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 CIFSSMBUnixSetFileInfo(xid, tcon, &args, netfid,
434 cifs_put_tlink(tlink);
438 /* Try to reacquire byte range locks that were released when session */
439 /* to server was lost */
440 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
444 /* BB list all locks open on this file and relock */
449 static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
454 struct cifs_sb_info *cifs_sb;
455 struct cifsTconInfo *tcon;
456 struct cifsInodeInfo *pCifsInode;
458 char *full_path = NULL;
460 int disposition = FILE_OPEN;
464 mutex_lock(&pCifsFile->fh_mutex);
465 if (!pCifsFile->invalidHandle) {
466 mutex_unlock(&pCifsFile->fh_mutex);
472 inode = pCifsFile->dentry->d_inode;
473 cifs_sb = CIFS_SB(inode->i_sb);
474 tcon = tlink_tcon(pCifsFile->tlink);
476 /* can not grab rename sem here because various ops, including
477 those that already have the rename sem can end up causing writepage
478 to get called and if the server was down that means we end up here,
479 and we can never tell if the caller already has the rename_sem */
480 full_path = build_path_from_dentry(pCifsFile->dentry);
481 if (full_path == NULL) {
483 mutex_unlock(&pCifsFile->fh_mutex);
488 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
489 inode, pCifsFile->f_flags, full_path);
496 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
497 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
498 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
501 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
502 * original open. Must mask them off for a reopen.
504 unsigned int oflags = pCifsFile->f_flags &
505 ~(O_CREAT | O_EXCL | O_TRUNC);
507 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
508 cifs_sb->mnt_file_mode /* ignored */,
509 oflags, &oplock, &netfid, xid);
511 cFYI(1, "posix reopen succeeded");
514 /* fallthrough to retry open the old way on errors, especially
515 in the reconnect path it is important to retry hard */
518 desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
520 /* Can not refresh inode by passing in file_info buf to be returned
521 by SMBOpen and then calling get_inode_info with returned buf
522 since file might have write behind data that needs to be flushed
523 and server version of file size can be stale. If we knew for sure
524 that inode was not dirty locally we could do this */
526 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
527 CREATE_NOT_DIR, &netfid, &oplock, NULL,
528 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
529 CIFS_MOUNT_MAP_SPECIAL_CHR);
531 mutex_unlock(&pCifsFile->fh_mutex);
532 cFYI(1, "cifs_open returned 0x%x", rc);
533 cFYI(1, "oplock: %d", oplock);
534 goto reopen_error_exit;
538 pCifsFile->netfid = netfid;
539 pCifsFile->invalidHandle = false;
540 mutex_unlock(&pCifsFile->fh_mutex);
541 pCifsInode = CIFS_I(inode);
544 rc = filemap_write_and_wait(inode->i_mapping);
545 mapping_set_error(inode->i_mapping, rc);
548 rc = cifs_get_inode_info_unix(&inode,
549 full_path, inode->i_sb, xid);
551 rc = cifs_get_inode_info(&inode,
552 full_path, NULL, inode->i_sb,
554 } /* else we are writing out data to server already
555 and could deadlock if we tried to flush data, and
556 since we do not know if we have data that would
557 invalidate the current end of file on the server
558 we can not go to the server to get the new inod
561 cifs_set_oplock_level(pCifsInode, oplock);
563 cifs_relock_file(pCifsFile);
571 int cifs_close(struct inode *inode, struct file *file)
573 cifsFileInfo_put(file->private_data);
574 file->private_data = NULL;
576 /* return code from the ->release op is always ignored */
580 int cifs_closedir(struct inode *inode, struct file *file)
584 struct cifsFileInfo *pCFileStruct = file->private_data;
587 cFYI(1, "Closedir inode = 0x%p", inode);
592 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
594 cFYI(1, "Freeing private data in close dir");
595 spin_lock(&cifs_file_list_lock);
596 if (!pCFileStruct->srch_inf.endOfSearch &&
597 !pCFileStruct->invalidHandle) {
598 pCFileStruct->invalidHandle = true;
599 spin_unlock(&cifs_file_list_lock);
600 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
601 cFYI(1, "Closing uncompleted readdir with rc %d",
603 /* not much we can do if it fails anyway, ignore rc */
606 spin_unlock(&cifs_file_list_lock);
607 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
609 cFYI(1, "closedir free smb buf in srch struct");
610 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
611 if (pCFileStruct->srch_inf.smallBuf)
612 cifs_small_buf_release(ptmp);
614 cifs_buf_release(ptmp);
616 cifs_put_tlink(pCFileStruct->tlink);
617 kfree(file->private_data);
618 file->private_data = NULL;
620 /* BB can we lock the filestruct while this is going on? */
625 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
626 __u64 offset, __u8 lockType)
628 struct cifsLockInfo *li =
629 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
635 mutex_lock(&fid->lock_mutex);
636 list_add(&li->llist, &fid->llist);
637 mutex_unlock(&fid->lock_mutex);
641 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
647 bool wait_flag = false;
648 struct cifs_sb_info *cifs_sb;
649 struct cifsTconInfo *tcon;
651 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
652 bool posix_locking = 0;
654 length = 1 + pfLock->fl_end - pfLock->fl_start;
658 cFYI(1, "Lock parm: 0x%x flockflags: "
659 "0x%x flocktype: 0x%x start: %lld end: %lld",
660 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
663 if (pfLock->fl_flags & FL_POSIX)
665 if (pfLock->fl_flags & FL_FLOCK)
667 if (pfLock->fl_flags & FL_SLEEP) {
668 cFYI(1, "Blocking lock");
671 if (pfLock->fl_flags & FL_ACCESS)
672 cFYI(1, "Process suspended by mandatory locking - "
673 "not implemented yet");
674 if (pfLock->fl_flags & FL_LEASE)
675 cFYI(1, "Lease on file - not implemented yet");
676 if (pfLock->fl_flags &
677 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
678 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
680 if (pfLock->fl_type == F_WRLCK) {
683 } else if (pfLock->fl_type == F_UNLCK) {
686 /* Check if unlock includes more than
688 } else if (pfLock->fl_type == F_RDLCK) {
690 lockType |= LOCKING_ANDX_SHARED_LOCK;
692 } else if (pfLock->fl_type == F_EXLCK) {
695 } else if (pfLock->fl_type == F_SHLCK) {
697 lockType |= LOCKING_ANDX_SHARED_LOCK;
700 cFYI(1, "Unknown type of lock");
702 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
703 tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
704 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
706 if ((tcon->ses->capabilities & CAP_UNIX) &&
707 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
708 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
710 /* BB add code here to normalize offset and length to
711 account for negative length which we can not accept over the
716 if (lockType & LOCKING_ANDX_SHARED_LOCK)
717 posix_lock_type = CIFS_RDLCK;
719 posix_lock_type = CIFS_WRLCK;
720 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
722 posix_lock_type, wait_flag);
727 /* BB we could chain these into one lock request BB */
728 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
729 0, 1, lockType, 0 /* wait flag */, 0);
731 rc = CIFSSMBLock(xid, tcon, netfid, length,
732 pfLock->fl_start, 1 /* numUnlock */ ,
733 0 /* numLock */ , lockType,
734 0 /* wait flag */, 0);
735 pfLock->fl_type = F_UNLCK;
737 cERROR(1, "Error unlocking previously locked "
738 "range %d during test of lock", rc);
742 /* if rc == ERR_SHARING_VIOLATION ? */
745 if (lockType & LOCKING_ANDX_SHARED_LOCK) {
746 pfLock->fl_type = F_WRLCK;
748 rc = CIFSSMBLock(xid, tcon, netfid, length,
749 pfLock->fl_start, 0, 1,
750 lockType | LOCKING_ANDX_SHARED_LOCK,
751 0 /* wait flag */, 0);
753 rc = CIFSSMBLock(xid, tcon, netfid,
754 length, pfLock->fl_start, 1, 0,
756 LOCKING_ANDX_SHARED_LOCK,
757 0 /* wait flag */, 0);
758 pfLock->fl_type = F_RDLCK;
760 cERROR(1, "Error unlocking "
761 "previously locked range %d "
762 "during test of lock", rc);
765 pfLock->fl_type = F_WRLCK;
775 if (!numLock && !numUnlock) {
776 /* if no lock or unlock then nothing
777 to do since we do not know what it is */
784 if (lockType & LOCKING_ANDX_SHARED_LOCK)
785 posix_lock_type = CIFS_RDLCK;
787 posix_lock_type = CIFS_WRLCK;
790 posix_lock_type = CIFS_UNLCK;
792 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
794 posix_lock_type, wait_flag);
796 struct cifsFileInfo *fid = file->private_data;
799 rc = CIFSSMBLock(xid, tcon, netfid, length,
800 pfLock->fl_start, 0, numLock, lockType,
804 /* For Windows locks we must store them. */
805 rc = store_file_lock(fid, length,
806 pfLock->fl_start, lockType);
808 } else if (numUnlock) {
809 /* For each stored lock that this unlock overlaps
810 completely, unlock it. */
812 struct cifsLockInfo *li, *tmp;
815 mutex_lock(&fid->lock_mutex);
816 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
817 if (pfLock->fl_start <= li->offset &&
818 (pfLock->fl_start + length) >=
819 (li->offset + li->length)) {
820 stored_rc = CIFSSMBLock(xid, tcon,
827 list_del(&li->llist);
832 mutex_unlock(&fid->lock_mutex);
836 if (pfLock->fl_flags & FL_POSIX)
837 posix_lock_file_wait(file, pfLock);
842 /* update the file size (if needed) after a write */
844 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
845 unsigned int bytes_written)
847 loff_t end_of_write = offset + bytes_written;
849 if (end_of_write > cifsi->server_eof)
850 cifsi->server_eof = end_of_write;
853 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
854 size_t write_size, loff_t *poffset)
856 struct inode *inode = file->f_path.dentry->d_inode;
858 unsigned int bytes_written = 0;
859 unsigned int total_written;
860 struct cifs_sb_info *cifs_sb;
861 struct cifsTconInfo *pTcon;
863 struct cifsFileInfo *open_file;
864 struct cifsInodeInfo *cifsi = CIFS_I(inode);
866 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
868 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
869 *poffset, file->f_path.dentry->d_name.name); */
871 if (file->private_data == NULL)
874 open_file = file->private_data;
875 pTcon = tlink_tcon(open_file->tlink);
877 rc = generic_write_checks(file, poffset, &write_size, 0);
883 for (total_written = 0; write_size > total_written;
884 total_written += bytes_written) {
886 while (rc == -EAGAIN) {
887 if (file->private_data == NULL) {
888 /* file has been closed on us */
890 /* if we have gotten here we have written some data
891 and blocked, and the file has been freed on us while
892 we blocked so return what we managed to write */
893 return total_written;
895 if (open_file->invalidHandle) {
896 /* we could deadlock if we called
897 filemap_fdatawait from here so tell
898 reopen_file not to flush data to server
900 rc = cifs_reopen_file(open_file, false);
905 rc = CIFSSMBWrite(xid, pTcon,
907 min_t(const int, cifs_sb->wsize,
908 write_size - total_written),
909 *poffset, &bytes_written,
910 NULL, write_data + total_written, 0);
912 if (rc || (bytes_written == 0)) {
920 cifs_update_eof(cifsi, *poffset, bytes_written);
921 *poffset += bytes_written;
925 cifs_stats_bytes_written(pTcon, total_written);
927 /* Do not update local mtime - server will set its actual value on write
928 * inode->i_ctime = inode->i_mtime =
929 * current_fs_time(inode->i_sb);*/
930 if (total_written > 0) {
931 spin_lock(&inode->i_lock);
932 if (*poffset > inode->i_size)
933 i_size_write(inode, *poffset);
934 spin_unlock(&inode->i_lock);
936 mark_inode_dirty_sync(inode);
939 return total_written;
942 static ssize_t cifs_write(struct cifsFileInfo *open_file,
943 const char *write_data, size_t write_size,
947 unsigned int bytes_written = 0;
948 unsigned int total_written;
949 struct cifs_sb_info *cifs_sb;
950 struct cifsTconInfo *pTcon;
952 struct dentry *dentry = open_file->dentry;
953 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
955 cifs_sb = CIFS_SB(dentry->d_sb);
957 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
958 *poffset, dentry->d_name.name);
960 pTcon = tlink_tcon(open_file->tlink);
964 for (total_written = 0; write_size > total_written;
965 total_written += bytes_written) {
967 while (rc == -EAGAIN) {
968 if (open_file->invalidHandle) {
969 /* we could deadlock if we called
970 filemap_fdatawait from here so tell
971 reopen_file not to flush data to
973 rc = cifs_reopen_file(open_file, false);
977 if (experimEnabled || (pTcon->ses->server &&
978 ((pTcon->ses->server->secMode &
979 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
984 len = min((size_t)cifs_sb->wsize,
985 write_size - total_written);
986 /* iov[0] is reserved for smb header */
987 iov[1].iov_base = (char *)write_data +
989 iov[1].iov_len = len;
990 rc = CIFSSMBWrite2(xid, pTcon,
991 open_file->netfid, len,
992 *poffset, &bytes_written,
995 rc = CIFSSMBWrite(xid, pTcon,
997 min_t(const int, cifs_sb->wsize,
998 write_size - total_written),
999 *poffset, &bytes_written,
1000 write_data + total_written,
1003 if (rc || (bytes_written == 0)) {
1011 cifs_update_eof(cifsi, *poffset, bytes_written);
1012 *poffset += bytes_written;
1016 cifs_stats_bytes_written(pTcon, total_written);
1018 if (total_written > 0) {
1019 spin_lock(&dentry->d_inode->i_lock);
1020 if (*poffset > dentry->d_inode->i_size)
1021 i_size_write(dentry->d_inode, *poffset);
1022 spin_unlock(&dentry->d_inode->i_lock);
1024 mark_inode_dirty_sync(dentry->d_inode);
1026 return total_written;
1029 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1032 struct cifsFileInfo *open_file = NULL;
1033 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1035 /* only filter by fsuid on multiuser mounts */
1036 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1039 spin_lock(&cifs_file_list_lock);
1040 /* we could simply get the first_list_entry since write-only entries
1041 are always at the end of the list but since the first entry might
1042 have a close pending, we go through the whole list */
1043 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1044 if (fsuid_only && open_file->uid != current_fsuid())
1046 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1047 if (!open_file->invalidHandle) {
1048 /* found a good file */
1049 /* lock it so it will not be closed on us */
1050 cifsFileInfo_get(open_file);
1051 spin_unlock(&cifs_file_list_lock);
1053 } /* else might as well continue, and look for
1054 another, or simply have the caller reopen it
1055 again rather than trying to fix this handle */
1056 } else /* write only file */
1057 break; /* write only files are last so must be done */
1059 spin_unlock(&cifs_file_list_lock);
1063 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1066 struct cifsFileInfo *open_file;
1067 struct cifs_sb_info *cifs_sb;
1068 bool any_available = false;
1071 /* Having a null inode here (because mapping->host was set to zero by
1072 the VFS or MM) should not happen but we had reports of on oops (due to
1073 it being zero) during stress testcases so we need to check for it */
1075 if (cifs_inode == NULL) {
1076 cERROR(1, "Null inode passed to cifs_writeable_file");
1081 cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1083 /* only filter by fsuid on multiuser mounts */
1084 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1087 spin_lock(&cifs_file_list_lock);
1089 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1090 if (!any_available && open_file->pid != current->tgid)
1092 if (fsuid_only && open_file->uid != current_fsuid())
1094 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1095 cifsFileInfo_get(open_file);
1097 if (!open_file->invalidHandle) {
1098 /* found a good writable file */
1099 spin_unlock(&cifs_file_list_lock);
1103 spin_unlock(&cifs_file_list_lock);
1105 /* Had to unlock since following call can block */
1106 rc = cifs_reopen_file(open_file, false);
1110 /* if it fails, try another handle if possible */
1111 cFYI(1, "wp failed on reopen file");
1112 cifsFileInfo_put(open_file);
1114 spin_lock(&cifs_file_list_lock);
1116 /* else we simply continue to the next entry. Thus
1117 we do not loop on reopen errors. If we
1118 can not reopen the file, for example if we
1119 reconnected to a server with another client
1120 racing to delete or lock the file we would not
1121 make progress if we restarted before the beginning
1122 of the loop here. */
1125 /* couldn't find useable FH with same pid, try any available */
1126 if (!any_available) {
1127 any_available = true;
1128 goto refind_writable;
1130 spin_unlock(&cifs_file_list_lock);
1134 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1136 struct address_space *mapping = page->mapping;
1137 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1140 int bytes_written = 0;
1141 struct cifs_sb_info *cifs_sb;
1142 struct inode *inode;
1143 struct cifsFileInfo *open_file;
1145 if (!mapping || !mapping->host)
1148 inode = page->mapping->host;
1149 cifs_sb = CIFS_SB(inode->i_sb);
1151 offset += (loff_t)from;
1152 write_data = kmap(page);
1155 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1160 /* racing with truncate? */
1161 if (offset > mapping->host->i_size) {
1163 return 0; /* don't care */
1166 /* check to make sure that we are not extending the file */
1167 if (mapping->host->i_size - offset < (loff_t)to)
1168 to = (unsigned)(mapping->host->i_size - offset);
1170 open_file = find_writable_file(CIFS_I(mapping->host), false);
1172 bytes_written = cifs_write(open_file, write_data,
1173 to - from, &offset);
1174 cifsFileInfo_put(open_file);
1175 /* Does mm or vfs already set times? */
1176 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1177 if ((bytes_written > 0) && (offset))
1179 else if (bytes_written < 0)
1182 cFYI(1, "No writeable filehandles for inode");
1190 static int cifs_writepages(struct address_space *mapping,
1191 struct writeback_control *wbc)
1193 unsigned int bytes_to_write;
1194 unsigned int bytes_written;
1195 struct cifs_sb_info *cifs_sb;
1199 int range_whole = 0;
1206 struct cifsFileInfo *open_file;
1207 struct cifsTconInfo *tcon;
1208 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1210 struct pagevec pvec;
1215 cifs_sb = CIFS_SB(mapping->host->i_sb);
1218 * If wsize is smaller that the page cache size, default to writing
1219 * one page at a time via cifs_writepage
1221 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1222 return generic_writepages(mapping, wbc);
1224 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1226 return generic_writepages(mapping, wbc);
1229 * if there's no open file, then this is likely to fail too,
1230 * but it'll at least handle the return. Maybe it should be
1233 open_file = find_writable_file(CIFS_I(mapping->host), false);
1236 return generic_writepages(mapping, wbc);
1239 tcon = tlink_tcon(open_file->tlink);
1240 if (!experimEnabled && tcon->ses->server->secMode &
1241 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1242 cifsFileInfo_put(open_file);
1244 return generic_writepages(mapping, wbc);
1246 cifsFileInfo_put(open_file);
1250 pagevec_init(&pvec, 0);
1251 if (wbc->range_cyclic) {
1252 index = mapping->writeback_index; /* Start from prev offset */
1255 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1256 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1257 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1262 while (!done && (index <= end) &&
1263 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1264 PAGECACHE_TAG_DIRTY,
1265 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1274 for (i = 0; i < nr_pages; i++) {
1275 page = pvec.pages[i];
1277 * At this point we hold neither mapping->tree_lock nor
1278 * lock on the page itself: the page may be truncated or
1279 * invalidated (changing page->mapping to NULL), or even
1280 * swizzled back from swapper_space to tmpfs file
1286 else if (!trylock_page(page))
1289 if (unlikely(page->mapping != mapping)) {
1294 if (!wbc->range_cyclic && page->index > end) {
1300 if (next && (page->index != next)) {
1301 /* Not next consecutive page */
1306 if (wbc->sync_mode != WB_SYNC_NONE)
1307 wait_on_page_writeback(page);
1309 if (PageWriteback(page) ||
1310 !clear_page_dirty_for_io(page)) {
1316 * This actually clears the dirty bit in the radix tree.
1317 * See cifs_writepage() for more commentary.
1319 set_page_writeback(page);
1321 if (page_offset(page) >= mapping->host->i_size) {
1324 end_page_writeback(page);
1329 * BB can we get rid of this? pages are held by pvec
1331 page_cache_get(page);
1333 len = min(mapping->host->i_size - page_offset(page),
1334 (loff_t)PAGE_CACHE_SIZE);
1336 /* reserve iov[0] for the smb header */
1338 iov[n_iov].iov_base = kmap(page);
1339 iov[n_iov].iov_len = len;
1340 bytes_to_write += len;
1344 offset = page_offset(page);
1346 next = page->index + 1;
1347 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1352 open_file = find_writable_file(CIFS_I(mapping->host),
1355 cERROR(1, "No writable handles for inode");
1358 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1359 bytes_to_write, offset,
1360 &bytes_written, iov, n_iov,
1362 cifsFileInfo_put(open_file);
1365 cFYI(1, "Write2 rc=%d, wrote=%u", rc, bytes_written);
1368 * For now, treat a short write as if nothing got
1369 * written. A zero length write however indicates
1370 * ENOSPC or EFBIG. We have no way to know which
1371 * though, so call it ENOSPC for now. EFBIG would
1372 * get translated to AS_EIO anyway.
1374 * FIXME: make it take into account the data that did
1378 if (bytes_written == 0)
1380 else if (bytes_written < bytes_to_write)
1384 /* retry on data-integrity flush */
1385 if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN)
1388 /* fix the stats and EOF */
1389 if (bytes_written > 0) {
1390 cifs_stats_bytes_written(tcon, bytes_written);
1391 cifs_update_eof(cifsi, offset, bytes_written);
1394 for (i = 0; i < n_iov; i++) {
1395 page = pvec.pages[first + i];
1396 /* on retryable write error, redirty page */
1398 redirty_page_for_writepage(wbc, page);
1403 end_page_writeback(page);
1404 page_cache_release(page);
1408 mapping_set_error(mapping, rc);
1412 if ((wbc->nr_to_write -= n_iov) <= 0)
1416 /* Need to re-find the pages we skipped */
1417 index = pvec.pages[0]->index + 1;
1419 pagevec_release(&pvec);
1421 if (!scanned && !done) {
1423 * We hit the last page and there is more work to be done: wrap
1424 * back to the start of the file
1430 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1431 mapping->writeback_index = index;
1438 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1444 /* BB add check for wbc flags */
1445 page_cache_get(page);
1446 if (!PageUptodate(page))
1447 cFYI(1, "ppw - page not up to date");
1450 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1452 * A writepage() implementation always needs to do either this,
1453 * or re-dirty the page with "redirty_page_for_writepage()" in
1454 * the case of a failure.
1456 * Just unlocking the page will cause the radix tree tag-bits
1457 * to fail to update with the state of the page correctly.
1459 set_page_writeback(page);
1460 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1461 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1463 end_page_writeback(page);
1464 page_cache_release(page);
1469 static int cifs_write_end(struct file *file, struct address_space *mapping,
1470 loff_t pos, unsigned len, unsigned copied,
1471 struct page *page, void *fsdata)
1474 struct inode *inode = mapping->host;
1476 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1479 if (PageChecked(page)) {
1481 SetPageUptodate(page);
1482 ClearPageChecked(page);
1483 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1484 SetPageUptodate(page);
1486 if (!PageUptodate(page)) {
1488 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1492 /* this is probably better than directly calling
1493 partialpage_write since in this function the file handle is
1494 known which we might as well leverage */
1495 /* BB check if anything else missing out of ppw
1496 such as updating last write time */
1497 page_data = kmap(page);
1498 rc = cifs_write(file->private_data, page_data + offset,
1500 /* if (rc < 0) should we set writebehind rc? */
1507 set_page_dirty(page);
1511 spin_lock(&inode->i_lock);
1512 if (pos > inode->i_size)
1513 i_size_write(inode, pos);
1514 spin_unlock(&inode->i_lock);
1518 page_cache_release(page);
1523 int cifs_fsync(struct file *file, int datasync)
1527 struct cifsTconInfo *tcon;
1528 struct cifsFileInfo *smbfile = file->private_data;
1529 struct inode *inode = file->f_path.dentry->d_inode;
1533 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1534 file->f_path.dentry->d_name.name, datasync);
1536 rc = filemap_write_and_wait(inode->i_mapping);
1538 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
1540 tcon = tlink_tcon(smbfile->tlink);
1541 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1542 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1549 /* static void cifs_sync_page(struct page *page)
1551 struct address_space *mapping;
1552 struct inode *inode;
1553 unsigned long index = page->index;
1554 unsigned int rpages = 0;
1557 cFYI(1, "sync page %p", page);
1558 mapping = page->mapping;
1561 inode = mapping->host;
1565 /* fill in rpages then
1566 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1568 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1578 * As file closes, flush all cached write data for this inode checking
1579 * for write behind errors.
1581 int cifs_flush(struct file *file, fl_owner_t id)
1583 struct inode *inode = file->f_path.dentry->d_inode;
1586 if (file->f_mode & FMODE_WRITE)
1587 rc = filemap_write_and_wait(inode->i_mapping);
1589 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1594 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1595 size_t read_size, loff_t *poffset)
1598 unsigned int bytes_read = 0;
1599 unsigned int total_read = 0;
1600 unsigned int current_read_size;
1601 struct cifs_sb_info *cifs_sb;
1602 struct cifsTconInfo *pTcon;
1604 struct cifsFileInfo *open_file;
1605 char *smb_read_data;
1606 char __user *current_offset;
1607 struct smb_com_read_rsp *pSMBr;
1610 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1612 if (file->private_data == NULL) {
1617 open_file = file->private_data;
1618 pTcon = tlink_tcon(open_file->tlink);
1620 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1621 cFYI(1, "attempting read on write only file instance");
1623 for (total_read = 0, current_offset = read_data;
1624 read_size > total_read;
1625 total_read += bytes_read, current_offset += bytes_read) {
1626 current_read_size = min_t(const int, read_size - total_read,
1629 smb_read_data = NULL;
1630 while (rc == -EAGAIN) {
1631 int buf_type = CIFS_NO_BUFFER;
1632 if (open_file->invalidHandle) {
1633 rc = cifs_reopen_file(open_file, true);
1637 rc = CIFSSMBRead(xid, pTcon,
1639 current_read_size, *poffset,
1640 &bytes_read, &smb_read_data,
1642 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1643 if (smb_read_data) {
1644 if (copy_to_user(current_offset,
1646 4 /* RFC1001 length field */ +
1647 le16_to_cpu(pSMBr->DataOffset),
1651 if (buf_type == CIFS_SMALL_BUFFER)
1652 cifs_small_buf_release(smb_read_data);
1653 else if (buf_type == CIFS_LARGE_BUFFER)
1654 cifs_buf_release(smb_read_data);
1655 smb_read_data = NULL;
1658 if (rc || (bytes_read == 0)) {
1666 cifs_stats_bytes_read(pTcon, bytes_read);
1667 *poffset += bytes_read;
1675 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1679 unsigned int bytes_read = 0;
1680 unsigned int total_read;
1681 unsigned int current_read_size;
1682 struct cifs_sb_info *cifs_sb;
1683 struct cifsTconInfo *pTcon;
1685 char *current_offset;
1686 struct cifsFileInfo *open_file;
1687 int buf_type = CIFS_NO_BUFFER;
1690 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1692 if (file->private_data == NULL) {
1697 open_file = file->private_data;
1698 pTcon = tlink_tcon(open_file->tlink);
1700 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1701 cFYI(1, "attempting read on write only file instance");
1703 for (total_read = 0, current_offset = read_data;
1704 read_size > total_read;
1705 total_read += bytes_read, current_offset += bytes_read) {
1706 current_read_size = min_t(const int, read_size - total_read,
1708 /* For windows me and 9x we do not want to request more
1709 than it negotiated since it will refuse the read then */
1711 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1712 current_read_size = min_t(const int, current_read_size,
1713 pTcon->ses->server->maxBuf - 128);
1716 while (rc == -EAGAIN) {
1717 if (open_file->invalidHandle) {
1718 rc = cifs_reopen_file(open_file, true);
1722 rc = CIFSSMBRead(xid, pTcon,
1724 current_read_size, *poffset,
1725 &bytes_read, ¤t_offset,
1728 if (rc || (bytes_read == 0)) {
1736 cifs_stats_bytes_read(pTcon, total_read);
1737 *poffset += bytes_read;
1744 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1749 rc = cifs_revalidate_file(file);
1751 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1755 rc = generic_file_mmap(file, vma);
1761 static void cifs_copy_cache_pages(struct address_space *mapping,
1762 struct list_head *pages, int bytes_read, char *data)
1767 while (bytes_read > 0) {
1768 if (list_empty(pages))
1771 page = list_entry(pages->prev, struct page, lru);
1772 list_del(&page->lru);
1774 if (add_to_page_cache_lru(page, mapping, page->index,
1776 page_cache_release(page);
1777 cFYI(1, "Add page cache failed");
1778 data += PAGE_CACHE_SIZE;
1779 bytes_read -= PAGE_CACHE_SIZE;
1782 page_cache_release(page);
1784 target = kmap_atomic(page, KM_USER0);
1786 if (PAGE_CACHE_SIZE > bytes_read) {
1787 memcpy(target, data, bytes_read);
1788 /* zero the tail end of this partial page */
1789 memset(target + bytes_read, 0,
1790 PAGE_CACHE_SIZE - bytes_read);
1793 memcpy(target, data, PAGE_CACHE_SIZE);
1794 bytes_read -= PAGE_CACHE_SIZE;
1796 kunmap_atomic(target, KM_USER0);
1798 flush_dcache_page(page);
1799 SetPageUptodate(page);
1801 data += PAGE_CACHE_SIZE;
1803 /* add page to FS-Cache */
1804 cifs_readpage_to_fscache(mapping->host, page);
1809 static int cifs_readpages(struct file *file, struct address_space *mapping,
1810 struct list_head *page_list, unsigned num_pages)
1816 struct cifs_sb_info *cifs_sb;
1817 struct cifsTconInfo *pTcon;
1818 unsigned int bytes_read = 0;
1819 unsigned int read_size, i;
1820 char *smb_read_data = NULL;
1821 struct smb_com_read_rsp *pSMBr;
1822 struct cifsFileInfo *open_file;
1823 int buf_type = CIFS_NO_BUFFER;
1826 if (file->private_data == NULL) {
1831 open_file = file->private_data;
1832 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1833 pTcon = tlink_tcon(open_file->tlink);
1836 * Reads as many pages as possible from fscache. Returns -ENOBUFS
1837 * immediately if the cookie is negative
1839 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
1844 cFYI(DBG2, "rpages: num pages %d", num_pages);
1845 for (i = 0; i < num_pages; ) {
1846 unsigned contig_pages;
1847 struct page *tmp_page;
1848 unsigned long expected_index;
1850 if (list_empty(page_list))
1853 page = list_entry(page_list->prev, struct page, lru);
1854 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1856 /* count adjacent pages that we will read into */
1859 list_entry(page_list->prev, struct page, lru)->index;
1860 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1861 if (tmp_page->index == expected_index) {
1867 if (contig_pages + i > num_pages)
1868 contig_pages = num_pages - i;
1870 /* for reads over a certain size could initiate async
1873 read_size = contig_pages * PAGE_CACHE_SIZE;
1874 /* Read size needs to be in multiples of one page */
1875 read_size = min_t(const unsigned int, read_size,
1876 cifs_sb->rsize & PAGE_CACHE_MASK);
1877 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
1878 read_size, contig_pages);
1880 while (rc == -EAGAIN) {
1881 if (open_file->invalidHandle) {
1882 rc = cifs_reopen_file(open_file, true);
1887 rc = CIFSSMBRead(xid, pTcon,
1890 &bytes_read, &smb_read_data,
1892 /* BB more RC checks ? */
1893 if (rc == -EAGAIN) {
1894 if (smb_read_data) {
1895 if (buf_type == CIFS_SMALL_BUFFER)
1896 cifs_small_buf_release(smb_read_data);
1897 else if (buf_type == CIFS_LARGE_BUFFER)
1898 cifs_buf_release(smb_read_data);
1899 smb_read_data = NULL;
1903 if ((rc < 0) || (smb_read_data == NULL)) {
1904 cFYI(1, "Read error in readpages: %d", rc);
1906 } else if (bytes_read > 0) {
1907 task_io_account_read(bytes_read);
1908 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1909 cifs_copy_cache_pages(mapping, page_list, bytes_read,
1910 smb_read_data + 4 /* RFC1001 hdr */ +
1911 le16_to_cpu(pSMBr->DataOffset));
1913 i += bytes_read >> PAGE_CACHE_SHIFT;
1914 cifs_stats_bytes_read(pTcon, bytes_read);
1915 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1916 i++; /* account for partial page */
1918 /* server copy of file can have smaller size
1920 /* BB do we need to verify this common case ?
1921 this case is ok - if we are at server EOF
1922 we will hit it on next read */
1927 cFYI(1, "No bytes read (%d) at offset %lld . "
1928 "Cleaning remaining pages from readahead list",
1929 bytes_read, offset);
1930 /* BB turn off caching and do new lookup on
1931 file size at server? */
1934 if (smb_read_data) {
1935 if (buf_type == CIFS_SMALL_BUFFER)
1936 cifs_small_buf_release(smb_read_data);
1937 else if (buf_type == CIFS_LARGE_BUFFER)
1938 cifs_buf_release(smb_read_data);
1939 smb_read_data = NULL;
1944 /* need to free smb_read_data buf before exit */
1945 if (smb_read_data) {
1946 if (buf_type == CIFS_SMALL_BUFFER)
1947 cifs_small_buf_release(smb_read_data);
1948 else if (buf_type == CIFS_LARGE_BUFFER)
1949 cifs_buf_release(smb_read_data);
1950 smb_read_data = NULL;
1958 static int cifs_readpage_worker(struct file *file, struct page *page,
1964 /* Is the page cached? */
1965 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
1969 page_cache_get(page);
1970 read_data = kmap(page);
1971 /* for reads over a certain size could initiate async read ahead */
1973 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1978 cFYI(1, "Bytes read %d", rc);
1980 file->f_path.dentry->d_inode->i_atime =
1981 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1983 if (PAGE_CACHE_SIZE > rc)
1984 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1986 flush_dcache_page(page);
1987 SetPageUptodate(page);
1989 /* send this page to the cache */
1990 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
1996 page_cache_release(page);
2002 static int cifs_readpage(struct file *file, struct page *page)
2004 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2010 if (file->private_data == NULL) {
2016 cFYI(1, "readpage %p at offset %d 0x%x\n",
2017 page, (int)offset, (int)offset);
2019 rc = cifs_readpage_worker(file, page, &offset);
2027 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2029 struct cifsFileInfo *open_file;
2031 spin_lock(&cifs_file_list_lock);
2032 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2033 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2034 spin_unlock(&cifs_file_list_lock);
2038 spin_unlock(&cifs_file_list_lock);
2042 /* We do not want to update the file size from server for inodes
2043 open for write - to avoid races with writepage extending
2044 the file - in the future we could consider allowing
2045 refreshing the inode only on increases in the file size
2046 but this is tricky to do without racing with writebehind
2047 page caching in the current Linux kernel design */
2048 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2053 if (is_inode_writable(cifsInode)) {
2054 /* This inode is open for write at least once */
2055 struct cifs_sb_info *cifs_sb;
2057 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2058 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2059 /* since no page cache to corrupt on directio
2060 we can change size safely */
2064 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2072 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2073 loff_t pos, unsigned len, unsigned flags,
2074 struct page **pagep, void **fsdata)
2076 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2077 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2078 loff_t page_start = pos & PAGE_MASK;
2083 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2085 page = grab_cache_page_write_begin(mapping, index, flags);
2091 if (PageUptodate(page))
2095 * If we write a full page it will be up to date, no need to read from
2096 * the server. If the write is short, we'll end up doing a sync write
2099 if (len == PAGE_CACHE_SIZE)
2103 * optimize away the read when we have an oplock, and we're not
2104 * expecting to use any of the data we'd be reading in. That
2105 * is, when the page lies beyond the EOF, or straddles the EOF
2106 * and the write will cover all of the existing data.
2108 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2109 i_size = i_size_read(mapping->host);
2110 if (page_start >= i_size ||
2111 (offset == 0 && (pos + len) >= i_size)) {
2112 zero_user_segments(page, 0, offset,
2116 * PageChecked means that the parts of the page
2117 * to which we're not writing are considered up
2118 * to date. Once the data is copied to the
2119 * page, it can be set uptodate.
2121 SetPageChecked(page);
2126 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2128 * might as well read a page, it is fast enough. If we get
2129 * an error, we don't need to return it. cifs_write_end will
2130 * do a sync write instead since PG_uptodate isn't set.
2132 cifs_readpage_worker(file, page, &page_start);
2134 /* we could try using another file handle if there is one -
2135 but how would we lock it to prevent close of that handle
2136 racing with this read? In any case
2137 this will be written out by write_end so is fine */
2144 static int cifs_release_page(struct page *page, gfp_t gfp)
2146 if (PagePrivate(page))
2149 return cifs_fscache_release_page(page, gfp);
2152 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2154 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2157 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2160 void cifs_oplock_break(struct work_struct *work)
2162 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2164 struct inode *inode = cfile->dentry->d_inode;
2165 struct cifsInodeInfo *cinode = CIFS_I(inode);
2168 if (inode && S_ISREG(inode->i_mode)) {
2169 if (cinode->clientCanCacheRead)
2170 break_lease(inode, O_RDONLY);
2172 break_lease(inode, O_WRONLY);
2173 rc = filemap_fdatawrite(inode->i_mapping);
2174 if (cinode->clientCanCacheRead == 0) {
2175 rc = filemap_fdatawait(inode->i_mapping);
2176 mapping_set_error(inode->i_mapping, rc);
2177 invalidate_remote_inode(inode);
2179 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2183 * releasing stale oplock after recent reconnect of smb session using
2184 * a now incorrect file handle is not a data integrity issue but do
2185 * not bother sending an oplock release if session to server still is
2186 * disconnected since oplock already released by the server
2188 if (!cfile->oplock_break_cancelled) {
2189 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2190 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false,
2191 cinode->clientCanCacheRead ? 1 : 0);
2192 cFYI(1, "Oplock release rc = %d", rc);
2196 * We might have kicked in before is_valid_oplock_break()
2197 * finished grabbing reference for us. Make sure it's done by
2198 * waiting for cifs_file_list_lock.
2200 spin_lock(&cifs_file_list_lock);
2201 spin_unlock(&cifs_file_list_lock);
2203 cifs_oplock_break_put(cfile);
2206 /* must be called while holding cifs_file_list_lock */
2207 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2209 cifs_sb_active(cfile->dentry->d_sb);
2210 cifsFileInfo_get(cfile);
2213 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2215 struct super_block *sb = cfile->dentry->d_sb;
2217 cifsFileInfo_put(cfile);
2218 cifs_sb_deactive(sb);
2221 const struct address_space_operations cifs_addr_ops = {
2222 .readpage = cifs_readpage,
2223 .readpages = cifs_readpages,
2224 .writepage = cifs_writepage,
2225 .writepages = cifs_writepages,
2226 .write_begin = cifs_write_begin,
2227 .write_end = cifs_write_end,
2228 .set_page_dirty = __set_page_dirty_nobuffers,
2229 .releasepage = cifs_release_page,
2230 .invalidatepage = cifs_invalidate_page,
2231 /* .sync_page = cifs_sync_page, */
2236 * cifs_readpages requires the server to support a buffer large enough to
2237 * contain the header plus one complete page of data. Otherwise, we need
2238 * to leave cifs_readpages out of the address space operations.
2240 const struct address_space_operations cifs_addr_ops_smallbuf = {
2241 .readpage = cifs_readpage,
2242 .writepage = cifs_writepage,
2243 .writepages = cifs_writepages,
2244 .write_begin = cifs_write_begin,
2245 .write_end = cifs_write_end,
2246 .set_page_dirty = __set_page_dirty_nobuffers,
2247 .releasepage = cifs_release_page,
2248 .invalidatepage = cifs_invalidate_page,
2249 /* .sync_page = cifs_sync_page, */