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> | |
16 | #include <linux/init.h> | |
17 | #include <linux/file.h> | |
18 | #include <linux/fs.h> | |
19 | #include <linux/personality.h> | |
20 | #include <linux/security.h> | |
21 | #include <linux/hugetlb.h> | |
22 | #include <linux/profile.h> | |
23 | #include <linux/module.h> | |
24 | #include <linux/mount.h> | |
25 | #include <linux/mempolicy.h> | |
26 | #include <linux/rmap.h> | |
27 | ||
28 | #include <asm/uaccess.h> | |
29 | #include <asm/cacheflush.h> | |
30 | #include <asm/tlb.h> | |
31 | ||
32 | /* | |
33 | * WARNING: the debugging will use recursive algorithms so never enable this | |
34 | * unless you know what you are doing. | |
35 | */ | |
36 | #undef DEBUG_MM_RB | |
37 | ||
38 | /* description of effects of mapping type and prot in current implementation. | |
39 | * this is due to the limited x86 page protection hardware. The expected | |
40 | * behavior is in parens: | |
41 | * | |
42 | * map_type prot | |
43 | * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC | |
44 | * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes | |
45 | * w: (no) no w: (no) no w: (yes) yes w: (no) no | |
46 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
47 | * | |
48 | * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes | |
49 | * w: (no) no w: (no) no w: (copy) copy w: (no) no | |
50 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
51 | * | |
52 | */ | |
53 | pgprot_t protection_map[16] = { | |
54 | __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, | |
55 | __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 | |
56 | }; | |
57 | ||
58 | int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ | |
59 | int sysctl_overcommit_ratio = 50; /* default is 50% */ | |
60 | int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; | |
61 | atomic_t vm_committed_space = ATOMIC_INIT(0); | |
62 | ||
63 | /* | |
64 | * Check that a process has enough memory to allocate a new virtual | |
65 | * mapping. 0 means there is enough memory for the allocation to | |
66 | * succeed and -ENOMEM implies there is not. | |
67 | * | |
68 | * We currently support three overcommit policies, which are set via the | |
69 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting | |
70 | * | |
71 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. | |
72 | * Additional code 2002 Jul 20 by Robert Love. | |
73 | * | |
74 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. | |
75 | * | |
76 | * Note this is a helper function intended to be used by LSMs which | |
77 | * wish to use this logic. | |
78 | */ | |
79 | int __vm_enough_memory(long pages, int cap_sys_admin) | |
80 | { | |
81 | unsigned long free, allowed; | |
82 | ||
83 | vm_acct_memory(pages); | |
84 | ||
85 | /* | |
86 | * Sometimes we want to use more memory than we have | |
87 | */ | |
88 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) | |
89 | return 0; | |
90 | ||
91 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { | |
92 | unsigned long n; | |
93 | ||
94 | free = get_page_cache_size(); | |
95 | free += nr_swap_pages; | |
96 | ||
97 | /* | |
98 | * Any slabs which are created with the | |
99 | * SLAB_RECLAIM_ACCOUNT flag claim to have contents | |
100 | * which are reclaimable, under pressure. The dentry | |
101 | * cache and most inode caches should fall into this | |
102 | */ | |
103 | free += atomic_read(&slab_reclaim_pages); | |
104 | ||
105 | /* | |
106 | * Leave the last 3% for root | |
107 | */ | |
108 | if (!cap_sys_admin) | |
109 | free -= free / 32; | |
110 | ||
111 | if (free > pages) | |
112 | return 0; | |
113 | ||
114 | /* | |
115 | * nr_free_pages() is very expensive on large systems, | |
116 | * only call if we're about to fail. | |
117 | */ | |
118 | n = nr_free_pages(); | |
119 | if (!cap_sys_admin) | |
120 | n -= n / 32; | |
121 | free += n; | |
122 | ||
123 | if (free > pages) | |
124 | return 0; | |
125 | vm_unacct_memory(pages); | |
126 | return -ENOMEM; | |
127 | } | |
128 | ||
129 | allowed = (totalram_pages - hugetlb_total_pages()) | |
130 | * sysctl_overcommit_ratio / 100; | |
131 | /* | |
132 | * Leave the last 3% for root | |
133 | */ | |
134 | if (!cap_sys_admin) | |
135 | allowed -= allowed / 32; | |
136 | allowed += total_swap_pages; | |
137 | ||
138 | /* Don't let a single process grow too big: | |
139 | leave 3% of the size of this process for other processes */ | |
140 | allowed -= current->mm->total_vm / 32; | |
141 | ||
142 | if (atomic_read(&vm_committed_space) < allowed) | |
143 | return 0; | |
144 | ||
145 | vm_unacct_memory(pages); | |
146 | ||
147 | return -ENOMEM; | |
148 | } | |
149 | ||
150 | EXPORT_SYMBOL(sysctl_overcommit_memory); | |
151 | EXPORT_SYMBOL(sysctl_overcommit_ratio); | |
152 | EXPORT_SYMBOL(sysctl_max_map_count); | |
153 | EXPORT_SYMBOL(vm_committed_space); | |
154 | EXPORT_SYMBOL(__vm_enough_memory); | |
155 | ||
156 | /* | |
157 | * Requires inode->i_mapping->i_mmap_lock | |
158 | */ | |
159 | static void __remove_shared_vm_struct(struct vm_area_struct *vma, | |
160 | struct file *file, struct address_space *mapping) | |
161 | { | |
162 | if (vma->vm_flags & VM_DENYWRITE) | |
163 | atomic_inc(&file->f_dentry->d_inode->i_writecount); | |
164 | if (vma->vm_flags & VM_SHARED) | |
165 | mapping->i_mmap_writable--; | |
166 | ||
167 | flush_dcache_mmap_lock(mapping); | |
168 | if (unlikely(vma->vm_flags & VM_NONLINEAR)) | |
169 | list_del_init(&vma->shared.vm_set.list); | |
170 | else | |
171 | vma_prio_tree_remove(vma, &mapping->i_mmap); | |
172 | flush_dcache_mmap_unlock(mapping); | |
173 | } | |
174 | ||
175 | /* | |
176 | * Remove one vm structure and free it. | |
177 | */ | |
178 | static void remove_vm_struct(struct vm_area_struct *vma) | |
179 | { | |
180 | struct file *file = vma->vm_file; | |
181 | ||
182 | might_sleep(); | |
183 | if (file) { | |
184 | struct address_space *mapping = file->f_mapping; | |
185 | spin_lock(&mapping->i_mmap_lock); | |
186 | __remove_shared_vm_struct(vma, file, mapping); | |
187 | spin_unlock(&mapping->i_mmap_lock); | |
188 | } | |
189 | if (vma->vm_ops && vma->vm_ops->close) | |
190 | vma->vm_ops->close(vma); | |
191 | if (file) | |
192 | fput(file); | |
193 | anon_vma_unlink(vma); | |
194 | mpol_free(vma_policy(vma)); | |
195 | kmem_cache_free(vm_area_cachep, vma); | |
196 | } | |
197 | ||
198 | /* | |
199 | * sys_brk() for the most part doesn't need the global kernel | |
200 | * lock, except when an application is doing something nasty | |
201 | * like trying to un-brk an area that has already been mapped | |
202 | * to a regular file. in this case, the unmapping will need | |
203 | * to invoke file system routines that need the global lock. | |
204 | */ | |
205 | asmlinkage unsigned long sys_brk(unsigned long brk) | |
206 | { | |
207 | unsigned long rlim, retval; | |
208 | unsigned long newbrk, oldbrk; | |
209 | struct mm_struct *mm = current->mm; | |
210 | ||
211 | down_write(&mm->mmap_sem); | |
212 | ||
213 | if (brk < mm->end_code) | |
214 | goto out; | |
215 | newbrk = PAGE_ALIGN(brk); | |
216 | oldbrk = PAGE_ALIGN(mm->brk); | |
217 | if (oldbrk == newbrk) | |
218 | goto set_brk; | |
219 | ||
220 | /* Always allow shrinking brk. */ | |
221 | if (brk <= mm->brk) { | |
222 | if (!do_munmap(mm, newbrk, oldbrk-newbrk)) | |
223 | goto set_brk; | |
224 | goto out; | |
225 | } | |
226 | ||
227 | /* Check against rlimit.. */ | |
228 | rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur; | |
229 | if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim) | |
230 | goto out; | |
231 | ||
232 | /* Check against existing mmap mappings. */ | |
233 | if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) | |
234 | goto out; | |
235 | ||
236 | /* Ok, looks good - let it rip. */ | |
237 | if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) | |
238 | goto out; | |
239 | set_brk: | |
240 | mm->brk = brk; | |
241 | out: | |
242 | retval = mm->brk; | |
243 | up_write(&mm->mmap_sem); | |
244 | return retval; | |
245 | } | |
246 | ||
247 | #ifdef DEBUG_MM_RB | |
248 | static int browse_rb(struct rb_root *root) | |
249 | { | |
250 | int i = 0, j; | |
251 | struct rb_node *nd, *pn = NULL; | |
252 | unsigned long prev = 0, pend = 0; | |
253 | ||
254 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
255 | struct vm_area_struct *vma; | |
256 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
257 | if (vma->vm_start < prev) | |
258 | printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1; | |
259 | if (vma->vm_start < pend) | |
260 | printk("vm_start %lx pend %lx\n", vma->vm_start, pend); | |
261 | if (vma->vm_start > vma->vm_end) | |
262 | printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start); | |
263 | i++; | |
264 | pn = nd; | |
265 | } | |
266 | j = 0; | |
267 | for (nd = pn; nd; nd = rb_prev(nd)) { | |
268 | j++; | |
269 | } | |
270 | if (i != j) | |
271 | printk("backwards %d, forwards %d\n", j, i), i = 0; | |
272 | return i; | |
273 | } | |
274 | ||
275 | void validate_mm(struct mm_struct *mm) | |
276 | { | |
277 | int bug = 0; | |
278 | int i = 0; | |
279 | struct vm_area_struct *tmp = mm->mmap; | |
280 | while (tmp) { | |
281 | tmp = tmp->vm_next; | |
282 | i++; | |
283 | } | |
284 | if (i != mm->map_count) | |
285 | printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1; | |
286 | i = browse_rb(&mm->mm_rb); | |
287 | if (i != mm->map_count) | |
288 | printk("map_count %d rb %d\n", mm->map_count, i), bug = 1; | |
289 | if (bug) | |
290 | BUG(); | |
291 | } | |
292 | #else | |
293 | #define validate_mm(mm) do { } while (0) | |
294 | #endif | |
295 | ||
296 | static struct vm_area_struct * | |
297 | find_vma_prepare(struct mm_struct *mm, unsigned long addr, | |
298 | struct vm_area_struct **pprev, struct rb_node ***rb_link, | |
299 | struct rb_node ** rb_parent) | |
300 | { | |
301 | struct vm_area_struct * vma; | |
302 | struct rb_node ** __rb_link, * __rb_parent, * rb_prev; | |
303 | ||
304 | __rb_link = &mm->mm_rb.rb_node; | |
305 | rb_prev = __rb_parent = NULL; | |
306 | vma = NULL; | |
307 | ||
308 | while (*__rb_link) { | |
309 | struct vm_area_struct *vma_tmp; | |
310 | ||
311 | __rb_parent = *__rb_link; | |
312 | vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); | |
313 | ||
314 | if (vma_tmp->vm_end > addr) { | |
315 | vma = vma_tmp; | |
316 | if (vma_tmp->vm_start <= addr) | |
317 | return vma; | |
318 | __rb_link = &__rb_parent->rb_left; | |
319 | } else { | |
320 | rb_prev = __rb_parent; | |
321 | __rb_link = &__rb_parent->rb_right; | |
322 | } | |
323 | } | |
324 | ||
325 | *pprev = NULL; | |
326 | if (rb_prev) | |
327 | *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); | |
328 | *rb_link = __rb_link; | |
329 | *rb_parent = __rb_parent; | |
330 | return vma; | |
331 | } | |
332 | ||
333 | static inline void | |
334 | __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma, | |
335 | struct vm_area_struct *prev, struct rb_node *rb_parent) | |
336 | { | |
337 | if (prev) { | |
338 | vma->vm_next = prev->vm_next; | |
339 | prev->vm_next = vma; | |
340 | } else { | |
341 | mm->mmap = vma; | |
342 | if (rb_parent) | |
343 | vma->vm_next = rb_entry(rb_parent, | |
344 | struct vm_area_struct, vm_rb); | |
345 | else | |
346 | vma->vm_next = NULL; | |
347 | } | |
348 | } | |
349 | ||
350 | void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, | |
351 | struct rb_node **rb_link, struct rb_node *rb_parent) | |
352 | { | |
353 | rb_link_node(&vma->vm_rb, rb_parent, rb_link); | |
354 | rb_insert_color(&vma->vm_rb, &mm->mm_rb); | |
355 | } | |
356 | ||
357 | static inline void __vma_link_file(struct vm_area_struct *vma) | |
358 | { | |
359 | struct file * file; | |
360 | ||
361 | file = vma->vm_file; | |
362 | if (file) { | |
363 | struct address_space *mapping = file->f_mapping; | |
364 | ||
365 | if (vma->vm_flags & VM_DENYWRITE) | |
366 | atomic_dec(&file->f_dentry->d_inode->i_writecount); | |
367 | if (vma->vm_flags & VM_SHARED) | |
368 | mapping->i_mmap_writable++; | |
369 | ||
370 | flush_dcache_mmap_lock(mapping); | |
371 | if (unlikely(vma->vm_flags & VM_NONLINEAR)) | |
372 | vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); | |
373 | else | |
374 | vma_prio_tree_insert(vma, &mapping->i_mmap); | |
375 | flush_dcache_mmap_unlock(mapping); | |
376 | } | |
377 | } | |
378 | ||
379 | static void | |
380 | __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
381 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
382 | struct rb_node *rb_parent) | |
383 | { | |
384 | __vma_link_list(mm, vma, prev, rb_parent); | |
385 | __vma_link_rb(mm, vma, rb_link, rb_parent); | |
386 | __anon_vma_link(vma); | |
387 | } | |
388 | ||
389 | static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
390 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
391 | struct rb_node *rb_parent) | |
392 | { | |
393 | struct address_space *mapping = NULL; | |
394 | ||
395 | if (vma->vm_file) | |
396 | mapping = vma->vm_file->f_mapping; | |
397 | ||
398 | if (mapping) { | |
399 | spin_lock(&mapping->i_mmap_lock); | |
400 | vma->vm_truncate_count = mapping->truncate_count; | |
401 | } | |
402 | anon_vma_lock(vma); | |
403 | ||
404 | __vma_link(mm, vma, prev, rb_link, rb_parent); | |
405 | __vma_link_file(vma); | |
406 | ||
407 | anon_vma_unlock(vma); | |
408 | if (mapping) | |
409 | spin_unlock(&mapping->i_mmap_lock); | |
410 | ||
411 | mm->map_count++; | |
412 | validate_mm(mm); | |
413 | } | |
414 | ||
415 | /* | |
416 | * Helper for vma_adjust in the split_vma insert case: | |
417 | * insert vm structure into list and rbtree and anon_vma, | |
418 | * but it has already been inserted into prio_tree earlier. | |
419 | */ | |
420 | static void | |
421 | __insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma) | |
422 | { | |
423 | struct vm_area_struct * __vma, * prev; | |
424 | struct rb_node ** rb_link, * rb_parent; | |
425 | ||
426 | __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent); | |
427 | if (__vma && __vma->vm_start < vma->vm_end) | |
428 | BUG(); | |
429 | __vma_link(mm, vma, prev, rb_link, rb_parent); | |
430 | mm->map_count++; | |
431 | } | |
432 | ||
433 | static inline void | |
434 | __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, | |
435 | struct vm_area_struct *prev) | |
436 | { | |
437 | prev->vm_next = vma->vm_next; | |
438 | rb_erase(&vma->vm_rb, &mm->mm_rb); | |
439 | if (mm->mmap_cache == vma) | |
440 | mm->mmap_cache = prev; | |
441 | } | |
442 | ||
443 | /* | |
444 | * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that | |
445 | * is already present in an i_mmap tree without adjusting the tree. | |
446 | * The following helper function should be used when such adjustments | |
447 | * are necessary. The "insert" vma (if any) is to be inserted | |
448 | * before we drop the necessary locks. | |
449 | */ | |
450 | void vma_adjust(struct vm_area_struct *vma, unsigned long start, | |
451 | unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) | |
452 | { | |
453 | struct mm_struct *mm = vma->vm_mm; | |
454 | struct vm_area_struct *next = vma->vm_next; | |
455 | struct vm_area_struct *importer = NULL; | |
456 | struct address_space *mapping = NULL; | |
457 | struct prio_tree_root *root = NULL; | |
458 | struct file *file = vma->vm_file; | |
459 | struct anon_vma *anon_vma = NULL; | |
460 | long adjust_next = 0; | |
461 | int remove_next = 0; | |
462 | ||
463 | if (next && !insert) { | |
464 | if (end >= next->vm_end) { | |
465 | /* | |
466 | * vma expands, overlapping all the next, and | |
467 | * perhaps the one after too (mprotect case 6). | |
468 | */ | |
469 | again: remove_next = 1 + (end > next->vm_end); | |
470 | end = next->vm_end; | |
471 | anon_vma = next->anon_vma; | |
472 | importer = vma; | |
473 | } else if (end > next->vm_start) { | |
474 | /* | |
475 | * vma expands, overlapping part of the next: | |
476 | * mprotect case 5 shifting the boundary up. | |
477 | */ | |
478 | adjust_next = (end - next->vm_start) >> PAGE_SHIFT; | |
479 | anon_vma = next->anon_vma; | |
480 | importer = vma; | |
481 | } else if (end < vma->vm_end) { | |
482 | /* | |
483 | * vma shrinks, and !insert tells it's not | |
484 | * split_vma inserting another: so it must be | |
485 | * mprotect case 4 shifting the boundary down. | |
486 | */ | |
487 | adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT); | |
488 | anon_vma = next->anon_vma; | |
489 | importer = next; | |
490 | } | |
491 | } | |
492 | ||
493 | if (file) { | |
494 | mapping = file->f_mapping; | |
495 | if (!(vma->vm_flags & VM_NONLINEAR)) | |
496 | root = &mapping->i_mmap; | |
497 | spin_lock(&mapping->i_mmap_lock); | |
498 | if (importer && | |
499 | vma->vm_truncate_count != next->vm_truncate_count) { | |
500 | /* | |
501 | * unmap_mapping_range might be in progress: | |
502 | * ensure that the expanding vma is rescanned. | |
503 | */ | |
504 | importer->vm_truncate_count = 0; | |
505 | } | |
506 | if (insert) { | |
507 | insert->vm_truncate_count = vma->vm_truncate_count; | |
508 | /* | |
509 | * Put into prio_tree now, so instantiated pages | |
510 | * are visible to arm/parisc __flush_dcache_page | |
511 | * throughout; but we cannot insert into address | |
512 | * space until vma start or end is updated. | |
513 | */ | |
514 | __vma_link_file(insert); | |
515 | } | |
516 | } | |
517 | ||
518 | /* | |
519 | * When changing only vma->vm_end, we don't really need | |
520 | * anon_vma lock: but is that case worth optimizing out? | |
521 | */ | |
522 | if (vma->anon_vma) | |
523 | anon_vma = vma->anon_vma; | |
524 | if (anon_vma) { | |
525 | spin_lock(&anon_vma->lock); | |
526 | /* | |
527 | * Easily overlooked: when mprotect shifts the boundary, | |
528 | * make sure the expanding vma has anon_vma set if the | |
529 | * shrinking vma had, to cover any anon pages imported. | |
530 | */ | |
531 | if (importer && !importer->anon_vma) { | |
532 | importer->anon_vma = anon_vma; | |
533 | __anon_vma_link(importer); | |
534 | } | |
535 | } | |
536 | ||
537 | if (root) { | |
538 | flush_dcache_mmap_lock(mapping); | |
539 | vma_prio_tree_remove(vma, root); | |
540 | if (adjust_next) | |
541 | vma_prio_tree_remove(next, root); | |
542 | } | |
543 | ||
544 | vma->vm_start = start; | |
545 | vma->vm_end = end; | |
546 | vma->vm_pgoff = pgoff; | |
547 | if (adjust_next) { | |
548 | next->vm_start += adjust_next << PAGE_SHIFT; | |
549 | next->vm_pgoff += adjust_next; | |
550 | } | |
551 | ||
552 | if (root) { | |
553 | if (adjust_next) | |
554 | vma_prio_tree_insert(next, root); | |
555 | vma_prio_tree_insert(vma, root); | |
556 | flush_dcache_mmap_unlock(mapping); | |
557 | } | |
558 | ||
559 | if (remove_next) { | |
560 | /* | |
561 | * vma_merge has merged next into vma, and needs | |
562 | * us to remove next before dropping the locks. | |
563 | */ | |
564 | __vma_unlink(mm, next, vma); | |
565 | if (file) | |
566 | __remove_shared_vm_struct(next, file, mapping); | |
567 | if (next->anon_vma) | |
568 | __anon_vma_merge(vma, next); | |
569 | } else if (insert) { | |
570 | /* | |
571 | * split_vma has split insert from vma, and needs | |
572 | * us to insert it before dropping the locks | |
573 | * (it may either follow vma or precede it). | |
574 | */ | |
575 | __insert_vm_struct(mm, insert); | |
576 | } | |
577 | ||
578 | if (anon_vma) | |
579 | spin_unlock(&anon_vma->lock); | |
580 | if (mapping) | |
581 | spin_unlock(&mapping->i_mmap_lock); | |
582 | ||
583 | if (remove_next) { | |
584 | if (file) | |
585 | fput(file); | |
586 | mm->map_count--; | |
587 | mpol_free(vma_policy(next)); | |
588 | kmem_cache_free(vm_area_cachep, next); | |
589 | /* | |
590 | * In mprotect's case 6 (see comments on vma_merge), | |
591 | * we must remove another next too. It would clutter | |
592 | * up the code too much to do both in one go. | |
593 | */ | |
594 | if (remove_next == 2) { | |
595 | next = vma->vm_next; | |
596 | goto again; | |
597 | } | |
598 | } | |
599 | ||
600 | validate_mm(mm); | |
601 | } | |
602 | ||
603 | /* | |
604 | * If the vma has a ->close operation then the driver probably needs to release | |
605 | * per-vma resources, so we don't attempt to merge those. | |
606 | */ | |
607 | #define VM_SPECIAL (VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED) | |
608 | ||
609 | static inline int is_mergeable_vma(struct vm_area_struct *vma, | |
610 | struct file *file, unsigned long vm_flags) | |
611 | { | |
612 | if (vma->vm_flags != vm_flags) | |
613 | return 0; | |
614 | if (vma->vm_file != file) | |
615 | return 0; | |
616 | if (vma->vm_ops && vma->vm_ops->close) | |
617 | return 0; | |
618 | return 1; | |
619 | } | |
620 | ||
621 | static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, | |
622 | struct anon_vma *anon_vma2) | |
623 | { | |
624 | return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2); | |
625 | } | |
626 | ||
627 | /* | |
628 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
629 | * in front of (at a lower virtual address and file offset than) the vma. | |
630 | * | |
631 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
632 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
633 | * | |
634 | * We don't check here for the merged mmap wrapping around the end of pagecache | |
635 | * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which | |
636 | * wrap, nor mmaps which cover the final page at index -1UL. | |
637 | */ | |
638 | static int | |
639 | can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, | |
640 | struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) | |
641 | { | |
642 | if (is_mergeable_vma(vma, file, vm_flags) && | |
643 | is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { | |
644 | if (vma->vm_pgoff == vm_pgoff) | |
645 | return 1; | |
646 | } | |
647 | return 0; | |
648 | } | |
649 | ||
650 | /* | |
651 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
652 | * beyond (at a higher virtual address and file offset than) the vma. | |
653 | * | |
654 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
655 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
656 | */ | |
657 | static int | |
658 | can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, | |
659 | struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) | |
660 | { | |
661 | if (is_mergeable_vma(vma, file, vm_flags) && | |
662 | is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { | |
663 | pgoff_t vm_pglen; | |
664 | vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; | |
665 | if (vma->vm_pgoff + vm_pglen == vm_pgoff) | |
666 | return 1; | |
667 | } | |
668 | return 0; | |
669 | } | |
670 | ||
671 | /* | |
672 | * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out | |
673 | * whether that can be merged with its predecessor or its successor. | |
674 | * Or both (it neatly fills a hole). | |
675 | * | |
676 | * In most cases - when called for mmap, brk or mremap - [addr,end) is | |
677 | * certain not to be mapped by the time vma_merge is called; but when | |
678 | * called for mprotect, it is certain to be already mapped (either at | |
679 | * an offset within prev, or at the start of next), and the flags of | |
680 | * this area are about to be changed to vm_flags - and the no-change | |
681 | * case has already been eliminated. | |
682 | * | |
683 | * The following mprotect cases have to be considered, where AAAA is | |
684 | * the area passed down from mprotect_fixup, never extending beyond one | |
685 | * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: | |
686 | * | |
687 | * AAAA AAAA AAAA AAAA | |
688 | * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX | |
689 | * cannot merge might become might become might become | |
690 | * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or | |
691 | * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or | |
692 | * mremap move: PPPPNNNNNNNN 8 | |
693 | * AAAA | |
694 | * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN | |
695 | * might become case 1 below case 2 below case 3 below | |
696 | * | |
697 | * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: | |
698 | * mprotect_fixup updates vm_flags & vm_page_prot on successful return. | |
699 | */ | |
700 | struct vm_area_struct *vma_merge(struct mm_struct *mm, | |
701 | struct vm_area_struct *prev, unsigned long addr, | |
702 | unsigned long end, unsigned long vm_flags, | |
703 | struct anon_vma *anon_vma, struct file *file, | |
704 | pgoff_t pgoff, struct mempolicy *policy) | |
705 | { | |
706 | pgoff_t pglen = (end - addr) >> PAGE_SHIFT; | |
707 | struct vm_area_struct *area, *next; | |
708 | ||
709 | /* | |
710 | * We later require that vma->vm_flags == vm_flags, | |
711 | * so this tests vma->vm_flags & VM_SPECIAL, too. | |
712 | */ | |
713 | if (vm_flags & VM_SPECIAL) | |
714 | return NULL; | |
715 | ||
716 | if (prev) | |
717 | next = prev->vm_next; | |
718 | else | |
719 | next = mm->mmap; | |
720 | area = next; | |
721 | if (next && next->vm_end == end) /* cases 6, 7, 8 */ | |
722 | next = next->vm_next; | |
723 | ||
724 | /* | |
725 | * Can it merge with the predecessor? | |
726 | */ | |
727 | if (prev && prev->vm_end == addr && | |
728 | mpol_equal(vma_policy(prev), policy) && | |
729 | can_vma_merge_after(prev, vm_flags, | |
730 | anon_vma, file, pgoff)) { | |
731 | /* | |
732 | * OK, it can. Can we now merge in the successor as well? | |
733 | */ | |
734 | if (next && end == next->vm_start && | |
735 | mpol_equal(policy, vma_policy(next)) && | |
736 | can_vma_merge_before(next, vm_flags, | |
737 | anon_vma, file, pgoff+pglen) && | |
738 | is_mergeable_anon_vma(prev->anon_vma, | |
739 | next->anon_vma)) { | |
740 | /* cases 1, 6 */ | |
741 | vma_adjust(prev, prev->vm_start, | |
742 | next->vm_end, prev->vm_pgoff, NULL); | |
743 | } else /* cases 2, 5, 7 */ | |
744 | vma_adjust(prev, prev->vm_start, | |
745 | end, prev->vm_pgoff, NULL); | |
746 | return prev; | |
747 | } | |
748 | ||
749 | /* | |
750 | * Can this new request be merged in front of next? | |
751 | */ | |
752 | if (next && end == next->vm_start && | |
753 | mpol_equal(policy, vma_policy(next)) && | |
754 | can_vma_merge_before(next, vm_flags, | |
755 | anon_vma, file, pgoff+pglen)) { | |
756 | if (prev && addr < prev->vm_end) /* case 4 */ | |
757 | vma_adjust(prev, prev->vm_start, | |
758 | addr, prev->vm_pgoff, NULL); | |
759 | else /* cases 3, 8 */ | |
760 | vma_adjust(area, addr, next->vm_end, | |
761 | next->vm_pgoff - pglen, NULL); | |
762 | return area; | |
763 | } | |
764 | ||
765 | return NULL; | |
766 | } | |
767 | ||
768 | /* | |
769 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check | |
770 | * neighbouring vmas for a suitable anon_vma, before it goes off | |
771 | * to allocate a new anon_vma. It checks because a repetitive | |
772 | * sequence of mprotects and faults may otherwise lead to distinct | |
773 | * anon_vmas being allocated, preventing vma merge in subsequent | |
774 | * mprotect. | |
775 | */ | |
776 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) | |
777 | { | |
778 | struct vm_area_struct *near; | |
779 | unsigned long vm_flags; | |
780 | ||
781 | near = vma->vm_next; | |
782 | if (!near) | |
783 | goto try_prev; | |
784 | ||
785 | /* | |
786 | * Since only mprotect tries to remerge vmas, match flags | |
787 | * which might be mprotected into each other later on. | |
788 | * Neither mlock nor madvise tries to remerge at present, | |
789 | * so leave their flags as obstructing a merge. | |
790 | */ | |
791 | vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC); | |
792 | vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC); | |
793 | ||
794 | if (near->anon_vma && vma->vm_end == near->vm_start && | |
795 | mpol_equal(vma_policy(vma), vma_policy(near)) && | |
796 | can_vma_merge_before(near, vm_flags, | |
797 | NULL, vma->vm_file, vma->vm_pgoff + | |
798 | ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT))) | |
799 | return near->anon_vma; | |
800 | try_prev: | |
801 | /* | |
802 | * It is potentially slow to have to call find_vma_prev here. | |
803 | * But it's only on the first write fault on the vma, not | |
804 | * every time, and we could devise a way to avoid it later | |
805 | * (e.g. stash info in next's anon_vma_node when assigning | |
806 | * an anon_vma, or when trying vma_merge). Another time. | |
807 | */ | |
808 | if (find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma) | |
809 | BUG(); | |
810 | if (!near) | |
811 | goto none; | |
812 | ||
813 | vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC); | |
814 | vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC); | |
815 | ||
816 | if (near->anon_vma && near->vm_end == vma->vm_start && | |
817 | mpol_equal(vma_policy(near), vma_policy(vma)) && | |
818 | can_vma_merge_after(near, vm_flags, | |
819 | NULL, vma->vm_file, vma->vm_pgoff)) | |
820 | return near->anon_vma; | |
821 | none: | |
822 | /* | |
823 | * There's no absolute need to look only at touching neighbours: | |
824 | * we could search further afield for "compatible" anon_vmas. | |
825 | * But it would probably just be a waste of time searching, | |
826 | * or lead to too many vmas hanging off the same anon_vma. | |
827 | * We're trying to allow mprotect remerging later on, | |
828 | * not trying to minimize memory used for anon_vmas. | |
829 | */ | |
830 | return NULL; | |
831 | } | |
832 | ||
833 | #ifdef CONFIG_PROC_FS | |
834 | void __vm_stat_account(struct mm_struct *mm, unsigned long flags, | |
835 | struct file *file, long pages) | |
836 | { | |
837 | const unsigned long stack_flags | |
838 | = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); | |
839 | ||
840 | #ifdef CONFIG_HUGETLB | |
841 | if (flags & VM_HUGETLB) { | |
842 | if (!(flags & VM_DONTCOPY)) | |
843 | mm->shared_vm += pages; | |
844 | return; | |
845 | } | |
846 | #endif /* CONFIG_HUGETLB */ | |
847 | ||
848 | if (file) { | |
849 | mm->shared_vm += pages; | |
850 | if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) | |
851 | mm->exec_vm += pages; | |
852 | } else if (flags & stack_flags) | |
853 | mm->stack_vm += pages; | |
854 | if (flags & (VM_RESERVED|VM_IO)) | |
855 | mm->reserved_vm += pages; | |
856 | } | |
857 | #endif /* CONFIG_PROC_FS */ | |
858 | ||
859 | /* | |
860 | * The caller must hold down_write(current->mm->mmap_sem). | |
861 | */ | |
862 | ||
863 | unsigned long do_mmap_pgoff(struct file * file, unsigned long addr, | |
864 | unsigned long len, unsigned long prot, | |
865 | unsigned long flags, unsigned long pgoff) | |
866 | { | |
867 | struct mm_struct * mm = current->mm; | |
868 | struct vm_area_struct * vma, * prev; | |
869 | struct inode *inode; | |
870 | unsigned int vm_flags; | |
871 | int correct_wcount = 0; | |
872 | int error; | |
873 | struct rb_node ** rb_link, * rb_parent; | |
874 | int accountable = 1; | |
875 | unsigned long charged = 0, reqprot = prot; | |
876 | ||
877 | if (file) { | |
878 | if (is_file_hugepages(file)) | |
879 | accountable = 0; | |
880 | ||
881 | if (!file->f_op || !file->f_op->mmap) | |
882 | return -ENODEV; | |
883 | ||
884 | if ((prot & PROT_EXEC) && | |
885 | (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)) | |
886 | return -EPERM; | |
887 | } | |
888 | /* | |
889 | * Does the application expect PROT_READ to imply PROT_EXEC? | |
890 | * | |
891 | * (the exception is when the underlying filesystem is noexec | |
892 | * mounted, in which case we dont add PROT_EXEC.) | |
893 | */ | |
894 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
895 | if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))) | |
896 | prot |= PROT_EXEC; | |
897 | ||
898 | if (!len) | |
899 | return -EINVAL; | |
900 | ||
901 | /* Careful about overflows.. */ | |
902 | len = PAGE_ALIGN(len); | |
903 | if (!len || len > TASK_SIZE) | |
904 | return -ENOMEM; | |
905 | ||
906 | /* offset overflow? */ | |
907 | if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) | |
908 | return -EOVERFLOW; | |
909 | ||
910 | /* Too many mappings? */ | |
911 | if (mm->map_count > sysctl_max_map_count) | |
912 | return -ENOMEM; | |
913 | ||
914 | /* Obtain the address to map to. we verify (or select) it and ensure | |
915 | * that it represents a valid section of the address space. | |
916 | */ | |
917 | addr = get_unmapped_area(file, addr, len, pgoff, flags); | |
918 | if (addr & ~PAGE_MASK) | |
919 | return addr; | |
920 | ||
921 | /* Do simple checking here so the lower-level routines won't have | |
922 | * to. we assume access permissions have been handled by the open | |
923 | * of the memory object, so we don't do any here. | |
924 | */ | |
925 | vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | | |
926 | mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; | |
927 | ||
928 | if (flags & MAP_LOCKED) { | |
929 | if (!can_do_mlock()) | |
930 | return -EPERM; | |
931 | vm_flags |= VM_LOCKED; | |
932 | } | |
933 | /* mlock MCL_FUTURE? */ | |
934 | if (vm_flags & VM_LOCKED) { | |
935 | unsigned long locked, lock_limit; | |
936 | locked = mm->locked_vm << PAGE_SHIFT; | |
937 | lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; | |
938 | locked += len; | |
939 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) | |
940 | return -EAGAIN; | |
941 | } | |
942 | ||
943 | inode = file ? file->f_dentry->d_inode : NULL; | |
944 | ||
945 | if (file) { | |
946 | switch (flags & MAP_TYPE) { | |
947 | case MAP_SHARED: | |
948 | if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) | |
949 | return -EACCES; | |
950 | ||
951 | /* | |
952 | * Make sure we don't allow writing to an append-only | |
953 | * file.. | |
954 | */ | |
955 | if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) | |
956 | return -EACCES; | |
957 | ||
958 | /* | |
959 | * Make sure there are no mandatory locks on the file. | |
960 | */ | |
961 | if (locks_verify_locked(inode)) | |
962 | return -EAGAIN; | |
963 | ||
964 | vm_flags |= VM_SHARED | VM_MAYSHARE; | |
965 | if (!(file->f_mode & FMODE_WRITE)) | |
966 | vm_flags &= ~(VM_MAYWRITE | VM_SHARED); | |
967 | ||
968 | /* fall through */ | |
969 | case MAP_PRIVATE: | |
970 | if (!(file->f_mode & FMODE_READ)) | |
971 | return -EACCES; | |
972 | break; | |
973 | ||
974 | default: | |
975 | return -EINVAL; | |
976 | } | |
977 | } else { | |
978 | switch (flags & MAP_TYPE) { | |
979 | case MAP_SHARED: | |
980 | vm_flags |= VM_SHARED | VM_MAYSHARE; | |
981 | break; | |
982 | case MAP_PRIVATE: | |
983 | /* | |
984 | * Set pgoff according to addr for anon_vma. | |
985 | */ | |
986 | pgoff = addr >> PAGE_SHIFT; | |
987 | break; | |
988 | default: | |
989 | return -EINVAL; | |
990 | } | |
991 | } | |
992 | ||
993 | error = security_file_mmap(file, reqprot, prot, flags); | |
994 | if (error) | |
995 | return error; | |
996 | ||
997 | /* Clear old maps */ | |
998 | error = -ENOMEM; | |
999 | munmap_back: | |
1000 | vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); | |
1001 | if (vma && vma->vm_start < addr + len) { | |
1002 | if (do_munmap(mm, addr, len)) | |
1003 | return -ENOMEM; | |
1004 | goto munmap_back; | |
1005 | } | |
1006 | ||
1007 | /* Check against address space limit. */ | |
1008 | if ((mm->total_vm << PAGE_SHIFT) + len | |
1009 | > current->signal->rlim[RLIMIT_AS].rlim_cur) | |
1010 | return -ENOMEM; | |
1011 | ||
1012 | if (accountable && (!(flags & MAP_NORESERVE) || | |
1013 | sysctl_overcommit_memory == OVERCOMMIT_NEVER)) { | |
1014 | if (vm_flags & VM_SHARED) { | |
1015 | /* Check memory availability in shmem_file_setup? */ | |
1016 | vm_flags |= VM_ACCOUNT; | |
1017 | } else if (vm_flags & VM_WRITE) { | |
1018 | /* | |
1019 | * Private writable mapping: check memory availability | |
1020 | */ | |
1021 | charged = len >> PAGE_SHIFT; | |
1022 | if (security_vm_enough_memory(charged)) | |
1023 | return -ENOMEM; | |
1024 | vm_flags |= VM_ACCOUNT; | |
1025 | } | |
1026 | } | |
1027 | ||
1028 | /* | |
1029 | * Can we just expand an old private anonymous mapping? | |
1030 | * The VM_SHARED test is necessary because shmem_zero_setup | |
1031 | * will create the file object for a shared anonymous map below. | |
1032 | */ | |
1033 | if (!file && !(vm_flags & VM_SHARED) && | |
1034 | vma_merge(mm, prev, addr, addr + len, vm_flags, | |
1035 | NULL, NULL, pgoff, NULL)) | |
1036 | goto out; | |
1037 | ||
1038 | /* | |
1039 | * Determine the object being mapped and call the appropriate | |
1040 | * specific mapper. the address has already been validated, but | |
1041 | * not unmapped, but the maps are removed from the list. | |
1042 | */ | |
1043 | vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL); | |
1044 | if (!vma) { | |
1045 | error = -ENOMEM; | |
1046 | goto unacct_error; | |
1047 | } | |
1048 | memset(vma, 0, sizeof(*vma)); | |
1049 | ||
1050 | vma->vm_mm = mm; | |
1051 | vma->vm_start = addr; | |
1052 | vma->vm_end = addr + len; | |
1053 | vma->vm_flags = vm_flags; | |
1054 | vma->vm_page_prot = protection_map[vm_flags & 0x0f]; | |
1055 | vma->vm_pgoff = pgoff; | |
1056 | ||
1057 | if (file) { | |
1058 | error = -EINVAL; | |
1059 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) | |
1060 | goto free_vma; | |
1061 | if (vm_flags & VM_DENYWRITE) { | |
1062 | error = deny_write_access(file); | |
1063 | if (error) | |
1064 | goto free_vma; | |
1065 | correct_wcount = 1; | |
1066 | } | |
1067 | vma->vm_file = file; | |
1068 | get_file(file); | |
1069 | error = file->f_op->mmap(file, vma); | |
1070 | if (error) | |
1071 | goto unmap_and_free_vma; | |
1072 | } else if (vm_flags & VM_SHARED) { | |
1073 | error = shmem_zero_setup(vma); | |
1074 | if (error) | |
1075 | goto free_vma; | |
1076 | } | |
1077 | ||
1078 | /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform | |
1079 | * shmem_zero_setup (perhaps called through /dev/zero's ->mmap) | |
1080 | * that memory reservation must be checked; but that reservation | |
1081 | * belongs to shared memory object, not to vma: so now clear it. | |
1082 | */ | |
1083 | if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT)) | |
1084 | vma->vm_flags &= ~VM_ACCOUNT; | |
1085 | ||
1086 | /* Can addr have changed?? | |
1087 | * | |
1088 | * Answer: Yes, several device drivers can do it in their | |
1089 | * f_op->mmap method. -DaveM | |
1090 | */ | |
1091 | addr = vma->vm_start; | |
1092 | pgoff = vma->vm_pgoff; | |
1093 | vm_flags = vma->vm_flags; | |
1094 | ||
1095 | if (!file || !vma_merge(mm, prev, addr, vma->vm_end, | |
1096 | vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) { | |
1097 | file = vma->vm_file; | |
1098 | vma_link(mm, vma, prev, rb_link, rb_parent); | |
1099 | if (correct_wcount) | |
1100 | atomic_inc(&inode->i_writecount); | |
1101 | } else { | |
1102 | if (file) { | |
1103 | if (correct_wcount) | |
1104 | atomic_inc(&inode->i_writecount); | |
1105 | fput(file); | |
1106 | } | |
1107 | mpol_free(vma_policy(vma)); | |
1108 | kmem_cache_free(vm_area_cachep, vma); | |
1109 | } | |
1110 | out: | |
1111 | mm->total_vm += len >> PAGE_SHIFT; | |
1112 | __vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); | |
1113 | if (vm_flags & VM_LOCKED) { | |
1114 | mm->locked_vm += len >> PAGE_SHIFT; | |
1115 | make_pages_present(addr, addr + len); | |
1116 | } | |
1117 | if (flags & MAP_POPULATE) { | |
1118 | up_write(&mm->mmap_sem); | |
1119 | sys_remap_file_pages(addr, len, 0, | |
1120 | pgoff, flags & MAP_NONBLOCK); | |
1121 | down_write(&mm->mmap_sem); | |
1122 | } | |
1123 | return addr; | |
1124 | ||
1125 | unmap_and_free_vma: | |
1126 | if (correct_wcount) | |
1127 | atomic_inc(&inode->i_writecount); | |
1128 | vma->vm_file = NULL; | |
1129 | fput(file); | |
1130 | ||
1131 | /* Undo any partial mapping done by a device driver. */ | |
1132 | zap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, NULL); | |
1133 | free_vma: | |
1134 | kmem_cache_free(vm_area_cachep, vma); | |
1135 | unacct_error: | |
1136 | if (charged) | |
1137 | vm_unacct_memory(charged); | |
1138 | return error; | |
1139 | } | |
1140 | ||
1141 | EXPORT_SYMBOL(do_mmap_pgoff); | |
1142 | ||
1143 | /* Get an address range which is currently unmapped. | |
1144 | * For shmat() with addr=0. | |
1145 | * | |
1146 | * Ugly calling convention alert: | |
1147 | * Return value with the low bits set means error value, | |
1148 | * ie | |
1149 | * if (ret & ~PAGE_MASK) | |
1150 | * error = ret; | |
1151 | * | |
1152 | * This function "knows" that -ENOMEM has the bits set. | |
1153 | */ | |
1154 | #ifndef HAVE_ARCH_UNMAPPED_AREA | |
1155 | unsigned long | |
1156 | arch_get_unmapped_area(struct file *filp, unsigned long addr, | |
1157 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1158 | { | |
1159 | struct mm_struct *mm = current->mm; | |
1160 | struct vm_area_struct *vma; | |
1161 | unsigned long start_addr; | |
1162 | ||
1163 | if (len > TASK_SIZE) | |
1164 | return -ENOMEM; | |
1165 | ||
1166 | if (addr) { | |
1167 | addr = PAGE_ALIGN(addr); | |
1168 | vma = find_vma(mm, addr); | |
1169 | if (TASK_SIZE - len >= addr && | |
1170 | (!vma || addr + len <= vma->vm_start)) | |
1171 | return addr; | |
1172 | } | |
1173 | start_addr = addr = mm->free_area_cache; | |
1174 | ||
1175 | full_search: | |
1176 | for (vma = find_vma(mm, addr); ; vma = vma->vm_next) { | |
1177 | /* At this point: (!vma || addr < vma->vm_end). */ | |
1178 | if (TASK_SIZE - len < addr) { | |
1179 | /* | |
1180 | * Start a new search - just in case we missed | |
1181 | * some holes. | |
1182 | */ | |
1183 | if (start_addr != TASK_UNMAPPED_BASE) { | |
1184 | start_addr = addr = TASK_UNMAPPED_BASE; | |
1185 | goto full_search; | |
1186 | } | |
1187 | return -ENOMEM; | |
1188 | } | |
1189 | if (!vma || addr + len <= vma->vm_start) { | |
1190 | /* | |
1191 | * Remember the place where we stopped the search: | |
1192 | */ | |
1193 | mm->free_area_cache = addr + len; | |
1194 | return addr; | |
1195 | } | |
1196 | addr = vma->vm_end; | |
1197 | } | |
1198 | } | |
1199 | #endif | |
1200 | ||
1201 | void arch_unmap_area(struct vm_area_struct *area) | |
1202 | { | |
1203 | /* | |
1204 | * Is this a new hole at the lowest possible address? | |
1205 | */ | |
1206 | if (area->vm_start >= TASK_UNMAPPED_BASE && | |
1207 | area->vm_start < area->vm_mm->free_area_cache) | |
1208 | area->vm_mm->free_area_cache = area->vm_start; | |
1209 | } | |
1210 | ||
1211 | /* | |
1212 | * This mmap-allocator allocates new areas top-down from below the | |
1213 | * stack's low limit (the base): | |
1214 | */ | |
1215 | #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN | |
1216 | unsigned long | |
1217 | arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, | |
1218 | const unsigned long len, const unsigned long pgoff, | |
1219 | const unsigned long flags) | |
1220 | { | |
1221 | struct vm_area_struct *vma; | |
1222 | struct mm_struct *mm = current->mm; | |
1223 | unsigned long addr = addr0; | |
1224 | ||
1225 | /* requested length too big for entire address space */ | |
1226 | if (len > TASK_SIZE) | |
1227 | return -ENOMEM; | |
1228 | ||
1229 | /* requesting a specific address */ | |
1230 | if (addr) { | |
1231 | addr = PAGE_ALIGN(addr); | |
1232 | vma = find_vma(mm, addr); | |
1233 | if (TASK_SIZE - len >= addr && | |
1234 | (!vma || addr + len <= vma->vm_start)) | |
1235 | return addr; | |
1236 | } | |
1237 | ||
1238 | /* either no address requested or can't fit in requested address hole */ | |
1239 | addr = mm->free_area_cache; | |
1240 | ||
1241 | /* make sure it can fit in the remaining address space */ | |
1242 | if (addr >= len) { | |
1243 | vma = find_vma(mm, addr-len); | |
1244 | if (!vma || addr <= vma->vm_start) | |
1245 | /* remember the address as a hint for next time */ | |
1246 | return (mm->free_area_cache = addr-len); | |
1247 | } | |
1248 | ||
1249 | addr = mm->mmap_base-len; | |
1250 | ||
1251 | do { | |
1252 | /* | |
1253 | * Lookup failure means no vma is above this address, | |
1254 | * else if new region fits below vma->vm_start, | |
1255 | * return with success: | |
1256 | */ | |
1257 | vma = find_vma(mm, addr); | |
1258 | if (!vma || addr+len <= vma->vm_start) | |
1259 | /* remember the address as a hint for next time */ | |
1260 | return (mm->free_area_cache = addr); | |
1261 | ||
1262 | /* try just below the current vma->vm_start */ | |
1263 | addr = vma->vm_start-len; | |
1264 | } while (len <= vma->vm_start); | |
1265 | ||
1266 | /* | |
1267 | * A failed mmap() very likely causes application failure, | |
1268 | * so fall back to the bottom-up function here. This scenario | |
1269 | * can happen with large stack limits and large mmap() | |
1270 | * allocations. | |
1271 | */ | |
1272 | mm->free_area_cache = TASK_UNMAPPED_BASE; | |
1273 | addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags); | |
1274 | /* | |
1275 | * Restore the topdown base: | |
1276 | */ | |
1277 | mm->free_area_cache = mm->mmap_base; | |
1278 | ||
1279 | return addr; | |
1280 | } | |
1281 | #endif | |
1282 | ||
1283 | void arch_unmap_area_topdown(struct vm_area_struct *area) | |
1284 | { | |
1285 | /* | |
1286 | * Is this a new hole at the highest possible address? | |
1287 | */ | |
1288 | if (area->vm_end > area->vm_mm->free_area_cache) | |
1289 | area->vm_mm->free_area_cache = area->vm_end; | |
1290 | ||
1291 | /* dont allow allocations above current base */ | |
1292 | if (area->vm_mm->free_area_cache > area->vm_mm->mmap_base) | |
1293 | area->vm_mm->free_area_cache = area->vm_mm->mmap_base; | |
1294 | } | |
1295 | ||
1296 | unsigned long | |
1297 | get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, | |
1298 | unsigned long pgoff, unsigned long flags) | |
1299 | { | |
1300 | if (flags & MAP_FIXED) { | |
1301 | unsigned long ret; | |
1302 | ||
1303 | if (addr > TASK_SIZE - len) | |
1304 | return -ENOMEM; | |
1305 | if (addr & ~PAGE_MASK) | |
1306 | return -EINVAL; | |
1307 | if (file && is_file_hugepages(file)) { | |
1308 | /* | |
1309 | * Check if the given range is hugepage aligned, and | |
1310 | * can be made suitable for hugepages. | |
1311 | */ | |
1312 | ret = prepare_hugepage_range(addr, len); | |
1313 | } else { | |
1314 | /* | |
1315 | * Ensure that a normal request is not falling in a | |
1316 | * reserved hugepage range. For some archs like IA-64, | |
1317 | * there is a separate region for hugepages. | |
1318 | */ | |
1319 | ret = is_hugepage_only_range(current->mm, addr, len); | |
1320 | } | |
1321 | if (ret) | |
1322 | return -EINVAL; | |
1323 | return addr; | |
1324 | } | |
1325 | ||
1326 | if (file && file->f_op && file->f_op->get_unmapped_area) | |
1327 | return file->f_op->get_unmapped_area(file, addr, len, | |
1328 | pgoff, flags); | |
1329 | ||
1330 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
1331 | } | |
1332 | ||
1333 | EXPORT_SYMBOL(get_unmapped_area); | |
1334 | ||
1335 | /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ | |
1336 | struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr) | |
1337 | { | |
1338 | struct vm_area_struct *vma = NULL; | |
1339 | ||
1340 | if (mm) { | |
1341 | /* Check the cache first. */ | |
1342 | /* (Cache hit rate is typically around 35%.) */ | |
1343 | vma = mm->mmap_cache; | |
1344 | if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) { | |
1345 | struct rb_node * rb_node; | |
1346 | ||
1347 | rb_node = mm->mm_rb.rb_node; | |
1348 | vma = NULL; | |
1349 | ||
1350 | while (rb_node) { | |
1351 | struct vm_area_struct * vma_tmp; | |
1352 | ||
1353 | vma_tmp = rb_entry(rb_node, | |
1354 | struct vm_area_struct, vm_rb); | |
1355 | ||
1356 | if (vma_tmp->vm_end > addr) { | |
1357 | vma = vma_tmp; | |
1358 | if (vma_tmp->vm_start <= addr) | |
1359 | break; | |
1360 | rb_node = rb_node->rb_left; | |
1361 | } else | |
1362 | rb_node = rb_node->rb_right; | |
1363 | } | |
1364 | if (vma) | |
1365 | mm->mmap_cache = vma; | |
1366 | } | |
1367 | } | |
1368 | return vma; | |
1369 | } | |
1370 | ||
1371 | EXPORT_SYMBOL(find_vma); | |
1372 | ||
1373 | /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */ | |
1374 | struct vm_area_struct * | |
1375 | find_vma_prev(struct mm_struct *mm, unsigned long addr, | |
1376 | struct vm_area_struct **pprev) | |
1377 | { | |
1378 | struct vm_area_struct *vma = NULL, *prev = NULL; | |
1379 | struct rb_node * rb_node; | |
1380 | if (!mm) | |
1381 | goto out; | |
1382 | ||
1383 | /* Guard against addr being lower than the first VMA */ | |
1384 | vma = mm->mmap; | |
1385 | ||
1386 | /* Go through the RB tree quickly. */ | |
1387 | rb_node = mm->mm_rb.rb_node; | |
1388 | ||
1389 | while (rb_node) { | |
1390 | struct vm_area_struct *vma_tmp; | |
1391 | vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
1392 | ||
1393 | if (addr < vma_tmp->vm_end) { | |
1394 | rb_node = rb_node->rb_left; | |
1395 | } else { | |
1396 | prev = vma_tmp; | |
1397 | if (!prev->vm_next || (addr < prev->vm_next->vm_end)) | |
1398 | break; | |
1399 | rb_node = rb_node->rb_right; | |
1400 | } | |
1401 | } | |
1402 | ||
1403 | out: | |
1404 | *pprev = prev; | |
1405 | return prev ? prev->vm_next : vma; | |
1406 | } | |
1407 | ||
1408 | /* | |
1409 | * Verify that the stack growth is acceptable and | |
1410 | * update accounting. This is shared with both the | |
1411 | * grow-up and grow-down cases. | |
1412 | */ | |
1413 | static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow) | |
1414 | { | |
1415 | struct mm_struct *mm = vma->vm_mm; | |
1416 | struct rlimit *rlim = current->signal->rlim; | |
1417 | ||
1418 | /* address space limit tests */ | |
1419 | if (mm->total_vm + grow > rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT) | |
1420 | return -ENOMEM; | |
1421 | ||
1422 | /* Stack limit test */ | |
1423 | if (size > rlim[RLIMIT_STACK].rlim_cur) | |
1424 | return -ENOMEM; | |
1425 | ||
1426 | /* mlock limit tests */ | |
1427 | if (vma->vm_flags & VM_LOCKED) { | |
1428 | unsigned long locked; | |
1429 | unsigned long limit; | |
1430 | locked = mm->locked_vm + grow; | |
1431 | limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT; | |
1432 | if (locked > limit && !capable(CAP_IPC_LOCK)) | |
1433 | return -ENOMEM; | |
1434 | } | |
1435 | ||
1436 | /* | |
1437 | * Overcommit.. This must be the final test, as it will | |
1438 | * update security statistics. | |
1439 | */ | |
1440 | if (security_vm_enough_memory(grow)) | |
1441 | return -ENOMEM; | |
1442 | ||
1443 | /* Ok, everything looks good - let it rip */ | |
1444 | mm->total_vm += grow; | |
1445 | if (vma->vm_flags & VM_LOCKED) | |
1446 | mm->locked_vm += grow; | |
1447 | __vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); | |
1448 | return 0; | |
1449 | } | |
1450 | ||
1451 | #ifdef CONFIG_STACK_GROWSUP | |
1452 | /* | |
1453 | * vma is the first one with address > vma->vm_end. Have to extend vma. | |
1454 | */ | |
1455 | int expand_stack(struct vm_area_struct * vma, unsigned long address) | |
1456 | { | |
1457 | int error; | |
1458 | ||
1459 | if (!(vma->vm_flags & VM_GROWSUP)) | |
1460 | return -EFAULT; | |
1461 | ||
1462 | /* | |
1463 | * We must make sure the anon_vma is allocated | |
1464 | * so that the anon_vma locking is not a noop. | |
1465 | */ | |
1466 | if (unlikely(anon_vma_prepare(vma))) | |
1467 | return -ENOMEM; | |
1468 | anon_vma_lock(vma); | |
1469 | ||
1470 | /* | |
1471 | * vma->vm_start/vm_end cannot change under us because the caller | |
1472 | * is required to hold the mmap_sem in read mode. We need the | |
1473 | * anon_vma lock to serialize against concurrent expand_stacks. | |
1474 | */ | |
1475 | address += 4 + PAGE_SIZE - 1; | |
1476 | address &= PAGE_MASK; | |
1477 | error = 0; | |
1478 | ||
1479 | /* Somebody else might have raced and expanded it already */ | |
1480 | if (address > vma->vm_end) { | |
1481 | unsigned long size, grow; | |
1482 | ||
1483 | size = address - vma->vm_start; | |
1484 | grow = (address - vma->vm_end) >> PAGE_SHIFT; | |
1485 | ||
1486 | error = acct_stack_growth(vma, size, grow); | |
1487 | if (!error) | |
1488 | vma->vm_end = address; | |
1489 | } | |
1490 | anon_vma_unlock(vma); | |
1491 | return error; | |
1492 | } | |
1493 | ||
1494 | struct vm_area_struct * | |
1495 | find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
1496 | { | |
1497 | struct vm_area_struct *vma, *prev; | |
1498 | ||
1499 | addr &= PAGE_MASK; | |
1500 | vma = find_vma_prev(mm, addr, &prev); | |
1501 | if (vma && (vma->vm_start <= addr)) | |
1502 | return vma; | |
1503 | if (!prev || expand_stack(prev, addr)) | |
1504 | return NULL; | |
1505 | if (prev->vm_flags & VM_LOCKED) { | |
1506 | make_pages_present(addr, prev->vm_end); | |
1507 | } | |
1508 | return prev; | |
1509 | } | |
1510 | #else | |
1511 | /* | |
1512 | * vma is the first one with address < vma->vm_start. Have to extend vma. | |
1513 | */ | |
1514 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
1515 | { | |
1516 | int error; | |
1517 | ||
1518 | /* | |
1519 | * We must make sure the anon_vma is allocated | |
1520 | * so that the anon_vma locking is not a noop. | |
1521 | */ | |
1522 | if (unlikely(anon_vma_prepare(vma))) | |
1523 | return -ENOMEM; | |
1524 | anon_vma_lock(vma); | |
1525 | ||
1526 | /* | |
1527 | * vma->vm_start/vm_end cannot change under us because the caller | |
1528 | * is required to hold the mmap_sem in read mode. We need the | |
1529 | * anon_vma lock to serialize against concurrent expand_stacks. | |
1530 | */ | |
1531 | address &= PAGE_MASK; | |
1532 | error = 0; | |
1533 | ||
1534 | /* Somebody else might have raced and expanded it already */ | |
1535 | if (address < vma->vm_start) { | |
1536 | unsigned long size, grow; | |
1537 | ||
1538 | size = vma->vm_end - address; | |
1539 | grow = (vma->vm_start - address) >> PAGE_SHIFT; | |
1540 | ||
1541 | error = acct_stack_growth(vma, size, grow); | |
1542 | if (!error) { | |
1543 | vma->vm_start = address; | |
1544 | vma->vm_pgoff -= grow; | |
1545 | } | |
1546 | } | |
1547 | anon_vma_unlock(vma); | |
1548 | return error; | |
1549 | } | |
1550 | ||
1551 | struct vm_area_struct * | |
1552 | find_extend_vma(struct mm_struct * mm, unsigned long addr) | |
1553 | { | |
1554 | struct vm_area_struct * vma; | |
1555 | unsigned long start; | |
1556 | ||
1557 | addr &= PAGE_MASK; | |
1558 | vma = find_vma(mm,addr); | |
1559 | if (!vma) | |
1560 | return NULL; | |
1561 | if (vma->vm_start <= addr) | |
1562 | return vma; | |
1563 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
1564 | return NULL; | |
1565 | start = vma->vm_start; | |
1566 | if (expand_stack(vma, addr)) | |
1567 | return NULL; | |
1568 | if (vma->vm_flags & VM_LOCKED) { | |
1569 | make_pages_present(addr, start); | |
1570 | } | |
1571 | return vma; | |
1572 | } | |
1573 | #endif | |
1574 | ||
1575 | /* | |
1576 | * Try to free as many page directory entries as we can, | |
1577 | * without having to work very hard at actually scanning | |
1578 | * the page tables themselves. | |
1579 | * | |
1580 | * Right now we try to free page tables if we have a nice | |
1581 | * PGDIR-aligned area that got free'd up. We could be more | |
1582 | * granular if we want to, but this is fast and simple, | |
1583 | * and covers the bad cases. | |
1584 | * | |
1585 | * "prev", if it exists, points to a vma before the one | |
1586 | * we just free'd - but there's no telling how much before. | |
1587 | */ | |
1588 | static void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *prev, | |
1589 | unsigned long start, unsigned long end) | |
1590 | { | |
1591 | unsigned long first = start & PGDIR_MASK; | |
1592 | unsigned long last = end + PGDIR_SIZE - 1; | |
1593 | struct mm_struct *mm = tlb->mm; | |
1594 | ||
1595 | if (last > MM_VM_SIZE(mm) || last < end) | |
1596 | last = MM_VM_SIZE(mm); | |
1597 | ||
1598 | if (!prev) { | |
1599 | prev = mm->mmap; | |
1600 | if (!prev) | |
1601 | goto no_mmaps; | |
1602 | if (prev->vm_end > start) { | |
1603 | if (last > prev->vm_start) | |
1604 | last = prev->vm_start; | |
1605 | goto no_mmaps; | |
1606 | } | |
1607 | } | |
1608 | for (;;) { | |
1609 | struct vm_area_struct *next = prev->vm_next; | |
1610 | ||
1611 | if (next) { | |
1612 | if (next->vm_start < start) { | |
1613 | prev = next; | |
1614 | continue; | |
1615 | } | |
1616 | if (last > next->vm_start) | |
1617 | last = next->vm_start; | |
1618 | } | |
1619 | if (prev->vm_end > first) | |
1620 | first = prev->vm_end; | |
1621 | break; | |
1622 | } | |
1623 | no_mmaps: | |
1624 | if (last < first) /* for arches with discontiguous pgd indices */ | |
1625 | return; | |
1626 | if (first < FIRST_USER_PGD_NR * PGDIR_SIZE) | |
1627 | first = FIRST_USER_PGD_NR * PGDIR_SIZE; | |
1628 | /* No point trying to free anything if we're in the same pte page */ | |
1629 | if ((first & PMD_MASK) < (last & PMD_MASK)) { | |
1630 | clear_page_range(tlb, first, last); | |
1631 | flush_tlb_pgtables(mm, first, last); | |
1632 | } | |
1633 | } | |
1634 | ||
1635 | /* Normal function to fix up a mapping | |
1636 | * This function is the default for when an area has no specific | |
1637 | * function. This may be used as part of a more specific routine. | |
1638 | * | |
1639 | * By the time this function is called, the area struct has been | |
1640 | * removed from the process mapping list. | |
1641 | */ | |
1642 | static void unmap_vma(struct mm_struct *mm, struct vm_area_struct *area) | |
1643 | { | |
1644 | size_t len = area->vm_end - area->vm_start; | |
1645 | ||
1646 | area->vm_mm->total_vm -= len >> PAGE_SHIFT; | |
1647 | if (area->vm_flags & VM_LOCKED) | |
1648 | area->vm_mm->locked_vm -= len >> PAGE_SHIFT; | |
1649 | vm_stat_unaccount(area); | |
1650 | area->vm_mm->unmap_area(area); | |
1651 | remove_vm_struct(area); | |
1652 | } | |
1653 | ||
1654 | /* | |
1655 | * Update the VMA and inode share lists. | |
1656 | * | |
1657 | * Ok - we have the memory areas we should free on the 'free' list, | |
1658 | * so release them, and do the vma updates. | |
1659 | */ | |
1660 | static void unmap_vma_list(struct mm_struct *mm, | |
1661 | struct vm_area_struct *mpnt) | |
1662 | { | |
1663 | do { | |
1664 | struct vm_area_struct *next = mpnt->vm_next; | |
1665 | unmap_vma(mm, mpnt); | |
1666 | mpnt = next; | |
1667 | } while (mpnt != NULL); | |
1668 | validate_mm(mm); | |
1669 | } | |
1670 | ||
1671 | /* | |
1672 | * Get rid of page table information in the indicated region. | |
1673 | * | |
1674 | * Called with the page table lock held. | |
1675 | */ | |
1676 | static void unmap_region(struct mm_struct *mm, | |
1677 | struct vm_area_struct *vma, | |
1678 | struct vm_area_struct *prev, | |
1679 | unsigned long start, | |
1680 | unsigned long end) | |
1681 | { | |
1682 | struct mmu_gather *tlb; | |
1683 | unsigned long nr_accounted = 0; | |
1684 | ||
1685 | lru_add_drain(); | |
1686 | tlb = tlb_gather_mmu(mm, 0); | |
1687 | unmap_vmas(&tlb, mm, vma, start, end, &nr_accounted, NULL); | |
1688 | vm_unacct_memory(nr_accounted); | |
1689 | ||
1690 | if (is_hugepage_only_range(mm, start, end - start)) | |
1691 | hugetlb_free_pgtables(tlb, prev, start, end); | |
1692 | else | |
1693 | free_pgtables(tlb, prev, start, end); | |
1694 | tlb_finish_mmu(tlb, start, end); | |
1695 | } | |
1696 | ||
1697 | /* | |
1698 | * Create a list of vma's touched by the unmap, removing them from the mm's | |
1699 | * vma list as we go.. | |
1700 | */ | |
1701 | static void | |
1702 | detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, | |
1703 | struct vm_area_struct *prev, unsigned long end) | |
1704 | { | |
1705 | struct vm_area_struct **insertion_point; | |
1706 | struct vm_area_struct *tail_vma = NULL; | |
1707 | ||
1708 | insertion_point = (prev ? &prev->vm_next : &mm->mmap); | |
1709 | do { | |
1710 | rb_erase(&vma->vm_rb, &mm->mm_rb); | |
1711 | mm->map_count--; | |
1712 | tail_vma = vma; | |
1713 | vma = vma->vm_next; | |
1714 | } while (vma && vma->vm_start < end); | |
1715 | *insertion_point = vma; | |
1716 | tail_vma->vm_next = NULL; | |
1717 | mm->mmap_cache = NULL; /* Kill the cache. */ | |
1718 | } | |
1719 | ||
1720 | /* | |
1721 | * Split a vma into two pieces at address 'addr', a new vma is allocated | |
1722 | * either for the first part or the the tail. | |
1723 | */ | |
1724 | int split_vma(struct mm_struct * mm, struct vm_area_struct * vma, | |
1725 | unsigned long addr, int new_below) | |
1726 | { | |
1727 | struct mempolicy *pol; | |
1728 | struct vm_area_struct *new; | |
1729 | ||
1730 | if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK)) | |
1731 | return -EINVAL; | |
1732 | ||
1733 | if (mm->map_count >= sysctl_max_map_count) | |
1734 | return -ENOMEM; | |
1735 | ||
1736 | new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL); | |
1737 | if (!new) | |
1738 | return -ENOMEM; | |
1739 | ||
1740 | /* most fields are the same, copy all, and then fixup */ | |
1741 | *new = *vma; | |
1742 | ||
1743 | if (new_below) | |
1744 | new->vm_end = addr; | |
1745 | else { | |
1746 | new->vm_start = addr; | |
1747 | new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); | |
1748 | } | |
1749 | ||
1750 | pol = mpol_copy(vma_policy(vma)); | |
1751 | if (IS_ERR(pol)) { | |
1752 | kmem_cache_free(vm_area_cachep, new); | |
1753 | return PTR_ERR(pol); | |
1754 | } | |
1755 | vma_set_policy(new, pol); | |
1756 | ||
1757 | if (new->vm_file) | |
1758 | get_file(new->vm_file); | |
1759 | ||
1760 | if (new->vm_ops && new->vm_ops->open) | |
1761 | new->vm_ops->open(new); | |
1762 | ||
1763 | if (new_below) | |
1764 | vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + | |
1765 | ((addr - new->vm_start) >> PAGE_SHIFT), new); | |
1766 | else | |
1767 | vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); | |
1768 | ||
1769 | return 0; | |
1770 | } | |
1771 | ||
1772 | /* Munmap is split into 2 main parts -- this part which finds | |
1773 | * what needs doing, and the areas themselves, which do the | |
1774 | * work. This now handles partial unmappings. | |
1775 | * Jeremy Fitzhardinge <jeremy@goop.org> | |
1776 | */ | |
1777 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | |
1778 | { | |
1779 | unsigned long end; | |
1780 | struct vm_area_struct *mpnt, *prev, *last; | |
1781 | ||
1782 | if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) | |
1783 | return -EINVAL; | |
1784 | ||
1785 | if ((len = PAGE_ALIGN(len)) == 0) | |
1786 | return -EINVAL; | |
1787 | ||
1788 | /* Find the first overlapping VMA */ | |
1789 | mpnt = find_vma_prev(mm, start, &prev); | |
1790 | if (!mpnt) | |
1791 | return 0; | |
1792 | /* we have start < mpnt->vm_end */ | |
1793 | ||
1794 | /* if it doesn't overlap, we have nothing.. */ | |
1795 | end = start + len; | |
1796 | if (mpnt->vm_start >= end) | |
1797 | return 0; | |
1798 | ||
1799 | /* | |
1800 | * If we need to split any vma, do it now to save pain later. | |
1801 | * | |
1802 | * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially | |
1803 | * unmapped vm_area_struct will remain in use: so lower split_vma | |
1804 | * places tmp vma above, and higher split_vma places tmp vma below. | |
1805 | */ | |
1806 | if (start > mpnt->vm_start) { | |
1807 | int error = split_vma(mm, mpnt, start, 0); | |
1808 | if (error) | |
1809 | return error; | |
1810 | prev = mpnt; | |
1811 | } | |
1812 | ||
1813 | /* Does it split the last one? */ | |
1814 | last = find_vma(mm, end); | |
1815 | if (last && end > last->vm_start) { | |
1816 | int error = split_vma(mm, last, end, 1); | |
1817 | if (error) | |
1818 | return error; | |
1819 | } | |
1820 | mpnt = prev? prev->vm_next: mm->mmap; | |
1821 | ||
1822 | /* | |
1823 | * Remove the vma's, and unmap the actual pages | |
1824 | */ | |
1825 | detach_vmas_to_be_unmapped(mm, mpnt, prev, end); | |
1826 | spin_lock(&mm->page_table_lock); | |
1827 | unmap_region(mm, mpnt, prev, start, end); | |
1828 | spin_unlock(&mm->page_table_lock); | |
1829 | ||
1830 | /* Fix up all other VM information */ | |
1831 | unmap_vma_list(mm, mpnt); | |
1832 | ||
1833 | return 0; | |
1834 | } | |
1835 | ||
1836 | EXPORT_SYMBOL(do_munmap); | |
1837 | ||
1838 | asmlinkage long sys_munmap(unsigned long addr, size_t len) | |
1839 | { | |
1840 | int ret; | |
1841 | struct mm_struct *mm = current->mm; | |
1842 | ||
1843 | profile_munmap(addr); | |
1844 | ||
1845 | down_write(&mm->mmap_sem); | |
1846 | ret = do_munmap(mm, addr, len); | |
1847 | up_write(&mm->mmap_sem); | |
1848 | return ret; | |
1849 | } | |
1850 | ||
1851 | static inline void verify_mm_writelocked(struct mm_struct *mm) | |
1852 | { | |
1853 | #ifdef CONFIG_DEBUG_KERNEL | |
1854 | if (unlikely(down_read_trylock(&mm->mmap_sem))) { | |
1855 | WARN_ON(1); | |
1856 | up_read(&mm->mmap_sem); | |
1857 | } | |
1858 | #endif | |
1859 | } | |
1860 | ||
1861 | /* | |
1862 | * this is really a simplified "do_mmap". it only handles | |
1863 | * anonymous maps. eventually we may be able to do some | |
1864 | * brk-specific accounting here. | |
1865 | */ | |
1866 | unsigned long do_brk(unsigned long addr, unsigned long len) | |
1867 | { | |
1868 | struct mm_struct * mm = current->mm; | |
1869 | struct vm_area_struct * vma, * prev; | |
1870 | unsigned long flags; | |
1871 | struct rb_node ** rb_link, * rb_parent; | |
1872 | pgoff_t pgoff = addr >> PAGE_SHIFT; | |
1873 | ||
1874 | len = PAGE_ALIGN(len); | |
1875 | if (!len) | |
1876 | return addr; | |
1877 | ||
1878 | if ((addr + len) > TASK_SIZE || (addr + len) < addr) | |
1879 | return -EINVAL; | |
1880 | ||
1881 | /* | |
1882 | * mlock MCL_FUTURE? | |
1883 | */ | |
1884 | if (mm->def_flags & VM_LOCKED) { | |
1885 | unsigned long locked, lock_limit; | |
1886 | locked = mm->locked_vm << PAGE_SHIFT; | |
1887 | lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; | |
1888 | locked += len; | |
1889 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) | |
1890 | return -EAGAIN; | |
1891 | } | |
1892 | ||
1893 | /* | |
1894 | * mm->mmap_sem is required to protect against another thread | |
1895 | * changing the mappings in case we sleep. | |
1896 | */ | |
1897 | verify_mm_writelocked(mm); | |
1898 | ||
1899 | /* | |
1900 | * Clear old maps. this also does some error checking for us | |
1901 | */ | |
1902 | munmap_back: | |
1903 | vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); | |
1904 | if (vma && vma->vm_start < addr + len) { | |
1905 | if (do_munmap(mm, addr, len)) | |
1906 | return -ENOMEM; | |
1907 | goto munmap_back; | |
1908 | } | |
1909 | ||
1910 | /* Check against address space limits *after* clearing old maps... */ | |
1911 | if ((mm->total_vm << PAGE_SHIFT) + len | |
1912 | > current->signal->rlim[RLIMIT_AS].rlim_cur) | |
1913 | return -ENOMEM; | |
1914 | ||
1915 | if (mm->map_count > sysctl_max_map_count) | |
1916 | return -ENOMEM; | |
1917 | ||
1918 | if (security_vm_enough_memory(len >> PAGE_SHIFT)) | |
1919 | return -ENOMEM; | |
1920 | ||
1921 | flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; | |
1922 | ||
1923 | /* Can we just expand an old private anonymous mapping? */ | |
1924 | if (vma_merge(mm, prev, addr, addr + len, flags, | |
1925 | NULL, NULL, pgoff, NULL)) | |
1926 | goto out; | |
1927 | ||
1928 | /* | |
1929 | * create a vma struct for an anonymous mapping | |
1930 | */ | |
1931 | vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL); | |
1932 | if (!vma) { | |
1933 | vm_unacct_memory(len >> PAGE_SHIFT); | |
1934 | return -ENOMEM; | |
1935 | } | |
1936 | memset(vma, 0, sizeof(*vma)); | |
1937 | ||
1938 | vma->vm_mm = mm; | |
1939 | vma->vm_start = addr; | |
1940 | vma->vm_end = addr + len; | |
1941 | vma->vm_pgoff = pgoff; | |
1942 | vma->vm_flags = flags; | |
1943 | vma->vm_page_prot = protection_map[flags & 0x0f]; | |
1944 | vma_link(mm, vma, prev, rb_link, rb_parent); | |
1945 | out: | |
1946 | mm->total_vm += len >> PAGE_SHIFT; | |
1947 | if (flags & VM_LOCKED) { | |
1948 | mm->locked_vm += len >> PAGE_SHIFT; | |
1949 | make_pages_present(addr, addr + len); | |
1950 | } | |
1951 | return addr; | |
1952 | } | |
1953 | ||
1954 | EXPORT_SYMBOL(do_brk); | |
1955 | ||
1956 | /* Release all mmaps. */ | |
1957 | void exit_mmap(struct mm_struct *mm) | |
1958 | { | |
1959 | struct mmu_gather *tlb; | |
1960 | struct vm_area_struct *vma; | |
1961 | unsigned long nr_accounted = 0; | |
1962 | ||
1963 | lru_add_drain(); | |
1964 | ||
1965 | spin_lock(&mm->page_table_lock); | |
1966 | ||
1967 | tlb = tlb_gather_mmu(mm, 1); | |
1968 | flush_cache_mm(mm); | |
1969 | /* Use ~0UL here to ensure all VMAs in the mm are unmapped */ | |
1970 | mm->map_count -= unmap_vmas(&tlb, mm, mm->mmap, 0, | |
1971 | ~0UL, &nr_accounted, NULL); | |
1972 | vm_unacct_memory(nr_accounted); | |
1973 | BUG_ON(mm->map_count); /* This is just debugging */ | |
1974 | clear_page_range(tlb, FIRST_USER_PGD_NR * PGDIR_SIZE, MM_VM_SIZE(mm)); | |
1975 | ||
1976 | tlb_finish_mmu(tlb, 0, MM_VM_SIZE(mm)); | |
1977 | ||
1978 | vma = mm->mmap; | |
1979 | mm->mmap = mm->mmap_cache = NULL; | |
1980 | mm->mm_rb = RB_ROOT; | |
1981 | set_mm_counter(mm, rss, 0); | |
1982 | mm->total_vm = 0; | |
1983 | mm->locked_vm = 0; | |
1984 | ||
1985 | spin_unlock(&mm->page_table_lock); | |
1986 | ||
1987 | /* | |
1988 | * Walk the list again, actually closing and freeing it | |
1989 | * without holding any MM locks. | |
1990 | */ | |
1991 | while (vma) { | |
1992 | struct vm_area_struct *next = vma->vm_next; | |
1993 | remove_vm_struct(vma); | |
1994 | vma = next; | |
1995 | } | |
1996 | } | |
1997 | ||
1998 | /* Insert vm structure into process list sorted by address | |
1999 | * and into the inode's i_mmap tree. If vm_file is non-NULL | |
2000 | * then i_mmap_lock is taken here. | |
2001 | */ | |
2002 | int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma) | |
2003 | { | |
2004 | struct vm_area_struct * __vma, * prev; | |
2005 | struct rb_node ** rb_link, * rb_parent; | |
2006 | ||
2007 | /* | |
2008 | * The vm_pgoff of a purely anonymous vma should be irrelevant | |
2009 | * until its first write fault, when page's anon_vma and index | |
2010 | * are set. But now set the vm_pgoff it will almost certainly | |
2011 | * end up with (unless mremap moves it elsewhere before that | |
2012 | * first wfault), so /proc/pid/maps tells a consistent story. | |
2013 | * | |
2014 | * By setting it to reflect the virtual start address of the | |
2015 | * vma, merges and splits can happen in a seamless way, just | |
2016 | * using the existing file pgoff checks and manipulations. | |
2017 | * Similarly in do_mmap_pgoff and in do_brk. | |
2018 | */ | |
2019 | if (!vma->vm_file) { | |
2020 | BUG_ON(vma->anon_vma); | |
2021 | vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; | |
2022 | } | |
2023 | __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent); | |
2024 | if (__vma && __vma->vm_start < vma->vm_end) | |
2025 | return -ENOMEM; | |
2026 | vma_link(mm, vma, prev, rb_link, rb_parent); | |
2027 | return 0; | |
2028 | } | |
2029 | ||
2030 | /* | |
2031 | * Copy the vma structure to a new location in the same mm, | |
2032 | * prior to moving page table entries, to effect an mremap move. | |
2033 | */ | |
2034 | struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, | |
2035 | unsigned long addr, unsigned long len, pgoff_t pgoff) | |
2036 | { | |
2037 | struct vm_area_struct *vma = *vmap; | |
2038 | unsigned long vma_start = vma->vm_start; | |
2039 | struct mm_struct *mm = vma->vm_mm; | |
2040 | struct vm_area_struct *new_vma, *prev; | |
2041 | struct rb_node **rb_link, *rb_parent; | |
2042 | struct mempolicy *pol; | |
2043 | ||
2044 | /* | |
2045 | * If anonymous vma has not yet been faulted, update new pgoff | |
2046 | * to match new location, to increase its chance of merging. | |
2047 | */ | |
2048 | if (!vma->vm_file && !vma->anon_vma) | |
2049 | pgoff = addr >> PAGE_SHIFT; | |
2050 | ||
2051 | find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); | |
2052 | new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, | |
2053 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); | |
2054 | if (new_vma) { | |
2055 | /* | |
2056 | * Source vma may have been merged into new_vma | |
2057 | */ | |
2058 | if (vma_start >= new_vma->vm_start && | |
2059 | vma_start < new_vma->vm_end) | |
2060 | *vmap = new_vma; | |
2061 | } else { | |
2062 | new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL); | |
2063 | if (new_vma) { | |
2064 | *new_vma = *vma; | |
2065 | pol = mpol_copy(vma_policy(vma)); | |
2066 | if (IS_ERR(pol)) { | |
2067 | kmem_cache_free(vm_area_cachep, new_vma); | |
2068 | return NULL; | |
2069 | } | |
2070 | vma_set_policy(new_vma, pol); | |
2071 | new_vma->vm_start = addr; | |
2072 | new_vma->vm_end = addr + len; | |
2073 | new_vma->vm_pgoff = pgoff; | |
2074 | if (new_vma->vm_file) | |
2075 | get_file(new_vma->vm_file); | |
2076 | if (new_vma->vm_ops && new_vma->vm_ops->open) | |
2077 | new_vma->vm_ops->open(new_vma); | |
2078 | vma_link(mm, new_vma, prev, rb_link, rb_parent); | |
2079 | } | |
2080 | } | |
2081 | return new_vma; | |
2082 | } |