blob: 52312ec93ff4ae5c0753cdb37af0991f1f733a4c [file] [log] [blame]
David Howells642fb4d2006-01-06 00:11:41 -08001/* 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>
Dimitri Gorokhovik131612d2006-12-29 16:48:24 -080014#include <linux/mm.h>
David Howells642fb4d2006-01-06 00:11:41 -080015#include <linux/pagemap.h>
16#include <linux/highmem.h>
17#include <linux/init.h>
18#include <linux/string.h>
David Howells642fb4d2006-01-06 00:11:41 -080019#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
28static int ramfs_nommu_setattr(struct dentry *, struct iattr *);
29
Christoph Hellwigf5e54d62006-06-28 04:26:44 -070030const struct address_space_operations ramfs_aops = {
David Howells642fb4d2006-01-06 00:11:41 -080031 .readpage = simple_readpage,
Nick Piggin800d15a2007-10-16 01:25:03 -070032 .write_begin = simple_write_begin,
33 .write_end = simple_write_end,
Ken Chen46626292007-02-10 01:43:17 -080034 .set_page_dirty = __set_page_dirty_no_writeback,
David Howells642fb4d2006-01-06 00:11:41 -080035};
36
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080037const struct file_operations ramfs_file_operations = {
David Howells642fb4d2006-01-06 00:11:41 -080038 .mmap = ramfs_nommu_mmap,
39 .get_unmapped_area = ramfs_nommu_get_unmapped_area,
Badari Pulavarty543ade12006-09-30 23:28:48 -070040 .read = do_sync_read,
41 .aio_read = generic_file_aio_read,
42 .write = do_sync_write,
43 .aio_write = generic_file_aio_write,
David Howells642fb4d2006-01-06 00:11:41 -080044 .fsync = simple_sync_file,
Jens Axboe5ffc4ef2007-06-01 11:49:19 +020045 .splice_read = generic_file_splice_read,
Octavian Purdila8b3d3562008-07-04 09:33:33 +020046 .splice_write = generic_file_splice_write,
David Howells642fb4d2006-01-06 00:11:41 -080047 .llseek = generic_file_llseek,
48};
49
Arjan van de Venc5ef1c42007-02-12 00:55:40 -080050const struct inode_operations ramfs_file_inode_operations = {
David Howells642fb4d2006-01-06 00:11:41 -080051 .setattr = ramfs_nommu_setattr,
52 .getattr = simple_getattr,
53};
54
55/*****************************************************************************/
56/*
57 * add a contiguous set of pages into a ramfs inode when it's truncated from
58 * size 0 on the assumption that it's going to be used for an mmap of shared
59 * memory
60 */
61static int ramfs_nommu_expand_for_mapping(struct inode *inode, size_t newsize)
62{
63 struct pagevec lru_pvec;
64 unsigned long npages, xpages, loop, limit;
65 struct page *pages;
66 unsigned order;
67 void *data;
68 int ret;
69
70 /* make various checks */
71 order = get_order(newsize);
72 if (unlikely(order >= MAX_ORDER))
73 goto too_big;
74
75 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
76 if (limit != RLIM_INFINITY && newsize > limit)
77 goto fsize_exceeded;
78
79 if (newsize > inode->i_sb->s_maxbytes)
80 goto too_big;
81
82 i_size_write(inode, newsize);
83
84 /* allocate enough contiguous pages to be able to satisfy the
85 * request */
86 pages = alloc_pages(mapping_gfp_mask(inode->i_mapping), order);
87 if (!pages)
88 return -ENOMEM;
89
90 /* split the high-order page into an array of single pages */
91 xpages = 1UL << order;
92 npages = (newsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
93
Nick Piggin84097512006-03-22 00:08:34 -080094 split_page(pages, order);
David Howells642fb4d2006-01-06 00:11:41 -080095
96 /* trim off any pages we don't actually require */
97 for (loop = npages; loop < xpages; loop++)
98 __free_page(pages + loop);
99
100 /* clear the memory we allocated */
101 newsize = PAGE_SIZE * npages;
102 data = page_address(pages);
103 memset(data, 0, newsize);
104
105 /* attach all the pages to the inode's address space */
106 pagevec_init(&lru_pvec, 0);
107 for (loop = 0; loop < npages; loop++) {
108 struct page *page = pages + loop;
109
110 ret = add_to_page_cache(page, inode->i_mapping, loop, GFP_KERNEL);
111 if (ret < 0)
112 goto add_error;
113
114 if (!pagevec_add(&lru_pvec, page))
115 __pagevec_lru_add(&lru_pvec);
116
117 unlock_page(page);
118 }
119
120 pagevec_lru_add(&lru_pvec);
121 return 0;
122
123 fsize_exceeded:
124 send_sig(SIGXFSZ, current, 0);
125 too_big:
126 return -EFBIG;
127
128 add_error:
129 page_cache_release(pages + loop);
130 for (loop++; loop < npages; loop++)
131 __free_page(pages + loop);
132 return ret;
133}
134
135/*****************************************************************************/
136/*
137 * check that file shrinkage doesn't leave any VMAs dangling in midair
138 */
139static int ramfs_nommu_check_mappings(struct inode *inode,
140 size_t newsize, size_t size)
141{
142 struct vm_area_struct *vma;
143 struct prio_tree_iter iter;
144
145 /* search for VMAs that fall within the dead zone */
146 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
147 newsize >> PAGE_SHIFT,
148 (size + PAGE_SIZE - 1) >> PAGE_SHIFT
149 ) {
150 /* found one - only interested if it's shared out of the page
151 * cache */
152 if (vma->vm_flags & VM_SHARED)
153 return -ETXTBSY; /* not quite true, but near enough */
154 }
155
156 return 0;
157}
158
159/*****************************************************************************/
160/*
161 *
162 */
163static int ramfs_nommu_resize(struct inode *inode, loff_t newsize, loff_t size)
164{
165 int ret;
166
167 /* assume a truncate from zero size is going to be for the purposes of
168 * shared mmap */
169 if (size == 0) {
170 if (unlikely(newsize >> 32))
171 return -EFBIG;
172
173 return ramfs_nommu_expand_for_mapping(inode, newsize);
174 }
175
176 /* check that a decrease in size doesn't cut off any shared mappings */
177 if (newsize < size) {
178 ret = ramfs_nommu_check_mappings(inode, newsize, size);
179 if (ret < 0)
180 return ret;
181 }
182
Bryan Wu3f0a6762007-05-31 11:31:55 +0800183 ret = vmtruncate(inode, newsize);
David Howells642fb4d2006-01-06 00:11:41 -0800184
185 return ret;
186}
187
188/*****************************************************************************/
189/*
190 * handle a change of attributes
191 * - we're specifically interested in a change of size
192 */
193static int ramfs_nommu_setattr(struct dentry *dentry, struct iattr *ia)
194{
195 struct inode *inode = dentry->d_inode;
196 unsigned int old_ia_valid = ia->ia_valid;
197 int ret = 0;
198
Bryan Wu85f60382007-06-05 11:02:01 +0800199 /* POSIX UID/GID verification for setting inode attributes */
200 ret = inode_change_ok(inode, ia);
201 if (ret)
202 return ret;
203
David Howells642fb4d2006-01-06 00:11:41 -0800204 /* by providing our own setattr() method, we skip this quotaism */
205 if ((old_ia_valid & ATTR_UID && ia->ia_uid != inode->i_uid) ||
206 (old_ia_valid & ATTR_GID && ia->ia_gid != inode->i_gid))
207 ret = DQUOT_TRANSFER(inode, ia) ? -EDQUOT : 0;
208
209 /* pick out size-changing events */
210 if (ia->ia_valid & ATTR_SIZE) {
211 loff_t size = i_size_read(inode);
212 if (ia->ia_size != size) {
213 ret = ramfs_nommu_resize(inode, ia->ia_size, size);
214 if (ret < 0 || ia->ia_valid == ATTR_SIZE)
215 goto out;
216 } else {
217 /* we skipped the truncate but must still update
218 * timestamps
219 */
220 ia->ia_valid |= ATTR_MTIME|ATTR_CTIME;
221 }
222 }
223
224 ret = inode_setattr(inode, ia);
225 out:
226 ia->ia_valid = old_ia_valid;
227 return ret;
228}
229
230/*****************************************************************************/
231/*
232 * try to determine where a shared mapping can be made
233 * - we require that:
234 * - the pages to be mapped must exist
235 * - the pages be physically contiguous in sequence
236 */
237unsigned long ramfs_nommu_get_unmapped_area(struct file *file,
238 unsigned long addr, unsigned long len,
239 unsigned long pgoff, unsigned long flags)
240{
241 unsigned long maxpages, lpages, nr, loop, ret;
Josef Sipeka57c4d62006-12-08 02:37:32 -0800242 struct inode *inode = file->f_path.dentry->d_inode;
David Howells642fb4d2006-01-06 00:11:41 -0800243 struct page **pages = NULL, **ptr, *page;
244 loff_t isize;
245
246 if (!(flags & MAP_SHARED))
247 return addr;
248
249 /* the mapping mustn't extend beyond the EOF */
250 lpages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
251 isize = i_size_read(inode);
252
253 ret = -EINVAL;
254 maxpages = (isize + PAGE_SIZE - 1) >> PAGE_SHIFT;
255 if (pgoff >= maxpages)
256 goto out;
257
258 if (maxpages - pgoff < lpages)
259 goto out;
260
261 /* gang-find the pages */
262 ret = -ENOMEM;
263 pages = kzalloc(lpages * sizeof(struct page *), GFP_KERNEL);
264 if (!pages)
265 goto out;
266
267 nr = find_get_pages(inode->i_mapping, pgoff, lpages, pages);
268 if (nr != lpages)
269 goto out; /* leave if some pages were missing */
270
271 /* check the pages for physical adjacency */
272 ptr = pages;
273 page = *ptr++;
274 page++;
275 for (loop = lpages; loop > 1; loop--)
276 if (*ptr++ != page++)
277 goto out;
278
279 /* okay - all conditions fulfilled */
280 ret = (unsigned long) page_address(pages[0]);
281
282 out:
283 if (pages) {
284 ptr = pages;
285 for (loop = lpages; loop > 0; loop--)
286 put_page(*ptr++);
287 kfree(pages);
288 }
289
290 return ret;
291}
292
293/*****************************************************************************/
294/*
David Howells21ff8212006-07-10 04:44:52 -0700295 * set up a mapping for shared memory segments
David Howells642fb4d2006-01-06 00:11:41 -0800296 */
297int ramfs_nommu_mmap(struct file *file, struct vm_area_struct *vma)
298{
David Howells2e92a3b2007-07-31 00:37:24 -0700299 if (!(vma->vm_flags & VM_SHARED))
300 return -ENOSYS;
301
302 file_accessed(file);
303 vma->vm_ops = &generic_file_vm_ops;
304 return 0;
David Howells642fb4d2006-01-06 00:11:41 -0800305}