2 #include <linux/slab.h>
3 #include <linux/string.h>
4 #include <linux/compiler.h>
5 #include <linux/export.h>
6 #include <linux/ctype.h>
8 #include <linux/sched.h>
9 #include <linux/security.h>
10 #include <linux/swap.h>
11 #include <linux/swapops.h>
12 #include <linux/mman.h>
13 #include <linux/hugetlb.h>
14 #include <linux/vmalloc.h>
16 #include <asm/sections.h>
17 #include <asm/uaccess.h>
21 static inline int is_kernel_rodata(unsigned long addr)
23 return addr >= (unsigned long)__start_rodata &&
24 addr < (unsigned long)__end_rodata;
28 * kfree_const - conditionally free memory
29 * @x: pointer to the memory
31 * Function calls kfree only if @x is not in .rodata section.
33 void kfree_const(const void *x)
35 if (!is_kernel_rodata((unsigned long)x))
38 EXPORT_SYMBOL(kfree_const);
41 * kstrdup - allocate space for and copy an existing string
42 * @s: the string to duplicate
43 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
45 char *kstrdup(const char *s, gfp_t gfp)
54 buf = kmalloc_track_caller(len, gfp);
59 EXPORT_SYMBOL(kstrdup);
62 * kstrdup_const - conditionally duplicate an existing const string
63 * @s: the string to duplicate
64 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
66 * Function returns source string if it is in .rodata section otherwise it
67 * fallbacks to kstrdup.
68 * Strings allocated by kstrdup_const should be freed by kfree_const.
70 const char *kstrdup_const(const char *s, gfp_t gfp)
72 if (is_kernel_rodata((unsigned long)s))
75 return kstrdup(s, gfp);
77 EXPORT_SYMBOL(kstrdup_const);
80 * kstrndup - allocate space for and copy an existing string
81 * @s: the string to duplicate
82 * @max: read at most @max chars from @s
83 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
85 char *kstrndup(const char *s, size_t max, gfp_t gfp)
93 len = strnlen(s, max);
94 buf = kmalloc_track_caller(len+1, gfp);
101 EXPORT_SYMBOL(kstrndup);
104 * kstrimdup - Trim and copy a %NUL terminated string.
105 * @s: the string to trim and duplicate
106 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
108 * Returns an address, which the caller must kfree, containing
109 * a duplicate of the passed string with leading and/or trailing
110 * whitespace (as defined by isspace) removed.
112 char *kstrimdup(const char *s, gfp_t gfp)
115 char *begin = skip_spaces(s);
116 size_t len = strlen(begin);
118 while (len && isspace(begin[len - 1]))
121 buf = kmalloc_track_caller(len + 1, gfp);
125 memcpy(buf, begin, len);
130 EXPORT_SYMBOL(kstrimdup);
133 * kmemdup - duplicate region of memory
135 * @src: memory region to duplicate
136 * @len: memory region length
137 * @gfp: GFP mask to use
139 void *kmemdup(const void *src, size_t len, gfp_t gfp)
143 p = kmalloc_track_caller(len, gfp);
148 EXPORT_SYMBOL(kmemdup);
151 * memdup_user - duplicate memory region from user space
153 * @src: source address in user space
154 * @len: number of bytes to copy
156 * Returns an ERR_PTR() on failure.
158 void *memdup_user(const void __user *src, size_t len)
163 * Always use GFP_KERNEL, since copy_from_user() can sleep and
164 * cause pagefault, which makes it pointless to use GFP_NOFS
167 p = kmalloc_track_caller(len, GFP_KERNEL);
169 return ERR_PTR(-ENOMEM);
171 if (copy_from_user(p, src, len)) {
173 return ERR_PTR(-EFAULT);
178 EXPORT_SYMBOL(memdup_user);
181 * strndup_user - duplicate an existing string from user space
182 * @s: The string to duplicate
183 * @n: Maximum number of bytes to copy, including the trailing NUL.
185 char *strndup_user(const char __user *s, long n)
190 length = strnlen_user(s, n);
193 return ERR_PTR(-EFAULT);
196 return ERR_PTR(-EINVAL);
198 p = memdup_user(s, length);
203 p[length - 1] = '\0';
207 EXPORT_SYMBOL(strndup_user);
210 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
212 * @src: source address in user space
213 * @len: number of bytes to copy
215 * Returns an ERR_PTR() on failure.
217 void *memdup_user_nul(const void __user *src, size_t len)
222 * Always use GFP_KERNEL, since copy_from_user() can sleep and
223 * cause pagefault, which makes it pointless to use GFP_NOFS
226 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
228 return ERR_PTR(-ENOMEM);
230 if (copy_from_user(p, src, len)) {
232 return ERR_PTR(-EFAULT);
238 EXPORT_SYMBOL(memdup_user_nul);
240 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
241 struct vm_area_struct *prev, struct rb_node *rb_parent)
243 struct vm_area_struct *next;
247 next = prev->vm_next;
252 next = rb_entry(rb_parent,
253 struct vm_area_struct, vm_rb);
262 /* Check if the vma is being used as a stack by this task */
263 int vma_is_stack_for_task(struct vm_area_struct *vma, struct task_struct *t)
265 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
268 #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
269 void arch_pick_mmap_layout(struct mm_struct *mm)
271 mm->mmap_base = TASK_UNMAPPED_BASE;
272 mm->get_unmapped_area = arch_get_unmapped_area;
277 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
278 * back to the regular GUP.
279 * If the architecture not support this function, simply return with no
282 int __weak __get_user_pages_fast(unsigned long start,
283 int nr_pages, int write, struct page **pages)
287 EXPORT_SYMBOL_GPL(__get_user_pages_fast);
290 * get_user_pages_fast() - pin user pages in memory
291 * @start: starting user address
292 * @nr_pages: number of pages from start to pin
293 * @write: whether pages will be written to
294 * @pages: array that receives pointers to the pages pinned.
295 * Should be at least nr_pages long.
297 * Returns number of pages pinned. This may be fewer than the number
298 * requested. If nr_pages is 0 or negative, returns 0. If no pages
299 * were pinned, returns -errno.
301 * get_user_pages_fast provides equivalent functionality to get_user_pages,
302 * operating on current and current->mm, with force=0 and vma=NULL. However
303 * unlike get_user_pages, it must be called without mmap_sem held.
305 * get_user_pages_fast may take mmap_sem and page table locks, so no
306 * assumptions can be made about lack of locking. get_user_pages_fast is to be
307 * implemented in a way that is advantageous (vs get_user_pages()) when the
308 * user memory area is already faulted in and present in ptes. However if the
309 * pages have to be faulted in, it may turn out to be slightly slower so
310 * callers need to carefully consider what to use. On many architectures,
311 * get_user_pages_fast simply falls back to get_user_pages.
313 int __weak get_user_pages_fast(unsigned long start,
314 int nr_pages, int write, struct page **pages)
316 return get_user_pages_unlocked(start, nr_pages, write, 0, pages);
318 EXPORT_SYMBOL_GPL(get_user_pages_fast);
320 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
321 unsigned long len, unsigned long prot,
322 unsigned long flag, unsigned long pgoff)
325 struct mm_struct *mm = current->mm;
326 unsigned long populate;
328 ret = security_mmap_file(file, prot, flag);
330 down_write(&mm->mmap_sem);
331 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
333 up_write(&mm->mmap_sem);
335 mm_populate(ret, populate);
340 unsigned long vm_mmap(struct file *file, unsigned long addr,
341 unsigned long len, unsigned long prot,
342 unsigned long flag, unsigned long offset)
344 if (unlikely(offset + PAGE_ALIGN(len) < offset))
346 if (unlikely(offset_in_page(offset)))
349 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
351 EXPORT_SYMBOL(vm_mmap);
353 void kvfree(const void *addr)
355 if (is_vmalloc_addr(addr))
360 EXPORT_SYMBOL(kvfree);
362 static inline void *__page_rmapping(struct page *page)
364 unsigned long mapping;
366 mapping = (unsigned long)page->mapping;
367 mapping &= ~PAGE_MAPPING_FLAGS;
369 return (void *)mapping;
372 /* Neutral page->mapping pointer to address_space or anon_vma or other */
373 void *page_rmapping(struct page *page)
375 page = compound_head(page);
376 return __page_rmapping(page);
379 struct anon_vma *page_anon_vma(struct page *page)
381 unsigned long mapping;
383 page = compound_head(page);
384 mapping = (unsigned long)page->mapping;
385 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
387 return __page_rmapping(page);
390 struct address_space *page_mapping(struct page *page)
392 struct address_space *mapping;
394 page = compound_head(page);
396 /* This happens if someone calls flush_dcache_page on slab page */
397 if (unlikely(PageSlab(page)))
400 if (unlikely(PageSwapCache(page))) {
403 entry.val = page_private(page);
404 return swap_address_space(entry);
407 mapping = page->mapping;
408 if ((unsigned long)mapping & PAGE_MAPPING_FLAGS)
413 /* Slow path of page_mapcount() for compound pages */
414 int __page_mapcount(struct page *page)
418 ret = atomic_read(&page->_mapcount) + 1;
419 page = compound_head(page);
420 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
421 if (PageDoubleMap(page))
425 EXPORT_SYMBOL_GPL(__page_mapcount);
427 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
428 int sysctl_overcommit_ratio __read_mostly = 50;
429 unsigned long sysctl_overcommit_kbytes __read_mostly;
430 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
431 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
432 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
434 int overcommit_ratio_handler(struct ctl_table *table, int write,
435 void __user *buffer, size_t *lenp,
440 ret = proc_dointvec(table, write, buffer, lenp, ppos);
441 if (ret == 0 && write)
442 sysctl_overcommit_kbytes = 0;
446 int overcommit_kbytes_handler(struct ctl_table *table, int write,
447 void __user *buffer, size_t *lenp,
452 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
453 if (ret == 0 && write)
454 sysctl_overcommit_ratio = 0;
459 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
461 unsigned long vm_commit_limit(void)
463 unsigned long allowed;
465 if (sysctl_overcommit_kbytes)
466 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
468 allowed = ((totalram_pages - hugetlb_total_pages())
469 * sysctl_overcommit_ratio / 100);
470 allowed += total_swap_pages;
476 * Make sure vm_committed_as in one cacheline and not cacheline shared with
477 * other variables. It can be updated by several CPUs frequently.
479 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
482 * The global memory commitment made in the system can be a metric
483 * that can be used to drive ballooning decisions when Linux is hosted
484 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
485 * balancing memory across competing virtual machines that are hosted.
486 * Several metrics drive this policy engine including the guest reported
489 unsigned long vm_memory_committed(void)
491 return percpu_counter_read_positive(&vm_committed_as);
493 EXPORT_SYMBOL_GPL(vm_memory_committed);
496 * Check that a process has enough memory to allocate a new virtual
497 * mapping. 0 means there is enough memory for the allocation to
498 * succeed and -ENOMEM implies there is not.
500 * We currently support three overcommit policies, which are set via the
501 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
503 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
504 * Additional code 2002 Jul 20 by Robert Love.
506 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
508 * Note this is a helper function intended to be used by LSMs which
509 * wish to use this logic.
511 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
513 long free, allowed, reserve;
515 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
516 -(s64)vm_committed_as_batch * num_online_cpus(),
517 "memory commitment underflow");
519 vm_acct_memory(pages);
522 * Sometimes we want to use more memory than we have
524 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
527 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
528 free = global_page_state(NR_FREE_PAGES);
529 free += global_page_state(NR_FILE_PAGES);
532 * shmem pages shouldn't be counted as free in this
533 * case, they can't be purged, only swapped out, and
534 * that won't affect the overall amount of available
535 * memory in the system.
537 free -= global_page_state(NR_SHMEM);
539 free += get_nr_swap_pages();
542 * Any slabs which are created with the
543 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
544 * which are reclaimable, under pressure. The dentry
545 * cache and most inode caches should fall into this
547 free += global_page_state(NR_SLAB_RECLAIMABLE);
550 * Leave reserved pages. The pages are not for anonymous pages.
552 if (free <= totalreserve_pages)
555 free -= totalreserve_pages;
558 * Reserve some for root
561 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
569 allowed = vm_commit_limit();
571 * Reserve some for root
574 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
577 * Don't let a single process grow so big a user can't recover
580 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
581 allowed -= min_t(long, mm->total_vm / 32, reserve);
584 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
587 vm_unacct_memory(pages);
593 * get_cmdline() - copy the cmdline value to a buffer.
594 * @task: the task whose cmdline value to copy.
595 * @buffer: the buffer to copy to.
596 * @buflen: the length of the buffer. Larger cmdline values are truncated
598 * Returns the size of the cmdline field copied. Note that the copy does
599 * not guarantee an ending NULL byte.
601 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
605 struct mm_struct *mm = get_task_mm(task);
606 unsigned long arg_start, arg_end, env_start, env_end;
610 goto out_mm; /* Shh! No looking before we're done */
612 down_read(&mm->mmap_sem);
613 arg_start = mm->arg_start;
614 arg_end = mm->arg_end;
615 env_start = mm->env_start;
616 env_end = mm->env_end;
617 up_read(&mm->mmap_sem);
619 len = arg_end - arg_start;
624 res = access_process_vm(task, arg_start, buffer, len, 0);
627 * If the nul at the end of args has been overwritten, then
628 * assume application is using setproctitle(3).
630 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
631 len = strnlen(buffer, res);
635 len = env_end - env_start;
636 if (len > buflen - res)
638 res += access_process_vm(task, env_start,
640 res = strnlen(buffer, res);