Thomas Gleixner | 457c899 | 2019-05-19 13:08:55 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
Andrew Morton | 16d6926 | 2008-07-25 19:44:36 -0700 | [diff] [blame] | 2 | #include <linux/mm.h> |
Matt Mackall | 30992c9 | 2006-01-08 01:01:43 -0800 | [diff] [blame] | 3 | #include <linux/slab.h> |
| 4 | #include <linux/string.h> |
Gideon Israel Dsouza | 3b32123 | 2014-04-07 15:37:26 -0700 | [diff] [blame] | 5 | #include <linux/compiler.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 6 | #include <linux/export.h> |
Davi Arnaut | 96840aa | 2006-03-24 03:18:42 -0800 | [diff] [blame] | 7 | #include <linux/err.h> |
Adrian Bunk | 3b8f14b | 2008-07-26 15:22:28 -0700 | [diff] [blame] | 8 | #include <linux/sched.h> |
Ingo Molnar | 6e84f31 | 2017-02-08 18:51:29 +0100 | [diff] [blame] | 9 | #include <linux/sched/mm.h> |
Daniel Jordan | 79eb597 | 2019-07-16 16:30:54 -0700 | [diff] [blame] | 10 | #include <linux/sched/signal.h> |
Ingo Molnar | 68db0cf | 2017-02-08 18:51:37 +0100 | [diff] [blame] | 11 | #include <linux/sched/task_stack.h> |
Al Viro | eb36c58 | 2012-05-30 20:17:35 -0400 | [diff] [blame] | 12 | #include <linux/security.h> |
Shaohua Li | 9800339 | 2013-02-22 16:34:35 -0800 | [diff] [blame] | 13 | #include <linux/swap.h> |
Shaohua Li | 33806f0 | 2013-02-22 16:34:37 -0800 | [diff] [blame] | 14 | #include <linux/swapops.h> |
Jerome Marchand | 00619bc | 2013-11-12 15:08:31 -0800 | [diff] [blame] | 15 | #include <linux/mman.h> |
| 16 | #include <linux/hugetlb.h> |
Al Viro | 39f1f78 | 2014-05-06 14:02:53 -0400 | [diff] [blame] | 17 | #include <linux/vmalloc.h> |
Mike Rapoport | 897ab3e | 2017-02-24 14:58:22 -0800 | [diff] [blame] | 18 | #include <linux/userfaultfd_k.h> |
Alexandre Ghiti | 649775b | 2019-09-23 15:38:37 -0700 | [diff] [blame] | 19 | #include <linux/elf.h> |
Alexandre Ghiti | 67f3977 | 2019-09-23 15:38:47 -0700 | [diff] [blame] | 20 | #include <linux/elf-randomize.h> |
| 21 | #include <linux/personality.h> |
Alexandre Ghiti | 649775b | 2019-09-23 15:38:37 -0700 | [diff] [blame] | 22 | #include <linux/random.h> |
Alexandre Ghiti | 67f3977 | 2019-09-23 15:38:47 -0700 | [diff] [blame] | 23 | #include <linux/processor.h> |
| 24 | #include <linux/sizes.h> |
| 25 | #include <linux/compat.h> |
Jerome Marchand | 00619bc | 2013-11-12 15:08:31 -0800 | [diff] [blame] | 26 | |
Linus Torvalds | 7c0f6ba | 2016-12-24 11:46:01 -0800 | [diff] [blame] | 27 | #include <linux/uaccess.h> |
Matt Mackall | 30992c9 | 2006-01-08 01:01:43 -0800 | [diff] [blame] | 28 | |
Namhyung Kim | 6038def | 2011-05-24 17:11:22 -0700 | [diff] [blame] | 29 | #include "internal.h" |
Kuan-Ying Lee | a5543c9 | 2021-06-18 13:18:59 +0800 | [diff] [blame] | 30 | #ifndef __GENKSYMS__ |
| 31 | #include <trace/hooks/syscall_check.h> |
| 32 | #endif |
Namhyung Kim | 6038def | 2011-05-24 17:11:22 -0700 | [diff] [blame] | 33 | |
Andrzej Hajda | a4bb1e4 | 2015-02-13 14:36:24 -0800 | [diff] [blame] | 34 | /** |
| 35 | * kfree_const - conditionally free memory |
| 36 | * @x: pointer to the memory |
| 37 | * |
| 38 | * Function calls kfree only if @x is not in .rodata section. |
| 39 | */ |
| 40 | void kfree_const(const void *x) |
| 41 | { |
| 42 | if (!is_kernel_rodata((unsigned long)x)) |
| 43 | kfree(x); |
| 44 | } |
| 45 | EXPORT_SYMBOL(kfree_const); |
| 46 | |
Matt Mackall | 30992c9 | 2006-01-08 01:01:43 -0800 | [diff] [blame] | 47 | /** |
Matt Mackall | 30992c9 | 2006-01-08 01:01:43 -0800 | [diff] [blame] | 48 | * kstrdup - allocate space for and copy an existing string |
Matt Mackall | 30992c9 | 2006-01-08 01:01:43 -0800 | [diff] [blame] | 49 | * @s: the string to duplicate |
| 50 | * @gfp: the GFP mask used in the kmalloc() call when allocating memory |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 51 | * |
| 52 | * Return: newly allocated copy of @s or %NULL in case of error |
Matt Mackall | 30992c9 | 2006-01-08 01:01:43 -0800 | [diff] [blame] | 53 | */ |
| 54 | char *kstrdup(const char *s, gfp_t gfp) |
| 55 | { |
| 56 | size_t len; |
| 57 | char *buf; |
| 58 | |
| 59 | if (!s) |
| 60 | return NULL; |
| 61 | |
| 62 | len = strlen(s) + 1; |
Christoph Hellwig | 1d2c8ee | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 63 | buf = kmalloc_track_caller(len, gfp); |
Matt Mackall | 30992c9 | 2006-01-08 01:01:43 -0800 | [diff] [blame] | 64 | if (buf) |
| 65 | memcpy(buf, s, len); |
| 66 | return buf; |
| 67 | } |
| 68 | EXPORT_SYMBOL(kstrdup); |
Davi Arnaut | 96840aa | 2006-03-24 03:18:42 -0800 | [diff] [blame] | 69 | |
Alexey Dobriyan | 1a2f67b | 2006-09-30 23:27:20 -0700 | [diff] [blame] | 70 | /** |
Andrzej Hajda | a4bb1e4 | 2015-02-13 14:36:24 -0800 | [diff] [blame] | 71 | * kstrdup_const - conditionally duplicate an existing const string |
| 72 | * @s: the string to duplicate |
| 73 | * @gfp: the GFP mask used in the kmalloc() call when allocating memory |
| 74 | * |
Bartosz Golaszewski | 295a173 | 2020-10-15 20:07:39 -0700 | [diff] [blame] | 75 | * Note: Strings allocated by kstrdup_const should be freed by kfree_const and |
| 76 | * must not be passed to krealloc(). |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 77 | * |
| 78 | * Return: source string if it is in .rodata section otherwise |
| 79 | * fallback to kstrdup. |
Andrzej Hajda | a4bb1e4 | 2015-02-13 14:36:24 -0800 | [diff] [blame] | 80 | */ |
| 81 | const char *kstrdup_const(const char *s, gfp_t gfp) |
| 82 | { |
| 83 | if (is_kernel_rodata((unsigned long)s)) |
| 84 | return s; |
| 85 | |
| 86 | return kstrdup(s, gfp); |
| 87 | } |
| 88 | EXPORT_SYMBOL(kstrdup_const); |
| 89 | |
| 90 | /** |
Jeremy Fitzhardinge | 1e66df3 | 2007-07-17 18:37:02 -0700 | [diff] [blame] | 91 | * kstrndup - allocate space for and copy an existing string |
| 92 | * @s: the string to duplicate |
| 93 | * @max: read at most @max chars from @s |
| 94 | * @gfp: the GFP mask used in the kmalloc() call when allocating memory |
David Howells | f351574 | 2017-07-04 17:25:02 +0100 | [diff] [blame] | 95 | * |
| 96 | * Note: Use kmemdup_nul() instead if the size is known exactly. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 97 | * |
| 98 | * Return: newly allocated copy of @s or %NULL in case of error |
Jeremy Fitzhardinge | 1e66df3 | 2007-07-17 18:37:02 -0700 | [diff] [blame] | 99 | */ |
| 100 | char *kstrndup(const char *s, size_t max, gfp_t gfp) |
| 101 | { |
| 102 | size_t len; |
| 103 | char *buf; |
| 104 | |
| 105 | if (!s) |
| 106 | return NULL; |
| 107 | |
| 108 | len = strnlen(s, max); |
| 109 | buf = kmalloc_track_caller(len+1, gfp); |
| 110 | if (buf) { |
| 111 | memcpy(buf, s, len); |
| 112 | buf[len] = '\0'; |
| 113 | } |
| 114 | return buf; |
| 115 | } |
| 116 | EXPORT_SYMBOL(kstrndup); |
| 117 | |
| 118 | /** |
Alexey Dobriyan | 1a2f67b | 2006-09-30 23:27:20 -0700 | [diff] [blame] | 119 | * kmemdup - duplicate region of memory |
| 120 | * |
| 121 | * @src: memory region to duplicate |
| 122 | * @len: memory region length |
| 123 | * @gfp: GFP mask to use |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 124 | * |
| 125 | * Return: newly allocated copy of @src or %NULL in case of error |
Alexey Dobriyan | 1a2f67b | 2006-09-30 23:27:20 -0700 | [diff] [blame] | 126 | */ |
| 127 | void *kmemdup(const void *src, size_t len, gfp_t gfp) |
| 128 | { |
| 129 | void *p; |
| 130 | |
Christoph Hellwig | 1d2c8ee | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 131 | p = kmalloc_track_caller(len, gfp); |
Alexey Dobriyan | 1a2f67b | 2006-09-30 23:27:20 -0700 | [diff] [blame] | 132 | if (p) |
| 133 | memcpy(p, src, len); |
| 134 | return p; |
| 135 | } |
| 136 | EXPORT_SYMBOL(kmemdup); |
| 137 | |
Christoph Lameter | ef2ad80 | 2007-07-17 04:03:21 -0700 | [diff] [blame] | 138 | /** |
David Howells | f351574 | 2017-07-04 17:25:02 +0100 | [diff] [blame] | 139 | * kmemdup_nul - Create a NUL-terminated string from unterminated data |
| 140 | * @s: The data to stringify |
| 141 | * @len: The size of the data |
| 142 | * @gfp: the GFP mask used in the kmalloc() call when allocating memory |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 143 | * |
| 144 | * Return: newly allocated copy of @s with NUL-termination or %NULL in |
| 145 | * case of error |
David Howells | f351574 | 2017-07-04 17:25:02 +0100 | [diff] [blame] | 146 | */ |
| 147 | char *kmemdup_nul(const char *s, size_t len, gfp_t gfp) |
| 148 | { |
| 149 | char *buf; |
| 150 | |
| 151 | if (!s) |
| 152 | return NULL; |
| 153 | |
| 154 | buf = kmalloc_track_caller(len + 1, gfp); |
| 155 | if (buf) { |
| 156 | memcpy(buf, s, len); |
| 157 | buf[len] = '\0'; |
| 158 | } |
| 159 | return buf; |
| 160 | } |
| 161 | EXPORT_SYMBOL(kmemdup_nul); |
| 162 | |
| 163 | /** |
Li Zefan | 610a77e | 2009-03-31 15:23:16 -0700 | [diff] [blame] | 164 | * memdup_user - duplicate memory region from user space |
| 165 | * |
| 166 | * @src: source address in user space |
| 167 | * @len: number of bytes to copy |
| 168 | * |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 169 | * Return: an ERR_PTR() on failure. Result is physically |
Al Viro | 50fd2f2 | 2018-01-07 13:06:15 -0500 | [diff] [blame] | 170 | * contiguous, to be freed by kfree(). |
Li Zefan | 610a77e | 2009-03-31 15:23:16 -0700 | [diff] [blame] | 171 | */ |
| 172 | void *memdup_user(const void __user *src, size_t len) |
| 173 | { |
| 174 | void *p; |
| 175 | |
Daniel Vetter | 6c8fcc0 | 2019-02-20 22:20:42 -0800 | [diff] [blame] | 176 | p = kmalloc_track_caller(len, GFP_USER | __GFP_NOWARN); |
Li Zefan | 610a77e | 2009-03-31 15:23:16 -0700 | [diff] [blame] | 177 | if (!p) |
| 178 | return ERR_PTR(-ENOMEM); |
| 179 | |
| 180 | if (copy_from_user(p, src, len)) { |
| 181 | kfree(p); |
| 182 | return ERR_PTR(-EFAULT); |
| 183 | } |
| 184 | |
| 185 | return p; |
| 186 | } |
| 187 | EXPORT_SYMBOL(memdup_user); |
| 188 | |
Al Viro | 50fd2f2 | 2018-01-07 13:06:15 -0500 | [diff] [blame] | 189 | /** |
| 190 | * vmemdup_user - duplicate memory region from user space |
| 191 | * |
| 192 | * @src: source address in user space |
| 193 | * @len: number of bytes to copy |
| 194 | * |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 195 | * Return: an ERR_PTR() on failure. Result may be not |
Al Viro | 50fd2f2 | 2018-01-07 13:06:15 -0500 | [diff] [blame] | 196 | * physically contiguous. Use kvfree() to free. |
| 197 | */ |
| 198 | void *vmemdup_user(const void __user *src, size_t len) |
| 199 | { |
| 200 | void *p; |
| 201 | |
| 202 | p = kvmalloc(len, GFP_USER); |
| 203 | if (!p) |
| 204 | return ERR_PTR(-ENOMEM); |
| 205 | |
| 206 | if (copy_from_user(p, src, len)) { |
| 207 | kvfree(p); |
| 208 | return ERR_PTR(-EFAULT); |
| 209 | } |
| 210 | |
| 211 | return p; |
| 212 | } |
| 213 | EXPORT_SYMBOL(vmemdup_user); |
| 214 | |
Mike Rapoport | b86181f | 2018-08-23 17:00:59 -0700 | [diff] [blame] | 215 | /** |
Davi Arnaut | 96840aa | 2006-03-24 03:18:42 -0800 | [diff] [blame] | 216 | * strndup_user - duplicate an existing string from user space |
Davi Arnaut | 96840aa | 2006-03-24 03:18:42 -0800 | [diff] [blame] | 217 | * @s: The string to duplicate |
| 218 | * @n: Maximum number of bytes to copy, including the trailing NUL. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 219 | * |
Andrew Morton | e914552 | 2019-04-05 18:39:34 -0700 | [diff] [blame] | 220 | * Return: newly allocated copy of @s or an ERR_PTR() in case of error |
Davi Arnaut | 96840aa | 2006-03-24 03:18:42 -0800 | [diff] [blame] | 221 | */ |
| 222 | char *strndup_user(const char __user *s, long n) |
| 223 | { |
| 224 | char *p; |
| 225 | long length; |
| 226 | |
| 227 | length = strnlen_user(s, n); |
| 228 | |
| 229 | if (!length) |
| 230 | return ERR_PTR(-EFAULT); |
| 231 | |
| 232 | if (length > n) |
| 233 | return ERR_PTR(-EINVAL); |
| 234 | |
Julia Lawall | 90d7404 | 2010-08-09 17:18:26 -0700 | [diff] [blame] | 235 | p = memdup_user(s, length); |
Davi Arnaut | 96840aa | 2006-03-24 03:18:42 -0800 | [diff] [blame] | 236 | |
Julia Lawall | 90d7404 | 2010-08-09 17:18:26 -0700 | [diff] [blame] | 237 | if (IS_ERR(p)) |
| 238 | return p; |
Davi Arnaut | 96840aa | 2006-03-24 03:18:42 -0800 | [diff] [blame] | 239 | |
| 240 | p[length - 1] = '\0'; |
| 241 | |
| 242 | return p; |
| 243 | } |
| 244 | EXPORT_SYMBOL(strndup_user); |
Andrew Morton | 16d6926 | 2008-07-25 19:44:36 -0700 | [diff] [blame] | 245 | |
Al Viro | e9d408e | 2015-12-24 00:06:05 -0500 | [diff] [blame] | 246 | /** |
| 247 | * memdup_user_nul - duplicate memory region from user space and NUL-terminate |
| 248 | * |
| 249 | * @src: source address in user space |
| 250 | * @len: number of bytes to copy |
| 251 | * |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 252 | * Return: an ERR_PTR() on failure. |
Al Viro | e9d408e | 2015-12-24 00:06:05 -0500 | [diff] [blame] | 253 | */ |
| 254 | void *memdup_user_nul(const void __user *src, size_t len) |
| 255 | { |
| 256 | char *p; |
| 257 | |
| 258 | /* |
| 259 | * Always use GFP_KERNEL, since copy_from_user() can sleep and |
| 260 | * cause pagefault, which makes it pointless to use GFP_NOFS |
| 261 | * or GFP_ATOMIC. |
| 262 | */ |
| 263 | p = kmalloc_track_caller(len + 1, GFP_KERNEL); |
| 264 | if (!p) |
| 265 | return ERR_PTR(-ENOMEM); |
| 266 | |
| 267 | if (copy_from_user(p, src, len)) { |
| 268 | kfree(p); |
| 269 | return ERR_PTR(-EFAULT); |
| 270 | } |
| 271 | p[len] = '\0'; |
| 272 | |
| 273 | return p; |
| 274 | } |
| 275 | EXPORT_SYMBOL(memdup_user_nul); |
| 276 | |
Namhyung Kim | 6038def | 2011-05-24 17:11:22 -0700 | [diff] [blame] | 277 | void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma, |
Wei Yang | aba6dfb | 2019-11-30 17:50:53 -0800 | [diff] [blame] | 278 | struct vm_area_struct *prev) |
Namhyung Kim | 6038def | 2011-05-24 17:11:22 -0700 | [diff] [blame] | 279 | { |
| 280 | struct vm_area_struct *next; |
| 281 | |
| 282 | vma->vm_prev = prev; |
| 283 | if (prev) { |
| 284 | next = prev->vm_next; |
| 285 | prev->vm_next = vma; |
| 286 | } else { |
Wei Yang | aba6dfb | 2019-11-30 17:50:53 -0800 | [diff] [blame] | 287 | next = mm->mmap; |
Namhyung Kim | 6038def | 2011-05-24 17:11:22 -0700 | [diff] [blame] | 288 | mm->mmap = vma; |
Namhyung Kim | 6038def | 2011-05-24 17:11:22 -0700 | [diff] [blame] | 289 | } |
| 290 | vma->vm_next = next; |
| 291 | if (next) |
| 292 | next->vm_prev = vma; |
| 293 | } |
| 294 | |
Wei Yang | 1b9fc5b2 | 2019-11-30 17:50:49 -0800 | [diff] [blame] | 295 | void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma) |
| 296 | { |
| 297 | struct vm_area_struct *prev, *next; |
| 298 | |
| 299 | next = vma->vm_next; |
| 300 | prev = vma->vm_prev; |
| 301 | if (prev) |
| 302 | prev->vm_next = next; |
| 303 | else |
| 304 | mm->mmap = next; |
| 305 | if (next) |
| 306 | next->vm_prev = prev; |
| 307 | } |
| 308 | |
Siddhesh Poyarekar | b764375 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 309 | /* Check if the vma is being used as a stack by this task */ |
Andy Lutomirski | d17af50 | 2016-09-30 10:58:58 -0700 | [diff] [blame] | 310 | int vma_is_stack_for_current(struct vm_area_struct *vma) |
Siddhesh Poyarekar | b764375 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 311 | { |
Andy Lutomirski | d17af50 | 2016-09-30 10:58:58 -0700 | [diff] [blame] | 312 | struct task_struct * __maybe_unused t = current; |
| 313 | |
Siddhesh Poyarekar | b764375 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 314 | return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t)); |
| 315 | } |
| 316 | |
Alexandre Ghiti | 649775b | 2019-09-23 15:38:37 -0700 | [diff] [blame] | 317 | #ifndef STACK_RND_MASK |
| 318 | #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */ |
| 319 | #endif |
| 320 | |
| 321 | unsigned long randomize_stack_top(unsigned long stack_top) |
| 322 | { |
| 323 | unsigned long random_variable = 0; |
| 324 | |
| 325 | if (current->flags & PF_RANDOMIZE) { |
| 326 | random_variable = get_random_long(); |
| 327 | random_variable &= STACK_RND_MASK; |
| 328 | random_variable <<= PAGE_SHIFT; |
| 329 | } |
| 330 | #ifdef CONFIG_STACK_GROWSUP |
| 331 | return PAGE_ALIGN(stack_top) + random_variable; |
| 332 | #else |
| 333 | return PAGE_ALIGN(stack_top) - random_variable; |
| 334 | #endif |
| 335 | } |
| 336 | |
Jason A. Donenfeld | 552ae8e | 2022-05-14 13:59:30 +0200 | [diff] [blame] | 337 | /** |
| 338 | * randomize_page - Generate a random, page aligned address |
| 339 | * @start: The smallest acceptable address the caller will take. |
| 340 | * @range: The size of the area, starting at @start, within which the |
| 341 | * random address must fall. |
| 342 | * |
| 343 | * If @start + @range would overflow, @range is capped. |
| 344 | * |
| 345 | * NOTE: Historical use of randomize_range, which this replaces, presumed that |
| 346 | * @start was already page aligned. We now align it regardless. |
| 347 | * |
| 348 | * Return: A page aligned address within [start, start + range). On error, |
| 349 | * @start is returned. |
| 350 | */ |
| 351 | unsigned long randomize_page(unsigned long start, unsigned long range) |
| 352 | { |
| 353 | if (!PAGE_ALIGNED(start)) { |
| 354 | range -= PAGE_ALIGN(start) - start; |
| 355 | start = PAGE_ALIGN(start); |
| 356 | } |
| 357 | |
| 358 | if (start > ULONG_MAX - range) |
| 359 | range = ULONG_MAX - start; |
| 360 | |
| 361 | range >>= PAGE_SHIFT; |
| 362 | |
| 363 | if (range == 0) |
| 364 | return start; |
| 365 | |
| 366 | return start + (get_random_long() % range << PAGE_SHIFT); |
| 367 | } |
| 368 | |
Alexandre Ghiti | 67f3977 | 2019-09-23 15:38:47 -0700 | [diff] [blame] | 369 | #ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT |
Alexandre Ghiti | e7142bf | 2019-09-23 15:38:50 -0700 | [diff] [blame] | 370 | unsigned long arch_randomize_brk(struct mm_struct *mm) |
| 371 | { |
| 372 | /* Is the current task 32bit ? */ |
| 373 | if (!IS_ENABLED(CONFIG_64BIT) || is_compat_task()) |
| 374 | return randomize_page(mm->brk, SZ_32M); |
| 375 | |
| 376 | return randomize_page(mm->brk, SZ_1G); |
| 377 | } |
| 378 | |
Alexandre Ghiti | 67f3977 | 2019-09-23 15:38:47 -0700 | [diff] [blame] | 379 | unsigned long arch_mmap_rnd(void) |
| 380 | { |
| 381 | unsigned long rnd; |
| 382 | |
| 383 | #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS |
| 384 | if (is_compat_task()) |
| 385 | rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1); |
| 386 | else |
| 387 | #endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */ |
| 388 | rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1); |
| 389 | |
| 390 | return rnd << PAGE_SHIFT; |
| 391 | } |
xieliujie | d80c70d | 2021-07-03 16:59:16 +0800 | [diff] [blame] | 392 | EXPORT_SYMBOL_GPL(arch_mmap_rnd); |
Alexandre Ghiti | 67f3977 | 2019-09-23 15:38:47 -0700 | [diff] [blame] | 393 | |
| 394 | static int mmap_is_legacy(struct rlimit *rlim_stack) |
| 395 | { |
| 396 | if (current->personality & ADDR_COMPAT_LAYOUT) |
| 397 | return 1; |
| 398 | |
| 399 | if (rlim_stack->rlim_cur == RLIM_INFINITY) |
| 400 | return 1; |
| 401 | |
| 402 | return sysctl_legacy_va_layout; |
| 403 | } |
| 404 | |
| 405 | /* |
| 406 | * Leave enough space between the mmap area and the stack to honour ulimit in |
| 407 | * the face of randomisation. |
| 408 | */ |
| 409 | #define MIN_GAP (SZ_128M) |
| 410 | #define MAX_GAP (STACK_TOP / 6 * 5) |
| 411 | |
| 412 | static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack) |
| 413 | { |
| 414 | unsigned long gap = rlim_stack->rlim_cur; |
| 415 | unsigned long pad = stack_guard_gap; |
| 416 | |
| 417 | /* Account for stack randomization if necessary */ |
| 418 | if (current->flags & PF_RANDOMIZE) |
| 419 | pad += (STACK_RND_MASK << PAGE_SHIFT); |
| 420 | |
| 421 | /* Values close to RLIM_INFINITY can overflow. */ |
| 422 | if (gap + pad > gap) |
| 423 | gap += pad; |
| 424 | |
| 425 | if (gap < MIN_GAP) |
| 426 | gap = MIN_GAP; |
| 427 | else if (gap > MAX_GAP) |
| 428 | gap = MAX_GAP; |
| 429 | |
| 430 | return PAGE_ALIGN(STACK_TOP - gap - rnd); |
| 431 | } |
| 432 | |
| 433 | void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack) |
| 434 | { |
| 435 | unsigned long random_factor = 0UL; |
| 436 | |
| 437 | if (current->flags & PF_RANDOMIZE) |
| 438 | random_factor = arch_mmap_rnd(); |
| 439 | |
| 440 | if (mmap_is_legacy(rlim_stack)) { |
| 441 | mm->mmap_base = TASK_UNMAPPED_BASE + random_factor; |
| 442 | mm->get_unmapped_area = arch_get_unmapped_area; |
| 443 | } else { |
| 444 | mm->mmap_base = mmap_base(random_factor, rlim_stack); |
| 445 | mm->get_unmapped_area = arch_get_unmapped_area_topdown; |
| 446 | } |
| 447 | } |
| 448 | #elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT) |
Kees Cook | 8f2af15 | 2018-04-10 16:34:53 -0700 | [diff] [blame] | 449 | void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack) |
Andrew Morton | 16d6926 | 2008-07-25 19:44:36 -0700 | [diff] [blame] | 450 | { |
| 451 | mm->mmap_base = TASK_UNMAPPED_BASE; |
| 452 | mm->get_unmapped_area = arch_get_unmapped_area; |
Andrew Morton | 16d6926 | 2008-07-25 19:44:36 -0700 | [diff] [blame] | 453 | } |
| 454 | #endif |
Rusty Russell | 912985d | 2008-08-12 17:52:52 -0500 | [diff] [blame] | 455 | |
Daniel Jordan | 79eb597 | 2019-07-16 16:30:54 -0700 | [diff] [blame] | 456 | /** |
| 457 | * __account_locked_vm - account locked pages to an mm's locked_vm |
| 458 | * @mm: mm to account against |
| 459 | * @pages: number of pages to account |
| 460 | * @inc: %true if @pages should be considered positive, %false if not |
| 461 | * @task: task used to check RLIMIT_MEMLOCK |
| 462 | * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped |
| 463 | * |
| 464 | * Assumes @task and @mm are valid (i.e. at least one reference on each), and |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 465 | * that mmap_lock is held as writer. |
Daniel Jordan | 79eb597 | 2019-07-16 16:30:54 -0700 | [diff] [blame] | 466 | * |
| 467 | * Return: |
| 468 | * * 0 on success |
| 469 | * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded. |
| 470 | */ |
| 471 | int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc, |
| 472 | struct task_struct *task, bool bypass_rlim) |
| 473 | { |
| 474 | unsigned long locked_vm, limit; |
| 475 | int ret = 0; |
| 476 | |
Michel Lespinasse | 42fc541 | 2020-06-08 21:33:44 -0700 | [diff] [blame] | 477 | mmap_assert_write_locked(mm); |
Daniel Jordan | 79eb597 | 2019-07-16 16:30:54 -0700 | [diff] [blame] | 478 | |
| 479 | locked_vm = mm->locked_vm; |
| 480 | if (inc) { |
| 481 | if (!bypass_rlim) { |
| 482 | limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT; |
| 483 | if (locked_vm + pages > limit) |
| 484 | ret = -ENOMEM; |
| 485 | } |
| 486 | if (!ret) |
| 487 | mm->locked_vm = locked_vm + pages; |
| 488 | } else { |
| 489 | WARN_ON_ONCE(pages > locked_vm); |
| 490 | mm->locked_vm = locked_vm - pages; |
| 491 | } |
| 492 | |
| 493 | pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid, |
| 494 | (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT, |
| 495 | locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK), |
| 496 | ret ? " - exceeded" : ""); |
| 497 | |
| 498 | return ret; |
| 499 | } |
| 500 | EXPORT_SYMBOL_GPL(__account_locked_vm); |
| 501 | |
| 502 | /** |
| 503 | * account_locked_vm - account locked pages to an mm's locked_vm |
| 504 | * @mm: mm to account against, may be NULL |
| 505 | * @pages: number of pages to account |
| 506 | * @inc: %true if @pages should be considered positive, %false if not |
| 507 | * |
| 508 | * Assumes a non-NULL @mm is valid (i.e. at least one reference on it). |
| 509 | * |
| 510 | * Return: |
| 511 | * * 0 on success, or if mm is NULL |
| 512 | * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded. |
| 513 | */ |
| 514 | int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc) |
| 515 | { |
| 516 | int ret; |
| 517 | |
| 518 | if (pages == 0 || !mm) |
| 519 | return 0; |
| 520 | |
Michel Lespinasse | d8ed45c | 2020-06-08 21:33:25 -0700 | [diff] [blame] | 521 | mmap_write_lock(mm); |
Daniel Jordan | 79eb597 | 2019-07-16 16:30:54 -0700 | [diff] [blame] | 522 | ret = __account_locked_vm(mm, pages, inc, current, |
| 523 | capable(CAP_IPC_LOCK)); |
Michel Lespinasse | d8ed45c | 2020-06-08 21:33:25 -0700 | [diff] [blame] | 524 | mmap_write_unlock(mm); |
Daniel Jordan | 79eb597 | 2019-07-16 16:30:54 -0700 | [diff] [blame] | 525 | |
| 526 | return ret; |
| 527 | } |
| 528 | EXPORT_SYMBOL_GPL(account_locked_vm); |
| 529 | |
Al Viro | eb36c58 | 2012-05-30 20:17:35 -0400 | [diff] [blame] | 530 | unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr, |
| 531 | unsigned long len, unsigned long prot, |
Michal Hocko | 9fbeb5a | 2016-05-23 16:25:30 -0700 | [diff] [blame] | 532 | unsigned long flag, unsigned long pgoff) |
Al Viro | eb36c58 | 2012-05-30 20:17:35 -0400 | [diff] [blame] | 533 | { |
| 534 | unsigned long ret; |
| 535 | struct mm_struct *mm = current->mm; |
Michel Lespinasse | 41badc1 | 2013-02-22 16:32:47 -0800 | [diff] [blame] | 536 | unsigned long populate; |
Mike Rapoport | 897ab3e | 2017-02-24 14:58:22 -0800 | [diff] [blame] | 537 | LIST_HEAD(uf); |
Al Viro | eb36c58 | 2012-05-30 20:17:35 -0400 | [diff] [blame] | 538 | |
| 539 | ret = security_mmap_file(file, prot, flag); |
| 540 | if (!ret) { |
Michel Lespinasse | d8ed45c | 2020-06-08 21:33:25 -0700 | [diff] [blame] | 541 | if (mmap_write_lock_killable(mm)) |
Michal Hocko | 9fbeb5a | 2016-05-23 16:25:30 -0700 | [diff] [blame] | 542 | return -EINTR; |
Peter Collingbourne | 45e5530 | 2020-08-06 23:23:37 -0700 | [diff] [blame] | 543 | ret = do_mmap(file, addr, len, prot, flag, pgoff, &populate, |
| 544 | &uf); |
Michel Lespinasse | d8ed45c | 2020-06-08 21:33:25 -0700 | [diff] [blame] | 545 | mmap_write_unlock(mm); |
Mike Rapoport | 897ab3e | 2017-02-24 14:58:22 -0800 | [diff] [blame] | 546 | userfaultfd_unmap_complete(mm, &uf); |
Michel Lespinasse | 41badc1 | 2013-02-22 16:32:47 -0800 | [diff] [blame] | 547 | if (populate) |
| 548 | mm_populate(ret, populate); |
Al Viro | eb36c58 | 2012-05-30 20:17:35 -0400 | [diff] [blame] | 549 | } |
Kuan-Ying Lee | a5543c9 | 2021-06-18 13:18:59 +0800 | [diff] [blame] | 550 | trace_android_vh_check_mmap_file(file, prot, flag, ret); |
Al Viro | eb36c58 | 2012-05-30 20:17:35 -0400 | [diff] [blame] | 551 | return ret; |
| 552 | } |
| 553 | |
| 554 | unsigned long vm_mmap(struct file *file, unsigned long addr, |
| 555 | unsigned long len, unsigned long prot, |
| 556 | unsigned long flag, unsigned long offset) |
| 557 | { |
| 558 | if (unlikely(offset + PAGE_ALIGN(len) < offset)) |
| 559 | return -EINVAL; |
Alexander Kuleshov | ea53cde | 2015-11-05 18:46:46 -0800 | [diff] [blame] | 560 | if (unlikely(offset_in_page(offset))) |
Al Viro | eb36c58 | 2012-05-30 20:17:35 -0400 | [diff] [blame] | 561 | return -EINVAL; |
| 562 | |
Michal Hocko | 9fbeb5a | 2016-05-23 16:25:30 -0700 | [diff] [blame] | 563 | return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT); |
Al Viro | eb36c58 | 2012-05-30 20:17:35 -0400 | [diff] [blame] | 564 | } |
| 565 | EXPORT_SYMBOL(vm_mmap); |
| 566 | |
Michal Hocko | a7c3e90 | 2017-05-08 15:57:09 -0700 | [diff] [blame] | 567 | /** |
| 568 | * kvmalloc_node - attempt to allocate physically contiguous memory, but upon |
| 569 | * failure, fall back to non-contiguous (vmalloc) allocation. |
| 570 | * @size: size of the request. |
| 571 | * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL. |
| 572 | * @node: numa node to allocate from |
| 573 | * |
| 574 | * Uses kmalloc to get the memory but if the allocation fails then falls back |
| 575 | * to the vmalloc allocator. Use kvfree for freeing the memory. |
| 576 | * |
Michal Hocko | cc965a2 | 2017-07-12 14:36:52 -0700 | [diff] [blame] | 577 | * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported. |
| 578 | * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is |
| 579 | * preferable to the vmalloc fallback, due to visible performance drawbacks. |
Michal Hocko | a7c3e90 | 2017-05-08 15:57:09 -0700 | [diff] [blame] | 580 | * |
Michal Hocko | ce91f6e | 2018-06-07 17:09:40 -0700 | [diff] [blame] | 581 | * Please note that any use of gfp flags outside of GFP_KERNEL is careful to not |
| 582 | * fall back to vmalloc. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 583 | * |
| 584 | * Return: pointer to the allocated memory of %NULL in case of failure |
Michal Hocko | a7c3e90 | 2017-05-08 15:57:09 -0700 | [diff] [blame] | 585 | */ |
| 586 | void *kvmalloc_node(size_t size, gfp_t flags, int node) |
| 587 | { |
| 588 | gfp_t kmalloc_flags = flags; |
| 589 | void *ret; |
| 590 | |
| 591 | /* |
| 592 | * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables) |
| 593 | * so the given set of flags has to be compatible. |
| 594 | */ |
Michal Hocko | ce91f6e | 2018-06-07 17:09:40 -0700 | [diff] [blame] | 595 | if ((flags & GFP_KERNEL) != GFP_KERNEL) |
| 596 | return kmalloc_node(size, flags, node); |
Michal Hocko | a7c3e90 | 2017-05-08 15:57:09 -0700 | [diff] [blame] | 597 | |
| 598 | /* |
Michal Hocko | 4f4f2ba | 2017-06-02 14:46:19 -0700 | [diff] [blame] | 599 | * We want to attempt a large physically contiguous block first because |
| 600 | * it is less likely to fragment multiple larger blocks and therefore |
| 601 | * contribute to a long term fragmentation less than vmalloc fallback. |
| 602 | * However make sure that larger requests are not too disruptive - no |
| 603 | * OOM killer and no allocation failure warnings as we have a fallback. |
Michal Hocko | a7c3e90 | 2017-05-08 15:57:09 -0700 | [diff] [blame] | 604 | */ |
Michal Hocko | 6c5ab65 | 2017-05-08 15:57:15 -0700 | [diff] [blame] | 605 | if (size > PAGE_SIZE) { |
| 606 | kmalloc_flags |= __GFP_NOWARN; |
| 607 | |
Michal Hocko | cc965a2 | 2017-07-12 14:36:52 -0700 | [diff] [blame] | 608 | if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL)) |
Michal Hocko | 6c5ab65 | 2017-05-08 15:57:15 -0700 | [diff] [blame] | 609 | kmalloc_flags |= __GFP_NORETRY; |
| 610 | } |
Michal Hocko | a7c3e90 | 2017-05-08 15:57:09 -0700 | [diff] [blame] | 611 | |
| 612 | ret = kmalloc_node(size, kmalloc_flags, node); |
| 613 | |
| 614 | /* |
| 615 | * It doesn't really make sense to fallback to vmalloc for sub page |
| 616 | * requests |
| 617 | */ |
| 618 | if (ret || size <= PAGE_SIZE) |
| 619 | return ret; |
| 620 | |
Linus Torvalds | 57a269a | 2021-07-14 09:45:49 -0700 | [diff] [blame] | 621 | /* Don't even allow crazy sizes */ |
Daniel Borkmann | e93f2be | 2022-03-04 15:26:32 +0100 | [diff] [blame] | 622 | if (unlikely(size > INT_MAX)) { |
| 623 | WARN_ON_ONCE(!(flags & __GFP_NOWARN)); |
Linus Torvalds | 57a269a | 2021-07-14 09:45:49 -0700 | [diff] [blame] | 624 | return NULL; |
Daniel Borkmann | e93f2be | 2022-03-04 15:26:32 +0100 | [diff] [blame] | 625 | } |
Linus Torvalds | 57a269a | 2021-07-14 09:45:49 -0700 | [diff] [blame] | 626 | |
Christoph Hellwig | 2b90594 | 2020-06-01 21:51:53 -0700 | [diff] [blame] | 627 | return __vmalloc_node(size, 1, flags, node, |
Michal Hocko | 8594a21 | 2017-05-12 15:46:41 -0700 | [diff] [blame] | 628 | __builtin_return_address(0)); |
Michal Hocko | a7c3e90 | 2017-05-08 15:57:09 -0700 | [diff] [blame] | 629 | } |
| 630 | EXPORT_SYMBOL(kvmalloc_node); |
| 631 | |
Mike Rapoport | ff4dc77 | 2018-08-23 17:01:02 -0700 | [diff] [blame] | 632 | /** |
Andrew Morton | 04b8e94 | 2018-09-04 15:45:55 -0700 | [diff] [blame] | 633 | * kvfree() - Free memory. |
| 634 | * @addr: Pointer to allocated memory. |
Mike Rapoport | ff4dc77 | 2018-08-23 17:01:02 -0700 | [diff] [blame] | 635 | * |
Andrew Morton | 04b8e94 | 2018-09-04 15:45:55 -0700 | [diff] [blame] | 636 | * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc(). |
| 637 | * It is slightly more efficient to use kfree() or vfree() if you are certain |
| 638 | * that you know which one to use. |
| 639 | * |
Andrey Ryabinin | 52414d3 | 2018-10-26 15:07:00 -0700 | [diff] [blame] | 640 | * Context: Either preemptible task context or not-NMI interrupt. |
Mike Rapoport | ff4dc77 | 2018-08-23 17:01:02 -0700 | [diff] [blame] | 641 | */ |
Al Viro | 39f1f78 | 2014-05-06 14:02:53 -0400 | [diff] [blame] | 642 | void kvfree(const void *addr) |
| 643 | { |
| 644 | if (is_vmalloc_addr(addr)) |
| 645 | vfree(addr); |
| 646 | else |
| 647 | kfree(addr); |
| 648 | } |
| 649 | EXPORT_SYMBOL(kvfree); |
| 650 | |
Waiman Long | d4eaa28 | 2020-06-04 16:48:21 -0700 | [diff] [blame] | 651 | /** |
| 652 | * kvfree_sensitive - Free a data object containing sensitive information. |
| 653 | * @addr: address of the data object to be freed. |
| 654 | * @len: length of the data object. |
| 655 | * |
| 656 | * Use the special memzero_explicit() function to clear the content of a |
| 657 | * kvmalloc'ed object containing sensitive data to make sure that the |
| 658 | * compiler won't optimize out the data clearing. |
| 659 | */ |
| 660 | void kvfree_sensitive(const void *addr, size_t len) |
| 661 | { |
| 662 | if (likely(!ZERO_OR_NULL_PTR(addr))) { |
| 663 | memzero_explicit((void *)addr, len); |
| 664 | kvfree(addr); |
| 665 | } |
| 666 | } |
| 667 | EXPORT_SYMBOL(kvfree_sensitive); |
| 668 | |
Dave Chinner | f5f3e54 | 2022-08-10 16:15:50 +0200 | [diff] [blame] | 669 | void *kvrealloc(const void *p, size_t oldsize, size_t newsize, gfp_t flags) |
| 670 | { |
| 671 | void *newp; |
| 672 | |
| 673 | if (oldsize >= newsize) |
| 674 | return (void *)p; |
| 675 | newp = kvmalloc(newsize, flags); |
| 676 | if (!newp) |
| 677 | return NULL; |
| 678 | memcpy(newp, p, oldsize); |
| 679 | kvfree(p); |
| 680 | return newp; |
| 681 | } |
| 682 | EXPORT_SYMBOL(kvrealloc); |
| 683 | |
Kirill A. Shutemov | e39155e | 2015-04-15 16:14:53 -0700 | [diff] [blame] | 684 | static inline void *__page_rmapping(struct page *page) |
| 685 | { |
| 686 | unsigned long mapping; |
| 687 | |
| 688 | mapping = (unsigned long)page->mapping; |
| 689 | mapping &= ~PAGE_MAPPING_FLAGS; |
| 690 | |
| 691 | return (void *)mapping; |
| 692 | } |
| 693 | |
| 694 | /* Neutral page->mapping pointer to address_space or anon_vma or other */ |
| 695 | void *page_rmapping(struct page *page) |
| 696 | { |
| 697 | page = compound_head(page); |
| 698 | return __page_rmapping(page); |
| 699 | } |
| 700 | |
Andrew Morton | 1aa8aea | 2016-05-19 17:12:00 -0700 | [diff] [blame] | 701 | /* |
| 702 | * Return true if this page is mapped into pagetables. |
| 703 | * For compound page it returns true if any subpage of compound page is mapped. |
| 704 | */ |
| 705 | bool page_mapped(struct page *page) |
| 706 | { |
| 707 | int i; |
| 708 | |
| 709 | if (likely(!PageCompound(page))) |
| 710 | return atomic_read(&page->_mapcount) >= 0; |
| 711 | page = compound_head(page); |
| 712 | if (atomic_read(compound_mapcount_ptr(page)) >= 0) |
| 713 | return true; |
| 714 | if (PageHuge(page)) |
| 715 | return false; |
Matthew Wilcox (Oracle) | d8c6546 | 2019-09-23 15:34:30 -0700 | [diff] [blame] | 716 | for (i = 0; i < compound_nr(page); i++) { |
Andrew Morton | 1aa8aea | 2016-05-19 17:12:00 -0700 | [diff] [blame] | 717 | if (atomic_read(&page[i]._mapcount) >= 0) |
| 718 | return true; |
| 719 | } |
| 720 | return false; |
| 721 | } |
| 722 | EXPORT_SYMBOL(page_mapped); |
| 723 | |
Kirill A. Shutemov | e39155e | 2015-04-15 16:14:53 -0700 | [diff] [blame] | 724 | struct anon_vma *page_anon_vma(struct page *page) |
| 725 | { |
| 726 | unsigned long mapping; |
| 727 | |
| 728 | page = compound_head(page); |
| 729 | mapping = (unsigned long)page->mapping; |
| 730 | if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
| 731 | return NULL; |
| 732 | return __page_rmapping(page); |
| 733 | } |
| 734 | |
Shaohua Li | 9800339 | 2013-02-22 16:34:35 -0800 | [diff] [blame] | 735 | struct address_space *page_mapping(struct page *page) |
| 736 | { |
Kirill A. Shutemov | 1c290f6 | 2016-01-15 16:52:07 -0800 | [diff] [blame] | 737 | struct address_space *mapping; |
| 738 | |
| 739 | page = compound_head(page); |
Shaohua Li | 9800339 | 2013-02-22 16:34:35 -0800 | [diff] [blame] | 740 | |
Mikulas Patocka | 03e5ac2 | 2014-01-14 17:56:40 -0800 | [diff] [blame] | 741 | /* This happens if someone calls flush_dcache_page on slab page */ |
| 742 | if (unlikely(PageSlab(page))) |
| 743 | return NULL; |
| 744 | |
Shaohua Li | 33806f0 | 2013-02-22 16:34:37 -0800 | [diff] [blame] | 745 | if (unlikely(PageSwapCache(page))) { |
| 746 | swp_entry_t entry; |
| 747 | |
| 748 | entry.val = page_private(page); |
Kirill A. Shutemov | e39155e | 2015-04-15 16:14:53 -0700 | [diff] [blame] | 749 | return swap_address_space(entry); |
| 750 | } |
| 751 | |
Kirill A. Shutemov | 1c290f6 | 2016-01-15 16:52:07 -0800 | [diff] [blame] | 752 | mapping = page->mapping; |
Minchan Kim | bda807d | 2016-07-26 15:23:05 -0700 | [diff] [blame] | 753 | if ((unsigned long)mapping & PAGE_MAPPING_ANON) |
Kirill A. Shutemov | e39155e | 2015-04-15 16:14:53 -0700 | [diff] [blame] | 754 | return NULL; |
Minchan Kim | bda807d | 2016-07-26 15:23:05 -0700 | [diff] [blame] | 755 | |
| 756 | return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS); |
Shaohua Li | 9800339 | 2013-02-22 16:34:35 -0800 | [diff] [blame] | 757 | } |
Minchan Kim | bda807d | 2016-07-26 15:23:05 -0700 | [diff] [blame] | 758 | EXPORT_SYMBOL(page_mapping); |
Shaohua Li | 9800339 | 2013-02-22 16:34:35 -0800 | [diff] [blame] | 759 | |
Huang Ying | cb9f753 | 2018-04-05 16:24:39 -0700 | [diff] [blame] | 760 | /* |
| 761 | * For file cache pages, return the address_space, otherwise return NULL |
| 762 | */ |
| 763 | struct address_space *page_mapping_file(struct page *page) |
| 764 | { |
| 765 | if (unlikely(PageSwapCache(page))) |
| 766 | return NULL; |
| 767 | return page_mapping(page); |
| 768 | } |
| 769 | |
Kirill A. Shutemov | b20ce5e | 2016-01-15 16:54:37 -0800 | [diff] [blame] | 770 | /* Slow path of page_mapcount() for compound pages */ |
| 771 | int __page_mapcount(struct page *page) |
| 772 | { |
| 773 | int ret; |
| 774 | |
| 775 | ret = atomic_read(&page->_mapcount) + 1; |
Kirill A. Shutemov | dd78fed | 2016-07-26 15:25:26 -0700 | [diff] [blame] | 776 | /* |
| 777 | * For file THP page->_mapcount contains total number of mapping |
| 778 | * of the page: no need to look into compound_mapcount. |
| 779 | */ |
| 780 | if (!PageAnon(page) && !PageHuge(page)) |
| 781 | return ret; |
Kirill A. Shutemov | b20ce5e | 2016-01-15 16:54:37 -0800 | [diff] [blame] | 782 | page = compound_head(page); |
| 783 | ret += atomic_read(compound_mapcount_ptr(page)) + 1; |
| 784 | if (PageDoubleMap(page)) |
| 785 | ret--; |
| 786 | return ret; |
| 787 | } |
| 788 | EXPORT_SYMBOL_GPL(__page_mapcount); |
| 789 | |
Andrey Ryabinin | 39a1aa8 | 2016-03-17 14:18:50 -0700 | [diff] [blame] | 790 | int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; |
| 791 | int sysctl_overcommit_ratio __read_mostly = 50; |
| 792 | unsigned long sysctl_overcommit_kbytes __read_mostly; |
| 793 | int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; |
| 794 | unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ |
| 795 | unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ |
| 796 | |
Christoph Hellwig | 3292739 | 2020-04-24 08:43:38 +0200 | [diff] [blame] | 797 | int overcommit_ratio_handler(struct ctl_table *table, int write, void *buffer, |
| 798 | size_t *lenp, loff_t *ppos) |
Jerome Marchand | 49f0ce5 | 2014-01-21 15:49:14 -0800 | [diff] [blame] | 799 | { |
| 800 | int ret; |
| 801 | |
| 802 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
| 803 | if (ret == 0 && write) |
| 804 | sysctl_overcommit_kbytes = 0; |
| 805 | return ret; |
| 806 | } |
| 807 | |
Feng Tang | 56f3547 | 2020-08-06 23:23:15 -0700 | [diff] [blame] | 808 | static void sync_overcommit_as(struct work_struct *dummy) |
| 809 | { |
| 810 | percpu_counter_sync(&vm_committed_as); |
| 811 | } |
| 812 | |
| 813 | int overcommit_policy_handler(struct ctl_table *table, int write, void *buffer, |
| 814 | size_t *lenp, loff_t *ppos) |
| 815 | { |
| 816 | struct ctl_table t; |
Chen Jun | 8382972 | 2021-09-24 15:44:06 -0700 | [diff] [blame] | 817 | int new_policy = -1; |
Feng Tang | 56f3547 | 2020-08-06 23:23:15 -0700 | [diff] [blame] | 818 | int ret; |
| 819 | |
| 820 | /* |
| 821 | * The deviation of sync_overcommit_as could be big with loose policy |
| 822 | * like OVERCOMMIT_ALWAYS/OVERCOMMIT_GUESS. When changing policy to |
| 823 | * strict OVERCOMMIT_NEVER, we need to reduce the deviation to comply |
| 824 | * with the strict "NEVER", and to avoid possible race condtion (even |
| 825 | * though user usually won't too frequently do the switching to policy |
| 826 | * OVERCOMMIT_NEVER), the switch is done in the following order: |
| 827 | * 1. changing the batch |
| 828 | * 2. sync percpu count on each CPU |
| 829 | * 3. switch the policy |
| 830 | */ |
| 831 | if (write) { |
| 832 | t = *table; |
| 833 | t.data = &new_policy; |
| 834 | ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); |
Chen Jun | 8382972 | 2021-09-24 15:44:06 -0700 | [diff] [blame] | 835 | if (ret || new_policy == -1) |
Feng Tang | 56f3547 | 2020-08-06 23:23:15 -0700 | [diff] [blame] | 836 | return ret; |
| 837 | |
| 838 | mm_compute_batch(new_policy); |
| 839 | if (new_policy == OVERCOMMIT_NEVER) |
| 840 | schedule_on_each_cpu(sync_overcommit_as); |
| 841 | sysctl_overcommit_memory = new_policy; |
| 842 | } else { |
| 843 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
| 844 | } |
| 845 | |
| 846 | return ret; |
| 847 | } |
| 848 | |
Christoph Hellwig | 3292739 | 2020-04-24 08:43:38 +0200 | [diff] [blame] | 849 | int overcommit_kbytes_handler(struct ctl_table *table, int write, void *buffer, |
| 850 | size_t *lenp, loff_t *ppos) |
Jerome Marchand | 49f0ce5 | 2014-01-21 15:49:14 -0800 | [diff] [blame] | 851 | { |
| 852 | int ret; |
| 853 | |
| 854 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
| 855 | if (ret == 0 && write) |
| 856 | sysctl_overcommit_ratio = 0; |
| 857 | return ret; |
| 858 | } |
| 859 | |
Jerome Marchand | 00619bc | 2013-11-12 15:08:31 -0800 | [diff] [blame] | 860 | /* |
| 861 | * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used |
| 862 | */ |
| 863 | unsigned long vm_commit_limit(void) |
| 864 | { |
Jerome Marchand | 49f0ce5 | 2014-01-21 15:49:14 -0800 | [diff] [blame] | 865 | unsigned long allowed; |
| 866 | |
| 867 | if (sysctl_overcommit_kbytes) |
| 868 | allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10); |
| 869 | else |
Arun KS | ca79b0c | 2018-12-28 00:34:29 -0800 | [diff] [blame] | 870 | allowed = ((totalram_pages() - hugetlb_total_pages()) |
Jerome Marchand | 49f0ce5 | 2014-01-21 15:49:14 -0800 | [diff] [blame] | 871 | * sysctl_overcommit_ratio / 100); |
| 872 | allowed += total_swap_pages; |
| 873 | |
| 874 | return allowed; |
Jerome Marchand | 00619bc | 2013-11-12 15:08:31 -0800 | [diff] [blame] | 875 | } |
| 876 | |
Andrey Ryabinin | 39a1aa8 | 2016-03-17 14:18:50 -0700 | [diff] [blame] | 877 | /* |
| 878 | * Make sure vm_committed_as in one cacheline and not cacheline shared with |
| 879 | * other variables. It can be updated by several CPUs frequently. |
| 880 | */ |
| 881 | struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; |
| 882 | |
| 883 | /* |
| 884 | * The global memory commitment made in the system can be a metric |
| 885 | * that can be used to drive ballooning decisions when Linux is hosted |
| 886 | * as a guest. On Hyper-V, the host implements a policy engine for dynamically |
| 887 | * balancing memory across competing virtual machines that are hosted. |
| 888 | * Several metrics drive this policy engine including the guest reported |
| 889 | * memory commitment. |
Feng Tang | 4e2ee51 | 2020-08-06 23:23:07 -0700 | [diff] [blame] | 890 | * |
| 891 | * The time cost of this is very low for small platforms, and for big |
| 892 | * platform like a 2S/36C/72T Skylake server, in worst case where |
| 893 | * vm_committed_as's spinlock is under severe contention, the time cost |
| 894 | * could be about 30~40 microseconds. |
Andrey Ryabinin | 39a1aa8 | 2016-03-17 14:18:50 -0700 | [diff] [blame] | 895 | */ |
| 896 | unsigned long vm_memory_committed(void) |
| 897 | { |
Feng Tang | 4e2ee51 | 2020-08-06 23:23:07 -0700 | [diff] [blame] | 898 | return percpu_counter_sum_positive(&vm_committed_as); |
Andrey Ryabinin | 39a1aa8 | 2016-03-17 14:18:50 -0700 | [diff] [blame] | 899 | } |
| 900 | EXPORT_SYMBOL_GPL(vm_memory_committed); |
| 901 | |
| 902 | /* |
| 903 | * Check that a process has enough memory to allocate a new virtual |
| 904 | * mapping. 0 means there is enough memory for the allocation to |
| 905 | * succeed and -ENOMEM implies there is not. |
| 906 | * |
| 907 | * We currently support three overcommit policies, which are set via the |
Mike Rapoport | ad56b73 | 2018-03-21 21:22:47 +0200 | [diff] [blame] | 908 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting.rst |
Andrey Ryabinin | 39a1aa8 | 2016-03-17 14:18:50 -0700 | [diff] [blame] | 909 | * |
| 910 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. |
| 911 | * Additional code 2002 Jul 20 by Robert Love. |
| 912 | * |
| 913 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. |
| 914 | * |
| 915 | * Note this is a helper function intended to be used by LSMs which |
| 916 | * wish to use this logic. |
| 917 | */ |
| 918 | int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) |
| 919 | { |
Johannes Weiner | 8c7829b | 2019-05-13 17:21:50 -0700 | [diff] [blame] | 920 | long allowed; |
Andrey Ryabinin | 39a1aa8 | 2016-03-17 14:18:50 -0700 | [diff] [blame] | 921 | |
Andrey Ryabinin | 39a1aa8 | 2016-03-17 14:18:50 -0700 | [diff] [blame] | 922 | vm_acct_memory(pages); |
| 923 | |
| 924 | /* |
| 925 | * Sometimes we want to use more memory than we have |
| 926 | */ |
| 927 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) |
| 928 | return 0; |
| 929 | |
| 930 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { |
Johannes Weiner | 8c7829b | 2019-05-13 17:21:50 -0700 | [diff] [blame] | 931 | if (pages > totalram_pages() + total_swap_pages) |
Andrey Ryabinin | 39a1aa8 | 2016-03-17 14:18:50 -0700 | [diff] [blame] | 932 | goto error; |
Johannes Weiner | 8c7829b | 2019-05-13 17:21:50 -0700 | [diff] [blame] | 933 | return 0; |
Andrey Ryabinin | 39a1aa8 | 2016-03-17 14:18:50 -0700 | [diff] [blame] | 934 | } |
| 935 | |
| 936 | allowed = vm_commit_limit(); |
| 937 | /* |
| 938 | * Reserve some for root |
| 939 | */ |
| 940 | if (!cap_sys_admin) |
| 941 | allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); |
| 942 | |
| 943 | /* |
| 944 | * Don't let a single process grow so big a user can't recover |
| 945 | */ |
| 946 | if (mm) { |
Johannes Weiner | 8c7829b | 2019-05-13 17:21:50 -0700 | [diff] [blame] | 947 | long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); |
| 948 | |
Andrey Ryabinin | 39a1aa8 | 2016-03-17 14:18:50 -0700 | [diff] [blame] | 949 | allowed -= min_t(long, mm->total_vm / 32, reserve); |
| 950 | } |
| 951 | |
| 952 | if (percpu_counter_read_positive(&vm_committed_as) < allowed) |
| 953 | return 0; |
| 954 | error: |
| 955 | vm_unacct_memory(pages); |
| 956 | |
| 957 | return -ENOMEM; |
| 958 | } |
| 959 | |
William Roberts | a909025 | 2014-02-11 10:11:59 -0800 | [diff] [blame] | 960 | /** |
| 961 | * get_cmdline() - copy the cmdline value to a buffer. |
| 962 | * @task: the task whose cmdline value to copy. |
| 963 | * @buffer: the buffer to copy to. |
| 964 | * @buflen: the length of the buffer. Larger cmdline values are truncated |
| 965 | * to this length. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 966 | * |
| 967 | * Return: the size of the cmdline field copied. Note that the copy does |
William Roberts | a909025 | 2014-02-11 10:11:59 -0800 | [diff] [blame] | 968 | * not guarantee an ending NULL byte. |
| 969 | */ |
| 970 | int get_cmdline(struct task_struct *task, char *buffer, int buflen) |
| 971 | { |
| 972 | int res = 0; |
| 973 | unsigned int len; |
| 974 | struct mm_struct *mm = get_task_mm(task); |
Mateusz Guzik | a3b609e | 2016-01-20 15:01:05 -0800 | [diff] [blame] | 975 | unsigned long arg_start, arg_end, env_start, env_end; |
William Roberts | a909025 | 2014-02-11 10:11:59 -0800 | [diff] [blame] | 976 | if (!mm) |
| 977 | goto out; |
| 978 | if (!mm->arg_end) |
| 979 | goto out_mm; /* Shh! No looking before we're done */ |
| 980 | |
Michal Koutný | bc81426 | 2019-05-31 22:30:19 -0700 | [diff] [blame] | 981 | spin_lock(&mm->arg_lock); |
Mateusz Guzik | a3b609e | 2016-01-20 15:01:05 -0800 | [diff] [blame] | 982 | arg_start = mm->arg_start; |
| 983 | arg_end = mm->arg_end; |
| 984 | env_start = mm->env_start; |
| 985 | env_end = mm->env_end; |
Michal Koutný | bc81426 | 2019-05-31 22:30:19 -0700 | [diff] [blame] | 986 | spin_unlock(&mm->arg_lock); |
Mateusz Guzik | a3b609e | 2016-01-20 15:01:05 -0800 | [diff] [blame] | 987 | |
| 988 | len = arg_end - arg_start; |
William Roberts | a909025 | 2014-02-11 10:11:59 -0800 | [diff] [blame] | 989 | |
| 990 | if (len > buflen) |
| 991 | len = buflen; |
| 992 | |
Lorenzo Stoakes | f307ab6 | 2016-10-13 01:20:20 +0100 | [diff] [blame] | 993 | res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE); |
William Roberts | a909025 | 2014-02-11 10:11:59 -0800 | [diff] [blame] | 994 | |
| 995 | /* |
| 996 | * If the nul at the end of args has been overwritten, then |
| 997 | * assume application is using setproctitle(3). |
| 998 | */ |
| 999 | if (res > 0 && buffer[res-1] != '\0' && len < buflen) { |
| 1000 | len = strnlen(buffer, res); |
| 1001 | if (len < res) { |
| 1002 | res = len; |
| 1003 | } else { |
Mateusz Guzik | a3b609e | 2016-01-20 15:01:05 -0800 | [diff] [blame] | 1004 | len = env_end - env_start; |
William Roberts | a909025 | 2014-02-11 10:11:59 -0800 | [diff] [blame] | 1005 | if (len > buflen - res) |
| 1006 | len = buflen - res; |
Mateusz Guzik | a3b609e | 2016-01-20 15:01:05 -0800 | [diff] [blame] | 1007 | res += access_process_vm(task, env_start, |
Lorenzo Stoakes | f307ab6 | 2016-10-13 01:20:20 +0100 | [diff] [blame] | 1008 | buffer+res, len, |
| 1009 | FOLL_FORCE); |
William Roberts | a909025 | 2014-02-11 10:11:59 -0800 | [diff] [blame] | 1010 | res = strnlen(buffer, res); |
| 1011 | } |
| 1012 | } |
| 1013 | out_mm: |
| 1014 | mmput(mm); |
| 1015 | out: |
| 1016 | return res; |
| 1017 | } |
Song Liu | 010c164 | 2019-09-23 15:38:19 -0700 | [diff] [blame] | 1018 | |
Catalin Marinas | 4d1a8a2 | 2019-11-27 09:53:44 +0000 | [diff] [blame] | 1019 | int __weak memcmp_pages(struct page *page1, struct page *page2) |
Song Liu | 010c164 | 2019-09-23 15:38:19 -0700 | [diff] [blame] | 1020 | { |
| 1021 | char *addr1, *addr2; |
| 1022 | int ret; |
| 1023 | |
| 1024 | addr1 = kmap_atomic(page1); |
| 1025 | addr2 = kmap_atomic(page2); |
| 1026 | ret = memcmp(addr1, addr2, PAGE_SIZE); |
| 1027 | kunmap_atomic(addr2); |
| 1028 | kunmap_atomic(addr1); |
| 1029 | return ret; |
| 1030 | } |