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 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, pages,
 287                                        write ? FOLL_WRITE : 0);
 288 }
 289 EXPORT_SYMBOL_GPL(get_user_pages_fast);
 290
 291 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
 292         unsigned long len, unsigned long prot,
 293         unsigned long flag, unsigned long pgoff)
 294 {
 295         unsigned long ret;
 296         struct mm_struct *mm = current->mm;
 297         unsigned long populate;
 298
 299         ret = security_mmap_file(file, prot, flag);
 300         if (!ret) {
 301                 if (down_write_killable(&mm->mmap_sem))
 302                         return -EINTR;
 303                 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
 304                                     &populate);
 305                 up_write(&mm->mmap_sem);
 306                 if (populate)
 307                         mm_populate(ret, populate);
 308         }
 309         return ret;
 310 }
 311
 312 unsigned long vm_mmap(struct file *file, unsigned long addr,
 313         unsigned long len, unsigned long prot,
 314         unsigned long flag, unsigned long offset)
 315 {
 316         if (unlikely(offset + PAGE_ALIGN(len) < offset))
 317                 return -EINVAL;
 318         if (unlikely(offset_in_page(offset)))
 319                 return -EINVAL;
 320
 321         return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
 322 }
 323 EXPORT_SYMBOL(vm_mmap);
 324
 325 void kvfree(const void *addr)
 326 {
 327         if (is_vmalloc_addr(addr))
 328                 vfree(addr);
 329         else
 330                 kfree(addr);
 331 }
 332 EXPORT_SYMBOL(kvfree);
 333
 334 static inline void *__page_rmapping(struct page *page)
 335 {
 336         unsigned long mapping;
 337
 338         mapping = (unsigned long)page->mapping;
 339         mapping &= ~PAGE_MAPPING_FLAGS;
 340
 341         return (void *)mapping;
 342 }
 343
 344 /* Neutral page->mapping pointer to address_space or anon_vma or other */
 345 void *page_rmapping(struct page *page)
 346 {
 347         page = compound_head(page);
 348         return __page_rmapping(page);
 349 }
 350
 351 /*
 352  * Return true if this page is mapped into pagetables.
 353  * For compound page it returns true if any subpage of compound page is mapped.
 354  */
 355 bool page_mapped(struct page *page)
 356 {
 357         int i;
 358
 359         if (likely(!PageCompound(page)))
 360                 return atomic_read(&page->_mapcount) >= 0;
 361         page = compound_head(page);
 362         if (atomic_read(compound_mapcount_ptr(page)) >= 0)
 363                 return true;
 364         if (PageHuge(page))
 365                 return false;
 366         for (i = 0; i < hpage_nr_pages(page); i++) {
 367                 if (atomic_read(&page[i]._mapcount) >= 0)
 368                         return true;
 369         }
 370         return false;
 371 }
 372 EXPORT_SYMBOL(page_mapped);
 373
 374 struct anon_vma *page_anon_vma(struct page *page)
 375 {
 376         unsigned long mapping;
 377
 378         page = compound_head(page);
 379         mapping = (unsigned long)page->mapping;
 380         if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
 381                 return NULL;
 382         return __page_rmapping(page);
 383 }
 384
 385 struct address_space *page_mapping(struct page *page)
 386 {
 387         struct address_space *mapping;
 388
 389         page = compound_head(page);
 390
 391         /* This happens if someone calls flush_dcache_page on slab page */
 392         if (unlikely(PageSlab(page)))
 393                 return NULL;
 394
 395         if (unlikely(PageSwapCache(page))) {
 396                 swp_entry_t entry;
 397
 398                 entry.val = page_private(page);
 399                 return swap_address_space(entry);
 400         }
 401
 402         mapping = page->mapping;
 403         if ((unsigned long)mapping & PAGE_MAPPING_ANON)
 404                 return NULL;
 405
 406         return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
 407 }
 408 EXPORT_SYMBOL(page_mapping);
 409
 410 /* Slow path of page_mapcount() for compound pages */
 411 int __page_mapcount(struct page *page)
 412 {
 413         int ret;
 414
 415         ret = atomic_read(&page->_mapcount) + 1;
 416         /*
 417          * For file THP page->_mapcount contains total number of mapping
 418          * of the page: no need to look into compound_mapcount.
 419          */
 420         if (!PageAnon(page) && !PageHuge(page))
 421                 return ret;
 422         page = compound_head(page);
 423         ret += atomic_read(compound_mapcount_ptr(page)) + 1;
 424         if (PageDoubleMap(page))
 425                 ret--;
 426         return ret;
 427 }
 428 EXPORT_SYMBOL_GPL(__page_mapcount);
 429
 430 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
 431 int sysctl_overcommit_ratio __read_mostly = 50;
 432 unsigned long sysctl_overcommit_kbytes __read_mostly;
 433 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
 434 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
 435 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
 436
 437 int overcommit_ratio_handler(struct ctl_table *table, int write,
 438                              void __user *buffer, size_t *lenp,
 439                              loff_t *ppos)
 440 {
 441         int ret;
 442
 443         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 444         if (ret == 0 && write)
 445                 sysctl_overcommit_kbytes = 0;
 446         return ret;
 447 }
 448
 449 int overcommit_kbytes_handler(struct ctl_table *table, int write,
 450                              void __user *buffer, size_t *lenp,
 451                              loff_t *ppos)
 452 {
 453         int ret;
 454
 455         ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 456         if (ret == 0 && write)
 457                 sysctl_overcommit_ratio = 0;
 458         return ret;
 459 }
 460
 461 /*
 462  * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
 463  */
 464 unsigned long vm_commit_limit(void)
 465 {
 466         unsigned long allowed;
 467
 468         if (sysctl_overcommit_kbytes)
 469                 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
 470         else
 471                 allowed = ((totalram_pages - hugetlb_total_pages())
 472                            * sysctl_overcommit_ratio / 100);
 473         allowed += total_swap_pages;
 474
 475         return allowed;
 476 }
 477
 478 /*
 479  * Make sure vm_committed_as in one cacheline and not cacheline shared with
 480  * other variables. It can be updated by several CPUs frequently.
 481  */
 482 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
 483
 484 /*
 485  * The global memory commitment made in the system can be a metric
 486  * that can be used to drive ballooning decisions when Linux is hosted
 487  * as a guest. On Hyper-V, the host implements a policy engine for dynamically
 488  * balancing memory across competing virtual machines that are hosted.
 489  * Several metrics drive this policy engine including the guest reported
 490  * memory commitment.
 491  */
 492 unsigned long vm_memory_committed(void)
 493 {
 494         return percpu_counter_read_positive(&vm_committed_as);
 495 }
 496 EXPORT_SYMBOL_GPL(vm_memory_committed);
 497
 498 /*
 499  * Check that a process has enough memory to allocate a new virtual
 500  * mapping. 0 means there is enough memory for the allocation to
 501  * succeed and -ENOMEM implies there is not.
 502  *
 503  * We currently support three overcommit policies, which are set via the
 504  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
 505  *
 506  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
 507  * Additional code 2002 Jul 20 by Robert Love.
 508  *
 509  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
 510  *
 511  * Note this is a helper function intended to be used by LSMs which
 512  * wish to use this logic.
 513  */
 514 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
 515 {
 516         long free, allowed, reserve;
 517
 518         VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
 519                         -(s64)vm_committed_as_batch * num_online_cpus(),
 520                         "memory commitment underflow");
 521
 522         vm_acct_memory(pages);
 523
 524         /*
 525          * Sometimes we want to use more memory than we have
 526          */
 527         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
 528                 return 0;
 529
 530         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
 531                 free = global_page_state(NR_FREE_PAGES);
 532                 free += global_node_page_state(NR_FILE_PAGES);
 533
 534                 /*
 535                  * shmem pages shouldn't be counted as free in this
 536                  * case, they can't be purged, only swapped out, and
 537                  * that won't affect the overall amount of available
 538                  * memory in the system.
 539                  */
 540                 free -= global_node_page_state(NR_SHMEM);
 541
 542                 free += get_nr_swap_pages();
 543
 544                 /*
 545                  * Any slabs which are created with the
 546                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
 547                  * which are reclaimable, under pressure.  The dentry
 548                  * cache and most inode caches should fall into this
 549                  */
 550                 free += global_page_state(NR_SLAB_RECLAIMABLE);
 551
 552                 /*
 553                  * Leave reserved pages. The pages are not for anonymous pages.
 554                  */
 555                 if (free <= totalreserve_pages)
 556                         goto error;
 557                 else
 558                         free -= totalreserve_pages;
 559
 560                 /*
 561                  * Reserve some for root
 562                  */
 563                 if (!cap_sys_admin)
 564                         free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
 565
 566                 if (free > pages)
 567                         return 0;
 568
 569                 goto error;
 570         }
 571
 572         allowed = vm_commit_limit();
 573         /*
 574          * Reserve some for root
 575          */
 576         if (!cap_sys_admin)
 577                 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
 578
 579         /*
 580          * Don't let a single process grow so big a user can't recover
 581          */
 582         if (mm) {
 583                 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
 584                 allowed -= min_t(long, mm->total_vm / 32, reserve);
 585         }
 586
 587         if (percpu_counter_read_positive(&vm_committed_as) < allowed)
 588                 return 0;
 589 error:
 590         vm_unacct_memory(pages);
 591
 592         return -ENOMEM;
 593 }
 594
 595 /**
 596  * get_cmdline() - copy the cmdline value to a buffer.
 597  * @task:     the task whose cmdline value to copy.
 598  * @buffer:   the buffer to copy to.
 599  * @buflen:   the length of the buffer. Larger cmdline values are truncated
 600  *            to this length.
 601  * Returns the size of the cmdline field copied. Note that the copy does
 602  * not guarantee an ending NULL byte.
 603  */
 604 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
 605 {
 606         int res = 0;
 607         unsigned int len;
 608         struct mm_struct *mm = get_task_mm(task);
 609         unsigned long arg_start, arg_end, env_start, env_end;
 610         if (!mm)
 611                 goto out;
 612         if (!mm->arg_end)
 613                 goto out_mm;    /* Shh! No looking before we're done */
 614
 615         down_read(&mm->mmap_sem);
 616         arg_start = mm->arg_start;
 617         arg_end = mm->arg_end;
 618         env_start = mm->env_start;
 619         env_end = mm->env_end;
 620         up_read(&mm->mmap_sem);
 621
 622         len = arg_end - arg_start;
 623
 624         if (len > buflen)
 625                 len = buflen;
 626
 627         res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
 628
 629         /*
 630          * If the nul at the end of args has been overwritten, then
 631          * assume application is using setproctitle(3).
 632          */
 633         if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
 634                 len = strnlen(buffer, res);
 635                 if (len < res) {
 636                         res = len;
 637                 } else {
 638                         len = env_end - env_start;
 639                         if (len > buflen - res)
 640                                 len = buflen - res;
 641                         res += access_process_vm(task, env_start,
 642                                                  buffer+res, len,
 643                                                  FOLL_FORCE);
 644                         res = strnlen(buffer, res);
 645                 }
 646         }
 647 out_mm:
 648         mmput(mm);
 649 out:
 650         return res;
 651 }