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