blob: 21ddf90f883d9fe8cf4c3d6f993100a4ca1559fd [file] [log] [blame]
Andrew Morton16d69262008-07-25 19:44:36 -07001#include <linux/mm.h>
Matt Mackall30992c92006-01-08 01:01:43 -08002#include <linux/slab.h>
3#include <linux/string.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -07004#include <linux/compiler.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -04005#include <linux/export.h>
Davi Arnaut96840aa2006-03-24 03:18:42 -08006#include <linux/err.h>
Adrian Bunk3b8f14b2008-07-26 15:22:28 -07007#include <linux/sched.h>
Ingo Molnar6e84f312017-02-08 18:51:29 +01008#include <linux/sched/mm.h>
Ingo Molnar68db0cf2017-02-08 18:51:37 +01009#include <linux/sched/task_stack.h>
Al Viroeb36c582012-05-30 20:17:35 -040010#include <linux/security.h>
Shaohua Li98003392013-02-22 16:34:35 -080011#include <linux/swap.h>
Shaohua Li33806f02013-02-22 16:34:37 -080012#include <linux/swapops.h>
Jerome Marchand00619bc2013-11-12 15:08:31 -080013#include <linux/mman.h>
14#include <linux/hugetlb.h>
Al Viro39f1f782014-05-06 14:02:53 -040015#include <linux/vmalloc.h>
Mike Rapoport897ab3e2017-02-24 14:58:22 -080016#include <linux/userfaultfd_k.h>
Jerome Marchand00619bc2013-11-12 15:08:31 -080017
Andrzej Hajdaa4bb1e42015-02-13 14:36:24 -080018#include <asm/sections.h>
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080019#include <linux/uaccess.h>
Matt Mackall30992c92006-01-08 01:01:43 -080020
Namhyung Kim6038def2011-05-24 17:11:22 -070021#include "internal.h"
22
Andrzej Hajdaa4bb1e42015-02-13 14:36:24 -080023static 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 */
35void kfree_const(const void *x)
36{
37 if (!is_kernel_rodata((unsigned long)x))
38 kfree(x);
39}
40EXPORT_SYMBOL(kfree_const);
41
Matt Mackall30992c92006-01-08 01:01:43 -080042/**
Matt Mackall30992c92006-01-08 01:01:43 -080043 * kstrdup - allocate space for and copy an existing string
Matt Mackall30992c92006-01-08 01:01:43 -080044 * @s: the string to duplicate
45 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
46 */
47char *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;
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -070056 buf = kmalloc_track_caller(len, gfp);
Matt Mackall30992c92006-01-08 01:01:43 -080057 if (buf)
58 memcpy(buf, s, len);
59 return buf;
60}
61EXPORT_SYMBOL(kstrdup);
Davi Arnaut96840aa2006-03-24 03:18:42 -080062
Alexey Dobriyan1a2f67b2006-09-30 23:27:20 -070063/**
Andrzej Hajdaa4bb1e42015-02-13 14:36:24 -080064 * 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 */
72const 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}
79EXPORT_SYMBOL(kstrdup_const);
80
81/**
Jeremy Fitzhardinge1e66df32007-07-17 18:37:02 -070082 * 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
David Howellsf3515742017-07-04 17:25:02 +010086 *
87 * Note: Use kmemdup_nul() instead if the size is known exactly.
Jeremy Fitzhardinge1e66df32007-07-17 18:37:02 -070088 */
89char *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}
105EXPORT_SYMBOL(kstrndup);
106
107/**
Alexey Dobriyan1a2f67b2006-09-30 23:27:20 -0700108 * kmemdup - duplicate region of memory
109 *
110 * @src: memory region to duplicate
111 * @len: memory region length
112 * @gfp: GFP mask to use
113 */
114void *kmemdup(const void *src, size_t len, gfp_t gfp)
115{
116 void *p;
117
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700118 p = kmalloc_track_caller(len, gfp);
Alexey Dobriyan1a2f67b2006-09-30 23:27:20 -0700119 if (p)
120 memcpy(p, src, len);
121 return p;
122}
123EXPORT_SYMBOL(kmemdup);
124
Christoph Lameteref2ad802007-07-17 04:03:21 -0700125/**
David Howellsf3515742017-07-04 17:25:02 +0100126 * 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 */
131char *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}
145EXPORT_SYMBOL(kmemdup_nul);
146
147/**
Li Zefan610a77e2009-03-31 15:23:16 -0700148 * memdup_user - duplicate memory region from user space
149 *
150 * @src: source address in user space
151 * @len: number of bytes to copy
152 *
153 * Returns an ERR_PTR() on failure.
154 */
155void *memdup_user(const void __user *src, size_t len)
156{
157 void *p;
158
159 /*
160 * Always use GFP_KERNEL, since copy_from_user() can sleep and
161 * cause pagefault, which makes it pointless to use GFP_NOFS
162 * or GFP_ATOMIC.
163 */
164 p = kmalloc_track_caller(len, GFP_KERNEL);
165 if (!p)
166 return ERR_PTR(-ENOMEM);
167
168 if (copy_from_user(p, src, len)) {
169 kfree(p);
170 return ERR_PTR(-EFAULT);
171 }
172
173 return p;
174}
175EXPORT_SYMBOL(memdup_user);
176
Davi Arnaut96840aa2006-03-24 03:18:42 -0800177/*
178 * strndup_user - duplicate an existing string from user space
Davi Arnaut96840aa2006-03-24 03:18:42 -0800179 * @s: The string to duplicate
180 * @n: Maximum number of bytes to copy, including the trailing NUL.
181 */
182char *strndup_user(const char __user *s, long n)
183{
184 char *p;
185 long length;
186
187 length = strnlen_user(s, n);
188
189 if (!length)
190 return ERR_PTR(-EFAULT);
191
192 if (length > n)
193 return ERR_PTR(-EINVAL);
194
Julia Lawall90d74042010-08-09 17:18:26 -0700195 p = memdup_user(s, length);
Davi Arnaut96840aa2006-03-24 03:18:42 -0800196
Julia Lawall90d74042010-08-09 17:18:26 -0700197 if (IS_ERR(p))
198 return p;
Davi Arnaut96840aa2006-03-24 03:18:42 -0800199
200 p[length - 1] = '\0';
201
202 return p;
203}
204EXPORT_SYMBOL(strndup_user);
Andrew Morton16d69262008-07-25 19:44:36 -0700205
Al Viroe9d408e2015-12-24 00:06:05 -0500206/**
207 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
208 *
209 * @src: source address in user space
210 * @len: number of bytes to copy
211 *
212 * Returns an ERR_PTR() on failure.
213 */
214void *memdup_user_nul(const void __user *src, size_t len)
215{
216 char *p;
217
218 /*
219 * Always use GFP_KERNEL, since copy_from_user() can sleep and
220 * cause pagefault, which makes it pointless to use GFP_NOFS
221 * or GFP_ATOMIC.
222 */
223 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
224 if (!p)
225 return ERR_PTR(-ENOMEM);
226
227 if (copy_from_user(p, src, len)) {
228 kfree(p);
229 return ERR_PTR(-EFAULT);
230 }
231 p[len] = '\0';
232
233 return p;
234}
235EXPORT_SYMBOL(memdup_user_nul);
236
Namhyung Kim6038def2011-05-24 17:11:22 -0700237void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
238 struct vm_area_struct *prev, struct rb_node *rb_parent)
239{
240 struct vm_area_struct *next;
241
242 vma->vm_prev = prev;
243 if (prev) {
244 next = prev->vm_next;
245 prev->vm_next = vma;
246 } else {
247 mm->mmap = vma;
248 if (rb_parent)
249 next = rb_entry(rb_parent,
250 struct vm_area_struct, vm_rb);
251 else
252 next = NULL;
253 }
254 vma->vm_next = next;
255 if (next)
256 next->vm_prev = vma;
257}
258
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700259/* Check if the vma is being used as a stack by this task */
Andy Lutomirskid17af502016-09-30 10:58:58 -0700260int vma_is_stack_for_current(struct vm_area_struct *vma)
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700261{
Andy Lutomirskid17af502016-09-30 10:58:58 -0700262 struct task_struct * __maybe_unused t = current;
263
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700264 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
265}
266
David Howellsefc1a3b2010-01-15 17:01:35 -0800267#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
Andrew Morton16d69262008-07-25 19:44:36 -0700268void arch_pick_mmap_layout(struct mm_struct *mm)
269{
270 mm->mmap_base = TASK_UNMAPPED_BASE;
271 mm->get_unmapped_area = arch_get_unmapped_area;
Andrew Morton16d69262008-07-25 19:44:36 -0700272}
273#endif
Rusty Russell912985d2008-08-12 17:52:52 -0500274
Xiao Guangrong45888a02010-08-22 19:08:57 +0800275/*
276 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
277 * back to the regular GUP.
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300278 * If the architecture not support this function, simply return with no
Xiao Guangrong45888a02010-08-22 19:08:57 +0800279 * page pinned
280 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700281int __weak __get_user_pages_fast(unsigned long start,
Xiao Guangrong45888a02010-08-22 19:08:57 +0800282 int nr_pages, int write, struct page **pages)
283{
284 return 0;
285}
286EXPORT_SYMBOL_GPL(__get_user_pages_fast);
287
Andy Grover9de100d2009-04-13 14:40:05 -0700288/**
289 * get_user_pages_fast() - pin user pages in memory
290 * @start: starting user address
291 * @nr_pages: number of pages from start to pin
292 * @write: whether pages will be written to
293 * @pages: array that receives pointers to the pages pinned.
294 * Should be at least nr_pages long.
295 *
Andy Grover9de100d2009-04-13 14:40:05 -0700296 * Returns number of pages pinned. This may be fewer than the number
297 * requested. If nr_pages is 0 or negative, returns 0. If no pages
298 * were pinned, returns -errno.
Nick Piggind2bf6be2009-06-16 15:31:39 -0700299 *
300 * get_user_pages_fast provides equivalent functionality to get_user_pages,
301 * operating on current and current->mm, with force=0 and vma=NULL. However
302 * unlike get_user_pages, it must be called without mmap_sem held.
303 *
304 * get_user_pages_fast may take mmap_sem and page table locks, so no
305 * assumptions can be made about lack of locking. get_user_pages_fast is to be
306 * implemented in a way that is advantageous (vs get_user_pages()) when the
307 * user memory area is already faulted in and present in ptes. However if the
308 * pages have to be faulted in, it may turn out to be slightly slower so
309 * callers need to carefully consider what to use. On many architectures,
310 * get_user_pages_fast simply falls back to get_user_pages.
Andy Grover9de100d2009-04-13 14:40:05 -0700311 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700312int __weak get_user_pages_fast(unsigned long start,
Rusty Russell912985d2008-08-12 17:52:52 -0500313 int nr_pages, int write, struct page **pages)
314{
Lorenzo Stoakesc1641542016-10-13 01:20:13 +0100315 return get_user_pages_unlocked(start, nr_pages, pages,
316 write ? FOLL_WRITE : 0);
Rusty Russell912985d2008-08-12 17:52:52 -0500317}
318EXPORT_SYMBOL_GPL(get_user_pages_fast);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +0200319
Al Viroeb36c582012-05-30 20:17:35 -0400320unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
321 unsigned long len, unsigned long prot,
Michal Hocko9fbeb5a2016-05-23 16:25:30 -0700322 unsigned long flag, unsigned long pgoff)
Al Viroeb36c582012-05-30 20:17:35 -0400323{
324 unsigned long ret;
325 struct mm_struct *mm = current->mm;
Michel Lespinasse41badc12013-02-22 16:32:47 -0800326 unsigned long populate;
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800327 LIST_HEAD(uf);
Al Viroeb36c582012-05-30 20:17:35 -0400328
329 ret = security_mmap_file(file, prot, flag);
330 if (!ret) {
Michal Hocko9fbeb5a2016-05-23 16:25:30 -0700331 if (down_write_killable(&mm->mmap_sem))
332 return -EINTR;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800333 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800334 &populate, &uf);
Al Viroeb36c582012-05-30 20:17:35 -0400335 up_write(&mm->mmap_sem);
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800336 userfaultfd_unmap_complete(mm, &uf);
Michel Lespinasse41badc12013-02-22 16:32:47 -0800337 if (populate)
338 mm_populate(ret, populate);
Al Viroeb36c582012-05-30 20:17:35 -0400339 }
340 return ret;
341}
342
343unsigned long vm_mmap(struct file *file, unsigned long addr,
344 unsigned long len, unsigned long prot,
345 unsigned long flag, unsigned long offset)
346{
347 if (unlikely(offset + PAGE_ALIGN(len) < offset))
348 return -EINVAL;
Alexander Kuleshovea53cde2015-11-05 18:46:46 -0800349 if (unlikely(offset_in_page(offset)))
Al Viroeb36c582012-05-30 20:17:35 -0400350 return -EINVAL;
351
Michal Hocko9fbeb5a2016-05-23 16:25:30 -0700352 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
Al Viroeb36c582012-05-30 20:17:35 -0400353}
354EXPORT_SYMBOL(vm_mmap);
355
Michal Hockoa7c3e902017-05-08 15:57:09 -0700356/**
357 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
358 * failure, fall back to non-contiguous (vmalloc) allocation.
359 * @size: size of the request.
360 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
361 * @node: numa node to allocate from
362 *
363 * Uses kmalloc to get the memory but if the allocation fails then falls back
364 * to the vmalloc allocator. Use kvfree for freeing the memory.
365 *
Michal Hocko6c5ab652017-05-08 15:57:15 -0700366 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported. __GFP_REPEAT
367 * is supported only for large (>32kB) allocations, and it should be used only if
368 * kmalloc is preferable to the vmalloc fallback, due to visible performance drawbacks.
Michal Hockoa7c3e902017-05-08 15:57:09 -0700369 *
370 * Any use of gfp flags outside of GFP_KERNEL should be consulted with mm people.
371 */
372void *kvmalloc_node(size_t size, gfp_t flags, int node)
373{
374 gfp_t kmalloc_flags = flags;
375 void *ret;
376
377 /*
378 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
379 * so the given set of flags has to be compatible.
380 */
381 WARN_ON_ONCE((flags & GFP_KERNEL) != GFP_KERNEL);
382
383 /*
Michal Hocko4f4f2ba2017-06-02 14:46:19 -0700384 * We want to attempt a large physically contiguous block first because
385 * it is less likely to fragment multiple larger blocks and therefore
386 * contribute to a long term fragmentation less than vmalloc fallback.
387 * However make sure that larger requests are not too disruptive - no
388 * OOM killer and no allocation failure warnings as we have a fallback.
Michal Hockoa7c3e902017-05-08 15:57:09 -0700389 */
Michal Hocko6c5ab652017-05-08 15:57:15 -0700390 if (size > PAGE_SIZE) {
391 kmalloc_flags |= __GFP_NOWARN;
392
393 /*
394 * We have to override __GFP_REPEAT by __GFP_NORETRY for !costly
395 * requests because there is no other way to tell the allocator
396 * that we want to fail rather than retry endlessly.
397 */
398 if (!(kmalloc_flags & __GFP_REPEAT) ||
399 (size <= PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
400 kmalloc_flags |= __GFP_NORETRY;
401 }
Michal Hockoa7c3e902017-05-08 15:57:09 -0700402
403 ret = kmalloc_node(size, kmalloc_flags, node);
404
405 /*
406 * It doesn't really make sense to fallback to vmalloc for sub page
407 * requests
408 */
409 if (ret || size <= PAGE_SIZE)
410 return ret;
411
Michal Hocko8594a212017-05-12 15:46:41 -0700412 return __vmalloc_node_flags_caller(size, node, flags,
413 __builtin_return_address(0));
Michal Hockoa7c3e902017-05-08 15:57:09 -0700414}
415EXPORT_SYMBOL(kvmalloc_node);
416
Al Viro39f1f782014-05-06 14:02:53 -0400417void kvfree(const void *addr)
418{
419 if (is_vmalloc_addr(addr))
420 vfree(addr);
421 else
422 kfree(addr);
423}
424EXPORT_SYMBOL(kvfree);
425
Kirill A. Shutemove39155e2015-04-15 16:14:53 -0700426static inline void *__page_rmapping(struct page *page)
427{
428 unsigned long mapping;
429
430 mapping = (unsigned long)page->mapping;
431 mapping &= ~PAGE_MAPPING_FLAGS;
432
433 return (void *)mapping;
434}
435
436/* Neutral page->mapping pointer to address_space or anon_vma or other */
437void *page_rmapping(struct page *page)
438{
439 page = compound_head(page);
440 return __page_rmapping(page);
441}
442
Andrew Morton1aa8aea2016-05-19 17:12:00 -0700443/*
444 * Return true if this page is mapped into pagetables.
445 * For compound page it returns true if any subpage of compound page is mapped.
446 */
447bool page_mapped(struct page *page)
448{
449 int i;
450
451 if (likely(!PageCompound(page)))
452 return atomic_read(&page->_mapcount) >= 0;
453 page = compound_head(page);
454 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
455 return true;
456 if (PageHuge(page))
457 return false;
458 for (i = 0; i < hpage_nr_pages(page); i++) {
459 if (atomic_read(&page[i]._mapcount) >= 0)
460 return true;
461 }
462 return false;
463}
464EXPORT_SYMBOL(page_mapped);
465
Kirill A. Shutemove39155e2015-04-15 16:14:53 -0700466struct anon_vma *page_anon_vma(struct page *page)
467{
468 unsigned long mapping;
469
470 page = compound_head(page);
471 mapping = (unsigned long)page->mapping;
472 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
473 return NULL;
474 return __page_rmapping(page);
475}
476
Shaohua Li98003392013-02-22 16:34:35 -0800477struct address_space *page_mapping(struct page *page)
478{
Kirill A. Shutemov1c290f62016-01-15 16:52:07 -0800479 struct address_space *mapping;
480
481 page = compound_head(page);
Shaohua Li98003392013-02-22 16:34:35 -0800482
Mikulas Patocka03e5ac22014-01-14 17:56:40 -0800483 /* This happens if someone calls flush_dcache_page on slab page */
484 if (unlikely(PageSlab(page)))
485 return NULL;
486
Shaohua Li33806f02013-02-22 16:34:37 -0800487 if (unlikely(PageSwapCache(page))) {
488 swp_entry_t entry;
489
490 entry.val = page_private(page);
Kirill A. Shutemove39155e2015-04-15 16:14:53 -0700491 return swap_address_space(entry);
492 }
493
Kirill A. Shutemov1c290f62016-01-15 16:52:07 -0800494 mapping = page->mapping;
Minchan Kimbda807d2016-07-26 15:23:05 -0700495 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
Kirill A. Shutemove39155e2015-04-15 16:14:53 -0700496 return NULL;
Minchan Kimbda807d2016-07-26 15:23:05 -0700497
498 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
Shaohua Li98003392013-02-22 16:34:35 -0800499}
Minchan Kimbda807d2016-07-26 15:23:05 -0700500EXPORT_SYMBOL(page_mapping);
Shaohua Li98003392013-02-22 16:34:35 -0800501
Kirill A. Shutemovb20ce5e2016-01-15 16:54:37 -0800502/* Slow path of page_mapcount() for compound pages */
503int __page_mapcount(struct page *page)
504{
505 int ret;
506
507 ret = atomic_read(&page->_mapcount) + 1;
Kirill A. Shutemovdd78fed2016-07-26 15:25:26 -0700508 /*
509 * For file THP page->_mapcount contains total number of mapping
510 * of the page: no need to look into compound_mapcount.
511 */
512 if (!PageAnon(page) && !PageHuge(page))
513 return ret;
Kirill A. Shutemovb20ce5e2016-01-15 16:54:37 -0800514 page = compound_head(page);
515 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
516 if (PageDoubleMap(page))
517 ret--;
518 return ret;
519}
520EXPORT_SYMBOL_GPL(__page_mapcount);
521
Andrey Ryabinin39a1aa82016-03-17 14:18:50 -0700522int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
523int sysctl_overcommit_ratio __read_mostly = 50;
524unsigned long sysctl_overcommit_kbytes __read_mostly;
525int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
526unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
527unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
528
Jerome Marchand49f0ce52014-01-21 15:49:14 -0800529int overcommit_ratio_handler(struct ctl_table *table, int write,
530 void __user *buffer, size_t *lenp,
531 loff_t *ppos)
532{
533 int ret;
534
535 ret = proc_dointvec(table, write, buffer, lenp, ppos);
536 if (ret == 0 && write)
537 sysctl_overcommit_kbytes = 0;
538 return ret;
539}
540
541int overcommit_kbytes_handler(struct ctl_table *table, int write,
542 void __user *buffer, size_t *lenp,
543 loff_t *ppos)
544{
545 int ret;
546
547 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
548 if (ret == 0 && write)
549 sysctl_overcommit_ratio = 0;
550 return ret;
551}
552
Jerome Marchand00619bc2013-11-12 15:08:31 -0800553/*
554 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
555 */
556unsigned long vm_commit_limit(void)
557{
Jerome Marchand49f0ce52014-01-21 15:49:14 -0800558 unsigned long allowed;
559
560 if (sysctl_overcommit_kbytes)
561 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
562 else
563 allowed = ((totalram_pages - hugetlb_total_pages())
564 * sysctl_overcommit_ratio / 100);
565 allowed += total_swap_pages;
566
567 return allowed;
Jerome Marchand00619bc2013-11-12 15:08:31 -0800568}
569
Andrey Ryabinin39a1aa82016-03-17 14:18:50 -0700570/*
571 * Make sure vm_committed_as in one cacheline and not cacheline shared with
572 * other variables. It can be updated by several CPUs frequently.
573 */
574struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
575
576/*
577 * The global memory commitment made in the system can be a metric
578 * that can be used to drive ballooning decisions when Linux is hosted
579 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
580 * balancing memory across competing virtual machines that are hosted.
581 * Several metrics drive this policy engine including the guest reported
582 * memory commitment.
583 */
584unsigned long vm_memory_committed(void)
585{
586 return percpu_counter_read_positive(&vm_committed_as);
587}
588EXPORT_SYMBOL_GPL(vm_memory_committed);
589
590/*
591 * Check that a process has enough memory to allocate a new virtual
592 * mapping. 0 means there is enough memory for the allocation to
593 * succeed and -ENOMEM implies there is not.
594 *
595 * We currently support three overcommit policies, which are set via the
596 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
597 *
598 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
599 * Additional code 2002 Jul 20 by Robert Love.
600 *
601 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
602 *
603 * Note this is a helper function intended to be used by LSMs which
604 * wish to use this logic.
605 */
606int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
607{
608 long free, allowed, reserve;
609
610 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
611 -(s64)vm_committed_as_batch * num_online_cpus(),
612 "memory commitment underflow");
613
614 vm_acct_memory(pages);
615
616 /*
617 * Sometimes we want to use more memory than we have
618 */
619 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
620 return 0;
621
622 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
623 free = global_page_state(NR_FREE_PAGES);
Mel Gorman11fb9982016-07-28 15:46:20 -0700624 free += global_node_page_state(NR_FILE_PAGES);
Andrey Ryabinin39a1aa82016-03-17 14:18:50 -0700625
626 /*
627 * shmem pages shouldn't be counted as free in this
628 * case, they can't be purged, only swapped out, and
629 * that won't affect the overall amount of available
630 * memory in the system.
631 */
Mel Gorman11fb9982016-07-28 15:46:20 -0700632 free -= global_node_page_state(NR_SHMEM);
Andrey Ryabinin39a1aa82016-03-17 14:18:50 -0700633
634 free += get_nr_swap_pages();
635
636 /*
637 * Any slabs which are created with the
638 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
639 * which are reclaimable, under pressure. The dentry
640 * cache and most inode caches should fall into this
641 */
642 free += global_page_state(NR_SLAB_RECLAIMABLE);
643
644 /*
645 * Leave reserved pages. The pages are not for anonymous pages.
646 */
647 if (free <= totalreserve_pages)
648 goto error;
649 else
650 free -= totalreserve_pages;
651
652 /*
653 * Reserve some for root
654 */
655 if (!cap_sys_admin)
656 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
657
658 if (free > pages)
659 return 0;
660
661 goto error;
662 }
663
664 allowed = vm_commit_limit();
665 /*
666 * Reserve some for root
667 */
668 if (!cap_sys_admin)
669 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
670
671 /*
672 * Don't let a single process grow so big a user can't recover
673 */
674 if (mm) {
675 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
676 allowed -= min_t(long, mm->total_vm / 32, reserve);
677 }
678
679 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
680 return 0;
681error:
682 vm_unacct_memory(pages);
683
684 return -ENOMEM;
685}
686
William Robertsa9090252014-02-11 10:11:59 -0800687/**
688 * get_cmdline() - copy the cmdline value to a buffer.
689 * @task: the task whose cmdline value to copy.
690 * @buffer: the buffer to copy to.
691 * @buflen: the length of the buffer. Larger cmdline values are truncated
692 * to this length.
693 * Returns the size of the cmdline field copied. Note that the copy does
694 * not guarantee an ending NULL byte.
695 */
696int get_cmdline(struct task_struct *task, char *buffer, int buflen)
697{
698 int res = 0;
699 unsigned int len;
700 struct mm_struct *mm = get_task_mm(task);
Mateusz Guzika3b609e2016-01-20 15:01:05 -0800701 unsigned long arg_start, arg_end, env_start, env_end;
William Robertsa9090252014-02-11 10:11:59 -0800702 if (!mm)
703 goto out;
704 if (!mm->arg_end)
705 goto out_mm; /* Shh! No looking before we're done */
706
Mateusz Guzika3b609e2016-01-20 15:01:05 -0800707 down_read(&mm->mmap_sem);
708 arg_start = mm->arg_start;
709 arg_end = mm->arg_end;
710 env_start = mm->env_start;
711 env_end = mm->env_end;
712 up_read(&mm->mmap_sem);
713
714 len = arg_end - arg_start;
William Robertsa9090252014-02-11 10:11:59 -0800715
716 if (len > buflen)
717 len = buflen;
718
Lorenzo Stoakesf307ab62016-10-13 01:20:20 +0100719 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
William Robertsa9090252014-02-11 10:11:59 -0800720
721 /*
722 * If the nul at the end of args has been overwritten, then
723 * assume application is using setproctitle(3).
724 */
725 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
726 len = strnlen(buffer, res);
727 if (len < res) {
728 res = len;
729 } else {
Mateusz Guzika3b609e2016-01-20 15:01:05 -0800730 len = env_end - env_start;
William Robertsa9090252014-02-11 10:11:59 -0800731 if (len > buflen - res)
732 len = buflen - res;
Mateusz Guzika3b609e2016-01-20 15:01:05 -0800733 res += access_process_vm(task, env_start,
Lorenzo Stoakesf307ab62016-10-13 01:20:20 +0100734 buffer+res, len,
735 FOLL_FORCE);
William Robertsa9090252014-02-11 10:11:59 -0800736 res = strnlen(buffer, res);
737 }
738 }
739out_mm:
740 mmput(mm);
741out:
742 return res;
743}