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 23 static 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 */ 35 void kfree_const(const void *x) 36 { 37 if (!is_kernel_rodata((unsigned long)x)) 38 kfree(x); 39 } 40 EXPORT_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 */ 47 char *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 } 61 EXPORT_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 */ 72 const 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 } 79 EXPORT_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 */ 89 char *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 } 105 EXPORT_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 */ 114 void *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 } 123 EXPORT_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 */ 131 char *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 } 145 EXPORT_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. 154 */ 155 void *memdup_user(const void __user *src, size_t len) 156 { 157 void *p; 158 159 /* 160 * Always use GFP_KERNEL, since copy_from_user() can sleep and 161 * cause pagefault, which makes it pointless to use GFP_NOFS 162 * or GFP_ATOMIC. 163 */ 164 p = kmalloc_track_caller(len, GFP_KERNEL); 165 if (!p) 166 return ERR_PTR(-ENOMEM); 167 168 if (copy_from_user(p, src, len)) { 169 kfree(p); 170 return ERR_PTR(-EFAULT); 171 } 172 173 return p; 174 } 175 EXPORT_SYMBOL(memdup_user); 176 177 /* 178 * strndup_user - duplicate an existing string from user space 179 * @s: The string to duplicate 180 * @n: Maximum number of bytes to copy, including the trailing NUL. 181 */ 182 char *strndup_user(const char __user *s, long n) 183 { 184 char *p; 185 long length; 186 187 length = strnlen_user(s, n); 188 189 if (!length) 190 return ERR_PTR(-EFAULT); 191 192 if (length > n) 193 return ERR_PTR(-EINVAL); 194 195 p = memdup_user(s, length); 196 197 if (IS_ERR(p)) 198 return p; 199 200 p[length - 1] = '\0'; 201 202 return p; 203 } 204 EXPORT_SYMBOL(strndup_user); 205 206 /** 207 * memdup_user_nul - duplicate memory region from user space and NUL-terminate 208 * 209 * @src: source address in user space 210 * @len: number of bytes to copy 211 * 212 * Returns an ERR_PTR() on failure. 213 */ 214 void *memdup_user_nul(const void __user *src, size_t len) 215 { 216 char *p; 217 218 /* 219 * Always use GFP_KERNEL, since copy_from_user() can sleep and 220 * cause pagefault, which makes it pointless to use GFP_NOFS 221 * or GFP_ATOMIC. 222 */ 223 p = kmalloc_track_caller(len + 1, GFP_KERNEL); 224 if (!p) 225 return ERR_PTR(-ENOMEM); 226 227 if (copy_from_user(p, src, len)) { 228 kfree(p); 229 return ERR_PTR(-EFAULT); 230 } 231 p[len] = '\0'; 232 233 return p; 234 } 235 EXPORT_SYMBOL(memdup_user_nul); 236 237 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma, 238 struct vm_area_struct *prev, struct rb_node *rb_parent) 239 { 240 struct vm_area_struct *next; 241 242 vma->vm_prev = prev; 243 if (prev) { 244 next = prev->vm_next; 245 prev->vm_next = vma; 246 } else { 247 mm->mmap = vma; 248 if (rb_parent) 249 next = rb_entry(rb_parent, 250 struct vm_area_struct, vm_rb); 251 else 252 next = NULL; 253 } 254 vma->vm_next = next; 255 if (next) 256 next->vm_prev = vma; 257 } 258 259 /* Check if the vma is being used as a stack by this task */ 260 int vma_is_stack_for_current(struct vm_area_struct *vma) 261 { 262 struct task_struct * __maybe_unused t = current; 263 264 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t)); 265 } 266 267 #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT) 268 void arch_pick_mmap_layout(struct mm_struct *mm) 269 { 270 mm->mmap_base = TASK_UNMAPPED_BASE; 271 mm->get_unmapped_area = arch_get_unmapped_area; 272 } 273 #endif 274 275 /* 276 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall 277 * back to the regular GUP. 278 * If the architecture not support this function, simply return with no 279 * page pinned 280 */ 281 int __weak __get_user_pages_fast(unsigned long start, 282 int nr_pages, int write, struct page **pages) 283 { 284 return 0; 285 } 286 EXPORT_SYMBOL_GPL(__get_user_pages_fast); 287 288 /** 289 * get_user_pages_fast() - pin user pages in memory 290 * @start: starting user address 291 * @nr_pages: number of pages from start to pin 292 * @write: whether pages will be written to 293 * @pages: array that receives pointers to the pages pinned. 294 * Should be at least nr_pages long. 295 * 296 * Returns number of pages pinned. This may be fewer than the number 297 * requested. If nr_pages is 0 or negative, returns 0. If no pages 298 * were pinned, returns -errno. 299 * 300 * get_user_pages_fast provides equivalent functionality to get_user_pages, 301 * operating on current and current->mm, with force=0 and vma=NULL. However 302 * unlike get_user_pages, it must be called without mmap_sem held. 303 * 304 * get_user_pages_fast may take mmap_sem and page table locks, so no 305 * assumptions can be made about lack of locking. get_user_pages_fast is to be 306 * implemented in a way that is advantageous (vs get_user_pages()) when the 307 * user memory area is already faulted in and present in ptes. However if the 308 * pages have to be faulted in, it may turn out to be slightly slower so 309 * callers need to carefully consider what to use. On many architectures, 310 * get_user_pages_fast simply falls back to get_user_pages. 311 */ 312 int __weak get_user_pages_fast(unsigned long start, 313 int nr_pages, int write, struct page **pages) 314 { 315 return get_user_pages_unlocked(start, nr_pages, pages, 316 write ? FOLL_WRITE : 0); 317 } 318 EXPORT_SYMBOL_GPL(get_user_pages_fast); 319 320 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr, 321 unsigned long len, unsigned long prot, 322 unsigned long flag, unsigned long pgoff) 323 { 324 unsigned long ret; 325 struct mm_struct *mm = current->mm; 326 unsigned long populate; 327 LIST_HEAD(uf); 328 329 ret = security_mmap_file(file, prot, flag); 330 if (!ret) { 331 if (down_write_killable(&mm->mmap_sem)) 332 return -EINTR; 333 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff, 334 &populate, &uf); 335 up_write(&mm->mmap_sem); 336 userfaultfd_unmap_complete(mm, &uf); 337 if (populate) 338 mm_populate(ret, populate); 339 } 340 return ret; 341 } 342 343 unsigned long vm_mmap(struct file *file, unsigned long addr, 344 unsigned long len, unsigned long prot, 345 unsigned long flag, unsigned long offset) 346 { 347 if (unlikely(offset + PAGE_ALIGN(len) < offset)) 348 return -EINVAL; 349 if (unlikely(offset_in_page(offset))) 350 return -EINVAL; 351 352 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT); 353 } 354 EXPORT_SYMBOL(vm_mmap); 355 356 /** 357 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon 358 * failure, fall back to non-contiguous (vmalloc) allocation. 359 * @size: size of the request. 360 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL. 361 * @node: numa node to allocate from 362 * 363 * Uses kmalloc to get the memory but if the allocation fails then falls back 364 * to the vmalloc allocator. Use kvfree for freeing the memory. 365 * 366 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported. 367 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is 368 * preferable to the vmalloc fallback, due to visible performance drawbacks. 369 * 370 * Any use of gfp flags outside of GFP_KERNEL should be consulted with mm people. 371 */ 372 void *kvmalloc_node(size_t size, gfp_t flags, int node) 373 { 374 gfp_t kmalloc_flags = flags; 375 void *ret; 376 377 /* 378 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables) 379 * so the given set of flags has to be compatible. 380 */ 381 WARN_ON_ONCE((flags & GFP_KERNEL) != GFP_KERNEL); 382 383 /* 384 * We want to attempt a large physically contiguous block first because 385 * it is less likely to fragment multiple larger blocks and therefore 386 * contribute to a long term fragmentation less than vmalloc fallback. 387 * However make sure that larger requests are not too disruptive - no 388 * OOM killer and no allocation failure warnings as we have a fallback. 389 */ 390 if (size > PAGE_SIZE) { 391 kmalloc_flags |= __GFP_NOWARN; 392 393 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL)) 394 kmalloc_flags |= __GFP_NORETRY; 395 } 396 397 ret = kmalloc_node(size, kmalloc_flags, node); 398 399 /* 400 * It doesn't really make sense to fallback to vmalloc for sub page 401 * requests 402 */ 403 if (ret || size <= PAGE_SIZE) 404 return ret; 405 406 return __vmalloc_node_flags_caller(size, node, flags, 407 __builtin_return_address(0)); 408 } 409 EXPORT_SYMBOL(kvmalloc_node); 410 411 void kvfree(const void *addr) 412 { 413 if (is_vmalloc_addr(addr)) 414 vfree(addr); 415 else 416 kfree(addr); 417 } 418 EXPORT_SYMBOL(kvfree); 419 420 static inline void *__page_rmapping(struct page *page) 421 { 422 unsigned long mapping; 423 424 mapping = (unsigned long)page->mapping; 425 mapping &= ~PAGE_MAPPING_FLAGS; 426 427 return (void *)mapping; 428 } 429 430 /* Neutral page->mapping pointer to address_space or anon_vma or other */ 431 void *page_rmapping(struct page *page) 432 { 433 page = compound_head(page); 434 return __page_rmapping(page); 435 } 436 437 /* 438 * Return true if this page is mapped into pagetables. 439 * For compound page it returns true if any subpage of compound page is mapped. 440 */ 441 bool page_mapped(struct page *page) 442 { 443 int i; 444 445 if (likely(!PageCompound(page))) 446 return atomic_read(&page->_mapcount) >= 0; 447 page = compound_head(page); 448 if (atomic_read(compound_mapcount_ptr(page)) >= 0) 449 return true; 450 if (PageHuge(page)) 451 return false; 452 for (i = 0; i < hpage_nr_pages(page); i++) { 453 if (atomic_read(&page[i]._mapcount) >= 0) 454 return true; 455 } 456 return false; 457 } 458 EXPORT_SYMBOL(page_mapped); 459 460 struct anon_vma *page_anon_vma(struct page *page) 461 { 462 unsigned long mapping; 463 464 page = compound_head(page); 465 mapping = (unsigned long)page->mapping; 466 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) 467 return NULL; 468 return __page_rmapping(page); 469 } 470 471 struct address_space *page_mapping(struct page *page) 472 { 473 struct address_space *mapping; 474 475 page = compound_head(page); 476 477 /* This happens if someone calls flush_dcache_page on slab page */ 478 if (unlikely(PageSlab(page))) 479 return NULL; 480 481 if (unlikely(PageSwapCache(page))) { 482 swp_entry_t entry; 483 484 entry.val = page_private(page); 485 return swap_address_space(entry); 486 } 487 488 mapping = page->mapping; 489 if ((unsigned long)mapping & PAGE_MAPPING_ANON) 490 return NULL; 491 492 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS); 493 } 494 EXPORT_SYMBOL(page_mapping); 495 496 /* Slow path of page_mapcount() for compound pages */ 497 int __page_mapcount(struct page *page) 498 { 499 int ret; 500 501 ret = atomic_read(&page->_mapcount) + 1; 502 /* 503 * For file THP page->_mapcount contains total number of mapping 504 * of the page: no need to look into compound_mapcount. 505 */ 506 if (!PageAnon(page) && !PageHuge(page)) 507 return ret; 508 page = compound_head(page); 509 ret += atomic_read(compound_mapcount_ptr(page)) + 1; 510 if (PageDoubleMap(page)) 511 ret--; 512 return ret; 513 } 514 EXPORT_SYMBOL_GPL(__page_mapcount); 515 516 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; 517 int sysctl_overcommit_ratio __read_mostly = 50; 518 unsigned long sysctl_overcommit_kbytes __read_mostly; 519 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; 520 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ 521 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ 522 523 int overcommit_ratio_handler(struct ctl_table *table, int write, 524 void __user *buffer, size_t *lenp, 525 loff_t *ppos) 526 { 527 int ret; 528 529 ret = proc_dointvec(table, write, buffer, lenp, ppos); 530 if (ret == 0 && write) 531 sysctl_overcommit_kbytes = 0; 532 return ret; 533 } 534 535 int overcommit_kbytes_handler(struct ctl_table *table, int write, 536 void __user *buffer, size_t *lenp, 537 loff_t *ppos) 538 { 539 int ret; 540 541 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); 542 if (ret == 0 && write) 543 sysctl_overcommit_ratio = 0; 544 return ret; 545 } 546 547 /* 548 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used 549 */ 550 unsigned long vm_commit_limit(void) 551 { 552 unsigned long allowed; 553 554 if (sysctl_overcommit_kbytes) 555 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10); 556 else 557 allowed = ((totalram_pages - hugetlb_total_pages()) 558 * sysctl_overcommit_ratio / 100); 559 allowed += total_swap_pages; 560 561 return allowed; 562 } 563 564 /* 565 * Make sure vm_committed_as in one cacheline and not cacheline shared with 566 * other variables. It can be updated by several CPUs frequently. 567 */ 568 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; 569 570 /* 571 * The global memory commitment made in the system can be a metric 572 * that can be used to drive ballooning decisions when Linux is hosted 573 * as a guest. On Hyper-V, the host implements a policy engine for dynamically 574 * balancing memory across competing virtual machines that are hosted. 575 * Several metrics drive this policy engine including the guest reported 576 * memory commitment. 577 */ 578 unsigned long vm_memory_committed(void) 579 { 580 return percpu_counter_read_positive(&vm_committed_as); 581 } 582 EXPORT_SYMBOL_GPL(vm_memory_committed); 583 584 /* 585 * Check that a process has enough memory to allocate a new virtual 586 * mapping. 0 means there is enough memory for the allocation to 587 * succeed and -ENOMEM implies there is not. 588 * 589 * We currently support three overcommit policies, which are set via the 590 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 591 * 592 * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 593 * Additional code 2002 Jul 20 by Robert Love. 594 * 595 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 596 * 597 * Note this is a helper function intended to be used by LSMs which 598 * wish to use this logic. 599 */ 600 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 601 { 602 long free, allowed, reserve; 603 604 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) < 605 -(s64)vm_committed_as_batch * num_online_cpus(), 606 "memory commitment underflow"); 607 608 vm_acct_memory(pages); 609 610 /* 611 * Sometimes we want to use more memory than we have 612 */ 613 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 614 return 0; 615 616 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 617 free = global_page_state(NR_FREE_PAGES); 618 free += global_node_page_state(NR_FILE_PAGES); 619 620 /* 621 * shmem pages shouldn't be counted as free in this 622 * case, they can't be purged, only swapped out, and 623 * that won't affect the overall amount of available 624 * memory in the system. 625 */ 626 free -= global_node_page_state(NR_SHMEM); 627 628 free += get_nr_swap_pages(); 629 630 /* 631 * Any slabs which are created with the 632 * SLAB_RECLAIM_ACCOUNT flag claim to have contents 633 * which are reclaimable, under pressure. The dentry 634 * cache and most inode caches should fall into this 635 */ 636 free += global_node_page_state(NR_SLAB_RECLAIMABLE); 637 638 /* 639 * Leave reserved pages. The pages are not for anonymous pages. 640 */ 641 if (free <= totalreserve_pages) 642 goto error; 643 else 644 free -= totalreserve_pages; 645 646 /* 647 * Reserve some for root 648 */ 649 if (!cap_sys_admin) 650 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 651 652 if (free > pages) 653 return 0; 654 655 goto error; 656 } 657 658 allowed = vm_commit_limit(); 659 /* 660 * Reserve some for root 661 */ 662 if (!cap_sys_admin) 663 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 664 665 /* 666 * Don't let a single process grow so big a user can't recover 667 */ 668 if (mm) { 669 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); 670 allowed -= min_t(long, mm->total_vm / 32, reserve); 671 } 672 673 if (percpu_counter_read_positive(&vm_committed_as) < allowed) 674 return 0; 675 error: 676 vm_unacct_memory(pages); 677 678 return -ENOMEM; 679 } 680 681 /** 682 * get_cmdline() - copy the cmdline value to a buffer. 683 * @task: the task whose cmdline value to copy. 684 * @buffer: the buffer to copy to. 685 * @buflen: the length of the buffer. Larger cmdline values are truncated 686 * to this length. 687 * Returns the size of the cmdline field copied. Note that the copy does 688 * not guarantee an ending NULL byte. 689 */ 690 int get_cmdline(struct task_struct *task, char *buffer, int buflen) 691 { 692 int res = 0; 693 unsigned int len; 694 struct mm_struct *mm = get_task_mm(task); 695 unsigned long arg_start, arg_end, env_start, env_end; 696 if (!mm) 697 goto out; 698 if (!mm->arg_end) 699 goto out_mm; /* Shh! No looking before we're done */ 700 701 down_read(&mm->mmap_sem); 702 arg_start = mm->arg_start; 703 arg_end = mm->arg_end; 704 env_start = mm->env_start; 705 env_end = mm->env_end; 706 up_read(&mm->mmap_sem); 707 708 len = arg_end - arg_start; 709 710 if (len > buflen) 711 len = buflen; 712 713 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE); 714 715 /* 716 * If the nul at the end of args has been overwritten, then 717 * assume application is using setproctitle(3). 718 */ 719 if (res > 0 && buffer[res-1] != '\0' && len < buflen) { 720 len = strnlen(buffer, res); 721 if (len < res) { 722 res = len; 723 } else { 724 len = env_end - env_start; 725 if (len > buflen - res) 726 len = buflen - res; 727 res += access_process_vm(task, env_start, 728 buffer+res, len, 729 FOLL_FORCE); 730 res = strnlen(buffer, res); 731 } 732 } 733 out_mm: 734 mmput(mm); 735 out: 736 return res; 737 } 738