Commit | Line | Data |
---|---|---|
5db53f3e JE |
1 | /* |
2 | * fs/logfs/file.c - prepare_write, commit_write and friends | |
3 | * | |
4 | * As should be obvious for Linux kernel code, license is GPLv2 | |
5 | * | |
6 | * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org> | |
7 | */ | |
8 | #include "logfs.h" | |
9 | #include <linux/sched.h> | |
10 | #include <linux/writeback.h> | |
11 | ||
12 | static int logfs_write_begin(struct file *file, struct address_space *mapping, | |
13 | loff_t pos, unsigned len, unsigned flags, | |
14 | struct page **pagep, void **fsdata) | |
15 | { | |
16 | struct inode *inode = mapping->host; | |
17 | struct page *page; | |
18 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
19 | ||
20 | page = grab_cache_page_write_begin(mapping, index, flags); | |
21 | if (!page) | |
22 | return -ENOMEM; | |
23 | *pagep = page; | |
24 | ||
25 | if ((len == PAGE_CACHE_SIZE) || PageUptodate(page)) | |
26 | return 0; | |
27 | if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) { | |
28 | unsigned start = pos & (PAGE_CACHE_SIZE - 1); | |
29 | unsigned end = start + len; | |
30 | ||
31 | /* Reading beyond i_size is simple: memset to zero */ | |
32 | zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE); | |
33 | return 0; | |
34 | } | |
35 | return logfs_readpage_nolock(page); | |
36 | } | |
37 | ||
38 | static int logfs_write_end(struct file *file, struct address_space *mapping, | |
39 | loff_t pos, unsigned len, unsigned copied, struct page *page, | |
40 | void *fsdata) | |
41 | { | |
42 | struct inode *inode = mapping->host; | |
43 | pgoff_t index = page->index; | |
44 | unsigned start = pos & (PAGE_CACHE_SIZE - 1); | |
45 | unsigned end = start + copied; | |
46 | int ret = 0; | |
47 | ||
48 | BUG_ON(PAGE_CACHE_SIZE != inode->i_sb->s_blocksize); | |
49 | BUG_ON(page->index > I3_BLOCKS); | |
50 | ||
51 | if (copied < len) { | |
52 | /* | |
53 | * Short write of a non-initialized paged. Just tell userspace | |
54 | * to retry the entire page. | |
55 | */ | |
56 | if (!PageUptodate(page)) { | |
57 | copied = 0; | |
58 | goto out; | |
59 | } | |
60 | } | |
61 | if (copied == 0) | |
62 | goto out; /* FIXME: do we need to update inode? */ | |
63 | ||
64 | if (i_size_read(inode) < (index << PAGE_CACHE_SHIFT) + end) { | |
65 | i_size_write(inode, (index << PAGE_CACHE_SHIFT) + end); | |
66 | mark_inode_dirty_sync(inode); | |
67 | } | |
68 | ||
69 | SetPageUptodate(page); | |
70 | if (!PageDirty(page)) { | |
71 | if (!get_page_reserve(inode, page)) | |
72 | __set_page_dirty_nobuffers(page); | |
73 | else | |
74 | ret = logfs_write_buf(inode, page, WF_LOCK); | |
75 | } | |
76 | out: | |
77 | unlock_page(page); | |
78 | page_cache_release(page); | |
79 | return ret ? ret : copied; | |
80 | } | |
81 | ||
82 | int logfs_readpage(struct file *file, struct page *page) | |
83 | { | |
84 | int ret; | |
85 | ||
86 | ret = logfs_readpage_nolock(page); | |
87 | unlock_page(page); | |
88 | return ret; | |
89 | } | |
90 | ||
91 | /* Clear the page's dirty flag in the radix tree. */ | |
92 | /* TODO: mucking with PageWriteback is silly. Add a generic function to clear | |
93 | * the dirty bit from the radix tree for filesystems that don't have to wait | |
94 | * for page writeback to finish (i.e. any compressing filesystem). | |
95 | */ | |
96 | static void clear_radix_tree_dirty(struct page *page) | |
97 | { | |
98 | BUG_ON(PagePrivate(page) || page->private); | |
99 | set_page_writeback(page); | |
100 | end_page_writeback(page); | |
101 | } | |
102 | ||
103 | static int __logfs_writepage(struct page *page) | |
104 | { | |
105 | struct inode *inode = page->mapping->host; | |
106 | int err; | |
107 | ||
108 | err = logfs_write_buf(inode, page, WF_LOCK); | |
109 | if (err) | |
110 | set_page_dirty(page); | |
111 | else | |
112 | clear_radix_tree_dirty(page); | |
113 | unlock_page(page); | |
114 | return err; | |
115 | } | |
116 | ||
117 | static int logfs_writepage(struct page *page, struct writeback_control *wbc) | |
118 | { | |
119 | struct inode *inode = page->mapping->host; | |
120 | loff_t i_size = i_size_read(inode); | |
121 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
122 | unsigned offset; | |
123 | u64 bix; | |
124 | level_t level; | |
125 | ||
126 | log_file("logfs_writepage(%lx, %lx, %p)\n", inode->i_ino, page->index, | |
127 | page); | |
128 | ||
129 | logfs_unpack_index(page->index, &bix, &level); | |
130 | ||
131 | /* Indirect blocks are never truncated */ | |
132 | if (level != 0) | |
133 | return __logfs_writepage(page); | |
134 | ||
135 | /* | |
136 | * TODO: everything below is a near-verbatim copy of nobh_writepage(). | |
137 | * The relevant bits should be factored out after logfs is merged. | |
138 | */ | |
139 | ||
140 | /* Is the page fully inside i_size? */ | |
141 | if (bix < end_index) | |
142 | return __logfs_writepage(page); | |
143 | ||
144 | /* Is the page fully outside i_size? (truncate in progress) */ | |
145 | offset = i_size & (PAGE_CACHE_SIZE-1); | |
146 | if (bix > end_index || offset == 0) { | |
147 | unlock_page(page); | |
148 | return 0; /* don't care */ | |
149 | } | |
150 | ||
151 | /* | |
152 | * The page straddles i_size. It must be zeroed out on each and every | |
153 | * writepage invokation because it may be mmapped. "A file is mapped | |
154 | * in multiples of the page size. For a file that is not a multiple of | |
155 | * the page size, the remaining memory is zeroed when mapped, and | |
156 | * writes to that region are not written out to the file." | |
157 | */ | |
158 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); | |
159 | return __logfs_writepage(page); | |
160 | } | |
161 | ||
162 | static void logfs_invalidatepage(struct page *page, unsigned long offset) | |
163 | { | |
164 | move_page_to_btree(page); | |
165 | BUG_ON(PagePrivate(page) || page->private); | |
166 | } | |
167 | ||
168 | static int logfs_releasepage(struct page *page, gfp_t only_xfs_uses_this) | |
169 | { | |
170 | return 0; /* None of these are easy to release */ | |
171 | } | |
172 | ||
173 | ||
174 | int logfs_ioctl(struct inode *inode, struct file *file, unsigned int cmd, | |
175 | unsigned long arg) | |
176 | { | |
177 | struct logfs_inode *li = logfs_inode(inode); | |
178 | unsigned int oldflags, flags; | |
179 | int err; | |
180 | ||
181 | switch (cmd) { | |
182 | case FS_IOC_GETFLAGS: | |
183 | flags = li->li_flags & LOGFS_FL_USER_VISIBLE; | |
184 | return put_user(flags, (int __user *)arg); | |
185 | case FS_IOC_SETFLAGS: | |
186 | if (IS_RDONLY(inode)) | |
187 | return -EROFS; | |
188 | ||
189 | if (!is_owner_or_cap(inode)) | |
190 | return -EACCES; | |
191 | ||
192 | err = get_user(flags, (int __user *)arg); | |
193 | if (err) | |
194 | return err; | |
195 | ||
196 | mutex_lock(&inode->i_mutex); | |
197 | oldflags = li->li_flags; | |
198 | flags &= LOGFS_FL_USER_MODIFIABLE; | |
199 | flags |= oldflags & ~LOGFS_FL_USER_MODIFIABLE; | |
200 | li->li_flags = flags; | |
201 | mutex_unlock(&inode->i_mutex); | |
202 | ||
203 | inode->i_ctime = CURRENT_TIME; | |
204 | mark_inode_dirty_sync(inode); | |
205 | return 0; | |
206 | ||
207 | default: | |
208 | return -ENOTTY; | |
209 | } | |
210 | } | |
211 | ||
212 | int logfs_fsync(struct file *file, struct dentry *dentry, int datasync) | |
213 | { | |
214 | struct super_block *sb = dentry->d_inode->i_sb; | |
215 | struct logfs_super *super = logfs_super(sb); | |
216 | ||
217 | /* FIXME: write anchor */ | |
218 | super->s_devops->sync(sb); | |
219 | return 0; | |
220 | } | |
221 | ||
222 | static int logfs_setattr(struct dentry *dentry, struct iattr *attr) | |
223 | { | |
224 | struct inode *inode = dentry->d_inode; | |
225 | int err = 0; | |
226 | ||
227 | if (attr->ia_valid & ATTR_SIZE) | |
228 | err = logfs_truncate(inode, attr->ia_size); | |
229 | attr->ia_valid &= ~ATTR_SIZE; | |
230 | ||
231 | if (!err) | |
232 | err = inode_change_ok(inode, attr); | |
233 | if (!err) | |
234 | err = inode_setattr(inode, attr); | |
235 | return err; | |
236 | } | |
237 | ||
238 | const struct inode_operations logfs_reg_iops = { | |
239 | .setattr = logfs_setattr, | |
240 | }; | |
241 | ||
242 | const struct file_operations logfs_reg_fops = { | |
243 | .aio_read = generic_file_aio_read, | |
244 | .aio_write = generic_file_aio_write, | |
245 | .fsync = logfs_fsync, | |
246 | .ioctl = logfs_ioctl, | |
247 | .llseek = generic_file_llseek, | |
248 | .mmap = generic_file_readonly_mmap, | |
249 | .open = generic_file_open, | |
250 | .read = do_sync_read, | |
251 | .write = do_sync_write, | |
252 | }; | |
253 | ||
254 | const struct address_space_operations logfs_reg_aops = { | |
255 | .invalidatepage = logfs_invalidatepage, | |
256 | .readpage = logfs_readpage, | |
257 | .releasepage = logfs_releasepage, | |
258 | .set_page_dirty = __set_page_dirty_nobuffers, | |
259 | .writepage = logfs_writepage, | |
260 | .writepages = generic_writepages, | |
261 | .write_begin = logfs_write_begin, | |
262 | .write_end = logfs_write_end, | |
263 | }; |