blob: 45fc3169e7b0fdb2677db32083615c12bd5f4b7c [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 *
Al Viro50fd2f22018-01-07 13:06:15 -0500153 * Returns an ERR_PTR() on failure. Result is physically
154 * contiguous, to be freed by kfree().
Li Zefan610a77e2009-03-31 15:23:16 -0700155 */
156void *memdup_user(const void __user *src, size_t len)
157{
158 void *p;
159
Al Viro6c2c97a2018-01-07 13:00:27 -0500160 p = kmalloc_track_caller(len, GFP_USER);
Li Zefan610a77e2009-03-31 15:23:16 -0700161 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}
171EXPORT_SYMBOL(memdup_user);
172
Al Viro50fd2f22018-01-07 13:06:15 -0500173/**
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 */
182void *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}
197EXPORT_SYMBOL(vmemdup_user);
198
Davi Arnaut96840aa2006-03-24 03:18:42 -0800199/*
200 * strndup_user - duplicate an existing string from user space
Davi Arnaut96840aa2006-03-24 03:18:42 -0800201 * @s: The string to duplicate
202 * @n: Maximum number of bytes to copy, including the trailing NUL.
203 */
204char *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
Julia Lawall90d74042010-08-09 17:18:26 -0700217 p = memdup_user(s, length);
Davi Arnaut96840aa2006-03-24 03:18:42 -0800218
Julia Lawall90d74042010-08-09 17:18:26 -0700219 if (IS_ERR(p))
220 return p;
Davi Arnaut96840aa2006-03-24 03:18:42 -0800221
222 p[length - 1] = '\0';
223
224 return p;
225}
226EXPORT_SYMBOL(strndup_user);
Andrew Morton16d69262008-07-25 19:44:36 -0700227
Al Viroe9d408e2015-12-24 00:06:05 -0500228/**
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 */
236void *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}
257EXPORT_SYMBOL(memdup_user_nul);
258
Namhyung Kim6038def2011-05-24 17:11:22 -0700259void __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
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700281/* Check if the vma is being used as a stack by this task */
Andy Lutomirskid17af502016-09-30 10:58:58 -0700282int vma_is_stack_for_current(struct vm_area_struct *vma)
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700283{
Andy Lutomirskid17af502016-09-30 10:58:58 -0700284 struct task_struct * __maybe_unused t = current;
285
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700286 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
287}
288
David Howellsefc1a3b2010-01-15 17:01:35 -0800289#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
Kees Cook8f2af152018-04-10 16:34:53 -0700290void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
Andrew Morton16d69262008-07-25 19:44:36 -0700291{
292 mm->mmap_base = TASK_UNMAPPED_BASE;
293 mm->get_unmapped_area = arch_get_unmapped_area;
Andrew Morton16d69262008-07-25 19:44:36 -0700294}
295#endif
Rusty Russell912985d2008-08-12 17:52:52 -0500296
Xiao Guangrong45888a02010-08-22 19:08:57 +0800297/*
298 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
299 * back to the regular GUP.
Michael S. Tsirkind0811072018-04-13 15:35:23 -0700300 * Note a difference with get_user_pages_fast: this always returns the
301 * number of pages pinned, 0 if no pages were pinned.
302 * If the architecture does not support this function, simply return with no
303 * pages pinned.
Xiao Guangrong45888a02010-08-22 19:08:57 +0800304 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700305int __weak __get_user_pages_fast(unsigned long start,
Xiao Guangrong45888a02010-08-22 19:08:57 +0800306 int nr_pages, int write, struct page **pages)
307{
308 return 0;
309}
310EXPORT_SYMBOL_GPL(__get_user_pages_fast);
311
Andy Grover9de100d2009-04-13 14:40:05 -0700312/**
313 * get_user_pages_fast() - pin user pages in memory
314 * @start: starting user address
315 * @nr_pages: number of pages from start to pin
316 * @write: whether pages will be written to
317 * @pages: array that receives pointers to the pages pinned.
318 * Should be at least nr_pages long.
319 *
Andy Grover9de100d2009-04-13 14:40:05 -0700320 * Returns number of pages pinned. This may be fewer than the number
321 * requested. If nr_pages is 0 or negative, returns 0. If no pages
322 * were pinned, returns -errno.
Nick Piggind2bf6be2009-06-16 15:31:39 -0700323 *
324 * get_user_pages_fast provides equivalent functionality to get_user_pages,
325 * operating on current and current->mm, with force=0 and vma=NULL. However
326 * unlike get_user_pages, it must be called without mmap_sem held.
327 *
328 * get_user_pages_fast may take mmap_sem and page table locks, so no
329 * assumptions can be made about lack of locking. get_user_pages_fast is to be
330 * implemented in a way that is advantageous (vs get_user_pages()) when the
331 * user memory area is already faulted in and present in ptes. However if the
332 * pages have to be faulted in, it may turn out to be slightly slower so
333 * callers need to carefully consider what to use. On many architectures,
334 * get_user_pages_fast simply falls back to get_user_pages.
Andy Grover9de100d2009-04-13 14:40:05 -0700335 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700336int __weak get_user_pages_fast(unsigned long start,
Rusty Russell912985d2008-08-12 17:52:52 -0500337 int nr_pages, int write, struct page **pages)
338{
Lorenzo Stoakesc1641542016-10-13 01:20:13 +0100339 return get_user_pages_unlocked(start, nr_pages, pages,
340 write ? FOLL_WRITE : 0);
Rusty Russell912985d2008-08-12 17:52:52 -0500341}
342EXPORT_SYMBOL_GPL(get_user_pages_fast);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +0200343
Al Viroeb36c582012-05-30 20:17:35 -0400344unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
345 unsigned long len, unsigned long prot,
Michal Hocko9fbeb5a2016-05-23 16:25:30 -0700346 unsigned long flag, unsigned long pgoff)
Al Viroeb36c582012-05-30 20:17:35 -0400347{
348 unsigned long ret;
349 struct mm_struct *mm = current->mm;
Michel Lespinasse41badc12013-02-22 16:32:47 -0800350 unsigned long populate;
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800351 LIST_HEAD(uf);
Al Viroeb36c582012-05-30 20:17:35 -0400352
353 ret = security_mmap_file(file, prot, flag);
354 if (!ret) {
Michal Hocko9fbeb5a2016-05-23 16:25:30 -0700355 if (down_write_killable(&mm->mmap_sem))
356 return -EINTR;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800357 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800358 &populate, &uf);
Al Viroeb36c582012-05-30 20:17:35 -0400359 up_write(&mm->mmap_sem);
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800360 userfaultfd_unmap_complete(mm, &uf);
Michel Lespinasse41badc12013-02-22 16:32:47 -0800361 if (populate)
362 mm_populate(ret, populate);
Al Viroeb36c582012-05-30 20:17:35 -0400363 }
364 return ret;
365}
366
367unsigned long vm_mmap(struct file *file, unsigned long addr,
368 unsigned long len, unsigned long prot,
369 unsigned long flag, unsigned long offset)
370{
371 if (unlikely(offset + PAGE_ALIGN(len) < offset))
372 return -EINVAL;
Alexander Kuleshovea53cde2015-11-05 18:46:46 -0800373 if (unlikely(offset_in_page(offset)))
Al Viroeb36c582012-05-30 20:17:35 -0400374 return -EINVAL;
375
Michal Hocko9fbeb5a2016-05-23 16:25:30 -0700376 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
Al Viroeb36c582012-05-30 20:17:35 -0400377}
378EXPORT_SYMBOL(vm_mmap);
379
Michal Hockoa7c3e902017-05-08 15:57:09 -0700380/**
381 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
382 * failure, fall back to non-contiguous (vmalloc) allocation.
383 * @size: size of the request.
384 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
385 * @node: numa node to allocate from
386 *
387 * Uses kmalloc to get the memory but if the allocation fails then falls back
388 * to the vmalloc allocator. Use kvfree for freeing the memory.
389 *
Michal Hockocc965a22017-07-12 14:36:52 -0700390 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
391 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
392 * preferable to the vmalloc fallback, due to visible performance drawbacks.
Michal Hockoa7c3e902017-05-08 15:57:09 -0700393 *
394 * Any use of gfp flags outside of GFP_KERNEL should be consulted with mm people.
395 */
396void *kvmalloc_node(size_t size, gfp_t flags, int node)
397{
398 gfp_t kmalloc_flags = flags;
399 void *ret;
400
401 /*
402 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
403 * so the given set of flags has to be compatible.
404 */
405 WARN_ON_ONCE((flags & GFP_KERNEL) != GFP_KERNEL);
406
407 /*
Michal Hocko4f4f2ba2017-06-02 14:46:19 -0700408 * We want to attempt a large physically contiguous block first because
409 * it is less likely to fragment multiple larger blocks and therefore
410 * contribute to a long term fragmentation less than vmalloc fallback.
411 * However make sure that larger requests are not too disruptive - no
412 * OOM killer and no allocation failure warnings as we have a fallback.
Michal Hockoa7c3e902017-05-08 15:57:09 -0700413 */
Michal Hocko6c5ab652017-05-08 15:57:15 -0700414 if (size > PAGE_SIZE) {
415 kmalloc_flags |= __GFP_NOWARN;
416
Michal Hockocc965a22017-07-12 14:36:52 -0700417 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
Michal Hocko6c5ab652017-05-08 15:57:15 -0700418 kmalloc_flags |= __GFP_NORETRY;
419 }
Michal Hockoa7c3e902017-05-08 15:57:09 -0700420
421 ret = kmalloc_node(size, kmalloc_flags, node);
422
423 /*
424 * It doesn't really make sense to fallback to vmalloc for sub page
425 * requests
426 */
427 if (ret || size <= PAGE_SIZE)
428 return ret;
429
Michal Hocko8594a212017-05-12 15:46:41 -0700430 return __vmalloc_node_flags_caller(size, node, flags,
431 __builtin_return_address(0));
Michal Hockoa7c3e902017-05-08 15:57:09 -0700432}
433EXPORT_SYMBOL(kvmalloc_node);
434
Al Viro39f1f782014-05-06 14:02:53 -0400435void kvfree(const void *addr)
436{
437 if (is_vmalloc_addr(addr))
438 vfree(addr);
439 else
440 kfree(addr);
441}
442EXPORT_SYMBOL(kvfree);
443
Kirill A. Shutemove39155e2015-04-15 16:14:53 -0700444static inline void *__page_rmapping(struct page *page)
445{
446 unsigned long mapping;
447
448 mapping = (unsigned long)page->mapping;
449 mapping &= ~PAGE_MAPPING_FLAGS;
450
451 return (void *)mapping;
452}
453
454/* Neutral page->mapping pointer to address_space or anon_vma or other */
455void *page_rmapping(struct page *page)
456{
457 page = compound_head(page);
458 return __page_rmapping(page);
459}
460
Andrew Morton1aa8aea2016-05-19 17:12:00 -0700461/*
462 * Return true if this page is mapped into pagetables.
463 * For compound page it returns true if any subpage of compound page is mapped.
464 */
465bool page_mapped(struct page *page)
466{
467 int i;
468
469 if (likely(!PageCompound(page)))
470 return atomic_read(&page->_mapcount) >= 0;
471 page = compound_head(page);
472 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
473 return true;
474 if (PageHuge(page))
475 return false;
476 for (i = 0; i < hpage_nr_pages(page); i++) {
477 if (atomic_read(&page[i]._mapcount) >= 0)
478 return true;
479 }
480 return false;
481}
482EXPORT_SYMBOL(page_mapped);
483
Kirill A. Shutemove39155e2015-04-15 16:14:53 -0700484struct anon_vma *page_anon_vma(struct page *page)
485{
486 unsigned long mapping;
487
488 page = compound_head(page);
489 mapping = (unsigned long)page->mapping;
490 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
491 return NULL;
492 return __page_rmapping(page);
493}
494
Shaohua Li98003392013-02-22 16:34:35 -0800495struct address_space *page_mapping(struct page *page)
496{
Kirill A. Shutemov1c290f62016-01-15 16:52:07 -0800497 struct address_space *mapping;
498
499 page = compound_head(page);
Shaohua Li98003392013-02-22 16:34:35 -0800500
Mikulas Patocka03e5ac22014-01-14 17:56:40 -0800501 /* This happens if someone calls flush_dcache_page on slab page */
502 if (unlikely(PageSlab(page)))
503 return NULL;
504
Shaohua Li33806f02013-02-22 16:34:37 -0800505 if (unlikely(PageSwapCache(page))) {
506 swp_entry_t entry;
507
508 entry.val = page_private(page);
Kirill A. Shutemove39155e2015-04-15 16:14:53 -0700509 return swap_address_space(entry);
510 }
511
Kirill A. Shutemov1c290f62016-01-15 16:52:07 -0800512 mapping = page->mapping;
Minchan Kimbda807d2016-07-26 15:23:05 -0700513 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
Kirill A. Shutemove39155e2015-04-15 16:14:53 -0700514 return NULL;
Minchan Kimbda807d2016-07-26 15:23:05 -0700515
516 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
Shaohua Li98003392013-02-22 16:34:35 -0800517}
Minchan Kimbda807d2016-07-26 15:23:05 -0700518EXPORT_SYMBOL(page_mapping);
Shaohua Li98003392013-02-22 16:34:35 -0800519
Huang Yingcb9f7532018-04-05 16:24:39 -0700520/*
521 * For file cache pages, return the address_space, otherwise return NULL
522 */
523struct address_space *page_mapping_file(struct page *page)
524{
525 if (unlikely(PageSwapCache(page)))
526 return NULL;
527 return page_mapping(page);
528}
529
Kirill A. Shutemovb20ce5e2016-01-15 16:54:37 -0800530/* Slow path of page_mapcount() for compound pages */
531int __page_mapcount(struct page *page)
532{
533 int ret;
534
535 ret = atomic_read(&page->_mapcount) + 1;
Kirill A. Shutemovdd78fed2016-07-26 15:25:26 -0700536 /*
537 * For file THP page->_mapcount contains total number of mapping
538 * of the page: no need to look into compound_mapcount.
539 */
540 if (!PageAnon(page) && !PageHuge(page))
541 return ret;
Kirill A. Shutemovb20ce5e2016-01-15 16:54:37 -0800542 page = compound_head(page);
543 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
544 if (PageDoubleMap(page))
545 ret--;
546 return ret;
547}
548EXPORT_SYMBOL_GPL(__page_mapcount);
549
Andrey Ryabinin39a1aa82016-03-17 14:18:50 -0700550int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
551int sysctl_overcommit_ratio __read_mostly = 50;
552unsigned long sysctl_overcommit_kbytes __read_mostly;
553int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
554unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
555unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
556
Jerome Marchand49f0ce52014-01-21 15:49:14 -0800557int overcommit_ratio_handler(struct ctl_table *table, int write,
558 void __user *buffer, size_t *lenp,
559 loff_t *ppos)
560{
561 int ret;
562
563 ret = proc_dointvec(table, write, buffer, lenp, ppos);
564 if (ret == 0 && write)
565 sysctl_overcommit_kbytes = 0;
566 return ret;
567}
568
569int overcommit_kbytes_handler(struct ctl_table *table, int write,
570 void __user *buffer, size_t *lenp,
571 loff_t *ppos)
572{
573 int ret;
574
575 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
576 if (ret == 0 && write)
577 sysctl_overcommit_ratio = 0;
578 return ret;
579}
580
Jerome Marchand00619bc2013-11-12 15:08:31 -0800581/*
582 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
583 */
584unsigned long vm_commit_limit(void)
585{
Jerome Marchand49f0ce52014-01-21 15:49:14 -0800586 unsigned long allowed;
587
588 if (sysctl_overcommit_kbytes)
589 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
590 else
591 allowed = ((totalram_pages - hugetlb_total_pages())
592 * sysctl_overcommit_ratio / 100);
593 allowed += total_swap_pages;
594
595 return allowed;
Jerome Marchand00619bc2013-11-12 15:08:31 -0800596}
597
Andrey Ryabinin39a1aa82016-03-17 14:18:50 -0700598/*
599 * Make sure vm_committed_as in one cacheline and not cacheline shared with
600 * other variables. It can be updated by several CPUs frequently.
601 */
602struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
603
604/*
605 * The global memory commitment made in the system can be a metric
606 * that can be used to drive ballooning decisions when Linux is hosted
607 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
608 * balancing memory across competing virtual machines that are hosted.
609 * Several metrics drive this policy engine including the guest reported
610 * memory commitment.
611 */
612unsigned long vm_memory_committed(void)
613{
614 return percpu_counter_read_positive(&vm_committed_as);
615}
616EXPORT_SYMBOL_GPL(vm_memory_committed);
617
618/*
619 * Check that a process has enough memory to allocate a new virtual
620 * mapping. 0 means there is enough memory for the allocation to
621 * succeed and -ENOMEM implies there is not.
622 *
623 * We currently support three overcommit policies, which are set via the
624 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
625 *
626 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
627 * Additional code 2002 Jul 20 by Robert Love.
628 *
629 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
630 *
631 * Note this is a helper function intended to be used by LSMs which
632 * wish to use this logic.
633 */
634int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
635{
636 long free, allowed, reserve;
637
638 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
639 -(s64)vm_committed_as_batch * num_online_cpus(),
640 "memory commitment underflow");
641
642 vm_acct_memory(pages);
643
644 /*
645 * Sometimes we want to use more memory than we have
646 */
647 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
648 return 0;
649
650 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Michal Hockoc41f0122017-09-06 16:23:36 -0700651 free = global_zone_page_state(NR_FREE_PAGES);
Mel Gorman11fb9982016-07-28 15:46:20 -0700652 free += global_node_page_state(NR_FILE_PAGES);
Andrey Ryabinin39a1aa82016-03-17 14:18:50 -0700653
654 /*
655 * shmem pages shouldn't be counted as free in this
656 * case, they can't be purged, only swapped out, and
657 * that won't affect the overall amount of available
658 * memory in the system.
659 */
Mel Gorman11fb9982016-07-28 15:46:20 -0700660 free -= global_node_page_state(NR_SHMEM);
Andrey Ryabinin39a1aa82016-03-17 14:18:50 -0700661
662 free += get_nr_swap_pages();
663
664 /*
665 * Any slabs which are created with the
666 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
667 * which are reclaimable, under pressure. The dentry
668 * cache and most inode caches should fall into this
669 */
Johannes Weinerd507e2e2017-08-10 15:23:31 -0700670 free += global_node_page_state(NR_SLAB_RECLAIMABLE);
Andrey Ryabinin39a1aa82016-03-17 14:18:50 -0700671
672 /*
Roman Gushchind79f7aa2018-04-10 16:27:47 -0700673 * Part of the kernel memory, which can be released
674 * under memory pressure.
675 */
676 free += global_node_page_state(
677 NR_INDIRECTLY_RECLAIMABLE_BYTES) >> PAGE_SHIFT;
678
679 /*
Andrey Ryabinin39a1aa82016-03-17 14:18:50 -0700680 * Leave reserved pages. The pages are not for anonymous pages.
681 */
682 if (free <= totalreserve_pages)
683 goto error;
684 else
685 free -= totalreserve_pages;
686
687 /*
688 * Reserve some for root
689 */
690 if (!cap_sys_admin)
691 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
692
693 if (free > pages)
694 return 0;
695
696 goto error;
697 }
698
699 allowed = vm_commit_limit();
700 /*
701 * Reserve some for root
702 */
703 if (!cap_sys_admin)
704 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
705
706 /*
707 * Don't let a single process grow so big a user can't recover
708 */
709 if (mm) {
710 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
711 allowed -= min_t(long, mm->total_vm / 32, reserve);
712 }
713
714 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
715 return 0;
716error:
717 vm_unacct_memory(pages);
718
719 return -ENOMEM;
720}
721
William Robertsa9090252014-02-11 10:11:59 -0800722/**
723 * get_cmdline() - copy the cmdline value to a buffer.
724 * @task: the task whose cmdline value to copy.
725 * @buffer: the buffer to copy to.
726 * @buflen: the length of the buffer. Larger cmdline values are truncated
727 * to this length.
728 * Returns the size of the cmdline field copied. Note that the copy does
729 * not guarantee an ending NULL byte.
730 */
731int get_cmdline(struct task_struct *task, char *buffer, int buflen)
732{
733 int res = 0;
734 unsigned int len;
735 struct mm_struct *mm = get_task_mm(task);
Mateusz Guzika3b609e2016-01-20 15:01:05 -0800736 unsigned long arg_start, arg_end, env_start, env_end;
William Robertsa9090252014-02-11 10:11:59 -0800737 if (!mm)
738 goto out;
739 if (!mm->arg_end)
740 goto out_mm; /* Shh! No looking before we're done */
741
Mateusz Guzika3b609e2016-01-20 15:01:05 -0800742 down_read(&mm->mmap_sem);
743 arg_start = mm->arg_start;
744 arg_end = mm->arg_end;
745 env_start = mm->env_start;
746 env_end = mm->env_end;
747 up_read(&mm->mmap_sem);
748
749 len = arg_end - arg_start;
William Robertsa9090252014-02-11 10:11:59 -0800750
751 if (len > buflen)
752 len = buflen;
753
Lorenzo Stoakesf307ab62016-10-13 01:20:20 +0100754 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
William Robertsa9090252014-02-11 10:11:59 -0800755
756 /*
757 * If the nul at the end of args has been overwritten, then
758 * assume application is using setproctitle(3).
759 */
760 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
761 len = strnlen(buffer, res);
762 if (len < res) {
763 res = len;
764 } else {
Mateusz Guzika3b609e2016-01-20 15:01:05 -0800765 len = env_end - env_start;
William Robertsa9090252014-02-11 10:11:59 -0800766 if (len > buflen - res)
767 len = buflen - res;
Mateusz Guzika3b609e2016-01-20 15:01:05 -0800768 res += access_process_vm(task, env_start,
Lorenzo Stoakesf307ab62016-10-13 01:20:20 +0100769 buffer+res, len,
770 FOLL_FORCE);
William Robertsa9090252014-02-11 10:11:59 -0800771 res = strnlen(buffer, res);
772 }
773 }
774out_mm:
775 mmput(mm);
776out:
777 return res;
778}