]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/mmap.c | |
3 | * | |
4 | * Written by obz. | |
5 | * | |
046c6884 | 6 | * Address space accounting code <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
7 | */ |
8 | ||
9 | #include <linux/slab.h> | |
4af3c9cc | 10 | #include <linux/backing-dev.h> |
1da177e4 LT |
11 | #include <linux/mm.h> |
12 | #include <linux/shm.h> | |
13 | #include <linux/mman.h> | |
14 | #include <linux/pagemap.h> | |
15 | #include <linux/swap.h> | |
16 | #include <linux/syscalls.h> | |
c59ede7b | 17 | #include <linux/capability.h> |
1da177e4 LT |
18 | #include <linux/init.h> |
19 | #include <linux/file.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/personality.h> | |
22 | #include <linux/security.h> | |
23 | #include <linux/hugetlb.h> | |
24 | #include <linux/profile.h> | |
b95f1b31 | 25 | #include <linux/export.h> |
1da177e4 LT |
26 | #include <linux/mount.h> |
27 | #include <linux/mempolicy.h> | |
28 | #include <linux/rmap.h> | |
cddb8a5c | 29 | #include <linux/mmu_notifier.h> |
cdd6c482 | 30 | #include <linux/perf_event.h> |
120a795d | 31 | #include <linux/audit.h> |
b15d00b6 | 32 | #include <linux/khugepaged.h> |
2b144498 | 33 | #include <linux/uprobes.h> |
d3737187 | 34 | #include <linux/rbtree_augmented.h> |
1da177e4 LT |
35 | |
36 | #include <asm/uaccess.h> | |
37 | #include <asm/cacheflush.h> | |
38 | #include <asm/tlb.h> | |
d6dd61c8 | 39 | #include <asm/mmu_context.h> |
1da177e4 | 40 | |
42b77728 JB |
41 | #include "internal.h" |
42 | ||
3a459756 KK |
43 | #ifndef arch_mmap_check |
44 | #define arch_mmap_check(addr, len, flags) (0) | |
45 | #endif | |
46 | ||
08e7d9b5 MS |
47 | #ifndef arch_rebalance_pgtables |
48 | #define arch_rebalance_pgtables(addr, len) (addr) | |
49 | #endif | |
50 | ||
e0da382c HD |
51 | static void unmap_region(struct mm_struct *mm, |
52 | struct vm_area_struct *vma, struct vm_area_struct *prev, | |
53 | unsigned long start, unsigned long end); | |
54 | ||
1da177e4 LT |
55 | /* description of effects of mapping type and prot in current implementation. |
56 | * this is due to the limited x86 page protection hardware. The expected | |
57 | * behavior is in parens: | |
58 | * | |
59 | * map_type prot | |
60 | * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC | |
61 | * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes | |
62 | * w: (no) no w: (no) no w: (yes) yes w: (no) no | |
63 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
64 | * | |
65 | * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes | |
66 | * w: (no) no w: (no) no w: (copy) copy w: (no) no | |
67 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
68 | * | |
69 | */ | |
70 | pgprot_t protection_map[16] = { | |
71 | __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, | |
72 | __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 | |
73 | }; | |
74 | ||
804af2cf HD |
75 | pgprot_t vm_get_page_prot(unsigned long vm_flags) |
76 | { | |
b845f313 DK |
77 | return __pgprot(pgprot_val(protection_map[vm_flags & |
78 | (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | | |
79 | pgprot_val(arch_vm_get_page_prot(vm_flags))); | |
804af2cf HD |
80 | } |
81 | EXPORT_SYMBOL(vm_get_page_prot); | |
82 | ||
34679d7e SL |
83 | int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */ |
84 | int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */ | |
c3d8c141 | 85 | int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; |
34679d7e SL |
86 | /* |
87 | * Make sure vm_committed_as in one cacheline and not cacheline shared with | |
88 | * other variables. It can be updated by several CPUs frequently. | |
89 | */ | |
90 | struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; | |
1da177e4 LT |
91 | |
92 | /* | |
93 | * Check that a process has enough memory to allocate a new virtual | |
94 | * mapping. 0 means there is enough memory for the allocation to | |
95 | * succeed and -ENOMEM implies there is not. | |
96 | * | |
97 | * We currently support three overcommit policies, which are set via the | |
98 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting | |
99 | * | |
100 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. | |
101 | * Additional code 2002 Jul 20 by Robert Love. | |
102 | * | |
103 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. | |
104 | * | |
105 | * Note this is a helper function intended to be used by LSMs which | |
106 | * wish to use this logic. | |
107 | */ | |
34b4e4aa | 108 | int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) |
1da177e4 LT |
109 | { |
110 | unsigned long free, allowed; | |
111 | ||
112 | vm_acct_memory(pages); | |
113 | ||
114 | /* | |
115 | * Sometimes we want to use more memory than we have | |
116 | */ | |
117 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) | |
118 | return 0; | |
119 | ||
120 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { | |
c15bef30 DF |
121 | free = global_page_state(NR_FREE_PAGES); |
122 | free += global_page_state(NR_FILE_PAGES); | |
123 | ||
124 | /* | |
125 | * shmem pages shouldn't be counted as free in this | |
126 | * case, they can't be purged, only swapped out, and | |
127 | * that won't affect the overall amount of available | |
128 | * memory in the system. | |
129 | */ | |
130 | free -= global_page_state(NR_SHMEM); | |
1da177e4 | 131 | |
1da177e4 LT |
132 | free += nr_swap_pages; |
133 | ||
134 | /* | |
135 | * Any slabs which are created with the | |
136 | * SLAB_RECLAIM_ACCOUNT flag claim to have contents | |
137 | * which are reclaimable, under pressure. The dentry | |
138 | * cache and most inode caches should fall into this | |
139 | */ | |
972d1a7b | 140 | free += global_page_state(NR_SLAB_RECLAIMABLE); |
1da177e4 | 141 | |
6d9f7839 HA |
142 | /* |
143 | * Leave reserved pages. The pages are not for anonymous pages. | |
144 | */ | |
c15bef30 | 145 | if (free <= totalreserve_pages) |
6d9f7839 HA |
146 | goto error; |
147 | else | |
c15bef30 | 148 | free -= totalreserve_pages; |
6d9f7839 HA |
149 | |
150 | /* | |
151 | * Leave the last 3% for root | |
152 | */ | |
1da177e4 | 153 | if (!cap_sys_admin) |
c15bef30 | 154 | free -= free / 32; |
1da177e4 LT |
155 | |
156 | if (free > pages) | |
157 | return 0; | |
6d9f7839 HA |
158 | |
159 | goto error; | |
1da177e4 LT |
160 | } |
161 | ||
162 | allowed = (totalram_pages - hugetlb_total_pages()) | |
163 | * sysctl_overcommit_ratio / 100; | |
164 | /* | |
165 | * Leave the last 3% for root | |
166 | */ | |
167 | if (!cap_sys_admin) | |
168 | allowed -= allowed / 32; | |
169 | allowed += total_swap_pages; | |
170 | ||
171 | /* Don't let a single process grow too big: | |
172 | leave 3% of the size of this process for other processes */ | |
731572d3 AC |
173 | if (mm) |
174 | allowed -= mm->total_vm / 32; | |
1da177e4 | 175 | |
00a62ce9 | 176 | if (percpu_counter_read_positive(&vm_committed_as) < allowed) |
1da177e4 | 177 | return 0; |
6d9f7839 | 178 | error: |
1da177e4 LT |
179 | vm_unacct_memory(pages); |
180 | ||
181 | return -ENOMEM; | |
182 | } | |
183 | ||
1da177e4 | 184 | /* |
3d48ae45 | 185 | * Requires inode->i_mapping->i_mmap_mutex |
1da177e4 LT |
186 | */ |
187 | static void __remove_shared_vm_struct(struct vm_area_struct *vma, | |
188 | struct file *file, struct address_space *mapping) | |
189 | { | |
190 | if (vma->vm_flags & VM_DENYWRITE) | |
d3ac7f89 | 191 | atomic_inc(&file->f_path.dentry->d_inode->i_writecount); |
1da177e4 LT |
192 | if (vma->vm_flags & VM_SHARED) |
193 | mapping->i_mmap_writable--; | |
194 | ||
195 | flush_dcache_mmap_lock(mapping); | |
196 | if (unlikely(vma->vm_flags & VM_NONLINEAR)) | |
6b2dbba8 | 197 | list_del_init(&vma->shared.nonlinear); |
1da177e4 | 198 | else |
6b2dbba8 | 199 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
1da177e4 LT |
200 | flush_dcache_mmap_unlock(mapping); |
201 | } | |
202 | ||
203 | /* | |
6b2dbba8 | 204 | * Unlink a file-based vm structure from its interval tree, to hide |
a8fb5618 | 205 | * vma from rmap and vmtruncate before freeing its page tables. |
1da177e4 | 206 | */ |
a8fb5618 | 207 | void unlink_file_vma(struct vm_area_struct *vma) |
1da177e4 LT |
208 | { |
209 | struct file *file = vma->vm_file; | |
210 | ||
1da177e4 LT |
211 | if (file) { |
212 | struct address_space *mapping = file->f_mapping; | |
3d48ae45 | 213 | mutex_lock(&mapping->i_mmap_mutex); |
1da177e4 | 214 | __remove_shared_vm_struct(vma, file, mapping); |
3d48ae45 | 215 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 | 216 | } |
a8fb5618 HD |
217 | } |
218 | ||
219 | /* | |
220 | * Close a vm structure and free it, returning the next. | |
221 | */ | |
222 | static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) | |
223 | { | |
224 | struct vm_area_struct *next = vma->vm_next; | |
225 | ||
a8fb5618 | 226 | might_sleep(); |
1da177e4 LT |
227 | if (vma->vm_ops && vma->vm_ops->close) |
228 | vma->vm_ops->close(vma); | |
e9714acf | 229 | if (vma->vm_file) |
a8fb5618 | 230 | fput(vma->vm_file); |
f0be3d32 | 231 | mpol_put(vma_policy(vma)); |
1da177e4 | 232 | kmem_cache_free(vm_area_cachep, vma); |
a8fb5618 | 233 | return next; |
1da177e4 LT |
234 | } |
235 | ||
e4eb1ff6 LT |
236 | static unsigned long do_brk(unsigned long addr, unsigned long len); |
237 | ||
6a6160a7 | 238 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 LT |
239 | { |
240 | unsigned long rlim, retval; | |
241 | unsigned long newbrk, oldbrk; | |
242 | struct mm_struct *mm = current->mm; | |
a5b4592c | 243 | unsigned long min_brk; |
1da177e4 LT |
244 | |
245 | down_write(&mm->mmap_sem); | |
246 | ||
a5b4592c | 247 | #ifdef CONFIG_COMPAT_BRK |
5520e894 JK |
248 | /* |
249 | * CONFIG_COMPAT_BRK can still be overridden by setting | |
250 | * randomize_va_space to 2, which will still cause mm->start_brk | |
251 | * to be arbitrarily shifted | |
252 | */ | |
4471a675 | 253 | if (current->brk_randomized) |
5520e894 JK |
254 | min_brk = mm->start_brk; |
255 | else | |
256 | min_brk = mm->end_data; | |
a5b4592c JK |
257 | #else |
258 | min_brk = mm->start_brk; | |
259 | #endif | |
260 | if (brk < min_brk) | |
1da177e4 | 261 | goto out; |
1e624196 RG |
262 | |
263 | /* | |
264 | * Check against rlimit here. If this check is done later after the test | |
265 | * of oldbrk with newbrk then it can escape the test and let the data | |
266 | * segment grow beyond its set limit the in case where the limit is | |
267 | * not page aligned -Ram Gupta | |
268 | */ | |
59e99e5b | 269 | rlim = rlimit(RLIMIT_DATA); |
c1d171a0 JK |
270 | if (rlim < RLIM_INFINITY && (brk - mm->start_brk) + |
271 | (mm->end_data - mm->start_data) > rlim) | |
1e624196 RG |
272 | goto out; |
273 | ||
1da177e4 LT |
274 | newbrk = PAGE_ALIGN(brk); |
275 | oldbrk = PAGE_ALIGN(mm->brk); | |
276 | if (oldbrk == newbrk) | |
277 | goto set_brk; | |
278 | ||
279 | /* Always allow shrinking brk. */ | |
280 | if (brk <= mm->brk) { | |
281 | if (!do_munmap(mm, newbrk, oldbrk-newbrk)) | |
282 | goto set_brk; | |
283 | goto out; | |
284 | } | |
285 | ||
1da177e4 LT |
286 | /* Check against existing mmap mappings. */ |
287 | if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) | |
288 | goto out; | |
289 | ||
290 | /* Ok, looks good - let it rip. */ | |
291 | if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) | |
292 | goto out; | |
293 | set_brk: | |
294 | mm->brk = brk; | |
295 | out: | |
296 | retval = mm->brk; | |
297 | up_write(&mm->mmap_sem); | |
298 | return retval; | |
299 | } | |
300 | ||
d3737187 ML |
301 | static long vma_compute_subtree_gap(struct vm_area_struct *vma) |
302 | { | |
303 | unsigned long max, subtree_gap; | |
304 | max = vma->vm_start; | |
305 | if (vma->vm_prev) | |
306 | max -= vma->vm_prev->vm_end; | |
307 | if (vma->vm_rb.rb_left) { | |
308 | subtree_gap = rb_entry(vma->vm_rb.rb_left, | |
309 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
310 | if (subtree_gap > max) | |
311 | max = subtree_gap; | |
312 | } | |
313 | if (vma->vm_rb.rb_right) { | |
314 | subtree_gap = rb_entry(vma->vm_rb.rb_right, | |
315 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
316 | if (subtree_gap > max) | |
317 | max = subtree_gap; | |
318 | } | |
319 | return max; | |
320 | } | |
321 | ||
ed8ea815 | 322 | #ifdef CONFIG_DEBUG_VM_RB |
1da177e4 LT |
323 | static int browse_rb(struct rb_root *root) |
324 | { | |
5a0768f6 | 325 | int i = 0, j, bug = 0; |
1da177e4 LT |
326 | struct rb_node *nd, *pn = NULL; |
327 | unsigned long prev = 0, pend = 0; | |
328 | ||
329 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
330 | struct vm_area_struct *vma; | |
331 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
5a0768f6 ML |
332 | if (vma->vm_start < prev) { |
333 | printk("vm_start %lx prev %lx\n", vma->vm_start, prev); | |
334 | bug = 1; | |
335 | } | |
336 | if (vma->vm_start < pend) { | |
1da177e4 | 337 | printk("vm_start %lx pend %lx\n", vma->vm_start, pend); |
5a0768f6 ML |
338 | bug = 1; |
339 | } | |
340 | if (vma->vm_start > vma->vm_end) { | |
341 | printk("vm_end %lx < vm_start %lx\n", | |
342 | vma->vm_end, vma->vm_start); | |
343 | bug = 1; | |
344 | } | |
345 | if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { | |
346 | printk("free gap %lx, correct %lx\n", | |
347 | vma->rb_subtree_gap, | |
348 | vma_compute_subtree_gap(vma)); | |
349 | bug = 1; | |
350 | } | |
1da177e4 LT |
351 | i++; |
352 | pn = nd; | |
d1af65d1 DM |
353 | prev = vma->vm_start; |
354 | pend = vma->vm_end; | |
1da177e4 LT |
355 | } |
356 | j = 0; | |
5a0768f6 | 357 | for (nd = pn; nd; nd = rb_prev(nd)) |
1da177e4 | 358 | j++; |
5a0768f6 ML |
359 | if (i != j) { |
360 | printk("backwards %d, forwards %d\n", j, i); | |
361 | bug = 1; | |
1da177e4 | 362 | } |
5a0768f6 | 363 | return bug ? -1 : i; |
1da177e4 LT |
364 | } |
365 | ||
d3737187 ML |
366 | static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) |
367 | { | |
368 | struct rb_node *nd; | |
369 | ||
370 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
371 | struct vm_area_struct *vma; | |
372 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
373 | BUG_ON(vma != ignore && | |
374 | vma->rb_subtree_gap != vma_compute_subtree_gap(vma)); | |
375 | } | |
376 | } | |
377 | ||
1da177e4 LT |
378 | void validate_mm(struct mm_struct *mm) |
379 | { | |
380 | int bug = 0; | |
381 | int i = 0; | |
5a0768f6 | 382 | unsigned long highest_address = 0; |
ed8ea815 ML |
383 | struct vm_area_struct *vma = mm->mmap; |
384 | while (vma) { | |
385 | struct anon_vma_chain *avc; | |
63c3b902 | 386 | vma_lock_anon_vma(vma); |
ed8ea815 ML |
387 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
388 | anon_vma_interval_tree_verify(avc); | |
63c3b902 | 389 | vma_unlock_anon_vma(vma); |
5a0768f6 | 390 | highest_address = vma->vm_end; |
ed8ea815 | 391 | vma = vma->vm_next; |
1da177e4 LT |
392 | i++; |
393 | } | |
5a0768f6 ML |
394 | if (i != mm->map_count) { |
395 | printk("map_count %d vm_next %d\n", mm->map_count, i); | |
396 | bug = 1; | |
397 | } | |
398 | if (highest_address != mm->highest_vm_end) { | |
399 | printk("mm->highest_vm_end %lx, found %lx\n", | |
400 | mm->highest_vm_end, highest_address); | |
401 | bug = 1; | |
402 | } | |
1da177e4 | 403 | i = browse_rb(&mm->mm_rb); |
5a0768f6 ML |
404 | if (i != mm->map_count) { |
405 | printk("map_count %d rb %d\n", mm->map_count, i); | |
406 | bug = 1; | |
407 | } | |
46a350ef | 408 | BUG_ON(bug); |
1da177e4 LT |
409 | } |
410 | #else | |
d3737187 | 411 | #define validate_mm_rb(root, ignore) do { } while (0) |
1da177e4 LT |
412 | #define validate_mm(mm) do { } while (0) |
413 | #endif | |
414 | ||
d3737187 ML |
415 | RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, |
416 | unsigned long, rb_subtree_gap, vma_compute_subtree_gap) | |
417 | ||
418 | /* | |
419 | * Update augmented rbtree rb_subtree_gap values after vma->vm_start or | |
420 | * vma->vm_prev->vm_end values changed, without modifying the vma's position | |
421 | * in the rbtree. | |
422 | */ | |
423 | static void vma_gap_update(struct vm_area_struct *vma) | |
424 | { | |
425 | /* | |
426 | * As it turns out, RB_DECLARE_CALLBACKS() already created a callback | |
427 | * function that does exacltly what we want. | |
428 | */ | |
429 | vma_gap_callbacks_propagate(&vma->vm_rb, NULL); | |
430 | } | |
431 | ||
432 | static inline void vma_rb_insert(struct vm_area_struct *vma, | |
433 | struct rb_root *root) | |
434 | { | |
435 | /* All rb_subtree_gap values must be consistent prior to insertion */ | |
436 | validate_mm_rb(root, NULL); | |
437 | ||
438 | rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
439 | } | |
440 | ||
441 | static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) | |
442 | { | |
443 | /* | |
444 | * All rb_subtree_gap values must be consistent prior to erase, | |
445 | * with the possible exception of the vma being erased. | |
446 | */ | |
447 | validate_mm_rb(root, vma); | |
448 | ||
449 | /* | |
450 | * Note rb_erase_augmented is a fairly large inline function, | |
451 | * so make sure we instantiate it only once with our desired | |
452 | * augmented rbtree callbacks. | |
453 | */ | |
454 | rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
455 | } | |
456 | ||
bf181b9f ML |
457 | /* |
458 | * vma has some anon_vma assigned, and is already inserted on that | |
459 | * anon_vma's interval trees. | |
460 | * | |
461 | * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the | |
462 | * vma must be removed from the anon_vma's interval trees using | |
463 | * anon_vma_interval_tree_pre_update_vma(). | |
464 | * | |
465 | * After the update, the vma will be reinserted using | |
466 | * anon_vma_interval_tree_post_update_vma(). | |
467 | * | |
468 | * The entire update must be protected by exclusive mmap_sem and by | |
469 | * the root anon_vma's mutex. | |
470 | */ | |
471 | static inline void | |
472 | anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) | |
473 | { | |
474 | struct anon_vma_chain *avc; | |
475 | ||
476 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
477 | anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); | |
478 | } | |
479 | ||
480 | static inline void | |
481 | anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) | |
482 | { | |
483 | struct anon_vma_chain *avc; | |
484 | ||
485 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
486 | anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); | |
487 | } | |
488 | ||
6597d783 HD |
489 | static int find_vma_links(struct mm_struct *mm, unsigned long addr, |
490 | unsigned long end, struct vm_area_struct **pprev, | |
491 | struct rb_node ***rb_link, struct rb_node **rb_parent) | |
1da177e4 | 492 | { |
6597d783 | 493 | struct rb_node **__rb_link, *__rb_parent, *rb_prev; |
1da177e4 LT |
494 | |
495 | __rb_link = &mm->mm_rb.rb_node; | |
496 | rb_prev = __rb_parent = NULL; | |
1da177e4 LT |
497 | |
498 | while (*__rb_link) { | |
499 | struct vm_area_struct *vma_tmp; | |
500 | ||
501 | __rb_parent = *__rb_link; | |
502 | vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); | |
503 | ||
504 | if (vma_tmp->vm_end > addr) { | |
6597d783 HD |
505 | /* Fail if an existing vma overlaps the area */ |
506 | if (vma_tmp->vm_start < end) | |
507 | return -ENOMEM; | |
1da177e4 LT |
508 | __rb_link = &__rb_parent->rb_left; |
509 | } else { | |
510 | rb_prev = __rb_parent; | |
511 | __rb_link = &__rb_parent->rb_right; | |
512 | } | |
513 | } | |
514 | ||
515 | *pprev = NULL; | |
516 | if (rb_prev) | |
517 | *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); | |
518 | *rb_link = __rb_link; | |
519 | *rb_parent = __rb_parent; | |
6597d783 | 520 | return 0; |
1da177e4 LT |
521 | } |
522 | ||
1da177e4 LT |
523 | void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, |
524 | struct rb_node **rb_link, struct rb_node *rb_parent) | |
525 | { | |
d3737187 ML |
526 | /* Update tracking information for the gap following the new vma. */ |
527 | if (vma->vm_next) | |
528 | vma_gap_update(vma->vm_next); | |
529 | else | |
530 | mm->highest_vm_end = vma->vm_end; | |
531 | ||
532 | /* | |
533 | * vma->vm_prev wasn't known when we followed the rbtree to find the | |
534 | * correct insertion point for that vma. As a result, we could not | |
535 | * update the vma vm_rb parents rb_subtree_gap values on the way down. | |
536 | * So, we first insert the vma with a zero rb_subtree_gap value | |
537 | * (to be consistent with what we did on the way down), and then | |
538 | * immediately update the gap to the correct value. Finally we | |
539 | * rebalance the rbtree after all augmented values have been set. | |
540 | */ | |
1da177e4 | 541 | rb_link_node(&vma->vm_rb, rb_parent, rb_link); |
d3737187 ML |
542 | vma->rb_subtree_gap = 0; |
543 | vma_gap_update(vma); | |
544 | vma_rb_insert(vma, &mm->mm_rb); | |
1da177e4 LT |
545 | } |
546 | ||
cb8f488c | 547 | static void __vma_link_file(struct vm_area_struct *vma) |
1da177e4 | 548 | { |
48aae425 | 549 | struct file *file; |
1da177e4 LT |
550 | |
551 | file = vma->vm_file; | |
552 | if (file) { | |
553 | struct address_space *mapping = file->f_mapping; | |
554 | ||
555 | if (vma->vm_flags & VM_DENYWRITE) | |
d3ac7f89 | 556 | atomic_dec(&file->f_path.dentry->d_inode->i_writecount); |
1da177e4 LT |
557 | if (vma->vm_flags & VM_SHARED) |
558 | mapping->i_mmap_writable++; | |
559 | ||
560 | flush_dcache_mmap_lock(mapping); | |
561 | if (unlikely(vma->vm_flags & VM_NONLINEAR)) | |
562 | vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); | |
563 | else | |
6b2dbba8 | 564 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
1da177e4 LT |
565 | flush_dcache_mmap_unlock(mapping); |
566 | } | |
567 | } | |
568 | ||
569 | static void | |
570 | __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
571 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
572 | struct rb_node *rb_parent) | |
573 | { | |
574 | __vma_link_list(mm, vma, prev, rb_parent); | |
575 | __vma_link_rb(mm, vma, rb_link, rb_parent); | |
1da177e4 LT |
576 | } |
577 | ||
578 | static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
579 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
580 | struct rb_node *rb_parent) | |
581 | { | |
582 | struct address_space *mapping = NULL; | |
583 | ||
584 | if (vma->vm_file) | |
585 | mapping = vma->vm_file->f_mapping; | |
586 | ||
97a89413 | 587 | if (mapping) |
3d48ae45 | 588 | mutex_lock(&mapping->i_mmap_mutex); |
1da177e4 LT |
589 | |
590 | __vma_link(mm, vma, prev, rb_link, rb_parent); | |
591 | __vma_link_file(vma); | |
592 | ||
1da177e4 | 593 | if (mapping) |
3d48ae45 | 594 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 LT |
595 | |
596 | mm->map_count++; | |
597 | validate_mm(mm); | |
598 | } | |
599 | ||
600 | /* | |
88f6b4c3 | 601 | * Helper for vma_adjust() in the split_vma insert case: insert a vma into the |
6b2dbba8 | 602 | * mm's list and rbtree. It has already been inserted into the interval tree. |
1da177e4 | 603 | */ |
48aae425 | 604 | static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 605 | { |
6597d783 | 606 | struct vm_area_struct *prev; |
48aae425 | 607 | struct rb_node **rb_link, *rb_parent; |
1da177e4 | 608 | |
6597d783 HD |
609 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
610 | &prev, &rb_link, &rb_parent)) | |
611 | BUG(); | |
1da177e4 LT |
612 | __vma_link(mm, vma, prev, rb_link, rb_parent); |
613 | mm->map_count++; | |
614 | } | |
615 | ||
616 | static inline void | |
617 | __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, | |
618 | struct vm_area_struct *prev) | |
619 | { | |
d3737187 | 620 | struct vm_area_struct *next; |
297c5eee | 621 | |
d3737187 ML |
622 | vma_rb_erase(vma, &mm->mm_rb); |
623 | prev->vm_next = next = vma->vm_next; | |
297c5eee LT |
624 | if (next) |
625 | next->vm_prev = prev; | |
1da177e4 LT |
626 | if (mm->mmap_cache == vma) |
627 | mm->mmap_cache = prev; | |
628 | } | |
629 | ||
630 | /* | |
631 | * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that | |
632 | * is already present in an i_mmap tree without adjusting the tree. | |
633 | * The following helper function should be used when such adjustments | |
634 | * are necessary. The "insert" vma (if any) is to be inserted | |
635 | * before we drop the necessary locks. | |
636 | */ | |
5beb4930 | 637 | int vma_adjust(struct vm_area_struct *vma, unsigned long start, |
1da177e4 LT |
638 | unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) |
639 | { | |
640 | struct mm_struct *mm = vma->vm_mm; | |
641 | struct vm_area_struct *next = vma->vm_next; | |
642 | struct vm_area_struct *importer = NULL; | |
643 | struct address_space *mapping = NULL; | |
6b2dbba8 | 644 | struct rb_root *root = NULL; |
012f1800 | 645 | struct anon_vma *anon_vma = NULL; |
1da177e4 | 646 | struct file *file = vma->vm_file; |
d3737187 | 647 | bool start_changed = false, end_changed = false; |
1da177e4 LT |
648 | long adjust_next = 0; |
649 | int remove_next = 0; | |
650 | ||
651 | if (next && !insert) { | |
287d97ac LT |
652 | struct vm_area_struct *exporter = NULL; |
653 | ||
1da177e4 LT |
654 | if (end >= next->vm_end) { |
655 | /* | |
656 | * vma expands, overlapping all the next, and | |
657 | * perhaps the one after too (mprotect case 6). | |
658 | */ | |
659 | again: remove_next = 1 + (end > next->vm_end); | |
660 | end = next->vm_end; | |
287d97ac | 661 | exporter = next; |
1da177e4 LT |
662 | importer = vma; |
663 | } else if (end > next->vm_start) { | |
664 | /* | |
665 | * vma expands, overlapping part of the next: | |
666 | * mprotect case 5 shifting the boundary up. | |
667 | */ | |
668 | adjust_next = (end - next->vm_start) >> PAGE_SHIFT; | |
287d97ac | 669 | exporter = next; |
1da177e4 LT |
670 | importer = vma; |
671 | } else if (end < vma->vm_end) { | |
672 | /* | |
673 | * vma shrinks, and !insert tells it's not | |
674 | * split_vma inserting another: so it must be | |
675 | * mprotect case 4 shifting the boundary down. | |
676 | */ | |
677 | adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT); | |
287d97ac | 678 | exporter = vma; |
1da177e4 LT |
679 | importer = next; |
680 | } | |
1da177e4 | 681 | |
5beb4930 RR |
682 | /* |
683 | * Easily overlooked: when mprotect shifts the boundary, | |
684 | * make sure the expanding vma has anon_vma set if the | |
685 | * shrinking vma had, to cover any anon pages imported. | |
686 | */ | |
287d97ac LT |
687 | if (exporter && exporter->anon_vma && !importer->anon_vma) { |
688 | if (anon_vma_clone(importer, exporter)) | |
5beb4930 | 689 | return -ENOMEM; |
287d97ac | 690 | importer->anon_vma = exporter->anon_vma; |
5beb4930 RR |
691 | } |
692 | } | |
693 | ||
1da177e4 LT |
694 | if (file) { |
695 | mapping = file->f_mapping; | |
682968e0 | 696 | if (!(vma->vm_flags & VM_NONLINEAR)) { |
1da177e4 | 697 | root = &mapping->i_mmap; |
cbc91f71 | 698 | uprobe_munmap(vma, vma->vm_start, vma->vm_end); |
682968e0 SD |
699 | |
700 | if (adjust_next) | |
cbc91f71 SD |
701 | uprobe_munmap(next, next->vm_start, |
702 | next->vm_end); | |
682968e0 SD |
703 | } |
704 | ||
3d48ae45 | 705 | mutex_lock(&mapping->i_mmap_mutex); |
1da177e4 | 706 | if (insert) { |
1da177e4 | 707 | /* |
6b2dbba8 | 708 | * Put into interval tree now, so instantiated pages |
1da177e4 LT |
709 | * are visible to arm/parisc __flush_dcache_page |
710 | * throughout; but we cannot insert into address | |
711 | * space until vma start or end is updated. | |
712 | */ | |
713 | __vma_link_file(insert); | |
714 | } | |
715 | } | |
716 | ||
94fcc585 AA |
717 | vma_adjust_trans_huge(vma, start, end, adjust_next); |
718 | ||
bf181b9f ML |
719 | anon_vma = vma->anon_vma; |
720 | if (!anon_vma && adjust_next) | |
721 | anon_vma = next->anon_vma; | |
722 | if (anon_vma) { | |
ca42b26a ML |
723 | VM_BUG_ON(adjust_next && next->anon_vma && |
724 | anon_vma != next->anon_vma); | |
012f1800 | 725 | anon_vma_lock(anon_vma); |
bf181b9f ML |
726 | anon_vma_interval_tree_pre_update_vma(vma); |
727 | if (adjust_next) | |
728 | anon_vma_interval_tree_pre_update_vma(next); | |
729 | } | |
012f1800 | 730 | |
1da177e4 LT |
731 | if (root) { |
732 | flush_dcache_mmap_lock(mapping); | |
6b2dbba8 | 733 | vma_interval_tree_remove(vma, root); |
1da177e4 | 734 | if (adjust_next) |
6b2dbba8 | 735 | vma_interval_tree_remove(next, root); |
1da177e4 LT |
736 | } |
737 | ||
d3737187 ML |
738 | if (start != vma->vm_start) { |
739 | vma->vm_start = start; | |
740 | start_changed = true; | |
741 | } | |
742 | if (end != vma->vm_end) { | |
743 | vma->vm_end = end; | |
744 | end_changed = true; | |
745 | } | |
1da177e4 LT |
746 | vma->vm_pgoff = pgoff; |
747 | if (adjust_next) { | |
748 | next->vm_start += adjust_next << PAGE_SHIFT; | |
749 | next->vm_pgoff += adjust_next; | |
750 | } | |
751 | ||
752 | if (root) { | |
753 | if (adjust_next) | |
6b2dbba8 ML |
754 | vma_interval_tree_insert(next, root); |
755 | vma_interval_tree_insert(vma, root); | |
1da177e4 LT |
756 | flush_dcache_mmap_unlock(mapping); |
757 | } | |
758 | ||
759 | if (remove_next) { | |
760 | /* | |
761 | * vma_merge has merged next into vma, and needs | |
762 | * us to remove next before dropping the locks. | |
763 | */ | |
764 | __vma_unlink(mm, next, vma); | |
765 | if (file) | |
766 | __remove_shared_vm_struct(next, file, mapping); | |
1da177e4 LT |
767 | } else if (insert) { |
768 | /* | |
769 | * split_vma has split insert from vma, and needs | |
770 | * us to insert it before dropping the locks | |
771 | * (it may either follow vma or precede it). | |
772 | */ | |
773 | __insert_vm_struct(mm, insert); | |
d3737187 ML |
774 | } else { |
775 | if (start_changed) | |
776 | vma_gap_update(vma); | |
777 | if (end_changed) { | |
778 | if (!next) | |
779 | mm->highest_vm_end = end; | |
780 | else if (!adjust_next) | |
781 | vma_gap_update(next); | |
782 | } | |
1da177e4 LT |
783 | } |
784 | ||
bf181b9f ML |
785 | if (anon_vma) { |
786 | anon_vma_interval_tree_post_update_vma(vma); | |
787 | if (adjust_next) | |
788 | anon_vma_interval_tree_post_update_vma(next); | |
012f1800 | 789 | anon_vma_unlock(anon_vma); |
bf181b9f | 790 | } |
1da177e4 | 791 | if (mapping) |
3d48ae45 | 792 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 | 793 | |
2b144498 | 794 | if (root) { |
7b2d81d4 | 795 | uprobe_mmap(vma); |
2b144498 SD |
796 | |
797 | if (adjust_next) | |
7b2d81d4 | 798 | uprobe_mmap(next); |
2b144498 SD |
799 | } |
800 | ||
1da177e4 | 801 | if (remove_next) { |
925d1c40 | 802 | if (file) { |
cbc91f71 | 803 | uprobe_munmap(next, next->vm_start, next->vm_end); |
1da177e4 | 804 | fput(file); |
925d1c40 | 805 | } |
5beb4930 RR |
806 | if (next->anon_vma) |
807 | anon_vma_merge(vma, next); | |
1da177e4 | 808 | mm->map_count--; |
f0be3d32 | 809 | mpol_put(vma_policy(next)); |
1da177e4 LT |
810 | kmem_cache_free(vm_area_cachep, next); |
811 | /* | |
812 | * In mprotect's case 6 (see comments on vma_merge), | |
813 | * we must remove another next too. It would clutter | |
814 | * up the code too much to do both in one go. | |
815 | */ | |
d3737187 ML |
816 | next = vma->vm_next; |
817 | if (remove_next == 2) | |
1da177e4 | 818 | goto again; |
d3737187 ML |
819 | else if (next) |
820 | vma_gap_update(next); | |
821 | else | |
822 | mm->highest_vm_end = end; | |
1da177e4 | 823 | } |
2b144498 | 824 | if (insert && file) |
7b2d81d4 | 825 | uprobe_mmap(insert); |
1da177e4 LT |
826 | |
827 | validate_mm(mm); | |
5beb4930 RR |
828 | |
829 | return 0; | |
1da177e4 LT |
830 | } |
831 | ||
832 | /* | |
833 | * If the vma has a ->close operation then the driver probably needs to release | |
834 | * per-vma resources, so we don't attempt to merge those. | |
835 | */ | |
1da177e4 LT |
836 | static inline int is_mergeable_vma(struct vm_area_struct *vma, |
837 | struct file *file, unsigned long vm_flags) | |
838 | { | |
0b173bc4 | 839 | if (vma->vm_flags ^ vm_flags) |
1da177e4 LT |
840 | return 0; |
841 | if (vma->vm_file != file) | |
842 | return 0; | |
843 | if (vma->vm_ops && vma->vm_ops->close) | |
844 | return 0; | |
845 | return 1; | |
846 | } | |
847 | ||
848 | static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, | |
965f55de SL |
849 | struct anon_vma *anon_vma2, |
850 | struct vm_area_struct *vma) | |
1da177e4 | 851 | { |
965f55de SL |
852 | /* |
853 | * The list_is_singular() test is to avoid merging VMA cloned from | |
854 | * parents. This can improve scalability caused by anon_vma lock. | |
855 | */ | |
856 | if ((!anon_vma1 || !anon_vma2) && (!vma || | |
857 | list_is_singular(&vma->anon_vma_chain))) | |
858 | return 1; | |
859 | return anon_vma1 == anon_vma2; | |
1da177e4 LT |
860 | } |
861 | ||
862 | /* | |
863 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
864 | * in front of (at a lower virtual address and file offset than) the vma. | |
865 | * | |
866 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
867 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
868 | * | |
869 | * We don't check here for the merged mmap wrapping around the end of pagecache | |
870 | * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which | |
871 | * wrap, nor mmaps which cover the final page at index -1UL. | |
872 | */ | |
873 | static int | |
874 | can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, | |
875 | struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) | |
876 | { | |
877 | if (is_mergeable_vma(vma, file, vm_flags) && | |
965f55de | 878 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 LT |
879 | if (vma->vm_pgoff == vm_pgoff) |
880 | return 1; | |
881 | } | |
882 | return 0; | |
883 | } | |
884 | ||
885 | /* | |
886 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
887 | * beyond (at a higher virtual address and file offset than) the vma. | |
888 | * | |
889 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
890 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
891 | */ | |
892 | static int | |
893 | can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, | |
894 | struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) | |
895 | { | |
896 | if (is_mergeable_vma(vma, file, vm_flags) && | |
965f55de | 897 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 LT |
898 | pgoff_t vm_pglen; |
899 | vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; | |
900 | if (vma->vm_pgoff + vm_pglen == vm_pgoff) | |
901 | return 1; | |
902 | } | |
903 | return 0; | |
904 | } | |
905 | ||
906 | /* | |
907 | * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out | |
908 | * whether that can be merged with its predecessor or its successor. | |
909 | * Or both (it neatly fills a hole). | |
910 | * | |
911 | * In most cases - when called for mmap, brk or mremap - [addr,end) is | |
912 | * certain not to be mapped by the time vma_merge is called; but when | |
913 | * called for mprotect, it is certain to be already mapped (either at | |
914 | * an offset within prev, or at the start of next), and the flags of | |
915 | * this area are about to be changed to vm_flags - and the no-change | |
916 | * case has already been eliminated. | |
917 | * | |
918 | * The following mprotect cases have to be considered, where AAAA is | |
919 | * the area passed down from mprotect_fixup, never extending beyond one | |
920 | * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: | |
921 | * | |
922 | * AAAA AAAA AAAA AAAA | |
923 | * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX | |
924 | * cannot merge might become might become might become | |
925 | * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or | |
926 | * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or | |
927 | * mremap move: PPPPNNNNNNNN 8 | |
928 | * AAAA | |
929 | * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN | |
930 | * might become case 1 below case 2 below case 3 below | |
931 | * | |
932 | * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: | |
933 | * mprotect_fixup updates vm_flags & vm_page_prot on successful return. | |
934 | */ | |
935 | struct vm_area_struct *vma_merge(struct mm_struct *mm, | |
936 | struct vm_area_struct *prev, unsigned long addr, | |
937 | unsigned long end, unsigned long vm_flags, | |
938 | struct anon_vma *anon_vma, struct file *file, | |
939 | pgoff_t pgoff, struct mempolicy *policy) | |
940 | { | |
941 | pgoff_t pglen = (end - addr) >> PAGE_SHIFT; | |
942 | struct vm_area_struct *area, *next; | |
5beb4930 | 943 | int err; |
1da177e4 LT |
944 | |
945 | /* | |
946 | * We later require that vma->vm_flags == vm_flags, | |
947 | * so this tests vma->vm_flags & VM_SPECIAL, too. | |
948 | */ | |
949 | if (vm_flags & VM_SPECIAL) | |
950 | return NULL; | |
951 | ||
952 | if (prev) | |
953 | next = prev->vm_next; | |
954 | else | |
955 | next = mm->mmap; | |
956 | area = next; | |
957 | if (next && next->vm_end == end) /* cases 6, 7, 8 */ | |
958 | next = next->vm_next; | |
959 | ||
960 | /* | |
961 | * Can it merge with the predecessor? | |
962 | */ | |
963 | if (prev && prev->vm_end == addr && | |
964 | mpol_equal(vma_policy(prev), policy) && | |
965 | can_vma_merge_after(prev, vm_flags, | |
966 | anon_vma, file, pgoff)) { | |
967 | /* | |
968 | * OK, it can. Can we now merge in the successor as well? | |
969 | */ | |
970 | if (next && end == next->vm_start && | |
971 | mpol_equal(policy, vma_policy(next)) && | |
972 | can_vma_merge_before(next, vm_flags, | |
973 | anon_vma, file, pgoff+pglen) && | |
974 | is_mergeable_anon_vma(prev->anon_vma, | |
965f55de | 975 | next->anon_vma, NULL)) { |
1da177e4 | 976 | /* cases 1, 6 */ |
5beb4930 | 977 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 LT |
978 | next->vm_end, prev->vm_pgoff, NULL); |
979 | } else /* cases 2, 5, 7 */ | |
5beb4930 | 980 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 | 981 | end, prev->vm_pgoff, NULL); |
5beb4930 RR |
982 | if (err) |
983 | return NULL; | |
b15d00b6 | 984 | khugepaged_enter_vma_merge(prev); |
1da177e4 LT |
985 | return prev; |
986 | } | |
987 | ||
988 | /* | |
989 | * Can this new request be merged in front of next? | |
990 | */ | |
991 | if (next && end == next->vm_start && | |
992 | mpol_equal(policy, vma_policy(next)) && | |
993 | can_vma_merge_before(next, vm_flags, | |
994 | anon_vma, file, pgoff+pglen)) { | |
995 | if (prev && addr < prev->vm_end) /* case 4 */ | |
5beb4930 | 996 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 LT |
997 | addr, prev->vm_pgoff, NULL); |
998 | else /* cases 3, 8 */ | |
5beb4930 | 999 | err = vma_adjust(area, addr, next->vm_end, |
1da177e4 | 1000 | next->vm_pgoff - pglen, NULL); |
5beb4930 RR |
1001 | if (err) |
1002 | return NULL; | |
b15d00b6 | 1003 | khugepaged_enter_vma_merge(area); |
1da177e4 LT |
1004 | return area; |
1005 | } | |
1006 | ||
1007 | return NULL; | |
1008 | } | |
1009 | ||
d0e9fe17 LT |
1010 | /* |
1011 | * Rough compatbility check to quickly see if it's even worth looking | |
1012 | * at sharing an anon_vma. | |
1013 | * | |
1014 | * They need to have the same vm_file, and the flags can only differ | |
1015 | * in things that mprotect may change. | |
1016 | * | |
1017 | * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that | |
1018 | * we can merge the two vma's. For example, we refuse to merge a vma if | |
1019 | * there is a vm_ops->close() function, because that indicates that the | |
1020 | * driver is doing some kind of reference counting. But that doesn't | |
1021 | * really matter for the anon_vma sharing case. | |
1022 | */ | |
1023 | static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) | |
1024 | { | |
1025 | return a->vm_end == b->vm_start && | |
1026 | mpol_equal(vma_policy(a), vma_policy(b)) && | |
1027 | a->vm_file == b->vm_file && | |
1028 | !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) && | |
1029 | b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); | |
1030 | } | |
1031 | ||
1032 | /* | |
1033 | * Do some basic sanity checking to see if we can re-use the anon_vma | |
1034 | * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be | |
1035 | * the same as 'old', the other will be the new one that is trying | |
1036 | * to share the anon_vma. | |
1037 | * | |
1038 | * NOTE! This runs with mm_sem held for reading, so it is possible that | |
1039 | * the anon_vma of 'old' is concurrently in the process of being set up | |
1040 | * by another page fault trying to merge _that_. But that's ok: if it | |
1041 | * is being set up, that automatically means that it will be a singleton | |
1042 | * acceptable for merging, so we can do all of this optimistically. But | |
1043 | * we do that ACCESS_ONCE() to make sure that we never re-load the pointer. | |
1044 | * | |
1045 | * IOW: that the "list_is_singular()" test on the anon_vma_chain only | |
1046 | * matters for the 'stable anon_vma' case (ie the thing we want to avoid | |
1047 | * is to return an anon_vma that is "complex" due to having gone through | |
1048 | * a fork). | |
1049 | * | |
1050 | * We also make sure that the two vma's are compatible (adjacent, | |
1051 | * and with the same memory policies). That's all stable, even with just | |
1052 | * a read lock on the mm_sem. | |
1053 | */ | |
1054 | static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) | |
1055 | { | |
1056 | if (anon_vma_compatible(a, b)) { | |
1057 | struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma); | |
1058 | ||
1059 | if (anon_vma && list_is_singular(&old->anon_vma_chain)) | |
1060 | return anon_vma; | |
1061 | } | |
1062 | return NULL; | |
1063 | } | |
1064 | ||
1da177e4 LT |
1065 | /* |
1066 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check | |
1067 | * neighbouring vmas for a suitable anon_vma, before it goes off | |
1068 | * to allocate a new anon_vma. It checks because a repetitive | |
1069 | * sequence of mprotects and faults may otherwise lead to distinct | |
1070 | * anon_vmas being allocated, preventing vma merge in subsequent | |
1071 | * mprotect. | |
1072 | */ | |
1073 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) | |
1074 | { | |
d0e9fe17 | 1075 | struct anon_vma *anon_vma; |
1da177e4 | 1076 | struct vm_area_struct *near; |
1da177e4 LT |
1077 | |
1078 | near = vma->vm_next; | |
1079 | if (!near) | |
1080 | goto try_prev; | |
1081 | ||
d0e9fe17 LT |
1082 | anon_vma = reusable_anon_vma(near, vma, near); |
1083 | if (anon_vma) | |
1084 | return anon_vma; | |
1da177e4 | 1085 | try_prev: |
9be34c9d | 1086 | near = vma->vm_prev; |
1da177e4 LT |
1087 | if (!near) |
1088 | goto none; | |
1089 | ||
d0e9fe17 LT |
1090 | anon_vma = reusable_anon_vma(near, near, vma); |
1091 | if (anon_vma) | |
1092 | return anon_vma; | |
1da177e4 LT |
1093 | none: |
1094 | /* | |
1095 | * There's no absolute need to look only at touching neighbours: | |
1096 | * we could search further afield for "compatible" anon_vmas. | |
1097 | * But it would probably just be a waste of time searching, | |
1098 | * or lead to too many vmas hanging off the same anon_vma. | |
1099 | * We're trying to allow mprotect remerging later on, | |
1100 | * not trying to minimize memory used for anon_vmas. | |
1101 | */ | |
1102 | return NULL; | |
1103 | } | |
1104 | ||
1105 | #ifdef CONFIG_PROC_FS | |
ab50b8ed | 1106 | void vm_stat_account(struct mm_struct *mm, unsigned long flags, |
1da177e4 LT |
1107 | struct file *file, long pages) |
1108 | { | |
1109 | const unsigned long stack_flags | |
1110 | = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); | |
1111 | ||
44de9d0c HS |
1112 | mm->total_vm += pages; |
1113 | ||
1da177e4 LT |
1114 | if (file) { |
1115 | mm->shared_vm += pages; | |
1116 | if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) | |
1117 | mm->exec_vm += pages; | |
1118 | } else if (flags & stack_flags) | |
1119 | mm->stack_vm += pages; | |
1da177e4 LT |
1120 | } |
1121 | #endif /* CONFIG_PROC_FS */ | |
1122 | ||
40401530 AV |
1123 | /* |
1124 | * If a hint addr is less than mmap_min_addr change hint to be as | |
1125 | * low as possible but still greater than mmap_min_addr | |
1126 | */ | |
1127 | static inline unsigned long round_hint_to_min(unsigned long hint) | |
1128 | { | |
1129 | hint &= PAGE_MASK; | |
1130 | if (((void *)hint != NULL) && | |
1131 | (hint < mmap_min_addr)) | |
1132 | return PAGE_ALIGN(mmap_min_addr); | |
1133 | return hint; | |
1134 | } | |
1135 | ||
1da177e4 | 1136 | /* |
27f5de79 | 1137 | * The caller must hold down_write(¤t->mm->mmap_sem). |
1da177e4 LT |
1138 | */ |
1139 | ||
e3fc629d | 1140 | unsigned long do_mmap_pgoff(struct file *file, unsigned long addr, |
1da177e4 LT |
1141 | unsigned long len, unsigned long prot, |
1142 | unsigned long flags, unsigned long pgoff) | |
1143 | { | |
1144 | struct mm_struct * mm = current->mm; | |
1da177e4 | 1145 | struct inode *inode; |
ca16d140 | 1146 | vm_flags_t vm_flags; |
1da177e4 | 1147 | |
1da177e4 LT |
1148 | /* |
1149 | * Does the application expect PROT_READ to imply PROT_EXEC? | |
1150 | * | |
1151 | * (the exception is when the underlying filesystem is noexec | |
1152 | * mounted, in which case we dont add PROT_EXEC.) | |
1153 | */ | |
1154 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
d3ac7f89 | 1155 | if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC))) |
1da177e4 LT |
1156 | prot |= PROT_EXEC; |
1157 | ||
1158 | if (!len) | |
1159 | return -EINVAL; | |
1160 | ||
7cd94146 EP |
1161 | if (!(flags & MAP_FIXED)) |
1162 | addr = round_hint_to_min(addr); | |
1163 | ||
1da177e4 LT |
1164 | /* Careful about overflows.. */ |
1165 | len = PAGE_ALIGN(len); | |
9206de95 | 1166 | if (!len) |
1da177e4 LT |
1167 | return -ENOMEM; |
1168 | ||
1169 | /* offset overflow? */ | |
1170 | if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) | |
1171 | return -EOVERFLOW; | |
1172 | ||
1173 | /* Too many mappings? */ | |
1174 | if (mm->map_count > sysctl_max_map_count) | |
1175 | return -ENOMEM; | |
1176 | ||
1177 | /* Obtain the address to map to. we verify (or select) it and ensure | |
1178 | * that it represents a valid section of the address space. | |
1179 | */ | |
1180 | addr = get_unmapped_area(file, addr, len, pgoff, flags); | |
1181 | if (addr & ~PAGE_MASK) | |
1182 | return addr; | |
1183 | ||
1184 | /* Do simple checking here so the lower-level routines won't have | |
1185 | * to. we assume access permissions have been handled by the open | |
1186 | * of the memory object, so we don't do any here. | |
1187 | */ | |
1188 | vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | | |
1189 | mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; | |
1190 | ||
cdf7b341 | 1191 | if (flags & MAP_LOCKED) |
1da177e4 LT |
1192 | if (!can_do_mlock()) |
1193 | return -EPERM; | |
ba470de4 | 1194 | |
1da177e4 LT |
1195 | /* mlock MCL_FUTURE? */ |
1196 | if (vm_flags & VM_LOCKED) { | |
1197 | unsigned long locked, lock_limit; | |
93ea1d0a CW |
1198 | locked = len >> PAGE_SHIFT; |
1199 | locked += mm->locked_vm; | |
59e99e5b | 1200 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
93ea1d0a | 1201 | lock_limit >>= PAGE_SHIFT; |
1da177e4 LT |
1202 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) |
1203 | return -EAGAIN; | |
1204 | } | |
1205 | ||
d3ac7f89 | 1206 | inode = file ? file->f_path.dentry->d_inode : NULL; |
1da177e4 LT |
1207 | |
1208 | if (file) { | |
1209 | switch (flags & MAP_TYPE) { | |
1210 | case MAP_SHARED: | |
1211 | if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) | |
1212 | return -EACCES; | |
1213 | ||
1214 | /* | |
1215 | * Make sure we don't allow writing to an append-only | |
1216 | * file.. | |
1217 | */ | |
1218 | if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) | |
1219 | return -EACCES; | |
1220 | ||
1221 | /* | |
1222 | * Make sure there are no mandatory locks on the file. | |
1223 | */ | |
1224 | if (locks_verify_locked(inode)) | |
1225 | return -EAGAIN; | |
1226 | ||
1227 | vm_flags |= VM_SHARED | VM_MAYSHARE; | |
1228 | if (!(file->f_mode & FMODE_WRITE)) | |
1229 | vm_flags &= ~(VM_MAYWRITE | VM_SHARED); | |
1230 | ||
1231 | /* fall through */ | |
1232 | case MAP_PRIVATE: | |
1233 | if (!(file->f_mode & FMODE_READ)) | |
1234 | return -EACCES; | |
d3ac7f89 | 1235 | if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { |
80c5606c LT |
1236 | if (vm_flags & VM_EXEC) |
1237 | return -EPERM; | |
1238 | vm_flags &= ~VM_MAYEXEC; | |
1239 | } | |
80c5606c LT |
1240 | |
1241 | if (!file->f_op || !file->f_op->mmap) | |
1242 | return -ENODEV; | |
1da177e4 LT |
1243 | break; |
1244 | ||
1245 | default: | |
1246 | return -EINVAL; | |
1247 | } | |
1248 | } else { | |
1249 | switch (flags & MAP_TYPE) { | |
1250 | case MAP_SHARED: | |
ce363942 TH |
1251 | /* |
1252 | * Ignore pgoff. | |
1253 | */ | |
1254 | pgoff = 0; | |
1da177e4 LT |
1255 | vm_flags |= VM_SHARED | VM_MAYSHARE; |
1256 | break; | |
1257 | case MAP_PRIVATE: | |
1258 | /* | |
1259 | * Set pgoff according to addr for anon_vma. | |
1260 | */ | |
1261 | pgoff = addr >> PAGE_SHIFT; | |
1262 | break; | |
1263 | default: | |
1264 | return -EINVAL; | |
1265 | } | |
1266 | } | |
1267 | ||
5a6fe125 | 1268 | return mmap_region(file, addr, len, flags, vm_flags, pgoff); |
0165ab44 | 1269 | } |
6be5ceb0 | 1270 | |
66f0dc48 HD |
1271 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1272 | unsigned long, prot, unsigned long, flags, | |
1273 | unsigned long, fd, unsigned long, pgoff) | |
1274 | { | |
1275 | struct file *file = NULL; | |
1276 | unsigned long retval = -EBADF; | |
1277 | ||
1278 | if (!(flags & MAP_ANONYMOUS)) { | |
120a795d | 1279 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1280 | if (unlikely(flags & MAP_HUGETLB)) |
1281 | return -EINVAL; | |
1282 | file = fget(fd); | |
1283 | if (!file) | |
1284 | goto out; | |
1285 | } else if (flags & MAP_HUGETLB) { | |
1286 | struct user_struct *user = NULL; | |
1287 | /* | |
1288 | * VM_NORESERVE is used because the reservations will be | |
1289 | * taken when vm_ops->mmap() is called | |
1290 | * A dummy user value is used because we are not locking | |
1291 | * memory so no accounting is necessary | |
1292 | */ | |
40716e29 | 1293 | file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len, |
42d7395f AK |
1294 | VM_NORESERVE, |
1295 | &user, HUGETLB_ANONHUGE_INODE, | |
1296 | (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); | |
66f0dc48 HD |
1297 | if (IS_ERR(file)) |
1298 | return PTR_ERR(file); | |
1299 | } | |
1300 | ||
1301 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); | |
1302 | ||
eb36c587 | 1303 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
66f0dc48 HD |
1304 | if (file) |
1305 | fput(file); | |
1306 | out: | |
1307 | return retval; | |
1308 | } | |
1309 | ||
a4679373 CH |
1310 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1311 | struct mmap_arg_struct { | |
1312 | unsigned long addr; | |
1313 | unsigned long len; | |
1314 | unsigned long prot; | |
1315 | unsigned long flags; | |
1316 | unsigned long fd; | |
1317 | unsigned long offset; | |
1318 | }; | |
1319 | ||
1320 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1321 | { | |
1322 | struct mmap_arg_struct a; | |
1323 | ||
1324 | if (copy_from_user(&a, arg, sizeof(a))) | |
1325 | return -EFAULT; | |
1326 | if (a.offset & ~PAGE_MASK) | |
1327 | return -EINVAL; | |
1328 | ||
1329 | return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, | |
1330 | a.offset >> PAGE_SHIFT); | |
1331 | } | |
1332 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1333 | ||
4e950f6f AD |
1334 | /* |
1335 | * Some shared mappigns will want the pages marked read-only | |
1336 | * to track write events. If so, we'll downgrade vm_page_prot | |
1337 | * to the private version (using protection_map[] without the | |
1338 | * VM_SHARED bit). | |
1339 | */ | |
1340 | int vma_wants_writenotify(struct vm_area_struct *vma) | |
1341 | { | |
ca16d140 | 1342 | vm_flags_t vm_flags = vma->vm_flags; |
4e950f6f AD |
1343 | |
1344 | /* If it was private or non-writable, the write bit is already clear */ | |
1345 | if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) | |
1346 | return 0; | |
1347 | ||
1348 | /* The backer wishes to know when pages are first written to? */ | |
1349 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) | |
1350 | return 1; | |
1351 | ||
1352 | /* The open routine did something to the protections already? */ | |
1353 | if (pgprot_val(vma->vm_page_prot) != | |
3ed75eb8 | 1354 | pgprot_val(vm_get_page_prot(vm_flags))) |
4e950f6f AD |
1355 | return 0; |
1356 | ||
1357 | /* Specialty mapping? */ | |
4b6e1e37 | 1358 | if (vm_flags & VM_PFNMAP) |
4e950f6f AD |
1359 | return 0; |
1360 | ||
1361 | /* Can the mapping track the dirty pages? */ | |
1362 | return vma->vm_file && vma->vm_file->f_mapping && | |
1363 | mapping_cap_account_dirty(vma->vm_file->f_mapping); | |
1364 | } | |
1365 | ||
fc8744ad LT |
1366 | /* |
1367 | * We account for memory if it's a private writeable mapping, | |
5a6fe125 | 1368 | * not hugepages and VM_NORESERVE wasn't set. |
fc8744ad | 1369 | */ |
ca16d140 | 1370 | static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) |
fc8744ad | 1371 | { |
5a6fe125 MG |
1372 | /* |
1373 | * hugetlb has its own accounting separate from the core VM | |
1374 | * VM_HUGETLB may not be set yet so we cannot check for that flag. | |
1375 | */ | |
1376 | if (file && is_file_hugepages(file)) | |
1377 | return 0; | |
1378 | ||
fc8744ad LT |
1379 | return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; |
1380 | } | |
1381 | ||
0165ab44 MS |
1382 | unsigned long mmap_region(struct file *file, unsigned long addr, |
1383 | unsigned long len, unsigned long flags, | |
ca16d140 | 1384 | vm_flags_t vm_flags, unsigned long pgoff) |
0165ab44 MS |
1385 | { |
1386 | struct mm_struct *mm = current->mm; | |
1387 | struct vm_area_struct *vma, *prev; | |
1388 | int correct_wcount = 0; | |
1389 | int error; | |
1390 | struct rb_node **rb_link, *rb_parent; | |
1391 | unsigned long charged = 0; | |
1392 | struct inode *inode = file ? file->f_path.dentry->d_inode : NULL; | |
1393 | ||
1da177e4 LT |
1394 | /* Clear old maps */ |
1395 | error = -ENOMEM; | |
1396 | munmap_back: | |
6597d783 | 1397 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { |
1da177e4 LT |
1398 | if (do_munmap(mm, addr, len)) |
1399 | return -ENOMEM; | |
1400 | goto munmap_back; | |
1401 | } | |
1402 | ||
1403 | /* Check against address space limit. */ | |
119f657c | 1404 | if (!may_expand_vm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
1405 | return -ENOMEM; |
1406 | ||
fc8744ad LT |
1407 | /* |
1408 | * Set 'VM_NORESERVE' if we should not account for the | |
5a6fe125 | 1409 | * memory use of this mapping. |
fc8744ad | 1410 | */ |
5a6fe125 MG |
1411 | if ((flags & MAP_NORESERVE)) { |
1412 | /* We honor MAP_NORESERVE if allowed to overcommit */ | |
1413 | if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) | |
1414 | vm_flags |= VM_NORESERVE; | |
1415 | ||
1416 | /* hugetlb applies strict overcommit unless MAP_NORESERVE */ | |
1417 | if (file && is_file_hugepages(file)) | |
1418 | vm_flags |= VM_NORESERVE; | |
1419 | } | |
cdfd4325 | 1420 | |
fc8744ad LT |
1421 | /* |
1422 | * Private writable mapping: check memory availability | |
1423 | */ | |
5a6fe125 | 1424 | if (accountable_mapping(file, vm_flags)) { |
fc8744ad | 1425 | charged = len >> PAGE_SHIFT; |
191c5424 | 1426 | if (security_vm_enough_memory_mm(mm, charged)) |
fc8744ad LT |
1427 | return -ENOMEM; |
1428 | vm_flags |= VM_ACCOUNT; | |
1da177e4 LT |
1429 | } |
1430 | ||
1431 | /* | |
de33c8db | 1432 | * Can we just expand an old mapping? |
1da177e4 | 1433 | */ |
de33c8db LT |
1434 | vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL); |
1435 | if (vma) | |
1436 | goto out; | |
1da177e4 LT |
1437 | |
1438 | /* | |
1439 | * Determine the object being mapped and call the appropriate | |
1440 | * specific mapper. the address has already been validated, but | |
1441 | * not unmapped, but the maps are removed from the list. | |
1442 | */ | |
c5e3b83e | 1443 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
1444 | if (!vma) { |
1445 | error = -ENOMEM; | |
1446 | goto unacct_error; | |
1447 | } | |
1da177e4 LT |
1448 | |
1449 | vma->vm_mm = mm; | |
1450 | vma->vm_start = addr; | |
1451 | vma->vm_end = addr + len; | |
1452 | vma->vm_flags = vm_flags; | |
3ed75eb8 | 1453 | vma->vm_page_prot = vm_get_page_prot(vm_flags); |
1da177e4 | 1454 | vma->vm_pgoff = pgoff; |
5beb4930 | 1455 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 | 1456 | |
ce8fea7a HD |
1457 | error = -EINVAL; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */ |
1458 | ||
1da177e4 | 1459 | if (file) { |
1da177e4 LT |
1460 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1461 | goto free_vma; | |
1462 | if (vm_flags & VM_DENYWRITE) { | |
1463 | error = deny_write_access(file); | |
1464 | if (error) | |
1465 | goto free_vma; | |
1466 | correct_wcount = 1; | |
1467 | } | |
cb0942b8 | 1468 | vma->vm_file = get_file(file); |
1da177e4 LT |
1469 | error = file->f_op->mmap(file, vma); |
1470 | if (error) | |
1471 | goto unmap_and_free_vma; | |
f8dbf0a7 HS |
1472 | |
1473 | /* Can addr have changed?? | |
1474 | * | |
1475 | * Answer: Yes, several device drivers can do it in their | |
1476 | * f_op->mmap method. -DaveM | |
1477 | */ | |
1478 | addr = vma->vm_start; | |
1479 | pgoff = vma->vm_pgoff; | |
1480 | vm_flags = vma->vm_flags; | |
1da177e4 | 1481 | } else if (vm_flags & VM_SHARED) { |
835ee797 AV |
1482 | if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP))) |
1483 | goto free_vma; | |
1da177e4 LT |
1484 | error = shmem_zero_setup(vma); |
1485 | if (error) | |
1486 | goto free_vma; | |
1487 | } | |
1488 | ||
c9d0bf24 MD |
1489 | if (vma_wants_writenotify(vma)) { |
1490 | pgprot_t pprot = vma->vm_page_prot; | |
1491 | ||
1492 | /* Can vma->vm_page_prot have changed?? | |
1493 | * | |
1494 | * Answer: Yes, drivers may have changed it in their | |
1495 | * f_op->mmap method. | |
1496 | * | |
1497 | * Ensures that vmas marked as uncached stay that way. | |
1498 | */ | |
1ddd439e | 1499 | vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED); |
c9d0bf24 MD |
1500 | if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot))) |
1501 | vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); | |
1502 | } | |
d08b3851 | 1503 | |
de33c8db LT |
1504 | vma_link(mm, vma, prev, rb_link, rb_parent); |
1505 | file = vma->vm_file; | |
4d3d5b41 ON |
1506 | |
1507 | /* Once vma denies write, undo our temporary denial count */ | |
1508 | if (correct_wcount) | |
1509 | atomic_inc(&inode->i_writecount); | |
1510 | out: | |
cdd6c482 | 1511 | perf_event_mmap(vma); |
0a4a9391 | 1512 | |
ab50b8ed | 1513 | vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); |
1da177e4 | 1514 | if (vm_flags & VM_LOCKED) { |
06f9d8c2 KM |
1515 | if (!mlock_vma_pages_range(vma, addr, addr + len)) |
1516 | mm->locked_vm += (len >> PAGE_SHIFT); | |
ba470de4 | 1517 | } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK)) |
54cb8821 | 1518 | make_pages_present(addr, addr + len); |
2b144498 | 1519 | |
c7a3a88c ON |
1520 | if (file) |
1521 | uprobe_mmap(vma); | |
2b144498 | 1522 | |
1da177e4 LT |
1523 | return addr; |
1524 | ||
1525 | unmap_and_free_vma: | |
1526 | if (correct_wcount) | |
1527 | atomic_inc(&inode->i_writecount); | |
1528 | vma->vm_file = NULL; | |
1529 | fput(file); | |
1530 | ||
1531 | /* Undo any partial mapping done by a device driver. */ | |
e0da382c HD |
1532 | unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); |
1533 | charged = 0; | |
1da177e4 LT |
1534 | free_vma: |
1535 | kmem_cache_free(vm_area_cachep, vma); | |
1536 | unacct_error: | |
1537 | if (charged) | |
1538 | vm_unacct_memory(charged); | |
1539 | return error; | |
1540 | } | |
1541 | ||
1da177e4 LT |
1542 | /* Get an address range which is currently unmapped. |
1543 | * For shmat() with addr=0. | |
1544 | * | |
1545 | * Ugly calling convention alert: | |
1546 | * Return value with the low bits set means error value, | |
1547 | * ie | |
1548 | * if (ret & ~PAGE_MASK) | |
1549 | * error = ret; | |
1550 | * | |
1551 | * This function "knows" that -ENOMEM has the bits set. | |
1552 | */ | |
1553 | #ifndef HAVE_ARCH_UNMAPPED_AREA | |
1554 | unsigned long | |
1555 | arch_get_unmapped_area(struct file *filp, unsigned long addr, | |
1556 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1557 | { | |
1558 | struct mm_struct *mm = current->mm; | |
1559 | struct vm_area_struct *vma; | |
1560 | unsigned long start_addr; | |
1561 | ||
1562 | if (len > TASK_SIZE) | |
1563 | return -ENOMEM; | |
1564 | ||
06abdfb4 BH |
1565 | if (flags & MAP_FIXED) |
1566 | return addr; | |
1567 | ||
1da177e4 LT |
1568 | if (addr) { |
1569 | addr = PAGE_ALIGN(addr); | |
1570 | vma = find_vma(mm, addr); | |
1571 | if (TASK_SIZE - len >= addr && | |
1572 | (!vma || addr + len <= vma->vm_start)) | |
1573 | return addr; | |
1574 | } | |
1363c3cd WW |
1575 | if (len > mm->cached_hole_size) { |
1576 | start_addr = addr = mm->free_area_cache; | |
1577 | } else { | |
1578 | start_addr = addr = TASK_UNMAPPED_BASE; | |
1579 | mm->cached_hole_size = 0; | |
1580 | } | |
1da177e4 LT |
1581 | |
1582 | full_search: | |
1583 | for (vma = find_vma(mm, addr); ; vma = vma->vm_next) { | |
1584 | /* At this point: (!vma || addr < vma->vm_end). */ | |
1585 | if (TASK_SIZE - len < addr) { | |
1586 | /* | |
1587 | * Start a new search - just in case we missed | |
1588 | * some holes. | |
1589 | */ | |
1590 | if (start_addr != TASK_UNMAPPED_BASE) { | |
1363c3cd WW |
1591 | addr = TASK_UNMAPPED_BASE; |
1592 | start_addr = addr; | |
1593 | mm->cached_hole_size = 0; | |
1da177e4 LT |
1594 | goto full_search; |
1595 | } | |
1596 | return -ENOMEM; | |
1597 | } | |
1598 | if (!vma || addr + len <= vma->vm_start) { | |
1599 | /* | |
1600 | * Remember the place where we stopped the search: | |
1601 | */ | |
1602 | mm->free_area_cache = addr + len; | |
1603 | return addr; | |
1604 | } | |
1363c3cd WW |
1605 | if (addr + mm->cached_hole_size < vma->vm_start) |
1606 | mm->cached_hole_size = vma->vm_start - addr; | |
1da177e4 LT |
1607 | addr = vma->vm_end; |
1608 | } | |
1609 | } | |
1610 | #endif | |
1611 | ||
1363c3cd | 1612 | void arch_unmap_area(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
1613 | { |
1614 | /* | |
1615 | * Is this a new hole at the lowest possible address? | |
1616 | */ | |
f44d2198 | 1617 | if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) |
1363c3cd | 1618 | mm->free_area_cache = addr; |
1da177e4 LT |
1619 | } |
1620 | ||
1621 | /* | |
1622 | * This mmap-allocator allocates new areas top-down from below the | |
1623 | * stack's low limit (the base): | |
1624 | */ | |
1625 | #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN | |
1626 | unsigned long | |
1627 | arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, | |
1628 | const unsigned long len, const unsigned long pgoff, | |
1629 | const unsigned long flags) | |
1630 | { | |
1631 | struct vm_area_struct *vma; | |
1632 | struct mm_struct *mm = current->mm; | |
b716ad95 | 1633 | unsigned long addr = addr0, start_addr; |
1da177e4 LT |
1634 | |
1635 | /* requested length too big for entire address space */ | |
1636 | if (len > TASK_SIZE) | |
1637 | return -ENOMEM; | |
1638 | ||
06abdfb4 BH |
1639 | if (flags & MAP_FIXED) |
1640 | return addr; | |
1641 | ||
1da177e4 LT |
1642 | /* requesting a specific address */ |
1643 | if (addr) { | |
1644 | addr = PAGE_ALIGN(addr); | |
1645 | vma = find_vma(mm, addr); | |
1646 | if (TASK_SIZE - len >= addr && | |
1647 | (!vma || addr + len <= vma->vm_start)) | |
1648 | return addr; | |
1649 | } | |
1650 | ||
1363c3cd WW |
1651 | /* check if free_area_cache is useful for us */ |
1652 | if (len <= mm->cached_hole_size) { | |
1653 | mm->cached_hole_size = 0; | |
1654 | mm->free_area_cache = mm->mmap_base; | |
1655 | } | |
1656 | ||
b716ad95 | 1657 | try_again: |
1da177e4 | 1658 | /* either no address requested or can't fit in requested address hole */ |
b716ad95 | 1659 | start_addr = addr = mm->free_area_cache; |
73219d17 | 1660 | |
b716ad95 XG |
1661 | if (addr < len) |
1662 | goto fail; | |
1da177e4 | 1663 | |
b716ad95 | 1664 | addr -= len; |
1da177e4 LT |
1665 | do { |
1666 | /* | |
1667 | * Lookup failure means no vma is above this address, | |
1668 | * else if new region fits below vma->vm_start, | |
1669 | * return with success: | |
1670 | */ | |
1671 | vma = find_vma(mm, addr); | |
1672 | if (!vma || addr+len <= vma->vm_start) | |
1673 | /* remember the address as a hint for next time */ | |
1674 | return (mm->free_area_cache = addr); | |
1675 | ||
1363c3cd WW |
1676 | /* remember the largest hole we saw so far */ |
1677 | if (addr + mm->cached_hole_size < vma->vm_start) | |
1678 | mm->cached_hole_size = vma->vm_start - addr; | |
1679 | ||
1da177e4 LT |
1680 | /* try just below the current vma->vm_start */ |
1681 | addr = vma->vm_start-len; | |
49a43876 | 1682 | } while (len < vma->vm_start); |
1da177e4 | 1683 | |
b716ad95 XG |
1684 | fail: |
1685 | /* | |
1686 | * if hint left us with no space for the requested | |
1687 | * mapping then try again: | |
1688 | * | |
1689 | * Note: this is different with the case of bottomup | |
1690 | * which does the fully line-search, but we use find_vma | |
1691 | * here that causes some holes skipped. | |
1692 | */ | |
1693 | if (start_addr != mm->mmap_base) { | |
1694 | mm->free_area_cache = mm->mmap_base; | |
1695 | mm->cached_hole_size = 0; | |
1696 | goto try_again; | |
1697 | } | |
1698 | ||
1da177e4 LT |
1699 | /* |
1700 | * A failed mmap() very likely causes application failure, | |
1701 | * so fall back to the bottom-up function here. This scenario | |
1702 | * can happen with large stack limits and large mmap() | |
1703 | * allocations. | |
1704 | */ | |
1363c3cd WW |
1705 | mm->cached_hole_size = ~0UL; |
1706 | mm->free_area_cache = TASK_UNMAPPED_BASE; | |
1da177e4 LT |
1707 | addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags); |
1708 | /* | |
1709 | * Restore the topdown base: | |
1710 | */ | |
1711 | mm->free_area_cache = mm->mmap_base; | |
1363c3cd | 1712 | mm->cached_hole_size = ~0UL; |
1da177e4 LT |
1713 | |
1714 | return addr; | |
1715 | } | |
1716 | #endif | |
1717 | ||
1363c3cd | 1718 | void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
1719 | { |
1720 | /* | |
1721 | * Is this a new hole at the highest possible address? | |
1722 | */ | |
1363c3cd WW |
1723 | if (addr > mm->free_area_cache) |
1724 | mm->free_area_cache = addr; | |
1da177e4 LT |
1725 | |
1726 | /* dont allow allocations above current base */ | |
1363c3cd WW |
1727 | if (mm->free_area_cache > mm->mmap_base) |
1728 | mm->free_area_cache = mm->mmap_base; | |
1da177e4 LT |
1729 | } |
1730 | ||
1731 | unsigned long | |
1732 | get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, | |
1733 | unsigned long pgoff, unsigned long flags) | |
1734 | { | |
06abdfb4 BH |
1735 | unsigned long (*get_area)(struct file *, unsigned long, |
1736 | unsigned long, unsigned long, unsigned long); | |
1737 | ||
9206de95 AV |
1738 | unsigned long error = arch_mmap_check(addr, len, flags); |
1739 | if (error) | |
1740 | return error; | |
1741 | ||
1742 | /* Careful about overflows.. */ | |
1743 | if (len > TASK_SIZE) | |
1744 | return -ENOMEM; | |
1745 | ||
06abdfb4 BH |
1746 | get_area = current->mm->get_unmapped_area; |
1747 | if (file && file->f_op && file->f_op->get_unmapped_area) | |
1748 | get_area = file->f_op->get_unmapped_area; | |
1749 | addr = get_area(file, addr, len, pgoff, flags); | |
1750 | if (IS_ERR_VALUE(addr)) | |
1751 | return addr; | |
1da177e4 | 1752 | |
07ab67c8 LT |
1753 | if (addr > TASK_SIZE - len) |
1754 | return -ENOMEM; | |
1755 | if (addr & ~PAGE_MASK) | |
1756 | return -EINVAL; | |
06abdfb4 | 1757 | |
9ac4ed4b AV |
1758 | addr = arch_rebalance_pgtables(addr, len); |
1759 | error = security_mmap_addr(addr); | |
1760 | return error ? error : addr; | |
1da177e4 LT |
1761 | } |
1762 | ||
1763 | EXPORT_SYMBOL(get_unmapped_area); | |
1764 | ||
1765 | /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ | |
48aae425 | 1766 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
1767 | { |
1768 | struct vm_area_struct *vma = NULL; | |
1769 | ||
841e31e5 RM |
1770 | if (WARN_ON_ONCE(!mm)) /* Remove this in linux-3.6 */ |
1771 | return NULL; | |
1772 | ||
1773 | /* Check the cache first. */ | |
1774 | /* (Cache hit rate is typically around 35%.) */ | |
1775 | vma = mm->mmap_cache; | |
1776 | if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) { | |
1777 | struct rb_node *rb_node; | |
1778 | ||
1779 | rb_node = mm->mm_rb.rb_node; | |
1780 | vma = NULL; | |
1781 | ||
1782 | while (rb_node) { | |
1783 | struct vm_area_struct *vma_tmp; | |
1784 | ||
1785 | vma_tmp = rb_entry(rb_node, | |
1786 | struct vm_area_struct, vm_rb); | |
1787 | ||
1788 | if (vma_tmp->vm_end > addr) { | |
1789 | vma = vma_tmp; | |
1790 | if (vma_tmp->vm_start <= addr) | |
1791 | break; | |
1792 | rb_node = rb_node->rb_left; | |
1793 | } else | |
1794 | rb_node = rb_node->rb_right; | |
1da177e4 | 1795 | } |
841e31e5 RM |
1796 | if (vma) |
1797 | mm->mmap_cache = vma; | |
1da177e4 LT |
1798 | } |
1799 | return vma; | |
1800 | } | |
1801 | ||
1802 | EXPORT_SYMBOL(find_vma); | |
1803 | ||
6bd4837d KM |
1804 | /* |
1805 | * Same as find_vma, but also return a pointer to the previous VMA in *pprev. | |
6bd4837d | 1806 | */ |
1da177e4 LT |
1807 | struct vm_area_struct * |
1808 | find_vma_prev(struct mm_struct *mm, unsigned long addr, | |
1809 | struct vm_area_struct **pprev) | |
1810 | { | |
6bd4837d | 1811 | struct vm_area_struct *vma; |
1da177e4 | 1812 | |
6bd4837d | 1813 | vma = find_vma(mm, addr); |
83cd904d MP |
1814 | if (vma) { |
1815 | *pprev = vma->vm_prev; | |
1816 | } else { | |
1817 | struct rb_node *rb_node = mm->mm_rb.rb_node; | |
1818 | *pprev = NULL; | |
1819 | while (rb_node) { | |
1820 | *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
1821 | rb_node = rb_node->rb_right; | |
1822 | } | |
1823 | } | |
6bd4837d | 1824 | return vma; |
1da177e4 LT |
1825 | } |
1826 | ||
1827 | /* | |
1828 | * Verify that the stack growth is acceptable and | |
1829 | * update accounting. This is shared with both the | |
1830 | * grow-up and grow-down cases. | |
1831 | */ | |
48aae425 | 1832 | static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) |
1da177e4 LT |
1833 | { |
1834 | struct mm_struct *mm = vma->vm_mm; | |
1835 | struct rlimit *rlim = current->signal->rlim; | |
0d59a01b | 1836 | unsigned long new_start; |
1da177e4 LT |
1837 | |
1838 | /* address space limit tests */ | |
119f657c | 1839 | if (!may_expand_vm(mm, grow)) |
1da177e4 LT |
1840 | return -ENOMEM; |
1841 | ||
1842 | /* Stack limit test */ | |
59e99e5b | 1843 | if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur)) |
1da177e4 LT |
1844 | return -ENOMEM; |
1845 | ||
1846 | /* mlock limit tests */ | |
1847 | if (vma->vm_flags & VM_LOCKED) { | |
1848 | unsigned long locked; | |
1849 | unsigned long limit; | |
1850 | locked = mm->locked_vm + grow; | |
59e99e5b JS |
1851 | limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); |
1852 | limit >>= PAGE_SHIFT; | |
1da177e4 LT |
1853 | if (locked > limit && !capable(CAP_IPC_LOCK)) |
1854 | return -ENOMEM; | |
1855 | } | |
1856 | ||
0d59a01b AL |
1857 | /* Check to ensure the stack will not grow into a hugetlb-only region */ |
1858 | new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : | |
1859 | vma->vm_end - size; | |
1860 | if (is_hugepage_only_range(vma->vm_mm, new_start, size)) | |
1861 | return -EFAULT; | |
1862 | ||
1da177e4 LT |
1863 | /* |
1864 | * Overcommit.. This must be the final test, as it will | |
1865 | * update security statistics. | |
1866 | */ | |
05fa199d | 1867 | if (security_vm_enough_memory_mm(mm, grow)) |
1da177e4 LT |
1868 | return -ENOMEM; |
1869 | ||
1870 | /* Ok, everything looks good - let it rip */ | |
1da177e4 LT |
1871 | if (vma->vm_flags & VM_LOCKED) |
1872 | mm->locked_vm += grow; | |
ab50b8ed | 1873 | vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); |
1da177e4 LT |
1874 | return 0; |
1875 | } | |
1876 | ||
46dea3d0 | 1877 | #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) |
1da177e4 | 1878 | /* |
46dea3d0 HD |
1879 | * PA-RISC uses this for its stack; IA64 for its Register Backing Store. |
1880 | * vma is the last one with address > vma->vm_end. Have to extend vma. | |
1da177e4 | 1881 | */ |
46dea3d0 | 1882 | int expand_upwards(struct vm_area_struct *vma, unsigned long address) |
1da177e4 LT |
1883 | { |
1884 | int error; | |
1885 | ||
1886 | if (!(vma->vm_flags & VM_GROWSUP)) | |
1887 | return -EFAULT; | |
1888 | ||
1889 | /* | |
1890 | * We must make sure the anon_vma is allocated | |
1891 | * so that the anon_vma locking is not a noop. | |
1892 | */ | |
1893 | if (unlikely(anon_vma_prepare(vma))) | |
1894 | return -ENOMEM; | |
bb4a340e | 1895 | vma_lock_anon_vma(vma); |
1da177e4 LT |
1896 | |
1897 | /* | |
1898 | * vma->vm_start/vm_end cannot change under us because the caller | |
1899 | * is required to hold the mmap_sem in read mode. We need the | |
1900 | * anon_vma lock to serialize against concurrent expand_stacks. | |
06b32f3a | 1901 | * Also guard against wrapping around to address 0. |
1da177e4 | 1902 | */ |
06b32f3a HD |
1903 | if (address < PAGE_ALIGN(address+4)) |
1904 | address = PAGE_ALIGN(address+4); | |
1905 | else { | |
bb4a340e | 1906 | vma_unlock_anon_vma(vma); |
06b32f3a HD |
1907 | return -ENOMEM; |
1908 | } | |
1da177e4 LT |
1909 | error = 0; |
1910 | ||
1911 | /* Somebody else might have raced and expanded it already */ | |
1912 | if (address > vma->vm_end) { | |
1913 | unsigned long size, grow; | |
1914 | ||
1915 | size = address - vma->vm_start; | |
1916 | grow = (address - vma->vm_end) >> PAGE_SHIFT; | |
1917 | ||
42c36f63 HD |
1918 | error = -ENOMEM; |
1919 | if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { | |
1920 | error = acct_stack_growth(vma, size, grow); | |
1921 | if (!error) { | |
bf181b9f | 1922 | anon_vma_interval_tree_pre_update_vma(vma); |
42c36f63 | 1923 | vma->vm_end = address; |
bf181b9f | 1924 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 ML |
1925 | if (vma->vm_next) |
1926 | vma_gap_update(vma->vm_next); | |
1927 | else | |
1928 | vma->vm_mm->highest_vm_end = address; | |
42c36f63 HD |
1929 | perf_event_mmap(vma); |
1930 | } | |
3af9e859 | 1931 | } |
1da177e4 | 1932 | } |
bb4a340e | 1933 | vma_unlock_anon_vma(vma); |
b15d00b6 | 1934 | khugepaged_enter_vma_merge(vma); |
ed8ea815 | 1935 | validate_mm(vma->vm_mm); |
1da177e4 LT |
1936 | return error; |
1937 | } | |
46dea3d0 HD |
1938 | #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ |
1939 | ||
1da177e4 LT |
1940 | /* |
1941 | * vma is the first one with address < vma->vm_start. Have to extend vma. | |
1942 | */ | |
d05f3169 | 1943 | int expand_downwards(struct vm_area_struct *vma, |
b6a2fea3 | 1944 | unsigned long address) |
1da177e4 LT |
1945 | { |
1946 | int error; | |
1947 | ||
1948 | /* | |
1949 | * We must make sure the anon_vma is allocated | |
1950 | * so that the anon_vma locking is not a noop. | |
1951 | */ | |
1952 | if (unlikely(anon_vma_prepare(vma))) | |
1953 | return -ENOMEM; | |
8869477a EP |
1954 | |
1955 | address &= PAGE_MASK; | |
e5467859 | 1956 | error = security_mmap_addr(address); |
8869477a EP |
1957 | if (error) |
1958 | return error; | |
1959 | ||
bb4a340e | 1960 | vma_lock_anon_vma(vma); |
1da177e4 LT |
1961 | |
1962 | /* | |
1963 | * vma->vm_start/vm_end cannot change under us because the caller | |
1964 | * is required to hold the mmap_sem in read mode. We need the | |
1965 | * anon_vma lock to serialize against concurrent expand_stacks. | |
1966 | */ | |
1da177e4 LT |
1967 | |
1968 | /* Somebody else might have raced and expanded it already */ | |
1969 | if (address < vma->vm_start) { | |
1970 | unsigned long size, grow; | |
1971 | ||
1972 | size = vma->vm_end - address; | |
1973 | grow = (vma->vm_start - address) >> PAGE_SHIFT; | |
1974 | ||
a626ca6a LT |
1975 | error = -ENOMEM; |
1976 | if (grow <= vma->vm_pgoff) { | |
1977 | error = acct_stack_growth(vma, size, grow); | |
1978 | if (!error) { | |
bf181b9f | 1979 | anon_vma_interval_tree_pre_update_vma(vma); |
a626ca6a LT |
1980 | vma->vm_start = address; |
1981 | vma->vm_pgoff -= grow; | |
bf181b9f | 1982 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 | 1983 | vma_gap_update(vma); |
a626ca6a LT |
1984 | perf_event_mmap(vma); |
1985 | } | |
1da177e4 LT |
1986 | } |
1987 | } | |
bb4a340e | 1988 | vma_unlock_anon_vma(vma); |
b15d00b6 | 1989 | khugepaged_enter_vma_merge(vma); |
ed8ea815 | 1990 | validate_mm(vma->vm_mm); |
1da177e4 LT |
1991 | return error; |
1992 | } | |
1993 | ||
b6a2fea3 OW |
1994 | #ifdef CONFIG_STACK_GROWSUP |
1995 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
1996 | { | |
1997 | return expand_upwards(vma, address); | |
1998 | } | |
1999 | ||
2000 | struct vm_area_struct * | |
2001 | find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
2002 | { | |
2003 | struct vm_area_struct *vma, *prev; | |
2004 | ||
2005 | addr &= PAGE_MASK; | |
2006 | vma = find_vma_prev(mm, addr, &prev); | |
2007 | if (vma && (vma->vm_start <= addr)) | |
2008 | return vma; | |
1c127185 | 2009 | if (!prev || expand_stack(prev, addr)) |
b6a2fea3 | 2010 | return NULL; |
ba470de4 | 2011 | if (prev->vm_flags & VM_LOCKED) { |
c58267c3 | 2012 | mlock_vma_pages_range(prev, addr, prev->vm_end); |
ba470de4 | 2013 | } |
b6a2fea3 OW |
2014 | return prev; |
2015 | } | |
2016 | #else | |
2017 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2018 | { | |
2019 | return expand_downwards(vma, address); | |
2020 | } | |
2021 | ||
1da177e4 LT |
2022 | struct vm_area_struct * |
2023 | find_extend_vma(struct mm_struct * mm, unsigned long addr) | |
2024 | { | |
2025 | struct vm_area_struct * vma; | |
2026 | unsigned long start; | |
2027 | ||
2028 | addr &= PAGE_MASK; | |
2029 | vma = find_vma(mm,addr); | |
2030 | if (!vma) | |
2031 | return NULL; | |
2032 | if (vma->vm_start <= addr) | |
2033 | return vma; | |
2034 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
2035 | return NULL; | |
2036 | start = vma->vm_start; | |
2037 | if (expand_stack(vma, addr)) | |
2038 | return NULL; | |
ba470de4 | 2039 | if (vma->vm_flags & VM_LOCKED) { |
c58267c3 | 2040 | mlock_vma_pages_range(vma, addr, start); |
ba470de4 | 2041 | } |
1da177e4 LT |
2042 | return vma; |
2043 | } | |
2044 | #endif | |
2045 | ||
1da177e4 | 2046 | /* |
2c0b3814 | 2047 | * Ok - we have the memory areas we should free on the vma list, |
1da177e4 | 2048 | * so release them, and do the vma updates. |
2c0b3814 HD |
2049 | * |
2050 | * Called with the mm semaphore held. | |
1da177e4 | 2051 | */ |
2c0b3814 | 2052 | static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 2053 | { |
4f74d2c8 LT |
2054 | unsigned long nr_accounted = 0; |
2055 | ||
365e9c87 HD |
2056 | /* Update high watermark before we lower total_vm */ |
2057 | update_hiwater_vm(mm); | |
1da177e4 | 2058 | do { |
2c0b3814 HD |
2059 | long nrpages = vma_pages(vma); |
2060 | ||
4f74d2c8 LT |
2061 | if (vma->vm_flags & VM_ACCOUNT) |
2062 | nr_accounted += nrpages; | |
2c0b3814 | 2063 | vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); |
a8fb5618 | 2064 | vma = remove_vma(vma); |
146425a3 | 2065 | } while (vma); |
4f74d2c8 | 2066 | vm_unacct_memory(nr_accounted); |
1da177e4 LT |
2067 | validate_mm(mm); |
2068 | } | |
2069 | ||
2070 | /* | |
2071 | * Get rid of page table information in the indicated region. | |
2072 | * | |
f10df686 | 2073 | * Called with the mm semaphore held. |
1da177e4 LT |
2074 | */ |
2075 | static void unmap_region(struct mm_struct *mm, | |
e0da382c HD |
2076 | struct vm_area_struct *vma, struct vm_area_struct *prev, |
2077 | unsigned long start, unsigned long end) | |
1da177e4 | 2078 | { |
e0da382c | 2079 | struct vm_area_struct *next = prev? prev->vm_next: mm->mmap; |
d16dfc55 | 2080 | struct mmu_gather tlb; |
1da177e4 LT |
2081 | |
2082 | lru_add_drain(); | |
d16dfc55 | 2083 | tlb_gather_mmu(&tlb, mm, 0); |
365e9c87 | 2084 | update_hiwater_rss(mm); |
4f74d2c8 | 2085 | unmap_vmas(&tlb, vma, start, end); |
d16dfc55 PZ |
2086 | free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, |
2087 | next ? next->vm_start : 0); | |
2088 | tlb_finish_mmu(&tlb, start, end); | |
1da177e4 LT |
2089 | } |
2090 | ||
2091 | /* | |
2092 | * Create a list of vma's touched by the unmap, removing them from the mm's | |
2093 | * vma list as we go.. | |
2094 | */ | |
2095 | static void | |
2096 | detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, | |
2097 | struct vm_area_struct *prev, unsigned long end) | |
2098 | { | |
2099 | struct vm_area_struct **insertion_point; | |
2100 | struct vm_area_struct *tail_vma = NULL; | |
1363c3cd | 2101 | unsigned long addr; |
1da177e4 LT |
2102 | |
2103 | insertion_point = (prev ? &prev->vm_next : &mm->mmap); | |
297c5eee | 2104 | vma->vm_prev = NULL; |
1da177e4 | 2105 | do { |
d3737187 | 2106 | vma_rb_erase(vma, &mm->mm_rb); |
1da177e4 LT |
2107 | mm->map_count--; |
2108 | tail_vma = vma; | |
2109 | vma = vma->vm_next; | |
2110 | } while (vma && vma->vm_start < end); | |
2111 | *insertion_point = vma; | |
d3737187 | 2112 | if (vma) { |
297c5eee | 2113 | vma->vm_prev = prev; |
d3737187 ML |
2114 | vma_gap_update(vma); |
2115 | } else | |
2116 | mm->highest_vm_end = prev ? prev->vm_end : 0; | |
1da177e4 | 2117 | tail_vma->vm_next = NULL; |
1363c3cd WW |
2118 | if (mm->unmap_area == arch_unmap_area) |
2119 | addr = prev ? prev->vm_end : mm->mmap_base; | |
2120 | else | |
2121 | addr = vma ? vma->vm_start : mm->mmap_base; | |
2122 | mm->unmap_area(mm, addr); | |
1da177e4 LT |
2123 | mm->mmap_cache = NULL; /* Kill the cache. */ |
2124 | } | |
2125 | ||
2126 | /* | |
659ace58 KM |
2127 | * __split_vma() bypasses sysctl_max_map_count checking. We use this on the |
2128 | * munmap path where it doesn't make sense to fail. | |
1da177e4 | 2129 | */ |
659ace58 | 2130 | static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma, |
1da177e4 LT |
2131 | unsigned long addr, int new_below) |
2132 | { | |
2133 | struct mempolicy *pol; | |
2134 | struct vm_area_struct *new; | |
5beb4930 | 2135 | int err = -ENOMEM; |
1da177e4 | 2136 | |
a5516438 AK |
2137 | if (is_vm_hugetlb_page(vma) && (addr & |
2138 | ~(huge_page_mask(hstate_vma(vma))))) | |
1da177e4 LT |
2139 | return -EINVAL; |
2140 | ||
e94b1766 | 2141 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 | 2142 | if (!new) |
5beb4930 | 2143 | goto out_err; |
1da177e4 LT |
2144 | |
2145 | /* most fields are the same, copy all, and then fixup */ | |
2146 | *new = *vma; | |
2147 | ||
5beb4930 RR |
2148 | INIT_LIST_HEAD(&new->anon_vma_chain); |
2149 | ||
1da177e4 LT |
2150 | if (new_below) |
2151 | new->vm_end = addr; | |
2152 | else { | |
2153 | new->vm_start = addr; | |
2154 | new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); | |
2155 | } | |
2156 | ||
846a16bf | 2157 | pol = mpol_dup(vma_policy(vma)); |
1da177e4 | 2158 | if (IS_ERR(pol)) { |
5beb4930 RR |
2159 | err = PTR_ERR(pol); |
2160 | goto out_free_vma; | |
1da177e4 LT |
2161 | } |
2162 | vma_set_policy(new, pol); | |
2163 | ||
5beb4930 RR |
2164 | if (anon_vma_clone(new, vma)) |
2165 | goto out_free_mpol; | |
2166 | ||
e9714acf | 2167 | if (new->vm_file) |
1da177e4 LT |
2168 | get_file(new->vm_file); |
2169 | ||
2170 | if (new->vm_ops && new->vm_ops->open) | |
2171 | new->vm_ops->open(new); | |
2172 | ||
2173 | if (new_below) | |
5beb4930 | 2174 | err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + |
1da177e4 LT |
2175 | ((addr - new->vm_start) >> PAGE_SHIFT), new); |
2176 | else | |
5beb4930 | 2177 | err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); |
1da177e4 | 2178 | |
5beb4930 RR |
2179 | /* Success. */ |
2180 | if (!err) | |
2181 | return 0; | |
2182 | ||
2183 | /* Clean everything up if vma_adjust failed. */ | |
58927533 RR |
2184 | if (new->vm_ops && new->vm_ops->close) |
2185 | new->vm_ops->close(new); | |
e9714acf | 2186 | if (new->vm_file) |
5beb4930 | 2187 | fput(new->vm_file); |
2aeadc30 | 2188 | unlink_anon_vmas(new); |
5beb4930 RR |
2189 | out_free_mpol: |
2190 | mpol_put(pol); | |
2191 | out_free_vma: | |
2192 | kmem_cache_free(vm_area_cachep, new); | |
2193 | out_err: | |
2194 | return err; | |
1da177e4 LT |
2195 | } |
2196 | ||
659ace58 KM |
2197 | /* |
2198 | * Split a vma into two pieces at address 'addr', a new vma is allocated | |
2199 | * either for the first part or the tail. | |
2200 | */ | |
2201 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, | |
2202 | unsigned long addr, int new_below) | |
2203 | { | |
2204 | if (mm->map_count >= sysctl_max_map_count) | |
2205 | return -ENOMEM; | |
2206 | ||
2207 | return __split_vma(mm, vma, addr, new_below); | |
2208 | } | |
2209 | ||
1da177e4 LT |
2210 | /* Munmap is split into 2 main parts -- this part which finds |
2211 | * what needs doing, and the areas themselves, which do the | |
2212 | * work. This now handles partial unmappings. | |
2213 | * Jeremy Fitzhardinge <jeremy@goop.org> | |
2214 | */ | |
2215 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | |
2216 | { | |
2217 | unsigned long end; | |
146425a3 | 2218 | struct vm_area_struct *vma, *prev, *last; |
1da177e4 LT |
2219 | |
2220 | if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) | |
2221 | return -EINVAL; | |
2222 | ||
2223 | if ((len = PAGE_ALIGN(len)) == 0) | |
2224 | return -EINVAL; | |
2225 | ||
2226 | /* Find the first overlapping VMA */ | |
9be34c9d | 2227 | vma = find_vma(mm, start); |
146425a3 | 2228 | if (!vma) |
1da177e4 | 2229 | return 0; |
9be34c9d | 2230 | prev = vma->vm_prev; |
146425a3 | 2231 | /* we have start < vma->vm_end */ |
1da177e4 LT |
2232 | |
2233 | /* if it doesn't overlap, we have nothing.. */ | |
2234 | end = start + len; | |
146425a3 | 2235 | if (vma->vm_start >= end) |
1da177e4 LT |
2236 | return 0; |
2237 | ||
2238 | /* | |
2239 | * If we need to split any vma, do it now to save pain later. | |
2240 | * | |
2241 | * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially | |
2242 | * unmapped vm_area_struct will remain in use: so lower split_vma | |
2243 | * places tmp vma above, and higher split_vma places tmp vma below. | |
2244 | */ | |
146425a3 | 2245 | if (start > vma->vm_start) { |
659ace58 KM |
2246 | int error; |
2247 | ||
2248 | /* | |
2249 | * Make sure that map_count on return from munmap() will | |
2250 | * not exceed its limit; but let map_count go just above | |
2251 | * its limit temporarily, to help free resources as expected. | |
2252 | */ | |
2253 | if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) | |
2254 | return -ENOMEM; | |
2255 | ||
2256 | error = __split_vma(mm, vma, start, 0); | |
1da177e4 LT |
2257 | if (error) |
2258 | return error; | |
146425a3 | 2259 | prev = vma; |
1da177e4 LT |
2260 | } |
2261 | ||
2262 | /* Does it split the last one? */ | |
2263 | last = find_vma(mm, end); | |
2264 | if (last && end > last->vm_start) { | |
659ace58 | 2265 | int error = __split_vma(mm, last, end, 1); |
1da177e4 LT |
2266 | if (error) |
2267 | return error; | |
2268 | } | |
146425a3 | 2269 | vma = prev? prev->vm_next: mm->mmap; |
1da177e4 | 2270 | |
ba470de4 RR |
2271 | /* |
2272 | * unlock any mlock()ed ranges before detaching vmas | |
2273 | */ | |
2274 | if (mm->locked_vm) { | |
2275 | struct vm_area_struct *tmp = vma; | |
2276 | while (tmp && tmp->vm_start < end) { | |
2277 | if (tmp->vm_flags & VM_LOCKED) { | |
2278 | mm->locked_vm -= vma_pages(tmp); | |
2279 | munlock_vma_pages_all(tmp); | |
2280 | } | |
2281 | tmp = tmp->vm_next; | |
2282 | } | |
2283 | } | |
2284 | ||
1da177e4 LT |
2285 | /* |
2286 | * Remove the vma's, and unmap the actual pages | |
2287 | */ | |
146425a3 HD |
2288 | detach_vmas_to_be_unmapped(mm, vma, prev, end); |
2289 | unmap_region(mm, vma, prev, start, end); | |
1da177e4 LT |
2290 | |
2291 | /* Fix up all other VM information */ | |
2c0b3814 | 2292 | remove_vma_list(mm, vma); |
1da177e4 LT |
2293 | |
2294 | return 0; | |
2295 | } | |
1da177e4 | 2296 | |
bfce281c | 2297 | int vm_munmap(unsigned long start, size_t len) |
1da177e4 LT |
2298 | { |
2299 | int ret; | |
bfce281c | 2300 | struct mm_struct *mm = current->mm; |
1da177e4 LT |
2301 | |
2302 | down_write(&mm->mmap_sem); | |
a46ef99d | 2303 | ret = do_munmap(mm, start, len); |
1da177e4 LT |
2304 | up_write(&mm->mmap_sem); |
2305 | return ret; | |
2306 | } | |
a46ef99d LT |
2307 | EXPORT_SYMBOL(vm_munmap); |
2308 | ||
2309 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) | |
2310 | { | |
2311 | profile_munmap(addr); | |
bfce281c | 2312 | return vm_munmap(addr, len); |
a46ef99d | 2313 | } |
1da177e4 LT |
2314 | |
2315 | static inline void verify_mm_writelocked(struct mm_struct *mm) | |
2316 | { | |
a241ec65 | 2317 | #ifdef CONFIG_DEBUG_VM |
1da177e4 LT |
2318 | if (unlikely(down_read_trylock(&mm->mmap_sem))) { |
2319 | WARN_ON(1); | |
2320 | up_read(&mm->mmap_sem); | |
2321 | } | |
2322 | #endif | |
2323 | } | |
2324 | ||
2325 | /* | |
2326 | * this is really a simplified "do_mmap". it only handles | |
2327 | * anonymous maps. eventually we may be able to do some | |
2328 | * brk-specific accounting here. | |
2329 | */ | |
e4eb1ff6 | 2330 | static unsigned long do_brk(unsigned long addr, unsigned long len) |
1da177e4 LT |
2331 | { |
2332 | struct mm_struct * mm = current->mm; | |
2333 | struct vm_area_struct * vma, * prev; | |
2334 | unsigned long flags; | |
2335 | struct rb_node ** rb_link, * rb_parent; | |
2336 | pgoff_t pgoff = addr >> PAGE_SHIFT; | |
3a459756 | 2337 | int error; |
1da177e4 LT |
2338 | |
2339 | len = PAGE_ALIGN(len); | |
2340 | if (!len) | |
2341 | return addr; | |
2342 | ||
3a459756 KK |
2343 | flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; |
2344 | ||
2c6a1016 AV |
2345 | error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); |
2346 | if (error & ~PAGE_MASK) | |
3a459756 KK |
2347 | return error; |
2348 | ||
1da177e4 LT |
2349 | /* |
2350 | * mlock MCL_FUTURE? | |
2351 | */ | |
2352 | if (mm->def_flags & VM_LOCKED) { | |
2353 | unsigned long locked, lock_limit; | |
93ea1d0a CW |
2354 | locked = len >> PAGE_SHIFT; |
2355 | locked += mm->locked_vm; | |
59e99e5b | 2356 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
93ea1d0a | 2357 | lock_limit >>= PAGE_SHIFT; |
1da177e4 LT |
2358 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) |
2359 | return -EAGAIN; | |
2360 | } | |
2361 | ||
2362 | /* | |
2363 | * mm->mmap_sem is required to protect against another thread | |
2364 | * changing the mappings in case we sleep. | |
2365 | */ | |
2366 | verify_mm_writelocked(mm); | |
2367 | ||
2368 | /* | |
2369 | * Clear old maps. this also does some error checking for us | |
2370 | */ | |
2371 | munmap_back: | |
6597d783 | 2372 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { |
1da177e4 LT |
2373 | if (do_munmap(mm, addr, len)) |
2374 | return -ENOMEM; | |
2375 | goto munmap_back; | |
2376 | } | |
2377 | ||
2378 | /* Check against address space limits *after* clearing old maps... */ | |
119f657c | 2379 | if (!may_expand_vm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
2380 | return -ENOMEM; |
2381 | ||
2382 | if (mm->map_count > sysctl_max_map_count) | |
2383 | return -ENOMEM; | |
2384 | ||
191c5424 | 2385 | if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
2386 | return -ENOMEM; |
2387 | ||
1da177e4 | 2388 | /* Can we just expand an old private anonymous mapping? */ |
ba470de4 RR |
2389 | vma = vma_merge(mm, prev, addr, addr + len, flags, |
2390 | NULL, NULL, pgoff, NULL); | |
2391 | if (vma) | |
1da177e4 LT |
2392 | goto out; |
2393 | ||
2394 | /* | |
2395 | * create a vma struct for an anonymous mapping | |
2396 | */ | |
c5e3b83e | 2397 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
2398 | if (!vma) { |
2399 | vm_unacct_memory(len >> PAGE_SHIFT); | |
2400 | return -ENOMEM; | |
2401 | } | |
1da177e4 | 2402 | |
5beb4930 | 2403 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 LT |
2404 | vma->vm_mm = mm; |
2405 | vma->vm_start = addr; | |
2406 | vma->vm_end = addr + len; | |
2407 | vma->vm_pgoff = pgoff; | |
2408 | vma->vm_flags = flags; | |
3ed75eb8 | 2409 | vma->vm_page_prot = vm_get_page_prot(flags); |
1da177e4 LT |
2410 | vma_link(mm, vma, prev, rb_link, rb_parent); |
2411 | out: | |
3af9e859 | 2412 | perf_event_mmap(vma); |
1da177e4 LT |
2413 | mm->total_vm += len >> PAGE_SHIFT; |
2414 | if (flags & VM_LOCKED) { | |
ba470de4 RR |
2415 | if (!mlock_vma_pages_range(vma, addr, addr + len)) |
2416 | mm->locked_vm += (len >> PAGE_SHIFT); | |
1da177e4 LT |
2417 | } |
2418 | return addr; | |
2419 | } | |
2420 | ||
e4eb1ff6 LT |
2421 | unsigned long vm_brk(unsigned long addr, unsigned long len) |
2422 | { | |
2423 | struct mm_struct *mm = current->mm; | |
2424 | unsigned long ret; | |
2425 | ||
2426 | down_write(&mm->mmap_sem); | |
2427 | ret = do_brk(addr, len); | |
2428 | up_write(&mm->mmap_sem); | |
2429 | return ret; | |
2430 | } | |
2431 | EXPORT_SYMBOL(vm_brk); | |
1da177e4 LT |
2432 | |
2433 | /* Release all mmaps. */ | |
2434 | void exit_mmap(struct mm_struct *mm) | |
2435 | { | |
d16dfc55 | 2436 | struct mmu_gather tlb; |
ba470de4 | 2437 | struct vm_area_struct *vma; |
1da177e4 LT |
2438 | unsigned long nr_accounted = 0; |
2439 | ||
d6dd61c8 | 2440 | /* mm's last user has gone, and its about to be pulled down */ |
cddb8a5c | 2441 | mmu_notifier_release(mm); |
d6dd61c8 | 2442 | |
ba470de4 RR |
2443 | if (mm->locked_vm) { |
2444 | vma = mm->mmap; | |
2445 | while (vma) { | |
2446 | if (vma->vm_flags & VM_LOCKED) | |
2447 | munlock_vma_pages_all(vma); | |
2448 | vma = vma->vm_next; | |
2449 | } | |
2450 | } | |
9480c53e JF |
2451 | |
2452 | arch_exit_mmap(mm); | |
2453 | ||
ba470de4 | 2454 | vma = mm->mmap; |
9480c53e JF |
2455 | if (!vma) /* Can happen if dup_mmap() received an OOM */ |
2456 | return; | |
2457 | ||
1da177e4 | 2458 | lru_add_drain(); |
1da177e4 | 2459 | flush_cache_mm(mm); |
d16dfc55 | 2460 | tlb_gather_mmu(&tlb, mm, 1); |
901608d9 | 2461 | /* update_hiwater_rss(mm) here? but nobody should be looking */ |
e0da382c | 2462 | /* Use -1 here to ensure all VMAs in the mm are unmapped */ |
4f74d2c8 | 2463 | unmap_vmas(&tlb, vma, 0, -1); |
9ba69294 | 2464 | |
d16dfc55 | 2465 | free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0); |
853f5e26 | 2466 | tlb_finish_mmu(&tlb, 0, -1); |
1da177e4 | 2467 | |
1da177e4 | 2468 | /* |
8f4f8c16 HD |
2469 | * Walk the list again, actually closing and freeing it, |
2470 | * with preemption enabled, without holding any MM locks. | |
1da177e4 | 2471 | */ |
4f74d2c8 LT |
2472 | while (vma) { |
2473 | if (vma->vm_flags & VM_ACCOUNT) | |
2474 | nr_accounted += vma_pages(vma); | |
a8fb5618 | 2475 | vma = remove_vma(vma); |
4f74d2c8 LT |
2476 | } |
2477 | vm_unacct_memory(nr_accounted); | |
e0da382c | 2478 | |
f9aed62a | 2479 | WARN_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT); |
1da177e4 LT |
2480 | } |
2481 | ||
2482 | /* Insert vm structure into process list sorted by address | |
2483 | * and into the inode's i_mmap tree. If vm_file is non-NULL | |
3d48ae45 | 2484 | * then i_mmap_mutex is taken here. |
1da177e4 | 2485 | */ |
6597d783 | 2486 | int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 2487 | { |
6597d783 HD |
2488 | struct vm_area_struct *prev; |
2489 | struct rb_node **rb_link, *rb_parent; | |
1da177e4 LT |
2490 | |
2491 | /* | |
2492 | * The vm_pgoff of a purely anonymous vma should be irrelevant | |
2493 | * until its first write fault, when page's anon_vma and index | |
2494 | * are set. But now set the vm_pgoff it will almost certainly | |
2495 | * end up with (unless mremap moves it elsewhere before that | |
2496 | * first wfault), so /proc/pid/maps tells a consistent story. | |
2497 | * | |
2498 | * By setting it to reflect the virtual start address of the | |
2499 | * vma, merges and splits can happen in a seamless way, just | |
2500 | * using the existing file pgoff checks and manipulations. | |
2501 | * Similarly in do_mmap_pgoff and in do_brk. | |
2502 | */ | |
2503 | if (!vma->vm_file) { | |
2504 | BUG_ON(vma->anon_vma); | |
2505 | vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; | |
2506 | } | |
6597d783 HD |
2507 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
2508 | &prev, &rb_link, &rb_parent)) | |
1da177e4 | 2509 | return -ENOMEM; |
2fd4ef85 | 2510 | if ((vma->vm_flags & VM_ACCOUNT) && |
34b4e4aa | 2511 | security_vm_enough_memory_mm(mm, vma_pages(vma))) |
2fd4ef85 | 2512 | return -ENOMEM; |
2b144498 | 2513 | |
1da177e4 LT |
2514 | vma_link(mm, vma, prev, rb_link, rb_parent); |
2515 | return 0; | |
2516 | } | |
2517 | ||
2518 | /* | |
2519 | * Copy the vma structure to a new location in the same mm, | |
2520 | * prior to moving page table entries, to effect an mremap move. | |
2521 | */ | |
2522 | struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, | |
38a76013 ML |
2523 | unsigned long addr, unsigned long len, pgoff_t pgoff, |
2524 | bool *need_rmap_locks) | |
1da177e4 LT |
2525 | { |
2526 | struct vm_area_struct *vma = *vmap; | |
2527 | unsigned long vma_start = vma->vm_start; | |
2528 | struct mm_struct *mm = vma->vm_mm; | |
2529 | struct vm_area_struct *new_vma, *prev; | |
2530 | struct rb_node **rb_link, *rb_parent; | |
2531 | struct mempolicy *pol; | |
948f017b | 2532 | bool faulted_in_anon_vma = true; |
1da177e4 LT |
2533 | |
2534 | /* | |
2535 | * If anonymous vma has not yet been faulted, update new pgoff | |
2536 | * to match new location, to increase its chance of merging. | |
2537 | */ | |
948f017b | 2538 | if (unlikely(!vma->vm_file && !vma->anon_vma)) { |
1da177e4 | 2539 | pgoff = addr >> PAGE_SHIFT; |
948f017b AA |
2540 | faulted_in_anon_vma = false; |
2541 | } | |
1da177e4 | 2542 | |
6597d783 HD |
2543 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) |
2544 | return NULL; /* should never get here */ | |
1da177e4 LT |
2545 | new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, |
2546 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); | |
2547 | if (new_vma) { | |
2548 | /* | |
2549 | * Source vma may have been merged into new_vma | |
2550 | */ | |
948f017b AA |
2551 | if (unlikely(vma_start >= new_vma->vm_start && |
2552 | vma_start < new_vma->vm_end)) { | |
2553 | /* | |
2554 | * The only way we can get a vma_merge with | |
2555 | * self during an mremap is if the vma hasn't | |
2556 | * been faulted in yet and we were allowed to | |
2557 | * reset the dst vma->vm_pgoff to the | |
2558 | * destination address of the mremap to allow | |
2559 | * the merge to happen. mremap must change the | |
2560 | * vm_pgoff linearity between src and dst vmas | |
2561 | * (in turn preventing a vma_merge) to be | |
2562 | * safe. It is only safe to keep the vm_pgoff | |
2563 | * linear if there are no pages mapped yet. | |
2564 | */ | |
2565 | VM_BUG_ON(faulted_in_anon_vma); | |
38a76013 | 2566 | *vmap = vma = new_vma; |
108d6642 | 2567 | } |
38a76013 | 2568 | *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); |
1da177e4 | 2569 | } else { |
e94b1766 | 2570 | new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
2571 | if (new_vma) { |
2572 | *new_vma = *vma; | |
523d4e20 ML |
2573 | new_vma->vm_start = addr; |
2574 | new_vma->vm_end = addr + len; | |
2575 | new_vma->vm_pgoff = pgoff; | |
846a16bf | 2576 | pol = mpol_dup(vma_policy(vma)); |
5beb4930 RR |
2577 | if (IS_ERR(pol)) |
2578 | goto out_free_vma; | |
523d4e20 | 2579 | vma_set_policy(new_vma, pol); |
5beb4930 RR |
2580 | INIT_LIST_HEAD(&new_vma->anon_vma_chain); |
2581 | if (anon_vma_clone(new_vma, vma)) | |
2582 | goto out_free_mempol; | |
e9714acf | 2583 | if (new_vma->vm_file) |
1da177e4 LT |
2584 | get_file(new_vma->vm_file); |
2585 | if (new_vma->vm_ops && new_vma->vm_ops->open) | |
2586 | new_vma->vm_ops->open(new_vma); | |
2587 | vma_link(mm, new_vma, prev, rb_link, rb_parent); | |
38a76013 | 2588 | *need_rmap_locks = false; |
1da177e4 LT |
2589 | } |
2590 | } | |
2591 | return new_vma; | |
5beb4930 RR |
2592 | |
2593 | out_free_mempol: | |
2594 | mpol_put(pol); | |
2595 | out_free_vma: | |
2596 | kmem_cache_free(vm_area_cachep, new_vma); | |
2597 | return NULL; | |
1da177e4 | 2598 | } |
119f657c | 2599 | |
2600 | /* | |
2601 | * Return true if the calling process may expand its vm space by the passed | |
2602 | * number of pages | |
2603 | */ | |
2604 | int may_expand_vm(struct mm_struct *mm, unsigned long npages) | |
2605 | { | |
2606 | unsigned long cur = mm->total_vm; /* pages */ | |
2607 | unsigned long lim; | |
2608 | ||
59e99e5b | 2609 | lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT; |
119f657c | 2610 | |
2611 | if (cur + npages > lim) | |
2612 | return 0; | |
2613 | return 1; | |
2614 | } | |
fa5dc22f RM |
2615 | |
2616 | ||
b1d0e4f5 NP |
2617 | static int special_mapping_fault(struct vm_area_struct *vma, |
2618 | struct vm_fault *vmf) | |
fa5dc22f | 2619 | { |
b1d0e4f5 | 2620 | pgoff_t pgoff; |
fa5dc22f RM |
2621 | struct page **pages; |
2622 | ||
b1d0e4f5 NP |
2623 | /* |
2624 | * special mappings have no vm_file, and in that case, the mm | |
2625 | * uses vm_pgoff internally. So we have to subtract it from here. | |
2626 | * We are allowed to do this because we are the mm; do not copy | |
2627 | * this code into drivers! | |
2628 | */ | |
2629 | pgoff = vmf->pgoff - vma->vm_pgoff; | |
fa5dc22f | 2630 | |
b1d0e4f5 NP |
2631 | for (pages = vma->vm_private_data; pgoff && *pages; ++pages) |
2632 | pgoff--; | |
fa5dc22f RM |
2633 | |
2634 | if (*pages) { | |
2635 | struct page *page = *pages; | |
2636 | get_page(page); | |
b1d0e4f5 NP |
2637 | vmf->page = page; |
2638 | return 0; | |
fa5dc22f RM |
2639 | } |
2640 | ||
b1d0e4f5 | 2641 | return VM_FAULT_SIGBUS; |
fa5dc22f RM |
2642 | } |
2643 | ||
2644 | /* | |
2645 | * Having a close hook prevents vma merging regardless of flags. | |
2646 | */ | |
2647 | static void special_mapping_close(struct vm_area_struct *vma) | |
2648 | { | |
2649 | } | |
2650 | ||
f0f37e2f | 2651 | static const struct vm_operations_struct special_mapping_vmops = { |
fa5dc22f | 2652 | .close = special_mapping_close, |
b1d0e4f5 | 2653 | .fault = special_mapping_fault, |
fa5dc22f RM |
2654 | }; |
2655 | ||
2656 | /* | |
2657 | * Called with mm->mmap_sem held for writing. | |
2658 | * Insert a new vma covering the given region, with the given flags. | |
2659 | * Its pages are supplied by the given array of struct page *. | |
2660 | * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. | |
2661 | * The region past the last page supplied will always produce SIGBUS. | |
2662 | * The array pointer and the pages it points to are assumed to stay alive | |
2663 | * for as long as this mapping might exist. | |
2664 | */ | |
2665 | int install_special_mapping(struct mm_struct *mm, | |
2666 | unsigned long addr, unsigned long len, | |
2667 | unsigned long vm_flags, struct page **pages) | |
2668 | { | |
462e635e | 2669 | int ret; |
fa5dc22f RM |
2670 | struct vm_area_struct *vma; |
2671 | ||
2672 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
2673 | if (unlikely(vma == NULL)) | |
2674 | return -ENOMEM; | |
2675 | ||
5beb4930 | 2676 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
fa5dc22f RM |
2677 | vma->vm_mm = mm; |
2678 | vma->vm_start = addr; | |
2679 | vma->vm_end = addr + len; | |
2680 | ||
2f98735c | 2681 | vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND; |
3ed75eb8 | 2682 | vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); |
fa5dc22f RM |
2683 | |
2684 | vma->vm_ops = &special_mapping_vmops; | |
2685 | vma->vm_private_data = pages; | |
2686 | ||
462e635e TO |
2687 | ret = insert_vm_struct(mm, vma); |
2688 | if (ret) | |
2689 | goto out; | |
fa5dc22f RM |
2690 | |
2691 | mm->total_vm += len >> PAGE_SHIFT; | |
2692 | ||
cdd6c482 | 2693 | perf_event_mmap(vma); |
089dd79d | 2694 | |
fa5dc22f | 2695 | return 0; |
462e635e TO |
2696 | |
2697 | out: | |
2698 | kmem_cache_free(vm_area_cachep, vma); | |
2699 | return ret; | |
fa5dc22f | 2700 | } |
7906d00c AA |
2701 | |
2702 | static DEFINE_MUTEX(mm_all_locks_mutex); | |
2703 | ||
454ed842 | 2704 | static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) |
7906d00c | 2705 | { |
bf181b9f | 2706 | if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
7906d00c AA |
2707 | /* |
2708 | * The LSB of head.next can't change from under us | |
2709 | * because we hold the mm_all_locks_mutex. | |
2710 | */ | |
2b575eb6 | 2711 | mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem); |
7906d00c AA |
2712 | /* |
2713 | * We can safely modify head.next after taking the | |
2b575eb6 | 2714 | * anon_vma->root->mutex. If some other vma in this mm shares |
7906d00c AA |
2715 | * the same anon_vma we won't take it again. |
2716 | * | |
2717 | * No need of atomic instructions here, head.next | |
2718 | * can't change from under us thanks to the | |
2b575eb6 | 2719 | * anon_vma->root->mutex. |
7906d00c AA |
2720 | */ |
2721 | if (__test_and_set_bit(0, (unsigned long *) | |
bf181b9f | 2722 | &anon_vma->root->rb_root.rb_node)) |
7906d00c AA |
2723 | BUG(); |
2724 | } | |
2725 | } | |
2726 | ||
454ed842 | 2727 | static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) |
7906d00c AA |
2728 | { |
2729 | if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
2730 | /* | |
2731 | * AS_MM_ALL_LOCKS can't change from under us because | |
2732 | * we hold the mm_all_locks_mutex. | |
2733 | * | |
2734 | * Operations on ->flags have to be atomic because | |
2735 | * even if AS_MM_ALL_LOCKS is stable thanks to the | |
2736 | * mm_all_locks_mutex, there may be other cpus | |
2737 | * changing other bitflags in parallel to us. | |
2738 | */ | |
2739 | if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) | |
2740 | BUG(); | |
3d48ae45 | 2741 | mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem); |
7906d00c AA |
2742 | } |
2743 | } | |
2744 | ||
2745 | /* | |
2746 | * This operation locks against the VM for all pte/vma/mm related | |
2747 | * operations that could ever happen on a certain mm. This includes | |
2748 | * vmtruncate, try_to_unmap, and all page faults. | |
2749 | * | |
2750 | * The caller must take the mmap_sem in write mode before calling | |
2751 | * mm_take_all_locks(). The caller isn't allowed to release the | |
2752 | * mmap_sem until mm_drop_all_locks() returns. | |
2753 | * | |
2754 | * mmap_sem in write mode is required in order to block all operations | |
2755 | * that could modify pagetables and free pages without need of | |
2756 | * altering the vma layout (for example populate_range() with | |
2757 | * nonlinear vmas). It's also needed in write mode to avoid new | |
2758 | * anon_vmas to be associated with existing vmas. | |
2759 | * | |
2760 | * A single task can't take more than one mm_take_all_locks() in a row | |
2761 | * or it would deadlock. | |
2762 | * | |
bf181b9f | 2763 | * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in |
7906d00c AA |
2764 | * mapping->flags avoid to take the same lock twice, if more than one |
2765 | * vma in this mm is backed by the same anon_vma or address_space. | |
2766 | * | |
2767 | * We can take all the locks in random order because the VM code | |
2b575eb6 | 2768 | * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never |
7906d00c AA |
2769 | * takes more than one of them in a row. Secondly we're protected |
2770 | * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex. | |
2771 | * | |
2772 | * mm_take_all_locks() and mm_drop_all_locks are expensive operations | |
2773 | * that may have to take thousand of locks. | |
2774 | * | |
2775 | * mm_take_all_locks() can fail if it's interrupted by signals. | |
2776 | */ | |
2777 | int mm_take_all_locks(struct mm_struct *mm) | |
2778 | { | |
2779 | struct vm_area_struct *vma; | |
5beb4930 | 2780 | struct anon_vma_chain *avc; |
7906d00c AA |
2781 | |
2782 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
2783 | ||
2784 | mutex_lock(&mm_all_locks_mutex); | |
2785 | ||
2786 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
2787 | if (signal_pending(current)) | |
2788 | goto out_unlock; | |
7906d00c | 2789 | if (vma->vm_file && vma->vm_file->f_mapping) |
454ed842 | 2790 | vm_lock_mapping(mm, vma->vm_file->f_mapping); |
7906d00c | 2791 | } |
7cd5a02f PZ |
2792 | |
2793 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
2794 | if (signal_pending(current)) | |
2795 | goto out_unlock; | |
2796 | if (vma->anon_vma) | |
5beb4930 RR |
2797 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
2798 | vm_lock_anon_vma(mm, avc->anon_vma); | |
7906d00c | 2799 | } |
7cd5a02f | 2800 | |
584cff54 | 2801 | return 0; |
7906d00c AA |
2802 | |
2803 | out_unlock: | |
584cff54 KC |
2804 | mm_drop_all_locks(mm); |
2805 | return -EINTR; | |
7906d00c AA |
2806 | } |
2807 | ||
2808 | static void vm_unlock_anon_vma(struct anon_vma *anon_vma) | |
2809 | { | |
bf181b9f | 2810 | if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
7906d00c AA |
2811 | /* |
2812 | * The LSB of head.next can't change to 0 from under | |
2813 | * us because we hold the mm_all_locks_mutex. | |
2814 | * | |
2815 | * We must however clear the bitflag before unlocking | |
bf181b9f | 2816 | * the vma so the users using the anon_vma->rb_root will |
7906d00c AA |
2817 | * never see our bitflag. |
2818 | * | |
2819 | * No need of atomic instructions here, head.next | |
2820 | * can't change from under us until we release the | |
2b575eb6 | 2821 | * anon_vma->root->mutex. |
7906d00c AA |
2822 | */ |
2823 | if (!__test_and_clear_bit(0, (unsigned long *) | |
bf181b9f | 2824 | &anon_vma->root->rb_root.rb_node)) |
7906d00c | 2825 | BUG(); |
cba48b98 | 2826 | anon_vma_unlock(anon_vma); |
7906d00c AA |
2827 | } |
2828 | } | |
2829 | ||
2830 | static void vm_unlock_mapping(struct address_space *mapping) | |
2831 | { | |
2832 | if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
2833 | /* | |
2834 | * AS_MM_ALL_LOCKS can't change to 0 from under us | |
2835 | * because we hold the mm_all_locks_mutex. | |
2836 | */ | |
3d48ae45 | 2837 | mutex_unlock(&mapping->i_mmap_mutex); |
7906d00c AA |
2838 | if (!test_and_clear_bit(AS_MM_ALL_LOCKS, |
2839 | &mapping->flags)) | |
2840 | BUG(); | |
2841 | } | |
2842 | } | |
2843 | ||
2844 | /* | |
2845 | * The mmap_sem cannot be released by the caller until | |
2846 | * mm_drop_all_locks() returns. | |
2847 | */ | |
2848 | void mm_drop_all_locks(struct mm_struct *mm) | |
2849 | { | |
2850 | struct vm_area_struct *vma; | |
5beb4930 | 2851 | struct anon_vma_chain *avc; |
7906d00c AA |
2852 | |
2853 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
2854 | BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); | |
2855 | ||
2856 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
2857 | if (vma->anon_vma) | |
5beb4930 RR |
2858 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
2859 | vm_unlock_anon_vma(avc->anon_vma); | |
7906d00c AA |
2860 | if (vma->vm_file && vma->vm_file->f_mapping) |
2861 | vm_unlock_mapping(vma->vm_file->f_mapping); | |
2862 | } | |
2863 | ||
2864 | mutex_unlock(&mm_all_locks_mutex); | |
2865 | } | |
8feae131 DH |
2866 | |
2867 | /* | |
2868 | * initialise the VMA slab | |
2869 | */ | |
2870 | void __init mmap_init(void) | |
2871 | { | |
00a62ce9 KM |
2872 | int ret; |
2873 | ||
2874 | ret = percpu_counter_init(&vm_committed_as, 0); | |
2875 | VM_BUG_ON(ret); | |
8feae131 | 2876 | } |