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_fault(struct vm_area_struct *area,
64 struct fault_data *fdata)
66 struct page *page = NULL;
67 sigset_t blocked, oldset;
70 mlog_entry("(area=%p, page offset=%lu)\n", area, fdata->pgoff);
72 ret = ocfs2_vm_op_block_sigs(&blocked, &oldset);
74 fdata->type = VM_FAULT_SIGBUS;
79 page = filemap_fault(area, fdata);
81 ret = ocfs2_vm_op_unblock_sigs(&oldset);
89 static int __ocfs2_page_mkwrite(struct inode *inode, struct buffer_head *di_bh,
93 struct address_space *mapping = inode->i_mapping;
94 loff_t pos = page->index << PAGE_CACHE_SHIFT;
95 unsigned int len = PAGE_CACHE_SIZE;
97 struct page *locked_page = NULL;
99 loff_t size = i_size_read(inode);
102 * Another node might have truncated while we were waiting on
105 last_index = size >> PAGE_CACHE_SHIFT;
106 if (page->index > last_index) {
112 * The i_size check above doesn't catch the case where nodes
113 * truncated and then re-extended the file. We'll re-check the
114 * page mapping after taking the page lock inside of
115 * ocfs2_write_begin_nolock().
117 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
123 * Call ocfs2_write_begin() and ocfs2_write_end() to take
124 * advantage of the allocation code there. We pass a write
125 * length of the whole page (chopped to i_size) to make sure
126 * the whole thing is allocated.
128 * Since we know the page is up to date, we don't have to
129 * worry about ocfs2_write_begin() skipping some buffer reads
130 * because the "write" would invalidate their data.
132 if (page->index == last_index)
133 len = size & ~PAGE_CACHE_MASK;
135 ret = ocfs2_write_begin_nolock(mapping, pos, len, 0, &locked_page,
136 &fsdata, di_bh, page);
143 ret = ocfs2_write_end_nolock(mapping, pos, len, len, locked_page,
155 static int ocfs2_page_mkwrite(struct vm_area_struct *vma, struct page *page)
157 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
158 struct buffer_head *di_bh = NULL;
159 sigset_t blocked, oldset;
162 ret = ocfs2_vm_op_block_sigs(&blocked, &oldset);
169 * The cluster locks taken will block a truncate from another
170 * node. Taking the data lock will also ensure that we don't
171 * attempt page truncation as part of a downconvert.
173 ret = ocfs2_meta_lock(inode, &di_bh, 1);
180 * The alloc sem should be enough to serialize with
181 * ocfs2_truncate_file() changing i_size as well as any thread
182 * modifying the inode btree.
184 down_write(&OCFS2_I(inode)->ip_alloc_sem);
186 ret = ocfs2_data_lock(inode, 1);
189 goto out_meta_unlock;
192 ret = __ocfs2_page_mkwrite(inode, di_bh, page);
194 ocfs2_data_unlock(inode, 1);
197 up_write(&OCFS2_I(inode)->ip_alloc_sem);
200 ocfs2_meta_unlock(inode, 1);
203 ret2 = ocfs2_vm_op_unblock_sigs(&oldset);
210 static struct vm_operations_struct ocfs2_file_vm_ops = {
211 .fault = ocfs2_fault,
212 .page_mkwrite = ocfs2_page_mkwrite,
215 int ocfs2_mmap(struct file *file, struct vm_area_struct *vma)
217 int ret = 0, lock_level = 0;
219 ret = ocfs2_meta_lock_atime(file->f_dentry->d_inode,
220 file->f_vfsmnt, &lock_level);
225 ocfs2_meta_unlock(file->f_dentry->d_inode, lock_level);
227 vma->vm_ops = &ocfs2_file_vm_ops;
228 vma->vm_flags |= VM_CAN_INVALIDATE | VM_CAN_NONLINEAR;