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/uaccess.h>
16
17#include "internal.h"
18
19#define CREATE_TRACE_POINTS
20#include <trace/events/kmem.h>
21
22/**
23 * kstrdup - allocate space for and copy an existing string
24 * @s: the string to duplicate
25 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
26 */
27char *kstrdup(const char *s, gfp_t gfp)
28{
29 size_t len;
30 char *buf;
31
32 if (!s)
33 return NULL;
34
35 len = strlen(s) + 1;
36 buf = kmalloc_track_caller(len, gfp);
37 if (buf)
38 memcpy(buf, s, len);
39 return buf;
40}
41EXPORT_SYMBOL(kstrdup);
42
43/**
44 * kstrndup - allocate space for and copy an existing string
45 * @s: the string to duplicate
46 * @max: read at most @max chars from @s
47 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
48 */
49char *kstrndup(const char *s, size_t max, gfp_t gfp)
50{
51 size_t len;
52 char *buf;
53
54 if (!s)
55 return NULL;
56
57 len = strnlen(s, max);
58 buf = kmalloc_track_caller(len+1, gfp);
59 if (buf) {
60 memcpy(buf, s, len);
61 buf[len] = '\0';
62 }
63 return buf;
64}
65EXPORT_SYMBOL(kstrndup);
66
67/**
68 * kmemdup - duplicate region of memory
69 *
70 * @src: memory region to duplicate
71 * @len: memory region length
72 * @gfp: GFP mask to use
73 */
74void *kmemdup(const void *src, size_t len, gfp_t gfp)
75{
76 void *p;
77
78 p = kmalloc_track_caller(len, gfp);
79 if (p)
80 memcpy(p, src, len);
81 return p;
82}
83EXPORT_SYMBOL(kmemdup);
84
85/**
86 * memdup_user - duplicate memory region from user space
87 *
88 * @src: source address in user space
89 * @len: number of bytes to copy
90 *
91 * Returns an ERR_PTR() on failure.
92 */
93void *memdup_user(const void __user *src, size_t len)
94{
95 void *p;
96
97 /*
98 * Always use GFP_KERNEL, since copy_from_user() can sleep and
99 * cause pagefault, which makes it pointless to use GFP_NOFS
100 * or GFP_ATOMIC.
101 */
102 p = kmalloc_track_caller(len, GFP_KERNEL);
103 if (!p)
104 return ERR_PTR(-ENOMEM);
105
106 if (copy_from_user(p, src, len)) {
107 kfree(p);
108 return ERR_PTR(-EFAULT);
109 }
110
111 return p;
112}
113EXPORT_SYMBOL(memdup_user);
114
115static __always_inline void *__do_krealloc(const void *p, size_t new_size,
116 gfp_t flags)
117{
118 void *ret;
119 size_t ks = 0;
120
121 if (p)
122 ks = ksize(p);
123
124 if (ks >= new_size)
125 return (void *)p;
126
127 ret = kmalloc_track_caller(new_size, flags);
128 if (ret && p)
129 memcpy(ret, p, ks);
130
131 return ret;
132}
133
134/**
135 * __krealloc - like krealloc() but don't free @p.
136 * @p: object to reallocate memory for.
137 * @new_size: how many bytes of memory are required.
138 * @flags: the type of memory to allocate.
139 *
140 * This function is like krealloc() except it never frees the originally
141 * allocated buffer. Use this if you don't want to free the buffer immediately
142 * like, for example, with RCU.
143 */
144void *__krealloc(const void *p, size_t new_size, gfp_t flags)
145{
146 if (unlikely(!new_size))
147 return ZERO_SIZE_PTR;
148
149 return __do_krealloc(p, new_size, flags);
150
151}
152EXPORT_SYMBOL(__krealloc);
153
154/**
155 * krealloc - reallocate memory. The contents will remain unchanged.
156 * @p: object to reallocate memory for.
157 * @new_size: how many bytes of memory are required.
158 * @flags: the type of memory to allocate.
159 *
160 * The contents of the object pointed to are preserved up to the
161 * lesser of the new and old sizes. If @p is %NULL, krealloc()
162 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
163 * %NULL pointer, the object pointed to is freed.
164 */
165void *krealloc(const void *p, size_t new_size, gfp_t flags)
166{
167 void *ret;
168
169 if (unlikely(!new_size)) {
170 kfree(p);
171 return ZERO_SIZE_PTR;
172 }
173
174 ret = __do_krealloc(p, new_size, flags);
175 if (ret && p != ret)
176 kfree(p);
177
178 return ret;
179}
180EXPORT_SYMBOL(krealloc);
181
182/**
183 * kzfree - like kfree but zero memory
184 * @p: object to free memory of
185 *
186 * The memory of the object @p points to is zeroed before freed.
187 * If @p is %NULL, kzfree() does nothing.
188 *
189 * Note: this function zeroes the whole allocated buffer which can be a good
190 * deal bigger than the requested buffer size passed to kmalloc(). So be
191 * careful when using this function in performance sensitive code.
192 */
193void kzfree(const void *p)
194{
195 size_t ks;
196 void *mem = (void *)p;
197
198 if (unlikely(ZERO_OR_NULL_PTR(mem)))
199 return;
200 ks = ksize(mem);
201 memset(mem, 0, ks);
202 kfree(mem);
203}
204EXPORT_SYMBOL(kzfree);
205
206/*
207 * strndup_user - duplicate an existing string from user space
208 * @s: The string to duplicate
209 * @n: Maximum number of bytes to copy, including the trailing NUL.
210 */
211char *strndup_user(const char __user *s, long n)
212{
213 char *p;
214 long length;
215
216 length = strnlen_user(s, n);
217
218 if (!length)
219 return ERR_PTR(-EFAULT);
220
221 if (length > n)
222 return ERR_PTR(-EINVAL);
223
224 p = memdup_user(s, length);
225
226 if (IS_ERR(p))
227 return p;
228
229 p[length - 1] = '\0';
230
231 return p;
232}
233EXPORT_SYMBOL(strndup_user);
234
235void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
236 struct vm_area_struct *prev, struct rb_node *rb_parent)
237{
238 struct vm_area_struct *next;
239
240 vma->vm_prev = prev;
241 if (prev) {
242 next = prev->vm_next;
243 prev->vm_next = vma;
244 } else {
245 mm->mmap = vma;
246 if (rb_parent)
247 next = rb_entry(rb_parent,
248 struct vm_area_struct, vm_rb);
249 else
250 next = NULL;
251 }
252 vma->vm_next = next;
253 if (next)
254 next->vm_prev = vma;
255}
256
257/* Check if the vma is being used as a stack by this task */
258static int vm_is_stack_for_task(struct task_struct *t,
259 struct vm_area_struct *vma)
260{
261 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
262}
263
264/*
265 * Check if the vma is being used as a stack.
266 * If is_group is non-zero, check in the entire thread group or else
267 * just check in the current task. Returns the pid of the task that
268 * the vma is stack for.
269 */
270pid_t vm_is_stack(struct task_struct *task,
271 struct vm_area_struct *vma, int in_group)
272{
273 pid_t ret = 0;
274
275 if (vm_is_stack_for_task(task, vma))
276 return task->pid;
277
278 if (in_group) {
279 struct task_struct *t;
280 rcu_read_lock();
281 if (!pid_alive(task))
282 goto done;
283
284 t = task;
285 do {
286 if (vm_is_stack_for_task(t, vma)) {
287 ret = t->pid;
288 goto done;
289 }
290 } while_each_thread(task, t);
291done:
292 rcu_read_unlock();
293 }
294
295 return ret;
296}
297
298#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
299void arch_pick_mmap_layout(struct mm_struct *mm)
300{
301 mm->mmap_base = TASK_UNMAPPED_BASE;
302 mm->get_unmapped_area = arch_get_unmapped_area;
303}
304#endif
305
306/*
307 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
308 * back to the regular GUP.
309 * If the architecture not support this function, simply return with no
310 * page pinned
311 */
312int __weak __get_user_pages_fast(unsigned long start,
313 int nr_pages, int write, struct page **pages)
314{
315 return 0;
316}
317EXPORT_SYMBOL_GPL(__get_user_pages_fast);
318
319/**
320 * get_user_pages_fast() - pin user pages in memory
321 * @start: starting user address
322 * @nr_pages: number of pages from start to pin
323 * @write: whether pages will be written to
324 * @pages: array that receives pointers to the pages pinned.
325 * Should be at least nr_pages long.
326 *
327 * Returns number of pages pinned. This may be fewer than the number
328 * requested. If nr_pages is 0 or negative, returns 0. If no pages
329 * were pinned, returns -errno.
330 *
331 * get_user_pages_fast provides equivalent functionality to get_user_pages,
332 * operating on current and current->mm, with force=0 and vma=NULL. However
333 * unlike get_user_pages, it must be called without mmap_sem held.
334 *
335 * get_user_pages_fast may take mmap_sem and page table locks, so no
336 * assumptions can be made about lack of locking. get_user_pages_fast is to be
337 * implemented in a way that is advantageous (vs get_user_pages()) when the
338 * user memory area is already faulted in and present in ptes. However if the
339 * pages have to be faulted in, it may turn out to be slightly slower so
340 * callers need to carefully consider what to use. On many architectures,
341 * get_user_pages_fast simply falls back to get_user_pages.
342 */
343int __weak get_user_pages_fast(unsigned long start,
344 int nr_pages, int write, struct page **pages)
345{
346 struct mm_struct *mm = current->mm;
347 int ret;
348
349 down_read(&mm->mmap_sem);
350 ret = get_user_pages(current, mm, start, nr_pages,
351 write, 0, pages, NULL);
352 up_read(&mm->mmap_sem);
353
354 return ret;
355}
356EXPORT_SYMBOL_GPL(get_user_pages_fast);
357
358unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
359 unsigned long len, unsigned long prot,
360 unsigned long flag, unsigned long pgoff)
361{
362 unsigned long ret;
363 struct mm_struct *mm = current->mm;
364 unsigned long populate;
365
366 ret = security_mmap_file(file, prot, flag);
367 if (!ret) {
368 down_write(&mm->mmap_sem);
369 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
370 &populate);
371 up_write(&mm->mmap_sem);
372 if (populate)
373 mm_populate(ret, populate);
374 }
375 return ret;
376}
377
378unsigned long vm_mmap(struct file *file, unsigned long addr,
379 unsigned long len, unsigned long prot,
380 unsigned long flag, unsigned long offset)
381{
382 if (unlikely(offset + PAGE_ALIGN(len) < offset))
383 return -EINVAL;
384 if (unlikely(offset & ~PAGE_MASK))
385 return -EINVAL;
386
387 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
388}
389EXPORT_SYMBOL(vm_mmap);
390
391void kvfree(const void *addr)
392{
393 if (is_vmalloc_addr(addr))
394 vfree(addr);
395 else
396 kfree(addr);
397}
398EXPORT_SYMBOL(kvfree);
399
400struct address_space *page_mapping(struct page *page)
401{
402 struct address_space *mapping = page->mapping;
403
404 /* This happens if someone calls flush_dcache_page on slab page */
405 if (unlikely(PageSlab(page)))
406 return NULL;
407
408 if (unlikely(PageSwapCache(page))) {
409 swp_entry_t entry;
410
411 entry.val = page_private(page);
412 mapping = swap_address_space(entry);
413 } else if ((unsigned long)mapping & PAGE_MAPPING_ANON)
414 mapping = NULL;
415 return mapping;
416}
417
418int overcommit_ratio_handler(struct ctl_table *table, int write,
419 void __user *buffer, size_t *lenp,
420 loff_t *ppos)
421{
422 int ret;
423
424 ret = proc_dointvec(table, write, buffer, lenp, ppos);
425 if (ret == 0 && write)
426 sysctl_overcommit_kbytes = 0;
427 return ret;
428}
429
430int overcommit_kbytes_handler(struct ctl_table *table, int write,
431 void __user *buffer, size_t *lenp,
432 loff_t *ppos)
433{
434 int ret;
435
436 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
437 if (ret == 0 && write)
438 sysctl_overcommit_ratio = 0;
439 return ret;
440}
441
442/*
443 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
444 */
445unsigned long vm_commit_limit(void)
446{
447 unsigned long allowed;
448
449 if (sysctl_overcommit_kbytes)
450 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
451 else
452 allowed = ((totalram_pages - hugetlb_total_pages())
453 * sysctl_overcommit_ratio / 100);
454 allowed += total_swap_pages;
455
456 return allowed;
457}
458
459/**
460 * get_cmdline() - copy the cmdline value to a buffer.
461 * @task: the task whose cmdline value to copy.
462 * @buffer: the buffer to copy to.
463 * @buflen: the length of the buffer. Larger cmdline values are truncated
464 * to this length.
465 * Returns the size of the cmdline field copied. Note that the copy does
466 * not guarantee an ending NULL byte.
467 */
468int get_cmdline(struct task_struct *task, char *buffer, int buflen)
469{
470 int res = 0;
471 unsigned int len;
472 struct mm_struct *mm = get_task_mm(task);
473 if (!mm)
474 goto out;
475 if (!mm->arg_end)
476 goto out_mm; /* Shh! No looking before we're done */
477
478 len = mm->arg_end - mm->arg_start;
479
480 if (len > buflen)
481 len = buflen;
482
483 res = access_process_vm(task, mm->arg_start, buffer, len, 0);
484
485 /*
486 * If the nul at the end of args has been overwritten, then
487 * assume application is using setproctitle(3).
488 */
489 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
490 len = strnlen(buffer, res);
491 if (len < res) {
492 res = len;
493 } else {
494 len = mm->env_end - mm->env_start;
495 if (len > buflen - res)
496 len = buflen - res;
497 res += access_process_vm(task, mm->env_start,
498 buffer+res, len, 0);
499 res = strnlen(buffer, res);
500 }
501 }
502out_mm:
503 mmput(mm);
504out:
505 return res;
506}
507
508/* Tracepoints definitions. */
509EXPORT_TRACEPOINT_SYMBOL(kmalloc);
510EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
511EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
512EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
513EXPORT_TRACEPOINT_SYMBOL(kfree);
514EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
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}