]>
Commit | Line | Data |
---|---|---|
1 | /* file-nommu.c: no-MMU version of ramfs | |
2 | * | |
3 | * Copyright (C) 2005 Red Hat, Inc. All Rights Reserved. | |
4 | * Written by David Howells (dhowells@redhat.com) | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License | |
8 | * as published by the Free Software Foundation; either version | |
9 | * 2 of the License, or (at your option) any later version. | |
10 | */ | |
11 | ||
12 | #include <linux/module.h> | |
13 | #include <linux/fs.h> | |
14 | #include <linux/mm.h> | |
15 | #include <linux/pagemap.h> | |
16 | #include <linux/highmem.h> | |
17 | #include <linux/init.h> | |
18 | #include <linux/string.h> | |
19 | #include <linux/backing-dev.h> | |
20 | #include <linux/ramfs.h> | |
21 | #include <linux/quotaops.h> | |
22 | #include <linux/pagevec.h> | |
23 | #include <linux/mman.h> | |
24 | ||
25 | #include <asm/uaccess.h> | |
26 | #include "internal.h" | |
27 | ||
28 | static int ramfs_nommu_setattr(struct dentry *, struct iattr *); | |
29 | ||
30 | const struct address_space_operations ramfs_aops = { | |
31 | .readpage = simple_readpage, | |
32 | .prepare_write = simple_prepare_write, | |
33 | .commit_write = simple_commit_write, | |
34 | .set_page_dirty = __set_page_dirty_no_writeback, | |
35 | }; | |
36 | ||
37 | const struct file_operations ramfs_file_operations = { | |
38 | .mmap = ramfs_nommu_mmap, | |
39 | .get_unmapped_area = ramfs_nommu_get_unmapped_area, | |
40 | .read = do_sync_read, | |
41 | .aio_read = generic_file_aio_read, | |
42 | .write = do_sync_write, | |
43 | .aio_write = generic_file_aio_write, | |
44 | .fsync = simple_sync_file, | |
45 | .splice_read = generic_file_splice_read, | |
46 | .llseek = generic_file_llseek, | |
47 | }; | |
48 | ||
49 | const struct inode_operations ramfs_file_inode_operations = { | |
50 | .setattr = ramfs_nommu_setattr, | |
51 | .getattr = simple_getattr, | |
52 | }; | |
53 | ||
54 | /*****************************************************************************/ | |
55 | /* | |
56 | * add a contiguous set of pages into a ramfs inode when it's truncated from | |
57 | * size 0 on the assumption that it's going to be used for an mmap of shared | |
58 | * memory | |
59 | */ | |
60 | static int ramfs_nommu_expand_for_mapping(struct inode *inode, size_t newsize) | |
61 | { | |
62 | struct pagevec lru_pvec; | |
63 | unsigned long npages, xpages, loop, limit; | |
64 | struct page *pages; | |
65 | unsigned order; | |
66 | void *data; | |
67 | int ret; | |
68 | ||
69 | /* make various checks */ | |
70 | order = get_order(newsize); | |
71 | if (unlikely(order >= MAX_ORDER)) | |
72 | goto too_big; | |
73 | ||
74 | limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; | |
75 | if (limit != RLIM_INFINITY && newsize > limit) | |
76 | goto fsize_exceeded; | |
77 | ||
78 | if (newsize > inode->i_sb->s_maxbytes) | |
79 | goto too_big; | |
80 | ||
81 | i_size_write(inode, newsize); | |
82 | ||
83 | /* allocate enough contiguous pages to be able to satisfy the | |
84 | * request */ | |
85 | pages = alloc_pages(mapping_gfp_mask(inode->i_mapping), order); | |
86 | if (!pages) | |
87 | return -ENOMEM; | |
88 | ||
89 | /* split the high-order page into an array of single pages */ | |
90 | xpages = 1UL << order; | |
91 | npages = (newsize + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
92 | ||
93 | split_page(pages, order); | |
94 | ||
95 | /* trim off any pages we don't actually require */ | |
96 | for (loop = npages; loop < xpages; loop++) | |
97 | __free_page(pages + loop); | |
98 | ||
99 | /* clear the memory we allocated */ | |
100 | newsize = PAGE_SIZE * npages; | |
101 | data = page_address(pages); | |
102 | memset(data, 0, newsize); | |
103 | ||
104 | /* attach all the pages to the inode's address space */ | |
105 | pagevec_init(&lru_pvec, 0); | |
106 | for (loop = 0; loop < npages; loop++) { | |
107 | struct page *page = pages + loop; | |
108 | ||
109 | ret = add_to_page_cache(page, inode->i_mapping, loop, GFP_KERNEL); | |
110 | if (ret < 0) | |
111 | goto add_error; | |
112 | ||
113 | if (!pagevec_add(&lru_pvec, page)) | |
114 | __pagevec_lru_add(&lru_pvec); | |
115 | ||
116 | unlock_page(page); | |
117 | } | |
118 | ||
119 | pagevec_lru_add(&lru_pvec); | |
120 | return 0; | |
121 | ||
122 | fsize_exceeded: | |
123 | send_sig(SIGXFSZ, current, 0); | |
124 | too_big: | |
125 | return -EFBIG; | |
126 | ||
127 | add_error: | |
128 | page_cache_release(pages + loop); | |
129 | for (loop++; loop < npages; loop++) | |
130 | __free_page(pages + loop); | |
131 | return ret; | |
132 | } | |
133 | ||
134 | /*****************************************************************************/ | |
135 | /* | |
136 | * check that file shrinkage doesn't leave any VMAs dangling in midair | |
137 | */ | |
138 | static int ramfs_nommu_check_mappings(struct inode *inode, | |
139 | size_t newsize, size_t size) | |
140 | { | |
141 | struct vm_area_struct *vma; | |
142 | struct prio_tree_iter iter; | |
143 | ||
144 | /* search for VMAs that fall within the dead zone */ | |
145 | vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap, | |
146 | newsize >> PAGE_SHIFT, | |
147 | (size + PAGE_SIZE - 1) >> PAGE_SHIFT | |
148 | ) { | |
149 | /* found one - only interested if it's shared out of the page | |
150 | * cache */ | |
151 | if (vma->vm_flags & VM_SHARED) | |
152 | return -ETXTBSY; /* not quite true, but near enough */ | |
153 | } | |
154 | ||
155 | return 0; | |
156 | } | |
157 | ||
158 | /*****************************************************************************/ | |
159 | /* | |
160 | * | |
161 | */ | |
162 | static int ramfs_nommu_resize(struct inode *inode, loff_t newsize, loff_t size) | |
163 | { | |
164 | int ret; | |
165 | ||
166 | /* assume a truncate from zero size is going to be for the purposes of | |
167 | * shared mmap */ | |
168 | if (size == 0) { | |
169 | if (unlikely(newsize >> 32)) | |
170 | return -EFBIG; | |
171 | ||
172 | return ramfs_nommu_expand_for_mapping(inode, newsize); | |
173 | } | |
174 | ||
175 | /* check that a decrease in size doesn't cut off any shared mappings */ | |
176 | if (newsize < size) { | |
177 | ret = ramfs_nommu_check_mappings(inode, newsize, size); | |
178 | if (ret < 0) | |
179 | return ret; | |
180 | } | |
181 | ||
182 | ret = vmtruncate(inode, newsize); | |
183 | ||
184 | return ret; | |
185 | } | |
186 | ||
187 | /*****************************************************************************/ | |
188 | /* | |
189 | * handle a change of attributes | |
190 | * - we're specifically interested in a change of size | |
191 | */ | |
192 | static int ramfs_nommu_setattr(struct dentry *dentry, struct iattr *ia) | |
193 | { | |
194 | struct inode *inode = dentry->d_inode; | |
195 | unsigned int old_ia_valid = ia->ia_valid; | |
196 | int ret = 0; | |
197 | ||
198 | /* POSIX UID/GID verification for setting inode attributes */ | |
199 | ret = inode_change_ok(inode, ia); | |
200 | if (ret) | |
201 | return ret; | |
202 | ||
203 | /* by providing our own setattr() method, we skip this quotaism */ | |
204 | if ((old_ia_valid & ATTR_UID && ia->ia_uid != inode->i_uid) || | |
205 | (old_ia_valid & ATTR_GID && ia->ia_gid != inode->i_gid)) | |
206 | ret = DQUOT_TRANSFER(inode, ia) ? -EDQUOT : 0; | |
207 | ||
208 | /* pick out size-changing events */ | |
209 | if (ia->ia_valid & ATTR_SIZE) { | |
210 | loff_t size = i_size_read(inode); | |
211 | if (ia->ia_size != size) { | |
212 | ret = ramfs_nommu_resize(inode, ia->ia_size, size); | |
213 | if (ret < 0 || ia->ia_valid == ATTR_SIZE) | |
214 | goto out; | |
215 | } else { | |
216 | /* we skipped the truncate but must still update | |
217 | * timestamps | |
218 | */ | |
219 | ia->ia_valid |= ATTR_MTIME|ATTR_CTIME; | |
220 | } | |
221 | } | |
222 | ||
223 | ret = inode_setattr(inode, ia); | |
224 | out: | |
225 | ia->ia_valid = old_ia_valid; | |
226 | return ret; | |
227 | } | |
228 | ||
229 | /*****************************************************************************/ | |
230 | /* | |
231 | * try to determine where a shared mapping can be made | |
232 | * - we require that: | |
233 | * - the pages to be mapped must exist | |
234 | * - the pages be physically contiguous in sequence | |
235 | */ | |
236 | unsigned long ramfs_nommu_get_unmapped_area(struct file *file, | |
237 | unsigned long addr, unsigned long len, | |
238 | unsigned long pgoff, unsigned long flags) | |
239 | { | |
240 | unsigned long maxpages, lpages, nr, loop, ret; | |
241 | struct inode *inode = file->f_path.dentry->d_inode; | |
242 | struct page **pages = NULL, **ptr, *page; | |
243 | loff_t isize; | |
244 | ||
245 | if (!(flags & MAP_SHARED)) | |
246 | return addr; | |
247 | ||
248 | /* the mapping mustn't extend beyond the EOF */ | |
249 | lpages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
250 | isize = i_size_read(inode); | |
251 | ||
252 | ret = -EINVAL; | |
253 | maxpages = (isize + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
254 | if (pgoff >= maxpages) | |
255 | goto out; | |
256 | ||
257 | if (maxpages - pgoff < lpages) | |
258 | goto out; | |
259 | ||
260 | /* gang-find the pages */ | |
261 | ret = -ENOMEM; | |
262 | pages = kzalloc(lpages * sizeof(struct page *), GFP_KERNEL); | |
263 | if (!pages) | |
264 | goto out; | |
265 | ||
266 | nr = find_get_pages(inode->i_mapping, pgoff, lpages, pages); | |
267 | if (nr != lpages) | |
268 | goto out; /* leave if some pages were missing */ | |
269 | ||
270 | /* check the pages for physical adjacency */ | |
271 | ptr = pages; | |
272 | page = *ptr++; | |
273 | page++; | |
274 | for (loop = lpages; loop > 1; loop--) | |
275 | if (*ptr++ != page++) | |
276 | goto out; | |
277 | ||
278 | /* okay - all conditions fulfilled */ | |
279 | ret = (unsigned long) page_address(pages[0]); | |
280 | ||
281 | out: | |
282 | if (pages) { | |
283 | ptr = pages; | |
284 | for (loop = lpages; loop > 0; loop--) | |
285 | put_page(*ptr++); | |
286 | kfree(pages); | |
287 | } | |
288 | ||
289 | return ret; | |
290 | } | |
291 | ||
292 | /*****************************************************************************/ | |
293 | /* | |
294 | * set up a mapping for shared memory segments | |
295 | */ | |
296 | int ramfs_nommu_mmap(struct file *file, struct vm_area_struct *vma) | |
297 | { | |
298 | if (!(vma->vm_flags & VM_SHARED)) | |
299 | return -ENOSYS; | |
300 | ||
301 | file_accessed(file); | |
302 | vma->vm_ops = &generic_file_vm_ops; | |
303 | return 0; | |
304 | } |