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"
47 static inline int cifs_convert_flags(unsigned int flags)
49 if ((flags & O_ACCMODE) == O_RDONLY)
51 else if ((flags & O_ACCMODE) == O_WRONLY)
53 else if ((flags & O_ACCMODE) == O_RDWR) {
54 /* GENERIC_ALL is too much permission to request
55 can cause unnecessary access denied on create */
56 /* return GENERIC_ALL; */
57 return (GENERIC_READ | GENERIC_WRITE);
60 return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
61 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
65 static u32 cifs_posix_convert_flags(unsigned int flags)
69 if ((flags & O_ACCMODE) == O_RDONLY)
70 posix_flags = SMB_O_RDONLY;
71 else if ((flags & O_ACCMODE) == O_WRONLY)
72 posix_flags = SMB_O_WRONLY;
73 else if ((flags & O_ACCMODE) == O_RDWR)
74 posix_flags = SMB_O_RDWR;
76 if (flags & O_CREAT) {
77 posix_flags |= SMB_O_CREAT;
79 posix_flags |= SMB_O_EXCL;
80 } else if (flags & O_EXCL)
81 cifs_dbg(FYI, "Application %s pid %d has incorrectly set O_EXCL flag but not O_CREAT on file open. Ignoring O_EXCL\n",
82 current->comm, current->tgid);
85 posix_flags |= SMB_O_TRUNC;
86 /* be safe and imply O_SYNC for O_DSYNC */
88 posix_flags |= SMB_O_SYNC;
89 if (flags & O_DIRECTORY)
90 posix_flags |= SMB_O_DIRECTORY;
91 if (flags & O_NOFOLLOW)
92 posix_flags |= SMB_O_NOFOLLOW;
94 posix_flags |= SMB_O_DIRECT;
99 static inline int cifs_get_disposition(unsigned int flags)
101 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
103 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
104 return FILE_OVERWRITE_IF;
105 else if ((flags & O_CREAT) == O_CREAT)
107 else if ((flags & O_TRUNC) == O_TRUNC)
108 return FILE_OVERWRITE;
113 int cifs_posix_open(char *full_path, struct inode **pinode,
114 struct super_block *sb, int mode, unsigned int f_flags,
115 __u32 *poplock, __u16 *pnetfid, unsigned int xid)
118 FILE_UNIX_BASIC_INFO *presp_data;
119 __u32 posix_flags = 0;
120 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
121 struct cifs_fattr fattr;
122 struct tcon_link *tlink;
123 struct cifs_tcon *tcon;
125 cifs_dbg(FYI, "posix open %s\n", full_path);
127 presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
128 if (presp_data == NULL)
131 tlink = cifs_sb_tlink(cifs_sb);
137 tcon = tlink_tcon(tlink);
138 mode &= ~current_umask();
140 posix_flags = cifs_posix_convert_flags(f_flags);
141 rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
142 poplock, full_path, cifs_sb->local_nls,
143 cifs_sb->mnt_cifs_flags &
144 CIFS_MOUNT_MAP_SPECIAL_CHR);
145 cifs_put_tlink(tlink);
150 if (presp_data->Type == cpu_to_le32(-1))
151 goto posix_open_ret; /* open ok, caller does qpathinfo */
154 goto posix_open_ret; /* caller does not need info */
156 cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
158 /* get new inode and set it up */
159 if (*pinode == NULL) {
160 cifs_fill_uniqueid(sb, &fattr);
161 *pinode = cifs_iget(sb, &fattr);
167 cifs_fattr_to_inode(*pinode, &fattr);
176 cifs_nt_open(char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
177 struct cifs_tcon *tcon, unsigned int f_flags, __u32 *oplock,
178 struct cifs_fid *fid, unsigned int xid)
183 int create_options = CREATE_NOT_DIR;
185 struct TCP_Server_Info *server = tcon->ses->server;
186 struct cifs_open_parms oparms;
188 if (!server->ops->open)
191 desired_access = cifs_convert_flags(f_flags);
193 /*********************************************************************
194 * open flag mapping table:
196 * POSIX Flag CIFS Disposition
197 * ---------- ----------------
198 * O_CREAT FILE_OPEN_IF
199 * O_CREAT | O_EXCL FILE_CREATE
200 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
201 * O_TRUNC FILE_OVERWRITE
202 * none of the above FILE_OPEN
204 * Note that there is not a direct match between disposition
205 * FILE_SUPERSEDE (ie create whether or not file exists although
206 * O_CREAT | O_TRUNC is similar but truncates the existing
207 * file rather than creating a new file as FILE_SUPERSEDE does
208 * (which uses the attributes / metadata passed in on open call)
210 *? O_SYNC is a reasonable match to CIFS writethrough flag
211 *? and the read write flags match reasonably. O_LARGEFILE
212 *? is irrelevant because largefile support is always used
213 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
214 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
215 *********************************************************************/
217 disposition = cifs_get_disposition(f_flags);
219 /* BB pass O_SYNC flag through on file attributes .. BB */
221 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
225 if (backup_cred(cifs_sb))
226 create_options |= CREATE_OPEN_BACKUP_INTENT;
229 oparms.cifs_sb = cifs_sb;
230 oparms.desired_access = desired_access;
231 oparms.create_options = create_options;
232 oparms.disposition = disposition;
233 oparms.path = full_path;
235 oparms.reconnect = false;
237 rc = server->ops->open(xid, &oparms, oplock, buf);
243 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
246 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
255 cifs_has_mand_locks(struct cifsInodeInfo *cinode)
257 struct cifs_fid_locks *cur;
258 bool has_locks = false;
260 down_read(&cinode->lock_sem);
261 list_for_each_entry(cur, &cinode->llist, llist) {
262 if (!list_empty(&cur->locks)) {
267 up_read(&cinode->lock_sem);
271 struct cifsFileInfo *
272 cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
273 struct tcon_link *tlink, __u32 oplock)
275 struct dentry *dentry = file->f_path.dentry;
276 struct inode *inode = dentry->d_inode;
277 struct cifsInodeInfo *cinode = CIFS_I(inode);
278 struct cifsFileInfo *cfile;
279 struct cifs_fid_locks *fdlocks;
280 struct cifs_tcon *tcon = tlink_tcon(tlink);
281 struct TCP_Server_Info *server = tcon->ses->server;
283 cfile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
287 fdlocks = kzalloc(sizeof(struct cifs_fid_locks), GFP_KERNEL);
293 INIT_LIST_HEAD(&fdlocks->locks);
294 fdlocks->cfile = cfile;
295 cfile->llist = fdlocks;
296 down_write(&cinode->lock_sem);
297 list_add(&fdlocks->llist, &cinode->llist);
298 up_write(&cinode->lock_sem);
301 cfile->pid = current->tgid;
302 cfile->uid = current_fsuid();
303 cfile->dentry = dget(dentry);
304 cfile->f_flags = file->f_flags;
305 cfile->invalidHandle = false;
306 cfile->tlink = cifs_get_tlink(tlink);
307 INIT_WORK(&cfile->oplock_break, cifs_oplock_break);
308 mutex_init(&cfile->fh_mutex);
310 cifs_sb_active(inode->i_sb);
313 * If the server returned a read oplock and we have mandatory brlocks,
314 * set oplock level to None.
316 if (server->ops->is_read_op(oplock) && cifs_has_mand_locks(cinode)) {
317 cifs_dbg(FYI, "Reset oplock val from read to None due to mand locks\n");
321 spin_lock(&cifs_file_list_lock);
322 if (fid->pending_open->oplock != CIFS_OPLOCK_NO_CHANGE && oplock)
323 oplock = fid->pending_open->oplock;
324 list_del(&fid->pending_open->olist);
326 fid->purge_cache = false;
327 server->ops->set_fid(cfile, fid, oplock);
329 list_add(&cfile->tlist, &tcon->openFileList);
330 /* if readable file instance put first in list*/
331 if (file->f_mode & FMODE_READ)
332 list_add(&cfile->flist, &cinode->openFileList);
334 list_add_tail(&cfile->flist, &cinode->openFileList);
335 spin_unlock(&cifs_file_list_lock);
337 if (fid->purge_cache)
338 cifs_zap_mapping(inode);
340 file->private_data = cfile;
344 struct cifsFileInfo *
345 cifsFileInfo_get(struct cifsFileInfo *cifs_file)
347 spin_lock(&cifs_file_list_lock);
348 cifsFileInfo_get_locked(cifs_file);
349 spin_unlock(&cifs_file_list_lock);
354 * Release a reference on the file private data. This may involve closing
355 * the filehandle out on the server. Must be called without holding
356 * cifs_file_list_lock.
358 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
360 struct inode *inode = cifs_file->dentry->d_inode;
361 struct cifs_tcon *tcon = tlink_tcon(cifs_file->tlink);
362 struct TCP_Server_Info *server = tcon->ses->server;
363 struct cifsInodeInfo *cifsi = CIFS_I(inode);
364 struct super_block *sb = inode->i_sb;
365 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
366 struct cifsLockInfo *li, *tmp;
368 struct cifs_pending_open open;
370 spin_lock(&cifs_file_list_lock);
371 if (--cifs_file->count > 0) {
372 spin_unlock(&cifs_file_list_lock);
376 if (server->ops->get_lease_key)
377 server->ops->get_lease_key(inode, &fid);
379 /* store open in pending opens to make sure we don't miss lease break */
380 cifs_add_pending_open_locked(&fid, cifs_file->tlink, &open);
382 /* remove it from the lists */
383 list_del(&cifs_file->flist);
384 list_del(&cifs_file->tlist);
386 if (list_empty(&cifsi->openFileList)) {
387 cifs_dbg(FYI, "closing last open instance for inode %p\n",
388 cifs_file->dentry->d_inode);
390 * In strict cache mode we need invalidate mapping on the last
391 * close because it may cause a error when we open this file
392 * again and get at least level II oplock.
394 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
395 set_bit(CIFS_INO_INVALID_MAPPING, &cifsi->flags);
396 cifs_set_oplock_level(cifsi, 0);
398 spin_unlock(&cifs_file_list_lock);
400 cancel_work_sync(&cifs_file->oplock_break);
402 if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
403 struct TCP_Server_Info *server = tcon->ses->server;
407 if (server->ops->close)
408 server->ops->close(xid, tcon, &cifs_file->fid);
412 cifs_del_pending_open(&open);
415 * Delete any outstanding lock records. We'll lose them when the file
418 down_write(&cifsi->lock_sem);
419 list_for_each_entry_safe(li, tmp, &cifs_file->llist->locks, llist) {
420 list_del(&li->llist);
421 cifs_del_lock_waiters(li);
424 list_del(&cifs_file->llist->llist);
425 kfree(cifs_file->llist);
426 up_write(&cifsi->lock_sem);
428 cifs_put_tlink(cifs_file->tlink);
429 dput(cifs_file->dentry);
430 cifs_sb_deactive(sb);
434 int cifs_open(struct inode *inode, struct file *file)
440 struct cifs_sb_info *cifs_sb;
441 struct TCP_Server_Info *server;
442 struct cifs_tcon *tcon;
443 struct tcon_link *tlink;
444 struct cifsFileInfo *cfile = NULL;
445 char *full_path = NULL;
446 bool posix_open_ok = false;
448 struct cifs_pending_open open;
452 cifs_sb = CIFS_SB(inode->i_sb);
453 tlink = cifs_sb_tlink(cifs_sb);
456 return PTR_ERR(tlink);
458 tcon = tlink_tcon(tlink);
459 server = tcon->ses->server;
461 full_path = build_path_from_dentry(file->f_path.dentry);
462 if (full_path == NULL) {
467 cifs_dbg(FYI, "inode = 0x%p file flags are 0x%x for %s\n",
468 inode, file->f_flags, full_path);
475 if (!tcon->broken_posix_open && tcon->unix_ext &&
476 cap_unix(tcon->ses) && (CIFS_UNIX_POSIX_PATH_OPS_CAP &
477 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
478 /* can not refresh inode info since size could be stale */
479 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
480 cifs_sb->mnt_file_mode /* ignored */,
481 file->f_flags, &oplock, &fid.netfid, xid);
483 cifs_dbg(FYI, "posix open succeeded\n");
484 posix_open_ok = true;
485 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
486 if (tcon->ses->serverNOS)
487 cifs_dbg(VFS, "server %s of type %s returned unexpected error on SMB posix open, disabling posix open support. Check if server update available.\n",
488 tcon->ses->serverName,
489 tcon->ses->serverNOS);
490 tcon->broken_posix_open = true;
491 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
492 (rc != -EOPNOTSUPP)) /* path not found or net err */
495 * Else fallthrough to retry open the old way on network i/o
500 if (server->ops->get_lease_key)
501 server->ops->get_lease_key(inode, &fid);
503 cifs_add_pending_open(&fid, tlink, &open);
505 if (!posix_open_ok) {
506 if (server->ops->get_lease_key)
507 server->ops->get_lease_key(inode, &fid);
509 rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
510 file->f_flags, &oplock, &fid, xid);
512 cifs_del_pending_open(&open);
517 cfile = cifs_new_fileinfo(&fid, file, tlink, oplock);
519 if (server->ops->close)
520 server->ops->close(xid, tcon, &fid);
521 cifs_del_pending_open(&open);
526 cifs_fscache_set_inode_cookie(inode, file);
528 if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
530 * Time to set mode which we can not set earlier due to
531 * problems creating new read-only files.
533 struct cifs_unix_set_info_args args = {
534 .mode = inode->i_mode,
535 .uid = INVALID_UID, /* no change */
536 .gid = INVALID_GID, /* no change */
537 .ctime = NO_CHANGE_64,
538 .atime = NO_CHANGE_64,
539 .mtime = NO_CHANGE_64,
542 CIFSSMBUnixSetFileInfo(xid, tcon, &args, fid.netfid,
549 cifs_put_tlink(tlink);
553 static int cifs_push_posix_locks(struct cifsFileInfo *cfile);
556 * Try to reacquire byte range locks that were released when session
557 * to server was lost.
560 cifs_relock_file(struct cifsFileInfo *cfile)
562 struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
563 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
564 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
567 down_read(&cinode->lock_sem);
568 if (cinode->can_cache_brlcks) {
569 /* can cache locks - no need to relock */
570 up_read(&cinode->lock_sem);
574 if (cap_unix(tcon->ses) &&
575 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
576 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
577 rc = cifs_push_posix_locks(cfile);
579 rc = tcon->ses->server->ops->push_mand_locks(cfile);
581 up_read(&cinode->lock_sem);
586 cifs_reopen_file(struct cifsFileInfo *cfile, bool can_flush)
591 struct cifs_sb_info *cifs_sb;
592 struct cifs_tcon *tcon;
593 struct TCP_Server_Info *server;
594 struct cifsInodeInfo *cinode;
596 char *full_path = NULL;
598 int disposition = FILE_OPEN;
599 int create_options = CREATE_NOT_DIR;
600 struct cifs_open_parms oparms;
603 mutex_lock(&cfile->fh_mutex);
604 if (!cfile->invalidHandle) {
605 mutex_unlock(&cfile->fh_mutex);
611 inode = cfile->dentry->d_inode;
612 cifs_sb = CIFS_SB(inode->i_sb);
613 tcon = tlink_tcon(cfile->tlink);
614 server = tcon->ses->server;
617 * Can not grab rename sem here because various ops, including those
618 * that already have the rename sem can end up causing writepage to get
619 * called and if the server was down that means we end up here, and we
620 * can never tell if the caller already has the rename_sem.
622 full_path = build_path_from_dentry(cfile->dentry);
623 if (full_path == NULL) {
625 mutex_unlock(&cfile->fh_mutex);
630 cifs_dbg(FYI, "inode = 0x%p file flags 0x%x for %s\n",
631 inode, cfile->f_flags, full_path);
633 if (tcon->ses->server->oplocks)
638 if (tcon->unix_ext && cap_unix(tcon->ses) &&
639 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
640 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
642 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
643 * original open. Must mask them off for a reopen.
645 unsigned int oflags = cfile->f_flags &
646 ~(O_CREAT | O_EXCL | O_TRUNC);
648 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
649 cifs_sb->mnt_file_mode /* ignored */,
650 oflags, &oplock, &cfile->fid.netfid, xid);
652 cifs_dbg(FYI, "posix reopen succeeded\n");
653 oparms.reconnect = true;
657 * fallthrough to retry open the old way on errors, especially
658 * in the reconnect path it is important to retry hard
662 desired_access = cifs_convert_flags(cfile->f_flags);
664 if (backup_cred(cifs_sb))
665 create_options |= CREATE_OPEN_BACKUP_INTENT;
667 if (server->ops->get_lease_key)
668 server->ops->get_lease_key(inode, &cfile->fid);
671 oparms.cifs_sb = cifs_sb;
672 oparms.desired_access = desired_access;
673 oparms.create_options = create_options;
674 oparms.disposition = disposition;
675 oparms.path = full_path;
676 oparms.fid = &cfile->fid;
677 oparms.reconnect = true;
680 * Can not refresh inode by passing in file_info buf to be returned by
681 * ops->open and then calling get_inode_info with returned buf since
682 * file might have write behind data that needs to be flushed and server
683 * version of file size can be stale. If we knew for sure that inode was
684 * not dirty locally we could do this.
686 rc = server->ops->open(xid, &oparms, &oplock, NULL);
687 if (rc == -ENOENT && oparms.reconnect == false) {
688 /* durable handle timeout is expired - open the file again */
689 rc = server->ops->open(xid, &oparms, &oplock, NULL);
690 /* indicate that we need to relock the file */
691 oparms.reconnect = true;
695 mutex_unlock(&cfile->fh_mutex);
696 cifs_dbg(FYI, "cifs_reopen returned 0x%x\n", rc);
697 cifs_dbg(FYI, "oplock: %d\n", oplock);
698 goto reopen_error_exit;
702 cfile->invalidHandle = false;
703 mutex_unlock(&cfile->fh_mutex);
704 cinode = CIFS_I(inode);
707 rc = filemap_write_and_wait(inode->i_mapping);
708 mapping_set_error(inode->i_mapping, rc);
711 rc = cifs_get_inode_info_unix(&inode, full_path,
714 rc = cifs_get_inode_info(&inode, full_path, NULL,
715 inode->i_sb, xid, NULL);
718 * Else we are writing out data to server already and could deadlock if
719 * we tried to flush data, and since we do not know if we have data that
720 * would invalidate the current end of file on the server we can not go
721 * to the server to get the new inode info.
724 server->ops->set_fid(cfile, &cfile->fid, oplock);
725 if (oparms.reconnect)
726 cifs_relock_file(cfile);
734 int cifs_close(struct inode *inode, struct file *file)
736 if (file->private_data != NULL) {
737 cifsFileInfo_put(file->private_data);
738 file->private_data = NULL;
741 /* return code from the ->release op is always ignored */
745 int cifs_closedir(struct inode *inode, struct file *file)
749 struct cifsFileInfo *cfile = file->private_data;
750 struct cifs_tcon *tcon;
751 struct TCP_Server_Info *server;
754 cifs_dbg(FYI, "Closedir inode = 0x%p\n", inode);
760 tcon = tlink_tcon(cfile->tlink);
761 server = tcon->ses->server;
763 cifs_dbg(FYI, "Freeing private data in close dir\n");
764 spin_lock(&cifs_file_list_lock);
765 if (!cfile->srch_inf.endOfSearch && !cfile->invalidHandle) {
766 cfile->invalidHandle = true;
767 spin_unlock(&cifs_file_list_lock);
768 if (server->ops->close_dir)
769 rc = server->ops->close_dir(xid, tcon, &cfile->fid);
772 cifs_dbg(FYI, "Closing uncompleted readdir with rc %d\n", rc);
773 /* not much we can do if it fails anyway, ignore rc */
776 spin_unlock(&cifs_file_list_lock);
778 buf = cfile->srch_inf.ntwrk_buf_start;
780 cifs_dbg(FYI, "closedir free smb buf in srch struct\n");
781 cfile->srch_inf.ntwrk_buf_start = NULL;
782 if (cfile->srch_inf.smallBuf)
783 cifs_small_buf_release(buf);
785 cifs_buf_release(buf);
788 cifs_put_tlink(cfile->tlink);
789 kfree(file->private_data);
790 file->private_data = NULL;
791 /* BB can we lock the filestruct while this is going on? */
796 static struct cifsLockInfo *
797 cifs_lock_init(__u64 offset, __u64 length, __u8 type)
799 struct cifsLockInfo *lock =
800 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
803 lock->offset = offset;
804 lock->length = length;
806 lock->pid = current->tgid;
807 INIT_LIST_HEAD(&lock->blist);
808 init_waitqueue_head(&lock->block_q);
813 cifs_del_lock_waiters(struct cifsLockInfo *lock)
815 struct cifsLockInfo *li, *tmp;
816 list_for_each_entry_safe(li, tmp, &lock->blist, blist) {
817 list_del_init(&li->blist);
818 wake_up(&li->block_q);
822 #define CIFS_LOCK_OP 0
823 #define CIFS_READ_OP 1
824 #define CIFS_WRITE_OP 2
826 /* @rw_check : 0 - no op, 1 - read, 2 - write */
828 cifs_find_fid_lock_conflict(struct cifs_fid_locks *fdlocks, __u64 offset,
829 __u64 length, __u8 type, struct cifsFileInfo *cfile,
830 struct cifsLockInfo **conf_lock, int rw_check)
832 struct cifsLockInfo *li;
833 struct cifsFileInfo *cur_cfile = fdlocks->cfile;
834 struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
836 list_for_each_entry(li, &fdlocks->locks, llist) {
837 if (offset + length <= li->offset ||
838 offset >= li->offset + li->length)
840 if (rw_check != CIFS_LOCK_OP && current->tgid == li->pid &&
841 server->ops->compare_fids(cfile, cur_cfile)) {
842 /* shared lock prevents write op through the same fid */
843 if (!(li->type & server->vals->shared_lock_type) ||
844 rw_check != CIFS_WRITE_OP)
847 if ((type & server->vals->shared_lock_type) &&
848 ((server->ops->compare_fids(cfile, cur_cfile) &&
849 current->tgid == li->pid) || type == li->type))
859 cifs_find_lock_conflict(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
860 __u8 type, struct cifsLockInfo **conf_lock,
864 struct cifs_fid_locks *cur;
865 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
867 list_for_each_entry(cur, &cinode->llist, llist) {
868 rc = cifs_find_fid_lock_conflict(cur, offset, length, type,
869 cfile, conf_lock, rw_check);
878 * Check if there is another lock that prevents us to set the lock (mandatory
879 * style). If such a lock exists, update the flock structure with its
880 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
881 * or leave it the same if we can't. Returns 0 if we don't need to request to
882 * the server or 1 otherwise.
885 cifs_lock_test(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
886 __u8 type, struct file_lock *flock)
889 struct cifsLockInfo *conf_lock;
890 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
891 struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
894 down_read(&cinode->lock_sem);
896 exist = cifs_find_lock_conflict(cfile, offset, length, type,
897 &conf_lock, CIFS_LOCK_OP);
899 flock->fl_start = conf_lock->offset;
900 flock->fl_end = conf_lock->offset + conf_lock->length - 1;
901 flock->fl_pid = conf_lock->pid;
902 if (conf_lock->type & server->vals->shared_lock_type)
903 flock->fl_type = F_RDLCK;
905 flock->fl_type = F_WRLCK;
906 } else if (!cinode->can_cache_brlcks)
909 flock->fl_type = F_UNLCK;
911 up_read(&cinode->lock_sem);
916 cifs_lock_add(struct cifsFileInfo *cfile, struct cifsLockInfo *lock)
918 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
919 down_write(&cinode->lock_sem);
920 list_add_tail(&lock->llist, &cfile->llist->locks);
921 up_write(&cinode->lock_sem);
925 * Set the byte-range lock (mandatory style). Returns:
926 * 1) 0, if we set the lock and don't need to request to the server;
927 * 2) 1, if no locks prevent us but we need to request to the server;
928 * 3) -EACCESS, if there is a lock that prevents us and wait is false.
931 cifs_lock_add_if(struct cifsFileInfo *cfile, struct cifsLockInfo *lock,
934 struct cifsLockInfo *conf_lock;
935 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
941 down_write(&cinode->lock_sem);
943 exist = cifs_find_lock_conflict(cfile, lock->offset, lock->length,
944 lock->type, &conf_lock, CIFS_LOCK_OP);
945 if (!exist && cinode->can_cache_brlcks) {
946 list_add_tail(&lock->llist, &cfile->llist->locks);
947 up_write(&cinode->lock_sem);
956 list_add_tail(&lock->blist, &conf_lock->blist);
957 up_write(&cinode->lock_sem);
958 rc = wait_event_interruptible(lock->block_q,
959 (lock->blist.prev == &lock->blist) &&
960 (lock->blist.next == &lock->blist));
963 down_write(&cinode->lock_sem);
964 list_del_init(&lock->blist);
967 up_write(&cinode->lock_sem);
972 * Check if there is another lock that prevents us to set the lock (posix
973 * style). If such a lock exists, update the flock structure with its
974 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
975 * or leave it the same if we can't. Returns 0 if we don't need to request to
976 * the server or 1 otherwise.
979 cifs_posix_lock_test(struct file *file, struct file_lock *flock)
982 struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
983 unsigned char saved_type = flock->fl_type;
985 if ((flock->fl_flags & FL_POSIX) == 0)
988 down_read(&cinode->lock_sem);
989 posix_test_lock(file, flock);
991 if (flock->fl_type == F_UNLCK && !cinode->can_cache_brlcks) {
992 flock->fl_type = saved_type;
996 up_read(&cinode->lock_sem);
1001 * Set the byte-range lock (posix style). Returns:
1002 * 1) 0, if we set the lock and don't need to request to the server;
1003 * 2) 1, if we need to request to the server;
1004 * 3) <0, if the error occurs while setting the lock.
1007 cifs_posix_lock_set(struct file *file, struct file_lock *flock)
1009 struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
1012 if ((flock->fl_flags & FL_POSIX) == 0)
1016 down_write(&cinode->lock_sem);
1017 if (!cinode->can_cache_brlcks) {
1018 up_write(&cinode->lock_sem);
1022 rc = posix_lock_file(file, flock, NULL);
1023 up_write(&cinode->lock_sem);
1024 if (rc == FILE_LOCK_DEFERRED) {
1025 rc = wait_event_interruptible(flock->fl_wait, !flock->fl_next);
1028 posix_unblock_lock(flock);
1034 cifs_push_mandatory_locks(struct cifsFileInfo *cfile)
1037 int rc = 0, stored_rc;
1038 struct cifsLockInfo *li, *tmp;
1039 struct cifs_tcon *tcon;
1040 unsigned int num, max_num, max_buf;
1041 LOCKING_ANDX_RANGE *buf, *cur;
1042 int types[] = {LOCKING_ANDX_LARGE_FILES,
1043 LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES};
1047 tcon = tlink_tcon(cfile->tlink);
1050 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1051 * and check it for zero before using.
1053 max_buf = tcon->ses->server->maxBuf;
1059 max_num = (max_buf - sizeof(struct smb_hdr)) /
1060 sizeof(LOCKING_ANDX_RANGE);
1061 buf = kzalloc(max_num * sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1067 for (i = 0; i < 2; i++) {
1070 list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1071 if (li->type != types[i])
1073 cur->Pid = cpu_to_le16(li->pid);
1074 cur->LengthLow = cpu_to_le32((u32)li->length);
1075 cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1076 cur->OffsetLow = cpu_to_le32((u32)li->offset);
1077 cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1078 if (++num == max_num) {
1079 stored_rc = cifs_lockv(xid, tcon,
1081 (__u8)li->type, 0, num,
1092 stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1093 (__u8)types[i], 0, num, buf);
1104 /* copied from fs/locks.c with a name change */
1105 #define cifs_for_each_lock(inode, lockp) \
1106 for (lockp = &inode->i_flock; *lockp != NULL; \
1107 lockp = &(*lockp)->fl_next)
1109 struct lock_to_push {
1110 struct list_head llist;
1119 cifs_push_posix_locks(struct cifsFileInfo *cfile)
1121 struct inode *inode = cfile->dentry->d_inode;
1122 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1123 struct file_lock *flock, **before;
1124 unsigned int count = 0, i = 0;
1125 int rc = 0, xid, type;
1126 struct list_head locks_to_send, *el;
1127 struct lock_to_push *lck, *tmp;
1132 spin_lock(&inode->i_lock);
1133 cifs_for_each_lock(inode, before) {
1134 if ((*before)->fl_flags & FL_POSIX)
1137 spin_unlock(&inode->i_lock);
1139 INIT_LIST_HEAD(&locks_to_send);
1142 * Allocating count locks is enough because no FL_POSIX locks can be
1143 * added to the list while we are holding cinode->lock_sem that
1144 * protects locking operations of this inode.
1146 for (; i < count; i++) {
1147 lck = kmalloc(sizeof(struct lock_to_push), GFP_KERNEL);
1152 list_add_tail(&lck->llist, &locks_to_send);
1155 el = locks_to_send.next;
1156 spin_lock(&inode->i_lock);
1157 cifs_for_each_lock(inode, before) {
1159 if ((flock->fl_flags & FL_POSIX) == 0)
1161 if (el == &locks_to_send) {
1163 * The list ended. We don't have enough allocated
1164 * structures - something is really wrong.
1166 cifs_dbg(VFS, "Can't push all brlocks!\n");
1169 length = 1 + flock->fl_end - flock->fl_start;
1170 if (flock->fl_type == F_RDLCK || flock->fl_type == F_SHLCK)
1174 lck = list_entry(el, struct lock_to_push, llist);
1175 lck->pid = flock->fl_pid;
1176 lck->netfid = cfile->fid.netfid;
1177 lck->length = length;
1179 lck->offset = flock->fl_start;
1182 spin_unlock(&inode->i_lock);
1184 list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1187 stored_rc = CIFSSMBPosixLock(xid, tcon, lck->netfid, lck->pid,
1188 lck->offset, lck->length, NULL,
1192 list_del(&lck->llist);
1200 list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1201 list_del(&lck->llist);
1208 cifs_push_locks(struct cifsFileInfo *cfile)
1210 struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
1211 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
1212 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1215 /* we are going to update can_cache_brlcks here - need a write access */
1216 down_write(&cinode->lock_sem);
1217 if (!cinode->can_cache_brlcks) {
1218 up_write(&cinode->lock_sem);
1222 if (cap_unix(tcon->ses) &&
1223 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1224 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1225 rc = cifs_push_posix_locks(cfile);
1227 rc = tcon->ses->server->ops->push_mand_locks(cfile);
1229 cinode->can_cache_brlcks = false;
1230 up_write(&cinode->lock_sem);
1235 cifs_read_flock(struct file_lock *flock, __u32 *type, int *lock, int *unlock,
1236 bool *wait_flag, struct TCP_Server_Info *server)
1238 if (flock->fl_flags & FL_POSIX)
1239 cifs_dbg(FYI, "Posix\n");
1240 if (flock->fl_flags & FL_FLOCK)
1241 cifs_dbg(FYI, "Flock\n");
1242 if (flock->fl_flags & FL_SLEEP) {
1243 cifs_dbg(FYI, "Blocking lock\n");
1246 if (flock->fl_flags & FL_ACCESS)
1247 cifs_dbg(FYI, "Process suspended by mandatory locking - not implemented yet\n");
1248 if (flock->fl_flags & FL_LEASE)
1249 cifs_dbg(FYI, "Lease on file - not implemented yet\n");
1250 if (flock->fl_flags &
1251 (~(FL_POSIX | FL_FLOCK | FL_SLEEP |
1252 FL_ACCESS | FL_LEASE | FL_CLOSE)))
1253 cifs_dbg(FYI, "Unknown lock flags 0x%x\n", flock->fl_flags);
1255 *type = server->vals->large_lock_type;
1256 if (flock->fl_type == F_WRLCK) {
1257 cifs_dbg(FYI, "F_WRLCK\n");
1258 *type |= server->vals->exclusive_lock_type;
1260 } else if (flock->fl_type == F_UNLCK) {
1261 cifs_dbg(FYI, "F_UNLCK\n");
1262 *type |= server->vals->unlock_lock_type;
1264 /* Check if unlock includes more than one lock range */
1265 } else if (flock->fl_type == F_RDLCK) {
1266 cifs_dbg(FYI, "F_RDLCK\n");
1267 *type |= server->vals->shared_lock_type;
1269 } else if (flock->fl_type == F_EXLCK) {
1270 cifs_dbg(FYI, "F_EXLCK\n");
1271 *type |= server->vals->exclusive_lock_type;
1273 } else if (flock->fl_type == F_SHLCK) {
1274 cifs_dbg(FYI, "F_SHLCK\n");
1275 *type |= server->vals->shared_lock_type;
1278 cifs_dbg(FYI, "Unknown type of lock\n");
1282 cifs_getlk(struct file *file, struct file_lock *flock, __u32 type,
1283 bool wait_flag, bool posix_lck, unsigned int xid)
1286 __u64 length = 1 + flock->fl_end - flock->fl_start;
1287 struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1288 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1289 struct TCP_Server_Info *server = tcon->ses->server;
1290 __u16 netfid = cfile->fid.netfid;
1293 int posix_lock_type;
1295 rc = cifs_posix_lock_test(file, flock);
1299 if (type & server->vals->shared_lock_type)
1300 posix_lock_type = CIFS_RDLCK;
1302 posix_lock_type = CIFS_WRLCK;
1303 rc = CIFSSMBPosixLock(xid, tcon, netfid, current->tgid,
1304 flock->fl_start, length, flock,
1305 posix_lock_type, wait_flag);
1309 rc = cifs_lock_test(cfile, flock->fl_start, length, type, flock);
1313 /* BB we could chain these into one lock request BB */
1314 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length, type,
1317 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1319 flock->fl_type = F_UNLCK;
1321 cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1326 if (type & server->vals->shared_lock_type) {
1327 flock->fl_type = F_WRLCK;
1331 type &= ~server->vals->exclusive_lock_type;
1333 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1334 type | server->vals->shared_lock_type,
1337 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1338 type | server->vals->shared_lock_type, 0, 1, false);
1339 flock->fl_type = F_RDLCK;
1341 cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1344 flock->fl_type = F_WRLCK;
1350 cifs_move_llist(struct list_head *source, struct list_head *dest)
1352 struct list_head *li, *tmp;
1353 list_for_each_safe(li, tmp, source)
1354 list_move(li, dest);
1358 cifs_free_llist(struct list_head *llist)
1360 struct cifsLockInfo *li, *tmp;
1361 list_for_each_entry_safe(li, tmp, llist, llist) {
1362 cifs_del_lock_waiters(li);
1363 list_del(&li->llist);
1369 cifs_unlock_range(struct cifsFileInfo *cfile, struct file_lock *flock,
1372 int rc = 0, stored_rc;
1373 int types[] = {LOCKING_ANDX_LARGE_FILES,
1374 LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES};
1376 unsigned int max_num, num, max_buf;
1377 LOCKING_ANDX_RANGE *buf, *cur;
1378 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1379 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
1380 struct cifsLockInfo *li, *tmp;
1381 __u64 length = 1 + flock->fl_end - flock->fl_start;
1382 struct list_head tmp_llist;
1384 INIT_LIST_HEAD(&tmp_llist);
1387 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1388 * and check it for zero before using.
1390 max_buf = tcon->ses->server->maxBuf;
1394 max_num = (max_buf - sizeof(struct smb_hdr)) /
1395 sizeof(LOCKING_ANDX_RANGE);
1396 buf = kzalloc(max_num * sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1400 down_write(&cinode->lock_sem);
1401 for (i = 0; i < 2; i++) {
1404 list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1405 if (flock->fl_start > li->offset ||
1406 (flock->fl_start + length) <
1407 (li->offset + li->length))
1409 if (current->tgid != li->pid)
1411 if (types[i] != li->type)
1413 if (cinode->can_cache_brlcks) {
1415 * We can cache brlock requests - simply remove
1416 * a lock from the file's list.
1418 list_del(&li->llist);
1419 cifs_del_lock_waiters(li);
1423 cur->Pid = cpu_to_le16(li->pid);
1424 cur->LengthLow = cpu_to_le32((u32)li->length);
1425 cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1426 cur->OffsetLow = cpu_to_le32((u32)li->offset);
1427 cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1429 * We need to save a lock here to let us add it again to
1430 * the file's list if the unlock range request fails on
1433 list_move(&li->llist, &tmp_llist);
1434 if (++num == max_num) {
1435 stored_rc = cifs_lockv(xid, tcon,
1437 li->type, num, 0, buf);
1440 * We failed on the unlock range
1441 * request - add all locks from the tmp
1442 * list to the head of the file's list.
1444 cifs_move_llist(&tmp_llist,
1445 &cfile->llist->locks);
1449 * The unlock range request succeed -
1450 * free the tmp list.
1452 cifs_free_llist(&tmp_llist);
1459 stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1460 types[i], num, 0, buf);
1462 cifs_move_llist(&tmp_llist,
1463 &cfile->llist->locks);
1466 cifs_free_llist(&tmp_llist);
1470 up_write(&cinode->lock_sem);
1476 cifs_setlk(struct file *file, struct file_lock *flock, __u32 type,
1477 bool wait_flag, bool posix_lck, int lock, int unlock,
1481 __u64 length = 1 + flock->fl_end - flock->fl_start;
1482 struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1483 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1484 struct TCP_Server_Info *server = tcon->ses->server;
1485 struct inode *inode = cfile->dentry->d_inode;
1488 int posix_lock_type;
1490 rc = cifs_posix_lock_set(file, flock);
1494 if (type & server->vals->shared_lock_type)
1495 posix_lock_type = CIFS_RDLCK;
1497 posix_lock_type = CIFS_WRLCK;
1500 posix_lock_type = CIFS_UNLCK;
1502 rc = CIFSSMBPosixLock(xid, tcon, cfile->fid.netfid,
1503 current->tgid, flock->fl_start, length,
1504 NULL, posix_lock_type, wait_flag);
1509 struct cifsLockInfo *lock;
1511 lock = cifs_lock_init(flock->fl_start, length, type);
1515 rc = cifs_lock_add_if(cfile, lock, wait_flag);
1524 * Windows 7 server can delay breaking lease from read to None
1525 * if we set a byte-range lock on a file - break it explicitly
1526 * before sending the lock to the server to be sure the next
1527 * read won't conflict with non-overlapted locks due to
1530 if (!CIFS_CACHE_WRITE(CIFS_I(inode)) &&
1531 CIFS_CACHE_READ(CIFS_I(inode))) {
1532 cifs_zap_mapping(inode);
1533 cifs_dbg(FYI, "Set no oplock for inode=%p due to mand locks\n",
1535 CIFS_I(inode)->oplock = 0;
1538 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1539 type, 1, 0, wait_flag);
1545 cifs_lock_add(cfile, lock);
1547 rc = server->ops->mand_unlock_range(cfile, flock, xid);
1550 if (flock->fl_flags & FL_POSIX)
1551 posix_lock_file_wait(file, flock);
1555 int cifs_lock(struct file *file, int cmd, struct file_lock *flock)
1558 int lock = 0, unlock = 0;
1559 bool wait_flag = false;
1560 bool posix_lck = false;
1561 struct cifs_sb_info *cifs_sb;
1562 struct cifs_tcon *tcon;
1563 struct cifsInodeInfo *cinode;
1564 struct cifsFileInfo *cfile;
1571 cifs_dbg(FYI, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld end: %lld\n",
1572 cmd, flock->fl_flags, flock->fl_type,
1573 flock->fl_start, flock->fl_end);
1575 cfile = (struct cifsFileInfo *)file->private_data;
1576 tcon = tlink_tcon(cfile->tlink);
1578 cifs_read_flock(flock, &type, &lock, &unlock, &wait_flag,
1581 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1582 netfid = cfile->fid.netfid;
1583 cinode = CIFS_I(file_inode(file));
1585 if (cap_unix(tcon->ses) &&
1586 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1587 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1590 * BB add code here to normalize offset and length to account for
1591 * negative length which we can not accept over the wire.
1593 if (IS_GETLK(cmd)) {
1594 rc = cifs_getlk(file, flock, type, wait_flag, posix_lck, xid);
1599 if (!lock && !unlock) {
1601 * if no lock or unlock then nothing to do since we do not
1608 rc = cifs_setlk(file, flock, type, wait_flag, posix_lck, lock, unlock,
1615 * update the file size (if needed) after a write. Should be called with
1616 * the inode->i_lock held
1619 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
1620 unsigned int bytes_written)
1622 loff_t end_of_write = offset + bytes_written;
1624 if (end_of_write > cifsi->server_eof)
1625 cifsi->server_eof = end_of_write;
1629 cifs_write(struct cifsFileInfo *open_file, __u32 pid, const char *write_data,
1630 size_t write_size, loff_t *offset)
1633 unsigned int bytes_written = 0;
1634 unsigned int total_written;
1635 struct cifs_sb_info *cifs_sb;
1636 struct cifs_tcon *tcon;
1637 struct TCP_Server_Info *server;
1639 struct dentry *dentry = open_file->dentry;
1640 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
1641 struct cifs_io_parms io_parms;
1643 cifs_sb = CIFS_SB(dentry->d_sb);
1645 cifs_dbg(FYI, "write %zd bytes to offset %lld of %s\n",
1646 write_size, *offset, dentry->d_name.name);
1648 tcon = tlink_tcon(open_file->tlink);
1649 server = tcon->ses->server;
1651 if (!server->ops->sync_write)
1656 for (total_written = 0; write_size > total_written;
1657 total_written += bytes_written) {
1659 while (rc == -EAGAIN) {
1663 if (open_file->invalidHandle) {
1664 /* we could deadlock if we called
1665 filemap_fdatawait from here so tell
1666 reopen_file not to flush data to
1668 rc = cifs_reopen_file(open_file, false);
1673 len = min(server->ops->wp_retry_size(dentry->d_inode),
1674 (unsigned int)write_size - total_written);
1675 /* iov[0] is reserved for smb header */
1676 iov[1].iov_base = (char *)write_data + total_written;
1677 iov[1].iov_len = len;
1679 io_parms.tcon = tcon;
1680 io_parms.offset = *offset;
1681 io_parms.length = len;
1682 rc = server->ops->sync_write(xid, open_file, &io_parms,
1683 &bytes_written, iov, 1);
1685 if (rc || (bytes_written == 0)) {
1693 spin_lock(&dentry->d_inode->i_lock);
1694 cifs_update_eof(cifsi, *offset, bytes_written);
1695 spin_unlock(&dentry->d_inode->i_lock);
1696 *offset += bytes_written;
1700 cifs_stats_bytes_written(tcon, total_written);
1702 if (total_written > 0) {
1703 spin_lock(&dentry->d_inode->i_lock);
1704 if (*offset > dentry->d_inode->i_size)
1705 i_size_write(dentry->d_inode, *offset);
1706 spin_unlock(&dentry->d_inode->i_lock);
1708 mark_inode_dirty_sync(dentry->d_inode);
1710 return total_written;
1713 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1716 struct cifsFileInfo *open_file = NULL;
1717 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1719 /* only filter by fsuid on multiuser mounts */
1720 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1723 spin_lock(&cifs_file_list_lock);
1724 /* we could simply get the first_list_entry since write-only entries
1725 are always at the end of the list but since the first entry might
1726 have a close pending, we go through the whole list */
1727 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1728 if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1730 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1731 if (!open_file->invalidHandle) {
1732 /* found a good file */
1733 /* lock it so it will not be closed on us */
1734 cifsFileInfo_get_locked(open_file);
1735 spin_unlock(&cifs_file_list_lock);
1737 } /* else might as well continue, and look for
1738 another, or simply have the caller reopen it
1739 again rather than trying to fix this handle */
1740 } else /* write only file */
1741 break; /* write only files are last so must be done */
1743 spin_unlock(&cifs_file_list_lock);
1747 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1750 struct cifsFileInfo *open_file, *inv_file = NULL;
1751 struct cifs_sb_info *cifs_sb;
1752 bool any_available = false;
1754 unsigned int refind = 0;
1756 /* Having a null inode here (because mapping->host was set to zero by
1757 the VFS or MM) should not happen but we had reports of on oops (due to
1758 it being zero) during stress testcases so we need to check for it */
1760 if (cifs_inode == NULL) {
1761 cifs_dbg(VFS, "Null inode passed to cifs_writeable_file\n");
1766 cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1768 /* only filter by fsuid on multiuser mounts */
1769 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1772 spin_lock(&cifs_file_list_lock);
1774 if (refind > MAX_REOPEN_ATT) {
1775 spin_unlock(&cifs_file_list_lock);
1778 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1779 if (!any_available && open_file->pid != current->tgid)
1781 if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1783 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1784 if (!open_file->invalidHandle) {
1785 /* found a good writable file */
1786 cifsFileInfo_get_locked(open_file);
1787 spin_unlock(&cifs_file_list_lock);
1791 inv_file = open_file;
1795 /* couldn't find useable FH with same pid, try any available */
1796 if (!any_available) {
1797 any_available = true;
1798 goto refind_writable;
1802 any_available = false;
1803 cifsFileInfo_get_locked(inv_file);
1806 spin_unlock(&cifs_file_list_lock);
1809 rc = cifs_reopen_file(inv_file, false);
1813 spin_lock(&cifs_file_list_lock);
1814 list_move_tail(&inv_file->flist,
1815 &cifs_inode->openFileList);
1816 spin_unlock(&cifs_file_list_lock);
1817 cifsFileInfo_put(inv_file);
1818 spin_lock(&cifs_file_list_lock);
1820 goto refind_writable;
1827 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1829 struct address_space *mapping = page->mapping;
1830 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1833 int bytes_written = 0;
1834 struct inode *inode;
1835 struct cifsFileInfo *open_file;
1837 if (!mapping || !mapping->host)
1840 inode = page->mapping->host;
1842 offset += (loff_t)from;
1843 write_data = kmap(page);
1846 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1851 /* racing with truncate? */
1852 if (offset > mapping->host->i_size) {
1854 return 0; /* don't care */
1857 /* check to make sure that we are not extending the file */
1858 if (mapping->host->i_size - offset < (loff_t)to)
1859 to = (unsigned)(mapping->host->i_size - offset);
1861 open_file = find_writable_file(CIFS_I(mapping->host), false);
1863 bytes_written = cifs_write(open_file, open_file->pid,
1864 write_data, to - from, &offset);
1865 cifsFileInfo_put(open_file);
1866 /* Does mm or vfs already set times? */
1867 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1868 if ((bytes_written > 0) && (offset))
1870 else if (bytes_written < 0)
1873 cifs_dbg(FYI, "No writeable filehandles for inode\n");
1881 static struct cifs_writedata *
1882 wdata_alloc_and_fillpages(pgoff_t tofind, struct address_space *mapping,
1883 pgoff_t end, pgoff_t *index,
1884 unsigned int *found_pages)
1886 unsigned int nr_pages;
1887 struct page **pages;
1888 struct cifs_writedata *wdata;
1890 wdata = cifs_writedata_alloc((unsigned int)tofind,
1891 cifs_writev_complete);
1896 * find_get_pages_tag seems to return a max of 256 on each
1897 * iteration, so we must call it several times in order to
1898 * fill the array or the wsize is effectively limited to
1899 * 256 * PAGE_CACHE_SIZE.
1902 pages = wdata->pages;
1904 nr_pages = find_get_pages_tag(mapping, index,
1905 PAGECACHE_TAG_DIRTY, tofind,
1907 *found_pages += nr_pages;
1910 } while (nr_pages && tofind && *index <= end);
1916 wdata_prepare_pages(struct cifs_writedata *wdata, unsigned int found_pages,
1917 struct address_space *mapping,
1918 struct writeback_control *wbc,
1919 pgoff_t end, pgoff_t *index, pgoff_t *next, bool *done)
1921 unsigned int nr_pages = 0, i;
1924 for (i = 0; i < found_pages; i++) {
1925 page = wdata->pages[i];
1927 * At this point we hold neither mapping->tree_lock nor
1928 * lock on the page itself: the page may be truncated or
1929 * invalidated (changing page->mapping to NULL), or even
1930 * swizzled back from swapper_space to tmpfs file
1936 else if (!trylock_page(page))
1939 if (unlikely(page->mapping != mapping)) {
1944 if (!wbc->range_cyclic && page->index > end) {
1950 if (*next && (page->index != *next)) {
1951 /* Not next consecutive page */
1956 if (wbc->sync_mode != WB_SYNC_NONE)
1957 wait_on_page_writeback(page);
1959 if (PageWriteback(page) ||
1960 !clear_page_dirty_for_io(page)) {
1966 * This actually clears the dirty bit in the radix tree.
1967 * See cifs_writepage() for more commentary.
1969 set_page_writeback(page);
1970 if (page_offset(page) >= i_size_read(mapping->host)) {
1973 end_page_writeback(page);
1977 wdata->pages[i] = page;
1978 *next = page->index + 1;
1982 /* reset index to refind any pages skipped */
1984 *index = wdata->pages[0]->index + 1;
1986 /* put any pages we aren't going to use */
1987 for (i = nr_pages; i < found_pages; i++) {
1988 page_cache_release(wdata->pages[i]);
1989 wdata->pages[i] = NULL;
1996 wdata_send_pages(struct cifs_writedata *wdata, unsigned int nr_pages,
1997 struct address_space *mapping, struct writeback_control *wbc)
2000 struct TCP_Server_Info *server;
2003 wdata->sync_mode = wbc->sync_mode;
2004 wdata->nr_pages = nr_pages;
2005 wdata->offset = page_offset(wdata->pages[0]);
2006 wdata->pagesz = PAGE_CACHE_SIZE;
2007 wdata->tailsz = min(i_size_read(mapping->host) -
2008 page_offset(wdata->pages[nr_pages - 1]),
2009 (loff_t)PAGE_CACHE_SIZE);
2010 wdata->bytes = ((nr_pages - 1) * PAGE_CACHE_SIZE) + wdata->tailsz;
2012 if (wdata->cfile != NULL)
2013 cifsFileInfo_put(wdata->cfile);
2014 wdata->cfile = find_writable_file(CIFS_I(mapping->host), false);
2015 if (!wdata->cfile) {
2016 cifs_dbg(VFS, "No writable handles for inode\n");
2019 wdata->pid = wdata->cfile->pid;
2020 server = tlink_tcon(wdata->cfile->tlink)->ses->server;
2021 rc = server->ops->async_writev(wdata, cifs_writedata_release);
2024 for (i = 0; i < nr_pages; ++i)
2025 unlock_page(wdata->pages[i]);
2030 static int cifs_writepages(struct address_space *mapping,
2031 struct writeback_control *wbc)
2033 struct cifs_sb_info *cifs_sb = CIFS_SB(mapping->host->i_sb);
2034 struct TCP_Server_Info *server;
2035 bool done = false, scanned = false, range_whole = false;
2037 struct cifs_writedata *wdata;
2041 * If wsize is smaller than the page cache size, default to writing
2042 * one page at a time via cifs_writepage
2044 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
2045 return generic_writepages(mapping, wbc);
2047 if (wbc->range_cyclic) {
2048 index = mapping->writeback_index; /* Start from prev offset */
2051 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2052 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2053 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2057 server = cifs_sb_master_tcon(cifs_sb)->ses->server;
2059 while (!done && index <= end) {
2060 unsigned int i, nr_pages, found_pages, wsize, credits;
2061 pgoff_t next = 0, tofind, saved_index = index;
2063 rc = server->ops->wait_mtu_credits(server, cifs_sb->wsize,
2068 tofind = min((wsize / PAGE_CACHE_SIZE) - 1, end - index) + 1;
2070 wdata = wdata_alloc_and_fillpages(tofind, mapping, end, &index,
2074 add_credits_and_wake_if(server, credits, 0);
2078 if (found_pages == 0) {
2079 kref_put(&wdata->refcount, cifs_writedata_release);
2080 add_credits_and_wake_if(server, credits, 0);
2084 nr_pages = wdata_prepare_pages(wdata, found_pages, mapping, wbc,
2085 end, &index, &next, &done);
2087 /* nothing to write? */
2088 if (nr_pages == 0) {
2089 kref_put(&wdata->refcount, cifs_writedata_release);
2090 add_credits_and_wake_if(server, credits, 0);
2094 wdata->credits = credits;
2096 rc = wdata_send_pages(wdata, nr_pages, mapping, wbc);
2098 /* send failure -- clean up the mess */
2100 add_credits_and_wake_if(server, wdata->credits, 0);
2101 for (i = 0; i < nr_pages; ++i) {
2103 redirty_page_for_writepage(wbc,
2106 SetPageError(wdata->pages[i]);
2107 end_page_writeback(wdata->pages[i]);
2108 page_cache_release(wdata->pages[i]);
2111 mapping_set_error(mapping, rc);
2113 kref_put(&wdata->refcount, cifs_writedata_release);
2115 if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN) {
2116 index = saved_index;
2120 wbc->nr_to_write -= nr_pages;
2121 if (wbc->nr_to_write <= 0)
2127 if (!scanned && !done) {
2129 * We hit the last page and there is more work to be done: wrap
2130 * back to the start of the file
2137 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2138 mapping->writeback_index = index;
2144 cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
2150 /* BB add check for wbc flags */
2151 page_cache_get(page);
2152 if (!PageUptodate(page))
2153 cifs_dbg(FYI, "ppw - page not up to date\n");
2156 * Set the "writeback" flag, and clear "dirty" in the radix tree.
2158 * A writepage() implementation always needs to do either this,
2159 * or re-dirty the page with "redirty_page_for_writepage()" in
2160 * the case of a failure.
2162 * Just unlocking the page will cause the radix tree tag-bits
2163 * to fail to update with the state of the page correctly.
2165 set_page_writeback(page);
2167 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
2168 if (rc == -EAGAIN && wbc->sync_mode == WB_SYNC_ALL)
2170 else if (rc == -EAGAIN)
2171 redirty_page_for_writepage(wbc, page);
2175 SetPageUptodate(page);
2176 end_page_writeback(page);
2177 page_cache_release(page);
2182 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
2184 int rc = cifs_writepage_locked(page, wbc);
2189 static int cifs_write_end(struct file *file, struct address_space *mapping,
2190 loff_t pos, unsigned len, unsigned copied,
2191 struct page *page, void *fsdata)
2194 struct inode *inode = mapping->host;
2195 struct cifsFileInfo *cfile = file->private_data;
2196 struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
2199 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2202 pid = current->tgid;
2204 cifs_dbg(FYI, "write_end for page %p from pos %lld with %d bytes\n",
2207 if (PageChecked(page)) {
2209 SetPageUptodate(page);
2210 ClearPageChecked(page);
2211 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
2212 SetPageUptodate(page);
2214 if (!PageUptodate(page)) {
2216 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
2220 /* this is probably better than directly calling
2221 partialpage_write since in this function the file handle is
2222 known which we might as well leverage */
2223 /* BB check if anything else missing out of ppw
2224 such as updating last write time */
2225 page_data = kmap(page);
2226 rc = cifs_write(cfile, pid, page_data + offset, copied, &pos);
2227 /* if (rc < 0) should we set writebehind rc? */
2234 set_page_dirty(page);
2238 spin_lock(&inode->i_lock);
2239 if (pos > inode->i_size)
2240 i_size_write(inode, pos);
2241 spin_unlock(&inode->i_lock);
2245 page_cache_release(page);
2250 int cifs_strict_fsync(struct file *file, loff_t start, loff_t end,
2255 struct cifs_tcon *tcon;
2256 struct TCP_Server_Info *server;
2257 struct cifsFileInfo *smbfile = file->private_data;
2258 struct inode *inode = file_inode(file);
2259 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2261 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2264 mutex_lock(&inode->i_mutex);
2268 cifs_dbg(FYI, "Sync file - name: %s datasync: 0x%x\n",
2269 file->f_path.dentry->d_name.name, datasync);
2271 if (!CIFS_CACHE_READ(CIFS_I(inode))) {
2272 rc = cifs_zap_mapping(inode);
2274 cifs_dbg(FYI, "rc: %d during invalidate phase\n", rc);
2275 rc = 0; /* don't care about it in fsync */
2279 tcon = tlink_tcon(smbfile->tlink);
2280 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2281 server = tcon->ses->server;
2282 if (server->ops->flush)
2283 rc = server->ops->flush(xid, tcon, &smbfile->fid);
2289 mutex_unlock(&inode->i_mutex);
2293 int cifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2297 struct cifs_tcon *tcon;
2298 struct TCP_Server_Info *server;
2299 struct cifsFileInfo *smbfile = file->private_data;
2300 struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2301 struct inode *inode = file->f_mapping->host;
2303 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2306 mutex_lock(&inode->i_mutex);
2310 cifs_dbg(FYI, "Sync file - name: %s datasync: 0x%x\n",
2311 file->f_path.dentry->d_name.name, datasync);
2313 tcon = tlink_tcon(smbfile->tlink);
2314 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2315 server = tcon->ses->server;
2316 if (server->ops->flush)
2317 rc = server->ops->flush(xid, tcon, &smbfile->fid);
2323 mutex_unlock(&inode->i_mutex);
2328 * As file closes, flush all cached write data for this inode checking
2329 * for write behind errors.
2331 int cifs_flush(struct file *file, fl_owner_t id)
2333 struct inode *inode = file_inode(file);
2336 if (file->f_mode & FMODE_WRITE)
2337 rc = filemap_write_and_wait(inode->i_mapping);
2339 cifs_dbg(FYI, "Flush inode %p file %p rc %d\n", inode, file, rc);
2345 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
2350 for (i = 0; i < num_pages; i++) {
2351 pages[i] = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2354 * save number of pages we have already allocated and
2355 * return with ENOMEM error
2364 for (i = 0; i < num_pages; i++)
2371 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
2376 clen = min_t(const size_t, len, wsize);
2377 num_pages = DIV_ROUND_UP(clen, PAGE_SIZE);
2386 cifs_uncached_writedata_release(struct kref *refcount)
2389 struct cifs_writedata *wdata = container_of(refcount,
2390 struct cifs_writedata, refcount);
2392 for (i = 0; i < wdata->nr_pages; i++)
2393 put_page(wdata->pages[i]);
2394 cifs_writedata_release(refcount);
2398 cifs_uncached_writev_complete(struct work_struct *work)
2400 struct cifs_writedata *wdata = container_of(work,
2401 struct cifs_writedata, work);
2402 struct inode *inode = wdata->cfile->dentry->d_inode;
2403 struct cifsInodeInfo *cifsi = CIFS_I(inode);
2405 spin_lock(&inode->i_lock);
2406 cifs_update_eof(cifsi, wdata->offset, wdata->bytes);
2407 if (cifsi->server_eof > inode->i_size)
2408 i_size_write(inode, cifsi->server_eof);
2409 spin_unlock(&inode->i_lock);
2411 complete(&wdata->done);
2413 kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2417 wdata_fill_from_iovec(struct cifs_writedata *wdata, struct iov_iter *from,
2418 size_t *len, unsigned long *num_pages)
2420 size_t save_len, copied, bytes, cur_len = *len;
2421 unsigned long i, nr_pages = *num_pages;
2424 for (i = 0; i < nr_pages; i++) {
2425 bytes = min_t(const size_t, cur_len, PAGE_SIZE);
2426 copied = copy_page_from_iter(wdata->pages[i], 0, bytes, from);
2429 * If we didn't copy as much as we expected, then that
2430 * may mean we trod into an unmapped area. Stop copying
2431 * at that point. On the next pass through the big
2432 * loop, we'll likely end up getting a zero-length
2433 * write and bailing out of it.
2438 cur_len = save_len - cur_len;
2442 * If we have no data to send, then that probably means that
2443 * the copy above failed altogether. That's most likely because
2444 * the address in the iovec was bogus. Return -EFAULT and let
2445 * the caller free anything we allocated and bail out.
2451 * i + 1 now represents the number of pages we actually used in
2452 * the copy phase above.
2459 cifs_write_from_iter(loff_t offset, size_t len, struct iov_iter *from,
2460 struct cifsFileInfo *open_file,
2461 struct cifs_sb_info *cifs_sb, struct list_head *wdata_list)
2465 unsigned long nr_pages, num_pages, i;
2466 struct cifs_writedata *wdata;
2467 struct iov_iter saved_from;
2468 loff_t saved_offset = offset;
2470 struct TCP_Server_Info *server;
2472 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2473 pid = open_file->pid;
2475 pid = current->tgid;
2477 server = tlink_tcon(open_file->tlink)->ses->server;
2478 memcpy(&saved_from, from, sizeof(struct iov_iter));
2481 unsigned int wsize, credits;
2483 rc = server->ops->wait_mtu_credits(server, cifs_sb->wsize,
2488 nr_pages = get_numpages(wsize, len, &cur_len);
2489 wdata = cifs_writedata_alloc(nr_pages,
2490 cifs_uncached_writev_complete);
2493 add_credits_and_wake_if(server, credits, 0);
2497 rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
2500 add_credits_and_wake_if(server, credits, 0);
2504 num_pages = nr_pages;
2505 rc = wdata_fill_from_iovec(wdata, from, &cur_len, &num_pages);
2507 for (i = 0; i < nr_pages; i++)
2508 put_page(wdata->pages[i]);
2510 add_credits_and_wake_if(server, credits, 0);
2515 * Bring nr_pages down to the number of pages we actually used,
2516 * and free any pages that we didn't use.
2518 for ( ; nr_pages > num_pages; nr_pages--)
2519 put_page(wdata->pages[nr_pages - 1]);
2521 wdata->sync_mode = WB_SYNC_ALL;
2522 wdata->nr_pages = nr_pages;
2523 wdata->offset = (__u64)offset;
2524 wdata->cfile = cifsFileInfo_get(open_file);
2526 wdata->bytes = cur_len;
2527 wdata->pagesz = PAGE_SIZE;
2528 wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
2529 wdata->credits = credits;
2531 if (!wdata->cfile->invalidHandle ||
2532 !cifs_reopen_file(wdata->cfile, false))
2533 rc = server->ops->async_writev(wdata,
2534 cifs_uncached_writedata_release);
2536 add_credits_and_wake_if(server, wdata->credits, 0);
2537 kref_put(&wdata->refcount,
2538 cifs_uncached_writedata_release);
2539 if (rc == -EAGAIN) {
2540 memcpy(from, &saved_from,
2541 sizeof(struct iov_iter));
2542 iov_iter_advance(from, offset - saved_offset);
2548 list_add_tail(&wdata->list, wdata_list);
2557 cifs_iovec_write(struct file *file, struct iov_iter *from, loff_t *poffset)
2560 ssize_t total_written = 0;
2561 struct cifsFileInfo *open_file;
2562 struct cifs_tcon *tcon;
2563 struct cifs_sb_info *cifs_sb;
2564 struct cifs_writedata *wdata, *tmp;
2565 struct list_head wdata_list;
2566 struct iov_iter saved_from;
2569 len = iov_iter_count(from);
2570 rc = generic_write_checks(file, poffset, &len, 0);
2577 iov_iter_truncate(from, len);
2579 INIT_LIST_HEAD(&wdata_list);
2580 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2581 open_file = file->private_data;
2582 tcon = tlink_tcon(open_file->tlink);
2584 if (!tcon->ses->server->ops->async_writev)
2587 memcpy(&saved_from, from, sizeof(struct iov_iter));
2589 rc = cifs_write_from_iter(*poffset, len, from, open_file, cifs_sb,
2593 * If at least one write was successfully sent, then discard any rc
2594 * value from the later writes. If the other write succeeds, then
2595 * we'll end up returning whatever was written. If it fails, then
2596 * we'll get a new rc value from that.
2598 if (!list_empty(&wdata_list))
2602 * Wait for and collect replies for any successful sends in order of
2603 * increasing offset. Once an error is hit or we get a fatal signal
2604 * while waiting, then return without waiting for any more replies.
2607 list_for_each_entry_safe(wdata, tmp, &wdata_list, list) {
2609 /* FIXME: freezable too? */
2610 rc = wait_for_completion_killable(&wdata->done);
2613 else if (wdata->result)
2616 total_written += wdata->bytes;
2618 /* resend call if it's a retryable error */
2619 if (rc == -EAGAIN) {
2620 struct list_head tmp_list;
2621 struct iov_iter tmp_from;
2623 INIT_LIST_HEAD(&tmp_list);
2624 list_del_init(&wdata->list);
2626 memcpy(&tmp_from, &saved_from,
2627 sizeof(struct iov_iter));
2628 iov_iter_advance(&tmp_from,
2629 wdata->offset - *poffset);
2631 rc = cifs_write_from_iter(wdata->offset,
2632 wdata->bytes, &tmp_from,
2633 open_file, cifs_sb, &tmp_list);
2635 list_splice(&tmp_list, &wdata_list);
2637 kref_put(&wdata->refcount,
2638 cifs_uncached_writedata_release);
2642 list_del_init(&wdata->list);
2643 kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2646 if (total_written > 0)
2647 *poffset += total_written;
2649 cifs_stats_bytes_written(tcon, total_written);
2650 return total_written ? total_written : (ssize_t)rc;
2653 ssize_t cifs_user_writev(struct kiocb *iocb, struct iov_iter *from)
2656 struct inode *inode;
2657 loff_t pos = iocb->ki_pos;
2659 inode = file_inode(iocb->ki_filp);
2662 * BB - optimize the way when signing is disabled. We can drop this
2663 * extra memory-to-memory copying and use iovec buffers for constructing
2667 written = cifs_iovec_write(iocb->ki_filp, from, &pos);
2669 set_bit(CIFS_INO_INVALID_MAPPING, &CIFS_I(inode)->flags);
2677 cifs_writev(struct kiocb *iocb, struct iov_iter *from)
2679 struct file *file = iocb->ki_filp;
2680 struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
2681 struct inode *inode = file->f_mapping->host;
2682 struct cifsInodeInfo *cinode = CIFS_I(inode);
2683 struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
2684 ssize_t rc = -EACCES;
2685 loff_t lock_pos = iocb->ki_pos;
2688 * We need to hold the sem to be sure nobody modifies lock list
2689 * with a brlock that prevents writing.
2691 down_read(&cinode->lock_sem);
2692 mutex_lock(&inode->i_mutex);
2693 if (file->f_flags & O_APPEND)
2694 lock_pos = i_size_read(inode);
2695 if (!cifs_find_lock_conflict(cfile, lock_pos, iov_iter_count(from),
2696 server->vals->exclusive_lock_type, NULL,
2698 rc = __generic_file_write_iter(iocb, from);
2699 mutex_unlock(&inode->i_mutex);
2704 err = generic_write_sync(file, iocb->ki_pos - rc, rc);
2709 mutex_unlock(&inode->i_mutex);
2711 up_read(&cinode->lock_sem);
2716 cifs_strict_writev(struct kiocb *iocb, struct iov_iter *from)
2718 struct inode *inode = file_inode(iocb->ki_filp);
2719 struct cifsInodeInfo *cinode = CIFS_I(inode);
2720 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2721 struct cifsFileInfo *cfile = (struct cifsFileInfo *)
2722 iocb->ki_filp->private_data;
2723 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
2726 written = cifs_get_writer(cinode);
2730 if (CIFS_CACHE_WRITE(cinode)) {
2731 if (cap_unix(tcon->ses) &&
2732 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability))
2733 && ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0)) {
2734 written = generic_file_write_iter(iocb, from);
2737 written = cifs_writev(iocb, from);
2741 * For non-oplocked files in strict cache mode we need to write the data
2742 * to the server exactly from the pos to pos+len-1 rather than flush all
2743 * affected pages because it may cause a error with mandatory locks on
2744 * these pages but not on the region from pos to ppos+len-1.
2746 written = cifs_user_writev(iocb, from);
2747 if (written > 0 && CIFS_CACHE_READ(cinode)) {
2749 * Windows 7 server can delay breaking level2 oplock if a write
2750 * request comes - break it on the client to prevent reading
2753 cifs_zap_mapping(inode);
2754 cifs_dbg(FYI, "Set no oplock for inode=%p after a write operation\n",
2759 cifs_put_writer(cinode);
2763 static struct cifs_readdata *
2764 cifs_readdata_alloc(unsigned int nr_pages, work_func_t complete)
2766 struct cifs_readdata *rdata;
2768 rdata = kzalloc(sizeof(*rdata) + (sizeof(struct page *) * nr_pages),
2770 if (rdata != NULL) {
2771 kref_init(&rdata->refcount);
2772 INIT_LIST_HEAD(&rdata->list);
2773 init_completion(&rdata->done);
2774 INIT_WORK(&rdata->work, complete);
2781 cifs_readdata_release(struct kref *refcount)
2783 struct cifs_readdata *rdata = container_of(refcount,
2784 struct cifs_readdata, refcount);
2787 cifsFileInfo_put(rdata->cfile);
2793 cifs_read_allocate_pages(struct cifs_readdata *rdata, unsigned int nr_pages)
2799 for (i = 0; i < nr_pages; i++) {
2800 page = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2805 rdata->pages[i] = page;
2809 for (i = 0; i < nr_pages; i++) {
2810 put_page(rdata->pages[i]);
2811 rdata->pages[i] = NULL;
2818 cifs_uncached_readdata_release(struct kref *refcount)
2820 struct cifs_readdata *rdata = container_of(refcount,
2821 struct cifs_readdata, refcount);
2824 for (i = 0; i < rdata->nr_pages; i++) {
2825 put_page(rdata->pages[i]);
2826 rdata->pages[i] = NULL;
2828 cifs_readdata_release(refcount);
2832 cifs_retry_async_readv(struct cifs_readdata *rdata)
2835 struct TCP_Server_Info *server;
2837 server = tlink_tcon(rdata->cfile->tlink)->ses->server;
2840 if (rdata->cfile->invalidHandle) {
2841 rc = cifs_reopen_file(rdata->cfile, true);
2845 rc = server->ops->async_readv(rdata);
2846 } while (rc == -EAGAIN);
2852 * cifs_readdata_to_iov - copy data from pages in response to an iovec
2853 * @rdata: the readdata response with list of pages holding data
2854 * @iter: destination for our data
2856 * This function copies data from a list of pages in a readdata response into
2857 * an array of iovecs. It will first calculate where the data should go
2858 * based on the info in the readdata and then copy the data into that spot.
2861 cifs_readdata_to_iov(struct cifs_readdata *rdata, struct iov_iter *iter)
2863 size_t remaining = rdata->bytes;
2866 for (i = 0; i < rdata->nr_pages; i++) {
2867 struct page *page = rdata->pages[i];
2868 size_t copy = min_t(size_t, remaining, PAGE_SIZE);
2869 size_t written = copy_page_to_iter(page, 0, copy, iter);
2870 remaining -= written;
2871 if (written < copy && iov_iter_count(iter) > 0)
2874 return remaining ? -EFAULT : 0;
2878 cifs_uncached_readv_complete(struct work_struct *work)
2880 struct cifs_readdata *rdata = container_of(work,
2881 struct cifs_readdata, work);
2883 complete(&rdata->done);
2884 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
2888 cifs_uncached_read_into_pages(struct TCP_Server_Info *server,
2889 struct cifs_readdata *rdata, unsigned int len)
2891 int total_read = 0, result = 0;
2893 unsigned int nr_pages = rdata->nr_pages;
2896 rdata->tailsz = PAGE_SIZE;
2897 for (i = 0; i < nr_pages; i++) {
2898 struct page *page = rdata->pages[i];
2900 if (len >= PAGE_SIZE) {
2901 /* enough data to fill the page */
2902 iov.iov_base = kmap(page);
2903 iov.iov_len = PAGE_SIZE;
2904 cifs_dbg(FYI, "%u: iov_base=%p iov_len=%zu\n",
2905 i, iov.iov_base, iov.iov_len);
2907 } else if (len > 0) {
2908 /* enough for partial page, fill and zero the rest */
2909 iov.iov_base = kmap(page);
2911 cifs_dbg(FYI, "%u: iov_base=%p iov_len=%zu\n",
2912 i, iov.iov_base, iov.iov_len);
2913 memset(iov.iov_base + len, '\0', PAGE_SIZE - len);
2914 rdata->tailsz = len;
2917 /* no need to hold page hostage */
2918 rdata->pages[i] = NULL;
2924 result = cifs_readv_from_socket(server, &iov, 1, iov.iov_len);
2929 total_read += result;
2932 return total_read > 0 && result != -EAGAIN ? total_read : result;
2935 ssize_t cifs_user_readv(struct kiocb *iocb, struct iov_iter *to)
2937 struct file *file = iocb->ki_filp;
2939 size_t len, cur_len;
2940 ssize_t total_read = 0;
2941 loff_t offset = iocb->ki_pos;
2942 unsigned int npages;
2943 struct cifs_sb_info *cifs_sb;
2944 struct cifs_tcon *tcon;
2945 struct cifsFileInfo *open_file;
2946 struct cifs_readdata *rdata, *tmp;
2947 struct list_head rdata_list;
2950 len = iov_iter_count(to);
2954 INIT_LIST_HEAD(&rdata_list);
2955 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2956 open_file = file->private_data;
2957 tcon = tlink_tcon(open_file->tlink);
2959 if (!tcon->ses->server->ops->async_readv)
2962 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2963 pid = open_file->pid;
2965 pid = current->tgid;
2967 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
2968 cifs_dbg(FYI, "attempting read on write only file instance\n");
2971 cur_len = min_t(const size_t, len - total_read, cifs_sb->rsize);
2972 npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
2974 /* allocate a readdata struct */
2975 rdata = cifs_readdata_alloc(npages,
2976 cifs_uncached_readv_complete);
2982 rc = cifs_read_allocate_pages(rdata, npages);
2986 rdata->cfile = cifsFileInfo_get(open_file);
2987 rdata->nr_pages = npages;
2988 rdata->offset = offset;
2989 rdata->bytes = cur_len;
2991 rdata->pagesz = PAGE_SIZE;
2992 rdata->read_into_pages = cifs_uncached_read_into_pages;
2994 rc = cifs_retry_async_readv(rdata);
2997 kref_put(&rdata->refcount,
2998 cifs_uncached_readdata_release);
3002 list_add_tail(&rdata->list, &rdata_list);
3007 /* if at least one read request send succeeded, then reset rc */
3008 if (!list_empty(&rdata_list))
3011 len = iov_iter_count(to);
3012 /* the loop below should proceed in the order of increasing offsets */
3013 list_for_each_entry_safe(rdata, tmp, &rdata_list, list) {
3016 /* FIXME: freezable sleep too? */
3017 rc = wait_for_completion_killable(&rdata->done);
3020 else if (rdata->result) {
3022 /* resend call if it's a retryable error */
3023 if (rc == -EAGAIN) {
3024 rc = cifs_retry_async_readv(rdata);
3028 rc = cifs_readdata_to_iov(rdata, to);
3032 list_del_init(&rdata->list);
3033 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3036 total_read = len - iov_iter_count(to);
3038 cifs_stats_bytes_read(tcon, total_read);
3040 /* mask nodata case */
3045 iocb->ki_pos += total_read;
3052 cifs_strict_readv(struct kiocb *iocb, struct iov_iter *to)
3054 struct inode *inode = file_inode(iocb->ki_filp);
3055 struct cifsInodeInfo *cinode = CIFS_I(inode);
3056 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
3057 struct cifsFileInfo *cfile = (struct cifsFileInfo *)
3058 iocb->ki_filp->private_data;
3059 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
3063 * In strict cache mode we need to read from the server all the time
3064 * if we don't have level II oplock because the server can delay mtime
3065 * change - so we can't make a decision about inode invalidating.
3066 * And we can also fail with pagereading if there are mandatory locks
3067 * on pages affected by this read but not on the region from pos to
3070 if (!CIFS_CACHE_READ(cinode))
3071 return cifs_user_readv(iocb, to);
3073 if (cap_unix(tcon->ses) &&
3074 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
3075 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
3076 return generic_file_read_iter(iocb, to);
3079 * We need to hold the sem to be sure nobody modifies lock list
3080 * with a brlock that prevents reading.
3082 down_read(&cinode->lock_sem);
3083 if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(to),
3084 tcon->ses->server->vals->shared_lock_type,
3085 NULL, CIFS_READ_OP))
3086 rc = generic_file_read_iter(iocb, to);
3087 up_read(&cinode->lock_sem);
3092 cifs_read(struct file *file, char *read_data, size_t read_size, loff_t *offset)
3095 unsigned int bytes_read = 0;
3096 unsigned int total_read;
3097 unsigned int current_read_size;
3099 struct cifs_sb_info *cifs_sb;
3100 struct cifs_tcon *tcon;
3101 struct TCP_Server_Info *server;
3104 struct cifsFileInfo *open_file;
3105 struct cifs_io_parms io_parms;
3106 int buf_type = CIFS_NO_BUFFER;
3110 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
3112 /* FIXME: set up handlers for larger reads and/or convert to async */
3113 rsize = min_t(unsigned int, cifs_sb->rsize, CIFSMaxBufSize);
3115 if (file->private_data == NULL) {
3120 open_file = file->private_data;
3121 tcon = tlink_tcon(open_file->tlink);
3122 server = tcon->ses->server;
3124 if (!server->ops->sync_read) {
3129 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3130 pid = open_file->pid;
3132 pid = current->tgid;
3134 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3135 cifs_dbg(FYI, "attempting read on write only file instance\n");
3137 for (total_read = 0, cur_offset = read_data; read_size > total_read;
3138 total_read += bytes_read, cur_offset += bytes_read) {
3139 current_read_size = min_t(uint, read_size - total_read, rsize);
3141 * For windows me and 9x we do not want to request more than it
3142 * negotiated since it will refuse the read then.
3144 if ((tcon->ses) && !(tcon->ses->capabilities &
3145 tcon->ses->server->vals->cap_large_files)) {
3146 current_read_size = min_t(uint, current_read_size,
3150 while (rc == -EAGAIN) {
3151 if (open_file->invalidHandle) {
3152 rc = cifs_reopen_file(open_file, true);
3157 io_parms.tcon = tcon;
3158 io_parms.offset = *offset;
3159 io_parms.length = current_read_size;
3160 rc = server->ops->sync_read(xid, open_file, &io_parms,
3161 &bytes_read, &cur_offset,
3164 if (rc || (bytes_read == 0)) {
3172 cifs_stats_bytes_read(tcon, total_read);
3173 *offset += bytes_read;
3181 * If the page is mmap'ed into a process' page tables, then we need to make
3182 * sure that it doesn't change while being written back.
3185 cifs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
3187 struct page *page = vmf->page;
3190 return VM_FAULT_LOCKED;
3193 static struct vm_operations_struct cifs_file_vm_ops = {
3194 .fault = filemap_fault,
3195 .map_pages = filemap_map_pages,
3196 .page_mkwrite = cifs_page_mkwrite,
3197 .remap_pages = generic_file_remap_pages,
3200 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
3203 struct inode *inode = file_inode(file);
3207 if (!CIFS_CACHE_READ(CIFS_I(inode))) {
3208 rc = cifs_zap_mapping(inode);
3213 rc = generic_file_mmap(file, vma);
3215 vma->vm_ops = &cifs_file_vm_ops;
3220 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
3225 rc = cifs_revalidate_file(file);
3227 cifs_dbg(FYI, "Validation prior to mmap failed, error=%d\n",
3232 rc = generic_file_mmap(file, vma);
3234 vma->vm_ops = &cifs_file_vm_ops;
3240 cifs_readv_complete(struct work_struct *work)
3243 struct cifs_readdata *rdata = container_of(work,
3244 struct cifs_readdata, work);
3246 for (i = 0; i < rdata->nr_pages; i++) {
3247 struct page *page = rdata->pages[i];
3249 lru_cache_add_file(page);
3251 if (rdata->result == 0) {
3252 flush_dcache_page(page);
3253 SetPageUptodate(page);
3258 if (rdata->result == 0)
3259 cifs_readpage_to_fscache(rdata->mapping->host, page);
3261 page_cache_release(page);
3262 rdata->pages[i] = NULL;
3264 kref_put(&rdata->refcount, cifs_readdata_release);
3268 cifs_readpages_read_into_pages(struct TCP_Server_Info *server,
3269 struct cifs_readdata *rdata, unsigned int len)
3271 int total_read = 0, result = 0;
3275 unsigned int nr_pages = rdata->nr_pages;
3278 /* determine the eof that the server (probably) has */
3279 eof = CIFS_I(rdata->mapping->host)->server_eof;
3280 eof_index = eof ? (eof - 1) >> PAGE_CACHE_SHIFT : 0;
3281 cifs_dbg(FYI, "eof=%llu eof_index=%lu\n", eof, eof_index);
3283 rdata->tailsz = PAGE_CACHE_SIZE;
3284 for (i = 0; i < nr_pages; i++) {
3285 struct page *page = rdata->pages[i];
3287 if (len >= PAGE_CACHE_SIZE) {
3288 /* enough data to fill the page */
3289 iov.iov_base = kmap(page);
3290 iov.iov_len = PAGE_CACHE_SIZE;
3291 cifs_dbg(FYI, "%u: idx=%lu iov_base=%p iov_len=%zu\n",
3292 i, page->index, iov.iov_base, iov.iov_len);
3293 len -= PAGE_CACHE_SIZE;
3294 } else if (len > 0) {
3295 /* enough for partial page, fill and zero the rest */
3296 iov.iov_base = kmap(page);
3298 cifs_dbg(FYI, "%u: idx=%lu iov_base=%p iov_len=%zu\n",
3299 i, page->index, iov.iov_base, iov.iov_len);
3300 memset(iov.iov_base + len,
3301 '\0', PAGE_CACHE_SIZE - len);
3302 rdata->tailsz = len;
3304 } else if (page->index > eof_index) {
3306 * The VFS will not try to do readahead past the
3307 * i_size, but it's possible that we have outstanding
3308 * writes with gaps in the middle and the i_size hasn't
3309 * caught up yet. Populate those with zeroed out pages
3310 * to prevent the VFS from repeatedly attempting to
3311 * fill them until the writes are flushed.
3313 zero_user(page, 0, PAGE_CACHE_SIZE);
3314 lru_cache_add_file(page);
3315 flush_dcache_page(page);
3316 SetPageUptodate(page);
3318 page_cache_release(page);
3319 rdata->pages[i] = NULL;
3323 /* no need to hold page hostage */
3324 lru_cache_add_file(page);
3326 page_cache_release(page);
3327 rdata->pages[i] = NULL;
3332 result = cifs_readv_from_socket(server, &iov, 1, iov.iov_len);
3337 total_read += result;
3340 return total_read > 0 && result != -EAGAIN ? total_read : result;
3344 readpages_get_pages(struct address_space *mapping, struct list_head *page_list,
3345 unsigned int rsize, struct list_head *tmplist,
3346 unsigned int *nr_pages, loff_t *offset, unsigned int *bytes)
3348 struct page *page, *tpage;
3349 unsigned int expected_index;
3352 page = list_entry(page_list->prev, struct page, lru);
3355 * Lock the page and put it in the cache. Since no one else
3356 * should have access to this page, we're safe to simply set
3357 * PG_locked without checking it first.
3359 __set_page_locked(page);
3360 rc = add_to_page_cache_locked(page, mapping,
3361 page->index, GFP_KERNEL);
3363 /* give up if we can't stick it in the cache */
3365 __clear_page_locked(page);
3369 /* move first page to the tmplist */
3370 *offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3371 *bytes = PAGE_CACHE_SIZE;
3373 list_move_tail(&page->lru, tmplist);
3375 /* now try and add more pages onto the request */
3376 expected_index = page->index + 1;
3377 list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
3378 /* discontinuity ? */
3379 if (page->index != expected_index)
3382 /* would this page push the read over the rsize? */
3383 if (*bytes + PAGE_CACHE_SIZE > rsize)
3386 __set_page_locked(page);
3387 if (add_to_page_cache_locked(page, mapping, page->index,
3389 __clear_page_locked(page);
3392 list_move_tail(&page->lru, tmplist);
3393 (*bytes) += PAGE_CACHE_SIZE;
3400 static int cifs_readpages(struct file *file, struct address_space *mapping,
3401 struct list_head *page_list, unsigned num_pages)
3404 struct list_head tmplist;
3405 struct cifsFileInfo *open_file = file->private_data;
3406 struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
3407 unsigned int rsize = cifs_sb->rsize;
3411 * Give up immediately if rsize is too small to read an entire page.
3412 * The VFS will fall back to readpage. We should never reach this
3413 * point however since we set ra_pages to 0 when the rsize is smaller
3414 * than a cache page.
3416 if (unlikely(rsize < PAGE_CACHE_SIZE))
3420 * Reads as many pages as possible from fscache. Returns -ENOBUFS
3421 * immediately if the cookie is negative
3423 * After this point, every page in the list might have PG_fscache set,
3424 * so we will need to clean that up off of every page we don't use.
3426 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
3431 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3432 pid = open_file->pid;
3434 pid = current->tgid;
3437 INIT_LIST_HEAD(&tmplist);
3439 cifs_dbg(FYI, "%s: file=%p mapping=%p num_pages=%u\n",
3440 __func__, file, mapping, num_pages);
3443 * Start with the page at end of list and move it to private
3444 * list. Do the same with any following pages until we hit
3445 * the rsize limit, hit an index discontinuity, or run out of
3446 * pages. Issue the async read and then start the loop again
3447 * until the list is empty.
3449 * Note that list order is important. The page_list is in
3450 * the order of declining indexes. When we put the pages in
3451 * the rdata->pages, then we want them in increasing order.
3453 while (!list_empty(page_list)) {
3454 unsigned int i, nr_pages, bytes;
3456 struct page *page, *tpage;
3457 struct cifs_readdata *rdata;
3459 rc = readpages_get_pages(mapping, page_list, rsize, &tmplist,
3460 &nr_pages, &offset, &bytes);
3464 rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
3466 /* best to give up if we're out of mem */
3467 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3468 list_del(&page->lru);
3469 lru_cache_add_file(page);
3471 page_cache_release(page);
3477 rdata->cfile = cifsFileInfo_get(open_file);
3478 rdata->mapping = mapping;
3479 rdata->offset = offset;
3480 rdata->bytes = bytes;
3482 rdata->pagesz = PAGE_CACHE_SIZE;
3483 rdata->read_into_pages = cifs_readpages_read_into_pages;
3485 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3486 list_del(&page->lru);
3487 rdata->pages[rdata->nr_pages++] = page;
3490 rc = cifs_retry_async_readv(rdata);
3492 for (i = 0; i < rdata->nr_pages; i++) {
3493 page = rdata->pages[i];
3494 lru_cache_add_file(page);
3496 page_cache_release(page);
3498 kref_put(&rdata->refcount, cifs_readdata_release);
3502 kref_put(&rdata->refcount, cifs_readdata_release);
3505 /* Any pages that have been shown to fscache but didn't get added to
3506 * the pagecache must be uncached before they get returned to the
3509 cifs_fscache_readpages_cancel(mapping->host, page_list);
3514 * cifs_readpage_worker must be called with the page pinned
3516 static int cifs_readpage_worker(struct file *file, struct page *page,
3522 /* Is the page cached? */
3523 rc = cifs_readpage_from_fscache(file_inode(file), page);
3527 read_data = kmap(page);
3528 /* for reads over a certain size could initiate async read ahead */
3530 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
3535 cifs_dbg(FYI, "Bytes read %d\n", rc);
3537 file_inode(file)->i_atime =
3538 current_fs_time(file_inode(file)->i_sb);
3540 if (PAGE_CACHE_SIZE > rc)
3541 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
3543 flush_dcache_page(page);
3544 SetPageUptodate(page);
3546 /* send this page to the cache */
3547 cifs_readpage_to_fscache(file_inode(file), page);
3559 static int cifs_readpage(struct file *file, struct page *page)
3561 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3567 if (file->private_data == NULL) {
3573 cifs_dbg(FYI, "readpage %p at offset %d 0x%x\n",
3574 page, (int)offset, (int)offset);
3576 rc = cifs_readpage_worker(file, page, &offset);
3582 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
3584 struct cifsFileInfo *open_file;
3586 spin_lock(&cifs_file_list_lock);
3587 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
3588 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
3589 spin_unlock(&cifs_file_list_lock);
3593 spin_unlock(&cifs_file_list_lock);
3597 /* We do not want to update the file size from server for inodes
3598 open for write - to avoid races with writepage extending
3599 the file - in the future we could consider allowing
3600 refreshing the inode only on increases in the file size
3601 but this is tricky to do without racing with writebehind
3602 page caching in the current Linux kernel design */
3603 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
3608 if (is_inode_writable(cifsInode)) {
3609 /* This inode is open for write at least once */
3610 struct cifs_sb_info *cifs_sb;
3612 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
3613 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
3614 /* since no page cache to corrupt on directio
3615 we can change size safely */
3619 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
3627 static int cifs_write_begin(struct file *file, struct address_space *mapping,
3628 loff_t pos, unsigned len, unsigned flags,
3629 struct page **pagep, void **fsdata)
3632 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
3633 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
3634 loff_t page_start = pos & PAGE_MASK;
3639 cifs_dbg(FYI, "write_begin from %lld len %d\n", (long long)pos, len);
3642 page = grab_cache_page_write_begin(mapping, index, flags);
3648 if (PageUptodate(page))
3652 * If we write a full page it will be up to date, no need to read from
3653 * the server. If the write is short, we'll end up doing a sync write
3656 if (len == PAGE_CACHE_SIZE)
3660 * optimize away the read when we have an oplock, and we're not
3661 * expecting to use any of the data we'd be reading in. That
3662 * is, when the page lies beyond the EOF, or straddles the EOF
3663 * and the write will cover all of the existing data.
3665 if (CIFS_CACHE_READ(CIFS_I(mapping->host))) {
3666 i_size = i_size_read(mapping->host);
3667 if (page_start >= i_size ||
3668 (offset == 0 && (pos + len) >= i_size)) {
3669 zero_user_segments(page, 0, offset,
3673 * PageChecked means that the parts of the page
3674 * to which we're not writing are considered up
3675 * to date. Once the data is copied to the
3676 * page, it can be set uptodate.
3678 SetPageChecked(page);
3683 if ((file->f_flags & O_ACCMODE) != O_WRONLY && !oncethru) {
3685 * might as well read a page, it is fast enough. If we get
3686 * an error, we don't need to return it. cifs_write_end will
3687 * do a sync write instead since PG_uptodate isn't set.
3689 cifs_readpage_worker(file, page, &page_start);
3690 page_cache_release(page);
3694 /* we could try using another file handle if there is one -
3695 but how would we lock it to prevent close of that handle
3696 racing with this read? In any case
3697 this will be written out by write_end so is fine */
3704 static int cifs_release_page(struct page *page, gfp_t gfp)
3706 if (PagePrivate(page))
3709 return cifs_fscache_release_page(page, gfp);
3712 static void cifs_invalidate_page(struct page *page, unsigned int offset,
3713 unsigned int length)
3715 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
3717 if (offset == 0 && length == PAGE_CACHE_SIZE)
3718 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
3721 static int cifs_launder_page(struct page *page)
3724 loff_t range_start = page_offset(page);
3725 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
3726 struct writeback_control wbc = {
3727 .sync_mode = WB_SYNC_ALL,
3729 .range_start = range_start,
3730 .range_end = range_end,
3733 cifs_dbg(FYI, "Launder page: %p\n", page);
3735 if (clear_page_dirty_for_io(page))
3736 rc = cifs_writepage_locked(page, &wbc);
3738 cifs_fscache_invalidate_page(page, page->mapping->host);
3743 cifs_pending_writers_wait(void *unused)
3749 void cifs_oplock_break(struct work_struct *work)
3751 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
3753 struct inode *inode = cfile->dentry->d_inode;
3754 struct cifsInodeInfo *cinode = CIFS_I(inode);
3755 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
3756 struct TCP_Server_Info *server = tcon->ses->server;
3759 wait_on_bit(&cinode->flags, CIFS_INODE_PENDING_WRITERS,
3760 cifs_pending_writers_wait, TASK_UNINTERRUPTIBLE);
3762 server->ops->downgrade_oplock(server, cinode,
3763 test_bit(CIFS_INODE_DOWNGRADE_OPLOCK_TO_L2, &cinode->flags));
3765 if (!CIFS_CACHE_WRITE(cinode) && CIFS_CACHE_READ(cinode) &&
3766 cifs_has_mand_locks(cinode)) {
3767 cifs_dbg(FYI, "Reset oplock to None for inode=%p due to mand locks\n",
3772 if (inode && S_ISREG(inode->i_mode)) {
3773 if (CIFS_CACHE_READ(cinode))
3774 break_lease(inode, O_RDONLY);
3776 break_lease(inode, O_WRONLY);
3777 rc = filemap_fdatawrite(inode->i_mapping);
3778 if (!CIFS_CACHE_READ(cinode)) {
3779 rc = filemap_fdatawait(inode->i_mapping);
3780 mapping_set_error(inode->i_mapping, rc);
3781 cifs_zap_mapping(inode);
3783 cifs_dbg(FYI, "Oplock flush inode %p rc %d\n", inode, rc);
3786 rc = cifs_push_locks(cfile);
3788 cifs_dbg(VFS, "Push locks rc = %d\n", rc);
3791 * releasing stale oplock after recent reconnect of smb session using
3792 * a now incorrect file handle is not a data integrity issue but do
3793 * not bother sending an oplock release if session to server still is
3794 * disconnected since oplock already released by the server
3796 if (!cfile->oplock_break_cancelled) {
3797 rc = tcon->ses->server->ops->oplock_response(tcon, &cfile->fid,
3799 cifs_dbg(FYI, "Oplock release rc = %d\n", rc);
3801 cifs_done_oplock_break(cinode);
3805 * The presence of cifs_direct_io() in the address space ops vector
3806 * allowes open() O_DIRECT flags which would have failed otherwise.
3808 * In the non-cached mode (mount with cache=none), we shunt off direct read and write requests
3809 * so this method should never be called.
3811 * Direct IO is not yet supported in the cached mode.
3814 cifs_direct_io(int rw, struct kiocb *iocb, struct iov_iter *iter,
3819 * Eventually need to support direct IO for non forcedirectio mounts
3825 const struct address_space_operations cifs_addr_ops = {
3826 .readpage = cifs_readpage,
3827 .readpages = cifs_readpages,
3828 .writepage = cifs_writepage,
3829 .writepages = cifs_writepages,
3830 .write_begin = cifs_write_begin,
3831 .write_end = cifs_write_end,
3832 .set_page_dirty = __set_page_dirty_nobuffers,
3833 .releasepage = cifs_release_page,
3834 .direct_IO = cifs_direct_io,
3835 .invalidatepage = cifs_invalidate_page,
3836 .launder_page = cifs_launder_page,
3840 * cifs_readpages requires the server to support a buffer large enough to
3841 * contain the header plus one complete page of data. Otherwise, we need
3842 * to leave cifs_readpages out of the address space operations.
3844 const struct address_space_operations cifs_addr_ops_smallbuf = {
3845 .readpage = cifs_readpage,
3846 .writepage = cifs_writepage,
3847 .writepages = cifs_writepages,
3848 .write_begin = cifs_write_begin,
3849 .write_end = cifs_write_end,
3850 .set_page_dirty = __set_page_dirty_nobuffers,
3851 .releasepage = cifs_release_page,
3852 .invalidatepage = cifs_invalidate_page,
3853 .launder_page = cifs_launder_page,