xref: /openbmc/linux/mm/util.c (revision 4e1a33b1)
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/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>
15 
16 #include <asm/sections.h>
17 #include <linux/uaccess.h>
18 
19 #include "internal.h"
20 
21 static inline int is_kernel_rodata(unsigned long addr)
22 {
23 	return addr >= (unsigned long)__start_rodata &&
24 		addr < (unsigned long)__end_rodata;
25 }
26 
27 /**
28  * kfree_const - conditionally free memory
29  * @x: pointer to the memory
30  *
31  * Function calls kfree only if @x is not in .rodata section.
32  */
33 void kfree_const(const void *x)
34 {
35 	if (!is_kernel_rodata((unsigned long)x))
36 		kfree(x);
37 }
38 EXPORT_SYMBOL(kfree_const);
39 
40 /**
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
44  */
45 char *kstrdup(const char *s, gfp_t gfp)
46 {
47 	size_t len;
48 	char *buf;
49 
50 	if (!s)
51 		return NULL;
52 
53 	len = strlen(s) + 1;
54 	buf = kmalloc_track_caller(len, gfp);
55 	if (buf)
56 		memcpy(buf, s, len);
57 	return buf;
58 }
59 EXPORT_SYMBOL(kstrdup);
60 
61 /**
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
65  *
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.
69  */
70 const char *kstrdup_const(const char *s, gfp_t gfp)
71 {
72 	if (is_kernel_rodata((unsigned long)s))
73 		return s;
74 
75 	return kstrdup(s, gfp);
76 }
77 EXPORT_SYMBOL(kstrdup_const);
78 
79 /**
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
84  */
85 char *kstrndup(const char *s, size_t max, gfp_t gfp)
86 {
87 	size_t len;
88 	char *buf;
89 
90 	if (!s)
91 		return NULL;
92 
93 	len = strnlen(s, max);
94 	buf = kmalloc_track_caller(len+1, gfp);
95 	if (buf) {
96 		memcpy(buf, s, len);
97 		buf[len] = '\0';
98 	}
99 	return buf;
100 }
101 EXPORT_SYMBOL(kstrndup);
102 
103 /**
104  * kmemdup - duplicate region of memory
105  *
106  * @src: memory region to duplicate
107  * @len: memory region length
108  * @gfp: GFP mask to use
109  */
110 void *kmemdup(const void *src, size_t len, gfp_t gfp)
111 {
112 	void *p;
113 
114 	p = kmalloc_track_caller(len, gfp);
115 	if (p)
116 		memcpy(p, src, len);
117 	return p;
118 }
119 EXPORT_SYMBOL(kmemdup);
120 
121 /**
122  * memdup_user - duplicate memory region from user space
123  *
124  * @src: source address in user space
125  * @len: number of bytes to copy
126  *
127  * Returns an ERR_PTR() on failure.
128  */
129 void *memdup_user(const void __user *src, size_t len)
130 {
131 	void *p;
132 
133 	/*
134 	 * Always use GFP_KERNEL, since copy_from_user() can sleep and
135 	 * cause pagefault, which makes it pointless to use GFP_NOFS
136 	 * or GFP_ATOMIC.
137 	 */
138 	p = kmalloc_track_caller(len, GFP_KERNEL);
139 	if (!p)
140 		return ERR_PTR(-ENOMEM);
141 
142 	if (copy_from_user(p, src, len)) {
143 		kfree(p);
144 		return ERR_PTR(-EFAULT);
145 	}
146 
147 	return p;
148 }
149 EXPORT_SYMBOL(memdup_user);
150 
151 /*
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.
155  */
156 char *strndup_user(const char __user *s, long n)
157 {
158 	char *p;
159 	long length;
160 
161 	length = strnlen_user(s, n);
162 
163 	if (!length)
164 		return ERR_PTR(-EFAULT);
165 
166 	if (length > n)
167 		return ERR_PTR(-EINVAL);
168 
169 	p = memdup_user(s, length);
170 
171 	if (IS_ERR(p))
172 		return p;
173 
174 	p[length - 1] = '\0';
175 
176 	return p;
177 }
178 EXPORT_SYMBOL(strndup_user);
179 
180 /**
181  * memdup_user_nul - duplicate memory region from user space and NUL-terminate
182  *
183  * @src: source address in user space
184  * @len: number of bytes to copy
185  *
186  * Returns an ERR_PTR() on failure.
187  */
188 void *memdup_user_nul(const void __user *src, size_t len)
189 {
190 	char *p;
191 
192 	/*
193 	 * Always use GFP_KERNEL, since copy_from_user() can sleep and
194 	 * cause pagefault, which makes it pointless to use GFP_NOFS
195 	 * or GFP_ATOMIC.
196 	 */
197 	p = kmalloc_track_caller(len + 1, GFP_KERNEL);
198 	if (!p)
199 		return ERR_PTR(-ENOMEM);
200 
201 	if (copy_from_user(p, src, len)) {
202 		kfree(p);
203 		return ERR_PTR(-EFAULT);
204 	}
205 	p[len] = '\0';
206 
207 	return p;
208 }
209 EXPORT_SYMBOL(memdup_user_nul);
210 
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)
213 {
214 	struct vm_area_struct *next;
215 
216 	vma->vm_prev = prev;
217 	if (prev) {
218 		next = prev->vm_next;
219 		prev->vm_next = vma;
220 	} else {
221 		mm->mmap = vma;
222 		if (rb_parent)
223 			next = rb_entry(rb_parent,
224 					struct vm_area_struct, vm_rb);
225 		else
226 			next = NULL;
227 	}
228 	vma->vm_next = next;
229 	if (next)
230 		next->vm_prev = vma;
231 }
232 
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)
235 {
236 	struct task_struct * __maybe_unused t = current;
237 
238 	return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
239 }
240 
241 #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
242 void arch_pick_mmap_layout(struct mm_struct *mm)
243 {
244 	mm->mmap_base = TASK_UNMAPPED_BASE;
245 	mm->get_unmapped_area = arch_get_unmapped_area;
246 }
247 #endif
248 
249 /*
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
253  * page pinned
254  */
255 int __weak __get_user_pages_fast(unsigned long start,
256 				 int nr_pages, int write, struct page **pages)
257 {
258 	return 0;
259 }
260 EXPORT_SYMBOL_GPL(__get_user_pages_fast);
261 
262 /**
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.
269  *
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.
273  *
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.
277  *
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.
285  */
286 int __weak get_user_pages_fast(unsigned long start,
287 				int nr_pages, int write, struct page **pages)
288 {
289 	return get_user_pages_unlocked(start, nr_pages, pages,
290 				       write ? FOLL_WRITE : 0);
291 }
292 EXPORT_SYMBOL_GPL(get_user_pages_fast);
293 
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)
297 {
298 	unsigned long ret;
299 	struct mm_struct *mm = current->mm;
300 	unsigned long populate;
301 	LIST_HEAD(uf);
302 
303 	ret = security_mmap_file(file, prot, flag);
304 	if (!ret) {
305 		if (down_write_killable(&mm->mmap_sem))
306 			return -EINTR;
307 		ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
308 				    &populate, &uf);
309 		up_write(&mm->mmap_sem);
310 		userfaultfd_unmap_complete(mm, &uf);
311 		if (populate)
312 			mm_populate(ret, populate);
313 	}
314 	return ret;
315 }
316 
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)
320 {
321 	if (unlikely(offset + PAGE_ALIGN(len) < offset))
322 		return -EINVAL;
323 	if (unlikely(offset_in_page(offset)))
324 		return -EINVAL;
325 
326 	return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
327 }
328 EXPORT_SYMBOL(vm_mmap);
329 
330 void kvfree(const void *addr)
331 {
332 	if (is_vmalloc_addr(addr))
333 		vfree(addr);
334 	else
335 		kfree(addr);
336 }
337 EXPORT_SYMBOL(kvfree);
338 
339 static inline void *__page_rmapping(struct page *page)
340 {
341 	unsigned long mapping;
342 
343 	mapping = (unsigned long)page->mapping;
344 	mapping &= ~PAGE_MAPPING_FLAGS;
345 
346 	return (void *)mapping;
347 }
348 
349 /* Neutral page->mapping pointer to address_space or anon_vma or other */
350 void *page_rmapping(struct page *page)
351 {
352 	page = compound_head(page);
353 	return __page_rmapping(page);
354 }
355 
356 /*
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.
359  */
360 bool page_mapped(struct page *page)
361 {
362 	int i;
363 
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)
368 		return true;
369 	if (PageHuge(page))
370 		return false;
371 	for (i = 0; i < hpage_nr_pages(page); i++) {
372 		if (atomic_read(&page[i]._mapcount) >= 0)
373 			return true;
374 	}
375 	return false;
376 }
377 EXPORT_SYMBOL(page_mapped);
378 
379 struct anon_vma *page_anon_vma(struct page *page)
380 {
381 	unsigned long mapping;
382 
383 	page = compound_head(page);
384 	mapping = (unsigned long)page->mapping;
385 	if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
386 		return NULL;
387 	return __page_rmapping(page);
388 }
389 
390 struct address_space *page_mapping(struct page *page)
391 {
392 	struct address_space *mapping;
393 
394 	page = compound_head(page);
395 
396 	/* This happens if someone calls flush_dcache_page on slab page */
397 	if (unlikely(PageSlab(page)))
398 		return NULL;
399 
400 	if (unlikely(PageSwapCache(page))) {
401 		swp_entry_t entry;
402 
403 		entry.val = page_private(page);
404 		return swap_address_space(entry);
405 	}
406 
407 	mapping = page->mapping;
408 	if ((unsigned long)mapping & PAGE_MAPPING_ANON)
409 		return NULL;
410 
411 	return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
412 }
413 EXPORT_SYMBOL(page_mapping);
414 
415 /* Slow path of page_mapcount() for compound pages */
416 int __page_mapcount(struct page *page)
417 {
418 	int ret;
419 
420 	ret = atomic_read(&page->_mapcount) + 1;
421 	/*
422 	 * For file THP page->_mapcount contains total number of mapping
423 	 * of the page: no need to look into compound_mapcount.
424 	 */
425 	if (!PageAnon(page) && !PageHuge(page))
426 		return ret;
427 	page = compound_head(page);
428 	ret += atomic_read(compound_mapcount_ptr(page)) + 1;
429 	if (PageDoubleMap(page))
430 		ret--;
431 	return ret;
432 }
433 EXPORT_SYMBOL_GPL(__page_mapcount);
434 
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 */
441 
442 int overcommit_ratio_handler(struct ctl_table *table, int write,
443 			     void __user *buffer, size_t *lenp,
444 			     loff_t *ppos)
445 {
446 	int ret;
447 
448 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
449 	if (ret == 0 && write)
450 		sysctl_overcommit_kbytes = 0;
451 	return ret;
452 }
453 
454 int overcommit_kbytes_handler(struct ctl_table *table, int write,
455 			     void __user *buffer, size_t *lenp,
456 			     loff_t *ppos)
457 {
458 	int ret;
459 
460 	ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
461 	if (ret == 0 && write)
462 		sysctl_overcommit_ratio = 0;
463 	return ret;
464 }
465 
466 /*
467  * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
468  */
469 unsigned long vm_commit_limit(void)
470 {
471 	unsigned long allowed;
472 
473 	if (sysctl_overcommit_kbytes)
474 		allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
475 	else
476 		allowed = ((totalram_pages - hugetlb_total_pages())
477 			   * sysctl_overcommit_ratio / 100);
478 	allowed += total_swap_pages;
479 
480 	return allowed;
481 }
482 
483 /*
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.
486  */
487 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
488 
489 /*
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
495  * memory commitment.
496  */
497 unsigned long vm_memory_committed(void)
498 {
499 	return percpu_counter_read_positive(&vm_committed_as);
500 }
501 EXPORT_SYMBOL_GPL(vm_memory_committed);
502 
503 /*
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.
507  *
508  * We currently support three overcommit policies, which are set via the
509  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
510  *
511  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
512  * Additional code 2002 Jul 20 by Robert Love.
513  *
514  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
515  *
516  * Note this is a helper function intended to be used by LSMs which
517  * wish to use this logic.
518  */
519 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
520 {
521 	long free, allowed, reserve;
522 
523 	VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
524 			-(s64)vm_committed_as_batch * num_online_cpus(),
525 			"memory commitment underflow");
526 
527 	vm_acct_memory(pages);
528 
529 	/*
530 	 * Sometimes we want to use more memory than we have
531 	 */
532 	if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
533 		return 0;
534 
535 	if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
536 		free = global_page_state(NR_FREE_PAGES);
537 		free += global_node_page_state(NR_FILE_PAGES);
538 
539 		/*
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.
544 		 */
545 		free -= global_node_page_state(NR_SHMEM);
546 
547 		free += get_nr_swap_pages();
548 
549 		/*
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
554 		 */
555 		free += global_page_state(NR_SLAB_RECLAIMABLE);
556 
557 		/*
558 		 * Leave reserved pages. The pages are not for anonymous pages.
559 		 */
560 		if (free <= totalreserve_pages)
561 			goto error;
562 		else
563 			free -= totalreserve_pages;
564 
565 		/*
566 		 * Reserve some for root
567 		 */
568 		if (!cap_sys_admin)
569 			free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
570 
571 		if (free > pages)
572 			return 0;
573 
574 		goto error;
575 	}
576 
577 	allowed = vm_commit_limit();
578 	/*
579 	 * Reserve some for root
580 	 */
581 	if (!cap_sys_admin)
582 		allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
583 
584 	/*
585 	 * Don't let a single process grow so big a user can't recover
586 	 */
587 	if (mm) {
588 		reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
589 		allowed -= min_t(long, mm->total_vm / 32, reserve);
590 	}
591 
592 	if (percpu_counter_read_positive(&vm_committed_as) < allowed)
593 		return 0;
594 error:
595 	vm_unacct_memory(pages);
596 
597 	return -ENOMEM;
598 }
599 
600 /**
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
605  *            to this length.
606  * Returns the size of the cmdline field copied. Note that the copy does
607  * not guarantee an ending NULL byte.
608  */
609 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
610 {
611 	int res = 0;
612 	unsigned int len;
613 	struct mm_struct *mm = get_task_mm(task);
614 	unsigned long arg_start, arg_end, env_start, env_end;
615 	if (!mm)
616 		goto out;
617 	if (!mm->arg_end)
618 		goto out_mm;	/* Shh! No looking before we're done */
619 
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);
626 
627 	len = arg_end - arg_start;
628 
629 	if (len > buflen)
630 		len = buflen;
631 
632 	res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
633 
634 	/*
635 	 * If the nul at the end of args has been overwritten, then
636 	 * assume application is using setproctitle(3).
637 	 */
638 	if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
639 		len = strnlen(buffer, res);
640 		if (len < res) {
641 			res = len;
642 		} else {
643 			len = env_end - env_start;
644 			if (len > buflen - res)
645 				len = buflen - res;
646 			res += access_process_vm(task, env_start,
647 						 buffer+res, len,
648 						 FOLL_FORCE);
649 			res = strnlen(buffer, res);
650 		}
651 	}
652 out_mm:
653 	mmput(mm);
654 out:
655 	return res;
656 }
657