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 47 static 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 62 static int __ocfs2_page_mkwrite(struct file *file, struct buffer_head *di_bh, 63 struct page *page) 64 { 65 int ret; 66 struct inode *inode = file->f_path.dentry->d_inode; 67 struct address_space *mapping = inode->i_mapping; 68 loff_t pos = page_offset(page); 69 unsigned int len = PAGE_CACHE_SIZE; 70 pgoff_t last_index; 71 struct page *locked_page = NULL; 72 void *fsdata; 73 loff_t size = i_size_read(inode); 74 75 /* 76 * Another node might have truncated while we were waiting on 77 * cluster locks. 78 * We don't check size == 0 before the shift. This is borrowed 79 * from do_generic_file_read. 80 */ 81 last_index = (size - 1) >> PAGE_CACHE_SHIFT; 82 if (unlikely(!size || page->index > last_index)) { 83 ret = -EINVAL; 84 goto out; 85 } 86 87 /* 88 * The i_size check above doesn't catch the case where nodes 89 * truncated and then re-extended the file. We'll re-check the 90 * page mapping after taking the page lock inside of 91 * ocfs2_write_begin_nolock(). 92 */ 93 if (!PageUptodate(page) || page->mapping != inode->i_mapping) { 94 /* 95 * the page has been umapped in ocfs2_data_downconvert_worker. 96 * So return 0 here and let VFS retry. 97 */ 98 ret = 0; 99 goto out; 100 } 101 102 /* 103 * Call ocfs2_write_begin() and ocfs2_write_end() to take 104 * advantage of the allocation code there. We pass a write 105 * length of the whole page (chopped to i_size) to make sure 106 * the whole thing is allocated. 107 * 108 * Since we know the page is up to date, we don't have to 109 * worry about ocfs2_write_begin() skipping some buffer reads 110 * because the "write" would invalidate their data. 111 */ 112 if (page->index == last_index) 113 len = ((size - 1) & ~PAGE_CACHE_MASK) + 1; 114 115 ret = ocfs2_write_begin_nolock(file, mapping, pos, len, 0, &locked_page, 116 &fsdata, di_bh, page); 117 if (ret) { 118 if (ret != -ENOSPC) 119 mlog_errno(ret); 120 goto out; 121 } 122 123 ret = ocfs2_write_end_nolock(mapping, pos, len, len, locked_page, 124 fsdata); 125 if (ret < 0) { 126 mlog_errno(ret); 127 goto out; 128 } 129 BUG_ON(ret != len); 130 ret = 0; 131 out: 132 return ret; 133 } 134 135 static int ocfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 136 { 137 struct page *page = vmf->page; 138 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 139 struct buffer_head *di_bh = NULL; 140 sigset_t oldset; 141 int ret; 142 143 ocfs2_block_signals(&oldset); 144 145 /* 146 * The cluster locks taken will block a truncate from another 147 * node. Taking the data lock will also ensure that we don't 148 * attempt page truncation as part of a downconvert. 149 */ 150 ret = ocfs2_inode_lock(inode, &di_bh, 1); 151 if (ret < 0) { 152 mlog_errno(ret); 153 goto out; 154 } 155 156 /* 157 * The alloc sem should be enough to serialize with 158 * ocfs2_truncate_file() changing i_size as well as any thread 159 * modifying the inode btree. 160 */ 161 down_write(&OCFS2_I(inode)->ip_alloc_sem); 162 163 ret = __ocfs2_page_mkwrite(vma->vm_file, di_bh, page); 164 165 up_write(&OCFS2_I(inode)->ip_alloc_sem); 166 167 brelse(di_bh); 168 ocfs2_inode_unlock(inode, 1); 169 170 out: 171 ocfs2_unblock_signals(&oldset); 172 if (ret) 173 ret = VM_FAULT_SIGBUS; 174 return ret; 175 } 176 177 static const struct vm_operations_struct ocfs2_file_vm_ops = { 178 .fault = ocfs2_fault, 179 .page_mkwrite = ocfs2_page_mkwrite, 180 }; 181 182 int ocfs2_mmap(struct file *file, struct vm_area_struct *vma) 183 { 184 int ret = 0, lock_level = 0; 185 186 ret = ocfs2_inode_lock_atime(file->f_dentry->d_inode, 187 file->f_vfsmnt, &lock_level); 188 if (ret < 0) { 189 mlog_errno(ret); 190 goto out; 191 } 192 ocfs2_inode_unlock(file->f_dentry->d_inode, lock_level); 193 out: 194 vma->vm_ops = &ocfs2_file_vm_ops; 195 vma->vm_flags |= VM_CAN_NONLINEAR; 196 return 0; 197 } 198 199