1 #include <linux/mm.h> 2 #include <linux/slab.h> 3 #include <linux/string.h> 4 #include <linux/compiler.h> 5 #include <linux/export.h> 6 #include <linux/err.h> 7 #include <linux/sched.h> 8 #include <linux/security.h> 9 #include <linux/swap.h> 10 #include <linux/swapops.h> 11 #include <linux/mman.h> 12 #include <linux/hugetlb.h> 13 #include <linux/vmalloc.h> 14 15 #include <asm/sections.h> 16 #include <asm/uaccess.h> 17 18 #include "internal.h" 19 20 static inline int is_kernel_rodata(unsigned long addr) 21 { 22 return addr >= (unsigned long)__start_rodata && 23 addr < (unsigned long)__end_rodata; 24 } 25 26 /** 27 * kfree_const - conditionally free memory 28 * @x: pointer to the memory 29 * 30 * Function calls kfree only if @x is not in .rodata section. 31 */ 32 void kfree_const(const void *x) 33 { 34 if (!is_kernel_rodata((unsigned long)x)) 35 kfree(x); 36 } 37 EXPORT_SYMBOL(kfree_const); 38 39 /** 40 * kstrdup - allocate space for and copy an existing string 41 * @s: the string to duplicate 42 * @gfp: the GFP mask used in the kmalloc() call when allocating memory 43 */ 44 char *kstrdup(const char *s, gfp_t gfp) 45 { 46 size_t len; 47 char *buf; 48 49 if (!s) 50 return NULL; 51 52 len = strlen(s) + 1; 53 buf = kmalloc_track_caller(len, gfp); 54 if (buf) 55 memcpy(buf, s, len); 56 return buf; 57 } 58 EXPORT_SYMBOL(kstrdup); 59 60 /** 61 * kstrdup_const - conditionally duplicate an existing const string 62 * @s: the string to duplicate 63 * @gfp: the GFP mask used in the kmalloc() call when allocating memory 64 * 65 * Function returns source string if it is in .rodata section otherwise it 66 * fallbacks to kstrdup. 67 * Strings allocated by kstrdup_const should be freed by kfree_const. 68 */ 69 const char *kstrdup_const(const char *s, gfp_t gfp) 70 { 71 if (is_kernel_rodata((unsigned long)s)) 72 return s; 73 74 return kstrdup(s, gfp); 75 } 76 EXPORT_SYMBOL(kstrdup_const); 77 78 /** 79 * kstrndup - allocate space for and copy an existing string 80 * @s: the string to duplicate 81 * @max: read at most @max chars from @s 82 * @gfp: the GFP mask used in the kmalloc() call when allocating memory 83 */ 84 char *kstrndup(const char *s, size_t max, gfp_t gfp) 85 { 86 size_t len; 87 char *buf; 88 89 if (!s) 90 return NULL; 91 92 len = strnlen(s, max); 93 buf = kmalloc_track_caller(len+1, gfp); 94 if (buf) { 95 memcpy(buf, s, len); 96 buf[len] = '\0'; 97 } 98 return buf; 99 } 100 EXPORT_SYMBOL(kstrndup); 101 102 /** 103 * kmemdup - duplicate region of memory 104 * 105 * @src: memory region to duplicate 106 * @len: memory region length 107 * @gfp: GFP mask to use 108 */ 109 void *kmemdup(const void *src, size_t len, gfp_t gfp) 110 { 111 void *p; 112 113 p = kmalloc_track_caller(len, gfp); 114 if (p) 115 memcpy(p, src, len); 116 return p; 117 } 118 EXPORT_SYMBOL(kmemdup); 119 120 /** 121 * memdup_user - duplicate memory region from user space 122 * 123 * @src: source address in user space 124 * @len: number of bytes to copy 125 * 126 * Returns an ERR_PTR() on failure. 127 */ 128 void *memdup_user(const void __user *src, size_t len) 129 { 130 void *p; 131 132 /* 133 * Always use GFP_KERNEL, since copy_from_user() can sleep and 134 * cause pagefault, which makes it pointless to use GFP_NOFS 135 * or GFP_ATOMIC. 136 */ 137 p = kmalloc_track_caller(len, GFP_KERNEL); 138 if (!p) 139 return ERR_PTR(-ENOMEM); 140 141 if (copy_from_user(p, src, len)) { 142 kfree(p); 143 return ERR_PTR(-EFAULT); 144 } 145 146 return p; 147 } 148 EXPORT_SYMBOL(memdup_user); 149 150 /* 151 * strndup_user - duplicate an existing string from user space 152 * @s: The string to duplicate 153 * @n: Maximum number of bytes to copy, including the trailing NUL. 154 */ 155 char *strndup_user(const char __user *s, long n) 156 { 157 char *p; 158 long length; 159 160 length = strnlen_user(s, n); 161 162 if (!length) 163 return ERR_PTR(-EFAULT); 164 165 if (length > n) 166 return ERR_PTR(-EINVAL); 167 168 p = memdup_user(s, length); 169 170 if (IS_ERR(p)) 171 return p; 172 173 p[length - 1] = '\0'; 174 175 return p; 176 } 177 EXPORT_SYMBOL(strndup_user); 178 179 /** 180 * memdup_user_nul - duplicate memory region from user space and NUL-terminate 181 * 182 * @src: source address in user space 183 * @len: number of bytes to copy 184 * 185 * Returns an ERR_PTR() on failure. 186 */ 187 void *memdup_user_nul(const void __user *src, size_t len) 188 { 189 char *p; 190 191 /* 192 * Always use GFP_KERNEL, since copy_from_user() can sleep and 193 * cause pagefault, which makes it pointless to use GFP_NOFS 194 * or GFP_ATOMIC. 195 */ 196 p = kmalloc_track_caller(len + 1, GFP_KERNEL); 197 if (!p) 198 return ERR_PTR(-ENOMEM); 199 200 if (copy_from_user(p, src, len)) { 201 kfree(p); 202 return ERR_PTR(-EFAULT); 203 } 204 p[len] = '\0'; 205 206 return p; 207 } 208 EXPORT_SYMBOL(memdup_user_nul); 209 210 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma, 211 struct vm_area_struct *prev, struct rb_node *rb_parent) 212 { 213 struct vm_area_struct *next; 214 215 vma->vm_prev = prev; 216 if (prev) { 217 next = prev->vm_next; 218 prev->vm_next = vma; 219 } else { 220 mm->mmap = vma; 221 if (rb_parent) 222 next = rb_entry(rb_parent, 223 struct vm_area_struct, vm_rb); 224 else 225 next = NULL; 226 } 227 vma->vm_next = next; 228 if (next) 229 next->vm_prev = vma; 230 } 231 232 /* Check if the vma is being used as a stack by this task */ 233 int vma_is_stack_for_task(struct vm_area_struct *vma, struct task_struct *t) 234 { 235 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t)); 236 } 237 238 #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT) 239 void arch_pick_mmap_layout(struct mm_struct *mm) 240 { 241 mm->mmap_base = TASK_UNMAPPED_BASE; 242 mm->get_unmapped_area = arch_get_unmapped_area; 243 } 244 #endif 245 246 /* 247 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall 248 * back to the regular GUP. 249 * If the architecture not support this function, simply return with no 250 * page pinned 251 */ 252 int __weak __get_user_pages_fast(unsigned long start, 253 int nr_pages, int write, struct page **pages) 254 { 255 return 0; 256 } 257 EXPORT_SYMBOL_GPL(__get_user_pages_fast); 258 259 /** 260 * get_user_pages_fast() - pin user pages in memory 261 * @start: starting user address 262 * @nr_pages: number of pages from start to pin 263 * @write: whether pages will be written to 264 * @pages: array that receives pointers to the pages pinned. 265 * Should be at least nr_pages long. 266 * 267 * Returns number of pages pinned. This may be fewer than the number 268 * requested. If nr_pages is 0 or negative, returns 0. If no pages 269 * were pinned, returns -errno. 270 * 271 * get_user_pages_fast provides equivalent functionality to get_user_pages, 272 * operating on current and current->mm, with force=0 and vma=NULL. However 273 * unlike get_user_pages, it must be called without mmap_sem held. 274 * 275 * get_user_pages_fast may take mmap_sem and page table locks, so no 276 * assumptions can be made about lack of locking. get_user_pages_fast is to be 277 * implemented in a way that is advantageous (vs get_user_pages()) when the 278 * user memory area is already faulted in and present in ptes. However if the 279 * pages have to be faulted in, it may turn out to be slightly slower so 280 * callers need to carefully consider what to use. On many architectures, 281 * get_user_pages_fast simply falls back to get_user_pages. 282 */ 283 int __weak get_user_pages_fast(unsigned long start, 284 int nr_pages, int write, struct page **pages) 285 { 286 return get_user_pages_unlocked(start, nr_pages, write, 0, pages); 287 } 288 EXPORT_SYMBOL_GPL(get_user_pages_fast); 289 290 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr, 291 unsigned long len, unsigned long prot, 292 unsigned long flag, unsigned long pgoff) 293 { 294 unsigned long ret; 295 struct mm_struct *mm = current->mm; 296 unsigned long populate; 297 298 ret = security_mmap_file(file, prot, flag); 299 if (!ret) { 300 down_write(&mm->mmap_sem); 301 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff, 302 &populate); 303 up_write(&mm->mmap_sem); 304 if (populate) 305 mm_populate(ret, populate); 306 } 307 return ret; 308 } 309 310 unsigned long vm_mmap(struct file *file, unsigned long addr, 311 unsigned long len, unsigned long prot, 312 unsigned long flag, unsigned long offset) 313 { 314 if (unlikely(offset + PAGE_ALIGN(len) < offset)) 315 return -EINVAL; 316 if (unlikely(offset_in_page(offset))) 317 return -EINVAL; 318 319 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT); 320 } 321 EXPORT_SYMBOL(vm_mmap); 322 323 void kvfree(const void *addr) 324 { 325 if (is_vmalloc_addr(addr)) 326 vfree(addr); 327 else 328 kfree(addr); 329 } 330 EXPORT_SYMBOL(kvfree); 331 332 static inline void *__page_rmapping(struct page *page) 333 { 334 unsigned long mapping; 335 336 mapping = (unsigned long)page->mapping; 337 mapping &= ~PAGE_MAPPING_FLAGS; 338 339 return (void *)mapping; 340 } 341 342 /* Neutral page->mapping pointer to address_space or anon_vma or other */ 343 void *page_rmapping(struct page *page) 344 { 345 page = compound_head(page); 346 return __page_rmapping(page); 347 } 348 349 /* 350 * Return true if this page is mapped into pagetables. 351 * For compound page it returns true if any subpage of compound page is mapped. 352 */ 353 bool page_mapped(struct page *page) 354 { 355 int i; 356 357 if (likely(!PageCompound(page))) 358 return atomic_read(&page->_mapcount) >= 0; 359 page = compound_head(page); 360 if (atomic_read(compound_mapcount_ptr(page)) >= 0) 361 return true; 362 if (PageHuge(page)) 363 return false; 364 for (i = 0; i < hpage_nr_pages(page); i++) { 365 if (atomic_read(&page[i]._mapcount) >= 0) 366 return true; 367 } 368 return false; 369 } 370 EXPORT_SYMBOL(page_mapped); 371 372 struct anon_vma *page_anon_vma(struct page *page) 373 { 374 unsigned long mapping; 375 376 page = compound_head(page); 377 mapping = (unsigned long)page->mapping; 378 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) 379 return NULL; 380 return __page_rmapping(page); 381 } 382 383 struct address_space *page_mapping(struct page *page) 384 { 385 struct address_space *mapping; 386 387 page = compound_head(page); 388 389 /* This happens if someone calls flush_dcache_page on slab page */ 390 if (unlikely(PageSlab(page))) 391 return NULL; 392 393 if (unlikely(PageSwapCache(page))) { 394 swp_entry_t entry; 395 396 entry.val = page_private(page); 397 return swap_address_space(entry); 398 } 399 400 mapping = page->mapping; 401 if ((unsigned long)mapping & PAGE_MAPPING_FLAGS) 402 return NULL; 403 return mapping; 404 } 405 406 /* Slow path of page_mapcount() for compound pages */ 407 int __page_mapcount(struct page *page) 408 { 409 int ret; 410 411 ret = atomic_read(&page->_mapcount) + 1; 412 page = compound_head(page); 413 ret += atomic_read(compound_mapcount_ptr(page)) + 1; 414 if (PageDoubleMap(page)) 415 ret--; 416 return ret; 417 } 418 EXPORT_SYMBOL_GPL(__page_mapcount); 419 420 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; 421 int sysctl_overcommit_ratio __read_mostly = 50; 422 unsigned long sysctl_overcommit_kbytes __read_mostly; 423 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; 424 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ 425 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ 426 427 int overcommit_ratio_handler(struct ctl_table *table, int write, 428 void __user *buffer, size_t *lenp, 429 loff_t *ppos) 430 { 431 int ret; 432 433 ret = proc_dointvec(table, write, buffer, lenp, ppos); 434 if (ret == 0 && write) 435 sysctl_overcommit_kbytes = 0; 436 return ret; 437 } 438 439 int overcommit_kbytes_handler(struct ctl_table *table, int write, 440 void __user *buffer, size_t *lenp, 441 loff_t *ppos) 442 { 443 int ret; 444 445 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); 446 if (ret == 0 && write) 447 sysctl_overcommit_ratio = 0; 448 return ret; 449 } 450 451 /* 452 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used 453 */ 454 unsigned long vm_commit_limit(void) 455 { 456 unsigned long allowed; 457 458 if (sysctl_overcommit_kbytes) 459 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10); 460 else 461 allowed = ((totalram_pages - hugetlb_total_pages()) 462 * sysctl_overcommit_ratio / 100); 463 allowed += total_swap_pages; 464 465 return allowed; 466 } 467 468 /* 469 * Make sure vm_committed_as in one cacheline and not cacheline shared with 470 * other variables. It can be updated by several CPUs frequently. 471 */ 472 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; 473 474 /* 475 * The global memory commitment made in the system can be a metric 476 * that can be used to drive ballooning decisions when Linux is hosted 477 * as a guest. On Hyper-V, the host implements a policy engine for dynamically 478 * balancing memory across competing virtual machines that are hosted. 479 * Several metrics drive this policy engine including the guest reported 480 * memory commitment. 481 */ 482 unsigned long vm_memory_committed(void) 483 { 484 return percpu_counter_read_positive(&vm_committed_as); 485 } 486 EXPORT_SYMBOL_GPL(vm_memory_committed); 487 488 /* 489 * Check that a process has enough memory to allocate a new virtual 490 * mapping. 0 means there is enough memory for the allocation to 491 * succeed and -ENOMEM implies there is not. 492 * 493 * We currently support three overcommit policies, which are set via the 494 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 495 * 496 * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 497 * Additional code 2002 Jul 20 by Robert Love. 498 * 499 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 500 * 501 * Note this is a helper function intended to be used by LSMs which 502 * wish to use this logic. 503 */ 504 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 505 { 506 long free, allowed, reserve; 507 508 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) < 509 -(s64)vm_committed_as_batch * num_online_cpus(), 510 "memory commitment underflow"); 511 512 vm_acct_memory(pages); 513 514 /* 515 * Sometimes we want to use more memory than we have 516 */ 517 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 518 return 0; 519 520 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 521 free = global_page_state(NR_FREE_PAGES); 522 free += global_page_state(NR_FILE_PAGES); 523 524 /* 525 * shmem pages shouldn't be counted as free in this 526 * case, they can't be purged, only swapped out, and 527 * that won't affect the overall amount of available 528 * memory in the system. 529 */ 530 free -= global_page_state(NR_SHMEM); 531 532 free += get_nr_swap_pages(); 533 534 /* 535 * Any slabs which are created with the 536 * SLAB_RECLAIM_ACCOUNT flag claim to have contents 537 * which are reclaimable, under pressure. The dentry 538 * cache and most inode caches should fall into this 539 */ 540 free += global_page_state(NR_SLAB_RECLAIMABLE); 541 542 /* 543 * Leave reserved pages. The pages are not for anonymous pages. 544 */ 545 if (free <= totalreserve_pages) 546 goto error; 547 else 548 free -= totalreserve_pages; 549 550 /* 551 * Reserve some for root 552 */ 553 if (!cap_sys_admin) 554 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 555 556 if (free > pages) 557 return 0; 558 559 goto error; 560 } 561 562 allowed = vm_commit_limit(); 563 /* 564 * Reserve some for root 565 */ 566 if (!cap_sys_admin) 567 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 568 569 /* 570 * Don't let a single process grow so big a user can't recover 571 */ 572 if (mm) { 573 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); 574 allowed -= min_t(long, mm->total_vm / 32, reserve); 575 } 576 577 if (percpu_counter_read_positive(&vm_committed_as) < allowed) 578 return 0; 579 error: 580 vm_unacct_memory(pages); 581 582 return -ENOMEM; 583 } 584 585 /** 586 * get_cmdline() - copy the cmdline value to a buffer. 587 * @task: the task whose cmdline value to copy. 588 * @buffer: the buffer to copy to. 589 * @buflen: the length of the buffer. Larger cmdline values are truncated 590 * to this length. 591 * Returns the size of the cmdline field copied. Note that the copy does 592 * not guarantee an ending NULL byte. 593 */ 594 int get_cmdline(struct task_struct *task, char *buffer, int buflen) 595 { 596 int res = 0; 597 unsigned int len; 598 struct mm_struct *mm = get_task_mm(task); 599 unsigned long arg_start, arg_end, env_start, env_end; 600 if (!mm) 601 goto out; 602 if (!mm->arg_end) 603 goto out_mm; /* Shh! No looking before we're done */ 604 605 down_read(&mm->mmap_sem); 606 arg_start = mm->arg_start; 607 arg_end = mm->arg_end; 608 env_start = mm->env_start; 609 env_end = mm->env_end; 610 up_read(&mm->mmap_sem); 611 612 len = arg_end - arg_start; 613 614 if (len > buflen) 615 len = buflen; 616 617 res = access_process_vm(task, arg_start, buffer, len, 0); 618 619 /* 620 * If the nul at the end of args has been overwritten, then 621 * assume application is using setproctitle(3). 622 */ 623 if (res > 0 && buffer[res-1] != '\0' && len < buflen) { 624 len = strnlen(buffer, res); 625 if (len < res) { 626 res = len; 627 } else { 628 len = env_end - env_start; 629 if (len > buflen - res) 630 len = buflen - res; 631 res += access_process_vm(task, env_start, 632 buffer+res, len, 0); 633 res = strnlen(buffer, res); 634 } 635 } 636 out_mm: 637 mmput(mm); 638 out: 639 return res; 640 } 641