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/sched/mm.h>
9 #include <linux/sched/task_stack.h>
10 #include <linux/security.h>
11 #include <linux/swap.h>
12 #include <linux/swapops.h>
13 #include <linux/mman.h>
14 #include <linux/hugetlb.h>
15 #include <linux/vmalloc.h>
16 #include <linux/userfaultfd_k.h>
18 #include <asm/sections.h>
19 #include <linux/uaccess.h>
23 static inline int is_kernel_rodata(unsigned long addr)
25 return addr >= (unsigned long)__start_rodata &&
26 addr < (unsigned long)__end_rodata;
30 * kfree_const - conditionally free memory
31 * @x: pointer to the memory
33 * Function calls kfree only if @x is not in .rodata section.
35 void kfree_const(const void *x)
37 if (!is_kernel_rodata((unsigned long)x))
40 EXPORT_SYMBOL(kfree_const);
43 * kstrdup - allocate space for and copy an existing string
44 * @s: the string to duplicate
45 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
47 char *kstrdup(const char *s, gfp_t gfp)
56 buf = kmalloc_track_caller(len, gfp);
61 EXPORT_SYMBOL(kstrdup);
64 * kstrdup_const - conditionally duplicate an existing const string
65 * @s: the string to duplicate
66 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
68 * Function returns source string if it is in .rodata section otherwise it
69 * fallbacks to kstrdup.
70 * Strings allocated by kstrdup_const should be freed by kfree_const.
72 const char *kstrdup_const(const char *s, gfp_t gfp)
74 if (is_kernel_rodata((unsigned long)s))
77 return kstrdup(s, gfp);
79 EXPORT_SYMBOL(kstrdup_const);
82 * kstrndup - allocate space for and copy an existing string
83 * @s: the string to duplicate
84 * @max: read at most @max chars from @s
85 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
87 char *kstrndup(const char *s, size_t max, gfp_t gfp)
95 len = strnlen(s, max);
96 buf = kmalloc_track_caller(len+1, gfp);
103 EXPORT_SYMBOL(kstrndup);
106 * kmemdup - duplicate region of memory
108 * @src: memory region to duplicate
109 * @len: memory region length
110 * @gfp: GFP mask to use
112 void *kmemdup(const void *src, size_t len, gfp_t gfp)
116 p = kmalloc_track_caller(len, gfp);
121 EXPORT_SYMBOL(kmemdup);
124 * memdup_user - duplicate memory region from user space
126 * @src: source address in user space
127 * @len: number of bytes to copy
129 * Returns an ERR_PTR() on failure.
131 void *memdup_user(const void __user *src, size_t len)
136 * Always use GFP_KERNEL, since copy_from_user() can sleep and
137 * cause pagefault, which makes it pointless to use GFP_NOFS
140 p = kmalloc_track_caller(len, GFP_KERNEL);
142 return ERR_PTR(-ENOMEM);
144 if (copy_from_user(p, src, len)) {
146 return ERR_PTR(-EFAULT);
151 EXPORT_SYMBOL(memdup_user);
154 * strndup_user - duplicate an existing string from user space
155 * @s: The string to duplicate
156 * @n: Maximum number of bytes to copy, including the trailing NUL.
158 char *strndup_user(const char __user *s, long n)
163 length = strnlen_user(s, n);
166 return ERR_PTR(-EFAULT);
169 return ERR_PTR(-EINVAL);
171 p = memdup_user(s, length);
176 p[length - 1] = '\0';
180 EXPORT_SYMBOL(strndup_user);
183 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
185 * @src: source address in user space
186 * @len: number of bytes to copy
188 * Returns an ERR_PTR() on failure.
190 void *memdup_user_nul(const void __user *src, size_t len)
195 * Always use GFP_KERNEL, since copy_from_user() can sleep and
196 * cause pagefault, which makes it pointless to use GFP_NOFS
199 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
201 return ERR_PTR(-ENOMEM);
203 if (copy_from_user(p, src, len)) {
205 return ERR_PTR(-EFAULT);
211 EXPORT_SYMBOL(memdup_user_nul);
213 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
214 struct vm_area_struct *prev, struct rb_node *rb_parent)
216 struct vm_area_struct *next;
220 next = prev->vm_next;
225 next = rb_entry(rb_parent,
226 struct vm_area_struct, vm_rb);
235 /* Check if the vma is being used as a stack by this task */
236 int vma_is_stack_for_current(struct vm_area_struct *vma)
238 struct task_struct * __maybe_unused t = current;
240 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
243 #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
244 void arch_pick_mmap_layout(struct mm_struct *mm)
246 mm->mmap_base = TASK_UNMAPPED_BASE;
247 mm->get_unmapped_area = arch_get_unmapped_area;
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
257 int __weak __get_user_pages_fast(unsigned long start,
258 int nr_pages, int write, struct page **pages)
262 EXPORT_SYMBOL_GPL(__get_user_pages_fast);
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.
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.
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.
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.
288 int __weak get_user_pages_fast(unsigned long start,
289 int nr_pages, int write, struct page **pages)
291 return get_user_pages_unlocked(start, nr_pages, pages,
292 write ? FOLL_WRITE : 0);
294 EXPORT_SYMBOL_GPL(get_user_pages_fast);
296 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
297 unsigned long len, unsigned long prot,
298 unsigned long flag, unsigned long pgoff)
301 struct mm_struct *mm = current->mm;
302 unsigned long populate;
305 ret = security_mmap_file(file, prot, flag);
307 if (down_write_killable(&mm->mmap_sem))
309 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
311 up_write(&mm->mmap_sem);
312 userfaultfd_unmap_complete(mm, &uf);
314 mm_populate(ret, populate);
319 unsigned long vm_mmap(struct file *file, unsigned long addr,
320 unsigned long len, unsigned long prot,
321 unsigned long flag, unsigned long offset)
323 if (unlikely(offset + PAGE_ALIGN(len) < offset))
325 if (unlikely(offset_in_page(offset)))
328 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
330 EXPORT_SYMBOL(vm_mmap);
333 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
334 * failure, fall back to non-contiguous (vmalloc) allocation.
335 * @size: size of the request.
336 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
337 * @node: numa node to allocate from
339 * Uses kmalloc to get the memory but if the allocation fails then falls back
340 * to the vmalloc allocator. Use kvfree for freeing the memory.
342 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported. __GFP_REPEAT
343 * is supported only for large (>32kB) allocations, and it should be used only if
344 * kmalloc is preferable to the vmalloc fallback, due to visible performance drawbacks.
346 * Any use of gfp flags outside of GFP_KERNEL should be consulted with mm people.
348 void *kvmalloc_node(size_t size, gfp_t flags, int node)
350 gfp_t kmalloc_flags = flags;
354 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
355 * so the given set of flags has to be compatible.
357 WARN_ON_ONCE((flags & GFP_KERNEL) != GFP_KERNEL);
360 * Make sure that larger requests are not too disruptive - no OOM
361 * killer and no allocation failure warnings as we have a fallback
363 if (size > PAGE_SIZE) {
364 kmalloc_flags |= __GFP_NOWARN;
367 * We have to override __GFP_REPEAT by __GFP_NORETRY for !costly
368 * requests because there is no other way to tell the allocator
369 * that we want to fail rather than retry endlessly.
371 if (!(kmalloc_flags & __GFP_REPEAT) ||
372 (size <= PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
373 kmalloc_flags |= __GFP_NORETRY;
376 ret = kmalloc_node(size, kmalloc_flags, node);
379 * It doesn't really make sense to fallback to vmalloc for sub page
382 if (ret || size <= PAGE_SIZE)
385 return __vmalloc_node_flags(size, node, flags);
387 EXPORT_SYMBOL(kvmalloc_node);
389 void kvfree(const void *addr)
391 if (is_vmalloc_addr(addr))
396 EXPORT_SYMBOL(kvfree);
398 static inline void *__page_rmapping(struct page *page)
400 unsigned long mapping;
402 mapping = (unsigned long)page->mapping;
403 mapping &= ~PAGE_MAPPING_FLAGS;
405 return (void *)mapping;
408 /* Neutral page->mapping pointer to address_space or anon_vma or other */
409 void *page_rmapping(struct page *page)
411 page = compound_head(page);
412 return __page_rmapping(page);
416 * Return true if this page is mapped into pagetables.
417 * For compound page it returns true if any subpage of compound page is mapped.
419 bool page_mapped(struct page *page)
423 if (likely(!PageCompound(page)))
424 return atomic_read(&page->_mapcount) >= 0;
425 page = compound_head(page);
426 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
430 for (i = 0; i < hpage_nr_pages(page); i++) {
431 if (atomic_read(&page[i]._mapcount) >= 0)
436 EXPORT_SYMBOL(page_mapped);
438 struct anon_vma *page_anon_vma(struct page *page)
440 unsigned long mapping;
442 page = compound_head(page);
443 mapping = (unsigned long)page->mapping;
444 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
446 return __page_rmapping(page);
449 struct address_space *page_mapping(struct page *page)
451 struct address_space *mapping;
453 page = compound_head(page);
455 /* This happens if someone calls flush_dcache_page on slab page */
456 if (unlikely(PageSlab(page)))
459 if (unlikely(PageSwapCache(page))) {
462 entry.val = page_private(page);
463 return swap_address_space(entry);
466 mapping = page->mapping;
467 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
470 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
472 EXPORT_SYMBOL(page_mapping);
474 /* Slow path of page_mapcount() for compound pages */
475 int __page_mapcount(struct page *page)
479 ret = atomic_read(&page->_mapcount) + 1;
481 * For file THP page->_mapcount contains total number of mapping
482 * of the page: no need to look into compound_mapcount.
484 if (!PageAnon(page) && !PageHuge(page))
486 page = compound_head(page);
487 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
488 if (PageDoubleMap(page))
492 EXPORT_SYMBOL_GPL(__page_mapcount);
494 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
495 int sysctl_overcommit_ratio __read_mostly = 50;
496 unsigned long sysctl_overcommit_kbytes __read_mostly;
497 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
498 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
499 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
501 int overcommit_ratio_handler(struct ctl_table *table, int write,
502 void __user *buffer, size_t *lenp,
507 ret = proc_dointvec(table, write, buffer, lenp, ppos);
508 if (ret == 0 && write)
509 sysctl_overcommit_kbytes = 0;
513 int overcommit_kbytes_handler(struct ctl_table *table, int write,
514 void __user *buffer, size_t *lenp,
519 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
520 if (ret == 0 && write)
521 sysctl_overcommit_ratio = 0;
526 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
528 unsigned long vm_commit_limit(void)
530 unsigned long allowed;
532 if (sysctl_overcommit_kbytes)
533 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
535 allowed = ((totalram_pages - hugetlb_total_pages())
536 * sysctl_overcommit_ratio / 100);
537 allowed += total_swap_pages;
543 * Make sure vm_committed_as in one cacheline and not cacheline shared with
544 * other variables. It can be updated by several CPUs frequently.
546 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
549 * The global memory commitment made in the system can be a metric
550 * that can be used to drive ballooning decisions when Linux is hosted
551 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
552 * balancing memory across competing virtual machines that are hosted.
553 * Several metrics drive this policy engine including the guest reported
556 unsigned long vm_memory_committed(void)
558 return percpu_counter_read_positive(&vm_committed_as);
560 EXPORT_SYMBOL_GPL(vm_memory_committed);
563 * Check that a process has enough memory to allocate a new virtual
564 * mapping. 0 means there is enough memory for the allocation to
565 * succeed and -ENOMEM implies there is not.
567 * We currently support three overcommit policies, which are set via the
568 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
570 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
571 * Additional code 2002 Jul 20 by Robert Love.
573 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
575 * Note this is a helper function intended to be used by LSMs which
576 * wish to use this logic.
578 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
580 long free, allowed, reserve;
582 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
583 -(s64)vm_committed_as_batch * num_online_cpus(),
584 "memory commitment underflow");
586 vm_acct_memory(pages);
589 * Sometimes we want to use more memory than we have
591 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
594 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
595 free = global_page_state(NR_FREE_PAGES);
596 free += global_node_page_state(NR_FILE_PAGES);
599 * shmem pages shouldn't be counted as free in this
600 * case, they can't be purged, only swapped out, and
601 * that won't affect the overall amount of available
602 * memory in the system.
604 free -= global_node_page_state(NR_SHMEM);
606 free += get_nr_swap_pages();
609 * Any slabs which are created with the
610 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
611 * which are reclaimable, under pressure. The dentry
612 * cache and most inode caches should fall into this
614 free += global_page_state(NR_SLAB_RECLAIMABLE);
617 * Leave reserved pages. The pages are not for anonymous pages.
619 if (free <= totalreserve_pages)
622 free -= totalreserve_pages;
625 * Reserve some for root
628 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
636 allowed = vm_commit_limit();
638 * Reserve some for root
641 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
644 * Don't let a single process grow so big a user can't recover
647 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
648 allowed -= min_t(long, mm->total_vm / 32, reserve);
651 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
654 vm_unacct_memory(pages);
660 * get_cmdline() - copy the cmdline value to a buffer.
661 * @task: the task whose cmdline value to copy.
662 * @buffer: the buffer to copy to.
663 * @buflen: the length of the buffer. Larger cmdline values are truncated
665 * Returns the size of the cmdline field copied. Note that the copy does
666 * not guarantee an ending NULL byte.
668 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
672 struct mm_struct *mm = get_task_mm(task);
673 unsigned long arg_start, arg_end, env_start, env_end;
677 goto out_mm; /* Shh! No looking before we're done */
679 down_read(&mm->mmap_sem);
680 arg_start = mm->arg_start;
681 arg_end = mm->arg_end;
682 env_start = mm->env_start;
683 env_end = mm->env_end;
684 up_read(&mm->mmap_sem);
686 len = arg_end - arg_start;
691 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
694 * If the nul at the end of args has been overwritten, then
695 * assume application is using setproctitle(3).
697 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
698 len = strnlen(buffer, res);
702 len = env_end - env_start;
703 if (len > buflen - res)
705 res += access_process_vm(task, env_start,
708 res = strnlen(buffer, res);