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 * 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
24 * Note the struct file* is not used at all. It may be NULL.
27 do_xip_mapping_read(struct address_space *mapping,
28 struct file_ra_state *_ra,
31 read_descriptor_t *desc,
34 struct inode *inode = mapping->host;
35 unsigned long index, end_index, offset;
38 BUG_ON(!mapping->a_ops->get_xip_page);
40 index = *ppos >> PAGE_CACHE_SHIFT;
41 offset = *ppos & ~PAGE_CACHE_MASK;
43 isize = i_size_read(inode);
47 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
50 unsigned long nr, ret;
52 /* nr is the maximum number of bytes to copy from this page */
54 if (index >= end_index) {
55 if (index > end_index)
57 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
64 page = mapping->a_ops->get_xip_page(mapping,
65 index*(PAGE_SIZE/512), 0);
68 if (unlikely(IS_ERR(page))) {
69 if (PTR_ERR(page) == -ENODATA) {
73 desc->error = PTR_ERR(page);
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.
82 if (mapping_writably_mapped(mapping))
83 flush_dcache_page(page);
86 * Ok, we have the page, so now we can copy it to user space...
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).
94 ret = actor(desc, page, offset, nr);
96 index += offset >> PAGE_CACHE_SHIFT;
97 offset &= ~PAGE_CACHE_MASK;
99 if (ret == nr && desc->count)
104 /* Did not get the page. Report it */
110 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
116 xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
118 read_descriptor_t desc;
120 if (!access_ok(VERIFY_WRITE, buf, len))
128 do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp,
129 ppos, &desc, file_read_actor);
136 EXPORT_SYMBOL_GPL(xip_file_read);
139 xip_file_sendfile(struct file *in_file, loff_t *ppos,
140 size_t count, read_actor_t actor, void *target)
142 read_descriptor_t desc;
149 desc.arg.data = target;
152 do_xip_mapping_read(in_file->f_mapping, &in_file->f_ra, in_file,
158 EXPORT_SYMBOL_GPL(xip_file_sendfile);
161 * __xip_unmap is invoked from xip_unmap and
164 * This function walks all vmas of the address_space and unmaps the
165 * ZERO_PAGE when found at pgoff. Should it go in rmap.c?
168 __xip_unmap (struct address_space * mapping,
171 struct vm_area_struct *vma;
172 struct mm_struct *mm;
173 struct prio_tree_iter iter;
174 unsigned long address;
180 spin_lock(&mapping->i_mmap_lock);
181 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
183 address = vma->vm_start +
184 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
185 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
187 pte = page_check_address(page, mm, address, &ptl);
189 /* Nuke the page table entry. */
190 flush_cache_page(vma, address, pte_pfn(*pte));
191 pteval = ptep_clear_flush(vma, address, pte);
192 page_remove_rmap(page, vma);
193 dec_mm_counter(mm, file_rss);
194 BUG_ON(pte_dirty(pteval));
195 pte_unmap_unlock(pte, ptl);
196 page_cache_release(page);
199 spin_unlock(&mapping->i_mmap_lock);
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.
206 * This function is derived from filemap_nopage, but used for execute in place
209 xip_file_nopage(struct vm_area_struct * area,
210 unsigned long address,
213 struct file *file = area->vm_file;
214 struct address_space *mapping = file->f_mapping;
215 struct inode *inode = mapping->host;
217 unsigned long size, pgoff, endoff;
219 pgoff = ((address - area->vm_start) >> PAGE_CACHE_SHIFT)
221 endoff = ((area->vm_end - area->vm_start) >> PAGE_CACHE_SHIFT)
224 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
229 page = mapping->a_ops->get_xip_page(mapping, pgoff*(PAGE_SIZE/512), 0);
233 if (PTR_ERR(page) != -ENODATA)
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);
245 /* unmap page at pgoff from all other vmas */
246 __xip_unmap(mapping, pgoff);
248 /* not shared and writable, use ZERO_PAGE() */
253 page_cache_get(page);
257 static struct vm_operations_struct xip_file_vm_ops = {
258 .nopage = xip_file_nopage,
261 int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
263 BUG_ON(!file->f_mapping->a_ops->get_xip_page);
266 vma->vm_ops = &xip_file_vm_ops;
269 EXPORT_SYMBOL_GPL(xip_file_mmap);
272 __xip_file_write(struct file *filp, const char __user *buf,
273 size_t count, loff_t pos, loff_t *ppos)
275 struct address_space * mapping = filp->f_mapping;
276 const struct address_space_operations *a_ops = mapping->a_ops;
277 struct inode *inode = mapping->host;
283 BUG_ON(!mapping->a_ops->get_xip_page);
287 unsigned long offset;
290 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
291 index = pos >> PAGE_CACHE_SHIFT;
292 bytes = PAGE_CACHE_SIZE - offset;
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
302 fault_in_pages_readable(buf, bytes);
304 page = a_ops->get_xip_page(mapping,
305 index*(PAGE_SIZE/512), 0);
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,
309 index*(PAGE_SIZE/512), 1);
311 /* unmap page at pgoff from all other vmas */
312 __xip_unmap(mapping, index);
316 status = PTR_ERR(page);
320 copied = filemap_copy_from_user(page, offset, buf, bytes);
321 flush_dcache_page(page);
322 if (likely(copied > 0)) {
332 if (unlikely(copied != bytes))
340 * No need to use i_size_read() here, the i_size
341 * cannot change under us because we hold i_mutex.
343 if (pos > inode->i_size) {
344 i_size_write(inode, pos);
345 mark_inode_dirty(inode);
348 return written ? written : status;
352 xip_file_write(struct file *filp, const char __user *buf, size_t len,
355 struct address_space *mapping = filp->f_mapping;
356 struct inode *inode = mapping->host;
361 mutex_lock(&inode->i_mutex);
363 if (!access_ok(VERIFY_READ, buf, len)) {
371 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
373 /* We can write back this queue in page reclaim */
374 current->backing_dev_info = mapping->backing_dev_info;
376 ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
382 ret = remove_suid(filp->f_path.dentry);
386 file_update_time(filp);
388 ret = __xip_file_write (filp, buf, count, pos, ppos);
391 current->backing_dev_info = NULL;
393 mutex_unlock(&inode->i_mutex);
396 EXPORT_SYMBOL_GPL(xip_file_write);
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
404 xip_truncate_page(struct address_space *mapping, loff_t from)
406 pgoff_t index = from >> PAGE_CACHE_SHIFT;
407 unsigned offset = from & (PAGE_CACHE_SIZE-1);
413 BUG_ON(!mapping->a_ops->get_xip_page);
415 blocksize = 1 << mapping->host->i_blkbits;
416 length = offset & (blocksize - 1);
418 /* Block boundary? Nothing to do */
422 length = blocksize - length;
424 page = mapping->a_ops->get_xip_page(mapping,
425 index*(PAGE_SIZE/512), 0);
428 if (unlikely(IS_ERR(page))) {
429 if (PTR_ERR(page) == -ENODATA)
430 /* Hole? No need to truncate */
433 return PTR_ERR(page);
435 kaddr = kmap_atomic(page, KM_USER0);
436 memset(kaddr + offset, 0, length);
437 kunmap_atomic(kaddr, KM_USER0);
439 flush_dcache_page(page);
442 EXPORT_SYMBOL_GPL(xip_truncate_page);