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 | ||
e8420a8e | 9 | #include <linux/kernel.h> |
1da177e4 | 10 | #include <linux/slab.h> |
4af3c9cc | 11 | #include <linux/backing-dev.h> |
1da177e4 LT |
12 | #include <linux/mm.h> |
13 | #include <linux/shm.h> | |
14 | #include <linux/mman.h> | |
15 | #include <linux/pagemap.h> | |
16 | #include <linux/swap.h> | |
17 | #include <linux/syscalls.h> | |
c59ede7b | 18 | #include <linux/capability.h> |
1da177e4 LT |
19 | #include <linux/init.h> |
20 | #include <linux/file.h> | |
21 | #include <linux/fs.h> | |
22 | #include <linux/personality.h> | |
23 | #include <linux/security.h> | |
24 | #include <linux/hugetlb.h> | |
25 | #include <linux/profile.h> | |
b95f1b31 | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/mount.h> |
28 | #include <linux/mempolicy.h> | |
29 | #include <linux/rmap.h> | |
cddb8a5c | 30 | #include <linux/mmu_notifier.h> |
cdd6c482 | 31 | #include <linux/perf_event.h> |
120a795d | 32 | #include <linux/audit.h> |
b15d00b6 | 33 | #include <linux/khugepaged.h> |
2b144498 | 34 | #include <linux/uprobes.h> |
d3737187 | 35 | #include <linux/rbtree_augmented.h> |
cf4aebc2 | 36 | #include <linux/sched/sysctl.h> |
1640879a AS |
37 | #include <linux/notifier.h> |
38 | #include <linux/memory.h> | |
1da177e4 LT |
39 | |
40 | #include <asm/uaccess.h> | |
41 | #include <asm/cacheflush.h> | |
42 | #include <asm/tlb.h> | |
d6dd61c8 | 43 | #include <asm/mmu_context.h> |
1da177e4 | 44 | |
42b77728 JB |
45 | #include "internal.h" |
46 | ||
3a459756 KK |
47 | #ifndef arch_mmap_check |
48 | #define arch_mmap_check(addr, len, flags) (0) | |
49 | #endif | |
50 | ||
08e7d9b5 MS |
51 | #ifndef arch_rebalance_pgtables |
52 | #define arch_rebalance_pgtables(addr, len) (addr) | |
53 | #endif | |
54 | ||
e0da382c HD |
55 | static void unmap_region(struct mm_struct *mm, |
56 | struct vm_area_struct *vma, struct vm_area_struct *prev, | |
57 | unsigned long start, unsigned long end); | |
58 | ||
1da177e4 LT |
59 | /* description of effects of mapping type and prot in current implementation. |
60 | * this is due to the limited x86 page protection hardware. The expected | |
61 | * behavior is in parens: | |
62 | * | |
63 | * map_type prot | |
64 | * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC | |
65 | * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes | |
66 | * w: (no) no w: (no) no w: (yes) yes w: (no) no | |
67 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
68 | * | |
69 | * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes | |
70 | * w: (no) no w: (no) no w: (copy) copy w: (no) no | |
71 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
72 | * | |
73 | */ | |
74 | pgprot_t protection_map[16] = { | |
75 | __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, | |
76 | __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 | |
77 | }; | |
78 | ||
804af2cf HD |
79 | pgprot_t vm_get_page_prot(unsigned long vm_flags) |
80 | { | |
b845f313 DK |
81 | return __pgprot(pgprot_val(protection_map[vm_flags & |
82 | (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | | |
83 | pgprot_val(arch_vm_get_page_prot(vm_flags))); | |
804af2cf HD |
84 | } |
85 | EXPORT_SYMBOL(vm_get_page_prot); | |
86 | ||
34679d7e SL |
87 | int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */ |
88 | int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */ | |
c3d8c141 | 89 | int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; |
c9b1d098 | 90 | unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ |
4eeab4f5 | 91 | unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ |
34679d7e SL |
92 | /* |
93 | * Make sure vm_committed_as in one cacheline and not cacheline shared with | |
94 | * other variables. It can be updated by several CPUs frequently. | |
95 | */ | |
96 | struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; | |
1da177e4 | 97 | |
997071bc S |
98 | /* |
99 | * The global memory commitment made in the system can be a metric | |
100 | * that can be used to drive ballooning decisions when Linux is hosted | |
101 | * as a guest. On Hyper-V, the host implements a policy engine for dynamically | |
102 | * balancing memory across competing virtual machines that are hosted. | |
103 | * Several metrics drive this policy engine including the guest reported | |
104 | * memory commitment. | |
105 | */ | |
106 | unsigned long vm_memory_committed(void) | |
107 | { | |
108 | return percpu_counter_read_positive(&vm_committed_as); | |
109 | } | |
110 | EXPORT_SYMBOL_GPL(vm_memory_committed); | |
111 | ||
1da177e4 LT |
112 | /* |
113 | * Check that a process has enough memory to allocate a new virtual | |
114 | * mapping. 0 means there is enough memory for the allocation to | |
115 | * succeed and -ENOMEM implies there is not. | |
116 | * | |
117 | * We currently support three overcommit policies, which are set via the | |
118 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting | |
119 | * | |
120 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. | |
121 | * Additional code 2002 Jul 20 by Robert Love. | |
122 | * | |
123 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. | |
124 | * | |
125 | * Note this is a helper function intended to be used by LSMs which | |
126 | * wish to use this logic. | |
127 | */ | |
34b4e4aa | 128 | int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) |
1da177e4 | 129 | { |
992f1cae | 130 | long free, allowed, reserve; |
1da177e4 LT |
131 | |
132 | vm_acct_memory(pages); | |
133 | ||
134 | /* | |
135 | * Sometimes we want to use more memory than we have | |
136 | */ | |
137 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) | |
138 | return 0; | |
139 | ||
140 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { | |
c15bef30 DF |
141 | free = global_page_state(NR_FREE_PAGES); |
142 | free += global_page_state(NR_FILE_PAGES); | |
143 | ||
144 | /* | |
145 | * shmem pages shouldn't be counted as free in this | |
146 | * case, they can't be purged, only swapped out, and | |
147 | * that won't affect the overall amount of available | |
148 | * memory in the system. | |
149 | */ | |
150 | free -= global_page_state(NR_SHMEM); | |
1da177e4 | 151 | |
ec8acf20 | 152 | free += get_nr_swap_pages(); |
1da177e4 LT |
153 | |
154 | /* | |
155 | * Any slabs which are created with the | |
156 | * SLAB_RECLAIM_ACCOUNT flag claim to have contents | |
157 | * which are reclaimable, under pressure. The dentry | |
158 | * cache and most inode caches should fall into this | |
159 | */ | |
972d1a7b | 160 | free += global_page_state(NR_SLAB_RECLAIMABLE); |
1da177e4 | 161 | |
6d9f7839 HA |
162 | /* |
163 | * Leave reserved pages. The pages are not for anonymous pages. | |
164 | */ | |
c15bef30 | 165 | if (free <= totalreserve_pages) |
6d9f7839 HA |
166 | goto error; |
167 | else | |
c15bef30 | 168 | free -= totalreserve_pages; |
6d9f7839 HA |
169 | |
170 | /* | |
4eeab4f5 | 171 | * Reserve some for root |
6d9f7839 | 172 | */ |
1da177e4 | 173 | if (!cap_sys_admin) |
4eeab4f5 | 174 | free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); |
1da177e4 LT |
175 | |
176 | if (free > pages) | |
177 | return 0; | |
6d9f7839 HA |
178 | |
179 | goto error; | |
1da177e4 LT |
180 | } |
181 | ||
182 | allowed = (totalram_pages - hugetlb_total_pages()) | |
183 | * sysctl_overcommit_ratio / 100; | |
184 | /* | |
4eeab4f5 | 185 | * Reserve some for root |
1da177e4 LT |
186 | */ |
187 | if (!cap_sys_admin) | |
4eeab4f5 | 188 | allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); |
1da177e4 LT |
189 | allowed += total_swap_pages; |
190 | ||
c9b1d098 AS |
191 | /* |
192 | * Don't let a single process grow so big a user can't recover | |
193 | */ | |
194 | if (mm) { | |
195 | reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); | |
992f1cae | 196 | allowed -= min_t(long, mm->total_vm / 32, reserve); |
c9b1d098 | 197 | } |
1da177e4 | 198 | |
00a62ce9 | 199 | if (percpu_counter_read_positive(&vm_committed_as) < allowed) |
1da177e4 | 200 | return 0; |
6d9f7839 | 201 | error: |
1da177e4 LT |
202 | vm_unacct_memory(pages); |
203 | ||
204 | return -ENOMEM; | |
205 | } | |
206 | ||
1da177e4 | 207 | /* |
3d48ae45 | 208 | * Requires inode->i_mapping->i_mmap_mutex |
1da177e4 LT |
209 | */ |
210 | static void __remove_shared_vm_struct(struct vm_area_struct *vma, | |
211 | struct file *file, struct address_space *mapping) | |
212 | { | |
213 | if (vma->vm_flags & VM_DENYWRITE) | |
496ad9aa | 214 | atomic_inc(&file_inode(file)->i_writecount); |
1da177e4 LT |
215 | if (vma->vm_flags & VM_SHARED) |
216 | mapping->i_mmap_writable--; | |
217 | ||
218 | flush_dcache_mmap_lock(mapping); | |
219 | if (unlikely(vma->vm_flags & VM_NONLINEAR)) | |
6b2dbba8 | 220 | list_del_init(&vma->shared.nonlinear); |
1da177e4 | 221 | else |
6b2dbba8 | 222 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
1da177e4 LT |
223 | flush_dcache_mmap_unlock(mapping); |
224 | } | |
225 | ||
226 | /* | |
6b2dbba8 | 227 | * Unlink a file-based vm structure from its interval tree, to hide |
a8fb5618 | 228 | * vma from rmap and vmtruncate before freeing its page tables. |
1da177e4 | 229 | */ |
a8fb5618 | 230 | void unlink_file_vma(struct vm_area_struct *vma) |
1da177e4 LT |
231 | { |
232 | struct file *file = vma->vm_file; | |
233 | ||
1da177e4 LT |
234 | if (file) { |
235 | struct address_space *mapping = file->f_mapping; | |
3d48ae45 | 236 | mutex_lock(&mapping->i_mmap_mutex); |
1da177e4 | 237 | __remove_shared_vm_struct(vma, file, mapping); |
3d48ae45 | 238 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 | 239 | } |
a8fb5618 HD |
240 | } |
241 | ||
242 | /* | |
243 | * Close a vm structure and free it, returning the next. | |
244 | */ | |
245 | static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) | |
246 | { | |
247 | struct vm_area_struct *next = vma->vm_next; | |
248 | ||
a8fb5618 | 249 | might_sleep(); |
1da177e4 LT |
250 | if (vma->vm_ops && vma->vm_ops->close) |
251 | vma->vm_ops->close(vma); | |
e9714acf | 252 | if (vma->vm_file) |
a8fb5618 | 253 | fput(vma->vm_file); |
f0be3d32 | 254 | mpol_put(vma_policy(vma)); |
1da177e4 | 255 | kmem_cache_free(vm_area_cachep, vma); |
a8fb5618 | 256 | return next; |
1da177e4 LT |
257 | } |
258 | ||
e4eb1ff6 LT |
259 | static unsigned long do_brk(unsigned long addr, unsigned long len); |
260 | ||
6a6160a7 | 261 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 LT |
262 | { |
263 | unsigned long rlim, retval; | |
264 | unsigned long newbrk, oldbrk; | |
265 | struct mm_struct *mm = current->mm; | |
1ad9a25d | 266 | struct vm_area_struct *next; |
a5b4592c | 267 | unsigned long min_brk; |
128557ff | 268 | bool populate; |
1da177e4 LT |
269 | |
270 | down_write(&mm->mmap_sem); | |
271 | ||
a5b4592c | 272 | #ifdef CONFIG_COMPAT_BRK |
5520e894 JK |
273 | /* |
274 | * CONFIG_COMPAT_BRK can still be overridden by setting | |
275 | * randomize_va_space to 2, which will still cause mm->start_brk | |
276 | * to be arbitrarily shifted | |
277 | */ | |
4471a675 | 278 | if (current->brk_randomized) |
5520e894 JK |
279 | min_brk = mm->start_brk; |
280 | else | |
281 | min_brk = mm->end_data; | |
a5b4592c JK |
282 | #else |
283 | min_brk = mm->start_brk; | |
284 | #endif | |
285 | if (brk < min_brk) | |
1da177e4 | 286 | goto out; |
1e624196 RG |
287 | |
288 | /* | |
289 | * Check against rlimit here. If this check is done later after the test | |
290 | * of oldbrk with newbrk then it can escape the test and let the data | |
291 | * segment grow beyond its set limit the in case where the limit is | |
292 | * not page aligned -Ram Gupta | |
293 | */ | |
59e99e5b | 294 | rlim = rlimit(RLIMIT_DATA); |
c1d171a0 JK |
295 | if (rlim < RLIM_INFINITY && (brk - mm->start_brk) + |
296 | (mm->end_data - mm->start_data) > rlim) | |
1e624196 RG |
297 | goto out; |
298 | ||
1da177e4 LT |
299 | newbrk = PAGE_ALIGN(brk); |
300 | oldbrk = PAGE_ALIGN(mm->brk); | |
301 | if (oldbrk == newbrk) | |
302 | goto set_brk; | |
303 | ||
304 | /* Always allow shrinking brk. */ | |
305 | if (brk <= mm->brk) { | |
306 | if (!do_munmap(mm, newbrk, oldbrk-newbrk)) | |
307 | goto set_brk; | |
308 | goto out; | |
309 | } | |
310 | ||
1da177e4 | 311 | /* Check against existing mmap mappings. */ |
1ad9a25d HD |
312 | next = find_vma(mm, oldbrk); |
313 | if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) | |
1da177e4 LT |
314 | goto out; |
315 | ||
316 | /* Ok, looks good - let it rip. */ | |
317 | if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) | |
318 | goto out; | |
128557ff | 319 | |
1da177e4 LT |
320 | set_brk: |
321 | mm->brk = brk; | |
128557ff ML |
322 | populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; |
323 | up_write(&mm->mmap_sem); | |
324 | if (populate) | |
325 | mm_populate(oldbrk, newbrk - oldbrk); | |
326 | return brk; | |
327 | ||
1da177e4 LT |
328 | out: |
329 | retval = mm->brk; | |
330 | up_write(&mm->mmap_sem); | |
331 | return retval; | |
332 | } | |
333 | ||
d3737187 ML |
334 | static long vma_compute_subtree_gap(struct vm_area_struct *vma) |
335 | { | |
1ad9a25d HD |
336 | unsigned long max, prev_end, subtree_gap; |
337 | ||
338 | /* | |
339 | * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we | |
340 | * allow two stack_guard_gaps between them here, and when choosing | |
341 | * an unmapped area; whereas when expanding we only require one. | |
342 | * That's a little inconsistent, but keeps the code here simpler. | |
343 | */ | |
344 | max = vm_start_gap(vma); | |
345 | if (vma->vm_prev) { | |
346 | prev_end = vm_end_gap(vma->vm_prev); | |
347 | if (max > prev_end) | |
348 | max -= prev_end; | |
349 | else | |
350 | max = 0; | |
351 | } | |
d3737187 ML |
352 | if (vma->vm_rb.rb_left) { |
353 | subtree_gap = rb_entry(vma->vm_rb.rb_left, | |
354 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
355 | if (subtree_gap > max) | |
356 | max = subtree_gap; | |
357 | } | |
358 | if (vma->vm_rb.rb_right) { | |
359 | subtree_gap = rb_entry(vma->vm_rb.rb_right, | |
360 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
361 | if (subtree_gap > max) | |
362 | max = subtree_gap; | |
363 | } | |
364 | return max; | |
365 | } | |
366 | ||
ed8ea815 | 367 | #ifdef CONFIG_DEBUG_VM_RB |
1da177e4 LT |
368 | static int browse_rb(struct rb_root *root) |
369 | { | |
5a0768f6 | 370 | int i = 0, j, bug = 0; |
1da177e4 LT |
371 | struct rb_node *nd, *pn = NULL; |
372 | unsigned long prev = 0, pend = 0; | |
373 | ||
374 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
375 | struct vm_area_struct *vma; | |
376 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
5a0768f6 ML |
377 | if (vma->vm_start < prev) { |
378 | printk("vm_start %lx prev %lx\n", vma->vm_start, prev); | |
379 | bug = 1; | |
380 | } | |
381 | if (vma->vm_start < pend) { | |
1da177e4 | 382 | printk("vm_start %lx pend %lx\n", vma->vm_start, pend); |
5a0768f6 ML |
383 | bug = 1; |
384 | } | |
385 | if (vma->vm_start > vma->vm_end) { | |
386 | printk("vm_end %lx < vm_start %lx\n", | |
387 | vma->vm_end, vma->vm_start); | |
388 | bug = 1; | |
389 | } | |
390 | if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { | |
391 | printk("free gap %lx, correct %lx\n", | |
392 | vma->rb_subtree_gap, | |
393 | vma_compute_subtree_gap(vma)); | |
394 | bug = 1; | |
395 | } | |
1da177e4 LT |
396 | i++; |
397 | pn = nd; | |
d1af65d1 DM |
398 | prev = vma->vm_start; |
399 | pend = vma->vm_end; | |
1da177e4 LT |
400 | } |
401 | j = 0; | |
5a0768f6 | 402 | for (nd = pn; nd; nd = rb_prev(nd)) |
1da177e4 | 403 | j++; |
5a0768f6 ML |
404 | if (i != j) { |
405 | printk("backwards %d, forwards %d\n", j, i); | |
406 | bug = 1; | |
1da177e4 | 407 | } |
5a0768f6 | 408 | return bug ? -1 : i; |
1da177e4 LT |
409 | } |
410 | ||
d3737187 ML |
411 | static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) |
412 | { | |
413 | struct rb_node *nd; | |
414 | ||
415 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
416 | struct vm_area_struct *vma; | |
417 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
418 | BUG_ON(vma != ignore && | |
419 | vma->rb_subtree_gap != vma_compute_subtree_gap(vma)); | |
1da177e4 | 420 | } |
1da177e4 LT |
421 | } |
422 | ||
423 | void validate_mm(struct mm_struct *mm) | |
424 | { | |
425 | int bug = 0; | |
426 | int i = 0; | |
5a0768f6 | 427 | unsigned long highest_address = 0; |
ed8ea815 ML |
428 | struct vm_area_struct *vma = mm->mmap; |
429 | while (vma) { | |
430 | struct anon_vma_chain *avc; | |
63c3b902 | 431 | vma_lock_anon_vma(vma); |
ed8ea815 ML |
432 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
433 | anon_vma_interval_tree_verify(avc); | |
63c3b902 | 434 | vma_unlock_anon_vma(vma); |
1ad9a25d | 435 | highest_address = vm_end_gap(vma); |
ed8ea815 | 436 | vma = vma->vm_next; |
1da177e4 LT |
437 | i++; |
438 | } | |
5a0768f6 ML |
439 | if (i != mm->map_count) { |
440 | printk("map_count %d vm_next %d\n", mm->map_count, i); | |
441 | bug = 1; | |
442 | } | |
443 | if (highest_address != mm->highest_vm_end) { | |
444 | printk("mm->highest_vm_end %lx, found %lx\n", | |
445 | mm->highest_vm_end, highest_address); | |
446 | bug = 1; | |
447 | } | |
1da177e4 | 448 | i = browse_rb(&mm->mm_rb); |
5a0768f6 ML |
449 | if (i != mm->map_count) { |
450 | printk("map_count %d rb %d\n", mm->map_count, i); | |
451 | bug = 1; | |
452 | } | |
46a350ef | 453 | BUG_ON(bug); |
1da177e4 LT |
454 | } |
455 | #else | |
d3737187 | 456 | #define validate_mm_rb(root, ignore) do { } while (0) |
1da177e4 LT |
457 | #define validate_mm(mm) do { } while (0) |
458 | #endif | |
459 | ||
d3737187 ML |
460 | RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, |
461 | unsigned long, rb_subtree_gap, vma_compute_subtree_gap) | |
462 | ||
463 | /* | |
464 | * Update augmented rbtree rb_subtree_gap values after vma->vm_start or | |
465 | * vma->vm_prev->vm_end values changed, without modifying the vma's position | |
466 | * in the rbtree. | |
467 | */ | |
468 | static void vma_gap_update(struct vm_area_struct *vma) | |
469 | { | |
470 | /* | |
471 | * As it turns out, RB_DECLARE_CALLBACKS() already created a callback | |
472 | * function that does exacltly what we want. | |
473 | */ | |
474 | vma_gap_callbacks_propagate(&vma->vm_rb, NULL); | |
475 | } | |
476 | ||
477 | static inline void vma_rb_insert(struct vm_area_struct *vma, | |
478 | struct rb_root *root) | |
479 | { | |
480 | /* All rb_subtree_gap values must be consistent prior to insertion */ | |
481 | validate_mm_rb(root, NULL); | |
482 | ||
483 | rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
484 | } | |
485 | ||
486 | static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) | |
487 | { | |
488 | /* | |
489 | * All rb_subtree_gap values must be consistent prior to erase, | |
490 | * with the possible exception of the vma being erased. | |
491 | */ | |
492 | validate_mm_rb(root, vma); | |
493 | ||
494 | /* | |
495 | * Note rb_erase_augmented is a fairly large inline function, | |
496 | * so make sure we instantiate it only once with our desired | |
497 | * augmented rbtree callbacks. | |
498 | */ | |
499 | rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
500 | } | |
501 | ||
bf181b9f ML |
502 | /* |
503 | * vma has some anon_vma assigned, and is already inserted on that | |
504 | * anon_vma's interval trees. | |
505 | * | |
506 | * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the | |
507 | * vma must be removed from the anon_vma's interval trees using | |
508 | * anon_vma_interval_tree_pre_update_vma(). | |
509 | * | |
510 | * After the update, the vma will be reinserted using | |
511 | * anon_vma_interval_tree_post_update_vma(). | |
512 | * | |
513 | * The entire update must be protected by exclusive mmap_sem and by | |
514 | * the root anon_vma's mutex. | |
515 | */ | |
516 | static inline void | |
517 | anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) | |
518 | { | |
519 | struct anon_vma_chain *avc; | |
520 | ||
521 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
522 | anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); | |
523 | } | |
524 | ||
525 | static inline void | |
526 | anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) | |
527 | { | |
528 | struct anon_vma_chain *avc; | |
529 | ||
530 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
531 | anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); | |
532 | } | |
533 | ||
6597d783 HD |
534 | static int find_vma_links(struct mm_struct *mm, unsigned long addr, |
535 | unsigned long end, struct vm_area_struct **pprev, | |
536 | struct rb_node ***rb_link, struct rb_node **rb_parent) | |
1da177e4 | 537 | { |
6597d783 | 538 | struct rb_node **__rb_link, *__rb_parent, *rb_prev; |
1da177e4 LT |
539 | |
540 | __rb_link = &mm->mm_rb.rb_node; | |
541 | rb_prev = __rb_parent = NULL; | |
1da177e4 LT |
542 | |
543 | while (*__rb_link) { | |
544 | struct vm_area_struct *vma_tmp; | |
545 | ||
546 | __rb_parent = *__rb_link; | |
547 | vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); | |
548 | ||
549 | if (vma_tmp->vm_end > addr) { | |
6597d783 HD |
550 | /* Fail if an existing vma overlaps the area */ |
551 | if (vma_tmp->vm_start < end) | |
552 | return -ENOMEM; | |
1da177e4 LT |
553 | __rb_link = &__rb_parent->rb_left; |
554 | } else { | |
555 | rb_prev = __rb_parent; | |
556 | __rb_link = &__rb_parent->rb_right; | |
557 | } | |
558 | } | |
559 | ||
560 | *pprev = NULL; | |
561 | if (rb_prev) | |
562 | *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); | |
563 | *rb_link = __rb_link; | |
564 | *rb_parent = __rb_parent; | |
6597d783 | 565 | return 0; |
1da177e4 LT |
566 | } |
567 | ||
e8420a8e CH |
568 | static unsigned long count_vma_pages_range(struct mm_struct *mm, |
569 | unsigned long addr, unsigned long end) | |
570 | { | |
571 | unsigned long nr_pages = 0; | |
572 | struct vm_area_struct *vma; | |
573 | ||
574 | /* Find first overlaping mapping */ | |
575 | vma = find_vma_intersection(mm, addr, end); | |
576 | if (!vma) | |
577 | return 0; | |
578 | ||
579 | nr_pages = (min(end, vma->vm_end) - | |
580 | max(addr, vma->vm_start)) >> PAGE_SHIFT; | |
581 | ||
582 | /* Iterate over the rest of the overlaps */ | |
583 | for (vma = vma->vm_next; vma; vma = vma->vm_next) { | |
584 | unsigned long overlap_len; | |
585 | ||
586 | if (vma->vm_start > end) | |
587 | break; | |
588 | ||
589 | overlap_len = min(end, vma->vm_end) - vma->vm_start; | |
590 | nr_pages += overlap_len >> PAGE_SHIFT; | |
591 | } | |
592 | ||
593 | return nr_pages; | |
594 | } | |
595 | ||
1da177e4 LT |
596 | void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, |
597 | struct rb_node **rb_link, struct rb_node *rb_parent) | |
598 | { | |
d3737187 ML |
599 | /* Update tracking information for the gap following the new vma. */ |
600 | if (vma->vm_next) | |
601 | vma_gap_update(vma->vm_next); | |
602 | else | |
1ad9a25d | 603 | mm->highest_vm_end = vm_end_gap(vma); |
d3737187 ML |
604 | |
605 | /* | |
606 | * vma->vm_prev wasn't known when we followed the rbtree to find the | |
607 | * correct insertion point for that vma. As a result, we could not | |
608 | * update the vma vm_rb parents rb_subtree_gap values on the way down. | |
609 | * So, we first insert the vma with a zero rb_subtree_gap value | |
610 | * (to be consistent with what we did on the way down), and then | |
611 | * immediately update the gap to the correct value. Finally we | |
612 | * rebalance the rbtree after all augmented values have been set. | |
613 | */ | |
1da177e4 | 614 | rb_link_node(&vma->vm_rb, rb_parent, rb_link); |
d3737187 ML |
615 | vma->rb_subtree_gap = 0; |
616 | vma_gap_update(vma); | |
617 | vma_rb_insert(vma, &mm->mm_rb); | |
1da177e4 LT |
618 | } |
619 | ||
cb8f488c | 620 | static void __vma_link_file(struct vm_area_struct *vma) |
1da177e4 | 621 | { |
48aae425 | 622 | struct file *file; |
1da177e4 LT |
623 | |
624 | file = vma->vm_file; | |
625 | if (file) { | |
626 | struct address_space *mapping = file->f_mapping; | |
627 | ||
628 | if (vma->vm_flags & VM_DENYWRITE) | |
496ad9aa | 629 | atomic_dec(&file_inode(file)->i_writecount); |
1da177e4 LT |
630 | if (vma->vm_flags & VM_SHARED) |
631 | mapping->i_mmap_writable++; | |
632 | ||
633 | flush_dcache_mmap_lock(mapping); | |
634 | if (unlikely(vma->vm_flags & VM_NONLINEAR)) | |
635 | vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); | |
636 | else | |
6b2dbba8 | 637 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
1da177e4 LT |
638 | flush_dcache_mmap_unlock(mapping); |
639 | } | |
640 | } | |
641 | ||
642 | static void | |
643 | __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
644 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
645 | struct rb_node *rb_parent) | |
646 | { | |
647 | __vma_link_list(mm, vma, prev, rb_parent); | |
648 | __vma_link_rb(mm, vma, rb_link, rb_parent); | |
1da177e4 LT |
649 | } |
650 | ||
651 | static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
652 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
653 | struct rb_node *rb_parent) | |
654 | { | |
655 | struct address_space *mapping = NULL; | |
656 | ||
657 | if (vma->vm_file) | |
658 | mapping = vma->vm_file->f_mapping; | |
659 | ||
97a89413 | 660 | if (mapping) |
3d48ae45 | 661 | mutex_lock(&mapping->i_mmap_mutex); |
1da177e4 LT |
662 | |
663 | __vma_link(mm, vma, prev, rb_link, rb_parent); | |
664 | __vma_link_file(vma); | |
665 | ||
1da177e4 | 666 | if (mapping) |
3d48ae45 | 667 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 LT |
668 | |
669 | mm->map_count++; | |
670 | validate_mm(mm); | |
671 | } | |
672 | ||
673 | /* | |
88f6b4c3 | 674 | * Helper for vma_adjust() in the split_vma insert case: insert a vma into the |
6b2dbba8 | 675 | * mm's list and rbtree. It has already been inserted into the interval tree. |
1da177e4 | 676 | */ |
48aae425 | 677 | static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 678 | { |
6597d783 | 679 | struct vm_area_struct *prev; |
48aae425 | 680 | struct rb_node **rb_link, *rb_parent; |
1da177e4 | 681 | |
6597d783 HD |
682 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
683 | &prev, &rb_link, &rb_parent)) | |
684 | BUG(); | |
1da177e4 LT |
685 | __vma_link(mm, vma, prev, rb_link, rb_parent); |
686 | mm->map_count++; | |
687 | } | |
688 | ||
689 | static inline void | |
690 | __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, | |
691 | struct vm_area_struct *prev) | |
692 | { | |
d3737187 | 693 | struct vm_area_struct *next; |
297c5eee | 694 | |
d3737187 ML |
695 | vma_rb_erase(vma, &mm->mm_rb); |
696 | prev->vm_next = next = vma->vm_next; | |
297c5eee LT |
697 | if (next) |
698 | next->vm_prev = prev; | |
1da177e4 LT |
699 | if (mm->mmap_cache == vma) |
700 | mm->mmap_cache = prev; | |
701 | } | |
702 | ||
703 | /* | |
704 | * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that | |
705 | * is already present in an i_mmap tree without adjusting the tree. | |
706 | * The following helper function should be used when such adjustments | |
707 | * are necessary. The "insert" vma (if any) is to be inserted | |
708 | * before we drop the necessary locks. | |
709 | */ | |
5beb4930 | 710 | int vma_adjust(struct vm_area_struct *vma, unsigned long start, |
1da177e4 LT |
711 | unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) |
712 | { | |
713 | struct mm_struct *mm = vma->vm_mm; | |
714 | struct vm_area_struct *next = vma->vm_next; | |
715 | struct vm_area_struct *importer = NULL; | |
716 | struct address_space *mapping = NULL; | |
6b2dbba8 | 717 | struct rb_root *root = NULL; |
012f1800 | 718 | struct anon_vma *anon_vma = NULL; |
1da177e4 | 719 | struct file *file = vma->vm_file; |
d3737187 | 720 | bool start_changed = false, end_changed = false; |
1da177e4 LT |
721 | long adjust_next = 0; |
722 | int remove_next = 0; | |
723 | ||
724 | if (next && !insert) { | |
287d97ac LT |
725 | struct vm_area_struct *exporter = NULL; |
726 | ||
1da177e4 LT |
727 | if (end >= next->vm_end) { |
728 | /* | |
729 | * vma expands, overlapping all the next, and | |
730 | * perhaps the one after too (mprotect case 6). | |
731 | */ | |
732 | again: remove_next = 1 + (end > next->vm_end); | |
733 | end = next->vm_end; | |
287d97ac | 734 | exporter = next; |
1da177e4 LT |
735 | importer = vma; |
736 | } else if (end > next->vm_start) { | |
737 | /* | |
738 | * vma expands, overlapping part of the next: | |
739 | * mprotect case 5 shifting the boundary up. | |
740 | */ | |
741 | adjust_next = (end - next->vm_start) >> PAGE_SHIFT; | |
287d97ac | 742 | exporter = next; |
1da177e4 LT |
743 | importer = vma; |
744 | } else if (end < vma->vm_end) { | |
745 | /* | |
746 | * vma shrinks, and !insert tells it's not | |
747 | * split_vma inserting another: so it must be | |
748 | * mprotect case 4 shifting the boundary down. | |
749 | */ | |
750 | adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT); | |
287d97ac | 751 | exporter = vma; |
1da177e4 LT |
752 | importer = next; |
753 | } | |
1da177e4 | 754 | |
5beb4930 RR |
755 | /* |
756 | * Easily overlooked: when mprotect shifts the boundary, | |
757 | * make sure the expanding vma has anon_vma set if the | |
758 | * shrinking vma had, to cover any anon pages imported. | |
759 | */ | |
287d97ac LT |
760 | if (exporter && exporter->anon_vma && !importer->anon_vma) { |
761 | if (anon_vma_clone(importer, exporter)) | |
5beb4930 | 762 | return -ENOMEM; |
287d97ac | 763 | importer->anon_vma = exporter->anon_vma; |
5beb4930 RR |
764 | } |
765 | } | |
766 | ||
1da177e4 LT |
767 | if (file) { |
768 | mapping = file->f_mapping; | |
682968e0 | 769 | if (!(vma->vm_flags & VM_NONLINEAR)) { |
1da177e4 | 770 | root = &mapping->i_mmap; |
cbc91f71 | 771 | uprobe_munmap(vma, vma->vm_start, vma->vm_end); |
682968e0 SD |
772 | |
773 | if (adjust_next) | |
cbc91f71 SD |
774 | uprobe_munmap(next, next->vm_start, |
775 | next->vm_end); | |
682968e0 SD |
776 | } |
777 | ||
3d48ae45 | 778 | mutex_lock(&mapping->i_mmap_mutex); |
1da177e4 | 779 | if (insert) { |
1da177e4 | 780 | /* |
6b2dbba8 | 781 | * Put into interval tree now, so instantiated pages |
1da177e4 LT |
782 | * are visible to arm/parisc __flush_dcache_page |
783 | * throughout; but we cannot insert into address | |
784 | * space until vma start or end is updated. | |
785 | */ | |
786 | __vma_link_file(insert); | |
787 | } | |
788 | } | |
789 | ||
94fcc585 AA |
790 | vma_adjust_trans_huge(vma, start, end, adjust_next); |
791 | ||
bf181b9f ML |
792 | anon_vma = vma->anon_vma; |
793 | if (!anon_vma && adjust_next) | |
794 | anon_vma = next->anon_vma; | |
795 | if (anon_vma) { | |
ca42b26a ML |
796 | VM_BUG_ON(adjust_next && next->anon_vma && |
797 | anon_vma != next->anon_vma); | |
4fc3f1d6 | 798 | anon_vma_lock_write(anon_vma); |
bf181b9f ML |
799 | anon_vma_interval_tree_pre_update_vma(vma); |
800 | if (adjust_next) | |
801 | anon_vma_interval_tree_pre_update_vma(next); | |
802 | } | |
012f1800 | 803 | |
1da177e4 LT |
804 | if (root) { |
805 | flush_dcache_mmap_lock(mapping); | |
6b2dbba8 | 806 | vma_interval_tree_remove(vma, root); |
1da177e4 | 807 | if (adjust_next) |
6b2dbba8 | 808 | vma_interval_tree_remove(next, root); |
1da177e4 LT |
809 | } |
810 | ||
d3737187 ML |
811 | if (start != vma->vm_start) { |
812 | vma->vm_start = start; | |
813 | start_changed = true; | |
814 | } | |
815 | if (end != vma->vm_end) { | |
816 | vma->vm_end = end; | |
817 | end_changed = true; | |
818 | } | |
1da177e4 LT |
819 | vma->vm_pgoff = pgoff; |
820 | if (adjust_next) { | |
821 | next->vm_start += adjust_next << PAGE_SHIFT; | |
822 | next->vm_pgoff += adjust_next; | |
823 | } | |
824 | ||
825 | if (root) { | |
826 | if (adjust_next) | |
6b2dbba8 ML |
827 | vma_interval_tree_insert(next, root); |
828 | vma_interval_tree_insert(vma, root); | |
1da177e4 LT |
829 | flush_dcache_mmap_unlock(mapping); |
830 | } | |
831 | ||
832 | if (remove_next) { | |
833 | /* | |
834 | * vma_merge has merged next into vma, and needs | |
835 | * us to remove next before dropping the locks. | |
836 | */ | |
837 | __vma_unlink(mm, next, vma); | |
838 | if (file) | |
839 | __remove_shared_vm_struct(next, file, mapping); | |
1da177e4 LT |
840 | } else if (insert) { |
841 | /* | |
842 | * split_vma has split insert from vma, and needs | |
843 | * us to insert it before dropping the locks | |
844 | * (it may either follow vma or precede it). | |
845 | */ | |
846 | __insert_vm_struct(mm, insert); | |
d3737187 ML |
847 | } else { |
848 | if (start_changed) | |
849 | vma_gap_update(vma); | |
850 | if (end_changed) { | |
851 | if (!next) | |
1ad9a25d | 852 | mm->highest_vm_end = vm_end_gap(vma); |
d3737187 ML |
853 | else if (!adjust_next) |
854 | vma_gap_update(next); | |
855 | } | |
1da177e4 LT |
856 | } |
857 | ||
bf181b9f ML |
858 | if (anon_vma) { |
859 | anon_vma_interval_tree_post_update_vma(vma); | |
860 | if (adjust_next) | |
861 | anon_vma_interval_tree_post_update_vma(next); | |
08b52706 | 862 | anon_vma_unlock_write(anon_vma); |
bf181b9f | 863 | } |
1da177e4 | 864 | if (mapping) |
3d48ae45 | 865 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 | 866 | |
2b144498 | 867 | if (root) { |
7b2d81d4 | 868 | uprobe_mmap(vma); |
2b144498 SD |
869 | |
870 | if (adjust_next) | |
7b2d81d4 | 871 | uprobe_mmap(next); |
2b144498 SD |
872 | } |
873 | ||
1da177e4 | 874 | if (remove_next) { |
925d1c40 | 875 | if (file) { |
cbc91f71 | 876 | uprobe_munmap(next, next->vm_start, next->vm_end); |
1da177e4 | 877 | fput(file); |
925d1c40 | 878 | } |
5beb4930 RR |
879 | if (next->anon_vma) |
880 | anon_vma_merge(vma, next); | |
1da177e4 | 881 | mm->map_count--; |
0bd6f78c | 882 | mpol_put(vma_policy(next)); |
1da177e4 LT |
883 | kmem_cache_free(vm_area_cachep, next); |
884 | /* | |
885 | * In mprotect's case 6 (see comments on vma_merge), | |
886 | * we must remove another next too. It would clutter | |
887 | * up the code too much to do both in one go. | |
888 | */ | |
d3737187 ML |
889 | next = vma->vm_next; |
890 | if (remove_next == 2) | |
1da177e4 | 891 | goto again; |
d3737187 ML |
892 | else if (next) |
893 | vma_gap_update(next); | |
894 | else | |
1ad9a25d | 895 | WARN_ON(mm->highest_vm_end != vm_end_gap(vma)); |
1da177e4 | 896 | } |
2b144498 | 897 | if (insert && file) |
7b2d81d4 | 898 | uprobe_mmap(insert); |
1da177e4 LT |
899 | |
900 | validate_mm(mm); | |
5beb4930 RR |
901 | |
902 | return 0; | |
1da177e4 LT |
903 | } |
904 | ||
905 | /* | |
906 | * If the vma has a ->close operation then the driver probably needs to release | |
907 | * per-vma resources, so we don't attempt to merge those. | |
908 | */ | |
1da177e4 LT |
909 | static inline int is_mergeable_vma(struct vm_area_struct *vma, |
910 | struct file *file, unsigned long vm_flags) | |
911 | { | |
0b173bc4 | 912 | if (vma->vm_flags ^ vm_flags) |
1da177e4 LT |
913 | return 0; |
914 | if (vma->vm_file != file) | |
915 | return 0; | |
916 | if (vma->vm_ops && vma->vm_ops->close) | |
917 | return 0; | |
918 | return 1; | |
919 | } | |
920 | ||
921 | static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, | |
965f55de SL |
922 | struct anon_vma *anon_vma2, |
923 | struct vm_area_struct *vma) | |
1da177e4 | 924 | { |
965f55de SL |
925 | /* |
926 | * The list_is_singular() test is to avoid merging VMA cloned from | |
927 | * parents. This can improve scalability caused by anon_vma lock. | |
928 | */ | |
929 | if ((!anon_vma1 || !anon_vma2) && (!vma || | |
930 | list_is_singular(&vma->anon_vma_chain))) | |
931 | return 1; | |
932 | return anon_vma1 == anon_vma2; | |
1da177e4 LT |
933 | } |
934 | ||
935 | /* | |
936 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
937 | * in front of (at a lower virtual address and file offset than) the vma. | |
938 | * | |
939 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
940 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
941 | * | |
942 | * We don't check here for the merged mmap wrapping around the end of pagecache | |
943 | * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which | |
944 | * wrap, nor mmaps which cover the final page at index -1UL. | |
945 | */ | |
946 | static int | |
947 | can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, | |
948 | struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) | |
949 | { | |
950 | if (is_mergeable_vma(vma, file, vm_flags) && | |
965f55de | 951 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 LT |
952 | if (vma->vm_pgoff == vm_pgoff) |
953 | return 1; | |
954 | } | |
955 | return 0; | |
956 | } | |
957 | ||
958 | /* | |
959 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
960 | * beyond (at a higher virtual address and file offset than) the vma. | |
961 | * | |
962 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
963 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
964 | */ | |
965 | static int | |
966 | can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, | |
967 | struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) | |
968 | { | |
969 | if (is_mergeable_vma(vma, file, vm_flags) && | |
965f55de | 970 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 LT |
971 | pgoff_t vm_pglen; |
972 | vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; | |
973 | if (vma->vm_pgoff + vm_pglen == vm_pgoff) | |
974 | return 1; | |
975 | } | |
976 | return 0; | |
977 | } | |
978 | ||
979 | /* | |
980 | * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out | |
981 | * whether that can be merged with its predecessor or its successor. | |
982 | * Or both (it neatly fills a hole). | |
983 | * | |
984 | * In most cases - when called for mmap, brk or mremap - [addr,end) is | |
985 | * certain not to be mapped by the time vma_merge is called; but when | |
986 | * called for mprotect, it is certain to be already mapped (either at | |
987 | * an offset within prev, or at the start of next), and the flags of | |
988 | * this area are about to be changed to vm_flags - and the no-change | |
989 | * case has already been eliminated. | |
990 | * | |
991 | * The following mprotect cases have to be considered, where AAAA is | |
992 | * the area passed down from mprotect_fixup, never extending beyond one | |
993 | * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: | |
994 | * | |
995 | * AAAA AAAA AAAA AAAA | |
996 | * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX | |
997 | * cannot merge might become might become might become | |
998 | * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or | |
999 | * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or | |
1000 | * mremap move: PPPPNNNNNNNN 8 | |
1001 | * AAAA | |
1002 | * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN | |
1003 | * might become case 1 below case 2 below case 3 below | |
1004 | * | |
1005 | * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: | |
1006 | * mprotect_fixup updates vm_flags & vm_page_prot on successful return. | |
1007 | */ | |
1008 | struct vm_area_struct *vma_merge(struct mm_struct *mm, | |
1009 | struct vm_area_struct *prev, unsigned long addr, | |
1010 | unsigned long end, unsigned long vm_flags, | |
1011 | struct anon_vma *anon_vma, struct file *file, | |
1012 | pgoff_t pgoff, struct mempolicy *policy) | |
1013 | { | |
1014 | pgoff_t pglen = (end - addr) >> PAGE_SHIFT; | |
1015 | struct vm_area_struct *area, *next; | |
5beb4930 | 1016 | int err; |
1da177e4 LT |
1017 | |
1018 | /* | |
1019 | * We later require that vma->vm_flags == vm_flags, | |
1020 | * so this tests vma->vm_flags & VM_SPECIAL, too. | |
1021 | */ | |
1022 | if (vm_flags & VM_SPECIAL) | |
1023 | return NULL; | |
1024 | ||
1025 | if (prev) | |
1026 | next = prev->vm_next; | |
1027 | else | |
1028 | next = mm->mmap; | |
1029 | area = next; | |
1030 | if (next && next->vm_end == end) /* cases 6, 7, 8 */ | |
1031 | next = next->vm_next; | |
1032 | ||
1033 | /* | |
1034 | * Can it merge with the predecessor? | |
1035 | */ | |
1036 | if (prev && prev->vm_end == addr && | |
1037 | mpol_equal(vma_policy(prev), policy) && | |
1038 | can_vma_merge_after(prev, vm_flags, | |
1039 | anon_vma, file, pgoff)) { | |
1040 | /* | |
1041 | * OK, it can. Can we now merge in the successor as well? | |
1042 | */ | |
1043 | if (next && end == next->vm_start && | |
1044 | mpol_equal(policy, vma_policy(next)) && | |
1045 | can_vma_merge_before(next, vm_flags, | |
1046 | anon_vma, file, pgoff+pglen) && | |
1047 | is_mergeable_anon_vma(prev->anon_vma, | |
965f55de | 1048 | next->anon_vma, NULL)) { |
1da177e4 | 1049 | /* cases 1, 6 */ |
5beb4930 | 1050 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 LT |
1051 | next->vm_end, prev->vm_pgoff, NULL); |
1052 | } else /* cases 2, 5, 7 */ | |
5beb4930 | 1053 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 | 1054 | end, prev->vm_pgoff, NULL); |
5beb4930 RR |
1055 | if (err) |
1056 | return NULL; | |
b15d00b6 | 1057 | khugepaged_enter_vma_merge(prev); |
1da177e4 LT |
1058 | return prev; |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * Can this new request be merged in front of next? | |
1063 | */ | |
1064 | if (next && end == next->vm_start && | |
1065 | mpol_equal(policy, vma_policy(next)) && | |
1066 | can_vma_merge_before(next, vm_flags, | |
1067 | anon_vma, file, pgoff+pglen)) { | |
1068 | if (prev && addr < prev->vm_end) /* case 4 */ | |
5beb4930 | 1069 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 LT |
1070 | addr, prev->vm_pgoff, NULL); |
1071 | else /* cases 3, 8 */ | |
5beb4930 | 1072 | err = vma_adjust(area, addr, next->vm_end, |
1da177e4 | 1073 | next->vm_pgoff - pglen, NULL); |
5beb4930 RR |
1074 | if (err) |
1075 | return NULL; | |
b15d00b6 | 1076 | khugepaged_enter_vma_merge(area); |
1da177e4 LT |
1077 | return area; |
1078 | } | |
1079 | ||
1080 | return NULL; | |
1081 | } | |
1082 | ||
d0e9fe17 LT |
1083 | /* |
1084 | * Rough compatbility check to quickly see if it's even worth looking | |
1085 | * at sharing an anon_vma. | |
1086 | * | |
1087 | * They need to have the same vm_file, and the flags can only differ | |
1088 | * in things that mprotect may change. | |
1089 | * | |
1090 | * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that | |
1091 | * we can merge the two vma's. For example, we refuse to merge a vma if | |
1092 | * there is a vm_ops->close() function, because that indicates that the | |
1093 | * driver is doing some kind of reference counting. But that doesn't | |
1094 | * really matter for the anon_vma sharing case. | |
1095 | */ | |
1096 | static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) | |
1097 | { | |
1098 | return a->vm_end == b->vm_start && | |
1099 | mpol_equal(vma_policy(a), vma_policy(b)) && | |
1100 | a->vm_file == b->vm_file && | |
1101 | !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) && | |
1102 | b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); | |
1103 | } | |
1104 | ||
1105 | /* | |
1106 | * Do some basic sanity checking to see if we can re-use the anon_vma | |
1107 | * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be | |
1108 | * the same as 'old', the other will be the new one that is trying | |
1109 | * to share the anon_vma. | |
1110 | * | |
1111 | * NOTE! This runs with mm_sem held for reading, so it is possible that | |
1112 | * the anon_vma of 'old' is concurrently in the process of being set up | |
1113 | * by another page fault trying to merge _that_. But that's ok: if it | |
1114 | * is being set up, that automatically means that it will be a singleton | |
1115 | * acceptable for merging, so we can do all of this optimistically. But | |
1116 | * we do that ACCESS_ONCE() to make sure that we never re-load the pointer. | |
1117 | * | |
1118 | * IOW: that the "list_is_singular()" test on the anon_vma_chain only | |
1119 | * matters for the 'stable anon_vma' case (ie the thing we want to avoid | |
1120 | * is to return an anon_vma that is "complex" due to having gone through | |
1121 | * a fork). | |
1122 | * | |
1123 | * We also make sure that the two vma's are compatible (adjacent, | |
1124 | * and with the same memory policies). That's all stable, even with just | |
1125 | * a read lock on the mm_sem. | |
1126 | */ | |
1127 | static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) | |
1128 | { | |
1129 | if (anon_vma_compatible(a, b)) { | |
1130 | struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma); | |
1131 | ||
1132 | if (anon_vma && list_is_singular(&old->anon_vma_chain)) | |
1133 | return anon_vma; | |
1134 | } | |
1135 | return NULL; | |
1136 | } | |
1137 | ||
1da177e4 LT |
1138 | /* |
1139 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check | |
1140 | * neighbouring vmas for a suitable anon_vma, before it goes off | |
1141 | * to allocate a new anon_vma. It checks because a repetitive | |
1142 | * sequence of mprotects and faults may otherwise lead to distinct | |
1143 | * anon_vmas being allocated, preventing vma merge in subsequent | |
1144 | * mprotect. | |
1145 | */ | |
1146 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) | |
1147 | { | |
d0e9fe17 | 1148 | struct anon_vma *anon_vma; |
1da177e4 | 1149 | struct vm_area_struct *near; |
1da177e4 LT |
1150 | |
1151 | near = vma->vm_next; | |
1152 | if (!near) | |
1153 | goto try_prev; | |
1154 | ||
d0e9fe17 LT |
1155 | anon_vma = reusable_anon_vma(near, vma, near); |
1156 | if (anon_vma) | |
1157 | return anon_vma; | |
1da177e4 | 1158 | try_prev: |
9be34c9d | 1159 | near = vma->vm_prev; |
1da177e4 LT |
1160 | if (!near) |
1161 | goto none; | |
1162 | ||
d0e9fe17 LT |
1163 | anon_vma = reusable_anon_vma(near, near, vma); |
1164 | if (anon_vma) | |
1165 | return anon_vma; | |
1da177e4 LT |
1166 | none: |
1167 | /* | |
1168 | * There's no absolute need to look only at touching neighbours: | |
1169 | * we could search further afield for "compatible" anon_vmas. | |
1170 | * But it would probably just be a waste of time searching, | |
1171 | * or lead to too many vmas hanging off the same anon_vma. | |
1172 | * We're trying to allow mprotect remerging later on, | |
1173 | * not trying to minimize memory used for anon_vmas. | |
1174 | */ | |
1175 | return NULL; | |
1176 | } | |
1177 | ||
1178 | #ifdef CONFIG_PROC_FS | |
ab50b8ed | 1179 | void vm_stat_account(struct mm_struct *mm, unsigned long flags, |
1da177e4 LT |
1180 | struct file *file, long pages) |
1181 | { | |
1182 | const unsigned long stack_flags | |
1183 | = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); | |
1184 | ||
44de9d0c HS |
1185 | mm->total_vm += pages; |
1186 | ||
1da177e4 LT |
1187 | if (file) { |
1188 | mm->shared_vm += pages; | |
1189 | if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) | |
1190 | mm->exec_vm += pages; | |
1191 | } else if (flags & stack_flags) | |
1192 | mm->stack_vm += pages; | |
1da177e4 LT |
1193 | } |
1194 | #endif /* CONFIG_PROC_FS */ | |
1195 | ||
40401530 AV |
1196 | /* |
1197 | * If a hint addr is less than mmap_min_addr change hint to be as | |
1198 | * low as possible but still greater than mmap_min_addr | |
1199 | */ | |
1200 | static inline unsigned long round_hint_to_min(unsigned long hint) | |
1201 | { | |
1202 | hint &= PAGE_MASK; | |
1203 | if (((void *)hint != NULL) && | |
1204 | (hint < mmap_min_addr)) | |
1205 | return PAGE_ALIGN(mmap_min_addr); | |
1206 | return hint; | |
1207 | } | |
1208 | ||
1da177e4 | 1209 | /* |
27f5de79 | 1210 | * The caller must hold down_write(¤t->mm->mmap_sem). |
1da177e4 LT |
1211 | */ |
1212 | ||
e3fc629d | 1213 | unsigned long do_mmap_pgoff(struct file *file, unsigned long addr, |
1da177e4 | 1214 | unsigned long len, unsigned long prot, |
bebeb3d6 | 1215 | unsigned long flags, unsigned long pgoff, |
41badc15 | 1216 | unsigned long *populate) |
1da177e4 LT |
1217 | { |
1218 | struct mm_struct * mm = current->mm; | |
1da177e4 | 1219 | struct inode *inode; |
ca16d140 | 1220 | vm_flags_t vm_flags; |
1da177e4 | 1221 | |
41badc15 | 1222 | *populate = 0; |
bebeb3d6 | 1223 | |
1da177e4 LT |
1224 | /* |
1225 | * Does the application expect PROT_READ to imply PROT_EXEC? | |
1226 | * | |
1227 | * (the exception is when the underlying filesystem is noexec | |
1228 | * mounted, in which case we dont add PROT_EXEC.) | |
1229 | */ | |
1230 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
d3ac7f89 | 1231 | if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC))) |
1da177e4 LT |
1232 | prot |= PROT_EXEC; |
1233 | ||
1234 | if (!len) | |
1235 | return -EINVAL; | |
1236 | ||
7cd94146 EP |
1237 | if (!(flags & MAP_FIXED)) |
1238 | addr = round_hint_to_min(addr); | |
1239 | ||
1da177e4 LT |
1240 | /* Careful about overflows.. */ |
1241 | len = PAGE_ALIGN(len); | |
9206de95 | 1242 | if (!len) |
1da177e4 LT |
1243 | return -ENOMEM; |
1244 | ||
1245 | /* offset overflow? */ | |
1246 | if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) | |
1247 | return -EOVERFLOW; | |
1248 | ||
1249 | /* Too many mappings? */ | |
1250 | if (mm->map_count > sysctl_max_map_count) | |
1251 | return -ENOMEM; | |
1252 | ||
1253 | /* Obtain the address to map to. we verify (or select) it and ensure | |
1254 | * that it represents a valid section of the address space. | |
1255 | */ | |
1256 | addr = get_unmapped_area(file, addr, len, pgoff, flags); | |
1257 | if (addr & ~PAGE_MASK) | |
1258 | return addr; | |
1259 | ||
1260 | /* Do simple checking here so the lower-level routines won't have | |
1261 | * to. we assume access permissions have been handled by the open | |
1262 | * of the memory object, so we don't do any here. | |
1263 | */ | |
1264 | vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | | |
1265 | mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; | |
1266 | ||
cdf7b341 | 1267 | if (flags & MAP_LOCKED) |
1da177e4 LT |
1268 | if (!can_do_mlock()) |
1269 | return -EPERM; | |
ba470de4 | 1270 | |
1da177e4 LT |
1271 | /* mlock MCL_FUTURE? */ |
1272 | if (vm_flags & VM_LOCKED) { | |
1273 | unsigned long locked, lock_limit; | |
93ea1d0a CW |
1274 | locked = len >> PAGE_SHIFT; |
1275 | locked += mm->locked_vm; | |
59e99e5b | 1276 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
93ea1d0a | 1277 | lock_limit >>= PAGE_SHIFT; |
1da177e4 LT |
1278 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) |
1279 | return -EAGAIN; | |
1280 | } | |
1281 | ||
496ad9aa | 1282 | inode = file ? file_inode(file) : NULL; |
1da177e4 LT |
1283 | |
1284 | if (file) { | |
1285 | switch (flags & MAP_TYPE) { | |
1286 | case MAP_SHARED: | |
1287 | if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) | |
1288 | return -EACCES; | |
1289 | ||
1290 | /* | |
1291 | * Make sure we don't allow writing to an append-only | |
1292 | * file.. | |
1293 | */ | |
1294 | if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) | |
1295 | return -EACCES; | |
1296 | ||
1297 | /* | |
1298 | * Make sure there are no mandatory locks on the file. | |
1299 | */ | |
1300 | if (locks_verify_locked(inode)) | |
1301 | return -EAGAIN; | |
1302 | ||
1303 | vm_flags |= VM_SHARED | VM_MAYSHARE; | |
1304 | if (!(file->f_mode & FMODE_WRITE)) | |
1305 | vm_flags &= ~(VM_MAYWRITE | VM_SHARED); | |
1306 | ||
1307 | /* fall through */ | |
1308 | case MAP_PRIVATE: | |
1309 | if (!(file->f_mode & FMODE_READ)) | |
1310 | return -EACCES; | |
d3ac7f89 | 1311 | if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { |
80c5606c LT |
1312 | if (vm_flags & VM_EXEC) |
1313 | return -EPERM; | |
1314 | vm_flags &= ~VM_MAYEXEC; | |
1315 | } | |
80c5606c LT |
1316 | |
1317 | if (!file->f_op || !file->f_op->mmap) | |
1318 | return -ENODEV; | |
1da177e4 LT |
1319 | break; |
1320 | ||
1321 | default: | |
1322 | return -EINVAL; | |
1323 | } | |
1324 | } else { | |
1325 | switch (flags & MAP_TYPE) { | |
1326 | case MAP_SHARED: | |
ce363942 TH |
1327 | /* |
1328 | * Ignore pgoff. | |
1329 | */ | |
1330 | pgoff = 0; | |
1da177e4 LT |
1331 | vm_flags |= VM_SHARED | VM_MAYSHARE; |
1332 | break; | |
1333 | case MAP_PRIVATE: | |
1334 | /* | |
1335 | * Set pgoff according to addr for anon_vma. | |
1336 | */ | |
1337 | pgoff = addr >> PAGE_SHIFT; | |
1338 | break; | |
1339 | default: | |
1340 | return -EINVAL; | |
1341 | } | |
1342 | } | |
1343 | ||
c22c0d63 ML |
1344 | /* |
1345 | * Set 'VM_NORESERVE' if we should not account for the | |
1346 | * memory use of this mapping. | |
1347 | */ | |
1348 | if (flags & MAP_NORESERVE) { | |
1349 | /* We honor MAP_NORESERVE if allowed to overcommit */ | |
1350 | if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) | |
1351 | vm_flags |= VM_NORESERVE; | |
1352 | ||
1353 | /* hugetlb applies strict overcommit unless MAP_NORESERVE */ | |
1354 | if (file && is_file_hugepages(file)) | |
1355 | vm_flags |= VM_NORESERVE; | |
1356 | } | |
1357 | ||
1358 | addr = mmap_region(file, addr, len, vm_flags, pgoff); | |
09a9f1d2 ML |
1359 | if (!IS_ERR_VALUE(addr) && |
1360 | ((vm_flags & VM_LOCKED) || | |
1361 | (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) | |
41badc15 | 1362 | *populate = len; |
bebeb3d6 | 1363 | return addr; |
0165ab44 | 1364 | } |
6be5ceb0 | 1365 | |
66f0dc48 HD |
1366 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1367 | unsigned long, prot, unsigned long, flags, | |
1368 | unsigned long, fd, unsigned long, pgoff) | |
1369 | { | |
1370 | struct file *file = NULL; | |
1371 | unsigned long retval = -EBADF; | |
1372 | ||
1373 | if (!(flags & MAP_ANONYMOUS)) { | |
120a795d | 1374 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1375 | if (unlikely(flags & MAP_HUGETLB)) |
1376 | return -EINVAL; | |
1377 | file = fget(fd); | |
1378 | if (!file) | |
1379 | goto out; | |
af73e4d9 NH |
1380 | if (is_file_hugepages(file)) |
1381 | len = ALIGN(len, huge_page_size(hstate_file(file))); | |
66f0dc48 HD |
1382 | } else if (flags & MAP_HUGETLB) { |
1383 | struct user_struct *user = NULL; | |
091d0d55 LZ |
1384 | struct hstate *hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & |
1385 | SHM_HUGE_MASK); | |
af73e4d9 | 1386 | |
091d0d55 LZ |
1387 | if (!hs) |
1388 | return -EINVAL; | |
1389 | ||
1390 | len = ALIGN(len, huge_page_size(hs)); | |
66f0dc48 HD |
1391 | /* |
1392 | * VM_NORESERVE is used because the reservations will be | |
1393 | * taken when vm_ops->mmap() is called | |
1394 | * A dummy user value is used because we are not locking | |
1395 | * memory so no accounting is necessary | |
1396 | */ | |
af73e4d9 | 1397 | file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, |
42d7395f AK |
1398 | VM_NORESERVE, |
1399 | &user, HUGETLB_ANONHUGE_INODE, | |
1400 | (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); | |
66f0dc48 HD |
1401 | if (IS_ERR(file)) |
1402 | return PTR_ERR(file); | |
1403 | } | |
1404 | ||
1405 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); | |
1406 | ||
eb36c587 | 1407 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
66f0dc48 HD |
1408 | if (file) |
1409 | fput(file); | |
1410 | out: | |
1411 | return retval; | |
1412 | } | |
1413 | ||
a4679373 CH |
1414 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1415 | struct mmap_arg_struct { | |
1416 | unsigned long addr; | |
1417 | unsigned long len; | |
1418 | unsigned long prot; | |
1419 | unsigned long flags; | |
1420 | unsigned long fd; | |
1421 | unsigned long offset; | |
1422 | }; | |
1423 | ||
1424 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1425 | { | |
1426 | struct mmap_arg_struct a; | |
1427 | ||
1428 | if (copy_from_user(&a, arg, sizeof(a))) | |
1429 | return -EFAULT; | |
1430 | if (a.offset & ~PAGE_MASK) | |
1431 | return -EINVAL; | |
1432 | ||
1433 | return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, | |
1434 | a.offset >> PAGE_SHIFT); | |
1435 | } | |
1436 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1437 | ||
4e950f6f AD |
1438 | /* |
1439 | * Some shared mappigns will want the pages marked read-only | |
1440 | * to track write events. If so, we'll downgrade vm_page_prot | |
1441 | * to the private version (using protection_map[] without the | |
1442 | * VM_SHARED bit). | |
1443 | */ | |
1444 | int vma_wants_writenotify(struct vm_area_struct *vma) | |
1445 | { | |
ca16d140 | 1446 | vm_flags_t vm_flags = vma->vm_flags; |
4e950f6f AD |
1447 | |
1448 | /* If it was private or non-writable, the write bit is already clear */ | |
1449 | if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) | |
1450 | return 0; | |
1451 | ||
1452 | /* The backer wishes to know when pages are first written to? */ | |
1453 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) | |
1454 | return 1; | |
1455 | ||
1456 | /* The open routine did something to the protections already? */ | |
1457 | if (pgprot_val(vma->vm_page_prot) != | |
3ed75eb8 | 1458 | pgprot_val(vm_get_page_prot(vm_flags))) |
4e950f6f AD |
1459 | return 0; |
1460 | ||
1461 | /* Specialty mapping? */ | |
4b6e1e37 | 1462 | if (vm_flags & VM_PFNMAP) |
4e950f6f AD |
1463 | return 0; |
1464 | ||
1465 | /* Can the mapping track the dirty pages? */ | |
1466 | return vma->vm_file && vma->vm_file->f_mapping && | |
1467 | mapping_cap_account_dirty(vma->vm_file->f_mapping); | |
1468 | } | |
1469 | ||
fc8744ad LT |
1470 | /* |
1471 | * We account for memory if it's a private writeable mapping, | |
5a6fe125 | 1472 | * not hugepages and VM_NORESERVE wasn't set. |
fc8744ad | 1473 | */ |
ca16d140 | 1474 | static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) |
fc8744ad | 1475 | { |
5a6fe125 MG |
1476 | /* |
1477 | * hugetlb has its own accounting separate from the core VM | |
1478 | * VM_HUGETLB may not be set yet so we cannot check for that flag. | |
1479 | */ | |
1480 | if (file && is_file_hugepages(file)) | |
1481 | return 0; | |
1482 | ||
fc8744ad LT |
1483 | return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; |
1484 | } | |
1485 | ||
0165ab44 | 1486 | unsigned long mmap_region(struct file *file, unsigned long addr, |
c22c0d63 | 1487 | unsigned long len, vm_flags_t vm_flags, unsigned long pgoff) |
0165ab44 MS |
1488 | { |
1489 | struct mm_struct *mm = current->mm; | |
1490 | struct vm_area_struct *vma, *prev; | |
1491 | int correct_wcount = 0; | |
1492 | int error; | |
1493 | struct rb_node **rb_link, *rb_parent; | |
1494 | unsigned long charged = 0; | |
496ad9aa | 1495 | struct inode *inode = file ? file_inode(file) : NULL; |
0165ab44 | 1496 | |
e8420a8e CH |
1497 | /* Check against address space limit. */ |
1498 | if (!may_expand_vm(mm, len >> PAGE_SHIFT)) { | |
1499 | unsigned long nr_pages; | |
1500 | ||
1501 | /* | |
1502 | * MAP_FIXED may remove pages of mappings that intersects with | |
1503 | * requested mapping. Account for the pages it would unmap. | |
1504 | */ | |
1505 | if (!(vm_flags & MAP_FIXED)) | |
1506 | return -ENOMEM; | |
1507 | ||
1508 | nr_pages = count_vma_pages_range(mm, addr, addr + len); | |
1509 | ||
1510 | if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages)) | |
1511 | return -ENOMEM; | |
1512 | } | |
1513 | ||
1da177e4 LT |
1514 | /* Clear old maps */ |
1515 | error = -ENOMEM; | |
1516 | munmap_back: | |
6597d783 | 1517 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { |
1da177e4 LT |
1518 | if (do_munmap(mm, addr, len)) |
1519 | return -ENOMEM; | |
1520 | goto munmap_back; | |
1521 | } | |
1522 | ||
fc8744ad LT |
1523 | /* |
1524 | * Private writable mapping: check memory availability | |
1525 | */ | |
5a6fe125 | 1526 | if (accountable_mapping(file, vm_flags)) { |
fc8744ad | 1527 | charged = len >> PAGE_SHIFT; |
191c5424 | 1528 | if (security_vm_enough_memory_mm(mm, charged)) |
fc8744ad LT |
1529 | return -ENOMEM; |
1530 | vm_flags |= VM_ACCOUNT; | |
1da177e4 LT |
1531 | } |
1532 | ||
1533 | /* | |
de33c8db | 1534 | * Can we just expand an old mapping? |
1da177e4 | 1535 | */ |
de33c8db LT |
1536 | vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL); |
1537 | if (vma) | |
1538 | goto out; | |
1da177e4 LT |
1539 | |
1540 | /* | |
1541 | * Determine the object being mapped and call the appropriate | |
1542 | * specific mapper. the address has already been validated, but | |
1543 | * not unmapped, but the maps are removed from the list. | |
1544 | */ | |
c5e3b83e | 1545 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
1546 | if (!vma) { |
1547 | error = -ENOMEM; | |
1548 | goto unacct_error; | |
1549 | } | |
1da177e4 LT |
1550 | |
1551 | vma->vm_mm = mm; | |
1552 | vma->vm_start = addr; | |
1553 | vma->vm_end = addr + len; | |
1554 | vma->vm_flags = vm_flags; | |
3ed75eb8 | 1555 | vma->vm_page_prot = vm_get_page_prot(vm_flags); |
1da177e4 | 1556 | vma->vm_pgoff = pgoff; |
5beb4930 | 1557 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 | 1558 | |
ce8fea7a HD |
1559 | error = -EINVAL; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */ |
1560 | ||
1da177e4 | 1561 | if (file) { |
1da177e4 LT |
1562 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1563 | goto free_vma; | |
1564 | if (vm_flags & VM_DENYWRITE) { | |
1565 | error = deny_write_access(file); | |
1566 | if (error) | |
1567 | goto free_vma; | |
1568 | correct_wcount = 1; | |
1569 | } | |
cb0942b8 | 1570 | vma->vm_file = get_file(file); |
1da177e4 LT |
1571 | error = file->f_op->mmap(file, vma); |
1572 | if (error) | |
1573 | goto unmap_and_free_vma; | |
f8dbf0a7 HS |
1574 | |
1575 | /* Can addr have changed?? | |
1576 | * | |
1577 | * Answer: Yes, several device drivers can do it in their | |
1578 | * f_op->mmap method. -DaveM | |
2897b4d2 JK |
1579 | * Bug: If addr is changed, prev, rb_link, rb_parent should |
1580 | * be updated for vma_link() | |
f8dbf0a7 | 1581 | */ |
2897b4d2 JK |
1582 | WARN_ON_ONCE(addr != vma->vm_start); |
1583 | ||
f8dbf0a7 HS |
1584 | addr = vma->vm_start; |
1585 | pgoff = vma->vm_pgoff; | |
1586 | vm_flags = vma->vm_flags; | |
1da177e4 | 1587 | } else if (vm_flags & VM_SHARED) { |
835ee797 AV |
1588 | if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP))) |
1589 | goto free_vma; | |
1da177e4 LT |
1590 | error = shmem_zero_setup(vma); |
1591 | if (error) | |
1592 | goto free_vma; | |
1593 | } | |
1594 | ||
c9d0bf24 MD |
1595 | if (vma_wants_writenotify(vma)) { |
1596 | pgprot_t pprot = vma->vm_page_prot; | |
1597 | ||
1598 | /* Can vma->vm_page_prot have changed?? | |
1599 | * | |
1600 | * Answer: Yes, drivers may have changed it in their | |
1601 | * f_op->mmap method. | |
1602 | * | |
1603 | * Ensures that vmas marked as uncached stay that way. | |
1604 | */ | |
1ddd439e | 1605 | vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED); |
c9d0bf24 MD |
1606 | if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot))) |
1607 | vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); | |
1608 | } | |
d08b3851 | 1609 | |
de33c8db LT |
1610 | vma_link(mm, vma, prev, rb_link, rb_parent); |
1611 | file = vma->vm_file; | |
4d3d5b41 ON |
1612 | |
1613 | /* Once vma denies write, undo our temporary denial count */ | |
1614 | if (correct_wcount) | |
1615 | atomic_inc(&inode->i_writecount); | |
1616 | out: | |
cdd6c482 | 1617 | perf_event_mmap(vma); |
0a4a9391 | 1618 | |
ab50b8ed | 1619 | vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); |
1da177e4 | 1620 | if (vm_flags & VM_LOCKED) { |
bebeb3d6 ML |
1621 | if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || |
1622 | vma == get_gate_vma(current->mm))) | |
06f9d8c2 | 1623 | mm->locked_vm += (len >> PAGE_SHIFT); |
bebeb3d6 ML |
1624 | else |
1625 | vma->vm_flags &= ~VM_LOCKED; | |
1626 | } | |
2b144498 | 1627 | |
c7a3a88c ON |
1628 | if (file) |
1629 | uprobe_mmap(vma); | |
2b144498 | 1630 | |
1da177e4 LT |
1631 | return addr; |
1632 | ||
1633 | unmap_and_free_vma: | |
1634 | if (correct_wcount) | |
1635 | atomic_inc(&inode->i_writecount); | |
1636 | vma->vm_file = NULL; | |
1637 | fput(file); | |
1638 | ||
1639 | /* Undo any partial mapping done by a device driver. */ | |
e0da382c HD |
1640 | unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); |
1641 | charged = 0; | |
1da177e4 LT |
1642 | free_vma: |
1643 | kmem_cache_free(vm_area_cachep, vma); | |
1644 | unacct_error: | |
1645 | if (charged) | |
1646 | vm_unacct_memory(charged); | |
1647 | return error; | |
1648 | } | |
1649 | ||
db4fbfb9 ML |
1650 | unsigned long unmapped_area(struct vm_unmapped_area_info *info) |
1651 | { | |
1652 | /* | |
1653 | * We implement the search by looking for an rbtree node that | |
1654 | * immediately follows a suitable gap. That is, | |
1655 | * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; | |
1656 | * - gap_end = vma->vm_start >= info->low_limit + length; | |
1657 | * - gap_end - gap_start >= length | |
1658 | */ | |
1659 | ||
1660 | struct mm_struct *mm = current->mm; | |
1661 | struct vm_area_struct *vma; | |
1662 | unsigned long length, low_limit, high_limit, gap_start, gap_end; | |
1663 | ||
1664 | /* Adjust search length to account for worst case alignment overhead */ | |
1665 | length = info->length + info->align_mask; | |
1666 | if (length < info->length) | |
1667 | return -ENOMEM; | |
1668 | ||
1669 | /* Adjust search limits by the desired length */ | |
1670 | if (info->high_limit < length) | |
1671 | return -ENOMEM; | |
1672 | high_limit = info->high_limit - length; | |
1673 | ||
1674 | if (info->low_limit > high_limit) | |
1675 | return -ENOMEM; | |
1676 | low_limit = info->low_limit + length; | |
1677 | ||
1678 | /* Check if rbtree root looks promising */ | |
1679 | if (RB_EMPTY_ROOT(&mm->mm_rb)) | |
1680 | goto check_highest; | |
1681 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); | |
1682 | if (vma->rb_subtree_gap < length) | |
1683 | goto check_highest; | |
1684 | ||
1685 | while (true) { | |
1686 | /* Visit left subtree if it looks promising */ | |
1ad9a25d | 1687 | gap_end = vm_start_gap(vma); |
db4fbfb9 ML |
1688 | if (gap_end >= low_limit && vma->vm_rb.rb_left) { |
1689 | struct vm_area_struct *left = | |
1690 | rb_entry(vma->vm_rb.rb_left, | |
1691 | struct vm_area_struct, vm_rb); | |
1692 | if (left->rb_subtree_gap >= length) { | |
1693 | vma = left; | |
1694 | continue; | |
1695 | } | |
1696 | } | |
1697 | ||
1ad9a25d | 1698 | gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; |
db4fbfb9 ML |
1699 | check_current: |
1700 | /* Check if current node has a suitable gap */ | |
1701 | if (gap_start > high_limit) | |
1702 | return -ENOMEM; | |
28ebf895 HD |
1703 | if (gap_end >= low_limit && |
1704 | gap_end > gap_start && gap_end - gap_start >= length) | |
db4fbfb9 ML |
1705 | goto found; |
1706 | ||
1707 | /* Visit right subtree if it looks promising */ | |
1708 | if (vma->vm_rb.rb_right) { | |
1709 | struct vm_area_struct *right = | |
1710 | rb_entry(vma->vm_rb.rb_right, | |
1711 | struct vm_area_struct, vm_rb); | |
1712 | if (right->rb_subtree_gap >= length) { | |
1713 | vma = right; | |
1714 | continue; | |
1715 | } | |
1716 | } | |
1717 | ||
1718 | /* Go back up the rbtree to find next candidate node */ | |
1719 | while (true) { | |
1720 | struct rb_node *prev = &vma->vm_rb; | |
1721 | if (!rb_parent(prev)) | |
1722 | goto check_highest; | |
1723 | vma = rb_entry(rb_parent(prev), | |
1724 | struct vm_area_struct, vm_rb); | |
1725 | if (prev == vma->vm_rb.rb_left) { | |
1ad9a25d HD |
1726 | gap_start = vm_end_gap(vma->vm_prev); |
1727 | gap_end = vm_start_gap(vma); | |
db4fbfb9 ML |
1728 | goto check_current; |
1729 | } | |
1730 | } | |
1731 | } | |
1732 | ||
1733 | check_highest: | |
1734 | /* Check highest gap, which does not precede any rbtree node */ | |
1735 | gap_start = mm->highest_vm_end; | |
1736 | gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ | |
1737 | if (gap_start > high_limit) | |
1738 | return -ENOMEM; | |
1739 | ||
1740 | found: | |
1741 | /* We found a suitable gap. Clip it with the original low_limit. */ | |
1742 | if (gap_start < info->low_limit) | |
1743 | gap_start = info->low_limit; | |
1744 | ||
1745 | /* Adjust gap address to the desired alignment */ | |
1746 | gap_start += (info->align_offset - gap_start) & info->align_mask; | |
1747 | ||
1748 | VM_BUG_ON(gap_start + info->length > info->high_limit); | |
1749 | VM_BUG_ON(gap_start + info->length > gap_end); | |
1750 | return gap_start; | |
1751 | } | |
1752 | ||
1753 | unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) | |
1754 | { | |
1755 | struct mm_struct *mm = current->mm; | |
1756 | struct vm_area_struct *vma; | |
1757 | unsigned long length, low_limit, high_limit, gap_start, gap_end; | |
1758 | ||
1759 | /* Adjust search length to account for worst case alignment overhead */ | |
1760 | length = info->length + info->align_mask; | |
1761 | if (length < info->length) | |
1762 | return -ENOMEM; | |
1763 | ||
1764 | /* | |
1765 | * Adjust search limits by the desired length. | |
1766 | * See implementation comment at top of unmapped_area(). | |
1767 | */ | |
1768 | gap_end = info->high_limit; | |
1769 | if (gap_end < length) | |
1770 | return -ENOMEM; | |
1771 | high_limit = gap_end - length; | |
1772 | ||
1773 | if (info->low_limit > high_limit) | |
1774 | return -ENOMEM; | |
1775 | low_limit = info->low_limit + length; | |
1776 | ||
1777 | /* Check highest gap, which does not precede any rbtree node */ | |
1778 | gap_start = mm->highest_vm_end; | |
1779 | if (gap_start <= high_limit) | |
1780 | goto found_highest; | |
1781 | ||
1782 | /* Check if rbtree root looks promising */ | |
1783 | if (RB_EMPTY_ROOT(&mm->mm_rb)) | |
1784 | return -ENOMEM; | |
1785 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); | |
1786 | if (vma->rb_subtree_gap < length) | |
1787 | return -ENOMEM; | |
1788 | ||
1789 | while (true) { | |
1790 | /* Visit right subtree if it looks promising */ | |
1ad9a25d | 1791 | gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; |
db4fbfb9 ML |
1792 | if (gap_start <= high_limit && vma->vm_rb.rb_right) { |
1793 | struct vm_area_struct *right = | |
1794 | rb_entry(vma->vm_rb.rb_right, | |
1795 | struct vm_area_struct, vm_rb); | |
1796 | if (right->rb_subtree_gap >= length) { | |
1797 | vma = right; | |
1798 | continue; | |
1799 | } | |
1800 | } | |
1801 | ||
1802 | check_current: | |
1803 | /* Check if current node has a suitable gap */ | |
1ad9a25d | 1804 | gap_end = vm_start_gap(vma); |
db4fbfb9 ML |
1805 | if (gap_end < low_limit) |
1806 | return -ENOMEM; | |
28ebf895 HD |
1807 | if (gap_start <= high_limit && |
1808 | gap_end > gap_start && gap_end - gap_start >= length) | |
db4fbfb9 ML |
1809 | goto found; |
1810 | ||
1811 | /* Visit left subtree if it looks promising */ | |
1812 | if (vma->vm_rb.rb_left) { | |
1813 | struct vm_area_struct *left = | |
1814 | rb_entry(vma->vm_rb.rb_left, | |
1815 | struct vm_area_struct, vm_rb); | |
1816 | if (left->rb_subtree_gap >= length) { | |
1817 | vma = left; | |
1818 | continue; | |
1819 | } | |
1820 | } | |
1821 | ||
1822 | /* Go back up the rbtree to find next candidate node */ | |
1823 | while (true) { | |
1824 | struct rb_node *prev = &vma->vm_rb; | |
1825 | if (!rb_parent(prev)) | |
1826 | return -ENOMEM; | |
1827 | vma = rb_entry(rb_parent(prev), | |
1828 | struct vm_area_struct, vm_rb); | |
1829 | if (prev == vma->vm_rb.rb_right) { | |
1830 | gap_start = vma->vm_prev ? | |
1ad9a25d | 1831 | vm_end_gap(vma->vm_prev) : 0; |
db4fbfb9 ML |
1832 | goto check_current; |
1833 | } | |
1834 | } | |
1835 | } | |
1836 | ||
1837 | found: | |
1838 | /* We found a suitable gap. Clip it with the original high_limit. */ | |
1839 | if (gap_end > info->high_limit) | |
1840 | gap_end = info->high_limit; | |
1841 | ||
1842 | found_highest: | |
1843 | /* Compute highest gap address at the desired alignment */ | |
1844 | gap_end -= info->length; | |
1845 | gap_end -= (gap_end - info->align_offset) & info->align_mask; | |
1846 | ||
1847 | VM_BUG_ON(gap_end < info->low_limit); | |
1848 | VM_BUG_ON(gap_end < gap_start); | |
1849 | return gap_end; | |
1850 | } | |
1851 | ||
1da177e4 LT |
1852 | /* Get an address range which is currently unmapped. |
1853 | * For shmat() with addr=0. | |
1854 | * | |
1855 | * Ugly calling convention alert: | |
1856 | * Return value with the low bits set means error value, | |
1857 | * ie | |
1858 | * if (ret & ~PAGE_MASK) | |
1859 | * error = ret; | |
1860 | * | |
1861 | * This function "knows" that -ENOMEM has the bits set. | |
1862 | */ | |
1863 | #ifndef HAVE_ARCH_UNMAPPED_AREA | |
1864 | unsigned long | |
1865 | arch_get_unmapped_area(struct file *filp, unsigned long addr, | |
1866 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1867 | { | |
1868 | struct mm_struct *mm = current->mm; | |
1ad9a25d | 1869 | struct vm_area_struct *vma, *prev; |
db4fbfb9 | 1870 | struct vm_unmapped_area_info info; |
1da177e4 | 1871 | |
3cbafaa7 | 1872 | if (len > TASK_SIZE - mmap_min_addr) |
1da177e4 LT |
1873 | return -ENOMEM; |
1874 | ||
06abdfb4 BH |
1875 | if (flags & MAP_FIXED) |
1876 | return addr; | |
1877 | ||
1da177e4 LT |
1878 | if (addr) { |
1879 | addr = PAGE_ALIGN(addr); | |
1ad9a25d | 1880 | vma = find_vma_prev(mm, addr, &prev); |
3cbafaa7 | 1881 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
1ad9a25d HD |
1882 | (!vma || addr + len <= vm_start_gap(vma)) && |
1883 | (!prev || addr >= vm_end_gap(prev))) | |
1da177e4 LT |
1884 | return addr; |
1885 | } | |
1da177e4 | 1886 | |
db4fbfb9 ML |
1887 | info.flags = 0; |
1888 | info.length = len; | |
1889 | info.low_limit = TASK_UNMAPPED_BASE; | |
1890 | info.high_limit = TASK_SIZE; | |
1891 | info.align_mask = 0; | |
1892 | return vm_unmapped_area(&info); | |
1da177e4 LT |
1893 | } |
1894 | #endif | |
1895 | ||
1363c3cd | 1896 | void arch_unmap_area(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
1897 | { |
1898 | /* | |
1899 | * Is this a new hole at the lowest possible address? | |
1900 | */ | |
f44d2198 | 1901 | if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) |
1363c3cd | 1902 | mm->free_area_cache = addr; |
1da177e4 LT |
1903 | } |
1904 | ||
1905 | /* | |
1906 | * This mmap-allocator allocates new areas top-down from below the | |
1907 | * stack's low limit (the base): | |
1908 | */ | |
1909 | #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN | |
1910 | unsigned long | |
1911 | arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, | |
1912 | const unsigned long len, const unsigned long pgoff, | |
1913 | const unsigned long flags) | |
1914 | { | |
1ad9a25d | 1915 | struct vm_area_struct *vma, *prev; |
1da177e4 | 1916 | struct mm_struct *mm = current->mm; |
db4fbfb9 ML |
1917 | unsigned long addr = addr0; |
1918 | struct vm_unmapped_area_info info; | |
1da177e4 LT |
1919 | |
1920 | /* requested length too big for entire address space */ | |
3cbafaa7 | 1921 | if (len > TASK_SIZE - mmap_min_addr) |
1da177e4 LT |
1922 | return -ENOMEM; |
1923 | ||
06abdfb4 BH |
1924 | if (flags & MAP_FIXED) |
1925 | return addr; | |
1926 | ||
1da177e4 LT |
1927 | /* requesting a specific address */ |
1928 | if (addr) { | |
1929 | addr = PAGE_ALIGN(addr); | |
1ad9a25d | 1930 | vma = find_vma_prev(mm, addr, &prev); |
3cbafaa7 | 1931 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
1ad9a25d HD |
1932 | (!vma || addr + len <= vm_start_gap(vma)) && |
1933 | (!prev || addr >= vm_end_gap(prev))) | |
1da177e4 LT |
1934 | return addr; |
1935 | } | |
1936 | ||
db4fbfb9 ML |
1937 | info.flags = VM_UNMAPPED_AREA_TOPDOWN; |
1938 | info.length = len; | |
3cbafaa7 | 1939 | info.low_limit = max(PAGE_SIZE, mmap_min_addr); |
db4fbfb9 ML |
1940 | info.high_limit = mm->mmap_base; |
1941 | info.align_mask = 0; | |
1942 | addr = vm_unmapped_area(&info); | |
b716ad95 | 1943 | |
1da177e4 LT |
1944 | /* |
1945 | * A failed mmap() very likely causes application failure, | |
1946 | * so fall back to the bottom-up function here. This scenario | |
1947 | * can happen with large stack limits and large mmap() | |
1948 | * allocations. | |
1949 | */ | |
db4fbfb9 ML |
1950 | if (addr & ~PAGE_MASK) { |
1951 | VM_BUG_ON(addr != -ENOMEM); | |
1952 | info.flags = 0; | |
1953 | info.low_limit = TASK_UNMAPPED_BASE; | |
1954 | info.high_limit = TASK_SIZE; | |
1955 | addr = vm_unmapped_area(&info); | |
1956 | } | |
1da177e4 LT |
1957 | |
1958 | return addr; | |
1959 | } | |
1960 | #endif | |
1961 | ||
1363c3cd | 1962 | void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
1963 | { |
1964 | /* | |
1965 | * Is this a new hole at the highest possible address? | |
1966 | */ | |
1363c3cd WW |
1967 | if (addr > mm->free_area_cache) |
1968 | mm->free_area_cache = addr; | |
1da177e4 LT |
1969 | |
1970 | /* dont allow allocations above current base */ | |
1363c3cd WW |
1971 | if (mm->free_area_cache > mm->mmap_base) |
1972 | mm->free_area_cache = mm->mmap_base; | |
1da177e4 LT |
1973 | } |
1974 | ||
1975 | unsigned long | |
1976 | get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, | |
1977 | unsigned long pgoff, unsigned long flags) | |
1978 | { | |
06abdfb4 BH |
1979 | unsigned long (*get_area)(struct file *, unsigned long, |
1980 | unsigned long, unsigned long, unsigned long); | |
1981 | ||
9206de95 AV |
1982 | unsigned long error = arch_mmap_check(addr, len, flags); |
1983 | if (error) | |
1984 | return error; | |
1985 | ||
1986 | /* Careful about overflows.. */ | |
1987 | if (len > TASK_SIZE) | |
1988 | return -ENOMEM; | |
1989 | ||
06abdfb4 BH |
1990 | get_area = current->mm->get_unmapped_area; |
1991 | if (file && file->f_op && file->f_op->get_unmapped_area) | |
1992 | get_area = file->f_op->get_unmapped_area; | |
1993 | addr = get_area(file, addr, len, pgoff, flags); | |
1994 | if (IS_ERR_VALUE(addr)) | |
1995 | return addr; | |
1da177e4 | 1996 | |
07ab67c8 LT |
1997 | if (addr > TASK_SIZE - len) |
1998 | return -ENOMEM; | |
1999 | if (addr & ~PAGE_MASK) | |
2000 | return -EINVAL; | |
06abdfb4 | 2001 | |
9ac4ed4b AV |
2002 | addr = arch_rebalance_pgtables(addr, len); |
2003 | error = security_mmap_addr(addr); | |
2004 | return error ? error : addr; | |
1da177e4 LT |
2005 | } |
2006 | ||
2007 | EXPORT_SYMBOL(get_unmapped_area); | |
2008 | ||
2009 | /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ | |
48aae425 | 2010 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
2011 | { |
2012 | struct vm_area_struct *vma = NULL; | |
2013 | ||
841e31e5 RM |
2014 | /* Check the cache first. */ |
2015 | /* (Cache hit rate is typically around 35%.) */ | |
b6a9b7f6 | 2016 | vma = ACCESS_ONCE(mm->mmap_cache); |
841e31e5 RM |
2017 | if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) { |
2018 | struct rb_node *rb_node; | |
2019 | ||
2020 | rb_node = mm->mm_rb.rb_node; | |
2021 | vma = NULL; | |
2022 | ||
2023 | while (rb_node) { | |
2024 | struct vm_area_struct *vma_tmp; | |
2025 | ||
2026 | vma_tmp = rb_entry(rb_node, | |
2027 | struct vm_area_struct, vm_rb); | |
2028 | ||
2029 | if (vma_tmp->vm_end > addr) { | |
2030 | vma = vma_tmp; | |
2031 | if (vma_tmp->vm_start <= addr) | |
2032 | break; | |
2033 | rb_node = rb_node->rb_left; | |
2034 | } else | |
2035 | rb_node = rb_node->rb_right; | |
1da177e4 | 2036 | } |
841e31e5 RM |
2037 | if (vma) |
2038 | mm->mmap_cache = vma; | |
1da177e4 LT |
2039 | } |
2040 | return vma; | |
2041 | } | |
2042 | ||
2043 | EXPORT_SYMBOL(find_vma); | |
2044 | ||
6bd4837d KM |
2045 | /* |
2046 | * Same as find_vma, but also return a pointer to the previous VMA in *pprev. | |
6bd4837d | 2047 | */ |
1da177e4 LT |
2048 | struct vm_area_struct * |
2049 | find_vma_prev(struct mm_struct *mm, unsigned long addr, | |
2050 | struct vm_area_struct **pprev) | |
2051 | { | |
6bd4837d | 2052 | struct vm_area_struct *vma; |
1da177e4 | 2053 | |
6bd4837d | 2054 | vma = find_vma(mm, addr); |
83cd904d MP |
2055 | if (vma) { |
2056 | *pprev = vma->vm_prev; | |
2057 | } else { | |
2058 | struct rb_node *rb_node = mm->mm_rb.rb_node; | |
2059 | *pprev = NULL; | |
2060 | while (rb_node) { | |
2061 | *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
2062 | rb_node = rb_node->rb_right; | |
2063 | } | |
2064 | } | |
6bd4837d | 2065 | return vma; |
1da177e4 LT |
2066 | } |
2067 | ||
2068 | /* | |
2069 | * Verify that the stack growth is acceptable and | |
2070 | * update accounting. This is shared with both the | |
2071 | * grow-up and grow-down cases. | |
2072 | */ | |
1ad9a25d HD |
2073 | static int acct_stack_growth(struct vm_area_struct *vma, |
2074 | unsigned long size, unsigned long grow) | |
1da177e4 LT |
2075 | { |
2076 | struct mm_struct *mm = vma->vm_mm; | |
2077 | struct rlimit *rlim = current->signal->rlim; | |
1ad9a25d | 2078 | unsigned long new_start; |
1da177e4 LT |
2079 | |
2080 | /* address space limit tests */ | |
119f657c | 2081 | if (!may_expand_vm(mm, grow)) |
1da177e4 LT |
2082 | return -ENOMEM; |
2083 | ||
2084 | /* Stack limit test */ | |
1ad9a25d | 2085 | if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur)) |
1da177e4 LT |
2086 | return -ENOMEM; |
2087 | ||
2088 | /* mlock limit tests */ | |
2089 | if (vma->vm_flags & VM_LOCKED) { | |
2090 | unsigned long locked; | |
2091 | unsigned long limit; | |
2092 | locked = mm->locked_vm + grow; | |
59e99e5b JS |
2093 | limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); |
2094 | limit >>= PAGE_SHIFT; | |
1da177e4 LT |
2095 | if (locked > limit && !capable(CAP_IPC_LOCK)) |
2096 | return -ENOMEM; | |
2097 | } | |
2098 | ||
0d59a01b AL |
2099 | /* Check to ensure the stack will not grow into a hugetlb-only region */ |
2100 | new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : | |
2101 | vma->vm_end - size; | |
2102 | if (is_hugepage_only_range(vma->vm_mm, new_start, size)) | |
2103 | return -EFAULT; | |
2104 | ||
1da177e4 LT |
2105 | /* |
2106 | * Overcommit.. This must be the final test, as it will | |
2107 | * update security statistics. | |
2108 | */ | |
05fa199d | 2109 | if (security_vm_enough_memory_mm(mm, grow)) |
1da177e4 LT |
2110 | return -ENOMEM; |
2111 | ||
2112 | /* Ok, everything looks good - let it rip */ | |
1da177e4 LT |
2113 | if (vma->vm_flags & VM_LOCKED) |
2114 | mm->locked_vm += grow; | |
ab50b8ed | 2115 | vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); |
1da177e4 LT |
2116 | return 0; |
2117 | } | |
2118 | ||
46dea3d0 | 2119 | #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) |
1da177e4 | 2120 | /* |
46dea3d0 HD |
2121 | * PA-RISC uses this for its stack; IA64 for its Register Backing Store. |
2122 | * vma is the last one with address > vma->vm_end. Have to extend vma. | |
1da177e4 | 2123 | */ |
46dea3d0 | 2124 | int expand_upwards(struct vm_area_struct *vma, unsigned long address) |
1da177e4 | 2125 | { |
1ad9a25d HD |
2126 | struct vm_area_struct *next; |
2127 | unsigned long gap_addr; | |
2128 | int error = 0; | |
1da177e4 LT |
2129 | |
2130 | if (!(vma->vm_flags & VM_GROWSUP)) | |
2131 | return -EFAULT; | |
2132 | ||
0fcba8f9 | 2133 | /* Guard against exceeding limits of the address space. */ |
1ad9a25d | 2134 | address &= PAGE_MASK; |
0fcba8f9 | 2135 | if (address >= TASK_SIZE) |
1ad9a25d | 2136 | return -ENOMEM; |
0fcba8f9 | 2137 | address += PAGE_SIZE; |
1ad9a25d HD |
2138 | |
2139 | /* Enforce stack_guard_gap */ | |
2140 | gap_addr = address + stack_guard_gap; | |
0fcba8f9 HD |
2141 | |
2142 | /* Guard against overflow */ | |
2143 | if (gap_addr < address || gap_addr > TASK_SIZE) | |
2144 | gap_addr = TASK_SIZE; | |
2145 | ||
1ad9a25d HD |
2146 | next = vma->vm_next; |
2147 | if (next && next->vm_start < gap_addr) { | |
2148 | if (!(next->vm_flags & VM_GROWSUP)) | |
2149 | return -ENOMEM; | |
2150 | /* Check that both stack segments have the same anon_vma? */ | |
2151 | } | |
2152 | ||
2153 | /* We must make sure the anon_vma is allocated. */ | |
1da177e4 LT |
2154 | if (unlikely(anon_vma_prepare(vma))) |
2155 | return -ENOMEM; | |
1da177e4 LT |
2156 | |
2157 | /* | |
2158 | * vma->vm_start/vm_end cannot change under us because the caller | |
2159 | * is required to hold the mmap_sem in read mode. We need the | |
2160 | * anon_vma lock to serialize against concurrent expand_stacks. | |
2161 | */ | |
1ad9a25d | 2162 | vma_lock_anon_vma(vma); |
1da177e4 LT |
2163 | |
2164 | /* Somebody else might have raced and expanded it already */ | |
2165 | if (address > vma->vm_end) { | |
2166 | unsigned long size, grow; | |
2167 | ||
2168 | size = address - vma->vm_start; | |
2169 | grow = (address - vma->vm_end) >> PAGE_SHIFT; | |
2170 | ||
42c36f63 HD |
2171 | error = -ENOMEM; |
2172 | if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { | |
2173 | error = acct_stack_growth(vma, size, grow); | |
2174 | if (!error) { | |
4128997b ML |
2175 | /* |
2176 | * vma_gap_update() doesn't support concurrent | |
2177 | * updates, but we only hold a shared mmap_sem | |
2178 | * lock here, so we need to protect against | |
2179 | * concurrent vma expansions. | |
2180 | * vma_lock_anon_vma() doesn't help here, as | |
2181 | * we don't guarantee that all growable vmas | |
2182 | * in a mm share the same root anon vma. | |
2183 | * So, we reuse mm->page_table_lock to guard | |
2184 | * against concurrent vma expansions. | |
2185 | */ | |
2186 | spin_lock(&vma->vm_mm->page_table_lock); | |
bf181b9f | 2187 | anon_vma_interval_tree_pre_update_vma(vma); |
42c36f63 | 2188 | vma->vm_end = address; |
bf181b9f | 2189 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 ML |
2190 | if (vma->vm_next) |
2191 | vma_gap_update(vma->vm_next); | |
2192 | else | |
1ad9a25d | 2193 | vma->vm_mm->highest_vm_end = vm_end_gap(vma); |
4128997b ML |
2194 | spin_unlock(&vma->vm_mm->page_table_lock); |
2195 | ||
42c36f63 HD |
2196 | perf_event_mmap(vma); |
2197 | } | |
3af9e859 | 2198 | } |
1da177e4 | 2199 | } |
bb4a340e | 2200 | vma_unlock_anon_vma(vma); |
b15d00b6 | 2201 | khugepaged_enter_vma_merge(vma); |
ed8ea815 | 2202 | validate_mm(vma->vm_mm); |
1da177e4 LT |
2203 | return error; |
2204 | } | |
46dea3d0 HD |
2205 | #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ |
2206 | ||
1da177e4 LT |
2207 | /* |
2208 | * vma is the first one with address < vma->vm_start. Have to extend vma. | |
2209 | */ | |
d05f3169 | 2210 | int expand_downwards(struct vm_area_struct *vma, |
b6a2fea3 | 2211 | unsigned long address) |
1da177e4 | 2212 | { |
1ad9a25d HD |
2213 | struct vm_area_struct *prev; |
2214 | unsigned long gap_addr; | |
1da177e4 LT |
2215 | int error; |
2216 | ||
8869477a | 2217 | address &= PAGE_MASK; |
e5467859 | 2218 | error = security_mmap_addr(address); |
8869477a EP |
2219 | if (error) |
2220 | return error; | |
2221 | ||
1ad9a25d HD |
2222 | /* Enforce stack_guard_gap */ |
2223 | gap_addr = address - stack_guard_gap; | |
2224 | if (gap_addr > address) | |
2225 | return -ENOMEM; | |
2226 | prev = vma->vm_prev; | |
2227 | if (prev && prev->vm_end > gap_addr) { | |
2228 | if (!(prev->vm_flags & VM_GROWSDOWN)) | |
2229 | return -ENOMEM; | |
2230 | /* Check that both stack segments have the same anon_vma? */ | |
2231 | } | |
2232 | ||
2233 | /* We must make sure the anon_vma is allocated. */ | |
2234 | if (unlikely(anon_vma_prepare(vma))) | |
2235 | return -ENOMEM; | |
1da177e4 LT |
2236 | |
2237 | /* | |
2238 | * vma->vm_start/vm_end cannot change under us because the caller | |
2239 | * is required to hold the mmap_sem in read mode. We need the | |
2240 | * anon_vma lock to serialize against concurrent expand_stacks. | |
2241 | */ | |
1ad9a25d | 2242 | vma_lock_anon_vma(vma); |
1da177e4 LT |
2243 | |
2244 | /* Somebody else might have raced and expanded it already */ | |
2245 | if (address < vma->vm_start) { | |
2246 | unsigned long size, grow; | |
2247 | ||
2248 | size = vma->vm_end - address; | |
2249 | grow = (vma->vm_start - address) >> PAGE_SHIFT; | |
2250 | ||
a626ca6a LT |
2251 | error = -ENOMEM; |
2252 | if (grow <= vma->vm_pgoff) { | |
2253 | error = acct_stack_growth(vma, size, grow); | |
2254 | if (!error) { | |
4128997b ML |
2255 | /* |
2256 | * vma_gap_update() doesn't support concurrent | |
2257 | * updates, but we only hold a shared mmap_sem | |
2258 | * lock here, so we need to protect against | |
2259 | * concurrent vma expansions. | |
2260 | * vma_lock_anon_vma() doesn't help here, as | |
2261 | * we don't guarantee that all growable vmas | |
2262 | * in a mm share the same root anon vma. | |
2263 | * So, we reuse mm->page_table_lock to guard | |
2264 | * against concurrent vma expansions. | |
2265 | */ | |
2266 | spin_lock(&vma->vm_mm->page_table_lock); | |
bf181b9f | 2267 | anon_vma_interval_tree_pre_update_vma(vma); |
a626ca6a LT |
2268 | vma->vm_start = address; |
2269 | vma->vm_pgoff -= grow; | |
bf181b9f | 2270 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 | 2271 | vma_gap_update(vma); |
4128997b ML |
2272 | spin_unlock(&vma->vm_mm->page_table_lock); |
2273 | ||
a626ca6a LT |
2274 | perf_event_mmap(vma); |
2275 | } | |
1da177e4 LT |
2276 | } |
2277 | } | |
bb4a340e | 2278 | vma_unlock_anon_vma(vma); |
b15d00b6 | 2279 | khugepaged_enter_vma_merge(vma); |
ed8ea815 | 2280 | validate_mm(vma->vm_mm); |
1da177e4 LT |
2281 | return error; |
2282 | } | |
2283 | ||
1ad9a25d HD |
2284 | /* enforced gap between the expanding stack and other mappings. */ |
2285 | unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; | |
2286 | ||
2287 | static int __init cmdline_parse_stack_guard_gap(char *p) | |
2288 | { | |
2289 | unsigned long val; | |
2290 | char *endptr; | |
2291 | ||
2292 | val = simple_strtoul(p, &endptr, 10); | |
2293 | if (!*endptr) | |
2294 | stack_guard_gap = val << PAGE_SHIFT; | |
2295 | ||
2296 | return 0; | |
2297 | } | |
2298 | __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); | |
2299 | ||
b6a2fea3 OW |
2300 | #ifdef CONFIG_STACK_GROWSUP |
2301 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2302 | { | |
2303 | return expand_upwards(vma, address); | |
2304 | } | |
2305 | ||
2306 | struct vm_area_struct * | |
2307 | find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
2308 | { | |
2309 | struct vm_area_struct *vma, *prev; | |
2310 | ||
2311 | addr &= PAGE_MASK; | |
2312 | vma = find_vma_prev(mm, addr, &prev); | |
2313 | if (vma && (vma->vm_start <= addr)) | |
2314 | return vma; | |
1c127185 | 2315 | if (!prev || expand_stack(prev, addr)) |
b6a2fea3 | 2316 | return NULL; |
cea10a19 ML |
2317 | if (prev->vm_flags & VM_LOCKED) |
2318 | __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL); | |
b6a2fea3 OW |
2319 | return prev; |
2320 | } | |
2321 | #else | |
2322 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2323 | { | |
2324 | return expand_downwards(vma, address); | |
2325 | } | |
2326 | ||
1da177e4 LT |
2327 | struct vm_area_struct * |
2328 | find_extend_vma(struct mm_struct * mm, unsigned long addr) | |
2329 | { | |
2330 | struct vm_area_struct * vma; | |
2331 | unsigned long start; | |
2332 | ||
2333 | addr &= PAGE_MASK; | |
2334 | vma = find_vma(mm,addr); | |
2335 | if (!vma) | |
2336 | return NULL; | |
2337 | if (vma->vm_start <= addr) | |
2338 | return vma; | |
2339 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
2340 | return NULL; | |
2341 | start = vma->vm_start; | |
2342 | if (expand_stack(vma, addr)) | |
2343 | return NULL; | |
cea10a19 ML |
2344 | if (vma->vm_flags & VM_LOCKED) |
2345 | __mlock_vma_pages_range(vma, addr, start, NULL); | |
1da177e4 LT |
2346 | return vma; |
2347 | } | |
2348 | #endif | |
2349 | ||
1da177e4 | 2350 | /* |
2c0b3814 | 2351 | * Ok - we have the memory areas we should free on the vma list, |
1da177e4 | 2352 | * so release them, and do the vma updates. |
2c0b3814 HD |
2353 | * |
2354 | * Called with the mm semaphore held. | |
1da177e4 | 2355 | */ |
2c0b3814 | 2356 | static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 2357 | { |
4f74d2c8 LT |
2358 | unsigned long nr_accounted = 0; |
2359 | ||
365e9c87 HD |
2360 | /* Update high watermark before we lower total_vm */ |
2361 | update_hiwater_vm(mm); | |
1da177e4 | 2362 | do { |
2c0b3814 HD |
2363 | long nrpages = vma_pages(vma); |
2364 | ||
4f74d2c8 LT |
2365 | if (vma->vm_flags & VM_ACCOUNT) |
2366 | nr_accounted += nrpages; | |
2c0b3814 | 2367 | vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); |
a8fb5618 | 2368 | vma = remove_vma(vma); |
146425a3 | 2369 | } while (vma); |
4f74d2c8 | 2370 | vm_unacct_memory(nr_accounted); |
1da177e4 LT |
2371 | validate_mm(mm); |
2372 | } | |
2373 | ||
2374 | /* | |
2375 | * Get rid of page table information in the indicated region. | |
2376 | * | |
f10df686 | 2377 | * Called with the mm semaphore held. |
1da177e4 LT |
2378 | */ |
2379 | static void unmap_region(struct mm_struct *mm, | |
e0da382c HD |
2380 | struct vm_area_struct *vma, struct vm_area_struct *prev, |
2381 | unsigned long start, unsigned long end) | |
1da177e4 | 2382 | { |
e0da382c | 2383 | struct vm_area_struct *next = prev? prev->vm_next: mm->mmap; |
d16dfc55 | 2384 | struct mmu_gather tlb; |
1da177e4 LT |
2385 | |
2386 | lru_add_drain(); | |
8e220cfd | 2387 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 2388 | update_hiwater_rss(mm); |
4f74d2c8 | 2389 | unmap_vmas(&tlb, vma, start, end); |
d16dfc55 | 2390 | free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, |
6ee8630e | 2391 | next ? next->vm_start : USER_PGTABLES_CEILING); |
d16dfc55 | 2392 | tlb_finish_mmu(&tlb, start, end); |
1da177e4 LT |
2393 | } |
2394 | ||
2395 | /* | |
2396 | * Create a list of vma's touched by the unmap, removing them from the mm's | |
2397 | * vma list as we go.. | |
2398 | */ | |
2399 | static void | |
2400 | detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, | |
2401 | struct vm_area_struct *prev, unsigned long end) | |
2402 | { | |
2403 | struct vm_area_struct **insertion_point; | |
2404 | struct vm_area_struct *tail_vma = NULL; | |
1363c3cd | 2405 | unsigned long addr; |
1da177e4 LT |
2406 | |
2407 | insertion_point = (prev ? &prev->vm_next : &mm->mmap); | |
297c5eee | 2408 | vma->vm_prev = NULL; |
1da177e4 | 2409 | do { |
d3737187 | 2410 | vma_rb_erase(vma, &mm->mm_rb); |
1da177e4 LT |
2411 | mm->map_count--; |
2412 | tail_vma = vma; | |
2413 | vma = vma->vm_next; | |
2414 | } while (vma && vma->vm_start < end); | |
2415 | *insertion_point = vma; | |
d3737187 | 2416 | if (vma) { |
297c5eee | 2417 | vma->vm_prev = prev; |
d3737187 ML |
2418 | vma_gap_update(vma); |
2419 | } else | |
1ad9a25d | 2420 | mm->highest_vm_end = prev ? vm_end_gap(prev) : 0; |
1da177e4 | 2421 | tail_vma->vm_next = NULL; |
1363c3cd WW |
2422 | if (mm->unmap_area == arch_unmap_area) |
2423 | addr = prev ? prev->vm_end : mm->mmap_base; | |
2424 | else | |
2425 | addr = vma ? vma->vm_start : mm->mmap_base; | |
2426 | mm->unmap_area(mm, addr); | |
1da177e4 LT |
2427 | mm->mmap_cache = NULL; /* Kill the cache. */ |
2428 | } | |
2429 | ||
2430 | /* | |
659ace58 KM |
2431 | * __split_vma() bypasses sysctl_max_map_count checking. We use this on the |
2432 | * munmap path where it doesn't make sense to fail. | |
1da177e4 | 2433 | */ |
659ace58 | 2434 | static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma, |
1da177e4 LT |
2435 | unsigned long addr, int new_below) |
2436 | { | |
2437 | struct mempolicy *pol; | |
2438 | struct vm_area_struct *new; | |
5beb4930 | 2439 | int err = -ENOMEM; |
1da177e4 | 2440 | |
a5516438 AK |
2441 | if (is_vm_hugetlb_page(vma) && (addr & |
2442 | ~(huge_page_mask(hstate_vma(vma))))) | |
1da177e4 LT |
2443 | return -EINVAL; |
2444 | ||
e94b1766 | 2445 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 | 2446 | if (!new) |
5beb4930 | 2447 | goto out_err; |
1da177e4 LT |
2448 | |
2449 | /* most fields are the same, copy all, and then fixup */ | |
2450 | *new = *vma; | |
2451 | ||
5beb4930 RR |
2452 | INIT_LIST_HEAD(&new->anon_vma_chain); |
2453 | ||
1da177e4 LT |
2454 | if (new_below) |
2455 | new->vm_end = addr; | |
2456 | else { | |
2457 | new->vm_start = addr; | |
2458 | new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); | |
2459 | } | |
2460 | ||
846a16bf | 2461 | pol = mpol_dup(vma_policy(vma)); |
1da177e4 | 2462 | if (IS_ERR(pol)) { |
5beb4930 RR |
2463 | err = PTR_ERR(pol); |
2464 | goto out_free_vma; | |
1da177e4 LT |
2465 | } |
2466 | vma_set_policy(new, pol); | |
2467 | ||
5beb4930 RR |
2468 | if (anon_vma_clone(new, vma)) |
2469 | goto out_free_mpol; | |
2470 | ||
e9714acf | 2471 | if (new->vm_file) |
1da177e4 LT |
2472 | get_file(new->vm_file); |
2473 | ||
2474 | if (new->vm_ops && new->vm_ops->open) | |
2475 | new->vm_ops->open(new); | |
2476 | ||
2477 | if (new_below) | |
5beb4930 | 2478 | err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + |
1da177e4 LT |
2479 | ((addr - new->vm_start) >> PAGE_SHIFT), new); |
2480 | else | |
5beb4930 | 2481 | err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); |
1da177e4 | 2482 | |
5beb4930 RR |
2483 | /* Success. */ |
2484 | if (!err) | |
2485 | return 0; | |
2486 | ||
2487 | /* Clean everything up if vma_adjust failed. */ | |
58927533 RR |
2488 | if (new->vm_ops && new->vm_ops->close) |
2489 | new->vm_ops->close(new); | |
e9714acf | 2490 | if (new->vm_file) |
5beb4930 | 2491 | fput(new->vm_file); |
2aeadc30 | 2492 | unlink_anon_vmas(new); |
5beb4930 RR |
2493 | out_free_mpol: |
2494 | mpol_put(pol); | |
2495 | out_free_vma: | |
2496 | kmem_cache_free(vm_area_cachep, new); | |
2497 | out_err: | |
2498 | return err; | |
1da177e4 LT |
2499 | } |
2500 | ||
659ace58 KM |
2501 | /* |
2502 | * Split a vma into two pieces at address 'addr', a new vma is allocated | |
2503 | * either for the first part or the tail. | |
2504 | */ | |
2505 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, | |
2506 | unsigned long addr, int new_below) | |
2507 | { | |
2508 | if (mm->map_count >= sysctl_max_map_count) | |
2509 | return -ENOMEM; | |
2510 | ||
2511 | return __split_vma(mm, vma, addr, new_below); | |
2512 | } | |
2513 | ||
1da177e4 LT |
2514 | /* Munmap is split into 2 main parts -- this part which finds |
2515 | * what needs doing, and the areas themselves, which do the | |
2516 | * work. This now handles partial unmappings. | |
2517 | * Jeremy Fitzhardinge <jeremy@goop.org> | |
2518 | */ | |
2519 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | |
2520 | { | |
2521 | unsigned long end; | |
146425a3 | 2522 | struct vm_area_struct *vma, *prev, *last; |
1da177e4 LT |
2523 | |
2524 | if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) | |
2525 | return -EINVAL; | |
2526 | ||
2527 | if ((len = PAGE_ALIGN(len)) == 0) | |
2528 | return -EINVAL; | |
2529 | ||
2530 | /* Find the first overlapping VMA */ | |
9be34c9d | 2531 | vma = find_vma(mm, start); |
146425a3 | 2532 | if (!vma) |
1da177e4 | 2533 | return 0; |
9be34c9d | 2534 | prev = vma->vm_prev; |
146425a3 | 2535 | /* we have start < vma->vm_end */ |
1da177e4 LT |
2536 | |
2537 | /* if it doesn't overlap, we have nothing.. */ | |
2538 | end = start + len; | |
146425a3 | 2539 | if (vma->vm_start >= end) |
1da177e4 LT |
2540 | return 0; |
2541 | ||
2542 | /* | |
2543 | * If we need to split any vma, do it now to save pain later. | |
2544 | * | |
2545 | * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially | |
2546 | * unmapped vm_area_struct will remain in use: so lower split_vma | |
2547 | * places tmp vma above, and higher split_vma places tmp vma below. | |
2548 | */ | |
146425a3 | 2549 | if (start > vma->vm_start) { |
659ace58 KM |
2550 | int error; |
2551 | ||
2552 | /* | |
2553 | * Make sure that map_count on return from munmap() will | |
2554 | * not exceed its limit; but let map_count go just above | |
2555 | * its limit temporarily, to help free resources as expected. | |
2556 | */ | |
2557 | if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) | |
2558 | return -ENOMEM; | |
2559 | ||
2560 | error = __split_vma(mm, vma, start, 0); | |
1da177e4 LT |
2561 | if (error) |
2562 | return error; | |
146425a3 | 2563 | prev = vma; |
1da177e4 LT |
2564 | } |
2565 | ||
2566 | /* Does it split the last one? */ | |
2567 | last = find_vma(mm, end); | |
2568 | if (last && end > last->vm_start) { | |
659ace58 | 2569 | int error = __split_vma(mm, last, end, 1); |
1da177e4 LT |
2570 | if (error) |
2571 | return error; | |
2572 | } | |
146425a3 | 2573 | vma = prev? prev->vm_next: mm->mmap; |
1da177e4 | 2574 | |
ba470de4 RR |
2575 | /* |
2576 | * unlock any mlock()ed ranges before detaching vmas | |
2577 | */ | |
2578 | if (mm->locked_vm) { | |
2579 | struct vm_area_struct *tmp = vma; | |
2580 | while (tmp && tmp->vm_start < end) { | |
2581 | if (tmp->vm_flags & VM_LOCKED) { | |
2582 | mm->locked_vm -= vma_pages(tmp); | |
2583 | munlock_vma_pages_all(tmp); | |
2584 | } | |
2585 | tmp = tmp->vm_next; | |
2586 | } | |
2587 | } | |
2588 | ||
1da177e4 LT |
2589 | /* |
2590 | * Remove the vma's, and unmap the actual pages | |
2591 | */ | |
146425a3 HD |
2592 | detach_vmas_to_be_unmapped(mm, vma, prev, end); |
2593 | unmap_region(mm, vma, prev, start, end); | |
1da177e4 LT |
2594 | |
2595 | /* Fix up all other VM information */ | |
2c0b3814 | 2596 | remove_vma_list(mm, vma); |
1da177e4 LT |
2597 | |
2598 | return 0; | |
2599 | } | |
1da177e4 | 2600 | |
bfce281c | 2601 | int vm_munmap(unsigned long start, size_t len) |
1da177e4 LT |
2602 | { |
2603 | int ret; | |
bfce281c | 2604 | struct mm_struct *mm = current->mm; |
1da177e4 LT |
2605 | |
2606 | down_write(&mm->mmap_sem); | |
a46ef99d | 2607 | ret = do_munmap(mm, start, len); |
1da177e4 LT |
2608 | up_write(&mm->mmap_sem); |
2609 | return ret; | |
2610 | } | |
a46ef99d LT |
2611 | EXPORT_SYMBOL(vm_munmap); |
2612 | ||
2613 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) | |
2614 | { | |
2615 | profile_munmap(addr); | |
bfce281c | 2616 | return vm_munmap(addr, len); |
a46ef99d | 2617 | } |
1da177e4 LT |
2618 | |
2619 | static inline void verify_mm_writelocked(struct mm_struct *mm) | |
2620 | { | |
a241ec65 | 2621 | #ifdef CONFIG_DEBUG_VM |
1da177e4 LT |
2622 | if (unlikely(down_read_trylock(&mm->mmap_sem))) { |
2623 | WARN_ON(1); | |
2624 | up_read(&mm->mmap_sem); | |
2625 | } | |
2626 | #endif | |
2627 | } | |
2628 | ||
2629 | /* | |
2630 | * this is really a simplified "do_mmap". it only handles | |
2631 | * anonymous maps. eventually we may be able to do some | |
2632 | * brk-specific accounting here. | |
2633 | */ | |
e4eb1ff6 | 2634 | static unsigned long do_brk(unsigned long addr, unsigned long len) |
1da177e4 LT |
2635 | { |
2636 | struct mm_struct * mm = current->mm; | |
2637 | struct vm_area_struct * vma, * prev; | |
2638 | unsigned long flags; | |
2639 | struct rb_node ** rb_link, * rb_parent; | |
2640 | pgoff_t pgoff = addr >> PAGE_SHIFT; | |
3a459756 | 2641 | int error; |
1da177e4 LT |
2642 | |
2643 | len = PAGE_ALIGN(len); | |
2644 | if (!len) | |
2645 | return addr; | |
2646 | ||
3a459756 KK |
2647 | flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; |
2648 | ||
2c6a1016 AV |
2649 | error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); |
2650 | if (error & ~PAGE_MASK) | |
3a459756 KK |
2651 | return error; |
2652 | ||
1da177e4 LT |
2653 | /* |
2654 | * mlock MCL_FUTURE? | |
2655 | */ | |
2656 | if (mm->def_flags & VM_LOCKED) { | |
2657 | unsigned long locked, lock_limit; | |
93ea1d0a CW |
2658 | locked = len >> PAGE_SHIFT; |
2659 | locked += mm->locked_vm; | |
59e99e5b | 2660 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
93ea1d0a | 2661 | lock_limit >>= PAGE_SHIFT; |
1da177e4 LT |
2662 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) |
2663 | return -EAGAIN; | |
2664 | } | |
2665 | ||
2666 | /* | |
2667 | * mm->mmap_sem is required to protect against another thread | |
2668 | * changing the mappings in case we sleep. | |
2669 | */ | |
2670 | verify_mm_writelocked(mm); | |
2671 | ||
2672 | /* | |
2673 | * Clear old maps. this also does some error checking for us | |
2674 | */ | |
2675 | munmap_back: | |
6597d783 | 2676 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { |
1da177e4 LT |
2677 | if (do_munmap(mm, addr, len)) |
2678 | return -ENOMEM; | |
2679 | goto munmap_back; | |
2680 | } | |
2681 | ||
2682 | /* Check against address space limits *after* clearing old maps... */ | |
119f657c | 2683 | if (!may_expand_vm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
2684 | return -ENOMEM; |
2685 | ||
2686 | if (mm->map_count > sysctl_max_map_count) | |
2687 | return -ENOMEM; | |
2688 | ||
191c5424 | 2689 | if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
2690 | return -ENOMEM; |
2691 | ||
1da177e4 | 2692 | /* Can we just expand an old private anonymous mapping? */ |
ba470de4 RR |
2693 | vma = vma_merge(mm, prev, addr, addr + len, flags, |
2694 | NULL, NULL, pgoff, NULL); | |
2695 | if (vma) | |
1da177e4 LT |
2696 | goto out; |
2697 | ||
2698 | /* | |
2699 | * create a vma struct for an anonymous mapping | |
2700 | */ | |
c5e3b83e | 2701 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
2702 | if (!vma) { |
2703 | vm_unacct_memory(len >> PAGE_SHIFT); | |
2704 | return -ENOMEM; | |
2705 | } | |
1da177e4 | 2706 | |
5beb4930 | 2707 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 LT |
2708 | vma->vm_mm = mm; |
2709 | vma->vm_start = addr; | |
2710 | vma->vm_end = addr + len; | |
2711 | vma->vm_pgoff = pgoff; | |
2712 | vma->vm_flags = flags; | |
3ed75eb8 | 2713 | vma->vm_page_prot = vm_get_page_prot(flags); |
1da177e4 LT |
2714 | vma_link(mm, vma, prev, rb_link, rb_parent); |
2715 | out: | |
3af9e859 | 2716 | perf_event_mmap(vma); |
1da177e4 | 2717 | mm->total_vm += len >> PAGE_SHIFT; |
128557ff ML |
2718 | if (flags & VM_LOCKED) |
2719 | mm->locked_vm += (len >> PAGE_SHIFT); | |
1da177e4 LT |
2720 | return addr; |
2721 | } | |
2722 | ||
e4eb1ff6 LT |
2723 | unsigned long vm_brk(unsigned long addr, unsigned long len) |
2724 | { | |
2725 | struct mm_struct *mm = current->mm; | |
2726 | unsigned long ret; | |
128557ff | 2727 | bool populate; |
e4eb1ff6 LT |
2728 | |
2729 | down_write(&mm->mmap_sem); | |
2730 | ret = do_brk(addr, len); | |
128557ff | 2731 | populate = ((mm->def_flags & VM_LOCKED) != 0); |
e4eb1ff6 | 2732 | up_write(&mm->mmap_sem); |
128557ff ML |
2733 | if (populate) |
2734 | mm_populate(addr, len); | |
e4eb1ff6 LT |
2735 | return ret; |
2736 | } | |
2737 | EXPORT_SYMBOL(vm_brk); | |
1da177e4 LT |
2738 | |
2739 | /* Release all mmaps. */ | |
2740 | void exit_mmap(struct mm_struct *mm) | |
2741 | { | |
d16dfc55 | 2742 | struct mmu_gather tlb; |
ba470de4 | 2743 | struct vm_area_struct *vma; |
1da177e4 LT |
2744 | unsigned long nr_accounted = 0; |
2745 | ||
d6dd61c8 | 2746 | /* mm's last user has gone, and its about to be pulled down */ |
cddb8a5c | 2747 | mmu_notifier_release(mm); |
d6dd61c8 | 2748 | |
ba470de4 RR |
2749 | if (mm->locked_vm) { |
2750 | vma = mm->mmap; | |
2751 | while (vma) { | |
2752 | if (vma->vm_flags & VM_LOCKED) | |
2753 | munlock_vma_pages_all(vma); | |
2754 | vma = vma->vm_next; | |
2755 | } | |
2756 | } | |
9480c53e JF |
2757 | |
2758 | arch_exit_mmap(mm); | |
2759 | ||
ba470de4 | 2760 | vma = mm->mmap; |
9480c53e JF |
2761 | if (!vma) /* Can happen if dup_mmap() received an OOM */ |
2762 | return; | |
2763 | ||
1da177e4 | 2764 | lru_add_drain(); |
1da177e4 | 2765 | flush_cache_mm(mm); |
8e220cfd | 2766 | tlb_gather_mmu(&tlb, mm, 0, -1); |
901608d9 | 2767 | /* update_hiwater_rss(mm) here? but nobody should be looking */ |
e0da382c | 2768 | /* Use -1 here to ensure all VMAs in the mm are unmapped */ |
4f74d2c8 | 2769 | unmap_vmas(&tlb, vma, 0, -1); |
9ba69294 | 2770 | |
6ee8630e | 2771 | free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); |
853f5e26 | 2772 | tlb_finish_mmu(&tlb, 0, -1); |
1da177e4 | 2773 | |
1da177e4 | 2774 | /* |
8f4f8c16 HD |
2775 | * Walk the list again, actually closing and freeing it, |
2776 | * with preemption enabled, without holding any MM locks. | |
1da177e4 | 2777 | */ |
4f74d2c8 LT |
2778 | while (vma) { |
2779 | if (vma->vm_flags & VM_ACCOUNT) | |
2780 | nr_accounted += vma_pages(vma); | |
a8fb5618 | 2781 | vma = remove_vma(vma); |
4f74d2c8 LT |
2782 | } |
2783 | vm_unacct_memory(nr_accounted); | |
e0da382c | 2784 | |
f9aed62a | 2785 | WARN_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT); |
1da177e4 LT |
2786 | } |
2787 | ||
2788 | /* Insert vm structure into process list sorted by address | |
2789 | * and into the inode's i_mmap tree. If vm_file is non-NULL | |
3d48ae45 | 2790 | * then i_mmap_mutex is taken here. |
1da177e4 | 2791 | */ |
6597d783 | 2792 | int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 2793 | { |
6597d783 HD |
2794 | struct vm_area_struct *prev; |
2795 | struct rb_node **rb_link, *rb_parent; | |
1da177e4 LT |
2796 | |
2797 | /* | |
2798 | * The vm_pgoff of a purely anonymous vma should be irrelevant | |
2799 | * until its first write fault, when page's anon_vma and index | |
2800 | * are set. But now set the vm_pgoff it will almost certainly | |
2801 | * end up with (unless mremap moves it elsewhere before that | |
2802 | * first wfault), so /proc/pid/maps tells a consistent story. | |
2803 | * | |
2804 | * By setting it to reflect the virtual start address of the | |
2805 | * vma, merges and splits can happen in a seamless way, just | |
2806 | * using the existing file pgoff checks and manipulations. | |
2807 | * Similarly in do_mmap_pgoff and in do_brk. | |
2808 | */ | |
2809 | if (!vma->vm_file) { | |
2810 | BUG_ON(vma->anon_vma); | |
2811 | vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; | |
2812 | } | |
6597d783 HD |
2813 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
2814 | &prev, &rb_link, &rb_parent)) | |
1da177e4 | 2815 | return -ENOMEM; |
2fd4ef85 | 2816 | if ((vma->vm_flags & VM_ACCOUNT) && |
34b4e4aa | 2817 | security_vm_enough_memory_mm(mm, vma_pages(vma))) |
2fd4ef85 | 2818 | return -ENOMEM; |
2b144498 | 2819 | |
1da177e4 LT |
2820 | vma_link(mm, vma, prev, rb_link, rb_parent); |
2821 | return 0; | |
2822 | } | |
2823 | ||
2824 | /* | |
2825 | * Copy the vma structure to a new location in the same mm, | |
2826 | * prior to moving page table entries, to effect an mremap move. | |
2827 | */ | |
2828 | struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, | |
38a76013 ML |
2829 | unsigned long addr, unsigned long len, pgoff_t pgoff, |
2830 | bool *need_rmap_locks) | |
1da177e4 LT |
2831 | { |
2832 | struct vm_area_struct *vma = *vmap; | |
2833 | unsigned long vma_start = vma->vm_start; | |
2834 | struct mm_struct *mm = vma->vm_mm; | |
2835 | struct vm_area_struct *new_vma, *prev; | |
2836 | struct rb_node **rb_link, *rb_parent; | |
2837 | struct mempolicy *pol; | |
948f017b | 2838 | bool faulted_in_anon_vma = true; |
1da177e4 LT |
2839 | |
2840 | /* | |
2841 | * If anonymous vma has not yet been faulted, update new pgoff | |
2842 | * to match new location, to increase its chance of merging. | |
2843 | */ | |
948f017b | 2844 | if (unlikely(!vma->vm_file && !vma->anon_vma)) { |
1da177e4 | 2845 | pgoff = addr >> PAGE_SHIFT; |
948f017b AA |
2846 | faulted_in_anon_vma = false; |
2847 | } | |
1da177e4 | 2848 | |
6597d783 HD |
2849 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) |
2850 | return NULL; /* should never get here */ | |
1da177e4 LT |
2851 | new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, |
2852 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); | |
2853 | if (new_vma) { | |
2854 | /* | |
2855 | * Source vma may have been merged into new_vma | |
2856 | */ | |
948f017b AA |
2857 | if (unlikely(vma_start >= new_vma->vm_start && |
2858 | vma_start < new_vma->vm_end)) { | |
2859 | /* | |
2860 | * The only way we can get a vma_merge with | |
2861 | * self during an mremap is if the vma hasn't | |
2862 | * been faulted in yet and we were allowed to | |
2863 | * reset the dst vma->vm_pgoff to the | |
2864 | * destination address of the mremap to allow | |
2865 | * the merge to happen. mremap must change the | |
2866 | * vm_pgoff linearity between src and dst vmas | |
2867 | * (in turn preventing a vma_merge) to be | |
2868 | * safe. It is only safe to keep the vm_pgoff | |
2869 | * linear if there are no pages mapped yet. | |
2870 | */ | |
2871 | VM_BUG_ON(faulted_in_anon_vma); | |
38a76013 | 2872 | *vmap = vma = new_vma; |
108d6642 | 2873 | } |
38a76013 | 2874 | *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); |
1da177e4 | 2875 | } else { |
e94b1766 | 2876 | new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
2877 | if (new_vma) { |
2878 | *new_vma = *vma; | |
523d4e20 ML |
2879 | new_vma->vm_start = addr; |
2880 | new_vma->vm_end = addr + len; | |
2881 | new_vma->vm_pgoff = pgoff; | |
846a16bf | 2882 | pol = mpol_dup(vma_policy(vma)); |
5beb4930 RR |
2883 | if (IS_ERR(pol)) |
2884 | goto out_free_vma; | |
523d4e20 | 2885 | vma_set_policy(new_vma, pol); |
5beb4930 RR |
2886 | INIT_LIST_HEAD(&new_vma->anon_vma_chain); |
2887 | if (anon_vma_clone(new_vma, vma)) | |
2888 | goto out_free_mempol; | |
e9714acf | 2889 | if (new_vma->vm_file) |
1da177e4 LT |
2890 | get_file(new_vma->vm_file); |
2891 | if (new_vma->vm_ops && new_vma->vm_ops->open) | |
2892 | new_vma->vm_ops->open(new_vma); | |
2893 | vma_link(mm, new_vma, prev, rb_link, rb_parent); | |
38a76013 | 2894 | *need_rmap_locks = false; |
1da177e4 LT |
2895 | } |
2896 | } | |
2897 | return new_vma; | |
5beb4930 RR |
2898 | |
2899 | out_free_mempol: | |
2900 | mpol_put(pol); | |
2901 | out_free_vma: | |
2902 | kmem_cache_free(vm_area_cachep, new_vma); | |
2903 | return NULL; | |
1da177e4 | 2904 | } |
119f657c AM |
2905 | |
2906 | /* | |
2907 | * Return true if the calling process may expand its vm space by the passed | |
2908 | * number of pages | |
2909 | */ | |
2910 | int may_expand_vm(struct mm_struct *mm, unsigned long npages) | |
2911 | { | |
2912 | unsigned long cur = mm->total_vm; /* pages */ | |
2913 | unsigned long lim; | |
2914 | ||
59e99e5b | 2915 | lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT; |
119f657c AM |
2916 | |
2917 | if (cur + npages > lim) | |
2918 | return 0; | |
2919 | return 1; | |
2920 | } | |
fa5dc22f RM |
2921 | |
2922 | ||
b1d0e4f5 NP |
2923 | static int special_mapping_fault(struct vm_area_struct *vma, |
2924 | struct vm_fault *vmf) | |
fa5dc22f | 2925 | { |
b1d0e4f5 | 2926 | pgoff_t pgoff; |
fa5dc22f RM |
2927 | struct page **pages; |
2928 | ||
b1d0e4f5 NP |
2929 | /* |
2930 | * special mappings have no vm_file, and in that case, the mm | |
2931 | * uses vm_pgoff internally. So we have to subtract it from here. | |
2932 | * We are allowed to do this because we are the mm; do not copy | |
2933 | * this code into drivers! | |
2934 | */ | |
2935 | pgoff = vmf->pgoff - vma->vm_pgoff; | |
fa5dc22f | 2936 | |
b1d0e4f5 NP |
2937 | for (pages = vma->vm_private_data; pgoff && *pages; ++pages) |
2938 | pgoff--; | |
fa5dc22f RM |
2939 | |
2940 | if (*pages) { | |
2941 | struct page *page = *pages; | |
2942 | get_page(page); | |
b1d0e4f5 NP |
2943 | vmf->page = page; |
2944 | return 0; | |
fa5dc22f RM |
2945 | } |
2946 | ||
b1d0e4f5 | 2947 | return VM_FAULT_SIGBUS; |
fa5dc22f RM |
2948 | } |
2949 | ||
2950 | /* | |
2951 | * Having a close hook prevents vma merging regardless of flags. | |
2952 | */ | |
2953 | static void special_mapping_close(struct vm_area_struct *vma) | |
2954 | { | |
2955 | } | |
2956 | ||
f0f37e2f | 2957 | static const struct vm_operations_struct special_mapping_vmops = { |
fa5dc22f | 2958 | .close = special_mapping_close, |
b1d0e4f5 | 2959 | .fault = special_mapping_fault, |
fa5dc22f RM |
2960 | }; |
2961 | ||
2962 | /* | |
2963 | * Called with mm->mmap_sem held for writing. | |
2964 | * Insert a new vma covering the given region, with the given flags. | |
2965 | * Its pages are supplied by the given array of struct page *. | |
2966 | * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. | |
2967 | * The region past the last page supplied will always produce SIGBUS. | |
2968 | * The array pointer and the pages it points to are assumed to stay alive | |
2969 | * for as long as this mapping might exist. | |
2970 | */ | |
2971 | int install_special_mapping(struct mm_struct *mm, | |
2972 | unsigned long addr, unsigned long len, | |
2973 | unsigned long vm_flags, struct page **pages) | |
2974 | { | |
462e635e | 2975 | int ret; |
fa5dc22f RM |
2976 | struct vm_area_struct *vma; |
2977 | ||
2978 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
2979 | if (unlikely(vma == NULL)) | |
2980 | return -ENOMEM; | |
2981 | ||
5beb4930 | 2982 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
fa5dc22f RM |
2983 | vma->vm_mm = mm; |
2984 | vma->vm_start = addr; | |
2985 | vma->vm_end = addr + len; | |
2986 | ||
2f98735c | 2987 | vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND; |
3ed75eb8 | 2988 | vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); |
fa5dc22f RM |
2989 | |
2990 | vma->vm_ops = &special_mapping_vmops; | |
2991 | vma->vm_private_data = pages; | |
2992 | ||
462e635e TO |
2993 | ret = insert_vm_struct(mm, vma); |
2994 | if (ret) | |
2995 | goto out; | |
fa5dc22f RM |
2996 | |
2997 | mm->total_vm += len >> PAGE_SHIFT; | |
2998 | ||
cdd6c482 | 2999 | perf_event_mmap(vma); |
089dd79d | 3000 | |
fa5dc22f | 3001 | return 0; |
462e635e TO |
3002 | |
3003 | out: | |
3004 | kmem_cache_free(vm_area_cachep, vma); | |
3005 | return ret; | |
fa5dc22f | 3006 | } |
7906d00c AA |
3007 | |
3008 | static DEFINE_MUTEX(mm_all_locks_mutex); | |
3009 | ||
454ed842 | 3010 | static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) |
7906d00c | 3011 | { |
bf181b9f | 3012 | if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
7906d00c AA |
3013 | /* |
3014 | * The LSB of head.next can't change from under us | |
3015 | * because we hold the mm_all_locks_mutex. | |
3016 | */ | |
572043c9 | 3017 | down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); |
7906d00c AA |
3018 | /* |
3019 | * We can safely modify head.next after taking the | |
5a505085 | 3020 | * anon_vma->root->rwsem. If some other vma in this mm shares |
7906d00c AA |
3021 | * the same anon_vma we won't take it again. |
3022 | * | |
3023 | * No need of atomic instructions here, head.next | |
3024 | * can't change from under us thanks to the | |
5a505085 | 3025 | * anon_vma->root->rwsem. |
7906d00c AA |
3026 | */ |
3027 | if (__test_and_set_bit(0, (unsigned long *) | |
bf181b9f | 3028 | &anon_vma->root->rb_root.rb_node)) |
7906d00c AA |
3029 | BUG(); |
3030 | } | |
3031 | } | |
3032 | ||
454ed842 | 3033 | static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) |
7906d00c AA |
3034 | { |
3035 | if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3036 | /* | |
3037 | * AS_MM_ALL_LOCKS can't change from under us because | |
3038 | * we hold the mm_all_locks_mutex. | |
3039 | * | |
3040 | * Operations on ->flags have to be atomic because | |
3041 | * even if AS_MM_ALL_LOCKS is stable thanks to the | |
3042 | * mm_all_locks_mutex, there may be other cpus | |
3043 | * changing other bitflags in parallel to us. | |
3044 | */ | |
3045 | if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) | |
3046 | BUG(); | |
3d48ae45 | 3047 | mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem); |
7906d00c AA |
3048 | } |
3049 | } | |
3050 | ||
3051 | /* | |
3052 | * This operation locks against the VM for all pte/vma/mm related | |
3053 | * operations that could ever happen on a certain mm. This includes | |
3054 | * vmtruncate, try_to_unmap, and all page faults. | |
3055 | * | |
3056 | * The caller must take the mmap_sem in write mode before calling | |
3057 | * mm_take_all_locks(). The caller isn't allowed to release the | |
3058 | * mmap_sem until mm_drop_all_locks() returns. | |
3059 | * | |
3060 | * mmap_sem in write mode is required in order to block all operations | |
3061 | * that could modify pagetables and free pages without need of | |
3062 | * altering the vma layout (for example populate_range() with | |
3063 | * nonlinear vmas). It's also needed in write mode to avoid new | |
3064 | * anon_vmas to be associated with existing vmas. | |
3065 | * | |
3066 | * A single task can't take more than one mm_take_all_locks() in a row | |
3067 | * or it would deadlock. | |
3068 | * | |
bf181b9f | 3069 | * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in |
7906d00c AA |
3070 | * mapping->flags avoid to take the same lock twice, if more than one |
3071 | * vma in this mm is backed by the same anon_vma or address_space. | |
3072 | * | |
3073 | * We can take all the locks in random order because the VM code | |
631b0cfd | 3074 | * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never |
7906d00c AA |
3075 | * takes more than one of them in a row. Secondly we're protected |
3076 | * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex. | |
3077 | * | |
3078 | * mm_take_all_locks() and mm_drop_all_locks are expensive operations | |
3079 | * that may have to take thousand of locks. | |
3080 | * | |
3081 | * mm_take_all_locks() can fail if it's interrupted by signals. | |
3082 | */ | |
3083 | int mm_take_all_locks(struct mm_struct *mm) | |
3084 | { | |
3085 | struct vm_area_struct *vma; | |
5beb4930 | 3086 | struct anon_vma_chain *avc; |
7906d00c AA |
3087 | |
3088 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
3089 | ||
3090 | mutex_lock(&mm_all_locks_mutex); | |
3091 | ||
3092 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3093 | if (signal_pending(current)) | |
3094 | goto out_unlock; | |
7906d00c | 3095 | if (vma->vm_file && vma->vm_file->f_mapping) |
454ed842 | 3096 | vm_lock_mapping(mm, vma->vm_file->f_mapping); |
7906d00c | 3097 | } |
7cd5a02f PZ |
3098 | |
3099 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3100 | if (signal_pending(current)) | |
3101 | goto out_unlock; | |
3102 | if (vma->anon_vma) | |
5beb4930 RR |
3103 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3104 | vm_lock_anon_vma(mm, avc->anon_vma); | |
7906d00c | 3105 | } |
7cd5a02f | 3106 | |
584cff54 | 3107 | return 0; |
7906d00c AA |
3108 | |
3109 | out_unlock: | |
584cff54 KC |
3110 | mm_drop_all_locks(mm); |
3111 | return -EINTR; | |
7906d00c AA |
3112 | } |
3113 | ||
3114 | static void vm_unlock_anon_vma(struct anon_vma *anon_vma) | |
3115 | { | |
bf181b9f | 3116 | if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
7906d00c AA |
3117 | /* |
3118 | * The LSB of head.next can't change to 0 from under | |
3119 | * us because we hold the mm_all_locks_mutex. | |
3120 | * | |
3121 | * We must however clear the bitflag before unlocking | |
bf181b9f | 3122 | * the vma so the users using the anon_vma->rb_root will |
7906d00c AA |
3123 | * never see our bitflag. |
3124 | * | |
3125 | * No need of atomic instructions here, head.next | |
3126 | * can't change from under us until we release the | |
5a505085 | 3127 | * anon_vma->root->rwsem. |
7906d00c AA |
3128 | */ |
3129 | if (!__test_and_clear_bit(0, (unsigned long *) | |
bf181b9f | 3130 | &anon_vma->root->rb_root.rb_node)) |
7906d00c | 3131 | BUG(); |
08b52706 | 3132 | anon_vma_unlock_write(anon_vma); |
7906d00c AA |
3133 | } |
3134 | } | |
3135 | ||
3136 | static void vm_unlock_mapping(struct address_space *mapping) | |
3137 | { | |
3138 | if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3139 | /* | |
3140 | * AS_MM_ALL_LOCKS can't change to 0 from under us | |
3141 | * because we hold the mm_all_locks_mutex. | |
3142 | */ | |
3d48ae45 | 3143 | mutex_unlock(&mapping->i_mmap_mutex); |
7906d00c AA |
3144 | if (!test_and_clear_bit(AS_MM_ALL_LOCKS, |
3145 | &mapping->flags)) | |
3146 | BUG(); | |
3147 | } | |
3148 | } | |
3149 | ||
3150 | /* | |
3151 | * The mmap_sem cannot be released by the caller until | |
3152 | * mm_drop_all_locks() returns. | |
3153 | */ | |
3154 | void mm_drop_all_locks(struct mm_struct *mm) | |
3155 | { | |
3156 | struct vm_area_struct *vma; | |
5beb4930 | 3157 | struct anon_vma_chain *avc; |
7906d00c AA |
3158 | |
3159 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
3160 | BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); | |
3161 | ||
3162 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3163 | if (vma->anon_vma) | |
5beb4930 RR |
3164 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3165 | vm_unlock_anon_vma(avc->anon_vma); | |
7906d00c AA |
3166 | if (vma->vm_file && vma->vm_file->f_mapping) |
3167 | vm_unlock_mapping(vma->vm_file->f_mapping); | |
3168 | } | |
3169 | ||
3170 | mutex_unlock(&mm_all_locks_mutex); | |
3171 | } | |
8feae131 DH |
3172 | |
3173 | /* | |
3174 | * initialise the VMA slab | |
3175 | */ | |
3176 | void __init mmap_init(void) | |
3177 | { | |
00a62ce9 KM |
3178 | int ret; |
3179 | ||
3180 | ret = percpu_counter_init(&vm_committed_as, 0); | |
3181 | VM_BUG_ON(ret); | |
8feae131 | 3182 | } |
c9b1d098 AS |
3183 | |
3184 | /* | |
3185 | * Initialise sysctl_user_reserve_kbytes. | |
3186 | * | |
3187 | * This is intended to prevent a user from starting a single memory hogging | |
3188 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER | |
3189 | * mode. | |
3190 | * | |
3191 | * The default value is min(3% of free memory, 128MB) | |
3192 | * 128MB is enough to recover with sshd/login, bash, and top/kill. | |
3193 | */ | |
1640879a | 3194 | static int init_user_reserve(void) |
c9b1d098 AS |
3195 | { |
3196 | unsigned long free_kbytes; | |
3197 | ||
3198 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3199 | ||
3200 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); | |
3201 | return 0; | |
3202 | } | |
3203 | module_init(init_user_reserve) | |
4eeab4f5 AS |
3204 | |
3205 | /* | |
3206 | * Initialise sysctl_admin_reserve_kbytes. | |
3207 | * | |
3208 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin | |
3209 | * to log in and kill a memory hogging process. | |
3210 | * | |
3211 | * Systems with more than 256MB will reserve 8MB, enough to recover | |
3212 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will | |
3213 | * only reserve 3% of free pages by default. | |
3214 | */ | |
1640879a | 3215 | static int init_admin_reserve(void) |
4eeab4f5 AS |
3216 | { |
3217 | unsigned long free_kbytes; | |
3218 | ||
3219 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3220 | ||
3221 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); | |
3222 | return 0; | |
3223 | } | |
3224 | module_init(init_admin_reserve) | |
1640879a AS |
3225 | |
3226 | /* | |
3227 | * Reinititalise user and admin reserves if memory is added or removed. | |
3228 | * | |
3229 | * The default user reserve max is 128MB, and the default max for the | |
3230 | * admin reserve is 8MB. These are usually, but not always, enough to | |
3231 | * enable recovery from a memory hogging process using login/sshd, a shell, | |
3232 | * and tools like top. It may make sense to increase or even disable the | |
3233 | * reserve depending on the existence of swap or variations in the recovery | |
3234 | * tools. So, the admin may have changed them. | |
3235 | * | |
3236 | * If memory is added and the reserves have been eliminated or increased above | |
3237 | * the default max, then we'll trust the admin. | |
3238 | * | |
3239 | * If memory is removed and there isn't enough free memory, then we | |
3240 | * need to reset the reserves. | |
3241 | * | |
3242 | * Otherwise keep the reserve set by the admin. | |
3243 | */ | |
3244 | static int reserve_mem_notifier(struct notifier_block *nb, | |
3245 | unsigned long action, void *data) | |
3246 | { | |
3247 | unsigned long tmp, free_kbytes; | |
3248 | ||
3249 | switch (action) { | |
3250 | case MEM_ONLINE: | |
3251 | /* Default max is 128MB. Leave alone if modified by operator. */ | |
3252 | tmp = sysctl_user_reserve_kbytes; | |
3253 | if (0 < tmp && tmp < (1UL << 17)) | |
3254 | init_user_reserve(); | |
3255 | ||
3256 | /* Default max is 8MB. Leave alone if modified by operator. */ | |
3257 | tmp = sysctl_admin_reserve_kbytes; | |
3258 | if (0 < tmp && tmp < (1UL << 13)) | |
3259 | init_admin_reserve(); | |
3260 | ||
3261 | break; | |
3262 | case MEM_OFFLINE: | |
3263 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3264 | ||
3265 | if (sysctl_user_reserve_kbytes > free_kbytes) { | |
3266 | init_user_reserve(); | |
3267 | pr_info("vm.user_reserve_kbytes reset to %lu\n", | |
3268 | sysctl_user_reserve_kbytes); | |
3269 | } | |
3270 | ||
3271 | if (sysctl_admin_reserve_kbytes > free_kbytes) { | |
3272 | init_admin_reserve(); | |
3273 | pr_info("vm.admin_reserve_kbytes reset to %lu\n", | |
3274 | sysctl_admin_reserve_kbytes); | |
3275 | } | |
3276 | break; | |
3277 | default: | |
3278 | break; | |
3279 | } | |
3280 | return NOTIFY_OK; | |
3281 | } | |
3282 | ||
3283 | static struct notifier_block reserve_mem_nb = { | |
3284 | .notifier_call = reserve_mem_notifier, | |
3285 | }; | |
3286 | ||
3287 | static int __meminit init_reserve_notifier(void) | |
3288 | { | |
3289 | if (register_hotmemory_notifier(&reserve_mem_nb)) | |
3290 | printk("Failed registering memory add/remove notifier for admin reserve"); | |
3291 | ||
3292 | return 0; | |
3293 | } | |
3294 | module_init(init_reserve_notifier) |