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 47 static int ocfs2_fault(struct vm_fault *vmf) 48 { 49 struct vm_area_struct *vma = vmf->vma; 50 sigset_t oldset; 51 int ret; 52 53 ocfs2_block_signals(&oldset); 54 ret = filemap_fault(vmf); 55 ocfs2_unblock_signals(&oldset); 56 57 trace_ocfs2_fault(OCFS2_I(vma->vm_file->f_mapping->host)->ip_blkno, 58 vma, vmf->page, vmf->pgoff); 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 = VM_FAULT_NOPAGE; 66 struct inode *inode = file_inode(file); 67 struct address_space *mapping = inode->i_mapping; 68 loff_t pos = page_offset(page); 69 unsigned int len = PAGE_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 last_index = (size - 1) >> PAGE_SHIFT; 76 77 /* 78 * There are cases that lead to the page no longer bebongs to the 79 * mapping. 80 * 1) pagecache truncates locally due to memory pressure. 81 * 2) pagecache truncates when another is taking EX lock against 82 * inode lock. see ocfs2_data_convert_worker. 83 * 84 * The i_size check doesn't catch the case where nodes truncated and 85 * then re-extended the file. We'll re-check the page mapping after 86 * taking the page lock inside of ocfs2_write_begin_nolock(). 87 * 88 * Let VM retry with these cases. 89 */ 90 if ((page->mapping != inode->i_mapping) || 91 (!PageUptodate(page)) || 92 (page_offset(page) >= size)) 93 goto out; 94 95 /* 96 * Call ocfs2_write_begin() and ocfs2_write_end() to take 97 * advantage of the allocation code there. We pass a write 98 * length of the whole page (chopped to i_size) to make sure 99 * the whole thing is allocated. 100 * 101 * Since we know the page is up to date, we don't have to 102 * worry about ocfs2_write_begin() skipping some buffer reads 103 * because the "write" would invalidate their data. 104 */ 105 if (page->index == last_index) 106 len = ((size - 1) & ~PAGE_MASK) + 1; 107 108 ret = ocfs2_write_begin_nolock(mapping, pos, len, OCFS2_WRITE_MMAP, 109 &locked_page, &fsdata, di_bh, page); 110 if (ret) { 111 if (ret != -ENOSPC) 112 mlog_errno(ret); 113 if (ret == -ENOMEM) 114 ret = VM_FAULT_OOM; 115 else 116 ret = VM_FAULT_SIGBUS; 117 goto out; 118 } 119 120 if (!locked_page) { 121 ret = VM_FAULT_NOPAGE; 122 goto out; 123 } 124 ret = ocfs2_write_end_nolock(mapping, pos, len, len, fsdata); 125 BUG_ON(ret != len); 126 ret = VM_FAULT_LOCKED; 127 out: 128 return ret; 129 } 130 131 static int ocfs2_page_mkwrite(struct vm_fault *vmf) 132 { 133 struct page *page = vmf->page; 134 struct inode *inode = file_inode(vmf->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 if (ret == -ENOMEM) 151 ret = VM_FAULT_OOM; 152 else 153 ret = VM_FAULT_SIGBUS; 154 goto out; 155 } 156 157 /* 158 * The alloc sem should be enough to serialize with 159 * ocfs2_truncate_file() changing i_size as well as any thread 160 * modifying the inode btree. 161 */ 162 down_write(&OCFS2_I(inode)->ip_alloc_sem); 163 164 ret = __ocfs2_page_mkwrite(vmf->vma->vm_file, di_bh, page); 165 166 up_write(&OCFS2_I(inode)->ip_alloc_sem); 167 168 brelse(di_bh); 169 ocfs2_inode_unlock(inode, 1); 170 171 out: 172 ocfs2_unblock_signals(&oldset); 173 sb_end_pagefault(inode->i_sb); 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_inode(file), 187 file->f_path.mnt, &lock_level, 1); 188 if (ret < 0) { 189 mlog_errno(ret); 190 goto out; 191 } 192 ocfs2_inode_unlock(file_inode(file), lock_level); 193 out: 194 vma->vm_ops = &ocfs2_file_vm_ops; 195 return 0; 196 } 197 198