Commit | Line | Data |
---|---|---|
16d69265 | 1 | #include <linux/mm.h> |
30992c97 MM |
2 | #include <linux/slab.h> |
3 | #include <linux/string.h> | |
3b32123d | 4 | #include <linux/compiler.h> |
b95f1b31 | 5 | #include <linux/export.h> |
96840aa0 | 6 | #include <linux/err.h> |
3b8f14b4 | 7 | #include <linux/sched.h> |
6e84f315 | 8 | #include <linux/sched/mm.h> |
68db0cf1 | 9 | #include <linux/sched/task_stack.h> |
eb36c587 | 10 | #include <linux/security.h> |
9800339b | 11 | #include <linux/swap.h> |
33806f06 | 12 | #include <linux/swapops.h> |
00619bcc JM |
13 | #include <linux/mman.h> |
14 | #include <linux/hugetlb.h> | |
39f1f78d | 15 | #include <linux/vmalloc.h> |
897ab3e0 | 16 | #include <linux/userfaultfd_k.h> |
00619bcc | 17 | |
a4bb1e43 | 18 | #include <asm/sections.h> |
7c0f6ba6 | 19 | #include <linux/uaccess.h> |
30992c97 | 20 | |
6038def0 NK |
21 | #include "internal.h" |
22 | ||
a4bb1e43 AH |
23 | static inline int is_kernel_rodata(unsigned long addr) |
24 | { | |
25 | return addr >= (unsigned long)__start_rodata && | |
26 | addr < (unsigned long)__end_rodata; | |
27 | } | |
28 | ||
29 | /** | |
30 | * kfree_const - conditionally free memory | |
31 | * @x: pointer to the memory | |
32 | * | |
33 | * Function calls kfree only if @x is not in .rodata section. | |
34 | */ | |
35 | void kfree_const(const void *x) | |
36 | { | |
37 | if (!is_kernel_rodata((unsigned long)x)) | |
38 | kfree(x); | |
39 | } | |
40 | EXPORT_SYMBOL(kfree_const); | |
41 | ||
30992c97 | 42 | /** |
30992c97 | 43 | * kstrdup - allocate space for and copy an existing string |
30992c97 MM |
44 | * @s: the string to duplicate |
45 | * @gfp: the GFP mask used in the kmalloc() call when allocating memory | |
46 | */ | |
47 | char *kstrdup(const char *s, gfp_t gfp) | |
48 | { | |
49 | size_t len; | |
50 | char *buf; | |
51 | ||
52 | if (!s) | |
53 | return NULL; | |
54 | ||
55 | len = strlen(s) + 1; | |
1d2c8eea | 56 | buf = kmalloc_track_caller(len, gfp); |
30992c97 MM |
57 | if (buf) |
58 | memcpy(buf, s, len); | |
59 | return buf; | |
60 | } | |
61 | EXPORT_SYMBOL(kstrdup); | |
96840aa0 | 62 | |
a4bb1e43 AH |
63 | /** |
64 | * kstrdup_const - conditionally duplicate an existing const string | |
65 | * @s: the string to duplicate | |
66 | * @gfp: the GFP mask used in the kmalloc() call when allocating memory | |
67 | * | |
68 | * Function returns source string if it is in .rodata section otherwise it | |
69 | * fallbacks to kstrdup. | |
70 | * Strings allocated by kstrdup_const should be freed by kfree_const. | |
71 | */ | |
72 | const char *kstrdup_const(const char *s, gfp_t gfp) | |
73 | { | |
74 | if (is_kernel_rodata((unsigned long)s)) | |
75 | return s; | |
76 | ||
77 | return kstrdup(s, gfp); | |
78 | } | |
79 | EXPORT_SYMBOL(kstrdup_const); | |
80 | ||
1e66df3e JF |
81 | /** |
82 | * kstrndup - allocate space for and copy an existing string | |
83 | * @s: the string to duplicate | |
84 | * @max: read at most @max chars from @s | |
85 | * @gfp: the GFP mask used in the kmalloc() call when allocating memory | |
f3515741 DH |
86 | * |
87 | * Note: Use kmemdup_nul() instead if the size is known exactly. | |
1e66df3e JF |
88 | */ |
89 | char *kstrndup(const char *s, size_t max, gfp_t gfp) | |
90 | { | |
91 | size_t len; | |
92 | char *buf; | |
93 | ||
94 | if (!s) | |
95 | return NULL; | |
96 | ||
97 | len = strnlen(s, max); | |
98 | buf = kmalloc_track_caller(len+1, gfp); | |
99 | if (buf) { | |
100 | memcpy(buf, s, len); | |
101 | buf[len] = '\0'; | |
102 | } | |
103 | return buf; | |
104 | } | |
105 | EXPORT_SYMBOL(kstrndup); | |
106 | ||
1a2f67b4 AD |
107 | /** |
108 | * kmemdup - duplicate region of memory | |
109 | * | |
110 | * @src: memory region to duplicate | |
111 | * @len: memory region length | |
112 | * @gfp: GFP mask to use | |
113 | */ | |
114 | void *kmemdup(const void *src, size_t len, gfp_t gfp) | |
115 | { | |
116 | void *p; | |
117 | ||
1d2c8eea | 118 | p = kmalloc_track_caller(len, gfp); |
1a2f67b4 AD |
119 | if (p) |
120 | memcpy(p, src, len); | |
121 | return p; | |
122 | } | |
123 | EXPORT_SYMBOL(kmemdup); | |
124 | ||
f3515741 DH |
125 | /** |
126 | * kmemdup_nul - Create a NUL-terminated string from unterminated data | |
127 | * @s: The data to stringify | |
128 | * @len: The size of the data | |
129 | * @gfp: the GFP mask used in the kmalloc() call when allocating memory | |
130 | */ | |
131 | char *kmemdup_nul(const char *s, size_t len, gfp_t gfp) | |
132 | { | |
133 | char *buf; | |
134 | ||
135 | if (!s) | |
136 | return NULL; | |
137 | ||
138 | buf = kmalloc_track_caller(len + 1, gfp); | |
139 | if (buf) { | |
140 | memcpy(buf, s, len); | |
141 | buf[len] = '\0'; | |
142 | } | |
143 | return buf; | |
144 | } | |
145 | EXPORT_SYMBOL(kmemdup_nul); | |
146 | ||
610a77e0 LZ |
147 | /** |
148 | * memdup_user - duplicate memory region from user space | |
149 | * | |
150 | * @src: source address in user space | |
151 | * @len: number of bytes to copy | |
152 | * | |
50fd2f29 AV |
153 | * Returns an ERR_PTR() on failure. Result is physically |
154 | * contiguous, to be freed by kfree(). | |
610a77e0 LZ |
155 | */ |
156 | void *memdup_user(const void __user *src, size_t len) | |
157 | { | |
158 | void *p; | |
159 | ||
6c2c97a2 | 160 | p = kmalloc_track_caller(len, GFP_USER); |
610a77e0 LZ |
161 | if (!p) |
162 | return ERR_PTR(-ENOMEM); | |
163 | ||
164 | if (copy_from_user(p, src, len)) { | |
165 | kfree(p); | |
166 | return ERR_PTR(-EFAULT); | |
167 | } | |
168 | ||
169 | return p; | |
170 | } | |
171 | EXPORT_SYMBOL(memdup_user); | |
172 | ||
50fd2f29 AV |
173 | /** |
174 | * vmemdup_user - duplicate memory region from user space | |
175 | * | |
176 | * @src: source address in user space | |
177 | * @len: number of bytes to copy | |
178 | * | |
179 | * Returns an ERR_PTR() on failure. Result may be not | |
180 | * physically contiguous. Use kvfree() to free. | |
181 | */ | |
182 | void *vmemdup_user(const void __user *src, size_t len) | |
183 | { | |
184 | void *p; | |
185 | ||
186 | p = kvmalloc(len, GFP_USER); | |
187 | if (!p) | |
188 | return ERR_PTR(-ENOMEM); | |
189 | ||
190 | if (copy_from_user(p, src, len)) { | |
191 | kvfree(p); | |
192 | return ERR_PTR(-EFAULT); | |
193 | } | |
194 | ||
195 | return p; | |
196 | } | |
197 | EXPORT_SYMBOL(vmemdup_user); | |
198 | ||
96840aa0 DA |
199 | /* |
200 | * strndup_user - duplicate an existing string from user space | |
96840aa0 DA |
201 | * @s: The string to duplicate |
202 | * @n: Maximum number of bytes to copy, including the trailing NUL. | |
203 | */ | |
204 | char *strndup_user(const char __user *s, long n) | |
205 | { | |
206 | char *p; | |
207 | long length; | |
208 | ||
209 | length = strnlen_user(s, n); | |
210 | ||
211 | if (!length) | |
212 | return ERR_PTR(-EFAULT); | |
213 | ||
214 | if (length > n) | |
215 | return ERR_PTR(-EINVAL); | |
216 | ||
90d74045 | 217 | p = memdup_user(s, length); |
96840aa0 | 218 | |
90d74045 JL |
219 | if (IS_ERR(p)) |
220 | return p; | |
96840aa0 DA |
221 | |
222 | p[length - 1] = '\0'; | |
223 | ||
224 | return p; | |
225 | } | |
226 | EXPORT_SYMBOL(strndup_user); | |
16d69265 | 227 | |
e9d408e1 AV |
228 | /** |
229 | * memdup_user_nul - duplicate memory region from user space and NUL-terminate | |
230 | * | |
231 | * @src: source address in user space | |
232 | * @len: number of bytes to copy | |
233 | * | |
234 | * Returns an ERR_PTR() on failure. | |
235 | */ | |
236 | void *memdup_user_nul(const void __user *src, size_t len) | |
237 | { | |
238 | char *p; | |
239 | ||
240 | /* | |
241 | * Always use GFP_KERNEL, since copy_from_user() can sleep and | |
242 | * cause pagefault, which makes it pointless to use GFP_NOFS | |
243 | * or GFP_ATOMIC. | |
244 | */ | |
245 | p = kmalloc_track_caller(len + 1, GFP_KERNEL); | |
246 | if (!p) | |
247 | return ERR_PTR(-ENOMEM); | |
248 | ||
249 | if (copy_from_user(p, src, len)) { | |
250 | kfree(p); | |
251 | return ERR_PTR(-EFAULT); | |
252 | } | |
253 | p[len] = '\0'; | |
254 | ||
255 | return p; | |
256 | } | |
257 | EXPORT_SYMBOL(memdup_user_nul); | |
258 | ||
6038def0 NK |
259 | void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma, |
260 | struct vm_area_struct *prev, struct rb_node *rb_parent) | |
261 | { | |
262 | struct vm_area_struct *next; | |
263 | ||
264 | vma->vm_prev = prev; | |
265 | if (prev) { | |
266 | next = prev->vm_next; | |
267 | prev->vm_next = vma; | |
268 | } else { | |
269 | mm->mmap = vma; | |
270 | if (rb_parent) | |
271 | next = rb_entry(rb_parent, | |
272 | struct vm_area_struct, vm_rb); | |
273 | else | |
274 | next = NULL; | |
275 | } | |
276 | vma->vm_next = next; | |
277 | if (next) | |
278 | next->vm_prev = vma; | |
279 | } | |
280 | ||
b7643757 | 281 | /* Check if the vma is being used as a stack by this task */ |
d17af505 | 282 | int vma_is_stack_for_current(struct vm_area_struct *vma) |
b7643757 | 283 | { |
d17af505 AL |
284 | struct task_struct * __maybe_unused t = current; |
285 | ||
b7643757 SP |
286 | return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t)); |
287 | } | |
288 | ||
efc1a3b1 | 289 | #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT) |
16d69265 AM |
290 | void arch_pick_mmap_layout(struct mm_struct *mm) |
291 | { | |
292 | mm->mmap_base = TASK_UNMAPPED_BASE; | |
293 | mm->get_unmapped_area = arch_get_unmapped_area; | |
16d69265 AM |
294 | } |
295 | #endif | |
912985dc | 296 | |
45888a0c XG |
297 | /* |
298 | * Like get_user_pages_fast() except its IRQ-safe in that it won't fall | |
299 | * back to the regular GUP. | |
25985edc | 300 | * If the architecture not support this function, simply return with no |
45888a0c XG |
301 | * page pinned |
302 | */ | |
3b32123d | 303 | int __weak __get_user_pages_fast(unsigned long start, |
45888a0c XG |
304 | int nr_pages, int write, struct page **pages) |
305 | { | |
306 | return 0; | |
307 | } | |
308 | EXPORT_SYMBOL_GPL(__get_user_pages_fast); | |
309 | ||
9de100d0 AG |
310 | /** |
311 | * get_user_pages_fast() - pin user pages in memory | |
312 | * @start: starting user address | |
313 | * @nr_pages: number of pages from start to pin | |
314 | * @write: whether pages will be written to | |
315 | * @pages: array that receives pointers to the pages pinned. | |
316 | * Should be at least nr_pages long. | |
317 | * | |
9de100d0 AG |
318 | * Returns number of pages pinned. This may be fewer than the number |
319 | * requested. If nr_pages is 0 or negative, returns 0. If no pages | |
320 | * were pinned, returns -errno. | |
d2bf6be8 NP |
321 | * |
322 | * get_user_pages_fast provides equivalent functionality to get_user_pages, | |
323 | * operating on current and current->mm, with force=0 and vma=NULL. However | |
324 | * unlike get_user_pages, it must be called without mmap_sem held. | |
325 | * | |
326 | * get_user_pages_fast may take mmap_sem and page table locks, so no | |
327 | * assumptions can be made about lack of locking. get_user_pages_fast is to be | |
328 | * implemented in a way that is advantageous (vs get_user_pages()) when the | |
329 | * user memory area is already faulted in and present in ptes. However if the | |
330 | * pages have to be faulted in, it may turn out to be slightly slower so | |
331 | * callers need to carefully consider what to use. On many architectures, | |
332 | * get_user_pages_fast simply falls back to get_user_pages. | |
9de100d0 | 333 | */ |
3b32123d | 334 | int __weak get_user_pages_fast(unsigned long start, |
912985dc RR |
335 | int nr_pages, int write, struct page **pages) |
336 | { | |
c164154f LS |
337 | return get_user_pages_unlocked(start, nr_pages, pages, |
338 | write ? FOLL_WRITE : 0); | |
912985dc RR |
339 | } |
340 | EXPORT_SYMBOL_GPL(get_user_pages_fast); | |
ca2b84cb | 341 | |
eb36c587 AV |
342 | unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr, |
343 | unsigned long len, unsigned long prot, | |
9fbeb5ab | 344 | unsigned long flag, unsigned long pgoff) |
eb36c587 AV |
345 | { |
346 | unsigned long ret; | |
347 | struct mm_struct *mm = current->mm; | |
41badc15 | 348 | unsigned long populate; |
897ab3e0 | 349 | LIST_HEAD(uf); |
eb36c587 AV |
350 | |
351 | ret = security_mmap_file(file, prot, flag); | |
352 | if (!ret) { | |
9fbeb5ab MH |
353 | if (down_write_killable(&mm->mmap_sem)) |
354 | return -EINTR; | |
bebeb3d6 | 355 | ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff, |
897ab3e0 | 356 | &populate, &uf); |
eb36c587 | 357 | up_write(&mm->mmap_sem); |
897ab3e0 | 358 | userfaultfd_unmap_complete(mm, &uf); |
41badc15 ML |
359 | if (populate) |
360 | mm_populate(ret, populate); | |
eb36c587 AV |
361 | } |
362 | return ret; | |
363 | } | |
364 | ||
365 | unsigned long vm_mmap(struct file *file, unsigned long addr, | |
366 | unsigned long len, unsigned long prot, | |
367 | unsigned long flag, unsigned long offset) | |
368 | { | |
369 | if (unlikely(offset + PAGE_ALIGN(len) < offset)) | |
370 | return -EINVAL; | |
ea53cde0 | 371 | if (unlikely(offset_in_page(offset))) |
eb36c587 AV |
372 | return -EINVAL; |
373 | ||
9fbeb5ab | 374 | return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT); |
eb36c587 AV |
375 | } |
376 | EXPORT_SYMBOL(vm_mmap); | |
377 | ||
a7c3e901 MH |
378 | /** |
379 | * kvmalloc_node - attempt to allocate physically contiguous memory, but upon | |
380 | * failure, fall back to non-contiguous (vmalloc) allocation. | |
381 | * @size: size of the request. | |
382 | * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL. | |
383 | * @node: numa node to allocate from | |
384 | * | |
385 | * Uses kmalloc to get the memory but if the allocation fails then falls back | |
386 | * to the vmalloc allocator. Use kvfree for freeing the memory. | |
387 | * | |
cc965a29 MH |
388 | * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported. |
389 | * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is | |
390 | * preferable to the vmalloc fallback, due to visible performance drawbacks. | |
a7c3e901 MH |
391 | * |
392 | * Any use of gfp flags outside of GFP_KERNEL should be consulted with mm people. | |
393 | */ | |
394 | void *kvmalloc_node(size_t size, gfp_t flags, int node) | |
395 | { | |
396 | gfp_t kmalloc_flags = flags; | |
397 | void *ret; | |
398 | ||
399 | /* | |
400 | * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables) | |
401 | * so the given set of flags has to be compatible. | |
402 | */ | |
403 | WARN_ON_ONCE((flags & GFP_KERNEL) != GFP_KERNEL); | |
404 | ||
405 | /* | |
4f4f2ba9 MH |
406 | * We want to attempt a large physically contiguous block first because |
407 | * it is less likely to fragment multiple larger blocks and therefore | |
408 | * contribute to a long term fragmentation less than vmalloc fallback. | |
409 | * However make sure that larger requests are not too disruptive - no | |
410 | * OOM killer and no allocation failure warnings as we have a fallback. | |
a7c3e901 | 411 | */ |
6c5ab651 MH |
412 | if (size > PAGE_SIZE) { |
413 | kmalloc_flags |= __GFP_NOWARN; | |
414 | ||
cc965a29 | 415 | if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL)) |
6c5ab651 MH |
416 | kmalloc_flags |= __GFP_NORETRY; |
417 | } | |
a7c3e901 MH |
418 | |
419 | ret = kmalloc_node(size, kmalloc_flags, node); | |
420 | ||
421 | /* | |
422 | * It doesn't really make sense to fallback to vmalloc for sub page | |
423 | * requests | |
424 | */ | |
425 | if (ret || size <= PAGE_SIZE) | |
426 | return ret; | |
427 | ||
8594a21c MH |
428 | return __vmalloc_node_flags_caller(size, node, flags, |
429 | __builtin_return_address(0)); | |
a7c3e901 MH |
430 | } |
431 | EXPORT_SYMBOL(kvmalloc_node); | |
432 | ||
39f1f78d AV |
433 | void kvfree(const void *addr) |
434 | { | |
435 | if (is_vmalloc_addr(addr)) | |
436 | vfree(addr); | |
437 | else | |
438 | kfree(addr); | |
439 | } | |
440 | EXPORT_SYMBOL(kvfree); | |
441 | ||
e39155ea KS |
442 | static inline void *__page_rmapping(struct page *page) |
443 | { | |
444 | unsigned long mapping; | |
445 | ||
446 | mapping = (unsigned long)page->mapping; | |
447 | mapping &= ~PAGE_MAPPING_FLAGS; | |
448 | ||
449 | return (void *)mapping; | |
450 | } | |
451 | ||
452 | /* Neutral page->mapping pointer to address_space or anon_vma or other */ | |
453 | void *page_rmapping(struct page *page) | |
454 | { | |
455 | page = compound_head(page); | |
456 | return __page_rmapping(page); | |
457 | } | |
458 | ||
1aa8aea5 AM |
459 | /* |
460 | * Return true if this page is mapped into pagetables. | |
461 | * For compound page it returns true if any subpage of compound page is mapped. | |
462 | */ | |
463 | bool page_mapped(struct page *page) | |
464 | { | |
465 | int i; | |
466 | ||
467 | if (likely(!PageCompound(page))) | |
468 | return atomic_read(&page->_mapcount) >= 0; | |
469 | page = compound_head(page); | |
470 | if (atomic_read(compound_mapcount_ptr(page)) >= 0) | |
471 | return true; | |
472 | if (PageHuge(page)) | |
473 | return false; | |
474 | for (i = 0; i < hpage_nr_pages(page); i++) { | |
475 | if (atomic_read(&page[i]._mapcount) >= 0) | |
476 | return true; | |
477 | } | |
478 | return false; | |
479 | } | |
480 | EXPORT_SYMBOL(page_mapped); | |
481 | ||
e39155ea KS |
482 | struct anon_vma *page_anon_vma(struct page *page) |
483 | { | |
484 | unsigned long mapping; | |
485 | ||
486 | page = compound_head(page); | |
487 | mapping = (unsigned long)page->mapping; | |
488 | if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) | |
489 | return NULL; | |
490 | return __page_rmapping(page); | |
491 | } | |
492 | ||
9800339b SL |
493 | struct address_space *page_mapping(struct page *page) |
494 | { | |
1c290f64 KS |
495 | struct address_space *mapping; |
496 | ||
497 | page = compound_head(page); | |
9800339b | 498 | |
03e5ac2f MP |
499 | /* This happens if someone calls flush_dcache_page on slab page */ |
500 | if (unlikely(PageSlab(page))) | |
501 | return NULL; | |
502 | ||
33806f06 SL |
503 | if (unlikely(PageSwapCache(page))) { |
504 | swp_entry_t entry; | |
505 | ||
506 | entry.val = page_private(page); | |
e39155ea KS |
507 | return swap_address_space(entry); |
508 | } | |
509 | ||
1c290f64 | 510 | mapping = page->mapping; |
bda807d4 | 511 | if ((unsigned long)mapping & PAGE_MAPPING_ANON) |
e39155ea | 512 | return NULL; |
bda807d4 MK |
513 | |
514 | return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS); | |
9800339b | 515 | } |
bda807d4 | 516 | EXPORT_SYMBOL(page_mapping); |
9800339b | 517 | |
cb9f753a HY |
518 | /* |
519 | * For file cache pages, return the address_space, otherwise return NULL | |
520 | */ | |
521 | struct address_space *page_mapping_file(struct page *page) | |
522 | { | |
523 | if (unlikely(PageSwapCache(page))) | |
524 | return NULL; | |
525 | return page_mapping(page); | |
526 | } | |
527 | ||
b20ce5e0 KS |
528 | /* Slow path of page_mapcount() for compound pages */ |
529 | int __page_mapcount(struct page *page) | |
530 | { | |
531 | int ret; | |
532 | ||
533 | ret = atomic_read(&page->_mapcount) + 1; | |
dd78fedd KS |
534 | /* |
535 | * For file THP page->_mapcount contains total number of mapping | |
536 | * of the page: no need to look into compound_mapcount. | |
537 | */ | |
538 | if (!PageAnon(page) && !PageHuge(page)) | |
539 | return ret; | |
b20ce5e0 KS |
540 | page = compound_head(page); |
541 | ret += atomic_read(compound_mapcount_ptr(page)) + 1; | |
542 | if (PageDoubleMap(page)) | |
543 | ret--; | |
544 | return ret; | |
545 | } | |
546 | EXPORT_SYMBOL_GPL(__page_mapcount); | |
547 | ||
39a1aa8e AR |
548 | int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; |
549 | int sysctl_overcommit_ratio __read_mostly = 50; | |
550 | unsigned long sysctl_overcommit_kbytes __read_mostly; | |
551 | int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; | |
552 | unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ | |
553 | unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ | |
554 | ||
49f0ce5f JM |
555 | int overcommit_ratio_handler(struct ctl_table *table, int write, |
556 | void __user *buffer, size_t *lenp, | |
557 | loff_t *ppos) | |
558 | { | |
559 | int ret; | |
560 | ||
561 | ret = proc_dointvec(table, write, buffer, lenp, ppos); | |
562 | if (ret == 0 && write) | |
563 | sysctl_overcommit_kbytes = 0; | |
564 | return ret; | |
565 | } | |
566 | ||
567 | int overcommit_kbytes_handler(struct ctl_table *table, int write, | |
568 | void __user *buffer, size_t *lenp, | |
569 | loff_t *ppos) | |
570 | { | |
571 | int ret; | |
572 | ||
573 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); | |
574 | if (ret == 0 && write) | |
575 | sysctl_overcommit_ratio = 0; | |
576 | return ret; | |
577 | } | |
578 | ||
00619bcc JM |
579 | /* |
580 | * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used | |
581 | */ | |
582 | unsigned long vm_commit_limit(void) | |
583 | { | |
49f0ce5f JM |
584 | unsigned long allowed; |
585 | ||
586 | if (sysctl_overcommit_kbytes) | |
587 | allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10); | |
588 | else | |
589 | allowed = ((totalram_pages - hugetlb_total_pages()) | |
590 | * sysctl_overcommit_ratio / 100); | |
591 | allowed += total_swap_pages; | |
592 | ||
593 | return allowed; | |
00619bcc JM |
594 | } |
595 | ||
39a1aa8e AR |
596 | /* |
597 | * Make sure vm_committed_as in one cacheline and not cacheline shared with | |
598 | * other variables. It can be updated by several CPUs frequently. | |
599 | */ | |
600 | struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; | |
601 | ||
602 | /* | |
603 | * The global memory commitment made in the system can be a metric | |
604 | * that can be used to drive ballooning decisions when Linux is hosted | |
605 | * as a guest. On Hyper-V, the host implements a policy engine for dynamically | |
606 | * balancing memory across competing virtual machines that are hosted. | |
607 | * Several metrics drive this policy engine including the guest reported | |
608 | * memory commitment. | |
609 | */ | |
610 | unsigned long vm_memory_committed(void) | |
611 | { | |
612 | return percpu_counter_read_positive(&vm_committed_as); | |
613 | } | |
614 | EXPORT_SYMBOL_GPL(vm_memory_committed); | |
615 | ||
616 | /* | |
617 | * Check that a process has enough memory to allocate a new virtual | |
618 | * mapping. 0 means there is enough memory for the allocation to | |
619 | * succeed and -ENOMEM implies there is not. | |
620 | * | |
621 | * We currently support three overcommit policies, which are set via the | |
622 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting | |
623 | * | |
624 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. | |
625 | * Additional code 2002 Jul 20 by Robert Love. | |
626 | * | |
627 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. | |
628 | * | |
629 | * Note this is a helper function intended to be used by LSMs which | |
630 | * wish to use this logic. | |
631 | */ | |
632 | int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) | |
633 | { | |
634 | long free, allowed, reserve; | |
635 | ||
636 | VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) < | |
637 | -(s64)vm_committed_as_batch * num_online_cpus(), | |
638 | "memory commitment underflow"); | |
639 | ||
640 | vm_acct_memory(pages); | |
641 | ||
642 | /* | |
643 | * Sometimes we want to use more memory than we have | |
644 | */ | |
645 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) | |
646 | return 0; | |
647 | ||
648 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { | |
c41f012a | 649 | free = global_zone_page_state(NR_FREE_PAGES); |
11fb9989 | 650 | free += global_node_page_state(NR_FILE_PAGES); |
39a1aa8e AR |
651 | |
652 | /* | |
653 | * shmem pages shouldn't be counted as free in this | |
654 | * case, they can't be purged, only swapped out, and | |
655 | * that won't affect the overall amount of available | |
656 | * memory in the system. | |
657 | */ | |
11fb9989 | 658 | free -= global_node_page_state(NR_SHMEM); |
39a1aa8e AR |
659 | |
660 | free += get_nr_swap_pages(); | |
661 | ||
662 | /* | |
663 | * Any slabs which are created with the | |
664 | * SLAB_RECLAIM_ACCOUNT flag claim to have contents | |
665 | * which are reclaimable, under pressure. The dentry | |
666 | * cache and most inode caches should fall into this | |
667 | */ | |
d507e2eb | 668 | free += global_node_page_state(NR_SLAB_RECLAIMABLE); |
39a1aa8e AR |
669 | |
670 | /* | |
671 | * Leave reserved pages. The pages are not for anonymous pages. | |
672 | */ | |
673 | if (free <= totalreserve_pages) | |
674 | goto error; | |
675 | else | |
676 | free -= totalreserve_pages; | |
677 | ||
678 | /* | |
679 | * Reserve some for root | |
680 | */ | |
681 | if (!cap_sys_admin) | |
682 | free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); | |
683 | ||
684 | if (free > pages) | |
685 | return 0; | |
686 | ||
687 | goto error; | |
688 | } | |
689 | ||
690 | allowed = vm_commit_limit(); | |
691 | /* | |
692 | * Reserve some for root | |
693 | */ | |
694 | if (!cap_sys_admin) | |
695 | allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); | |
696 | ||
697 | /* | |
698 | * Don't let a single process grow so big a user can't recover | |
699 | */ | |
700 | if (mm) { | |
701 | reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); | |
702 | allowed -= min_t(long, mm->total_vm / 32, reserve); | |
703 | } | |
704 | ||
705 | if (percpu_counter_read_positive(&vm_committed_as) < allowed) | |
706 | return 0; | |
707 | error: | |
708 | vm_unacct_memory(pages); | |
709 | ||
710 | return -ENOMEM; | |
711 | } | |
712 | ||
a9090253 WR |
713 | /** |
714 | * get_cmdline() - copy the cmdline value to a buffer. | |
715 | * @task: the task whose cmdline value to copy. | |
716 | * @buffer: the buffer to copy to. | |
717 | * @buflen: the length of the buffer. Larger cmdline values are truncated | |
718 | * to this length. | |
719 | * Returns the size of the cmdline field copied. Note that the copy does | |
720 | * not guarantee an ending NULL byte. | |
721 | */ | |
722 | int get_cmdline(struct task_struct *task, char *buffer, int buflen) | |
723 | { | |
724 | int res = 0; | |
725 | unsigned int len; | |
726 | struct mm_struct *mm = get_task_mm(task); | |
a3b609ef | 727 | unsigned long arg_start, arg_end, env_start, env_end; |
a9090253 WR |
728 | if (!mm) |
729 | goto out; | |
730 | if (!mm->arg_end) | |
731 | goto out_mm; /* Shh! No looking before we're done */ | |
732 | ||
a3b609ef MG |
733 | down_read(&mm->mmap_sem); |
734 | arg_start = mm->arg_start; | |
735 | arg_end = mm->arg_end; | |
736 | env_start = mm->env_start; | |
737 | env_end = mm->env_end; | |
738 | up_read(&mm->mmap_sem); | |
739 | ||
740 | len = arg_end - arg_start; | |
a9090253 WR |
741 | |
742 | if (len > buflen) | |
743 | len = buflen; | |
744 | ||
f307ab6d | 745 | res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE); |
a9090253 WR |
746 | |
747 | /* | |
748 | * If the nul at the end of args has been overwritten, then | |
749 | * assume application is using setproctitle(3). | |
750 | */ | |
751 | if (res > 0 && buffer[res-1] != '\0' && len < buflen) { | |
752 | len = strnlen(buffer, res); | |
753 | if (len < res) { | |
754 | res = len; | |
755 | } else { | |
a3b609ef | 756 | len = env_end - env_start; |
a9090253 WR |
757 | if (len > buflen - res) |
758 | len = buflen - res; | |
a3b609ef | 759 | res += access_process_vm(task, env_start, |
f307ab6d LS |
760 | buffer+res, len, |
761 | FOLL_FORCE); | |
a9090253 WR |
762 | res = strnlen(buffer, res); |
763 | } | |
764 | } | |
765 | out_mm: | |
766 | mmput(mm); | |
767 | out: | |
768 | return res; | |
769 | } |