4 * vfs operations that deal with files
6 * Copyright (C) International Business Machines Corp., 2002,2010
7 * Author(s): Steve French (sfrench@us.ibm.com)
8 * Jeremy Allison (jra@samba.org)
10 * This library is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU Lesser General Public License as published
12 * by the Free Software Foundation; either version 2.1 of the License, or
13 * (at your option) any later version.
15 * This library is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
18 * the GNU Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public License
21 * along with this library; if not, write to the Free Software
22 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
25 #include <linux/backing-dev.h>
26 #include <linux/stat.h>
27 #include <linux/fcntl.h>
28 #include <linux/pagemap.h>
29 #include <linux/pagevec.h>
30 #include <linux/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/delay.h>
33 #include <linux/mount.h>
34 #include <linux/slab.h>
35 #include <asm/div64.h>
39 #include "cifsproto.h"
40 #include "cifs_unicode.h"
41 #include "cifs_debug.h"
42 #include "cifs_fs_sb.h"
45 static inline int cifs_convert_flags(unsigned int flags)
47 if ((flags & O_ACCMODE) == O_RDONLY)
49 else if ((flags & O_ACCMODE) == O_WRONLY)
51 else if ((flags & O_ACCMODE) == O_RDWR) {
52 /* GENERIC_ALL is too much permission to request
53 can cause unnecessary access denied on create */
54 /* return GENERIC_ALL; */
55 return (GENERIC_READ | GENERIC_WRITE);
58 return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
59 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
63 static u32 cifs_posix_convert_flags(unsigned int flags)
67 if ((flags & O_ACCMODE) == O_RDONLY)
68 posix_flags = SMB_O_RDONLY;
69 else if ((flags & O_ACCMODE) == O_WRONLY)
70 posix_flags = SMB_O_WRONLY;
71 else if ((flags & O_ACCMODE) == O_RDWR)
72 posix_flags = SMB_O_RDWR;
75 posix_flags |= SMB_O_CREAT;
77 posix_flags |= SMB_O_EXCL;
79 posix_flags |= SMB_O_TRUNC;
80 /* be safe and imply O_SYNC for O_DSYNC */
82 posix_flags |= SMB_O_SYNC;
83 if (flags & O_DIRECTORY)
84 posix_flags |= SMB_O_DIRECTORY;
85 if (flags & O_NOFOLLOW)
86 posix_flags |= SMB_O_NOFOLLOW;
88 posix_flags |= SMB_O_DIRECT;
93 static inline int cifs_get_disposition(unsigned int flags)
95 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
97 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
98 return FILE_OVERWRITE_IF;
99 else if ((flags & O_CREAT) == O_CREAT)
101 else if ((flags & O_TRUNC) == O_TRUNC)
102 return FILE_OVERWRITE;
107 static inline int cifs_open_inode_helper(struct inode *inode,
108 struct cifsTconInfo *pTcon, __u32 oplock, FILE_ALL_INFO *buf,
109 char *full_path, int xid)
111 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
112 struct timespec temp;
115 if (pCifsInode->clientCanCacheRead) {
116 /* we have the inode open somewhere else
117 no need to discard cache data */
118 goto client_can_cache;
121 /* BB need same check in cifs_create too? */
122 /* if not oplocked, invalidate inode pages if mtime or file
124 temp = cifs_NTtimeToUnix(buf->LastWriteTime);
125 if (timespec_equal(&inode->i_mtime, &temp) &&
127 (loff_t)le64_to_cpu(buf->EndOfFile))) {
128 cFYI(1, "inode unchanged on server");
130 if (inode->i_mapping) {
131 /* BB no need to lock inode until after invalidate
132 since namei code should already have it locked? */
133 rc = filemap_write_and_wait(inode->i_mapping);
135 pCifsInode->write_behind_rc = rc;
137 cFYI(1, "invalidating remote inode since open detected it "
139 invalidate_remote_inode(inode);
144 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
147 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
150 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
151 pCifsInode->clientCanCacheAll = true;
152 pCifsInode->clientCanCacheRead = true;
153 cFYI(1, "Exclusive Oplock granted on inode %p", inode);
154 } else if ((oplock & 0xF) == OPLOCK_READ)
155 pCifsInode->clientCanCacheRead = true;
160 int cifs_posix_open(char *full_path, struct inode **pinode,
161 struct super_block *sb, int mode, unsigned int f_flags,
162 __u32 *poplock, __u16 *pnetfid, int xid)
165 FILE_UNIX_BASIC_INFO *presp_data;
166 __u32 posix_flags = 0;
167 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
168 struct cifs_fattr fattr;
169 struct tcon_link *tlink;
170 struct cifsTconInfo *tcon;
172 cFYI(1, "posix open %s", full_path);
174 presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
175 if (presp_data == NULL)
178 tlink = cifs_sb_tlink(cifs_sb);
184 tcon = tlink_tcon(tlink);
185 mode &= ~current_umask();
187 posix_flags = cifs_posix_convert_flags(f_flags);
188 rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
189 poplock, full_path, cifs_sb->local_nls,
190 cifs_sb->mnt_cifs_flags &
191 CIFS_MOUNT_MAP_SPECIAL_CHR);
192 cifs_put_tlink(tlink);
197 if (presp_data->Type == cpu_to_le32(-1))
198 goto posix_open_ret; /* open ok, caller does qpathinfo */
201 goto posix_open_ret; /* caller does not need info */
203 cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
205 /* get new inode and set it up */
206 if (*pinode == NULL) {
207 cifs_fill_uniqueid(sb, &fattr);
208 *pinode = cifs_iget(sb, &fattr);
214 cifs_fattr_to_inode(*pinode, &fattr);
222 struct cifsFileInfo *
223 cifs_new_fileinfo(__u16 fileHandle, struct file *file,
224 struct tcon_link *tlink, __u32 oplock)
226 struct dentry *dentry = file->f_path.dentry;
227 struct inode *inode = dentry->d_inode;
228 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
229 struct cifsFileInfo *pCifsFile;
231 pCifsFile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
232 if (pCifsFile == NULL)
235 pCifsFile->count = 1;
236 pCifsFile->netfid = fileHandle;
237 pCifsFile->pid = current->tgid;
238 pCifsFile->uid = current_fsuid();
239 pCifsFile->dentry = dget(dentry);
240 pCifsFile->f_flags = file->f_flags;
241 pCifsFile->invalidHandle = false;
242 pCifsFile->tlink = cifs_get_tlink(tlink);
243 mutex_init(&pCifsFile->fh_mutex);
244 mutex_init(&pCifsFile->lock_mutex);
245 INIT_LIST_HEAD(&pCifsFile->llist);
246 INIT_WORK(&pCifsFile->oplock_break, cifs_oplock_break);
248 spin_lock(&cifs_file_list_lock);
249 list_add(&pCifsFile->tlist, &(tlink_tcon(tlink)->openFileList));
250 /* if readable file instance put first in list*/
251 if (file->f_mode & FMODE_READ)
252 list_add(&pCifsFile->flist, &pCifsInode->openFileList);
254 list_add_tail(&pCifsFile->flist, &pCifsInode->openFileList);
255 spin_unlock(&cifs_file_list_lock);
257 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
258 pCifsInode->clientCanCacheAll = true;
259 pCifsInode->clientCanCacheRead = true;
260 cFYI(1, "Exclusive Oplock inode %p", inode);
261 } else if ((oplock & 0xF) == OPLOCK_READ)
262 pCifsInode->clientCanCacheRead = true;
264 file->private_data = pCifsFile;
269 * Release a reference on the file private data. This may involve closing
270 * the filehandle out on the server. Must be called without holding
271 * cifs_file_list_lock.
273 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
275 struct cifsTconInfo *tcon = tlink_tcon(cifs_file->tlink);
276 struct cifsInodeInfo *cifsi = CIFS_I(cifs_file->dentry->d_inode);
277 struct cifsLockInfo *li, *tmp;
279 spin_lock(&cifs_file_list_lock);
280 if (--cifs_file->count > 0) {
281 spin_unlock(&cifs_file_list_lock);
285 /* remove it from the lists */
286 list_del(&cifs_file->flist);
287 list_del(&cifs_file->tlist);
289 if (list_empty(&cifsi->openFileList)) {
290 cFYI(1, "closing last open instance for inode %p",
291 cifs_file->dentry->d_inode);
292 cifsi->clientCanCacheRead = false;
293 cifsi->clientCanCacheAll = false;
295 spin_unlock(&cifs_file_list_lock);
297 if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
301 rc = CIFSSMBClose(xid, tcon, cifs_file->netfid);
305 /* Delete any outstanding lock records. We'll lose them when the file
308 mutex_lock(&cifs_file->lock_mutex);
309 list_for_each_entry_safe(li, tmp, &cifs_file->llist, llist) {
310 list_del(&li->llist);
313 mutex_unlock(&cifs_file->lock_mutex);
315 cifs_put_tlink(cifs_file->tlink);
316 dput(cifs_file->dentry);
320 int cifs_open(struct inode *inode, struct file *file)
325 struct cifs_sb_info *cifs_sb;
326 struct cifsTconInfo *tcon;
327 struct tcon_link *tlink;
328 struct cifsFileInfo *pCifsFile = NULL;
329 struct cifsInodeInfo *pCifsInode;
330 char *full_path = NULL;
334 FILE_ALL_INFO *buf = NULL;
338 cifs_sb = CIFS_SB(inode->i_sb);
339 tlink = cifs_sb_tlink(cifs_sb);
342 return PTR_ERR(tlink);
344 tcon = tlink_tcon(tlink);
346 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
348 full_path = build_path_from_dentry(file->f_path.dentry);
349 if (full_path == NULL) {
354 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
355 inode, file->f_flags, full_path);
362 if (!tcon->broken_posix_open && tcon->unix_ext &&
363 (tcon->ses->capabilities & CAP_UNIX) &&
364 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
365 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
366 /* can not refresh inode info since size could be stale */
367 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
368 cifs_sb->mnt_file_mode /* ignored */,
369 file->f_flags, &oplock, &netfid, xid);
371 cFYI(1, "posix open succeeded");
373 pCifsFile = cifs_new_fileinfo(netfid, file, tlink,
375 if (pCifsFile == NULL) {
376 CIFSSMBClose(xid, tcon, netfid);
380 cifs_fscache_set_inode_cookie(inode, file);
383 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
384 if (tcon->ses->serverNOS)
385 cERROR(1, "server %s of type %s returned"
386 " unexpected error on SMB posix open"
387 ", disabling posix open support."
388 " Check if server update available.",
389 tcon->ses->serverName,
390 tcon->ses->serverNOS);
391 tcon->broken_posix_open = true;
392 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
393 (rc != -EOPNOTSUPP)) /* path not found or net err */
395 /* else fallthrough to retry open the old way on network i/o
399 desiredAccess = cifs_convert_flags(file->f_flags);
401 /*********************************************************************
402 * open flag mapping table:
404 * POSIX Flag CIFS Disposition
405 * ---------- ----------------
406 * O_CREAT FILE_OPEN_IF
407 * O_CREAT | O_EXCL FILE_CREATE
408 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
409 * O_TRUNC FILE_OVERWRITE
410 * none of the above FILE_OPEN
412 * Note that there is not a direct match between disposition
413 * FILE_SUPERSEDE (ie create whether or not file exists although
414 * O_CREAT | O_TRUNC is similar but truncates the existing
415 * file rather than creating a new file as FILE_SUPERSEDE does
416 * (which uses the attributes / metadata passed in on open call)
418 *? O_SYNC is a reasonable match to CIFS writethrough flag
419 *? and the read write flags match reasonably. O_LARGEFILE
420 *? is irrelevant because largefile support is always used
421 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
422 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
423 *********************************************************************/
425 disposition = cifs_get_disposition(file->f_flags);
427 /* BB pass O_SYNC flag through on file attributes .. BB */
429 /* Also refresh inode by passing in file_info buf returned by SMBOpen
430 and calling get_inode_info with returned buf (at least helps
431 non-Unix server case) */
433 /* BB we can not do this if this is the second open of a file
434 and the first handle has writebehind data, we might be
435 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
436 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
442 if (tcon->ses->capabilities & CAP_NT_SMBS)
443 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
444 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
445 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
446 & CIFS_MOUNT_MAP_SPECIAL_CHR);
448 rc = -EIO; /* no NT SMB support fall into legacy open below */
451 /* Old server, try legacy style OpenX */
452 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
453 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
454 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
455 & CIFS_MOUNT_MAP_SPECIAL_CHR);
458 cFYI(1, "cifs_open returned 0x%x", rc);
462 rc = cifs_open_inode_helper(inode, tcon, oplock, buf, full_path, xid);
466 pCifsFile = cifs_new_fileinfo(netfid, file, tlink, oplock);
467 if (pCifsFile == NULL) {
472 cifs_fscache_set_inode_cookie(inode, file);
474 if (oplock & CIFS_CREATE_ACTION) {
475 /* time to set mode which we can not set earlier due to
476 problems creating new read-only files */
477 if (tcon->unix_ext) {
478 struct cifs_unix_set_info_args args = {
479 .mode = inode->i_mode,
482 .ctime = NO_CHANGE_64,
483 .atime = NO_CHANGE_64,
484 .mtime = NO_CHANGE_64,
487 CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
489 cifs_sb->mnt_cifs_flags &
490 CIFS_MOUNT_MAP_SPECIAL_CHR);
498 cifs_put_tlink(tlink);
502 /* Try to reacquire byte range locks that were released when session */
503 /* to server was lost */
504 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
508 /* BB list all locks open on this file and relock */
513 static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
518 struct cifs_sb_info *cifs_sb;
519 struct cifsTconInfo *tcon;
520 struct cifsInodeInfo *pCifsInode;
522 char *full_path = NULL;
524 int disposition = FILE_OPEN;
528 mutex_lock(&pCifsFile->fh_mutex);
529 if (!pCifsFile->invalidHandle) {
530 mutex_unlock(&pCifsFile->fh_mutex);
536 inode = pCifsFile->dentry->d_inode;
537 cifs_sb = CIFS_SB(inode->i_sb);
538 tcon = tlink_tcon(pCifsFile->tlink);
540 /* can not grab rename sem here because various ops, including
541 those that already have the rename sem can end up causing writepage
542 to get called and if the server was down that means we end up here,
543 and we can never tell if the caller already has the rename_sem */
544 full_path = build_path_from_dentry(pCifsFile->dentry);
545 if (full_path == NULL) {
547 mutex_unlock(&pCifsFile->fh_mutex);
552 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
553 inode, pCifsFile->f_flags, full_path);
560 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
561 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
562 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
565 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
566 * original open. Must mask them off for a reopen.
568 unsigned int oflags = pCifsFile->f_flags &
569 ~(O_CREAT | O_EXCL | O_TRUNC);
571 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
572 cifs_sb->mnt_file_mode /* ignored */,
573 oflags, &oplock, &netfid, xid);
575 cFYI(1, "posix reopen succeeded");
578 /* fallthrough to retry open the old way on errors, especially
579 in the reconnect path it is important to retry hard */
582 desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
584 /* Can not refresh inode by passing in file_info buf to be returned
585 by SMBOpen and then calling get_inode_info with returned buf
586 since file might have write behind data that needs to be flushed
587 and server version of file size can be stale. If we knew for sure
588 that inode was not dirty locally we could do this */
590 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
591 CREATE_NOT_DIR, &netfid, &oplock, NULL,
592 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
593 CIFS_MOUNT_MAP_SPECIAL_CHR);
595 mutex_unlock(&pCifsFile->fh_mutex);
596 cFYI(1, "cifs_open returned 0x%x", rc);
597 cFYI(1, "oplock: %d", oplock);
598 goto reopen_error_exit;
602 pCifsFile->netfid = netfid;
603 pCifsFile->invalidHandle = false;
604 mutex_unlock(&pCifsFile->fh_mutex);
605 pCifsInode = CIFS_I(inode);
608 rc = filemap_write_and_wait(inode->i_mapping);
610 CIFS_I(inode)->write_behind_rc = rc;
612 pCifsInode->clientCanCacheAll = false;
613 pCifsInode->clientCanCacheRead = false;
615 rc = cifs_get_inode_info_unix(&inode,
616 full_path, inode->i_sb, xid);
618 rc = cifs_get_inode_info(&inode,
619 full_path, NULL, inode->i_sb,
621 } /* else we are writing out data to server already
622 and could deadlock if we tried to flush data, and
623 since we do not know if we have data that would
624 invalidate the current end of file on the server
625 we can not go to the server to get the new inod
627 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
628 pCifsInode->clientCanCacheAll = true;
629 pCifsInode->clientCanCacheRead = true;
630 cFYI(1, "Exclusive Oplock granted on inode %p",
631 pCifsFile->dentry->d_inode);
632 } else if ((oplock & 0xF) == OPLOCK_READ) {
633 pCifsInode->clientCanCacheRead = true;
634 pCifsInode->clientCanCacheAll = false;
636 pCifsInode->clientCanCacheRead = false;
637 pCifsInode->clientCanCacheAll = false;
639 cifs_relock_file(pCifsFile);
647 int cifs_close(struct inode *inode, struct file *file)
649 cifsFileInfo_put(file->private_data);
650 file->private_data = NULL;
652 /* return code from the ->release op is always ignored */
656 int cifs_closedir(struct inode *inode, struct file *file)
660 struct cifsFileInfo *pCFileStruct = file->private_data;
663 cFYI(1, "Closedir inode = 0x%p", inode);
668 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
670 cFYI(1, "Freeing private data in close dir");
671 spin_lock(&cifs_file_list_lock);
672 if (!pCFileStruct->srch_inf.endOfSearch &&
673 !pCFileStruct->invalidHandle) {
674 pCFileStruct->invalidHandle = true;
675 spin_unlock(&cifs_file_list_lock);
676 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
677 cFYI(1, "Closing uncompleted readdir with rc %d",
679 /* not much we can do if it fails anyway, ignore rc */
682 spin_unlock(&cifs_file_list_lock);
683 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
685 cFYI(1, "closedir free smb buf in srch struct");
686 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
687 if (pCFileStruct->srch_inf.smallBuf)
688 cifs_small_buf_release(ptmp);
690 cifs_buf_release(ptmp);
692 cifs_put_tlink(pCFileStruct->tlink);
693 kfree(file->private_data);
694 file->private_data = NULL;
696 /* BB can we lock the filestruct while this is going on? */
701 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
702 __u64 offset, __u8 lockType)
704 struct cifsLockInfo *li =
705 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
711 mutex_lock(&fid->lock_mutex);
712 list_add(&li->llist, &fid->llist);
713 mutex_unlock(&fid->lock_mutex);
717 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
723 bool wait_flag = false;
724 struct cifs_sb_info *cifs_sb;
725 struct cifsTconInfo *tcon;
727 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
728 bool posix_locking = 0;
730 length = 1 + pfLock->fl_end - pfLock->fl_start;
734 cFYI(1, "Lock parm: 0x%x flockflags: "
735 "0x%x flocktype: 0x%x start: %lld end: %lld",
736 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
739 if (pfLock->fl_flags & FL_POSIX)
741 if (pfLock->fl_flags & FL_FLOCK)
743 if (pfLock->fl_flags & FL_SLEEP) {
744 cFYI(1, "Blocking lock");
747 if (pfLock->fl_flags & FL_ACCESS)
748 cFYI(1, "Process suspended by mandatory locking - "
749 "not implemented yet");
750 if (pfLock->fl_flags & FL_LEASE)
751 cFYI(1, "Lease on file - not implemented yet");
752 if (pfLock->fl_flags &
753 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
754 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
756 if (pfLock->fl_type == F_WRLCK) {
759 } else if (pfLock->fl_type == F_UNLCK) {
762 /* Check if unlock includes more than
764 } else if (pfLock->fl_type == F_RDLCK) {
766 lockType |= LOCKING_ANDX_SHARED_LOCK;
768 } else if (pfLock->fl_type == F_EXLCK) {
771 } else if (pfLock->fl_type == F_SHLCK) {
773 lockType |= LOCKING_ANDX_SHARED_LOCK;
776 cFYI(1, "Unknown type of lock");
778 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
779 tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
781 if (file->private_data == NULL) {
786 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
788 if ((tcon->ses->capabilities & CAP_UNIX) &&
789 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
790 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
792 /* BB add code here to normalize offset and length to
793 account for negative length which we can not accept over the
798 if (lockType & LOCKING_ANDX_SHARED_LOCK)
799 posix_lock_type = CIFS_RDLCK;
801 posix_lock_type = CIFS_WRLCK;
802 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
804 posix_lock_type, wait_flag);
809 /* BB we could chain these into one lock request BB */
810 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
811 0, 1, lockType, 0 /* wait flag */ );
813 rc = CIFSSMBLock(xid, tcon, netfid, length,
814 pfLock->fl_start, 1 /* numUnlock */ ,
815 0 /* numLock */ , lockType,
817 pfLock->fl_type = F_UNLCK;
819 cERROR(1, "Error unlocking previously locked "
820 "range %d during test of lock", rc);
824 /* if rc == ERR_SHARING_VIOLATION ? */
827 if (lockType & LOCKING_ANDX_SHARED_LOCK) {
828 pfLock->fl_type = F_WRLCK;
830 rc = CIFSSMBLock(xid, tcon, netfid, length,
831 pfLock->fl_start, 0, 1,
832 lockType | LOCKING_ANDX_SHARED_LOCK,
835 rc = CIFSSMBLock(xid, tcon, netfid,
836 length, pfLock->fl_start, 1, 0,
838 LOCKING_ANDX_SHARED_LOCK,
840 pfLock->fl_type = F_RDLCK;
842 cERROR(1, "Error unlocking "
843 "previously locked range %d "
844 "during test of lock", rc);
847 pfLock->fl_type = F_WRLCK;
857 if (!numLock && !numUnlock) {
858 /* if no lock or unlock then nothing
859 to do since we do not know what it is */
866 if (lockType & LOCKING_ANDX_SHARED_LOCK)
867 posix_lock_type = CIFS_RDLCK;
869 posix_lock_type = CIFS_WRLCK;
872 posix_lock_type = CIFS_UNLCK;
874 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
876 posix_lock_type, wait_flag);
878 struct cifsFileInfo *fid = file->private_data;
881 rc = CIFSSMBLock(xid, tcon, netfid, length,
883 0, numLock, lockType, wait_flag);
886 /* For Windows locks we must store them. */
887 rc = store_file_lock(fid, length,
888 pfLock->fl_start, lockType);
890 } else if (numUnlock) {
891 /* For each stored lock that this unlock overlaps
892 completely, unlock it. */
894 struct cifsLockInfo *li, *tmp;
897 mutex_lock(&fid->lock_mutex);
898 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
899 if (pfLock->fl_start <= li->offset &&
900 (pfLock->fl_start + length) >=
901 (li->offset + li->length)) {
902 stored_rc = CIFSSMBLock(xid, tcon,
904 li->length, li->offset,
905 1, 0, li->type, false);
909 list_del(&li->llist);
914 mutex_unlock(&fid->lock_mutex);
918 if (pfLock->fl_flags & FL_POSIX)
919 posix_lock_file_wait(file, pfLock);
925 * Set the timeout on write requests past EOF. For some servers (Windows)
926 * these calls can be very long.
928 * If we're writing >10M past the EOF we give a 180s timeout. Anything less
929 * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
930 * The 10M cutoff is totally arbitrary. A better scheme for this would be
931 * welcome if someone wants to suggest one.
933 * We may be able to do a better job with this if there were some way to
934 * declare that a file should be sparse.
937 cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
939 if (offset <= cifsi->server_eof)
941 else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
942 return CIFS_VLONG_OP;
947 /* update the file size (if needed) after a write */
949 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
950 unsigned int bytes_written)
952 loff_t end_of_write = offset + bytes_written;
954 if (end_of_write > cifsi->server_eof)
955 cifsi->server_eof = end_of_write;
958 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
959 size_t write_size, loff_t *poffset)
962 unsigned int bytes_written = 0;
963 unsigned int total_written;
964 struct cifs_sb_info *cifs_sb;
965 struct cifsTconInfo *pTcon;
967 struct cifsFileInfo *open_file;
968 struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
970 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
972 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
973 *poffset, file->f_path.dentry->d_name.name); */
975 if (file->private_data == NULL)
978 open_file = file->private_data;
979 pTcon = tlink_tcon(open_file->tlink);
981 rc = generic_write_checks(file, poffset, &write_size, 0);
987 long_op = cifs_write_timeout(cifsi, *poffset);
988 for (total_written = 0; write_size > total_written;
989 total_written += bytes_written) {
991 while (rc == -EAGAIN) {
992 if (file->private_data == NULL) {
993 /* file has been closed on us */
995 /* if we have gotten here we have written some data
996 and blocked, and the file has been freed on us while
997 we blocked so return what we managed to write */
998 return total_written;
1000 if (open_file->invalidHandle) {
1001 /* we could deadlock if we called
1002 filemap_fdatawait from here so tell
1003 reopen_file not to flush data to server
1005 rc = cifs_reopen_file(open_file, false);
1010 rc = CIFSSMBWrite(xid, pTcon,
1012 min_t(const int, cifs_sb->wsize,
1013 write_size - total_written),
1014 *poffset, &bytes_written,
1015 NULL, write_data + total_written, long_op);
1017 if (rc || (bytes_written == 0)) {
1025 cifs_update_eof(cifsi, *poffset, bytes_written);
1026 *poffset += bytes_written;
1028 long_op = CIFS_STD_OP; /* subsequent writes fast -
1029 15 seconds is plenty */
1032 cifs_stats_bytes_written(pTcon, total_written);
1034 /* since the write may have blocked check these pointers again */
1035 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1036 struct inode *inode = file->f_path.dentry->d_inode;
1037 /* Do not update local mtime - server will set its actual value on write
1038 * inode->i_ctime = inode->i_mtime =
1039 * current_fs_time(inode->i_sb);*/
1040 if (total_written > 0) {
1041 spin_lock(&inode->i_lock);
1042 if (*poffset > file->f_path.dentry->d_inode->i_size)
1043 i_size_write(file->f_path.dentry->d_inode,
1045 spin_unlock(&inode->i_lock);
1047 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1050 return total_written;
1053 static ssize_t cifs_write(struct cifsFileInfo *open_file,
1054 const char *write_data, size_t write_size,
1058 unsigned int bytes_written = 0;
1059 unsigned int total_written;
1060 struct cifs_sb_info *cifs_sb;
1061 struct cifsTconInfo *pTcon;
1063 struct dentry *dentry = open_file->dentry;
1064 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
1066 cifs_sb = CIFS_SB(dentry->d_sb);
1068 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
1069 *poffset, dentry->d_name.name);
1071 pTcon = tlink_tcon(open_file->tlink);
1075 long_op = cifs_write_timeout(cifsi, *poffset);
1076 for (total_written = 0; write_size > total_written;
1077 total_written += bytes_written) {
1079 while (rc == -EAGAIN) {
1080 if (open_file->invalidHandle) {
1081 /* we could deadlock if we called
1082 filemap_fdatawait from here so tell
1083 reopen_file not to flush data to
1085 rc = cifs_reopen_file(open_file, false);
1089 if (experimEnabled || (pTcon->ses->server &&
1090 ((pTcon->ses->server->secMode &
1091 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1096 len = min((size_t)cifs_sb->wsize,
1097 write_size - total_written);
1098 /* iov[0] is reserved for smb header */
1099 iov[1].iov_base = (char *)write_data +
1101 iov[1].iov_len = len;
1102 rc = CIFSSMBWrite2(xid, pTcon,
1103 open_file->netfid, len,
1104 *poffset, &bytes_written,
1107 rc = CIFSSMBWrite(xid, pTcon,
1109 min_t(const int, cifs_sb->wsize,
1110 write_size - total_written),
1111 *poffset, &bytes_written,
1112 write_data + total_written,
1115 if (rc || (bytes_written == 0)) {
1123 cifs_update_eof(cifsi, *poffset, bytes_written);
1124 *poffset += bytes_written;
1126 long_op = CIFS_STD_OP; /* subsequent writes fast -
1127 15 seconds is plenty */
1130 cifs_stats_bytes_written(pTcon, total_written);
1132 if (total_written > 0) {
1133 spin_lock(&dentry->d_inode->i_lock);
1134 if (*poffset > dentry->d_inode->i_size)
1135 i_size_write(dentry->d_inode, *poffset);
1136 spin_unlock(&dentry->d_inode->i_lock);
1138 mark_inode_dirty_sync(dentry->d_inode);
1140 return total_written;
1143 #ifdef CONFIG_CIFS_EXPERIMENTAL
1144 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1147 struct cifsFileInfo *open_file = NULL;
1148 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1150 /* only filter by fsuid on multiuser mounts */
1151 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1154 spin_lock(&cifs_file_list_lock);
1155 /* we could simply get the first_list_entry since write-only entries
1156 are always at the end of the list but since the first entry might
1157 have a close pending, we go through the whole list */
1158 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1159 if (fsuid_only && open_file->uid != current_fsuid())
1161 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1162 if (!open_file->invalidHandle) {
1163 /* found a good file */
1164 /* lock it so it will not be closed on us */
1165 cifsFileInfo_get(open_file);
1166 spin_unlock(&cifs_file_list_lock);
1168 } /* else might as well continue, and look for
1169 another, or simply have the caller reopen it
1170 again rather than trying to fix this handle */
1171 } else /* write only file */
1172 break; /* write only files are last so must be done */
1174 spin_unlock(&cifs_file_list_lock);
1179 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1182 struct cifsFileInfo *open_file;
1183 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1184 bool any_available = false;
1187 /* Having a null inode here (because mapping->host was set to zero by
1188 the VFS or MM) should not happen but we had reports of on oops (due to
1189 it being zero) during stress testcases so we need to check for it */
1191 if (cifs_inode == NULL) {
1192 cERROR(1, "Null inode passed to cifs_writeable_file");
1197 /* only filter by fsuid on multiuser mounts */
1198 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1201 spin_lock(&cifs_file_list_lock);
1203 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1204 if (!any_available && open_file->pid != current->tgid)
1206 if (fsuid_only && open_file->uid != current_fsuid())
1208 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1209 cifsFileInfo_get(open_file);
1211 if (!open_file->invalidHandle) {
1212 /* found a good writable file */
1213 spin_unlock(&cifs_file_list_lock);
1217 spin_unlock(&cifs_file_list_lock);
1219 /* Had to unlock since following call can block */
1220 rc = cifs_reopen_file(open_file, false);
1224 /* if it fails, try another handle if possible */
1225 cFYI(1, "wp failed on reopen file");
1226 cifsFileInfo_put(open_file);
1228 spin_lock(&cifs_file_list_lock);
1230 /* else we simply continue to the next entry. Thus
1231 we do not loop on reopen errors. If we
1232 can not reopen the file, for example if we
1233 reconnected to a server with another client
1234 racing to delete or lock the file we would not
1235 make progress if we restarted before the beginning
1236 of the loop here. */
1239 /* couldn't find useable FH with same pid, try any available */
1240 if (!any_available) {
1241 any_available = true;
1242 goto refind_writable;
1244 spin_unlock(&cifs_file_list_lock);
1248 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1250 struct address_space *mapping = page->mapping;
1251 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1254 int bytes_written = 0;
1255 struct cifs_sb_info *cifs_sb;
1256 struct inode *inode;
1257 struct cifsFileInfo *open_file;
1259 if (!mapping || !mapping->host)
1262 inode = page->mapping->host;
1263 cifs_sb = CIFS_SB(inode->i_sb);
1265 offset += (loff_t)from;
1266 write_data = kmap(page);
1269 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1274 /* racing with truncate? */
1275 if (offset > mapping->host->i_size) {
1277 return 0; /* don't care */
1280 /* check to make sure that we are not extending the file */
1281 if (mapping->host->i_size - offset < (loff_t)to)
1282 to = (unsigned)(mapping->host->i_size - offset);
1284 open_file = find_writable_file(CIFS_I(mapping->host), false);
1286 bytes_written = cifs_write(open_file, write_data,
1287 to - from, &offset);
1288 cifsFileInfo_put(open_file);
1289 /* Does mm or vfs already set times? */
1290 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1291 if ((bytes_written > 0) && (offset))
1293 else if (bytes_written < 0)
1296 cFYI(1, "No writeable filehandles for inode");
1304 static int cifs_writepages(struct address_space *mapping,
1305 struct writeback_control *wbc)
1307 struct backing_dev_info *bdi = mapping->backing_dev_info;
1308 unsigned int bytes_to_write;
1309 unsigned int bytes_written;
1310 struct cifs_sb_info *cifs_sb;
1314 int range_whole = 0;
1321 struct cifsFileInfo *open_file;
1322 struct cifsTconInfo *tcon;
1323 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1325 struct pagevec pvec;
1331 * BB: Is this meaningful for a non-block-device file system?
1332 * If it is, we should test it again after we do I/O
1334 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1335 wbc->encountered_congestion = 1;
1339 cifs_sb = CIFS_SB(mapping->host->i_sb);
1342 * If wsize is smaller that the page cache size, default to writing
1343 * one page at a time via cifs_writepage
1345 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1346 return generic_writepages(mapping, wbc);
1348 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1350 return generic_writepages(mapping, wbc);
1353 * if there's no open file, then this is likely to fail too,
1354 * but it'll at least handle the return. Maybe it should be
1357 open_file = find_writable_file(CIFS_I(mapping->host), false);
1360 return generic_writepages(mapping, wbc);
1363 tcon = tlink_tcon(open_file->tlink);
1364 if (!experimEnabled && tcon->ses->server->secMode &
1365 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1366 cifsFileInfo_put(open_file);
1367 return generic_writepages(mapping, wbc);
1369 cifsFileInfo_put(open_file);
1373 pagevec_init(&pvec, 0);
1374 if (wbc->range_cyclic) {
1375 index = mapping->writeback_index; /* Start from prev offset */
1378 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1379 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1380 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1385 while (!done && (index <= end) &&
1386 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1387 PAGECACHE_TAG_DIRTY,
1388 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1397 for (i = 0; i < nr_pages; i++) {
1398 page = pvec.pages[i];
1400 * At this point we hold neither mapping->tree_lock nor
1401 * lock on the page itself: the page may be truncated or
1402 * invalidated (changing page->mapping to NULL), or even
1403 * swizzled back from swapper_space to tmpfs file
1409 else if (!trylock_page(page))
1412 if (unlikely(page->mapping != mapping)) {
1417 if (!wbc->range_cyclic && page->index > end) {
1423 if (next && (page->index != next)) {
1424 /* Not next consecutive page */
1429 if (wbc->sync_mode != WB_SYNC_NONE)
1430 wait_on_page_writeback(page);
1432 if (PageWriteback(page) ||
1433 !clear_page_dirty_for_io(page)) {
1439 * This actually clears the dirty bit in the radix tree.
1440 * See cifs_writepage() for more commentary.
1442 set_page_writeback(page);
1444 if (page_offset(page) >= mapping->host->i_size) {
1447 end_page_writeback(page);
1452 * BB can we get rid of this? pages are held by pvec
1454 page_cache_get(page);
1456 len = min(mapping->host->i_size - page_offset(page),
1457 (loff_t)PAGE_CACHE_SIZE);
1459 /* reserve iov[0] for the smb header */
1461 iov[n_iov].iov_base = kmap(page);
1462 iov[n_iov].iov_len = len;
1463 bytes_to_write += len;
1467 offset = page_offset(page);
1469 next = page->index + 1;
1470 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1474 open_file = find_writable_file(CIFS_I(mapping->host),
1477 cERROR(1, "No writable handles for inode");
1480 long_op = cifs_write_timeout(cifsi, offset);
1481 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1482 bytes_to_write, offset,
1483 &bytes_written, iov, n_iov,
1485 cifsFileInfo_put(open_file);
1486 cifs_update_eof(cifsi, offset, bytes_written);
1489 if (rc || bytes_written < bytes_to_write) {
1490 cERROR(1, "Write2 ret %d, wrote %d",
1492 /* BB what if continued retry is
1493 requested via mount flags? */
1495 set_bit(AS_ENOSPC, &mapping->flags);
1497 set_bit(AS_EIO, &mapping->flags);
1499 cifs_stats_bytes_written(tcon, bytes_written);
1502 for (i = 0; i < n_iov; i++) {
1503 page = pvec.pages[first + i];
1504 /* Should we also set page error on
1505 success rc but too little data written? */
1506 /* BB investigate retry logic on temporary
1507 server crash cases and how recovery works
1508 when page marked as error */
1513 end_page_writeback(page);
1514 page_cache_release(page);
1516 if ((wbc->nr_to_write -= n_iov) <= 0)
1520 /* Need to re-find the pages we skipped */
1521 index = pvec.pages[0]->index + 1;
1523 pagevec_release(&pvec);
1525 if (!scanned && !done) {
1527 * We hit the last page and there is more work to be done: wrap
1528 * back to the start of the file
1534 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1535 mapping->writeback_index = index;
1542 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1548 /* BB add check for wbc flags */
1549 page_cache_get(page);
1550 if (!PageUptodate(page))
1551 cFYI(1, "ppw - page not up to date");
1554 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1556 * A writepage() implementation always needs to do either this,
1557 * or re-dirty the page with "redirty_page_for_writepage()" in
1558 * the case of a failure.
1560 * Just unlocking the page will cause the radix tree tag-bits
1561 * to fail to update with the state of the page correctly.
1563 set_page_writeback(page);
1564 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1565 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1567 end_page_writeback(page);
1568 page_cache_release(page);
1573 static int cifs_write_end(struct file *file, struct address_space *mapping,
1574 loff_t pos, unsigned len, unsigned copied,
1575 struct page *page, void *fsdata)
1578 struct inode *inode = mapping->host;
1580 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1583 if (PageChecked(page)) {
1585 SetPageUptodate(page);
1586 ClearPageChecked(page);
1587 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1588 SetPageUptodate(page);
1590 if (!PageUptodate(page)) {
1592 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1596 /* this is probably better than directly calling
1597 partialpage_write since in this function the file handle is
1598 known which we might as well leverage */
1599 /* BB check if anything else missing out of ppw
1600 such as updating last write time */
1601 page_data = kmap(page);
1602 rc = cifs_write(file->private_data, page_data + offset,
1604 /* if (rc < 0) should we set writebehind rc? */
1611 set_page_dirty(page);
1615 spin_lock(&inode->i_lock);
1616 if (pos > inode->i_size)
1617 i_size_write(inode, pos);
1618 spin_unlock(&inode->i_lock);
1622 page_cache_release(page);
1627 int cifs_fsync(struct file *file, int datasync)
1631 struct cifsTconInfo *tcon;
1632 struct cifsFileInfo *smbfile = file->private_data;
1633 struct inode *inode = file->f_path.dentry->d_inode;
1637 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1638 file->f_path.dentry->d_name.name, datasync);
1640 rc = filemap_write_and_wait(inode->i_mapping);
1642 rc = CIFS_I(inode)->write_behind_rc;
1643 CIFS_I(inode)->write_behind_rc = 0;
1644 tcon = tlink_tcon(smbfile->tlink);
1645 if (!rc && tcon && smbfile &&
1646 !(CIFS_SB(inode->i_sb)->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1647 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1654 /* static void cifs_sync_page(struct page *page)
1656 struct address_space *mapping;
1657 struct inode *inode;
1658 unsigned long index = page->index;
1659 unsigned int rpages = 0;
1662 cFYI(1, "sync page %p", page);
1663 mapping = page->mapping;
1666 inode = mapping->host;
1670 /* fill in rpages then
1671 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1673 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1683 * As file closes, flush all cached write data for this inode checking
1684 * for write behind errors.
1686 int cifs_flush(struct file *file, fl_owner_t id)
1688 struct inode *inode = file->f_path.dentry->d_inode;
1691 if (file->f_mode & FMODE_WRITE) {
1692 rc = filemap_write_and_wait(inode->i_mapping);
1693 /* reset wb rc if we were able to write out dirty pages */
1695 rc = CIFS_I(inode)->write_behind_rc;
1696 CIFS_I(inode)->write_behind_rc = 0;
1700 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1705 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1706 size_t read_size, loff_t *poffset)
1709 unsigned int bytes_read = 0;
1710 unsigned int total_read = 0;
1711 unsigned int current_read_size;
1712 struct cifs_sb_info *cifs_sb;
1713 struct cifsTconInfo *pTcon;
1715 struct cifsFileInfo *open_file;
1716 char *smb_read_data;
1717 char __user *current_offset;
1718 struct smb_com_read_rsp *pSMBr;
1721 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1723 if (file->private_data == NULL) {
1728 open_file = file->private_data;
1729 pTcon = tlink_tcon(open_file->tlink);
1731 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1732 cFYI(1, "attempting read on write only file instance");
1734 for (total_read = 0, current_offset = read_data;
1735 read_size > total_read;
1736 total_read += bytes_read, current_offset += bytes_read) {
1737 current_read_size = min_t(const int, read_size - total_read,
1740 smb_read_data = NULL;
1741 while (rc == -EAGAIN) {
1742 int buf_type = CIFS_NO_BUFFER;
1743 if (open_file->invalidHandle) {
1744 rc = cifs_reopen_file(open_file, true);
1748 rc = CIFSSMBRead(xid, pTcon,
1750 current_read_size, *poffset,
1751 &bytes_read, &smb_read_data,
1753 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1754 if (smb_read_data) {
1755 if (copy_to_user(current_offset,
1757 4 /* RFC1001 length field */ +
1758 le16_to_cpu(pSMBr->DataOffset),
1762 if (buf_type == CIFS_SMALL_BUFFER)
1763 cifs_small_buf_release(smb_read_data);
1764 else if (buf_type == CIFS_LARGE_BUFFER)
1765 cifs_buf_release(smb_read_data);
1766 smb_read_data = NULL;
1769 if (rc || (bytes_read == 0)) {
1777 cifs_stats_bytes_read(pTcon, bytes_read);
1778 *poffset += bytes_read;
1786 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1790 unsigned int bytes_read = 0;
1791 unsigned int total_read;
1792 unsigned int current_read_size;
1793 struct cifs_sb_info *cifs_sb;
1794 struct cifsTconInfo *pTcon;
1796 char *current_offset;
1797 struct cifsFileInfo *open_file;
1798 int buf_type = CIFS_NO_BUFFER;
1801 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1803 if (file->private_data == NULL) {
1808 open_file = file->private_data;
1809 pTcon = tlink_tcon(open_file->tlink);
1811 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1812 cFYI(1, "attempting read on write only file instance");
1814 for (total_read = 0, current_offset = read_data;
1815 read_size > total_read;
1816 total_read += bytes_read, current_offset += bytes_read) {
1817 current_read_size = min_t(const int, read_size - total_read,
1819 /* For windows me and 9x we do not want to request more
1820 than it negotiated since it will refuse the read then */
1822 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1823 current_read_size = min_t(const int, current_read_size,
1824 pTcon->ses->server->maxBuf - 128);
1827 while (rc == -EAGAIN) {
1828 if (open_file->invalidHandle) {
1829 rc = cifs_reopen_file(open_file, true);
1833 rc = CIFSSMBRead(xid, pTcon,
1835 current_read_size, *poffset,
1836 &bytes_read, ¤t_offset,
1839 if (rc || (bytes_read == 0)) {
1847 cifs_stats_bytes_read(pTcon, total_read);
1848 *poffset += bytes_read;
1855 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1860 rc = cifs_revalidate_file(file);
1862 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1866 rc = generic_file_mmap(file, vma);
1872 static void cifs_copy_cache_pages(struct address_space *mapping,
1873 struct list_head *pages, int bytes_read, char *data)
1878 while (bytes_read > 0) {
1879 if (list_empty(pages))
1882 page = list_entry(pages->prev, struct page, lru);
1883 list_del(&page->lru);
1885 if (add_to_page_cache_lru(page, mapping, page->index,
1887 page_cache_release(page);
1888 cFYI(1, "Add page cache failed");
1889 data += PAGE_CACHE_SIZE;
1890 bytes_read -= PAGE_CACHE_SIZE;
1893 page_cache_release(page);
1895 target = kmap_atomic(page, KM_USER0);
1897 if (PAGE_CACHE_SIZE > bytes_read) {
1898 memcpy(target, data, bytes_read);
1899 /* zero the tail end of this partial page */
1900 memset(target + bytes_read, 0,
1901 PAGE_CACHE_SIZE - bytes_read);
1904 memcpy(target, data, PAGE_CACHE_SIZE);
1905 bytes_read -= PAGE_CACHE_SIZE;
1907 kunmap_atomic(target, KM_USER0);
1909 flush_dcache_page(page);
1910 SetPageUptodate(page);
1912 data += PAGE_CACHE_SIZE;
1914 /* add page to FS-Cache */
1915 cifs_readpage_to_fscache(mapping->host, page);
1920 static int cifs_readpages(struct file *file, struct address_space *mapping,
1921 struct list_head *page_list, unsigned num_pages)
1927 struct cifs_sb_info *cifs_sb;
1928 struct cifsTconInfo *pTcon;
1929 unsigned int bytes_read = 0;
1930 unsigned int read_size, i;
1931 char *smb_read_data = NULL;
1932 struct smb_com_read_rsp *pSMBr;
1933 struct cifsFileInfo *open_file;
1934 int buf_type = CIFS_NO_BUFFER;
1937 if (file->private_data == NULL) {
1942 open_file = file->private_data;
1943 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1944 pTcon = tlink_tcon(open_file->tlink);
1947 * Reads as many pages as possible from fscache. Returns -ENOBUFS
1948 * immediately if the cookie is negative
1950 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
1955 cFYI(DBG2, "rpages: num pages %d", num_pages);
1956 for (i = 0; i < num_pages; ) {
1957 unsigned contig_pages;
1958 struct page *tmp_page;
1959 unsigned long expected_index;
1961 if (list_empty(page_list))
1964 page = list_entry(page_list->prev, struct page, lru);
1965 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1967 /* count adjacent pages that we will read into */
1970 list_entry(page_list->prev, struct page, lru)->index;
1971 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1972 if (tmp_page->index == expected_index) {
1978 if (contig_pages + i > num_pages)
1979 contig_pages = num_pages - i;
1981 /* for reads over a certain size could initiate async
1984 read_size = contig_pages * PAGE_CACHE_SIZE;
1985 /* Read size needs to be in multiples of one page */
1986 read_size = min_t(const unsigned int, read_size,
1987 cifs_sb->rsize & PAGE_CACHE_MASK);
1988 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
1989 read_size, contig_pages);
1991 while (rc == -EAGAIN) {
1992 if (open_file->invalidHandle) {
1993 rc = cifs_reopen_file(open_file, true);
1998 rc = CIFSSMBRead(xid, pTcon,
2001 &bytes_read, &smb_read_data,
2003 /* BB more RC checks ? */
2004 if (rc == -EAGAIN) {
2005 if (smb_read_data) {
2006 if (buf_type == CIFS_SMALL_BUFFER)
2007 cifs_small_buf_release(smb_read_data);
2008 else if (buf_type == CIFS_LARGE_BUFFER)
2009 cifs_buf_release(smb_read_data);
2010 smb_read_data = NULL;
2014 if ((rc < 0) || (smb_read_data == NULL)) {
2015 cFYI(1, "Read error in readpages: %d", rc);
2017 } else if (bytes_read > 0) {
2018 task_io_account_read(bytes_read);
2019 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2020 cifs_copy_cache_pages(mapping, page_list, bytes_read,
2021 smb_read_data + 4 /* RFC1001 hdr */ +
2022 le16_to_cpu(pSMBr->DataOffset));
2024 i += bytes_read >> PAGE_CACHE_SHIFT;
2025 cifs_stats_bytes_read(pTcon, bytes_read);
2026 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2027 i++; /* account for partial page */
2029 /* server copy of file can have smaller size
2031 /* BB do we need to verify this common case ?
2032 this case is ok - if we are at server EOF
2033 we will hit it on next read */
2038 cFYI(1, "No bytes read (%d) at offset %lld . "
2039 "Cleaning remaining pages from readahead list",
2040 bytes_read, offset);
2041 /* BB turn off caching and do new lookup on
2042 file size at server? */
2045 if (smb_read_data) {
2046 if (buf_type == CIFS_SMALL_BUFFER)
2047 cifs_small_buf_release(smb_read_data);
2048 else if (buf_type == CIFS_LARGE_BUFFER)
2049 cifs_buf_release(smb_read_data);
2050 smb_read_data = NULL;
2055 /* need to free smb_read_data buf before exit */
2056 if (smb_read_data) {
2057 if (buf_type == CIFS_SMALL_BUFFER)
2058 cifs_small_buf_release(smb_read_data);
2059 else if (buf_type == CIFS_LARGE_BUFFER)
2060 cifs_buf_release(smb_read_data);
2061 smb_read_data = NULL;
2069 static int cifs_readpage_worker(struct file *file, struct page *page,
2075 /* Is the page cached? */
2076 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2080 page_cache_get(page);
2081 read_data = kmap(page);
2082 /* for reads over a certain size could initiate async read ahead */
2084 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2089 cFYI(1, "Bytes read %d", rc);
2091 file->f_path.dentry->d_inode->i_atime =
2092 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2094 if (PAGE_CACHE_SIZE > rc)
2095 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2097 flush_dcache_page(page);
2098 SetPageUptodate(page);
2100 /* send this page to the cache */
2101 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2107 page_cache_release(page);
2113 static int cifs_readpage(struct file *file, struct page *page)
2115 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2121 if (file->private_data == NULL) {
2127 cFYI(1, "readpage %p at offset %d 0x%x\n",
2128 page, (int)offset, (int)offset);
2130 rc = cifs_readpage_worker(file, page, &offset);
2138 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2140 struct cifsFileInfo *open_file;
2142 spin_lock(&cifs_file_list_lock);
2143 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2144 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2145 spin_unlock(&cifs_file_list_lock);
2149 spin_unlock(&cifs_file_list_lock);
2153 /* We do not want to update the file size from server for inodes
2154 open for write - to avoid races with writepage extending
2155 the file - in the future we could consider allowing
2156 refreshing the inode only on increases in the file size
2157 but this is tricky to do without racing with writebehind
2158 page caching in the current Linux kernel design */
2159 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2164 if (is_inode_writable(cifsInode)) {
2165 /* This inode is open for write at least once */
2166 struct cifs_sb_info *cifs_sb;
2168 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2169 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2170 /* since no page cache to corrupt on directio
2171 we can change size safely */
2175 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2183 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2184 loff_t pos, unsigned len, unsigned flags,
2185 struct page **pagep, void **fsdata)
2187 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2188 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2189 loff_t page_start = pos & PAGE_MASK;
2194 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2196 page = grab_cache_page_write_begin(mapping, index, flags);
2202 if (PageUptodate(page))
2206 * If we write a full page it will be up to date, no need to read from
2207 * the server. If the write is short, we'll end up doing a sync write
2210 if (len == PAGE_CACHE_SIZE)
2214 * optimize away the read when we have an oplock, and we're not
2215 * expecting to use any of the data we'd be reading in. That
2216 * is, when the page lies beyond the EOF, or straddles the EOF
2217 * and the write will cover all of the existing data.
2219 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2220 i_size = i_size_read(mapping->host);
2221 if (page_start >= i_size ||
2222 (offset == 0 && (pos + len) >= i_size)) {
2223 zero_user_segments(page, 0, offset,
2227 * PageChecked means that the parts of the page
2228 * to which we're not writing are considered up
2229 * to date. Once the data is copied to the
2230 * page, it can be set uptodate.
2232 SetPageChecked(page);
2237 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2239 * might as well read a page, it is fast enough. If we get
2240 * an error, we don't need to return it. cifs_write_end will
2241 * do a sync write instead since PG_uptodate isn't set.
2243 cifs_readpage_worker(file, page, &page_start);
2245 /* we could try using another file handle if there is one -
2246 but how would we lock it to prevent close of that handle
2247 racing with this read? In any case
2248 this will be written out by write_end so is fine */
2255 static int cifs_release_page(struct page *page, gfp_t gfp)
2257 if (PagePrivate(page))
2260 return cifs_fscache_release_page(page, gfp);
2263 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2265 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2268 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2271 void cifs_oplock_break(struct work_struct *work)
2273 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2275 struct inode *inode = cfile->dentry->d_inode;
2276 struct cifsInodeInfo *cinode = CIFS_I(inode);
2279 if (inode && S_ISREG(inode->i_mode)) {
2280 if (cinode->clientCanCacheRead)
2281 break_lease(inode, O_RDONLY);
2283 break_lease(inode, O_WRONLY);
2284 rc = filemap_fdatawrite(inode->i_mapping);
2285 if (cinode->clientCanCacheRead == 0) {
2286 waitrc = filemap_fdatawait(inode->i_mapping);
2287 invalidate_remote_inode(inode);
2292 cinode->write_behind_rc = rc;
2293 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2297 * releasing stale oplock after recent reconnect of smb session using
2298 * a now incorrect file handle is not a data integrity issue but do
2299 * not bother sending an oplock release if session to server still is
2300 * disconnected since oplock already released by the server
2302 if (!cfile->oplock_break_cancelled) {
2303 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2304 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false);
2305 cFYI(1, "Oplock release rc = %d", rc);
2309 * We might have kicked in before is_valid_oplock_break()
2310 * finished grabbing reference for us. Make sure it's done by
2311 * waiting for GlobalSMSSeslock.
2313 spin_lock(&cifs_file_list_lock);
2314 spin_unlock(&cifs_file_list_lock);
2316 cifs_oplock_break_put(cfile);
2319 /* must be called while holding cifs_file_list_lock */
2320 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2322 cifs_sb_active(cfile->dentry->d_sb);
2323 cifsFileInfo_get(cfile);
2326 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2328 cifsFileInfo_put(cfile);
2329 cifs_sb_deactive(cfile->dentry->d_sb);
2332 const struct address_space_operations cifs_addr_ops = {
2333 .readpage = cifs_readpage,
2334 .readpages = cifs_readpages,
2335 .writepage = cifs_writepage,
2336 .writepages = cifs_writepages,
2337 .write_begin = cifs_write_begin,
2338 .write_end = cifs_write_end,
2339 .set_page_dirty = __set_page_dirty_nobuffers,
2340 .releasepage = cifs_release_page,
2341 .invalidatepage = cifs_invalidate_page,
2342 /* .sync_page = cifs_sync_page, */
2347 * cifs_readpages requires the server to support a buffer large enough to
2348 * contain the header plus one complete page of data. Otherwise, we need
2349 * to leave cifs_readpages out of the address space operations.
2351 const struct address_space_operations cifs_addr_ops_smallbuf = {
2352 .readpage = cifs_readpage,
2353 .writepage = cifs_writepage,
2354 .writepages = cifs_writepages,
2355 .write_begin = cifs_write_begin,
2356 .write_end = cifs_write_end,
2357 .set_page_dirty = __set_page_dirty_nobuffers,
2358 .releasepage = cifs_release_page,
2359 .invalidatepage = cifs_invalidate_page,
2360 /* .sync_page = cifs_sync_page, */