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 static ssize_t cifs_write(struct cifsFileInfo *open_file,
861 const char *write_data, size_t write_size,
865 unsigned int bytes_written = 0;
866 unsigned int total_written;
867 struct cifs_sb_info *cifs_sb;
868 struct cifsTconInfo *pTcon;
870 struct dentry *dentry = open_file->dentry;
871 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
873 cifs_sb = CIFS_SB(dentry->d_sb);
875 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
876 *poffset, dentry->d_name.name);
878 pTcon = tlink_tcon(open_file->tlink);
882 for (total_written = 0; write_size > total_written;
883 total_written += bytes_written) {
885 while (rc == -EAGAIN) {
889 if (open_file->invalidHandle) {
890 /* we could deadlock if we called
891 filemap_fdatawait from here so tell
892 reopen_file not to flush data to
894 rc = cifs_reopen_file(open_file, false);
899 len = min((size_t)cifs_sb->wsize,
900 write_size - total_written);
901 /* iov[0] is reserved for smb header */
902 iov[1].iov_base = (char *)write_data + total_written;
903 iov[1].iov_len = len;
904 rc = CIFSSMBWrite2(xid, pTcon, open_file->netfid, len,
905 *poffset, &bytes_written, iov, 1, 0);
907 if (rc || (bytes_written == 0)) {
915 cifs_update_eof(cifsi, *poffset, bytes_written);
916 *poffset += bytes_written;
920 cifs_stats_bytes_written(pTcon, total_written);
922 if (total_written > 0) {
923 spin_lock(&dentry->d_inode->i_lock);
924 if (*poffset > dentry->d_inode->i_size)
925 i_size_write(dentry->d_inode, *poffset);
926 spin_unlock(&dentry->d_inode->i_lock);
928 mark_inode_dirty_sync(dentry->d_inode);
930 return total_written;
933 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
936 struct cifsFileInfo *open_file = NULL;
937 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
939 /* only filter by fsuid on multiuser mounts */
940 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
943 spin_lock(&cifs_file_list_lock);
944 /* we could simply get the first_list_entry since write-only entries
945 are always at the end of the list but since the first entry might
946 have a close pending, we go through the whole list */
947 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
948 if (fsuid_only && open_file->uid != current_fsuid())
950 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
951 if (!open_file->invalidHandle) {
952 /* found a good file */
953 /* lock it so it will not be closed on us */
954 cifsFileInfo_get(open_file);
955 spin_unlock(&cifs_file_list_lock);
957 } /* else might as well continue, and look for
958 another, or simply have the caller reopen it
959 again rather than trying to fix this handle */
960 } else /* write only file */
961 break; /* write only files are last so must be done */
963 spin_unlock(&cifs_file_list_lock);
967 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
970 struct cifsFileInfo *open_file;
971 struct cifs_sb_info *cifs_sb;
972 bool any_available = false;
975 /* Having a null inode here (because mapping->host was set to zero by
976 the VFS or MM) should not happen but we had reports of on oops (due to
977 it being zero) during stress testcases so we need to check for it */
979 if (cifs_inode == NULL) {
980 cERROR(1, "Null inode passed to cifs_writeable_file");
985 cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
987 /* only filter by fsuid on multiuser mounts */
988 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
991 spin_lock(&cifs_file_list_lock);
993 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
994 if (!any_available && open_file->pid != current->tgid)
996 if (fsuid_only && open_file->uid != current_fsuid())
998 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
999 cifsFileInfo_get(open_file);
1001 if (!open_file->invalidHandle) {
1002 /* found a good writable file */
1003 spin_unlock(&cifs_file_list_lock);
1007 spin_unlock(&cifs_file_list_lock);
1009 /* Had to unlock since following call can block */
1010 rc = cifs_reopen_file(open_file, false);
1014 /* if it fails, try another handle if possible */
1015 cFYI(1, "wp failed on reopen file");
1016 cifsFileInfo_put(open_file);
1018 spin_lock(&cifs_file_list_lock);
1020 /* else we simply continue to the next entry. Thus
1021 we do not loop on reopen errors. If we
1022 can not reopen the file, for example if we
1023 reconnected to a server with another client
1024 racing to delete or lock the file we would not
1025 make progress if we restarted before the beginning
1026 of the loop here. */
1029 /* couldn't find useable FH with same pid, try any available */
1030 if (!any_available) {
1031 any_available = true;
1032 goto refind_writable;
1034 spin_unlock(&cifs_file_list_lock);
1038 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1040 struct address_space *mapping = page->mapping;
1041 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1044 int bytes_written = 0;
1045 struct inode *inode;
1046 struct cifsFileInfo *open_file;
1048 if (!mapping || !mapping->host)
1051 inode = page->mapping->host;
1053 offset += (loff_t)from;
1054 write_data = kmap(page);
1057 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1062 /* racing with truncate? */
1063 if (offset > mapping->host->i_size) {
1065 return 0; /* don't care */
1068 /* check to make sure that we are not extending the file */
1069 if (mapping->host->i_size - offset < (loff_t)to)
1070 to = (unsigned)(mapping->host->i_size - offset);
1072 open_file = find_writable_file(CIFS_I(mapping->host), false);
1074 bytes_written = cifs_write(open_file, write_data,
1075 to - from, &offset);
1076 cifsFileInfo_put(open_file);
1077 /* Does mm or vfs already set times? */
1078 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1079 if ((bytes_written > 0) && (offset))
1081 else if (bytes_written < 0)
1084 cFYI(1, "No writeable filehandles for inode");
1092 static int cifs_writepages(struct address_space *mapping,
1093 struct writeback_control *wbc)
1095 struct cifs_sb_info *cifs_sb = CIFS_SB(mapping->host->i_sb);
1096 bool done = false, scanned = false, range_whole = false;
1098 struct cifs_writedata *wdata;
1103 * If wsize is smaller than the page cache size, default to writing
1104 * one page at a time via cifs_writepage
1106 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1107 return generic_writepages(mapping, wbc);
1109 if (wbc->range_cyclic) {
1110 index = mapping->writeback_index; /* Start from prev offset */
1113 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1114 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1115 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1120 while (!done && index <= end) {
1121 unsigned int i, nr_pages, found_pages;
1122 pgoff_t next = 0, tofind;
1123 struct page **pages;
1125 tofind = min((cifs_sb->wsize / PAGE_CACHE_SIZE) - 1,
1128 wdata = cifs_writedata_alloc((unsigned int)tofind);
1135 * find_get_pages_tag seems to return a max of 256 on each
1136 * iteration, so we must call it several times in order to
1137 * fill the array or the wsize is effectively limited to
1138 * 256 * PAGE_CACHE_SIZE.
1141 pages = wdata->pages;
1143 nr_pages = find_get_pages_tag(mapping, &index,
1144 PAGECACHE_TAG_DIRTY,
1146 found_pages += nr_pages;
1149 } while (nr_pages && tofind && index <= end);
1151 if (found_pages == 0) {
1152 kref_put(&wdata->refcount, cifs_writedata_release);
1157 for (i = 0; i < found_pages; i++) {
1158 page = wdata->pages[i];
1160 * At this point we hold neither mapping->tree_lock nor
1161 * lock on the page itself: the page may be truncated or
1162 * invalidated (changing page->mapping to NULL), or even
1163 * swizzled back from swapper_space to tmpfs file
1169 else if (!trylock_page(page))
1172 if (unlikely(page->mapping != mapping)) {
1177 if (!wbc->range_cyclic && page->index > end) {
1183 if (next && (page->index != next)) {
1184 /* Not next consecutive page */
1189 if (wbc->sync_mode != WB_SYNC_NONE)
1190 wait_on_page_writeback(page);
1192 if (PageWriteback(page) ||
1193 !clear_page_dirty_for_io(page)) {
1199 * This actually clears the dirty bit in the radix tree.
1200 * See cifs_writepage() for more commentary.
1202 set_page_writeback(page);
1204 if (page_offset(page) >= mapping->host->i_size) {
1207 end_page_writeback(page);
1211 wdata->pages[i] = page;
1212 next = page->index + 1;
1216 /* reset index to refind any pages skipped */
1218 index = wdata->pages[0]->index + 1;
1220 /* put any pages we aren't going to use */
1221 for (i = nr_pages; i < found_pages; i++) {
1222 page_cache_release(wdata->pages[i]);
1223 wdata->pages[i] = NULL;
1226 /* nothing to write? */
1227 if (nr_pages == 0) {
1228 kref_put(&wdata->refcount, cifs_writedata_release);
1232 wdata->sync_mode = wbc->sync_mode;
1233 wdata->nr_pages = nr_pages;
1234 wdata->offset = page_offset(wdata->pages[0]);
1237 if (wdata->cfile != NULL)
1238 cifsFileInfo_put(wdata->cfile);
1239 wdata->cfile = find_writable_file(CIFS_I(mapping->host),
1241 if (!wdata->cfile) {
1242 cERROR(1, "No writable handles for inode");
1246 rc = cifs_async_writev(wdata);
1247 } while (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN);
1249 for (i = 0; i < nr_pages; ++i)
1250 unlock_page(wdata->pages[i]);
1252 /* send failure -- clean up the mess */
1254 for (i = 0; i < nr_pages; ++i) {
1256 redirty_page_for_writepage(wbc,
1259 SetPageError(wdata->pages[i]);
1260 end_page_writeback(wdata->pages[i]);
1261 page_cache_release(wdata->pages[i]);
1264 mapping_set_error(mapping, rc);
1266 kref_put(&wdata->refcount, cifs_writedata_release);
1268 wbc->nr_to_write -= nr_pages;
1269 if (wbc->nr_to_write <= 0)
1275 if (!scanned && !done) {
1277 * We hit the last page and there is more work to be done: wrap
1278 * back to the start of the file
1285 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1286 mapping->writeback_index = index;
1292 cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
1298 /* BB add check for wbc flags */
1299 page_cache_get(page);
1300 if (!PageUptodate(page))
1301 cFYI(1, "ppw - page not up to date");
1304 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1306 * A writepage() implementation always needs to do either this,
1307 * or re-dirty the page with "redirty_page_for_writepage()" in
1308 * the case of a failure.
1310 * Just unlocking the page will cause the radix tree tag-bits
1311 * to fail to update with the state of the page correctly.
1313 set_page_writeback(page);
1315 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1316 if (rc == -EAGAIN && wbc->sync_mode == WB_SYNC_ALL)
1318 else if (rc == -EAGAIN)
1319 redirty_page_for_writepage(wbc, page);
1323 SetPageUptodate(page);
1324 end_page_writeback(page);
1325 page_cache_release(page);
1330 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1332 int rc = cifs_writepage_locked(page, wbc);
1337 static int cifs_write_end(struct file *file, struct address_space *mapping,
1338 loff_t pos, unsigned len, unsigned copied,
1339 struct page *page, void *fsdata)
1342 struct inode *inode = mapping->host;
1344 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1347 if (PageChecked(page)) {
1349 SetPageUptodate(page);
1350 ClearPageChecked(page);
1351 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1352 SetPageUptodate(page);
1354 if (!PageUptodate(page)) {
1356 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1360 /* this is probably better than directly calling
1361 partialpage_write since in this function the file handle is
1362 known which we might as well leverage */
1363 /* BB check if anything else missing out of ppw
1364 such as updating last write time */
1365 page_data = kmap(page);
1366 rc = cifs_write(file->private_data, page_data + offset,
1368 /* if (rc < 0) should we set writebehind rc? */
1375 set_page_dirty(page);
1379 spin_lock(&inode->i_lock);
1380 if (pos > inode->i_size)
1381 i_size_write(inode, pos);
1382 spin_unlock(&inode->i_lock);
1386 page_cache_release(page);
1391 int cifs_strict_fsync(struct file *file, int datasync)
1395 struct cifsTconInfo *tcon;
1396 struct cifsFileInfo *smbfile = file->private_data;
1397 struct inode *inode = file->f_path.dentry->d_inode;
1398 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
1402 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1403 file->f_path.dentry->d_name.name, datasync);
1405 if (!CIFS_I(inode)->clientCanCacheRead) {
1406 rc = cifs_invalidate_mapping(inode);
1408 cFYI(1, "rc: %d during invalidate phase", rc);
1409 rc = 0; /* don't care about it in fsync */
1413 tcon = tlink_tcon(smbfile->tlink);
1414 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1415 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1421 int cifs_fsync(struct file *file, int datasync)
1425 struct cifsTconInfo *tcon;
1426 struct cifsFileInfo *smbfile = file->private_data;
1427 struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1431 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1432 file->f_path.dentry->d_name.name, datasync);
1434 tcon = tlink_tcon(smbfile->tlink);
1435 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1436 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1443 * As file closes, flush all cached write data for this inode checking
1444 * for write behind errors.
1446 int cifs_flush(struct file *file, fl_owner_t id)
1448 struct inode *inode = file->f_path.dentry->d_inode;
1451 if (file->f_mode & FMODE_WRITE)
1452 rc = filemap_write_and_wait(inode->i_mapping);
1454 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1460 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
1465 for (i = 0; i < num_pages; i++) {
1466 pages[i] = alloc_page(__GFP_HIGHMEM);
1469 * save number of pages we have already allocated and
1470 * return with ENOMEM error
1481 for (i = 0; i < num_pages; i++)
1487 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
1492 clen = min_t(const size_t, len, wsize);
1493 num_pages = clen / PAGE_CACHE_SIZE;
1494 if (clen % PAGE_CACHE_SIZE)
1504 cifs_iovec_write(struct file *file, const struct iovec *iov,
1505 unsigned long nr_segs, loff_t *poffset)
1507 unsigned int written;
1508 unsigned long num_pages, npages, i;
1509 size_t copied, len, cur_len;
1510 ssize_t total_written = 0;
1511 struct kvec *to_send;
1512 struct page **pages;
1514 struct inode *inode;
1515 struct cifsFileInfo *open_file;
1516 struct cifsTconInfo *pTcon;
1517 struct cifs_sb_info *cifs_sb;
1520 len = iov_length(iov, nr_segs);
1524 rc = generic_write_checks(file, poffset, &len, 0);
1528 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1529 num_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
1531 pages = kmalloc(sizeof(struct pages *)*num_pages, GFP_KERNEL);
1535 to_send = kmalloc(sizeof(struct kvec)*(num_pages + 1), GFP_KERNEL);
1541 rc = cifs_write_allocate_pages(pages, num_pages);
1549 open_file = file->private_data;
1550 pTcon = tlink_tcon(open_file->tlink);
1551 inode = file->f_path.dentry->d_inode;
1553 iov_iter_init(&it, iov, nr_segs, len, 0);
1557 size_t save_len = cur_len;
1558 for (i = 0; i < npages; i++) {
1559 copied = min_t(const size_t, cur_len, PAGE_CACHE_SIZE);
1560 copied = iov_iter_copy_from_user(pages[i], &it, 0,
1563 iov_iter_advance(&it, copied);
1564 to_send[i+1].iov_base = kmap(pages[i]);
1565 to_send[i+1].iov_len = copied;
1568 cur_len = save_len - cur_len;
1571 if (open_file->invalidHandle) {
1572 rc = cifs_reopen_file(open_file, false);
1576 rc = CIFSSMBWrite2(xid, pTcon, open_file->netfid,
1577 cur_len, *poffset, &written,
1578 to_send, npages, 0);
1579 } while (rc == -EAGAIN);
1581 for (i = 0; i < npages; i++)
1586 total_written += written;
1587 cifs_update_eof(CIFS_I(inode), *poffset, written);
1588 *poffset += written;
1589 } else if (rc < 0) {
1595 /* get length and number of kvecs of the next write */
1596 npages = get_numpages(cifs_sb->wsize, len, &cur_len);
1599 if (total_written > 0) {
1600 spin_lock(&inode->i_lock);
1601 if (*poffset > inode->i_size)
1602 i_size_write(inode, *poffset);
1603 spin_unlock(&inode->i_lock);
1606 cifs_stats_bytes_written(pTcon, total_written);
1607 mark_inode_dirty_sync(inode);
1609 for (i = 0; i < num_pages; i++)
1614 return total_written;
1617 ssize_t cifs_user_writev(struct kiocb *iocb, const struct iovec *iov,
1618 unsigned long nr_segs, loff_t pos)
1621 struct inode *inode;
1623 inode = iocb->ki_filp->f_path.dentry->d_inode;
1626 * BB - optimize the way when signing is disabled. We can drop this
1627 * extra memory-to-memory copying and use iovec buffers for constructing
1631 written = cifs_iovec_write(iocb->ki_filp, iov, nr_segs, &pos);
1633 CIFS_I(inode)->invalid_mapping = true;
1640 ssize_t cifs_strict_writev(struct kiocb *iocb, const struct iovec *iov,
1641 unsigned long nr_segs, loff_t pos)
1643 struct inode *inode;
1645 inode = iocb->ki_filp->f_path.dentry->d_inode;
1647 if (CIFS_I(inode)->clientCanCacheAll)
1648 return generic_file_aio_write(iocb, iov, nr_segs, pos);
1651 * In strict cache mode we need to write the data to the server exactly
1652 * from the pos to pos+len-1 rather than flush all affected pages
1653 * because it may cause a error with mandatory locks on these pages but
1654 * not on the region from pos to ppos+len-1.
1657 return cifs_user_writev(iocb, iov, nr_segs, pos);
1661 cifs_iovec_read(struct file *file, const struct iovec *iov,
1662 unsigned long nr_segs, loff_t *poffset)
1667 unsigned int bytes_read = 0;
1668 size_t len, cur_len;
1670 struct cifs_sb_info *cifs_sb;
1671 struct cifsTconInfo *pTcon;
1672 struct cifsFileInfo *open_file;
1673 struct smb_com_read_rsp *pSMBr;
1679 len = iov_length(iov, nr_segs);
1684 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1686 open_file = file->private_data;
1687 pTcon = tlink_tcon(open_file->tlink);
1689 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1690 cFYI(1, "attempting read on write only file instance");
1692 for (total_read = 0; total_read < len; total_read += bytes_read) {
1693 cur_len = min_t(const size_t, len - total_read, cifs_sb->rsize);
1697 while (rc == -EAGAIN) {
1698 int buf_type = CIFS_NO_BUFFER;
1699 if (open_file->invalidHandle) {
1700 rc = cifs_reopen_file(open_file, true);
1704 rc = CIFSSMBRead(xid, pTcon, open_file->netfid,
1705 cur_len, *poffset, &bytes_read,
1706 &read_data, &buf_type);
1707 pSMBr = (struct smb_com_read_rsp *)read_data;
1709 char *data_offset = read_data + 4 +
1710 le16_to_cpu(pSMBr->DataOffset);
1711 if (memcpy_toiovecend(iov, data_offset,
1712 iov_offset, bytes_read))
1714 if (buf_type == CIFS_SMALL_BUFFER)
1715 cifs_small_buf_release(read_data);
1716 else if (buf_type == CIFS_LARGE_BUFFER)
1717 cifs_buf_release(read_data);
1719 iov_offset += bytes_read;
1723 if (rc || (bytes_read == 0)) {
1731 cifs_stats_bytes_read(pTcon, bytes_read);
1732 *poffset += bytes_read;
1740 ssize_t cifs_user_readv(struct kiocb *iocb, const struct iovec *iov,
1741 unsigned long nr_segs, loff_t pos)
1745 read = cifs_iovec_read(iocb->ki_filp, iov, nr_segs, &pos);
1752 ssize_t cifs_strict_readv(struct kiocb *iocb, const struct iovec *iov,
1753 unsigned long nr_segs, loff_t pos)
1755 struct inode *inode;
1757 inode = iocb->ki_filp->f_path.dentry->d_inode;
1759 if (CIFS_I(inode)->clientCanCacheRead)
1760 return generic_file_aio_read(iocb, iov, nr_segs, pos);
1763 * In strict cache mode we need to read from the server all the time
1764 * if we don't have level II oplock because the server can delay mtime
1765 * change - so we can't make a decision about inode invalidating.
1766 * And we can also fail with pagereading if there are mandatory locks
1767 * on pages affected by this read but not on the region from pos to
1771 return cifs_user_readv(iocb, iov, nr_segs, pos);
1774 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1778 unsigned int bytes_read = 0;
1779 unsigned int total_read;
1780 unsigned int current_read_size;
1781 struct cifs_sb_info *cifs_sb;
1782 struct cifsTconInfo *pTcon;
1784 char *current_offset;
1785 struct cifsFileInfo *open_file;
1786 int buf_type = CIFS_NO_BUFFER;
1789 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1791 if (file->private_data == NULL) {
1796 open_file = file->private_data;
1797 pTcon = tlink_tcon(open_file->tlink);
1799 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1800 cFYI(1, "attempting read on write only file instance");
1802 for (total_read = 0, current_offset = read_data;
1803 read_size > total_read;
1804 total_read += bytes_read, current_offset += bytes_read) {
1805 current_read_size = min_t(const int, read_size - total_read,
1807 /* For windows me and 9x we do not want to request more
1808 than it negotiated since it will refuse the read then */
1810 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1811 current_read_size = min_t(const int, current_read_size,
1812 pTcon->ses->server->maxBuf - 128);
1815 while (rc == -EAGAIN) {
1816 if (open_file->invalidHandle) {
1817 rc = cifs_reopen_file(open_file, true);
1821 rc = CIFSSMBRead(xid, pTcon,
1823 current_read_size, *poffset,
1824 &bytes_read, ¤t_offset,
1827 if (rc || (bytes_read == 0)) {
1835 cifs_stats_bytes_read(pTcon, total_read);
1836 *poffset += bytes_read;
1844 * If the page is mmap'ed into a process' page tables, then we need to make
1845 * sure that it doesn't change while being written back.
1848 cifs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1850 struct page *page = vmf->page;
1853 return VM_FAULT_LOCKED;
1856 static struct vm_operations_struct cifs_file_vm_ops = {
1857 .fault = filemap_fault,
1858 .page_mkwrite = cifs_page_mkwrite,
1861 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
1864 struct inode *inode = file->f_path.dentry->d_inode;
1868 if (!CIFS_I(inode)->clientCanCacheRead) {
1869 rc = cifs_invalidate_mapping(inode);
1874 rc = generic_file_mmap(file, vma);
1876 vma->vm_ops = &cifs_file_vm_ops;
1881 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1886 rc = cifs_revalidate_file(file);
1888 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1892 rc = generic_file_mmap(file, vma);
1894 vma->vm_ops = &cifs_file_vm_ops;
1900 static void cifs_copy_cache_pages(struct address_space *mapping,
1901 struct list_head *pages, int bytes_read, char *data)
1906 while (bytes_read > 0) {
1907 if (list_empty(pages))
1910 page = list_entry(pages->prev, struct page, lru);
1911 list_del(&page->lru);
1913 if (add_to_page_cache_lru(page, mapping, page->index,
1915 page_cache_release(page);
1916 cFYI(1, "Add page cache failed");
1917 data += PAGE_CACHE_SIZE;
1918 bytes_read -= PAGE_CACHE_SIZE;
1921 page_cache_release(page);
1923 target = kmap_atomic(page, KM_USER0);
1925 if (PAGE_CACHE_SIZE > bytes_read) {
1926 memcpy(target, data, bytes_read);
1927 /* zero the tail end of this partial page */
1928 memset(target + bytes_read, 0,
1929 PAGE_CACHE_SIZE - bytes_read);
1932 memcpy(target, data, PAGE_CACHE_SIZE);
1933 bytes_read -= PAGE_CACHE_SIZE;
1935 kunmap_atomic(target, KM_USER0);
1937 flush_dcache_page(page);
1938 SetPageUptodate(page);
1940 data += PAGE_CACHE_SIZE;
1942 /* add page to FS-Cache */
1943 cifs_readpage_to_fscache(mapping->host, page);
1948 static int cifs_readpages(struct file *file, struct address_space *mapping,
1949 struct list_head *page_list, unsigned num_pages)
1955 struct cifs_sb_info *cifs_sb;
1956 struct cifsTconInfo *pTcon;
1957 unsigned int bytes_read = 0;
1958 unsigned int read_size, i;
1959 char *smb_read_data = NULL;
1960 struct smb_com_read_rsp *pSMBr;
1961 struct cifsFileInfo *open_file;
1962 int buf_type = CIFS_NO_BUFFER;
1965 if (file->private_data == NULL) {
1970 open_file = file->private_data;
1971 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1972 pTcon = tlink_tcon(open_file->tlink);
1975 * Reads as many pages as possible from fscache. Returns -ENOBUFS
1976 * immediately if the cookie is negative
1978 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
1983 cFYI(DBG2, "rpages: num pages %d", num_pages);
1984 for (i = 0; i < num_pages; ) {
1985 unsigned contig_pages;
1986 struct page *tmp_page;
1987 unsigned long expected_index;
1989 if (list_empty(page_list))
1992 page = list_entry(page_list->prev, struct page, lru);
1993 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1995 /* count adjacent pages that we will read into */
1998 list_entry(page_list->prev, struct page, lru)->index;
1999 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2000 if (tmp_page->index == expected_index) {
2006 if (contig_pages + i > num_pages)
2007 contig_pages = num_pages - i;
2009 /* for reads over a certain size could initiate async
2012 read_size = contig_pages * PAGE_CACHE_SIZE;
2013 /* Read size needs to be in multiples of one page */
2014 read_size = min_t(const unsigned int, read_size,
2015 cifs_sb->rsize & PAGE_CACHE_MASK);
2016 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
2017 read_size, contig_pages);
2019 while (rc == -EAGAIN) {
2020 if (open_file->invalidHandle) {
2021 rc = cifs_reopen_file(open_file, true);
2026 rc = CIFSSMBRead(xid, pTcon,
2029 &bytes_read, &smb_read_data,
2031 /* BB more RC checks ? */
2032 if (rc == -EAGAIN) {
2033 if (smb_read_data) {
2034 if (buf_type == CIFS_SMALL_BUFFER)
2035 cifs_small_buf_release(smb_read_data);
2036 else if (buf_type == CIFS_LARGE_BUFFER)
2037 cifs_buf_release(smb_read_data);
2038 smb_read_data = NULL;
2042 if ((rc < 0) || (smb_read_data == NULL)) {
2043 cFYI(1, "Read error in readpages: %d", rc);
2045 } else if (bytes_read > 0) {
2046 task_io_account_read(bytes_read);
2047 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2048 cifs_copy_cache_pages(mapping, page_list, bytes_read,
2049 smb_read_data + 4 /* RFC1001 hdr */ +
2050 le16_to_cpu(pSMBr->DataOffset));
2052 i += bytes_read >> PAGE_CACHE_SHIFT;
2053 cifs_stats_bytes_read(pTcon, bytes_read);
2054 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2055 i++; /* account for partial page */
2057 /* server copy of file can have smaller size
2059 /* BB do we need to verify this common case ?
2060 this case is ok - if we are at server EOF
2061 we will hit it on next read */
2066 cFYI(1, "No bytes read (%d) at offset %lld . "
2067 "Cleaning remaining pages from readahead list",
2068 bytes_read, offset);
2069 /* BB turn off caching and do new lookup on
2070 file size at server? */
2073 if (smb_read_data) {
2074 if (buf_type == CIFS_SMALL_BUFFER)
2075 cifs_small_buf_release(smb_read_data);
2076 else if (buf_type == CIFS_LARGE_BUFFER)
2077 cifs_buf_release(smb_read_data);
2078 smb_read_data = NULL;
2083 /* need to free smb_read_data buf before exit */
2084 if (smb_read_data) {
2085 if (buf_type == CIFS_SMALL_BUFFER)
2086 cifs_small_buf_release(smb_read_data);
2087 else if (buf_type == CIFS_LARGE_BUFFER)
2088 cifs_buf_release(smb_read_data);
2089 smb_read_data = NULL;
2097 static int cifs_readpage_worker(struct file *file, struct page *page,
2103 /* Is the page cached? */
2104 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2108 page_cache_get(page);
2109 read_data = kmap(page);
2110 /* for reads over a certain size could initiate async read ahead */
2112 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2117 cFYI(1, "Bytes read %d", rc);
2119 file->f_path.dentry->d_inode->i_atime =
2120 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2122 if (PAGE_CACHE_SIZE > rc)
2123 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2125 flush_dcache_page(page);
2126 SetPageUptodate(page);
2128 /* send this page to the cache */
2129 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2135 page_cache_release(page);
2141 static int cifs_readpage(struct file *file, struct page *page)
2143 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2149 if (file->private_data == NULL) {
2155 cFYI(1, "readpage %p at offset %d 0x%x\n",
2156 page, (int)offset, (int)offset);
2158 rc = cifs_readpage_worker(file, page, &offset);
2166 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2168 struct cifsFileInfo *open_file;
2170 spin_lock(&cifs_file_list_lock);
2171 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2172 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2173 spin_unlock(&cifs_file_list_lock);
2177 spin_unlock(&cifs_file_list_lock);
2181 /* We do not want to update the file size from server for inodes
2182 open for write - to avoid races with writepage extending
2183 the file - in the future we could consider allowing
2184 refreshing the inode only on increases in the file size
2185 but this is tricky to do without racing with writebehind
2186 page caching in the current Linux kernel design */
2187 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2192 if (is_inode_writable(cifsInode)) {
2193 /* This inode is open for write at least once */
2194 struct cifs_sb_info *cifs_sb;
2196 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2197 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2198 /* since no page cache to corrupt on directio
2199 we can change size safely */
2203 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2211 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2212 loff_t pos, unsigned len, unsigned flags,
2213 struct page **pagep, void **fsdata)
2215 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2216 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2217 loff_t page_start = pos & PAGE_MASK;
2222 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2224 page = grab_cache_page_write_begin(mapping, index, flags);
2230 if (PageUptodate(page))
2234 * If we write a full page it will be up to date, no need to read from
2235 * the server. If the write is short, we'll end up doing a sync write
2238 if (len == PAGE_CACHE_SIZE)
2242 * optimize away the read when we have an oplock, and we're not
2243 * expecting to use any of the data we'd be reading in. That
2244 * is, when the page lies beyond the EOF, or straddles the EOF
2245 * and the write will cover all of the existing data.
2247 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2248 i_size = i_size_read(mapping->host);
2249 if (page_start >= i_size ||
2250 (offset == 0 && (pos + len) >= i_size)) {
2251 zero_user_segments(page, 0, offset,
2255 * PageChecked means that the parts of the page
2256 * to which we're not writing are considered up
2257 * to date. Once the data is copied to the
2258 * page, it can be set uptodate.
2260 SetPageChecked(page);
2265 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2267 * might as well read a page, it is fast enough. If we get
2268 * an error, we don't need to return it. cifs_write_end will
2269 * do a sync write instead since PG_uptodate isn't set.
2271 cifs_readpage_worker(file, page, &page_start);
2273 /* we could try using another file handle if there is one -
2274 but how would we lock it to prevent close of that handle
2275 racing with this read? In any case
2276 this will be written out by write_end so is fine */
2283 static int cifs_release_page(struct page *page, gfp_t gfp)
2285 if (PagePrivate(page))
2288 return cifs_fscache_release_page(page, gfp);
2291 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2293 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2296 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2299 static int cifs_launder_page(struct page *page)
2302 loff_t range_start = page_offset(page);
2303 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
2304 struct writeback_control wbc = {
2305 .sync_mode = WB_SYNC_ALL,
2307 .range_start = range_start,
2308 .range_end = range_end,
2311 cFYI(1, "Launder page: %p", page);
2313 if (clear_page_dirty_for_io(page))
2314 rc = cifs_writepage_locked(page, &wbc);
2316 cifs_fscache_invalidate_page(page, page->mapping->host);
2320 void cifs_oplock_break(struct work_struct *work)
2322 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2324 struct inode *inode = cfile->dentry->d_inode;
2325 struct cifsInodeInfo *cinode = CIFS_I(inode);
2328 if (inode && S_ISREG(inode->i_mode)) {
2329 if (cinode->clientCanCacheRead)
2330 break_lease(inode, O_RDONLY);
2332 break_lease(inode, O_WRONLY);
2333 rc = filemap_fdatawrite(inode->i_mapping);
2334 if (cinode->clientCanCacheRead == 0) {
2335 rc = filemap_fdatawait(inode->i_mapping);
2336 mapping_set_error(inode->i_mapping, rc);
2337 invalidate_remote_inode(inode);
2339 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2343 * releasing stale oplock after recent reconnect of smb session using
2344 * a now incorrect file handle is not a data integrity issue but do
2345 * not bother sending an oplock release if session to server still is
2346 * disconnected since oplock already released by the server
2348 if (!cfile->oplock_break_cancelled) {
2349 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2350 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false,
2351 cinode->clientCanCacheRead ? 1 : 0);
2352 cFYI(1, "Oplock release rc = %d", rc);
2356 * We might have kicked in before is_valid_oplock_break()
2357 * finished grabbing reference for us. Make sure it's done by
2358 * waiting for cifs_file_list_lock.
2360 spin_lock(&cifs_file_list_lock);
2361 spin_unlock(&cifs_file_list_lock);
2363 cifs_oplock_break_put(cfile);
2366 /* must be called while holding cifs_file_list_lock */
2367 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2369 cifs_sb_active(cfile->dentry->d_sb);
2370 cifsFileInfo_get(cfile);
2373 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2375 struct super_block *sb = cfile->dentry->d_sb;
2377 cifsFileInfo_put(cfile);
2378 cifs_sb_deactive(sb);
2381 const struct address_space_operations cifs_addr_ops = {
2382 .readpage = cifs_readpage,
2383 .readpages = cifs_readpages,
2384 .writepage = cifs_writepage,
2385 .writepages = cifs_writepages,
2386 .write_begin = cifs_write_begin,
2387 .write_end = cifs_write_end,
2388 .set_page_dirty = __set_page_dirty_nobuffers,
2389 .releasepage = cifs_release_page,
2390 .invalidatepage = cifs_invalidate_page,
2391 .launder_page = cifs_launder_page,
2395 * cifs_readpages requires the server to support a buffer large enough to
2396 * contain the header plus one complete page of data. Otherwise, we need
2397 * to leave cifs_readpages out of the address space operations.
2399 const struct address_space_operations cifs_addr_ops_smallbuf = {
2400 .readpage = cifs_readpage,
2401 .writepage = cifs_writepage,
2402 .writepages = cifs_writepages,
2403 .write_begin = cifs_write_begin,
2404 .write_end = cifs_write_end,
2405 .set_page_dirty = __set_page_dirty_nobuffers,
2406 .releasepage = cifs_release_page,
2407 .invalidatepage = cifs_invalidate_page,
2408 .launder_page = cifs_launder_page,