memdup_user(): switch to GFP_USER
[linux-2.6-block.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>
6e84f315 8#include <linux/sched/mm.h>
68db0cf1 9#include <linux/sched/task_stack.h>
eb36c587 10#include <linux/security.h>
9800339b 11#include <linux/swap.h>
33806f06 12#include <linux/swapops.h>
00619bcc
JM
13#include <linux/mman.h>
14#include <linux/hugetlb.h>
39f1f78d 15#include <linux/vmalloc.h>
897ab3e0 16#include <linux/userfaultfd_k.h>
00619bcc 17
a4bb1e43 18#include <asm/sections.h>
7c0f6ba6 19#include <linux/uaccess.h>
30992c97 20
6038def0
NK
21#include "internal.h"
22
a4bb1e43
AH
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
30992c97 42/**
30992c97 43 * kstrdup - allocate space for and copy an existing string
30992c97
MM
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;
1d2c8eea 56 buf = kmalloc_track_caller(len, gfp);
30992c97
MM
57 if (buf)
58 memcpy(buf, s, len);
59 return buf;
60}
61EXPORT_SYMBOL(kstrdup);
96840aa0 62
a4bb1e43
AH
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
1e66df3e
JF
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
f3515741
DH
86 *
87 * Note: Use kmemdup_nul() instead if the size is known exactly.
1e66df3e
JF
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
1a2f67b4
AD
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
1d2c8eea 118 p = kmalloc_track_caller(len, gfp);
1a2f67b4
AD
119 if (p)
120 memcpy(p, src, len);
121 return p;
122}
123EXPORT_SYMBOL(kmemdup);
124
f3515741
DH
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
610a77e0
LZ
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.
154 */
155void *memdup_user(const void __user *src, size_t len)
156{
157 void *p;
158
6c2c97a2 159 p = kmalloc_track_caller(len, GFP_USER);
610a77e0
LZ
160 if (!p)
161 return ERR_PTR(-ENOMEM);
162
163 if (copy_from_user(p, src, len)) {
164 kfree(p);
165 return ERR_PTR(-EFAULT);
166 }
167
168 return p;
169}
170EXPORT_SYMBOL(memdup_user);
171
96840aa0
DA
172/*
173 * strndup_user - duplicate an existing string from user space
96840aa0
DA
174 * @s: The string to duplicate
175 * @n: Maximum number of bytes to copy, including the trailing NUL.
176 */
177char *strndup_user(const char __user *s, long n)
178{
179 char *p;
180 long length;
181
182 length = strnlen_user(s, n);
183
184 if (!length)
185 return ERR_PTR(-EFAULT);
186
187 if (length > n)
188 return ERR_PTR(-EINVAL);
189
90d74045 190 p = memdup_user(s, length);
96840aa0 191
90d74045
JL
192 if (IS_ERR(p))
193 return p;
96840aa0
DA
194
195 p[length - 1] = '\0';
196
197 return p;
198}
199EXPORT_SYMBOL(strndup_user);
16d69265 200
e9d408e1
AV
201/**
202 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
203 *
204 * @src: source address in user space
205 * @len: number of bytes to copy
206 *
207 * Returns an ERR_PTR() on failure.
208 */
209void *memdup_user_nul(const void __user *src, size_t len)
210{
211 char *p;
212
213 /*
214 * Always use GFP_KERNEL, since copy_from_user() can sleep and
215 * cause pagefault, which makes it pointless to use GFP_NOFS
216 * or GFP_ATOMIC.
217 */
218 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
219 if (!p)
220 return ERR_PTR(-ENOMEM);
221
222 if (copy_from_user(p, src, len)) {
223 kfree(p);
224 return ERR_PTR(-EFAULT);
225 }
226 p[len] = '\0';
227
228 return p;
229}
230EXPORT_SYMBOL(memdup_user_nul);
231
6038def0
NK
232void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
233 struct vm_area_struct *prev, struct rb_node *rb_parent)
234{
235 struct vm_area_struct *next;
236
237 vma->vm_prev = prev;
238 if (prev) {
239 next = prev->vm_next;
240 prev->vm_next = vma;
241 } else {
242 mm->mmap = vma;
243 if (rb_parent)
244 next = rb_entry(rb_parent,
245 struct vm_area_struct, vm_rb);
246 else
247 next = NULL;
248 }
249 vma->vm_next = next;
250 if (next)
251 next->vm_prev = vma;
252}
253
b7643757 254/* Check if the vma is being used as a stack by this task */
d17af505 255int vma_is_stack_for_current(struct vm_area_struct *vma)
b7643757 256{
d17af505
AL
257 struct task_struct * __maybe_unused t = current;
258
b7643757
SP
259 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
260}
261
efc1a3b1 262#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
16d69265
AM
263void arch_pick_mmap_layout(struct mm_struct *mm)
264{
265 mm->mmap_base = TASK_UNMAPPED_BASE;
266 mm->get_unmapped_area = arch_get_unmapped_area;
16d69265
AM
267}
268#endif
912985dc 269
45888a0c
XG
270/*
271 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
272 * back to the regular GUP.
25985edc 273 * If the architecture not support this function, simply return with no
45888a0c
XG
274 * page pinned
275 */
3b32123d 276int __weak __get_user_pages_fast(unsigned long start,
45888a0c
XG
277 int nr_pages, int write, struct page **pages)
278{
279 return 0;
280}
281EXPORT_SYMBOL_GPL(__get_user_pages_fast);
282
9de100d0
AG
283/**
284 * get_user_pages_fast() - pin user pages in memory
285 * @start: starting user address
286 * @nr_pages: number of pages from start to pin
287 * @write: whether pages will be written to
288 * @pages: array that receives pointers to the pages pinned.
289 * Should be at least nr_pages long.
290 *
9de100d0
AG
291 * Returns number of pages pinned. This may be fewer than the number
292 * requested. If nr_pages is 0 or negative, returns 0. If no pages
293 * were pinned, returns -errno.
d2bf6be8
NP
294 *
295 * get_user_pages_fast provides equivalent functionality to get_user_pages,
296 * operating on current and current->mm, with force=0 and vma=NULL. However
297 * unlike get_user_pages, it must be called without mmap_sem held.
298 *
299 * get_user_pages_fast may take mmap_sem and page table locks, so no
300 * assumptions can be made about lack of locking. get_user_pages_fast is to be
301 * implemented in a way that is advantageous (vs get_user_pages()) when the
302 * user memory area is already faulted in and present in ptes. However if the
303 * pages have to be faulted in, it may turn out to be slightly slower so
304 * callers need to carefully consider what to use. On many architectures,
305 * get_user_pages_fast simply falls back to get_user_pages.
9de100d0 306 */
3b32123d 307int __weak get_user_pages_fast(unsigned long start,
912985dc
RR
308 int nr_pages, int write, struct page **pages)
309{
c164154f
LS
310 return get_user_pages_unlocked(start, nr_pages, pages,
311 write ? FOLL_WRITE : 0);
912985dc
RR
312}
313EXPORT_SYMBOL_GPL(get_user_pages_fast);
ca2b84cb 314
eb36c587
AV
315unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
316 unsigned long len, unsigned long prot,
9fbeb5ab 317 unsigned long flag, unsigned long pgoff)
eb36c587
AV
318{
319 unsigned long ret;
320 struct mm_struct *mm = current->mm;
41badc15 321 unsigned long populate;
897ab3e0 322 LIST_HEAD(uf);
eb36c587
AV
323
324 ret = security_mmap_file(file, prot, flag);
325 if (!ret) {
9fbeb5ab
MH
326 if (down_write_killable(&mm->mmap_sem))
327 return -EINTR;
bebeb3d6 328 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
897ab3e0 329 &populate, &uf);
eb36c587 330 up_write(&mm->mmap_sem);
897ab3e0 331 userfaultfd_unmap_complete(mm, &uf);
41badc15
ML
332 if (populate)
333 mm_populate(ret, populate);
eb36c587
AV
334 }
335 return ret;
336}
337
338unsigned long vm_mmap(struct file *file, unsigned long addr,
339 unsigned long len, unsigned long prot,
340 unsigned long flag, unsigned long offset)
341{
342 if (unlikely(offset + PAGE_ALIGN(len) < offset))
343 return -EINVAL;
ea53cde0 344 if (unlikely(offset_in_page(offset)))
eb36c587
AV
345 return -EINVAL;
346
9fbeb5ab 347 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
eb36c587
AV
348}
349EXPORT_SYMBOL(vm_mmap);
350
a7c3e901
MH
351/**
352 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
353 * failure, fall back to non-contiguous (vmalloc) allocation.
354 * @size: size of the request.
355 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
356 * @node: numa node to allocate from
357 *
358 * Uses kmalloc to get the memory but if the allocation fails then falls back
359 * to the vmalloc allocator. Use kvfree for freeing the memory.
360 *
cc965a29
MH
361 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
362 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
363 * preferable to the vmalloc fallback, due to visible performance drawbacks.
a7c3e901
MH
364 *
365 * Any use of gfp flags outside of GFP_KERNEL should be consulted with mm people.
366 */
367void *kvmalloc_node(size_t size, gfp_t flags, int node)
368{
369 gfp_t kmalloc_flags = flags;
370 void *ret;
371
372 /*
373 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
374 * so the given set of flags has to be compatible.
375 */
376 WARN_ON_ONCE((flags & GFP_KERNEL) != GFP_KERNEL);
377
378 /*
4f4f2ba9
MH
379 * We want to attempt a large physically contiguous block first because
380 * it is less likely to fragment multiple larger blocks and therefore
381 * contribute to a long term fragmentation less than vmalloc fallback.
382 * However make sure that larger requests are not too disruptive - no
383 * OOM killer and no allocation failure warnings as we have a fallback.
a7c3e901 384 */
6c5ab651
MH
385 if (size > PAGE_SIZE) {
386 kmalloc_flags |= __GFP_NOWARN;
387
cc965a29 388 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
6c5ab651
MH
389 kmalloc_flags |= __GFP_NORETRY;
390 }
a7c3e901
MH
391
392 ret = kmalloc_node(size, kmalloc_flags, node);
393
394 /*
395 * It doesn't really make sense to fallback to vmalloc for sub page
396 * requests
397 */
398 if (ret || size <= PAGE_SIZE)
399 return ret;
400
8594a21c
MH
401 return __vmalloc_node_flags_caller(size, node, flags,
402 __builtin_return_address(0));
a7c3e901
MH
403}
404EXPORT_SYMBOL(kvmalloc_node);
405
39f1f78d
AV
406void kvfree(const void *addr)
407{
408 if (is_vmalloc_addr(addr))
409 vfree(addr);
410 else
411 kfree(addr);
412}
413EXPORT_SYMBOL(kvfree);
414
e39155ea
KS
415static inline void *__page_rmapping(struct page *page)
416{
417 unsigned long mapping;
418
419 mapping = (unsigned long)page->mapping;
420 mapping &= ~PAGE_MAPPING_FLAGS;
421
422 return (void *)mapping;
423}
424
425/* Neutral page->mapping pointer to address_space or anon_vma or other */
426void *page_rmapping(struct page *page)
427{
428 page = compound_head(page);
429 return __page_rmapping(page);
430}
431
1aa8aea5
AM
432/*
433 * Return true if this page is mapped into pagetables.
434 * For compound page it returns true if any subpage of compound page is mapped.
435 */
436bool page_mapped(struct page *page)
437{
438 int i;
439
440 if (likely(!PageCompound(page)))
441 return atomic_read(&page->_mapcount) >= 0;
442 page = compound_head(page);
443 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
444 return true;
445 if (PageHuge(page))
446 return false;
447 for (i = 0; i < hpage_nr_pages(page); i++) {
448 if (atomic_read(&page[i]._mapcount) >= 0)
449 return true;
450 }
451 return false;
452}
453EXPORT_SYMBOL(page_mapped);
454
e39155ea
KS
455struct anon_vma *page_anon_vma(struct page *page)
456{
457 unsigned long mapping;
458
459 page = compound_head(page);
460 mapping = (unsigned long)page->mapping;
461 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
462 return NULL;
463 return __page_rmapping(page);
464}
465
9800339b
SL
466struct address_space *page_mapping(struct page *page)
467{
1c290f64
KS
468 struct address_space *mapping;
469
470 page = compound_head(page);
9800339b 471
03e5ac2f
MP
472 /* This happens if someone calls flush_dcache_page on slab page */
473 if (unlikely(PageSlab(page)))
474 return NULL;
475
33806f06
SL
476 if (unlikely(PageSwapCache(page))) {
477 swp_entry_t entry;
478
479 entry.val = page_private(page);
e39155ea
KS
480 return swap_address_space(entry);
481 }
482
1c290f64 483 mapping = page->mapping;
bda807d4 484 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
e39155ea 485 return NULL;
bda807d4
MK
486
487 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
9800339b 488}
bda807d4 489EXPORT_SYMBOL(page_mapping);
9800339b 490
b20ce5e0
KS
491/* Slow path of page_mapcount() for compound pages */
492int __page_mapcount(struct page *page)
493{
494 int ret;
495
496 ret = atomic_read(&page->_mapcount) + 1;
dd78fedd
KS
497 /*
498 * For file THP page->_mapcount contains total number of mapping
499 * of the page: no need to look into compound_mapcount.
500 */
501 if (!PageAnon(page) && !PageHuge(page))
502 return ret;
b20ce5e0
KS
503 page = compound_head(page);
504 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
505 if (PageDoubleMap(page))
506 ret--;
507 return ret;
508}
509EXPORT_SYMBOL_GPL(__page_mapcount);
510
39a1aa8e
AR
511int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
512int sysctl_overcommit_ratio __read_mostly = 50;
513unsigned long sysctl_overcommit_kbytes __read_mostly;
514int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
515unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
516unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
517
49f0ce5f
JM
518int overcommit_ratio_handler(struct ctl_table *table, int write,
519 void __user *buffer, size_t *lenp,
520 loff_t *ppos)
521{
522 int ret;
523
524 ret = proc_dointvec(table, write, buffer, lenp, ppos);
525 if (ret == 0 && write)
526 sysctl_overcommit_kbytes = 0;
527 return ret;
528}
529
530int overcommit_kbytes_handler(struct ctl_table *table, int write,
531 void __user *buffer, size_t *lenp,
532 loff_t *ppos)
533{
534 int ret;
535
536 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
537 if (ret == 0 && write)
538 sysctl_overcommit_ratio = 0;
539 return ret;
540}
541
00619bcc
JM
542/*
543 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
544 */
545unsigned long vm_commit_limit(void)
546{
49f0ce5f
JM
547 unsigned long allowed;
548
549 if (sysctl_overcommit_kbytes)
550 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
551 else
552 allowed = ((totalram_pages - hugetlb_total_pages())
553 * sysctl_overcommit_ratio / 100);
554 allowed += total_swap_pages;
555
556 return allowed;
00619bcc
JM
557}
558
39a1aa8e
AR
559/*
560 * Make sure vm_committed_as in one cacheline and not cacheline shared with
561 * other variables. It can be updated by several CPUs frequently.
562 */
563struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
564
565/*
566 * The global memory commitment made in the system can be a metric
567 * that can be used to drive ballooning decisions when Linux is hosted
568 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
569 * balancing memory across competing virtual machines that are hosted.
570 * Several metrics drive this policy engine including the guest reported
571 * memory commitment.
572 */
573unsigned long vm_memory_committed(void)
574{
575 return percpu_counter_read_positive(&vm_committed_as);
576}
577EXPORT_SYMBOL_GPL(vm_memory_committed);
578
579/*
580 * Check that a process has enough memory to allocate a new virtual
581 * mapping. 0 means there is enough memory for the allocation to
582 * succeed and -ENOMEM implies there is not.
583 *
584 * We currently support three overcommit policies, which are set via the
585 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
586 *
587 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
588 * Additional code 2002 Jul 20 by Robert Love.
589 *
590 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
591 *
592 * Note this is a helper function intended to be used by LSMs which
593 * wish to use this logic.
594 */
595int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
596{
597 long free, allowed, reserve;
598
599 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
600 -(s64)vm_committed_as_batch * num_online_cpus(),
601 "memory commitment underflow");
602
603 vm_acct_memory(pages);
604
605 /*
606 * Sometimes we want to use more memory than we have
607 */
608 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
609 return 0;
610
611 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
c41f012a 612 free = global_zone_page_state(NR_FREE_PAGES);
11fb9989 613 free += global_node_page_state(NR_FILE_PAGES);
39a1aa8e
AR
614
615 /*
616 * shmem pages shouldn't be counted as free in this
617 * case, they can't be purged, only swapped out, and
618 * that won't affect the overall amount of available
619 * memory in the system.
620 */
11fb9989 621 free -= global_node_page_state(NR_SHMEM);
39a1aa8e
AR
622
623 free += get_nr_swap_pages();
624
625 /*
626 * Any slabs which are created with the
627 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
628 * which are reclaimable, under pressure. The dentry
629 * cache and most inode caches should fall into this
630 */
d507e2eb 631 free += global_node_page_state(NR_SLAB_RECLAIMABLE);
39a1aa8e
AR
632
633 /*
634 * Leave reserved pages. The pages are not for anonymous pages.
635 */
636 if (free <= totalreserve_pages)
637 goto error;
638 else
639 free -= totalreserve_pages;
640
641 /*
642 * Reserve some for root
643 */
644 if (!cap_sys_admin)
645 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
646
647 if (free > pages)
648 return 0;
649
650 goto error;
651 }
652
653 allowed = vm_commit_limit();
654 /*
655 * Reserve some for root
656 */
657 if (!cap_sys_admin)
658 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
659
660 /*
661 * Don't let a single process grow so big a user can't recover
662 */
663 if (mm) {
664 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
665 allowed -= min_t(long, mm->total_vm / 32, reserve);
666 }
667
668 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
669 return 0;
670error:
671 vm_unacct_memory(pages);
672
673 return -ENOMEM;
674}
675
a9090253
WR
676/**
677 * get_cmdline() - copy the cmdline value to a buffer.
678 * @task: the task whose cmdline value to copy.
679 * @buffer: the buffer to copy to.
680 * @buflen: the length of the buffer. Larger cmdline values are truncated
681 * to this length.
682 * Returns the size of the cmdline field copied. Note that the copy does
683 * not guarantee an ending NULL byte.
684 */
685int get_cmdline(struct task_struct *task, char *buffer, int buflen)
686{
687 int res = 0;
688 unsigned int len;
689 struct mm_struct *mm = get_task_mm(task);
a3b609ef 690 unsigned long arg_start, arg_end, env_start, env_end;
a9090253
WR
691 if (!mm)
692 goto out;
693 if (!mm->arg_end)
694 goto out_mm; /* Shh! No looking before we're done */
695
a3b609ef
MG
696 down_read(&mm->mmap_sem);
697 arg_start = mm->arg_start;
698 arg_end = mm->arg_end;
699 env_start = mm->env_start;
700 env_end = mm->env_end;
701 up_read(&mm->mmap_sem);
702
703 len = arg_end - arg_start;
a9090253
WR
704
705 if (len > buflen)
706 len = buflen;
707
f307ab6d 708 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
a9090253
WR
709
710 /*
711 * If the nul at the end of args has been overwritten, then
712 * assume application is using setproctitle(3).
713 */
714 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
715 len = strnlen(buffer, res);
716 if (len < res) {
717 res = len;
718 } else {
a3b609ef 719 len = env_end - env_start;
a9090253
WR
720 if (len > buflen - res)
721 len = buflen - res;
a3b609ef 722 res += access_process_vm(task, env_start,
f307ab6d
LS
723 buffer+res, len,
724 FOLL_FORCE);
a9090253
WR
725 res = strnlen(buffer, res);
726 }
727 }
728out_mm:
729 mmput(mm);
730out:
731 return res;
732}