mm/util.c

   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  * Note: Use kmemdup_nul() instead if the size is known exactly.
  85  */
  86 char *kstrndup(const char *s, size_t max, gfp_t gfp)
  87 {
  88         size_t len;
  89         char *buf;
  90
  91         if (!s)
  92                 return NULL;
  93
  94         len = strnlen(s, max);
  95         buf = kmalloc_track_caller(len+1, gfp);
  96         if (buf) {
  97                 memcpy(buf, s, len);
  98                 buf[len] = '\0';
  99         }
 100         return buf;
 101 }
 102 EXPORT_SYMBOL(kstrndup);
 103
 104 /**
 105  * kmemdup - duplicate region of memory
 106  *
 107  * @src: memory region to duplicate
 108  * @len: memory region length
 109  * @gfp: GFP mask to use
 110  */
 111 void *kmemdup(const void *src, size_t len, gfp_t gfp)
 112 {
 113         void *p;
 114
 115         p = kmalloc_track_caller(len, gfp);
 116         if (p)
 117                 memcpy(p, src, len);
 118         return p;
 119 }
 120 EXPORT_SYMBOL(kmemdup);
 121
 122 /**
 123  * kmemdup_nul - Create a NUL-terminated string from unterminated data
 124  * @s: The data to stringify
 125  * @len: The size of the data
 126  * @gfp: the GFP mask used in the kmalloc() call when allocating memory
 127  */
 128 char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
 129 {
 130         char *buf;
 131
 132         if (!s)
 133                 return NULL;
 134
 135         buf = kmalloc_track_caller(len + 1, gfp);
 136         if (buf) {
 137                 memcpy(buf, s, len);
 138                 buf[len] = '\0';
 139         }
 140         return buf;
 141 }
 142 EXPORT_SYMBOL(kmemdup_nul);
 143
 144 /**
 145  * memdup_user - duplicate memory region from user space
 146  *
 147  * @src: source address in user space
 148  * @len: number of bytes to copy
 149  *
 150  * Returns an ERR_PTR() on failure.
 151  */
 152 void *memdup_user(const void __user *src, size_t len)
 153 {
 154         void *p;
 155
 156         /*
 157          * Always use GFP_KERNEL, since copy_from_user() can sleep and
 158          * cause pagefault, which makes it pointless to use GFP_NOFS
 159          * or GFP_ATOMIC.
 160          */
 161         p = kmalloc_track_caller(len, GFP_KERNEL);
 162         if (!p)
 163                 return ERR_PTR(-ENOMEM);
 164
 165         if (copy_from_user(p, src, len)) {
 166                 kfree(p);
 167                 return ERR_PTR(-EFAULT);
 168         }
 169
 170         return p;
 171 }
 172 EXPORT_SYMBOL(memdup_user);
 173
 174 /*
 175  * strndup_user - duplicate an existing string from user space
 176  * @s: The string to duplicate
 177  * @n: Maximum number of bytes to copy, including the trailing NUL.
 178  */
 179 char *strndup_user(const char __user *s, long n)
 180 {
 181         char *p;
 182         long length;
 183
 184         length = strnlen_user(s, n);
 185
 186         if (!length)
 187                 return ERR_PTR(-EFAULT);
 188
 189         if (length > n)
 190                 return ERR_PTR(-EINVAL);
 191
 192         p = memdup_user(s, length);
 193
 194         if (IS_ERR(p))
 195                 return p;
 196
 197         p[length - 1] = '\0';
 198
 199         return p;
 200 }
 201 EXPORT_SYMBOL(strndup_user);
 202
 203 /**
 204  * memdup_user_nul - duplicate memory region from user space and NUL-terminate
 205  *
 206  * @src: source address in user space
 207  * @len: number of bytes to copy
 208  *
 209  * Returns an ERR_PTR() on failure.
 210  */
 211 void *memdup_user_nul(const void __user *src, size_t len)
 212 {
 213         char *p;
 214
 215         /*
 216          * Always use GFP_KERNEL, since copy_from_user() can sleep and
 217          * cause pagefault, which makes it pointless to use GFP_NOFS
 218          * or GFP_ATOMIC.
 219          */
 220         p = kmalloc_track_caller(len + 1, GFP_KERNEL);
 221         if (!p)
 222                 return ERR_PTR(-ENOMEM);
 223
 224         if (copy_from_user(p, src, len)) {
 225                 kfree(p);
 226                 return ERR_PTR(-EFAULT);
 227         }
 228         p[len] = '\0';
 229
 230         return p;
 231 }
 232 EXPORT_SYMBOL(memdup_user_nul);
 233
 234 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
 235                 struct vm_area_struct *prev, struct rb_node *rb_parent)
 236 {
 237         struct vm_area_struct *next;
 238
 239         vma->vm_prev = prev;
 240         if (prev) {
 241                 next = prev->vm_next;
 242                 prev->vm_next = vma;
 243         } else {
 244                 mm->mmap = vma;
 245                 if (rb_parent)
 246                         next = rb_entry(rb_parent,
 247                                         struct vm_area_struct, vm_rb);
 248                 else
 249                         next = NULL;
 250         }
 251         vma->vm_next = next;
 252         if (next)
 253                 next->vm_prev = vma;
 254 }
 255
 256 /* Check if the vma is being used as a stack by this task */
 257 int vma_is_stack_for_current(struct vm_area_struct *vma)
 258 {
 259         struct task_struct * __maybe_unused t = current;
 260
 261         return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
 262 }
 263
 264 #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
 265 void arch_pick_mmap_layout(struct mm_struct *mm)
 266 {
 267         mm->mmap_base = TASK_UNMAPPED_BASE;
 268         mm->get_unmapped_area = arch_get_unmapped_area;
 269 }
 270 #endif
 271
 272 /*
 273  * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
 274  * back to the regular GUP.
 275  * If the architecture not support this function, simply return with no
 276  * page pinned
 277  */
 278 int __weak __get_user_pages_fast(unsigned long start,
 279                                  int nr_pages, int write, struct page **pages)
 280 {
 281         return 0;
 282 }
 283 EXPORT_SYMBOL_GPL(__get_user_pages_fast);
 284
 285 /**
 286  * get_user_pages_fast() - pin user pages in memory
 287  * @start:      starting user address
 288  * @nr_pages:   number of pages from start to pin
 289  * @write:      whether pages will be written to
 290  * @pages:      array that receives pointers to the pages pinned.
 291  *              Should be at least nr_pages long.
 292  *
 293  * Returns number of pages pinned. This may be fewer than the number
 294  * requested. If nr_pages is 0 or negative, returns 0. If no pages
 295  * were pinned, returns -errno.
 296  *
 297  * get_user_pages_fast provides equivalent functionality to get_user_pages,
 298  * operating on current and current->mm, with force=0 and vma=NULL. However
 299  * unlike get_user_pages, it must be called without mmap_sem held.
 300  *
 301  * get_user_pages_fast may take mmap_sem and page table locks, so no
 302  * assumptions can be made about lack of locking. get_user_pages_fast is to be
 303  * implemented in a way that is advantageous (vs get_user_pages()) when the
 304  * user memory area is already faulted in and present in ptes. However if the
 305  * pages have to be faulted in, it may turn out to be slightly slower so
 306  * callers need to carefully consider what to use. On many architectures,
 307  * get_user_pages_fast simply falls back to get_user_pages.
 308  */
 309 int __weak get_user_pages_fast(unsigned long start,
 310                                 int nr_pages, int write, struct page **pages)
 311 {
 312         return get_user_pages_unlocked(start, nr_pages, pages,
 313                                        write ? FOLL_WRITE : 0);
 314 }
 315 EXPORT_SYMBOL_GPL(get_user_pages_fast);
 316
 317 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
 318         unsigned long len, unsigned long prot,
 319         unsigned long flag, unsigned long pgoff)
 320 {
 321         unsigned long ret;
 322         struct mm_struct *mm = current->mm;
 323         unsigned long populate;
 324
 325         ret = security_mmap_file(file, prot, flag);
 326         if (!ret) {
 327                 if (down_write_killable(&mm->mmap_sem))
 328                         return -EINTR;
 329                 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
 330                                     &populate);
 331                 up_write(&mm->mmap_sem);
 332                 if (populate)
 333                         mm_populate(ret, populate);
 334         }
 335         return ret;
 336 }
 337
 338 unsigned long vm_mmap(struct file *file, unsigned long addr,
 339         unsigned long len, unsigned long prot,
 340         unsigned long flag, unsigned long offset)
 341 {
 342         if (unlikely(offset + PAGE_ALIGN(len) < offset))
 343                 return -EINVAL;
 344         if (unlikely(offset_in_page(offset)))
 345                 return -EINVAL;
 346
 347         return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
 348 }
 349 EXPORT_SYMBOL(vm_mmap);
 350
 351 void kvfree(const void *addr)
 352 {
 353         if (is_vmalloc_addr(addr))
 354                 vfree(addr);
 355         else
 356                 kfree(addr);
 357 }
 358 EXPORT_SYMBOL(kvfree);
 359
 360 static inline void *__page_rmapping(struct page *page)
 361 {
 362         unsigned long mapping;
 363
 364         mapping = (unsigned long)page->mapping;
 365         mapping &= ~PAGE_MAPPING_FLAGS;
 366
 367         return (void *)mapping;
 368 }
 369
 370 /* Neutral page->mapping pointer to address_space or anon_vma or other */
 371 void *page_rmapping(struct page *page)
 372 {
 373         page = compound_head(page);
 374         return __page_rmapping(page);
 375 }
 376
 377 /*
 378  * Return true if this page is mapped into pagetables.
 379  * For compound page it returns true if any subpage of compound page is mapped.
 380  */
 381 bool page_mapped(struct page *page)
 382 {
 383         int i;
 384
 385         if (likely(!PageCompound(page)))
 386                 return atomic_read(&page->_mapcount) >= 0;
 387         page = compound_head(page);
 388         if (atomic_read(compound_mapcount_ptr(page)) >= 0)
 389                 return true;
 390         if (PageHuge(page))
 391                 return false;
 392         for (i = 0; i < (1 << compound_order(page)); i++) {
 393                 if (atomic_read(&page[i]._mapcount) >= 0)
 394                         return true;
 395         }
 396         return false;
 397 }
 398 EXPORT_SYMBOL(page_mapped);
 399
 400 struct anon_vma *page_anon_vma(struct page *page)
 401 {
 402         unsigned long mapping;
 403
 404         page = compound_head(page);
 405         mapping = (unsigned long)page->mapping;
 406         if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
 407                 return NULL;
 408         return __page_rmapping(page);
 409 }
 410
 411 struct address_space *page_mapping(struct page *page)
 412 {
 413         struct address_space *mapping;
 414
 415         page = compound_head(page);
 416
 417         /* This happens if someone calls flush_dcache_page on slab page */
 418         if (unlikely(PageSlab(page)))
 419                 return NULL;
 420
 421         if (unlikely(PageSwapCache(page))) {
 422                 swp_entry_t entry;
 423
 424                 entry.val = page_private(page);
 425                 return swap_address_space(entry);
 426         }
 427
 428         mapping = page->mapping;
 429         if ((unsigned long)mapping & PAGE_MAPPING_ANON)
 430                 return NULL;
 431
 432         return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
 433 }
 434 EXPORT_SYMBOL(page_mapping);
 435
 436 /* Slow path of page_mapcount() for compound pages */
 437 int __page_mapcount(struct page *page)
 438 {
 439         int ret;
 440
 441         ret = atomic_read(&page->_mapcount) + 1;
 442         /*
 443          * For file THP page->_mapcount contains total number of mapping
 444          * of the page: no need to look into compound_mapcount.
 445          */
 446         if (!PageAnon(page) && !PageHuge(page))
 447                 return ret;
 448         page = compound_head(page);
 449         ret += atomic_read(compound_mapcount_ptr(page)) + 1;
 450         if (PageDoubleMap(page))
 451                 ret--;
 452         return ret;
 453 }
 454 EXPORT_SYMBOL_GPL(__page_mapcount);
 455
 456 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
 457 int sysctl_overcommit_ratio __read_mostly = 50;
 458 unsigned long sysctl_overcommit_kbytes __read_mostly;
 459 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
 460 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
 461 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
 462
 463 int overcommit_ratio_handler(struct ctl_table *table, int write,
 464                              void __user *buffer, size_t *lenp,
 465                              loff_t *ppos)
 466 {
 467         int ret;
 468
 469         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 470         if (ret == 0 && write)
 471                 sysctl_overcommit_kbytes = 0;
 472         return ret;
 473 }
 474
 475 int overcommit_kbytes_handler(struct ctl_table *table, int write,
 476                              void __user *buffer, size_t *lenp,
 477                              loff_t *ppos)
 478 {
 479         int ret;
 480
 481         ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 482         if (ret == 0 && write)
 483                 sysctl_overcommit_ratio = 0;
 484         return ret;
 485 }
 486
 487 /*
 488  * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
 489  */
 490 unsigned long vm_commit_limit(void)
 491 {
 492         unsigned long allowed;
 493
 494         if (sysctl_overcommit_kbytes)
 495                 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
 496         else
 497                 allowed = ((totalram_pages - hugetlb_total_pages())
 498                            * sysctl_overcommit_ratio / 100);
 499         allowed += total_swap_pages;
 500
 501         return allowed;
 502 }
 503
 504 /*
 505  * Make sure vm_committed_as in one cacheline and not cacheline shared with
 506  * other variables. It can be updated by several CPUs frequently.
 507  */
 508 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
 509
 510 /*
 511  * The global memory commitment made in the system can be a metric
 512  * that can be used to drive ballooning decisions when Linux is hosted
 513  * as a guest. On Hyper-V, the host implements a policy engine for dynamically
 514  * balancing memory across competing virtual machines that are hosted.
 515  * Several metrics drive this policy engine including the guest reported
 516  * memory commitment.
 517  */
 518 unsigned long vm_memory_committed(void)
 519 {
 520         return percpu_counter_read_positive(&vm_committed_as);
 521 }
 522 EXPORT_SYMBOL_GPL(vm_memory_committed);
 523
 524 /*
 525  * Check that a process has enough memory to allocate a new virtual
 526  * mapping. 0 means there is enough memory for the allocation to
 527  * succeed and -ENOMEM implies there is not.
 528  *
 529  * We currently support three overcommit policies, which are set via the
 530  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
 531  *
 532  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
 533  * Additional code 2002 Jul 20 by Robert Love.
 534  *
 535  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
 536  *
 537  * Note this is a helper function intended to be used by LSMs which
 538  * wish to use this logic.
 539  */
 540 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
 541 {
 542         long free, allowed, reserve;
 543
 544         VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
 545                         -(s64)vm_committed_as_batch * num_online_cpus(),
 546                         "memory commitment underflow");
 547
 548         vm_acct_memory(pages);
 549
 550         /*
 551          * Sometimes we want to use more memory than we have
 552          */
 553         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
 554                 return 0;
 555
 556         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
 557                 free = global_page_state(NR_FREE_PAGES);
 558                 free += global_node_page_state(NR_FILE_PAGES);
 559
 560                 /*
 561                  * shmem pages shouldn't be counted as free in this
 562                  * case, they can't be purged, only swapped out, and
 563                  * that won't affect the overall amount of available
 564                  * memory in the system.
 565                  */
 566                 free -= global_node_page_state(NR_SHMEM);
 567
 568                 free += get_nr_swap_pages();
 569
 570                 /*
 571                  * Any slabs which are created with the
 572                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
 573                  * which are reclaimable, under pressure.  The dentry
 574                  * cache and most inode caches should fall into this
 575                  */
 576                 free += global_page_state(NR_SLAB_RECLAIMABLE);
 577
 578                 /*
 579                  * Leave reserved pages. The pages are not for anonymous pages.
 580                  */
 581                 if (free <= totalreserve_pages)
 582                         goto error;
 583                 else
 584                         free -= totalreserve_pages;
 585
 586                 /*
 587                  * Reserve some for root
 588                  */
 589                 if (!cap_sys_admin)
 590                         free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
 591
 592                 if (free > pages)
 593                         return 0;
 594
 595                 goto error;
 596         }
 597
 598         allowed = vm_commit_limit();
 599         /*
 600          * Reserve some for root
 601          */
 602         if (!cap_sys_admin)
 603                 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
 604
 605         /*
 606          * Don't let a single process grow so big a user can't recover
 607          */
 608         if (mm) {
 609                 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
 610                 allowed -= min_t(long, mm->total_vm / 32, reserve);
 611         }
 612
 613         if (percpu_counter_read_positive(&vm_committed_as) < allowed)
 614                 return 0;
 615 error:
 616         vm_unacct_memory(pages);
 617
 618         return -ENOMEM;
 619 }
 620
 621 /**
 622  * get_cmdline() - copy the cmdline value to a buffer.
 623  * @task:     the task whose cmdline value to copy.
 624  * @buffer:   the buffer to copy to.
 625  * @buflen:   the length of the buffer. Larger cmdline values are truncated
 626  *            to this length.
 627  * Returns the size of the cmdline field copied. Note that the copy does
 628  * not guarantee an ending NULL byte.
 629  */
 630 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
 631 {
 632         int res = 0;
 633         unsigned int len;
 634         struct mm_struct *mm = get_task_mm(task);
 635         unsigned long arg_start, arg_end, env_start, env_end;
 636         if (!mm)
 637                 goto out;
 638         if (!mm->arg_end)
 639                 goto out_mm;    /* Shh! No looking before we're done */
 640
 641         down_read(&mm->mmap_sem);
 642         arg_start = mm->arg_start;
 643         arg_end = mm->arg_end;
 644         env_start = mm->env_start;
 645         env_end = mm->env_end;
 646         up_read(&mm->mmap_sem);
 647
 648         len = arg_end - arg_start;
 649
 650         if (len > buflen)
 651                 len = buflen;
 652
 653         res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
 654
 655         /*
 656          * If the nul at the end of args has been overwritten, then
 657          * assume application is using setproctitle(3).
 658          */
 659         if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
 660                 len = strnlen(buffer, res);
 661                 if (len < res) {
 662                         res = len;
 663                 } else {
 664                         len = env_end - env_start;
 665                         if (len > buflen - res)
 666                                 len = buflen - res;
 667                         res += access_process_vm(task, env_start,
 668                                                  buffer+res, len,
 669                                                  FOLL_FORCE);
 670                         res = strnlen(buffer, res);
 671                 }
 672         }
 673 out_mm:
 674         mmput(mm);
 675 out:
 676         return res;
 677 }