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 int cifs_open(struct inode *inode, struct file *file)
227 struct cifs_sb_info *cifs_sb;
228 struct cifsTconInfo *tcon;
229 struct tcon_link *tlink;
230 struct cifsFileInfo *pCifsFile = NULL;
231 struct cifsInodeInfo *pCifsInode;
232 char *full_path = NULL;
236 FILE_ALL_INFO *buf = NULL;
240 cifs_sb = CIFS_SB(inode->i_sb);
241 tlink = cifs_sb_tlink(cifs_sb);
244 return PTR_ERR(tlink);
246 tcon = tlink_tcon(tlink);
248 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
250 full_path = build_path_from_dentry(file->f_path.dentry);
251 if (full_path == NULL) {
256 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
257 inode, file->f_flags, full_path);
264 if (!tcon->broken_posix_open && tcon->unix_ext &&
265 (tcon->ses->capabilities & CAP_UNIX) &&
266 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
267 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
268 /* can not refresh inode info since size could be stale */
269 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
270 cifs_sb->mnt_file_mode /* ignored */,
271 file->f_flags, &oplock, &netfid, xid);
273 cFYI(1, "posix open succeeded");
275 pCifsFile = cifs_new_fileinfo(inode, netfid, file,
276 tlink, oflags, oplock);
277 if (pCifsFile == NULL) {
278 CIFSSMBClose(xid, tcon, netfid);
282 cifs_fscache_set_inode_cookie(inode, file);
285 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
286 if (tcon->ses->serverNOS)
287 cERROR(1, "server %s of type %s returned"
288 " unexpected error on SMB posix open"
289 ", disabling posix open support."
290 " Check if server update available.",
291 tcon->ses->serverName,
292 tcon->ses->serverNOS);
293 tcon->broken_posix_open = true;
294 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
295 (rc != -EOPNOTSUPP)) /* path not found or net err */
297 /* else fallthrough to retry open the old way on network i/o
301 desiredAccess = cifs_convert_flags(file->f_flags);
303 /*********************************************************************
304 * open flag mapping table:
306 * POSIX Flag CIFS Disposition
307 * ---------- ----------------
308 * O_CREAT FILE_OPEN_IF
309 * O_CREAT | O_EXCL FILE_CREATE
310 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
311 * O_TRUNC FILE_OVERWRITE
312 * none of the above FILE_OPEN
314 * Note that there is not a direct match between disposition
315 * FILE_SUPERSEDE (ie create whether or not file exists although
316 * O_CREAT | O_TRUNC is similar but truncates the existing
317 * file rather than creating a new file as FILE_SUPERSEDE does
318 * (which uses the attributes / metadata passed in on open call)
320 *? O_SYNC is a reasonable match to CIFS writethrough flag
321 *? and the read write flags match reasonably. O_LARGEFILE
322 *? is irrelevant because largefile support is always used
323 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
324 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
325 *********************************************************************/
327 disposition = cifs_get_disposition(file->f_flags);
329 /* BB pass O_SYNC flag through on file attributes .. BB */
331 /* Also refresh inode by passing in file_info buf returned by SMBOpen
332 and calling get_inode_info with returned buf (at least helps
333 non-Unix server case) */
335 /* BB we can not do this if this is the second open of a file
336 and the first handle has writebehind data, we might be
337 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
338 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
344 if (tcon->ses->capabilities & CAP_NT_SMBS)
345 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
346 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
347 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
348 & CIFS_MOUNT_MAP_SPECIAL_CHR);
350 rc = -EIO; /* no NT SMB support fall into legacy open below */
353 /* Old server, try legacy style OpenX */
354 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
355 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
356 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
357 & CIFS_MOUNT_MAP_SPECIAL_CHR);
360 cFYI(1, "cifs_open returned 0x%x", rc);
364 rc = cifs_open_inode_helper(inode, tcon, oplock, buf, full_path, xid);
368 pCifsFile = cifs_new_fileinfo(inode, netfid, file, tlink,
369 file->f_flags, oplock);
370 if (pCifsFile == NULL) {
375 cifs_fscache_set_inode_cookie(inode, file);
377 if (oplock & CIFS_CREATE_ACTION) {
378 /* time to set mode which we can not set earlier due to
379 problems creating new read-only files */
380 if (tcon->unix_ext) {
381 struct cifs_unix_set_info_args args = {
382 .mode = inode->i_mode,
385 .ctime = NO_CHANGE_64,
386 .atime = NO_CHANGE_64,
387 .mtime = NO_CHANGE_64,
390 CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
392 cifs_sb->mnt_cifs_flags &
393 CIFS_MOUNT_MAP_SPECIAL_CHR);
401 cifs_put_tlink(tlink);
405 /* Try to reacquire byte range locks that were released when session */
406 /* to server was lost */
407 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
411 /* BB list all locks open on this file and relock */
416 static int cifs_reopen_file(struct file *file, bool can_flush)
421 struct cifs_sb_info *cifs_sb;
422 struct cifsTconInfo *tcon;
423 struct cifsFileInfo *pCifsFile;
424 struct cifsInodeInfo *pCifsInode;
426 char *full_path = NULL;
428 int disposition = FILE_OPEN;
431 if (file->private_data)
432 pCifsFile = file->private_data;
437 mutex_lock(&pCifsFile->fh_mutex);
438 if (!pCifsFile->invalidHandle) {
439 mutex_unlock(&pCifsFile->fh_mutex);
445 if (file->f_path.dentry == NULL) {
446 cERROR(1, "no valid name if dentry freed");
449 goto reopen_error_exit;
452 inode = file->f_path.dentry->d_inode;
454 cERROR(1, "inode not valid");
457 goto reopen_error_exit;
460 cifs_sb = CIFS_SB(inode->i_sb);
461 tcon = tlink_tcon(pCifsFile->tlink);
463 /* can not grab rename sem here because various ops, including
464 those that already have the rename sem can end up causing writepage
465 to get called and if the server was down that means we end up here,
466 and we can never tell if the caller already has the rename_sem */
467 full_path = build_path_from_dentry(file->f_path.dentry);
468 if (full_path == NULL) {
471 mutex_unlock(&pCifsFile->fh_mutex);
476 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
477 inode, file->f_flags, full_path);
484 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
485 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
486 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
489 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
490 * original open. Must mask them off for a reopen.
492 unsigned int oflags = file->f_flags & ~(O_CREAT|O_EXCL|O_TRUNC);
494 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
495 cifs_sb->mnt_file_mode /* ignored */,
496 oflags, &oplock, &netfid, xid);
498 cFYI(1, "posix reopen succeeded");
501 /* fallthrough to retry open the old way on errors, especially
502 in the reconnect path it is important to retry hard */
505 desiredAccess = cifs_convert_flags(file->f_flags);
507 /* Can not refresh inode by passing in file_info buf to be returned
508 by SMBOpen and then calling get_inode_info with returned buf
509 since file might have write behind data that needs to be flushed
510 and server version of file size can be stale. If we knew for sure
511 that inode was not dirty locally we could do this */
513 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
514 CREATE_NOT_DIR, &netfid, &oplock, NULL,
515 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
516 CIFS_MOUNT_MAP_SPECIAL_CHR);
518 mutex_unlock(&pCifsFile->fh_mutex);
519 cFYI(1, "cifs_open returned 0x%x", rc);
520 cFYI(1, "oplock: %d", oplock);
523 pCifsFile->netfid = netfid;
524 pCifsFile->invalidHandle = false;
525 mutex_unlock(&pCifsFile->fh_mutex);
526 pCifsInode = CIFS_I(inode);
529 rc = filemap_write_and_wait(inode->i_mapping);
531 CIFS_I(inode)->write_behind_rc = rc;
532 /* temporarily disable caching while we
533 go to server to get inode info */
534 pCifsInode->clientCanCacheAll = false;
535 pCifsInode->clientCanCacheRead = false;
537 rc = cifs_get_inode_info_unix(&inode,
538 full_path, inode->i_sb, xid);
540 rc = cifs_get_inode_info(&inode,
541 full_path, NULL, inode->i_sb,
543 } /* else we are writing out data to server already
544 and could deadlock if we tried to flush data, and
545 since we do not know if we have data that would
546 invalidate the current end of file on the server
547 we can not go to the server to get the new inod
549 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
550 pCifsInode->clientCanCacheAll = true;
551 pCifsInode->clientCanCacheRead = true;
552 cFYI(1, "Exclusive Oplock granted on inode %p",
553 file->f_path.dentry->d_inode);
554 } else if ((oplock & 0xF) == OPLOCK_READ) {
555 pCifsInode->clientCanCacheRead = true;
556 pCifsInode->clientCanCacheAll = false;
558 pCifsInode->clientCanCacheRead = false;
559 pCifsInode->clientCanCacheAll = false;
561 cifs_relock_file(pCifsFile);
569 int cifs_close(struct inode *inode, struct file *file)
573 struct cifs_sb_info *cifs_sb;
574 struct cifsTconInfo *pTcon;
575 struct cifsFileInfo *pSMBFile = file->private_data;
579 cifs_sb = CIFS_SB(inode->i_sb);
580 pTcon = tlink_tcon(pSMBFile->tlink);
582 struct cifsLockInfo *li, *tmp;
583 write_lock(&GlobalSMBSeslock);
584 pSMBFile->closePend = true;
586 /* no sense reconnecting to close a file that is
588 if (!pTcon->need_reconnect) {
589 write_unlock(&GlobalSMBSeslock);
591 while ((atomic_read(&pSMBFile->count) != 1)
592 && (timeout <= 2048)) {
593 /* Give write a better chance to get to
594 server ahead of the close. We do not
595 want to add a wait_q here as it would
596 increase the memory utilization as
597 the struct would be in each open file,
598 but this should give enough time to
600 cFYI(DBG2, "close delay, write pending");
604 if (!pTcon->need_reconnect &&
605 !pSMBFile->invalidHandle)
606 rc = CIFSSMBClose(xid, pTcon,
609 write_unlock(&GlobalSMBSeslock);
611 write_unlock(&GlobalSMBSeslock);
613 /* Delete any outstanding lock records.
614 We'll lose them when the file is closed anyway. */
615 mutex_lock(&pSMBFile->lock_mutex);
616 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
617 list_del(&li->llist);
620 mutex_unlock(&pSMBFile->lock_mutex);
622 write_lock(&GlobalSMBSeslock);
623 list_del(&pSMBFile->flist);
624 list_del(&pSMBFile->tlist);
625 write_unlock(&GlobalSMBSeslock);
626 cifsFileInfo_put(file->private_data);
627 file->private_data = NULL;
631 read_lock(&GlobalSMBSeslock);
632 if (list_empty(&(CIFS_I(inode)->openFileList))) {
633 cFYI(1, "closing last open instance for inode %p", inode);
634 /* if the file is not open we do not know if we can cache info
635 on this inode, much less write behind and read ahead */
636 CIFS_I(inode)->clientCanCacheRead = false;
637 CIFS_I(inode)->clientCanCacheAll = false;
639 read_unlock(&GlobalSMBSeslock);
640 if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
641 rc = CIFS_I(inode)->write_behind_rc;
646 int cifs_closedir(struct inode *inode, struct file *file)
650 struct cifsFileInfo *pCFileStruct = file->private_data;
653 cFYI(1, "Closedir inode = 0x%p", inode);
658 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
660 cFYI(1, "Freeing private data in close dir");
661 write_lock(&GlobalSMBSeslock);
662 if (!pCFileStruct->srch_inf.endOfSearch &&
663 !pCFileStruct->invalidHandle) {
664 pCFileStruct->invalidHandle = true;
665 write_unlock(&GlobalSMBSeslock);
666 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
667 cFYI(1, "Closing uncompleted readdir with rc %d",
669 /* not much we can do if it fails anyway, ignore rc */
672 write_unlock(&GlobalSMBSeslock);
673 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
675 cFYI(1, "closedir free smb buf in srch struct");
676 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
677 if (pCFileStruct->srch_inf.smallBuf)
678 cifs_small_buf_release(ptmp);
680 cifs_buf_release(ptmp);
682 cifs_put_tlink(pCFileStruct->tlink);
683 kfree(file->private_data);
684 file->private_data = NULL;
686 /* BB can we lock the filestruct while this is going on? */
691 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
692 __u64 offset, __u8 lockType)
694 struct cifsLockInfo *li =
695 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
701 mutex_lock(&fid->lock_mutex);
702 list_add(&li->llist, &fid->llist);
703 mutex_unlock(&fid->lock_mutex);
707 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
713 bool wait_flag = false;
714 struct cifs_sb_info *cifs_sb;
715 struct cifsTconInfo *tcon;
717 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
718 bool posix_locking = 0;
720 length = 1 + pfLock->fl_end - pfLock->fl_start;
724 cFYI(1, "Lock parm: 0x%x flockflags: "
725 "0x%x flocktype: 0x%x start: %lld end: %lld",
726 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
729 if (pfLock->fl_flags & FL_POSIX)
731 if (pfLock->fl_flags & FL_FLOCK)
733 if (pfLock->fl_flags & FL_SLEEP) {
734 cFYI(1, "Blocking lock");
737 if (pfLock->fl_flags & FL_ACCESS)
738 cFYI(1, "Process suspended by mandatory locking - "
739 "not implemented yet");
740 if (pfLock->fl_flags & FL_LEASE)
741 cFYI(1, "Lease on file - not implemented yet");
742 if (pfLock->fl_flags &
743 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
744 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
746 if (pfLock->fl_type == F_WRLCK) {
749 } else if (pfLock->fl_type == F_UNLCK) {
752 /* Check if unlock includes more than
754 } else if (pfLock->fl_type == F_RDLCK) {
756 lockType |= LOCKING_ANDX_SHARED_LOCK;
758 } else if (pfLock->fl_type == F_EXLCK) {
761 } else if (pfLock->fl_type == F_SHLCK) {
763 lockType |= LOCKING_ANDX_SHARED_LOCK;
766 cFYI(1, "Unknown type of lock");
768 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
769 tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
771 if (file->private_data == NULL) {
776 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
778 if ((tcon->ses->capabilities & CAP_UNIX) &&
779 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
780 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
782 /* BB add code here to normalize offset and length to
783 account for negative length which we can not accept over the
788 if (lockType & LOCKING_ANDX_SHARED_LOCK)
789 posix_lock_type = CIFS_RDLCK;
791 posix_lock_type = CIFS_WRLCK;
792 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
794 posix_lock_type, wait_flag);
799 /* BB we could chain these into one lock request BB */
800 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
801 0, 1, lockType, 0 /* wait flag */ );
803 rc = CIFSSMBLock(xid, tcon, netfid, length,
804 pfLock->fl_start, 1 /* numUnlock */ ,
805 0 /* numLock */ , lockType,
807 pfLock->fl_type = F_UNLCK;
809 cERROR(1, "Error unlocking previously locked "
810 "range %d during test of lock", rc);
814 /* if rc == ERR_SHARING_VIOLATION ? */
817 if (lockType & LOCKING_ANDX_SHARED_LOCK) {
818 pfLock->fl_type = F_WRLCK;
820 rc = CIFSSMBLock(xid, tcon, netfid, length,
821 pfLock->fl_start, 0, 1,
822 lockType | LOCKING_ANDX_SHARED_LOCK,
825 rc = CIFSSMBLock(xid, tcon, netfid,
826 length, pfLock->fl_start, 1, 0,
828 LOCKING_ANDX_SHARED_LOCK,
830 pfLock->fl_type = F_RDLCK;
832 cERROR(1, "Error unlocking "
833 "previously locked range %d "
834 "during test of lock", rc);
837 pfLock->fl_type = F_WRLCK;
847 if (!numLock && !numUnlock) {
848 /* if no lock or unlock then nothing
849 to do since we do not know what it is */
856 if (lockType & LOCKING_ANDX_SHARED_LOCK)
857 posix_lock_type = CIFS_RDLCK;
859 posix_lock_type = CIFS_WRLCK;
862 posix_lock_type = CIFS_UNLCK;
864 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
866 posix_lock_type, wait_flag);
868 struct cifsFileInfo *fid = file->private_data;
871 rc = CIFSSMBLock(xid, tcon, netfid, length,
873 0, numLock, lockType, wait_flag);
876 /* For Windows locks we must store them. */
877 rc = store_file_lock(fid, length,
878 pfLock->fl_start, lockType);
880 } else if (numUnlock) {
881 /* For each stored lock that this unlock overlaps
882 completely, unlock it. */
884 struct cifsLockInfo *li, *tmp;
887 mutex_lock(&fid->lock_mutex);
888 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
889 if (pfLock->fl_start <= li->offset &&
890 (pfLock->fl_start + length) >=
891 (li->offset + li->length)) {
892 stored_rc = CIFSSMBLock(xid, tcon,
894 li->length, li->offset,
895 1, 0, li->type, false);
899 list_del(&li->llist);
904 mutex_unlock(&fid->lock_mutex);
908 if (pfLock->fl_flags & FL_POSIX)
909 posix_lock_file_wait(file, pfLock);
915 * Set the timeout on write requests past EOF. For some servers (Windows)
916 * these calls can be very long.
918 * If we're writing >10M past the EOF we give a 180s timeout. Anything less
919 * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
920 * The 10M cutoff is totally arbitrary. A better scheme for this would be
921 * welcome if someone wants to suggest one.
923 * We may be able to do a better job with this if there were some way to
924 * declare that a file should be sparse.
927 cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
929 if (offset <= cifsi->server_eof)
931 else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
932 return CIFS_VLONG_OP;
937 /* update the file size (if needed) after a write */
939 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
940 unsigned int bytes_written)
942 loff_t end_of_write = offset + bytes_written;
944 if (end_of_write > cifsi->server_eof)
945 cifsi->server_eof = end_of_write;
948 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
949 size_t write_size, loff_t *poffset)
952 unsigned int bytes_written = 0;
953 unsigned int total_written;
954 struct cifs_sb_info *cifs_sb;
955 struct cifsTconInfo *pTcon;
957 struct cifsFileInfo *open_file;
958 struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
960 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
962 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
963 *poffset, file->f_path.dentry->d_name.name); */
965 if (file->private_data == NULL)
968 open_file = file->private_data;
969 pTcon = tlink_tcon(open_file->tlink);
971 rc = generic_write_checks(file, poffset, &write_size, 0);
977 long_op = cifs_write_timeout(cifsi, *poffset);
978 for (total_written = 0; write_size > total_written;
979 total_written += bytes_written) {
981 while (rc == -EAGAIN) {
982 if (file->private_data == NULL) {
983 /* file has been closed on us */
985 /* if we have gotten here we have written some data
986 and blocked, and the file has been freed on us while
987 we blocked so return what we managed to write */
988 return total_written;
990 if (open_file->closePend) {
993 return total_written;
997 if (open_file->invalidHandle) {
998 /* we could deadlock if we called
999 filemap_fdatawait from here so tell
1000 reopen_file not to flush data to server
1002 rc = cifs_reopen_file(file, false);
1007 rc = CIFSSMBWrite(xid, pTcon,
1009 min_t(const int, cifs_sb->wsize,
1010 write_size - total_written),
1011 *poffset, &bytes_written,
1012 NULL, write_data + total_written, long_op);
1014 if (rc || (bytes_written == 0)) {
1022 cifs_update_eof(cifsi, *poffset, bytes_written);
1023 *poffset += bytes_written;
1025 long_op = CIFS_STD_OP; /* subsequent writes fast -
1026 15 seconds is plenty */
1029 cifs_stats_bytes_written(pTcon, total_written);
1031 /* since the write may have blocked check these pointers again */
1032 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1033 struct inode *inode = file->f_path.dentry->d_inode;
1034 /* Do not update local mtime - server will set its actual value on write
1035 * inode->i_ctime = inode->i_mtime =
1036 * current_fs_time(inode->i_sb);*/
1037 if (total_written > 0) {
1038 spin_lock(&inode->i_lock);
1039 if (*poffset > file->f_path.dentry->d_inode->i_size)
1040 i_size_write(file->f_path.dentry->d_inode,
1042 spin_unlock(&inode->i_lock);
1044 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1047 return total_written;
1050 static ssize_t cifs_write(struct file *file, const char *write_data,
1051 size_t write_size, loff_t *poffset)
1054 unsigned int bytes_written = 0;
1055 unsigned int total_written;
1056 struct cifs_sb_info *cifs_sb;
1057 struct cifsTconInfo *pTcon;
1059 struct cifsFileInfo *open_file;
1060 struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
1062 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1064 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
1065 *poffset, file->f_path.dentry->d_name.name);
1067 if (file->private_data == NULL)
1069 open_file = file->private_data;
1070 pTcon = tlink_tcon(open_file->tlink);
1074 long_op = cifs_write_timeout(cifsi, *poffset);
1075 for (total_written = 0; write_size > total_written;
1076 total_written += bytes_written) {
1078 while (rc == -EAGAIN) {
1079 if (file->private_data == NULL) {
1080 /* file has been closed on us */
1082 /* if we have gotten here we have written some data
1083 and blocked, and the file has been freed on us
1084 while we blocked so return what we managed to
1086 return total_written;
1088 if (open_file->closePend) {
1091 return total_written;
1095 if (open_file->invalidHandle) {
1096 /* we could deadlock if we called
1097 filemap_fdatawait from here so tell
1098 reopen_file not to flush data to
1100 rc = cifs_reopen_file(file, false);
1104 if (experimEnabled || (pTcon->ses->server &&
1105 ((pTcon->ses->server->secMode &
1106 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1111 len = min((size_t)cifs_sb->wsize,
1112 write_size - total_written);
1113 /* iov[0] is reserved for smb header */
1114 iov[1].iov_base = (char *)write_data +
1116 iov[1].iov_len = len;
1117 rc = CIFSSMBWrite2(xid, pTcon,
1118 open_file->netfid, len,
1119 *poffset, &bytes_written,
1122 rc = CIFSSMBWrite(xid, pTcon,
1124 min_t(const int, cifs_sb->wsize,
1125 write_size - total_written),
1126 *poffset, &bytes_written,
1127 write_data + total_written,
1130 if (rc || (bytes_written == 0)) {
1138 cifs_update_eof(cifsi, *poffset, bytes_written);
1139 *poffset += bytes_written;
1141 long_op = CIFS_STD_OP; /* subsequent writes fast -
1142 15 seconds is plenty */
1145 cifs_stats_bytes_written(pTcon, total_written);
1147 /* since the write may have blocked check these pointers again */
1148 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1149 /*BB We could make this contingent on superblock ATIME flag too */
1150 /* file->f_path.dentry->d_inode->i_ctime =
1151 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1152 if (total_written > 0) {
1153 spin_lock(&file->f_path.dentry->d_inode->i_lock);
1154 if (*poffset > file->f_path.dentry->d_inode->i_size)
1155 i_size_write(file->f_path.dentry->d_inode,
1157 spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1159 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1162 return total_written;
1165 #ifdef CONFIG_CIFS_EXPERIMENTAL
1166 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1169 struct cifsFileInfo *open_file = NULL;
1170 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1172 /* only filter by fsuid on multiuser mounts */
1173 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1176 read_lock(&GlobalSMBSeslock);
1177 /* we could simply get the first_list_entry since write-only entries
1178 are always at the end of the list but since the first entry might
1179 have a close pending, we go through the whole list */
1180 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1181 if (open_file->closePend)
1183 if (fsuid_only && open_file->uid != current_fsuid())
1185 if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
1186 (open_file->pfile->f_flags & O_RDONLY))) {
1187 if (!open_file->invalidHandle) {
1188 /* found a good file */
1189 /* lock it so it will not be closed on us */
1190 cifsFileInfo_get(open_file);
1191 read_unlock(&GlobalSMBSeslock);
1193 } /* else might as well continue, and look for
1194 another, or simply have the caller reopen it
1195 again rather than trying to fix this handle */
1196 } else /* write only file */
1197 break; /* write only files are last so must be done */
1199 read_unlock(&GlobalSMBSeslock);
1204 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1207 struct cifsFileInfo *open_file;
1208 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1209 bool any_available = false;
1212 /* Having a null inode here (because mapping->host was set to zero by
1213 the VFS or MM) should not happen but we had reports of on oops (due to
1214 it being zero) during stress testcases so we need to check for it */
1216 if (cifs_inode == NULL) {
1217 cERROR(1, "Null inode passed to cifs_writeable_file");
1222 /* only filter by fsuid on multiuser mounts */
1223 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1226 read_lock(&GlobalSMBSeslock);
1228 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1229 if (open_file->closePend)
1231 if (!any_available && open_file->pid != current->tgid)
1233 if (fsuid_only && open_file->uid != current_fsuid())
1235 if (open_file->pfile &&
1236 ((open_file->pfile->f_flags & O_RDWR) ||
1237 (open_file->pfile->f_flags & O_WRONLY))) {
1238 cifsFileInfo_get(open_file);
1240 if (!open_file->invalidHandle) {
1241 /* found a good writable file */
1242 read_unlock(&GlobalSMBSeslock);
1246 read_unlock(&GlobalSMBSeslock);
1247 /* Had to unlock since following call can block */
1248 rc = cifs_reopen_file(open_file->pfile, false);
1250 if (!open_file->closePend)
1252 else { /* start over in case this was deleted */
1253 /* since the list could be modified */
1254 read_lock(&GlobalSMBSeslock);
1255 cifsFileInfo_put(open_file);
1256 goto refind_writable;
1260 /* if it fails, try another handle if possible -
1261 (we can not do this if closePending since
1262 loop could be modified - in which case we
1263 have to start at the beginning of the list
1264 again. Note that it would be bad
1265 to hold up writepages here (rather than
1266 in caller) with continuous retries */
1267 cFYI(1, "wp failed on reopen file");
1268 read_lock(&GlobalSMBSeslock);
1269 /* can not use this handle, no write
1270 pending on this one after all */
1271 cifsFileInfo_put(open_file);
1273 if (open_file->closePend) /* list could have changed */
1274 goto refind_writable;
1275 /* else we simply continue to the next entry. Thus
1276 we do not loop on reopen errors. If we
1277 can not reopen the file, for example if we
1278 reconnected to a server with another client
1279 racing to delete or lock the file we would not
1280 make progress if we restarted before the beginning
1281 of the loop here. */
1284 /* couldn't find useable FH with same pid, try any available */
1285 if (!any_available) {
1286 any_available = true;
1287 goto refind_writable;
1289 read_unlock(&GlobalSMBSeslock);
1293 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1295 struct address_space *mapping = page->mapping;
1296 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1299 int bytes_written = 0;
1300 struct cifs_sb_info *cifs_sb;
1301 struct inode *inode;
1302 struct cifsFileInfo *open_file;
1304 if (!mapping || !mapping->host)
1307 inode = page->mapping->host;
1308 cifs_sb = CIFS_SB(inode->i_sb);
1310 offset += (loff_t)from;
1311 write_data = kmap(page);
1314 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1319 /* racing with truncate? */
1320 if (offset > mapping->host->i_size) {
1322 return 0; /* don't care */
1325 /* check to make sure that we are not extending the file */
1326 if (mapping->host->i_size - offset < (loff_t)to)
1327 to = (unsigned)(mapping->host->i_size - offset);
1329 open_file = find_writable_file(CIFS_I(mapping->host), false);
1331 bytes_written = cifs_write(open_file->pfile, write_data,
1333 cifsFileInfo_put(open_file);
1334 /* Does mm or vfs already set times? */
1335 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1336 if ((bytes_written > 0) && (offset))
1338 else if (bytes_written < 0)
1341 cFYI(1, "No writeable filehandles for inode");
1349 static int cifs_writepages(struct address_space *mapping,
1350 struct writeback_control *wbc)
1352 struct backing_dev_info *bdi = mapping->backing_dev_info;
1353 unsigned int bytes_to_write;
1354 unsigned int bytes_written;
1355 struct cifs_sb_info *cifs_sb;
1359 int range_whole = 0;
1366 struct cifsFileInfo *open_file;
1367 struct cifsTconInfo *tcon;
1368 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1370 struct pagevec pvec;
1376 * BB: Is this meaningful for a non-block-device file system?
1377 * If it is, we should test it again after we do I/O
1379 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1380 wbc->encountered_congestion = 1;
1384 cifs_sb = CIFS_SB(mapping->host->i_sb);
1387 * If wsize is smaller that the page cache size, default to writing
1388 * one page at a time via cifs_writepage
1390 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1391 return generic_writepages(mapping, wbc);
1393 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1395 return generic_writepages(mapping, wbc);
1398 * if there's no open file, then this is likely to fail too,
1399 * but it'll at least handle the return. Maybe it should be
1402 open_file = find_writable_file(CIFS_I(mapping->host), false);
1405 return generic_writepages(mapping, wbc);
1408 tcon = tlink_tcon(open_file->tlink);
1409 if (!experimEnabled && tcon->ses->server->secMode &
1410 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1411 cifsFileInfo_put(open_file);
1412 return generic_writepages(mapping, wbc);
1414 cifsFileInfo_put(open_file);
1418 pagevec_init(&pvec, 0);
1419 if (wbc->range_cyclic) {
1420 index = mapping->writeback_index; /* Start from prev offset */
1423 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1424 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1425 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1430 while (!done && (index <= end) &&
1431 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1432 PAGECACHE_TAG_DIRTY,
1433 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1442 for (i = 0; i < nr_pages; i++) {
1443 page = pvec.pages[i];
1445 * At this point we hold neither mapping->tree_lock nor
1446 * lock on the page itself: the page may be truncated or
1447 * invalidated (changing page->mapping to NULL), or even
1448 * swizzled back from swapper_space to tmpfs file
1454 else if (!trylock_page(page))
1457 if (unlikely(page->mapping != mapping)) {
1462 if (!wbc->range_cyclic && page->index > end) {
1468 if (next && (page->index != next)) {
1469 /* Not next consecutive page */
1474 if (wbc->sync_mode != WB_SYNC_NONE)
1475 wait_on_page_writeback(page);
1477 if (PageWriteback(page) ||
1478 !clear_page_dirty_for_io(page)) {
1484 * This actually clears the dirty bit in the radix tree.
1485 * See cifs_writepage() for more commentary.
1487 set_page_writeback(page);
1489 if (page_offset(page) >= mapping->host->i_size) {
1492 end_page_writeback(page);
1497 * BB can we get rid of this? pages are held by pvec
1499 page_cache_get(page);
1501 len = min(mapping->host->i_size - page_offset(page),
1502 (loff_t)PAGE_CACHE_SIZE);
1504 /* reserve iov[0] for the smb header */
1506 iov[n_iov].iov_base = kmap(page);
1507 iov[n_iov].iov_len = len;
1508 bytes_to_write += len;
1512 offset = page_offset(page);
1514 next = page->index + 1;
1515 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1519 open_file = find_writable_file(CIFS_I(mapping->host),
1522 cERROR(1, "No writable handles for inode");
1525 long_op = cifs_write_timeout(cifsi, offset);
1526 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1527 bytes_to_write, offset,
1528 &bytes_written, iov, n_iov,
1530 cifsFileInfo_put(open_file);
1531 cifs_update_eof(cifsi, offset, bytes_written);
1534 if (rc || bytes_written < bytes_to_write) {
1535 cERROR(1, "Write2 ret %d, wrote %d",
1537 /* BB what if continued retry is
1538 requested via mount flags? */
1540 set_bit(AS_ENOSPC, &mapping->flags);
1542 set_bit(AS_EIO, &mapping->flags);
1544 cifs_stats_bytes_written(tcon, bytes_written);
1547 for (i = 0; i < n_iov; i++) {
1548 page = pvec.pages[first + i];
1549 /* Should we also set page error on
1550 success rc but too little data written? */
1551 /* BB investigate retry logic on temporary
1552 server crash cases and how recovery works
1553 when page marked as error */
1558 end_page_writeback(page);
1559 page_cache_release(page);
1561 if ((wbc->nr_to_write -= n_iov) <= 0)
1565 /* Need to re-find the pages we skipped */
1566 index = pvec.pages[0]->index + 1;
1568 pagevec_release(&pvec);
1570 if (!scanned && !done) {
1572 * We hit the last page and there is more work to be done: wrap
1573 * back to the start of the file
1579 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1580 mapping->writeback_index = index;
1587 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1593 /* BB add check for wbc flags */
1594 page_cache_get(page);
1595 if (!PageUptodate(page))
1596 cFYI(1, "ppw - page not up to date");
1599 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1601 * A writepage() implementation always needs to do either this,
1602 * or re-dirty the page with "redirty_page_for_writepage()" in
1603 * the case of a failure.
1605 * Just unlocking the page will cause the radix tree tag-bits
1606 * to fail to update with the state of the page correctly.
1608 set_page_writeback(page);
1609 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1610 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1612 end_page_writeback(page);
1613 page_cache_release(page);
1618 static int cifs_write_end(struct file *file, struct address_space *mapping,
1619 loff_t pos, unsigned len, unsigned copied,
1620 struct page *page, void *fsdata)
1623 struct inode *inode = mapping->host;
1625 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1628 if (PageChecked(page)) {
1630 SetPageUptodate(page);
1631 ClearPageChecked(page);
1632 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1633 SetPageUptodate(page);
1635 if (!PageUptodate(page)) {
1637 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1641 /* this is probably better than directly calling
1642 partialpage_write since in this function the file handle is
1643 known which we might as well leverage */
1644 /* BB check if anything else missing out of ppw
1645 such as updating last write time */
1646 page_data = kmap(page);
1647 rc = cifs_write(file, page_data + offset, copied, &pos);
1648 /* if (rc < 0) should we set writebehind rc? */
1655 set_page_dirty(page);
1659 spin_lock(&inode->i_lock);
1660 if (pos > inode->i_size)
1661 i_size_write(inode, pos);
1662 spin_unlock(&inode->i_lock);
1666 page_cache_release(page);
1671 int cifs_fsync(struct file *file, int datasync)
1675 struct cifsTconInfo *tcon;
1676 struct cifsFileInfo *smbfile = file->private_data;
1677 struct inode *inode = file->f_path.dentry->d_inode;
1681 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1682 file->f_path.dentry->d_name.name, datasync);
1684 rc = filemap_write_and_wait(inode->i_mapping);
1686 rc = CIFS_I(inode)->write_behind_rc;
1687 CIFS_I(inode)->write_behind_rc = 0;
1688 tcon = tlink_tcon(smbfile->tlink);
1689 if (!rc && tcon && smbfile &&
1690 !(CIFS_SB(inode->i_sb)->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1691 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1698 /* static void cifs_sync_page(struct page *page)
1700 struct address_space *mapping;
1701 struct inode *inode;
1702 unsigned long index = page->index;
1703 unsigned int rpages = 0;
1706 cFYI(1, "sync page %p", page);
1707 mapping = page->mapping;
1710 inode = mapping->host;
1714 /* fill in rpages then
1715 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1717 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1727 * As file closes, flush all cached write data for this inode checking
1728 * for write behind errors.
1730 int cifs_flush(struct file *file, fl_owner_t id)
1732 struct inode *inode = file->f_path.dentry->d_inode;
1735 /* Rather than do the steps manually:
1736 lock the inode for writing
1737 loop through pages looking for write behind data (dirty pages)
1738 coalesce into contiguous 16K (or smaller) chunks to write to server
1739 send to server (prefer in parallel)
1740 deal with writebehind errors
1741 unlock inode for writing
1742 filemapfdatawrite appears easier for the time being */
1744 rc = filemap_fdatawrite(inode->i_mapping);
1745 /* reset wb rc if we were able to write out dirty pages */
1747 rc = CIFS_I(inode)->write_behind_rc;
1748 CIFS_I(inode)->write_behind_rc = 0;
1751 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1756 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1757 size_t read_size, loff_t *poffset)
1760 unsigned int bytes_read = 0;
1761 unsigned int total_read = 0;
1762 unsigned int current_read_size;
1763 struct cifs_sb_info *cifs_sb;
1764 struct cifsTconInfo *pTcon;
1766 struct cifsFileInfo *open_file;
1767 char *smb_read_data;
1768 char __user *current_offset;
1769 struct smb_com_read_rsp *pSMBr;
1772 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1774 if (file->private_data == NULL) {
1779 open_file = file->private_data;
1780 pTcon = tlink_tcon(open_file->tlink);
1782 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1783 cFYI(1, "attempting read on write only file instance");
1785 for (total_read = 0, current_offset = read_data;
1786 read_size > total_read;
1787 total_read += bytes_read, current_offset += bytes_read) {
1788 current_read_size = min_t(const int, read_size - total_read,
1791 smb_read_data = NULL;
1792 while (rc == -EAGAIN) {
1793 int buf_type = CIFS_NO_BUFFER;
1794 if ((open_file->invalidHandle) &&
1795 (!open_file->closePend)) {
1796 rc = cifs_reopen_file(file, true);
1800 rc = CIFSSMBRead(xid, pTcon,
1802 current_read_size, *poffset,
1803 &bytes_read, &smb_read_data,
1805 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1806 if (smb_read_data) {
1807 if (copy_to_user(current_offset,
1809 4 /* RFC1001 length field */ +
1810 le16_to_cpu(pSMBr->DataOffset),
1814 if (buf_type == CIFS_SMALL_BUFFER)
1815 cifs_small_buf_release(smb_read_data);
1816 else if (buf_type == CIFS_LARGE_BUFFER)
1817 cifs_buf_release(smb_read_data);
1818 smb_read_data = NULL;
1821 if (rc || (bytes_read == 0)) {
1829 cifs_stats_bytes_read(pTcon, bytes_read);
1830 *poffset += bytes_read;
1838 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1842 unsigned int bytes_read = 0;
1843 unsigned int total_read;
1844 unsigned int current_read_size;
1845 struct cifs_sb_info *cifs_sb;
1846 struct cifsTconInfo *pTcon;
1848 char *current_offset;
1849 struct cifsFileInfo *open_file;
1850 int buf_type = CIFS_NO_BUFFER;
1853 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1855 if (file->private_data == NULL) {
1860 open_file = file->private_data;
1861 pTcon = tlink_tcon(open_file->tlink);
1863 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1864 cFYI(1, "attempting read on write only file instance");
1866 for (total_read = 0, current_offset = read_data;
1867 read_size > total_read;
1868 total_read += bytes_read, current_offset += bytes_read) {
1869 current_read_size = min_t(const int, read_size - total_read,
1871 /* For windows me and 9x we do not want to request more
1872 than it negotiated since it will refuse the read then */
1874 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1875 current_read_size = min_t(const int, current_read_size,
1876 pTcon->ses->server->maxBuf - 128);
1879 while (rc == -EAGAIN) {
1880 if ((open_file->invalidHandle) &&
1881 (!open_file->closePend)) {
1882 rc = cifs_reopen_file(file, true);
1886 rc = CIFSSMBRead(xid, pTcon,
1888 current_read_size, *poffset,
1889 &bytes_read, ¤t_offset,
1892 if (rc || (bytes_read == 0)) {
1900 cifs_stats_bytes_read(pTcon, total_read);
1901 *poffset += bytes_read;
1908 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1913 rc = cifs_revalidate_file(file);
1915 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1919 rc = generic_file_mmap(file, vma);
1925 static void cifs_copy_cache_pages(struct address_space *mapping,
1926 struct list_head *pages, int bytes_read, char *data)
1931 while (bytes_read > 0) {
1932 if (list_empty(pages))
1935 page = list_entry(pages->prev, struct page, lru);
1936 list_del(&page->lru);
1938 if (add_to_page_cache_lru(page, mapping, page->index,
1940 page_cache_release(page);
1941 cFYI(1, "Add page cache failed");
1942 data += PAGE_CACHE_SIZE;
1943 bytes_read -= PAGE_CACHE_SIZE;
1946 page_cache_release(page);
1948 target = kmap_atomic(page, KM_USER0);
1950 if (PAGE_CACHE_SIZE > bytes_read) {
1951 memcpy(target, data, bytes_read);
1952 /* zero the tail end of this partial page */
1953 memset(target + bytes_read, 0,
1954 PAGE_CACHE_SIZE - bytes_read);
1957 memcpy(target, data, PAGE_CACHE_SIZE);
1958 bytes_read -= PAGE_CACHE_SIZE;
1960 kunmap_atomic(target, KM_USER0);
1962 flush_dcache_page(page);
1963 SetPageUptodate(page);
1965 data += PAGE_CACHE_SIZE;
1967 /* add page to FS-Cache */
1968 cifs_readpage_to_fscache(mapping->host, page);
1973 static int cifs_readpages(struct file *file, struct address_space *mapping,
1974 struct list_head *page_list, unsigned num_pages)
1980 struct cifs_sb_info *cifs_sb;
1981 struct cifsTconInfo *pTcon;
1982 unsigned int bytes_read = 0;
1983 unsigned int read_size, i;
1984 char *smb_read_data = NULL;
1985 struct smb_com_read_rsp *pSMBr;
1986 struct cifsFileInfo *open_file;
1987 int buf_type = CIFS_NO_BUFFER;
1990 if (file->private_data == NULL) {
1995 open_file = file->private_data;
1996 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1997 pTcon = tlink_tcon(open_file->tlink);
2000 * Reads as many pages as possible from fscache. Returns -ENOBUFS
2001 * immediately if the cookie is negative
2003 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
2008 cFYI(DBG2, "rpages: num pages %d", num_pages);
2009 for (i = 0; i < num_pages; ) {
2010 unsigned contig_pages;
2011 struct page *tmp_page;
2012 unsigned long expected_index;
2014 if (list_empty(page_list))
2017 page = list_entry(page_list->prev, struct page, lru);
2018 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2020 /* count adjacent pages that we will read into */
2023 list_entry(page_list->prev, struct page, lru)->index;
2024 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2025 if (tmp_page->index == expected_index) {
2031 if (contig_pages + i > num_pages)
2032 contig_pages = num_pages - i;
2034 /* for reads over a certain size could initiate async
2037 read_size = contig_pages * PAGE_CACHE_SIZE;
2038 /* Read size needs to be in multiples of one page */
2039 read_size = min_t(const unsigned int, read_size,
2040 cifs_sb->rsize & PAGE_CACHE_MASK);
2041 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
2042 read_size, contig_pages);
2044 while (rc == -EAGAIN) {
2045 if ((open_file->invalidHandle) &&
2046 (!open_file->closePend)) {
2047 rc = cifs_reopen_file(file, true);
2052 rc = CIFSSMBRead(xid, pTcon,
2055 &bytes_read, &smb_read_data,
2057 /* BB more RC checks ? */
2058 if (rc == -EAGAIN) {
2059 if (smb_read_data) {
2060 if (buf_type == CIFS_SMALL_BUFFER)
2061 cifs_small_buf_release(smb_read_data);
2062 else if (buf_type == CIFS_LARGE_BUFFER)
2063 cifs_buf_release(smb_read_data);
2064 smb_read_data = NULL;
2068 if ((rc < 0) || (smb_read_data == NULL)) {
2069 cFYI(1, "Read error in readpages: %d", rc);
2071 } else if (bytes_read > 0) {
2072 task_io_account_read(bytes_read);
2073 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2074 cifs_copy_cache_pages(mapping, page_list, bytes_read,
2075 smb_read_data + 4 /* RFC1001 hdr */ +
2076 le16_to_cpu(pSMBr->DataOffset));
2078 i += bytes_read >> PAGE_CACHE_SHIFT;
2079 cifs_stats_bytes_read(pTcon, bytes_read);
2080 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2081 i++; /* account for partial page */
2083 /* server copy of file can have smaller size
2085 /* BB do we need to verify this common case ?
2086 this case is ok - if we are at server EOF
2087 we will hit it on next read */
2092 cFYI(1, "No bytes read (%d) at offset %lld . "
2093 "Cleaning remaining pages from readahead list",
2094 bytes_read, offset);
2095 /* BB turn off caching and do new lookup on
2096 file size at server? */
2099 if (smb_read_data) {
2100 if (buf_type == CIFS_SMALL_BUFFER)
2101 cifs_small_buf_release(smb_read_data);
2102 else if (buf_type == CIFS_LARGE_BUFFER)
2103 cifs_buf_release(smb_read_data);
2104 smb_read_data = NULL;
2109 /* need to free smb_read_data buf before exit */
2110 if (smb_read_data) {
2111 if (buf_type == CIFS_SMALL_BUFFER)
2112 cifs_small_buf_release(smb_read_data);
2113 else if (buf_type == CIFS_LARGE_BUFFER)
2114 cifs_buf_release(smb_read_data);
2115 smb_read_data = NULL;
2123 static int cifs_readpage_worker(struct file *file, struct page *page,
2129 /* Is the page cached? */
2130 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2134 page_cache_get(page);
2135 read_data = kmap(page);
2136 /* for reads over a certain size could initiate async read ahead */
2138 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2143 cFYI(1, "Bytes read %d", rc);
2145 file->f_path.dentry->d_inode->i_atime =
2146 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2148 if (PAGE_CACHE_SIZE > rc)
2149 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2151 flush_dcache_page(page);
2152 SetPageUptodate(page);
2154 /* send this page to the cache */
2155 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2161 page_cache_release(page);
2167 static int cifs_readpage(struct file *file, struct page *page)
2169 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2175 if (file->private_data == NULL) {
2181 cFYI(1, "readpage %p at offset %d 0x%x\n",
2182 page, (int)offset, (int)offset);
2184 rc = cifs_readpage_worker(file, page, &offset);
2192 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2194 struct cifsFileInfo *open_file;
2196 read_lock(&GlobalSMBSeslock);
2197 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2198 if (open_file->closePend)
2200 if (open_file->pfile &&
2201 ((open_file->pfile->f_flags & O_RDWR) ||
2202 (open_file->pfile->f_flags & O_WRONLY))) {
2203 read_unlock(&GlobalSMBSeslock);
2207 read_unlock(&GlobalSMBSeslock);
2211 /* We do not want to update the file size from server for inodes
2212 open for write - to avoid races with writepage extending
2213 the file - in the future we could consider allowing
2214 refreshing the inode only on increases in the file size
2215 but this is tricky to do without racing with writebehind
2216 page caching in the current Linux kernel design */
2217 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2222 if (is_inode_writable(cifsInode)) {
2223 /* This inode is open for write at least once */
2224 struct cifs_sb_info *cifs_sb;
2226 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2227 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2228 /* since no page cache to corrupt on directio
2229 we can change size safely */
2233 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2241 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2242 loff_t pos, unsigned len, unsigned flags,
2243 struct page **pagep, void **fsdata)
2245 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2246 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2247 loff_t page_start = pos & PAGE_MASK;
2252 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2254 page = grab_cache_page_write_begin(mapping, index, flags);
2260 if (PageUptodate(page))
2264 * If we write a full page it will be up to date, no need to read from
2265 * the server. If the write is short, we'll end up doing a sync write
2268 if (len == PAGE_CACHE_SIZE)
2272 * optimize away the read when we have an oplock, and we're not
2273 * expecting to use any of the data we'd be reading in. That
2274 * is, when the page lies beyond the EOF, or straddles the EOF
2275 * and the write will cover all of the existing data.
2277 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2278 i_size = i_size_read(mapping->host);
2279 if (page_start >= i_size ||
2280 (offset == 0 && (pos + len) >= i_size)) {
2281 zero_user_segments(page, 0, offset,
2285 * PageChecked means that the parts of the page
2286 * to which we're not writing are considered up
2287 * to date. Once the data is copied to the
2288 * page, it can be set uptodate.
2290 SetPageChecked(page);
2295 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2297 * might as well read a page, it is fast enough. If we get
2298 * an error, we don't need to return it. cifs_write_end will
2299 * do a sync write instead since PG_uptodate isn't set.
2301 cifs_readpage_worker(file, page, &page_start);
2303 /* we could try using another file handle if there is one -
2304 but how would we lock it to prevent close of that handle
2305 racing with this read? In any case
2306 this will be written out by write_end so is fine */
2313 static int cifs_release_page(struct page *page, gfp_t gfp)
2315 if (PagePrivate(page))
2318 return cifs_fscache_release_page(page, gfp);
2321 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2323 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2326 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2329 void cifs_oplock_break(struct work_struct *work)
2331 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2333 struct inode *inode = cfile->dentry->d_inode;
2334 struct cifsInodeInfo *cinode = CIFS_I(inode);
2337 if (inode && S_ISREG(inode->i_mode)) {
2338 if (cinode->clientCanCacheRead)
2339 break_lease(inode, O_RDONLY);
2341 break_lease(inode, O_WRONLY);
2342 rc = filemap_fdatawrite(inode->i_mapping);
2343 if (cinode->clientCanCacheRead == 0) {
2344 waitrc = filemap_fdatawait(inode->i_mapping);
2345 invalidate_remote_inode(inode);
2350 cinode->write_behind_rc = rc;
2351 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2355 * releasing stale oplock after recent reconnect of smb session using
2356 * a now incorrect file handle is not a data integrity issue but do
2357 * not bother sending an oplock release if session to server still is
2358 * disconnected since oplock already released by the server
2360 if (!cfile->closePend && !cfile->oplock_break_cancelled) {
2361 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2362 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false);
2363 cFYI(1, "Oplock release rc = %d", rc);
2367 * We might have kicked in before is_valid_oplock_break()
2368 * finished grabbing reference for us. Make sure it's done by
2369 * waiting for GlobalSMSSeslock.
2371 write_lock(&GlobalSMBSeslock);
2372 write_unlock(&GlobalSMBSeslock);
2374 cifs_oplock_break_put(cfile);
2377 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2379 cifs_sb_active(cfile->dentry->d_sb);
2380 cifsFileInfo_get(cfile);
2383 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2385 cifsFileInfo_put(cfile);
2386 cifs_sb_deactive(cfile->dentry->d_sb);
2389 const struct address_space_operations cifs_addr_ops = {
2390 .readpage = cifs_readpage,
2391 .readpages = cifs_readpages,
2392 .writepage = cifs_writepage,
2393 .writepages = cifs_writepages,
2394 .write_begin = cifs_write_begin,
2395 .write_end = cifs_write_end,
2396 .set_page_dirty = __set_page_dirty_nobuffers,
2397 .releasepage = cifs_release_page,
2398 .invalidatepage = cifs_invalidate_page,
2399 /* .sync_page = cifs_sync_page, */
2404 * cifs_readpages requires the server to support a buffer large enough to
2405 * contain the header plus one complete page of data. Otherwise, we need
2406 * to leave cifs_readpages out of the address space operations.
2408 const struct address_space_operations cifs_addr_ops_smallbuf = {
2409 .readpage = cifs_readpage,
2410 .writepage = cifs_writepage,
2411 .writepages = cifs_writepages,
2412 .write_begin = cifs_write_begin,
2413 .write_end = cifs_write_end,
2414 .set_page_dirty = __set_page_dirty_nobuffers,
2415 .releasepage = cifs_release_page,
2416 .invalidatepage = cifs_invalidate_page,
2417 /* .sync_page = cifs_sync_page, */