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/uaccess.h>
  16
  17 #include "internal.h"
  18
  19 /**
  20  * kstrdup - allocate space for and copy an existing string
  21  * @s: the string to duplicate
  22  * @gfp: the GFP mask used in the kmalloc() call when allocating memory
  23  */
  24 char *kstrdup(const char *s, gfp_t gfp)
  25 {
  26         size_t len;
  27         char *buf;
  28
  29         if (!s)
  30                 return NULL;
  31
  32         len = strlen(s) + 1;
  33         buf = kmalloc_track_caller(len, gfp);
  34         if (buf)
  35                 memcpy(buf, s, len);
  36         return buf;
  37 }
  38 EXPORT_SYMBOL(kstrdup);
  39
  40 /**
  41  * kstrndup - allocate space for and copy an existing string
  42  * @s: the string to duplicate
  43  * @max: read at most @max chars from @s
  44  * @gfp: the GFP mask used in the kmalloc() call when allocating memory
  45  */
  46 char *kstrndup(const char *s, size_t max, gfp_t gfp)
  47 {
  48         size_t len;
  49         char *buf;
  50
  51         if (!s)
  52                 return NULL;
  53
  54         len = strnlen(s, max);
  55         buf = kmalloc_track_caller(len+1, gfp);
  56         if (buf) {
  57                 memcpy(buf, s, len);
  58                 buf[len] = '\0';
  59         }
  60         return buf;
  61 }
  62 EXPORT_SYMBOL(kstrndup);
  63
  64 /**
  65  * kmemdup - duplicate region of memory
  66  *
  67  * @src: memory region to duplicate
  68  * @len: memory region length
  69  * @gfp: GFP mask to use
  70  */
  71 void *kmemdup(const void *src, size_t len, gfp_t gfp)
  72 {
  73         void *p;
  74
  75         p = kmalloc_track_caller(len, gfp);
  76         if (p)
  77                 memcpy(p, src, len);
  78         return p;
  79 }
  80 EXPORT_SYMBOL(kmemdup);
  81
  82 /**
  83  * memdup_user - duplicate memory region from user space
  84  *
  85  * @src: source address in user space
  86  * @len: number of bytes to copy
  87  *
  88  * Returns an ERR_PTR() on failure.
  89  */
  90 void *memdup_user(const void __user *src, size_t len)
  91 {
  92         void *p;
  93
  94         /*
  95          * Always use GFP_KERNEL, since copy_from_user() can sleep and
  96          * cause pagefault, which makes it pointless to use GFP_NOFS
  97          * or GFP_ATOMIC.
  98          */
  99         p = kmalloc_track_caller(len, GFP_KERNEL);
 100         if (!p)
 101                 return ERR_PTR(-ENOMEM);
 102
 103         if (copy_from_user(p, src, len)) {
 104                 kfree(p);
 105                 return ERR_PTR(-EFAULT);
 106         }
 107
 108         return p;
 109 }
 110 EXPORT_SYMBOL(memdup_user);
 111
 112 /*
 113  * strndup_user - duplicate an existing string from user space
 114  * @s: The string to duplicate
 115  * @n: Maximum number of bytes to copy, including the trailing NUL.
 116  */
 117 char *strndup_user(const char __user *s, long n)
 118 {
 119         char *p;
 120         long length;
 121
 122         length = strnlen_user(s, n);
 123
 124         if (!length)
 125                 return ERR_PTR(-EFAULT);
 126
 127         if (length > n)
 128                 return ERR_PTR(-EINVAL);
 129
 130         p = memdup_user(s, length);
 131
 132         if (IS_ERR(p))
 133                 return p;
 134
 135         p[length - 1] = '\0';
 136
 137         return p;
 138 }
 139 EXPORT_SYMBOL(strndup_user);
 140
 141 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
 142                 struct vm_area_struct *prev, struct rb_node *rb_parent)
 143 {
 144         struct vm_area_struct *next;
 145
 146         vma->vm_prev = prev;
 147         if (prev) {
 148                 next = prev->vm_next;
 149                 prev->vm_next = vma;
 150         } else {
 151                 mm->mmap = vma;
 152                 if (rb_parent)
 153                         next = rb_entry(rb_parent,
 154                                         struct vm_area_struct, vm_rb);
 155                 else
 156                         next = NULL;
 157         }
 158         vma->vm_next = next;
 159         if (next)
 160                 next->vm_prev = vma;
 161 }
 162
 163 /* Check if the vma is being used as a stack by this task */
 164 static int vm_is_stack_for_task(struct task_struct *t,
 165                                 struct vm_area_struct *vma)
 166 {
 167         return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
 168 }
 169
 170 /*
 171  * Check if the vma is being used as a stack.
 172  * If is_group is non-zero, check in the entire thread group or else
 173  * just check in the current task. Returns the task_struct of the task
 174  * that the vma is stack for. Must be called under rcu_read_lock().
 175  */
 176 struct task_struct *task_of_stack(struct task_struct *task,
 177                                 struct vm_area_struct *vma, bool in_group)
 178 {
 179         if (vm_is_stack_for_task(task, vma))
 180                 return task;
 181
 182         if (in_group) {
 183                 struct task_struct *t;
 184
 185                 for_each_thread(task, t) {
 186                         if (vm_is_stack_for_task(t, vma))
 187                                 return t;
 188                 }
 189         }
 190
 191         return NULL;
 192 }
 193
 194 #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
 195 void arch_pick_mmap_layout(struct mm_struct *mm)
 196 {
 197         mm->mmap_base = TASK_UNMAPPED_BASE;
 198         mm->get_unmapped_area = arch_get_unmapped_area;
 199 }
 200 #endif
 201
 202 /*
 203  * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
 204  * back to the regular GUP.
 205  * If the architecture not support this function, simply return with no
 206  * page pinned
 207  */
 208 int __weak __get_user_pages_fast(unsigned long start,
 209                                  int nr_pages, int write, struct page **pages)
 210 {
 211         return 0;
 212 }
 213 EXPORT_SYMBOL_GPL(__get_user_pages_fast);
 214
 215 /**
 216  * get_user_pages_fast() - pin user pages in memory
 217  * @start:      starting user address
 218  * @nr_pages:   number of pages from start to pin
 219  * @write:      whether pages will be written to
 220  * @pages:      array that receives pointers to the pages pinned.
 221  *              Should be at least nr_pages long.
 222  *
 223  * Returns number of pages pinned. This may be fewer than the number
 224  * requested. If nr_pages is 0 or negative, returns 0. If no pages
 225  * were pinned, returns -errno.
 226  *
 227  * get_user_pages_fast provides equivalent functionality to get_user_pages,
 228  * operating on current and current->mm, with force=0 and vma=NULL. However
 229  * unlike get_user_pages, it must be called without mmap_sem held.
 230  *
 231  * get_user_pages_fast may take mmap_sem and page table locks, so no
 232  * assumptions can be made about lack of locking. get_user_pages_fast is to be
 233  * implemented in a way that is advantageous (vs get_user_pages()) when the
 234  * user memory area is already faulted in and present in ptes. However if the
 235  * pages have to be faulted in, it may turn out to be slightly slower so
 236  * callers need to carefully consider what to use. On many architectures,
 237  * get_user_pages_fast simply falls back to get_user_pages.
 238  */
 239 int __weak get_user_pages_fast(unsigned long start,
 240                                 int nr_pages, int write, struct page **pages)
 241 {
 242         struct mm_struct *mm = current->mm;
 243         return get_user_pages_unlocked(current, mm, start, nr_pages,
 244                                        write, 0, pages);
 245 }
 246 EXPORT_SYMBOL_GPL(get_user_pages_fast);
 247
 248 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
 249         unsigned long len, unsigned long prot,
 250         unsigned long flag, unsigned long pgoff)
 251 {
 252         unsigned long ret;
 253         struct mm_struct *mm = current->mm;
 254         unsigned long populate;
 255
 256         ret = security_mmap_file(file, prot, flag);
 257         if (!ret) {
 258                 down_write(&mm->mmap_sem);
 259                 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
 260                                     &populate);
 261                 up_write(&mm->mmap_sem);
 262                 if (populate)
 263                         mm_populate(ret, populate);
 264         }
 265         return ret;
 266 }
 267
 268 unsigned long vm_mmap(struct file *file, unsigned long addr,
 269         unsigned long len, unsigned long prot,
 270         unsigned long flag, unsigned long offset)
 271 {
 272         if (unlikely(offset + PAGE_ALIGN(len) < offset))
 273                 return -EINVAL;
 274         if (unlikely(offset & ~PAGE_MASK))
 275                 return -EINVAL;
 276
 277         return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
 278 }
 279 EXPORT_SYMBOL(vm_mmap);
 280
 281 void kvfree(const void *addr)
 282 {
 283         if (is_vmalloc_addr(addr))
 284                 vfree(addr);
 285         else
 286                 kfree(addr);
 287 }
 288 EXPORT_SYMBOL(kvfree);
 289
 290 struct address_space *page_mapping(struct page *page)
 291 {
 292         struct address_space *mapping = page->mapping;
 293
 294         /* This happens if someone calls flush_dcache_page on slab page */
 295         if (unlikely(PageSlab(page)))
 296                 return NULL;
 297
 298         if (unlikely(PageSwapCache(page))) {
 299                 swp_entry_t entry;
 300
 301                 entry.val = page_private(page);
 302                 mapping = swap_address_space(entry);
 303         } else if ((unsigned long)mapping & PAGE_MAPPING_ANON)
 304                 mapping = NULL;
 305         return mapping;
 306 }
 307
 308 int overcommit_ratio_handler(struct ctl_table *table, int write,
 309                              void __user *buffer, size_t *lenp,
 310                              loff_t *ppos)
 311 {
 312         int ret;
 313
 314         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 315         if (ret == 0 && write)
 316                 sysctl_overcommit_kbytes = 0;
 317         return ret;
 318 }
 319
 320 int overcommit_kbytes_handler(struct ctl_table *table, int write,
 321                              void __user *buffer, size_t *lenp,
 322                              loff_t *ppos)
 323 {
 324         int ret;
 325
 326         ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 327         if (ret == 0 && write)
 328                 sysctl_overcommit_ratio = 0;
 329         return ret;
 330 }
 331
 332 /*
 333  * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
 334  */
 335 unsigned long vm_commit_limit(void)
 336 {
 337         unsigned long allowed;
 338
 339         if (sysctl_overcommit_kbytes)
 340                 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
 341         else
 342                 allowed = ((totalram_pages - hugetlb_total_pages())
 343                            * sysctl_overcommit_ratio / 100);
 344         allowed += total_swap_pages;
 345
 346         return allowed;
 347 }
 348
 349 /**
 350  * get_cmdline() - copy the cmdline value to a buffer.
 351  * @task:     the task whose cmdline value to copy.
 352  * @buffer:   the buffer to copy to.
 353  * @buflen:   the length of the buffer. Larger cmdline values are truncated
 354  *            to this length.
 355  * Returns the size of the cmdline field copied. Note that the copy does
 356  * not guarantee an ending NULL byte.
 357  */
 358 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
 359 {
 360         int res = 0;
 361         unsigned int len;
 362         struct mm_struct *mm = get_task_mm(task);
 363         if (!mm)
 364                 goto out;
 365         if (!mm->arg_end)
 366                 goto out_mm;    /* Shh! No looking before we're done */
 367
 368         len = mm->arg_end - mm->arg_start;
 369
 370         if (len > buflen)
 371                 len = buflen;
 372
 373         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 374
 375         /*
 376          * If the nul at the end of args has been overwritten, then
 377          * assume application is using setproctitle(3).
 378          */
 379         if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
 380                 len = strnlen(buffer, res);
 381                 if (len < res) {
 382                         res = len;
 383                 } else {
 384                         len = mm->env_end - mm->env_start;
 385                         if (len > buflen - res)
 386                                 len = buflen - res;
 387                         res += access_process_vm(task, mm->env_start,
 388                                                  buffer+res, len, 0);
 389                         res = strnlen(buffer, res);
 390                 }
 391         }
 392 out_mm:
 393         mmput(mm);
 394 out:
 395         return res;
 396 }