4 * vfs operations that deal with files
6 * Copyright (C) International Business Machines Corp., 2002,2010
7 * Author(s): Steve French (sfrench@us.ibm.com)
8 * Jeremy Allison (jra@samba.org)
10 * This library is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU Lesser General Public License as published
12 * by the Free Software Foundation; either version 2.1 of the License, or
13 * (at your option) any later version.
15 * This library is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
18 * the GNU Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public License
21 * along with this library; if not, write to the Free Software
22 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
25 #include <linux/backing-dev.h>
26 #include <linux/stat.h>
27 #include <linux/fcntl.h>
28 #include <linux/pagemap.h>
29 #include <linux/pagevec.h>
30 #include <linux/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/delay.h>
33 #include <linux/mount.h>
34 #include <linux/slab.h>
35 #include <asm/div64.h>
39 #include "cifsproto.h"
40 #include "cifs_unicode.h"
41 #include "cifs_debug.h"
42 #include "cifs_fs_sb.h"
45 static inline int cifs_convert_flags(unsigned int flags)
47 if ((flags & O_ACCMODE) == O_RDONLY)
49 else if ((flags & O_ACCMODE) == O_WRONLY)
51 else if ((flags & O_ACCMODE) == O_RDWR) {
52 /* GENERIC_ALL is too much permission to request
53 can cause unnecessary access denied on create */
54 /* return GENERIC_ALL; */
55 return (GENERIC_READ | GENERIC_WRITE);
58 return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
59 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
63 static u32 cifs_posix_convert_flags(unsigned int flags)
67 if ((flags & O_ACCMODE) == O_RDONLY)
68 posix_flags = SMB_O_RDONLY;
69 else if ((flags & O_ACCMODE) == O_WRONLY)
70 posix_flags = SMB_O_WRONLY;
71 else if ((flags & O_ACCMODE) == O_RDWR)
72 posix_flags = SMB_O_RDWR;
75 posix_flags |= SMB_O_CREAT;
77 posix_flags |= SMB_O_EXCL;
79 posix_flags |= SMB_O_TRUNC;
80 /* be safe and imply O_SYNC for O_DSYNC */
82 posix_flags |= SMB_O_SYNC;
83 if (flags & O_DIRECTORY)
84 posix_flags |= SMB_O_DIRECTORY;
85 if (flags & O_NOFOLLOW)
86 posix_flags |= SMB_O_NOFOLLOW;
88 posix_flags |= SMB_O_DIRECT;
93 static inline int cifs_get_disposition(unsigned int flags)
95 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
97 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
98 return FILE_OVERWRITE_IF;
99 else if ((flags & O_CREAT) == O_CREAT)
101 else if ((flags & O_TRUNC) == O_TRUNC)
102 return FILE_OVERWRITE;
107 static inline int cifs_open_inode_helper(struct inode *inode,
108 struct cifsTconInfo *pTcon, __u32 oplock, FILE_ALL_INFO *buf,
109 char *full_path, int xid)
111 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
112 struct timespec temp;
115 if (pCifsInode->clientCanCacheRead) {
116 /* we have the inode open somewhere else
117 no need to discard cache data */
118 goto client_can_cache;
121 /* BB need same check in cifs_create too? */
122 /* if not oplocked, invalidate inode pages if mtime or file
124 temp = cifs_NTtimeToUnix(buf->LastWriteTime);
125 if (timespec_equal(&inode->i_mtime, &temp) &&
127 (loff_t)le64_to_cpu(buf->EndOfFile))) {
128 cFYI(1, "inode unchanged on server");
130 if (inode->i_mapping) {
131 /* BB no need to lock inode until after invalidate
132 since namei code should already have it locked? */
133 rc = filemap_write_and_wait(inode->i_mapping);
134 mapping_set_error(inode->i_mapping, rc);
136 cFYI(1, "invalidating remote inode since open detected it "
138 invalidate_remote_inode(inode);
143 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
146 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
149 cifs_set_oplock_level(inode, oplock);
154 int cifs_posix_open(char *full_path, struct inode **pinode,
155 struct super_block *sb, int mode, unsigned int f_flags,
156 __u32 *poplock, __u16 *pnetfid, int xid)
159 FILE_UNIX_BASIC_INFO *presp_data;
160 __u32 posix_flags = 0;
161 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
162 struct cifs_fattr fattr;
163 struct tcon_link *tlink;
164 struct cifsTconInfo *tcon;
166 cFYI(1, "posix open %s", full_path);
168 presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
169 if (presp_data == NULL)
172 tlink = cifs_sb_tlink(cifs_sb);
178 tcon = tlink_tcon(tlink);
179 mode &= ~current_umask();
181 posix_flags = cifs_posix_convert_flags(f_flags);
182 rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
183 poplock, full_path, cifs_sb->local_nls,
184 cifs_sb->mnt_cifs_flags &
185 CIFS_MOUNT_MAP_SPECIAL_CHR);
186 cifs_put_tlink(tlink);
191 if (presp_data->Type == cpu_to_le32(-1))
192 goto posix_open_ret; /* open ok, caller does qpathinfo */
195 goto posix_open_ret; /* caller does not need info */
197 cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
199 /* get new inode and set it up */
200 if (*pinode == NULL) {
201 cifs_fill_uniqueid(sb, &fattr);
202 *pinode = cifs_iget(sb, &fattr);
208 cifs_fattr_to_inode(*pinode, &fattr);
216 struct cifsFileInfo *
217 cifs_new_fileinfo(__u16 fileHandle, struct file *file,
218 struct tcon_link *tlink, __u32 oplock)
220 struct dentry *dentry = file->f_path.dentry;
221 struct inode *inode = dentry->d_inode;
222 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
223 struct cifsFileInfo *pCifsFile;
225 pCifsFile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
226 if (pCifsFile == NULL)
229 pCifsFile->count = 1;
230 pCifsFile->netfid = fileHandle;
231 pCifsFile->pid = current->tgid;
232 pCifsFile->uid = current_fsuid();
233 pCifsFile->dentry = dget(dentry);
234 pCifsFile->f_flags = file->f_flags;
235 pCifsFile->invalidHandle = false;
236 pCifsFile->tlink = cifs_get_tlink(tlink);
237 mutex_init(&pCifsFile->fh_mutex);
238 mutex_init(&pCifsFile->lock_mutex);
239 INIT_LIST_HEAD(&pCifsFile->llist);
240 INIT_WORK(&pCifsFile->oplock_break, cifs_oplock_break);
242 spin_lock(&cifs_file_list_lock);
243 list_add(&pCifsFile->tlist, &(tlink_tcon(tlink)->openFileList));
244 /* if readable file instance put first in list*/
245 if (file->f_mode & FMODE_READ)
246 list_add(&pCifsFile->flist, &pCifsInode->openFileList);
248 list_add_tail(&pCifsFile->flist, &pCifsInode->openFileList);
249 spin_unlock(&cifs_file_list_lock);
251 cifs_set_oplock_level(inode, oplock);
253 file->private_data = pCifsFile;
258 * Release a reference on the file private data. This may involve closing
259 * the filehandle out on the server. Must be called without holding
260 * cifs_file_list_lock.
262 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
264 struct inode *inode = cifs_file->dentry->d_inode;
265 struct cifsTconInfo *tcon = tlink_tcon(cifs_file->tlink);
266 struct cifsInodeInfo *cifsi = CIFS_I(inode);
267 struct cifsLockInfo *li, *tmp;
269 spin_lock(&cifs_file_list_lock);
270 if (--cifs_file->count > 0) {
271 spin_unlock(&cifs_file_list_lock);
275 /* remove it from the lists */
276 list_del(&cifs_file->flist);
277 list_del(&cifs_file->tlist);
279 if (list_empty(&cifsi->openFileList)) {
280 cFYI(1, "closing last open instance for inode %p",
281 cifs_file->dentry->d_inode);
282 cifs_set_oplock_level(inode, 0);
284 spin_unlock(&cifs_file_list_lock);
286 if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
290 rc = CIFSSMBClose(xid, tcon, cifs_file->netfid);
294 /* Delete any outstanding lock records. We'll lose them when the file
297 mutex_lock(&cifs_file->lock_mutex);
298 list_for_each_entry_safe(li, tmp, &cifs_file->llist, llist) {
299 list_del(&li->llist);
302 mutex_unlock(&cifs_file->lock_mutex);
304 cifs_put_tlink(cifs_file->tlink);
305 dput(cifs_file->dentry);
309 int cifs_open(struct inode *inode, struct file *file)
314 struct cifs_sb_info *cifs_sb;
315 struct cifsTconInfo *tcon;
316 struct tcon_link *tlink;
317 struct cifsFileInfo *pCifsFile = NULL;
318 struct cifsInodeInfo *pCifsInode;
319 char *full_path = NULL;
323 FILE_ALL_INFO *buf = NULL;
327 cifs_sb = CIFS_SB(inode->i_sb);
328 tlink = cifs_sb_tlink(cifs_sb);
331 return PTR_ERR(tlink);
333 tcon = tlink_tcon(tlink);
335 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
337 full_path = build_path_from_dentry(file->f_path.dentry);
338 if (full_path == NULL) {
343 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
344 inode, file->f_flags, full_path);
351 if (!tcon->broken_posix_open && tcon->unix_ext &&
352 (tcon->ses->capabilities & CAP_UNIX) &&
353 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
354 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
355 /* can not refresh inode info since size could be stale */
356 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
357 cifs_sb->mnt_file_mode /* ignored */,
358 file->f_flags, &oplock, &netfid, xid);
360 cFYI(1, "posix open succeeded");
362 pCifsFile = cifs_new_fileinfo(netfid, file, tlink,
364 if (pCifsFile == NULL) {
365 CIFSSMBClose(xid, tcon, netfid);
369 cifs_fscache_set_inode_cookie(inode, file);
372 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
373 if (tcon->ses->serverNOS)
374 cERROR(1, "server %s of type %s returned"
375 " unexpected error on SMB posix open"
376 ", disabling posix open support."
377 " Check if server update available.",
378 tcon->ses->serverName,
379 tcon->ses->serverNOS);
380 tcon->broken_posix_open = true;
381 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
382 (rc != -EOPNOTSUPP)) /* path not found or net err */
384 /* else fallthrough to retry open the old way on network i/o
388 desiredAccess = cifs_convert_flags(file->f_flags);
390 /*********************************************************************
391 * open flag mapping table:
393 * POSIX Flag CIFS Disposition
394 * ---------- ----------------
395 * O_CREAT FILE_OPEN_IF
396 * O_CREAT | O_EXCL FILE_CREATE
397 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
398 * O_TRUNC FILE_OVERWRITE
399 * none of the above FILE_OPEN
401 * Note that there is not a direct match between disposition
402 * FILE_SUPERSEDE (ie create whether or not file exists although
403 * O_CREAT | O_TRUNC is similar but truncates the existing
404 * file rather than creating a new file as FILE_SUPERSEDE does
405 * (which uses the attributes / metadata passed in on open call)
407 *? O_SYNC is a reasonable match to CIFS writethrough flag
408 *? and the read write flags match reasonably. O_LARGEFILE
409 *? is irrelevant because largefile support is always used
410 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
411 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
412 *********************************************************************/
414 disposition = cifs_get_disposition(file->f_flags);
416 /* BB pass O_SYNC flag through on file attributes .. BB */
418 /* Also refresh inode by passing in file_info buf returned by SMBOpen
419 and calling get_inode_info with returned buf (at least helps
420 non-Unix server case) */
422 /* BB we can not do this if this is the second open of a file
423 and the first handle has writebehind data, we might be
424 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
425 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
431 if (tcon->ses->capabilities & CAP_NT_SMBS)
432 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
433 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
434 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
435 & CIFS_MOUNT_MAP_SPECIAL_CHR);
437 rc = -EIO; /* no NT SMB support fall into legacy open below */
440 /* Old server, try legacy style OpenX */
441 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
442 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
443 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
444 & CIFS_MOUNT_MAP_SPECIAL_CHR);
447 cFYI(1, "cifs_open returned 0x%x", rc);
451 rc = cifs_open_inode_helper(inode, tcon, oplock, buf, full_path, xid);
455 pCifsFile = cifs_new_fileinfo(netfid, file, tlink, oplock);
456 if (pCifsFile == NULL) {
461 cifs_fscache_set_inode_cookie(inode, file);
463 if (oplock & CIFS_CREATE_ACTION) {
464 /* time to set mode which we can not set earlier due to
465 problems creating new read-only files */
466 if (tcon->unix_ext) {
467 struct cifs_unix_set_info_args args = {
468 .mode = inode->i_mode,
471 .ctime = NO_CHANGE_64,
472 .atime = NO_CHANGE_64,
473 .mtime = NO_CHANGE_64,
476 CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
478 cifs_sb->mnt_cifs_flags &
479 CIFS_MOUNT_MAP_SPECIAL_CHR);
487 cifs_put_tlink(tlink);
491 /* Try to reacquire byte range locks that were released when session */
492 /* to server was lost */
493 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
497 /* BB list all locks open on this file and relock */
502 static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
507 struct cifs_sb_info *cifs_sb;
508 struct cifsTconInfo *tcon;
509 struct cifsInodeInfo *pCifsInode;
511 char *full_path = NULL;
513 int disposition = FILE_OPEN;
517 mutex_lock(&pCifsFile->fh_mutex);
518 if (!pCifsFile->invalidHandle) {
519 mutex_unlock(&pCifsFile->fh_mutex);
525 inode = pCifsFile->dentry->d_inode;
526 cifs_sb = CIFS_SB(inode->i_sb);
527 tcon = tlink_tcon(pCifsFile->tlink);
529 /* can not grab rename sem here because various ops, including
530 those that already have the rename sem can end up causing writepage
531 to get called and if the server was down that means we end up here,
532 and we can never tell if the caller already has the rename_sem */
533 full_path = build_path_from_dentry(pCifsFile->dentry);
534 if (full_path == NULL) {
536 mutex_unlock(&pCifsFile->fh_mutex);
541 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
542 inode, pCifsFile->f_flags, full_path);
549 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
550 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
551 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
554 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
555 * original open. Must mask them off for a reopen.
557 unsigned int oflags = pCifsFile->f_flags &
558 ~(O_CREAT | O_EXCL | O_TRUNC);
560 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
561 cifs_sb->mnt_file_mode /* ignored */,
562 oflags, &oplock, &netfid, xid);
564 cFYI(1, "posix reopen succeeded");
567 /* fallthrough to retry open the old way on errors, especially
568 in the reconnect path it is important to retry hard */
571 desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
573 /* Can not refresh inode by passing in file_info buf to be returned
574 by SMBOpen and then calling get_inode_info with returned buf
575 since file might have write behind data that needs to be flushed
576 and server version of file size can be stale. If we knew for sure
577 that inode was not dirty locally we could do this */
579 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
580 CREATE_NOT_DIR, &netfid, &oplock, NULL,
581 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
582 CIFS_MOUNT_MAP_SPECIAL_CHR);
584 mutex_unlock(&pCifsFile->fh_mutex);
585 cFYI(1, "cifs_open returned 0x%x", rc);
586 cFYI(1, "oplock: %d", oplock);
587 goto reopen_error_exit;
591 pCifsFile->netfid = netfid;
592 pCifsFile->invalidHandle = false;
593 mutex_unlock(&pCifsFile->fh_mutex);
594 pCifsInode = CIFS_I(inode);
597 rc = filemap_write_and_wait(inode->i_mapping);
598 mapping_set_error(inode->i_mapping, rc);
601 rc = cifs_get_inode_info_unix(&inode,
602 full_path, inode->i_sb, xid);
604 rc = cifs_get_inode_info(&inode,
605 full_path, NULL, inode->i_sb,
607 } /* else we are writing out data to server already
608 and could deadlock if we tried to flush data, and
609 since we do not know if we have data that would
610 invalidate the current end of file on the server
611 we can not go to the server to get the new inod
614 cifs_set_oplock_level(inode, oplock);
616 cifs_relock_file(pCifsFile);
624 int cifs_close(struct inode *inode, struct file *file)
626 cifsFileInfo_put(file->private_data);
627 file->private_data = NULL;
629 /* return code from the ->release op is always ignored */
633 int cifs_closedir(struct inode *inode, struct file *file)
637 struct cifsFileInfo *pCFileStruct = file->private_data;
640 cFYI(1, "Closedir inode = 0x%p", inode);
645 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
647 cFYI(1, "Freeing private data in close dir");
648 spin_lock(&cifs_file_list_lock);
649 if (!pCFileStruct->srch_inf.endOfSearch &&
650 !pCFileStruct->invalidHandle) {
651 pCFileStruct->invalidHandle = true;
652 spin_unlock(&cifs_file_list_lock);
653 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
654 cFYI(1, "Closing uncompleted readdir with rc %d",
656 /* not much we can do if it fails anyway, ignore rc */
659 spin_unlock(&cifs_file_list_lock);
660 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
662 cFYI(1, "closedir free smb buf in srch struct");
663 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
664 if (pCFileStruct->srch_inf.smallBuf)
665 cifs_small_buf_release(ptmp);
667 cifs_buf_release(ptmp);
669 cifs_put_tlink(pCFileStruct->tlink);
670 kfree(file->private_data);
671 file->private_data = NULL;
673 /* BB can we lock the filestruct while this is going on? */
678 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
679 __u64 offset, __u8 lockType)
681 struct cifsLockInfo *li =
682 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
688 mutex_lock(&fid->lock_mutex);
689 list_add(&li->llist, &fid->llist);
690 mutex_unlock(&fid->lock_mutex);
694 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
700 bool wait_flag = false;
701 struct cifs_sb_info *cifs_sb;
702 struct cifsTconInfo *tcon;
704 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
705 bool posix_locking = 0;
707 length = 1 + pfLock->fl_end - pfLock->fl_start;
711 cFYI(1, "Lock parm: 0x%x flockflags: "
712 "0x%x flocktype: 0x%x start: %lld end: %lld",
713 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
716 if (pfLock->fl_flags & FL_POSIX)
718 if (pfLock->fl_flags & FL_FLOCK)
720 if (pfLock->fl_flags & FL_SLEEP) {
721 cFYI(1, "Blocking lock");
724 if (pfLock->fl_flags & FL_ACCESS)
725 cFYI(1, "Process suspended by mandatory locking - "
726 "not implemented yet");
727 if (pfLock->fl_flags & FL_LEASE)
728 cFYI(1, "Lease on file - not implemented yet");
729 if (pfLock->fl_flags &
730 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
731 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
733 if (pfLock->fl_type == F_WRLCK) {
736 } else if (pfLock->fl_type == F_UNLCK) {
739 /* Check if unlock includes more than
741 } else if (pfLock->fl_type == F_RDLCK) {
743 lockType |= LOCKING_ANDX_SHARED_LOCK;
745 } else if (pfLock->fl_type == F_EXLCK) {
748 } else if (pfLock->fl_type == F_SHLCK) {
750 lockType |= LOCKING_ANDX_SHARED_LOCK;
753 cFYI(1, "Unknown type of lock");
755 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
756 tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
758 if (file->private_data == NULL) {
763 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
765 if ((tcon->ses->capabilities & CAP_UNIX) &&
766 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
767 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
769 /* BB add code here to normalize offset and length to
770 account for negative length which we can not accept over the
775 if (lockType & LOCKING_ANDX_SHARED_LOCK)
776 posix_lock_type = CIFS_RDLCK;
778 posix_lock_type = CIFS_WRLCK;
779 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
781 posix_lock_type, wait_flag);
786 /* BB we could chain these into one lock request BB */
787 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
788 0, 1, lockType, 0 /* wait flag */ );
790 rc = CIFSSMBLock(xid, tcon, netfid, length,
791 pfLock->fl_start, 1 /* numUnlock */ ,
792 0 /* numLock */ , lockType,
794 pfLock->fl_type = F_UNLCK;
796 cERROR(1, "Error unlocking previously locked "
797 "range %d during test of lock", rc);
801 /* if rc == ERR_SHARING_VIOLATION ? */
804 if (lockType & LOCKING_ANDX_SHARED_LOCK) {
805 pfLock->fl_type = F_WRLCK;
807 rc = CIFSSMBLock(xid, tcon, netfid, length,
808 pfLock->fl_start, 0, 1,
809 lockType | LOCKING_ANDX_SHARED_LOCK,
812 rc = CIFSSMBLock(xid, tcon, netfid,
813 length, pfLock->fl_start, 1, 0,
815 LOCKING_ANDX_SHARED_LOCK,
817 pfLock->fl_type = F_RDLCK;
819 cERROR(1, "Error unlocking "
820 "previously locked range %d "
821 "during test of lock", rc);
824 pfLock->fl_type = F_WRLCK;
834 if (!numLock && !numUnlock) {
835 /* if no lock or unlock then nothing
836 to do since we do not know what it is */
843 if (lockType & LOCKING_ANDX_SHARED_LOCK)
844 posix_lock_type = CIFS_RDLCK;
846 posix_lock_type = CIFS_WRLCK;
849 posix_lock_type = CIFS_UNLCK;
851 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
853 posix_lock_type, wait_flag);
855 struct cifsFileInfo *fid = file->private_data;
858 rc = CIFSSMBLock(xid, tcon, netfid, length,
860 0, numLock, lockType, wait_flag);
863 /* For Windows locks we must store them. */
864 rc = store_file_lock(fid, length,
865 pfLock->fl_start, lockType);
867 } else if (numUnlock) {
868 /* For each stored lock that this unlock overlaps
869 completely, unlock it. */
871 struct cifsLockInfo *li, *tmp;
874 mutex_lock(&fid->lock_mutex);
875 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
876 if (pfLock->fl_start <= li->offset &&
877 (pfLock->fl_start + length) >=
878 (li->offset + li->length)) {
879 stored_rc = CIFSSMBLock(xid, tcon,
881 li->length, li->offset,
882 1, 0, li->type, false);
886 list_del(&li->llist);
891 mutex_unlock(&fid->lock_mutex);
895 if (pfLock->fl_flags & FL_POSIX)
896 posix_lock_file_wait(file, pfLock);
902 * Set the timeout on write requests past EOF. For some servers (Windows)
903 * these calls can be very long.
905 * If we're writing >10M past the EOF we give a 180s timeout. Anything less
906 * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
907 * The 10M cutoff is totally arbitrary. A better scheme for this would be
908 * welcome if someone wants to suggest one.
910 * We may be able to do a better job with this if there were some way to
911 * declare that a file should be sparse.
914 cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
916 if (offset <= cifsi->server_eof)
918 else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
919 return CIFS_VLONG_OP;
924 /* update the file size (if needed) after a write */
926 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
927 unsigned int bytes_written)
929 loff_t end_of_write = offset + bytes_written;
931 if (end_of_write > cifsi->server_eof)
932 cifsi->server_eof = end_of_write;
935 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
936 size_t write_size, loff_t *poffset)
938 struct inode *inode = file->f_path.dentry->d_inode;
940 unsigned int bytes_written = 0;
941 unsigned int total_written;
942 struct cifs_sb_info *cifs_sb;
943 struct cifsTconInfo *pTcon;
945 struct cifsFileInfo *open_file;
946 struct cifsInodeInfo *cifsi = CIFS_I(inode);
948 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
950 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
951 *poffset, file->f_path.dentry->d_name.name); */
953 if (file->private_data == NULL)
956 open_file = file->private_data;
957 pTcon = tlink_tcon(open_file->tlink);
959 rc = generic_write_checks(file, poffset, &write_size, 0);
965 long_op = cifs_write_timeout(cifsi, *poffset);
966 for (total_written = 0; write_size > total_written;
967 total_written += bytes_written) {
969 while (rc == -EAGAIN) {
970 if (file->private_data == NULL) {
971 /* file has been closed on us */
973 /* if we have gotten here we have written some data
974 and blocked, and the file has been freed on us while
975 we blocked so return what we managed to write */
976 return total_written;
978 if (open_file->invalidHandle) {
979 /* we could deadlock if we called
980 filemap_fdatawait from here so tell
981 reopen_file not to flush data to server
983 rc = cifs_reopen_file(open_file, false);
988 rc = CIFSSMBWrite(xid, pTcon,
990 min_t(const int, cifs_sb->wsize,
991 write_size - total_written),
992 *poffset, &bytes_written,
993 NULL, write_data + total_written, long_op);
995 if (rc || (bytes_written == 0)) {
1003 cifs_update_eof(cifsi, *poffset, bytes_written);
1004 *poffset += bytes_written;
1006 long_op = CIFS_STD_OP; /* subsequent writes fast -
1007 15 seconds is plenty */
1010 cifs_stats_bytes_written(pTcon, total_written);
1012 /* Do not update local mtime - server will set its actual value on write
1013 * inode->i_ctime = inode->i_mtime =
1014 * current_fs_time(inode->i_sb);*/
1015 if (total_written > 0) {
1016 spin_lock(&inode->i_lock);
1017 if (*poffset > inode->i_size)
1018 i_size_write(inode, *poffset);
1019 spin_unlock(&inode->i_lock);
1021 mark_inode_dirty_sync(inode);
1024 return total_written;
1027 static ssize_t cifs_write(struct cifsFileInfo *open_file,
1028 const char *write_data, size_t write_size,
1032 unsigned int bytes_written = 0;
1033 unsigned int total_written;
1034 struct cifs_sb_info *cifs_sb;
1035 struct cifsTconInfo *pTcon;
1037 struct dentry *dentry = open_file->dentry;
1038 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
1040 cifs_sb = CIFS_SB(dentry->d_sb);
1042 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
1043 *poffset, dentry->d_name.name);
1045 pTcon = tlink_tcon(open_file->tlink);
1049 long_op = cifs_write_timeout(cifsi, *poffset);
1050 for (total_written = 0; write_size > total_written;
1051 total_written += bytes_written) {
1053 while (rc == -EAGAIN) {
1054 if (open_file->invalidHandle) {
1055 /* we could deadlock if we called
1056 filemap_fdatawait from here so tell
1057 reopen_file not to flush data to
1059 rc = cifs_reopen_file(open_file, false);
1063 if (experimEnabled || (pTcon->ses->server &&
1064 ((pTcon->ses->server->secMode &
1065 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1070 len = min((size_t)cifs_sb->wsize,
1071 write_size - total_written);
1072 /* iov[0] is reserved for smb header */
1073 iov[1].iov_base = (char *)write_data +
1075 iov[1].iov_len = len;
1076 rc = CIFSSMBWrite2(xid, pTcon,
1077 open_file->netfid, len,
1078 *poffset, &bytes_written,
1081 rc = CIFSSMBWrite(xid, pTcon,
1083 min_t(const int, cifs_sb->wsize,
1084 write_size - total_written),
1085 *poffset, &bytes_written,
1086 write_data + total_written,
1089 if (rc || (bytes_written == 0)) {
1097 cifs_update_eof(cifsi, *poffset, bytes_written);
1098 *poffset += bytes_written;
1100 long_op = CIFS_STD_OP; /* subsequent writes fast -
1101 15 seconds is plenty */
1104 cifs_stats_bytes_written(pTcon, total_written);
1106 if (total_written > 0) {
1107 spin_lock(&dentry->d_inode->i_lock);
1108 if (*poffset > dentry->d_inode->i_size)
1109 i_size_write(dentry->d_inode, *poffset);
1110 spin_unlock(&dentry->d_inode->i_lock);
1112 mark_inode_dirty_sync(dentry->d_inode);
1114 return total_written;
1117 #ifdef CONFIG_CIFS_EXPERIMENTAL
1118 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1121 struct cifsFileInfo *open_file = NULL;
1122 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1124 /* only filter by fsuid on multiuser mounts */
1125 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1128 spin_lock(&cifs_file_list_lock);
1129 /* we could simply get the first_list_entry since write-only entries
1130 are always at the end of the list but since the first entry might
1131 have a close pending, we go through the whole list */
1132 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1133 if (fsuid_only && open_file->uid != current_fsuid())
1135 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1136 if (!open_file->invalidHandle) {
1137 /* found a good file */
1138 /* lock it so it will not be closed on us */
1139 cifsFileInfo_get(open_file);
1140 spin_unlock(&cifs_file_list_lock);
1142 } /* else might as well continue, and look for
1143 another, or simply have the caller reopen it
1144 again rather than trying to fix this handle */
1145 } else /* write only file */
1146 break; /* write only files are last so must be done */
1148 spin_unlock(&cifs_file_list_lock);
1153 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1156 struct cifsFileInfo *open_file;
1157 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1158 bool any_available = false;
1161 /* Having a null inode here (because mapping->host was set to zero by
1162 the VFS or MM) should not happen but we had reports of on oops (due to
1163 it being zero) during stress testcases so we need to check for it */
1165 if (cifs_inode == NULL) {
1166 cERROR(1, "Null inode passed to cifs_writeable_file");
1171 /* only filter by fsuid on multiuser mounts */
1172 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1175 spin_lock(&cifs_file_list_lock);
1177 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1178 if (!any_available && open_file->pid != current->tgid)
1180 if (fsuid_only && open_file->uid != current_fsuid())
1182 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1183 cifsFileInfo_get(open_file);
1185 if (!open_file->invalidHandle) {
1186 /* found a good writable file */
1187 spin_unlock(&cifs_file_list_lock);
1191 spin_unlock(&cifs_file_list_lock);
1193 /* Had to unlock since following call can block */
1194 rc = cifs_reopen_file(open_file, false);
1198 /* if it fails, try another handle if possible */
1199 cFYI(1, "wp failed on reopen file");
1200 cifsFileInfo_put(open_file);
1202 spin_lock(&cifs_file_list_lock);
1204 /* else we simply continue to the next entry. Thus
1205 we do not loop on reopen errors. If we
1206 can not reopen the file, for example if we
1207 reconnected to a server with another client
1208 racing to delete or lock the file we would not
1209 make progress if we restarted before the beginning
1210 of the loop here. */
1213 /* couldn't find useable FH with same pid, try any available */
1214 if (!any_available) {
1215 any_available = true;
1216 goto refind_writable;
1218 spin_unlock(&cifs_file_list_lock);
1222 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1224 struct address_space *mapping = page->mapping;
1225 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1228 int bytes_written = 0;
1229 struct cifs_sb_info *cifs_sb;
1230 struct inode *inode;
1231 struct cifsFileInfo *open_file;
1233 if (!mapping || !mapping->host)
1236 inode = page->mapping->host;
1237 cifs_sb = CIFS_SB(inode->i_sb);
1239 offset += (loff_t)from;
1240 write_data = kmap(page);
1243 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1248 /* racing with truncate? */
1249 if (offset > mapping->host->i_size) {
1251 return 0; /* don't care */
1254 /* check to make sure that we are not extending the file */
1255 if (mapping->host->i_size - offset < (loff_t)to)
1256 to = (unsigned)(mapping->host->i_size - offset);
1258 open_file = find_writable_file(CIFS_I(mapping->host), false);
1260 bytes_written = cifs_write(open_file, write_data,
1261 to - from, &offset);
1262 cifsFileInfo_put(open_file);
1263 /* Does mm or vfs already set times? */
1264 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1265 if ((bytes_written > 0) && (offset))
1267 else if (bytes_written < 0)
1270 cFYI(1, "No writeable filehandles for inode");
1278 static int cifs_writepages(struct address_space *mapping,
1279 struct writeback_control *wbc)
1281 unsigned int bytes_to_write;
1282 unsigned int bytes_written;
1283 struct cifs_sb_info *cifs_sb;
1287 int range_whole = 0;
1294 struct cifsFileInfo *open_file;
1295 struct cifsTconInfo *tcon;
1296 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1298 struct pagevec pvec;
1303 cifs_sb = CIFS_SB(mapping->host->i_sb);
1306 * If wsize is smaller that the page cache size, default to writing
1307 * one page at a time via cifs_writepage
1309 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1310 return generic_writepages(mapping, wbc);
1312 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1314 return generic_writepages(mapping, wbc);
1317 * if there's no open file, then this is likely to fail too,
1318 * but it'll at least handle the return. Maybe it should be
1321 open_file = find_writable_file(CIFS_I(mapping->host), false);
1324 return generic_writepages(mapping, wbc);
1327 tcon = tlink_tcon(open_file->tlink);
1328 if (!experimEnabled && tcon->ses->server->secMode &
1329 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1330 cifsFileInfo_put(open_file);
1332 return generic_writepages(mapping, wbc);
1334 cifsFileInfo_put(open_file);
1338 pagevec_init(&pvec, 0);
1339 if (wbc->range_cyclic) {
1340 index = mapping->writeback_index; /* Start from prev offset */
1343 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1344 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1345 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1350 while (!done && (index <= end) &&
1351 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1352 PAGECACHE_TAG_DIRTY,
1353 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1362 for (i = 0; i < nr_pages; i++) {
1363 page = pvec.pages[i];
1365 * At this point we hold neither mapping->tree_lock nor
1366 * lock on the page itself: the page may be truncated or
1367 * invalidated (changing page->mapping to NULL), or even
1368 * swizzled back from swapper_space to tmpfs file
1374 else if (!trylock_page(page))
1377 if (unlikely(page->mapping != mapping)) {
1382 if (!wbc->range_cyclic && page->index > end) {
1388 if (next && (page->index != next)) {
1389 /* Not next consecutive page */
1394 if (wbc->sync_mode != WB_SYNC_NONE)
1395 wait_on_page_writeback(page);
1397 if (PageWriteback(page) ||
1398 !clear_page_dirty_for_io(page)) {
1404 * This actually clears the dirty bit in the radix tree.
1405 * See cifs_writepage() for more commentary.
1407 set_page_writeback(page);
1409 if (page_offset(page) >= mapping->host->i_size) {
1412 end_page_writeback(page);
1417 * BB can we get rid of this? pages are held by pvec
1419 page_cache_get(page);
1421 len = min(mapping->host->i_size - page_offset(page),
1422 (loff_t)PAGE_CACHE_SIZE);
1424 /* reserve iov[0] for the smb header */
1426 iov[n_iov].iov_base = kmap(page);
1427 iov[n_iov].iov_len = len;
1428 bytes_to_write += len;
1432 offset = page_offset(page);
1434 next = page->index + 1;
1435 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1439 open_file = find_writable_file(CIFS_I(mapping->host),
1442 cERROR(1, "No writable handles for inode");
1445 long_op = cifs_write_timeout(cifsi, offset);
1446 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1447 bytes_to_write, offset,
1448 &bytes_written, iov, n_iov,
1450 cifsFileInfo_put(open_file);
1451 cifs_update_eof(cifsi, offset, bytes_written);
1454 if (rc || bytes_written < bytes_to_write) {
1455 cERROR(1, "Write2 ret %d, wrote %d",
1457 mapping_set_error(mapping, rc);
1459 cifs_stats_bytes_written(tcon, bytes_written);
1462 for (i = 0; i < n_iov; i++) {
1463 page = pvec.pages[first + i];
1464 /* Should we also set page error on
1465 success rc but too little data written? */
1466 /* BB investigate retry logic on temporary
1467 server crash cases and how recovery works
1468 when page marked as error */
1473 end_page_writeback(page);
1474 page_cache_release(page);
1476 if ((wbc->nr_to_write -= n_iov) <= 0)
1480 /* Need to re-find the pages we skipped */
1481 index = pvec.pages[0]->index + 1;
1483 pagevec_release(&pvec);
1485 if (!scanned && !done) {
1487 * We hit the last page and there is more work to be done: wrap
1488 * back to the start of the file
1494 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1495 mapping->writeback_index = index;
1502 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1508 /* BB add check for wbc flags */
1509 page_cache_get(page);
1510 if (!PageUptodate(page))
1511 cFYI(1, "ppw - page not up to date");
1514 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1516 * A writepage() implementation always needs to do either this,
1517 * or re-dirty the page with "redirty_page_for_writepage()" in
1518 * the case of a failure.
1520 * Just unlocking the page will cause the radix tree tag-bits
1521 * to fail to update with the state of the page correctly.
1523 set_page_writeback(page);
1524 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1525 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1527 end_page_writeback(page);
1528 page_cache_release(page);
1533 static int cifs_write_end(struct file *file, struct address_space *mapping,
1534 loff_t pos, unsigned len, unsigned copied,
1535 struct page *page, void *fsdata)
1538 struct inode *inode = mapping->host;
1540 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1543 if (PageChecked(page)) {
1545 SetPageUptodate(page);
1546 ClearPageChecked(page);
1547 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1548 SetPageUptodate(page);
1550 if (!PageUptodate(page)) {
1552 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1556 /* this is probably better than directly calling
1557 partialpage_write since in this function the file handle is
1558 known which we might as well leverage */
1559 /* BB check if anything else missing out of ppw
1560 such as updating last write time */
1561 page_data = kmap(page);
1562 rc = cifs_write(file->private_data, page_data + offset,
1564 /* if (rc < 0) should we set writebehind rc? */
1571 set_page_dirty(page);
1575 spin_lock(&inode->i_lock);
1576 if (pos > inode->i_size)
1577 i_size_write(inode, pos);
1578 spin_unlock(&inode->i_lock);
1582 page_cache_release(page);
1587 int cifs_fsync(struct file *file, int datasync)
1591 struct cifsTconInfo *tcon;
1592 struct cifsFileInfo *smbfile = file->private_data;
1593 struct inode *inode = file->f_path.dentry->d_inode;
1597 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1598 file->f_path.dentry->d_name.name, datasync);
1600 rc = filemap_write_and_wait(inode->i_mapping);
1602 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
1604 tcon = tlink_tcon(smbfile->tlink);
1605 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1606 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1613 /* static void cifs_sync_page(struct page *page)
1615 struct address_space *mapping;
1616 struct inode *inode;
1617 unsigned long index = page->index;
1618 unsigned int rpages = 0;
1621 cFYI(1, "sync page %p", page);
1622 mapping = page->mapping;
1625 inode = mapping->host;
1629 /* fill in rpages then
1630 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1632 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1642 * As file closes, flush all cached write data for this inode checking
1643 * for write behind errors.
1645 int cifs_flush(struct file *file, fl_owner_t id)
1647 struct inode *inode = file->f_path.dentry->d_inode;
1650 if (file->f_mode & FMODE_WRITE)
1651 rc = filemap_write_and_wait(inode->i_mapping);
1653 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1658 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1659 size_t read_size, loff_t *poffset)
1662 unsigned int bytes_read = 0;
1663 unsigned int total_read = 0;
1664 unsigned int current_read_size;
1665 struct cifs_sb_info *cifs_sb;
1666 struct cifsTconInfo *pTcon;
1668 struct cifsFileInfo *open_file;
1669 char *smb_read_data;
1670 char __user *current_offset;
1671 struct smb_com_read_rsp *pSMBr;
1674 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1676 if (file->private_data == NULL) {
1681 open_file = file->private_data;
1682 pTcon = tlink_tcon(open_file->tlink);
1684 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1685 cFYI(1, "attempting read on write only file instance");
1687 for (total_read = 0, current_offset = read_data;
1688 read_size > total_read;
1689 total_read += bytes_read, current_offset += bytes_read) {
1690 current_read_size = min_t(const int, read_size - total_read,
1693 smb_read_data = NULL;
1694 while (rc == -EAGAIN) {
1695 int buf_type = CIFS_NO_BUFFER;
1696 if (open_file->invalidHandle) {
1697 rc = cifs_reopen_file(open_file, true);
1701 rc = CIFSSMBRead(xid, pTcon,
1703 current_read_size, *poffset,
1704 &bytes_read, &smb_read_data,
1706 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1707 if (smb_read_data) {
1708 if (copy_to_user(current_offset,
1710 4 /* RFC1001 length field */ +
1711 le16_to_cpu(pSMBr->DataOffset),
1715 if (buf_type == CIFS_SMALL_BUFFER)
1716 cifs_small_buf_release(smb_read_data);
1717 else if (buf_type == CIFS_LARGE_BUFFER)
1718 cifs_buf_release(smb_read_data);
1719 smb_read_data = NULL;
1722 if (rc || (bytes_read == 0)) {
1730 cifs_stats_bytes_read(pTcon, bytes_read);
1731 *poffset += bytes_read;
1739 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1743 unsigned int bytes_read = 0;
1744 unsigned int total_read;
1745 unsigned int current_read_size;
1746 struct cifs_sb_info *cifs_sb;
1747 struct cifsTconInfo *pTcon;
1749 char *current_offset;
1750 struct cifsFileInfo *open_file;
1751 int buf_type = CIFS_NO_BUFFER;
1754 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1756 if (file->private_data == NULL) {
1761 open_file = file->private_data;
1762 pTcon = tlink_tcon(open_file->tlink);
1764 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1765 cFYI(1, "attempting read on write only file instance");
1767 for (total_read = 0, current_offset = read_data;
1768 read_size > total_read;
1769 total_read += bytes_read, current_offset += bytes_read) {
1770 current_read_size = min_t(const int, read_size - total_read,
1772 /* For windows me and 9x we do not want to request more
1773 than it negotiated since it will refuse the read then */
1775 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1776 current_read_size = min_t(const int, current_read_size,
1777 pTcon->ses->server->maxBuf - 128);
1780 while (rc == -EAGAIN) {
1781 if (open_file->invalidHandle) {
1782 rc = cifs_reopen_file(open_file, true);
1786 rc = CIFSSMBRead(xid, pTcon,
1788 current_read_size, *poffset,
1789 &bytes_read, ¤t_offset,
1792 if (rc || (bytes_read == 0)) {
1800 cifs_stats_bytes_read(pTcon, total_read);
1801 *poffset += bytes_read;
1808 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1813 rc = cifs_revalidate_file(file);
1815 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1819 rc = generic_file_mmap(file, vma);
1825 static void cifs_copy_cache_pages(struct address_space *mapping,
1826 struct list_head *pages, int bytes_read, char *data)
1831 while (bytes_read > 0) {
1832 if (list_empty(pages))
1835 page = list_entry(pages->prev, struct page, lru);
1836 list_del(&page->lru);
1838 if (add_to_page_cache_lru(page, mapping, page->index,
1840 page_cache_release(page);
1841 cFYI(1, "Add page cache failed");
1842 data += PAGE_CACHE_SIZE;
1843 bytes_read -= PAGE_CACHE_SIZE;
1846 page_cache_release(page);
1848 target = kmap_atomic(page, KM_USER0);
1850 if (PAGE_CACHE_SIZE > bytes_read) {
1851 memcpy(target, data, bytes_read);
1852 /* zero the tail end of this partial page */
1853 memset(target + bytes_read, 0,
1854 PAGE_CACHE_SIZE - bytes_read);
1857 memcpy(target, data, PAGE_CACHE_SIZE);
1858 bytes_read -= PAGE_CACHE_SIZE;
1860 kunmap_atomic(target, KM_USER0);
1862 flush_dcache_page(page);
1863 SetPageUptodate(page);
1865 data += PAGE_CACHE_SIZE;
1867 /* add page to FS-Cache */
1868 cifs_readpage_to_fscache(mapping->host, page);
1873 static int cifs_readpages(struct file *file, struct address_space *mapping,
1874 struct list_head *page_list, unsigned num_pages)
1880 struct cifs_sb_info *cifs_sb;
1881 struct cifsTconInfo *pTcon;
1882 unsigned int bytes_read = 0;
1883 unsigned int read_size, i;
1884 char *smb_read_data = NULL;
1885 struct smb_com_read_rsp *pSMBr;
1886 struct cifsFileInfo *open_file;
1887 int buf_type = CIFS_NO_BUFFER;
1890 if (file->private_data == NULL) {
1895 open_file = file->private_data;
1896 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1897 pTcon = tlink_tcon(open_file->tlink);
1900 * Reads as many pages as possible from fscache. Returns -ENOBUFS
1901 * immediately if the cookie is negative
1903 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
1908 cFYI(DBG2, "rpages: num pages %d", num_pages);
1909 for (i = 0; i < num_pages; ) {
1910 unsigned contig_pages;
1911 struct page *tmp_page;
1912 unsigned long expected_index;
1914 if (list_empty(page_list))
1917 page = list_entry(page_list->prev, struct page, lru);
1918 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1920 /* count adjacent pages that we will read into */
1923 list_entry(page_list->prev, struct page, lru)->index;
1924 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1925 if (tmp_page->index == expected_index) {
1931 if (contig_pages + i > num_pages)
1932 contig_pages = num_pages - i;
1934 /* for reads over a certain size could initiate async
1937 read_size = contig_pages * PAGE_CACHE_SIZE;
1938 /* Read size needs to be in multiples of one page */
1939 read_size = min_t(const unsigned int, read_size,
1940 cifs_sb->rsize & PAGE_CACHE_MASK);
1941 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
1942 read_size, contig_pages);
1944 while (rc == -EAGAIN) {
1945 if (open_file->invalidHandle) {
1946 rc = cifs_reopen_file(open_file, true);
1951 rc = CIFSSMBRead(xid, pTcon,
1954 &bytes_read, &smb_read_data,
1956 /* BB more RC checks ? */
1957 if (rc == -EAGAIN) {
1958 if (smb_read_data) {
1959 if (buf_type == CIFS_SMALL_BUFFER)
1960 cifs_small_buf_release(smb_read_data);
1961 else if (buf_type == CIFS_LARGE_BUFFER)
1962 cifs_buf_release(smb_read_data);
1963 smb_read_data = NULL;
1967 if ((rc < 0) || (smb_read_data == NULL)) {
1968 cFYI(1, "Read error in readpages: %d", rc);
1970 } else if (bytes_read > 0) {
1971 task_io_account_read(bytes_read);
1972 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1973 cifs_copy_cache_pages(mapping, page_list, bytes_read,
1974 smb_read_data + 4 /* RFC1001 hdr */ +
1975 le16_to_cpu(pSMBr->DataOffset));
1977 i += bytes_read >> PAGE_CACHE_SHIFT;
1978 cifs_stats_bytes_read(pTcon, bytes_read);
1979 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1980 i++; /* account for partial page */
1982 /* server copy of file can have smaller size
1984 /* BB do we need to verify this common case ?
1985 this case is ok - if we are at server EOF
1986 we will hit it on next read */
1991 cFYI(1, "No bytes read (%d) at offset %lld . "
1992 "Cleaning remaining pages from readahead list",
1993 bytes_read, offset);
1994 /* BB turn off caching and do new lookup on
1995 file size at server? */
1998 if (smb_read_data) {
1999 if (buf_type == CIFS_SMALL_BUFFER)
2000 cifs_small_buf_release(smb_read_data);
2001 else if (buf_type == CIFS_LARGE_BUFFER)
2002 cifs_buf_release(smb_read_data);
2003 smb_read_data = NULL;
2008 /* need to free smb_read_data buf before exit */
2009 if (smb_read_data) {
2010 if (buf_type == CIFS_SMALL_BUFFER)
2011 cifs_small_buf_release(smb_read_data);
2012 else if (buf_type == CIFS_LARGE_BUFFER)
2013 cifs_buf_release(smb_read_data);
2014 smb_read_data = NULL;
2022 static int cifs_readpage_worker(struct file *file, struct page *page,
2028 /* Is the page cached? */
2029 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2033 page_cache_get(page);
2034 read_data = kmap(page);
2035 /* for reads over a certain size could initiate async read ahead */
2037 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2042 cFYI(1, "Bytes read %d", rc);
2044 file->f_path.dentry->d_inode->i_atime =
2045 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2047 if (PAGE_CACHE_SIZE > rc)
2048 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2050 flush_dcache_page(page);
2051 SetPageUptodate(page);
2053 /* send this page to the cache */
2054 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2060 page_cache_release(page);
2066 static int cifs_readpage(struct file *file, struct page *page)
2068 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2074 if (file->private_data == NULL) {
2080 cFYI(1, "readpage %p at offset %d 0x%x\n",
2081 page, (int)offset, (int)offset);
2083 rc = cifs_readpage_worker(file, page, &offset);
2091 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2093 struct cifsFileInfo *open_file;
2095 spin_lock(&cifs_file_list_lock);
2096 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2097 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2098 spin_unlock(&cifs_file_list_lock);
2102 spin_unlock(&cifs_file_list_lock);
2106 /* We do not want to update the file size from server for inodes
2107 open for write - to avoid races with writepage extending
2108 the file - in the future we could consider allowing
2109 refreshing the inode only on increases in the file size
2110 but this is tricky to do without racing with writebehind
2111 page caching in the current Linux kernel design */
2112 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2117 if (is_inode_writable(cifsInode)) {
2118 /* This inode is open for write at least once */
2119 struct cifs_sb_info *cifs_sb;
2121 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2122 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2123 /* since no page cache to corrupt on directio
2124 we can change size safely */
2128 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2136 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2137 loff_t pos, unsigned len, unsigned flags,
2138 struct page **pagep, void **fsdata)
2140 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2141 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2142 loff_t page_start = pos & PAGE_MASK;
2147 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2149 page = grab_cache_page_write_begin(mapping, index, flags);
2155 if (PageUptodate(page))
2159 * If we write a full page it will be up to date, no need to read from
2160 * the server. If the write is short, we'll end up doing a sync write
2163 if (len == PAGE_CACHE_SIZE)
2167 * optimize away the read when we have an oplock, and we're not
2168 * expecting to use any of the data we'd be reading in. That
2169 * is, when the page lies beyond the EOF, or straddles the EOF
2170 * and the write will cover all of the existing data.
2172 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2173 i_size = i_size_read(mapping->host);
2174 if (page_start >= i_size ||
2175 (offset == 0 && (pos + len) >= i_size)) {
2176 zero_user_segments(page, 0, offset,
2180 * PageChecked means that the parts of the page
2181 * to which we're not writing are considered up
2182 * to date. Once the data is copied to the
2183 * page, it can be set uptodate.
2185 SetPageChecked(page);
2190 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2192 * might as well read a page, it is fast enough. If we get
2193 * an error, we don't need to return it. cifs_write_end will
2194 * do a sync write instead since PG_uptodate isn't set.
2196 cifs_readpage_worker(file, page, &page_start);
2198 /* we could try using another file handle if there is one -
2199 but how would we lock it to prevent close of that handle
2200 racing with this read? In any case
2201 this will be written out by write_end so is fine */
2208 static int cifs_release_page(struct page *page, gfp_t gfp)
2210 if (PagePrivate(page))
2213 return cifs_fscache_release_page(page, gfp);
2216 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2218 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2221 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2224 void cifs_oplock_break(struct work_struct *work)
2226 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2228 struct inode *inode = cfile->dentry->d_inode;
2229 struct cifsInodeInfo *cinode = CIFS_I(inode);
2232 if (inode && S_ISREG(inode->i_mode)) {
2233 if (cinode->clientCanCacheRead)
2234 break_lease(inode, O_RDONLY);
2236 break_lease(inode, O_WRONLY);
2237 rc = filemap_fdatawrite(inode->i_mapping);
2238 if (cinode->clientCanCacheRead == 0) {
2239 rc = filemap_fdatawait(inode->i_mapping);
2240 mapping_set_error(inode->i_mapping, rc);
2241 invalidate_remote_inode(inode);
2243 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2247 * releasing stale oplock after recent reconnect of smb session using
2248 * a now incorrect file handle is not a data integrity issue but do
2249 * not bother sending an oplock release if session to server still is
2250 * disconnected since oplock already released by the server
2252 if (!cfile->oplock_break_cancelled) {
2253 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2254 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false);
2255 cFYI(1, "Oplock release rc = %d", rc);
2259 * We might have kicked in before is_valid_oplock_break()
2260 * finished grabbing reference for us. Make sure it's done by
2261 * waiting for cifs_file_list_lock.
2263 spin_lock(&cifs_file_list_lock);
2264 spin_unlock(&cifs_file_list_lock);
2266 cifs_oplock_break_put(cfile);
2269 /* must be called while holding cifs_file_list_lock */
2270 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2272 cifs_sb_active(cfile->dentry->d_sb);
2273 cifsFileInfo_get(cfile);
2276 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2278 cifsFileInfo_put(cfile);
2279 cifs_sb_deactive(cfile->dentry->d_sb);
2282 const struct address_space_operations cifs_addr_ops = {
2283 .readpage = cifs_readpage,
2284 .readpages = cifs_readpages,
2285 .writepage = cifs_writepage,
2286 .writepages = cifs_writepages,
2287 .write_begin = cifs_write_begin,
2288 .write_end = cifs_write_end,
2289 .set_page_dirty = __set_page_dirty_nobuffers,
2290 .releasepage = cifs_release_page,
2291 .invalidatepage = cifs_invalidate_page,
2292 /* .sync_page = cifs_sync_page, */
2297 * cifs_readpages requires the server to support a buffer large enough to
2298 * contain the header plus one complete page of data. Otherwise, we need
2299 * to leave cifs_readpages out of the address space operations.
2301 const struct address_space_operations cifs_addr_ops_smallbuf = {
2302 .readpage = cifs_readpage,
2303 .writepage = cifs_writepage,
2304 .writepages = cifs_writepages,
2305 .write_begin = cifs_write_begin,
2306 .write_end = cifs_write_end,
2307 .set_page_dirty = __set_page_dirty_nobuffers,
2308 .releasepage = cifs_release_page,
2309 .invalidatepage = cifs_invalidate_page,
2310 /* .sync_page = cifs_sync_page, */