Linux Audio

Check our new training course

Loading...
v3.1
  1#include <linux/mm.h>
  2#include <linux/slab.h>
  3#include <linux/string.h>
  4#include <linux/module.h>
 
  5#include <linux/err.h>
  6#include <linux/sched.h>
 
 
 
 
 
 
 
 
  7#include <asm/uaccess.h>
  8
  9#include "internal.h"
 10
 11#define CREATE_TRACE_POINTS
 12#include <trace/events/kmem.h>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 13
 14/**
 15 * kstrdup - allocate space for and copy an existing string
 16 * @s: the string to duplicate
 17 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
 18 */
 19char *kstrdup(const char *s, gfp_t gfp)
 20{
 21	size_t len;
 22	char *buf;
 23
 24	if (!s)
 25		return NULL;
 26
 27	len = strlen(s) + 1;
 28	buf = kmalloc_track_caller(len, gfp);
 29	if (buf)
 30		memcpy(buf, s, len);
 31	return buf;
 32}
 33EXPORT_SYMBOL(kstrdup);
 34
 35/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 36 * kstrndup - allocate space for and copy an existing string
 37 * @s: the string to duplicate
 38 * @max: read at most @max chars from @s
 39 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
 40 */
 41char *kstrndup(const char *s, size_t max, gfp_t gfp)
 42{
 43	size_t len;
 44	char *buf;
 45
 46	if (!s)
 47		return NULL;
 48
 49	len = strnlen(s, max);
 50	buf = kmalloc_track_caller(len+1, gfp);
 51	if (buf) {
 52		memcpy(buf, s, len);
 53		buf[len] = '\0';
 54	}
 55	return buf;
 56}
 57EXPORT_SYMBOL(kstrndup);
 58
 59/**
 60 * kmemdup - duplicate region of memory
 61 *
 62 * @src: memory region to duplicate
 63 * @len: memory region length
 64 * @gfp: GFP mask to use
 65 */
 66void *kmemdup(const void *src, size_t len, gfp_t gfp)
 67{
 68	void *p;
 69
 70	p = kmalloc_track_caller(len, gfp);
 71	if (p)
 72		memcpy(p, src, len);
 73	return p;
 74}
 75EXPORT_SYMBOL(kmemdup);
 76
 77/**
 78 * memdup_user - duplicate memory region from user space
 79 *
 80 * @src: source address in user space
 81 * @len: number of bytes to copy
 82 *
 83 * Returns an ERR_PTR() on failure.
 84 */
 85void *memdup_user(const void __user *src, size_t len)
 86{
 87	void *p;
 88
 89	/*
 90	 * Always use GFP_KERNEL, since copy_from_user() can sleep and
 91	 * cause pagefault, which makes it pointless to use GFP_NOFS
 92	 * or GFP_ATOMIC.
 93	 */
 94	p = kmalloc_track_caller(len, GFP_KERNEL);
 95	if (!p)
 96		return ERR_PTR(-ENOMEM);
 97
 98	if (copy_from_user(p, src, len)) {
 99		kfree(p);
100		return ERR_PTR(-EFAULT);
101	}
102
103	return p;
104}
105EXPORT_SYMBOL(memdup_user);
106
107/**
108 * __krealloc - like krealloc() but don't free @p.
109 * @p: object to reallocate memory for.
110 * @new_size: how many bytes of memory are required.
111 * @flags: the type of memory to allocate.
112 *
113 * This function is like krealloc() except it never frees the originally
114 * allocated buffer. Use this if you don't want to free the buffer immediately
115 * like, for example, with RCU.
116 */
117void *__krealloc(const void *p, size_t new_size, gfp_t flags)
118{
119	void *ret;
120	size_t ks = 0;
121
122	if (unlikely(!new_size))
123		return ZERO_SIZE_PTR;
124
125	if (p)
126		ks = ksize(p);
127
128	if (ks >= new_size)
129		return (void *)p;
130
131	ret = kmalloc_track_caller(new_size, flags);
132	if (ret && p)
133		memcpy(ret, p, ks);
134
135	return ret;
136}
137EXPORT_SYMBOL(__krealloc);
138
139/**
140 * krealloc - reallocate memory. The contents will remain unchanged.
141 * @p: object to reallocate memory for.
142 * @new_size: how many bytes of memory are required.
143 * @flags: the type of memory to allocate.
144 *
145 * The contents of the object pointed to are preserved up to the
146 * lesser of the new and old sizes.  If @p is %NULL, krealloc()
147 * behaves exactly like kmalloc().  If @size is 0 and @p is not a
148 * %NULL pointer, the object pointed to is freed.
149 */
150void *krealloc(const void *p, size_t new_size, gfp_t flags)
151{
152	void *ret;
153
154	if (unlikely(!new_size)) {
155		kfree(p);
156		return ZERO_SIZE_PTR;
157	}
158
159	ret = __krealloc(p, new_size, flags);
160	if (ret && p != ret)
161		kfree(p);
162
163	return ret;
164}
165EXPORT_SYMBOL(krealloc);
166
167/**
168 * kzfree - like kfree but zero memory
169 * @p: object to free memory of
170 *
171 * The memory of the object @p points to is zeroed before freed.
172 * If @p is %NULL, kzfree() does nothing.
173 *
174 * Note: this function zeroes the whole allocated buffer which can be a good
175 * deal bigger than the requested buffer size passed to kmalloc(). So be
176 * careful when using this function in performance sensitive code.
177 */
178void kzfree(const void *p)
179{
180	size_t ks;
181	void *mem = (void *)p;
182
183	if (unlikely(ZERO_OR_NULL_PTR(mem)))
184		return;
185	ks = ksize(mem);
186	memset(mem, 0, ks);
187	kfree(mem);
188}
189EXPORT_SYMBOL(kzfree);
190
191/*
192 * strndup_user - duplicate an existing string from user space
193 * @s: The string to duplicate
194 * @n: Maximum number of bytes to copy, including the trailing NUL.
195 */
196char *strndup_user(const char __user *s, long n)
197{
198	char *p;
199	long length;
200
201	length = strnlen_user(s, n);
202
203	if (!length)
204		return ERR_PTR(-EFAULT);
205
206	if (length > n)
207		return ERR_PTR(-EINVAL);
208
209	p = memdup_user(s, length);
210
211	if (IS_ERR(p))
212		return p;
213
214	p[length - 1] = '\0';
215
216	return p;
217}
218EXPORT_SYMBOL(strndup_user);
219
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
220void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
221		struct vm_area_struct *prev, struct rb_node *rb_parent)
222{
223	struct vm_area_struct *next;
224
225	vma->vm_prev = prev;
226	if (prev) {
227		next = prev->vm_next;
228		prev->vm_next = vma;
229	} else {
230		mm->mmap = vma;
231		if (rb_parent)
232			next = rb_entry(rb_parent,
233					struct vm_area_struct, vm_rb);
234		else
235			next = NULL;
236	}
237	vma->vm_next = next;
238	if (next)
239		next->vm_prev = vma;
240}
241
 
 
 
 
 
 
242#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
243void arch_pick_mmap_layout(struct mm_struct *mm)
244{
245	mm->mmap_base = TASK_UNMAPPED_BASE;
246	mm->get_unmapped_area = arch_get_unmapped_area;
247	mm->unmap_area = arch_unmap_area;
248}
249#endif
250
251/*
252 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
253 * back to the regular GUP.
254 * If the architecture not support this function, simply return with no
255 * page pinned
256 */
257int __attribute__((weak)) __get_user_pages_fast(unsigned long start,
258				 int nr_pages, int write, struct page **pages)
259{
260	return 0;
261}
262EXPORT_SYMBOL_GPL(__get_user_pages_fast);
263
264/**
265 * get_user_pages_fast() - pin user pages in memory
266 * @start:	starting user address
267 * @nr_pages:	number of pages from start to pin
268 * @write:	whether pages will be written to
269 * @pages:	array that receives pointers to the pages pinned.
270 *		Should be at least nr_pages long.
271 *
272 * Returns number of pages pinned. This may be fewer than the number
273 * requested. If nr_pages is 0 or negative, returns 0. If no pages
274 * were pinned, returns -errno.
275 *
276 * get_user_pages_fast provides equivalent functionality to get_user_pages,
277 * operating on current and current->mm, with force=0 and vma=NULL. However
278 * unlike get_user_pages, it must be called without mmap_sem held.
279 *
280 * get_user_pages_fast may take mmap_sem and page table locks, so no
281 * assumptions can be made about lack of locking. get_user_pages_fast is to be
282 * implemented in a way that is advantageous (vs get_user_pages()) when the
283 * user memory area is already faulted in and present in ptes. However if the
284 * pages have to be faulted in, it may turn out to be slightly slower so
285 * callers need to carefully consider what to use. On many architectures,
286 * get_user_pages_fast simply falls back to get_user_pages.
287 */
288int __attribute__((weak)) get_user_pages_fast(unsigned long start,
289				int nr_pages, int write, struct page **pages)
290{
 
 
 
 
 
 
 
 
 
291	struct mm_struct *mm = current->mm;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
292	int ret;
293
294	down_read(&mm->mmap_sem);
295	ret = get_user_pages(current, mm, start, nr_pages,
296					write, 0, pages, NULL);
297	up_read(&mm->mmap_sem);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
298
 
 
 
299	return ret;
300}
301EXPORT_SYMBOL_GPL(get_user_pages_fast);
302
303/* Tracepoints definitions. */
304EXPORT_TRACEPOINT_SYMBOL(kmalloc);
305EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
306EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
307EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
308EXPORT_TRACEPOINT_SYMBOL(kfree);
309EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
v4.6
  1#include <linux/mm.h>
  2#include <linux/slab.h>
  3#include <linux/string.h>
  4#include <linux/compiler.h>
  5#include <linux/export.h>
  6#include <linux/err.h>
  7#include <linux/sched.h>
  8#include <linux/security.h>
  9#include <linux/swap.h>
 10#include <linux/swapops.h>
 11#include <linux/mman.h>
 12#include <linux/hugetlb.h>
 13#include <linux/vmalloc.h>
 14
 15#include <asm/sections.h>
 16#include <asm/uaccess.h>
 17
 18#include "internal.h"
 19
 20static inline int is_kernel_rodata(unsigned long addr)
 21{
 22	return addr >= (unsigned long)__start_rodata &&
 23		addr < (unsigned long)__end_rodata;
 24}
 25
 26/**
 27 * kfree_const - conditionally free memory
 28 * @x: pointer to the memory
 29 *
 30 * Function calls kfree only if @x is not in .rodata section.
 31 */
 32void kfree_const(const void *x)
 33{
 34	if (!is_kernel_rodata((unsigned long)x))
 35		kfree(x);
 36}
 37EXPORT_SYMBOL(kfree_const);
 38
 39/**
 40 * kstrdup - allocate space for and copy an existing string
 41 * @s: the string to duplicate
 42 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
 43 */
 44char *kstrdup(const char *s, gfp_t gfp)
 45{
 46	size_t len;
 47	char *buf;
 48
 49	if (!s)
 50		return NULL;
 51
 52	len = strlen(s) + 1;
 53	buf = kmalloc_track_caller(len, gfp);
 54	if (buf)
 55		memcpy(buf, s, len);
 56	return buf;
 57}
 58EXPORT_SYMBOL(kstrdup);
 59
 60/**
 61 * kstrdup_const - conditionally duplicate an existing const string
 62 * @s: the string to duplicate
 63 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
 64 *
 65 * Function returns source string if it is in .rodata section otherwise it
 66 * fallbacks to kstrdup.
 67 * Strings allocated by kstrdup_const should be freed by kfree_const.
 68 */
 69const char *kstrdup_const(const char *s, gfp_t gfp)
 70{
 71	if (is_kernel_rodata((unsigned long)s))
 72		return s;
 73
 74	return kstrdup(s, gfp);
 75}
 76EXPORT_SYMBOL(kstrdup_const);
 77
 78/**
 79 * kstrndup - allocate space for and copy an existing string
 80 * @s: the string to duplicate
 81 * @max: read at most @max chars from @s
 82 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
 83 */
 84char *kstrndup(const char *s, size_t max, gfp_t gfp)
 85{
 86	size_t len;
 87	char *buf;
 88
 89	if (!s)
 90		return NULL;
 91
 92	len = strnlen(s, max);
 93	buf = kmalloc_track_caller(len+1, gfp);
 94	if (buf) {
 95		memcpy(buf, s, len);
 96		buf[len] = '\0';
 97	}
 98	return buf;
 99}
100EXPORT_SYMBOL(kstrndup);
101
102/**
103 * kmemdup - duplicate region of memory
104 *
105 * @src: memory region to duplicate
106 * @len: memory region length
107 * @gfp: GFP mask to use
108 */
109void *kmemdup(const void *src, size_t len, gfp_t gfp)
110{
111	void *p;
112
113	p = kmalloc_track_caller(len, gfp);
114	if (p)
115		memcpy(p, src, len);
116	return p;
117}
118EXPORT_SYMBOL(kmemdup);
119
120/**
121 * memdup_user - duplicate memory region from user space
122 *
123 * @src: source address in user space
124 * @len: number of bytes to copy
125 *
126 * Returns an ERR_PTR() on failure.
127 */
128void *memdup_user(const void __user *src, size_t len)
129{
130	void *p;
131
132	/*
133	 * Always use GFP_KERNEL, since copy_from_user() can sleep and
134	 * cause pagefault, which makes it pointless to use GFP_NOFS
135	 * or GFP_ATOMIC.
136	 */
137	p = kmalloc_track_caller(len, GFP_KERNEL);
138	if (!p)
139		return ERR_PTR(-ENOMEM);
140
141	if (copy_from_user(p, src, len)) {
142		kfree(p);
143		return ERR_PTR(-EFAULT);
144	}
145
146	return p;
147}
148EXPORT_SYMBOL(memdup_user);
149
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
150/*
151 * strndup_user - duplicate an existing string from user space
152 * @s: The string to duplicate
153 * @n: Maximum number of bytes to copy, including the trailing NUL.
154 */
155char *strndup_user(const char __user *s, long n)
156{
157	char *p;
158	long length;
159
160	length = strnlen_user(s, n);
161
162	if (!length)
163		return ERR_PTR(-EFAULT);
164
165	if (length > n)
166		return ERR_PTR(-EINVAL);
167
168	p = memdup_user(s, length);
169
170	if (IS_ERR(p))
171		return p;
172
173	p[length - 1] = '\0';
174
175	return p;
176}
177EXPORT_SYMBOL(strndup_user);
178
179/**
180 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
181 *
182 * @src: source address in user space
183 * @len: number of bytes to copy
184 *
185 * Returns an ERR_PTR() on failure.
186 */
187void *memdup_user_nul(const void __user *src, size_t len)
188{
189	char *p;
190
191	/*
192	 * Always use GFP_KERNEL, since copy_from_user() can sleep and
193	 * cause pagefault, which makes it pointless to use GFP_NOFS
194	 * or GFP_ATOMIC.
195	 */
196	p = kmalloc_track_caller(len + 1, GFP_KERNEL);
197	if (!p)
198		return ERR_PTR(-ENOMEM);
199
200	if (copy_from_user(p, src, len)) {
201		kfree(p);
202		return ERR_PTR(-EFAULT);
203	}
204	p[len] = '\0';
205
206	return p;
207}
208EXPORT_SYMBOL(memdup_user_nul);
209
210void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
211		struct vm_area_struct *prev, struct rb_node *rb_parent)
212{
213	struct vm_area_struct *next;
214
215	vma->vm_prev = prev;
216	if (prev) {
217		next = prev->vm_next;
218		prev->vm_next = vma;
219	} else {
220		mm->mmap = vma;
221		if (rb_parent)
222			next = rb_entry(rb_parent,
223					struct vm_area_struct, vm_rb);
224		else
225			next = NULL;
226	}
227	vma->vm_next = next;
228	if (next)
229		next->vm_prev = vma;
230}
231
232/* Check if the vma is being used as a stack by this task */
233int vma_is_stack_for_task(struct vm_area_struct *vma, struct task_struct *t)
234{
235	return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
236}
237
238#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
239void arch_pick_mmap_layout(struct mm_struct *mm)
240{
241	mm->mmap_base = TASK_UNMAPPED_BASE;
242	mm->get_unmapped_area = arch_get_unmapped_area;
 
243}
244#endif
245
246/*
247 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
248 * back to the regular GUP.
249 * If the architecture not support this function, simply return with no
250 * page pinned
251 */
252int __weak __get_user_pages_fast(unsigned long start,
253				 int nr_pages, int write, struct page **pages)
254{
255	return 0;
256}
257EXPORT_SYMBOL_GPL(__get_user_pages_fast);
258
259/**
260 * get_user_pages_fast() - pin user pages in memory
261 * @start:	starting user address
262 * @nr_pages:	number of pages from start to pin
263 * @write:	whether pages will be written to
264 * @pages:	array that receives pointers to the pages pinned.
265 *		Should be at least nr_pages long.
266 *
267 * Returns number of pages pinned. This may be fewer than the number
268 * requested. If nr_pages is 0 or negative, returns 0. If no pages
269 * were pinned, returns -errno.
270 *
271 * get_user_pages_fast provides equivalent functionality to get_user_pages,
272 * operating on current and current->mm, with force=0 and vma=NULL. However
273 * unlike get_user_pages, it must be called without mmap_sem held.
274 *
275 * get_user_pages_fast may take mmap_sem and page table locks, so no
276 * assumptions can be made about lack of locking. get_user_pages_fast is to be
277 * implemented in a way that is advantageous (vs get_user_pages()) when the
278 * user memory area is already faulted in and present in ptes. However if the
279 * pages have to be faulted in, it may turn out to be slightly slower so
280 * callers need to carefully consider what to use. On many architectures,
281 * get_user_pages_fast simply falls back to get_user_pages.
282 */
283int __weak get_user_pages_fast(unsigned long start,
284				int nr_pages, int write, struct page **pages)
285{
286	return get_user_pages_unlocked(start, nr_pages, write, 0, pages);
287}
288EXPORT_SYMBOL_GPL(get_user_pages_fast);
289
290unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
291	unsigned long len, unsigned long prot,
292	unsigned long flag, unsigned long pgoff)
293{
294	unsigned long ret;
295	struct mm_struct *mm = current->mm;
296	unsigned long populate;
297
298	ret = security_mmap_file(file, prot, flag);
299	if (!ret) {
300		down_write(&mm->mmap_sem);
301		ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
302				    &populate);
303		up_write(&mm->mmap_sem);
304		if (populate)
305			mm_populate(ret, populate);
306	}
307	return ret;
308}
309
310unsigned long vm_mmap(struct file *file, unsigned long addr,
311	unsigned long len, unsigned long prot,
312	unsigned long flag, unsigned long offset)
313{
314	if (unlikely(offset + PAGE_ALIGN(len) < offset))
315		return -EINVAL;
316	if (unlikely(offset_in_page(offset)))
317		return -EINVAL;
318
319	return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
320}
321EXPORT_SYMBOL(vm_mmap);
322
323void kvfree(const void *addr)
324{
325	if (is_vmalloc_addr(addr))
326		vfree(addr);
327	else
328		kfree(addr);
329}
330EXPORT_SYMBOL(kvfree);
331
332static inline void *__page_rmapping(struct page *page)
333{
334	unsigned long mapping;
335
336	mapping = (unsigned long)page->mapping;
337	mapping &= ~PAGE_MAPPING_FLAGS;
338
339	return (void *)mapping;
340}
341
342/* Neutral page->mapping pointer to address_space or anon_vma or other */
343void *page_rmapping(struct page *page)
344{
345	page = compound_head(page);
346	return __page_rmapping(page);
347}
348
349struct anon_vma *page_anon_vma(struct page *page)
350{
351	unsigned long mapping;
352
353	page = compound_head(page);
354	mapping = (unsigned long)page->mapping;
355	if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
356		return NULL;
357	return __page_rmapping(page);
358}
359
360struct address_space *page_mapping(struct page *page)
361{
362	struct address_space *mapping;
363
364	page = compound_head(page);
365
366	/* This happens if someone calls flush_dcache_page on slab page */
367	if (unlikely(PageSlab(page)))
368		return NULL;
369
370	if (unlikely(PageSwapCache(page))) {
371		swp_entry_t entry;
372
373		entry.val = page_private(page);
374		return swap_address_space(entry);
375	}
376
377	mapping = page->mapping;
378	if ((unsigned long)mapping & PAGE_MAPPING_FLAGS)
379		return NULL;
380	return mapping;
381}
382
383/* Slow path of page_mapcount() for compound pages */
384int __page_mapcount(struct page *page)
385{
386	int ret;
387
388	ret = atomic_read(&page->_mapcount) + 1;
389	page = compound_head(page);
390	ret += atomic_read(compound_mapcount_ptr(page)) + 1;
391	if (PageDoubleMap(page))
392		ret--;
393	return ret;
394}
395EXPORT_SYMBOL_GPL(__page_mapcount);
396
397int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
398int sysctl_overcommit_ratio __read_mostly = 50;
399unsigned long sysctl_overcommit_kbytes __read_mostly;
400int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
401unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
402unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
403
404int overcommit_ratio_handler(struct ctl_table *table, int write,
405			     void __user *buffer, size_t *lenp,
406			     loff_t *ppos)
407{
408	int ret;
409
410	ret = proc_dointvec(table, write, buffer, lenp, ppos);
411	if (ret == 0 && write)
412		sysctl_overcommit_kbytes = 0;
413	return ret;
414}
 
415
416int overcommit_kbytes_handler(struct ctl_table *table, int write,
417			     void __user *buffer, size_t *lenp,
418			     loff_t *ppos)
419{
420	int ret;
421
422	ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
423	if (ret == 0 && write)
424		sysctl_overcommit_ratio = 0;
425	return ret;
426}
427
428/*
429 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
430 */
431unsigned long vm_commit_limit(void)
432{
433	unsigned long allowed;
434
435	if (sysctl_overcommit_kbytes)
436		allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
437	else
438		allowed = ((totalram_pages - hugetlb_total_pages())
439			   * sysctl_overcommit_ratio / 100);
440	allowed += total_swap_pages;
441
442	return allowed;
443}
444
445/*
446 * Make sure vm_committed_as in one cacheline and not cacheline shared with
447 * other variables. It can be updated by several CPUs frequently.
448 */
449struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
450
451/*
452 * The global memory commitment made in the system can be a metric
453 * that can be used to drive ballooning decisions when Linux is hosted
454 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
455 * balancing memory across competing virtual machines that are hosted.
456 * Several metrics drive this policy engine including the guest reported
457 * memory commitment.
458 */
459unsigned long vm_memory_committed(void)
460{
461	return percpu_counter_read_positive(&vm_committed_as);
462}
463EXPORT_SYMBOL_GPL(vm_memory_committed);
464
465/*
466 * Check that a process has enough memory to allocate a new virtual
467 * mapping. 0 means there is enough memory for the allocation to
468 * succeed and -ENOMEM implies there is not.
469 *
470 * We currently support three overcommit policies, which are set via the
471 * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
472 *
473 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
474 * Additional code 2002 Jul 20 by Robert Love.
475 *
476 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
477 *
478 * Note this is a helper function intended to be used by LSMs which
479 * wish to use this logic.
480 */
481int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
482{
483	long free, allowed, reserve;
484
485	VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
486			-(s64)vm_committed_as_batch * num_online_cpus(),
487			"memory commitment underflow");
488
489	vm_acct_memory(pages);
490
491	/*
492	 * Sometimes we want to use more memory than we have
493	 */
494	if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
495		return 0;
496
497	if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
498		free = global_page_state(NR_FREE_PAGES);
499		free += global_page_state(NR_FILE_PAGES);
500
501		/*
502		 * shmem pages shouldn't be counted as free in this
503		 * case, they can't be purged, only swapped out, and
504		 * that won't affect the overall amount of available
505		 * memory in the system.
506		 */
507		free -= global_page_state(NR_SHMEM);
508
509		free += get_nr_swap_pages();
510
511		/*
512		 * Any slabs which are created with the
513		 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
514		 * which are reclaimable, under pressure.  The dentry
515		 * cache and most inode caches should fall into this
516		 */
517		free += global_page_state(NR_SLAB_RECLAIMABLE);
518
519		/*
520		 * Leave reserved pages. The pages are not for anonymous pages.
521		 */
522		if (free <= totalreserve_pages)
523			goto error;
524		else
525			free -= totalreserve_pages;
526
527		/*
528		 * Reserve some for root
529		 */
530		if (!cap_sys_admin)
531			free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
532
533		if (free > pages)
534			return 0;
535
536		goto error;
537	}
538
539	allowed = vm_commit_limit();
540	/*
541	 * Reserve some for root
542	 */
543	if (!cap_sys_admin)
544		allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
545
546	/*
547	 * Don't let a single process grow so big a user can't recover
548	 */
549	if (mm) {
550		reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
551		allowed -= min_t(long, mm->total_vm / 32, reserve);
552	}
553
554	if (percpu_counter_read_positive(&vm_committed_as) < allowed)
555		return 0;
556error:
557	vm_unacct_memory(pages);
558
559	return -ENOMEM;
560}
561
562/**
563 * get_cmdline() - copy the cmdline value to a buffer.
564 * @task:     the task whose cmdline value to copy.
565 * @buffer:   the buffer to copy to.
566 * @buflen:   the length of the buffer. Larger cmdline values are truncated
567 *            to this length.
568 * Returns the size of the cmdline field copied. Note that the copy does
569 * not guarantee an ending NULL byte.
570 */
571int get_cmdline(struct task_struct *task, char *buffer, int buflen)
572{
573	int res = 0;
574	unsigned int len;
575	struct mm_struct *mm = get_task_mm(task);
576	unsigned long arg_start, arg_end, env_start, env_end;
577	if (!mm)
578		goto out;
579	if (!mm->arg_end)
580		goto out_mm;	/* Shh! No looking before we're done */
581
582	down_read(&mm->mmap_sem);
583	arg_start = mm->arg_start;
584	arg_end = mm->arg_end;
585	env_start = mm->env_start;
586	env_end = mm->env_end;
587	up_read(&mm->mmap_sem);
588
589	len = arg_end - arg_start;
590
591	if (len > buflen)
592		len = buflen;
593
594	res = access_process_vm(task, arg_start, buffer, len, 0);
595
596	/*
597	 * If the nul at the end of args has been overwritten, then
598	 * assume application is using setproctitle(3).
599	 */
600	if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
601		len = strnlen(buffer, res);
602		if (len < res) {
603			res = len;
604		} else {
605			len = env_end - env_start;
606			if (len > buflen - res)
607				len = buflen - res;
608			res += access_process_vm(task, env_start,
609						 buffer+res, len, 0);
610			res = strnlen(buffer, res);
611		}
612	}
613out_mm:
614	mmput(mm);
615out:
616	return res;
617}