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/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>
17
18#include <asm/sections.h>
19#include <linux/uaccess.h>
20
21#include "internal.h"
22
23static inline int is_kernel_rodata(unsigned long addr)
24{
25 return addr >= (unsigned long)__start_rodata &&
26 addr < (unsigned long)__end_rodata;
27}
28
29/**
30 * kfree_const - conditionally free memory
31 * @x: pointer to the memory
32 *
33 * Function calls kfree only if @x is not in .rodata section.
34 */
35void kfree_const(const void *x)
36{
37 if (!is_kernel_rodata((unsigned long)x))
38 kfree(x);
39}
40EXPORT_SYMBOL(kfree_const);
41
42/**
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
46 */
47char *kstrdup(const char *s, gfp_t gfp)
48{
49 size_t len;
50 char *buf;
51
52 if (!s)
53 return NULL;
54
55 len = strlen(s) + 1;
56 buf = kmalloc_track_caller(len, gfp);
57 if (buf)
58 memcpy(buf, s, len);
59 return buf;
60}
61EXPORT_SYMBOL(kstrdup);
62
63/**
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
67 *
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.
71 */
72const char *kstrdup_const(const char *s, gfp_t gfp)
73{
74 if (is_kernel_rodata((unsigned long)s))
75 return s;
76
77 return kstrdup(s, gfp);
78}
79EXPORT_SYMBOL(kstrdup_const);
80
81/**
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
86 *
87 * Note: Use kmemdup_nul() instead if the size is known exactly.
88 */
89char *kstrndup(const char *s, size_t max, gfp_t gfp)
90{
91 size_t len;
92 char *buf;
93
94 if (!s)
95 return NULL;
96
97 len = strnlen(s, max);
98 buf = kmalloc_track_caller(len+1, gfp);
99 if (buf) {
100 memcpy(buf, s, len);
101 buf[len] = '\0';
102 }
103 return buf;
104}
105EXPORT_SYMBOL(kstrndup);
106
107/**
108 * kmemdup - duplicate region of memory
109 *
110 * @src: memory region to duplicate
111 * @len: memory region length
112 * @gfp: GFP mask to use
113 */
114void *kmemdup(const void *src, size_t len, gfp_t gfp)
115{
116 void *p;
117
118 p = kmalloc_track_caller(len, gfp);
119 if (p)
120 memcpy(p, src, len);
121 return p;
122}
123EXPORT_SYMBOL(kmemdup);
124
125/**
126 * kmemdup_nul - Create a NUL-terminated string from unterminated data
127 * @s: The data to stringify
128 * @len: The size of the data
129 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
130 */
131char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
132{
133 char *buf;
134
135 if (!s)
136 return NULL;
137
138 buf = kmalloc_track_caller(len + 1, gfp);
139 if (buf) {
140 memcpy(buf, s, len);
141 buf[len] = '\0';
142 }
143 return buf;
144}
145EXPORT_SYMBOL(kmemdup_nul);
146
147/**
148 * memdup_user - duplicate memory region from user space
149 *
150 * @src: source address in user space
151 * @len: number of bytes to copy
152 *
153 * Returns an ERR_PTR() on failure. Result is physically
154 * contiguous, to be freed by kfree().
155 */
156void *memdup_user(const void __user *src, size_t len)
157{
158 void *p;
159
160 p = kmalloc_track_caller(len, GFP_USER);
161 if (!p)
162 return ERR_PTR(-ENOMEM);
163
164 if (copy_from_user(p, src, len)) {
165 kfree(p);
166 return ERR_PTR(-EFAULT);
167 }
168
169 return p;
170}
171EXPORT_SYMBOL(memdup_user);
172
173/**
174 * vmemdup_user - duplicate memory region from user space
175 *
176 * @src: source address in user space
177 * @len: number of bytes to copy
178 *
179 * Returns an ERR_PTR() on failure. Result may be not
180 * physically contiguous. Use kvfree() to free.
181 */
182void *vmemdup_user(const void __user *src, size_t len)
183{
184 void *p;
185
186 p = kvmalloc(len, GFP_USER);
187 if (!p)
188 return ERR_PTR(-ENOMEM);
189
190 if (copy_from_user(p, src, len)) {
191 kvfree(p);
192 return ERR_PTR(-EFAULT);
193 }
194
195 return p;
196}
197EXPORT_SYMBOL(vmemdup_user);
198
199/*
200 * strndup_user - duplicate an existing string from user space
201 * @s: The string to duplicate
202 * @n: Maximum number of bytes to copy, including the trailing NUL.
203 */
204char *strndup_user(const char __user *s, long n)
205{
206 char *p;
207 long length;
208
209 length = strnlen_user(s, n);
210
211 if (!length)
212 return ERR_PTR(-EFAULT);
213
214 if (length > n)
215 return ERR_PTR(-EINVAL);
216
217 p = memdup_user(s, length);
218
219 if (IS_ERR(p))
220 return p;
221
222 p[length - 1] = '\0';
223
224 return p;
225}
226EXPORT_SYMBOL(strndup_user);
227
228/**
229 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
230 *
231 * @src: source address in user space
232 * @len: number of bytes to copy
233 *
234 * Returns an ERR_PTR() on failure.
235 */
236void *memdup_user_nul(const void __user *src, size_t len)
237{
238 char *p;
239
240 /*
241 * Always use GFP_KERNEL, since copy_from_user() can sleep and
242 * cause pagefault, which makes it pointless to use GFP_NOFS
243 * or GFP_ATOMIC.
244 */
245 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
246 if (!p)
247 return ERR_PTR(-ENOMEM);
248
249 if (copy_from_user(p, src, len)) {
250 kfree(p);
251 return ERR_PTR(-EFAULT);
252 }
253 p[len] = '\0';
254
255 return p;
256}
257EXPORT_SYMBOL(memdup_user_nul);
258
259void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
260 struct vm_area_struct *prev, struct rb_node *rb_parent)
261{
262 struct vm_area_struct *next;
263
264 vma->vm_prev = prev;
265 if (prev) {
266 next = prev->vm_next;
267 prev->vm_next = vma;
268 } else {
269 mm->mmap = vma;
270 if (rb_parent)
271 next = rb_entry(rb_parent,
272 struct vm_area_struct, vm_rb);
273 else
274 next = NULL;
275 }
276 vma->vm_next = next;
277 if (next)
278 next->vm_prev = vma;
279}
280
281/* Check if the vma is being used as a stack by this task */
282int vma_is_stack_for_current(struct vm_area_struct *vma)
283{
284 struct task_struct * __maybe_unused t = current;
285
286 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
287}
288
289#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
290void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
291{
292 mm->mmap_base = TASK_UNMAPPED_BASE;
293 mm->get_unmapped_area = arch_get_unmapped_area;
294}
295#endif
296
297/*
298 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
299 * back to the regular GUP.
300 * Note a difference with get_user_pages_fast: this always returns the
301 * number of pages pinned, 0 if no pages were pinned.
302 * If the architecture does not support this function, simply return with no
303 * pages pinned.
304 */
305int __weak __get_user_pages_fast(unsigned long start,
306 int nr_pages, int write, struct page **pages)
307{
308 return 0;
309}
310EXPORT_SYMBOL_GPL(__get_user_pages_fast);
311
312/**
313 * get_user_pages_fast() - pin user pages in memory
314 * @start: starting user address
315 * @nr_pages: number of pages from start to pin
316 * @write: whether pages will be written to
317 * @pages: array that receives pointers to the pages pinned.
318 * Should be at least nr_pages long.
319 *
320 * Returns number of pages pinned. This may be fewer than the number
321 * requested. If nr_pages is 0 or negative, returns 0. If no pages
322 * were pinned, returns -errno.
323 *
324 * get_user_pages_fast provides equivalent functionality to get_user_pages,
325 * operating on current and current->mm, with force=0 and vma=NULL. However
326 * unlike get_user_pages, it must be called without mmap_sem held.
327 *
328 * get_user_pages_fast may take mmap_sem and page table locks, so no
329 * assumptions can be made about lack of locking. get_user_pages_fast is to be
330 * implemented in a way that is advantageous (vs get_user_pages()) when the
331 * user memory area is already faulted in and present in ptes. However if the
332 * pages have to be faulted in, it may turn out to be slightly slower so
333 * callers need to carefully consider what to use. On many architectures,
334 * get_user_pages_fast simply falls back to get_user_pages.
335 */
336int __weak get_user_pages_fast(unsigned long start,
337 int nr_pages, int write, struct page **pages)
338{
339 return get_user_pages_unlocked(start, nr_pages, pages,
340 write ? FOLL_WRITE : 0);
341}
342EXPORT_SYMBOL_GPL(get_user_pages_fast);
343
344unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
345 unsigned long len, unsigned long prot,
346 unsigned long flag, unsigned long pgoff)
347{
348 unsigned long ret;
349 struct mm_struct *mm = current->mm;
350 unsigned long populate;
351 LIST_HEAD(uf);
352
353 ret = security_mmap_file(file, prot, flag);
354 if (!ret) {
355 if (down_write_killable(&mm->mmap_sem))
356 return -EINTR;
357 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
358 &populate, &uf);
359 up_write(&mm->mmap_sem);
360 userfaultfd_unmap_complete(mm, &uf);
361 if (populate)
362 mm_populate(ret, populate);
363 }
364 return ret;
365}
366
367unsigned long vm_mmap(struct file *file, unsigned long addr,
368 unsigned long len, unsigned long prot,
369 unsigned long flag, unsigned long offset)
370{
371 if (unlikely(offset + PAGE_ALIGN(len) < offset))
372 return -EINVAL;
373 if (unlikely(offset_in_page(offset)))
374 return -EINVAL;
375
376 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
377}
378EXPORT_SYMBOL(vm_mmap);
379
380/**
381 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
382 * failure, fall back to non-contiguous (vmalloc) allocation.
383 * @size: size of the request.
384 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
385 * @node: numa node to allocate from
386 *
387 * Uses kmalloc to get the memory but if the allocation fails then falls back
388 * to the vmalloc allocator. Use kvfree for freeing the memory.
389 *
390 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
391 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
392 * preferable to the vmalloc fallback, due to visible performance drawbacks.
393 *
394 * Any use of gfp flags outside of GFP_KERNEL should be consulted with mm people.
395 */
396void *kvmalloc_node(size_t size, gfp_t flags, int node)
397{
398 gfp_t kmalloc_flags = flags;
399 void *ret;
400
401 /*
402 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
403 * so the given set of flags has to be compatible.
404 */
405 WARN_ON_ONCE((flags & GFP_KERNEL) != GFP_KERNEL);
406
407 /*
408 * We want to attempt a large physically contiguous block first because
409 * it is less likely to fragment multiple larger blocks and therefore
410 * contribute to a long term fragmentation less than vmalloc fallback.
411 * However make sure that larger requests are not too disruptive - no
412 * OOM killer and no allocation failure warnings as we have a fallback.
413 */
414 if (size > PAGE_SIZE) {
415 kmalloc_flags |= __GFP_NOWARN;
416
417 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
418 kmalloc_flags |= __GFP_NORETRY;
419 }
420
421 ret = kmalloc_node(size, kmalloc_flags, node);
422
423 /*
424 * It doesn't really make sense to fallback to vmalloc for sub page
425 * requests
426 */
427 if (ret || size <= PAGE_SIZE)
428 return ret;
429
430 return __vmalloc_node_flags_caller(size, node, flags,
431 __builtin_return_address(0));
432}
433EXPORT_SYMBOL(kvmalloc_node);
434
435void kvfree(const void *addr)
436{
437 if (is_vmalloc_addr(addr))
438 vfree(addr);
439 else
440 kfree(addr);
441}
442EXPORT_SYMBOL(kvfree);
443
444static inline void *__page_rmapping(struct page *page)
445{
446 unsigned long mapping;
447
448 mapping = (unsigned long)page->mapping;
449 mapping &= ~PAGE_MAPPING_FLAGS;
450
451 return (void *)mapping;
452}
453
454/* Neutral page->mapping pointer to address_space or anon_vma or other */
455void *page_rmapping(struct page *page)
456{
457 page = compound_head(page);
458 return __page_rmapping(page);
459}
460
461/*
462 * Return true if this page is mapped into pagetables.
463 * For compound page it returns true if any subpage of compound page is mapped.
464 */
465bool page_mapped(struct page *page)
466{
467 int i;
468
469 if (likely(!PageCompound(page)))
470 return atomic_read(&page->_mapcount) >= 0;
471 page = compound_head(page);
472 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
473 return true;
474 if (PageHuge(page))
475 return false;
476 for (i = 0; i < hpage_nr_pages(page); i++) {
477 if (atomic_read(&page[i]._mapcount) >= 0)
478 return true;
479 }
480 return false;
481}
482EXPORT_SYMBOL(page_mapped);
483
484struct anon_vma *page_anon_vma(struct page *page)
485{
486 unsigned long mapping;
487
488 page = compound_head(page);
489 mapping = (unsigned long)page->mapping;
490 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
491 return NULL;
492 return __page_rmapping(page);
493}
494
495struct address_space *page_mapping(struct page *page)
496{
497 struct address_space *mapping;
498
499 page = compound_head(page);
500
501 /* This happens if someone calls flush_dcache_page on slab page */
502 if (unlikely(PageSlab(page)))
503 return NULL;
504
505 if (unlikely(PageSwapCache(page))) {
506 swp_entry_t entry;
507
508 entry.val = page_private(page);
509 return swap_address_space(entry);
510 }
511
512 mapping = page->mapping;
513 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
514 return NULL;
515
516 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
517}
518EXPORT_SYMBOL(page_mapping);
519
520/*
521 * For file cache pages, return the address_space, otherwise return NULL
522 */
523struct address_space *page_mapping_file(struct page *page)
524{
525 if (unlikely(PageSwapCache(page)))
526 return NULL;
527 return page_mapping(page);
528}
529
530/* Slow path of page_mapcount() for compound pages */
531int __page_mapcount(struct page *page)
532{
533 int ret;
534
535 ret = atomic_read(&page->_mapcount) + 1;
536 /*
537 * For file THP page->_mapcount contains total number of mapping
538 * of the page: no need to look into compound_mapcount.
539 */
540 if (!PageAnon(page) && !PageHuge(page))
541 return ret;
542 page = compound_head(page);
543 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
544 if (PageDoubleMap(page))
545 ret--;
546 return ret;
547}
548EXPORT_SYMBOL_GPL(__page_mapcount);
549
550int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
551int sysctl_overcommit_ratio __read_mostly = 50;
552unsigned long sysctl_overcommit_kbytes __read_mostly;
553int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
554unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
555unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
556
557int overcommit_ratio_handler(struct ctl_table *table, int write,
558 void __user *buffer, size_t *lenp,
559 loff_t *ppos)
560{
561 int ret;
562
563 ret = proc_dointvec(table, write, buffer, lenp, ppos);
564 if (ret == 0 && write)
565 sysctl_overcommit_kbytes = 0;
566 return ret;
567}
568
569int overcommit_kbytes_handler(struct ctl_table *table, int write,
570 void __user *buffer, size_t *lenp,
571 loff_t *ppos)
572{
573 int ret;
574
575 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
576 if (ret == 0 && write)
577 sysctl_overcommit_ratio = 0;
578 return ret;
579}
580
581/*
582 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
583 */
584unsigned long vm_commit_limit(void)
585{
586 unsigned long allowed;
587
588 if (sysctl_overcommit_kbytes)
589 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
590 else
591 allowed = ((totalram_pages - hugetlb_total_pages())
592 * sysctl_overcommit_ratio / 100);
593 allowed += total_swap_pages;
594
595 return allowed;
596}
597
598/*
599 * Make sure vm_committed_as in one cacheline and not cacheline shared with
600 * other variables. It can be updated by several CPUs frequently.
601 */
602struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
603
604/*
605 * The global memory commitment made in the system can be a metric
606 * that can be used to drive ballooning decisions when Linux is hosted
607 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
608 * balancing memory across competing virtual machines that are hosted.
609 * Several metrics drive this policy engine including the guest reported
610 * memory commitment.
611 */
612unsigned long vm_memory_committed(void)
613{
614 return percpu_counter_read_positive(&vm_committed_as);
615}
616EXPORT_SYMBOL_GPL(vm_memory_committed);
617
618/*
619 * Check that a process has enough memory to allocate a new virtual
620 * mapping. 0 means there is enough memory for the allocation to
621 * succeed and -ENOMEM implies there is not.
622 *
623 * We currently support three overcommit policies, which are set via the
624 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
625 *
626 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
627 * Additional code 2002 Jul 20 by Robert Love.
628 *
629 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
630 *
631 * Note this is a helper function intended to be used by LSMs which
632 * wish to use this logic.
633 */
634int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
635{
636 long free, allowed, reserve;
637
638 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
639 -(s64)vm_committed_as_batch * num_online_cpus(),
640 "memory commitment underflow");
641
642 vm_acct_memory(pages);
643
644 /*
645 * Sometimes we want to use more memory than we have
646 */
647 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
648 return 0;
649
650 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
651 free = global_zone_page_state(NR_FREE_PAGES);
652 free += global_node_page_state(NR_FILE_PAGES);
653
654 /*
655 * shmem pages shouldn't be counted as free in this
656 * case, they can't be purged, only swapped out, and
657 * that won't affect the overall amount of available
658 * memory in the system.
659 */
660 free -= global_node_page_state(NR_SHMEM);
661
662 free += get_nr_swap_pages();
663
664 /*
665 * Any slabs which are created with the
666 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
667 * which are reclaimable, under pressure. The dentry
668 * cache and most inode caches should fall into this
669 */
670 free += global_node_page_state(NR_SLAB_RECLAIMABLE);
671
672 /*
673 * Part of the kernel memory, which can be released
674 * under memory pressure.
675 */
676 free += global_node_page_state(
677 NR_INDIRECTLY_RECLAIMABLE_BYTES) >> PAGE_SHIFT;
678
679 /*
680 * Leave reserved pages. The pages are not for anonymous pages.
681 */
682 if (free <= totalreserve_pages)
683 goto error;
684 else
685 free -= totalreserve_pages;
686
687 /*
688 * Reserve some for root
689 */
690 if (!cap_sys_admin)
691 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
692
693 if (free > pages)
694 return 0;
695
696 goto error;
697 }
698
699 allowed = vm_commit_limit();
700 /*
701 * Reserve some for root
702 */
703 if (!cap_sys_admin)
704 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
705
706 /*
707 * Don't let a single process grow so big a user can't recover
708 */
709 if (mm) {
710 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
711 allowed -= min_t(long, mm->total_vm / 32, reserve);
712 }
713
714 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
715 return 0;
716error:
717 vm_unacct_memory(pages);
718
719 return -ENOMEM;
720}
721
722/**
723 * get_cmdline() - copy the cmdline value to a buffer.
724 * @task: the task whose cmdline value to copy.
725 * @buffer: the buffer to copy to.
726 * @buflen: the length of the buffer. Larger cmdline values are truncated
727 * to this length.
728 * Returns the size of the cmdline field copied. Note that the copy does
729 * not guarantee an ending NULL byte.
730 */
731int get_cmdline(struct task_struct *task, char *buffer, int buflen)
732{
733 int res = 0;
734 unsigned int len;
735 struct mm_struct *mm = get_task_mm(task);
736 unsigned long arg_start, arg_end, env_start, env_end;
737 if (!mm)
738 goto out;
739 if (!mm->arg_end)
740 goto out_mm; /* Shh! No looking before we're done */
741
742 down_read(&mm->mmap_sem);
743 arg_start = mm->arg_start;
744 arg_end = mm->arg_end;
745 env_start = mm->env_start;
746 env_end = mm->env_end;
747 up_read(&mm->mmap_sem);
748
749 len = arg_end - arg_start;
750
751 if (len > buflen)
752 len = buflen;
753
754 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
755
756 /*
757 * If the nul at the end of args has been overwritten, then
758 * assume application is using setproctitle(3).
759 */
760 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
761 len = strnlen(buffer, res);
762 if (len < res) {
763 res = len;
764 } else {
765 len = env_end - env_start;
766 if (len > buflen - res)
767 len = buflen - res;
768 res += access_process_vm(task, env_start,
769 buffer+res, len,
770 FOLL_FORCE);
771 res = strnlen(buffer, res);
772 }
773 }
774out_mm:
775 mmput(mm);
776out:
777 return res;
778}
1// SPDX-License-Identifier: GPL-2.0-only
2#include <linux/mm.h>
3#include <linux/slab.h>
4#include <linux/string.h>
5#include <linux/compiler.h>
6#include <linux/export.h>
7#include <linux/err.h>
8#include <linux/sched.h>
9#include <linux/sched/mm.h>
10#include <linux/sched/signal.h>
11#include <linux/sched/task_stack.h>
12#include <linux/security.h>
13#include <linux/swap.h>
14#include <linux/swapops.h>
15#include <linux/mman.h>
16#include <linux/hugetlb.h>
17#include <linux/vmalloc.h>
18#include <linux/userfaultfd_k.h>
19#include <linux/elf.h>
20#include <linux/elf-randomize.h>
21#include <linux/personality.h>
22#include <linux/random.h>
23#include <linux/processor.h>
24#include <linux/sizes.h>
25#include <linux/compat.h>
26
27#include <linux/uaccess.h>
28
29#include "internal.h"
30
31/**
32 * kfree_const - conditionally free memory
33 * @x: pointer to the memory
34 *
35 * Function calls kfree only if @x is not in .rodata section.
36 */
37void kfree_const(const void *x)
38{
39 if (!is_kernel_rodata((unsigned long)x))
40 kfree(x);
41}
42EXPORT_SYMBOL(kfree_const);
43
44/**
45 * kstrdup - allocate space for and copy an existing string
46 * @s: the string to duplicate
47 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
48 *
49 * Return: newly allocated copy of @s or %NULL in case of error
50 */
51char *kstrdup(const char *s, gfp_t gfp)
52{
53 size_t len;
54 char *buf;
55
56 if (!s)
57 return NULL;
58
59 len = strlen(s) + 1;
60 buf = kmalloc_track_caller(len, gfp);
61 if (buf)
62 memcpy(buf, s, len);
63 return buf;
64}
65EXPORT_SYMBOL(kstrdup);
66
67/**
68 * kstrdup_const - conditionally duplicate an existing const string
69 * @s: the string to duplicate
70 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
71 *
72 * Note: Strings allocated by kstrdup_const should be freed by kfree_const and
73 * must not be passed to krealloc().
74 *
75 * Return: source string if it is in .rodata section otherwise
76 * fallback to kstrdup.
77 */
78const char *kstrdup_const(const char *s, gfp_t gfp)
79{
80 if (is_kernel_rodata((unsigned long)s))
81 return s;
82
83 return kstrdup(s, gfp);
84}
85EXPORT_SYMBOL(kstrdup_const);
86
87/**
88 * kstrndup - allocate space for and copy an existing string
89 * @s: the string to duplicate
90 * @max: read at most @max chars from @s
91 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
92 *
93 * Note: Use kmemdup_nul() instead if the size is known exactly.
94 *
95 * Return: newly allocated copy of @s or %NULL in case of error
96 */
97char *kstrndup(const char *s, size_t max, gfp_t gfp)
98{
99 size_t len;
100 char *buf;
101
102 if (!s)
103 return NULL;
104
105 len = strnlen(s, max);
106 buf = kmalloc_track_caller(len+1, gfp);
107 if (buf) {
108 memcpy(buf, s, len);
109 buf[len] = '\0';
110 }
111 return buf;
112}
113EXPORT_SYMBOL(kstrndup);
114
115/**
116 * kmemdup - duplicate region of memory
117 *
118 * @src: memory region to duplicate
119 * @len: memory region length
120 * @gfp: GFP mask to use
121 *
122 * Return: newly allocated copy of @src or %NULL in case of error
123 */
124void *kmemdup(const void *src, size_t len, gfp_t gfp)
125{
126 void *p;
127
128 p = kmalloc_track_caller(len, gfp);
129 if (p)
130 memcpy(p, src, len);
131 return p;
132}
133EXPORT_SYMBOL(kmemdup);
134
135/**
136 * kmemdup_nul - Create a NUL-terminated string from unterminated data
137 * @s: The data to stringify
138 * @len: The size of the data
139 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
140 *
141 * Return: newly allocated copy of @s with NUL-termination or %NULL in
142 * case of error
143 */
144char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
145{
146 char *buf;
147
148 if (!s)
149 return NULL;
150
151 buf = kmalloc_track_caller(len + 1, gfp);
152 if (buf) {
153 memcpy(buf, s, len);
154 buf[len] = '\0';
155 }
156 return buf;
157}
158EXPORT_SYMBOL(kmemdup_nul);
159
160/**
161 * memdup_user - duplicate memory region from user space
162 *
163 * @src: source address in user space
164 * @len: number of bytes to copy
165 *
166 * Return: an ERR_PTR() on failure. Result is physically
167 * contiguous, to be freed by kfree().
168 */
169void *memdup_user(const void __user *src, size_t len)
170{
171 void *p;
172
173 p = kmalloc_track_caller(len, GFP_USER | __GFP_NOWARN);
174 if (!p)
175 return ERR_PTR(-ENOMEM);
176
177 if (copy_from_user(p, src, len)) {
178 kfree(p);
179 return ERR_PTR(-EFAULT);
180 }
181
182 return p;
183}
184EXPORT_SYMBOL(memdup_user);
185
186/**
187 * vmemdup_user - duplicate memory region from user space
188 *
189 * @src: source address in user space
190 * @len: number of bytes to copy
191 *
192 * Return: an ERR_PTR() on failure. Result may be not
193 * physically contiguous. Use kvfree() to free.
194 */
195void *vmemdup_user(const void __user *src, size_t len)
196{
197 void *p;
198
199 p = kvmalloc(len, GFP_USER);
200 if (!p)
201 return ERR_PTR(-ENOMEM);
202
203 if (copy_from_user(p, src, len)) {
204 kvfree(p);
205 return ERR_PTR(-EFAULT);
206 }
207
208 return p;
209}
210EXPORT_SYMBOL(vmemdup_user);
211
212/**
213 * strndup_user - duplicate an existing string from user space
214 * @s: The string to duplicate
215 * @n: Maximum number of bytes to copy, including the trailing NUL.
216 *
217 * Return: newly allocated copy of @s or an ERR_PTR() in case of error
218 */
219char *strndup_user(const char __user *s, long n)
220{
221 char *p;
222 long length;
223
224 length = strnlen_user(s, n);
225
226 if (!length)
227 return ERR_PTR(-EFAULT);
228
229 if (length > n)
230 return ERR_PTR(-EINVAL);
231
232 p = memdup_user(s, length);
233
234 if (IS_ERR(p))
235 return p;
236
237 p[length - 1] = '\0';
238
239 return p;
240}
241EXPORT_SYMBOL(strndup_user);
242
243/**
244 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
245 *
246 * @src: source address in user space
247 * @len: number of bytes to copy
248 *
249 * Return: an ERR_PTR() on failure.
250 */
251void *memdup_user_nul(const void __user *src, size_t len)
252{
253 char *p;
254
255 /*
256 * Always use GFP_KERNEL, since copy_from_user() can sleep and
257 * cause pagefault, which makes it pointless to use GFP_NOFS
258 * or GFP_ATOMIC.
259 */
260 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
261 if (!p)
262 return ERR_PTR(-ENOMEM);
263
264 if (copy_from_user(p, src, len)) {
265 kfree(p);
266 return ERR_PTR(-EFAULT);
267 }
268 p[len] = '\0';
269
270 return p;
271}
272EXPORT_SYMBOL(memdup_user_nul);
273
274void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
275 struct vm_area_struct *prev)
276{
277 struct vm_area_struct *next;
278
279 vma->vm_prev = prev;
280 if (prev) {
281 next = prev->vm_next;
282 prev->vm_next = vma;
283 } else {
284 next = mm->mmap;
285 mm->mmap = vma;
286 }
287 vma->vm_next = next;
288 if (next)
289 next->vm_prev = vma;
290}
291
292void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma)
293{
294 struct vm_area_struct *prev, *next;
295
296 next = vma->vm_next;
297 prev = vma->vm_prev;
298 if (prev)
299 prev->vm_next = next;
300 else
301 mm->mmap = next;
302 if (next)
303 next->vm_prev = prev;
304}
305
306/* Check if the vma is being used as a stack by this task */
307int vma_is_stack_for_current(struct vm_area_struct *vma)
308{
309 struct task_struct * __maybe_unused t = current;
310
311 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
312}
313
314/*
315 * Change backing file, only valid to use during initial VMA setup.
316 */
317void vma_set_file(struct vm_area_struct *vma, struct file *file)
318{
319 /* Changing an anonymous vma with this is illegal */
320 get_file(file);
321 swap(vma->vm_file, file);
322 fput(file);
323}
324EXPORT_SYMBOL(vma_set_file);
325
326#ifndef STACK_RND_MASK
327#define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
328#endif
329
330unsigned long randomize_stack_top(unsigned long stack_top)
331{
332 unsigned long random_variable = 0;
333
334 if (current->flags & PF_RANDOMIZE) {
335 random_variable = get_random_long();
336 random_variable &= STACK_RND_MASK;
337 random_variable <<= PAGE_SHIFT;
338 }
339#ifdef CONFIG_STACK_GROWSUP
340 return PAGE_ALIGN(stack_top) + random_variable;
341#else
342 return PAGE_ALIGN(stack_top) - random_variable;
343#endif
344}
345
346#ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
347unsigned long arch_randomize_brk(struct mm_struct *mm)
348{
349 /* Is the current task 32bit ? */
350 if (!IS_ENABLED(CONFIG_64BIT) || is_compat_task())
351 return randomize_page(mm->brk, SZ_32M);
352
353 return randomize_page(mm->brk, SZ_1G);
354}
355
356unsigned long arch_mmap_rnd(void)
357{
358 unsigned long rnd;
359
360#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
361 if (is_compat_task())
362 rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1);
363 else
364#endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
365 rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1);
366
367 return rnd << PAGE_SHIFT;
368}
369
370static int mmap_is_legacy(struct rlimit *rlim_stack)
371{
372 if (current->personality & ADDR_COMPAT_LAYOUT)
373 return 1;
374
375 if (rlim_stack->rlim_cur == RLIM_INFINITY)
376 return 1;
377
378 return sysctl_legacy_va_layout;
379}
380
381/*
382 * Leave enough space between the mmap area and the stack to honour ulimit in
383 * the face of randomisation.
384 */
385#define MIN_GAP (SZ_128M)
386#define MAX_GAP (STACK_TOP / 6 * 5)
387
388static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
389{
390 unsigned long gap = rlim_stack->rlim_cur;
391 unsigned long pad = stack_guard_gap;
392
393 /* Account for stack randomization if necessary */
394 if (current->flags & PF_RANDOMIZE)
395 pad += (STACK_RND_MASK << PAGE_SHIFT);
396
397 /* Values close to RLIM_INFINITY can overflow. */
398 if (gap + pad > gap)
399 gap += pad;
400
401 if (gap < MIN_GAP)
402 gap = MIN_GAP;
403 else if (gap > MAX_GAP)
404 gap = MAX_GAP;
405
406 return PAGE_ALIGN(STACK_TOP - gap - rnd);
407}
408
409void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
410{
411 unsigned long random_factor = 0UL;
412
413 if (current->flags & PF_RANDOMIZE)
414 random_factor = arch_mmap_rnd();
415
416 if (mmap_is_legacy(rlim_stack)) {
417 mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
418 mm->get_unmapped_area = arch_get_unmapped_area;
419 } else {
420 mm->mmap_base = mmap_base(random_factor, rlim_stack);
421 mm->get_unmapped_area = arch_get_unmapped_area_topdown;
422 }
423}
424#elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
425void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
426{
427 mm->mmap_base = TASK_UNMAPPED_BASE;
428 mm->get_unmapped_area = arch_get_unmapped_area;
429}
430#endif
431
432/**
433 * __account_locked_vm - account locked pages to an mm's locked_vm
434 * @mm: mm to account against
435 * @pages: number of pages to account
436 * @inc: %true if @pages should be considered positive, %false if not
437 * @task: task used to check RLIMIT_MEMLOCK
438 * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
439 *
440 * Assumes @task and @mm are valid (i.e. at least one reference on each), and
441 * that mmap_lock is held as writer.
442 *
443 * Return:
444 * * 0 on success
445 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
446 */
447int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
448 struct task_struct *task, bool bypass_rlim)
449{
450 unsigned long locked_vm, limit;
451 int ret = 0;
452
453 mmap_assert_write_locked(mm);
454
455 locked_vm = mm->locked_vm;
456 if (inc) {
457 if (!bypass_rlim) {
458 limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
459 if (locked_vm + pages > limit)
460 ret = -ENOMEM;
461 }
462 if (!ret)
463 mm->locked_vm = locked_vm + pages;
464 } else {
465 WARN_ON_ONCE(pages > locked_vm);
466 mm->locked_vm = locked_vm - pages;
467 }
468
469 pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
470 (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
471 locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
472 ret ? " - exceeded" : "");
473
474 return ret;
475}
476EXPORT_SYMBOL_GPL(__account_locked_vm);
477
478/**
479 * account_locked_vm - account locked pages to an mm's locked_vm
480 * @mm: mm to account against, may be NULL
481 * @pages: number of pages to account
482 * @inc: %true if @pages should be considered positive, %false if not
483 *
484 * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
485 *
486 * Return:
487 * * 0 on success, or if mm is NULL
488 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
489 */
490int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
491{
492 int ret;
493
494 if (pages == 0 || !mm)
495 return 0;
496
497 mmap_write_lock(mm);
498 ret = __account_locked_vm(mm, pages, inc, current,
499 capable(CAP_IPC_LOCK));
500 mmap_write_unlock(mm);
501
502 return ret;
503}
504EXPORT_SYMBOL_GPL(account_locked_vm);
505
506unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
507 unsigned long len, unsigned long prot,
508 unsigned long flag, unsigned long pgoff)
509{
510 unsigned long ret;
511 struct mm_struct *mm = current->mm;
512 unsigned long populate;
513 LIST_HEAD(uf);
514
515 ret = security_mmap_file(file, prot, flag);
516 if (!ret) {
517 if (mmap_write_lock_killable(mm))
518 return -EINTR;
519 ret = do_mmap(file, addr, len, prot, flag, pgoff, &populate,
520 &uf);
521 mmap_write_unlock(mm);
522 userfaultfd_unmap_complete(mm, &uf);
523 if (populate)
524 mm_populate(ret, populate);
525 }
526 return ret;
527}
528
529unsigned long vm_mmap(struct file *file, unsigned long addr,
530 unsigned long len, unsigned long prot,
531 unsigned long flag, unsigned long offset)
532{
533 if (unlikely(offset + PAGE_ALIGN(len) < offset))
534 return -EINVAL;
535 if (unlikely(offset_in_page(offset)))
536 return -EINVAL;
537
538 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
539}
540EXPORT_SYMBOL(vm_mmap);
541
542/**
543 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
544 * failure, fall back to non-contiguous (vmalloc) allocation.
545 * @size: size of the request.
546 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
547 * @node: numa node to allocate from
548 *
549 * Uses kmalloc to get the memory but if the allocation fails then falls back
550 * to the vmalloc allocator. Use kvfree for freeing the memory.
551 *
552 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
553 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
554 * preferable to the vmalloc fallback, due to visible performance drawbacks.
555 *
556 * Please note that any use of gfp flags outside of GFP_KERNEL is careful to not
557 * fall back to vmalloc.
558 *
559 * Return: pointer to the allocated memory of %NULL in case of failure
560 */
561void *kvmalloc_node(size_t size, gfp_t flags, int node)
562{
563 gfp_t kmalloc_flags = flags;
564 void *ret;
565
566 /*
567 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
568 * so the given set of flags has to be compatible.
569 */
570 if ((flags & GFP_KERNEL) != GFP_KERNEL)
571 return kmalloc_node(size, flags, node);
572
573 /*
574 * We want to attempt a large physically contiguous block first because
575 * it is less likely to fragment multiple larger blocks and therefore
576 * contribute to a long term fragmentation less than vmalloc fallback.
577 * However make sure that larger requests are not too disruptive - no
578 * OOM killer and no allocation failure warnings as we have a fallback.
579 */
580 if (size > PAGE_SIZE) {
581 kmalloc_flags |= __GFP_NOWARN;
582
583 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
584 kmalloc_flags |= __GFP_NORETRY;
585 }
586
587 ret = kmalloc_node(size, kmalloc_flags, node);
588
589 /*
590 * It doesn't really make sense to fallback to vmalloc for sub page
591 * requests
592 */
593 if (ret || size <= PAGE_SIZE)
594 return ret;
595
596 /* Don't even allow crazy sizes */
597 if (WARN_ON_ONCE(size > INT_MAX))
598 return NULL;
599
600 return __vmalloc_node(size, 1, flags, node,
601 __builtin_return_address(0));
602}
603EXPORT_SYMBOL(kvmalloc_node);
604
605/**
606 * kvfree() - Free memory.
607 * @addr: Pointer to allocated memory.
608 *
609 * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
610 * It is slightly more efficient to use kfree() or vfree() if you are certain
611 * that you know which one to use.
612 *
613 * Context: Either preemptible task context or not-NMI interrupt.
614 */
615void kvfree(const void *addr)
616{
617 if (is_vmalloc_addr(addr))
618 vfree(addr);
619 else
620 kfree(addr);
621}
622EXPORT_SYMBOL(kvfree);
623
624/**
625 * kvfree_sensitive - Free a data object containing sensitive information.
626 * @addr: address of the data object to be freed.
627 * @len: length of the data object.
628 *
629 * Use the special memzero_explicit() function to clear the content of a
630 * kvmalloc'ed object containing sensitive data to make sure that the
631 * compiler won't optimize out the data clearing.
632 */
633void kvfree_sensitive(const void *addr, size_t len)
634{
635 if (likely(!ZERO_OR_NULL_PTR(addr))) {
636 memzero_explicit((void *)addr, len);
637 kvfree(addr);
638 }
639}
640EXPORT_SYMBOL(kvfree_sensitive);
641
642static inline void *__page_rmapping(struct page *page)
643{
644 unsigned long mapping;
645
646 mapping = (unsigned long)page->mapping;
647 mapping &= ~PAGE_MAPPING_FLAGS;
648
649 return (void *)mapping;
650}
651
652/* Neutral page->mapping pointer to address_space or anon_vma or other */
653void *page_rmapping(struct page *page)
654{
655 page = compound_head(page);
656 return __page_rmapping(page);
657}
658
659/*
660 * Return true if this page is mapped into pagetables.
661 * For compound page it returns true if any subpage of compound page is mapped.
662 */
663bool page_mapped(struct page *page)
664{
665 int i;
666
667 if (likely(!PageCompound(page)))
668 return atomic_read(&page->_mapcount) >= 0;
669 page = compound_head(page);
670 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
671 return true;
672 if (PageHuge(page))
673 return false;
674 for (i = 0; i < compound_nr(page); i++) {
675 if (atomic_read(&page[i]._mapcount) >= 0)
676 return true;
677 }
678 return false;
679}
680EXPORT_SYMBOL(page_mapped);
681
682struct anon_vma *page_anon_vma(struct page *page)
683{
684 unsigned long mapping;
685
686 page = compound_head(page);
687 mapping = (unsigned long)page->mapping;
688 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
689 return NULL;
690 return __page_rmapping(page);
691}
692
693struct address_space *page_mapping(struct page *page)
694{
695 struct address_space *mapping;
696
697 page = compound_head(page);
698
699 /* This happens if someone calls flush_dcache_page on slab page */
700 if (unlikely(PageSlab(page)))
701 return NULL;
702
703 if (unlikely(PageSwapCache(page))) {
704 swp_entry_t entry;
705
706 entry.val = page_private(page);
707 return swap_address_space(entry);
708 }
709
710 mapping = page->mapping;
711 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
712 return NULL;
713
714 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
715}
716EXPORT_SYMBOL(page_mapping);
717
718/* Slow path of page_mapcount() for compound pages */
719int __page_mapcount(struct page *page)
720{
721 int ret;
722
723 ret = atomic_read(&page->_mapcount) + 1;
724 /*
725 * For file THP page->_mapcount contains total number of mapping
726 * of the page: no need to look into compound_mapcount.
727 */
728 if (!PageAnon(page) && !PageHuge(page))
729 return ret;
730 page = compound_head(page);
731 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
732 if (PageDoubleMap(page))
733 ret--;
734 return ret;
735}
736EXPORT_SYMBOL_GPL(__page_mapcount);
737
738void copy_huge_page(struct page *dst, struct page *src)
739{
740 unsigned i, nr = compound_nr(src);
741
742 for (i = 0; i < nr; i++) {
743 cond_resched();
744 copy_highpage(nth_page(dst, i), nth_page(src, i));
745 }
746}
747
748int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
749int sysctl_overcommit_ratio __read_mostly = 50;
750unsigned long sysctl_overcommit_kbytes __read_mostly;
751int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
752unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
753unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
754
755int overcommit_ratio_handler(struct ctl_table *table, int write, void *buffer,
756 size_t *lenp, loff_t *ppos)
757{
758 int ret;
759
760 ret = proc_dointvec(table, write, buffer, lenp, ppos);
761 if (ret == 0 && write)
762 sysctl_overcommit_kbytes = 0;
763 return ret;
764}
765
766static void sync_overcommit_as(struct work_struct *dummy)
767{
768 percpu_counter_sync(&vm_committed_as);
769}
770
771int overcommit_policy_handler(struct ctl_table *table, int write, void *buffer,
772 size_t *lenp, loff_t *ppos)
773{
774 struct ctl_table t;
775 int new_policy = -1;
776 int ret;
777
778 /*
779 * The deviation of sync_overcommit_as could be big with loose policy
780 * like OVERCOMMIT_ALWAYS/OVERCOMMIT_GUESS. When changing policy to
781 * strict OVERCOMMIT_NEVER, we need to reduce the deviation to comply
782 * with the strict "NEVER", and to avoid possible race condition (even
783 * though user usually won't too frequently do the switching to policy
784 * OVERCOMMIT_NEVER), the switch is done in the following order:
785 * 1. changing the batch
786 * 2. sync percpu count on each CPU
787 * 3. switch the policy
788 */
789 if (write) {
790 t = *table;
791 t.data = &new_policy;
792 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
793 if (ret || new_policy == -1)
794 return ret;
795
796 mm_compute_batch(new_policy);
797 if (new_policy == OVERCOMMIT_NEVER)
798 schedule_on_each_cpu(sync_overcommit_as);
799 sysctl_overcommit_memory = new_policy;
800 } else {
801 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
802 }
803
804 return ret;
805}
806
807int overcommit_kbytes_handler(struct ctl_table *table, int write, void *buffer,
808 size_t *lenp, loff_t *ppos)
809{
810 int ret;
811
812 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
813 if (ret == 0 && write)
814 sysctl_overcommit_ratio = 0;
815 return ret;
816}
817
818/*
819 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
820 */
821unsigned long vm_commit_limit(void)
822{
823 unsigned long allowed;
824
825 if (sysctl_overcommit_kbytes)
826 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
827 else
828 allowed = ((totalram_pages() - hugetlb_total_pages())
829 * sysctl_overcommit_ratio / 100);
830 allowed += total_swap_pages;
831
832 return allowed;
833}
834
835/*
836 * Make sure vm_committed_as in one cacheline and not cacheline shared with
837 * other variables. It can be updated by several CPUs frequently.
838 */
839struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
840
841/*
842 * The global memory commitment made in the system can be a metric
843 * that can be used to drive ballooning decisions when Linux is hosted
844 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
845 * balancing memory across competing virtual machines that are hosted.
846 * Several metrics drive this policy engine including the guest reported
847 * memory commitment.
848 *
849 * The time cost of this is very low for small platforms, and for big
850 * platform like a 2S/36C/72T Skylake server, in worst case where
851 * vm_committed_as's spinlock is under severe contention, the time cost
852 * could be about 30~40 microseconds.
853 */
854unsigned long vm_memory_committed(void)
855{
856 return percpu_counter_sum_positive(&vm_committed_as);
857}
858EXPORT_SYMBOL_GPL(vm_memory_committed);
859
860/*
861 * Check that a process has enough memory to allocate a new virtual
862 * mapping. 0 means there is enough memory for the allocation to
863 * succeed and -ENOMEM implies there is not.
864 *
865 * We currently support three overcommit policies, which are set via the
866 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting.rst
867 *
868 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
869 * Additional code 2002 Jul 20 by Robert Love.
870 *
871 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
872 *
873 * Note this is a helper function intended to be used by LSMs which
874 * wish to use this logic.
875 */
876int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
877{
878 long allowed;
879
880 vm_acct_memory(pages);
881
882 /*
883 * Sometimes we want to use more memory than we have
884 */
885 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
886 return 0;
887
888 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
889 if (pages > totalram_pages() + total_swap_pages)
890 goto error;
891 return 0;
892 }
893
894 allowed = vm_commit_limit();
895 /*
896 * Reserve some for root
897 */
898 if (!cap_sys_admin)
899 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
900
901 /*
902 * Don't let a single process grow so big a user can't recover
903 */
904 if (mm) {
905 long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
906
907 allowed -= min_t(long, mm->total_vm / 32, reserve);
908 }
909
910 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
911 return 0;
912error:
913 vm_unacct_memory(pages);
914
915 return -ENOMEM;
916}
917
918/**
919 * get_cmdline() - copy the cmdline value to a buffer.
920 * @task: the task whose cmdline value to copy.
921 * @buffer: the buffer to copy to.
922 * @buflen: the length of the buffer. Larger cmdline values are truncated
923 * to this length.
924 *
925 * Return: the size of the cmdline field copied. Note that the copy does
926 * not guarantee an ending NULL byte.
927 */
928int get_cmdline(struct task_struct *task, char *buffer, int buflen)
929{
930 int res = 0;
931 unsigned int len;
932 struct mm_struct *mm = get_task_mm(task);
933 unsigned long arg_start, arg_end, env_start, env_end;
934 if (!mm)
935 goto out;
936 if (!mm->arg_end)
937 goto out_mm; /* Shh! No looking before we're done */
938
939 spin_lock(&mm->arg_lock);
940 arg_start = mm->arg_start;
941 arg_end = mm->arg_end;
942 env_start = mm->env_start;
943 env_end = mm->env_end;
944 spin_unlock(&mm->arg_lock);
945
946 len = arg_end - arg_start;
947
948 if (len > buflen)
949 len = buflen;
950
951 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
952
953 /*
954 * If the nul at the end of args has been overwritten, then
955 * assume application is using setproctitle(3).
956 */
957 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
958 len = strnlen(buffer, res);
959 if (len < res) {
960 res = len;
961 } else {
962 len = env_end - env_start;
963 if (len > buflen - res)
964 len = buflen - res;
965 res += access_process_vm(task, env_start,
966 buffer+res, len,
967 FOLL_FORCE);
968 res = strnlen(buffer, res);
969 }
970 }
971out_mm:
972 mmput(mm);
973out:
974 return res;
975}
976
977int __weak memcmp_pages(struct page *page1, struct page *page2)
978{
979 char *addr1, *addr2;
980 int ret;
981
982 addr1 = kmap_atomic(page1);
983 addr2 = kmap_atomic(page2);
984 ret = memcmp(addr1, addr2, PAGE_SIZE);
985 kunmap_atomic(addr2);
986 kunmap_atomic(addr1);
987 return ret;
988}
989
990#ifdef CONFIG_PRINTK
991/**
992 * mem_dump_obj - Print available provenance information
993 * @object: object for which to find provenance information.
994 *
995 * This function uses pr_cont(), so that the caller is expected to have
996 * printed out whatever preamble is appropriate. The provenance information
997 * depends on the type of object and on how much debugging is enabled.
998 * For example, for a slab-cache object, the slab name is printed, and,
999 * if available, the return address and stack trace from the allocation
1000 * and last free path of that object.
1001 */
1002void mem_dump_obj(void *object)
1003{
1004 const char *type;
1005
1006 if (kmem_valid_obj(object)) {
1007 kmem_dump_obj(object);
1008 return;
1009 }
1010
1011 if (vmalloc_dump_obj(object))
1012 return;
1013
1014 if (virt_addr_valid(object))
1015 type = "non-slab/vmalloc memory";
1016 else if (object == NULL)
1017 type = "NULL pointer";
1018 else if (object == ZERO_SIZE_PTR)
1019 type = "zero-size pointer";
1020 else
1021 type = "non-paged memory";
1022
1023 pr_cont(" %s\n", type);
1024}
1025EXPORT_SYMBOL_GPL(mem_dump_obj);
1026#endif
1027
1028/*
1029 * A driver might set a page logically offline -- PageOffline() -- and
1030 * turn the page inaccessible in the hypervisor; after that, access to page
1031 * content can be fatal.
1032 *
1033 * Some special PFN walkers -- i.e., /proc/kcore -- read content of random
1034 * pages after checking PageOffline(); however, these PFN walkers can race
1035 * with drivers that set PageOffline().
1036 *
1037 * page_offline_freeze()/page_offline_thaw() allows for a subsystem to
1038 * synchronize with such drivers, achieving that a page cannot be set
1039 * PageOffline() while frozen.
1040 *
1041 * page_offline_begin()/page_offline_end() is used by drivers that care about
1042 * such races when setting a page PageOffline().
1043 */
1044static DECLARE_RWSEM(page_offline_rwsem);
1045
1046void page_offline_freeze(void)
1047{
1048 down_read(&page_offline_rwsem);
1049}
1050
1051void page_offline_thaw(void)
1052{
1053 up_read(&page_offline_rwsem);
1054}
1055
1056void page_offline_begin(void)
1057{
1058 down_write(&page_offline_rwsem);
1059}
1060EXPORT_SYMBOL(page_offline_begin);
1061
1062void page_offline_end(void)
1063{
1064 up_write(&page_offline_rwsem);
1065}
1066EXPORT_SYMBOL(page_offline_end);