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 <asm/div64.h>
39 #include "cifsproto.h"
40 #include "cifs_unicode.h"
41 #include "cifs_debug.h"
42 #include "cifs_fs_sb.h"
45 static inline int cifs_convert_flags(unsigned int flags)
47 if ((flags & O_ACCMODE) == O_RDONLY)
49 else if ((flags & O_ACCMODE) == O_WRONLY)
51 else if ((flags & O_ACCMODE) == O_RDWR) {
52 /* GENERIC_ALL is too much permission to request
53 can cause unnecessary access denied on create */
54 /* return GENERIC_ALL; */
55 return (GENERIC_READ | GENERIC_WRITE);
58 return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
59 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
63 static u32 cifs_posix_convert_flags(unsigned int flags)
67 if ((flags & O_ACCMODE) == O_RDONLY)
68 posix_flags = SMB_O_RDONLY;
69 else if ((flags & O_ACCMODE) == O_WRONLY)
70 posix_flags = SMB_O_WRONLY;
71 else if ((flags & O_ACCMODE) == O_RDWR)
72 posix_flags = SMB_O_RDWR;
75 posix_flags |= SMB_O_CREAT;
77 posix_flags |= SMB_O_EXCL;
79 posix_flags |= SMB_O_TRUNC;
80 /* be safe and imply O_SYNC for O_DSYNC */
82 posix_flags |= SMB_O_SYNC;
83 if (flags & O_DIRECTORY)
84 posix_flags |= SMB_O_DIRECTORY;
85 if (flags & O_NOFOLLOW)
86 posix_flags |= SMB_O_NOFOLLOW;
88 posix_flags |= SMB_O_DIRECT;
93 static inline int cifs_get_disposition(unsigned int flags)
95 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
97 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
98 return FILE_OVERWRITE_IF;
99 else if ((flags & O_CREAT) == O_CREAT)
101 else if ((flags & O_TRUNC) == O_TRUNC)
102 return FILE_OVERWRITE;
107 static inline int cifs_open_inode_helper(struct inode *inode,
108 struct cifsTconInfo *pTcon, __u32 oplock, FILE_ALL_INFO *buf,
109 char *full_path, int xid)
111 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
112 struct timespec temp;
115 if (pCifsInode->clientCanCacheRead) {
116 /* we have the inode open somewhere else
117 no need to discard cache data */
118 goto client_can_cache;
121 /* BB need same check in cifs_create too? */
122 /* if not oplocked, invalidate inode pages if mtime or file
124 temp = cifs_NTtimeToUnix(buf->LastWriteTime);
125 if (timespec_equal(&inode->i_mtime, &temp) &&
127 (loff_t)le64_to_cpu(buf->EndOfFile))) {
128 cFYI(1, "inode unchanged on server");
130 if (inode->i_mapping) {
131 /* BB no need to lock inode until after invalidate
132 since namei code should already have it locked? */
133 rc = filemap_write_and_wait(inode->i_mapping);
135 pCifsInode->write_behind_rc = rc;
137 cFYI(1, "invalidating remote inode since open detected it "
139 invalidate_remote_inode(inode);
144 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
147 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
150 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
151 pCifsInode->clientCanCacheAll = true;
152 pCifsInode->clientCanCacheRead = true;
153 cFYI(1, "Exclusive Oplock granted on inode %p", inode);
154 } else if ((oplock & 0xF) == OPLOCK_READ)
155 pCifsInode->clientCanCacheRead = true;
160 int cifs_posix_open(char *full_path, struct inode **pinode,
161 struct super_block *sb, int mode, unsigned int f_flags,
162 __u32 *poplock, __u16 *pnetfid, int xid)
165 FILE_UNIX_BASIC_INFO *presp_data;
166 __u32 posix_flags = 0;
167 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
168 struct cifs_fattr fattr;
169 struct tcon_link *tlink;
170 struct cifsTconInfo *tcon;
172 cFYI(1, "posix open %s", full_path);
174 presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
175 if (presp_data == NULL)
178 tlink = cifs_sb_tlink(cifs_sb);
184 tcon = tlink_tcon(tlink);
185 mode &= ~current_umask();
187 posix_flags = cifs_posix_convert_flags(f_flags);
188 rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
189 poplock, full_path, cifs_sb->local_nls,
190 cifs_sb->mnt_cifs_flags &
191 CIFS_MOUNT_MAP_SPECIAL_CHR);
192 cifs_put_tlink(tlink);
197 if (presp_data->Type == cpu_to_le32(-1))
198 goto posix_open_ret; /* open ok, caller does qpathinfo */
201 goto posix_open_ret; /* caller does not need info */
203 cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
205 /* get new inode and set it up */
206 if (*pinode == NULL) {
207 cifs_fill_uniqueid(sb, &fattr);
208 *pinode = cifs_iget(sb, &fattr);
214 cifs_fattr_to_inode(*pinode, &fattr);
222 int cifs_open(struct inode *inode, struct file *file)
227 struct cifs_sb_info *cifs_sb;
228 struct cifsTconInfo *tcon;
229 struct tcon_link *tlink;
230 struct cifsFileInfo *pCifsFile = NULL;
231 struct cifsInodeInfo *pCifsInode;
232 char *full_path = NULL;
236 FILE_ALL_INFO *buf = NULL;
240 cifs_sb = CIFS_SB(inode->i_sb);
241 tlink = cifs_sb_tlink(cifs_sb);
244 return PTR_ERR(tlink);
246 tcon = tlink_tcon(tlink);
248 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
250 full_path = build_path_from_dentry(file->f_path.dentry);
251 if (full_path == NULL) {
256 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
257 inode, file->f_flags, full_path);
264 if (!tcon->broken_posix_open && tcon->unix_ext &&
265 (tcon->ses->capabilities & CAP_UNIX) &&
266 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
267 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
268 /* can not refresh inode info since size could be stale */
269 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
270 cifs_sb->mnt_file_mode /* ignored */,
271 file->f_flags, &oplock, &netfid, xid);
273 cFYI(1, "posix open succeeded");
275 pCifsFile = cifs_new_fileinfo(inode, netfid, file,
277 if (pCifsFile == NULL) {
278 CIFSSMBClose(xid, tcon, netfid);
282 cifs_fscache_set_inode_cookie(inode, file);
285 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
286 if (tcon->ses->serverNOS)
287 cERROR(1, "server %s of type %s returned"
288 " unexpected error on SMB posix open"
289 ", disabling posix open support."
290 " Check if server update available.",
291 tcon->ses->serverName,
292 tcon->ses->serverNOS);
293 tcon->broken_posix_open = true;
294 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
295 (rc != -EOPNOTSUPP)) /* path not found or net err */
297 /* else fallthrough to retry open the old way on network i/o
301 desiredAccess = cifs_convert_flags(file->f_flags);
303 /*********************************************************************
304 * open flag mapping table:
306 * POSIX Flag CIFS Disposition
307 * ---------- ----------------
308 * O_CREAT FILE_OPEN_IF
309 * O_CREAT | O_EXCL FILE_CREATE
310 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
311 * O_TRUNC FILE_OVERWRITE
312 * none of the above FILE_OPEN
314 * Note that there is not a direct match between disposition
315 * FILE_SUPERSEDE (ie create whether or not file exists although
316 * O_CREAT | O_TRUNC is similar but truncates the existing
317 * file rather than creating a new file as FILE_SUPERSEDE does
318 * (which uses the attributes / metadata passed in on open call)
320 *? O_SYNC is a reasonable match to CIFS writethrough flag
321 *? and the read write flags match reasonably. O_LARGEFILE
322 *? is irrelevant because largefile support is always used
323 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
324 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
325 *********************************************************************/
327 disposition = cifs_get_disposition(file->f_flags);
329 /* BB pass O_SYNC flag through on file attributes .. BB */
331 /* Also refresh inode by passing in file_info buf returned by SMBOpen
332 and calling get_inode_info with returned buf (at least helps
333 non-Unix server case) */
335 /* BB we can not do this if this is the second open of a file
336 and the first handle has writebehind data, we might be
337 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
338 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
344 if (tcon->ses->capabilities & CAP_NT_SMBS)
345 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
346 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
347 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
348 & CIFS_MOUNT_MAP_SPECIAL_CHR);
350 rc = -EIO; /* no NT SMB support fall into legacy open below */
353 /* Old server, try legacy style OpenX */
354 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
355 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
356 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
357 & CIFS_MOUNT_MAP_SPECIAL_CHR);
360 cFYI(1, "cifs_open returned 0x%x", rc);
364 rc = cifs_open_inode_helper(inode, tcon, oplock, buf, full_path, xid);
368 pCifsFile = cifs_new_fileinfo(inode, netfid, file, tlink, oplock);
369 if (pCifsFile == NULL) {
374 cifs_fscache_set_inode_cookie(inode, file);
376 if (oplock & CIFS_CREATE_ACTION) {
377 /* time to set mode which we can not set earlier due to
378 problems creating new read-only files */
379 if (tcon->unix_ext) {
380 struct cifs_unix_set_info_args args = {
381 .mode = inode->i_mode,
384 .ctime = NO_CHANGE_64,
385 .atime = NO_CHANGE_64,
386 .mtime = NO_CHANGE_64,
389 CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
391 cifs_sb->mnt_cifs_flags &
392 CIFS_MOUNT_MAP_SPECIAL_CHR);
400 cifs_put_tlink(tlink);
404 /* Try to reacquire byte range locks that were released when session */
405 /* to server was lost */
406 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
410 /* BB list all locks open on this file and relock */
415 static int cifs_reopen_file(struct file *file, bool can_flush)
420 struct cifs_sb_info *cifs_sb;
421 struct cifsTconInfo *tcon;
422 struct cifsFileInfo *pCifsFile;
423 struct cifsInodeInfo *pCifsInode;
425 char *full_path = NULL;
427 int disposition = FILE_OPEN;
430 if (file->private_data)
431 pCifsFile = file->private_data;
436 mutex_lock(&pCifsFile->fh_mutex);
437 if (!pCifsFile->invalidHandle) {
438 mutex_unlock(&pCifsFile->fh_mutex);
444 if (file->f_path.dentry == NULL) {
445 cERROR(1, "no valid name if dentry freed");
448 goto reopen_error_exit;
451 inode = file->f_path.dentry->d_inode;
453 cERROR(1, "inode not valid");
456 goto reopen_error_exit;
459 cifs_sb = CIFS_SB(inode->i_sb);
460 tcon = tlink_tcon(pCifsFile->tlink);
462 /* can not grab rename sem here because various ops, including
463 those that already have the rename sem can end up causing writepage
464 to get called and if the server was down that means we end up here,
465 and we can never tell if the caller already has the rename_sem */
466 full_path = build_path_from_dentry(file->f_path.dentry);
467 if (full_path == NULL) {
470 mutex_unlock(&pCifsFile->fh_mutex);
475 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
476 inode, file->f_flags, full_path);
483 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
484 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
485 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
488 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
489 * original open. Must mask them off for a reopen.
491 unsigned int oflags = file->f_flags & ~(O_CREAT|O_EXCL|O_TRUNC);
493 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
494 cifs_sb->mnt_file_mode /* ignored */,
495 oflags, &oplock, &netfid, xid);
497 cFYI(1, "posix reopen succeeded");
500 /* fallthrough to retry open the old way on errors, especially
501 in the reconnect path it is important to retry hard */
504 desiredAccess = cifs_convert_flags(file->f_flags);
506 /* Can not refresh inode by passing in file_info buf to be returned
507 by SMBOpen and then calling get_inode_info with returned buf
508 since file might have write behind data that needs to be flushed
509 and server version of file size can be stale. If we knew for sure
510 that inode was not dirty locally we could do this */
512 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
513 CREATE_NOT_DIR, &netfid, &oplock, NULL,
514 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
515 CIFS_MOUNT_MAP_SPECIAL_CHR);
517 mutex_unlock(&pCifsFile->fh_mutex);
518 cFYI(1, "cifs_open returned 0x%x", rc);
519 cFYI(1, "oplock: %d", oplock);
522 pCifsFile->netfid = netfid;
523 pCifsFile->invalidHandle = false;
524 mutex_unlock(&pCifsFile->fh_mutex);
525 pCifsInode = CIFS_I(inode);
528 rc = filemap_write_and_wait(inode->i_mapping);
530 CIFS_I(inode)->write_behind_rc = rc;
531 /* temporarily disable caching while we
532 go to server to get inode info */
533 pCifsInode->clientCanCacheAll = false;
534 pCifsInode->clientCanCacheRead = false;
536 rc = cifs_get_inode_info_unix(&inode,
537 full_path, inode->i_sb, xid);
539 rc = cifs_get_inode_info(&inode,
540 full_path, NULL, inode->i_sb,
542 } /* else we are writing out data to server already
543 and could deadlock if we tried to flush data, and
544 since we do not know if we have data that would
545 invalidate the current end of file on the server
546 we can not go to the server to get the new inod
548 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
549 pCifsInode->clientCanCacheAll = true;
550 pCifsInode->clientCanCacheRead = true;
551 cFYI(1, "Exclusive Oplock granted on inode %p",
552 file->f_path.dentry->d_inode);
553 } else if ((oplock & 0xF) == OPLOCK_READ) {
554 pCifsInode->clientCanCacheRead = true;
555 pCifsInode->clientCanCacheAll = false;
557 pCifsInode->clientCanCacheRead = false;
558 pCifsInode->clientCanCacheAll = false;
560 cifs_relock_file(pCifsFile);
568 int cifs_close(struct inode *inode, struct file *file)
572 struct cifs_sb_info *cifs_sb;
573 struct cifsTconInfo *pTcon;
574 struct cifsFileInfo *pSMBFile = file->private_data;
578 cifs_sb = CIFS_SB(inode->i_sb);
579 pTcon = tlink_tcon(pSMBFile->tlink);
581 struct cifsLockInfo *li, *tmp;
582 write_lock(&GlobalSMBSeslock);
583 pSMBFile->closePend = true;
585 /* no sense reconnecting to close a file that is
587 if (!pTcon->need_reconnect) {
588 write_unlock(&GlobalSMBSeslock);
590 while ((atomic_read(&pSMBFile->count) != 1)
591 && (timeout <= 2048)) {
592 /* Give write a better chance to get to
593 server ahead of the close. We do not
594 want to add a wait_q here as it would
595 increase the memory utilization as
596 the struct would be in each open file,
597 but this should give enough time to
599 cFYI(DBG2, "close delay, write pending");
603 if (!pTcon->need_reconnect &&
604 !pSMBFile->invalidHandle)
605 rc = CIFSSMBClose(xid, pTcon,
608 write_unlock(&GlobalSMBSeslock);
610 write_unlock(&GlobalSMBSeslock);
612 /* Delete any outstanding lock records.
613 We'll lose them when the file is closed anyway. */
614 mutex_lock(&pSMBFile->lock_mutex);
615 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
616 list_del(&li->llist);
619 mutex_unlock(&pSMBFile->lock_mutex);
621 write_lock(&GlobalSMBSeslock);
622 list_del(&pSMBFile->flist);
623 list_del(&pSMBFile->tlist);
624 write_unlock(&GlobalSMBSeslock);
625 cifsFileInfo_put(file->private_data);
626 file->private_data = NULL;
630 read_lock(&GlobalSMBSeslock);
631 if (list_empty(&(CIFS_I(inode)->openFileList))) {
632 cFYI(1, "closing last open instance for inode %p", inode);
633 /* if the file is not open we do not know if we can cache info
634 on this inode, much less write behind and read ahead */
635 CIFS_I(inode)->clientCanCacheRead = false;
636 CIFS_I(inode)->clientCanCacheAll = false;
638 read_unlock(&GlobalSMBSeslock);
639 if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
640 rc = CIFS_I(inode)->write_behind_rc;
645 int cifs_closedir(struct inode *inode, struct file *file)
649 struct cifsFileInfo *pCFileStruct = file->private_data;
652 cFYI(1, "Closedir inode = 0x%p", inode);
657 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
659 cFYI(1, "Freeing private data in close dir");
660 write_lock(&GlobalSMBSeslock);
661 if (!pCFileStruct->srch_inf.endOfSearch &&
662 !pCFileStruct->invalidHandle) {
663 pCFileStruct->invalidHandle = true;
664 write_unlock(&GlobalSMBSeslock);
665 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
666 cFYI(1, "Closing uncompleted readdir with rc %d",
668 /* not much we can do if it fails anyway, ignore rc */
671 write_unlock(&GlobalSMBSeslock);
672 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
674 cFYI(1, "closedir free smb buf in srch struct");
675 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
676 if (pCFileStruct->srch_inf.smallBuf)
677 cifs_small_buf_release(ptmp);
679 cifs_buf_release(ptmp);
681 cifs_put_tlink(pCFileStruct->tlink);
682 kfree(file->private_data);
683 file->private_data = NULL;
685 /* BB can we lock the filestruct while this is going on? */
690 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
691 __u64 offset, __u8 lockType)
693 struct cifsLockInfo *li =
694 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
700 mutex_lock(&fid->lock_mutex);
701 list_add(&li->llist, &fid->llist);
702 mutex_unlock(&fid->lock_mutex);
706 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
712 bool wait_flag = false;
713 struct cifs_sb_info *cifs_sb;
714 struct cifsTconInfo *tcon;
716 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
717 bool posix_locking = 0;
719 length = 1 + pfLock->fl_end - pfLock->fl_start;
723 cFYI(1, "Lock parm: 0x%x flockflags: "
724 "0x%x flocktype: 0x%x start: %lld end: %lld",
725 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
728 if (pfLock->fl_flags & FL_POSIX)
730 if (pfLock->fl_flags & FL_FLOCK)
732 if (pfLock->fl_flags & FL_SLEEP) {
733 cFYI(1, "Blocking lock");
736 if (pfLock->fl_flags & FL_ACCESS)
737 cFYI(1, "Process suspended by mandatory locking - "
738 "not implemented yet");
739 if (pfLock->fl_flags & FL_LEASE)
740 cFYI(1, "Lease on file - not implemented yet");
741 if (pfLock->fl_flags &
742 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
743 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
745 if (pfLock->fl_type == F_WRLCK) {
748 } else if (pfLock->fl_type == F_UNLCK) {
751 /* Check if unlock includes more than
753 } else if (pfLock->fl_type == F_RDLCK) {
755 lockType |= LOCKING_ANDX_SHARED_LOCK;
757 } else if (pfLock->fl_type == F_EXLCK) {
760 } else if (pfLock->fl_type == F_SHLCK) {
762 lockType |= LOCKING_ANDX_SHARED_LOCK;
765 cFYI(1, "Unknown type of lock");
767 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
768 tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
770 if (file->private_data == NULL) {
775 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
777 if ((tcon->ses->capabilities & CAP_UNIX) &&
778 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
779 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
781 /* BB add code here to normalize offset and length to
782 account for negative length which we can not accept over the
787 if (lockType & LOCKING_ANDX_SHARED_LOCK)
788 posix_lock_type = CIFS_RDLCK;
790 posix_lock_type = CIFS_WRLCK;
791 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
793 posix_lock_type, wait_flag);
798 /* BB we could chain these into one lock request BB */
799 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
800 0, 1, lockType, 0 /* wait flag */ );
802 rc = CIFSSMBLock(xid, tcon, netfid, length,
803 pfLock->fl_start, 1 /* numUnlock */ ,
804 0 /* numLock */ , lockType,
806 pfLock->fl_type = F_UNLCK;
808 cERROR(1, "Error unlocking previously locked "
809 "range %d during test of lock", rc);
813 /* if rc == ERR_SHARING_VIOLATION ? */
816 if (lockType & LOCKING_ANDX_SHARED_LOCK) {
817 pfLock->fl_type = F_WRLCK;
819 rc = CIFSSMBLock(xid, tcon, netfid, length,
820 pfLock->fl_start, 0, 1,
821 lockType | LOCKING_ANDX_SHARED_LOCK,
824 rc = CIFSSMBLock(xid, tcon, netfid,
825 length, pfLock->fl_start, 1, 0,
827 LOCKING_ANDX_SHARED_LOCK,
829 pfLock->fl_type = F_RDLCK;
831 cERROR(1, "Error unlocking "
832 "previously locked range %d "
833 "during test of lock", rc);
836 pfLock->fl_type = F_WRLCK;
846 if (!numLock && !numUnlock) {
847 /* if no lock or unlock then nothing
848 to do since we do not know what it is */
855 if (lockType & LOCKING_ANDX_SHARED_LOCK)
856 posix_lock_type = CIFS_RDLCK;
858 posix_lock_type = CIFS_WRLCK;
861 posix_lock_type = CIFS_UNLCK;
863 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
865 posix_lock_type, wait_flag);
867 struct cifsFileInfo *fid = file->private_data;
870 rc = CIFSSMBLock(xid, tcon, netfid, length,
872 0, numLock, lockType, wait_flag);
875 /* For Windows locks we must store them. */
876 rc = store_file_lock(fid, length,
877 pfLock->fl_start, lockType);
879 } else if (numUnlock) {
880 /* For each stored lock that this unlock overlaps
881 completely, unlock it. */
883 struct cifsLockInfo *li, *tmp;
886 mutex_lock(&fid->lock_mutex);
887 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
888 if (pfLock->fl_start <= li->offset &&
889 (pfLock->fl_start + length) >=
890 (li->offset + li->length)) {
891 stored_rc = CIFSSMBLock(xid, tcon,
893 li->length, li->offset,
894 1, 0, li->type, false);
898 list_del(&li->llist);
903 mutex_unlock(&fid->lock_mutex);
907 if (pfLock->fl_flags & FL_POSIX)
908 posix_lock_file_wait(file, pfLock);
914 * Set the timeout on write requests past EOF. For some servers (Windows)
915 * these calls can be very long.
917 * If we're writing >10M past the EOF we give a 180s timeout. Anything less
918 * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
919 * The 10M cutoff is totally arbitrary. A better scheme for this would be
920 * welcome if someone wants to suggest one.
922 * We may be able to do a better job with this if there were some way to
923 * declare that a file should be sparse.
926 cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
928 if (offset <= cifsi->server_eof)
930 else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
931 return CIFS_VLONG_OP;
936 /* update the file size (if needed) after a write */
938 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
939 unsigned int bytes_written)
941 loff_t end_of_write = offset + bytes_written;
943 if (end_of_write > cifsi->server_eof)
944 cifsi->server_eof = end_of_write;
947 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
948 size_t write_size, loff_t *poffset)
951 unsigned int bytes_written = 0;
952 unsigned int total_written;
953 struct cifs_sb_info *cifs_sb;
954 struct cifsTconInfo *pTcon;
956 struct cifsFileInfo *open_file;
957 struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
959 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
961 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
962 *poffset, file->f_path.dentry->d_name.name); */
964 if (file->private_data == NULL)
967 open_file = file->private_data;
968 pTcon = tlink_tcon(open_file->tlink);
970 rc = generic_write_checks(file, poffset, &write_size, 0);
976 long_op = cifs_write_timeout(cifsi, *poffset);
977 for (total_written = 0; write_size > total_written;
978 total_written += bytes_written) {
980 while (rc == -EAGAIN) {
981 if (file->private_data == NULL) {
982 /* file has been closed on us */
984 /* if we have gotten here we have written some data
985 and blocked, and the file has been freed on us while
986 we blocked so return what we managed to write */
987 return total_written;
989 if (open_file->closePend) {
992 return total_written;
996 if (open_file->invalidHandle) {
997 /* we could deadlock if we called
998 filemap_fdatawait from here so tell
999 reopen_file not to flush data to server
1001 rc = cifs_reopen_file(file, false);
1006 rc = CIFSSMBWrite(xid, pTcon,
1008 min_t(const int, cifs_sb->wsize,
1009 write_size - total_written),
1010 *poffset, &bytes_written,
1011 NULL, write_data + total_written, long_op);
1013 if (rc || (bytes_written == 0)) {
1021 cifs_update_eof(cifsi, *poffset, bytes_written);
1022 *poffset += bytes_written;
1024 long_op = CIFS_STD_OP; /* subsequent writes fast -
1025 15 seconds is plenty */
1028 cifs_stats_bytes_written(pTcon, total_written);
1030 /* since the write may have blocked check these pointers again */
1031 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1032 struct inode *inode = file->f_path.dentry->d_inode;
1033 /* Do not update local mtime - server will set its actual value on write
1034 * inode->i_ctime = inode->i_mtime =
1035 * current_fs_time(inode->i_sb);*/
1036 if (total_written > 0) {
1037 spin_lock(&inode->i_lock);
1038 if (*poffset > file->f_path.dentry->d_inode->i_size)
1039 i_size_write(file->f_path.dentry->d_inode,
1041 spin_unlock(&inode->i_lock);
1043 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1046 return total_written;
1049 static ssize_t cifs_write(struct file *file, const char *write_data,
1050 size_t write_size, loff_t *poffset)
1053 unsigned int bytes_written = 0;
1054 unsigned int total_written;
1055 struct cifs_sb_info *cifs_sb;
1056 struct cifsTconInfo *pTcon;
1058 struct cifsFileInfo *open_file;
1059 struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
1061 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1063 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
1064 *poffset, file->f_path.dentry->d_name.name);
1066 if (file->private_data == NULL)
1068 open_file = file->private_data;
1069 pTcon = tlink_tcon(open_file->tlink);
1073 long_op = cifs_write_timeout(cifsi, *poffset);
1074 for (total_written = 0; write_size > total_written;
1075 total_written += bytes_written) {
1077 while (rc == -EAGAIN) {
1078 if (file->private_data == NULL) {
1079 /* file has been closed on us */
1081 /* if we have gotten here we have written some data
1082 and blocked, and the file has been freed on us
1083 while we blocked so return what we managed to
1085 return total_written;
1087 if (open_file->closePend) {
1090 return total_written;
1094 if (open_file->invalidHandle) {
1095 /* we could deadlock if we called
1096 filemap_fdatawait from here so tell
1097 reopen_file not to flush data to
1099 rc = cifs_reopen_file(file, false);
1103 if (experimEnabled || (pTcon->ses->server &&
1104 ((pTcon->ses->server->secMode &
1105 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1110 len = min((size_t)cifs_sb->wsize,
1111 write_size - total_written);
1112 /* iov[0] is reserved for smb header */
1113 iov[1].iov_base = (char *)write_data +
1115 iov[1].iov_len = len;
1116 rc = CIFSSMBWrite2(xid, pTcon,
1117 open_file->netfid, len,
1118 *poffset, &bytes_written,
1121 rc = CIFSSMBWrite(xid, pTcon,
1123 min_t(const int, cifs_sb->wsize,
1124 write_size - total_written),
1125 *poffset, &bytes_written,
1126 write_data + total_written,
1129 if (rc || (bytes_written == 0)) {
1137 cifs_update_eof(cifsi, *poffset, bytes_written);
1138 *poffset += bytes_written;
1140 long_op = CIFS_STD_OP; /* subsequent writes fast -
1141 15 seconds is plenty */
1144 cifs_stats_bytes_written(pTcon, total_written);
1146 /* since the write may have blocked check these pointers again */
1147 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1148 /*BB We could make this contingent on superblock ATIME flag too */
1149 /* file->f_path.dentry->d_inode->i_ctime =
1150 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1151 if (total_written > 0) {
1152 spin_lock(&file->f_path.dentry->d_inode->i_lock);
1153 if (*poffset > file->f_path.dentry->d_inode->i_size)
1154 i_size_write(file->f_path.dentry->d_inode,
1156 spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1158 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1161 return total_written;
1164 #ifdef CONFIG_CIFS_EXPERIMENTAL
1165 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1168 struct cifsFileInfo *open_file = NULL;
1169 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1171 /* only filter by fsuid on multiuser mounts */
1172 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1175 read_lock(&GlobalSMBSeslock);
1176 /* we could simply get the first_list_entry since write-only entries
1177 are always at the end of the list but since the first entry might
1178 have a close pending, we go through the whole list */
1179 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1180 if (open_file->closePend)
1182 if (fsuid_only && open_file->uid != current_fsuid())
1184 if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
1185 (open_file->pfile->f_flags & O_RDONLY))) {
1186 if (!open_file->invalidHandle) {
1187 /* found a good file */
1188 /* lock it so it will not be closed on us */
1189 cifsFileInfo_get(open_file);
1190 read_unlock(&GlobalSMBSeslock);
1192 } /* else might as well continue, and look for
1193 another, or simply have the caller reopen it
1194 again rather than trying to fix this handle */
1195 } else /* write only file */
1196 break; /* write only files are last so must be done */
1198 read_unlock(&GlobalSMBSeslock);
1203 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1206 struct cifsFileInfo *open_file;
1207 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1208 bool any_available = false;
1211 /* Having a null inode here (because mapping->host was set to zero by
1212 the VFS or MM) should not happen but we had reports of on oops (due to
1213 it being zero) during stress testcases so we need to check for it */
1215 if (cifs_inode == NULL) {
1216 cERROR(1, "Null inode passed to cifs_writeable_file");
1221 /* only filter by fsuid on multiuser mounts */
1222 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1225 read_lock(&GlobalSMBSeslock);
1227 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1228 if (open_file->closePend)
1230 if (!any_available && open_file->pid != current->tgid)
1232 if (fsuid_only && open_file->uid != current_fsuid())
1234 if (open_file->pfile &&
1235 ((open_file->pfile->f_flags & O_RDWR) ||
1236 (open_file->pfile->f_flags & O_WRONLY))) {
1237 cifsFileInfo_get(open_file);
1239 if (!open_file->invalidHandle) {
1240 /* found a good writable file */
1241 read_unlock(&GlobalSMBSeslock);
1245 read_unlock(&GlobalSMBSeslock);
1246 /* Had to unlock since following call can block */
1247 rc = cifs_reopen_file(open_file->pfile, false);
1249 if (!open_file->closePend)
1251 else { /* start over in case this was deleted */
1252 /* since the list could be modified */
1253 read_lock(&GlobalSMBSeslock);
1254 cifsFileInfo_put(open_file);
1255 goto refind_writable;
1259 /* if it fails, try another handle if possible -
1260 (we can not do this if closePending since
1261 loop could be modified - in which case we
1262 have to start at the beginning of the list
1263 again. Note that it would be bad
1264 to hold up writepages here (rather than
1265 in caller) with continuous retries */
1266 cFYI(1, "wp failed on reopen file");
1267 read_lock(&GlobalSMBSeslock);
1268 /* can not use this handle, no write
1269 pending on this one after all */
1270 cifsFileInfo_put(open_file);
1272 if (open_file->closePend) /* list could have changed */
1273 goto refind_writable;
1274 /* else we simply continue to the next entry. Thus
1275 we do not loop on reopen errors. If we
1276 can not reopen the file, for example if we
1277 reconnected to a server with another client
1278 racing to delete or lock the file we would not
1279 make progress if we restarted before the beginning
1280 of the loop here. */
1283 /* couldn't find useable FH with same pid, try any available */
1284 if (!any_available) {
1285 any_available = true;
1286 goto refind_writable;
1288 read_unlock(&GlobalSMBSeslock);
1292 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1294 struct address_space *mapping = page->mapping;
1295 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1298 int bytes_written = 0;
1299 struct cifs_sb_info *cifs_sb;
1300 struct inode *inode;
1301 struct cifsFileInfo *open_file;
1303 if (!mapping || !mapping->host)
1306 inode = page->mapping->host;
1307 cifs_sb = CIFS_SB(inode->i_sb);
1309 offset += (loff_t)from;
1310 write_data = kmap(page);
1313 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1318 /* racing with truncate? */
1319 if (offset > mapping->host->i_size) {
1321 return 0; /* don't care */
1324 /* check to make sure that we are not extending the file */
1325 if (mapping->host->i_size - offset < (loff_t)to)
1326 to = (unsigned)(mapping->host->i_size - offset);
1328 open_file = find_writable_file(CIFS_I(mapping->host), false);
1330 bytes_written = cifs_write(open_file->pfile, write_data,
1332 cifsFileInfo_put(open_file);
1333 /* Does mm or vfs already set times? */
1334 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1335 if ((bytes_written > 0) && (offset))
1337 else if (bytes_written < 0)
1340 cFYI(1, "No writeable filehandles for inode");
1348 static int cifs_writepages(struct address_space *mapping,
1349 struct writeback_control *wbc)
1351 struct backing_dev_info *bdi = mapping->backing_dev_info;
1352 unsigned int bytes_to_write;
1353 unsigned int bytes_written;
1354 struct cifs_sb_info *cifs_sb;
1358 int range_whole = 0;
1365 struct cifsFileInfo *open_file;
1366 struct cifsTconInfo *tcon;
1367 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1369 struct pagevec pvec;
1375 * BB: Is this meaningful for a non-block-device file system?
1376 * If it is, we should test it again after we do I/O
1378 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1379 wbc->encountered_congestion = 1;
1383 cifs_sb = CIFS_SB(mapping->host->i_sb);
1386 * If wsize is smaller that the page cache size, default to writing
1387 * one page at a time via cifs_writepage
1389 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1390 return generic_writepages(mapping, wbc);
1392 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1394 return generic_writepages(mapping, wbc);
1397 * if there's no open file, then this is likely to fail too,
1398 * but it'll at least handle the return. Maybe it should be
1401 open_file = find_writable_file(CIFS_I(mapping->host), false);
1404 return generic_writepages(mapping, wbc);
1407 tcon = tlink_tcon(open_file->tlink);
1408 if (!experimEnabled && tcon->ses->server->secMode &
1409 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1410 cifsFileInfo_put(open_file);
1411 return generic_writepages(mapping, wbc);
1413 cifsFileInfo_put(open_file);
1417 pagevec_init(&pvec, 0);
1418 if (wbc->range_cyclic) {
1419 index = mapping->writeback_index; /* Start from prev offset */
1422 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1423 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1424 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1429 while (!done && (index <= end) &&
1430 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1431 PAGECACHE_TAG_DIRTY,
1432 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1441 for (i = 0; i < nr_pages; i++) {
1442 page = pvec.pages[i];
1444 * At this point we hold neither mapping->tree_lock nor
1445 * lock on the page itself: the page may be truncated or
1446 * invalidated (changing page->mapping to NULL), or even
1447 * swizzled back from swapper_space to tmpfs file
1453 else if (!trylock_page(page))
1456 if (unlikely(page->mapping != mapping)) {
1461 if (!wbc->range_cyclic && page->index > end) {
1467 if (next && (page->index != next)) {
1468 /* Not next consecutive page */
1473 if (wbc->sync_mode != WB_SYNC_NONE)
1474 wait_on_page_writeback(page);
1476 if (PageWriteback(page) ||
1477 !clear_page_dirty_for_io(page)) {
1483 * This actually clears the dirty bit in the radix tree.
1484 * See cifs_writepage() for more commentary.
1486 set_page_writeback(page);
1488 if (page_offset(page) >= mapping->host->i_size) {
1491 end_page_writeback(page);
1496 * BB can we get rid of this? pages are held by pvec
1498 page_cache_get(page);
1500 len = min(mapping->host->i_size - page_offset(page),
1501 (loff_t)PAGE_CACHE_SIZE);
1503 /* reserve iov[0] for the smb header */
1505 iov[n_iov].iov_base = kmap(page);
1506 iov[n_iov].iov_len = len;
1507 bytes_to_write += len;
1511 offset = page_offset(page);
1513 next = page->index + 1;
1514 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1518 open_file = find_writable_file(CIFS_I(mapping->host),
1521 cERROR(1, "No writable handles for inode");
1524 long_op = cifs_write_timeout(cifsi, offset);
1525 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1526 bytes_to_write, offset,
1527 &bytes_written, iov, n_iov,
1529 cifsFileInfo_put(open_file);
1530 cifs_update_eof(cifsi, offset, bytes_written);
1533 if (rc || bytes_written < bytes_to_write) {
1534 cERROR(1, "Write2 ret %d, wrote %d",
1536 /* BB what if continued retry is
1537 requested via mount flags? */
1539 set_bit(AS_ENOSPC, &mapping->flags);
1541 set_bit(AS_EIO, &mapping->flags);
1543 cifs_stats_bytes_written(tcon, bytes_written);
1546 for (i = 0; i < n_iov; i++) {
1547 page = pvec.pages[first + i];
1548 /* Should we also set page error on
1549 success rc but too little data written? */
1550 /* BB investigate retry logic on temporary
1551 server crash cases and how recovery works
1552 when page marked as error */
1557 end_page_writeback(page);
1558 page_cache_release(page);
1560 if ((wbc->nr_to_write -= n_iov) <= 0)
1564 /* Need to re-find the pages we skipped */
1565 index = pvec.pages[0]->index + 1;
1567 pagevec_release(&pvec);
1569 if (!scanned && !done) {
1571 * We hit the last page and there is more work to be done: wrap
1572 * back to the start of the file
1578 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1579 mapping->writeback_index = index;
1586 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1592 /* BB add check for wbc flags */
1593 page_cache_get(page);
1594 if (!PageUptodate(page))
1595 cFYI(1, "ppw - page not up to date");
1598 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1600 * A writepage() implementation always needs to do either this,
1601 * or re-dirty the page with "redirty_page_for_writepage()" in
1602 * the case of a failure.
1604 * Just unlocking the page will cause the radix tree tag-bits
1605 * to fail to update with the state of the page correctly.
1607 set_page_writeback(page);
1608 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1609 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1611 end_page_writeback(page);
1612 page_cache_release(page);
1617 static int cifs_write_end(struct file *file, struct address_space *mapping,
1618 loff_t pos, unsigned len, unsigned copied,
1619 struct page *page, void *fsdata)
1622 struct inode *inode = mapping->host;
1624 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1627 if (PageChecked(page)) {
1629 SetPageUptodate(page);
1630 ClearPageChecked(page);
1631 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1632 SetPageUptodate(page);
1634 if (!PageUptodate(page)) {
1636 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1640 /* this is probably better than directly calling
1641 partialpage_write since in this function the file handle is
1642 known which we might as well leverage */
1643 /* BB check if anything else missing out of ppw
1644 such as updating last write time */
1645 page_data = kmap(page);
1646 rc = cifs_write(file, page_data + offset, copied, &pos);
1647 /* if (rc < 0) should we set writebehind rc? */
1654 set_page_dirty(page);
1658 spin_lock(&inode->i_lock);
1659 if (pos > inode->i_size)
1660 i_size_write(inode, pos);
1661 spin_unlock(&inode->i_lock);
1665 page_cache_release(page);
1670 int cifs_fsync(struct file *file, int datasync)
1674 struct cifsTconInfo *tcon;
1675 struct cifsFileInfo *smbfile = file->private_data;
1676 struct inode *inode = file->f_path.dentry->d_inode;
1680 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1681 file->f_path.dentry->d_name.name, datasync);
1683 rc = filemap_write_and_wait(inode->i_mapping);
1685 rc = CIFS_I(inode)->write_behind_rc;
1686 CIFS_I(inode)->write_behind_rc = 0;
1687 tcon = tlink_tcon(smbfile->tlink);
1688 if (!rc && tcon && smbfile &&
1689 !(CIFS_SB(inode->i_sb)->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1690 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1697 /* static void cifs_sync_page(struct page *page)
1699 struct address_space *mapping;
1700 struct inode *inode;
1701 unsigned long index = page->index;
1702 unsigned int rpages = 0;
1705 cFYI(1, "sync page %p", page);
1706 mapping = page->mapping;
1709 inode = mapping->host;
1713 /* fill in rpages then
1714 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1716 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1726 * As file closes, flush all cached write data for this inode checking
1727 * for write behind errors.
1729 int cifs_flush(struct file *file, fl_owner_t id)
1731 struct inode *inode = file->f_path.dentry->d_inode;
1734 /* Rather than do the steps manually:
1735 lock the inode for writing
1736 loop through pages looking for write behind data (dirty pages)
1737 coalesce into contiguous 16K (or smaller) chunks to write to server
1738 send to server (prefer in parallel)
1739 deal with writebehind errors
1740 unlock inode for writing
1741 filemapfdatawrite appears easier for the time being */
1743 rc = filemap_fdatawrite(inode->i_mapping);
1744 /* reset wb rc if we were able to write out dirty pages */
1746 rc = CIFS_I(inode)->write_behind_rc;
1747 CIFS_I(inode)->write_behind_rc = 0;
1750 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1755 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1756 size_t read_size, loff_t *poffset)
1759 unsigned int bytes_read = 0;
1760 unsigned int total_read = 0;
1761 unsigned int current_read_size;
1762 struct cifs_sb_info *cifs_sb;
1763 struct cifsTconInfo *pTcon;
1765 struct cifsFileInfo *open_file;
1766 char *smb_read_data;
1767 char __user *current_offset;
1768 struct smb_com_read_rsp *pSMBr;
1771 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1773 if (file->private_data == NULL) {
1778 open_file = file->private_data;
1779 pTcon = tlink_tcon(open_file->tlink);
1781 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1782 cFYI(1, "attempting read on write only file instance");
1784 for (total_read = 0, current_offset = read_data;
1785 read_size > total_read;
1786 total_read += bytes_read, current_offset += bytes_read) {
1787 current_read_size = min_t(const int, read_size - total_read,
1790 smb_read_data = NULL;
1791 while (rc == -EAGAIN) {
1792 int buf_type = CIFS_NO_BUFFER;
1793 if ((open_file->invalidHandle) &&
1794 (!open_file->closePend)) {
1795 rc = cifs_reopen_file(file, true);
1799 rc = CIFSSMBRead(xid, pTcon,
1801 current_read_size, *poffset,
1802 &bytes_read, &smb_read_data,
1804 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1805 if (smb_read_data) {
1806 if (copy_to_user(current_offset,
1808 4 /* RFC1001 length field */ +
1809 le16_to_cpu(pSMBr->DataOffset),
1813 if (buf_type == CIFS_SMALL_BUFFER)
1814 cifs_small_buf_release(smb_read_data);
1815 else if (buf_type == CIFS_LARGE_BUFFER)
1816 cifs_buf_release(smb_read_data);
1817 smb_read_data = NULL;
1820 if (rc || (bytes_read == 0)) {
1828 cifs_stats_bytes_read(pTcon, bytes_read);
1829 *poffset += bytes_read;
1837 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1841 unsigned int bytes_read = 0;
1842 unsigned int total_read;
1843 unsigned int current_read_size;
1844 struct cifs_sb_info *cifs_sb;
1845 struct cifsTconInfo *pTcon;
1847 char *current_offset;
1848 struct cifsFileInfo *open_file;
1849 int buf_type = CIFS_NO_BUFFER;
1852 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1854 if (file->private_data == NULL) {
1859 open_file = file->private_data;
1860 pTcon = tlink_tcon(open_file->tlink);
1862 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1863 cFYI(1, "attempting read on write only file instance");
1865 for (total_read = 0, current_offset = read_data;
1866 read_size > total_read;
1867 total_read += bytes_read, current_offset += bytes_read) {
1868 current_read_size = min_t(const int, read_size - total_read,
1870 /* For windows me and 9x we do not want to request more
1871 than it negotiated since it will refuse the read then */
1873 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1874 current_read_size = min_t(const int, current_read_size,
1875 pTcon->ses->server->maxBuf - 128);
1878 while (rc == -EAGAIN) {
1879 if ((open_file->invalidHandle) &&
1880 (!open_file->closePend)) {
1881 rc = cifs_reopen_file(file, true);
1885 rc = CIFSSMBRead(xid, pTcon,
1887 current_read_size, *poffset,
1888 &bytes_read, ¤t_offset,
1891 if (rc || (bytes_read == 0)) {
1899 cifs_stats_bytes_read(pTcon, total_read);
1900 *poffset += bytes_read;
1907 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1912 rc = cifs_revalidate_file(file);
1914 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1918 rc = generic_file_mmap(file, vma);
1924 static void cifs_copy_cache_pages(struct address_space *mapping,
1925 struct list_head *pages, int bytes_read, char *data)
1930 while (bytes_read > 0) {
1931 if (list_empty(pages))
1934 page = list_entry(pages->prev, struct page, lru);
1935 list_del(&page->lru);
1937 if (add_to_page_cache_lru(page, mapping, page->index,
1939 page_cache_release(page);
1940 cFYI(1, "Add page cache failed");
1941 data += PAGE_CACHE_SIZE;
1942 bytes_read -= PAGE_CACHE_SIZE;
1945 page_cache_release(page);
1947 target = kmap_atomic(page, KM_USER0);
1949 if (PAGE_CACHE_SIZE > bytes_read) {
1950 memcpy(target, data, bytes_read);
1951 /* zero the tail end of this partial page */
1952 memset(target + bytes_read, 0,
1953 PAGE_CACHE_SIZE - bytes_read);
1956 memcpy(target, data, PAGE_CACHE_SIZE);
1957 bytes_read -= PAGE_CACHE_SIZE;
1959 kunmap_atomic(target, KM_USER0);
1961 flush_dcache_page(page);
1962 SetPageUptodate(page);
1964 data += PAGE_CACHE_SIZE;
1966 /* add page to FS-Cache */
1967 cifs_readpage_to_fscache(mapping->host, page);
1972 static int cifs_readpages(struct file *file, struct address_space *mapping,
1973 struct list_head *page_list, unsigned num_pages)
1979 struct cifs_sb_info *cifs_sb;
1980 struct cifsTconInfo *pTcon;
1981 unsigned int bytes_read = 0;
1982 unsigned int read_size, i;
1983 char *smb_read_data = NULL;
1984 struct smb_com_read_rsp *pSMBr;
1985 struct cifsFileInfo *open_file;
1986 int buf_type = CIFS_NO_BUFFER;
1989 if (file->private_data == NULL) {
1994 open_file = file->private_data;
1995 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1996 pTcon = tlink_tcon(open_file->tlink);
1999 * Reads as many pages as possible from fscache. Returns -ENOBUFS
2000 * immediately if the cookie is negative
2002 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
2007 cFYI(DBG2, "rpages: num pages %d", num_pages);
2008 for (i = 0; i < num_pages; ) {
2009 unsigned contig_pages;
2010 struct page *tmp_page;
2011 unsigned long expected_index;
2013 if (list_empty(page_list))
2016 page = list_entry(page_list->prev, struct page, lru);
2017 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2019 /* count adjacent pages that we will read into */
2022 list_entry(page_list->prev, struct page, lru)->index;
2023 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2024 if (tmp_page->index == expected_index) {
2030 if (contig_pages + i > num_pages)
2031 contig_pages = num_pages - i;
2033 /* for reads over a certain size could initiate async
2036 read_size = contig_pages * PAGE_CACHE_SIZE;
2037 /* Read size needs to be in multiples of one page */
2038 read_size = min_t(const unsigned int, read_size,
2039 cifs_sb->rsize & PAGE_CACHE_MASK);
2040 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
2041 read_size, contig_pages);
2043 while (rc == -EAGAIN) {
2044 if ((open_file->invalidHandle) &&
2045 (!open_file->closePend)) {
2046 rc = cifs_reopen_file(file, true);
2051 rc = CIFSSMBRead(xid, pTcon,
2054 &bytes_read, &smb_read_data,
2056 /* BB more RC checks ? */
2057 if (rc == -EAGAIN) {
2058 if (smb_read_data) {
2059 if (buf_type == CIFS_SMALL_BUFFER)
2060 cifs_small_buf_release(smb_read_data);
2061 else if (buf_type == CIFS_LARGE_BUFFER)
2062 cifs_buf_release(smb_read_data);
2063 smb_read_data = NULL;
2067 if ((rc < 0) || (smb_read_data == NULL)) {
2068 cFYI(1, "Read error in readpages: %d", rc);
2070 } else if (bytes_read > 0) {
2071 task_io_account_read(bytes_read);
2072 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2073 cifs_copy_cache_pages(mapping, page_list, bytes_read,
2074 smb_read_data + 4 /* RFC1001 hdr */ +
2075 le16_to_cpu(pSMBr->DataOffset));
2077 i += bytes_read >> PAGE_CACHE_SHIFT;
2078 cifs_stats_bytes_read(pTcon, bytes_read);
2079 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2080 i++; /* account for partial page */
2082 /* server copy of file can have smaller size
2084 /* BB do we need to verify this common case ?
2085 this case is ok - if we are at server EOF
2086 we will hit it on next read */
2091 cFYI(1, "No bytes read (%d) at offset %lld . "
2092 "Cleaning remaining pages from readahead list",
2093 bytes_read, offset);
2094 /* BB turn off caching and do new lookup on
2095 file size at server? */
2098 if (smb_read_data) {
2099 if (buf_type == CIFS_SMALL_BUFFER)
2100 cifs_small_buf_release(smb_read_data);
2101 else if (buf_type == CIFS_LARGE_BUFFER)
2102 cifs_buf_release(smb_read_data);
2103 smb_read_data = NULL;
2108 /* need to free smb_read_data buf before exit */
2109 if (smb_read_data) {
2110 if (buf_type == CIFS_SMALL_BUFFER)
2111 cifs_small_buf_release(smb_read_data);
2112 else if (buf_type == CIFS_LARGE_BUFFER)
2113 cifs_buf_release(smb_read_data);
2114 smb_read_data = NULL;
2122 static int cifs_readpage_worker(struct file *file, struct page *page,
2128 /* Is the page cached? */
2129 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2133 page_cache_get(page);
2134 read_data = kmap(page);
2135 /* for reads over a certain size could initiate async read ahead */
2137 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2142 cFYI(1, "Bytes read %d", rc);
2144 file->f_path.dentry->d_inode->i_atime =
2145 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2147 if (PAGE_CACHE_SIZE > rc)
2148 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2150 flush_dcache_page(page);
2151 SetPageUptodate(page);
2153 /* send this page to the cache */
2154 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2160 page_cache_release(page);
2166 static int cifs_readpage(struct file *file, struct page *page)
2168 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2174 if (file->private_data == NULL) {
2180 cFYI(1, "readpage %p at offset %d 0x%x\n",
2181 page, (int)offset, (int)offset);
2183 rc = cifs_readpage_worker(file, page, &offset);
2191 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2193 struct cifsFileInfo *open_file;
2195 read_lock(&GlobalSMBSeslock);
2196 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2197 if (open_file->closePend)
2199 if (open_file->pfile &&
2200 ((open_file->pfile->f_flags & O_RDWR) ||
2201 (open_file->pfile->f_flags & O_WRONLY))) {
2202 read_unlock(&GlobalSMBSeslock);
2206 read_unlock(&GlobalSMBSeslock);
2210 /* We do not want to update the file size from server for inodes
2211 open for write - to avoid races with writepage extending
2212 the file - in the future we could consider allowing
2213 refreshing the inode only on increases in the file size
2214 but this is tricky to do without racing with writebehind
2215 page caching in the current Linux kernel design */
2216 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2221 if (is_inode_writable(cifsInode)) {
2222 /* This inode is open for write at least once */
2223 struct cifs_sb_info *cifs_sb;
2225 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2226 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2227 /* since no page cache to corrupt on directio
2228 we can change size safely */
2232 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2240 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2241 loff_t pos, unsigned len, unsigned flags,
2242 struct page **pagep, void **fsdata)
2244 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2245 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2246 loff_t page_start = pos & PAGE_MASK;
2251 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2253 page = grab_cache_page_write_begin(mapping, index, flags);
2259 if (PageUptodate(page))
2263 * If we write a full page it will be up to date, no need to read from
2264 * the server. If the write is short, we'll end up doing a sync write
2267 if (len == PAGE_CACHE_SIZE)
2271 * optimize away the read when we have an oplock, and we're not
2272 * expecting to use any of the data we'd be reading in. That
2273 * is, when the page lies beyond the EOF, or straddles the EOF
2274 * and the write will cover all of the existing data.
2276 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2277 i_size = i_size_read(mapping->host);
2278 if (page_start >= i_size ||
2279 (offset == 0 && (pos + len) >= i_size)) {
2280 zero_user_segments(page, 0, offset,
2284 * PageChecked means that the parts of the page
2285 * to which we're not writing are considered up
2286 * to date. Once the data is copied to the
2287 * page, it can be set uptodate.
2289 SetPageChecked(page);
2294 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2296 * might as well read a page, it is fast enough. If we get
2297 * an error, we don't need to return it. cifs_write_end will
2298 * do a sync write instead since PG_uptodate isn't set.
2300 cifs_readpage_worker(file, page, &page_start);
2302 /* we could try using another file handle if there is one -
2303 but how would we lock it to prevent close of that handle
2304 racing with this read? In any case
2305 this will be written out by write_end so is fine */
2312 static int cifs_release_page(struct page *page, gfp_t gfp)
2314 if (PagePrivate(page))
2317 return cifs_fscache_release_page(page, gfp);
2320 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2322 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2325 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2328 void cifs_oplock_break(struct work_struct *work)
2330 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2332 struct inode *inode = cfile->dentry->d_inode;
2333 struct cifsInodeInfo *cinode = CIFS_I(inode);
2336 if (inode && S_ISREG(inode->i_mode)) {
2337 if (cinode->clientCanCacheRead)
2338 break_lease(inode, O_RDONLY);
2340 break_lease(inode, O_WRONLY);
2341 rc = filemap_fdatawrite(inode->i_mapping);
2342 if (cinode->clientCanCacheRead == 0) {
2343 waitrc = filemap_fdatawait(inode->i_mapping);
2344 invalidate_remote_inode(inode);
2349 cinode->write_behind_rc = rc;
2350 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2354 * releasing stale oplock after recent reconnect of smb session using
2355 * a now incorrect file handle is not a data integrity issue but do
2356 * not bother sending an oplock release if session to server still is
2357 * disconnected since oplock already released by the server
2359 if (!cfile->closePend && !cfile->oplock_break_cancelled) {
2360 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2361 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false);
2362 cFYI(1, "Oplock release rc = %d", rc);
2366 * We might have kicked in before is_valid_oplock_break()
2367 * finished grabbing reference for us. Make sure it's done by
2368 * waiting for GlobalSMSSeslock.
2370 write_lock(&GlobalSMBSeslock);
2371 write_unlock(&GlobalSMBSeslock);
2373 cifs_oplock_break_put(cfile);
2376 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2378 cifs_sb_active(cfile->dentry->d_sb);
2379 cifsFileInfo_get(cfile);
2382 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2384 cifsFileInfo_put(cfile);
2385 cifs_sb_deactive(cfile->dentry->d_sb);
2388 const struct address_space_operations cifs_addr_ops = {
2389 .readpage = cifs_readpage,
2390 .readpages = cifs_readpages,
2391 .writepage = cifs_writepage,
2392 .writepages = cifs_writepages,
2393 .write_begin = cifs_write_begin,
2394 .write_end = cifs_write_end,
2395 .set_page_dirty = __set_page_dirty_nobuffers,
2396 .releasepage = cifs_release_page,
2397 .invalidatepage = cifs_invalidate_page,
2398 /* .sync_page = cifs_sync_page, */
2403 * cifs_readpages requires the server to support a buffer large enough to
2404 * contain the header plus one complete page of data. Otherwise, we need
2405 * to leave cifs_readpages out of the address space operations.
2407 const struct address_space_operations cifs_addr_ops_smallbuf = {
2408 .readpage = cifs_readpage,
2409 .writepage = cifs_writepage,
2410 .writepages = cifs_writepages,
2411 .write_begin = cifs_write_begin,
2412 .write_end = cifs_write_end,
2413 .set_page_dirty = __set_page_dirty_nobuffers,
2414 .releasepage = cifs_release_page,
2415 .invalidatepage = cifs_invalidate_page,
2416 /* .sync_page = cifs_sync_page, */