4 * vfs operations that deal with files
6 * Copyright (C) International Business Machines Corp., 2002,2010
7 * Author(s): Steve French (sfrench@us.ibm.com)
8 * Jeremy Allison (jra@samba.org)
10 * This library is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU Lesser General Public License as published
12 * by the Free Software Foundation; either version 2.1 of the License, or
13 * (at your option) any later version.
15 * This library is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
18 * the GNU Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public License
21 * along with this library; if not, write to the Free Software
22 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
25 #include <linux/backing-dev.h>
26 #include <linux/stat.h>
27 #include <linux/fcntl.h>
28 #include <linux/pagemap.h>
29 #include <linux/pagevec.h>
30 #include <linux/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/delay.h>
33 #include <linux/mount.h>
34 #include <linux/slab.h>
35 #include <asm/div64.h>
39 #include "cifsproto.h"
40 #include "cifs_unicode.h"
41 #include "cifs_debug.h"
42 #include "cifs_fs_sb.h"
45 static inline int cifs_convert_flags(unsigned int flags)
47 if ((flags & O_ACCMODE) == O_RDONLY)
49 else if ((flags & O_ACCMODE) == O_WRONLY)
51 else if ((flags & O_ACCMODE) == O_RDWR) {
52 /* GENERIC_ALL is too much permission to request
53 can cause unnecessary access denied on create */
54 /* return GENERIC_ALL; */
55 return (GENERIC_READ | GENERIC_WRITE);
58 return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
59 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
63 static u32 cifs_posix_convert_flags(unsigned int flags)
67 if ((flags & O_ACCMODE) == O_RDONLY)
68 posix_flags = SMB_O_RDONLY;
69 else if ((flags & O_ACCMODE) == O_WRONLY)
70 posix_flags = SMB_O_WRONLY;
71 else if ((flags & O_ACCMODE) == O_RDWR)
72 posix_flags = SMB_O_RDWR;
75 posix_flags |= SMB_O_CREAT;
77 posix_flags |= SMB_O_EXCL;
79 posix_flags |= SMB_O_TRUNC;
80 /* be safe and imply O_SYNC for O_DSYNC */
82 posix_flags |= SMB_O_SYNC;
83 if (flags & O_DIRECTORY)
84 posix_flags |= SMB_O_DIRECTORY;
85 if (flags & O_NOFOLLOW)
86 posix_flags |= SMB_O_NOFOLLOW;
88 posix_flags |= SMB_O_DIRECT;
93 static inline int cifs_get_disposition(unsigned int flags)
95 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
97 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
98 return FILE_OVERWRITE_IF;
99 else if ((flags & O_CREAT) == O_CREAT)
101 else if ((flags & O_TRUNC) == O_TRUNC)
102 return FILE_OVERWRITE;
107 int cifs_posix_open(char *full_path, struct inode **pinode,
108 struct super_block *sb, int mode, unsigned int f_flags,
109 __u32 *poplock, __u16 *pnetfid, int xid)
112 FILE_UNIX_BASIC_INFO *presp_data;
113 __u32 posix_flags = 0;
114 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
115 struct cifs_fattr fattr;
116 struct tcon_link *tlink;
117 struct cifsTconInfo *tcon;
119 cFYI(1, "posix open %s", full_path);
121 presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
122 if (presp_data == NULL)
125 tlink = cifs_sb_tlink(cifs_sb);
131 tcon = tlink_tcon(tlink);
132 mode &= ~current_umask();
134 posix_flags = cifs_posix_convert_flags(f_flags);
135 rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
136 poplock, full_path, cifs_sb->local_nls,
137 cifs_sb->mnt_cifs_flags &
138 CIFS_MOUNT_MAP_SPECIAL_CHR);
139 cifs_put_tlink(tlink);
144 if (presp_data->Type == cpu_to_le32(-1))
145 goto posix_open_ret; /* open ok, caller does qpathinfo */
148 goto posix_open_ret; /* caller does not need info */
150 cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
152 /* get new inode and set it up */
153 if (*pinode == NULL) {
154 cifs_fill_uniqueid(sb, &fattr);
155 *pinode = cifs_iget(sb, &fattr);
161 cifs_fattr_to_inode(*pinode, &fattr);
170 cifs_nt_open(char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
171 struct cifsTconInfo *tcon, unsigned int f_flags, __u32 *poplock,
172 __u16 *pnetfid, int xid)
179 desiredAccess = cifs_convert_flags(f_flags);
181 /*********************************************************************
182 * open flag mapping table:
184 * POSIX Flag CIFS Disposition
185 * ---------- ----------------
186 * O_CREAT FILE_OPEN_IF
187 * O_CREAT | O_EXCL FILE_CREATE
188 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
189 * O_TRUNC FILE_OVERWRITE
190 * none of the above FILE_OPEN
192 * Note that there is not a direct match between disposition
193 * FILE_SUPERSEDE (ie create whether or not file exists although
194 * O_CREAT | O_TRUNC is similar but truncates the existing
195 * file rather than creating a new file as FILE_SUPERSEDE does
196 * (which uses the attributes / metadata passed in on open call)
198 *? O_SYNC is a reasonable match to CIFS writethrough flag
199 *? and the read write flags match reasonably. O_LARGEFILE
200 *? is irrelevant because largefile support is always used
201 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
202 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
203 *********************************************************************/
205 disposition = cifs_get_disposition(f_flags);
207 /* BB pass O_SYNC flag through on file attributes .. BB */
209 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
213 if (tcon->ses->capabilities & CAP_NT_SMBS)
214 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
215 desiredAccess, CREATE_NOT_DIR, pnetfid, poplock, buf,
216 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
217 & CIFS_MOUNT_MAP_SPECIAL_CHR);
219 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
220 desiredAccess, CREATE_NOT_DIR, pnetfid, poplock, buf,
221 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
222 & CIFS_MOUNT_MAP_SPECIAL_CHR);
228 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
231 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
239 struct cifsFileInfo *
240 cifs_new_fileinfo(__u16 fileHandle, struct file *file,
241 struct tcon_link *tlink, __u32 oplock)
243 struct dentry *dentry = file->f_path.dentry;
244 struct inode *inode = dentry->d_inode;
245 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
246 struct cifsFileInfo *pCifsFile;
248 pCifsFile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
249 if (pCifsFile == NULL)
252 pCifsFile->count = 1;
253 pCifsFile->netfid = fileHandle;
254 pCifsFile->pid = current->tgid;
255 pCifsFile->uid = current_fsuid();
256 pCifsFile->dentry = dget(dentry);
257 pCifsFile->f_flags = file->f_flags;
258 pCifsFile->invalidHandle = false;
259 pCifsFile->tlink = cifs_get_tlink(tlink);
260 mutex_init(&pCifsFile->fh_mutex);
261 mutex_init(&pCifsFile->lock_mutex);
262 INIT_LIST_HEAD(&pCifsFile->llist);
263 INIT_WORK(&pCifsFile->oplock_break, cifs_oplock_break);
265 spin_lock(&cifs_file_list_lock);
266 list_add(&pCifsFile->tlist, &(tlink_tcon(tlink)->openFileList));
267 /* if readable file instance put first in list*/
268 if (file->f_mode & FMODE_READ)
269 list_add(&pCifsFile->flist, &pCifsInode->openFileList);
271 list_add_tail(&pCifsFile->flist, &pCifsInode->openFileList);
272 spin_unlock(&cifs_file_list_lock);
274 cifs_set_oplock_level(pCifsInode, oplock);
276 file->private_data = pCifsFile;
281 * Release a reference on the file private data. This may involve closing
282 * the filehandle out on the server. Must be called without holding
283 * cifs_file_list_lock.
285 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
287 struct inode *inode = cifs_file->dentry->d_inode;
288 struct cifsTconInfo *tcon = tlink_tcon(cifs_file->tlink);
289 struct cifsInodeInfo *cifsi = CIFS_I(inode);
290 struct cifsLockInfo *li, *tmp;
292 spin_lock(&cifs_file_list_lock);
293 if (--cifs_file->count > 0) {
294 spin_unlock(&cifs_file_list_lock);
298 /* remove it from the lists */
299 list_del(&cifs_file->flist);
300 list_del(&cifs_file->tlist);
302 if (list_empty(&cifsi->openFileList)) {
303 cFYI(1, "closing last open instance for inode %p",
304 cifs_file->dentry->d_inode);
305 cifs_set_oplock_level(cifsi, 0);
307 spin_unlock(&cifs_file_list_lock);
309 if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
313 rc = CIFSSMBClose(xid, tcon, cifs_file->netfid);
317 /* Delete any outstanding lock records. We'll lose them when the file
320 mutex_lock(&cifs_file->lock_mutex);
321 list_for_each_entry_safe(li, tmp, &cifs_file->llist, llist) {
322 list_del(&li->llist);
325 mutex_unlock(&cifs_file->lock_mutex);
327 cifs_put_tlink(cifs_file->tlink);
328 dput(cifs_file->dentry);
332 int cifs_open(struct inode *inode, struct file *file)
337 struct cifs_sb_info *cifs_sb;
338 struct cifsTconInfo *tcon;
339 struct tcon_link *tlink;
340 struct cifsFileInfo *pCifsFile = NULL;
341 struct cifsInodeInfo *pCifsInode;
342 char *full_path = NULL;
343 bool posix_open_ok = false;
348 cifs_sb = CIFS_SB(inode->i_sb);
349 tlink = cifs_sb_tlink(cifs_sb);
352 return PTR_ERR(tlink);
354 tcon = tlink_tcon(tlink);
356 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
358 full_path = build_path_from_dentry(file->f_path.dentry);
359 if (full_path == NULL) {
364 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
365 inode, file->f_flags, full_path);
372 if (!tcon->broken_posix_open && tcon->unix_ext &&
373 (tcon->ses->capabilities & CAP_UNIX) &&
374 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
375 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
376 /* can not refresh inode info since size could be stale */
377 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
378 cifs_sb->mnt_file_mode /* ignored */,
379 file->f_flags, &oplock, &netfid, xid);
381 cFYI(1, "posix open succeeded");
382 posix_open_ok = true;
383 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
384 if (tcon->ses->serverNOS)
385 cERROR(1, "server %s of type %s returned"
386 " unexpected error on SMB posix open"
387 ", disabling posix open support."
388 " Check if server update available.",
389 tcon->ses->serverName,
390 tcon->ses->serverNOS);
391 tcon->broken_posix_open = true;
392 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
393 (rc != -EOPNOTSUPP)) /* path not found or net err */
395 /* else fallthrough to retry open the old way on network i/o
399 if (!posix_open_ok) {
400 rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
401 file->f_flags, &oplock, &netfid, xid);
406 pCifsFile = cifs_new_fileinfo(netfid, file, tlink, oplock);
407 if (pCifsFile == NULL) {
408 CIFSSMBClose(xid, tcon, netfid);
413 cifs_fscache_set_inode_cookie(inode, file);
415 if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
416 /* time to set mode which we can not set earlier due to
417 problems creating new read-only files */
418 struct cifs_unix_set_info_args args = {
419 .mode = inode->i_mode,
422 .ctime = NO_CHANGE_64,
423 .atime = NO_CHANGE_64,
424 .mtime = NO_CHANGE_64,
427 CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
429 cifs_sb->mnt_cifs_flags &
430 CIFS_MOUNT_MAP_SPECIAL_CHR);
436 cifs_put_tlink(tlink);
440 /* Try to reacquire byte range locks that were released when session */
441 /* to server was lost */
442 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
446 /* BB list all locks open on this file and relock */
451 static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
456 struct cifs_sb_info *cifs_sb;
457 struct cifsTconInfo *tcon;
458 struct cifsInodeInfo *pCifsInode;
460 char *full_path = NULL;
462 int disposition = FILE_OPEN;
466 mutex_lock(&pCifsFile->fh_mutex);
467 if (!pCifsFile->invalidHandle) {
468 mutex_unlock(&pCifsFile->fh_mutex);
474 inode = pCifsFile->dentry->d_inode;
475 cifs_sb = CIFS_SB(inode->i_sb);
476 tcon = tlink_tcon(pCifsFile->tlink);
478 /* can not grab rename sem here because various ops, including
479 those that already have the rename sem can end up causing writepage
480 to get called and if the server was down that means we end up here,
481 and we can never tell if the caller already has the rename_sem */
482 full_path = build_path_from_dentry(pCifsFile->dentry);
483 if (full_path == NULL) {
485 mutex_unlock(&pCifsFile->fh_mutex);
490 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
491 inode, pCifsFile->f_flags, full_path);
498 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
499 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
500 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
503 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
504 * original open. Must mask them off for a reopen.
506 unsigned int oflags = pCifsFile->f_flags &
507 ~(O_CREAT | O_EXCL | O_TRUNC);
509 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
510 cifs_sb->mnt_file_mode /* ignored */,
511 oflags, &oplock, &netfid, xid);
513 cFYI(1, "posix reopen succeeded");
516 /* fallthrough to retry open the old way on errors, especially
517 in the reconnect path it is important to retry hard */
520 desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
522 /* Can not refresh inode by passing in file_info buf to be returned
523 by SMBOpen and then calling get_inode_info with returned buf
524 since file might have write behind data that needs to be flushed
525 and server version of file size can be stale. If we knew for sure
526 that inode was not dirty locally we could do this */
528 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
529 CREATE_NOT_DIR, &netfid, &oplock, NULL,
530 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
531 CIFS_MOUNT_MAP_SPECIAL_CHR);
533 mutex_unlock(&pCifsFile->fh_mutex);
534 cFYI(1, "cifs_open returned 0x%x", rc);
535 cFYI(1, "oplock: %d", oplock);
536 goto reopen_error_exit;
540 pCifsFile->netfid = netfid;
541 pCifsFile->invalidHandle = false;
542 mutex_unlock(&pCifsFile->fh_mutex);
543 pCifsInode = CIFS_I(inode);
546 rc = filemap_write_and_wait(inode->i_mapping);
547 mapping_set_error(inode->i_mapping, rc);
550 rc = cifs_get_inode_info_unix(&inode,
551 full_path, inode->i_sb, xid);
553 rc = cifs_get_inode_info(&inode,
554 full_path, NULL, inode->i_sb,
556 } /* else we are writing out data to server already
557 and could deadlock if we tried to flush data, and
558 since we do not know if we have data that would
559 invalidate the current end of file on the server
560 we can not go to the server to get the new inod
563 cifs_set_oplock_level(pCifsInode, oplock);
565 cifs_relock_file(pCifsFile);
573 int cifs_close(struct inode *inode, struct file *file)
575 cifsFileInfo_put(file->private_data);
576 file->private_data = NULL;
578 /* return code from the ->release op is always ignored */
582 int cifs_closedir(struct inode *inode, struct file *file)
586 struct cifsFileInfo *pCFileStruct = file->private_data;
589 cFYI(1, "Closedir inode = 0x%p", inode);
594 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
596 cFYI(1, "Freeing private data in close dir");
597 spin_lock(&cifs_file_list_lock);
598 if (!pCFileStruct->srch_inf.endOfSearch &&
599 !pCFileStruct->invalidHandle) {
600 pCFileStruct->invalidHandle = true;
601 spin_unlock(&cifs_file_list_lock);
602 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
603 cFYI(1, "Closing uncompleted readdir with rc %d",
605 /* not much we can do if it fails anyway, ignore rc */
608 spin_unlock(&cifs_file_list_lock);
609 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
611 cFYI(1, "closedir free smb buf in srch struct");
612 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
613 if (pCFileStruct->srch_inf.smallBuf)
614 cifs_small_buf_release(ptmp);
616 cifs_buf_release(ptmp);
618 cifs_put_tlink(pCFileStruct->tlink);
619 kfree(file->private_data);
620 file->private_data = NULL;
622 /* BB can we lock the filestruct while this is going on? */
627 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
628 __u64 offset, __u8 lockType)
630 struct cifsLockInfo *li =
631 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
637 mutex_lock(&fid->lock_mutex);
638 list_add(&li->llist, &fid->llist);
639 mutex_unlock(&fid->lock_mutex);
643 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
649 bool wait_flag = false;
650 struct cifs_sb_info *cifs_sb;
651 struct cifsTconInfo *tcon;
653 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
654 bool posix_locking = 0;
656 length = 1 + pfLock->fl_end - pfLock->fl_start;
660 cFYI(1, "Lock parm: 0x%x flockflags: "
661 "0x%x flocktype: 0x%x start: %lld end: %lld",
662 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
665 if (pfLock->fl_flags & FL_POSIX)
667 if (pfLock->fl_flags & FL_FLOCK)
669 if (pfLock->fl_flags & FL_SLEEP) {
670 cFYI(1, "Blocking lock");
673 if (pfLock->fl_flags & FL_ACCESS)
674 cFYI(1, "Process suspended by mandatory locking - "
675 "not implemented yet");
676 if (pfLock->fl_flags & FL_LEASE)
677 cFYI(1, "Lease on file - not implemented yet");
678 if (pfLock->fl_flags &
679 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
680 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
682 if (pfLock->fl_type == F_WRLCK) {
685 } else if (pfLock->fl_type == F_UNLCK) {
688 /* Check if unlock includes more than
690 } else if (pfLock->fl_type == F_RDLCK) {
692 lockType |= LOCKING_ANDX_SHARED_LOCK;
694 } else if (pfLock->fl_type == F_EXLCK) {
697 } else if (pfLock->fl_type == F_SHLCK) {
699 lockType |= LOCKING_ANDX_SHARED_LOCK;
702 cFYI(1, "Unknown type of lock");
704 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
705 tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
706 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
708 if ((tcon->ses->capabilities & CAP_UNIX) &&
709 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
710 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
712 /* BB add code here to normalize offset and length to
713 account for negative length which we can not accept over the
718 if (lockType & LOCKING_ANDX_SHARED_LOCK)
719 posix_lock_type = CIFS_RDLCK;
721 posix_lock_type = CIFS_WRLCK;
722 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
724 posix_lock_type, wait_flag);
729 /* BB we could chain these into one lock request BB */
730 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
731 0, 1, lockType, 0 /* wait flag */ );
733 rc = CIFSSMBLock(xid, tcon, netfid, length,
734 pfLock->fl_start, 1 /* numUnlock */ ,
735 0 /* numLock */ , lockType,
737 pfLock->fl_type = F_UNLCK;
739 cERROR(1, "Error unlocking previously locked "
740 "range %d during test of lock", rc);
744 /* if rc == ERR_SHARING_VIOLATION ? */
747 if (lockType & LOCKING_ANDX_SHARED_LOCK) {
748 pfLock->fl_type = F_WRLCK;
750 rc = CIFSSMBLock(xid, tcon, netfid, length,
751 pfLock->fl_start, 0, 1,
752 lockType | LOCKING_ANDX_SHARED_LOCK,
755 rc = CIFSSMBLock(xid, tcon, netfid,
756 length, pfLock->fl_start, 1, 0,
758 LOCKING_ANDX_SHARED_LOCK,
760 pfLock->fl_type = F_RDLCK;
762 cERROR(1, "Error unlocking "
763 "previously locked range %d "
764 "during test of lock", rc);
767 pfLock->fl_type = F_WRLCK;
777 if (!numLock && !numUnlock) {
778 /* if no lock or unlock then nothing
779 to do since we do not know what it is */
786 if (lockType & LOCKING_ANDX_SHARED_LOCK)
787 posix_lock_type = CIFS_RDLCK;
789 posix_lock_type = CIFS_WRLCK;
792 posix_lock_type = CIFS_UNLCK;
794 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
796 posix_lock_type, wait_flag);
798 struct cifsFileInfo *fid = file->private_data;
801 rc = CIFSSMBLock(xid, tcon, netfid, length,
803 0, numLock, lockType, wait_flag);
806 /* For Windows locks we must store them. */
807 rc = store_file_lock(fid, length,
808 pfLock->fl_start, lockType);
810 } else if (numUnlock) {
811 /* For each stored lock that this unlock overlaps
812 completely, unlock it. */
814 struct cifsLockInfo *li, *tmp;
817 mutex_lock(&fid->lock_mutex);
818 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
819 if (pfLock->fl_start <= li->offset &&
820 (pfLock->fl_start + length) >=
821 (li->offset + li->length)) {
822 stored_rc = CIFSSMBLock(xid, tcon,
824 li->length, li->offset,
825 1, 0, li->type, false);
829 list_del(&li->llist);
834 mutex_unlock(&fid->lock_mutex);
838 if (pfLock->fl_flags & FL_POSIX)
839 posix_lock_file_wait(file, pfLock);
845 * Set the timeout on write requests past EOF. For some servers (Windows)
846 * these calls can be very long.
848 * If we're writing >10M past the EOF we give a 180s timeout. Anything less
849 * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
850 * The 10M cutoff is totally arbitrary. A better scheme for this would be
851 * welcome if someone wants to suggest one.
853 * We may be able to do a better job with this if there were some way to
854 * declare that a file should be sparse.
857 cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
859 if (offset <= cifsi->server_eof)
861 else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
862 return CIFS_VLONG_OP;
867 /* update the file size (if needed) after a write */
869 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
870 unsigned int bytes_written)
872 loff_t end_of_write = offset + bytes_written;
874 if (end_of_write > cifsi->server_eof)
875 cifsi->server_eof = end_of_write;
878 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
879 size_t write_size, loff_t *poffset)
881 struct inode *inode = file->f_path.dentry->d_inode;
883 unsigned int bytes_written = 0;
884 unsigned int total_written;
885 struct cifs_sb_info *cifs_sb;
886 struct cifsTconInfo *pTcon;
888 struct cifsFileInfo *open_file;
889 struct cifsInodeInfo *cifsi = CIFS_I(inode);
891 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
893 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
894 *poffset, file->f_path.dentry->d_name.name); */
896 if (file->private_data == NULL)
899 open_file = file->private_data;
900 pTcon = tlink_tcon(open_file->tlink);
902 rc = generic_write_checks(file, poffset, &write_size, 0);
908 long_op = cifs_write_timeout(cifsi, *poffset);
909 for (total_written = 0; write_size > total_written;
910 total_written += bytes_written) {
912 while (rc == -EAGAIN) {
913 if (file->private_data == NULL) {
914 /* file has been closed on us */
916 /* if we have gotten here we have written some data
917 and blocked, and the file has been freed on us while
918 we blocked so return what we managed to write */
919 return total_written;
921 if (open_file->invalidHandle) {
922 /* we could deadlock if we called
923 filemap_fdatawait from here so tell
924 reopen_file not to flush data to server
926 rc = cifs_reopen_file(open_file, false);
931 rc = CIFSSMBWrite(xid, pTcon,
933 min_t(const int, cifs_sb->wsize,
934 write_size - total_written),
935 *poffset, &bytes_written,
936 NULL, write_data + total_written, long_op);
938 if (rc || (bytes_written == 0)) {
946 cifs_update_eof(cifsi, *poffset, bytes_written);
947 *poffset += bytes_written;
949 long_op = CIFS_STD_OP; /* subsequent writes fast -
950 15 seconds is plenty */
953 cifs_stats_bytes_written(pTcon, total_written);
955 /* Do not update local mtime - server will set its actual value on write
956 * inode->i_ctime = inode->i_mtime =
957 * current_fs_time(inode->i_sb);*/
958 if (total_written > 0) {
959 spin_lock(&inode->i_lock);
960 if (*poffset > inode->i_size)
961 i_size_write(inode, *poffset);
962 spin_unlock(&inode->i_lock);
964 mark_inode_dirty_sync(inode);
967 return total_written;
970 static ssize_t cifs_write(struct cifsFileInfo *open_file,
971 const char *write_data, size_t write_size,
975 unsigned int bytes_written = 0;
976 unsigned int total_written;
977 struct cifs_sb_info *cifs_sb;
978 struct cifsTconInfo *pTcon;
980 struct dentry *dentry = open_file->dentry;
981 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
983 cifs_sb = CIFS_SB(dentry->d_sb);
985 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
986 *poffset, dentry->d_name.name);
988 pTcon = tlink_tcon(open_file->tlink);
992 long_op = cifs_write_timeout(cifsi, *poffset);
993 for (total_written = 0; write_size > total_written;
994 total_written += bytes_written) {
996 while (rc == -EAGAIN) {
997 if (open_file->invalidHandle) {
998 /* we could deadlock if we called
999 filemap_fdatawait from here so tell
1000 reopen_file not to flush data to
1002 rc = cifs_reopen_file(open_file, false);
1006 if (experimEnabled || (pTcon->ses->server &&
1007 ((pTcon->ses->server->secMode &
1008 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1013 len = min((size_t)cifs_sb->wsize,
1014 write_size - total_written);
1015 /* iov[0] is reserved for smb header */
1016 iov[1].iov_base = (char *)write_data +
1018 iov[1].iov_len = len;
1019 rc = CIFSSMBWrite2(xid, pTcon,
1020 open_file->netfid, len,
1021 *poffset, &bytes_written,
1024 rc = CIFSSMBWrite(xid, pTcon,
1026 min_t(const int, cifs_sb->wsize,
1027 write_size - total_written),
1028 *poffset, &bytes_written,
1029 write_data + total_written,
1032 if (rc || (bytes_written == 0)) {
1040 cifs_update_eof(cifsi, *poffset, bytes_written);
1041 *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 if (total_written > 0) {
1050 spin_lock(&dentry->d_inode->i_lock);
1051 if (*poffset > dentry->d_inode->i_size)
1052 i_size_write(dentry->d_inode, *poffset);
1053 spin_unlock(&dentry->d_inode->i_lock);
1055 mark_inode_dirty_sync(dentry->d_inode);
1057 return total_written;
1060 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1063 struct cifsFileInfo *open_file = NULL;
1064 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1066 /* only filter by fsuid on multiuser mounts */
1067 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1070 spin_lock(&cifs_file_list_lock);
1071 /* we could simply get the first_list_entry since write-only entries
1072 are always at the end of the list but since the first entry might
1073 have a close pending, we go through the whole list */
1074 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1075 if (fsuid_only && open_file->uid != current_fsuid())
1077 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1078 if (!open_file->invalidHandle) {
1079 /* found a good file */
1080 /* lock it so it will not be closed on us */
1081 cifsFileInfo_get(open_file);
1082 spin_unlock(&cifs_file_list_lock);
1084 } /* else might as well continue, and look for
1085 another, or simply have the caller reopen it
1086 again rather than trying to fix this handle */
1087 } else /* write only file */
1088 break; /* write only files are last so must be done */
1090 spin_unlock(&cifs_file_list_lock);
1094 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1097 struct cifsFileInfo *open_file;
1098 struct cifs_sb_info *cifs_sb;
1099 bool any_available = false;
1102 /* Having a null inode here (because mapping->host was set to zero by
1103 the VFS or MM) should not happen but we had reports of on oops (due to
1104 it being zero) during stress testcases so we need to check for it */
1106 if (cifs_inode == NULL) {
1107 cERROR(1, "Null inode passed to cifs_writeable_file");
1112 cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1114 /* only filter by fsuid on multiuser mounts */
1115 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1118 spin_lock(&cifs_file_list_lock);
1120 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1121 if (!any_available && open_file->pid != current->tgid)
1123 if (fsuid_only && open_file->uid != current_fsuid())
1125 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1126 cifsFileInfo_get(open_file);
1128 if (!open_file->invalidHandle) {
1129 /* found a good writable file */
1130 spin_unlock(&cifs_file_list_lock);
1134 spin_unlock(&cifs_file_list_lock);
1136 /* Had to unlock since following call can block */
1137 rc = cifs_reopen_file(open_file, false);
1141 /* if it fails, try another handle if possible */
1142 cFYI(1, "wp failed on reopen file");
1143 cifsFileInfo_put(open_file);
1145 spin_lock(&cifs_file_list_lock);
1147 /* else we simply continue to the next entry. Thus
1148 we do not loop on reopen errors. If we
1149 can not reopen the file, for example if we
1150 reconnected to a server with another client
1151 racing to delete or lock the file we would not
1152 make progress if we restarted before the beginning
1153 of the loop here. */
1156 /* couldn't find useable FH with same pid, try any available */
1157 if (!any_available) {
1158 any_available = true;
1159 goto refind_writable;
1161 spin_unlock(&cifs_file_list_lock);
1165 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1167 struct address_space *mapping = page->mapping;
1168 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1171 int bytes_written = 0;
1172 struct cifs_sb_info *cifs_sb;
1173 struct inode *inode;
1174 struct cifsFileInfo *open_file;
1176 if (!mapping || !mapping->host)
1179 inode = page->mapping->host;
1180 cifs_sb = CIFS_SB(inode->i_sb);
1182 offset += (loff_t)from;
1183 write_data = kmap(page);
1186 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1191 /* racing with truncate? */
1192 if (offset > mapping->host->i_size) {
1194 return 0; /* don't care */
1197 /* check to make sure that we are not extending the file */
1198 if (mapping->host->i_size - offset < (loff_t)to)
1199 to = (unsigned)(mapping->host->i_size - offset);
1201 open_file = find_writable_file(CIFS_I(mapping->host), false);
1203 bytes_written = cifs_write(open_file, write_data,
1204 to - from, &offset);
1205 cifsFileInfo_put(open_file);
1206 /* Does mm or vfs already set times? */
1207 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1208 if ((bytes_written > 0) && (offset))
1210 else if (bytes_written < 0)
1213 cFYI(1, "No writeable filehandles for inode");
1221 static int cifs_writepages(struct address_space *mapping,
1222 struct writeback_control *wbc)
1224 unsigned int bytes_to_write;
1225 unsigned int bytes_written;
1226 struct cifs_sb_info *cifs_sb;
1230 int range_whole = 0;
1237 struct cifsFileInfo *open_file;
1238 struct cifsTconInfo *tcon;
1239 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1241 struct pagevec pvec;
1246 cifs_sb = CIFS_SB(mapping->host->i_sb);
1249 * If wsize is smaller that the page cache size, default to writing
1250 * one page at a time via cifs_writepage
1252 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1253 return generic_writepages(mapping, wbc);
1255 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1257 return generic_writepages(mapping, wbc);
1260 * if there's no open file, then this is likely to fail too,
1261 * but it'll at least handle the return. Maybe it should be
1264 open_file = find_writable_file(CIFS_I(mapping->host), false);
1267 return generic_writepages(mapping, wbc);
1270 tcon = tlink_tcon(open_file->tlink);
1271 if (!experimEnabled && tcon->ses->server->secMode &
1272 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1273 cifsFileInfo_put(open_file);
1275 return generic_writepages(mapping, wbc);
1277 cifsFileInfo_put(open_file);
1281 pagevec_init(&pvec, 0);
1282 if (wbc->range_cyclic) {
1283 index = mapping->writeback_index; /* Start from prev offset */
1286 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1287 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1288 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1293 while (!done && (index <= end) &&
1294 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1295 PAGECACHE_TAG_DIRTY,
1296 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1305 for (i = 0; i < nr_pages; i++) {
1306 page = pvec.pages[i];
1308 * At this point we hold neither mapping->tree_lock nor
1309 * lock on the page itself: the page may be truncated or
1310 * invalidated (changing page->mapping to NULL), or even
1311 * swizzled back from swapper_space to tmpfs file
1317 else if (!trylock_page(page))
1320 if (unlikely(page->mapping != mapping)) {
1325 if (!wbc->range_cyclic && page->index > end) {
1331 if (next && (page->index != next)) {
1332 /* Not next consecutive page */
1337 if (wbc->sync_mode != WB_SYNC_NONE)
1338 wait_on_page_writeback(page);
1340 if (PageWriteback(page) ||
1341 !clear_page_dirty_for_io(page)) {
1347 * This actually clears the dirty bit in the radix tree.
1348 * See cifs_writepage() for more commentary.
1350 set_page_writeback(page);
1352 if (page_offset(page) >= mapping->host->i_size) {
1355 end_page_writeback(page);
1360 * BB can we get rid of this? pages are held by pvec
1362 page_cache_get(page);
1364 len = min(mapping->host->i_size - page_offset(page),
1365 (loff_t)PAGE_CACHE_SIZE);
1367 /* reserve iov[0] for the smb header */
1369 iov[n_iov].iov_base = kmap(page);
1370 iov[n_iov].iov_len = len;
1371 bytes_to_write += len;
1375 offset = page_offset(page);
1377 next = page->index + 1;
1378 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1382 open_file = find_writable_file(CIFS_I(mapping->host),
1385 cERROR(1, "No writable handles for inode");
1388 long_op = cifs_write_timeout(cifsi, offset);
1389 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1390 bytes_to_write, offset,
1391 &bytes_written, iov, n_iov,
1393 cifsFileInfo_put(open_file);
1394 cifs_update_eof(cifsi, offset, bytes_written);
1397 if (rc || bytes_written < bytes_to_write) {
1398 cERROR(1, "Write2 ret %d, wrote %d",
1400 mapping_set_error(mapping, rc);
1402 cifs_stats_bytes_written(tcon, bytes_written);
1405 for (i = 0; i < n_iov; i++) {
1406 page = pvec.pages[first + i];
1407 /* Should we also set page error on
1408 success rc but too little data written? */
1409 /* BB investigate retry logic on temporary
1410 server crash cases and how recovery works
1411 when page marked as error */
1416 end_page_writeback(page);
1417 page_cache_release(page);
1419 if ((wbc->nr_to_write -= n_iov) <= 0)
1423 /* Need to re-find the pages we skipped */
1424 index = pvec.pages[0]->index + 1;
1426 pagevec_release(&pvec);
1428 if (!scanned && !done) {
1430 * We hit the last page and there is more work to be done: wrap
1431 * back to the start of the file
1437 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1438 mapping->writeback_index = index;
1445 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1451 /* BB add check for wbc flags */
1452 page_cache_get(page);
1453 if (!PageUptodate(page))
1454 cFYI(1, "ppw - page not up to date");
1457 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1459 * A writepage() implementation always needs to do either this,
1460 * or re-dirty the page with "redirty_page_for_writepage()" in
1461 * the case of a failure.
1463 * Just unlocking the page will cause the radix tree tag-bits
1464 * to fail to update with the state of the page correctly.
1466 set_page_writeback(page);
1467 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1468 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1470 end_page_writeback(page);
1471 page_cache_release(page);
1476 static int cifs_write_end(struct file *file, struct address_space *mapping,
1477 loff_t pos, unsigned len, unsigned copied,
1478 struct page *page, void *fsdata)
1481 struct inode *inode = mapping->host;
1483 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1486 if (PageChecked(page)) {
1488 SetPageUptodate(page);
1489 ClearPageChecked(page);
1490 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1491 SetPageUptodate(page);
1493 if (!PageUptodate(page)) {
1495 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1499 /* this is probably better than directly calling
1500 partialpage_write since in this function the file handle is
1501 known which we might as well leverage */
1502 /* BB check if anything else missing out of ppw
1503 such as updating last write time */
1504 page_data = kmap(page);
1505 rc = cifs_write(file->private_data, page_data + offset,
1507 /* if (rc < 0) should we set writebehind rc? */
1514 set_page_dirty(page);
1518 spin_lock(&inode->i_lock);
1519 if (pos > inode->i_size)
1520 i_size_write(inode, pos);
1521 spin_unlock(&inode->i_lock);
1525 page_cache_release(page);
1530 int cifs_fsync(struct file *file, int datasync)
1534 struct cifsTconInfo *tcon;
1535 struct cifsFileInfo *smbfile = file->private_data;
1536 struct inode *inode = file->f_path.dentry->d_inode;
1540 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1541 file->f_path.dentry->d_name.name, datasync);
1543 rc = filemap_write_and_wait(inode->i_mapping);
1545 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
1547 tcon = tlink_tcon(smbfile->tlink);
1548 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1549 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1556 /* static void cifs_sync_page(struct page *page)
1558 struct address_space *mapping;
1559 struct inode *inode;
1560 unsigned long index = page->index;
1561 unsigned int rpages = 0;
1564 cFYI(1, "sync page %p", page);
1565 mapping = page->mapping;
1568 inode = mapping->host;
1572 /* fill in rpages then
1573 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1575 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1585 * As file closes, flush all cached write data for this inode checking
1586 * for write behind errors.
1588 int cifs_flush(struct file *file, fl_owner_t id)
1590 struct inode *inode = file->f_path.dentry->d_inode;
1593 if (file->f_mode & FMODE_WRITE)
1594 rc = filemap_write_and_wait(inode->i_mapping);
1596 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1601 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1602 size_t read_size, loff_t *poffset)
1605 unsigned int bytes_read = 0;
1606 unsigned int total_read = 0;
1607 unsigned int current_read_size;
1608 struct cifs_sb_info *cifs_sb;
1609 struct cifsTconInfo *pTcon;
1611 struct cifsFileInfo *open_file;
1612 char *smb_read_data;
1613 char __user *current_offset;
1614 struct smb_com_read_rsp *pSMBr;
1617 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1619 if (file->private_data == NULL) {
1624 open_file = file->private_data;
1625 pTcon = tlink_tcon(open_file->tlink);
1627 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1628 cFYI(1, "attempting read on write only file instance");
1630 for (total_read = 0, current_offset = read_data;
1631 read_size > total_read;
1632 total_read += bytes_read, current_offset += bytes_read) {
1633 current_read_size = min_t(const int, read_size - total_read,
1636 smb_read_data = NULL;
1637 while (rc == -EAGAIN) {
1638 int buf_type = CIFS_NO_BUFFER;
1639 if (open_file->invalidHandle) {
1640 rc = cifs_reopen_file(open_file, true);
1644 rc = CIFSSMBRead(xid, pTcon,
1646 current_read_size, *poffset,
1647 &bytes_read, &smb_read_data,
1649 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1650 if (smb_read_data) {
1651 if (copy_to_user(current_offset,
1653 4 /* RFC1001 length field */ +
1654 le16_to_cpu(pSMBr->DataOffset),
1658 if (buf_type == CIFS_SMALL_BUFFER)
1659 cifs_small_buf_release(smb_read_data);
1660 else if (buf_type == CIFS_LARGE_BUFFER)
1661 cifs_buf_release(smb_read_data);
1662 smb_read_data = NULL;
1665 if (rc || (bytes_read == 0)) {
1673 cifs_stats_bytes_read(pTcon, bytes_read);
1674 *poffset += bytes_read;
1682 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1686 unsigned int bytes_read = 0;
1687 unsigned int total_read;
1688 unsigned int current_read_size;
1689 struct cifs_sb_info *cifs_sb;
1690 struct cifsTconInfo *pTcon;
1692 char *current_offset;
1693 struct cifsFileInfo *open_file;
1694 int buf_type = CIFS_NO_BUFFER;
1697 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1699 if (file->private_data == NULL) {
1704 open_file = file->private_data;
1705 pTcon = tlink_tcon(open_file->tlink);
1707 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1708 cFYI(1, "attempting read on write only file instance");
1710 for (total_read = 0, current_offset = read_data;
1711 read_size > total_read;
1712 total_read += bytes_read, current_offset += bytes_read) {
1713 current_read_size = min_t(const int, read_size - total_read,
1715 /* For windows me and 9x we do not want to request more
1716 than it negotiated since it will refuse the read then */
1718 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1719 current_read_size = min_t(const int, current_read_size,
1720 pTcon->ses->server->maxBuf - 128);
1723 while (rc == -EAGAIN) {
1724 if (open_file->invalidHandle) {
1725 rc = cifs_reopen_file(open_file, true);
1729 rc = CIFSSMBRead(xid, pTcon,
1731 current_read_size, *poffset,
1732 &bytes_read, ¤t_offset,
1735 if (rc || (bytes_read == 0)) {
1743 cifs_stats_bytes_read(pTcon, total_read);
1744 *poffset += bytes_read;
1751 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1756 rc = cifs_revalidate_file(file);
1758 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1762 rc = generic_file_mmap(file, vma);
1768 static void cifs_copy_cache_pages(struct address_space *mapping,
1769 struct list_head *pages, int bytes_read, char *data)
1774 while (bytes_read > 0) {
1775 if (list_empty(pages))
1778 page = list_entry(pages->prev, struct page, lru);
1779 list_del(&page->lru);
1781 if (add_to_page_cache_lru(page, mapping, page->index,
1783 page_cache_release(page);
1784 cFYI(1, "Add page cache failed");
1785 data += PAGE_CACHE_SIZE;
1786 bytes_read -= PAGE_CACHE_SIZE;
1789 page_cache_release(page);
1791 target = kmap_atomic(page, KM_USER0);
1793 if (PAGE_CACHE_SIZE > bytes_read) {
1794 memcpy(target, data, bytes_read);
1795 /* zero the tail end of this partial page */
1796 memset(target + bytes_read, 0,
1797 PAGE_CACHE_SIZE - bytes_read);
1800 memcpy(target, data, PAGE_CACHE_SIZE);
1801 bytes_read -= PAGE_CACHE_SIZE;
1803 kunmap_atomic(target, KM_USER0);
1805 flush_dcache_page(page);
1806 SetPageUptodate(page);
1808 data += PAGE_CACHE_SIZE;
1810 /* add page to FS-Cache */
1811 cifs_readpage_to_fscache(mapping->host, page);
1816 static int cifs_readpages(struct file *file, struct address_space *mapping,
1817 struct list_head *page_list, unsigned num_pages)
1823 struct cifs_sb_info *cifs_sb;
1824 struct cifsTconInfo *pTcon;
1825 unsigned int bytes_read = 0;
1826 unsigned int read_size, i;
1827 char *smb_read_data = NULL;
1828 struct smb_com_read_rsp *pSMBr;
1829 struct cifsFileInfo *open_file;
1830 int buf_type = CIFS_NO_BUFFER;
1833 if (file->private_data == NULL) {
1838 open_file = file->private_data;
1839 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1840 pTcon = tlink_tcon(open_file->tlink);
1843 * Reads as many pages as possible from fscache. Returns -ENOBUFS
1844 * immediately if the cookie is negative
1846 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
1851 cFYI(DBG2, "rpages: num pages %d", num_pages);
1852 for (i = 0; i < num_pages; ) {
1853 unsigned contig_pages;
1854 struct page *tmp_page;
1855 unsigned long expected_index;
1857 if (list_empty(page_list))
1860 page = list_entry(page_list->prev, struct page, lru);
1861 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1863 /* count adjacent pages that we will read into */
1866 list_entry(page_list->prev, struct page, lru)->index;
1867 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1868 if (tmp_page->index == expected_index) {
1874 if (contig_pages + i > num_pages)
1875 contig_pages = num_pages - i;
1877 /* for reads over a certain size could initiate async
1880 read_size = contig_pages * PAGE_CACHE_SIZE;
1881 /* Read size needs to be in multiples of one page */
1882 read_size = min_t(const unsigned int, read_size,
1883 cifs_sb->rsize & PAGE_CACHE_MASK);
1884 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
1885 read_size, contig_pages);
1887 while (rc == -EAGAIN) {
1888 if (open_file->invalidHandle) {
1889 rc = cifs_reopen_file(open_file, true);
1894 rc = CIFSSMBRead(xid, pTcon,
1897 &bytes_read, &smb_read_data,
1899 /* BB more RC checks ? */
1900 if (rc == -EAGAIN) {
1901 if (smb_read_data) {
1902 if (buf_type == CIFS_SMALL_BUFFER)
1903 cifs_small_buf_release(smb_read_data);
1904 else if (buf_type == CIFS_LARGE_BUFFER)
1905 cifs_buf_release(smb_read_data);
1906 smb_read_data = NULL;
1910 if ((rc < 0) || (smb_read_data == NULL)) {
1911 cFYI(1, "Read error in readpages: %d", rc);
1913 } else if (bytes_read > 0) {
1914 task_io_account_read(bytes_read);
1915 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1916 cifs_copy_cache_pages(mapping, page_list, bytes_read,
1917 smb_read_data + 4 /* RFC1001 hdr */ +
1918 le16_to_cpu(pSMBr->DataOffset));
1920 i += bytes_read >> PAGE_CACHE_SHIFT;
1921 cifs_stats_bytes_read(pTcon, bytes_read);
1922 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1923 i++; /* account for partial page */
1925 /* server copy of file can have smaller size
1927 /* BB do we need to verify this common case ?
1928 this case is ok - if we are at server EOF
1929 we will hit it on next read */
1934 cFYI(1, "No bytes read (%d) at offset %lld . "
1935 "Cleaning remaining pages from readahead list",
1936 bytes_read, offset);
1937 /* BB turn off caching and do new lookup on
1938 file size at server? */
1941 if (smb_read_data) {
1942 if (buf_type == CIFS_SMALL_BUFFER)
1943 cifs_small_buf_release(smb_read_data);
1944 else if (buf_type == CIFS_LARGE_BUFFER)
1945 cifs_buf_release(smb_read_data);
1946 smb_read_data = NULL;
1951 /* need to free smb_read_data buf before exit */
1952 if (smb_read_data) {
1953 if (buf_type == CIFS_SMALL_BUFFER)
1954 cifs_small_buf_release(smb_read_data);
1955 else if (buf_type == CIFS_LARGE_BUFFER)
1956 cifs_buf_release(smb_read_data);
1957 smb_read_data = NULL;
1965 static int cifs_readpage_worker(struct file *file, struct page *page,
1971 /* Is the page cached? */
1972 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
1976 page_cache_get(page);
1977 read_data = kmap(page);
1978 /* for reads over a certain size could initiate async read ahead */
1980 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1985 cFYI(1, "Bytes read %d", rc);
1987 file->f_path.dentry->d_inode->i_atime =
1988 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1990 if (PAGE_CACHE_SIZE > rc)
1991 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1993 flush_dcache_page(page);
1994 SetPageUptodate(page);
1996 /* send this page to the cache */
1997 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2003 page_cache_release(page);
2009 static int cifs_readpage(struct file *file, struct page *page)
2011 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2017 if (file->private_data == NULL) {
2023 cFYI(1, "readpage %p at offset %d 0x%x\n",
2024 page, (int)offset, (int)offset);
2026 rc = cifs_readpage_worker(file, page, &offset);
2034 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2036 struct cifsFileInfo *open_file;
2038 spin_lock(&cifs_file_list_lock);
2039 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2040 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2041 spin_unlock(&cifs_file_list_lock);
2045 spin_unlock(&cifs_file_list_lock);
2049 /* We do not want to update the file size from server for inodes
2050 open for write - to avoid races with writepage extending
2051 the file - in the future we could consider allowing
2052 refreshing the inode only on increases in the file size
2053 but this is tricky to do without racing with writebehind
2054 page caching in the current Linux kernel design */
2055 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2060 if (is_inode_writable(cifsInode)) {
2061 /* This inode is open for write at least once */
2062 struct cifs_sb_info *cifs_sb;
2064 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2065 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2066 /* since no page cache to corrupt on directio
2067 we can change size safely */
2071 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2079 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2080 loff_t pos, unsigned len, unsigned flags,
2081 struct page **pagep, void **fsdata)
2083 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2084 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2085 loff_t page_start = pos & PAGE_MASK;
2090 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2092 page = grab_cache_page_write_begin(mapping, index, flags);
2098 if (PageUptodate(page))
2102 * If we write a full page it will be up to date, no need to read from
2103 * the server. If the write is short, we'll end up doing a sync write
2106 if (len == PAGE_CACHE_SIZE)
2110 * optimize away the read when we have an oplock, and we're not
2111 * expecting to use any of the data we'd be reading in. That
2112 * is, when the page lies beyond the EOF, or straddles the EOF
2113 * and the write will cover all of the existing data.
2115 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2116 i_size = i_size_read(mapping->host);
2117 if (page_start >= i_size ||
2118 (offset == 0 && (pos + len) >= i_size)) {
2119 zero_user_segments(page, 0, offset,
2123 * PageChecked means that the parts of the page
2124 * to which we're not writing are considered up
2125 * to date. Once the data is copied to the
2126 * page, it can be set uptodate.
2128 SetPageChecked(page);
2133 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2135 * might as well read a page, it is fast enough. If we get
2136 * an error, we don't need to return it. cifs_write_end will
2137 * do a sync write instead since PG_uptodate isn't set.
2139 cifs_readpage_worker(file, page, &page_start);
2141 /* we could try using another file handle if there is one -
2142 but how would we lock it to prevent close of that handle
2143 racing with this read? In any case
2144 this will be written out by write_end so is fine */
2151 static int cifs_release_page(struct page *page, gfp_t gfp)
2153 if (PagePrivate(page))
2156 return cifs_fscache_release_page(page, gfp);
2159 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2161 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2164 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2167 void cifs_oplock_break(struct work_struct *work)
2169 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2171 struct inode *inode = cfile->dentry->d_inode;
2172 struct cifsInodeInfo *cinode = CIFS_I(inode);
2175 if (inode && S_ISREG(inode->i_mode)) {
2176 if (cinode->clientCanCacheRead)
2177 break_lease(inode, O_RDONLY);
2179 break_lease(inode, O_WRONLY);
2180 rc = filemap_fdatawrite(inode->i_mapping);
2181 if (cinode->clientCanCacheRead == 0) {
2182 rc = filemap_fdatawait(inode->i_mapping);
2183 mapping_set_error(inode->i_mapping, rc);
2184 invalidate_remote_inode(inode);
2186 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2190 * releasing stale oplock after recent reconnect of smb session using
2191 * a now incorrect file handle is not a data integrity issue but do
2192 * not bother sending an oplock release if session to server still is
2193 * disconnected since oplock already released by the server
2195 if (!cfile->oplock_break_cancelled) {
2196 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2197 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false);
2198 cFYI(1, "Oplock release rc = %d", rc);
2202 * We might have kicked in before is_valid_oplock_break()
2203 * finished grabbing reference for us. Make sure it's done by
2204 * waiting for cifs_file_list_lock.
2206 spin_lock(&cifs_file_list_lock);
2207 spin_unlock(&cifs_file_list_lock);
2209 cifs_oplock_break_put(cfile);
2212 /* must be called while holding cifs_file_list_lock */
2213 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2215 cifs_sb_active(cfile->dentry->d_sb);
2216 cifsFileInfo_get(cfile);
2219 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2221 struct super_block *sb = cfile->dentry->d_sb;
2223 cifsFileInfo_put(cfile);
2224 cifs_sb_deactive(sb);
2227 const struct address_space_operations cifs_addr_ops = {
2228 .readpage = cifs_readpage,
2229 .readpages = cifs_readpages,
2230 .writepage = cifs_writepage,
2231 .writepages = cifs_writepages,
2232 .write_begin = cifs_write_begin,
2233 .write_end = cifs_write_end,
2234 .set_page_dirty = __set_page_dirty_nobuffers,
2235 .releasepage = cifs_release_page,
2236 .invalidatepage = cifs_invalidate_page,
2237 /* .sync_page = cifs_sync_page, */
2242 * cifs_readpages requires the server to support a buffer large enough to
2243 * contain the header plus one complete page of data. Otherwise, we need
2244 * to leave cifs_readpages out of the address space operations.
2246 const struct address_space_operations cifs_addr_ops_smallbuf = {
2247 .readpage = cifs_readpage,
2248 .writepage = cifs_writepage,
2249 .writepages = cifs_writepages,
2250 .write_begin = cifs_write_begin,
2251 .write_end = cifs_write_end,
2252 .set_page_dirty = __set_page_dirty_nobuffers,
2253 .releasepage = cifs_release_page,
2254 .invalidatepage = cifs_invalidate_page,
2255 /* .sync_page = cifs_sync_page, */