filemap_xip.c revision 83c54070ee1a2d05c89793884bea1a03f2851ed4 
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 <linux/sched.h>
17#include <asm/tlbflush.h>
18#include "filemap.h"
19
20/*
21 * We do use our own empty page to avoid interference with other users
22 * of ZERO_PAGE(), such as /dev/zero
23 */
24static struct page *__xip_sparse_page;
25
26static struct page *xip_sparse_page(void)
27{
28	if (!__xip_sparse_page) {
29		unsigned long zeroes = get_zeroed_page(GFP_HIGHUSER);
30		if (zeroes) {
31			static DEFINE_SPINLOCK(xip_alloc_lock);
32			spin_lock(&xip_alloc_lock);
33			if (!__xip_sparse_page)
34				__xip_sparse_page = virt_to_page(zeroes);
35			else
36				free_page(zeroes);
37			spin_unlock(&xip_alloc_lock);
38		}
39	}
40	return __xip_sparse_page;
41}
42
43/*
44 * This is a file read routine for execute in place files, and uses
45 * the mapping->a_ops->get_xip_page() function for the actual low-level
46 * stuff.
47 *
48 * Note the struct file* is not used at all.  It may be NULL.
49 */
50static void
51do_xip_mapping_read(struct address_space *mapping,
52		    struct file_ra_state *_ra,
53		    struct file *filp,
54		    loff_t *ppos,
55		    read_descriptor_t *desc,
56		    read_actor_t actor)
57{
58	struct inode *inode = mapping->host;
59	unsigned long index, end_index, offset;
60	loff_t isize;
61
62	BUG_ON(!mapping->a_ops->get_xip_page);
63
64	index = *ppos >> PAGE_CACHE_SHIFT;
65	offset = *ppos & ~PAGE_CACHE_MASK;
66
67	isize = i_size_read(inode);
68	if (!isize)
69		goto out;
70
71	end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
72	for (;;) {
73		struct page *page;
74		unsigned long nr, ret;
75
76		/* nr is the maximum number of bytes to copy from this page */
77		nr = PAGE_CACHE_SIZE;
78		if (index >= end_index) {
79			if (index > end_index)
80				goto out;
81			nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
82			if (nr <= offset) {
83				goto out;
84			}
85		}
86		nr = nr - offset;
87
88		page = mapping->a_ops->get_xip_page(mapping,
89			index*(PAGE_SIZE/512), 0);
90		if (!page)
91			goto no_xip_page;
92		if (unlikely(IS_ERR(page))) {
93			if (PTR_ERR(page) == -ENODATA) {
94				/* sparse */
95				page = ZERO_PAGE(0);
96			} else {
97				desc->error = PTR_ERR(page);
98				goto out;
99			}
100		}
101
102		/* If users can be writing to this page using arbitrary
103		 * virtual addresses, take care about potential aliasing
104		 * before reading the page on the kernel side.
105		 */
106		if (mapping_writably_mapped(mapping))
107			flush_dcache_page(page);
108
109		/*
110		 * Ok, we have the page, so now we can copy it to user space...
111		 *
112		 * The actor routine returns how many bytes were actually used..
113		 * NOTE! This may not be the same as how much of a user buffer
114		 * we filled up (we may be padding etc), so we can only update
115		 * "pos" here (the actor routine has to update the user buffer
116		 * pointers and the remaining count).
117		 */
118		ret = actor(desc, page, offset, nr);
119		offset += ret;
120		index += offset >> PAGE_CACHE_SHIFT;
121		offset &= ~PAGE_CACHE_MASK;
122
123		if (ret == nr && desc->count)
124			continue;
125		goto out;
126
127no_xip_page:
128		/* Did not get the page. Report it */
129		desc->error = -EIO;
130		goto out;
131	}
132
133out:
134	*ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
135	if (filp)
136		file_accessed(filp);
137}
138
139ssize_t
140xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
141{
142	read_descriptor_t desc;
143
144	if (!access_ok(VERIFY_WRITE, buf, len))
145		return -EFAULT;
146
147	desc.written = 0;
148	desc.arg.buf = buf;
149	desc.count = len;
150	desc.error = 0;
151
152	do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp,
153			    ppos, &desc, file_read_actor);
154
155	if (desc.written)
156		return desc.written;
157	else
158		return desc.error;
159}
160EXPORT_SYMBOL_GPL(xip_file_read);
161
162/*
163 * __xip_unmap is invoked from xip_unmap and
164 * xip_write
165 *
166 * This function walks all vmas of the address_space and unmaps the
167 * __xip_sparse_page when found at pgoff.
168 */
169static void
170__xip_unmap (struct address_space * mapping,
171		     unsigned long pgoff)
172{
173	struct vm_area_struct *vma;
174	struct mm_struct *mm;
175	struct prio_tree_iter iter;
176	unsigned long address;
177	pte_t *pte;
178	pte_t pteval;
179	spinlock_t *ptl;
180	struct page *page;
181
182	page = __xip_sparse_page;
183	if (!page)
184		return;
185
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);
192		pte = page_check_address(page, mm, address, &ptl);
193		if (pte) {
194			/* Nuke the page table entry. */
195			flush_cache_page(vma, address, pte_pfn(*pte));
196			pteval = ptep_clear_flush(vma, address, pte);
197			page_remove_rmap(page, vma);
198			dec_mm_counter(mm, file_rss);
199			BUG_ON(pte_dirty(pteval));
200			pte_unmap_unlock(pte, ptl);
201			page_cache_release(page);
202		}
203	}
204	spin_unlock(&mapping->i_mmap_lock);
205}
206
207/*
208 * xip_fault() is invoked via the vma operations vector for a
209 * mapped memory region to read in file data during a page fault.
210 *
211 * This function is derived from filemap_fault, but used for execute in place
212 */
213static int xip_file_fault(struct vm_area_struct *area, struct vm_fault *vmf)
214{
215	struct file *file = area->vm_file;
216	struct address_space *mapping = file->f_mapping;
217	struct inode *inode = mapping->host;
218	struct page *page;
219	pgoff_t size;
220
221	/* XXX: are VM_FAULT_ codes OK? */
222
223	size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
224	if (vmf->pgoff >= size)
225		return VM_FAULT_SIGBUS;
226
227	page = mapping->a_ops->get_xip_page(mapping,
228					vmf->pgoff*(PAGE_SIZE/512), 0);
229	if (!IS_ERR(page))
230		goto out;
231	if (PTR_ERR(page) != -ENODATA)
232		return VM_FAULT_OOM;
233
234	/* sparse block */
235	if ((area->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
236	    (area->vm_flags & (VM_SHARED| VM_MAYSHARE)) &&
237	    (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
238		/* maybe shared writable, allocate new block */
239		page = mapping->a_ops->get_xip_page(mapping,
240					vmf->pgoff*(PAGE_SIZE/512), 1);
241		if (IS_ERR(page))
242			return VM_FAULT_SIGBUS;
243		/* unmap page at pgoff from all other vmas */
244		__xip_unmap(mapping, vmf->pgoff);
245	} else {
246		/* not shared and writable, use xip_sparse_page() */
247		page = xip_sparse_page();
248		if (!page)
249			return VM_FAULT_OOM;
250	}
251
252out:
253	page_cache_get(page);
254	vmf->page = page;
255	return 0;
256}
257
258static struct vm_operations_struct xip_file_vm_ops = {
259	.fault	= xip_file_fault,
260};
261
262int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
263{
264	BUG_ON(!file->f_mapping->a_ops->get_xip_page);
265
266	file_accessed(file);
267	vma->vm_ops = &xip_file_vm_ops;
268	vma->vm_flags |= VM_CAN_NONLINEAR;
269	return 0;
270}
271EXPORT_SYMBOL_GPL(xip_file_mmap);
272
273static ssize_t
274__xip_file_write(struct file *filp, const char __user *buf,
275		  size_t count, loff_t pos, loff_t *ppos)
276{
277	struct address_space * mapping = filp->f_mapping;
278	const struct address_space_operations *a_ops = mapping->a_ops;
279	struct inode 	*inode = mapping->host;
280	long		status = 0;
281	struct page	*page;
282	size_t		bytes;
283	ssize_t		written = 0;
284
285	BUG_ON(!mapping->a_ops->get_xip_page);
286
287	do {
288		unsigned long index;
289		unsigned long offset;
290		size_t copied;
291
292		offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
293		index = pos >> PAGE_CACHE_SHIFT;
294		bytes = PAGE_CACHE_SIZE - offset;
295		if (bytes > count)
296			bytes = count;
297
298		/*
299		 * Bring in the user page that we will copy from _first_.
300		 * Otherwise there's a nasty deadlock on copying from the
301		 * same page as we're writing to, without it being marked
302		 * up-to-date.
303		 */
304		fault_in_pages_readable(buf, bytes);
305
306		page = a_ops->get_xip_page(mapping,
307					   index*(PAGE_SIZE/512), 0);
308		if (IS_ERR(page) && (PTR_ERR(page) == -ENODATA)) {
309			/* we allocate a new page unmap it */
310			page = a_ops->get_xip_page(mapping,
311						   index*(PAGE_SIZE/512), 1);
312			if (!IS_ERR(page))
313				/* unmap page at pgoff from all other vmas */
314				__xip_unmap(mapping, index);
315		}
316
317		if (IS_ERR(page)) {
318			status = PTR_ERR(page);
319			break;
320		}
321
322		copied = filemap_copy_from_user(page, offset, buf, bytes);
323		flush_dcache_page(page);
324		if (likely(copied > 0)) {
325			status = copied;
326
327			if (status >= 0) {
328				written += status;
329				count -= status;
330				pos += status;
331				buf += status;
332			}
333		}
334		if (unlikely(copied != bytes))
335			if (status >= 0)
336				status = -EFAULT;
337		if (status < 0)
338			break;
339	} while (count);
340	*ppos = pos;
341	/*
342	 * No need to use i_size_read() here, the i_size
343	 * cannot change under us because we hold i_mutex.
344	 */
345	if (pos > inode->i_size) {
346		i_size_write(inode, pos);
347		mark_inode_dirty(inode);
348	}
349
350	return written ? written : status;
351}
352
353ssize_t
354xip_file_write(struct file *filp, const char __user *buf, size_t len,
355	       loff_t *ppos)
356{
357	struct address_space *mapping = filp->f_mapping;
358	struct inode *inode = mapping->host;
359	size_t count;
360	loff_t pos;
361	ssize_t ret;
362
363	mutex_lock(&inode->i_mutex);
364
365	if (!access_ok(VERIFY_READ, buf, len)) {
366		ret=-EFAULT;
367		goto out_up;
368	}
369
370	pos = *ppos;
371	count = len;
372
373	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
374
375	/* We can write back this queue in page reclaim */
376	current->backing_dev_info = mapping->backing_dev_info;
377
378	ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
379	if (ret)
380		goto out_backing;
381	if (count == 0)
382		goto out_backing;
383
384	ret = remove_suid(filp->f_path.dentry);
385	if (ret)
386		goto out_backing;
387
388	file_update_time(filp);
389
390	ret = __xip_file_write (filp, buf, count, pos, ppos);
391
392 out_backing:
393	current->backing_dev_info = NULL;
394 out_up:
395	mutex_unlock(&inode->i_mutex);
396	return ret;
397}
398EXPORT_SYMBOL_GPL(xip_file_write);
399
400/*
401 * truncate a page used for execute in place
402 * functionality is analog to block_truncate_page but does use get_xip_page
403 * to get the page instead of page cache
404 */
405int
406xip_truncate_page(struct address_space *mapping, loff_t from)
407{
408	pgoff_t index = from >> PAGE_CACHE_SHIFT;
409	unsigned offset = from & (PAGE_CACHE_SIZE-1);
410	unsigned blocksize;
411	unsigned length;
412	struct page *page;
413
414	BUG_ON(!mapping->a_ops->get_xip_page);
415
416	blocksize = 1 << mapping->host->i_blkbits;
417	length = offset & (blocksize - 1);
418
419	/* Block boundary? Nothing to do */
420	if (!length)
421		return 0;
422
423	length = blocksize - length;
424
425	page = mapping->a_ops->get_xip_page(mapping,
426					    index*(PAGE_SIZE/512), 0);
427	if (!page)
428		return -ENOMEM;
429	if (unlikely(IS_ERR(page))) {
430		if (PTR_ERR(page) == -ENODATA)
431			/* Hole? No need to truncate */
432			return 0;
433		else
434			return PTR_ERR(page);
435	}
436	zero_user_page(page, offset, length, KM_USER0);
437	return 0;
438}
439EXPORT_SYMBOL_GPL(xip_truncate_page);
440