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);
134 mapping_set_error(inode->i_mapping, rc);
136 cFYI(1, "invalidating remote inode since open detected it "
138 invalidate_remote_inode(inode);
143 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
146 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
149 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
150 pCifsInode->clientCanCacheAll = true;
151 pCifsInode->clientCanCacheRead = true;
152 cFYI(1, "Exclusive Oplock granted on inode %p", inode);
153 } else if ((oplock & 0xF) == OPLOCK_READ)
154 pCifsInode->clientCanCacheRead = true;
159 int cifs_posix_open(char *full_path, struct inode **pinode,
160 struct super_block *sb, int mode, unsigned int f_flags,
161 __u32 *poplock, __u16 *pnetfid, int xid)
164 FILE_UNIX_BASIC_INFO *presp_data;
165 __u32 posix_flags = 0;
166 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
167 struct cifs_fattr fattr;
168 struct tcon_link *tlink;
169 struct cifsTconInfo *tcon;
171 cFYI(1, "posix open %s", full_path);
173 presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
174 if (presp_data == NULL)
177 tlink = cifs_sb_tlink(cifs_sb);
183 tcon = tlink_tcon(tlink);
184 mode &= ~current_umask();
186 posix_flags = cifs_posix_convert_flags(f_flags);
187 rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
188 poplock, full_path, cifs_sb->local_nls,
189 cifs_sb->mnt_cifs_flags &
190 CIFS_MOUNT_MAP_SPECIAL_CHR);
191 cifs_put_tlink(tlink);
196 if (presp_data->Type == cpu_to_le32(-1))
197 goto posix_open_ret; /* open ok, caller does qpathinfo */
200 goto posix_open_ret; /* caller does not need info */
202 cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
204 /* get new inode and set it up */
205 if (*pinode == NULL) {
206 cifs_fill_uniqueid(sb, &fattr);
207 *pinode = cifs_iget(sb, &fattr);
213 cifs_fattr_to_inode(*pinode, &fattr);
221 struct cifsFileInfo *
222 cifs_new_fileinfo(__u16 fileHandle, struct file *file,
223 struct tcon_link *tlink, __u32 oplock)
225 struct dentry *dentry = file->f_path.dentry;
226 struct inode *inode = dentry->d_inode;
227 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
228 struct cifsFileInfo *pCifsFile;
230 pCifsFile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
231 if (pCifsFile == NULL)
234 pCifsFile->count = 1;
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->tlink = cifs_get_tlink(tlink);
242 mutex_init(&pCifsFile->fh_mutex);
243 mutex_init(&pCifsFile->lock_mutex);
244 INIT_LIST_HEAD(&pCifsFile->llist);
245 INIT_WORK(&pCifsFile->oplock_break, cifs_oplock_break);
247 spin_lock(&cifs_file_list_lock);
248 list_add(&pCifsFile->tlist, &(tlink_tcon(tlink)->openFileList));
249 /* if readable file instance put first in list*/
250 if (file->f_mode & FMODE_READ)
251 list_add(&pCifsFile->flist, &pCifsInode->openFileList);
253 list_add_tail(&pCifsFile->flist, &pCifsInode->openFileList);
254 spin_unlock(&cifs_file_list_lock);
256 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
257 pCifsInode->clientCanCacheAll = true;
258 pCifsInode->clientCanCacheRead = true;
259 cFYI(1, "Exclusive Oplock inode %p", inode);
260 } else if ((oplock & 0xF) == OPLOCK_READ)
261 pCifsInode->clientCanCacheRead = true;
263 file->private_data = pCifsFile;
268 * Release a reference on the file private data. This may involve closing
269 * the filehandle out on the server. Must be called without holding
270 * cifs_file_list_lock.
272 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
274 struct cifsTconInfo *tcon = tlink_tcon(cifs_file->tlink);
275 struct cifsInodeInfo *cifsi = CIFS_I(cifs_file->dentry->d_inode);
276 struct cifsLockInfo *li, *tmp;
278 spin_lock(&cifs_file_list_lock);
279 if (--cifs_file->count > 0) {
280 spin_unlock(&cifs_file_list_lock);
284 /* remove it from the lists */
285 list_del(&cifs_file->flist);
286 list_del(&cifs_file->tlist);
288 if (list_empty(&cifsi->openFileList)) {
289 cFYI(1, "closing last open instance for inode %p",
290 cifs_file->dentry->d_inode);
291 cifsi->clientCanCacheRead = false;
292 cifsi->clientCanCacheAll = false;
294 spin_unlock(&cifs_file_list_lock);
296 if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
300 rc = CIFSSMBClose(xid, tcon, cifs_file->netfid);
304 /* Delete any outstanding lock records. We'll lose them when the file
307 mutex_lock(&cifs_file->lock_mutex);
308 list_for_each_entry_safe(li, tmp, &cifs_file->llist, llist) {
309 list_del(&li->llist);
312 mutex_unlock(&cifs_file->lock_mutex);
314 cifs_put_tlink(cifs_file->tlink);
315 dput(cifs_file->dentry);
319 int cifs_open(struct inode *inode, struct file *file)
324 struct cifs_sb_info *cifs_sb;
325 struct cifsTconInfo *tcon;
326 struct tcon_link *tlink;
327 struct cifsFileInfo *pCifsFile = NULL;
328 struct cifsInodeInfo *pCifsInode;
329 char *full_path = NULL;
333 FILE_ALL_INFO *buf = NULL;
337 cifs_sb = CIFS_SB(inode->i_sb);
338 tlink = cifs_sb_tlink(cifs_sb);
341 return PTR_ERR(tlink);
343 tcon = tlink_tcon(tlink);
345 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
347 full_path = build_path_from_dentry(file->f_path.dentry);
348 if (full_path == NULL) {
353 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
354 inode, file->f_flags, full_path);
361 if (!tcon->broken_posix_open && tcon->unix_ext &&
362 (tcon->ses->capabilities & CAP_UNIX) &&
363 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
364 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
365 /* can not refresh inode info since size could be stale */
366 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
367 cifs_sb->mnt_file_mode /* ignored */,
368 file->f_flags, &oplock, &netfid, xid);
370 cFYI(1, "posix open succeeded");
372 pCifsFile = cifs_new_fileinfo(netfid, file, tlink,
374 if (pCifsFile == NULL) {
375 CIFSSMBClose(xid, tcon, netfid);
379 cifs_fscache_set_inode_cookie(inode, file);
382 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
383 if (tcon->ses->serverNOS)
384 cERROR(1, "server %s of type %s returned"
385 " unexpected error on SMB posix open"
386 ", disabling posix open support."
387 " Check if server update available.",
388 tcon->ses->serverName,
389 tcon->ses->serverNOS);
390 tcon->broken_posix_open = true;
391 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
392 (rc != -EOPNOTSUPP)) /* path not found or net err */
394 /* else fallthrough to retry open the old way on network i/o
398 desiredAccess = cifs_convert_flags(file->f_flags);
400 /*********************************************************************
401 * open flag mapping table:
403 * POSIX Flag CIFS Disposition
404 * ---------- ----------------
405 * O_CREAT FILE_OPEN_IF
406 * O_CREAT | O_EXCL FILE_CREATE
407 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
408 * O_TRUNC FILE_OVERWRITE
409 * none of the above FILE_OPEN
411 * Note that there is not a direct match between disposition
412 * FILE_SUPERSEDE (ie create whether or not file exists although
413 * O_CREAT | O_TRUNC is similar but truncates the existing
414 * file rather than creating a new file as FILE_SUPERSEDE does
415 * (which uses the attributes / metadata passed in on open call)
417 *? O_SYNC is a reasonable match to CIFS writethrough flag
418 *? and the read write flags match reasonably. O_LARGEFILE
419 *? is irrelevant because largefile support is always used
420 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
421 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
422 *********************************************************************/
424 disposition = cifs_get_disposition(file->f_flags);
426 /* BB pass O_SYNC flag through on file attributes .. BB */
428 /* Also refresh inode by passing in file_info buf returned by SMBOpen
429 and calling get_inode_info with returned buf (at least helps
430 non-Unix server case) */
432 /* BB we can not do this if this is the second open of a file
433 and the first handle has writebehind data, we might be
434 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
435 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
441 if (tcon->ses->capabilities & CAP_NT_SMBS)
442 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
443 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
444 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
445 & CIFS_MOUNT_MAP_SPECIAL_CHR);
447 rc = -EIO; /* no NT SMB support fall into legacy open below */
450 /* Old server, try legacy style OpenX */
451 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
452 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
453 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
454 & CIFS_MOUNT_MAP_SPECIAL_CHR);
457 cFYI(1, "cifs_open returned 0x%x", rc);
461 rc = cifs_open_inode_helper(inode, tcon, oplock, buf, full_path, xid);
465 pCifsFile = cifs_new_fileinfo(netfid, file, tlink, oplock);
466 if (pCifsFile == NULL) {
471 cifs_fscache_set_inode_cookie(inode, file);
473 if (oplock & CIFS_CREATE_ACTION) {
474 /* time to set mode which we can not set earlier due to
475 problems creating new read-only files */
476 if (tcon->unix_ext) {
477 struct cifs_unix_set_info_args args = {
478 .mode = inode->i_mode,
481 .ctime = NO_CHANGE_64,
482 .atime = NO_CHANGE_64,
483 .mtime = NO_CHANGE_64,
486 CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
488 cifs_sb->mnt_cifs_flags &
489 CIFS_MOUNT_MAP_SPECIAL_CHR);
497 cifs_put_tlink(tlink);
501 /* Try to reacquire byte range locks that were released when session */
502 /* to server was lost */
503 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
507 /* BB list all locks open on this file and relock */
512 static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
517 struct cifs_sb_info *cifs_sb;
518 struct cifsTconInfo *tcon;
519 struct cifsInodeInfo *pCifsInode;
521 char *full_path = NULL;
523 int disposition = FILE_OPEN;
527 mutex_lock(&pCifsFile->fh_mutex);
528 if (!pCifsFile->invalidHandle) {
529 mutex_unlock(&pCifsFile->fh_mutex);
535 inode = pCifsFile->dentry->d_inode;
536 cifs_sb = CIFS_SB(inode->i_sb);
537 tcon = tlink_tcon(pCifsFile->tlink);
539 /* can not grab rename sem here because various ops, including
540 those that already have the rename sem can end up causing writepage
541 to get called and if the server was down that means we end up here,
542 and we can never tell if the caller already has the rename_sem */
543 full_path = build_path_from_dentry(pCifsFile->dentry);
544 if (full_path == NULL) {
546 mutex_unlock(&pCifsFile->fh_mutex);
551 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
552 inode, pCifsFile->f_flags, full_path);
559 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
560 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
561 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
564 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
565 * original open. Must mask them off for a reopen.
567 unsigned int oflags = pCifsFile->f_flags &
568 ~(O_CREAT | O_EXCL | O_TRUNC);
570 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
571 cifs_sb->mnt_file_mode /* ignored */,
572 oflags, &oplock, &netfid, xid);
574 cFYI(1, "posix reopen succeeded");
577 /* fallthrough to retry open the old way on errors, especially
578 in the reconnect path it is important to retry hard */
581 desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
583 /* Can not refresh inode by passing in file_info buf to be returned
584 by SMBOpen and then calling get_inode_info with returned buf
585 since file might have write behind data that needs to be flushed
586 and server version of file size can be stale. If we knew for sure
587 that inode was not dirty locally we could do this */
589 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
590 CREATE_NOT_DIR, &netfid, &oplock, NULL,
591 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
592 CIFS_MOUNT_MAP_SPECIAL_CHR);
594 mutex_unlock(&pCifsFile->fh_mutex);
595 cFYI(1, "cifs_open returned 0x%x", rc);
596 cFYI(1, "oplock: %d", oplock);
597 goto reopen_error_exit;
601 pCifsFile->netfid = netfid;
602 pCifsFile->invalidHandle = false;
603 mutex_unlock(&pCifsFile->fh_mutex);
604 pCifsInode = CIFS_I(inode);
607 rc = filemap_write_and_wait(inode->i_mapping);
608 mapping_set_error(inode->i_mapping, rc);
610 pCifsInode->clientCanCacheAll = false;
611 pCifsInode->clientCanCacheRead = false;
613 rc = cifs_get_inode_info_unix(&inode,
614 full_path, inode->i_sb, xid);
616 rc = cifs_get_inode_info(&inode,
617 full_path, NULL, inode->i_sb,
619 } /* else we are writing out data to server already
620 and could deadlock if we tried to flush data, and
621 since we do not know if we have data that would
622 invalidate the current end of file on the server
623 we can not go to the server to get the new inod
625 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
626 pCifsInode->clientCanCacheAll = true;
627 pCifsInode->clientCanCacheRead = true;
628 cFYI(1, "Exclusive Oplock granted on inode %p",
629 pCifsFile->dentry->d_inode);
630 } else if ((oplock & 0xF) == OPLOCK_READ) {
631 pCifsInode->clientCanCacheRead = true;
632 pCifsInode->clientCanCacheAll = false;
634 pCifsInode->clientCanCacheRead = false;
635 pCifsInode->clientCanCacheAll = false;
637 cifs_relock_file(pCifsFile);
645 int cifs_close(struct inode *inode, struct file *file)
647 cifsFileInfo_put(file->private_data);
648 file->private_data = NULL;
650 /* return code from the ->release op is always ignored */
654 int cifs_closedir(struct inode *inode, struct file *file)
658 struct cifsFileInfo *pCFileStruct = file->private_data;
661 cFYI(1, "Closedir inode = 0x%p", inode);
666 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
668 cFYI(1, "Freeing private data in close dir");
669 spin_lock(&cifs_file_list_lock);
670 if (!pCFileStruct->srch_inf.endOfSearch &&
671 !pCFileStruct->invalidHandle) {
672 pCFileStruct->invalidHandle = true;
673 spin_unlock(&cifs_file_list_lock);
674 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
675 cFYI(1, "Closing uncompleted readdir with rc %d",
677 /* not much we can do if it fails anyway, ignore rc */
680 spin_unlock(&cifs_file_list_lock);
681 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
683 cFYI(1, "closedir free smb buf in srch struct");
684 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
685 if (pCFileStruct->srch_inf.smallBuf)
686 cifs_small_buf_release(ptmp);
688 cifs_buf_release(ptmp);
690 cifs_put_tlink(pCFileStruct->tlink);
691 kfree(file->private_data);
692 file->private_data = NULL;
694 /* BB can we lock the filestruct while this is going on? */
699 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
700 __u64 offset, __u8 lockType)
702 struct cifsLockInfo *li =
703 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
709 mutex_lock(&fid->lock_mutex);
710 list_add(&li->llist, &fid->llist);
711 mutex_unlock(&fid->lock_mutex);
715 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
721 bool wait_flag = false;
722 struct cifs_sb_info *cifs_sb;
723 struct cifsTconInfo *tcon;
725 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
726 bool posix_locking = 0;
728 length = 1 + pfLock->fl_end - pfLock->fl_start;
732 cFYI(1, "Lock parm: 0x%x flockflags: "
733 "0x%x flocktype: 0x%x start: %lld end: %lld",
734 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
737 if (pfLock->fl_flags & FL_POSIX)
739 if (pfLock->fl_flags & FL_FLOCK)
741 if (pfLock->fl_flags & FL_SLEEP) {
742 cFYI(1, "Blocking lock");
745 if (pfLock->fl_flags & FL_ACCESS)
746 cFYI(1, "Process suspended by mandatory locking - "
747 "not implemented yet");
748 if (pfLock->fl_flags & FL_LEASE)
749 cFYI(1, "Lease on file - not implemented yet");
750 if (pfLock->fl_flags &
751 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
752 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
754 if (pfLock->fl_type == F_WRLCK) {
757 } else if (pfLock->fl_type == F_UNLCK) {
760 /* Check if unlock includes more than
762 } else if (pfLock->fl_type == F_RDLCK) {
764 lockType |= LOCKING_ANDX_SHARED_LOCK;
766 } else if (pfLock->fl_type == F_EXLCK) {
769 } else if (pfLock->fl_type == F_SHLCK) {
771 lockType |= LOCKING_ANDX_SHARED_LOCK;
774 cFYI(1, "Unknown type of lock");
776 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
777 tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
779 if (file->private_data == NULL) {
784 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
786 if ((tcon->ses->capabilities & CAP_UNIX) &&
787 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
788 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
790 /* BB add code here to normalize offset and length to
791 account for negative length which we can not accept over the
796 if (lockType & LOCKING_ANDX_SHARED_LOCK)
797 posix_lock_type = CIFS_RDLCK;
799 posix_lock_type = CIFS_WRLCK;
800 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
802 posix_lock_type, wait_flag);
807 /* BB we could chain these into one lock request BB */
808 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
809 0, 1, lockType, 0 /* wait flag */ );
811 rc = CIFSSMBLock(xid, tcon, netfid, length,
812 pfLock->fl_start, 1 /* numUnlock */ ,
813 0 /* numLock */ , lockType,
815 pfLock->fl_type = F_UNLCK;
817 cERROR(1, "Error unlocking previously locked "
818 "range %d during test of lock", rc);
822 /* if rc == ERR_SHARING_VIOLATION ? */
825 if (lockType & LOCKING_ANDX_SHARED_LOCK) {
826 pfLock->fl_type = F_WRLCK;
828 rc = CIFSSMBLock(xid, tcon, netfid, length,
829 pfLock->fl_start, 0, 1,
830 lockType | LOCKING_ANDX_SHARED_LOCK,
833 rc = CIFSSMBLock(xid, tcon, netfid,
834 length, pfLock->fl_start, 1, 0,
836 LOCKING_ANDX_SHARED_LOCK,
838 pfLock->fl_type = F_RDLCK;
840 cERROR(1, "Error unlocking "
841 "previously locked range %d "
842 "during test of lock", rc);
845 pfLock->fl_type = F_WRLCK;
855 if (!numLock && !numUnlock) {
856 /* if no lock or unlock then nothing
857 to do since we do not know what it is */
864 if (lockType & LOCKING_ANDX_SHARED_LOCK)
865 posix_lock_type = CIFS_RDLCK;
867 posix_lock_type = CIFS_WRLCK;
870 posix_lock_type = CIFS_UNLCK;
872 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
874 posix_lock_type, wait_flag);
876 struct cifsFileInfo *fid = file->private_data;
879 rc = CIFSSMBLock(xid, tcon, netfid, length,
881 0, numLock, lockType, wait_flag);
884 /* For Windows locks we must store them. */
885 rc = store_file_lock(fid, length,
886 pfLock->fl_start, lockType);
888 } else if (numUnlock) {
889 /* For each stored lock that this unlock overlaps
890 completely, unlock it. */
892 struct cifsLockInfo *li, *tmp;
895 mutex_lock(&fid->lock_mutex);
896 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
897 if (pfLock->fl_start <= li->offset &&
898 (pfLock->fl_start + length) >=
899 (li->offset + li->length)) {
900 stored_rc = CIFSSMBLock(xid, tcon,
902 li->length, li->offset,
903 1, 0, li->type, false);
907 list_del(&li->llist);
912 mutex_unlock(&fid->lock_mutex);
916 if (pfLock->fl_flags & FL_POSIX)
917 posix_lock_file_wait(file, pfLock);
923 * Set the timeout on write requests past EOF. For some servers (Windows)
924 * these calls can be very long.
926 * If we're writing >10M past the EOF we give a 180s timeout. Anything less
927 * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
928 * The 10M cutoff is totally arbitrary. A better scheme for this would be
929 * welcome if someone wants to suggest one.
931 * We may be able to do a better job with this if there were some way to
932 * declare that a file should be sparse.
935 cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
937 if (offset <= cifsi->server_eof)
939 else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
940 return CIFS_VLONG_OP;
945 /* update the file size (if needed) after a write */
947 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
948 unsigned int bytes_written)
950 loff_t end_of_write = offset + bytes_written;
952 if (end_of_write > cifsi->server_eof)
953 cifsi->server_eof = end_of_write;
956 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
957 size_t write_size, loff_t *poffset)
960 unsigned int bytes_written = 0;
961 unsigned int total_written;
962 struct cifs_sb_info *cifs_sb;
963 struct cifsTconInfo *pTcon;
965 struct cifsFileInfo *open_file;
966 struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
968 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
970 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
971 *poffset, file->f_path.dentry->d_name.name); */
973 if (file->private_data == NULL)
976 open_file = file->private_data;
977 pTcon = tlink_tcon(open_file->tlink);
979 rc = generic_write_checks(file, poffset, &write_size, 0);
985 long_op = cifs_write_timeout(cifsi, *poffset);
986 for (total_written = 0; write_size > total_written;
987 total_written += bytes_written) {
989 while (rc == -EAGAIN) {
990 if (file->private_data == NULL) {
991 /* file has been closed on us */
993 /* if we have gotten here we have written some data
994 and blocked, and the file has been freed on us while
995 we blocked so return what we managed to write */
996 return total_written;
998 if (open_file->invalidHandle) {
999 /* we could deadlock if we called
1000 filemap_fdatawait from here so tell
1001 reopen_file not to flush data to server
1003 rc = cifs_reopen_file(open_file, false);
1008 rc = CIFSSMBWrite(xid, pTcon,
1010 min_t(const int, cifs_sb->wsize,
1011 write_size - total_written),
1012 *poffset, &bytes_written,
1013 NULL, write_data + total_written, long_op);
1015 if (rc || (bytes_written == 0)) {
1023 cifs_update_eof(cifsi, *poffset, bytes_written);
1024 *poffset += bytes_written;
1026 long_op = CIFS_STD_OP; /* subsequent writes fast -
1027 15 seconds is plenty */
1030 cifs_stats_bytes_written(pTcon, total_written);
1032 /* since the write may have blocked check these pointers again */
1033 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1034 struct inode *inode = file->f_path.dentry->d_inode;
1035 /* Do not update local mtime - server will set its actual value on write
1036 * inode->i_ctime = inode->i_mtime =
1037 * current_fs_time(inode->i_sb);*/
1038 if (total_written > 0) {
1039 spin_lock(&inode->i_lock);
1040 if (*poffset > file->f_path.dentry->d_inode->i_size)
1041 i_size_write(file->f_path.dentry->d_inode,
1043 spin_unlock(&inode->i_lock);
1045 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1048 return total_written;
1051 static ssize_t cifs_write(struct cifsFileInfo *open_file,
1052 const char *write_data, size_t write_size,
1056 unsigned int bytes_written = 0;
1057 unsigned int total_written;
1058 struct cifs_sb_info *cifs_sb;
1059 struct cifsTconInfo *pTcon;
1061 struct dentry *dentry = open_file->dentry;
1062 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
1064 cifs_sb = CIFS_SB(dentry->d_sb);
1066 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
1067 *poffset, dentry->d_name.name);
1069 pTcon = tlink_tcon(open_file->tlink);
1073 long_op = cifs_write_timeout(cifsi, *poffset);
1074 for (total_written = 0; write_size > total_written;
1075 total_written += bytes_written) {
1077 while (rc == -EAGAIN) {
1078 if (open_file->invalidHandle) {
1079 /* we could deadlock if we called
1080 filemap_fdatawait from here so tell
1081 reopen_file not to flush data to
1083 rc = cifs_reopen_file(open_file, false);
1087 if (experimEnabled || (pTcon->ses->server &&
1088 ((pTcon->ses->server->secMode &
1089 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1094 len = min((size_t)cifs_sb->wsize,
1095 write_size - total_written);
1096 /* iov[0] is reserved for smb header */
1097 iov[1].iov_base = (char *)write_data +
1099 iov[1].iov_len = len;
1100 rc = CIFSSMBWrite2(xid, pTcon,
1101 open_file->netfid, len,
1102 *poffset, &bytes_written,
1105 rc = CIFSSMBWrite(xid, pTcon,
1107 min_t(const int, cifs_sb->wsize,
1108 write_size - total_written),
1109 *poffset, &bytes_written,
1110 write_data + total_written,
1113 if (rc || (bytes_written == 0)) {
1121 cifs_update_eof(cifsi, *poffset, bytes_written);
1122 *poffset += bytes_written;
1124 long_op = CIFS_STD_OP; /* subsequent writes fast -
1125 15 seconds is plenty */
1128 cifs_stats_bytes_written(pTcon, total_written);
1130 if (total_written > 0) {
1131 spin_lock(&dentry->d_inode->i_lock);
1132 if (*poffset > dentry->d_inode->i_size)
1133 i_size_write(dentry->d_inode, *poffset);
1134 spin_unlock(&dentry->d_inode->i_lock);
1136 mark_inode_dirty_sync(dentry->d_inode);
1138 return total_written;
1141 #ifdef CONFIG_CIFS_EXPERIMENTAL
1142 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1145 struct cifsFileInfo *open_file = NULL;
1146 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1148 /* only filter by fsuid on multiuser mounts */
1149 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1152 spin_lock(&cifs_file_list_lock);
1153 /* we could simply get the first_list_entry since write-only entries
1154 are always at the end of the list but since the first entry might
1155 have a close pending, we go through the whole list */
1156 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1157 if (fsuid_only && open_file->uid != current_fsuid())
1159 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1160 if (!open_file->invalidHandle) {
1161 /* found a good file */
1162 /* lock it so it will not be closed on us */
1163 cifsFileInfo_get(open_file);
1164 spin_unlock(&cifs_file_list_lock);
1166 } /* else might as well continue, and look for
1167 another, or simply have the caller reopen it
1168 again rather than trying to fix this handle */
1169 } else /* write only file */
1170 break; /* write only files are last so must be done */
1172 spin_unlock(&cifs_file_list_lock);
1177 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1180 struct cifsFileInfo *open_file;
1181 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1182 bool any_available = false;
1185 /* Having a null inode here (because mapping->host was set to zero by
1186 the VFS or MM) should not happen but we had reports of on oops (due to
1187 it being zero) during stress testcases so we need to check for it */
1189 if (cifs_inode == NULL) {
1190 cERROR(1, "Null inode passed to cifs_writeable_file");
1195 /* only filter by fsuid on multiuser mounts */
1196 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1199 spin_lock(&cifs_file_list_lock);
1201 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1202 if (!any_available && open_file->pid != current->tgid)
1204 if (fsuid_only && open_file->uid != current_fsuid())
1206 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1207 cifsFileInfo_get(open_file);
1209 if (!open_file->invalidHandle) {
1210 /* found a good writable file */
1211 spin_unlock(&cifs_file_list_lock);
1215 spin_unlock(&cifs_file_list_lock);
1217 /* Had to unlock since following call can block */
1218 rc = cifs_reopen_file(open_file, false);
1222 /* if it fails, try another handle if possible */
1223 cFYI(1, "wp failed on reopen file");
1224 cifsFileInfo_put(open_file);
1226 spin_lock(&cifs_file_list_lock);
1228 /* else we simply continue to the next entry. Thus
1229 we do not loop on reopen errors. If we
1230 can not reopen the file, for example if we
1231 reconnected to a server with another client
1232 racing to delete or lock the file we would not
1233 make progress if we restarted before the beginning
1234 of the loop here. */
1237 /* couldn't find useable FH with same pid, try any available */
1238 if (!any_available) {
1239 any_available = true;
1240 goto refind_writable;
1242 spin_unlock(&cifs_file_list_lock);
1246 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1248 struct address_space *mapping = page->mapping;
1249 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1252 int bytes_written = 0;
1253 struct cifs_sb_info *cifs_sb;
1254 struct inode *inode;
1255 struct cifsFileInfo *open_file;
1257 if (!mapping || !mapping->host)
1260 inode = page->mapping->host;
1261 cifs_sb = CIFS_SB(inode->i_sb);
1263 offset += (loff_t)from;
1264 write_data = kmap(page);
1267 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1272 /* racing with truncate? */
1273 if (offset > mapping->host->i_size) {
1275 return 0; /* don't care */
1278 /* check to make sure that we are not extending the file */
1279 if (mapping->host->i_size - offset < (loff_t)to)
1280 to = (unsigned)(mapping->host->i_size - offset);
1282 open_file = find_writable_file(CIFS_I(mapping->host), false);
1284 bytes_written = cifs_write(open_file, write_data,
1285 to - from, &offset);
1286 cifsFileInfo_put(open_file);
1287 /* Does mm or vfs already set times? */
1288 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1289 if ((bytes_written > 0) && (offset))
1291 else if (bytes_written < 0)
1294 cFYI(1, "No writeable filehandles for inode");
1302 static int cifs_writepages(struct address_space *mapping,
1303 struct writeback_control *wbc)
1305 struct backing_dev_info *bdi = mapping->backing_dev_info;
1306 unsigned int bytes_to_write;
1307 unsigned int bytes_written;
1308 struct cifs_sb_info *cifs_sb;
1312 int range_whole = 0;
1319 struct cifsFileInfo *open_file;
1320 struct cifsTconInfo *tcon;
1321 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1323 struct pagevec pvec;
1329 * BB: Is this meaningful for a non-block-device file system?
1330 * If it is, we should test it again after we do I/O
1332 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1333 wbc->encountered_congestion = 1;
1337 cifs_sb = CIFS_SB(mapping->host->i_sb);
1340 * If wsize is smaller that the page cache size, default to writing
1341 * one page at a time via cifs_writepage
1343 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1344 return generic_writepages(mapping, wbc);
1346 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1348 return generic_writepages(mapping, wbc);
1351 * if there's no open file, then this is likely to fail too,
1352 * but it'll at least handle the return. Maybe it should be
1355 open_file = find_writable_file(CIFS_I(mapping->host), false);
1358 return generic_writepages(mapping, wbc);
1361 tcon = tlink_tcon(open_file->tlink);
1362 if (!experimEnabled && tcon->ses->server->secMode &
1363 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1364 cifsFileInfo_put(open_file);
1365 return generic_writepages(mapping, wbc);
1367 cifsFileInfo_put(open_file);
1371 pagevec_init(&pvec, 0);
1372 if (wbc->range_cyclic) {
1373 index = mapping->writeback_index; /* Start from prev offset */
1376 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1377 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1378 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1383 while (!done && (index <= end) &&
1384 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1385 PAGECACHE_TAG_DIRTY,
1386 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1395 for (i = 0; i < nr_pages; i++) {
1396 page = pvec.pages[i];
1398 * At this point we hold neither mapping->tree_lock nor
1399 * lock on the page itself: the page may be truncated or
1400 * invalidated (changing page->mapping to NULL), or even
1401 * swizzled back from swapper_space to tmpfs file
1407 else if (!trylock_page(page))
1410 if (unlikely(page->mapping != mapping)) {
1415 if (!wbc->range_cyclic && page->index > end) {
1421 if (next && (page->index != next)) {
1422 /* Not next consecutive page */
1427 if (wbc->sync_mode != WB_SYNC_NONE)
1428 wait_on_page_writeback(page);
1430 if (PageWriteback(page) ||
1431 !clear_page_dirty_for_io(page)) {
1437 * This actually clears the dirty bit in the radix tree.
1438 * See cifs_writepage() for more commentary.
1440 set_page_writeback(page);
1442 if (page_offset(page) >= mapping->host->i_size) {
1445 end_page_writeback(page);
1450 * BB can we get rid of this? pages are held by pvec
1452 page_cache_get(page);
1454 len = min(mapping->host->i_size - page_offset(page),
1455 (loff_t)PAGE_CACHE_SIZE);
1457 /* reserve iov[0] for the smb header */
1459 iov[n_iov].iov_base = kmap(page);
1460 iov[n_iov].iov_len = len;
1461 bytes_to_write += len;
1465 offset = page_offset(page);
1467 next = page->index + 1;
1468 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1472 open_file = find_writable_file(CIFS_I(mapping->host),
1475 cERROR(1, "No writable handles for inode");
1478 long_op = cifs_write_timeout(cifsi, offset);
1479 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1480 bytes_to_write, offset,
1481 &bytes_written, iov, n_iov,
1483 cifsFileInfo_put(open_file);
1484 cifs_update_eof(cifsi, offset, bytes_written);
1487 if (rc || bytes_written < bytes_to_write) {
1488 cERROR(1, "Write2 ret %d, wrote %d",
1490 mapping_set_error(mapping, rc);
1492 cifs_stats_bytes_written(tcon, bytes_written);
1495 for (i = 0; i < n_iov; i++) {
1496 page = pvec.pages[first + i];
1497 /* Should we also set page error on
1498 success rc but too little data written? */
1499 /* BB investigate retry logic on temporary
1500 server crash cases and how recovery works
1501 when page marked as error */
1506 end_page_writeback(page);
1507 page_cache_release(page);
1509 if ((wbc->nr_to_write -= n_iov) <= 0)
1513 /* Need to re-find the pages we skipped */
1514 index = pvec.pages[0]->index + 1;
1516 pagevec_release(&pvec);
1518 if (!scanned && !done) {
1520 * We hit the last page and there is more work to be done: wrap
1521 * back to the start of the file
1527 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1528 mapping->writeback_index = index;
1535 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1541 /* BB add check for wbc flags */
1542 page_cache_get(page);
1543 if (!PageUptodate(page))
1544 cFYI(1, "ppw - page not up to date");
1547 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1549 * A writepage() implementation always needs to do either this,
1550 * or re-dirty the page with "redirty_page_for_writepage()" in
1551 * the case of a failure.
1553 * Just unlocking the page will cause the radix tree tag-bits
1554 * to fail to update with the state of the page correctly.
1556 set_page_writeback(page);
1557 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1558 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1560 end_page_writeback(page);
1561 page_cache_release(page);
1566 static int cifs_write_end(struct file *file, struct address_space *mapping,
1567 loff_t pos, unsigned len, unsigned copied,
1568 struct page *page, void *fsdata)
1571 struct inode *inode = mapping->host;
1573 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1576 if (PageChecked(page)) {
1578 SetPageUptodate(page);
1579 ClearPageChecked(page);
1580 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1581 SetPageUptodate(page);
1583 if (!PageUptodate(page)) {
1585 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1589 /* this is probably better than directly calling
1590 partialpage_write since in this function the file handle is
1591 known which we might as well leverage */
1592 /* BB check if anything else missing out of ppw
1593 such as updating last write time */
1594 page_data = kmap(page);
1595 rc = cifs_write(file->private_data, page_data + offset,
1597 /* if (rc < 0) should we set writebehind rc? */
1604 set_page_dirty(page);
1608 spin_lock(&inode->i_lock);
1609 if (pos > inode->i_size)
1610 i_size_write(inode, pos);
1611 spin_unlock(&inode->i_lock);
1615 page_cache_release(page);
1620 int cifs_fsync(struct file *file, int datasync)
1624 struct cifsTconInfo *tcon;
1625 struct cifsFileInfo *smbfile = file->private_data;
1626 struct inode *inode = file->f_path.dentry->d_inode;
1630 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1631 file->f_path.dentry->d_name.name, datasync);
1633 rc = filemap_write_and_wait(inode->i_mapping);
1635 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
1637 tcon = tlink_tcon(smbfile->tlink);
1638 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1639 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1646 /* static void cifs_sync_page(struct page *page)
1648 struct address_space *mapping;
1649 struct inode *inode;
1650 unsigned long index = page->index;
1651 unsigned int rpages = 0;
1654 cFYI(1, "sync page %p", page);
1655 mapping = page->mapping;
1658 inode = mapping->host;
1662 /* fill in rpages then
1663 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1665 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1675 * As file closes, flush all cached write data for this inode checking
1676 * for write behind errors.
1678 int cifs_flush(struct file *file, fl_owner_t id)
1680 struct inode *inode = file->f_path.dentry->d_inode;
1683 if (file->f_mode & FMODE_WRITE)
1684 rc = filemap_write_and_wait(inode->i_mapping);
1686 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1691 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1692 size_t read_size, loff_t *poffset)
1695 unsigned int bytes_read = 0;
1696 unsigned int total_read = 0;
1697 unsigned int current_read_size;
1698 struct cifs_sb_info *cifs_sb;
1699 struct cifsTconInfo *pTcon;
1701 struct cifsFileInfo *open_file;
1702 char *smb_read_data;
1703 char __user *current_offset;
1704 struct smb_com_read_rsp *pSMBr;
1707 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1709 if (file->private_data == NULL) {
1714 open_file = file->private_data;
1715 pTcon = tlink_tcon(open_file->tlink);
1717 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1718 cFYI(1, "attempting read on write only file instance");
1720 for (total_read = 0, current_offset = read_data;
1721 read_size > total_read;
1722 total_read += bytes_read, current_offset += bytes_read) {
1723 current_read_size = min_t(const int, read_size - total_read,
1726 smb_read_data = NULL;
1727 while (rc == -EAGAIN) {
1728 int buf_type = CIFS_NO_BUFFER;
1729 if (open_file->invalidHandle) {
1730 rc = cifs_reopen_file(open_file, true);
1734 rc = CIFSSMBRead(xid, pTcon,
1736 current_read_size, *poffset,
1737 &bytes_read, &smb_read_data,
1739 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1740 if (smb_read_data) {
1741 if (copy_to_user(current_offset,
1743 4 /* RFC1001 length field */ +
1744 le16_to_cpu(pSMBr->DataOffset),
1748 if (buf_type == CIFS_SMALL_BUFFER)
1749 cifs_small_buf_release(smb_read_data);
1750 else if (buf_type == CIFS_LARGE_BUFFER)
1751 cifs_buf_release(smb_read_data);
1752 smb_read_data = NULL;
1755 if (rc || (bytes_read == 0)) {
1763 cifs_stats_bytes_read(pTcon, bytes_read);
1764 *poffset += bytes_read;
1772 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1776 unsigned int bytes_read = 0;
1777 unsigned int total_read;
1778 unsigned int current_read_size;
1779 struct cifs_sb_info *cifs_sb;
1780 struct cifsTconInfo *pTcon;
1782 char *current_offset;
1783 struct cifsFileInfo *open_file;
1784 int buf_type = CIFS_NO_BUFFER;
1787 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1789 if (file->private_data == NULL) {
1794 open_file = file->private_data;
1795 pTcon = tlink_tcon(open_file->tlink);
1797 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1798 cFYI(1, "attempting read on write only file instance");
1800 for (total_read = 0, current_offset = read_data;
1801 read_size > total_read;
1802 total_read += bytes_read, current_offset += bytes_read) {
1803 current_read_size = min_t(const int, read_size - total_read,
1805 /* For windows me and 9x we do not want to request more
1806 than it negotiated since it will refuse the read then */
1808 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1809 current_read_size = min_t(const int, current_read_size,
1810 pTcon->ses->server->maxBuf - 128);
1813 while (rc == -EAGAIN) {
1814 if (open_file->invalidHandle) {
1815 rc = cifs_reopen_file(open_file, true);
1819 rc = CIFSSMBRead(xid, pTcon,
1821 current_read_size, *poffset,
1822 &bytes_read, ¤t_offset,
1825 if (rc || (bytes_read == 0)) {
1833 cifs_stats_bytes_read(pTcon, total_read);
1834 *poffset += bytes_read;
1841 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1846 rc = cifs_revalidate_file(file);
1848 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1852 rc = generic_file_mmap(file, vma);
1858 static void cifs_copy_cache_pages(struct address_space *mapping,
1859 struct list_head *pages, int bytes_read, char *data)
1864 while (bytes_read > 0) {
1865 if (list_empty(pages))
1868 page = list_entry(pages->prev, struct page, lru);
1869 list_del(&page->lru);
1871 if (add_to_page_cache_lru(page, mapping, page->index,
1873 page_cache_release(page);
1874 cFYI(1, "Add page cache failed");
1875 data += PAGE_CACHE_SIZE;
1876 bytes_read -= PAGE_CACHE_SIZE;
1879 page_cache_release(page);
1881 target = kmap_atomic(page, KM_USER0);
1883 if (PAGE_CACHE_SIZE > bytes_read) {
1884 memcpy(target, data, bytes_read);
1885 /* zero the tail end of this partial page */
1886 memset(target + bytes_read, 0,
1887 PAGE_CACHE_SIZE - bytes_read);
1890 memcpy(target, data, PAGE_CACHE_SIZE);
1891 bytes_read -= PAGE_CACHE_SIZE;
1893 kunmap_atomic(target, KM_USER0);
1895 flush_dcache_page(page);
1896 SetPageUptodate(page);
1898 data += PAGE_CACHE_SIZE;
1900 /* add page to FS-Cache */
1901 cifs_readpage_to_fscache(mapping->host, page);
1906 static int cifs_readpages(struct file *file, struct address_space *mapping,
1907 struct list_head *page_list, unsigned num_pages)
1913 struct cifs_sb_info *cifs_sb;
1914 struct cifsTconInfo *pTcon;
1915 unsigned int bytes_read = 0;
1916 unsigned int read_size, i;
1917 char *smb_read_data = NULL;
1918 struct smb_com_read_rsp *pSMBr;
1919 struct cifsFileInfo *open_file;
1920 int buf_type = CIFS_NO_BUFFER;
1923 if (file->private_data == NULL) {
1928 open_file = file->private_data;
1929 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1930 pTcon = tlink_tcon(open_file->tlink);
1933 * Reads as many pages as possible from fscache. Returns -ENOBUFS
1934 * immediately if the cookie is negative
1936 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
1941 cFYI(DBG2, "rpages: num pages %d", num_pages);
1942 for (i = 0; i < num_pages; ) {
1943 unsigned contig_pages;
1944 struct page *tmp_page;
1945 unsigned long expected_index;
1947 if (list_empty(page_list))
1950 page = list_entry(page_list->prev, struct page, lru);
1951 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1953 /* count adjacent pages that we will read into */
1956 list_entry(page_list->prev, struct page, lru)->index;
1957 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1958 if (tmp_page->index == expected_index) {
1964 if (contig_pages + i > num_pages)
1965 contig_pages = num_pages - i;
1967 /* for reads over a certain size could initiate async
1970 read_size = contig_pages * PAGE_CACHE_SIZE;
1971 /* Read size needs to be in multiples of one page */
1972 read_size = min_t(const unsigned int, read_size,
1973 cifs_sb->rsize & PAGE_CACHE_MASK);
1974 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
1975 read_size, contig_pages);
1977 while (rc == -EAGAIN) {
1978 if (open_file->invalidHandle) {
1979 rc = cifs_reopen_file(open_file, true);
1984 rc = CIFSSMBRead(xid, pTcon,
1987 &bytes_read, &smb_read_data,
1989 /* BB more RC checks ? */
1990 if (rc == -EAGAIN) {
1991 if (smb_read_data) {
1992 if (buf_type == CIFS_SMALL_BUFFER)
1993 cifs_small_buf_release(smb_read_data);
1994 else if (buf_type == CIFS_LARGE_BUFFER)
1995 cifs_buf_release(smb_read_data);
1996 smb_read_data = NULL;
2000 if ((rc < 0) || (smb_read_data == NULL)) {
2001 cFYI(1, "Read error in readpages: %d", rc);
2003 } else if (bytes_read > 0) {
2004 task_io_account_read(bytes_read);
2005 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2006 cifs_copy_cache_pages(mapping, page_list, bytes_read,
2007 smb_read_data + 4 /* RFC1001 hdr */ +
2008 le16_to_cpu(pSMBr->DataOffset));
2010 i += bytes_read >> PAGE_CACHE_SHIFT;
2011 cifs_stats_bytes_read(pTcon, bytes_read);
2012 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2013 i++; /* account for partial page */
2015 /* server copy of file can have smaller size
2017 /* BB do we need to verify this common case ?
2018 this case is ok - if we are at server EOF
2019 we will hit it on next read */
2024 cFYI(1, "No bytes read (%d) at offset %lld . "
2025 "Cleaning remaining pages from readahead list",
2026 bytes_read, offset);
2027 /* BB turn off caching and do new lookup on
2028 file size at server? */
2031 if (smb_read_data) {
2032 if (buf_type == CIFS_SMALL_BUFFER)
2033 cifs_small_buf_release(smb_read_data);
2034 else if (buf_type == CIFS_LARGE_BUFFER)
2035 cifs_buf_release(smb_read_data);
2036 smb_read_data = NULL;
2041 /* need to free smb_read_data buf before exit */
2042 if (smb_read_data) {
2043 if (buf_type == CIFS_SMALL_BUFFER)
2044 cifs_small_buf_release(smb_read_data);
2045 else if (buf_type == CIFS_LARGE_BUFFER)
2046 cifs_buf_release(smb_read_data);
2047 smb_read_data = NULL;
2055 static int cifs_readpage_worker(struct file *file, struct page *page,
2061 /* Is the page cached? */
2062 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2066 page_cache_get(page);
2067 read_data = kmap(page);
2068 /* for reads over a certain size could initiate async read ahead */
2070 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2075 cFYI(1, "Bytes read %d", rc);
2077 file->f_path.dentry->d_inode->i_atime =
2078 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2080 if (PAGE_CACHE_SIZE > rc)
2081 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2083 flush_dcache_page(page);
2084 SetPageUptodate(page);
2086 /* send this page to the cache */
2087 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2093 page_cache_release(page);
2099 static int cifs_readpage(struct file *file, struct page *page)
2101 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2107 if (file->private_data == NULL) {
2113 cFYI(1, "readpage %p at offset %d 0x%x\n",
2114 page, (int)offset, (int)offset);
2116 rc = cifs_readpage_worker(file, page, &offset);
2124 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2126 struct cifsFileInfo *open_file;
2128 spin_lock(&cifs_file_list_lock);
2129 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2130 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2131 spin_unlock(&cifs_file_list_lock);
2135 spin_unlock(&cifs_file_list_lock);
2139 /* We do not want to update the file size from server for inodes
2140 open for write - to avoid races with writepage extending
2141 the file - in the future we could consider allowing
2142 refreshing the inode only on increases in the file size
2143 but this is tricky to do without racing with writebehind
2144 page caching in the current Linux kernel design */
2145 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2150 if (is_inode_writable(cifsInode)) {
2151 /* This inode is open for write at least once */
2152 struct cifs_sb_info *cifs_sb;
2154 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2155 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2156 /* since no page cache to corrupt on directio
2157 we can change size safely */
2161 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2169 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2170 loff_t pos, unsigned len, unsigned flags,
2171 struct page **pagep, void **fsdata)
2173 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2174 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2175 loff_t page_start = pos & PAGE_MASK;
2180 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2182 page = grab_cache_page_write_begin(mapping, index, flags);
2188 if (PageUptodate(page))
2192 * If we write a full page it will be up to date, no need to read from
2193 * the server. If the write is short, we'll end up doing a sync write
2196 if (len == PAGE_CACHE_SIZE)
2200 * optimize away the read when we have an oplock, and we're not
2201 * expecting to use any of the data we'd be reading in. That
2202 * is, when the page lies beyond the EOF, or straddles the EOF
2203 * and the write will cover all of the existing data.
2205 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2206 i_size = i_size_read(mapping->host);
2207 if (page_start >= i_size ||
2208 (offset == 0 && (pos + len) >= i_size)) {
2209 zero_user_segments(page, 0, offset,
2213 * PageChecked means that the parts of the page
2214 * to which we're not writing are considered up
2215 * to date. Once the data is copied to the
2216 * page, it can be set uptodate.
2218 SetPageChecked(page);
2223 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2225 * might as well read a page, it is fast enough. If we get
2226 * an error, we don't need to return it. cifs_write_end will
2227 * do a sync write instead since PG_uptodate isn't set.
2229 cifs_readpage_worker(file, page, &page_start);
2231 /* we could try using another file handle if there is one -
2232 but how would we lock it to prevent close of that handle
2233 racing with this read? In any case
2234 this will be written out by write_end so is fine */
2241 static int cifs_release_page(struct page *page, gfp_t gfp)
2243 if (PagePrivate(page))
2246 return cifs_fscache_release_page(page, gfp);
2249 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2251 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2254 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2257 void cifs_oplock_break(struct work_struct *work)
2259 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2261 struct inode *inode = cfile->dentry->d_inode;
2262 struct cifsInodeInfo *cinode = CIFS_I(inode);
2265 if (inode && S_ISREG(inode->i_mode)) {
2266 if (cinode->clientCanCacheRead)
2267 break_lease(inode, O_RDONLY);
2269 break_lease(inode, O_WRONLY);
2270 rc = filemap_fdatawrite(inode->i_mapping);
2271 if (cinode->clientCanCacheRead == 0) {
2272 rc = filemap_fdatawait(inode->i_mapping);
2273 mapping_set_error(inode->i_mapping, rc);
2274 invalidate_remote_inode(inode);
2276 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2280 * releasing stale oplock after recent reconnect of smb session using
2281 * a now incorrect file handle is not a data integrity issue but do
2282 * not bother sending an oplock release if session to server still is
2283 * disconnected since oplock already released by the server
2285 if (!cfile->oplock_break_cancelled) {
2286 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2287 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false);
2288 cFYI(1, "Oplock release rc = %d", rc);
2292 * We might have kicked in before is_valid_oplock_break()
2293 * finished grabbing reference for us. Make sure it's done by
2294 * waiting for cifs_file_list_lock.
2296 spin_lock(&cifs_file_list_lock);
2297 spin_unlock(&cifs_file_list_lock);
2299 cifs_oplock_break_put(cfile);
2302 /* must be called while holding cifs_file_list_lock */
2303 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2305 cifs_sb_active(cfile->dentry->d_sb);
2306 cifsFileInfo_get(cfile);
2309 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2311 cifsFileInfo_put(cfile);
2312 cifs_sb_deactive(cfile->dentry->d_sb);
2315 const struct address_space_operations cifs_addr_ops = {
2316 .readpage = cifs_readpage,
2317 .readpages = cifs_readpages,
2318 .writepage = cifs_writepage,
2319 .writepages = cifs_writepages,
2320 .write_begin = cifs_write_begin,
2321 .write_end = cifs_write_end,
2322 .set_page_dirty = __set_page_dirty_nobuffers,
2323 .releasepage = cifs_release_page,
2324 .invalidatepage = cifs_invalidate_page,
2325 /* .sync_page = cifs_sync_page, */
2330 * cifs_readpages requires the server to support a buffer large enough to
2331 * contain the header plus one complete page of data. Otherwise, we need
2332 * to leave cifs_readpages out of the address space operations.
2334 const struct address_space_operations cifs_addr_ops_smallbuf = {
2335 .readpage = cifs_readpage,
2336 .writepage = cifs_writepage,
2337 .writepages = cifs_writepages,
2338 .write_begin = cifs_write_begin,
2339 .write_end = cifs_write_end,
2340 .set_page_dirty = __set_page_dirty_nobuffers,
2341 .releasepage = cifs_release_page,
2342 .invalidatepage = cifs_invalidate_page,
2343 /* .sync_page = cifs_sync_page, */