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