blob: 4e74ad60339a2ebf255c23fb1020086ec1901c4c [file] [log] [blame]
Carsten Otteceffc072005-06-23 22:05:25 -07001/*
2 * linux/mm/filemap_xip.c
3 *
4 * Copyright (C) 2005 IBM Corporation
5 * Author: Carsten Otte <cotte@de.ibm.com>
6 *
7 * derived from linux/mm/filemap.c - Copyright (C) Linus Torvalds
8 *
9 */
10
11#include <linux/fs.h>
12#include <linux/pagemap.h>
13#include <linux/module.h>
14#include <linux/uio.h>
15#include <linux/rmap.h>
16#include <asm/tlbflush.h>
17#include "filemap.h"
18
19/*
20 * This is a file read routine for execute in place files, and uses
21 * the mapping->a_ops->get_xip_page() function for the actual low-level
22 * stuff.
23 *
24 * Note the struct file* is not used at all. It may be NULL.
25 */
26static void
27do_xip_mapping_read(struct address_space *mapping,
28 struct file_ra_state *_ra,
29 struct file *filp,
30 loff_t *ppos,
31 read_descriptor_t *desc,
32 read_actor_t actor)
33{
34 struct inode *inode = mapping->host;
35 unsigned long index, end_index, offset;
36 loff_t isize;
37
38 BUG_ON(!mapping->a_ops->get_xip_page);
39
40 index = *ppos >> PAGE_CACHE_SHIFT;
41 offset = *ppos & ~PAGE_CACHE_MASK;
42
43 isize = i_size_read(inode);
44 if (!isize)
45 goto out;
46
47 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
48 for (;;) {
49 struct page *page;
50 unsigned long nr, ret;
51
52 /* nr is the maximum number of bytes to copy from this page */
53 nr = PAGE_CACHE_SIZE;
54 if (index >= end_index) {
55 if (index > end_index)
56 goto out;
57 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
58 if (nr <= offset) {
59 goto out;
60 }
61 }
62 nr = nr - offset;
63
64 page = mapping->a_ops->get_xip_page(mapping,
65 index*(PAGE_SIZE/512), 0);
66 if (!page)
67 goto no_xip_page;
68 if (unlikely(IS_ERR(page))) {
69 if (PTR_ERR(page) == -ENODATA) {
70 /* sparse */
Carsten Otteafa597b2005-07-15 03:56:30 -070071 page = ZERO_PAGE(0);
Carsten Otteceffc072005-06-23 22:05:25 -070072 } else {
73 desc->error = PTR_ERR(page);
74 goto out;
75 }
Carsten Otteafa597b2005-07-15 03:56:30 -070076 }
Carsten Otteceffc072005-06-23 22:05:25 -070077
78 /* If users can be writing to this page using arbitrary
79 * virtual addresses, take care about potential aliasing
80 * before reading the page on the kernel side.
81 */
82 if (mapping_writably_mapped(mapping))
83 flush_dcache_page(page);
84
85 /*
Carsten Otteafa597b2005-07-15 03:56:30 -070086 * Ok, we have the page, so now we can copy it to user space...
Carsten Otteceffc072005-06-23 22:05:25 -070087 *
88 * The actor routine returns how many bytes were actually used..
89 * NOTE! This may not be the same as how much of a user buffer
90 * we filled up (we may be padding etc), so we can only update
91 * "pos" here (the actor routine has to update the user buffer
92 * pointers and the remaining count).
93 */
94 ret = actor(desc, page, offset, nr);
95 offset += ret;
96 index += offset >> PAGE_CACHE_SHIFT;
97 offset &= ~PAGE_CACHE_MASK;
98
99 if (ret == nr && desc->count)
100 continue;
101 goto out;
102
103no_xip_page:
104 /* Did not get the page. Report it */
105 desc->error = -EIO;
106 goto out;
107 }
108
109out:
110 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
111 if (filp)
112 file_accessed(filp);
113}
114
Carsten Otteceffc072005-06-23 22:05:25 -0700115ssize_t
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700116xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
Carsten Otteceffc072005-06-23 22:05:25 -0700117{
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700118 read_descriptor_t desc;
Carsten Otteceffc072005-06-23 22:05:25 -0700119
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700120 if (!access_ok(VERIFY_WRITE, buf, len))
121 return -EFAULT;
122
123 desc.written = 0;
124 desc.arg.buf = buf;
125 desc.count = len;
126 desc.error = 0;
127
128 do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp,
129 ppos, &desc, file_read_actor);
130
131 if (desc.written)
132 return desc.written;
133 else
134 return desc.error;
Carsten Otteceffc072005-06-23 22:05:25 -0700135}
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700136EXPORT_SYMBOL_GPL(xip_file_read);
Carsten Otteceffc072005-06-23 22:05:25 -0700137
138ssize_t
139xip_file_sendfile(struct file *in_file, loff_t *ppos,
140 size_t count, read_actor_t actor, void *target)
141{
142 read_descriptor_t desc;
143
144 if (!count)
145 return 0;
146
147 desc.written = 0;
148 desc.count = count;
149 desc.arg.data = target;
150 desc.error = 0;
151
152 do_xip_mapping_read(in_file->f_mapping, &in_file->f_ra, in_file,
153 ppos, &desc, actor);
154 if (desc.written)
155 return desc.written;
156 return desc.error;
157}
158EXPORT_SYMBOL_GPL(xip_file_sendfile);
159
160/*
161 * __xip_unmap is invoked from xip_unmap and
162 * xip_write
163 *
164 * This function walks all vmas of the address_space and unmaps the
Carsten Otteafa597b2005-07-15 03:56:30 -0700165 * ZERO_PAGE when found at pgoff. Should it go in rmap.c?
Carsten Otteceffc072005-06-23 22:05:25 -0700166 */
167static void
168__xip_unmap (struct address_space * mapping,
169 unsigned long pgoff)
170{
171 struct vm_area_struct *vma;
172 struct mm_struct *mm;
173 struct prio_tree_iter iter;
174 unsigned long address;
175 pte_t *pte;
176 pte_t pteval;
Hugh Dickins67b02f12005-10-29 18:16:31 -0700177 struct page *page;
Carsten Otteceffc072005-06-23 22:05:25 -0700178
179 spin_lock(&mapping->i_mmap_lock);
180 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
181 mm = vma->vm_mm;
182 address = vma->vm_start +
183 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
184 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Hugh Dickins67b02f12005-10-29 18:16:31 -0700185 page = ZERO_PAGE(address);
Carsten Otteceffc072005-06-23 22:05:25 -0700186 /*
187 * We need the page_table_lock to protect us from page faults,
188 * munmap, fork, etc...
189 */
Nick Pigginb5810032005-10-29 18:16:12 -0700190 pte = page_check_address(page, mm, address);
Carsten Otteceffc072005-06-23 22:05:25 -0700191 if (!IS_ERR(pte)) {
192 /* Nuke the page table entry. */
Geert Uytterhoeven082ff0a2005-07-12 13:58:18 -0700193 flush_cache_page(vma, address, pte_pfn(*pte));
Carsten Otteceffc072005-06-23 22:05:25 -0700194 pteval = ptep_clear_flush(vma, address, pte);
Nick Pigginb5810032005-10-29 18:16:12 -0700195 page_remove_rmap(page);
196 dec_mm_counter(mm, file_rss);
Carsten Otteceffc072005-06-23 22:05:25 -0700197 BUG_ON(pte_dirty(pteval));
198 pte_unmap(pte);
199 spin_unlock(&mm->page_table_lock);
Nick Pigginb5810032005-10-29 18:16:12 -0700200 page_cache_release(page);
Carsten Otteceffc072005-06-23 22:05:25 -0700201 }
202 }
203 spin_unlock(&mapping->i_mmap_lock);
204}
205
206/*
207 * xip_nopage() is invoked via the vma operations vector for a
208 * mapped memory region to read in file data during a page fault.
209 *
210 * This function is derived from filemap_nopage, but used for execute in place
211 */
212static struct page *
213xip_file_nopage(struct vm_area_struct * area,
214 unsigned long address,
215 int *type)
216{
217 struct file *file = area->vm_file;
218 struct address_space *mapping = file->f_mapping;
219 struct inode *inode = mapping->host;
220 struct page *page;
221 unsigned long size, pgoff, endoff;
222
223 pgoff = ((address - area->vm_start) >> PAGE_CACHE_SHIFT)
224 + area->vm_pgoff;
225 endoff = ((area->vm_end - area->vm_start) >> PAGE_CACHE_SHIFT)
226 + area->vm_pgoff;
227
228 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
229 if (pgoff >= size) {
230 return NULL;
231 }
232
233 page = mapping->a_ops->get_xip_page(mapping, pgoff*(PAGE_SIZE/512), 0);
234 if (!IS_ERR(page)) {
Nick Pigginb5810032005-10-29 18:16:12 -0700235 goto out;
Carsten Otteceffc072005-06-23 22:05:25 -0700236 }
237 if (PTR_ERR(page) != -ENODATA)
238 return NULL;
239
240 /* sparse block */
241 if ((area->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
242 (area->vm_flags & (VM_SHARED| VM_MAYSHARE)) &&
243 (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
244 /* maybe shared writable, allocate new block */
245 page = mapping->a_ops->get_xip_page (mapping,
246 pgoff*(PAGE_SIZE/512), 1);
247 if (IS_ERR(page))
248 return NULL;
Carsten Otteceffc072005-06-23 22:05:25 -0700249 /* unmap page at pgoff from all other vmas */
250 __xip_unmap(mapping, pgoff);
251 } else {
Carsten Otteafa597b2005-07-15 03:56:30 -0700252 /* not shared and writable, use ZERO_PAGE() */
253 page = ZERO_PAGE(address);
Carsten Otteceffc072005-06-23 22:05:25 -0700254 }
255
Nick Pigginb5810032005-10-29 18:16:12 -0700256out:
257 page_cache_get(page);
Carsten Otteceffc072005-06-23 22:05:25 -0700258 return page;
259}
260
261static struct vm_operations_struct xip_file_vm_ops = {
262 .nopage = xip_file_nopage,
263};
264
265int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
266{
267 BUG_ON(!file->f_mapping->a_ops->get_xip_page);
268
269 file_accessed(file);
270 vma->vm_ops = &xip_file_vm_ops;
271 return 0;
272}
273EXPORT_SYMBOL_GPL(xip_file_mmap);
274
275static ssize_t
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700276__xip_file_write(struct file *filp, const char __user *buf,
277 size_t count, loff_t pos, loff_t *ppos)
Carsten Otteceffc072005-06-23 22:05:25 -0700278{
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700279 struct address_space * mapping = filp->f_mapping;
Carsten Otteceffc072005-06-23 22:05:25 -0700280 struct address_space_operations *a_ops = mapping->a_ops;
281 struct inode *inode = mapping->host;
282 long status = 0;
283 struct page *page;
284 size_t bytes;
Carsten Otteceffc072005-06-23 22:05:25 -0700285 ssize_t written = 0;
286
287 BUG_ON(!mapping->a_ops->get_xip_page);
288
Carsten Otteceffc072005-06-23 22:05:25 -0700289 do {
290 unsigned long index;
291 unsigned long offset;
292 size_t copied;
293
294 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
295 index = pos >> PAGE_CACHE_SHIFT;
296 bytes = PAGE_CACHE_SIZE - offset;
297 if (bytes > count)
298 bytes = count;
299
300 /*
301 * Bring in the user page that we will copy from _first_.
302 * Otherwise there's a nasty deadlock on copying from the
303 * same page as we're writing to, without it being marked
304 * up-to-date.
305 */
306 fault_in_pages_readable(buf, bytes);
307
308 page = a_ops->get_xip_page(mapping,
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700309 index*(PAGE_SIZE/512), 0);
Carsten Otteceffc072005-06-23 22:05:25 -0700310 if (IS_ERR(page) && (PTR_ERR(page) == -ENODATA)) {
311 /* we allocate a new page unmap it */
312 page = a_ops->get_xip_page(mapping,
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700313 index*(PAGE_SIZE/512), 1);
Carsten Otteceffc072005-06-23 22:05:25 -0700314 if (!IS_ERR(page))
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700315 /* unmap page at pgoff from all other vmas */
316 __xip_unmap(mapping, index);
Carsten Otteceffc072005-06-23 22:05:25 -0700317 }
318
319 if (IS_ERR(page)) {
320 status = PTR_ERR(page);
321 break;
322 }
323
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700324 copied = filemap_copy_from_user(page, offset, buf, bytes);
Carsten Otteceffc072005-06-23 22:05:25 -0700325 flush_dcache_page(page);
326 if (likely(copied > 0)) {
327 status = copied;
328
329 if (status >= 0) {
330 written += status;
331 count -= status;
332 pos += status;
333 buf += status;
Carsten Otteceffc072005-06-23 22:05:25 -0700334 }
335 }
336 if (unlikely(copied != bytes))
337 if (status >= 0)
338 status = -EFAULT;
339 if (status < 0)
340 break;
341 } while (count);
342 *ppos = pos;
343 /*
344 * No need to use i_size_read() here, the i_size
345 * cannot change under us because we hold i_sem.
346 */
347 if (pos > inode->i_size) {
348 i_size_write(inode, pos);
349 mark_inode_dirty(inode);
350 }
351
352 return written ? written : status;
353}
354
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700355ssize_t
356xip_file_write(struct file *filp, const char __user *buf, size_t len,
357 loff_t *ppos)
Carsten Otteceffc072005-06-23 22:05:25 -0700358{
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700359 struct address_space *mapping = filp->f_mapping;
360 struct inode *inode = mapping->host;
361 size_t count;
362 loff_t pos;
363 ssize_t ret;
Carsten Otteceffc072005-06-23 22:05:25 -0700364
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700365 down(&inode->i_sem);
Carsten Otteceffc072005-06-23 22:05:25 -0700366
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700367 if (!access_ok(VERIFY_READ, buf, len)) {
368 ret=-EFAULT;
369 goto out_up;
Carsten Otteceffc072005-06-23 22:05:25 -0700370 }
371
Carsten Otteceffc072005-06-23 22:05:25 -0700372 pos = *ppos;
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700373 count = len;
Carsten Otteceffc072005-06-23 22:05:25 -0700374
375 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
376
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700377 /* We can write back this queue in page reclaim */
378 current->backing_dev_info = mapping->backing_dev_info;
Carsten Otteceffc072005-06-23 22:05:25 -0700379
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700380 ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
381 if (ret)
382 goto out_backing;
Carsten Otteceffc072005-06-23 22:05:25 -0700383 if (count == 0)
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700384 goto out_backing;
Carsten Otteceffc072005-06-23 22:05:25 -0700385
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700386 ret = remove_suid(filp->f_dentry);
387 if (ret)
388 goto out_backing;
Carsten Otteceffc072005-06-23 22:05:25 -0700389
390 inode_update_time(inode, 1);
391
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700392 ret = __xip_file_write (filp, buf, count, pos, ppos);
Carsten Otteceffc072005-06-23 22:05:25 -0700393
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700394 out_backing:
395 current->backing_dev_info = NULL;
396 out_up:
Carsten Otteceffc072005-06-23 22:05:25 -0700397 up(&inode->i_sem);
398 return ret;
399}
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700400EXPORT_SYMBOL_GPL(xip_file_write);
Carsten Otteceffc072005-06-23 22:05:25 -0700401
402/*
403 * truncate a page used for execute in place
404 * functionality is analog to block_truncate_page but does use get_xip_page
405 * to get the page instead of page cache
406 */
407int
408xip_truncate_page(struct address_space *mapping, loff_t from)
409{
410 pgoff_t index = from >> PAGE_CACHE_SHIFT;
411 unsigned offset = from & (PAGE_CACHE_SIZE-1);
412 unsigned blocksize;
413 unsigned length;
414 struct page *page;
415 void *kaddr;
Carsten Otteceffc072005-06-23 22:05:25 -0700416
417 BUG_ON(!mapping->a_ops->get_xip_page);
418
419 blocksize = 1 << mapping->host->i_blkbits;
420 length = offset & (blocksize - 1);
421
422 /* Block boundary? Nothing to do */
423 if (!length)
424 return 0;
425
426 length = blocksize - length;
427
428 page = mapping->a_ops->get_xip_page(mapping,
429 index*(PAGE_SIZE/512), 0);
Carsten Otteceffc072005-06-23 22:05:25 -0700430 if (!page)
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700431 return -ENOMEM;
Carsten Otteceffc072005-06-23 22:05:25 -0700432 if (unlikely(IS_ERR(page))) {
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700433 if (PTR_ERR(page) == -ENODATA)
Carsten Otteceffc072005-06-23 22:05:25 -0700434 /* Hole? No need to truncate */
435 return 0;
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700436 else
437 return PTR_ERR(page);
Carsten Otteafa597b2005-07-15 03:56:30 -0700438 }
Carsten Otteceffc072005-06-23 22:05:25 -0700439 kaddr = kmap_atomic(page, KM_USER0);
440 memset(kaddr + offset, 0, length);
441 kunmap_atomic(kaddr, KM_USER0);
442
443 flush_dcache_page(page);
Carsten Otteeb6fe0c2005-06-23 22:05:28 -0700444 return 0;
Carsten Otteceffc072005-06-23 22:05:25 -0700445}
446EXPORT_SYMBOL_GPL(xip_truncate_page);