Commit | Line | Data |
---|---|---|
642fb4d1 DH |
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/pagemap.h> | |
15 | #include <linux/highmem.h> | |
16 | #include <linux/init.h> | |
17 | #include <linux/string.h> | |
18 | #include <linux/smp_lock.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 | ||
f5e54d6e | 30 | const struct address_space_operations ramfs_aops = { |
642fb4d1 DH |
31 | .readpage = simple_readpage, |
32 | .prepare_write = simple_prepare_write, | |
33 | .commit_write = simple_commit_write | |
34 | }; | |
35 | ||
4b6f5d20 | 36 | const struct file_operations ramfs_file_operations = { |
642fb4d1 DH |
37 | .mmap = ramfs_nommu_mmap, |
38 | .get_unmapped_area = ramfs_nommu_get_unmapped_area, | |
543ade1f BP |
39 | .read = do_sync_read, |
40 | .aio_read = generic_file_aio_read, | |
41 | .write = do_sync_write, | |
42 | .aio_write = generic_file_aio_write, | |
642fb4d1 DH |
43 | .fsync = simple_sync_file, |
44 | .sendfile = generic_file_sendfile, | |
45 | .llseek = generic_file_llseek, | |
46 | }; | |
47 | ||
48 | struct inode_operations ramfs_file_inode_operations = { | |
49 | .setattr = ramfs_nommu_setattr, | |
50 | .getattr = simple_getattr, | |
51 | }; | |
52 | ||
53 | /*****************************************************************************/ | |
54 | /* | |
55 | * add a contiguous set of pages into a ramfs inode when it's truncated from | |
56 | * size 0 on the assumption that it's going to be used for an mmap of shared | |
57 | * memory | |
58 | */ | |
59 | static int ramfs_nommu_expand_for_mapping(struct inode *inode, size_t newsize) | |
60 | { | |
61 | struct pagevec lru_pvec; | |
62 | unsigned long npages, xpages, loop, limit; | |
63 | struct page *pages; | |
64 | unsigned order; | |
65 | void *data; | |
66 | int ret; | |
67 | ||
68 | /* make various checks */ | |
69 | order = get_order(newsize); | |
70 | if (unlikely(order >= MAX_ORDER)) | |
71 | goto too_big; | |
72 | ||
73 | limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; | |
74 | if (limit != RLIM_INFINITY && newsize > limit) | |
75 | goto fsize_exceeded; | |
76 | ||
77 | if (newsize > inode->i_sb->s_maxbytes) | |
78 | goto too_big; | |
79 | ||
80 | i_size_write(inode, newsize); | |
81 | ||
82 | /* allocate enough contiguous pages to be able to satisfy the | |
83 | * request */ | |
84 | pages = alloc_pages(mapping_gfp_mask(inode->i_mapping), order); | |
85 | if (!pages) | |
86 | return -ENOMEM; | |
87 | ||
88 | /* split the high-order page into an array of single pages */ | |
89 | xpages = 1UL << order; | |
90 | npages = (newsize + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
91 | ||
84097518 | 92 | split_page(pages, order); |
642fb4d1 DH |
93 | |
94 | /* trim off any pages we don't actually require */ | |
95 | for (loop = npages; loop < xpages; loop++) | |
96 | __free_page(pages + loop); | |
97 | ||
98 | /* clear the memory we allocated */ | |
99 | newsize = PAGE_SIZE * npages; | |
100 | data = page_address(pages); | |
101 | memset(data, 0, newsize); | |
102 | ||
103 | /* attach all the pages to the inode's address space */ | |
104 | pagevec_init(&lru_pvec, 0); | |
105 | for (loop = 0; loop < npages; loop++) { | |
106 | struct page *page = pages + loop; | |
107 | ||
108 | ret = add_to_page_cache(page, inode->i_mapping, loop, GFP_KERNEL); | |
109 | if (ret < 0) | |
110 | goto add_error; | |
111 | ||
112 | if (!pagevec_add(&lru_pvec, page)) | |
113 | __pagevec_lru_add(&lru_pvec); | |
114 | ||
115 | unlock_page(page); | |
116 | } | |
117 | ||
118 | pagevec_lru_add(&lru_pvec); | |
119 | return 0; | |
120 | ||
121 | fsize_exceeded: | |
122 | send_sig(SIGXFSZ, current, 0); | |
123 | too_big: | |
124 | return -EFBIG; | |
125 | ||
126 | add_error: | |
127 | page_cache_release(pages + loop); | |
128 | for (loop++; loop < npages; loop++) | |
129 | __free_page(pages + loop); | |
130 | return ret; | |
131 | } | |
132 | ||
133 | /*****************************************************************************/ | |
134 | /* | |
135 | * check that file shrinkage doesn't leave any VMAs dangling in midair | |
136 | */ | |
137 | static int ramfs_nommu_check_mappings(struct inode *inode, | |
138 | size_t newsize, size_t size) | |
139 | { | |
140 | struct vm_area_struct *vma; | |
141 | struct prio_tree_iter iter; | |
142 | ||
143 | /* search for VMAs that fall within the dead zone */ | |
144 | vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap, | |
145 | newsize >> PAGE_SHIFT, | |
146 | (size + PAGE_SIZE - 1) >> PAGE_SHIFT | |
147 | ) { | |
148 | /* found one - only interested if it's shared out of the page | |
149 | * cache */ | |
150 | if (vma->vm_flags & VM_SHARED) | |
151 | return -ETXTBSY; /* not quite true, but near enough */ | |
152 | } | |
153 | ||
154 | return 0; | |
155 | } | |
156 | ||
157 | /*****************************************************************************/ | |
158 | /* | |
159 | * | |
160 | */ | |
161 | static int ramfs_nommu_resize(struct inode *inode, loff_t newsize, loff_t size) | |
162 | { | |
163 | int ret; | |
164 | ||
165 | /* assume a truncate from zero size is going to be for the purposes of | |
166 | * shared mmap */ | |
167 | if (size == 0) { | |
168 | if (unlikely(newsize >> 32)) | |
169 | return -EFBIG; | |
170 | ||
171 | return ramfs_nommu_expand_for_mapping(inode, newsize); | |
172 | } | |
173 | ||
174 | /* check that a decrease in size doesn't cut off any shared mappings */ | |
175 | if (newsize < size) { | |
176 | ret = ramfs_nommu_check_mappings(inode, newsize, size); | |
177 | if (ret < 0) | |
178 | return ret; | |
179 | } | |
180 | ||
181 | ret = vmtruncate(inode, size); | |
182 | ||
183 | return ret; | |
184 | } | |
185 | ||
186 | /*****************************************************************************/ | |
187 | /* | |
188 | * handle a change of attributes | |
189 | * - we're specifically interested in a change of size | |
190 | */ | |
191 | static int ramfs_nommu_setattr(struct dentry *dentry, struct iattr *ia) | |
192 | { | |
193 | struct inode *inode = dentry->d_inode; | |
194 | unsigned int old_ia_valid = ia->ia_valid; | |
195 | int ret = 0; | |
196 | ||
197 | /* by providing our own setattr() method, we skip this quotaism */ | |
198 | if ((old_ia_valid & ATTR_UID && ia->ia_uid != inode->i_uid) || | |
199 | (old_ia_valid & ATTR_GID && ia->ia_gid != inode->i_gid)) | |
200 | ret = DQUOT_TRANSFER(inode, ia) ? -EDQUOT : 0; | |
201 | ||
202 | /* pick out size-changing events */ | |
203 | if (ia->ia_valid & ATTR_SIZE) { | |
204 | loff_t size = i_size_read(inode); | |
205 | if (ia->ia_size != size) { | |
206 | ret = ramfs_nommu_resize(inode, ia->ia_size, size); | |
207 | if (ret < 0 || ia->ia_valid == ATTR_SIZE) | |
208 | goto out; | |
209 | } else { | |
210 | /* we skipped the truncate but must still update | |
211 | * timestamps | |
212 | */ | |
213 | ia->ia_valid |= ATTR_MTIME|ATTR_CTIME; | |
214 | } | |
215 | } | |
216 | ||
217 | ret = inode_setattr(inode, ia); | |
218 | out: | |
219 | ia->ia_valid = old_ia_valid; | |
220 | return ret; | |
221 | } | |
222 | ||
223 | /*****************************************************************************/ | |
224 | /* | |
225 | * try to determine where a shared mapping can be made | |
226 | * - we require that: | |
227 | * - the pages to be mapped must exist | |
228 | * - the pages be physically contiguous in sequence | |
229 | */ | |
230 | unsigned long ramfs_nommu_get_unmapped_area(struct file *file, | |
231 | unsigned long addr, unsigned long len, | |
232 | unsigned long pgoff, unsigned long flags) | |
233 | { | |
234 | unsigned long maxpages, lpages, nr, loop, ret; | |
235 | struct inode *inode = file->f_dentry->d_inode; | |
236 | struct page **pages = NULL, **ptr, *page; | |
237 | loff_t isize; | |
238 | ||
239 | if (!(flags & MAP_SHARED)) | |
240 | return addr; | |
241 | ||
242 | /* the mapping mustn't extend beyond the EOF */ | |
243 | lpages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
244 | isize = i_size_read(inode); | |
245 | ||
246 | ret = -EINVAL; | |
247 | maxpages = (isize + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
248 | if (pgoff >= maxpages) | |
249 | goto out; | |
250 | ||
251 | if (maxpages - pgoff < lpages) | |
252 | goto out; | |
253 | ||
254 | /* gang-find the pages */ | |
255 | ret = -ENOMEM; | |
256 | pages = kzalloc(lpages * sizeof(struct page *), GFP_KERNEL); | |
257 | if (!pages) | |
258 | goto out; | |
259 | ||
260 | nr = find_get_pages(inode->i_mapping, pgoff, lpages, pages); | |
261 | if (nr != lpages) | |
262 | goto out; /* leave if some pages were missing */ | |
263 | ||
264 | /* check the pages for physical adjacency */ | |
265 | ptr = pages; | |
266 | page = *ptr++; | |
267 | page++; | |
268 | for (loop = lpages; loop > 1; loop--) | |
269 | if (*ptr++ != page++) | |
270 | goto out; | |
271 | ||
272 | /* okay - all conditions fulfilled */ | |
273 | ret = (unsigned long) page_address(pages[0]); | |
274 | ||
275 | out: | |
276 | if (pages) { | |
277 | ptr = pages; | |
278 | for (loop = lpages; loop > 0; loop--) | |
279 | put_page(*ptr++); | |
280 | kfree(pages); | |
281 | } | |
282 | ||
283 | return ret; | |
284 | } | |
285 | ||
286 | /*****************************************************************************/ | |
287 | /* | |
21ff8216 | 288 | * set up a mapping for shared memory segments |
642fb4d1 DH |
289 | */ |
290 | int ramfs_nommu_mmap(struct file *file, struct vm_area_struct *vma) | |
291 | { | |
21ff8216 | 292 | return vma->vm_flags & VM_SHARED ? 0 : -ENOSYS; |
642fb4d1 | 293 | } |