4 * vfs operations that deal with files
6 * Copyright (C) International Business Machines Corp., 2002,2007
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 <asm/div64.h>
37 #include "cifsproto.h"
38 #include "cifs_unicode.h"
39 #include "cifs_debug.h"
40 #include "cifs_fs_sb.h"
42 static inline struct cifsFileInfo *cifs_init_private(
43 struct cifsFileInfo *private_data, struct inode *inode,
44 struct file *file, __u16 netfid)
46 memset(private_data, 0, sizeof(struct cifsFileInfo));
47 private_data->netfid = netfid;
48 private_data->pid = current->tgid;
49 init_MUTEX(&private_data->fh_sem);
50 mutex_init(&private_data->lock_mutex);
51 INIT_LIST_HEAD(&private_data->llist);
52 private_data->pfile = file; /* needed for writepage */
53 private_data->pInode = inode;
54 private_data->invalidHandle = false;
55 private_data->closePend = false;
56 /* we have to track num writers to the inode, since writepages
57 does not tell us which handle the write is for so there can
58 be a close (overlapping with write) of the filehandle that
59 cifs_writepages chose to use */
60 atomic_set(&private_data->wrtPending, 0);
65 static inline int cifs_convert_flags(unsigned int flags)
67 if ((flags & O_ACCMODE) == O_RDONLY)
69 else if ((flags & O_ACCMODE) == O_WRONLY)
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 return (GENERIC_READ | GENERIC_WRITE);
78 return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
79 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
83 static inline fmode_t cifs_posix_convert_flags(unsigned int flags)
85 fmode_t posix_flags = 0;
87 if ((flags & O_ACCMODE) == O_RDONLY)
88 posix_flags = FMODE_READ;
89 else if ((flags & O_ACCMODE) == O_WRONLY)
90 posix_flags = FMODE_WRITE;
91 else if ((flags & O_ACCMODE) == O_RDWR) {
92 /* GENERIC_ALL is too much permission to request
93 can cause unnecessary access denied on create */
94 /* return GENERIC_ALL; */
95 posix_flags = FMODE_READ | FMODE_WRITE;
97 /* can not map O_CREAT or O_EXCL or O_TRUNC flags when
98 reopening a file. They had their effect on the original open */
100 posix_flags |= (fmode_t)O_APPEND;
102 posix_flags |= (fmode_t)O_SYNC;
103 if (flags & O_DIRECTORY)
104 posix_flags |= (fmode_t)O_DIRECTORY;
105 if (flags & O_NOFOLLOW)
106 posix_flags |= (fmode_t)O_NOFOLLOW;
107 if (flags & O_DIRECT)
108 posix_flags |= (fmode_t)O_DIRECT;
113 static inline int cifs_get_disposition(unsigned int flags)
115 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
117 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
118 return FILE_OVERWRITE_IF;
119 else if ((flags & O_CREAT) == O_CREAT)
121 else if ((flags & O_TRUNC) == O_TRUNC)
122 return FILE_OVERWRITE;
127 /* all arguments to this function must be checked for validity in caller */
128 static inline int cifs_posix_open_inode_helper(struct inode *inode,
129 struct file *file, struct cifsInodeInfo *pCifsInode,
130 struct cifsFileInfo *pCifsFile, int oplock, u16 netfid)
132 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
133 /* struct timespec temp; */ /* BB REMOVEME BB */
135 file->private_data = kmalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
136 if (file->private_data == NULL)
138 pCifsFile = cifs_init_private(file->private_data, inode, file, netfid);
139 write_lock(&GlobalSMBSeslock);
140 list_add(&pCifsFile->tlist, &cifs_sb->tcon->openFileList);
142 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
143 if (pCifsInode == NULL) {
144 write_unlock(&GlobalSMBSeslock);
148 /* want handles we can use to read with first
149 in the list so we do not have to walk the
150 list to search for one in write_begin */
151 if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
152 list_add_tail(&pCifsFile->flist,
153 &pCifsInode->openFileList);
155 list_add(&pCifsFile->flist,
156 &pCifsInode->openFileList);
159 if (pCifsInode->clientCanCacheRead) {
160 /* we have the inode open somewhere else
161 no need to discard cache data */
162 goto psx_client_can_cache;
165 /* BB FIXME need to fix this check to move it earlier into posix_open
166 BB fIX following section BB FIXME */
168 /* if not oplocked, invalidate inode pages if mtime or file
170 /* temp = cifs_NTtimeToUnix(le64_to_cpu(buf->LastWriteTime));
171 if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
172 (file->f_path.dentry->d_inode->i_size ==
173 (loff_t)le64_to_cpu(buf->EndOfFile))) {
174 cFYI(1, ("inode unchanged on server"));
176 if (file->f_path.dentry->d_inode->i_mapping) {
177 rc = filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
179 CIFS_I(file->f_path.dentry->d_inode)->write_behind_rc = rc;
181 cFYI(1, ("invalidating remote inode since open detected it "
183 invalidate_remote_inode(file->f_path.dentry->d_inode);
186 psx_client_can_cache:
187 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
188 pCifsInode->clientCanCacheAll = true;
189 pCifsInode->clientCanCacheRead = true;
190 cFYI(1, ("Exclusive Oplock granted on inode %p",
191 file->f_path.dentry->d_inode));
192 } else if ((oplock & 0xF) == OPLOCK_READ)
193 pCifsInode->clientCanCacheRead = true;
195 /* will have to change the unlock if we reenable the
196 filemap_fdatawrite (which does not seem necessary */
197 write_unlock(&GlobalSMBSeslock);
201 /* all arguments to this function must be checked for validity in caller */
202 static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
203 struct cifsInodeInfo *pCifsInode, struct cifsFileInfo *pCifsFile,
204 struct cifsTconInfo *pTcon, int *oplock, FILE_ALL_INFO *buf,
205 char *full_path, int xid)
207 struct timespec temp;
210 /* want handles we can use to read with first
211 in the list so we do not have to walk the
212 list to search for one in write_begin */
213 if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
214 list_add_tail(&pCifsFile->flist,
215 &pCifsInode->openFileList);
217 list_add(&pCifsFile->flist,
218 &pCifsInode->openFileList);
220 write_unlock(&GlobalSMBSeslock);
221 if (pCifsInode->clientCanCacheRead) {
222 /* we have the inode open somewhere else
223 no need to discard cache data */
224 goto client_can_cache;
227 /* BB need same check in cifs_create too? */
228 /* if not oplocked, invalidate inode pages if mtime or file
230 temp = cifs_NTtimeToUnix(le64_to_cpu(buf->LastWriteTime));
231 if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
232 (file->f_path.dentry->d_inode->i_size ==
233 (loff_t)le64_to_cpu(buf->EndOfFile))) {
234 cFYI(1, ("inode unchanged on server"));
236 if (file->f_path.dentry->d_inode->i_mapping) {
237 /* BB no need to lock inode until after invalidate
238 since namei code should already have it locked? */
239 rc = filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
241 CIFS_I(file->f_path.dentry->d_inode)->write_behind_rc = rc;
243 cFYI(1, ("invalidating remote inode since open detected it "
245 invalidate_remote_inode(file->f_path.dentry->d_inode);
250 rc = cifs_get_inode_info_unix(&file->f_path.dentry->d_inode,
251 full_path, inode->i_sb, xid);
253 rc = cifs_get_inode_info(&file->f_path.dentry->d_inode,
254 full_path, buf, inode->i_sb, xid, NULL);
256 if ((*oplock & 0xF) == OPLOCK_EXCLUSIVE) {
257 pCifsInode->clientCanCacheAll = true;
258 pCifsInode->clientCanCacheRead = true;
259 cFYI(1, ("Exclusive Oplock granted on inode %p",
260 file->f_path.dentry->d_inode));
261 } else if ((*oplock & 0xF) == OPLOCK_READ)
262 pCifsInode->clientCanCacheRead = true;
267 int cifs_open(struct inode *inode, struct file *file)
271 struct cifs_sb_info *cifs_sb;
272 struct cifsTconInfo *tcon;
273 struct cifsFileInfo *pCifsFile;
274 struct cifsInodeInfo *pCifsInode;
275 struct list_head *tmp;
276 char *full_path = NULL;
280 FILE_ALL_INFO *buf = NULL;
284 cifs_sb = CIFS_SB(inode->i_sb);
285 tcon = cifs_sb->tcon;
287 if (file->f_flags & O_CREAT) {
288 /* search inode for this file and fill in file->private_data */
289 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
290 read_lock(&GlobalSMBSeslock);
291 list_for_each(tmp, &pCifsInode->openFileList) {
292 pCifsFile = list_entry(tmp, struct cifsFileInfo,
294 if ((pCifsFile->pfile == NULL) &&
295 (pCifsFile->pid == current->tgid)) {
296 /* mode set in cifs_create */
298 /* needed for writepage */
299 pCifsFile->pfile = file;
301 file->private_data = pCifsFile;
305 read_unlock(&GlobalSMBSeslock);
306 if (file->private_data != NULL) {
311 if (file->f_flags & O_EXCL)
312 cERROR(1, ("could not find file instance for "
313 "new file %p", file));
317 full_path = build_path_from_dentry(file->f_path.dentry);
318 if (full_path == NULL) {
323 cFYI(1, ("inode = 0x%p file flags are 0x%x for %s",
324 inode, file->f_flags, full_path));
331 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
332 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
333 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
334 int oflags = (int) cifs_posix_convert_flags(file->f_flags);
335 /* can not refresh inode info since size could be stale */
336 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
337 cifs_sb->mnt_file_mode /* ignored */,
338 oflags, &oplock, &netfid, xid);
340 cFYI(1, ("posix open succeeded"));
341 /* no need for special case handling of setting mode
342 on read only files needed here */
344 cifs_posix_open_inode_helper(inode, file, pCifsInode,
345 pCifsFile, oplock, netfid);
347 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
348 (rc != -EOPNOTSUPP)) /* path not found or net err */
350 /* fallthrough to retry open the old way on operation
351 not supported or DFS errors */
354 desiredAccess = cifs_convert_flags(file->f_flags);
356 /*********************************************************************
357 * open flag mapping table:
359 * POSIX Flag CIFS Disposition
360 * ---------- ----------------
361 * O_CREAT FILE_OPEN_IF
362 * O_CREAT | O_EXCL FILE_CREATE
363 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
364 * O_TRUNC FILE_OVERWRITE
365 * none of the above FILE_OPEN
367 * Note that there is not a direct match between disposition
368 * FILE_SUPERSEDE (ie create whether or not file exists although
369 * O_CREAT | O_TRUNC is similar but truncates the existing
370 * file rather than creating a new file as FILE_SUPERSEDE does
371 * (which uses the attributes / metadata passed in on open call)
373 *? O_SYNC is a reasonable match to CIFS writethrough flag
374 *? and the read write flags match reasonably. O_LARGEFILE
375 *? is irrelevant because largefile support is always used
376 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
377 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
378 *********************************************************************/
380 disposition = cifs_get_disposition(file->f_flags);
382 /* BB pass O_SYNC flag through on file attributes .. BB */
384 /* Also refresh inode by passing in file_info buf returned by SMBOpen
385 and calling get_inode_info with returned buf (at least helps
386 non-Unix server case) */
388 /* BB we can not do this if this is the second open of a file
389 and the first handle has writebehind data, we might be
390 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
391 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
397 if (cifs_sb->tcon->ses->capabilities & CAP_NT_SMBS)
398 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
399 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
400 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
401 & CIFS_MOUNT_MAP_SPECIAL_CHR);
403 rc = -EIO; /* no NT SMB support fall into legacy open below */
406 /* Old server, try legacy style OpenX */
407 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
408 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
409 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
410 & CIFS_MOUNT_MAP_SPECIAL_CHR);
413 cFYI(1, ("cifs_open returned 0x%x", rc));
417 kmalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
418 if (file->private_data == NULL) {
422 pCifsFile = cifs_init_private(file->private_data, inode, file, netfid);
423 write_lock(&GlobalSMBSeslock);
424 list_add(&pCifsFile->tlist, &tcon->openFileList);
426 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
428 rc = cifs_open_inode_helper(inode, file, pCifsInode,
430 &oplock, buf, full_path, xid);
432 write_unlock(&GlobalSMBSeslock);
435 if (oplock & CIFS_CREATE_ACTION) {
436 /* time to set mode which we can not set earlier due to
437 problems creating new read-only files */
438 if (tcon->unix_ext) {
439 struct cifs_unix_set_info_args args = {
440 .mode = inode->i_mode,
443 .ctime = NO_CHANGE_64,
444 .atime = NO_CHANGE_64,
445 .mtime = NO_CHANGE_64,
448 CIFSSMBUnixSetInfo(xid, tcon, full_path, &args,
450 cifs_sb->mnt_cifs_flags &
451 CIFS_MOUNT_MAP_SPECIAL_CHR);
462 /* Try to reacquire byte range locks that were released when session */
463 /* to server was lost */
464 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
468 /* BB list all locks open on this file and relock */
473 static int cifs_reopen_file(struct file *file, bool can_flush)
477 struct cifs_sb_info *cifs_sb;
478 struct cifsTconInfo *tcon;
479 struct cifsFileInfo *pCifsFile;
480 struct cifsInodeInfo *pCifsInode;
482 char *full_path = NULL;
484 int disposition = FILE_OPEN;
487 if (file->private_data)
488 pCifsFile = (struct cifsFileInfo *)file->private_data;
493 down(&pCifsFile->fh_sem);
494 if (!pCifsFile->invalidHandle) {
495 up(&pCifsFile->fh_sem);
500 if (file->f_path.dentry == NULL) {
501 cERROR(1, ("no valid name if dentry freed"));
504 goto reopen_error_exit;
507 inode = file->f_path.dentry->d_inode;
509 cERROR(1, ("inode not valid"));
512 goto reopen_error_exit;
515 cifs_sb = CIFS_SB(inode->i_sb);
516 tcon = cifs_sb->tcon;
518 /* can not grab rename sem here because various ops, including
519 those that already have the rename sem can end up causing writepage
520 to get called and if the server was down that means we end up here,
521 and we can never tell if the caller already has the rename_sem */
522 full_path = build_path_from_dentry(file->f_path.dentry);
523 if (full_path == NULL) {
526 up(&pCifsFile->fh_sem);
531 cFYI(1, ("inode = 0x%p file flags 0x%x for %s",
532 inode, file->f_flags, full_path));
539 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
540 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
541 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
542 int oflags = (int) cifs_posix_convert_flags(file->f_flags);
543 /* can not refresh inode info since size could be stale */
544 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
545 cifs_sb->mnt_file_mode /* ignored */,
546 oflags, &oplock, &netfid, xid);
548 cFYI(1, ("posix reopen succeeded"));
551 /* fallthrough to retry open the old way on errors, especially
552 in the reconnect path it is important to retry hard */
555 desiredAccess = cifs_convert_flags(file->f_flags);
557 /* Can not refresh inode by passing in file_info buf to be returned
558 by SMBOpen and then calling get_inode_info with returned buf
559 since file might have write behind data that needs to be flushed
560 and server version of file size can be stale. If we knew for sure
561 that inode was not dirty locally we could do this */
563 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
564 CREATE_NOT_DIR, &netfid, &oplock, NULL,
565 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
566 CIFS_MOUNT_MAP_SPECIAL_CHR);
568 up(&pCifsFile->fh_sem);
569 cFYI(1, ("cifs_open returned 0x%x", rc));
570 cFYI(1, ("oplock: %d", oplock));
573 pCifsFile->netfid = netfid;
574 pCifsFile->invalidHandle = false;
575 up(&pCifsFile->fh_sem);
576 pCifsInode = CIFS_I(inode);
579 rc = filemap_write_and_wait(inode->i_mapping);
581 CIFS_I(inode)->write_behind_rc = rc;
582 /* temporarily disable caching while we
583 go to server to get inode info */
584 pCifsInode->clientCanCacheAll = false;
585 pCifsInode->clientCanCacheRead = false;
587 rc = cifs_get_inode_info_unix(&inode,
588 full_path, inode->i_sb, xid);
590 rc = cifs_get_inode_info(&inode,
591 full_path, NULL, inode->i_sb,
593 } /* else we are writing out data to server already
594 and could deadlock if we tried to flush data, and
595 since we do not know if we have data that would
596 invalidate the current end of file on the server
597 we can not go to the server to get the new inod
599 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
600 pCifsInode->clientCanCacheAll = true;
601 pCifsInode->clientCanCacheRead = true;
602 cFYI(1, ("Exclusive Oplock granted on inode %p",
603 file->f_path.dentry->d_inode));
604 } else if ((oplock & 0xF) == OPLOCK_READ) {
605 pCifsInode->clientCanCacheRead = true;
606 pCifsInode->clientCanCacheAll = false;
608 pCifsInode->clientCanCacheRead = false;
609 pCifsInode->clientCanCacheAll = false;
611 cifs_relock_file(pCifsFile);
619 int cifs_close(struct inode *inode, struct file *file)
623 struct cifs_sb_info *cifs_sb;
624 struct cifsTconInfo *pTcon;
625 struct cifsFileInfo *pSMBFile =
626 (struct cifsFileInfo *)file->private_data;
630 cifs_sb = CIFS_SB(inode->i_sb);
631 pTcon = cifs_sb->tcon;
633 struct cifsLockInfo *li, *tmp;
634 write_lock(&GlobalSMBSeslock);
635 pSMBFile->closePend = true;
637 /* no sense reconnecting to close a file that is
639 if (!pTcon->need_reconnect) {
640 write_unlock(&GlobalSMBSeslock);
642 while ((atomic_read(&pSMBFile->wrtPending) != 0)
643 && (timeout <= 2048)) {
644 /* Give write a better chance to get to
645 server ahead of the close. We do not
646 want to add a wait_q here as it would
647 increase the memory utilization as
648 the struct would be in each open file,
649 but this should give enough time to
652 ("close delay, write pending"));
656 if (atomic_read(&pSMBFile->wrtPending))
657 cERROR(1, ("close with pending write"));
658 if (!pTcon->need_reconnect &&
659 !pSMBFile->invalidHandle)
660 rc = CIFSSMBClose(xid, pTcon,
663 write_unlock(&GlobalSMBSeslock);
665 write_unlock(&GlobalSMBSeslock);
667 /* Delete any outstanding lock records.
668 We'll lose them when the file is closed anyway. */
669 mutex_lock(&pSMBFile->lock_mutex);
670 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
671 list_del(&li->llist);
674 mutex_unlock(&pSMBFile->lock_mutex);
676 write_lock(&GlobalSMBSeslock);
677 list_del(&pSMBFile->flist);
678 list_del(&pSMBFile->tlist);
679 write_unlock(&GlobalSMBSeslock);
681 /* We waited above to give the SMBWrite a chance to issue
682 on the wire (so we do not get SMBWrite returning EBADF
683 if writepages is racing with close. Note that writepages
684 does not specify a file handle, so it is possible for a file
685 to be opened twice, and the application close the "wrong"
686 file handle - in these cases we delay long enough to allow
687 the SMBWrite to get on the wire before the SMB Close.
688 We allow total wait here over 45 seconds, more than
689 oplock break time, and more than enough to allow any write
690 to complete on the server, or to time out on the client */
691 while ((atomic_read(&pSMBFile->wrtPending) != 0)
692 && (timeout <= 50000)) {
693 cERROR(1, ("writes pending, delay free of handle"));
697 kfree(file->private_data);
698 file->private_data = NULL;
702 read_lock(&GlobalSMBSeslock);
703 if (list_empty(&(CIFS_I(inode)->openFileList))) {
704 cFYI(1, ("closing last open instance for inode %p", inode));
705 /* if the file is not open we do not know if we can cache info
706 on this inode, much less write behind and read ahead */
707 CIFS_I(inode)->clientCanCacheRead = false;
708 CIFS_I(inode)->clientCanCacheAll = false;
710 read_unlock(&GlobalSMBSeslock);
711 if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
712 rc = CIFS_I(inode)->write_behind_rc;
717 int cifs_closedir(struct inode *inode, struct file *file)
721 struct cifsFileInfo *pCFileStruct =
722 (struct cifsFileInfo *)file->private_data;
725 cFYI(1, ("Closedir inode = 0x%p", inode));
730 struct cifsTconInfo *pTcon;
731 struct cifs_sb_info *cifs_sb =
732 CIFS_SB(file->f_path.dentry->d_sb);
734 pTcon = cifs_sb->tcon;
736 cFYI(1, ("Freeing private data in close dir"));
737 write_lock(&GlobalSMBSeslock);
738 if (!pCFileStruct->srch_inf.endOfSearch &&
739 !pCFileStruct->invalidHandle) {
740 pCFileStruct->invalidHandle = true;
741 write_unlock(&GlobalSMBSeslock);
742 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
743 cFYI(1, ("Closing uncompleted readdir with rc %d",
745 /* not much we can do if it fails anyway, ignore rc */
748 write_unlock(&GlobalSMBSeslock);
749 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
751 cFYI(1, ("closedir free smb buf in srch struct"));
752 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
753 if (pCFileStruct->srch_inf.smallBuf)
754 cifs_small_buf_release(ptmp);
756 cifs_buf_release(ptmp);
758 kfree(file->private_data);
759 file->private_data = NULL;
761 /* BB can we lock the filestruct while this is going on? */
766 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
767 __u64 offset, __u8 lockType)
769 struct cifsLockInfo *li =
770 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
776 mutex_lock(&fid->lock_mutex);
777 list_add(&li->llist, &fid->llist);
778 mutex_unlock(&fid->lock_mutex);
782 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
788 bool wait_flag = false;
789 struct cifs_sb_info *cifs_sb;
790 struct cifsTconInfo *tcon;
792 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
793 bool posix_locking = 0;
795 length = 1 + pfLock->fl_end - pfLock->fl_start;
799 cFYI(1, ("Lock parm: 0x%x flockflags: "
800 "0x%x flocktype: 0x%x start: %lld end: %lld",
801 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
804 if (pfLock->fl_flags & FL_POSIX)
806 if (pfLock->fl_flags & FL_FLOCK)
808 if (pfLock->fl_flags & FL_SLEEP) {
809 cFYI(1, ("Blocking lock"));
812 if (pfLock->fl_flags & FL_ACCESS)
813 cFYI(1, ("Process suspended by mandatory locking - "
814 "not implemented yet"));
815 if (pfLock->fl_flags & FL_LEASE)
816 cFYI(1, ("Lease on file - not implemented yet"));
817 if (pfLock->fl_flags &
818 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
819 cFYI(1, ("Unknown lock flags 0x%x", pfLock->fl_flags));
821 if (pfLock->fl_type == F_WRLCK) {
822 cFYI(1, ("F_WRLCK "));
824 } else if (pfLock->fl_type == F_UNLCK) {
825 cFYI(1, ("F_UNLCK"));
827 /* Check if unlock includes more than
829 } else if (pfLock->fl_type == F_RDLCK) {
830 cFYI(1, ("F_RDLCK"));
831 lockType |= LOCKING_ANDX_SHARED_LOCK;
833 } else if (pfLock->fl_type == F_EXLCK) {
834 cFYI(1, ("F_EXLCK"));
836 } else if (pfLock->fl_type == F_SHLCK) {
837 cFYI(1, ("F_SHLCK"));
838 lockType |= LOCKING_ANDX_SHARED_LOCK;
841 cFYI(1, ("Unknown type of lock"));
843 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
844 tcon = cifs_sb->tcon;
846 if (file->private_data == NULL) {
850 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
852 if ((tcon->ses->capabilities & CAP_UNIX) &&
853 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
854 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
856 /* BB add code here to normalize offset and length to
857 account for negative length which we can not accept over the
862 if (lockType & LOCKING_ANDX_SHARED_LOCK)
863 posix_lock_type = CIFS_RDLCK;
865 posix_lock_type = CIFS_WRLCK;
866 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
868 posix_lock_type, wait_flag);
873 /* BB we could chain these into one lock request BB */
874 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
875 0, 1, lockType, 0 /* wait flag */ );
877 rc = CIFSSMBLock(xid, tcon, netfid, length,
878 pfLock->fl_start, 1 /* numUnlock */ ,
879 0 /* numLock */ , lockType,
881 pfLock->fl_type = F_UNLCK;
883 cERROR(1, ("Error unlocking previously locked "
884 "range %d during test of lock", rc));
888 /* if rc == ERR_SHARING_VIOLATION ? */
889 rc = 0; /* do not change lock type to unlock
890 since range in use */
897 if (!numLock && !numUnlock) {
898 /* if no lock or unlock then nothing
899 to do since we do not know what it is */
906 if (lockType & LOCKING_ANDX_SHARED_LOCK)
907 posix_lock_type = CIFS_RDLCK;
909 posix_lock_type = CIFS_WRLCK;
912 posix_lock_type = CIFS_UNLCK;
914 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
916 posix_lock_type, wait_flag);
918 struct cifsFileInfo *fid =
919 (struct cifsFileInfo *)file->private_data;
922 rc = CIFSSMBLock(xid, tcon, netfid, length,
924 0, numLock, lockType, wait_flag);
927 /* For Windows locks we must store them. */
928 rc = store_file_lock(fid, length,
929 pfLock->fl_start, lockType);
931 } else if (numUnlock) {
932 /* For each stored lock that this unlock overlaps
933 completely, unlock it. */
935 struct cifsLockInfo *li, *tmp;
938 mutex_lock(&fid->lock_mutex);
939 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
940 if (pfLock->fl_start <= li->offset &&
941 (pfLock->fl_start + length) >=
942 (li->offset + li->length)) {
943 stored_rc = CIFSSMBLock(xid, tcon,
945 li->length, li->offset,
946 1, 0, li->type, false);
950 list_del(&li->llist);
954 mutex_unlock(&fid->lock_mutex);
958 if (pfLock->fl_flags & FL_POSIX)
959 posix_lock_file_wait(file, pfLock);
964 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
965 size_t write_size, loff_t *poffset)
968 unsigned int bytes_written = 0;
969 unsigned int total_written;
970 struct cifs_sb_info *cifs_sb;
971 struct cifsTconInfo *pTcon;
973 struct cifsFileInfo *open_file;
975 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
977 pTcon = cifs_sb->tcon;
980 (" write %d bytes to offset %lld of %s", write_size,
981 *poffset, file->f_path.dentry->d_name.name)); */
983 if (file->private_data == NULL)
985 open_file = (struct cifsFileInfo *) file->private_data;
987 rc = generic_write_checks(file, poffset, &write_size, 0);
993 if (*poffset > file->f_path.dentry->d_inode->i_size)
994 long_op = CIFS_VLONG_OP; /* writes past EOF take long time */
996 long_op = CIFS_LONG_OP;
998 for (total_written = 0; write_size > total_written;
999 total_written += bytes_written) {
1001 while (rc == -EAGAIN) {
1002 if (file->private_data == NULL) {
1003 /* file has been closed on us */
1005 /* if we have gotten here we have written some data
1006 and blocked, and the file has been freed on us while
1007 we blocked so return what we managed to write */
1008 return total_written;
1010 if (open_file->closePend) {
1013 return total_written;
1017 if (open_file->invalidHandle) {
1018 /* we could deadlock if we called
1019 filemap_fdatawait from here so tell
1020 reopen_file not to flush data to server
1022 rc = cifs_reopen_file(file, false);
1027 rc = CIFSSMBWrite(xid, pTcon,
1029 min_t(const int, cifs_sb->wsize,
1030 write_size - total_written),
1031 *poffset, &bytes_written,
1032 NULL, write_data + total_written, long_op);
1034 if (rc || (bytes_written == 0)) {
1042 *poffset += bytes_written;
1043 long_op = CIFS_STD_OP; /* subsequent writes fast -
1044 15 seconds is plenty */
1047 cifs_stats_bytes_written(pTcon, total_written);
1049 /* since the write may have blocked check these pointers again */
1050 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1051 struct inode *inode = file->f_path.dentry->d_inode;
1052 /* Do not update local mtime - server will set its actual value on write
1053 * inode->i_ctime = inode->i_mtime =
1054 * current_fs_time(inode->i_sb);*/
1055 if (total_written > 0) {
1056 spin_lock(&inode->i_lock);
1057 if (*poffset > file->f_path.dentry->d_inode->i_size)
1058 i_size_write(file->f_path.dentry->d_inode,
1060 spin_unlock(&inode->i_lock);
1062 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1065 return total_written;
1068 static ssize_t cifs_write(struct file *file, const char *write_data,
1069 size_t write_size, loff_t *poffset)
1072 unsigned int bytes_written = 0;
1073 unsigned int total_written;
1074 struct cifs_sb_info *cifs_sb;
1075 struct cifsTconInfo *pTcon;
1077 struct cifsFileInfo *open_file;
1079 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1081 pTcon = cifs_sb->tcon;
1083 cFYI(1, ("write %zd bytes to offset %lld of %s", write_size,
1084 *poffset, file->f_path.dentry->d_name.name));
1086 if (file->private_data == NULL)
1088 open_file = (struct cifsFileInfo *)file->private_data;
1092 if (*poffset > file->f_path.dentry->d_inode->i_size)
1093 long_op = CIFS_VLONG_OP; /* writes past EOF can be slow */
1095 long_op = CIFS_LONG_OP;
1097 for (total_written = 0; write_size > total_written;
1098 total_written += bytes_written) {
1100 while (rc == -EAGAIN) {
1101 if (file->private_data == NULL) {
1102 /* file has been closed on us */
1104 /* if we have gotten here we have written some data
1105 and blocked, and the file has been freed on us
1106 while we blocked so return what we managed to
1108 return total_written;
1110 if (open_file->closePend) {
1113 return total_written;
1117 if (open_file->invalidHandle) {
1118 /* we could deadlock if we called
1119 filemap_fdatawait from here so tell
1120 reopen_file not to flush data to
1122 rc = cifs_reopen_file(file, false);
1126 if (experimEnabled || (pTcon->ses->server &&
1127 ((pTcon->ses->server->secMode &
1128 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1133 len = min((size_t)cifs_sb->wsize,
1134 write_size - total_written);
1135 /* iov[0] is reserved for smb header */
1136 iov[1].iov_base = (char *)write_data +
1138 iov[1].iov_len = len;
1139 rc = CIFSSMBWrite2(xid, pTcon,
1140 open_file->netfid, len,
1141 *poffset, &bytes_written,
1144 rc = CIFSSMBWrite(xid, pTcon,
1146 min_t(const int, cifs_sb->wsize,
1147 write_size - total_written),
1148 *poffset, &bytes_written,
1149 write_data + total_written,
1152 if (rc || (bytes_written == 0)) {
1160 *poffset += bytes_written;
1161 long_op = CIFS_STD_OP; /* subsequent writes fast -
1162 15 seconds is plenty */
1165 cifs_stats_bytes_written(pTcon, total_written);
1167 /* since the write may have blocked check these pointers again */
1168 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1169 /*BB We could make this contingent on superblock ATIME flag too */
1170 /* file->f_path.dentry->d_inode->i_ctime =
1171 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1172 if (total_written > 0) {
1173 spin_lock(&file->f_path.dentry->d_inode->i_lock);
1174 if (*poffset > file->f_path.dentry->d_inode->i_size)
1175 i_size_write(file->f_path.dentry->d_inode,
1177 spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1179 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1182 return total_written;
1185 #ifdef CONFIG_CIFS_EXPERIMENTAL
1186 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode)
1188 struct cifsFileInfo *open_file = NULL;
1190 read_lock(&GlobalSMBSeslock);
1191 /* we could simply get the first_list_entry since write-only entries
1192 are always at the end of the list but since the first entry might
1193 have a close pending, we go through the whole list */
1194 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1195 if (open_file->closePend)
1197 if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
1198 (open_file->pfile->f_flags & O_RDONLY))) {
1199 if (!open_file->invalidHandle) {
1200 /* found a good file */
1201 /* lock it so it will not be closed on us */
1202 atomic_inc(&open_file->wrtPending);
1203 read_unlock(&GlobalSMBSeslock);
1205 } /* else might as well continue, and look for
1206 another, or simply have the caller reopen it
1207 again rather than trying to fix this handle */
1208 } else /* write only file */
1209 break; /* write only files are last so must be done */
1211 read_unlock(&GlobalSMBSeslock);
1216 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1218 struct cifsFileInfo *open_file;
1219 bool any_available = false;
1222 /* Having a null inode here (because mapping->host was set to zero by
1223 the VFS or MM) should not happen but we had reports of on oops (due to
1224 it being zero) during stress testcases so we need to check for it */
1226 if (cifs_inode == NULL) {
1227 cERROR(1, ("Null inode passed to cifs_writeable_file"));
1232 read_lock(&GlobalSMBSeslock);
1234 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1235 if (open_file->closePend ||
1236 (!any_available && open_file->pid != current->tgid))
1239 if (open_file->pfile &&
1240 ((open_file->pfile->f_flags & O_RDWR) ||
1241 (open_file->pfile->f_flags & O_WRONLY))) {
1242 atomic_inc(&open_file->wrtPending);
1244 if (!open_file->invalidHandle) {
1245 /* found a good writable file */
1246 read_unlock(&GlobalSMBSeslock);
1250 read_unlock(&GlobalSMBSeslock);
1251 /* Had to unlock since following call can block */
1252 rc = cifs_reopen_file(open_file->pfile, false);
1254 if (!open_file->closePend)
1256 else { /* start over in case this was deleted */
1257 /* since the list could be modified */
1258 read_lock(&GlobalSMBSeslock);
1259 atomic_dec(&open_file->wrtPending);
1260 goto refind_writable;
1264 /* if it fails, try another handle if possible -
1265 (we can not do this if closePending since
1266 loop could be modified - in which case we
1267 have to start at the beginning of the list
1268 again. Note that it would be bad
1269 to hold up writepages here (rather than
1270 in caller) with continuous retries */
1271 cFYI(1, ("wp failed on reopen file"));
1272 read_lock(&GlobalSMBSeslock);
1273 /* can not use this handle, no write
1274 pending on this one after all */
1275 atomic_dec(&open_file->wrtPending);
1277 if (open_file->closePend) /* list could have changed */
1278 goto refind_writable;
1279 /* else we simply continue to the next entry. Thus
1280 we do not loop on reopen errors. If we
1281 can not reopen the file, for example if we
1282 reconnected to a server with another client
1283 racing to delete or lock the file we would not
1284 make progress if we restarted before the beginning
1285 of the loop here. */
1288 /* couldn't find useable FH with same pid, try any available */
1289 if (!any_available) {
1290 any_available = true;
1291 goto refind_writable;
1293 read_unlock(&GlobalSMBSeslock);
1297 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1299 struct address_space *mapping = page->mapping;
1300 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1303 int bytes_written = 0;
1304 struct cifs_sb_info *cifs_sb;
1305 struct cifsTconInfo *pTcon;
1306 struct inode *inode;
1307 struct cifsFileInfo *open_file;
1309 if (!mapping || !mapping->host)
1312 inode = page->mapping->host;
1313 cifs_sb = CIFS_SB(inode->i_sb);
1314 pTcon = cifs_sb->tcon;
1316 offset += (loff_t)from;
1317 write_data = kmap(page);
1320 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1325 /* racing with truncate? */
1326 if (offset > mapping->host->i_size) {
1328 return 0; /* don't care */
1331 /* check to make sure that we are not extending the file */
1332 if (mapping->host->i_size - offset < (loff_t)to)
1333 to = (unsigned)(mapping->host->i_size - offset);
1335 open_file = find_writable_file(CIFS_I(mapping->host));
1337 bytes_written = cifs_write(open_file->pfile, write_data,
1339 atomic_dec(&open_file->wrtPending);
1340 /* Does mm or vfs already set times? */
1341 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1342 if ((bytes_written > 0) && (offset))
1344 else if (bytes_written < 0)
1347 cFYI(1, ("No writeable filehandles for inode"));
1355 static int cifs_writepages(struct address_space *mapping,
1356 struct writeback_control *wbc)
1358 struct backing_dev_info *bdi = mapping->backing_dev_info;
1359 unsigned int bytes_to_write;
1360 unsigned int bytes_written;
1361 struct cifs_sb_info *cifs_sb;
1365 int range_whole = 0;
1372 struct cifsFileInfo *open_file;
1374 struct pagevec pvec;
1379 cifs_sb = CIFS_SB(mapping->host->i_sb);
1382 * If wsize is smaller that the page cache size, default to writing
1383 * one page at a time via cifs_writepage
1385 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1386 return generic_writepages(mapping, wbc);
1388 if ((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1389 if (cifs_sb->tcon->ses->server->secMode &
1390 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1391 if (!experimEnabled)
1392 return generic_writepages(mapping, wbc);
1394 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1396 return generic_writepages(mapping, wbc);
1400 * BB: Is this meaningful for a non-block-device file system?
1401 * If it is, we should test it again after we do I/O
1403 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1404 wbc->encountered_congestion = 1;
1411 pagevec_init(&pvec, 0);
1412 if (wbc->range_cyclic) {
1413 index = mapping->writeback_index; /* Start from prev offset */
1416 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1417 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1418 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1423 while (!done && (index <= end) &&
1424 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1425 PAGECACHE_TAG_DIRTY,
1426 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1435 for (i = 0; i < nr_pages; i++) {
1436 page = pvec.pages[i];
1438 * At this point we hold neither mapping->tree_lock nor
1439 * lock on the page itself: the page may be truncated or
1440 * invalidated (changing page->mapping to NULL), or even
1441 * swizzled back from swapper_space to tmpfs file
1447 else if (!trylock_page(page))
1450 if (unlikely(page->mapping != mapping)) {
1455 if (!wbc->range_cyclic && page->index > end) {
1461 if (next && (page->index != next)) {
1462 /* Not next consecutive page */
1467 if (wbc->sync_mode != WB_SYNC_NONE)
1468 wait_on_page_writeback(page);
1470 if (PageWriteback(page) ||
1471 !clear_page_dirty_for_io(page)) {
1477 * This actually clears the dirty bit in the radix tree.
1478 * See cifs_writepage() for more commentary.
1480 set_page_writeback(page);
1482 if (page_offset(page) >= mapping->host->i_size) {
1485 end_page_writeback(page);
1490 * BB can we get rid of this? pages are held by pvec
1492 page_cache_get(page);
1494 len = min(mapping->host->i_size - page_offset(page),
1495 (loff_t)PAGE_CACHE_SIZE);
1497 /* reserve iov[0] for the smb header */
1499 iov[n_iov].iov_base = kmap(page);
1500 iov[n_iov].iov_len = len;
1501 bytes_to_write += len;
1505 offset = page_offset(page);
1507 next = page->index + 1;
1508 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1512 /* Search for a writable handle every time we call
1513 * CIFSSMBWrite2. We can't rely on the last handle
1514 * we used to still be valid
1516 open_file = find_writable_file(CIFS_I(mapping->host));
1518 cERROR(1, ("No writable handles for inode"));
1521 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1523 bytes_to_write, offset,
1524 &bytes_written, iov, n_iov,
1526 atomic_dec(&open_file->wrtPending);
1527 if (rc || bytes_written < bytes_to_write) {
1528 cERROR(1, ("Write2 ret %d, wrote %d",
1529 rc, bytes_written));
1530 /* BB what if continued retry is
1531 requested via mount flags? */
1533 set_bit(AS_ENOSPC, &mapping->flags);
1535 set_bit(AS_EIO, &mapping->flags);
1537 cifs_stats_bytes_written(cifs_sb->tcon,
1541 for (i = 0; i < n_iov; i++) {
1542 page = pvec.pages[first + i];
1543 /* Should we also set page error on
1544 success rc but too little data written? */
1545 /* BB investigate retry logic on temporary
1546 server crash cases and how recovery works
1547 when page marked as error */
1552 end_page_writeback(page);
1553 page_cache_release(page);
1555 if ((wbc->nr_to_write -= n_iov) <= 0)
1559 /* Need to re-find the pages we skipped */
1560 index = pvec.pages[0]->index + 1;
1562 pagevec_release(&pvec);
1564 if (!scanned && !done) {
1566 * We hit the last page and there is more work to be done: wrap
1567 * back to the start of the file
1573 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1574 mapping->writeback_index = index;
1581 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1587 /* BB add check for wbc flags */
1588 page_cache_get(page);
1589 if (!PageUptodate(page))
1590 cFYI(1, ("ppw - page not up to date"));
1593 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1595 * A writepage() implementation always needs to do either this,
1596 * or re-dirty the page with "redirty_page_for_writepage()" in
1597 * the case of a failure.
1599 * Just unlocking the page will cause the radix tree tag-bits
1600 * to fail to update with the state of the page correctly.
1602 set_page_writeback(page);
1603 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1604 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1606 end_page_writeback(page);
1607 page_cache_release(page);
1612 static int cifs_write_end(struct file *file, struct address_space *mapping,
1613 loff_t pos, unsigned len, unsigned copied,
1614 struct page *page, void *fsdata)
1617 struct inode *inode = mapping->host;
1619 cFYI(1, ("write_end for page %p from pos %lld with %d bytes",
1620 page, pos, copied));
1622 if (PageChecked(page)) {
1624 SetPageUptodate(page);
1625 ClearPageChecked(page);
1626 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1627 SetPageUptodate(page);
1629 if (!PageUptodate(page)) {
1631 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1635 /* this is probably better than directly calling
1636 partialpage_write since in this function the file handle is
1637 known which we might as well leverage */
1638 /* BB check if anything else missing out of ppw
1639 such as updating last write time */
1640 page_data = kmap(page);
1641 rc = cifs_write(file, page_data + offset, copied, &pos);
1642 /* if (rc < 0) should we set writebehind rc? */
1649 set_page_dirty(page);
1653 spin_lock(&inode->i_lock);
1654 if (pos > inode->i_size)
1655 i_size_write(inode, pos);
1656 spin_unlock(&inode->i_lock);
1660 page_cache_release(page);
1665 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1669 struct cifsTconInfo *tcon;
1670 struct cifsFileInfo *smbfile =
1671 (struct cifsFileInfo *)file->private_data;
1672 struct inode *inode = file->f_path.dentry->d_inode;
1676 cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1677 dentry->d_name.name, datasync));
1679 rc = filemap_write_and_wait(inode->i_mapping);
1681 rc = CIFS_I(inode)->write_behind_rc;
1682 CIFS_I(inode)->write_behind_rc = 0;
1683 tcon = CIFS_SB(inode->i_sb)->tcon;
1684 if (!rc && tcon && smbfile &&
1685 !(CIFS_SB(inode->i_sb)->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1686 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1693 /* static void cifs_sync_page(struct page *page)
1695 struct address_space *mapping;
1696 struct inode *inode;
1697 unsigned long index = page->index;
1698 unsigned int rpages = 0;
1701 cFYI(1, ("sync page %p",page));
1702 mapping = page->mapping;
1705 inode = mapping->host;
1709 /* fill in rpages then
1710 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1712 /* cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1722 * As file closes, flush all cached write data for this inode checking
1723 * for write behind errors.
1725 int cifs_flush(struct file *file, fl_owner_t id)
1727 struct inode *inode = file->f_path.dentry->d_inode;
1730 /* Rather than do the steps manually:
1731 lock the inode for writing
1732 loop through pages looking for write behind data (dirty pages)
1733 coalesce into contiguous 16K (or smaller) chunks to write to server
1734 send to server (prefer in parallel)
1735 deal with writebehind errors
1736 unlock inode for writing
1737 filemapfdatawrite appears easier for the time being */
1739 rc = filemap_fdatawrite(inode->i_mapping);
1740 /* reset wb rc if we were able to write out dirty pages */
1742 rc = CIFS_I(inode)->write_behind_rc;
1743 CIFS_I(inode)->write_behind_rc = 0;
1746 cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
1751 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1752 size_t read_size, loff_t *poffset)
1755 unsigned int bytes_read = 0;
1756 unsigned int total_read = 0;
1757 unsigned int current_read_size;
1758 struct cifs_sb_info *cifs_sb;
1759 struct cifsTconInfo *pTcon;
1761 struct cifsFileInfo *open_file;
1762 char *smb_read_data;
1763 char __user *current_offset;
1764 struct smb_com_read_rsp *pSMBr;
1767 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1768 pTcon = cifs_sb->tcon;
1770 if (file->private_data == NULL) {
1774 open_file = (struct cifsFileInfo *)file->private_data;
1776 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1777 cFYI(1, ("attempting read on write only file instance"));
1779 for (total_read = 0, current_offset = read_data;
1780 read_size > total_read;
1781 total_read += bytes_read, current_offset += bytes_read) {
1782 current_read_size = min_t(const int, read_size - total_read,
1785 smb_read_data = NULL;
1786 while (rc == -EAGAIN) {
1787 int buf_type = CIFS_NO_BUFFER;
1788 if ((open_file->invalidHandle) &&
1789 (!open_file->closePend)) {
1790 rc = cifs_reopen_file(file, true);
1794 rc = CIFSSMBRead(xid, pTcon,
1796 current_read_size, *poffset,
1797 &bytes_read, &smb_read_data,
1799 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1800 if (smb_read_data) {
1801 if (copy_to_user(current_offset,
1803 4 /* RFC1001 length field */ +
1804 le16_to_cpu(pSMBr->DataOffset),
1808 if (buf_type == CIFS_SMALL_BUFFER)
1809 cifs_small_buf_release(smb_read_data);
1810 else if (buf_type == CIFS_LARGE_BUFFER)
1811 cifs_buf_release(smb_read_data);
1812 smb_read_data = NULL;
1815 if (rc || (bytes_read == 0)) {
1823 cifs_stats_bytes_read(pTcon, bytes_read);
1824 *poffset += bytes_read;
1832 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1836 unsigned int bytes_read = 0;
1837 unsigned int total_read;
1838 unsigned int current_read_size;
1839 struct cifs_sb_info *cifs_sb;
1840 struct cifsTconInfo *pTcon;
1842 char *current_offset;
1843 struct cifsFileInfo *open_file;
1844 int buf_type = CIFS_NO_BUFFER;
1847 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1848 pTcon = cifs_sb->tcon;
1850 if (file->private_data == NULL) {
1854 open_file = (struct cifsFileInfo *)file->private_data;
1856 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1857 cFYI(1, ("attempting read on write only file instance"));
1859 for (total_read = 0, current_offset = read_data;
1860 read_size > total_read;
1861 total_read += bytes_read, current_offset += bytes_read) {
1862 current_read_size = min_t(const int, read_size - total_read,
1864 /* For windows me and 9x we do not want to request more
1865 than it negotiated since it will refuse the read then */
1867 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1868 current_read_size = min_t(const int, current_read_size,
1869 pTcon->ses->server->maxBuf - 128);
1872 while (rc == -EAGAIN) {
1873 if ((open_file->invalidHandle) &&
1874 (!open_file->closePend)) {
1875 rc = cifs_reopen_file(file, true);
1879 rc = CIFSSMBRead(xid, pTcon,
1881 current_read_size, *poffset,
1882 &bytes_read, ¤t_offset,
1885 if (rc || (bytes_read == 0)) {
1893 cifs_stats_bytes_read(pTcon, total_read);
1894 *poffset += bytes_read;
1901 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1903 struct dentry *dentry = file->f_path.dentry;
1907 rc = cifs_revalidate(dentry);
1909 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1913 rc = generic_file_mmap(file, vma);
1919 static void cifs_copy_cache_pages(struct address_space *mapping,
1920 struct list_head *pages, int bytes_read, char *data,
1921 struct pagevec *plru_pvec)
1926 while (bytes_read > 0) {
1927 if (list_empty(pages))
1930 page = list_entry(pages->prev, struct page, lru);
1931 list_del(&page->lru);
1933 if (add_to_page_cache(page, mapping, page->index,
1935 page_cache_release(page);
1936 cFYI(1, ("Add page cache failed"));
1937 data += PAGE_CACHE_SIZE;
1938 bytes_read -= PAGE_CACHE_SIZE;
1942 target = kmap_atomic(page, KM_USER0);
1944 if (PAGE_CACHE_SIZE > bytes_read) {
1945 memcpy(target, data, bytes_read);
1946 /* zero the tail end of this partial page */
1947 memset(target + bytes_read, 0,
1948 PAGE_CACHE_SIZE - bytes_read);
1951 memcpy(target, data, PAGE_CACHE_SIZE);
1952 bytes_read -= PAGE_CACHE_SIZE;
1954 kunmap_atomic(target, KM_USER0);
1956 flush_dcache_page(page);
1957 SetPageUptodate(page);
1959 if (!pagevec_add(plru_pvec, page))
1960 __pagevec_lru_add_file(plru_pvec);
1961 data += PAGE_CACHE_SIZE;
1966 static int cifs_readpages(struct file *file, struct address_space *mapping,
1967 struct list_head *page_list, unsigned num_pages)
1973 struct cifs_sb_info *cifs_sb;
1974 struct cifsTconInfo *pTcon;
1975 unsigned int bytes_read = 0;
1976 unsigned int read_size, i;
1977 char *smb_read_data = NULL;
1978 struct smb_com_read_rsp *pSMBr;
1979 struct pagevec lru_pvec;
1980 struct cifsFileInfo *open_file;
1981 int buf_type = CIFS_NO_BUFFER;
1984 if (file->private_data == NULL) {
1988 open_file = (struct cifsFileInfo *)file->private_data;
1989 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1990 pTcon = cifs_sb->tcon;
1992 pagevec_init(&lru_pvec, 0);
1993 cFYI(DBG2, ("rpages: num pages %d", num_pages));
1994 for (i = 0; i < num_pages; ) {
1995 unsigned contig_pages;
1996 struct page *tmp_page;
1997 unsigned long expected_index;
1999 if (list_empty(page_list))
2002 page = list_entry(page_list->prev, struct page, lru);
2003 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2005 /* count adjacent pages that we will read into */
2008 list_entry(page_list->prev, struct page, lru)->index;
2009 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2010 if (tmp_page->index == expected_index) {
2016 if (contig_pages + i > num_pages)
2017 contig_pages = num_pages - i;
2019 /* for reads over a certain size could initiate async
2022 read_size = contig_pages * PAGE_CACHE_SIZE;
2023 /* Read size needs to be in multiples of one page */
2024 read_size = min_t(const unsigned int, read_size,
2025 cifs_sb->rsize & PAGE_CACHE_MASK);
2026 cFYI(DBG2, ("rpages: read size 0x%x contiguous pages %d",
2027 read_size, contig_pages));
2029 while (rc == -EAGAIN) {
2030 if ((open_file->invalidHandle) &&
2031 (!open_file->closePend)) {
2032 rc = cifs_reopen_file(file, true);
2037 rc = CIFSSMBRead(xid, pTcon,
2040 &bytes_read, &smb_read_data,
2042 /* BB more RC checks ? */
2043 if (rc == -EAGAIN) {
2044 if (smb_read_data) {
2045 if (buf_type == CIFS_SMALL_BUFFER)
2046 cifs_small_buf_release(smb_read_data);
2047 else if (buf_type == CIFS_LARGE_BUFFER)
2048 cifs_buf_release(smb_read_data);
2049 smb_read_data = NULL;
2053 if ((rc < 0) || (smb_read_data == NULL)) {
2054 cFYI(1, ("Read error in readpages: %d", rc));
2056 } else if (bytes_read > 0) {
2057 task_io_account_read(bytes_read);
2058 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2059 cifs_copy_cache_pages(mapping, page_list, bytes_read,
2060 smb_read_data + 4 /* RFC1001 hdr */ +
2061 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
2063 i += bytes_read >> PAGE_CACHE_SHIFT;
2064 cifs_stats_bytes_read(pTcon, bytes_read);
2065 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2066 i++; /* account for partial page */
2068 /* server copy of file can have smaller size
2070 /* BB do we need to verify this common case ?
2071 this case is ok - if we are at server EOF
2072 we will hit it on next read */
2077 cFYI(1, ("No bytes read (%d) at offset %lld . "
2078 "Cleaning remaining pages from readahead list",
2079 bytes_read, offset));
2080 /* BB turn off caching and do new lookup on
2081 file size at server? */
2084 if (smb_read_data) {
2085 if (buf_type == CIFS_SMALL_BUFFER)
2086 cifs_small_buf_release(smb_read_data);
2087 else if (buf_type == CIFS_LARGE_BUFFER)
2088 cifs_buf_release(smb_read_data);
2089 smb_read_data = NULL;
2094 pagevec_lru_add_file(&lru_pvec);
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;
2109 static int cifs_readpage_worker(struct file *file, struct page *page,
2115 page_cache_get(page);
2116 read_data = kmap(page);
2117 /* for reads over a certain size could initiate async read ahead */
2119 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2124 cFYI(1, ("Bytes read %d", rc));
2126 file->f_path.dentry->d_inode->i_atime =
2127 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2129 if (PAGE_CACHE_SIZE > rc)
2130 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2132 flush_dcache_page(page);
2133 SetPageUptodate(page);
2138 page_cache_release(page);
2142 static int cifs_readpage(struct file *file, struct page *page)
2144 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2150 if (file->private_data == NULL) {
2155 cFYI(1, ("readpage %p at offset %d 0x%x\n",
2156 page, (int)offset, (int)offset));
2158 rc = cifs_readpage_worker(file, page, &offset);
2166 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2168 struct cifsFileInfo *open_file;
2170 read_lock(&GlobalSMBSeslock);
2171 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2172 if (open_file->closePend)
2174 if (open_file->pfile &&
2175 ((open_file->pfile->f_flags & O_RDWR) ||
2176 (open_file->pfile->f_flags & O_WRONLY))) {
2177 read_unlock(&GlobalSMBSeslock);
2181 read_unlock(&GlobalSMBSeslock);
2185 /* We do not want to update the file size from server for inodes
2186 open for write - to avoid races with writepage extending
2187 the file - in the future we could consider allowing
2188 refreshing the inode only on increases in the file size
2189 but this is tricky to do without racing with writebehind
2190 page caching in the current Linux kernel design */
2191 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2196 if (is_inode_writable(cifsInode)) {
2197 /* This inode is open for write at least once */
2198 struct cifs_sb_info *cifs_sb;
2200 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2201 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2202 /* since no page cache to corrupt on directio
2203 we can change size safely */
2207 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2215 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2216 loff_t pos, unsigned len, unsigned flags,
2217 struct page **pagep, void **fsdata)
2219 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2220 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2221 loff_t page_start = pos & PAGE_MASK;
2226 cFYI(1, ("write_begin from %lld len %d", (long long)pos, len));
2228 page = grab_cache_page_write_begin(mapping, index, flags);
2234 if (PageUptodate(page))
2238 * If we write a full page it will be up to date, no need to read from
2239 * the server. If the write is short, we'll end up doing a sync write
2242 if (len == PAGE_CACHE_SIZE)
2246 * optimize away the read when we have an oplock, and we're not
2247 * expecting to use any of the data we'd be reading in. That
2248 * is, when the page lies beyond the EOF, or straddles the EOF
2249 * and the write will cover all of the existing data.
2251 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2252 i_size = i_size_read(mapping->host);
2253 if (page_start >= i_size ||
2254 (offset == 0 && (pos + len) >= i_size)) {
2255 zero_user_segments(page, 0, offset,
2259 * PageChecked means that the parts of the page
2260 * to which we're not writing are considered up
2261 * to date. Once the data is copied to the
2262 * page, it can be set uptodate.
2264 SetPageChecked(page);
2269 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2271 * might as well read a page, it is fast enough. If we get
2272 * an error, we don't need to return it. cifs_write_end will
2273 * do a sync write instead since PG_uptodate isn't set.
2275 cifs_readpage_worker(file, page, &page_start);
2277 /* we could try using another file handle if there is one -
2278 but how would we lock it to prevent close of that handle
2279 racing with this read? In any case
2280 this will be written out by write_end so is fine */
2287 const struct address_space_operations cifs_addr_ops = {
2288 .readpage = cifs_readpage,
2289 .readpages = cifs_readpages,
2290 .writepage = cifs_writepage,
2291 .writepages = cifs_writepages,
2292 .write_begin = cifs_write_begin,
2293 .write_end = cifs_write_end,
2294 .set_page_dirty = __set_page_dirty_nobuffers,
2295 /* .sync_page = cifs_sync_page, */
2300 * cifs_readpages requires the server to support a buffer large enough to
2301 * contain the header plus one complete page of data. Otherwise, we need
2302 * to leave cifs_readpages out of the address space operations.
2304 const struct address_space_operations cifs_addr_ops_smallbuf = {
2305 .readpage = cifs_readpage,
2306 .writepage = cifs_writepage,
2307 .writepages = cifs_writepages,
2308 .write_begin = cifs_write_begin,
2309 .write_end = cifs_write_end,
2310 .set_page_dirty = __set_page_dirty_nobuffers,
2311 /* .sync_page = cifs_sync_page, */