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)
959 struct inode *inode = file->f_path.dentry->d_inode;
961 unsigned int bytes_written = 0;
962 unsigned int total_written;
963 struct cifs_sb_info *cifs_sb;
964 struct cifsTconInfo *pTcon;
966 struct cifsFileInfo *open_file;
967 struct cifsInodeInfo *cifsi = CIFS_I(inode);
969 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
971 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
972 *poffset, file->f_path.dentry->d_name.name); */
974 if (file->private_data == NULL)
977 open_file = file->private_data;
978 pTcon = tlink_tcon(open_file->tlink);
980 rc = generic_write_checks(file, poffset, &write_size, 0);
986 long_op = cifs_write_timeout(cifsi, *poffset);
987 for (total_written = 0; write_size > total_written;
988 total_written += bytes_written) {
990 while (rc == -EAGAIN) {
991 if (file->private_data == NULL) {
992 /* file has been closed on us */
994 /* if we have gotten here we have written some data
995 and blocked, and the file has been freed on us while
996 we blocked so return what we managed to write */
997 return total_written;
999 if (open_file->invalidHandle) {
1000 /* we could deadlock if we called
1001 filemap_fdatawait from here so tell
1002 reopen_file not to flush data to server
1004 rc = cifs_reopen_file(open_file, false);
1009 rc = CIFSSMBWrite(xid, pTcon,
1011 min_t(const int, cifs_sb->wsize,
1012 write_size - total_written),
1013 *poffset, &bytes_written,
1014 NULL, write_data + total_written, long_op);
1016 if (rc || (bytes_written == 0)) {
1024 cifs_update_eof(cifsi, *poffset, bytes_written);
1025 *poffset += bytes_written;
1027 long_op = CIFS_STD_OP; /* subsequent writes fast -
1028 15 seconds is plenty */
1031 cifs_stats_bytes_written(pTcon, total_written);
1033 /* Do not update local mtime - server will set its actual value on write
1034 * inode->i_ctime = inode->i_mtime =
1035 * current_fs_time(inode->i_sb);*/
1036 if (total_written > 0) {
1037 spin_lock(&inode->i_lock);
1038 if (*poffset > inode->i_size)
1039 i_size_write(inode, *poffset);
1040 spin_unlock(&inode->i_lock);
1042 mark_inode_dirty_sync(inode);
1045 return total_written;
1048 static ssize_t cifs_write(struct cifsFileInfo *open_file,
1049 const char *write_data, size_t write_size,
1053 unsigned int bytes_written = 0;
1054 unsigned int total_written;
1055 struct cifs_sb_info *cifs_sb;
1056 struct cifsTconInfo *pTcon;
1058 struct dentry *dentry = open_file->dentry;
1059 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
1061 cifs_sb = CIFS_SB(dentry->d_sb);
1063 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
1064 *poffset, dentry->d_name.name);
1066 pTcon = tlink_tcon(open_file->tlink);
1070 long_op = cifs_write_timeout(cifsi, *poffset);
1071 for (total_written = 0; write_size > total_written;
1072 total_written += bytes_written) {
1074 while (rc == -EAGAIN) {
1075 if (open_file->invalidHandle) {
1076 /* we could deadlock if we called
1077 filemap_fdatawait from here so tell
1078 reopen_file not to flush data to
1080 rc = cifs_reopen_file(open_file, false);
1084 if (experimEnabled || (pTcon->ses->server &&
1085 ((pTcon->ses->server->secMode &
1086 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1091 len = min((size_t)cifs_sb->wsize,
1092 write_size - total_written);
1093 /* iov[0] is reserved for smb header */
1094 iov[1].iov_base = (char *)write_data +
1096 iov[1].iov_len = len;
1097 rc = CIFSSMBWrite2(xid, pTcon,
1098 open_file->netfid, len,
1099 *poffset, &bytes_written,
1102 rc = CIFSSMBWrite(xid, pTcon,
1104 min_t(const int, cifs_sb->wsize,
1105 write_size - total_written),
1106 *poffset, &bytes_written,
1107 write_data + total_written,
1110 if (rc || (bytes_written == 0)) {
1118 cifs_update_eof(cifsi, *poffset, bytes_written);
1119 *poffset += bytes_written;
1121 long_op = CIFS_STD_OP; /* subsequent writes fast -
1122 15 seconds is plenty */
1125 cifs_stats_bytes_written(pTcon, total_written);
1127 if (total_written > 0) {
1128 spin_lock(&dentry->d_inode->i_lock);
1129 if (*poffset > dentry->d_inode->i_size)
1130 i_size_write(dentry->d_inode, *poffset);
1131 spin_unlock(&dentry->d_inode->i_lock);
1133 mark_inode_dirty_sync(dentry->d_inode);
1135 return total_written;
1138 #ifdef CONFIG_CIFS_EXPERIMENTAL
1139 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1142 struct cifsFileInfo *open_file = NULL;
1143 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1145 /* only filter by fsuid on multiuser mounts */
1146 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1149 spin_lock(&cifs_file_list_lock);
1150 /* we could simply get the first_list_entry since write-only entries
1151 are always at the end of the list but since the first entry might
1152 have a close pending, we go through the whole list */
1153 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1154 if (fsuid_only && open_file->uid != current_fsuid())
1156 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1157 if (!open_file->invalidHandle) {
1158 /* found a good file */
1159 /* lock it so it will not be closed on us */
1160 cifsFileInfo_get(open_file);
1161 spin_unlock(&cifs_file_list_lock);
1163 } /* else might as well continue, and look for
1164 another, or simply have the caller reopen it
1165 again rather than trying to fix this handle */
1166 } else /* write only file */
1167 break; /* write only files are last so must be done */
1169 spin_unlock(&cifs_file_list_lock);
1174 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1177 struct cifsFileInfo *open_file;
1178 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1179 bool any_available = false;
1182 /* Having a null inode here (because mapping->host was set to zero by
1183 the VFS or MM) should not happen but we had reports of on oops (due to
1184 it being zero) during stress testcases so we need to check for it */
1186 if (cifs_inode == NULL) {
1187 cERROR(1, "Null inode passed to cifs_writeable_file");
1192 /* only filter by fsuid on multiuser mounts */
1193 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1196 spin_lock(&cifs_file_list_lock);
1198 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1199 if (!any_available && open_file->pid != current->tgid)
1201 if (fsuid_only && open_file->uid != current_fsuid())
1203 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1204 cifsFileInfo_get(open_file);
1206 if (!open_file->invalidHandle) {
1207 /* found a good writable file */
1208 spin_unlock(&cifs_file_list_lock);
1212 spin_unlock(&cifs_file_list_lock);
1214 /* Had to unlock since following call can block */
1215 rc = cifs_reopen_file(open_file, false);
1219 /* if it fails, try another handle if possible */
1220 cFYI(1, "wp failed on reopen file");
1221 cifsFileInfo_put(open_file);
1223 spin_lock(&cifs_file_list_lock);
1225 /* else we simply continue to the next entry. Thus
1226 we do not loop on reopen errors. If we
1227 can not reopen the file, for example if we
1228 reconnected to a server with another client
1229 racing to delete or lock the file we would not
1230 make progress if we restarted before the beginning
1231 of the loop here. */
1234 /* couldn't find useable FH with same pid, try any available */
1235 if (!any_available) {
1236 any_available = true;
1237 goto refind_writable;
1239 spin_unlock(&cifs_file_list_lock);
1243 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1245 struct address_space *mapping = page->mapping;
1246 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1249 int bytes_written = 0;
1250 struct cifs_sb_info *cifs_sb;
1251 struct inode *inode;
1252 struct cifsFileInfo *open_file;
1254 if (!mapping || !mapping->host)
1257 inode = page->mapping->host;
1258 cifs_sb = CIFS_SB(inode->i_sb);
1260 offset += (loff_t)from;
1261 write_data = kmap(page);
1264 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1269 /* racing with truncate? */
1270 if (offset > mapping->host->i_size) {
1272 return 0; /* don't care */
1275 /* check to make sure that we are not extending the file */
1276 if (mapping->host->i_size - offset < (loff_t)to)
1277 to = (unsigned)(mapping->host->i_size - offset);
1279 open_file = find_writable_file(CIFS_I(mapping->host), false);
1281 bytes_written = cifs_write(open_file, write_data,
1282 to - from, &offset);
1283 cifsFileInfo_put(open_file);
1284 /* Does mm or vfs already set times? */
1285 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1286 if ((bytes_written > 0) && (offset))
1288 else if (bytes_written < 0)
1291 cFYI(1, "No writeable filehandles for inode");
1299 static int cifs_writepages(struct address_space *mapping,
1300 struct writeback_control *wbc)
1302 unsigned int bytes_to_write;
1303 unsigned int bytes_written;
1304 struct cifs_sb_info *cifs_sb;
1308 int range_whole = 0;
1315 struct cifsFileInfo *open_file;
1316 struct cifsTconInfo *tcon;
1317 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1319 struct pagevec pvec;
1324 cifs_sb = CIFS_SB(mapping->host->i_sb);
1327 * If wsize is smaller that the page cache size, default to writing
1328 * one page at a time via cifs_writepage
1330 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1331 return generic_writepages(mapping, wbc);
1333 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1335 return generic_writepages(mapping, wbc);
1338 * if there's no open file, then this is likely to fail too,
1339 * but it'll at least handle the return. Maybe it should be
1342 open_file = find_writable_file(CIFS_I(mapping->host), false);
1345 return generic_writepages(mapping, wbc);
1348 tcon = tlink_tcon(open_file->tlink);
1349 if (!experimEnabled && tcon->ses->server->secMode &
1350 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1351 cifsFileInfo_put(open_file);
1353 return generic_writepages(mapping, wbc);
1355 cifsFileInfo_put(open_file);
1359 pagevec_init(&pvec, 0);
1360 if (wbc->range_cyclic) {
1361 index = mapping->writeback_index; /* Start from prev offset */
1364 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1365 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1366 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1371 while (!done && (index <= end) &&
1372 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1373 PAGECACHE_TAG_DIRTY,
1374 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1383 for (i = 0; i < nr_pages; i++) {
1384 page = pvec.pages[i];
1386 * At this point we hold neither mapping->tree_lock nor
1387 * lock on the page itself: the page may be truncated or
1388 * invalidated (changing page->mapping to NULL), or even
1389 * swizzled back from swapper_space to tmpfs file
1395 else if (!trylock_page(page))
1398 if (unlikely(page->mapping != mapping)) {
1403 if (!wbc->range_cyclic && page->index > end) {
1409 if (next && (page->index != next)) {
1410 /* Not next consecutive page */
1415 if (wbc->sync_mode != WB_SYNC_NONE)
1416 wait_on_page_writeback(page);
1418 if (PageWriteback(page) ||
1419 !clear_page_dirty_for_io(page)) {
1425 * This actually clears the dirty bit in the radix tree.
1426 * See cifs_writepage() for more commentary.
1428 set_page_writeback(page);
1430 if (page_offset(page) >= mapping->host->i_size) {
1433 end_page_writeback(page);
1438 * BB can we get rid of this? pages are held by pvec
1440 page_cache_get(page);
1442 len = min(mapping->host->i_size - page_offset(page),
1443 (loff_t)PAGE_CACHE_SIZE);
1445 /* reserve iov[0] for the smb header */
1447 iov[n_iov].iov_base = kmap(page);
1448 iov[n_iov].iov_len = len;
1449 bytes_to_write += len;
1453 offset = page_offset(page);
1455 next = page->index + 1;
1456 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1460 open_file = find_writable_file(CIFS_I(mapping->host),
1463 cERROR(1, "No writable handles for inode");
1466 long_op = cifs_write_timeout(cifsi, offset);
1467 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1468 bytes_to_write, offset,
1469 &bytes_written, iov, n_iov,
1471 cifsFileInfo_put(open_file);
1472 cifs_update_eof(cifsi, offset, bytes_written);
1475 if (rc || bytes_written < bytes_to_write) {
1476 cERROR(1, "Write2 ret %d, wrote %d",
1478 mapping_set_error(mapping, rc);
1480 cifs_stats_bytes_written(tcon, bytes_written);
1483 for (i = 0; i < n_iov; i++) {
1484 page = pvec.pages[first + i];
1485 /* Should we also set page error on
1486 success rc but too little data written? */
1487 /* BB investigate retry logic on temporary
1488 server crash cases and how recovery works
1489 when page marked as error */
1494 end_page_writeback(page);
1495 page_cache_release(page);
1497 if ((wbc->nr_to_write -= n_iov) <= 0)
1501 /* Need to re-find the pages we skipped */
1502 index = pvec.pages[0]->index + 1;
1504 pagevec_release(&pvec);
1506 if (!scanned && !done) {
1508 * We hit the last page and there is more work to be done: wrap
1509 * back to the start of the file
1515 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1516 mapping->writeback_index = index;
1523 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1529 /* BB add check for wbc flags */
1530 page_cache_get(page);
1531 if (!PageUptodate(page))
1532 cFYI(1, "ppw - page not up to date");
1535 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1537 * A writepage() implementation always needs to do either this,
1538 * or re-dirty the page with "redirty_page_for_writepage()" in
1539 * the case of a failure.
1541 * Just unlocking the page will cause the radix tree tag-bits
1542 * to fail to update with the state of the page correctly.
1544 set_page_writeback(page);
1545 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1546 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1548 end_page_writeback(page);
1549 page_cache_release(page);
1554 static int cifs_write_end(struct file *file, struct address_space *mapping,
1555 loff_t pos, unsigned len, unsigned copied,
1556 struct page *page, void *fsdata)
1559 struct inode *inode = mapping->host;
1561 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1564 if (PageChecked(page)) {
1566 SetPageUptodate(page);
1567 ClearPageChecked(page);
1568 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1569 SetPageUptodate(page);
1571 if (!PageUptodate(page)) {
1573 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1577 /* this is probably better than directly calling
1578 partialpage_write since in this function the file handle is
1579 known which we might as well leverage */
1580 /* BB check if anything else missing out of ppw
1581 such as updating last write time */
1582 page_data = kmap(page);
1583 rc = cifs_write(file->private_data, page_data + offset,
1585 /* if (rc < 0) should we set writebehind rc? */
1592 set_page_dirty(page);
1596 spin_lock(&inode->i_lock);
1597 if (pos > inode->i_size)
1598 i_size_write(inode, pos);
1599 spin_unlock(&inode->i_lock);
1603 page_cache_release(page);
1608 int cifs_fsync(struct file *file, int datasync)
1612 struct cifsTconInfo *tcon;
1613 struct cifsFileInfo *smbfile = file->private_data;
1614 struct inode *inode = file->f_path.dentry->d_inode;
1618 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1619 file->f_path.dentry->d_name.name, datasync);
1621 rc = filemap_write_and_wait(inode->i_mapping);
1623 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
1625 tcon = tlink_tcon(smbfile->tlink);
1626 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1627 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1634 /* static void cifs_sync_page(struct page *page)
1636 struct address_space *mapping;
1637 struct inode *inode;
1638 unsigned long index = page->index;
1639 unsigned int rpages = 0;
1642 cFYI(1, "sync page %p", page);
1643 mapping = page->mapping;
1646 inode = mapping->host;
1650 /* fill in rpages then
1651 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1653 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1663 * As file closes, flush all cached write data for this inode checking
1664 * for write behind errors.
1666 int cifs_flush(struct file *file, fl_owner_t id)
1668 struct inode *inode = file->f_path.dentry->d_inode;
1671 if (file->f_mode & FMODE_WRITE)
1672 rc = filemap_write_and_wait(inode->i_mapping);
1674 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1679 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1680 size_t read_size, loff_t *poffset)
1683 unsigned int bytes_read = 0;
1684 unsigned int total_read = 0;
1685 unsigned int current_read_size;
1686 struct cifs_sb_info *cifs_sb;
1687 struct cifsTconInfo *pTcon;
1689 struct cifsFileInfo *open_file;
1690 char *smb_read_data;
1691 char __user *current_offset;
1692 struct smb_com_read_rsp *pSMBr;
1695 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1697 if (file->private_data == NULL) {
1702 open_file = file->private_data;
1703 pTcon = tlink_tcon(open_file->tlink);
1705 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1706 cFYI(1, "attempting read on write only file instance");
1708 for (total_read = 0, current_offset = read_data;
1709 read_size > total_read;
1710 total_read += bytes_read, current_offset += bytes_read) {
1711 current_read_size = min_t(const int, read_size - total_read,
1714 smb_read_data = NULL;
1715 while (rc == -EAGAIN) {
1716 int buf_type = CIFS_NO_BUFFER;
1717 if (open_file->invalidHandle) {
1718 rc = cifs_reopen_file(open_file, true);
1722 rc = CIFSSMBRead(xid, pTcon,
1724 current_read_size, *poffset,
1725 &bytes_read, &smb_read_data,
1727 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1728 if (smb_read_data) {
1729 if (copy_to_user(current_offset,
1731 4 /* RFC1001 length field */ +
1732 le16_to_cpu(pSMBr->DataOffset),
1736 if (buf_type == CIFS_SMALL_BUFFER)
1737 cifs_small_buf_release(smb_read_data);
1738 else if (buf_type == CIFS_LARGE_BUFFER)
1739 cifs_buf_release(smb_read_data);
1740 smb_read_data = NULL;
1743 if (rc || (bytes_read == 0)) {
1751 cifs_stats_bytes_read(pTcon, bytes_read);
1752 *poffset += bytes_read;
1760 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1764 unsigned int bytes_read = 0;
1765 unsigned int total_read;
1766 unsigned int current_read_size;
1767 struct cifs_sb_info *cifs_sb;
1768 struct cifsTconInfo *pTcon;
1770 char *current_offset;
1771 struct cifsFileInfo *open_file;
1772 int buf_type = CIFS_NO_BUFFER;
1775 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1777 if (file->private_data == NULL) {
1782 open_file = file->private_data;
1783 pTcon = tlink_tcon(open_file->tlink);
1785 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1786 cFYI(1, "attempting read on write only file instance");
1788 for (total_read = 0, current_offset = read_data;
1789 read_size > total_read;
1790 total_read += bytes_read, current_offset += bytes_read) {
1791 current_read_size = min_t(const int, read_size - total_read,
1793 /* For windows me and 9x we do not want to request more
1794 than it negotiated since it will refuse the read then */
1796 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1797 current_read_size = min_t(const int, current_read_size,
1798 pTcon->ses->server->maxBuf - 128);
1801 while (rc == -EAGAIN) {
1802 if (open_file->invalidHandle) {
1803 rc = cifs_reopen_file(open_file, true);
1807 rc = CIFSSMBRead(xid, pTcon,
1809 current_read_size, *poffset,
1810 &bytes_read, ¤t_offset,
1813 if (rc || (bytes_read == 0)) {
1821 cifs_stats_bytes_read(pTcon, total_read);
1822 *poffset += bytes_read;
1829 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1834 rc = cifs_revalidate_file(file);
1836 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1840 rc = generic_file_mmap(file, vma);
1846 static void cifs_copy_cache_pages(struct address_space *mapping,
1847 struct list_head *pages, int bytes_read, char *data)
1852 while (bytes_read > 0) {
1853 if (list_empty(pages))
1856 page = list_entry(pages->prev, struct page, lru);
1857 list_del(&page->lru);
1859 if (add_to_page_cache_lru(page, mapping, page->index,
1861 page_cache_release(page);
1862 cFYI(1, "Add page cache failed");
1863 data += PAGE_CACHE_SIZE;
1864 bytes_read -= PAGE_CACHE_SIZE;
1867 page_cache_release(page);
1869 target = kmap_atomic(page, KM_USER0);
1871 if (PAGE_CACHE_SIZE > bytes_read) {
1872 memcpy(target, data, bytes_read);
1873 /* zero the tail end of this partial page */
1874 memset(target + bytes_read, 0,
1875 PAGE_CACHE_SIZE - bytes_read);
1878 memcpy(target, data, PAGE_CACHE_SIZE);
1879 bytes_read -= PAGE_CACHE_SIZE;
1881 kunmap_atomic(target, KM_USER0);
1883 flush_dcache_page(page);
1884 SetPageUptodate(page);
1886 data += PAGE_CACHE_SIZE;
1888 /* add page to FS-Cache */
1889 cifs_readpage_to_fscache(mapping->host, page);
1894 static int cifs_readpages(struct file *file, struct address_space *mapping,
1895 struct list_head *page_list, unsigned num_pages)
1901 struct cifs_sb_info *cifs_sb;
1902 struct cifsTconInfo *pTcon;
1903 unsigned int bytes_read = 0;
1904 unsigned int read_size, i;
1905 char *smb_read_data = NULL;
1906 struct smb_com_read_rsp *pSMBr;
1907 struct cifsFileInfo *open_file;
1908 int buf_type = CIFS_NO_BUFFER;
1911 if (file->private_data == NULL) {
1916 open_file = file->private_data;
1917 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1918 pTcon = tlink_tcon(open_file->tlink);
1921 * Reads as many pages as possible from fscache. Returns -ENOBUFS
1922 * immediately if the cookie is negative
1924 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
1929 cFYI(DBG2, "rpages: num pages %d", num_pages);
1930 for (i = 0; i < num_pages; ) {
1931 unsigned contig_pages;
1932 struct page *tmp_page;
1933 unsigned long expected_index;
1935 if (list_empty(page_list))
1938 page = list_entry(page_list->prev, struct page, lru);
1939 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1941 /* count adjacent pages that we will read into */
1944 list_entry(page_list->prev, struct page, lru)->index;
1945 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1946 if (tmp_page->index == expected_index) {
1952 if (contig_pages + i > num_pages)
1953 contig_pages = num_pages - i;
1955 /* for reads over a certain size could initiate async
1958 read_size = contig_pages * PAGE_CACHE_SIZE;
1959 /* Read size needs to be in multiples of one page */
1960 read_size = min_t(const unsigned int, read_size,
1961 cifs_sb->rsize & PAGE_CACHE_MASK);
1962 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
1963 read_size, contig_pages);
1965 while (rc == -EAGAIN) {
1966 if (open_file->invalidHandle) {
1967 rc = cifs_reopen_file(open_file, true);
1972 rc = CIFSSMBRead(xid, pTcon,
1975 &bytes_read, &smb_read_data,
1977 /* BB more RC checks ? */
1978 if (rc == -EAGAIN) {
1979 if (smb_read_data) {
1980 if (buf_type == CIFS_SMALL_BUFFER)
1981 cifs_small_buf_release(smb_read_data);
1982 else if (buf_type == CIFS_LARGE_BUFFER)
1983 cifs_buf_release(smb_read_data);
1984 smb_read_data = NULL;
1988 if ((rc < 0) || (smb_read_data == NULL)) {
1989 cFYI(1, "Read error in readpages: %d", rc);
1991 } else if (bytes_read > 0) {
1992 task_io_account_read(bytes_read);
1993 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1994 cifs_copy_cache_pages(mapping, page_list, bytes_read,
1995 smb_read_data + 4 /* RFC1001 hdr */ +
1996 le16_to_cpu(pSMBr->DataOffset));
1998 i += bytes_read >> PAGE_CACHE_SHIFT;
1999 cifs_stats_bytes_read(pTcon, bytes_read);
2000 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2001 i++; /* account for partial page */
2003 /* server copy of file can have smaller size
2005 /* BB do we need to verify this common case ?
2006 this case is ok - if we are at server EOF
2007 we will hit it on next read */
2012 cFYI(1, "No bytes read (%d) at offset %lld . "
2013 "Cleaning remaining pages from readahead list",
2014 bytes_read, offset);
2015 /* BB turn off caching and do new lookup on
2016 file size at server? */
2019 if (smb_read_data) {
2020 if (buf_type == CIFS_SMALL_BUFFER)
2021 cifs_small_buf_release(smb_read_data);
2022 else if (buf_type == CIFS_LARGE_BUFFER)
2023 cifs_buf_release(smb_read_data);
2024 smb_read_data = NULL;
2029 /* need to free smb_read_data buf before exit */
2030 if (smb_read_data) {
2031 if (buf_type == CIFS_SMALL_BUFFER)
2032 cifs_small_buf_release(smb_read_data);
2033 else if (buf_type == CIFS_LARGE_BUFFER)
2034 cifs_buf_release(smb_read_data);
2035 smb_read_data = NULL;
2043 static int cifs_readpage_worker(struct file *file, struct page *page,
2049 /* Is the page cached? */
2050 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2054 page_cache_get(page);
2055 read_data = kmap(page);
2056 /* for reads over a certain size could initiate async read ahead */
2058 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2063 cFYI(1, "Bytes read %d", rc);
2065 file->f_path.dentry->d_inode->i_atime =
2066 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2068 if (PAGE_CACHE_SIZE > rc)
2069 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2071 flush_dcache_page(page);
2072 SetPageUptodate(page);
2074 /* send this page to the cache */
2075 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2081 page_cache_release(page);
2087 static int cifs_readpage(struct file *file, struct page *page)
2089 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2095 if (file->private_data == NULL) {
2101 cFYI(1, "readpage %p at offset %d 0x%x\n",
2102 page, (int)offset, (int)offset);
2104 rc = cifs_readpage_worker(file, page, &offset);
2112 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2114 struct cifsFileInfo *open_file;
2116 spin_lock(&cifs_file_list_lock);
2117 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2118 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2119 spin_unlock(&cifs_file_list_lock);
2123 spin_unlock(&cifs_file_list_lock);
2127 /* We do not want to update the file size from server for inodes
2128 open for write - to avoid races with writepage extending
2129 the file - in the future we could consider allowing
2130 refreshing the inode only on increases in the file size
2131 but this is tricky to do without racing with writebehind
2132 page caching in the current Linux kernel design */
2133 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2138 if (is_inode_writable(cifsInode)) {
2139 /* This inode is open for write at least once */
2140 struct cifs_sb_info *cifs_sb;
2142 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2143 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2144 /* since no page cache to corrupt on directio
2145 we can change size safely */
2149 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2157 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2158 loff_t pos, unsigned len, unsigned flags,
2159 struct page **pagep, void **fsdata)
2161 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2162 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2163 loff_t page_start = pos & PAGE_MASK;
2168 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2170 page = grab_cache_page_write_begin(mapping, index, flags);
2176 if (PageUptodate(page))
2180 * If we write a full page it will be up to date, no need to read from
2181 * the server. If the write is short, we'll end up doing a sync write
2184 if (len == PAGE_CACHE_SIZE)
2188 * optimize away the read when we have an oplock, and we're not
2189 * expecting to use any of the data we'd be reading in. That
2190 * is, when the page lies beyond the EOF, or straddles the EOF
2191 * and the write will cover all of the existing data.
2193 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2194 i_size = i_size_read(mapping->host);
2195 if (page_start >= i_size ||
2196 (offset == 0 && (pos + len) >= i_size)) {
2197 zero_user_segments(page, 0, offset,
2201 * PageChecked means that the parts of the page
2202 * to which we're not writing are considered up
2203 * to date. Once the data is copied to the
2204 * page, it can be set uptodate.
2206 SetPageChecked(page);
2211 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2213 * might as well read a page, it is fast enough. If we get
2214 * an error, we don't need to return it. cifs_write_end will
2215 * do a sync write instead since PG_uptodate isn't set.
2217 cifs_readpage_worker(file, page, &page_start);
2219 /* we could try using another file handle if there is one -
2220 but how would we lock it to prevent close of that handle
2221 racing with this read? In any case
2222 this will be written out by write_end so is fine */
2229 static int cifs_release_page(struct page *page, gfp_t gfp)
2231 if (PagePrivate(page))
2234 return cifs_fscache_release_page(page, gfp);
2237 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2239 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2242 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2245 void cifs_oplock_break(struct work_struct *work)
2247 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2249 struct inode *inode = cfile->dentry->d_inode;
2250 struct cifsInodeInfo *cinode = CIFS_I(inode);
2253 if (inode && S_ISREG(inode->i_mode)) {
2254 if (cinode->clientCanCacheRead)
2255 break_lease(inode, O_RDONLY);
2257 break_lease(inode, O_WRONLY);
2258 rc = filemap_fdatawrite(inode->i_mapping);
2259 if (cinode->clientCanCacheRead == 0) {
2260 rc = filemap_fdatawait(inode->i_mapping);
2261 mapping_set_error(inode->i_mapping, rc);
2262 invalidate_remote_inode(inode);
2264 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2268 * releasing stale oplock after recent reconnect of smb session using
2269 * a now incorrect file handle is not a data integrity issue but do
2270 * not bother sending an oplock release if session to server still is
2271 * disconnected since oplock already released by the server
2273 if (!cfile->oplock_break_cancelled) {
2274 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2275 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false);
2276 cFYI(1, "Oplock release rc = %d", rc);
2280 * We might have kicked in before is_valid_oplock_break()
2281 * finished grabbing reference for us. Make sure it's done by
2282 * waiting for cifs_file_list_lock.
2284 spin_lock(&cifs_file_list_lock);
2285 spin_unlock(&cifs_file_list_lock);
2287 cifs_oplock_break_put(cfile);
2290 /* must be called while holding cifs_file_list_lock */
2291 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2293 cifs_sb_active(cfile->dentry->d_sb);
2294 cifsFileInfo_get(cfile);
2297 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2299 cifsFileInfo_put(cfile);
2300 cifs_sb_deactive(cfile->dentry->d_sb);
2303 const struct address_space_operations cifs_addr_ops = {
2304 .readpage = cifs_readpage,
2305 .readpages = cifs_readpages,
2306 .writepage = cifs_writepage,
2307 .writepages = cifs_writepages,
2308 .write_begin = cifs_write_begin,
2309 .write_end = cifs_write_end,
2310 .set_page_dirty = __set_page_dirty_nobuffers,
2311 .releasepage = cifs_release_page,
2312 .invalidatepage = cifs_invalidate_page,
2313 /* .sync_page = cifs_sync_page, */
2318 * cifs_readpages requires the server to support a buffer large enough to
2319 * contain the header plus one complete page of data. Otherwise, we need
2320 * to leave cifs_readpages out of the address space operations.
2322 const struct address_space_operations cifs_addr_ops_smallbuf = {
2323 .readpage = cifs_readpage,
2324 .writepage = cifs_writepage,
2325 .writepages = cifs_writepages,
2326 .write_begin = cifs_write_begin,
2327 .write_end = cifs_write_end,
2328 .set_page_dirty = __set_page_dirty_nobuffers,
2329 .releasepage = cifs_release_page,
2330 .invalidatepage = cifs_invalidate_page,
2331 /* .sync_page = cifs_sync_page, */