]>
Commit | Line | Data |
---|---|---|
328970de | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
fa60ce2c | 2 | /* |
ccd979bd MF |
3 | * mmap.c |
4 | * | |
5 | * Code to deal with the mess that is clustered mmap. | |
6 | * | |
7 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | |
ccd979bd MF |
8 | */ |
9 | ||
10 | #include <linux/fs.h> | |
11 | #include <linux/types.h> | |
ccd979bd MF |
12 | #include <linux/highmem.h> |
13 | #include <linux/pagemap.h> | |
14 | #include <linux/uio.h> | |
15 | #include <linux/signal.h> | |
16 | #include <linux/rbtree.h> | |
17 | ||
ccd979bd MF |
18 | #include <cluster/masklog.h> |
19 | ||
20 | #include "ocfs2.h" | |
21 | ||
7307de80 | 22 | #include "aops.h" |
ccd979bd MF |
23 | #include "dlmglue.h" |
24 | #include "file.h" | |
25 | #include "inode.h" | |
26 | #include "mmap.h" | |
e4b963f1 | 27 | #include "super.h" |
614a9e84 | 28 | #include "ocfs2_trace.h" |
ccd979bd | 29 | |
7307de80 | 30 | |
c6137fe3 | 31 | static vm_fault_t ocfs2_fault(struct vm_fault *vmf) |
ccd979bd | 32 | { |
11bac800 | 33 | struct vm_area_struct *vma = vmf->vma; |
e4b963f1 | 34 | sigset_t oldset; |
c6137fe3 | 35 | vm_fault_t ret; |
ccd979bd | 36 | |
e4b963f1 | 37 | ocfs2_block_signals(&oldset); |
11bac800 | 38 | ret = filemap_fault(vmf); |
e4b963f1 | 39 | ocfs2_unblock_signals(&oldset); |
ccd979bd | 40 | |
11bac800 DJ |
41 | trace_ocfs2_fault(OCFS2_I(vma->vm_file->f_mapping->host)->ip_blkno, |
42 | vma, vmf->page, vmf->pgoff); | |
d0217ac0 | 43 | return ret; |
ccd979bd MF |
44 | } |
45 | ||
c6137fe3 SJ |
46 | static vm_fault_t __ocfs2_page_mkwrite(struct file *file, |
47 | struct buffer_head *di_bh, struct page *page) | |
7307de80 | 48 | { |
c6137fe3 SJ |
49 | int err; |
50 | vm_fault_t ret = VM_FAULT_NOPAGE; | |
496ad9aa | 51 | struct inode *inode = file_inode(file); |
7307de80 | 52 | struct address_space *mapping = inode->i_mapping; |
18336338 | 53 | loff_t pos = page_offset(page); |
09cbfeaf | 54 | unsigned int len = PAGE_SIZE; |
7307de80 MF |
55 | pgoff_t last_index; |
56 | struct page *locked_page = NULL; | |
57 | void *fsdata; | |
58 | loff_t size = i_size_read(inode); | |
ccd979bd | 59 | |
09cbfeaf | 60 | last_index = (size - 1) >> PAGE_SHIFT; |
7307de80 MF |
61 | |
62 | /* | |
cc989e78 | 63 | * There are cases that lead to the page no longer belonging to the |
5cffff9e WW |
64 | * mapping. |
65 | * 1) pagecache truncates locally due to memory pressure. | |
66 | * 2) pagecache truncates when another is taking EX lock against | |
67 | * inode lock. see ocfs2_data_convert_worker. | |
68 | * | |
69 | * The i_size check doesn't catch the case where nodes truncated and | |
70 | * then re-extended the file. We'll re-check the page mapping after | |
71 | * taking the page lock inside of ocfs2_write_begin_nolock(). | |
72 | * | |
73 | * Let VM retry with these cases. | |
7307de80 | 74 | */ |
5cffff9e WW |
75 | if ((page->mapping != inode->i_mapping) || |
76 | (!PageUptodate(page)) || | |
77 | (page_offset(page) >= size)) | |
7307de80 | 78 | goto out; |
7307de80 MF |
79 | |
80 | /* | |
81 | * Call ocfs2_write_begin() and ocfs2_write_end() to take | |
82 | * advantage of the allocation code there. We pass a write | |
83 | * length of the whole page (chopped to i_size) to make sure | |
84 | * the whole thing is allocated. | |
85 | * | |
86 | * Since we know the page is up to date, we don't have to | |
87 | * worry about ocfs2_write_begin() skipping some buffer reads | |
88 | * because the "write" would invalidate their data. | |
89 | */ | |
90 | if (page->index == last_index) | |
09cbfeaf | 91 | len = ((size - 1) & ~PAGE_MASK) + 1; |
7307de80 | 92 | |
c6137fe3 | 93 | err = ocfs2_write_begin_nolock(mapping, pos, len, OCFS2_WRITE_MMAP, |
c1ad1e3c | 94 | &locked_page, &fsdata, di_bh, page); |
c6137fe3 SJ |
95 | if (err) { |
96 | if (err != -ENOSPC) | |
97 | mlog_errno(err); | |
98 | ret = vmf_error(err); | |
7307de80 MF |
99 | goto out; |
100 | } | |
101 | ||
5cffff9e WW |
102 | if (!locked_page) { |
103 | ret = VM_FAULT_NOPAGE; | |
7307de80 MF |
104 | goto out; |
105 | } | |
c6137fe3 SJ |
106 | err = ocfs2_write_end_nolock(mapping, pos, len, len, fsdata); |
107 | BUG_ON(err != len); | |
5cffff9e | 108 | ret = VM_FAULT_LOCKED; |
7307de80 MF |
109 | out: |
110 | return ret; | |
111 | } | |
112 | ||
c6137fe3 | 113 | static vm_fault_t ocfs2_page_mkwrite(struct vm_fault *vmf) |
ccd979bd | 114 | { |
c2ec175c | 115 | struct page *page = vmf->page; |
11bac800 | 116 | struct inode *inode = file_inode(vmf->vma->vm_file); |
7307de80 | 117 | struct buffer_head *di_bh = NULL; |
e4b963f1 | 118 | sigset_t oldset; |
c6137fe3 SJ |
119 | int err; |
120 | vm_fault_t ret; | |
7307de80 | 121 | |
fef6925c | 122 | sb_start_pagefault(inode->i_sb); |
e4b963f1 | 123 | ocfs2_block_signals(&oldset); |
7307de80 MF |
124 | |
125 | /* | |
126 | * The cluster locks taken will block a truncate from another | |
127 | * node. Taking the data lock will also ensure that we don't | |
128 | * attempt page truncation as part of a downconvert. | |
129 | */ | |
c6137fe3 SJ |
130 | err = ocfs2_inode_lock(inode, &di_bh, 1); |
131 | if (err < 0) { | |
132 | mlog_errno(err); | |
133 | ret = vmf_error(err); | |
7307de80 MF |
134 | goto out; |
135 | } | |
25899dee | 136 | |
89488984 | 137 | /* |
7307de80 MF |
138 | * The alloc sem should be enough to serialize with |
139 | * ocfs2_truncate_file() changing i_size as well as any thread | |
140 | * modifying the inode btree. | |
89488984 | 141 | */ |
7307de80 MF |
142 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
143 | ||
11bac800 | 144 | ret = __ocfs2_page_mkwrite(vmf->vma->vm_file, di_bh, page); |
7307de80 | 145 | |
7307de80 MF |
146 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
147 | ||
148 | brelse(di_bh); | |
e63aecb6 | 149 | ocfs2_inode_unlock(inode, 1); |
7307de80 MF |
150 | |
151 | out: | |
e4b963f1 | 152 | ocfs2_unblock_signals(&oldset); |
fef6925c | 153 | sb_end_pagefault(inode->i_sb); |
7307de80 MF |
154 | return ret; |
155 | } | |
156 | ||
f0f37e2f | 157 | static const struct vm_operations_struct ocfs2_file_vm_ops = { |
54cb8821 | 158 | .fault = ocfs2_fault, |
7307de80 MF |
159 | .page_mkwrite = ocfs2_page_mkwrite, |
160 | }; | |
161 | ||
162 | int ocfs2_mmap(struct file *file, struct vm_area_struct *vma) | |
163 | { | |
164 | int ret = 0, lock_level = 0; | |
165 | ||
496ad9aa | 166 | ret = ocfs2_inode_lock_atime(file_inode(file), |
c4c2416a | 167 | file->f_path.mnt, &lock_level, 1); |
25899dee TY |
168 | if (ret < 0) { |
169 | mlog_errno(ret); | |
170 | goto out; | |
171 | } | |
496ad9aa | 172 | ocfs2_inode_unlock(file_inode(file), lock_level); |
25899dee | 173 | out: |
ccd979bd MF |
174 | vma->vm_ops = &ocfs2_file_vm_ops; |
175 | return 0; | |
176 | } | |
177 |