2 #include <linux/slab.h>
3 #include <linux/string.h>
4 #include <linux/compiler.h>
5 #include <linux/export.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>
15 #include <asm/sections.h>
16 #include <asm/uaccess.h>
20 static inline int is_kernel_rodata(unsigned long addr
)
22 return addr
>= (unsigned long)__start_rodata
&&
23 addr
< (unsigned long)__end_rodata
;
27 * kfree_const - conditionally free memory
28 * @x: pointer to the memory
30 * Function calls kfree only if @x is not in .rodata section.
32 void kfree_const(const void *x
)
34 if (!is_kernel_rodata((unsigned long)x
))
37 EXPORT_SYMBOL(kfree_const
);
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
44 char *kstrdup(const char *s
, gfp_t gfp
)
53 buf
= kmalloc_track_caller(len
, gfp
);
58 EXPORT_SYMBOL(kstrdup
);
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
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.
69 const char *kstrdup_const(const char *s
, gfp_t gfp
)
71 if (is_kernel_rodata((unsigned long)s
))
74 return kstrdup(s
, gfp
);
76 EXPORT_SYMBOL(kstrdup_const
);
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
84 char *kstrndup(const char *s
, size_t max
, gfp_t gfp
)
92 len
= strnlen(s
, max
);
93 buf
= kmalloc_track_caller(len
+1, gfp
);
100 EXPORT_SYMBOL(kstrndup
);
103 * kmemdup - duplicate region of memory
105 * @src: memory region to duplicate
106 * @len: memory region length
107 * @gfp: GFP mask to use
109 void *kmemdup(const void *src
, size_t len
, gfp_t gfp
)
113 p
= kmalloc_track_caller(len
, gfp
);
118 EXPORT_SYMBOL(kmemdup
);
121 * memdup_user - duplicate memory region from user space
123 * @src: source address in user space
124 * @len: number of bytes to copy
126 * Returns an ERR_PTR() on failure.
128 void *memdup_user(const void __user
*src
, size_t len
)
133 * Always use GFP_KERNEL, since copy_from_user() can sleep and
134 * cause pagefault, which makes it pointless to use GFP_NOFS
137 p
= kmalloc_track_caller(len
, GFP_KERNEL
);
139 return ERR_PTR(-ENOMEM
);
141 if (copy_from_user(p
, src
, len
)) {
143 return ERR_PTR(-EFAULT
);
148 EXPORT_SYMBOL(memdup_user
);
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.
155 char *strndup_user(const char __user
*s
, long n
)
160 length
= strnlen_user(s
, n
);
163 return ERR_PTR(-EFAULT
);
166 return ERR_PTR(-EINVAL
);
168 p
= memdup_user(s
, length
);
173 p
[length
- 1] = '\0';
177 EXPORT_SYMBOL(strndup_user
);
180 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
182 * @src: source address in user space
183 * @len: number of bytes to copy
185 * Returns an ERR_PTR() on failure.
187 void *memdup_user_nul(const void __user
*src
, size_t len
)
192 * Always use GFP_KERNEL, since copy_from_user() can sleep and
193 * cause pagefault, which makes it pointless to use GFP_NOFS
196 p
= kmalloc_track_caller(len
+ 1, GFP_KERNEL
);
198 return ERR_PTR(-ENOMEM
);
200 if (copy_from_user(p
, src
, len
)) {
202 return ERR_PTR(-EFAULT
);
208 EXPORT_SYMBOL(memdup_user_nul
);
210 void __vma_link_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
211 struct vm_area_struct
*prev
, struct rb_node
*rb_parent
)
213 struct vm_area_struct
*next
;
217 next
= prev
->vm_next
;
222 next
= rb_entry(rb_parent
,
223 struct vm_area_struct
, vm_rb
);
232 /* Check if the vma is being used as a stack by this task */
233 int vma_is_stack_for_task(struct vm_area_struct
*vma
, struct task_struct
*t
)
235 return (vma
->vm_start
<= KSTK_ESP(t
) && vma
->vm_end
>= KSTK_ESP(t
));
238 #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
239 void arch_pick_mmap_layout(struct mm_struct
*mm
)
241 mm
->mmap_base
= TASK_UNMAPPED_BASE
;
242 mm
->get_unmapped_area
= arch_get_unmapped_area
;
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
252 int __weak
__get_user_pages_fast(unsigned long start
,
253 int nr_pages
, int write
, struct page
**pages
)
257 EXPORT_SYMBOL_GPL(__get_user_pages_fast
);
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.
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.
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.
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.
283 int __weak
get_user_pages_fast(unsigned long start
,
284 int nr_pages
, int write
, struct page
**pages
)
286 struct mm_struct
*mm
= current
->mm
;
287 return get_user_pages_unlocked(current
, mm
, start
, nr_pages
,
290 EXPORT_SYMBOL_GPL(get_user_pages_fast
);
292 unsigned long vm_mmap_pgoff(struct file
*file
, unsigned long addr
,
293 unsigned long len
, unsigned long prot
,
294 unsigned long flag
, unsigned long pgoff
)
297 struct mm_struct
*mm
= current
->mm
;
298 unsigned long populate
;
300 ret
= security_mmap_file(file
, prot
, flag
);
302 down_write(&mm
->mmap_sem
);
303 ret
= do_mmap_pgoff(file
, addr
, len
, prot
, flag
, pgoff
,
305 up_write(&mm
->mmap_sem
);
307 mm_populate(ret
, populate
);
312 unsigned long vm_mmap(struct file
*file
, unsigned long addr
,
313 unsigned long len
, unsigned long prot
,
314 unsigned long flag
, unsigned long offset
)
316 if (unlikely(offset
+ PAGE_ALIGN(len
) < offset
))
318 if (unlikely(offset_in_page(offset
)))
321 return vm_mmap_pgoff(file
, addr
, len
, prot
, flag
, offset
>> PAGE_SHIFT
);
323 EXPORT_SYMBOL(vm_mmap
);
325 void kvfree(const void *addr
)
327 if (is_vmalloc_addr(addr
))
332 EXPORT_SYMBOL(kvfree
);
334 static inline void *__page_rmapping(struct page
*page
)
336 unsigned long mapping
;
338 mapping
= (unsigned long)page
->mapping
;
339 mapping
&= ~PAGE_MAPPING_FLAGS
;
341 return (void *)mapping
;
344 /* Neutral page->mapping pointer to address_space or anon_vma or other */
345 void *page_rmapping(struct page
*page
)
347 page
= compound_head(page
);
348 return __page_rmapping(page
);
351 struct anon_vma
*page_anon_vma(struct page
*page
)
353 unsigned long mapping
;
355 page
= compound_head(page
);
356 mapping
= (unsigned long)page
->mapping
;
357 if ((mapping
& PAGE_MAPPING_FLAGS
) != PAGE_MAPPING_ANON
)
359 return __page_rmapping(page
);
362 struct address_space
*page_mapping(struct page
*page
)
364 struct address_space
*mapping
;
366 page
= compound_head(page
);
368 /* This happens if someone calls flush_dcache_page on slab page */
369 if (unlikely(PageSlab(page
)))
372 if (unlikely(PageSwapCache(page
))) {
375 entry
.val
= page_private(page
);
376 return swap_address_space(entry
);
379 mapping
= page
->mapping
;
380 if ((unsigned long)mapping
& PAGE_MAPPING_FLAGS
)
385 /* Slow path of page_mapcount() for compound pages */
386 int __page_mapcount(struct page
*page
)
390 ret
= atomic_read(&page
->_mapcount
) + 1;
391 page
= compound_head(page
);
392 ret
+= atomic_read(compound_mapcount_ptr(page
)) + 1;
393 if (PageDoubleMap(page
))
397 EXPORT_SYMBOL_GPL(__page_mapcount
);
399 int sysctl_overcommit_memory __read_mostly
= OVERCOMMIT_GUESS
;
400 int sysctl_overcommit_ratio __read_mostly
= 50;
401 unsigned long sysctl_overcommit_kbytes __read_mostly
;
402 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
403 unsigned long sysctl_user_reserve_kbytes __read_mostly
= 1UL << 17; /* 128MB */
404 unsigned long sysctl_admin_reserve_kbytes __read_mostly
= 1UL << 13; /* 8MB */
406 int overcommit_ratio_handler(struct ctl_table
*table
, int write
,
407 void __user
*buffer
, size_t *lenp
,
412 ret
= proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
413 if (ret
== 0 && write
)
414 sysctl_overcommit_kbytes
= 0;
418 int overcommit_kbytes_handler(struct ctl_table
*table
, int write
,
419 void __user
*buffer
, size_t *lenp
,
424 ret
= proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
425 if (ret
== 0 && write
)
426 sysctl_overcommit_ratio
= 0;
431 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
433 unsigned long vm_commit_limit(void)
435 unsigned long allowed
;
437 if (sysctl_overcommit_kbytes
)
438 allowed
= sysctl_overcommit_kbytes
>> (PAGE_SHIFT
- 10);
440 allowed
= ((totalram_pages
- hugetlb_total_pages())
441 * sysctl_overcommit_ratio
/ 100);
442 allowed
+= total_swap_pages
;
448 * Make sure vm_committed_as in one cacheline and not cacheline shared with
449 * other variables. It can be updated by several CPUs frequently.
451 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp
;
454 * The global memory commitment made in the system can be a metric
455 * that can be used to drive ballooning decisions when Linux is hosted
456 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
457 * balancing memory across competing virtual machines that are hosted.
458 * Several metrics drive this policy engine including the guest reported
461 unsigned long vm_memory_committed(void)
463 return percpu_counter_read_positive(&vm_committed_as
);
465 EXPORT_SYMBOL_GPL(vm_memory_committed
);
468 * Check that a process has enough memory to allocate a new virtual
469 * mapping. 0 means there is enough memory for the allocation to
470 * succeed and -ENOMEM implies there is not.
472 * We currently support three overcommit policies, which are set via the
473 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
475 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
476 * Additional code 2002 Jul 20 by Robert Love.
478 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
480 * Note this is a helper function intended to be used by LSMs which
481 * wish to use this logic.
483 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
485 long free
, allowed
, reserve
;
487 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as
) <
488 -(s64
)vm_committed_as_batch
* num_online_cpus(),
489 "memory commitment underflow");
491 vm_acct_memory(pages
);
494 * Sometimes we want to use more memory than we have
496 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
499 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
500 free
= global_page_state(NR_FREE_PAGES
);
501 free
+= global_page_state(NR_FILE_PAGES
);
504 * shmem pages shouldn't be counted as free in this
505 * case, they can't be purged, only swapped out, and
506 * that won't affect the overall amount of available
507 * memory in the system.
509 free
-= global_page_state(NR_SHMEM
);
511 free
+= get_nr_swap_pages();
514 * Any slabs which are created with the
515 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
516 * which are reclaimable, under pressure. The dentry
517 * cache and most inode caches should fall into this
519 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
522 * Leave reserved pages. The pages are not for anonymous pages.
524 if (free
<= totalreserve_pages
)
527 free
-= totalreserve_pages
;
530 * Reserve some for root
533 free
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
541 allowed
= vm_commit_limit();
543 * Reserve some for root
546 allowed
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
549 * Don't let a single process grow so big a user can't recover
552 reserve
= sysctl_user_reserve_kbytes
>> (PAGE_SHIFT
- 10);
553 allowed
-= min_t(long, mm
->total_vm
/ 32, reserve
);
556 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
559 vm_unacct_memory(pages
);
565 * get_cmdline() - copy the cmdline value to a buffer.
566 * @task: the task whose cmdline value to copy.
567 * @buffer: the buffer to copy to.
568 * @buflen: the length of the buffer. Larger cmdline values are truncated
570 * Returns the size of the cmdline field copied. Note that the copy does
571 * not guarantee an ending NULL byte.
573 int get_cmdline(struct task_struct
*task
, char *buffer
, int buflen
)
577 struct mm_struct
*mm
= get_task_mm(task
);
578 unsigned long arg_start
, arg_end
, env_start
, env_end
;
582 goto out_mm
; /* Shh! No looking before we're done */
584 down_read(&mm
->mmap_sem
);
585 arg_start
= mm
->arg_start
;
586 arg_end
= mm
->arg_end
;
587 env_start
= mm
->env_start
;
588 env_end
= mm
->env_end
;
589 up_read(&mm
->mmap_sem
);
591 len
= arg_end
- arg_start
;
596 res
= access_process_vm(task
, arg_start
, buffer
, len
, 0);
599 * If the nul at the end of args has been overwritten, then
600 * assume application is using setproctitle(3).
602 if (res
> 0 && buffer
[res
-1] != '\0' && len
< buflen
) {
603 len
= strnlen(buffer
, res
);
607 len
= env_end
- env_start
;
608 if (len
> buflen
- res
)
610 res
+= access_process_vm(task
, env_start
,
612 res
= strnlen(buffer
, res
);