2 * linux/mm/filemap_xip.c
4 * Copyright (C) 2005 IBM Corporation
5 * Author: Carsten Otte <cotte@de.ibm.com>
7 * derived from linux/mm/filemap.c - Copyright (C) Linus Torvalds
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>
20 * We do use our own empty page to avoid interference with other users
21 * of ZERO_PAGE(), such as /dev/zero
23 static struct page *__xip_sparse_page;
25 static struct page *xip_sparse_page(void)
27 if (!__xip_sparse_page) {
28 unsigned long zeroes = get_zeroed_page(GFP_HIGHUSER);
30 static DEFINE_SPINLOCK(xip_alloc_lock);
31 spin_lock(&xip_alloc_lock);
32 if (!__xip_sparse_page)
33 __xip_sparse_page = virt_to_page(zeroes);
36 spin_unlock(&xip_alloc_lock);
39 return __xip_sparse_page;
43 * This is a file read routine for execute in place files, and uses
44 * the mapping->a_ops->get_xip_page() function for the actual low-level
47 * Note the struct file* is not used at all. It may be NULL.
50 do_xip_mapping_read(struct address_space *mapping,
51 struct file_ra_state *_ra,
54 read_descriptor_t *desc,
57 struct inode *inode = mapping->host;
58 unsigned long index, end_index, offset;
61 BUG_ON(!mapping->a_ops->get_xip_page);
63 index = *ppos >> PAGE_CACHE_SHIFT;
64 offset = *ppos & ~PAGE_CACHE_MASK;
66 isize = i_size_read(inode);
70 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
73 unsigned long nr, ret;
75 /* nr is the maximum number of bytes to copy from this page */
77 if (index >= end_index) {
78 if (index > end_index)
80 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
87 page = mapping->a_ops->get_xip_page(mapping,
88 index*(PAGE_SIZE/512), 0);
91 if (unlikely(IS_ERR(page))) {
92 if (PTR_ERR(page) == -ENODATA) {
96 desc->error = PTR_ERR(page);
101 /* If users can be writing to this page using arbitrary
102 * virtual addresses, take care about potential aliasing
103 * before reading the page on the kernel side.
105 if (mapping_writably_mapped(mapping))
106 flush_dcache_page(page);
109 * Ok, we have the page, so now we can copy it to user space...
111 * The actor routine returns how many bytes were actually used..
112 * NOTE! This may not be the same as how much of a user buffer
113 * we filled up (we may be padding etc), so we can only update
114 * "pos" here (the actor routine has to update the user buffer
115 * pointers and the remaining count).
117 ret = actor(desc, page, offset, nr);
119 index += offset >> PAGE_CACHE_SHIFT;
120 offset &= ~PAGE_CACHE_MASK;
122 if (ret == nr && desc->count)
127 /* Did not get the page. Report it */
133 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
139 xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
141 read_descriptor_t desc;
143 if (!access_ok(VERIFY_WRITE, buf, len))
151 do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp,
152 ppos, &desc, file_read_actor);
159 EXPORT_SYMBOL_GPL(xip_file_read);
162 xip_file_sendfile(struct file *in_file, loff_t *ppos,
163 size_t count, read_actor_t actor, void *target)
165 read_descriptor_t desc;
172 desc.arg.data = target;
175 do_xip_mapping_read(in_file->f_mapping, &in_file->f_ra, in_file,
181 EXPORT_SYMBOL_GPL(xip_file_sendfile);
184 * __xip_unmap is invoked from xip_unmap and
187 * This function walks all vmas of the address_space and unmaps the
188 * __xip_sparse_page when found at pgoff.
191 __xip_unmap (struct address_space * mapping,
194 struct vm_area_struct *vma;
195 struct mm_struct *mm;
196 struct prio_tree_iter iter;
197 unsigned long address;
203 page = __xip_sparse_page;
207 spin_lock(&mapping->i_mmap_lock);
208 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
210 address = vma->vm_start +
211 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
212 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
213 pte = page_check_address(page, mm, address, &ptl);
215 /* Nuke the page table entry. */
216 flush_cache_page(vma, address, pte_pfn(*pte));
217 pteval = ptep_clear_flush(vma, address, pte);
218 page_remove_rmap(page, vma);
219 dec_mm_counter(mm, file_rss);
220 BUG_ON(pte_dirty(pteval));
221 pte_unmap_unlock(pte, ptl);
222 page_cache_release(page);
225 spin_unlock(&mapping->i_mmap_lock);
229 * xip_nopage() is invoked via the vma operations vector for a
230 * mapped memory region to read in file data during a page fault.
232 * This function is derived from filemap_nopage, but used for execute in place
235 xip_file_nopage(struct vm_area_struct * area,
236 unsigned long address,
239 struct file *file = area->vm_file;
240 struct address_space *mapping = file->f_mapping;
241 struct inode *inode = mapping->host;
243 unsigned long size, pgoff, endoff;
245 pgoff = ((address - area->vm_start) >> PAGE_CACHE_SHIFT)
247 endoff = ((area->vm_end - area->vm_start) >> PAGE_CACHE_SHIFT)
250 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
252 return NOPAGE_SIGBUS;
254 page = mapping->a_ops->get_xip_page(mapping, pgoff*(PAGE_SIZE/512), 0);
257 if (PTR_ERR(page) != -ENODATA)
258 return NOPAGE_SIGBUS;
261 if ((area->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
262 (area->vm_flags & (VM_SHARED| VM_MAYSHARE)) &&
263 (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
264 /* maybe shared writable, allocate new block */
265 page = mapping->a_ops->get_xip_page (mapping,
266 pgoff*(PAGE_SIZE/512), 1);
268 return NOPAGE_SIGBUS;
269 /* unmap page at pgoff from all other vmas */
270 __xip_unmap(mapping, pgoff);
272 /* not shared and writable, use xip_sparse_page() */
273 page = xip_sparse_page();
279 page_cache_get(page);
283 static struct vm_operations_struct xip_file_vm_ops = {
284 .nopage = xip_file_nopage,
287 int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
289 BUG_ON(!file->f_mapping->a_ops->get_xip_page);
292 vma->vm_ops = &xip_file_vm_ops;
295 EXPORT_SYMBOL_GPL(xip_file_mmap);
298 __xip_file_write(struct file *filp, const char __user *buf,
299 size_t count, loff_t pos, loff_t *ppos)
301 struct address_space * mapping = filp->f_mapping;
302 const struct address_space_operations *a_ops = mapping->a_ops;
303 struct inode *inode = mapping->host;
309 BUG_ON(!mapping->a_ops->get_xip_page);
313 unsigned long offset;
316 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
317 index = pos >> PAGE_CACHE_SHIFT;
318 bytes = PAGE_CACHE_SIZE - offset;
323 * Bring in the user page that we will copy from _first_.
324 * Otherwise there's a nasty deadlock on copying from the
325 * same page as we're writing to, without it being marked
328 fault_in_pages_readable(buf, bytes);
330 page = a_ops->get_xip_page(mapping,
331 index*(PAGE_SIZE/512), 0);
332 if (IS_ERR(page) && (PTR_ERR(page) == -ENODATA)) {
333 /* we allocate a new page unmap it */
334 page = a_ops->get_xip_page(mapping,
335 index*(PAGE_SIZE/512), 1);
337 /* unmap page at pgoff from all other vmas */
338 __xip_unmap(mapping, index);
342 status = PTR_ERR(page);
346 copied = filemap_copy_from_user(page, offset, buf, bytes);
347 flush_dcache_page(page);
348 if (likely(copied > 0)) {
358 if (unlikely(copied != bytes))
366 * No need to use i_size_read() here, the i_size
367 * cannot change under us because we hold i_mutex.
369 if (pos > inode->i_size) {
370 i_size_write(inode, pos);
371 mark_inode_dirty(inode);
374 return written ? written : status;
378 xip_file_write(struct file *filp, const char __user *buf, size_t len,
381 struct address_space *mapping = filp->f_mapping;
382 struct inode *inode = mapping->host;
387 mutex_lock(&inode->i_mutex);
389 if (!access_ok(VERIFY_READ, buf, len)) {
397 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
399 /* We can write back this queue in page reclaim */
400 current->backing_dev_info = mapping->backing_dev_info;
402 ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
408 ret = remove_suid(filp->f_path.dentry);
412 file_update_time(filp);
414 ret = __xip_file_write (filp, buf, count, pos, ppos);
417 current->backing_dev_info = NULL;
419 mutex_unlock(&inode->i_mutex);
422 EXPORT_SYMBOL_GPL(xip_file_write);
425 * truncate a page used for execute in place
426 * functionality is analog to block_truncate_page but does use get_xip_page
427 * to get the page instead of page cache
430 xip_truncate_page(struct address_space *mapping, loff_t from)
432 pgoff_t index = from >> PAGE_CACHE_SHIFT;
433 unsigned offset = from & (PAGE_CACHE_SIZE-1);
439 BUG_ON(!mapping->a_ops->get_xip_page);
441 blocksize = 1 << mapping->host->i_blkbits;
442 length = offset & (blocksize - 1);
444 /* Block boundary? Nothing to do */
448 length = blocksize - length;
450 page = mapping->a_ops->get_xip_page(mapping,
451 index*(PAGE_SIZE/512), 0);
454 if (unlikely(IS_ERR(page))) {
455 if (PTR_ERR(page) == -ENODATA)
456 /* Hole? No need to truncate */
459 return PTR_ERR(page);
461 kaddr = kmap_atomic(page, KM_USER0);
462 memset(kaddr + offset, 0, length);
463 kunmap_atomic(kaddr, KM_USER0);
465 flush_dcache_page(page);
468 EXPORT_SYMBOL_GPL(xip_truncate_page);