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>
37 #include <asm/div64.h>
41 #include "cifsproto.h"
42 #include "cifs_unicode.h"
43 #include "cifs_debug.h"
44 #include "cifs_fs_sb.h"
46 #include "smbdirect.h"
48 static inline int cifs_convert_flags(unsigned int flags
)
50 if ((flags
& O_ACCMODE
) == O_RDONLY
)
52 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
54 else if ((flags
& O_ACCMODE
) == O_RDWR
) {
55 /* GENERIC_ALL is too much permission to request
56 can cause unnecessary access denied on create */
57 /* return GENERIC_ALL; */
58 return (GENERIC_READ
| GENERIC_WRITE
);
61 return (READ_CONTROL
| FILE_WRITE_ATTRIBUTES
| FILE_READ_ATTRIBUTES
|
62 FILE_WRITE_EA
| FILE_APPEND_DATA
| FILE_WRITE_DATA
|
66 static u32
cifs_posix_convert_flags(unsigned int flags
)
70 if ((flags
& O_ACCMODE
) == O_RDONLY
)
71 posix_flags
= SMB_O_RDONLY
;
72 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
73 posix_flags
= SMB_O_WRONLY
;
74 else if ((flags
& O_ACCMODE
) == O_RDWR
)
75 posix_flags
= SMB_O_RDWR
;
77 if (flags
& O_CREAT
) {
78 posix_flags
|= SMB_O_CREAT
;
80 posix_flags
|= SMB_O_EXCL
;
81 } else if (flags
& O_EXCL
)
82 cifs_dbg(FYI
, "Application %s pid %d has incorrectly set O_EXCL flag but not O_CREAT on file open. Ignoring O_EXCL\n",
83 current
->comm
, current
->tgid
);
86 posix_flags
|= SMB_O_TRUNC
;
87 /* be safe and imply O_SYNC for O_DSYNC */
89 posix_flags
|= SMB_O_SYNC
;
90 if (flags
& O_DIRECTORY
)
91 posix_flags
|= SMB_O_DIRECTORY
;
92 if (flags
& O_NOFOLLOW
)
93 posix_flags
|= SMB_O_NOFOLLOW
;
95 posix_flags
|= SMB_O_DIRECT
;
100 static inline int cifs_get_disposition(unsigned int flags
)
102 if ((flags
& (O_CREAT
| O_EXCL
)) == (O_CREAT
| O_EXCL
))
104 else if ((flags
& (O_CREAT
| O_TRUNC
)) == (O_CREAT
| O_TRUNC
))
105 return FILE_OVERWRITE_IF
;
106 else if ((flags
& O_CREAT
) == O_CREAT
)
108 else if ((flags
& O_TRUNC
) == O_TRUNC
)
109 return FILE_OVERWRITE
;
114 int cifs_posix_open(char *full_path
, struct inode
**pinode
,
115 struct super_block
*sb
, int mode
, unsigned int f_flags
,
116 __u32
*poplock
, __u16
*pnetfid
, unsigned int xid
)
119 FILE_UNIX_BASIC_INFO
*presp_data
;
120 __u32 posix_flags
= 0;
121 struct cifs_sb_info
*cifs_sb
= CIFS_SB(sb
);
122 struct cifs_fattr fattr
;
123 struct tcon_link
*tlink
;
124 struct cifs_tcon
*tcon
;
126 cifs_dbg(FYI
, "posix open %s\n", full_path
);
128 presp_data
= kzalloc(sizeof(FILE_UNIX_BASIC_INFO
), GFP_KERNEL
);
129 if (presp_data
== NULL
)
132 tlink
= cifs_sb_tlink(cifs_sb
);
138 tcon
= tlink_tcon(tlink
);
139 mode
&= ~current_umask();
141 posix_flags
= cifs_posix_convert_flags(f_flags
);
142 rc
= CIFSPOSIXCreate(xid
, tcon
, posix_flags
, mode
, pnetfid
, presp_data
,
143 poplock
, full_path
, cifs_sb
->local_nls
,
144 cifs_remap(cifs_sb
));
145 cifs_put_tlink(tlink
);
150 if (presp_data
->Type
== cpu_to_le32(-1))
151 goto posix_open_ret
; /* open ok, caller does qpathinfo */
154 goto posix_open_ret
; /* caller does not need info */
156 cifs_unix_basic_to_fattr(&fattr
, presp_data
, cifs_sb
);
158 /* get new inode and set it up */
159 if (*pinode
== NULL
) {
160 cifs_fill_uniqueid(sb
, &fattr
);
161 *pinode
= cifs_iget(sb
, &fattr
);
167 cifs_fattr_to_inode(*pinode
, &fattr
);
176 cifs_nt_open(char *full_path
, struct inode
*inode
, struct cifs_sb_info
*cifs_sb
,
177 struct cifs_tcon
*tcon
, unsigned int f_flags
, __u32
*oplock
,
178 struct cifs_fid
*fid
, unsigned int xid
)
183 int create_options
= CREATE_NOT_DIR
;
185 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
186 struct cifs_open_parms oparms
;
188 if (!server
->ops
->open
)
191 desired_access
= cifs_convert_flags(f_flags
);
193 /*********************************************************************
194 * open flag mapping table:
196 * POSIX Flag CIFS Disposition
197 * ---------- ----------------
198 * O_CREAT FILE_OPEN_IF
199 * O_CREAT | O_EXCL FILE_CREATE
200 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
201 * O_TRUNC FILE_OVERWRITE
202 * none of the above FILE_OPEN
204 * Note that there is not a direct match between disposition
205 * FILE_SUPERSEDE (ie create whether or not file exists although
206 * O_CREAT | O_TRUNC is similar but truncates the existing
207 * file rather than creating a new file as FILE_SUPERSEDE does
208 * (which uses the attributes / metadata passed in on open call)
210 *? O_SYNC is a reasonable match to CIFS writethrough flag
211 *? and the read write flags match reasonably. O_LARGEFILE
212 *? is irrelevant because largefile support is always used
213 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
214 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
215 *********************************************************************/
217 disposition
= cifs_get_disposition(f_flags
);
219 /* BB pass O_SYNC flag through on file attributes .. BB */
221 buf
= kmalloc(sizeof(FILE_ALL_INFO
), GFP_KERNEL
);
225 if (backup_cred(cifs_sb
))
226 create_options
|= CREATE_OPEN_BACKUP_INTENT
;
228 /* O_SYNC also has bit for O_DSYNC so following check picks up either */
229 if (f_flags
& O_SYNC
)
230 create_options
|= CREATE_WRITE_THROUGH
;
232 if (f_flags
& O_DIRECT
)
233 create_options
|= CREATE_NO_BUFFER
;
236 oparms
.cifs_sb
= cifs_sb
;
237 oparms
.desired_access
= desired_access
;
238 oparms
.create_options
= create_options
;
239 oparms
.disposition
= disposition
;
240 oparms
.path
= full_path
;
242 oparms
.reconnect
= false;
244 rc
= server
->ops
->open(xid
, &oparms
, oplock
, buf
);
250 rc
= cifs_get_inode_info_unix(&inode
, full_path
, inode
->i_sb
,
253 rc
= cifs_get_inode_info(&inode
, full_path
, buf
, inode
->i_sb
,
257 server
->ops
->close(xid
, tcon
, fid
);
268 cifs_has_mand_locks(struct cifsInodeInfo
*cinode
)
270 struct cifs_fid_locks
*cur
;
271 bool has_locks
= false;
273 down_read(&cinode
->lock_sem
);
274 list_for_each_entry(cur
, &cinode
->llist
, llist
) {
275 if (!list_empty(&cur
->locks
)) {
280 up_read(&cinode
->lock_sem
);
284 struct cifsFileInfo
*
285 cifs_new_fileinfo(struct cifs_fid
*fid
, struct file
*file
,
286 struct tcon_link
*tlink
, __u32 oplock
)
288 struct dentry
*dentry
= file_dentry(file
);
289 struct inode
*inode
= d_inode(dentry
);
290 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
291 struct cifsFileInfo
*cfile
;
292 struct cifs_fid_locks
*fdlocks
;
293 struct cifs_tcon
*tcon
= tlink_tcon(tlink
);
294 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
296 cfile
= kzalloc(sizeof(struct cifsFileInfo
), GFP_KERNEL
);
300 fdlocks
= kzalloc(sizeof(struct cifs_fid_locks
), GFP_KERNEL
);
306 INIT_LIST_HEAD(&fdlocks
->locks
);
307 fdlocks
->cfile
= cfile
;
308 cfile
->llist
= fdlocks
;
309 down_write(&cinode
->lock_sem
);
310 list_add(&fdlocks
->llist
, &cinode
->llist
);
311 up_write(&cinode
->lock_sem
);
314 cfile
->pid
= current
->tgid
;
315 cfile
->uid
= current_fsuid();
316 cfile
->dentry
= dget(dentry
);
317 cfile
->f_flags
= file
->f_flags
;
318 cfile
->invalidHandle
= false;
319 cfile
->tlink
= cifs_get_tlink(tlink
);
320 INIT_WORK(&cfile
->oplock_break
, cifs_oplock_break
);
321 mutex_init(&cfile
->fh_mutex
);
322 spin_lock_init(&cfile
->file_info_lock
);
324 cifs_sb_active(inode
->i_sb
);
327 * If the server returned a read oplock and we have mandatory brlocks,
328 * set oplock level to None.
330 if (server
->ops
->is_read_op(oplock
) && cifs_has_mand_locks(cinode
)) {
331 cifs_dbg(FYI
, "Reset oplock val from read to None due to mand locks\n");
335 spin_lock(&tcon
->open_file_lock
);
336 if (fid
->pending_open
->oplock
!= CIFS_OPLOCK_NO_CHANGE
&& oplock
)
337 oplock
= fid
->pending_open
->oplock
;
338 list_del(&fid
->pending_open
->olist
);
340 fid
->purge_cache
= false;
341 server
->ops
->set_fid(cfile
, fid
, oplock
);
343 list_add(&cfile
->tlist
, &tcon
->openFileList
);
344 atomic_inc(&tcon
->num_local_opens
);
346 /* if readable file instance put first in list*/
347 spin_lock(&cinode
->open_file_lock
);
348 if (file
->f_mode
& FMODE_READ
)
349 list_add(&cfile
->flist
, &cinode
->openFileList
);
351 list_add_tail(&cfile
->flist
, &cinode
->openFileList
);
352 spin_unlock(&cinode
->open_file_lock
);
353 spin_unlock(&tcon
->open_file_lock
);
355 if (fid
->purge_cache
)
356 cifs_zap_mapping(inode
);
358 file
->private_data
= cfile
;
362 struct cifsFileInfo
*
363 cifsFileInfo_get(struct cifsFileInfo
*cifs_file
)
365 spin_lock(&cifs_file
->file_info_lock
);
366 cifsFileInfo_get_locked(cifs_file
);
367 spin_unlock(&cifs_file
->file_info_lock
);
372 * cifsFileInfo_put - release a reference of file priv data
374 * Always potentially wait for oplock handler. See _cifsFileInfo_put().
376 void cifsFileInfo_put(struct cifsFileInfo
*cifs_file
)
378 _cifsFileInfo_put(cifs_file
, true);
382 * _cifsFileInfo_put - release a reference of file priv data
384 * This may involve closing the filehandle @cifs_file out on the
385 * server. Must be called without holding tcon->open_file_lock and
386 * cifs_file->file_info_lock.
388 * If @wait_for_oplock_handler is true and we are releasing the last
389 * reference, wait for any running oplock break handler of the file
390 * and cancel any pending one. If calling this function from the
391 * oplock break handler, you need to pass false.
394 void _cifsFileInfo_put(struct cifsFileInfo
*cifs_file
, bool wait_oplock_handler
)
396 struct inode
*inode
= d_inode(cifs_file
->dentry
);
397 struct cifs_tcon
*tcon
= tlink_tcon(cifs_file
->tlink
);
398 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
399 struct cifsInodeInfo
*cifsi
= CIFS_I(inode
);
400 struct super_block
*sb
= inode
->i_sb
;
401 struct cifs_sb_info
*cifs_sb
= CIFS_SB(sb
);
402 struct cifsLockInfo
*li
, *tmp
;
404 struct cifs_pending_open open
;
405 bool oplock_break_cancelled
;
407 spin_lock(&tcon
->open_file_lock
);
409 spin_lock(&cifs_file
->file_info_lock
);
410 if (--cifs_file
->count
> 0) {
411 spin_unlock(&cifs_file
->file_info_lock
);
412 spin_unlock(&tcon
->open_file_lock
);
415 spin_unlock(&cifs_file
->file_info_lock
);
417 if (server
->ops
->get_lease_key
)
418 server
->ops
->get_lease_key(inode
, &fid
);
420 /* store open in pending opens to make sure we don't miss lease break */
421 cifs_add_pending_open_locked(&fid
, cifs_file
->tlink
, &open
);
423 /* remove it from the lists */
424 spin_lock(&cifsi
->open_file_lock
);
425 list_del(&cifs_file
->flist
);
426 spin_unlock(&cifsi
->open_file_lock
);
427 list_del(&cifs_file
->tlist
);
428 atomic_dec(&tcon
->num_local_opens
);
430 if (list_empty(&cifsi
->openFileList
)) {
431 cifs_dbg(FYI
, "closing last open instance for inode %p\n",
432 d_inode(cifs_file
->dentry
));
434 * In strict cache mode we need invalidate mapping on the last
435 * close because it may cause a error when we open this file
436 * again and get at least level II oplock.
438 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_STRICT_IO
)
439 set_bit(CIFS_INO_INVALID_MAPPING
, &cifsi
->flags
);
440 cifs_set_oplock_level(cifsi
, 0);
443 spin_unlock(&tcon
->open_file_lock
);
445 oplock_break_cancelled
= wait_oplock_handler
?
446 cancel_work_sync(&cifs_file
->oplock_break
) : false;
448 if (!tcon
->need_reconnect
&& !cifs_file
->invalidHandle
) {
449 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
453 if (server
->ops
->close
)
454 server
->ops
->close(xid
, tcon
, &cifs_file
->fid
);
458 if (oplock_break_cancelled
)
459 cifs_done_oplock_break(cifsi
);
461 cifs_del_pending_open(&open
);
464 * Delete any outstanding lock records. We'll lose them when the file
467 down_write(&cifsi
->lock_sem
);
468 list_for_each_entry_safe(li
, tmp
, &cifs_file
->llist
->locks
, llist
) {
469 list_del(&li
->llist
);
470 cifs_del_lock_waiters(li
);
473 list_del(&cifs_file
->llist
->llist
);
474 kfree(cifs_file
->llist
);
475 up_write(&cifsi
->lock_sem
);
477 cifs_put_tlink(cifs_file
->tlink
);
478 dput(cifs_file
->dentry
);
479 cifs_sb_deactive(sb
);
483 int cifs_open(struct inode
*inode
, struct file
*file
)
489 struct cifs_sb_info
*cifs_sb
;
490 struct TCP_Server_Info
*server
;
491 struct cifs_tcon
*tcon
;
492 struct tcon_link
*tlink
;
493 struct cifsFileInfo
*cfile
= NULL
;
494 char *full_path
= NULL
;
495 bool posix_open_ok
= false;
497 struct cifs_pending_open open
;
501 cifs_sb
= CIFS_SB(inode
->i_sb
);
502 tlink
= cifs_sb_tlink(cifs_sb
);
505 return PTR_ERR(tlink
);
507 tcon
= tlink_tcon(tlink
);
508 server
= tcon
->ses
->server
;
510 full_path
= build_path_from_dentry(file_dentry(file
));
511 if (full_path
== NULL
) {
516 cifs_dbg(FYI
, "inode = 0x%p file flags are 0x%x for %s\n",
517 inode
, file
->f_flags
, full_path
);
519 if (file
->f_flags
& O_DIRECT
&&
520 cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_STRICT_IO
) {
521 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NO_BRL
)
522 file
->f_op
= &cifs_file_direct_nobrl_ops
;
524 file
->f_op
= &cifs_file_direct_ops
;
532 if (!tcon
->broken_posix_open
&& tcon
->unix_ext
&&
533 cap_unix(tcon
->ses
) && (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
534 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
535 /* can not refresh inode info since size could be stale */
536 rc
= cifs_posix_open(full_path
, &inode
, inode
->i_sb
,
537 cifs_sb
->mnt_file_mode
/* ignored */,
538 file
->f_flags
, &oplock
, &fid
.netfid
, xid
);
540 cifs_dbg(FYI
, "posix open succeeded\n");
541 posix_open_ok
= true;
542 } else if ((rc
== -EINVAL
) || (rc
== -EOPNOTSUPP
)) {
543 if (tcon
->ses
->serverNOS
)
544 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",
545 tcon
->ses
->serverName
,
546 tcon
->ses
->serverNOS
);
547 tcon
->broken_posix_open
= true;
548 } else if ((rc
!= -EIO
) && (rc
!= -EREMOTE
) &&
549 (rc
!= -EOPNOTSUPP
)) /* path not found or net err */
552 * Else fallthrough to retry open the old way on network i/o
557 if (server
->ops
->get_lease_key
)
558 server
->ops
->get_lease_key(inode
, &fid
);
560 cifs_add_pending_open(&fid
, tlink
, &open
);
562 if (!posix_open_ok
) {
563 if (server
->ops
->get_lease_key
)
564 server
->ops
->get_lease_key(inode
, &fid
);
566 rc
= cifs_nt_open(full_path
, inode
, cifs_sb
, tcon
,
567 file
->f_flags
, &oplock
, &fid
, xid
);
569 cifs_del_pending_open(&open
);
574 cfile
= cifs_new_fileinfo(&fid
, file
, tlink
, oplock
);
576 if (server
->ops
->close
)
577 server
->ops
->close(xid
, tcon
, &fid
);
578 cifs_del_pending_open(&open
);
583 cifs_fscache_set_inode_cookie(inode
, file
);
585 if ((oplock
& CIFS_CREATE_ACTION
) && !posix_open_ok
&& tcon
->unix_ext
) {
587 * Time to set mode which we can not set earlier due to
588 * problems creating new read-only files.
590 struct cifs_unix_set_info_args args
= {
591 .mode
= inode
->i_mode
,
592 .uid
= INVALID_UID
, /* no change */
593 .gid
= INVALID_GID
, /* no change */
594 .ctime
= NO_CHANGE_64
,
595 .atime
= NO_CHANGE_64
,
596 .mtime
= NO_CHANGE_64
,
599 CIFSSMBUnixSetFileInfo(xid
, tcon
, &args
, fid
.netfid
,
606 cifs_put_tlink(tlink
);
610 static int cifs_push_posix_locks(struct cifsFileInfo
*cfile
);
613 * Try to reacquire byte range locks that were released when session
614 * to server was lost.
617 cifs_relock_file(struct cifsFileInfo
*cfile
)
619 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cfile
->dentry
->d_sb
);
620 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
621 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
624 down_read_nested(&cinode
->lock_sem
, SINGLE_DEPTH_NESTING
);
625 if (cinode
->can_cache_brlcks
) {
626 /* can cache locks - no need to relock */
627 up_read(&cinode
->lock_sem
);
631 if (cap_unix(tcon
->ses
) &&
632 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
633 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
634 rc
= cifs_push_posix_locks(cfile
);
636 rc
= tcon
->ses
->server
->ops
->push_mand_locks(cfile
);
638 up_read(&cinode
->lock_sem
);
643 cifs_reopen_file(struct cifsFileInfo
*cfile
, bool can_flush
)
648 struct cifs_sb_info
*cifs_sb
;
649 struct cifs_tcon
*tcon
;
650 struct TCP_Server_Info
*server
;
651 struct cifsInodeInfo
*cinode
;
653 char *full_path
= NULL
;
655 int disposition
= FILE_OPEN
;
656 int create_options
= CREATE_NOT_DIR
;
657 struct cifs_open_parms oparms
;
660 mutex_lock(&cfile
->fh_mutex
);
661 if (!cfile
->invalidHandle
) {
662 mutex_unlock(&cfile
->fh_mutex
);
668 inode
= d_inode(cfile
->dentry
);
669 cifs_sb
= CIFS_SB(inode
->i_sb
);
670 tcon
= tlink_tcon(cfile
->tlink
);
671 server
= tcon
->ses
->server
;
674 * Can not grab rename sem here because various ops, including those
675 * that already have the rename sem can end up causing writepage to get
676 * called and if the server was down that means we end up here, and we
677 * can never tell if the caller already has the rename_sem.
679 full_path
= build_path_from_dentry(cfile
->dentry
);
680 if (full_path
== NULL
) {
682 mutex_unlock(&cfile
->fh_mutex
);
687 cifs_dbg(FYI
, "inode = 0x%p file flags 0x%x for %s\n",
688 inode
, cfile
->f_flags
, full_path
);
690 if (tcon
->ses
->server
->oplocks
)
695 if (tcon
->unix_ext
&& cap_unix(tcon
->ses
) &&
696 (CIFS_UNIX_POSIX_PATH_OPS_CAP
&
697 le64_to_cpu(tcon
->fsUnixInfo
.Capability
))) {
699 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
700 * original open. Must mask them off for a reopen.
702 unsigned int oflags
= cfile
->f_flags
&
703 ~(O_CREAT
| O_EXCL
| O_TRUNC
);
705 rc
= cifs_posix_open(full_path
, NULL
, inode
->i_sb
,
706 cifs_sb
->mnt_file_mode
/* ignored */,
707 oflags
, &oplock
, &cfile
->fid
.netfid
, xid
);
709 cifs_dbg(FYI
, "posix reopen succeeded\n");
710 oparms
.reconnect
= true;
714 * fallthrough to retry open the old way on errors, especially
715 * in the reconnect path it is important to retry hard
719 desired_access
= cifs_convert_flags(cfile
->f_flags
);
721 if (backup_cred(cifs_sb
))
722 create_options
|= CREATE_OPEN_BACKUP_INTENT
;
724 if (server
->ops
->get_lease_key
)
725 server
->ops
->get_lease_key(inode
, &cfile
->fid
);
728 oparms
.cifs_sb
= cifs_sb
;
729 oparms
.desired_access
= desired_access
;
730 oparms
.create_options
= create_options
;
731 oparms
.disposition
= disposition
;
732 oparms
.path
= full_path
;
733 oparms
.fid
= &cfile
->fid
;
734 oparms
.reconnect
= true;
737 * Can not refresh inode by passing in file_info buf to be returned by
738 * ops->open and then calling get_inode_info with returned buf since
739 * file might have write behind data that needs to be flushed and server
740 * version of file size can be stale. If we knew for sure that inode was
741 * not dirty locally we could do this.
743 rc
= server
->ops
->open(xid
, &oparms
, &oplock
, NULL
);
744 if (rc
== -ENOENT
&& oparms
.reconnect
== false) {
745 /* durable handle timeout is expired - open the file again */
746 rc
= server
->ops
->open(xid
, &oparms
, &oplock
, NULL
);
747 /* indicate that we need to relock the file */
748 oparms
.reconnect
= true;
752 mutex_unlock(&cfile
->fh_mutex
);
753 cifs_dbg(FYI
, "cifs_reopen returned 0x%x\n", rc
);
754 cifs_dbg(FYI
, "oplock: %d\n", oplock
);
755 goto reopen_error_exit
;
759 cfile
->invalidHandle
= false;
760 mutex_unlock(&cfile
->fh_mutex
);
761 cinode
= CIFS_I(inode
);
764 rc
= filemap_write_and_wait(inode
->i_mapping
);
765 if (!is_interrupt_error(rc
))
766 mapping_set_error(inode
->i_mapping
, rc
);
769 rc
= cifs_get_inode_info_unix(&inode
, full_path
,
772 rc
= cifs_get_inode_info(&inode
, full_path
, NULL
,
773 inode
->i_sb
, xid
, NULL
);
776 * Else we are writing out data to server already and could deadlock if
777 * we tried to flush data, and since we do not know if we have data that
778 * would invalidate the current end of file on the server we can not go
779 * to the server to get the new inode info.
783 * If the server returned a read oplock and we have mandatory brlocks,
784 * set oplock level to None.
786 if (server
->ops
->is_read_op(oplock
) && cifs_has_mand_locks(cinode
)) {
787 cifs_dbg(FYI
, "Reset oplock val from read to None due to mand locks\n");
791 server
->ops
->set_fid(cfile
, &cfile
->fid
, oplock
);
792 if (oparms
.reconnect
)
793 cifs_relock_file(cfile
);
801 int cifs_close(struct inode
*inode
, struct file
*file
)
803 if (file
->private_data
!= NULL
) {
804 cifsFileInfo_put(file
->private_data
);
805 file
->private_data
= NULL
;
808 /* return code from the ->release op is always ignored */
813 cifs_reopen_persistent_handles(struct cifs_tcon
*tcon
)
815 struct cifsFileInfo
*open_file
;
816 struct list_head
*tmp
;
817 struct list_head
*tmp1
;
818 struct list_head tmp_list
;
820 if (!tcon
->use_persistent
|| !tcon
->need_reopen_files
)
823 tcon
->need_reopen_files
= false;
825 cifs_dbg(FYI
, "Reopen persistent handles");
826 INIT_LIST_HEAD(&tmp_list
);
828 /* list all files open on tree connection, reopen resilient handles */
829 spin_lock(&tcon
->open_file_lock
);
830 list_for_each(tmp
, &tcon
->openFileList
) {
831 open_file
= list_entry(tmp
, struct cifsFileInfo
, tlist
);
832 if (!open_file
->invalidHandle
)
834 cifsFileInfo_get(open_file
);
835 list_add_tail(&open_file
->rlist
, &tmp_list
);
837 spin_unlock(&tcon
->open_file_lock
);
839 list_for_each_safe(tmp
, tmp1
, &tmp_list
) {
840 open_file
= list_entry(tmp
, struct cifsFileInfo
, rlist
);
841 if (cifs_reopen_file(open_file
, false /* do not flush */))
842 tcon
->need_reopen_files
= true;
843 list_del_init(&open_file
->rlist
);
844 cifsFileInfo_put(open_file
);
848 int cifs_closedir(struct inode
*inode
, struct file
*file
)
852 struct cifsFileInfo
*cfile
= file
->private_data
;
853 struct cifs_tcon
*tcon
;
854 struct TCP_Server_Info
*server
;
857 cifs_dbg(FYI
, "Closedir inode = 0x%p\n", inode
);
863 tcon
= tlink_tcon(cfile
->tlink
);
864 server
= tcon
->ses
->server
;
866 cifs_dbg(FYI
, "Freeing private data in close dir\n");
867 spin_lock(&cfile
->file_info_lock
);
868 if (server
->ops
->dir_needs_close(cfile
)) {
869 cfile
->invalidHandle
= true;
870 spin_unlock(&cfile
->file_info_lock
);
871 if (server
->ops
->close_dir
)
872 rc
= server
->ops
->close_dir(xid
, tcon
, &cfile
->fid
);
875 cifs_dbg(FYI
, "Closing uncompleted readdir with rc %d\n", rc
);
876 /* not much we can do if it fails anyway, ignore rc */
879 spin_unlock(&cfile
->file_info_lock
);
881 buf
= cfile
->srch_inf
.ntwrk_buf_start
;
883 cifs_dbg(FYI
, "closedir free smb buf in srch struct\n");
884 cfile
->srch_inf
.ntwrk_buf_start
= NULL
;
885 if (cfile
->srch_inf
.smallBuf
)
886 cifs_small_buf_release(buf
);
888 cifs_buf_release(buf
);
891 cifs_put_tlink(cfile
->tlink
);
892 kfree(file
->private_data
);
893 file
->private_data
= NULL
;
894 /* BB can we lock the filestruct while this is going on? */
899 static struct cifsLockInfo
*
900 cifs_lock_init(__u64 offset
, __u64 length
, __u8 type
, __u16 flags
)
902 struct cifsLockInfo
*lock
=
903 kmalloc(sizeof(struct cifsLockInfo
), GFP_KERNEL
);
906 lock
->offset
= offset
;
907 lock
->length
= length
;
909 lock
->pid
= current
->tgid
;
911 INIT_LIST_HEAD(&lock
->blist
);
912 init_waitqueue_head(&lock
->block_q
);
917 cifs_del_lock_waiters(struct cifsLockInfo
*lock
)
919 struct cifsLockInfo
*li
, *tmp
;
920 list_for_each_entry_safe(li
, tmp
, &lock
->blist
, blist
) {
921 list_del_init(&li
->blist
);
922 wake_up(&li
->block_q
);
926 #define CIFS_LOCK_OP 0
927 #define CIFS_READ_OP 1
928 #define CIFS_WRITE_OP 2
930 /* @rw_check : 0 - no op, 1 - read, 2 - write */
932 cifs_find_fid_lock_conflict(struct cifs_fid_locks
*fdlocks
, __u64 offset
,
933 __u64 length
, __u8 type
, __u16 flags
,
934 struct cifsFileInfo
*cfile
,
935 struct cifsLockInfo
**conf_lock
, int rw_check
)
937 struct cifsLockInfo
*li
;
938 struct cifsFileInfo
*cur_cfile
= fdlocks
->cfile
;
939 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
941 list_for_each_entry(li
, &fdlocks
->locks
, llist
) {
942 if (offset
+ length
<= li
->offset
||
943 offset
>= li
->offset
+ li
->length
)
945 if (rw_check
!= CIFS_LOCK_OP
&& current
->tgid
== li
->pid
&&
946 server
->ops
->compare_fids(cfile
, cur_cfile
)) {
947 /* shared lock prevents write op through the same fid */
948 if (!(li
->type
& server
->vals
->shared_lock_type
) ||
949 rw_check
!= CIFS_WRITE_OP
)
952 if ((type
& server
->vals
->shared_lock_type
) &&
953 ((server
->ops
->compare_fids(cfile
, cur_cfile
) &&
954 current
->tgid
== li
->pid
) || type
== li
->type
))
956 if (rw_check
== CIFS_LOCK_OP
&&
957 (flags
& FL_OFDLCK
) && (li
->flags
& FL_OFDLCK
) &&
958 server
->ops
->compare_fids(cfile
, cur_cfile
))
968 cifs_find_lock_conflict(struct cifsFileInfo
*cfile
, __u64 offset
, __u64 length
,
969 __u8 type
, __u16 flags
,
970 struct cifsLockInfo
**conf_lock
, int rw_check
)
973 struct cifs_fid_locks
*cur
;
974 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
976 list_for_each_entry(cur
, &cinode
->llist
, llist
) {
977 rc
= cifs_find_fid_lock_conflict(cur
, offset
, length
, type
,
978 flags
, cfile
, conf_lock
,
988 * Check if there is another lock that prevents us to set the lock (mandatory
989 * style). If such a lock exists, update the flock structure with its
990 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
991 * or leave it the same if we can't. Returns 0 if we don't need to request to
992 * the server or 1 otherwise.
995 cifs_lock_test(struct cifsFileInfo
*cfile
, __u64 offset
, __u64 length
,
996 __u8 type
, struct file_lock
*flock
)
999 struct cifsLockInfo
*conf_lock
;
1000 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
1001 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
1004 down_read(&cinode
->lock_sem
);
1006 exist
= cifs_find_lock_conflict(cfile
, offset
, length
, type
,
1007 flock
->fl_flags
, &conf_lock
,
1010 flock
->fl_start
= conf_lock
->offset
;
1011 flock
->fl_end
= conf_lock
->offset
+ conf_lock
->length
- 1;
1012 flock
->fl_pid
= conf_lock
->pid
;
1013 if (conf_lock
->type
& server
->vals
->shared_lock_type
)
1014 flock
->fl_type
= F_RDLCK
;
1016 flock
->fl_type
= F_WRLCK
;
1017 } else if (!cinode
->can_cache_brlcks
)
1020 flock
->fl_type
= F_UNLCK
;
1022 up_read(&cinode
->lock_sem
);
1027 cifs_lock_add(struct cifsFileInfo
*cfile
, struct cifsLockInfo
*lock
)
1029 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
1030 down_write(&cinode
->lock_sem
);
1031 list_add_tail(&lock
->llist
, &cfile
->llist
->locks
);
1032 up_write(&cinode
->lock_sem
);
1036 * Set the byte-range lock (mandatory style). Returns:
1037 * 1) 0, if we set the lock and don't need to request to the server;
1038 * 2) 1, if no locks prevent us but we need to request to the server;
1039 * 3) -EACCES, if there is a lock that prevents us and wait is false.
1042 cifs_lock_add_if(struct cifsFileInfo
*cfile
, struct cifsLockInfo
*lock
,
1045 struct cifsLockInfo
*conf_lock
;
1046 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
1052 down_write(&cinode
->lock_sem
);
1054 exist
= cifs_find_lock_conflict(cfile
, lock
->offset
, lock
->length
,
1055 lock
->type
, lock
->flags
, &conf_lock
,
1057 if (!exist
&& cinode
->can_cache_brlcks
) {
1058 list_add_tail(&lock
->llist
, &cfile
->llist
->locks
);
1059 up_write(&cinode
->lock_sem
);
1068 list_add_tail(&lock
->blist
, &conf_lock
->blist
);
1069 up_write(&cinode
->lock_sem
);
1070 rc
= wait_event_interruptible(lock
->block_q
,
1071 (lock
->blist
.prev
== &lock
->blist
) &&
1072 (lock
->blist
.next
== &lock
->blist
));
1075 down_write(&cinode
->lock_sem
);
1076 list_del_init(&lock
->blist
);
1079 up_write(&cinode
->lock_sem
);
1084 * Check if there is another lock that prevents us to set the lock (posix
1085 * style). If such a lock exists, update the flock structure with its
1086 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
1087 * or leave it the same if we can't. Returns 0 if we don't need to request to
1088 * the server or 1 otherwise.
1091 cifs_posix_lock_test(struct file
*file
, struct file_lock
*flock
)
1094 struct cifsInodeInfo
*cinode
= CIFS_I(file_inode(file
));
1095 unsigned char saved_type
= flock
->fl_type
;
1097 if ((flock
->fl_flags
& FL_POSIX
) == 0)
1100 down_read(&cinode
->lock_sem
);
1101 posix_test_lock(file
, flock
);
1103 if (flock
->fl_type
== F_UNLCK
&& !cinode
->can_cache_brlcks
) {
1104 flock
->fl_type
= saved_type
;
1108 up_read(&cinode
->lock_sem
);
1113 * Set the byte-range lock (posix style). Returns:
1114 * 1) 0, if we set the lock and don't need to request to the server;
1115 * 2) 1, if we need to request to the server;
1116 * 3) <0, if the error occurs while setting the lock.
1119 cifs_posix_lock_set(struct file
*file
, struct file_lock
*flock
)
1121 struct cifsInodeInfo
*cinode
= CIFS_I(file_inode(file
));
1124 if ((flock
->fl_flags
& FL_POSIX
) == 0)
1128 down_write(&cinode
->lock_sem
);
1129 if (!cinode
->can_cache_brlcks
) {
1130 up_write(&cinode
->lock_sem
);
1134 rc
= posix_lock_file(file
, flock
, NULL
);
1135 up_write(&cinode
->lock_sem
);
1136 if (rc
== FILE_LOCK_DEFERRED
) {
1137 rc
= wait_event_interruptible(flock
->fl_wait
, !flock
->fl_blocker
);
1140 locks_delete_block(flock
);
1146 cifs_push_mandatory_locks(struct cifsFileInfo
*cfile
)
1149 int rc
= 0, stored_rc
;
1150 struct cifsLockInfo
*li
, *tmp
;
1151 struct cifs_tcon
*tcon
;
1152 unsigned int num
, max_num
, max_buf
;
1153 LOCKING_ANDX_RANGE
*buf
, *cur
;
1154 static const int types
[] = {
1155 LOCKING_ANDX_LARGE_FILES
,
1156 LOCKING_ANDX_SHARED_LOCK
| LOCKING_ANDX_LARGE_FILES
1161 tcon
= tlink_tcon(cfile
->tlink
);
1164 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1165 * and check it before using.
1167 max_buf
= tcon
->ses
->server
->maxBuf
;
1168 if (max_buf
< (sizeof(struct smb_hdr
) + sizeof(LOCKING_ANDX_RANGE
))) {
1173 BUILD_BUG_ON(sizeof(struct smb_hdr
) + sizeof(LOCKING_ANDX_RANGE
) >
1175 max_buf
= min_t(unsigned int, max_buf
- sizeof(struct smb_hdr
),
1177 max_num
= (max_buf
- sizeof(struct smb_hdr
)) /
1178 sizeof(LOCKING_ANDX_RANGE
);
1179 buf
= kcalloc(max_num
, sizeof(LOCKING_ANDX_RANGE
), GFP_KERNEL
);
1185 for (i
= 0; i
< 2; i
++) {
1188 list_for_each_entry_safe(li
, tmp
, &cfile
->llist
->locks
, llist
) {
1189 if (li
->type
!= types
[i
])
1191 cur
->Pid
= cpu_to_le16(li
->pid
);
1192 cur
->LengthLow
= cpu_to_le32((u32
)li
->length
);
1193 cur
->LengthHigh
= cpu_to_le32((u32
)(li
->length
>>32));
1194 cur
->OffsetLow
= cpu_to_le32((u32
)li
->offset
);
1195 cur
->OffsetHigh
= cpu_to_le32((u32
)(li
->offset
>>32));
1196 if (++num
== max_num
) {
1197 stored_rc
= cifs_lockv(xid
, tcon
,
1199 (__u8
)li
->type
, 0, num
,
1210 stored_rc
= cifs_lockv(xid
, tcon
, cfile
->fid
.netfid
,
1211 (__u8
)types
[i
], 0, num
, buf
);
1223 hash_lockowner(fl_owner_t owner
)
1225 return cifs_lock_secret
^ hash32_ptr((const void *)owner
);
1228 struct lock_to_push
{
1229 struct list_head llist
;
1238 cifs_push_posix_locks(struct cifsFileInfo
*cfile
)
1240 struct inode
*inode
= d_inode(cfile
->dentry
);
1241 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1242 struct file_lock
*flock
;
1243 struct file_lock_context
*flctx
= inode
->i_flctx
;
1244 unsigned int count
= 0, i
;
1245 int rc
= 0, xid
, type
;
1246 struct list_head locks_to_send
, *el
;
1247 struct lock_to_push
*lck
, *tmp
;
1255 spin_lock(&flctx
->flc_lock
);
1256 list_for_each(el
, &flctx
->flc_posix
) {
1259 spin_unlock(&flctx
->flc_lock
);
1261 INIT_LIST_HEAD(&locks_to_send
);
1264 * Allocating count locks is enough because no FL_POSIX locks can be
1265 * added to the list while we are holding cinode->lock_sem that
1266 * protects locking operations of this inode.
1268 for (i
= 0; i
< count
; i
++) {
1269 lck
= kmalloc(sizeof(struct lock_to_push
), GFP_KERNEL
);
1274 list_add_tail(&lck
->llist
, &locks_to_send
);
1277 el
= locks_to_send
.next
;
1278 spin_lock(&flctx
->flc_lock
);
1279 list_for_each_entry(flock
, &flctx
->flc_posix
, fl_list
) {
1280 if (el
== &locks_to_send
) {
1282 * The list ended. We don't have enough allocated
1283 * structures - something is really wrong.
1285 cifs_dbg(VFS
, "Can't push all brlocks!\n");
1288 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1289 if (flock
->fl_type
== F_RDLCK
|| flock
->fl_type
== F_SHLCK
)
1293 lck
= list_entry(el
, struct lock_to_push
, llist
);
1294 lck
->pid
= hash_lockowner(flock
->fl_owner
);
1295 lck
->netfid
= cfile
->fid
.netfid
;
1296 lck
->length
= length
;
1298 lck
->offset
= flock
->fl_start
;
1300 spin_unlock(&flctx
->flc_lock
);
1302 list_for_each_entry_safe(lck
, tmp
, &locks_to_send
, llist
) {
1305 stored_rc
= CIFSSMBPosixLock(xid
, tcon
, lck
->netfid
, lck
->pid
,
1306 lck
->offset
, lck
->length
, NULL
,
1310 list_del(&lck
->llist
);
1318 list_for_each_entry_safe(lck
, tmp
, &locks_to_send
, llist
) {
1319 list_del(&lck
->llist
);
1326 cifs_push_locks(struct cifsFileInfo
*cfile
)
1328 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cfile
->dentry
->d_sb
);
1329 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
1330 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1333 /* we are going to update can_cache_brlcks here - need a write access */
1334 down_write(&cinode
->lock_sem
);
1335 if (!cinode
->can_cache_brlcks
) {
1336 up_write(&cinode
->lock_sem
);
1340 if (cap_unix(tcon
->ses
) &&
1341 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
1342 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
1343 rc
= cifs_push_posix_locks(cfile
);
1345 rc
= tcon
->ses
->server
->ops
->push_mand_locks(cfile
);
1347 cinode
->can_cache_brlcks
= false;
1348 up_write(&cinode
->lock_sem
);
1353 cifs_read_flock(struct file_lock
*flock
, __u32
*type
, int *lock
, int *unlock
,
1354 bool *wait_flag
, struct TCP_Server_Info
*server
)
1356 if (flock
->fl_flags
& FL_POSIX
)
1357 cifs_dbg(FYI
, "Posix\n");
1358 if (flock
->fl_flags
& FL_FLOCK
)
1359 cifs_dbg(FYI
, "Flock\n");
1360 if (flock
->fl_flags
& FL_SLEEP
) {
1361 cifs_dbg(FYI
, "Blocking lock\n");
1364 if (flock
->fl_flags
& FL_ACCESS
)
1365 cifs_dbg(FYI
, "Process suspended by mandatory locking - not implemented yet\n");
1366 if (flock
->fl_flags
& FL_LEASE
)
1367 cifs_dbg(FYI
, "Lease on file - not implemented yet\n");
1368 if (flock
->fl_flags
&
1369 (~(FL_POSIX
| FL_FLOCK
| FL_SLEEP
|
1370 FL_ACCESS
| FL_LEASE
| FL_CLOSE
| FL_OFDLCK
)))
1371 cifs_dbg(FYI
, "Unknown lock flags 0x%x\n", flock
->fl_flags
);
1373 *type
= server
->vals
->large_lock_type
;
1374 if (flock
->fl_type
== F_WRLCK
) {
1375 cifs_dbg(FYI
, "F_WRLCK\n");
1376 *type
|= server
->vals
->exclusive_lock_type
;
1378 } else if (flock
->fl_type
== F_UNLCK
) {
1379 cifs_dbg(FYI
, "F_UNLCK\n");
1380 *type
|= server
->vals
->unlock_lock_type
;
1382 /* Check if unlock includes more than one lock range */
1383 } else if (flock
->fl_type
== F_RDLCK
) {
1384 cifs_dbg(FYI
, "F_RDLCK\n");
1385 *type
|= server
->vals
->shared_lock_type
;
1387 } else if (flock
->fl_type
== F_EXLCK
) {
1388 cifs_dbg(FYI
, "F_EXLCK\n");
1389 *type
|= server
->vals
->exclusive_lock_type
;
1391 } else if (flock
->fl_type
== F_SHLCK
) {
1392 cifs_dbg(FYI
, "F_SHLCK\n");
1393 *type
|= server
->vals
->shared_lock_type
;
1396 cifs_dbg(FYI
, "Unknown type of lock\n");
1400 cifs_getlk(struct file
*file
, struct file_lock
*flock
, __u32 type
,
1401 bool wait_flag
, bool posix_lck
, unsigned int xid
)
1404 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1405 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
1406 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1407 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
1408 __u16 netfid
= cfile
->fid
.netfid
;
1411 int posix_lock_type
;
1413 rc
= cifs_posix_lock_test(file
, flock
);
1417 if (type
& server
->vals
->shared_lock_type
)
1418 posix_lock_type
= CIFS_RDLCK
;
1420 posix_lock_type
= CIFS_WRLCK
;
1421 rc
= CIFSSMBPosixLock(xid
, tcon
, netfid
,
1422 hash_lockowner(flock
->fl_owner
),
1423 flock
->fl_start
, length
, flock
,
1424 posix_lock_type
, wait_flag
);
1428 rc
= cifs_lock_test(cfile
, flock
->fl_start
, length
, type
, flock
);
1432 /* BB we could chain these into one lock request BB */
1433 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
, type
,
1436 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1438 flock
->fl_type
= F_UNLCK
;
1440 cifs_dbg(VFS
, "Error unlocking previously locked range %d during test of lock\n",
1445 if (type
& server
->vals
->shared_lock_type
) {
1446 flock
->fl_type
= F_WRLCK
;
1450 type
&= ~server
->vals
->exclusive_lock_type
;
1452 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1453 type
| server
->vals
->shared_lock_type
,
1456 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1457 type
| server
->vals
->shared_lock_type
, 0, 1, false);
1458 flock
->fl_type
= F_RDLCK
;
1460 cifs_dbg(VFS
, "Error unlocking previously locked range %d during test of lock\n",
1463 flock
->fl_type
= F_WRLCK
;
1469 cifs_move_llist(struct list_head
*source
, struct list_head
*dest
)
1471 struct list_head
*li
, *tmp
;
1472 list_for_each_safe(li
, tmp
, source
)
1473 list_move(li
, dest
);
1477 cifs_free_llist(struct list_head
*llist
)
1479 struct cifsLockInfo
*li
, *tmp
;
1480 list_for_each_entry_safe(li
, tmp
, llist
, llist
) {
1481 cifs_del_lock_waiters(li
);
1482 list_del(&li
->llist
);
1488 cifs_unlock_range(struct cifsFileInfo
*cfile
, struct file_lock
*flock
,
1491 int rc
= 0, stored_rc
;
1492 static const int types
[] = {
1493 LOCKING_ANDX_LARGE_FILES
,
1494 LOCKING_ANDX_SHARED_LOCK
| LOCKING_ANDX_LARGE_FILES
1497 unsigned int max_num
, num
, max_buf
;
1498 LOCKING_ANDX_RANGE
*buf
, *cur
;
1499 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1500 struct cifsInodeInfo
*cinode
= CIFS_I(d_inode(cfile
->dentry
));
1501 struct cifsLockInfo
*li
, *tmp
;
1502 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1503 struct list_head tmp_llist
;
1505 INIT_LIST_HEAD(&tmp_llist
);
1508 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1509 * and check it before using.
1511 max_buf
= tcon
->ses
->server
->maxBuf
;
1512 if (max_buf
< (sizeof(struct smb_hdr
) + sizeof(LOCKING_ANDX_RANGE
)))
1515 BUILD_BUG_ON(sizeof(struct smb_hdr
) + sizeof(LOCKING_ANDX_RANGE
) >
1517 max_buf
= min_t(unsigned int, max_buf
- sizeof(struct smb_hdr
),
1519 max_num
= (max_buf
- sizeof(struct smb_hdr
)) /
1520 sizeof(LOCKING_ANDX_RANGE
);
1521 buf
= kcalloc(max_num
, sizeof(LOCKING_ANDX_RANGE
), GFP_KERNEL
);
1525 down_write(&cinode
->lock_sem
);
1526 for (i
= 0; i
< 2; i
++) {
1529 list_for_each_entry_safe(li
, tmp
, &cfile
->llist
->locks
, llist
) {
1530 if (flock
->fl_start
> li
->offset
||
1531 (flock
->fl_start
+ length
) <
1532 (li
->offset
+ li
->length
))
1534 if (current
->tgid
!= li
->pid
)
1536 if (types
[i
] != li
->type
)
1538 if (cinode
->can_cache_brlcks
) {
1540 * We can cache brlock requests - simply remove
1541 * a lock from the file's list.
1543 list_del(&li
->llist
);
1544 cifs_del_lock_waiters(li
);
1548 cur
->Pid
= cpu_to_le16(li
->pid
);
1549 cur
->LengthLow
= cpu_to_le32((u32
)li
->length
);
1550 cur
->LengthHigh
= cpu_to_le32((u32
)(li
->length
>>32));
1551 cur
->OffsetLow
= cpu_to_le32((u32
)li
->offset
);
1552 cur
->OffsetHigh
= cpu_to_le32((u32
)(li
->offset
>>32));
1554 * We need to save a lock here to let us add it again to
1555 * the file's list if the unlock range request fails on
1558 list_move(&li
->llist
, &tmp_llist
);
1559 if (++num
== max_num
) {
1560 stored_rc
= cifs_lockv(xid
, tcon
,
1562 li
->type
, num
, 0, buf
);
1565 * We failed on the unlock range
1566 * request - add all locks from the tmp
1567 * list to the head of the file's list.
1569 cifs_move_llist(&tmp_llist
,
1570 &cfile
->llist
->locks
);
1574 * The unlock range request succeed -
1575 * free the tmp list.
1577 cifs_free_llist(&tmp_llist
);
1584 stored_rc
= cifs_lockv(xid
, tcon
, cfile
->fid
.netfid
,
1585 types
[i
], num
, 0, buf
);
1587 cifs_move_llist(&tmp_llist
,
1588 &cfile
->llist
->locks
);
1591 cifs_free_llist(&tmp_llist
);
1595 up_write(&cinode
->lock_sem
);
1601 cifs_setlk(struct file
*file
, struct file_lock
*flock
, __u32 type
,
1602 bool wait_flag
, bool posix_lck
, int lock
, int unlock
,
1606 __u64 length
= 1 + flock
->fl_end
- flock
->fl_start
;
1607 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
1608 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
1609 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
1610 struct inode
*inode
= d_inode(cfile
->dentry
);
1613 int posix_lock_type
;
1615 rc
= cifs_posix_lock_set(file
, flock
);
1619 if (type
& server
->vals
->shared_lock_type
)
1620 posix_lock_type
= CIFS_RDLCK
;
1622 posix_lock_type
= CIFS_WRLCK
;
1625 posix_lock_type
= CIFS_UNLCK
;
1627 rc
= CIFSSMBPosixLock(xid
, tcon
, cfile
->fid
.netfid
,
1628 hash_lockowner(flock
->fl_owner
),
1629 flock
->fl_start
, length
,
1630 NULL
, posix_lock_type
, wait_flag
);
1635 struct cifsLockInfo
*lock
;
1637 lock
= cifs_lock_init(flock
->fl_start
, length
, type
,
1642 rc
= cifs_lock_add_if(cfile
, lock
, wait_flag
);
1651 * Windows 7 server can delay breaking lease from read to None
1652 * if we set a byte-range lock on a file - break it explicitly
1653 * before sending the lock to the server to be sure the next
1654 * read won't conflict with non-overlapted locks due to
1657 if (!CIFS_CACHE_WRITE(CIFS_I(inode
)) &&
1658 CIFS_CACHE_READ(CIFS_I(inode
))) {
1659 cifs_zap_mapping(inode
);
1660 cifs_dbg(FYI
, "Set no oplock for inode=%p due to mand locks\n",
1662 CIFS_I(inode
)->oplock
= 0;
1665 rc
= server
->ops
->mand_lock(xid
, cfile
, flock
->fl_start
, length
,
1666 type
, 1, 0, wait_flag
);
1672 cifs_lock_add(cfile
, lock
);
1674 rc
= server
->ops
->mand_unlock_range(cfile
, flock
, xid
);
1677 if (flock
->fl_flags
& FL_POSIX
) {
1679 * If this is a request to remove all locks because we
1680 * are closing the file, it doesn't matter if the
1681 * unlocking failed as both cifs.ko and the SMB server
1682 * remove the lock on file close
1685 cifs_dbg(VFS
, "%s failed rc=%d\n", __func__
, rc
);
1686 if (!(flock
->fl_flags
& FL_CLOSE
))
1689 rc
= locks_lock_file_wait(file
, flock
);
1694 int cifs_lock(struct file
*file
, int cmd
, struct file_lock
*flock
)
1697 int lock
= 0, unlock
= 0;
1698 bool wait_flag
= false;
1699 bool posix_lck
= false;
1700 struct cifs_sb_info
*cifs_sb
;
1701 struct cifs_tcon
*tcon
;
1702 struct cifsFileInfo
*cfile
;
1708 cifs_dbg(FYI
, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld end: %lld\n",
1709 cmd
, flock
->fl_flags
, flock
->fl_type
,
1710 flock
->fl_start
, flock
->fl_end
);
1712 cfile
= (struct cifsFileInfo
*)file
->private_data
;
1713 tcon
= tlink_tcon(cfile
->tlink
);
1715 cifs_read_flock(flock
, &type
, &lock
, &unlock
, &wait_flag
,
1717 cifs_sb
= CIFS_FILE_SB(file
);
1719 if (cap_unix(tcon
->ses
) &&
1720 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
1721 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
1724 * BB add code here to normalize offset and length to account for
1725 * negative length which we can not accept over the wire.
1727 if (IS_GETLK(cmd
)) {
1728 rc
= cifs_getlk(file
, flock
, type
, wait_flag
, posix_lck
, xid
);
1733 if (!lock
&& !unlock
) {
1735 * if no lock or unlock then nothing to do since we do not
1742 rc
= cifs_setlk(file
, flock
, type
, wait_flag
, posix_lck
, lock
, unlock
,
1749 * update the file size (if needed) after a write. Should be called with
1750 * the inode->i_lock held
1753 cifs_update_eof(struct cifsInodeInfo
*cifsi
, loff_t offset
,
1754 unsigned int bytes_written
)
1756 loff_t end_of_write
= offset
+ bytes_written
;
1758 if (end_of_write
> cifsi
->server_eof
)
1759 cifsi
->server_eof
= end_of_write
;
1763 cifs_write(struct cifsFileInfo
*open_file
, __u32 pid
, const char *write_data
,
1764 size_t write_size
, loff_t
*offset
)
1767 unsigned int bytes_written
= 0;
1768 unsigned int total_written
;
1769 struct cifs_tcon
*tcon
;
1770 struct TCP_Server_Info
*server
;
1772 struct dentry
*dentry
= open_file
->dentry
;
1773 struct cifsInodeInfo
*cifsi
= CIFS_I(d_inode(dentry
));
1774 struct cifs_io_parms io_parms
;
1776 cifs_dbg(FYI
, "write %zd bytes to offset %lld of %pd\n",
1777 write_size
, *offset
, dentry
);
1779 tcon
= tlink_tcon(open_file
->tlink
);
1780 server
= tcon
->ses
->server
;
1782 if (!server
->ops
->sync_write
)
1787 for (total_written
= 0; write_size
> total_written
;
1788 total_written
+= bytes_written
) {
1790 while (rc
== -EAGAIN
) {
1794 if (open_file
->invalidHandle
) {
1795 /* we could deadlock if we called
1796 filemap_fdatawait from here so tell
1797 reopen_file not to flush data to
1799 rc
= cifs_reopen_file(open_file
, false);
1804 len
= min(server
->ops
->wp_retry_size(d_inode(dentry
)),
1805 (unsigned int)write_size
- total_written
);
1806 /* iov[0] is reserved for smb header */
1807 iov
[1].iov_base
= (char *)write_data
+ total_written
;
1808 iov
[1].iov_len
= len
;
1810 io_parms
.tcon
= tcon
;
1811 io_parms
.offset
= *offset
;
1812 io_parms
.length
= len
;
1813 rc
= server
->ops
->sync_write(xid
, &open_file
->fid
,
1814 &io_parms
, &bytes_written
, iov
, 1);
1816 if (rc
|| (bytes_written
== 0)) {
1824 spin_lock(&d_inode(dentry
)->i_lock
);
1825 cifs_update_eof(cifsi
, *offset
, bytes_written
);
1826 spin_unlock(&d_inode(dentry
)->i_lock
);
1827 *offset
+= bytes_written
;
1831 cifs_stats_bytes_written(tcon
, total_written
);
1833 if (total_written
> 0) {
1834 spin_lock(&d_inode(dentry
)->i_lock
);
1835 if (*offset
> d_inode(dentry
)->i_size
)
1836 i_size_write(d_inode(dentry
), *offset
);
1837 spin_unlock(&d_inode(dentry
)->i_lock
);
1839 mark_inode_dirty_sync(d_inode(dentry
));
1841 return total_written
;
1844 struct cifsFileInfo
*find_readable_file(struct cifsInodeInfo
*cifs_inode
,
1847 struct cifsFileInfo
*open_file
= NULL
;
1848 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1850 /* only filter by fsuid on multiuser mounts */
1851 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1854 spin_lock(&cifs_inode
->open_file_lock
);
1855 /* we could simply get the first_list_entry since write-only entries
1856 are always at the end of the list but since the first entry might
1857 have a close pending, we go through the whole list */
1858 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1859 if (fsuid_only
&& !uid_eq(open_file
->uid
, current_fsuid()))
1861 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_READ
) {
1862 if (!open_file
->invalidHandle
) {
1863 /* found a good file */
1864 /* lock it so it will not be closed on us */
1865 cifsFileInfo_get(open_file
);
1866 spin_unlock(&cifs_inode
->open_file_lock
);
1868 } /* else might as well continue, and look for
1869 another, or simply have the caller reopen it
1870 again rather than trying to fix this handle */
1871 } else /* write only file */
1872 break; /* write only files are last so must be done */
1874 spin_unlock(&cifs_inode
->open_file_lock
);
1878 /* Return -EBADF if no handle is found and general rc otherwise */
1880 cifs_get_writable_file(struct cifsInodeInfo
*cifs_inode
, bool fsuid_only
,
1881 struct cifsFileInfo
**ret_file
)
1883 struct cifsFileInfo
*open_file
, *inv_file
= NULL
;
1884 struct cifs_sb_info
*cifs_sb
;
1885 bool any_available
= false;
1887 unsigned int refind
= 0;
1892 * Having a null inode here (because mapping->host was set to zero by
1893 * the VFS or MM) should not happen but we had reports of on oops (due
1894 * to it being zero) during stress testcases so we need to check for it
1897 if (cifs_inode
== NULL
) {
1898 cifs_dbg(VFS
, "Null inode passed to cifs_writeable_file\n");
1903 cifs_sb
= CIFS_SB(cifs_inode
->vfs_inode
.i_sb
);
1905 /* only filter by fsuid on multiuser mounts */
1906 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_MULTIUSER
))
1909 spin_lock(&cifs_inode
->open_file_lock
);
1911 if (refind
> MAX_REOPEN_ATT
) {
1912 spin_unlock(&cifs_inode
->open_file_lock
);
1915 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
1916 if (!any_available
&& open_file
->pid
!= current
->tgid
)
1918 if (fsuid_only
&& !uid_eq(open_file
->uid
, current_fsuid()))
1920 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
1921 if (!open_file
->invalidHandle
) {
1922 /* found a good writable file */
1923 cifsFileInfo_get(open_file
);
1924 spin_unlock(&cifs_inode
->open_file_lock
);
1925 *ret_file
= open_file
;
1929 inv_file
= open_file
;
1933 /* couldn't find useable FH with same pid, try any available */
1934 if (!any_available
) {
1935 any_available
= true;
1936 goto refind_writable
;
1940 any_available
= false;
1941 cifsFileInfo_get(inv_file
);
1944 spin_unlock(&cifs_inode
->open_file_lock
);
1947 rc
= cifs_reopen_file(inv_file
, false);
1949 *ret_file
= inv_file
;
1953 spin_lock(&cifs_inode
->open_file_lock
);
1954 list_move_tail(&inv_file
->flist
, &cifs_inode
->openFileList
);
1955 spin_unlock(&cifs_inode
->open_file_lock
);
1956 cifsFileInfo_put(inv_file
);
1959 spin_lock(&cifs_inode
->open_file_lock
);
1960 goto refind_writable
;
1966 struct cifsFileInfo
*
1967 find_writable_file(struct cifsInodeInfo
*cifs_inode
, bool fsuid_only
)
1969 struct cifsFileInfo
*cfile
;
1972 rc
= cifs_get_writable_file(cifs_inode
, fsuid_only
, &cfile
);
1974 cifs_dbg(FYI
, "couldn't find writable handle rc=%d", rc
);
1980 cifs_get_writable_path(struct cifs_tcon
*tcon
, const char *name
,
1981 struct cifsFileInfo
**ret_file
)
1983 struct list_head
*tmp
;
1984 struct cifsFileInfo
*cfile
;
1985 struct cifsInodeInfo
*cinode
;
1990 spin_lock(&tcon
->open_file_lock
);
1991 list_for_each(tmp
, &tcon
->openFileList
) {
1992 cfile
= list_entry(tmp
, struct cifsFileInfo
,
1994 full_path
= build_path_from_dentry(cfile
->dentry
);
1995 if (full_path
== NULL
) {
1996 spin_unlock(&tcon
->open_file_lock
);
1999 if (strcmp(full_path
, name
)) {
2005 cinode
= CIFS_I(d_inode(cfile
->dentry
));
2006 spin_unlock(&tcon
->open_file_lock
);
2007 return cifs_get_writable_file(cinode
, 0, ret_file
);
2010 spin_unlock(&tcon
->open_file_lock
);
2015 cifs_get_readable_path(struct cifs_tcon
*tcon
, const char *name
,
2016 struct cifsFileInfo
**ret_file
)
2018 struct list_head
*tmp
;
2019 struct cifsFileInfo
*cfile
;
2020 struct cifsInodeInfo
*cinode
;
2025 spin_lock(&tcon
->open_file_lock
);
2026 list_for_each(tmp
, &tcon
->openFileList
) {
2027 cfile
= list_entry(tmp
, struct cifsFileInfo
,
2029 full_path
= build_path_from_dentry(cfile
->dentry
);
2030 if (full_path
== NULL
) {
2031 spin_unlock(&tcon
->open_file_lock
);
2034 if (strcmp(full_path
, name
)) {
2040 cinode
= CIFS_I(d_inode(cfile
->dentry
));
2041 spin_unlock(&tcon
->open_file_lock
);
2042 *ret_file
= find_readable_file(cinode
, 0);
2043 return *ret_file
? 0 : -ENOENT
;
2046 spin_unlock(&tcon
->open_file_lock
);
2050 static int cifs_partialpagewrite(struct page
*page
, unsigned from
, unsigned to
)
2052 struct address_space
*mapping
= page
->mapping
;
2053 loff_t offset
= (loff_t
)page
->index
<< PAGE_SHIFT
;
2056 int bytes_written
= 0;
2057 struct inode
*inode
;
2058 struct cifsFileInfo
*open_file
;
2060 if (!mapping
|| !mapping
->host
)
2063 inode
= page
->mapping
->host
;
2065 offset
+= (loff_t
)from
;
2066 write_data
= kmap(page
);
2069 if ((to
> PAGE_SIZE
) || (from
> to
)) {
2074 /* racing with truncate? */
2075 if (offset
> mapping
->host
->i_size
) {
2077 return 0; /* don't care */
2080 /* check to make sure that we are not extending the file */
2081 if (mapping
->host
->i_size
- offset
< (loff_t
)to
)
2082 to
= (unsigned)(mapping
->host
->i_size
- offset
);
2084 rc
= cifs_get_writable_file(CIFS_I(mapping
->host
), false, &open_file
);
2086 bytes_written
= cifs_write(open_file
, open_file
->pid
,
2087 write_data
, to
- from
, &offset
);
2088 cifsFileInfo_put(open_file
);
2089 /* Does mm or vfs already set times? */
2090 inode
->i_atime
= inode
->i_mtime
= current_time(inode
);
2091 if ((bytes_written
> 0) && (offset
))
2093 else if (bytes_written
< 0)
2098 cifs_dbg(FYI
, "No writable handle for write page rc=%d\n", rc
);
2099 if (!is_retryable_error(rc
))
2107 static struct cifs_writedata
*
2108 wdata_alloc_and_fillpages(pgoff_t tofind
, struct address_space
*mapping
,
2109 pgoff_t end
, pgoff_t
*index
,
2110 unsigned int *found_pages
)
2112 struct cifs_writedata
*wdata
;
2114 wdata
= cifs_writedata_alloc((unsigned int)tofind
,
2115 cifs_writev_complete
);
2119 *found_pages
= find_get_pages_range_tag(mapping
, index
, end
,
2120 PAGECACHE_TAG_DIRTY
, tofind
, wdata
->pages
);
2125 wdata_prepare_pages(struct cifs_writedata
*wdata
, unsigned int found_pages
,
2126 struct address_space
*mapping
,
2127 struct writeback_control
*wbc
,
2128 pgoff_t end
, pgoff_t
*index
, pgoff_t
*next
, bool *done
)
2130 unsigned int nr_pages
= 0, i
;
2133 for (i
= 0; i
< found_pages
; i
++) {
2134 page
= wdata
->pages
[i
];
2136 * At this point we hold neither the i_pages lock nor the
2137 * page lock: the page may be truncated or invalidated
2138 * (changing page->mapping to NULL), or even swizzled
2139 * back from swapper_space to tmpfs file mapping
2144 else if (!trylock_page(page
))
2147 if (unlikely(page
->mapping
!= mapping
)) {
2152 if (!wbc
->range_cyclic
&& page
->index
> end
) {
2158 if (*next
&& (page
->index
!= *next
)) {
2159 /* Not next consecutive page */
2164 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
2165 wait_on_page_writeback(page
);
2167 if (PageWriteback(page
) ||
2168 !clear_page_dirty_for_io(page
)) {
2174 * This actually clears the dirty bit in the radix tree.
2175 * See cifs_writepage() for more commentary.
2177 set_page_writeback(page
);
2178 if (page_offset(page
) >= i_size_read(mapping
->host
)) {
2181 end_page_writeback(page
);
2185 wdata
->pages
[i
] = page
;
2186 *next
= page
->index
+ 1;
2190 /* reset index to refind any pages skipped */
2192 *index
= wdata
->pages
[0]->index
+ 1;
2194 /* put any pages we aren't going to use */
2195 for (i
= nr_pages
; i
< found_pages
; i
++) {
2196 put_page(wdata
->pages
[i
]);
2197 wdata
->pages
[i
] = NULL
;
2204 wdata_send_pages(struct cifs_writedata
*wdata
, unsigned int nr_pages
,
2205 struct address_space
*mapping
, struct writeback_control
*wbc
)
2208 struct TCP_Server_Info
*server
=
2209 tlink_tcon(wdata
->cfile
->tlink
)->ses
->server
;
2211 wdata
->sync_mode
= wbc
->sync_mode
;
2212 wdata
->nr_pages
= nr_pages
;
2213 wdata
->offset
= page_offset(wdata
->pages
[0]);
2214 wdata
->pagesz
= PAGE_SIZE
;
2215 wdata
->tailsz
= min(i_size_read(mapping
->host
) -
2216 page_offset(wdata
->pages
[nr_pages
- 1]),
2218 wdata
->bytes
= ((nr_pages
- 1) * PAGE_SIZE
) + wdata
->tailsz
;
2219 wdata
->pid
= wdata
->cfile
->pid
;
2221 rc
= adjust_credits(server
, &wdata
->credits
, wdata
->bytes
);
2225 if (wdata
->cfile
->invalidHandle
)
2228 rc
= server
->ops
->async_writev(wdata
, cifs_writedata_release
);
2233 static int cifs_writepages(struct address_space
*mapping
,
2234 struct writeback_control
*wbc
)
2236 struct inode
*inode
= mapping
->host
;
2237 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
2238 struct TCP_Server_Info
*server
;
2239 bool done
= false, scanned
= false, range_whole
= false;
2241 struct cifs_writedata
*wdata
;
2242 struct cifsFileInfo
*cfile
= NULL
;
2248 * If wsize is smaller than the page cache size, default to writing
2249 * one page at a time via cifs_writepage
2251 if (cifs_sb
->wsize
< PAGE_SIZE
)
2252 return generic_writepages(mapping
, wbc
);
2255 if (wbc
->range_cyclic
) {
2256 index
= mapping
->writeback_index
; /* Start from prev offset */
2259 index
= wbc
->range_start
>> PAGE_SHIFT
;
2260 end
= wbc
->range_end
>> PAGE_SHIFT
;
2261 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2265 server
= cifs_sb_master_tcon(cifs_sb
)->ses
->server
;
2267 while (!done
&& index
<= end
) {
2268 unsigned int i
, nr_pages
, found_pages
, wsize
;
2269 pgoff_t next
= 0, tofind
, saved_index
= index
;
2270 struct cifs_credits credits_on_stack
;
2271 struct cifs_credits
*credits
= &credits_on_stack
;
2272 int get_file_rc
= 0;
2275 cifsFileInfo_put(cfile
);
2277 rc
= cifs_get_writable_file(CIFS_I(inode
), false, &cfile
);
2279 /* in case of an error store it to return later */
2283 rc
= server
->ops
->wait_mtu_credits(server
, cifs_sb
->wsize
,
2290 tofind
= min((wsize
/ PAGE_SIZE
) - 1, end
- index
) + 1;
2292 wdata
= wdata_alloc_and_fillpages(tofind
, mapping
, end
, &index
,
2297 add_credits_and_wake_if(server
, credits
, 0);
2301 if (found_pages
== 0) {
2302 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2303 add_credits_and_wake_if(server
, credits
, 0);
2307 nr_pages
= wdata_prepare_pages(wdata
, found_pages
, mapping
, wbc
,
2308 end
, &index
, &next
, &done
);
2310 /* nothing to write? */
2311 if (nr_pages
== 0) {
2312 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2313 add_credits_and_wake_if(server
, credits
, 0);
2317 wdata
->credits
= credits_on_stack
;
2318 wdata
->cfile
= cfile
;
2321 if (!wdata
->cfile
) {
2322 cifs_dbg(VFS
, "No writable handle in writepages rc=%d\n",
2324 if (is_retryable_error(get_file_rc
))
2329 rc
= wdata_send_pages(wdata
, nr_pages
, mapping
, wbc
);
2331 for (i
= 0; i
< nr_pages
; ++i
)
2332 unlock_page(wdata
->pages
[i
]);
2334 /* send failure -- clean up the mess */
2336 add_credits_and_wake_if(server
, &wdata
->credits
, 0);
2337 for (i
= 0; i
< nr_pages
; ++i
) {
2338 if (is_retryable_error(rc
))
2339 redirty_page_for_writepage(wbc
,
2342 SetPageError(wdata
->pages
[i
]);
2343 end_page_writeback(wdata
->pages
[i
]);
2344 put_page(wdata
->pages
[i
]);
2346 if (!is_retryable_error(rc
))
2347 mapping_set_error(mapping
, rc
);
2349 kref_put(&wdata
->refcount
, cifs_writedata_release
);
2351 if (wbc
->sync_mode
== WB_SYNC_ALL
&& rc
== -EAGAIN
) {
2352 index
= saved_index
;
2356 /* Return immediately if we received a signal during writing */
2357 if (is_interrupt_error(rc
)) {
2362 if (rc
!= 0 && saved_rc
== 0)
2365 wbc
->nr_to_write
-= nr_pages
;
2366 if (wbc
->nr_to_write
<= 0)
2372 if (!scanned
&& !done
) {
2374 * We hit the last page and there is more work to be done: wrap
2375 * back to the start of the file
2385 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2386 mapping
->writeback_index
= index
;
2389 cifsFileInfo_put(cfile
);
2395 cifs_writepage_locked(struct page
*page
, struct writeback_control
*wbc
)
2401 /* BB add check for wbc flags */
2403 if (!PageUptodate(page
))
2404 cifs_dbg(FYI
, "ppw - page not up to date\n");
2407 * Set the "writeback" flag, and clear "dirty" in the radix tree.
2409 * A writepage() implementation always needs to do either this,
2410 * or re-dirty the page with "redirty_page_for_writepage()" in
2411 * the case of a failure.
2413 * Just unlocking the page will cause the radix tree tag-bits
2414 * to fail to update with the state of the page correctly.
2416 set_page_writeback(page
);
2418 rc
= cifs_partialpagewrite(page
, 0, PAGE_SIZE
);
2419 if (is_retryable_error(rc
)) {
2420 if (wbc
->sync_mode
== WB_SYNC_ALL
&& rc
== -EAGAIN
)
2422 redirty_page_for_writepage(wbc
, page
);
2423 } else if (rc
!= 0) {
2425 mapping_set_error(page
->mapping
, rc
);
2427 SetPageUptodate(page
);
2429 end_page_writeback(page
);
2435 static int cifs_writepage(struct page
*page
, struct writeback_control
*wbc
)
2437 int rc
= cifs_writepage_locked(page
, wbc
);
2442 static int cifs_write_end(struct file
*file
, struct address_space
*mapping
,
2443 loff_t pos
, unsigned len
, unsigned copied
,
2444 struct page
*page
, void *fsdata
)
2447 struct inode
*inode
= mapping
->host
;
2448 struct cifsFileInfo
*cfile
= file
->private_data
;
2449 struct cifs_sb_info
*cifs_sb
= CIFS_SB(cfile
->dentry
->d_sb
);
2452 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2455 pid
= current
->tgid
;
2457 cifs_dbg(FYI
, "write_end for page %p from pos %lld with %d bytes\n",
2460 if (PageChecked(page
)) {
2462 SetPageUptodate(page
);
2463 ClearPageChecked(page
);
2464 } else if (!PageUptodate(page
) && copied
== PAGE_SIZE
)
2465 SetPageUptodate(page
);
2467 if (!PageUptodate(page
)) {
2469 unsigned offset
= pos
& (PAGE_SIZE
- 1);
2473 /* this is probably better than directly calling
2474 partialpage_write since in this function the file handle is
2475 known which we might as well leverage */
2476 /* BB check if anything else missing out of ppw
2477 such as updating last write time */
2478 page_data
= kmap(page
);
2479 rc
= cifs_write(cfile
, pid
, page_data
+ offset
, copied
, &pos
);
2480 /* if (rc < 0) should we set writebehind rc? */
2487 set_page_dirty(page
);
2491 spin_lock(&inode
->i_lock
);
2492 if (pos
> inode
->i_size
)
2493 i_size_write(inode
, pos
);
2494 spin_unlock(&inode
->i_lock
);
2503 int cifs_strict_fsync(struct file
*file
, loff_t start
, loff_t end
,
2508 struct cifs_tcon
*tcon
;
2509 struct TCP_Server_Info
*server
;
2510 struct cifsFileInfo
*smbfile
= file
->private_data
;
2511 struct inode
*inode
= file_inode(file
);
2512 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
2514 rc
= file_write_and_wait_range(file
, start
, end
);
2520 cifs_dbg(FYI
, "Sync file - name: %pD datasync: 0x%x\n",
2523 if (!CIFS_CACHE_READ(CIFS_I(inode
))) {
2524 rc
= cifs_zap_mapping(inode
);
2526 cifs_dbg(FYI
, "rc: %d during invalidate phase\n", rc
);
2527 rc
= 0; /* don't care about it in fsync */
2531 tcon
= tlink_tcon(smbfile
->tlink
);
2532 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOSSYNC
)) {
2533 server
= tcon
->ses
->server
;
2534 if (server
->ops
->flush
)
2535 rc
= server
->ops
->flush(xid
, tcon
, &smbfile
->fid
);
2544 int cifs_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
2548 struct cifs_tcon
*tcon
;
2549 struct TCP_Server_Info
*server
;
2550 struct cifsFileInfo
*smbfile
= file
->private_data
;
2551 struct cifs_sb_info
*cifs_sb
= CIFS_FILE_SB(file
);
2553 rc
= file_write_and_wait_range(file
, start
, end
);
2559 cifs_dbg(FYI
, "Sync file - name: %pD datasync: 0x%x\n",
2562 tcon
= tlink_tcon(smbfile
->tlink
);
2563 if (!(cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOSSYNC
)) {
2564 server
= tcon
->ses
->server
;
2565 if (server
->ops
->flush
)
2566 rc
= server
->ops
->flush(xid
, tcon
, &smbfile
->fid
);
2576 * As file closes, flush all cached write data for this inode checking
2577 * for write behind errors.
2579 int cifs_flush(struct file
*file
, fl_owner_t id
)
2581 struct inode
*inode
= file_inode(file
);
2584 if (file
->f_mode
& FMODE_WRITE
)
2585 rc
= filemap_write_and_wait(inode
->i_mapping
);
2587 cifs_dbg(FYI
, "Flush inode %p file %p rc %d\n", inode
, file
, rc
);
2593 cifs_write_allocate_pages(struct page
**pages
, unsigned long num_pages
)
2598 for (i
= 0; i
< num_pages
; i
++) {
2599 pages
[i
] = alloc_page(GFP_KERNEL
|__GFP_HIGHMEM
);
2602 * save number of pages we have already allocated and
2603 * return with ENOMEM error
2612 for (i
= 0; i
< num_pages
; i
++)
2619 size_t get_numpages(const size_t wsize
, const size_t len
, size_t *cur_len
)
2624 clen
= min_t(const size_t, len
, wsize
);
2625 num_pages
= DIV_ROUND_UP(clen
, PAGE_SIZE
);
2634 cifs_uncached_writedata_release(struct kref
*refcount
)
2637 struct cifs_writedata
*wdata
= container_of(refcount
,
2638 struct cifs_writedata
, refcount
);
2640 kref_put(&wdata
->ctx
->refcount
, cifs_aio_ctx_release
);
2641 for (i
= 0; i
< wdata
->nr_pages
; i
++)
2642 put_page(wdata
->pages
[i
]);
2643 cifs_writedata_release(refcount
);
2646 static void collect_uncached_write_data(struct cifs_aio_ctx
*ctx
);
2649 cifs_uncached_writev_complete(struct work_struct
*work
)
2651 struct cifs_writedata
*wdata
= container_of(work
,
2652 struct cifs_writedata
, work
);
2653 struct inode
*inode
= d_inode(wdata
->cfile
->dentry
);
2654 struct cifsInodeInfo
*cifsi
= CIFS_I(inode
);
2656 spin_lock(&inode
->i_lock
);
2657 cifs_update_eof(cifsi
, wdata
->offset
, wdata
->bytes
);
2658 if (cifsi
->server_eof
> inode
->i_size
)
2659 i_size_write(inode
, cifsi
->server_eof
);
2660 spin_unlock(&inode
->i_lock
);
2662 complete(&wdata
->done
);
2663 collect_uncached_write_data(wdata
->ctx
);
2664 /* the below call can possibly free the last ref to aio ctx */
2665 kref_put(&wdata
->refcount
, cifs_uncached_writedata_release
);
2669 wdata_fill_from_iovec(struct cifs_writedata
*wdata
, struct iov_iter
*from
,
2670 size_t *len
, unsigned long *num_pages
)
2672 size_t save_len
, copied
, bytes
, cur_len
= *len
;
2673 unsigned long i
, nr_pages
= *num_pages
;
2676 for (i
= 0; i
< nr_pages
; i
++) {
2677 bytes
= min_t(const size_t, cur_len
, PAGE_SIZE
);
2678 copied
= copy_page_from_iter(wdata
->pages
[i
], 0, bytes
, from
);
2681 * If we didn't copy as much as we expected, then that
2682 * may mean we trod into an unmapped area. Stop copying
2683 * at that point. On the next pass through the big
2684 * loop, we'll likely end up getting a zero-length
2685 * write and bailing out of it.
2690 cur_len
= save_len
- cur_len
;
2694 * If we have no data to send, then that probably means that
2695 * the copy above failed altogether. That's most likely because
2696 * the address in the iovec was bogus. Return -EFAULT and let
2697 * the caller free anything we allocated and bail out.
2703 * i + 1 now represents the number of pages we actually used in
2704 * the copy phase above.
2711 cifs_resend_wdata(struct cifs_writedata
*wdata
, struct list_head
*wdata_list
,
2712 struct cifs_aio_ctx
*ctx
)
2715 struct cifs_credits credits
;
2717 struct TCP_Server_Info
*server
=
2718 tlink_tcon(wdata
->cfile
->tlink
)->ses
->server
;
2721 if (wdata
->cfile
->invalidHandle
) {
2722 rc
= cifs_reopen_file(wdata
->cfile
, false);
2731 * Wait for credits to resend this wdata.
2732 * Note: we are attempting to resend the whole wdata not in
2736 rc
= server
->ops
->wait_mtu_credits(server
, wdata
->bytes
,
2741 if (wsize
< wdata
->bytes
) {
2742 add_credits_and_wake_if(server
, &credits
, 0);
2745 } while (wsize
< wdata
->bytes
);
2746 wdata
->credits
= credits
;
2748 rc
= adjust_credits(server
, &wdata
->credits
, wdata
->bytes
);
2751 if (wdata
->cfile
->invalidHandle
)
2754 rc
= server
->ops
->async_writev(wdata
,
2755 cifs_uncached_writedata_release
);
2758 /* If the write was successfully sent, we are done */
2760 list_add_tail(&wdata
->list
, wdata_list
);
2764 /* Roll back credits and retry if needed */
2765 add_credits_and_wake_if(server
, &wdata
->credits
, 0);
2766 } while (rc
== -EAGAIN
);
2769 kref_put(&wdata
->refcount
, cifs_uncached_writedata_release
);
2774 cifs_write_from_iter(loff_t offset
, size_t len
, struct iov_iter
*from
,
2775 struct cifsFileInfo
*open_file
,
2776 struct cifs_sb_info
*cifs_sb
, struct list_head
*wdata_list
,
2777 struct cifs_aio_ctx
*ctx
)
2781 unsigned long nr_pages
, num_pages
, i
;
2782 struct cifs_writedata
*wdata
;
2783 struct iov_iter saved_from
= *from
;
2784 loff_t saved_offset
= offset
;
2786 struct TCP_Server_Info
*server
;
2787 struct page
**pagevec
;
2791 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
2792 pid
= open_file
->pid
;
2794 pid
= current
->tgid
;
2796 server
= tlink_tcon(open_file
->tlink
)->ses
->server
;
2801 struct cifs_credits credits_on_stack
;
2802 struct cifs_credits
*credits
= &credits_on_stack
;
2804 if (open_file
->invalidHandle
) {
2805 rc
= cifs_reopen_file(open_file
, false);
2812 rc
= server
->ops
->wait_mtu_credits(server
, cifs_sb
->wsize
,
2817 cur_len
= min_t(const size_t, len
, wsize
);
2819 if (ctx
->direct_io
) {
2822 result
= iov_iter_get_pages_alloc(
2823 from
, &pagevec
, cur_len
, &start
);
2826 "direct_writev couldn't get user pages "
2827 "(rc=%zd) iter type %d iov_offset %zd "
2830 from
->iov_offset
, from
->count
);
2834 add_credits_and_wake_if(server
, credits
, 0);
2837 cur_len
= (size_t)result
;
2838 iov_iter_advance(from
, cur_len
);
2841 (cur_len
+ start
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
2843 wdata
= cifs_writedata_direct_alloc(pagevec
,
2844 cifs_uncached_writev_complete
);
2847 add_credits_and_wake_if(server
, credits
, 0);
2852 wdata
->page_offset
= start
;
2855 cur_len
- (PAGE_SIZE
- start
) -
2856 (nr_pages
- 2) * PAGE_SIZE
:
2859 nr_pages
= get_numpages(wsize
, len
, &cur_len
);
2860 wdata
= cifs_writedata_alloc(nr_pages
,
2861 cifs_uncached_writev_complete
);
2864 add_credits_and_wake_if(server
, credits
, 0);
2868 rc
= cifs_write_allocate_pages(wdata
->pages
, nr_pages
);
2870 kvfree(wdata
->pages
);
2872 add_credits_and_wake_if(server
, credits
, 0);
2876 num_pages
= nr_pages
;
2877 rc
= wdata_fill_from_iovec(
2878 wdata
, from
, &cur_len
, &num_pages
);
2880 for (i
= 0; i
< nr_pages
; i
++)
2881 put_page(wdata
->pages
[i
]);
2882 kvfree(wdata
->pages
);
2884 add_credits_and_wake_if(server
, credits
, 0);
2889 * Bring nr_pages down to the number of pages we
2890 * actually used, and free any pages that we didn't use.
2892 for ( ; nr_pages
> num_pages
; nr_pages
--)
2893 put_page(wdata
->pages
[nr_pages
- 1]);
2895 wdata
->tailsz
= cur_len
- ((nr_pages
- 1) * PAGE_SIZE
);
2898 wdata
->sync_mode
= WB_SYNC_ALL
;
2899 wdata
->nr_pages
= nr_pages
;
2900 wdata
->offset
= (__u64
)offset
;
2901 wdata
->cfile
= cifsFileInfo_get(open_file
);
2903 wdata
->bytes
= cur_len
;
2904 wdata
->pagesz
= PAGE_SIZE
;
2905 wdata
->credits
= credits_on_stack
;
2907 kref_get(&ctx
->refcount
);
2909 rc
= adjust_credits(server
, &wdata
->credits
, wdata
->bytes
);
2912 if (wdata
->cfile
->invalidHandle
)
2915 rc
= server
->ops
->async_writev(wdata
,
2916 cifs_uncached_writedata_release
);
2920 add_credits_and_wake_if(server
, &wdata
->credits
, 0);
2921 kref_put(&wdata
->refcount
,
2922 cifs_uncached_writedata_release
);
2923 if (rc
== -EAGAIN
) {
2925 iov_iter_advance(from
, offset
- saved_offset
);
2931 list_add_tail(&wdata
->list
, wdata_list
);
2940 static void collect_uncached_write_data(struct cifs_aio_ctx
*ctx
)
2942 struct cifs_writedata
*wdata
, *tmp
;
2943 struct cifs_tcon
*tcon
;
2944 struct cifs_sb_info
*cifs_sb
;
2945 struct dentry
*dentry
= ctx
->cfile
->dentry
;
2948 tcon
= tlink_tcon(ctx
->cfile
->tlink
);
2949 cifs_sb
= CIFS_SB(dentry
->d_sb
);
2951 mutex_lock(&ctx
->aio_mutex
);
2953 if (list_empty(&ctx
->list
)) {
2954 mutex_unlock(&ctx
->aio_mutex
);
2960 * Wait for and collect replies for any successful sends in order of
2961 * increasing offset. Once an error is hit, then return without waiting
2962 * for any more replies.
2965 list_for_each_entry_safe(wdata
, tmp
, &ctx
->list
, list
) {
2967 if (!try_wait_for_completion(&wdata
->done
)) {
2968 mutex_unlock(&ctx
->aio_mutex
);
2975 ctx
->total_len
+= wdata
->bytes
;
2977 /* resend call if it's a retryable error */
2978 if (rc
== -EAGAIN
) {
2979 struct list_head tmp_list
;
2980 struct iov_iter tmp_from
= ctx
->iter
;
2982 INIT_LIST_HEAD(&tmp_list
);
2983 list_del_init(&wdata
->list
);
2986 rc
= cifs_resend_wdata(
2987 wdata
, &tmp_list
, ctx
);
2989 iov_iter_advance(&tmp_from
,
2990 wdata
->offset
- ctx
->pos
);
2992 rc
= cifs_write_from_iter(wdata
->offset
,
2993 wdata
->bytes
, &tmp_from
,
2994 ctx
->cfile
, cifs_sb
, &tmp_list
,
2997 kref_put(&wdata
->refcount
,
2998 cifs_uncached_writedata_release
);
3001 list_splice(&tmp_list
, &ctx
->list
);
3005 list_del_init(&wdata
->list
);
3006 kref_put(&wdata
->refcount
, cifs_uncached_writedata_release
);
3009 cifs_stats_bytes_written(tcon
, ctx
->total_len
);
3010 set_bit(CIFS_INO_INVALID_MAPPING
, &CIFS_I(dentry
->d_inode
)->flags
);
3012 ctx
->rc
= (rc
== 0) ? ctx
->total_len
: rc
;
3014 mutex_unlock(&ctx
->aio_mutex
);
3016 if (ctx
->iocb
&& ctx
->iocb
->ki_complete
)
3017 ctx
->iocb
->ki_complete(ctx
->iocb
, ctx
->rc
, 0);
3019 complete(&ctx
->done
);
3022 static ssize_t
__cifs_writev(
3023 struct kiocb
*iocb
, struct iov_iter
*from
, bool direct
)
3025 struct file
*file
= iocb
->ki_filp
;
3026 ssize_t total_written
= 0;
3027 struct cifsFileInfo
*cfile
;
3028 struct cifs_tcon
*tcon
;
3029 struct cifs_sb_info
*cifs_sb
;
3030 struct cifs_aio_ctx
*ctx
;
3031 struct iov_iter saved_from
= *from
;
3032 size_t len
= iov_iter_count(from
);
3036 * iov_iter_get_pages_alloc doesn't work with ITER_KVEC.
3037 * In this case, fall back to non-direct write function.
3038 * this could be improved by getting pages directly in ITER_KVEC
3040 if (direct
&& from
->type
& ITER_KVEC
) {
3041 cifs_dbg(FYI
, "use non-direct cifs_writev for kvec I/O\n");
3045 rc
= generic_write_checks(iocb
, from
);
3049 cifs_sb
= CIFS_FILE_SB(file
);
3050 cfile
= file
->private_data
;
3051 tcon
= tlink_tcon(cfile
->tlink
);
3053 if (!tcon
->ses
->server
->ops
->async_writev
)
3056 ctx
= cifs_aio_ctx_alloc();
3060 ctx
->cfile
= cifsFileInfo_get(cfile
);
3062 if (!is_sync_kiocb(iocb
))
3065 ctx
->pos
= iocb
->ki_pos
;
3068 ctx
->direct_io
= true;
3072 rc
= setup_aio_ctx_iter(ctx
, from
, WRITE
);
3074 kref_put(&ctx
->refcount
, cifs_aio_ctx_release
);
3079 /* grab a lock here due to read response handlers can access ctx */
3080 mutex_lock(&ctx
->aio_mutex
);
3082 rc
= cifs_write_from_iter(iocb
->ki_pos
, ctx
->len
, &saved_from
,
3083 cfile
, cifs_sb
, &ctx
->list
, ctx
);
3086 * If at least one write was successfully sent, then discard any rc
3087 * value from the later writes. If the other write succeeds, then
3088 * we'll end up returning whatever was written. If it fails, then
3089 * we'll get a new rc value from that.
3091 if (!list_empty(&ctx
->list
))
3094 mutex_unlock(&ctx
->aio_mutex
);
3097 kref_put(&ctx
->refcount
, cifs_aio_ctx_release
);
3101 if (!is_sync_kiocb(iocb
)) {
3102 kref_put(&ctx
->refcount
, cifs_aio_ctx_release
);
3103 return -EIOCBQUEUED
;
3106 rc
= wait_for_completion_killable(&ctx
->done
);
3108 mutex_lock(&ctx
->aio_mutex
);
3109 ctx
->rc
= rc
= -EINTR
;
3110 total_written
= ctx
->total_len
;
3111 mutex_unlock(&ctx
->aio_mutex
);
3114 total_written
= ctx
->total_len
;
3117 kref_put(&ctx
->refcount
, cifs_aio_ctx_release
);
3119 if (unlikely(!total_written
))
3122 iocb
->ki_pos
+= total_written
;
3123 return total_written
;
3126 ssize_t
cifs_direct_writev(struct kiocb
*iocb
, struct iov_iter
*from
)
3128 return __cifs_writev(iocb
, from
, true);
3131 ssize_t
cifs_user_writev(struct kiocb
*iocb
, struct iov_iter
*from
)
3133 return __cifs_writev(iocb
, from
, false);
3137 cifs_writev(struct kiocb
*iocb
, struct iov_iter
*from
)
3139 struct file
*file
= iocb
->ki_filp
;
3140 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)file
->private_data
;
3141 struct inode
*inode
= file
->f_mapping
->host
;
3142 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
3143 struct TCP_Server_Info
*server
= tlink_tcon(cfile
->tlink
)->ses
->server
;
3148 * We need to hold the sem to be sure nobody modifies lock list
3149 * with a brlock that prevents writing.
3151 down_read(&cinode
->lock_sem
);
3153 rc
= generic_write_checks(iocb
, from
);
3157 if (!cifs_find_lock_conflict(cfile
, iocb
->ki_pos
, iov_iter_count(from
),
3158 server
->vals
->exclusive_lock_type
, 0,
3159 NULL
, CIFS_WRITE_OP
))
3160 rc
= __generic_file_write_iter(iocb
, from
);
3164 up_read(&cinode
->lock_sem
);
3165 inode_unlock(inode
);
3168 rc
= generic_write_sync(iocb
, rc
);
3173 cifs_strict_writev(struct kiocb
*iocb
, struct iov_iter
*from
)
3175 struct inode
*inode
= file_inode(iocb
->ki_filp
);
3176 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
3177 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
3178 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)
3179 iocb
->ki_filp
->private_data
;
3180 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
3183 written
= cifs_get_writer(cinode
);
3187 if (CIFS_CACHE_WRITE(cinode
)) {
3188 if (cap_unix(tcon
->ses
) &&
3189 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
))
3190 && ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0)) {
3191 written
= generic_file_write_iter(iocb
, from
);
3194 written
= cifs_writev(iocb
, from
);
3198 * For non-oplocked files in strict cache mode we need to write the data
3199 * to the server exactly from the pos to pos+len-1 rather than flush all
3200 * affected pages because it may cause a error with mandatory locks on
3201 * these pages but not on the region from pos to ppos+len-1.
3203 written
= cifs_user_writev(iocb
, from
);
3204 if (CIFS_CACHE_READ(cinode
)) {
3206 * We have read level caching and we have just sent a write
3207 * request to the server thus making data in the cache stale.
3208 * Zap the cache and set oplock/lease level to NONE to avoid
3209 * reading stale data from the cache. All subsequent read
3210 * operations will read new data from the server.
3212 cifs_zap_mapping(inode
);
3213 cifs_dbg(FYI
, "Set Oplock/Lease to NONE for inode=%p after write\n",
3218 cifs_put_writer(cinode
);
3222 static struct cifs_readdata
*
3223 cifs_readdata_direct_alloc(struct page
**pages
, work_func_t complete
)
3225 struct cifs_readdata
*rdata
;
3227 rdata
= kzalloc(sizeof(*rdata
), GFP_KERNEL
);
3228 if (rdata
!= NULL
) {
3229 rdata
->pages
= pages
;
3230 kref_init(&rdata
->refcount
);
3231 INIT_LIST_HEAD(&rdata
->list
);
3232 init_completion(&rdata
->done
);
3233 INIT_WORK(&rdata
->work
, complete
);
3239 static struct cifs_readdata
*
3240 cifs_readdata_alloc(unsigned int nr_pages
, work_func_t complete
)
3242 struct page
**pages
=
3243 kcalloc(nr_pages
, sizeof(struct page
*), GFP_KERNEL
);
3244 struct cifs_readdata
*ret
= NULL
;
3247 ret
= cifs_readdata_direct_alloc(pages
, complete
);
3256 cifs_readdata_release(struct kref
*refcount
)
3258 struct cifs_readdata
*rdata
= container_of(refcount
,
3259 struct cifs_readdata
, refcount
);
3260 #ifdef CONFIG_CIFS_SMB_DIRECT
3262 smbd_deregister_mr(rdata
->mr
);
3267 cifsFileInfo_put(rdata
->cfile
);
3269 kvfree(rdata
->pages
);
3274 cifs_read_allocate_pages(struct cifs_readdata
*rdata
, unsigned int nr_pages
)
3280 for (i
= 0; i
< nr_pages
; i
++) {
3281 page
= alloc_page(GFP_KERNEL
|__GFP_HIGHMEM
);
3286 rdata
->pages
[i
] = page
;
3290 unsigned int nr_page_failed
= i
;
3292 for (i
= 0; i
< nr_page_failed
; i
++) {
3293 put_page(rdata
->pages
[i
]);
3294 rdata
->pages
[i
] = NULL
;
3301 cifs_uncached_readdata_release(struct kref
*refcount
)
3303 struct cifs_readdata
*rdata
= container_of(refcount
,
3304 struct cifs_readdata
, refcount
);
3307 kref_put(&rdata
->ctx
->refcount
, cifs_aio_ctx_release
);
3308 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
3309 put_page(rdata
->pages
[i
]);
3311 cifs_readdata_release(refcount
);
3315 * cifs_readdata_to_iov - copy data from pages in response to an iovec
3316 * @rdata: the readdata response with list of pages holding data
3317 * @iter: destination for our data
3319 * This function copies data from a list of pages in a readdata response into
3320 * an array of iovecs. It will first calculate where the data should go
3321 * based on the info in the readdata and then copy the data into that spot.
3324 cifs_readdata_to_iov(struct cifs_readdata
*rdata
, struct iov_iter
*iter
)
3326 size_t remaining
= rdata
->got_bytes
;
3329 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
3330 struct page
*page
= rdata
->pages
[i
];
3331 size_t copy
= min_t(size_t, remaining
, PAGE_SIZE
);
3334 if (unlikely(iov_iter_is_pipe(iter
))) {
3335 void *addr
= kmap_atomic(page
);
3337 written
= copy_to_iter(addr
, copy
, iter
);
3338 kunmap_atomic(addr
);
3340 written
= copy_page_to_iter(page
, 0, copy
, iter
);
3341 remaining
-= written
;
3342 if (written
< copy
&& iov_iter_count(iter
) > 0)
3345 return remaining
? -EFAULT
: 0;
3348 static void collect_uncached_read_data(struct cifs_aio_ctx
*ctx
);
3351 cifs_uncached_readv_complete(struct work_struct
*work
)
3353 struct cifs_readdata
*rdata
= container_of(work
,
3354 struct cifs_readdata
, work
);
3356 complete(&rdata
->done
);
3357 collect_uncached_read_data(rdata
->ctx
);
3358 /* the below call can possibly free the last ref to aio ctx */
3359 kref_put(&rdata
->refcount
, cifs_uncached_readdata_release
);
3363 uncached_fill_pages(struct TCP_Server_Info
*server
,
3364 struct cifs_readdata
*rdata
, struct iov_iter
*iter
,
3369 unsigned int nr_pages
= rdata
->nr_pages
;
3370 unsigned int page_offset
= rdata
->page_offset
;
3372 rdata
->got_bytes
= 0;
3373 rdata
->tailsz
= PAGE_SIZE
;
3374 for (i
= 0; i
< nr_pages
; i
++) {
3375 struct page
*page
= rdata
->pages
[i
];
3377 unsigned int segment_size
= rdata
->pagesz
;
3380 segment_size
-= page_offset
;
3386 /* no need to hold page hostage */
3387 rdata
->pages
[i
] = NULL
;
3394 if (len
>= segment_size
)
3395 /* enough data to fill the page */
3398 rdata
->tailsz
= len
;
3402 result
= copy_page_from_iter(
3403 page
, page_offset
, n
, iter
);
3404 #ifdef CONFIG_CIFS_SMB_DIRECT
3409 result
= cifs_read_page_from_socket(
3410 server
, page
, page_offset
, n
);
3414 rdata
->got_bytes
+= result
;
3417 return rdata
->got_bytes
> 0 && result
!= -ECONNABORTED
?
3418 rdata
->got_bytes
: result
;
3422 cifs_uncached_read_into_pages(struct TCP_Server_Info
*server
,
3423 struct cifs_readdata
*rdata
, unsigned int len
)
3425 return uncached_fill_pages(server
, rdata
, NULL
, len
);
3429 cifs_uncached_copy_into_pages(struct TCP_Server_Info
*server
,
3430 struct cifs_readdata
*rdata
,
3431 struct iov_iter
*iter
)
3433 return uncached_fill_pages(server
, rdata
, iter
, iter
->count
);
3436 static int cifs_resend_rdata(struct cifs_readdata
*rdata
,
3437 struct list_head
*rdata_list
,
3438 struct cifs_aio_ctx
*ctx
)
3441 struct cifs_credits credits
;
3443 struct TCP_Server_Info
*server
=
3444 tlink_tcon(rdata
->cfile
->tlink
)->ses
->server
;
3447 if (rdata
->cfile
->invalidHandle
) {
3448 rc
= cifs_reopen_file(rdata
->cfile
, true);
3456 * Wait for credits to resend this rdata.
3457 * Note: we are attempting to resend the whole rdata not in
3461 rc
= server
->ops
->wait_mtu_credits(server
, rdata
->bytes
,
3467 if (rsize
< rdata
->bytes
) {
3468 add_credits_and_wake_if(server
, &credits
, 0);
3471 } while (rsize
< rdata
->bytes
);
3472 rdata
->credits
= credits
;
3474 rc
= adjust_credits(server
, &rdata
->credits
, rdata
->bytes
);
3476 if (rdata
->cfile
->invalidHandle
)
3479 rc
= server
->ops
->async_readv(rdata
);
3482 /* If the read was successfully sent, we are done */
3484 /* Add to aio pending list */
3485 list_add_tail(&rdata
->list
, rdata_list
);
3489 /* Roll back credits and retry if needed */
3490 add_credits_and_wake_if(server
, &rdata
->credits
, 0);
3491 } while (rc
== -EAGAIN
);
3494 kref_put(&rdata
->refcount
, cifs_uncached_readdata_release
);
3499 cifs_send_async_read(loff_t offset
, size_t len
, struct cifsFileInfo
*open_file
,
3500 struct cifs_sb_info
*cifs_sb
, struct list_head
*rdata_list
,
3501 struct cifs_aio_ctx
*ctx
)
3503 struct cifs_readdata
*rdata
;
3504 unsigned int npages
, rsize
;
3505 struct cifs_credits credits_on_stack
;
3506 struct cifs_credits
*credits
= &credits_on_stack
;
3510 struct TCP_Server_Info
*server
;
3511 struct page
**pagevec
;
3513 struct iov_iter direct_iov
= ctx
->iter
;
3515 server
= tlink_tcon(open_file
->tlink
)->ses
->server
;
3517 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
3518 pid
= open_file
->pid
;
3520 pid
= current
->tgid
;
3523 iov_iter_advance(&direct_iov
, offset
- ctx
->pos
);
3526 if (open_file
->invalidHandle
) {
3527 rc
= cifs_reopen_file(open_file
, true);
3534 rc
= server
->ops
->wait_mtu_credits(server
, cifs_sb
->rsize
,
3539 cur_len
= min_t(const size_t, len
, rsize
);
3541 if (ctx
->direct_io
) {
3544 result
= iov_iter_get_pages_alloc(
3545 &direct_iov
, &pagevec
,
3549 "couldn't get user pages (rc=%zd)"
3551 " iov_offset %zd count %zd\n",
3552 result
, direct_iov
.type
,
3553 direct_iov
.iov_offset
,
3558 add_credits_and_wake_if(server
, credits
, 0);
3561 cur_len
= (size_t)result
;
3562 iov_iter_advance(&direct_iov
, cur_len
);
3564 rdata
= cifs_readdata_direct_alloc(
3565 pagevec
, cifs_uncached_readv_complete
);
3567 add_credits_and_wake_if(server
, credits
, 0);
3572 npages
= (cur_len
+ start
+ PAGE_SIZE
-1) / PAGE_SIZE
;
3573 rdata
->page_offset
= start
;
3574 rdata
->tailsz
= npages
> 1 ?
3575 cur_len
-(PAGE_SIZE
-start
)-(npages
-2)*PAGE_SIZE
:
3580 npages
= DIV_ROUND_UP(cur_len
, PAGE_SIZE
);
3581 /* allocate a readdata struct */
3582 rdata
= cifs_readdata_alloc(npages
,
3583 cifs_uncached_readv_complete
);
3585 add_credits_and_wake_if(server
, credits
, 0);
3590 rc
= cifs_read_allocate_pages(rdata
, npages
);
3592 kvfree(rdata
->pages
);
3594 add_credits_and_wake_if(server
, credits
, 0);
3598 rdata
->tailsz
= PAGE_SIZE
;
3601 rdata
->cfile
= cifsFileInfo_get(open_file
);
3602 rdata
->nr_pages
= npages
;
3603 rdata
->offset
= offset
;
3604 rdata
->bytes
= cur_len
;
3606 rdata
->pagesz
= PAGE_SIZE
;
3607 rdata
->read_into_pages
= cifs_uncached_read_into_pages
;
3608 rdata
->copy_into_pages
= cifs_uncached_copy_into_pages
;
3609 rdata
->credits
= credits_on_stack
;
3611 kref_get(&ctx
->refcount
);
3613 rc
= adjust_credits(server
, &rdata
->credits
, rdata
->bytes
);
3616 if (rdata
->cfile
->invalidHandle
)
3619 rc
= server
->ops
->async_readv(rdata
);
3623 add_credits_and_wake_if(server
, &rdata
->credits
, 0);
3624 kref_put(&rdata
->refcount
,
3625 cifs_uncached_readdata_release
);
3626 if (rc
== -EAGAIN
) {
3627 iov_iter_revert(&direct_iov
, cur_len
);
3633 list_add_tail(&rdata
->list
, rdata_list
);
3642 collect_uncached_read_data(struct cifs_aio_ctx
*ctx
)
3644 struct cifs_readdata
*rdata
, *tmp
;
3645 struct iov_iter
*to
= &ctx
->iter
;
3646 struct cifs_sb_info
*cifs_sb
;
3649 cifs_sb
= CIFS_SB(ctx
->cfile
->dentry
->d_sb
);
3651 mutex_lock(&ctx
->aio_mutex
);
3653 if (list_empty(&ctx
->list
)) {
3654 mutex_unlock(&ctx
->aio_mutex
);
3659 /* the loop below should proceed in the order of increasing offsets */
3661 list_for_each_entry_safe(rdata
, tmp
, &ctx
->list
, list
) {
3663 if (!try_wait_for_completion(&rdata
->done
)) {
3664 mutex_unlock(&ctx
->aio_mutex
);
3668 if (rdata
->result
== -EAGAIN
) {
3669 /* resend call if it's a retryable error */
3670 struct list_head tmp_list
;
3671 unsigned int got_bytes
= rdata
->got_bytes
;
3673 list_del_init(&rdata
->list
);
3674 INIT_LIST_HEAD(&tmp_list
);
3677 * Got a part of data and then reconnect has
3678 * happened -- fill the buffer and continue
3681 if (got_bytes
&& got_bytes
< rdata
->bytes
) {
3683 if (!ctx
->direct_io
)
3684 rc
= cifs_readdata_to_iov(rdata
, to
);
3686 kref_put(&rdata
->refcount
,
3687 cifs_uncached_readdata_release
);
3692 if (ctx
->direct_io
) {
3694 * Re-use rdata as this is a
3697 rc
= cifs_resend_rdata(
3701 rc
= cifs_send_async_read(
3702 rdata
->offset
+ got_bytes
,
3703 rdata
->bytes
- got_bytes
,
3704 rdata
->cfile
, cifs_sb
,
3707 kref_put(&rdata
->refcount
,
3708 cifs_uncached_readdata_release
);
3711 list_splice(&tmp_list
, &ctx
->list
);
3714 } else if (rdata
->result
)
3716 else if (!ctx
->direct_io
)
3717 rc
= cifs_readdata_to_iov(rdata
, to
);
3719 /* if there was a short read -- discard anything left */
3720 if (rdata
->got_bytes
&& rdata
->got_bytes
< rdata
->bytes
)
3723 ctx
->total_len
+= rdata
->got_bytes
;
3725 list_del_init(&rdata
->list
);
3726 kref_put(&rdata
->refcount
, cifs_uncached_readdata_release
);
3729 if (!ctx
->direct_io
)
3730 ctx
->total_len
= ctx
->len
- iov_iter_count(to
);
3732 /* mask nodata case */
3736 ctx
->rc
= (rc
== 0) ? ctx
->total_len
: rc
;
3738 mutex_unlock(&ctx
->aio_mutex
);
3740 if (ctx
->iocb
&& ctx
->iocb
->ki_complete
)
3741 ctx
->iocb
->ki_complete(ctx
->iocb
, ctx
->rc
, 0);
3743 complete(&ctx
->done
);
3746 static ssize_t
__cifs_readv(
3747 struct kiocb
*iocb
, struct iov_iter
*to
, bool direct
)
3750 struct file
*file
= iocb
->ki_filp
;
3751 struct cifs_sb_info
*cifs_sb
;
3752 struct cifsFileInfo
*cfile
;
3753 struct cifs_tcon
*tcon
;
3754 ssize_t rc
, total_read
= 0;
3755 loff_t offset
= iocb
->ki_pos
;
3756 struct cifs_aio_ctx
*ctx
;
3759 * iov_iter_get_pages_alloc() doesn't work with ITER_KVEC,
3760 * fall back to data copy read path
3761 * this could be improved by getting pages directly in ITER_KVEC
3763 if (direct
&& to
->type
& ITER_KVEC
) {
3764 cifs_dbg(FYI
, "use non-direct cifs_user_readv for kvec I/O\n");
3768 len
= iov_iter_count(to
);
3772 cifs_sb
= CIFS_FILE_SB(file
);
3773 cfile
= file
->private_data
;
3774 tcon
= tlink_tcon(cfile
->tlink
);
3776 if (!tcon
->ses
->server
->ops
->async_readv
)
3779 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
3780 cifs_dbg(FYI
, "attempting read on write only file instance\n");
3782 ctx
= cifs_aio_ctx_alloc();
3786 ctx
->cfile
= cifsFileInfo_get(cfile
);
3788 if (!is_sync_kiocb(iocb
))
3791 if (iter_is_iovec(to
))
3792 ctx
->should_dirty
= true;
3796 ctx
->direct_io
= true;
3800 rc
= setup_aio_ctx_iter(ctx
, to
, READ
);
3802 kref_put(&ctx
->refcount
, cifs_aio_ctx_release
);
3808 /* grab a lock here due to read response handlers can access ctx */
3809 mutex_lock(&ctx
->aio_mutex
);
3811 rc
= cifs_send_async_read(offset
, len
, cfile
, cifs_sb
, &ctx
->list
, ctx
);
3813 /* if at least one read request send succeeded, then reset rc */
3814 if (!list_empty(&ctx
->list
))
3817 mutex_unlock(&ctx
->aio_mutex
);
3820 kref_put(&ctx
->refcount
, cifs_aio_ctx_release
);
3824 if (!is_sync_kiocb(iocb
)) {
3825 kref_put(&ctx
->refcount
, cifs_aio_ctx_release
);
3826 return -EIOCBQUEUED
;
3829 rc
= wait_for_completion_killable(&ctx
->done
);
3831 mutex_lock(&ctx
->aio_mutex
);
3832 ctx
->rc
= rc
= -EINTR
;
3833 total_read
= ctx
->total_len
;
3834 mutex_unlock(&ctx
->aio_mutex
);
3837 total_read
= ctx
->total_len
;
3840 kref_put(&ctx
->refcount
, cifs_aio_ctx_release
);
3843 iocb
->ki_pos
+= total_read
;
3849 ssize_t
cifs_direct_readv(struct kiocb
*iocb
, struct iov_iter
*to
)
3851 return __cifs_readv(iocb
, to
, true);
3854 ssize_t
cifs_user_readv(struct kiocb
*iocb
, struct iov_iter
*to
)
3856 return __cifs_readv(iocb
, to
, false);
3860 cifs_strict_readv(struct kiocb
*iocb
, struct iov_iter
*to
)
3862 struct inode
*inode
= file_inode(iocb
->ki_filp
);
3863 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
3864 struct cifs_sb_info
*cifs_sb
= CIFS_SB(inode
->i_sb
);
3865 struct cifsFileInfo
*cfile
= (struct cifsFileInfo
*)
3866 iocb
->ki_filp
->private_data
;
3867 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
3871 * In strict cache mode we need to read from the server all the time
3872 * if we don't have level II oplock because the server can delay mtime
3873 * change - so we can't make a decision about inode invalidating.
3874 * And we can also fail with pagereading if there are mandatory locks
3875 * on pages affected by this read but not on the region from pos to
3878 if (!CIFS_CACHE_READ(cinode
))
3879 return cifs_user_readv(iocb
, to
);
3881 if (cap_unix(tcon
->ses
) &&
3882 (CIFS_UNIX_FCNTL_CAP
& le64_to_cpu(tcon
->fsUnixInfo
.Capability
)) &&
3883 ((cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_NOPOSIXBRL
) == 0))
3884 return generic_file_read_iter(iocb
, to
);
3887 * We need to hold the sem to be sure nobody modifies lock list
3888 * with a brlock that prevents reading.
3890 down_read(&cinode
->lock_sem
);
3891 if (!cifs_find_lock_conflict(cfile
, iocb
->ki_pos
, iov_iter_count(to
),
3892 tcon
->ses
->server
->vals
->shared_lock_type
,
3893 0, NULL
, CIFS_READ_OP
))
3894 rc
= generic_file_read_iter(iocb
, to
);
3895 up_read(&cinode
->lock_sem
);
3900 cifs_read(struct file
*file
, char *read_data
, size_t read_size
, loff_t
*offset
)
3903 unsigned int bytes_read
= 0;
3904 unsigned int total_read
;
3905 unsigned int current_read_size
;
3907 struct cifs_sb_info
*cifs_sb
;
3908 struct cifs_tcon
*tcon
;
3909 struct TCP_Server_Info
*server
;
3912 struct cifsFileInfo
*open_file
;
3913 struct cifs_io_parms io_parms
;
3914 int buf_type
= CIFS_NO_BUFFER
;
3918 cifs_sb
= CIFS_FILE_SB(file
);
3920 /* FIXME: set up handlers for larger reads and/or convert to async */
3921 rsize
= min_t(unsigned int, cifs_sb
->rsize
, CIFSMaxBufSize
);
3923 if (file
->private_data
== NULL
) {
3928 open_file
= file
->private_data
;
3929 tcon
= tlink_tcon(open_file
->tlink
);
3930 server
= tcon
->ses
->server
;
3932 if (!server
->ops
->sync_read
) {
3937 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
3938 pid
= open_file
->pid
;
3940 pid
= current
->tgid
;
3942 if ((file
->f_flags
& O_ACCMODE
) == O_WRONLY
)
3943 cifs_dbg(FYI
, "attempting read on write only file instance\n");
3945 for (total_read
= 0, cur_offset
= read_data
; read_size
> total_read
;
3946 total_read
+= bytes_read
, cur_offset
+= bytes_read
) {
3948 current_read_size
= min_t(uint
, read_size
- total_read
,
3951 * For windows me and 9x we do not want to request more
3952 * than it negotiated since it will refuse the read
3955 if ((tcon
->ses
) && !(tcon
->ses
->capabilities
&
3956 tcon
->ses
->server
->vals
->cap_large_files
)) {
3957 current_read_size
= min_t(uint
,
3958 current_read_size
, CIFSMaxBufSize
);
3960 if (open_file
->invalidHandle
) {
3961 rc
= cifs_reopen_file(open_file
, true);
3966 io_parms
.tcon
= tcon
;
3967 io_parms
.offset
= *offset
;
3968 io_parms
.length
= current_read_size
;
3969 rc
= server
->ops
->sync_read(xid
, &open_file
->fid
, &io_parms
,
3970 &bytes_read
, &cur_offset
,
3972 } while (rc
== -EAGAIN
);
3974 if (rc
|| (bytes_read
== 0)) {
3982 cifs_stats_bytes_read(tcon
, total_read
);
3983 *offset
+= bytes_read
;
3991 * If the page is mmap'ed into a process' page tables, then we need to make
3992 * sure that it doesn't change while being written back.
3995 cifs_page_mkwrite(struct vm_fault
*vmf
)
3997 struct page
*page
= vmf
->page
;
4000 return VM_FAULT_LOCKED
;
4003 static const struct vm_operations_struct cifs_file_vm_ops
= {
4004 .fault
= filemap_fault
,
4005 .map_pages
= filemap_map_pages
,
4006 .page_mkwrite
= cifs_page_mkwrite
,
4009 int cifs_file_strict_mmap(struct file
*file
, struct vm_area_struct
*vma
)
4012 struct inode
*inode
= file_inode(file
);
4016 if (!CIFS_CACHE_READ(CIFS_I(inode
)))
4017 rc
= cifs_zap_mapping(inode
);
4019 rc
= generic_file_mmap(file
, vma
);
4021 vma
->vm_ops
= &cifs_file_vm_ops
;
4027 int cifs_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
4033 rc
= cifs_revalidate_file(file
);
4035 cifs_dbg(FYI
, "Validation prior to mmap failed, error=%d\n",
4038 rc
= generic_file_mmap(file
, vma
);
4040 vma
->vm_ops
= &cifs_file_vm_ops
;
4047 cifs_readv_complete(struct work_struct
*work
)
4049 unsigned int i
, got_bytes
;
4050 struct cifs_readdata
*rdata
= container_of(work
,
4051 struct cifs_readdata
, work
);
4053 got_bytes
= rdata
->got_bytes
;
4054 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
4055 struct page
*page
= rdata
->pages
[i
];
4057 lru_cache_add_file(page
);
4059 if (rdata
->result
== 0 ||
4060 (rdata
->result
== -EAGAIN
&& got_bytes
)) {
4061 flush_dcache_page(page
);
4062 SetPageUptodate(page
);
4067 if (rdata
->result
== 0 ||
4068 (rdata
->result
== -EAGAIN
&& got_bytes
))
4069 cifs_readpage_to_fscache(rdata
->mapping
->host
, page
);
4071 got_bytes
-= min_t(unsigned int, PAGE_SIZE
, got_bytes
);
4074 rdata
->pages
[i
] = NULL
;
4076 kref_put(&rdata
->refcount
, cifs_readdata_release
);
4080 readpages_fill_pages(struct TCP_Server_Info
*server
,
4081 struct cifs_readdata
*rdata
, struct iov_iter
*iter
,
4088 unsigned int nr_pages
= rdata
->nr_pages
;
4089 unsigned int page_offset
= rdata
->page_offset
;
4091 /* determine the eof that the server (probably) has */
4092 eof
= CIFS_I(rdata
->mapping
->host
)->server_eof
;
4093 eof_index
= eof
? (eof
- 1) >> PAGE_SHIFT
: 0;
4094 cifs_dbg(FYI
, "eof=%llu eof_index=%lu\n", eof
, eof_index
);
4096 rdata
->got_bytes
= 0;
4097 rdata
->tailsz
= PAGE_SIZE
;
4098 for (i
= 0; i
< nr_pages
; i
++) {
4099 struct page
*page
= rdata
->pages
[i
];
4100 unsigned int to_read
= rdata
->pagesz
;
4104 to_read
-= page_offset
;
4110 if (len
>= to_read
) {
4112 } else if (len
> 0) {
4113 /* enough for partial page, fill and zero the rest */
4114 zero_user(page
, len
+ page_offset
, to_read
- len
);
4115 n
= rdata
->tailsz
= len
;
4117 } else if (page
->index
> eof_index
) {
4119 * The VFS will not try to do readahead past the
4120 * i_size, but it's possible that we have outstanding
4121 * writes with gaps in the middle and the i_size hasn't
4122 * caught up yet. Populate those with zeroed out pages
4123 * to prevent the VFS from repeatedly attempting to
4124 * fill them until the writes are flushed.
4126 zero_user(page
, 0, PAGE_SIZE
);
4127 lru_cache_add_file(page
);
4128 flush_dcache_page(page
);
4129 SetPageUptodate(page
);
4132 rdata
->pages
[i
] = NULL
;
4136 /* no need to hold page hostage */
4137 lru_cache_add_file(page
);
4140 rdata
->pages
[i
] = NULL
;
4146 result
= copy_page_from_iter(
4147 page
, page_offset
, n
, iter
);
4148 #ifdef CONFIG_CIFS_SMB_DIRECT
4153 result
= cifs_read_page_from_socket(
4154 server
, page
, page_offset
, n
);
4158 rdata
->got_bytes
+= result
;
4161 return rdata
->got_bytes
> 0 && result
!= -ECONNABORTED
?
4162 rdata
->got_bytes
: result
;
4166 cifs_readpages_read_into_pages(struct TCP_Server_Info
*server
,
4167 struct cifs_readdata
*rdata
, unsigned int len
)
4169 return readpages_fill_pages(server
, rdata
, NULL
, len
);
4173 cifs_readpages_copy_into_pages(struct TCP_Server_Info
*server
,
4174 struct cifs_readdata
*rdata
,
4175 struct iov_iter
*iter
)
4177 return readpages_fill_pages(server
, rdata
, iter
, iter
->count
);
4181 readpages_get_pages(struct address_space
*mapping
, struct list_head
*page_list
,
4182 unsigned int rsize
, struct list_head
*tmplist
,
4183 unsigned int *nr_pages
, loff_t
*offset
, unsigned int *bytes
)
4185 struct page
*page
, *tpage
;
4186 unsigned int expected_index
;
4188 gfp_t gfp
= readahead_gfp_mask(mapping
);
4190 INIT_LIST_HEAD(tmplist
);
4192 page
= lru_to_page(page_list
);
4195 * Lock the page and put it in the cache. Since no one else
4196 * should have access to this page, we're safe to simply set
4197 * PG_locked without checking it first.
4199 __SetPageLocked(page
);
4200 rc
= add_to_page_cache_locked(page
, mapping
,
4203 /* give up if we can't stick it in the cache */
4205 __ClearPageLocked(page
);
4209 /* move first page to the tmplist */
4210 *offset
= (loff_t
)page
->index
<< PAGE_SHIFT
;
4213 list_move_tail(&page
->lru
, tmplist
);
4215 /* now try and add more pages onto the request */
4216 expected_index
= page
->index
+ 1;
4217 list_for_each_entry_safe_reverse(page
, tpage
, page_list
, lru
) {
4218 /* discontinuity ? */
4219 if (page
->index
!= expected_index
)
4222 /* would this page push the read over the rsize? */
4223 if (*bytes
+ PAGE_SIZE
> rsize
)
4226 __SetPageLocked(page
);
4227 if (add_to_page_cache_locked(page
, mapping
, page
->index
, gfp
)) {
4228 __ClearPageLocked(page
);
4231 list_move_tail(&page
->lru
, tmplist
);
4232 (*bytes
) += PAGE_SIZE
;
4239 static int cifs_readpages(struct file
*file
, struct address_space
*mapping
,
4240 struct list_head
*page_list
, unsigned num_pages
)
4243 struct list_head tmplist
;
4244 struct cifsFileInfo
*open_file
= file
->private_data
;
4245 struct cifs_sb_info
*cifs_sb
= CIFS_FILE_SB(file
);
4246 struct TCP_Server_Info
*server
;
4252 * Reads as many pages as possible from fscache. Returns -ENOBUFS
4253 * immediately if the cookie is negative
4255 * After this point, every page in the list might have PG_fscache set,
4256 * so we will need to clean that up off of every page we don't use.
4258 rc
= cifs_readpages_from_fscache(mapping
->host
, mapping
, page_list
,
4265 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_RWPIDFORWARD
)
4266 pid
= open_file
->pid
;
4268 pid
= current
->tgid
;
4271 server
= tlink_tcon(open_file
->tlink
)->ses
->server
;
4273 cifs_dbg(FYI
, "%s: file=%p mapping=%p num_pages=%u\n",
4274 __func__
, file
, mapping
, num_pages
);
4277 * Start with the page at end of list and move it to private
4278 * list. Do the same with any following pages until we hit
4279 * the rsize limit, hit an index discontinuity, or run out of
4280 * pages. Issue the async read and then start the loop again
4281 * until the list is empty.
4283 * Note that list order is important. The page_list is in
4284 * the order of declining indexes. When we put the pages in
4285 * the rdata->pages, then we want them in increasing order.
4287 while (!list_empty(page_list
)) {
4288 unsigned int i
, nr_pages
, bytes
, rsize
;
4290 struct page
*page
, *tpage
;
4291 struct cifs_readdata
*rdata
;
4292 struct cifs_credits credits_on_stack
;
4293 struct cifs_credits
*credits
= &credits_on_stack
;
4295 if (open_file
->invalidHandle
) {
4296 rc
= cifs_reopen_file(open_file
, true);
4303 rc
= server
->ops
->wait_mtu_credits(server
, cifs_sb
->rsize
,
4309 * Give up immediately if rsize is too small to read an entire
4310 * page. The VFS will fall back to readpage. We should never
4311 * reach this point however since we set ra_pages to 0 when the
4312 * rsize is smaller than a cache page.
4314 if (unlikely(rsize
< PAGE_SIZE
)) {
4315 add_credits_and_wake_if(server
, credits
, 0);
4320 rc
= readpages_get_pages(mapping
, page_list
, rsize
, &tmplist
,
4321 &nr_pages
, &offset
, &bytes
);
4323 add_credits_and_wake_if(server
, credits
, 0);
4327 rdata
= cifs_readdata_alloc(nr_pages
, cifs_readv_complete
);
4329 /* best to give up if we're out of mem */
4330 list_for_each_entry_safe(page
, tpage
, &tmplist
, lru
) {
4331 list_del(&page
->lru
);
4332 lru_cache_add_file(page
);
4337 add_credits_and_wake_if(server
, credits
, 0);
4341 rdata
->cfile
= cifsFileInfo_get(open_file
);
4342 rdata
->mapping
= mapping
;
4343 rdata
->offset
= offset
;
4344 rdata
->bytes
= bytes
;
4346 rdata
->pagesz
= PAGE_SIZE
;
4347 rdata
->tailsz
= PAGE_SIZE
;
4348 rdata
->read_into_pages
= cifs_readpages_read_into_pages
;
4349 rdata
->copy_into_pages
= cifs_readpages_copy_into_pages
;
4350 rdata
->credits
= credits_on_stack
;
4352 list_for_each_entry_safe(page
, tpage
, &tmplist
, lru
) {
4353 list_del(&page
->lru
);
4354 rdata
->pages
[rdata
->nr_pages
++] = page
;
4357 rc
= adjust_credits(server
, &rdata
->credits
, rdata
->bytes
);
4360 if (rdata
->cfile
->invalidHandle
)
4363 rc
= server
->ops
->async_readv(rdata
);
4367 add_credits_and_wake_if(server
, &rdata
->credits
, 0);
4368 for (i
= 0; i
< rdata
->nr_pages
; i
++) {
4369 page
= rdata
->pages
[i
];
4370 lru_cache_add_file(page
);
4374 /* Fallback to the readpage in error/reconnect cases */
4375 kref_put(&rdata
->refcount
, cifs_readdata_release
);
4379 kref_put(&rdata
->refcount
, cifs_readdata_release
);
4382 /* Any pages that have been shown to fscache but didn't get added to
4383 * the pagecache must be uncached before they get returned to the
4386 cifs_fscache_readpages_cancel(mapping
->host
, page_list
);
4392 * cifs_readpage_worker must be called with the page pinned
4394 static int cifs_readpage_worker(struct file
*file
, struct page
*page
,
4400 /* Is the page cached? */
4401 rc
= cifs_readpage_from_fscache(file_inode(file
), page
);
4405 read_data
= kmap(page
);
4406 /* for reads over a certain size could initiate async read ahead */
4408 rc
= cifs_read(file
, read_data
, PAGE_SIZE
, poffset
);
4413 cifs_dbg(FYI
, "Bytes read %d\n", rc
);
4415 /* we do not want atime to be less than mtime, it broke some apps */
4416 file_inode(file
)->i_atime
= current_time(file_inode(file
));
4417 if (timespec64_compare(&(file_inode(file
)->i_atime
), &(file_inode(file
)->i_mtime
)))
4418 file_inode(file
)->i_atime
= file_inode(file
)->i_mtime
;
4420 file_inode(file
)->i_atime
= current_time(file_inode(file
));
4423 memset(read_data
+ rc
, 0, PAGE_SIZE
- rc
);
4425 flush_dcache_page(page
);
4426 SetPageUptodate(page
);
4428 /* send this page to the cache */
4429 cifs_readpage_to_fscache(file_inode(file
), page
);
4441 static int cifs_readpage(struct file
*file
, struct page
*page
)
4443 loff_t offset
= (loff_t
)page
->index
<< PAGE_SHIFT
;
4449 if (file
->private_data
== NULL
) {
4455 cifs_dbg(FYI
, "readpage %p at offset %d 0x%x\n",
4456 page
, (int)offset
, (int)offset
);
4458 rc
= cifs_readpage_worker(file
, page
, &offset
);
4464 static int is_inode_writable(struct cifsInodeInfo
*cifs_inode
)
4466 struct cifsFileInfo
*open_file
;
4468 spin_lock(&cifs_inode
->open_file_lock
);
4469 list_for_each_entry(open_file
, &cifs_inode
->openFileList
, flist
) {
4470 if (OPEN_FMODE(open_file
->f_flags
) & FMODE_WRITE
) {
4471 spin_unlock(&cifs_inode
->open_file_lock
);
4475 spin_unlock(&cifs_inode
->open_file_lock
);
4479 /* We do not want to update the file size from server for inodes
4480 open for write - to avoid races with writepage extending
4481 the file - in the future we could consider allowing
4482 refreshing the inode only on increases in the file size
4483 but this is tricky to do without racing with writebehind
4484 page caching in the current Linux kernel design */
4485 bool is_size_safe_to_change(struct cifsInodeInfo
*cifsInode
, __u64 end_of_file
)
4490 if (is_inode_writable(cifsInode
)) {
4491 /* This inode is open for write at least once */
4492 struct cifs_sb_info
*cifs_sb
;
4494 cifs_sb
= CIFS_SB(cifsInode
->vfs_inode
.i_sb
);
4495 if (cifs_sb
->mnt_cifs_flags
& CIFS_MOUNT_DIRECT_IO
) {
4496 /* since no page cache to corrupt on directio
4497 we can change size safely */
4501 if (i_size_read(&cifsInode
->vfs_inode
) < end_of_file
)
4509 static int cifs_write_begin(struct file
*file
, struct address_space
*mapping
,
4510 loff_t pos
, unsigned len
, unsigned flags
,
4511 struct page
**pagep
, void **fsdata
)
4514 pgoff_t index
= pos
>> PAGE_SHIFT
;
4515 loff_t offset
= pos
& (PAGE_SIZE
- 1);
4516 loff_t page_start
= pos
& PAGE_MASK
;
4521 cifs_dbg(FYI
, "write_begin from %lld len %d\n", (long long)pos
, len
);
4524 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
4530 if (PageUptodate(page
))
4534 * If we write a full page it will be up to date, no need to read from
4535 * the server. If the write is short, we'll end up doing a sync write
4538 if (len
== PAGE_SIZE
)
4542 * optimize away the read when we have an oplock, and we're not
4543 * expecting to use any of the data we'd be reading in. That
4544 * is, when the page lies beyond the EOF, or straddles the EOF
4545 * and the write will cover all of the existing data.
4547 if (CIFS_CACHE_READ(CIFS_I(mapping
->host
))) {
4548 i_size
= i_size_read(mapping
->host
);
4549 if (page_start
>= i_size
||
4550 (offset
== 0 && (pos
+ len
) >= i_size
)) {
4551 zero_user_segments(page
, 0, offset
,
4555 * PageChecked means that the parts of the page
4556 * to which we're not writing are considered up
4557 * to date. Once the data is copied to the
4558 * page, it can be set uptodate.
4560 SetPageChecked(page
);
4565 if ((file
->f_flags
& O_ACCMODE
) != O_WRONLY
&& !oncethru
) {
4567 * might as well read a page, it is fast enough. If we get
4568 * an error, we don't need to return it. cifs_write_end will
4569 * do a sync write instead since PG_uptodate isn't set.
4571 cifs_readpage_worker(file
, page
, &page_start
);
4576 /* we could try using another file handle if there is one -
4577 but how would we lock it to prevent close of that handle
4578 racing with this read? In any case
4579 this will be written out by write_end so is fine */
4586 static int cifs_release_page(struct page
*page
, gfp_t gfp
)
4588 if (PagePrivate(page
))
4591 return cifs_fscache_release_page(page
, gfp
);
4594 static void cifs_invalidate_page(struct page
*page
, unsigned int offset
,
4595 unsigned int length
)
4597 struct cifsInodeInfo
*cifsi
= CIFS_I(page
->mapping
->host
);
4599 if (offset
== 0 && length
== PAGE_SIZE
)
4600 cifs_fscache_invalidate_page(page
, &cifsi
->vfs_inode
);
4603 static int cifs_launder_page(struct page
*page
)
4606 loff_t range_start
= page_offset(page
);
4607 loff_t range_end
= range_start
+ (loff_t
)(PAGE_SIZE
- 1);
4608 struct writeback_control wbc
= {
4609 .sync_mode
= WB_SYNC_ALL
,
4611 .range_start
= range_start
,
4612 .range_end
= range_end
,
4615 cifs_dbg(FYI
, "Launder page: %p\n", page
);
4617 if (clear_page_dirty_for_io(page
))
4618 rc
= cifs_writepage_locked(page
, &wbc
);
4620 cifs_fscache_invalidate_page(page
, page
->mapping
->host
);
4624 void cifs_oplock_break(struct work_struct
*work
)
4626 struct cifsFileInfo
*cfile
= container_of(work
, struct cifsFileInfo
,
4628 struct inode
*inode
= d_inode(cfile
->dentry
);
4629 struct cifsInodeInfo
*cinode
= CIFS_I(inode
);
4630 struct cifs_tcon
*tcon
= tlink_tcon(cfile
->tlink
);
4631 struct TCP_Server_Info
*server
= tcon
->ses
->server
;
4634 wait_on_bit(&cinode
->flags
, CIFS_INODE_PENDING_WRITERS
,
4635 TASK_UNINTERRUPTIBLE
);
4637 server
->ops
->downgrade_oplock(server
, cinode
,
4638 test_bit(CIFS_INODE_DOWNGRADE_OPLOCK_TO_L2
, &cinode
->flags
));
4640 if (!CIFS_CACHE_WRITE(cinode
) && CIFS_CACHE_READ(cinode
) &&
4641 cifs_has_mand_locks(cinode
)) {
4642 cifs_dbg(FYI
, "Reset oplock to None for inode=%p due to mand locks\n",
4647 if (inode
&& S_ISREG(inode
->i_mode
)) {
4648 if (CIFS_CACHE_READ(cinode
))
4649 break_lease(inode
, O_RDONLY
);
4651 break_lease(inode
, O_WRONLY
);
4652 rc
= filemap_fdatawrite(inode
->i_mapping
);
4653 if (!CIFS_CACHE_READ(cinode
)) {
4654 rc
= filemap_fdatawait(inode
->i_mapping
);
4655 mapping_set_error(inode
->i_mapping
, rc
);
4656 cifs_zap_mapping(inode
);
4658 cifs_dbg(FYI
, "Oplock flush inode %p rc %d\n", inode
, rc
);
4661 rc
= cifs_push_locks(cfile
);
4663 cifs_dbg(VFS
, "Push locks rc = %d\n", rc
);
4666 * releasing stale oplock after recent reconnect of smb session using
4667 * a now incorrect file handle is not a data integrity issue but do
4668 * not bother sending an oplock release if session to server still is
4669 * disconnected since oplock already released by the server
4671 if (!cfile
->oplock_break_cancelled
) {
4672 rc
= tcon
->ses
->server
->ops
->oplock_response(tcon
, &cfile
->fid
,
4674 cifs_dbg(FYI
, "Oplock release rc = %d\n", rc
);
4676 _cifsFileInfo_put(cfile
, false /* do not wait for ourself */);
4677 cifs_done_oplock_break(cinode
);
4681 * The presence of cifs_direct_io() in the address space ops vector
4682 * allowes open() O_DIRECT flags which would have failed otherwise.
4684 * In the non-cached mode (mount with cache=none), we shunt off direct read and write requests
4685 * so this method should never be called.
4687 * Direct IO is not yet supported in the cached mode.
4690 cifs_direct_io(struct kiocb
*iocb
, struct iov_iter
*iter
)
4694 * Eventually need to support direct IO for non forcedirectio mounts
4700 const struct address_space_operations cifs_addr_ops
= {
4701 .readpage
= cifs_readpage
,
4702 .readpages
= cifs_readpages
,
4703 .writepage
= cifs_writepage
,
4704 .writepages
= cifs_writepages
,
4705 .write_begin
= cifs_write_begin
,
4706 .write_end
= cifs_write_end
,
4707 .set_page_dirty
= __set_page_dirty_nobuffers
,
4708 .releasepage
= cifs_release_page
,
4709 .direct_IO
= cifs_direct_io
,
4710 .invalidatepage
= cifs_invalidate_page
,
4711 .launder_page
= cifs_launder_page
,
4715 * cifs_readpages requires the server to support a buffer large enough to
4716 * contain the header plus one complete page of data. Otherwise, we need
4717 * to leave cifs_readpages out of the address space operations.
4719 const struct address_space_operations cifs_addr_ops_smallbuf
= {
4720 .readpage
= cifs_readpage
,
4721 .writepage
= cifs_writepage
,
4722 .writepages
= cifs_writepages
,
4723 .write_begin
= cifs_write_begin
,
4724 .write_end
= cifs_write_end
,
4725 .set_page_dirty
= __set_page_dirty_nobuffers
,
4726 .releasepage
= cifs_release_page
,
4727 .invalidatepage
= cifs_invalidate_page
,
4728 .launder_page
= cifs_launder_page
,