]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/mmap.c | |
3 | * | |
4 | * Written by obz. | |
5 | * | |
6 | * Address space accounting code <alan@redhat.com> | |
7 | */ | |
8 | ||
9 | #include <linux/slab.h> | |
10 | #include <linux/mm.h> | |
11 | #include <linux/shm.h> | |
12 | #include <linux/mman.h> | |
13 | #include <linux/pagemap.h> | |
14 | #include <linux/swap.h> | |
15 | #include <linux/syscalls.h> | |
c59ede7b | 16 | #include <linux/capability.h> |
1da177e4 LT |
17 | #include <linux/init.h> |
18 | #include <linux/file.h> | |
19 | #include <linux/fs.h> | |
20 | #include <linux/personality.h> | |
21 | #include <linux/security.h> | |
22 | #include <linux/hugetlb.h> | |
23 | #include <linux/profile.h> | |
24 | #include <linux/module.h> | |
25 | #include <linux/mount.h> | |
26 | #include <linux/mempolicy.h> | |
27 | #include <linux/rmap.h> | |
28 | ||
29 | #include <asm/uaccess.h> | |
30 | #include <asm/cacheflush.h> | |
31 | #include <asm/tlb.h> | |
32 | ||
e0da382c HD |
33 | static void unmap_region(struct mm_struct *mm, |
34 | struct vm_area_struct *vma, struct vm_area_struct *prev, | |
35 | unsigned long start, unsigned long end); | |
36 | ||
1da177e4 LT |
37 | /* |
38 | * WARNING: the debugging will use recursive algorithms so never enable this | |
39 | * unless you know what you are doing. | |
40 | */ | |
41 | #undef DEBUG_MM_RB | |
42 | ||
43 | /* description of effects of mapping type and prot in current implementation. | |
44 | * this is due to the limited x86 page protection hardware. The expected | |
45 | * behavior is in parens: | |
46 | * | |
47 | * map_type prot | |
48 | * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC | |
49 | * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes | |
50 | * w: (no) no w: (no) no w: (yes) yes w: (no) no | |
51 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
52 | * | |
53 | * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes | |
54 | * w: (no) no w: (no) no w: (copy) copy w: (no) no | |
55 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
56 | * | |
57 | */ | |
58 | pgprot_t protection_map[16] = { | |
59 | __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, | |
60 | __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 | |
61 | }; | |
62 | ||
63 | int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ | |
64 | int sysctl_overcommit_ratio = 50; /* default is 50% */ | |
c3d8c141 | 65 | int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; |
1da177e4 LT |
66 | atomic_t vm_committed_space = ATOMIC_INIT(0); |
67 | ||
68 | /* | |
69 | * Check that a process has enough memory to allocate a new virtual | |
70 | * mapping. 0 means there is enough memory for the allocation to | |
71 | * succeed and -ENOMEM implies there is not. | |
72 | * | |
73 | * We currently support three overcommit policies, which are set via the | |
74 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting | |
75 | * | |
76 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. | |
77 | * Additional code 2002 Jul 20 by Robert Love. | |
78 | * | |
79 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. | |
80 | * | |
81 | * Note this is a helper function intended to be used by LSMs which | |
82 | * wish to use this logic. | |
83 | */ | |
84 | int __vm_enough_memory(long pages, int cap_sys_admin) | |
85 | { | |
86 | unsigned long free, allowed; | |
87 | ||
88 | vm_acct_memory(pages); | |
89 | ||
90 | /* | |
91 | * Sometimes we want to use more memory than we have | |
92 | */ | |
93 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) | |
94 | return 0; | |
95 | ||
96 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { | |
97 | unsigned long n; | |
98 | ||
99 | free = get_page_cache_size(); | |
100 | free += nr_swap_pages; | |
101 | ||
102 | /* | |
103 | * Any slabs which are created with the | |
104 | * SLAB_RECLAIM_ACCOUNT flag claim to have contents | |
105 | * which are reclaimable, under pressure. The dentry | |
106 | * cache and most inode caches should fall into this | |
107 | */ | |
108 | free += atomic_read(&slab_reclaim_pages); | |
109 | ||
110 | /* | |
111 | * Leave the last 3% for root | |
112 | */ | |
113 | if (!cap_sys_admin) | |
114 | free -= free / 32; | |
115 | ||
116 | if (free > pages) | |
117 | return 0; | |
118 | ||
119 | /* | |
120 | * nr_free_pages() is very expensive on large systems, | |
121 | * only call if we're about to fail. | |
122 | */ | |
123 | n = nr_free_pages(); | |
124 | if (!cap_sys_admin) | |
125 | n -= n / 32; | |
126 | free += n; | |
127 | ||
128 | if (free > pages) | |
129 | return 0; | |
130 | vm_unacct_memory(pages); | |
131 | return -ENOMEM; | |
132 | } | |
133 | ||
134 | allowed = (totalram_pages - hugetlb_total_pages()) | |
135 | * sysctl_overcommit_ratio / 100; | |
136 | /* | |
137 | * Leave the last 3% for root | |
138 | */ | |
139 | if (!cap_sys_admin) | |
140 | allowed -= allowed / 32; | |
141 | allowed += total_swap_pages; | |
142 | ||
143 | /* Don't let a single process grow too big: | |
144 | leave 3% of the size of this process for other processes */ | |
145 | allowed -= current->mm->total_vm / 32; | |
146 | ||
2f60f8d3 SD |
147 | /* |
148 | * cast `allowed' as a signed long because vm_committed_space | |
149 | * sometimes has a negative value | |
150 | */ | |
151 | if (atomic_read(&vm_committed_space) < (long)allowed) | |
1da177e4 LT |
152 | return 0; |
153 | ||
154 | vm_unacct_memory(pages); | |
155 | ||
156 | return -ENOMEM; | |
157 | } | |
158 | ||
1da177e4 LT |
159 | EXPORT_SYMBOL(__vm_enough_memory); |
160 | ||
161 | /* | |
162 | * Requires inode->i_mapping->i_mmap_lock | |
163 | */ | |
164 | static void __remove_shared_vm_struct(struct vm_area_struct *vma, | |
165 | struct file *file, struct address_space *mapping) | |
166 | { | |
167 | if (vma->vm_flags & VM_DENYWRITE) | |
168 | atomic_inc(&file->f_dentry->d_inode->i_writecount); | |
169 | if (vma->vm_flags & VM_SHARED) | |
170 | mapping->i_mmap_writable--; | |
171 | ||
172 | flush_dcache_mmap_lock(mapping); | |
173 | if (unlikely(vma->vm_flags & VM_NONLINEAR)) | |
174 | list_del_init(&vma->shared.vm_set.list); | |
175 | else | |
176 | vma_prio_tree_remove(vma, &mapping->i_mmap); | |
177 | flush_dcache_mmap_unlock(mapping); | |
178 | } | |
179 | ||
180 | /* | |
a8fb5618 HD |
181 | * Unlink a file-based vm structure from its prio_tree, to hide |
182 | * vma from rmap and vmtruncate before freeing its page tables. | |
1da177e4 | 183 | */ |
a8fb5618 | 184 | void unlink_file_vma(struct vm_area_struct *vma) |
1da177e4 LT |
185 | { |
186 | struct file *file = vma->vm_file; | |
187 | ||
1da177e4 LT |
188 | if (file) { |
189 | struct address_space *mapping = file->f_mapping; | |
190 | spin_lock(&mapping->i_mmap_lock); | |
191 | __remove_shared_vm_struct(vma, file, mapping); | |
192 | spin_unlock(&mapping->i_mmap_lock); | |
193 | } | |
a8fb5618 HD |
194 | } |
195 | ||
196 | /* | |
197 | * Close a vm structure and free it, returning the next. | |
198 | */ | |
199 | static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) | |
200 | { | |
201 | struct vm_area_struct *next = vma->vm_next; | |
202 | ||
a8fb5618 | 203 | might_sleep(); |
1da177e4 LT |
204 | if (vma->vm_ops && vma->vm_ops->close) |
205 | vma->vm_ops->close(vma); | |
a8fb5618 HD |
206 | if (vma->vm_file) |
207 | fput(vma->vm_file); | |
1da177e4 LT |
208 | mpol_free(vma_policy(vma)); |
209 | kmem_cache_free(vm_area_cachep, vma); | |
a8fb5618 | 210 | return next; |
1da177e4 LT |
211 | } |
212 | ||
1da177e4 LT |
213 | asmlinkage unsigned long sys_brk(unsigned long brk) |
214 | { | |
215 | unsigned long rlim, retval; | |
216 | unsigned long newbrk, oldbrk; | |
217 | struct mm_struct *mm = current->mm; | |
218 | ||
219 | down_write(&mm->mmap_sem); | |
220 | ||
221 | if (brk < mm->end_code) | |
222 | goto out; | |
223 | newbrk = PAGE_ALIGN(brk); | |
224 | oldbrk = PAGE_ALIGN(mm->brk); | |
225 | if (oldbrk == newbrk) | |
226 | goto set_brk; | |
227 | ||
228 | /* Always allow shrinking brk. */ | |
229 | if (brk <= mm->brk) { | |
230 | if (!do_munmap(mm, newbrk, oldbrk-newbrk)) | |
231 | goto set_brk; | |
232 | goto out; | |
233 | } | |
234 | ||
235 | /* Check against rlimit.. */ | |
236 | rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur; | |
237 | if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim) | |
238 | goto out; | |
239 | ||
240 | /* Check against existing mmap mappings. */ | |
241 | if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) | |
242 | goto out; | |
243 | ||
244 | /* Ok, looks good - let it rip. */ | |
245 | if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) | |
246 | goto out; | |
247 | set_brk: | |
248 | mm->brk = brk; | |
249 | out: | |
250 | retval = mm->brk; | |
251 | up_write(&mm->mmap_sem); | |
252 | return retval; | |
253 | } | |
254 | ||
255 | #ifdef DEBUG_MM_RB | |
256 | static int browse_rb(struct rb_root *root) | |
257 | { | |
258 | int i = 0, j; | |
259 | struct rb_node *nd, *pn = NULL; | |
260 | unsigned long prev = 0, pend = 0; | |
261 | ||
262 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
263 | struct vm_area_struct *vma; | |
264 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
265 | if (vma->vm_start < prev) | |
266 | printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1; | |
267 | if (vma->vm_start < pend) | |
268 | printk("vm_start %lx pend %lx\n", vma->vm_start, pend); | |
269 | if (vma->vm_start > vma->vm_end) | |
270 | printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start); | |
271 | i++; | |
272 | pn = nd; | |
273 | } | |
274 | j = 0; | |
275 | for (nd = pn; nd; nd = rb_prev(nd)) { | |
276 | j++; | |
277 | } | |
278 | if (i != j) | |
279 | printk("backwards %d, forwards %d\n", j, i), i = 0; | |
280 | return i; | |
281 | } | |
282 | ||
283 | void validate_mm(struct mm_struct *mm) | |
284 | { | |
285 | int bug = 0; | |
286 | int i = 0; | |
287 | struct vm_area_struct *tmp = mm->mmap; | |
288 | while (tmp) { | |
289 | tmp = tmp->vm_next; | |
290 | i++; | |
291 | } | |
292 | if (i != mm->map_count) | |
293 | printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1; | |
294 | i = browse_rb(&mm->mm_rb); | |
295 | if (i != mm->map_count) | |
296 | printk("map_count %d rb %d\n", mm->map_count, i), bug = 1; | |
46a350ef | 297 | BUG_ON(bug); |
1da177e4 LT |
298 | } |
299 | #else | |
300 | #define validate_mm(mm) do { } while (0) | |
301 | #endif | |
302 | ||
303 | static struct vm_area_struct * | |
304 | find_vma_prepare(struct mm_struct *mm, unsigned long addr, | |
305 | struct vm_area_struct **pprev, struct rb_node ***rb_link, | |
306 | struct rb_node ** rb_parent) | |
307 | { | |
308 | struct vm_area_struct * vma; | |
309 | struct rb_node ** __rb_link, * __rb_parent, * rb_prev; | |
310 | ||
311 | __rb_link = &mm->mm_rb.rb_node; | |
312 | rb_prev = __rb_parent = NULL; | |
313 | vma = NULL; | |
314 | ||
315 | while (*__rb_link) { | |
316 | struct vm_area_struct *vma_tmp; | |
317 | ||
318 | __rb_parent = *__rb_link; | |
319 | vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); | |
320 | ||
321 | if (vma_tmp->vm_end > addr) { | |
322 | vma = vma_tmp; | |
323 | if (vma_tmp->vm_start <= addr) | |
324 | return vma; | |
325 | __rb_link = &__rb_parent->rb_left; | |
326 | } else { | |
327 | rb_prev = __rb_parent; | |
328 | __rb_link = &__rb_parent->rb_right; | |
329 | } | |
330 | } | |
331 | ||
332 | *pprev = NULL; | |
333 | if (rb_prev) | |
334 | *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); | |
335 | *rb_link = __rb_link; | |
336 | *rb_parent = __rb_parent; | |
337 | return vma; | |
338 | } | |
339 | ||
340 | static inline void | |
341 | __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma, | |
342 | struct vm_area_struct *prev, struct rb_node *rb_parent) | |
343 | { | |
344 | if (prev) { | |
345 | vma->vm_next = prev->vm_next; | |
346 | prev->vm_next = vma; | |
347 | } else { | |
348 | mm->mmap = vma; | |
349 | if (rb_parent) | |
350 | vma->vm_next = rb_entry(rb_parent, | |
351 | struct vm_area_struct, vm_rb); | |
352 | else | |
353 | vma->vm_next = NULL; | |
354 | } | |
355 | } | |
356 | ||
357 | void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, | |
358 | struct rb_node **rb_link, struct rb_node *rb_parent) | |
359 | { | |
360 | rb_link_node(&vma->vm_rb, rb_parent, rb_link); | |
361 | rb_insert_color(&vma->vm_rb, &mm->mm_rb); | |
362 | } | |
363 | ||
364 | static inline void __vma_link_file(struct vm_area_struct *vma) | |
365 | { | |
366 | struct file * file; | |
367 | ||
368 | file = vma->vm_file; | |
369 | if (file) { | |
370 | struct address_space *mapping = file->f_mapping; | |
371 | ||
372 | if (vma->vm_flags & VM_DENYWRITE) | |
373 | atomic_dec(&file->f_dentry->d_inode->i_writecount); | |
374 | if (vma->vm_flags & VM_SHARED) | |
375 | mapping->i_mmap_writable++; | |
376 | ||
377 | flush_dcache_mmap_lock(mapping); | |
378 | if (unlikely(vma->vm_flags & VM_NONLINEAR)) | |
379 | vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); | |
380 | else | |
381 | vma_prio_tree_insert(vma, &mapping->i_mmap); | |
382 | flush_dcache_mmap_unlock(mapping); | |
383 | } | |
384 | } | |
385 | ||
386 | static void | |
387 | __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
388 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
389 | struct rb_node *rb_parent) | |
390 | { | |
391 | __vma_link_list(mm, vma, prev, rb_parent); | |
392 | __vma_link_rb(mm, vma, rb_link, rb_parent); | |
393 | __anon_vma_link(vma); | |
394 | } | |
395 | ||
396 | static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
397 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
398 | struct rb_node *rb_parent) | |
399 | { | |
400 | struct address_space *mapping = NULL; | |
401 | ||
402 | if (vma->vm_file) | |
403 | mapping = vma->vm_file->f_mapping; | |
404 | ||
405 | if (mapping) { | |
406 | spin_lock(&mapping->i_mmap_lock); | |
407 | vma->vm_truncate_count = mapping->truncate_count; | |
408 | } | |
409 | anon_vma_lock(vma); | |
410 | ||
411 | __vma_link(mm, vma, prev, rb_link, rb_parent); | |
412 | __vma_link_file(vma); | |
413 | ||
414 | anon_vma_unlock(vma); | |
415 | if (mapping) | |
416 | spin_unlock(&mapping->i_mmap_lock); | |
417 | ||
418 | mm->map_count++; | |
419 | validate_mm(mm); | |
420 | } | |
421 | ||
422 | /* | |
423 | * Helper for vma_adjust in the split_vma insert case: | |
424 | * insert vm structure into list and rbtree and anon_vma, | |
425 | * but it has already been inserted into prio_tree earlier. | |
426 | */ | |
427 | static void | |
428 | __insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma) | |
429 | { | |
430 | struct vm_area_struct * __vma, * prev; | |
431 | struct rb_node ** rb_link, * rb_parent; | |
432 | ||
433 | __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent); | |
46a350ef | 434 | BUG_ON(__vma && __vma->vm_start < vma->vm_end); |
1da177e4 LT |
435 | __vma_link(mm, vma, prev, rb_link, rb_parent); |
436 | mm->map_count++; | |
437 | } | |
438 | ||
439 | static inline void | |
440 | __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, | |
441 | struct vm_area_struct *prev) | |
442 | { | |
443 | prev->vm_next = vma->vm_next; | |
444 | rb_erase(&vma->vm_rb, &mm->mm_rb); | |
445 | if (mm->mmap_cache == vma) | |
446 | mm->mmap_cache = prev; | |
447 | } | |
448 | ||
449 | /* | |
450 | * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that | |
451 | * is already present in an i_mmap tree without adjusting the tree. | |
452 | * The following helper function should be used when such adjustments | |
453 | * are necessary. The "insert" vma (if any) is to be inserted | |
454 | * before we drop the necessary locks. | |
455 | */ | |
456 | void vma_adjust(struct vm_area_struct *vma, unsigned long start, | |
457 | unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) | |
458 | { | |
459 | struct mm_struct *mm = vma->vm_mm; | |
460 | struct vm_area_struct *next = vma->vm_next; | |
461 | struct vm_area_struct *importer = NULL; | |
462 | struct address_space *mapping = NULL; | |
463 | struct prio_tree_root *root = NULL; | |
464 | struct file *file = vma->vm_file; | |
465 | struct anon_vma *anon_vma = NULL; | |
466 | long adjust_next = 0; | |
467 | int remove_next = 0; | |
468 | ||
469 | if (next && !insert) { | |
470 | if (end >= next->vm_end) { | |
471 | /* | |
472 | * vma expands, overlapping all the next, and | |
473 | * perhaps the one after too (mprotect case 6). | |
474 | */ | |
475 | again: remove_next = 1 + (end > next->vm_end); | |
476 | end = next->vm_end; | |
477 | anon_vma = next->anon_vma; | |
478 | importer = vma; | |
479 | } else if (end > next->vm_start) { | |
480 | /* | |
481 | * vma expands, overlapping part of the next: | |
482 | * mprotect case 5 shifting the boundary up. | |
483 | */ | |
484 | adjust_next = (end - next->vm_start) >> PAGE_SHIFT; | |
485 | anon_vma = next->anon_vma; | |
486 | importer = vma; | |
487 | } else if (end < vma->vm_end) { | |
488 | /* | |
489 | * vma shrinks, and !insert tells it's not | |
490 | * split_vma inserting another: so it must be | |
491 | * mprotect case 4 shifting the boundary down. | |
492 | */ | |
493 | adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT); | |
494 | anon_vma = next->anon_vma; | |
495 | importer = next; | |
496 | } | |
497 | } | |
498 | ||
499 | if (file) { | |
500 | mapping = file->f_mapping; | |
501 | if (!(vma->vm_flags & VM_NONLINEAR)) | |
502 | root = &mapping->i_mmap; | |
503 | spin_lock(&mapping->i_mmap_lock); | |
504 | if (importer && | |
505 | vma->vm_truncate_count != next->vm_truncate_count) { | |
506 | /* | |
507 | * unmap_mapping_range might be in progress: | |
508 | * ensure that the expanding vma is rescanned. | |
509 | */ | |
510 | importer->vm_truncate_count = 0; | |
511 | } | |
512 | if (insert) { | |
513 | insert->vm_truncate_count = vma->vm_truncate_count; | |
514 | /* | |
515 | * Put into prio_tree now, so instantiated pages | |
516 | * are visible to arm/parisc __flush_dcache_page | |
517 | * throughout; but we cannot insert into address | |
518 | * space until vma start or end is updated. | |
519 | */ | |
520 | __vma_link_file(insert); | |
521 | } | |
522 | } | |
523 | ||
524 | /* | |
525 | * When changing only vma->vm_end, we don't really need | |
526 | * anon_vma lock: but is that case worth optimizing out? | |
527 | */ | |
528 | if (vma->anon_vma) | |
529 | anon_vma = vma->anon_vma; | |
530 | if (anon_vma) { | |
531 | spin_lock(&anon_vma->lock); | |
532 | /* | |
533 | * Easily overlooked: when mprotect shifts the boundary, | |
534 | * make sure the expanding vma has anon_vma set if the | |
535 | * shrinking vma had, to cover any anon pages imported. | |
536 | */ | |
537 | if (importer && !importer->anon_vma) { | |
538 | importer->anon_vma = anon_vma; | |
539 | __anon_vma_link(importer); | |
540 | } | |
541 | } | |
542 | ||
543 | if (root) { | |
544 | flush_dcache_mmap_lock(mapping); | |
545 | vma_prio_tree_remove(vma, root); | |
546 | if (adjust_next) | |
547 | vma_prio_tree_remove(next, root); | |
548 | } | |
549 | ||
550 | vma->vm_start = start; | |
551 | vma->vm_end = end; | |
552 | vma->vm_pgoff = pgoff; | |
553 | if (adjust_next) { | |
554 | next->vm_start += adjust_next << PAGE_SHIFT; | |
555 | next->vm_pgoff += adjust_next; | |
556 | } | |
557 | ||
558 | if (root) { | |
559 | if (adjust_next) | |
560 | vma_prio_tree_insert(next, root); | |
561 | vma_prio_tree_insert(vma, root); | |
562 | flush_dcache_mmap_unlock(mapping); | |
563 | } | |
564 | ||
565 | if (remove_next) { | |
566 | /* | |
567 | * vma_merge has merged next into vma, and needs | |
568 | * us to remove next before dropping the locks. | |
569 | */ | |
570 | __vma_unlink(mm, next, vma); | |
571 | if (file) | |
572 | __remove_shared_vm_struct(next, file, mapping); | |
573 | if (next->anon_vma) | |
574 | __anon_vma_merge(vma, next); | |
575 | } else if (insert) { | |
576 | /* | |
577 | * split_vma has split insert from vma, and needs | |
578 | * us to insert it before dropping the locks | |
579 | * (it may either follow vma or precede it). | |
580 | */ | |
581 | __insert_vm_struct(mm, insert); | |
582 | } | |
583 | ||
584 | if (anon_vma) | |
585 | spin_unlock(&anon_vma->lock); | |
586 | if (mapping) | |
587 | spin_unlock(&mapping->i_mmap_lock); | |
588 | ||
589 | if (remove_next) { | |
590 | if (file) | |
591 | fput(file); | |
592 | mm->map_count--; | |
593 | mpol_free(vma_policy(next)); | |
594 | kmem_cache_free(vm_area_cachep, next); | |
595 | /* | |
596 | * In mprotect's case 6 (see comments on vma_merge), | |
597 | * we must remove another next too. It would clutter | |
598 | * up the code too much to do both in one go. | |
599 | */ | |
600 | if (remove_next == 2) { | |
601 | next = vma->vm_next; | |
602 | goto again; | |
603 | } | |
604 | } | |
605 | ||
606 | validate_mm(mm); | |
607 | } | |
608 | ||
609 | /* | |
610 | * If the vma has a ->close operation then the driver probably needs to release | |
611 | * per-vma resources, so we don't attempt to merge those. | |
612 | */ | |
a6f563db | 613 | #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP) |
1da177e4 LT |
614 | |
615 | static inline int is_mergeable_vma(struct vm_area_struct *vma, | |
616 | struct file *file, unsigned long vm_flags) | |
617 | { | |
618 | if (vma->vm_flags != vm_flags) | |
619 | return 0; | |
620 | if (vma->vm_file != file) | |
621 | return 0; | |
622 | if (vma->vm_ops && vma->vm_ops->close) | |
623 | return 0; | |
624 | return 1; | |
625 | } | |
626 | ||
627 | static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, | |
628 | struct anon_vma *anon_vma2) | |
629 | { | |
630 | return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2); | |
631 | } | |
632 | ||
633 | /* | |
634 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
635 | * in front of (at a lower virtual address and file offset than) the vma. | |
636 | * | |
637 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
638 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
639 | * | |
640 | * We don't check here for the merged mmap wrapping around the end of pagecache | |
641 | * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which | |
642 | * wrap, nor mmaps which cover the final page at index -1UL. | |
643 | */ | |
644 | static int | |
645 | can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, | |
646 | struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) | |
647 | { | |
648 | if (is_mergeable_vma(vma, file, vm_flags) && | |
649 | is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { | |
650 | if (vma->vm_pgoff == vm_pgoff) | |
651 | return 1; | |
652 | } | |
653 | return 0; | |
654 | } | |
655 | ||
656 | /* | |
657 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
658 | * beyond (at a higher virtual address and file offset than) the vma. | |
659 | * | |
660 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
661 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
662 | */ | |
663 | static int | |
664 | can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, | |
665 | struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) | |
666 | { | |
667 | if (is_mergeable_vma(vma, file, vm_flags) && | |
668 | is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { | |
669 | pgoff_t vm_pglen; | |
670 | vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; | |
671 | if (vma->vm_pgoff + vm_pglen == vm_pgoff) | |
672 | return 1; | |
673 | } | |
674 | return 0; | |
675 | } | |
676 | ||
677 | /* | |
678 | * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out | |
679 | * whether that can be merged with its predecessor or its successor. | |
680 | * Or both (it neatly fills a hole). | |
681 | * | |
682 | * In most cases - when called for mmap, brk or mremap - [addr,end) is | |
683 | * certain not to be mapped by the time vma_merge is called; but when | |
684 | * called for mprotect, it is certain to be already mapped (either at | |
685 | * an offset within prev, or at the start of next), and the flags of | |
686 | * this area are about to be changed to vm_flags - and the no-change | |
687 | * case has already been eliminated. | |
688 | * | |
689 | * The following mprotect cases have to be considered, where AAAA is | |
690 | * the area passed down from mprotect_fixup, never extending beyond one | |
691 | * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: | |
692 | * | |
693 | * AAAA AAAA AAAA AAAA | |
694 | * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX | |
695 | * cannot merge might become might become might become | |
696 | * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or | |
697 | * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or | |
698 | * mremap move: PPPPNNNNNNNN 8 | |
699 | * AAAA | |
700 | * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN | |
701 | * might become case 1 below case 2 below case 3 below | |
702 | * | |
703 | * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: | |
704 | * mprotect_fixup updates vm_flags & vm_page_prot on successful return. | |
705 | */ | |
706 | struct vm_area_struct *vma_merge(struct mm_struct *mm, | |
707 | struct vm_area_struct *prev, unsigned long addr, | |
708 | unsigned long end, unsigned long vm_flags, | |
709 | struct anon_vma *anon_vma, struct file *file, | |
710 | pgoff_t pgoff, struct mempolicy *policy) | |
711 | { | |
712 | pgoff_t pglen = (end - addr) >> PAGE_SHIFT; | |
713 | struct vm_area_struct *area, *next; | |
714 | ||
715 | /* | |
716 | * We later require that vma->vm_flags == vm_flags, | |
717 | * so this tests vma->vm_flags & VM_SPECIAL, too. | |
718 | */ | |
719 | if (vm_flags & VM_SPECIAL) | |
720 | return NULL; | |
721 | ||
722 | if (prev) | |
723 | next = prev->vm_next; | |
724 | else | |
725 | next = mm->mmap; | |
726 | area = next; | |
727 | if (next && next->vm_end == end) /* cases 6, 7, 8 */ | |
728 | next = next->vm_next; | |
729 | ||
730 | /* | |
731 | * Can it merge with the predecessor? | |
732 | */ | |
733 | if (prev && prev->vm_end == addr && | |
734 | mpol_equal(vma_policy(prev), policy) && | |
735 | can_vma_merge_after(prev, vm_flags, | |
736 | anon_vma, file, pgoff)) { | |
737 | /* | |
738 | * OK, it can. Can we now merge in the successor as well? | |
739 | */ | |
740 | if (next && end == next->vm_start && | |
741 | mpol_equal(policy, vma_policy(next)) && | |
742 | can_vma_merge_before(next, vm_flags, | |
743 | anon_vma, file, pgoff+pglen) && | |
744 | is_mergeable_anon_vma(prev->anon_vma, | |
745 | next->anon_vma)) { | |
746 | /* cases 1, 6 */ | |
747 | vma_adjust(prev, prev->vm_start, | |
748 | next->vm_end, prev->vm_pgoff, NULL); | |
749 | } else /* cases 2, 5, 7 */ | |
750 | vma_adjust(prev, prev->vm_start, | |
751 | end, prev->vm_pgoff, NULL); | |
752 | return prev; | |
753 | } | |
754 | ||
755 | /* | |
756 | * Can this new request be merged in front of next? | |
757 | */ | |
758 | if (next && end == next->vm_start && | |
759 | mpol_equal(policy, vma_policy(next)) && | |
760 | can_vma_merge_before(next, vm_flags, | |
761 | anon_vma, file, pgoff+pglen)) { | |
762 | if (prev && addr < prev->vm_end) /* case 4 */ | |
763 | vma_adjust(prev, prev->vm_start, | |
764 | addr, prev->vm_pgoff, NULL); | |
765 | else /* cases 3, 8 */ | |
766 | vma_adjust(area, addr, next->vm_end, | |
767 | next->vm_pgoff - pglen, NULL); | |
768 | return area; | |
769 | } | |
770 | ||
771 | return NULL; | |
772 | } | |
773 | ||
774 | /* | |
775 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check | |
776 | * neighbouring vmas for a suitable anon_vma, before it goes off | |
777 | * to allocate a new anon_vma. It checks because a repetitive | |
778 | * sequence of mprotects and faults may otherwise lead to distinct | |
779 | * anon_vmas being allocated, preventing vma merge in subsequent | |
780 | * mprotect. | |
781 | */ | |
782 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) | |
783 | { | |
784 | struct vm_area_struct *near; | |
785 | unsigned long vm_flags; | |
786 | ||
787 | near = vma->vm_next; | |
788 | if (!near) | |
789 | goto try_prev; | |
790 | ||
791 | /* | |
792 | * Since only mprotect tries to remerge vmas, match flags | |
793 | * which might be mprotected into each other later on. | |
794 | * Neither mlock nor madvise tries to remerge at present, | |
795 | * so leave their flags as obstructing a merge. | |
796 | */ | |
797 | vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC); | |
798 | vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC); | |
799 | ||
800 | if (near->anon_vma && vma->vm_end == near->vm_start && | |
801 | mpol_equal(vma_policy(vma), vma_policy(near)) && | |
802 | can_vma_merge_before(near, vm_flags, | |
803 | NULL, vma->vm_file, vma->vm_pgoff + | |
804 | ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT))) | |
805 | return near->anon_vma; | |
806 | try_prev: | |
807 | /* | |
808 | * It is potentially slow to have to call find_vma_prev here. | |
809 | * But it's only on the first write fault on the vma, not | |
810 | * every time, and we could devise a way to avoid it later | |
811 | * (e.g. stash info in next's anon_vma_node when assigning | |
812 | * an anon_vma, or when trying vma_merge). Another time. | |
813 | */ | |
46a350ef | 814 | BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma); |
1da177e4 LT |
815 | if (!near) |
816 | goto none; | |
817 | ||
818 | vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC); | |
819 | vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC); | |
820 | ||
821 | if (near->anon_vma && near->vm_end == vma->vm_start && | |
822 | mpol_equal(vma_policy(near), vma_policy(vma)) && | |
823 | can_vma_merge_after(near, vm_flags, | |
824 | NULL, vma->vm_file, vma->vm_pgoff)) | |
825 | return near->anon_vma; | |
826 | none: | |
827 | /* | |
828 | * There's no absolute need to look only at touching neighbours: | |
829 | * we could search further afield for "compatible" anon_vmas. | |
830 | * But it would probably just be a waste of time searching, | |
831 | * or lead to too many vmas hanging off the same anon_vma. | |
832 | * We're trying to allow mprotect remerging later on, | |
833 | * not trying to minimize memory used for anon_vmas. | |
834 | */ | |
835 | return NULL; | |
836 | } | |
837 | ||
838 | #ifdef CONFIG_PROC_FS | |
ab50b8ed | 839 | void vm_stat_account(struct mm_struct *mm, unsigned long flags, |
1da177e4 LT |
840 | struct file *file, long pages) |
841 | { | |
842 | const unsigned long stack_flags | |
843 | = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); | |
844 | ||
1da177e4 LT |
845 | if (file) { |
846 | mm->shared_vm += pages; | |
847 | if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) | |
848 | mm->exec_vm += pages; | |
849 | } else if (flags & stack_flags) | |
850 | mm->stack_vm += pages; | |
851 | if (flags & (VM_RESERVED|VM_IO)) | |
852 | mm->reserved_vm += pages; | |
853 | } | |
854 | #endif /* CONFIG_PROC_FS */ | |
855 | ||
856 | /* | |
857 | * The caller must hold down_write(current->mm->mmap_sem). | |
858 | */ | |
859 | ||
860 | unsigned long do_mmap_pgoff(struct file * file, unsigned long addr, | |
861 | unsigned long len, unsigned long prot, | |
862 | unsigned long flags, unsigned long pgoff) | |
863 | { | |
864 | struct mm_struct * mm = current->mm; | |
865 | struct vm_area_struct * vma, * prev; | |
866 | struct inode *inode; | |
867 | unsigned int vm_flags; | |
868 | int correct_wcount = 0; | |
869 | int error; | |
870 | struct rb_node ** rb_link, * rb_parent; | |
871 | int accountable = 1; | |
872 | unsigned long charged = 0, reqprot = prot; | |
873 | ||
874 | if (file) { | |
875 | if (is_file_hugepages(file)) | |
876 | accountable = 0; | |
877 | ||
878 | if (!file->f_op || !file->f_op->mmap) | |
879 | return -ENODEV; | |
880 | ||
881 | if ((prot & PROT_EXEC) && | |
882 | (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)) | |
883 | return -EPERM; | |
884 | } | |
885 | /* | |
886 | * Does the application expect PROT_READ to imply PROT_EXEC? | |
887 | * | |
888 | * (the exception is when the underlying filesystem is noexec | |
889 | * mounted, in which case we dont add PROT_EXEC.) | |
890 | */ | |
891 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
892 | if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))) | |
893 | prot |= PROT_EXEC; | |
894 | ||
895 | if (!len) | |
896 | return -EINVAL; | |
897 | ||
898 | /* Careful about overflows.. */ | |
899 | len = PAGE_ALIGN(len); | |
900 | if (!len || len > TASK_SIZE) | |
901 | return -ENOMEM; | |
902 | ||
903 | /* offset overflow? */ | |
904 | if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) | |
905 | return -EOVERFLOW; | |
906 | ||
907 | /* Too many mappings? */ | |
908 | if (mm->map_count > sysctl_max_map_count) | |
909 | return -ENOMEM; | |
910 | ||
911 | /* Obtain the address to map to. we verify (or select) it and ensure | |
912 | * that it represents a valid section of the address space. | |
913 | */ | |
914 | addr = get_unmapped_area(file, addr, len, pgoff, flags); | |
915 | if (addr & ~PAGE_MASK) | |
916 | return addr; | |
917 | ||
918 | /* Do simple checking here so the lower-level routines won't have | |
919 | * to. we assume access permissions have been handled by the open | |
920 | * of the memory object, so we don't do any here. | |
921 | */ | |
922 | vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | | |
923 | mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; | |
924 | ||
925 | if (flags & MAP_LOCKED) { | |
926 | if (!can_do_mlock()) | |
927 | return -EPERM; | |
928 | vm_flags |= VM_LOCKED; | |
929 | } | |
930 | /* mlock MCL_FUTURE? */ | |
931 | if (vm_flags & VM_LOCKED) { | |
932 | unsigned long locked, lock_limit; | |
93ea1d0a CW |
933 | locked = len >> PAGE_SHIFT; |
934 | locked += mm->locked_vm; | |
1da177e4 | 935 | lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; |
93ea1d0a | 936 | lock_limit >>= PAGE_SHIFT; |
1da177e4 LT |
937 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) |
938 | return -EAGAIN; | |
939 | } | |
940 | ||
941 | inode = file ? file->f_dentry->d_inode : NULL; | |
942 | ||
943 | if (file) { | |
944 | switch (flags & MAP_TYPE) { | |
945 | case MAP_SHARED: | |
946 | if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) | |
947 | return -EACCES; | |
948 | ||
949 | /* | |
950 | * Make sure we don't allow writing to an append-only | |
951 | * file.. | |
952 | */ | |
953 | if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) | |
954 | return -EACCES; | |
955 | ||
956 | /* | |
957 | * Make sure there are no mandatory locks on the file. | |
958 | */ | |
959 | if (locks_verify_locked(inode)) | |
960 | return -EAGAIN; | |
961 | ||
962 | vm_flags |= VM_SHARED | VM_MAYSHARE; | |
963 | if (!(file->f_mode & FMODE_WRITE)) | |
964 | vm_flags &= ~(VM_MAYWRITE | VM_SHARED); | |
965 | ||
966 | /* fall through */ | |
967 | case MAP_PRIVATE: | |
968 | if (!(file->f_mode & FMODE_READ)) | |
969 | return -EACCES; | |
970 | break; | |
971 | ||
972 | default: | |
973 | return -EINVAL; | |
974 | } | |
975 | } else { | |
976 | switch (flags & MAP_TYPE) { | |
977 | case MAP_SHARED: | |
978 | vm_flags |= VM_SHARED | VM_MAYSHARE; | |
979 | break; | |
980 | case MAP_PRIVATE: | |
981 | /* | |
982 | * Set pgoff according to addr for anon_vma. | |
983 | */ | |
984 | pgoff = addr >> PAGE_SHIFT; | |
985 | break; | |
986 | default: | |
987 | return -EINVAL; | |
988 | } | |
989 | } | |
990 | ||
991 | error = security_file_mmap(file, reqprot, prot, flags); | |
992 | if (error) | |
993 | return error; | |
994 | ||
995 | /* Clear old maps */ | |
996 | error = -ENOMEM; | |
997 | munmap_back: | |
998 | vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); | |
999 | if (vma && vma->vm_start < addr + len) { | |
1000 | if (do_munmap(mm, addr, len)) | |
1001 | return -ENOMEM; | |
1002 | goto munmap_back; | |
1003 | } | |
1004 | ||
1005 | /* Check against address space limit. */ | |
119f657c | 1006 | if (!may_expand_vm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
1007 | return -ENOMEM; |
1008 | ||
1009 | if (accountable && (!(flags & MAP_NORESERVE) || | |
1010 | sysctl_overcommit_memory == OVERCOMMIT_NEVER)) { | |
1011 | if (vm_flags & VM_SHARED) { | |
1012 | /* Check memory availability in shmem_file_setup? */ | |
1013 | vm_flags |= VM_ACCOUNT; | |
1014 | } else if (vm_flags & VM_WRITE) { | |
1015 | /* | |
1016 | * Private writable mapping: check memory availability | |
1017 | */ | |
1018 | charged = len >> PAGE_SHIFT; | |
1019 | if (security_vm_enough_memory(charged)) | |
1020 | return -ENOMEM; | |
1021 | vm_flags |= VM_ACCOUNT; | |
1022 | } | |
1023 | } | |
1024 | ||
1025 | /* | |
1026 | * Can we just expand an old private anonymous mapping? | |
1027 | * The VM_SHARED test is necessary because shmem_zero_setup | |
1028 | * will create the file object for a shared anonymous map below. | |
1029 | */ | |
1030 | if (!file && !(vm_flags & VM_SHARED) && | |
1031 | vma_merge(mm, prev, addr, addr + len, vm_flags, | |
1032 | NULL, NULL, pgoff, NULL)) | |
1033 | goto out; | |
1034 | ||
1035 | /* | |
1036 | * Determine the object being mapped and call the appropriate | |
1037 | * specific mapper. the address has already been validated, but | |
1038 | * not unmapped, but the maps are removed from the list. | |
1039 | */ | |
c5e3b83e | 1040 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
1041 | if (!vma) { |
1042 | error = -ENOMEM; | |
1043 | goto unacct_error; | |
1044 | } | |
1da177e4 LT |
1045 | |
1046 | vma->vm_mm = mm; | |
1047 | vma->vm_start = addr; | |
1048 | vma->vm_end = addr + len; | |
1049 | vma->vm_flags = vm_flags; | |
1050 | vma->vm_page_prot = protection_map[vm_flags & 0x0f]; | |
1051 | vma->vm_pgoff = pgoff; | |
1052 | ||
1053 | if (file) { | |
1054 | error = -EINVAL; | |
1055 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) | |
1056 | goto free_vma; | |
1057 | if (vm_flags & VM_DENYWRITE) { | |
1058 | error = deny_write_access(file); | |
1059 | if (error) | |
1060 | goto free_vma; | |
1061 | correct_wcount = 1; | |
1062 | } | |
1063 | vma->vm_file = file; | |
1064 | get_file(file); | |
1065 | error = file->f_op->mmap(file, vma); | |
1066 | if (error) | |
1067 | goto unmap_and_free_vma; | |
1068 | } else if (vm_flags & VM_SHARED) { | |
1069 | error = shmem_zero_setup(vma); | |
1070 | if (error) | |
1071 | goto free_vma; | |
1072 | } | |
1073 | ||
1074 | /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform | |
1075 | * shmem_zero_setup (perhaps called through /dev/zero's ->mmap) | |
1076 | * that memory reservation must be checked; but that reservation | |
1077 | * belongs to shared memory object, not to vma: so now clear it. | |
1078 | */ | |
1079 | if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT)) | |
1080 | vma->vm_flags &= ~VM_ACCOUNT; | |
1081 | ||
1082 | /* Can addr have changed?? | |
1083 | * | |
1084 | * Answer: Yes, several device drivers can do it in their | |
1085 | * f_op->mmap method. -DaveM | |
1086 | */ | |
1087 | addr = vma->vm_start; | |
1088 | pgoff = vma->vm_pgoff; | |
1089 | vm_flags = vma->vm_flags; | |
1090 | ||
1091 | if (!file || !vma_merge(mm, prev, addr, vma->vm_end, | |
1092 | vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) { | |
1093 | file = vma->vm_file; | |
1094 | vma_link(mm, vma, prev, rb_link, rb_parent); | |
1095 | if (correct_wcount) | |
1096 | atomic_inc(&inode->i_writecount); | |
1097 | } else { | |
1098 | if (file) { | |
1099 | if (correct_wcount) | |
1100 | atomic_inc(&inode->i_writecount); | |
1101 | fput(file); | |
1102 | } | |
1103 | mpol_free(vma_policy(vma)); | |
1104 | kmem_cache_free(vm_area_cachep, vma); | |
1105 | } | |
1106 | out: | |
1107 | mm->total_vm += len >> PAGE_SHIFT; | |
ab50b8ed | 1108 | vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); |
1da177e4 LT |
1109 | if (vm_flags & VM_LOCKED) { |
1110 | mm->locked_vm += len >> PAGE_SHIFT; | |
1111 | make_pages_present(addr, addr + len); | |
1112 | } | |
1113 | if (flags & MAP_POPULATE) { | |
1114 | up_write(&mm->mmap_sem); | |
1115 | sys_remap_file_pages(addr, len, 0, | |
1116 | pgoff, flags & MAP_NONBLOCK); | |
1117 | down_write(&mm->mmap_sem); | |
1118 | } | |
1119 | return addr; | |
1120 | ||
1121 | unmap_and_free_vma: | |
1122 | if (correct_wcount) | |
1123 | atomic_inc(&inode->i_writecount); | |
1124 | vma->vm_file = NULL; | |
1125 | fput(file); | |
1126 | ||
1127 | /* Undo any partial mapping done by a device driver. */ | |
e0da382c HD |
1128 | unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); |
1129 | charged = 0; | |
1da177e4 LT |
1130 | free_vma: |
1131 | kmem_cache_free(vm_area_cachep, vma); | |
1132 | unacct_error: | |
1133 | if (charged) | |
1134 | vm_unacct_memory(charged); | |
1135 | return error; | |
1136 | } | |
1137 | ||
1138 | EXPORT_SYMBOL(do_mmap_pgoff); | |
1139 | ||
1140 | /* Get an address range which is currently unmapped. | |
1141 | * For shmat() with addr=0. | |
1142 | * | |
1143 | * Ugly calling convention alert: | |
1144 | * Return value with the low bits set means error value, | |
1145 | * ie | |
1146 | * if (ret & ~PAGE_MASK) | |
1147 | * error = ret; | |
1148 | * | |
1149 | * This function "knows" that -ENOMEM has the bits set. | |
1150 | */ | |
1151 | #ifndef HAVE_ARCH_UNMAPPED_AREA | |
1152 | unsigned long | |
1153 | arch_get_unmapped_area(struct file *filp, unsigned long addr, | |
1154 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1155 | { | |
1156 | struct mm_struct *mm = current->mm; | |
1157 | struct vm_area_struct *vma; | |
1158 | unsigned long start_addr; | |
1159 | ||
1160 | if (len > TASK_SIZE) | |
1161 | return -ENOMEM; | |
1162 | ||
1163 | if (addr) { | |
1164 | addr = PAGE_ALIGN(addr); | |
1165 | vma = find_vma(mm, addr); | |
1166 | if (TASK_SIZE - len >= addr && | |
1167 | (!vma || addr + len <= vma->vm_start)) | |
1168 | return addr; | |
1169 | } | |
1363c3cd WW |
1170 | if (len > mm->cached_hole_size) { |
1171 | start_addr = addr = mm->free_area_cache; | |
1172 | } else { | |
1173 | start_addr = addr = TASK_UNMAPPED_BASE; | |
1174 | mm->cached_hole_size = 0; | |
1175 | } | |
1da177e4 LT |
1176 | |
1177 | full_search: | |
1178 | for (vma = find_vma(mm, addr); ; vma = vma->vm_next) { | |
1179 | /* At this point: (!vma || addr < vma->vm_end). */ | |
1180 | if (TASK_SIZE - len < addr) { | |
1181 | /* | |
1182 | * Start a new search - just in case we missed | |
1183 | * some holes. | |
1184 | */ | |
1185 | if (start_addr != TASK_UNMAPPED_BASE) { | |
1363c3cd WW |
1186 | addr = TASK_UNMAPPED_BASE; |
1187 | start_addr = addr; | |
1188 | mm->cached_hole_size = 0; | |
1da177e4 LT |
1189 | goto full_search; |
1190 | } | |
1191 | return -ENOMEM; | |
1192 | } | |
1193 | if (!vma || addr + len <= vma->vm_start) { | |
1194 | /* | |
1195 | * Remember the place where we stopped the search: | |
1196 | */ | |
1197 | mm->free_area_cache = addr + len; | |
1198 | return addr; | |
1199 | } | |
1363c3cd WW |
1200 | if (addr + mm->cached_hole_size < vma->vm_start) |
1201 | mm->cached_hole_size = vma->vm_start - addr; | |
1da177e4 LT |
1202 | addr = vma->vm_end; |
1203 | } | |
1204 | } | |
1205 | #endif | |
1206 | ||
1363c3cd | 1207 | void arch_unmap_area(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
1208 | { |
1209 | /* | |
1210 | * Is this a new hole at the lowest possible address? | |
1211 | */ | |
1363c3cd WW |
1212 | if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) { |
1213 | mm->free_area_cache = addr; | |
1214 | mm->cached_hole_size = ~0UL; | |
1215 | } | |
1da177e4 LT |
1216 | } |
1217 | ||
1218 | /* | |
1219 | * This mmap-allocator allocates new areas top-down from below the | |
1220 | * stack's low limit (the base): | |
1221 | */ | |
1222 | #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN | |
1223 | unsigned long | |
1224 | arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, | |
1225 | const unsigned long len, const unsigned long pgoff, | |
1226 | const unsigned long flags) | |
1227 | { | |
1228 | struct vm_area_struct *vma; | |
1229 | struct mm_struct *mm = current->mm; | |
1230 | unsigned long addr = addr0; | |
1231 | ||
1232 | /* requested length too big for entire address space */ | |
1233 | if (len > TASK_SIZE) | |
1234 | return -ENOMEM; | |
1235 | ||
1236 | /* requesting a specific address */ | |
1237 | if (addr) { | |
1238 | addr = PAGE_ALIGN(addr); | |
1239 | vma = find_vma(mm, addr); | |
1240 | if (TASK_SIZE - len >= addr && | |
1241 | (!vma || addr + len <= vma->vm_start)) | |
1242 | return addr; | |
1243 | } | |
1244 | ||
1363c3cd WW |
1245 | /* check if free_area_cache is useful for us */ |
1246 | if (len <= mm->cached_hole_size) { | |
1247 | mm->cached_hole_size = 0; | |
1248 | mm->free_area_cache = mm->mmap_base; | |
1249 | } | |
1250 | ||
1da177e4 LT |
1251 | /* either no address requested or can't fit in requested address hole */ |
1252 | addr = mm->free_area_cache; | |
1253 | ||
1254 | /* make sure it can fit in the remaining address space */ | |
49a43876 | 1255 | if (addr > len) { |
1da177e4 LT |
1256 | vma = find_vma(mm, addr-len); |
1257 | if (!vma || addr <= vma->vm_start) | |
1258 | /* remember the address as a hint for next time */ | |
1259 | return (mm->free_area_cache = addr-len); | |
1260 | } | |
1261 | ||
73219d17 CW |
1262 | if (mm->mmap_base < len) |
1263 | goto bottomup; | |
1264 | ||
1da177e4 LT |
1265 | addr = mm->mmap_base-len; |
1266 | ||
1267 | do { | |
1268 | /* | |
1269 | * Lookup failure means no vma is above this address, | |
1270 | * else if new region fits below vma->vm_start, | |
1271 | * return with success: | |
1272 | */ | |
1273 | vma = find_vma(mm, addr); | |
1274 | if (!vma || addr+len <= vma->vm_start) | |
1275 | /* remember the address as a hint for next time */ | |
1276 | return (mm->free_area_cache = addr); | |
1277 | ||
1363c3cd WW |
1278 | /* remember the largest hole we saw so far */ |
1279 | if (addr + mm->cached_hole_size < vma->vm_start) | |
1280 | mm->cached_hole_size = vma->vm_start - addr; | |
1281 | ||
1da177e4 LT |
1282 | /* try just below the current vma->vm_start */ |
1283 | addr = vma->vm_start-len; | |
49a43876 | 1284 | } while (len < vma->vm_start); |
1da177e4 | 1285 | |
73219d17 | 1286 | bottomup: |
1da177e4 LT |
1287 | /* |
1288 | * A failed mmap() very likely causes application failure, | |
1289 | * so fall back to the bottom-up function here. This scenario | |
1290 | * can happen with large stack limits and large mmap() | |
1291 | * allocations. | |
1292 | */ | |
1363c3cd WW |
1293 | mm->cached_hole_size = ~0UL; |
1294 | mm->free_area_cache = TASK_UNMAPPED_BASE; | |
1da177e4 LT |
1295 | addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags); |
1296 | /* | |
1297 | * Restore the topdown base: | |
1298 | */ | |
1299 | mm->free_area_cache = mm->mmap_base; | |
1363c3cd | 1300 | mm->cached_hole_size = ~0UL; |
1da177e4 LT |
1301 | |
1302 | return addr; | |
1303 | } | |
1304 | #endif | |
1305 | ||
1363c3cd | 1306 | void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
1307 | { |
1308 | /* | |
1309 | * Is this a new hole at the highest possible address? | |
1310 | */ | |
1363c3cd WW |
1311 | if (addr > mm->free_area_cache) |
1312 | mm->free_area_cache = addr; | |
1da177e4 LT |
1313 | |
1314 | /* dont allow allocations above current base */ | |
1363c3cd WW |
1315 | if (mm->free_area_cache > mm->mmap_base) |
1316 | mm->free_area_cache = mm->mmap_base; | |
1da177e4 LT |
1317 | } |
1318 | ||
1319 | unsigned long | |
1320 | get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, | |
1321 | unsigned long pgoff, unsigned long flags) | |
1322 | { | |
07ab67c8 | 1323 | unsigned long ret; |
1da177e4 | 1324 | |
07ab67c8 LT |
1325 | if (!(flags & MAP_FIXED)) { |
1326 | unsigned long (*get_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long); | |
1da177e4 | 1327 | |
07ab67c8 LT |
1328 | get_area = current->mm->get_unmapped_area; |
1329 | if (file && file->f_op && file->f_op->get_unmapped_area) | |
1330 | get_area = file->f_op->get_unmapped_area; | |
1331 | addr = get_area(file, addr, len, pgoff, flags); | |
1332 | if (IS_ERR_VALUE(addr)) | |
1333 | return addr; | |
1334 | } | |
1da177e4 | 1335 | |
07ab67c8 LT |
1336 | if (addr > TASK_SIZE - len) |
1337 | return -ENOMEM; | |
1338 | if (addr & ~PAGE_MASK) | |
1339 | return -EINVAL; | |
1340 | if (file && is_file_hugepages(file)) { | |
1341 | /* | |
1342 | * Check if the given range is hugepage aligned, and | |
1343 | * can be made suitable for hugepages. | |
1344 | */ | |
1345 | ret = prepare_hugepage_range(addr, len); | |
1346 | } else { | |
1347 | /* | |
1348 | * Ensure that a normal request is not falling in a | |
1349 | * reserved hugepage range. For some archs like IA-64, | |
1350 | * there is a separate region for hugepages. | |
1351 | */ | |
1352 | ret = is_hugepage_only_range(current->mm, addr, len); | |
1353 | } | |
1354 | if (ret) | |
1355 | return -EINVAL; | |
1356 | return addr; | |
1da177e4 LT |
1357 | } |
1358 | ||
1359 | EXPORT_SYMBOL(get_unmapped_area); | |
1360 | ||
1361 | /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ | |
1362 | struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr) | |
1363 | { | |
1364 | struct vm_area_struct *vma = NULL; | |
1365 | ||
1366 | if (mm) { | |
1367 | /* Check the cache first. */ | |
1368 | /* (Cache hit rate is typically around 35%.) */ | |
1369 | vma = mm->mmap_cache; | |
1370 | if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) { | |
1371 | struct rb_node * rb_node; | |
1372 | ||
1373 | rb_node = mm->mm_rb.rb_node; | |
1374 | vma = NULL; | |
1375 | ||
1376 | while (rb_node) { | |
1377 | struct vm_area_struct * vma_tmp; | |
1378 | ||
1379 | vma_tmp = rb_entry(rb_node, | |
1380 | struct vm_area_struct, vm_rb); | |
1381 | ||
1382 | if (vma_tmp->vm_end > addr) { | |
1383 | vma = vma_tmp; | |
1384 | if (vma_tmp->vm_start <= addr) | |
1385 | break; | |
1386 | rb_node = rb_node->rb_left; | |
1387 | } else | |
1388 | rb_node = rb_node->rb_right; | |
1389 | } | |
1390 | if (vma) | |
1391 | mm->mmap_cache = vma; | |
1392 | } | |
1393 | } | |
1394 | return vma; | |
1395 | } | |
1396 | ||
1397 | EXPORT_SYMBOL(find_vma); | |
1398 | ||
1399 | /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */ | |
1400 | struct vm_area_struct * | |
1401 | find_vma_prev(struct mm_struct *mm, unsigned long addr, | |
1402 | struct vm_area_struct **pprev) | |
1403 | { | |
1404 | struct vm_area_struct *vma = NULL, *prev = NULL; | |
1405 | struct rb_node * rb_node; | |
1406 | if (!mm) | |
1407 | goto out; | |
1408 | ||
1409 | /* Guard against addr being lower than the first VMA */ | |
1410 | vma = mm->mmap; | |
1411 | ||
1412 | /* Go through the RB tree quickly. */ | |
1413 | rb_node = mm->mm_rb.rb_node; | |
1414 | ||
1415 | while (rb_node) { | |
1416 | struct vm_area_struct *vma_tmp; | |
1417 | vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
1418 | ||
1419 | if (addr < vma_tmp->vm_end) { | |
1420 | rb_node = rb_node->rb_left; | |
1421 | } else { | |
1422 | prev = vma_tmp; | |
1423 | if (!prev->vm_next || (addr < prev->vm_next->vm_end)) | |
1424 | break; | |
1425 | rb_node = rb_node->rb_right; | |
1426 | } | |
1427 | } | |
1428 | ||
1429 | out: | |
1430 | *pprev = prev; | |
1431 | return prev ? prev->vm_next : vma; | |
1432 | } | |
1433 | ||
1434 | /* | |
1435 | * Verify that the stack growth is acceptable and | |
1436 | * update accounting. This is shared with both the | |
1437 | * grow-up and grow-down cases. | |
1438 | */ | |
1439 | static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow) | |
1440 | { | |
1441 | struct mm_struct *mm = vma->vm_mm; | |
1442 | struct rlimit *rlim = current->signal->rlim; | |
1443 | ||
1444 | /* address space limit tests */ | |
119f657c | 1445 | if (!may_expand_vm(mm, grow)) |
1da177e4 LT |
1446 | return -ENOMEM; |
1447 | ||
1448 | /* Stack limit test */ | |
1449 | if (size > rlim[RLIMIT_STACK].rlim_cur) | |
1450 | return -ENOMEM; | |
1451 | ||
1452 | /* mlock limit tests */ | |
1453 | if (vma->vm_flags & VM_LOCKED) { | |
1454 | unsigned long locked; | |
1455 | unsigned long limit; | |
1456 | locked = mm->locked_vm + grow; | |
1457 | limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT; | |
1458 | if (locked > limit && !capable(CAP_IPC_LOCK)) | |
1459 | return -ENOMEM; | |
1460 | } | |
1461 | ||
1462 | /* | |
1463 | * Overcommit.. This must be the final test, as it will | |
1464 | * update security statistics. | |
1465 | */ | |
1466 | if (security_vm_enough_memory(grow)) | |
1467 | return -ENOMEM; | |
1468 | ||
1469 | /* Ok, everything looks good - let it rip */ | |
1470 | mm->total_vm += grow; | |
1471 | if (vma->vm_flags & VM_LOCKED) | |
1472 | mm->locked_vm += grow; | |
ab50b8ed | 1473 | vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); |
1da177e4 LT |
1474 | return 0; |
1475 | } | |
1476 | ||
46dea3d0 | 1477 | #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) |
1da177e4 | 1478 | /* |
46dea3d0 HD |
1479 | * PA-RISC uses this for its stack; IA64 for its Register Backing Store. |
1480 | * vma is the last one with address > vma->vm_end. Have to extend vma. | |
1da177e4 | 1481 | */ |
9ab88515 | 1482 | #ifndef CONFIG_IA64 |
46dea3d0 HD |
1483 | static inline |
1484 | #endif | |
1485 | int expand_upwards(struct vm_area_struct *vma, unsigned long address) | |
1da177e4 LT |
1486 | { |
1487 | int error; | |
1488 | ||
1489 | if (!(vma->vm_flags & VM_GROWSUP)) | |
1490 | return -EFAULT; | |
1491 | ||
1492 | /* | |
1493 | * We must make sure the anon_vma is allocated | |
1494 | * so that the anon_vma locking is not a noop. | |
1495 | */ | |
1496 | if (unlikely(anon_vma_prepare(vma))) | |
1497 | return -ENOMEM; | |
1498 | anon_vma_lock(vma); | |
1499 | ||
1500 | /* | |
1501 | * vma->vm_start/vm_end cannot change under us because the caller | |
1502 | * is required to hold the mmap_sem in read mode. We need the | |
1503 | * anon_vma lock to serialize against concurrent expand_stacks. | |
1504 | */ | |
1505 | address += 4 + PAGE_SIZE - 1; | |
1506 | address &= PAGE_MASK; | |
1507 | error = 0; | |
1508 | ||
1509 | /* Somebody else might have raced and expanded it already */ | |
1510 | if (address > vma->vm_end) { | |
1511 | unsigned long size, grow; | |
1512 | ||
1513 | size = address - vma->vm_start; | |
1514 | grow = (address - vma->vm_end) >> PAGE_SHIFT; | |
1515 | ||
1516 | error = acct_stack_growth(vma, size, grow); | |
1517 | if (!error) | |
1518 | vma->vm_end = address; | |
1519 | } | |
1520 | anon_vma_unlock(vma); | |
1521 | return error; | |
1522 | } | |
46dea3d0 HD |
1523 | #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ |
1524 | ||
1525 | #ifdef CONFIG_STACK_GROWSUP | |
1526 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
1527 | { | |
1528 | return expand_upwards(vma, address); | |
1529 | } | |
1da177e4 LT |
1530 | |
1531 | struct vm_area_struct * | |
1532 | find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
1533 | { | |
1534 | struct vm_area_struct *vma, *prev; | |
1535 | ||
1536 | addr &= PAGE_MASK; | |
1537 | vma = find_vma_prev(mm, addr, &prev); | |
1538 | if (vma && (vma->vm_start <= addr)) | |
1539 | return vma; | |
1540 | if (!prev || expand_stack(prev, addr)) | |
1541 | return NULL; | |
1542 | if (prev->vm_flags & VM_LOCKED) { | |
1543 | make_pages_present(addr, prev->vm_end); | |
1544 | } | |
1545 | return prev; | |
1546 | } | |
1547 | #else | |
1548 | /* | |
1549 | * vma is the first one with address < vma->vm_start. Have to extend vma. | |
1550 | */ | |
1551 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
1552 | { | |
1553 | int error; | |
1554 | ||
1555 | /* | |
1556 | * We must make sure the anon_vma is allocated | |
1557 | * so that the anon_vma locking is not a noop. | |
1558 | */ | |
1559 | if (unlikely(anon_vma_prepare(vma))) | |
1560 | return -ENOMEM; | |
1561 | anon_vma_lock(vma); | |
1562 | ||
1563 | /* | |
1564 | * vma->vm_start/vm_end cannot change under us because the caller | |
1565 | * is required to hold the mmap_sem in read mode. We need the | |
1566 | * anon_vma lock to serialize against concurrent expand_stacks. | |
1567 | */ | |
1568 | address &= PAGE_MASK; | |
1569 | error = 0; | |
1570 | ||
1571 | /* Somebody else might have raced and expanded it already */ | |
1572 | if (address < vma->vm_start) { | |
1573 | unsigned long size, grow; | |
1574 | ||
1575 | size = vma->vm_end - address; | |
1576 | grow = (vma->vm_start - address) >> PAGE_SHIFT; | |
1577 | ||
1578 | error = acct_stack_growth(vma, size, grow); | |
1579 | if (!error) { | |
1580 | vma->vm_start = address; | |
1581 | vma->vm_pgoff -= grow; | |
1582 | } | |
1583 | } | |
1584 | anon_vma_unlock(vma); | |
1585 | return error; | |
1586 | } | |
1587 | ||
1588 | struct vm_area_struct * | |
1589 | find_extend_vma(struct mm_struct * mm, unsigned long addr) | |
1590 | { | |
1591 | struct vm_area_struct * vma; | |
1592 | unsigned long start; | |
1593 | ||
1594 | addr &= PAGE_MASK; | |
1595 | vma = find_vma(mm,addr); | |
1596 | if (!vma) | |
1597 | return NULL; | |
1598 | if (vma->vm_start <= addr) | |
1599 | return vma; | |
1600 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
1601 | return NULL; | |
1602 | start = vma->vm_start; | |
1603 | if (expand_stack(vma, addr)) | |
1604 | return NULL; | |
1605 | if (vma->vm_flags & VM_LOCKED) { | |
1606 | make_pages_present(addr, start); | |
1607 | } | |
1608 | return vma; | |
1609 | } | |
1610 | #endif | |
1611 | ||
1da177e4 | 1612 | /* |
2c0b3814 | 1613 | * Ok - we have the memory areas we should free on the vma list, |
1da177e4 | 1614 | * so release them, and do the vma updates. |
2c0b3814 HD |
1615 | * |
1616 | * Called with the mm semaphore held. | |
1da177e4 | 1617 | */ |
2c0b3814 | 1618 | static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 1619 | { |
365e9c87 HD |
1620 | /* Update high watermark before we lower total_vm */ |
1621 | update_hiwater_vm(mm); | |
1da177e4 | 1622 | do { |
2c0b3814 HD |
1623 | long nrpages = vma_pages(vma); |
1624 | ||
1625 | mm->total_vm -= nrpages; | |
1626 | if (vma->vm_flags & VM_LOCKED) | |
1627 | mm->locked_vm -= nrpages; | |
1628 | vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); | |
a8fb5618 | 1629 | vma = remove_vma(vma); |
146425a3 | 1630 | } while (vma); |
1da177e4 LT |
1631 | validate_mm(mm); |
1632 | } | |
1633 | ||
1634 | /* | |
1635 | * Get rid of page table information in the indicated region. | |
1636 | * | |
f10df686 | 1637 | * Called with the mm semaphore held. |
1da177e4 LT |
1638 | */ |
1639 | static void unmap_region(struct mm_struct *mm, | |
e0da382c HD |
1640 | struct vm_area_struct *vma, struct vm_area_struct *prev, |
1641 | unsigned long start, unsigned long end) | |
1da177e4 | 1642 | { |
e0da382c | 1643 | struct vm_area_struct *next = prev? prev->vm_next: mm->mmap; |
1da177e4 LT |
1644 | struct mmu_gather *tlb; |
1645 | unsigned long nr_accounted = 0; | |
1646 | ||
1647 | lru_add_drain(); | |
1648 | tlb = tlb_gather_mmu(mm, 0); | |
365e9c87 | 1649 | update_hiwater_rss(mm); |
508034a3 | 1650 | unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL); |
1da177e4 | 1651 | vm_unacct_memory(nr_accounted); |
e2cdef8c | 1652 | free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS, |
e0da382c | 1653 | next? next->vm_start: 0); |
1da177e4 LT |
1654 | tlb_finish_mmu(tlb, start, end); |
1655 | } | |
1656 | ||
1657 | /* | |
1658 | * Create a list of vma's touched by the unmap, removing them from the mm's | |
1659 | * vma list as we go.. | |
1660 | */ | |
1661 | static void | |
1662 | detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, | |
1663 | struct vm_area_struct *prev, unsigned long end) | |
1664 | { | |
1665 | struct vm_area_struct **insertion_point; | |
1666 | struct vm_area_struct *tail_vma = NULL; | |
1363c3cd | 1667 | unsigned long addr; |
1da177e4 LT |
1668 | |
1669 | insertion_point = (prev ? &prev->vm_next : &mm->mmap); | |
1670 | do { | |
1671 | rb_erase(&vma->vm_rb, &mm->mm_rb); | |
1672 | mm->map_count--; | |
1673 | tail_vma = vma; | |
1674 | vma = vma->vm_next; | |
1675 | } while (vma && vma->vm_start < end); | |
1676 | *insertion_point = vma; | |
1677 | tail_vma->vm_next = NULL; | |
1363c3cd WW |
1678 | if (mm->unmap_area == arch_unmap_area) |
1679 | addr = prev ? prev->vm_end : mm->mmap_base; | |
1680 | else | |
1681 | addr = vma ? vma->vm_start : mm->mmap_base; | |
1682 | mm->unmap_area(mm, addr); | |
1da177e4 LT |
1683 | mm->mmap_cache = NULL; /* Kill the cache. */ |
1684 | } | |
1685 | ||
1686 | /* | |
1687 | * Split a vma into two pieces at address 'addr', a new vma is allocated | |
1688 | * either for the first part or the the tail. | |
1689 | */ | |
1690 | int split_vma(struct mm_struct * mm, struct vm_area_struct * vma, | |
1691 | unsigned long addr, int new_below) | |
1692 | { | |
1693 | struct mempolicy *pol; | |
1694 | struct vm_area_struct *new; | |
1695 | ||
1696 | if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK)) | |
1697 | return -EINVAL; | |
1698 | ||
1699 | if (mm->map_count >= sysctl_max_map_count) | |
1700 | return -ENOMEM; | |
1701 | ||
1702 | new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL); | |
1703 | if (!new) | |
1704 | return -ENOMEM; | |
1705 | ||
1706 | /* most fields are the same, copy all, and then fixup */ | |
1707 | *new = *vma; | |
1708 | ||
1709 | if (new_below) | |
1710 | new->vm_end = addr; | |
1711 | else { | |
1712 | new->vm_start = addr; | |
1713 | new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); | |
1714 | } | |
1715 | ||
1716 | pol = mpol_copy(vma_policy(vma)); | |
1717 | if (IS_ERR(pol)) { | |
1718 | kmem_cache_free(vm_area_cachep, new); | |
1719 | return PTR_ERR(pol); | |
1720 | } | |
1721 | vma_set_policy(new, pol); | |
1722 | ||
1723 | if (new->vm_file) | |
1724 | get_file(new->vm_file); | |
1725 | ||
1726 | if (new->vm_ops && new->vm_ops->open) | |
1727 | new->vm_ops->open(new); | |
1728 | ||
1729 | if (new_below) | |
1730 | vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + | |
1731 | ((addr - new->vm_start) >> PAGE_SHIFT), new); | |
1732 | else | |
1733 | vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); | |
1734 | ||
1735 | return 0; | |
1736 | } | |
1737 | ||
1738 | /* Munmap is split into 2 main parts -- this part which finds | |
1739 | * what needs doing, and the areas themselves, which do the | |
1740 | * work. This now handles partial unmappings. | |
1741 | * Jeremy Fitzhardinge <jeremy@goop.org> | |
1742 | */ | |
1743 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | |
1744 | { | |
1745 | unsigned long end; | |
146425a3 | 1746 | struct vm_area_struct *vma, *prev, *last; |
1da177e4 LT |
1747 | |
1748 | if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) | |
1749 | return -EINVAL; | |
1750 | ||
1751 | if ((len = PAGE_ALIGN(len)) == 0) | |
1752 | return -EINVAL; | |
1753 | ||
1754 | /* Find the first overlapping VMA */ | |
146425a3 HD |
1755 | vma = find_vma_prev(mm, start, &prev); |
1756 | if (!vma) | |
1da177e4 | 1757 | return 0; |
146425a3 | 1758 | /* we have start < vma->vm_end */ |
1da177e4 LT |
1759 | |
1760 | /* if it doesn't overlap, we have nothing.. */ | |
1761 | end = start + len; | |
146425a3 | 1762 | if (vma->vm_start >= end) |
1da177e4 LT |
1763 | return 0; |
1764 | ||
1765 | /* | |
1766 | * If we need to split any vma, do it now to save pain later. | |
1767 | * | |
1768 | * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially | |
1769 | * unmapped vm_area_struct will remain in use: so lower split_vma | |
1770 | * places tmp vma above, and higher split_vma places tmp vma below. | |
1771 | */ | |
146425a3 HD |
1772 | if (start > vma->vm_start) { |
1773 | int error = split_vma(mm, vma, start, 0); | |
1da177e4 LT |
1774 | if (error) |
1775 | return error; | |
146425a3 | 1776 | prev = vma; |
1da177e4 LT |
1777 | } |
1778 | ||
1779 | /* Does it split the last one? */ | |
1780 | last = find_vma(mm, end); | |
1781 | if (last && end > last->vm_start) { | |
1782 | int error = split_vma(mm, last, end, 1); | |
1783 | if (error) | |
1784 | return error; | |
1785 | } | |
146425a3 | 1786 | vma = prev? prev->vm_next: mm->mmap; |
1da177e4 LT |
1787 | |
1788 | /* | |
1789 | * Remove the vma's, and unmap the actual pages | |
1790 | */ | |
146425a3 HD |
1791 | detach_vmas_to_be_unmapped(mm, vma, prev, end); |
1792 | unmap_region(mm, vma, prev, start, end); | |
1da177e4 LT |
1793 | |
1794 | /* Fix up all other VM information */ | |
2c0b3814 | 1795 | remove_vma_list(mm, vma); |
1da177e4 LT |
1796 | |
1797 | return 0; | |
1798 | } | |
1799 | ||
1800 | EXPORT_SYMBOL(do_munmap); | |
1801 | ||
1802 | asmlinkage long sys_munmap(unsigned long addr, size_t len) | |
1803 | { | |
1804 | int ret; | |
1805 | struct mm_struct *mm = current->mm; | |
1806 | ||
1807 | profile_munmap(addr); | |
1808 | ||
1809 | down_write(&mm->mmap_sem); | |
1810 | ret = do_munmap(mm, addr, len); | |
1811 | up_write(&mm->mmap_sem); | |
1812 | return ret; | |
1813 | } | |
1814 | ||
1815 | static inline void verify_mm_writelocked(struct mm_struct *mm) | |
1816 | { | |
a241ec65 | 1817 | #ifdef CONFIG_DEBUG_VM |
1da177e4 LT |
1818 | if (unlikely(down_read_trylock(&mm->mmap_sem))) { |
1819 | WARN_ON(1); | |
1820 | up_read(&mm->mmap_sem); | |
1821 | } | |
1822 | #endif | |
1823 | } | |
1824 | ||
1825 | /* | |
1826 | * this is really a simplified "do_mmap". it only handles | |
1827 | * anonymous maps. eventually we may be able to do some | |
1828 | * brk-specific accounting here. | |
1829 | */ | |
1830 | unsigned long do_brk(unsigned long addr, unsigned long len) | |
1831 | { | |
1832 | struct mm_struct * mm = current->mm; | |
1833 | struct vm_area_struct * vma, * prev; | |
1834 | unsigned long flags; | |
1835 | struct rb_node ** rb_link, * rb_parent; | |
1836 | pgoff_t pgoff = addr >> PAGE_SHIFT; | |
1837 | ||
1838 | len = PAGE_ALIGN(len); | |
1839 | if (!len) | |
1840 | return addr; | |
1841 | ||
1842 | if ((addr + len) > TASK_SIZE || (addr + len) < addr) | |
1843 | return -EINVAL; | |
1844 | ||
1845 | /* | |
1846 | * mlock MCL_FUTURE? | |
1847 | */ | |
1848 | if (mm->def_flags & VM_LOCKED) { | |
1849 | unsigned long locked, lock_limit; | |
93ea1d0a CW |
1850 | locked = len >> PAGE_SHIFT; |
1851 | locked += mm->locked_vm; | |
1da177e4 | 1852 | lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; |
93ea1d0a | 1853 | lock_limit >>= PAGE_SHIFT; |
1da177e4 LT |
1854 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) |
1855 | return -EAGAIN; | |
1856 | } | |
1857 | ||
1858 | /* | |
1859 | * mm->mmap_sem is required to protect against another thread | |
1860 | * changing the mappings in case we sleep. | |
1861 | */ | |
1862 | verify_mm_writelocked(mm); | |
1863 | ||
1864 | /* | |
1865 | * Clear old maps. this also does some error checking for us | |
1866 | */ | |
1867 | munmap_back: | |
1868 | vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); | |
1869 | if (vma && vma->vm_start < addr + len) { | |
1870 | if (do_munmap(mm, addr, len)) | |
1871 | return -ENOMEM; | |
1872 | goto munmap_back; | |
1873 | } | |
1874 | ||
1875 | /* Check against address space limits *after* clearing old maps... */ | |
119f657c | 1876 | if (!may_expand_vm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
1877 | return -ENOMEM; |
1878 | ||
1879 | if (mm->map_count > sysctl_max_map_count) | |
1880 | return -ENOMEM; | |
1881 | ||
1882 | if (security_vm_enough_memory(len >> PAGE_SHIFT)) | |
1883 | return -ENOMEM; | |
1884 | ||
1885 | flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; | |
1886 | ||
1887 | /* Can we just expand an old private anonymous mapping? */ | |
1888 | if (vma_merge(mm, prev, addr, addr + len, flags, | |
1889 | NULL, NULL, pgoff, NULL)) | |
1890 | goto out; | |
1891 | ||
1892 | /* | |
1893 | * create a vma struct for an anonymous mapping | |
1894 | */ | |
c5e3b83e | 1895 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
1896 | if (!vma) { |
1897 | vm_unacct_memory(len >> PAGE_SHIFT); | |
1898 | return -ENOMEM; | |
1899 | } | |
1da177e4 LT |
1900 | |
1901 | vma->vm_mm = mm; | |
1902 | vma->vm_start = addr; | |
1903 | vma->vm_end = addr + len; | |
1904 | vma->vm_pgoff = pgoff; | |
1905 | vma->vm_flags = flags; | |
1906 | vma->vm_page_prot = protection_map[flags & 0x0f]; | |
1907 | vma_link(mm, vma, prev, rb_link, rb_parent); | |
1908 | out: | |
1909 | mm->total_vm += len >> PAGE_SHIFT; | |
1910 | if (flags & VM_LOCKED) { | |
1911 | mm->locked_vm += len >> PAGE_SHIFT; | |
1912 | make_pages_present(addr, addr + len); | |
1913 | } | |
1914 | return addr; | |
1915 | } | |
1916 | ||
1917 | EXPORT_SYMBOL(do_brk); | |
1918 | ||
1919 | /* Release all mmaps. */ | |
1920 | void exit_mmap(struct mm_struct *mm) | |
1921 | { | |
1922 | struct mmu_gather *tlb; | |
e0da382c | 1923 | struct vm_area_struct *vma = mm->mmap; |
1da177e4 | 1924 | unsigned long nr_accounted = 0; |
ee39b37b | 1925 | unsigned long end; |
1da177e4 LT |
1926 | |
1927 | lru_add_drain(); | |
1da177e4 | 1928 | flush_cache_mm(mm); |
e0da382c | 1929 | tlb = tlb_gather_mmu(mm, 1); |
365e9c87 | 1930 | /* Don't update_hiwater_rss(mm) here, do_exit already did */ |
e0da382c | 1931 | /* Use -1 here to ensure all VMAs in the mm are unmapped */ |
508034a3 | 1932 | end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL); |
1da177e4 | 1933 | vm_unacct_memory(nr_accounted); |
e2cdef8c | 1934 | free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0); |
ee39b37b | 1935 | tlb_finish_mmu(tlb, 0, end); |
1da177e4 | 1936 | |
1da177e4 | 1937 | /* |
8f4f8c16 HD |
1938 | * Walk the list again, actually closing and freeing it, |
1939 | * with preemption enabled, without holding any MM locks. | |
1da177e4 | 1940 | */ |
a8fb5618 HD |
1941 | while (vma) |
1942 | vma = remove_vma(vma); | |
e0da382c | 1943 | |
e2cdef8c | 1944 | BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT); |
1da177e4 LT |
1945 | } |
1946 | ||
1947 | /* Insert vm structure into process list sorted by address | |
1948 | * and into the inode's i_mmap tree. If vm_file is non-NULL | |
1949 | * then i_mmap_lock is taken here. | |
1950 | */ | |
1951 | int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma) | |
1952 | { | |
1953 | struct vm_area_struct * __vma, * prev; | |
1954 | struct rb_node ** rb_link, * rb_parent; | |
1955 | ||
1956 | /* | |
1957 | * The vm_pgoff of a purely anonymous vma should be irrelevant | |
1958 | * until its first write fault, when page's anon_vma and index | |
1959 | * are set. But now set the vm_pgoff it will almost certainly | |
1960 | * end up with (unless mremap moves it elsewhere before that | |
1961 | * first wfault), so /proc/pid/maps tells a consistent story. | |
1962 | * | |
1963 | * By setting it to reflect the virtual start address of the | |
1964 | * vma, merges and splits can happen in a seamless way, just | |
1965 | * using the existing file pgoff checks and manipulations. | |
1966 | * Similarly in do_mmap_pgoff and in do_brk. | |
1967 | */ | |
1968 | if (!vma->vm_file) { | |
1969 | BUG_ON(vma->anon_vma); | |
1970 | vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; | |
1971 | } | |
1972 | __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent); | |
1973 | if (__vma && __vma->vm_start < vma->vm_end) | |
1974 | return -ENOMEM; | |
2fd4ef85 HD |
1975 | if ((vma->vm_flags & VM_ACCOUNT) && |
1976 | security_vm_enough_memory(vma_pages(vma))) | |
1977 | return -ENOMEM; | |
1da177e4 LT |
1978 | vma_link(mm, vma, prev, rb_link, rb_parent); |
1979 | return 0; | |
1980 | } | |
1981 | ||
1982 | /* | |
1983 | * Copy the vma structure to a new location in the same mm, | |
1984 | * prior to moving page table entries, to effect an mremap move. | |
1985 | */ | |
1986 | struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, | |
1987 | unsigned long addr, unsigned long len, pgoff_t pgoff) | |
1988 | { | |
1989 | struct vm_area_struct *vma = *vmap; | |
1990 | unsigned long vma_start = vma->vm_start; | |
1991 | struct mm_struct *mm = vma->vm_mm; | |
1992 | struct vm_area_struct *new_vma, *prev; | |
1993 | struct rb_node **rb_link, *rb_parent; | |
1994 | struct mempolicy *pol; | |
1995 | ||
1996 | /* | |
1997 | * If anonymous vma has not yet been faulted, update new pgoff | |
1998 | * to match new location, to increase its chance of merging. | |
1999 | */ | |
2000 | if (!vma->vm_file && !vma->anon_vma) | |
2001 | pgoff = addr >> PAGE_SHIFT; | |
2002 | ||
2003 | find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); | |
2004 | new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, | |
2005 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); | |
2006 | if (new_vma) { | |
2007 | /* | |
2008 | * Source vma may have been merged into new_vma | |
2009 | */ | |
2010 | if (vma_start >= new_vma->vm_start && | |
2011 | vma_start < new_vma->vm_end) | |
2012 | *vmap = new_vma; | |
2013 | } else { | |
2014 | new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL); | |
2015 | if (new_vma) { | |
2016 | *new_vma = *vma; | |
2017 | pol = mpol_copy(vma_policy(vma)); | |
2018 | if (IS_ERR(pol)) { | |
2019 | kmem_cache_free(vm_area_cachep, new_vma); | |
2020 | return NULL; | |
2021 | } | |
2022 | vma_set_policy(new_vma, pol); | |
2023 | new_vma->vm_start = addr; | |
2024 | new_vma->vm_end = addr + len; | |
2025 | new_vma->vm_pgoff = pgoff; | |
2026 | if (new_vma->vm_file) | |
2027 | get_file(new_vma->vm_file); | |
2028 | if (new_vma->vm_ops && new_vma->vm_ops->open) | |
2029 | new_vma->vm_ops->open(new_vma); | |
2030 | vma_link(mm, new_vma, prev, rb_link, rb_parent); | |
2031 | } | |
2032 | } | |
2033 | return new_vma; | |
2034 | } | |
119f657c | 2035 | |
2036 | /* | |
2037 | * Return true if the calling process may expand its vm space by the passed | |
2038 | * number of pages | |
2039 | */ | |
2040 | int may_expand_vm(struct mm_struct *mm, unsigned long npages) | |
2041 | { | |
2042 | unsigned long cur = mm->total_vm; /* pages */ | |
2043 | unsigned long lim; | |
2044 | ||
2045 | lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT; | |
2046 | ||
2047 | if (cur + npages > lim) | |
2048 | return 0; | |
2049 | return 1; | |
2050 | } |