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((size_t)cifs_sb->wsize,
1674 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 bool done = false, scanned = false, range_whole = false;
2036 struct cifs_writedata *wdata;
2040 * If wsize is smaller than the page cache size, default to writing
2041 * one page at a time via cifs_writepage
2043 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
2044 return generic_writepages(mapping, wbc);
2046 if (wbc->range_cyclic) {
2047 index = mapping->writeback_index; /* Start from prev offset */
2050 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2051 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2052 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2057 while (!done && index <= end) {
2058 unsigned int i, nr_pages, found_pages;
2059 pgoff_t next = 0, tofind, saved_index = index;
2061 tofind = min((cifs_sb->wsize / PAGE_CACHE_SIZE) - 1,
2064 wdata = wdata_alloc_and_fillpages(tofind, mapping, end, &index,
2071 if (found_pages == 0) {
2072 kref_put(&wdata->refcount, cifs_writedata_release);
2076 nr_pages = wdata_prepare_pages(wdata, found_pages, mapping, wbc,
2077 end, &index, &next, &done);
2079 /* nothing to write? */
2080 if (nr_pages == 0) {
2081 kref_put(&wdata->refcount, cifs_writedata_release);
2085 rc = wdata_send_pages(wdata, nr_pages, mapping, wbc);
2087 /* send failure -- clean up the mess */
2089 for (i = 0; i < nr_pages; ++i) {
2091 redirty_page_for_writepage(wbc,
2094 SetPageError(wdata->pages[i]);
2095 end_page_writeback(wdata->pages[i]);
2096 page_cache_release(wdata->pages[i]);
2099 mapping_set_error(mapping, rc);
2101 kref_put(&wdata->refcount, cifs_writedata_release);
2103 if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN) {
2104 index = saved_index;
2108 wbc->nr_to_write -= nr_pages;
2109 if (wbc->nr_to_write <= 0)
2115 if (!scanned && !done) {
2117 * We hit the last page and there is more work to be done: wrap
2118 * back to the start of the file
2125 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2126 mapping->writeback_index = index;
2132 cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
2138 /* BB add check for wbc flags */
2139 page_cache_get(page);
2140 if (!PageUptodate(page))
2141 cifs_dbg(FYI, "ppw - page not up to date\n");
2144 * Set the "writeback" flag, and clear "dirty" in the radix tree.
2146 * A writepage() implementation always needs to do either this,
2147 * or re-dirty the page with "redirty_page_for_writepage()" in
2148 * the case of a failure.
2150 * Just unlocking the page will cause the radix tree tag-bits
2151 * to fail to update with the state of the page correctly.
2153 set_page_writeback(page);
2155 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
2156 if (rc == -EAGAIN && wbc->sync_mode == WB_SYNC_ALL)
2158 else if (rc == -EAGAIN)
2159 redirty_page_for_writepage(wbc, page);
2163 SetPageUptodate(page);
2164 end_page_writeback(page);
2165 page_cache_release(page);
2170 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
2172 int rc = cifs_writepage_locked(page, wbc);
2177 static int cifs_write_end(struct file *file, struct address_space *mapping,
2178 loff_t pos, unsigned len, unsigned copied,
2179 struct page *page, void *fsdata)
2182 struct inode *inode = mapping->host;
2183 struct cifsFileInfo *cfile = file->private_data;
2184 struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
2187 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2190 pid = current->tgid;
2192 cifs_dbg(FYI, "write_end for page %p from pos %lld with %d bytes\n",
2195 if (PageChecked(page)) {
2197 SetPageUptodate(page);
2198 ClearPageChecked(page);
2199 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
2200 SetPageUptodate(page);
2202 if (!PageUptodate(page)) {
2204 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
2208 /* this is probably better than directly calling
2209 partialpage_write since in this function the file handle is
2210 known which we might as well leverage */
2211 /* BB check if anything else missing out of ppw
2212 such as updating last write time */
2213 page_data = kmap(page);
2214 rc = cifs_write(cfile, pid, page_data + offset, copied, &pos);
2215 /* if (rc < 0) should we set writebehind rc? */
2222 set_page_dirty(page);
2226 spin_lock(&inode->i_lock);
2227 if (pos > inode->i_size)
2228 i_size_write(inode, pos);
2229 spin_unlock(&inode->i_lock);
2233 page_cache_release(page);
2238 int cifs_strict_fsync(struct file *file, loff_t start, loff_t end,
2243 struct cifs_tcon *tcon;
2244 struct TCP_Server_Info *server;
2245 struct cifsFileInfo *smbfile = file->private_data;
2246 struct inode *inode = file_inode(file);
2247 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2249 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2252 mutex_lock(&inode->i_mutex);
2256 cifs_dbg(FYI, "Sync file - name: %s datasync: 0x%x\n",
2257 file->f_path.dentry->d_name.name, datasync);
2259 if (!CIFS_CACHE_READ(CIFS_I(inode))) {
2260 rc = cifs_zap_mapping(inode);
2262 cifs_dbg(FYI, "rc: %d during invalidate phase\n", rc);
2263 rc = 0; /* don't care about it in fsync */
2267 tcon = tlink_tcon(smbfile->tlink);
2268 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2269 server = tcon->ses->server;
2270 if (server->ops->flush)
2271 rc = server->ops->flush(xid, tcon, &smbfile->fid);
2277 mutex_unlock(&inode->i_mutex);
2281 int cifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2285 struct cifs_tcon *tcon;
2286 struct TCP_Server_Info *server;
2287 struct cifsFileInfo *smbfile = file->private_data;
2288 struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2289 struct inode *inode = file->f_mapping->host;
2291 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2294 mutex_lock(&inode->i_mutex);
2298 cifs_dbg(FYI, "Sync file - name: %s datasync: 0x%x\n",
2299 file->f_path.dentry->d_name.name, datasync);
2301 tcon = tlink_tcon(smbfile->tlink);
2302 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2303 server = tcon->ses->server;
2304 if (server->ops->flush)
2305 rc = server->ops->flush(xid, tcon, &smbfile->fid);
2311 mutex_unlock(&inode->i_mutex);
2316 * As file closes, flush all cached write data for this inode checking
2317 * for write behind errors.
2319 int cifs_flush(struct file *file, fl_owner_t id)
2321 struct inode *inode = file_inode(file);
2324 if (file->f_mode & FMODE_WRITE)
2325 rc = filemap_write_and_wait(inode->i_mapping);
2327 cifs_dbg(FYI, "Flush inode %p file %p rc %d\n", inode, file, rc);
2333 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
2338 for (i = 0; i < num_pages; i++) {
2339 pages[i] = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2342 * save number of pages we have already allocated and
2343 * return with ENOMEM error
2352 for (i = 0; i < num_pages; i++)
2359 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
2364 clen = min_t(const size_t, len, wsize);
2365 num_pages = DIV_ROUND_UP(clen, PAGE_SIZE);
2374 cifs_uncached_writedata_release(struct kref *refcount)
2377 struct cifs_writedata *wdata = container_of(refcount,
2378 struct cifs_writedata, refcount);
2380 for (i = 0; i < wdata->nr_pages; i++)
2381 put_page(wdata->pages[i]);
2382 cifs_writedata_release(refcount);
2386 cifs_uncached_writev_complete(struct work_struct *work)
2388 struct cifs_writedata *wdata = container_of(work,
2389 struct cifs_writedata, work);
2390 struct inode *inode = wdata->cfile->dentry->d_inode;
2391 struct cifsInodeInfo *cifsi = CIFS_I(inode);
2393 spin_lock(&inode->i_lock);
2394 cifs_update_eof(cifsi, wdata->offset, wdata->bytes);
2395 if (cifsi->server_eof > inode->i_size)
2396 i_size_write(inode, cifsi->server_eof);
2397 spin_unlock(&inode->i_lock);
2399 complete(&wdata->done);
2401 kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2404 /* attempt to send write to server, retry on any -EAGAIN errors */
2406 cifs_uncached_retry_writev(struct cifs_writedata *wdata)
2409 struct TCP_Server_Info *server;
2411 server = tlink_tcon(wdata->cfile->tlink)->ses->server;
2414 if (wdata->cfile->invalidHandle) {
2415 rc = cifs_reopen_file(wdata->cfile, false);
2419 rc = server->ops->async_writev(wdata,
2420 cifs_uncached_writedata_release);
2421 } while (rc == -EAGAIN);
2427 wdata_fill_from_iovec(struct cifs_writedata *wdata, struct iov_iter *from,
2428 size_t *len, unsigned long *num_pages)
2430 size_t save_len, copied, bytes, cur_len = *len;
2431 unsigned long i, nr_pages = *num_pages;
2434 for (i = 0; i < nr_pages; i++) {
2435 bytes = min_t(const size_t, cur_len, PAGE_SIZE);
2436 copied = copy_page_from_iter(wdata->pages[i], 0, bytes, from);
2439 * If we didn't copy as much as we expected, then that
2440 * may mean we trod into an unmapped area. Stop copying
2441 * at that point. On the next pass through the big
2442 * loop, we'll likely end up getting a zero-length
2443 * write and bailing out of it.
2448 cur_len = save_len - cur_len;
2452 * If we have no data to send, then that probably means that
2453 * the copy above failed altogether. That's most likely because
2454 * the address in the iovec was bogus. Return -EFAULT and let
2455 * the caller free anything we allocated and bail out.
2461 * i + 1 now represents the number of pages we actually used in
2462 * the copy phase above.
2469 cifs_write_from_iter(loff_t offset, size_t len, struct iov_iter *from,
2470 struct cifsFileInfo *open_file,
2471 struct cifs_sb_info *cifs_sb, struct list_head *wdata_list)
2475 unsigned long nr_pages, num_pages, i;
2476 struct cifs_writedata *wdata;
2479 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2480 pid = open_file->pid;
2482 pid = current->tgid;
2485 nr_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
2486 wdata = cifs_writedata_alloc(nr_pages,
2487 cifs_uncached_writev_complete);
2493 rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
2499 num_pages = nr_pages;
2500 rc = wdata_fill_from_iovec(wdata, from, &cur_len, &num_pages);
2502 for (i = 0; i < nr_pages; i++)
2503 put_page(wdata->pages[i]);
2509 * Bring nr_pages down to the number of pages we actually used,
2510 * and free any pages that we didn't use.
2512 for ( ; nr_pages > num_pages; nr_pages--)
2513 put_page(wdata->pages[nr_pages - 1]);
2515 wdata->sync_mode = WB_SYNC_ALL;
2516 wdata->nr_pages = nr_pages;
2517 wdata->offset = (__u64)offset;
2518 wdata->cfile = cifsFileInfo_get(open_file);
2520 wdata->bytes = cur_len;
2521 wdata->pagesz = PAGE_SIZE;
2522 wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
2523 rc = cifs_uncached_retry_writev(wdata);
2525 kref_put(&wdata->refcount,
2526 cifs_uncached_writedata_release);
2530 list_add_tail(&wdata->list, wdata_list);
2539 cifs_iovec_write(struct file *file, struct iov_iter *from, loff_t *poffset)
2542 ssize_t total_written = 0;
2543 struct cifsFileInfo *open_file;
2544 struct cifs_tcon *tcon;
2545 struct cifs_sb_info *cifs_sb;
2546 struct cifs_writedata *wdata, *tmp;
2547 struct list_head wdata_list;
2550 len = iov_iter_count(from);
2551 rc = generic_write_checks(file, poffset, &len, 0);
2558 iov_iter_truncate(from, len);
2560 INIT_LIST_HEAD(&wdata_list);
2561 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2562 open_file = file->private_data;
2563 tcon = tlink_tcon(open_file->tlink);
2565 if (!tcon->ses->server->ops->async_writev)
2568 rc = cifs_write_from_iter(*poffset, len, from, open_file, cifs_sb,
2572 * If at least one write was successfully sent, then discard any rc
2573 * value from the later writes. If the other write succeeds, then
2574 * we'll end up returning whatever was written. If it fails, then
2575 * we'll get a new rc value from that.
2577 if (!list_empty(&wdata_list))
2581 * Wait for and collect replies for any successful sends in order of
2582 * increasing offset. Once an error is hit or we get a fatal signal
2583 * while waiting, then return without waiting for any more replies.
2586 list_for_each_entry_safe(wdata, tmp, &wdata_list, list) {
2588 /* FIXME: freezable too? */
2589 rc = wait_for_completion_killable(&wdata->done);
2592 else if (wdata->result)
2595 total_written += wdata->bytes;
2597 /* resend call if it's a retryable error */
2598 if (rc == -EAGAIN) {
2599 rc = cifs_uncached_retry_writev(wdata);
2603 list_del_init(&wdata->list);
2604 kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2607 if (total_written > 0)
2608 *poffset += total_written;
2610 cifs_stats_bytes_written(tcon, total_written);
2611 return total_written ? total_written : (ssize_t)rc;
2614 ssize_t cifs_user_writev(struct kiocb *iocb, struct iov_iter *from)
2617 struct inode *inode;
2618 loff_t pos = iocb->ki_pos;
2620 inode = file_inode(iocb->ki_filp);
2623 * BB - optimize the way when signing is disabled. We can drop this
2624 * extra memory-to-memory copying and use iovec buffers for constructing
2628 written = cifs_iovec_write(iocb->ki_filp, from, &pos);
2630 set_bit(CIFS_INO_INVALID_MAPPING, &CIFS_I(inode)->flags);
2638 cifs_writev(struct kiocb *iocb, struct iov_iter *from)
2640 struct file *file = iocb->ki_filp;
2641 struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
2642 struct inode *inode = file->f_mapping->host;
2643 struct cifsInodeInfo *cinode = CIFS_I(inode);
2644 struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
2645 ssize_t rc = -EACCES;
2646 loff_t lock_pos = iocb->ki_pos;
2649 * We need to hold the sem to be sure nobody modifies lock list
2650 * with a brlock that prevents writing.
2652 down_read(&cinode->lock_sem);
2653 mutex_lock(&inode->i_mutex);
2654 if (file->f_flags & O_APPEND)
2655 lock_pos = i_size_read(inode);
2656 if (!cifs_find_lock_conflict(cfile, lock_pos, iov_iter_count(from),
2657 server->vals->exclusive_lock_type, NULL,
2659 rc = __generic_file_write_iter(iocb, from);
2660 mutex_unlock(&inode->i_mutex);
2665 err = generic_write_sync(file, iocb->ki_pos - rc, rc);
2670 mutex_unlock(&inode->i_mutex);
2672 up_read(&cinode->lock_sem);
2677 cifs_strict_writev(struct kiocb *iocb, struct iov_iter *from)
2679 struct inode *inode = file_inode(iocb->ki_filp);
2680 struct cifsInodeInfo *cinode = CIFS_I(inode);
2681 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2682 struct cifsFileInfo *cfile = (struct cifsFileInfo *)
2683 iocb->ki_filp->private_data;
2684 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
2687 written = cifs_get_writer(cinode);
2691 if (CIFS_CACHE_WRITE(cinode)) {
2692 if (cap_unix(tcon->ses) &&
2693 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability))
2694 && ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0)) {
2695 written = generic_file_write_iter(iocb, from);
2698 written = cifs_writev(iocb, from);
2702 * For non-oplocked files in strict cache mode we need to write the data
2703 * to the server exactly from the pos to pos+len-1 rather than flush all
2704 * affected pages because it may cause a error with mandatory locks on
2705 * these pages but not on the region from pos to ppos+len-1.
2707 written = cifs_user_writev(iocb, from);
2708 if (written > 0 && CIFS_CACHE_READ(cinode)) {
2710 * Windows 7 server can delay breaking level2 oplock if a write
2711 * request comes - break it on the client to prevent reading
2714 cifs_zap_mapping(inode);
2715 cifs_dbg(FYI, "Set no oplock for inode=%p after a write operation\n",
2720 cifs_put_writer(cinode);
2724 static struct cifs_readdata *
2725 cifs_readdata_alloc(unsigned int nr_pages, work_func_t complete)
2727 struct cifs_readdata *rdata;
2729 rdata = kzalloc(sizeof(*rdata) + (sizeof(struct page *) * nr_pages),
2731 if (rdata != NULL) {
2732 kref_init(&rdata->refcount);
2733 INIT_LIST_HEAD(&rdata->list);
2734 init_completion(&rdata->done);
2735 INIT_WORK(&rdata->work, complete);
2742 cifs_readdata_release(struct kref *refcount)
2744 struct cifs_readdata *rdata = container_of(refcount,
2745 struct cifs_readdata, refcount);
2748 cifsFileInfo_put(rdata->cfile);
2754 cifs_read_allocate_pages(struct cifs_readdata *rdata, unsigned int nr_pages)
2760 for (i = 0; i < nr_pages; i++) {
2761 page = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2766 rdata->pages[i] = page;
2770 for (i = 0; i < nr_pages; i++) {
2771 put_page(rdata->pages[i]);
2772 rdata->pages[i] = NULL;
2779 cifs_uncached_readdata_release(struct kref *refcount)
2781 struct cifs_readdata *rdata = container_of(refcount,
2782 struct cifs_readdata, refcount);
2785 for (i = 0; i < rdata->nr_pages; i++) {
2786 put_page(rdata->pages[i]);
2787 rdata->pages[i] = NULL;
2789 cifs_readdata_release(refcount);
2793 cifs_retry_async_readv(struct cifs_readdata *rdata)
2796 struct TCP_Server_Info *server;
2798 server = tlink_tcon(rdata->cfile->tlink)->ses->server;
2801 if (rdata->cfile->invalidHandle) {
2802 rc = cifs_reopen_file(rdata->cfile, true);
2806 rc = server->ops->async_readv(rdata);
2807 } while (rc == -EAGAIN);
2813 * cifs_readdata_to_iov - copy data from pages in response to an iovec
2814 * @rdata: the readdata response with list of pages holding data
2815 * @iter: destination for our data
2817 * This function copies data from a list of pages in a readdata response into
2818 * an array of iovecs. It will first calculate where the data should go
2819 * based on the info in the readdata and then copy the data into that spot.
2822 cifs_readdata_to_iov(struct cifs_readdata *rdata, struct iov_iter *iter)
2824 size_t remaining = rdata->bytes;
2827 for (i = 0; i < rdata->nr_pages; i++) {
2828 struct page *page = rdata->pages[i];
2829 size_t copy = min_t(size_t, remaining, PAGE_SIZE);
2830 size_t written = copy_page_to_iter(page, 0, copy, iter);
2831 remaining -= written;
2832 if (written < copy && iov_iter_count(iter) > 0)
2835 return remaining ? -EFAULT : 0;
2839 cifs_uncached_readv_complete(struct work_struct *work)
2841 struct cifs_readdata *rdata = container_of(work,
2842 struct cifs_readdata, work);
2844 complete(&rdata->done);
2845 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
2849 cifs_uncached_read_into_pages(struct TCP_Server_Info *server,
2850 struct cifs_readdata *rdata, unsigned int len)
2852 int total_read = 0, result = 0;
2854 unsigned int nr_pages = rdata->nr_pages;
2857 rdata->tailsz = PAGE_SIZE;
2858 for (i = 0; i < nr_pages; i++) {
2859 struct page *page = rdata->pages[i];
2861 if (len >= PAGE_SIZE) {
2862 /* enough data to fill the page */
2863 iov.iov_base = kmap(page);
2864 iov.iov_len = PAGE_SIZE;
2865 cifs_dbg(FYI, "%u: iov_base=%p iov_len=%zu\n",
2866 i, iov.iov_base, iov.iov_len);
2868 } else if (len > 0) {
2869 /* enough for partial page, fill and zero the rest */
2870 iov.iov_base = kmap(page);
2872 cifs_dbg(FYI, "%u: iov_base=%p iov_len=%zu\n",
2873 i, iov.iov_base, iov.iov_len);
2874 memset(iov.iov_base + len, '\0', PAGE_SIZE - len);
2875 rdata->tailsz = len;
2878 /* no need to hold page hostage */
2879 rdata->pages[i] = NULL;
2885 result = cifs_readv_from_socket(server, &iov, 1, iov.iov_len);
2890 total_read += result;
2893 return total_read > 0 && result != -EAGAIN ? total_read : result;
2896 ssize_t cifs_user_readv(struct kiocb *iocb, struct iov_iter *to)
2898 struct file *file = iocb->ki_filp;
2900 size_t len, cur_len;
2901 ssize_t total_read = 0;
2902 loff_t offset = iocb->ki_pos;
2903 unsigned int npages;
2904 struct cifs_sb_info *cifs_sb;
2905 struct cifs_tcon *tcon;
2906 struct cifsFileInfo *open_file;
2907 struct cifs_readdata *rdata, *tmp;
2908 struct list_head rdata_list;
2911 len = iov_iter_count(to);
2915 INIT_LIST_HEAD(&rdata_list);
2916 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2917 open_file = file->private_data;
2918 tcon = tlink_tcon(open_file->tlink);
2920 if (!tcon->ses->server->ops->async_readv)
2923 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2924 pid = open_file->pid;
2926 pid = current->tgid;
2928 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
2929 cifs_dbg(FYI, "attempting read on write only file instance\n");
2932 cur_len = min_t(const size_t, len - total_read, cifs_sb->rsize);
2933 npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
2935 /* allocate a readdata struct */
2936 rdata = cifs_readdata_alloc(npages,
2937 cifs_uncached_readv_complete);
2943 rc = cifs_read_allocate_pages(rdata, npages);
2947 rdata->cfile = cifsFileInfo_get(open_file);
2948 rdata->nr_pages = npages;
2949 rdata->offset = offset;
2950 rdata->bytes = cur_len;
2952 rdata->pagesz = PAGE_SIZE;
2953 rdata->read_into_pages = cifs_uncached_read_into_pages;
2955 rc = cifs_retry_async_readv(rdata);
2958 kref_put(&rdata->refcount,
2959 cifs_uncached_readdata_release);
2963 list_add_tail(&rdata->list, &rdata_list);
2968 /* if at least one read request send succeeded, then reset rc */
2969 if (!list_empty(&rdata_list))
2972 len = iov_iter_count(to);
2973 /* the loop below should proceed in the order of increasing offsets */
2974 list_for_each_entry_safe(rdata, tmp, &rdata_list, list) {
2977 /* FIXME: freezable sleep too? */
2978 rc = wait_for_completion_killable(&rdata->done);
2981 else if (rdata->result) {
2983 /* resend call if it's a retryable error */
2984 if (rc == -EAGAIN) {
2985 rc = cifs_retry_async_readv(rdata);
2989 rc = cifs_readdata_to_iov(rdata, to);
2993 list_del_init(&rdata->list);
2994 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
2997 total_read = len - iov_iter_count(to);
2999 cifs_stats_bytes_read(tcon, total_read);
3001 /* mask nodata case */
3006 iocb->ki_pos += total_read;
3013 cifs_strict_readv(struct kiocb *iocb, struct iov_iter *to)
3015 struct inode *inode = file_inode(iocb->ki_filp);
3016 struct cifsInodeInfo *cinode = CIFS_I(inode);
3017 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
3018 struct cifsFileInfo *cfile = (struct cifsFileInfo *)
3019 iocb->ki_filp->private_data;
3020 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
3024 * In strict cache mode we need to read from the server all the time
3025 * if we don't have level II oplock because the server can delay mtime
3026 * change - so we can't make a decision about inode invalidating.
3027 * And we can also fail with pagereading if there are mandatory locks
3028 * on pages affected by this read but not on the region from pos to
3031 if (!CIFS_CACHE_READ(cinode))
3032 return cifs_user_readv(iocb, to);
3034 if (cap_unix(tcon->ses) &&
3035 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
3036 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
3037 return generic_file_read_iter(iocb, to);
3040 * We need to hold the sem to be sure nobody modifies lock list
3041 * with a brlock that prevents reading.
3043 down_read(&cinode->lock_sem);
3044 if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(to),
3045 tcon->ses->server->vals->shared_lock_type,
3046 NULL, CIFS_READ_OP))
3047 rc = generic_file_read_iter(iocb, to);
3048 up_read(&cinode->lock_sem);
3053 cifs_read(struct file *file, char *read_data, size_t read_size, loff_t *offset)
3056 unsigned int bytes_read = 0;
3057 unsigned int total_read;
3058 unsigned int current_read_size;
3060 struct cifs_sb_info *cifs_sb;
3061 struct cifs_tcon *tcon;
3062 struct TCP_Server_Info *server;
3065 struct cifsFileInfo *open_file;
3066 struct cifs_io_parms io_parms;
3067 int buf_type = CIFS_NO_BUFFER;
3071 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
3073 /* FIXME: set up handlers for larger reads and/or convert to async */
3074 rsize = min_t(unsigned int, cifs_sb->rsize, CIFSMaxBufSize);
3076 if (file->private_data == NULL) {
3081 open_file = file->private_data;
3082 tcon = tlink_tcon(open_file->tlink);
3083 server = tcon->ses->server;
3085 if (!server->ops->sync_read) {
3090 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3091 pid = open_file->pid;
3093 pid = current->tgid;
3095 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3096 cifs_dbg(FYI, "attempting read on write only file instance\n");
3098 for (total_read = 0, cur_offset = read_data; read_size > total_read;
3099 total_read += bytes_read, cur_offset += bytes_read) {
3100 current_read_size = min_t(uint, read_size - total_read, rsize);
3102 * For windows me and 9x we do not want to request more than it
3103 * negotiated since it will refuse the read then.
3105 if ((tcon->ses) && !(tcon->ses->capabilities &
3106 tcon->ses->server->vals->cap_large_files)) {
3107 current_read_size = min_t(uint, current_read_size,
3111 while (rc == -EAGAIN) {
3112 if (open_file->invalidHandle) {
3113 rc = cifs_reopen_file(open_file, true);
3118 io_parms.tcon = tcon;
3119 io_parms.offset = *offset;
3120 io_parms.length = current_read_size;
3121 rc = server->ops->sync_read(xid, open_file, &io_parms,
3122 &bytes_read, &cur_offset,
3125 if (rc || (bytes_read == 0)) {
3133 cifs_stats_bytes_read(tcon, total_read);
3134 *offset += bytes_read;
3142 * If the page is mmap'ed into a process' page tables, then we need to make
3143 * sure that it doesn't change while being written back.
3146 cifs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
3148 struct page *page = vmf->page;
3151 return VM_FAULT_LOCKED;
3154 static struct vm_operations_struct cifs_file_vm_ops = {
3155 .fault = filemap_fault,
3156 .map_pages = filemap_map_pages,
3157 .page_mkwrite = cifs_page_mkwrite,
3158 .remap_pages = generic_file_remap_pages,
3161 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
3164 struct inode *inode = file_inode(file);
3168 if (!CIFS_CACHE_READ(CIFS_I(inode))) {
3169 rc = cifs_zap_mapping(inode);
3174 rc = generic_file_mmap(file, vma);
3176 vma->vm_ops = &cifs_file_vm_ops;
3181 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
3186 rc = cifs_revalidate_file(file);
3188 cifs_dbg(FYI, "Validation prior to mmap failed, error=%d\n",
3193 rc = generic_file_mmap(file, vma);
3195 vma->vm_ops = &cifs_file_vm_ops;
3201 cifs_readv_complete(struct work_struct *work)
3204 struct cifs_readdata *rdata = container_of(work,
3205 struct cifs_readdata, work);
3207 for (i = 0; i < rdata->nr_pages; i++) {
3208 struct page *page = rdata->pages[i];
3210 lru_cache_add_file(page);
3212 if (rdata->result == 0) {
3213 flush_dcache_page(page);
3214 SetPageUptodate(page);
3219 if (rdata->result == 0)
3220 cifs_readpage_to_fscache(rdata->mapping->host, page);
3222 page_cache_release(page);
3223 rdata->pages[i] = NULL;
3225 kref_put(&rdata->refcount, cifs_readdata_release);
3229 cifs_readpages_read_into_pages(struct TCP_Server_Info *server,
3230 struct cifs_readdata *rdata, unsigned int len)
3232 int total_read = 0, result = 0;
3236 unsigned int nr_pages = rdata->nr_pages;
3239 /* determine the eof that the server (probably) has */
3240 eof = CIFS_I(rdata->mapping->host)->server_eof;
3241 eof_index = eof ? (eof - 1) >> PAGE_CACHE_SHIFT : 0;
3242 cifs_dbg(FYI, "eof=%llu eof_index=%lu\n", eof, eof_index);
3244 rdata->tailsz = PAGE_CACHE_SIZE;
3245 for (i = 0; i < nr_pages; i++) {
3246 struct page *page = rdata->pages[i];
3248 if (len >= PAGE_CACHE_SIZE) {
3249 /* enough data to fill the page */
3250 iov.iov_base = kmap(page);
3251 iov.iov_len = PAGE_CACHE_SIZE;
3252 cifs_dbg(FYI, "%u: idx=%lu iov_base=%p iov_len=%zu\n",
3253 i, page->index, iov.iov_base, iov.iov_len);
3254 len -= PAGE_CACHE_SIZE;
3255 } else if (len > 0) {
3256 /* enough for partial page, fill and zero the rest */
3257 iov.iov_base = kmap(page);
3259 cifs_dbg(FYI, "%u: idx=%lu iov_base=%p iov_len=%zu\n",
3260 i, page->index, iov.iov_base, iov.iov_len);
3261 memset(iov.iov_base + len,
3262 '\0', PAGE_CACHE_SIZE - len);
3263 rdata->tailsz = len;
3265 } else if (page->index > eof_index) {
3267 * The VFS will not try to do readahead past the
3268 * i_size, but it's possible that we have outstanding
3269 * writes with gaps in the middle and the i_size hasn't
3270 * caught up yet. Populate those with zeroed out pages
3271 * to prevent the VFS from repeatedly attempting to
3272 * fill them until the writes are flushed.
3274 zero_user(page, 0, PAGE_CACHE_SIZE);
3275 lru_cache_add_file(page);
3276 flush_dcache_page(page);
3277 SetPageUptodate(page);
3279 page_cache_release(page);
3280 rdata->pages[i] = NULL;
3284 /* no need to hold page hostage */
3285 lru_cache_add_file(page);
3287 page_cache_release(page);
3288 rdata->pages[i] = NULL;
3293 result = cifs_readv_from_socket(server, &iov, 1, iov.iov_len);
3298 total_read += result;
3301 return total_read > 0 && result != -EAGAIN ? total_read : result;
3304 static int cifs_readpages(struct file *file, struct address_space *mapping,
3305 struct list_head *page_list, unsigned num_pages)
3308 struct list_head tmplist;
3309 struct cifsFileInfo *open_file = file->private_data;
3310 struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
3311 unsigned int rsize = cifs_sb->rsize;
3315 * Give up immediately if rsize is too small to read an entire page.
3316 * The VFS will fall back to readpage. We should never reach this
3317 * point however since we set ra_pages to 0 when the rsize is smaller
3318 * than a cache page.
3320 if (unlikely(rsize < PAGE_CACHE_SIZE))
3324 * Reads as many pages as possible from fscache. Returns -ENOBUFS
3325 * immediately if the cookie is negative
3327 * After this point, every page in the list might have PG_fscache set,
3328 * so we will need to clean that up off of every page we don't use.
3330 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
3335 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3336 pid = open_file->pid;
3338 pid = current->tgid;
3341 INIT_LIST_HEAD(&tmplist);
3343 cifs_dbg(FYI, "%s: file=%p mapping=%p num_pages=%u\n",
3344 __func__, file, mapping, num_pages);
3347 * Start with the page at end of list and move it to private
3348 * list. Do the same with any following pages until we hit
3349 * the rsize limit, hit an index discontinuity, or run out of
3350 * pages. Issue the async read and then start the loop again
3351 * until the list is empty.
3353 * Note that list order is important. The page_list is in
3354 * the order of declining indexes. When we put the pages in
3355 * the rdata->pages, then we want them in increasing order.
3357 while (!list_empty(page_list)) {
3359 unsigned int bytes = PAGE_CACHE_SIZE;
3360 unsigned int expected_index;
3361 unsigned int nr_pages = 1;
3363 struct page *page, *tpage;
3364 struct cifs_readdata *rdata;
3366 page = list_entry(page_list->prev, struct page, lru);
3369 * Lock the page and put it in the cache. Since no one else
3370 * should have access to this page, we're safe to simply set
3371 * PG_locked without checking it first.
3373 __set_page_locked(page);
3374 rc = add_to_page_cache_locked(page, mapping,
3375 page->index, GFP_KERNEL);
3377 /* give up if we can't stick it in the cache */
3379 __clear_page_locked(page);
3383 /* move first page to the tmplist */
3384 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3385 list_move_tail(&page->lru, &tmplist);
3387 /* now try and add more pages onto the request */
3388 expected_index = page->index + 1;
3389 list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
3390 /* discontinuity ? */
3391 if (page->index != expected_index)
3394 /* would this page push the read over the rsize? */
3395 if (bytes + PAGE_CACHE_SIZE > rsize)
3398 __set_page_locked(page);
3399 if (add_to_page_cache_locked(page, mapping,
3400 page->index, GFP_KERNEL)) {
3401 __clear_page_locked(page);
3404 list_move_tail(&page->lru, &tmplist);
3405 bytes += PAGE_CACHE_SIZE;
3410 rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
3412 /* best to give up if we're out of mem */
3413 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3414 list_del(&page->lru);
3415 lru_cache_add_file(page);
3417 page_cache_release(page);
3423 rdata->cfile = cifsFileInfo_get(open_file);
3424 rdata->mapping = mapping;
3425 rdata->offset = offset;
3426 rdata->bytes = bytes;
3428 rdata->pagesz = PAGE_CACHE_SIZE;
3429 rdata->read_into_pages = cifs_readpages_read_into_pages;
3431 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3432 list_del(&page->lru);
3433 rdata->pages[rdata->nr_pages++] = page;
3436 rc = cifs_retry_async_readv(rdata);
3438 for (i = 0; i < rdata->nr_pages; i++) {
3439 page = rdata->pages[i];
3440 lru_cache_add_file(page);
3442 page_cache_release(page);
3444 kref_put(&rdata->refcount, cifs_readdata_release);
3448 kref_put(&rdata->refcount, cifs_readdata_release);
3451 /* Any pages that have been shown to fscache but didn't get added to
3452 * the pagecache must be uncached before they get returned to the
3455 cifs_fscache_readpages_cancel(mapping->host, page_list);
3460 * cifs_readpage_worker must be called with the page pinned
3462 static int cifs_readpage_worker(struct file *file, struct page *page,
3468 /* Is the page cached? */
3469 rc = cifs_readpage_from_fscache(file_inode(file), page);
3473 read_data = kmap(page);
3474 /* for reads over a certain size could initiate async read ahead */
3476 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
3481 cifs_dbg(FYI, "Bytes read %d\n", rc);
3483 file_inode(file)->i_atime =
3484 current_fs_time(file_inode(file)->i_sb);
3486 if (PAGE_CACHE_SIZE > rc)
3487 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
3489 flush_dcache_page(page);
3490 SetPageUptodate(page);
3492 /* send this page to the cache */
3493 cifs_readpage_to_fscache(file_inode(file), page);
3505 static int cifs_readpage(struct file *file, struct page *page)
3507 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3513 if (file->private_data == NULL) {
3519 cifs_dbg(FYI, "readpage %p at offset %d 0x%x\n",
3520 page, (int)offset, (int)offset);
3522 rc = cifs_readpage_worker(file, page, &offset);
3528 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
3530 struct cifsFileInfo *open_file;
3532 spin_lock(&cifs_file_list_lock);
3533 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
3534 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
3535 spin_unlock(&cifs_file_list_lock);
3539 spin_unlock(&cifs_file_list_lock);
3543 /* We do not want to update the file size from server for inodes
3544 open for write - to avoid races with writepage extending
3545 the file - in the future we could consider allowing
3546 refreshing the inode only on increases in the file size
3547 but this is tricky to do without racing with writebehind
3548 page caching in the current Linux kernel design */
3549 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
3554 if (is_inode_writable(cifsInode)) {
3555 /* This inode is open for write at least once */
3556 struct cifs_sb_info *cifs_sb;
3558 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
3559 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
3560 /* since no page cache to corrupt on directio
3561 we can change size safely */
3565 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
3573 static int cifs_write_begin(struct file *file, struct address_space *mapping,
3574 loff_t pos, unsigned len, unsigned flags,
3575 struct page **pagep, void **fsdata)
3578 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
3579 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
3580 loff_t page_start = pos & PAGE_MASK;
3585 cifs_dbg(FYI, "write_begin from %lld len %d\n", (long long)pos, len);
3588 page = grab_cache_page_write_begin(mapping, index, flags);
3594 if (PageUptodate(page))
3598 * If we write a full page it will be up to date, no need to read from
3599 * the server. If the write is short, we'll end up doing a sync write
3602 if (len == PAGE_CACHE_SIZE)
3606 * optimize away the read when we have an oplock, and we're not
3607 * expecting to use any of the data we'd be reading in. That
3608 * is, when the page lies beyond the EOF, or straddles the EOF
3609 * and the write will cover all of the existing data.
3611 if (CIFS_CACHE_READ(CIFS_I(mapping->host))) {
3612 i_size = i_size_read(mapping->host);
3613 if (page_start >= i_size ||
3614 (offset == 0 && (pos + len) >= i_size)) {
3615 zero_user_segments(page, 0, offset,
3619 * PageChecked means that the parts of the page
3620 * to which we're not writing are considered up
3621 * to date. Once the data is copied to the
3622 * page, it can be set uptodate.
3624 SetPageChecked(page);
3629 if ((file->f_flags & O_ACCMODE) != O_WRONLY && !oncethru) {
3631 * might as well read a page, it is fast enough. If we get
3632 * an error, we don't need to return it. cifs_write_end will
3633 * do a sync write instead since PG_uptodate isn't set.
3635 cifs_readpage_worker(file, page, &page_start);
3636 page_cache_release(page);
3640 /* we could try using another file handle if there is one -
3641 but how would we lock it to prevent close of that handle
3642 racing with this read? In any case
3643 this will be written out by write_end so is fine */
3650 static int cifs_release_page(struct page *page, gfp_t gfp)
3652 if (PagePrivate(page))
3655 return cifs_fscache_release_page(page, gfp);
3658 static void cifs_invalidate_page(struct page *page, unsigned int offset,
3659 unsigned int length)
3661 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
3663 if (offset == 0 && length == PAGE_CACHE_SIZE)
3664 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
3667 static int cifs_launder_page(struct page *page)
3670 loff_t range_start = page_offset(page);
3671 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
3672 struct writeback_control wbc = {
3673 .sync_mode = WB_SYNC_ALL,
3675 .range_start = range_start,
3676 .range_end = range_end,
3679 cifs_dbg(FYI, "Launder page: %p\n", page);
3681 if (clear_page_dirty_for_io(page))
3682 rc = cifs_writepage_locked(page, &wbc);
3684 cifs_fscache_invalidate_page(page, page->mapping->host);
3689 cifs_pending_writers_wait(void *unused)
3695 void cifs_oplock_break(struct work_struct *work)
3697 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
3699 struct inode *inode = cfile->dentry->d_inode;
3700 struct cifsInodeInfo *cinode = CIFS_I(inode);
3701 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
3702 struct TCP_Server_Info *server = tcon->ses->server;
3705 wait_on_bit(&cinode->flags, CIFS_INODE_PENDING_WRITERS,
3706 cifs_pending_writers_wait, TASK_UNINTERRUPTIBLE);
3708 server->ops->downgrade_oplock(server, cinode,
3709 test_bit(CIFS_INODE_DOWNGRADE_OPLOCK_TO_L2, &cinode->flags));
3711 if (!CIFS_CACHE_WRITE(cinode) && CIFS_CACHE_READ(cinode) &&
3712 cifs_has_mand_locks(cinode)) {
3713 cifs_dbg(FYI, "Reset oplock to None for inode=%p due to mand locks\n",
3718 if (inode && S_ISREG(inode->i_mode)) {
3719 if (CIFS_CACHE_READ(cinode))
3720 break_lease(inode, O_RDONLY);
3722 break_lease(inode, O_WRONLY);
3723 rc = filemap_fdatawrite(inode->i_mapping);
3724 if (!CIFS_CACHE_READ(cinode)) {
3725 rc = filemap_fdatawait(inode->i_mapping);
3726 mapping_set_error(inode->i_mapping, rc);
3727 cifs_zap_mapping(inode);
3729 cifs_dbg(FYI, "Oplock flush inode %p rc %d\n", inode, rc);
3732 rc = cifs_push_locks(cfile);
3734 cifs_dbg(VFS, "Push locks rc = %d\n", rc);
3737 * releasing stale oplock after recent reconnect of smb session using
3738 * a now incorrect file handle is not a data integrity issue but do
3739 * not bother sending an oplock release if session to server still is
3740 * disconnected since oplock already released by the server
3742 if (!cfile->oplock_break_cancelled) {
3743 rc = tcon->ses->server->ops->oplock_response(tcon, &cfile->fid,
3745 cifs_dbg(FYI, "Oplock release rc = %d\n", rc);
3747 cifs_done_oplock_break(cinode);
3751 * The presence of cifs_direct_io() in the address space ops vector
3752 * allowes open() O_DIRECT flags which would have failed otherwise.
3754 * In the non-cached mode (mount with cache=none), we shunt off direct read and write requests
3755 * so this method should never be called.
3757 * Direct IO is not yet supported in the cached mode.
3760 cifs_direct_io(int rw, struct kiocb *iocb, struct iov_iter *iter,
3765 * Eventually need to support direct IO for non forcedirectio mounts
3771 const struct address_space_operations cifs_addr_ops = {
3772 .readpage = cifs_readpage,
3773 .readpages = cifs_readpages,
3774 .writepage = cifs_writepage,
3775 .writepages = cifs_writepages,
3776 .write_begin = cifs_write_begin,
3777 .write_end = cifs_write_end,
3778 .set_page_dirty = __set_page_dirty_nobuffers,
3779 .releasepage = cifs_release_page,
3780 .direct_IO = cifs_direct_io,
3781 .invalidatepage = cifs_invalidate_page,
3782 .launder_page = cifs_launder_page,
3786 * cifs_readpages requires the server to support a buffer large enough to
3787 * contain the header plus one complete page of data. Otherwise, we need
3788 * to leave cifs_readpages out of the address space operations.
3790 const struct address_space_operations cifs_addr_ops_smallbuf = {
3791 .readpage = cifs_readpage,
3792 .writepage = cifs_writepage,
3793 .writepages = cifs_writepages,
3794 .write_begin = cifs_write_begin,
3795 .write_end = cifs_write_end,
3796 .set_page_dirty = __set_page_dirty_nobuffers,
3797 .releasepage = cifs_release_page,
3798 .invalidatepage = cifs_invalidate_page,
3799 .launder_page = cifs_launder_page,