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 static inline int cifs_open_inode_helper(struct inode *inode,
108 struct cifsTconInfo *pTcon, __u32 oplock, FILE_ALL_INFO *buf,
109 char *full_path, int xid)
111 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
112 struct timespec temp;
115 if (pCifsInode->clientCanCacheRead) {
116 /* we have the inode open somewhere else
117 no need to discard cache data */
118 goto client_can_cache;
121 /* BB need same check in cifs_create too? */
122 /* if not oplocked, invalidate inode pages if mtime or file
124 temp = cifs_NTtimeToUnix(buf->LastWriteTime);
125 if (timespec_equal(&inode->i_mtime, &temp) &&
127 (loff_t)le64_to_cpu(buf->EndOfFile))) {
128 cFYI(1, "inode unchanged on server");
130 if (inode->i_mapping) {
131 /* BB no need to lock inode until after invalidate
132 since namei code should already have it locked? */
133 rc = filemap_write_and_wait(inode->i_mapping);
135 pCifsInode->write_behind_rc = rc;
137 cFYI(1, "invalidating remote inode since open detected it "
139 invalidate_remote_inode(inode);
144 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
147 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
150 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
151 pCifsInode->clientCanCacheAll = true;
152 pCifsInode->clientCanCacheRead = true;
153 cFYI(1, "Exclusive Oplock granted on inode %p", inode);
154 } else if ((oplock & 0xF) == OPLOCK_READ)
155 pCifsInode->clientCanCacheRead = true;
160 int cifs_posix_open(char *full_path, struct inode **pinode,
161 struct super_block *sb, int mode, unsigned int f_flags,
162 __u32 *poplock, __u16 *pnetfid, int xid)
165 FILE_UNIX_BASIC_INFO *presp_data;
166 __u32 posix_flags = 0;
167 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
168 struct cifs_fattr fattr;
169 struct tcon_link *tlink;
170 struct cifsTconInfo *tcon;
172 cFYI(1, "posix open %s", full_path);
174 presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
175 if (presp_data == NULL)
178 tlink = cifs_sb_tlink(cifs_sb);
184 tcon = tlink_tcon(tlink);
185 mode &= ~current_umask();
187 posix_flags = cifs_posix_convert_flags(f_flags);
188 rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
189 poplock, full_path, cifs_sb->local_nls,
190 cifs_sb->mnt_cifs_flags &
191 CIFS_MOUNT_MAP_SPECIAL_CHR);
192 cifs_put_tlink(tlink);
197 if (presp_data->Type == cpu_to_le32(-1))
198 goto posix_open_ret; /* open ok, caller does qpathinfo */
201 goto posix_open_ret; /* caller does not need info */
203 cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
205 /* get new inode and set it up */
206 if (*pinode == NULL) {
207 cifs_fill_uniqueid(sb, &fattr);
208 *pinode = cifs_iget(sb, &fattr);
214 cifs_fattr_to_inode(*pinode, &fattr);
222 struct cifsFileInfo *
223 cifs_new_fileinfo(__u16 fileHandle, struct file *file,
224 struct tcon_link *tlink, __u32 oplock)
226 struct dentry *dentry = file->f_path.dentry;
227 struct inode *inode = dentry->d_inode;
228 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
229 struct cifsFileInfo *pCifsFile;
231 pCifsFile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
232 if (pCifsFile == NULL)
235 pCifsFile->netfid = fileHandle;
236 pCifsFile->pid = current->tgid;
237 pCifsFile->uid = current_fsuid();
238 pCifsFile->dentry = dget(dentry);
239 pCifsFile->f_flags = file->f_flags;
240 pCifsFile->invalidHandle = false;
241 pCifsFile->closePend = false;
242 pCifsFile->tlink = cifs_get_tlink(tlink);
243 mutex_init(&pCifsFile->fh_mutex);
244 mutex_init(&pCifsFile->lock_mutex);
245 INIT_LIST_HEAD(&pCifsFile->llist);
246 atomic_set(&pCifsFile->count, 1);
247 INIT_WORK(&pCifsFile->oplock_break, cifs_oplock_break);
249 write_lock(&GlobalSMBSeslock);
250 list_add(&pCifsFile->tlist, &(tlink_tcon(tlink)->openFileList));
251 /* if readable file instance put first in list*/
252 if (file->f_mode & FMODE_READ)
253 list_add(&pCifsFile->flist, &pCifsInode->openFileList);
255 list_add_tail(&pCifsFile->flist, &pCifsInode->openFileList);
256 write_unlock(&GlobalSMBSeslock);
258 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
259 pCifsInode->clientCanCacheAll = true;
260 pCifsInode->clientCanCacheRead = true;
261 cFYI(1, "Exclusive Oplock inode %p", inode);
262 } else if ((oplock & 0xF) == OPLOCK_READ)
263 pCifsInode->clientCanCacheRead = true;
265 file->private_data = pCifsFile;
269 int cifs_open(struct inode *inode, struct file *file)
274 struct cifs_sb_info *cifs_sb;
275 struct cifsTconInfo *tcon;
276 struct tcon_link *tlink;
277 struct cifsFileInfo *pCifsFile = NULL;
278 struct cifsInodeInfo *pCifsInode;
279 char *full_path = NULL;
283 FILE_ALL_INFO *buf = NULL;
287 cifs_sb = CIFS_SB(inode->i_sb);
288 tlink = cifs_sb_tlink(cifs_sb);
291 return PTR_ERR(tlink);
293 tcon = tlink_tcon(tlink);
295 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
297 full_path = build_path_from_dentry(file->f_path.dentry);
298 if (full_path == NULL) {
303 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
304 inode, file->f_flags, full_path);
311 if (!tcon->broken_posix_open && tcon->unix_ext &&
312 (tcon->ses->capabilities & CAP_UNIX) &&
313 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
314 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
315 /* can not refresh inode info since size could be stale */
316 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
317 cifs_sb->mnt_file_mode /* ignored */,
318 file->f_flags, &oplock, &netfid, xid);
320 cFYI(1, "posix open succeeded");
322 pCifsFile = cifs_new_fileinfo(netfid, file, tlink,
324 if (pCifsFile == NULL) {
325 CIFSSMBClose(xid, tcon, netfid);
329 cifs_fscache_set_inode_cookie(inode, file);
332 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
333 if (tcon->ses->serverNOS)
334 cERROR(1, "server %s of type %s returned"
335 " unexpected error on SMB posix open"
336 ", disabling posix open support."
337 " Check if server update available.",
338 tcon->ses->serverName,
339 tcon->ses->serverNOS);
340 tcon->broken_posix_open = true;
341 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
342 (rc != -EOPNOTSUPP)) /* path not found or net err */
344 /* else fallthrough to retry open the old way on network i/o
348 desiredAccess = cifs_convert_flags(file->f_flags);
350 /*********************************************************************
351 * open flag mapping table:
353 * POSIX Flag CIFS Disposition
354 * ---------- ----------------
355 * O_CREAT FILE_OPEN_IF
356 * O_CREAT | O_EXCL FILE_CREATE
357 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
358 * O_TRUNC FILE_OVERWRITE
359 * none of the above FILE_OPEN
361 * Note that there is not a direct match between disposition
362 * FILE_SUPERSEDE (ie create whether or not file exists although
363 * O_CREAT | O_TRUNC is similar but truncates the existing
364 * file rather than creating a new file as FILE_SUPERSEDE does
365 * (which uses the attributes / metadata passed in on open call)
367 *? O_SYNC is a reasonable match to CIFS writethrough flag
368 *? and the read write flags match reasonably. O_LARGEFILE
369 *? is irrelevant because largefile support is always used
370 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
371 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
372 *********************************************************************/
374 disposition = cifs_get_disposition(file->f_flags);
376 /* BB pass O_SYNC flag through on file attributes .. BB */
378 /* Also refresh inode by passing in file_info buf returned by SMBOpen
379 and calling get_inode_info with returned buf (at least helps
380 non-Unix server case) */
382 /* BB we can not do this if this is the second open of a file
383 and the first handle has writebehind data, we might be
384 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
385 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
391 if (tcon->ses->capabilities & CAP_NT_SMBS)
392 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
393 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
394 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
395 & CIFS_MOUNT_MAP_SPECIAL_CHR);
397 rc = -EIO; /* no NT SMB support fall into legacy open below */
400 /* Old server, try legacy style OpenX */
401 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
402 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
403 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
404 & CIFS_MOUNT_MAP_SPECIAL_CHR);
407 cFYI(1, "cifs_open returned 0x%x", rc);
411 rc = cifs_open_inode_helper(inode, tcon, oplock, buf, full_path, xid);
415 pCifsFile = cifs_new_fileinfo(netfid, file, tlink, oplock);
416 if (pCifsFile == NULL) {
421 cifs_fscache_set_inode_cookie(inode, file);
423 if (oplock & CIFS_CREATE_ACTION) {
424 /* time to set mode which we can not set earlier due to
425 problems creating new read-only files */
426 if (tcon->unix_ext) {
427 struct cifs_unix_set_info_args args = {
428 .mode = inode->i_mode,
431 .ctime = NO_CHANGE_64,
432 .atime = NO_CHANGE_64,
433 .mtime = NO_CHANGE_64,
436 CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
438 cifs_sb->mnt_cifs_flags &
439 CIFS_MOUNT_MAP_SPECIAL_CHR);
447 cifs_put_tlink(tlink);
451 /* Try to reacquire byte range locks that were released when session */
452 /* to server was lost */
453 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
457 /* BB list all locks open on this file and relock */
462 static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
467 struct cifs_sb_info *cifs_sb;
468 struct cifsTconInfo *tcon;
469 struct cifsInodeInfo *pCifsInode;
471 char *full_path = NULL;
473 int disposition = FILE_OPEN;
477 mutex_lock(&pCifsFile->fh_mutex);
478 if (!pCifsFile->invalidHandle) {
479 mutex_unlock(&pCifsFile->fh_mutex);
485 inode = pCifsFile->dentry->d_inode;
486 cifs_sb = CIFS_SB(inode->i_sb);
487 tcon = tlink_tcon(pCifsFile->tlink);
489 /* can not grab rename sem here because various ops, including
490 those that already have the rename sem can end up causing writepage
491 to get called and if the server was down that means we end up here,
492 and we can never tell if the caller already has the rename_sem */
493 full_path = build_path_from_dentry(pCifsFile->dentry);
494 if (full_path == NULL) {
496 mutex_unlock(&pCifsFile->fh_mutex);
501 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
502 inode, pCifsFile->f_flags, full_path);
509 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
510 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
511 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
514 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
515 * original open. Must mask them off for a reopen.
517 unsigned int oflags = pCifsFile->f_flags &
518 ~(O_CREAT | O_EXCL | O_TRUNC);
520 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
521 cifs_sb->mnt_file_mode /* ignored */,
522 oflags, &oplock, &netfid, xid);
524 cFYI(1, "posix reopen succeeded");
527 /* fallthrough to retry open the old way on errors, especially
528 in the reconnect path it is important to retry hard */
531 desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
533 /* Can not refresh inode by passing in file_info buf to be returned
534 by SMBOpen and then calling get_inode_info with returned buf
535 since file might have write behind data that needs to be flushed
536 and server version of file size can be stale. If we knew for sure
537 that inode was not dirty locally we could do this */
539 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
540 CREATE_NOT_DIR, &netfid, &oplock, NULL,
541 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
542 CIFS_MOUNT_MAP_SPECIAL_CHR);
544 mutex_unlock(&pCifsFile->fh_mutex);
545 cFYI(1, "cifs_open returned 0x%x", rc);
546 cFYI(1, "oplock: %d", oplock);
547 goto reopen_error_exit;
551 pCifsFile->netfid = netfid;
552 pCifsFile->invalidHandle = false;
553 mutex_unlock(&pCifsFile->fh_mutex);
554 pCifsInode = CIFS_I(inode);
557 rc = filemap_write_and_wait(inode->i_mapping);
559 CIFS_I(inode)->write_behind_rc = rc;
561 pCifsInode->clientCanCacheAll = false;
562 pCifsInode->clientCanCacheRead = false;
564 rc = cifs_get_inode_info_unix(&inode,
565 full_path, inode->i_sb, xid);
567 rc = cifs_get_inode_info(&inode,
568 full_path, NULL, inode->i_sb,
570 } /* else we are writing out data to server already
571 and could deadlock if we tried to flush data, and
572 since we do not know if we have data that would
573 invalidate the current end of file on the server
574 we can not go to the server to get the new inod
576 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
577 pCifsInode->clientCanCacheAll = true;
578 pCifsInode->clientCanCacheRead = true;
579 cFYI(1, "Exclusive Oplock granted on inode %p",
580 pCifsFile->dentry->d_inode);
581 } else if ((oplock & 0xF) == OPLOCK_READ) {
582 pCifsInode->clientCanCacheRead = true;
583 pCifsInode->clientCanCacheAll = false;
585 pCifsInode->clientCanCacheRead = false;
586 pCifsInode->clientCanCacheAll = false;
588 cifs_relock_file(pCifsFile);
596 int cifs_close(struct inode *inode, struct file *file)
600 struct cifs_sb_info *cifs_sb;
601 struct cifsTconInfo *pTcon;
602 struct cifsFileInfo *pSMBFile = file->private_data;
606 cifs_sb = CIFS_SB(inode->i_sb);
607 pTcon = tlink_tcon(pSMBFile->tlink);
609 struct cifsLockInfo *li, *tmp;
610 write_lock(&GlobalSMBSeslock);
611 pSMBFile->closePend = true;
613 /* no sense reconnecting to close a file that is
615 if (!pTcon->need_reconnect) {
616 write_unlock(&GlobalSMBSeslock);
618 while ((atomic_read(&pSMBFile->count) != 1)
619 && (timeout <= 2048)) {
620 /* Give write a better chance to get to
621 server ahead of the close. We do not
622 want to add a wait_q here as it would
623 increase the memory utilization as
624 the struct would be in each open file,
625 but this should give enough time to
627 cFYI(DBG2, "close delay, write pending");
631 if (!pTcon->need_reconnect &&
632 !pSMBFile->invalidHandle)
633 rc = CIFSSMBClose(xid, pTcon,
636 write_unlock(&GlobalSMBSeslock);
638 write_unlock(&GlobalSMBSeslock);
640 /* Delete any outstanding lock records.
641 We'll lose them when the file is closed anyway. */
642 mutex_lock(&pSMBFile->lock_mutex);
643 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
644 list_del(&li->llist);
647 mutex_unlock(&pSMBFile->lock_mutex);
649 write_lock(&GlobalSMBSeslock);
650 list_del(&pSMBFile->flist);
651 list_del(&pSMBFile->tlist);
652 write_unlock(&GlobalSMBSeslock);
653 cifsFileInfo_put(file->private_data);
654 file->private_data = NULL;
658 read_lock(&GlobalSMBSeslock);
659 if (list_empty(&(CIFS_I(inode)->openFileList))) {
660 cFYI(1, "closing last open instance for inode %p", inode);
661 /* if the file is not open we do not know if we can cache info
662 on this inode, much less write behind and read ahead */
663 CIFS_I(inode)->clientCanCacheRead = false;
664 CIFS_I(inode)->clientCanCacheAll = false;
666 read_unlock(&GlobalSMBSeslock);
667 if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
668 rc = CIFS_I(inode)->write_behind_rc;
673 int cifs_closedir(struct inode *inode, struct file *file)
677 struct cifsFileInfo *pCFileStruct = file->private_data;
680 cFYI(1, "Closedir inode = 0x%p", inode);
685 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
687 cFYI(1, "Freeing private data in close dir");
688 write_lock(&GlobalSMBSeslock);
689 if (!pCFileStruct->srch_inf.endOfSearch &&
690 !pCFileStruct->invalidHandle) {
691 pCFileStruct->invalidHandle = true;
692 write_unlock(&GlobalSMBSeslock);
693 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
694 cFYI(1, "Closing uncompleted readdir with rc %d",
696 /* not much we can do if it fails anyway, ignore rc */
699 write_unlock(&GlobalSMBSeslock);
700 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
702 cFYI(1, "closedir free smb buf in srch struct");
703 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
704 if (pCFileStruct->srch_inf.smallBuf)
705 cifs_small_buf_release(ptmp);
707 cifs_buf_release(ptmp);
709 cifs_put_tlink(pCFileStruct->tlink);
710 kfree(file->private_data);
711 file->private_data = NULL;
713 /* BB can we lock the filestruct while this is going on? */
718 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
719 __u64 offset, __u8 lockType)
721 struct cifsLockInfo *li =
722 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
728 mutex_lock(&fid->lock_mutex);
729 list_add(&li->llist, &fid->llist);
730 mutex_unlock(&fid->lock_mutex);
734 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
740 bool wait_flag = false;
741 struct cifs_sb_info *cifs_sb;
742 struct cifsTconInfo *tcon;
744 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
745 bool posix_locking = 0;
747 length = 1 + pfLock->fl_end - pfLock->fl_start;
751 cFYI(1, "Lock parm: 0x%x flockflags: "
752 "0x%x flocktype: 0x%x start: %lld end: %lld",
753 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
756 if (pfLock->fl_flags & FL_POSIX)
758 if (pfLock->fl_flags & FL_FLOCK)
760 if (pfLock->fl_flags & FL_SLEEP) {
761 cFYI(1, "Blocking lock");
764 if (pfLock->fl_flags & FL_ACCESS)
765 cFYI(1, "Process suspended by mandatory locking - "
766 "not implemented yet");
767 if (pfLock->fl_flags & FL_LEASE)
768 cFYI(1, "Lease on file - not implemented yet");
769 if (pfLock->fl_flags &
770 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
771 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
773 if (pfLock->fl_type == F_WRLCK) {
776 } else if (pfLock->fl_type == F_UNLCK) {
779 /* Check if unlock includes more than
781 } else if (pfLock->fl_type == F_RDLCK) {
783 lockType |= LOCKING_ANDX_SHARED_LOCK;
785 } else if (pfLock->fl_type == F_EXLCK) {
788 } else if (pfLock->fl_type == F_SHLCK) {
790 lockType |= LOCKING_ANDX_SHARED_LOCK;
793 cFYI(1, "Unknown type of lock");
795 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
796 tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
798 if (file->private_data == NULL) {
803 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
805 if ((tcon->ses->capabilities & CAP_UNIX) &&
806 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
807 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
809 /* BB add code here to normalize offset and length to
810 account for negative length which we can not accept over the
815 if (lockType & LOCKING_ANDX_SHARED_LOCK)
816 posix_lock_type = CIFS_RDLCK;
818 posix_lock_type = CIFS_WRLCK;
819 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
821 posix_lock_type, wait_flag);
826 /* BB we could chain these into one lock request BB */
827 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
828 0, 1, lockType, 0 /* wait flag */ );
830 rc = CIFSSMBLock(xid, tcon, netfid, length,
831 pfLock->fl_start, 1 /* numUnlock */ ,
832 0 /* numLock */ , lockType,
834 pfLock->fl_type = F_UNLCK;
836 cERROR(1, "Error unlocking previously locked "
837 "range %d during test of lock", rc);
841 /* if rc == ERR_SHARING_VIOLATION ? */
844 if (lockType & LOCKING_ANDX_SHARED_LOCK) {
845 pfLock->fl_type = F_WRLCK;
847 rc = CIFSSMBLock(xid, tcon, netfid, length,
848 pfLock->fl_start, 0, 1,
849 lockType | LOCKING_ANDX_SHARED_LOCK,
852 rc = CIFSSMBLock(xid, tcon, netfid,
853 length, pfLock->fl_start, 1, 0,
855 LOCKING_ANDX_SHARED_LOCK,
857 pfLock->fl_type = F_RDLCK;
859 cERROR(1, "Error unlocking "
860 "previously locked range %d "
861 "during test of lock", rc);
864 pfLock->fl_type = F_WRLCK;
874 if (!numLock && !numUnlock) {
875 /* if no lock or unlock then nothing
876 to do since we do not know what it is */
883 if (lockType & LOCKING_ANDX_SHARED_LOCK)
884 posix_lock_type = CIFS_RDLCK;
886 posix_lock_type = CIFS_WRLCK;
889 posix_lock_type = CIFS_UNLCK;
891 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
893 posix_lock_type, wait_flag);
895 struct cifsFileInfo *fid = file->private_data;
898 rc = CIFSSMBLock(xid, tcon, netfid, length,
900 0, numLock, lockType, wait_flag);
903 /* For Windows locks we must store them. */
904 rc = store_file_lock(fid, length,
905 pfLock->fl_start, lockType);
907 } else if (numUnlock) {
908 /* For each stored lock that this unlock overlaps
909 completely, unlock it. */
911 struct cifsLockInfo *li, *tmp;
914 mutex_lock(&fid->lock_mutex);
915 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
916 if (pfLock->fl_start <= li->offset &&
917 (pfLock->fl_start + length) >=
918 (li->offset + li->length)) {
919 stored_rc = CIFSSMBLock(xid, tcon,
921 li->length, li->offset,
922 1, 0, li->type, false);
926 list_del(&li->llist);
931 mutex_unlock(&fid->lock_mutex);
935 if (pfLock->fl_flags & FL_POSIX)
936 posix_lock_file_wait(file, pfLock);
942 * Set the timeout on write requests past EOF. For some servers (Windows)
943 * these calls can be very long.
945 * If we're writing >10M past the EOF we give a 180s timeout. Anything less
946 * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
947 * The 10M cutoff is totally arbitrary. A better scheme for this would be
948 * welcome if someone wants to suggest one.
950 * We may be able to do a better job with this if there were some way to
951 * declare that a file should be sparse.
954 cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
956 if (offset <= cifsi->server_eof)
958 else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
959 return CIFS_VLONG_OP;
964 /* update the file size (if needed) after a write */
966 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
967 unsigned int bytes_written)
969 loff_t end_of_write = offset + bytes_written;
971 if (end_of_write > cifsi->server_eof)
972 cifsi->server_eof = end_of_write;
975 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
976 size_t write_size, loff_t *poffset)
979 unsigned int bytes_written = 0;
980 unsigned int total_written;
981 struct cifs_sb_info *cifs_sb;
982 struct cifsTconInfo *pTcon;
984 struct cifsFileInfo *open_file;
985 struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
987 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
989 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
990 *poffset, file->f_path.dentry->d_name.name); */
992 if (file->private_data == NULL)
995 open_file = file->private_data;
996 pTcon = tlink_tcon(open_file->tlink);
998 rc = generic_write_checks(file, poffset, &write_size, 0);
1004 long_op = cifs_write_timeout(cifsi, *poffset);
1005 for (total_written = 0; write_size > total_written;
1006 total_written += bytes_written) {
1008 while (rc == -EAGAIN) {
1009 if (file->private_data == NULL) {
1010 /* file has been closed on us */
1012 /* if we have gotten here we have written some data
1013 and blocked, and the file has been freed on us while
1014 we blocked so return what we managed to write */
1015 return total_written;
1017 if (open_file->closePend) {
1020 return total_written;
1024 if (open_file->invalidHandle) {
1025 /* we could deadlock if we called
1026 filemap_fdatawait from here so tell
1027 reopen_file not to flush data to server
1029 rc = cifs_reopen_file(open_file, false);
1034 rc = CIFSSMBWrite(xid, pTcon,
1036 min_t(const int, cifs_sb->wsize,
1037 write_size - total_written),
1038 *poffset, &bytes_written,
1039 NULL, write_data + total_written, long_op);
1041 if (rc || (bytes_written == 0)) {
1049 cifs_update_eof(cifsi, *poffset, bytes_written);
1050 *poffset += bytes_written;
1052 long_op = CIFS_STD_OP; /* subsequent writes fast -
1053 15 seconds is plenty */
1056 cifs_stats_bytes_written(pTcon, total_written);
1058 /* since the write may have blocked check these pointers again */
1059 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1060 struct inode *inode = file->f_path.dentry->d_inode;
1061 /* Do not update local mtime - server will set its actual value on write
1062 * inode->i_ctime = inode->i_mtime =
1063 * current_fs_time(inode->i_sb);*/
1064 if (total_written > 0) {
1065 spin_lock(&inode->i_lock);
1066 if (*poffset > file->f_path.dentry->d_inode->i_size)
1067 i_size_write(file->f_path.dentry->d_inode,
1069 spin_unlock(&inode->i_lock);
1071 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1074 return total_written;
1077 static ssize_t cifs_write(struct cifsFileInfo *open_file,
1078 const char *write_data, size_t write_size,
1082 unsigned int bytes_written = 0;
1083 unsigned int total_written;
1084 struct cifs_sb_info *cifs_sb;
1085 struct cifsTconInfo *pTcon;
1087 struct dentry *dentry = open_file->dentry;
1088 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
1090 cifs_sb = CIFS_SB(dentry->d_sb);
1092 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
1093 *poffset, dentry->d_name.name);
1095 pTcon = tlink_tcon(open_file->tlink);
1099 long_op = cifs_write_timeout(cifsi, *poffset);
1100 for (total_written = 0; write_size > total_written;
1101 total_written += bytes_written) {
1103 while (rc == -EAGAIN) {
1104 if (open_file->closePend) {
1107 return total_written;
1111 if (open_file->invalidHandle) {
1112 /* we could deadlock if we called
1113 filemap_fdatawait from here so tell
1114 reopen_file not to flush data to
1116 rc = cifs_reopen_file(open_file, false);
1120 if (experimEnabled || (pTcon->ses->server &&
1121 ((pTcon->ses->server->secMode &
1122 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1127 len = min((size_t)cifs_sb->wsize,
1128 write_size - total_written);
1129 /* iov[0] is reserved for smb header */
1130 iov[1].iov_base = (char *)write_data +
1132 iov[1].iov_len = len;
1133 rc = CIFSSMBWrite2(xid, pTcon,
1134 open_file->netfid, len,
1135 *poffset, &bytes_written,
1138 rc = CIFSSMBWrite(xid, pTcon,
1140 min_t(const int, cifs_sb->wsize,
1141 write_size - total_written),
1142 *poffset, &bytes_written,
1143 write_data + total_written,
1146 if (rc || (bytes_written == 0)) {
1154 cifs_update_eof(cifsi, *poffset, bytes_written);
1155 *poffset += bytes_written;
1157 long_op = CIFS_STD_OP; /* subsequent writes fast -
1158 15 seconds is plenty */
1161 cifs_stats_bytes_written(pTcon, total_written);
1163 if (total_written > 0) {
1164 spin_lock(&dentry->d_inode->i_lock);
1165 if (*poffset > dentry->d_inode->i_size)
1166 i_size_write(dentry->d_inode, *poffset);
1167 spin_unlock(&dentry->d_inode->i_lock);
1169 mark_inode_dirty_sync(dentry->d_inode);
1171 return total_written;
1174 #ifdef CONFIG_CIFS_EXPERIMENTAL
1175 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1178 struct cifsFileInfo *open_file = NULL;
1179 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1181 /* only filter by fsuid on multiuser mounts */
1182 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1185 read_lock(&GlobalSMBSeslock);
1186 /* we could simply get the first_list_entry since write-only entries
1187 are always at the end of the list but since the first entry might
1188 have a close pending, we go through the whole list */
1189 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1190 if (open_file->closePend)
1192 if (fsuid_only && open_file->uid != current_fsuid())
1194 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1195 if (!open_file->invalidHandle) {
1196 /* found a good file */
1197 /* lock it so it will not be closed on us */
1198 cifsFileInfo_get(open_file);
1199 read_unlock(&GlobalSMBSeslock);
1201 } /* else might as well continue, and look for
1202 another, or simply have the caller reopen it
1203 again rather than trying to fix this handle */
1204 } else /* write only file */
1205 break; /* write only files are last so must be done */
1207 read_unlock(&GlobalSMBSeslock);
1212 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1215 struct cifsFileInfo *open_file;
1216 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1217 bool any_available = false;
1220 /* Having a null inode here (because mapping->host was set to zero by
1221 the VFS or MM) should not happen but we had reports of on oops (due to
1222 it being zero) during stress testcases so we need to check for it */
1224 if (cifs_inode == NULL) {
1225 cERROR(1, "Null inode passed to cifs_writeable_file");
1230 /* only filter by fsuid on multiuser mounts */
1231 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1234 read_lock(&GlobalSMBSeslock);
1236 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1237 if (open_file->closePend)
1239 if (!any_available && open_file->pid != current->tgid)
1241 if (fsuid_only && open_file->uid != current_fsuid())
1243 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1244 cifsFileInfo_get(open_file);
1246 if (!open_file->invalidHandle) {
1247 /* found a good writable file */
1248 read_unlock(&GlobalSMBSeslock);
1252 read_unlock(&GlobalSMBSeslock);
1253 /* Had to unlock since following call can block */
1254 rc = cifs_reopen_file(open_file, false);
1256 if (!open_file->closePend)
1258 else { /* start over in case this was deleted */
1259 /* since the list could be modified */
1260 read_lock(&GlobalSMBSeslock);
1261 cifsFileInfo_put(open_file);
1262 goto refind_writable;
1266 /* if it fails, try another handle if possible -
1267 (we can not do this if closePending since
1268 loop could be modified - in which case we
1269 have to start at the beginning of the list
1270 again. Note that it would be bad
1271 to hold up writepages here (rather than
1272 in caller) with continuous retries */
1273 cFYI(1, "wp failed on reopen file");
1274 read_lock(&GlobalSMBSeslock);
1275 /* can not use this handle, no write
1276 pending on this one after all */
1277 cifsFileInfo_put(open_file);
1279 if (open_file->closePend) /* list could have changed */
1280 goto refind_writable;
1281 /* else we simply continue to the next entry. Thus
1282 we do not loop on reopen errors. If we
1283 can not reopen the file, for example if we
1284 reconnected to a server with another client
1285 racing to delete or lock the file we would not
1286 make progress if we restarted before the beginning
1287 of the loop here. */
1290 /* couldn't find useable FH with same pid, try any available */
1291 if (!any_available) {
1292 any_available = true;
1293 goto refind_writable;
1295 read_unlock(&GlobalSMBSeslock);
1299 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1301 struct address_space *mapping = page->mapping;
1302 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1305 int bytes_written = 0;
1306 struct cifs_sb_info *cifs_sb;
1307 struct inode *inode;
1308 struct cifsFileInfo *open_file;
1310 if (!mapping || !mapping->host)
1313 inode = page->mapping->host;
1314 cifs_sb = CIFS_SB(inode->i_sb);
1316 offset += (loff_t)from;
1317 write_data = kmap(page);
1320 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1325 /* racing with truncate? */
1326 if (offset > mapping->host->i_size) {
1328 return 0; /* don't care */
1331 /* check to make sure that we are not extending the file */
1332 if (mapping->host->i_size - offset < (loff_t)to)
1333 to = (unsigned)(mapping->host->i_size - offset);
1335 open_file = find_writable_file(CIFS_I(mapping->host), false);
1337 bytes_written = cifs_write(open_file, write_data,
1338 to - from, &offset);
1339 cifsFileInfo_put(open_file);
1340 /* Does mm or vfs already set times? */
1341 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1342 if ((bytes_written > 0) && (offset))
1344 else if (bytes_written < 0)
1347 cFYI(1, "No writeable filehandles for inode");
1355 static int cifs_writepages(struct address_space *mapping,
1356 struct writeback_control *wbc)
1358 struct backing_dev_info *bdi = mapping->backing_dev_info;
1359 unsigned int bytes_to_write;
1360 unsigned int bytes_written;
1361 struct cifs_sb_info *cifs_sb;
1365 int range_whole = 0;
1372 struct cifsFileInfo *open_file;
1373 struct cifsTconInfo *tcon;
1374 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1376 struct pagevec pvec;
1382 * BB: Is this meaningful for a non-block-device file system?
1383 * If it is, we should test it again after we do I/O
1385 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1386 wbc->encountered_congestion = 1;
1390 cifs_sb = CIFS_SB(mapping->host->i_sb);
1393 * If wsize is smaller that the page cache size, default to writing
1394 * one page at a time via cifs_writepage
1396 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1397 return generic_writepages(mapping, wbc);
1399 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1401 return generic_writepages(mapping, wbc);
1404 * if there's no open file, then this is likely to fail too,
1405 * but it'll at least handle the return. Maybe it should be
1408 open_file = find_writable_file(CIFS_I(mapping->host), false);
1411 return generic_writepages(mapping, wbc);
1414 tcon = tlink_tcon(open_file->tlink);
1415 if (!experimEnabled && tcon->ses->server->secMode &
1416 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1417 cifsFileInfo_put(open_file);
1418 return generic_writepages(mapping, wbc);
1420 cifsFileInfo_put(open_file);
1424 pagevec_init(&pvec, 0);
1425 if (wbc->range_cyclic) {
1426 index = mapping->writeback_index; /* Start from prev offset */
1429 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1430 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1431 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1436 while (!done && (index <= end) &&
1437 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1438 PAGECACHE_TAG_DIRTY,
1439 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1448 for (i = 0; i < nr_pages; i++) {
1449 page = pvec.pages[i];
1451 * At this point we hold neither mapping->tree_lock nor
1452 * lock on the page itself: the page may be truncated or
1453 * invalidated (changing page->mapping to NULL), or even
1454 * swizzled back from swapper_space to tmpfs file
1460 else if (!trylock_page(page))
1463 if (unlikely(page->mapping != mapping)) {
1468 if (!wbc->range_cyclic && page->index > end) {
1474 if (next && (page->index != next)) {
1475 /* Not next consecutive page */
1480 if (wbc->sync_mode != WB_SYNC_NONE)
1481 wait_on_page_writeback(page);
1483 if (PageWriteback(page) ||
1484 !clear_page_dirty_for_io(page)) {
1490 * This actually clears the dirty bit in the radix tree.
1491 * See cifs_writepage() for more commentary.
1493 set_page_writeback(page);
1495 if (page_offset(page) >= mapping->host->i_size) {
1498 end_page_writeback(page);
1503 * BB can we get rid of this? pages are held by pvec
1505 page_cache_get(page);
1507 len = min(mapping->host->i_size - page_offset(page),
1508 (loff_t)PAGE_CACHE_SIZE);
1510 /* reserve iov[0] for the smb header */
1512 iov[n_iov].iov_base = kmap(page);
1513 iov[n_iov].iov_len = len;
1514 bytes_to_write += len;
1518 offset = page_offset(page);
1520 next = page->index + 1;
1521 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1525 open_file = find_writable_file(CIFS_I(mapping->host),
1528 cERROR(1, "No writable handles for inode");
1531 long_op = cifs_write_timeout(cifsi, offset);
1532 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1533 bytes_to_write, offset,
1534 &bytes_written, iov, n_iov,
1536 cifsFileInfo_put(open_file);
1537 cifs_update_eof(cifsi, offset, bytes_written);
1540 if (rc || bytes_written < bytes_to_write) {
1541 cERROR(1, "Write2 ret %d, wrote %d",
1543 /* BB what if continued retry is
1544 requested via mount flags? */
1546 set_bit(AS_ENOSPC, &mapping->flags);
1548 set_bit(AS_EIO, &mapping->flags);
1550 cifs_stats_bytes_written(tcon, bytes_written);
1553 for (i = 0; i < n_iov; i++) {
1554 page = pvec.pages[first + i];
1555 /* Should we also set page error on
1556 success rc but too little data written? */
1557 /* BB investigate retry logic on temporary
1558 server crash cases and how recovery works
1559 when page marked as error */
1564 end_page_writeback(page);
1565 page_cache_release(page);
1567 if ((wbc->nr_to_write -= n_iov) <= 0)
1571 /* Need to re-find the pages we skipped */
1572 index = pvec.pages[0]->index + 1;
1574 pagevec_release(&pvec);
1576 if (!scanned && !done) {
1578 * We hit the last page and there is more work to be done: wrap
1579 * back to the start of the file
1585 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1586 mapping->writeback_index = index;
1593 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1599 /* BB add check for wbc flags */
1600 page_cache_get(page);
1601 if (!PageUptodate(page))
1602 cFYI(1, "ppw - page not up to date");
1605 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1607 * A writepage() implementation always needs to do either this,
1608 * or re-dirty the page with "redirty_page_for_writepage()" in
1609 * the case of a failure.
1611 * Just unlocking the page will cause the radix tree tag-bits
1612 * to fail to update with the state of the page correctly.
1614 set_page_writeback(page);
1615 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1616 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1618 end_page_writeback(page);
1619 page_cache_release(page);
1624 static int cifs_write_end(struct file *file, struct address_space *mapping,
1625 loff_t pos, unsigned len, unsigned copied,
1626 struct page *page, void *fsdata)
1629 struct inode *inode = mapping->host;
1631 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1634 if (PageChecked(page)) {
1636 SetPageUptodate(page);
1637 ClearPageChecked(page);
1638 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1639 SetPageUptodate(page);
1641 if (!PageUptodate(page)) {
1643 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1647 /* this is probably better than directly calling
1648 partialpage_write since in this function the file handle is
1649 known which we might as well leverage */
1650 /* BB check if anything else missing out of ppw
1651 such as updating last write time */
1652 page_data = kmap(page);
1653 rc = cifs_write(file->private_data, page_data + offset,
1655 /* if (rc < 0) should we set writebehind rc? */
1662 set_page_dirty(page);
1666 spin_lock(&inode->i_lock);
1667 if (pos > inode->i_size)
1668 i_size_write(inode, pos);
1669 spin_unlock(&inode->i_lock);
1673 page_cache_release(page);
1678 int cifs_fsync(struct file *file, int datasync)
1682 struct cifsTconInfo *tcon;
1683 struct cifsFileInfo *smbfile = file->private_data;
1684 struct inode *inode = file->f_path.dentry->d_inode;
1688 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1689 file->f_path.dentry->d_name.name, datasync);
1691 rc = filemap_write_and_wait(inode->i_mapping);
1693 rc = CIFS_I(inode)->write_behind_rc;
1694 CIFS_I(inode)->write_behind_rc = 0;
1695 tcon = tlink_tcon(smbfile->tlink);
1696 if (!rc && tcon && smbfile &&
1697 !(CIFS_SB(inode->i_sb)->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1698 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1705 /* static void cifs_sync_page(struct page *page)
1707 struct address_space *mapping;
1708 struct inode *inode;
1709 unsigned long index = page->index;
1710 unsigned int rpages = 0;
1713 cFYI(1, "sync page %p", page);
1714 mapping = page->mapping;
1717 inode = mapping->host;
1721 /* fill in rpages then
1722 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1724 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1734 * As file closes, flush all cached write data for this inode checking
1735 * for write behind errors.
1737 int cifs_flush(struct file *file, fl_owner_t id)
1739 struct inode *inode = file->f_path.dentry->d_inode;
1742 /* Rather than do the steps manually:
1743 lock the inode for writing
1744 loop through pages looking for write behind data (dirty pages)
1745 coalesce into contiguous 16K (or smaller) chunks to write to server
1746 send to server (prefer in parallel)
1747 deal with writebehind errors
1748 unlock inode for writing
1749 filemapfdatawrite appears easier for the time being */
1751 rc = filemap_fdatawrite(inode->i_mapping);
1752 /* reset wb rc if we were able to write out dirty pages */
1754 rc = CIFS_I(inode)->write_behind_rc;
1755 CIFS_I(inode)->write_behind_rc = 0;
1758 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1763 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1764 size_t read_size, loff_t *poffset)
1767 unsigned int bytes_read = 0;
1768 unsigned int total_read = 0;
1769 unsigned int current_read_size;
1770 struct cifs_sb_info *cifs_sb;
1771 struct cifsTconInfo *pTcon;
1773 struct cifsFileInfo *open_file;
1774 char *smb_read_data;
1775 char __user *current_offset;
1776 struct smb_com_read_rsp *pSMBr;
1779 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1781 if (file->private_data == NULL) {
1786 open_file = file->private_data;
1787 pTcon = tlink_tcon(open_file->tlink);
1789 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1790 cFYI(1, "attempting read on write only file instance");
1792 for (total_read = 0, current_offset = read_data;
1793 read_size > total_read;
1794 total_read += bytes_read, current_offset += bytes_read) {
1795 current_read_size = min_t(const int, read_size - total_read,
1798 smb_read_data = NULL;
1799 while (rc == -EAGAIN) {
1800 int buf_type = CIFS_NO_BUFFER;
1801 if ((open_file->invalidHandle) &&
1802 (!open_file->closePend)) {
1803 rc = cifs_reopen_file(open_file, true);
1807 rc = CIFSSMBRead(xid, pTcon,
1809 current_read_size, *poffset,
1810 &bytes_read, &smb_read_data,
1812 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1813 if (smb_read_data) {
1814 if (copy_to_user(current_offset,
1816 4 /* RFC1001 length field */ +
1817 le16_to_cpu(pSMBr->DataOffset),
1821 if (buf_type == CIFS_SMALL_BUFFER)
1822 cifs_small_buf_release(smb_read_data);
1823 else if (buf_type == CIFS_LARGE_BUFFER)
1824 cifs_buf_release(smb_read_data);
1825 smb_read_data = NULL;
1828 if (rc || (bytes_read == 0)) {
1836 cifs_stats_bytes_read(pTcon, bytes_read);
1837 *poffset += bytes_read;
1845 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1849 unsigned int bytes_read = 0;
1850 unsigned int total_read;
1851 unsigned int current_read_size;
1852 struct cifs_sb_info *cifs_sb;
1853 struct cifsTconInfo *pTcon;
1855 char *current_offset;
1856 struct cifsFileInfo *open_file;
1857 int buf_type = CIFS_NO_BUFFER;
1860 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1862 if (file->private_data == NULL) {
1867 open_file = file->private_data;
1868 pTcon = tlink_tcon(open_file->tlink);
1870 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1871 cFYI(1, "attempting read on write only file instance");
1873 for (total_read = 0, current_offset = read_data;
1874 read_size > total_read;
1875 total_read += bytes_read, current_offset += bytes_read) {
1876 current_read_size = min_t(const int, read_size - total_read,
1878 /* For windows me and 9x we do not want to request more
1879 than it negotiated since it will refuse the read then */
1881 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1882 current_read_size = min_t(const int, current_read_size,
1883 pTcon->ses->server->maxBuf - 128);
1886 while (rc == -EAGAIN) {
1887 if ((open_file->invalidHandle) &&
1888 (!open_file->closePend)) {
1889 rc = cifs_reopen_file(open_file, true);
1893 rc = CIFSSMBRead(xid, pTcon,
1895 current_read_size, *poffset,
1896 &bytes_read, ¤t_offset,
1899 if (rc || (bytes_read == 0)) {
1907 cifs_stats_bytes_read(pTcon, total_read);
1908 *poffset += bytes_read;
1915 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1920 rc = cifs_revalidate_file(file);
1922 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1926 rc = generic_file_mmap(file, vma);
1932 static void cifs_copy_cache_pages(struct address_space *mapping,
1933 struct list_head *pages, int bytes_read, char *data)
1938 while (bytes_read > 0) {
1939 if (list_empty(pages))
1942 page = list_entry(pages->prev, struct page, lru);
1943 list_del(&page->lru);
1945 if (add_to_page_cache_lru(page, mapping, page->index,
1947 page_cache_release(page);
1948 cFYI(1, "Add page cache failed");
1949 data += PAGE_CACHE_SIZE;
1950 bytes_read -= PAGE_CACHE_SIZE;
1953 page_cache_release(page);
1955 target = kmap_atomic(page, KM_USER0);
1957 if (PAGE_CACHE_SIZE > bytes_read) {
1958 memcpy(target, data, bytes_read);
1959 /* zero the tail end of this partial page */
1960 memset(target + bytes_read, 0,
1961 PAGE_CACHE_SIZE - bytes_read);
1964 memcpy(target, data, PAGE_CACHE_SIZE);
1965 bytes_read -= PAGE_CACHE_SIZE;
1967 kunmap_atomic(target, KM_USER0);
1969 flush_dcache_page(page);
1970 SetPageUptodate(page);
1972 data += PAGE_CACHE_SIZE;
1974 /* add page to FS-Cache */
1975 cifs_readpage_to_fscache(mapping->host, page);
1980 static int cifs_readpages(struct file *file, struct address_space *mapping,
1981 struct list_head *page_list, unsigned num_pages)
1987 struct cifs_sb_info *cifs_sb;
1988 struct cifsTconInfo *pTcon;
1989 unsigned int bytes_read = 0;
1990 unsigned int read_size, i;
1991 char *smb_read_data = NULL;
1992 struct smb_com_read_rsp *pSMBr;
1993 struct cifsFileInfo *open_file;
1994 int buf_type = CIFS_NO_BUFFER;
1997 if (file->private_data == NULL) {
2002 open_file = file->private_data;
2003 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2004 pTcon = tlink_tcon(open_file->tlink);
2007 * Reads as many pages as possible from fscache. Returns -ENOBUFS
2008 * immediately if the cookie is negative
2010 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
2015 cFYI(DBG2, "rpages: num pages %d", num_pages);
2016 for (i = 0; i < num_pages; ) {
2017 unsigned contig_pages;
2018 struct page *tmp_page;
2019 unsigned long expected_index;
2021 if (list_empty(page_list))
2024 page = list_entry(page_list->prev, struct page, lru);
2025 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2027 /* count adjacent pages that we will read into */
2030 list_entry(page_list->prev, struct page, lru)->index;
2031 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2032 if (tmp_page->index == expected_index) {
2038 if (contig_pages + i > num_pages)
2039 contig_pages = num_pages - i;
2041 /* for reads over a certain size could initiate async
2044 read_size = contig_pages * PAGE_CACHE_SIZE;
2045 /* Read size needs to be in multiples of one page */
2046 read_size = min_t(const unsigned int, read_size,
2047 cifs_sb->rsize & PAGE_CACHE_MASK);
2048 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
2049 read_size, contig_pages);
2051 while (rc == -EAGAIN) {
2052 if ((open_file->invalidHandle) &&
2053 (!open_file->closePend)) {
2054 rc = cifs_reopen_file(open_file, true);
2059 rc = CIFSSMBRead(xid, pTcon,
2062 &bytes_read, &smb_read_data,
2064 /* BB more RC checks ? */
2065 if (rc == -EAGAIN) {
2066 if (smb_read_data) {
2067 if (buf_type == CIFS_SMALL_BUFFER)
2068 cifs_small_buf_release(smb_read_data);
2069 else if (buf_type == CIFS_LARGE_BUFFER)
2070 cifs_buf_release(smb_read_data);
2071 smb_read_data = NULL;
2075 if ((rc < 0) || (smb_read_data == NULL)) {
2076 cFYI(1, "Read error in readpages: %d", rc);
2078 } else if (bytes_read > 0) {
2079 task_io_account_read(bytes_read);
2080 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2081 cifs_copy_cache_pages(mapping, page_list, bytes_read,
2082 smb_read_data + 4 /* RFC1001 hdr */ +
2083 le16_to_cpu(pSMBr->DataOffset));
2085 i += bytes_read >> PAGE_CACHE_SHIFT;
2086 cifs_stats_bytes_read(pTcon, bytes_read);
2087 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2088 i++; /* account for partial page */
2090 /* server copy of file can have smaller size
2092 /* BB do we need to verify this common case ?
2093 this case is ok - if we are at server EOF
2094 we will hit it on next read */
2099 cFYI(1, "No bytes read (%d) at offset %lld . "
2100 "Cleaning remaining pages from readahead list",
2101 bytes_read, offset);
2102 /* BB turn off caching and do new lookup on
2103 file size at server? */
2106 if (smb_read_data) {
2107 if (buf_type == CIFS_SMALL_BUFFER)
2108 cifs_small_buf_release(smb_read_data);
2109 else if (buf_type == CIFS_LARGE_BUFFER)
2110 cifs_buf_release(smb_read_data);
2111 smb_read_data = NULL;
2116 /* need to free smb_read_data buf before exit */
2117 if (smb_read_data) {
2118 if (buf_type == CIFS_SMALL_BUFFER)
2119 cifs_small_buf_release(smb_read_data);
2120 else if (buf_type == CIFS_LARGE_BUFFER)
2121 cifs_buf_release(smb_read_data);
2122 smb_read_data = NULL;
2130 static int cifs_readpage_worker(struct file *file, struct page *page,
2136 /* Is the page cached? */
2137 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2141 page_cache_get(page);
2142 read_data = kmap(page);
2143 /* for reads over a certain size could initiate async read ahead */
2145 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2150 cFYI(1, "Bytes read %d", rc);
2152 file->f_path.dentry->d_inode->i_atime =
2153 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2155 if (PAGE_CACHE_SIZE > rc)
2156 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2158 flush_dcache_page(page);
2159 SetPageUptodate(page);
2161 /* send this page to the cache */
2162 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2168 page_cache_release(page);
2174 static int cifs_readpage(struct file *file, struct page *page)
2176 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2182 if (file->private_data == NULL) {
2188 cFYI(1, "readpage %p at offset %d 0x%x\n",
2189 page, (int)offset, (int)offset);
2191 rc = cifs_readpage_worker(file, page, &offset);
2199 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2201 struct cifsFileInfo *open_file;
2203 read_lock(&GlobalSMBSeslock);
2204 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2205 if (open_file->closePend)
2207 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2208 read_unlock(&GlobalSMBSeslock);
2212 read_unlock(&GlobalSMBSeslock);
2216 /* We do not want to update the file size from server for inodes
2217 open for write - to avoid races with writepage extending
2218 the file - in the future we could consider allowing
2219 refreshing the inode only on increases in the file size
2220 but this is tricky to do without racing with writebehind
2221 page caching in the current Linux kernel design */
2222 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2227 if (is_inode_writable(cifsInode)) {
2228 /* This inode is open for write at least once */
2229 struct cifs_sb_info *cifs_sb;
2231 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2232 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2233 /* since no page cache to corrupt on directio
2234 we can change size safely */
2238 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2246 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2247 loff_t pos, unsigned len, unsigned flags,
2248 struct page **pagep, void **fsdata)
2250 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2251 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2252 loff_t page_start = pos & PAGE_MASK;
2257 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2259 page = grab_cache_page_write_begin(mapping, index, flags);
2265 if (PageUptodate(page))
2269 * If we write a full page it will be up to date, no need to read from
2270 * the server. If the write is short, we'll end up doing a sync write
2273 if (len == PAGE_CACHE_SIZE)
2277 * optimize away the read when we have an oplock, and we're not
2278 * expecting to use any of the data we'd be reading in. That
2279 * is, when the page lies beyond the EOF, or straddles the EOF
2280 * and the write will cover all of the existing data.
2282 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2283 i_size = i_size_read(mapping->host);
2284 if (page_start >= i_size ||
2285 (offset == 0 && (pos + len) >= i_size)) {
2286 zero_user_segments(page, 0, offset,
2290 * PageChecked means that the parts of the page
2291 * to which we're not writing are considered up
2292 * to date. Once the data is copied to the
2293 * page, it can be set uptodate.
2295 SetPageChecked(page);
2300 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2302 * might as well read a page, it is fast enough. If we get
2303 * an error, we don't need to return it. cifs_write_end will
2304 * do a sync write instead since PG_uptodate isn't set.
2306 cifs_readpage_worker(file, page, &page_start);
2308 /* we could try using another file handle if there is one -
2309 but how would we lock it to prevent close of that handle
2310 racing with this read? In any case
2311 this will be written out by write_end so is fine */
2318 static int cifs_release_page(struct page *page, gfp_t gfp)
2320 if (PagePrivate(page))
2323 return cifs_fscache_release_page(page, gfp);
2326 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2328 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2331 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2334 void cifs_oplock_break(struct work_struct *work)
2336 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2338 struct inode *inode = cfile->dentry->d_inode;
2339 struct cifsInodeInfo *cinode = CIFS_I(inode);
2342 if (inode && S_ISREG(inode->i_mode)) {
2343 if (cinode->clientCanCacheRead)
2344 break_lease(inode, O_RDONLY);
2346 break_lease(inode, O_WRONLY);
2347 rc = filemap_fdatawrite(inode->i_mapping);
2348 if (cinode->clientCanCacheRead == 0) {
2349 waitrc = filemap_fdatawait(inode->i_mapping);
2350 invalidate_remote_inode(inode);
2355 cinode->write_behind_rc = rc;
2356 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2360 * releasing stale oplock after recent reconnect of smb session using
2361 * a now incorrect file handle is not a data integrity issue but do
2362 * not bother sending an oplock release if session to server still is
2363 * disconnected since oplock already released by the server
2365 if (!cfile->closePend && !cfile->oplock_break_cancelled) {
2366 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2367 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false);
2368 cFYI(1, "Oplock release rc = %d", rc);
2372 * We might have kicked in before is_valid_oplock_break()
2373 * finished grabbing reference for us. Make sure it's done by
2374 * waiting for GlobalSMSSeslock.
2376 write_lock(&GlobalSMBSeslock);
2377 write_unlock(&GlobalSMBSeslock);
2379 cifs_oplock_break_put(cfile);
2382 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2384 cifs_sb_active(cfile->dentry->d_sb);
2385 cifsFileInfo_get(cfile);
2388 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2390 cifsFileInfo_put(cfile);
2391 cifs_sb_deactive(cfile->dentry->d_sb);
2394 const struct address_space_operations cifs_addr_ops = {
2395 .readpage = cifs_readpage,
2396 .readpages = cifs_readpages,
2397 .writepage = cifs_writepage,
2398 .writepages = cifs_writepages,
2399 .write_begin = cifs_write_begin,
2400 .write_end = cifs_write_end,
2401 .set_page_dirty = __set_page_dirty_nobuffers,
2402 .releasepage = cifs_release_page,
2403 .invalidatepage = cifs_invalidate_page,
2404 /* .sync_page = cifs_sync_page, */
2409 * cifs_readpages requires the server to support a buffer large enough to
2410 * contain the header plus one complete page of data. Otherwise, we need
2411 * to leave cifs_readpages out of the address space operations.
2413 const struct address_space_operations cifs_addr_ops_smallbuf = {
2414 .readpage = cifs_readpage,
2415 .writepage = cifs_writepage,
2416 .writepages = cifs_writepages,
2417 .write_begin = cifs_write_begin,
2418 .write_end = cifs_write_end,
2419 .set_page_dirty = __set_page_dirty_nobuffers,
2420 .releasepage = cifs_release_page,
2421 .invalidatepage = cifs_invalidate_page,
2422 /* .sync_page = cifs_sync_page, */