Loading...
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/uaccess.h>
16
17#include "internal.h"
18
19#define CREATE_TRACE_POINTS
20#include <trace/events/kmem.h>
21
22/**
23 * kstrdup - allocate space for and copy an existing string
24 * @s: the string to duplicate
25 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
26 */
27char *kstrdup(const char *s, gfp_t gfp)
28{
29 size_t len;
30 char *buf;
31
32 if (!s)
33 return NULL;
34
35 len = strlen(s) + 1;
36 buf = kmalloc_track_caller(len, gfp);
37 if (buf)
38 memcpy(buf, s, len);
39 return buf;
40}
41EXPORT_SYMBOL(kstrdup);
42
43/**
44 * kstrndup - allocate space for and copy an existing string
45 * @s: the string to duplicate
46 * @max: read at most @max chars from @s
47 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
48 */
49char *kstrndup(const char *s, size_t max, gfp_t gfp)
50{
51 size_t len;
52 char *buf;
53
54 if (!s)
55 return NULL;
56
57 len = strnlen(s, max);
58 buf = kmalloc_track_caller(len+1, gfp);
59 if (buf) {
60 memcpy(buf, s, len);
61 buf[len] = '\0';
62 }
63 return buf;
64}
65EXPORT_SYMBOL(kstrndup);
66
67/**
68 * kmemdup - duplicate region of memory
69 *
70 * @src: memory region to duplicate
71 * @len: memory region length
72 * @gfp: GFP mask to use
73 */
74void *kmemdup(const void *src, size_t len, gfp_t gfp)
75{
76 void *p;
77
78 p = kmalloc_track_caller(len, gfp);
79 if (p)
80 memcpy(p, src, len);
81 return p;
82}
83EXPORT_SYMBOL(kmemdup);
84
85/**
86 * memdup_user - duplicate memory region from user space
87 *
88 * @src: source address in user space
89 * @len: number of bytes to copy
90 *
91 * Returns an ERR_PTR() on failure.
92 */
93void *memdup_user(const void __user *src, size_t len)
94{
95 void *p;
96
97 /*
98 * Always use GFP_KERNEL, since copy_from_user() can sleep and
99 * cause pagefault, which makes it pointless to use GFP_NOFS
100 * or GFP_ATOMIC.
101 */
102 p = kmalloc_track_caller(len, GFP_KERNEL);
103 if (!p)
104 return ERR_PTR(-ENOMEM);
105
106 if (copy_from_user(p, src, len)) {
107 kfree(p);
108 return ERR_PTR(-EFAULT);
109 }
110
111 return p;
112}
113EXPORT_SYMBOL(memdup_user);
114
115static __always_inline void *__do_krealloc(const void *p, size_t new_size,
116 gfp_t flags)
117{
118 void *ret;
119 size_t ks = 0;
120
121 if (p)
122 ks = ksize(p);
123
124 if (ks >= new_size)
125 return (void *)p;
126
127 ret = kmalloc_track_caller(new_size, flags);
128 if (ret && p)
129 memcpy(ret, p, ks);
130
131 return ret;
132}
133
134/**
135 * __krealloc - like krealloc() but don't free @p.
136 * @p: object to reallocate memory for.
137 * @new_size: how many bytes of memory are required.
138 * @flags: the type of memory to allocate.
139 *
140 * This function is like krealloc() except it never frees the originally
141 * allocated buffer. Use this if you don't want to free the buffer immediately
142 * like, for example, with RCU.
143 */
144void *__krealloc(const void *p, size_t new_size, gfp_t flags)
145{
146 if (unlikely(!new_size))
147 return ZERO_SIZE_PTR;
148
149 return __do_krealloc(p, new_size, flags);
150
151}
152EXPORT_SYMBOL(__krealloc);
153
154/**
155 * krealloc - reallocate memory. The contents will remain unchanged.
156 * @p: object to reallocate memory for.
157 * @new_size: how many bytes of memory are required.
158 * @flags: the type of memory to allocate.
159 *
160 * The contents of the object pointed to are preserved up to the
161 * lesser of the new and old sizes. If @p is %NULL, krealloc()
162 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
163 * %NULL pointer, the object pointed to is freed.
164 */
165void *krealloc(const void *p, size_t new_size, gfp_t flags)
166{
167 void *ret;
168
169 if (unlikely(!new_size)) {
170 kfree(p);
171 return ZERO_SIZE_PTR;
172 }
173
174 ret = __do_krealloc(p, new_size, flags);
175 if (ret && p != ret)
176 kfree(p);
177
178 return ret;
179}
180EXPORT_SYMBOL(krealloc);
181
182/**
183 * kzfree - like kfree but zero memory
184 * @p: object to free memory of
185 *
186 * The memory of the object @p points to is zeroed before freed.
187 * If @p is %NULL, kzfree() does nothing.
188 *
189 * Note: this function zeroes the whole allocated buffer which can be a good
190 * deal bigger than the requested buffer size passed to kmalloc(). So be
191 * careful when using this function in performance sensitive code.
192 */
193void kzfree(const void *p)
194{
195 size_t ks;
196 void *mem = (void *)p;
197
198 if (unlikely(ZERO_OR_NULL_PTR(mem)))
199 return;
200 ks = ksize(mem);
201 memset(mem, 0, ks);
202 kfree(mem);
203}
204EXPORT_SYMBOL(kzfree);
205
206/*
207 * strndup_user - duplicate an existing string from user space
208 * @s: The string to duplicate
209 * @n: Maximum number of bytes to copy, including the trailing NUL.
210 */
211char *strndup_user(const char __user *s, long n)
212{
213 char *p;
214 long length;
215
216 length = strnlen_user(s, n);
217
218 if (!length)
219 return ERR_PTR(-EFAULT);
220
221 if (length > n)
222 return ERR_PTR(-EINVAL);
223
224 p = memdup_user(s, length);
225
226 if (IS_ERR(p))
227 return p;
228
229 p[length - 1] = '\0';
230
231 return p;
232}
233EXPORT_SYMBOL(strndup_user);
234
235void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
236 struct vm_area_struct *prev, struct rb_node *rb_parent)
237{
238 struct vm_area_struct *next;
239
240 vma->vm_prev = prev;
241 if (prev) {
242 next = prev->vm_next;
243 prev->vm_next = vma;
244 } else {
245 mm->mmap = vma;
246 if (rb_parent)
247 next = rb_entry(rb_parent,
248 struct vm_area_struct, vm_rb);
249 else
250 next = NULL;
251 }
252 vma->vm_next = next;
253 if (next)
254 next->vm_prev = vma;
255}
256
257/* Check if the vma is being used as a stack by this task */
258static int vm_is_stack_for_task(struct task_struct *t,
259 struct vm_area_struct *vma)
260{
261 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
262}
263
264/*
265 * Check if the vma is being used as a stack.
266 * If is_group is non-zero, check in the entire thread group or else
267 * just check in the current task. Returns the pid of the task that
268 * the vma is stack for.
269 */
270pid_t vm_is_stack(struct task_struct *task,
271 struct vm_area_struct *vma, int in_group)
272{
273 pid_t ret = 0;
274
275 if (vm_is_stack_for_task(task, vma))
276 return task->pid;
277
278 if (in_group) {
279 struct task_struct *t;
280 rcu_read_lock();
281 if (!pid_alive(task))
282 goto done;
283
284 t = task;
285 do {
286 if (vm_is_stack_for_task(t, vma)) {
287 ret = t->pid;
288 goto done;
289 }
290 } while_each_thread(task, t);
291done:
292 rcu_read_unlock();
293 }
294
295 return ret;
296}
297
298#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
299void arch_pick_mmap_layout(struct mm_struct *mm)
300{
301 mm->mmap_base = TASK_UNMAPPED_BASE;
302 mm->get_unmapped_area = arch_get_unmapped_area;
303}
304#endif
305
306/*
307 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
308 * back to the regular GUP.
309 * If the architecture not support this function, simply return with no
310 * page pinned
311 */
312int __weak __get_user_pages_fast(unsigned long start,
313 int nr_pages, int write, struct page **pages)
314{
315 return 0;
316}
317EXPORT_SYMBOL_GPL(__get_user_pages_fast);
318
319/**
320 * get_user_pages_fast() - pin user pages in memory
321 * @start: starting user address
322 * @nr_pages: number of pages from start to pin
323 * @write: whether pages will be written to
324 * @pages: array that receives pointers to the pages pinned.
325 * Should be at least nr_pages long.
326 *
327 * Returns number of pages pinned. This may be fewer than the number
328 * requested. If nr_pages is 0 or negative, returns 0. If no pages
329 * were pinned, returns -errno.
330 *
331 * get_user_pages_fast provides equivalent functionality to get_user_pages,
332 * operating on current and current->mm, with force=0 and vma=NULL. However
333 * unlike get_user_pages, it must be called without mmap_sem held.
334 *
335 * get_user_pages_fast may take mmap_sem and page table locks, so no
336 * assumptions can be made about lack of locking. get_user_pages_fast is to be
337 * implemented in a way that is advantageous (vs get_user_pages()) when the
338 * user memory area is already faulted in and present in ptes. However if the
339 * pages have to be faulted in, it may turn out to be slightly slower so
340 * callers need to carefully consider what to use. On many architectures,
341 * get_user_pages_fast simply falls back to get_user_pages.
342 */
343int __weak get_user_pages_fast(unsigned long start,
344 int nr_pages, int write, struct page **pages)
345{
346 struct mm_struct *mm = current->mm;
347 int ret;
348
349 down_read(&mm->mmap_sem);
350 ret = get_user_pages(current, mm, start, nr_pages,
351 write, 0, pages, NULL);
352 up_read(&mm->mmap_sem);
353
354 return ret;
355}
356EXPORT_SYMBOL_GPL(get_user_pages_fast);
357
358unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
359 unsigned long len, unsigned long prot,
360 unsigned long flag, unsigned long pgoff)
361{
362 unsigned long ret;
363 struct mm_struct *mm = current->mm;
364 unsigned long populate;
365
366 ret = security_mmap_file(file, prot, flag);
367 if (!ret) {
368 down_write(&mm->mmap_sem);
369 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
370 &populate);
371 up_write(&mm->mmap_sem);
372 if (populate)
373 mm_populate(ret, populate);
374 }
375 return ret;
376}
377
378unsigned long vm_mmap(struct file *file, unsigned long addr,
379 unsigned long len, unsigned long prot,
380 unsigned long flag, unsigned long offset)
381{
382 if (unlikely(offset + PAGE_ALIGN(len) < offset))
383 return -EINVAL;
384 if (unlikely(offset & ~PAGE_MASK))
385 return -EINVAL;
386
387 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
388}
389EXPORT_SYMBOL(vm_mmap);
390
391void kvfree(const void *addr)
392{
393 if (is_vmalloc_addr(addr))
394 vfree(addr);
395 else
396 kfree(addr);
397}
398EXPORT_SYMBOL(kvfree);
399
400struct address_space *page_mapping(struct page *page)
401{
402 struct address_space *mapping = page->mapping;
403
404 /* This happens if someone calls flush_dcache_page on slab page */
405 if (unlikely(PageSlab(page)))
406 return NULL;
407
408 if (unlikely(PageSwapCache(page))) {
409 swp_entry_t entry;
410
411 entry.val = page_private(page);
412 mapping = swap_address_space(entry);
413 } else if ((unsigned long)mapping & PAGE_MAPPING_ANON)
414 mapping = NULL;
415 return mapping;
416}
417
418int overcommit_ratio_handler(struct ctl_table *table, int write,
419 void __user *buffer, size_t *lenp,
420 loff_t *ppos)
421{
422 int ret;
423
424 ret = proc_dointvec(table, write, buffer, lenp, ppos);
425 if (ret == 0 && write)
426 sysctl_overcommit_kbytes = 0;
427 return ret;
428}
429
430int overcommit_kbytes_handler(struct ctl_table *table, int write,
431 void __user *buffer, size_t *lenp,
432 loff_t *ppos)
433{
434 int ret;
435
436 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
437 if (ret == 0 && write)
438 sysctl_overcommit_ratio = 0;
439 return ret;
440}
441
442/*
443 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
444 */
445unsigned long vm_commit_limit(void)
446{
447 unsigned long allowed;
448
449 if (sysctl_overcommit_kbytes)
450 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
451 else
452 allowed = ((totalram_pages - hugetlb_total_pages())
453 * sysctl_overcommit_ratio / 100);
454 allowed += total_swap_pages;
455
456 return allowed;
457}
458
459/**
460 * get_cmdline() - copy the cmdline value to a buffer.
461 * @task: the task whose cmdline value to copy.
462 * @buffer: the buffer to copy to.
463 * @buflen: the length of the buffer. Larger cmdline values are truncated
464 * to this length.
465 * Returns the size of the cmdline field copied. Note that the copy does
466 * not guarantee an ending NULL byte.
467 */
468int get_cmdline(struct task_struct *task, char *buffer, int buflen)
469{
470 int res = 0;
471 unsigned int len;
472 struct mm_struct *mm = get_task_mm(task);
473 if (!mm)
474 goto out;
475 if (!mm->arg_end)
476 goto out_mm; /* Shh! No looking before we're done */
477
478 len = mm->arg_end - mm->arg_start;
479
480 if (len > buflen)
481 len = buflen;
482
483 res = access_process_vm(task, mm->arg_start, buffer, len, 0);
484
485 /*
486 * If the nul at the end of args has been overwritten, then
487 * assume application is using setproctitle(3).
488 */
489 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
490 len = strnlen(buffer, res);
491 if (len < res) {
492 res = len;
493 } else {
494 len = mm->env_end - mm->env_start;
495 if (len > buflen - res)
496 len = buflen - res;
497 res += access_process_vm(task, mm->env_start,
498 buffer+res, len, 0);
499 res = strnlen(buffer, res);
500 }
501 }
502out_mm:
503 mmput(mm);
504out:
505 return res;
506}
507
508/* Tracepoints definitions. */
509EXPORT_TRACEPOINT_SYMBOL(kmalloc);
510EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
511EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
512EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
513EXPORT_TRACEPOINT_SYMBOL(kfree);
514EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
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 <linux/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_current(struct vm_area_struct *vma)
234{
235 struct task_struct * __maybe_unused t = current;
236
237 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
238}
239
240#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
241void arch_pick_mmap_layout(struct mm_struct *mm)
242{
243 mm->mmap_base = TASK_UNMAPPED_BASE;
244 mm->get_unmapped_area = arch_get_unmapped_area;
245}
246#endif
247
248/*
249 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
250 * back to the regular GUP.
251 * If the architecture not support this function, simply return with no
252 * page pinned
253 */
254int __weak __get_user_pages_fast(unsigned long start,
255 int nr_pages, int write, struct page **pages)
256{
257 return 0;
258}
259EXPORT_SYMBOL_GPL(__get_user_pages_fast);
260
261/**
262 * get_user_pages_fast() - pin user pages in memory
263 * @start: starting user address
264 * @nr_pages: number of pages from start to pin
265 * @write: whether pages will be written to
266 * @pages: array that receives pointers to the pages pinned.
267 * Should be at least nr_pages long.
268 *
269 * Returns number of pages pinned. This may be fewer than the number
270 * requested. If nr_pages is 0 or negative, returns 0. If no pages
271 * were pinned, returns -errno.
272 *
273 * get_user_pages_fast provides equivalent functionality to get_user_pages,
274 * operating on current and current->mm, with force=0 and vma=NULL. However
275 * unlike get_user_pages, it must be called without mmap_sem held.
276 *
277 * get_user_pages_fast may take mmap_sem and page table locks, so no
278 * assumptions can be made about lack of locking. get_user_pages_fast is to be
279 * implemented in a way that is advantageous (vs get_user_pages()) when the
280 * user memory area is already faulted in and present in ptes. However if the
281 * pages have to be faulted in, it may turn out to be slightly slower so
282 * callers need to carefully consider what to use. On many architectures,
283 * get_user_pages_fast simply falls back to get_user_pages.
284 */
285int __weak get_user_pages_fast(unsigned long start,
286 int nr_pages, int write, struct page **pages)
287{
288 return get_user_pages_unlocked(start, nr_pages, pages,
289 write ? FOLL_WRITE : 0);
290}
291EXPORT_SYMBOL_GPL(get_user_pages_fast);
292
293unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
294 unsigned long len, unsigned long prot,
295 unsigned long flag, unsigned long pgoff)
296{
297 unsigned long ret;
298 struct mm_struct *mm = current->mm;
299 unsigned long populate;
300
301 ret = security_mmap_file(file, prot, flag);
302 if (!ret) {
303 if (down_write_killable(&mm->mmap_sem))
304 return -EINTR;
305 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
306 &populate);
307 up_write(&mm->mmap_sem);
308 if (populate)
309 mm_populate(ret, populate);
310 }
311 return ret;
312}
313
314unsigned long vm_mmap(struct file *file, unsigned long addr,
315 unsigned long len, unsigned long prot,
316 unsigned long flag, unsigned long offset)
317{
318 if (unlikely(offset + PAGE_ALIGN(len) < offset))
319 return -EINVAL;
320 if (unlikely(offset_in_page(offset)))
321 return -EINVAL;
322
323 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
324}
325EXPORT_SYMBOL(vm_mmap);
326
327void kvfree(const void *addr)
328{
329 if (is_vmalloc_addr(addr))
330 vfree(addr);
331 else
332 kfree(addr);
333}
334EXPORT_SYMBOL(kvfree);
335
336static inline void *__page_rmapping(struct page *page)
337{
338 unsigned long mapping;
339
340 mapping = (unsigned long)page->mapping;
341 mapping &= ~PAGE_MAPPING_FLAGS;
342
343 return (void *)mapping;
344}
345
346/* Neutral page->mapping pointer to address_space or anon_vma or other */
347void *page_rmapping(struct page *page)
348{
349 page = compound_head(page);
350 return __page_rmapping(page);
351}
352
353/*
354 * Return true if this page is mapped into pagetables.
355 * For compound page it returns true if any subpage of compound page is mapped.
356 */
357bool page_mapped(struct page *page)
358{
359 int i;
360
361 if (likely(!PageCompound(page)))
362 return atomic_read(&page->_mapcount) >= 0;
363 page = compound_head(page);
364 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
365 return true;
366 if (PageHuge(page))
367 return false;
368 for (i = 0; i < hpage_nr_pages(page); i++) {
369 if (atomic_read(&page[i]._mapcount) >= 0)
370 return true;
371 }
372 return false;
373}
374EXPORT_SYMBOL(page_mapped);
375
376struct anon_vma *page_anon_vma(struct page *page)
377{
378 unsigned long mapping;
379
380 page = compound_head(page);
381 mapping = (unsigned long)page->mapping;
382 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
383 return NULL;
384 return __page_rmapping(page);
385}
386
387struct address_space *page_mapping(struct page *page)
388{
389 struct address_space *mapping;
390
391 page = compound_head(page);
392
393 /* This happens if someone calls flush_dcache_page on slab page */
394 if (unlikely(PageSlab(page)))
395 return NULL;
396
397 if (unlikely(PageSwapCache(page))) {
398 swp_entry_t entry;
399
400 entry.val = page_private(page);
401 return swap_address_space(entry);
402 }
403
404 mapping = page->mapping;
405 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
406 return NULL;
407
408 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
409}
410EXPORT_SYMBOL(page_mapping);
411
412/* Slow path of page_mapcount() for compound pages */
413int __page_mapcount(struct page *page)
414{
415 int ret;
416
417 ret = atomic_read(&page->_mapcount) + 1;
418 /*
419 * For file THP page->_mapcount contains total number of mapping
420 * of the page: no need to look into compound_mapcount.
421 */
422 if (!PageAnon(page) && !PageHuge(page))
423 return ret;
424 page = compound_head(page);
425 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
426 if (PageDoubleMap(page))
427 ret--;
428 return ret;
429}
430EXPORT_SYMBOL_GPL(__page_mapcount);
431
432int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
433int sysctl_overcommit_ratio __read_mostly = 50;
434unsigned long sysctl_overcommit_kbytes __read_mostly;
435int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
436unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
437unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
438
439int overcommit_ratio_handler(struct ctl_table *table, int write,
440 void __user *buffer, size_t *lenp,
441 loff_t *ppos)
442{
443 int ret;
444
445 ret = proc_dointvec(table, write, buffer, lenp, ppos);
446 if (ret == 0 && write)
447 sysctl_overcommit_kbytes = 0;
448 return ret;
449}
450
451int overcommit_kbytes_handler(struct ctl_table *table, int write,
452 void __user *buffer, size_t *lenp,
453 loff_t *ppos)
454{
455 int ret;
456
457 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
458 if (ret == 0 && write)
459 sysctl_overcommit_ratio = 0;
460 return ret;
461}
462
463/*
464 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
465 */
466unsigned long vm_commit_limit(void)
467{
468 unsigned long allowed;
469
470 if (sysctl_overcommit_kbytes)
471 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
472 else
473 allowed = ((totalram_pages - hugetlb_total_pages())
474 * sysctl_overcommit_ratio / 100);
475 allowed += total_swap_pages;
476
477 return allowed;
478}
479
480/*
481 * Make sure vm_committed_as in one cacheline and not cacheline shared with
482 * other variables. It can be updated by several CPUs frequently.
483 */
484struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
485
486/*
487 * The global memory commitment made in the system can be a metric
488 * that can be used to drive ballooning decisions when Linux is hosted
489 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
490 * balancing memory across competing virtual machines that are hosted.
491 * Several metrics drive this policy engine including the guest reported
492 * memory commitment.
493 */
494unsigned long vm_memory_committed(void)
495{
496 return percpu_counter_read_positive(&vm_committed_as);
497}
498EXPORT_SYMBOL_GPL(vm_memory_committed);
499
500/*
501 * Check that a process has enough memory to allocate a new virtual
502 * mapping. 0 means there is enough memory for the allocation to
503 * succeed and -ENOMEM implies there is not.
504 *
505 * We currently support three overcommit policies, which are set via the
506 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
507 *
508 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
509 * Additional code 2002 Jul 20 by Robert Love.
510 *
511 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
512 *
513 * Note this is a helper function intended to be used by LSMs which
514 * wish to use this logic.
515 */
516int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
517{
518 long free, allowed, reserve;
519
520 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
521 -(s64)vm_committed_as_batch * num_online_cpus(),
522 "memory commitment underflow");
523
524 vm_acct_memory(pages);
525
526 /*
527 * Sometimes we want to use more memory than we have
528 */
529 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
530 return 0;
531
532 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
533 free = global_page_state(NR_FREE_PAGES);
534 free += global_node_page_state(NR_FILE_PAGES);
535
536 /*
537 * shmem pages shouldn't be counted as free in this
538 * case, they can't be purged, only swapped out, and
539 * that won't affect the overall amount of available
540 * memory in the system.
541 */
542 free -= global_node_page_state(NR_SHMEM);
543
544 free += get_nr_swap_pages();
545
546 /*
547 * Any slabs which are created with the
548 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
549 * which are reclaimable, under pressure. The dentry
550 * cache and most inode caches should fall into this
551 */
552 free += global_page_state(NR_SLAB_RECLAIMABLE);
553
554 /*
555 * Leave reserved pages. The pages are not for anonymous pages.
556 */
557 if (free <= totalreserve_pages)
558 goto error;
559 else
560 free -= totalreserve_pages;
561
562 /*
563 * Reserve some for root
564 */
565 if (!cap_sys_admin)
566 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
567
568 if (free > pages)
569 return 0;
570
571 goto error;
572 }
573
574 allowed = vm_commit_limit();
575 /*
576 * Reserve some for root
577 */
578 if (!cap_sys_admin)
579 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
580
581 /*
582 * Don't let a single process grow so big a user can't recover
583 */
584 if (mm) {
585 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
586 allowed -= min_t(long, mm->total_vm / 32, reserve);
587 }
588
589 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
590 return 0;
591error:
592 vm_unacct_memory(pages);
593
594 return -ENOMEM;
595}
596
597/**
598 * get_cmdline() - copy the cmdline value to a buffer.
599 * @task: the task whose cmdline value to copy.
600 * @buffer: the buffer to copy to.
601 * @buflen: the length of the buffer. Larger cmdline values are truncated
602 * to this length.
603 * Returns the size of the cmdline field copied. Note that the copy does
604 * not guarantee an ending NULL byte.
605 */
606int get_cmdline(struct task_struct *task, char *buffer, int buflen)
607{
608 int res = 0;
609 unsigned int len;
610 struct mm_struct *mm = get_task_mm(task);
611 unsigned long arg_start, arg_end, env_start, env_end;
612 if (!mm)
613 goto out;
614 if (!mm->arg_end)
615 goto out_mm; /* Shh! No looking before we're done */
616
617 down_read(&mm->mmap_sem);
618 arg_start = mm->arg_start;
619 arg_end = mm->arg_end;
620 env_start = mm->env_start;
621 env_end = mm->env_end;
622 up_read(&mm->mmap_sem);
623
624 len = arg_end - arg_start;
625
626 if (len > buflen)
627 len = buflen;
628
629 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
630
631 /*
632 * If the nul at the end of args has been overwritten, then
633 * assume application is using setproctitle(3).
634 */
635 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
636 len = strnlen(buffer, res);
637 if (len < res) {
638 res = len;
639 } else {
640 len = env_end - env_start;
641 if (len > buflen - res)
642 len = buflen - res;
643 res += access_process_vm(task, env_start,
644 buffer+res, len,
645 FOLL_FORCE);
646 res = strnlen(buffer, res);
647 }
648 }
649out_mm:
650 mmput(mm);
651out:
652 return res;
653}