1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Code to deal with the mess that is clustered mmap.
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program 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 the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/highmem.h>
30 #include <linux/pagemap.h>
31 #include <linux/uio.h>
32 #include <linux/signal.h>
33 #include <linux/rbtree.h>
35 #define MLOG_MASK_PREFIX ML_FILE_IO
36 #include <cluster/masklog.h>
46 static inline int ocfs2_vm_op_block_sigs(sigset_t
*blocked
, sigset_t
*oldset
)
48 /* The best way to deal with signals in the vm path is
49 * to block them upfront, rather than allowing the
50 * locking paths to return -ERESTARTSYS. */
53 /* We should technically never get a bad return value
55 return sigprocmask(SIG_BLOCK
, blocked
, oldset
);
58 static inline int ocfs2_vm_op_unblock_sigs(sigset_t
*oldset
)
60 return sigprocmask(SIG_SETMASK
, oldset
, NULL
);
63 static struct page
*ocfs2_nopage(struct vm_area_struct
* area
,
64 unsigned long address
,
67 struct page
*page
= NOPAGE_SIGBUS
;
68 sigset_t blocked
, oldset
;
71 mlog_entry("(area=%p, address=%lu, type=%p)\n", area
, address
,
74 ret
= ocfs2_vm_op_block_sigs(&blocked
, &oldset
);
80 page
= filemap_nopage(area
, address
, type
);
82 ret
= ocfs2_vm_op_unblock_sigs(&oldset
);
90 static int __ocfs2_page_mkwrite(struct inode
*inode
, struct buffer_head
*di_bh
,
94 struct address_space
*mapping
= inode
->i_mapping
;
95 loff_t pos
= page
->index
<< PAGE_CACHE_SHIFT
;
96 unsigned int len
= PAGE_CACHE_SIZE
;
98 struct page
*locked_page
= NULL
;
100 loff_t size
= i_size_read(inode
);
103 * Another node might have truncated while we were waiting on
106 last_index
= size
>> PAGE_CACHE_SHIFT
;
107 if (page
->index
> last_index
) {
113 * The i_size check above doesn't catch the case where nodes
114 * truncated and then re-extended the file. We'll re-check the
115 * page mapping after taking the page lock inside of
116 * ocfs2_write_begin_nolock().
118 if (!PageUptodate(page
) || page
->mapping
!= inode
->i_mapping
) {
124 * Call ocfs2_write_begin() and ocfs2_write_end() to take
125 * advantage of the allocation code there. We pass a write
126 * length of the whole page (chopped to i_size) to make sure
127 * the whole thing is allocated.
129 * Since we know the page is up to date, we don't have to
130 * worry about ocfs2_write_begin() skipping some buffer reads
131 * because the "write" would invalidate their data.
133 if (page
->index
== last_index
)
134 len
= size
& ~PAGE_CACHE_MASK
;
136 ret
= ocfs2_write_begin_nolock(mapping
, pos
, len
, 0, &locked_page
,
137 &fsdata
, di_bh
, page
);
144 ret
= ocfs2_write_end_nolock(mapping
, pos
, len
, len
, locked_page
,
156 static int ocfs2_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
158 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
159 struct buffer_head
*di_bh
= NULL
;
160 sigset_t blocked
, oldset
;
163 ret
= ocfs2_vm_op_block_sigs(&blocked
, &oldset
);
170 * The cluster locks taken will block a truncate from another
171 * node. Taking the data lock will also ensure that we don't
172 * attempt page truncation as part of a downconvert.
174 ret
= ocfs2_meta_lock(inode
, &di_bh
, 1);
181 * The alloc sem should be enough to serialize with
182 * ocfs2_truncate_file() changing i_size as well as any thread
183 * modifying the inode btree.
185 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
187 ret
= ocfs2_data_lock(inode
, 1);
190 goto out_meta_unlock
;
193 ret
= __ocfs2_page_mkwrite(inode
, di_bh
, page
);
195 ocfs2_data_unlock(inode
, 1);
198 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
201 ocfs2_meta_unlock(inode
, 1);
204 ret2
= ocfs2_vm_op_unblock_sigs(&oldset
);
211 static struct vm_operations_struct ocfs2_file_vm_ops
= {
212 .nopage
= ocfs2_nopage
,
213 .page_mkwrite
= ocfs2_page_mkwrite
,
216 int ocfs2_mmap(struct file
*file
, struct vm_area_struct
*vma
)
218 int ret
= 0, lock_level
= 0;
220 ret
= ocfs2_meta_lock_atime(file
->f_dentry
->d_inode
,
221 file
->f_vfsmnt
, &lock_level
);
226 ocfs2_meta_unlock(file
->f_dentry
->d_inode
, lock_level
);
228 vma
->vm_ops
= &ocfs2_file_vm_ops
;
229 vma
->vm_flags
|= VM_CAN_INVALIDATE
;