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 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
 180                 struct vm_area_struct *prev, struct rb_node *rb_parent)
 181 {
 182         struct vm_area_struct *next;
 183
 184         vma->vm_prev = prev;
 185         if (prev) {
 186                 next = prev->vm_next;
 187                 prev->vm_next = vma;
 188         } else {
 189                 mm->mmap = vma;
 190                 if (rb_parent)
 191                         next = rb_entry(rb_parent,
 192                                         struct vm_area_struct, vm_rb);
 193                 else
 194                         next = NULL;
 195         }
 196         vma->vm_next = next;
 197         if (next)
 198                 next->vm_prev = vma;
 199 }
 200
 201 /* Check if the vma is being used as a stack by this task */
 202 static int vm_is_stack_for_task(struct task_struct *t,
 203                                 struct vm_area_struct *vma)
 204 {
 205         return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
 206 }
 207
 208 /*
 209  * Check if the vma is being used as a stack.
 210  * If is_group is non-zero, check in the entire thread group or else
 211  * just check in the current task. Returns the task_struct of the task
 212  * that the vma is stack for. Must be called under rcu_read_lock().
 213  */
 214 struct task_struct *task_of_stack(struct task_struct *task,
 215                                 struct vm_area_struct *vma, bool in_group)
 216 {
 217         if (vm_is_stack_for_task(task, vma))
 218                 return task;
 219
 220         if (in_group) {
 221                 struct task_struct *t;
 222
 223                 for_each_thread(task, t) {
 224                         if (vm_is_stack_for_task(t, vma))
 225                                 return t;
 226                 }
 227         }
 228
 229         return NULL;
 230 }
 231
 232 #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
 233 void arch_pick_mmap_layout(struct mm_struct *mm)
 234 {
 235         mm->mmap_base = TASK_UNMAPPED_BASE;
 236         mm->get_unmapped_area = arch_get_unmapped_area;
 237 }
 238 #endif
 239
 240 /*
 241  * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
 242  * back to the regular GUP.
 243  * If the architecture not support this function, simply return with no
 244  * page pinned
 245  */
 246 int __weak __get_user_pages_fast(unsigned long start,
 247                                  int nr_pages, int write, struct page **pages)
 248 {
 249         return 0;
 250 }
 251 EXPORT_SYMBOL_GPL(__get_user_pages_fast);
 252
 253 /**
 254  * get_user_pages_fast() - pin user pages in memory
 255  * @start:      starting user address
 256  * @nr_pages:   number of pages from start to pin
 257  * @write:      whether pages will be written to
 258  * @pages:      array that receives pointers to the pages pinned.
 259  *              Should be at least nr_pages long.
 260  *
 261  * Returns number of pages pinned. This may be fewer than the number
 262  * requested. If nr_pages is 0 or negative, returns 0. If no pages
 263  * were pinned, returns -errno.
 264  *
 265  * get_user_pages_fast provides equivalent functionality to get_user_pages,
 266  * operating on current and current->mm, with force=0 and vma=NULL. However
 267  * unlike get_user_pages, it must be called without mmap_sem held.
 268  *
 269  * get_user_pages_fast may take mmap_sem and page table locks, so no
 270  * assumptions can be made about lack of locking. get_user_pages_fast is to be
 271  * implemented in a way that is advantageous (vs get_user_pages()) when the
 272  * user memory area is already faulted in and present in ptes. However if the
 273  * pages have to be faulted in, it may turn out to be slightly slower so
 274  * callers need to carefully consider what to use. On many architectures,
 275  * get_user_pages_fast simply falls back to get_user_pages.
 276  */
 277 int __weak get_user_pages_fast(unsigned long start,
 278                                 int nr_pages, int write, struct page **pages)
 279 {
 280         struct mm_struct *mm = current->mm;
 281         return get_user_pages_unlocked(current, mm, start, nr_pages,
 282                                        write, 0, pages);
 283 }
 284 EXPORT_SYMBOL_GPL(get_user_pages_fast);
 285
 286 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
 287         unsigned long len, unsigned long prot,
 288         unsigned long flag, unsigned long pgoff)
 289 {
 290         unsigned long ret;
 291         struct mm_struct *mm = current->mm;
 292         unsigned long populate;
 293
 294         ret = security_mmap_file(file, prot, flag);
 295         if (!ret) {
 296                 down_write(&mm->mmap_sem);
 297                 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
 298                                     &populate);
 299                 up_write(&mm->mmap_sem);
 300                 if (populate)
 301                         mm_populate(ret, populate);
 302         }
 303         return ret;
 304 }
 305
 306 unsigned long vm_mmap(struct file *file, unsigned long addr,
 307         unsigned long len, unsigned long prot,
 308         unsigned long flag, unsigned long offset)
 309 {
 310         if (unlikely(offset + PAGE_ALIGN(len) < offset))
 311                 return -EINVAL;
 312         if (unlikely(offset & ~PAGE_MASK))
 313                 return -EINVAL;
 314
 315         return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
 316 }
 317 EXPORT_SYMBOL(vm_mmap);
 318
 319 void kvfree(const void *addr)
 320 {
 321         if (is_vmalloc_addr(addr))
 322                 vfree(addr);
 323         else
 324                 kfree(addr);
 325 }
 326 EXPORT_SYMBOL(kvfree);
 327
 328 static inline void *__page_rmapping(struct page *page)
 329 {
 330         unsigned long mapping;
 331
 332         mapping = (unsigned long)page->mapping;
 333         mapping &= ~PAGE_MAPPING_FLAGS;
 334
 335         return (void *)mapping;
 336 }
 337
 338 /* Neutral page->mapping pointer to address_space or anon_vma or other */
 339 void *page_rmapping(struct page *page)
 340 {
 341         page = compound_head(page);
 342         return __page_rmapping(page);
 343 }
 344
 345 struct anon_vma *page_anon_vma(struct page *page)
 346 {
 347         unsigned long mapping;
 348
 349         page = compound_head(page);
 350         mapping = (unsigned long)page->mapping;
 351         if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
 352                 return NULL;
 353         return __page_rmapping(page);
 354 }
 355
 356 struct address_space *page_mapping(struct page *page)
 357 {
 358         unsigned long mapping;
 359
 360         /* This happens if someone calls flush_dcache_page on slab page */
 361         if (unlikely(PageSlab(page)))
 362                 return NULL;
 363
 364         if (unlikely(PageSwapCache(page))) {
 365                 swp_entry_t entry;
 366
 367                 entry.val = page_private(page);
 368                 return swap_address_space(entry);
 369         }
 370
 371         mapping = (unsigned long)page->mapping;
 372         if (mapping & PAGE_MAPPING_FLAGS)
 373                 return NULL;
 374         return page->mapping;
 375 }
 376
 377 int overcommit_ratio_handler(struct ctl_table *table, int write,
 378                              void __user *buffer, size_t *lenp,
 379                              loff_t *ppos)
 380 {
 381         int ret;
 382
 383         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 384         if (ret == 0 && write)
 385                 sysctl_overcommit_kbytes = 0;
 386         return ret;
 387 }
 388
 389 int overcommit_kbytes_handler(struct ctl_table *table, int write,
 390                              void __user *buffer, size_t *lenp,
 391                              loff_t *ppos)
 392 {
 393         int ret;
 394
 395         ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 396         if (ret == 0 && write)
 397                 sysctl_overcommit_ratio = 0;
 398         return ret;
 399 }
 400
 401 /*
 402  * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
 403  */
 404 unsigned long vm_commit_limit(void)
 405 {
 406         unsigned long allowed;
 407
 408         if (sysctl_overcommit_kbytes)
 409                 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
 410         else
 411                 allowed = ((totalram_pages - hugetlb_total_pages())
 412                            * sysctl_overcommit_ratio / 100);
 413         allowed += total_swap_pages;
 414
 415         return allowed;
 416 }
 417
 418 /**
 419  * get_cmdline() - copy the cmdline value to a buffer.
 420  * @task:     the task whose cmdline value to copy.
 421  * @buffer:   the buffer to copy to.
 422  * @buflen:   the length of the buffer. Larger cmdline values are truncated
 423  *            to this length.
 424  * Returns the size of the cmdline field copied. Note that the copy does
 425  * not guarantee an ending NULL byte.
 426  */
 427 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
 428 {
 429         int res = 0;
 430         unsigned int len;
 431         struct mm_struct *mm = get_task_mm(task);
 432         if (!mm)
 433                 goto out;
 434         if (!mm->arg_end)
 435                 goto out_mm;    /* Shh! No looking before we're done */
 436
 437         len = mm->arg_end - mm->arg_start;
 438
 439         if (len > buflen)
 440                 len = buflen;
 441
 442         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 443
 444         /*
 445          * If the nul at the end of args has been overwritten, then
 446          * assume application is using setproctitle(3).
 447          */
 448         if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
 449                 len = strnlen(buffer, res);
 450                 if (len < res) {
 451                         res = len;
 452                 } else {
 453                         len = mm->env_end - mm->env_start;
 454                         if (len > buflen - res)
 455                                 len = buflen - res;
 456                         res += access_process_vm(task, mm->env_start,
 457                                                  buffer+res, len, 0);
 458                         res = strnlen(buffer, res);
 459                 }
 460         }
 461 out_mm:
 462         mmput(mm);
 463 out:
 464         return res;
 465 }