net: ipv6 bind to device issue
[linux-2.6-block.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>
cddb8a5c 16#include <linux/mmu_notifier.h>
e8edc6e0 17#include <linux/sched.h>
538f8ea6
NP
18#include <linux/seqlock.h>
19#include <linux/mutex.h>
5a0e3ad6 20#include <linux/gfp.h>
ceffc078 21#include <asm/tlbflush.h>
70688e4d 22#include <asm/io.h>
ceffc078 23
a76c0b97
CO
24/*
25 * We do use our own empty page to avoid interference with other users
26 * of ZERO_PAGE(), such as /dev/zero
27 */
538f8ea6
NP
28static DEFINE_MUTEX(xip_sparse_mutex);
29static seqcount_t xip_sparse_seq = SEQCNT_ZERO;
a76c0b97
CO
30static struct page *__xip_sparse_page;
31
538f8ea6 32/* called under xip_sparse_mutex */
a76c0b97
CO
33static struct page *xip_sparse_page(void)
34{
35 if (!__xip_sparse_page) {
c51b1a16
AM
36 struct page *page = alloc_page(GFP_HIGHUSER | __GFP_ZERO);
37
538f8ea6
NP
38 if (page)
39 __xip_sparse_page = page;
a76c0b97
CO
40 }
41 return __xip_sparse_page;
42}
43
ceffc078
CO
44/*
45 * This is a file read routine for execute in place files, and uses
70688e4d 46 * the mapping->a_ops->get_xip_mem() function for the actual low-level
ceffc078
CO
47 * stuff.
48 *
49 * Note the struct file* is not used at all. It may be NULL.
50 */
70688e4d 51static ssize_t
ceffc078
CO
52do_xip_mapping_read(struct address_space *mapping,
53 struct file_ra_state *_ra,
54 struct file *filp,
70688e4d
NP
55 char __user *buf,
56 size_t len,
57 loff_t *ppos)
ceffc078
CO
58{
59 struct inode *inode = mapping->host;
2004dc8e
JK
60 pgoff_t index, end_index;
61 unsigned long offset;
70688e4d
NP
62 loff_t isize, pos;
63 size_t copied = 0, error = 0;
ceffc078 64
70688e4d 65 BUG_ON(!mapping->a_ops->get_xip_mem);
ceffc078 66
70688e4d
NP
67 pos = *ppos;
68 index = pos >> PAGE_CACHE_SHIFT;
69 offset = pos & ~PAGE_CACHE_MASK;
ceffc078
CO
70
71 isize = i_size_read(inode);
72 if (!isize)
73 goto out;
74
75 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
70688e4d
NP
76 do {
77 unsigned long nr, left;
78 void *xip_mem;
79 unsigned long xip_pfn;
80 int zero = 0;
ceffc078
CO
81
82 /* nr is the maximum number of bytes to copy from this page */
83 nr = PAGE_CACHE_SIZE;
84 if (index >= end_index) {
85 if (index > end_index)
86 goto out;
87 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
88 if (nr <= offset) {
89 goto out;
90 }
91 }
92 nr = nr - offset;
58984ce2
MS
93 if (nr > len - copied)
94 nr = len - copied;
ceffc078 95
70688e4d
NP
96 error = mapping->a_ops->get_xip_mem(mapping, index, 0,
97 &xip_mem, &xip_pfn);
98 if (unlikely(error)) {
99 if (error == -ENODATA) {
ceffc078 100 /* sparse */
70688e4d
NP
101 zero = 1;
102 } else
ceffc078 103 goto out;
afa597ba 104 }
ceffc078
CO
105
106 /* If users can be writing to this page using arbitrary
107 * virtual addresses, take care about potential aliasing
108 * before reading the page on the kernel side.
109 */
110 if (mapping_writably_mapped(mapping))
70688e4d 111 /* address based flush */ ;
ceffc078
CO
112
113 /*
70688e4d 114 * Ok, we have the mem, so now we can copy it to user space...
ceffc078
CO
115 *
116 * The actor routine returns how many bytes were actually used..
117 * NOTE! This may not be the same as how much of a user buffer
118 * we filled up (we may be padding etc), so we can only update
119 * "pos" here (the actor routine has to update the user buffer
120 * pointers and the remaining count).
121 */
70688e4d
NP
122 if (!zero)
123 left = __copy_to_user(buf+copied, xip_mem+offset, nr);
124 else
125 left = __clear_user(buf + copied, nr);
ceffc078 126
70688e4d
NP
127 if (left) {
128 error = -EFAULT;
129 goto out;
130 }
ceffc078 131
70688e4d
NP
132 copied += (nr - left);
133 offset += (nr - left);
134 index += offset >> PAGE_CACHE_SHIFT;
135 offset &= ~PAGE_CACHE_MASK;
136 } while (copied < len);
ceffc078
CO
137
138out:
70688e4d 139 *ppos = pos + copied;
ceffc078
CO
140 if (filp)
141 file_accessed(filp);
70688e4d
NP
142
143 return (copied ? copied : error);
ceffc078
CO
144}
145
ceffc078 146ssize_t
eb6fe0c3 147xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
ceffc078 148{
eb6fe0c3
CO
149 if (!access_ok(VERIFY_WRITE, buf, len))
150 return -EFAULT;
ceffc078 151
70688e4d
NP
152 return do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp,
153 buf, len, ppos);
ceffc078 154}
eb6fe0c3 155EXPORT_SYMBOL_GPL(xip_file_read);
ceffc078 156
ceffc078
CO
157/*
158 * __xip_unmap is invoked from xip_unmap and
159 * xip_write
160 *
161 * This function walks all vmas of the address_space and unmaps the
a76c0b97 162 * __xip_sparse_page when found at pgoff.
ceffc078
CO
163 */
164static void
165__xip_unmap (struct address_space * mapping,
166 unsigned long pgoff)
167{
168 struct vm_area_struct *vma;
169 struct mm_struct *mm;
170 struct prio_tree_iter iter;
171 unsigned long address;
172 pte_t *pte;
173 pte_t pteval;
c0718806 174 spinlock_t *ptl;
67b02f11 175 struct page *page;
538f8ea6
NP
176 unsigned count;
177 int locked = 0;
178
179 count = read_seqcount_begin(&xip_sparse_seq);
ceffc078 180
a76c0b97
CO
181 page = __xip_sparse_page;
182 if (!page)
183 return;
184
538f8ea6 185retry:
ceffc078
CO
186 spin_lock(&mapping->i_mmap_lock);
187 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
188 mm = vma->vm_mm;
189 address = vma->vm_start +
190 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
191 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
479db0bf 192 pte = page_check_address(page, mm, address, &ptl, 1);
c0718806 193 if (pte) {
ceffc078 194 /* Nuke the page table entry. */
082ff0a9 195 flush_cache_page(vma, address, pte_pfn(*pte));
cddb8a5c 196 pteval = ptep_clear_flush_notify(vma, address, pte);
edc315fd 197 page_remove_rmap(page);
d559db08 198 dec_mm_counter(mm, MM_FILEPAGES);
ceffc078 199 BUG_ON(pte_dirty(pteval));
c0718806 200 pte_unmap_unlock(pte, ptl);
b5810039 201 page_cache_release(page);
ceffc078
CO
202 }
203 }
204 spin_unlock(&mapping->i_mmap_lock);
538f8ea6
NP
205
206 if (locked) {
207 mutex_unlock(&xip_sparse_mutex);
208 } else if (read_seqcount_retry(&xip_sparse_seq, count)) {
209 mutex_lock(&xip_sparse_mutex);
210 locked = 1;
211 goto retry;
212 }
ceffc078
CO
213}
214
215/*
54cb8821 216 * xip_fault() is invoked via the vma operations vector for a
ceffc078
CO
217 * mapped memory region to read in file data during a page fault.
218 *
54cb8821 219 * This function is derived from filemap_fault, but used for execute in place
ceffc078 220 */
70688e4d 221static int xip_file_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
ceffc078 222{
70688e4d 223 struct file *file = vma->vm_file;
ceffc078
CO
224 struct address_space *mapping = file->f_mapping;
225 struct inode *inode = mapping->host;
54cb8821 226 pgoff_t size;
70688e4d
NP
227 void *xip_mem;
228 unsigned long xip_pfn;
229 struct page *page;
230 int error;
ceffc078 231
54cb8821 232 /* XXX: are VM_FAULT_ codes OK? */
538f8ea6 233again:
ceffc078 234 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
d0217ac0
NP
235 if (vmf->pgoff >= size)
236 return VM_FAULT_SIGBUS;
ceffc078 237
70688e4d
NP
238 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 0,
239 &xip_mem, &xip_pfn);
240 if (likely(!error))
241 goto found;
242 if (error != -ENODATA)
d0217ac0 243 return VM_FAULT_OOM;
ceffc078
CO
244
245 /* sparse block */
70688e4d
NP
246 if ((vma->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
247 (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) &&
ceffc078 248 (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
70688e4d
NP
249 int err;
250
ceffc078 251 /* maybe shared writable, allocate new block */
14bac5ac 252 mutex_lock(&xip_sparse_mutex);
70688e4d
NP
253 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 1,
254 &xip_mem, &xip_pfn);
14bac5ac 255 mutex_unlock(&xip_sparse_mutex);
70688e4d 256 if (error)
d0217ac0 257 return VM_FAULT_SIGBUS;
70688e4d 258 /* unmap sparse mappings at pgoff from all other vmas */
d0217ac0 259 __xip_unmap(mapping, vmf->pgoff);
70688e4d
NP
260
261found:
262 err = vm_insert_mixed(vma, (unsigned long)vmf->virtual_address,
263 xip_pfn);
264 if (err == -ENOMEM)
265 return VM_FAULT_OOM;
266 BUG_ON(err);
267 return VM_FAULT_NOPAGE;
ceffc078 268 } else {
538f8ea6
NP
269 int err, ret = VM_FAULT_OOM;
270
271 mutex_lock(&xip_sparse_mutex);
272 write_seqcount_begin(&xip_sparse_seq);
273 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 0,
274 &xip_mem, &xip_pfn);
275 if (unlikely(!error)) {
276 write_seqcount_end(&xip_sparse_seq);
277 mutex_unlock(&xip_sparse_mutex);
278 goto again;
279 }
280 if (error != -ENODATA)
281 goto out;
a76c0b97
CO
282 /* not shared and writable, use xip_sparse_page() */
283 page = xip_sparse_page();
d0217ac0 284 if (!page)
538f8ea6
NP
285 goto out;
286 err = vm_insert_page(vma, (unsigned long)vmf->virtual_address,
287 page);
288 if (err == -ENOMEM)
289 goto out;
290
291 ret = VM_FAULT_NOPAGE;
292out:
293 write_seqcount_end(&xip_sparse_seq);
294 mutex_unlock(&xip_sparse_mutex);
ceffc078 295
538f8ea6 296 return ret;
70688e4d 297 }
ceffc078
CO
298}
299
f0f37e2f 300static const struct vm_operations_struct xip_file_vm_ops = {
54cb8821 301 .fault = xip_file_fault,
ceffc078
CO
302};
303
304int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
305{
70688e4d 306 BUG_ON(!file->f_mapping->a_ops->get_xip_mem);
ceffc078
CO
307
308 file_accessed(file);
309 vma->vm_ops = &xip_file_vm_ops;
70688e4d 310 vma->vm_flags |= VM_CAN_NONLINEAR | VM_MIXEDMAP;
ceffc078
CO
311 return 0;
312}
313EXPORT_SYMBOL_GPL(xip_file_mmap);
314
315static ssize_t
eb6fe0c3
CO
316__xip_file_write(struct file *filp, const char __user *buf,
317 size_t count, loff_t pos, loff_t *ppos)
ceffc078 318{
eb6fe0c3 319 struct address_space * mapping = filp->f_mapping;
f5e54d6e 320 const struct address_space_operations *a_ops = mapping->a_ops;
ceffc078
CO
321 struct inode *inode = mapping->host;
322 long status = 0;
ceffc078 323 size_t bytes;
ceffc078
CO
324 ssize_t written = 0;
325
70688e4d 326 BUG_ON(!mapping->a_ops->get_xip_mem);
ceffc078 327
ceffc078
CO
328 do {
329 unsigned long index;
330 unsigned long offset;
331 size_t copied;
70688e4d
NP
332 void *xip_mem;
333 unsigned long xip_pfn;
ceffc078
CO
334
335 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
336 index = pos >> PAGE_CACHE_SHIFT;
337 bytes = PAGE_CACHE_SIZE - offset;
338 if (bytes > count)
339 bytes = count;
340
70688e4d
NP
341 status = a_ops->get_xip_mem(mapping, index, 0,
342 &xip_mem, &xip_pfn);
343 if (status == -ENODATA) {
ceffc078 344 /* we allocate a new page unmap it */
14bac5ac 345 mutex_lock(&xip_sparse_mutex);
70688e4d
NP
346 status = a_ops->get_xip_mem(mapping, index, 1,
347 &xip_mem, &xip_pfn);
14bac5ac 348 mutex_unlock(&xip_sparse_mutex);
70688e4d 349 if (!status)
eb6fe0c3
CO
350 /* unmap page at pgoff from all other vmas */
351 __xip_unmap(mapping, index);
ceffc078
CO
352 }
353
70688e4d 354 if (status)
ceffc078 355 break;
ceffc078 356
4a9e5ef1 357 copied = bytes -
70688e4d 358 __copy_from_user_nocache(xip_mem + offset, buf, bytes);
4a9e5ef1 359
ceffc078
CO
360 if (likely(copied > 0)) {
361 status = copied;
362
363 if (status >= 0) {
364 written += status;
365 count -= status;
366 pos += status;
367 buf += status;
ceffc078
CO
368 }
369 }
370 if (unlikely(copied != bytes))
371 if (status >= 0)
372 status = -EFAULT;
373 if (status < 0)
374 break;
375 } while (count);
376 *ppos = pos;
377 /*
378 * No need to use i_size_read() here, the i_size
1b1dcc1b 379 * cannot change under us because we hold i_mutex.
ceffc078
CO
380 */
381 if (pos > inode->i_size) {
382 i_size_write(inode, pos);
383 mark_inode_dirty(inode);
384 }
385
386 return written ? written : status;
387}
388
eb6fe0c3
CO
389ssize_t
390xip_file_write(struct file *filp, const char __user *buf, size_t len,
391 loff_t *ppos)
ceffc078 392{
eb6fe0c3
CO
393 struct address_space *mapping = filp->f_mapping;
394 struct inode *inode = mapping->host;
395 size_t count;
396 loff_t pos;
397 ssize_t ret;
ceffc078 398
1b1dcc1b 399 mutex_lock(&inode->i_mutex);
ceffc078 400
eb6fe0c3
CO
401 if (!access_ok(VERIFY_READ, buf, len)) {
402 ret=-EFAULT;
403 goto out_up;
ceffc078
CO
404 }
405
ceffc078 406 pos = *ppos;
eb6fe0c3 407 count = len;
ceffc078
CO
408
409 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
410
eb6fe0c3
CO
411 /* We can write back this queue in page reclaim */
412 current->backing_dev_info = mapping->backing_dev_info;
ceffc078 413
eb6fe0c3
CO
414 ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
415 if (ret)
416 goto out_backing;
ceffc078 417 if (count == 0)
eb6fe0c3 418 goto out_backing;
ceffc078 419
2f1936b8 420 ret = file_remove_suid(filp);
eb6fe0c3
CO
421 if (ret)
422 goto out_backing;
ceffc078 423
870f4817 424 file_update_time(filp);
ceffc078 425
eb6fe0c3 426 ret = __xip_file_write (filp, buf, count, pos, ppos);
ceffc078 427
eb6fe0c3
CO
428 out_backing:
429 current->backing_dev_info = NULL;
430 out_up:
1b1dcc1b 431 mutex_unlock(&inode->i_mutex);
ceffc078
CO
432 return ret;
433}
eb6fe0c3 434EXPORT_SYMBOL_GPL(xip_file_write);
ceffc078
CO
435
436/*
437 * truncate a page used for execute in place
70688e4d 438 * functionality is analog to block_truncate_page but does use get_xip_mem
ceffc078
CO
439 * to get the page instead of page cache
440 */
441int
442xip_truncate_page(struct address_space *mapping, loff_t from)
443{
444 pgoff_t index = from >> PAGE_CACHE_SHIFT;
445 unsigned offset = from & (PAGE_CACHE_SIZE-1);
446 unsigned blocksize;
447 unsigned length;
70688e4d
NP
448 void *xip_mem;
449 unsigned long xip_pfn;
450 int err;
ceffc078 451
70688e4d 452 BUG_ON(!mapping->a_ops->get_xip_mem);
ceffc078
CO
453
454 blocksize = 1 << mapping->host->i_blkbits;
455 length = offset & (blocksize - 1);
456
457 /* Block boundary? Nothing to do */
458 if (!length)
459 return 0;
460
461 length = blocksize - length;
462
70688e4d
NP
463 err = mapping->a_ops->get_xip_mem(mapping, index, 0,
464 &xip_mem, &xip_pfn);
465 if (unlikely(err)) {
466 if (err == -ENODATA)
ceffc078
CO
467 /* Hole? No need to truncate */
468 return 0;
eb6fe0c3 469 else
70688e4d 470 return err;
afa597ba 471 }
70688e4d 472 memset(xip_mem + offset, 0, length);
eb6fe0c3 473 return 0;
ceffc078
CO
474}
475EXPORT_SYMBOL_GPL(xip_truncate_page);