]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - mm/util.c
mm: move most file-based accounting to the node
[mirror_ubuntu-bionic-kernel.git] / mm / util.c
CommitLineData
16d69265 1#include <linux/mm.h>
30992c97
MM
2#include <linux/slab.h>
3#include <linux/string.h>
3b32123d 4#include <linux/compiler.h>
b95f1b31 5#include <linux/export.h>
96840aa0 6#include <linux/err.h>
3b8f14b4 7#include <linux/sched.h>
eb36c587 8#include <linux/security.h>
9800339b 9#include <linux/swap.h>
33806f06 10#include <linux/swapops.h>
00619bcc
JM
11#include <linux/mman.h>
12#include <linux/hugetlb.h>
39f1f78d 13#include <linux/vmalloc.h>
00619bcc 14
a4bb1e43 15#include <asm/sections.h>
96840aa0 16#include <asm/uaccess.h>
30992c97 17
6038def0
NK
18#include "internal.h"
19
a4bb1e43
AH
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
30992c97 39/**
30992c97 40 * kstrdup - allocate space for and copy an existing string
30992c97
MM
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;
1d2c8eea 53 buf = kmalloc_track_caller(len, gfp);
30992c97
MM
54 if (buf)
55 memcpy(buf, s, len);
56 return buf;
57}
58EXPORT_SYMBOL(kstrdup);
96840aa0 59
a4bb1e43
AH
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
1e66df3e
JF
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
1a2f67b4
AD
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
1d2c8eea 113 p = kmalloc_track_caller(len, gfp);
1a2f67b4
AD
114 if (p)
115 memcpy(p, src, len);
116 return p;
117}
118EXPORT_SYMBOL(kmemdup);
119
610a77e0
LZ
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
96840aa0
DA
150/*
151 * strndup_user - duplicate an existing string from user space
96840aa0
DA
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
90d74045 168 p = memdup_user(s, length);
96840aa0 169
90d74045
JL
170 if (IS_ERR(p))
171 return p;
96840aa0
DA
172
173 p[length - 1] = '\0';
174
175 return p;
176}
177EXPORT_SYMBOL(strndup_user);
16d69265 178
e9d408e1
AV
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
6038def0
NK
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
b7643757 232/* Check if the vma is being used as a stack by this task */
65376df5 233int vma_is_stack_for_task(struct vm_area_struct *vma, struct task_struct *t)
b7643757
SP
234{
235 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
236}
237
efc1a3b1 238#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
16d69265
AM
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;
16d69265
AM
243}
244#endif
912985dc 245
45888a0c
XG
246/*
247 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
248 * back to the regular GUP.
25985edc 249 * If the architecture not support this function, simply return with no
45888a0c
XG
250 * page pinned
251 */
3b32123d 252int __weak __get_user_pages_fast(unsigned long start,
45888a0c
XG
253 int nr_pages, int write, struct page **pages)
254{
255 return 0;
256}
257EXPORT_SYMBOL_GPL(__get_user_pages_fast);
258
9de100d0
AG
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 *
9de100d0
AG
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.
d2bf6be8
NP
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.
9de100d0 282 */
3b32123d 283int __weak get_user_pages_fast(unsigned long start,
912985dc
RR
284 int nr_pages, int write, struct page **pages)
285{
cde70140 286 return get_user_pages_unlocked(start, nr_pages, write, 0, pages);
912985dc
RR
287}
288EXPORT_SYMBOL_GPL(get_user_pages_fast);
ca2b84cb 289
eb36c587
AV
290unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
291 unsigned long len, unsigned long prot,
9fbeb5ab 292 unsigned long flag, unsigned long pgoff)
eb36c587
AV
293{
294 unsigned long ret;
295 struct mm_struct *mm = current->mm;
41badc15 296 unsigned long populate;
eb36c587
AV
297
298 ret = security_mmap_file(file, prot, flag);
299 if (!ret) {
9fbeb5ab
MH
300 if (down_write_killable(&mm->mmap_sem))
301 return -EINTR;
bebeb3d6
ML
302 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
303 &populate);
eb36c587 304 up_write(&mm->mmap_sem);
41badc15
ML
305 if (populate)
306 mm_populate(ret, populate);
eb36c587
AV
307 }
308 return ret;
309}
310
311unsigned long vm_mmap(struct file *file, unsigned long addr,
312 unsigned long len, unsigned long prot,
313 unsigned long flag, unsigned long offset)
314{
315 if (unlikely(offset + PAGE_ALIGN(len) < offset))
316 return -EINVAL;
ea53cde0 317 if (unlikely(offset_in_page(offset)))
eb36c587
AV
318 return -EINVAL;
319
9fbeb5ab 320 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
eb36c587
AV
321}
322EXPORT_SYMBOL(vm_mmap);
323
39f1f78d
AV
324void kvfree(const void *addr)
325{
326 if (is_vmalloc_addr(addr))
327 vfree(addr);
328 else
329 kfree(addr);
330}
331EXPORT_SYMBOL(kvfree);
332
e39155ea
KS
333static inline void *__page_rmapping(struct page *page)
334{
335 unsigned long mapping;
336
337 mapping = (unsigned long)page->mapping;
338 mapping &= ~PAGE_MAPPING_FLAGS;
339
340 return (void *)mapping;
341}
342
343/* Neutral page->mapping pointer to address_space or anon_vma or other */
344void *page_rmapping(struct page *page)
345{
346 page = compound_head(page);
347 return __page_rmapping(page);
348}
349
1aa8aea5
AM
350/*
351 * Return true if this page is mapped into pagetables.
352 * For compound page it returns true if any subpage of compound page is mapped.
353 */
354bool page_mapped(struct page *page)
355{
356 int i;
357
358 if (likely(!PageCompound(page)))
359 return atomic_read(&page->_mapcount) >= 0;
360 page = compound_head(page);
361 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
362 return true;
363 if (PageHuge(page))
364 return false;
365 for (i = 0; i < hpage_nr_pages(page); i++) {
366 if (atomic_read(&page[i]._mapcount) >= 0)
367 return true;
368 }
369 return false;
370}
371EXPORT_SYMBOL(page_mapped);
372
e39155ea
KS
373struct anon_vma *page_anon_vma(struct page *page)
374{
375 unsigned long mapping;
376
377 page = compound_head(page);
378 mapping = (unsigned long)page->mapping;
379 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
380 return NULL;
381 return __page_rmapping(page);
382}
383
9800339b
SL
384struct address_space *page_mapping(struct page *page)
385{
1c290f64
KS
386 struct address_space *mapping;
387
388 page = compound_head(page);
9800339b 389
03e5ac2f
MP
390 /* This happens if someone calls flush_dcache_page on slab page */
391 if (unlikely(PageSlab(page)))
392 return NULL;
393
33806f06
SL
394 if (unlikely(PageSwapCache(page))) {
395 swp_entry_t entry;
396
397 entry.val = page_private(page);
e39155ea
KS
398 return swap_address_space(entry);
399 }
400
1c290f64 401 mapping = page->mapping;
bda807d4 402 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
e39155ea 403 return NULL;
bda807d4
MK
404
405 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
9800339b 406}
bda807d4 407EXPORT_SYMBOL(page_mapping);
9800339b 408
b20ce5e0
KS
409/* Slow path of page_mapcount() for compound pages */
410int __page_mapcount(struct page *page)
411{
412 int ret;
413
414 ret = atomic_read(&page->_mapcount) + 1;
dd78fedd
KS
415 /*
416 * For file THP page->_mapcount contains total number of mapping
417 * of the page: no need to look into compound_mapcount.
418 */
419 if (!PageAnon(page) && !PageHuge(page))
420 return ret;
b20ce5e0
KS
421 page = compound_head(page);
422 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
423 if (PageDoubleMap(page))
424 ret--;
425 return ret;
426}
427EXPORT_SYMBOL_GPL(__page_mapcount);
428
39a1aa8e
AR
429int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
430int sysctl_overcommit_ratio __read_mostly = 50;
431unsigned long sysctl_overcommit_kbytes __read_mostly;
432int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
433unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
434unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
435
49f0ce5f
JM
436int overcommit_ratio_handler(struct ctl_table *table, int write,
437 void __user *buffer, size_t *lenp,
438 loff_t *ppos)
439{
440 int ret;
441
442 ret = proc_dointvec(table, write, buffer, lenp, ppos);
443 if (ret == 0 && write)
444 sysctl_overcommit_kbytes = 0;
445 return ret;
446}
447
448int overcommit_kbytes_handler(struct ctl_table *table, int write,
449 void __user *buffer, size_t *lenp,
450 loff_t *ppos)
451{
452 int ret;
453
454 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
455 if (ret == 0 && write)
456 sysctl_overcommit_ratio = 0;
457 return ret;
458}
459
00619bcc
JM
460/*
461 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
462 */
463unsigned long vm_commit_limit(void)
464{
49f0ce5f
JM
465 unsigned long allowed;
466
467 if (sysctl_overcommit_kbytes)
468 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
469 else
470 allowed = ((totalram_pages - hugetlb_total_pages())
471 * sysctl_overcommit_ratio / 100);
472 allowed += total_swap_pages;
473
474 return allowed;
00619bcc
JM
475}
476
39a1aa8e
AR
477/*
478 * Make sure vm_committed_as in one cacheline and not cacheline shared with
479 * other variables. It can be updated by several CPUs frequently.
480 */
481struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
482
483/*
484 * The global memory commitment made in the system can be a metric
485 * that can be used to drive ballooning decisions when Linux is hosted
486 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
487 * balancing memory across competing virtual machines that are hosted.
488 * Several metrics drive this policy engine including the guest reported
489 * memory commitment.
490 */
491unsigned long vm_memory_committed(void)
492{
493 return percpu_counter_read_positive(&vm_committed_as);
494}
495EXPORT_SYMBOL_GPL(vm_memory_committed);
496
497/*
498 * Check that a process has enough memory to allocate a new virtual
499 * mapping. 0 means there is enough memory for the allocation to
500 * succeed and -ENOMEM implies there is not.
501 *
502 * We currently support three overcommit policies, which are set via the
503 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
504 *
505 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
506 * Additional code 2002 Jul 20 by Robert Love.
507 *
508 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
509 *
510 * Note this is a helper function intended to be used by LSMs which
511 * wish to use this logic.
512 */
513int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
514{
515 long free, allowed, reserve;
516
517 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
518 -(s64)vm_committed_as_batch * num_online_cpus(),
519 "memory commitment underflow");
520
521 vm_acct_memory(pages);
522
523 /*
524 * Sometimes we want to use more memory than we have
525 */
526 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
527 return 0;
528
529 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
530 free = global_page_state(NR_FREE_PAGES);
11fb9989 531 free += global_node_page_state(NR_FILE_PAGES);
39a1aa8e
AR
532
533 /*
534 * shmem pages shouldn't be counted as free in this
535 * case, they can't be purged, only swapped out, and
536 * that won't affect the overall amount of available
537 * memory in the system.
538 */
11fb9989 539 free -= global_node_page_state(NR_SHMEM);
39a1aa8e
AR
540
541 free += get_nr_swap_pages();
542
543 /*
544 * Any slabs which are created with the
545 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
546 * which are reclaimable, under pressure. The dentry
547 * cache and most inode caches should fall into this
548 */
549 free += global_page_state(NR_SLAB_RECLAIMABLE);
550
551 /*
552 * Leave reserved pages. The pages are not for anonymous pages.
553 */
554 if (free <= totalreserve_pages)
555 goto error;
556 else
557 free -= totalreserve_pages;
558
559 /*
560 * Reserve some for root
561 */
562 if (!cap_sys_admin)
563 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
564
565 if (free > pages)
566 return 0;
567
568 goto error;
569 }
570
571 allowed = vm_commit_limit();
572 /*
573 * Reserve some for root
574 */
575 if (!cap_sys_admin)
576 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
577
578 /*
579 * Don't let a single process grow so big a user can't recover
580 */
581 if (mm) {
582 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
583 allowed -= min_t(long, mm->total_vm / 32, reserve);
584 }
585
586 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
587 return 0;
588error:
589 vm_unacct_memory(pages);
590
591 return -ENOMEM;
592}
593
a9090253
WR
594/**
595 * get_cmdline() - copy the cmdline value to a buffer.
596 * @task: the task whose cmdline value to copy.
597 * @buffer: the buffer to copy to.
598 * @buflen: the length of the buffer. Larger cmdline values are truncated
599 * to this length.
600 * Returns the size of the cmdline field copied. Note that the copy does
601 * not guarantee an ending NULL byte.
602 */
603int get_cmdline(struct task_struct *task, char *buffer, int buflen)
604{
605 int res = 0;
606 unsigned int len;
607 struct mm_struct *mm = get_task_mm(task);
a3b609ef 608 unsigned long arg_start, arg_end, env_start, env_end;
a9090253
WR
609 if (!mm)
610 goto out;
611 if (!mm->arg_end)
612 goto out_mm; /* Shh! No looking before we're done */
613
a3b609ef
MG
614 down_read(&mm->mmap_sem);
615 arg_start = mm->arg_start;
616 arg_end = mm->arg_end;
617 env_start = mm->env_start;
618 env_end = mm->env_end;
619 up_read(&mm->mmap_sem);
620
621 len = arg_end - arg_start;
a9090253
WR
622
623 if (len > buflen)
624 len = buflen;
625
a3b609ef 626 res = access_process_vm(task, arg_start, buffer, len, 0);
a9090253
WR
627
628 /*
629 * If the nul at the end of args has been overwritten, then
630 * assume application is using setproctitle(3).
631 */
632 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
633 len = strnlen(buffer, res);
634 if (len < res) {
635 res = len;
636 } else {
a3b609ef 637 len = env_end - env_start;
a9090253
WR
638 if (len > buflen - res)
639 len = buflen - res;
a3b609ef 640 res += access_process_vm(task, env_start,
a9090253
WR
641 buffer+res, len, 0);
642 res = strnlen(buffer, res);
643 }
644 }
645out_mm:
646 mmput(mm);
647out:
648 return res;
649}