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 */ | |
afa597ba | 71 | page = ZERO_PAGE(0); |
ceffc078 CO |
72 | } else { |
73 | desc->error = PTR_ERR(page); | |
74 | goto out; | |
75 | } | |
afa597ba | 76 | } |
ceffc078 CO |
77 | |
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. | |
81 | */ | |
82 | if (mapping_writably_mapped(mapping)) | |
83 | flush_dcache_page(page); | |
84 | ||
85 | /* | |
afa597ba | 86 | * Ok, we have the page, so now we can copy it to user space... |
ceffc078 CO |
87 | * |
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). | |
93 | */ | |
94 | ret = actor(desc, page, offset, nr); | |
95 | offset += ret; | |
96 | index += offset >> PAGE_CACHE_SHIFT; | |
97 | offset &= ~PAGE_CACHE_MASK; | |
98 | ||
99 | if (ret == nr && desc->count) | |
100 | continue; | |
101 | goto out; | |
102 | ||
103 | no_xip_page: | |
104 | /* Did not get the page. Report it */ | |
105 | desc->error = -EIO; | |
106 | goto out; | |
107 | } | |
108 | ||
109 | out: | |
110 | *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; | |
111 | if (filp) | |
112 | file_accessed(filp); | |
113 | } | |
114 | ||
ceffc078 | 115 | ssize_t |
eb6fe0c3 | 116 | xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos) |
ceffc078 | 117 | { |
eb6fe0c3 | 118 | read_descriptor_t desc; |
ceffc078 | 119 | |
eb6fe0c3 CO |
120 | if (!access_ok(VERIFY_WRITE, buf, len)) |
121 | return -EFAULT; | |
ceffc078 | 122 | |
eb6fe0c3 CO |
123 | desc.written = 0; |
124 | desc.arg.buf = buf; | |
125 | desc.count = len; | |
126 | desc.error = 0; | |
ceffc078 | 127 | |
eb6fe0c3 CO |
128 | do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp, |
129 | ppos, &desc, file_read_actor); | |
130 | ||
131 | if (desc.written) | |
132 | return desc.written; | |
133 | else | |
134 | return desc.error; | |
ceffc078 | 135 | } |
eb6fe0c3 | 136 | EXPORT_SYMBOL_GPL(xip_file_read); |
ceffc078 CO |
137 | |
138 | ssize_t | |
139 | xip_file_sendfile(struct file *in_file, loff_t *ppos, | |
140 | size_t count, read_actor_t actor, void *target) | |
141 | { | |
142 | read_descriptor_t desc; | |
143 | ||
144 | if (!count) | |
145 | return 0; | |
146 | ||
147 | desc.written = 0; | |
148 | desc.count = count; | |
149 | desc.arg.data = target; | |
150 | desc.error = 0; | |
151 | ||
152 | do_xip_mapping_read(in_file->f_mapping, &in_file->f_ra, in_file, | |
153 | ppos, &desc, actor); | |
154 | if (desc.written) | |
155 | return desc.written; | |
156 | return desc.error; | |
157 | } | |
158 | EXPORT_SYMBOL_GPL(xip_file_sendfile); | |
159 | ||
160 | /* | |
161 | * __xip_unmap is invoked from xip_unmap and | |
162 | * xip_write | |
163 | * | |
164 | * This function walks all vmas of the address_space and unmaps the | |
afa597ba | 165 | * ZERO_PAGE when found at pgoff. Should it go in rmap.c? |
ceffc078 CO |
166 | */ |
167 | static void | |
168 | __xip_unmap (struct address_space * mapping, | |
169 | unsigned long pgoff) | |
170 | { | |
171 | struct vm_area_struct *vma; | |
172 | struct mm_struct *mm; | |
173 | struct prio_tree_iter iter; | |
174 | unsigned long address; | |
175 | pte_t *pte; | |
176 | pte_t pteval; | |
b5810039 | 177 | struct page *page = ZERO_PAGE(address); |
ceffc078 CO |
178 | |
179 | spin_lock(&mapping->i_mmap_lock); | |
180 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
181 | mm = vma->vm_mm; | |
182 | address = vma->vm_start + | |
183 | ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); | |
184 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
185 | /* | |
186 | * We need the page_table_lock to protect us from page faults, | |
187 | * munmap, fork, etc... | |
188 | */ | |
b5810039 | 189 | pte = page_check_address(page, mm, address); |
ceffc078 CO |
190 | if (!IS_ERR(pte)) { |
191 | /* Nuke the page table entry. */ | |
082ff0a9 | 192 | flush_cache_page(vma, address, pte_pfn(*pte)); |
ceffc078 | 193 | pteval = ptep_clear_flush(vma, address, pte); |
b5810039 NP |
194 | page_remove_rmap(page); |
195 | dec_mm_counter(mm, file_rss); | |
ceffc078 CO |
196 | BUG_ON(pte_dirty(pteval)); |
197 | pte_unmap(pte); | |
198 | spin_unlock(&mm->page_table_lock); | |
b5810039 | 199 | page_cache_release(page); |
ceffc078 CO |
200 | } |
201 | } | |
202 | spin_unlock(&mapping->i_mmap_lock); | |
203 | } | |
204 | ||
205 | /* | |
206 | * xip_nopage() is invoked via the vma operations vector for a | |
207 | * mapped memory region to read in file data during a page fault. | |
208 | * | |
209 | * This function is derived from filemap_nopage, but used for execute in place | |
210 | */ | |
211 | static struct page * | |
212 | xip_file_nopage(struct vm_area_struct * area, | |
213 | unsigned long address, | |
214 | int *type) | |
215 | { | |
216 | struct file *file = area->vm_file; | |
217 | struct address_space *mapping = file->f_mapping; | |
218 | struct inode *inode = mapping->host; | |
219 | struct page *page; | |
220 | unsigned long size, pgoff, endoff; | |
221 | ||
222 | pgoff = ((address - area->vm_start) >> PAGE_CACHE_SHIFT) | |
223 | + area->vm_pgoff; | |
224 | endoff = ((area->vm_end - area->vm_start) >> PAGE_CACHE_SHIFT) | |
225 | + area->vm_pgoff; | |
226 | ||
227 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
228 | if (pgoff >= size) { | |
229 | return NULL; | |
230 | } | |
231 | ||
232 | page = mapping->a_ops->get_xip_page(mapping, pgoff*(PAGE_SIZE/512), 0); | |
233 | if (!IS_ERR(page)) { | |
b5810039 | 234 | goto out; |
ceffc078 CO |
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; | |
ceffc078 CO |
248 | /* unmap page at pgoff from all other vmas */ |
249 | __xip_unmap(mapping, pgoff); | |
250 | } else { | |
afa597ba CO |
251 | /* not shared and writable, use ZERO_PAGE() */ |
252 | page = ZERO_PAGE(address); | |
ceffc078 CO |
253 | } |
254 | ||
b5810039 NP |
255 | out: |
256 | page_cache_get(page); | |
ceffc078 CO |
257 | return page; |
258 | } | |
259 | ||
260 | static struct vm_operations_struct xip_file_vm_ops = { | |
261 | .nopage = xip_file_nopage, | |
262 | }; | |
263 | ||
264 | int xip_file_mmap(struct file * file, struct vm_area_struct * vma) | |
265 | { | |
266 | BUG_ON(!file->f_mapping->a_ops->get_xip_page); | |
267 | ||
268 | file_accessed(file); | |
269 | vma->vm_ops = &xip_file_vm_ops; | |
270 | return 0; | |
271 | } | |
272 | EXPORT_SYMBOL_GPL(xip_file_mmap); | |
273 | ||
274 | static ssize_t | |
eb6fe0c3 CO |
275 | __xip_file_write(struct file *filp, const char __user *buf, |
276 | size_t count, loff_t pos, loff_t *ppos) | |
ceffc078 | 277 | { |
eb6fe0c3 | 278 | struct address_space * mapping = filp->f_mapping; |
ceffc078 CO |
279 | struct address_space_operations *a_ops = mapping->a_ops; |
280 | struct inode *inode = mapping->host; | |
281 | long status = 0; | |
282 | struct page *page; | |
283 | size_t bytes; | |
ceffc078 CO |
284 | ssize_t written = 0; |
285 | ||
286 | BUG_ON(!mapping->a_ops->get_xip_page); | |
287 | ||
ceffc078 CO |
288 | do { |
289 | unsigned long index; | |
290 | unsigned long offset; | |
291 | size_t copied; | |
292 | ||
293 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ | |
294 | index = pos >> PAGE_CACHE_SHIFT; | |
295 | bytes = PAGE_CACHE_SIZE - offset; | |
296 | if (bytes > count) | |
297 | bytes = count; | |
298 | ||
299 | /* | |
300 | * Bring in the user page that we will copy from _first_. | |
301 | * Otherwise there's a nasty deadlock on copying from the | |
302 | * same page as we're writing to, without it being marked | |
303 | * up-to-date. | |
304 | */ | |
305 | fault_in_pages_readable(buf, bytes); | |
306 | ||
307 | page = a_ops->get_xip_page(mapping, | |
eb6fe0c3 | 308 | index*(PAGE_SIZE/512), 0); |
ceffc078 CO |
309 | if (IS_ERR(page) && (PTR_ERR(page) == -ENODATA)) { |
310 | /* we allocate a new page unmap it */ | |
311 | page = a_ops->get_xip_page(mapping, | |
eb6fe0c3 | 312 | index*(PAGE_SIZE/512), 1); |
ceffc078 | 313 | if (!IS_ERR(page)) |
eb6fe0c3 CO |
314 | /* unmap page at pgoff from all other vmas */ |
315 | __xip_unmap(mapping, index); | |
ceffc078 CO |
316 | } |
317 | ||
318 | if (IS_ERR(page)) { | |
319 | status = PTR_ERR(page); | |
320 | break; | |
321 | } | |
322 | ||
eb6fe0c3 | 323 | copied = filemap_copy_from_user(page, offset, buf, bytes); |
ceffc078 CO |
324 | flush_dcache_page(page); |
325 | if (likely(copied > 0)) { | |
326 | status = copied; | |
327 | ||
328 | if (status >= 0) { | |
329 | written += status; | |
330 | count -= status; | |
331 | pos += status; | |
332 | buf += status; | |
ceffc078 CO |
333 | } |
334 | } | |
335 | if (unlikely(copied != bytes)) | |
336 | if (status >= 0) | |
337 | status = -EFAULT; | |
338 | if (status < 0) | |
339 | break; | |
340 | } while (count); | |
341 | *ppos = pos; | |
342 | /* | |
343 | * No need to use i_size_read() here, the i_size | |
344 | * cannot change under us because we hold i_sem. | |
345 | */ | |
346 | if (pos > inode->i_size) { | |
347 | i_size_write(inode, pos); | |
348 | mark_inode_dirty(inode); | |
349 | } | |
350 | ||
351 | return written ? written : status; | |
352 | } | |
353 | ||
eb6fe0c3 CO |
354 | ssize_t |
355 | xip_file_write(struct file *filp, const char __user *buf, size_t len, | |
356 | loff_t *ppos) | |
ceffc078 | 357 | { |
eb6fe0c3 CO |
358 | struct address_space *mapping = filp->f_mapping; |
359 | struct inode *inode = mapping->host; | |
360 | size_t count; | |
361 | loff_t pos; | |
362 | ssize_t ret; | |
ceffc078 | 363 | |
eb6fe0c3 | 364 | down(&inode->i_sem); |
ceffc078 | 365 | |
eb6fe0c3 CO |
366 | if (!access_ok(VERIFY_READ, buf, len)) { |
367 | ret=-EFAULT; | |
368 | goto out_up; | |
ceffc078 CO |
369 | } |
370 | ||
ceffc078 | 371 | pos = *ppos; |
eb6fe0c3 | 372 | count = len; |
ceffc078 CO |
373 | |
374 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
375 | ||
eb6fe0c3 CO |
376 | /* We can write back this queue in page reclaim */ |
377 | current->backing_dev_info = mapping->backing_dev_info; | |
ceffc078 | 378 | |
eb6fe0c3 CO |
379 | ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode)); |
380 | if (ret) | |
381 | goto out_backing; | |
ceffc078 | 382 | if (count == 0) |
eb6fe0c3 | 383 | goto out_backing; |
ceffc078 | 384 | |
eb6fe0c3 CO |
385 | ret = remove_suid(filp->f_dentry); |
386 | if (ret) | |
387 | goto out_backing; | |
ceffc078 CO |
388 | |
389 | inode_update_time(inode, 1); | |
390 | ||
eb6fe0c3 | 391 | ret = __xip_file_write (filp, buf, count, pos, ppos); |
ceffc078 | 392 | |
eb6fe0c3 CO |
393 | out_backing: |
394 | current->backing_dev_info = NULL; | |
395 | out_up: | |
ceffc078 CO |
396 | up(&inode->i_sem); |
397 | return ret; | |
398 | } | |
eb6fe0c3 | 399 | EXPORT_SYMBOL_GPL(xip_file_write); |
ceffc078 CO |
400 | |
401 | /* | |
402 | * truncate a page used for execute in place | |
403 | * functionality is analog to block_truncate_page but does use get_xip_page | |
404 | * to get the page instead of page cache | |
405 | */ | |
406 | int | |
407 | xip_truncate_page(struct address_space *mapping, loff_t from) | |
408 | { | |
409 | pgoff_t index = from >> PAGE_CACHE_SHIFT; | |
410 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
411 | unsigned blocksize; | |
412 | unsigned length; | |
413 | struct page *page; | |
414 | void *kaddr; | |
ceffc078 CO |
415 | |
416 | BUG_ON(!mapping->a_ops->get_xip_page); | |
417 | ||
418 | blocksize = 1 << mapping->host->i_blkbits; | |
419 | length = offset & (blocksize - 1); | |
420 | ||
421 | /* Block boundary? Nothing to do */ | |
422 | if (!length) | |
423 | return 0; | |
424 | ||
425 | length = blocksize - length; | |
426 | ||
427 | page = mapping->a_ops->get_xip_page(mapping, | |
428 | index*(PAGE_SIZE/512), 0); | |
ceffc078 | 429 | if (!page) |
eb6fe0c3 | 430 | return -ENOMEM; |
ceffc078 | 431 | if (unlikely(IS_ERR(page))) { |
eb6fe0c3 | 432 | if (PTR_ERR(page) == -ENODATA) |
ceffc078 CO |
433 | /* Hole? No need to truncate */ |
434 | return 0; | |
eb6fe0c3 CO |
435 | else |
436 | return PTR_ERR(page); | |
afa597ba | 437 | } |
ceffc078 CO |
438 | kaddr = kmap_atomic(page, KM_USER0); |
439 | memset(kaddr + offset, 0, length); | |
440 | kunmap_atomic(kaddr, KM_USER0); | |
441 | ||
442 | flush_dcache_page(page); | |
eb6fe0c3 | 443 | return 0; |
ceffc078 CO |
444 | } |
445 | EXPORT_SYMBOL_GPL(xip_truncate_page); |