]> bbs.cooldavid.org Git - net-next-2.6.git/blame - mm/filemap_xip.c
[PATCH] holepunch: fix mmap_sem i_mutex deadlock
[net-next-2.6.git] / mm / filemap_xip.c
CommitLineData
ceffc078
CO
1/*
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 */
afa597ba 71 page = ZERO_PAGE(0);
ceffc078
CO
72 } else {
73 desc->error = PTR_ERR(page);
74 goto out;
75 }
afa597ba 76 }
ceffc078
CO
77
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 /*
afa597ba 86 * Ok, we have the page, so now we can copy it to user space...
ceffc078
CO
87 *
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
ceffc078 115ssize_t
eb6fe0c3 116xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
ceffc078 117{
eb6fe0c3 118 read_descriptor_t desc;
ceffc078 119
eb6fe0c3
CO
120 if (!access_ok(VERIFY_WRITE, buf, len))
121 return -EFAULT;
ceffc078 122
eb6fe0c3
CO
123 desc.written = 0;
124 desc.arg.buf = buf;
125 desc.count = len;
126 desc.error = 0;
ceffc078 127
eb6fe0c3
CO
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;
ceffc078 135}
eb6fe0c3 136EXPORT_SYMBOL_GPL(xip_file_read);
ceffc078
CO
137
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
afa597ba 165 * ZERO_PAGE when found at pgoff. Should it go in rmap.c?
ceffc078
CO
166 */
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;
c0718806 177 spinlock_t *ptl;
67b02f11 178 struct page *page;
ceffc078
CO
179
180 spin_lock(&mapping->i_mmap_lock);
181 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
182 mm = vma->vm_mm;
183 address = vma->vm_start +
184 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
185 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
701dfbc1 186 page = ZERO_PAGE(0);
c0718806
HD
187 pte = page_check_address(page, mm, address, &ptl);
188 if (pte) {
ceffc078 189 /* Nuke the page table entry. */
082ff0a9 190 flush_cache_page(vma, address, pte_pfn(*pte));
ceffc078 191 pteval = ptep_clear_flush(vma, address, pte);
7de6b805 192 page_remove_rmap(page, vma);
b5810039 193 dec_mm_counter(mm, file_rss);
ceffc078 194 BUG_ON(pte_dirty(pteval));
c0718806 195 pte_unmap_unlock(pte, ptl);
b5810039 196 page_cache_release(page);
ceffc078
CO
197 }
198 }
199 spin_unlock(&mapping->i_mmap_lock);
200}
201
202/*
203 * xip_nopage() is invoked via the vma operations vector for a
204 * mapped memory region to read in file data during a page fault.
205 *
206 * This function is derived from filemap_nopage, but used for execute in place
207 */
208static struct page *
209xip_file_nopage(struct vm_area_struct * area,
210 unsigned long address,
211 int *type)
212{
213 struct file *file = area->vm_file;
214 struct address_space *mapping = file->f_mapping;
215 struct inode *inode = mapping->host;
216 struct page *page;
217 unsigned long size, pgoff, endoff;
218
219 pgoff = ((address - area->vm_start) >> PAGE_CACHE_SHIFT)
220 + area->vm_pgoff;
221 endoff = ((area->vm_end - area->vm_start) >> PAGE_CACHE_SHIFT)
222 + area->vm_pgoff;
223
224 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
225 if (pgoff >= size) {
226 return NULL;
227 }
228
229 page = mapping->a_ops->get_xip_page(mapping, pgoff*(PAGE_SIZE/512), 0);
230 if (!IS_ERR(page)) {
b5810039 231 goto out;
ceffc078
CO
232 }
233 if (PTR_ERR(page) != -ENODATA)
234 return NULL;
235
236 /* sparse block */
237 if ((area->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
238 (area->vm_flags & (VM_SHARED| VM_MAYSHARE)) &&
239 (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
240 /* maybe shared writable, allocate new block */
241 page = mapping->a_ops->get_xip_page (mapping,
242 pgoff*(PAGE_SIZE/512), 1);
243 if (IS_ERR(page))
244 return NULL;
ceffc078
CO
245 /* unmap page at pgoff from all other vmas */
246 __xip_unmap(mapping, pgoff);
247 } else {
afa597ba 248 /* not shared and writable, use ZERO_PAGE() */
701dfbc1 249 page = ZERO_PAGE(0);
ceffc078
CO
250 }
251
b5810039
NP
252out:
253 page_cache_get(page);
ceffc078
CO
254 return page;
255}
256
257static struct vm_operations_struct xip_file_vm_ops = {
258 .nopage = xip_file_nopage,
259};
260
261int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
262{
263 BUG_ON(!file->f_mapping->a_ops->get_xip_page);
264
265 file_accessed(file);
266 vma->vm_ops = &xip_file_vm_ops;
267 return 0;
268}
269EXPORT_SYMBOL_GPL(xip_file_mmap);
270
271static ssize_t
eb6fe0c3
CO
272__xip_file_write(struct file *filp, const char __user *buf,
273 size_t count, loff_t pos, loff_t *ppos)
ceffc078 274{
eb6fe0c3 275 struct address_space * mapping = filp->f_mapping;
f5e54d6e 276 const struct address_space_operations *a_ops = mapping->a_ops;
ceffc078
CO
277 struct inode *inode = mapping->host;
278 long status = 0;
279 struct page *page;
280 size_t bytes;
ceffc078
CO
281 ssize_t written = 0;
282
283 BUG_ON(!mapping->a_ops->get_xip_page);
284
ceffc078
CO
285 do {
286 unsigned long index;
287 unsigned long offset;
288 size_t copied;
289
290 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
291 index = pos >> PAGE_CACHE_SHIFT;
292 bytes = PAGE_CACHE_SIZE - offset;
293 if (bytes > count)
294 bytes = count;
295
296 /*
297 * Bring in the user page that we will copy from _first_.
298 * Otherwise there's a nasty deadlock on copying from the
299 * same page as we're writing to, without it being marked
300 * up-to-date.
301 */
302 fault_in_pages_readable(buf, bytes);
303
304 page = a_ops->get_xip_page(mapping,
eb6fe0c3 305 index*(PAGE_SIZE/512), 0);
ceffc078
CO
306 if (IS_ERR(page) && (PTR_ERR(page) == -ENODATA)) {
307 /* we allocate a new page unmap it */
308 page = a_ops->get_xip_page(mapping,
eb6fe0c3 309 index*(PAGE_SIZE/512), 1);
ceffc078 310 if (!IS_ERR(page))
eb6fe0c3
CO
311 /* unmap page at pgoff from all other vmas */
312 __xip_unmap(mapping, index);
ceffc078
CO
313 }
314
315 if (IS_ERR(page)) {
316 status = PTR_ERR(page);
317 break;
318 }
319
eb6fe0c3 320 copied = filemap_copy_from_user(page, offset, buf, bytes);
ceffc078
CO
321 flush_dcache_page(page);
322 if (likely(copied > 0)) {
323 status = copied;
324
325 if (status >= 0) {
326 written += status;
327 count -= status;
328 pos += status;
329 buf += status;
ceffc078
CO
330 }
331 }
332 if (unlikely(copied != bytes))
333 if (status >= 0)
334 status = -EFAULT;
335 if (status < 0)
336 break;
337 } while (count);
338 *ppos = pos;
339 /*
340 * No need to use i_size_read() here, the i_size
1b1dcc1b 341 * cannot change under us because we hold i_mutex.
ceffc078
CO
342 */
343 if (pos > inode->i_size) {
344 i_size_write(inode, pos);
345 mark_inode_dirty(inode);
346 }
347
348 return written ? written : status;
349}
350
eb6fe0c3
CO
351ssize_t
352xip_file_write(struct file *filp, const char __user *buf, size_t len,
353 loff_t *ppos)
ceffc078 354{
eb6fe0c3
CO
355 struct address_space *mapping = filp->f_mapping;
356 struct inode *inode = mapping->host;
357 size_t count;
358 loff_t pos;
359 ssize_t ret;
ceffc078 360
1b1dcc1b 361 mutex_lock(&inode->i_mutex);
ceffc078 362
eb6fe0c3
CO
363 if (!access_ok(VERIFY_READ, buf, len)) {
364 ret=-EFAULT;
365 goto out_up;
ceffc078
CO
366 }
367
ceffc078 368 pos = *ppos;
eb6fe0c3 369 count = len;
ceffc078
CO
370
371 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
372
eb6fe0c3
CO
373 /* We can write back this queue in page reclaim */
374 current->backing_dev_info = mapping->backing_dev_info;
ceffc078 375
eb6fe0c3
CO
376 ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
377 if (ret)
378 goto out_backing;
ceffc078 379 if (count == 0)
eb6fe0c3 380 goto out_backing;
ceffc078 381
d3ac7f89 382 ret = remove_suid(filp->f_path.dentry);
eb6fe0c3
CO
383 if (ret)
384 goto out_backing;
ceffc078 385
870f4817 386 file_update_time(filp);
ceffc078 387
eb6fe0c3 388 ret = __xip_file_write (filp, buf, count, pos, ppos);
ceffc078 389
eb6fe0c3
CO
390 out_backing:
391 current->backing_dev_info = NULL;
392 out_up:
1b1dcc1b 393 mutex_unlock(&inode->i_mutex);
ceffc078
CO
394 return ret;
395}
eb6fe0c3 396EXPORT_SYMBOL_GPL(xip_file_write);
ceffc078
CO
397
398/*
399 * truncate a page used for execute in place
400 * functionality is analog to block_truncate_page but does use get_xip_page
401 * to get the page instead of page cache
402 */
403int
404xip_truncate_page(struct address_space *mapping, loff_t from)
405{
406 pgoff_t index = from >> PAGE_CACHE_SHIFT;
407 unsigned offset = from & (PAGE_CACHE_SIZE-1);
408 unsigned blocksize;
409 unsigned length;
410 struct page *page;
411 void *kaddr;
ceffc078
CO
412
413 BUG_ON(!mapping->a_ops->get_xip_page);
414
415 blocksize = 1 << mapping->host->i_blkbits;
416 length = offset & (blocksize - 1);
417
418 /* Block boundary? Nothing to do */
419 if (!length)
420 return 0;
421
422 length = blocksize - length;
423
424 page = mapping->a_ops->get_xip_page(mapping,
425 index*(PAGE_SIZE/512), 0);
ceffc078 426 if (!page)
eb6fe0c3 427 return -ENOMEM;
ceffc078 428 if (unlikely(IS_ERR(page))) {
eb6fe0c3 429 if (PTR_ERR(page) == -ENODATA)
ceffc078
CO
430 /* Hole? No need to truncate */
431 return 0;
eb6fe0c3
CO
432 else
433 return PTR_ERR(page);
afa597ba 434 }
ceffc078
CO
435 kaddr = kmap_atomic(page, KM_USER0);
436 memset(kaddr + offset, 0, length);
437 kunmap_atomic(kaddr, KM_USER0);
438
439 flush_dcache_page(page);
eb6fe0c3 440 return 0;
ceffc078
CO
441}
442EXPORT_SYMBOL_GPL(xip_truncate_page);