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"
47 static inline int cifs_convert_flags(unsigned int flags
)
49 if ((flags
& O_ACCMODE
) == O_RDONLY
)
51 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
53 else if ((flags
& O_ACCMODE
) == O_RDWR
) {
54 /* GENERIC_ALL is too much permission to request
55 can cause unnecessary access denied on create */
56 /* return GENERIC_ALL; */
57 return (GENERIC_READ
| GENERIC_WRITE
);
60 return (READ_CONTROL
| FILE_WRITE_ATTRIBUTES
| FILE_READ_ATTRIBUTES
|
61 FILE_WRITE_EA
| FILE_APPEND_DATA
| FILE_WRITE_DATA
|
65 static u32
cifs_posix_convert_flags(unsigned int flags
)
69 if ((flags
& O_ACCMODE
) == O_RDONLY
)
70 posix_flags
= SMB_O_RDONLY
;
71 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
72 posix_flags
= SMB_O_WRONLY
;
73 else if ((flags
& O_ACCMODE
) == O_RDWR
)
74 posix_flags
= SMB_O_RDWR
;
76 if (flags
& O_CREAT
) {
77 posix_flags
|= SMB_O_CREAT
;
79 posix_flags
|= SMB_O_EXCL
;
80 } else if (flags
& O_EXCL
)
81 cifs_dbg(FYI
, "Application %s pid %d has incorrectly set O_EXCL flag but not O_CREAT on file open. Ignoring O_EXCL\n",
82 current
->comm
, current
->tgid
);
85 posix_flags
|= SMB_O_TRUNC
;
86 /* be safe and imply O_SYNC for O_DSYNC */
88 posix_flags
|= SMB_O_SYNC
;
89 if (flags
& O_DIRECTORY
)
90 posix_flags
|= SMB_O_DIRECTORY
;
91 if (flags
& O_NOFOLLOW
)
92 posix_flags
|= SMB_O_NOFOLLOW
;
94 posix_flags
|= SMB_O_DIRECT
;
99 static inline int cifs_get_disposition(unsigned int flags
)
101 if ((flags
& (O_CREAT
| O_EXCL
)) == (O_CREAT
| O_EXCL
))
103 else if ((flags
& (O_CREAT
| O_TRUNC
)) == (O_CREAT
| O_TRUNC
))
104 return FILE_OVERWRITE_IF
;
105 else if ((flags
& O_CREAT
) == O_CREAT
)
107 else if ((flags
& O_TRUNC
) == O_TRUNC
)
108 return FILE_OVERWRITE
;
113 int cifs_posix_open(char *full_path
, struct inode
**pinode
,
114 struct super_block
*sb
, int mode
, unsigned int f_flags
,
115 __u32
*poplock
, __u16
*pnetfid
, unsigned int xid
)
118 FILE_UNIX_BASIC_INFO
*presp_data
;
119 __u32 posix_flags
= 0;
120 struct cifs_sb_info
*cifs_sb
= CIFS_SB(sb
);
121 struct cifs_fattr fattr
;
122 struct tcon_link
*tlink
;
123 struct cifs_tcon
*tcon
;
125 cifs_dbg(FYI
, "posix open %s\n", full_path
);
127 presp_data
= kzalloc(sizeof(FILE_UNIX_BASIC_INFO
), GFP_KERNEL
);
128 if (presp_data
== NULL
)
131 tlink
= cifs_sb_tlink(cifs_sb
);
137 tcon
= tlink_tcon(tlink
);
138 mode
&= ~current_umask();
140 posix_flags
= cifs_posix_convert_flags(f_flags
);
141 rc
= CIFSPOSIXCreate(xid
, tcon
, posix_flags
, mode
, pnetfid
, presp_data
,
142 poplock
, full_path
, cifs_sb
->local_nls
,
143 cifs_remap(cifs_sb
));
144 cifs_put_tlink(tlink
);
149 if (presp_data
->Type
== cpu_to_le32(-1))
150 goto posix_open_ret
; /* open ok, caller does qpathinfo */
153 goto posix_open_ret
; /* caller does not need info */
155 cifs_unix_basic_to_fattr(&fattr
, presp_data
, cifs_sb
);
157 /* get new inode and set it up */
158 if (*pinode
== NULL
) {
159 cifs_fill_uniqueid(sb
, &fattr
);
160 *pinode
= cifs_iget(sb
, &fattr
);
166 cifs_fattr_to_inode(*pinode
, &fattr
);
175 cifs_nt_open(char *full_path
, struct inode
*inode
, struct cifs_sb_info
*cifs_sb
,
176 struct cifs_tcon
*tcon
, unsigned int f_flags
, __u32
*oplock
,
177 struct cifs_fid
*fid
, unsigned int xid
)
182 int create_options
= CREATE_NOT_DIR
;
184 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
185 struct cifs_open_parms oparms
;
187 if (!server
->ops
->open
)
190 desired_access
= cifs_convert_flags(f_flags
);
192 /*********************************************************************
193 * open flag mapping table:
195 * POSIX Flag CIFS Disposition
196 * ---------- ----------------
197 * O_CREAT FILE_OPEN_IF
198 * O_CREAT | O_EXCL FILE_CREATE
199 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
200 * O_TRUNC FILE_OVERWRITE
201 * none of the above FILE_OPEN
203 * Note that there is not a direct match between disposition
204 * FILE_SUPERSEDE (ie create whether or not file exists although
205 * O_CREAT | O_TRUNC is similar but truncates the existing
206 * file rather than creating a new file as FILE_SUPERSEDE does
207 * (which uses the attributes / metadata passed in on open call)
209 *? O_SYNC is a reasonable match to CIFS writethrough flag
210 *? and the read write flags match reasonably. O_LARGEFILE
211 *? is irrelevant because largefile support is always used
212 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
213 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
214 *********************************************************************/
216 disposition
= cifs_get_disposition(f_flags
);
218 /* BB pass O_SYNC flag through on file attributes .. BB */
220 buf
= kmalloc(sizeof(FILE_ALL_INFO
), GFP_KERNEL
);
224 if (backup_cred(cifs_sb
))
225 create_options
|= CREATE_OPEN_BACKUP_INTENT
;
228 oparms
.cifs_sb
= cifs_sb
;
229 oparms
.desired_access
= desired_access
;
230 oparms
.create_options
= create_options
;
231 oparms
.disposition
= disposition
;
232 oparms
.path
= full_path
;
234 oparms
.reconnect
= false;
236 rc
= server
->ops
->open(xid
, &oparms
, oplock
, buf
);
242 rc
= cifs_get_inode_info_unix(&inode
, full_path
, inode
->i_sb
,
245 rc
= cifs_get_inode_info(&inode
, full_path
, buf
, inode
->i_sb
,
254 cifs_has_mand_locks(struct cifsInodeInfo
*cinode
)
256 struct cifs_fid_locks
*cur
;
257 bool has_locks
= false;
259 down_read(&cinode
->lock_sem
);
260 list_for_each_entry(cur
, &cinode
->llist
, llist
) {
261 if (!list_empty(&cur
->locks
)) {
266 up_read(&cinode
->lock_sem
);
270 struct cifsFileInfo
*
271 cifs_new_fileinfo(struct cifs_fid
*fid
, struct file
*file
,
272 struct tcon_link
*tlink
, __u32 oplock
)
274 struct dentry
*dentry
= file_dentry(file
);
275 struct inode
*inode
= d_inode(dentry
);
276 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
277 struct cifsFileInfo
*cfile
;
278 struct cifs_fid_locks
*fdlocks
;
279 struct cifs_tcon
*tcon
= tlink_tcon(tlink
);
280 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
282 cfile
= kzalloc(sizeof(struct cifsFileInfo
), GFP_KERNEL
);
286 fdlocks
= kzalloc(sizeof(struct cifs_fid_locks
), GFP_KERNEL
);
292 INIT_LIST_HEAD(&fdlocks
->locks
);
293 fdlocks
->cfile
= cfile
;
294 cfile
->llist
= fdlocks
;
295 down_write(&cinode
->lock_sem
);
296 list_add(&fdlocks
->llist
, &cinode
->llist
);
297 up_write(&cinode
->lock_sem
);
300 cfile
->pid
= current
->tgid
;
301 cfile
->uid
= current_fsuid();
302 cfile
->dentry
= dget(dentry
);
303 cfile
->f_flags
= file
->f_flags
;
304 cfile
->invalidHandle
= false;
305 cfile
->tlink
= cifs_get_tlink(tlink
);
306 INIT_WORK(&cfile
->oplock_break
, cifs_oplock_break
);
307 mutex_init(&cfile
->fh_mutex
);
309 cifs_sb_active(inode
->i_sb
);
312 * If the server returned a read oplock and we have mandatory brlocks,
313 * set oplock level to None.
315 if (server
->ops
->is_read_op(oplock
) && cifs_has_mand_locks(cinode
)) {
316 cifs_dbg(FYI
, "Reset oplock val from read to None due to mand locks\n");
320 spin_lock(&cifs_file_list_lock
);
321 if (fid
->pending_open
->oplock
!= CIFS_OPLOCK_NO_CHANGE
&& oplock
)
322 oplock
= fid
->pending_open
->oplock
;
323 list_del(&fid
->pending_open
->olist
);
325 fid
->purge_cache
= false;
326 server
->ops
->set_fid(cfile
, fid
, oplock
);
328 list_add(&cfile
->tlist
, &tcon
->openFileList
);
329 /* if readable file instance put first in list*/
330 if (file
->f_mode
& FMODE_READ
)
331 list_add(&cfile
->flist
, &cinode
->openFileList
);
333 list_add_tail(&cfile
->flist
, &cinode
->openFileList
);
334 spin_unlock(&cifs_file_list_lock
);
336 if (fid
->purge_cache
)
337 cifs_zap_mapping(inode
);
339 file
->private_data
= cfile
;
343 struct cifsFileInfo
*
344 cifsFileInfo_get(struct cifsFileInfo
*cifs_file
)
346 spin_lock(&cifs_file_list_lock
);
347 cifsFileInfo_get_locked(cifs_file
);
348 spin_unlock(&cifs_file_list_lock
);
353 * Release a reference on the file private data. This may involve closing
354 * the filehandle out on the server. Must be called without holding
355 * cifs_file_list_lock.
357 void cifsFileInfo_put(struct cifsFileInfo
*cifs_file
)
359 struct inode
*inode
= d_inode(cifs_file
->dentry
);
360 struct cifs_tcon
*tcon
= tlink_tcon(cifs_file
->tlink
);
361 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
362 struct cifsInodeInfo
*cifsi
= CIFS_I(inode
);
363 struct super_block
*sb
= inode
->i_sb
;
364 struct cifs_sb_info
*cifs_sb
= CIFS_SB(sb
);
365 struct cifsLockInfo
*li
, *tmp
;
367 struct cifs_pending_open open
;
368 bool oplock_break_cancelled
;
370 spin_lock(&cifs_file_list_lock
);
371 if (--cifs_file
->count
> 0) {
372 spin_unlock(&cifs_file_list_lock
);
376 if (server
->ops
->get_lease_key
)
377 server
->ops
->get_lease_key(inode
, &fid
);
379 /* store open in pending opens to make sure we don't miss lease break */
380 cifs_add_pending_open_locked(&fid
, cifs_file
->tlink
, &open
);
382 /* remove it from the lists */
383 list_del(&cifs_file
->flist
);
384 list_del(&cifs_file
->tlist
);
386 if (list_empty(&cifsi
->openFileList
)) {
387 cifs_dbg(FYI
, "closing last open instance for inode %p\n",
388 d_inode(cifs_file
->dentry
));
390 * In strict cache mode we need invalidate mapping on the last
391 * close because it may cause a error when we open this file
392 * again and get at least level II oplock.
394 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_STRICT_IO
)
395 set_bit(CIFS_INO_INVALID_MAPPING
, &cifsi
->flags
);
396 cifs_set_oplock_level(cifsi
, 0);
398 spin_unlock(&cifs_file_list_lock
);
400 oplock_break_cancelled
= cancel_work_sync(&cifs_file
->oplock_break
);
402 if (!tcon
->need_reconnect
&& !cifs_file
->invalidHandle
) {
403 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
407 if (server
->ops
->close
)
408 server
->ops
->close(xid
, tcon
, &cifs_file
->fid
);
412 if (oplock_break_cancelled
)
413 cifs_done_oplock_break(cifsi
);
415 cifs_del_pending_open(&open
);
418 * Delete any outstanding lock records. We'll lose them when the file
421 down_write(&cifsi
->lock_sem
);
422 list_for_each_entry_safe(li
, tmp
, &cifs_file
->llist
->locks
, llist
) {
423 list_del(&li
->llist
);
424 cifs_del_lock_waiters(li
);
427 list_del(&cifs_file
->llist
->llist
);
428 kfree(cifs_file
->llist
);
429 up_write(&cifsi
->lock_sem
);
431 cifs_put_tlink(cifs_file
->tlink
);
432 dput(cifs_file
->dentry
);
433 cifs_sb_deactive(sb
);
437 int cifs_open(struct inode
*inode
, struct file
*file
)
443 struct cifs_sb_info
*cifs_sb
;
444 struct TCP_Server_Info
*server
;
445 struct cifs_tcon
*tcon
;
446 struct tcon_link
*tlink
;
447 struct cifsFileInfo
*cfile
= NULL
;
448 char *full_path
= NULL
;
449 bool posix_open_ok
= false;
451 struct cifs_pending_open open
;
455 cifs_sb
= CIFS_SB(inode
->i_sb
);
456 tlink
= cifs_sb_tlink(cifs_sb
);
459 return PTR_ERR(tlink
);
461 tcon
= tlink_tcon(tlink
);
462 server
= tcon
->ses
->server
;
464 full_path
= build_path_from_dentry(file_dentry(file
));
465 if (full_path
== NULL
) {
470 cifs_dbg(FYI
, "inode = 0x%p file flags are 0x%x for %s\n",
471 inode
, file
->f_flags
, full_path
);
473 if (file
->f_flags
& O_DIRECT
&&
474 cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_STRICT_IO
) {
475 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NO_BRL
)
476 file
->f_op
= &cifs_file_direct_nobrl_ops
;
478 file
->f_op
= &cifs_file_direct_ops
;
486 if (!tcon
->broken_posix_open
&& tcon
->unix_ext
&&
487 cap_unix(tcon
->ses
) && (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
488 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
489 /* can not refresh inode info since size could be stale */
490 rc
= cifs_posix_open(full_path
, &inode
, inode
->i_sb
,
491 cifs_sb
->mnt_file_mode
/* ignored */,
492 file
->f_flags
, &oplock
, &fid
.netfid
, xid
);
494 cifs_dbg(FYI
, "posix open succeeded\n");
495 posix_open_ok
= true;
496 } else if ((rc
== -EINVAL
) || (rc
== -EOPNOTSUPP
)) {
497 if (tcon
->ses
->serverNOS
)
498 cifs_dbg(VFS
, "server %s of type %s returned unexpected error on SMB posix open, disabling posix open support. Check if server update available.\n",
499 tcon
->ses
->serverName
,
500 tcon
->ses
->serverNOS
);
501 tcon
->broken_posix_open
= true;
502 } else if ((rc
!= -EIO
) && (rc
!= -EREMOTE
) &&
503 (rc
!= -EOPNOTSUPP
)) /* path not found or net err */
506 * Else fallthrough to retry open the old way on network i/o
511 if (server
->ops
->get_lease_key
)
512 server
->ops
->get_lease_key(inode
, &fid
);
514 cifs_add_pending_open(&fid
, tlink
, &open
);
516 if (!posix_open_ok
) {
517 if (server
->ops
->get_lease_key
)
518 server
->ops
->get_lease_key(inode
, &fid
);
520 rc
= cifs_nt_open(full_path
, inode
, cifs_sb
, tcon
,
521 file
->f_flags
, &oplock
, &fid
, xid
);
523 cifs_del_pending_open(&open
);
528 cfile
= cifs_new_fileinfo(&fid
, file
, tlink
, oplock
);
530 if (server
->ops
->close
)
531 server
->ops
->close(xid
, tcon
, &fid
);
532 cifs_del_pending_open(&open
);
537 cifs_fscache_set_inode_cookie(inode
, file
);
539 if ((oplock
& CIFS_CREATE_ACTION
) && !posix_open_ok
&& tcon
->unix_ext
) {
541 * Time to set mode which we can not set earlier due to
542 * problems creating new read-only files.
544 struct cifs_unix_set_info_args args
= {
545 .mode
= inode
->i_mode
,
546 .uid
= INVALID_UID
, /* no change */
547 .gid
= INVALID_GID
, /* no change */
548 .ctime
= NO_CHANGE_64
,
549 .atime
= NO_CHANGE_64
,
550 .mtime
= NO_CHANGE_64
,
553 CIFSSMBUnixSetFileInfo(xid
, tcon
, &args
, fid
.netfid
,
560 cifs_put_tlink(tlink
);
564 static int cifs_push_posix_locks(struct cifsFileInfo
*cfile
);
567 * Try to reacquire byte range locks that were released when session
568 * to server was lost.
571 cifs_relock_file(struct cifsFileInfo
*cfile
)
573 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cfile
->dentry
->d_sb
);
574 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
575 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
578 down_read(&cinode
->lock_sem
);
579 if (cinode
->can_cache_brlcks
) {
580 /* can cache locks - no need to relock */
581 up_read(&cinode
->lock_sem
);
585 if (cap_unix(tcon
->ses
) &&
586 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
587 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
588 rc
= cifs_push_posix_locks(cfile
);
590 rc
= tcon
->ses
->server
->ops
->push_mand_locks(cfile
);
592 up_read(&cinode
->lock_sem
);
597 cifs_reopen_file(struct cifsFileInfo
*cfile
, bool can_flush
)
602 struct cifs_sb_info
*cifs_sb
;
603 struct cifs_tcon
*tcon
;
604 struct TCP_Server_Info
*server
;
605 struct cifsInodeInfo
*cinode
;
607 char *full_path
= NULL
;
609 int disposition
= FILE_OPEN
;
610 int create_options
= CREATE_NOT_DIR
;
611 struct cifs_open_parms oparms
;
614 mutex_lock(&cfile
->fh_mutex
);
615 if (!cfile
->invalidHandle
) {
616 mutex_unlock(&cfile
->fh_mutex
);
622 inode
= d_inode(cfile
->dentry
);
623 cifs_sb
= CIFS_SB(inode
->i_sb
);
624 tcon
= tlink_tcon(cfile
->tlink
);
625 server
= tcon
->ses
->server
;
628 * Can not grab rename sem here because various ops, including those
629 * that already have the rename sem can end up causing writepage to get
630 * called and if the server was down that means we end up here, and we
631 * can never tell if the caller already has the rename_sem.
633 full_path
= build_path_from_dentry(cfile
->dentry
);
634 if (full_path
== NULL
) {
636 mutex_unlock(&cfile
->fh_mutex
);
641 cifs_dbg(FYI
, "inode = 0x%p file flags 0x%x for %s\n",
642 inode
, cfile
->f_flags
, full_path
);
644 if (tcon
->ses
->server
->oplocks
)
649 if (tcon
->unix_ext
&& cap_unix(tcon
->ses
) &&
650 (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
651 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
653 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
654 * original open. Must mask them off for a reopen.
656 unsigned int oflags
= cfile
->f_flags
&
657 ~(O_CREAT
| O_EXCL
| O_TRUNC
);
659 rc
= cifs_posix_open(full_path
, NULL
, inode
->i_sb
,
660 cifs_sb
->mnt_file_mode
/* ignored */,
661 oflags
, &oplock
, &cfile
->fid
.netfid
, xid
);
663 cifs_dbg(FYI
, "posix reopen succeeded\n");
664 oparms
.reconnect
= true;
668 * fallthrough to retry open the old way on errors, especially
669 * in the reconnect path it is important to retry hard
673 desired_access
= cifs_convert_flags(cfile
->f_flags
);
675 if (backup_cred(cifs_sb
))
676 create_options
|= CREATE_OPEN_BACKUP_INTENT
;
678 if (server
->ops
->get_lease_key
)
679 server
->ops
->get_lease_key(inode
, &cfile
->fid
);
682 oparms
.cifs_sb
= cifs_sb
;
683 oparms
.desired_access
= desired_access
;
684 oparms
.create_options
= create_options
;
685 oparms
.disposition
= disposition
;
686 oparms
.path
= full_path
;
687 oparms
.fid
= &cfile
->fid
;
688 oparms
.reconnect
= true;
691 * Can not refresh inode by passing in file_info buf to be returned by
692 * ops->open and then calling get_inode_info with returned buf since
693 * file might have write behind data that needs to be flushed and server
694 * version of file size can be stale. If we knew for sure that inode was
695 * not dirty locally we could do this.
697 rc
= server
->ops
->open(xid
, &oparms
, &oplock
, NULL
);
698 if (rc
== -ENOENT
&& oparms
.reconnect
== false) {
699 /* durable handle timeout is expired - open the file again */
700 rc
= server
->ops
->open(xid
, &oparms
, &oplock
, NULL
);
701 /* indicate that we need to relock the file */
702 oparms
.reconnect
= true;
706 mutex_unlock(&cfile
->fh_mutex
);
707 cifs_dbg(FYI
, "cifs_reopen returned 0x%x\n", rc
);
708 cifs_dbg(FYI
, "oplock: %d\n", oplock
);
709 goto reopen_error_exit
;
713 cfile
->invalidHandle
= false;
714 mutex_unlock(&cfile
->fh_mutex
);
715 cinode
= CIFS_I(inode
);
718 rc
= filemap_write_and_wait(inode
->i_mapping
);
719 mapping_set_error(inode
->i_mapping
, rc
);
722 rc
= cifs_get_inode_info_unix(&inode
, full_path
,
725 rc
= cifs_get_inode_info(&inode
, full_path
, NULL
,
726 inode
->i_sb
, xid
, NULL
);
729 * Else we are writing out data to server already and could deadlock if
730 * we tried to flush data, and since we do not know if we have data that
731 * would invalidate the current end of file on the server we can not go
732 * to the server to get the new inode info.
735 server
->ops
->set_fid(cfile
, &cfile
->fid
, oplock
);
736 if (oparms
.reconnect
)
737 cifs_relock_file(cfile
);
745 int cifs_close(struct inode
*inode
, struct file
*file
)
747 if (file
->private_data
!= NULL
) {
748 cifsFileInfo_put(file
->private_data
);
749 file
->private_data
= NULL
;
752 /* return code from the ->release op is always ignored */
756 int cifs_closedir(struct inode
*inode
, struct file
*file
)
760 struct cifsFileInfo
*cfile
= file
->private_data
;
761 struct cifs_tcon
*tcon
;
762 struct TCP_Server_Info
*server
;
765 cifs_dbg(FYI
, "Closedir inode = 0x%p\n", inode
);
771 tcon
= tlink_tcon(cfile
->tlink
);
772 server
= tcon
->ses
->server
;
774 cifs_dbg(FYI
, "Freeing private data in close dir\n");
775 spin_lock(&cifs_file_list_lock
);
776 if (server
->ops
->dir_needs_close(cfile
)) {
777 cfile
->invalidHandle
= true;
778 spin_unlock(&cifs_file_list_lock
);
779 if (server
->ops
->close_dir
)
780 rc
= server
->ops
->close_dir(xid
, tcon
, &cfile
->fid
);
783 cifs_dbg(FYI
, "Closing uncompleted readdir with rc %d\n", rc
);
784 /* not much we can do if it fails anyway, ignore rc */
787 spin_unlock(&cifs_file_list_lock
);
789 buf
= cfile
->srch_inf
.ntwrk_buf_start
;
791 cifs_dbg(FYI
, "closedir free smb buf in srch struct\n");
792 cfile
->srch_inf
.ntwrk_buf_start
= NULL
;
793 if (cfile
->srch_inf
.smallBuf
)
794 cifs_small_buf_release(buf
);
796 cifs_buf_release(buf
);
799 cifs_put_tlink(cfile
->tlink
);
800 kfree(file
->private_data
);
801 file
->private_data
= NULL
;
802 /* BB can we lock the filestruct while this is going on? */
807 static struct cifsLockInfo
*
808 cifs_lock_init(__u64 offset
, __u64 length
, __u8 type
)
810 struct cifsLockInfo
*lock
=
811 kmalloc(sizeof(struct cifsLockInfo
), GFP_KERNEL
);
814 lock
->offset
= offset
;
815 lock
->length
= length
;
817 lock
->pid
= current
->tgid
;
818 INIT_LIST_HEAD(&lock
->blist
);
819 init_waitqueue_head(&lock
->block_q
);
824 cifs_del_lock_waiters(struct cifsLockInfo
*lock
)
826 struct cifsLockInfo
*li
, *tmp
;
827 list_for_each_entry_safe(li
, tmp
, &lock
->blist
, blist
) {
828 list_del_init(&li
->blist
);
829 wake_up(&li
->block_q
);
833 #define CIFS_LOCK_OP 0
834 #define CIFS_READ_OP 1
835 #define CIFS_WRITE_OP 2
837 /* @rw_check : 0 - no op, 1 - read, 2 - write */
839 cifs_find_fid_lock_conflict(struct cifs_fid_locks
*fdlocks
, __u64 offset
,
840 __u64 length
, __u8 type
, struct cifsFileInfo
*cfile
,
841 struct cifsLockInfo
**conf_lock
, int rw_check
)
843 struct cifsLockInfo
*li
;
844 struct cifsFileInfo
*cur_cfile
= fdlocks
->cfile
;
845 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
847 list_for_each_entry(li
, &fdlocks
->locks
, llist
) {
848 if (offset
+ length
<= li
->offset
||
849 offset
>= li
->offset
+ li
->length
)
851 if (rw_check
!= CIFS_LOCK_OP
&& current
->tgid
== li
->pid
&&
852 server
->ops
->compare_fids(cfile
, cur_cfile
)) {
853 /* shared lock prevents write op through the same fid */
854 if (!(li
->type
& server
->vals
->shared_lock_type
) ||
855 rw_check
!= CIFS_WRITE_OP
)
858 if ((type
& server
->vals
->shared_lock_type
) &&
859 ((server
->ops
->compare_fids(cfile
, cur_cfile
) &&
860 current
->tgid
== li
->pid
) || type
== li
->type
))
870 cifs_find_lock_conflict(struct cifsFileInfo
*cfile
, __u64 offset
, __u64 length
,
871 __u8 type
, struct cifsLockInfo
**conf_lock
,
875 struct cifs_fid_locks
*cur
;
876 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
878 list_for_each_entry(cur
, &cinode
->llist
, llist
) {
879 rc
= cifs_find_fid_lock_conflict(cur
, offset
, length
, type
,
880 cfile
, conf_lock
, rw_check
);
889 * Check if there is another lock that prevents us to set the lock (mandatory
890 * style). If such a lock exists, update the flock structure with its
891 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
892 * or leave it the same if we can't. Returns 0 if we don't need to request to
893 * the server or 1 otherwise.
896 cifs_lock_test(struct cifsFileInfo
*cfile
, __u64 offset
, __u64 length
,
897 __u8 type
, struct file_lock
*flock
)
900 struct cifsLockInfo
*conf_lock
;
901 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
902 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
905 down_read(&cinode
->lock_sem
);
907 exist
= cifs_find_lock_conflict(cfile
, offset
, length
, type
,
908 &conf_lock
, CIFS_LOCK_OP
);
910 flock
->fl_start
= conf_lock
->offset
;
911 flock
->fl_end
= conf_lock
->offset
+ conf_lock
->length
- 1;
912 flock
->fl_pid
= conf_lock
->pid
;
913 if (conf_lock
->type
& server
->vals
->shared_lock_type
)
914 flock
->fl_type
= F_RDLCK
;
916 flock
->fl_type
= F_WRLCK
;
917 } else if (!cinode
->can_cache_brlcks
)
920 flock
->fl_type
= F_UNLCK
;
922 up_read(&cinode
->lock_sem
);
927 cifs_lock_add(struct cifsFileInfo
*cfile
, struct cifsLockInfo
*lock
)
929 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
930 down_write(&cinode
->lock_sem
);
931 list_add_tail(&lock
->llist
, &cfile
->llist
->locks
);
932 up_write(&cinode
->lock_sem
);
936 * Set the byte-range lock (mandatory style). Returns:
937 * 1) 0, if we set the lock and don't need to request to the server;
938 * 2) 1, if no locks prevent us but we need to request to the server;
939 * 3) -EACCESS, if there is a lock that prevents us and wait is false.
942 cifs_lock_add_if(struct cifsFileInfo
*cfile
, struct cifsLockInfo
*lock
,
945 struct cifsLockInfo
*conf_lock
;
946 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
952 down_write(&cinode
->lock_sem
);
954 exist
= cifs_find_lock_conflict(cfile
, lock
->offset
, lock
->length
,
955 lock
->type
, &conf_lock
, CIFS_LOCK_OP
);
956 if (!exist
&& cinode
->can_cache_brlcks
) {
957 list_add_tail(&lock
->llist
, &cfile
->llist
->locks
);
958 up_write(&cinode
->lock_sem
);
967 list_add_tail(&lock
->blist
, &conf_lock
->blist
);
968 up_write(&cinode
->lock_sem
);
969 rc
= wait_event_interruptible(lock
->block_q
,
970 (lock
->blist
.prev
== &lock
->blist
) &&
971 (lock
->blist
.next
== &lock
->blist
));
974 down_write(&cinode
->lock_sem
);
975 list_del_init(&lock
->blist
);
978 up_write(&cinode
->lock_sem
);
983 * Check if there is another lock that prevents us to set the lock (posix
984 * style). If such a lock exists, update the flock structure with its
985 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
986 * or leave it the same if we can't. Returns 0 if we don't need to request to
987 * the server or 1 otherwise.
990 cifs_posix_lock_test(struct file
*file
, struct file_lock
*flock
)
993 struct cifsInodeInfo
*cinode
= CIFS_I(file_inode(file
));
994 unsigned char saved_type
= flock
->fl_type
;
996 if ((flock
->fl_flags
& FL_POSIX
) == 0)
999 down_read(&cinode
->lock_sem
);
1000 posix_test_lock(file
, flock
);
1002 if (flock
->fl_type
== F_UNLCK
&& !cinode
->can_cache_brlcks
) {
1003 flock
->fl_type
= saved_type
;
1007 up_read(&cinode
->lock_sem
);
1012 * Set the byte-range lock (posix style). Returns:
1013 * 1) 0, if we set the lock and don't need to request to the server;
1014 * 2) 1, if we need to request to the server;
1015 * 3) <0, if the error occurs while setting the lock.
1018 cifs_posix_lock_set(struct file
*file
, struct file_lock
*flock
)
1020 struct cifsInodeInfo
*cinode
= CIFS_I(file_inode(file
));
1023 if ((flock
->fl_flags
& FL_POSIX
) == 0)
1027 down_write(&cinode
->lock_sem
);
1028 if (!cinode
->can_cache_brlcks
) {
1029 up_write(&cinode
->lock_sem
);
1033 rc
= posix_lock_file(file
, flock
, NULL
);
1034 up_write(&cinode
->lock_sem
);
1035 if (rc
== FILE_LOCK_DEFERRED
) {
1036 rc
= wait_event_interruptible(flock
->fl_wait
, !flock
->fl_next
);
1039 posix_unblock_lock(flock
);
1045 cifs_push_mandatory_locks(struct cifsFileInfo
*cfile
)
1048 int rc
= 0, stored_rc
;
1049 struct cifsLockInfo
*li
, *tmp
;
1050 struct cifs_tcon
*tcon
;
1051 unsigned int num
, max_num
, max_buf
;
1052 LOCKING_ANDX_RANGE
*buf
, *cur
;
1053 int types
[] = {LOCKING_ANDX_LARGE_FILES
,
1054 LOCKING_ANDX_SHARED_LOCK
| LOCKING_ANDX_LARGE_FILES
};
1058 tcon
= tlink_tcon(cfile
->tlink
);
1061 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1062 * and check it for zero before using.
1064 max_buf
= tcon
->ses
->server
->maxBuf
;
1070 max_num
= (max_buf
- sizeof(struct smb_hdr
)) /
1071 sizeof(LOCKING_ANDX_RANGE
);
1072 buf
= kcalloc(max_num
, sizeof(LOCKING_ANDX_RANGE
), GFP_KERNEL
);
1078 for (i
= 0; i
< 2; i
++) {
1081 list_for_each_entry_safe(li
, tmp
, &cfile
->llist
->locks
, llist
) {
1082 if (li
->type
!= types
[i
])
1084 cur
->Pid
= cpu_to_le16(li
->pid
);
1085 cur
->LengthLow
= cpu_to_le32((u32
)li
->length
);
1086 cur
->LengthHigh
= cpu_to_le32((u32
)(li
->length
>>32));
1087 cur
->OffsetLow
= cpu_to_le32((u32
)li
->offset
);
1088 cur
->OffsetHigh
= cpu_to_le32((u32
)(li
->offset
>>32));
1089 if (++num
== max_num
) {
1090 stored_rc
= cifs_lockv(xid
, tcon
,
1092 (__u8
)li
->type
, 0, num
,
1103 stored_rc
= cifs_lockv(xid
, tcon
, cfile
->fid
.netfid
,
1104 (__u8
)types
[i
], 0, num
, buf
);
1116 hash_lockowner(fl_owner_t owner
)
1118 return cifs_lock_secret
^ hash32_ptr((const void *)owner
);
1121 struct lock_to_push
{
1122 struct list_head llist
;
1131 cifs_push_posix_locks(struct cifsFileInfo
*cfile
)
1133 struct inode
*inode
= d_inode(cfile
->dentry
);
1134 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1135 struct file_lock
*flock
;
1136 struct file_lock_context
*flctx
= inode
->i_flctx
;
1137 unsigned int count
= 0, i
;
1138 int rc
= 0, xid
, type
;
1139 struct list_head locks_to_send
, *el
;
1140 struct lock_to_push
*lck
, *tmp
;
1148 spin_lock(&flctx
->flc_lock
);
1149 list_for_each(el
, &flctx
->flc_posix
) {
1152 spin_unlock(&flctx
->flc_lock
);
1154 INIT_LIST_HEAD(&locks_to_send
);
1157 * Allocating count locks is enough because no FL_POSIX locks can be
1158 * added to the list while we are holding cinode->lock_sem that
1159 * protects locking operations of this inode.
1161 for (i
= 0; i
< count
; i
++) {
1162 lck
= kmalloc(sizeof(struct lock_to_push
), GFP_KERNEL
);
1167 list_add_tail(&lck
->llist
, &locks_to_send
);
1170 el
= locks_to_send
.next
;
1171 spin_lock(&flctx
->flc_lock
);
1172 list_for_each_entry(flock
, &flctx
->flc_posix
, fl_list
) {
1173 if (el
== &locks_to_send
) {
1175 * The list ended. We don't have enough allocated
1176 * structures - something is really wrong.
1178 cifs_dbg(VFS
, "Can't push all brlocks!\n");
1181 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1182 if (flock
->fl_type
== F_RDLCK
|| flock
->fl_type
== F_SHLCK
)
1186 lck
= list_entry(el
, struct lock_to_push
, llist
);
1187 lck
->pid
= hash_lockowner(flock
->fl_owner
);
1188 lck
->netfid
= cfile
->fid
.netfid
;
1189 lck
->length
= length
;
1191 lck
->offset
= flock
->fl_start
;
1193 spin_unlock(&flctx
->flc_lock
);
1195 list_for_each_entry_safe(lck
, tmp
, &locks_to_send
, llist
) {
1198 stored_rc
= CIFSSMBPosixLock(xid
, tcon
, lck
->netfid
, lck
->pid
,
1199 lck
->offset
, lck
->length
, NULL
,
1203 list_del(&lck
->llist
);
1211 list_for_each_entry_safe(lck
, tmp
, &locks_to_send
, llist
) {
1212 list_del(&lck
->llist
);
1219 cifs_push_locks(struct cifsFileInfo
*cfile
)
1221 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cfile
->dentry
->d_sb
);
1222 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
1223 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1226 /* we are going to update can_cache_brlcks here - need a write access */
1227 down_write(&cinode
->lock_sem
);
1228 if (!cinode
->can_cache_brlcks
) {
1229 up_write(&cinode
->lock_sem
);
1233 if (cap_unix(tcon
->ses
) &&
1234 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
1235 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
1236 rc
= cifs_push_posix_locks(cfile
);
1238 rc
= tcon
->ses
->server
->ops
->push_mand_locks(cfile
);
1240 cinode
->can_cache_brlcks
= false;
1241 up_write(&cinode
->lock_sem
);
1246 cifs_read_flock(struct file_lock
*flock
, __u32
*type
, int *lock
, int *unlock
,
1247 bool *wait_flag
, struct TCP_Server_Info
*server
)
1249 if (flock
->fl_flags
& FL_POSIX
)
1250 cifs_dbg(FYI
, "Posix\n");
1251 if (flock
->fl_flags
& FL_FLOCK
)
1252 cifs_dbg(FYI
, "Flock\n");
1253 if (flock
->fl_flags
& FL_SLEEP
) {
1254 cifs_dbg(FYI
, "Blocking lock\n");
1257 if (flock
->fl_flags
& FL_ACCESS
)
1258 cifs_dbg(FYI
, "Process suspended by mandatory locking - not implemented yet\n");
1259 if (flock
->fl_flags
& FL_LEASE
)
1260 cifs_dbg(FYI
, "Lease on file - not implemented yet\n");
1261 if (flock
->fl_flags
&
1262 (~(FL_POSIX
| FL_FLOCK
| FL_SLEEP
|
1263 FL_ACCESS
| FL_LEASE
| FL_CLOSE
)))
1264 cifs_dbg(FYI
, "Unknown lock flags 0x%x\n", flock
->fl_flags
);
1266 *type
= server
->vals
->large_lock_type
;
1267 if (flock
->fl_type
== F_WRLCK
) {
1268 cifs_dbg(FYI
, "F_WRLCK\n");
1269 *type
|= server
->vals
->exclusive_lock_type
;
1271 } else if (flock
->fl_type
== F_UNLCK
) {
1272 cifs_dbg(FYI
, "F_UNLCK\n");
1273 *type
|= server
->vals
->unlock_lock_type
;
1275 /* Check if unlock includes more than one lock range */
1276 } else if (flock
->fl_type
== F_RDLCK
) {
1277 cifs_dbg(FYI
, "F_RDLCK\n");
1278 *type
|= server
->vals
->shared_lock_type
;
1280 } else if (flock
->fl_type
== F_EXLCK
) {
1281 cifs_dbg(FYI
, "F_EXLCK\n");
1282 *type
|= server
->vals
->exclusive_lock_type
;
1284 } else if (flock
->fl_type
== F_SHLCK
) {
1285 cifs_dbg(FYI
, "F_SHLCK\n");
1286 *type
|= server
->vals
->shared_lock_type
;
1289 cifs_dbg(FYI
, "Unknown type of lock\n");
1293 cifs_getlk(struct file
*file
, struct file_lock
*flock
, __u32 type
,
1294 bool wait_flag
, bool posix_lck
, unsigned int xid
)
1297 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1298 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
1299 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1300 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
1301 __u16 netfid
= cfile
->fid
.netfid
;
1304 int posix_lock_type
;
1306 rc
= cifs_posix_lock_test(file
, flock
);
1310 if (type
& server
->vals
->shared_lock_type
)
1311 posix_lock_type
= CIFS_RDLCK
;
1313 posix_lock_type
= CIFS_WRLCK
;
1314 rc
= CIFSSMBPosixLock(xid
, tcon
, netfid
,
1315 hash_lockowner(flock
->fl_owner
),
1316 flock
->fl_start
, length
, flock
,
1317 posix_lock_type
, wait_flag
);
1321 rc
= cifs_lock_test(cfile
, flock
->fl_start
, length
, type
, flock
);
1325 /* BB we could chain these into one lock request BB */
1326 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
, type
,
1329 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1331 flock
->fl_type
= F_UNLCK
;
1333 cifs_dbg(VFS
, "Error unlocking previously locked range %d during test of lock\n",
1338 if (type
& server
->vals
->shared_lock_type
) {
1339 flock
->fl_type
= F_WRLCK
;
1343 type
&= ~server
->vals
->exclusive_lock_type
;
1345 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1346 type
| server
->vals
->shared_lock_type
,
1349 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1350 type
| server
->vals
->shared_lock_type
, 0, 1, false);
1351 flock
->fl_type
= F_RDLCK
;
1353 cifs_dbg(VFS
, "Error unlocking previously locked range %d during test of lock\n",
1356 flock
->fl_type
= F_WRLCK
;
1362 cifs_move_llist(struct list_head
*source
, struct list_head
*dest
)
1364 struct list_head
*li
, *tmp
;
1365 list_for_each_safe(li
, tmp
, source
)
1366 list_move(li
, dest
);
1370 cifs_free_llist(struct list_head
*llist
)
1372 struct cifsLockInfo
*li
, *tmp
;
1373 list_for_each_entry_safe(li
, tmp
, llist
, llist
) {
1374 cifs_del_lock_waiters(li
);
1375 list_del(&li
->llist
);
1381 cifs_unlock_range(struct cifsFileInfo
*cfile
, struct file_lock
*flock
,
1384 int rc
= 0, stored_rc
;
1385 int types
[] = {LOCKING_ANDX_LARGE_FILES
,
1386 LOCKING_ANDX_SHARED_LOCK
| LOCKING_ANDX_LARGE_FILES
};
1388 unsigned int max_num
, num
, max_buf
;
1389 LOCKING_ANDX_RANGE
*buf
, *cur
;
1390 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1391 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
1392 struct cifsLockInfo
*li
, *tmp
;
1393 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1394 struct list_head tmp_llist
;
1396 INIT_LIST_HEAD(&tmp_llist
);
1399 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1400 * and check it for zero before using.
1402 max_buf
= tcon
->ses
->server
->maxBuf
;
1406 max_num
= (max_buf
- sizeof(struct smb_hdr
)) /
1407 sizeof(LOCKING_ANDX_RANGE
);
1408 buf
= kcalloc(max_num
, sizeof(LOCKING_ANDX_RANGE
), GFP_KERNEL
);
1412 down_write(&cinode
->lock_sem
);
1413 for (i
= 0; i
< 2; i
++) {
1416 list_for_each_entry_safe(li
, tmp
, &cfile
->llist
->locks
, llist
) {
1417 if (flock
->fl_start
> li
->offset
||
1418 (flock
->fl_start
+ length
) <
1419 (li
->offset
+ li
->length
))
1421 if (current
->tgid
!= li
->pid
)
1423 if (types
[i
] != li
->type
)
1425 if (cinode
->can_cache_brlcks
) {
1427 * We can cache brlock requests - simply remove
1428 * a lock from the file's list.
1430 list_del(&li
->llist
);
1431 cifs_del_lock_waiters(li
);
1435 cur
->Pid
= cpu_to_le16(li
->pid
);
1436 cur
->LengthLow
= cpu_to_le32((u32
)li
->length
);
1437 cur
->LengthHigh
= cpu_to_le32((u32
)(li
->length
>>32));
1438 cur
->OffsetLow
= cpu_to_le32((u32
)li
->offset
);
1439 cur
->OffsetHigh
= cpu_to_le32((u32
)(li
->offset
>>32));
1441 * We need to save a lock here to let us add it again to
1442 * the file's list if the unlock range request fails on
1445 list_move(&li
->llist
, &tmp_llist
);
1446 if (++num
== max_num
) {
1447 stored_rc
= cifs_lockv(xid
, tcon
,
1449 li
->type
, num
, 0, buf
);
1452 * We failed on the unlock range
1453 * request - add all locks from the tmp
1454 * list to the head of the file's list.
1456 cifs_move_llist(&tmp_llist
,
1457 &cfile
->llist
->locks
);
1461 * The unlock range request succeed -
1462 * free the tmp list.
1464 cifs_free_llist(&tmp_llist
);
1471 stored_rc
= cifs_lockv(xid
, tcon
, cfile
->fid
.netfid
,
1472 types
[i
], num
, 0, buf
);
1474 cifs_move_llist(&tmp_llist
,
1475 &cfile
->llist
->locks
);
1478 cifs_free_llist(&tmp_llist
);
1482 up_write(&cinode
->lock_sem
);
1488 cifs_setlk(struct file
*file
, struct file_lock
*flock
, __u32 type
,
1489 bool wait_flag
, bool posix_lck
, int lock
, int unlock
,
1493 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1494 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
1495 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1496 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
1497 struct inode
*inode
= d_inode(cfile
->dentry
);
1500 int posix_lock_type
;
1502 rc
= cifs_posix_lock_set(file
, flock
);
1506 if (type
& server
->vals
->shared_lock_type
)
1507 posix_lock_type
= CIFS_RDLCK
;
1509 posix_lock_type
= CIFS_WRLCK
;
1512 posix_lock_type
= CIFS_UNLCK
;
1514 rc
= CIFSSMBPosixLock(xid
, tcon
, cfile
->fid
.netfid
,
1515 hash_lockowner(flock
->fl_owner
),
1516 flock
->fl_start
, length
,
1517 NULL
, posix_lock_type
, wait_flag
);
1522 struct cifsLockInfo
*lock
;
1524 lock
= cifs_lock_init(flock
->fl_start
, length
, type
);
1528 rc
= cifs_lock_add_if(cfile
, lock
, wait_flag
);
1537 * Windows 7 server can delay breaking lease from read to None
1538 * if we set a byte-range lock on a file - break it explicitly
1539 * before sending the lock to the server to be sure the next
1540 * read won't conflict with non-overlapted locks due to
1543 if (!CIFS_CACHE_WRITE(CIFS_I(inode
)) &&
1544 CIFS_CACHE_READ(CIFS_I(inode
))) {
1545 cifs_zap_mapping(inode
);
1546 cifs_dbg(FYI
, "Set no oplock for inode=%p due to mand locks\n",
1548 CIFS_I(inode
)->oplock
= 0;
1551 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1552 type
, 1, 0, wait_flag
);
1558 cifs_lock_add(cfile
, lock
);
1560 rc
= server
->ops
->mand_unlock_range(cfile
, flock
, xid
);
1563 if (flock
->fl_flags
& FL_POSIX
&& !rc
)
1564 rc
= locks_lock_file_wait(file
, flock
);
1568 int cifs_lock(struct file
*file
, int cmd
, struct file_lock
*flock
)
1571 int lock
= 0, unlock
= 0;
1572 bool wait_flag
= false;
1573 bool posix_lck
= false;
1574 struct cifs_sb_info
*cifs_sb
;
1575 struct cifs_tcon
*tcon
;
1576 struct cifsInodeInfo
*cinode
;
1577 struct cifsFileInfo
*cfile
;
1584 cifs_dbg(FYI
, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld end: %lld\n",
1585 cmd
, flock
->fl_flags
, flock
->fl_type
,
1586 flock
->fl_start
, flock
->fl_end
);
1588 cfile
= (struct cifsFileInfo
*)file
->private_data
;
1589 tcon
= tlink_tcon(cfile
->tlink
);
1591 cifs_read_flock(flock
, &type
, &lock
, &unlock
, &wait_flag
,
1594 cifs_sb
= CIFS_FILE_SB(file
);
1595 netfid
= cfile
->fid
.netfid
;
1596 cinode
= CIFS_I(file_inode(file
));
1598 if (cap_unix(tcon
->ses
) &&
1599 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
1600 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
1603 * BB add code here to normalize offset and length to account for
1604 * negative length which we can not accept over the wire.
1606 if (IS_GETLK(cmd
)) {
1607 rc
= cifs_getlk(file
, flock
, type
, wait_flag
, posix_lck
, xid
);
1612 if (!lock
&& !unlock
) {
1614 * if no lock or unlock then nothing to do since we do not
1621 rc
= cifs_setlk(file
, flock
, type
, wait_flag
, posix_lck
, lock
, unlock
,
1628 * update the file size (if needed) after a write. Should be called with
1629 * the inode->i_lock held
1632 cifs_update_eof(struct cifsInodeInfo
*cifsi
, loff_t offset
,
1633 unsigned int bytes_written
)
1635 loff_t end_of_write
= offset
+ bytes_written
;
1637 if (end_of_write
> cifsi
->server_eof
)
1638 cifsi
->server_eof
= end_of_write
;
1642 cifs_write(struct cifsFileInfo
*open_file
, __u32 pid
, const char *write_data
,
1643 size_t write_size
, loff_t
*offset
)
1646 unsigned int bytes_written
= 0;
1647 unsigned int total_written
;
1648 struct cifs_sb_info
*cifs_sb
;
1649 struct cifs_tcon
*tcon
;
1650 struct TCP_Server_Info
*server
;
1652 struct dentry
*dentry
= open_file
->dentry
;
1653 struct cifsInodeInfo
*cifsi
= CIFS_I(d_inode(dentry
));
1654 struct cifs_io_parms io_parms
;
1656 cifs_sb
= CIFS_SB(dentry
->d_sb
);
1658 cifs_dbg(FYI
, "write %zd bytes to offset %lld of %pd\n",
1659 write_size
, *offset
, dentry
);
1661 tcon
= tlink_tcon(open_file
->tlink
);
1662 server
= tcon
->ses
->server
;
1664 if (!server
->ops
->sync_write
)
1669 for (total_written
= 0; write_size
> total_written
;
1670 total_written
+= bytes_written
) {
1672 while (rc
== -EAGAIN
) {
1676 if (open_file
->invalidHandle
) {
1677 /* we could deadlock if we called
1678 filemap_fdatawait from here so tell
1679 reopen_file not to flush data to
1681 rc
= cifs_reopen_file(open_file
, false);
1686 len
= min(server
->ops
->wp_retry_size(d_inode(dentry
)),
1687 (unsigned int)write_size
- total_written
);
1688 /* iov[0] is reserved for smb header */
1689 iov
[1].iov_base
= (char *)write_data
+ total_written
;
1690 iov
[1].iov_len
= len
;
1692 io_parms
.tcon
= tcon
;
1693 io_parms
.offset
= *offset
;
1694 io_parms
.length
= len
;
1695 rc
= server
->ops
->sync_write(xid
, &open_file
->fid
,
1696 &io_parms
, &bytes_written
, iov
, 1);
1698 if (rc
|| (bytes_written
== 0)) {
1706 spin_lock(&d_inode(dentry
)->i_lock
);
1707 cifs_update_eof(cifsi
, *offset
, bytes_written
);
1708 spin_unlock(&d_inode(dentry
)->i_lock
);
1709 *offset
+= bytes_written
;
1713 cifs_stats_bytes_written(tcon
, total_written
);
1715 if (total_written
> 0) {
1716 spin_lock(&d_inode(dentry
)->i_lock
);
1717 if (*offset
> d_inode(dentry
)->i_size
)
1718 i_size_write(d_inode(dentry
), *offset
);
1719 spin_unlock(&d_inode(dentry
)->i_lock
);
1721 mark_inode_dirty_sync(d_inode(dentry
));
1723 return total_written
;
1726 struct cifsFileInfo
*find_readable_file(struct cifsInodeInfo
*cifs_inode
,
1729 struct cifsFileInfo
*open_file
= NULL
;
1730 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1732 /* only filter by fsuid on multiuser mounts */
1733 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1736 spin_lock(&cifs_file_list_lock
);
1737 /* we could simply get the first_list_entry since write-only entries
1738 are always at the end of the list but since the first entry might
1739 have a close pending, we go through the whole list */
1740 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1741 if (fsuid_only
&& !uid_eq(open_file
->uid
, current_fsuid()))
1743 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_READ
) {
1744 if (!open_file
->invalidHandle
) {
1745 /* found a good file */
1746 /* lock it so it will not be closed on us */
1747 cifsFileInfo_get_locked(open_file
);
1748 spin_unlock(&cifs_file_list_lock
);
1750 } /* else might as well continue, and look for
1751 another, or simply have the caller reopen it
1752 again rather than trying to fix this handle */
1753 } else /* write only file */
1754 break; /* write only files are last so must be done */
1756 spin_unlock(&cifs_file_list_lock
);
1760 struct cifsFileInfo
*find_writable_file(struct cifsInodeInfo
*cifs_inode
,
1763 struct cifsFileInfo
*open_file
, *inv_file
= NULL
;
1764 struct cifs_sb_info
*cifs_sb
;
1765 bool any_available
= false;
1767 unsigned int refind
= 0;
1769 /* Having a null inode here (because mapping->host was set to zero by
1770 the VFS or MM) should not happen but we had reports of on oops (due to
1771 it being zero) during stress testcases so we need to check for it */
1773 if (cifs_inode
== NULL
) {
1774 cifs_dbg(VFS
, "Null inode passed to cifs_writeable_file\n");
1779 cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1781 /* only filter by fsuid on multiuser mounts */
1782 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1785 spin_lock(&cifs_file_list_lock
);
1787 if (refind
> MAX_REOPEN_ATT
) {
1788 spin_unlock(&cifs_file_list_lock
);
1791 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1792 if (!any_available
&& open_file
->pid
!= current
->tgid
)
1794 if (fsuid_only
&& !uid_eq(open_file
->uid
, current_fsuid()))
1796 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
1797 if (!open_file
->invalidHandle
) {
1798 /* found a good writable file */
1799 cifsFileInfo_get_locked(open_file
);
1800 spin_unlock(&cifs_file_list_lock
);
1804 inv_file
= open_file
;
1808 /* couldn't find useable FH with same pid, try any available */
1809 if (!any_available
) {
1810 any_available
= true;
1811 goto refind_writable
;
1815 any_available
= false;
1816 cifsFileInfo_get_locked(inv_file
);
1819 spin_unlock(&cifs_file_list_lock
);
1822 rc
= cifs_reopen_file(inv_file
, false);
1826 spin_lock(&cifs_file_list_lock
);
1827 list_move_tail(&inv_file
->flist
,
1828 &cifs_inode
->openFileList
);
1829 spin_unlock(&cifs_file_list_lock
);
1830 cifsFileInfo_put(inv_file
);
1831 spin_lock(&cifs_file_list_lock
);
1834 goto refind_writable
;
1841 static int cifs_partialpagewrite(struct page
*page
, unsigned from
, unsigned to
)
1843 struct address_space
*mapping
= page
->mapping
;
1844 loff_t offset
= (loff_t
)page
->index
<< PAGE_SHIFT
;
1847 int bytes_written
= 0;
1848 struct inode
*inode
;
1849 struct cifsFileInfo
*open_file
;
1851 if (!mapping
|| !mapping
->host
)
1854 inode
= page
->mapping
->host
;
1856 offset
+= (loff_t
)from
;
1857 write_data
= kmap(page
);
1860 if ((to
> PAGE_SIZE
) || (from
> to
)) {
1865 /* racing with truncate? */
1866 if (offset
> mapping
->host
->i_size
) {
1868 return 0; /* don't care */
1871 /* check to make sure that we are not extending the file */
1872 if (mapping
->host
->i_size
- offset
< (loff_t
)to
)
1873 to
= (unsigned)(mapping
->host
->i_size
- offset
);
1875 open_file
= find_writable_file(CIFS_I(mapping
->host
), false);
1877 bytes_written
= cifs_write(open_file
, open_file
->pid
,
1878 write_data
, to
- from
, &offset
);
1879 cifsFileInfo_put(open_file
);
1880 /* Does mm or vfs already set times? */
1881 inode
->i_atime
= inode
->i_mtime
= current_time(inode
);
1882 if ((bytes_written
> 0) && (offset
))
1884 else if (bytes_written
< 0)
1887 cifs_dbg(FYI
, "No writeable filehandles for inode\n");
1895 static struct cifs_writedata
*
1896 wdata_alloc_and_fillpages(pgoff_t tofind
, struct address_space
*mapping
,
1897 pgoff_t end
, pgoff_t
*index
,
1898 unsigned int *found_pages
)
1900 unsigned int nr_pages
;
1901 struct page
**pages
;
1902 struct cifs_writedata
*wdata
;
1904 wdata
= cifs_writedata_alloc((unsigned int)tofind
,
1905 cifs_writev_complete
);
1910 * find_get_pages_tag seems to return a max of 256 on each
1911 * iteration, so we must call it several times in order to
1912 * fill the array or the wsize is effectively limited to
1916 pages
= wdata
->pages
;
1918 nr_pages
= find_get_pages_tag(mapping
, index
,
1919 PAGECACHE_TAG_DIRTY
, tofind
,
1921 *found_pages
+= nr_pages
;
1924 } while (nr_pages
&& tofind
&& *index
<= end
);
1930 wdata_prepare_pages(struct cifs_writedata
*wdata
, unsigned int found_pages
,
1931 struct address_space
*mapping
,
1932 struct writeback_control
*wbc
,
1933 pgoff_t end
, pgoff_t
*index
, pgoff_t
*next
, bool *done
)
1935 unsigned int nr_pages
= 0, i
;
1938 for (i
= 0; i
< found_pages
; i
++) {
1939 page
= wdata
->pages
[i
];
1941 * At this point we hold neither mapping->tree_lock nor
1942 * lock on the page itself: the page may be truncated or
1943 * invalidated (changing page->mapping to NULL), or even
1944 * swizzled back from swapper_space to tmpfs file
1950 else if (!trylock_page(page
))
1953 if (unlikely(page
->mapping
!= mapping
)) {
1958 if (!wbc
->range_cyclic
&& page
->index
> end
) {
1964 if (*next
&& (page
->index
!= *next
)) {
1965 /* Not next consecutive page */
1970 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
1971 wait_on_page_writeback(page
);
1973 if (PageWriteback(page
) ||
1974 !clear_page_dirty_for_io(page
)) {
1980 * This actually clears the dirty bit in the radix tree.
1981 * See cifs_writepage() for more commentary.
1983 set_page_writeback(page
);
1984 if (page_offset(page
) >= i_size_read(mapping
->host
)) {
1987 end_page_writeback(page
);
1991 wdata
->pages
[i
] = page
;
1992 *next
= page
->index
+ 1;
1996 /* reset index to refind any pages skipped */
1998 *index
= wdata
->pages
[0]->index
+ 1;
2000 /* put any pages we aren't going to use */
2001 for (i
= nr_pages
; i
< found_pages
; i
++) {
2002 put_page(wdata
->pages
[i
]);
2003 wdata
->pages
[i
] = NULL
;
2010 wdata_send_pages(struct cifs_writedata
*wdata
, unsigned int nr_pages
,
2011 struct address_space
*mapping
, struct writeback_control
*wbc
)
2014 struct TCP_Server_Info
*server
;
2017 wdata
->sync_mode
= wbc
->sync_mode
;
2018 wdata
->nr_pages
= nr_pages
;
2019 wdata
->offset
= page_offset(wdata
->pages
[0]);
2020 wdata
->pagesz
= PAGE_SIZE
;
2021 wdata
->tailsz
= min(i_size_read(mapping
->host
) -
2022 page_offset(wdata
->pages
[nr_pages
- 1]),
2024 wdata
->bytes
= ((nr_pages
- 1) * PAGE_SIZE
) + wdata
->tailsz
;
2026 if (wdata
->cfile
!= NULL
)
2027 cifsFileInfo_put(wdata
->cfile
);
2028 wdata
->cfile
= find_writable_file(CIFS_I(mapping
->host
), false);
2029 if (!wdata
->cfile
) {
2030 cifs_dbg(VFS
, "No writable handles for inode\n");
2033 wdata
->pid
= wdata
->cfile
->pid
;
2034 server
= tlink_tcon(wdata
->cfile
->tlink
)->ses
->server
;
2035 rc
= server
->ops
->async_writev(wdata
, cifs_writedata_release
);
2038 for (i
= 0; i
< nr_pages
; ++i
)
2039 unlock_page(wdata
->pages
[i
]);
2044 static int cifs_writepages(struct address_space
*mapping
,
2045 struct writeback_control
*wbc
)
2047 struct cifs_sb_info
*cifs_sb
= CIFS_SB(mapping
->host
->i_sb
);
2048 struct TCP_Server_Info
*server
;
2049 bool done
= false, scanned
= false, range_whole
= false;
2051 struct cifs_writedata
*wdata
;
2055 * If wsize is smaller than the page cache size, default to writing
2056 * one page at a time via cifs_writepage
2058 if (cifs_sb
->wsize
< PAGE_SIZE
)
2059 return generic_writepages(mapping
, wbc
);
2061 if (wbc
->range_cyclic
) {
2062 index
= mapping
->writeback_index
; /* Start from prev offset */
2065 index
= wbc
->range_start
>> PAGE_SHIFT
;
2066 end
= wbc
->range_end
>> PAGE_SHIFT
;
2067 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2071 server
= cifs_sb_master_tcon(cifs_sb
)->ses
->server
;
2073 while (!done
&& index
<= end
) {
2074 unsigned int i
, nr_pages
, found_pages
, wsize
, credits
;
2075 pgoff_t next
= 0, tofind
, saved_index
= index
;
2077 rc
= server
->ops
->wait_mtu_credits(server
, cifs_sb
->wsize
,
2082 tofind
= min((wsize
/ PAGE_SIZE
) - 1, end
- index
) + 1;
2084 wdata
= wdata_alloc_and_fillpages(tofind
, mapping
, end
, &index
,
2088 add_credits_and_wake_if(server
, credits
, 0);
2092 if (found_pages
== 0) {
2093 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2094 add_credits_and_wake_if(server
, credits
, 0);
2098 nr_pages
= wdata_prepare_pages(wdata
, found_pages
, mapping
, wbc
,
2099 end
, &index
, &next
, &done
);
2101 /* nothing to write? */
2102 if (nr_pages
== 0) {
2103 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2104 add_credits_and_wake_if(server
, credits
, 0);
2108 wdata
->credits
= credits
;
2110 rc
= wdata_send_pages(wdata
, nr_pages
, mapping
, wbc
);
2112 /* send failure -- clean up the mess */
2114 add_credits_and_wake_if(server
, wdata
->credits
, 0);
2115 for (i
= 0; i
< nr_pages
; ++i
) {
2117 redirty_page_for_writepage(wbc
,
2120 SetPageError(wdata
->pages
[i
]);
2121 end_page_writeback(wdata
->pages
[i
]);
2122 put_page(wdata
->pages
[i
]);
2125 mapping_set_error(mapping
, rc
);
2127 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2129 if (wbc
->sync_mode
== WB_SYNC_ALL
&& rc
== -EAGAIN
) {
2130 index
= saved_index
;
2134 wbc
->nr_to_write
-= nr_pages
;
2135 if (wbc
->nr_to_write
<= 0)
2141 if (!scanned
&& !done
) {
2143 * We hit the last page and there is more work to be done: wrap
2144 * back to the start of the file
2151 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2152 mapping
->writeback_index
= index
;
2158 cifs_writepage_locked(struct page
*page
, struct writeback_control
*wbc
)
2164 /* BB add check for wbc flags */
2166 if (!PageUptodate(page
))
2167 cifs_dbg(FYI
, "ppw - page not up to date\n");
2170 * Set the "writeback" flag, and clear "dirty" in the radix tree.
2172 * A writepage() implementation always needs to do either this,
2173 * or re-dirty the page with "redirty_page_for_writepage()" in
2174 * the case of a failure.
2176 * Just unlocking the page will cause the radix tree tag-bits
2177 * to fail to update with the state of the page correctly.
2179 set_page_writeback(page
);
2181 rc
= cifs_partialpagewrite(page
, 0, PAGE_SIZE
);
2182 if (rc
== -EAGAIN
&& wbc
->sync_mode
== WB_SYNC_ALL
)
2184 else if (rc
== -EAGAIN
)
2185 redirty_page_for_writepage(wbc
, page
);
2189 SetPageUptodate(page
);
2190 end_page_writeback(page
);
2196 static int cifs_writepage(struct page
*page
, struct writeback_control
*wbc
)
2198 int rc
= cifs_writepage_locked(page
, wbc
);
2203 static int cifs_write_end(struct file
*file
, struct address_space
*mapping
,
2204 loff_t pos
, unsigned len
, unsigned copied
,
2205 struct page
*page
, void *fsdata
)
2208 struct inode
*inode
= mapping
->host
;
2209 struct cifsFileInfo
*cfile
= file
->private_data
;
2210 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cfile
->dentry
->d_sb
);
2213 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2216 pid
= current
->tgid
;
2218 cifs_dbg(FYI
, "write_end for page %p from pos %lld with %d bytes\n",
2221 if (PageChecked(page
)) {
2223 SetPageUptodate(page
);
2224 ClearPageChecked(page
);
2225 } else if (!PageUptodate(page
) && copied
== PAGE_SIZE
)
2226 SetPageUptodate(page
);
2228 if (!PageUptodate(page
)) {
2230 unsigned offset
= pos
& (PAGE_SIZE
- 1);
2234 /* this is probably better than directly calling
2235 partialpage_write since in this function the file handle is
2236 known which we might as well leverage */
2237 /* BB check if anything else missing out of ppw
2238 such as updating last write time */
2239 page_data
= kmap(page
);
2240 rc
= cifs_write(cfile
, pid
, page_data
+ offset
, copied
, &pos
);
2241 /* if (rc < 0) should we set writebehind rc? */
2248 set_page_dirty(page
);
2252 spin_lock(&inode
->i_lock
);
2253 if (pos
> inode
->i_size
)
2254 i_size_write(inode
, pos
);
2255 spin_unlock(&inode
->i_lock
);
2264 int cifs_strict_fsync(struct file
*file
, loff_t start
, loff_t end
,
2269 struct cifs_tcon
*tcon
;
2270 struct TCP_Server_Info
*server
;
2271 struct cifsFileInfo
*smbfile
= file
->private_data
;
2272 struct inode
*inode
= file_inode(file
);
2273 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
2275 rc
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
2282 cifs_dbg(FYI
, "Sync file - name: %pD datasync: 0x%x\n",
2285 if (!CIFS_CACHE_READ(CIFS_I(inode
))) {
2286 rc
= cifs_zap_mapping(inode
);
2288 cifs_dbg(FYI
, "rc: %d during invalidate phase\n", rc
);
2289 rc
= 0; /* don't care about it in fsync */
2293 tcon
= tlink_tcon(smbfile
->tlink
);
2294 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOSSYNC
)) {
2295 server
= tcon
->ses
->server
;
2296 if (server
->ops
->flush
)
2297 rc
= server
->ops
->flush(xid
, tcon
, &smbfile
->fid
);
2303 inode_unlock(inode
);
2307 int cifs_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
2311 struct cifs_tcon
*tcon
;
2312 struct TCP_Server_Info
*server
;
2313 struct cifsFileInfo
*smbfile
= file
->private_data
;
2314 struct cifs_sb_info
*cifs_sb
= CIFS_FILE_SB(file
);
2315 struct inode
*inode
= file
->f_mapping
->host
;
2317 rc
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
2324 cifs_dbg(FYI
, "Sync file - name: %pD datasync: 0x%x\n",
2327 tcon
= tlink_tcon(smbfile
->tlink
);
2328 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOSSYNC
)) {
2329 server
= tcon
->ses
->server
;
2330 if (server
->ops
->flush
)
2331 rc
= server
->ops
->flush(xid
, tcon
, &smbfile
->fid
);
2337 inode_unlock(inode
);
2342 * As file closes, flush all cached write data for this inode checking
2343 * for write behind errors.
2345 int cifs_flush(struct file
*file
, fl_owner_t id
)
2347 struct inode
*inode
= file_inode(file
);
2350 if (file
->f_mode
& FMODE_WRITE
)
2351 rc
= filemap_write_and_wait(inode
->i_mapping
);
2353 cifs_dbg(FYI
, "Flush inode %p file %p rc %d\n", inode
, file
, rc
);
2359 cifs_write_allocate_pages(struct page
**pages
, unsigned long num_pages
)
2364 for (i
= 0; i
< num_pages
; i
++) {
2365 pages
[i
] = alloc_page(GFP_KERNEL
|__GFP_HIGHMEM
);
2368 * save number of pages we have already allocated and
2369 * return with ENOMEM error
2378 for (i
= 0; i
< num_pages
; i
++)
2385 size_t get_numpages(const size_t wsize
, const size_t len
, size_t *cur_len
)
2390 clen
= min_t(const size_t, len
, wsize
);
2391 num_pages
= DIV_ROUND_UP(clen
, PAGE_SIZE
);
2400 cifs_uncached_writedata_release(struct kref
*refcount
)
2403 struct cifs_writedata
*wdata
= container_of(refcount
,
2404 struct cifs_writedata
, refcount
);
2406 for (i
= 0; i
< wdata
->nr_pages
; i
++)
2407 put_page(wdata
->pages
[i
]);
2408 cifs_writedata_release(refcount
);
2412 cifs_uncached_writev_complete(struct work_struct
*work
)
2414 struct cifs_writedata
*wdata
= container_of(work
,
2415 struct cifs_writedata
, work
);
2416 struct inode
*inode
= d_inode(wdata
->cfile
->dentry
);
2417 struct cifsInodeInfo
*cifsi
= CIFS_I(inode
);
2419 spin_lock(&inode
->i_lock
);
2420 cifs_update_eof(cifsi
, wdata
->offset
, wdata
->bytes
);
2421 if (cifsi
->server_eof
> inode
->i_size
)
2422 i_size_write(inode
, cifsi
->server_eof
);
2423 spin_unlock(&inode
->i_lock
);
2425 complete(&wdata
->done
);
2427 kref_put(&wdata
->refcount
, cifs_uncached_writedata_release
);
2431 wdata_fill_from_iovec(struct cifs_writedata
*wdata
, struct iov_iter
*from
,
2432 size_t *len
, unsigned long *num_pages
)
2434 size_t save_len
, copied
, bytes
, cur_len
= *len
;
2435 unsigned long i
, nr_pages
= *num_pages
;
2438 for (i
= 0; i
< nr_pages
; i
++) {
2439 bytes
= min_t(const size_t, cur_len
, PAGE_SIZE
);
2440 copied
= copy_page_from_iter(wdata
->pages
[i
], 0, bytes
, from
);
2443 * If we didn't copy as much as we expected, then that
2444 * may mean we trod into an unmapped area. Stop copying
2445 * at that point. On the next pass through the big
2446 * loop, we'll likely end up getting a zero-length
2447 * write and bailing out of it.
2452 cur_len
= save_len
- cur_len
;
2456 * If we have no data to send, then that probably means that
2457 * the copy above failed altogether. That's most likely because
2458 * the address in the iovec was bogus. Return -EFAULT and let
2459 * the caller free anything we allocated and bail out.
2465 * i + 1 now represents the number of pages we actually used in
2466 * the copy phase above.
2473 cifs_write_from_iter(loff_t offset
, size_t len
, struct iov_iter
*from
,
2474 struct cifsFileInfo
*open_file
,
2475 struct cifs_sb_info
*cifs_sb
, struct list_head
*wdata_list
)
2479 unsigned long nr_pages
, num_pages
, i
;
2480 struct cifs_writedata
*wdata
;
2481 struct iov_iter saved_from
= *from
;
2482 loff_t saved_offset
= offset
;
2484 struct TCP_Server_Info
*server
;
2486 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2487 pid
= open_file
->pid
;
2489 pid
= current
->tgid
;
2491 server
= tlink_tcon(open_file
->tlink
)->ses
->server
;
2494 unsigned int wsize
, credits
;
2496 rc
= server
->ops
->wait_mtu_credits(server
, cifs_sb
->wsize
,
2501 nr_pages
= get_numpages(wsize
, len
, &cur_len
);
2502 wdata
= cifs_writedata_alloc(nr_pages
,
2503 cifs_uncached_writev_complete
);
2506 add_credits_and_wake_if(server
, credits
, 0);
2510 rc
= cifs_write_allocate_pages(wdata
->pages
, nr_pages
);
2513 add_credits_and_wake_if(server
, credits
, 0);
2517 num_pages
= nr_pages
;
2518 rc
= wdata_fill_from_iovec(wdata
, from
, &cur_len
, &num_pages
);
2520 for (i
= 0; i
< nr_pages
; i
++)
2521 put_page(wdata
->pages
[i
]);
2523 add_credits_and_wake_if(server
, credits
, 0);
2528 * Bring nr_pages down to the number of pages we actually used,
2529 * and free any pages that we didn't use.
2531 for ( ; nr_pages
> num_pages
; nr_pages
--)
2532 put_page(wdata
->pages
[nr_pages
- 1]);
2534 wdata
->sync_mode
= WB_SYNC_ALL
;
2535 wdata
->nr_pages
= nr_pages
;
2536 wdata
->offset
= (__u64
)offset
;
2537 wdata
->cfile
= cifsFileInfo_get(open_file
);
2539 wdata
->bytes
= cur_len
;
2540 wdata
->pagesz
= PAGE_SIZE
;
2541 wdata
->tailsz
= cur_len
- ((nr_pages
- 1) * PAGE_SIZE
);
2542 wdata
->credits
= credits
;
2544 if (!wdata
->cfile
->invalidHandle
||
2545 !cifs_reopen_file(wdata
->cfile
, false))
2546 rc
= server
->ops
->async_writev(wdata
,
2547 cifs_uncached_writedata_release
);
2549 add_credits_and_wake_if(server
, wdata
->credits
, 0);
2550 kref_put(&wdata
->refcount
,
2551 cifs_uncached_writedata_release
);
2552 if (rc
== -EAGAIN
) {
2554 iov_iter_advance(from
, offset
- saved_offset
);
2560 list_add_tail(&wdata
->list
, wdata_list
);
2568 ssize_t
cifs_user_writev(struct kiocb
*iocb
, struct iov_iter
*from
)
2570 struct file
*file
= iocb
->ki_filp
;
2571 ssize_t total_written
= 0;
2572 struct cifsFileInfo
*open_file
;
2573 struct cifs_tcon
*tcon
;
2574 struct cifs_sb_info
*cifs_sb
;
2575 struct cifs_writedata
*wdata
, *tmp
;
2576 struct list_head wdata_list
;
2577 struct iov_iter saved_from
= *from
;
2581 * BB - optimize the way when signing is disabled. We can drop this
2582 * extra memory-to-memory copying and use iovec buffers for constructing
2586 rc
= generic_write_checks(iocb
, from
);
2590 INIT_LIST_HEAD(&wdata_list
);
2591 cifs_sb
= CIFS_FILE_SB(file
);
2592 open_file
= file
->private_data
;
2593 tcon
= tlink_tcon(open_file
->tlink
);
2595 if (!tcon
->ses
->server
->ops
->async_writev
)
2598 rc
= cifs_write_from_iter(iocb
->ki_pos
, iov_iter_count(from
), from
,
2599 open_file
, cifs_sb
, &wdata_list
);
2602 * If at least one write was successfully sent, then discard any rc
2603 * value from the later writes. If the other write succeeds, then
2604 * we'll end up returning whatever was written. If it fails, then
2605 * we'll get a new rc value from that.
2607 if (!list_empty(&wdata_list
))
2611 * Wait for and collect replies for any successful sends in order of
2612 * increasing offset. Once an error is hit or we get a fatal signal
2613 * while waiting, then return without waiting for any more replies.
2616 list_for_each_entry_safe(wdata
, tmp
, &wdata_list
, list
) {
2618 /* FIXME: freezable too? */
2619 rc
= wait_for_completion_killable(&wdata
->done
);
2622 else if (wdata
->result
)
2625 total_written
+= wdata
->bytes
;
2627 /* resend call if it's a retryable error */
2628 if (rc
== -EAGAIN
) {
2629 struct list_head tmp_list
;
2630 struct iov_iter tmp_from
= saved_from
;
2632 INIT_LIST_HEAD(&tmp_list
);
2633 list_del_init(&wdata
->list
);
2635 iov_iter_advance(&tmp_from
,
2636 wdata
->offset
- iocb
->ki_pos
);
2638 rc
= cifs_write_from_iter(wdata
->offset
,
2639 wdata
->bytes
, &tmp_from
,
2640 open_file
, cifs_sb
, &tmp_list
);
2642 list_splice(&tmp_list
, &wdata_list
);
2644 kref_put(&wdata
->refcount
,
2645 cifs_uncached_writedata_release
);
2649 list_del_init(&wdata
->list
);
2650 kref_put(&wdata
->refcount
, cifs_uncached_writedata_release
);
2653 if (unlikely(!total_written
))
2656 iocb
->ki_pos
+= total_written
;
2657 set_bit(CIFS_INO_INVALID_MAPPING
, &CIFS_I(file_inode(file
))->flags
);
2658 cifs_stats_bytes_written(tcon
, total_written
);
2659 return total_written
;
2663 cifs_writev(struct kiocb
*iocb
, struct iov_iter
*from
)
2665 struct file
*file
= iocb
->ki_filp
;
2666 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
2667 struct inode
*inode
= file
->f_mapping
->host
;
2668 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
2669 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
2673 * We need to hold the sem to be sure nobody modifies lock list
2674 * with a brlock that prevents writing.
2676 down_read(&cinode
->lock_sem
);
2679 rc
= generic_write_checks(iocb
, from
);
2683 if (!cifs_find_lock_conflict(cfile
, iocb
->ki_pos
, iov_iter_count(from
),
2684 server
->vals
->exclusive_lock_type
, NULL
,
2686 rc
= __generic_file_write_iter(iocb
, from
);
2690 inode_unlock(inode
);
2693 rc
= generic_write_sync(iocb
, rc
);
2694 up_read(&cinode
->lock_sem
);
2699 cifs_strict_writev(struct kiocb
*iocb
, struct iov_iter
*from
)
2701 struct inode
*inode
= file_inode(iocb
->ki_filp
);
2702 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
2703 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
2704 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)
2705 iocb
->ki_filp
->private_data
;
2706 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
2709 written
= cifs_get_writer(cinode
);
2713 if (CIFS_CACHE_WRITE(cinode
)) {
2714 if (cap_unix(tcon
->ses
) &&
2715 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
))
2716 && ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0)) {
2717 written
= generic_file_write_iter(iocb
, from
);
2720 written
= cifs_writev(iocb
, from
);
2724 * For non-oplocked files in strict cache mode we need to write the data
2725 * to the server exactly from the pos to pos+len-1 rather than flush all
2726 * affected pages because it may cause a error with mandatory locks on
2727 * these pages but not on the region from pos to ppos+len-1.
2729 written
= cifs_user_writev(iocb
, from
);
2730 if (written
> 0 && CIFS_CACHE_READ(cinode
)) {
2732 * Windows 7 server can delay breaking level2 oplock if a write
2733 * request comes - break it on the client to prevent reading
2736 cifs_zap_mapping(inode
);
2737 cifs_dbg(FYI
, "Set no oplock for inode=%p after a write operation\n",
2742 cifs_put_writer(cinode
);
2746 static struct cifs_readdata
*
2747 cifs_readdata_alloc(unsigned int nr_pages
, work_func_t complete
)
2749 struct cifs_readdata
*rdata
;
2751 rdata
= kzalloc(sizeof(*rdata
) + (sizeof(struct page
*) * nr_pages
),
2753 if (rdata
!= NULL
) {
2754 kref_init(&rdata
->refcount
);
2755 INIT_LIST_HEAD(&rdata
->list
);
2756 init_completion(&rdata
->done
);
2757 INIT_WORK(&rdata
->work
, complete
);
2764 cifs_readdata_release(struct kref
*refcount
)
2766 struct cifs_readdata
*rdata
= container_of(refcount
,
2767 struct cifs_readdata
, refcount
);
2770 cifsFileInfo_put(rdata
->cfile
);
2776 cifs_read_allocate_pages(struct cifs_readdata
*rdata
, unsigned int nr_pages
)
2782 for (i
= 0; i
< nr_pages
; i
++) {
2783 page
= alloc_page(GFP_KERNEL
|__GFP_HIGHMEM
);
2788 rdata
->pages
[i
] = page
;
2792 for (i
= 0; i
< nr_pages
; i
++) {
2793 put_page(rdata
->pages
[i
]);
2794 rdata
->pages
[i
] = NULL
;
2801 cifs_uncached_readdata_release(struct kref
*refcount
)
2803 struct cifs_readdata
*rdata
= container_of(refcount
,
2804 struct cifs_readdata
, refcount
);
2807 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
2808 put_page(rdata
->pages
[i
]);
2809 rdata
->pages
[i
] = NULL
;
2811 cifs_readdata_release(refcount
);
2815 * cifs_readdata_to_iov - copy data from pages in response to an iovec
2816 * @rdata: the readdata response with list of pages holding data
2817 * @iter: destination for our data
2819 * This function copies data from a list of pages in a readdata response into
2820 * an array of iovecs. It will first calculate where the data should go
2821 * based on the info in the readdata and then copy the data into that spot.
2824 cifs_readdata_to_iov(struct cifs_readdata
*rdata
, struct iov_iter
*iter
)
2826 size_t remaining
= rdata
->got_bytes
;
2829 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
2830 struct page
*page
= rdata
->pages
[i
];
2831 size_t copy
= min_t(size_t, remaining
, PAGE_SIZE
);
2832 size_t written
= copy_page_to_iter(page
, 0, copy
, iter
);
2833 remaining
-= written
;
2834 if (written
< copy
&& iov_iter_count(iter
) > 0)
2837 return remaining
? -EFAULT
: 0;
2841 cifs_uncached_readv_complete(struct work_struct
*work
)
2843 struct cifs_readdata
*rdata
= container_of(work
,
2844 struct cifs_readdata
, work
);
2846 complete(&rdata
->done
);
2847 kref_put(&rdata
->refcount
, cifs_uncached_readdata_release
);
2851 cifs_uncached_read_into_pages(struct TCP_Server_Info
*server
,
2852 struct cifs_readdata
*rdata
, unsigned int len
)
2856 unsigned int nr_pages
= rdata
->nr_pages
;
2858 rdata
->got_bytes
= 0;
2859 rdata
->tailsz
= PAGE_SIZE
;
2860 for (i
= 0; i
< nr_pages
; i
++) {
2861 struct page
*page
= rdata
->pages
[i
];
2865 /* no need to hold page hostage */
2866 rdata
->pages
[i
] = NULL
;
2872 if (len
>= PAGE_SIZE
) {
2873 /* enough data to fill the page */
2877 zero_user(page
, len
, PAGE_SIZE
- len
);
2878 rdata
->tailsz
= len
;
2881 result
= cifs_read_page_from_socket(server
, page
, n
);
2885 rdata
->got_bytes
+= result
;
2888 return rdata
->got_bytes
> 0 && result
!= -ECONNABORTED
?
2889 rdata
->got_bytes
: result
;
2893 cifs_send_async_read(loff_t offset
, size_t len
, struct cifsFileInfo
*open_file
,
2894 struct cifs_sb_info
*cifs_sb
, struct list_head
*rdata_list
)
2896 struct cifs_readdata
*rdata
;
2897 unsigned int npages
, rsize
, credits
;
2901 struct TCP_Server_Info
*server
;
2903 server
= tlink_tcon(open_file
->tlink
)->ses
->server
;
2905 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2906 pid
= open_file
->pid
;
2908 pid
= current
->tgid
;
2911 rc
= server
->ops
->wait_mtu_credits(server
, cifs_sb
->rsize
,
2916 cur_len
= min_t(const size_t, len
, rsize
);
2917 npages
= DIV_ROUND_UP(cur_len
, PAGE_SIZE
);
2919 /* allocate a readdata struct */
2920 rdata
= cifs_readdata_alloc(npages
,
2921 cifs_uncached_readv_complete
);
2923 add_credits_and_wake_if(server
, credits
, 0);
2928 rc
= cifs_read_allocate_pages(rdata
, npages
);
2932 rdata
->cfile
= cifsFileInfo_get(open_file
);
2933 rdata
->nr_pages
= npages
;
2934 rdata
->offset
= offset
;
2935 rdata
->bytes
= cur_len
;
2937 rdata
->pagesz
= PAGE_SIZE
;
2938 rdata
->read_into_pages
= cifs_uncached_read_into_pages
;
2939 rdata
->credits
= credits
;
2941 if (!rdata
->cfile
->invalidHandle
||
2942 !cifs_reopen_file(rdata
->cfile
, true))
2943 rc
= server
->ops
->async_readv(rdata
);
2946 add_credits_and_wake_if(server
, rdata
->credits
, 0);
2947 kref_put(&rdata
->refcount
,
2948 cifs_uncached_readdata_release
);
2954 list_add_tail(&rdata
->list
, rdata_list
);
2962 ssize_t
cifs_user_readv(struct kiocb
*iocb
, struct iov_iter
*to
)
2964 struct file
*file
= iocb
->ki_filp
;
2967 ssize_t total_read
= 0;
2968 loff_t offset
= iocb
->ki_pos
;
2969 struct cifs_sb_info
*cifs_sb
;
2970 struct cifs_tcon
*tcon
;
2971 struct cifsFileInfo
*open_file
;
2972 struct cifs_readdata
*rdata
, *tmp
;
2973 struct list_head rdata_list
;
2975 len
= iov_iter_count(to
);
2979 INIT_LIST_HEAD(&rdata_list
);
2980 cifs_sb
= CIFS_FILE_SB(file
);
2981 open_file
= file
->private_data
;
2982 tcon
= tlink_tcon(open_file
->tlink
);
2984 if (!tcon
->ses
->server
->ops
->async_readv
)
2987 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
2988 cifs_dbg(FYI
, "attempting read on write only file instance\n");
2990 rc
= cifs_send_async_read(offset
, len
, open_file
, cifs_sb
, &rdata_list
);
2992 /* if at least one read request send succeeded, then reset rc */
2993 if (!list_empty(&rdata_list
))
2996 len
= iov_iter_count(to
);
2997 /* the loop below should proceed in the order of increasing offsets */
2999 list_for_each_entry_safe(rdata
, tmp
, &rdata_list
, list
) {
3001 /* FIXME: freezable sleep too? */
3002 rc
= wait_for_completion_killable(&rdata
->done
);
3005 else if (rdata
->result
== -EAGAIN
) {
3006 /* resend call if it's a retryable error */
3007 struct list_head tmp_list
;
3008 unsigned int got_bytes
= rdata
->got_bytes
;
3010 list_del_init(&rdata
->list
);
3011 INIT_LIST_HEAD(&tmp_list
);
3014 * Got a part of data and then reconnect has
3015 * happened -- fill the buffer and continue
3018 if (got_bytes
&& got_bytes
< rdata
->bytes
) {
3019 rc
= cifs_readdata_to_iov(rdata
, to
);
3021 kref_put(&rdata
->refcount
,
3022 cifs_uncached_readdata_release
);
3027 rc
= cifs_send_async_read(
3028 rdata
->offset
+ got_bytes
,
3029 rdata
->bytes
- got_bytes
,
3030 rdata
->cfile
, cifs_sb
,
3033 list_splice(&tmp_list
, &rdata_list
);
3035 kref_put(&rdata
->refcount
,
3036 cifs_uncached_readdata_release
);
3038 } else if (rdata
->result
)
3041 rc
= cifs_readdata_to_iov(rdata
, to
);
3043 /* if there was a short read -- discard anything left */
3044 if (rdata
->got_bytes
&& rdata
->got_bytes
< rdata
->bytes
)
3047 list_del_init(&rdata
->list
);
3048 kref_put(&rdata
->refcount
, cifs_uncached_readdata_release
);
3051 total_read
= len
- iov_iter_count(to
);
3053 cifs_stats_bytes_read(tcon
, total_read
);
3055 /* mask nodata case */
3060 iocb
->ki_pos
+= total_read
;
3067 cifs_strict_readv(struct kiocb
*iocb
, struct iov_iter
*to
)
3069 struct inode
*inode
= file_inode(iocb
->ki_filp
);
3070 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
3071 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
3072 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)
3073 iocb
->ki_filp
->private_data
;
3074 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
3078 * In strict cache mode we need to read from the server all the time
3079 * if we don't have level II oplock because the server can delay mtime
3080 * change - so we can't make a decision about inode invalidating.
3081 * And we can also fail with pagereading if there are mandatory locks
3082 * on pages affected by this read but not on the region from pos to
3085 if (!CIFS_CACHE_READ(cinode
))
3086 return cifs_user_readv(iocb
, to
);
3088 if (cap_unix(tcon
->ses
) &&
3089 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
3090 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
3091 return generic_file_read_iter(iocb
, to
);
3094 * We need to hold the sem to be sure nobody modifies lock list
3095 * with a brlock that prevents reading.
3097 down_read(&cinode
->lock_sem
);
3098 if (!cifs_find_lock_conflict(cfile
, iocb
->ki_pos
, iov_iter_count(to
),
3099 tcon
->ses
->server
->vals
->shared_lock_type
,
3100 NULL
, CIFS_READ_OP
))
3101 rc
= generic_file_read_iter(iocb
, to
);
3102 up_read(&cinode
->lock_sem
);
3107 cifs_read(struct file
*file
, char *read_data
, size_t read_size
, loff_t
*offset
)
3110 unsigned int bytes_read
= 0;
3111 unsigned int total_read
;
3112 unsigned int current_read_size
;
3114 struct cifs_sb_info
*cifs_sb
;
3115 struct cifs_tcon
*tcon
;
3116 struct TCP_Server_Info
*server
;
3119 struct cifsFileInfo
*open_file
;
3120 struct cifs_io_parms io_parms
;
3121 int buf_type
= CIFS_NO_BUFFER
;
3125 cifs_sb
= CIFS_FILE_SB(file
);
3127 /* FIXME: set up handlers for larger reads and/or convert to async */
3128 rsize
= min_t(unsigned int, cifs_sb
->rsize
, CIFSMaxBufSize
);
3130 if (file
->private_data
== NULL
) {
3135 open_file
= file
->private_data
;
3136 tcon
= tlink_tcon(open_file
->tlink
);
3137 server
= tcon
->ses
->server
;
3139 if (!server
->ops
->sync_read
) {
3144 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
3145 pid
= open_file
->pid
;
3147 pid
= current
->tgid
;
3149 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
3150 cifs_dbg(FYI
, "attempting read on write only file instance\n");
3152 for (total_read
= 0, cur_offset
= read_data
; read_size
> total_read
;
3153 total_read
+= bytes_read
, cur_offset
+= bytes_read
) {
3155 current_read_size
= min_t(uint
, read_size
- total_read
,
3158 * For windows me and 9x we do not want to request more
3159 * than it negotiated since it will refuse the read
3162 if ((tcon
->ses
) && !(tcon
->ses
->capabilities
&
3163 tcon
->ses
->server
->vals
->cap_large_files
)) {
3164 current_read_size
= min_t(uint
,
3165 current_read_size
, CIFSMaxBufSize
);
3167 if (open_file
->invalidHandle
) {
3168 rc
= cifs_reopen_file(open_file
, true);
3173 io_parms
.tcon
= tcon
;
3174 io_parms
.offset
= *offset
;
3175 io_parms
.length
= current_read_size
;
3176 rc
= server
->ops
->sync_read(xid
, &open_file
->fid
, &io_parms
,
3177 &bytes_read
, &cur_offset
,
3179 } while (rc
== -EAGAIN
);
3181 if (rc
|| (bytes_read
== 0)) {
3189 cifs_stats_bytes_read(tcon
, total_read
);
3190 *offset
+= bytes_read
;
3198 * If the page is mmap'ed into a process' page tables, then we need to make
3199 * sure that it doesn't change while being written back.
3202 cifs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
3204 struct page
*page
= vmf
->page
;
3207 return VM_FAULT_LOCKED
;
3210 static const struct vm_operations_struct cifs_file_vm_ops
= {
3211 .fault
= filemap_fault
,
3212 .map_pages
= filemap_map_pages
,
3213 .page_mkwrite
= cifs_page_mkwrite
,
3216 int cifs_file_strict_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3219 struct inode
*inode
= file_inode(file
);
3223 if (!CIFS_CACHE_READ(CIFS_I(inode
))) {
3224 rc
= cifs_zap_mapping(inode
);
3229 rc
= generic_file_mmap(file
, vma
);
3231 vma
->vm_ops
= &cifs_file_vm_ops
;
3236 int cifs_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3241 rc
= cifs_revalidate_file(file
);
3243 cifs_dbg(FYI
, "Validation prior to mmap failed, error=%d\n",
3248 rc
= generic_file_mmap(file
, vma
);
3250 vma
->vm_ops
= &cifs_file_vm_ops
;
3256 cifs_readv_complete(struct work_struct
*work
)
3258 unsigned int i
, got_bytes
;
3259 struct cifs_readdata
*rdata
= container_of(work
,
3260 struct cifs_readdata
, work
);
3262 got_bytes
= rdata
->got_bytes
;
3263 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
3264 struct page
*page
= rdata
->pages
[i
];
3266 lru_cache_add_file(page
);
3268 if (rdata
->result
== 0 ||
3269 (rdata
->result
== -EAGAIN
&& got_bytes
)) {
3270 flush_dcache_page(page
);
3271 SetPageUptodate(page
);
3276 if (rdata
->result
== 0 ||
3277 (rdata
->result
== -EAGAIN
&& got_bytes
))
3278 cifs_readpage_to_fscache(rdata
->mapping
->host
, page
);
3280 got_bytes
-= min_t(unsigned int, PAGE_SIZE
, got_bytes
);
3283 rdata
->pages
[i
] = NULL
;
3285 kref_put(&rdata
->refcount
, cifs_readdata_release
);
3289 cifs_readpages_read_into_pages(struct TCP_Server_Info
*server
,
3290 struct cifs_readdata
*rdata
, unsigned int len
)
3296 unsigned int nr_pages
= rdata
->nr_pages
;
3298 /* determine the eof that the server (probably) has */
3299 eof
= CIFS_I(rdata
->mapping
->host
)->server_eof
;
3300 eof_index
= eof
? (eof
- 1) >> PAGE_SHIFT
: 0;
3301 cifs_dbg(FYI
, "eof=%llu eof_index=%lu\n", eof
, eof_index
);
3303 rdata
->got_bytes
= 0;
3304 rdata
->tailsz
= PAGE_SIZE
;
3305 for (i
= 0; i
< nr_pages
; i
++) {
3306 struct page
*page
= rdata
->pages
[i
];
3307 size_t n
= PAGE_SIZE
;
3309 if (len
>= PAGE_SIZE
) {
3311 } else if (len
> 0) {
3312 /* enough for partial page, fill and zero the rest */
3313 zero_user(page
, len
, PAGE_SIZE
- len
);
3314 n
= rdata
->tailsz
= len
;
3316 } else if (page
->index
> eof_index
) {
3318 * The VFS will not try to do readahead past the
3319 * i_size, but it's possible that we have outstanding
3320 * writes with gaps in the middle and the i_size hasn't
3321 * caught up yet. Populate those with zeroed out pages
3322 * to prevent the VFS from repeatedly attempting to
3323 * fill them until the writes are flushed.
3325 zero_user(page
, 0, PAGE_SIZE
);
3326 lru_cache_add_file(page
);
3327 flush_dcache_page(page
);
3328 SetPageUptodate(page
);
3331 rdata
->pages
[i
] = NULL
;
3335 /* no need to hold page hostage */
3336 lru_cache_add_file(page
);
3339 rdata
->pages
[i
] = NULL
;
3344 result
= cifs_read_page_from_socket(server
, page
, n
);
3348 rdata
->got_bytes
+= result
;
3351 return rdata
->got_bytes
> 0 && result
!= -ECONNABORTED
?
3352 rdata
->got_bytes
: result
;
3356 readpages_get_pages(struct address_space
*mapping
, struct list_head
*page_list
,
3357 unsigned int rsize
, struct list_head
*tmplist
,
3358 unsigned int *nr_pages
, loff_t
*offset
, unsigned int *bytes
)
3360 struct page
*page
, *tpage
;
3361 unsigned int expected_index
;
3363 gfp_t gfp
= readahead_gfp_mask(mapping
);
3365 INIT_LIST_HEAD(tmplist
);
3367 page
= list_entry(page_list
->prev
, struct page
, lru
);
3370 * Lock the page and put it in the cache. Since no one else
3371 * should have access to this page, we're safe to simply set
3372 * PG_locked without checking it first.
3374 __SetPageLocked(page
);
3375 rc
= add_to_page_cache_locked(page
, mapping
,
3378 /* give up if we can't stick it in the cache */
3380 __ClearPageLocked(page
);
3384 /* move first page to the tmplist */
3385 *offset
= (loff_t
)page
->index
<< PAGE_SHIFT
;
3388 list_move_tail(&page
->lru
, tmplist
);
3390 /* now try and add more pages onto the request */
3391 expected_index
= page
->index
+ 1;
3392 list_for_each_entry_safe_reverse(page
, tpage
, page_list
, lru
) {
3393 /* discontinuity ? */
3394 if (page
->index
!= expected_index
)
3397 /* would this page push the read over the rsize? */
3398 if (*bytes
+ PAGE_SIZE
> rsize
)
3401 __SetPageLocked(page
);
3402 if (add_to_page_cache_locked(page
, mapping
, page
->index
, gfp
)) {
3403 __ClearPageLocked(page
);
3406 list_move_tail(&page
->lru
, tmplist
);
3407 (*bytes
) += PAGE_SIZE
;
3414 static int cifs_readpages(struct file
*file
, struct address_space
*mapping
,
3415 struct list_head
*page_list
, unsigned num_pages
)
3418 struct list_head tmplist
;
3419 struct cifsFileInfo
*open_file
= file
->private_data
;
3420 struct cifs_sb_info
*cifs_sb
= CIFS_FILE_SB(file
);
3421 struct TCP_Server_Info
*server
;
3425 * Reads as many pages as possible from fscache. Returns -ENOBUFS
3426 * immediately if the cookie is negative
3428 * After this point, every page in the list might have PG_fscache set,
3429 * so we will need to clean that up off of every page we don't use.
3431 rc
= cifs_readpages_from_fscache(mapping
->host
, mapping
, page_list
,
3436 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
3437 pid
= open_file
->pid
;
3439 pid
= current
->tgid
;
3442 server
= tlink_tcon(open_file
->tlink
)->ses
->server
;
3444 cifs_dbg(FYI
, "%s: file=%p mapping=%p num_pages=%u\n",
3445 __func__
, file
, mapping
, num_pages
);
3448 * Start with the page at end of list and move it to private
3449 * list. Do the same with any following pages until we hit
3450 * the rsize limit, hit an index discontinuity, or run out of
3451 * pages. Issue the async read and then start the loop again
3452 * until the list is empty.
3454 * Note that list order is important. The page_list is in
3455 * the order of declining indexes. When we put the pages in
3456 * the rdata->pages, then we want them in increasing order.
3458 while (!list_empty(page_list
)) {
3459 unsigned int i
, nr_pages
, bytes
, rsize
;
3461 struct page
*page
, *tpage
;
3462 struct cifs_readdata
*rdata
;
3465 rc
= server
->ops
->wait_mtu_credits(server
, cifs_sb
->rsize
,
3471 * Give up immediately if rsize is too small to read an entire
3472 * page. The VFS will fall back to readpage. We should never
3473 * reach this point however since we set ra_pages to 0 when the
3474 * rsize is smaller than a cache page.
3476 if (unlikely(rsize
< PAGE_SIZE
)) {
3477 add_credits_and_wake_if(server
, credits
, 0);
3481 rc
= readpages_get_pages(mapping
, page_list
, rsize
, &tmplist
,
3482 &nr_pages
, &offset
, &bytes
);
3484 add_credits_and_wake_if(server
, credits
, 0);
3488 rdata
= cifs_readdata_alloc(nr_pages
, cifs_readv_complete
);
3490 /* best to give up if we're out of mem */
3491 list_for_each_entry_safe(page
, tpage
, &tmplist
, lru
) {
3492 list_del(&page
->lru
);
3493 lru_cache_add_file(page
);
3498 add_credits_and_wake_if(server
, credits
, 0);
3502 rdata
->cfile
= cifsFileInfo_get(open_file
);
3503 rdata
->mapping
= mapping
;
3504 rdata
->offset
= offset
;
3505 rdata
->bytes
= bytes
;
3507 rdata
->pagesz
= PAGE_SIZE
;
3508 rdata
->read_into_pages
= cifs_readpages_read_into_pages
;
3509 rdata
->credits
= credits
;
3511 list_for_each_entry_safe(page
, tpage
, &tmplist
, lru
) {
3512 list_del(&page
->lru
);
3513 rdata
->pages
[rdata
->nr_pages
++] = page
;
3516 if (!rdata
->cfile
->invalidHandle
||
3517 !cifs_reopen_file(rdata
->cfile
, true))
3518 rc
= server
->ops
->async_readv(rdata
);
3520 add_credits_and_wake_if(server
, rdata
->credits
, 0);
3521 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
3522 page
= rdata
->pages
[i
];
3523 lru_cache_add_file(page
);
3527 /* Fallback to the readpage in error/reconnect cases */
3528 kref_put(&rdata
->refcount
, cifs_readdata_release
);
3532 kref_put(&rdata
->refcount
, cifs_readdata_release
);
3535 /* Any pages that have been shown to fscache but didn't get added to
3536 * the pagecache must be uncached before they get returned to the
3539 cifs_fscache_readpages_cancel(mapping
->host
, page_list
);
3544 * cifs_readpage_worker must be called with the page pinned
3546 static int cifs_readpage_worker(struct file
*file
, struct page
*page
,
3552 /* Is the page cached? */
3553 rc
= cifs_readpage_from_fscache(file_inode(file
), page
);
3557 read_data
= kmap(page
);
3558 /* for reads over a certain size could initiate async read ahead */
3560 rc
= cifs_read(file
, read_data
, PAGE_SIZE
, poffset
);
3565 cifs_dbg(FYI
, "Bytes read %d\n", rc
);
3567 file_inode(file
)->i_atime
=
3568 current_time(file_inode(file
));
3571 memset(read_data
+ rc
, 0, PAGE_SIZE
- rc
);
3573 flush_dcache_page(page
);
3574 SetPageUptodate(page
);
3576 /* send this page to the cache */
3577 cifs_readpage_to_fscache(file_inode(file
), page
);
3589 static int cifs_readpage(struct file
*file
, struct page
*page
)
3591 loff_t offset
= (loff_t
)page
->index
<< PAGE_SHIFT
;
3597 if (file
->private_data
== NULL
) {
3603 cifs_dbg(FYI
, "readpage %p at offset %d 0x%x\n",
3604 page
, (int)offset
, (int)offset
);
3606 rc
= cifs_readpage_worker(file
, page
, &offset
);
3612 static int is_inode_writable(struct cifsInodeInfo
*cifs_inode
)
3614 struct cifsFileInfo
*open_file
;
3616 spin_lock(&cifs_file_list_lock
);
3617 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
3618 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
3619 spin_unlock(&cifs_file_list_lock
);
3623 spin_unlock(&cifs_file_list_lock
);
3627 /* We do not want to update the file size from server for inodes
3628 open for write - to avoid races with writepage extending
3629 the file - in the future we could consider allowing
3630 refreshing the inode only on increases in the file size
3631 but this is tricky to do without racing with writebehind
3632 page caching in the current Linux kernel design */
3633 bool is_size_safe_to_change(struct cifsInodeInfo
*cifsInode
, __u64 end_of_file
)
3638 if (is_inode_writable(cifsInode
)) {
3639 /* This inode is open for write at least once */
3640 struct cifs_sb_info
*cifs_sb
;
3642 cifs_sb
= CIFS_SB(cifsInode
->vfs_inode
.i_sb
);
3643 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_DIRECT_IO
) {
3644 /* since no page cache to corrupt on directio
3645 we can change size safely */
3649 if (i_size_read(&cifsInode
->vfs_inode
) < end_of_file
)
3657 static int cifs_write_begin(struct file
*file
, struct address_space
*mapping
,
3658 loff_t pos
, unsigned len
, unsigned flags
,
3659 struct page
**pagep
, void **fsdata
)
3662 pgoff_t index
= pos
>> PAGE_SHIFT
;
3663 loff_t offset
= pos
& (PAGE_SIZE
- 1);
3664 loff_t page_start
= pos
& PAGE_MASK
;
3669 cifs_dbg(FYI
, "write_begin from %lld len %d\n", (long long)pos
, len
);
3672 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3678 if (PageUptodate(page
))
3682 * If we write a full page it will be up to date, no need to read from
3683 * the server. If the write is short, we'll end up doing a sync write
3686 if (len
== PAGE_SIZE
)
3690 * optimize away the read when we have an oplock, and we're not
3691 * expecting to use any of the data we'd be reading in. That
3692 * is, when the page lies beyond the EOF, or straddles the EOF
3693 * and the write will cover all of the existing data.
3695 if (CIFS_CACHE_READ(CIFS_I(mapping
->host
))) {
3696 i_size
= i_size_read(mapping
->host
);
3697 if (page_start
>= i_size
||
3698 (offset
== 0 && (pos
+ len
) >= i_size
)) {
3699 zero_user_segments(page
, 0, offset
,
3703 * PageChecked means that the parts of the page
3704 * to which we're not writing are considered up
3705 * to date. Once the data is copied to the
3706 * page, it can be set uptodate.
3708 SetPageChecked(page
);
3713 if ((file
->f_flags
& O_ACCMODE
) != O_WRONLY
&& !oncethru
) {
3715 * might as well read a page, it is fast enough. If we get
3716 * an error, we don't need to return it. cifs_write_end will
3717 * do a sync write instead since PG_uptodate isn't set.
3719 cifs_readpage_worker(file
, page
, &page_start
);
3724 /* we could try using another file handle if there is one -
3725 but how would we lock it to prevent close of that handle
3726 racing with this read? In any case
3727 this will be written out by write_end so is fine */
3734 static int cifs_release_page(struct page
*page
, gfp_t gfp
)
3736 if (PagePrivate(page
))
3739 return cifs_fscache_release_page(page
, gfp
);
3742 static void cifs_invalidate_page(struct page
*page
, unsigned int offset
,
3743 unsigned int length
)
3745 struct cifsInodeInfo
*cifsi
= CIFS_I(page
->mapping
->host
);
3747 if (offset
== 0 && length
== PAGE_SIZE
)
3748 cifs_fscache_invalidate_page(page
, &cifsi
->vfs_inode
);
3751 static int cifs_launder_page(struct page
*page
)
3754 loff_t range_start
= page_offset(page
);
3755 loff_t range_end
= range_start
+ (loff_t
)(PAGE_SIZE
- 1);
3756 struct writeback_control wbc
= {
3757 .sync_mode
= WB_SYNC_ALL
,
3759 .range_start
= range_start
,
3760 .range_end
= range_end
,
3763 cifs_dbg(FYI
, "Launder page: %p\n", page
);
3765 if (clear_page_dirty_for_io(page
))
3766 rc
= cifs_writepage_locked(page
, &wbc
);
3768 cifs_fscache_invalidate_page(page
, page
->mapping
->host
);
3772 void cifs_oplock_break(struct work_struct
*work
)
3774 struct cifsFileInfo
*cfile
= container_of(work
, struct cifsFileInfo
,
3776 struct inode
*inode
= d_inode(cfile
->dentry
);
3777 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
3778 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
3779 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
3782 wait_on_bit(&cinode
->flags
, CIFS_INODE_PENDING_WRITERS
,
3783 TASK_UNINTERRUPTIBLE
);
3785 server
->ops
->downgrade_oplock(server
, cinode
,
3786 test_bit(CIFS_INODE_DOWNGRADE_OPLOCK_TO_L2
, &cinode
->flags
));
3788 if (!CIFS_CACHE_WRITE(cinode
) && CIFS_CACHE_READ(cinode
) &&
3789 cifs_has_mand_locks(cinode
)) {
3790 cifs_dbg(FYI
, "Reset oplock to None for inode=%p due to mand locks\n",
3795 if (inode
&& S_ISREG(inode
->i_mode
)) {
3796 if (CIFS_CACHE_READ(cinode
))
3797 break_lease(inode
, O_RDONLY
);
3799 break_lease(inode
, O_WRONLY
);
3800 rc
= filemap_fdatawrite(inode
->i_mapping
);
3801 if (!CIFS_CACHE_READ(cinode
)) {
3802 rc
= filemap_fdatawait(inode
->i_mapping
);
3803 mapping_set_error(inode
->i_mapping
, rc
);
3804 cifs_zap_mapping(inode
);
3806 cifs_dbg(FYI
, "Oplock flush inode %p rc %d\n", inode
, rc
);
3809 rc
= cifs_push_locks(cfile
);
3811 cifs_dbg(VFS
, "Push locks rc = %d\n", rc
);
3814 * releasing stale oplock after recent reconnect of smb session using
3815 * a now incorrect file handle is not a data integrity issue but do
3816 * not bother sending an oplock release if session to server still is
3817 * disconnected since oplock already released by the server
3819 if (!cfile
->oplock_break_cancelled
) {
3820 rc
= tcon
->ses
->server
->ops
->oplock_response(tcon
, &cfile
->fid
,
3822 cifs_dbg(FYI
, "Oplock release rc = %d\n", rc
);
3824 cifs_done_oplock_break(cinode
);
3828 * The presence of cifs_direct_io() in the address space ops vector
3829 * allowes open() O_DIRECT flags which would have failed otherwise.
3831 * In the non-cached mode (mount with cache=none), we shunt off direct read and write requests
3832 * so this method should never be called.
3834 * Direct IO is not yet supported in the cached mode.
3837 cifs_direct_io(struct kiocb
*iocb
, struct iov_iter
*iter
)
3841 * Eventually need to support direct IO for non forcedirectio mounts
3847 const struct address_space_operations cifs_addr_ops
= {
3848 .readpage
= cifs_readpage
,
3849 .readpages
= cifs_readpages
,
3850 .writepage
= cifs_writepage
,
3851 .writepages
= cifs_writepages
,
3852 .write_begin
= cifs_write_begin
,
3853 .write_end
= cifs_write_end
,
3854 .set_page_dirty
= __set_page_dirty_nobuffers
,
3855 .releasepage
= cifs_release_page
,
3856 .direct_IO
= cifs_direct_io
,
3857 .invalidatepage
= cifs_invalidate_page
,
3858 .launder_page
= cifs_launder_page
,
3862 * cifs_readpages requires the server to support a buffer large enough to
3863 * contain the header plus one complete page of data. Otherwise, we need
3864 * to leave cifs_readpages out of the address space operations.
3866 const struct address_space_operations cifs_addr_ops_smallbuf
= {
3867 .readpage
= cifs_readpage
,
3868 .writepage
= cifs_writepage
,
3869 .writepages
= cifs_writepages
,
3870 .write_begin
= cifs_write_begin
,
3871 .write_end
= cifs_write_end
,
3872 .set_page_dirty
= __set_page_dirty_nobuffers
,
3873 .releasepage
= cifs_release_page
,
3874 .invalidatepage
= cifs_invalidate_page
,
3875 .launder_page
= cifs_launder_page
,