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/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
15#include <asm/sections.h>
16#include <asm/uaccess.h>
17
18#include "internal.h"
19
20static inline int is_kernel_rodata(unsigned long addr)
21{
22 return addr >= (unsigned long)__start_rodata &&
23 addr < (unsigned long)__end_rodata;
24}
25
26/**
27 * kfree_const - conditionally free memory
28 * @x: pointer to the memory
29 *
30 * Function calls kfree only if @x is not in .rodata section.
31 */
32void kfree_const(const void *x)
33{
34 if (!is_kernel_rodata((unsigned long)x))
35 kfree(x);
36}
37EXPORT_SYMBOL(kfree_const);
38
39/**
40 * kstrdup - allocate space for and copy an existing string
41 * @s: the string to duplicate
42 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
43 */
44char *kstrdup(const char *s, gfp_t gfp)
45{
46 size_t len;
47 char *buf;
48
49 if (!s)
50 return NULL;
51
52 len = strlen(s) + 1;
53 buf = kmalloc_track_caller(len, gfp);
54 if (buf)
55 memcpy(buf, s, len);
56 return buf;
57}
58EXPORT_SYMBOL(kstrdup);
59
60/**
61 * kstrdup_const - conditionally duplicate an existing const string
62 * @s: the string to duplicate
63 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
64 *
65 * Function returns source string if it is in .rodata section otherwise it
66 * fallbacks to kstrdup.
67 * Strings allocated by kstrdup_const should be freed by kfree_const.
68 */
69const char *kstrdup_const(const char *s, gfp_t gfp)
70{
71 if (is_kernel_rodata((unsigned long)s))
72 return s;
73
74 return kstrdup(s, gfp);
75}
76EXPORT_SYMBOL(kstrdup_const);
77
78/**
79 * kstrndup - allocate space for and copy an existing string
80 * @s: the string to duplicate
81 * @max: read at most @max chars from @s
82 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
83 */
84char *kstrndup(const char *s, size_t max, gfp_t gfp)
85{
86 size_t len;
87 char *buf;
88
89 if (!s)
90 return NULL;
91
92 len = strnlen(s, max);
93 buf = kmalloc_track_caller(len+1, gfp);
94 if (buf) {
95 memcpy(buf, s, len);
96 buf[len] = '\0';
97 }
98 return buf;
99}
100EXPORT_SYMBOL(kstrndup);
101
102/**
103 * kmemdup - duplicate region of memory
104 *
105 * @src: memory region to duplicate
106 * @len: memory region length
107 * @gfp: GFP mask to use
108 */
109void *kmemdup(const void *src, size_t len, gfp_t gfp)
110{
111 void *p;
112
113 p = kmalloc_track_caller(len, gfp);
114 if (p)
115 memcpy(p, src, len);
116 return p;
117}
118EXPORT_SYMBOL(kmemdup);
119
120/**
121 * memdup_user - duplicate memory region from user space
122 *
123 * @src: source address in user space
124 * @len: number of bytes to copy
125 *
126 * Returns an ERR_PTR() on failure.
127 */
128void *memdup_user(const void __user *src, size_t len)
129{
130 void *p;
131
132 /*
133 * Always use GFP_KERNEL, since copy_from_user() can sleep and
134 * cause pagefault, which makes it pointless to use GFP_NOFS
135 * or GFP_ATOMIC.
136 */
137 p = kmalloc_track_caller(len, GFP_KERNEL);
138 if (!p)
139 return ERR_PTR(-ENOMEM);
140
141 if (copy_from_user(p, src, len)) {
142 kfree(p);
143 return ERR_PTR(-EFAULT);
144 }
145
146 return p;
147}
148EXPORT_SYMBOL(memdup_user);
149
150/*
151 * strndup_user - duplicate an existing string from user space
152 * @s: The string to duplicate
153 * @n: Maximum number of bytes to copy, including the trailing NUL.
154 */
155char *strndup_user(const char __user *s, long n)
156{
157 char *p;
158 long length;
159
160 length = strnlen_user(s, n);
161
162 if (!length)
163 return ERR_PTR(-EFAULT);
164
165 if (length > n)
166 return ERR_PTR(-EINVAL);
167
168 p = memdup_user(s, length);
169
170 if (IS_ERR(p))
171 return p;
172
173 p[length - 1] = '\0';
174
175 return p;
176}
177EXPORT_SYMBOL(strndup_user);
178
179/**
180 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
181 *
182 * @src: source address in user space
183 * @len: number of bytes to copy
184 *
185 * Returns an ERR_PTR() on failure.
186 */
187void *memdup_user_nul(const void __user *src, size_t len)
188{
189 char *p;
190
191 /*
192 * Always use GFP_KERNEL, since copy_from_user() can sleep and
193 * cause pagefault, which makes it pointless to use GFP_NOFS
194 * or GFP_ATOMIC.
195 */
196 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
197 if (!p)
198 return ERR_PTR(-ENOMEM);
199
200 if (copy_from_user(p, src, len)) {
201 kfree(p);
202 return ERR_PTR(-EFAULT);
203 }
204 p[len] = '\0';
205
206 return p;
207}
208EXPORT_SYMBOL(memdup_user_nul);
209
210void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
211 struct vm_area_struct *prev, struct rb_node *rb_parent)
212{
213 struct vm_area_struct *next;
214
215 vma->vm_prev = prev;
216 if (prev) {
217 next = prev->vm_next;
218 prev->vm_next = vma;
219 } else {
220 mm->mmap = vma;
221 if (rb_parent)
222 next = rb_entry(rb_parent,
223 struct vm_area_struct, vm_rb);
224 else
225 next = NULL;
226 }
227 vma->vm_next = next;
228 if (next)
229 next->vm_prev = vma;
230}
231
232/* Check if the vma is being used as a stack by this task */
233int vma_is_stack_for_task(struct vm_area_struct *vma, struct task_struct *t)
234{
235 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
236}
237
238#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
239void arch_pick_mmap_layout(struct mm_struct *mm)
240{
241 mm->mmap_base = TASK_UNMAPPED_BASE;
242 mm->get_unmapped_area = arch_get_unmapped_area;
243}
244#endif
245
246/*
247 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
248 * back to the regular GUP.
249 * If the architecture not support this function, simply return with no
250 * page pinned
251 */
252int __weak __get_user_pages_fast(unsigned long start,
253 int nr_pages, int write, struct page **pages)
254{
255 return 0;
256}
257EXPORT_SYMBOL_GPL(__get_user_pages_fast);
258
259/**
260 * get_user_pages_fast() - pin user pages in memory
261 * @start: starting user address
262 * @nr_pages: number of pages from start to pin
263 * @write: whether pages will be written to
264 * @pages: array that receives pointers to the pages pinned.
265 * Should be at least nr_pages long.
266 *
267 * Returns number of pages pinned. This may be fewer than the number
268 * requested. If nr_pages is 0 or negative, returns 0. If no pages
269 * were pinned, returns -errno.
270 *
271 * get_user_pages_fast provides equivalent functionality to get_user_pages,
272 * operating on current and current->mm, with force=0 and vma=NULL. However
273 * unlike get_user_pages, it must be called without mmap_sem held.
274 *
275 * get_user_pages_fast may take mmap_sem and page table locks, so no
276 * assumptions can be made about lack of locking. get_user_pages_fast is to be
277 * implemented in a way that is advantageous (vs get_user_pages()) when the
278 * user memory area is already faulted in and present in ptes. However if the
279 * pages have to be faulted in, it may turn out to be slightly slower so
280 * callers need to carefully consider what to use. On many architectures,
281 * get_user_pages_fast simply falls back to get_user_pages.
282 */
283int __weak get_user_pages_fast(unsigned long start,
284 int nr_pages, int write, struct page **pages)
285{
286 return get_user_pages_unlocked(start, nr_pages, write, 0, pages);
287}
288EXPORT_SYMBOL_GPL(get_user_pages_fast);
289
290unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
291 unsigned long len, unsigned long prot,
292 unsigned long flag, unsigned long pgoff)
293{
294 unsigned long ret;
295 struct mm_struct *mm = current->mm;
296 unsigned long populate;
297
298 ret = security_mmap_file(file, prot, flag);
299 if (!ret) {
300 down_write(&mm->mmap_sem);
301 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
302 &populate);
303 up_write(&mm->mmap_sem);
304 if (populate)
305 mm_populate(ret, populate);
306 }
307 return ret;
308}
309
310unsigned long vm_mmap(struct file *file, unsigned long addr,
311 unsigned long len, unsigned long prot,
312 unsigned long flag, unsigned long offset)
313{
314 if (unlikely(offset + PAGE_ALIGN(len) < offset))
315 return -EINVAL;
316 if (unlikely(offset_in_page(offset)))
317 return -EINVAL;
318
319 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
320}
321EXPORT_SYMBOL(vm_mmap);
322
323void kvfree(const void *addr)
324{
325 if (is_vmalloc_addr(addr))
326 vfree(addr);
327 else
328 kfree(addr);
329}
330EXPORT_SYMBOL(kvfree);
331
332static inline void *__page_rmapping(struct page *page)
333{
334 unsigned long mapping;
335
336 mapping = (unsigned long)page->mapping;
337 mapping &= ~PAGE_MAPPING_FLAGS;
338
339 return (void *)mapping;
340}
341
342/* Neutral page->mapping pointer to address_space or anon_vma or other */
343void *page_rmapping(struct page *page)
344{
345 page = compound_head(page);
346 return __page_rmapping(page);
347}
348
349struct anon_vma *page_anon_vma(struct page *page)
350{
351 unsigned long mapping;
352
353 page = compound_head(page);
354 mapping = (unsigned long)page->mapping;
355 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
356 return NULL;
357 return __page_rmapping(page);
358}
359
360struct address_space *page_mapping(struct page *page)
361{
362 struct address_space *mapping;
363
364 page = compound_head(page);
365
366 /* This happens if someone calls flush_dcache_page on slab page */
367 if (unlikely(PageSlab(page)))
368 return NULL;
369
370 if (unlikely(PageSwapCache(page))) {
371 swp_entry_t entry;
372
373 entry.val = page_private(page);
374 return swap_address_space(entry);
375 }
376
377 mapping = page->mapping;
378 if ((unsigned long)mapping & PAGE_MAPPING_FLAGS)
379 return NULL;
380 return mapping;
381}
382
383/* Slow path of page_mapcount() for compound pages */
384int __page_mapcount(struct page *page)
385{
386 int ret;
387
388 ret = atomic_read(&page->_mapcount) + 1;
389 page = compound_head(page);
390 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
391 if (PageDoubleMap(page))
392 ret--;
393 return ret;
394}
395EXPORT_SYMBOL_GPL(__page_mapcount);
396
397int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
398int sysctl_overcommit_ratio __read_mostly = 50;
399unsigned long sysctl_overcommit_kbytes __read_mostly;
400int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
401unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
402unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
403
404int overcommit_ratio_handler(struct ctl_table *table, int write,
405 void __user *buffer, size_t *lenp,
406 loff_t *ppos)
407{
408 int ret;
409
410 ret = proc_dointvec(table, write, buffer, lenp, ppos);
411 if (ret == 0 && write)
412 sysctl_overcommit_kbytes = 0;
413 return ret;
414}
415
416int overcommit_kbytes_handler(struct ctl_table *table, int write,
417 void __user *buffer, size_t *lenp,
418 loff_t *ppos)
419{
420 int ret;
421
422 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
423 if (ret == 0 && write)
424 sysctl_overcommit_ratio = 0;
425 return ret;
426}
427
428/*
429 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
430 */
431unsigned long vm_commit_limit(void)
432{
433 unsigned long allowed;
434
435 if (sysctl_overcommit_kbytes)
436 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
437 else
438 allowed = ((totalram_pages - hugetlb_total_pages())
439 * sysctl_overcommit_ratio / 100);
440 allowed += total_swap_pages;
441
442 return allowed;
443}
444
445/*
446 * Make sure vm_committed_as in one cacheline and not cacheline shared with
447 * other variables. It can be updated by several CPUs frequently.
448 */
449struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
450
451/*
452 * The global memory commitment made in the system can be a metric
453 * that can be used to drive ballooning decisions when Linux is hosted
454 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
455 * balancing memory across competing virtual machines that are hosted.
456 * Several metrics drive this policy engine including the guest reported
457 * memory commitment.
458 */
459unsigned long vm_memory_committed(void)
460{
461 return percpu_counter_read_positive(&vm_committed_as);
462}
463EXPORT_SYMBOL_GPL(vm_memory_committed);
464
465/*
466 * Check that a process has enough memory to allocate a new virtual
467 * mapping. 0 means there is enough memory for the allocation to
468 * succeed and -ENOMEM implies there is not.
469 *
470 * We currently support three overcommit policies, which are set via the
471 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
472 *
473 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
474 * Additional code 2002 Jul 20 by Robert Love.
475 *
476 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
477 *
478 * Note this is a helper function intended to be used by LSMs which
479 * wish to use this logic.
480 */
481int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
482{
483 long free, allowed, reserve;
484
485 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
486 -(s64)vm_committed_as_batch * num_online_cpus(),
487 "memory commitment underflow");
488
489 vm_acct_memory(pages);
490
491 /*
492 * Sometimes we want to use more memory than we have
493 */
494 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
495 return 0;
496
497 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
498 free = global_page_state(NR_FREE_PAGES);
499 free += global_page_state(NR_FILE_PAGES);
500
501 /*
502 * shmem pages shouldn't be counted as free in this
503 * case, they can't be purged, only swapped out, and
504 * that won't affect the overall amount of available
505 * memory in the system.
506 */
507 free -= global_page_state(NR_SHMEM);
508
509 free += get_nr_swap_pages();
510
511 /*
512 * Any slabs which are created with the
513 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
514 * which are reclaimable, under pressure. The dentry
515 * cache and most inode caches should fall into this
516 */
517 free += global_page_state(NR_SLAB_RECLAIMABLE);
518
519 /*
520 * Leave reserved pages. The pages are not for anonymous pages.
521 */
522 if (free <= totalreserve_pages)
523 goto error;
524 else
525 free -= totalreserve_pages;
526
527 /*
528 * Reserve some for root
529 */
530 if (!cap_sys_admin)
531 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
532
533 if (free > pages)
534 return 0;
535
536 goto error;
537 }
538
539 allowed = vm_commit_limit();
540 /*
541 * Reserve some for root
542 */
543 if (!cap_sys_admin)
544 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
545
546 /*
547 * Don't let a single process grow so big a user can't recover
548 */
549 if (mm) {
550 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
551 allowed -= min_t(long, mm->total_vm / 32, reserve);
552 }
553
554 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
555 return 0;
556error:
557 vm_unacct_memory(pages);
558
559 return -ENOMEM;
560}
561
562/**
563 * get_cmdline() - copy the cmdline value to a buffer.
564 * @task: the task whose cmdline value to copy.
565 * @buffer: the buffer to copy to.
566 * @buflen: the length of the buffer. Larger cmdline values are truncated
567 * to this length.
568 * Returns the size of the cmdline field copied. Note that the copy does
569 * not guarantee an ending NULL byte.
570 */
571int get_cmdline(struct task_struct *task, char *buffer, int buflen)
572{
573 int res = 0;
574 unsigned int len;
575 struct mm_struct *mm = get_task_mm(task);
576 unsigned long arg_start, arg_end, env_start, env_end;
577 if (!mm)
578 goto out;
579 if (!mm->arg_end)
580 goto out_mm; /* Shh! No looking before we're done */
581
582 down_read(&mm->mmap_sem);
583 arg_start = mm->arg_start;
584 arg_end = mm->arg_end;
585 env_start = mm->env_start;
586 env_end = mm->env_end;
587 up_read(&mm->mmap_sem);
588
589 len = arg_end - arg_start;
590
591 if (len > buflen)
592 len = buflen;
593
594 res = access_process_vm(task, arg_start, buffer, len, 0);
595
596 /*
597 * If the nul at the end of args has been overwritten, then
598 * assume application is using setproctitle(3).
599 */
600 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
601 len = strnlen(buffer, res);
602 if (len < res) {
603 res = len;
604 } else {
605 len = env_end - env_start;
606 if (len > buflen - res)
607 len = buflen - res;
608 res += access_process_vm(task, env_start,
609 buffer+res, len, 0);
610 res = strnlen(buffer, res);
611 }
612 }
613out_mm:
614 mmput(mm);
615out:
616 return res;
617}
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}