1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * mmap.c
5 *
6 * Code to deal with the mess that is clustered mmap.
7 *
8 * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9 *
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.
14 *
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.
19 *
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.
24 */
25
26#include <linux/fs.h>
27#include <linux/types.h>
28#include <linux/highmem.h>
29#include <linux/pagemap.h>
30#include <linux/uio.h>
31#include <linux/signal.h>
32#include <linux/rbtree.h>
33
34#define MLOG_MASK_PREFIX ML_FILE_IO
35#include <cluster/masklog.h>
36
37#include "ocfs2.h"
38
39#include "aops.h"
40#include "dlmglue.h"
41#include "file.h"
42#include "inode.h"
43#include "mmap.h"
44#include "super.h"
45
46
47static int ocfs2_fault(struct vm_area_struct *area, struct vm_fault *vmf)
48{
49	sigset_t oldset;
50	int ret;
51
52	mlog_entry("(area=%p, page offset=%lu)\n", area, vmf->pgoff);
53
54	ocfs2_block_signals(&oldset);
55	ret = filemap_fault(area, vmf);
56	ocfs2_unblock_signals(&oldset);
57
58	mlog_exit_ptr(vmf->page);
59	return ret;
60}
61
62static int __ocfs2_page_mkwrite(struct inode *inode, struct buffer_head *di_bh,
63				struct page *page)
64{
65	int ret;
66	struct address_space *mapping = inode->i_mapping;
67	loff_t pos = page_offset(page);
68	unsigned int len = PAGE_CACHE_SIZE;
69	pgoff_t last_index;
70	struct page *locked_page = NULL;
71	void *fsdata;
72	loff_t size = i_size_read(inode);
73
74	/*
75	 * Another node might have truncated while we were waiting on
76	 * cluster locks.
77	 * We don't check size == 0 before the shift. This is borrowed
78	 * from do_generic_file_read.
79	 */
80	last_index = (size - 1) >> PAGE_CACHE_SHIFT;
81	if (unlikely(!size || page->index > last_index)) {
82		ret = -EINVAL;
83		goto out;
84	}
85
86	/*
87	 * The i_size check above doesn't catch the case where nodes
88	 * truncated and then re-extended the file. We'll re-check the
89	 * page mapping after taking the page lock inside of
90	 * ocfs2_write_begin_nolock().
91	 */
92	if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
93		/*
94		 * the page has been umapped in ocfs2_data_downconvert_worker.
95		 * So return 0 here and let VFS retry.
96		 */
97		ret = 0;
98		goto out;
99	}
100
101	/*
102	 * Call ocfs2_write_begin() and ocfs2_write_end() to take
103	 * advantage of the allocation code there. We pass a write
104	 * length of the whole page (chopped to i_size) to make sure
105	 * the whole thing is allocated.
106	 *
107	 * Since we know the page is up to date, we don't have to
108	 * worry about ocfs2_write_begin() skipping some buffer reads
109	 * because the "write" would invalidate their data.
110	 */
111	if (page->index == last_index)
112		len = ((size - 1) & ~PAGE_CACHE_MASK) + 1;
113
114	ret = ocfs2_write_begin_nolock(mapping, pos, len, 0, &locked_page,
115				       &fsdata, di_bh, page);
116	if (ret) {
117		if (ret != -ENOSPC)
118			mlog_errno(ret);
119		goto out;
120	}
121
122	ret = ocfs2_write_end_nolock(mapping, pos, len, len, locked_page,
123				     fsdata);
124	if (ret < 0) {
125		mlog_errno(ret);
126		goto out;
127	}
128	BUG_ON(ret != len);
129	ret = 0;
130out:
131	return ret;
132}
133
134static int ocfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
135{
136	struct page *page = vmf->page;
137	struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
138	struct buffer_head *di_bh = NULL;
139	sigset_t oldset;
140	int ret;
141
142	ocfs2_block_signals(&oldset);
143
144	/*
145	 * The cluster locks taken will block a truncate from another
146	 * node. Taking the data lock will also ensure that we don't
147	 * attempt page truncation as part of a downconvert.
148	 */
149	ret = ocfs2_inode_lock(inode, &di_bh, 1);
150	if (ret < 0) {
151		mlog_errno(ret);
152		goto out;
153	}
154
155	/*
156	 * The alloc sem should be enough to serialize with
157	 * ocfs2_truncate_file() changing i_size as well as any thread
158	 * modifying the inode btree.
159	 */
160	down_write(&OCFS2_I(inode)->ip_alloc_sem);
161
162	ret = __ocfs2_page_mkwrite(inode, di_bh, page);
163
164	up_write(&OCFS2_I(inode)->ip_alloc_sem);
165
166	brelse(di_bh);
167	ocfs2_inode_unlock(inode, 1);
168
169out:
170	ocfs2_unblock_signals(&oldset);
171	if (ret)
172		ret = VM_FAULT_SIGBUS;
173	return ret;
174}
175
176static const struct vm_operations_struct ocfs2_file_vm_ops = {
177	.fault		= ocfs2_fault,
178	.page_mkwrite	= ocfs2_page_mkwrite,
179};
180
181int ocfs2_mmap(struct file *file, struct vm_area_struct *vma)
182{
183	int ret = 0, lock_level = 0;
184
185	ret = ocfs2_inode_lock_atime(file->f_dentry->d_inode,
186				    file->f_vfsmnt, &lock_level);
187	if (ret < 0) {
188		mlog_errno(ret);
189		goto out;
190	}
191	ocfs2_inode_unlock(file->f_dentry->d_inode, lock_level);
192out:
193	vma->vm_ops = &ocfs2_file_vm_ops;
194	vma->vm_flags |= VM_CAN_NONLINEAR;
195	return 0;
196}
197