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>
14 #include <linux/userfaultfd_k.h>
16 #include <asm/sections.h>
17 #include <linux/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 * kmemdup - duplicate region of memory
106 * @src: memory region to duplicate
107 * @len: memory region length
108 * @gfp: GFP mask to use
110 void *kmemdup(const void *src, size_t len, gfp_t gfp)
114 p = kmalloc_track_caller(len, gfp);
119 EXPORT_SYMBOL(kmemdup);
122 * memdup_user - duplicate memory region from user space
124 * @src: source address in user space
125 * @len: number of bytes to copy
127 * Returns an ERR_PTR() on failure.
129 void *memdup_user(const void __user *src, size_t len)
134 * Always use GFP_KERNEL, since copy_from_user() can sleep and
135 * cause pagefault, which makes it pointless to use GFP_NOFS
138 p = kmalloc_track_caller(len, GFP_KERNEL);
140 return ERR_PTR(-ENOMEM);
142 if (copy_from_user(p, src, len)) {
144 return ERR_PTR(-EFAULT);
149 EXPORT_SYMBOL(memdup_user);
152 * strndup_user - duplicate an existing string from user space
153 * @s: The string to duplicate
154 * @n: Maximum number of bytes to copy, including the trailing NUL.
156 char *strndup_user(const char __user *s, long n)
161 length = strnlen_user(s, n);
164 return ERR_PTR(-EFAULT);
167 return ERR_PTR(-EINVAL);
169 p = memdup_user(s, length);
174 p[length - 1] = '\0';
178 EXPORT_SYMBOL(strndup_user);
181 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
183 * @src: source address in user space
184 * @len: number of bytes to copy
186 * Returns an ERR_PTR() on failure.
188 void *memdup_user_nul(const void __user *src, size_t len)
193 * Always use GFP_KERNEL, since copy_from_user() can sleep and
194 * cause pagefault, which makes it pointless to use GFP_NOFS
197 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
199 return ERR_PTR(-ENOMEM);
201 if (copy_from_user(p, src, len)) {
203 return ERR_PTR(-EFAULT);
209 EXPORT_SYMBOL(memdup_user_nul);
211 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
212 struct vm_area_struct *prev, struct rb_node *rb_parent)
214 struct vm_area_struct *next;
218 next = prev->vm_next;
223 next = rb_entry(rb_parent,
224 struct vm_area_struct, vm_rb);
233 /* Check if the vma is being used as a stack by this task */
234 int vma_is_stack_for_current(struct vm_area_struct *vma)
236 struct task_struct * __maybe_unused t = current;
238 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
241 #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
242 void arch_pick_mmap_layout(struct mm_struct *mm)
244 mm->mmap_base = TASK_UNMAPPED_BASE;
245 mm->get_unmapped_area = arch_get_unmapped_area;
250 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
251 * back to the regular GUP.
252 * If the architecture not support this function, simply return with no
255 int __weak __get_user_pages_fast(unsigned long start,
256 int nr_pages, int write, struct page **pages)
260 EXPORT_SYMBOL_GPL(__get_user_pages_fast);
263 * get_user_pages_fast() - pin user pages in memory
264 * @start: starting user address
265 * @nr_pages: number of pages from start to pin
266 * @write: whether pages will be written to
267 * @pages: array that receives pointers to the pages pinned.
268 * Should be at least nr_pages long.
270 * Returns number of pages pinned. This may be fewer than the number
271 * requested. If nr_pages is 0 or negative, returns 0. If no pages
272 * were pinned, returns -errno.
274 * get_user_pages_fast provides equivalent functionality to get_user_pages,
275 * operating on current and current->mm, with force=0 and vma=NULL. However
276 * unlike get_user_pages, it must be called without mmap_sem held.
278 * get_user_pages_fast may take mmap_sem and page table locks, so no
279 * assumptions can be made about lack of locking. get_user_pages_fast is to be
280 * implemented in a way that is advantageous (vs get_user_pages()) when the
281 * user memory area is already faulted in and present in ptes. However if the
282 * pages have to be faulted in, it may turn out to be slightly slower so
283 * callers need to carefully consider what to use. On many architectures,
284 * get_user_pages_fast simply falls back to get_user_pages.
286 int __weak get_user_pages_fast(unsigned long start,
287 int nr_pages, int write, struct page **pages)
289 return get_user_pages_unlocked(start, nr_pages, pages,
290 write ? FOLL_WRITE : 0);
292 EXPORT_SYMBOL_GPL(get_user_pages_fast);
294 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
295 unsigned long len, unsigned long prot,
296 unsigned long flag, unsigned long pgoff)
299 struct mm_struct *mm = current->mm;
300 unsigned long populate;
303 ret = security_mmap_file(file, prot, flag);
305 if (down_write_killable(&mm->mmap_sem))
307 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
309 up_write(&mm->mmap_sem);
310 userfaultfd_unmap_complete(mm, &uf);
312 mm_populate(ret, populate);
317 unsigned long vm_mmap(struct file *file, unsigned long addr,
318 unsigned long len, unsigned long prot,
319 unsigned long flag, unsigned long offset)
321 if (unlikely(offset + PAGE_ALIGN(len) < offset))
323 if (unlikely(offset_in_page(offset)))
326 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
328 EXPORT_SYMBOL(vm_mmap);
330 void kvfree(const void *addr)
332 if (is_vmalloc_addr(addr))
337 EXPORT_SYMBOL(kvfree);
339 static inline void *__page_rmapping(struct page *page)
341 unsigned long mapping;
343 mapping = (unsigned long)page->mapping;
344 mapping &= ~PAGE_MAPPING_FLAGS;
346 return (void *)mapping;
349 /* Neutral page->mapping pointer to address_space or anon_vma or other */
350 void *page_rmapping(struct page *page)
352 page = compound_head(page);
353 return __page_rmapping(page);
357 * Return true if this page is mapped into pagetables.
358 * For compound page it returns true if any subpage of compound page is mapped.
360 bool page_mapped(struct page *page)
364 if (likely(!PageCompound(page)))
365 return atomic_read(&page->_mapcount) >= 0;
366 page = compound_head(page);
367 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
371 for (i = 0; i < hpage_nr_pages(page); i++) {
372 if (atomic_read(&page[i]._mapcount) >= 0)
377 EXPORT_SYMBOL(page_mapped);
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_ANON)
411 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
413 EXPORT_SYMBOL(page_mapping);
415 /* Slow path of page_mapcount() for compound pages */
416 int __page_mapcount(struct page *page)
420 ret = atomic_read(&page->_mapcount) + 1;
422 * For file THP page->_mapcount contains total number of mapping
423 * of the page: no need to look into compound_mapcount.
425 if (!PageAnon(page) && !PageHuge(page))
427 page = compound_head(page);
428 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
429 if (PageDoubleMap(page))
433 EXPORT_SYMBOL_GPL(__page_mapcount);
435 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
436 int sysctl_overcommit_ratio __read_mostly = 50;
437 unsigned long sysctl_overcommit_kbytes __read_mostly;
438 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
439 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
440 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
442 int overcommit_ratio_handler(struct ctl_table *table, int write,
443 void __user *buffer, size_t *lenp,
448 ret = proc_dointvec(table, write, buffer, lenp, ppos);
449 if (ret == 0 && write)
450 sysctl_overcommit_kbytes = 0;
454 int overcommit_kbytes_handler(struct ctl_table *table, int write,
455 void __user *buffer, size_t *lenp,
460 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
461 if (ret == 0 && write)
462 sysctl_overcommit_ratio = 0;
467 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
469 unsigned long vm_commit_limit(void)
471 unsigned long allowed;
473 if (sysctl_overcommit_kbytes)
474 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
476 allowed = ((totalram_pages - hugetlb_total_pages())
477 * sysctl_overcommit_ratio / 100);
478 allowed += total_swap_pages;
484 * Make sure vm_committed_as in one cacheline and not cacheline shared with
485 * other variables. It can be updated by several CPUs frequently.
487 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
490 * The global memory commitment made in the system can be a metric
491 * that can be used to drive ballooning decisions when Linux is hosted
492 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
493 * balancing memory across competing virtual machines that are hosted.
494 * Several metrics drive this policy engine including the guest reported
497 unsigned long vm_memory_committed(void)
499 return percpu_counter_read_positive(&vm_committed_as);
501 EXPORT_SYMBOL_GPL(vm_memory_committed);
504 * Check that a process has enough memory to allocate a new virtual
505 * mapping. 0 means there is enough memory for the allocation to
506 * succeed and -ENOMEM implies there is not.
508 * We currently support three overcommit policies, which are set via the
509 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
511 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
512 * Additional code 2002 Jul 20 by Robert Love.
514 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
516 * Note this is a helper function intended to be used by LSMs which
517 * wish to use this logic.
519 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
521 long free, allowed, reserve;
523 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
524 -(s64)vm_committed_as_batch * num_online_cpus(),
525 "memory commitment underflow");
527 vm_acct_memory(pages);
530 * Sometimes we want to use more memory than we have
532 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
535 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
536 free = global_page_state(NR_FREE_PAGES);
537 free += global_node_page_state(NR_FILE_PAGES);
540 * shmem pages shouldn't be counted as free in this
541 * case, they can't be purged, only swapped out, and
542 * that won't affect the overall amount of available
543 * memory in the system.
545 free -= global_node_page_state(NR_SHMEM);
547 free += get_nr_swap_pages();
550 * Any slabs which are created with the
551 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
552 * which are reclaimable, under pressure. The dentry
553 * cache and most inode caches should fall into this
555 free += global_page_state(NR_SLAB_RECLAIMABLE);
558 * Leave reserved pages. The pages are not for anonymous pages.
560 if (free <= totalreserve_pages)
563 free -= totalreserve_pages;
566 * Reserve some for root
569 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
577 allowed = vm_commit_limit();
579 * Reserve some for root
582 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
585 * Don't let a single process grow so big a user can't recover
588 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
589 allowed -= min_t(long, mm->total_vm / 32, reserve);
592 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
595 vm_unacct_memory(pages);
601 * get_cmdline() - copy the cmdline value to a buffer.
602 * @task: the task whose cmdline value to copy.
603 * @buffer: the buffer to copy to.
604 * @buflen: the length of the buffer. Larger cmdline values are truncated
606 * Returns the size of the cmdline field copied. Note that the copy does
607 * not guarantee an ending NULL byte.
609 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
613 struct mm_struct *mm = get_task_mm(task);
614 unsigned long arg_start, arg_end, env_start, env_end;
618 goto out_mm; /* Shh! No looking before we're done */
620 down_read(&mm->mmap_sem);
621 arg_start = mm->arg_start;
622 arg_end = mm->arg_end;
623 env_start = mm->env_start;
624 env_end = mm->env_end;
625 up_read(&mm->mmap_sem);
627 len = arg_end - arg_start;
632 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
635 * If the nul at the end of args has been overwritten, then
636 * assume application is using setproctitle(3).
638 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
639 len = strnlen(buffer, res);
643 len = env_end - env_start;
644 if (len > buflen - res)
646 res += access_process_vm(task, env_start,
649 res = strnlen(buffer, res);