blob: 9cd41083e99123eca1c48085fb39809e6b906b40 [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- 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>
Mark Fashehccd979b2005-12-15 14:31:24 -080028#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
Mark Fashehccd979b2005-12-15 14:31:24 -080034#include <cluster/masklog.h>
35
36#include "ocfs2.h"
37
Mark Fasheh7307de82007-05-09 15:16:19 -070038#include "aops.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080039#include "dlmglue.h"
40#include "file.h"
41#include "inode.h"
42#include "mmap.h"
Joel Beckere4b963f2009-09-02 17:17:36 -070043#include "super.h"
Tao Ma614a9e82011-02-22 21:59:46 +080044#include "ocfs2_trace.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080045
Mark Fasheh7307de82007-05-09 15:16:19 -070046
Nick Piggind0217ac2007-07-19 01:47:03 -070047static int ocfs2_fault(struct vm_area_struct *area, struct vm_fault *vmf)
Mark Fashehccd979b2005-12-15 14:31:24 -080048{
Joel Beckere4b963f2009-09-02 17:17:36 -070049 sigset_t oldset;
50 int ret;
Mark Fashehccd979b2005-12-15 14:31:24 -080051
Joel Beckere4b963f2009-09-02 17:17:36 -070052 ocfs2_block_signals(&oldset);
Nick Piggind0217ac2007-07-19 01:47:03 -070053 ret = filemap_fault(area, vmf);
Joel Beckere4b963f2009-09-02 17:17:36 -070054 ocfs2_unblock_signals(&oldset);
Mark Fashehccd979b2005-12-15 14:31:24 -080055
Tao Ma614a9e82011-02-22 21:59:46 +080056 trace_ocfs2_fault(OCFS2_I(area->vm_file->f_mapping->host)->ip_blkno,
57 area, vmf->page, vmf->pgoff);
Nick Piggind0217ac2007-07-19 01:47:03 -070058 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -080059}
60
Tao Ma0378da0f2010-08-12 10:25:28 +080061static int __ocfs2_page_mkwrite(struct file *file, struct buffer_head *di_bh,
Mark Fasheh7307de82007-05-09 15:16:19 -070062 struct page *page)
63{
Wengang Wang5cffff92011-07-24 10:36:54 -070064 int ret = VM_FAULT_NOPAGE;
Tao Ma0378da0f2010-08-12 10:25:28 +080065 struct inode *inode = file->f_path.dentry->d_inode;
Mark Fasheh7307de82007-05-09 15:16:19 -070066 struct address_space *mapping = inode->i_mapping;
Nick Piggin18336332007-07-20 00:31:45 -070067 loff_t pos = page_offset(page);
Mark Fasheh7307de82007-05-09 15:16:19 -070068 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
Tao Maf63afdb2010-07-17 21:45:49 +080074 last_index = (size - 1) >> PAGE_CACHE_SHIFT;
Mark Fasheh7307de82007-05-09 15:16:19 -070075
76 /*
Wengang Wang5cffff92011-07-24 10:36:54 -070077 * 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.
Mark Fasheh7307de82007-05-09 15:16:19 -070088 */
Wengang Wang5cffff92011-07-24 10:36:54 -070089 if ((page->mapping != inode->i_mapping) ||
90 (!PageUptodate(page)) ||
91 (page_offset(page) >= size))
Mark Fasheh7307de82007-05-09 15:16:19 -070092 goto out;
Mark Fasheh7307de82007-05-09 15:16:19 -070093
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)
Tao Maf63afdb2010-07-17 21:45:49 +0800105 len = ((size - 1) & ~PAGE_CACHE_MASK) + 1;
Mark Fasheh7307de82007-05-09 15:16:19 -0700106
Tao Ma0378da0f2010-08-12 10:25:28 +0800107 ret = ocfs2_write_begin_nolock(file, mapping, pos, len, 0, &locked_page,
Mark Fasheh7307de82007-05-09 15:16:19 -0700108 &fsdata, di_bh, page);
109 if (ret) {
110 if (ret != -ENOSPC)
111 mlog_errno(ret);
Wengang Wang5cffff92011-07-24 10:36:54 -0700112 if (ret == -ENOMEM)
113 ret = VM_FAULT_OOM;
114 else
115 ret = VM_FAULT_SIGBUS;
Mark Fasheh7307de82007-05-09 15:16:19 -0700116 goto out;
117 }
118
Wengang Wang5cffff92011-07-24 10:36:54 -0700119 if (!locked_page) {
120 ret = VM_FAULT_NOPAGE;
Mark Fasheh7307de82007-05-09 15:16:19 -0700121 goto out;
122 }
Wengang Wang5cffff92011-07-24 10:36:54 -0700123 ret = ocfs2_write_end_nolock(mapping, pos, len, len, locked_page,
124 fsdata);
Mark Fasheh7307de82007-05-09 15:16:19 -0700125 BUG_ON(ret != len);
Wengang Wang5cffff92011-07-24 10:36:54 -0700126 ret = VM_FAULT_LOCKED;
Mark Fasheh7307de82007-05-09 15:16:19 -0700127out:
128 return ret;
129}
130
Nick Pigginc2ec1752009-03-31 15:23:21 -0700131static int ocfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Mark Fasheh7307de82007-05-09 15:16:19 -0700132{
Nick Pigginc2ec1752009-03-31 15:23:21 -0700133 struct page *page = vmf->page;
Mark Fasheh7307de82007-05-09 15:16:19 -0700134 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
135 struct buffer_head *di_bh = NULL;
Joel Beckere4b963f2009-09-02 17:17:36 -0700136 sigset_t oldset;
137 int ret;
Mark Fasheh7307de82007-05-09 15:16:19 -0700138
Joel Beckere4b963f2009-09-02 17:17:36 -0700139 ocfs2_block_signals(&oldset);
Mark Fasheh7307de82007-05-09 15:16:19 -0700140
141 /*
142 * The cluster locks taken will block a truncate from another
143 * node. Taking the data lock will also ensure that we don't
144 * attempt page truncation as part of a downconvert.
145 */
Mark Fashehe63aecb62007-10-18 15:30:42 -0700146 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fasheh7307de82007-05-09 15:16:19 -0700147 if (ret < 0) {
148 mlog_errno(ret);
149 goto out;
150 }
151
152 /*
153 * The alloc sem should be enough to serialize with
154 * ocfs2_truncate_file() changing i_size as well as any thread
155 * modifying the inode btree.
156 */
157 down_write(&OCFS2_I(inode)->ip_alloc_sem);
158
Tao Ma0378da0f2010-08-12 10:25:28 +0800159 ret = __ocfs2_page_mkwrite(vma->vm_file, di_bh, page);
Mark Fasheh7307de82007-05-09 15:16:19 -0700160
Mark Fasheh7307de82007-05-09 15:16:19 -0700161 up_write(&OCFS2_I(inode)->ip_alloc_sem);
162
163 brelse(di_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700164 ocfs2_inode_unlock(inode, 1);
Mark Fasheh7307de82007-05-09 15:16:19 -0700165
166out:
Joel Beckere4b963f2009-09-02 17:17:36 -0700167 ocfs2_unblock_signals(&oldset);
Mark Fasheh7307de82007-05-09 15:16:19 -0700168 return ret;
169}
170
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400171static const struct vm_operations_struct ocfs2_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -0700172 .fault = ocfs2_fault,
Mark Fasheh7307de82007-05-09 15:16:19 -0700173 .page_mkwrite = ocfs2_page_mkwrite,
Mark Fashehccd979b2005-12-15 14:31:24 -0800174};
175
Christoph Hellwig870f4812006-01-09 20:52:01 -0800176int ocfs2_mmap(struct file *file, struct vm_area_struct *vma)
Mark Fashehccd979b2005-12-15 14:31:24 -0800177{
Tiger Yang25899de2006-11-15 15:49:02 +0800178 int ret = 0, lock_level = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800179
Mark Fashehe63aecb62007-10-18 15:30:42 -0700180 ret = ocfs2_inode_lock_atime(file->f_dentry->d_inode,
Tiger Yang25899de2006-11-15 15:49:02 +0800181 file->f_vfsmnt, &lock_level);
182 if (ret < 0) {
183 mlog_errno(ret);
184 goto out;
185 }
Mark Fashehe63aecb62007-10-18 15:30:42 -0700186 ocfs2_inode_unlock(file->f_dentry->d_inode, lock_level);
Tiger Yang25899de2006-11-15 15:49:02 +0800187out:
Mark Fashehccd979b2005-12-15 14:31:24 -0800188 vma->vm_ops = &ocfs2_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -0700189 vma->vm_flags |= VM_CAN_NONLINEAR;
Mark Fashehccd979b2005-12-15 14:31:24 -0800190 return 0;
191}
192