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         struct mm_struct *mm = current->mm;
 287         return get_user_pages_unlocked(current, mm, start, nr_pages,
 288                                        write, 0, pages);
 289 }
 290 EXPORT_SYMBOL_GPL(get_user_pages_fast);
 291
 292 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
 293         unsigned long len, unsigned long prot,
 294         unsigned long flag, unsigned long pgoff)
 295 {
 296         unsigned long ret;
 297         struct mm_struct *mm = current->mm;
 298         unsigned long populate;
 299
 300         ret = security_mmap_file(file, prot, flag);
 301         if (!ret) {
 302                 down_write(&mm->mmap_sem);
 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 struct anon_vma *page_anon_vma(struct page *page)
 352 {
 353         unsigned long mapping;
 354
 355         page = compound_head(page);
 356         mapping = (unsigned long)page->mapping;
 357         if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
 358                 return NULL;
 359         return __page_rmapping(page);
 360 }
 361
 362 struct address_space *page_mapping(struct page *page)
 363 {
 364         struct address_space *mapping;
 365
 366         page = compound_head(page);
 367
 368         /* This happens if someone calls flush_dcache_page on slab page */
 369         if (unlikely(PageSlab(page)))
 370                 return NULL;
 371
 372         if (unlikely(PageSwapCache(page))) {
 373                 swp_entry_t entry;
 374
 375                 entry.val = page_private(page);
 376                 return swap_address_space(entry);
 377         }
 378
 379         mapping = page->mapping;
 380         if ((unsigned long)mapping & PAGE_MAPPING_FLAGS)
 381                 return NULL;
 382         return mapping;
 383 }
 384
 385 /* Slow path of page_mapcount() for compound pages */
 386 int __page_mapcount(struct page *page)
 387 {
 388         int ret;
 389
 390         ret = atomic_read(&page->_mapcount) + 1;
 391         page = compound_head(page);
 392         ret += atomic_read(compound_mapcount_ptr(page)) + 1;
 393         if (PageDoubleMap(page))
 394                 ret--;
 395         return ret;
 396 }
 397 EXPORT_SYMBOL_GPL(__page_mapcount);
 398
 399 int overcommit_ratio_handler(struct ctl_table *table, int write,
 400                              void __user *buffer, size_t *lenp,
 401                              loff_t *ppos)
 402 {
 403         int ret;
 404
 405         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 406         if (ret == 0 && write)
 407                 sysctl_overcommit_kbytes = 0;
 408         return ret;
 409 }
 410
 411 int overcommit_kbytes_handler(struct ctl_table *table, int write,
 412                              void __user *buffer, size_t *lenp,
 413                              loff_t *ppos)
 414 {
 415         int ret;
 416
 417         ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 418         if (ret == 0 && write)
 419                 sysctl_overcommit_ratio = 0;
 420         return ret;
 421 }
 422
 423 /*
 424  * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
 425  */
 426 unsigned long vm_commit_limit(void)
 427 {
 428         unsigned long allowed;
 429
 430         if (sysctl_overcommit_kbytes)
 431                 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
 432         else
 433                 allowed = ((totalram_pages - hugetlb_total_pages())
 434                            * sysctl_overcommit_ratio / 100);
 435         allowed += total_swap_pages;
 436
 437         return allowed;
 438 }
 439
 440 /**
 441  * get_cmdline() - copy the cmdline value to a buffer.
 442  * @task:     the task whose cmdline value to copy.
 443  * @buffer:   the buffer to copy to.
 444  * @buflen:   the length of the buffer. Larger cmdline values are truncated
 445  *            to this length.
 446  * Returns the size of the cmdline field copied. Note that the copy does
 447  * not guarantee an ending NULL byte.
 448  */
 449 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
 450 {
 451         int res = 0;
 452         unsigned int len;
 453         struct mm_struct *mm = get_task_mm(task);
 454         unsigned long arg_start, arg_end, env_start, env_end;
 455         if (!mm)
 456                 goto out;
 457         if (!mm->arg_end)
 458                 goto out_mm;    /* Shh! No looking before we're done */
 459
 460         down_read(&mm->mmap_sem);
 461         arg_start = mm->arg_start;
 462         arg_end = mm->arg_end;
 463         env_start = mm->env_start;
 464         env_end = mm->env_end;
 465         up_read(&mm->mmap_sem);
 466
 467         len = arg_end - arg_start;
 468
 469         if (len > buflen)
 470                 len = buflen;
 471
 472         res = access_process_vm(task, arg_start, buffer, len, 0);
 473
 474         /*
 475          * If the nul at the end of args has been overwritten, then
 476          * assume application is using setproctitle(3).
 477          */
 478         if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
 479                 len = strnlen(buffer, res);
 480                 if (len < res) {
 481                         res = len;
 482                 } else {
 483                         len = env_end - env_start;
 484                         if (len > buflen - res)
 485                                 len = buflen - res;
 486                         res += access_process_vm(task, env_start,
 487                                                  buffer+res, len, 0);
 488                         res = strnlen(buffer, res);
 489                 }
 490         }
 491 out_mm:
 492         mmput(mm);
 493 out:
 494         return res;
 495 }