Commit | Line | Data |
---|---|---|
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> | |
16 | #include <asm/tlbflush.h> | |
17 | #include "filemap.h" | |
18 | ||
19 | /* | |
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 | |
22 | * stuff. | |
23 | * | |
24 | * Note the struct file* is not used at all. It may be NULL. | |
25 | */ | |
26 | static void | |
27 | do_xip_mapping_read(struct address_space *mapping, | |
28 | struct file_ra_state *_ra, | |
29 | struct file *filp, | |
30 | loff_t *ppos, | |
31 | read_descriptor_t *desc, | |
32 | read_actor_t actor) | |
33 | { | |
34 | struct inode *inode = mapping->host; | |
35 | unsigned long index, end_index, offset; | |
36 | loff_t isize; | |
37 | ||
38 | BUG_ON(!mapping->a_ops->get_xip_page); | |
39 | ||
40 | index = *ppos >> PAGE_CACHE_SHIFT; | |
41 | offset = *ppos & ~PAGE_CACHE_MASK; | |
42 | ||
43 | isize = i_size_read(inode); | |
44 | if (!isize) | |
45 | goto out; | |
46 | ||
47 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
48 | for (;;) { | |
49 | struct page *page; | |
50 | unsigned long nr, ret; | |
51 | ||
52 | /* nr is the maximum number of bytes to copy from this page */ | |
53 | nr = PAGE_CACHE_SIZE; | |
54 | if (index >= end_index) { | |
55 | if (index > end_index) | |
56 | goto out; | |
57 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
58 | if (nr <= offset) { | |
59 | goto out; | |
60 | } | |
61 | } | |
62 | nr = nr - offset; | |
63 | ||
64 | page = mapping->a_ops->get_xip_page(mapping, | |
65 | index*(PAGE_SIZE/512), 0); | |
66 | if (!page) | |
67 | goto no_xip_page; | |
68 | if (unlikely(IS_ERR(page))) { | |
69 | if (PTR_ERR(page) == -ENODATA) { | |
70 | /* sparse */ | |
71 | page = virt_to_page(empty_zero_page); | |
72 | } else { | |
73 | desc->error = PTR_ERR(page); | |
74 | goto out; | |
75 | } | |
76 | } else | |
77 | BUG_ON(!PageUptodate(page)); | |
78 | ||
79 | /* If users can be writing to this page using arbitrary | |
80 | * virtual addresses, take care about potential aliasing | |
81 | * before reading the page on the kernel side. | |
82 | */ | |
83 | if (mapping_writably_mapped(mapping)) | |
84 | flush_dcache_page(page); | |
85 | ||
86 | /* | |
87 | * Ok, we have the page, and it's up-to-date, so | |
88 | * now we can copy it to user space... | |
89 | * | |
90 | * The actor routine returns how many bytes were actually used.. | |
91 | * NOTE! This may not be the same as how much of a user buffer | |
92 | * we filled up (we may be padding etc), so we can only update | |
93 | * "pos" here (the actor routine has to update the user buffer | |
94 | * pointers and the remaining count). | |
95 | */ | |
96 | ret = actor(desc, page, offset, nr); | |
97 | offset += ret; | |
98 | index += offset >> PAGE_CACHE_SHIFT; | |
99 | offset &= ~PAGE_CACHE_MASK; | |
100 | ||
101 | if (ret == nr && desc->count) | |
102 | continue; | |
103 | goto out; | |
104 | ||
105 | no_xip_page: | |
106 | /* Did not get the page. Report it */ | |
107 | desc->error = -EIO; | |
108 | goto out; | |
109 | } | |
110 | ||
111 | out: | |
112 | *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; | |
113 | if (filp) | |
114 | file_accessed(filp); | |
115 | } | |
116 | ||
ceffc078 | 117 | ssize_t |
eb6fe0c3 | 118 | xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos) |
ceffc078 | 119 | { |
eb6fe0c3 | 120 | read_descriptor_t desc; |
ceffc078 | 121 | |
eb6fe0c3 CO |
122 | if (!access_ok(VERIFY_WRITE, buf, len)) |
123 | return -EFAULT; | |
ceffc078 | 124 | |
eb6fe0c3 CO |
125 | desc.written = 0; |
126 | desc.arg.buf = buf; | |
127 | desc.count = len; | |
128 | desc.error = 0; | |
ceffc078 | 129 | |
eb6fe0c3 CO |
130 | do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp, |
131 | ppos, &desc, file_read_actor); | |
132 | ||
133 | if (desc.written) | |
134 | return desc.written; | |
135 | else | |
136 | return desc.error; | |
ceffc078 | 137 | } |
eb6fe0c3 | 138 | EXPORT_SYMBOL_GPL(xip_file_read); |
ceffc078 CO |
139 | |
140 | ssize_t | |
141 | xip_file_sendfile(struct file *in_file, loff_t *ppos, | |
142 | size_t count, read_actor_t actor, void *target) | |
143 | { | |
144 | read_descriptor_t desc; | |
145 | ||
146 | if (!count) | |
147 | return 0; | |
148 | ||
149 | desc.written = 0; | |
150 | desc.count = count; | |
151 | desc.arg.data = target; | |
152 | desc.error = 0; | |
153 | ||
154 | do_xip_mapping_read(in_file->f_mapping, &in_file->f_ra, in_file, | |
155 | ppos, &desc, actor); | |
156 | if (desc.written) | |
157 | return desc.written; | |
158 | return desc.error; | |
159 | } | |
160 | EXPORT_SYMBOL_GPL(xip_file_sendfile); | |
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 | * empty_zero_page when found at pgoff. Should it go in rmap.c? | |
168 | */ | |
169 | static 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 | ||
180 | spin_lock(&mapping->i_mmap_lock); | |
181 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
182 | mm = vma->vm_mm; | |
183 | address = vma->vm_start + | |
184 | ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); | |
185 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
186 | /* | |
187 | * We need the page_table_lock to protect us from page faults, | |
188 | * munmap, fork, etc... | |
189 | */ | |
190 | pte = page_check_address(virt_to_page(empty_zero_page), mm, | |
191 | address); | |
192 | if (!IS_ERR(pte)) { | |
193 | /* Nuke the page table entry. */ | |
194 | flush_cache_page(vma, address, pte_pfn(pte)); | |
195 | pteval = ptep_clear_flush(vma, address, pte); | |
196 | BUG_ON(pte_dirty(pteval)); | |
197 | pte_unmap(pte); | |
198 | spin_unlock(&mm->page_table_lock); | |
199 | } | |
200 | } | |
201 | spin_unlock(&mapping->i_mmap_lock); | |
202 | } | |
203 | ||
204 | /* | |
205 | * xip_nopage() is invoked via the vma operations vector for a | |
206 | * mapped memory region to read in file data during a page fault. | |
207 | * | |
208 | * This function is derived from filemap_nopage, but used for execute in place | |
209 | */ | |
210 | static struct page * | |
211 | xip_file_nopage(struct vm_area_struct * area, | |
212 | unsigned long address, | |
213 | int *type) | |
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 | unsigned long size, pgoff, endoff; | |
220 | ||
221 | pgoff = ((address - area->vm_start) >> PAGE_CACHE_SHIFT) | |
222 | + area->vm_pgoff; | |
223 | endoff = ((area->vm_end - area->vm_start) >> PAGE_CACHE_SHIFT) | |
224 | + area->vm_pgoff; | |
225 | ||
226 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
227 | if (pgoff >= size) { | |
228 | return NULL; | |
229 | } | |
230 | ||
231 | page = mapping->a_ops->get_xip_page(mapping, pgoff*(PAGE_SIZE/512), 0); | |
232 | if (!IS_ERR(page)) { | |
233 | BUG_ON(!PageUptodate(page)); | |
234 | return page; | |
235 | } | |
236 | if (PTR_ERR(page) != -ENODATA) | |
237 | return NULL; | |
238 | ||
239 | /* sparse block */ | |
240 | if ((area->vm_flags & (VM_WRITE | VM_MAYWRITE)) && | |
241 | (area->vm_flags & (VM_SHARED| VM_MAYSHARE)) && | |
242 | (!(mapping->host->i_sb->s_flags & MS_RDONLY))) { | |
243 | /* maybe shared writable, allocate new block */ | |
244 | page = mapping->a_ops->get_xip_page (mapping, | |
245 | pgoff*(PAGE_SIZE/512), 1); | |
246 | if (IS_ERR(page)) | |
247 | return NULL; | |
248 | BUG_ON(!PageUptodate(page)); | |
249 | /* unmap page at pgoff from all other vmas */ | |
250 | __xip_unmap(mapping, pgoff); | |
251 | } else { | |
252 | /* not shared and writable, use empty_zero_page */ | |
253 | page = virt_to_page(empty_zero_page); | |
254 | } | |
255 | ||
256 | return page; | |
257 | } | |
258 | ||
259 | static struct vm_operations_struct xip_file_vm_ops = { | |
260 | .nopage = xip_file_nopage, | |
261 | }; | |
262 | ||
263 | int xip_file_mmap(struct file * file, struct vm_area_struct * vma) | |
264 | { | |
265 | BUG_ON(!file->f_mapping->a_ops->get_xip_page); | |
266 | ||
267 | file_accessed(file); | |
268 | vma->vm_ops = &xip_file_vm_ops; | |
269 | return 0; | |
270 | } | |
271 | EXPORT_SYMBOL_GPL(xip_file_mmap); | |
272 | ||
273 | static ssize_t | |
eb6fe0c3 CO |
274 | __xip_file_write(struct file *filp, const char __user *buf, |
275 | size_t count, loff_t pos, loff_t *ppos) | |
ceffc078 | 276 | { |
eb6fe0c3 | 277 | struct address_space * mapping = filp->f_mapping; |
ceffc078 CO |
278 | 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; | |
ceffc078 CO |
283 | ssize_t written = 0; |
284 | ||
285 | BUG_ON(!mapping->a_ops->get_xip_page); | |
286 | ||
ceffc078 CO |
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, | |
eb6fe0c3 | 307 | index*(PAGE_SIZE/512), 0); |
ceffc078 CO |
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, | |
eb6fe0c3 | 311 | index*(PAGE_SIZE/512), 1); |
ceffc078 | 312 | if (!IS_ERR(page)) |
eb6fe0c3 CO |
313 | /* unmap page at pgoff from all other vmas */ |
314 | __xip_unmap(mapping, index); | |
ceffc078 CO |
315 | } |
316 | ||
317 | if (IS_ERR(page)) { | |
318 | status = PTR_ERR(page); | |
319 | break; | |
320 | } | |
321 | ||
322 | BUG_ON(!PageUptodate(page)); | |
323 | ||
eb6fe0c3 | 324 | copied = filemap_copy_from_user(page, offset, buf, bytes); |
ceffc078 CO |
325 | flush_dcache_page(page); |
326 | if (likely(copied > 0)) { | |
327 | status = copied; | |
328 | ||
329 | if (status >= 0) { | |
330 | written += status; | |
331 | count -= status; | |
332 | pos += status; | |
333 | buf += status; | |
ceffc078 CO |
334 | } |
335 | } | |
336 | if (unlikely(copied != bytes)) | |
337 | if (status >= 0) | |
338 | status = -EFAULT; | |
339 | if (status < 0) | |
340 | break; | |
341 | } while (count); | |
342 | *ppos = pos; | |
343 | /* | |
344 | * No need to use i_size_read() here, the i_size | |
345 | * cannot change under us because we hold i_sem. | |
346 | */ | |
347 | if (pos > inode->i_size) { | |
348 | i_size_write(inode, pos); | |
349 | mark_inode_dirty(inode); | |
350 | } | |
351 | ||
352 | return written ? written : status; | |
353 | } | |
354 | ||
eb6fe0c3 CO |
355 | ssize_t |
356 | xip_file_write(struct file *filp, const char __user *buf, size_t len, | |
357 | loff_t *ppos) | |
ceffc078 | 358 | { |
eb6fe0c3 CO |
359 | struct address_space *mapping = filp->f_mapping; |
360 | struct inode *inode = mapping->host; | |
361 | size_t count; | |
362 | loff_t pos; | |
363 | ssize_t ret; | |
ceffc078 | 364 | |
eb6fe0c3 | 365 | down(&inode->i_sem); |
ceffc078 | 366 | |
eb6fe0c3 CO |
367 | if (!access_ok(VERIFY_READ, buf, len)) { |
368 | ret=-EFAULT; | |
369 | goto out_up; | |
ceffc078 CO |
370 | } |
371 | ||
ceffc078 | 372 | pos = *ppos; |
eb6fe0c3 | 373 | count = len; |
ceffc078 CO |
374 | |
375 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
376 | ||
eb6fe0c3 CO |
377 | /* We can write back this queue in page reclaim */ |
378 | current->backing_dev_info = mapping->backing_dev_info; | |
ceffc078 | 379 | |
eb6fe0c3 CO |
380 | ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode)); |
381 | if (ret) | |
382 | goto out_backing; | |
ceffc078 | 383 | if (count == 0) |
eb6fe0c3 | 384 | goto out_backing; |
ceffc078 | 385 | |
eb6fe0c3 CO |
386 | ret = remove_suid(filp->f_dentry); |
387 | if (ret) | |
388 | goto out_backing; | |
ceffc078 CO |
389 | |
390 | inode_update_time(inode, 1); | |
391 | ||
eb6fe0c3 | 392 | ret = __xip_file_write (filp, buf, count, pos, ppos); |
ceffc078 | 393 | |
eb6fe0c3 CO |
394 | out_backing: |
395 | current->backing_dev_info = NULL; | |
396 | out_up: | |
ceffc078 CO |
397 | up(&inode->i_sem); |
398 | return ret; | |
399 | } | |
eb6fe0c3 | 400 | EXPORT_SYMBOL_GPL(xip_file_write); |
ceffc078 CO |
401 | |
402 | /* | |
403 | * truncate a page used for execute in place | |
404 | * functionality is analog to block_truncate_page but does use get_xip_page | |
405 | * to get the page instead of page cache | |
406 | */ | |
407 | int | |
408 | xip_truncate_page(struct address_space *mapping, loff_t from) | |
409 | { | |
410 | pgoff_t index = from >> PAGE_CACHE_SHIFT; | |
411 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
412 | unsigned blocksize; | |
413 | unsigned length; | |
414 | struct page *page; | |
415 | void *kaddr; | |
ceffc078 CO |
416 | |
417 | BUG_ON(!mapping->a_ops->get_xip_page); | |
418 | ||
419 | blocksize = 1 << mapping->host->i_blkbits; | |
420 | length = offset & (blocksize - 1); | |
421 | ||
422 | /* Block boundary? Nothing to do */ | |
423 | if (!length) | |
424 | return 0; | |
425 | ||
426 | length = blocksize - length; | |
427 | ||
428 | page = mapping->a_ops->get_xip_page(mapping, | |
429 | index*(PAGE_SIZE/512), 0); | |
ceffc078 | 430 | if (!page) |
eb6fe0c3 | 431 | return -ENOMEM; |
ceffc078 | 432 | if (unlikely(IS_ERR(page))) { |
eb6fe0c3 | 433 | if (PTR_ERR(page) == -ENODATA) |
ceffc078 CO |
434 | /* Hole? No need to truncate */ |
435 | return 0; | |
eb6fe0c3 CO |
436 | else |
437 | return PTR_ERR(page); | |
ceffc078 CO |
438 | } else |
439 | BUG_ON(!PageUptodate(page)); | |
440 | kaddr = kmap_atomic(page, KM_USER0); | |
441 | memset(kaddr + offset, 0, length); | |
442 | kunmap_atomic(kaddr, KM_USER0); | |
443 | ||
444 | flush_dcache_page(page); | |
eb6fe0c3 | 445 | return 0; |
ceffc078 CO |
446 | } |
447 | EXPORT_SYMBOL_GPL(xip_truncate_page); |