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 inline fmode_t cifs_posix_convert_flags(unsigned int flags)
65 fmode_t posix_flags = 0;
67 if ((flags & O_ACCMODE) == O_RDONLY)
68 posix_flags = FMODE_READ;
69 else if ((flags & O_ACCMODE) == O_WRONLY)
70 posix_flags = FMODE_WRITE;
71 else if ((flags & O_ACCMODE) == O_RDWR) {
72 /* GENERIC_ALL is too much permission to request
73 can cause unnecessary access denied on create */
74 /* return GENERIC_ALL; */
75 posix_flags = FMODE_READ | FMODE_WRITE;
77 /* can not map O_CREAT or O_EXCL or O_TRUNC flags when
78 reopening a file. They had their effect on the original open */
80 posix_flags |= (fmode_t)O_APPEND;
82 posix_flags |= (fmode_t)O_DSYNC;
84 posix_flags |= (fmode_t)__O_SYNC;
85 if (flags & O_DIRECTORY)
86 posix_flags |= (fmode_t)O_DIRECTORY;
87 if (flags & O_NOFOLLOW)
88 posix_flags |= (fmode_t)O_NOFOLLOW;
90 posix_flags |= (fmode_t)O_DIRECT;
95 static inline int cifs_get_disposition(unsigned int flags)
97 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
99 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
100 return FILE_OVERWRITE_IF;
101 else if ((flags & O_CREAT) == O_CREAT)
103 else if ((flags & O_TRUNC) == O_TRUNC)
104 return FILE_OVERWRITE;
109 /* all arguments to this function must be checked for validity in caller */
111 cifs_posix_open_inode_helper(struct inode *inode, struct file *file,
112 struct cifsInodeInfo *pCifsInode, __u32 oplock,
116 write_lock(&GlobalSMBSeslock);
118 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
119 if (pCifsInode == NULL) {
120 write_unlock(&GlobalSMBSeslock);
124 if (pCifsInode->clientCanCacheRead) {
125 /* we have the inode open somewhere else
126 no need to discard cache data */
127 goto psx_client_can_cache;
130 /* BB FIXME need to fix this check to move it earlier into posix_open
131 BB fIX following section BB FIXME */
133 /* if not oplocked, invalidate inode pages if mtime or file
135 /* temp = cifs_NTtimeToUnix(le64_to_cpu(buf->LastWriteTime));
136 if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
137 (file->f_path.dentry->d_inode->i_size ==
138 (loff_t)le64_to_cpu(buf->EndOfFile))) {
139 cFYI(1, "inode unchanged on server");
141 if (file->f_path.dentry->d_inode->i_mapping) {
142 rc = filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
144 CIFS_I(file->f_path.dentry->d_inode)->write_behind_rc = rc;
146 cFYI(1, "invalidating remote inode since open detected it "
148 invalidate_remote_inode(file->f_path.dentry->d_inode);
151 psx_client_can_cache:
152 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
153 pCifsInode->clientCanCacheAll = true;
154 pCifsInode->clientCanCacheRead = true;
155 cFYI(1, "Exclusive Oplock granted on inode %p",
156 file->f_path.dentry->d_inode);
157 } else if ((oplock & 0xF) == OPLOCK_READ)
158 pCifsInode->clientCanCacheRead = true;
160 /* will have to change the unlock if we reenable the
161 filemap_fdatawrite (which does not seem necessary */
162 write_unlock(&GlobalSMBSeslock);
166 /* all arguments to this function must be checked for validity in caller */
167 static inline int cifs_open_inode_helper(struct inode *inode,
168 struct cifsTconInfo *pTcon, __u32 oplock, FILE_ALL_INFO *buf,
169 char *full_path, int xid)
171 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
172 struct timespec temp;
175 if (pCifsInode->clientCanCacheRead) {
176 /* we have the inode open somewhere else
177 no need to discard cache data */
178 goto client_can_cache;
181 /* BB need same check in cifs_create too? */
182 /* if not oplocked, invalidate inode pages if mtime or file
184 temp = cifs_NTtimeToUnix(buf->LastWriteTime);
185 if (timespec_equal(&inode->i_mtime, &temp) &&
187 (loff_t)le64_to_cpu(buf->EndOfFile))) {
188 cFYI(1, "inode unchanged on server");
190 if (inode->i_mapping) {
191 /* BB no need to lock inode until after invalidate
192 since namei code should already have it locked? */
193 rc = filemap_write_and_wait(inode->i_mapping);
195 pCifsInode->write_behind_rc = rc;
197 cFYI(1, "invalidating remote inode since open detected it "
199 invalidate_remote_inode(inode);
204 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
207 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
210 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
211 pCifsInode->clientCanCacheAll = true;
212 pCifsInode->clientCanCacheRead = true;
213 cFYI(1, "Exclusive Oplock granted on inode %p", inode);
214 } else if ((oplock & 0xF) == OPLOCK_READ)
215 pCifsInode->clientCanCacheRead = true;
220 int cifs_open(struct inode *inode, struct file *file)
225 struct cifs_sb_info *cifs_sb;
226 struct cifsTconInfo *tcon;
227 struct tcon_link *tlink;
228 struct cifsFileInfo *pCifsFile = NULL;
229 struct cifsInodeInfo *pCifsInode;
230 char *full_path = NULL;
234 FILE_ALL_INFO *buf = NULL;
238 cifs_sb = CIFS_SB(inode->i_sb);
239 tlink = cifs_sb_tlink(cifs_sb);
242 return PTR_ERR(tlink);
244 tcon = tlink_tcon(tlink);
246 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
248 full_path = build_path_from_dentry(file->f_path.dentry);
249 if (full_path == NULL) {
254 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
255 inode, file->f_flags, full_path);
262 if (!tcon->broken_posix_open && tcon->unix_ext &&
263 (tcon->ses->capabilities & CAP_UNIX) &&
264 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
265 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
266 int oflags = (int) cifs_posix_convert_flags(file->f_flags);
267 oflags |= SMB_O_CREAT;
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 oflags, &oplock, &netfid, xid);
273 cFYI(1, "posix open succeeded");
274 /* no need for special case handling of setting mode
275 on read only files needed here */
277 rc = cifs_posix_open_inode_helper(inode, file,
278 pCifsInode, oplock, netfid);
280 CIFSSMBClose(xid, tcon, netfid);
284 pCifsFile = cifs_new_fileinfo(inode, netfid, file,
286 tcon, oflags, oplock);
287 if (pCifsFile == NULL) {
288 CIFSSMBClose(xid, tcon, netfid);
292 cifs_fscache_set_inode_cookie(inode, file);
295 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
296 if (tcon->ses->serverNOS)
297 cERROR(1, "server %s of type %s returned"
298 " unexpected error on SMB posix open"
299 ", disabling posix open support."
300 " Check if server update available.",
301 tcon->ses->serverName,
302 tcon->ses->serverNOS);
303 tcon->broken_posix_open = true;
304 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
305 (rc != -EOPNOTSUPP)) /* path not found or net err */
307 /* else fallthrough to retry open the old way on network i/o
311 desiredAccess = cifs_convert_flags(file->f_flags);
313 /*********************************************************************
314 * open flag mapping table:
316 * POSIX Flag CIFS Disposition
317 * ---------- ----------------
318 * O_CREAT FILE_OPEN_IF
319 * O_CREAT | O_EXCL FILE_CREATE
320 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
321 * O_TRUNC FILE_OVERWRITE
322 * none of the above FILE_OPEN
324 * Note that there is not a direct match between disposition
325 * FILE_SUPERSEDE (ie create whether or not file exists although
326 * O_CREAT | O_TRUNC is similar but truncates the existing
327 * file rather than creating a new file as FILE_SUPERSEDE does
328 * (which uses the attributes / metadata passed in on open call)
330 *? O_SYNC is a reasonable match to CIFS writethrough flag
331 *? and the read write flags match reasonably. O_LARGEFILE
332 *? is irrelevant because largefile support is always used
333 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
334 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
335 *********************************************************************/
337 disposition = cifs_get_disposition(file->f_flags);
339 /* BB pass O_SYNC flag through on file attributes .. BB */
341 /* Also refresh inode by passing in file_info buf returned by SMBOpen
342 and calling get_inode_info with returned buf (at least helps
343 non-Unix server case) */
345 /* BB we can not do this if this is the second open of a file
346 and the first handle has writebehind data, we might be
347 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
348 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
354 if (tcon->ses->capabilities & CAP_NT_SMBS)
355 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
356 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
357 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
358 & CIFS_MOUNT_MAP_SPECIAL_CHR);
360 rc = -EIO; /* no NT SMB support fall into legacy open below */
363 /* Old server, try legacy style OpenX */
364 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
365 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
366 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
367 & CIFS_MOUNT_MAP_SPECIAL_CHR);
370 cFYI(1, "cifs_open returned 0x%x", rc);
374 rc = cifs_open_inode_helper(inode, tcon, oplock, buf, full_path, xid);
378 pCifsFile = cifs_new_fileinfo(inode, netfid, file, file->f_path.mnt,
379 tcon, file->f_flags, oplock);
380 if (pCifsFile == NULL) {
385 cifs_fscache_set_inode_cookie(inode, file);
387 if (oplock & CIFS_CREATE_ACTION) {
388 /* time to set mode which we can not set earlier due to
389 problems creating new read-only files */
390 if (tcon->unix_ext) {
391 struct cifs_unix_set_info_args args = {
392 .mode = inode->i_mode,
395 .ctime = NO_CHANGE_64,
396 .atime = NO_CHANGE_64,
397 .mtime = NO_CHANGE_64,
400 CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
402 cifs_sb->mnt_cifs_flags &
403 CIFS_MOUNT_MAP_SPECIAL_CHR);
411 cifs_put_tlink(tlink);
415 /* Try to reacquire byte range locks that were released when session */
416 /* to server was lost */
417 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
421 /* BB list all locks open on this file and relock */
426 static int cifs_reopen_file(struct file *file, bool can_flush)
431 struct cifs_sb_info *cifs_sb;
432 struct cifsTconInfo *tcon;
433 struct cifsFileInfo *pCifsFile;
434 struct cifsInodeInfo *pCifsInode;
436 char *full_path = NULL;
438 int disposition = FILE_OPEN;
441 if (file->private_data)
442 pCifsFile = file->private_data;
447 mutex_lock(&pCifsFile->fh_mutex);
448 if (!pCifsFile->invalidHandle) {
449 mutex_unlock(&pCifsFile->fh_mutex);
455 if (file->f_path.dentry == NULL) {
456 cERROR(1, "no valid name if dentry freed");
459 goto reopen_error_exit;
462 inode = file->f_path.dentry->d_inode;
464 cERROR(1, "inode not valid");
467 goto reopen_error_exit;
470 cifs_sb = CIFS_SB(inode->i_sb);
471 tcon = pCifsFile->tcon;
473 /* can not grab rename sem here because various ops, including
474 those that already have the rename sem can end up causing writepage
475 to get called and if the server was down that means we end up here,
476 and we can never tell if the caller already has the rename_sem */
477 full_path = build_path_from_dentry(file->f_path.dentry);
478 if (full_path == NULL) {
481 mutex_unlock(&pCifsFile->fh_mutex);
486 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
487 inode, file->f_flags, full_path);
494 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
495 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
496 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
497 int oflags = (int) cifs_posix_convert_flags(file->f_flags);
498 /* can not refresh inode info since size could be stale */
499 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
500 cifs_sb->mnt_file_mode /* ignored */,
501 oflags, &oplock, &netfid, xid);
503 cFYI(1, "posix reopen succeeded");
506 /* fallthrough to retry open the old way on errors, especially
507 in the reconnect path it is important to retry hard */
510 desiredAccess = cifs_convert_flags(file->f_flags);
512 /* Can not refresh inode by passing in file_info buf to be returned
513 by SMBOpen and then calling get_inode_info with returned buf
514 since file might have write behind data that needs to be flushed
515 and server version of file size can be stale. If we knew for sure
516 that inode was not dirty locally we could do this */
518 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
519 CREATE_NOT_DIR, &netfid, &oplock, NULL,
520 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
521 CIFS_MOUNT_MAP_SPECIAL_CHR);
523 mutex_unlock(&pCifsFile->fh_mutex);
524 cFYI(1, "cifs_open returned 0x%x", rc);
525 cFYI(1, "oplock: %d", oplock);
528 pCifsFile->netfid = netfid;
529 pCifsFile->invalidHandle = false;
530 mutex_unlock(&pCifsFile->fh_mutex);
531 pCifsInode = CIFS_I(inode);
534 rc = filemap_write_and_wait(inode->i_mapping);
536 CIFS_I(inode)->write_behind_rc = rc;
537 /* temporarily disable caching while we
538 go to server to get inode info */
539 pCifsInode->clientCanCacheAll = false;
540 pCifsInode->clientCanCacheRead = false;
542 rc = cifs_get_inode_info_unix(&inode,
543 full_path, inode->i_sb, xid);
545 rc = cifs_get_inode_info(&inode,
546 full_path, NULL, inode->i_sb,
548 } /* else we are writing out data to server already
549 and could deadlock if we tried to flush data, and
550 since we do not know if we have data that would
551 invalidate the current end of file on the server
552 we can not go to the server to get the new inod
554 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
555 pCifsInode->clientCanCacheAll = true;
556 pCifsInode->clientCanCacheRead = true;
557 cFYI(1, "Exclusive Oplock granted on inode %p",
558 file->f_path.dentry->d_inode);
559 } else if ((oplock & 0xF) == OPLOCK_READ) {
560 pCifsInode->clientCanCacheRead = true;
561 pCifsInode->clientCanCacheAll = false;
563 pCifsInode->clientCanCacheRead = false;
564 pCifsInode->clientCanCacheAll = false;
566 cifs_relock_file(pCifsFile);
574 int cifs_close(struct inode *inode, struct file *file)
578 struct cifs_sb_info *cifs_sb;
579 struct cifsTconInfo *pTcon;
580 struct cifsFileInfo *pSMBFile = file->private_data;
584 cifs_sb = CIFS_SB(inode->i_sb);
585 pTcon = pSMBFile->tcon;
587 struct cifsLockInfo *li, *tmp;
588 write_lock(&GlobalSMBSeslock);
589 pSMBFile->closePend = true;
591 /* no sense reconnecting to close a file that is
593 if (!pTcon->need_reconnect) {
594 write_unlock(&GlobalSMBSeslock);
596 while ((atomic_read(&pSMBFile->count) != 1)
597 && (timeout <= 2048)) {
598 /* Give write a better chance to get to
599 server ahead of the close. We do not
600 want to add a wait_q here as it would
601 increase the memory utilization as
602 the struct would be in each open file,
603 but this should give enough time to
605 cFYI(DBG2, "close delay, write pending");
609 if (!pTcon->need_reconnect &&
610 !pSMBFile->invalidHandle)
611 rc = CIFSSMBClose(xid, pTcon,
614 write_unlock(&GlobalSMBSeslock);
616 write_unlock(&GlobalSMBSeslock);
618 /* Delete any outstanding lock records.
619 We'll lose them when the file is closed anyway. */
620 mutex_lock(&pSMBFile->lock_mutex);
621 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
622 list_del(&li->llist);
625 mutex_unlock(&pSMBFile->lock_mutex);
627 write_lock(&GlobalSMBSeslock);
628 list_del(&pSMBFile->flist);
629 list_del(&pSMBFile->tlist);
630 write_unlock(&GlobalSMBSeslock);
631 cifsFileInfo_put(file->private_data);
632 file->private_data = NULL;
636 read_lock(&GlobalSMBSeslock);
637 if (list_empty(&(CIFS_I(inode)->openFileList))) {
638 cFYI(1, "closing last open instance for inode %p", inode);
639 /* if the file is not open we do not know if we can cache info
640 on this inode, much less write behind and read ahead */
641 CIFS_I(inode)->clientCanCacheRead = false;
642 CIFS_I(inode)->clientCanCacheAll = false;
644 read_unlock(&GlobalSMBSeslock);
645 if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
646 rc = CIFS_I(inode)->write_behind_rc;
651 int cifs_closedir(struct inode *inode, struct file *file)
655 struct cifsFileInfo *pCFileStruct = file->private_data;
658 cFYI(1, "Closedir inode = 0x%p", inode);
663 struct cifsTconInfo *pTcon = pCFileStruct->tcon;
665 cFYI(1, "Freeing private data in close dir");
666 write_lock(&GlobalSMBSeslock);
667 if (!pCFileStruct->srch_inf.endOfSearch &&
668 !pCFileStruct->invalidHandle) {
669 pCFileStruct->invalidHandle = true;
670 write_unlock(&GlobalSMBSeslock);
671 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
672 cFYI(1, "Closing uncompleted readdir with rc %d",
674 /* not much we can do if it fails anyway, ignore rc */
677 write_unlock(&GlobalSMBSeslock);
678 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
680 cFYI(1, "closedir free smb buf in srch struct");
681 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
682 if (pCFileStruct->srch_inf.smallBuf)
683 cifs_small_buf_release(ptmp);
685 cifs_buf_release(ptmp);
687 kfree(file->private_data);
688 file->private_data = NULL;
690 /* BB can we lock the filestruct while this is going on? */
695 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
696 __u64 offset, __u8 lockType)
698 struct cifsLockInfo *li =
699 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
705 mutex_lock(&fid->lock_mutex);
706 list_add(&li->llist, &fid->llist);
707 mutex_unlock(&fid->lock_mutex);
711 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
717 bool wait_flag = false;
718 struct cifs_sb_info *cifs_sb;
719 struct cifsTconInfo *tcon;
721 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
722 bool posix_locking = 0;
724 length = 1 + pfLock->fl_end - pfLock->fl_start;
728 cFYI(1, "Lock parm: 0x%x flockflags: "
729 "0x%x flocktype: 0x%x start: %lld end: %lld",
730 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
733 if (pfLock->fl_flags & FL_POSIX)
735 if (pfLock->fl_flags & FL_FLOCK)
737 if (pfLock->fl_flags & FL_SLEEP) {
738 cFYI(1, "Blocking lock");
741 if (pfLock->fl_flags & FL_ACCESS)
742 cFYI(1, "Process suspended by mandatory locking - "
743 "not implemented yet");
744 if (pfLock->fl_flags & FL_LEASE)
745 cFYI(1, "Lease on file - not implemented yet");
746 if (pfLock->fl_flags &
747 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
748 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
750 if (pfLock->fl_type == F_WRLCK) {
753 } else if (pfLock->fl_type == F_UNLCK) {
756 /* Check if unlock includes more than
758 } else if (pfLock->fl_type == F_RDLCK) {
760 lockType |= LOCKING_ANDX_SHARED_LOCK;
762 } else if (pfLock->fl_type == F_EXLCK) {
765 } else if (pfLock->fl_type == F_SHLCK) {
767 lockType |= LOCKING_ANDX_SHARED_LOCK;
770 cFYI(1, "Unknown type of lock");
772 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
773 tcon = ((struct cifsFileInfo *)file->private_data)->tcon;
775 if (file->private_data == NULL) {
780 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
782 if ((tcon->ses->capabilities & CAP_UNIX) &&
783 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
784 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
786 /* BB add code here to normalize offset and length to
787 account for negative length which we can not accept over the
792 if (lockType & LOCKING_ANDX_SHARED_LOCK)
793 posix_lock_type = CIFS_RDLCK;
795 posix_lock_type = CIFS_WRLCK;
796 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
798 posix_lock_type, wait_flag);
803 /* BB we could chain these into one lock request BB */
804 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
805 0, 1, lockType, 0 /* wait flag */ );
807 rc = CIFSSMBLock(xid, tcon, netfid, length,
808 pfLock->fl_start, 1 /* numUnlock */ ,
809 0 /* numLock */ , lockType,
811 pfLock->fl_type = F_UNLCK;
813 cERROR(1, "Error unlocking previously locked "
814 "range %d during test of lock", rc);
818 /* if rc == ERR_SHARING_VIOLATION ? */
821 if (lockType & LOCKING_ANDX_SHARED_LOCK) {
822 pfLock->fl_type = F_WRLCK;
824 rc = CIFSSMBLock(xid, tcon, netfid, length,
825 pfLock->fl_start, 0, 1,
826 lockType | LOCKING_ANDX_SHARED_LOCK,
829 rc = CIFSSMBLock(xid, tcon, netfid,
830 length, pfLock->fl_start, 1, 0,
832 LOCKING_ANDX_SHARED_LOCK,
834 pfLock->fl_type = F_RDLCK;
836 cERROR(1, "Error unlocking "
837 "previously locked range %d "
838 "during test of lock", rc);
841 pfLock->fl_type = F_WRLCK;
851 if (!numLock && !numUnlock) {
852 /* if no lock or unlock then nothing
853 to do since we do not know what it is */
860 if (lockType & LOCKING_ANDX_SHARED_LOCK)
861 posix_lock_type = CIFS_RDLCK;
863 posix_lock_type = CIFS_WRLCK;
866 posix_lock_type = CIFS_UNLCK;
868 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
870 posix_lock_type, wait_flag);
872 struct cifsFileInfo *fid = file->private_data;
875 rc = CIFSSMBLock(xid, tcon, netfid, length,
877 0, numLock, lockType, wait_flag);
880 /* For Windows locks we must store them. */
881 rc = store_file_lock(fid, length,
882 pfLock->fl_start, lockType);
884 } else if (numUnlock) {
885 /* For each stored lock that this unlock overlaps
886 completely, unlock it. */
888 struct cifsLockInfo *li, *tmp;
891 mutex_lock(&fid->lock_mutex);
892 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
893 if (pfLock->fl_start <= li->offset &&
894 (pfLock->fl_start + length) >=
895 (li->offset + li->length)) {
896 stored_rc = CIFSSMBLock(xid, tcon,
898 li->length, li->offset,
899 1, 0, li->type, false);
903 list_del(&li->llist);
908 mutex_unlock(&fid->lock_mutex);
912 if (pfLock->fl_flags & FL_POSIX)
913 posix_lock_file_wait(file, pfLock);
919 * Set the timeout on write requests past EOF. For some servers (Windows)
920 * these calls can be very long.
922 * If we're writing >10M past the EOF we give a 180s timeout. Anything less
923 * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
924 * The 10M cutoff is totally arbitrary. A better scheme for this would be
925 * welcome if someone wants to suggest one.
927 * We may be able to do a better job with this if there were some way to
928 * declare that a file should be sparse.
931 cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
933 if (offset <= cifsi->server_eof)
935 else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
936 return CIFS_VLONG_OP;
941 /* update the file size (if needed) after a write */
943 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
944 unsigned int bytes_written)
946 loff_t end_of_write = offset + bytes_written;
948 if (end_of_write > cifsi->server_eof)
949 cifsi->server_eof = end_of_write;
952 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
953 size_t write_size, loff_t *poffset)
956 unsigned int bytes_written = 0;
957 unsigned int total_written;
958 struct cifs_sb_info *cifs_sb;
959 struct cifsTconInfo *pTcon;
961 struct cifsFileInfo *open_file;
962 struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
964 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
966 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
967 *poffset, file->f_path.dentry->d_name.name); */
969 if (file->private_data == NULL)
972 open_file = file->private_data;
973 pTcon = open_file->tcon;
975 rc = generic_write_checks(file, poffset, &write_size, 0);
981 long_op = cifs_write_timeout(cifsi, *poffset);
982 for (total_written = 0; write_size > total_written;
983 total_written += bytes_written) {
985 while (rc == -EAGAIN) {
986 if (file->private_data == NULL) {
987 /* file has been closed on us */
989 /* if we have gotten here we have written some data
990 and blocked, and the file has been freed on us while
991 we blocked so return what we managed to write */
992 return total_written;
994 if (open_file->closePend) {
997 return total_written;
1001 if (open_file->invalidHandle) {
1002 /* we could deadlock if we called
1003 filemap_fdatawait from here so tell
1004 reopen_file not to flush data to server
1006 rc = cifs_reopen_file(file, false);
1011 rc = CIFSSMBWrite(xid, pTcon,
1013 min_t(const int, cifs_sb->wsize,
1014 write_size - total_written),
1015 *poffset, &bytes_written,
1016 NULL, write_data + total_written, long_op);
1018 if (rc || (bytes_written == 0)) {
1026 cifs_update_eof(cifsi, *poffset, bytes_written);
1027 *poffset += bytes_written;
1029 long_op = CIFS_STD_OP; /* subsequent writes fast -
1030 15 seconds is plenty */
1033 cifs_stats_bytes_written(pTcon, total_written);
1035 /* since the write may have blocked check these pointers again */
1036 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1037 struct inode *inode = file->f_path.dentry->d_inode;
1038 /* Do not update local mtime - server will set its actual value on write
1039 * inode->i_ctime = inode->i_mtime =
1040 * current_fs_time(inode->i_sb);*/
1041 if (total_written > 0) {
1042 spin_lock(&inode->i_lock);
1043 if (*poffset > file->f_path.dentry->d_inode->i_size)
1044 i_size_write(file->f_path.dentry->d_inode,
1046 spin_unlock(&inode->i_lock);
1048 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1051 return total_written;
1054 static ssize_t cifs_write(struct file *file, const char *write_data,
1055 size_t write_size, loff_t *poffset)
1058 unsigned int bytes_written = 0;
1059 unsigned int total_written;
1060 struct cifs_sb_info *cifs_sb;
1061 struct cifsTconInfo *pTcon;
1063 struct cifsFileInfo *open_file;
1064 struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
1066 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1068 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
1069 *poffset, file->f_path.dentry->d_name.name);
1071 if (file->private_data == NULL)
1073 open_file = file->private_data;
1074 pTcon = open_file->tcon;
1078 long_op = cifs_write_timeout(cifsi, *poffset);
1079 for (total_written = 0; write_size > total_written;
1080 total_written += bytes_written) {
1082 while (rc == -EAGAIN) {
1083 if (file->private_data == NULL) {
1084 /* file has been closed on us */
1086 /* if we have gotten here we have written some data
1087 and blocked, and the file has been freed on us
1088 while we blocked so return what we managed to
1090 return total_written;
1092 if (open_file->closePend) {
1095 return total_written;
1099 if (open_file->invalidHandle) {
1100 /* we could deadlock if we called
1101 filemap_fdatawait from here so tell
1102 reopen_file not to flush data to
1104 rc = cifs_reopen_file(file, false);
1108 if (experimEnabled || (pTcon->ses->server &&
1109 ((pTcon->ses->server->secMode &
1110 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1115 len = min((size_t)cifs_sb->wsize,
1116 write_size - total_written);
1117 /* iov[0] is reserved for smb header */
1118 iov[1].iov_base = (char *)write_data +
1120 iov[1].iov_len = len;
1121 rc = CIFSSMBWrite2(xid, pTcon,
1122 open_file->netfid, len,
1123 *poffset, &bytes_written,
1126 rc = CIFSSMBWrite(xid, pTcon,
1128 min_t(const int, cifs_sb->wsize,
1129 write_size - total_written),
1130 *poffset, &bytes_written,
1131 write_data + total_written,
1134 if (rc || (bytes_written == 0)) {
1142 cifs_update_eof(cifsi, *poffset, bytes_written);
1143 *poffset += bytes_written;
1145 long_op = CIFS_STD_OP; /* subsequent writes fast -
1146 15 seconds is plenty */
1149 cifs_stats_bytes_written(pTcon, total_written);
1151 /* since the write may have blocked check these pointers again */
1152 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1153 /*BB We could make this contingent on superblock ATIME flag too */
1154 /* file->f_path.dentry->d_inode->i_ctime =
1155 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1156 if (total_written > 0) {
1157 spin_lock(&file->f_path.dentry->d_inode->i_lock);
1158 if (*poffset > file->f_path.dentry->d_inode->i_size)
1159 i_size_write(file->f_path.dentry->d_inode,
1161 spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1163 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1166 return total_written;
1169 #ifdef CONFIG_CIFS_EXPERIMENTAL
1170 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode)
1172 struct cifsFileInfo *open_file = NULL;
1174 read_lock(&GlobalSMBSeslock);
1175 /* we could simply get the first_list_entry since write-only entries
1176 are always at the end of the list but since the first entry might
1177 have a close pending, we go through the whole list */
1178 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1179 if (open_file->closePend)
1181 if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
1182 (open_file->pfile->f_flags & O_RDONLY))) {
1183 if (!open_file->invalidHandle) {
1184 /* found a good file */
1185 /* lock it so it will not be closed on us */
1186 cifsFileInfo_get(open_file);
1187 read_unlock(&GlobalSMBSeslock);
1189 } /* else might as well continue, and look for
1190 another, or simply have the caller reopen it
1191 again rather than trying to fix this handle */
1192 } else /* write only file */
1193 break; /* write only files are last so must be done */
1195 read_unlock(&GlobalSMBSeslock);
1200 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1202 struct cifsFileInfo *open_file;
1203 bool any_available = false;
1206 /* Having a null inode here (because mapping->host was set to zero by
1207 the VFS or MM) should not happen but we had reports of on oops (due to
1208 it being zero) during stress testcases so we need to check for it */
1210 if (cifs_inode == NULL) {
1211 cERROR(1, "Null inode passed to cifs_writeable_file");
1216 read_lock(&GlobalSMBSeslock);
1218 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1219 if (open_file->closePend ||
1220 (!any_available && open_file->pid != current->tgid))
1223 if (open_file->pfile &&
1224 ((open_file->pfile->f_flags & O_RDWR) ||
1225 (open_file->pfile->f_flags & O_WRONLY))) {
1226 cifsFileInfo_get(open_file);
1228 if (!open_file->invalidHandle) {
1229 /* found a good writable file */
1230 read_unlock(&GlobalSMBSeslock);
1234 read_unlock(&GlobalSMBSeslock);
1235 /* Had to unlock since following call can block */
1236 rc = cifs_reopen_file(open_file->pfile, false);
1238 if (!open_file->closePend)
1240 else { /* start over in case this was deleted */
1241 /* since the list could be modified */
1242 read_lock(&GlobalSMBSeslock);
1243 cifsFileInfo_put(open_file);
1244 goto refind_writable;
1248 /* if it fails, try another handle if possible -
1249 (we can not do this if closePending since
1250 loop could be modified - in which case we
1251 have to start at the beginning of the list
1252 again. Note that it would be bad
1253 to hold up writepages here (rather than
1254 in caller) with continuous retries */
1255 cFYI(1, "wp failed on reopen file");
1256 read_lock(&GlobalSMBSeslock);
1257 /* can not use this handle, no write
1258 pending on this one after all */
1259 cifsFileInfo_put(open_file);
1261 if (open_file->closePend) /* list could have changed */
1262 goto refind_writable;
1263 /* else we simply continue to the next entry. Thus
1264 we do not loop on reopen errors. If we
1265 can not reopen the file, for example if we
1266 reconnected to a server with another client
1267 racing to delete or lock the file we would not
1268 make progress if we restarted before the beginning
1269 of the loop here. */
1272 /* couldn't find useable FH with same pid, try any available */
1273 if (!any_available) {
1274 any_available = true;
1275 goto refind_writable;
1277 read_unlock(&GlobalSMBSeslock);
1281 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1283 struct address_space *mapping = page->mapping;
1284 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1287 int bytes_written = 0;
1288 struct cifs_sb_info *cifs_sb;
1289 struct inode *inode;
1290 struct cifsFileInfo *open_file;
1292 if (!mapping || !mapping->host)
1295 inode = page->mapping->host;
1296 cifs_sb = CIFS_SB(inode->i_sb);
1298 offset += (loff_t)from;
1299 write_data = kmap(page);
1302 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1307 /* racing with truncate? */
1308 if (offset > mapping->host->i_size) {
1310 return 0; /* don't care */
1313 /* check to make sure that we are not extending the file */
1314 if (mapping->host->i_size - offset < (loff_t)to)
1315 to = (unsigned)(mapping->host->i_size - offset);
1317 open_file = find_writable_file(CIFS_I(mapping->host));
1319 bytes_written = cifs_write(open_file->pfile, write_data,
1321 cifsFileInfo_put(open_file);
1322 /* Does mm or vfs already set times? */
1323 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1324 if ((bytes_written > 0) && (offset))
1326 else if (bytes_written < 0)
1329 cFYI(1, "No writeable filehandles for inode");
1337 static int cifs_writepages(struct address_space *mapping,
1338 struct writeback_control *wbc)
1340 struct backing_dev_info *bdi = mapping->backing_dev_info;
1341 unsigned int bytes_to_write;
1342 unsigned int bytes_written;
1343 struct cifs_sb_info *cifs_sb;
1347 int range_whole = 0;
1354 struct cifsFileInfo *open_file;
1355 struct cifsTconInfo *tcon;
1356 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1358 struct pagevec pvec;
1364 * BB: Is this meaningful for a non-block-device file system?
1365 * If it is, we should test it again after we do I/O
1367 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1368 wbc->encountered_congestion = 1;
1372 cifs_sb = CIFS_SB(mapping->host->i_sb);
1375 * If wsize is smaller that the page cache size, default to writing
1376 * one page at a time via cifs_writepage
1378 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1379 return generic_writepages(mapping, wbc);
1381 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1383 return generic_writepages(mapping, wbc);
1386 * if there's no open file, then this is likely to fail too,
1387 * but it'll at least handle the return. Maybe it should be
1390 open_file = find_writable_file(CIFS_I(mapping->host));
1393 return generic_writepages(mapping, wbc);
1396 tcon = open_file->tcon;
1397 if (!experimEnabled && tcon->ses->server->secMode &
1398 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1399 cifsFileInfo_put(open_file);
1400 return generic_writepages(mapping, wbc);
1402 cifsFileInfo_put(open_file);
1406 pagevec_init(&pvec, 0);
1407 if (wbc->range_cyclic) {
1408 index = mapping->writeback_index; /* Start from prev offset */
1411 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1412 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1413 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1418 while (!done && (index <= end) &&
1419 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1420 PAGECACHE_TAG_DIRTY,
1421 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1430 for (i = 0; i < nr_pages; i++) {
1431 page = pvec.pages[i];
1433 * At this point we hold neither mapping->tree_lock nor
1434 * lock on the page itself: the page may be truncated or
1435 * invalidated (changing page->mapping to NULL), or even
1436 * swizzled back from swapper_space to tmpfs file
1442 else if (!trylock_page(page))
1445 if (unlikely(page->mapping != mapping)) {
1450 if (!wbc->range_cyclic && page->index > end) {
1456 if (next && (page->index != next)) {
1457 /* Not next consecutive page */
1462 if (wbc->sync_mode != WB_SYNC_NONE)
1463 wait_on_page_writeback(page);
1465 if (PageWriteback(page) ||
1466 !clear_page_dirty_for_io(page)) {
1472 * This actually clears the dirty bit in the radix tree.
1473 * See cifs_writepage() for more commentary.
1475 set_page_writeback(page);
1477 if (page_offset(page) >= mapping->host->i_size) {
1480 end_page_writeback(page);
1485 * BB can we get rid of this? pages are held by pvec
1487 page_cache_get(page);
1489 len = min(mapping->host->i_size - page_offset(page),
1490 (loff_t)PAGE_CACHE_SIZE);
1492 /* reserve iov[0] for the smb header */
1494 iov[n_iov].iov_base = kmap(page);
1495 iov[n_iov].iov_len = len;
1496 bytes_to_write += len;
1500 offset = page_offset(page);
1502 next = page->index + 1;
1503 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1507 open_file = find_writable_file(CIFS_I(mapping->host));
1509 cERROR(1, "No writable handles for inode");
1512 long_op = cifs_write_timeout(cifsi, offset);
1513 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1514 bytes_to_write, offset,
1515 &bytes_written, iov, n_iov,
1517 cifsFileInfo_put(open_file);
1518 cifs_update_eof(cifsi, offset, bytes_written);
1521 if (rc || bytes_written < bytes_to_write) {
1522 cERROR(1, "Write2 ret %d, wrote %d",
1524 /* BB what if continued retry is
1525 requested via mount flags? */
1527 set_bit(AS_ENOSPC, &mapping->flags);
1529 set_bit(AS_EIO, &mapping->flags);
1531 cifs_stats_bytes_written(tcon, bytes_written);
1534 for (i = 0; i < n_iov; i++) {
1535 page = pvec.pages[first + i];
1536 /* Should we also set page error on
1537 success rc but too little data written? */
1538 /* BB investigate retry logic on temporary
1539 server crash cases and how recovery works
1540 when page marked as error */
1545 end_page_writeback(page);
1546 page_cache_release(page);
1548 if ((wbc->nr_to_write -= n_iov) <= 0)
1552 /* Need to re-find the pages we skipped */
1553 index = pvec.pages[0]->index + 1;
1555 pagevec_release(&pvec);
1557 if (!scanned && !done) {
1559 * We hit the last page and there is more work to be done: wrap
1560 * back to the start of the file
1566 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1567 mapping->writeback_index = index;
1574 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1580 /* BB add check for wbc flags */
1581 page_cache_get(page);
1582 if (!PageUptodate(page))
1583 cFYI(1, "ppw - page not up to date");
1586 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1588 * A writepage() implementation always needs to do either this,
1589 * or re-dirty the page with "redirty_page_for_writepage()" in
1590 * the case of a failure.
1592 * Just unlocking the page will cause the radix tree tag-bits
1593 * to fail to update with the state of the page correctly.
1595 set_page_writeback(page);
1596 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1597 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1599 end_page_writeback(page);
1600 page_cache_release(page);
1605 static int cifs_write_end(struct file *file, struct address_space *mapping,
1606 loff_t pos, unsigned len, unsigned copied,
1607 struct page *page, void *fsdata)
1610 struct inode *inode = mapping->host;
1612 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1615 if (PageChecked(page)) {
1617 SetPageUptodate(page);
1618 ClearPageChecked(page);
1619 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1620 SetPageUptodate(page);
1622 if (!PageUptodate(page)) {
1624 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1628 /* this is probably better than directly calling
1629 partialpage_write since in this function the file handle is
1630 known which we might as well leverage */
1631 /* BB check if anything else missing out of ppw
1632 such as updating last write time */
1633 page_data = kmap(page);
1634 rc = cifs_write(file, page_data + offset, copied, &pos);
1635 /* if (rc < 0) should we set writebehind rc? */
1642 set_page_dirty(page);
1646 spin_lock(&inode->i_lock);
1647 if (pos > inode->i_size)
1648 i_size_write(inode, pos);
1649 spin_unlock(&inode->i_lock);
1653 page_cache_release(page);
1658 int cifs_fsync(struct file *file, int datasync)
1662 struct cifsTconInfo *tcon;
1663 struct cifsFileInfo *smbfile = file->private_data;
1664 struct inode *inode = file->f_path.dentry->d_inode;
1668 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1669 file->f_path.dentry->d_name.name, datasync);
1671 rc = filemap_write_and_wait(inode->i_mapping);
1673 rc = CIFS_I(inode)->write_behind_rc;
1674 CIFS_I(inode)->write_behind_rc = 0;
1675 tcon = smbfile->tcon;
1676 if (!rc && tcon && smbfile &&
1677 !(CIFS_SB(inode->i_sb)->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1678 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1685 /* static void cifs_sync_page(struct page *page)
1687 struct address_space *mapping;
1688 struct inode *inode;
1689 unsigned long index = page->index;
1690 unsigned int rpages = 0;
1693 cFYI(1, "sync page %p", page);
1694 mapping = page->mapping;
1697 inode = mapping->host;
1701 /* fill in rpages then
1702 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1704 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1714 * As file closes, flush all cached write data for this inode checking
1715 * for write behind errors.
1717 int cifs_flush(struct file *file, fl_owner_t id)
1719 struct inode *inode = file->f_path.dentry->d_inode;
1722 /* Rather than do the steps manually:
1723 lock the inode for writing
1724 loop through pages looking for write behind data (dirty pages)
1725 coalesce into contiguous 16K (or smaller) chunks to write to server
1726 send to server (prefer in parallel)
1727 deal with writebehind errors
1728 unlock inode for writing
1729 filemapfdatawrite appears easier for the time being */
1731 rc = filemap_fdatawrite(inode->i_mapping);
1732 /* reset wb rc if we were able to write out dirty pages */
1734 rc = CIFS_I(inode)->write_behind_rc;
1735 CIFS_I(inode)->write_behind_rc = 0;
1738 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1743 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1744 size_t read_size, loff_t *poffset)
1747 unsigned int bytes_read = 0;
1748 unsigned int total_read = 0;
1749 unsigned int current_read_size;
1750 struct cifs_sb_info *cifs_sb;
1751 struct cifsTconInfo *pTcon;
1753 struct cifsFileInfo *open_file;
1754 char *smb_read_data;
1755 char __user *current_offset;
1756 struct smb_com_read_rsp *pSMBr;
1759 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1761 if (file->private_data == NULL) {
1766 open_file = file->private_data;
1767 pTcon = open_file->tcon;
1769 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1770 cFYI(1, "attempting read on write only file instance");
1772 for (total_read = 0, current_offset = read_data;
1773 read_size > total_read;
1774 total_read += bytes_read, current_offset += bytes_read) {
1775 current_read_size = min_t(const int, read_size - total_read,
1778 smb_read_data = NULL;
1779 while (rc == -EAGAIN) {
1780 int buf_type = CIFS_NO_BUFFER;
1781 if ((open_file->invalidHandle) &&
1782 (!open_file->closePend)) {
1783 rc = cifs_reopen_file(file, true);
1787 rc = CIFSSMBRead(xid, pTcon,
1789 current_read_size, *poffset,
1790 &bytes_read, &smb_read_data,
1792 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1793 if (smb_read_data) {
1794 if (copy_to_user(current_offset,
1796 4 /* RFC1001 length field */ +
1797 le16_to_cpu(pSMBr->DataOffset),
1801 if (buf_type == CIFS_SMALL_BUFFER)
1802 cifs_small_buf_release(smb_read_data);
1803 else if (buf_type == CIFS_LARGE_BUFFER)
1804 cifs_buf_release(smb_read_data);
1805 smb_read_data = NULL;
1808 if (rc || (bytes_read == 0)) {
1816 cifs_stats_bytes_read(pTcon, bytes_read);
1817 *poffset += bytes_read;
1825 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1829 unsigned int bytes_read = 0;
1830 unsigned int total_read;
1831 unsigned int current_read_size;
1832 struct cifs_sb_info *cifs_sb;
1833 struct cifsTconInfo *pTcon;
1835 char *current_offset;
1836 struct cifsFileInfo *open_file;
1837 int buf_type = CIFS_NO_BUFFER;
1840 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1842 if (file->private_data == NULL) {
1847 open_file = file->private_data;
1848 pTcon = open_file->tcon;
1850 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1851 cFYI(1, "attempting read on write only file instance");
1853 for (total_read = 0, current_offset = read_data;
1854 read_size > total_read;
1855 total_read += bytes_read, current_offset += bytes_read) {
1856 current_read_size = min_t(const int, read_size - total_read,
1858 /* For windows me and 9x we do not want to request more
1859 than it negotiated since it will refuse the read then */
1861 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1862 current_read_size = min_t(const int, current_read_size,
1863 pTcon->ses->server->maxBuf - 128);
1866 while (rc == -EAGAIN) {
1867 if ((open_file->invalidHandle) &&
1868 (!open_file->closePend)) {
1869 rc = cifs_reopen_file(file, true);
1873 rc = CIFSSMBRead(xid, pTcon,
1875 current_read_size, *poffset,
1876 &bytes_read, ¤t_offset,
1879 if (rc || (bytes_read == 0)) {
1887 cifs_stats_bytes_read(pTcon, total_read);
1888 *poffset += bytes_read;
1895 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1900 rc = cifs_revalidate_file(file);
1902 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1906 rc = generic_file_mmap(file, vma);
1912 static void cifs_copy_cache_pages(struct address_space *mapping,
1913 struct list_head *pages, int bytes_read, char *data)
1918 while (bytes_read > 0) {
1919 if (list_empty(pages))
1922 page = list_entry(pages->prev, struct page, lru);
1923 list_del(&page->lru);
1925 if (add_to_page_cache_lru(page, mapping, page->index,
1927 page_cache_release(page);
1928 cFYI(1, "Add page cache failed");
1929 data += PAGE_CACHE_SIZE;
1930 bytes_read -= PAGE_CACHE_SIZE;
1933 page_cache_release(page);
1935 target = kmap_atomic(page, KM_USER0);
1937 if (PAGE_CACHE_SIZE > bytes_read) {
1938 memcpy(target, data, bytes_read);
1939 /* zero the tail end of this partial page */
1940 memset(target + bytes_read, 0,
1941 PAGE_CACHE_SIZE - bytes_read);
1944 memcpy(target, data, PAGE_CACHE_SIZE);
1945 bytes_read -= PAGE_CACHE_SIZE;
1947 kunmap_atomic(target, KM_USER0);
1949 flush_dcache_page(page);
1950 SetPageUptodate(page);
1952 data += PAGE_CACHE_SIZE;
1954 /* add page to FS-Cache */
1955 cifs_readpage_to_fscache(mapping->host, page);
1960 static int cifs_readpages(struct file *file, struct address_space *mapping,
1961 struct list_head *page_list, unsigned num_pages)
1967 struct cifs_sb_info *cifs_sb;
1968 struct cifsTconInfo *pTcon;
1969 unsigned int bytes_read = 0;
1970 unsigned int read_size, i;
1971 char *smb_read_data = NULL;
1972 struct smb_com_read_rsp *pSMBr;
1973 struct cifsFileInfo *open_file;
1974 int buf_type = CIFS_NO_BUFFER;
1977 if (file->private_data == NULL) {
1982 open_file = file->private_data;
1983 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1984 pTcon = open_file->tcon;
1987 * Reads as many pages as possible from fscache. Returns -ENOBUFS
1988 * immediately if the cookie is negative
1990 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
1995 cFYI(DBG2, "rpages: num pages %d", num_pages);
1996 for (i = 0; i < num_pages; ) {
1997 unsigned contig_pages;
1998 struct page *tmp_page;
1999 unsigned long expected_index;
2001 if (list_empty(page_list))
2004 page = list_entry(page_list->prev, struct page, lru);
2005 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2007 /* count adjacent pages that we will read into */
2010 list_entry(page_list->prev, struct page, lru)->index;
2011 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2012 if (tmp_page->index == expected_index) {
2018 if (contig_pages + i > num_pages)
2019 contig_pages = num_pages - i;
2021 /* for reads over a certain size could initiate async
2024 read_size = contig_pages * PAGE_CACHE_SIZE;
2025 /* Read size needs to be in multiples of one page */
2026 read_size = min_t(const unsigned int, read_size,
2027 cifs_sb->rsize & PAGE_CACHE_MASK);
2028 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
2029 read_size, contig_pages);
2031 while (rc == -EAGAIN) {
2032 if ((open_file->invalidHandle) &&
2033 (!open_file->closePend)) {
2034 rc = cifs_reopen_file(file, true);
2039 rc = CIFSSMBRead(xid, pTcon,
2042 &bytes_read, &smb_read_data,
2044 /* BB more RC checks ? */
2045 if (rc == -EAGAIN) {
2046 if (smb_read_data) {
2047 if (buf_type == CIFS_SMALL_BUFFER)
2048 cifs_small_buf_release(smb_read_data);
2049 else if (buf_type == CIFS_LARGE_BUFFER)
2050 cifs_buf_release(smb_read_data);
2051 smb_read_data = NULL;
2055 if ((rc < 0) || (smb_read_data == NULL)) {
2056 cFYI(1, "Read error in readpages: %d", rc);
2058 } else if (bytes_read > 0) {
2059 task_io_account_read(bytes_read);
2060 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2061 cifs_copy_cache_pages(mapping, page_list, bytes_read,
2062 smb_read_data + 4 /* RFC1001 hdr */ +
2063 le16_to_cpu(pSMBr->DataOffset));
2065 i += bytes_read >> PAGE_CACHE_SHIFT;
2066 cifs_stats_bytes_read(pTcon, bytes_read);
2067 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2068 i++; /* account for partial page */
2070 /* server copy of file can have smaller size
2072 /* BB do we need to verify this common case ?
2073 this case is ok - if we are at server EOF
2074 we will hit it on next read */
2079 cFYI(1, "No bytes read (%d) at offset %lld . "
2080 "Cleaning remaining pages from readahead list",
2081 bytes_read, offset);
2082 /* BB turn off caching and do new lookup on
2083 file size at server? */
2086 if (smb_read_data) {
2087 if (buf_type == CIFS_SMALL_BUFFER)
2088 cifs_small_buf_release(smb_read_data);
2089 else if (buf_type == CIFS_LARGE_BUFFER)
2090 cifs_buf_release(smb_read_data);
2091 smb_read_data = NULL;
2096 /* need to free smb_read_data buf before exit */
2097 if (smb_read_data) {
2098 if (buf_type == CIFS_SMALL_BUFFER)
2099 cifs_small_buf_release(smb_read_data);
2100 else if (buf_type == CIFS_LARGE_BUFFER)
2101 cifs_buf_release(smb_read_data);
2102 smb_read_data = NULL;
2110 static int cifs_readpage_worker(struct file *file, struct page *page,
2116 /* Is the page cached? */
2117 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2121 page_cache_get(page);
2122 read_data = kmap(page);
2123 /* for reads over a certain size could initiate async read ahead */
2125 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2130 cFYI(1, "Bytes read %d", rc);
2132 file->f_path.dentry->d_inode->i_atime =
2133 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2135 if (PAGE_CACHE_SIZE > rc)
2136 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2138 flush_dcache_page(page);
2139 SetPageUptodate(page);
2141 /* send this page to the cache */
2142 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2148 page_cache_release(page);
2154 static int cifs_readpage(struct file *file, struct page *page)
2156 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2162 if (file->private_data == NULL) {
2168 cFYI(1, "readpage %p at offset %d 0x%x\n",
2169 page, (int)offset, (int)offset);
2171 rc = cifs_readpage_worker(file, page, &offset);
2179 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2181 struct cifsFileInfo *open_file;
2183 read_lock(&GlobalSMBSeslock);
2184 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2185 if (open_file->closePend)
2187 if (open_file->pfile &&
2188 ((open_file->pfile->f_flags & O_RDWR) ||
2189 (open_file->pfile->f_flags & O_WRONLY))) {
2190 read_unlock(&GlobalSMBSeslock);
2194 read_unlock(&GlobalSMBSeslock);
2198 /* We do not want to update the file size from server for inodes
2199 open for write - to avoid races with writepage extending
2200 the file - in the future we could consider allowing
2201 refreshing the inode only on increases in the file size
2202 but this is tricky to do without racing with writebehind
2203 page caching in the current Linux kernel design */
2204 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2209 if (is_inode_writable(cifsInode)) {
2210 /* This inode is open for write at least once */
2211 struct cifs_sb_info *cifs_sb;
2213 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2214 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2215 /* since no page cache to corrupt on directio
2216 we can change size safely */
2220 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2228 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2229 loff_t pos, unsigned len, unsigned flags,
2230 struct page **pagep, void **fsdata)
2232 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2233 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2234 loff_t page_start = pos & PAGE_MASK;
2239 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2241 page = grab_cache_page_write_begin(mapping, index, flags);
2247 if (PageUptodate(page))
2251 * If we write a full page it will be up to date, no need to read from
2252 * the server. If the write is short, we'll end up doing a sync write
2255 if (len == PAGE_CACHE_SIZE)
2259 * optimize away the read when we have an oplock, and we're not
2260 * expecting to use any of the data we'd be reading in. That
2261 * is, when the page lies beyond the EOF, or straddles the EOF
2262 * and the write will cover all of the existing data.
2264 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2265 i_size = i_size_read(mapping->host);
2266 if (page_start >= i_size ||
2267 (offset == 0 && (pos + len) >= i_size)) {
2268 zero_user_segments(page, 0, offset,
2272 * PageChecked means that the parts of the page
2273 * to which we're not writing are considered up
2274 * to date. Once the data is copied to the
2275 * page, it can be set uptodate.
2277 SetPageChecked(page);
2282 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2284 * might as well read a page, it is fast enough. If we get
2285 * an error, we don't need to return it. cifs_write_end will
2286 * do a sync write instead since PG_uptodate isn't set.
2288 cifs_readpage_worker(file, page, &page_start);
2290 /* we could try using another file handle if there is one -
2291 but how would we lock it to prevent close of that handle
2292 racing with this read? In any case
2293 this will be written out by write_end so is fine */
2300 static int cifs_release_page(struct page *page, gfp_t gfp)
2302 if (PagePrivate(page))
2305 return cifs_fscache_release_page(page, gfp);
2308 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2310 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2313 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2316 void cifs_oplock_break(struct work_struct *work)
2318 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2320 struct inode *inode = cfile->pInode;
2321 struct cifsInodeInfo *cinode = CIFS_I(inode);
2324 if (inode && S_ISREG(inode->i_mode)) {
2325 if (cinode->clientCanCacheRead)
2326 break_lease(inode, O_RDONLY);
2328 break_lease(inode, O_WRONLY);
2329 rc = filemap_fdatawrite(inode->i_mapping);
2330 if (cinode->clientCanCacheRead == 0) {
2331 waitrc = filemap_fdatawait(inode->i_mapping);
2332 invalidate_remote_inode(inode);
2337 cinode->write_behind_rc = rc;
2338 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2342 * releasing stale oplock after recent reconnect of smb session using
2343 * a now incorrect file handle is not a data integrity issue but do
2344 * not bother sending an oplock release if session to server still is
2345 * disconnected since oplock already released by the server
2347 if (!cfile->closePend && !cfile->oplock_break_cancelled) {
2348 rc = CIFSSMBLock(0, cfile->tcon, cfile->netfid, 0, 0, 0, 0,
2349 LOCKING_ANDX_OPLOCK_RELEASE, false);
2350 cFYI(1, "Oplock release rc = %d", rc);
2354 * We might have kicked in before is_valid_oplock_break()
2355 * finished grabbing reference for us. Make sure it's done by
2356 * waiting for GlobalSMSSeslock.
2358 write_lock(&GlobalSMBSeslock);
2359 write_unlock(&GlobalSMBSeslock);
2361 cifs_oplock_break_put(cfile);
2364 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2367 cifsFileInfo_get(cfile);
2370 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2373 cifsFileInfo_put(cfile);
2376 const struct address_space_operations cifs_addr_ops = {
2377 .readpage = cifs_readpage,
2378 .readpages = cifs_readpages,
2379 .writepage = cifs_writepage,
2380 .writepages = cifs_writepages,
2381 .write_begin = cifs_write_begin,
2382 .write_end = cifs_write_end,
2383 .set_page_dirty = __set_page_dirty_nobuffers,
2384 .releasepage = cifs_release_page,
2385 .invalidatepage = cifs_invalidate_page,
2386 /* .sync_page = cifs_sync_page, */
2391 * cifs_readpages requires the server to support a buffer large enough to
2392 * contain the header plus one complete page of data. Otherwise, we need
2393 * to leave cifs_readpages out of the address space operations.
2395 const struct address_space_operations cifs_addr_ops_smallbuf = {
2396 .readpage = cifs_readpage,
2397 .writepage = cifs_writepage,
2398 .writepages = cifs_writepages,
2399 .write_begin = cifs_write_begin,
2400 .write_end = cifs_write_end,
2401 .set_page_dirty = __set_page_dirty_nobuffers,
2402 .releasepage = cifs_release_page,
2403 .invalidatepage = cifs_invalidate_page,
2404 /* .sync_page = cifs_sync_page, */