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 cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
291 struct cifsLockInfo *li, *tmp;
293 spin_lock(&cifs_file_list_lock);
294 if (--cifs_file->count > 0) {
295 spin_unlock(&cifs_file_list_lock);
299 /* remove it from the lists */
300 list_del(&cifs_file->flist);
301 list_del(&cifs_file->tlist);
303 if (list_empty(&cifsi->openFileList)) {
304 cFYI(1, "closing last open instance for inode %p",
305 cifs_file->dentry->d_inode);
307 /* in strict cache mode we need invalidate mapping on the last
308 close because it may cause a error when we open this file
309 again and get at least level II oplock */
310 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
311 CIFS_I(inode)->invalid_mapping = true;
313 cifs_set_oplock_level(cifsi, 0);
315 spin_unlock(&cifs_file_list_lock);
317 if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
321 rc = CIFSSMBClose(xid, tcon, cifs_file->netfid);
325 /* Delete any outstanding lock records. We'll lose them when the file
328 mutex_lock(&cifs_file->lock_mutex);
329 list_for_each_entry_safe(li, tmp, &cifs_file->llist, llist) {
330 list_del(&li->llist);
333 mutex_unlock(&cifs_file->lock_mutex);
335 cifs_put_tlink(cifs_file->tlink);
336 dput(cifs_file->dentry);
340 int cifs_open(struct inode *inode, struct file *file)
345 struct cifs_sb_info *cifs_sb;
346 struct cifsTconInfo *tcon;
347 struct tcon_link *tlink;
348 struct cifsFileInfo *pCifsFile = NULL;
349 struct cifsInodeInfo *pCifsInode;
350 char *full_path = NULL;
351 bool posix_open_ok = false;
356 cifs_sb = CIFS_SB(inode->i_sb);
357 tlink = cifs_sb_tlink(cifs_sb);
360 return PTR_ERR(tlink);
362 tcon = tlink_tcon(tlink);
364 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
366 full_path = build_path_from_dentry(file->f_path.dentry);
367 if (full_path == NULL) {
372 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
373 inode, file->f_flags, full_path);
380 if (!tcon->broken_posix_open && tcon->unix_ext &&
381 (tcon->ses->capabilities & CAP_UNIX) &&
382 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
383 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
384 /* can not refresh inode info since size could be stale */
385 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
386 cifs_sb->mnt_file_mode /* ignored */,
387 file->f_flags, &oplock, &netfid, xid);
389 cFYI(1, "posix open succeeded");
390 posix_open_ok = true;
391 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
392 if (tcon->ses->serverNOS)
393 cERROR(1, "server %s of type %s returned"
394 " unexpected error on SMB posix open"
395 ", disabling posix open support."
396 " Check if server update available.",
397 tcon->ses->serverName,
398 tcon->ses->serverNOS);
399 tcon->broken_posix_open = true;
400 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
401 (rc != -EOPNOTSUPP)) /* path not found or net err */
403 /* else fallthrough to retry open the old way on network i/o
407 if (!posix_open_ok) {
408 rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
409 file->f_flags, &oplock, &netfid, xid);
414 pCifsFile = cifs_new_fileinfo(netfid, file, tlink, oplock);
415 if (pCifsFile == NULL) {
416 CIFSSMBClose(xid, tcon, netfid);
421 cifs_fscache_set_inode_cookie(inode, file);
423 if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
424 /* time to set mode which we can not set earlier due to
425 problems creating new read-only files */
426 struct cifs_unix_set_info_args args = {
427 .mode = inode->i_mode,
430 .ctime = NO_CHANGE_64,
431 .atime = NO_CHANGE_64,
432 .mtime = NO_CHANGE_64,
435 CIFSSMBUnixSetFileInfo(xid, tcon, &args, netfid,
442 cifs_put_tlink(tlink);
446 /* Try to reacquire byte range locks that were released when session */
447 /* to server was lost */
448 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
452 /* BB list all locks open on this file and relock */
457 static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
462 struct cifs_sb_info *cifs_sb;
463 struct cifsTconInfo *tcon;
464 struct cifsInodeInfo *pCifsInode;
466 char *full_path = NULL;
468 int disposition = FILE_OPEN;
472 mutex_lock(&pCifsFile->fh_mutex);
473 if (!pCifsFile->invalidHandle) {
474 mutex_unlock(&pCifsFile->fh_mutex);
480 inode = pCifsFile->dentry->d_inode;
481 cifs_sb = CIFS_SB(inode->i_sb);
482 tcon = tlink_tcon(pCifsFile->tlink);
484 /* can not grab rename sem here because various ops, including
485 those that already have the rename sem can end up causing writepage
486 to get called and if the server was down that means we end up here,
487 and we can never tell if the caller already has the rename_sem */
488 full_path = build_path_from_dentry(pCifsFile->dentry);
489 if (full_path == NULL) {
491 mutex_unlock(&pCifsFile->fh_mutex);
496 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
497 inode, pCifsFile->f_flags, full_path);
504 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
505 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
506 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
509 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
510 * original open. Must mask them off for a reopen.
512 unsigned int oflags = pCifsFile->f_flags &
513 ~(O_CREAT | O_EXCL | O_TRUNC);
515 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
516 cifs_sb->mnt_file_mode /* ignored */,
517 oflags, &oplock, &netfid, xid);
519 cFYI(1, "posix reopen succeeded");
522 /* fallthrough to retry open the old way on errors, especially
523 in the reconnect path it is important to retry hard */
526 desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
528 /* Can not refresh inode by passing in file_info buf to be returned
529 by SMBOpen and then calling get_inode_info with returned buf
530 since file might have write behind data that needs to be flushed
531 and server version of file size can be stale. If we knew for sure
532 that inode was not dirty locally we could do this */
534 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
535 CREATE_NOT_DIR, &netfid, &oplock, NULL,
536 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
537 CIFS_MOUNT_MAP_SPECIAL_CHR);
539 mutex_unlock(&pCifsFile->fh_mutex);
540 cFYI(1, "cifs_open returned 0x%x", rc);
541 cFYI(1, "oplock: %d", oplock);
542 goto reopen_error_exit;
546 pCifsFile->netfid = netfid;
547 pCifsFile->invalidHandle = false;
548 mutex_unlock(&pCifsFile->fh_mutex);
549 pCifsInode = CIFS_I(inode);
552 rc = filemap_write_and_wait(inode->i_mapping);
553 mapping_set_error(inode->i_mapping, rc);
556 rc = cifs_get_inode_info_unix(&inode,
557 full_path, inode->i_sb, xid);
559 rc = cifs_get_inode_info(&inode,
560 full_path, NULL, inode->i_sb,
562 } /* else we are writing out data to server already
563 and could deadlock if we tried to flush data, and
564 since we do not know if we have data that would
565 invalidate the current end of file on the server
566 we can not go to the server to get the new inod
569 cifs_set_oplock_level(pCifsInode, oplock);
571 cifs_relock_file(pCifsFile);
579 int cifs_close(struct inode *inode, struct file *file)
581 cifsFileInfo_put(file->private_data);
582 file->private_data = NULL;
584 /* return code from the ->release op is always ignored */
588 int cifs_closedir(struct inode *inode, struct file *file)
592 struct cifsFileInfo *pCFileStruct = file->private_data;
595 cFYI(1, "Closedir inode = 0x%p", inode);
600 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
602 cFYI(1, "Freeing private data in close dir");
603 spin_lock(&cifs_file_list_lock);
604 if (!pCFileStruct->srch_inf.endOfSearch &&
605 !pCFileStruct->invalidHandle) {
606 pCFileStruct->invalidHandle = true;
607 spin_unlock(&cifs_file_list_lock);
608 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
609 cFYI(1, "Closing uncompleted readdir with rc %d",
611 /* not much we can do if it fails anyway, ignore rc */
614 spin_unlock(&cifs_file_list_lock);
615 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
617 cFYI(1, "closedir free smb buf in srch struct");
618 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
619 if (pCFileStruct->srch_inf.smallBuf)
620 cifs_small_buf_release(ptmp);
622 cifs_buf_release(ptmp);
624 cifs_put_tlink(pCFileStruct->tlink);
625 kfree(file->private_data);
626 file->private_data = NULL;
628 /* BB can we lock the filestruct while this is going on? */
633 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
634 __u64 offset, __u8 lockType)
636 struct cifsLockInfo *li =
637 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
643 mutex_lock(&fid->lock_mutex);
644 list_add(&li->llist, &fid->llist);
645 mutex_unlock(&fid->lock_mutex);
649 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
655 bool wait_flag = false;
656 struct cifs_sb_info *cifs_sb;
657 struct cifsTconInfo *tcon;
659 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
660 bool posix_locking = 0;
662 length = 1 + pfLock->fl_end - pfLock->fl_start;
666 cFYI(1, "Lock parm: 0x%x flockflags: "
667 "0x%x flocktype: 0x%x start: %lld end: %lld",
668 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
671 if (pfLock->fl_flags & FL_POSIX)
673 if (pfLock->fl_flags & FL_FLOCK)
675 if (pfLock->fl_flags & FL_SLEEP) {
676 cFYI(1, "Blocking lock");
679 if (pfLock->fl_flags & FL_ACCESS)
680 cFYI(1, "Process suspended by mandatory locking - "
681 "not implemented yet");
682 if (pfLock->fl_flags & FL_LEASE)
683 cFYI(1, "Lease on file - not implemented yet");
684 if (pfLock->fl_flags &
685 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
686 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
688 if (pfLock->fl_type == F_WRLCK) {
691 } else if (pfLock->fl_type == F_UNLCK) {
694 /* Check if unlock includes more than
696 } else if (pfLock->fl_type == F_RDLCK) {
698 lockType |= LOCKING_ANDX_SHARED_LOCK;
700 } else if (pfLock->fl_type == F_EXLCK) {
703 } else if (pfLock->fl_type == F_SHLCK) {
705 lockType |= LOCKING_ANDX_SHARED_LOCK;
708 cFYI(1, "Unknown type of lock");
710 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
711 tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
712 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
714 if ((tcon->ses->capabilities & CAP_UNIX) &&
715 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
716 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
718 /* BB add code here to normalize offset and length to
719 account for negative length which we can not accept over the
724 if (lockType & LOCKING_ANDX_SHARED_LOCK)
725 posix_lock_type = CIFS_RDLCK;
727 posix_lock_type = CIFS_WRLCK;
728 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
730 posix_lock_type, wait_flag);
735 /* BB we could chain these into one lock request BB */
736 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
737 0, 1, lockType, 0 /* wait flag */, 0);
739 rc = CIFSSMBLock(xid, tcon, netfid, length,
740 pfLock->fl_start, 1 /* numUnlock */ ,
741 0 /* numLock */ , lockType,
742 0 /* wait flag */, 0);
743 pfLock->fl_type = F_UNLCK;
745 cERROR(1, "Error unlocking previously locked "
746 "range %d during test of lock", rc);
750 /* if rc == ERR_SHARING_VIOLATION ? */
753 if (lockType & LOCKING_ANDX_SHARED_LOCK) {
754 pfLock->fl_type = F_WRLCK;
756 rc = CIFSSMBLock(xid, tcon, netfid, length,
757 pfLock->fl_start, 0, 1,
758 lockType | LOCKING_ANDX_SHARED_LOCK,
759 0 /* wait flag */, 0);
761 rc = CIFSSMBLock(xid, tcon, netfid,
762 length, pfLock->fl_start, 1, 0,
764 LOCKING_ANDX_SHARED_LOCK,
765 0 /* wait flag */, 0);
766 pfLock->fl_type = F_RDLCK;
768 cERROR(1, "Error unlocking "
769 "previously locked range %d "
770 "during test of lock", rc);
773 pfLock->fl_type = F_WRLCK;
783 if (!numLock && !numUnlock) {
784 /* if no lock or unlock then nothing
785 to do since we do not know what it is */
792 if (lockType & LOCKING_ANDX_SHARED_LOCK)
793 posix_lock_type = CIFS_RDLCK;
795 posix_lock_type = CIFS_WRLCK;
798 posix_lock_type = CIFS_UNLCK;
800 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
802 posix_lock_type, wait_flag);
804 struct cifsFileInfo *fid = file->private_data;
807 rc = CIFSSMBLock(xid, tcon, netfid, length,
808 pfLock->fl_start, 0, numLock, lockType,
812 /* For Windows locks we must store them. */
813 rc = store_file_lock(fid, length,
814 pfLock->fl_start, lockType);
816 } else if (numUnlock) {
817 /* For each stored lock that this unlock overlaps
818 completely, unlock it. */
820 struct cifsLockInfo *li, *tmp;
823 mutex_lock(&fid->lock_mutex);
824 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
825 if (pfLock->fl_start <= li->offset &&
826 (pfLock->fl_start + length) >=
827 (li->offset + li->length)) {
828 stored_rc = CIFSSMBLock(xid, tcon,
835 list_del(&li->llist);
840 mutex_unlock(&fid->lock_mutex);
844 if (pfLock->fl_flags & FL_POSIX)
845 posix_lock_file_wait(file, pfLock);
850 /* update the file size (if needed) after a write */
852 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
853 unsigned int bytes_written)
855 loff_t end_of_write = offset + bytes_written;
857 if (end_of_write > cifsi->server_eof)
858 cifsi->server_eof = end_of_write;
861 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
862 size_t write_size, loff_t *poffset)
864 struct inode *inode = file->f_path.dentry->d_inode;
866 unsigned int bytes_written = 0;
867 unsigned int total_written;
868 struct cifs_sb_info *cifs_sb;
869 struct cifsTconInfo *pTcon;
871 struct cifsFileInfo *open_file;
872 struct cifsInodeInfo *cifsi = CIFS_I(inode);
874 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
876 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
877 *poffset, file->f_path.dentry->d_name.name); */
879 if (file->private_data == NULL)
882 open_file = file->private_data;
883 pTcon = tlink_tcon(open_file->tlink);
885 rc = generic_write_checks(file, poffset, &write_size, 0);
891 for (total_written = 0; write_size > total_written;
892 total_written += bytes_written) {
894 while (rc == -EAGAIN) {
895 if (file->private_data == NULL) {
896 /* file has been closed on us */
898 /* if we have gotten here we have written some data
899 and blocked, and the file has been freed on us while
900 we blocked so return what we managed to write */
901 return total_written;
903 if (open_file->invalidHandle) {
904 /* we could deadlock if we called
905 filemap_fdatawait from here so tell
906 reopen_file not to flush data to server
908 rc = cifs_reopen_file(open_file, false);
913 rc = CIFSSMBWrite(xid, pTcon,
915 min_t(const int, cifs_sb->wsize,
916 write_size - total_written),
917 *poffset, &bytes_written,
918 NULL, write_data + total_written, 0);
920 if (rc || (bytes_written == 0)) {
928 cifs_update_eof(cifsi, *poffset, bytes_written);
929 *poffset += bytes_written;
933 cifs_stats_bytes_written(pTcon, total_written);
935 /* Do not update local mtime - server will set its actual value on write
936 * inode->i_ctime = inode->i_mtime =
937 * current_fs_time(inode->i_sb);*/
938 if (total_written > 0) {
939 spin_lock(&inode->i_lock);
940 if (*poffset > inode->i_size)
941 i_size_write(inode, *poffset);
942 spin_unlock(&inode->i_lock);
944 mark_inode_dirty_sync(inode);
947 return total_written;
950 static ssize_t cifs_write(struct cifsFileInfo *open_file,
951 const char *write_data, size_t write_size,
955 unsigned int bytes_written = 0;
956 unsigned int total_written;
957 struct cifs_sb_info *cifs_sb;
958 struct cifsTconInfo *pTcon;
960 struct dentry *dentry = open_file->dentry;
961 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
963 cifs_sb = CIFS_SB(dentry->d_sb);
965 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
966 *poffset, dentry->d_name.name);
968 pTcon = tlink_tcon(open_file->tlink);
972 for (total_written = 0; write_size > total_written;
973 total_written += bytes_written) {
975 while (rc == -EAGAIN) {
976 if (open_file->invalidHandle) {
977 /* we could deadlock if we called
978 filemap_fdatawait from here so tell
979 reopen_file not to flush data to
981 rc = cifs_reopen_file(open_file, false);
985 if (experimEnabled || (pTcon->ses->server &&
986 ((pTcon->ses->server->secMode &
987 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
992 len = min((size_t)cifs_sb->wsize,
993 write_size - total_written);
994 /* iov[0] is reserved for smb header */
995 iov[1].iov_base = (char *)write_data +
997 iov[1].iov_len = len;
998 rc = CIFSSMBWrite2(xid, pTcon,
999 open_file->netfid, len,
1000 *poffset, &bytes_written,
1003 rc = CIFSSMBWrite(xid, pTcon,
1005 min_t(const int, cifs_sb->wsize,
1006 write_size - total_written),
1007 *poffset, &bytes_written,
1008 write_data + total_written,
1011 if (rc || (bytes_written == 0)) {
1019 cifs_update_eof(cifsi, *poffset, bytes_written);
1020 *poffset += bytes_written;
1024 cifs_stats_bytes_written(pTcon, total_written);
1026 if (total_written > 0) {
1027 spin_lock(&dentry->d_inode->i_lock);
1028 if (*poffset > dentry->d_inode->i_size)
1029 i_size_write(dentry->d_inode, *poffset);
1030 spin_unlock(&dentry->d_inode->i_lock);
1032 mark_inode_dirty_sync(dentry->d_inode);
1034 return total_written;
1037 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1040 struct cifsFileInfo *open_file = NULL;
1041 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1043 /* only filter by fsuid on multiuser mounts */
1044 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1047 spin_lock(&cifs_file_list_lock);
1048 /* we could simply get the first_list_entry since write-only entries
1049 are always at the end of the list but since the first entry might
1050 have a close pending, we go through the whole list */
1051 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1052 if (fsuid_only && open_file->uid != current_fsuid())
1054 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1055 if (!open_file->invalidHandle) {
1056 /* found a good file */
1057 /* lock it so it will not be closed on us */
1058 cifsFileInfo_get(open_file);
1059 spin_unlock(&cifs_file_list_lock);
1061 } /* else might as well continue, and look for
1062 another, or simply have the caller reopen it
1063 again rather than trying to fix this handle */
1064 } else /* write only file */
1065 break; /* write only files are last so must be done */
1067 spin_unlock(&cifs_file_list_lock);
1071 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1074 struct cifsFileInfo *open_file;
1075 struct cifs_sb_info *cifs_sb;
1076 bool any_available = false;
1079 /* Having a null inode here (because mapping->host was set to zero by
1080 the VFS or MM) should not happen but we had reports of on oops (due to
1081 it being zero) during stress testcases so we need to check for it */
1083 if (cifs_inode == NULL) {
1084 cERROR(1, "Null inode passed to cifs_writeable_file");
1089 cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1091 /* only filter by fsuid on multiuser mounts */
1092 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1095 spin_lock(&cifs_file_list_lock);
1097 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1098 if (!any_available && open_file->pid != current->tgid)
1100 if (fsuid_only && open_file->uid != current_fsuid())
1102 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1103 cifsFileInfo_get(open_file);
1105 if (!open_file->invalidHandle) {
1106 /* found a good writable file */
1107 spin_unlock(&cifs_file_list_lock);
1111 spin_unlock(&cifs_file_list_lock);
1113 /* Had to unlock since following call can block */
1114 rc = cifs_reopen_file(open_file, false);
1118 /* if it fails, try another handle if possible */
1119 cFYI(1, "wp failed on reopen file");
1120 cifsFileInfo_put(open_file);
1122 spin_lock(&cifs_file_list_lock);
1124 /* else we simply continue to the next entry. Thus
1125 we do not loop on reopen errors. If we
1126 can not reopen the file, for example if we
1127 reconnected to a server with another client
1128 racing to delete or lock the file we would not
1129 make progress if we restarted before the beginning
1130 of the loop here. */
1133 /* couldn't find useable FH with same pid, try any available */
1134 if (!any_available) {
1135 any_available = true;
1136 goto refind_writable;
1138 spin_unlock(&cifs_file_list_lock);
1142 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1144 struct address_space *mapping = page->mapping;
1145 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1148 int bytes_written = 0;
1149 struct cifs_sb_info *cifs_sb;
1150 struct inode *inode;
1151 struct cifsFileInfo *open_file;
1153 if (!mapping || !mapping->host)
1156 inode = page->mapping->host;
1157 cifs_sb = CIFS_SB(inode->i_sb);
1159 offset += (loff_t)from;
1160 write_data = kmap(page);
1163 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1168 /* racing with truncate? */
1169 if (offset > mapping->host->i_size) {
1171 return 0; /* don't care */
1174 /* check to make sure that we are not extending the file */
1175 if (mapping->host->i_size - offset < (loff_t)to)
1176 to = (unsigned)(mapping->host->i_size - offset);
1178 open_file = find_writable_file(CIFS_I(mapping->host), false);
1180 bytes_written = cifs_write(open_file, write_data,
1181 to - from, &offset);
1182 cifsFileInfo_put(open_file);
1183 /* Does mm or vfs already set times? */
1184 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1185 if ((bytes_written > 0) && (offset))
1187 else if (bytes_written < 0)
1190 cFYI(1, "No writeable filehandles for inode");
1198 static int cifs_writepages(struct address_space *mapping,
1199 struct writeback_control *wbc)
1201 unsigned int bytes_to_write;
1202 unsigned int bytes_written;
1203 struct cifs_sb_info *cifs_sb;
1207 int range_whole = 0;
1214 struct cifsFileInfo *open_file;
1215 struct cifsTconInfo *tcon;
1216 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1218 struct pagevec pvec;
1223 cifs_sb = CIFS_SB(mapping->host->i_sb);
1226 * If wsize is smaller that the page cache size, default to writing
1227 * one page at a time via cifs_writepage
1229 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1230 return generic_writepages(mapping, wbc);
1232 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1234 return generic_writepages(mapping, wbc);
1237 * if there's no open file, then this is likely to fail too,
1238 * but it'll at least handle the return. Maybe it should be
1241 open_file = find_writable_file(CIFS_I(mapping->host), false);
1244 return generic_writepages(mapping, wbc);
1247 tcon = tlink_tcon(open_file->tlink);
1248 if (!experimEnabled && tcon->ses->server->secMode &
1249 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1250 cifsFileInfo_put(open_file);
1252 return generic_writepages(mapping, wbc);
1254 cifsFileInfo_put(open_file);
1258 pagevec_init(&pvec, 0);
1259 if (wbc->range_cyclic) {
1260 index = mapping->writeback_index; /* Start from prev offset */
1263 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1264 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1265 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1270 while (!done && (index <= end) &&
1271 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1272 PAGECACHE_TAG_DIRTY,
1273 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1282 for (i = 0; i < nr_pages; i++) {
1283 page = pvec.pages[i];
1285 * At this point we hold neither mapping->tree_lock nor
1286 * lock on the page itself: the page may be truncated or
1287 * invalidated (changing page->mapping to NULL), or even
1288 * swizzled back from swapper_space to tmpfs file
1294 else if (!trylock_page(page))
1297 if (unlikely(page->mapping != mapping)) {
1302 if (!wbc->range_cyclic && page->index > end) {
1308 if (next && (page->index != next)) {
1309 /* Not next consecutive page */
1314 if (wbc->sync_mode != WB_SYNC_NONE)
1315 wait_on_page_writeback(page);
1317 if (PageWriteback(page) ||
1318 !clear_page_dirty_for_io(page)) {
1324 * This actually clears the dirty bit in the radix tree.
1325 * See cifs_writepage() for more commentary.
1327 set_page_writeback(page);
1329 if (page_offset(page) >= mapping->host->i_size) {
1332 end_page_writeback(page);
1337 * BB can we get rid of this? pages are held by pvec
1339 page_cache_get(page);
1341 len = min(mapping->host->i_size - page_offset(page),
1342 (loff_t)PAGE_CACHE_SIZE);
1344 /* reserve iov[0] for the smb header */
1346 iov[n_iov].iov_base = kmap(page);
1347 iov[n_iov].iov_len = len;
1348 bytes_to_write += len;
1352 offset = page_offset(page);
1354 next = page->index + 1;
1355 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1360 open_file = find_writable_file(CIFS_I(mapping->host),
1363 cERROR(1, "No writable handles for inode");
1366 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1367 bytes_to_write, offset,
1368 &bytes_written, iov, n_iov,
1370 cifsFileInfo_put(open_file);
1373 cFYI(1, "Write2 rc=%d, wrote=%u", rc, bytes_written);
1376 * For now, treat a short write as if nothing got
1377 * written. A zero length write however indicates
1378 * ENOSPC or EFBIG. We have no way to know which
1379 * though, so call it ENOSPC for now. EFBIG would
1380 * get translated to AS_EIO anyway.
1382 * FIXME: make it take into account the data that did
1386 if (bytes_written == 0)
1388 else if (bytes_written < bytes_to_write)
1392 /* retry on data-integrity flush */
1393 if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN)
1396 /* fix the stats and EOF */
1397 if (bytes_written > 0) {
1398 cifs_stats_bytes_written(tcon, bytes_written);
1399 cifs_update_eof(cifsi, offset, bytes_written);
1402 for (i = 0; i < n_iov; i++) {
1403 page = pvec.pages[first + i];
1404 /* on retryable write error, redirty page */
1406 redirty_page_for_writepage(wbc, page);
1411 end_page_writeback(page);
1412 page_cache_release(page);
1416 mapping_set_error(mapping, rc);
1420 if ((wbc->nr_to_write -= n_iov) <= 0)
1424 /* Need to re-find the pages we skipped */
1425 index = pvec.pages[0]->index + 1;
1427 pagevec_release(&pvec);
1429 if (!scanned && !done) {
1431 * We hit the last page and there is more work to be done: wrap
1432 * back to the start of the file
1438 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1439 mapping->writeback_index = index;
1446 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1452 /* BB add check for wbc flags */
1453 page_cache_get(page);
1454 if (!PageUptodate(page))
1455 cFYI(1, "ppw - page not up to date");
1458 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1460 * A writepage() implementation always needs to do either this,
1461 * or re-dirty the page with "redirty_page_for_writepage()" in
1462 * the case of a failure.
1464 * Just unlocking the page will cause the radix tree tag-bits
1465 * to fail to update with the state of the page correctly.
1467 set_page_writeback(page);
1468 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1469 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1471 end_page_writeback(page);
1472 page_cache_release(page);
1477 static int cifs_write_end(struct file *file, struct address_space *mapping,
1478 loff_t pos, unsigned len, unsigned copied,
1479 struct page *page, void *fsdata)
1482 struct inode *inode = mapping->host;
1484 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1487 if (PageChecked(page)) {
1489 SetPageUptodate(page);
1490 ClearPageChecked(page);
1491 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1492 SetPageUptodate(page);
1494 if (!PageUptodate(page)) {
1496 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1500 /* this is probably better than directly calling
1501 partialpage_write since in this function the file handle is
1502 known which we might as well leverage */
1503 /* BB check if anything else missing out of ppw
1504 such as updating last write time */
1505 page_data = kmap(page);
1506 rc = cifs_write(file->private_data, page_data + offset,
1508 /* if (rc < 0) should we set writebehind rc? */
1515 set_page_dirty(page);
1519 spin_lock(&inode->i_lock);
1520 if (pos > inode->i_size)
1521 i_size_write(inode, pos);
1522 spin_unlock(&inode->i_lock);
1526 page_cache_release(page);
1531 int cifs_strict_fsync(struct file *file, int datasync)
1535 struct cifsTconInfo *tcon;
1536 struct cifsFileInfo *smbfile = file->private_data;
1537 struct inode *inode = file->f_path.dentry->d_inode;
1538 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
1542 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1543 file->f_path.dentry->d_name.name, datasync);
1545 if (!CIFS_I(inode)->clientCanCacheRead)
1546 cifs_invalidate_mapping(inode);
1548 tcon = tlink_tcon(smbfile->tlink);
1549 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1550 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1556 int cifs_fsync(struct file *file, int datasync)
1560 struct cifsTconInfo *tcon;
1561 struct cifsFileInfo *smbfile = file->private_data;
1562 struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1566 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1567 file->f_path.dentry->d_name.name, datasync);
1569 tcon = tlink_tcon(smbfile->tlink);
1570 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1571 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1577 /* static void cifs_sync_page(struct page *page)
1579 struct address_space *mapping;
1580 struct inode *inode;
1581 unsigned long index = page->index;
1582 unsigned int rpages = 0;
1585 cFYI(1, "sync page %p", page);
1586 mapping = page->mapping;
1589 inode = mapping->host;
1593 /* fill in rpages then
1594 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1596 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1606 * As file closes, flush all cached write data for this inode checking
1607 * for write behind errors.
1609 int cifs_flush(struct file *file, fl_owner_t id)
1611 struct inode *inode = file->f_path.dentry->d_inode;
1614 if (file->f_mode & FMODE_WRITE)
1615 rc = filemap_write_and_wait(inode->i_mapping);
1617 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1623 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
1628 for (i = 0; i < num_pages; i++) {
1629 pages[i] = alloc_page(__GFP_HIGHMEM);
1632 * save number of pages we have already allocated and
1633 * return with ENOMEM error
1644 for (i = 0; i < num_pages; i++)
1650 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
1655 clen = min_t(const size_t, len, wsize);
1656 num_pages = clen / PAGE_CACHE_SIZE;
1657 if (clen % PAGE_CACHE_SIZE)
1667 cifs_iovec_write(struct file *file, const struct iovec *iov,
1668 unsigned long nr_segs, loff_t *poffset)
1670 size_t total_written = 0, written = 0;
1671 unsigned long num_pages, npages;
1672 size_t copied, len, cur_len, i;
1673 struct kvec *to_send;
1674 struct page **pages;
1676 struct inode *inode;
1677 struct cifsFileInfo *open_file;
1678 struct cifsTconInfo *pTcon;
1679 struct cifs_sb_info *cifs_sb;
1682 len = iov_length(iov, nr_segs);
1686 rc = generic_write_checks(file, poffset, &len, 0);
1690 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1691 num_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
1693 pages = kmalloc(sizeof(struct pages *)*num_pages, GFP_KERNEL);
1697 to_send = kmalloc(sizeof(struct kvec)*(num_pages + 1), GFP_KERNEL);
1703 rc = cifs_write_allocate_pages(pages, num_pages);
1711 open_file = file->private_data;
1712 pTcon = tlink_tcon(open_file->tlink);
1713 inode = file->f_path.dentry->d_inode;
1715 iov_iter_init(&it, iov, nr_segs, len, 0);
1719 size_t save_len = cur_len;
1720 for (i = 0; i < npages; i++) {
1721 copied = min_t(const size_t, cur_len, PAGE_CACHE_SIZE);
1722 copied = iov_iter_copy_from_user(pages[i], &it, 0,
1725 iov_iter_advance(&it, copied);
1726 to_send[i+1].iov_base = kmap(pages[i]);
1727 to_send[i+1].iov_len = copied;
1730 cur_len = save_len - cur_len;
1733 if (open_file->invalidHandle) {
1734 rc = cifs_reopen_file(open_file, false);
1738 rc = CIFSSMBWrite2(xid, pTcon, open_file->netfid,
1739 cur_len, *poffset, &written,
1740 to_send, npages, 0);
1741 } while (rc == -EAGAIN);
1743 for (i = 0; i < npages; i++)
1748 total_written += written;
1749 cifs_update_eof(CIFS_I(inode), *poffset, written);
1750 *poffset += written;
1751 } else if (rc < 0) {
1757 /* get length and number of kvecs of the next write */
1758 npages = get_numpages(cifs_sb->wsize, len, &cur_len);
1761 if (total_written > 0) {
1762 spin_lock(&inode->i_lock);
1763 if (*poffset > inode->i_size)
1764 i_size_write(inode, *poffset);
1765 spin_unlock(&inode->i_lock);
1768 cifs_stats_bytes_written(pTcon, total_written);
1769 mark_inode_dirty_sync(inode);
1771 for (i = 0; i < num_pages; i++)
1776 return total_written;
1779 static ssize_t cifs_user_writev(struct kiocb *iocb, const struct iovec *iov,
1780 unsigned long nr_segs, loff_t pos)
1783 struct inode *inode;
1785 inode = iocb->ki_filp->f_path.dentry->d_inode;
1788 * BB - optimize the way when signing is disabled. We can drop this
1789 * extra memory-to-memory copying and use iovec buffers for constructing
1793 written = cifs_iovec_write(iocb->ki_filp, iov, nr_segs, &pos);
1795 CIFS_I(inode)->invalid_mapping = true;
1802 ssize_t cifs_strict_writev(struct kiocb *iocb, const struct iovec *iov,
1803 unsigned long nr_segs, loff_t pos)
1805 struct inode *inode;
1807 inode = iocb->ki_filp->f_path.dentry->d_inode;
1809 if (CIFS_I(inode)->clientCanCacheAll)
1810 return generic_file_aio_write(iocb, iov, nr_segs, pos);
1813 * In strict cache mode we need to write the data to the server exactly
1814 * from the pos to pos+len-1 rather than flush all affected pages
1815 * because it may cause a error with mandatory locks on these pages but
1816 * not on the region from pos to ppos+len-1.
1819 return cifs_user_writev(iocb, iov, nr_segs, pos);
1823 cifs_iovec_read(struct file *file, const struct iovec *iov,
1824 unsigned long nr_segs, loff_t *poffset)
1828 unsigned int total_read, bytes_read = 0;
1829 size_t len, cur_len;
1831 struct cifs_sb_info *cifs_sb;
1832 struct cifsTconInfo *pTcon;
1833 struct cifsFileInfo *open_file;
1834 struct smb_com_read_rsp *pSMBr;
1840 len = iov_length(iov, nr_segs);
1845 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1847 open_file = file->private_data;
1848 pTcon = tlink_tcon(open_file->tlink);
1850 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1851 cFYI(1, "attempting read on write only file instance");
1853 for (total_read = 0; total_read < len; total_read += bytes_read) {
1854 cur_len = min_t(const size_t, len - total_read, cifs_sb->rsize);
1858 while (rc == -EAGAIN) {
1859 int buf_type = CIFS_NO_BUFFER;
1860 if (open_file->invalidHandle) {
1861 rc = cifs_reopen_file(open_file, true);
1865 rc = CIFSSMBRead(xid, pTcon, open_file->netfid,
1866 cur_len, *poffset, &bytes_read,
1867 &read_data, &buf_type);
1868 pSMBr = (struct smb_com_read_rsp *)read_data;
1870 char *data_offset = read_data + 4 +
1871 le16_to_cpu(pSMBr->DataOffset);
1872 if (memcpy_toiovecend(iov, data_offset,
1873 iov_offset, bytes_read))
1875 if (buf_type == CIFS_SMALL_BUFFER)
1876 cifs_small_buf_release(read_data);
1877 else if (buf_type == CIFS_LARGE_BUFFER)
1878 cifs_buf_release(read_data);
1880 iov_offset += bytes_read;
1884 if (rc || (bytes_read == 0)) {
1892 cifs_stats_bytes_read(pTcon, bytes_read);
1893 *poffset += bytes_read;
1901 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1902 size_t read_size, loff_t *poffset)
1905 iov.iov_base = read_data;
1906 iov.iov_len = read_size;
1908 return cifs_iovec_read(file, &iov, 1, poffset);
1911 static ssize_t cifs_user_readv(struct kiocb *iocb, const struct iovec *iov,
1912 unsigned long nr_segs, loff_t pos)
1916 read = cifs_iovec_read(iocb->ki_filp, iov, nr_segs, &pos);
1923 ssize_t cifs_strict_readv(struct kiocb *iocb, const struct iovec *iov,
1924 unsigned long nr_segs, loff_t pos)
1926 struct inode *inode;
1928 inode = iocb->ki_filp->f_path.dentry->d_inode;
1930 if (CIFS_I(inode)->clientCanCacheRead)
1931 return generic_file_aio_read(iocb, iov, nr_segs, pos);
1934 * In strict cache mode we need to read from the server all the time
1935 * if we don't have level II oplock because the server can delay mtime
1936 * change - so we can't make a decision about inode invalidating.
1937 * And we can also fail with pagereading if there are mandatory locks
1938 * on pages affected by this read but not on the region from pos to
1942 return cifs_user_readv(iocb, iov, nr_segs, pos);
1945 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1949 unsigned int bytes_read = 0;
1950 unsigned int total_read;
1951 unsigned int current_read_size;
1952 struct cifs_sb_info *cifs_sb;
1953 struct cifsTconInfo *pTcon;
1955 char *current_offset;
1956 struct cifsFileInfo *open_file;
1957 int buf_type = CIFS_NO_BUFFER;
1960 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1962 if (file->private_data == NULL) {
1967 open_file = file->private_data;
1968 pTcon = tlink_tcon(open_file->tlink);
1970 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1971 cFYI(1, "attempting read on write only file instance");
1973 for (total_read = 0, current_offset = read_data;
1974 read_size > total_read;
1975 total_read += bytes_read, current_offset += bytes_read) {
1976 current_read_size = min_t(const int, read_size - total_read,
1978 /* For windows me and 9x we do not want to request more
1979 than it negotiated since it will refuse the read then */
1981 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1982 current_read_size = min_t(const int, current_read_size,
1983 pTcon->ses->server->maxBuf - 128);
1986 while (rc == -EAGAIN) {
1987 if (open_file->invalidHandle) {
1988 rc = cifs_reopen_file(open_file, true);
1992 rc = CIFSSMBRead(xid, pTcon,
1994 current_read_size, *poffset,
1995 &bytes_read, ¤t_offset,
1998 if (rc || (bytes_read == 0)) {
2006 cifs_stats_bytes_read(pTcon, total_read);
2007 *poffset += bytes_read;
2014 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
2017 struct inode *inode = file->f_path.dentry->d_inode;
2021 if (!CIFS_I(inode)->clientCanCacheRead)
2022 cifs_invalidate_mapping(inode);
2024 rc = generic_file_mmap(file, vma);
2029 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
2034 rc = cifs_revalidate_file(file);
2036 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
2040 rc = generic_file_mmap(file, vma);
2046 static void cifs_copy_cache_pages(struct address_space *mapping,
2047 struct list_head *pages, int bytes_read, char *data)
2052 while (bytes_read > 0) {
2053 if (list_empty(pages))
2056 page = list_entry(pages->prev, struct page, lru);
2057 list_del(&page->lru);
2059 if (add_to_page_cache_lru(page, mapping, page->index,
2061 page_cache_release(page);
2062 cFYI(1, "Add page cache failed");
2063 data += PAGE_CACHE_SIZE;
2064 bytes_read -= PAGE_CACHE_SIZE;
2067 page_cache_release(page);
2069 target = kmap_atomic(page, KM_USER0);
2071 if (PAGE_CACHE_SIZE > bytes_read) {
2072 memcpy(target, data, bytes_read);
2073 /* zero the tail end of this partial page */
2074 memset(target + bytes_read, 0,
2075 PAGE_CACHE_SIZE - bytes_read);
2078 memcpy(target, data, PAGE_CACHE_SIZE);
2079 bytes_read -= PAGE_CACHE_SIZE;
2081 kunmap_atomic(target, KM_USER0);
2083 flush_dcache_page(page);
2084 SetPageUptodate(page);
2086 data += PAGE_CACHE_SIZE;
2088 /* add page to FS-Cache */
2089 cifs_readpage_to_fscache(mapping->host, page);
2094 static int cifs_readpages(struct file *file, struct address_space *mapping,
2095 struct list_head *page_list, unsigned num_pages)
2101 struct cifs_sb_info *cifs_sb;
2102 struct cifsTconInfo *pTcon;
2103 unsigned int bytes_read = 0;
2104 unsigned int read_size, i;
2105 char *smb_read_data = NULL;
2106 struct smb_com_read_rsp *pSMBr;
2107 struct cifsFileInfo *open_file;
2108 int buf_type = CIFS_NO_BUFFER;
2111 if (file->private_data == NULL) {
2116 open_file = file->private_data;
2117 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2118 pTcon = tlink_tcon(open_file->tlink);
2121 * Reads as many pages as possible from fscache. Returns -ENOBUFS
2122 * immediately if the cookie is negative
2124 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
2129 cFYI(DBG2, "rpages: num pages %d", num_pages);
2130 for (i = 0; i < num_pages; ) {
2131 unsigned contig_pages;
2132 struct page *tmp_page;
2133 unsigned long expected_index;
2135 if (list_empty(page_list))
2138 page = list_entry(page_list->prev, struct page, lru);
2139 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2141 /* count adjacent pages that we will read into */
2144 list_entry(page_list->prev, struct page, lru)->index;
2145 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2146 if (tmp_page->index == expected_index) {
2152 if (contig_pages + i > num_pages)
2153 contig_pages = num_pages - i;
2155 /* for reads over a certain size could initiate async
2158 read_size = contig_pages * PAGE_CACHE_SIZE;
2159 /* Read size needs to be in multiples of one page */
2160 read_size = min_t(const unsigned int, read_size,
2161 cifs_sb->rsize & PAGE_CACHE_MASK);
2162 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
2163 read_size, contig_pages);
2165 while (rc == -EAGAIN) {
2166 if (open_file->invalidHandle) {
2167 rc = cifs_reopen_file(open_file, true);
2172 rc = CIFSSMBRead(xid, pTcon,
2175 &bytes_read, &smb_read_data,
2177 /* BB more RC checks ? */
2178 if (rc == -EAGAIN) {
2179 if (smb_read_data) {
2180 if (buf_type == CIFS_SMALL_BUFFER)
2181 cifs_small_buf_release(smb_read_data);
2182 else if (buf_type == CIFS_LARGE_BUFFER)
2183 cifs_buf_release(smb_read_data);
2184 smb_read_data = NULL;
2188 if ((rc < 0) || (smb_read_data == NULL)) {
2189 cFYI(1, "Read error in readpages: %d", rc);
2191 } else if (bytes_read > 0) {
2192 task_io_account_read(bytes_read);
2193 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2194 cifs_copy_cache_pages(mapping, page_list, bytes_read,
2195 smb_read_data + 4 /* RFC1001 hdr */ +
2196 le16_to_cpu(pSMBr->DataOffset));
2198 i += bytes_read >> PAGE_CACHE_SHIFT;
2199 cifs_stats_bytes_read(pTcon, bytes_read);
2200 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2201 i++; /* account for partial page */
2203 /* server copy of file can have smaller size
2205 /* BB do we need to verify this common case ?
2206 this case is ok - if we are at server EOF
2207 we will hit it on next read */
2212 cFYI(1, "No bytes read (%d) at offset %lld . "
2213 "Cleaning remaining pages from readahead list",
2214 bytes_read, offset);
2215 /* BB turn off caching and do new lookup on
2216 file size at server? */
2219 if (smb_read_data) {
2220 if (buf_type == CIFS_SMALL_BUFFER)
2221 cifs_small_buf_release(smb_read_data);
2222 else if (buf_type == CIFS_LARGE_BUFFER)
2223 cifs_buf_release(smb_read_data);
2224 smb_read_data = NULL;
2229 /* need to free smb_read_data buf before exit */
2230 if (smb_read_data) {
2231 if (buf_type == CIFS_SMALL_BUFFER)
2232 cifs_small_buf_release(smb_read_data);
2233 else if (buf_type == CIFS_LARGE_BUFFER)
2234 cifs_buf_release(smb_read_data);
2235 smb_read_data = NULL;
2243 static int cifs_readpage_worker(struct file *file, struct page *page,
2249 /* Is the page cached? */
2250 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2254 page_cache_get(page);
2255 read_data = kmap(page);
2256 /* for reads over a certain size could initiate async read ahead */
2258 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2263 cFYI(1, "Bytes read %d", rc);
2265 file->f_path.dentry->d_inode->i_atime =
2266 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2268 if (PAGE_CACHE_SIZE > rc)
2269 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2271 flush_dcache_page(page);
2272 SetPageUptodate(page);
2274 /* send this page to the cache */
2275 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2281 page_cache_release(page);
2287 static int cifs_readpage(struct file *file, struct page *page)
2289 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2295 if (file->private_data == NULL) {
2301 cFYI(1, "readpage %p at offset %d 0x%x\n",
2302 page, (int)offset, (int)offset);
2304 rc = cifs_readpage_worker(file, page, &offset);
2312 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2314 struct cifsFileInfo *open_file;
2316 spin_lock(&cifs_file_list_lock);
2317 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2318 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2319 spin_unlock(&cifs_file_list_lock);
2323 spin_unlock(&cifs_file_list_lock);
2327 /* We do not want to update the file size from server for inodes
2328 open for write - to avoid races with writepage extending
2329 the file - in the future we could consider allowing
2330 refreshing the inode only on increases in the file size
2331 but this is tricky to do without racing with writebehind
2332 page caching in the current Linux kernel design */
2333 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2338 if (is_inode_writable(cifsInode)) {
2339 /* This inode is open for write at least once */
2340 struct cifs_sb_info *cifs_sb;
2342 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2343 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2344 /* since no page cache to corrupt on directio
2345 we can change size safely */
2349 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2357 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2358 loff_t pos, unsigned len, unsigned flags,
2359 struct page **pagep, void **fsdata)
2361 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2362 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2363 loff_t page_start = pos & PAGE_MASK;
2368 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2370 page = grab_cache_page_write_begin(mapping, index, flags);
2376 if (PageUptodate(page))
2380 * If we write a full page it will be up to date, no need to read from
2381 * the server. If the write is short, we'll end up doing a sync write
2384 if (len == PAGE_CACHE_SIZE)
2388 * optimize away the read when we have an oplock, and we're not
2389 * expecting to use any of the data we'd be reading in. That
2390 * is, when the page lies beyond the EOF, or straddles the EOF
2391 * and the write will cover all of the existing data.
2393 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2394 i_size = i_size_read(mapping->host);
2395 if (page_start >= i_size ||
2396 (offset == 0 && (pos + len) >= i_size)) {
2397 zero_user_segments(page, 0, offset,
2401 * PageChecked means that the parts of the page
2402 * to which we're not writing are considered up
2403 * to date. Once the data is copied to the
2404 * page, it can be set uptodate.
2406 SetPageChecked(page);
2411 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2413 * might as well read a page, it is fast enough. If we get
2414 * an error, we don't need to return it. cifs_write_end will
2415 * do a sync write instead since PG_uptodate isn't set.
2417 cifs_readpage_worker(file, page, &page_start);
2419 /* we could try using another file handle if there is one -
2420 but how would we lock it to prevent close of that handle
2421 racing with this read? In any case
2422 this will be written out by write_end so is fine */
2429 static int cifs_release_page(struct page *page, gfp_t gfp)
2431 if (PagePrivate(page))
2434 return cifs_fscache_release_page(page, gfp);
2437 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2439 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2442 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2445 void cifs_oplock_break(struct work_struct *work)
2447 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2449 struct inode *inode = cfile->dentry->d_inode;
2450 struct cifsInodeInfo *cinode = CIFS_I(inode);
2453 if (inode && S_ISREG(inode->i_mode)) {
2454 if (cinode->clientCanCacheRead)
2455 break_lease(inode, O_RDONLY);
2457 break_lease(inode, O_WRONLY);
2458 rc = filemap_fdatawrite(inode->i_mapping);
2459 if (cinode->clientCanCacheRead == 0) {
2460 rc = filemap_fdatawait(inode->i_mapping);
2461 mapping_set_error(inode->i_mapping, rc);
2462 invalidate_remote_inode(inode);
2464 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2468 * releasing stale oplock after recent reconnect of smb session using
2469 * a now incorrect file handle is not a data integrity issue but do
2470 * not bother sending an oplock release if session to server still is
2471 * disconnected since oplock already released by the server
2473 if (!cfile->oplock_break_cancelled) {
2474 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2475 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false,
2476 cinode->clientCanCacheRead ? 1 : 0);
2477 cFYI(1, "Oplock release rc = %d", rc);
2481 * We might have kicked in before is_valid_oplock_break()
2482 * finished grabbing reference for us. Make sure it's done by
2483 * waiting for cifs_file_list_lock.
2485 spin_lock(&cifs_file_list_lock);
2486 spin_unlock(&cifs_file_list_lock);
2488 cifs_oplock_break_put(cfile);
2491 /* must be called while holding cifs_file_list_lock */
2492 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2494 cifs_sb_active(cfile->dentry->d_sb);
2495 cifsFileInfo_get(cfile);
2498 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2500 struct super_block *sb = cfile->dentry->d_sb;
2502 cifsFileInfo_put(cfile);
2503 cifs_sb_deactive(sb);
2506 const struct address_space_operations cifs_addr_ops = {
2507 .readpage = cifs_readpage,
2508 .readpages = cifs_readpages,
2509 .writepage = cifs_writepage,
2510 .writepages = cifs_writepages,
2511 .write_begin = cifs_write_begin,
2512 .write_end = cifs_write_end,
2513 .set_page_dirty = __set_page_dirty_nobuffers,
2514 .releasepage = cifs_release_page,
2515 .invalidatepage = cifs_invalidate_page,
2516 /* .sync_page = cifs_sync_page, */
2521 * cifs_readpages requires the server to support a buffer large enough to
2522 * contain the header plus one complete page of data. Otherwise, we need
2523 * to leave cifs_readpages out of the address space operations.
2525 const struct address_space_operations cifs_addr_ops_smallbuf = {
2526 .readpage = cifs_readpage,
2527 .writepage = cifs_writepage,
2528 .writepages = cifs_writepages,
2529 .write_begin = cifs_write_begin,
2530 .write_end = cifs_write_end,
2531 .set_page_dirty = __set_page_dirty_nobuffers,
2532 .releasepage = cifs_release_page,
2533 .invalidatepage = cifs_invalidate_page,
2534 /* .sync_page = cifs_sync_page, */