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 <linux/mmu_notifier.h>
17 #include <linux/sched.h>
18 #include <linux/seqlock.h>
19 #include <linux/mutex.h>
20 #include <asm/tlbflush.h>
24 * We do use our own empty page to avoid interference with other users
25 * of ZERO_PAGE(), such as /dev/zero
27 static DEFINE_MUTEX(xip_sparse_mutex);
28 static seqcount_t xip_sparse_seq = SEQCNT_ZERO;
29 static struct page *__xip_sparse_page;
31 /* called under xip_sparse_mutex */
32 static struct page *xip_sparse_page(void)
34 if (!__xip_sparse_page) {
35 struct page *page = alloc_page(GFP_HIGHUSER | __GFP_ZERO);
38 __xip_sparse_page = page;
40 return __xip_sparse_page;
44 * This is a file read routine for execute in place files, and uses
45 * the mapping->a_ops->get_xip_mem() function for the actual low-level
48 * Note the struct file* is not used at all. It may be NULL.
51 do_xip_mapping_read(struct address_space *mapping,
52 struct file_ra_state *_ra,
58 struct inode *inode = mapping->host;
59 pgoff_t index, end_index;
62 size_t copied = 0, error = 0;
64 BUG_ON(!mapping->a_ops->get_xip_mem);
67 index = pos >> PAGE_CACHE_SHIFT;
68 offset = pos & ~PAGE_CACHE_MASK;
70 isize = i_size_read(inode);
74 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
76 unsigned long nr, left;
78 unsigned long xip_pfn;
81 /* nr is the maximum number of bytes to copy from this page */
83 if (index >= end_index) {
84 if (index > end_index)
86 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
95 error = mapping->a_ops->get_xip_mem(mapping, index, 0,
97 if (unlikely(error)) {
98 if (error == -ENODATA) {
105 /* If users can be writing to this page using arbitrary
106 * virtual addresses, take care about potential aliasing
107 * before reading the page on the kernel side.
109 if (mapping_writably_mapped(mapping))
110 /* address based flush */ ;
113 * Ok, we have the mem, so now we can copy it to user space...
115 * The actor routine returns how many bytes were actually used..
116 * NOTE! This may not be the same as how much of a user buffer
117 * we filled up (we may be padding etc), so we can only update
118 * "pos" here (the actor routine has to update the user buffer
119 * pointers and the remaining count).
122 left = __copy_to_user(buf+copied, xip_mem+offset, nr);
124 left = __clear_user(buf + copied, nr);
131 copied += (nr - left);
132 offset += (nr - left);
133 index += offset >> PAGE_CACHE_SHIFT;
134 offset &= ~PAGE_CACHE_MASK;
135 } while (copied < len);
138 *ppos = pos + copied;
142 return (copied ? copied : error);
146 xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
148 if (!access_ok(VERIFY_WRITE, buf, len))
151 return do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp,
154 EXPORT_SYMBOL_GPL(xip_file_read);
157 * __xip_unmap is invoked from xip_unmap and
160 * This function walks all vmas of the address_space and unmaps the
161 * __xip_sparse_page when found at pgoff.
164 __xip_unmap (struct address_space * mapping,
167 struct vm_area_struct *vma;
168 struct mm_struct *mm;
169 struct prio_tree_iter iter;
170 unsigned long address;
178 count = read_seqcount_begin(&xip_sparse_seq);
180 page = __xip_sparse_page;
185 spin_lock(&mapping->i_mmap_lock);
186 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
188 address = vma->vm_start +
189 ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
190 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
191 pte = page_check_address(page, mm, address, &ptl, 1);
193 /* Nuke the page table entry. */
194 flush_cache_page(vma, address, pte_pfn(*pte));
195 pteval = ptep_clear_flush_notify(vma, address, pte);
196 page_remove_rmap(page, vma);
197 dec_mm_counter(mm, file_rss);
198 BUG_ON(pte_dirty(pteval));
199 pte_unmap_unlock(pte, ptl);
200 page_cache_release(page);
203 spin_unlock(&mapping->i_mmap_lock);
206 mutex_unlock(&xip_sparse_mutex);
207 } else if (read_seqcount_retry(&xip_sparse_seq, count)) {
208 mutex_lock(&xip_sparse_mutex);
215 * xip_fault() is invoked via the vma operations vector for a
216 * mapped memory region to read in file data during a page fault.
218 * This function is derived from filemap_fault, but used for execute in place
220 static int xip_file_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
222 struct file *file = vma->vm_file;
223 struct address_space *mapping = file->f_mapping;
224 struct inode *inode = mapping->host;
227 unsigned long xip_pfn;
231 /* XXX: are VM_FAULT_ codes OK? */
233 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
234 if (vmf->pgoff >= size)
235 return VM_FAULT_SIGBUS;
237 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 0,
241 if (error != -ENODATA)
245 if ((vma->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
246 (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) &&
247 (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
250 /* maybe shared writable, allocate new block */
251 mutex_lock(&xip_sparse_mutex);
252 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 1,
254 mutex_unlock(&xip_sparse_mutex);
256 return VM_FAULT_SIGBUS;
257 /* unmap sparse mappings at pgoff from all other vmas */
258 __xip_unmap(mapping, vmf->pgoff);
261 err = vm_insert_mixed(vma, (unsigned long)vmf->virtual_address,
266 return VM_FAULT_NOPAGE;
268 int err, ret = VM_FAULT_OOM;
270 mutex_lock(&xip_sparse_mutex);
271 write_seqcount_begin(&xip_sparse_seq);
272 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 0,
274 if (unlikely(!error)) {
275 write_seqcount_end(&xip_sparse_seq);
276 mutex_unlock(&xip_sparse_mutex);
279 if (error != -ENODATA)
281 /* not shared and writable, use xip_sparse_page() */
282 page = xip_sparse_page();
285 err = vm_insert_page(vma, (unsigned long)vmf->virtual_address,
290 ret = VM_FAULT_NOPAGE;
292 write_seqcount_end(&xip_sparse_seq);
293 mutex_unlock(&xip_sparse_mutex);
299 static struct vm_operations_struct xip_file_vm_ops = {
300 .fault = xip_file_fault,
303 int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
305 BUG_ON(!file->f_mapping->a_ops->get_xip_mem);
308 vma->vm_ops = &xip_file_vm_ops;
309 vma->vm_flags |= VM_CAN_NONLINEAR | VM_MIXEDMAP;
312 EXPORT_SYMBOL_GPL(xip_file_mmap);
315 __xip_file_write(struct file *filp, const char __user *buf,
316 size_t count, loff_t pos, loff_t *ppos)
318 struct address_space * mapping = filp->f_mapping;
319 const struct address_space_operations *a_ops = mapping->a_ops;
320 struct inode *inode = mapping->host;
325 BUG_ON(!mapping->a_ops->get_xip_mem);
329 unsigned long offset;
332 unsigned long xip_pfn;
334 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
335 index = pos >> PAGE_CACHE_SHIFT;
336 bytes = PAGE_CACHE_SIZE - offset;
340 status = a_ops->get_xip_mem(mapping, index, 0,
342 if (status == -ENODATA) {
343 /* we allocate a new page unmap it */
344 mutex_lock(&xip_sparse_mutex);
345 status = a_ops->get_xip_mem(mapping, index, 1,
347 mutex_unlock(&xip_sparse_mutex);
349 /* unmap page at pgoff from all other vmas */
350 __xip_unmap(mapping, index);
357 __copy_from_user_nocache(xip_mem + offset, buf, bytes);
359 if (likely(copied > 0)) {
369 if (unlikely(copied != bytes))
377 * No need to use i_size_read() here, the i_size
378 * cannot change under us because we hold i_mutex.
380 if (pos > inode->i_size) {
381 i_size_write(inode, pos);
382 mark_inode_dirty(inode);
385 return written ? written : status;
389 xip_file_write(struct file *filp, const char __user *buf, size_t len,
392 struct address_space *mapping = filp->f_mapping;
393 struct inode *inode = mapping->host;
398 mutex_lock(&inode->i_mutex);
400 if (!access_ok(VERIFY_READ, buf, len)) {
408 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
410 /* We can write back this queue in page reclaim */
411 current->backing_dev_info = mapping->backing_dev_info;
413 ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
419 ret = file_remove_suid(filp);
423 file_update_time(filp);
425 ret = __xip_file_write (filp, buf, count, pos, ppos);
428 current->backing_dev_info = NULL;
430 mutex_unlock(&inode->i_mutex);
433 EXPORT_SYMBOL_GPL(xip_file_write);
436 * truncate a page used for execute in place
437 * functionality is analog to block_truncate_page but does use get_xip_mem
438 * to get the page instead of page cache
441 xip_truncate_page(struct address_space *mapping, loff_t from)
443 pgoff_t index = from >> PAGE_CACHE_SHIFT;
444 unsigned offset = from & (PAGE_CACHE_SIZE-1);
448 unsigned long xip_pfn;
451 BUG_ON(!mapping->a_ops->get_xip_mem);
453 blocksize = 1 << mapping->host->i_blkbits;
454 length = offset & (blocksize - 1);
456 /* Block boundary? Nothing to do */
460 length = blocksize - length;
462 err = mapping->a_ops->get_xip_mem(mapping, index, 0,
466 /* Hole? No need to truncate */
471 memset(xip_mem + offset, 0, length);
474 EXPORT_SYMBOL_GPL(xip_truncate_page);