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#include <cluster/masklog.h>
35
36#include "ocfs2.h"
37
38#include "aops.h"
39#include "dlmglue.h"
40#include "file.h"
41#include "inode.h"
42#include "mmap.h"
43#include "super.h"
44#include "ocfs2_trace.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	ocfs2_block_signals(&oldset);
53	ret = filemap_fault(area, vmf);
54	ocfs2_unblock_signals(&oldset);
55
56	trace_ocfs2_fault(OCFS2_I(area->vm_file->f_mapping->host)->ip_blkno,
57			  area, vmf->page, vmf->pgoff);
58	return ret;
59}
60
61static int __ocfs2_page_mkwrite(struct file *file, struct buffer_head *di_bh,
62				struct page *page)
63{
64	int ret = VM_FAULT_NOPAGE;
65	struct inode *inode = file_inode(file);
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	last_index = (size - 1) >> PAGE_CACHE_SHIFT;
75
76	/*
77	 * There are cases that lead to the page no longer bebongs to the
78	 * mapping.
79	 * 1) pagecache truncates locally due to memory pressure.
80	 * 2) pagecache truncates when another is taking EX lock against
81	 * inode lock. see ocfs2_data_convert_worker.
82	 *
83	 * The i_size check doesn't catch the case where nodes truncated and
84	 * then re-extended the file. We'll re-check the page mapping after
85	 * taking the page lock inside of ocfs2_write_begin_nolock().
86	 *
87	 * Let VM retry with these cases.
88	 */
89	if ((page->mapping != inode->i_mapping) ||
90	    (!PageUptodate(page)) ||
91	    (page_offset(page) >= size))
92		goto out;
93
94	/*
95	 * Call ocfs2_write_begin() and ocfs2_write_end() to take
96	 * advantage of the allocation code there. We pass a write
97	 * length of the whole page (chopped to i_size) to make sure
98	 * the whole thing is allocated.
99	 *
100	 * Since we know the page is up to date, we don't have to
101	 * worry about ocfs2_write_begin() skipping some buffer reads
102	 * because the "write" would invalidate their data.
103	 */
104	if (page->index == last_index)
105		len = ((size - 1) & ~PAGE_CACHE_MASK) + 1;
106
107	ret = ocfs2_write_begin_nolock(file, mapping, pos, len, 0, &locked_page,
108				       &fsdata, di_bh, page);
109	if (ret) {
110		if (ret != -ENOSPC)
111			mlog_errno(ret);
112		if (ret == -ENOMEM)
113			ret = VM_FAULT_OOM;
114		else
115			ret = VM_FAULT_SIGBUS;
116		goto out;
117	}
118
119	if (!locked_page) {
120		ret = VM_FAULT_NOPAGE;
121		goto out;
122	}
123	ret = ocfs2_write_end_nolock(mapping, pos, len, len, locked_page,
124				     fsdata);
125	BUG_ON(ret != len);
126	ret = VM_FAULT_LOCKED;
127out:
128	return ret;
129}
130
131static int ocfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
132{
133	struct page *page = vmf->page;
134	struct inode *inode = file_inode(vma->vm_file);
135	struct buffer_head *di_bh = NULL;
136	sigset_t oldset;
137	int ret;
138
139	sb_start_pagefault(inode->i_sb);
140	ocfs2_block_signals(&oldset);
141
142	/*
143	 * The cluster locks taken will block a truncate from another
144	 * node. Taking the data lock will also ensure that we don't
145	 * attempt page truncation as part of a downconvert.
146	 */
147	ret = ocfs2_inode_lock(inode, &di_bh, 1);
148	if (ret < 0) {
149		mlog_errno(ret);
150		goto out;
151	}
152
153	/*
154	 * The alloc sem should be enough to serialize with
155	 * ocfs2_truncate_file() changing i_size as well as any thread
156	 * modifying the inode btree.
157	 */
158	down_write(&OCFS2_I(inode)->ip_alloc_sem);
159
160	ret = __ocfs2_page_mkwrite(vma->vm_file, di_bh, page);
161
162	up_write(&OCFS2_I(inode)->ip_alloc_sem);
163
164	brelse(di_bh);
165	ocfs2_inode_unlock(inode, 1);
166
167out:
168	ocfs2_unblock_signals(&oldset);
169	sb_end_pagefault(inode->i_sb);
170	return ret;
171}
172
173static const struct vm_operations_struct ocfs2_file_vm_ops = {
174	.fault		= ocfs2_fault,
175	.page_mkwrite	= ocfs2_page_mkwrite,
176};
177
178int ocfs2_mmap(struct file *file, struct vm_area_struct *vma)
179{
180	int ret = 0, lock_level = 0;
181
182	ret = ocfs2_inode_lock_atime(file_inode(file),
183				    file->f_path.mnt, &lock_level);
184	if (ret < 0) {
185		mlog_errno(ret);
186		goto out;
187	}
188	ocfs2_inode_unlock(file_inode(file), lock_level);
189out:
190	vma->vm_ops = &ocfs2_file_vm_ops;
191	return 0;
192}
193
194