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 int cifs_posix_open(char *full_path, struct inode **pinode,
108 struct super_block *sb, int mode, unsigned int f_flags,
109 __u32 *poplock, __u16 *pnetfid, int xid)
112 FILE_UNIX_BASIC_INFO *presp_data;
113 __u32 posix_flags = 0;
114 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
115 struct cifs_fattr fattr;
116 struct tcon_link *tlink;
117 struct cifsTconInfo *tcon;
119 cFYI(1, "posix open %s", full_path);
121 presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
122 if (presp_data == NULL)
125 tlink = cifs_sb_tlink(cifs_sb);
131 tcon = tlink_tcon(tlink);
132 mode &= ~current_umask();
134 posix_flags = cifs_posix_convert_flags(f_flags);
135 rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
136 poplock, full_path, cifs_sb->local_nls,
137 cifs_sb->mnt_cifs_flags &
138 CIFS_MOUNT_MAP_SPECIAL_CHR);
139 cifs_put_tlink(tlink);
144 if (presp_data->Type == cpu_to_le32(-1))
145 goto posix_open_ret; /* open ok, caller does qpathinfo */
148 goto posix_open_ret; /* caller does not need info */
150 cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
152 /* get new inode and set it up */
153 if (*pinode == NULL) {
154 cifs_fill_uniqueid(sb, &fattr);
155 *pinode = cifs_iget(sb, &fattr);
161 cifs_fattr_to_inode(*pinode, &fattr);
170 cifs_nt_open(char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
171 struct cifsTconInfo *tcon, unsigned int f_flags, __u32 *poplock,
172 __u16 *pnetfid, int xid)
179 desiredAccess = cifs_convert_flags(f_flags);
181 /*********************************************************************
182 * open flag mapping table:
184 * POSIX Flag CIFS Disposition
185 * ---------- ----------------
186 * O_CREAT FILE_OPEN_IF
187 * O_CREAT | O_EXCL FILE_CREATE
188 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
189 * O_TRUNC FILE_OVERWRITE
190 * none of the above FILE_OPEN
192 * Note that there is not a direct match between disposition
193 * FILE_SUPERSEDE (ie create whether or not file exists although
194 * O_CREAT | O_TRUNC is similar but truncates the existing
195 * file rather than creating a new file as FILE_SUPERSEDE does
196 * (which uses the attributes / metadata passed in on open call)
198 *? O_SYNC is a reasonable match to CIFS writethrough flag
199 *? and the read write flags match reasonably. O_LARGEFILE
200 *? is irrelevant because largefile support is always used
201 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
202 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
203 *********************************************************************/
205 disposition = cifs_get_disposition(f_flags);
207 /* BB pass O_SYNC flag through on file attributes .. BB */
209 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
213 if (tcon->ses->capabilities & CAP_NT_SMBS)
214 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
215 desiredAccess, CREATE_NOT_DIR, pnetfid, poplock, buf,
216 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
217 & CIFS_MOUNT_MAP_SPECIAL_CHR);
219 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
220 desiredAccess, CREATE_NOT_DIR, pnetfid, poplock, buf,
221 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
222 & CIFS_MOUNT_MAP_SPECIAL_CHR);
228 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
231 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
239 struct cifsFileInfo *
240 cifs_new_fileinfo(__u16 fileHandle, struct file *file,
241 struct tcon_link *tlink, __u32 oplock)
243 struct dentry *dentry = file->f_path.dentry;
244 struct inode *inode = dentry->d_inode;
245 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
246 struct cifsFileInfo *pCifsFile;
248 pCifsFile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
249 if (pCifsFile == NULL)
252 pCifsFile->count = 1;
253 pCifsFile->netfid = fileHandle;
254 pCifsFile->pid = current->tgid;
255 pCifsFile->uid = current_fsuid();
256 pCifsFile->dentry = dget(dentry);
257 pCifsFile->f_flags = file->f_flags;
258 pCifsFile->invalidHandle = false;
259 pCifsFile->tlink = cifs_get_tlink(tlink);
260 mutex_init(&pCifsFile->fh_mutex);
261 mutex_init(&pCifsFile->lock_mutex);
262 INIT_LIST_HEAD(&pCifsFile->llist);
263 INIT_WORK(&pCifsFile->oplock_break, cifs_oplock_break);
265 spin_lock(&cifs_file_list_lock);
266 list_add(&pCifsFile->tlist, &(tlink_tcon(tlink)->openFileList));
267 /* if readable file instance put first in list*/
268 if (file->f_mode & FMODE_READ)
269 list_add(&pCifsFile->flist, &pCifsInode->openFileList);
271 list_add_tail(&pCifsFile->flist, &pCifsInode->openFileList);
272 spin_unlock(&cifs_file_list_lock);
274 cifs_set_oplock_level(pCifsInode, oplock);
276 file->private_data = pCifsFile;
281 * Release a reference on the file private data. This may involve closing
282 * the filehandle out on the server. Must be called without holding
283 * cifs_file_list_lock.
285 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
287 struct inode *inode = cifs_file->dentry->d_inode;
288 struct cifsTconInfo *tcon = tlink_tcon(cifs_file->tlink);
289 struct cifsInodeInfo *cifsi = CIFS_I(inode);
290 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
291 struct cifsLockInfo *li, *tmp;
293 spin_lock(&cifs_file_list_lock);
294 if (--cifs_file->count > 0) {
295 spin_unlock(&cifs_file_list_lock);
299 /* remove it from the lists */
300 list_del(&cifs_file->flist);
301 list_del(&cifs_file->tlist);
303 if (list_empty(&cifsi->openFileList)) {
304 cFYI(1, "closing last open instance for inode %p",
305 cifs_file->dentry->d_inode);
307 /* in strict cache mode we need invalidate mapping on the last
308 close because it may cause a error when we open this file
309 again and get at least level II oplock */
310 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
311 CIFS_I(inode)->invalid_mapping = true;
313 cifs_set_oplock_level(cifsi, 0);
315 spin_unlock(&cifs_file_list_lock);
317 if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
321 rc = CIFSSMBClose(xid, tcon, cifs_file->netfid);
325 /* Delete any outstanding lock records. We'll lose them when the file
328 mutex_lock(&cifs_file->lock_mutex);
329 list_for_each_entry_safe(li, tmp, &cifs_file->llist, llist) {
330 list_del(&li->llist);
333 mutex_unlock(&cifs_file->lock_mutex);
335 cifs_put_tlink(cifs_file->tlink);
336 dput(cifs_file->dentry);
340 int cifs_open(struct inode *inode, struct file *file)
345 struct cifs_sb_info *cifs_sb;
346 struct cifsTconInfo *tcon;
347 struct tcon_link *tlink;
348 struct cifsFileInfo *pCifsFile = NULL;
349 char *full_path = NULL;
350 bool posix_open_ok = false;
355 cifs_sb = CIFS_SB(inode->i_sb);
356 tlink = cifs_sb_tlink(cifs_sb);
359 return PTR_ERR(tlink);
361 tcon = tlink_tcon(tlink);
363 full_path = build_path_from_dentry(file->f_path.dentry);
364 if (full_path == NULL) {
369 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
370 inode, file->f_flags, full_path);
377 if (!tcon->broken_posix_open && tcon->unix_ext &&
378 (tcon->ses->capabilities & CAP_UNIX) &&
379 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
380 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
381 /* can not refresh inode info since size could be stale */
382 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
383 cifs_sb->mnt_file_mode /* ignored */,
384 file->f_flags, &oplock, &netfid, xid);
386 cFYI(1, "posix open succeeded");
387 posix_open_ok = true;
388 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
389 if (tcon->ses->serverNOS)
390 cERROR(1, "server %s of type %s returned"
391 " unexpected error on SMB posix open"
392 ", disabling posix open support."
393 " Check if server update available.",
394 tcon->ses->serverName,
395 tcon->ses->serverNOS);
396 tcon->broken_posix_open = true;
397 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
398 (rc != -EOPNOTSUPP)) /* path not found or net err */
400 /* else fallthrough to retry open the old way on network i/o
404 if (!posix_open_ok) {
405 rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
406 file->f_flags, &oplock, &netfid, xid);
411 pCifsFile = cifs_new_fileinfo(netfid, file, tlink, oplock);
412 if (pCifsFile == NULL) {
413 CIFSSMBClose(xid, tcon, netfid);
418 cifs_fscache_set_inode_cookie(inode, file);
420 if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
421 /* time to set mode which we can not set earlier due to
422 problems creating new read-only files */
423 struct cifs_unix_set_info_args args = {
424 .mode = inode->i_mode,
427 .ctime = NO_CHANGE_64,
428 .atime = NO_CHANGE_64,
429 .mtime = NO_CHANGE_64,
432 CIFSSMBUnixSetFileInfo(xid, tcon, &args, netfid,
439 cifs_put_tlink(tlink);
443 /* Try to reacquire byte range locks that were released when session */
444 /* to server was lost */
445 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
449 /* BB list all locks open on this file and relock */
454 static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
459 struct cifs_sb_info *cifs_sb;
460 struct cifsTconInfo *tcon;
461 struct cifsInodeInfo *pCifsInode;
463 char *full_path = NULL;
465 int disposition = FILE_OPEN;
469 mutex_lock(&pCifsFile->fh_mutex);
470 if (!pCifsFile->invalidHandle) {
471 mutex_unlock(&pCifsFile->fh_mutex);
477 inode = pCifsFile->dentry->d_inode;
478 cifs_sb = CIFS_SB(inode->i_sb);
479 tcon = tlink_tcon(pCifsFile->tlink);
481 /* can not grab rename sem here because various ops, including
482 those that already have the rename sem can end up causing writepage
483 to get called and if the server was down that means we end up here,
484 and we can never tell if the caller already has the rename_sem */
485 full_path = build_path_from_dentry(pCifsFile->dentry);
486 if (full_path == NULL) {
488 mutex_unlock(&pCifsFile->fh_mutex);
493 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
494 inode, pCifsFile->f_flags, full_path);
501 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
502 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
503 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
506 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
507 * original open. Must mask them off for a reopen.
509 unsigned int oflags = pCifsFile->f_flags &
510 ~(O_CREAT | O_EXCL | O_TRUNC);
512 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
513 cifs_sb->mnt_file_mode /* ignored */,
514 oflags, &oplock, &netfid, xid);
516 cFYI(1, "posix reopen succeeded");
519 /* fallthrough to retry open the old way on errors, especially
520 in the reconnect path it is important to retry hard */
523 desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
525 /* Can not refresh inode by passing in file_info buf to be returned
526 by SMBOpen and then calling get_inode_info with returned buf
527 since file might have write behind data that needs to be flushed
528 and server version of file size can be stale. If we knew for sure
529 that inode was not dirty locally we could do this */
531 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
532 CREATE_NOT_DIR, &netfid, &oplock, NULL,
533 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
534 CIFS_MOUNT_MAP_SPECIAL_CHR);
536 mutex_unlock(&pCifsFile->fh_mutex);
537 cFYI(1, "cifs_open returned 0x%x", rc);
538 cFYI(1, "oplock: %d", oplock);
539 goto reopen_error_exit;
543 pCifsFile->netfid = netfid;
544 pCifsFile->invalidHandle = false;
545 mutex_unlock(&pCifsFile->fh_mutex);
546 pCifsInode = CIFS_I(inode);
549 rc = filemap_write_and_wait(inode->i_mapping);
550 mapping_set_error(inode->i_mapping, rc);
553 rc = cifs_get_inode_info_unix(&inode,
554 full_path, inode->i_sb, xid);
556 rc = cifs_get_inode_info(&inode,
557 full_path, NULL, inode->i_sb,
559 } /* else we are writing out data to server already
560 and could deadlock if we tried to flush data, and
561 since we do not know if we have data that would
562 invalidate the current end of file on the server
563 we can not go to the server to get the new inod
566 cifs_set_oplock_level(pCifsInode, oplock);
568 cifs_relock_file(pCifsFile);
576 int cifs_close(struct inode *inode, struct file *file)
578 if (file->private_data != NULL) {
579 cifsFileInfo_put(file->private_data);
580 file->private_data = NULL;
583 /* return code from the ->release op is always ignored */
587 int cifs_closedir(struct inode *inode, struct file *file)
591 struct cifsFileInfo *pCFileStruct = file->private_data;
594 cFYI(1, "Closedir inode = 0x%p", inode);
599 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
601 cFYI(1, "Freeing private data in close dir");
602 spin_lock(&cifs_file_list_lock);
603 if (!pCFileStruct->srch_inf.endOfSearch &&
604 !pCFileStruct->invalidHandle) {
605 pCFileStruct->invalidHandle = true;
606 spin_unlock(&cifs_file_list_lock);
607 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
608 cFYI(1, "Closing uncompleted readdir with rc %d",
610 /* not much we can do if it fails anyway, ignore rc */
613 spin_unlock(&cifs_file_list_lock);
614 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
616 cFYI(1, "closedir free smb buf in srch struct");
617 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
618 if (pCFileStruct->srch_inf.smallBuf)
619 cifs_small_buf_release(ptmp);
621 cifs_buf_release(ptmp);
623 cifs_put_tlink(pCFileStruct->tlink);
624 kfree(file->private_data);
625 file->private_data = NULL;
627 /* BB can we lock the filestruct while this is going on? */
632 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
633 __u64 offset, __u8 lockType)
635 struct cifsLockInfo *li =
636 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
642 mutex_lock(&fid->lock_mutex);
643 list_add(&li->llist, &fid->llist);
644 mutex_unlock(&fid->lock_mutex);
648 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
654 bool wait_flag = false;
655 struct cifs_sb_info *cifs_sb;
656 struct cifsTconInfo *tcon;
658 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
659 bool posix_locking = 0;
661 length = 1 + pfLock->fl_end - pfLock->fl_start;
665 cFYI(1, "Lock parm: 0x%x flockflags: "
666 "0x%x flocktype: 0x%x start: %lld end: %lld",
667 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
670 if (pfLock->fl_flags & FL_POSIX)
672 if (pfLock->fl_flags & FL_FLOCK)
674 if (pfLock->fl_flags & FL_SLEEP) {
675 cFYI(1, "Blocking lock");
678 if (pfLock->fl_flags & FL_ACCESS)
679 cFYI(1, "Process suspended by mandatory locking - "
680 "not implemented yet");
681 if (pfLock->fl_flags & FL_LEASE)
682 cFYI(1, "Lease on file - not implemented yet");
683 if (pfLock->fl_flags &
684 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
685 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
687 if (pfLock->fl_type == F_WRLCK) {
690 } else if (pfLock->fl_type == F_UNLCK) {
693 /* Check if unlock includes more than
695 } else if (pfLock->fl_type == F_RDLCK) {
697 lockType |= LOCKING_ANDX_SHARED_LOCK;
699 } else if (pfLock->fl_type == F_EXLCK) {
702 } else if (pfLock->fl_type == F_SHLCK) {
704 lockType |= LOCKING_ANDX_SHARED_LOCK;
707 cFYI(1, "Unknown type of lock");
709 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
710 tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
711 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
713 if ((tcon->ses->capabilities & CAP_UNIX) &&
714 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
715 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
717 /* BB add code here to normalize offset and length to
718 account for negative length which we can not accept over the
723 if (lockType & LOCKING_ANDX_SHARED_LOCK)
724 posix_lock_type = CIFS_RDLCK;
726 posix_lock_type = CIFS_WRLCK;
727 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
729 posix_lock_type, wait_flag);
734 /* BB we could chain these into one lock request BB */
735 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
736 0, 1, lockType, 0 /* wait flag */, 0);
738 rc = CIFSSMBLock(xid, tcon, netfid, length,
739 pfLock->fl_start, 1 /* numUnlock */ ,
740 0 /* numLock */ , lockType,
741 0 /* wait flag */, 0);
742 pfLock->fl_type = F_UNLCK;
744 cERROR(1, "Error unlocking previously locked "
745 "range %d during test of lock", rc);
749 /* if rc == ERR_SHARING_VIOLATION ? */
752 if (lockType & LOCKING_ANDX_SHARED_LOCK) {
753 pfLock->fl_type = F_WRLCK;
755 rc = CIFSSMBLock(xid, tcon, netfid, length,
756 pfLock->fl_start, 0, 1,
757 lockType | LOCKING_ANDX_SHARED_LOCK,
758 0 /* wait flag */, 0);
760 rc = CIFSSMBLock(xid, tcon, netfid,
761 length, pfLock->fl_start, 1, 0,
763 LOCKING_ANDX_SHARED_LOCK,
764 0 /* wait flag */, 0);
765 pfLock->fl_type = F_RDLCK;
767 cERROR(1, "Error unlocking "
768 "previously locked range %d "
769 "during test of lock", rc);
772 pfLock->fl_type = F_WRLCK;
782 if (!numLock && !numUnlock) {
783 /* if no lock or unlock then nothing
784 to do since we do not know what it is */
791 if (lockType & LOCKING_ANDX_SHARED_LOCK)
792 posix_lock_type = CIFS_RDLCK;
794 posix_lock_type = CIFS_WRLCK;
797 posix_lock_type = CIFS_UNLCK;
799 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
801 posix_lock_type, wait_flag);
803 struct cifsFileInfo *fid = file->private_data;
806 rc = CIFSSMBLock(xid, tcon, netfid, length,
807 pfLock->fl_start, 0, numLock, lockType,
811 /* For Windows locks we must store them. */
812 rc = store_file_lock(fid, length,
813 pfLock->fl_start, lockType);
815 } else if (numUnlock) {
816 /* For each stored lock that this unlock overlaps
817 completely, unlock it. */
819 struct cifsLockInfo *li, *tmp;
822 mutex_lock(&fid->lock_mutex);
823 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
824 if (pfLock->fl_start <= li->offset &&
825 (pfLock->fl_start + length) >=
826 (li->offset + li->length)) {
827 stored_rc = CIFSSMBLock(xid, tcon,
834 list_del(&li->llist);
839 mutex_unlock(&fid->lock_mutex);
843 if (pfLock->fl_flags & FL_POSIX)
844 posix_lock_file_wait(file, pfLock);
849 /* update the file size (if needed) after a write */
851 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
852 unsigned int bytes_written)
854 loff_t end_of_write = offset + bytes_written;
856 if (end_of_write > cifsi->server_eof)
857 cifsi->server_eof = end_of_write;
860 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
861 size_t write_size, loff_t *poffset)
863 struct inode *inode = file->f_path.dentry->d_inode;
865 unsigned int bytes_written = 0;
866 unsigned int total_written;
867 struct cifs_sb_info *cifs_sb;
868 struct cifsTconInfo *pTcon;
870 struct cifsFileInfo *open_file;
871 struct cifsInodeInfo *cifsi = CIFS_I(inode);
873 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
875 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
876 *poffset, file->f_path.dentry->d_name.name); */
878 if (file->private_data == NULL)
881 open_file = file->private_data;
882 pTcon = tlink_tcon(open_file->tlink);
884 rc = generic_write_checks(file, poffset, &write_size, 0);
890 for (total_written = 0; write_size > total_written;
891 total_written += bytes_written) {
893 while (rc == -EAGAIN) {
894 if (file->private_data == NULL) {
895 /* file has been closed on us */
897 /* if we have gotten here we have written some data
898 and blocked, and the file has been freed on us while
899 we blocked so return what we managed to write */
900 return total_written;
902 if (open_file->invalidHandle) {
903 /* we could deadlock if we called
904 filemap_fdatawait from here so tell
905 reopen_file not to flush data to server
907 rc = cifs_reopen_file(open_file, false);
912 rc = CIFSSMBWrite(xid, pTcon,
914 min_t(const int, cifs_sb->wsize,
915 write_size - total_written),
916 *poffset, &bytes_written,
917 NULL, write_data + total_written, 0);
919 if (rc || (bytes_written == 0)) {
927 cifs_update_eof(cifsi, *poffset, bytes_written);
928 *poffset += bytes_written;
932 cifs_stats_bytes_written(pTcon, total_written);
934 /* Do not update local mtime - server will set its actual value on write
935 * inode->i_ctime = inode->i_mtime =
936 * current_fs_time(inode->i_sb);*/
937 if (total_written > 0) {
938 spin_lock(&inode->i_lock);
939 if (*poffset > inode->i_size)
940 i_size_write(inode, *poffset);
941 spin_unlock(&inode->i_lock);
943 mark_inode_dirty_sync(inode);
946 return total_written;
949 static ssize_t cifs_write(struct cifsFileInfo *open_file,
950 const char *write_data, size_t write_size,
954 unsigned int bytes_written = 0;
955 unsigned int total_written;
956 struct cifs_sb_info *cifs_sb;
957 struct cifsTconInfo *pTcon;
959 struct dentry *dentry = open_file->dentry;
960 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
962 cifs_sb = CIFS_SB(dentry->d_sb);
964 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
965 *poffset, dentry->d_name.name);
967 pTcon = tlink_tcon(open_file->tlink);
971 for (total_written = 0; write_size > total_written;
972 total_written += bytes_written) {
974 while (rc == -EAGAIN) {
975 if (open_file->invalidHandle) {
976 /* we could deadlock if we called
977 filemap_fdatawait from here so tell
978 reopen_file not to flush data to
980 rc = cifs_reopen_file(open_file, false);
984 if (experimEnabled || (pTcon->ses->server &&
985 ((pTcon->ses->server->secMode &
986 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
991 len = min((size_t)cifs_sb->wsize,
992 write_size - total_written);
993 /* iov[0] is reserved for smb header */
994 iov[1].iov_base = (char *)write_data +
996 iov[1].iov_len = len;
997 rc = CIFSSMBWrite2(xid, pTcon,
998 open_file->netfid, len,
999 *poffset, &bytes_written,
1002 rc = CIFSSMBWrite(xid, pTcon,
1004 min_t(const int, cifs_sb->wsize,
1005 write_size - total_written),
1006 *poffset, &bytes_written,
1007 write_data + total_written,
1010 if (rc || (bytes_written == 0)) {
1018 cifs_update_eof(cifsi, *poffset, bytes_written);
1019 *poffset += bytes_written;
1023 cifs_stats_bytes_written(pTcon, total_written);
1025 if (total_written > 0) {
1026 spin_lock(&dentry->d_inode->i_lock);
1027 if (*poffset > dentry->d_inode->i_size)
1028 i_size_write(dentry->d_inode, *poffset);
1029 spin_unlock(&dentry->d_inode->i_lock);
1031 mark_inode_dirty_sync(dentry->d_inode);
1033 return total_written;
1036 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1039 struct cifsFileInfo *open_file = NULL;
1040 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1042 /* only filter by fsuid on multiuser mounts */
1043 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1046 spin_lock(&cifs_file_list_lock);
1047 /* we could simply get the first_list_entry since write-only entries
1048 are always at the end of the list but since the first entry might
1049 have a close pending, we go through the whole list */
1050 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1051 if (fsuid_only && open_file->uid != current_fsuid())
1053 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1054 if (!open_file->invalidHandle) {
1055 /* found a good file */
1056 /* lock it so it will not be closed on us */
1057 cifsFileInfo_get(open_file);
1058 spin_unlock(&cifs_file_list_lock);
1060 } /* else might as well continue, and look for
1061 another, or simply have the caller reopen it
1062 again rather than trying to fix this handle */
1063 } else /* write only file */
1064 break; /* write only files are last so must be done */
1066 spin_unlock(&cifs_file_list_lock);
1070 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1073 struct cifsFileInfo *open_file;
1074 struct cifs_sb_info *cifs_sb;
1075 bool any_available = false;
1078 /* Having a null inode here (because mapping->host was set to zero by
1079 the VFS or MM) should not happen but we had reports of on oops (due to
1080 it being zero) during stress testcases so we need to check for it */
1082 if (cifs_inode == NULL) {
1083 cERROR(1, "Null inode passed to cifs_writeable_file");
1088 cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1090 /* only filter by fsuid on multiuser mounts */
1091 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1094 spin_lock(&cifs_file_list_lock);
1096 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1097 if (!any_available && open_file->pid != current->tgid)
1099 if (fsuid_only && open_file->uid != current_fsuid())
1101 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1102 cifsFileInfo_get(open_file);
1104 if (!open_file->invalidHandle) {
1105 /* found a good writable file */
1106 spin_unlock(&cifs_file_list_lock);
1110 spin_unlock(&cifs_file_list_lock);
1112 /* Had to unlock since following call can block */
1113 rc = cifs_reopen_file(open_file, false);
1117 /* if it fails, try another handle if possible */
1118 cFYI(1, "wp failed on reopen file");
1119 cifsFileInfo_put(open_file);
1121 spin_lock(&cifs_file_list_lock);
1123 /* else we simply continue to the next entry. Thus
1124 we do not loop on reopen errors. If we
1125 can not reopen the file, for example if we
1126 reconnected to a server with another client
1127 racing to delete or lock the file we would not
1128 make progress if we restarted before the beginning
1129 of the loop here. */
1132 /* couldn't find useable FH with same pid, try any available */
1133 if (!any_available) {
1134 any_available = true;
1135 goto refind_writable;
1137 spin_unlock(&cifs_file_list_lock);
1141 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1143 struct address_space *mapping = page->mapping;
1144 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1147 int bytes_written = 0;
1148 struct inode *inode;
1149 struct cifsFileInfo *open_file;
1151 if (!mapping || !mapping->host)
1154 inode = page->mapping->host;
1156 offset += (loff_t)from;
1157 write_data = kmap(page);
1160 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1165 /* racing with truncate? */
1166 if (offset > mapping->host->i_size) {
1168 return 0; /* don't care */
1171 /* check to make sure that we are not extending the file */
1172 if (mapping->host->i_size - offset < (loff_t)to)
1173 to = (unsigned)(mapping->host->i_size - offset);
1175 open_file = find_writable_file(CIFS_I(mapping->host), false);
1177 bytes_written = cifs_write(open_file, write_data,
1178 to - from, &offset);
1179 cifsFileInfo_put(open_file);
1180 /* Does mm or vfs already set times? */
1181 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1182 if ((bytes_written > 0) && (offset))
1184 else if (bytes_written < 0)
1187 cFYI(1, "No writeable filehandles for inode");
1195 static int cifs_writepages(struct address_space *mapping,
1196 struct writeback_control *wbc)
1198 unsigned int bytes_to_write;
1199 unsigned int bytes_written;
1200 struct cifs_sb_info *cifs_sb;
1204 int range_whole = 0;
1211 struct cifsFileInfo *open_file;
1212 struct cifsTconInfo *tcon;
1213 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1215 struct pagevec pvec;
1220 cifs_sb = CIFS_SB(mapping->host->i_sb);
1223 * If wsize is smaller that the page cache size, default to writing
1224 * one page at a time via cifs_writepage
1226 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1227 return generic_writepages(mapping, wbc);
1229 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1231 return generic_writepages(mapping, wbc);
1234 * if there's no open file, then this is likely to fail too,
1235 * but it'll at least handle the return. Maybe it should be
1238 open_file = find_writable_file(CIFS_I(mapping->host), false);
1241 return generic_writepages(mapping, wbc);
1244 tcon = tlink_tcon(open_file->tlink);
1245 if (!experimEnabled && tcon->ses->server->secMode &
1246 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1247 cifsFileInfo_put(open_file);
1249 return generic_writepages(mapping, wbc);
1251 cifsFileInfo_put(open_file);
1255 pagevec_init(&pvec, 0);
1256 if (wbc->range_cyclic) {
1257 index = mapping->writeback_index; /* Start from prev offset */
1260 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1261 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1262 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1267 while (!done && (index <= end) &&
1268 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1269 PAGECACHE_TAG_DIRTY,
1270 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1279 for (i = 0; i < nr_pages; i++) {
1280 page = pvec.pages[i];
1282 * At this point we hold neither mapping->tree_lock nor
1283 * lock on the page itself: the page may be truncated or
1284 * invalidated (changing page->mapping to NULL), or even
1285 * swizzled back from swapper_space to tmpfs file
1291 else if (!trylock_page(page))
1294 if (unlikely(page->mapping != mapping)) {
1299 if (!wbc->range_cyclic && page->index > end) {
1305 if (next && (page->index != next)) {
1306 /* Not next consecutive page */
1311 if (wbc->sync_mode != WB_SYNC_NONE)
1312 wait_on_page_writeback(page);
1314 if (PageWriteback(page) ||
1315 !clear_page_dirty_for_io(page)) {
1321 * This actually clears the dirty bit in the radix tree.
1322 * See cifs_writepage() for more commentary.
1324 set_page_writeback(page);
1326 if (page_offset(page) >= mapping->host->i_size) {
1329 end_page_writeback(page);
1334 * BB can we get rid of this? pages are held by pvec
1336 page_cache_get(page);
1338 len = min(mapping->host->i_size - page_offset(page),
1339 (loff_t)PAGE_CACHE_SIZE);
1341 /* reserve iov[0] for the smb header */
1343 iov[n_iov].iov_base = kmap(page);
1344 iov[n_iov].iov_len = len;
1345 bytes_to_write += len;
1349 offset = page_offset(page);
1351 next = page->index + 1;
1352 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1357 open_file = find_writable_file(CIFS_I(mapping->host),
1360 cERROR(1, "No writable handles for inode");
1363 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1364 bytes_to_write, offset,
1365 &bytes_written, iov, n_iov,
1367 cifsFileInfo_put(open_file);
1370 cFYI(1, "Write2 rc=%d, wrote=%u", rc, bytes_written);
1373 * For now, treat a short write as if nothing got
1374 * written. A zero length write however indicates
1375 * ENOSPC or EFBIG. We have no way to know which
1376 * though, so call it ENOSPC for now. EFBIG would
1377 * get translated to AS_EIO anyway.
1379 * FIXME: make it take into account the data that did
1383 if (bytes_written == 0)
1385 else if (bytes_written < bytes_to_write)
1389 /* retry on data-integrity flush */
1390 if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN)
1393 /* fix the stats and EOF */
1394 if (bytes_written > 0) {
1395 cifs_stats_bytes_written(tcon, bytes_written);
1396 cifs_update_eof(cifsi, offset, bytes_written);
1399 for (i = 0; i < n_iov; i++) {
1400 page = pvec.pages[first + i];
1401 /* on retryable write error, redirty page */
1403 redirty_page_for_writepage(wbc, page);
1408 end_page_writeback(page);
1409 page_cache_release(page);
1413 mapping_set_error(mapping, rc);
1417 if ((wbc->nr_to_write -= n_iov) <= 0)
1421 /* Need to re-find the pages we skipped */
1422 index = pvec.pages[0]->index + 1;
1424 pagevec_release(&pvec);
1426 if (!scanned && !done) {
1428 * We hit the last page and there is more work to be done: wrap
1429 * back to the start of the file
1435 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1436 mapping->writeback_index = index;
1443 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1449 /* BB add check for wbc flags */
1450 page_cache_get(page);
1451 if (!PageUptodate(page))
1452 cFYI(1, "ppw - page not up to date");
1455 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1457 * A writepage() implementation always needs to do either this,
1458 * or re-dirty the page with "redirty_page_for_writepage()" in
1459 * the case of a failure.
1461 * Just unlocking the page will cause the radix tree tag-bits
1462 * to fail to update with the state of the page correctly.
1464 set_page_writeback(page);
1465 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1466 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1468 end_page_writeback(page);
1469 page_cache_release(page);
1474 static int cifs_write_end(struct file *file, struct address_space *mapping,
1475 loff_t pos, unsigned len, unsigned copied,
1476 struct page *page, void *fsdata)
1479 struct inode *inode = mapping->host;
1481 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1484 if (PageChecked(page)) {
1486 SetPageUptodate(page);
1487 ClearPageChecked(page);
1488 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1489 SetPageUptodate(page);
1491 if (!PageUptodate(page)) {
1493 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1497 /* this is probably better than directly calling
1498 partialpage_write since in this function the file handle is
1499 known which we might as well leverage */
1500 /* BB check if anything else missing out of ppw
1501 such as updating last write time */
1502 page_data = kmap(page);
1503 rc = cifs_write(file->private_data, page_data + offset,
1505 /* if (rc < 0) should we set writebehind rc? */
1512 set_page_dirty(page);
1516 spin_lock(&inode->i_lock);
1517 if (pos > inode->i_size)
1518 i_size_write(inode, pos);
1519 spin_unlock(&inode->i_lock);
1523 page_cache_release(page);
1528 int cifs_strict_fsync(struct file *file, int datasync)
1532 struct cifsTconInfo *tcon;
1533 struct cifsFileInfo *smbfile = file->private_data;
1534 struct inode *inode = file->f_path.dentry->d_inode;
1535 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
1539 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1540 file->f_path.dentry->d_name.name, datasync);
1542 if (!CIFS_I(inode)->clientCanCacheRead)
1543 cifs_invalidate_mapping(inode);
1545 tcon = tlink_tcon(smbfile->tlink);
1546 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1547 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1553 int cifs_fsync(struct file *file, int datasync)
1557 struct cifsTconInfo *tcon;
1558 struct cifsFileInfo *smbfile = file->private_data;
1559 struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1563 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1564 file->f_path.dentry->d_name.name, datasync);
1566 tcon = tlink_tcon(smbfile->tlink);
1567 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1568 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1575 * As file closes, flush all cached write data for this inode checking
1576 * for write behind errors.
1578 int cifs_flush(struct file *file, fl_owner_t id)
1580 struct inode *inode = file->f_path.dentry->d_inode;
1583 if (file->f_mode & FMODE_WRITE)
1584 rc = filemap_write_and_wait(inode->i_mapping);
1586 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1592 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
1597 for (i = 0; i < num_pages; i++) {
1598 pages[i] = alloc_page(__GFP_HIGHMEM);
1601 * save number of pages we have already allocated and
1602 * return with ENOMEM error
1613 for (i = 0; i < num_pages; i++)
1619 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
1624 clen = min_t(const size_t, len, wsize);
1625 num_pages = clen / PAGE_CACHE_SIZE;
1626 if (clen % PAGE_CACHE_SIZE)
1636 cifs_iovec_write(struct file *file, const struct iovec *iov,
1637 unsigned long nr_segs, loff_t *poffset)
1639 unsigned int written;
1640 unsigned long num_pages, npages, i;
1641 size_t copied, len, cur_len;
1642 ssize_t total_written = 0;
1643 struct kvec *to_send;
1644 struct page **pages;
1646 struct inode *inode;
1647 struct cifsFileInfo *open_file;
1648 struct cifsTconInfo *pTcon;
1649 struct cifs_sb_info *cifs_sb;
1652 len = iov_length(iov, nr_segs);
1656 rc = generic_write_checks(file, poffset, &len, 0);
1660 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1661 num_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
1663 pages = kmalloc(sizeof(struct pages *)*num_pages, GFP_KERNEL);
1667 to_send = kmalloc(sizeof(struct kvec)*(num_pages + 1), GFP_KERNEL);
1673 rc = cifs_write_allocate_pages(pages, num_pages);
1681 open_file = file->private_data;
1682 pTcon = tlink_tcon(open_file->tlink);
1683 inode = file->f_path.dentry->d_inode;
1685 iov_iter_init(&it, iov, nr_segs, len, 0);
1689 size_t save_len = cur_len;
1690 for (i = 0; i < npages; i++) {
1691 copied = min_t(const size_t, cur_len, PAGE_CACHE_SIZE);
1692 copied = iov_iter_copy_from_user(pages[i], &it, 0,
1695 iov_iter_advance(&it, copied);
1696 to_send[i+1].iov_base = kmap(pages[i]);
1697 to_send[i+1].iov_len = copied;
1700 cur_len = save_len - cur_len;
1703 if (open_file->invalidHandle) {
1704 rc = cifs_reopen_file(open_file, false);
1708 rc = CIFSSMBWrite2(xid, pTcon, open_file->netfid,
1709 cur_len, *poffset, &written,
1710 to_send, npages, 0);
1711 } while (rc == -EAGAIN);
1713 for (i = 0; i < npages; i++)
1718 total_written += written;
1719 cifs_update_eof(CIFS_I(inode), *poffset, written);
1720 *poffset += written;
1721 } else if (rc < 0) {
1727 /* get length and number of kvecs of the next write */
1728 npages = get_numpages(cifs_sb->wsize, len, &cur_len);
1731 if (total_written > 0) {
1732 spin_lock(&inode->i_lock);
1733 if (*poffset > inode->i_size)
1734 i_size_write(inode, *poffset);
1735 spin_unlock(&inode->i_lock);
1738 cifs_stats_bytes_written(pTcon, total_written);
1739 mark_inode_dirty_sync(inode);
1741 for (i = 0; i < num_pages; i++)
1746 return total_written;
1749 static ssize_t cifs_user_writev(struct kiocb *iocb, const struct iovec *iov,
1750 unsigned long nr_segs, loff_t pos)
1753 struct inode *inode;
1755 inode = iocb->ki_filp->f_path.dentry->d_inode;
1758 * BB - optimize the way when signing is disabled. We can drop this
1759 * extra memory-to-memory copying and use iovec buffers for constructing
1763 written = cifs_iovec_write(iocb->ki_filp, iov, nr_segs, &pos);
1765 CIFS_I(inode)->invalid_mapping = true;
1772 ssize_t cifs_strict_writev(struct kiocb *iocb, const struct iovec *iov,
1773 unsigned long nr_segs, loff_t pos)
1775 struct inode *inode;
1777 inode = iocb->ki_filp->f_path.dentry->d_inode;
1779 if (CIFS_I(inode)->clientCanCacheAll)
1780 return generic_file_aio_write(iocb, iov, nr_segs, pos);
1783 * In strict cache mode we need to write the data to the server exactly
1784 * from the pos to pos+len-1 rather than flush all affected pages
1785 * because it may cause a error with mandatory locks on these pages but
1786 * not on the region from pos to ppos+len-1.
1789 return cifs_user_writev(iocb, iov, nr_segs, pos);
1793 cifs_iovec_read(struct file *file, const struct iovec *iov,
1794 unsigned long nr_segs, loff_t *poffset)
1799 unsigned int bytes_read = 0;
1800 size_t len, cur_len;
1802 struct cifs_sb_info *cifs_sb;
1803 struct cifsTconInfo *pTcon;
1804 struct cifsFileInfo *open_file;
1805 struct smb_com_read_rsp *pSMBr;
1811 len = iov_length(iov, nr_segs);
1816 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1818 open_file = file->private_data;
1819 pTcon = tlink_tcon(open_file->tlink);
1821 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1822 cFYI(1, "attempting read on write only file instance");
1824 for (total_read = 0; total_read < len; total_read += bytes_read) {
1825 cur_len = min_t(const size_t, len - total_read, cifs_sb->rsize);
1829 while (rc == -EAGAIN) {
1830 int buf_type = CIFS_NO_BUFFER;
1831 if (open_file->invalidHandle) {
1832 rc = cifs_reopen_file(open_file, true);
1836 rc = CIFSSMBRead(xid, pTcon, open_file->netfid,
1837 cur_len, *poffset, &bytes_read,
1838 &read_data, &buf_type);
1839 pSMBr = (struct smb_com_read_rsp *)read_data;
1841 char *data_offset = read_data + 4 +
1842 le16_to_cpu(pSMBr->DataOffset);
1843 if (memcpy_toiovecend(iov, data_offset,
1844 iov_offset, bytes_read))
1846 if (buf_type == CIFS_SMALL_BUFFER)
1847 cifs_small_buf_release(read_data);
1848 else if (buf_type == CIFS_LARGE_BUFFER)
1849 cifs_buf_release(read_data);
1851 iov_offset += bytes_read;
1855 if (rc || (bytes_read == 0)) {
1863 cifs_stats_bytes_read(pTcon, bytes_read);
1864 *poffset += bytes_read;
1872 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1873 size_t read_size, loff_t *poffset)
1876 iov.iov_base = read_data;
1877 iov.iov_len = read_size;
1879 return cifs_iovec_read(file, &iov, 1, poffset);
1882 static ssize_t cifs_user_readv(struct kiocb *iocb, const struct iovec *iov,
1883 unsigned long nr_segs, loff_t pos)
1887 read = cifs_iovec_read(iocb->ki_filp, iov, nr_segs, &pos);
1894 ssize_t cifs_strict_readv(struct kiocb *iocb, const struct iovec *iov,
1895 unsigned long nr_segs, loff_t pos)
1897 struct inode *inode;
1899 inode = iocb->ki_filp->f_path.dentry->d_inode;
1901 if (CIFS_I(inode)->clientCanCacheRead)
1902 return generic_file_aio_read(iocb, iov, nr_segs, pos);
1905 * In strict cache mode we need to read from the server all the time
1906 * if we don't have level II oplock because the server can delay mtime
1907 * change - so we can't make a decision about inode invalidating.
1908 * And we can also fail with pagereading if there are mandatory locks
1909 * on pages affected by this read but not on the region from pos to
1913 return cifs_user_readv(iocb, iov, nr_segs, pos);
1916 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1920 unsigned int bytes_read = 0;
1921 unsigned int total_read;
1922 unsigned int current_read_size;
1923 struct cifs_sb_info *cifs_sb;
1924 struct cifsTconInfo *pTcon;
1926 char *current_offset;
1927 struct cifsFileInfo *open_file;
1928 int buf_type = CIFS_NO_BUFFER;
1931 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1933 if (file->private_data == NULL) {
1938 open_file = file->private_data;
1939 pTcon = tlink_tcon(open_file->tlink);
1941 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1942 cFYI(1, "attempting read on write only file instance");
1944 for (total_read = 0, current_offset = read_data;
1945 read_size > total_read;
1946 total_read += bytes_read, current_offset += bytes_read) {
1947 current_read_size = min_t(const int, read_size - total_read,
1949 /* For windows me and 9x we do not want to request more
1950 than it negotiated since it will refuse the read then */
1952 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1953 current_read_size = min_t(const int, current_read_size,
1954 pTcon->ses->server->maxBuf - 128);
1957 while (rc == -EAGAIN) {
1958 if (open_file->invalidHandle) {
1959 rc = cifs_reopen_file(open_file, true);
1963 rc = CIFSSMBRead(xid, pTcon,
1965 current_read_size, *poffset,
1966 &bytes_read, ¤t_offset,
1969 if (rc || (bytes_read == 0)) {
1977 cifs_stats_bytes_read(pTcon, total_read);
1978 *poffset += bytes_read;
1985 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
1988 struct inode *inode = file->f_path.dentry->d_inode;
1992 if (!CIFS_I(inode)->clientCanCacheRead)
1993 cifs_invalidate_mapping(inode);
1995 rc = generic_file_mmap(file, vma);
2000 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
2005 rc = cifs_revalidate_file(file);
2007 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
2011 rc = generic_file_mmap(file, vma);
2017 static void cifs_copy_cache_pages(struct address_space *mapping,
2018 struct list_head *pages, int bytes_read, char *data)
2023 while (bytes_read > 0) {
2024 if (list_empty(pages))
2027 page = list_entry(pages->prev, struct page, lru);
2028 list_del(&page->lru);
2030 if (add_to_page_cache_lru(page, mapping, page->index,
2032 page_cache_release(page);
2033 cFYI(1, "Add page cache failed");
2034 data += PAGE_CACHE_SIZE;
2035 bytes_read -= PAGE_CACHE_SIZE;
2038 page_cache_release(page);
2040 target = kmap_atomic(page, KM_USER0);
2042 if (PAGE_CACHE_SIZE > bytes_read) {
2043 memcpy(target, data, bytes_read);
2044 /* zero the tail end of this partial page */
2045 memset(target + bytes_read, 0,
2046 PAGE_CACHE_SIZE - bytes_read);
2049 memcpy(target, data, PAGE_CACHE_SIZE);
2050 bytes_read -= PAGE_CACHE_SIZE;
2052 kunmap_atomic(target, KM_USER0);
2054 flush_dcache_page(page);
2055 SetPageUptodate(page);
2057 data += PAGE_CACHE_SIZE;
2059 /* add page to FS-Cache */
2060 cifs_readpage_to_fscache(mapping->host, page);
2065 static int cifs_readpages(struct file *file, struct address_space *mapping,
2066 struct list_head *page_list, unsigned num_pages)
2072 struct cifs_sb_info *cifs_sb;
2073 struct cifsTconInfo *pTcon;
2074 unsigned int bytes_read = 0;
2075 unsigned int read_size, i;
2076 char *smb_read_data = NULL;
2077 struct smb_com_read_rsp *pSMBr;
2078 struct cifsFileInfo *open_file;
2079 int buf_type = CIFS_NO_BUFFER;
2082 if (file->private_data == NULL) {
2087 open_file = file->private_data;
2088 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2089 pTcon = tlink_tcon(open_file->tlink);
2092 * Reads as many pages as possible from fscache. Returns -ENOBUFS
2093 * immediately if the cookie is negative
2095 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
2100 cFYI(DBG2, "rpages: num pages %d", num_pages);
2101 for (i = 0; i < num_pages; ) {
2102 unsigned contig_pages;
2103 struct page *tmp_page;
2104 unsigned long expected_index;
2106 if (list_empty(page_list))
2109 page = list_entry(page_list->prev, struct page, lru);
2110 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2112 /* count adjacent pages that we will read into */
2115 list_entry(page_list->prev, struct page, lru)->index;
2116 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2117 if (tmp_page->index == expected_index) {
2123 if (contig_pages + i > num_pages)
2124 contig_pages = num_pages - i;
2126 /* for reads over a certain size could initiate async
2129 read_size = contig_pages * PAGE_CACHE_SIZE;
2130 /* Read size needs to be in multiples of one page */
2131 read_size = min_t(const unsigned int, read_size,
2132 cifs_sb->rsize & PAGE_CACHE_MASK);
2133 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
2134 read_size, contig_pages);
2136 while (rc == -EAGAIN) {
2137 if (open_file->invalidHandle) {
2138 rc = cifs_reopen_file(open_file, true);
2143 rc = CIFSSMBRead(xid, pTcon,
2146 &bytes_read, &smb_read_data,
2148 /* BB more RC checks ? */
2149 if (rc == -EAGAIN) {
2150 if (smb_read_data) {
2151 if (buf_type == CIFS_SMALL_BUFFER)
2152 cifs_small_buf_release(smb_read_data);
2153 else if (buf_type == CIFS_LARGE_BUFFER)
2154 cifs_buf_release(smb_read_data);
2155 smb_read_data = NULL;
2159 if ((rc < 0) || (smb_read_data == NULL)) {
2160 cFYI(1, "Read error in readpages: %d", rc);
2162 } else if (bytes_read > 0) {
2163 task_io_account_read(bytes_read);
2164 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2165 cifs_copy_cache_pages(mapping, page_list, bytes_read,
2166 smb_read_data + 4 /* RFC1001 hdr */ +
2167 le16_to_cpu(pSMBr->DataOffset));
2169 i += bytes_read >> PAGE_CACHE_SHIFT;
2170 cifs_stats_bytes_read(pTcon, bytes_read);
2171 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2172 i++; /* account for partial page */
2174 /* server copy of file can have smaller size
2176 /* BB do we need to verify this common case ?
2177 this case is ok - if we are at server EOF
2178 we will hit it on next read */
2183 cFYI(1, "No bytes read (%d) at offset %lld . "
2184 "Cleaning remaining pages from readahead list",
2185 bytes_read, offset);
2186 /* BB turn off caching and do new lookup on
2187 file size at server? */
2190 if (smb_read_data) {
2191 if (buf_type == CIFS_SMALL_BUFFER)
2192 cifs_small_buf_release(smb_read_data);
2193 else if (buf_type == CIFS_LARGE_BUFFER)
2194 cifs_buf_release(smb_read_data);
2195 smb_read_data = NULL;
2200 /* need to free smb_read_data buf before exit */
2201 if (smb_read_data) {
2202 if (buf_type == CIFS_SMALL_BUFFER)
2203 cifs_small_buf_release(smb_read_data);
2204 else if (buf_type == CIFS_LARGE_BUFFER)
2205 cifs_buf_release(smb_read_data);
2206 smb_read_data = NULL;
2214 static int cifs_readpage_worker(struct file *file, struct page *page,
2220 /* Is the page cached? */
2221 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2225 page_cache_get(page);
2226 read_data = kmap(page);
2227 /* for reads over a certain size could initiate async read ahead */
2229 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2234 cFYI(1, "Bytes read %d", rc);
2236 file->f_path.dentry->d_inode->i_atime =
2237 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2239 if (PAGE_CACHE_SIZE > rc)
2240 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2242 flush_dcache_page(page);
2243 SetPageUptodate(page);
2245 /* send this page to the cache */
2246 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2252 page_cache_release(page);
2258 static int cifs_readpage(struct file *file, struct page *page)
2260 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2266 if (file->private_data == NULL) {
2272 cFYI(1, "readpage %p at offset %d 0x%x\n",
2273 page, (int)offset, (int)offset);
2275 rc = cifs_readpage_worker(file, page, &offset);
2283 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2285 struct cifsFileInfo *open_file;
2287 spin_lock(&cifs_file_list_lock);
2288 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2289 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2290 spin_unlock(&cifs_file_list_lock);
2294 spin_unlock(&cifs_file_list_lock);
2298 /* We do not want to update the file size from server for inodes
2299 open for write - to avoid races with writepage extending
2300 the file - in the future we could consider allowing
2301 refreshing the inode only on increases in the file size
2302 but this is tricky to do without racing with writebehind
2303 page caching in the current Linux kernel design */
2304 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2309 if (is_inode_writable(cifsInode)) {
2310 /* This inode is open for write at least once */
2311 struct cifs_sb_info *cifs_sb;
2313 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2314 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2315 /* since no page cache to corrupt on directio
2316 we can change size safely */
2320 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2328 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2329 loff_t pos, unsigned len, unsigned flags,
2330 struct page **pagep, void **fsdata)
2332 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2333 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2334 loff_t page_start = pos & PAGE_MASK;
2339 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2341 page = grab_cache_page_write_begin(mapping, index, flags);
2347 if (PageUptodate(page))
2351 * If we write a full page it will be up to date, no need to read from
2352 * the server. If the write is short, we'll end up doing a sync write
2355 if (len == PAGE_CACHE_SIZE)
2359 * optimize away the read when we have an oplock, and we're not
2360 * expecting to use any of the data we'd be reading in. That
2361 * is, when the page lies beyond the EOF, or straddles the EOF
2362 * and the write will cover all of the existing data.
2364 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2365 i_size = i_size_read(mapping->host);
2366 if (page_start >= i_size ||
2367 (offset == 0 && (pos + len) >= i_size)) {
2368 zero_user_segments(page, 0, offset,
2372 * PageChecked means that the parts of the page
2373 * to which we're not writing are considered up
2374 * to date. Once the data is copied to the
2375 * page, it can be set uptodate.
2377 SetPageChecked(page);
2382 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2384 * might as well read a page, it is fast enough. If we get
2385 * an error, we don't need to return it. cifs_write_end will
2386 * do a sync write instead since PG_uptodate isn't set.
2388 cifs_readpage_worker(file, page, &page_start);
2390 /* we could try using another file handle if there is one -
2391 but how would we lock it to prevent close of that handle
2392 racing with this read? In any case
2393 this will be written out by write_end so is fine */
2400 static int cifs_release_page(struct page *page, gfp_t gfp)
2402 if (PagePrivate(page))
2405 return cifs_fscache_release_page(page, gfp);
2408 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2410 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2413 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2416 void cifs_oplock_break(struct work_struct *work)
2418 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2420 struct inode *inode = cfile->dentry->d_inode;
2421 struct cifsInodeInfo *cinode = CIFS_I(inode);
2424 if (inode && S_ISREG(inode->i_mode)) {
2425 if (cinode->clientCanCacheRead)
2426 break_lease(inode, O_RDONLY);
2428 break_lease(inode, O_WRONLY);
2429 rc = filemap_fdatawrite(inode->i_mapping);
2430 if (cinode->clientCanCacheRead == 0) {
2431 rc = filemap_fdatawait(inode->i_mapping);
2432 mapping_set_error(inode->i_mapping, rc);
2433 invalidate_remote_inode(inode);
2435 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2439 * releasing stale oplock after recent reconnect of smb session using
2440 * a now incorrect file handle is not a data integrity issue but do
2441 * not bother sending an oplock release if session to server still is
2442 * disconnected since oplock already released by the server
2444 if (!cfile->oplock_break_cancelled) {
2445 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2446 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false,
2447 cinode->clientCanCacheRead ? 1 : 0);
2448 cFYI(1, "Oplock release rc = %d", rc);
2452 * We might have kicked in before is_valid_oplock_break()
2453 * finished grabbing reference for us. Make sure it's done by
2454 * waiting for cifs_file_list_lock.
2456 spin_lock(&cifs_file_list_lock);
2457 spin_unlock(&cifs_file_list_lock);
2459 cifs_oplock_break_put(cfile);
2462 /* must be called while holding cifs_file_list_lock */
2463 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2465 cifs_sb_active(cfile->dentry->d_sb);
2466 cifsFileInfo_get(cfile);
2469 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2471 struct super_block *sb = cfile->dentry->d_sb;
2473 cifsFileInfo_put(cfile);
2474 cifs_sb_deactive(sb);
2477 const struct address_space_operations cifs_addr_ops = {
2478 .readpage = cifs_readpage,
2479 .readpages = cifs_readpages,
2480 .writepage = cifs_writepage,
2481 .writepages = cifs_writepages,
2482 .write_begin = cifs_write_begin,
2483 .write_end = cifs_write_end,
2484 .set_page_dirty = __set_page_dirty_nobuffers,
2485 .releasepage = cifs_release_page,
2486 .invalidatepage = cifs_invalidate_page,
2491 * cifs_readpages requires the server to support a buffer large enough to
2492 * contain the header plus one complete page of data. Otherwise, we need
2493 * to leave cifs_readpages out of the address space operations.
2495 const struct address_space_operations cifs_addr_ops_smallbuf = {
2496 .readpage = cifs_readpage,
2497 .writepage = cifs_writepage,
2498 .writepages = cifs_writepages,
2499 .write_begin = cifs_write_begin,
2500 .write_end = cifs_write_end,
2501 .set_page_dirty = __set_page_dirty_nobuffers,
2502 .releasepage = cifs_release_page,
2503 .invalidatepage = cifs_invalidate_page,