
->shared_lock_type
) &&
788 ((server
->ops
->compare_fids(cfile
, cur_cfile
) &&
789 current
->tgid
== li
->pid
) || type
== li
->type
))
799 cifs_find_lock_conflict(struct cifsFileInfo
*cfile
, __u64 offset
, __u64 length
,
800 __u8 type
, struct cifsLockInfo
**conf_lock
,
804 struct cifs_fid_locks
*cur
;
805 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
807 list_for_each_entry(cur
, &cinode
->llist
, llist
) {
808 rc
= cifs_find_fid_lock_conflict(cur
, offset
, length
, type
,
809 cfile
, conf_lock
, rw_check
);
818 * Check if there is another lock that prevents us to set the lock (mandatory
819 * style). If such a lock exists, update the flock structure with its
820 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
821 * or leave it the same if we can't. Returns 0 if we don't need to request to
822 * the server or 1 otherwise.
825 cifs_lock_test(struct cifsFileInfo
*cfile
, __u64 offset
, __u64 length
,
826 __u8 type
, struct file_lock
*flock
)
829 struct cifsLockInfo
*conf_lock
;
830 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
831 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
834 down_read(&cinode
->lock_sem
);
836 exist
= cifs_find_lock_conflict(cfile
, offset
, length
, type
,
837 &conf_lock
, CIFS_LOCK_OP
);
839 flock
->fl_start
= conf_lock
->offset
;
840 flock
->fl_end
= conf_lock
->offset
+ conf_lock
->length
- 1;
841 flock
->fl_pid
= conf_lock
->pid
;
842 if (conf_lock
->type
& server
->vals
->shared_lock_type
)
843 flock
->fl_type
= F_RDLCK
;
845 flock
->fl_type
= F_WRLCK
;
846 } else if (!cinode
->can_cache_brlcks
)
849 flock
->fl_type
= F_UNLCK
;
851 up_read(&cinode
->lock_sem
);
856 cifs_lock_add(struct cifsFileInfo
*cfile
, struct cifsLockInfo
*lock
)
858 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
859 down_write(&cinode
->lock_sem
);
860 list_add_tail(&lock
->llist
, &cfile
->llist
->locks
);
861 up_write(&cinode
->lock_sem
);
865 * Set the byte-range lock (mandatory style). Returns:
866 * 1) 0, if we set the lock and don't need to request to the server;
867 * 2) 1, if no locks prevent us but we need to request to the server;
868 * 3) -EACCESS, if there is a lock that prevents us and wait is false.
871 cifs_lock_add_if(struct cifsFileInfo
*cfile
, struct cifsLockInfo
*lock
,
874 struct cifsLockInfo
*conf_lock
;
875 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
881 down_write(&cinode
->lock_sem
);
883 exist
= cifs_find_lock_conflict(cfile
, lock
->offset
, lock
->length
,
884 lock
->type
, &conf_lock
, CIFS_LOCK_OP
);
885 if (!exist
&& cinode
->can_cache_brlcks
) {
886 list_add_tail(&lock
->llist
, &cfile
->llist
->locks
);
887 up_write(&cinode
->lock_sem
);
896 list_add_tail(&lock
->blist
, &conf_lock
->blist
);
897 up_write(&cinode
->lock_sem
);
898 rc
= wait_event_interruptible(lock
->block_q
,
899 (lock
->blist
.prev
== &lock
->blist
) &&
900 (lock
->blist
.next
== &lock
->blist
));
903 down_write(&cinode
->lock_sem
);
904 list_del_init(&lock
->blist
);
907 up_write(&cinode
->lock_sem
);
912 * Check if there is another lock that prevents us to set the lock (posix
913 * style). If such a lock exists, update the flock structure with its
914 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
915 * or leave it the same if we can't. Returns 0 if we don't need to request to
916 * the server or 1 otherwise.
919 cifs_posix_lock_test(struct file
*file
, struct file_lock
*flock
)
922 struct cifsInodeInfo
*cinode
= CIFS_I(file
->f_path
.dentry
->d_inode
);
923 unsigned char saved_type
= flock
->fl_type
;
925 if ((flock
->fl_flags
& FL_POSIX
) == 0)
928 down_read(&cinode
->lock_sem
);
929 posix_test_lock(file
, flock
);
931 if (flock
->fl_type
== F_UNLCK
&& !cinode
->can_cache_brlcks
) {
932 flock
->fl_type
= saved_type
;
936 up_read(&cinode
->lock_sem
);
941 * Set the byte-range lock (posix style). Returns:
942 * 1) 0, if we set the lock and don't need to request to the server;
943 * 2) 1, if we need to request to the server;
944 * 3) <0, if the error occurs while setting the lock.
947 cifs_posix_lock_set(struct file
*file
, struct file_lock
*flock
)
949 struct cifsInodeInfo
*cinode
= CIFS_I(file
->f_path
.dentry
->d_inode
);
952 if ((flock
->fl_flags
& FL_POSIX
) == 0)
956 down_write(&cinode
->lock_sem
);
957 if (!cinode
->can_cache_brlcks
) {
958 up_write(&cinode
->lock_sem
);
962 rc
= posix_lock_file(file
, flock
, NULL
);
963 up_write(&cinode
->lock_sem
);
964 if (rc
== FILE_LOCK_DEFERRED
) {
965 rc
= wait_event_interruptible(flock
->fl_wait
, !flock
->fl_next
);
968 locks_delete_block(flock
);
974 cifs_push_mandatory_locks(struct cifsFileInfo
*cfile
)
977 int rc
= 0, stored_rc
;
978 struct cifsLockInfo
*li
, *tmp
;
979 struct cifs_tcon
*tcon
;
980 unsigned int num
, max_num
, max_buf
;
981 LOCKING_ANDX_RANGE
*buf
, *cur
;
982 int types
[] = {LOCKING_ANDX_LARGE_FILES
,
983 LOCKING_ANDX_SHARED_LOCK
| LOCKING_ANDX_LARGE_FILES
};
987 tcon
= tlink_tcon(cfile
->tlink
);
990 * Accessing maxBuf is racy with cifs_reconnect - need to store value
991 * and check it for zero before using.
993 max_buf
= tcon
->ses
->server
->maxBuf
;
999 max_num
= (max_buf
- sizeof(struct smb_hdr
)) /
1000 sizeof(LOCKING_ANDX_RANGE
);
1001 buf
= kzalloc(max_num
* sizeof(LOCKING_ANDX_RANGE
), GFP_KERNEL
);
1007 for (i
= 0; i
< 2; i
++) {
1010 list_for_each_entry_safe(li
, tmp
, &cfile
->llist
->locks
, llist
) {
1011 if (li
->type
!= types
[i
])
1013 cur
->Pid
= cpu_to_le16(li
->pid
);
1014 cur
->LengthLow
= cpu_to_le32((u32
)li
->length
);
1015 cur
->LengthHigh
= cpu_to_le32((u32
)(li
->length
>>32));
1016 cur
->OffsetLow
= cpu_to_le32((u32
)li
->offset
);
1017 cur
->OffsetHigh
= cpu_to_le32((u32
)(li
->offset
>>32));
1018 if (++num
== max_num
) {
1019 stored_rc
= cifs_lockv(xid
, tcon
,
1021 (__u8
)li
->type
, 0, num
,
1032 stored_rc
= cifs_lockv(xid
, tcon
, cfile
->fid
.netfid
,
1033 (__u8
)types
[i
], 0, num
, buf
);
1044 /* copied from fs/locks.c with a name change */
1045 #define cifs_for_each_lock(inode, lockp) \
1046 for (lockp = &inode->i_flock; *lockp != NULL; \
1047 lockp = &(*lockp)->fl_next)
1049 struct lock_to_push
{
1050 struct list_head llist
;
1059 cifs_push_posix_locks(struct cifsFileInfo
*cfile
)
1061 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1062 struct file_lock
*flock
, **before
;
1063 unsigned int count
= 0, i
= 0;
1064 int rc
= 0, xid
, type
;
1065 struct list_head locks_to_send
, *el
;
1066 struct lock_to_push
*lck
, *tmp
;
1072 cifs_for_each_lock(cfile
->dentry
->d_inode
, before
) {
1073 if ((*before
)->fl_flags
& FL_POSIX
)
1078 INIT_LIST_HEAD(&locks_to_send
);
1081 * Allocating count locks is enough because no FL_POSIX locks can be
1082 * added to the list while we are holding cinode->lock_sem that
1083 * protects locking operations of this inode.
1085 for (; i
< count
; i
++) {
1086 lck
= kmalloc(sizeof(struct lock_to_push
), GFP_KERNEL
);
1091 list_add_tail(&lck
->llist
, &locks_to_send
);
1094 el
= locks_to_send
.next
;
1096 cifs_for_each_lock(cfile
->dentry
->d_inode
, before
) {
1098 if ((flock
->fl_flags
& FL_POSIX
) == 0)
1100 if (el
== &locks_to_send
) {
1102 * The list ended. We don't have enough allocated
1103 * structures - something is really wrong.
1105 cERROR(1, "Can't push all brlocks!");
1108 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1109 if (flock
->fl_type
== F_RDLCK
|| flock
->fl_type
== F_SHLCK
)
1113 lck
= list_entry(el
, struct lock_to_push
, llist
);
1114 lck
->pid
= flock
->fl_pid
;
1115 lck
->netfid
= cfile
->fid
.netfid
;
1116 lck
->length
= length
;
1118 lck
->offset
= flock
->fl_start
;
1123 list_for_each_entry_safe(lck
, tmp
, &locks_to_send
, llist
) {
1126 stored_rc
= CIFSSMBPosixLock(xid
, tcon
, lck
->netfid
, lck
->pid
,
1127 lck
->offset
, lck
->length
, NULL
,
1131 list_del(&lck
->llist
);
1139 list_for_each_entry_safe(lck
, tmp
, &locks_to_send
, llist
) {
1140 list_del(&lck
->llist
);
1147 cifs_push_locks(struct cifsFileInfo
*cfile
)
1149 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cfile
->dentry
->d_sb
);
1150 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
1151 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1154 /* we are going to update can_cache_brlcks here - need a write access */
1155 down_write(&cinode
->lock_sem
);
1156 if (!cinode
->can_cache_brlcks
) {
1157 up_write(&cinode
->lock_sem
);
1161 if (cap_unix(tcon
->ses
) &&
1162 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
1163 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
1164 rc
= cifs_push_posix_locks(cfile
);
1166 rc
= tcon
->ses
->server
->ops
->push_mand_locks(cfile
);
1168 cinode
->can_cache_brlcks
= false;
1169 up_write(&cinode
->lock_sem
);
1174 cifs_read_flock(struct file_lock
*flock
, __u32
*type
, int *lock
, int *unlock
,
1175 bool *wait_flag
, struct TCP_Server_Info
*server
)
1177 if (flock
->fl_flags
& FL_POSIX
)
1179 if (flock
->fl_flags
& FL_FLOCK
)
1181 if (flock
->fl_flags
& FL_SLEEP
) {
1182 cFYI(1, "Blocking lock");
1185 if (flock
->fl_flags
& FL_ACCESS
)
1186 cFYI(1, "Process suspended by mandatory locking - "
1187 "not implemented yet");
1188 if (flock
->fl_flags
& FL_LEASE
)
1189 cFYI(1, "Lease on file - not implemented yet");
1190 if (flock
->fl_flags
&
1191 (~(FL_POSIX
| FL_FLOCK
| FL_SLEEP
|
1192 FL_ACCESS
| FL_LEASE
| FL_CLOSE
)))
1193 cFYI(1, "Unknown lock flags 0x%x", flock
->fl_flags
);
1195 *type
= server
->vals
->large_lock_type
;
1196 if (flock
->fl_type
== F_WRLCK
) {
1197 cFYI(1, "F_WRLCK ");
1198 *type
|= server
->vals
->exclusive_lock_type
;
1200 } else if (flock
->fl_type
== F_UNLCK
) {
1202 *type
|= server
->vals
->unlock_lock_type
;
1204 /* Check if unlock includes more than one lock range */
1205 } else if (flock
->fl_type
== F_RDLCK
) {
1207 *type
|= server
->vals
->shared_lock_type
;
1209 } else if (flock
->fl_type
== F_EXLCK
) {
1211 *type
|= server
->vals
->exclusive_lock_type
;
1213 } else if (flock
->fl_type
== F_SHLCK
) {
1215 *type
|= server
->vals
->shared_lock_type
;
1218 cFYI(1, "Unknown type of lock");
1222 cifs_getlk(struct file
*file
, struct file_lock
*flock
, __u32 type
,
1223 bool wait_flag
, bool posix_lck
, unsigned int xid
)
1226 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1227 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
1228 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1229 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
1230 __u16 netfid
= cfile
->fid
.netfid
;
1233 int posix_lock_type
;
1235 rc
= cifs_posix_lock_test(file
, flock
);
1239 if (type
& server
->vals
->shared_lock_type
)
1240 posix_lock_type
= CIFS_RDLCK
;
1242 posix_lock_type
= CIFS_WRLCK
;
1243 rc
= CIFSSMBPosixLock(xid
, tcon
, netfid
, current
->tgid
,
1244 flock
->fl_start
, length
, flock
,
1245 posix_lock_type
, wait_flag
);
1249 rc
= cifs_lock_test(cfile
, flock
->fl_start
, length
, type
, flock
);
1253 /* BB we could chain these into one lock request BB */
1254 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
, type
,
1257 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1259 flock
->fl_type
= F_UNLCK
;
1261 cERROR(1, "Error unlocking previously locked "
1262 "range %d during test of lock", rc
);
1266 if (type
& server
->vals
->shared_lock_type
) {
1267 flock
->fl_type
= F_WRLCK
;
1271 type
&= ~server
->vals
->exclusive_lock_type
;
1273 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1274 type
| server
->vals
->shared_lock_type
,
1277 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1278 type
| server
->vals
->shared_lock_type
, 0, 1, false);
1279 flock
->fl_type
= F_RDLCK
;
1281 cERROR(1, "Error unlocking previously locked "
1282 "range %d during test of lock", rc
);
1284 flock
->fl_type
= F_WRLCK
;
1290 cifs_move_llist(struct list_head
*source
, struct list_head
*dest
)
1292 struct list_head
*li
, *tmp
;
1293 list_for_each_safe(li
, tmp
, source
)
1294 list_move(li
, dest
);
1298 cifs_free_llist(struct list_head
*llist
)
1300 struct cifsLockInfo
*li
, *tmp
;
1301 list_for_each_entry_safe(li
, tmp
, llist
, llist
) {
1302 cifs_del_lock_waiters(li
);
1303 list_del(&li
->llist
);
1309 cifs_unlock_range(struct cifsFileInfo
*cfile
, struct file_lock
*flock
,
1312 int rc
= 0, stored_rc
;
1313 int types
[] = {LOCKING_ANDX_LARGE_FILES
,
1314 LOCKING_ANDX_SHARED_LOCK
| LOCKING_ANDX_LARGE_FILES
};
1316 unsigned int max_num
, num
, max_buf
;
1317 LOCKING_ANDX_RANGE
*buf
, *cur
;
1318 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1319 struct cifsInodeInfo
*cinode
= CIFS_I(cfile
->dentry
->d_inode
);
1320 struct cifsLockInfo
*li
, *tmp
;
1321 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1322 struct list_head tmp_llist
;
1324 INIT_LIST_HEAD(&tmp_llist
);
1327 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1328 * and check it for zero before using.
1330 max_buf
= tcon
->ses
->server
->maxBuf
;
1334 max_num
= (max_buf
- sizeof(struct smb_hdr
)) /
1335 sizeof(LOCKING_ANDX_RANGE
);
1336 buf
= kzalloc(max_num
* sizeof(LOCKING_ANDX_RANGE
), GFP_KERNEL
);
1340 down_write(&cinode
->lock_sem
);
1341 for (i
= 0; i
< 2; i
++) {
1344 list_for_each_entry_safe(li
, tmp
, &cfile
->llist
->locks
, llist
) {
1345 if (flock
->fl_start
> li
->offset
||
1346 (flock
->fl_start
+ length
) <
1347 (li
->offset
+ li
->length
))
1349 if (current
->tgid
!= li
->pid
)
1351 if (types
[i
] != li
->type
)
1353 if (cinode
->can_cache_brlcks
) {
1355 * We can cache brlock requests - simply remove
1356 * a lock from the file's list.
1358 list_del(&li
->llist
);
1359 cifs_del_lock_waiters(li
);
1363 cur
->Pid
= cpu_to_le16(li
->pid
);
1364 cur
->LengthLow
= cpu_to_le32((u32
)li
->length
);
1365 cur
->LengthHigh
= cpu_to_le32((u32
)(li
->length
>>32));
1366 cur
->OffsetLow
= cpu_to_le32((u32
)li
->offset
);
1367 cur
->OffsetHigh
= cpu_to_le32((u32
)(li
->offset
>>32));
1369 * We need to save a lock here to let us add it again to
1370 * the file's list if the unlock range request fails on
1373 list_move(&li
->llist
, &tmp_llist
);
1374 if (++num
== max_num
) {
1375 stored_rc
= cifs_lockv(xid
, tcon
,
1377 li
->type
, num
, 0, buf
);
1380 * We failed on the unlock range
1381 * request - add all locks from the tmp
1382 * list to the head of the file's list.
1384 cifs_move_llist(&tmp_llist
,
1385 &cfile
->llist
->locks
);
1389 * The unlock range request succeed -
1390 * free the tmp list.
1392 cifs_free_llist(&tmp_llist
);
1399 stored_rc
= cifs_lockv(xid
, tcon
, cfile
->fid
.netfid
,
1400 types
[i
], num
, 0, buf
);
1402 cifs_move_llist(&tmp_llist
,
1403 &cfile
->llist
->locks
);
1406 cifs_free_llist(&tmp_llist
);
1410 up_write(&cinode
->lock_sem
);
1416 cifs_setlk(struct file
*file
, struct file_lock
*flock
, __u32 type
,
1417 bool wait_flag
, bool posix_lck
, int lock
, int unlock
,
1421 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1422 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
1423 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1424 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
1427 int posix_lock_type
;
1429 rc
= cifs_posix_lock_set(file
, flock
);
1433 if (type
& server
->vals
->shared_lock_type
)
1434 posix_lock_type
= CIFS_RDLCK
;
1436 posix_lock_type
= CIFS_WRLCK
;
1439 posix_lock_type
= CIFS_UNLCK
;
1441 rc
= CIFSSMBPosixLock(xid
, tcon
, cfile
->fid
.netfid
,
1442 current
->tgid
, flock
->fl_start
, length
,
1443 NULL
, posix_lock_type
, wait_flag
);
1448 struct cifsLockInfo
*lock
;
1450 lock
= cifs_lock_init(flock
->fl_start
, length
, type
);
1454 rc
= cifs_lock_add_if(cfile
, lock
, wait_flag
);
1462 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1463 type
, 1, 0, wait_flag
);
1469 cifs_lock_add(cfile
, lock
);
1471 rc
= server
->ops
->mand_unlock_range(cfile
, flock
, xid
);
1474 if (flock
->fl_flags
& FL_POSIX
)
1475 posix_lock_file_wait(file
, flock
);
1479 int cifs_lock(struct file
*file
, int cmd
, struct file_lock
*flock
)
1482 int lock
= 0, unlock
= 0;
1483 bool wait_flag
= false;
1484 bool posix_lck
= false;
1485 struct cifs_sb_info
*cifs_sb
;
1486 struct cifs_tcon
*tcon
;
1487 struct cifsInodeInfo
*cinode
;
1488 struct cifsFileInfo
*cfile
;
1495 cFYI(1, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld "
1496 "end: %lld", cmd
, flock
->fl_flags
, flock
->fl_type
,
1497 flock
->fl_start
, flock
->fl_end
);
1499 cfile
= (struct cifsFileInfo
*)file
->private_data
;
1500 tcon
= tlink_tcon(cfile
->tlink
);
1502 cifs_read_flock(flock
, &type
, &lock
, &unlock
, &wait_flag
,
1505 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
1506 netfid
= cfile
->fid
.netfid
;
1507 cinode
= CIFS_I(file
->f_path
.dentry
->d_inode
);
1509 if (cap_unix(tcon
->ses
) &&
1510 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
1511 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
1514 * BB add code here to normalize offset and length to account for
1515 * negative length which we can not accept over the wire.
1517 if (IS_GETLK(cmd
)) {
1518 rc
= cifs_getlk(file
, flock
, type
, wait_flag
, posix_lck
, xid
);
1523 if (!lock
&& !unlock
) {
1525 * if no lock or unlock then nothing to do since we do not
1532 rc
= cifs_setlk(file
, flock
, type
, wait_flag
, posix_lck
, lock
, unlock
,
1539 * update the file size (if needed) after a write. Should be called with
1540 * the inode->i_lock held
1543 cifs_update_eof(struct cifsInodeInfo
*cifsi
, loff_t offset
,
1544 unsigned int bytes_written
)
1546 loff_t end_of_write
= offset
+ bytes_written
;
1548 if (end_of_write
> cifsi
->server_eof
)
1549 cifsi
->server_eof
= end_of_write
;
1553 cifs_write(struct cifsFileInfo
*open_file
, __u32 pid
, const char *write_data
,
1554 size_t write_size
, loff_t
*offset
)
1557 unsigned int bytes_written
= 0;
1558 unsigned int total_written
;
1559 struct cifs_sb_info
*cifs_sb
;
1560 struct cifs_tcon
*tcon
;
1561 struct TCP_Server_Info
*server
;
1563 struct dentry
*dentry
= open_file
->dentry
;
1564 struct cifsInodeInfo
*cifsi
= CIFS_I(dentry
->d_inode
);
1565 struct cifs_io_parms io_parms
;
1567 cifs_sb
= CIFS_SB(dentry
->d_sb
);
1569 cFYI(1, "write %zd bytes to offset %lld of %s", write_size
,
1570 *offset
, dentry
->d_name
.name
);
1572 tcon
= tlink_tcon(open_file
->tlink
);
1573 server
= tcon
->ses
->server
;
1575 if (!server
->ops
->sync_write
)
1580 for (total_written
= 0; write_size
> total_written
;
1581 total_written
+= bytes_written
) {
1583 while (rc
== -EAGAIN
) {
1587 if (open_file
->invalidHandle
) {
1588 /* we could deadlock if we called
1589 filemap_fdatawait from here so tell
1590 reopen_file not to flush data to
1592 rc
= cifs_reopen_file(open_file
, false);
1597 len
= min((size_t)cifs_sb
->wsize
,
1598 write_size
- total_written
);
1599 /* iov[0] is reserved for smb header */
1600 iov
[1].iov_base
= (char *)write_data
+ total_written
;
1601 iov
[1].iov_len
= len
;
1603 io_parms
.tcon
= tcon
;
1604 io_parms
.offset
= *offset
;
1605 io_parms
.length
= len
;
1606 rc
= server
->ops
->sync_write(xid
, open_file
, &io_parms
,
1607 &bytes_written
, iov
, 1);
1609 if (rc
|| (bytes_written
== 0)) {
1617 spin_lock(&dentry
->d_inode
->i_lock
);
1618 cifs_update_eof(cifsi
, *offset
, bytes_written
);
1619 spin_unlock(&dentry
->d_inode
->i_lock
);
1620 *offset
+= bytes_written
;
1624 cifs_stats_bytes_written(tcon
, total_written
);
1626 if (total_written
> 0) {
1627 spin_lock(&dentry
->d_inode
->i_lock
);
1628 if (*offset
> dentry
->d_inode
->i_size
)
1629 i_size_write(dentry
->d_inode
, *offset
);
1630 spin_unlock(&dentry
->d_inode
->i_lock
);
1632 mark_inode_dirty_sync(dentry
->d_inode
);
1634 return total_written
;
1637 struct cifsFileInfo
*find_readable_file(struct cifsInodeInfo
*cifs_inode
,
1640 struct cifsFileInfo
*open_file
= NULL
;
1641 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1643 /* only filter by fsuid on multiuser mounts */
1644 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1647 spin_lock(&cifs_file_list_lock
);
1648 /* we could simply get the first_list_entry since write-only entries
1649 are always at the end of the list but since the first entry might
1650 have a close pending, we go through the whole list */
1651 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1652 if (fsuid_only
&& open_file
->uid
!= current_fsuid())
1654 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_READ
) {
1655 if (!open_file
->invalidHandle
) {
1656 /* found a good file */
1657 /* lock it so it will not be closed on us */
1658 cifsFileInfo_get_locked(open_file
);
1659 spin_unlock(&cifs_file_list_lock
);
1661 } /* else might as well continue, and look for
1662 another, or simply have the caller reopen it
1663 again rather than trying to fix this handle */
1664 } else /* write only file */
1665 break; /* write only files are last so must be done */
1667 spin_unlock(&cifs_file_list_lock
);
1671 struct cifsFileInfo
*find_writable_file(struct cifsInodeInfo
*cifs_inode
,
1674 struct cifsFileInfo
*open_file
, *inv_file
= NULL
;
1675 struct cifs_sb_info
*cifs_sb
;
1676 bool any_available
= false;
1678 unsigned int refind
= 0;
1680 /* Having a null inode here (because mapping->host was set to zero by
1681 the VFS or MM) should not happen but we had reports of on oops (due to
1682 it being zero) during stress testcases so we need to check for it */
1684 if (cifs_inode
== NULL
) {
1685 cERROR(1, "Null inode passed to cifs_writeable_file");
1690 cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1692 /* only filter by fsuid on multiuser mounts */
1693 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1696 spin_lock(&cifs_file_list_lock
);
1698 if (refind
> MAX_REOPEN_ATT
) {
1699 spin_unlock(&cifs_file_list_lock
);
1702 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1703 if (!any_available
&& open_file
->pid
!= current
->tgid
)
1705 if (fsuid_only
&& open_file
->uid
!= current_fsuid())
1707 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
1708 if (!open_file
->invalidHandle
) {
1709 /* found a good writable file */
1710 cifsFileInfo_get_locked(open_file
);
1711 spin_unlock(&cifs_file_list_lock
);
1715 inv_file
= open_file
;
1719 /* couldn't find useable FH with same pid, try any available */
1720 if (!any_available
) {
1721 any_available
= true;
1722 goto refind_writable
;
1726 any_available
= false;
1727 cifsFileInfo_get_locked(inv_file
);
1730 spin_unlock(&cifs_file_list_lock
);
1733 rc
= cifs_reopen_file(inv_file
, false);
1737 spin_lock(&cifs_file_list_lock
);
1738 list_move_tail(&inv_file
->flist
,
1739 &cifs_inode
->openFileList
);
1740 spin_unlock(&cifs_file_list_lock
);
1741 cifsFileInfo_put(inv_file
);
1742 spin_lock(&cifs_file_list_lock
);
1744 goto refind_writable
;
1751 static int cifs_partialpagewrite(struct page
*page
, unsigned from
, unsigned to
)
1753 struct address_space
*mapping
= page
->mapping
;
1754 loff_t offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
1757 int bytes_written
= 0;
1758 struct inode
*inode
;
1759 struct cifsFileInfo
*open_file
;
1761 if (!mapping
|| !mapping
->host
)
1764 inode
= page
->mapping
->host
;
1766 offset
+= (loff_t
)from
;
1767 write_data
= kmap(page
);
1770 if ((to
> PAGE_CACHE_SIZE
) || (from
> to
)) {
1775 /* racing with truncate? */
1776 if (offset
> mapping
->host
->i_size
) {
1778 return 0; /* don't care */
1781 /* check to make sure that we are not extending the file */
1782 if (mapping
->host
->i_size
- offset
< (loff_t
)to
)
1783 to
= (unsigned)(mapping
->host
->i_size
- offset
);
1785 open_file
= find_writable_file(CIFS_I(mapping
->host
), false);
1787 bytes_written
= cifs_write(open_file
, open_file
->pid
,
1788 write_data
, to
- from
, &offset
);
1789 cifsFileInfo_put(open_file
);
1790 /* Does mm or vfs already set times? */
1791 inode
->i_atime
= inode
->i_mtime
= current_fs_time(inode
->i_sb
);
1792 if ((bytes_written
> 0) && (offset
))
1794 else if (bytes_written
< 0)
1797 cFYI(1, "No writeable filehandles for inode");
1805 static int cifs_writepages(struct address_space
*mapping
,
1806 struct writeback_control
*wbc
)
1808 struct cifs_sb_info
*cifs_sb
= CIFS_SB(mapping
->host
->i_sb
);
1809 bool done
= false, scanned
= false, range_whole
= false;
1811 struct cifs_writedata
*wdata
;
1812 struct TCP_Server_Info
*server
;
1817 * If wsize is smaller than the page cache size, default to writing
1818 * one page at a time via cifs_writepage
1820 if (cifs_sb
->wsize
< PAGE_CACHE_SIZE
)
1821 return generic_writepages(mapping
, wbc
);
1823 if (wbc
->range_cyclic
) {
1824 index
= mapping
->writeback_index
; /* Start from prev offset */
1827 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
1828 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
1829 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
1834 while (!done
&& index
<= end
) {
1835 unsigned int i
, nr_pages
, found_pages
;
1836 pgoff_t next
= 0, tofind
;
1837 struct page
**pages
;
1839 tofind
= min((cifs_sb
->wsize
/ PAGE_CACHE_SIZE
) - 1,
1842 wdata
= cifs_writedata_alloc((unsigned int)tofind
,
1843 cifs_writev_complete
);
1850 * find_get_pages_tag seems to return a max of 256 on each
1851 * iteration, so we must call it several times in order to
1852 * fill the array or the wsize is effectively limited to
1853 * 256 * PAGE_CACHE_SIZE.
1856 pages
= wdata
->pages
;
1858 nr_pages
= find_get_pages_tag(mapping
, &index
,
1859 PAGECACHE_TAG_DIRTY
,
1861 found_pages
+= nr_pages
;
1864 } while (nr_pages
&& tofind
&& index
<= end
);
1866 if (found_pages
== 0) {
1867 kref_put(&wdata
->refcount
, cifs_writedata_release
);
1872 for (i
= 0; i
< found_pages
; i
++) {
1873 page
= wdata
->pages
[i
];
1875 * At this point we hold neither mapping->tree_lock nor
1876 * lock on the page itself: the page may be truncated or
1877 * invalidated (changing page->mapping to NULL), or even
1878 * swizzled back from swapper_space to tmpfs file
1884 else if (!trylock_page(page
))
1887 if (unlikely(page
->mapping
!= mapping
)) {
1892 if (!wbc
->range_cyclic
&& page
->index
> end
) {
1898 if (next
&& (page
->index
!= next
)) {
1899 /* Not next consecutive page */
1904 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
1905 wait_on_page_writeback(page
);
1907 if (PageWriteback(page
) ||
1908 !clear_page_dirty_for_io(page
)) {
1914 * This actually clears the dirty bit in the radix tree.
1915 * See cifs_writepage() for more commentary.
1917 set_page_writeback(page
);
1919 if (page_offset(page
) >= i_size_read(mapping
->host
)) {
1922 end_page_writeback(page
);
1926 wdata
->pages
[i
] = page
;
1927 next
= page
->index
+ 1;
1931 /* reset index to refind any pages skipped */
1933 index
= wdata
->pages
[0]->index
+ 1;
1935 /* put any pages we aren't going to use */
1936 for (i
= nr_pages
; i
< found_pages
; i
++) {
1937 page_cache_release(wdata
->pages
[i
]);
1938 wdata
->pages
[i
] = NULL
;
1941 /* nothing to write? */
1942 if (nr_pages
== 0) {
1943 kref_put(&wdata
->refcount
, cifs_writedata_release
);
1947 wdata
->sync_mode
= wbc
->sync_mode
;
1948 wdata
->nr_pages
= nr_pages
;
1949 wdata
->offset
= page_offset(wdata
->pages
[0]);
1950 wdata
->pagesz
= PAGE_CACHE_SIZE
;
1952 min(i_size_read(mapping
->host
) -
1953 page_offset(wdata
->pages
[nr_pages
- 1]),
1954 (loff_t
)PAGE_CACHE_SIZE
);
1955 wdata
->bytes
= ((nr_pages
- 1) * PAGE_CACHE_SIZE
) +
1959 if (wdata
->cfile
!= NULL
)
1960 cifsFileInfo_put(wdata
->cfile
);
1961 wdata
->cfile
= find_writable_file(CIFS_I(mapping
->host
),
1963 if (!wdata
->cfile
) {
1964 cERROR(1, "No writable handles for inode");
1968 wdata
->pid
= wdata
->cfile
->pid
;
1969 server
= tlink_tcon(wdata
->cfile
->tlink
)->ses
->server
;
1970 rc
= server
->ops
->async_writev(wdata
);
1971 } while (wbc
->sync_mode
== WB_SYNC_ALL
&& rc
== -EAGAIN
);
1973 for (i
= 0; i
< nr_pages
; ++i
)
1974 unlock_page(wdata
->pages
[i
]);
1976 /* send failure -- clean up the mess */
1978 for (i
= 0; i
< nr_pages
; ++i
) {
1980 redirty_page_for_writepage(wbc
,
1983 SetPageError(wdata
->pages
[i
]);
1984 end_page_writeback(wdata
->pages
[i
]);
1985 page_cache_release(wdata
->pages
[i
]);
1988 mapping_set_error(mapping
, rc
);
1990 kref_put(&wdata
->refcount
, cifs_writedata_release
);
1992 wbc
->nr_to_write
-= nr_pages
;
1993 if (wbc
->nr_to_write
<= 0)
1999 if (!scanned
&& !done
) {
2001 * We hit the last page and there is more work to be done: wrap
2002 * back to the start of the file
2009 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2010 mapping
->writeback_index
= index
;
2016 cifs_writepage_locked(struct page
*page
, struct writeback_control
*wbc
)
2022 /* BB add check for wbc flags */
2023 page_cache_get(page
);
2024 if (!PageUptodate(page
))
2025 cFYI(1, "ppw - page not up to date");
2028 * Set the "writeback" flag, and clear "dirty" in the radix tree.
2030 * A writepage() implementation always needs to do either this,
2031 * or re-dirty the page with "redirty_page_for_writepage()" in
2032 * the case of a failure.
2034 * Just unlocking the page will cause the radix tree tag-bits
2035 * to fail to update with the state of the page correctly.
2037 set_page_writeback(page
);
2039 rc
= cifs_partialpagewrite(page
, 0, PAGE_CACHE_SIZE
);
2040 if (rc
== -EAGAIN
&& wbc
->sync_mode
== WB_SYNC_ALL
)
2042 else if (rc
== -EAGAIN
)
2043 redirty_page_for_writepage(wbc
, page
);
2047 SetPageUptodate(page
);
2048 end_page_writeback(page
);
2049 page_cache_release(page
);
2054 static int cifs_writepage(struct page
*page
, struct writeback_control
*wbc
)
2056 int rc
= cifs_writepage_locked(page
, wbc
);
2061 static int cifs_write_end(struct file
*file
, struct address_space
*mapping
,
2062 loff_t pos
, unsigned len
, unsigned copied
,
2063 struct page
*page
, void *fsdata
)
2066 struct inode
*inode
= mapping
->host
;
2067 struct cifsFileInfo
*cfile
= file
->private_data
;
2068 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cfile
->dentry
->d_sb
);
2071 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2074 pid
= current
->tgid
;
2076 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
2079 if (PageChecked(page
)) {
2081 SetPageUptodate(page
);
2082 ClearPageChecked(page
);
2083 } else if (!PageUptodate(page
) && copied
== PAGE_CACHE_SIZE
)
2084 SetPageUptodate(page
);
2086 if (!PageUptodate(page
)) {
2088 unsigned offset
= pos
& (PAGE_CACHE_SIZE
- 1);
2092 /* this is probably better than directly calling
2093 partialpage_write since in this function the file handle is
2094 known which we might as well leverage */
2095 /* BB check if anything else missing out of ppw
2096 such as updating last write time */
2097 page_data
= kmap(page
);
2098 rc
= cifs_write(cfile
, pid
, page_data
+ offset
, copied
, &pos
);
2099 /* if (rc < 0) should we set writebehind rc? */
2106 set_page_dirty(page
);
2110 spin_lock(&inode
->i_lock
);
2111 if (pos
> inode
->i_size
)
2112 i_size_write(inode
, pos
);
2113 spin_unlock(&inode
->i_lock
);
2117 page_cache_release(page
);
2122 int cifs_strict_fsync(struct file
*file
, loff_t start
, loff_t end
,
2127 struct cifs_tcon
*tcon
;
2128 struct TCP_Server_Info
*server
;
2129 struct cifsFileInfo
*smbfile
= file
->private_data
;
2130 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
2131 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
2133 rc
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
2136 mutex_lock(&inode
->i_mutex
);
2140 cFYI(1, "Sync file - name: %s datasync: 0x%x",
2141 file
->f_path
.dentry
->d_name
.name
, datasync
);
2143 if (!CIFS_I(inode
)->clientCanCacheRead
) {
2144 rc
= cifs_invalidate_mapping(inode
);
2146 cFYI(1, "rc: %d during invalidate phase", rc
);
2147 rc
= 0; /* don't care about it in fsync */
2151 tcon
= tlink_tcon(smbfile
->tlink
);
2152 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOSSYNC
)) {
2153 server
= tcon
->ses
->server
;
2154 if (server
->ops
->flush
)
2155 rc
= server
->ops
->flush(xid
, tcon
, &smbfile
->fid
);
2161 mutex_unlock(&inode
->i_mutex
);
2165 int cifs_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
2169 struct cifs_tcon
*tcon
;
2170 struct TCP_Server_Info
*server
;
2171 struct cifsFileInfo
*smbfile
= file
->private_data
;
2172 struct cifs_sb_info
*cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
2173 struct inode
*inode
= file
->f_mapping
->host
;
2175 rc
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
2178 mutex_lock(&inode
->i_mutex
);
2182 cFYI(1, "Sync file - name: %s datasync: 0x%x",
2183 file
->f_path
.dentry
->d_name
.name
, datasync
);
2185 tcon
= tlink_tcon(smbfile
->tlink
);
2186 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOSSYNC
)) {
2187 server
= tcon
->ses
->server
;
2188 if (server
->ops
->flush
)
2189 rc
= server
->ops
->flush(xid
, tcon
, &smbfile
->fid
);
2195 mutex_unlock(&inode
->i_mutex
);
2200 * As file closes, flush all cached write data for this inode checking
2201 * for write behind errors.
2203 int cifs_flush(struct file
*file
, fl_owner_t id
)
2205 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
2208 if (file
->f_mode
& FMODE_WRITE
)
2209 rc
= filemap_write_and_wait(inode
->i_mapping
);
2211 cFYI(1, "Flush inode %p file %p rc %d", inode
, file
, rc
);
2217 cifs_write_allocate_pages(struct page
**pages
, unsigned long num_pages
)
2222 for (i
= 0; i
< num_pages
; i
++) {
2223 pages
[i
] = alloc_page(GFP_KERNEL
|__GFP_HIGHMEM
);
2226 * save number of pages we have already allocated and
2227 * return with ENOMEM error
2236 for (i
= 0; i
< num_pages
; i
++)
2243 size_t get_numpages(const size_t wsize
, const size_t len
, size_t *cur_len
)
2248 clen
= min_t(const size_t, len
, wsize
);
2249 num_pages
= DIV_ROUND_UP(clen
, PAGE_SIZE
);
2258 cifs_uncached_writev_complete(struct work_struct
*work
)
2261 struct cifs_writedata
*wdata
= container_of(work
,
2262 struct cifs_writedata
, work
);
2263 struct inode
*inode
= wdata
->cfile
->dentry
->d_inode
;
2264 struct cifsInodeInfo
*cifsi
= CIFS_I(inode
);
2266 spin_lock(&inode
->i_lock
);
2267 cifs_update_eof(cifsi
, wdata
->offset
, wdata
->bytes
);
2268 if (cifsi
->server_eof
> inode
->i_size
)
2269 i_size_write(inode
, cifsi
->server_eof
);
2270 spin_unlock(&inode
->i_lock
);
2272 complete(&wdata
->done
);
2274 if (wdata
->result
!= -EAGAIN
) {
2275 for (i
= 0; i
< wdata
->nr_pages
; i
++)
2276 put_page(wdata
->pages
[i
]);
2279 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2282 /* attempt to send write to server, retry on any -EAGAIN errors */
2284 cifs_uncached_retry_writev(struct cifs_writedata
*wdata
)
2287 struct TCP_Server_Info
*server
;
2289 server
= tlink_tcon(wdata
->cfile
->tlink
)->ses
->server
;
2292 if (wdata
->cfile
->invalidHandle
) {
2293 rc
= cifs_reopen_file(wdata
->cfile
, false);
2297 rc
= server
->ops
->async_writev(wdata
);
2298 } while (rc
== -EAGAIN
);
2304 cifs_iovec_write(struct file
*file
, const struct iovec
*iov
,
2305 unsigned long nr_segs
, loff_t
*poffset
)
2307 unsigned long nr_pages
, i
;
2308 size_t copied
, len
, cur_len
;
2309 ssize_t total_written
= 0;
2312 struct cifsFileInfo
*open_file
;
2313 struct cifs_tcon
*tcon
;
2314 struct cifs_sb_info
*cifs_sb
;
2315 struct cifs_writedata
*wdata
, *tmp
;
2316 struct list_head wdata_list
;
2320 len
= iov_length(iov
, nr_segs
);
2324 rc
= generic_write_checks(file
, poffset
, &len
, 0);
2328 INIT_LIST_HEAD(&wdata_list
);
2329 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
2330 open_file
= file
->private_data
;
2331 tcon
= tlink_tcon(open_file
->tlink
);
2333 if (!tcon
->ses
->server
->ops
->async_writev
)
2338 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2339 pid
= open_file
->pid
;
2341 pid
= current
->tgid
;
2343 iov_iter_init(&it
, iov
, nr_segs
, len
, 0);
2347 nr_pages
= get_numpages(cifs_sb
->wsize
, len
, &cur_len
);
2348 wdata
= cifs_writedata_alloc(nr_pages
,
2349 cifs_uncached_writev_complete
);
2355 rc
= cifs_write_allocate_pages(wdata
->pages
, nr_pages
);
2362 for (i
= 0; i
< nr_pages
; i
++) {
2363 copied
= min_t(const size_t, cur_len
, PAGE_SIZE
);
2364 copied
= iov_iter_copy_from_user(wdata
->pages
[i
], &it
,
2367 iov_iter_advance(&it
, copied
);
2369 cur_len
= save_len
- cur_len
;
2371 wdata
->sync_mode
= WB_SYNC_ALL
;
2372 wdata
->nr_pages
= nr_pages
;
2373 wdata
->offset
= (__u64
)offset
;
2374 wdata
->cfile
= cifsFileInfo_get(open_file
);
2376 wdata
->bytes
= cur_len
;
2377 wdata
->pagesz
= PAGE_SIZE
;
2378 wdata
->tailsz
= cur_len
- ((nr_pages
- 1) * PAGE_SIZE
);
2379 rc
= cifs_uncached_retry_writev(wdata
);
2381 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2385 list_add_tail(&wdata
->list
, &wdata_list
);
2391 * If at least one write was successfully sent, then discard any rc
2392 * value from the later writes. If the other write succeeds, then
2393 * we'll end up returning whatever was written. If it fails, then
2394 * we'll get a new rc value from that.
2396 if (!list_empty(&wdata_list
))
2400 * Wait for and collect replies for any successful sends in order of
2401 * increasing offset. Once an error is hit or we get a fatal signal
2402 * while waiting, then return without waiting for any more replies.
2405 list_for_each_entry_safe(wdata
, tmp
, &wdata_list
, list
) {
2407 /* FIXME: freezable too? */
2408 rc
= wait_for_completion_killable(&wdata
->done
);
2411 else if (wdata
->result
)
2414 total_written
+= wdata
->bytes
;
2416 /* resend call if it's a retryable error */
2417 if (rc
== -EAGAIN
) {
2418 rc
= cifs_uncached_retry_writev(wdata
);
2422 list_del_init(&wdata
->list
);
2423 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2426 if (total_written
> 0)
2427 *poffset
+= total_written
;
2429 cifs_stats_bytes_written(tcon
, total_written
);
2430 return total_written
? total_written
: (ssize_t
)rc
;
2433 ssize_t
cifs_user_writev(struct kiocb
*iocb
, const struct iovec
*iov
,
2434 unsigned long nr_segs
, loff_t pos
)
2437 struct inode
*inode
;
2439 inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2442 * BB - optimize the way when signing is disabled. We can drop this
2443 * extra memory-to-memory copying and use iovec buffers for constructing
2447 written
= cifs_iovec_write(iocb
->ki_filp
, iov
, nr_segs
, &pos
);
2449 CIFS_I(inode
)->invalid_mapping
= true;
2457 cifs_writev(struct kiocb
*iocb
, const struct iovec
*iov
,
2458 unsigned long nr_segs
, loff_t pos
)
2460 struct file
*file
= iocb
->ki_filp
;
2461 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
2462 struct inode
*inode
= file
->f_mapping
->host
;
2463 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
2464 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
2465 ssize_t rc
= -EACCES
;
2467 BUG_ON(iocb
->ki_pos
!= pos
);
2469 sb_start_write(inode
->i_sb
);
2472 * We need to hold the sem to be sure nobody modifies lock list
2473 * with a brlock that prevents writing.
2475 down_read(&cinode
->lock_sem
);
2476 if (!cifs_find_lock_conflict(cfile
, pos
, iov_length(iov
, nr_segs
),
2477 server
->vals
->exclusive_lock_type
, NULL
,
2479 mutex_lock(&inode
->i_mutex
);
2480 rc
= __generic_file_aio_write(iocb
, iov
, nr_segs
,
2482 mutex_unlock(&inode
->i_mutex
);
2485 if (rc
> 0 || rc
== -EIOCBQUEUED
) {
2488 err
= generic_write_sync(file
, pos
, rc
);
2489 if (err
< 0 && rc
> 0)
2493 up_read(&cinode
->lock_sem
);
2494 sb_end_write(inode
->i_sb
);
2499 cifs_strict_writev(struct kiocb
*iocb
, const struct iovec
*iov
,
2500 unsigned long nr_segs
, loff_t pos
)
2502 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2503 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
2504 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
2505 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)
2506 iocb
->ki_filp
->private_data
;
2507 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
2509 #ifdef CONFIG_CIFS_SMB2
2511 * If we have an oplock for read and want to write a data to the file
2512 * we need to store it in the page cache and then push it to the server
2513 * to be sure the next read will get a valid data.
2515 if (!cinode
->clientCanCacheAll
&& cinode
->clientCanCacheRead
) {
2519 written
= generic_file_aio_write(iocb
, iov
, nr_segs
, pos
);
2520 rc
= filemap_fdatawrite(inode
->i_mapping
);
2529 * For non-oplocked files in strict cache mode we need to write the data
2530 * to the server exactly from the pos to pos+len-1 rather than flush all
2531 * affected pages because it may cause a error with mandatory locks on
2532 * these pages but not on the region from pos to ppos+len-1.
2535 if (!cinode
->clientCanCacheAll
)
2536 return cifs_user_writev(iocb
, iov
, nr_segs
, pos
);
2538 if (cap_unix(tcon
->ses
) &&
2539 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
2540 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
2541 return generic_file_aio_write(iocb
, iov
, nr_segs
, pos
);
2543 return cifs_writev(iocb
, iov
, nr_segs
, pos
);
2546 static struct cifs_readdata
*
2547 cifs_readdata_alloc(unsigned int nr_pages
, work_func_t complete
)
2549 struct cifs_readdata
*rdata
;
2551 rdata
= kzalloc(sizeof(*rdata
) + (sizeof(struct page
*) * nr_pages
),
2553 if (rdata
!= NULL
) {
2554 kref_init(&rdata
->refcount
);
2555 INIT_LIST_HEAD(&rdata
->list
);
2556 init_completion(&rdata
->done
);
2557 INIT_WORK(&rdata
->work
, complete
);
2564 cifs_readdata_release(struct kref
*refcount
)
2566 struct cifs_readdata
*rdata
= container_of(refcount
,
2567 struct cifs_readdata
, refcount
);
2570 cifsFileInfo_put(rdata
->cfile
);
2576 cifs_read_allocate_pages(struct cifs_readdata
*rdata
, unsigned int nr_pages
)
2582 for (i
= 0; i
< nr_pages
; i
++) {
2583 page
= alloc_page(GFP_KERNEL
|__GFP_HIGHMEM
);
2588 rdata
->pages
[i
] = page
;
2592 for (i
= 0; i
< nr_pages
; i
++) {
2593 put_page(rdata
->pages
[i
]);
2594 rdata
->pages
[i
] = NULL
;
2601 cifs_uncached_readdata_release(struct kref
*refcount
)
2603 struct cifs_readdata
*rdata
= container_of(refcount
,
2604 struct cifs_readdata
, refcount
);
2607 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
2608 put_page(rdata
->pages
[i
]);
2609 rdata
->pages
[i
] = NULL
;
2611 cifs_readdata_release(refcount
);
2615 cifs_retry_async_readv(struct cifs_readdata
*rdata
)
2618 struct TCP_Server_Info
*server
;
2620 server
= tlink_tcon(rdata
->cfile
->tlink
)->ses
->server
;
2623 if (rdata
->cfile
->invalidHandle
) {
2624 rc
= cifs_reopen_file(rdata
->cfile
, true);
2628 rc
= server
->ops
->async_readv(rdata
);
2629 } while (rc
== -EAGAIN
);
2635 * cifs_readdata_to_iov - copy data from pages in response to an iovec
2636 * @rdata: the readdata response with list of pages holding data
2637 * @iov: vector in which we should copy the data
2638 * @nr_segs: number of segments in vector
2639 * @offset: offset into file of the first iovec
2640 * @copied: used to return the amount of data copied to the iov
2642 * This function copies data from a list of pages in a readdata response into
2643 * an array of iovecs. It will first calculate where the data should go
2644 * based on the info in the readdata and then copy the data into that spot.
2647 cifs_readdata_to_iov(struct cifs_readdata
*rdata
, const struct iovec
*iov
,
2648 unsigned long nr_segs
, loff_t offset
, ssize_t
*copied
)
2652 size_t pos
= rdata
->offset
- offset
;
2653 ssize_t remaining
= rdata
->bytes
;
2654 unsigned char *pdata
;
2657 /* set up iov_iter and advance to the correct offset */
2658 iov_iter_init(&ii
, iov
, nr_segs
, iov_length(iov
, nr_segs
), 0);
2659 iov_iter_advance(&ii
, pos
);
2662 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
2664 struct page
*page
= rdata
->pages
[i
];
2666 /* copy a whole page or whatever's left */
2667 copy
= min_t(ssize_t
, remaining
, PAGE_SIZE
);
2669 /* ...but limit it to whatever space is left in the iov */
2670 copy
= min_t(ssize_t
, copy
, iov_iter_count(&ii
));
2672 /* go while there's data to be copied and no errors */
2675 rc
= memcpy_toiovecend(ii
.iov
, pdata
, ii
.iov_offset
,
2681 iov_iter_advance(&ii
, copy
);
2690 cifs_uncached_readv_complete(struct work_struct
*work
)
2692 struct cifs_readdata
*rdata
= container_of(work
,
2693 struct cifs_readdata
, work
);
2695 complete(&rdata
->done
);
2696 kref_put(&rdata
->refcount
, cifs_uncached_readdata_release
);
2700 cifs_uncached_read_into_pages(struct TCP_Server_Info
*server
,
2701 struct cifs_readdata
*rdata
, unsigned int len
)
2703 int total_read
= 0, result
= 0;
2705 unsigned int nr_pages
= rdata
->nr_pages
;
2708 rdata
->tailsz
= PAGE_SIZE
;
2709 for (i
= 0; i
< nr_pages
; i
++) {
2710 struct page
*page
= rdata
->pages
[i
];
2712 if (len
>= PAGE_SIZE
) {
2713 /* enough data to fill the page */
2714 iov
.iov_base
= kmap(page
);
2715 iov
.iov_len
= PAGE_SIZE
;
2716 cFYI(1, "%u: iov_base=%p iov_len=%zu",
2717 i
, iov
.iov_base
, iov
.iov_len
);
2719 } else if (len
> 0) {
2720 /* enough for partial page, fill and zero the rest */
2721 iov
.iov_base
= kmap(page
);
2723 cFYI(1, "%u: iov_base=%p iov_len=%zu",
2724 i
, iov
.iov_base
, iov
.iov_len
);
2725 memset(iov
.iov_base
+ len
, '\0', PAGE_SIZE
- len
);
2726 rdata
->tailsz
= len
;
2729 /* no need to hold page hostage */
2730 rdata
->pages
[i
] = NULL
;
2736 result
= cifs_readv_from_socket(server
, &iov
, 1, iov
.iov_len
);
2741 total_read
+= result
;
2744 return total_read
> 0 ? total_read
: result
;
2748 cifs_iovec_read(struct file
*file
, const struct iovec
*iov
,
2749 unsigned long nr_segs
, loff_t
*poffset
)
2752 size_t len
, cur_len
;
2753 ssize_t total_read
= 0;
2754 loff_t offset
= *poffset
;
2755 unsigned int npages
;
2756 struct cifs_sb_info
*cifs_sb
;
2757 struct cifs_tcon
*tcon
;
2758 struct cifsFileInfo
*open_file
;
2759 struct cifs_readdata
*rdata
, *tmp
;
2760 struct list_head rdata_list
;
2766 len
= iov_length(iov
, nr_segs
);
2770 INIT_LIST_HEAD(&rdata_list
);
2771 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
2772 open_file
= file
->private_data
;
2773 tcon
= tlink_tcon(open_file
->tlink
);
2775 if (!tcon
->ses
->server
->ops
->async_readv
)
2778 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2779 pid
= open_file
->pid
;
2781 pid
= current
->tgid
;
2783 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
2784 cFYI(1, "attempting read on write only file instance");
2787 cur_len
= min_t(const size_t, len
- total_read
, cifs_sb
->rsize
);
2788 npages
= DIV_ROUND_UP(cur_len
, PAGE_SIZE
);
2790 /* allocate a readdata struct */
2791 rdata
= cifs_readdata_alloc(npages
,
2792 cifs_uncached_readv_complete
);
2798 rc
= cifs_read_allocate_pages(rdata
, npages
);
2802 rdata
->cfile
= cifsFileInfo_get(open_file
);
2803 rdata
->nr_pages
= npages
;
2804 rdata
->offset
= offset
;
2805 rdata
->bytes
= cur_len
;
2807 rdata
->pagesz
= PAGE_SIZE
;
2808 rdata
->read_into_pages
= cifs_uncached_read_into_pages
;
2810 rc
= cifs_retry_async_readv(rdata
);
2813 kref_put(&rdata
->refcount
,
2814 cifs_uncached_readdata_release
);
2818 list_add_tail(&rdata
->list
, &rdata_list
);
2823 /* if at least one read request send succeeded, then reset rc */
2824 if (!list_empty(&rdata_list
))
2827 /* the loop below should proceed in the order of increasing offsets */
2829 list_for_each_entry_safe(rdata
, tmp
, &rdata_list
, list
) {
2833 /* FIXME: freezable sleep too? */
2834 rc
= wait_for_completion_killable(&rdata
->done
);
2837 else if (rdata
->result
)
2840 rc
= cifs_readdata_to_iov(rdata
, iov
,
2843 total_read
+= copied
;
2846 /* resend call if it's a retryable error */
2847 if (rc
== -EAGAIN
) {
2848 rc
= cifs_retry_async_readv(rdata
);
2852 list_del_init(&rdata
->list
);
2853 kref_put(&rdata
->refcount
, cifs_uncached_readdata_release
);
2856 cifs_stats_bytes_read(tcon
, total_read
);
2857 *poffset
+= total_read
;
2859 /* mask nodata case */
2863 return total_read
? total_read
: rc
;
2866 ssize_t
cifs_user_readv(struct kiocb
*iocb
, const struct iovec
*iov
,
2867 unsigned long nr_segs
, loff_t pos
)
2871 read
= cifs_iovec_read(iocb
->ki_filp
, iov
, nr_segs
, &pos
);
2879 cifs_strict_readv(struct kiocb
*iocb
, const struct iovec
*iov
,
2880 unsigned long nr_segs
, loff_t pos
)
2882 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2883 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
2884 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
2885 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)
2886 iocb
->ki_filp
->private_data
;
2887 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
2891 * In strict cache mode we need to read from the server all the time
2892 * if we don't have level II oplock because the server can delay mtime
2893 * change - so we can't make a decision about inode invalidating.
2894 * And we can also fail with pagereading if there are mandatory locks
2895 * on pages affected by this read but not on the region from pos to
2898 if (!cinode
->clientCanCacheRead
)
2899 return cifs_user_readv(iocb
, iov
, nr_segs
, pos
);
2901 if (cap_unix(tcon
->ses
) &&
2902 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
2903 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
2904 return generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
2907 * We need to hold the sem to be sure nobody modifies lock list
2908 * with a brlock that prevents reading.
2910 down_read(&cinode
->lock_sem
);
2911 if (!cifs_find_lock_conflict(cfile
, pos
, iov_length(iov
, nr_segs
),
2912 tcon
->ses
->server
->vals
->shared_lock_type
,
2913 NULL
, CIFS_READ_OP
))
2914 rc
= generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
2915 up_read(&cinode
->lock_sem
);
2920 cifs_read(struct file
*file
, char *read_data
, size_t read_size
, loff_t
*offset
)
2923 unsigned int bytes_read
= 0;
2924 unsigned int total_read
;
2925 unsigned int current_read_size
;
2927 struct cifs_sb_info
*cifs_sb
;
2928 struct cifs_tcon
*tcon
;
2929 struct TCP_Server_Info
*server
;
2932 struct cifsFileInfo
*open_file
;
2933 struct cifs_io_parms io_parms
;
2934 int buf_type
= CIFS_NO_BUFFER
;
2938 cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
2940 /* FIXME: set up handlers for larger reads and/or convert to async */
2941 rsize
= min_t(unsigned int, cifs_sb
->rsize
, CIFSMaxBufSize
);
2943 if (file
->private_data
== NULL
) {
2948 open_file
= file
->private_data
;
2949 tcon
= tlink_tcon(open_file
->tlink
);
2950 server
= tcon
->ses
->server
;
2952 if (!server
->ops
->sync_read
) {
2957 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2958 pid
= open_file
->pid
;
2960 pid
= current
->tgid
;
2962 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
2963 cFYI(1, "attempting read on write only file instance");
2965 for (total_read
= 0, cur_offset
= read_data
; read_size
> total_read
;
2966 total_read
+= bytes_read
, cur_offset
+= bytes_read
) {
2967 current_read_size
= min_t(uint
, read_size
- total_read
, rsize
);
2969 * For windows me and 9x we do not want to request more than it
2970 * negotiated since it will refuse the read then.
2972 if ((tcon
->ses
) && !(tcon
->ses
->capabilities
&
2973 tcon
->ses
->server
->vals
->cap_large_files
)) {
2974 current_read_size
= min_t(uint
, current_read_size
,
2978 while (rc
== -EAGAIN
) {
2979 if (open_file
->invalidHandle
) {
2980 rc
= cifs_reopen_file(open_file
, true);
2985 io_parms
.tcon
= tcon
;
2986 io_parms
.offset
= *offset
;
2987 io_parms
.length
= current_read_size
;
2988 rc
= server
->ops
->sync_read(xid
, open_file
, &io_parms
,
2989 &bytes_read
, &cur_offset
,
2992 if (rc
|| (bytes_read
== 0)) {
3000 cifs_stats_bytes_read(tcon
, total_read
);
3001 *offset
+= bytes_read
;
3009 * If the page is mmap'ed into a process' page tables, then we need to make
3010 * sure that it doesn't change while being written back.
3013 cifs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
3015 struct page
*page
= vmf
->page
;
3018 return VM_FAULT_LOCKED
;
3021 static struct vm_operations_struct cifs_file_vm_ops
= {
3022 .fault
= filemap_fault
,
3023 .page_mkwrite
= cifs_page_mkwrite
,
3024 .remap_pages
= generic_file_remap_pages
,
3027 int cifs_file_strict_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3030 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
3034 if (!CIFS_I(inode
)->clientCanCacheRead
) {
3035 rc
= cifs_invalidate_mapping(inode
);
3040 rc
= generic_file_mmap(file
, vma
);
3042 vma
->vm_ops
= &cifs_file_vm_ops
;
3047 int cifs_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3052 rc
= cifs_revalidate_file(file
);
3054 cFYI(1, "Validation prior to mmap failed, error=%d", rc
);
3058 rc
= generic_file_mmap(file
, vma
);
3060 vma
->vm_ops
= &cifs_file_vm_ops
;
3066 cifs_readv_complete(struct work_struct
*work
)
3069 struct cifs_readdata
*rdata
= container_of(work
,
3070 struct cifs_readdata
, work
);
3072 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
3073 struct page
*page
= rdata
->pages
[i
];
3075 lru_cache_add_file(page
);
3077 if (rdata
->result
== 0) {
3078 flush_dcache_page(page
);
3079 SetPageUptodate(page
);
3084 if (rdata
->result
== 0)
3085 cifs_readpage_to_fscache(rdata
->mapping
->host
, page
);
3087 page_cache_release(page
);
3088 rdata
->pages
[i
] = NULL
;
3090 kref_put(&rdata
->refcount
, cifs_readdata_release
);
3094 cifs_readpages_read_into_pages(struct TCP_Server_Info
*server
,
3095 struct cifs_readdata
*rdata
, unsigned int len
)
3097 int total_read
= 0, result
= 0;
3101 unsigned int nr_pages
= rdata
->nr_pages
;
3104 /* determine the eof that the server (probably) has */
3105 eof
= CIFS_I(rdata
->mapping
->host
)->server_eof
;
3106 eof_index
= eof
? (eof
- 1) >> PAGE_CACHE_SHIFT
: 0;
3107 cFYI(1, "eof=%llu eof_index=%lu", eof
, eof_index
);
3109 rdata
->tailsz
= PAGE_CACHE_SIZE
;
3110 for (i
= 0; i
< nr_pages
; i
++) {
3111 struct page
*page
= rdata
->pages
[i
];
3113 if (len
>= PAGE_CACHE_SIZE
) {
3114 /* enough data to fill the page */
3115 iov
.iov_base
= kmap(page
);
3116 iov
.iov_len
= PAGE_CACHE_SIZE
;
3117 cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
3118 i
, page
->index
, iov
.iov_base
, iov
.iov_len
);
3119 len
-= PAGE_CACHE_SIZE
;
3120 } else if (len
> 0) {
3121 /* enough for partial page, fill and zero the rest */
3122 iov
.iov_base
= kmap(page
);
3124 cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
3125 i
, page
->index
, iov
.iov_base
, iov
.iov_len
);
3126 memset(iov
.iov_base
+ len
,
3127 '\0', PAGE_CACHE_SIZE
- len
);
3128 rdata
->tailsz
= len
;
3130 } else if (page
->index
> eof_index
) {
3132 * The VFS will not try to do readahead past the
3133 * i_size, but it's possible that we have outstanding
3134 * writes with gaps in the middle and the i_size hasn't
3135 * caught up yet. Populate those with zeroed out pages
3136 * to prevent the VFS from repeatedly attempting to
3137 * fill them until the writes are flushed.
3139 zero_user(page
, 0, PAGE_CACHE_SIZE
);
3140 lru_cache_add_file(page
);
3141 flush_dcache_page(page
);
3142 SetPageUptodate(page
);
3144 page_cache_release(page
);
3145 rdata
->pages
[i
] = NULL
;
3149 /* no need to hold page hostage */
3150 lru_cache_add_file(page
);
3152 page_cache_release(page
);
3153 rdata
->pages
[i
] = NULL
;
3158 result
= cifs_readv_from_socket(server
, &iov
, 1, iov
.iov_len
);
3163 total_read
+= result
;
3166 return total_read
> 0 ? total_read
: result
;
3169 static int cifs_readpages(struct file
*file
, struct address_space
*mapping
,
3170 struct list_head
*page_list
, unsigned num_pages
)
3173 struct list_head tmplist
;
3174 struct cifsFileInfo
*open_file
= file
->private_data
;
3175 struct cifs_sb_info
*cifs_sb
= CIFS_SB(file
->f_path
.dentry
->d_sb
);
3176 unsigned int rsize
= cifs_sb
->rsize
;
3180 * Give up immediately if rsize is too small to read an entire page.
3181 * The VFS will fall back to readpage. We should never reach this
3182 * point however since we set ra_pages to 0 when the rsize is smaller
3183 * than a cache page.
3185 if (unlikely(rsize
< PAGE_CACHE_SIZE
))
3189 * Reads as many pages as possible from fscache. Returns -ENOBUFS
3190 * immediately if the cookie is negative
3192 rc
= cifs_readpages_from_fscache(mapping
->host
, mapping
, page_list
,
3197 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
3198 pid
= open_file
->pid
;
3200 pid
= current
->tgid
;
3203 INIT_LIST_HEAD(&tmplist
);
3205 cFYI(1, "%s: file=%p mapping=%p num_pages=%u", __func__
, file
,
3206 mapping
, num_pages
);
3209 * Start with the page at end of list and move it to private
3210 * list. Do the same with any following pages until we hit
3211 * the rsize limit, hit an index discontinuity, or run out of
3212 * pages. Issue the async read and then start the loop again
3213 * until the list is empty.
3215 * Note that list order is important. The page_list is in
3216 * the order of declining indexes. When we put the pages in
3217 * the rdata->pages, then we want them in increasing order.
3219 while (!list_empty(page_list
)) {
3221 unsigned int bytes
= PAGE_CACHE_SIZE
;
3222 unsigned int expected_index
;
3223 unsigned int nr_pages
= 1;
3225 struct page
*page
, *tpage
;
3226 struct cifs_readdata
*rdata
;
3228 page
= list_entry(page_list
->prev
, struct page
, lru
);
3231 * Lock the page and put it in the cache. Since no one else
3232 * should have access to this page, we're safe to simply set
3233 * PG_locked without checking it first.
3235 __set_page_locked(page
);
3236 rc
= add_to_page_cache_locked(page
, mapping
,
3237 page
->index
, GFP_KERNEL
);
3239 /* give up if we can't stick it in the cache */
3241 __clear_page_locked(page
);
3245 /* move first page to the tmplist */
3246 offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
3247 list_move_tail(&page
->lru
, &tmplist
);
3249 /* now try and add more pages onto the request */
3250 expected_index
= page
->index
+ 1;
3251 list_for_each_entry_safe_reverse(page
, tpage
, page_list
, lru
) {
3252 /* discontinuity ? */
3253 if (page
->index
!= expected_index
)
3256 /* would this page push the read over the rsize? */
3257 if (bytes
+ PAGE_CACHE_SIZE
> rsize
)
3260 __set_page_locked(page
);
3261 if (add_to_page_cache_locked(page
, mapping
,
3262 page
->index
, GFP_KERNEL
)) {
3263 __clear_page_locked(page
);
3266 list_move_tail(&page
->lru
, &tmplist
);
3267 bytes
+= PAGE_CACHE_SIZE
;
3272 rdata
= cifs_readdata_alloc(nr_pages
, cifs_readv_complete
);
3274 /* best to give up if we're out of mem */
3275 list_for_each_entry_safe(page
, tpage
, &tmplist
, lru
) {
3276 list_del(&page
->lru
);
3277 lru_cache_add_file(page
);
3279 page_cache_release(page
);
3285 rdata
->cfile
= cifsFileInfo_get(open_file
);
3286 rdata
->mapping
= mapping
;
3287 rdata
->offset
= offset
;
3288 rdata
->bytes
= bytes
;
3290 rdata
->pagesz
= PAGE_CACHE_SIZE
;
3291 rdata
->read_into_pages
= cifs_readpages_read_into_pages
;
3293 list_for_each_entry_safe(page
, tpage
, &tmplist
, lru
) {
3294 list_del(&page
->lru
);
3295 rdata
->pages
[rdata
->nr_pages
++] = page
;
3298 rc
= cifs_retry_async_readv(rdata
);
3300 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
3301 page
= rdata
->pages
[i
];
3302 lru_cache_add_file(page
);
3304 page_cache_release(page
);
3306 kref_put(&rdata
->refcount
, cifs_readdata_release
);
3310 kref_put(&rdata
->refcount
, cifs_readdata_release
);
3316 static int cifs_readpage_worker(struct file
*file
, struct page
*page
,
3322 /* Is the page cached? */
3323 rc
= cifs_readpage_from_fscache(file
->f_path
.dentry
->d_inode
, page
);
3327 page_cache_get(page
);
3328 read_data
= kmap(page
);
3329 /* for reads over a certain size could initiate async read ahead */
3331 rc
= cifs_read(file
, read_data
, PAGE_CACHE_SIZE
, poffset
);
3336 cFYI(1, "Bytes read %d", rc
);
3338 file
->f_path
.dentry
->d_inode
->i_atime
=
3339 current_fs_time(file
->f_path
.dentry
->d_inode
->i_sb
);
3341 if (PAGE_CACHE_SIZE
> rc
)
3342 memset(read_data
+ rc
, 0, PAGE_CACHE_SIZE
- rc
);
3344 flush_dcache_page(page
);
3345 SetPageUptodate(page
);
3347 /* send this page to the cache */
3348 cifs_readpage_to_fscache(file
->f_path
.dentry
->d_inode
, page
);
3354 page_cache_release(page
);
3360 static int cifs_readpage(struct file
*file
, struct page
*page
)
3362 loff_t offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
3368 if (file
->private_data
== NULL
) {
3374 cFYI(1, "readpage %p at offset %d 0x%x",
3375 page
, (int)offset
, (int)offset
);
3377 rc
= cifs_readpage_worker(file
, page
, &offset
);
3385 static int is_inode_writable(struct cifsInodeInfo
*cifs_inode
)
3387 struct cifsFileInfo
*open_file
;
3389 spin_lock(&cifs_file_list_lock
);
3390 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
3391 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
3392 spin_unlock(&cifs_file_list_lock
);
3396 spin_unlock(&cifs_file_list_lock
);
3400 /* We do not want to update the file size from server for inodes
3401 open for write - to avoid races with writepage extending
3402 the file - in the future we could consider allowing
3403 refreshing the inode only on increases in the file size
3404 but this is tricky to do without racing with writebehind
3405 page caching in the current Linux kernel design */
3406 bool is_size_safe_to_change(struct cifsInodeInfo
*cifsInode
, __u64 end_of_file
)
3411 if (is_inode_writable(cifsInode
)) {
3412 /* This inode is open for write at least once */
3413 struct cifs_sb_info
*cifs_sb
;
3415 cifs_sb
= CIFS_SB(cifsInode
->vfs_inode
.i_sb
);
3416 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_DIRECT_IO
) {
3417 /* since no page cache to corrupt on directio
3418 we can change size safely */
3422 if (i_size_read(&cifsInode
->vfs_inode
) < end_of_file
)
3430 static int cifs_write_begin(struct file
*file
, struct address_space
*mapping
,
3431 loff_t pos
, unsigned len
, unsigned flags
,
3432 struct page
**pagep
, void **fsdata
)
3434 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
3435 loff_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
3436 loff_t page_start
= pos
& PAGE_MASK
;
3441 cFYI(1, "write_begin from %lld len %d", (long long)pos
, len
);
3443 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3449 if (PageUptodate(page
))
3453 * If we write a full page it will be up to date, no need to read from
3454 * the server. If the write is short, we'll end up doing a sync write
3457 if (len
== PAGE_CACHE_SIZE
)
3461 * optimize away the read when we have an oplock, and we're not
3462 * expecting to use any of the data we'd be reading in. That
3463 * is, when the page lies beyond the EOF, or straddles the EOF
3464 * and the write will cover all of the existing data.
3466 if (CIFS_I(mapping
->host
)->clientCanCacheRead
) {
3467 i_size
= i_size_read(mapping
->host
);
3468 if (page_start
>= i_size
||
3469 (offset
== 0 && (pos
+ len
) >= i_size
)) {
3470 zero_user_segments(page
, 0, offset
,
3474 * PageChecked means that the parts of the page
3475 * to which we're not writing are considered up
3476 * to date. Once the data is copied to the
3477 * page, it can be set uptodate.
3479 SetPageChecked(page
);
3484 if ((file
->f_flags
& O_ACCMODE
) != O_WRONLY
) {
3486 * might as well read a page, it is fast enough. If we get
3487 * an error, we don't need to return it. cifs_write_end will
3488 * do a sync write instead since PG_uptodate isn't set.
3490 cifs_readpage_worker(file
, page
, &page_start
);
3492 /* we could try using another file handle if there is one -
3493 but how would we lock it to prevent close of that handle
3494 racing with this read? In any case
3495 this will be written out by write_end so is fine */
3502 static int cifs_release_page(struct page
*page
, gfp_t gfp
)
3504 if (PagePrivate(page
))
3507 return cifs_fscache_release_page(page
, gfp
);
3510 static void cifs_invalidate_page(struct page
*page
, unsigned long offset
)
3512 struct cifsInodeInfo
*cifsi
= CIFS_I(page
->mapping
->host
);
3515 cifs_fscache_invalidate_page(page
, &cifsi
->vfs_inode
);
3518 static int cifs_launder_page(struct page
*page
)
3521 loff_t range_start
= page_offset(page
);
3522 loff_t range_end
= range_start
+ (loff_t
)(PAGE_CACHE_SIZE
- 1);
3523 struct writeback_control wbc
= {
3524 .sync_mode
= WB_SYNC_ALL
,
3526 .range_start
= range_start
,
3527 .range_end
= range_end
,
3530 cFYI(1, "Launder page: %p", page
);
3532 if (clear_page_dirty_for_io(page
))
3533 rc
= cifs_writepage_locked(page
, &wbc
);
3535 cifs_fscache_invalidate_page(page
, page
->mapping
->host
);
3539 void cifs_oplock_break(struct work_struct
*work
)
3541 struct cifsFileInfo
*cfile
= container_of(work
, struct cifsFileInfo
,
3543 struct inode
*inode
= cfile
->dentry
->d_inode
;
3544 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
3545 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
3548 if (inode
&& S_ISREG(inode
->i_mode
)) {
3549 if (cinode
->clientCanCacheRead
)
3550 break_lease(inode
, O_RDONLY
);
3552 break_lease(inode
, O_WRONLY
);
3553 rc
= filemap_fdatawrite(inode
->i_mapping
);
3554 if (cinode
->clientCanCacheRead
== 0) {
3555 rc
= filemap_fdatawait(inode
->i_mapping
);
3556 mapping_set_error(inode
->i_mapping
, rc
);
3557 invalidate_remote_inode(inode
);
3559 cFYI(1, "Oplock flush inode %p rc %d", inode
, rc
);
3562 rc
= cifs_push_locks(cfile
);
3564 cERROR(1, "Push locks rc = %d", rc
);
3567 * releasing stale oplock after recent reconnect of smb session using
3568 * a now incorrect file handle is not a data integrity issue but do
3569 * not bother sending an oplock release if session to server still is
3570 * disconnected since oplock already released by the server
3572 if (!cfile
->oplock_break_cancelled
) {
3573 rc
= tcon
->ses
->server
->ops
->oplock_response(tcon
, &cfile
->fid
,
3575 cFYI(1, "Oplock release rc = %d", rc
);
3579 const struct address_space_operations cifs_addr_ops
= {
3580 .readpage
= cifs_readpage
,
3581 .readpages
= cifs_readpages
,
3582 .writepage
= cifs_writepage
,
3583 .writepages
= cifs_writepages
,
3584 .write_begin
= cifs_write_begin
,
3585 .write_end
= cifs_write_end
,
3586 .set_page_dirty
= __set_page_dirty_nobuffers
,
3587 .releasepage
= cifs_release_page
,
3588 .invalidatepage
= cifs_invalidate_page
,
3589 .launder_page
= cifs_launder_page
,
3593 * cifs_readpages requires the server to support a buffer large enough to
3594 * contain the header plus one complete page of data. Otherwise, we need
3595 * to leave cifs_readpages out of the address space operations.
3597 const struct address_space_operations cifs_addr_ops_smallbuf
= {
3598 .readpage
= cifs_readpage
,
3599 .writepage
= cifs_writepage
,
3600 .writepages
= cifs_writepages
,
3601 .write_begin
= cifs_write_begin
,
3602 .write_end
= cifs_write_end
,
3603 .set_page_dirty
= __set_page_dirty_nobuffers
,
3604 .releasepage
= cifs_release_page
,
3605 .invalidatepage
= cifs_invalidate_page
,
3606 .launder_page
= cifs_launder_page
,