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 <linux/swap.h>
36 #include <asm/div64.h>
40 #include "cifsproto.h"
41 #include "cifs_unicode.h"
42 #include "cifs_debug.h"
43 #include "cifs_fs_sb.h"
46 static inline int cifs_convert_flags(unsigned int flags)
48 if ((flags & O_ACCMODE) == O_RDONLY)
50 else if ((flags & O_ACCMODE) == O_WRONLY)
52 else if ((flags & O_ACCMODE) == O_RDWR) {
53 /* GENERIC_ALL is too much permission to request
54 can cause unnecessary access denied on create */
55 /* return GENERIC_ALL; */
56 return (GENERIC_READ | GENERIC_WRITE);
59 return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
60 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
64 static u32 cifs_posix_convert_flags(unsigned int flags)
68 if ((flags & O_ACCMODE) == O_RDONLY)
69 posix_flags = SMB_O_RDONLY;
70 else if ((flags & O_ACCMODE) == O_WRONLY)
71 posix_flags = SMB_O_WRONLY;
72 else if ((flags & O_ACCMODE) == O_RDWR)
73 posix_flags = SMB_O_RDWR;
76 posix_flags |= SMB_O_CREAT;
78 posix_flags |= SMB_O_EXCL;
80 posix_flags |= SMB_O_TRUNC;
81 /* be safe and imply O_SYNC for O_DSYNC */
83 posix_flags |= SMB_O_SYNC;
84 if (flags & O_DIRECTORY)
85 posix_flags |= SMB_O_DIRECTORY;
86 if (flags & O_NOFOLLOW)
87 posix_flags |= SMB_O_NOFOLLOW;
89 posix_flags |= SMB_O_DIRECT;
94 static inline int cifs_get_disposition(unsigned int flags)
96 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
98 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
99 return FILE_OVERWRITE_IF;
100 else if ((flags & O_CREAT) == O_CREAT)
102 else if ((flags & O_TRUNC) == O_TRUNC)
103 return FILE_OVERWRITE;
108 int cifs_posix_open(char *full_path, struct inode **pinode,
109 struct super_block *sb, int mode, unsigned int f_flags,
110 __u32 *poplock, __u16 *pnetfid, unsigned int xid)
113 FILE_UNIX_BASIC_INFO *presp_data;
114 __u32 posix_flags = 0;
115 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
116 struct cifs_fattr fattr;
117 struct tcon_link *tlink;
118 struct cifs_tcon *tcon;
120 cFYI(1, "posix open %s", full_path);
122 presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
123 if (presp_data == NULL)
126 tlink = cifs_sb_tlink(cifs_sb);
132 tcon = tlink_tcon(tlink);
133 mode &= ~current_umask();
135 posix_flags = cifs_posix_convert_flags(f_flags);
136 rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
137 poplock, full_path, cifs_sb->local_nls,
138 cifs_sb->mnt_cifs_flags &
139 CIFS_MOUNT_MAP_SPECIAL_CHR);
140 cifs_put_tlink(tlink);
145 if (presp_data->Type == cpu_to_le32(-1))
146 goto posix_open_ret; /* open ok, caller does qpathinfo */
149 goto posix_open_ret; /* caller does not need info */
151 cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
153 /* get new inode and set it up */
154 if (*pinode == NULL) {
155 cifs_fill_uniqueid(sb, &fattr);
156 *pinode = cifs_iget(sb, &fattr);
162 cifs_fattr_to_inode(*pinode, &fattr);
171 cifs_nt_open(char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
172 struct cifs_tcon *tcon, unsigned int f_flags, __u32 *oplock,
173 struct cifs_fid *fid, unsigned int xid)
178 int create_options = CREATE_NOT_DIR;
181 if (!tcon->ses->server->ops->open)
184 desired_access = cifs_convert_flags(f_flags);
186 /*********************************************************************
187 * open flag mapping table:
189 * POSIX Flag CIFS Disposition
190 * ---------- ----------------
191 * O_CREAT FILE_OPEN_IF
192 * O_CREAT | O_EXCL FILE_CREATE
193 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
194 * O_TRUNC FILE_OVERWRITE
195 * none of the above FILE_OPEN
197 * Note that there is not a direct match between disposition
198 * FILE_SUPERSEDE (ie create whether or not file exists although
199 * O_CREAT | O_TRUNC is similar but truncates the existing
200 * file rather than creating a new file as FILE_SUPERSEDE does
201 * (which uses the attributes / metadata passed in on open call)
203 *? O_SYNC is a reasonable match to CIFS writethrough flag
204 *? and the read write flags match reasonably. O_LARGEFILE
205 *? is irrelevant because largefile support is always used
206 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
207 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
208 *********************************************************************/
210 disposition = cifs_get_disposition(f_flags);
212 /* BB pass O_SYNC flag through on file attributes .. BB */
214 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
218 if (backup_cred(cifs_sb))
219 create_options |= CREATE_OPEN_BACKUP_INTENT;
221 rc = tcon->ses->server->ops->open(xid, tcon, full_path, disposition,
222 desired_access, create_options, fid,
223 oplock, buf, cifs_sb);
229 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
232 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
240 struct cifsFileInfo *
241 cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
242 struct tcon_link *tlink, __u32 oplock)
244 struct dentry *dentry = file->f_path.dentry;
245 struct inode *inode = dentry->d_inode;
246 struct cifsInodeInfo *cinode = CIFS_I(inode);
247 struct cifsFileInfo *cfile;
249 cfile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
254 cfile->pid = current->tgid;
255 cfile->uid = current_fsuid();
256 cfile->dentry = dget(dentry);
257 cfile->f_flags = file->f_flags;
258 cfile->invalidHandle = false;
259 cfile->tlink = cifs_get_tlink(tlink);
260 mutex_init(&cfile->fh_mutex);
261 INIT_WORK(&cfile->oplock_break, cifs_oplock_break);
262 INIT_LIST_HEAD(&cfile->llist);
263 tlink_tcon(tlink)->ses->server->ops->set_fid(cfile, fid, oplock);
265 spin_lock(&cifs_file_list_lock);
266 list_add(&cfile->tlist, &(tlink_tcon(tlink)->openFileList));
267 /* if readable file instance put first in list*/
268 if (file->f_mode & FMODE_READ)
269 list_add(&cfile->flist, &cinode->openFileList);
271 list_add_tail(&cfile->flist, &cinode->openFileList);
272 spin_unlock(&cifs_file_list_lock);
274 file->private_data = cfile;
278 static void cifs_del_lock_waiters(struct cifsLockInfo *lock);
280 struct cifsFileInfo *
281 cifsFileInfo_get(struct cifsFileInfo *cifs_file)
283 spin_lock(&cifs_file_list_lock);
284 cifsFileInfo_get_locked(cifs_file);
285 spin_unlock(&cifs_file_list_lock);
290 * Release a reference on the file private data. This may involve closing
291 * the filehandle out on the server. Must be called without holding
292 * cifs_file_list_lock.
294 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
296 struct inode *inode = cifs_file->dentry->d_inode;
297 struct cifs_tcon *tcon = tlink_tcon(cifs_file->tlink);
298 struct cifsInodeInfo *cifsi = CIFS_I(inode);
299 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
300 struct cifsLockInfo *li, *tmp;
302 spin_lock(&cifs_file_list_lock);
303 if (--cifs_file->count > 0) {
304 spin_unlock(&cifs_file_list_lock);
308 /* remove it from the lists */
309 list_del(&cifs_file->flist);
310 list_del(&cifs_file->tlist);
312 if (list_empty(&cifsi->openFileList)) {
313 cFYI(1, "closing last open instance for inode %p",
314 cifs_file->dentry->d_inode);
316 /* in strict cache mode we need invalidate mapping on the last
317 close because it may cause a error when we open this file
318 again and get at least level II oplock */
319 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
320 CIFS_I(inode)->invalid_mapping = true;
322 cifs_set_oplock_level(cifsi, 0);
324 spin_unlock(&cifs_file_list_lock);
326 cancel_work_sync(&cifs_file->oplock_break);
328 if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
332 rc = CIFSSMBClose(xid, tcon, cifs_file->fid.netfid);
336 /* Delete any outstanding lock records. We'll lose them when the file
339 mutex_lock(&cifsi->lock_mutex);
340 list_for_each_entry_safe(li, tmp, &cifs_file->llist, llist) {
341 list_del(&li->llist);
342 cifs_del_lock_waiters(li);
345 mutex_unlock(&cifsi->lock_mutex);
347 cifs_put_tlink(cifs_file->tlink);
348 dput(cifs_file->dentry);
352 int cifs_open(struct inode *inode, struct file *file)
357 struct cifs_sb_info *cifs_sb;
358 struct cifs_tcon *tcon;
359 struct tcon_link *tlink;
360 struct cifsFileInfo *cfile = NULL;
361 char *full_path = NULL;
362 bool posix_open_ok = false;
367 cifs_sb = CIFS_SB(inode->i_sb);
368 tlink = cifs_sb_tlink(cifs_sb);
371 return PTR_ERR(tlink);
373 tcon = tlink_tcon(tlink);
375 full_path = build_path_from_dentry(file->f_path.dentry);
376 if (full_path == NULL) {
381 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
382 inode, file->f_flags, full_path);
384 if (tcon->ses->server->oplocks)
389 if (!tcon->broken_posix_open && tcon->unix_ext &&
390 cap_unix(tcon->ses) && (CIFS_UNIX_POSIX_PATH_OPS_CAP &
391 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
392 /* can not refresh inode info since size could be stale */
393 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
394 cifs_sb->mnt_file_mode /* ignored */,
395 file->f_flags, &oplock, &fid.netfid, xid);
397 cFYI(1, "posix open succeeded");
398 posix_open_ok = true;
399 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
400 if (tcon->ses->serverNOS)
401 cERROR(1, "server %s of type %s returned"
402 " unexpected error on SMB posix open"
403 ", disabling posix open support."
404 " Check if server update available.",
405 tcon->ses->serverName,
406 tcon->ses->serverNOS);
407 tcon->broken_posix_open = true;
408 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
409 (rc != -EOPNOTSUPP)) /* path not found or net err */
412 * Else fallthrough to retry open the old way on network i/o
417 if (!posix_open_ok) {
418 rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
419 file->f_flags, &oplock, &fid, xid);
424 cfile = cifs_new_fileinfo(&fid, file, tlink, oplock);
426 CIFSSMBClose(xid, tcon, fid.netfid);
431 cifs_fscache_set_inode_cookie(inode, file);
433 if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
435 * Time to set mode which we can not set earlier due to
436 * problems creating new read-only files.
438 struct cifs_unix_set_info_args args = {
439 .mode = inode->i_mode,
442 .ctime = NO_CHANGE_64,
443 .atime = NO_CHANGE_64,
444 .mtime = NO_CHANGE_64,
447 CIFSSMBUnixSetFileInfo(xid, tcon, &args, fid.netfid,
454 cifs_put_tlink(tlink);
458 /* Try to reacquire byte range locks that were released when session */
459 /* to server was lost */
460 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
464 /* BB list all locks open on this file and relock */
469 static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
474 struct cifs_sb_info *cifs_sb;
475 struct cifs_tcon *tcon;
476 struct cifsInodeInfo *pCifsInode;
478 char *full_path = NULL;
480 int disposition = FILE_OPEN;
481 int create_options = CREATE_NOT_DIR;
485 mutex_lock(&pCifsFile->fh_mutex);
486 if (!pCifsFile->invalidHandle) {
487 mutex_unlock(&pCifsFile->fh_mutex);
493 inode = pCifsFile->dentry->d_inode;
494 cifs_sb = CIFS_SB(inode->i_sb);
495 tcon = tlink_tcon(pCifsFile->tlink);
497 /* can not grab rename sem here because various ops, including
498 those that already have the rename sem can end up causing writepage
499 to get called and if the server was down that means we end up here,
500 and we can never tell if the caller already has the rename_sem */
501 full_path = build_path_from_dentry(pCifsFile->dentry);
502 if (full_path == NULL) {
504 mutex_unlock(&pCifsFile->fh_mutex);
509 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
510 inode, pCifsFile->f_flags, full_path);
512 if (tcon->ses->server->oplocks)
517 if (tcon->unix_ext && cap_unix(tcon->ses) &&
518 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
519 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
521 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
522 * original open. Must mask them off for a reopen.
524 unsigned int oflags = pCifsFile->f_flags &
525 ~(O_CREAT | O_EXCL | O_TRUNC);
527 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
528 cifs_sb->mnt_file_mode /* ignored */,
529 oflags, &oplock, &netfid, xid);
531 cFYI(1, "posix reopen succeeded");
534 /* fallthrough to retry open the old way on errors, especially
535 in the reconnect path it is important to retry hard */
538 desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
540 if (backup_cred(cifs_sb))
541 create_options |= CREATE_OPEN_BACKUP_INTENT;
543 /* Can not refresh inode by passing in file_info buf to be returned
544 by SMBOpen and then calling get_inode_info with returned buf
545 since file might have write behind data that needs to be flushed
546 and server version of file size can be stale. If we knew for sure
547 that inode was not dirty locally we could do this */
549 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
550 create_options, &netfid, &oplock, NULL,
551 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
552 CIFS_MOUNT_MAP_SPECIAL_CHR);
554 mutex_unlock(&pCifsFile->fh_mutex);
555 cFYI(1, "cifs_open returned 0x%x", rc);
556 cFYI(1, "oplock: %d", oplock);
557 goto reopen_error_exit;
561 pCifsFile->fid.netfid = netfid;
562 pCifsFile->invalidHandle = false;
563 mutex_unlock(&pCifsFile->fh_mutex);
564 pCifsInode = CIFS_I(inode);
567 rc = filemap_write_and_wait(inode->i_mapping);
568 mapping_set_error(inode->i_mapping, rc);
571 rc = cifs_get_inode_info_unix(&inode,
572 full_path, inode->i_sb, xid);
574 rc = cifs_get_inode_info(&inode,
575 full_path, NULL, inode->i_sb,
577 } /* else we are writing out data to server already
578 and could deadlock if we tried to flush data, and
579 since we do not know if we have data that would
580 invalidate the current end of file on the server
581 we can not go to the server to get the new inod
584 cifs_set_oplock_level(pCifsInode, oplock);
586 cifs_relock_file(pCifsFile);
594 int cifs_close(struct inode *inode, struct file *file)
596 if (file->private_data != NULL) {
597 cifsFileInfo_put(file->private_data);
598 file->private_data = NULL;
601 /* return code from the ->release op is always ignored */
605 int cifs_closedir(struct inode *inode, struct file *file)
609 struct cifsFileInfo *cfile = file->private_data;
612 cFYI(1, "Closedir inode = 0x%p", inode);
617 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
619 cFYI(1, "Freeing private data in close dir");
620 spin_lock(&cifs_file_list_lock);
621 if (!cfile->srch_inf.endOfSearch && !cfile->invalidHandle) {
622 cfile->invalidHandle = true;
623 spin_unlock(&cifs_file_list_lock);
624 rc = CIFSFindClose(xid, tcon, cfile->fid.netfid);
625 cFYI(1, "Closing uncompleted readdir with rc %d", rc);
626 /* not much we can do if it fails anyway, ignore rc */
629 spin_unlock(&cifs_file_list_lock);
630 tmp = cfile->srch_inf.ntwrk_buf_start;
632 cFYI(1, "closedir free smb buf in srch struct");
633 cfile->srch_inf.ntwrk_buf_start = NULL;
634 if (cfile->srch_inf.smallBuf)
635 cifs_small_buf_release(tmp);
637 cifs_buf_release(tmp);
639 cifs_put_tlink(cfile->tlink);
640 kfree(file->private_data);
641 file->private_data = NULL;
643 /* BB can we lock the filestruct while this is going on? */
648 static struct cifsLockInfo *
649 cifs_lock_init(__u64 offset, __u64 length, __u8 type)
651 struct cifsLockInfo *lock =
652 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
655 lock->offset = offset;
656 lock->length = length;
658 lock->pid = current->tgid;
659 INIT_LIST_HEAD(&lock->blist);
660 init_waitqueue_head(&lock->block_q);
665 cifs_del_lock_waiters(struct cifsLockInfo *lock)
667 struct cifsLockInfo *li, *tmp;
668 list_for_each_entry_safe(li, tmp, &lock->blist, blist) {
669 list_del_init(&li->blist);
670 wake_up(&li->block_q);
675 cifs_find_fid_lock_conflict(struct cifsFileInfo *cfile, __u64 offset,
676 __u64 length, __u8 type, struct cifsFileInfo *cur,
677 struct cifsLockInfo **conf_lock)
679 struct cifsLockInfo *li;
680 struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
682 list_for_each_entry(li, &cfile->llist, llist) {
683 if (offset + length <= li->offset ||
684 offset >= li->offset + li->length)
686 else if ((type & server->vals->shared_lock_type) &&
687 ((server->ops->compare_fids(cur, cfile) &&
688 current->tgid == li->pid) || type == li->type))
699 cifs_find_lock_conflict(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
700 __u8 type, struct cifsLockInfo **conf_lock)
703 struct cifsFileInfo *fid, *tmp;
704 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
706 spin_lock(&cifs_file_list_lock);
707 list_for_each_entry_safe(fid, tmp, &cinode->openFileList, flist) {
708 rc = cifs_find_fid_lock_conflict(fid, offset, length, type,
713 spin_unlock(&cifs_file_list_lock);
719 * Check if there is another lock that prevents us to set the lock (mandatory
720 * style). If such a lock exists, update the flock structure with its
721 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
722 * or leave it the same if we can't. Returns 0 if we don't need to request to
723 * the server or 1 otherwise.
726 cifs_lock_test(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
727 __u8 type, struct file_lock *flock)
730 struct cifsLockInfo *conf_lock;
731 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
732 struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
735 mutex_lock(&cinode->lock_mutex);
737 exist = cifs_find_lock_conflict(cfile, offset, length, type,
740 flock->fl_start = conf_lock->offset;
741 flock->fl_end = conf_lock->offset + conf_lock->length - 1;
742 flock->fl_pid = conf_lock->pid;
743 if (conf_lock->type & server->vals->shared_lock_type)
744 flock->fl_type = F_RDLCK;
746 flock->fl_type = F_WRLCK;
747 } else if (!cinode->can_cache_brlcks)
750 flock->fl_type = F_UNLCK;
752 mutex_unlock(&cinode->lock_mutex);
757 cifs_lock_add(struct cifsFileInfo *cfile, struct cifsLockInfo *lock)
759 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
760 mutex_lock(&cinode->lock_mutex);
761 list_add_tail(&lock->llist, &cfile->llist);
762 mutex_unlock(&cinode->lock_mutex);
766 * Set the byte-range lock (mandatory style). Returns:
767 * 1) 0, if we set the lock and don't need to request to the server;
768 * 2) 1, if no locks prevent us but we need to request to the server;
769 * 3) -EACCESS, if there is a lock that prevents us and wait is false.
772 cifs_lock_add_if(struct cifsFileInfo *cfile, struct cifsLockInfo *lock,
775 struct cifsLockInfo *conf_lock;
776 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
782 mutex_lock(&cinode->lock_mutex);
784 exist = cifs_find_lock_conflict(cfile, lock->offset, lock->length,
785 lock->type, &conf_lock);
786 if (!exist && cinode->can_cache_brlcks) {
787 list_add_tail(&lock->llist, &cfile->llist);
788 mutex_unlock(&cinode->lock_mutex);
797 list_add_tail(&lock->blist, &conf_lock->blist);
798 mutex_unlock(&cinode->lock_mutex);
799 rc = wait_event_interruptible(lock->block_q,
800 (lock->blist.prev == &lock->blist) &&
801 (lock->blist.next == &lock->blist));
804 mutex_lock(&cinode->lock_mutex);
805 list_del_init(&lock->blist);
808 mutex_unlock(&cinode->lock_mutex);
813 * Check if there is another lock that prevents us to set the lock (posix
814 * style). If such a lock exists, update the flock structure with its
815 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
816 * or leave it the same if we can't. Returns 0 if we don't need to request to
817 * the server or 1 otherwise.
820 cifs_posix_lock_test(struct file *file, struct file_lock *flock)
823 struct cifsInodeInfo *cinode = CIFS_I(file->f_path.dentry->d_inode);
824 unsigned char saved_type = flock->fl_type;
826 if ((flock->fl_flags & FL_POSIX) == 0)
829 mutex_lock(&cinode->lock_mutex);
830 posix_test_lock(file, flock);
832 if (flock->fl_type == F_UNLCK && !cinode->can_cache_brlcks) {
833 flock->fl_type = saved_type;
837 mutex_unlock(&cinode->lock_mutex);
842 * Set the byte-range lock (posix style). Returns:
843 * 1) 0, if we set the lock and don't need to request to the server;
844 * 2) 1, if we need to request to the server;
845 * 3) <0, if the error occurs while setting the lock.
848 cifs_posix_lock_set(struct file *file, struct file_lock *flock)
850 struct cifsInodeInfo *cinode = CIFS_I(file->f_path.dentry->d_inode);
853 if ((flock->fl_flags & FL_POSIX) == 0)
857 mutex_lock(&cinode->lock_mutex);
858 if (!cinode->can_cache_brlcks) {
859 mutex_unlock(&cinode->lock_mutex);
863 rc = posix_lock_file(file, flock, NULL);
864 mutex_unlock(&cinode->lock_mutex);
865 if (rc == FILE_LOCK_DEFERRED) {
866 rc = wait_event_interruptible(flock->fl_wait, !flock->fl_next);
869 locks_delete_block(flock);
875 cifs_push_mandatory_locks(struct cifsFileInfo *cfile)
878 int rc = 0, stored_rc;
879 struct cifsLockInfo *li, *tmp;
880 struct cifs_tcon *tcon;
881 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
882 unsigned int num, max_num, max_buf;
883 LOCKING_ANDX_RANGE *buf, *cur;
884 int types[] = {LOCKING_ANDX_LARGE_FILES,
885 LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES};
889 tcon = tlink_tcon(cfile->tlink);
891 mutex_lock(&cinode->lock_mutex);
892 if (!cinode->can_cache_brlcks) {
893 mutex_unlock(&cinode->lock_mutex);
899 * Accessing maxBuf is racy with cifs_reconnect - need to store value
900 * and check it for zero before using.
902 max_buf = tcon->ses->server->maxBuf;
904 mutex_unlock(&cinode->lock_mutex);
909 max_num = (max_buf - sizeof(struct smb_hdr)) /
910 sizeof(LOCKING_ANDX_RANGE);
911 buf = kzalloc(max_num * sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
913 mutex_unlock(&cinode->lock_mutex);
918 for (i = 0; i < 2; i++) {
921 list_for_each_entry_safe(li, tmp, &cfile->llist, llist) {
922 if (li->type != types[i])
924 cur->Pid = cpu_to_le16(li->pid);
925 cur->LengthLow = cpu_to_le32((u32)li->length);
926 cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
927 cur->OffsetLow = cpu_to_le32((u32)li->offset);
928 cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
929 if (++num == max_num) {
930 stored_rc = cifs_lockv(xid, tcon,
932 (__u8)li->type, 0, num,
943 stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
944 (__u8)types[i], 0, num, buf);
950 cinode->can_cache_brlcks = false;
951 mutex_unlock(&cinode->lock_mutex);
958 /* copied from fs/locks.c with a name change */
959 #define cifs_for_each_lock(inode, lockp) \
960 for (lockp = &inode->i_flock; *lockp != NULL; \
961 lockp = &(*lockp)->fl_next)
963 struct lock_to_push {
964 struct list_head llist;
973 cifs_push_posix_locks(struct cifsFileInfo *cfile)
975 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
976 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
977 struct file_lock *flock, **before;
978 unsigned int count = 0, i = 0;
979 int rc = 0, xid, type;
980 struct list_head locks_to_send, *el;
981 struct lock_to_push *lck, *tmp;
986 mutex_lock(&cinode->lock_mutex);
987 if (!cinode->can_cache_brlcks) {
988 mutex_unlock(&cinode->lock_mutex);
994 cifs_for_each_lock(cfile->dentry->d_inode, before) {
995 if ((*before)->fl_flags & FL_POSIX)
1000 INIT_LIST_HEAD(&locks_to_send);
1003 * Allocating count locks is enough because no FL_POSIX locks can be
1004 * added to the list while we are holding cinode->lock_mutex that
1005 * protects locking operations of this inode.
1007 for (; i < count; i++) {
1008 lck = kmalloc(sizeof(struct lock_to_push), GFP_KERNEL);
1013 list_add_tail(&lck->llist, &locks_to_send);
1016 el = locks_to_send.next;
1018 cifs_for_each_lock(cfile->dentry->d_inode, before) {
1020 if ((flock->fl_flags & FL_POSIX) == 0)
1022 if (el == &locks_to_send) {
1024 * The list ended. We don't have enough allocated
1025 * structures - something is really wrong.
1027 cERROR(1, "Can't push all brlocks!");
1030 length = 1 + flock->fl_end - flock->fl_start;
1031 if (flock->fl_type == F_RDLCK || flock->fl_type == F_SHLCK)
1035 lck = list_entry(el, struct lock_to_push, llist);
1036 lck->pid = flock->fl_pid;
1037 lck->netfid = cfile->fid.netfid;
1038 lck->length = length;
1040 lck->offset = flock->fl_start;
1045 list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1048 stored_rc = CIFSSMBPosixLock(xid, tcon, lck->netfid, lck->pid,
1049 lck->offset, lck->length, NULL,
1053 list_del(&lck->llist);
1058 cinode->can_cache_brlcks = false;
1059 mutex_unlock(&cinode->lock_mutex);
1064 list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1065 list_del(&lck->llist);
1072 cifs_push_locks(struct cifsFileInfo *cfile)
1074 struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
1075 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1077 if (cap_unix(tcon->ses) &&
1078 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1079 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1080 return cifs_push_posix_locks(cfile);
1082 return cifs_push_mandatory_locks(cfile);
1086 cifs_read_flock(struct file_lock *flock, __u32 *type, int *lock, int *unlock,
1087 bool *wait_flag, struct TCP_Server_Info *server)
1089 if (flock->fl_flags & FL_POSIX)
1091 if (flock->fl_flags & FL_FLOCK)
1093 if (flock->fl_flags & FL_SLEEP) {
1094 cFYI(1, "Blocking lock");
1097 if (flock->fl_flags & FL_ACCESS)
1098 cFYI(1, "Process suspended by mandatory locking - "
1099 "not implemented yet");
1100 if (flock->fl_flags & FL_LEASE)
1101 cFYI(1, "Lease on file - not implemented yet");
1102 if (flock->fl_flags &
1103 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
1104 cFYI(1, "Unknown lock flags 0x%x", flock->fl_flags);
1106 *type = server->vals->large_lock_type;
1107 if (flock->fl_type == F_WRLCK) {
1108 cFYI(1, "F_WRLCK ");
1109 *type |= server->vals->exclusive_lock_type;
1111 } else if (flock->fl_type == F_UNLCK) {
1113 *type |= server->vals->unlock_lock_type;
1115 /* Check if unlock includes more than one lock range */
1116 } else if (flock->fl_type == F_RDLCK) {
1118 *type |= server->vals->shared_lock_type;
1120 } else if (flock->fl_type == F_EXLCK) {
1122 *type |= server->vals->exclusive_lock_type;
1124 } else if (flock->fl_type == F_SHLCK) {
1126 *type |= server->vals->shared_lock_type;
1129 cFYI(1, "Unknown type of lock");
1133 cifs_mandatory_lock(unsigned int xid, struct cifsFileInfo *cfile, __u64 offset,
1134 __u64 length, __u32 type, int lock, int unlock, bool wait)
1136 return CIFSSMBLock(xid, tlink_tcon(cfile->tlink), cfile->fid.netfid,
1137 current->tgid, length, offset, unlock, lock,
1138 (__u8)type, wait, 0);
1142 cifs_getlk(struct file *file, struct file_lock *flock, __u32 type,
1143 bool wait_flag, bool posix_lck, unsigned int xid)
1146 __u64 length = 1 + flock->fl_end - flock->fl_start;
1147 struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1148 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1149 struct TCP_Server_Info *server = tcon->ses->server;
1150 __u16 netfid = cfile->fid.netfid;
1153 int posix_lock_type;
1155 rc = cifs_posix_lock_test(file, flock);
1159 if (type & server->vals->shared_lock_type)
1160 posix_lock_type = CIFS_RDLCK;
1162 posix_lock_type = CIFS_WRLCK;
1163 rc = CIFSSMBPosixLock(xid, tcon, netfid, current->tgid,
1164 flock->fl_start, length, flock,
1165 posix_lock_type, wait_flag);
1169 rc = cifs_lock_test(cfile, flock->fl_start, length, type, flock);
1173 /* BB we could chain these into one lock request BB */
1174 rc = cifs_mandatory_lock(xid, cfile, flock->fl_start, length, type,
1177 rc = cifs_mandatory_lock(xid, cfile, flock->fl_start, length,
1179 flock->fl_type = F_UNLCK;
1181 cERROR(1, "Error unlocking previously locked "
1182 "range %d during test of lock", rc);
1186 if (type & server->vals->shared_lock_type) {
1187 flock->fl_type = F_WRLCK;
1191 rc = cifs_mandatory_lock(xid, cfile, flock->fl_start, length,
1192 type | server->vals->shared_lock_type, 1, 0,
1195 rc = cifs_mandatory_lock(xid, cfile, flock->fl_start, length,
1196 type | server->vals->shared_lock_type,
1198 flock->fl_type = F_RDLCK;
1200 cERROR(1, "Error unlocking previously locked "
1201 "range %d during test of lock", rc);
1203 flock->fl_type = F_WRLCK;
1209 cifs_move_llist(struct list_head *source, struct list_head *dest)
1211 struct list_head *li, *tmp;
1212 list_for_each_safe(li, tmp, source)
1213 list_move(li, dest);
1217 cifs_free_llist(struct list_head *llist)
1219 struct cifsLockInfo *li, *tmp;
1220 list_for_each_entry_safe(li, tmp, llist, llist) {
1221 cifs_del_lock_waiters(li);
1222 list_del(&li->llist);
1228 cifs_unlock_range(struct cifsFileInfo *cfile, struct file_lock *flock,
1231 int rc = 0, stored_rc;
1232 int types[] = {LOCKING_ANDX_LARGE_FILES,
1233 LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES};
1235 unsigned int max_num, num, max_buf;
1236 LOCKING_ANDX_RANGE *buf, *cur;
1237 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1238 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
1239 struct cifsLockInfo *li, *tmp;
1240 __u64 length = 1 + flock->fl_end - flock->fl_start;
1241 struct list_head tmp_llist;
1243 INIT_LIST_HEAD(&tmp_llist);
1246 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1247 * and check it for zero before using.
1249 max_buf = tcon->ses->server->maxBuf;
1253 max_num = (max_buf - sizeof(struct smb_hdr)) /
1254 sizeof(LOCKING_ANDX_RANGE);
1255 buf = kzalloc(max_num * sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1259 mutex_lock(&cinode->lock_mutex);
1260 for (i = 0; i < 2; i++) {
1263 list_for_each_entry_safe(li, tmp, &cfile->llist, llist) {
1264 if (flock->fl_start > li->offset ||
1265 (flock->fl_start + length) <
1266 (li->offset + li->length))
1268 if (current->tgid != li->pid)
1270 if (types[i] != li->type)
1272 if (cinode->can_cache_brlcks) {
1274 * We can cache brlock requests - simply remove
1275 * a lock from the file's list.
1277 list_del(&li->llist);
1278 cifs_del_lock_waiters(li);
1282 cur->Pid = cpu_to_le16(li->pid);
1283 cur->LengthLow = cpu_to_le32((u32)li->length);
1284 cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1285 cur->OffsetLow = cpu_to_le32((u32)li->offset);
1286 cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1288 * We need to save a lock here to let us add it again to
1289 * the file's list if the unlock range request fails on
1292 list_move(&li->llist, &tmp_llist);
1293 if (++num == max_num) {
1294 stored_rc = cifs_lockv(xid, tcon,
1296 li->type, num, 0, buf);
1299 * We failed on the unlock range
1300 * request - add all locks from the tmp
1301 * list to the head of the file's list.
1303 cifs_move_llist(&tmp_llist,
1308 * The unlock range request succeed -
1309 * free the tmp list.
1311 cifs_free_llist(&tmp_llist);
1318 stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1319 types[i], num, 0, buf);
1321 cifs_move_llist(&tmp_llist, &cfile->llist);
1324 cifs_free_llist(&tmp_llist);
1328 mutex_unlock(&cinode->lock_mutex);
1334 cifs_setlk(struct file *file, struct file_lock *flock, __u32 type,
1335 bool wait_flag, bool posix_lck, int lock, int unlock,
1339 __u64 length = 1 + flock->fl_end - flock->fl_start;
1340 struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1341 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1342 struct TCP_Server_Info *server = tcon->ses->server;
1343 __u16 netfid = cfile->fid.netfid;
1346 int posix_lock_type;
1348 rc = cifs_posix_lock_set(file, flock);
1352 if (type & server->vals->shared_lock_type)
1353 posix_lock_type = CIFS_RDLCK;
1355 posix_lock_type = CIFS_WRLCK;
1358 posix_lock_type = CIFS_UNLCK;
1360 rc = CIFSSMBPosixLock(xid, tcon, netfid, current->tgid,
1361 flock->fl_start, length, NULL,
1362 posix_lock_type, wait_flag);
1367 struct cifsLockInfo *lock;
1369 lock = cifs_lock_init(flock->fl_start, length, type);
1373 rc = cifs_lock_add_if(cfile, lock, wait_flag);
1379 rc = cifs_mandatory_lock(xid, cfile, flock->fl_start, length,
1380 type, 1, 0, wait_flag);
1386 cifs_lock_add(cfile, lock);
1388 rc = cifs_unlock_range(cfile, flock, xid);
1391 if (flock->fl_flags & FL_POSIX)
1392 posix_lock_file_wait(file, flock);
1396 int cifs_lock(struct file *file, int cmd, struct file_lock *flock)
1399 int lock = 0, unlock = 0;
1400 bool wait_flag = false;
1401 bool posix_lck = false;
1402 struct cifs_sb_info *cifs_sb;
1403 struct cifs_tcon *tcon;
1404 struct cifsInodeInfo *cinode;
1405 struct cifsFileInfo *cfile;
1412 cFYI(1, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld "
1413 "end: %lld", cmd, flock->fl_flags, flock->fl_type,
1414 flock->fl_start, flock->fl_end);
1416 cfile = (struct cifsFileInfo *)file->private_data;
1417 tcon = tlink_tcon(cfile->tlink);
1419 cifs_read_flock(flock, &type, &lock, &unlock, &wait_flag,
1422 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1423 netfid = cfile->fid.netfid;
1424 cinode = CIFS_I(file->f_path.dentry->d_inode);
1426 if (cap_unix(tcon->ses) &&
1427 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1428 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1431 * BB add code here to normalize offset and length to account for
1432 * negative length which we can not accept over the wire.
1434 if (IS_GETLK(cmd)) {
1435 rc = cifs_getlk(file, flock, type, wait_flag, posix_lck, xid);
1440 if (!lock && !unlock) {
1442 * if no lock or unlock then nothing to do since we do not
1449 rc = cifs_setlk(file, flock, type, wait_flag, posix_lck, lock, unlock,
1456 * update the file size (if needed) after a write. Should be called with
1457 * the inode->i_lock held
1460 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
1461 unsigned int bytes_written)
1463 loff_t end_of_write = offset + bytes_written;
1465 if (end_of_write > cifsi->server_eof)
1466 cifsi->server_eof = end_of_write;
1469 static ssize_t cifs_write(struct cifsFileInfo *open_file, __u32 pid,
1470 const char *write_data, size_t write_size,
1474 unsigned int bytes_written = 0;
1475 unsigned int total_written;
1476 struct cifs_sb_info *cifs_sb;
1477 struct cifs_tcon *pTcon;
1479 struct dentry *dentry = open_file->dentry;
1480 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
1481 struct cifs_io_parms io_parms;
1483 cifs_sb = CIFS_SB(dentry->d_sb);
1485 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
1486 *poffset, dentry->d_name.name);
1488 pTcon = tlink_tcon(open_file->tlink);
1492 for (total_written = 0; write_size > total_written;
1493 total_written += bytes_written) {
1495 while (rc == -EAGAIN) {
1499 if (open_file->invalidHandle) {
1500 /* we could deadlock if we called
1501 filemap_fdatawait from here so tell
1502 reopen_file not to flush data to
1504 rc = cifs_reopen_file(open_file, false);
1509 len = min((size_t)cifs_sb->wsize,
1510 write_size - total_written);
1511 /* iov[0] is reserved for smb header */
1512 iov[1].iov_base = (char *)write_data + total_written;
1513 iov[1].iov_len = len;
1514 io_parms.netfid = open_file->fid.netfid;
1516 io_parms.tcon = pTcon;
1517 io_parms.offset = *poffset;
1518 io_parms.length = len;
1519 rc = CIFSSMBWrite2(xid, &io_parms, &bytes_written, iov,
1522 if (rc || (bytes_written == 0)) {
1530 spin_lock(&dentry->d_inode->i_lock);
1531 cifs_update_eof(cifsi, *poffset, bytes_written);
1532 spin_unlock(&dentry->d_inode->i_lock);
1533 *poffset += bytes_written;
1537 cifs_stats_bytes_written(pTcon, total_written);
1539 if (total_written > 0) {
1540 spin_lock(&dentry->d_inode->i_lock);
1541 if (*poffset > dentry->d_inode->i_size)
1542 i_size_write(dentry->d_inode, *poffset);
1543 spin_unlock(&dentry->d_inode->i_lock);
1545 mark_inode_dirty_sync(dentry->d_inode);
1547 return total_written;
1550 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1553 struct cifsFileInfo *open_file = NULL;
1554 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1556 /* only filter by fsuid on multiuser mounts */
1557 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1560 spin_lock(&cifs_file_list_lock);
1561 /* we could simply get the first_list_entry since write-only entries
1562 are always at the end of the list but since the first entry might
1563 have a close pending, we go through the whole list */
1564 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1565 if (fsuid_only && open_file->uid != current_fsuid())
1567 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1568 if (!open_file->invalidHandle) {
1569 /* found a good file */
1570 /* lock it so it will not be closed on us */
1571 cifsFileInfo_get_locked(open_file);
1572 spin_unlock(&cifs_file_list_lock);
1574 } /* else might as well continue, and look for
1575 another, or simply have the caller reopen it
1576 again rather than trying to fix this handle */
1577 } else /* write only file */
1578 break; /* write only files are last so must be done */
1580 spin_unlock(&cifs_file_list_lock);
1584 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1587 struct cifsFileInfo *open_file, *inv_file = NULL;
1588 struct cifs_sb_info *cifs_sb;
1589 bool any_available = false;
1591 unsigned int refind = 0;
1593 /* Having a null inode here (because mapping->host was set to zero by
1594 the VFS or MM) should not happen but we had reports of on oops (due to
1595 it being zero) during stress testcases so we need to check for it */
1597 if (cifs_inode == NULL) {
1598 cERROR(1, "Null inode passed to cifs_writeable_file");
1603 cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1605 /* only filter by fsuid on multiuser mounts */
1606 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1609 spin_lock(&cifs_file_list_lock);
1611 if (refind > MAX_REOPEN_ATT) {
1612 spin_unlock(&cifs_file_list_lock);
1615 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1616 if (!any_available && open_file->pid != current->tgid)
1618 if (fsuid_only && open_file->uid != current_fsuid())
1620 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1621 if (!open_file->invalidHandle) {
1622 /* found a good writable file */
1623 cifsFileInfo_get_locked(open_file);
1624 spin_unlock(&cifs_file_list_lock);
1628 inv_file = open_file;
1632 /* couldn't find useable FH with same pid, try any available */
1633 if (!any_available) {
1634 any_available = true;
1635 goto refind_writable;
1639 any_available = false;
1640 cifsFileInfo_get_locked(inv_file);
1643 spin_unlock(&cifs_file_list_lock);
1646 rc = cifs_reopen_file(inv_file, false);
1650 spin_lock(&cifs_file_list_lock);
1651 list_move_tail(&inv_file->flist,
1652 &cifs_inode->openFileList);
1653 spin_unlock(&cifs_file_list_lock);
1654 cifsFileInfo_put(inv_file);
1655 spin_lock(&cifs_file_list_lock);
1657 goto refind_writable;
1664 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1666 struct address_space *mapping = page->mapping;
1667 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1670 int bytes_written = 0;
1671 struct inode *inode;
1672 struct cifsFileInfo *open_file;
1674 if (!mapping || !mapping->host)
1677 inode = page->mapping->host;
1679 offset += (loff_t)from;
1680 write_data = kmap(page);
1683 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1688 /* racing with truncate? */
1689 if (offset > mapping->host->i_size) {
1691 return 0; /* don't care */
1694 /* check to make sure that we are not extending the file */
1695 if (mapping->host->i_size - offset < (loff_t)to)
1696 to = (unsigned)(mapping->host->i_size - offset);
1698 open_file = find_writable_file(CIFS_I(mapping->host), false);
1700 bytes_written = cifs_write(open_file, open_file->pid,
1701 write_data, to - from, &offset);
1702 cifsFileInfo_put(open_file);
1703 /* Does mm or vfs already set times? */
1704 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1705 if ((bytes_written > 0) && (offset))
1707 else if (bytes_written < 0)
1710 cFYI(1, "No writeable filehandles for inode");
1719 * Marshal up the iov array, reserving the first one for the header. Also,
1723 cifs_writepages_marshal_iov(struct kvec *iov, struct cifs_writedata *wdata)
1726 struct inode *inode = wdata->cfile->dentry->d_inode;
1727 loff_t size = i_size_read(inode);
1729 /* marshal up the pages into iov array */
1731 for (i = 0; i < wdata->nr_pages; i++) {
1732 iov[i + 1].iov_len = min(size - page_offset(wdata->pages[i]),
1733 (loff_t)PAGE_CACHE_SIZE);
1734 iov[i + 1].iov_base = kmap(wdata->pages[i]);
1735 wdata->bytes += iov[i + 1].iov_len;
1739 static int cifs_writepages(struct address_space *mapping,
1740 struct writeback_control *wbc)
1742 struct cifs_sb_info *cifs_sb = CIFS_SB(mapping->host->i_sb);
1743 bool done = false, scanned = false, range_whole = false;
1745 struct cifs_writedata *wdata;
1750 * If wsize is smaller than the page cache size, default to writing
1751 * one page at a time via cifs_writepage
1753 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1754 return generic_writepages(mapping, wbc);
1756 if (wbc->range_cyclic) {
1757 index = mapping->writeback_index; /* Start from prev offset */
1760 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1761 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1762 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1767 while (!done && index <= end) {
1768 unsigned int i, nr_pages, found_pages;
1769 pgoff_t next = 0, tofind;
1770 struct page **pages;
1772 tofind = min((cifs_sb->wsize / PAGE_CACHE_SIZE) - 1,
1775 wdata = cifs_writedata_alloc((unsigned int)tofind,
1776 cifs_writev_complete);
1783 * find_get_pages_tag seems to return a max of 256 on each
1784 * iteration, so we must call it several times in order to
1785 * fill the array or the wsize is effectively limited to
1786 * 256 * PAGE_CACHE_SIZE.
1789 pages = wdata->pages;
1791 nr_pages = find_get_pages_tag(mapping, &index,
1792 PAGECACHE_TAG_DIRTY,
1794 found_pages += nr_pages;
1797 } while (nr_pages && tofind && index <= end);
1799 if (found_pages == 0) {
1800 kref_put(&wdata->refcount, cifs_writedata_release);
1805 for (i = 0; i < found_pages; i++) {
1806 page = wdata->pages[i];
1808 * At this point we hold neither mapping->tree_lock nor
1809 * lock on the page itself: the page may be truncated or
1810 * invalidated (changing page->mapping to NULL), or even
1811 * swizzled back from swapper_space to tmpfs file
1817 else if (!trylock_page(page))
1820 if (unlikely(page->mapping != mapping)) {
1825 if (!wbc->range_cyclic && page->index > end) {
1831 if (next && (page->index != next)) {
1832 /* Not next consecutive page */
1837 if (wbc->sync_mode != WB_SYNC_NONE)
1838 wait_on_page_writeback(page);
1840 if (PageWriteback(page) ||
1841 !clear_page_dirty_for_io(page)) {
1847 * This actually clears the dirty bit in the radix tree.
1848 * See cifs_writepage() for more commentary.
1850 set_page_writeback(page);
1852 if (page_offset(page) >= mapping->host->i_size) {
1855 end_page_writeback(page);
1859 wdata->pages[i] = page;
1860 next = page->index + 1;
1864 /* reset index to refind any pages skipped */
1866 index = wdata->pages[0]->index + 1;
1868 /* put any pages we aren't going to use */
1869 for (i = nr_pages; i < found_pages; i++) {
1870 page_cache_release(wdata->pages[i]);
1871 wdata->pages[i] = NULL;
1874 /* nothing to write? */
1875 if (nr_pages == 0) {
1876 kref_put(&wdata->refcount, cifs_writedata_release);
1880 wdata->sync_mode = wbc->sync_mode;
1881 wdata->nr_pages = nr_pages;
1882 wdata->offset = page_offset(wdata->pages[0]);
1883 wdata->marshal_iov = cifs_writepages_marshal_iov;
1886 if (wdata->cfile != NULL)
1887 cifsFileInfo_put(wdata->cfile);
1888 wdata->cfile = find_writable_file(CIFS_I(mapping->host),
1890 if (!wdata->cfile) {
1891 cERROR(1, "No writable handles for inode");
1895 wdata->pid = wdata->cfile->pid;
1896 rc = cifs_async_writev(wdata);
1897 } while (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN);
1899 for (i = 0; i < nr_pages; ++i)
1900 unlock_page(wdata->pages[i]);
1902 /* send failure -- clean up the mess */
1904 for (i = 0; i < nr_pages; ++i) {
1906 redirty_page_for_writepage(wbc,
1909 SetPageError(wdata->pages[i]);
1910 end_page_writeback(wdata->pages[i]);
1911 page_cache_release(wdata->pages[i]);
1914 mapping_set_error(mapping, rc);
1916 kref_put(&wdata->refcount, cifs_writedata_release);
1918 wbc->nr_to_write -= nr_pages;
1919 if (wbc->nr_to_write <= 0)
1925 if (!scanned && !done) {
1927 * We hit the last page and there is more work to be done: wrap
1928 * back to the start of the file
1935 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1936 mapping->writeback_index = index;
1942 cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
1948 /* BB add check for wbc flags */
1949 page_cache_get(page);
1950 if (!PageUptodate(page))
1951 cFYI(1, "ppw - page not up to date");
1954 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1956 * A writepage() implementation always needs to do either this,
1957 * or re-dirty the page with "redirty_page_for_writepage()" in
1958 * the case of a failure.
1960 * Just unlocking the page will cause the radix tree tag-bits
1961 * to fail to update with the state of the page correctly.
1963 set_page_writeback(page);
1965 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1966 if (rc == -EAGAIN && wbc->sync_mode == WB_SYNC_ALL)
1968 else if (rc == -EAGAIN)
1969 redirty_page_for_writepage(wbc, page);
1973 SetPageUptodate(page);
1974 end_page_writeback(page);
1975 page_cache_release(page);
1980 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1982 int rc = cifs_writepage_locked(page, wbc);
1987 static int cifs_write_end(struct file *file, struct address_space *mapping,
1988 loff_t pos, unsigned len, unsigned copied,
1989 struct page *page, void *fsdata)
1992 struct inode *inode = mapping->host;
1993 struct cifsFileInfo *cfile = file->private_data;
1994 struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
1997 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2000 pid = current->tgid;
2002 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
2005 if (PageChecked(page)) {
2007 SetPageUptodate(page);
2008 ClearPageChecked(page);
2009 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
2010 SetPageUptodate(page);
2012 if (!PageUptodate(page)) {
2014 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
2018 /* this is probably better than directly calling
2019 partialpage_write since in this function the file handle is
2020 known which we might as well leverage */
2021 /* BB check if anything else missing out of ppw
2022 such as updating last write time */
2023 page_data = kmap(page);
2024 rc = cifs_write(cfile, pid, page_data + offset, copied, &pos);
2025 /* if (rc < 0) should we set writebehind rc? */
2032 set_page_dirty(page);
2036 spin_lock(&inode->i_lock);
2037 if (pos > inode->i_size)
2038 i_size_write(inode, pos);
2039 spin_unlock(&inode->i_lock);
2043 page_cache_release(page);
2048 int cifs_strict_fsync(struct file *file, loff_t start, loff_t end,
2053 struct cifs_tcon *tcon;
2054 struct cifsFileInfo *smbfile = file->private_data;
2055 struct inode *inode = file->f_path.dentry->d_inode;
2056 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2058 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2061 mutex_lock(&inode->i_mutex);
2065 cFYI(1, "Sync file - name: %s datasync: 0x%x",
2066 file->f_path.dentry->d_name.name, datasync);
2068 if (!CIFS_I(inode)->clientCanCacheRead) {
2069 rc = cifs_invalidate_mapping(inode);
2071 cFYI(1, "rc: %d during invalidate phase", rc);
2072 rc = 0; /* don't care about it in fsync */
2076 tcon = tlink_tcon(smbfile->tlink);
2077 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
2078 rc = CIFSSMBFlush(xid, tcon, smbfile->fid.netfid);
2081 mutex_unlock(&inode->i_mutex);
2085 int cifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2089 struct cifs_tcon *tcon;
2090 struct cifsFileInfo *smbfile = file->private_data;
2091 struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2092 struct inode *inode = file->f_mapping->host;
2094 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2097 mutex_lock(&inode->i_mutex);
2101 cFYI(1, "Sync file - name: %s datasync: 0x%x",
2102 file->f_path.dentry->d_name.name, datasync);
2104 tcon = tlink_tcon(smbfile->tlink);
2105 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
2106 rc = CIFSSMBFlush(xid, tcon, smbfile->fid.netfid);
2109 mutex_unlock(&inode->i_mutex);
2114 * As file closes, flush all cached write data for this inode checking
2115 * for write behind errors.
2117 int cifs_flush(struct file *file, fl_owner_t id)
2119 struct inode *inode = file->f_path.dentry->d_inode;
2122 if (file->f_mode & FMODE_WRITE)
2123 rc = filemap_write_and_wait(inode->i_mapping);
2125 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
2131 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
2136 for (i = 0; i < num_pages; i++) {
2137 pages[i] = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2140 * save number of pages we have already allocated and
2141 * return with ENOMEM error
2150 for (i = 0; i < num_pages; i++)
2157 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
2162 clen = min_t(const size_t, len, wsize);
2163 num_pages = DIV_ROUND_UP(clen, PAGE_SIZE);
2172 cifs_uncached_marshal_iov(struct kvec *iov, struct cifs_writedata *wdata)
2175 size_t bytes = wdata->bytes;
2177 /* marshal up the pages into iov array */
2178 for (i = 0; i < wdata->nr_pages; i++) {
2179 iov[i + 1].iov_len = min_t(size_t, bytes, PAGE_SIZE);
2180 iov[i + 1].iov_base = kmap(wdata->pages[i]);
2181 bytes -= iov[i + 1].iov_len;
2186 cifs_uncached_writev_complete(struct work_struct *work)
2189 struct cifs_writedata *wdata = container_of(work,
2190 struct cifs_writedata, work);
2191 struct inode *inode = wdata->cfile->dentry->d_inode;
2192 struct cifsInodeInfo *cifsi = CIFS_I(inode);
2194 spin_lock(&inode->i_lock);
2195 cifs_update_eof(cifsi, wdata->offset, wdata->bytes);
2196 if (cifsi->server_eof > inode->i_size)
2197 i_size_write(inode, cifsi->server_eof);
2198 spin_unlock(&inode->i_lock);
2200 complete(&wdata->done);
2202 if (wdata->result != -EAGAIN) {
2203 for (i = 0; i < wdata->nr_pages; i++)
2204 put_page(wdata->pages[i]);
2207 kref_put(&wdata->refcount, cifs_writedata_release);
2210 /* attempt to send write to server, retry on any -EAGAIN errors */
2212 cifs_uncached_retry_writev(struct cifs_writedata *wdata)
2217 if (wdata->cfile->invalidHandle) {
2218 rc = cifs_reopen_file(wdata->cfile, false);
2222 rc = cifs_async_writev(wdata);
2223 } while (rc == -EAGAIN);
2229 cifs_iovec_write(struct file *file, const struct iovec *iov,
2230 unsigned long nr_segs, loff_t *poffset)
2232 unsigned long nr_pages, i;
2233 size_t copied, len, cur_len;
2234 ssize_t total_written = 0;
2237 struct cifsFileInfo *open_file;
2238 struct cifs_tcon *tcon;
2239 struct cifs_sb_info *cifs_sb;
2240 struct cifs_writedata *wdata, *tmp;
2241 struct list_head wdata_list;
2245 len = iov_length(iov, nr_segs);
2249 rc = generic_write_checks(file, poffset, &len, 0);
2253 INIT_LIST_HEAD(&wdata_list);
2254 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2255 open_file = file->private_data;
2256 tcon = tlink_tcon(open_file->tlink);
2259 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2260 pid = open_file->pid;
2262 pid = current->tgid;
2264 iov_iter_init(&it, iov, nr_segs, len, 0);
2268 nr_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
2269 wdata = cifs_writedata_alloc(nr_pages,
2270 cifs_uncached_writev_complete);
2276 rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
2283 for (i = 0; i < nr_pages; i++) {
2284 copied = min_t(const size_t, cur_len, PAGE_SIZE);
2285 copied = iov_iter_copy_from_user(wdata->pages[i], &it,
2288 iov_iter_advance(&it, copied);
2290 cur_len = save_len - cur_len;
2292 wdata->sync_mode = WB_SYNC_ALL;
2293 wdata->nr_pages = nr_pages;
2294 wdata->offset = (__u64)offset;
2295 wdata->cfile = cifsFileInfo_get(open_file);
2297 wdata->bytes = cur_len;
2298 wdata->marshal_iov = cifs_uncached_marshal_iov;
2299 rc = cifs_uncached_retry_writev(wdata);
2301 kref_put(&wdata->refcount, cifs_writedata_release);
2305 list_add_tail(&wdata->list, &wdata_list);
2311 * If at least one write was successfully sent, then discard any rc
2312 * value from the later writes. If the other write succeeds, then
2313 * we'll end up returning whatever was written. If it fails, then
2314 * we'll get a new rc value from that.
2316 if (!list_empty(&wdata_list))
2320 * Wait for and collect replies for any successful sends in order of
2321 * increasing offset. Once an error is hit or we get a fatal signal
2322 * while waiting, then return without waiting for any more replies.
2325 list_for_each_entry_safe(wdata, tmp, &wdata_list, list) {
2327 /* FIXME: freezable too? */
2328 rc = wait_for_completion_killable(&wdata->done);
2331 else if (wdata->result)
2334 total_written += wdata->bytes;
2336 /* resend call if it's a retryable error */
2337 if (rc == -EAGAIN) {
2338 rc = cifs_uncached_retry_writev(wdata);
2342 list_del_init(&wdata->list);
2343 kref_put(&wdata->refcount, cifs_writedata_release);
2346 if (total_written > 0)
2347 *poffset += total_written;
2349 cifs_stats_bytes_written(tcon, total_written);
2350 return total_written ? total_written : (ssize_t)rc;
2353 ssize_t cifs_user_writev(struct kiocb *iocb, const struct iovec *iov,
2354 unsigned long nr_segs, loff_t pos)
2357 struct inode *inode;
2359 inode = iocb->ki_filp->f_path.dentry->d_inode;
2362 * BB - optimize the way when signing is disabled. We can drop this
2363 * extra memory-to-memory copying and use iovec buffers for constructing
2367 written = cifs_iovec_write(iocb->ki_filp, iov, nr_segs, &pos);
2369 CIFS_I(inode)->invalid_mapping = true;
2376 ssize_t cifs_strict_writev(struct kiocb *iocb, const struct iovec *iov,
2377 unsigned long nr_segs, loff_t pos)
2379 struct inode *inode;
2381 inode = iocb->ki_filp->f_path.dentry->d_inode;
2383 if (CIFS_I(inode)->clientCanCacheAll)
2384 return generic_file_aio_write(iocb, iov, nr_segs, pos);
2387 * In strict cache mode we need to write the data to the server exactly
2388 * from the pos to pos+len-1 rather than flush all affected pages
2389 * because it may cause a error with mandatory locks on these pages but
2390 * not on the region from pos to ppos+len-1.
2393 return cifs_user_writev(iocb, iov, nr_segs, pos);
2396 static struct cifs_readdata *
2397 cifs_readdata_alloc(unsigned int nr_vecs, work_func_t complete)
2399 struct cifs_readdata *rdata;
2401 rdata = kzalloc(sizeof(*rdata) +
2402 sizeof(struct kvec) * nr_vecs, GFP_KERNEL);
2403 if (rdata != NULL) {
2404 kref_init(&rdata->refcount);
2405 INIT_LIST_HEAD(&rdata->list);
2406 init_completion(&rdata->done);
2407 INIT_WORK(&rdata->work, complete);
2408 INIT_LIST_HEAD(&rdata->pages);
2414 cifs_readdata_release(struct kref *refcount)
2416 struct cifs_readdata *rdata = container_of(refcount,
2417 struct cifs_readdata, refcount);
2420 cifsFileInfo_put(rdata->cfile);
2426 cifs_read_allocate_pages(struct list_head *list, unsigned int npages)
2429 struct page *page, *tpage;
2432 for (i = 0; i < npages; i++) {
2433 page = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2438 list_add(&page->lru, list);
2442 list_for_each_entry_safe(page, tpage, list, lru) {
2443 list_del(&page->lru);
2451 cifs_uncached_readdata_release(struct kref *refcount)
2453 struct page *page, *tpage;
2454 struct cifs_readdata *rdata = container_of(refcount,
2455 struct cifs_readdata, refcount);
2457 list_for_each_entry_safe(page, tpage, &rdata->pages, lru) {
2458 list_del(&page->lru);
2461 cifs_readdata_release(refcount);
2465 cifs_retry_async_readv(struct cifs_readdata *rdata)
2470 if (rdata->cfile->invalidHandle) {
2471 rc = cifs_reopen_file(rdata->cfile, true);
2475 rc = cifs_async_readv(rdata);
2476 } while (rc == -EAGAIN);
2482 * cifs_readdata_to_iov - copy data from pages in response to an iovec
2483 * @rdata: the readdata response with list of pages holding data
2484 * @iov: vector in which we should copy the data
2485 * @nr_segs: number of segments in vector
2486 * @offset: offset into file of the first iovec
2487 * @copied: used to return the amount of data copied to the iov
2489 * This function copies data from a list of pages in a readdata response into
2490 * an array of iovecs. It will first calculate where the data should go
2491 * based on the info in the readdata and then copy the data into that spot.
2494 cifs_readdata_to_iov(struct cifs_readdata *rdata, const struct iovec *iov,
2495 unsigned long nr_segs, loff_t offset, ssize_t *copied)
2499 size_t pos = rdata->offset - offset;
2500 struct page *page, *tpage;
2501 ssize_t remaining = rdata->bytes;
2502 unsigned char *pdata;
2504 /* set up iov_iter and advance to the correct offset */
2505 iov_iter_init(&ii, iov, nr_segs, iov_length(iov, nr_segs), 0);
2506 iov_iter_advance(&ii, pos);
2509 list_for_each_entry_safe(page, tpage, &rdata->pages, lru) {
2512 /* copy a whole page or whatever's left */
2513 copy = min_t(ssize_t, remaining, PAGE_SIZE);
2515 /* ...but limit it to whatever space is left in the iov */
2516 copy = min_t(ssize_t, copy, iov_iter_count(&ii));
2518 /* go while there's data to be copied and no errors */
2521 rc = memcpy_toiovecend(ii.iov, pdata, ii.iov_offset,
2527 iov_iter_advance(&ii, copy);
2531 list_del(&page->lru);
2539 cifs_uncached_readv_complete(struct work_struct *work)
2541 struct cifs_readdata *rdata = container_of(work,
2542 struct cifs_readdata, work);
2544 /* if the result is non-zero then the pages weren't kmapped */
2545 if (rdata->result == 0) {
2548 list_for_each_entry(page, &rdata->pages, lru)
2552 complete(&rdata->done);
2553 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
2557 cifs_uncached_read_marshal_iov(struct cifs_readdata *rdata,
2558 unsigned int remaining)
2561 struct page *page, *tpage;
2564 list_for_each_entry_safe(page, tpage, &rdata->pages, lru) {
2565 if (remaining >= PAGE_SIZE) {
2566 /* enough data to fill the page */
2567 rdata->iov[rdata->nr_iov].iov_base = kmap(page);
2568 rdata->iov[rdata->nr_iov].iov_len = PAGE_SIZE;
2569 cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
2570 rdata->nr_iov, page->index,
2571 rdata->iov[rdata->nr_iov].iov_base,
2572 rdata->iov[rdata->nr_iov].iov_len);
2575 remaining -= PAGE_SIZE;
2576 } else if (remaining > 0) {
2577 /* enough for partial page, fill and zero the rest */
2578 rdata->iov[rdata->nr_iov].iov_base = kmap(page);
2579 rdata->iov[rdata->nr_iov].iov_len = remaining;
2580 cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
2581 rdata->nr_iov, page->index,
2582 rdata->iov[rdata->nr_iov].iov_base,
2583 rdata->iov[rdata->nr_iov].iov_len);
2584 memset(rdata->iov[rdata->nr_iov].iov_base + remaining,
2585 '\0', PAGE_SIZE - remaining);
2590 /* no need to hold page hostage */
2591 list_del(&page->lru);
2600 cifs_iovec_read(struct file *file, const struct iovec *iov,
2601 unsigned long nr_segs, loff_t *poffset)
2604 size_t len, cur_len;
2605 ssize_t total_read = 0;
2606 loff_t offset = *poffset;
2607 unsigned int npages;
2608 struct cifs_sb_info *cifs_sb;
2609 struct cifs_tcon *tcon;
2610 struct cifsFileInfo *open_file;
2611 struct cifs_readdata *rdata, *tmp;
2612 struct list_head rdata_list;
2618 len = iov_length(iov, nr_segs);
2622 INIT_LIST_HEAD(&rdata_list);
2623 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2624 open_file = file->private_data;
2625 tcon = tlink_tcon(open_file->tlink);
2627 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2628 pid = open_file->pid;
2630 pid = current->tgid;
2632 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
2633 cFYI(1, "attempting read on write only file instance");
2636 cur_len = min_t(const size_t, len - total_read, cifs_sb->rsize);
2637 npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
2639 /* allocate a readdata struct */
2640 rdata = cifs_readdata_alloc(npages,
2641 cifs_uncached_readv_complete);
2647 rc = cifs_read_allocate_pages(&rdata->pages, npages);
2651 rdata->cfile = cifsFileInfo_get(open_file);
2652 rdata->offset = offset;
2653 rdata->bytes = cur_len;
2655 rdata->marshal_iov = cifs_uncached_read_marshal_iov;
2657 rc = cifs_retry_async_readv(rdata);
2660 kref_put(&rdata->refcount,
2661 cifs_uncached_readdata_release);
2665 list_add_tail(&rdata->list, &rdata_list);
2670 /* if at least one read request send succeeded, then reset rc */
2671 if (!list_empty(&rdata_list))
2674 /* the loop below should proceed in the order of increasing offsets */
2676 list_for_each_entry_safe(rdata, tmp, &rdata_list, list) {
2680 /* FIXME: freezable sleep too? */
2681 rc = wait_for_completion_killable(&rdata->done);
2684 else if (rdata->result)
2687 rc = cifs_readdata_to_iov(rdata, iov,
2690 total_read += copied;
2693 /* resend call if it's a retryable error */
2694 if (rc == -EAGAIN) {
2695 rc = cifs_retry_async_readv(rdata);
2699 list_del_init(&rdata->list);
2700 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
2703 cifs_stats_bytes_read(tcon, total_read);
2704 *poffset += total_read;
2706 return total_read ? total_read : rc;
2709 ssize_t cifs_user_readv(struct kiocb *iocb, const struct iovec *iov,
2710 unsigned long nr_segs, loff_t pos)
2714 read = cifs_iovec_read(iocb->ki_filp, iov, nr_segs, &pos);
2721 ssize_t cifs_strict_readv(struct kiocb *iocb, const struct iovec *iov,
2722 unsigned long nr_segs, loff_t pos)
2724 struct inode *inode;
2726 inode = iocb->ki_filp->f_path.dentry->d_inode;
2728 if (CIFS_I(inode)->clientCanCacheRead)
2729 return generic_file_aio_read(iocb, iov, nr_segs, pos);
2732 * In strict cache mode we need to read from the server all the time
2733 * if we don't have level II oplock because the server can delay mtime
2734 * change - so we can't make a decision about inode invalidating.
2735 * And we can also fail with pagereading if there are mandatory locks
2736 * on pages affected by this read but not on the region from pos to
2740 return cifs_user_readv(iocb, iov, nr_segs, pos);
2743 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
2747 unsigned int bytes_read = 0;
2748 unsigned int total_read;
2749 unsigned int current_read_size;
2751 struct cifs_sb_info *cifs_sb;
2752 struct cifs_tcon *tcon;
2754 char *current_offset;
2755 struct cifsFileInfo *open_file;
2756 struct cifs_io_parms io_parms;
2757 int buf_type = CIFS_NO_BUFFER;
2761 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2763 /* FIXME: set up handlers for larger reads and/or convert to async */
2764 rsize = min_t(unsigned int, cifs_sb->rsize, CIFSMaxBufSize);
2766 if (file->private_data == NULL) {
2771 open_file = file->private_data;
2772 tcon = tlink_tcon(open_file->tlink);
2774 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2775 pid = open_file->pid;
2777 pid = current->tgid;
2779 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
2780 cFYI(1, "attempting read on write only file instance");
2782 for (total_read = 0, current_offset = read_data;
2783 read_size > total_read;
2784 total_read += bytes_read, current_offset += bytes_read) {
2785 current_read_size = min_t(uint, read_size - total_read, rsize);
2787 * For windows me and 9x we do not want to request more than it
2788 * negotiated since it will refuse the read then.
2790 if ((tcon->ses) && !(tcon->ses->capabilities &
2791 tcon->ses->server->vals->cap_large_files)) {
2792 current_read_size = min_t(uint, current_read_size,
2796 while (rc == -EAGAIN) {
2797 if (open_file->invalidHandle) {
2798 rc = cifs_reopen_file(open_file, true);
2802 io_parms.netfid = open_file->fid.netfid;
2804 io_parms.tcon = tcon;
2805 io_parms.offset = *poffset;
2806 io_parms.length = current_read_size;
2807 rc = CIFSSMBRead(xid, &io_parms, &bytes_read,
2808 ¤t_offset, &buf_type);
2810 if (rc || (bytes_read == 0)) {
2818 cifs_stats_bytes_read(tcon, total_read);
2819 *poffset += bytes_read;
2827 * If the page is mmap'ed into a process' page tables, then we need to make
2828 * sure that it doesn't change while being written back.
2831 cifs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2833 struct page *page = vmf->page;
2836 return VM_FAULT_LOCKED;
2839 static struct vm_operations_struct cifs_file_vm_ops = {
2840 .fault = filemap_fault,
2841 .page_mkwrite = cifs_page_mkwrite,
2844 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
2847 struct inode *inode = file->f_path.dentry->d_inode;
2851 if (!CIFS_I(inode)->clientCanCacheRead) {
2852 rc = cifs_invalidate_mapping(inode);
2857 rc = generic_file_mmap(file, vma);
2859 vma->vm_ops = &cifs_file_vm_ops;
2864 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
2869 rc = cifs_revalidate_file(file);
2871 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
2875 rc = generic_file_mmap(file, vma);
2877 vma->vm_ops = &cifs_file_vm_ops;
2883 cifs_readv_complete(struct work_struct *work)
2885 struct cifs_readdata *rdata = container_of(work,
2886 struct cifs_readdata, work);
2887 struct page *page, *tpage;
2889 list_for_each_entry_safe(page, tpage, &rdata->pages, lru) {
2890 list_del(&page->lru);
2891 lru_cache_add_file(page);
2893 if (rdata->result == 0) {
2895 flush_dcache_page(page);
2896 SetPageUptodate(page);
2901 if (rdata->result == 0)
2902 cifs_readpage_to_fscache(rdata->mapping->host, page);
2904 page_cache_release(page);
2906 kref_put(&rdata->refcount, cifs_readdata_release);
2910 cifs_readpages_marshal_iov(struct cifs_readdata *rdata, unsigned int remaining)
2913 struct page *page, *tpage;
2917 /* determine the eof that the server (probably) has */
2918 eof = CIFS_I(rdata->mapping->host)->server_eof;
2919 eof_index = eof ? (eof - 1) >> PAGE_CACHE_SHIFT : 0;
2920 cFYI(1, "eof=%llu eof_index=%lu", eof, eof_index);
2923 list_for_each_entry_safe(page, tpage, &rdata->pages, lru) {
2924 if (remaining >= PAGE_CACHE_SIZE) {
2925 /* enough data to fill the page */
2926 rdata->iov[rdata->nr_iov].iov_base = kmap(page);
2927 rdata->iov[rdata->nr_iov].iov_len = PAGE_CACHE_SIZE;
2928 cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
2929 rdata->nr_iov, page->index,
2930 rdata->iov[rdata->nr_iov].iov_base,
2931 rdata->iov[rdata->nr_iov].iov_len);
2933 len += PAGE_CACHE_SIZE;
2934 remaining -= PAGE_CACHE_SIZE;
2935 } else if (remaining > 0) {
2936 /* enough for partial page, fill and zero the rest */
2937 rdata->iov[rdata->nr_iov].iov_base = kmap(page);
2938 rdata->iov[rdata->nr_iov].iov_len = remaining;
2939 cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
2940 rdata->nr_iov, page->index,
2941 rdata->iov[rdata->nr_iov].iov_base,
2942 rdata->iov[rdata->nr_iov].iov_len);
2943 memset(rdata->iov[rdata->nr_iov].iov_base + remaining,
2944 '\0', PAGE_CACHE_SIZE - remaining);
2948 } else if (page->index > eof_index) {
2950 * The VFS will not try to do readahead past the
2951 * i_size, but it's possible that we have outstanding
2952 * writes with gaps in the middle and the i_size hasn't
2953 * caught up yet. Populate those with zeroed out pages
2954 * to prevent the VFS from repeatedly attempting to
2955 * fill them until the writes are flushed.
2957 zero_user(page, 0, PAGE_CACHE_SIZE);
2958 list_del(&page->lru);
2959 lru_cache_add_file(page);
2960 flush_dcache_page(page);
2961 SetPageUptodate(page);
2963 page_cache_release(page);
2965 /* no need to hold page hostage */
2966 list_del(&page->lru);
2967 lru_cache_add_file(page);
2969 page_cache_release(page);
2976 static int cifs_readpages(struct file *file, struct address_space *mapping,
2977 struct list_head *page_list, unsigned num_pages)
2980 struct list_head tmplist;
2981 struct cifsFileInfo *open_file = file->private_data;
2982 struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2983 unsigned int rsize = cifs_sb->rsize;
2987 * Give up immediately if rsize is too small to read an entire page.
2988 * The VFS will fall back to readpage. We should never reach this
2989 * point however since we set ra_pages to 0 when the rsize is smaller
2990 * than a cache page.
2992 if (unlikely(rsize < PAGE_CACHE_SIZE))
2996 * Reads as many pages as possible from fscache. Returns -ENOBUFS
2997 * immediately if the cookie is negative
2999 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
3004 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3005 pid = open_file->pid;
3007 pid = current->tgid;
3010 INIT_LIST_HEAD(&tmplist);
3012 cFYI(1, "%s: file=%p mapping=%p num_pages=%u", __func__, file,
3013 mapping, num_pages);
3016 * Start with the page at end of list and move it to private
3017 * list. Do the same with any following pages until we hit
3018 * the rsize limit, hit an index discontinuity, or run out of
3019 * pages. Issue the async read and then start the loop again
3020 * until the list is empty.
3022 * Note that list order is important. The page_list is in
3023 * the order of declining indexes. When we put the pages in
3024 * the rdata->pages, then we want them in increasing order.
3026 while (!list_empty(page_list)) {
3027 unsigned int bytes = PAGE_CACHE_SIZE;
3028 unsigned int expected_index;
3029 unsigned int nr_pages = 1;
3031 struct page *page, *tpage;
3032 struct cifs_readdata *rdata;
3034 page = list_entry(page_list->prev, struct page, lru);
3037 * Lock the page and put it in the cache. Since no one else
3038 * should have access to this page, we're safe to simply set
3039 * PG_locked without checking it first.
3041 __set_page_locked(page);
3042 rc = add_to_page_cache_locked(page, mapping,
3043 page->index, GFP_KERNEL);
3045 /* give up if we can't stick it in the cache */
3047 __clear_page_locked(page);
3051 /* move first page to the tmplist */
3052 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3053 list_move_tail(&page->lru, &tmplist);
3055 /* now try and add more pages onto the request */
3056 expected_index = page->index + 1;
3057 list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
3058 /* discontinuity ? */
3059 if (page->index != expected_index)
3062 /* would this page push the read over the rsize? */
3063 if (bytes + PAGE_CACHE_SIZE > rsize)
3066 __set_page_locked(page);
3067 if (add_to_page_cache_locked(page, mapping,
3068 page->index, GFP_KERNEL)) {
3069 __clear_page_locked(page);
3072 list_move_tail(&page->lru, &tmplist);
3073 bytes += PAGE_CACHE_SIZE;
3078 rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
3080 /* best to give up if we're out of mem */
3081 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3082 list_del(&page->lru);
3083 lru_cache_add_file(page);
3085 page_cache_release(page);
3091 rdata->cfile = cifsFileInfo_get(open_file);
3092 rdata->mapping = mapping;
3093 rdata->offset = offset;
3094 rdata->bytes = bytes;
3096 rdata->marshal_iov = cifs_readpages_marshal_iov;
3097 list_splice_init(&tmplist, &rdata->pages);
3099 rc = cifs_retry_async_readv(rdata);
3101 list_for_each_entry_safe(page, tpage, &rdata->pages,
3103 list_del(&page->lru);
3104 lru_cache_add_file(page);
3106 page_cache_release(page);
3108 kref_put(&rdata->refcount, cifs_readdata_release);
3112 kref_put(&rdata->refcount, cifs_readdata_release);
3118 static int cifs_readpage_worker(struct file *file, struct page *page,
3124 /* Is the page cached? */
3125 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
3129 page_cache_get(page);
3130 read_data = kmap(page);
3131 /* for reads over a certain size could initiate async read ahead */
3133 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
3138 cFYI(1, "Bytes read %d", rc);
3140 file->f_path.dentry->d_inode->i_atime =
3141 current_fs_time(file->f_path.dentry->d_inode->i_sb);
3143 if (PAGE_CACHE_SIZE > rc)
3144 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
3146 flush_dcache_page(page);
3147 SetPageUptodate(page);
3149 /* send this page to the cache */
3150 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
3156 page_cache_release(page);
3162 static int cifs_readpage(struct file *file, struct page *page)
3164 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3170 if (file->private_data == NULL) {
3176 cFYI(1, "readpage %p at offset %d 0x%x",
3177 page, (int)offset, (int)offset);
3179 rc = cifs_readpage_worker(file, page, &offset);
3187 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
3189 struct cifsFileInfo *open_file;
3191 spin_lock(&cifs_file_list_lock);
3192 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
3193 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
3194 spin_unlock(&cifs_file_list_lock);
3198 spin_unlock(&cifs_file_list_lock);
3202 /* We do not want to update the file size from server for inodes
3203 open for write - to avoid races with writepage extending
3204 the file - in the future we could consider allowing
3205 refreshing the inode only on increases in the file size
3206 but this is tricky to do without racing with writebehind
3207 page caching in the current Linux kernel design */
3208 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
3213 if (is_inode_writable(cifsInode)) {
3214 /* This inode is open for write at least once */
3215 struct cifs_sb_info *cifs_sb;
3217 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
3218 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
3219 /* since no page cache to corrupt on directio
3220 we can change size safely */
3224 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
3232 static int cifs_write_begin(struct file *file, struct address_space *mapping,
3233 loff_t pos, unsigned len, unsigned flags,
3234 struct page **pagep, void **fsdata)
3236 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
3237 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
3238 loff_t page_start = pos & PAGE_MASK;
3243 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
3245 page = grab_cache_page_write_begin(mapping, index, flags);
3251 if (PageUptodate(page))
3255 * If we write a full page it will be up to date, no need to read from
3256 * the server. If the write is short, we'll end up doing a sync write
3259 if (len == PAGE_CACHE_SIZE)
3263 * optimize away the read when we have an oplock, and we're not
3264 * expecting to use any of the data we'd be reading in. That
3265 * is, when the page lies beyond the EOF, or straddles the EOF
3266 * and the write will cover all of the existing data.
3268 if (CIFS_I(mapping->host)->clientCanCacheRead) {
3269 i_size = i_size_read(mapping->host);
3270 if (page_start >= i_size ||
3271 (offset == 0 && (pos + len) >= i_size)) {
3272 zero_user_segments(page, 0, offset,
3276 * PageChecked means that the parts of the page
3277 * to which we're not writing are considered up
3278 * to date. Once the data is copied to the
3279 * page, it can be set uptodate.
3281 SetPageChecked(page);
3286 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
3288 * might as well read a page, it is fast enough. If we get
3289 * an error, we don't need to return it. cifs_write_end will
3290 * do a sync write instead since PG_uptodate isn't set.
3292 cifs_readpage_worker(file, page, &page_start);
3294 /* we could try using another file handle if there is one -
3295 but how would we lock it to prevent close of that handle
3296 racing with this read? In any case
3297 this will be written out by write_end so is fine */
3304 static int cifs_release_page(struct page *page, gfp_t gfp)
3306 if (PagePrivate(page))
3309 return cifs_fscache_release_page(page, gfp);
3312 static void cifs_invalidate_page(struct page *page, unsigned long offset)
3314 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
3317 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
3320 static int cifs_launder_page(struct page *page)
3323 loff_t range_start = page_offset(page);
3324 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
3325 struct writeback_control wbc = {
3326 .sync_mode = WB_SYNC_ALL,
3328 .range_start = range_start,
3329 .range_end = range_end,
3332 cFYI(1, "Launder page: %p", page);
3334 if (clear_page_dirty_for_io(page))
3335 rc = cifs_writepage_locked(page, &wbc);
3337 cifs_fscache_invalidate_page(page, page->mapping->host);
3341 void cifs_oplock_break(struct work_struct *work)
3343 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
3345 struct inode *inode = cfile->dentry->d_inode;
3346 struct cifsInodeInfo *cinode = CIFS_I(inode);
3349 if (inode && S_ISREG(inode->i_mode)) {
3350 if (cinode->clientCanCacheRead)
3351 break_lease(inode, O_RDONLY);
3353 break_lease(inode, O_WRONLY);
3354 rc = filemap_fdatawrite(inode->i_mapping);
3355 if (cinode->clientCanCacheRead == 0) {
3356 rc = filemap_fdatawait(inode->i_mapping);
3357 mapping_set_error(inode->i_mapping, rc);
3358 invalidate_remote_inode(inode);
3360 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
3363 rc = cifs_push_locks(cfile);
3365 cERROR(1, "Push locks rc = %d", rc);
3368 * releasing stale oplock after recent reconnect of smb session using
3369 * a now incorrect file handle is not a data integrity issue but do
3370 * not bother sending an oplock release if session to server still is
3371 * disconnected since oplock already released by the server
3373 if (!cfile->oplock_break_cancelled) {
3374 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->fid.netfid,
3375 current->tgid, 0, 0, 0, 0,
3376 LOCKING_ANDX_OPLOCK_RELEASE, false,
3377 cinode->clientCanCacheRead ? 1 : 0);
3378 cFYI(1, "Oplock release rc = %d", rc);
3382 const struct address_space_operations cifs_addr_ops = {
3383 .readpage = cifs_readpage,
3384 .readpages = cifs_readpages,
3385 .writepage = cifs_writepage,
3386 .writepages = cifs_writepages,
3387 .write_begin = cifs_write_begin,
3388 .write_end = cifs_write_end,
3389 .set_page_dirty = __set_page_dirty_nobuffers,
3390 .releasepage = cifs_release_page,
3391 .invalidatepage = cifs_invalidate_page,
3392 .launder_page = cifs_launder_page,
3396 * cifs_readpages requires the server to support a buffer large enough to
3397 * contain the header plus one complete page of data. Otherwise, we need
3398 * to leave cifs_readpages out of the address space operations.
3400 const struct address_space_operations cifs_addr_ops_smallbuf = {
3401 .readpage = cifs_readpage,
3402 .writepage = cifs_writepage,
3403 .writepages = cifs_writepages,
3404 .write_begin = cifs_write_begin,
3405 .write_end = cifs_write_end,
3406 .set_page_dirty = __set_page_dirty_nobuffers,
3407 .releasepage = cifs_release_page,
3408 .invalidatepage = cifs_invalidate_page,
3409 .launder_page = cifs_launder_page,